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 |
|---|---|---|---|---|---|
Long-term effects of the growing population of HIV-treated people in Southern Africa on individuals and the public health sector at large are not yet understood . This study proposes a novel ‘ratio’ model that relates CD4+ T-cell counts of HIV-infected individuals to the CD4+ count reference values from healthy populations . We use mixed-effects regression to fit the model to data from 1616 children ( median age 4 . 3 years at ART initiation ) and 14 , 542 adults ( median age 36 years at ART initiation ) . We found that the scaled carrying capacity , maximum CD4+ count relative to an HIV-negative individual of similar age , and baseline scaled CD4+ counts were closer to healthy values in children than in adults . Post-ART initiation , CD4+ growth rate was inversely correlated with baseline CD4+ T-cell counts , and consequently higher in adults than children . Our results highlight the impacts of age on dynamics of the immune system of healthy and HIV-infected individuals .
The key outcome variable is scaled CD4+ T-cell count . In both infected adults and children , the cell counts post-ART initiation were scaled by reference values from healthy populations , to obtain the outcome variable . For HIV-infected children these reference values were calculated from the cross-sectional data ( see description in Appendix 1 ) of healthy children at specific ages , due to the large variability in CD4+ T-cell counts in the early years of life . For the reference values by age , a single exponential model was fitted to the healthy children’s cross-sectional data and continuous population estimates were simulated ( see Appendix 1—table 1 and Appendix 1—Figure 1 ) . These were within the normal CD4+ T-cell counts ranges published in South Africa ( Lawrie et al . , 2009; Lawrie et al . , 2015 ) and elsewhere ( Idigbe et al . , 2010; Pediatric AIDS Clinical Trials Group et al . , 2003 ) . We then scaled all HIV-infected children’s CD4+ T-cell counts as follows: ( 8 ) zi , ja=xi , jay ( a ) , 1≤i ≤N , 0 ≤j ≤ ni , 0 ≤a ≤ 203 , where zi , j is the scaled CD4+ T-cell counts of patient i of age a ( in months ) at time j ( measured as time since ART initiation ) ; xi , j is the CD4+ T-cell counts of an HIV-infected child i of age a and y ( a ) is the CD4+ T-cell counts of a healthy child of similar age a as patient i . Scaling CD4+ T-cell counts of HIV-infected children by that of healthy children of similar ages enabled the comparison of CD4+ T-cell counts responses across ages , while simultaneously accounting for the child’s growth and immune system development . For adults , normal reference values were estimated using values obtained from the literature ( Malaza et al . , 2013; Lawrie et al . , 2009 , Ngowi et al . , 2009; Lugada et al . , 2004; Institute of Human Virology/Plateau State Specialist Hospital AIDS Prevention in Nigeria Study Team et al . , 2005; Oladepo et al . , 2009 ) ( see details in Appendix 1—table 2 and Appendix 1—figure 2 ) . Although healthy adult CD4+ T-cell counts are known to vary with age , this variability is small compared to that observed in children , with an average difference of 100 cells between 25 and 60 years old healthy adults [Oladepo et al . , 2009] vs 1500 cells difference between 3 months and 15 years old healthy children [Lawrie et al . , 2015] . Similarly , average CD4+ T-cell counts differences between male and female adults in South Africa were in the range of 80–150 cells ( Malaza et al . , 2013; Lawrie et al . , 2009 ) . We evaluated results obtained from scaling HIV-infected CD4+ counts by simulated age and sex-dependent reference CD4+ counts values , and found minor or no difference with those obtained from CD4+ T-cell counts scaled by a single value . We obtained exactly the same population estimates and very similar individual parameter estimates . These were evaluated and we found no significant differences ( see Supplementary file 2 ) . Thus , due to the unavailability of individual age-specific data for a South African healthy adult population , a single normal reference value , that is , a constant y of 800 CD4+ T-cells per µL , was used to scale all CD4+ T-cell count observations for adults on ART ( zi= xi , jy ) . Our analysis was restricted to patients with available baseline ( i . e . at antiretroviral treatment initiation ) CD4+ T-cell counts and sufficient observations to estimate all model parameters ( Figure 1 ) . We categorized as baseline any CD4+ count measurement that was taken within 15 days before or after the ART initiation date . Two model-fitting scenarios were defined based on the availability of enough observations to estimate the model parameters: Scenario 1: In which baseline scaled CD4+ T-cell counts were estimated , including 1312 children and 12 , 238 adults , with a minimum of five CD4+ T-cell counts measurements and no missing values for our variables of interest ( specified below ) . Scenario 2: Where baseline scaled CD4+ T-cell counts were used as a predictor , including 1616 children and 14 , 542 adults , with a minimum of four CD4+ T-cell counts measurements and no missing values for our variables of interest . These variables were CD4+ T-cell counts since ART initiation; viral load , age and body mass index at ART initiation; sex and suppression ( or not ) of viral load within 12 months of ART initiation . In this paper we will mostly discuss results of scenario 1 , where all parameters are estimated . The methodology used and most of the obtained results are applicable for scenario 2 , when baseline CD4+ T-cells counts data are available . We also present results of scenario 2 . Under both scenarios , in the initial model fitting , all parameters were assumed to vary per individual ( i . e . there were random effects on all parameters ) and their distributions was set as log-normal . Thus , for each parameter θ , the distribution of individual values was log ( θi ) ∼ ℵ ( μ , ω2 ) , with mean μ and variance ω2 . Random effects were assumed to be independent and the identity diagonal matrix was used for the variance–covariance structure . This assumption was later relaxed and different variance–covariance structures were subsequently evaluated . To compare models , the Akaike and Schwarz information criteria ( AIC and BIC respectively ) , computed by importance sampling , were used ( see Comets et al . , 2017 ) . Models with a full variance–covariance matrix were retained as this structure gave the lowest AIC and BIC values . Our choice of covariates was based on literature reviews ( Pinzone et al . , 2012; Sempa et al . , 2017 ) , data availability , and biological plausibility . We considered baseline characteristics that were measured within 15 days before or after ART initiation , namely age , z-score body-mass-index ( BMI , for children ) , and viral load . Other covariates included sex and viral suppression ( or not ) within 12 months of ART initiation . Patient age at ART initiation was expressed in months . Z-score BMI at baseline for children was calculated using WHO-Igrowup’s package ( WHO Multicentre Growth Reference Study Group , , 2006 ) . ‘Suppress’ , a binary , was defined as reaching an undetectable viral load ( <1000 copies/mL ) or not , within 12 months of ART initiation . Baseline viral load was log transformed to simplify model fitting . Baseline age and sex effects were included for all parameters , while viral load suppression , baseline log viral load and BMI z-scores for children , were only included for the parameters describing the longitudinal immune responses of individuals on ART , namely: scaled carrying capacity post-ART , CD4+ T-cell growth rate post-ART and baseline scaled CD4+ T-cell count . Final adjusted models were obtained using a backward–forward stepwise approach using p-value criterion for covariate inclusion and AIC , BIC for the overall models . To evaluate the robustness of our results , we compared the estimated parameters of the full analysis described above using only three covariates: sex , baseline age , and baseline viral load , with those from models with all covariates . All graphs and computations were produced in the statistical environment R ( R Development Core Team , 2017 ) . Nonlinear mixed model fitting employed the saemix R package ( Comets et al . , 2005 ) , which uses a stochastic approximation , expectation-maximization algorithm for parameter estimation . All final adult and children models ( Figures 3 and 4 ) converged relatively rapidly towards their estimated values , as shown by the log-likelihood graphs ( Figure 3—figure supplement 1 and Figure 4—figure supplement 1 ) . Even when initial values were marginally varied , the final models converged to similar estimates .
Cohort characteristics are summarized in Table 1 . For scenario 1 , the median number of clinical visits was 8 per patient , in both adults IQR ( 6 , 9 ) and children IQR ( 6 , 10 ) , with a median follow-up time of 3 years for adults IQR ( 2 , 5 ) and 4 years for children IQR ( 3 , 5 ) . At baseline , that is , ART initiation , the median age for children was 4 . 5 years IQR ( 1 . 4 , 7 . 9 ) and 36 years IQR ( 30 . 7 , 43 ) for adults . A clinical WHO stage III was most common in all patients at ART initiation . Among adults , there were more females than males ( 56 . 7% versus 43 . 2% respectively ) , but only slightly more females among children ( 51 . 3% versus 48 . 6% males ) . All patients in our data sets initiated therapy between 1997 and 2013 , with 95% of them initiating in the period , 2003–2012 . About 53 . 4% of children reached a CD4+ T-cell count of 500 per µL within 12 months of ART initiation , while only 4 . 4% of adults reached the same threshold . This number increases to 10 . 1% after 2 years of therapy for adults . The median baseline CD4+ T-cell counts for children and adults were low at 404 per µL IQR ( 159 . 7 , 706 . 2 ) and 128 per µL IQR ( 62 , 196 ) respectively ( Table 1 ) . Higher counts at ART initiation were more common in younger ( 0–5 year’s age group ) versus older children ( >5 years ) . In adults , no differences were found in CD4+ T-cell counts at baseline , for different age groups . Both children and adults that presented with higher counts at baseline had lower viral loads versus those with low baseline counts . The performance of both asymptotic and ratio models , in terms of AIC and BIC criteria , was found to be strongly dependent on the structure selected for the variance–covariance matrices ( see Supplementary file 3 ) . Initially , both models were defined with random effects on all parameters and a diagonal matrix structure for the random effects , that is , individual random effects were assumed independent . Different matrix structures were then compared and those with full matrices , that is , in which random effects are correlated , were found to be the best . For both adults and children , the AIC and BIC for the Ratio model were smaller than for the Asymptotic model when baseline scaled CD4+ T-cell counts were estimated ( Scenario 1 ) . When the baseline scaled CD4+ T-cell counts were used as a predictor ( Scenario 2 ) this reversed . However , the number of parameters estimated by the two models was different . For the Asymptotic model there were three for scenario one vs two for scenario 2 , a 33% change , and for the ratio model there five for scenario one vs four for scenario 2 , a 20% change ( Table 2 ) . Given the different numbers of parameters , the Ratio performed well in comparison to the Asymptotic model . In terms of parameter estimations , both models estimated baseline scaled CD4+ of similar magnitudes , for HIV-infected adults and children ( Table 3 and Supplementary file 4 – Tables 1 and 2 ) . The asymptotic model predicts a sharp increase of CD4+ counts , which rapidly reaches an asymptote , while the ratio model predicts a sharp increase that is followed by a slower but still gradual increase of scaled CD4+ , up to more than 10 years after ART initiation ( Figures 3 and 4 ) . We note that the parameter estimates obtained using a baseline CD4 scaling constant of 800 cells/µL for healthy adults were comparable to those obtained using age-dependent healthy adults CD4 values ( Supplementary file 2 ) . Similarly , the estimates obtained using a baseline CD4 scaling constant of 800 cells/µl for healthy adults were also comparable to those obtained without scaling ( i . e . using a scaling constant of 1 cell/µL ) for all parameters except z0 , whose estimate increased by about 800-fold ( Supplementary file 2 ) as expected . We found no major differences in the estimated fixed effects when adjusting the models for age and sex only , versus adjusting for all our covariates . The minor differences between both fits were the sizes of estimated parameters ( Supplementary file 4 – Table 2 ) , a few inconsistences in the effects of age on particular parameters ( adults and children ) and sex effects on post-ART scaled carrying capacity ( adults only ) . Therefore , we only describe results for the models adjusted for all covariates . These results are presented in Table 3 . Note that a sample of individual fits is shown in Figure 3—figure supplement 2 and Figure 4—figure supplement 2 .
Fitting our model to the data demonstrated that larger values of baseline scaled CD4+ T-cell counts are associated with larger values of scaled CD4+ T-cell counts at any subsequent time . This is the consequence of equation 4 , which is in line with the findings of prior studies in which higher baseline CD4+ T-cell counts were associated with a higher final or plateauing value for CD4+ T-cell counts ( Moore and Keruly , 2007 ) . Similarly , patients with very low baseline CD4+ T-cell counts , <350 cells/µL , often fail to reach normal values even after long durations of therapy ( Nakanjako , 2016; Swiss HIV Cohort Study et al . , 2005; Moore and Keruly , 2007; Kelley et al . , 2009 ) . We also found that the scaled carrying capacity of CD4+ T-cells in HIV-infected individuals was negatively correlated with the baseline scaled CD4+ T-cell counts in both adults and children , respectively , −0 . 64 and −0 . 81 . This seems reasonable in that the closer an individual is to their normal or optimal CD4+ T-cell value at ART initiation , the less cell population expansion is required to reach normal levels . Our parameter estimates demonstrated that the scaled cellular carrying capacity was higher in individuals on ART than in those who were healthy . In both HIV-infected adults and children the scaled carrying capacity post-ART was greater than 1 , meaning that baseline CD4+ T-cell counts were lower than their corresponding long-term homeostatic optimum . That is , individual CD4+ T-cell counts had to grow to reach normal levels . This is consistent with the notion that an impaired immune system usually experiences repair following treatment initiation . In contrast , the scaled capacity for healthy adults and children were both lower than unity . In children the value was 0 . 68 , which is consistent with a mechanistic understanding of the dynamics of healthy CD4+ T-cell counts in early life . In particular , there is an increase until the age of approximately 1 year and then a decrease thereafter ( Bains et al . , 2009 ) . Total blood and body volume increase throughout childhood associated with the shrinkage of the thymus , which is accompanied by a reduction of naïve T-cell production ( Hapuarachchi et al . , 2013; Hazenberg et al . , 2004 ) . In combination , this results in decreasing CD4+ T-cell counts per volume with age . In healthy adults , we found a scaled carrying capacity of 0 . 49 , suggesting a decreasing CD4+ T-cell trend with age . This is in agreement with other studies from elsewhere which found that CD4+ T-cell counts decreased from young adulthood to middle age ( Lugada et al . , 2004; Zeh et al . , 2011 ) . However , one study in healthy South African adults has described CD4+ T-cell counts increasing slightly until the age of 64 years old ( Malaza et al . , 2013 ) . We found that post-ART CD4+ T-cell growth rate depends inversely on the cell count at treatment initiation , that is , they are negatively correlated ( Table 4 ) . This implies that the higher the scaled CD4+ T-cell counts of an individual at ART initiation , that is , the closer it is to its healthy/normal true value , the lower the rate of recovery will be . This is reasonable given a decrease in the need to achieve normal levels and is in agreement with the findings of prior studies ( Lawn et al . , 2006; Sachsenberg et al . , 1998 ) . We found that CD4+ T-cell growth rates were higher in individuals on ART than in healthy individuals . This agrees with prior studies in which cell growth rates in adults on ART were estimated to be sixfold to tenfold greater than in healthy adults ( Sachsenberg et al . , 1998; Hazenberg et al . , 2000 ) . This is understandable bearing in mind the immune system’s effort after treatment initiation to re-fill the void of peripheral CD4+ T-cells destroyed during HIV infection . Biological studies have described this as an initial redistribution of memory T-cells from the lymph nodes into the blood stream , followed by homeostatic proliferation and production of naïve cells by the thymus ( Tsukamoto et al . , 2009; Autran , 1999 ) . Thus , given that HIV-infected individuals have much lower CD4+ T-cell counts they require higher CD4+ T-cell growth rates than healthy individuals . This behaviour was captured by our model . Interestingly , the differences in our estimates for CD4+ T-cell growth rate in healthy individuals versus patients on ART were slightly lower than those found elsewhere ( Sachsenberg et al . , 1998 ) . This may be due to: the differences in the total time of follow-up of our patients , selection bias in the original population data , differences in the demographics of the populations studied , the types of treatments administered , and changes in the WHO guidelines for ART initiation over time . The WHO guidelines for minimal CD4+ T-cell counts at initiation changed from 200 , to 350 , to 500 and later to initiation at diagnosis . Thus , post-ART CD4+ T-cell growth rate in HIV-infected individual might be smaller now than it may have been on average in the past , as CD4+ T-cell counts at ART initiation are now higher . Baseline scaled CD4+ T-cell count was lower in adults compared to children ( 0 . 2 vs 0 . 8 ) , that is , adults started treatment when their immune systems were more compromised compared to children . This could be due to the fact that children ( <17 years old ) are more likely vertically infected and , thus , they are more likely to be diagnosed early , owing to early testing and follow-up in the South African program for prevention of mother-to-child transmission . Adults , on the other hand , might be infected for an extended period and are consequently more likely to be more highly immune-compromised prior to diagnosis , compared to children . We found that older age is associated with a lower value of baseline scaled CD4+ T-cell count in children and a higher value in adults . Vertical HIV- infection and an extended duration without treatment may lead to greater immune compromise in older children . In contrast , higher baseline scaled CD4+ T-cell counts in older adults suggest that they may have been more health conscious than younger adults ( Prohaska et al . , 1985 ) , that is , younger adults tend to seek treatment later than older adults . Such findings might also be explained by the negative correlation between post-ART CD4+ T-cell growth rates and baseline CD4+ T-cell counts: younger adults have a higher growth rate as they start at a lower CD4+ T-cell count value ( Cornell et al . , 2012 ) . Thus , their CD4+ T-cell rebound value may be higher than that of older adults over a similar period of time ( Means et al . , 2016; International epidemiological Database to Evaluate AIDS ( IeDEA ) West Africa Collaboration et al . , 2012 ) . Our results make no inferences regarding the period an individual takes to reach their normal CD4+ T-cell count value . We found that adult females have lower post-ART scaled carrying capacities than men , meaning they need a lower CD4+ T-cell count rebound to reach normality . This suggests that female adults tend to initiate treatment earlier than men ( Cornell and Myer , 2013 ) . This was also validated by the fact that baseline scaled CD4+ T-cell counts were higher in females than in males . A prior study has also found that women tend to have higher plateauing CD4+ T-cell counts than men ( Means et al . , 2016 ) . In our study , women had lower rates of recovery than males which also supports our finding that higher baseline scaled CD4+ T-cell count is associated with a lower CD4+ T-cell growth rate . As they have higher CD4+ T-cell counts at ART initiation , and given the inverse correlation between the post-ART CD4+ T-cell growth rate and baseline scaled CD4+ T-cell count , women demonstrate lower growth rates . Previous studies have found that adult males and females spend a similar time on therapy before reaching their rebound set-point CD4+ T-cell count ( EuroSIDA group et al . , 2003; Patterson et al . , 2007 ) . We do not consider this a contradiction of our results as an individual with a higher baseline CD4+ T-cell count will have a lower cell growth rate compared to an individual that started ART with a low CD4+ count . Thus , after an equivalent period following ART initiation similar increases in CD4+ T-cell counts may have occurred . We believe that the diverging opinions regarding the effects of sex on immune outcomes in adults post-ART initiation might be explained by variations in the definitions of ‘immunological outcomes’ by different authors and variations in analyses conducted . This study has strengths and limitations . To ensure that patients had sufficient data to enable parameterization of the model , only subsets of the full data set were used in the analysis . This prevented over-fitting , but it might have introduced selection bias . However , with the exception of the percentage of people that suppressed viral load within 12 months of starting ART ( Table 1 ) , comparison of the summary statistics of the subset versus the full data set demonstrated that they were similar . Further , the results obtained from adjusting and not adjusting the model parameters with the variable , viral load suppression within 12 months , were also in agreement ( see Supplementary file 4 ) . We consequently believe that the subset used was representative of the full data set . Our model does not distinguish between naïve and memory sub-types of CD4+ T-cells . Prior studies have shown that these have different dynamics ( Di Mascio et al . , 2006; Bajaria et al . , 2002 ) . However , in routine HIV monitoring , for which we had data , the different subtypes are not measured . As ART treatment is for life , separating analyses for children from that for adults does not account for children growing into adulthood . Future studies might express the scaled carrying capacity as a function of age rather than as a variable parameter . Ignoring the thymus’s contribution to CD4+ T-cell recovery , particularly in the earlier years of life , might have resulted in an overestimation of r and s parameters which are aggregate growth rates . Lastly , scaling the CD4+ T-cell counts of adults by a single average healthy CD4+ count value for adults might have introduced bias in the estimates of the scaled carrying capacity of healthy adults , due to variations of CD4+ T-cell counts across different age and demographic groups . Our study does provide insight into the effect of ageing on immune system dynamics in adults and children on ART compared to healthy individuals . The ratio model provides a more accurate estimation of CD4+ counts reconstitution than the asymptotic model as well as the ability to compare different immune system outcomes , for both healthy and HIV-treated individuals . Using scaled CD4+ T-cell counts allows for the evaluation of CD4+ counts trajectories , which is not possible with unscaled CD4+ T-cell counts . We found large variations in CD4+ T-cell growth rates and scaled carrying capacities between individuals , highlighting the need to evaluate ART outcomes on an individual level . This calls for improved patient monitoring strategies . The strong inverse correlation between baseline scaled CD4+ T-cell count and the scaled carrying capacity emphasizes the importance of early ART initiation , regardless of age or state of disease progression . We found that post-ART CD4+ T-cell growth rate is not associated with a patient’s age , but it is associated with higher baseline viral load . With the expansion of an aging population on ART , understanding long-term effects of the treatment on the immune system is critical to ensure that the best care is delivered to HIV-infected patients . | The human immunodeficiency virus ( HIV ) remains an ongoing global pandemic . There is currently no cure for HIV , but antiretroviral therapies can keep the virus in check and allow individuals with HIV to live longer , healthier lives . These drugs work in two ways . They block the ability of the virus to multiply and they allow numbers of an important type of infection-fighting cell called CD4+ T cells to rebound . As more patients with HIV survive and transition from one life stage to the next , it is critical to understand how long-term antiretroviral therapies will affect normal age-related changes in their immune systems . The health of an immune system can be evaluated by looking at the number of CD4+ T cells an individual has , though this will vary by age and location . Clinicians use the same metrics to assess the immune health of individuals with HIV , however , as they age , it becomes a challenge to identify if a patient’s immune system recovers normally or insufficiently . Thus , learning more about age-related differences in CD4+ T cells in people living with HIV may help improve their care . Using data from 1 , 616 children and 14 , 542 adults from South Africa , Ujeneza et al . created a simple mathematical model that can compare the immune system of person with HIV with the immune system of a similarly aged healthy individual . The model shows that among individuals with HIV receiving antiretroviral therapies , children have CD4+ T-cell numbers that are closest to the numbers seen in healthy individuals of the same age . This suggests that children may be more able to recover immune system function than adults after beginning treatment . Children also start antiretroviral therapies before their immune system has been severely damaged , while adults tend to start treatment much later when they have fewer CD4+ T cells left . Ujeneza et al . show that the fewer CD4+ T cells a person has when they start treatment , the faster the number of these cells grows after starting treatment . This suggests that the more damaged the immune system is , the harder it works to recover . This reinforces the need to identify people infected with HIV as soon as possible through testing and to begin treatment promptly . The new model may help clinicians and policy makers develop screening and treatment protocols tailored to the specific needs of children and adults living with HIV . | [
"Abstract",
"Introduction",
"Results",
"Discussion"
] | [
"microbiology",
"and",
"infectious",
"disease"
] | 2021 | A mechanistic model for long-term immunological outcomes in South African HIV-infected children and adults receiving ART |
Synaptic transmission consists of fast and slow components of neurotransmitter release . Here we show that these components are mediated by distinct exocytic proteins . The Caenorhabditis elegans unc-13 gene is required for SV exocytosis , and encodes long and short isoforms ( UNC-13L and S ) . Fast release was mediated by UNC-13L , whereas slow release required both UNC-13 proteins and was inhibited by Tomosyn . The spatial location of each protein correlated with its effect . Proteins adjacent to the dense projection mediated fast release , while those controlling slow release were more distal or diffuse . Two UNC-13L domains accelerated release . C2A , which binds RIM ( a protein associated with calcium channels ) , anchored UNC-13 at active zones and shortened the latency of release . A calmodulin binding site accelerated release but had little effect on UNC-13’s spatial localization . These results suggest that UNC-13L , UNC-13S , and Tomosyn form a molecular code that dictates the timing of neurotransmitter release .
The amount of neurotransmitter released at a synapse , and consequently synaptic strength , is often limited by the number of synapse vesicles ( SVs ) available for release . To become competent to undergo calcium-evoked fusion , SVs must first physically attach to the plasma membrane ( termed docking ) , and then must undergo a second process that has been termed priming . Docking and priming are both promoted by UNC-13/Munc13 proteins , whereas Tomosyn inhibits both processes ( Verhage and Sorensen , 2008 ) . Mammals express four Munc13 isoforms ( Munc13-1 , ubMunc13-2 , bMunc13-2 , and Munc13-3 ) , while Caenorhabditis elegans expresses two . It is not known if each Munc13 protein has distinct functions . Do they mediate different forms of plasticity ? Do they promote exocytosis of different subpopulations of SVs ? At most synapses , neurotransmitter release comprises multiple populations of SVs that fuse with distinct kinetics and distinct release probabilities ( Neher and Sakaba , 2008 ) . These kinetic components of release are thought to be mediated by SVs in different spatial domains of the nerve terminal . Rapid ( or synchronous ) release occurs within a few milliseconds and is proposed to consist of fusion of SVs that are close to calcium entry sites . Delayed release occurs over tens to hundreds of milliseconds and is thought to be mediated by fusion of SVs that are farther from calcium channels . Although a great deal is known about how synchronous and delayed release are regulated by calcium and by activity ( Zucker and Regehr , 2002 ) , relatively little is known about the molecules that define these kinetic components of release . It is generally believed that distinct calcium sensors are utilized for fast and slow release . Many studies suggest that the calcium sensor for fast release is Synaptotagmin I that binds calcium with low affinity and rapid kinetics ( Chapman , 2008; Pang and Sudhof , 2010 ) . A recent study proposed that the calcium sensor for slow release is Doc2α that binds calcium with high affinity and slow kinetics ( Yao et al . , 2011 ) . Beyond the use of distinct calcium sensors , it is unclear if fast and slow release are distinguished by other exocytic proteins . Several important questions remain unanswered . Do SVs equilibrate between the slow and fast pools ? Is there a molecular code that determines whether SVs enter the fast or slow pool ? Are fast and slow SVs regulated by distinct mechanisms ? We utilized the C . elegans cholinergic neuromuscular junction ( NMJ ) as a model to address these questions . Transmission at this synapse is mediated by graded release of neurotransmitter , whereby release varies with the strength of depolarization ( Liu et al . , 2009 ) . When activity is low , transmission consists of endogenous excitatory postsynaptic currents ( EPSCs ) that are evoked by fusion of a single SV ( Liu et al . , 2005 ) . Following forced depolarization of the motor neuron ( with a brief current pulse ) , large evoked EPSCs are produced consisting of the synchronous release of several hundred SVs . The ultrastructure of this synapse has been extensively characterized using the high-pressure freeze preservation technique . Each synapse contains a single dense projection that is surrounded by a pool of docked SVs , where docking occurs over a 1-μm domain of the presynaptic plasma membrane ( Weimer et al . , 2006; Hammarlund et al . , 2007 ) . In immunostained electron micrographs , UNC-2 calcium channels are enriched at dense projections , indicating that they are the sites of calcium entry ( Gracheva et al . , 2008 ) . The active zone proteins UNC-10/RIM and UNC-13 promote docking of distinct population of SVs . In mutants lacking UNC-10/RIM , there was a reduction of docked SVs within 50 nm of the dense projection while SVs docked more distally were unaffected ( Weimer et al . , 2006; Gracheva et al . , 2008 ) . By contrast , in unc-13 mutants , there was a uniform reduction of docked SVs across the entire 1-μm domain ( Gracheva et al . , 2006; Weimer et al . , 2006; Hammarlund et al . , 2007 ) . Corresponding patterns of docking defects were observed in mouse Munc13 and RIM knockouts ( Siksou et al . , 2009; Han et al . , 2011 ) . UNC-10/RIM and UNC-13 immunostaining was also concentrated near dense projections in electron micrographs , consistent with their playing a relatively direct role in SV docking ( Weimer et al . , 2006 ) . These results suggest that different proteins promote docking and fusion of distinct SV subpopulations . The functional significance of this spatial SV docking pattern remains unclear . Do proximal and distally docked SVs represent functionally distinct populations of SVs ? Do they mediate different forms of neurotransmitter release or different forms of synaptic plasticity ? Although this synapse has been characterized electrophysiologically , the mechanisms regulating release kinetics have not been analyzed . Here we show that ACh release at this synapse consists of fast and slow components , and that these components are mediated by distinct SV priming factors . Fast release is mediated by exocytosis of SVs that are docked and primed by a long UNC-13 isoform ( UNC-13L ) . Slow release is mediated by two UNC-13 isoforms ( UNC-13L and S ) and is inhibited by a third priming protein , the syntaxin-binding protein Tomosyn . These results suggest that UNC-13L , UNC-13S , and Tomosyn form a protein code that dictates the timing of neurotransmitter release .
The unc-13 gene contains two promoters . The upstream promoter drives expression of long isoforms , while a downstream promoter ( which lies in the intron between exons 13 and 14 ) drives expression of short isoforms ( hereafter referred to as UNC-13L and UNC-13S , respectively ) . UNC-13L and UNC-13S share a common domain ( R , encoded by exons 15–31 ) that contains the C1 , C2B , C2C , and MUN domains ( Figure 1A; Kohn et al . , 2000 ) . UNC-13L proteins have an N-terminal domain ( NTD , exons 1–13 ) containing the C2A domain , which the UNC-13S NTD ( exon 14 ) lacks ( Figure 1A ) . Rodents have analogous long ( Munc13-1 and ubMunc13-2 ) and short ( bMunc13-2 and Munc13-3 ) Munc13 proteins ( Brose et al . , 2000 ) . In long Munc13 proteins , the C2A domain mediates formation of Munc13 homodimers and Munc13/RIM heterodimers ( Lu et al . , 2006 ) . C . elegans UNC-13L ( mCherry-tagged ) and UNC-10 RIM ( GFP-tagged ) proteins were colocalized at presynaptic terminals ( Figure 1B–D ) , consistent with immuno-EM studies indicating that they are concentrated near dense projections ( Weimer et al . , 2006 ) . By contrast , UNC-13S had a diffuse distribution in axons and was less colocalized with UNC-10 RIM ( Figure 1B–D ) . Their different localization patterns suggested that UNC-13L and UNC-13S may mediate different forms of release and , consequently , could have distinct effects on synaptic transmission . 10 . 7554/eLife . 00967 . 003Figure 1 . UNC-13L and UNC-13S have distinct localization patterns in axons . ( A ) The domain structure of UNC-13L ( encoded by exons 1–31 , excluding exon 14 ) and UNC-13S ( exons 14–31 ) are illustrated . Ligands for each domain are indicated . The R region ( 1205 aa , comprising C1 , C2B , MUN , and C2C domains ) is shared between the two isoforms and is encoded by exons 15–31 . Each isoform has a unique NTD . The L-NTD ( 610 aa ) is encoded by exons 1–13 and contains a C2A domain ( aa 1–96 ) and a predicted calmodulin binding site ( green box , aa 556–610 ) . The S-NTD ( 259 aa ) is encoded by exon 14 . The e51 and e1091 alleles correspond to nonsense mutations in L-NTD exons ( exons 11 and 12 , respectively ) . The s69 allele corresponds to a 5 base pair deletion in an R domain exon ( exon 21 ) . ( B–D ) The localization of UNC-10/RIM in dorsal cord axons is compared to that of UNC-13L and UNC-13S . GFP-tagged UNC-10 and mCherry-tagged UNC-13L and UNC-13S were expressed in the DA and DB motor neurons of wild-type animals ( using the unc-129 promoter ) . Representative images ( B ) , line scans ( C ) of active zones ( identified by UNC-10/RIM fluorescence ) , and correlation coefficients ( D ) for UNC-10/RIM and UNC-13 fluorescence at synaptic puncta are shown . Line scans show normalized fluorescence values for UNC-10 ( green ) and UNC-13 ( red ) . The UNC-10 trace is the averaged line scan from 20 puncta . The red traces represent UNC-13 line scans at 10 representative active zones . Scale bar indicates 1 μm . Values that differ significantly are indicated ( ***p<0 . 001 ) . The number of animals analyzed is indicated for each genotype . Error bars indicate SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 00967 . 003 To determine how each UNC-13 protein contributes to synaptic function , we constructed animals that selectively express one or the other isoform . For these experiments , we introduced transgenes that express either UNC-13S or UNC-13L into unc-13 ( s69 ) mutants . The s69 allele corresponds to a 5 base pair deletion in an exon shared by UNC-13S and UNC-13L ( exon 21 ) , which shifts the reading frame thereby inactivating both isoforms ( Figure 1A; Kohn et al . , 2000 ) . Hereafter , we refer to these transgenic strains as UNC-13S- and UNC-13L-rescued animals . Both UNC-13 transgenes rescued the locomotion rate defects of unc-13 ( s69 ) mutants ( Figure 2A , B ) , indicating that both isoforms are able to restore synaptic function . 10 . 7554/eLife . 00967 . 004Figure 2 . UNC-13L and UNC-13S both reconstitute locomotion behavior in unc-13 mutants . ( A and B ) Locomotion behavior was analyzed for the indicated genotypes . ( A ) Representative 20 s locomotion trajectories are shown ( n = 10 animals for each genotype ) . The starting points for each trajectory were aligned for clarity . ( B ) Locomotion rates are compared . UNC-13L ( KP6893 ) and UNC-13S ( KP6899 ) refer to unc-13 ( s69 ) mutants containing the indicated transgenes . ( C and D ) Endogenous EPSCs were recorded from adult body wall muscles of the indicated genotypes . Representative traces ( C ) and summary data ( D ) are shown . Values that differ significantly from wild-type controls are indicated ( ***p<0 . 001 , *p<0 . 05 ) . The number of animals analyzed is indicated for each genotype . Error bars indicate SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 00967 . 00410 . 7554/eLife . 00967 . 005Figure 2—figure supplement 1 . UNC-13L and UNC-13S both restore SV priming . Sucrose-evoked EPSCs were recorded from adult body wall muscles of the indicated genotypes . Representative traces ( A ) and summary data ( B ) are shown . The bar above each trace indicates the 1 s application of sucrose . ( B ) The total synaptic charge transfer was measured for 1 s after initiation of the sucrose stimulus for each genotype . UNC-13L and UNC-13S refer to unc-13 ( s69 ) mutants containing the indicated transgene . Values that differ significantly from wild-type controls are indicated ( ***p<0 . 001 , **p<0 . 01 ) . The number of animals analyzed is indicated for each genotype . Error bars indicate SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 00967 . 005 To more directly assess changes in synaptic transmission , we recorded EPSCs from adult body wall muscles . We analyzed endogenous EPSCs that are postsynaptic currents evoked by fusion of a single SV ( Liu et al . , 2005 ) . Both UNC-13 transgenes rescued the endogenous EPSC defects of unc-13 ( s69 ) mutants ( Figure 2C , D ) . Endogenous EPSC rates and amplitudes were restored to wild-type levels in UNC-13L rescue . UNC-13S rescue exhibited lower EPSC rates , and slightly smaller endogenous EPSC amplitudes than in wild-type controls . Thus , expressing either UNC-13L or UNC-13S reconstituted synaptic transmission . For both locomotion and endogenous EPSCs , UNC-13L produced better rescue than UNC-13S . To confirm that UNC-13 transgenes rescued synaptic function by restoring SV priming , we measured the EPSC evoked by treatment with hypertonic sucrose ( Figure 2—figure supplement 1 ) . Hypertonic sucrose evokes release of docked and primed SVs in a calcium-independent manner ( Rosenmund and Stevens , 1996 ) . The sucrose-evoked charge observed in UNC-13L- and UNC-13S-rescued animals was significantly higher than in unc-13 ( s69 ) mutants , indicating that both isoforms efficiently promote SV priming . The sucrose-evoked charge observed in strains containing UNC-13S transgenes was significantly larger than those observed in wild-type controls , consistent with our prior studies ( Madison et al . , 2005 ) presumably because the primed pool of SVs is limited by UNC-13S expression levels . To assess changes in the kinetics of ACh release , we recorded stimulus-evoked EPSCs ( Figure 3A ) . To estimate the amount of ACh released during an evoked response , we calculated quantal content , which corresponds to the ratio of the total synaptic charge in an evoked EPSC to that in an endogenous EPSC ( Figure 3B ) . Both transgenes partially rescued the unc-13 ( s69 ) quantal content defect ( UNC-13L rescued 34% WT , UNC-13S rescued 86% WT , and unc-13 mutant 0 . 5% WT ) . 10 . 7554/eLife . 00967 . 006Figure 3 . UNC-13L and UNC-13S mediate fast and slow release , respectively . Stimulus-evoked EPSCs were recorded from adults body wall muscles of the indicated genotypes . Averaged traces ( A ) , quantal content ( B ) , latency ( 0–20% rise time ) , activation ( 10–90% rise time ) , decay kinetics ( C ) , and the cumulative charge transfer ( D ) are shown . Quantal content was calculated as the ratio of the charge transfer occurring during an evoked response to that occurring during an endogenous EPSC . Values that differ significantly from wild-type controls are indicated ( ***p<0 . 001 , *p<0 . 05 ) . The number of animals analyzed is indicated for each genotype . Error bars indicate SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 00967 . 00610 . 7554/eLife . 00967 . 007Figure 3—figure supplement 1 . Controls for experiments shown in Figure 3 . ( A–C ) Rescue of unc-13 ( s69 ) by UNC-13L constructs containing a c-terminal mCherry tag . Stimulus-evoked EPSCs were recorded from body wall muscles of wild-type adults of the indicated genotypes . Averaged traces ( A ) , cumulative charge transfer ( B ) , and activation kinetics ( C ) are shown . mCherry-tagged UNC-13L-mediated evoked responses that have faster activation kinetics than wild-type controls . ( D–F ) Rescue of unc-13 ( s69 ) by coexpression of UNC-13L and UNC-13S constructs . Stimulus-evoked EPSCs were recorded from body wall muscles of wild-type adults of the indicated genotypes . Averaged traces ( D ) , cumulative charge transfer ( E ) , quantal content , activation , and decay kinetics ( F ) are shown . In strains coexpressing UNC-13S and UNC-13L , quantal content was slightly larger than in wild-type controls , whereas activation and decay kinetics were not significantly altered ( G and H ) . Changes in the intrinsic kinetics of muscle ACh responses cannot account for changes in evoked EPSC kinetics . Endogenous EPSCs were recorded from the indicated genotypes . Averaged traces ( G ) and decay kinetics ( H ) are shown ( I and J ) . Changes in the composition of postsynaptic ACh receptors cannot account for the change in evoked EPSC kinetics . Evoked EPSCs were recorded from the indicated genotypes . Averaged responses ( I , left ) , normalized responses ( I , right ) , and activation and decay kinetics ( J ) are shown . Mutants lacking homopentameric ACR-16 receptors and those lacking heteropentameric levamisole receptors ( unc-29 mutants ) were analyzed . Values that significantly differ from wild-type controls are indicated ( ***p<0 . 001 , *p<0 . 05 ) . The number of animals analyzed is indicated for each genotype . Error bars indicate SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 00967 . 007 Evoked EPSC kinetics in UNC-13L- and UNC-13S-rescued animals were significantly different ( Figure 3C , D ) . In UNC-13L-rescued animals , evoked EPSCs exhibited a shorter lag between the stimulus and release ( i . e . , 0–20% rise time ) as well as faster activation and decay kinetics than in wild-type controls , which resulted in a faster rate of charge transfer during evoked responses . Similarly , evoked responses in animals rescued with mCherry-tagged UNC-13L also exhibited faster activation and charge transfer than in wild-type controls , indicating that fusion to mCherry did not disrupt UNC-13L function ( Figure 3—figure supplement 1A–C ) . By contrast , UNC-13S-evoked EPSCs activated more slowly and exhibited slower charge transfer . Coexpression of both UNC-13 isoforms in unc-13 ( s69 ) produced evoked responses with kinetics similar to wild-type controls ( Figure 3—figure supplement 1D–F ) . These results suggest that UNC-13L and UNC-13S mediate fast and slow ACh release , respectively . The altered EPSC kinetics observed in UNC-13 rescue strains could result from changes in the kinetics of muscle responses to ACh . This possibility seems unlikely because the UNC-13 transgenes are not expressed in muscles . Nevertheless , we did two experiments to test this possibility ( Figure 3—figure supplement 1G–J ) . First , the kinetics of endogenous EPSCs were unaltered in UNC-13L and UNC-13S rescue , indicating that the intrinsic kinetics of muscle ACh responses were not affected ( Figure 3—figure supplement 1G , H ) . Second , we analyzed the contribution of postsynaptic receptor composition to evoked EPSC kinetics ( Figure 3—figure supplement 1I , J ) . Body muscles express two classes of nicotinic ACh receptors , homomeric ACR-16 receptors and heteropentameric levamisole receptors ( LevRs ) , which have distinct kinetic properties . ACR-16 receptors exhibit fast desensitization , whereas LevRs desensitize slowly ( Francis et al . , 2005; Touroutine et al . , 2005 ) . Evoked EPSC decay kinetics was slower in acr-16 mutants and was unaltered in unc-29 Lev mutants . Evoked EPSC activation kinetics were unaltered in both acr-16 and unc-29 Lev mutants ( Figure 3—figure supplement 1J ) . Thus , changes in postsynaptic receptor composition cannot account for the altered evoked EPSC kinetics observed in UNC-13L- and UNC-13S-rescued animals . Instead , these results suggest that the altered evoked EPSC kinetics observed in UNC-13 transgenic animals are caused by changes in the kinetics of ACh release . The lag ( or coupling ) between calcium entry and exocytosis is determined by how rapidly calcium reaches primed SVs and binds to their calcium sensors . Tighter coupling is expected when SVs are physically closer to calcium channels or when they have higher intrinsic calcium sensitivity . Thus , differences in evoked EPSCs kinetics could be mediated by differential coupling of primed SVs to the sites of calcium entry . To test this idea , we analyzed the effects of the calcium chelator EGTA on evoked responses . Because EGTA binds calcium slowly , it inhibits SV fusions that occur later in time , that is the SVs that are more loosely coupled with calcium entry ( Atluri and Regehr , 1996 ) . To determine if fast and slow ACh release at the C . elegans NMJ can be distinguished by their calcium coupling , we analyzed the effect of EGTA on evoked EPSCs ( Figure 4 ) . We recorded evoked release after adding a membrane-permeant form of EGTA to the extracellular recording solution ( 40 μM EGTA-AM ) . EGTA-AM is loaded into both motor neurons and body muscles . In our experiments , body muscle EPSCs are recorded using patch pipettes containing 5 mM EGTA . Consequently , any effect of EGTA-AM on synaptic transmission must be caused by changes in presynaptic calcium . In wild-type animals , EGTA treatment significantly reduced the quantal content ( 56% decrease , p<0 . 001 ) of evoked EPSCs ( Figure 4 ) . EGTA treatment accelerated both the activation and decay of evoked EPSCs ( Figure 4—figure supplement 1A , B ) , resulting in accelerated charge transfer during evoked responses ( Figure 4—figure supplement 1C ) . EGTA had no effect on the latency of evoked release ( Figure 4—figure supplement 1C ) . Thus , in wild-type controls , evoked release comprises two SV pools: a fast pool tightly coupled to calcium entry and a slow loosely coupled pool . 10 . 7554/eLife . 00967 . 008Figure 4 . EGTA inhibits UNC-13S-mediated secretion but not that of UNC-13L . The effect of EGTA-AM on evoked responses was analyzed for the indicated genotypes . Averaged traces and quantal content are shown in control saline and after EGTA-AM treatment . Values that differ significantly are indicated ( ***p<0 . 001; n . s . , not significant ) . The number of animals analyzed is indicated for each genotype . Error bars indicate SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 00967 . 00810 . 7554/eLife . 00967 . 009Figure 4—figure supplement 1 . Changes in calcium influx cannot account for the altered evoked EPSC kinetics . Stimulus-evoked EPSCs were recorded from wild-type body wall muscles in control saline ( 1 . 0 mM CaCl2 ) , control saline with 40 μM EGTA-AM , and in saline with reduced extracellular calcium ( 0 . 5 mM CaCl2 ) . Averaged traces ( A ) , and summary data ( B and C ) are shown . ( B ) Amplitude , latency , activation , and decay kinetics are shown for the evoked responses in all three recording conditions . ( C ) Cumulative transfer of charge during evoked responses is plotted for the three recording conditions . Values that differ significantly are indicated ( ***p<0 . 001; **p<0 . 01; *p<0 . 05; n . s . , not significant ) . The number of animals analyzed is indicated for each genotype . Error bars indicate SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 00967 . 009 The effects of EGTA on EPSC kinetics could be caused by decreased presynaptic calcium levels or by faster decay of the presynaptic calcium transient . To distinguish between these possibilities , we analyzed the effect of external calcium concentration on evoked release ( Figure 4—figure supplement 1 ) . In reduced external calcium ( 0 . 5 mM CaCl2 ) , evoked EPSC amplitudes were very similar to those observed in 1 mM CaCl2 with EGTA ( Figure 4—figure supplement 1A , B ) . However , unlike EGTA treatment , reducing external calcium had no effect on EPSC activation and decay kinetics ( Figure 4—figure supplement 1B , C ) . Thus , the effects of EGTA on EPSC kinetics are unlikely to be caused by a uniform decrease in presynaptic calcium and are more likely a consequence of accelerated decay of the presynaptic calcium transient , as previously proposed ( DiGregorio et al . , 1999 ) . Differences in the evoked EPSC kinetics observed in UNC-13L- and UNC-13S-rescued animals could be mediated by altered coupling of primed SVs to calcium entry . Consistent with this idea , the effect of EGTA on the quantal content of evoked EPSCs was virtually eliminated in UNC-13L-rescued animals ( Figure 4 ) , indicating that SVs primed by UNC-13L bind calcium more rapidly than in wild-type controls . By contrast , in UNC-13S-rescued animals , the EGTA-resistant component of evoked responses was dramatically reduced ( Figure 4 ) , indicating that UNC-13S primed SVs bind calcium more slowly than in wild-type controls . Collectively , these results suggest that different UNC-13 proteins mediate fast and slow release . UNC-13L is concentrated at active zones , promotes fusion of SVs that are tightly coupled to calcium entry , thereby mediating fast ACh release . UNC-13S is diffusely localized in axons , promotes fusion of SVs that are loosely coupled to calcium entry , thereby mediating slow ACh release . The preceding experiments suggest that UNC-13L and UNC-13S ( when transgenically expressed ) can function independently to promote fast and slow release . Because transgenes are typically expressed at higher levels than endogenous genes , we wanted to determine if endogenously expressed UNC-13 isoforms also function independently . To address this question , we analyzed unc-13 mutants containing mutations that selectively inactivate UNC-13L . The e51 and e1091 alleles correspond to nonsense mutations in exons 11 and 12 ( respectively ) of unc-13 ( Kohn et al . , 2000 ) . These mutations truncate UNC-13L prior to the R domain ( encoded by exons 15–31 ) ( Figure 1A ) . The e51 and e1091 mutations are not included in the unc-13S mRNA ( encoded by exons 14–31 ) ( Kohn et al . , 2000 ) . Thus , e51 and e1091 both inactivate UNC-13L while leaving UNC-13S intact . Hereafter , we refer to e51 and e1091 as unc-13 [L−S+] mutations . If endogenous UNC-13S functions independently of UNC-13L , synaptic transmission in unc-13 [L−S+] mutants should be similar to that observed in UNC-13S-rescued animals . Contrary to this idea , the evoked EPSC ( Figure 5A , B ) and endogenous EPSC ( Figure 5—figure supplement 1A , B ) defects observed in unc-13 [L−S+] mutants were dramatically stronger than in UNC-13S rescue , but were slightly less severe than in unc-13 ( s69 ) mutants , consistent with prior studies ( Richmond et al . , 1999 ) . Thus , unlike UNC-13S-rescued animals , mutants lacking endogenous UNC-13L had severe synaptic defects . To confirm that the transmission defect observed in unc-13 [L−S+] was caused by decreased SV priming , we analyzed the sucrose-evoked EPSC . The sucrose-evoked charge was also dramatically reduced in unc-13 ( e1091 ) [L−S+] ( 22% WT ) but was slightly larger than in unc-13 ( s69 ) mutants ( 10% WT , p<0 . 05 ) ( Figure 5—figure supplement 1C , D ) ( McEwen et al . , 2006 ) . Thus , inactivating UNC-13L decreased the priming activities of both UNC-13S and UNC-13L . 10 . 7554/eLife . 00967 . 010Figure 5 . UNC-13L is required for the priming activity of endogenous UNC-13S . Stimulus-evoked EPSCs ( A and B ) were recorded from adult body wall muscles of the indicated genotypes . Averaged responses ( A ) , and summary data ( B ) for quantal content , activation and decay kinetics are shown . ( C and D ) Expression of UNC-13L in unc-13 ( e1091 ) [L−S+] mutants restores both EGTA sensitive and resistant forms of release . By contrast , UNC-13L expression in unc-13 ( s69 ) reconstitutes only the EGTA-resistant component of release . Evoked EPSCs were recorded in control saline and after EGTA-AM treatment . Averaged responses ( C ) and quantal content ( D ) are shown . Values that differ significantly are indicated ( ***p<0 . 001; *p<0 . 05; n . s . , not significant ) . The number of animals analyzed is indicated for each genotype . Error bars indicate SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 00967 . 01010 . 7554/eLife . 00967 . 011Figure 5—figure supplement 1 . UNC-13L is required for the priming activity of endogenously expressed UNC-13S . Endogenous EPSCs ( A and B ) and sucrose-evoked EPSCs ( C and D ) were recorded from the indicated genotypes . Representative traces ( A and C ) and summary data ( B and D ) are shown . For sucrose responses , the bar above each trace indicates the 1 s application of sucrose . The total synaptic charge transfer was measured for 1 s after initiation of the sucrose stimulus for each genotype . Values that differ significantly ( ***p<0 . 001 , *p<0 . 05 ) . The number of animals analyzed is indicated for each genotype . Error bars indicate SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 00967 . 011 The unc-13 [L−S+]-evoked EPSC defect was fully rescued by a transgene expressing UNC-13L ( Figure 5A , B ) , indicating that this defect was caused by inactivation of UNC-13L . The quantal content observed in UNC-13L rescued unc-13 [L−S+] mutants was indistinguishable from that in wild-type controls and was significantly larger than in UNC-13L-rescued unc-13 ( s69 ) mutants ( Figure 5B ) . The activation and decay kinetics of evoked EPSCs in UNC-13L-rescued unc-13 ( e1091 ) [L−S+] mutants were indistinguishable from wild-type controls ( Figure 5A , B ) . EGTA treatment decreased the amplitude and quantal content of evoked EPSCs in UNC-13L-rescued unc-13 ( e1091 ) [L−S+] mutants ( Figure 5C , D ) . Thus , restoring UNC-13L expression in unc-13 [L−S+] mutants reconstituted both fast and slow release . By contrast , expressing UNC-13L in unc-13 ( s69 ) mutants reconstituted only fast release . These results suggest that endogenous UNC-13S requires UNC-13L to perform its function . The discrepancy between the UNC-13S rescue and unc-13 [L−S+] mutant phenotypes could result from a failure to express UNC-13S in unc-13 [L−S+] mutants . To test this idea , we measured the abundance of UNC-13S mRNA by quantitative PCR . We found that UNC-13S mRNA levels were modestly increased in unc-13 [L−S+] mutants compared to wild-type controls ( 38% increase , p<0 . 01 ) . Consequently , decreased expression of the UNC-13S promoter is unlikely to explain the synaptic transmission defects observed in mutants lacking UNC-13L . Another potential model to explain the unc-13 [L−S+] synaptic defect is that UNC-13S function is actively inhibited by an endogenous regulatory mechanism . In this scenario , overexpressing UNC-13S could titrate out the inhibitor ( through direct binding or indirectly via a shared binding partner ) , thereby restoring synaptic function . Several studies suggest that the Tomosyn protein could provide this inhibitory function . In mutants lacking Tomosyn , neurotransmitter release is increased ( Gracheva et al . , 2006; McEwen et al . , 2006 ) . In electron micrographs , anti-TOM-1 antibodies labeled a presynaptic domain that is distal to dense projections , and little staining was observed adjacent to dense projections ( Gracheva et al . , 2007 ) . In tom-1 Tomosyn mutants , docking of distal SVs was significantly increased ( Gracheva et al . , 2006 ) . Prompted by these results , we tested the idea that TOM-1 specifically inhibits slow ACh release during evoked responses . Analysis of the kinetics of evoked responses suggests that ACh was released more slowly in tom-1 mutants . In particular , evoked EPSC decay was significantly slower in tom-1 mutants ( Figure 6 ) . The altered kinetics of tom-1-evoked responses are unlikely to result from a change in the intrinsic kinetics of muscle ACh responses because endogenous EPSC kinetics in tom-1 mutants were indistinguishable from those in wild-type controls ( Figure 3—figure supplement 1G , H ) . The increased evoked EPSC amplitude and quantal content observed in tom-1 mutants were both eliminated by EGTA treatment , indicating that the increased ACh release was mediated by fusion of SVs that are loosely coupled to calcium entry ( Figure 6 ) . Collectively , these results suggest that evoked ACh release was slower and more prolonged in tom-1 mutants , and that this kinetic change was mediated by increased exocytosis of SVs that are docked distal to the site of calcium entry . Thus , TOM-1 potently inhibits slow ACh release and has more limited effects on fast release . 10 . 7554/eLife . 00967 . 012Figure 6 . Tomosyn inhibits slow release . Stimulus-evoked EPSCs were recorded from body wall muscles in control saline and after the addition of EGTA-AM . Averaged traces ( A ) and summary data ( B ) for quantal content , activation , and decay kinetics are shown . Values that differ significantly are indicated ( ***p<0 . 001; n . s . , not significant ) . The number of animals analyzed is indicated for each genotype . Error bars indicate SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 00967 . 012 To determine if TOM-1 inhibits UNC-13S function , we analyzed tom-1 unc-13 double mutants ( Figure 7 ) . As previously reported , the quantal content of evoked EPSCs ( Figure 7B ) and the total sucrose-evoked charge ( Figure 7—figure supplement 1A , B ) were significantly increased in tom-1 single mutants , indicating increased SV priming ( Gracheva et al . , 2006; McEwen et al . , 2006 ) . Evoked ACh release was nearly completely eliminated in tom-1 unc-13 ( s69 ) double mutants ( Figure 7A , B ) , implying that UNC-13 priming activity is required for TOM-1’s effects on ACh release . By contrast , evoked ACh release in tom-1 unc-13 [L−S+] double mutants was significantly larger than in unc-13 [L−S+] single mutants ( Figure 7A , B ) . Increased ACh release in tom-1 unc-13 [L−S+] double mutants was accompanied by a corresponding increase in the sucrose-evoked charge ( Figure 7—figure supplement 1A , B ) , implying that TOM-1 inhibits UNC-13S priming activity . Consistent with improved synaptic function , the locomotion rate of tom-1 unc-13 [L−S+] double mutants was significantly higher than in unc-13 [L−S+] single mutants ( Figure 7C , D ) . Thus , electrophysiological recordings and behavioral studies indicate that synaptic function in tom-1 unc-13 [L−S+] double mutants was significantly greater than in tom-1 unc-13 ( s69 ) double mutants . The only genetic difference between these strains is that tom-1 unc-13 [L−S+] double mutants retain endogenous UNC-13S , while tom-1 unc-13 ( s69 ) double mutants lack UNC-13S . Consequently , these results suggest that Tomosyn inhibits the priming activity of endogenously expressed UNC-13S . 10 . 7554/eLife . 00967 . 013Figure 7 . Tomosyn inhibits UNC-13S-mediated secretion . Inactivating TOM-1 Tomosyn in unc-13 [L−S+] ( e1091 ) mutants restored ACh release ( A and B ) and locomotion behavior ( C and D ) . ( A and B ) Stimulus-evoked EPSCs were analyzed for the indicated genotypes . Averaged traces ( A ) and quantal content ( B ) are shown . ( C and D ) Locomotion behavior was analyzed in the indicated strains . Representative 20 s locomotion trajectories are shown ( n = 10 animals for each genotype ) ( C ) . The starting points for each trajectory were aligned for clarity . ( D ) Locomotion rates are compared for the indicated genotypes . Values that differ significantly are indicated ( ***p<0 . 001; **p<0 . 01; n . s . , not significant ) . The number of animals analyzed is indicated for each genotype . Error bars indicate SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 00967 . 01310 . 7554/eLife . 00967 . 014Figure 7—figure supplement 1 . TOM-1 inhibits the priming activity of UNC-13S . Sucrose-evoked EPSCs ( A and B ) and stimulus-evoked EPSCs ( C and D ) were recorded from the indicated genotypes . Representative traces of sucrose responses ( A ) , averaged traces of evoked responses ( C ) , and summary data ( B and D ) are shown . Values that differ significantly are indicated ( ***p<0 . 001; **p<0 . 01 ) . The number of animals analyzed is indicated for each genotype . Error bars indicate SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 00967 . 01410 . 7554/eLife . 00967 . 015Figure 7—figure supplement 2 . The latency of evoked responses is prolonged and calcium coupling is loosened in unc-13 ( e1091 ) [L−S+] and in unc-10 RIM mutants . Stimulus-evoked EPSCs were recorded from body wall muscles in control saline and after the addition of EGTA-AM . Averaged traces ( A and C ) and summary data for activation kinetics and latency of release ( B ) , quantal content ( D ) , and EGTA resistance ( E ) . Values that differ significantly are indicated ( ***p<0 . 001; *p<0 . 05 ) . The number of animals analyzed is indicated for each genotype . Error bars indicate SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 00967 . 015 To determine if Tomosyn also inhibits UNC-13L , we analyzed synaptic transmission in UNC-13S- and UNC-13L-rescued animals following inactivation of Tomosyn ( Figure 7—figure supplement 1C , D ) . Inactivating Tomosyn in UNC-13S animals significantly increased the quantal content ( 263% increase , p<0 . 001 ) . By contrast , inactivating Tomosyn in UNC-13L-rescued animals had more modest effects on the quantal content ( 96% increase , p<0 . 001 ) . These results suggest that Tomosyn potently inhibits UNC-13S-mediated secretion and has weaker effects on UNC-13L . As previously reported ( Gracheva et al . , 2006; McEwen et al . , 2006 ) , ACh release and SV priming were slightly higher in tom-1 unc-13 ( s69 ) double mutants than in unc-13 ( s69 ) single mutants ( Figure 7—figure supplement 1 ) . These results suggest that Tomosyn also inhibits an UNC-13-independent form of release . Thus far , our results suggest that fast release is mediated by UNC-13L , which is concentrated at dense projections , and by fusion of SVs that are tightly coupled to calcium channels . These results suggest that mutants lacking UNC-13L , or its binding partner UNC-10/RIM , should have slower release . To test this idea , we compared release kinetics in unc-13 ( e1091 ) [L−S+] and unc-10 RIM mutants . Because both mutations dramatically reduce release , we analyzed double mutants lacking TOM-1/Tomosyn to increase the sensitivity of our experiments . The activation kinetics of evoked EPSCs in tom-1 unc-13 ( e1091 ) and in tom-1; unc-10 double mutants were significantly slower than those in tom-1 single mutants ( Figure 7—figure supplement 2A , B ) . In particular , the lag between the stimulus and activation of evoked release was prolonged in both double mutants ( Figure 7—figure supplement 2B ) . Evoked responses in tom-1 unc-13 ( e1091 ) and in tom-1; unc-10 double mutants were also significantly less EGTA-resistant than those in tom-1 single mutants ( Figure 7—figure supplement 2C , E ) . Collectively , these results suggest that UNC-13L and UNC-10/RIM are required for fast release . The role of RIM in promoting fast release is likely conserved , as mouse and fly RIM knockouts exhibit similar changes in release kinetics and calcium coupling ( Han et al . , 2011; Müller et al . , 2012 ) . Mutants lacking RIM also have decreased calcium channel density and decreased calcium entry at active zones ( Han et al . , 2011; Kaeser et al . , 2011; Müller et al . , 2012 ) . Therefore , unc-10 RIM mutations may alter the spatial and temporal extent of the calcium transient , which would contribute to changes in release kinetics . The two UNC-13 isoforms differ only in their NTDs ( Figure 1A ) . To test the functional importance of the NTDs , we expressed a truncated UNC-13 protein containing only the common region ( UNC-13R ) ( Figure 8—figure supplement 1 ) . UNC-13R had a diffuse distribution in axons ( Figure 8A–C ) , and produced evoked responses with activation kinetics and EGTA sensitivity indistinguishable from those in UNC-13S rescue ( Figure 8D–H ) . Thus , sequences in the UNC-13S NTD were not required for the spatial location , looser calcium coupling , and slower release kinetics of UNC-13S . 10 . 7554/eLife . 00967 . 016Figure 8 . The C2A domain anchors UNC-13 at active zones and shortens the latency of release . Fusing the C2A domain to the R region ( C2AR ) increased recruitment of UNC-13 to active zones , shortened the latency of release , but did not tighten the coupling of release to calcium entry . Deleting the C2A domain in full length UNC-13L ( ΔC2A ) modestly decreased targeting to active zones , did not lengthen the latency of release , and did not loosen the coupling of exocytosis and calcium entry . ( A and C ) The distribution of mCherry-tagged UNC-13R , C2AR , and ΔC2A at active zones ( labeled with GFP-tagged UNC-10/RIM ) are shown . All tagged proteins were expressed in DA and DB motor neurons of wild type animals ( using the unc-129 promoter ) . Representative images ( A ) , line scans ( B ) , and correlation coefficients ( C ) for UNC-10/RIM and UNC-13 fluorescence at synaptic puncta are shown . Scale bar indicates 1 μm . ( D–H ) Stimulus-evoked EPSCs were recorded from body wall muscles in control saline and after the addition of EGTA-AM . Averaged traces ( D ) and summary data for quantal content ( E ) , EGTA resistance ( F ) , time course of initial evoked charge transfer ( 0–5 ms , G ) , and the latency of release ( 0–20% rise time , H ) are shown . Values that differ significantly are indicated ( ***p<0 . 001; **p<0 . 01; *p<0 . 05; n . s . , not significant ) . The number of animals analyzed is indicated for each genotype . Error bars indicate SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 00967 . 01610 . 7554/eLife . 00967 . 017Figure 8—figure supplement 1 . Chimeric UNC-13 proteins utilized in Figures 8 and 9 . The domain structure of UNC-13L , UNC-13S , ΔC2A , R , CaMR , and C2AR are illustrated . Ligands for each domain are indicated . Amino acid coordinates indicated refer to the UNC-13L cDNA sequence ( ZK524 . 2e ) . Each isoform has a unique NTD . The L-NTD ( 610 aa ) is encoded by exons 1–13 and contains a C2A domain ( aa 1–96 ) and a predicted calmodulin binding site ( green box , aa 556–610 ) . The S-NTD ( 259 aa ) is encoded by exon 14 . The ΔC2A construct contains an N-terminal truncation of UNC-13L , lacking amino acids 1–96 . The R construct contains the R domain ( aa 611–1816 , comprising C2A , C1 , C2B , MUN , C2C domains ) . The CaMR construct contains a predicted CaM binding site in the L-NTD ( aa 556–610 ) fused to the R domain ( aa 611–1816 ) . The C2AR construct contains the C2A domain ( aa 1–96 ) fused to the R domain ( aa 611–1816 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 00967 . 017 Faster kinetics and tighter calcium coupling could be caused by the closer spatial relationship of UNC-13L to the dense projection ( Figure 1B–D ) ( Weimer et al . , 2006 ) . To test this idea , we analyzed a chimeric protein containing the C2A domain ( which binds RIM ) fused to the R domain ( C2AR ) ( Figure 8—figure supplement 1 ) . As expected , mCherry-tagged C2AR was more punctate , and exhibited greater colocalization with UNC-10/RIM than did mCherry-tagged UNC-13R ( Figure 8A–C ) . Thus , the C2A domain increased UNC-13 recruitment to active zones . Evoked EPSCs in C2AR had slightly shorter latencies than those in UNC-13R; however , the EGTA resistance of evoked responses in C2AR and UNC-13R were similar ( Figure 8D–H ) . Thus , fusing the C2A and R domains recapitulated the spatial localization of UNC-13L but had no effect on calcium coupling and only a modest effect on release kinetics . These results suggest that differences in the spatial localization of UNC-13S and L cannot fully account for differences in the kinetics of neurotransmitter release mediated by these proteins . To further test C2A’s role in release kinetics , we analyzed an UNC-13L deletion mutant lacking the C2A domain ( ΔC2A ) ( Figure 8—figure supplement 1 ) . UNC-10/RIM was significantly more colocalized with ΔC2A than with UNC-13R ( Figure 8A–C ) . ΔC2A evoked EPSCs had shorter latencies , faster activation kinetics , and greater EGTA resistance than those in UNC-13R-rescued animals ( Figure 8D–H ) . Thus , the C2A domain was not required for UNC-13 anchoring at active zones , for tight coupling of exocytosis to calcium entry , nor for rapid release kinetics . Our analysis of ΔC2A suggests that in addition to the C2A domain , other sequences in the UNC-13L NTD contribute to the faster release kinetics . The NTDs of Munc13 proteins have amphipathic helixes that bind calmodulin ( Junge et al . , 2004; Lipstein et al . , 2012 ) . Calmodulin binds to Munc13 in an extended conformation , where the calmodulin C-terminal lobe is anchored to hydrophobic residues at positions 1 , 5 , and 8/10 of the Munc13 amphipathic helix , while the N-terminal lobe is anchored at position 26 of the helix ( Rodriguez-Castaneda et al . , 2010; Lipstein et al . , 2012 ) . A similar amphipathic helix is present at the same location in the UNC-13L NTD ( Figure 9A ) . 10 . 7554/eLife . 00967 . 018Figure 9 . A predicted calmodulin-binding site in UNC-13L accelerates release . The predicted calmodulin ( CaM ) -binding site of UNC-13L ( residues 556–610 ) was fused to the R domain . CaMR had a diffuse axonal distribution , was poorly colocalized with UNC-10/RIM , but exhibited faster and more EGTA-resistant evoked ACh release . The effects of the CaM-binding site on release were eliminated by the F574R mutation , which is predicted to disrupt calmodulin binding ( Junge et al . , 2004 ) . ( A ) Position of the putative CaM-binding site in UNC-13L and its alignment with the rat Munc13-1 CaM-binding site sequence are shown . Hydrophobic residues anchoring the C-terminal ( green ) and N-terminal ( red ) lobes of CaM , and predicted apo-calmodulin binding sites ( blue ) are indicated ( Rodriguez-Castaneda et al . , 2010; Lipstein et al . , 2012 ) . ( B–D ) The distribution of mCherry-tagged CaMR at active zones ( labeled with GFP-tagged UNC-10/RIM ) are shown . Tagged proteins were expressed in DA and DB motor neurons of wild-type animals ( using the unc-129 promoter ) . Representative images ( B ) , line scans ( C ) , and correlation coefficients ( D ) for UNC-10/RIM and CaMR fluorescence at synaptic puncta are shown . Scale bar indicates 1 μm . ( E and H ) Stimulus-evoked EPSCs were recorded from body wall muscles in control saline and after the addition of EGTA-AM . Averaged traces ( E ) and summary data for time course of initial evoked charge transfer ( 0–5 ms , F ) , quantal content ( G ) , EGTA resistance ( H ) , and the latency of release ( 0–20% rise time , I ) are shown . Values that differ significantly are indicated ( **p<0 . 01; *p<0 . 05; n . s . , not significant ) . The number of animals analyzed is indicated for each genotype . Error bars indicate SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 00967 . 018 To test its functional importance , we constructed a chimeric UNC-13 protein containing the predicted calmodulin-binding site ( residues 556–610 ) fused to the R domain ( CaMR ) ( Figure 8—figure supplement 1 ) . An mCherry-tagged CaMR had a diffuse distribution in axons and was not colocalized with UNC-10/RIM , similar to the distribution of UNC-13R ( Figure 9B–D ) . Thus , the CaM-binding site had little effect on UNC-13’s spatial distribution . CaMR-mediated evoked responses had significantly faster activation kinetics and were significantly more EGTA resistant than those in UNC-13R-rescued animals ( Figure 9E–I ) . Both these effects were eliminated by a mutation predicted to disrupt calmodulin binding , CaMR[F574R] ( Junge et al . , 2004 ) . In particular , CaMR-evoked responses exhibited faster charge transfer , shorter latency ( i . e . , 0–20% rise time ) , and greater EGTA resistance than those in CaMR[F574R] ( Figure 9F–I ) . Collectively , these results suggest that the UNC-13L NTD has two effects on release . The C2A domain anchors UNC-13 at active zones , while the calmodulin binding site domain accelerates release kinetics and tightens calcium coupling .
Prior studies showed that TOM-1 inhibits SV priming , and that it does so by binding to plasma membrane SNAREs , forming inactive SNARE complexes ( Gracheva et al . , 2006; McEwen et al . , 2006 ) . Here we show that Tomosyn’s effect on synaptic transmission is mediated primarily by inhibiting UNC-13S-mediated release . The increased evoked ACh release in tom-1 mutants was eliminated by EGTA treatment , indicating that the extra release was mediated by fusion of SVs that are loosely coupled to calcium entry . In transgenic animals , TOM-1 inhibited UNC-13S-mediated slow release far more potently than UNC-13L-mediated fast release . The apparent selectivity for inhibiting slow release could result from the distal localization of the TOM-1 protein , as immuno-EM studies indicate that TOM-1 staining peaks ∼300 nm from the dense projection ( Gracheva et al . , 2007 ) . We previously reported that the synaptic abundance of UNC-13S is increased in tom-1 mutants , perhaps because the rate of distal SV fusions had increased ( McEwen et al . , 2006 ) . Taken together , these results strongly support the idea that Tomosyn inhibits priming and fusion of distal SVs . Several results suggest that fast and slow release were mediated by fusion of SVs that are docked to distinct spatial domains of the nerve terminal . Immuno-EM studies indicate that proteins mediating fast release ( i . e . , UNC-10/RIM and UNC-13L ) are localized near dense projections ( Weimer et al . , 2006; Gracheva et al . , 2008 ) . Slow release is controlled by UNC-13S , which has a diffuse localization , and by TOM-1 for which immuno-EM studies indicate a distal location ( Gracheva et al . , 2007 ) . Analysis of SV docking by high-pressure freeze preservation and electron microscopy also suggests that fast and slow release are mediated by SV fusions in distinct spatial domains . Each NMJ has a single dense projection surrounded by 38 docked SVs extending radially ( Hammarlund et al . , 2007 ) . A mutation inactivating both UNC-13 isoforms decreased SV docking across the entire active zone ( i . e . , <350 nm from the dense projection ) ( Gracheva et al . , 2006; Weimer et al . , 2006; Hammarlund et al . , 2007 ) . By contrast , docking of proximal SVs was significantly reduced in unc-13 ( e1091 ) [L−S+] mutants ( <100 nm from dense projections ) and in unc-10 RIM mutants ( <50 nm from dense projections ) , whereas docking of distal SVs was unaffected in both mutants ( Weimer et al . , 2006; Hammarlund et al . , 2007; Gracheva et al . , 2008 ) . We found that unc-13 ( e1091 ) [L−S+] and unc-10 RIM mutations made evoked release slower and more loosely coupled to calcium entry , both indicating that SVs docked proximal to the dense projection are required for fast release . By contrast , inactivating TOM-1/Tomosyn increased docking of distal SVs ( Gracheva et al . , 2006 ) and made ACh release slower and more loosely coupled to calcium entry . Finally , the proportion of docked SVs that are <100 nm from the dense projection ( 15/38 , or ∼40% ) ( Hammarlund et al . , 2007 ) is similar to the fraction of evoked release that was EGTA resistant in our conditions . Thus , electrophysiological recordings , ultrastructural analysis of SV docking , immuno-EM localization of priming proteins ( UNC-10/RIM , UNC-13L , and TOM-1 ) , and sensitivity to EGTA inhibition all support a model whereby fast ACh release is mediated by the subpopulation of SVs that are docked adjacent to dense projections ( <100 nm ) , while slow release is mediated by fusion of distal SVs ( 100–350 nm from dense projections ) . Are differences in the spatial location of primed vesicles sufficient to explain the kinetics of fast and slow release ? Several results argue against this idea . The rise time of fast and slow release differ by ∼2 ms . If this difference was caused solely by a diffusional delay in calcium binding , slow SVs would have to be 3 μm farther from calcium channels than fast SVs ( assuming Dca = 200 µm2/s ) . Thus , it seems likely that exocytosis of UNC-13S primed SVs is intrinsically slower than that of UNC-13L primed vesicles . We show that UNC-13L accelerates release by two mechanisms . The C2A domain increased UNC-13 anchoring at active zones , and produced a slight decrease in the delay between stimulation and the onset of release . Both effects support the idea that the C2A domain caused primed SVs to become physically closer to calcium channels , thereby decreasing the time required for calcium to reach the vesicle’s calcium sensor . Other UNC-13L sequences ( beyond C2A ) also contribute to its physical location because the ΔC2A protein was strongly colocalized with UNC-10/RIM . The calmodulin-binding site in UNC-13L’s NTD also accelerated release . The analogous calmodulin-binding site in mouse Munc13-1 and Munc13-2 is required for augmentation , a form of short-term plasticity ( Junge et al . , 2004 ) . Based on these results , it was proposed that calmodulin binding to Munc13 comprises a calcium sensor for short-term plasticity . Our results suggest that this calmodulin-binding site also accelerates release . Thus , the faster kinetics of UNC-13L-mediated release results from both tighter spatial coupling to calcium channels and faster intrinsic kinetics of exocytosis . In addition to promoting fast release , UNC-13L also acts in conjunction with UNC-13S to promote slow release . Mutations inactivating UNC-13L dramatically reduced fast and slow release , and both forms of release were re-instated by restoring UNC-13L expression . Thus , UNC-13L promotes but is not required for slow release . The UNC-13L requirement for slow release was reduced ( but not eliminated ) when the total number of primed SVs was increased ( by inactivating Tomosyn or by overexpressing UNC-13S ) . UNC-13L could regulate slow release by a variety of mechanisms . UNC-13S mRNA levels were not decreased in unc-13 [L−S+] mutants , implying that UNC-13L did not alter expression of the UNC-13S promoter . UNC-13L could directly bind to UNC-13S thereby altering its activity or stability . UNC-13L could prevent inhibition of UNC-13S by Tomosyn . Finally , UNC-13L could increase the release probability of UNC-13S primed SVs . Further experiments will be required to determine how UNC-13L regulates slow release . Our results do not exclude the idea that other potential mechanisms also contribute to differences in UNC-13L and UNC-13S release kinetics . For example , UNC-13 and TOM-1 could regulate the kinetics or duration of calcium entry or could alter the spatial distribution of open calcium channels . Further experiments will be required to test these ideas . Our results suggest that fast and slow release are mediated by exocytosis of distinct subpopulations of SVs , and that distinct priming molecules mediate fast and slow SV fusions . In this manner , fast and slow release could be independently adjusted by modulatory pathways . It is intriguing that the fast and slow release components appear to function in parallel . We were able to construct mutants in which evoked release is mediated purely by either the fast or slow mechanism . Thus , both forms of release can occur in isolation . These results imply that fast SVs are not derived from maturation of slow SVs , and vice versa . Our results are more compatible with the idea that fast and slow release comprise independent exocytosis mechanisms that operate in parallel . Fast and slow release share a requirement for UNC-13L , which provides a potential mechanism for coordinately regulating changes in these two release pathways . Adjusting the contributions of fast and slow release will alter the duration of postsynaptic responses . This could have several important consequences for circuit function . Prolonged EPSCs will result in increased integration of synaptic inputs , which could alter the rate of postsynaptic spiking . Shifting the kinetics of postsynaptic responses could also alter spike timing-dependent plasticity . In sensory circuits , prolonged postsynaptic responses could alter the temporal integration of sensory inputs , and may alter the kinetics of sensory evoked responses . All these changes could have profound effects on circuit development , cognition , and behavior . Consistent with this idea , we recently showed that mutations linked to Autism alter the kinetics of neurotransmitter release ( Hu et al . , 2012 ) . For these reasons , there is significant interest in identifying molecular mechanisms that adjust the kinetics of neurotransmitter release . In summary , we propose that the kinetics of neurotransmitter release are dictated by a protein code consisting of fast ( UNC-10/RIM , UNC-13L ) and slow ( UNC-13L , UNC-13S , and Tomosyn ) priming factors , which mediate fusion of proximal and distal SVs , respectively . All these molecules are conserved across phylogeny , and orthologous proteins are found at both NMJs and central synapses . Thus , we propose that the basic mechanisms governing the kinetics of neurotransmitter release are ancient , and will apply to many synapses . Consistent with this idea , a recent study showed that Munc13 proteins that lack C2A domains ( bMunc13-2 and Munc13-3 ) are located more distally and mediate slow release at the Calyx of Held ( Chen et al . , 2013 ) .
Strain maintenance and genetic manipulation were performed as described ( Brenner , 1974 ) . Animals were cultivated at 20°C on agar nematode growth media seeded with OP50 bacteria . For electrophysiology , HB101 Escherichia coli was utilized . The following strains were utilized in this study:Wild type , N2 bristolKP6901 unc-13 ( s69 ) KP6902 unc-13 ( e1091 ) KP6903 unc-13 ( e51 ) KP3707 tom-1 ( nu468 ) KP3810 tom-1 ( nu468 ) unc-13 ( s69 ) KP6904 tom-1 ( nu468 ) unc-13 ( e1091 ) KP6905 tom-1 ( nu468 ) unc-13 ( e51 ) KP7162 tom-1 ( nu468 ) ;unc-10 ( md1117 ) KP5982 acr-16 ( ok789 ) KP5984 unc-29 ( x29 ) KP6893 nuEx1515 [Psnb-1::UNC-13L];unc-13 ( s69 ) KP6894 nuEx1516 [Psnb-1::UNC-13L];unc-13 ( e1091 ) KP6895 nuEx1517 [Psnb-1::UNC-13L];tom-1 ( nu468 ) unc-13 ( s69 ) KP6896 nuIs46 [Punc-13::UNC-13S::GFP];tom-1 ( nu468 ) unc-13 ( s69 ) KP6897 nuIs486 [Punc-129::UNC-13L::mCherry]KP6898 nuEx1518 [Pacr-2::UNC-10::mCherry]KP3928 nuIs165 [Punc-129::UNC-10::GFP]KP6899 nuIs46 [Punc-13::UNC-13S::GFP];unc-13 ( s69 ) KP7151 nuEx1590 [Psnb-1::UNC-13SL];unc-13 ( s69 ) KP7152 nuEx1591 [Psnb-1::UNC-13ΔC2A];unc-13 ( s69 ) KP7153 nuEx1592 [Psnb-1::UNC-13R];unc-13 ( s69 ) KP7154 nuEx1593 [Psnb-1::UNC-13C2AR];unc-13 ( s69 ) KP7155 nuEx1594 [Psnb-1::UNC-13CaMR];unc-13 ( s69 ) KP7156 nuEx1595 [Psnb-1::UNC-13caMR ( F574R ) ];unc-13 ( s69 ) KP7157 nuEx1596 [Punc-129::UNC-13ΔC2A::mCherry]KP7158 nuEx1597 [Punc-129::UNC-13C2AR::mCherry]KP7159 nuEx1598 [Punc-129::UNC-13CaMR::mCherry]KP7160 nuEx1599 [Punc-129::UNC-13S::mCherry]KP7161 nuIs490 [Punc-129::UNC-13R::mCherry] Transgenic strains were isolated by microinjection of various plasmids using either Pmyo-2::NLS-GFP ( KP#1106 ) or Pmyo-2::NLS-mCherry ( KP#1480 ) as coinjection markers . Integrated transgenes were obtained by UV irradiation of strains carrying extrachromosomal arrays . All integrated transgenes were out-crossed at least six times . UNC-13L and UNC-13S expression constructs contained full-length cDNAs , with the exceptions noted below . For UNC-13L , we utilized a full-length ZK524 . 2e cDNA ( spanning exons 1–31 , excluding exon 14 ) , whose expression as an SL1 spliced mRNA we confirmed by qPCR . The ZK524 . 2e cDNA corresponds to the LR mRNA previously described ( Kohn et al . , 2000 ) . For UNC-13S , we confirmed expression of an SL1 spliced mRNA containing exons 14–31 , which corresponds to the MR mRNA previously described ( Kohn et al . , 2000 ) . We could not detect expression of ZK524 . 2c , a WormBase predicted short UNC-13 mRNA , by qPCR; therefore , we utilized the full-length MR cDNA for UNC-13S imaging and an MR minigene ( nuIs46 ) ( Nurrish et al . , 1999 ) for electrophysiological recordings and behavior . Other UNC-13 constructs are as follows ( all residue coordinates refer to the ZK524 . 2e protein sequence ) : ΔC2A ( KP#1907 ) , ZK524 . 2e cDNA lacking the C2A domain ( residues 1–96 ) ; UNC-13R ( KP#1908 ) , cDNA containing exons 15–31 ( residues 611–1816 ) ; C2AR ( KP#1909 ) , cDNA containing the C2A domain ( residues 1–96 ) fused to the R domain; CaMR ( KP#1910 ) , and F574R ( KP#1911 ) , cDNAs containing the predicted CaM binding site ( residues 556–610 ) or a mutated binding site fused to the R domain . For electrophysiological recordings , transgenes were expressed by the downstream unc-13 promoter ( UNC-13S rescue ) or the snb-1 promoter ( all others ) . For imaging experiments , UNC-13 ( c-terminal mCherry-tagged ) and UNC-10 ( nuIs165 , c-terminal GFP tagged ) ( Ch’ng et al . , 2008 ) transgenes were expressed in DA and DB neurons , using the unc-129 promoter . Worm tracking and analysis were performed as previously described ( Dittman and Kaplan , 2008 ) with minor modifications . Briefly , worms were reared at 20°C and moved to room temperature 30 min before imaging . Young adult animals were picked to agar plates with no bacterial lawn ( 30 worms per plate ) . Locomotion was analyzed 10 min after the worms were removed from food . 30 s videos of individual animals were captured at 5× magnification and 4 Hz frame rate on a Zeiss Discovery Stereomicroscope using Axiovision software . The center of mass was recorded for each animal on each video frame using the object tracking software in the Axiovision software . The trajectories were then analyzed using custom software written in Igor Pro 5 . 0 ( Wavemetrics , Lake Oswego , OR ) . All quantitative imaging was done on a Zeiss Axioskop , using an Olympus PlanAPO 100 × 1 . 4 NA objective and a CoolSNAP HQ CCD camera ( Roper , Trenton , NJ ) . Worms were immobilized with 30 mg/ml BDM ( Sigma ) . Image stacks were captured and maximum intensity projections were obtained using Metamorph 7 . 1 software ( Molecular Devices , Sunnyvale , CA ) . GFP fluorescence was normalized to the absolute mean fluorescence of 0 . 5-mm FluoSphere beads ( Molecular Probes , Eugene , OR ) . For dorsal cord imaging , young adult worms , in which the dorsal cords were oriented toward the objective , were imaged in the region midway between the posterior gonad bend at the tail . Line scans of dorsal cord fluorescence were analyzed in Igor Pro ( WaveMetrics ) using custom-written software ( Dittman and Kaplan , 2006 ) . Electrophysiology was done on dissected adult C . elegans as previously described ( Richmond et al . , 1999 ) . Worms were superfused in an extracellular solution containing 127 mM NaCl , 5 mM KCl , 26 mM NaHCO3 , 1 . 25 mM NaH2PO4 , 20 mM glucose , 1 mM CaCl2 , and 4 mM MgCl2 , bubbled with 5% CO2 , 95% O2 at 20°C . Whole-cell recordings were carried out at −60 mV using an internal solution containing 105 mM CsCH3SO3 , 10 mM CsCl , 15 mM CsF , 4 mM MgCl2 , 5 mM EGTA , 0 . 25 mM CaCl2 , 10 mM HEPES , and 4 mM Na2ATP , adjusted to pH 7 . 2 using CsOH . Under these conditions , we only observed endogenous acetylcholine EPSCs . For endogenous GABA IPSC recordings , the holding potential was 0 mV , at which we only observe GABAergic postsynaptic currents . All recording conditions were as described ( Hu et al . , 2012 ) . Stimulus-evoked EPSCs were stimulated by placing a borosilicate pipette ( 5–10 μm ) near the ventral nerve cord ( one muscle distance from the recording pipette ) and applying a 0 . 4 ms , 30 μA square pulse using a stimulus current generator ( WPI ) . On average , this stimulus evoked ∼85% of the charge evoked by hypertonic sucrose , suggesting that a large fraction of the primed pool was released . Statistical significance was determined using a two-tailed Student’s t-test . | Neurons communicate with one another at junctions called synapses . When an electrical signal known as an action potential travels along a neuron and arrives at a synapse , the neuron releases a package of transmitter chemicals into the synapse . These chemicals then diffuse across the gap and bind to receptors on a second neuron , conveying the signal to the target neuron . The strength of a synapse depends in part on the number of packages , or vesicles , of transmitter chemicals that are available for release . Most synapses contain multiple populations of vesicles: some that are released within a few milliseconds of the arrival of an action potential , and others that are released more slowly . The vesicles that are released rapidly are found close to sites at which calcium ions enter the neuron , whereas the others are located further from these sites . However , little is known about the molecular basis of the differences between fast and slow vesicle release . Now Hu et al . have studied the proteins involved in these two processes in C . elegans , a nematode worm that is often used in neuroscience because it has a simple nervous system , consisting of just 302 neurons , and a well-characterized genome . Hu et al . showed that the release of synaptic vesicles at the neuromuscular junction between neurons and muscles in C . elegans also has slow and fast components . A long form of UNC-13 , which is also found in mammals , promotes fast release of transmitter vesicles . Slow release is mediated by an independent pathway that involves both long and short UNC-13 proteins , as well as a protein called Tomosyn . As in mammals , long UNC-13 is localized to the sites at which calcium ions enter neurons , whereas short UNC-13 is more widely distributed throughout neurons . The work of Hu et al . provides a molecular explanation for how the timing of transmitter release is determined . Because the UNC-13 and Tomosyn proteins are evolutionarily conserved , this mechanism is likely to be present in other animals too . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"cell",
"biology",
"neuroscience"
] | 2013 | UNC-13L, UNC-13S, and Tomosyn form a protein code for fast and slow neurotransmitter release in Caenorhabditis elegans |
Human organogenesis is when severe developmental abnormalities commonly originate . However , understanding this critical embryonic phase has relied upon inference from patient phenotypes and assumptions from in vitro stem cell models and non-human vertebrates . We report an integrated transcriptomic atlas of human organogenesis . By lineage-guided principal components analysis , we uncover novel relatedness of particular developmental genes across different organs and tissues and identified unique transcriptional codes which correctly predicted the cause of many congenital disorders . By inference , our model pinpoints co-enriched genes as new causes of developmental disorders such as cleft palate and congenital heart disease . The data revealed more than 6000 novel transcripts , over 90% of which fulfil criteria as long non-coding RNAs correlated with the protein-coding genome over megabase distances . Taken together , we have uncovered cryptic transcriptional programs used by the human embryo and established a new resource for the molecular understanding of human organogenesis and its associated disorders .
Embryogenesis encompasses the progression from fertilized zygote to blastocyst and through gastrulation to establish the three germ layers of ectoderm , mesoderm and endoderm , from which all organs and tissues subsequently arise during organogenesis . Remarkably little is known about this latter phase of assembling organs and tissues in human due to the restricted availability of human embryonic tissue and its tiny size . Previous transcriptomics post-implantation have sampled either the whole embryo by expression microarray ( Fang et al . , 2010 ) , thus lacking organ-specific resolution and the vast majority of long non-coding ( lnc ) transcription; or included lnc expression by massively parallel short-read RNA sequencing ( RNA-seq ) but focussed on single sites such as limb bud ( Cotney et al . , 2013 ) or pancreas ( Cebola et al . , 2015 ) . RNA-seq from NIH Roadmap and other studies during or after the end of the first trimester of pregnancy falls after the embryonic period ( which ends at 56–58 days post-conception ( Carnegie Stage 23 ) ) and commonly reflects near terminal differentiation within heterogeneous fetal organs and tissues ( Jaffe et al . , 2015; Roadmap Epigenomics Consortium , 2015; Roost et al . , 2015 ) . As a consequence of these combined deficiencies , we set about compiling global transcriptomic data during the critical phase of human organogenesis , sampling each germ layer and including sites of mixed origin that are subject to major developmental disorders such as cleft palate and limb abnormalities ( Figure 1a ) . 10 . 7554/eLife . 15657 . 003Figure 1 . Profiling the transcriptomes underlying organogenesis in human embryos . ( a ) Human embryo showing the 15 tissues and organs subjected to RNA-seq . ( b ) High dynamic range of human embryonic RNA-seq . The combined dataset ( black ) included expression of >90% of annotated protein-coding genes ( GENCODE18 [Harrow et al . , 2012] ) . ( c ) Human embryogenesis possesses a distinctive transcriptome . Human embryonic read counts were compared with equivalent fetal datasets from NIH Roadmap ( Roadmap Epigenomics Consortium , 2015 ) using edgeR ( Robinson et al . , 2010 ) and a false discovery rate ( FDR ) of 0 . 05 ( see Materials and methods , Supplementary file 1B ) . Negative log10 p-values are shown for selected biological process Gene Ontology ( GO ) terms with significant enrichment in either the embryonic or fetal gene sets following Fisher's exact test applied using the elimination algorithm ( Alexa and Rahnenfuhrer , 2010 ) ( Supplementary file 1C contains the full list of enriched terms ) . ( d ) Selected sites illustrate the highly specific expression of HOX genes within the human embryo . DOI: http://dx . doi . org/10 . 7554/eLife . 15657 . 00310 . 7554/eLife . 15657 . 004Figure 1—figure supplement 1 . Transcription factor atlas of human organogenesis . Heatmap of gene expression for all transcription factors annotated on KEGG ( http://www . genome . jp/kegg-bin/get_htext ? hsa03000 ) across the fifteen different human embryonic organs and tissues . Absolute maximum values of expression for each gene are represented to the left ( green ) . The relative expression of each transcription factor across tissues is shown to the right ( blue ) set against its own maximum value ( threshold for inclusion , read count >100 in at least one tissue ) . A high resolution version , text-searchable for each individual transcription factor , is available as Supplementary file 5 . B , brain; R , retinal pigmented epithelium; P , palate; Th , thyroid / parathyroid; Lu , lung; S , stomach; Pan , pancreas; L , liver; To , tongue; H , heart / left ventricle; Te , testis; A , adrenal; K , kidney; UL , upper limb; and LL , lower limb . DOI: http://dx . doi . org/10 . 7554/eLife . 15657 . 00410 . 7554/eLife . 15657 . 005Figure 1—figure supplement 2 . Heatmap of user-defined transcription factors indicates organ and tissue specificity during human organogenesis . To validate that tissue-specific signatures should be readily attainable from the global dataset several transcription factors for each organ or tissue were selected based on recognized published roles and mutant mouse phenotypes ( data available from Mouse Genome Informatics , www . informatics . jax . org ) . The heatmap demonstrates clear tissue-specificity . DOI: http://dx . doi . org/10 . 7554/eLife . 15657 . 00510 . 7554/eLife . 15657 . 006Figure 1—figure supplement 3 . Principal components analysis of the human embryonic transcriptomes . Across the four principal components biological replicates clustered together but from global pairwise correlations only the brain and to a lesser extent the liver were clearly distinct from the other organs and tissues ( either extreme of principal component 2 ) . As part of the reason why the liver was distinctive the five most abundant genes ( ALB , AFP and three fetal hemoglobins ) accounted for >20% of the data whereas in the other datasets the top 5 genes were responsible for only ~5% of transcription . The overall conclusion was that the simple principal components analysis failed to segregate clearly the individual transcriptomes of the different organs and tissues , an outcome that led to the development of the LgPCA methodology . Four samples from two human pluripotent stem cell ( PSC ) lines , H1 and HUES64 ( NIH Roadmap datasets ) , are included here because they were subsequently included in the LgPCA analysis ( Figure 2 ) . The PSC lines are clearly distinct from the primary human embryonic tissue samples ( negative loadings in principal component 1 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 15657 . 00610 . 7554/eLife . 15657 . 007Figure 1—figure supplement 4 . Heatmap of RNA-seq samples . Samples are clustered based on Spearman’s rank correlation across all annotated genes . RNA-seq batch is indicated in the colored key to the left . In this study , RNA sequencing was performed in 3 batches . The pancreas RNA-seq was re-used from a previous study ( Cebola et al . , 2015 ) . Four samples from two human pluripotent stem cell ( PSC ) lines , H1 and HUES64 ( NIH Roadmap datasets ) , are included here because they were subsequently included in the LgPCA analysis ( Figure 2 ) . The PSC lines are clearly distinct from the primary human embryonic tissue samples . DOI: http://dx . doi . org/10 . 7554/eLife . 15657 . 00710 . 7554/eLife . 15657 . 008Figure 1—figure supplement 5 . NMF Metagene analysis . ( a ) Subsets of tissue-specific genes ( ‘metagenes’ ) were found using non-negative matrix factorisation ( NMF ) ( Gaujoux and Seoighe , 2010 ) . The initial screen using the co-phenetic distance suggested 11 exclusive metagenes . The NMF was re-run 200 times to assess consistency of sample groupings between runs . The resulting metagenes were discriminatory for liver , heart / left ventricle , adrenal gland , RPE , brain and thyroid / parathyroid while others sample types formed heterogeneous clusters: for instance , lung , stomach and tongue ( metagene 9 ) ; kidney & testis ( metagene 3 ) ; and limbs and palate ( metagene 6 ) . ( b ) NMF metagene analysis demonstrates enrichment of expression for those genes comprising metagene 2 ( liver ) in fresh human hepatocytes and human embryonic stem cells differentiated towards hepatocytes but not in human embryonic fibroblasts [sequence data from ( Du et al . , 2014 ) ] compared to the other metagenes . DOI: http://dx . doi . org/10 . 7554/eLife . 15657 . 008 Organs and tissues from fifteen human embryonic sites were sequenced in two sets of biological replicates ( except pancreas and tongue ) to generate 28 strand-specific RNA-seq datasets with 44–90 million uniquely mapped reads per replicate ( Figure 1a; Supplementary file 1A , which contains information on embryonic stages ) . Global transcription rates across all organs and tissues were comparable over a high dynamic range; approximately 70% of protein-coding genes contained 100–10 , 000 mapped reads ( Figure 1b; Supplementary file 2 ) . We assessed whether our human embryo datasets identified earlier developmental processes than currently available fetal data ( Roadmap Epigenomics Consortium , 2015 ) . There were three-fold the number of differentially expressed genes in the fetal datasets but equivalent enrichment of gene ontology ( GO ) terms in the embryo , including many early developmental processes such as morphogenesis of an epithelial bud , anterior/posterior pattern specification and embryonic morphogenesis . These were in contrast to homeostatic processes enriched in the fetal dataset ( Figure 1c; Supplementary file 1B–C ) . Sampling gene expression across multiple sites allowed us to set about deciphering the precise transcriptomic codes responsible for the development of the different human embryonic organs and tissues . While ZNF and ZSCAN family members were broadly expressed discrete site-specific expression was more apparent for individual members of other transcription factor families ( Figure 1—figure supplement 1 ) exemplified by the HOX gene clusters ( Figure 1d ) . User-defined sets of up to five developmental transcription factors characteristic for a particular organ or tissue displayed very high levels of tissue specificity ( Figure 1—figure supplement 2 ) . However , while principal components ( PC ) analysis ( PCA ) or clustering grouped biological replicates , relationships between different organs and tissues other than the distinctiveness of brain and liver were not resolved ( Figure 1—figure supplements 3–4 ) . Non-negative matrix factorisation ( NMF ) also allows unbiased clustering of gene expression ( Gaujoux and Seoighe , 2010 ) . By setting the parameters such that representative genes were only extracted once , we identified eleven non-overlapping ‘metagenes’ from the complete expression dataset with clear tissue-specific signals for thyroid , liver , RPE , brain , heart and adrenal gland ( Figure 1—figure supplement 5; Supplementary file 1D ) . We hypothesized that these new signals might allow benchmarking to assess the fidelity of in vitro differentiated stem cells , similar to a previous report ( Roost et al . , 2015 ) . As an exemplar , we chose hepatocyte differentiation for which RNA-seq data are available including positive ( primary adult hepatocytes ) and negative ( human embryonic fibroblasts ) control data ( Du et al . , 2014 ) . Clear enrichment for the stem cell-derived hepatocytes and the primary hepatocytes ( but not the fibroblasts ) was apparent in metagene 2 , the cluster of 39 genes indicative of human embryonic liver . From this starting point , we wanted to move beyond the unique organ-specific signatures to study how patterns of gene expression co-varied across tissues . While relaxing NMF parameters would allow non-exclusive gene selection across metagenes , we also wanted to capture differences in gene expression between organs ( e . g . aspects of what is not expressed as a contributing factor to an organ’s identity ) . Moreover , different embryonic organs are related according to developmental lineage . We reasoned that being able to apply a lineage structure would create natural assemblies of co-regulated genes ( Figure 2a ) . Accordingly , we adapted a method from spatial ecology and phylogenetics ( Jombart et al . , 2010a , 2010b ) to constrain PCA by imposing a hierarchical developmental lineage and termed this approach LgPCA . We also included RNA-seq from undifferentiated human embryonic stem cells ( Roadmap Epigenomics Consortium , 2015 ) to represent pre-gastrulation human biology . Of the total thirty-one principal components ( PCs ) arising from LgPCA the first fifteen now identified patterns of gene expression across groups of related tissues in addition to unique organ-specific signatures ( Figure 2b ) while PCs 16–31 sampled heterogeneity within individual organs and tissues ( Figure 2—figure supplement 1 ) . In keeping with this transition PCs 1–4 ordered samples reminiscent of very early developmental events: pluripotency ( extreme positive loadings in PC1; ‘PC1 high’ ) , early brain formation ( extreme negative loadings in PC2; ‘PC2 low’ ) , foregut endoderm ( PC4 low ) and intermediate mesoderm ( PC4 high ) . PCs 5–15 resolved the individual organs and tissues; for instance low PC5 loadings discriminated liver from the other foregut endoderm derivatives . Heatmaps illustrated the composite or tissue-specific signals emanating from the genes with the most extreme PC loadings which also underlay appropriate developmental gene ontology ( GO ) terms ( Figure 2c–d and Supplementary file 1E–F ) . 10 . 7554/eLife . 15657 . 009Figure 2 . Lineage-guided PCA discovers unique transcriptional signatures regulating human organogenesis . ( a ) Interpreting gene expression profiles is dependent upon the underlying developmental lineage . Similar expression profiles in closely related tissues imply fewer regulatory events . ( b ) Lineage-guided principal components analysis ( LgPCA ) constrains PCA by imposing a developmental lineage on the different organs and tissues . The first 15 PCs are shown including biological replicates for the human embryonic organs and tissues integrated with human embryonic stem cell data ( Roadmap Epigenomics Consortium , 2015 ) . PC scores for the 15 different dimensions are shown in black ( positive/high ) or white ( negative/low ) with scale ( extremeness ) indicated by circle size ( sign/direction is arbitrary ) . Unique transcriptional signatures were resolved for broad organ groupings ( e . g . foregut endoderm derivatives , low scores in PC4 ) , single organs or tissues ( e . g . palate , high scores in PC13 ) or across tissues unrelated by germ layer but connected by multisystem congenital disorders ( e . g . heart and limb , low scores in PC13 ) . ( c ) Heatmaps of quantile normalised expression values of the most extreme 50 genes for selected PCs from the LgPCA . ( d ) Gene Ontology ( GO ) terms and their underlying genes illustrate the specific signatures from the LgPCA ( further examples in Supplementary file 1F ) . DOI: http://dx . doi . org/10 . 7554/eLife . 15657 . 00910 . 7554/eLife . 15657 . 010Figure 2—figure supplement 1 . Lineage-guided principal components analysis ( LgPCA ) for all 31 PCs . LgPCA showing all 31 PCs illustrating that global patterns ( i . e . strong lineage and organ or tissue level signatures ) emerge from the earlier PCs ( ≤PC15 to the left ) while local patterns ( e . g . heterogeneity between samples ) become evident at ≥PC16 , to the right ) . Many individual PCs gave very clear organ or tissue-specific signatures , however , the transcriptomes of most organs and tissues can also be represented by a composite of patterns visible across a number of different PCs . DOI: http://dx . doi . org/10 . 7554/eLife . 15657 . 010 Identifying the master regulators that differentially orchestrate organogenesis across the body has not previously been possible directly in human embryos . We undertook this in two different ways , both based on studying the 1000 genes with the most extreme loadings in PCs 1–15 that identified gene co-expression patterns across tissues and within individual organs ( Figure 2b; Supplementary file 1E ) . We interrogated these gene sets for regulatory networks based on the enrichment of transcription factor motifs ( Janky et al . , 2014 ) . Numerous well known master regulators were recovered alongside previously unappreciated factors for either broad tissue groups ( e . g . foregut endoderm derivatives ) or individual organs ( Figure 3a ) . As proof-of-principle , this also included proven regulators of human pluripotency , NANOG , OCT4 and MYC , at an extreme of PC1 . Remarkably , in several instances approaching half of the 1000 genes with the most extreme PC loadings imputed co-regulation by a single transcription factor , such as HNF4A in the liver or SRF in the heart . Alongside NR5A1 , the data predicted RUNX and BAD as novel regulators of human adrenal and gonadal development ( Figure 3a ) . As a second approach to study gene regulation , we extracted the transcription factors ( typically <5% ) from amongst the 1000 most extreme genes in PCs 1–15 ( Supplementary file 1G ) . We searched the Mouse Genome Informatics database ( MGI ) and in 309/594 instances there was a relevant mouse mutation phenotype supporting the notion that the transcription factors identified by LgPCA are key regulators of human organogenesis . At the lowest extreme of PC5 ( liver ) the twenty-two transcription factors contained all three of those required for reprogramming fibroblasts directly to hepatocytes ( Huang et al . , 2014 ) . This suggests novel fate programming roles for transcription factors at the extremes of other PCs ( including new potential regulators of pluripotency amongst the sixteen factors containing zinc fingers in PC1 ) . In keeping with these regulatory roles , the extreme PC loadings in the LgPCA data also prioritized those transcription factors responsible for major congenital disorders ( Supplementary file 1G ) . Because LgPCA is not limited to individual organs this included a novel ability to predict multisystem abnormalities such as the combined heart and limb defects of Holt-Oram syndrome ( OMIM 142900 , TBX5 , PC13 low ) or the palate and limb abnormalities associated with mutations in TP63 ( OMIM 603543 , PC3 high ) . 10 . 7554/eLife . 15657 . 011Figure 3 . LgPCA points to master regulators of human organogenesis and the causes of human congenital disorders . ( a ) Predicted regulation by iRegulon ( Janky et al . , 2014 ) of the most extreme 1000 genes for different PCs identifies known and unexpected transcription factors regulating human organogenesis . In several examples , individual transcription factors ( e . g . REST , NR5A1 , HNF4A , FOXA1 and SRF ) were predicted to regulate nearly half of the most extreme 1000 genes . ( b ) Transcription factors at the extremes of individual PCs in the LgPCA are responsible for a diverse range of congenital disorders ( red names in the ovals for heart and testis; full details in Supplementary file 1G ) . To validate the utility of these data , we conservatively selected some of the earliest critical regions for these disorders ( two ‘Proven’ examples on the left; all 53 listed in Supplementary file 1H ) . LgPCA frequently isolated the correct transcription factor from an average of 111 genes across >10 Mb , shown for NKX2-5 in congenital heart disease and SOX9 in campomelic dysplasia . Beyond this validation LgPCA similarly predicts causative transcription factors ( blue ) for many unresolved congenital disorders such as developmental heart abnormalities in Chr1p36 deletion syndrome and sex reversal / disorders of sex differentiation ( DSD ) ( all 13 examples in Supplementary file 1H ) . DOI: http://dx . doi . org/10 . 7554/eLife . 15657 . 011 Mutations in genes encoding transcription factors are over-represented causes of congenital disorders , most likely due to their critical function during organogenesis and inadequacy when haploinsufficient . The enrichment of transcription factors with specific disease-associations at the extremes of the LgPCA implicates the co-enriched genes as leading candidates for unanswered clinical syndromes . To test this model we identified some of the earliest chromosomal mapping or patient deletion data for the known disease-associated transcription factors from Supplementary file 1G . 53 disorders were suitable for assessment with an average critical region of 13 . 7 Mb each containing an average of 111 protein-coding genes ( Supplementary file 1H ) . Strikingly , in 37 instances ( 73% ) LgPCA uniquely selected the correct transcription factor and in 48 instances ( 91% ) narrowed the field down to three or fewer transcription factors . When applied to 13 syndromes ( mostly deletion disorders ) where the causative gene remains unresolved clear predictions of causality emerge , for instance in cleft palate ( DLX5 , DLX6 , LHX8 and FOXF2 ) or cerebellar disorders ( ZIC1 and ZIC4 ) ( Supplementary file 1H ) . Frequently , there is an appropriate mutant mouse phenotype such as CASZ1 in cardiac malformations , part of Chr1p36 deletion syndrome , or SOX10 in the 46 , XX disorder of sex differentiation ( DSD ) linked to duplication on Chr22 ( Figure 3b and Supplementary file 1H ) . Non-coding transcription has emerged as a critical regulator of cell and developmental biology ( Goff and Rinn , 2015 ) . A dedicated programme operating during human organogenesis seemed likely as 81 out of the 1571 genes enriched in embryogenesis compared to the fetal datasets were annotated long intergenic non-coding ( LINC ) transcripts ( Supplementary file 1B ) . To look beyond this we assembled strand-specific transcripts not recognized by current genome annotation [GENCODE 18 ( Harrow et al . , 2012 ) ] and systematically named them individually according to recommended criteria ( Mattick and Rinn , 2015 ) . 6251 unique loci accounted for in excess of 9 Mb of novel polyadenylated transcription from the human genome ( Figure 4a and Supplementary file 1I ) . The vast majority of transcripts fulfilled criteria as lnc RNAs by assessment of coding potential ( CPAT score <0 . 2 ) ( Figure 4b ) , length over 200 base pairs ( bp ) and an absence of reads spanning splice junctions to currently annotated genes ( Mattick and Rinn , 2015 ) . These lncRNAs were classified as either bidirectional , antisense or overlapping , or by exclusion intergenic , according to orientation and position in relation to the annotated genome ( Mattick and Rinn , 2015 ) . Transcripts were most commonly 500–1 , 500 bp but could extend to over 600 Kb ( Figure 4c ) and showed high tissue-specificity with the median Tau value ( Yanai et al . , 2005 ) of 0 . 86 , much higher than for protein-coding genes ( 0 . 63 ) but consistent with previously annotated non-coding genes ( 0 . 89 ) . We investigated the association between this novel human embryonic transcriptome and the annotated genome . Reduced physical distance to expressed annotated genes markedly increased the likelihood of novel transcript co-expression ( Figure 4e ) , although the best correlations were by no means always with the closest gene ( Figure 4f–g ) . The median distance to the closest annotated gene was 7 . 7 Kb ( Figure 4—figure supplement 1 ) while on average the best correlation was at 188 Kb ( random prediction was 476 Kb ) . Over half ( 3634 ) of the lnc transcripts were classified positionally as LINC RNAs . While LINC RNAs can harbour important regulatory function , how to forecast their relationship ( s ) with the protein-coding genome and prioritize the investigation of thousands of new transcripts is immensely challenging ( Goff and Rinn , 2015 ) . As a first step , the multi-tissue nature of our dataset allowed intricate correlative patterns to be deciphered implying putative relationships; for instance over a 2 Mb window and across numerous genes on chromosome 7 between HE-LINC-C7T121 and TBX20 , which encodes a developmental cardiac transcription factor mutated in a wide range of congenital heart disease ( Figure 4h ) . 10 . 7554/eLife . 15657 . 012Figure 4 . 6251 novel transcripts identified during human organogenesis show low coding probability and high tissue-specificity . ( a ) Novel transcript models were merged across tissues ( n = 9180; Supplementary file 4 ) , assessed for coding potential using CPAT and classified ( Mattick and Rinn , 2015 ) as overlapping ( OT ) , antisense ( AS ) , bidirectional ( BI ) , intergenic noncoding ( LINC ) and/or transcripts of uncertain coding potential ( TUCP , if CPAT >0 . 2 ) . LINC or TUCP transcripts were numbered sequentially ( T number ) along each chromosome ( C , either X , Y or 1–22 ) whereas BI , AS and OT transcripts were named by association with the annotated gene ( ‘Z’ ) . A small proportion of transcripts fulfilled dual criteria as BI/AS/OT and TUCP . 6251 unique , non-overlapping , filtered transcript models were identified ( the longest from each locus , >200 bp; Supplementary file 1I ) . ( b ) Histogram of coding probability determined using CPAT ( Wang et al . , 2013 ) . 9% of transcripts were classed as TUCP . The small proportion with clear open reading frames ( CPAT score = 1 . 0 ) were predominantly OT transcripts . ( c ) Distribution by size of transcript . 114 transcripts were >10 Kb . ( d ) Tissue specificity was calculated using Tau ( Yanai et al . , 2005 ) based on the mean normalized read counts for each tissue or organ site . 80% of transcripts showed Tau values >0 . 7 indicating high tissue specificity . Details on exon and read counts , and proximity to surrounding genes are shown in Figure 4—figure supplement 1 . ( e ) Box and whisker plots show the correlation between expression of the novel transcripts and surrounding annotated genes within set chromosomal distances of the novel transcriptional start site . Mean correlation was near zero beyond 1 Mb . ( f ) Histogram showing the correlation ( r ) between expression of each novel transcript and its closest annotated gene . One quarter of novel transcripts show a correlation ( r > 0 . 71 ) with the nearest gene; another quarter shows minimal correlation ( r = ±0 . 14 ) . There was no strong anticorrelation . g-h , Expression of the novel transcript is not always correlated with the immediately adjacent gene , illustrated by heatmaps across the 15 organs and tissues . ( g ) Expression of the novel transcript , HE-LINC-C6T24 , located just over 2 Kb from FOXQ1 , correlates strongly with FOXF2 , approximately 65 Kb distant . ( h ) Heatmap demonstrates the poor correlation of expression between HE-LINC-C7T121 and most of the nine genes within 1 Mb on Chr7 but near perfect correlation with TBX20 located ~0 . 7 Mb away beyond two intervening genes . DOI: http://dx . doi . org/10 . 7554/eLife . 15657 . 01210 . 7554/eLife . 15657 . 013Figure 4—figure supplement 1 . Exon and read counts and distance to the nearest annotated gene for the novel human embryonic transcripts . ( a–c ) Histograms showing the number of exons ( a ) , maximum read count for each transcript in any one tissue ( b ) , and total reads ( i . e . summed across all tissues ) for each transcript ( c ) . ( d ) Distance to the transcriptional start site ( TSS ) of the nearest annotated gene ( GENCODE18 ) from the TSS of the novel transcript . DOI: http://dx . doi . org/10 . 7554/eLife . 15657 . 013 Taken together , this study reports the first comprehensive transcriptomic atlas during human organogenesis to complement parallel initiatives from later development and adulthood ( Jaffe et al . , 2015; Roadmap Epigenomics Consortium , 2015; Roost et al . , 2015 ) . Subjecting transcription from many sites to a method of analysis that incorporated developmental lineage deciphered novel genetic signatures , predicted causality in many human developmental disorders and associated novel non-coding transcription with expression from the surrounding protein-coding genome . At present , the data arise from a relatively narrow window of embryonic development but set the stage for future longitudinal studies for individual organs over time . The tiny amounts and scarcity of human embryonic tissue also necessitated aspects of pooling across different Carnegie stages for some sites but it is striking that this had no impact on ascertaining organ and tissue-specific transcriptomic signatures by LgPCA . The integrated data are expected to be particularly valuable to stem cell researchers examining the fidelity of PSC differentiation in vitro or searching for transcription factors for direct reprogramming of chosen cell lineages . Finally , the discovery of a major new programme of non-coding transcription adds a fresh layer of detail on the spatiotemporal regulation of the human genome .
Human embryonic material was collected under ethical approval , informed consent and according to the Codes of Practice of the Human Tissue Authority and staged by the Carnegie classification as described previously ( Jennings et al . , 2013 ) . This clinical material was collected on site overseen by our research team with immediate transfer to the laboratory . Individually identified tissues and organs ( details in Supplementary file 1A ) were immediately dissected . The adrenal gland , whole brain , heart , kidney , liver , entire limb buds , lung , stomach , testis , thyroid and anterior two-thirds of the tongue were readily identifiable as discrete organs and tissues . All visible adherent mesenchyme was removed from organs and tissues under a dissecting microscope . For the adrenal gland , this includes the capsule which allowed separation from the kidney . The ureter , emerging from the renal pelvis , was removed separately from the kidney . For the heart , a window of tissue was removed from the lateral wall of the left ventricle . A segment of the liver was dissected from each embryo that avoided the developing gall bladder . The trachea was removed from the lung at its entry point into the lung parenchyma . The stomach was identified between the gastro-oesophageal and pyloric junctions . The testis was dissected free from the attached mesonephros . While the thyroid was readily visualized as a discrete organ in the neck , it unavoidably contained the developing parathyroids and thus this tissue type was referred to throughout as ‘thyroid/PTH’ . The palatal shelves were dissected on either side of the midline . The eye was dissected and the RPE peeled off mechanically from its posterior surface with validation possible under the dissecting microscope because the RPE is very darkly pigmented compared to the other ocular structures . All samples were collected into Trizol ( Thermofischer ) or Tri reagent ( Sigma-Aldrich ) for total RNA isolation as individual tissue or organ types followed by treatment with DNaseI ( Sigma-Aldrich ) . Once the quality of each RNA sample had been confirmed , samples were pooled in order to obtain sufficient RNA for each biological replicate ( Supplementary file 1A ) . The pancreas dataset derived from the same tissue collection was re-used from a previous study ( Cebola et al . , 2015 ) . Quality and integrity of total RNA samples were assessed using a 2100 Bioanalyzer or a 2200 TapeStation ( Agilent Technologies ) according to the manufacturer’s instructions . RNA sequencing ( RNA-seq ) libraries were generated using the TruSeq Stranded mRNA assay ( Illumina , Inc . ) according to the manufacturer’s protocol . Briefly , total RNA ( 0 . 1–4 µg ) was used as input material from which polyadenylated mRNA was purified using poly-T , oligo-attached , magnetic beads . The mRNA was then fragmented using divalent cations under elevated temperature and then reverse transcribed into first strand cDNA using random primers . Second strand cDNA was then synthesized using DNA Polymerase I and RNase H . Following a single 'A' base addition , adapters were ligated to the cDNA fragments , and the products purified and enriched by PCR to create the final cDNA library . Adapter indices were used to multiplex libraries , which were pooled prior to cluster generation using a cBot instrument . The loaded flow-cell was then sequenced ( paired-end; 101 + 101 cycles , plus indices ) on an Illumina HiSeq2000 or HiSeq2500 instrument . Demultiplexing of the output data ( allowing one mismatch ) and BCL-to-Fastq conversion was performed with CASAVA 1 . 8 . 3 . The RNA-seq was conducted in three batches at different times as a necessity of how human embryonic tissue was collected over time ( Supplementary file 1A ) . Where organs were sequenced across batches ( palate , RPE , kidney , testis , adrenal gland , heart / left ventricle and liver ) biological replicates clustered together ( Figure 1—figure supplement 4 ) . RNA-seq reads from the Illumina platform were mapped to the human genome ( hg19 ) strand-specifically using TopHat 2 . 0 . 9 ( Trapnell et al . , 2012 ) and the GENCODE 18 gene annotation set ( Harrow et al . , 2012 ) . We also remapped the published pancreas RNA-seq dataset ( Cebola et al . , 2015 ) obtained from material isolated previously in our laboratory . Additionally , a dataset of hepatocyte differentiation RNA-seq ( Du et al . , 2014 GEO: GSE54066 ) was downloaded , re-mapped and quantified as per our own data . Commonly applied RNA-seq normalisation methods such as TMM assume a small proportion of differentially expressed genes in any one dataset ( Dillies et al . , 2013 ) . Because the highly distinct tissues surveyed here differed strongly on the scale of thousands of genes ( for instance liver versus brain ) we used quantile normalisation which gave a lower median coefficient of variation than either no or TMM normalization . Read counts from the different datasets were quantile normalized using the R package preprocessCore ( Bolstad , 2007 ) . Tissue-specificity was scored per gene using Tau ( Yanai et al . , 2005 ) on normalized read counts across all samples . Initial genome-wide relationships were assessed using PCA ( Figure 1—figure supplement 3 ) and hierarchical clustering ( heatmap , Figure 1—figure supplement 4 ) . To compare our samples with RNA-seq from the NIH Roadmap project ( Roadmap Epigenomics Consortium , 2015 ) uniquely mapped strand-specific RNA-seq reads were counted into a set of non-redundant exon annotations ( custom made from GENCODE 18 annotations ) using bedtools intersect ( Quinlan and Hall , 2010 ) . Exon level counts were then summed into a single total per gene per sample . Counts were quantile normalized across samples . NIH roadmap samples ( Roadmap Epigenomics Consortium , 2015 ) used in this study are listed in Supplementary file 1J . For the analysis of human embryonic RNA-seq with comparable Roadmap fetal data ( adrenal gland , heart , kidney , lung , limbs , stomach and testis ) a single pairwise differential expression test was undertaken using the R package edgeR ( Robinson et al . , 2010 ) and an FDR < 0 . 05 . Non-negative matrix factorisation ( NMF ) searches complex expression data , comprising thousands of genes , for a small number of characteristic ‘metagenes’ ( Gaujoux and Seoighe , 2010 ) . NMF was performed using the nmf R package ( version NMF_0 . 20 . 5 ) ( Gaujoux and Seoighe , 2010 ) to extract tissue-specific metagenes . Non-normalised read counts were filtered to remove all Y-linked genes , the X-inactivation gene XIST and genes with fewer than 100 reads across all samples . Initially 50 runs each of ranks 11–18 and using the default ‘Brunet’ algorithm ( Brunet et al . , 2004 ) were performed to find an optimal factorisation ‘rank’ ( r ) . The maximal cophenetic distance was used to select the value of r . Subsequently , 200 runs using the optimal rank were performed to assess consistency of sample groupings between runs . Non-overlapping ( i . e . tissue-specific ) gene sets were extracted from each metagene by filtering on basis contribution >0 . 8 . The LgPCA approach was adapted from established phylogenetic PCA methodology ( Jombart et al . , 2010b ) and performed using quantile-normalized , gene-level read counts , a high memory ( 512 Gb ) compute node and the ppca function from the adephylo R package ( Jombart et al . , 2010a ) . A broad user-defined guide tree ( Figure 2b ) based on well-established knowledge of mammalian gastrulation and downstream lineage relationships was imposed on the different organ and tissue types following which the adephylo R package weighted the principal components by the lineage auto-correlation between samples; increased if related samples were similar and lessened if related samples were more different . As in the description from Jombart and colleagues the resulting components represented ‘global’ structures ( where similarity is high between related samples ) and ‘local’ structures ( where related samples are dissimilar ) ( Jombart et al . , 2010b ) . We used the LgPCA to extract all the global patterns from the data ( PCs 1–15 ) . These patterns were not apparent if lineage relationships were not included nor were they altered if any one tissue , such as palate , was altered within the broad lineage structure ( data not shown ) . The global patterns in PCs 1–15 infer co-regulatory patterns of gene expression across human organogenesis . The ‘local’ patterns thereafter captured heterogeneity between tissue replicates ( Figure 2—figure supplement 1 ) ( while PC7 separated the two PSC populations these RNA-seq datasets represent separate cell lines from NIH Roadmap ) . We used the Abouheif distance as implemented in adephylo ( Jombart et al . , 2010a ) , which takes into account the topology of the specified tree but does not use branch lengths . For the comparison of the embryonic versus fetal datasets Gene Ontology term enrichment was performed on upregulated genes ( FDR < 0 . 05 ) using Fisher’s exact test with the elimination algorithm of the R package topGO ( Alexa and Rahnenfuhrer , 2010 ) . For the LgPCA , annotated ontology nodes ( >10 genes ) were tested for each loadings vector for each PC against background using the Wilcoxon test . Tests were performed sequentially moving up the separate GO ontologies ( Biological Process ( BP ) , Molecular Function ( MF ) and Cellular Component ( CC ) ) , excluding significant scoring genes from later tests ( the topGO ‘elim’ method ) . iRegulon is a computational method which tests for enrichment amongst precomputed motif datasets to decipher transcriptional regulatory networks in a set of co-expressed genes . The 1000 genes with the most extreme loadings at either end of each PC ( ‘high’ and ‘low’ ) from the LgPCA were loaded into Cytoscape ( version 3 . 2 . 1 ) ( Shannon et al . , 2003 ) and used as queries to the iRegulon plug-in ( version 1 . 3 , build 1024 ) ( Janky et al . , 2014 ) . 20 Kb was examined centred on the transcriptional start site ( TSS ) under default settings . Sample-specific transcriptomes were assembled with Cufflinks ( version 2 . 2 . 0 ) ( Trapnell et al . , 2010 ) . Transcriptomes were combined ( ‘cuffmerge’; -min-isoform-fraction = 0 . 1 ) and compared with the original GENCODE 18 reference ( ‘cuffcompare’ ) . We filtered out known transcripts using the ‘Transfrag class codes’ ( http://cole-trapnell-lab . github . io/cufflinks/cuffcompare/#transfrag-class-codes ) to retain only wholly intronic ( ‘i’ , of which there were none ) , unknown ( ‘u’ ) , antisense ( x ) and overlapping ( ‘o’ ) transcripts . We discarded all other classes including pre-mRNA ( class ‘e’ ) , novel-isoforms spliced to known exons ( class ‘j’ ) , and 3’ run-ons within 2 kb of the end of the transcript annotation ( class ‘p’ ) . In addition , some remaining non-spliced transcripts may theoretically represent first or last exon ( UTR ) extensions; to delimit these , we calculated the distance on the same strand to the closest downstream transcription start site ( to consider potential 5’ UTR extension ) and upstream transcription termination site ( to consider potential 3’ UTR extension ) . Names were assigned to these novel transcripts following suggested criteria ( Mattick and Rinn , 2015 ) ( Figure 4a ) with the sole adaptation that bidirectional ( BI ) transcripts were defined as having their TSS within 1 Kb of the TSS of the associated annotated gene . No transcripts mapped to the same strand within the introns of any annotated gene excluding the possibility of unspliced transcripts from annotated genes being erroneously defined as novel transcripts . All transcript sequences ( annotated and unannotated ) were scored for protein-coding potential using CPAT ( based on human training data included with CPAT ) ( Wang et al . , 2013 ) . A threshold of >0 . 2 was used to define ‘Transcripts of Uncertain Coding Potential’ ( TUCP ) . Where there were multiple transcripts from a single locus , the longest transcript was retained in assembling the final dataset of 6251 novel transcripts . Transcript level read counts for the embryonic samples and NIH Roadmap samples ( Supplementary file 1J ) were generated for the merged transcriptome using bedtools multicov ( vers 2 . 21 . 0 ) ( Quinlan and Hall , 2010 ) . The correlations between each of the 6251 transcripts and all annotated genes within 1 Mb were calculated using only the human embryonic data from this study . Mapping coordinates against multiple genome versions using a range of common pipelines and summary count data are hosted at www . manchester . ac . uk/human-developmental-biology . To view data in the UCSC genome browser , a trackhub is available: http://www . humandevelopmentalbiology . manchester . ac . uk/data/hub_manchester_hdb/hub . txt . | Individual organs and tissues form in human embryos during the first two months of pregnancy . Any errors during this crucial stage of human development can result in miscarriage or serious birth defects . Yet remarkably little is known about how this process works . What is known has been inferred from studies of how other animals develop , human stem cells grown in a laboratory , and babies born with genetic conditions that cause developmental problems . Genes control the way that organs and tissues form , and are switched on or off in complex patterns during development to ensure that particular cells develop into one type of organ and not another . When genes are switched on , their DNA is copied into molecules called RNA . Many RNA molecules are used as templates to make proteins , which then perform critical roles in cell processes . One way to find out which genes are activated during development is to identify which RNAs are made by cells in the embryo . Here , Gerrard , Berry et al . used a technique called RNA-sequencing to identify the RNAs that human embryos make while their organs and tissues form . The RNA came from many different tissues including the heart , limbs and the roof of the mouth . Gerrard , Berry et al . developed a new computational model that used the identity of the RNAs to decode the precise patterns of gene activity in the tissues . The model correctly identified many genes that were already known to cause developmental problems when faulty , and identified numerous others that are now predicted to cause developmental defects in humans . Gerrard , Berry et al . also discovered over 6 , 000 RNAs in the human embryos that are unlikely to code for proteins . These “non-coding” RNAs may have other roles in cells , such as switching off genes , and many of them appear to be specific to human embryos . Together , these findings have uncovered new patterns of gene activity that drive development in human embryos and provide a resource for studying how organs and tissues form . Future challenges are to understand what controls these patterns of gene activity , and how the patterns change over time . | [
"Abstract",
"Introduction",
"Materials",
"and",
"methods"
] | [
"stem",
"cells",
"and",
"regenerative",
"medicine",
"developmental",
"biology",
"tools",
"and",
"resources"
] | 2016 | An integrative transcriptomic atlas of organogenesis in human embryos |
Bacterial mRNAs are organized into operons consisting of discrete open reading frames ( ORFs ) in a single polycistronic mRNA . Individual ORFs on the mRNA are differentially translated , with rates varying as much as 100-fold . The signals controlling differential translation are poorly understood . Our genome-wide mRNA secondary structure analysis indicated that operonic mRNAs are comprised of ORF-wide units of secondary structure that vary across ORF boundaries such that adjacent ORFs on the same mRNA molecule are structurally distinct . ORF translation rate is strongly correlated with its mRNA structure in vivo , and correlation persists , albeit in a reduced form , with its structure when translation is inhibited and with that of in vitro refolded mRNA . These data suggest that intrinsic ORF mRNA structure encodes a rough blueprint for translation efficiency . This structure is then amplified by translation , in a self-reinforcing loop , to provide the structure that ultimately specifies the translation of each ORF .
Protein synthesis is the most energetically costly process in bacteria , consuming up to 50% of cellular energy . Thus , to optimize cellular efficiency , the rate of synthesis of each protein must be carefully controlled . In bacteria , operons are central to this process . Open reading frames ( ORFs ) with related functions are organized into operons that are transcribed as a single mRNA ensuring that operonic genes are transcriptionally co-regulated in response to various conditions ( Jacob and Monod , 1961 ) . Additionally , the translation of each ORF in the operon is precisely tuned to cellular need . Indeed , the rate of protein production ( i . e . the translation efficiency ) of adjacent ORFs on a single mRNA can vary by as much as 100-fold , and members of protein complexes encoded on a single mRNA are generally translated in proportion to their stoichiometry ( Li et al . , 2014 ) . Understanding how the cell achieves optimal energy utilization critically depends on understanding how mRNA sequence features reliably drive ORF-specific translation . A number of mRNA features have been identified as contributing to the rate of translation of an ORF . Both the strength and accessibility of a Shine-Dalgarno ( SD ) sequence upstream from the initiation codon ( Steitz and Jakes , 1975 ) have been implicated in translatability . In support of the importance of SD accessibility , highly stable mRNA structures in direct proximity to the initiation codon diminish translatability ( de Smit and van Duin , 1990; Hall et al . , 1982; Lodish , 1970 ) and rare codons that disfavor RNA structure are enriched in positions immediately following the translation start site ( Bentele et al . , 2013; Eyre-Walker and Bulmer , 1993; Scharff et al . , 2011 ) . Moreover , several studies examining either synthetic ORFs with a few bases difference ( e . g . alterations to GFP ) , or fluorescent reporter assays studying the effect of multiple codon changes in the 5’ UTR and N-terminal coding sequences , find that models based on their predicted RNA structure at the translation start site are relatively successful at predicting their differences in translatability ( Goodman et al . , 2013; Kudla et al . , 2009 ) . Most recently , codon usage has emerged as an important variable for translation . A large study examining thousands of foreign ORFs concluded that except for the very initial nucleotides of the ORF , codon usage rather than mRNA folding propensity was the critical determinant for translatability ( Boël et al . , 2016 ) . While these mRNA features are of clear value for predicting the translatability of exogenously expressed ORFs , several considerations suggest that they may not capture the key features that have evolved to set the translation efficiency of endogenous genes . First , all the high-throughput studies overexpressed the mRNAs they studied , which is known to perturb the charged tRNA pool and introduce biases in codon usage ( Dittmar et al . , 2005; Elf et al . , 2003 ) . Second , in the Boel et al . manuscript and in other studies , mRNAs were transcribed by T7 RNA polymerase , which not only elongates significantly faster than E . coli RNA polymerase but also removes the influence of the many endogenous E . coli RNA polymerase binding proteins that modulate its elongation rate . Where examined , such RNAs exhibit altered folding patterns ( Lewicki et al . , 1993; Chao et al . , 1995; Pan et al . , 1999 ) . Thus , these transcripts likely have non-native structure . Third , these studies all used foreign mRNAs , which had not been subjected to evolution for precise tuning in E . coli . Finally , these studies primarily measured protein abundance , a quantity that is dependent on mRNA and protein stability as well as on the efficiency at which each ORF is translated . The goal of this work is to understand how E . coli establishes the relative expression of adjacent ORFs on the same mRNA . To accomplish this , we systematically assessed the translational efficiency ( TE ) of every ORF mRNA and then examined which of its features ( e . g . secondary structure , codon usage , and the strength of ribosome binding site ) correlated with its translatability . The translation efficiency of endogenous messages in E . coli could be probed with existing global technologies ( Li et al . , 2014; Oh et al . , 2011; Ingolia et al . , 2009 ) and the effects of codon usage with two metrics , tAI ( tRNA adaptation index ) ( Tuller et al . , 2010; dos Reis et al . , 2004 ) and codon influence ( Boël et al . , 2016 ) . However , in vivo mRNA structure has not previously been empirically evaluated at the global level in E . coli . We therefore adapted the dimethyl sulfate ( DMS ) -seq technique ( Rouskin et al . , 2014 ) , which uses next-generation sequencing to determine chemical accessibility of RNA to DMS , to E . coli . Our studies point to a critical role of intrinsic ORF-wide differences in mRNA structure in allowing differential translation of ORFs sharing the same operonic mRNAs .
New genomic technologies enable the determination of RNA structure on a global scale ( Ding et al . , 2014; Rouskin et al . , 2014; Wan et al . , 2014 ) . DMS-seq uses next-generation sequencing to determine chemical accessibility of RNA to DMS ( dimethyl sulfate ) , a reagent that reacts with unpaired adenosine and cytosine nucleotides ( Inoue and Cech , 1985 ) . We adapted DMS-seq to E . coli to monitor global in vivo RNA structure ( Figure 1A ) . By exploring the effect of coverage on reproducibility , we find that a read coverage of ~15 reads/nucleotide is sufficient for reproducible structure determination ( Figure 1B ) , and used this cutoff in all subsequent analyses . Structures determined from E . coli-adapted DMS-seq are in excellent agreement with both the 16S rRNA crystal structure ( Figure 1C ) ( Zhang et al . , 2009 ) , and a mutationally verified E . coli mRNA structure ( Figure 1D ) ( Wikström et al . , 1992 ) . 10 . 7554/eLife . 22037 . 003Figure 1 . DMS-seq effectively probes RNA structures in E . coli . ( A ) Schematic for obtaining mRNA structure and translation efficiency using DMS-seq , mRNA-seq , and ribosome profiling from the same sample . ( B ) Plot showing the effect of DMS-seq read coverage on the reproducibility of structure determination . X-axis: DMS-seq read depth cutoff ( reads/nucleotide ) ; Y-axis: median of Pearson’s R values calculated by comparing two replicates of in vivo DMS-seq signals of the first 200nt of ORFs passing the DMS-seq depth cutoff indicated in X-axis . A read coverage of ~15 reads/nucleotide is sufficient for reproducible structure determination . ( C ) Receiver operating characteristic ( ROC ) curve on the in vivo DMS-seq signals for A and C bases in the 16S rRNA using the E . coli ribosome crystal structure ( Zhang et al . , 2009 ) as a model . True positives are defined as bases that are both unpaired and solvent-accessible , and true negatives are bases that are paired . The total number of evaluated A/C bases is 438 . Signal threshold of 0 . 2 has 90% agreement with the crystal structure . ( D ) Structural prediction for rimM . The predicted rimM structure is based on a minimum free-energy prediction constrained by our DMS-seq measurements , using the same 0 . 2 threshold used for the 16S rRNA in ( B ) , which agrees with the rimM structure proposed and mutationally verified in Wikström et al . ( 1992 ) . The DMS-seq signal across rimM is shown below the structure . The color bar indicates the intensity of the DMS-seq signal at each position . ( E ) Calculation of the Gini index from the DMS-seq signal is indicated schematically by comparing highly structured regions to less structured regions . For a region of mRNA , the cumulative fraction of the total DMS-seq signal is plotted against the cumulative fraction of the total number of positions as a Lorenz Curve . The extent to which the curve sags below the diagonal indicates the degree of inequality of distribution , which is quantified by the Gini index defined as the ratio of the area between the diagonal line and the Lorenz Curve ( a ) to the area below the diagonal line ( a + b ) . A high Gini index indicates high level of mRNA structure , and vice versa . ( F ) Histogram of Gini indices of E . coli ORFs calculated from in vivo DMS-seq data at 37°C . All ORFs selected have ≥15 DMS-seq reads/nt ( N = 1116 ) . The Gini index of 16S rRNA and rimM , and the mean of Gini indices of in vitro heat-denatured mRNAs at 95°C are indicated . DOI: http://dx . doi . org/10 . 7554/eLife . 22037 . 003 We quantified the degree of secondary structure for each ORF using the Gini index metric , which measures the variability in reactivity of A and C residues to DMS in the region being examined ( Rouskin et al . , 2014 ) . A low Gini index indicates a relatively even distribution of DMS-seq reads and occurs in unstructured regions of the mRNA . A high Gini index occurs when a subset of residues is strongly protected from DMS reactivity and indicates a high degree of structure ( Figure 1E ) . We found that the degree of RNA secondary structure varied greatly between ORFs: a small number are nearly as structured as rRNA , whereas some are close to the denatured state ( Figure 1F ) . In contrast to the large variation in the degree of secondary structure among ORFs ( Figure 1F ) , individual ORFs generally have fairly consistent Gini scores across their bodies – for example , the first and second halves of each ORF exhibit highly correlated Gini scores ( Figure 2A ) . We tested whether the Gini scores across ORFs remained correlated in the absence of translation in two different conditions . First , we examined Gini scores of in vivo mRNA when translation initiation was inhibited with kasugamycin . We achieved rapid inhibition by using a ΔgcvB mutant , which has enhanced kasugamycin uptake rates ( Figure 2—figure supplement 1A; Shiver et al . , 2016 ) . Using ∆gcvB mutant is critical for this experiment because kasugamycin uptake by wild-type ( WT ) cells is slow enough to allow massive degradation of mRNA before ribosomes are cleared ( see extended methods for protocol and Figure 2—figure supplement 1B–C for method validation , including demonstrating that ∆gcvB does not alter global mRNA structure ) . Second , we examined Gini scores of purified mRNA refolded in vitro at 37°C . In both cases , the translation-independent mRNA structures obtained from DMS-seq indicated that Gini scores across the ORF mRNAs remain correlated ( Figure 2B–C ) . This correlation also holds true for computationally predicted mRNA structure of ORFs ( Figure 2D ) . Moreover , the degree/extent of mRNA structures ( henceforth referred to as structure ) determined in these various ways are highly correlated with each other ( Figure 2E–F ) . We conclude that mRNA is organized in ORF-wide structures that depend on the intrinsic sequence of the mRNA . 10 . 7554/eLife . 22037 . 004Figure 2 . E . coli mRNAs have intrinsic ORF-wide secondary structures . ( A–C ) Plots comparing the Gini indices of the first half of the ORF against those of the second half of the ORF for: A . in vivo modified mRNA from cells growing at 37°C; B . in vivo modified mRNA from cells treated with kasugamycin ( ksg ) at 37°C ( no translating ribosomes ) ; C . in vitro mRNA modified at 37°C . In this and all subsequent figures , analysis is performed only on those ORFs with ≥15 DMS-seq reads per nucleotide , with N ( the number of ORFs analyzed in each condition ) , and ρ ( the Spearman's rank correlation coefficient ) indicated . The ksg-treated sample has fewer ORFs passing the ≥15 DMS-seq reads/nt filter , likely due to mRNA degradation when translation is eliminated . Data calculated using different sets of ORFs are summarized in Supplementary file 1–3 . ( D ) Plot comparing the computationally predicted mRNA structure ( - minimum free energy / nucleotide or -ΔG/nt ) of the first half of the ORF against that of the second half of the ORF for the 480 ORFs in the ksg-treated DMS-seq dataset . ( E ) Correlation between Gini indices of the entire ORF calculated from in vivo mRNA vs in vivo untranslated mRNA ( ksg-treated cells ) for the 465 ORFs in both datasets . The dashed grey line represents the y = x diagonal line . ( F ) Correlation between Gini indices of the entire ORF calculated from in vivo mRNA vs in vitro refolded mRNA for the 708 ORFs shared in both datasets . ( G ) Plot comparing Gini indices for adjacent ORFs in operons ( N = 326; see Materials and methods for details ) . The dashed grey line represents the y = x diagonal line . DOI: http://dx . doi . org/10 . 7554/eLife . 22037 . 00410 . 7554/eLife . 22037 . 005Figure 2—figure supplement 1 . mRNA structure is organized around open reading frames . ( A ) Relative 35S-methionine incorporation of WT and ΔgcvB cells after treatment of kasugamycin at 37°C , normalized against the total incorporated radioactivity measured immediately before treatment ( t = 0 ) . The relative translation decreases to 6% in ΔgcvB cells after 2-min treatment . This is the time-point used for probing mRNA structure without the majority of translating ribosomes in vivo . ( B ) Sucrose gradient analysis showing the polysome run-off in ΔgcvB cells after 2 min of kasugamycin ( ksg ) treatment at 37°C . X-axis: elution time of different fractions of sucrose gradient . Y-axis: relative UV absorbance . Monosome enrichment after ksg treatment was previously seen ( Kaberdina et al . , 2009 ) . ( C ) Comparison of Gini indices of entire ORF bodies in WT and ΔgcvB cells indicates that ∆gcvB does not affect global mRNA structure . The 351 ORFs in common between WT and ΔgcvB cells were analyzed . ( D ) Lorenz curves calculated from in vivo DMS-seq data of ORFs in the rpsF-priB-rpsR-rplI operon . Gini indices of ORFs calculated from the Lorenz curves are indicated . ( E ) Scatter plot comparing Gini indices of adjacent non-overlapping ( N = 253 ) and overlapping ( N = 73 ) ORFs within operons . Overlapping ORFs are ORF pairs for which the annotated stop codon of the upstream ORF overlaps or is 3’ of the start codon of the downstream ORF . The dashed grey line represents the y = x diagonal line . DOI: http://dx . doi . org/10 . 7554/eLife . 22037 . 005 We next examined whether structural correlation extends to adjacent ORFs on the same polycistronic ( operonic ) mRNA . We considered only those operons in which each ORF has an approximately equivalent mRNA levels , thus excluding those with significant internal promoters or terminators ( see Materials and methods ) . Within operonic ( polycistronic ) mRNAs , the mRNA structure of adjacent ORFs can differ significantly ( Figure 2G and Figure 2—figure supplement 1D ) , even when the start and stop codons of the adjacent ORFs overlap ( Figure 2—figure supplement 1E ) . Thus , characteristic mRNA structures are a property of individual ORF mRNAs rather than of the entire polycistronic transcript . We next explored the relationship between the level of ORF-wide mRNA structure identified above with the TE of that ORF . We previously demonstrated that the overall rate of protein production can be accurately measured by an ORF’s average ribosome footprint density ( number of footprints per unit length of the ORF ) , showing that protein copy number per cell determined from average ribosome footprint density was in superb agreement with that obtained by individually quantifying stable proteins in E . coli ( Li et al . , 2014 ) . Here , we build on that validated parameter , defining TE as the rate of protein production per mRNA , measured by normalizing average ribosome footprint density of an ORF with its mRNA abundance ( i . e . RPKM of mRNA sequencing ) ( Ingolia et al . , 2009; Li et al . , 2014; Oh et al . , 2011 ) , with both measurements obtained from the same biological samples ( see Materials and methods ) . Importantly , this metric is not affected by differences in either mRNA or protein abundance or stability ( Li , 2015 ) . We found that the TE’s of E . coli endogenous ORFs in operonic mRNAs were highly negatively correlated with their level of ORF-wide mRNA structure ( ρ = −0 . 75 , Figure 3A: well-translated ORFs have less mRNA structure , while poorly translated ORFs have more structure ) . Consistent with the fact that the ORF-wide mRNA structures of adjacent ORFs in an operon can differ significantly ( Figure 2G ) , the TE’s of adjacent ORFs can also differ significantly ( Figure 3B , Figure 3—figure supplement 1A ) . Notably , ORF pairs with overlapping start and stop codons , believed to be translationally coupled ( Aksoy et al . , 1984; Oppenheim and Yanofsky , 1980; Schümperli et al . , 1982; Yates and Nomura , 1981 ) , show essentially as much variability in their relative translation as non-overlapping ORF pairs ( p=0 . 06 , K-S test , Figure 3B ) , suggesting that the extent of coupling is variable . We then expanded this analysis beyond operons to all ORFs and found that the level of mRNA structure and TE are highly anti-correlated on all endogenous open reading frames ( ρ = −0 . 76 , Figure 3C ) . Importantly , the Gini scores of ORFs calculated from control RNA samples without DMS modification were not correlated to TE ( ρ = 0 . 05 , Figure 3—figure supplement 1B ) , indicating that Gini scores calculated from DMS-seq indeed reflect the level of mRNA structure and the potential sequencing bias/noise does not contribute to the correlation between TE and mRNA structure . 10 . 7554/eLife . 22037 . 006Figure 3 . Translational efficiency ( TE ) is highly correlated with ORF mRNA structure . ( A ) Plots comparing the Gini indices of ORFs in polycistronic operons calculated from in vivo DMS-seq to their TEs ( N = 483 ) . ( B ) Histograms of TE ratios between adjacent non-overlapping ( N = 253 ) or overlapping ( N = 73 ) ORFs in operons ( see Materials and methods for details ) . Overlapping ORFs are ORF pairs for which the annotated stop codon of the upstream ORF overlaps or is 3’ of the start codon of the downstream ORF . ( C–E ) Plots comparing the Gini indices of endogenous ORF mRNAs calculated from DMS-seq data of: C . in vivo RNA; D . in vivo RNA with no translating ribosomes ( Ksg treated cells ) ; E . in vitro modified refolded mRNA , to their TEs . For this and all subsequent panels , data calculated using different sets of ORFs are summarized in Supplementary file 1–3 . ( F ) Plot comparing computationally predicted mRNA structure ( - minimum free energy / nucleotide; -ΔG/nt ) of the entire ORF body to TE . ( G ) Plots of the difference in the Gini index between untranslated ( ksg-treated ) and translated in vivo mRNA against their TE for the 465 ORFs in both datasets . X-axis: Gini index ( in vivo untranslated ) – Gini index ( in vivo ) , normalized by the average of the two . DOI: http://dx . doi . org/10 . 7554/eLife . 22037 . 00610 . 7554/eLife . 22037 . 007Figure 3—figure supplement 1 . Correlation between the mRNA structural level and translation efficiency . ( A ) An example showing the results of mRNA-seq , ribosome profiling and DMS-seq of the rpsF-priB-rpsR-rplI operon , with translation efficiency ( TE ) and Gini index of each ORF indicated . ( B ) Plot of Gini index of unmodified mRNA ( using the DMS sequencing library preparation protocol but without DMS treatment ) calculated across the entire ORF body against TE . ( C ) Bar plot comparing the absolute value of Spearman’s rank correlation coefficient ( ρ ) between TE and mRNA structure of different portions of ORF mRNAs in vivo ( blue ) , in vivo no ribosome ( ksg-treated; green ) and in vitro ( red ) . The portions of ORFs analyzed are shown schematically with the black bars underneath the ORF . The actual correlation between TE and in vivo ORF-wide RNA structure is indicated . DOI: http://dx . doi . org/10 . 7554/eLife . 22037 . 007 Translation itself can influence mRNA structure as the helicase activity of translating ribosomes is likely to decrease the mRNA structure of highly translated ORFs more than that of poorly translated ORFs . We asked whether TE is correlated solely to the mRNA structure that results from ribosome unwinding or whether it is also correlated to the intrinsic mRNA structure that exists in the absence of translation . We find that when translation is inhibited in vivo ( e . g . following kasugamycin treatment ) , the absolute correlation of TE to structure remains high but decreases somewhat ( ρ = −0 . 58 , Figure 3D ) , and that there is a small further decrease in correlation when mRNAs are refolded in vitro ( ρ = −0 . 48 , Figure 3E ) . Additionally , computationally predicted structures of entire ORFs also show robust correlation to their TE’s ( ρ = −0 . 52 , Figure 3F ) . The results are very similar when we confine ourselves to the 421 ORFs with ≥15 DMS reads/nucleotide in all conditions ( Figure 4—figure supplement 2 ) . We further dissected the influence of translation on ORF mRNA structure by determining how the difference in Gini score of in vivo mRNA with and without translation is related to its TE . We found that there is a tendency for mRNA to be more structured ( higher Gini index ) in the absence of translation ( Figure 2E ) and that mRNAs with the highest TEs had the greatest difference in their Gini’s ( ρ = 0 . 52 , Figure 3G ) . These data are consistent with the idea that unwinding by ribosomes contributes to the in vivo structure of highly translated genes . The decreased correlation of untranslated ORF mRNA structure to TE may result from removing the contribution of unwinding by translating ribosomes . Previous work on ORF translatability has pointed to the important role of sequences around the ORF start site . Using the ORFs that are common in all datasets and separated by ≥20 nt from the upstream ORF , we examined the correlation of TE with the level of mRNA structure only around the start site or extending further into the ORF ( Figure 3—figure supplement 1C ) . We find increasing correlation with TE as successively larger regions of the ORF are considered in the structural analysis ( −20nt to 40nt , 0 to 60nt , and 0 to 100nt relative to the gene start ) . Notably , the correlation of TE with extent of structure in either the first or second halves of the ORF are very similar , and the highest correlation is with the Gini of the ORF-wide mRNA structure . In toto , these analyses indicate that the linear sequences of bacterial mRNAs encode not only ORFs , but also ORF-wide secondary structures . These structures provide a rough blueprint for the TE of that ORF . Instructions from this blueprint are augmented by ribosomes and additional factors ( see Discussion ) . We next examined the ability of the Shine-Dalgarno sequence and codon usage to predict TE . Data for all ORFs are presented in Figure 4 , and that for the 421 ORFs in common between conditions are presented in Figure 4—figure supplement 2 . 10 . 7554/eLife . 22037 . 008Figure 4 . Correlation of Other mRNA features with TE . ( A–C ) Plots comparing tAI ( tRNA adaptation index ) of the entire ORF against: A . translation efficiency ( TE , protein synthesis rate per mRNA ) ; B . protein synthesis rate ( average ribosome footprint density ) ; C . mRNA abundance ( RPKM mRNA sequencing ) of the ORF . For this and the following panels of this figure , the 1116 ORFs in the in vivo RNA DMS-seq dataset are analyzed ( Supplementary file 1 ) . ( D–F ) Plots comparing codon influence across the entire ORF defined from overexpressing exogenous genes ( Boël et al . , 2016 ) against: D . translation efficiency; E . protein synthesis rate; F . mRNA abundance of the ORF . ( G ) Average ribosome occupancy at leucine codons in endogeneous genes when overexpressing a control plasmid ( p-CTRL without a mini ORF ) or plasmids with a heterologous CUA mini-ORF ( p-CUA ) or a CUG mini-ORF ( p-CUG ) . The ribosome occupancy at each leucine codon was normalized by the average ribosome density of the ORF . The relative ribosome occupancy of that specific leucine codon was averaged across ORFs and normalized to that of the cells with control plasmid . ( H–I ) Gene expression changes with the control plasmid and heterologous overexpression of CUA codon mini-ORF ( H ) or CUG codon mini-ORF ( I ) . The average ribosome footprint density of individual genes ( see Materials and methods for details ) was plotted in log2 scale . DOI: http://dx . doi . org/10 . 7554/eLife . 22037 . 00810 . 7554/eLife . 22037 . 009Figure 4—source data 1 . Linear regression model to predict TE based on different mRNA features . A multiple linear regression model is applied to predict TE based on the following features: mRNA structure level of various portions of the ORF , codon usage predicted by tAI ( dos Reis et al . , 2004; Tuller et al . , 2010 ) , codon influence metric ( Boël et al . , 2016 ) , and the strength of Shine-Dalgarno sequence ( using the RBS Calculator established by Salis et al from https://github . com/hsalis/Ribosome-Binding-Site-Calculator-v1 . 0 ) . ( A ) Comparison between the experimentally measured TE and the model-predicted TE . The red dashed line indicates the y = x diagonal line . ( B ) Relative contribution of the factors in predicting TE , calculated from stepwise regression . Y-axis: R2 of different models with stepwise addition of individual factors . Asterisks indicate significant improvement of model ( based on ANOVA , with significance codes: 0 ‘***’ 0 . 001 ‘**’ 0 . 01 ‘*’ 0 . 05 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 22037 . 00910 . 7554/eLife . 22037 . 010Figure 4—figure supplement 1 . Effect of SD strength , tAI , and codon influence on predicting TE of endogenous genes . ( A ) Plot of predicted Shine-Dalgarno strength ( see Materials and methods ) against in vivo translation efficiency . Genes with Gini indices in a tight range ( 0 . 5–0 . 52 ) are indicated in cyan . ( B–C ) Plot comparing Gini indices of ORFs of in vivo mRNA against: B . protein synthesis rate ( average ribosome footprint density ) ; C . mRNA abundance ( RPKM mRNA sequencing ) of ORFs . For this and the following panels of this figure the 1116 ORFs in the in vivo RNA DMS-seq dataset are analyzed ( Supplementary file 1 ) . ( D–E ) Plots of computationally predicted mRNA structure ( -minimum free folding energy per nucleotide ) against: D . tAI ( tRNA adaptation index ) ; E . codon influence across the entire ORF body defined from overexpressing exogenous genes ( Boël et al . , 2016 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 22037 . 01010 . 7554/eLife . 22037 . 011Figure 4—figure supplement 2 . Comparison of the relative significance of different mRNA features in predicting TE . Top: Bar plot comparing the absolute values of Spearman’s rank correlation coefficient ( ρ ) in various conditions with Gini indices of the dataset for each condition ( gray bars ) , or for the 421 ORFs present in all three datasets ( blue bars ) . Error bars indicate the 95% confidence intervals estimated from the bootstrapping distribution of the correlation coefficients ( replicate number = 1000 ) . Data calculated using 421 common ORFs among conditions are summarized in Supplementary file 4 . Bottom: The table shows comparisons between the absolute values of Spearman’s ρ values for pairwise features for the dataset comprised of the 421 ORFs ( blue bars above ) . p-values were calculated from the K-S test comparing two bootstrapping distributions of ρ . DOI: http://dx . doi . org/10 . 7554/eLife . 22037 . 011 Consistent with earlier studies ( Li et al . , 2014 ) , we found that the strength of the Shine-Dalgarno sequence does not have predictive power for TE , even after controlling for structure as measured by Gini index ( Figure 4—figure supplement 1A ) . Codon usage , quantified by tAI ( tRNA adaptation index ) ( dos Reis et al . , 2004; Tuller et al . , 2010 ) modestly correlates with TE ( ρ = 0 . 34 , Figure 4A ) . Interestingly , codon usage correlates more strongly with the overall rate of translation ( i . e . average ribosome footprint density , ρ = 0 . 61 ) and ORF mRNA abundance ( RPKM of mRNA sequencing , ρ = 0 . 48 ) than with their TE’s ( Figure 4B–C ) . In contrast , the Gini score exhibits its highest correlation with TE ( ρ = −0 . 76 , Figure 3C ) and is poorly correlated with mRNA abundance ( ρ = −0 . 05 ) ( Figure 4—figure supplement 1C ) . This suggests that codon bias may be evolutionally selected to correspond to the ORF expression level rather than to its translation efficiency . Additionally , there is evidence that codon usage correlates with mRNA half-life in both eukaryotes and prokaryotes ( Boël et al . , 2016; Presnyak et al . , 2015 ) . ORF-wide codon usage ( tAI ) and intrinsic mRNA structure appear to be largely independent variables , as they show little correlation with each other ( Figure 4—figure supplement 1D ) . Although a novel metric quantifying codon influence was highly successful at predicting protein production from overexpressed exogenous genes transcribed by T7 RNA polymerase ( Boël et al . , 2016 ) , it is relatively weakly correlated with TE under physiological conditions for endogenous genes ( ρ = 0 . 29 ) ( Figure 4D ) . This suggests that the codons providing efficient translation of an over-expressed transgene may differ from the efficient codons for an endogenous gene , as overexpression causes amino acid starvation and concomitant alteration of charged tRNA pools ( Plotkin and Kudla , 2011; Welch et al . , 2009; Dittmar et al . , 2005; Elf et al . , 2003 ) . Overall , ORF-wide mRNA secondary structure is by far the strongest and most significant predictor of endogenous TE compared to the other factors discussed above ( Figure 4—figure supplement 2 ) . A linear regression model that includes the addition of the Boel metric , tAI , and Shine-Dalgarno sequence strength showed marginal improvement in the predictive power compared to the ORF-wide structure alone ( Figure 4—source data 1 ) . Therefore , rather than being a driver for TE , codon optimization may be critical for highly expressed genes due to higher demand for these tRNAs and may play a role in setting the appropriate mRNA half-life . Our results thus far indicate that the rules for endogenous translation differ from those for overexpressed genes , particularly in the role of codon choice . Considering the fact that the expression of each tRNA species is tuned to the endogenous usage of its cognate codon ( s ) ( Dong et al . , 1996 ) , overexpressed transgenes are likely to perturb the balance between codon usage and tRNA abundance , creating a global translation defect ( Shah et al . , 2013 ) . To directly test this hypothesis , we evaluated the effects of transgene overexpression containing one codon at a time . We constructed a synthetic gene with only one sense codon after the initiating codon and expressed this minimal ORF to directly assess the influence of a single tRNA and amino acid without additional complications from the protein product . When the minimal ORF contains the rare leucine codon CUA , which has only one cognate tRNA , we observed elevated ribosome occupancy at CUA codons in endogenous genes ( Figure 4G ) . In particular , slow translation at CUA codons in the leuL leader sequence triggers the expression of leucine biosynthetic genes ( Figure 4H ) , whereas overexpressing the minimal ORF with the common leucine codon CUG or without any coding sequence does not change the expression level or ribosome occupancy at leucine codons of endogenous genes ( Figure 4G and I ) . These results suggest that overexpression of a rare codon and not a common codon can deplete the pool of free cognate aa-tRNA molecules , leading to global perturbation of translation . Cells expressing a transgene that contain more rare codons are thus under a different physiological state compared to WT cells solely expressing endogenous genes . Bacterial operons are densely packed with ORFs , as the majority of adjacent ORFs ( 62% ) are separated by only 25nt or less ( Figure 5A ) . Our finding that ORF mRNAs have a roughly similar degree of structure ( Gini index ) throughout their entire length ( Figure 2A–D ) , but that the degree of structure of adjacent ORF mRNAs on polycistronic transcripts can differ significantly ( Figure 2G ) suggests that mRNA structure undergoes a sharp transition at ORF boundaries . 10 . 7554/eLife . 22037 . 012Figure 5 . ORFs are isolated from each other by forming ORF-specific RNA structures . ( A ) Cumulative distribution of spacing between adjacent ORFs within operons of E . coli . X-axis: distance from 3’ of the stop codon of upstream genes ( gene A ) to 5’ of the start codon of downstream genes ( gene B ) . ( B–D ) Correlation between local mRNA structure quantified by Gini index and TE of adjacent ORFs in the same operon . X-axis: distance from the 5’ of start codon of downstream ORFs ( gene B ) . Y-axis: the absolute value of correlation ( Spearman’s ρ ) of local Gini indices , calculated from DMS-seq of in vivo mRNA ( B ) , in vivo untranslated mRNA ( ksg-treated ) ( C ) or in vitro modified mRNA ( D ) , against TE of the upstream ( gene A; dashed line ) or the downstream ( gene B; solid line ) gene . Gini indices were calculated within 300 nt windows scanning across the boundary between adjacent ORFs within operons . The correlation to TE is plotted at the center of each 300 nt window . ( E ) Meta-gene analysis of mRNA structure in the vicinity of translation initiation sites . Structure was predicted by applying the DMS-seq constrained minimum free-energy model calculated from in vivo mRNA ( blue ) , in vivo untranslated mRNA ( ksg-treated; green ) or in vitro modified mRNA ( red ) . Mean predicted base-pairing probability of each nucleotide ( averaged across genes ) was plotted across the boundary between adjacent ORFs within operons . ( F ) Plot of directionality of RNA folding at ORF boundaries . At each position , the probability of base pairing with every other position was calculated for each ORF examined . The average sum probability of base-pairing with any nucleotide in a 60 nt window upstream and in a 60 nt window downstream was calculated . Y-axis: the ratio of the downstream base-pairing probability to the upstream base-pairing probability at each position ( X-axis ) . The black arrows indicate preferential folding direction . DOI: http://dx . doi . org/10 . 7554/eLife . 22037 . 01210 . 7554/eLife . 22037 . 013Figure 5—figure supplement 1 . Structural isolation between mRNA of adjacent ORFs on the same operons . ( A ) The absolute value of correlation ( Spearman's ρ ) between computationally predicted local mRNA structure at the ORF boundary , quantified by predicted ∆G of minimum free-energy structure , and the in vivo TE of the upstream or downstream ORF . ∆G index was calculated for 300 nt windows that scan gene bodies , using genes that extend through the 300 nt window being examined , and is plotted at the center of each window . ( B ) Meta-gene analysis of mRNA structure in the vicinity of translation initiation sites at the boundary of overlapping ( red ) or non-overlapping ( grey ) ORF pairs . Subsets of non-overlapping ORF pairs with the same number as overlapping ORF pairs were selected to make the TE ratio between adjacent ORFs indistinguishable between overlapping and selected non-overlapping ORF pairs . 20 random selections past the criteria were plotted in gray . For each time , mRNA structure was predicted by applying the in vivo DMS-seq constrained minimum free-energy model . Mean predicted base-pairing probability of each nucleotide was plotted across the boundary between adjacent ORFs within operons . Red arrow indicates a downstream shift of low base-pairing region due to ORF overlapping . ( C ) Plot of directionality of RNA folding at overlapping ( red ) or non-overlapping ( grey ) ORF boundaries . Subsets of non-overlapping ORF pairs were selected as for ( B ) . At each position , the probability of base pairing with every other position was calculated for each ORF examined . The average sum probability of base-pairing with any nucleotide in a 60 nt window upstream and in a 60 nt window downstream was calculated . Y-axis: the ratio of the downstream base-pairing probability to the upstream base-pairing probability at each position ( X-axis ) . The red arrows suggest that between overlapping ORFs , the mRNA folding insulation at ORF boundaries may be weaker and also shifted ( due to ORF overlapping ) compared to non-overlapping ORFs . DOI: http://dx . doi . org/10 . 7554/eLife . 22037 . 013 We examined the structural organization of mRNA at ORF boundaries in polycistronic mRNAs . We find that the local degree of mRNA folding immediately downstream of the start site correlates with the TE of the downstream gene , but that this correlation rapidly diminishes upstream of the start site . Conversely , local mRNA structure upstream of the start site is only correlated with the TE of the upstream ORF ( Figure 5B ) . This is true not only for mRNAs that are being translated ( WT cells; Figure 5B ) but also for untranslated mRNAs ( kasugamycin-treated cells; Figure 5C ) , in vitro refolded mRNAs ( Figure 5D ) and computationally predicted mRNA structures ( Figure 5—figure supplement 1A ) . Thus , mRNA structure undergoes a sharp transition at ORF boundaries , and polycistronic mRNAs consist of distinct ORF-length structural domains . The close packing of ORF mRNAs raises the issue of how they maintain distinct structural domains , and suggests that bacterial ORFs may be marked not only by start and stop codons , but also by features that assist within-ORF mRNA folding . To investigate this , we computationally predicted the structure of mRNA extending −250 to +250 nt from the translation start at the boundary of adjacent ORF pairs within the same operon . Because folding algorithms often predict a large ensemble of possible folds for a long stretch of RNA , we constrained the predictions by forcing positions that were highly DMS-modified to be unpaired in the predicted structures . Consistent with previous studies ( Eyre-Walker and Bulmer , 1993; Scharff et al . , 2011; Bentele et al . , 2013; Del Campo et al . , 2015 ) , we found a lack of structure in the immediate vicinity of the start sites for most ORFs ( Figure 5E ) . Downstream from this structure-free zone ( 25–50 nt ) , endogenous mRNA has a high propensity to base pair with regions further downstream , that is pairing within the same ORF ( Figure 5F ) . Conversely , nucleotides located 25-50nt upstream of the start site have a strong preference for interacting with regions further upstream in the preceding ORF ( Figure 5F ) . Importantly , in vivo mRNA without translating ribosomes and in vitro probed mRNA ( Figure 5E–F ) also showed such preferences . Thus , the sharp transition in the directionality of base-pairing around start sites is driven by the mRNA sequence itself , promoting ORF-centric units of secondary structure . We experimentally investigated the effects of disrupting a region that promotes independent mRNA folding within adjacent ORFs . The dusB-fis operon is composed of a highly structured upstream gene ( dusB ) and a poorly structured downstream gene ( fis ) separated by 25 nucleotides . The two ORFs have an ~100 fold difference in TE ( Figure 6A ) . Previous work indicated that the upstream dusB gene has a stem-loop structure near the 3’ end of the gene; that mutationally disrupting the stem-loop ( Mutation M3; Figure 6A ) decreased translation of fis; and that restoring base pairing by a second mutation ( M2 ) restored fis translation for unknown reasons ( Nafissi et al . , 2012 ) . After confirming these results ( Figure 6B ) , we performed DMS-seq on WT and mutant cells to determine whether destroying the stem-loop decreased fis translation by reducing the structural isolation of dusB and fis . 10 . 7554/eLife . 22037 . 014Figure 6 . Disruption of structural isolation between dusB and fis affects fis translation . ( A ) mRNA structure at the 3’ end of dusB , with mutations M3 and M2 indicated . Translation efficiencies ( TEs ) of dusB and fis in WT cells are 0 . 02 and 2 . 06 , respectively . ( B ) The dusB-M3 mutation decreases Fis expression and is rescued by the complementary M2 mutation . Western blot compares Fis protein amounts in WT , dusB-M3 and dusB-M3:2 double mutant cells , with RpoB protein as an internal control . ( C ) Scatter plots comparing Gini indices of ORFs in WT cells to those in dusB-M3 or in dusB-M3:2 double mutant cells . Outlier test: fis , Bonferonni p-value=1 . 02e−05 ( dusB-M3 ) ; p-value>0 . 05 ( dusB-M3:2 ) . ( D ) Normalized DMS-seq signals at the dusB-fis boundary region from different samples as indicated . Positions of M3 and M2 are highlighted , with asterisks indicating mutated nucleotides . X-axis: distance from 5’ end of the fis start codon . Y-axis: normalized DMS-seq signals . Dashed line: threshold ( 0 . 2 ) above which the A/C bases are predicted to be unpaired ( see Materials and methods ) . ( E ) mRNA structure at the dusB-fis boundary region of WT or dusB-M3:2 cells , predicated by constraining a minimum free-energy model with DMS-seq measurements . Locations of mutations M3 and M2 are as indicated . ( F ) mRNA structure at the dusB-fis boundary region of dusB-M3 mutant cells , predicated by constraining a minimum free-energy model with DMS-seq measurements . CGG residues labeled with asterisks indicate the M3 mutation . DOI: http://dx . doi . org/10 . 7554/eLife . 22037 . 01410 . 7554/eLife . 22037 . 015Figure 6—figure supplement 1 . mRNA secondary structure at the dusB-fis boundary region of WT cells and dusB mutants . Gini indices calculating within 100 nt rolling windows ( plotted at the center of windows ) at the boundary region of dusB-fis operon . X-axis: distance from the 5’ of fis start codon . Different samples are color-coded as indicated . DOI: http://dx . doi . org/10 . 7554/eLife . 22037 . 015 A model of the structure of the dusB-fis interface constrained by DMS-seq data ( Figure 6D ) indicates that the dusB and fis ORFs are structurally distinct in WT and double mutant ( M3/M2 or M3:2 cells ) ( Figure 6E ) , but that M3 increases the structure of fis mRNA ( Figure 6C and Figure 6—figure supplement 1 ) . In the M3 mutant , the -58 ~ −53nt region ( blue ) pairs with the +9 ~ +14 nt region of fis ( red ) , rather than forming a stem-loop structure within dusB as it does in WT and M3:2 cells ( Figure 6E–F ) . The increased structure of fis mRNA in the M3 mutant starts at the ORF boundary and spreads across the entire downstream coding region of fis ( Figure 6—figure supplement 1 ) . Thus , mutation M3 induces long-range interactions between mRNA of the dusB and fis ORFs , which are normally structurally insulated from each other . In toto , our results suggest that specific sequences isolate mRNA folding within adjacent ORFs thereby minimizing structural crosstalk between adjacent ORFs . Disruption of structural boundaries affects both local and long-range mRNA folding , which is likely to be critical for programming the degree of translational isolation between ORFs on the same mRNA .
Translation is a highly controlled process in bacteria , making it critical to understand the mRNA features contributing to differential translatability . Numerous studies have investigated the important question of which features control protein production from overexpressed , foreign ORF mRNAs , identifying codon usage and local structure around the translation start site as key variables . However , these studies have left open the question of which mRNA features regulate endogenous translation . The importance of this question is highlighted by the observation that the rate of protein production from each ORF in a polycistronic mRNA can vary as much as 100-fold . Our global study now examines this issue . Our principal finding is that ORF mRNAs have modular structures within polycistronic mRNAs and that ORF-wide mRNA structure rather than codon usage correlates most strongly with the translation efficiency of endogenous ORFs . Our analysis of mRNA structure revealed the unanticipated finding that operonic mRNAs have modular structures . Each ORF mRNA in the operon has a characteristic degree of structure , with highly correlated Gini scores between their first and second halves . This correlation persists in the absence of translation , when mRNAs are refolded in vitro and when structure is determined computationally ( Figure 2A–D ) . In stark contrast , there is little correlation between the extent of structure in adjacent ORFs ( Figure 2G ) . Additionally , and consistent with earlier computational findings , we observe a small ~25 nucleotide region beginning at the translation start that is more unstructured than the remainder of the ORF ( Figure 5E ) . Thus , polycistronic mRNAs consist of a series of ORF-wide modules each with characteristic but different extents of structure , punctuated by regions of low basepairing at the translation start site ( Figure 7 ) . Maintenance of a common degree of structure throughout an ORF suggests that this parameter , like reduced structure at the start of ORFs , is a selective force in the evolution of ORF sequence , providing yet another constraint on mRNA sequence beyond codon adaptation ( Sharp and Li , 1987 ) . 10 . 7554/eLife . 22037 . 016Figure 7 . Model of operon mRNA structural organization . Polycistronic mRNAs are organized into ORF-centric modules with characteristic but different extents of mRNA structure , punctuated by regions of low basepairing close to the translational start site ( A ) . The intrinsic ORF-wide mRNA structure is highly predictive of translation efficiency ( B ) , and is amplified by translation , in a self-reinforcing loop , to provide the mRNA structure that ultimately specifies the translation of each ORF in an operon . DOI: http://dx . doi . org/10 . 7554/eLife . 22037 . 016 We find that the TE of each ORF correlates very highly , and most strongly with the ORF-wide extent of mRNA structure . We have begun to deconvolute the ‘chicken and egg’ problem of whether mRNA structure is a cause or a consequence of translation by examining the correlation of TE to ORF-wide structure when translation is inhibited . This removes the ribosome contribution but retains vectorial folding , RNA binding proteins and in vivo concentrations of salts and macromolecules . Untranslated mRNA structure is highly correlated with TE but less so than translated mRNA ( Figure 3C–D ) . Moreover , the difference in the mRNA structure of an ORF with and without translation is highly correlated to its TE ( Figure 3G ) . Thus , poorly translated mRNAs , have virtually identical extents of structure with and without translation , but more highly translated RNAs become increasingly more unstructured . Finally , computationally predicted structures or those obtained from in vitro refolded mRNAs correlate somewhat more poorly with TE ( ρ = −0 . 52 or −0 . 48 respectively ) than the structure of untranslated in vivo mRNA ( ρ = −0 . 58 ) ( Figure 3D–F ) . This suggests that features of the in vivo cell , besides translation by the ribosome , may also affect mRNA structure . Taken together , these results suggest that the intrinsic mRNA sequence itself encodes a rough blueprint for the ORF-centric mRNA structures that are predictive of TE . These structures are then amplified by translation and other features of the living cell , in a self-reinforcing loop , to provide the structure that ultimately specifies the translation of each ORF . Interestingly , E . coli has chosen to insert a predominance of its low TE ORFs into operons where adjacent genes have moderately high TE’s ( for example , the rpsF-priB-rpsR-rplI operon shown in Figure 3—figure supplement 1A ) . All of its 10 lowest TE ORFs , and 86% of its ORFs in the bottom 10% of TEs are located in operons , compared with 58% of all ORFs . The evolutionary advantage of this arrangement is not known , but may relate either to decreasing transcriptional noise or to mRNA stabilization . The necessity for achieving widely different TEs for adjacent ORFs in operons may have driven the evolution of the ORF-centric mRNA folding strategy . As the translation termination codon of most ORFs is separated by less than 25nt of untranslated mRNA from the start site of the downstream ORF , the abundant ribosomes of the highly translated ORF could transiently open the structure of the poorly translated ORF and increase the accessibility of its start site . The propensity for in-ORF mRNA folding at both the beginning and ends of ORFs may prevent the upstream ORF from influencing the structure and hence TE of the downstream ORF , effectively insulating each ORF from its neighbors . We have identified small regions , located about 25–50 nucleotides both downstream and upstream of ORF translation start sites that preferentially pair within their ORFs . These regions may reinforce the folding barriers between adjacent ORF mRNAs , as we demonstrated for the dusB-fis operon . Interestingly , RNA polymerase pausing is enriched at translation start sites ( Larson et al . , 2014 ) and this may reinforce ORF-centric structural insulation by allowing ORFs to fold independently during the pioneer round of translation . It is likely that the extent to which adjacent ORFs are insulated has been tuned . Approximately 15% of ORF pairs have overlapping stop and start codons and translational coupling has been demonstrated in some cases ( Aksoy et al . , 1984; Oppenheim and Yanofsky , 1980; Schümperli et al . , 1982; Yates and Nomura , 1981 ) . This overlap may enable upstream ribosomes to influence downstream ORF translation by unwinding mRNA structure , thereby promoting translational coupling . Indeed , it is likely that the propensity for in-ORF basepairing is slightly weaker for overlapping ORF pairs than for non-overlapping ORF pairs ( Figure 5—figure supplement 1B–C ) . The precise role of modular ORF structures that provide a rough blueprint for TE has not yet been established . It is certainly possible that the mRNA structure of an entire or a significant fraction of the ORF is required to define translation initiation , as has been demonstrated experimentally for rimM ( Wikström et al . , 1992 ) . Alternatively , a constant degree of ORF-wide mRNA structure may be the most robust way to ensure the appropriate amount of mRNA structure around the translation initiation site . In support of this idea , a recent study using in situ codon mutation of the E . coli essential gene infA showed that mutations of codons even far downstream from the start of the gene can be deleterious if they disrupt the native 5’ RNA conformation via long-range structural interactions predicted computationally ( Kelsic et al . , 2016 ) . ORF-wide structures may also play additional roles . For high TE ( poorly structured ) ORFs , extended lack of structure may provide the landing pad necessary to capture a large pool of non-specifically bound 30S subunits to wait for opening of the SD and start codon , the so-called ‘standby model’ of translation initiation ( Adhin and van Duin , 1990; de Smit and van Duin , 2003 ) . Additionally , the ORF-centric mRNA folding strategy may have been driven by the necessity for adjacent ORFs to have discrete , often significantly different TEs . Finally , ORF-wide mRNA structures may help set the rate of endonucleolytic cleavage . The function of these modules is an important area for future inquiry . Although the TE of endogenous ORFs is primarily predicted by the extent of its mRNA structure , translatability of overexpressed foreign ORFs appears to be strongly driven by codon usage and tRNA limitation . This difference may arise from the fact that codon usage and tRNA abundance are largely balanced under physiological conditions , but become imbalanced when foreign ORFs are overexpressed , and we have directly demonstrated that this is the case ( Figure 4G–I ) . This suggests that synthetic biologists and the cell tune translation in different ways . However , synthetic biologists struggle to robustly program differential translation of ORFs on the same mRNA . Our finding that polycistronic mRNAs consist of ORF-wide modules with set amounts of structure that are insulated from their neighbors may be key to this issue . Design approaches that incorporate appropriate mRNA structures may have the potential to produce the finely tuned synthesis rates observed in natural operons .
E . coli K-12 MG1655 ( RRID:SCR_002433 ) was used as the WT strain . All culture experiments were performed in MOPS medium supplemented with 0 . 2% glucose , all amino acids except methionine , vitamins , bases and micronutrients ( Teknova , Hollister CA ) . Cells were grown in an overnight liquid culture at 37°C , diluted to an OD420 = 0 . 001 in fresh medium and grown until OD420 reached 0 . 4 where samples were collected . Multiple deletion strains were generated by transduction of FRT-flanked deletion alleles from the Keio collection ( Baba et al . , 2006 ) followed by marker excision by Flp recombinase ( Cherepanov and Wackernagel , 1995 ) . All major experiments were biologically repeated for at least twice ( see raw data files for sequencing data ) . In the experiment testing the effects of overexpressing the rare CUA leucine codon and the common CUG leucine codon , plasmids with pBR322 origin of replication was constructed to have a mini ORF ATGCTATAA or ATGCTGTAA driven by an IPTG-inducible promoter . The plasmid also contains lacIq to increase the expression of lac repressor . MG1655 containing the control plasmid ( without mini ORFs ) and MG1655 containing the plasmid with CUA or CUG mini ORF were grown overnight in MOPS rich glucose medium with 100 µg/ml ampicillin , diluted 1:1000 into 250 ml pre-warmed fresh medium containing 1 mM IPTG next morning . Cells were grown at 200 rpm at 37°C and harvested when OD600 reached 0 . 3 by vacuum filtration . The protocol for bacterial ribosome profiling with flash freezing was described ( Li et al . , 2014 ) . Briefly , 200 mL of cell culture were filtered rapidly and the resulting cell pellet was flash-frozen in liquid nitrogen and combined with 650 µL of frozen lysis buffer ( 10 mM MgCl2 , 100 mM NH4Cl , 20 mM Tris-HCl pH 8 . 0 , 0 . 1% Nonidet P40 , 0 . 4% Triton X-100 , 100 U/mL DNase I ( Roche , St . Louis MO ) , 1 mM chloramphenicol ) . Cells were pulverized in 10 mL canisters pre-chilled in liquid nitrogen . Lysate containing 0 . 5 mg of RNA was digested for 1 hr with 750 U of micrococcal nuclease ( Roche ) at 25°C . The ribosome-protected RNA fragments were isolated using a sucrose gradient followed by hot acid phenol extraction . Library generation was performed using the previously described strategy ( Li et al . , 2014 ) detailed below . For experiments performed in parallel with ribosome profiling , total RNA was phenol extracted from the same lysate that was used for ribosome footprinting . For experiments performed independently of ribosome profilng experiments , and for total mRNA used for in vitro DMS-seq experiments , 4 mL of OD420 = 0 . 4 culture was added to 500 µL of ice-cold stop solution ( 475 µL of 100% EtOH and 25 µL acid phenol ) , vortexed , spun for 2 min at 8000 rpm , and the cell pellet was flash frozen in liquid nitrogen . Total RNA was then hot acid phenol extracted . For mRNA-seq experiments , ribosomal RNA and small RNA were removed from the total RNA with MICROBExpress ( Ambion , Grand Island NY ) or Ribozero ( Epicenter , Madison WI ) and MEGAclear ( Ambion ) , respectively . mRNA was randomly fragmented as described ( Ingolia et al . , 2009 ) . The fragmented mRNA sample was converted to a complementary DNA library with the same strategy as for ribosome footprints . The footprints and mRNA fragments were ligated to miRNA cloning linker-1 ( IDT ) 5rApp/CTGTAGGCACCATCAAT/3ddC/ , using a recombinantly expressed truncated T4 RNA ligase 2 K227Q produced in our laboratory . The ligated RNA fragments were reverse transcribed using the primer 5'/5Phos/GATCGTCGGACTGTAGAACTCTGAACCTGTCGGTGGTCGCC GTATCATT/iSp18/CACTCA/iSp18/CAAGCAGAAGACGGCATACGAATTGATG GTGCCTACAG 3' . The resulting cDNA was circularized with CircLigase ( Epicentre ) , and PCR amplification was done as described previously ( Ingolia et al . , 2009 ) . For in vivo DMS modification , 15 mL of exponentially growing E . coli were incubated with 750 µL DMS . Incubation was performed for 2 min at 37°C . For kasugamycin ( ksg ) experiments , ksg was added to a final concentration of 10 mg/mL to ΔgcvB cells for 2 min at 37°C prior to DMS modification . Untreated ΔgcvB cells were also modified by DMS and collected in parallel as control . DMS was quenched by adding 30 mL 0°C stop solution ( 30% β-mercaptoethanol , 25% isoamyl alcohol ) , after which cells were quickly put on ice , collected by centrifugation at 8000 g and 4°C for 2 min , and washed with 8 mL 30% BME solution . Cells were then resuspended in 450 µL total RNA lysis buffer ( 10 mM EDTA , 50 mM sodium acetate pH 5 . 5 ) , and total RNA was purified with hot acid phenol ( Ambion ) . For in vitro DMS modifications , mRNA was collected from cells that were not treated with DMS . Two micrograms of mRNA was denatured at 95°C for 2 min , cooled on ice and refolded in 90 µL RNA folding buffer ( 10 mM Tris pH 8 . 0 , 100 mM NaCl , 6 mM MgCl2 ) at 37°C for 30 min then incubated in either . 2% DMS for 1 min ( 95°C ) or 4% DMS for 5 min ( 37°C ) . The DMS reaction was quenched using 30% BME , 0 . 3 M sodium acetate pH 5 . 5 and precipitated with 2 µL GlycoBlue and 1X volume of isopropanol . Sequencing libraries were prepared as described ( Rouskin et al . , 2014 ) . Specifically , DMS treated mRNA samples were denatured for 2 min at 95°C and fragmented at 95°C for 2 min in 1x RNA fragmentation buffer ( Zn2+ based , Ambion ) . The reaction was stopped by adding 1/10 vol of 10X Stop solution ( Ambion ) and quickly placed on ice . The fragmented RNA was run on a 10% TBU ( Tris borate urea ) gel for 60 min . Fragments of 60–70 nucleotides in size were visualized by blue light ( Invitrogen , Carlsbad CA ) and excised . Reverse transcription was performed in a 20 µL volume at 52°C using Superscript III ( Invitrogen ) , and truncated reverse transcription products of 25–45 nucleotides were extracted by gel purification . Sequencing was performed on an Illumina HiSeq 2000 or 4000 . Sequence alignment with Bowtie v . 0 . 12 . 0 mapped the footprint data to the reference genomes NC_000913 . fna obtained from the NCBI Reference Sequence Bank . Sequencing data from mutated strains were aligned to appropriately modified genome . For ribosome footprint and mRNA-seq samples , the center residues that were at least 12 nucleotides from either end were given a score of 1/N in which N equals the number of positions leftover after discarding the 5' and 3' ends . For DMS-seq samples , read counts were assigned to the base immediately 5' of the 5' end of each read , which is the base that was DMS-modified . Data analysis was performed with custom scripts written for R 2 . 15 . 2 and Python 2 . 6 . 6 . To calculate mRNA abundance , the number of mRNA sequencing reads mapped to a gene , following a Winsorization applied to trim the top and bottom 5% of reads , was divided by the length of the gene to yield the number of reads corresponding to the message per thousand bases of message per million sequencing reads ( RPKM ) . The protein synthesis rate of individual ORFs was measured by average ribosome footprint density of the ORF calculated as described in ( Li et al . , 2014 ) . First , genes with less than 128 reads mapped and genes with unconventional translation events were excluded from the analysis , which include ( 1 ) genes encoding selenoproteins ( e . g . fdhF , fdoG , fdnG ) ; ( 2 ) proteins with nearly identical coding sequences ( e . g . gadA and gadB , ynaE and ydfK , ldrA and ldrC , ybfD and yhhI , tfaR and tfaQ , rzoD and rzoR , pinR and pinQ ) . Second , sequencing reads from ribosome profiling mapped to the first and last five codons of the gene were excluded to remove effects of translation and termination . Third , correction for the variations in translation elongation rate was done in three steps as described in Li et al . ( 2014 ) : ( 1 ) correcting for the elevated ribosome footprint density observed for the first 50–100 codons ( Oh et al . , 2011 ) ; ( 2 ) correcting for the elevated density at the ribosomal anti-Shine-Dalgarno ( aSD ) site ( Li et al . , 2012 ) ; ( 3 ) correcting for other possible ribosome pausing using 90% Winsorization , by removing the top and bottom 5% of the ribosome profiling signal for each gene . Finally , the average ribosome footprint density of a gene was calculated by dividing the corrected number of mapped ribosome footprint reads by the corrected length of the gene . Translation efficiency of a gene was calculated by normalizing the average ribosome footprint density by the mRNA abundance of the gene ( defined above ) . The average ribosome footprint density ( i . e . protein synthesis rate ) , mRNA abundance , and translation efficiency of genes from different samples are listed in Supplementary file 1–4 . For identification of unpaired bases , raw DMS-seq data was normalized to the most highly reactive residue after removing outliers by 95% Winsorisation ( all data above the 95th percentile is set to the 95th percentile ) . Bases with DMS-seq signal greater that 20% of the signal on the most highly reactive residue ( after normalization ) were called ‘unpaired’ . For determination of rimM mRNA structures , a Viennafold ( Hofacker , 2003 ) ( http://rna . tbi . univie . ac . at/ ) minimum free-energy model of the indicated region was generated , constrained by bases experimentally determined to be unpaired in the indicated dataset . Color-coding by DMS signal was done using VARNA ( http://varna . lri . fr/ ) . The secondary structure models for E . coli ribosomal RNAs were downloaded from Comparative RNA Website and Project database ( http://www . rna . icmb . utexas . edu/DAT/3C/Structure/index . php ) . The crystal structure model was downloaded from Protein Data Bank ( http://www . pdb . org , PDB entries 3I1M , 3I1N , 3I1O , and 3I1P ) . The solvent-accessible surface area was calculated in PyMOL , and DMS was modeled as a sphere with 2 . 5 Å radius ( representing a conservative estimate for accessibility because DMS is a flat molecule ) . Accessible residues were defined as residues with solvent accessibility area of greater than 2 Å2 . Unpaired residues in DMS-seq data were identified as described above . True positive bases were defined as bases that are both unpaired in the secondary structure model and solvent-accessible in the crystal structure model . True negative bases were defined as bases than are paired ( A-U or C-G specifically ) in the secondary structure model . Accuracy was calculated as the number of true positive bases plus the number of true negative bases divided by all tested bases . The R package ‘ineq’ ( https://github . com/cran/ineq ) was used to calculate Gini indices over As and Cs in the region specified for each experiment . For each DMS-seq sample , Gini indices were calculated only for genes that had greater than an average of 15 reads per nucleotide across the ORF . Genes with discontinuous mRNA-seq reads ( due to an early termination event or an internal promoter , 1% of genes ) were excluded from the analysis . Specifically , Gini indices were calculated on mRNA-seq data , and a cutoff was created based on two standard deviations from the mean . The Gini indices of genes from different samples were listed in Supplementary file 1–4 . Adjacent ORFs in annotated operons often have differing levels of mRNA-seq reads , suggesting that they are not always on the same mRNA molecule . To identify adjacent ORFs expressed as a single operon , we assessed mRNA-seq data for equivalent mean message level and for signal continuity , as described below . Equivalent mean message level was assessed by first determining the variability in mean mRNA-seq read density within individual ORFs . There is a single transcript that extends over the entire body of the large majority of ORFs , and so the variability in mean read density level in the first half of each ORF was compared to mean read density in the second half of each ORF , and the variability in this distribution was used to define a cut-off for ORFs on a single message . Adjacent ORFs that fell within a 2σ cutoff in mean level ( calculated to be a 1 . 5-fold difference in mRNA level ) were determined to have equivalent mRNA level and were then assessed for signal continuity . Signal continuity was assessed by first determining the distribution of read density in windows within messages . Gini index of mRNA-seq signal was calculated for all 80nt windows within ORF bodies , and the variability in this distribution was again used to define a cutoff for continuous mRNA regions . Gini index were then calculated for 80nt windows tiling the region between adjacent ORFs . Gene pairs that fell within a 2σ cutoff defined by the intra-ORF distribution were considered to be a pair of adjacent ORFs on a single message . To determine the directionality of mRNA base pairing at ORF boundaries , sequence from −250 to +250 nt relative to the translation start site of the downstream gene was extracted for each adjacent pair of ORFs . A Viennafold ( Hofacker , 2003 ) ( http://rna . tbi . univie . ac . at/ ) minimum free-energy model of each 500nt sequence was generated ( constrained by DMS-seq data ) . The predicted probability of each base interacting with every other base in each mRNA structure model was then extracted from the Viennafold output . For each position , the probability of that position base pairing with any position within the upstream or downstream 60nt was then calculated . The ratio between summed upstream over downstream interaction probability across all mRNAs was then calculated for each position . 1 µC of Perkin Elmer EasyTag 35S labeled methionine ( Product # NEG709A ) was mixed with 5 µL 288 µmol unlabeled methionine and 24 µL media . At the time of capture , 900 µL of culture was added to methionine mix , and was labeled on a shaker for 1 min at 37°C . After labeling , 100 µL of ice-cold 50% trichloracetic acid ( TCA ) was added to the sample , which was vortexed and placed on ice for at least 20 min to allow precipitation . Samples were then counted by running 100 µL of sample through a 25 mm APFC glass fiber filter ( Millipore APFC02500 , Hainesport NJ ) pre-wetted with 750 µL of 5% TCA on a vacuum stand , and washing three times with 750 µL 5% TCA and three times with 750 µL 100% ethanol . Filters were then placed in MP Ecolume scintillation fluid and counted . We used the RBS Calculator established by Salis et al downloaded from http://www . github . com/hsalis/Ribosome-Binding-Site-Calculator-v1 . 0 to predict the strength of Shine-Dalgarno sequence . The measurement of tAI ( tRNA adaptation index ) was adapted from the previous works ( Tuller et al . , 2010; dos Reis et al . , 2004 ) , which gauges the availability of tRNAs for each codon within a gene . tAI incorporates different efficiency weights of the wobble interactions between codons and anticodons , with wi is defined as the relative adaptiveness value of codon i of a gene ( Tuller et al . , 2010 ) . The final tAI of a gene is the geometric mean of all its codons as shown below . tAI= ( ∏k=1Lωikωik ) 1/L ik is the kth codon of the gene and L is the length of the gene ( excluding start and stop codons ) . Wild-type , dusB-M3 , and dusB-M3:2 cells were grown in MOPS rich medium at 37°C till log phase ( OD420 ~0 . 3 ) . 1 mL cells were collected , resuspended in 30 µL SDS loading buffer , and boiled for 5 min . 10 µL of cell lysate was subject to Blot 12% Bis-Tris plus gel ( ThermoFisher scientific , Grand Island NY ) . Proteins were transferred to a nitrocellulose membrane ( BIO-RAD , Hercules CA ) . The membrane was first incubated with rabbit polyclonal anti-Fis antibody ( a kind gift from Dr . Reid C . Johnson at UCLA ) and mouse monoclonal anti-RNAP β subunit antibody ( abcam , Cambridge MA ) , and then incubated with goat anti-rabbit IgG IRDye 800CW and anti-mouse IgG IRDye 680RD secondary antibodies ( LI-COR , Lincoln NE ) . The blots were visualized and quantified by an Odyssey imaging system ( LI-COR ) . The amount of Fis protein in each sample was normalized against the amount of RNAP β subunit in the same sample . All the processed and raw datasets of sequencing experiments were uploaded to NCBI GEO database with accession number GSE77617 . | Proteins make up much of the biological machinery inside cells and perform the essential tasks needed to keep each cell alive . Cells contain thousands of different proteins and the instructions needed to build each protein are encoded in genes . However , these instructions cannot be used directly to manufacture the proteins . Instead , a messenger molecule called mRNA is needed to carry the information stored within genes to the parts of the cell where proteins are made . In bacteria , one mRNA molecule can include information from several genes . This group of genes is called an operon and produces a set of proteins that perform a shared task . Although these proteins work together , some of them are needed in greater numbers than others . Because they are all made using information from the same mRNA , some instructions on the mRNA must be read more times than others . It is unclear how bacterial cells control how many proteins are produced from each part of one mRNA but it is thought to relate to the three-dimensional shape of the molecule itself . Burkhardt , Rouskin , Zhang et al . have now examined the production of proteins from mRNAs in the commonly studied bacterium , Escherichia coli . The results showed that each set of instructions on the mRNA formed a three-dimensional structure that corresponds to the amount of protein produced from that portion of the mRNA . When this three-dimensional structure is more stable or rigid , the corresponding instructions tended to produce fewer proteins than if the structure was relatively simple and unstable . Further investigation showed that these three-dimensional mRNA structures could form spontaneously outside of cells , suggesting that molecules other than the mRNA itself have a relatively small role in controlling the number of proteins produced . This also suggests that the entire structure of each mRNA is important and is likely to be essential for cell survival . The next step is to understand why bacteria organise their genes in this way and how the different mRNA structures control how proteins are produced . Moreover , because many bacteria are used like biological factories to produce a variety of commercially useful molecules , these new insights have the potential to enhance a number of manufacturing processes . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"computational",
"and",
"systems",
"biology",
"microbiology",
"and",
"infectious",
"disease"
] | 2017 | Operon mRNAs are organized into ORF-centric structures that predict translation efficiency |
The endoplasmic reticulum ( ER ) -localized Hsp70 chaperone BiP affects protein folding homeostasis and the response to ER stress . Reversible inactivating covalent modification of BiP is believed to contribute to the balance between chaperones and unfolded ER proteins , but the nature of this modification has so far been hinted at indirectly . We report that deletion of FICD , a gene encoding an ER-localized AMPylating enzyme , abolished detectable modification of endogenous BiP enhancing ER buffering of unfolded protein stress in mammalian cells , whilst deregulated FICD activity had the opposite effect . In vitro , FICD AMPylated BiP to completion on a single residue , Thr518 . AMPylation increased , in a strictly FICD-dependent manner , as the flux of proteins entering the ER was attenuated in vivo . In vitro , Thr518 AMPylation enhanced peptide dissociation from BiP 6-fold and abolished stimulation of ATP hydrolysis by J-domain cofactor . These findings expose the molecular basis for covalent inactivation of BiP .
Protein folding homeostasis in the endoplasmic reticulum ( ER ) is defended by signal transduction pathways that match the complement of chaperones and enzymes to the burden of unfolded protein within the compartment . Transcriptional activation of genes that enhance the capacity of the ER to process its clients and regulated translation initiation , which controls the flux of unfolded proteins into the ER , comprise the unfolded protein response ( UPR ) vital to the well-being of cells , tissues and organs ( Balch et al . , 2008; Walter and Ron , 2011 ) . Acting alongside this coherent UPR are rapid , activity-dependent post-translational changes in the disposition of the major ER chaperone BiP . Given the dominant role of BiP in protein folding homeostasis in the ER , the latter stands to have considerable biological significance . The UPR regulates the abundance of BiP transcriptionally ( Chang et al . , 1989; Kozutsumi et al . , 1988 ) , but this is a latent homeostatic process manifesting over hours and days . On a much shorter time scale BiP’s oligomeric state is observed to change , with fewer oligomers present as levels of unfolded protein increase . The architecture of BiP oligomers is consistent with their role as a rapidly accessible repository of inactive BiP that the cell may draw upon in short notice to cope with rapid fluctuations in unfolded protein load ( Preissler et al . , 2015 ) . BiP is also subject to activity-dependent post-translational modification ( s ) . This is reflected in transfer of metabolic label from intracellular pools of tritiated adenosine and 32P phosphate onto BiP , covalently modifying the protein ( Carlsson and Lazarides , 1983; Hendershot et al . , 1988 ) and imparting upon it a lower isoelectric point ( pI ) ( Carlsson and Lazarides , 1983; Laitusis et al . , 1999 ) . These covalent transformation ( s ) of BiP correlate inversely with the burden of unfolded proteins in the ER ( Chambers et al . , 2012; Laitusis et al . , 1999; Leno and Ledford , 1989 ) and because the modified form of BiP was under-represented in complex with substrates ( Hendershot et al . , 1988 ) , were deemed to reflect inactivating modification ( s ) of the chaperone ( Chambers et al . , 2012 ) . Despite considerable effort , the chemical nature of the modification ( s ) was never directly ascertained . ADP-ribosylation seemed a good candidate as it fits the labeling profile ( comprised of adenosine and phosphate ) , the acidic nature of the modification and its susceptibility to the effect of the ADP-ribosylation inhibitor novobiocin ( Laitusis et al . , 1999 ) . Early work even suggested the presence of an enzymatic activity in cell lysates that could transfer a label from 32P NAD+ onto BiP in vitro ( Carlsson and Lazarides , 1983 ) ; but molecular characterization of this enzymatic activity proved elusive . A consensus emerged in regards to the region of BiP that bears the modification , as both 32P orthophosphate and 3H adenosine labeling mapped consistently to a cyanogen bromide ( CnBr ) cleavage fragment in the C-terminal substrate binding domain of BiP ( Thr434 to Met541 ) ( Chambers et al . , 2012; Gaut , 1997 ) . Mutations in two arginine residues on that fragment ( Arg470 and Arg492 ) markedly attenuated radiolabeling in vivo of FLAG-tagged BiP expressed in cells , suggesting that they might be the modification sites ( Chambers et al . , 2012 ) , but peptides with a mass consistent with ADP-ribosylated BiP were never uncovered and the assignments thus remained tentative . Recently it has been reported that BiP is subject to a different modification , AMPylation , which results in the formation of an phosphodiester bond between the alpha phosphate of ATP and a hydroxyl amino acid side chain ( with release of pyrophosphate ) . This modification of BiP is effected by a broadly conserved , ER-localized enzyme , FICD ( also known as HYPE ) ( Ham et al . , 2014; Sanyal et al . , 2015 ) . There is lack of unanimity in regards to the circumstances and consequences of BiP AMPylation: the Orth lab provides evidence that AMPylation is an inactivation modification , induced when client protein burden is low ( Ham et al . , 2014 ) , whereas the Mattoo lab suggests that it is an activating modification induced by ER stress ( Sanyal et al . , 2015 ) . AMPylation can explain both the metabolic labeling profile of BiP ( from adenosine and phosphate pools ) as well as the acidity of the modified form and possibly even the sensitivity to novobiocin ( an ATP mimetic ) . However , the assignment of AMPylation to Ser365 or Thr366 in the nucleotide binding domain of BiP ( Ham et al . , 2014; Sanyal et al . , 2015 ) is at odds with the cyanogen bromide cleavage pattern of metabolically labeled BiP , suggesting a complex scenario of multiple modifications with alternative outcomes . Here we report on a functional and quantitative analysis of BiP modification by FICD in vitro and in vivo . Our findings fit best a parsimonious model whereby AMPylation of BiP on a single residue , Thr518 , is the only quantitatively significant modification of the chaperone and provide clues to its functional significance .
Changes in the disposition of BiP induced by manipulation of conditions in the ER are conveniently tracked by native gel electrophoresis ( native-PAGE ) and immunoblotting ( Freiden et al . , 1992 ) coupled with site-directed proteolysis ( Preissler et al . , 2015 ) ( Figure 1A ) . Inhibition of protein synthesis led to accumulation of a prominent high mobility species tentatively named ‘B’ form , whereas depletion of ER calcium progressively drew on this pool to promote the assembly of BiP oligomers ( Preissler et al . , 2015 ) ( Figure 1B and C ) . The ‘B’ form of BiP was also noted for its relative resistance to cleavage in vitro by a bacterial protease , SubA , which cuts BiP within a highly conserved linker sequence connecting the nucleotide binding domain ( NBD ) and substrate binding domain ( SBD ) ( Paton et al . , 2006 ) ( Figure 1A and D ) . Resistance to cleavage by SubA is also a feature of the acidic , modified form of BiP ( Chambers et al . , 2012 ) , suggesting that the ‘B’ form observed on native-PAGE might be related to the covalently modified form of BiP ( previously believed to be the ADP-ribosylated form ) . 10 . 7554/eLife . 12621 . 003Figure 1 . Native gel electrophoresis tracks activity state-dependent changes in BiP’s quaternary structure . ( A ) Schematic representation of BiP’s domain organization in the ATP- and ADP-bound states . BiP consists of an N-terminal nucleotide binding domain ( NBD , pink ) and a C-terminal substrate binding domain ( SBD , blue ) connected by a hydrophobic interdomain linker peptide ( green ) . The SBD is composed of a two-layered β-sandwich subdomain ( SBDβ ) containing the substrate binding crevice and a helical lid structure ( SBDα ) . In the ATP-bound conformation the NBD and SBD form extensive contacts , which involves the linker region , and the SBD is in the open conformation ( SBDα extended ) allowing for interactions with substrates ( dark blue ) at high association and dissociation rates . Upon ATP hydrolysis to ADP the inter-subunit contacts are lost leading to exposure of the linker , packing of SBDα against SBDβ , and strong reduction of substrate interaction kinetics . Cleavage of BiP by the linker-specific protease SubA ( scissors ) is favored in the ADP-bound state . ( B ) Immunoblot of endogenous BiP resolved by native gel electrophoresis . Where indicated the CHO-K1 cells were exposed to cycloheximide ( CHX; 100 µg/mL ) or thapsigargin ( Tg; 0 . 5 µM ) for the indicated time . The major species visible on the native gel are numbered by order of descending mobility ( I-III ) and the monomeric ‘B’ form induced by CHX treatment and the ‘A’ form detectable in untreated cells are marked . Immunoblots of the same samples resolved by SDS-PAGE report on total BiP and total eIF2α ( which also serves as a loading control ) and phosphorylated eIF2α to reveal the impact of thapsigargin action . Note the emergence of the ‘B’ form in CHX-treated cells , which is blocked by thapsigargin . ( C ) BiP immunoblot , as in “A” . Cells were first exposed to cycloheximide to build a pool of the ‘B’ form of BiP and then challenged with thapsigargin ( in the continued presence of cycloheximide ) . Note the disappearance of the ‘B’ form of BiP and the emergence of BiP oligomers in the thapsigargin-treated cells . ( D ) BiP immunoblot of lysates from untreated ( -CHX ) and cycloheximide-treated ( +CHX ) cells . Where indicated the lysates were exposed to the SubA protease that cleaves BiP’s interdomain linker in vitro , before loading onto the native gel . The cleavage products detected by the antiserum used on the native gel are noted ( CP , upper panel ) , as are the full-length BiP ( FL ) and its substrate binding domain ( SBD ) on the SDS-PAGE gel below ( the nucleotide binding domain is very weakly reactive with the antiserum ) . eIF2α serves as a loading control . Note the resistance of the ‘B’ form of BiP to cleavage by SubA . DOI: http://dx . doi . org/10 . 7554/eLife . 12621 . 003 To examine the potential role of FICD in elaboration of the ‘B’ form of BiP , we inactivated both alleles of FICD in hamster CHO-K1 cells by CRISPR-Cas9-mediated genome editing , truncating the coding sequence N-terminal to the FIC-domain active site ( Figure 2A ) . The SubA-resistant ‘B’ form , prominent in cycloheximide-treated wildtype cells , was conspicuously absent in FICD-/- cells ( compare lanes 2 and 5 in Figure 2B ) , whilst other aspects of BiP’s structural dynamism ( such as oligomerization in response to ER calcium depletion ) remained largely unchanged . FICD inactivation also led to disappearance of the acidic form of BiP , present in lysates of cycloheximide-treated wildtype cells resolved by isoelectric focusing PAGE ( IEF-PAGE , Figure 2C ) . FICD-mediated AMPylation is strongly autoinhibited by intramolecular competition for ATP γ-phosphate binding by glutamate 234 , whose mutation to glycine ( FICDE234G ) de-represses FICD , whereas replacement of histidine 363 , which is located in the highly conserved catalytic FIC motif , with alanine ( FICDH363A ) abolishes AMPylation activity ( Engel et al . , 2012 ) . In keeping with these observations , enforced overexpression of the constitutively active FICDE234G led to the re-appearance of an acidic form of BiP in the FICD-/- cells ( Figure 2D ) . Wildtype FICD did not promote BiP’s acidic form , even when strongly overexpressed , suggesting the existence of robust inhibitory mechanisms restraining potentially deleterious effects of protein modification ( Engel et al . , 2012 ) . Such tight regulation is a feature common to FIC enzymes ( Garcia-Pino et al . , 2014 ) but the signals leading to derepression under physiological circumstances are unknown so far . 10 . 7554/eLife . 12621 . 004Figure 2 . FICD deletion abolishes BiP modification in cultured cells . ( A ) Schematic illustration of the hamster FICD protein . The transmembrane region ( TM ) , the two tetratricopeptide repeats ( TPR ) as well as the FIC-domain ( purple ) with its core sub-domain ( dark purple ) and the catalytic loop sequence are shown . Numbers represent amino acid positions . The amino acid sequence surrounding the mutations introduced into the CHO-K1 FICD-/- clone ( #49 ) by CRISPR-Cas9-mediated genome editing are noted . Both alleles result in premature termination of translation deleting the active site ( * ) . ( B ) Immunoblots of endogenous BiP from wildtype ( wt ) or FICD-/- CHO-K1 cell lysates from which ATP was either depleted by incubation with hexokinase and glucose ( -ATP , top panel ) or to which ATP ( 1 mM ) had been added ( +ATP , bottom panel ) , resolved by native-PAGE . Where indicated the cells were exposed to cycloheximide ( CHX , 100 µg/ml ) or thapsigargin ( Tg , 0 . 5 µM ) for 3 hr before lysis . The major species visible on the native gels are numbered by order of descending mobility ( I-III ) and the monomeric ‘B’ form induced by CHX treatment and the ‘A’ form , prominent in ATP-replete lysates of untreated cells , are marked . Immunoblots of the same samples resolved by SDS-PAGE report on total BiP loaded and on eIF2α as a loading control . The ATP-supplemented lysates ( 3 µg/µl protein ) were in addition exposed to SubA ( 30 ng/µl ) for 10 min at room temperature prior to separation by SDS-PAGE and immunodetection of BiP . The intact protein and the substrate binding domain ( SBD ) , which are detected by the antibodies against a C-terminal epitope of BiP , are indicated . The asterisk marks a band of unknown identity . Note that neither CHX-dependent conversion of endogenous BiP into the monomeric ‘B’ form nor the CHX-mediated resistance of BiP towards proteolytic cleavage by SubA , were observed in FICD-/- cells . ( C ) Immunoblot of endogenous BiP from wildtype and FICD-/- CHO-K1 cell lysates resolved on an isoelectric focusing ( IEF ) gel . Where indicated the cells have been exposed to CHX ( 100 µg/ml ) for 3 hr before lysis . Note that the more acidic ( ‘B’ ) form of BiP associated with CHX treatment was absent in FICD-/- cells . ( D ) IEF immunoblot of endogenous BiP from CHO-K1 FICD-/- cells transfected with plasmids encoding wildtype GST-FICD , the constitutively active GST-FICDE234G or the inactive GST-FICDE234G-H363A mutant . Mock transfected cells were analyzed as a control . The cells were treated with CHX ( 100 µg/ml ) for 3 hr before lysis . A pulldown with GSH-Sepharose beads was performed with the same lysates to analyze expression levels of the plasmid-encoded GST-FICD fusion proteins . Note that formation of the acidic ( ‘B’ ) form of BiP was restored by expression of catalytically active GST-FICDE234G protein ( despite its comparatively low expression level ) but neither by expression of the catalytically inactive GST-FICDE234G-H363A mutant nor the regulated wildtype enzyme . DOI: http://dx . doi . org/10 . 7554/eLife . 12621 . 00410 . 7554/eLife . 12621 . 005Figure 2—figure supplement 1 . Time-dependent changes in BiP abundance , BiP ‘B’ form and the fraction of BiP resistant to cleavage by SubA in cells exposed to the reversible ER stress-inducing agent 2-deoxy-D-glucose ( 2-DG ) . ( A ) Immunoblots of BiP from lysates of untreated cells or cells exposed to 2-deoxy-D-glucose ( 2-DG , 3 mM ) . Where indicated , 2-DG was washed out before lysis ( in the presence of ATP ) . The samples in the top panels were exposed to SubA before denaturing SDS-PAGE . The migration of the undigested BiP ( FL ) , the nucleotide binding domain ( NBD ) and substrate binding domain ( SBD ) are indicated . The total content of BiP in the sample ( before digestion with SubA ) and the eIF2α signal ( provided as a loading control ) are also shown . Asterisks mark non-specific bands . Immunoblots of BiP from the same samples resolved by native-PAGE before and after cleavage with SubA are shown below . BiPr denotes SubA-resistant BiP detected on native gels and the positions of the monomeric A and FICD-dependent ‘B’ form , and BiP oligomers ( II ) are indicated . ( B ) Trace of time-dependent changes in total BiP abundance from the samples in “A” [normalized to the signal in the untreated ( t=0 ) sample of each genotype] and in the fraction of BiP digested by SubA from the samples exposed to 2-DG . The fraction digested , FD= ( NBD+SBD ) ( FL+NBD+SBD ) was derived from the top panels in “A” . ( C ) As in “B” but for cells exposed to 2-DG for 8 hr followed by washout . ( D ) Plot of time-dependent changes in the abundance of FICD and RPL27 ( 60S ribosomal protein L27 , a reference gene ) mRNA normalized to PPIA ( cyclophilin A ) mRNA internal control in wildtype cells exposed to 2-DG . The observations made in the key time points of this detailed experiment have been reproduced independently twice . Note the progressive increase in the fraction of BiP susceptible to cleavage by SubA , observed in the wildtype cells following exposure to 2-DG and the correlated disappearance of the ‘B’ form on native-PAGE gels and the correlated re-emergence of ‘B’ form and SubA-resistant BiP during the 2-DG washout; neither of which are evident in the FICD-/- cells . DOI: http://dx . doi . org/10 . 7554/eLife . 12621 . 005 The FICD-/- cells provided a convenient experimental reference against which to gage time dependent changes in the abundance of modified BiP following the imposition of ER stress and during its resolution in wildtype cells . A decline in ‘B’ form and an increase in susceptibility to digestion by SubA were noted within 2 hr of exposure of cells to the reversible ER stress inducing agent 2-deoxyglucose ( Figure 2—figure supplement 1A and B ) and persisted for up to 8 hr thereafter . This decline in ‘B’ form occurred in the face of a marked increase in FICD mRNA ( Figure 2—figure supplement 1D ) ( Ham et al . , 2014; Sanyal et al . , 2015 ) , a finding consistent with regulatory mechanisms that contravene the increase in mRNA . Washout of 2-deoxyglucose was associated with progressive increase in ‘B’ form and the emergence of resistance to cleavage by SubA ( Figure 2—figure supplement 1A and C ) . These changes fit well the previously observed inverse correlation between levels of ER stress and the abundance of modified acidic BiP ( Chambers et al . , 2012; Laitusis et al . , 1999 ) . Together with evidence that FICD was both necessary and sufficient for elaboration of the acidic , modified form of BiP on IEF-PAGE , these observations lent strong support to the notion that the ‘B’ form of BiP , observed on native gels , reflects the same or a related species . The high mobility of the FICD-dependent ‘B’ form and its indifference to the effects of ATP ( Figure 2B , lower panel ) ( which promotes dissociation of BiP from its client proteins in vitro ) are consistent with a role for the FICD-mediated modification in BiP inactivation . To examine these relationships in further detail we measured the effects of FICD on BiP in vitro , in a system constituted of pure components . Whether purified as a GST-fusion protein from bacteria or from overexpressing mammalian cells , active FICDE234G promoted the ATP-dependent in vitro appearance of a ‘B’ form of recombinant BiP purified from bacteria . Like the endogenous ‘B’ form , found in cells , the ‘B’ form constituted in vitro was also partially resistant to cleavage by SubA ( Figure 3A ) . Emergence of the ‘B’ form in vitro correlated with the acquisition of a faster migrating acidic form of BiP on IEF-PAGE ( Figure 3B ) . FICD converted all the detectable BiP to a single acidic form in a time-dependent manner , which , like the endogenous acidic from of BiP , was also relatively resistant to cleavage by SubA ( Figure 3B , lane 8 ) . Likewise , addition of purified active FICDE234G to lysates from FICD-/- cells entirely converted endogenous BiP into the acidic form ( Figure 3C ) . Of note , within the resolution of IEF-PAGE , there is no evidence for heterogeneity in BiP modification [also see ( Carlsson and Lazarides , 1983; Laitusis et al . , 1999 ) ] . This observation is consistent either with processive modification of BiP or modification occurring on a single site in any given BiP molecule . 10 . 7554/eLife . 12621 . 006Figure 3 . AMPylation of purified BiP in vitro recapitulates features of BiP modified in vivo . ( A ) Coomassie ( CBB ) -stained native-PAGE gel ( left panel ) or SDS-PAGE gel ( right panel ) of recombinant BiP purified from bacteria ( 10 µM ) exposed to ATP ( 1 . 5 mM ) in the absence or presence of recombinant active GST-FICDE234G or inactive GST-FICDE234G-H363A ( 0 . 5 µM ) purified from E . coli ( 45 min at 30°C ) . The interdomain linker-specific protease SubA ( 30 ng/µl ) was added afterwards for 10 min where indicated . The major species on the native gel are numbered by order of descending mobility ( I-III ) and the monomeric ‘A’ and ‘B’ forms of BiP are indicated . Full-length BiP , the nucleotide binding domain ( NBD , also resolved on the native gel ) and the substrate binding domain ( SBD ) are annotated on the SDS-PAGE gel . Note the quantitative AMPylation-dependent conversion of BiP into the monomeric ‘B’ form on native gels and the increased resistance of modified BiP to cleavage by SubA . Also note that upon incubation with ATP unmodified BiP forms a second slower migrating monomeric species similar to the ‘B’ form , which likely reflects an alternative ( e . g . ATP-bound ) conformation . ( B ) Immunoblot of an IEF gel of purified hamster BiP19-654 ( 15 µM ) after in vitro AMPylation with GST-FICDE234G or as a control with inactive GST-FICDE234G-H363A ( both at 0 . 75 µM ) in presence of ATP ( 1 . 5 mM ) for the indicated times at 30°C followed by treatment with or without SubA ( 30 ng/µl ) for 10 min . The two forms of full-length BiP ( ‘A’ and the more acidic ‘B’ form ) as well as faint signals that likely represent the unmodified and modified SBD are indicated . ( C ) Immunoblot of endogenous BiP from lysates of cycloheximide-treated CHO-K1 FICD-/- cells resolved by IEF-PAGE following exposure in vitro to purified active or inactive FICD ( as in "B" above ) . Note the conversion of all the detectable BiP to an acidic form in the sample exposed to active FICD . ( D ) Autoradiograph and Coomassie stain of an SDS-PAGE gel of BiP exposed in vitro to active GST-FICDE234G coupled to GSH-Sepharose beads ( or GST alone as a control ) in the presence of α-32P-ATP as a substrate . After the AMPylation reaction the samples were treated further with increasing concentrations of SubA ( 0 . 03 ng/µl , 0 . 1 ng/µl and 20 ng/µl , lanes 3-5 ) where indicated . Note the confinement of the radiolabel to the SBD fragment of cleaved BiP . DOI: http://dx . doi . org/10 . 7554/eLife . 12621 . 00610 . 7554/eLife . 12621 . 007Figure 3—figure supplement 1 . Comparison of the differential susceptibility of unmodified and AMPylated BiP to cleavage by SubA in vitro . ( A ) Schema of the experimental design . ATP hydrolysis-deficient BiPT229A protein was AMPylated in presence of radioactive α-32P-ATP with catalytically active GST-FICDE234G coupled to GSH-Sepharose beads . The enzyme-containing beads were removed by centrifugation and excess of non-radioactive ATP was added to the supernatant to competitively displace non-covalently bound α-32P-ATP from BiPT229A . Unbound nucleotides were then removed by passing the sample through a desalting column . Trace amounts ( 1 . 5 µg ) of the recovered 32P-labeled AMPylated BiPT229A protein were added to excess of unmodified BiPT229A protein ( 60 µg; a mass ratio of 40:1 ) and the combined sample was supplemented with ATP and treated for 30 min with increasing concentrations of the SubA protease ( 0 . 08 to 120 ng/µl ) before denaturing SDS-PAGE , Coomassie staining and autoradiography . ( B ) Autoradiograph and Coomassie stain ( CBB ) of an SDS-PAGE gel of BiP from samples described above . In the bottom panel , the Coomassie stain and radioactive signals of the full-length BiP and the substrate binding domain ( SBD ) were quantified and normalized to the values in lane 1 , which were set arbitrarily to 1 for full-length BiP and to 0 for the SBD ( graph ) . The Coomassie stain signal reports of the fate of unmodified BiPT229A , whereas the radioactive signals report exclusively on the modified BiPT229A in the combined sample . NBD denotes the nucleotide binding domain . ( C ) A shorter exposure ( 1 hr versus 8 hr ) of the autoradiograph shown in “B” , above . It reveals the substantial conservation of the radioactive signal emanating from intact BiP across the time course ( the green plot in “B” above ) , which is obscured by changes in the band width introduced by the progressive digestion of the unlabeled ( and unmodified ) intact BiP in the sample . DOI: http://dx . doi . org/10 . 7554/eLife . 12621 . 007 FICD efficiently transferred the 32P label from the alpha phosphate of ATP onto BiP , consistent with AMPylation ( Figure 3D ) . Radiolabeled BiP too was resistant to cleavage by SubA indicating that BiP radiolabeled in vitro by FICDE234G reports faithfully on the behavior of endogenous BiP ( Figure 3—figure supplement 1 ) . Interestingly , cleavage of radiolabeled BiP ( by prolonged incubation with SubA ) led to emergence of a radiolabeled substrate binding domain fragment ( Leu417-Leu654 ) ; the nucleotide binding domain fragment , though conspicuous on the Coomassie-stained gel , was entirely devoid of label ( Figure 3D and Figure 3—figure supplement 1B ) . This in vitro labeling pattern fits well with the metabolic labeling of BiP by 32P orthophosphate or 3H adenosine label donors in vivo , as the label was mapped to a single CnBr cleavage fragment spanning Thr434 to Met541 ( Chambers et al . , 2012; Gaut , 1997 ) . In contrast , FICD-mediated transfer of 32P label from ATP to BiP’s substrate binding domain , observed here in vitro , can not be reconciled with reported AMPylation of Thr366 or Ser365 ( Ham et al . , 2014; Sanyal et al . , 2015 ) . The migration of purified recombinant BiP on IEF and native gels suggests that most or all of the protein was modified in vitro with FICDE234G , yielding homogenous preparations of modified BiP , whereas BiP treated with the inactive mutant FICDE234G-H363A remained unmodified ( Figure 3A and 3B ) . These preparations were examined by electrospray ionization mass spectrometry and produced molecular masses of 71 , 060 . 1 and 71 , 395 . 6 Da for unmodified and modified BiP , respectively ( Figure 4A and Figure 4—figure supplement 1 ) . These two species were the only proteinaceous ions detected in the samples; the trailing shoulder of the ion-current traces ( Figure 4—figure supplement 1A ) being comprised entirely of singly charged low molecular mass contaminants . The mass difference of 335 Da is consistent with a single covalent modification by AMP ( 329 . 12 Da ) in modified BiP . The difference between the predicted mass of a molecule of AMP bound via an ester linkage ( 329 . 12 Da ) and the observed mass difference between the modified and unmodified BiP ( 335 Da ) is reflective of the broad peaks associated with these measurements . These observations indicate a single quantitative modification of BiP by AMPylation and also fit the simple pattern of modified BiP migration on the IEF-PAGE gel . An analogous mass spectrometry experiment with endogenous BiP immunopurified from cycloheximide-treated wildtype CHO-K1 cells also revealed only two molecular masses ( 70 , 538 . 3 Da and 70 , 866 . 1 Da ) with a mass difference of 327 . 8 Da , consistent with modification of BiP by a single AMP moiety in vivo . BiP from AMPylation-deficient FICD-/- cells gave rise to a single species with a molecular mass ( 70 , 532 . 2 Da ) that is similar to the predicted mass of unmodified mature BiP ( Figure 4—figure supplement 2 ) . 10 . 7554/eLife . 12621 . 008Figure 4 . FICD-mediated incorporation of a single AMP molecule onto the substrate binding domain of BiP in vitro . ( A ) Electrospray mass spectra of bacterially expressed hamster BiP ( 27-654 , with a His6-tag ) after reverse-phase HPLC purification . The spectra contain protein ions with between 36 and 100 associated protons ( the number of protons are indicated for the major species ) . The inset shows the data reconstructed onto a true mass scale . The sample in top panel is of unmodified BiP and that in bottom panel of BiP that had been modified in vitro with GST-FICDE234G and ATP . ( B ) Amino acid sequence of Chinese hamster BiP ( with the cleaved signal peptide in lower case letters ) . The SubA cleavage site ( after L416 ) is marked by the grey arrow and the predicted Arg-C and Asp-N AMPylated proteolytic cleavage fragments are delineated by the colour coded horizontal lines above the protein sequence . ( C ) LC-MS spectra of peptides derived from recombinant BiP digested with Arg-C or Asp-N before ( “No enzyme” ) and after in vitro modification with GST-FICDE234G and ATP . The m/z ratio of the signals is displayed in the abscissa and their relative intensity in the ordinate . The interval of the liquid chromatogram at which the peptides in question eluted is depicted above each paired sample [tLC ( min ) ] . Note the absence of any signal corresponding to the doubly-charged non-AMPylated 511-532 Arg-C fragment in the spectrum derived from BiP after exposure to GST-FICDE234G and ATP ( left-most lower panel ) and the absence of signals corresponding to the doubly-charged AMP modified peptides in the spectra derived from the Arg-C or Asp-N digests of BiP that had not been exposed to GST-FICDE234G and ATP ( central and rightmost upper panels ) . DOI: http://dx . doi . org/10 . 7554/eLife . 12621 . 00810 . 7554/eLife . 12621 . 009Figure 4—figure supplement 1 . Chromatographic profile and reconstructed mass spectrum of unmodified BiP and BiP modified in vitro with FICD and ATP . ( A ) Total ion current chromatograms from the reverse-phase separation of unmodified and AMPylated BiP , respectively . The peak regions denoted by the red arrow were used to generate the mass spectra shown in “B” . ( B ) Electrospray ionization mass measurements of intact unmodified and modified BiP , after reconstruction onto a true mass scale . These data were derived from the ion current chromatograms presented in Figure 4A . Note the homogeneity of the peak corresponding to modified BiP , consistent with quantitative modification of BiP by a single AMP molecule . The ions eluting as a shoulder after the main peaks shown in “A” comprise singly charged non-proteinaceous species . DOI: http://dx . doi . org/10 . 7554/eLife . 12621 . 00910 . 7554/eLife . 12621 . 010Figure 4—figure supplement 2 . Modification of BiP with a single AMP molecule in vivo . ( A ) Electrospray ionization mass spectrum of endogenous BiP eluted from a reverse-phase HPLC column after immunoaffinity purification from wildtype CHO-K1 lysates . The cells were treated with cycloheximide ( 100 µg/ml ) for 3 hr before lysis to enhance BiP modification . The spectrum in the upper panel contains peaks for multiply charged protein ions ( with between 36 and 100 associated protons ) , which were used to reconstruct the molecular masses of BiP ( lower panel ) . Note that only two major peaks ( 70 , 538 . 3 Da and 70 , 866 . 1 Da ) were obtained . The difference ( 327 . 8 Da ) between these reconstructed masses is consistent with the attachment of a single AMP moiety per BiP molecule in vivo . ( B ) Experiment as in “A” with BiP isolated from cycloheximide-treated CHO-K1 FICD-/- cells . Only a single molecular mass was reconstructed ( 70 , 532 . 2 Da ) , consistent with the inability of the mutant cells to AMPylate BiP . Note that in both experiments the masses of non-AMPylated BiP were slightly larger ( 59 . 7 Da and 53 . 6 Da , respectively ) than that predicted of mature endogenous BiP ( 70 , 478 . 6 Da ) , which may be explained by another quantitative ( likely irreversible ) post-translational modification of BiP ( e . g . acetylation of a lysine residue ) . DOI: http://dx . doi . org/10 . 7554/eLife . 12621 . 01010 . 7554/eLife . 12621 . 011Figure 4—figure supplement 3 . Evidence for the absence of modification of Ser365 or Thr366 in mono-AMPylated BiP . ( A ) . Amino acid sequence of Chinese hamster BiP ( with the cleaved signal peptide in lower case letters ) with Ser365 and Thr366 highlighted in red . The SubA cleavage site is marked by the grey arrow and the predicted proteolytic cleavage fragments generated by the four endoproteases are delineated by the color coded horizontal lines above the protein sequence . ( B ) LC-MS spectra of peptides derived from the indicated digests of recombinant BiP before ( “No enzyme” ) and after in vitro modification ( that went to completion , see Figure 3B ) with GST-FICDE234G and ATP . The m/z ratio of the signals is displayed in the abscissa and their relative intensity in the ordinate . The sequence of the peptide encompassing unmodified Ser365 or Thr366 is indicated . Note the contrast between the abundance of signal corresponding to the mass of the unmodified peptides encompassing Ser365 or Thr366 in the samples of AMPylated BiP ( lower panel in each of the four pairs ) and the absence of the peptide corresponding in mass to the non-AMPylated 511-532 Arg-C fragment in the spectrum from the same sample of AMPylated BiP ( Figure 4C , left-most lower panel ) . DOI: http://dx . doi . org/10 . 7554/eLife . 12621 . 011 To identify the AMPylation site ( s ) , samples of unmodified BiP or BiP modified in vitro with FICDE234G were treated with several proteases and the peptide digests were analyzed by liquid chromatography and tandem mass spectrometry ( LC-MS/MS ) . A unique peptide ( doubly charged peak of 1374 . 15 m/z ) corresponding in mass to an AMPylation of residues 511-532 was observed in samples digested with Arg-C , and another peptide ( doubly charged peak of 916 . 94 m/z ) corresponding to AMPylation of residues 515-528 was detected in the Asp-N digest ( Figure 4B and C ) . Moreover , another doubly charged peak ( 1209 . 62 m/z ) corresponding to the unmodified BiP 511-532 fragment was present in the Arg-C digests of unmodified BiP and absent from the spectrum of the modified form ( Figure 4C , left panels ) enhancing the confidence in these assignments . No masses corresponding to peptides overlapping BiP 511-532 [the region identified in the Arg-C and Asp-N digests to contain AMPylation site ( s ) ] , were detected in either the chymotrypsin or trypsin digests of AMPylated BiP . However , all four digests contained peptides covering Ser365 and Thr366 , but these were all unmodified peptides ( Figure 4—figure supplement 3 ) . In contrast to the unmodified 511–532 Arg-C fragment that was absent from the spectrum of the modified BiP ( Figure 4C ) , in all four digests the relative intensity of the signals from peptides encompassing Ser365 and Thr366 was undiminished by modification with FICDE234G ( Figure 4—figure supplement 3B ) ; this despite evidence that all the detectable BiP molecules in the sample were modified ( Figure 3B and 4 ) . The mass spectrometry-based evidence for non-modification of Ser365 or Thr366 also fits with the observation that under the in vitro conditions studied here , FICDE234G ( using α-32P-ATP as a substrate ) selectively modified a BiP fragment C-terminal to the SubA cleavage site ( at Leu416 ) with no evidence for modification of the N-terminal nucleotide binding domain ( Figure 3D and Figure 3—figure supplement 1 ) . The LC-MS/MS fragment spectra of AMP modified peptides from Arg-C and Asp-N digests obtained by high energy collisional induced dissociation ( HCD ) or electron transfer dissociation ( ETD ) did not directly identify the modified residue ( s ) . However , the specificity of FICD for AMPylation of hydroxyl side chains and the overlap of modified peptides from the two digests narrows the modification site to three residues: Thr518 , Thr525 and Thr527 . Both BiPT525A and BiPT527A remained substrates for AMPylation in vitro , whereas the BiPT518A mutation abolished all modification of BiP ( Figure 5A and B ) . 10 . 7554/eLife . 12621 . 012Figure 5 . Mutation of threonine 518 in the substrate binding domain of BiP abolishes its AMPylation in vitro . ( A ) Autoradiograph and Coomassie ( CBB ) stain of an SDS-PAGE gel of recombinant bacterially-expressed wildtype ( wt ) BiP and the indicated mutants exposed in vitro to active GST-FICDE234G coupled to GSH-Sepharose beads ( lanes 2-5 ) or GST alone as a control ( lane 1 ) in the presence of α-32P-ATP as a substrate . ( B ) Coomassie-stained native gel of wildtype BiP and the indicated mutants ( all at 20 µM ) , following exposure to ATP ( 1 . 5 mM ) , GST-FICDE234G ( 0 . 8 µM ) , both or neither ( for 45 min at 30°C ) . Where indicated the samples were afterwards exposed to SubA ( 30 ng/µl , 10 min at room temperature ) . ( C ) As in “A” above , with a different set of mutant BiP proteins . ( D ) As in “B” above , with a different set of mutant BiP proteins . Note that both the T518E and T518A mutations ( in panel “B” above ) affect the mobility of the ‘A’ form of BiP and forestall further changes in mobility by FICD , but only the T518E mutation mimics enzyme-mediated AMPylation by promoting a ‘B’ form-like state partially resistant to cleavage by SubA . DOI: http://dx . doi . org/10 . 7554/eLife . 12621 . 012 BiPT366A and BiPR470E or BiPR492E [the latter are mutations that attenuate labeling of FLAG-tagged BiP in vivo by 32P orthophosphate or 3H adenosine ( Chambers et al . , 2012 ) ] were all modified by FICDE234G in vitro with varying efficiency , as was the substrate binding-deficient mutant BiPV461F ( Figure 5C ) . The substitution of Thr518 by either alanine or glutamic acid moderately impaired the ability of BiP to form oligomers and slightly modified the mobility of the monomeric ( ‘A’ ) form of BiP on native gels , whereas mutating Thr366 had no effect on oligomerization ( Figure 5D ) . BiPT518E could not be AMPylated in vitro ( Figure 5C ) . Yet , whether exposed to FICDE234G or not , BiPT518E resembled AMPylated BiP by its migration on native-PAGE and by its resistance to SubA-mediated proteolytic cleavage ( Figure 5D ) . These observations are consistent with the bulky and negatively charged side chain of glutamic acid mimicking some aspect of AMPylation at Thr518 , promoting a linker-protected conformation . Together these observations support Thr518 as the only FICD-mediated AMPylation site in vitro and suggest that changes at Thr518 introduced by AMPylation ( or mutation ) have significant structural and functional consequences . To evaluate sites of FICD-mediated BiP modification in vivo , we took advantage of a CHO-K1 cell line in which the FICD gene had been inactivated by CRISPR-Cas9-mediated gene editing ( Figure 2 ) . Wildtype and FICD knockout CHO-K1 cells were cultured in media containing arginine with stable isotopes of either "light" or "heavy" nitrogen and carbon ( resulting in a net mass difference of 10 Da per arginine residue ) and exposed to cycloheximide ( to enhance FICD-dependent modification of BiP ) . Endogenous BiP was recovered by immunoaffinity purification , digested with proteases and the resultant peptides were subjected to LC-MS/MS analysis . The SILAC ( stable isotope labeling with amino acids in cell culture ) procedure was designed to quantify relative differences in the abundance of unmodified and modified species in differentially labeled cultures of cells ( Figure 6A ) . 10 . 7554/eLife . 12621 . 013Figure 6 . Reciprocal loss of unmodified and gain of AMPylated BiP511-532 purified from CHO-K1 cells treated with cycloheximide . ( A ) Schema of the design of the SILAC experiment to quantify relative changes in abundance unmodified and AMPylated BiP peptides from untreated and cycloheximide ( CHX ) -treated wildtype and FICD-/- mutant CHO-K1 cells . ( B ) LC-MS spectra of unmodified and modified quadruply-charged BiP511-532 peptides from a SILAC experiment where untreated "heavy" and cycloheximide-treated "light" samples from wildtype CHO-K1 cells were digested by Arg-C . The raw peptide abundance measurements were normalized to the recovery of a doubly-charged reference peptide , BiP61-74 , from the same SILAC samples ( Figure 6—figure supplement 1 ) to arrive at the normalized ratio of the signal in the paired samples ( Rnrl ) . Note that unmodified BiP511-532 is depleted and AMPylated BiP511-532 is reciprocally enriched in cycloheximide-treated wildtype cells . ( C ) Analysis as in “B” ( above ) applied to the indicated paired SILAC samples . Note that unmodified BiP511-532 is depleted by cycloheximide treatment only in wildtype cells and AMPylated BiP511-532 is only detected in wildtype cells . These observations were reproduced in a second independent SILAC experiment . DOI: http://dx . doi . org/10 . 7554/eLife . 12621 . 01310 . 7554/eLife . 12621 . 014Figure 6—figure supplement 1 . No change detected in abundance of unmodified BiP337-367 purified from CHO-K1 cells treated with cycloheximide . Shown are LC-MS spectra of unmodified quadruply-charged BiP337-367 peptides from a SILAC experiment in which untreated “heavy” and cycloheximide-treated “light” samples from wildtype cells or FICD-/- cells were digested by Arg-C . The spectrum of a doubly-charged reference peptide , BiP61-74 , from the same SILAC samples is provided for normalization . Note the absence of change in abundance of peptides encompassing Thr366 in the SILAC sample digested with Arg-C . These observations were reproduced in a second independent SILAC experiment . DOI: http://dx . doi . org/10 . 7554/eLife . 12621 . 01410 . 7554/eLife . 12621 . 015Figure 6—figure supplement 2 . Fragmentation spectra of unmodified and AMPylated BiP Arg-C peptide 511-532 pinpoints AMPylation to Thr518 . ( A ) High energy collision dissociation ( HCD ) fragmentation spectra of unmodified and AMPylated BiP511-532 peptides obtained from Arg-C digests of endogenous BiP ( ‘1’ , upper panels ) immunopurified from cycloheximide-treated wildtype CHO-K1 cells grown in medium containing “light” arginine ( R0 from SILAC sample A in Figure 6A ) and unmodified or in vitro AMPylated recombinant BiP ( ‘2’ , lower panels ) purified from bacteria . The most informative BiP511-532 b and y ions encompassing threonine residues Thr512 , Thr525 and Thr527 are indicated ( dotted lines ) . Note that the signal from unmodified fragment y16 , which includes Thr518 ( red ) , is detectable in spectra of unmodified BiP511-532 but is absent in spectra of AMPylated BiP511-532 ( close-up view ) . ( B ) The amino acid sequences of the b and y ions identified in BiP511-532 fragmentation spectra of unmodified and AMPylated endogenous BiP ( ‘1’ ) or purified recombinant BiP ( ‘2’ ) as shown in “A” are listed below the sequence of the intact BiP511-532 peptide . Threonine residues Thr512 , Thr525 and Thr527 are indicated in green . Fragment ions highlighted in “A” are marked in blue . Note the selective absence of fragment ions spanning Thr518 ( red ) from spectra of AMPylated BiP511-532 . DOI: http://dx . doi . org/10 . 7554/eLife . 12621 . 015 Peptides from Arg-C digests corresponding in mass to the unmodified and AMPylated BiP511-532 were noted in spectra derived from both untreated and cycloheximide-treated wildtype cells . The unmodified species was 1 . 5-fold more abundant in the untreated sample , whereas the AMPylated species was 1 . 5-fold more abundant in the cycloheximide-treated sample ( Figure 6B ) . Unmodified BiP511-532 was 2 . 5-fold more abundant in the cycloheximide-treated FICD knockout sample than in the cycloheximide-treated wildtype sample , whereas no modified BiP511-532 was detected in the cycloheximide-treated FICD knockout sample ( Figure 6C ) . These figures attest to a large fraction of AMPylated BiP in CHO-K1 cells growing in SILAC media ( especially when compared to cells growing in complete media , Figure 2C ) ; perhaps a reflection of depletion wrought by dialysis of the serum . The region encompassing Ser365 and Thr366 was well represented by peptides from the Arg-C digest . The quantification data showed no change in the abundance of the unmodified Arg-C peptide encompassing Ser365 and Thr366 ( BiP337-367 , Figure 6—figure supplement 1 ) . The contrast between the consistent cycloheximide-dependent and FICD genotype-dependent variation in abundance of the peptide containing Thr518 and absence of change in the abundance of peptides containing Ser365 and Thr366 , indicates that the latter residues are unlikely to be major sites of BiP AMPylation in CHO-K1 cells . The mass spectra of the intact protein indicates that FICD modifies any given BiP molecule on a single site both in vitro and in vivo ( Figure 4 and figure supplements ) and the peptide fragment analysis indicates that the modification site ( s ) are all encompassed by peptides spanning 511-532 ( Figure 6 and figure supplements ) . The fragmentation spectra of Arg-C peptides 511-532 from both endogenous BiP AMPylated in vivo , or recombinant BiP AMPylated in vitro , yielded a collection of y-ions and b-ions that included all the residues on the 511-532 peptide except Thr518 , whereas the latter residue was conspicuously represented in the ion series from the fragmentation spectrum of unmodified endogenous and recombinant BiP ( Figure 6—figure supplement 2 ) . The HCD procedure likely fragments the modified peptide in an unpredictable manner , rendering it unrecognizable , however , in relief , the fragmentation pattern reveals Thr518 as the only modification site of BiP , in vitro and in vivo . In vitro radiolabeling revealed that amino acid substitutions distant from the primary AMPylation site ( Thr518 ) notably influenced modification efficiency . Such point mutations might have indirectly influenced the accessibility of Thr518 to FICD by altering the ability of BiP proteins to interact with other molecules or by changing their overall conformation . In particular , we observed stronger modification of BiPV461F and BiPT366A compared to wildtype BiP ( Figure 5C ) . The V461F mutation may enhance access of FICD to BiP simply by reducing competing oligomerization interactions amongst BiP proteins . In contrast , Thr366 is located in the nucleotide binding domain and mutations at this position may interfere with ATP binding or hydrolysis and thus affect the conformation of BiP . The relationship between BiP’s conformation and the efficiency with which it is modified by FICD was systematically tested by analysis of a series of well-characterized BiP mutants that favor certain conformational states based on their altered abilities to interact with and hydrolyze nucleotides ( Gaut and Hendershot , 1993; Petrova et al . , 2008; Wei et al . , 1995 ) . BiPE201G and BiPT229A , mutants that bind ATP and undergo allosteric transitions but are defective in nucleotide hydrolysis ( Gaut and Hendershot , 1993; Petrova et al . , 2008; Wei et al . , 1995 ) , exhibited enhanced AMPylation rates compared to wildtype BiP ( Figure 7A ) . By contrast , BiPT37G , which is defective in ATP hydrolysis and in the allosteric transitions upon ATP binding , showed slightly reduced AMPylation , whereas mutants that are severely defective in adopting the ATP-bound conformation , such as the ATP binding-deficient BiPG226D or BiPADDA , which is locked in the domain-uncoupled state due to a four residue substitution in its interdomain linker ( Laufen et al . , 1999; Preissler et al . , 2015 ) , remained unmodified by FICD ( Figure 7A and B ) . Consistent with an important role for the conformation of BiP in determining its ability to serve as a substrate for FICD , the isolated BiP substrate binding domain ( purified from bacteria as a fusion to yeast Smt3 ) was not modified by FICDE234G ( Figure 7—figure supplement 1 ) . 10 . 7554/eLife . 12621 . 016Figure 7 . AMPylation of BiP is sensitive to its conformational state . ( A ) Autoradiograph and Coomassie ( CBB ) stain of an SDS-PAGE gel of wildtype ( wt ) BiP and the indicated mutants exposed in vitro to active GST-FICDE234G coupled to GSH-Sepharose beads ( lanes 2-5 ) or GST alone as a control ( lane 1 ) in the presence of α-32P-ATP as a substrate . ( B ) Bar graph of densitometric quantification of radiolabeled BiP proteins from in vitro AMPylation reactions as in “A” . The radioactive signals were normalized to the amount of loaded protein ( CBB signal ) and the values for wildtype BiP protein were arbitrarily set to 1 . Mean values ± SD of three independent experiments are shown . ( C ) Coomassie-stained native gels of the indicated BiP mutants ( all at 20 µM ) following exposure to bacterially expressed GST-FICDE234G ( 0 . 8 µM ) in presence of 1 . 5 mM ATP for the indicated time . Note the reduced ability of BiPE201G to form discrete oligomers in presence of ATP , which may be due to altered substrate interaction characteristics . Also note the sharpness of the bands of modified BiPE201G and BiPT229A , which suggest a high degree conformational uniformity and strongly reduced substrate interactions . ( D ) Plot of time-dependent accumulation of the ‘B’ form of BiP from experiments as shown in “C” . Initial values were set to 0% and end-point values were set to 100% for each of the BiP versions , respectively . Mean values ± SD of three independent experiments are shown . Non-linear regression analysis was performed to determine t1/2max values , which were 9 . 9 min , 4 . 0 min and 2 . 3 min for wildtype BiP , BiPT229A and BiPE201G , respectively . DOI: http://dx . doi . org/10 . 7554/eLife . 12621 . 01610 . 7554/eLife . 12621 . 017Figure 7—figure supplement 1 . The isolated BiP substrate binding domain is not measurably AMPylated by FICD in vitro . Autoradiograph and Coomassie ( CBB ) stain of an SDS-PAGE gel of wildtype BiP and a fusion of the isolated substrate binding domain ( SBD ) to Smt3 ( Smt3-SBD ) following exposure in vitro to active GST-FICDE243G coupled to GSH-Sepharose beads ( lanes 2-3 ) or GST alone as a control , ( lane 1 ) in the presence of α-32P-ATP as a substrate . DOI: http://dx . doi . org/10 . 7554/eLife . 12621 . 01710 . 7554/eLife . 12621 . 018Figure 7—figure supplement 2 . Loop 7 , 8 of the BiP substrate binding domain is destabilized in the ATP-bound conformation . The structure of the substrate binding domain ( SBD ) of human BiP in the apo/ADP state ( PDB 5E86 ) and ATP state ( PDB 5E84 ) rendered in cartoon form with the loop encompassing Thr518 ( L7 , 8 ) in stick diagram . L1 , 2 and L3 , 4 delimiting the peptide binding grove , the flanking L5 , 6 and Thr518 are indicated for orientation . Note that the web of polar interactions stabilizing L7 , 8 in the ADP state ( T518 side chain to N520 amine; T518 side chain to D515 side chain; D515 amine to N520 carboxylate; D515 carboxylate to G519 amine; D515 carboxylate to T518 amine and D515 side chain to G517 carboxylate ) is disrupted in the ATP state; disruption that would be maintained by AMPylation of Thr518 . DOI: http://dx . doi . org/10 . 7554/eLife . 12621 . 018 Kinetic differences in modification rates of mutant BiP molecules , suggested by the radiolabeling experiments , were further confirmed by tracking the time-dependent formation of the ‘B’ form of BiP following exposure to FICDE234G ( Figure 7C ) . The t1/2max values for in vitro AMPylation of BiPT229A ( 4 . 0 min ) and BiPE201G ( 2 . 3 min ) were significantly shorter than for wildtype BiP ( 9 . 9 min ) ( Figure 7D ) . ATP is a substrate for FICD , nonetheless , these important kinetic differences in modification were unlikely to reflect the trivial consequence of varying contributions of ATP hydrolysis ( by BiP ) to the availability of substrate nucleotide for the AMPylation reaction . Accelerated modification was not shared by the ATP hydrolysis-defective BiPT37G . Furthermore , BiP AMPylation was performed in the presence of millimolar ATP , whereas the affinity of FICDE234G for nucleotide is measured in the 100 nM range ( Bunney et al . , 2014 ) . Together , these observations indicate that modification of BiP is modulated by its conformation and BiP in the compact ATP-bound conformation is the preferred substrate for FICD . The recently determined crystal structure of human BiP substrate binding domain supports this conclusion in that the loop encompassing the Thr518 AMPylation site is stabilized by six polar interactions present in the apo/ADP conformation that are lost in the ATP conformation ( Yang et al . , 2015 ) , freeing the Thr518 side-chain to react with the active site of FICD ( Figure 7—figure supplement 2 ) . Impaired oligomerization of AMPylated BiP and its restricted conformational flexibility ( reflected in resistance to cleavage by SubA ) suggest that the modification affects BiP function and activity . Hsp70s have low basal ATP hydrolysis rates that are stimulated by J protein co-factors ( Liberek et al . , 1991 ) . Therefore , we compared the basal and J protein-stimulated ATPase activity of unmodified BiP and BiP modified to completion . AMPylation lowered the basal ATP hydrolysis rate of BiP by ~50% ( Figure 8A and B ) and , while the ATPase activity of unmodified BiP was enhanced by the presence of a J-domain in a concentration-dependent manner , AMPylated BiP was almost entirely resistant to J-mediated stimulation of ATP hydrolysis ( Figure 8C ) . 10 . 7554/eLife . 12621 . 019Figure 8 . Functional consequences of BiP AMPylation in vitro . ( A ) Bar diagram of ATP hydrolysis by BiP and BiP AMPylated to completion ( BiP-AMP ) , as reflected in phosphate release ( detected colorimetrically ) . Samples containing either purified BiP or BiP-AMP ( both at 5 µM ) were incubated with 3 mM ATP for 1 hr at 30°C and free orthophosphate generated by ATP hydrolysis was measured . Samples lacking BiP or ATP report on the assay background . Bar graph shows mean absorbance values ± SD at 612 nm ( A612 ) of the complex between free orthophosphate and the malachite green dye after background subtraction of three repeats ( n = 3 ) . ( B ) Bar diagram of Vmax values for basal ATPase activities of unmodified and AMPylated BiP , derived from experiments as in “A” . Mean values ± SD are shown ( n = 3 ) . ( C ) Measurement of J protein-stimulated ATPase activity of unmodified and AMPylated BiP . Samples of purified BiP or BiP-AMP ( both at 1 . 5 µM ) were incubated in presence of 2 mM ATP with isolated J-domain of ERdj6 ( J ) at the indicated concentrations for 3 hr at 30°C and released orthophosphate was detected as in “A” . The control reactions contained 4 µM of non-functional ERdj6H422Q J-domain ( carrying a mutation in the critical HPD motif; J* ) instead of the wildtype J-domain . Shown is the J protein-dependent change in ATP hydrolysis rate of BiP and BiP-AMP relative to their basal ATP hydrolysis rates in absence of J protein ( set to 1 ) of four experiments ( values ± SD , n = 4 ) . ( D ) Plot of concentration-dependent steady-state binding of substrate peptide by unmodified or AMPylated BiP . Fluorescence polarization ( FP ) of 1 µM lucifer yellow-labeled BiP substrate peptide ( HTFPAVLGSC ) was measured after incubation with purified BiP or BiP-AMP at the indicated concentrations for 24 hr at 30°C in presence of 1 mM ADP . Mean values of a representative experiment performed in triplicates are shown . ( E ) Plot of time-dependent release of fluorescently-labeled substrate peptide from unmodified or AMPylated BiP , following injection of 400-fold excess of unlabeled substrate peptide . Fluorescence polarization ( FP ) signal of lucifer yellow-labeled substrate peptide ( 1 µM ) bound to BiP or BiP-AMP ( both at 40 µM ) in presence of 1 mM ADP ( as in “D” ) was measured after addition of 400-fold excess of unlabeled substrate peptide ( 0 . 4 mM ) at t = 0 . The initial values ( after background subtraction ) were set to 100% and non-linear regression analysis was performed on the first 60 min of peptide competition ( inset ) . Mean values of a representative experiment performed in triplicates are plotted on the graph . The mean dissociation rate constants ( koff ) ± SD for BiP = 0 . 037 ± 0 . 006 min-1 and BiP-AMP = 0 . 212 ± 0 . 021 min-1 as well as the mean half-lives ( t1/2 ) ± SD for BiP = 19 . 3 ± 2 . 9 min and BiP-AMP = 3 . 3 ± 0 . 3 min were calculated based on three independent experiments . DOI: http://dx . doi . org/10 . 7554/eLife . 12621 . 01910 . 7554/eLife . 12621 . 020Figure 8—source data 1 . Data from three independent repeats ( each performed in triplicates ) of the experiment presented in Figure 8E are shown . The insert on top of each graph shows the absolute fluorescence polarization ( FP ) signals in arbitrary units ( A . U . ) of a reference sample containing only free fluorescent substrate peptide , which were used to create normalized FP traces of samples containing BiP + peptide . The initial values ( after reference signal subtraction ) were set to 100% . The fit to a single phase decay curve ( tabulated here ) was better than to a two phase model . The fit values from the three experiments were used to calculate the average values for “koff” and the half-lives . Experiment 3 is shown in Figure 8E . DOI: http://dx . doi . org/10 . 7554/eLife . 12621 . 020 AMPylation barely influenced steady-state binding of substrate peptide to BiP: The dissociation constants of the complex between unmodified BiP or AMPylated BiP and a well-characterized substrate peptide , HTFPAVLGSC , measured at equilibrium ( in the presence of ADP ) by fluorescence polarization , were similar ( KdBiP 16 . 7 µM and KdBiP-AMP 11 . 5 µM; Figure 8D ) and within the range previously reported ( Chambers et al . , 2012; Marcinowski et al . , 2011 ) . However , challenge with excess of unlabeled peptide revealed a markedly higher “off” rate of the complex between AMPylated BiP and the bound peptide ( koffBiP 0 . 037 ± 0 . 006 min-1 vs . koffBiP-AMP 0 . 212 ± 0 . 021 min-1; Figure 8E ) . These observations support the view that AMPylated BiP preferentially populated an ATP-like state with high substrate “on” and “off” rates , even when associated with ADP , whereas in the ATP-bound state AMPylated BiP was rendered inactive by its inability to respond to J protein co-factor . AMPylation weakens the interaction of BiP with its substrates and was therefore hypothesized to reverse BiP’s ability to repress the UPR transducers - whether imposed directly by binding to their regulatory lumenal domains ( Bertolotti et al . , 2000 ) or indirectly by competing for unfolded protein substrates ( Pincus et al . , 2010 ) , BiP repression of the UPR involves stable engagement of clients in its substrate binding domain . To test the derivative prediction that enforced expression of an active FICD would lead to UPR activation , wildtype FICD or a catalytically-dead or active FICDE234G were introduced by transient transfection into CHO-K1 cells bearing an integrated CHOP::GFP UPR reporter . The dormant reporter was activated by tunicamycin-induced ER stress ( Figure 9A , panels 1 and 2; a positive control ) , and by acquisition of a constitutively active FICDE234G but not by a catalytically dead FICDE234G-H363A nor by the regulated wildtype enzyme ( Figure 9A , panels 3-5 ) . The robustness of these observations and the independence of UPR activation from endogenous FICD – an important point given the evidence for FICD dimerization ( Bunney et al . , 2014 ) – are attested to by similar observations made in FICD knockout cells ( Figure 9—figure supplement 1 ) . 10 . 7554/eLife . 12621 . 021Figure 9 . Overexpression of active FICDE234G activates the UPR . ( A ) Flow cytometry analysis of CHO-K1 CHOP::GFP UPR reporter cells transiently transfected with plasmids encoding wildtype FICD , the constitutively active FICDE234G or the inactive FICDE234G-H363A mutant alongside an mCherry transfection marker . Mock transfected cells that were treated with the UPR-inducing compound tunicamycin ( 2 . 5 µg/ml ) for 16 hr before the analysis were included as controls . Note the accumulation of CHOP::GFP-positive ( and mCherry-negative ) cells in Q3 in the untransfected tunicamycin-treated samples and enhanced UPR reporter activation in cells transfected with plasmid encoding active FICDE234G and a co-expressed mCherry marker ( reflected in the large number of double-positive cells in Q2 ) . ( B ) Native gel immunoblot of endogenous BiP from lysates ( supplemented with 1 mM ATP ) of Flp-In T-REx 293 cells that carry a stable transgene encoding a doxycycline-inducible form of the active FICDE234G mutant . The cells were treated for the indicated time with doxycycline ( Dox ) prior to lysis . The major species visible on the native gel are numbered by order of descending mobility ( I-III ) and the monomeric ‘B’ and ‘A’ forms are marked . Immunoblots of the same samples resolved by SDS-PAGE report on FICDE234G expression , total BiP loaded and on eIF2α as a loading control . In addition , samples of the lysates were treated with SubA ( 30 ng/µl ) for 10 min at room temperature before separation of proteins by SDS-PAGE and immunoblotting . Full-length BiP ( FL ) , the nucleotide binding domain ( NBD ) and the substrate binding domain ( SBD ) are indicated . The ratios between the quantified signals of full-length BiP and total BiP ( following cleavage by SubA ) were normalized to the value observed in lane 1 ( arbitrarily to 1 ) and are indicated below . The asterisks mark bands of unknown identity . Note the correlation between FICDE234G expression , the appearance of the monomeric ‘B’ form of BiP on the native gel as well as the increasing resistance of BiP towards cleavage by SubA . ( C ) Analysis of Flp-In T-REx 293 cells upon doxycycline-induced expression of inactive FICDE234G-H363A or active FICDE234G as in “B” above . ( D ) Proliferation assay with Flp-In T-REx 293 cells upon doxycycline-induced expression of inactive FICDE234G-H363A or active FICDE234G for the indicated times . Shown are mean values ± SD relative to uninduced cells ( set to 1 ) of three independent experiments ( n = 3 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 12621 . 02110 . 7554/eLife . 12621 . 022Figure 9—figure supplement 1 . Overexpression of active FICDE234G induces UPR in FICD-/- cells . Flow cytometry analysis of CHO-K1 FICD-/- CHOP::GFP UPR reporter cells transiently transfected with plasmids encoding wildtype FICD , the constitutively active FICDE234G or the inactive FICDE234G-H363A mutant as in Figure 9A . DOI: http://dx . doi . org/10 . 7554/eLife . 12621 . 022 The induction of the UPR reporter in CHO-K1 cells was mirrored by the behavior of endogenous BiP , whose levels increased in a time-dependent manner in human HEK 293 cells upon activation of a conditional allele encoding FICDE234G ( Figure 9B ) . Qualitatively similar observations were made in two separate clones of FICDE234G-expressing cells [Figure 9C; the conspicuous basal ( Dox-independent ) SubA resistance of endogenous BiP , lanes 5 and 7 , likely reflecting basal leakiness of the expression system] . These findings are readily explained by the engagement of a feedback mechanism ( the UPR ) that defends a level of active BiP and compensates for the accumulation of inert , AMPylated BiP ( ‘B’ form ) . As suggested previously ( Sanyal et al . , 2015 ) , the capacity of the cells to compensate for the consequences of deregulated FICD activity is limited; reflected here by the adverse effect of sustained FICDE234G expression on cell viability ( Figure 9D ) . The aforementioned observations support the inactivating nature of FICD-mediated BiP modification and suggest that in some circumstances cells deficient in FICD may have elevated levels of active BiP . Given the strong repressive role of BiP on the UPR , enhanced availability of active chaperone attendant upon the absence of FICD and the consequential enhanced capacity to buffer unfolded protein stress might attenuate UPR signaling . Strong feedback control operative in the UPR specifies that such attenuation , were it to occur , would be transient and so more likely to be detected by monitoring early events in the UPR , such as PERK and IRE1α autophosphorylation , than later events , such as reporter gene activity . Furthermore , “over-chaperoning” attendant upon the absence of FICD would likely be revealed by manipulating the flux of proteins into the ER of a secretory cell . These theoretical considerations led us to AR42j pancreatic acinar cells - a secretory cell type in which PERK and IRE1α autophosphorylation are easy to detect ( Bertolotti et al . , 2000 ) . As expected , FICD inactivation by CRISPR-Cas9-mediated gene editing in AR42j cells eliminated both the acidic form of BiP from IEF-PAGE and the ‘B’ form from native-PAGE ( Figure 10A-C and Figure 10—figure supplement 1B ) . Wildtype and FICD-lacking AR42j cells were exposed to cycloheximide to build up pools of modified BiP ( in the wildtype cells ) and then to the ER stress-inducing agent DTT , whose ability to activate the UPR does not require ongoing protein synthesis ( Bertolotti et al . , 2000 ) . FICD knockout led to a conspicuous temporal delay in PERK and IRE1α autophosphorylation , observed in two independently derived AR42j FICD-/- clones ( Figure 10D ) . Sluggish activation of the UPR in the FICD-/- cells was also reflected in attenuated PERK-dependent repression of protein synthesis imposed by DTT on cells pre-treated with cycloheximide . In this experiment , wildtype or FICD-/- AR42j cells were pre-exposed to cycloheximide for 3 hr followed by brief washout to allow recovery of protein synthesis , which was measured by incorporation of puromycin label into newly synthesized proteins . In wildtype cells , DTT led to marked attenuation of protein synthesis , regardless of pre-exposure to cycloheximide ( Figure 10E , compare lanes 1 and 2 and 3 and 4 ) , however pre-exposure to cycloheximide attenuated subsequent translational repression by DTT-mediated PERK activation in FICD-/- cells ( Figure 10E , lanes 5-8 ) . 10 . 7554/eLife . 12621 . 023Figure 10 . Over-chaperoning in FICD-deficient cells delays UPR signaling . ( A ) Schematic illustration of the rat FICD protein . Protein domains are highlighted and the mutations introduced by CRISPR-Cas9-mediated genome editing are presented as in Figure 2A . ( B ) Isoelectric focusing ( IEF ) followed by immunoblot of endogenous BiP from wildtype ( wt ) and FICD-/- AR42j cell lysates . ( C ) Native gel immunoblots of endogenous BiP from ATP-depleted wildtype and FICD-/-AR42j cell lysates ( D ) Endogenous PERK , IRE1α , BiP and actin immunoblots of an SDS-PAGE gel on which lysates of wildtype and FICD-/- AR42j cells were resolved . The cells were pre-exposed to cycloheximide ( CHX , 100 µg/ml ) for 3 hr followed by treatment with 1 mM DTT for the indicated times . Slower migrating phosphorylated PERK is marked ( PERK-P ) . To detect phosphorylated IRE1α ( IRE1α-P ) samples of the same lysates were resolved on a Phos-tag gel . Actin served as a loading control . Asterisks mark bands of unknown identity . Note the delayed phosphorylation ( activation ) of PERK and IRE1α in the two FICD-/- clones . The experiment was performed four times with comparable results . ( E ) Immunoblots of puromycinylated proteins , PERK , IRE1α , BiP and actin from wildtype and FICD-/- AR42j cells . Where indicated , the cells were pre-treated with CHX ( 100 µg/ml ) for 3 hr followed by washout and exposure to puromycin ( Puro , 10 µg/ml ) in presence or absence of the reducing agent DTT ( 1 mM ) for 30 min . Puromycin incorporation into nascent chains ( reporting on protein synthesis rates ) , and PERK-P signals expressed relative to lane 1 ( arbitrarily set to 1 ) are plotted in the bar graph below ( mean values ± SD of three independent experiments , n = 3 ) . The asterisk indicates a band of unknown identity . Note persistent protein synthesis in FICD-/- cells that were exposed to DTT after CHX pre-treatment ( lane 8 ) . Also note the lower basal levels of phosphorylated forms of IRE1α and PERK in untreated FICD-/- cells . DOI: http://dx . doi . org/10 . 7554/eLife . 12621 . 02310 . 7554/eLife . 12621 . 024Figure 10—source data 1 . Data from three independent repeats used for quantification shown in the graph in Figure 10E . DOI: http://dx . doi . org/10 . 7554/eLife . 12621 . 02410 . 7554/eLife . 12621 . 025Figure 10—source data 2 . Source file of the flow cytometry data used to generate the plot in Figure 10—figure supplement 2D . DOI: http://dx . doi . org/10 . 7554/eLife . 12621 . 02510 . 7554/eLife . 12621 . 026Figure 10—figure supplement 1 . Undetectable FICD protein in AR42j FICD-/- cell lysates . ( A ) Immunoblot analysis of the sensitivity of anti-FICD antibodies . Indicated amounts of purified bacterially-expressed mouse FICD104-458 were applied to SDS-PAGE gels followed by immunoblotting with chicken anti-FICD antibodies ( 1:1000 ) and commercial rabbit anti-FICD antibodies ( 1:1000 ) . Note: the detection limit of 10 ng antigen and our inability to detect FICD by direct immunoblot of lysate of ~5 µL packed cell volume , imply a concentration in the ER of less than 1 µM . ( B ) FICD from lysates of wildtype ( wt ) and FICD-/- AR42j cells was immunopurified with chicken anti-FICD antibodies ( IP 1 ) or rabbit anti-FICD antibodies ( IP 2 ) . The recovered proteins were resolved on SDS-PAGE gels and FICD was detected by immunoblotting with rabbit anti-FICD antibodies ( IP 1 ) or with chicken anti-FICD antibodies ( IP 2 ) . Note that no FICD signal was detected in the samples from FICD-/- cell lysates . DOI: http://dx . doi . org/10 . 7554/eLife . 12621 . 02610 . 7554/eLife . 12621 . 027Figure 10—figure supplement 2 . Absence of FICD does not measurably affect the induction kinetics of the transcriptional response to unfolded protein stress . ( A ) Schematic illustration of the hamster FICD protein . Protein domains are highlighted and the mutations introduced by CRISPR-Cas9-mediated genome editing into CHO-K1 CHOP::GFP UPR reporter cell line are presented ( as in Figure 2A ) . ( B–C ) Confirmation of the mutant phenotype by isoelectric focusing ( IEF ) and native gel electrophoresis followed by immunodetection of endogenous BiP ( as in Figure 2B-C ) . ( D ) Flow cytometry analysis of the CHOP::GFP UPR reporter gene activity in wildtype and CHO-K1 FICD-/- CHOP::GFP UPR reporter cells upon exposure to tunicamycin ( 2 . 5 µg/ml ) for the indicated time . DOI: http://dx . doi . org/10 . 7554/eLife . 12621 . 027 The aforementioned defect in UPR activation in FICD-/- cells was both fleeting and dependent on the pre-imposition of a low incoming protein flux regime on the ER ( attained experimentally by cycloheximide pre-treatment ) , as FICD deletion had no measureable effect on the kinetics of CHOP::GFP expression induced by treatment with tunicamycin ( Figure 10—figure supplement 2 ) . These observations are inconsistent with the role proposed for FICD in signaling by the PERK branch of the UPR ( Sanyal et al . , 2015 ) and suggest instead that FICD-mediated BiP inactivation adjusted the level of active BiP to transient fluctuations in unfolded protein flux into the ER . In the absence of this mechanism , FICD knockout cells experienced an excess of functional chaperone , which was revealed here as attenuated activation of the earliest steps of the UPR .
Previous glimpses at BiP modification have been largely indirect: Inferred from transfer of radiolabel from intracellular 32P phosphate or 3H adenosine pools to an acidic form of BiP and have led to the conclusion that the modification ( s ) consists of phosphorylation or ADP-ribosylation or both . Attempts to pinpoint these hypothesized modifications directly were unsuccessful . The discovery of FICD-mediated BiP AMPylation ( Ham et al . , 2014; Sanyal et al . , 2015 ) was thus an important breakthrough whose significance is enhanced further by our finding that elimination of FICD abolishes all evidence for endogenous BiP modification in vivo as detected by IEF- and native-PAGE , and changes in mass of the endogenous protein . These findings unify and clarify the pre-existing observations regarding BiP modification and indicate that FICD-mediated AMPylation on Thr518 is the major , if not the only covalent change in this chaperone detectable by the existing methods . Furthermore , FICD-mediated AMPylation on Thr518 is sufficient to account for the previously observed key features of modified BiP , namely its acidity and inertness . Our findings are consonant with metabolic labeling experiments that mapped the modification to the C-terminal substrate binding domain of BiP and provide positive evidence for absence of significant FICD-mediated modification of Ser365 or Thr366 , neither in vitro nor in vivo . Provided ATP as a co-substrate , FICD modifies BiP in vitro at a single site with a mass expected of AMPylation . AMPylation maps solely to peptides encompassing Thr518 and these are modified to high stoichiometry in vitro . Furthermore , an unbiased chemical proteomic approach to profiling proteins that are modified by FICD in a cell lysate in vitro led to the identification of BiP Thr518 as an AMPylation site , whereas no modification at Ser365 or Thr366 was reported ( Broncel et al . , 2015 ) . It is thus tempting to speculate that AMPylation on Thr518 is the only quantitatively significant modification of BiP carried out by FICD in vivo too . However , the discovery that an FIC-domain containing bacterial enzyme AMPylates eukaryotic targets ( Kinch et al . , 2009; Yarbrough et al . , 2009 ) was closely followed by the realization that the domain may also participate in transfer of part of other pyrophosphate-containing metabolites , leading to UMPylation , mono-phosphorylation and conjugation of phosphocholine ( reviewed in Garcia-Pino et al . , 2014 ) . Thus , one must keep an open mind in regards to the possibility of parallel FICD-mediated modification of BiP ( and possibly other ER proteins ) by metabolites ( other than ATP ) found in the ER . The phenotype of FICD deletion in Drosophila is restricted to a defect in light sensing and involves disruption of neurotransmitter ( histamine ) re-uptake via glial cells . The subcellular loci of Drosophila FICD action are unclear: Consistent with an ER-based function , the protein co-fractionates with BiP . However , HRP-tagged endogenous FICD led to prominent staining of capitate projections , a glial cell plasma membrane component of photoreceptor synapses ( Rahman et al . , 2012 ) . Therefore , it is possible that FICD may have substrates other than BiP that figure prominently in its action in specific cell types . Thr518 of hamster BiP is highly conserved in other eukaryotes . The crystal structure of the isolated substrate binding domain of human BiP ( PDB 5E86 ) and yeast Kar2/BiP ( PDB 3H0X ) ( locked in the ADP-like state ) , shows the corresponding residues – human Thr518 and Kar2 Thr538 – to be located on a loop connecting beta strands 7 and 8 ( L7 , 8 ) , with its side-chain stabilized by a network of polar contacts with the conserved side chain of Asp515/Asp535 and the backbone amine of Asn520/Lys540 . However , both interactions are disrupted in the substrate binding domain of human BiP in the ATP state ( PDB 5E84 ) . Exposure of the Thr518 side chain to the solvent is likely enhanced further by the loss of four other polar interactions that stabilize L7 , 8 in the ADP state ( Yang et al . , 2015 ) . Therefore , mobilization of the side chain by allosteric signals attendant upon ATP binding is suggested to enhance exposure of Thr518 to solvent explaining the selective FICD-mediated modification of ATP-bound BiP and the absence of modification of mutant variants of BiP that are defective in domain coupling ( BiPADDA , BiPG226D and to a lesser extent BiPT37G ) . These recent observations on human BiP ( Yang et al . , 2015 ) also showcase the long range allostery common to all Hsp70 proteins ( Mayer , 2013 ) : the identity of the nucleotide bound in the nucleotide binding domain affects stability of L7 , 8 on the opposite side of the protein . It seems plausible that a modification affecting the disposition of this loop could reach the nucleotide binding domain through a reciprocal set of allosteric interactions to explain the effects of AMPylation on nucleotide hydrolysis and responsiveness to J proteins . An earlier study from our lab revealed that mutations affecting residues R470 and R492 consistently diminished labeling of overexpressed mutant BiP from 32P orthophosphate or 3H adenosine pools in vivo ( Chambers et al . , 2012 ) , leading us to suggest that these residues undergo ADP-ribosylation in vivo . The current findings indicate that our earlier conclusions were in error and suggest that the important structural role of R470 [in all Hsp70s ( Fernandez-Saiz et al . , 2006 ) ] and R492 [specifically in BiP ( Yang et al . , 2015 ) ] may have given rise to defective allosteric transitions by the R470 and R492 BiP mutants . Such defects appear to have a more conspicuous effect on modification of overexpressed BiP in vivo than on the modification of pure recombinant BiP by FICD in vitro . The structures of nucleotide-bound Hsp70s also provide hints to the functional consequences of BiP AMPylation . Residues comprising beta strand 8 in the substrate binding domain of the ADP-bound form of BiP and DnaK are dramatically delocalized in the ATP-bound form ( Kityk et al . , 2012; Qi et al . , 2013; Yang et al . , 2015 ) enabling the domain movements that underlie the functional allosteric transitions ( Mayer , 2013 ) . By altering the conformation of the preceding loop , AMPylation of Thr518 could facilitate the melting of beta strand 8 , thus favoring a new state of the chaperone . Our observations suggest that this new state would resemble the ATP-bound state in terms of high substrate “off” rates , but would differ from it in terms of ( un ) responsiveness to J protein-driven ATP hydrolysis . It is notable in this regard that the ATP-like conformation can be uncoupled from responsiveness to J protein , as a mutation of BiP’s substrate binding domain ( G461P/G468P ) that locks the protein in an ATP-like conformation is also refractory to J protein-mediated stimulation of ATP hydrolysis ( Yang et al . , 2015 ) . These structural considerations fit with the established inverse correlation between activity of BiP ( imposed by the burden of client proteins ) and the extent of modification: Nutrient deprivation and protein synthesis inhibitors , which lower the flux of unfolded proteins into the ER , increase the acidic form of BiP in cultured cells ( Laitusis et al . , 1999; Ledford and Jacobs , 1986 ) and animal tissues ( Chambers et al . , 2012 ) , whereas imposition of ER stress leads to less modified BiP ( Chambers et al . , 2012; Ham et al . , 2014; Hendershot et al . , 1988; Laitusis et al . , 1999; Leno and Ledford , 1989 and our observations here ) . All this suggests a simple mechanism whereby substrate-free , ATP-associated BiP partitions between two mutually exclusive fates: entering the substrate binding cycle or AMPylation . The former is driven by the concentration of unfolded client proteins and catalyzed by J proteins , whereas the latter imposes on BiP an inactive conformation that disfavors the J protein-mediated ATP hydrolysis required for high affinity client binding ( De Los Rios and Barducci , 2014; Misselwitz et al . , 1998 ) . High client “off” rates ensure that AMPylated BiP would neither interfere with protein folding by interacting excessively with substrates , nor would it unduly repress the UPR transducers . And futile cycles of ATP hydrolysis would be obviated by refractoriness to stimulation by J protein . However , given the reversibility of the modification , this inert BiP would serve as a repository for active chaperone were it needed ( Figure 11 , a model ) . 10 . 7554/eLife . 12621 . 028Figure 11 . Schema depicting the hypothesized relationship between AMPylation and the BiP chaperone cycle . FICD-mediated AMPylation on Thr518 allosterically traps BiP in a low substrate-affinity ATP-like state that is refractory to J protein-mediated stimulation of its ATPase activity . Removal of the modification by a phosphodiesterase allows BiP to re-join the chaperone cycle ( depicted in the lower portion of the cartoon ) . DOI: http://dx . doi . org/10 . 7554/eLife . 12621 . 028 Whilst such a simple kinetic competition model could account for the inverse relationship between unfolded protein load and AMPylation , there are hints of other layers of refining regulation . Like other type II or III FIC domain-containing proteins , FICD is intrinsically repressed by the insertion of a conserved glutamate into the ATP binding pocket , delocalizing the co-substrate ATP ( Bunney et al . , 2014; Engel et al . , 2012 ) . Relief of such repression , imposed experimentally by the E234G mutation ( Engel et al . , 2012 ) or by a nanobody ( Truttmann et al . , 2015 ) , is likely attained physiologically through allosteric mechanisms . Evidence that FICD binds unfolded proteins directly ( Sanyal et al . , 2015 ) hint at the possibility that such allostery might be responsive to the burden of unfolded proteins in the ER or might be regulated by FICD’s oligomeric state or co-factor binding ( Bunney et al . , 2014 ) . Coupling of FICD’s intrinsic activity to protein folding homeostasis in the ER could conserve energy by limiting the futile cycles of AMPylation and de-AMPylation prescribed by regulation based solely on kinetic competition with substrates for BiP in the ATP-bound state . Furthermore , regulation of FICD enzymatic activity would explain the dissociation between transcriptional induction of the FICD gene during the UPR and the emergence of modified BiP , which , as shown here , is delayed until after resolution of the stress . Recent findings suggest the existence of an alternative inactive state of BiP effected by oligomerization . The architecture of the oligomers , whereby one BiP molecule engages the interdomain linker of another as a conventional substrate peptide ( Preissler et al . , 2015 ) , implies that BiP protomers are locked in the ADP-bound state and unlikely to serve as a substrate for AMPylation . This conjecture fits the observation that depletion of ER calcium , which promotes BiP oligomerization at the expense of substrate binding , is associated with gradual depletion of the FICD-dependent modified ‘B’ form of BiP ( Preissler et al . , 2015 ) . The benefits of ensuring an adequate reserve of chaperones to cope with the unfolded protein burden are clearly revealed by the consequences of defects in UPR signaling ( Walter and Ron , 2011 ) . But the emergence of active mechanisms to downregulate BiP activity suggest that an excess of functional BiP might have a cost too . This cost is less obvious than that associated with UPR defects , as FICD inactivation results in no conspicuous growth defect in cultured cells . Nonetheless , our findings indicate that in circumstances of limited client protein load , the absence of FICD shifts the equilibrium in the ER in favor of chaperones over their clients ( at least as reflected in attenuated UPR signaling ) . BiP overexpression works in the same direction and is known to impose a measure of inefficiency in the secretion of certain proteins ( Dorner et al . , 1992 ) , it will thus be interesting to learn if FICD inactivation also promotes inefficiency in secretion and if so , under what circumstances .
Supplementary file 1 lists the plasmids used , their lab names , description and notes their first appearance in the figures and their corresponding label , and provides a published reference , where available . A combination of PCR-based manipulations , restriction digests and site-directed mutagenesis procedures were used to mobilize the coding sequences and produce in-frame fusions with the affinity tags [GST , hexahistidine ( His6 ) and His6-Smt3 epitopes] or mCherry fluorescent marker , and to create the point mutations indicated in the text . All cells were grown in tissue culture dishes or multi-well plates ( Corning ) at 37°C and 5% CO2 and the following cell line-specific media were used: CHO-K1 cells ( ATCC CCL-61 ) were phenotypically validated as proline auxotrophs and their Cricitulus griseus origin was confirmed by genomic sequencing . The cells were cultured in Nutrient mixture F-12 Ham ( Sigma , UK ) supplemented with 10% ( v/v ) serum ( FetalClone II; HyClone , South Logan , UT ) , 1 x Penicillin-Streptomycin ( Sigma ) and 2 mM L-glutamine ( Sigma ) . AR42j cells ( ATCC CRL-1492 ) were phenotypically validated by documenting inducibility of amylase expression by dexamethasone and their Rattus sp . origin was confirmed by genomic sequencing . They were cultured in DMEM ( Sigma ) supplemented with 10% ( v/v ) serum ( FetalClone II; HyClone ) , 1 x Penicillin-Streptomycin ( Sigma ) , 2 mM L-glutamine ( Sigma ) , 1 x non-essential amino acids ( Sigma ) and 50 µM β-mercaptoethanol ( Gibco; Life Technologies , UK ) . HEK293T cells ( ATCC CRL-3216 ) were cultured in DMEM ( Sigma ) supplemented with 10% ( v/v ) serum ( FetalClone II; HyClone ) , 1 x Penicillin-Streptomycin ( Sigma ) and 2 mM L-glutamine ( Sigma ) . Flp-In T-REx 293 cells ( Invitrogen , UK ) were cultured in the same medium except that 10% ( v/v ) tetracycline-free serum ( FBS Premium; PAN-Biotech , Germany ) was used . All experiments with untransfected cells were performed at cell densities of 60-90% confluence . For pharmacological treatments the drugs were first diluted into pre-warmed culture medium , mixed and immediately applied to the cells by medium exchange . Unless indicated otherwise the following final concentrations were used: 100 µg/ml cycloheximide ( Sigma ) , 0 . 5 µM thapsigargin ( Calbiochem , Germany ) , 1 mM DTT , 2 . 5 µg/ml tunicamycin ( Melford , UK ) , 10 µg/ml puromycin ( Calbiochem ) , and 3 mM 2-deoxy-D-glucose ( ACROS Organics , Belgium ) . Three single guide RNA sequences ( plasmids UK1448 , UK1449 and UK1450; Supplementary file 1 ) for targeting the third exon of Cricetulus griseus ( Chinese hamster ) FICD were selected from the CRISPy database [URL: http://staff . biosustain . dtu . dk/laeb/crispy/ , ( Ronda et al . , 2014 ) ] and duplex DNA oligonucleotides of the sequences were inserted into the pSpCas9 ( BB ) -2A-GFP plasmid ( plasmid UK1359; Supplementary file 1 ) following published procedures ( Ran et al . , 2013 ) . 2 x 105 CHO-K1 or CHO-K1 CHOP::GFP reporter cells ( Novoa et al . , 2001 ) were plated in 6-well plates . Twenty-four hr later the cells were transfected with a combination of guide RNA/Cas9 plasmids UK1448 and UK1449 or UK1448 and UK1450 ( 2 µg total plasmid DNA per transfection ) using Lipofectamine LTX ( Invitrogen ) . Thirty-six hr after transfection the cells were washed with PBS , resuspended in PBS containing 4 mM EDTA and 0 . 5% ( w/v ) BSA , and GFP-positive cells were individually sorted by fluorescence-activated cell sorting ( FACS ) into 96-well plates using a MoFlo Cell Sorter ( Beckman Coulter ) . Clones were then analyzed by a PCR-based assay to detect FICD mutations as described ( Klampfl et al . , 2013 ) . Briefly , primers were designed for the region encompassing the FICD RNA guide target sites and the reverse primer was labeled with 6-carboxyfluorescein ( 6-FAM ) on the 5’ end . A PCR reaction was set up using 5 µl of AmpliTaq Gold 360 Master Mix ( Applied Biosystems , UK ) , 0 . 6 µl of a mix of 10 µM forward and labeled reverse primers , 3 . 4 µl H2O and 1 µl genomic DNA ( approximately 10 ng/µl ) . PCR was performed as follows: 95°C for 10 min , 10 x ( 94°C for 15 s , 59°C for 15 s , 72°C for 30 s ) , 20 x ( 89°C for 15 s , 59°C for 15 s , 72°C for 30 s ) , 72°C for 20 min . PCR products were diluted 1:100 in water and fragment length was determined on a 3130xl Genetic Analyzer ( Applied Biosystems ) and the data were analyzed using the Gene Mapper software ( Applied Biosystems ) . Clones for which frameshift-causing insertions or deletions were detected for both alleles were sequenced to confirm the FICD mutations . FICD knockouts in the AR42j cell line were created as described above , using two single guide RNA sequences ( plasmids UK1503 and UK1504; Supplementary file 1 ) for targeting the third exon of Rattus norvegicus ( rat ) FICD selected using the CRISPR Design tool [URL: http://crispr . mit . edu/ ( Zhang Lab ) ] . These cells were electroporated using the Neon transfection system ( Life Technologies , UK ) as described ( Tsunoda et al . , 2014 ) . The effect of FICD overexpression was analyzed in mammalian cell clones carrying stable transgenes that encode doxycycline-inducible mutant versions of FICD . To generate stable cell lines , plasmid DNA encoding FICDE234G ( UK1440 ) and FICDE234G-H363A ( UK1446 ) were introduced into the Flp-In T-REx 293 cell line ( Invitrogen ) as described by the manufacturer's protocol . Briefly , HindIII-XhoI DNA fragments containing the coding sequences for FICDE234G and FICDE234G-H363A were sub-cloned into pcDNA5/FRT/TO ( Invitrogen ) . The resulting expression plasmids were individually co-transfected along with the Flp-recombinase expression vector pOG44 ( Invitrogen ) using Lipofectamine 2000 ( Invitrogen ) into Flp-In T-REx 293 cells . Isogenic clones expressing the transgene under a doxycycline-inducible promoter were selected for resistance to blasticidin ( 3 µg/ml; Thermo Fisher , UK ) and hygromycin ( 250 µg/ml; Invitrogen ) and sensitivity to zeocin ( 50 µg/ml; Invitrogen ) . Flp-In T-REx 293 cells with stable transgenes encoding FICDE234G or FICDE234G-H363A were plated at a density of 8 x 103 cells per well on a 24-well tissue culture plate and grown for 24 hr . The cells were then treated with 0 . 1 µg/ml doxycycline ( Melford ) for the indicated times . Afterwards , the cells were washed once in regular medium and then maintained in regular medium for three days . Following the recovery period , the medium was replaced with fresh medium containing 0 . 02 mM WST-1 ( Dojindo , Germany ) and 0 . 02 mM 1-methoxy phenazine methosulfate ( Sigma ) , and the cells were incubated for 60 min at 37°C before absorbance was measured at 440 nm . Each experiment was performed in duplicates and repeated three times . The effect of FICD overexpression ( wildtype , FICDE234G and FICDE234G-H363A mutant versions ) on the unfolded protein response was studied by transient transfection of wildtype and FICD-/- CHO-K1 CHOP::GFP UPR reporter cell lines with plasmids UK1397 , UK1398 or UK1443 ( see Supplementary file 1 ) using Lipofectamine LTX . Where indicated , cells were treated 24 hr after transfection with the ER stress-inducing agent tunicamycin ( 2 . 5 μg/µl ) for 16 hr to activate CHOP::GFP ( which results in enhanced GFP production ) . Cells were analyzed by dual-channel flow cytometry with an LSRFortessa cell analyzer ( BD Biosciences ) as described previously ( Tsunoda et al . , 2014 ) . GFP ( excitation laser 488 nm , filter 530/30 ) and mCherry signals ( excitation laser 561 , filter 610/20 ) were detected . The data were processed using FlowJo and median reporter values were plotted using GraphPad Prism ( GraphPad Software ) . Cell lysis was performed as described ( Preissler et al . , 2015 ) . Mammalian cells were grown on 10 cm dishes and treated as indicated . At the end of each experiment the dishes were placed on ice and washed twice with ice-cold PBS . The cells were detached with a cell scraper in PBS containing 1 mM EDTA , transferred to a 1 . 5 ml reaction tube and centrifuged for 5 min at 370 g at 4°C . The cells were lysed in HG lysis buffer [20 mM HEPES-KOH pH 7 . 4 , 150 NaCl , 2 mM MgCl2 , 10 mM D-glucose , 10% ( v/v ) glycerol , 1% ( v/v ) Triton X-100] supplemented with protease inhibitiors [1 mM phenylmethylsulfonyl fluoride ( PMSF ) , 2 µg/ml pepstatin , 2 µg/ml leupeptin , 4 µg/ml aprotinin] for 10 min on ice and centrifuged at 21 , 000 g for 10 min at 4°C . The protein concentrations of the cleared lysates were determined using the Bio-Rad protein assay reagent ( Bio-Rad , Germany ) and normalized with lysis buffer ( usually the protein concentration was adjusted to 1 . 2 µg/µl ) . Where indicated the lysates were treated with SubA protease as described . For SDS polyacrylamide gel electrophoresis ( SDS-PAGE ) analysis , the proteins were denatured by addition of SDS sample buffer and heating for 10 min at 75°C followed by separation on 12% SDS polyacrylamide gels . For detection of endogenous BiP by native-PAGE ( see below ) the cells were lysed in HG lysis buffer containing 2 x protease inhibitors ( 2 mM PMSF , 4 µg/ml pepstatin , 4 µg/ml leupeptin , 8 µg/ml aprotinin ) with or without 100 U/ml hexokinase ( Type III from baker’s yeast; Sigma ) as described above and samples were loaded immediately on native or SDS polyacrylamide gels . Phosphatase inhibitors ( 10 mM tetrasodium pyrophosphate , 100 mM sodium fluoride , 17 . 5 mM β-glycerophosphate ) were added to the lysis buffer in experiments where phosphorylated proteins were detected . Native-PAGE was performed as described ( Preissler et al . , 2015 ) . Tris-glycine polyacrylamide gels consisting of a 4 . 5% stacking gel and a 7 . 5% separation gel were used to separate purified protein or proteins from mammalian cell lysates under non-denaturing conditions to detect BiP oligomers . The gels were run in a Mini-PROTEAN electrophoresis chamber ( Bio-Rad ) in running buffer ( 25 mM Tris , 192 mM glycine , pH ~8 . 8 ) at 120 V for 2 hr when cell lysates were applied or for 1:45 hr when His6-tagged purified BiP proteins were analyzed . The proteins were then visualized by staining with InstantBlue Coomassie solution ( expedeon , UK ) or transferred for immunodetection to a polyvinylidene difluoride ( PVDF ) membrane in blotting buffer ( 48 mM Tris , 39 mM glycine , pH ~9 . 2 ) containing 0 . 04 ( w/v ) SDS for 16 hr at 30 V . The membrane was washed after the transfer for 20 min in blotting buffer supplemented with 20% ( v/v ) methanol before blocking . Seven µg of purified BiP protein was loaded per lane on a native gel to detect BiP oligomers by Coomassie staining and volumes of lysates corresponding to 30 µg of protein ( CHO-K1 and AR42j cells ) or 90 µg protein ( Flp-In T-REx 293 cells ) were loaded per lane . Proteins were separated by SDS-PAGE or native-PAGE and transferred to PVDF membranes . The membranes were blocked with 5% ( w/v ) dried skimmed milk in TBS ( 25 mM Tris-HCl pH 7 . 5 , 150 mM NaCl ) and probed with primary antibodies followed by IRDye fluorescently labeled secondary antibodies ( Li-Cor , UK ) . The membranes were scanned with an Odyssey near-infrared imager ( Li-Cor ) and where indicated densitometric quantification of the immunoblot signals was performed with ImageJ ( NIH ) . Primary antibodies and antisera against hamster BiP [chicken anti-BiP; ( Avezov et al . , 2013 ) ] , eIF2α [mouse anti-eIF2α; ( Scorsone et al . , 1987 ) ] , phosphorylated eIF2α [rabbit anti-eIF2α Phospho ( pS51 ) ; Epitomics cat . # 1090-1 , UK] , and FICD [rabbit anti-FICD; LifeSpan BioSciences cat . # LS-C80941 , UK or chicken anti-FICD ( see below ) ] were used . To separate unmodified IRE1α and phosphorylated IRE1α lysates from AR42j cells were loaded on SDS polyacrylamaide gels ( Mini-PROTEAN system; Bio-Rad ) consisting of a 4% polyacrylamide stacking gel and a 7% polyacrylamide separation gel . The separation gel contained 50 µM MnCl2 and 25 µM Phos-tag Acrylamide AAL-107 ( NARD Institute Ltd . , Japan ) . The gels were run for 3 hr at 300 V , incubated for 10 min in immunoblot transfer buffer containing 1 mM EDTA and proteins were transferred to PVDF membranes . Wildtype and mutant versions of N-terminally hexahistidine- ( His6- ) tagged hamster BiP proteins ( see Supplementary file 1 and “Plasmid construction” ) were expressed in M15 E . coli cells ( Qiagen , Germany ) as described ( Preissler et al . , 2015 ) . Bacterial cultures were grown at 37°C in LB medium containing 100 µg/ml ampicillin and 50 µg/ml kanamycin to an optical density ( OD600 nm ) of 0 . 8 and expression was induced with 1 mM isopropylthio β-D-1-galactopyranoside ( IPTG ) . After incubation for 6 hr at 37°C the cells were harvested by centrifugation and lysed with a high-pressure homogenizer ( EmulsiFlex-C3; Avestin ) in buffer A [50 mM Tris-HCl pH 7 . 5 , 500 mM NaCl , 1 mM MgCl2 , 0 . 2% ( v/v ) Triton X-100 , 10% ( v/v ) glycerol , 20 mM imidazole] containing protease inhibitors ( 2 mM PMSF , 4 µg/ml pepstatin , 4 µg/ml leupeptin , 8 µg/ml aprotinin ) and 0 . 1 mg/ml DNaseI . The lysates were cleared by centrifugation ( 30 min at 25 , 000 g ) and incubated with 1 ml nickel affinity matrix ( Ni-NTA agarose; Qiagen ) per 1 l of expression culture for 2 hr at 4°C . The beads were washed five times with 20 bed volumes of buffer A sequentially supplemented with ( i ) 30 mM imidazole , ( ii ) 1% ( v/v ) Triton X-100 , ( iii ) 1 M NaCl , ( iv ) 5 mM Mg2+-ATP , or ( v ) 0 . 5 M Tris-HCl pH 7 . 5 . Bound BiP proteins were eluted in buffer B [50 mM Tris-HCl pH 7 . 5 , 500 mM NaCl , 1 mM MgCl2 , 10% ( v/v ) glycerol , 250 mM imidazole] and dialyzed against HKM buffer ( 50 mM HEPES-KOH pH 7 . 4 , 150 mM KCl , 10 mM MgCl2 ) . The purified proteins were concentrated using centrifugal filters ( Amicon Ultra , 30 kDa MWCO; Merck Millipore , UK ) , snap-frozen in liquid nitrogen and stored at -80°C . The purification of mature hamster BiP ( 19-654 ) was described in detail in Preissler et al . , 2015 . Mammalian expressed GST-tagged recombinant FICDE234G was produced in HEK293T cells . The cells were grown on 10 cm tissue culture dishes to a density of 80% and transfected either with plasmid encoding only GST or GST-FICDE234G ( UK1415 ) using the polyethylenimine ( PEI ) method with the following adaptations: 14 µg plasmid DNA was added to 700 µl Opti-MEM ( Gibco; Life Technologies , UK ) mixed with 56 µl of PEI ( 1 mg/ml stock ) and added dropwise to the cells after incubation for 20 min at room temperature ( ten dishes were usually transfected per preparation ) . Thirty-six hr after transfection the cells were lysed as described above with Triton lysis buffer [20 mM HEPES-KOH pH 7 . 4 , 150 mM NaCl , 1 mM EDTA , 10% ( v/v ) glycerol , 1% ( v/v ) Triton X-100] containing protease inhibitors . The combined lysate ( from ten dishes ) was incubated with 200 µl GSH-Sepharose beads ( Glutathione Sepharose 4B; GE Healthcare , UK ) for 3 hr at 4°C . The beads were recovered by centrifugation for 5 min at 500 g , washed twice with Triton lysis buffer containing 300 mM NaCl and twice with AMPylation buffer [25 mM HEPES-KOH pH 7 . 4 , 100 mM KCl , 4 mM MgCl2 , 1 mM CaCl2 , 0 . 1% ( v/v ) Triton X-100] . The beads with bound protein were stored in AMPylation buffer containing 50% ( v/v ) glycerol at -20°C and used for radioactive in vitro AMPylation reactions . For bacterial expression of N-terminally GST-tagged FICD proteins and ERdj6 J-domains , plasmid DNA encoding active GST-FICDE234G ( UK1479 ) , inactive GST-FICDE234G-H363A ( UK1480 ) , GST-J ( UK185 ) or mutant GST-J* ( UK186 ) was transformed into BL21 T7 Express lysY/Iq E . coli cells ( New England BioLabs cat . # C3013 , UK ) . Bacterial cultures were grown at 37°C in LB medium containing 100 µg/ml ampicillin to OD600 nm 0 . 8 , shifted to 20°C and expression was induced with 0 . 1 mM IPTG for 16 hr . Afterwards , the cells were harvested and lysed as described above in lysis buffer [50 mM Tris-HCl pH 7 . 5 , 500 mM NaCl , 1 mM MgCl2 , 2 mM dithiothreitol ( DTT ) , 0 . 2% ( v/v ) Triton X-100 , 10% ( v/v ) glycerol] containing protease inhibitors and DNaseI . The lysates were cleared by centrifugation ( 30 min at 25 , 000 g ) and incubated with 0 . 7 ml GSH-Sepharose beads per 1 l of expression culture for 3 hr at 4°C . The beads were transferred to a column and washed with 20 ml wash buffer B [50 mM Tris-HCl pH 7 . 5 , 500 mM NaCl , 1 mM DTT , 0 . 2% ( v/v ) Triton X-100 , 10% ( v/v ) glycerol] containing protease inhibitors , 20 ml wash buffer C [50 mM Tris-HCl pH 7 . 5 , 300 mM NaCl , 10 mM MgCl2 , 1 mM DTT , 0 . 1% ( v/v ) Triton X-100 , 10% ( v/v ) glycerol] containing protease inhibitors and 20 ml wash buffer C sequentially supplemented with ( i ) 1% ( v/v ) Triton X-100 , ( ii ) 1 M NaCl , ( iii ) 3 mM ATP , ( iv ) or 0 . 5 M Tris-HCl pH 7 . 5 . Bound BiP proteins were then eluted in elution buffer [50 mM HEPES-KOH pH 7 . 4 , 100 mM KCl , 4 mM MgCl2 , 1 mM CaCl2 , 0 . 1% ( v/v ) Triton X-100 , 10% ( v/v ) glycerol , 40 mM reduced glutathione] and concentrated protein solutions were frozen in liquid nitrogen and stored at -80°C . N-terminally His6-Smt3 tagged wildtype mouse FICD lumenal domain ( L104-P458 ) ( UK1564 ) was encoded on a pET28b plasmid ( Novagen , UK ) and expressed in E . coli BL21 T7 Express lysY/Iq cells . Bacterial cultures ( 3 liters ) were grown at 37°C in LB medium containing 100 µg/ml ampicillin to OD600 nm 0 . 8 and expression was induced with 0 . 1 mM IPTG . After incubation for 16 hr at 20°C the cells were sedimented by centrifugation and pellets were lysed as described above in lysis buffer [50 mM Tris-HCl pH 7 . 4 , 500 mM NaCl , 1 mM MgCl2 , 10% ( v/v ) glycerol , 30 mM imidazole , 1 mM tris ( 2-carboxyethyl ) phosphine ( TCEP ) ] supplemented with protease inhibitors and DNaseI . The lysates were cleared by centrifugation for 30 min at 33 , 000 g , followed by addition of 0 . 2% ( v/v ) Triton X-100 and 0 . 002% ( v/v ) sodium deoxycholate , and incubated with 1 ml Ni-NTA agarose ( Qiagen ) for 2 . 5 hr at 4°C . The beads were washed with 25 bed volumes of wash buffer [50 mM Tris-HCl pH 7 . 4 , 500 mM NaCl , 10% ( v/v ) glycerol , 0 . 2% ( v/v ) Triton X-100 , 1 mM TCEP , protease inhibitors] , once with buffer D [50 mM Tris-HCl pH 7 . 4 , 300 mM NaCl , 10% ( v/v ) glycerol , 0 . 1% ( v/v ) Triton X-100 , 10 mM MgCl2 , 1 mM TCEP , and protease inhibitors] , and then with buffers Ei-iv [buffer D supplemented with ( i ) 1% ( v/v ) Triton X-100 , ( ii ) 1 M NaCl , ( iii ) 3 mM ATP , or ( iv ) 0 . 5 M Tris-HCl pH 7 . 4] . Bound mFICD protein was eluted in buffer F [50 mM Tris-HCl pH 7 . 4 , 100 mM NaCl , 4 mM MgCl2 , 1 mM CaCl2 , 10% ( v/v ) glycerol , 250 mM imidazole , 1 mM TCEP ) , and dialyzed against 50 mM HEPES-NaOH pH 7 . 4 , 100 mM NaCl , 0 . 5 mM 2-mercaptoethanol in the presence of Ulp1 ( 2 µg per mg of eluted protein ) for 16 hr at 4°C to cleave the His6-Smt3 tag . The dialyzed and cleaved protein was then diluted 1:2 in 25 mM HEPES-NaOH pH 8 . 1 to reduce the ionic strength , loaded on a anion-exchange chromatography column ( Mono Q 5/50 GL; GE Healthcare ) and eluted with a linear salt gradient from 50 to 500 mM NaCl in 25 mM HEPES-NaOH pH 7 . 4 . The mFICD-containing fractions were pooled and further purified by size-exclusion chromatography ( Superdex 200 10/300; GE Healthcare ) in gel filtration buffer ( 25 mM HEPES-NaOH pH 7 . 4 , 150 mM NaCl ) . The eluted fractions were combined and provided to Aves Labs ( Tigard , OR ) as an immunogen for production of chicken anti-FICD antibodies . The absence of FICD protein in AR42j FICD-/- cells was confirmed by immunoaffinity purification as the endogenous protein could not be detected by immunoblotting of total lysates . Protein extracts from wildtype and FICD-/- AR42j cells were prepared in HG lysis buffer containing protease inhibitors , cleared , normalized and equal volumes of the lysates ( 4 mg total protein ) were incubated with 15 µl UltraLink Hydrazine Resin ( Pierce cat . # 53149 , UK ) on which FICD-specific chicken IgY antibodies have been covalently immobilized according to the manufacturer’s instructions ( chicken anti-FICD ) or rabbit polyclonal IgG antibodies against human FICD ( rabbit anti-FICD; LifeSpan BioSciences ) bound to Protein A Sepharose 4B beads ( Invitrogen cat . #10-1041 ) , for 16 hr at 4°C . The beads were then recovered by centrifugation for 1 min at 1 , 000 g and washed three times with HG lysis buffer . Bound proteins were eluted in 40 µl 2 x SDS sample buffer for 10 min at 70°C and equal volumes of the samples were loaded on a 12% SDS polyacrylamide gel , and endogenous FICD was detected by immunoblotting with rabbit anti-FICD or chicken anti-FICD antibodies . Samples of the normalized lysates ( 25 µg ) were loaded and BiP was detected as an “input” control . Unless indicated otherwise , in vitro AMPylation was performed in AMPylation buffer ( described above ) by incubation of purified wildtype or mutant BiP proteins at 20 µM with bacterially expressed GST-FICDE234G ( or as a control with inactive GST-FICDE234G-H363A ) at 0 . 8 µM in presence of 1 . 5 mM ATP for 45 min at 30°C . The final volume of a typical reaction was 15 µl . The samples were then treated with or without the BiP linker-specific protease SubA at 30 ng/µl for 10 min at room temperature . Volumes corresponding to 7 µg BiP protein were loaded per lane on native or SDS polyacrylamide gels and visualized by Coomassie staining . The in vitro AMPylation time-course experiments presented in Figure 7C and D were performed likewise but after incubation for the indicated times at 30°C the reactions were stopped on ice by addition of 40 mM EDTA and samples were run immediately on native gels . Radioactive in vitro AMPylation was performed by mixing 7 . 5 µg of purified wildtype or mutant BiP proteins with 10 µM non-radiolabeled ATP and 0 . 185 MBq α-32P-ATP ( EasyTide; Perkin Elmer , UK ) in a total volume of 30 µl in AMPylation buffer . The reactions were incubated with 5 µl GSH-Sepharose beads coupled with mammalian expressed GST-FICDE234G ( or GST only ) in 1 . 5 ml reaction tubes at 30°C on a thermomixer ( Eppendorf ) while shaking at 800 rpm . After 45 min [in these experiments a time before the reaction went to completion to reveal differences in AMPylation amongst BiP mutants ( Figure 5A and C and Figure 7A ) ] the beads were sedimented by centrifugation and the reaction was terminated by addition of SDS loading buffer to the supernatants and heating for 5 min at 75°C . Equal volumes of the samples were loaded on 12% SDS polyacrylamide gels . Proteins were visualized by Coomassie staining and radioactive signals were detected with a Typhoon Trio imager ( GE Healthcare ) upon overnight exposure of the dried gels to a storage phosphor screen . To test the sensitivity of AMPylated BiP towards cleavage by SubA ( Figure 3—figure supplement 1 ) , 6 µg of purified BiPT229A were labeled in 60 µl AMPylation buffer containing 6 . 7 MBq of α-32P-ATP and 10 µl of GST-FICDE234G-coupled beads as described above . After incubation for 30 min at 30°C non-radioactive ATP was added at 5 µM and the reaction was allowed to proceed for another 60 min . The beads were then sedimented by centrifugation and 1 . 5 mM ATP was added to the supernatant . To remove nucleotides the sample was passed through a Sephadex G-50 MicroSpin column ( illustra AutoSeq G-50; GE Healthcare ) equilibrated with AMPylation buffer . 1 . 5 µg of the recovered protein was mixed with 60 µg unmodified BiPT229A ( at 0 . 6 mg/ml final ) in presence of 1 . 5 mM ATP and the combined sample was divided into fractions to which SubA ( from a dilution series ) was added to yield final concentrations between 0 . 08 and 120 ng/µl . After incubation for 30 min at room temperature the proteins were denatured with SDS sample buffer and equal volumes were applied to SDS-PAGE , Coomassie-stained and signals were detected by autoradiography . Twenty mg of purified wiltype BiP or BiPV461F protein were in vitro AMPylated for 2 hr at 30°C with 1 mg bacterially expressed GST-FICDE234G in presence of 1 . 5 mM ATP in AMPylation buffer in a final volume of 12 ml . BiP proteins were then bound to 400 µl nickel affinity matrix , washed with AMPylation buffer and eluted in AMPylation buffer containing 250 mM imidazole . The concentrated eluate was divided into aliquots and immediately frozen in liquid nitrogen and stored at -80°C . For applications that required complete absence of detergent or imidazole the affinity chromatography eluate was passed through a Centri•Pure P25 desalting column ( emp BIOTECH , Germany ) equilibrated in HKM buffer followed by concentration of the proteins using centrifugal filters ( Amicon Ultra , 30 kDa MWCO; Merck Millipore ) before freezing . The in vitro modified proteins were used for mass spectrometry analyses and functional assays ( see below ) . Unmodified BiP from parallel mock in vitro AMPylation reactions , to which no enzyme has been added , served as controls in all assays . The experimental strategy for the stable isotope labeling by amino acids in cell culture ( SILAC ) experiment is outlined in Figure 6A and samples were prepared as follows: Wildtype and FICD-/- CHO-K1 cells were adapted to Ham’s F12 medium minus L-arginine and L-lysine for SILAC ( Pierce cat . # 88424 ) supplemented with 10% ( v/v ) dialyzed fetal bovine serum ( Gibco; Life Technologies cat . # 26400-044 ) , 1 x Penicillin-Streptomycin ( Sigma ) , 2 mM L-glutamine ( Sigma ) , 280 mg/l L-proline ( Sigma cat . # P5607-25G ) , 62 . 5 mg/l “light” L-lysine monohydrochloride ( Sigma cat . # L8662-25G ) and 60 . 5 mg/l “light” L-arginine monohydrochloride ( Sigma cat . # A6969-25G ) and incubated as described above . Once adapted , the cells were cultured in SILAC medium containing either 60 . 5 mg/l “light” ( as above ) or “heavy” L-arginine monohydrochloride ( R10; U-13C6 , U-15N4; Cambridge Isotope Laboratories , Inc . cat . # CNLM-539 , Andover , MO ) for several passages ( > 15 cell divisions ) before expansion in five 10 cm dishes per sample . The cells were grown to ~70% confluence and medium was exchanged 14 hr before treatment with or without 100 µg/ml cycloheximide for 3 hr . After washing with ice-cold PBS the cells were collected in PBS containing 1 mM EDTA and lysed in Triton lysis buffer [20 mM HEPES-KOH pH 7 . 4 , 150 mM NaCl , 1 mM EDTA , 1% ( v/v ) Triton X-100 , 10% ( v/v ) glycerol] containing protease inhibitors . The protein concentrations of the cleared lysates were normalized and equal volumes were then mixed ( 1:1 ratio ) as indicated . Each combined sample was pre-cleared for 15 min at 4°C with 30 µl blocked UltraLink Hydrazine Resin followed by immunoaffinity purification for 2 hr at 4°C with 30 µl of the same resin to which BiP-specific IgY antibodies were covalently coupled ( Preissler et al . , 2015 ) . The beads were washed twice for 10 min with lysis buffer , once with lysis buffer containing 350 mM NaCl and once again with lysis buffer , before elution in 2x non-reducing SDS sample buffer for 10 min at 70°C . Afterwards , the beads were removed by centrifugation and the supernatants were supplemented with 125 mM DTT , heated for 5 min at 70°C and loaded on a 12% SDS polyacrylamide gel . The gel was stained with Coomassie and bands corresponding in size to BiP were cut out for in-gel digest with Arg-C endopeptidase , and analysis by mass spectrometry . Wildtype and FICD-/- CHO-K1 cells were grown in standard medium ( ten 10 cm dishes per sample ) and treated for 3 hr with 100 µg/ml cycloheximide before cell lysis with Triton lysis buffer supplemented with protease inhibitors and BiP was immunoaffinity purified as described above with the following modifications: The lysates ( ~4 mg protein per sample ) were pre-cleared with 40 µl blocked UltraLink Hydrazine Resin for 30 min at 4°C followed by incubation for 2 hr with 60 µl of the resin carrying covalently attached BiP-specific IgY antibodies . The beads were washed sequentially for 10 min with lysis buffer , lysis buffer containing 350 mM NaCl , HG buffer ( see above ) containing 1 mM ATP , and once again with lysis buffer . Bound protein was eluted in 150 µl of 0 . 2 M glycine-HCl pH 2 . 5 for 10 min at 16°C followed by neutralization of the recovered sample supernatants with 1/10 volume of 1 . 5 M Tris-HCl pH 8 . 8 . Afterwards , protein was precipitated with isopropanol and analyzed by mass spectrometry . Molecular masses of intact BiP proteins were measured by electrospray ionization mass spectrometry using a Q-Trap 4000 ( ABSciex ) after ‘on-line’ reverse-phase chromatographic purification of samples as described previously ( Carroll et al . , 2009 ) . The instrument was operated in MS mode , and was calibrated with a mixture of myoglobin and trypsinogen . Reconstructed molecular masses were calculated with Bioanalyst ( ABSciex ) and masses from amino acid sequences with MassLynx ( Waters ) . To analyze masses of BiP peptides , proteins were reduced , alkylated and digested “in-gel” . In vitro AMPylated BiP protein was digested with the proteolytic enzymes as indicated and SILAC samples were digested using Arg-C . The resulting peptides were analyzed by LC-MS/MS using either an Orbitrap XL coupled to a nanoAcquity UHPLC or a Q Exactive coupled to a RSLC3000 UHPLC . Data was processed in Proteome Discoverer 1 . 4 using Sequest to search a UniProt E . coli database ( downloaded 29/04/15 , 4 , 378 entries ) with the sequence for Chinese hamster ( Cricetulus griseus ) BiP added or a Chinese Hamster database ( downloaded 19/02/14 , 23 , 884 entries ) . Oxidation ( M ) , deamidation ( N/Q ) and AMPylation ( S/T ) were set as variable modifications and carbamidomethylation ( C ) as a fixed modification . FDR calculations were performed by Percolator and peptides were filtered to 1% . Peak areas were determined using the Precursor Ion Quantifier node in Proteome Discoverer . Purified untagged hamster BiP19-654 at 15 µM was in vitro AMPylated with 0 . 75 µM GST-FICDE234G ( or as a control with inactive GST-FICDE234G-H363A ) in presence of 1 . 5 mM ATP for the indicated times at 30°C as described above . The reactions were then treated without or with 30 ng/µl SubA for 10 min at room temperature in a final volume of 20 µl and analyzed by IEF as described previously ( Chambers et al . , 2012; Laitusis et al . , 1999 ) with modifications: The samples were diluted with 180 µl of IEF sample buffer [8 M urea , 5% ( w/v ) CHAPS , 50 mM DTT , 2% ( v/v ) Pharmalyte ( pH 4 . 5-5 . 4; GE Healthcare ) ] and loaded on a 3 . 75% polyacrylamide gel containing 8 . 8 M urea , 1 . 25% ( w/v ) CHAPS and 5% ( v/v ) Pharmalyte . The loaded samples were overlaid with 0 . 5 M urea and 2% ( v/v ) Pharmalyte solution before separation . 10 mM glutamic acid and 50 mM histidine were used as anode and cathode buffers , respectively . After the run ( 100V , 10 min; 250V , 1 hr; 300V , 1 hr; 500V , 30 min ) proteins were transferred to a nitrocellulose membrane in blotting buffer [25 mM Tris-HCl pH 9 . 2 , 190 mM glycine , 0 . 01% ( w/v ) SDS , 10% ( v/v ) methanol] for 3 hr at 300 mA and immunodetected with BiP-specific antibodies as described above . Lysate samples from mammalian cells ( CHO-K1 and AR42j ) were prepared for analysis by IEF as follows: Cells were grown in 10 cm dishes to approximately 90% confluence , washed with ice-cold TBS , resuspended in 1 ml TBS , sedimented by centrifugation , and lysed in 30 x its packed cell volume of IEF lysis buffer [8 . 8 M urea , 5% ( w/v ) CHAPS , 1 µM sodium pyrophosphate , 2 mM imidodiphosphate , 50 mM DTT and 2% ( v/v ) Pharmalyte] at room temperature for 5 min . The lysates were then centrifuged at 20 , 238 g for 10 min at room temperature , the supernatants were transferred to a new tube and then centrifuged again for 60 min . The cleared lysates ( 50 µl ) were then purified using Sephadex G-50 MicroSpin columns equilibrated with IEF sample buffer , diluted 1:2 in IEF sample buffer and 15 µl were loaded on a 3 . 75% polyacrylamide gel followed by BiP immunoblotting as described above . Malachite green ( MG ) reaction solution was prepared freshly by mixing stock solutions of MG dye [0 . 111% ( w/v ) malachite green , 6 N sulphuric acid] , 7 . 5% ( w/v ) ammonium molybdate and 11% ( v/v ) Tween 20 in a 10:2 . 5:0 . 2 ratio . Samples of purified BiP proteins ( 5 µM ) were incubated in HKM with 3 mM ATP in a final volume of 20 µl for 60 min at 30°C . 15 µl of each sample were then diluted with 135 µl water on a 96-well plate , mixed with 50 µl MG reaction solution and incubated for 2 min at room temperature . 20 µl of a 34% ( w/v ) sodium citrate solution were added to quench the reactions and after incubation for further 30 min the absorbance was measured at 612 nm with a plate reader ( Infinite F500; Tecan ) . Standard curves were prepared with serial dilutions of KH2PO4 and served as a reference to calculate Vmax values . Statistical analysis by unpaired t-test was performed using Graphpad Prism version 6 . Binding of substrate peptide by purified BiP proteins was measured as previously described ( Chambers et al . , 2012; Marcinowski et al . , 2011 ) with modifications: Unmodified or in vitro AMPylated BiP proteins were incubated at the indicated concentrations ( between 0 and 40 µM ) with 1 µM lucifer yellow ( LY ) -labeled substrate peptide ( HTFPAVLGSC ) in presence of 1 mM ADP for 24 hr at 30°C in FP buffer [37 . 5 mM HEPES-KOH pH 7 . 4 , 125 mM KCl , 7 mM MgCl2 , 0 . 5 mM CaCl2 , 125 mM imidazole , 0 . 05% ( v/v ) Triton X-100] . 20 µl of each sample were transferred to a 386-well polystyrene microplate ( µClear , black; greiner bio-one cat . # 781096 , UK ) and fluorescence polarization of the lucifer yellow fluorophore was measured ( excitation λ = 430 nm; emission λ = 535 nm ) at room temperature in a plate reader . For analysis of competitive substrate dissociation , BiP proteins ( 40 µM ) were incubated with LY-labeled substrate peptide ( 1 µM ) as described above . Afterwards , 400-fold molar excess of unlabeled substrate peptide ( HTFPAVL ) was added to the pre-formed BiP:substrate complexes and the change in fluorescence polarization was measured over time . From each value the background signal of a reference sample containing only LY-labeled peptide ( without BiP protein ) were subtracted . RNA from CHO-K1 cells was extracted using RNA Stat-60 ( Amsbio , UK ) according to the manufacturer’s instructions . 1 µg RNA was reverse transcribed with Oligo ( dT ) 18 primer using RevertAid Reverse Transcriptase ( Thermo Scientific , UK ) . Quantitative PCR analysis was performed using the Power SYBR Green PCR Master Mix ( Applied Biosystems ) according to the manufacturer’s instructions on a 7900HT Fast Real-Time PCR System ( Applied Biosystems ) . FICD , RPL27 ( 60S ribosomal protein L27 ) and PPIA ( cyclophilin A ) were amplified using the following primers: FICD-F 5’-GGGCTGCTGACTGTTAAGACTA-3’ , FICD-R 5’-CCTTGACGCTTCATCTCCA-3’ , RPL27-F 5’-ACAATCACCTCATGCCCACAAG-3’ , RPL27-R 5’-GCGTTTCAGGGCTGGGTCTC-3’ , PPIA-F 5’-TTCCTCCTTTCACAGAATTATCCC-3’ , and PPIA-R 5’-CTGCCGCCAGTGCCATTATG-3’ . Relative quantities of amplified cDNAs were then determined using SDS 2 . 4 . 1 software ( Applied Biosystems ) and normalized to PPIA mRNA . | Newly made proteins are unstructured chains of amino acids that must fold into particular shapes to work effectively . Proteins that are destined to be exported from the cell undergo folding in a compartment of the cell called the endoplasmic reticulum . This folding is assisted by chaperones ( which are themselves also proteins ) . Cells adjust the number of chaperones so that there are enough to cope with the burden of unfolded proteins . However , in the endoplasmic reticulum , the known mechanisms that regulate the production of chaperones are too slow to track the rapid fluctuations in the production of unfolded proteins . This suggests that other means exist to balance active chaperones and unfolded proteins that go beyond merely controlling chaperone or unfolded protein abundance . An important chaperone protein of the endoplasmic reticulum , called BiP , is chemically modified when the production of unfolded proteins declines , and loses the modification when more unfolded proteins are produced . This suggests that the modification might adjust BiP’s activity so that it can handle the unfolded proteins that are present . However , previous studies have failed to agree about the nature of the chemical modification and how it affects how BiP works . Preissler , Rato et al . compared the activity of BiP in normal mammalian cells and in cells engineered to lack an enzyme called FICD . This enzyme attaches a molecule of adenosine mono-phosphate ( AMP ) to proteins in a process known as AMPylation . The experiments revealed that AMPylation is the modification of BiP that tracks how many unfolded proteins are in the cell . Further studies showed that AMP attaches to a single amino acid of BiP , number 518 , a threonine . Reconstructing the AMPylation of threonine 518 in a test tube caused the modified BiP to lose its ability to engage with unfolded proteins . Overall , Preissler , Rato et al . ’s results indicate that cells inactivate BiP by AMPylating threonine 518 as the number of unfolded proteins decreases , and remove the modification to re-activate BiP in response to mounting levels of unfolded proteins . Further studies are now needed to determine how AMPylation inactivates BiP and to understand how the FICD enzyme is regulated so that it performs AMPylation at the right time . It also remains to be explored how important the regulation of BiP activity by AMPylation is for living cells . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"biochemistry",
"and",
"chemical",
"biology",
"cell",
"biology"
] | 2015 | AMPylation matches BiP activity to client protein load in the endoplasmic reticulum |
Subcellular asymmetry directed by the planar cell polarity ( PCP ) signaling pathway orients numerous morphogenetic events in both invertebrates and vertebrates . Here , we describe a morphogenetic movement in which the intertwined socket and shaft cells of the Drosophila anterior wing margin mechanosensory bristles undergo PCP-directed apical rotation , inducing twisting that results in a helical structure of defined chirality . We show that the Frizzled/Vang PCP signaling module coordinates polarity among and between bristles and surrounding cells to direct this rotation . Furthermore , we show that dynamic interplay between two isoforms of the Prickle protein determines right- or left-handed bristle morphogenesis . We provide evidence that , Frizzled/Vang signaling couples to the Fat/Dachsous PCP directional signal in opposite directions depending on whether Pkpk or Pksple predominates . Dynamic interplay between Pk isoforms is likely to be an important determinant of PCP outcomes in diverse contexts . Similar mechanisms may orient other lateralizing morphogenetic processes .
PCP signaling controls the polarization of cells within the plane of an epithelium , orienting asymmetric cellular structures , cell divisions and cell migration . In flies , PCP signaling controls the orientation of trichomes ( hairs ) on the adult cuticle , orientation of ommatidia in the eye , and orientation of cell divisions , though the full range of phenotypic outputs has not been explored . While much focus has been placed on mechanistic studies in flies , medically important developmental defects and physiological processes in vertebrates are also under control of PCP signaling , motivating mechanistic studies in flies that might inform similar studies in vertebrates . Defects in the core PCP mechanism result in open neural tube defects , conotruncal heart defects , deafness , situs inversus and heterotaxy ( reviewed in Butler and Wallingford , 2017; Henderson et al . , 2018; Blum and Ott , 2018 ) . PCP is also believed to participate in both early and late stages of cancer progression and in wound healing . PCP polarizes skin and hair , the ependyma and renal tubules . Paralogs of the PCP component Prickle are mutated in an epilepsy-ataxia syndrome ( Tao et al . , 2011; Mei et al . , 2013; Bassuk et al . , 2008; Ehaideb et al . , 2014; Paemka et al . , 2015 ) . Mutations in ‘global’ PCP components have been associated with a human disorder of neuronal migration and proliferation ( Zakaria et al . , 2014 ) and in developmental renal disorders ( Zhang et al . , 2019 ) . Work in Drosophila indicates that at least two molecular modules contribute to PCP signaling . The core module acts both to amplify molecular asymmetry , and to coordinate polarization between neighboring cells , producing a local alignment of polarity . Proteins in the core module , including the serpentine protein Frizzled ( Fz ) , the seven-pass atypical cadherin Flamingo ( Fmi; a . k . a . Starry night ) , the 4-pass protein Van Gogh ( Vang; a . k . a . Strabismus ) , and the cytosolic/peripheral membrane proteins Dishevelled ( Dsh ) , Diego ( Dgo ) , and the PET/Lim domain protein Prickle ( Pk ) adopt asymmetric subcellular localizations that predict the morphological polarity pattern such as hairs in the fly wing ( reviewed in Zallen , 2007; Butler and Wallingford , 2017 ) . These proteins communicate at cell boundaries , recruiting one group to the distal side of cells , and the other to the proximal side , through the function of an incompletely understood feedback mechanism , thereby aligning the polarity of adjacent cells . A global module serves to provide directional information to the core module by converting tissue level expression gradients to asymmetric subcellular Fat ( Ft ) - Dachsous ( Ds ) heterodimer localization ( reviewed in Matis and Axelrod , 2013; Butler and Wallingford , 2017; Zallen , 2007 ) . The atypical cadherins Ft and Ds form heterodimers which may orient in either of two directions at cell-cell junctions . The Golgi resident protein Four-jointed ( Fj ) acts on both Ft and Ds as an ectokinase to make Ft a stronger ligand , and Ds a weaker ligand , for the other . Graded Fj and Ds expression therefore result in the conversion of transcriptional gradients to subcellular gradients , producing a larger fraction of Ft-Ds heterodimers in one orientation relative to the other . Other less well defined sources of global directional information appear to act in partially overlapping , tissue dependent ways ( Wu et al . , 2013; Sharp and Axelrod , 2016 ) . Various Drosophila tissues depend primarily on either of two isoforms of Pk , Prickleprickle ( Pkpk ) and Pricklespiny-legs ( Pksple ) ( Gubb et al . , 1999 ) . These isoforms have been proposed to determine the direction in which core PCP signaling responds to directional information provided by the Ft/Ds/Fj system ( Olofsson and Axelrod , 2014; Hogan et al . , 2011; Ambegaonkar and Irvine , 2015; Ayukawa et al . , 2014 ) . Pksple binds directly to Ds , orienting Pksple-dependent core signaling with respect to the Ds and Fj gradients ( Ayukawa et al . , 2014; Ambegaonkar and Irvine , 2015 ) , while Pkpk-dependent core signaling has been proposed to couple less directly through a mechanism in which the Ft/Ds/Fj module directs the polarity of an apical microtubule cytoskeleton on which vesicles containing core proteins Fz , Dsh and Fmi undergo directionally biased trafficking ( Matis et al . , 2014; Olofsson and Axelrod , 2014; Shimada et al . , 2006; Harumoto et al . , 2010 ) . Others , however , have argued that Pkpk-dependent core signaling is instead uncoupled from the Ft/Ds/Fj signal ( Merkel et al . , 2014; Ambegaonkar and Irvine , 2015; Casal et al . , 2006; Lawrence et al . , 2007; Brittle et al . , 2012 ) . The most intensively studied morphogenetic responses to PCP signaling in Drosophila occur in epithelia such as wing and abdomen , in which cellular projections called trichomes ( hairs ) grow in a polarized fashion from the apical surface , and in ommatidia of the eye , in which photoreceptor clusters achieve chirality via directional cell fate signaling . Chaete ( bristles ) , which serve as sensory organs , are also polarized by PCP signaling ( Schweisguth , 2015; Gubb and García-Bellido , 1982 ) . Bristles comprise the 4–5 progeny of a sensory mother cell ( SMC ) , one of which , the shaft , extends a process above the epithelium such that it tilts in a defined direction with respect to the tissue . In mechanosensory microchaete on the notum , one daughter of the SMC divides to produce the shaft and a socket cell that surrounds the shaft where it emerges from the epithelial surface; the other SOP daughter divides to produce a glial cell , a sheath cell and a neuron . Studies of microchaete polarity have shown that the initial division of the SMC is polarized by PCP in the epithelium from which it derives , such that the two daughters are born in defined positions with respect to each other ( Gho and Schweisguth , 1998; Bellaïche et al . , 2001 ) . However , subsequent events that ultimately determine the direction of shaft polarity have not been described . We chose to study bristle polarity on the anterior margin of the wing ( AWM ) . A row of stout mechanosensory bristles and a row of curved chemosensory bristles are on the dorsal surface , and a mixed row of mechano and chemosensory bristles is on the ventral side ( Figure 1A , Figure 1—figure supplement 1A ) . All of these bristles tilt toward the distal end of the wing in wildtype . In pkpk mutant wings , and in wings overexpressing Pksple , a large fraction of the AWM bristles point proximally rather than distally; the pkpk phenotype is suppressed by mutation of dsh , implicating the core PCP signaling mechanism in this process ( Gubb et al . , 1999 ) . However , the morphogenetic process resulting in polarity and the genetics of its apparent control by PCP signaling have not been explored . Here , focusing on the dorsal mechanosensory bristles , we report our analysis of the underlying morphogenesis leading to AWM bristle polarization , and show that polarization results from a corkscrew-like helical morphogenetic process involving the shaft and socket cells . Furthermore , our results reveal how interplay between Pkpk and Pksple control the handedness of the helical growth , and how the Ft/Ds/Fj system directs it in opposite orientations depending on whether the core PCP mechanism operates in a Pkpk- or Pksple-dependent mode .
To begin to characterize the determinants of AWM mechanosensory bristle polarity , we labeled the externally exposed socket and shaft cells in wildtype ( wt ) and pkpk mutants with anti-Su ( H ) ( Suppressor of hairless ) antibody and phalloidin , respectively . In wildtype control wings at 36 hr apf , the apical ends of socket cells are tilted toward the distal end of the wing and are interspersed with the actin bundles of the shafts ( Figure 1C , C’ , Figure 1—figure supplement 1B–B’’ ) . The shafts are interspersed between socket cells , and appear to ascend along the proximal side of the adjacent socket cell and pass through an opening at its apical surface . Consistent with polarity patterns of adult bristles , actin bundles of pkpk mutant shafts near the distal end of the wing show a reversed , proximal , tilt , whereas shafts in the proximal and the very most distal regions show the normal distal tilt ( Figure 2A , A’ , Figure 1—figure supplement 1C–C’’ ) . Between these regions , shafts show a smooth transition between proximal and distal tilt , with some shafts pointing straight up ( neutral tilt ) . The socket cells tilt at angles that correlate with the polarity of neighboring actin bundles in pkpk mutant wings: P-D tilt at the proximal region and D-P tilt at the distal region , with smooth transition between those regions ( Figure 1—figure supplement 1C–C’’ ) . Furthermore , shaft actin bundles with reversed polarity appear to be positioned on the distal side of the socket cell through which they pass , opposite to their relationship in wildtype and to their relationship in the proximal region in pkpk mutant wings where their tilt shows the normal , distal , direction ( Figure 1—figure supplement 1B–B’’ , compare with Figure 1—figure supplement 1C–C’’ ) . Previous studies of AWM bristle ultrastructure have been insufficiently detailed to appreciate the determinants of polarity ( Hartenstein and Posakony , 1989; Palka et al . , 1979 ) . To better understand the structures of shaft-socket pairs , and to unambiguously determine the relationship between sibling shaft-socket cell pairs , random individual bristle lineages were labeled by clonal expression of RFP , filling the cell bodies of labeled shaft and socket cells . Simultaneous staining of the socket cells ( identified by Su ( H ) expression ) allows one to identify the sibling shaft-socket cell pairs . 3D confocal reconstructions facilitate examination from a variety of viewpoints and enable the positions and shapes of cell bodies , nuclei , shaft , and apical opening of the socket cells to be visualized ( Figure 1B–B’’’ , Figure 1—video 1 ) . These views allow us to see that the wildtype nucleus and main portion of the shaft cell is dorsal and extends slightly posterior to that of the sibling socket cell . The base of the shaft rises from the cell body , wraps clockwise along its socket sibling ( as viewed from the apical side in a right wing ) , and then rises through a groove in the socket that extends apically along the proximal side of the socket , finally emerging through the donut shaped apical surface of the socket cell ( Figure 1D–D’’ , Figure 1—video 1 ) . In contrast , in the distal bristles of pkpk mutant wings where bristles are reversed , the shaft is positioned within an oppositely oriented groove on the distal side of the sibling socket cell . The shaft nucleus is dorsal and posterior to the socket nucleus , similar to their arrangement in wildtype , though their alignment is not as regular . Therefore , the structure of pkpk mutant bristles is roughly a mirror image of that in wildtype , with nuclei in similar positions , but the shaft bending in a counterclockwise direction and rising through a groove on the opposite side of the socket cell compared to wildtype ( Figure 2B–B’’ ) . In the proximal region of the pkpk mutant wing , the relationship of shaft-socket siblings resembles that in wildtype , reflecting the normal bristle polarity in that region ( Figure 1—figure supplement 1C–C’’ and Figure 3—figure supplement 1B ) . Therefore , the P or D position of shaft relative to the sibling socket cell correlates to the normal or reversed bristle polarity in proximal and distal regions of pkpk mutants , suggesting that the shaft position relative to the socket determines bristle polarity ( tilt ) . Since the relative position of the shaft to the socket appears to be important for bristle polarity , we wished to identify the developmental process by which this relationship is achieved . In microchaete of the notum , the orientation of the initial division of the sensory organ precursor cell is specified by PCP signaling ( Schweisguth , 2015 ) . Assuming the orientations of subsequent daughter cell divisions are similarly regulated , the shaft-socket relationship may be determined by their relative positions at their birth . A similar process might occur in AWM mechanosensory bristles . Alternatively , a post-division morphogenetic process may determine their final configurations . To examine this process , we analyzed 3D structures of shaft-socket sibling pairs at earlier developmental stages . In wildtype bristles at 24 hr apf , the shaft cell nucleus is posterior and just slightly dorsal to the socket cell nucleus , similar to their positions at 36 hr apf ( Figure 3A , B ) . The extending shaft is just reaching the apical surface , and is positioned in a groove on the dorsal side of the socket cell . At the apical surface , the shaft , sits in a shallow indentation in the apical surface of the socket cell , which adopts a crescent shape with the opening of the crescent pointing in the dorsal direction . Over time , from 28 to 32 to 36 hr apf , as the shaft continues to extend above the apical surface , the socket cell crescent rotates clockwise , and gradually closes to form a ring around the shaft ( Figure 3A , B , Table 2 , Table 1 , Figure 3—videos 1 , 2 , 3 ) . Because the apical surface rotates while the nuclei remain relatively stationary , the shaft and socket twist to form a left-handed helical shape . As the shaft grows out above the socket cell , it points distally . The cell bodies , are initially relatively flat in the dorsal-ventral direction , but as they grow , extend dorsally , becoming flatter in the anterior-posterior direction . Quantification of rotation was performed by measuring rotation angles , as diagrammed in Figure 3B , from 3D images of socket cells captured at different time points ( c . f . Figure 3C ) , and data are displayed in rose plots ( Figure 3F , Figure 2—figure supplement 1A , Table 2 ) . Rotation at 32 hr averages 55° in the clockwise direction . Note that throughout , we describe analyses of right wings . In all cases , left wings develop as mirror images of right wings . Based on stereotypical patterns from images of timed , fixed samples of the apical surface , we infer that a margin cell at the dorsal side of the shaft-socket pair rotates together with the pair toward the proximal side , generating new junctions with the neighboring socket and margin cells and widening the gap between shaft-socket pairs ( Figure 2—figure supplement 1B ) . During these events , the number of margin cells surrounding the shaft-socket pair is maintained and cell-cell junctions are remodeled . To characterize the events leading to reversed bristle polarity in pkpk mutants , rotation angles of shaft-socket pairs during development were analyzed as above . At 24 hr apf , shaft cells were positioned dorsal to their socket sibling cells , as in wildtype . At later times , apical rotation proceeded counterclockwise , opposite to the wildtype direction , in the distal bristles that adopt a reversed polarity , giving rise to a right-handed helical shape in contrast to the left-handed helical shape of wildtype pairs ( Figure 3E , Figure 2—figure supplement 1A , Table 3 ) . Shaft-socket pairs in the proximal region rotated clockwise , corresponding to their normal polarity , and bristles in the region between normal and reversed bristles rotated very little , corresponding to their neutral , upright , polarity ( Figure 3—figure supplement 1B ) . Therefore , bristle polarity does not depend on the birth positions of shaft and socket cells , which are born by 14 hr apf ( Hartenstein and Posakony , 1989 ) , but rather on the direction of apical rotation of the developing shaft-socket pair , beginning shortly after 24 hr apf . Since pkpk mutants show counterclockwise rotation leading to reversed shaft-socket positioning ( D-P ) and bristle reversal , and Pksple overexpression similarly reverses bristle polarity as previously reported ( Figure 3—figure supplement 1E , E’ and Gubb et al . , 1999 ) , we surmised that Pksple induces the counterclockwise helical growth that leads to D-P orientation of shaft-socket pairs and reversed bristle polarity . Consistent with this , counterclockwise rotation of shaft-socket pairs and proximal polarity in pkpk mutants is suppressed by removing pksple ( in pkpk-sple13/pk-sple13; Figure 3—figure supplement 1A–D’ , Table 3 , Gubb et al . , 1999 ) , and pksple overexpression induces counterclockwise rotation similar to that in pkpk mutants ( Figure 3—figure supplement 1E , E’ Table 2 , Table 3 ) . Pksple is therefore needed to reverse shaft-socket rotation in pkpk mutants . Overexpression of Pksple induced counterclockwise rotation and reversed polarity in a wider region than in pkpk mutants ( compare Figure 3—figure supplement 1B to E ) . Thus , endogenous Pksple is only poised to act at the distal margin , but exogenous Pksple can reverse most , if not all , bristles . Though the potential for Pksple to reverse bristle polarity is unmasked in the absence of Pkpk , it plays no essential role in wildtype polarization , as pksple bristles fully rotate , or perhaps marginally over-rotate ( 59 . 3°±4 . 2° vs 54 . 6°±2 . 9° , p=0 . 0702; Table 3 , Figure 3—figure supplement 1C , C’ ) . Surprisingly , pkpk-sple13 mutant bristles rotate only moderately less than wildtype bristles ( 47 . 6°±6 . 6° vs . 54 . 6°±2 . 9° , p=0 . 0143; Table 3 ) , suggesting that Pkpk might play only a modest role in controlling the magnitude of rotation in wildtype . We propose that this is due to residual core PCP signaling activity observed in the absence of Pk ( Strutt and Strutt , 2007; Lawrence et al . , 2004; Adler et al . , 2000 ) , and the implications of this result are considered more fully in the Discussion . We have shown that in wildtype bristles , Pkpk antagonizes Pksple to direct clockwise rotation of shaft-socket pairs , and that Pksple , when overexpressed , outcompetes Pkpk to direct counterclockwise rotation . Although the idea that Pkpk and Pksple antagonize each other has been previously proposed ( Gubb et al . , 1999 ) , how this occurs has been obscured in part by the inability to specifically visualize the endogenous expression of each isoform . We therefore modified the endogenous genomic sequence encoding Pkpk or Pksple by appending a V5 tag to the N-terminus , facilitating the tissue and cellular level evaluation of their native expression patterns at various developmental stages . Both tagged isoforms support wildtype polarity development in all tissues and various controls suggest that expression of these genomically tagged versions reflect that of the native loci ( Figure 4 , Figure 4—figure supplements 1 and 2 ) . Here , we describe their expression in developing wings . Notably , V5::Pk and V5::Sple reveal that throughout wing development , expression is spatiotemporally dynamic . Early in wing development , Pkpk protein is strongly expressed and is present in most or all cells . In discs , Pkpk is relatively elevated in AWM proneural cells . Following a slight dip in levels around 8 hr apf , expression levels climb , peaking around 32 hr apf , when wing hairs emerge , and then decline with little detectable expression remaining by 40 hr apf ( Figure 4A–A’’ , Figure 4—figure supplement 1B–H , K–K’’ ) . Expression of Pksple is below detection in discs ( Figure 4—figure supplement 1B , I , J ) , makes a small peak at around 8 hr apf , and is then not detectable between 16 and 28 hr apf . Beginning around 28 hr apf , Pksple expression is specifically detected in dorsal AWM cells and in vein L3 . This expression persists through 32 hr apf , when wing hairs emerge , and weaker expression in other veins becomes apparent ( Figure 4B–B” , Figure 4—figure supplement 1B , L , L’ ) . At these times , Pksple is below detection levels on Western blots . Beginning sometime after 32 hr apf , Pksple expression increases in most cells , with its level equaling that of the declining Pkpk by 36 hr apf and reaching its highest level at around 40 hr apf ( the latest time we examined ) , when Pkpk is no longer detected . At 40 hr apf , when Pksple is at peak expression in most of the wing blade , expression has disappeared from vein cells ( Figure 4—figure supplement 1B , L–L” ) . Most pertinent to bristle development , at the 24 hr apf AWM , Pkpk is expressed at apical junctions at similar levels in anterior margin and socket cells ( Figure 4C , C’ , Figure 4—figure supplement 2A ) . By 28 hr apf , Pkpk expression begins to decrease during bristle-socket rotation , first in socket cells , and later in all cells at and near the margin ( Figure 4—figure supplement 2A ) . At the same time , Pksple expression becomes evident and increases over time , first uniformly in cells near the margin , and gradually becoming strongest in the socket and shaft cells ( Figure 4D , D’ , F–F’’ , Figure 4—figure supplement 2B , compare with Pkpk ( V5::Pk ) in Figure 4E ) . The timing of the shift from Pkpk to Pksple expression is accelerated at the margin relative to the interior of the wing . Mosaic experiments demonstrate that Pkpk localizes proximally in socket and other margin cells , and that , importantly , Pksple colocalizes with Pkpk to the proximal side of socket cells . Proximal Pksple localization is an unexpected observation based on previous studies in which overexpressed Pksple localized at the distal junctions of hair cells ( Ayukawa et al . , 2014; Ambegaonkar and Irvine , 2015 ) . Because Pksple expression is stronger in socket cells compared to margin cells by 32 hr apf , it has the useful property of effectively being expressed as a mosaic , allowing its localization to be scored without inducing clones ( Figure 4F–F’’ , H–H’’ , I–I’’ ) . As Pksple activity reverses rotation direction of shaft-socket pairs in pkpk mutant wings , Pksple protein localization was analyzed in pkpk mutant wings . Because anti-Pk[C] antibody recognizes the common region of Pkpk and Pksple , the antibody reveals Pksple isoform localization in pkpk mutants . In the region of pkpk mutant wings where P-D reversal occurs , Pksple protein localized at the distal side of socket cells , whereas in the proximal region where polarity is not reversed , Pksple protein shows minimal asymmetry ( Figure 4G–G’’ , Figure 4—figure supplement 2C–C’’’ ) . Distal localization of Pksple in the region of polarity reversal was verified by clonal knockdown of pksple in pkpk mutant wings . Notably , the socket-shaft pairs that lacked both Pksple and Pkpk failed to rotate ( Figure 4—figure supplement 2D–E’’’ ) . These results suggest that Pkpk normally inhibits Pksple from localizing distally by recruiting it to the proximal junction of socket cells ( and likely also in nearby margin cells , although this is hard to visualize due to lower expression in those cells ) . Furthermore , the distal localization of Pksple in pkpk mutants correlates with its ability to determine counter-clockwise rotation of shaft-socket pairs on a cell-by-cell basis , suggesting that this localization is likely the determinant of counter-clockwise rotation . Our results thus far show that the direction of Pk polarization , whether Pkpk , Pksple or both , corresponds to the direction of bristle polarization . Suppression of polarity reversal in pkpk mutants by dsh implicates the core PCP signaling mechanism in this process Gubb et al . ( 1999 ) . We therefore asked whether the remaining components of the core PCP signaling mechanism contribute to AWM bristle polarization . As with dsh mutation , knock down of fz or vang in pkpk mutants suppressed reversal of bristle tilt , shaft-socket orientation , and rotation direction ( Figure 5A–D ) . Core PCP signaling is therefore required for reversed , Pksple-dependent polarity . To assess a potential contribution of core PCP signaling to Pkpk-dependent bristle polarization , the anterior region of adult wings from fmi RNAi , fz , vang , and dsh mutants were analyzed , and the rotation of shaft-socket pairs was evaluated for each genotype ( Figure 5E–H ) . Adult mechanosensory bristles of core PCP mutants are less tilted toward the distal direction than those of wildtype and the tilting angles are somewhat irregular , with some bristles tilting out of the plane of the wing . Consistent with the adult wing defects , in pupal wings the shaft position relative to the socket varies , sometimes abruptly , in the same mutant wing , showing less local correlation than in wildtype . Quantification reveals substantial under-rotation of shaft-socket pairs , and a broader distribution of rotation angles than in wild type ( Figure 5E–H , compare with Figure 3D , F , Table 2 ) . Thus , careful morphological analysis reveals that core PCP signaling is required for normal , Pkpk-dependent polarity as well as reversed , Pksple-dependent polarity . Consistent with a role for core PCP components in mediating rotation of shaft-socket pairs , junctional asymmetry of Fz::EGFP and Vang::EYFP is well preserved between socket and margin cells and between adjacent margin cells ( Figure 5I–J’ ) . Little accumulation was observed at junctions between shaft and socket cells . Mosaic analyses demonstrated the expected distal localization of Fz and proximal localization of Vang at both margin cell-margin cell and margin cell-socket cell junctions , suggesting that PCP signaling likely occurs between margin cells and between margin and socket cells ( Figure 5K , L ) . Similarly , the reversed bristle polarity observed in pkpk mutants was accompanied by reversed Vang localization in pkpk mutant socket and margin cells , consistent with the idea that the direction of core PCP polarization is reversed in pkpk mutants ( Figure 5—figure supplement 1 ) . To functionally test polarity propagation between margin and socket cells by core PCP components , fz or vang knock-down clones were generated , and clones at the AWM were analyzed ( Figure 6 , Figure 6—figure supplement 1 ) . fz or vang knock-down clones , whether in just margin cells , just shaft-socket pairs , or both , showed non-autonomy as assessed by sequestration of Fz ( fz clones ) or Vang ( vang clones ) at the clone borders . Near fz RNAi clones , distal cells , including both bristle and hair cells , were re-oriented: sockets on the distal side of the clones showed counter-clockwise rotation , and hairs on the distal side grew toward the clones . Similarly , sockets on the proximal side of vang RNAi clones rotated counter-clockwise ( Figure 6—figure supplement 1 ) . These results indicate that core PCP signaling propagates between bristle and margin cells to control the rotation direction of shaft-socket pairs . We have previously proposed that a signal from the Ft/Ds/Fj system provides a directional cue to orient core PCP signaling in some tissues ( Ma et al . , 2003; Yang et al . , 2002; Matis et al . , 2014; Olofsson et al . , 2014 ) , although others have argued that this system operates in parallel with core PCP signaling ( Casal et al . , 2006; Lawrence et al . , 2007; Brittle et al . , 2012 ) . An asymmetry of Ft-Ds heterodimers , with a small excess of Ds displayed on the distal side of the cell , and Ft on the proximal side , has been observed , and is proposed to provide this signal ( Ambegaonkar et al . , 2012; Bosveld et al . , 2012; Brittle et al . , 2012 ) . We have also proposed that the core PCP module differentially interprets directional signals from Ds when operating in Pkpk- or Pksple-dependent modes , with Pksple directing localization of the Fmi-Vang complex to the side where Ds is in excess , while the Fmi-Vang complex localizes to the opposite side when functioning in a Pkpk-dependent manner [ ( Olofsson and Axelrod , 2014 ) ; see also Lawrence et al . ( 2004 ) . Pksple has been shown to bind to Ds and the associated Dachs protein , providing a mechanism for orienting Pksple-dependent core function to the Ft/Ds/Fj signal ( Ayukawa et al . , 2014; Ambegaonkar and Irvine , 2015 ) , whereas a less direct , microtubule-dependent mechanism was proposed to mediate this response when Pkpk is predominant ( Shimada et al . , 2006; Harumoto et al . , 2010; Olofsson and Axelrod , 2014; Matis et al . , 2014 ) . In contrast , some have proposed that Pkpk-dependent core signaling is instead uncoupled from the Ft/Ds/Fj signal ( Merkel et al . , 2014; Ambegaonkar and Irvine , 2015 ) . We reasoned that our reagents would allow us to analyze effects of the Ft/Ds/Fj signal on each Pk isoform . We first examined the phenotype resulting from knockdown of Ds ( Figure 7C ) . As previously observed ( Adler et al . , 1998 ) , normal bristle tilt was substantially disturbed . The pattern of disturbance consistently showed regions of coordinated polarity that smoothly transition through neutral polarity to adjacent regions of opposite polarity , although the number and position of those domains varied . The effectively mosaic expression of V5::Sple localization allows one to observe precisely correlated regions of distal and proximal Sple localization corresponding to the regions of reversed and normal polarity , respectively , with relatively symmetric localization in the intervening transitions ( Figure 7C ) . These results suggest that local core PCP signaling maintains polarity correlation among immediate neighbors but that alignment to the tissue axis is eliminated in the absence of Ds . We then asked if the Pksple-dependent reversed polarity in pkpk mutants depends on Ds . When ds was knocked down in pkpk mutants , bristle reversal was blocked , producing a phenotype similar to that of core mutants , and the distal localization of Pksple was no longer observed ( Figure 7A–D ) . Therefore , the reversed , Pksple-dependent polarity in pkpk mutants requires Ds , and we interpret this to indicate that the Ds global signal recruits Pksple to sites of enriched Ds ( distal ) in the absence of Pkpk ( Figure 7B; compare with 7D ) , which drives reversal of shaft-socket orientation and therefore reversal of bristle polarity . In ds knockdowns , we are unable to readily interpret the localization pattern of V5::Pk because levels are similar in socket and margin cells . Nonetheless , other results suggest that the local polarity correlation in the absence of Ds is mediated primarily by asymmetric localization of Pkpk rather than the asymmetric localization of Pksple that we can observe . First , recall that in wildtype , Pksple is recruited to colocalize with Pkpk at proximal sites , so Pkpk is likely to similarly recruit the colocalization of Pksple in the absence of Ds . Consistent with this idea , when ds was knocked down in pkpk mutants , neither proximal nor distal localization of Pksple was observed in socket cells , and local correlation of bristle polarity was weak ( Figure 7D ) . Thus , the local domains of correlated asymmetric Pksple localization in ds knock-down socket cells depend on the presence of Pkpk; Pksple alone is insufficient to facilitate local signaling between neighbors . Finally , removing Pksple in ds knock-down wings failed to significantly modify ds knock-down effects on the polarity of bristles ( and also hairs ) while removing both Pkpk and Pksple does ( Figure 7E , F , Figure 7—figure supplement 1A–D ) , confirming that ds knock-down affects the Pkpk-mediated , rather than the Pksple-mediated , PCP signal for wing bristle ( and hair ) polarity . Taken together , these observations suggest that the locally correlated domains of polarity observed in ds knock-down wings depend on Pkpk activity . These and previous results indicate that Pkpk is the principal isoform functioning in core PCP signaling during bristle polarization . They are most consistent with , though do not definitively show , that in wildtype , the Ds global signal directs orientation of core signaling such that Vang and Pkpk localize to the proximal side ( and incidentally colocalizing Pksple to the proximal side ) to establish normal polarity . The alternative possibility is that Ds activity is permissive , and some other signal directs this orientation of Pkpk-dependent core polarization . The proposal that Ds is instructive for orienting Pkpk-dependent core signaling , while consistent with our prior interpretation of coupling between the Ft/Ds/Fj signal and Pkpk-dependent core PCP signaling in wing hair polarization ( Ma et al . , 2003; Olofsson et al . , 2014; Sharp and Axelrod , 2016; Yang et al . , 2002 ) , is at odds with other reports asserting that while under Pkpk control , core PCP directionality is uncoupled from the Ft/Ds/Fj signal ( Merkel et al . , 2014; Ambegaonkar and Irvine , 2015 ) . Rigorous testing of this hypothesis requires reorienting the Ft/Ds/Fj signal and assessing the isoform dependence of the response . It was previously shown that reversing the gradient of Ds expression near the distal part of the wing under control of distal-less-GAL4 ( dll >2 x-ds ) reverses wing hair polarity ( Harumoto et al . , 2010 ) . Assuming that hair polarity is determined by Pkpk , for which ample evidence exists , and that it depends on core signaling , this result would demonstrate coupling of Pkpk-dependent core PCP signaling to the Ds signal . We rigorously tested this assumption by testing the core signaling and Pk isoform dependence of this response ( Figure 7G–K ) . dll >2 x-ds reverses polarity of a substantial swath of wing hairs , precisely in the region where the Dll expression gradient is expected to be steepest ( Figure 7—figure supplement 1E ) . dll >2 x-ds , however does not reverse AWM bristle polarity , as it does not produce a proximal-to-distal expression gradient at the AWM . Because our results show that AWM bristle and wing hair polarization show indistinguishable responses to Ft/Ds and to core PCP manipulations , we propose that wing hairs are a suitable readout for this assay . We first asked whether dll >2 x-ds-driven hair polarity reversal depends on core module activity by removing Fz , and found that ectopic Ds-dependent reversal is blocked in a fz mutant background ( Figure 7G , H ) . Furthermore , ectopic Ds re-orients the core PCP protein orientation ( Figure 7—figure supplement 1F–I ) . These results rule out the possibility that ectopic Ds reverses polarity through a pathway that does not include the core PCP module . We then tested the Pk isoform dependence of reversal , and found that it is almost entirely abolished upon removal of Pkpk ( pkpk or pkpk-sple ) , but is largely unchanged upon removal of Pksple ( pksple ) ( Figure 7G , I–K ) . This result decisively demonstrates that Pkpk-dependent core PCP signaling in wing hair polarization is oriented by the Ft/Ds/Fj signal , and strongly suggests that the same coupling occurs during Pkpk-dependent AWM bristle polarization .
Producing structures of defined chirality requires directional information on three Cartesian axes . Our results indicate that in determining bristle chirality , PCP provides directional information along the proximal-distal axis . The apical-basal axis is defined by the epithelium , while the dorsal-ventral axis is likely defined by the dorsal-ventral compartment boundary . We have shown that the polarity of wing margin bristles ( proximal or distal tilt ) is determined by controlling the handedness of helical growth . The entwined shaft and socket cells undergo an apical clockwise or counterclockwise rotation that results in a left-handed or right-handed helical structure , placing the shaft to the proximal ( wildtype ) or distal ( pkpk mutant ) side of the socket cell . The direction of rotation depends on PCP signaling among and between margin and socket cells . Helical cellular structures of defined handedness , such as the bristles resulting from properly directed rotation , have been noted in bacterial and plant species , but few examples have been described in animals . The entwined twisting of the shaft and socket is a coordinated morphogenetic event , and the apparent stereotyped junctional rearrangement of additional margin cells suggests that at least some other cells are involved as well . We do not know in which cell or cells mechanics are regulated to drive this morphogenesis . One possibility is that an internal cytoskeletal mechanism induces the helical growth of the socket and/or shaft cells . Another possibility is that the side of the socket cell crescent marked by Pk at 24 hr is anchored , while the other side grows to wrap around the shaft , inducing junctional rearrangements and propelling the rotation of the apical portion of the shaft relative to the socket cell . Apical rotation could then cause twisting of the more basal portions of the socket cell , and could in turn direct the shaft to the corresponding side of the socket cell . The precise location at which the PCP signal is required to determine rotation direction is unclear . Because we observe asymmetric core complexes at margin-socket cell junctions , but very little at shaft-margin or shaft-socket junctions , we hypothesize that interaction between the socket and surrounding margin cells is the essential determinant of rotation . PCP proteins at these junctions could control junctional dynamics , as is known to occur in other systems ( Huebner and Wallingford , 2018 ) . This will require further investigation . Core PCP signaling participates in regulating rotation , as the magnitude of rotation is substantially impaired in the core mutants fz , dsh , vang and fmi . Nonetheless , we note that a small amount of clockwise rotation still occurs in these mutants . We hypothesize that tissue scale mechanical forces may drive this rotation , though we do not rule out the possibility that some other signaling activity may also be involved . Compared to other core proteins , the impact of removing Pkpk on the magnitude of rotation is subtle . The clockwise rotation in pkpk-sple mutants is only slightly less than in wildtype . This result is reminiscent of Pkpk function in polarizing wing hairs: polarity in pkpk mutants is strongly perturbed due to the presence of Pksple , while hair polarity in pkpk-sple mutants is only weakly perturbed ( Gubb et al . , 1999 ) . These findings are consistent with previous proposals that the core PCP mechanism retains a residual capacity to propagate some asymmetry in the absence of Pk ( Strutt and Strutt , 2007; Lawrence et al . , 2004; Adler et al . , 2000 ) . The core PCP signal , in addition to executing directed rotation in response to Pkpk or Pksple , coordinates polarity between neighboring bristles . Local correlation between rotation angles is strong when core signaling is intact , even in the absence of the Ft/Ds signal , but is weak when core signaling is disrupted . This is analogous to the proposed mechanism for locally coordinating polarity between adjacent wing hairs . It is important to note that the local polarity signal must pass through intervening margin cells to signal from bristle to bristle . Our data suggest that whether the Pkpk or Pksple isoform dominates to control the direction of PCP signaling depends not only on the relative amounts of each isoform , but also on the dynamics of expression and its effect on competition for participation in the core complex . During rotation of AWM bristle shaft-socket pairs , both Pkpk and Pksple isoforms are detected at the apical junction of the socket with an inverse temporal relationship; high expression of Pkpk decreases during rotation , while the initially undetectable level of Pksple protein increases . In these conditions , the system is controlled by Pkpk , and both Pkpk and Pksple localize proximally , thus orienting the core complex in its wildtype configuration . We hypothesize that Pksple is recruited by Pkpk through their known ability to interact heterotypically ( Ayukawa et al . , 2014; Ambegaonkar and Irvine , 2015 ) . This ability of Pkpk and Pksple to colocalize has not been previously observed in wildtype conditions . Notably , however , in the wing , ectopic Pksple localization follows the expected position of Pkpk when Ds and Dachs cues are removed , though each were not independently visualized ( Ambegaonkar and Irvine , 2015 ) . Conversely , Pksple overexpression was seen to recruit Pkpk to the distal side of wing cells ( Ayukawa et al . , 2014 ) , reversing hair polarity as it does bristle polarity . We suggest that the temporal expression pattern in the AWM allows the system to initiate polarization under Pkpk control , and that the gradually accumulating Pksple colocalizes with Pkpk rather than outcompeting established proximal localization . Because bristles in pksple mutant wings fully polarize , the proximal Pksple is inconsequential for normal bristle polarization . In contrast , overexpression of Pksple , producing early and sustained high level expression , enables it to outcompete endogenous Pkpk and reverse polarity by driving localization to the distal side through its interaction with Ds and Dachs , likely recruiting Pkpk along with it . Similarly , in pkpk mutants , endogenous Pksple , free from recruitment to the proximal side , localizes distally and reverses polarity . We infer that during the critical period for determining bristle rotation direction in wildtype , Pksple does not reach a sufficient level to outcompete Pkpk and reverse the rotation . That pkpk mutation only reverses polarity of a region of distal bristles , whereas Pksple overexpression can reverse polarity of most or all AWM bristles , indicates that endogenous Pksple is only poised to act in a limited region of the margin . This may reflect subtle differences in the timing of its expression increase across the margin . Alternatively , it may reflect differences in the strength of the Ft/Ds/Fj signal across the margin . Our analyses do not have sufficient resolution to distinguish these possibilities . Dynamic isoform expression appears to have important consequences for other aspects of wing development . Hair polarity is determined by Pkpk ( at around 32 hr apf ) , but it can be inferred that some Pksple is already present , as is evident from the difference between the hair polarity patterns of pkpk and pkpk-sple mutants ( Gubb et al . , 1999 ) , and as confirmed by our expression analyses . We suggest that hair polarity does not fully reverse in pkpk mutants either because levels of Pksple are not yet high enough or because expression is primarily in veins and at the AWM at the time hair polarity is fixed . In contrast , the polarity of ridges , established later in wing development [ ( Merkel et al . , 2014 ) ; ( Doyle et al . , 2008 ) notwithstanding] , depends on Pksple . We propose that by the time ridge polarity is determined , the amount of Pksple has increased and the amount of Pkpk has decreased sufficiently to allow Pksple to exert control of ridge polarization . Though likely unimportant for normal development , the somewhat earlier expression of Pksple in veins relative to the intervein regions may contribute to polarity discontinuities observed in pkpk mutant wings , especially around L3 ( Gubb and García-Bellido , 1982; Hogan et al . , 2011; Merkel et al . , 2014 ) . Ft-Ds polarity appears to also be distorted around veins ( Merkel et al . , 2014 ) . Pkpk and Pksple expression dynamics are likely at play in determining the PCP response in other tissues as well . The idea that Ds controls the direction of core PCP signaling was first proposed by Adler based on wing hair polarity phenotypes ( Adler et al . , 1998 ) . We subsequently studied the Ft/Ds/Fj system and similarly concluded that it directs core PCP protein localization in the wing ( Ma et al . , 2003 ) , a Pkpk-dependent process , and polarization of ommatidia in the eye ( Yang et al . , 2002 ) , a Pksple-dependent process . We proposed that coupling in wing hair polarization is necessarily weak ( Ma et al . , 2003 ) , and the more recently proposed model in which Ft/Ds/Fj orient microtubules to orient directional trafficking of Fz , Dsh and Fmi-containing vesicles ( Matis et al . , 2014; Olofsson and Axelrod , 2014; Shimada et al . , 2006; Harumoto et al . , 2010 ) is consistent with a weak coupling mechanism in Pkpk-dependent processes . The finding of direct binding of Pksple to Ds and Dachs ( Ayukawa et al . , 2014; Ambegaonkar and Irvine , 2015 ) suggests a model for more direct and potentially stronger coupling of Pksple-dependent processes to the Ft/Ds/Fj system . The idea of coupling in Pk-dependent signaling has been controversial , and based largely on correlation , subsequent studies have led to the argument that Pksple-dependent core signaling is coupled , but Pkpk-dependent signaling is uncoupled from the the Ft/Ds/Fj system ( Merkel et al . , 2014; Ambegaonkar and Irvine , 2015 ) . Yet others have suggested that the Ft/Ds/Fj and core PCP systems always function in parallel rather than being coupled ( Lawrence et al . , 2007; Casal et al . , 2006 ) . Here , we report strong evidence that Pkpk-dependent core PCP signaling is responsive to the Ds signal , at least in polarizing wing hairs . We propose that the same is the case in polarizing bristles that are controlled by essentially similar responses to Pk isoforms and to the Ft/Ds/Fj system . In bristles , we directly observe the requirement for Ds to distally localize Pksple when Pkpk is absent , confirming Pksple coupling . The evidence that Pkpk-dependent core signaling is coupled to the upstream Ft/Ds/Fj signal is less apparent . In wildtype , correct bristle polarization requires Pkpk to prevent reversal by recruiting Pksple to the proximal side , though as noted above , proximal Pksple plays no essential role . But absent the need to antagonize Pksple coupling , is there evidence that core signaling in the presence of just Pkpk ( pksple mutant ) is coupled to Ft/Ds/Fj ? When the Ft/Ds/Fj system is intact , Pkpk localizes proximally , but without Ds or Ft , Pkpk localizes proximally and distally in random domains , driving domains of correct and reversed rotation analogous to the random but locally correlated domains of hair polarity in ft or ds mutant wing tissue ( Adler et al . , 1998; Ma et al . , 2003 ) . The same random domains of bristle polarity are seen when only Pkpk is available ( ds knockdown in a pksple mutant ) . This result demonstrates that the Ft/Ds/Fj system is required for correct polarization of the core PCP system while solely under Pkpk control , though it cannot distinguish a permissive from an instructive role . An instructive role is , however , concordant with its instructive role in directing Pkpk-dependent wing hair polarity . The proposal that Pkpk-dependent core signaling is coupled to and responds to the Ft/Ds/Fj signal in bristle polarization might at first appear to conflict with the observation that properly oriented rotation proceeds to a significant extent in the absence of both Pkpk and Pksple . This is explained by pointing out that our model for coupling invokes Ft/Ds/Fj directed microtubule-based transport of Fz and Dsh , but that the involvement of Pkpk is indirect ( Matis et al . , 2014; Olofsson and Axelrod , 2014; Shimada et al . , 2006; Harumoto et al . , 2010 ) . As have others , we propose that Pkpk functions to amplify the asymmetry introduced by this transport , but that some asymmetry , and communication of polarity information between cells , can still occur in its absence ( Strutt and Strutt , 2007; Lawrence et al . , 2004; Adler et al . , 2000 ) . In other words , Ft/Ds/Fj coupling to Pkpk-dependent core PCP signaling is not directed by Pkpk , but rather , is permitted to occur because the Pksple-dependent mechanism is not operating to override it . Because core module function is required , this activity does not result from Pk isoform action influencing Ft/Ds/Fj output independent of core signaling , as has been recently suggested in another context ( Casal et al . , 2018 ) . It is important to caution that the model for the relationship between Ft/Ds/Fj and core signaling presented here does not necessarily extend to their relationship in other tissues where their interactions may well be different , and that experiments done in other tissues may not be directly relevant to wing hair and AWM bristles . The results presented here indicate that mapping the spatiotemporal dynamics of Pk isoform expression is essential to understanding how various developmental events can be differentially coupled to upstream global directional signals in a given tissue . Chirality , or left-right laterality , is a key feature of many organs in invertebrates and vertebrates . In Drosophila , rotation of the gut and of the male genitalia occurs in a defined direction to produce such laterality . In vertebrates , rotation of the gut and heart tube also leads to left-right asymmetry in these organs . In many cases , PCP has been implicated in control of this lateralization ( Blum and Ott , 2018 ) . In the Drosophila hindgut , both core PCP and the Ft/Ds system play essential roles in directing normal dextral rotation ( González-Morales et al . , 2015 ) . Though the forces that drive these organ rotations are not well understood , left-right asymmetries in actomyosin distribution , cell shape , and localization of other cellular structures , together with PCP dependence ( Harris , 2018; Blum and Ott , 2018 ) , indicate that chirality at the cellular level is an important determinant of rotational direction . Indeed , chirality of isolated cells from looping chick heart has been directly demonstrated ( Ray et al . , 2018 ) . We therefore propose that regulation of chiral shaft-socket cell pair rotation may share much in common with the mechanisms that determine larger organ laterality , and its investigation could therefore yield insights that will enlighten understanding of organ rotation and laterality .
Drosophila melanogaster flies were grown on standard cornmeal/agar/molasses media at 25°C . FLP-on ( using the actP >CD2>GAL4 construct for trans-gene expression ) and FLP/FRT mitotic clones were generated by incubating third-instar larvae at 37°C for 1 hr . 36 to 48 hr later , white prepupae were collected and aged to the desired developmental time point prior to dissection and fixation . Drosophila mutant alleles and transgenic stocks are described in the Key resources table and detailed chromosomes and genotypes are provided below . pkpk-sple13 ( FBst0044230 ) , pkpk-sple14 ( Gubb , 1993 ) , pkpk30 ( FBst0044229 ) , pksple1 ( FBst0000422 ) , vangA3 ( Taylor et al . , 1998 ) , vangstbm6 ( FBst0006918 ) , fzR52 ( Krasnow and Adler , 1994 ) , dsh1 ( FBst0005298 ) , UAS-pksple ( FBst0041780 ) , UAS-pkRNAi ( VDRC ID: 101480 ) , UAS-fmiRNAi ( FBst0026022 ) , UAS-fzRNAi ( FBst0034321 ) , UAS-vangRNAi ( FBst0034354 ) , UAS-dsRNAi ( FBst0032964 ) , UAS-ds ( Matakatsu and Blair , 2004 ) , dll-Gal4 ( FBst0030558 ) , MS1096-Gal4 ( FBst0008860 ) , armP-fz::EGFP ( Strutt , 2001 ) , actP-vang::EYFP ( Strutt , 2002 ) , actP >CD2>vang::EYFP ( Strutt , 2002 ) , ci-Gal4 ( Croker et al . , 2006 ) , UAS-mCherry ( FBst0038424 ) , actP >CD2>Gal4 UAS-RFP ( FBst0030558 ) . Pupal wings were dissected at indicated developmental time points after puparium formation ( apf ) . Pupae were removed from their pupal cases and fixed for 60–90 min in 4% paraformaldehyde in PBS at 4°C . Wings were then dissected and extracted from the cuticle , and were washed two times in PBST ( PBS with 0 . 1% Triton X-100 ) . After blocking for 1 hr in 5% Bovine serum Albumin in PBST at 4°C , wings were incubated with primary antibodies overnight at 4°C in the blocking solution . Incubations with secondary antibodies were done for 90 min at room temperature in PBST . Washes in PBST were performed three times after primary and secondary antibody incubation , and incubations in phalloidin ( 1:200 dilution ) in PBST were done for 15 min followed by wash at room temperature before mounting if required . Stained wings were mounted in 15 μl Vectashield mounting medium ( Vector Laboratories ) . Primary antibodies were as follows: goat polyclonal anti-Su ( H ) ( 1:200 dilution , Santa Cruz , sc-15183 ) , mouse monoclonal anti-V5 ( 1:200 dilution , Thermo-fisher , R960-25 ) , guinea pig polyclonal anti-Pk[C] ( 1:800 dilution , Olofsson et al . , 2014 ) , rat monoclonal anti-dEcad ( 1:200 dilution , DSHB ) . Secondary antibodies from Thermo Fisher Scientific were as follows: 488-donkey anti-mouse , 488-goat anti-guinea pig , 546-donkey anti-goat , 633-goat anti-guinea pig , 633-goat anti-rat , 647-donkey anti-mouse . Alexa 635 and Alexa 350 conjugated phalloidin were from Thermo Fisher Scientific . Adult wings were dissected and washed with 70% EtOH and mounted in DPX ( Sigma ) solution . All adult wings were imaged on a Nikon Eclipse E1000M equipped with a Spot Flex camera ( Model 15 . 2 64 MP ) . All immunofluorescence images were taken with a Leica TCS SP8 AOBS confocal microscope and processed with LAS X ( Leica ) and Adobe Photoshop . For three dimensional wing margin images , 50 to 100 z-stacks , each with 0 . 2 μm thickness , were collected and combined using 3D reconstitution software ( Leica ) . Scale bars are not provided for three dimensional images due to errors introduced by perspective , but approximate scale can be inferred from related two- dimensional images . To measure the apical rotation angles of socket cells , a horizontal line linking centers of circles around apical surfaces of socket cells was drawn , and perpendicular lines intersecting the center of each socket cell apex were drawn ( black lines in Figure 3C ) . Vectors from each socket cell center passing through the center of the apical opening of the socket circles ( blue vectors in Figure 3C ) were drawn and angles between the vertical lines and the vectors were measured with Image J software . Statistical analysis was performed and rose plots generated using Oriana four software . Comparisons were made using Student’s t-test and p values are reported . Summary statistics are provided in Table 2 . For qualitative results such as expression patterns , a minimum of 20 biological replicates from at least two independent experiments were examined and representative images are shown . To generate the donor plasmid with homology arms of the pkpk genomic sequence and the V5::3Xmyc::APEX2 tag sequence , a 1 . 5 kb 3’ homology arm ( HR2 ) flanking the pkpk gRNA2 cleavage site was amplified and assembled into the pDsRed-attp ( Addgene , 51019 ) plasmid cut with SapI to make the pDsREd-attP-pkpkHR2 plasmid . To generate the donor plasmid for tagging pkpk , three DNA fragments including 5’ 1 . 5 kb homology arm ( HR1; containing a 1 . 2 kb homology arm flanking the pkpk gRNA1 cleavage site and a 5’ 0 . 3 kb sequence of the start codon ) , V5::3Xmyc::APEX2 tag with a linker sequence , and the fragment starting from the start codon of pkpk to the cleavage site targeted by the pkpk gRNA2 , were assembled into the pDsREd-attP-pkpkHR2 . To prevent the donor sequence from being cleaved by Cas9 , a point mutation was introduced in the PAM sequence of the HR1 using the NEB point-mutagenesis kit after sub-cloning the fragment into the pCR-Blunt-II-TOPO vector ( Thermo-Fisher , K280002 ) . The HR1 fragment bearing the mutant PAM sequence was then amplified for the assembly process . All three fragments were assembled into the pDsREd-attP-pkpkHR2 plasmid cut with EcoRI and NheI . To generate the donor plasmid for tagging pksple , similar strategies were applied . Briefly , 1 . 2 kb 5’ homology arm containing the mutant PAM sequence , the V5::3Xmyc::APEX2 tag with a linker sequence , and the fragment from the start codon of pksple to the cleavage site of pksple gRNA2 were assembled into the pDsREd-attP-pkspleHR2 ( bearing the 1 . 25 kb 3’ homology arm , HR2 ) plasmid . The donor plasmids containing the tag sequence and pkpk homology , or pksple homology , arms , were sequenced and then injected into the stable transgenic flies expressing two pkpk gRNAs , or pksple gRNAs , and nosCas9 , respectively , to generate recombinants . DsRed signal in the fly eyes was monitored for selecting the recombinants by BestGene , and dsRed and flanking sequences were removed by the Cre-Lox site-specific recombination method . The resulting modified alleles are referred to in the text as V5::Pk and V5::Sple for simplicity . Third-instar larval wing discs and pupal wings at appropriate developmental stages were dissected and lysed in protein loading buffer . Lysates from eight discs or wings were loaded per lane for SDS-PAGE analysis and western blots were performed using standard procedures . Antibodies: Guinea pig polyclonal anti-Pk[C] ( 1:1000 dilution , the same antibody used for immunostaining ) , mouse monoclonal anti-V5 ( 1:2000 dilution , the same antibody used for immunostaining ) , mouse monoclonal anti-γ-Tubulin ( 1:1000 dilution , Sigma-Aldrich , T6557 ) . Secondary antibodies were Peroxidase-conjugated goat anti-guinea pig ( 1:10000 ) and goat anti-mouse ( 1:10000 ) antibodies ( both from Jackson Immuno Research ) , and detection used SuperSignal West Pico Chemiluminescent Substrate ( Thermo-Fisher , 34080 ) | Our right and left hands are mirror images of each other and cannot be precisely superimposed . This property , known as chirality , is vital for many tissues and organs to form correctly in humans and other animals . For example , fruit flies have hair-like sensory organs on the edges of their wings known as bristles . One of the cells in each bristle forms a shaft that generally tilts away from the main body of the fly and is anchored in place by another cell known as the socket . A signaling pathway known as PCP signaling controls the directions in which many chiral tissues and organs in animals form . The pathway contains two signaling modules: the global module collects “directional” information about the orientation of the body and sends it to the core module , which interprets this information to control how the tissue or organ grows . Fruit flies have two different versions of one of the core module components – known as Prickle and Spiny legs – that are thought to alter the direction the core module responds to the information it receives . Mutant flies known as pkpk mutants are unable to make Prickle and their wing bristles tilt in the opposite direction compared to those in normal flies , but it was not clear exactly why this happens . To address this question , Cho et al . studied PCP signaling in the wings of normal and pkpk mutant flies . The experiments showed that Prickle directed the bristles on the right wing of a normal fly to grow in left-handed corkscrew-like patterns in which the emerging shaft and socket of each bristle twisted around each other . As a result , the bristles tilted away from the bodies of the flies . In the pkpk mutants , however , Spiny legs substituted for Prickle , causing the equivalent bristles to grow in a right-handed corkscrew pattern and tilt towards the body . The findings of Cho et al . show that PCP signaling controls the direction fly bristles grow by selectively using Prickle and Spiny legs . In the future , this work may also aid efforts to develop effective screening and treatments for birth defects that result from the failure of chiral tissues and organs to form properly . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"cell",
"biology"
] | 2020 | Prickle isoforms determine handedness of helical morphogenesis |
Lipids are critical to cellular function and it is generally accepted that lipid turnover is rapid and dysregulation in turnover results in disease ( Dawidowicz 1987; Phillips et al . , 2009; Liu et al . , 2013 ) . In this study , we present an intriguing counter-example by demonstrating that in the center of the human ocular lens , there is no lipid turnover in fiber cells during the entire human lifespan . This discovery , combined with prior demonstration of pronounced changes in the lens lipid composition over a lifetime ( Hughes et al . , 2012 ) , suggests that some lipid classes break down in the body over several decades , whereas others are stable . Such substantial changes in lens cell membranes may play a role in the genesis of age-related eye disorders . Whether long-lived lipids are present in other tissues is not yet known , but this may prove to be important in understanding the development of age-related diseases .
The membrane lipid composition of most tissues is dynamic and alters within days in response to diet ( Katan et al . , 1997; Owen et al . , 2004 ) and weeks in response to exercise ( Mitchell et al . , 2004 ) . Indeed , cellular phospholipid turnover can be detected within minutes after intravenous injection of fatty acids ( Thies et al . , 1994; Shetty et al . , 1996; Rapoport et al . , 1997 ) . However , in tissues where access to the metabolic machinery for membrane renewal is restricted , it is possible that some lipids are longer lived but this has never been shown . If long-lived lipids are present , the properties of the tissues in which they are incorporated may be influenced by the decomposition or modification of the lipidome over time . One tissue that can be used to examine this hypothesis is the ocular lens . There is no cellular turnover in the human lens ( Lynnerup et al . , 2008 ) and its center—a region known as the nucleus—is devoid of cellular organelles ( Bassnett and Beebe , 1992 ) . In this study , accelerator mass spectrometry ( AMS ) was utilized to measure carbon-14 ( 14C ) levels in lipids extracted from the human lens nucleus of donors covering a range of birth dates from 1948 to 1993 . This approach exploits the global pulse of artificial atmospheric 14CO2 resulting from above-ground testing of nuclear weapons that occurred from 1955 until 1963 ( Libby et al . , 1964; Harkness and Walton , 1972 ) . The concentration of 14C has decreased exponentially from 1963 until the present day due to the exchange of atmospheric CO2 in the oceans and biosphere . The lens grows continuously throughout life by the addition of fiber cells to a pre-existing lens that was present at birth ( the nucleus ) ( Lynnerup et al . , 2008 ) . Therefore , in the absence of turnover of cellular components , the 14C level of an individual's lens nucleus reflects the 14C abundance of the year in which they were born . If turnover occurs rapidly , 14C abundance reflects current levels present in the atmosphere , while an intermediate value would suggest a slower rate of exchange . Given this information , measurement of the amount of 14C present in a class of biomolecules allows the determination of the date of biosynthesis and the time the system ceased exchanging carbon with its surroundings , or the rate of exchange with its surroundings .
We carefully dissected nuclei from individual human lenses of 23 donors of known birth dates . The average human lens nucleus is approximately 6–7 mm in diameter ( Hermans et al . , 2007 ) . Therefore , to avoid contamination of fiber cells laid down postnatally , we cut a cylinder of 4 . 5 mm in diameter in the axial plane using a trephine , then removed 1 mm from either end as previously described ( Friedrich and Truscott , 2009 ) . Total lipids present in each lens nucleus were obtained using a well-established method that reports high yield of lipids and minimal protein contamination ( Folch et al . , 1957 ) . Since protein is the major component of lenses by mass and lens proteins are known to be present since birth ( Stewart et al . , 2013 ) , the residual protein content of lipid extracts was determined using a standard BCA assay ( Smith et al . , 1985 ) . Residual protein was found to represent less than 0 . 5% of the total weight of carbon in the extract , and therefore a negligible contribution to the 14C measurement . Radiocarbon content in lens lipids was determined by AMS and was found to closely match the atmospheric levels of the date of birth ( Figure 1A ) . 10 . 7554/eLife . 06003 . 003Figure 1 . Analysis of lens membrane lipid 14C content demonstrates a lack of molecular turnover . ( A ) The fraction of modern 14C present in the membrane lipids of human lens nuclear regions . The lipid samples ( • ) are superimposed over the levels of artificial 14CO2 present in the atmosphere in the northern hemisphere ( light gray ) and the southern hemisphere ( dark gray ) from 1950 until 1990 ( Hua et al . , 2013 ) . ( B ) The correlation between the predicted year of birth as calculated from the measured fraction of modern 14C present in lens membrane lipids and the actual year of birth of each individual . The slope was approximately one ( 0 . 98 ± 0 . 04 ) and the y-intercept was indistinguishable from zero within the measured error ( 39 ± 75 ) . Vertical error bars: ± sigma . Horizontal error bars: year of birth ± six months . DOI: http://dx . doi . org/10 . 7554/eLife . 06003 . 003 As shown in Figure 1B , the fraction of 14C ( f14C ) present in the lipid extract from a human lens nucleus was found to be a highly accurate predictor of the year of birth ( R2 = 0 . 9683 ) , indicating negligible lipid turnover during the human lifespan . The values obtained for f14C for human lens nuclear lipids ( and thus a predictor of the year of birth ) obtained in the present study are consistent with previous reports on the f14C for insoluble protein fraction ( Stewart et al . , 2013 ) and total protein ( Lynnerup et al . , 2008 ) of the human lens nucleus . Although radiocarbon dating has been previously utilized to examine the longevity of human tissues ( e . g . , tooth enamel [Spalding et al . , 2005] and lenses [Lynnerup et al . , 2008] ) and biomolecules ( e . g . , insoluble proteins [Stewart et al . , 2013] and DNA [Bhardwaj et al . , 2006; Spalding et al . , 2008; Bergmann et al . , 2009] ) , the results presented here are the first demonstration of the existence of long-lived lipids in any animal . The lack of molecular turnover in the human lens lipidome may be due to its unique pattern of growth . The lens grows continuously throughout life and pre-existing cells are encapsulated by newly formed fiber cells ( Bassnett and Beebe , 1992 ) such that the adult lens nucleus is comprised of cells biosynthesized in utero . The composition of human lens membranes is quite different from those of other mammals ( Borchman et al . , 2004 ) , possibly reflecting our long lifespan . In human lenses , sphingomyelins are more abundant than glycerophospholipids ( Borchman et al . , 1994 ) . The chemical stability of sphingomyelins is illustrated by the fact that they have been discovered , intact in a 40 , 000-year-old woolly mammoth ( Kreps et al . , 1979 ) . Furthermore , the higher degree of saturation found in sphingomyelins present in the human lens ( i . e . , predominantly dihydrosphingomyelins ) , may confer unique chemical and physical stability to the membranes of lens fiber cells ( Yappert and Borchman , 2004 ) . The data obtained here on lipid longevity are consistent with the results of quantitative molecular analysis of lipids isolated from human lens nuclei of different ages . In these studies , the total amount of glycerophospholipid classes such as phosphatidylcholines and phosphatidylethanolamines was found to decrease significantly with age , whereas the content of dihydrosphingomyelin and sphingomyelin remained relatively stable ( Hughes et al . , 2012 ) . Presumably , these results reflect the relative chemical stability of the different lipid classes under the conditions experienced within the lens over a lifetime . These findings reveal that some cells contain long-lived lipids and this discovery may have significant implications for other post-mitotic cells . If other post-mitotic cells contain long-lived lipids , their age-related deterioration may play a significant role in the properties of their respective individual organs , and possibly overall body function . With our aging population and increased prevalence of age-related diseases , a greater knowledge of long-lived biomolecules and their gradual deterioration is imperative .
HPLC-grade chloroform and methanol were purchased from Crown Scientific ( Moorebank , NSW , Australia ) . Analytical-grade sodium chloride and a Bicinchoninic Acid ( BCA ) assay kit were purchased from Sigma Aldrich ( Sydney , NSW , Australia ) . Glassware and stainless steel utensils were used throughout and were washed several times with 5% nitric acid , rinsed with deionized water , and dried at 70°C . Quartz tubes for lipid collection and combustion were baked before use at 600°C in a stream of pure oxygen for at least 4 hr . All work was approved by the human research ethics committees at the University of Sydney ( #7292 ) and the University of Wollongong ( HE 99/001 ) . The nuclear regions of human lenses ( n = 23 ) were obtained using a 4 . 5-mm trephine as described previously ( Friedrich and Truscott , 2009 ) . Donor year of birth was determined from Sydney Eye Bank records . Following dissection , each lens nucleus was homogenized in 1 ml of chloroform:methanol ( 2:1 vol/vol ) , and lipids were extracted according to Folch et al ( 1957 ) . The chloroform phase containing lipids was transferred into the baked quartz combustion tubes . Chloroform was removed by evaporation in a water bath at ∼50°C followed by drying under vacuum . Copper ( II ) oxide and silver wire that were previously pre-baked in oxygen were then added to the tubes , which were subsequently flame sealed . Lipids were then combusted in sealed tubes at 900°C overnight . CO2 was collected from the breakseals and dried by passing through a cryotrap ( −78°C ) . The amount of CO2 was determined and transferred into the small volume graphitization apparatus for graphite target production . 14C/12C isotopic ratios were measured on the Small Tandem for Applied Research ( STAR ) accelerator , which has greater than 0 . 5% precision for samples above 50 μg . Typical sample sizes were in the range of 70–120 μg of carbon . As the small weight of each sample made them susceptible to contamination , blanks that were subjected to the same procedural steps as the lens samples ( including extraction steps ) were processed with each batch of samples . Blanks produced a residual solvent carbon mass of 10–20 μg following evaporation and were measured for radiocarbon on the Australian National Tandem for Applied Research ( ANTARES ) accelerator ( Fink et al . , 2004 ) , which provides greater accuracy for samples less than 50 μg carbon . The isotopic ratio of 14C/12C or 14C/13C in samples was determined and normalized on the internationally agreed standard reference materials , oxalic acid I and oxalic acid II . Raw data were corrected for background count rate in the AMS instruments by measuring radiocarbon-free unprocessed commercial graphite and geological Ceylonese graphite . Each lens sample measurement was corrected for the mass of blanks as previously described ( Hua et al . , 2004 ) . Each batch of samples ( approximately 5–6 samples per batch ) was processed on different CO2 handling lines , resulting in variations in the precision of radiocarbon content that was measured . All procedures were initially optimized using bovine lens lipid extracts . Calendar dates were obtained by calibrating the radiocarbon determinations with the online version of the CALIbomb software ( 14CHRONO Centre of the Queen's University of Belfast , 2014 ) using the southern hemisphere data sets for the bomb pulse ( Hua et al . , 2013 ) and the tree ring southern hemispheric curve for data points prior to the 1950s ( Hogg et al . , 2013 ) . To ascertain whether the presence of lipids in solution could result in the retention of solvent , a separate batch of samples and blanks was processed with 13C-enriched methanol . Solvent mixtures with 13C enriched to 10% methanol ( corresponding to +900‰ δ13C ) were prepared . Lipids were extracted following standard procedures in parallel with unlabeled solvents and their δ13C determined by Isotope Ratio Mass Spectrometry ( IRMS ) . While both δ13C results were in the normal range of −20 to −24‰ , a small enrichment was observed at −22 . 7 ± 0 . 1 and −21 . 2 ± 0 . 7‰ for unlabeled and labeled solvents , respectively . Should these values represent isotopic exchange or retention of solvent by lipids , the carbon weight fraction of this contamination would be ∼0 . 013% and is therefore negligible . Lipid extracts ( n = 5 ) were dried under a stream of nitrogen at 37°C and reconstituted in 100 μl phosphate buffered saline . The amount of protein in each sample was determined using a standard BCA assay as described previously ( Smith et al . , 1985 ) and calculated as a fraction of the total amount of lipid in the extract . | Every cell is surrounded by a membrane made primarily of molecules called lipids . This membrane protects the cell and controls which molecules pass into and out of it . To keep the membrane in good working order , its lipids are regularly broken down and replaced with fresh molecules . However , some cells—such as the cells that make up most of the lens of the eye—lack easy access to the cell machinery that renews the membrane . The lens grows throughout life by adding new cells to the outside of the lens , but the center of the lens—also known as the lens nucleus—contains the same cells that were present at birth . This raises the question of whether the lipids in the membranes of these cells also remain in the cells for life . From 1955 to 1963 , above-ground test explosions of nuclear weapons caused a large amount of a radioactive form of carbon called carbon-14 to be released into the atmosphere . In subsequent years , these levels have decreased again as the carbon-14 is absorbed into the oceans or incorporated into biological molecules—like lipids . This doesn't affect the molecules , as carbon-14 works just like normal carbon . However , as the proportion of carbon-14 in a group of molecules reflects the amount of carbon-14 in the atmosphere when the molecule was made , this allows the age of the molecule to be determined . Hughes et al . used a technique called mass spectrometry to measure the carbon-14 in lens nuclei donated by 23 people who were born between 1948 and 1993 . This revealed that the proportion of carbon-14 in the total carbon content of the lipids in the nucleus could be used to accurately predict the year of birth of the donor . Therefore , the lipids in your lenses when you are born remain with you for your entire life . This finding could help us to understand age-related sight disorders , such as cataracts . Further research could also investigate whether there are any similarly long-lasting lipids in other body tissues , and whether these affect how other age-related diseases develop . | [
"Abstract",
"Introduction",
"Results",
"and",
"discussion",
"Materials",
"and",
"methods"
] | [
"short",
"report",
"cell",
"biology"
] | 2015 | No turnover in lens lipids for the entire human lifespan |
White adipose tissue ( WAT ) inflammation contributes to the development of insulin resistance in obesity . While the role of adipose tissue macrophage ( ATM ) pro-inflammatory signalling in the development of insulin resistance has been established , it is less clear how WAT inflammation is initiated . Here , we show that ATMs isolated from obese mice and humans exhibit markers of increased rate of de novo phosphatidylcholine ( PC ) biosynthesis . Macrophage-specific knockout of phosphocholine cytidylyltransferase A ( CCTα ) , the rate-limiting enzyme of de novo PC biosynthesis pathway , alleviated obesity-induced WAT inflammation and insulin resistance . Mechanistically , CCTα-deficient macrophages showed reduced ER stress and inflammation in response to palmitate . Surprisingly , this was not due to lower exogenous palmitate incorporation into cellular PCs . Instead , CCTα-null macrophages had lower membrane PC turnover , leading to elevated membrane polyunsaturated fatty acid levels that negated the pro-inflammatory effects of palmitate . Our results reveal a causal link between obesity-associated increase in de novo PC synthesis , accelerated PC turnover and pro-inflammatory activation of ATMs .
Obesity-related metabolic disorders are among the most prevalent causes of death worldwide . Secondary complications of obesity have been suggested to be caused by the functional failure of white adipose tissue ( WAT ) , leading to ectopic lipid deposition , lipotoxicity and systemic insulin resistance ( Virtue and Vidal-Puig , 2010 ) . Obesity is associated with a chronic low-grade inflammation , characterised by immune cell infiltration to WAT , a switch of adipose tissue macrophage ( ATM ) polarisation from a tissue-remodelling ( M2 ) to a pro-inflammatory ( M1 ) state and elevated production of pro-inflammatory , insulin-desensitising cytokines , such as tumour necrosis factor α ( TNFα ) . Over the last decade , multiple genetic and pharmacological approaches have defined a causal role of macrophage-driven WAT inflammation in the development of insulin resistance ( Hotamisligil , 2017 ) . However , specific pathophysiological mechanisms triggering pro-inflammatory activation of ATMs during obesity are poorly understood . Our previous work identified that the lipid composition of ATMs undergoes both quantitative and qualitative changes during obesity ( Prieur et al . , 2011 ) . Qualitative changes in the lipid composition of both plasma and endoplasmic reticulum ( ER ) membranes represent a major factor promoting insulin resistance ( Fu et al . , 2011; Holzer et al . , 2011; Wei et al . , 2016 ) . Obesity-associated alterations in ER lipid composition lead to a cellular process termed ER stress , which invokes an adaptive unfolded protein response ( UPR ) ( Hou et al . , 2014 ) . In macrophages , the UPR is coupled to the activation of intracellular inflammatory signalling pathways that cause WAT inflammation and insulin resistance ( Robblee et al . , 2016; Shan et al . , 2017; Suzuki et al . , 2017 ) . Furthermore , M1 macrophages are characterised by increased endogenous fatty acid synthesis , which stabilises lipid rafts within plasma membrane to allow pro-inflammatory signal transduction in obesity ( Wei et al . , 2016 ) . While phospholipids ( PLs ) are the main constituents of plasma and ER membranes , the importance of PL biosynthesis in ATMs during obesity has not yet been investigated . The concept that macrophage ER stress could be induced during obesity due to changes in membrane composition is in line with the known physiological changes in lipid metabolism that occur during obesity . Obesity is associated with increased circulating saturated fatty acids ( SFAs ) , which cause cellular ER stress by being incorporated into membrane PLs , leading to a decreased membrane fluidity due to increased membrane PL acyl chain saturation . Increased SFA-mediated ER rigidification is directly sensed by the transmembrane domains of UPR-transducing proteins ( Robblee et al . , 2016; Volmer et al . , 2013 ) . ER saturation and the resulting UPR can be counteracted by both endogenously and exogenously derived mono- and polyunsaturated fatty acids ( MUFAs and PUFAs ) , and PUFA-containing phospholipids , in particular phosphatidylcholines ( PCs ) ( Ariyama et al . , 2010; Gianfrancesco et al . , 2019; Robblee et al . , 2016; Rong et al . , 2013 ) . PC is the most abundant PL in mammalian cells . Most cells can synthesise PC de novo through the Kennedy pathway , involving the transfer of phosphocholine onto diacylglycerol moiety . PCs synthesised de novo predominantly contain saturated ( SFAs ) and monounsaturated fatty acids ( MUFAs ) , while polyunsaturated fatty acids ( PUFAs ) are incorporated into PCs via the Lands cycle , involving a hydrolysis of a single fatty acyl chain and esterification of a free PUFA to a resulting lysophosphatidylcholine ( lysoPC ) ( Shindou et al . , 2013 ) . Metabolic flux through the de novo PC synthesis pathway and cellular PC levels are greatly increased in differentiating macrophages ( Ecker et al . , 2010 ) . Furthermore , pro-inflammatory signalling via toll-like receptor 4 ( TLR4 ) increases the rate of choline uptake and de novo PC synthesis in macrophages ( Sanchez-Lopez et al . , 2019; Snider et al . , 2018; Tian et al . , 2008 ) . However , de novo PC synthesis in mature macrophages is not coupled to the expansion of the cellular PC pool , as it is counteracted by phospholipase D activity , leading to a rapid turnover of membrane PCs ( Jackowski et al . , 1997 ) . The consequences of altered PC turnover in metabolic disease are not currently known . Conceptually , the rate of de novo PC synthesis and turnover should affect PC remodelling via Lands cycle . The role of the Lands cycle in ER stress function has been studied by genetic manipulation of the enzyme lysoPC-acyltransferase 3 ( LPCAT3 ) , the major LPCAT isoform in macrophages ( Jiang et al . , 2018 ) . LPCAT3 overexpression increases the rate of PUFA incorporation into PCs and protects cells from palmitate-induced ER stress , while the loss of LPCAT3 sensitises cells to palmitate lipotoxicity ( Rong et al . , 2013 ) . We have previously shown that during obesity , ATMs acquire an M1 phenotype concomitantly with their intracellular lipid accumulation ( Prieur et al . , 2011 ) . Here , we demonstrate that markers of de novo PC synthesis are increased in ATMs isolated from obese mice and humans . Lepob/ob mice with a myeloid cell-specific reduction in de novo PC synthesis rate display reduced adipose tissue inflammation and improved metabolic profile compared to controls . Mechanistically , we show that reducing the activity of the de novo PC synthesis pathway by 30% does not reduce total cellular PC levels in macrophages . Instead , the reduction in PC synthesis is balanced by a reduction in PC degradation , maintaining the cellular PC pool size but increasing the half-life of PCs . The extended PC half-life leads to increased incorporation of PUFAs into PCs by allowing more time for PC remodelling . Elevated PC PUFA content protects macrophages from palmitate-induced ER stress and pro-inflammatory activation .
In order to identify intrinsic metabolic pathways associated with the phenotypic switch of ATMs towards an M1 polarisation state , we reanalysed our published microarray dataset ( GSE36669 ) of epididymal WAT ( eWAT ) macrophages isolated from WT and Lepob/ob animals using inferred metabolic flux analysis ( Cubuk et al . , 2018; Çubuk et al . , 2019 ) . We focused on the pathways that were unchanged or downregulated in 5-week-old Lepob/ob ATMs , which are predominantly M2-polarised , but upregulated at 16 weeks of age , when eWAT of Lepob/ob mice is inflamed ( Prieur et al . , 2011 ) . Among the metabolic pathways that fitted these criteria was de novo PC biosynthesis ( Figure 1a–b ) , with a lower inferred activity score in 5-week-old , but higher score in 16-week-old Lepob/ob ATMs compared to age-matched WT controls ( Figure 1d ) . Further analysis of the processes that were unchanged at 5 weeks but upregulated at 16 weeks in Lepob/ob ATMs revealed several pathways related to PL metabolism ( Figure 1d , Supplementary file 1 ) . The activity of de novo PE biosynthesis pathway was not modulated in Lepob/ob ATMs ( Figure 1—figure supplement 1a–b ) . In order to determine whether increased inferred activity of the de novo PC synthesis pathway in obesity was specific to ATMs , or also occurred in other tissue-resident macrophages , we performed global transcriptomic comparison between liver macrophages isolated from 14-week-old Lepob/ob and control mice . Unlike ATMs , liver macrophages isolated from obese mice showed similar expression levels of de novo PC biosynthesis pathway constituents compared to controls ( Figure 1—figure supplement 2 ) . In accordance with our ATM transcriptomic analysis , the expression of Pcyt1a , encoding phosphocholine cytidylyltransferase A ( CCTα ) , the rate-limiting enzyme in de novo PC synthesis pathway , was unchanged at 5 weeks , but increased at 16 weeks in Lepob/ob ATMs compared to WT controls when measured by qPCR ( Figure 1e ) . In contrast , the Pcyt1a paralogue Pcyt1b was down-regulated at 5 weeks and not modulated at 16 weeks in Lepob/ob ATMs ( Figure 1—figure supplement 1c ) . Furthermore , Pcyt1a expression in macrophages isolated from the WAT of obese individuals was positively correlated with BMI ( Figure 1f ) . Of note , out of all analysed tissue macrophage populations publicly available in Immgen database ( Heng et al . , 2008 ) , ATMs had the highest expression of Pcyt1a transcript ( Figure 1—figure supplement 3 ) . Next , we reanalysed our previously published lipid profiles from Lepob/ob ATMs ( Prieur et al . , 2011 ) , focusing only on measured PL species . Relative to the total PL amount , both PC abundance and PC:PE molar ratio tended to increase ( Figure 1—figure supplement 4a–b ) , and palmitate- and stearate-containing lysoPC species were upregulated in 16-week-old Lepob/ob ATMs compared to WT controls ( Figure 1g ) . It has been previously shown that hepatic lysoPC levels are reduced when the balance between PC synthesis and LPCAT activity is perturbed . Specifically , increasing LPCAT3 activity without changing de novo PC synthesis or breakdown reduces LysoPC levels , as LPCAT3 re-esterifies LysoPC into PC ( Rong et al . , 2013 ) . Conversely , in obese WAT , despite the upregulation of Lpcat3 transcript at both 5 and 16 weeks in Lepob/ob ATMs ( Figure 1—figure supplement 5a ) , we observed increased lysoPC species in Lepob/ob ATMs ( Figure 1g ) . The elevation in LysoPCs was therefore consistent with obesity causing a disproportional increase in the rate of both de novo PC synthesis and hydrolysis that exceeded the capacity of LPCAT3 to re-esterify Lyso-PC back into PC ( Figure 1—figure supplement 5b ) . To test if increased PC biosynthesis in ATMs affected whole-organism metabolic homeostasis , we investigated mice with Pcyt1a deletion in myeloid cells ( CCTα mKO ) that have been described previously ( Tian et al . , 2008 ) . Initially , we sought to validate whether the loss of Pcyt1a would impact macrophage differentiation or function in vitro and in vivo . As indicated by the normal surface expression of macrophage markers F4/80 , CD206 and CD301 , unaltered bacterial phagocytosis and normal TNFα and IL-6 cytokine secretion in response to LPS , the differentiation of CCTα-null bone marrow cells into macrophages ( BMDMs ) was not impaired ( Figure 2—figure supplement 1a–c ) . Pcyt1a transcript levels in BMDMs on a C57Bl/6J genetic background were reduced by ~50% , which translated into ~80% decrease in CCTα protein expression and ~30% decrease in de novo PC synthesis rate compared to controls ( Figure 2—figure supplement 1d ) . In vivo , the expression of macrophage mRNA markers in eWAT and liver was comparable between CCTα mKO and control animals ( Figure 2—figure supplement 2a–b ) . Overall , these results confirmed that loss of CCTα reduced de novo PC biosynthesis rate in macrophages without altering their development or function . CCTα mKO mice exhibited similar growth rates and metabolic tissue weights compared to controls ( Figure 2—figure supplement 3a–b ) . No differences in glucose or insulin tolerance tests were observed between CCTα mKO and control groups ( Figure 2—figure supplement 3c–d ) . In accordance , the expression levels of insulin-regulated metabolic genes were similar in eWAT and liver of CCTα mKO and control mice ( Figure 2—figure supplement 2a–b ) . We next evaluated the importance of increased macrophage de novo PC synthesis in obesity . We first confirmed that bone marrow transplantation did not alter the induction of Pcyt1a in the Lepob/ob genetic background ( Figure 2—figure supplement 4 ) . Next , we transplanted CCTα mKO or control bone marrow into irradiated Lepob/ob animals ( Figure 2a ) . While no differences in post-irradiation body weight gain , WAT and liver mass were observed ( Figure 2b–c ) , Lepob/ob mice carrying CCTα mKO bone marrow tended to have improved glucose tolerance and exhibited increased sensitivity to exogenous insulin compared to controls ( Figure 2d–e ) . Overall , macrophage-specific Pcyt1a deletion did not affect ATM development , adipose tissue function and glucose metabolism in lean animals , but improved systemic glucose handling in Lepob/ob mice , the model of obesity in which we originally observed an induction of Pcyt1a in the ATM population . While the metabolic effects of macrophage-specific Pcyt1a deletion on a Lepob/ob background were modest , they were consistent with the relatively small reduction ( 30% ) in de novo PC biosynthesis rate we observed in BMDMs in vitro . We next sought to determine how Pcyt1a deficiency in macrophages improved glucose metabolism in obese mice . First , we performed whole transcriptome comparison of eWAT isolated from CCTα mKO Lepob/ob BMT and control animals . Pathway analysis of the transcriptomic data revealed an increase in transcripts associated with insulin sensitivity and glucose metabolism , while pathways related to ER stress and macrophage-driven inflammation were supressed in CCTα mKO compared to control BMT Lepob/ob mice ( Figure 3a , Supplementary files 2a-b ) . RNA sequencing results were also confirmed by qPCR ( Figure 2—figure supplement 1a ) . In accordance to gene expression data , insulin-responsive AKT phosphorylation was increased in the eWAT of CCTα mKO Lepob/ob animals compared to controls ( Figure 3b–c ) . Furthermore , while we found no differences in total ATM number , eWAT macrophages showed a shift from M1 to M2 polarisation in CCTα mKO compared to controls ( Figure 3d ) . No differences in the number of crown-like structures ( CLS ) and eWAT adipocyte area were observed between genotypes ( Figure 3—figure supplement 2a–c ) . Finally , unlike eWAT , the expression of pro-inflammatory and insulin-responsive marker genes in the liver were similar between genotypes ( Figure 2—figure supplement 1b ) . Similarly , insulin-responsive AKT phosphorylation was comparable between the livers of both genotypes ( Figure 3—figure supplement 3a–b ) and only tended to be increased in the skeletal muscles of CCTα mKO compared to control BMT Lepob/ob mice ( Figure 3—figure supplement 3c–d ) , indicating that macrophage-specific Pcyt1a deletion had a stronger impact on white adipose tissue insulin signalling , compared to liver and muscle . Altogether , we have found that reducing de novo PC biosynthesis rate in macrophages alleviates WAT inflammation and insulin resistance in obese mice , without affecting total ATM and CLS number . Next , we sought to investigate the molecular events that reduce WAT inflammation in obese animals carrying CCTα mKO bone marrow . For this purpose , we utilised an in vitro model of BMDMs exposed to high palmitate concentrations . We selected palmitate concentrations that have previously been reported to induce ER stress and pro-inflammatory activation , thus mimicking the effects of obesity on ATMs ( Robblee et al . , 2016 ) . We observed diminished Tnf transcript levels in palmitate-treated CCTα-null macrophages compared to controls ( Figure 4a ) . Reduced inflammation in BMDMs was accompanied by a lower ER stress response to palmitate , as indicated by lower induction of ER stress marker gene expression and reduced stress-responsive kinase activation in CCTα-null BMDMs compared to controls ( Figure 4b–d ) . Furthermore , CCTα-null BMDMs were less susceptible to palmitate-induced cell death than controls ( Figure 4—figure supplement 1a ) . While Pcyt1a deficiency was protective against cytotoxicity in response to palmitate , it was detrimental in response to other ER stressors , including thapsigargin ( Figure 4—figure supplement 1b ) and free cholesterol ( Zhang et al . , 2000 ) . Finally , cultured peritoneal macrophages isolated from CCTα mKO animals also showed reduced ER stress response to palmitate compared to controls ( Figure 4—figure supplement 2 ) . Overall , CCTα-null macrophages were protected against palmitate-induced ER stress and subsequent cytotoxicity and inflammation . We next investigated how mitigating CCTα activity caused a reduction in palmitate-induced ER stress . As the de novo PC biosynthesis pathway had been suggested to control the flux of exogenous palmitate into cellular PCs ( Robblee et al . , 2016 ) , we hypothesised that CCTα-null BMDMs would have a reduced rate of palmitate incorporation into their membranes . In order to test our hypothesis , we traced the incorporation of exogenous palmitate into cellular PCs over time . Surprisingly , CCTα-null and control BMDMs showed no differences in the rate of radiolabelled palmitate appearance in total lipid or PC fractions ( Figure 5a–b ) . We then attempted to validate our unexpected findings using acute pharmacological inhibition of CCTα by miltefosine . Miltefosine reduced the rate of de novo PC synthesis in palmitate-treated BMDMs in a dose-response manner ( Figure 5c ) . In contrast , only 100 μM concentration of miltefosine showed an inhibitory effect on the incorporation of palmitate into membrane PCs ( Figure 5c ) . Importantly , and in line with the evidence from our genetic model , the dose of miltefosine that reduced de novo PC biosynthesis rate by 30% ( as we have observed in CCTα-null BMDMs , Figure 2—figure supplement 1d ) had no effect on the rate of incorporation of exogenous palmitate into cellular PCs in BMDMs ( Figure 5c ) . It has been proposed that the Kennedy pathway is coupled to endogenous cellular fatty acid synthesis ( Ecker et al . , 2010; Ridgway and Lagace , 2003 ) . In accordance , acetate incorporation into cellular PCs and total lipids showed a similar fold decrease ( approximately 30% ) as the reduction in de novo PC synthesis rate in CCTα-null macrophages ( Figure 5d–e ) . Overall , we found that reducing the rate of de novo PC synthesis in macrophages proportionally decreased the rate of incorporation of lipids derived from de novo lipogenesis , which are known to be incorporated by the Kennedy pathway , but did not affect the rate of exogenous palmitate incorporation into membrane lipids . As our tracer experiments could not explain the diminished ER stress response observed in CCTα-null BMDMs in response to palmitate , we performed global lipidomic analysis of CCTα-null and control BMDMs . As described previously ( Tian et al . , 2008 ) , total quantities of PC and PE in macrophages were unaffected by Pcyt1a deletion ( Figure 6—figure supplement 1a ) . Unexpectedly , CCTα-null macrophages showed an enrichment in PUFA-containing PC levels compared to controls ( Figure 6a , Figure 6—source data 1 ) . Consistent with such observation , the expression of sterol regulatory element-binding protein 1 ( SREBP1 ) target genes , which are known to be downregulated by high levels of PUFA-containing PLs in the ER ( Hagen et al . , 2010 ) , was lower in CCTα-null macrophages than controls ( Figure 6—figure supplement 2 ) . Furthermore , changes in PE composition were similar to qualitative PC changes in CCTα-null and control cells ( Figure 6—figure supplement 1b ) , indicating that reducing de novo PC biosynthesis rate promotes PUFA accumulation in membrane PLs . We observed similar levels of saturated PC species between CCTα-null and control BMDMs under basal conditions ( Figure 6b ) . Interestingly , the increased abundance of PUFA-containing PCs in CCTα-null BMDMs was at the expense of decreased mono- and diunsaturated PC species ( Figure 6a–b ) . We confirmed these findings by analysing total BMDM fatty acid composition , which showed increased relative abundance of arachidonic ( 20:4n6 ) , docosapentaenoic ( 22:5n6 ) and docosahexanoic acids ( 22:6n3 ) , while palmitoleic and oleic acid levels were reduced in CCTα-null cells compared to controls ( Figure 6c ) . As expected , palmitate treatment caused a large increase in the abundance of saturated PC species in BMDMs ( Figure 6—figure supplement 3b ) . Interestingly , in palmitate-treated macrophages , PC saturation was increased mostly at the expense of reduced mono- and diunsaturated PCs , while the proportion of 3 or more double bond-containing PC species was largely unaffected by palmitate treatment ( Figure 6b and Figure 6—figure supplement 3b ) . Consequently , CCTα-null BMDMs showed higher membrane PUFA levels than controls even after prolonged treatment with palmitate , leading to diminished palmitate-induced PC saturation ( Figure 6b and Figure 6—figure supplement 4 ) . Prolonged palmitate treatment elevated PC:PE ratio to a similar extent in both CCTα-null and control BMDMs , suggesting that changes in total PC levels or PC:PE ratio were unlikely to explain differences in ER stress response between genotypes ( Figure 6—figure supplement 3a ) . Our lipidomics analysis demonstrated that reduced CCTα activity could affect the levels of fatty acids other than palmitate and specifically resulted in the preferential accumulation of long chain PUFAs . These results were consistent with increased remodelling of PCs , most likely by LPCAT3 . Furthermore , our lipid analysis also suggested an explanation for the lower ER stress in response to palmitate . Considerable literature has demonstrated that increased PUFA content in cellular membranes is protective against palmitate induced ER stress ( Rong et al . , 2013; Yang et al . , 2011 ) , suggesting a mechanistic explanation for the protective effects of Pcyt1a deletion against palmitate toxicity . In order to experimentally demonstrate that CCTα-null macrophages exhibited lower ER stress in response to palmitate due to qualitative changes in membrane PL composition , we performed a rescue experiment of palmitate-treated BMDMs using exogenous arachidonic acid . Compared to palmitate treatment alone , 10:1 molar mixture of palmitate and arachidonate reduced Tnf and ER stress marker gene expression to the same basal level in both CCTα-null and control BMDMs ( Figure 6d ) , suggesting that elevated PUFA levels negate the inflammatory effects of palmitate in macrophages lacking Pcyt1a . Overall , our results were consistent with a loss of CCTα resulting in a shift to a PUFA-rich membrane fatty acid composition that was protective against exogenous palmitate . Finally , we set out to explain how reduced CCTα activity could lead to an alteration in the fatty acid composition of PC . Our hypothesis was that the increased half-life of membrane PCs in CCTα-null macrophages might allow more time for PCs to be remodelled to contain PUFAs via the Lands cycle . CCTα-null macrophages have previously been shown to have reduced PC turnover rates without changes in total PC levels ( Tian et al . , 2008 ) and our experimental findings confirmed these results , as we have observed reduced de novo PC synthesis rate and unchanged PC abundance in CCTα-null BMDMs compared to controls ( Figure 2—figure supplement 1d and Figure 6—figure supplement 1a ) . We next sought to confirm that the increased levels of PUFAs in CCTα-null macrophages were due to a lower turnover of PC . To do so , we performed a pulse-chase experiment using 3H-arachidonic acid . Indeed , CCTα-null BMDMs had increased retention of arachidonic acid in their membranes compared to control cells ( Figure 7a ) . Overall , our results showed that the rate of PC turnover in macrophages is negatively associated with PUFA retention in PLs . Importantly , while the data from Pcyt1a-deficient models demonstrated lower PC turnover , increased levels of long-chain PUFAs and lower PUFA turnover rates , this data came from congenic models lacking Pcyt1a . To exclude this being a phenomenon unique to genetically manipulating Pcyt1a , we sought to manipulate PC turnover rates via an alternative route . To do so , we pharmacologically blocked PC hydrolysis by inhibiting phospholipase D activity using 1-butanol . In accordance with our hypothesis , reducing PC turnover by inhibiting PC conversion to phosphatidic acid phenocopied the effects of reduced CCTα activity on cellular membrane fatty acid composition ( Figure 7b ) .
Here , we demonstrate for the first time that obesity is characterised by an increase in de novo PC synthesis pathway in ATMs . We show that the increase in ATM de novo PC biosynthesis during obesity is pathophysiological using a macrophage-specific genetic model of reduced CCTα activity . Reducing de novo PC synthesis rate in ATMs alleviates obesity-induced WAT inflammation and improves systemic glucose metabolism . Mechanistically , we show that decreasing CCTα activity in macrophages does not reduce PC levels , but instead leads to a compensatory reduction in PC degradation and maintenance of normal PC levels . Because of this reduced PC turnover , more time is afforded for PC remodelling enzymes to act on PCs , leading to an increase in PUFA-containing PC species that are protective against ER stress in response to palmitate . Our results reveal a novel relationship between the regulation of de novo PC synthesis , PC turnover and membrane saturation . Our study extends on existing findings regarding the importance of CCTα in mediating SFA-induced lipotoxicity in cultured macrophages . Two recent independent reports have suggested that de novo PC synthesis is responsible for exogenous palmitate incorporation into membrane PCs ( Gianfrancesco et al . , 2019; Robblee et al . , 2016 ) . Robblee et al . ( 2016 ) based their conclusions on data obtained from pharmacologically inhibiting CCTα using miltefosine at a dose of 100 μM , which inhibits de novo PC synthesis rate to a level that also leads to a reduction in exogenous palmitate incorporation , a result we reproduce here ( Figure 5C ) . However , lower doses of miltefosine reduce de novo PC synthesis rate without affecting palmitate incorporation into cell membranes . At 25 μM mitefosine , we detect a 30% reduction in de novo PC synthesis rate with no reduction in palmitate incorporation , which phenocopies our genetic model of Pcyt1a deficiency in BMDMs . Importantly , our data show that a 30% reduction in de novo PC synthesis rate is sufficient to ameliorate ER stress , highlighting the capacity for changes in PC biosynthetic rate to regulate ER stress in a manner that does not require changes in palmitate incorporation . In this publication , we describe such a mechanism; that of reduced CCTα activity leading to increased PC half-life and thus permitting the establishment of a membrane composition that is protective against palmitate-induced ER stress . Similarly , Gianfrancesco et al . ( 2019 ) have utilised siRNA-mediated PCYT1A knockdown in cultured human macrophages to achieve approximately 50% reduction in de novo PC synthesis rate and demonstrated that it reduces SFA-induced inflammation . Due to a lack of tracer experiment in their study , we believe that reduced IL-1b secretion in their PCYT1A knockdown model is due to increased membrane PUFA content , as our experimental findings demonstrate that at 50 μM mitefosine , we detect a 50% reduction in de novo PC synthesis rate with no significant reduction in palmitate incorporation . Further support for a role of PC turnover in regulating ER stress comes from recent work investigating TLR4 signalling . It has recently been shown that palmitate is not a direct TLR4 agonist , but instead requires TLR4 activation-induced changes in intracellular metabolism in order to promote ER stress and inflammation in macrophages ( Lancaster et al . , 2018 ) . Importantly , it has previously been demonstrated that TLR4 activation increases the rate of de novo PC synthesis and PC turnover in macrophages in order to provide a supply of new membranes for secretory vesicle formation in Golgi apparatus ( Sanchez-Lopez et al . , 2019; Snider et al . , 2018; Tian et al . , 2008 ) . As such , our findings are in line with a mechanism in which basal TLR4 activation increases sensitivity of cells to palmitate by increasing PC turnover . In support of this concept , we demonstrate that decreasing de novo PC synthesis protects macrophages from palmitate-driven ER stress and inflammation . Furthermore , two recent reports have demonstrated that TLR4 activation in macrophages increases the transcription of Slc44a1 , encoding choline transporter CTL1 ( Sanchez-Lopez et al . , 2019; Snider et al . , 2018 ) . As we observed increased Slc44a1 transcript levels in Lepob/ob ATMs compared to controls , in future it will be of interest to test whether the increase in Pcyt1a transcript in ATMs isolated from obese mouse and human WAT is dependent on TLR4 activation . While ATMs undergo pro-inflammatory activation during obesity , liver macrophages do not ( Morgantini et al . , 2019 ) . It is likely that for this reason we observed an increased de novo PC synthesis pathway activity in ATMs , but not in liver macrophages isolated from Lepob/ob mice . While speculative , the absence of pro-inflammatory activation and normal de novo PC synthesis rate in liver macrophages could explain why hepatic genes related to metabolism and inflammation were comparable between CCTα-null Lepob/ob BMT and control mice . The observed effect size of macrophage-specific Pcyt1a deletion on systemic insulin sensitivity on the Lepob/ob genetic background is relatively small . This could potentially be explained by two factors: 1 ) relatively small decrease ( 30% ) in de novo PC synthesis rate in Lyz2Cre/+ Pcyt1afl/fl macrophages , due to the poor penetrance of Lyz2Cre/+ on Pcyt1a allele , as described previously ( Zhang et al . , 2000 ) ; 2 ) Macrophage-specific Pcyt1a deletion having a stronger impact on white adipose tissue insulin sensitivity , compared to muscle and liver . In either case , our data demonstrates a link between the increase in de novo PC biosynthesis rate in ATMs and the development of adipose tissue inflammation and insulin resistance . Finally , inactivating mutations in PCYT1A gene have been linked to several human pathologies , including retinal dystrophy , spondylometaphyseal dysplasia and lipodystrophy ( Hoover-Fong et al . , 2014; Payne et al . , 2014; Yamamoto et al . , 2014 ) . Our results suggest that besides controlling the production of bulk cellular PC mass , CCTα activity can affect the fatty acid composition of cell membranes by regulating their turnover , thus potentially explaining why homozygous PCYT1A mutations manifest in specific tissue disorders , and not in a systemic failure of proliferating cells . Such specificity is also reflected in our CCTα-null Lepob/ob BMT model , where Pcyt1a deletion altered the inflammatory profile of ATMs , but not liver macrophages . In future , it would be interesting to study the relative impact of Pcyt1a deletion on different cell types , as well as same cell types present in different tissue microenvironments .
This research has been regulated under the Animals ( Scientific Procedures ) Act 1986 Amendment Regulations 2012 following ethical review by the University of Cambridge Animal Welfare and Ethical Review Body ( AWERB ) . Mice were housed 3–4 per cage in a temperature-controlled room ( 21°C ) with a 12 hr light/dark cycle , with ‘lights on’ corresponding to six am . Animals had ad-libitum access to food and water . A standard chow diet ( DS-105 , Safe Diets ) was administered to all animals from weaning , consisting of 64 . 3% carbohydrate , 22 . 4% protein and 13 . 3% lipid of total calories . Only male mice were used for in vivo experiments . Male and female mice ( 8–20 weeks of age ) were used for in vitro BMDM cultures . Macrophage-specific Pcyt1a knockout mouse ( CCTα mKO ) was generated by crossing a mouse model containing loxP sequences surrounding Pcyt1a alleles ( Pcyt1afl/fl ) to the Lyz2Cre/+ ( Clausen et al . , 1999 ) mouse . Pcyt1afl/fl mouse was generated by Prof . Ira Tabas and Dr . Susan Jackowski as described ( Zhang et al . , 2000 ) , and was gifted to us on a mixed C57Bl/6J , 129/Sv genetic background by Dr . Suzanne Jackowski . Pcyt1afl/fl and Lyz2Cre/+ lines were backcrossed to a C57Bl/6J genetic background using Marker-Assisted Accelerated Backcrossing ( MAX-BAX , Charles River , UK ) technology until SNP genotyping confirmed >99% background purity . All experimental macrophage-specific knockout mice were produced by crossing Lyz2+/+ with Lyz2Cre/+ animals on a floxed/floxed background , yielding a 1:1 Mendelian ratio of control ( floxed/floxed Lyz2+/+ ) to knockout ( floxed/floxed Lyz2Cre/+ ) offspring . 4–6 week-old WT or Lepob/ob host mice for bone marrow transplant were purchased from Jackson laboratories and were allowed to acclimatise for at least 2 weeks before the experiment . At 8 weeks of cage , mice were split into two groups of equal average body weight and equal average fed blood glucose concentration ( Respective BW and glucose values ± SEM for control and CCTα mKO BMT: 47 . 1 ± 0 . 78 g and 46 . 8+1 . 68 g , p=0 . 87; 27 . 9 ± 3 . 05 mM and 24 , 4+3 . 46 mM , p=0 . 45 ) . All mice were given 1% Baytril antibiotic in drinking water a day before irradiation . All mice received two doses of 5 . 5 Gy of radiation using Caesium 60 source . Two hours post irradiation , donor bone marrow cells ( 10 million/mouse ) were injected into the tail veins of the irradiated mice . The cells from one donor mouse were used for up to two host mice . Host mice were then housed at 3–4/cage , with 1–2 mice carrying Pcyt1afl/fl bone marrow and 1–2 - Pcyt1afl/fl Lyz2Cre/+ bone marrow in each cage . Mice were kept on 1% Baytril for 1 month , monitored and weighed regularly until 12 weeks of age . A single Lepob/ob mouse carrying Pcyt1afl/fl bone marrow had to be culled due to health reasons . Mice were then housed under standard housing conditions throughout the duration of the study . Protein lysates were diluted in NuPAGE LDS sample buffer ( NP0007 , Thermofisher Scientific ) containing 2 . 5% 2-mercaptoethanol and boiled at 95°C for 5 min . 10 μg of protein was then separated by electrophoresis using NuPAGE SDS-polyacrylamide gels ( Thermofisher Scientific ) and transferred to nitrocellulose membranes using the iBlot Dry Blotting System ( Thermofisher Scientific ) . Membranes were blocked for 1 hr in 5% fat-free milk ( Marvel ) or 5% BSA in Tris-buffered saline containing 0 . 05% Tween ( TBST ) at room temperature and incubated overnight at 4°C with the appropriate primary antibody . Bound primary antibodies were detected using peroxidase-coupled secondary anti-rabbit antibody ( 7074 , Cell signalling ) and enhanced chemiluminescence ( WBLUF0500 , Millipore ) . Blots were exposed digitally using the ChemiDoc MP System ( Bio-Rad ) , and bands were quantified using Image Lab software ( Bio-Rad ) . The expression of proteins was normalised to a housekeeping protein ( β-actin ) , and the phosphorylation status was determined by normalising to a respective total protein . All protein quantification data is expressed as arbitrary units . Adipose tissue samples for histology were placed in 10% formalin overnight , then transferred to 70% ethanol before embedding in paraffin . Different 4 μm sections were obtained from FFPE blocks and extra-coated with paraffin to preserve tissue integrity . After incubating overnight at 37°C , sections were dewaxed using xylene and 100% industrial methylated spirits , then washed under running water for 5 min and kept in TBST . The sections were stained as follows: 1 ) blocking endogenous peroxidases for 5 min ( DAKO Real Peroxidase Blocking solution , S2023 ) ; 2 ) wash in TBST; 3 ) blocking using serum for 20 min; 4 ) primary anti-F4/80 antibody incubation for 60 min; 5 ) wash in TBST for 5 min; 6 ) 30 min incubation with MOM ImmPress Polymer Reagent ( MP-2400 ) ; 7 ) wash in TBST; 8 ) DAB solution ( 5–10 min ) prepared according to the manufacturer’s instruction ( DAB Peroxidase substrate kit , SK-4100 ) ; 9 ) wash in TBST; 10 ) 1 min incubation with Dako REAL Haematoxylin ( S2020 ) . The sections where then washed in tap water , dehydrated in graded alcohols , cleared in xylene and mounted . All eWAT slides were scanned using a Zeiss Axio Scan Z1 and analysed using HALO software ( Indica Labs , Corrales , NM ) . The ‘tissue classifier module’ , utilising a state-of-the-art machine learning algorithm to identify tissue types based on colour , texture , and contextual features , was used to distinguish areas containing F4/80-positive cells ( marked in dark green ) . ‘Vacuole Quantification module’ was then applied to analyse the adipocytes in dark green areas and in a whole section . Intact vacuoles completely surrounded by F4/80-positive cells were considered as CLS , while whole section vacuole analysis was used to determine average adipocyte area . The analyses were performed on the whole section to avoid selection bias; tissue edges were excluded using manual annotation . Halo was ‘trained-by-example’ on randomly selected images , and then the analysis was extended on the whole batch of sections with HALO automated pipeline . Total lipids from cells were extracted using a modified Folch extraction method . Glass pipettes were used throughout the procedure in order to avoid plastic-bound lipid contamination . 1 ml of HPCL-grade chloroform: methanol 2:1 v/v mixture was added to cell samples in a glass vial . Where applicable , appropriate amounts ( calculated by approximating the average abundance of every fatty acid within the sample and adding matching amounts of standard ) of 1 , 2-diundecanoyl-sn-glycero-3-phosphocholine ( phospholipid standard , 850330C , Sigma ) were included in extraction mixture as internal standard . Samples were homogenised by vortexing for 15 s . 200 μl of HPLC-grade water was added to each sample before vortexing for 2 min and centrifuging at 4000 g for 10 min . 700 μl of the lower lipid fraction was transferred to a 7 ml glass tube . A second extraction was performed by adding 700 μl of fresh HPLC-grade chloroform followed by vortexing and centrifugation as above . 900 μl of lower lipid fraction was collected and pooled with the first 700 μl fraction ( total 1600 μl ) . Collected lipid fractions were dried under nitrogen stream . Dried lipids were stored at −20°C for subsequent processing , or resuspended in 100 µl chloroform , transferred to scintillation vials containing 5 ml of Opti-Fluor scintillation liquid ( 6013199 , Perkin Elmer ) and subjected to LSC . BMDMs were treated in 6-well plates as indicated in legend and lipids were labelled as described . Lipids from cells were extracted and solubilised in 50 µl of HPLC-grade chloroform . 20 µl of lipids were then spotted at the bottom of 20 cm x 20 cm thin layer chromatography ( TLC ) silica plates ( Z292974 , Sigma ) . TLC plates were placed into hermetic glass chambers containing 250 ml of 65:25:4 chloroform: methanol: ammonium hydroxide v/v solution for phospholipid separation . Plates were allowed to develop until the solvent front was approximately 2 cm below the top of the plate . Plates were dried under laminar flow and incubated with radiographic films ( 47410 , Fujifilm ) in the dark for 1–3 days at room temperature . Radiographic films were developed using automated film developer and scanned . ImageJ software ( NIH ) was used to calculate the density of the bands on scanned radiograms . To the previously dried lipid samples , 60 µL of the lipid internal standard was added ( methanol containing CE ( 18:0 ) d6 , Ceramide ( 16:0 ) d31 , FA ( 15:0 ) d29 , LPC ( 14:0 ) d42 , PA ( 34:1 ) d31 , PC ( 34:1 ) d31 , PE ( 34:1 ) d31 , PG ( 34:1 ) d31 , PI ( 34:1 ) d31 , PS ( 16:0 ) d62 , SM ( 16:0 ) d31 , TG ( 45:0 ) d29 , and TG ( 54:0 ) d35 , all at 10 µg/mL ) . The samples were then thoroughly vortexed , then dried under a gently stream of nitrogen . The samples were then reconstituted by adding 740 µL of 4:1 mix of isopropanol and acetonitrile , respectively , and vortexed ensuring there was no undissolved material . The samples were then analysed by LC-MS analysis . Chromatographic separation was achieved using Acquity UPLC CSH C18 ( 50 mm x 2 . 1 mm , 1 . 7 µm ) LC column with a Shimadzu UPLC system ( Shimadzu UK Limited , Wolverton , Milton Keynes ) . The column was maintained at 55°C with a flow rate of 0 . 5 mL/min . A binary mobile phase system was used with mobile phase A; 60:40 acetonitrile to water , respectively , with 10 mM ammonium formate , and mobile phase B; 90:10 isopropanol to acetonitrile , respectively , with 10 mM ammonium formate . The gradient profile was as follows; at 0 min_40% mobile phase B , at 0 . 4 min_43% mobile phase B , at 0 . 45 min_50% mobile phase B , at 2 . 4 min_54% mobile phase B , at 2 . 45 min_70% mobile phase B , at 7 min_99% mobile phase B , at 8 min_99% mobile phase B , at 8 . 3 min_40% mobile phase B , at 10 min_40% mobile phase B . Mass spectrometry detection was performed on an Exactive Orbitrap mass spectrometer ( Thermo Scientific , Hemel Hempstead , UK ) operating in positive/negative ion switching mode . Heated electrospray source was used , the sheath gas was set to 40 ( arbitrary units ) , the aux gas set to 15 ( arbitrary units ) and the capillary temperature set to 300°C . The instrument was operated in full scan mode from m/z 150–1200 Da . Data processing was completed using Thermo Xcalibur Quan browser integration software ( Thermo Scientific , Hemel Hempstead , UK ) . The identification of the lipid species was determined by an MS-signal for the corresponding theoretically calculated m/z accurate mass found at the expected retention time . The semi-quantitation of the lipids was calculated by the integration of the analyte MS-signal relative to the lipid class internal standard concentration . In order to derive FFAs and esterified fatty acids from complex lipids into FAMEs , 750 μl of HPLC-grade chloroform: methanol 1:1 v/v solution was added to previously dried lipids in 7 ml glass vials . 125 μl of 10% boron trifluoride in methanol ( 134821 , Sigma ) was then added into each vial . Vials were sealed and incubated in an oven at 80°C for 90 min in order to hydrolyse fatty acid-glycerol and fatty acid-cholesterol ester bonds and form FAMEs . Samples were allowed to cool , and 1 ml of HPLC-grade n-Hexane and 500 μl of HPLC-grade water were added . Samples were briefly vortexed and centrifuged at 2000 g using benchtop centrifuge . The upper organic layer was transferred into 2 ml gas chromatography glass vials and dried under nitrogen stream . Gas chromatography-mass spectrometry was performed with Agilent 7890B gas chromatography system linked to Agilent 5977A mass spectrometer , using AS3000 auto sampler . A TR-FAME column ( length: 30 m , inter diameter: 0 . 25 mm , film size: 0 . 25 μm , 260M142P , Thermofisher Scientific ) was used with helium as carrier gas . Inlet temperature was set at 230°C . Dried FAME samples were re-suspended in 100 µl HPLC-grade n-Hexane . 1 µl of this solution was injected for analysis . The oven programme used for separation was as follows: 100°C hold for 2 min , ramp at 25 °C/min to 150°C , ramp at 2 . 5 °C/min to 162°C and hold for 3 . 8 min , ramp at 4 . 5 °C/min to 173°C and hold for 5 min , ramp at 5 °C/min to 210°C , ramp at 40 °C/min to 230°C and hold for 0 . 5 min . Carrier gas flow was set to constant 1 . 5 ml/min . If the height of any FAME peaks exceeded 108 units , sample was re-injected with 10:1 – 100:1 split ratio . Identification of FAME peaks was based on retention time and made by comparison with those in external standards ( Food industry FAME mix , 35077 , Restek ) . Peak integration and quantification was performed using MassHunter Workstation Quantitative Analysis software ( version B . 07 . 00 , Agilent ) . Specific high-abundance ions from total ion chromatogram were chosen to calculate each fatty acid peak . The values for each fatty acid were expressed in molar percentages by dividing the area of each peak by the sum of all peak areas for a given sample . This analysis accounted for differences in total lipid content between samples . All data from experiments is represented as a mean , with error bars showing standard error of the mean and the number of replicates stated in legend . Some data is represented as a fold-change , and it is stated in legend to what value the data represented was normalised to generate the fold-change . Statistical tests used are also stated in legend . A student’s t-test was used to compare two groups; one-way analysis of variance ( ANOVA ) was used to compare more than two groups , followed by Bonferonni’s post-hoc test . Where more than one factor influenced the variable being measured , 2-way ANOVA was used to test for a significant effect of each factor as well as an interaction between factors . All statistical tests were performed and graphs were generated using GraphPad Prism six software . Graphs and figures were edited for presentation using Adobe Illustrator CC 2015 software . Metabolizer algorithm used to analyse microarray data can be accessed at http://metabolizer . babelomics . org and its methodology is presented in recent publications ( Çubuk et al . , 2019; Cubuk et al . , 2018 ) . | Although inflammation can be good for the body and help fight off infection , in certain cases it can also be harmful . When immune cells switch on at the wrong time , they can cause damage to cells and tissues . Fat tissue has its own population of immune cells called adipose tissue macrophages that remove dead fat cells and keep the tissue working . However , obesity changes the behaviour of these macrophages so they switch on as though they were fighting an infection and make the fat tissue inflamed . The signals produced by these activated macrophages stop fat tissue working , and this can lead to type 2 diabetes . The trigger that activates macrophages in obesity is not yet clear , but some evidence suggests that it is due to the type of fat available . Fats come in two main forms: saturated , which can lead to high cholesterol , or unsaturated which can reduce the risk of high blood pressure . An increase in saturated fats can cause cells , including macrophages , to become stressed . Researchers showed in 2011 that macrophages in the fatty tissue of obese mice accumulate fat and become inflamed , but it was unclear which types of fat , if any , were driving the inflammation . Now , Petkevicius et al . – including some of the researchers involved in the 2011 work – report that macrophages in the fatty tissue of obese mice make excess phosphatidylcholine , a fat normally found in the cell membrane . Phosphatidylcholine is a type of fat known as a phospholipid and it is made up of two subunits called fatty acids that can either be saturated or unsaturated . In obese people and mice , fatty tissue produces too much of the enzyme that makes phosphatidylcholine , called CCTa . Petkevicius et al . showed that partially removing the CCTa gene from macrophages reduces inflammation , but , unexpectedly , the amount of phosphatidylcholine in the cells stays the same . This is because macrophages respond to the halt in phosphatidylcholine production by removing less of the phospholipid from the membrane . This gives the macrophages time to exchange the saturated fatty acids in phosphatidylcholine for unsaturated fatty acids . Therefore , the longer phosphatidylcholine stays in the membrane , the more likely it is to contain unsaturated fatty acids . Further experiments demonstrated that this change counteracts the effects caused by excess saturated fats , protecting the cells and reducing inflammation . Although the understanding of obesity is still in its early stages , this study adds another piece of the puzzle . If we can understand why fat stops working in obesity , and how this leads to disease , it could aid the design of new treatments for type 2 diabetes . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"cell",
"biology"
] | 2019 | Accelerated phosphatidylcholine turnover in macrophages promotes adipose tissue inflammation in obesity |
The amphipod crustacean Parhyale hawaiensis is a blossoming model system for studies of developmental mechanisms and more recently regeneration . We have sequenced the genome allowing annotation of all key signaling pathways , transcription factors , and non-coding RNAs that will enhance ongoing functional studies . Parhyale is a member of the Malacostraca clade , which includes crustacean food crop species . We analysed the immunity related genes of Parhyale as an important comparative system for these species , where immunity related aquaculture problems have increased as farming has intensified . We also find that Parhyale and other species within Multicrustacea contain the enzyme sets necessary to perform lignocellulose digestion ( 'wood eating' ) , suggesting this ability may predate the diversification of this lineage . Our data provide an essential resource for further development of Parhyale as an experimental model . The first malacostracan genome will underpin ongoing comparative work in food crop species and research investigating lignocellulose as an energy source .
Very few members of the Animal Kingdom hold the esteemed position of major model system for understanding living systems . Inventions in molecular and cellular biology increasingly facilitate the emergence of new experimental systems for developmental genetic studies . The morphological and ecological diversity of the phylum Arthropoda makes them an ideal group of animals for comparative studies encompassing embryology , adaptation of adult body plans and life history evolution ( Akam , 2000; Budd and Telford , 2009; Peel et al . , 2005; Scholtz and Wolff , 2013 ) . While the most widely studied group are Hexapods , reflected by over a hundred sequencing projects available in the NCBI genome database , genomic data in the other three sub-phyla in Arthropoda are still relatively sparse . Recent molecular and morphological studies have placed crustaceans along with hexapods into a pancrustacean clade ( Figure 1A ) , revealing that crustaceans are paraphyletic ( Mallatt et al . , 2004; Cook et al . , 2005; Regier et al . , 2005; Ertas et al . , 2009; Richter , 2002 ) . Previously , the only available fully sequenced crustacean genome was that of the water flea Daphnia which is a member of the Branchiopoda ( Colbourne et al . , 2011 ) . A growing number of transcriptomes for larger phylogenetic analyses have led to differing hypotheses of the relationships of the major pancrustacean groups ( Figure 1B ) ( Meusemann et al . , 2010; Regier et al . , 2010; Oakley et al . , 2013; von Reumont et al . , 2012 ) . The genome of the amphipod crustacean Parhyale hawaiensis addresses the paucity of high quality non-hexapod genomes among the pancrustacean group , and will help to resolve relationships within this group as more genomes and complete proteomes become available ( Rivarola-Duarte et al . , 2014; Kenny et al . , 2014 ) . Crucially , genome sequence data is also necessary to further advance research in Parhyale , currently the most tractable crustacean model system . This is particularly true for the application of powerful functional genomic approaches , such as genome editing ( Cong et al . , 2013; Serano et al . , 2015; Martin et al . , 2015; Mali et al . , 2013; Jinek et al . , 2012; Gilles and Averof , 2014 ) . 10 . 7554/eLife . 20062 . 003Figure 1 . Introduction . ( A ) Phylogenetic relationship of Arthropods showing the Chelicerata as an outgroup to Mandibulata and the Pancrustacea clade which includes crustaceans and insects . Species listed for each clade have ongoing or complete genomes . Species include Crustacea: Parhyale hawaiensis , D . pulex; Hexapoda: Drosophila melanogaster , Apis mellifera , Bombyx mori , Aedis aegypti , Tribolium castaneum; Myriapoda: Strigamia maritima , Trigoniulus corallines; Chelicerata: Ixodes scapularis , Tetranychus urticae , Mesobuthus martensii , Stegodyphus mimosarum . ( B ) One of the unresolved issues concerns the placement of the Branchiopoda either together with the Cephalocarida , Remipedia and Hexapoda ( Allotriocarida hypothesis A ) or with the Copepoda , Thecostraca and Malacostraca ( Vericrustacea hypothesis B ) . ( C ) Life cycle of Parhyale that takes about two months at 26C . Parhyale is a direct developer and a sexually dimorphic species . The fertilized egg undergoes stereotyped total cleavages and each blastomere becomes committed to a particular germ layer already at the 8-cell stage depicted in ( D ) . The three macromeres Er , El , and Ep give rise to the anterior right , anterior left , and posterior ectoderm , respectively , while the fourth macromere Mav gives rise to the visceral mesoderm and anterior head somatic mesoderm . Among the 4 micromeres , the mr and ml micromeres give rise to the right and left somatic trunk mesoderm , en gives rise to the endoderm , and g gives rise to the germline . DOI: http://dx . doi . org/10 . 7554/eLife . 20062 . 003 Parhyale is a member of the diverse Malacostraca clade with thousands of extant species including economically and nutritionally important groups such as shrimps , crabs , crayfish and lobsters , as well as common garden animals like woodlice . They are found in all marine , fresh water , and higher humidity terrestrial environments . Apart from attracting research interest as an economically important food crop , this group of animals has been used to study developmental biology and the evolution of morphological diversity ( for example with respect to Hox genes ) ( Martin et al . , 2015; Averof and Patel , 1997; Liubicich et al . , 2009; Pavlopoulos et al . , 2009 ) , stem cell biology ( Konstantinides and Averof , 2014; Benton et al . , 2014 ) , innate immunity processes ( Vazquez et al . , 2009; Hauton , 2012 ) and recently the cellular mechanisms of regeneration ( Konstantinides and Averof , 2014; Benton et al . , 2014; Alwes et al . , 2016 ) . In addition , members of the Malacostraca , specifically both Amphipods and Isopods , are thought to be capable of 'wood eating' or lignocellulose digestion and to have microbiota-free digestive systems ( King et al . , 2010; Kern et al . , 2013; Boyle and Mitchell , 1978; Zimmer et al . , 2002 ) . The life history of Parhyale makes it a versatile model organism amenable to experimental manipulations ( Figure 1C ) ( Wolff and Gerberding , 2015 ) . Gravid females lay eggs every 2 weeks upon reaching sexual maturity and hundreds of eggs can be easily collected at all stages of embryogenesis . Embryogenesis takes about 10 days at 26°C and has been described in detail with an accurate staging system ( Browne et al . , 2005 ) . Early embryos display an invariant cell lineage with each blastomere at the 8-cell stage contributing to a specific germ layer ( Figure 1D ) ( Browne et al . , 2005; Gerberding et al . , 2002 ) . Embryonic and post-embryonic stages are amenable to experimental manipulations and direct observation in vivo ( Gerberding et al . , 2002; Extavour , 2005; Rehm et al . , 2009a , 2009b , 2009c , 2009d; Price et al . , 2010; Alwes et al . , 2011; Hannibal et al . , 2012; Kontarakis and Pavlopoulos , 2014; Nast and Extavour , 2014; Chaw and Patel , 2012; Pavlopoulos and Averof , 2005 ) . These can be combined with transgenic approaches ( Pavlopoulos and Averof , 2005; Kontarakis et al . , 2011; Kontarakis and Pavlopoulos , 2014; Pavlopoulos et al . , 2009 ) , RNA interference ( RNAi ) ( Liubicich et al . , 2009 ) and morpholino-mediated gene knockdown ( Ozhan-Kizil et al . , 2009 ) , and transgene-based lineage tracing ( Konstantinides and Averof , 2014 ) . Most recently the utility of the clustered regularly interspaced short palindromic repeats ( CRISPR ) /CRISPR-associated ( Cas ) system for targeted genome editing has been elegantly demonstrated during the systematic study of Parhyale Hox genes ( Martin et al . , 2015; Serano et al . , 2015 ) . This arsenal of experimental tools ( Table 1 ) has already established Parhyale as an attractive model system for biological research . 10 . 7554/eLife . 20062 . 004Table 1 . Experimental resources . Available experimental resources in Parhyale and corresponding references . DOI: http://dx . doi . org/10 . 7554/eLife . 20062 . 004Experimental ResourcesReferencesEmbryological manipulations Cell microinjection , isolation , ablation ( Gerberding et al . , 2002; Extavour , 2005; Price et al . , 2010; Alwes et al . , 2011; Hannibal et al . , 2012; Rehm et al . , 2009; Rehm et al . , 2009; Kontarakis and Pavlopoulos , 2014; Nast and Extavour , 2014 ) Gene expression studies In situ hybridization , antibody staining ( Rehm et al . , 2009; Rehm et al . , 2009 ) Gene knock-down RNA interference , morpholinos ( Liubicich et al . , 2009; Ozhan-Kizil et al . , 2009 ) Transgenesis Transposon-based , integrase-based ( Pavlopoulos and Averof , 2005; Kontarakis et al . , 2011; Kontarakis and Pavlopoulos , 2014 ) Gene trapping Exon/enhancer trapping , iTRAC ( trap conversion ) ( Kontarakis et al . , 2011 ) Gene misexpressionHeat-inducible ( Pavlopoulos et al . , 2009 ) Gene knock-outCRISPR/Cas ( Martin et al . , 2015 ) Gene knock-in CRISPR/Cas homology-dependent or homology-independent ( Serano et al . , 2015 ) Live imaging Bright-field , confocal , light-sheet microscopy ( Alwes et al . , 2011; Hannibal et al . , 2012; Chaw and Patel , 2012; Alwes et al . , 2016 ) So far , work in Parhyale has been constrained by the lack of a reference genome and other standardized genome-wide resources . To address this limitation , we have sequenced , assembled and annotated the genome . At an estimated size of 3 . 6 Gb , this genome represents one of the largest animal genomes tackled to date . The large size has not been the only challenge of the Parhyale genome , that also exhibits some of the highest levels of sequence repetitiveness and polymorphism reported among published genomes . We provide information in our assembly regarding polymorphism to facilitate functional genomic approaches sensitive to levels of sequence similarity , particularly homology-dependent genome editing approaches . We analysed a number of key features of the genome as foundations for new areas of research in Parhyale , including innate immunity in crustaceans , lignocellulose digestion , non-coding RNA biology , and epigenetic control of the genome . Our data bring Parhyale to the forefront of developing model systems for a broad swathe of important bioscience research questions .
The Parhyale genome contains 23 pairs ( 2n=46 ) of chromosomes ( Figure 2 ) and with an estimated size of 3 . 6 Gb , it is currently the second largest reported arthropod genome after the locust genome ( Parchem et al . , 2010; Wang et al . , 2014 ) . Sequencing was performed on genomic DNA isolated from a single adult male taken from a line derived from a single female and expanded after two rounds of sib-mating . We performed k-mer analyses of the trimmed reads to assess the impact of repeats and polymorphism on the assembly process . We analyzed k-mer frequencies ( Figure 3A ) and compared k-mer representation between our different sequencing libraries . We observed a 93% intersection of unique k-mers among sequencing libraries , indicating that the informational content was consistent between libraries ( Source code 1 ) . The k-mer analysis revealed a bimodal distribution of error-free k-mers ( Figure 3A ) . The higher-frequency peak corresponded to k-mers present on both haplotypes ( i . e . homozygous regions ) , while the lower-frequency peak had half the coverage and corresponded to k-mers present on one haplotype ( i . e . heterozygous regions ) ( Simpson and Durbin , 2012 ) . We concluded that the single sequenced adult Parhyale exhibits very high levels of heterozygosity , similar to the highly heterozygous oyster genome ( see below ) . 10 . 7554/eLife . 20062 . 005Figure 2 . Parhyale karyotype . ( A ) Frequency of the number of chromosomes observed in 42 mitotic spreads . Forty-six chromosomes were observed in more than half of all preparations . ( B ) Representative image of Hoechst-stained chromosomes . DOI: http://dx . doi . org/10 . 7554/eLife . 20062 . 00510 . 7554/eLife . 20062 . 006Figure 3 . Parhyale genome assembly metrics . ( A ) K-mer frequency spectra of all reads for k-lengths ranging from 20 to 50 . ( B ) K-mer branching analysis showing the frequency of k-mer branches classified as variants compared to Homo sapiens ( human ) , Crassostrea gigas ( oyster ) , and Saccharomyces cerevisiae ( yeast ) . ( C ) K-mer branching analysis showing the frequency of k-mer branches classified as repetitive compared to H . sapiens , C . gigas and S . cerevisiae . ( D ) Histogram of read coverages of assembled contigs . ( E ) The number of contigs with an identity ranging from 70–95% to another contig in the set of assembled contigs . ( F ) Collapsed contigs ( green ) are contigs with at least 95% identity with a longer primary contig ( red ) . These contigs were removed prior to scaffolding and added back as potential heterozygous contigs after scaffolding . DOI: http://dx . doi . org/10 . 7554/eLife . 20062 . 006 In order to quantify global heterozygosity and repeat content of the genome we assessed the de-Bruijn graphs generated from the trimmed reads to observe the frequency of both variant and repeat branches ( Simpson , 2014 ) ( Figure 3B and C ) . We found that the frequency of the variant branches was 10x higher than that observed in the human genome and very similar to levels in the highly polymorphic genome of the oyster Crassostrea gigas ( Zhang et al . , 2012 ) . We also observed a frequency of repeat branches approximately 4x higher than those observed in both the human and oyster genomes ( Figure 3C ) , suggesting that the big size of the Parhyale genome can be in large part attributed to the expansion of repetitive sequences . These metrics suggested that both contig assembly and scaffolding with mate-pair reads were likely to be challenging due to high heterozygosity and repeat content . After an initial contig assembly we remapped reads to assess coverage of each contig . We observed a major peak centered around 75x coverage and a smaller peak at 150x coverage . Contigs with lower 75x coverage represent regions of the genome that assembled into separate haplotypes and had half the frequency of mapped sequencing reads , reflecting high levels of heterozygosity . This resulted in independent assembly of haplotypes for much of the genome ( Figure 3D ) . One of the prime goals in sequencing the Parhyale genome was to achieve an assembly that could assist functional genetic and genomic approaches in this species . Different strategies have been employed to sequence highly heterozygous diploid genomes of non-model and wild-type samples ( Kajitani et al . , 2014 ) . We aimed for an assembly representative of different haplotypes , allowing manipulations to be targeted to different allelic variants in the assembly . This could be particularly important for homology dependent strategies that are likely to be sensitive to polymorphism . However , the presence of alternative haplotypes could lead to poor scaffolding between contigs as many mate-pair reads may not map uniquely to one contig and distinguish between haplotypes in the assembly . To alleviate this problem we used a strategy to conservatively identify pairs of allelic contigs and proceeded to use only one in the scaffolding process . First , we estimated levels of similarity ( identity and alignment length ) between all assembled contigs to identify independently assembled allelic regions ( Figure 3E ) . We then kept the longer contig of each pair for scaffolding using our mate-pair libraries ( Figure 3F ) , after which we added back the shorter allelic contigs to produce the final genome assembly ( Figure 4A ) . 10 . 7554/eLife . 20062 . 007Figure 4 . Workflows of assembly , annotation , and proteome generation . ( A ) Flowchart of the genome assembly . Two shotgun libraries and four mate-pair libraries with the indicated average sizes were prepared from a single male animal and sequenced to a predicted depth of 115x coverage after read filtering , based on a predicted size of 3 . 6 Gbp . Contigs were assembled at two different k-lengths with Abyss and the two assemblies were merged with GAM-NGS . Filtered contigs were scaffolded with SSPACE . ( B ) The final scaffolded assembly was annotated with a combination of Evidence Modeler to generate 847 high quality gene models and Augustus for the final set of 28 , 155 predictions . These protein-coding gene models were generated based on a Parhyale transcriptome consolidated from multiple developmental stages and conditions , their homology to the species indicated , and ab initio predictions with GeneMark and SNAP . ( C ) The Parhyale proteome contains 28 , 666 entries based on the consolidated transcriptome and gene predictions . The transcriptome contains 292 , 924 coding and non-coding RNAs , 96% of which could be mapped to the assembled genome . DOI: http://dx . doi . org/10 . 7554/eLife . 20062 . 00710 . 7554/eLife . 20062 . 008Figure 4—source data 1 . Catalog of repeat elements in Parhyale genome assembly . Description of repeat content in the Parhyale genome . DOI: http://dx . doi . org/10 . 7554/eLife . 20062 . 00810 . 7554/eLife . 20062 . 009Figure 4—source data 2 . Software and Data . List of programs and bioinformatic tools and publicly available sequence data used in this study . DOI: http://dx . doi . org/10 . 7554/eLife . 20062 . 00910 . 7554/eLife . 20062 . 010Figure 4—figure supplement 1 . CEGMA assessment of Parhyale transcriptome and genome . ( A ) CEGMA genes present in the transcriptome assembly scored by BLAST identity ( y axis ) and proportion of coverage ( relative length , x axis ) ( B ) CEGMA genes present in the genome assembly scored by BLAST identity ( y axis ) and proportion of coverage ( relative length , x axis ) . In this analysis coverage reduced . DOI: http://dx . doi . org/10 . 7554/eLife . 20062 . 010 RepeatModeler and RepeatMasker were used on the final assembly to find repetitive regions , which were subsequently classified into families of transposable elements or short tandem repeats ( Source code 2 ) . We found 1473 different repeat element sequences representing 57% of the assembly ( Figure 4—source data 1 ) . The Parhyale assembly comprises of 133 , 035 scaffolds ( 90% of assembly ) , 259 , 343 unplaced contigs ( 4% of assembly ) , and 584 , 392 shorter , potentially allelic contigs ( 6% of assembly ) , with a total length of 4 . 02 Gb ( Table 2 ) . The N50 length of the scaffolds is 81 , 190 bp . The final genome assembly was annotated with Augustus trained with high confidence gene models derived from assembled transcriptomes , gene homology , and ab initio predictions . This resulted in 28 , 155 final gene models ( Figure 4B; Source code 3 ) across 14 , 805 genic scaffolds and 357 unplaced contigs with an N50 of 161 , 819 , bp and an N90 of 52 , 952 bp . 10 . 7554/eLife . 20062 . 011Table 2 . Assembly statistics . Length metrics of assembled scaffolds and contigs . DOI: http://dx . doi . org/10 . 7554/eLife . 20062 . 011# sequencesN90N50N10Sum lengthMax length# Nsscaffolds133 , 03514 , 79981 , 190289 , 7053 . 63 GB1 , 285 , 3851 . 10 GBunplaced contigs259 , 3433046271779146 MB40 , 22223 , 431hetero . contigs584 , 3922654021038240 MB24 , 461627genic scaffolds15 , 16052952161 , 8194338361 . 49 GB1 , 285 , 385323 MB Parhyale has a mean coding gene size ( introns and ORFs ) of 20 kb ( median of 7 . 2 kb ) , which is longer than D . pulex ( mean: 2 kb , median: 1 . 2 kb ) , while shorter than genes in Homo sapiens ( mean: 52 . 9 kb , median: 18 . 5 kb ) . This difference in gene length was consistent across reciprocal blast pairs where ratios of gene lengths revealed Parhyale genes were longer than Caenorhabditis elegans , D . pulex , and Drosophila melanogaster and similar to H . sapiens . ( Figure 5A ) . The mean intron size in Parhyale is 5 . 4 kb , similar to intron size in H . sapiens ( 5 . 9 kb ) but dramatically longer than introns in D . pulex ( 0 . 3 kb ) , D . melanogaster ( 0 . 3 kb ) and C . elegans ( 1 kb ) ( Figure 5B ) . 10 . 7554/eLife . 20062 . 012Figure 5 . Parhyale genome comparisons . ( A ) Box plots comparing gene sizes between Parhyale and humans ( H . sapiens ) , water fleas ( D . pulex ) , flies ( D . melanogaster ) and nematodes ( C . elegans ) . Ratios were calculated by dividing the size of the top blast hit in each species with the corresponding Parhyale gene size . ( B ) Box plots showing the distribution of intron sizes in the same species used in A . ( C ) Comparison between Parhyale and representative proteomes from the indicated animal taxa . Colored bars indicate the number of blast hits recovered across various thresholds of E-values . The top hit value represents the number of proteins with a top hit corresponding to the respective species . ( D ) Cladogram showing the number of shared orthologous protein groups at various taxonomic levels , as well as the number of clade-specific groups . A total of 123 , 341 orthogroups were identified with Orthofinder across the 16 genomes used in this analysis . Within Pancrustacea , 37 orthogroups were shared between Branchiopoda and Hexapoda ( supporting the Allotriocarida hypothesis ) and 49 orthogroups were shared between Branchiopoda and Amphipoda ( supporting the Vericrustacea hypothesis ) . DOI: http://dx . doi . org/10 . 7554/eLife . 20062 . 01210 . 7554/eLife . 20062 . 013Figure 5—source data 1 . List of proteins currently unique to Parhyale . List of proteins in Parhyale without identity to other species . DOI: http://dx . doi . org/10 . 7554/eLife . 20062 . 01310 . 7554/eLife . 20062 . 014Figure 5—source data 2 . List of genes likely to be specific to the MalacostracaList of genes likely to be specific to the Malacostraca . DOI: http://dx . doi . org/10 . 7554/eLife . 20062 . 01410 . 7554/eLife . 20062 . 015Figure 5—source data 3 . Orthofinder analysis . Orthofinder analysis using the Parhyale predicted proteome . DOI: http://dx . doi . org/10 . 7554/eLife . 20062 . 01510 . 7554/eLife . 20062 . 016Figure 5—figure supplement 1 . Expanded gene families in Parhyale . Histograms showing number of paralogs in each listed species for ( A ) sidestep , ( B ) lachesin , ( C ) neurotrimin/DPR , ( D ) APN and ( E ) cathepsin genes for gene families over represented in Parhyale . DOI: http://dx . doi . org/10 . 7554/eLife . 20062 . 016 For downstream analyses of Parhyale protein coding content , a final proteome consisting of 28 , 666 proteins was generated by combining candidate coding sequences identified with TransDecoder ( Haas et al . , 2013 ) from mixed stage transcriptomes . Almost certainly the high number of predicted gene models and proteins is an overestimation due to fragmented genes , very different isoforms or unresolved alleles , that will be consolidated as annotation of the Parhyale genome improves . We also included additional high confidence gene predictions that were not found in the transcriptome ( Figure 4C ) . The canonical proteome dataset was annotated with both Pfam , KEGG , and BLAST against Uniprot . Assembly quality was further evaluated by alignment to core eukaryotic genes defined by the Core Eukaryotic Genes Mapping Approach ( CEGMA ) database ( Parra et al . , 2007 ) . We identified 244/248 CEGMA orthology groups from the assembled genome alone and 247/248 with a combination of genome and mapped transcriptome data ( Figure 4—figure supplement 1 ) . Additionally , 96% of over 280 , 000 identified transcripts , most of which are fragments that do not contain a large ORF , also mapped to the assembled genome . Together these data suggest that our assembly is close to complete with respect to protein coding genes and transcribed regions that are captured by deep RNA sequencing . To estimate the level of heterozygosity in genes we first identified transcribed regions of the genome by mapping back transcripts to the assembly . Where these regions appeared in a single contig in the assembly , heterozygosity was calculated using information from mapped reads . Where these regions appeared in more than one contig , because haplotypes had assembled independently , heterozygosity was calculated using an alignment of the genomic sequences corresponding to mapped transcripts and information from mapped reads . This allowed us to calculate heterozygosity for each gene within the sequenced individual ( Source code 4 ) . We then calculated the genomic coverage of all transcribed regions in the genome and found , as expected , they fell broadly into two categories with higher and lower read coverage ( Figure 6A; Source code 4 ) . Genes that fell within the higher read coverage group had a lower mean heterozygosity ( 1 . 09% of bases displaying polymorphism ) , which is expected as more reads were successfully mapped . Genes that fell within the lower read coverage group had a higher heterozygosity ( 2 . 68% ) , as reads mapped independently to each haplotype ( Figure 6B ) ( Simpson , 2014 ) . Thus , we conclude that heterozygosity that influences read mapping and assembly of transcribed regions , and not just non-coding parts of the assembly . 10 . 7554/eLife . 20062 . 017Figure 6 . Variation analyses of predicted genes . ( A ) A read coverage histogram of predicted genes . Reads were first mapped to the genome , then coverage was calculated for transcribed regions of each defined locus . ( B ) A coverage distribution plot showing that genes in the lower coverage region ( <105x coverage , peak at 75x ) have a higher level of heterozygosity than genes in the higher coverage region ( >105 coverage and <250 , peak at approximately 150x coverage ) . ( C ) Distribution plot indicating that mean level of population variance is similar for genes in the higher and lower coverage regions . DOI: http://dx . doi . org/10 . 7554/eLife . 20062 . 01710 . 7554/eLife . 20062 . 018Figure 6—source data 1 . Polymorphism in Parhyale devlopmental genes . Description of polymorphism in previously identfied Parhyale developmental genes . DOI: http://dx . doi . org/10 . 7554/eLife . 20062 . 01810 . 7554/eLife . 20062 . 019Figure 6—figure supplement 1 . Confirmation of polymorphisms in the wider laboratory population of Parhyale . ( A ) An example of laboratory population polymorphism in exon 1 of the gene aristalless . As well as heterozygoisty in the single Chicago-F male sequenced ( pink and purple bases ) there is additional polymorphism detectable in the transcriptome ( green bases ) ( B ) Further examples of polymorphism in the laboratory population in 5 developmental genes . DOI: http://dx . doi . org/10 . 7554/eLife . 20062 . 019 The assembled Parhyale transcriptome was derived from various laboratory populations , hence we expected to see additional polymorphism beyond that detected in the two haplotypes of the individual male we sequenced . Analysing all genes using the transcriptome we found additional variations in transcribed regions not found in the genome of the sequenced individual . In addition to polymorphisms that agreed with heterozygosity in the genome sequence we observed that the rate of additional variations is not substantially different between genes from the higher ( 0 . 88% ) versus lower coverage group genes ( 0 . 73%; Figure 6C ) . This analysis suggests that within captive laboratory populations of Parhyale there is considerable additional polymorphism distributed across genes , irrespective of whether or not they have relatively low or high heterozygosity in the individual male we sequenced . In addition the single male we have sequenced provides an accurate reflection of polymorphism of the wider laboratory population and the established Chicago-F strain does not by chance contain unusually divergent haplotypes . We also performed an assessment of polymorphism on previously cloned Parhyale developmental genes , and found some examples of startling levels of variation . ( Figure 6—figure supplement 1 and source data 1 ) . For example , we found that the cDNAs of the germ line determinants , nanos ( 78 SNPS , 34 non-synonymous substitutions and one 6 bp indel ) and vasa ( 37 SNPs , 7 non-synonymous substitutions and a one 6 bp indel ) can have more variability within laboratory Parhyale populations than might be observed for orthologs between closely related species ( Figure 6—source data 1 ) . To further evaluate the extent of polymorphism across the genome , we mapped the genomic reads to a set of previously Sanger-sequenced BAC clones of the Parhyale Hox cluster from the same Chicago-F line from which we sequenced the genome of an adult male . ( Serano et al . , 2015 ) . We detected SNPs at a rate of 1 . 3 to 2 . 5% among the BACs ( Table 3 ) and also additional sequence differences between the BACs and genomic reads , confirming that additional polymorhism exists in the Chicago-F line beyond that detected between in the haplotypes of the individual male we sequenced . 10 . 7554/eLife . 20062 . 020Table 3 . BAC variant statistics . Level of heterozygosity of each BAC sequence determined by mapping genomic reads to each BAC individually . Population variance rate represents additional alleles found ( i . e . more than 2 alleles ) from genomic reads . DOI: http://dx . doi . org/10 . 7554/eLife . 20062 . 020BAC IDLengthHeterozygosityPop . VariancePA81-D11140 , 2641 . 6540 . 568PA40-O15129 , 9572 . 4460 . 647PA76-H18141 , 8441 . 8240 . 199PA120-H17126 , 7662 . 6731 . 120PA222-D11128 , 5421 . 3441 . 404PA31-H15140 , 1432 . 7930 . 051PA284-I07141 , 3902 . 0460 . 450PA221-A05148 , 7031 . 8621 . 427PA93-L04139 , 9552 . 1770 . 742PA272-M04134 , 7441 . 9250 . 982PA179-K23137 , 2392 . 6710 . 990PA92-D22126 , 8482 . 6500 . 802PA268-E13135 , 3341 . 6781 . 322PA264-B19108 , 5711 . 5750 . 157PA24-C06141 , 4461 . 9461 . 488 Overlapping regions of the contiguous BACs gave us the opportunity to directly compare Chicago-F haplotypes and accurately observe polynucleotide polymorphisms , that are difficult to detect with short reads that do not map when polymorphisms are large , but are resolved by longer Sanger reads . ( Figure 7A ) . Since the BAC clones were generated from a pool of Chicago-F animals , we expected each sequenced BAC to be representative of one haplotype . Overlapping regions between BAC clones could potentially represent one or two haplotypes . We found that the genomic reads supported the SNPs observed between the overlapping BAC regions . We found relatively few base positions with evidence supporting the existence of a third allele . This analysis revealed many insertion/deletion ( indels ) with some cases of indels larger than 100 base pairs ( Figure 7B ) . The finding that polynucleotide polymorphisms are prevalent between the haplotypes of the Chicago-F is another reason , in addition to regions of high SNP heterozygosity in the genome sequence , for the extensive independent assembly of haplotypes . Taken togther these data mean that special attention will have to be given to those functional genomic approaches that are dependent on homology , such as CRISPR/Cas9 based knock in strategies . 10 . 7554/eLife . 20062 . 021Figure 7 . Variation observed in contiguous BAC sequences . ( A ) Schematic diagram of the contiguous BAC clones tiling across the HOX cluster and their% sequence identities . 'Overlap length' refers to the lengths ( bp ) of the overlapping regions between two BAC clones . 'BAC supported single nucleotide polymorphisms ( SNPs ) ' refer to the number of SNPs found in the overlapping regions by pairwise alignment . 'Genomic reads supported SNPs' refer to the number of SNPs identified in the overlapping regions by mapping all reads to the BAC clones and performing variant calling with GATK . 'BAC + Genomic reads supported SNPs' refer to the number of SNPs identified from the overlapping regions by pairwise alignment that are supported by reads . 'Third allele' refers to presence of an additional polymorphism not detected by genomic reads . 'Number of INDELs' refer to the number of all insertion or deletions found in the contiguous region . 'Number of INDELs >100' are insertion or deletions greater than or equal to 100 . ( B ) Position versus indel lengths across each overlapping BAC region . DOI: http://dx . doi . org/10 . 7554/eLife . 20062 . 021 Assessment of conservation of the proteome using BLAST against a selection of metazoan proteomes was congruent with broad phylogenetic expectations . These analyses included crustacean proteomes likely to be incomplete as they come from limited transcriptome datasets , but nonetheless highlighted genes likely to be specific to the Malacostraca ( Figure 5C , Figure 5—source data 2 ) . To better understand global gene content evolution we generated clusters of orthologous and paralogous gene families comparing the Parhyale proteome with other complete proteomes across the Metazoa using Orthofinder ( Emms and Kelly , 2015 ) ( Figure 5D; Figure 5—source data 3 ) . Amongst proteins conserved in protostomes and deuterostomes we saw no evidence for widespread gene duplication in the lineage leading to Parhyale . We identified orthologous and paralogous protein groups across 16 species with 2900 and 2532 orthologous groups containing proteins found only in Panarthropoda and Arthropoda respectively . We identified 855 orthologous groups that were shared exclusively by Mandibulata , 772 shared by Pancrustacea and 135 shared by Crustacea . There were 9877 Parhyale proteins that could not be assigned to an orthologous group , potentially representing rapidly evolving or lineage specific proteins ( Figure 5—source data 1 ) . Amongst these proteins we found 609 proteins ( 2 . 1% of the proteome ) that had paralogs within Parhyale , suggesting that younger and/or more divergent Parhyale genes have undergone some considerable level of gene duplication events . Our analysis of shared orthologous groups was equivocal with regard to alternative hypotheses on the relationships among pancrustacean subgroups: 44 groups of orthologous proteins are shared among the multicrustacea clade ( uniting the Malacostraca , Copepoda and Thecostraca ) , 37 groups are shared among the Allocarida ( Branchiopoda and Hexapoda ) and 49 groups are shared among the Vericrustacea ( Branchiopoda and Multicrustacea . To further analyse the evolution of the Parhyale proteome we examined protein families that appeared to be expanded ( z-score >2 ) , compared to other taxa ( Figure 5—figure supplement 1 , Source code 5 ) . We conservatively identified 29 gene families that are expanded in Parhyale . Gene family expansions include the Sidestep ( 55 genes ) and Lachesin ( 42 ) immunoglobulin superfamily proteins as well as nephrins ( 33 genes ) and neurotrimins ( 44 genes ) , which are thought to be involved in immunity , neural cell adhesion , permeability barriers and axon guidance ( Strigini et al . , 2006; Garver et al . , 2008; Siebert et al . , 2009 ) . Other Parhyale gene expansions include APN ( aminopeptidase N ) ( 38 genes ) and cathepsin-like genes ( 30 genes ) , involved in proteolytic digestion ( Deraison et al . , 2004 ) . Components of all common metazoan cell-signalling pathways are largely conserved in Parhyale . At least 13 Wnt subfamilies were present in the cnidarian-bilaterian ancestor . Wnt3 has been lost in protostomes that retain 12 Wnt genes ( Prud'homme et al . , 2002; Cho et al . , 2010-07; Janssen et al . , 2010 ) . Some sampled ecdysozoans have undergone significant Wnt gene loss , for example C . elegans has only 5 Wnt genes ( Hilliard and Bargmann , 2006 ) . At most 9 Wnt genes are present in any individual hexapod species ( Bolognesi et al . , 2008 ) , with wnt2 and wnt4 potentially lost before the hexapod radiation ( Hogvall et al . , 2014 ) . The Parhyale genome encodes 6 of the 13 Wnt subfamily genes; wnt1 , wnt4 , wnt5 , wnt10 , wnt11 and wnt16 ( Figure 8 ) . Wnt genes are known to have been ancestrally clustered ( Holstein , 2012 ) . We observed that wnt1 and wnt10 are linked in a single scaffold ( phaw_30 . 0003199 ) ; given the loss of wnt6 and wnt9 , this may be the remnant of the ancient wnt9-1-6-10 cluster conserved in some protostomes . 10 . 7554/eLife . 20062 . 022Figure 8 . Comparison of Wnt family members across Metazoa . Comparison of Wnt genes across Metazoa . Tree on the left illustrates the phylogenetic relationships of species used . Dotted lines in the phylogenetic tree illustrate the alternative hypothesis of Branchiopoda + Hexapoda versus Branchiopoda + Multicrustacea . Colour boxes indicate the presence of certain Wnt subfamily members ( wnt1 to wnt11 , wnt16 and wntA ) in each species . Empty boxes indicate the loss of particular Wnt genes . Two overlapping colour boxes represent duplicated Wnt genes . DOI: http://dx . doi . org/10 . 7554/eLife . 20062 . 02210 . 7554/eLife . 20062 . 023Figure 8—source data 1 . List of Parhyale transcription factors by family . List of Parhyale transcript IDs for all transcription factors in the proteome , grouped by transcription factor family . DOI: http://dx . doi . org/10 . 7554/eLife . 20062 . 02310 . 7554/eLife . 20062 . 024Figure 8—source data 2 . Wnt , TGFβ and FGF signaling pathways . Parhyale transcript IDs for Wnt , Wnt ligand , FGF , FGFR and TGFβ pathway genes . DOI: http://dx . doi . org/10 . 7554/eLife . 20062 . 02410 . 7554/eLife . 20062 . 025Figure 8—source data 3 . Homeobox transcription factors . Annotation of homeobox transcription factor genes in Parhyale . DOI: http://dx . doi . org/10 . 7554/eLife . 20062 . 02510 . 7554/eLife . 20062 . 026Figure 8—figure supplement 1 . Phylogenetic tree of FGF and FGR molecules ( A ) Phylogenetic tree of arthropod and vertebrate FGFs , including two FGFs from Parhyale ( B ) Phylogenetic tree of arthropod and vertebrate FGFRs , including a single FGFR in Parhyale . DOI: http://dx . doi . org/10 . 7554/eLife . 20062 . 02610 . 7554/eLife . 20062 . 027Figure 8—figure supplement 2 . Phylogenetic tree of CERS homeobox family genes . A phylogenetic tree highlighting an expansion of CERS homeobox family genes in Parhyale . DOI: http://dx . doi . org/10 . 7554/eLife . 20062 . 027 We could identify 2 Fibroblast Growth Factor ( FGF ) genes and only a single FGF receptor ( FGFR ) in the Parhyale genome , suggesting one FGFR has been lost in the malacostracan lineage ( Figure 8——figure supplement 1 ) . Within the Transforming Growth Factor beta ( TGF-β ) signaling pathway we found 2 genes from the activin subfamily ( an activin receptor and a myostatin ) , 7 genes from the Bone Morphogen Protein ( BMP ) subfamily and 2 genes from the inhibin subfamily . Of the BMP genes , Parhyale has a single decapentaplegic homologue ( Figure 8—source data 2 ) . Other components of the TGF-β pathway were identified such as the neuroblastoma suppressor of tumorigenicity ( NBL1/DAN ) , present in Aedes aegypti and Tribolium castaneum but absent in D . melanogaster and D . pulex , and TGFB-induced factor homeobox 1 ( TGIF1 ) which is a Smad2-binding protein within the pathway present in arthropods but absent in nematodes ( C . elegans and Brugia malayi;Figure 8—source data 2 ) . We identified homologues of PITX2 , a downstream target of the TGF-β pathway involved in endoderm and mesoderm formation present in vertebrates and crustaceans ( Parhyale and D . pulex ) but not in insects and nematodes ( Ryan et al . , 1998 ) . With the exception of SMAD7 and SMAD8/9 , all other SMADs ( SMAD1 , SMAD2/3 , SMAD4 , SMAD6 ) are found in arthropods sampled , including Parhyale . Components of other pathways interacting with TGF-β signaling like the JNK , Par6 , ROCK1/RhoA , p38 and Akt pathways were also recovered and annotated in the Parhyale genome ( Figure 8—source data 2 ) . We identified major Notch signaling components including Notch , Delta , Deltex , Fringe and modulators of the Notch pathway such as Dvl and Numb . Members of the gamma-secretase complex ( Nicastrin , Presenillin , and APH1 ) were also present as well as to other co-repressors of the Notch pathway such as Groucho and CtBP ( Nagel et al . , 2005 ) . A genome wide survey to annotate all potential transcription factors ( TFs ) discovered a total of 1143 proteins with DNA binding domains that belonged to all the major families previously identified . Importantly , we observed a large expansion of TFs containing the zinc-finger ( ZF ) -C2H2 domain , that was previously observed in a trancriptomic study of Parhyale ( Zeng et al . , 2011 ) . Parhyale has 699 ZF-C2H2-containing genes ( Chung et al . , 2002–12] , which is comparable to the number found in H . sapiens ( Najafabadi et al . , 2015 ) , but significantly expanded compared to other arthropod species like D . melanogaster encoding 326 members ( Figure 8—source data 1 ) . The Parhyale genome contains 126 homeobox-containing genes ( Figure 9; Figure 8—source data 3 ) , which is higher than the numbers reported for other arthropods ( 104 genes in D . melanogaster , 93 genes in the honey bee Apis melllifera , and 113 in the centipede Strigamia maritima ) ( Chipman et al . , 2014 ) . We identified a Parhyale specific expansion in the Ceramide Synthase ( CERS ) homeobox proteins , which include members with divergent homeodomains ( Pewzner-Jung et al . , 2006 ) . H . sapiens have six CERS genes , but only five with homeodomains ( Holland et al . , 2007 ) . We observed an expansion to 12 CERS genes in Parhyale , compared to 1–4 genes found in other arthropods ( Zhong and Holland , 2011 ) ( Figure 8—figure supplement 2 ) . In phylogenetic analyses all 12 CERS genes in Parhyale clustered together with a CERS from another amphipod Echinogammarus veneris , suggesting that this is recent expansion in the amphipod lineage . 10 . 7554/eLife . 20062 . 028Figure 9 . Homeodomain protein family tree . The overview of homeodomain radiation and phylogenetic relationships among homeodomain proteins from Arthropoda ( P . hawaiensis , D . melanogaster and A . mellifera ) , Chordata ( H . sapiens and B . floridae ) , and Cnidaria ( N . vectensis ) . Six major homeodomain classes are illustrated ( SINE , TALE , POU , LIM , ANTP and PRD ) with histograms indicating the number of genes in each species belonging to a given class . DOI: http://dx . doi . org/10 . 7554/eLife . 20062 . 028 Parhyale contains a complement of 9 canonical Hox genes that exhibit both spatial and temporal colinearity in their expression along the anterior-posterior body axis ( Serano et al . , 2015 ) . Chromosome walking experiments had shown that the Hox genes labial ( lab ) and proboscipedia ( pb ) are linked and that Deformed ( Dfd ) , Sex combs reduced ( Scr ) , Antennapedia ( Antp ) and Ultrabithorax ( Ubx ) are also contiguous in a cluster ( Serano et al . , 2015 ) . Previous experiments in D . melanogaster had shown that the proximity of nascent transcripts in RNA fluorescent in situ hybridizations ( FISH ) coincide with the position of the corresponding genes in the genomic DNA ( Kosman et al . , 2004; Ronshaugen and Levine , 2004 ) . Thus , we obtained additional information on Hox gene linkage by examining nascent Hox transcripts in cells where Hox genes are co-expressed . We first validated this methodology in Parhyale embryos by confirming with FISH , the known linkage of Dfd with Scr in the first maxillary segment where they are co-expressed ( Figure 10A–A“ ) . As a negative control , we detected no linkage between engrailed1 ( en1 ) and Ubx or abd-A transcripts ( Figure 10B - B“ , C - C“ ) . We then demonstrated the tightly coupled transcripts of lab with Dfd ( co-expressed in the second antennal segment , Figure 10D - D“ ) , Ubx and abd-A ( co-expressed in the posterior thoracic segments , Figure 10E - E“ ) , and abd-A with Abd-B ( co-expressed in the anterior abdominal segments , ( Figure 10F - F“ ) . Collectively , all evidence supports the linkage of all analysed Hox genes into a single cluster as shown in ( Figure 10G - G“ ) . The relative orientation and distance between certain Hox genes still needs to be worked out . So far , we have not been able to confirm that Hox3 is also part of the cluster due to the difficulty in visualizing nascent transcripts for Hox3 together with pb or Dfd . Despite these caveats , Parhyale provides an excellent arthropod model system to understand these still enigmatic phenomena of Hox gene clustering and spatio-temporal colinearity , and compare the underlying mechanisms to other well-studied vertebrate and invertebrate models ( Kmita and Duboule , 2003 ) . 10 . 7554/eLife . 20062 . 029Figure 10 . Evidence for an intact Hox cluster in Parhyale . ( A–F’’ ) Double fluorescent in situ hybridizations ( FISH ) for nascent transcripts of genes . ( A–A’’ ) Deformed ( Dfd ) and Sex combs reduced ( Scr ) , ( B-B’’ ) engrailed 1 ( en1 ) and Ultrabithorax ( Ubx ) , ( C–C’’ ) en1 and abdominal-A ( abd-A ) , ( D–D’’ ) labial ( lab ) and Dfd , ( E–E’’ ) Ubx and abd-A , and ( F–F’’ ) Abdominal-B ( Abd-B ) and abd-A . Cell nuclei are stained with DAPI ( blue ) in panels A–F and outlined with white dotted lines in panels A'–F' and A'' . Co-localization of nascent transcript dots in A , D , E and F suggest the proximity of the corresponding Hox genes in the genomic DNA . As negative controls , the en1 nascent transcripts in B and C do not co-localize with those of Hox genes Ubx or abd-A . ( G ) Schematic representation of the predicted configuration of the Hox cluster in Parhyale . Previously identified genomic linkages are indicated with solid black lines , whereas linkages established by FISH are shown with dotted gray lines . The arcs connecting the green and red dots represent the linkages identified in D , E and F , respectively . The position of the Hox3 gene is still uncertain . Scale bars are 5 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 20062 . 029 The ParaHox and NK gene clusters encode other ANTP class homeobox genes closely related to Hox genes ( Brooke et al . , 1998 ) . In Parhyale , we found 2 caudal ( Cdx ) and 1 Gsx ParaHox genes . Compared to hexapods , we identified expansions in some NK-like genes , including 5 Bar homeobox genes ( BarH1/2 ) , 2 developing brain homeobox genes ( DBX ) and 6 muscle segment homeobox genes ( MSX/Drop ) . Evidence from several bilaterian genomes suggests that NK genes are clustered together ( Pollard and Holland , 2000; Jagla et al . , 2001; Luke et al . , 2003; Castro and Holland , 2003] . In the current assembly of the Parhyale genome , we identified an NK2-3 gene and an NK3 gene on the same scaffold ( phaw_30 . 0004720 ) and the tandem duplication of an NK2 gene on another scaffold ( phaw_30 . 0004663 ) . Within the ANTP class , we also observed 1 mesenchyme homeobox ( Meox ) , 1 motor neuron homeobox ( MNX/Exex ) and 3 even-skipped homeobox ( Evx ) genes . Lignocellulosic ( plant ) biomass is the most abundant raw material on our planet and holds great promise as a source for the production of bio-fuels ( Himmel et al . , 2007 ) . Understanding how some animals and their symbionts achieve lignocellulose digestion is a promising research avenue for exploiting lignocellulose-rich material ( Wilson , 2011; Cragg et al . , 2015 ) . Amongst Metazoans , research into the ability to depolymerize plant biomass into useful catabolites is largely restricted to terrestrial species such as ruminants , termites and beetles . These animals rely on mutualistic associations with microbial endosymbionts that provide cellulolytic enzymes known as glycosyl hydrolases ( GHs ) ( Duan et al . , 2009; Warnecke et al . , 2007 ) ( Figure 11 ) . Much less studied is lignocellulose digestion in aquatic animals despite the fact that lignocellulose represents a major energy source in aquatic environments , particularly for benthic invertebrates ( Distel et al . , 2011 ) . Recently , it has been suggested that the marine wood-boring Isopod Limnoria quadripunctata and the amphipod Chelura terebrans may have sterile microbe-free digestive systems and they produce all required enzymes for lignocellulose digestion ( King et al . , 2010; Green Etxabe , 2013; Kern et al . , 2013 ) . Significantly , these species have been shown to have endogenous GH7 family enzymes with cellobiohydrolase ( beta-1 , 4-exoglucanase ) activity , previously thought to be absent from animal genomes . From an evolutionary perspective , it is likely that GH7 coding genes were acquired by these species via horizontal gene transfer from a protist symbiont . 10 . 7554/eLife . 20062 . 030Figure 11 . Lignocellulose digestion overview . ( A ) Simplified drawing of lignocellulose structure . The main component of lignocellulose is cellulose , which is a-1 , 4-linked chain of glucose monosaccharides . Cellulose and lignin are organized in structures called microfibrils , which in turn form macrofibrils . ( B ) Summary of cellulolytic enzymes and reactions involved in the breakdown of cellulose into glucose . -1 , 4-endoclucanases of the GH9 family catalyze the hydrolysis of crystalline cellulose into cellulose chains . -1 , 4-exoclucanases of the GH7 family break down cellulose chains into cellobiose ( glucose disaccharide ) that can be converted to glucose by -glucosidases . ( C ) Adult Parhyale feeding on a slice of carrot . DOI: http://dx . doi . org/10 . 7554/eLife . 20062 . 030 Parhyale is a detrivore that can be sustained on a diet of carrots ( Figure 11C ) , suggesting that they too may be able to depolymerize lignocellulose for energy ( Figure 11A and B ) . We searched for GH family genes in Parhyale using the classification system of the CAZy ( Carbohydrate-Active enZYmes ) database ( Cantarel et al . , 2009 ) and the annotation of protein domains in predicted genes with PFAM ( Finn et al . , 2006 ) . We identified 73 GH genes with complete GH catalytic domains that were classified into 17 families ( Figure 12—source data 1 ) including 3 members of the GH7 family . Phylogenetic analysis of Parhyale GH7s show high sequence similarity to the known GH7 genes in L . quadripunctata and the amphipod C . terebrans ( Kern et al . , 2013 ) ( Figure 12A; Figure 12—figure supplement 1 ) . GH7 family genes were also identified in the transcriptomes of three more species spanning the multicrustacea clade: Echinogammarus veneris ( amphipod ) , Eucyclops serrulatus ( copepod ) and Calanus finmarchicus ( copepod ) . As previously reported , we also discovered a closely related GH7 gene in the branchiopod Daphnia ( Figure 12A ) ( Cragg et al . , 2015 ) . This finding supports the grouping of Branchiopoda with Multicrustacea ( rather than with Hexapoda ) and the acquisition of a GH7 gene by a vericrustacean ancestor . Alternatively , this suggests an even earlier acquisition of a GH7 gene by a crustacean ancestor with subsequent loss of the GH7 family gene in the lineage leading to insects . 10 . 7554/eLife . 20062 . 031Figure 12 . Phylogenetic analysis of GH7 and GH9 family proteins . ( A ) Phylogenetic tree showing the relationship between GH7 family proteins of Parhyale , other crustaceans ( Malacostraca , Branchiopoda , Copepoda ) , fungi and symbiotic protists ( root ) . UniProt and GenBank accessions are listed next to the species names . ( B ) Phylogenetic tree showing the relationship between GH9 family proteins of Parhyale , crustaceans , insects , molluscs , echinoderms , amoeba , bacteria and plants ( root ) . UniProt and GenBank accessions are listed next to the species names . Both trees were constructed with RAxML using the WAG+G model from multiple alignments of protein sequences created with MUSCLE . DOI: http://dx . doi . org/10 . 7554/eLife . 20062 . 03110 . 7554/eLife . 20062 . 032Figure 12—source data 1 . Catalog of GH family genes in Parhyale . IDs of all Parhyale GH genes and analyis of GH family membership across available malacostracan data sets . DOI: http://dx . doi . org/10 . 7554/eLife . 20062 . 03210 . 7554/eLife . 20062 . 033Figure 12—figure supplement 1 . Alignment of GH7 family genes . Alignment of GH7 family genes in Parhyale with those from Chelura terebans and Limnoria quadripunctata . DOI: http://dx . doi . org/10 . 7554/eLife . 20062 . 033 GH families 5 , 9 , 10 , and 45 encode beta-1 , 4-endoglucanases which are also required for lignocellulose digestion and are commonly found across Metazoa . We found 3 GH9 family genes with complete catalytic domains in the Parhyale genome as well as in the other three multicrustacean species ( Figure 12B ) . These GH9 enzymes exhibited a high sequence similarity to their homologues in the isopod Limnoria and in a number of termites . Beta-glucosidases are the third class of enzyme required for digestion of lignocellulose . They have been classified into a number of GH families: 1 , 3 , 5 , 9 and 30 , with GH1 representing the largest group ( Cantarel et al . , 2009 ) . In Parhyale , we found 7 beta-glucosidases from the GH30 family and 3 from the GH9 family , but none from the GH1 family . Understanding lignocellulose digestion in animals using complex mutualistic interactions with microbes has proven to be a difficult task . The study of 'wood-eating' in Parhyale can offer new insights into lignocellulose digestion in the absence of gut microbes , and the unique opportunity to apply molecular genetic approaches to understand the activity of glycosyl hydrolases in the digestive system . Lignocellulose digestion may also have implications for gut immunity in some crustaceans , since these reactions have been reported to take place in a sterile gut ( Boyle and Mitchell , 1978; Zimmer et al . , 2002 ) . Immunity research in Malacostracans has attracted interest due to the rapid rise in aquaculture related problems ( Vazquez et al . , 2009; Stentiford et al . , 2012; Hauton , 2012 ) . Malacostracan food crops represent a huge global industry ( >$40 Billion at point of first sale ) , and reliance on this crop as a source of animal protein is likely to increase in line with human population growth ( Stentiford et al . , 2012 ) . Here we provide an overview of immune-related genes in Parhyale that were identified by mapping proteins to the ImmunoDB database ( Waterhouse et al . , 2007 ) . The ability of the innate immune system to identify pathogen-derived molecules is mediated by pattern recognition receptors ( PRRs ) ( Janeway and Medzhitov , 2002 ) . Several groups of invertebrate PRRs have been characterized , i . e . thioester-containing proteins ( TEP ) , Toll-like receptors ( TLR ) , peptidoglycan recognition proteins ( PGRP ) , C-type lectins , galectins , fibrinogen-related proteins ( FREP ) , gram-negative binding proteins ( GNBP ) , Down Syndrome Cell Adhesion Molecules ( Dscam ) and lipopolysaccharides and beta-1 , 3-glucan binding proteins ( LGBP ) . The functions of PGRPs have been described in detail in insects like D . melanogaster ( Werner et al . , 2003 ) and the PGRP family has also been reported in Vertebrates , Molluscs and Echinoderms ( Liu et al . , 2001; Rehman et al . , 2001 ) . Surprisingly , we found no PGRP genes in the Parhyale genome . PGRPs were also not found in other sequence datasets from Branchiopoda , Copepoda and Malacostraca ( Figure 13A ) , raising the possibility of their close phylogenetic relationship ( like the GH7 genes ) . In the absence of PGRPs , the freshwater crayfish Pacifastacus leniusculus relies on a Lysine-type peptidoglycan and serine proteinases , SPH1 and SPH2 that forms a complex with LGBP during immune response ( Liu et al . , 2011 ) . In Parhyale , we found one LGBP gene and two serine proteinases with high sequence identity to SPH1/2 in Pacifastacus . The D . pulex genome has also an expanded set of Gram-negative binding proteins ( proteins similar to LGBP ) suggesting a compensatory mechanism for the lost PGRPs ( McTaggart et al . , 2009 ) . Interestingly , we found a putative PGRP in the Remipede Speleonectes tulumensis ( Figure 13A ) providing further support for sister group relationship of Remipedia and Hexapoda ( von Reumont et al . , 2012 ) . 10 . 7554/eLife . 20062 . 034Figure 13 . Comparison of innate immunity genes . ( A ) Phylogenetic tree of peptidoglycan recognition proteins ( PGRPs ) . With the exception of Remipedes , PGRPs were not found in Crustaceans . PGRPs have been found in Arthropods , including insects , Myriapods and Chelicerates . ( B ) Phylogenetic tree of Toll-like receptors ( TLRs ) generated from five Crustaceans , three Hexapods , two Chelicerates , one Myriapod and one vertebrate species . ( C ) Genomic organization of the Parhyale Dscam locus showing the individual exons and exon arrays encoding the immunoglobulin ( IG ) and fibronectin ( FN ) domains of the protein . ( D ) Structure of the Parhyale Dscam locus and comparison with the ( E ) Dscam loci from Daphnia pulex , Daphnia magna and Drosophila melanogaster . The white boxes represent the number of predicted exons in each species encoding the signal peptide ( red ) , the IGs ( blue ) , the FNs and transmembrane ( yellow ) domains of the protein . The number of alternatively spliced exons in the arrays encoding the hypervariable regions IG2 ( exon 4 in all species ) , IG3 ( exon 6 in all species ) and IG7 ( exon 14 in Parhyale , 11 in D . pulex and 9 in Drosophila ) are indicated under each species schematic in the purple , green and magenta boxes , respectively . Abbreviations of species used: Parhyale hawaiensis ( Phaw ) , Bombyx mori ( Bmor ) , Aedes aegypti ( Aaeg ) , Drosophila melanogaster ( Dmel ) , Apis mellifera ( Amel ) , Speleonectes tulumensis ( Stul ) , Strigamia maritima ( Smar ) , Stegodyphus mimosarum ( Smim ) , Ixodes scapularis ( Isca ) , Amblyomma americanum ( Aame ) , Nephila pilipes ( Npil ) , Rhipicephalus microplus ( Rmic ) , Ixodes ricinus ( Iric ) , Amblyomma cajennense ( Acaj ) , Anopheles gambiae ( Agam ) , Daphnia pulex ( Apul ) , Tribolium castaneum ( Tcas ) , Litopenaeus vannamei ( Lvan ) , Lepeophtheirus salmonis ( Lsal ) , Eucyclops serrulatus ( Eser ) , Homo sapiens ( H . sap ) . Both trees were constructed with RAxML using the WAG+G model from multiple alignments of protein sequences created with MUSCLE . DOI: http://dx . doi . org/10 . 7554/eLife . 20062 . 03410 . 7554/eLife . 20062 . 035Figure 13—source data 1 . Catalog of innate immunity related genes in Parhyale . Parhyale IDs and numbers of immune related genes in comparison to other species . DOI: http://dx . doi . org/10 . 7554/eLife . 20062 . 03510 . 7554/eLife . 20062 . 036Figure 13—figure supplement 1 . Overview of Parhyale Dscam structure and hypervariable regions ( A ) Overview of domain structure of Parhyale Dscam protein and position of primers used to assess use of exons in 3 hypervariable regions . ( B ) Sequence alignments of cloned hypervariable regions in IG2 and ( C ) IG3 and ( D ) IG7 . ( E ) Alignment of crustacean DsCam proteins . DOI: http://dx . doi . org/10 . 7554/eLife . 20062 . 036 Innate immunity in insects is transduced by three major signaling pathways: the Immune Deficiency ( Imd ) , Toll and Janus kinase/signal transducer and activator of transcription ( JAK/STAT ) pathways ( Dostert et al . , 2005; Tanji et al . , 2007 ) . We found 16 members of the Toll family in Parhyale including 10 Toll-like receptors ( TLRs ) ( Figure 13B ) . Some TLRs have been also implicated in embryonic tissue morphogenesis in Parhyale and other arthropods ( Benton et al . , 2016 ) . Additionally , we identified 7 Imd and 25 JAK/STAT pathway members including two negative regulators: suppressor of cytokine signaling ( SOCS ) , and protein inhibitor of activated STAT ( PIAS ) ( Arbouzova and Zeidler , 2006 ) ( Figure 13—source data 1 ) . The blood of arthropods ( hemolymph ) contains hemocyanin which is a copper-binding protein involved in the transport of oxygen , and circulating blood cells called hemocytes for the phagocytosis of pathogens . Phagocytosis by hemocytes is facilitated by the evolutionarily conserved gene family , the thioester-containing proteins ( TEPs ) ( Levashina et al . , 2001 ) . Previously sequenced Pancrustacean species contained between 2 to 52 TEPs . We find 5 TEPs in the Parhyale genome . Arthropod hemocyanins themselves are structurally related to phenoloxidases ( PO; ( Decker and Jaenicke , 2004 ) and can be converted into POs by conformational changes under specific conditions ( Lee et al . , 2004 ) . POs are involved in several biological processes ( like the melanization immune response , wound healing and cuticle sclerotization ) and we identified 7 PO genes in Parhyale . Interestingly , hemocyanins and PO activity have been shown to be highly abundant together with glycosyl hydrolases in the digestive system of Isopods and Amphipods , raising a potential mechanistic link between gut sterility and degradation of lignocellulose ( King et al . , 2010; Zimmer et al . , 2002 ) . Another well-studied transmembrane protein essential for neuronal wiring and adaptive immune responses in insects is the immunoglobulin ( Ig ) -superfamily receptor Down syndrome cell adhesion molecule ( Dscam ) ( Schmucker et al . , 2000; Watson et al . , 2005 ) . Alternative splicing of Dscam transcripts can result in thousands of different isoforms that have a common architecture but have sequence variations encoded by blocks of alternative spliced exons . The D . melanogaster Dscam locus encodes 12 alternative forms of exon 4 ( encoding the N-terminal half of Ig2 ) , 48 alternative forms of exon 6 ( encoding the N-terminal half of Ig3 ) , 33 alternative forms of exon 9 ( encoding Ig7 ) , and 2 alternative forms of exon 17 ( encoding transmembrane domains ) resulting in a total of 38 , 016 possible combinations . The Dscam locus in Parhyale ( and in other crustaceans analysed ) has a similar organization to insects; tandem arrays of multiple exons encode the N-terminal halves of Ig2 ( exon 4 array with at least 13 variants ) and Ig3 ( exon 6 array with at least 20 variants ) and the entire Ig7 domain ( exon 14 array with at least 13 variants ) resulting in at least 3380 possible combinations ( Figure 13C–E ) . The alternative splicing of hypervariable exons in Parhyale was confirmed by sequencing of cDNA clones amplified with Dscam-specific primers . Almost the entire Dscam gene is represented in a single genomic scaffold and exhibits high amino-acid sequence conservation with other crustacean Dscams ( Figure 13——figure supplement 1 ) . The number of Dscam isoforms predicted in Parhyale is similar to that predicted for Daphnia species ( Brites et al . , 2008 ) . It remains an open question whether the higher number of isoforms observed in insects coincides with the evolution of additional Dscam functions compared to crustaceans . From a functional genomics perspective , the Parhyale immune system appears to be a good representative of the malacostrocan or even multicrustacean clade that can be studied in detail with existing tools and resources . Non-coding RNAs are a central , but still a relatively poorly understood part of eukaryotic genomes . In animal genomes , different classes of small RNAs are key for genome surveillance , host defense against viruses and parasitic elements in the genome , and regulation of gene expression through transcriptional , post-transcriptional and epigenetic control mechanisms ( Castel and Martienssen , 2013; Aravin et al . , 2001; Caplen et al . , 2001; Brennecke et al . , 2007; Gu et al . , 2009; Lee et al . , 2012; He and Hannon , 2004; Thomson et al . , 2006; Filipowicz et al . , 2008 ) . The nature of these non-coding RNAs , as well as the proteins involved in their biogenesis and function , can vary between animals . For example , some nematodes have Piwi-interacting short RNAs ( piRNAs ) , while others have replaced these by alternate small RNA based mechanisms to compensate for their loss ( Sarkies et al . , 2015 ) . As a first step , we surveyed the Parhyale genome for known conserved protein components of the small interfering RNA ( siRNA/RNAi ) and the piRNA pathways ( Table 4 ) . We found key components of all major small RNA pathways , including 4 argonaute family members , 2 PIWI family members , and orthologs of D . melanogaster Dicer-1 and Dicer-2 , drosha and loquacious , ( Figure 14—figure supplement 1 ) . Among Argonaute genes , Parhyale has 1 AGO-1 ortholog and 3 AGO-2 orthologs , which is presumably a malacostraca-specific expansion . While Parhyale only has 2 PIWI family members , other crustacean lineages have clearly undergone independent expansions of this protein family . Unlike in C . elegans , many mammals , fish and insects ( but not D . melanogaster ) , we did not find any evidence in the Parhyale genome for the SID-1 ( systemic RNA interference defective ) transmembrane protein that is essential for systemic RNAi ( Dong and Friedrich , 2005;Honeybee Genome Sequencing Consortium , 2006; Xu and Han , 2008 ) . Species without a SID-1 ortholog can silence genes only in a cell-autonomous manner ( Roignant et al . , 2003 ) . This feature has important implications for future design of RNAi experiments in Parhyale . 10 . 7554/eLife . 20062 . 037Table 4 . Small RNA processing pathway members . The Parhyale orthologs of small RNA processing pathway members . DOI: http://dx . doi . org/10 . 7554/eLife . 20062 . 037GeneCountsGen IDArmitage2phaw_30_tra_m . 006391 phaw_30_tra_m . 007425Spindle_E3phaw_30_tra_m . 000091 phaw_30_tra_m . 020806 phaw_30_tra_m . 018110rm627phaw_30_tra_m . 014329 phaw_30_tra_m . 012297 phaw_30_tra_m . 004444 phaw_30_tra_m . 012605 phaw_30_tra_m . 001849 phaw_30_tra_m . 006468 phaw_30_tra_m . 023485Piwi/aubergine2phaw_30_tra_m . 011247 phaw_30_tra_m . 016012Dicer 11phaw_30_tra_m . 001257Dicer 21phaw_30_tra_m . 021619argonaute 11phaw_30_tra_m . 006642arogonaute 23phaw_30_tra_m . 021514 phaw_30_tra_m . 018276 phaw_30_tra_m . 012367Loquacious2phaw_30_tra_m . 006389 phaw_30_tra_m . 000074Drosha1phaw_30_tra_m . 015433 We also assessed the miRNA and putative long non-coding RNAs ( lncRNA ) content of Parhyale using both MiRPara and Rfam ( Wu et al . , 2011; Nawrocki et al . , 2015 ) . We annotated 1405 homologues of known non-coding RNAs using Rfam . This includes 980 predicted tRNAs , 45 rRNA of the large ribosomal subunit , 10 rRNA of the small ribosomal subunit , 175 snRNA components of the major spliceosome ( U1 , U2 , U4 , U5 and U6 ) , 5 snRNA components of the minor spliceosome ( U11 , U12 , U4atac and U6atac ) , 43 ribozymes , 38 snoRNAs , 71 conserved cis-regulatory element derived RNAs and 42 highly conserved miRNA genes ( Source code 6 ) . Parhyale long non-coding RNAs ( lncRNAs ) were identified from the transcriptome using a series of filters to remove coding transcripts producing a list of 220 , 284 putative lncRNAs ( 32 , 223 of which are multi-exonic ) . Only one Parhyale lncRNA has clear homology to another annotated lncRNA , the sphinx lncRNA from D . melanogaster ( Wang et al . , 2002 ) . We then performed a more exhaustive search for miRNAs using MiRPara ( Source code 6 ) and a previously published Parhyale small RNA read dataset ( Blythe et al . , 2012 ) . We identified 1403 potential miRNA precursors represented by 100 or more reads . Combining MiRPara and Rfam results , we annotated 31 out of the 34 miRNA families found in all Bilateria , 12 miRNAs specific to Protostomia , 4 miRNAs specific to Arthropoda and 5 miRNAs previously found to be specific to Mandibulata ( Figure 14 ) . We did not identify mir-125 , mir-283 and mir-1993 in the Parhyale genome . The absence of mir-1993 is consistent with reports that this miRNA was lost during Arthropod evolution ( Wheeler et al . , 2009 ) . While we did not identify mir-125 , we observed that mir-100 and let-7 occurred in a cluster on the same scaffold ( Figure 14—figure supplement 2 ) , where mir-125 is also present in other animals . The absence of mir-125 has been also reported for the centipede genome ( Chipman et al . , 2014 ) . mir-100 is one of the most primitive miRNAs shared by Bilateria and Cnidaria ( Grimson et al . , 2008; Wheeler et al . , 2009 ) . The distance between mir-100 and let-7 genes within the cluster can vary substantially between different species . Both genes in Parhyale are localized within a 9 . 3kb region ( Figure 14—figure supplement 2 ) as compared to 3 . 8kb in the mosquito Anopheles gambiae and 100bp in the beetle Tribolium ( Behura , 2007 ) . Similar to D . melanogaster and the polychaete Platynereis dumerilii , we found that Parhyale mir-100 and let-7 are co-transcribed as a single , polycistronic lncRNA . We also found another cluster with miR-71 and mir-2 family members which is conserved across many invertebrates ( Marco et al . , 2014 ) ( Figure 14—figure supplement 2 ) . 10 . 7554/eLife . 20062 . 038Figure 14 . Evolution of miRNA families in Eumetazoans . Phylogenetic tree showing the gains ( in green ) and losses ( in red ) of miRNA families at various taxonomic levels of the Eumetazoan tree leading to Parhyale . miRNAs marked with plain characters were identified by MirPara with small RNA sequencing read support . miRNAs marked with bold characters were identified by Rfam and MirPara with small RNA sequencing read support . DOI: http://dx . doi . org/10 . 7554/eLife . 20062 . 03810 . 7554/eLife . 20062 . 039Figure 14—source data 1 . RFAM based annotation of the Parhyale genome . RFAM annotation of the Parhyale genome . DOI: http://dx . doi . org/10 . 7554/eLife . 20062 . 03910 . 7554/eLife . 20062 . 040Figure 14—figure supplement 1 . Phylogenetic trees of Dicer and PIWI/AGO genes . ( A ) Phylogenetic tree of Dicer family genes , including two Dicer genes from Parhyale . ( B ) Phylogenetic tree of PIWI/AGO genes , including several Parhyale genes . DOI: http://dx . doi . org/10 . 7554/eLife . 20062 . 04010 . 7554/eLife . 20062 . 041Figure 14—figure supplement 2 . Examples of miRNAs in the Parhyale genome . ( A ) Parhyale mir-100 and let-7 and clustered together in the intron of a putative lncRNA ( B ) A Parhyale mir-71/mir-2 family cluster ( C ) Parhyale mir-10 is in a conserved position in the genome between the Dfd and Scr Hox genes ( D ) Alignment of the predicted mir-10 precursor with mir-10 precursors from other species . DOI: http://dx . doi . org/10 . 7554/eLife . 20062 . 041 Conserved linkages have also been observed between miRNAs and Hox genes in Bilateria ( Enright et al . , 2003a; Tanzer et al . , 2005; Lemons and McGinnis , 2006; Stark et al . , 2008; Shippy et al . , 2008 ) . For example , the phylogenetically conserved mir-10 is present within both vertebrate and invertebrate Hox clusters between Hoxb4/Dfd and Hoxb5/Scr ( Enright et al . , 2003b ) . In the Parhyale genome and Hox BAC sequences , we found that mir-10 is also located between Dfd and Src on BAC clone PA179-K23 and scaffold phaw_30 . 0001203 ( Figure 14——figure supplement 2 ) . However , we could not detect mir-iab-4 near the Ubx and AbdA genes in Parhyale , the location where it is found in other arthropods/insects ( Cumberledge et al . , 1990 ) . Preliminary evidence regarding the presence of PIWI proteins and other piRNA pathway proteins also suggests that the piRNA pathway is likely active in Parhyale , although piRNAs themselves await to be surveyed . The opportunity to study these piRNA , miRNA and siRNA pathways in a genetically tractable crustacean system will shed further light into the regulation and evolution of these pathways and their contribution to morphological diversity . Methylation of cytosine residues ( m5C ) in CpG dinucleotides in animal genomes is regulated by a conserved multi-family group of DNA methyltransferases ( DNMTs ) with diverse roles in the epigenetic control of gene expression , genome stability and chromosome dynamics ( Zemach et al . , 2010; Law and Jacobsen , 2010; Jones , 2012 ) . The phylogenetic distribution of DNMTs in Metazoa suggests that the bilaterian ancestor had at least one member of the Dnmt1 and Dnmt3 families ( involved in de novo methylation and maintenance of DNA methylation ) and the Dnmt2 family ( involved in tRNA methylation ) , as well as additional RNA methyltransferases ( Jones and Liang , 2009; Jeltsch et al . , 2016 ) . Many animal groups have lost some of these DNA methyltransferases , for example DNMT1 and 3 are absent from D . melanogaster and flatworms ( Goll et al . , 2006; Jaber-Hijazi et al . , 2013 ) , while DNMT2 is absent from nematodes C . elegans and C . briggsae . The Parhyale genome encodes members of all 3 families DNMT1 , DNMT3 and DNMT2 , as well as 2 orthologs of conserved methyl-CpG-binding proteins and a single orthologue of Tet2 , an enzyme involved in DNA demethylation ( Hackett et al . , 2013 ) ( Figure 15A and Figure 15—source data 1 ) . 10 . 7554/eLife . 20062 . 042Figure 15 . Analysis of Parhyale genome methylation . ( A ) Phylogenetic tree showing the families and numbers of DNA methyltransferases ( DNMTs ) present in the genomes of indicated species . Parhyale has one copy from each DNMT family . ( B ) Amounts of methylation detected in the Parhyale genome . Amount of methylation is presented as percentage of reads showing methylation in bisulfite sequencing data . DNA methylation was analyzed in all sequence contexts ( CG shown in dark , CHG in blue and CHH in red ) and was detected preferentially in CpG sites . ( C ) Histograms showing mean percentages of methylation in different fractions of the genome: DNA transposons ( DNA ) , long terminal repeat transposable elements ( LTR ) , rolling circle transposable elements ( RC ) , long interspersed elements ( LINE ) , coding sequences ( cds ) , introns , promoters , and the rest of the genome . DOI: http://dx . doi . org/10 . 7554/eLife . 20062 . 04210 . 7554/eLife . 20062 . 043Figure 15—source data 1 . Genes involved with epigenetic modification . Catalog of Parhyale genes involved in DNA methylation and histone modifications . DOI: http://dx . doi . org/10 . 7554/eLife . 20062 . 043 We used genome wide bisulfite sequencing to confirm the presence and also assess the distribution of CpG dinucleotide methylation . Our results indicated that 20–30% of Parhyale DNA is methylated at CpG dinucleotides ( Figure 15B ) . The Parhyale methylation pattern is similar to that observed in vertebrates , with high levels of methylation detected in transposable elements and other repetitive elements , in promoters and gene bodies ( Figure 15C ) . A particular class of rolling-circle transposons are very highly methylated in the genome , potentially implicating methylation in silencing these elements . For comparison , about 1% or less of CpG-associated cytosines are methylated in insects like Drosophila , Apis , Bombyx and Tribolium . ( Feng et al . , 2010; Jeltsch , 2010; Zemach et al . , 2010 ) . These data represent the first documentation of a crustacean methylome . Considering the utility of Parhyale for genetic and genomic research , we anticipate future investigations to shed light on the functional importance and spatiotemporal dynamics of epigenetic modifications during normal development and regeneration , as well as their relevance to equivalent processes in vertebrate systems . Parhyale has already emerged as a powerful model for developmental genetic research where the expression and function of genes can be studied in the context of stereotyped cellular processes and with a single-cell resolution . Several experimental approaches and standardized resources have been established to study coding and non-coding sequences ( Table 1 ) . These functional studies will be enhanced by the availability of the assembled and annotated genome presented here . As a first application of these resources , we tested the efficiency of the CRISPR/Cas system for targeted genome editing in Parhyale ( Mali et al . , 2013; Jinek et al . , 2012; Cong et al . , 2013; Gilles and Averof , 2014; Martin et al . , 2015; Serano et al . , 2015 ) . In these studies , we targeted the Distal-less patterning gene ( called PhDll-e ) ( Liubicich et al . , 2009 ) that has a widely-conserved and highly-specific role in animal limb development ( Panganiban et al . , 1997 ) . We first genotyped our wild-type laboratory culture and found two PhDll-e alleles with 23 SNPs and 1 indel in their coding sequences and untranslated regions . For PhDll-e knock-out , two sgRNAs targeting both alleles in their coding sequences downstream of the start codon and upstream of the DNA-binding homeodomain were injected individually into 1-cell-stage embryos ( G0 generation ) together with a transient source of Cas9 ( Figure 16—figure supplement 1 A-B ) . Both sgRNAs gave rise to animals with truncated limbs ( Figure 16A and B ) ; the first sgRNA at a relatively low percentage around 9% and the second one at very high frequencies ranging between 53% and 76% ( Figure 16—figure supplement 1 ) . Genotyping experiments revealed that injected embryos carried PhDll-e alleles modified at the site targeted by each sgRNA ( Figure 16—figure supplement 1B ) . The number of modified PhDll-e alleles recovered from G0s varied from two , in cases of early bi-allelic editing at the 1-cell-stage , to three or more , in cases of later-stage modifications by Cas9 ( Figure 16—figure supplement 1C ) . We isolated indels of varying length that were either disrupting the open reading frame , likely producing loss-of-function alleles or were introducing in-frame mutations potentially representing functional alleles ( Figure 16—figure supplement 1C–D ) . In one experiment with the most efficient sgRNA , we raised the injected animals to adulthood and set pairwise crosses between 17 fertile G0s ( 10 male and 7 female ) : 88% ( 15/17 ) of these founders gave rise to G1 offspring with truncated limbs , presumably by transmitting PhDll-e alleles modified by Cas9 in their germlines . We tested this by genotyping individual G1s from two of these crosses and found that embryos bearing truncated limbs were homozygous for loss-of-function alleles with out-of-frame deletions , while their wild-type siblings carried one loss-of-function allele and one functional allele with an in-frame deletion ( Figure 16—figure supplement 1 D ) . 10 . 7554/eLife . 20062 . 044Figure 16 . CRISPR/Cas9-based genome editing in Parhyale . ( A ) Wild-type morphology . ( B ) Mutant Parhyale with truncated limbs after CRISPR-mediated knock-out ( DllKO ) of the limb patterning gene Distal-less ( PhDll-e ) . Panels show ventral views of juveniles stained for cuticle and color-coded by depth with anterior to the left . ( C ) Fluorescent tagging of PhDll-e expressed in most limbs ( shown in cyan ) by CRISPR-mediated knock-in ( DllKI ) using the non-homologous-end-joining repair mechanism . Panel shows a lateral view with anterior to the left and dorsal to the top of a live embryo ( stage S22 ) with merged bright-field and fluorescence channels . Yolk autofluorescence produces a dorsal crescent of fluorescence in the gut . Scale bars are 100 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 20062 . 04410 . 7554/eLife . 20062 . 045Figure 16—figure supplement 1 . CRISPR experiments targeting the Distalless locus . CRSIPR/Cas-based targeted genome editing in Parhyale . ( A ) Summary of gene knock-out experiments . ( B ) Illustration of the targeted PhDll-e ( Dll ) cDNA showing the 5’ and 3’ untranslated regions ( UTRs ) , the coding sequence with the homeodomain ( black box ) and the positions targeted by the two sgRNAs Dll1 and Dll2 . ( C ) Genotyping of a mosaic mutant embryo ( F0 generation ) with truncated appendages that was injected with Cas9 protein and the Dll1 sgRNA ( Dll1+PAM sequence in red ) . This animal carried multiple Dll alleles with deletions ( in yellow ) or insertions ( in cyan ) in the region targeted by Dll1 downstream of the start codon ( in green ) . Most of these alleles likely encoded truncated non-functional proteins , while a few alleles likely encoded functional proteins missing a few aminoacids at the targeted region ( putative number of aminoacids shown on the right ) . ( D ) Genotyping of wild-type and mutant embryos ( F1 generation ) from two separate crosses ( top and bottom black boxes ) of F0 animals injected with Cas9 protein and the Dll2 sgRNA ( Dll2+PAM sequence in red ) . Each mutant F1 carried two non-functional Dll alleles encoding truncated proteins , while their wild-type siblings carried one functional allele and one non-functional allele ( putative number of aminoacids shown on the right ) . ( E ) Summary of targeted gene knock-in based on the non-homologous end joining repair mechanism . ( F ) Schematic representation of the endogenous Dll locus with the non coding sequences shown in blue and the coding sequences in cyan ( left ) , and of the tagging plasmid carrying a copy of the Dll coding sequence ( in green ) , the T2A self-cleaving peptide ( in purple ) , a fusion of the Parhyale histone H2B with the Ruby 2 monomeric red fluorescent protein ( in magenta ) and the Dll 3’UTR ( in dark green ) . The Dll2+PAM sequences ( underlined ) and flanking sequences in the Dll locus and plasmid are shown in cyan and green , respectively . A single nucleotide substitution ( A>T shown in magenta ) right after the PAM sequence was introduced on purpose in the plasmid to discriminate the tagged sequence from the original one . The left and right junctions between the endogenous and inserted sequences were recovered by PCR from transgenic animals with fluorescent limbs using the indicated pairs of primers ( magenta and green , respectively ) . The tagged Dll locus is likely encoding a functional Dll protein ( with a small 7-aminoacid deletion in the region targeted by Dll2 and a stretch of T2A aminoacids in its C-terminus ) and a nuclear fluorescent reporter ( with the remaining T2A aminoacids in its N-terminus ) . DOI: http://dx . doi . org/10 . 7554/eLife . 20062 . 045 The non-homologous end joining ( NHEJ ) repair mechanism operating in the injected cells can be exploited not only for gene knock-out experiments described above , but also for CRISPR knock-in approaches where an exogenous DNA molecule is inserted into the targeted locus in a homology-independent manner . This homology-independent approach could be particularly useful for Parhyale that exhibits high levels of heterozygosity and polymorphisms in the targeted laboratory populations , especially in introns and intergenic regions . To this end , we co-injected into 1-cell-stage embryos the Cas9 protein together with the strongest sgRNA and a tagging plasmid . The plasmid was designed in such a way that upon its linearization by the same sgRNA and Cas9 and its integration into the PhDll-e locus in the appropriate orientation and open reading frame , it would restore the endogenous PhDll-e coding sequence in a bicistronic mRNA also expressing a nuclear fluorescent reporter . Among injected G0s , about 7% exhibited a nuclear fluorescence signal in the distal ( telopodite and exopodite ) parts of developing appendages ( Figure 16C and Figure 16—figure supplement 1 E ) , which are the limb segments that were missing in the knock-out experiments ( Figure 16B ) . Genotyping of one of these embryos demonstrated that the tagged PhDll-e locus was indeed encoding a functional PhDll-e protein with a small in-frame deletion around the targeted region ( Figure 16—figure supplement 1 F ) . These results , together with the other recent applications of the CRISPR/Cas system to study Hox genes in Parhyale ( Martin et al . , 2015; Serano et al . , 2015 ) , demonstrate that the ability to manipulate the fertilized eggs together with the slow tempo of early cleavages can result in very high targeting frequencies and low levels of mosaicism for both knock-out and knock-in approaches . Considering the usefulness of the genome-wide resources described in this report , we anticipate that the Parhyale embryo will prove an extremely powerful system for fast and reliable G0 screens of gene expression and function . In this article we described the first complete genome of a malacostracan crustacean species , the genome of the marine amphipod Parhyale hawaiensis . At an estimated size of 3 . 6 Gb , it is among the largest genomes submitted to NCBI . The Parhyale genome reported here is that of a single adult male from a sib-bred line called Chicago-F . We find Parhyale has an abundance of repetitive sequence and high levels of heterozygosity in the individual sequenced . Combined with analysis of available transcriptome sequences and independently sequenced genomic BAC clones , we conclude high levels of heterozygosity are representative of high levels of single and polynucleotide polymorphisms in the broader laboratory population . Our comparative bioinformatics analyses suggest that the expansion of repetitive sequences and the increase in gene size due to an expansion of intron size have contributed to the large size of the genome . Despite these challenges , the Parhyale genome and associated transcriptomic resources reported here provide a useful assembly of most genic regions in the genome and a comprehensive description of the Parhyale transcriptome and proteome . Parhyale has emerged since the early 2000’s as an attractive animal model for developmental genetic and molecular cell biology research . It fulfills several desirable biological and technical requirements as an experimental model , including a relatively short life-cycle , year-round breeding under standardized laboratory conditions , availability of thousands of eggs for experimentation on a daily basis , and amenability to various embryological , cellular , molecular genetic and genomic approaches . In addition , Parhyale has stereotyped cell lineages and cell behaviors , a direct mode of development , a remarkable appendage diversity and the capacity to regenerate limbs post-embryonically . These qualities can be utilized to address fundamental long-standing questions in developmental biology , like cell fate specification , nervous system development , organ morphogenesis and regeneration ( Stamataki and Pavlopoulos , 2016 ) . Research on these topics will benefit enormously from the standardized genome-wide resources reported here . Forward and reverse genetic analyses using both unbiased screens and candidate gene approaches have already been devised successfully in Parhyale ( Table 1 ) . The availability of coding and non-coding sequences for all identified signaling pathway components , transcription factors and various classes of non-coding RNAs will dramatically accelerate the study of the expression and function of genes implicated in the aforementioned processes . Equally importantly , our analyses highlight additional areas where Parhyale could serve as a new experimental model to address other questions of broad biomedical interest . From a functional genomics perspective , the Parhyale immune system appears to be a good representative of the malacostracan or even the multicrustacean clade that can be studied in detail with existing tools and resources . Besides the evolutionary implications and the characterization of alternative strategies used by arthropods to defend against pathogens , a deeper mechanistic understanding of the Parhyale immune system will be relevant to aquaculture . Some of the greatest setbacks in the crustacean farming industry are caused by severe disease outbreaks . Parhyale is closely related to farmed crustaceans ( primarily shrimps , prawns and crayfish ) and the knowledge acquired from studying its innate immunity could help enhance the sustainability of this industry by preventing or controlling infectious diseases ( Stentiford et al . , 2012; Johnson et al . , 2008; Lu et al . , 2008; Rajesh Kumar et al . , 2008; Rowley and Pope , 2012 ) . An immune-related problem that will be also interesting to explore in Parhyale concerns the possibility of a sterile digestive tract similar to that proposed for limnoriid Isopods ( King et al . , 2010 ) . Parhyale , like limnoriid Isopods , encodes and expresses all enzymes required for lignocellulose digestion , suggesting that it is able to ’digest wood' by itself without symbiotic microbial partners . Of course , a lot of work still needs to be invested in the characterization of the cellulolytic system in Parhyale before any comparisons can be made with other well-established symbiotic digestion systems of lignocellulose . Nevertheless , the possibility of an experimentally tractable animal model that serves as a living bioreactor to convert lignocellulose into simpler metabolites , suggests that future research in Parhyale may also have a strong biotechnological potential , especially for the production of biofuels from the most abundant and cheapest raw material , plant biomass . Although more high-quality genomes with a broader phylogenetic coverage are still needed for meaningful evolutionary comparisons , our observations from analysing the Parhyale genome and other crustacean data sets also contribute to the ongoing debate on the relationships between crustacean groups . While the analysis of shared orthologous groups did not provide clear support for either the Allotriocarida hypothesis ( uniting Branchiopoda with Hexapoda ) or the Vericrustacea hypothesis ( uniting Branchiopoda with Malacostraca ) , we noted the presence of GH7 genes and the absence of PGRP genes in branchiopod and multicrustacean genomes supporting the Vericrustacea hypothesis . It still remains to be proven how reliable these two characters will be to distinguish between these alternative phylogenetic affinities . Finally , Parhyale was introduced recently as a new model for limb regeneration ( Konstantinides and Averof , 2014 ) . In some respects , including the segmented body plan , the presence of a blood system and the contribution of lineage-committed adult stem cells to newly formed tissues , regeneration in Parhyale may resemble the process in vertebrates more than other established invertebrate models ( e . g . planarians , hydra ) . Regenerative research in Parhyale has been founded on transgenic approaches to label specific populations of cells and will be further assisted by the resources presented here . Likewise , we expect that the new genomic information and CRISPR-based genome editing methodologies together with all other facets of Parhyale biology will open other new research avenues not yet imagined .
About 10 µg of genomic DNA were isolated from a single adult male from the Chicago-F isofemale line established in 2001 ( Parchem et al . , 2010 ) . The animal was starved for one week and treated for 3 days with penicillin-streptomycin ( 100x , Gibco/Thermo Fisher Scientific ) , tetracycline hydrochloride ( 20 µg/ml , Sigma-Aldrich ) and amphotericin B ( 200x , Gibco/Thermo Fisher Scientific ) . It was then flash frozen in liquid nitrogen , homogenized manually with a pestle in a 1 . 5 ml microtube ( Kimble Kontes ) in 600 µl of Lysis buffer ( 100 mM Tris-HCl pH 8 , 100 mM NaCl , 50 mM EDTA , 0 . 5% SDS , 200 µg/ml Proteinase K , 20 µg/ml RNAse A ) . The lysate was incubated for 3 hr at 37°C , followed by phenol/chloroform extractions and ethanol precipitation . The condensed genomic DNA was fished out with a Pasteur pipette , washed in 70% ethanol , air-dried , resuspended in nuclease-free water and analysed on a Qubit fluorometer ( Thermo Fisher Scientific ) and on a Bioanalyzer ( Agilent Technologies ) . All genome libraries were prepared from this sample: 1 µg of genomic DNA was used to generate the shotgun libraries using the TruSeq DNA Sample Prep kit ( Illumina ) combined with size-selection on a LabChip XT fractionation system ( Caliper Life Sciences Inc ) to yield 2 shotgun libraries with average fragment sizes 431 bp and 432 bp , respectively; 4 µg of genomic DNA were used to generate 4 mate-pair libraries with average fragment sizes 5 . 5 kb , 7 . 3 kb , 9 . 3 kb and 13 . 8 kb using the Nextera Mate Pair Sample Preparation kit ( Illumina ) combined with agarose size selection . All libraries were sequenced on a HiSeq 2500 instrument ( Illumina ) using paired-end 150 nt reads . For chromosome spreads , tissue was obtained from embryos at stages 14–18 ( Browne et al . , 2005 ) . Eggs were taken from the mother and incubated for 1–2 hr in isotonic colchicine solution ( 0 . 05% colchicine , artificial sea water ) . After colchicine incubation , the embryonic tissue was dissected from the egg and placed in hypotonic solution ( 0 . 075 M KCl ) for 25 min . For tissue fixation , we replaced the hypotonic solution with freshly prepared ice-chilled Carnoy’s fixative ( six parts ethanol , three parts methanol and one part anhydrous acetic acid ) for 25 min . The fixed tissue was minced with a pair of fine tungsten needles in Carnoy’s solution and the resulting cell suspension was dropped with a siliconized Pasteur pipette from a height of about 5 cm onto a carefully cleaned ice-chilled microscopic slide . After partial evaporation of the Carnoy’s fixative the slides were briefly exposed a few times to hot water vapors to rehydrate the tissue . The slides were then dried on a 75°C metal block in a water bath . Finally , the slides with prepared chromosomes were aged overnight at 60°C . After DNA staining either with Hoechst ( H33342 , Molecular Probes ) or with DAPI ( Invitrogen ) , chromosomes were counted on a Zeiss Axioplan II Imaging equipped with C-Apochromat 63x/1 . 2 NA objective and a PCO pixelfly camera . FIJI was used to improve image quality ( contrast and brightness ) and FIJI plugin 'Cell Counter’ was used to determine the number of chromosomes . The Parhyale raw data and assembled data are available on the NCBI website . Genome assembly was done with Abyss ( Simpson et al . , 2009 ) at two different k-mer settings ( 70 , 120 ) and merged with GAM-NGS . Scaffolding was performed with SSPACE ( Boetzer et al . , 2011 ) . We chose cut-offs of >95% overlap length and >95% identity when removing shorter allelic contigs before scaffolding as these gave better scaffolding results as assessed by assembly metrics . Transcriptome assembly was performed with Trinity ( Haas et al . , 2013 ) . The completeness of the genome and transcriptome was assessed by blasting against CEGMA genes ( Parra et al . , 2007 ) and visualized by plotting the orthologue hit ratio versus e-value . K-mer analysis of variant and repetitive branching was performed with String Graph Assembler’s preqc module ( Simpson , 2014 ) . K-mer intersection analysis was performed using jellyfish2 ( Marçais and Kingsford , 2011 ) . Repetitive elements were annotated with RepeatModeler and RepeatMasker . An in-depth description of the assembly process and repeat masking is detailed in source code 1 and 2 . Parhyale transcriptome assembly was generated from Illumina reads collected from diverse embryonic stages ( Stages 19 , 20 , 22 , 23 , 25 , and 28 ) , and adult thoracic limbs and regenerating thoracic limbs ( 3 and 6 days post amputation ) . For the embryonic samples , RNA was extracted using Trizol; PolyA+ libraries were prepared with the Truseq V1 kit ( Illumina ) , starting with 0 . 6–3 . 5 μg of total mRNA , and sequenced on the Illumina Hiseq 2000 as paired-end 100 base reads , at the QB3 Vincent J . Coates Genomics Sequencing Laboratory . For the limb samples , RNA was extracted using Trizol; PolyA+ libraries were prepared with the Truseq V2 kit ( Illumina ) , starting with 1 μg of total mRNA , and sequenced on the Illumina Hiseq 2500 as paired-end 100 base reads , at the IGBMC Microarray and Sequencing platform . 260 million reads from embryos and 180 million reads from limbs were used for the transcriptome assembly . Prior to the assembly we trimmed adapter and index sequences using cutadapt ( Martin , 2011 ) . We also removed spliced leader sequences: GAATTTTCACTGTTCCCTTTACCACGTTTTACTG , TTACCAATCACCCCTTTACCAAGCGTTTACTG , CCCTTTACCAACTCTTAACTG , CCCTTTACCAACTTTACTG using cutadapt with 0 . 2 error allowance to remove all potential variants ( Douris et al . , 2009 ) . To assemble the transcriptome we used Trinity ( version trinityrnaseq_r20140413 ) ( Haas et al . , 2013 ) with settings: -min_kmer_cov 2 , -path_reinforcement_distance 50 . Gene prediction was done with a combination of Evidence Modeler ( Haas et al . , 2008 ) and Augustus ( Stanke and Waack , 2003 ) . The transcriptome was first mapped to the genome using GMAP ( Wu and Watanabe , 2005 ) . A secondary transcriptome reference assembly was performed with STAR/Cufflinks ( Trapnell et al . , 2010; Dobin et al . , 2013 ) . The transcriptome mapping and Cufflinks assembly was processed through the PASA pipeline ( Haas et al . , 2008 ) to consolidate the annotations . The PASA dataset , a set of Exonerate ( Slater and Birney , 2005 ) mapped Uniprot proteins , and Ab inito GeneMark ( Lukashin and Borodovsky , 1998 ) predictions were consolidated with Evidence Modeler to produce a set of gene annotations . A high confidence set of gene models from Evidence Modeler containing evidence from all three sources was used to train Augustus . Evidence from RepeatMasker ( Smit et al . , 2013 , PASA and Exonerate were then used to generate Augustus gene predictions . A final list of genes for down-stream analysis was generated using both transcriptome and gene predictions ( canonical proteome dataset ) . Detailed methods are described in Source code 3 . For variant analysis on the BAC clones , the short shot-gun library genomic reads were mapped to the BAC clones individually . GATK was then used to call variants . For variant analysis on the genic regions , transcript sequences used to generate the canonical proteome dataset were first aligned to the genome assembly . Genome alignments of less than 30 base pairs were discarded . The possible genome alignments were sorted based on number of mismatches with the top alignment having the least amount of mismatches . For each transcript , the top two genome aligments were used to call potential variants . Trascripts or parts of transcripts where there were more than five genomic mapping loci were discarded as potentially highly conserved domains or repetitive regions . Detailed methods of this process are described in Source code 4 . Parhyale genes ( nucleotide sequences ) were downloaded from GenBank . Each gene was used as a query for blastn against the Parhyale genome using the Geneious software ( Kearse et al . , 2012 ) . In each case two reference contig hits were observed where both had E values of close to zero . A new sequence called geneX_snp was created and this sequence was annotated with the snps and/or indels present in the alternative genomic contigs . To determine the occurrence of synonymous and non-synonymous substitutions , the original query and the newly created sequence ( with polymorphisms annotated ) were in silico translated into protein sequences followed by pairwise alignment . Regions showing amino acid changes were annotated as non-synonymous substitutions . Five random genes from the catalogue were selected for PCR , cloning and Sanger sequencing to confirm genomic polymorphisms and assess further polymorphism in the lab popultaion . Primers for genomic PCR designed to capture and amplify exon regions are listed as the following: dachshund ( PH1F = 5’- GGTGCGCTAAATTGAAGAAATTACG-3’ and PH1R = 5’- ACTCAGAGGGTAATAGTAACAGAA-3’ ) , distalless exon 2 ( PH2F = 5’-CACGGCCCGGCACTAACTATCTC-3’ and PH2R = 5’-GTAATATATCTTACAACAACGACTGAC-3’ ) , distalless exon 3 ( PH3F = 5’-GGTGAACGGGCCGGAGTCTC-3’ and PH3R = 5’-GCTGTGGGTGCTGTGGGT-3’ ) , homothorax ( PH4F = 5’-TCGGGGTGTAAAAAGGACTCTG-3’ and PH4R = 5’-AACATAGGAACTCACCTGGTGC-3’ ) , orthodenticle ( PH5F = 5’-TTTGCCACTAACACATATTTCGAAA-3’ and PH5R = 5’-TCCCAAGTAGATGATCCCTGGAT-3’ ) and prospero ( PH6F = 5’-TACACTGCAACATCCGATGACTTA-3’ and PH6R = 5’-CGTGTTATGTTCTCTCGTGGCTTC-3’ ) . Evolutionary analyses and comparative genomics were performed with 16 species: D . melanogaster , A . gambiae , D . pulex , L . salmonis , S . maritima , S . mimosarum , M . martensii , I . scapularis , H . dujardini , C . elegans , B . malayi , T . spiralis , M . musculus , H . sapiens , and B . floridae . For orthologous group analyses , gene families were identified using OrthoFinder ( Emms and Kelly , 2015 ) . The canonical proteome was used as a query in BlastP against proteomes from 16 species to generate a distance matrix for OrthoFinder to normalize and then cluster with MCL . Detailed methods are described in Source code 5 . For the comparative BLAST analysis , five additional transcriptome datasets were used from the following crustacean species: Litopenaeus vannamei , Echinogammarus veneris , Eucyclops serrulatus , Calanus finmarchicus , Speleonectes tulumensis . Embryo fixation and in-situ hybridization was performed according to ( Rehm et al . , 2009 ) . To enhance the nascent nuclear signal over mature cytoplasmic transcript , we used either early germband embryos ( Stages 11 – 15 ) in which expression of lab , Dfd , and Scr are just starting ( Serano et al . , 2015 ) , or probes that contain almost exclusively intron sequence ( Ubx , abd-A , Abd-B , and en1 ) . Lab , Dfd , and Scr probes are described in ( Serano et al . , 2015 ) . Template for the intron-spanning probes were amplified using the following primers: en1-Intron1 , AAGACACGACGAGCATCCTG and CTGTGTATGGCTACCCGTCC; Ubx-Intron1 , GGTATGACAGCCGTCCAACA and AGAGTGCCAAGGATACCCGA; abd-A , CGATATACCCAGTCCGGTGC and TCATCAGCGAGGGCACAATT; Abd-B , GCTGCAGGATATCCACACGA and TGCAGTTGCCGCCATAGTAA . A T7-adapter was appended to the 5’ end of each reverse primer to enable direct transcription from PCR product . Probes were labeled with either Digoxigenin ( DIG ) or Dinitrophenol ( DNP ) conjugated UTPs , and visualized using sheep -DIG ( Roche ) and donkey -Sheep AlexaFluor 555 ( Thermo Fischer Scientific ) , or Rabbit -DNP ( Thermo Fischer Scientific ) and Donkey -Rabbit AlexaFluor 488 ( Jackson ImmunoResearch ) , respectively . Preparations were imaged on an LSM 780 scanning laser confocal ( Zeiss ) , and processed using Volocity software ( Perkin-Elmer ) . The identification of GH family genes was done by obtaining Pfam annotations ( Finn et al . , 2006 ) for the Parhyale canonical proteome . Pfam domains were classified into different GH families based on the CAZy database ( Cantarel et al . , 2009 ) . For immune-related genes , best-reciprocal blast was performed with ImmunoDB genes ( Waterhouse et al . , 2007 ) . Multiple sequence alignments of protein sequences for gene families of FGF , FGFR , CERS , GH7 , GH9 , PGRP , Toll-like receptors , DICER , Piwi and Argonaute were performed using MUSCLE ( Edgar , 2004 ) . Phylogenetic tree construction was performed with RAxML ( Stamatakis , 2014 ) using the WAG+G model from MUSCLE multiple alignments . Libraries for DNA methylation analysis by bisulfite sequencing were constructed from 100ng of genomic DNA extracted from one Parhyale male individual , using the Illumina Truseq DNA methylation kit according to manufacturers instructions . Alignments to the Parhyale genome were generated using the core Bismark module from the program Bismark ( Krueger and Andrews , 2011 ) , having first artificially joined the Parhyale contigs to generate 10 pseudo-contigs as the program is limited as to the number of separate contigs it can analyse . We then generated genome-wide cytosine coverage maps using the bismark_methylation_extraction module with the parameter 'CX specified to generate annotations of CG , CHH and CHG sites . In order to analyse genome-wide methylation patterns only cytosines with more than a 10 read depth of coverage were selected . Overall methylation levels at CG , CHH and CHG sites were generated using a custom Perl script . To analyse which regions were methylated we mapped back from the joined contigs to the original contigs and assigned these to functional regions based on RepeatMasker ( Smit et al . , 2013 ) and transcript annotations of repeats and genes respectively . To generate overall plots of methylation levels in different features we averaged over all sites mapping to particular features , focusing on CG methylation and measuring the% methylation at each site as the number of reads showing methylation divided by the total number of reads covering the site . Meta gene plots over particular features were generated similarly except that sites mapping within a series of 100 bp wide bins from 1000 bp upstream of the feature start site and onward were collated . For the identification of Dscam in the Parhyale , we used the Dscam protein sequence from crustaceans D . pulex ( Brites et al . , 2008 ) and L . vannamei ( Chou et al . , 2009-12 ) as queries to probe the assembled genome using tBlastN . A 300kb region on scaffold phaw_30 . 0003392 was found corresponding to the Parhyale Dscam extending from IG1 to FN6 exons . This sequence was annotated using transcriptome data together with manual searches for open reading frames to identify IG , FN exons and exon-intron boundaries ( Figure 13—supplemental figure 1 ) . Hypervariable regions of IG2 , IG3 and IG7 were also annotated accordingly on the scaffold ( Figure 13—supplemental figure 1 ) . This region represents a bona fide Dscam paralog as it matches the canonical extracellular Dscam domain structure of nine IGs – four FNs – one IG and two FNs . Parhyale mRNA extractions were performed using the Zymo Research Direct-zol RNA MiniPrep kit according to manufacturer’s instructions . Total RNA extract was used for cDNA synthesis using the Qiagen QuantiTect Reverse Transcription Kit according to manufacturer’s instructions . To identify and confirm potential hypervariable regions from the Parhyale ( Ph-Dscam ) transcript , three regions of Ph-Dscam corresponding to IG2 , IG3 and IG7 exons respectively were amplified using the following primer pairs . IG2 region: DF1 = 5’-CCCTCGTGTTCCCGCCCTTCAAC-3’ DR1 = 5’-GCGATGTGCAGCTCTCCAGAGGG-3’ IG3 region: DF2 = 5’-TCTGGAGAGCTGCACATCGCTAAT-3’ DR2 = 5’-GTGGTCATTGCGTACGAAGCACTG-3’ IG7 region: DF3 = 5’-CGGATACCCCATCGACTCCATCG-3’ DR3 = 5’-GAAGCCGTCAGCCTTGCATTCAA-3’ PCR of each region was performed using Phusion High-fidelity polymerase from Thermo Fisher Scientific and thermal cycling was done as the following: 98C 30s , followed by 30 cycles of 98C 10s , 67C 30s , 72C 1m30s , and then 72C 5m . PCR products were cloned into pGEMT-Easy vector and a total of 81 clones were selected and Sanger sequenced and in silico translated in the correct reading frame using Geneious ( R7; ( Kearse et al . , 2012 ) for multiple sequence alignment . Parhyale non-protein-coding RNAs were identified using two independent approaches . Infernal 1 . 1 . 1 ( Nawrocki and Eddy , 2013 ) was used with the RFAM 12 . 0 database ( Nawrocki et al . , 2015 ) to scan the genome to identify potential non-coding RNAs . Additionally , MiRPara ( Wu et al . , 2011 ) was used to scan the genome for potential miRNA precursors . These potential precursors were further filtered using small RNA read mapping and miRBase mapping ( Griffiths-Jones et al . , 2008 ) . Putative lncRNAs were identified from the transcriptome by applying filtering criteria including removal of known and predicted coding RNAs . Detailed methods are available in Supplementary Data 11 . To genotype our wild-type population , extraction of total RNA and preparation of cDNA from embryos were carried out as previously described ( Pavlopoulos et al . , 2009 ) . The PhDll-e cDNA was amplified with primers PhDlle_2For ( 5’-TTTGTCAGGGATCTGCCATT-3’ ) and PhDlle_1852Rev ( 5’-TAGCGGCTGACGGTTGTTAC-3’ ) , purified with the DNA Clean and Concentrator kit ( Zymo Research ) , cloned with the Zero Blunt TOPO PCR Cloning Kit ( Thermo Fisher Scientific ) and sequenced with primers M13 forward ( 5’- GTAAAACGACGGCCAG-3’ ) and M13 reverse ( 5’- CAGGAAACAGCTATGAC-3’ ) . Each template for sgRNA synthesis was prepared by annealing and PCR amplification of the sgRNA-specific forward primer Dll1: ( 18 nt PhDll-e-targeted sequence underlined ) 5’-GAAATTAATACGACTCACTATA AGAGTTGTTACCAAAGAAGTTTTAGAGCTAGAAATAGC-3’ or Dll2: ( 20 nt PhDll-e-targeted sequence underlined ) 5’-GAAATTAATACGACTCACTAT AGGCTTCCCCGCCGCCATGTAGTTTTAGAGCTAGAAATAGC-3’ together with the universal reverse primer: 5’-AAAAGCACCGACTCGGTGCCACTTTTTCAAGTTGATAA CGGACTAGCCTTATTTTAACTTGCTATTTCTAGCTCTAAAAC-3’ using the Phusion DNA polymerase ( New England Biolabs ) . Each PCR product was gel-purified with the Zymoclean DNA recovery kit ( Zymo Research ) and 150 ng of DNA were used as template in an in vitro transcription reaction with the Megashortscript T7 kit ( Thermo Fisher Scientific ) . A 4-hr incubation at 37°C was followed by DNAse digestion , phenol/chloroform extraction , ethanol precipitation and storage in ethanol at −20° C according to the manufacturer’s instructions . Before microinjection , a small aliquot of the sgRNA was centrifuged , the pellet was washed with 70% ethanol , resuspended in nuclease-free water and quantified on a Nanodrop spectrophotometer ( Thermo Scientific ) . The Cas9 was provided either as in vitro synthesized caped mRNA or as recombinant protein . Cas9 mRNA synthesis was carried out as previously described ( Kontarakis and Pavlopoulos , 2014 ) using plasmid T7-Cas9 ( a gift from David Stern and Justin Crocker ) linearized with EcoRI digestion . The lyophilized Cas9 protein ( PNA Bio Inc ) was resuspended in nuclease-free water at a concentration of 1 . 25 µg/µl and small aliquots were stored at −80°C . For microinjections , we mixed 400 ng/µl of Cas9 protein with 40–200 ng/µl sgRNA , incubated at 37°C for 5 min , transferred on ice , added the inert dye phenol red ( 5x from Sigma-Aldrich ) and , for knock-in experiments , the tagging plasmid at a concentration of 10 ng/µl . The injection mix was centrifuged for 20 min at 4°C and the cleared solution was microinjected into 1-cell-stage embryos as previously described ( Kontarakis and Pavlopoulos , 2014 ) . In the knock-out experiments , embryos were scored for phenotypes under a bright-field stereomicroscope 7–8 days after injection ( stage S25-S27 ) when organogenesis is almost complete and the limbs are clearly visible through the transparent egg shell . To image the cuticle , anaesthetized hatchlings were fixed in 2% paraformaldehyde in 1xPBS for 24 hr at room temperature . The samples were then washed in PTx ( 1xPBS containing 1% TritonX-100 ) and stained with 1 mg/ml Congo Red ( Sigma-Aldrich ) in PTx at room temperature with agitation for 24 hr . Stained samples were washed in PTx and mounted in 70% glycerol for imaging . Serial optical sections were obtained at 2 µm intervals with the 562 nm laser line on a Zeiss 710 confocal microscope using the Plan-Apochromat 10x/0 . 45 NA objective . Images were processed with Fiji ( http://fiji . sc ) and Photoshop ( Adobe Systems Inc ) . This methodology enabled us to also extract genomic DNA for genotyping from the same imaged specimen . Each specimen was disrupted with a disposable pestle in a 1 . 5 ml microtube ( Kimble Kontes ) in 50 µl of Squishing buffer ( 10 mM Tris-HCl pH 8 , 1 mM EDTA , 25 mM NaCl , 200 µg/ml Proteinase K ) . The lysate was incubated at 37°C for a minimum of 2 hr , followed by heat inactivation of the Proteinase K for 5 min at 95°C , centrifugation at full speed for 5 min and transferring of the cleared lysate to a new tube . To recover the sequences in the PhDll-e locus targeted by the Dll1 and Dll2 sgRNAs , 5 µl of the lysate were used as template in a 50 µl PCR reaction with the Phusion DNA polymerase ( New England Biolabs ) and primers 313For ( 5’-TGGTTTTAGCAACAGTGAAGTGA-3’ ) and 557Rev ( 5’-GACTGGGAGCGTGAGGGTA-3’ ) . The amplified products were purified with the DNA Clean and Concentrator kit ( Zymo Research ) , cloned with the Zero Blunt TOPO PCR Cloning Kit ( Thermo Fisher Scientific ) and sequenced with the M13 forward primer . For the knock-in experiments , we constructed the tagging plasmid pCRISPR-NHEJ-KI-Dll-T2A-H2B-Ruby2 that contained the PhDll-e coding sequence fused in-frame with the T2A self-cleaving peptide , the Parhyale histone H2B and the Ruby 2 monomeric red fluorescent protein , followed by the PhDll-e 3’UTR and the pGEM-T Easy vector backbone ( Promega ) . This tagging plasmid has a modular design with unique restriction sites for easy exchange of any desired part . More details are available upon request . Embryos co-injected with the Cas9 protein , the Dll2 sgRNA and the pCRISPR-NHEJ-KI-Dll-T2A-H2B-Ruby2 tagging plasmid were screened for nuclear fluorescence in the developing appendages under an Olympus MVX10 epi-fluorescence stereomicroscope . To image expression , live embryos at stage S22 were mounted in 0 . 5% SeaPlaque low-melting agarose ( Lonza ) in glass bottom microwell dishes ( MatTek Corporation ) and scanned as described above acquiring both the fluorescence and transmitted light on an inverted Zeiss 880 confocal microscope . To recover the chromosome-plasmid junctions , genomic DNA was extracted from transgenic siblings with fluorescent limbs and used as template in PCR reaction as described above with primer pair 313For and H2BRev ( 5’-TTACTTAGAAGAAGTGTACTTTG-3’ ) for the left junction and primer pair M13 forward and 557Rev for the right junction . Amplified products were purified and cloned as described above and sequenced with the M13 forward and M13 reverse primers . | The marine crustacean known as Parhyale hawaiensis is related to prawns , shrimps and crabs and is found at tropical coastlines around the world . This species has recently attracted scientific interest as a possible new model to study how animal embryos develop before birth and , because Parhyale can rapidly regrow lost limbs , how tissues and organs regenerate . Indeed , Parhyale has many characteristics that make it a good model organism , being small , fast-growing and easy to keep and care for in the laboratory . Several research tools have already been developed to make it easier to study Parhyale . This includes the creation of a system for using the popular gene editing technology , CRISPR , in this animal . However , one critical resource that is available for most model organisms was missing; the complete sequence of all the genetic information of this crustacean , also known as its genome , was not available . Kao , Lai , Stamataki et al . have now compiled the Parhyale genome – which is slightly larger than the human genome – and studied its genetics . Analysis revealed that Parhyale has genes that allow it to fully digest plant material . This is unusual because most animals that do this rely upon the help of bacteria . Kao , Lai , Stamataki et al . also identified genes that provide some of the first insights into the immune system of crustaceans , which protects these creatures from diseases . Kao , Lai , Stamataki et al . have provided a resource and findings that could help to establish Parhyale as a popular model organism for studying several ideas in biology , including organ regeneration and embryonic development . Understanding how Parhyale digests plant matter , for example , could progress the biofuel industry towards efficient production of greener energy . Insights from its immune system could also be adapted to make farmed shrimp and prawns more resistant to infections , boosting seafood production . | [
"Abstract",
"Introduction",
"Results",
"and",
"discussion",
"Materials",
"and",
"methods"
] | [
"developmental",
"biology",
"tools",
"and",
"resources",
"genetics",
"and",
"genomics"
] | 2016 | The genome of the crustacean Parhyale hawaiensis, a model for animal development, regeneration, immunity and lignocellulose digestion |
Genome-wide association studies have found variation within the complement factor H gene family links to host susceptibility to meningococcal disease caused by infection with Neisseria meningitidis ( Davila et al . , 2010 ) . Mechanistic insights have been challenging since variation within this locus is complex and biological roles of the factor H-related proteins , unlike factor H , are incompletely understood . N . meningitidis subverts immune responses by hijacking a host-immune regulator , complement factor H ( CFH ) , to the bacterial surface ( Schneider et al . , 2006; Madico et al . , 2007; Schneider et al . , 2009 ) . We demonstrate that complement factor-H related 3 ( CFHR3 ) promotes immune activation by acting as an antagonist of CFH . Conserved sequences between CFH and CFHR3 mean that the bacterium cannot sufficiently distinguish between these two serum proteins to allow it to hijack the regulator alone . The level of protection from complement attack achieved by circulating N . meningitidis therefore depends on the relative levels of CFH and CFHR3 in serum . These data may explain the association between genetic variation in both CFH and CFHR3 and susceptibility to meningococcal disease .
Neisseria meningitidis is an important cause of rapidly progressive septicaemia and meningitis in children and young adults ( Stephens et al . , 2007 ) ; case fatality rates for bacteraemic disease remain at around 10% , and a significant proportion of survivors are left with long-term sequelae ( Vyse et al . , 2013 ) . For most individuals , however , this human-specific bacterium is primarily a non-pathogenic commensal of the nasopharynx , and up to 40% of the population are healthy carriers ( Caugant and Maiden , 2009 ) . The mechanisms underlying human genetic influences that govern the development of invasive disease or asymptomatic carriage are incompletely understood . It is known , however , that complement activation is critical for protection against disease , evident from the susceptibility of individuals with genetic deficiency of either the alternative complement pathway ( AP ) or terminal pathway activation , and among those receiving therapy that prevents terminal pathway activation ( Schneider et al . , 2007; McKeage , 2011 ) . Amongst several strategies that promote complement evasion , N . meningitidis recruits the human negative complement regulator , complement factor H ( CFH ) , to its surface by expressing factor H binding protein ( fHbp ) ( Schneider et al . , 2006 , 2009; Madico et al . , 2007 ) . fHbp can be divided into three variant groups ( V1 , V2 , and V3 ) , which have >85% sequence identity within the groups but only 60–70% similarity between groups ( Masignani et al . , 2003; Brehony et al . , 2009 ) . fHbps from all variant groups bind CFH with a KD in the low nanomolar range ( Johnson et al . , 2012 ) . Therefore , it is significant that the only single nucleotide polymorphisms ( SNPs ) associated with meningococcal disease in a recent genome-wide association study ( GWAS ) were those among the CFH–CFHR locus . These included SNPs within CFH and the downstream gene , CFHR3 ( Davila et al . , 2010 ) . While CFH has been extensively studied ( reviewed in Makou et al . ( 2013 ) ) , the role of the serum protein CFHR3 has not been unambiguously established ( reviewed in Jozsi and Meri ( 2014 ) ) .
We recently demonstrated that three of the five CFHR proteins ( i . e . , CFHR1 , 2 , and 5 ) compete with CFH for binding to C3b on heterologous erythrocytes , thereby promoting complement activation ( Goicoechea de Jorge et al . , 2013 ) . Given the association of CFHR3 with meningococcal disease ( Davila et al . , 2010 ) , we investigated whether this protein and CFHR4 also act as CFH antagonists in this assay . We generated full-length CFHR3 and CFHR4B ( which both consist of five complement control protein [CCP] domains , Figure 1A ) and truncated versions of these proteins , and we examined their ability to influence complement activation by the AP , which is regulated by CFH ( Makou et al . , 2013 ) . Erythrocyte haemolysis assays demonstrate that CFHR3 and CFHR4B are also CFH antagonists on this surface , with their two C-terminal domains enabling them to compete with CFH for binding sites on cell complement fragments such as C3b ( Figure 1B , Figure 1—figure supplement 1A , B ) . Additional cell-surface recognition sites within other domains ( Figure 1A ) increase the activity of the full-length CFHR3 and CFHR4B as competitive antagonists of CFH ( Figure 1B ) at physiologically relevant concentrations of ∼1 μM ( Fritsche et al . , 2010; Hebecker and Jozsi , 2012 ) . 10 . 7554/eLife . 04008 . 003Figure 1 . CFHR3 binds N . meningitidis fHbp at the same site as CFH and promotes complement activation . ( A ) Complement control protein ( CCP ) domains of CFH , CFHR3 , and CFHR4 are shown with the sequence identity to CFH and between the CFHRs indicated . Key functional regions of CFH are noted . ( B ) CFH-dependent alternative pathway ( AP ) haemolytic assay ( Goicoechea de Jorge et al . , 2013 ) . Using a CFH dose that reduced lysis of guinea pig erythrocytes to 50% , the addition of increasing concentrations of either full-length or C-terminal fragments of CFHR3 or CFHR4 resulted in a dose-dependent increase in lysis . The two N-terminal CCP domains from CFHR3 had no effect ( not shown ) . ( C ) Surface plasmon resonance ( SPR ) was used to investigate interactions of CFHRs with fHbp . Full-length and fragments of CFH and CFHRs were injected ( black bar ) over surface bound fHbp3 . 28 . Only CFHR312 , full-length CFHR3 , or CFH67 bind to fHbp . ( D ) Dissociation constants ( KD ) measured by SPR ( CFHR312 ) or microscale thermophoresis ( CFHR3 ) indicated the fHbp binding site is within the two N-terminal domains of CFHR3 . ( E ) SPR with 200 nM protein flowing over V3 . 28 fHbp demonstrates that the single point mutation E255A , which ablates CFH binding , also ablates CFHR12 binding . DOI: http://dx . doi . org/10 . 7554/eLife . 04008 . 00310 . 7554/eLife . 04008 . 004Figure 1—figure supplement 1 . Increased haemolysis on addition of CFHR3 is dependent on presence of CFH and interaction of CFHR3 full length and N-terminal fragment with V3 . 28 fHbp . ( A ) Haemolysis of guinea pig erythrocytes by 5% CFH depleted serum ( CompTech , TX , USA ) was measured upon addition of a range of CFHR3 concentrations ( 4 nM–2 μM ) in Mg-EGTA to ensure activation by the alternative pathway ( AP ) alone . Haemolysis is presented as a percentage of maximal lysis by H2O . No significant differences in haemolysis were observed under any of the conditions . ( B ) Supplementing the CFH deficient serum ( CompTech , TX , USA ) with 100 nM CFH reduced lysis to ∼20% . In this CFH-supplemented experiment addition of 2 mM CFHR3 led to a significant ( p = 0 . 003 , two-tailed , unpaired t test ) increase in haemolysis as seen with the naturally CFH sufficient serum in Figure 1 . ( C ) Full-length CFHR3 interacts strongly with the carboxymethyldextran matrix used to immobilise proteins in SPR , therefore , microscale thermophoeresis ( Seidel et al . , 2013 ) , a solution technique , was used to measure the affinity for V3 . 28 fHbp . The data plotted are mean ± SE for nine repeats , which were simultaneously fit to derive the KD and error estimate shown . ( D ) SPR was used to measure the KD of the interaction with V3 . 28 fHbp ( see ‘Materials and methods’ ) . Data from a single experiment are shown whilst the KD and fit statistics are mean ± SD for four independent repeats . DOI: http://dx . doi . org/10 . 7554/eLife . 04008 . 004 N . meningitidis fHbp recruits CFH by high-affinity interactions primarily with domain 6 of CFH ( CFH6 ) with some limited contact to CFH7 ( Schneider et al . , 2009 ) . Since the first domains of the CFHRs share between 40% to 90% identity with CFH6 ( Figure 1A , Figure 2 ) , we next investigated whether fHbp also bound the CFHRs . Our findings ( Figure 1C ) demonstrate that CFHR3 binds fHbp with a KD of ∼3 nM , similar to affinities for CFH ( Schneider et al . , 2009; Johnson et al . , 2012 ) ; no significant binding was detected for the other CFHRs with fHbp . Of note , the first two domains of CFHR3 ( CFHR312; domain 1 shares 91% identity with CFH6 , Figure 1A ) and the full-length protein bind fHbp with similar affinities ( Figure 1E , Figure 1—figure supplement 1C , D ) . Furthermore , a point mutation within fHbp ( E255A ) that reduces binding to CFH by more than three orders of magnitude ( Johnson et al . , 2012 ) similarly reduces CFHR3 binding ( Figure 1E ) . The nature of the CFH and CFHR3 interactions with fHbp are therefore similar , with these complement factors occupying overlapping sites on fHbp , which is consistent with the high degree of conservation of their fHbp binding sites ( Figure 2 ) . 10 . 7554/eLife . 04008 . 005Figure 2 . Amino acid differences between CFHR3 and CFH in the fHbp binding site . ( A ) Sequence alignment between CFHR312 and CFH67 . Differences are highlighted in red text , the two differences within the fHbp binding site are highlighted in yellow and residues contacting fHbp in the CFH-fHbpV3 structure ( PDB ID 4AYM [Johnson et al . , 2012] ) , chains A and D are indicated by an asterisk below the CFH sequence . ( B ) CFH67 is shown as a grey surface with residues that differ in CFHR3 highlighted in red; fHbp is shown as a green ribbon . ( C ) Rotated 90° around x compared to the view in B . ( D ) Looking down onto the fHbp binding site on CFH the two , conservative , sequence differences between the CFH67 and CFHR312 found in the binding site are indicated in red on the grey surface . DOI: http://dx . doi . org/10 . 7554/eLife . 04008 . 005 Next we investigated whether CFHR3 binds to the surface of N . meningitidis . We generated a CFHR3-specific monoclonal antibody ( mAb ) ( Figure 3—figure supplement 1 ) , designated HSL1 . By flow cytometry we demonstrated that serum CFHR3 bound N . meningitidis M1239 ( which expresses V3 . 28 fHbp ) in a fHbp-dependent manner . No binding was detected following incubation of bacteria in sera from an individual homozygous for an allele in which there is combined deletion of the CFHR1 and CFHR3 genes ( ΔCFHR3/1 ) ( ΔCFHR3/1 , Figure 3A , B , C ) . We therefore tested the functional consequences of CFHR3 binding by comparing the effect of adding full length or domains of CFHR3 to N . meningitidis prior to incubation in normal human serum ( NHS ) . Addition of full-length CFHR3 or CFHR312 , which contains the fHbp binding site ( Figure 1C ) but has no CFH antagonist activity ( Figure 1B ) , resulted in significantly reduced bacterial survival ( CFHR3 51%; CFHR312 70%; no protein or CFHR345 83% , Figure 3D ) ; in contrast , CFHR3 had no detectable effect in the presence of heat-inactivated NHS ( which lacks complement activity , Figure 3D ) . Full-length CFHR3 has a more marked effect on bacterial survival than CFHR312 , consistent with the ability of the full-length CHHR3 to bind to C3b deposited on the bacterial surface ( via CFHR345 ) as well as competing with CFH for fHbp ( via CFHR312 ) . As CFHR345 does not bind fHbp ( Figure 1C ) , its CFH antagonist activity is not effectively localized to the meningococcal surface , providing an explanation for why these domains have no effect on bacterial survival ( Figure 3D ) . Similarly , addition of CFHR3 has no effect on serum survival of bacteria that do not express any CFH-binding protein ( Figure 3—figure supplement 2A , N . meningitidis strain lacking fHbp expression; Figure 3—figure supplement 2B , Escherichia coli strain with no expression of any CFH-binding protein ) . 10 . 7554/eLife . 04008 . 006Figure 3 . N . meningitidis binds CFHR3 on its surface in an fHbp-dependent manner promoting complement-mediated lysis . ( A–C ) Flow cytometry of N . meningitidis demonstrates that CFHR3 binds to fHbp on the bacterial surface . ( D ) Sensitivity of N . meningitidis strain M1239 to complement-mediated cell lysis after pre-incubation with 1 µM CFHR3 , CFHR312 , CFHR345 , or no protein . Faded symbols are bacteria incubated in identical conditions except using heat-inactivated NHS instead of NHS . Data presented as percentage survival relative to bacteria incubated in heat-inactivated NHS without additional CFHR3 . Line shows the mean of three independent biological experiments , each with three technical replicates; individual results indicated . Significance was calculated using a two-tailed unpaired t test comparing with values obtained with no additional protein: ****p < 0 . 0001; *p < 0 . 05; ns p > 0 . 05 . DOI: http://dx . doi . org/10 . 7554/eLife . 04008 . 00610 . 7554/eLife . 04008 . 007Figure 3—figure supplement 1 . Characterisation of the novel anti-CFHR3 antibody HSL1 . ( A ) Western blotting of recombinant CFHR3 and sera from individuals sufficient or deficient in CFHR3 demonstrates that HSL1 is specific for CFHR3 . ( B ) For screening of mAbs , ELISA plates ( F96 maxisorp , Nunc ) were coated with C-terminal pair of CCP domains of CFH and the CFHRs ( 5 µg/ml , 50 µl per well ) overnight at room temperature before blocking with 2% skimmed milk in PBS 0 . 05% Tween 20 . Plates were incubated with neat hybridoma culture supernatant and binding was detected with goat anti-mouse HRP ( 1 in 5000 , Dako ) followed by ELISA substrate and stop solution ( Roche ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04008 . 00710 . 7554/eLife . 04008 . 008Figure 3—figure supplement 2 . The effect of CFHR312 on bacterial survival is dose and fHbp dependent . ( A ) Survival of N . meningitidis M1239 ( filled circles ) and M1239Δfhbp ( unfilled circles ) in the presence of NHS following pre-incubation in different concentrations of CFHR312 ( indicated ) . Data presented as percent survival relative to bacteria incubated in the absence of serum and CFHR312 . Error bars are ±SEM of three independent experiments . ( B ) Serum survival of E . coli DH5α ( which does not express a CFH-binding surface protein ) is not altered by pre-incubation with CFHR3 . Data presented as percent survival relative to bacteria incubated in the absence of serum and CFHR3 . Error bars are ±SEM of three separate experiments repeated four times . Significance was calculated using a standard one-way ANOVA with Tukey's multiple comparison test ( n . s . , not significant; ****p < 0 . 0001 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04008 . 008 The finding that CFHR3 binding impairs bacterial survival in NHS led us to consider whether fHbp can distinguish between the limited sequence differences between CFH and CFHR3 to generate specificity for CFH . We previously identified a series of alanine substitutions in fHbp that significantly reduce or ablate CFH binding ( Johnson et al . , 2012 ) . Therefore , we measured the contribution of fHbp residues to interactions with CFHR3 to determine whether any amino acid changes altered the specificity of fHbp for the two complement factors ( Figure 4A , Figure 4—source data 1 ) . The results demonstrate that the majority of mutations that impair CFH binding ( Johnson et al . , 2012 ) also significantly reduce CFHR3 binding , so do not alter specificity for CFH; point mutations at only a few sites led to modest alterations in specificity at levels that are unlikely to be physiologically relevant given the relative affinities for CFHR3/CFH and their abundance in serum ( CFHR3 1–1 . 6 μM [Fritsche et al . , 2010; Hebecker and Jozsi , 2012]; CFH 1–4 μM [Hakobyan et al . , 2008] ) . Interestingly the V1 fHbp used as the background for the mutations ( V1 . 1 ) did reveal a degree of specificity for CFH over CFHR3 ( in the order of ∼20-fold ) in contrast to V2 and V3 fHbps , suggesting that certain fHbps can discriminate between CFH and CFHR3 . 10 . 7554/eLife . 04008 . 009Figure 4 . Key residues in fHbp that mediate binding to CFHR3 . ( A ) SPR was used to determine KDs for both CFH67 and CFHR312 binding to a panel of fHbps bearing alanine substitutions within the CFH67 binding site . ∼50 mutations were screened in each of three fHbps ( V1 . 1 , 2 . 21 and 3 . 28 [Johnson et al . , 2012] ) . Mutations altered KDs for both CFH67 and CFHR312 but some mutations led to some specificity for binding one or the other protein ( see full data in Extended Data File 1 ) . ( B ) KDs were determined for a panel of fHbps from disease causing N . meningitidis . V1 fHbps have some ability to preferentially bind CFH67 vs CFHR312 , whilst V2 and V3 fHbps were less able to distinguish the molecules ( full data in Extended Data File 2 ) . ( C ) AP-mediated bacterial killing of isogenic strains expressing different fHbps in the presence of CFHR3 . Data are coloured as in panel ( B ) and bacterial survival in serum is increased in the strain expressing the fHbpV1 . 1 ( that has the maximal specificity for CFH vs CFHR3 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04008 . 00910 . 7554/eLife . 04008 . 010Figure 4—source data 1 . Excel spreadsheet with full details of KDs for CFH and CFHR312 derived from 1:1 Langmuir fits of SPR data for alanine scanning mutants and natural variant fHbp sequences . DOI: http://dx . doi . org/10 . 7554/eLife . 04008 . 010 To investigate this further , we measured binding of CFH and CFHR3 to 25 naturally occurring fHbps that are prevalent in disease isolates of N . meningitidis ( Figure 4B , Figure 4—source data 1 ) ( Lucidarme et al . , 2009 ) . This demonstrated that V2 and V3 fHbps displayed no selective affinity for CFH over CFHR3 ( average selectivity being 1 . 8 ± 1 . 4 and 2 . 4 ± 1 . 0 , respectively ) . In contrast , V1 fHbps had lower affinities for CFHR3 compared to CFH ( average selectivity 8 . 9 ± 4 . 6-fold ) , with 12 of the 14 V1 fHbp variants binding CFH 5- to 20-fold more tightly than CFHR3 . Modeling the effect on the occupancy of fHbp with CFH indicates that certain V1 fHbps will be bound by CFH in the physiological situation regardless of CFHR3 levels ( Figure 5 ) . To examine whether this has any functional consequence , we examined complement AP killing of isogenic strains expressing fHbp sequences with different abilities to discriminate between CFH and CFHR3 ( Figure 4C ) ; AP killing was assessed to exclude any confounding influence of antibodies in sera preferentially recognizing different fHbps . Bacteria expressing V3 . 28 fHbp , which has identical affinities for CFH and CFHR3 , were the most sensitive to AP killing , while the strain expressing the V1 . 1 fHbp sequence ( which has ∼20-fold tighter binding to CFH than CFHR3 ) was the least sensitive; bacteria with fHbps with intermediate CFH specificity displayed intermediate levels of protection . This suggests that , even though the most specific fHbps bind CFHR3 with a KD significantly below the physiological concentration ( Figure 4—source data 1 ) , the ability of fHbp to favour CFH binding promotes bacterial survival and offers a potential explanation for the prevalence of strains expressing V1 fHbp causing invasive disease ( Lucidarme et al . , 2009 ) . 10 . 7554/eLife . 04008 . 011Figure 5 . Modelling the occupancy of bacterial cell surface fHbp assuming different levels of specificity for CFH over CFHR3 . Each molecule of fHbp on the bacterial surface can bind with one molecule of CFHR3 or one molecule of CFH . Assuming that the number of molecules of either of these is significantly greater than the number of molecules of fHbp then the percentage of bacterial surface fHbp molecules binding CFH at equilibrium will be determined by: the relative amounts of both present in the individual; and the affinities of that fHbp sequence for CFH vs CFHR3 ( essentially a model of competitive antagonism , see equation below ) . Natural fHbp sequences are relatively constant in their KD for CFH ( KD ∼3 nM [Johnson et al . , 2012] and this work ) , therefore we modelled the effect of the KD for CFHR3 varying between 3 nM and 300 nM against this constant KD for CFH under two conditions . In red we assume that both CFH and CFHR3 are present in serum at 1 . 5 μM , whilst in blue we model the effect if the concentration of CFH is twice that of CFHR3 ( i . e . , 3 μM vs 1 . 5 μM ) . This range of relative concentrations covers those generally reported for CFH and CFHR3 ( except in the case of ΔCFHR3/1 individuals where the concentration of CFHR3 will be zero and 100% of the fHbp will bind CFH ) . The grey shaded areas indicate the regions of the x axis into which the families of natural variant fHbp sequences studied in Figure 3 fall . The equation used to generate these curves is:Occupancy CFH=[CFH]KD ( CFH ) [CFH]KD ( CFH ) +[CFHR3]KD ( CFHR3 ) +1∗100 . DOI: http://dx . doi . org/10 . 7554/eLife . 04008 . 011 To define the basis for the effect of CFHR3 on bacterial survival , we next examined the impact of CFHR3 on levels of CFH bound to the bacterial surface . N . meningitidis incubated in the ΔCFHR3/1 sera acquire high levels of CFH on the bacterial surface ( Figure 6A , B , C ) . However , addition of recombinant CFHR3 reduces the amount of bound CFH in a dose-dependent manner , demonstrating antagonism between CFH and CFHR3 for binding to the bacterial surface . When the amount of recombinant CFHR3 is added at approximately equimolar levels to those in NHS , the level of CFH bound on the bacterial surface is only half that seen in the CFHR3-deficient sera ( 1–1 . 6 μM [Fritsche et al . , 2010] ) . 10 . 7554/eLife . 04008 . 012Figure 6 . Susceptibility to N . meningitidis will be altered by the circulating levels of CFH and CFHR3 that compete for binding to fHbp . ( A–C ) Flow cytometry demonstrates that addition of increasing amounts of CFHR3 ( from 0 μM [light blue] to 1 μM [black] ) to sera from a ΔCFHR3–1 individual leads to decreasing levels of CFH on the bacterial surface ( A and C ) and increasing levels of CFHR3 ( B and C ) . ( D ) Model of how differing levels of CFHR3 and CFH in an individual combine to alter the amount of CFH hijacked to the bacterial surface , hence the level of complement-mediated bacterial killing and an individual's susceptibility to meningococcal disease . DOI: http://dx . doi . org/10 . 7554/eLife . 04008 . 012 In summary , our work demonstrates that susceptibility to an important infectious disease is governed by the competition between two complement factors with opposing effects ( CFH and CFHR3 ) for overlapping binding sites on a single pathogen molecule . Our data predict that the risk of developing meningococcal disease will be governed by the relative abundance of CFH and CFHR3 in individuals and the affinity of these molecules for the specific fHbp on the bacterial surface ( Figure 6D ) . It is notable that the ΔCFHR3/1 allele has a high prevalence in sub-Saharan Africa ( Holmes et al . , 2013 ) where there are high rates of epidemic disease . Our mechanism is consistent with this association , since this allele would enhance the interaction between CFH and fHbp and favour pathogen survival . This is particularly true as the ΔCFHR3/1 deletion is associated with increased plasma CFH levels , most likely due to the loss of a regulatory locus ( Ansari et al . , 2013; Zhu et al . , 2014 ) . However , host polymorphisms in CFHR3 might have less impact on strains of bacteria that express fHbp that more selectively bind CFH . Therefore , susceptibility to meningococcal disease will be influenced by polymorphisms and copy number variation affecting two host complement factors and by the relative affinity of fHbp on an infecting strain for CFH and CFHR3 . This illustrates why genetic association studies in infectious diseases are likely to be technically challenging since these will be influenced not only by sequence and structural variation within the host but also by genetic variation in pathogen .
Sequences encoding CFHR145 , CFHR312 , CFHR345 and CFHR445 were amplified using primer pairs CFHR1-4-Forward ( 5′-GAAGGAGATATACCATGACGGGAAAATGTGGGCC-3′ ) /CFHR1-5-Reverse ( 5′-GGCCCACATTTTCCCGTCATGGTATATCTCCTTC-3′ ) , CFHR3-1-Forward ( 5′-GATATACATATGAAACCTTGTGATTTTCCAGACATTAAAC-3′ ) /CFHR3-2-Reverse ( 5′-GGATCCTCGAGCTAACGGATGCATCTGGGAGTAG-3′ ) , CFHR3-4-Forward ( 5′-GGCAGCCATATGTCAGAAAAGTGTGGGCCTCCTC-3′ ) /CFHR3-5-Reverse ( 5′- CGGATCCTCGAGTTATTCGCATCTGGGGTATTCCACTATC-3′ ) and CFHR3-4-Forward/CFHR4-5-Reverse ( 5′- CGGATCCTCGAGTTATTCGCATCTGGGGTATTCCACTATG-3′ ) , respectively . The sequence encoding CFHR589 was amplified using primer pair CFHR5-8-Forward ( 5′-CCACATATGGCATATTGTGGGCCCCCTCCATC-3′ ) /CFHR5-9-Reverse ( 5′-GGATCCTCGAGTCATTCACATATAGGATATTCAAATTTC-3′ ) . Inserts were cloned between the NdeI and XhoI sites of a modified version of vector pET-15b ( Novagen , Merck Millipore , Billerica , MA ) , which has the NcoI site replaced with an NdeI site . A vector encoding FH19–20 was generated from that encoding CFHR145 using site-directed mutagenesis with primers L183S-Forward ( 5′-GGACAGCCAAACAGAAGCTTTATTCGAGAACAGGTGAA-3′ ) /L183S-Reverse ( 5′-TTCACCTGTTCTCGAATAAAGCTTCTGTTTGGCTGTCC-3′ ) and A189V-Forward ( 5′-TTTGAGAACAGGTGAATCAGTTGAATTTGTGTGTAAACGGG-3′ ) /A189V-Reverse ( 5′- CCCGTTTACACACAAATTCAACTGATTCACCTGTTCTCAAA-3′ ) . The gene encoding full-length CFHR3 was amplified using primers encoding a histidine ( His ) tag ( CFHR3-Forward 5′-ATTCGCGGCCGCCCCACCATGTTGTTACTAATCAAT-3′ and CFHR3-Reverse 5′-CACCATCACCATCACCATTAGCTCGAGGAA-3′ ) and cloned into a modified version of the pCAGGS plasmid ( Shimizu et al . , 1999 ) bearing the CMV-EI enhancer , the chicken β-actin promoter , and intron 1 of the simian virus 40 poly ( A ) signal . The gene encoding full-length CFHR4 was amplified using primer pair CFHR4-Forward ( 5′- GGACTGTCTAGAGACTCAGATCCCCATCCGCTCAAGCAGGCCACCATGTTGTTACTAATCAATGTCATTCTGACC-3′ ) /CFHR4-Reverse ( 5′ GTGTCGGGTCGACCCTTCGCATCTGGGGTATTCCAC-3′ ) prior to ligation into vector pEF-BOS ( Mizushima and Nagata , 1990 ) between the XbaI and SalI sites giving a C-terminal hexahistidine tag . Recombinant His-tagged fHbp was prepared from a selection of natural variant fHbps and alanine substitution mutants of fHbp V1 , V2 , and V3 as described previously ( Johnson et al . , 2012 ) . All three variant types were represented within 27 natural fHbp variant sequences , numbered according to an allele ID code as detailed on the Neisseria Multi Locus Sequence Typing website ( http://pubmlst . org/neisseria/ ) developed by Keith Jolley and sited at the University of Oxford ( Jolley and Maiden , 2010 ) . CFH67 was prepared as previously described ( Prosser et al . , 2007 ) . CFH19–20 , CFHR145 , CFHR345 , CFHR445 , and CFHR589 were expressed in E . coli and refolded from inclusion bodies prior to size exclusion chromatography ( Superdex S-75 , GE Healthcare , Little Chalfont , UK , buffer 50 mM Tris pH 7 . 5 , 150 mM NaCl ) using the same method previously described for CFHR245 ( Goicoechea de Jorge et al . , 2013 ) . CFHR112 was expressed in Kluyveromyces lactis as previously described ( Goicoechea de Jorge et al . , 2013 ) . CFHR312 was prepared in an identical manner to CFH67 with substitution of the final size exclusion chromatography step for purification using heparin affinity ( Heparin FF column , GE Healthcare , buffer A: 50 mM Tris pH 7 . 5 , 10 mM NaCl , buffer B: 50 mM Tris pH7 . 5 , 1 M NaCl , elution gradient: 50% over 15 CV ) . Recombinant His-tagged CFHR3 was expressed in HEK293 cells using Lipofectmine 2000 ( Invitrogen ) as per manufacturer's instructions . Recombinant His-tagged CFHR3 supernatant was purified by a single affinity chromatography step using a His-Trap HP ( GE Healthcare ) as per manufacturer's instructions . The bound protein was eluted with 500 mM Imidazole and then dialysed against 10 mM sodium phosphate pH7 . 8 . CFHR4 was expressed transiently in cell line HEK293S GntI−/− from vector pEF-BOS . Transfection was performed using 293Fectin ( Life Technologies , Carlsbad , CA ) as per the manufacturers instructions . Cells were cultured in suspension using Freestyle 293 Expression medium ( Life Technologies ) at 37°C with 10% CO2 for 96 hr prior to harvesting the media . CFHR4 was purified from the media using a Ni-NTA column ( Qiagen ) as per the manufacturer’s instructions . Purified protein was dialysed into phosphate buffered saline ( PBS ) prior to use . All surface plasmon resonance measurements were made at 25°C using either a Biacore 3000 ( GE Healthcare ) or ProteOn XPR36 ( BioRad , Hercules , CA ) instrument with buffer HBS-EP ( 10 mM Hepes pH7 . 4 , 150 mM NaCl , 3 mM EDTA , 0 . 05% surfactant-P20 ) . Proteins were immobilised via primary amine coupling with a mock activated–deactivated reference channel included on each . Measurement of the affinity between CFHR312 and fHbp V3 was performed by immobilising approximately 1200 RU fHbp on the surface of a CM5 sensor chip ( GE Healthcare ) . Dilution series of CFHR312 ranging from 12 . 5 nM–0 . 2 nM were flowed over the surface at 40 µl/min using KINJECT with a contact time of 300 s and dissociation time of 1200 s . The surface was regenerated with 10 mM glycine pH 3 . 0 between each injection . Curves were mock subtracted and fit with a 1:1 Langmuir model using BiaEvaluation ( GE Healthcare ) . Analysis of binding of CFHRs and CFHR fragments to fHbp V3 was performed using a CM5 chip with approximately 1700 RU fHbp immobilised . 100 µl of each analyte at a minimum concentration of 200 nM was flowed over the surface at 20 µl/min using the KINJECT command with a dissociation time of 400 s . The surface was regenerated using 10 mM glycine pH 3 . 0 between injections . Curves were mock subtracted using BiaEvaluation . Measurement of the large panel of fHbp mutants was performed by immobilization of the proteins on ProteOn GLM sensor chips ( BioRad ) . Increasing concentrations of CFHR312 or fH67 were injected over the flow channels at 60 µl/min for 240 s and allowed to dissociate for 1200 s . The surface was regenerated with 10 mM glycine pH 3 . 0 between each injection . Curves were mock subtracted and fit with a 1:1 Langmuir model using ProteOn manager software ( BioRad ) . fHbp V3 was labelled using the RED-NHS labelling kit ( NanoTemper Technologies , Munich , Germany ) as per the manufacturer’s instructions . Dilution series of CFHR312 in the range of 740 nM–23 pM were set up using HBS-EP . Each sample also contained 4 nM labelled fHbp . Thermophoresis was measured using a Monolith NT . 115 instrument ( NanoTemper Technologies ) at 22°C utilising hydrophilic treated capillaries ( NanoTemper Technologies ) , 80% LED power and 60% LED power . All data were measured in triplicate across three independent dilution series . Data were analysed using the signal from Thermophoresis + T-Jump ( NT Analysis software version 1 . 5 . 41 , NanoTemper Technologies ) . The effect of CFHR3 , CFHR312 , CFHR345 , CFHR4 , and CFHR445 upon AP activation was assessed using haemolytic assays that were performed in an identical manner to that described in Goicoechea de Jorge et al . ( 2013 ) . All measurements were performed in triplicate and are presented as haemolysis relative to the level of lysis without addition of CFHR proteins ( 0% ) and 100% lysis by H2O . Haemolytic assays with CFH-deficient serum ( CompTech , Tyler , TX ) were performed with 5% serum . N . meningitidis was grown on brain heart infusion ( BHI , Oxoid ) agar supplemented with 5% Levinthal's base ( 500 ml defibrinated horse blood autoclaved with 1 l BHI ) overnight at 37°C in the presence of 5% CO2 ( Lucidarme et al . , 2009; Jongerius et al . , 2013 ) . M1239Δfhbp was constructed as previously ( Jongerius et al . , 2013 ) and isogenic strains were constructed by complementing the strains with fHbp V1 . P1 , V1 . P14 , V2 . P22 , or V3 . P28 amplified using primers listed in Table 1 . Polymerase chain reaction products were ligated into pGCC4 ( Mehr and Seifert , 1998 ) . Transformation of N . meningitidis strain M1239Δfhbp was performed as described previously ( Exley et al . , 2005 ) . 10 . 7554/eLife . 04008 . 013Table 1 . Primers used to generate bacterial strainsDOI: http://dx . doi . org/10 . 7554/eLife . 04008 . 013Primer nameStrain amplifiedRefPrimer Sequence ( restriction site underlined ) pgcc4V1 . 1 FH44/76 ( Jongerius et al . , 2013 ) CGGTTAATTAAGGAGTAATTTTTGTGAATCGAACTGCCTTCTGCTpgcc4V1 . 1 RH44/76 ( Jongerius et al . , 2013 ) CGGTTAATTAATTATTGCTTGGCGGCpgcc4V1 . 14 FNZ98/254CGGTTAATTAAGGAGTAATTTTTGTGAACCGAACTGCCpgcc4V1 . 14 RNZ98/254CGGTTAATTAATTATTGCTTGGCGGCAAGACpgcc4V2 . 22FFAM18CGGTTAATTAAGGAGTAATTTTTGTGAACCGAACTGCCTTCTGCTpgcc4V2 . 22RFAM18CGGTTAATTAACTACTGTTTGCCGGCGATGCpgcc4V3 . 28 FM1239CGGTTAATTAAGGAGTAATTTTTGTGAATCGAACTGCCTTCTGCTpgcc4V3 . 28 RM1239CGGTTAATTAACTACTGTTTGCCGGCGATGC NHS were obtained by collecting whole venous blood and allowing it to coagulate at room temperature for 60 min before centrifuging at 3000×g for 20 min at 4°C . Sera were heat inactivated at 56°C for 30 min prior to use . N . meningitidis was grown overnight on BHI agar , re-suspended in PBS then diluted to 1 × 105 CFU/ml in DMEM ( Sigma ) . A total of 1 × 104 CFU of bacteria were pre-incubated with 1 µM CFHR3 , 1 µM CFHR3 domains or PBS for 10 min prior to incubation with 5% human sera or 5% heat inactivated sera for 30 min at 37°C in the presence of CO2 . Bacterial survival was determined by plating onto BHI agar . Relative survival was calculated from samples containing no CFHR3 and incubated with heat-inactivated sera and expressed as percentage survival . For the AP assays , bacteria were grown on BHI agar supplemented with 1 mM IPTG . Other complement pathways were inhibited by the addition of 5 mM MgCl2/10 mM EGTA , and bacteria were incubated in 25% sera as above . Statistical significance was tested using the one-way ANOVA multiple comparisons as implemented in GraphPad Prism v . 6 . 0 ( GraphPad Software Inc . ) to compare means ±S . E . M . using a p < 0 . 01 cutoff for significance . Recombinant CFHR345 conjugated to KLH ( Imject , Pierce ) was used as an antigen . Female BALB/c mice were immunised with 100 μg of recombinant protein emulsified in TitreMax Gold ( Sigma , UK ) subcutaneously on day 1 , then with 100 μg of CFHR345 in PBS intraperitoneally ( IP ) on days 21 and 42 . A booster of 200 μg of recombinant protein in PBS was given IP 3 days prior to cell fusion on day 63 . NS0 myeloma cells were grown in RPMI 1640 ( Lonza , UK ) supplemented with 10% foetal bovine serum ( FBS , Sigma , UK ) , 2 mM L-glutamine , penicillin ( 100 U/ml ) , and streptomycin ( 100 μg/ml ) ( Galfre and Milstein , 1981 ) . Splenectomy was performed on day 66 , and splenocytes fused to NS0 myeloma in polyethylene glycol ( PEG 1500 , Roche , UK ) ( Mason et al . , 1983 ) . Fused cells were then plated into 24-well plates in RPMI 1640 containing 2 mM L-glutamine , penicillin ( 100 U/ml ) , streptomycin ( 100 μg/ml ) , and 1% Ultroser G ( Pal France ) . After 24 hr , 2% hypoxanthine adminopterin thymidine ( HAT , Life technologies ) was added to the wells . Hybridoma clone supernatants were screened by ELISA against recombinant CFHR345 . Positive clones were re-plated and grown for a further 5 days before screening again by ELISA against the CFHR345 and all five CFHR proteins . Positive hybridoma cells were sub-cultured into 96-well plates at a concentration of 1 cell per well in RPMI 1640 containing 10% FBS , 2 mM L-glutamine , penicillin ( 100 U/ml ) , streptomycin ( 100 μg/ml ) , Ultroser G ( 1% ) , and BM Condimed H1 ( 1% , Roche , UK ) . After 5 days , wells containing single colonies were screened as before . Bacteria ( 1 × 109 CFU/ml ) were fixed in 1 ml of 3% formaldehyde for 2 hr then washed with PBS . To evaluate CFHR3 binding , 5 × 107 CFU/ml were re-suspended in 10 μl of NHS or 10 µl of sera from an individual with ΔCFHR3/1 for 30 min at room temperature . After two washes in 0 . 05% BSA/PBS , binding was detected following incubation with the anti-CFHR3 mAb ( HSL1 this study ) for 30 min at 4°C in 50 μl of PBS . Cells were washed twice in 0 . 05% BSA , then resuspended in 50 μl of goat anti-mouse IgG-Alexa Fluor 647 conjugate ( 1 in 1000 dilution in PBS; Molecular Probes , Life Technologies ) and incubated for 30 min at 4°C . Samples were run on a FACSCalibur ( BD Biosciences ) , and at least 104 events recorded before results were analysed by calculating the geometric mean FL-4 in FlowJo vX software ( Tree Star ) . To evaluate the influence of CFHR3 on CFH binding to bacteria , 5 × 107 CFU/ml of N . meningitidis M1239 were incubated in 10 µl of ΔCFHR3/1 sera and twofold dilutions of recombinant full-length CFHR3 ( from 1 µM ) for 30 min at room temperature . CFHR3 binding was detected with the anti-CFHR3 mAb ( HSL1 this study ) or an anti-human CFH mAb ( MRC OX24 , 0 . 1 μg/ml ) ( Sim et al . , 1983 ) , followed by incubating with goat anti-mouse IgG-Alexa Fluor 647 conjugate ( 1 in 1000 dilution in PBS; Molecular Probes , Life Technologies ) and analysed by flow cytometry . At least 104 events were recorded before results were analysed by calculating the geometric mean FL-4 in FlowJo vX software ( Tree Star ) . Maximum fluorescence for CFHR3 was normalised to the addition of 1 µM exogenous CFHR3 , whereas the maximum fluorescence for CFH was normalised without the addition of exogenous CFHR3 . Statistical significance was tested using the one-way ANOVA multiple comparisons as implemented in GraphPad Prism v . 6 . 0 ( GraphPad Software Inc . ) to compare means ±S . E . M . using a p < 0 . 01 cutoff for significance . | Meningitis is a potentially life-threatening condition whereby the membranes that cover and protect the brain and spinal cord become inflamed . Often meningitis is caused by a viral or bacterial infection—such as infection by a bacterium called Neisseria meningitidis , also known as meningococcus . However , not everyone that comes into contact with this bacterium will develop meningitis; 40% of the population is thought to carry N . meningitidis at the back of the nasal cavity and yet show no signs of the disease . It remains unclear why some people exposed to N . meningitidis develop meningitis while others do not; however recent research revealed that part of the immune system called the complement system plays a role in susceptibility to meningitis . The complement system is a collection of small proteins that work together to support the actions of the cells of the immune system . When activated , complement proteins trigger a cascade of events that helps to destroy the pathogen . Several mechanisms exist to keep the complement proteins in check—for example , a protein called complement factor H ( or CFH ) protects host cells from being attacked by other complement proteins . N . meningitidis can undermine the complement system by expressing a protein that binds to CFH and firmly fixes CFH to its cell surface . While the CFH-binding protein helps explain why some people are unable to mount the appropriate immune response to infection by N . meningitidis , it does not explain why some carriers of the pathogen do not develop meningitis . Now , Caesar et al . have examined a protein called CFH related-3 ( or CFHR3 ) , and discovered that CFHR3 competes with CFH for the binding protein on N . meningitidis . CFHR3 is structurally similar to CFH , but it is unable to regulate or silence the complement system . Caesar et al . explain that susceptibility to meningococcal disease is determined by how much CFH and how much CFHR3 each individual has , and that those with less CFHR3 will be more susceptible to N . meningitidis . An individual's genes will affect how much CFH and CFHR3 they have , while the genes of the bacterium can influence how strongly the CFH binding protein binds to either of these human proteins . Caesar et al . suggest that these two factors determine whether or not an individual will develop meningitis or simply carry the bacterium without any ill effects . Caesar et al . 's findings highlight the different ways that people's genes can determine how they respond to an invading pathogen . The findings also suggest that it is important to consider variation in the levels of these complement proteins across a population when planning immunisation schedules . | [
"Abstract",
"Introduction",
"Results",
"and",
"discussion",
"Materials",
"and",
"methods"
] | [
"microbiology",
"and",
"infectious",
"disease",
"immunology",
"and",
"inflammation"
] | 2014 | Competition between antagonistic complement factors for a single protein on N. meningitidis rules disease susceptibility |
Mucin 5AC ( MUC5AC ) is secreted by goblet cells of the respiratory tract and , surprisingly , also expressed de novo in mucus secreting cancer lines . siRNA-mediated knockdown of 7343 human gene products in a human colonic cancer goblet cell line ( HT29-18N2 ) revealed new proteins , including a Ca2+-activated channel TRPM5 , for MUC5AC secretion . TRPM5 was required for PMA and ATP-induced secretion of MUC5AC from the post-Golgi secretory granules . Stable knockdown of TRPM5 reduced a TRPM5-like current and ATP-mediated Ca2+ signal . ATP-induced MUC5AC secretion depended strongly on Ca2+ influx , which was markedly reduced in TRPM5 knockdown cells . The difference in ATP-induced Ca2+ entry between control and TRPM5 knockdown cells was abrogated in the absence of extracellular Ca2+ and by inhibition of the Na+/Ca2+ exchanger ( NCX ) . Accordingly , MUC5AC secretion was reduced by inhibition of NCX . Thus TRPM5 activation by ATP couples TRPM5-mediated Na+ entry to promote Ca2+ uptake via an NCX to trigger MUC5AC secretion .
Mucus is secreted by specialized cells that line the respiratory and digestive tract to protect against pathogens and other forms of cellular abuse . The secretion of mucus is therefore essential for the normal physiology of the wet mucosal epithelium ( Rubin , 2010 ) . The secretory or gel-forming mucin , Mucin 5AC ( MUC5AC ) is one of the major components of the mucus in the airways , and hyper- or hyposecretion of this component is a hallmark of a number of chronic obstructive pulmonary diseases ( COPD ) ( Rose and Voynow , 2006 ) . MUC5AC is also expressed at low levels in the gastrointestinal tract and , surprisingly , expressed de novo , and upregulated in colonic mucus from cancer and ulcerative colitis patients ( Bartman et al . , 1999; Kocer et al . , 2002; Byrd and Bresalier , 2004; Forgue-Lafitte et al . , 2007; Bu et al . , 2010 ) . MUC5AC is also expressed in response to parasitic infection , which is probably its additional physiological role ( Hasnain et al . , 2011 ) . The gel-forming mucins are giant filamentous glycoproteins that are synthesized in the Endoplasmic Reticulum ( ER ) and exported to the Golgi complex where they undergo extensive modification in their oligosaccharide chains . The apparent molecular weight of the gel-forming MUC5AC increases from 500 kD of monomeric unglycosylated ER form ( van Klinken et al . , 1998 ) to 2 . 2 MD ( Thornton et al . , 1996 ) by glycosylation and oligomerization during its transit through the Golgi apparatus to a secreted form that reaches up to 40 MD in apparent molecular weight ( Sheehan et al . , 2000 ) . The heavily glycosylated mucins are sorted , condensed and packed into mucin-secreting granules ( MSG ) . The MSG fuse with the plasma membrane , in a signal-dependent manner , and the condensed mucins expand their volume up to 1000-fold upon secretion ( Verdugo , 1993 ) . The signaling events that lead to mucin secretion in the airways involve mainly , but not exclusively , P2Y purinergic and muscarinic receptor activation by ATP and acetylcholine , respectively . The subsequent generation of diacylglycerol ( DAG ) and inositol 1 , 4 , 5-triphosphate ( IP3 ) activate protein kinase-C ( PKC ) and cause the release of Ca2+ from the ER to promote mucus secretion ( Bou-Hanna et al . , 1994; Abdullah et al . , 1996 , 1997; Bertrand et al . , 2004; Ehre et al . , 2007 ) . The progress to date on the components involved in the trafficking of mucins has recently been thoroughly reviewed ( Davis and Dickey , 2008 ) . Basically , mucins are packed ( somehow ) into MSG at the trans-Golgi network ( TGN ) . MSGs undergo fusion to produce mature condensed granules that are stored in the cytoplasm . The cortical actin acts as a barrier that is reorganized in a Ca2+-dependent reaction through the input of PKCε-dependent phosphorylation of MARCKS ( Wollman and Meyer , 2012 ) . The passage of mature MSGs through the actin network also requires Myo II and V . The proteins involved in the docking , priming and fusion of the MSGs are reported to include: Rab3d , Rab27 , Hsc70 , cysteine string protein , Synaptotagmin 2 , Munc13-2 , Munc13-4 , Munc18b , Syntaxin 2 , 3 , 11 , and VAMP8 . However , it is not known how many of these proteins are directly involved in mucin secretion and for some , such as the MARCKS protein , the mechanism is controversial ( Stumpo et al . , 1995; Arbuzova et al . , 2002 ) . The exact myosin involved in the trafficking of MSGs across the actin barrier remains unclear ( Rose et al . , 2003; Neco et al . , 2004; Jerdeva et al . , 2005 ) . More importantly , the mechanism of Ca2+-dependent signaling and the components involved in this signaling cascade are not fully characterized . To date , transport studies have been based on truncated GFP-mucin variants ( Perez-Vilar et al . , 2005 ) and time-consuming techniques such as combinations of density gradient centrifugation and agarose gel electrophoresis ( Sheehan et al . , 2004 ) . It has therefore been difficult to identify new components involved in mucin secretion and to decipher their mechanism of action . As stated above , human cancer cells and cells from patients with ulcerative colitis express and secrete MUC5AC . These cells and cell lines therefore provide a convenient means to address the mechanism MUC5AC secretion . We have established a quantitative assay to measure the secretion of MUC5AC from a human goblet cell line . The procedure was used to screen 7343 human gene products and we describe here the identification and involvement of transient receptor potential melastatin 5 ( TRPM5 ) channel in MUC5AC secretion .
The human colonic adenocarcinoma cells HT29-18N2 ( N2 ) differentiate to goblet cells upon starvation in protein-free medium ( Phillips et al . , 1995 ) , which increases the production of MUC5AC . Immunofluorescence analysis of accumulated MUC5AC in secretory granules ( Figure 1A ) shows the differences between starved and nonstarved cells . The increase in protein production of MUC5AC after starvation was confirmed by dot-blotting cell lysates of nonstarved and starved N2 cells ( Figure 1B ) . Quantification of the dot blot revealed a 45-fold increase of MUC5AC protein levels in starved N2 cells compared to nonstarved N2 cells . Our findings with the dot-blot procedure confirm the lack of MUC5AC production in Hela cells ( Figure 1B , C ) . MUC5AC mRNA analysis by quantitative real-time PCR also confirmed increased MUC5AC mRNA levels in starved cells ( Figure 1D ) . The fusion of MUC5AC-containing granules with the plasma membrane requires an external signal , which results in the production of DAG and the release of Ca2+ from internal stores . To induce mucin secretion from the starved N2 cells , we used the DAG mimic , phorbol-12-myristate-13-acetate ( PMA ) . Starved goblet cells were treated for 2 hr with 2 µM PMA to induce MUC5AC secretion ( Figure 1E ) . The extracellular MUC5AC expands and coats the cell surface ( Figure 1E ) . We took advantage of the stickiness of the mucin film to quantitate secreted MUC5AC . After 2 hr incubation with PMA , the cells were fixed with paraformaldehyde followed by incubation with an anti-MUC5AC antibody and a secondary fluorescent-labeled antibody to visualize secreted mucin ( Figure 1E ) . To detect the intracellular pool of MUC5AC after PMA-induced release , the cells were washed extensively to remove secreted MUC5AC and then fixed with paraformaldehyde , permeabilized and processed for immunofluorescence microscopy with an anti-MUC5AC antibody as described above ( Figure 1E ) . 10 . 7554/eLife . 00658 . 003Figure 1 . Mucin synthesis and secretion from goblet cells . ( A ) Nonstarved and starved N2 cells were fixed and analyzed by immunofluorescence microscopy with an anti-MUC5AC antibody ( green ) . The nuclear DNA was stained with DAPI ( blue ) to localize the position of the nucleus . ( B ) Dot blot of total lysates of nonstarved , starved N2 and HeLa cells were probed with anti-MUC5AC and anti-actin antibody . ( C ) The dot blots in ( B ) were quantified and normalized to actin levels . The y-axis represents relative values with respect to the values of nonstarved N2 cells . Average values ± SEM are plotted as bar graphs ( N = 3 ) . ( D ) Nonstarved and 5 days starved N2 cells were lysed and total RNA was extracted for quantitative real-time PCR analysis . The values for MUC5AC mRNA levels were normalized to the values of the housekeeping gene HPRT1 . The y-axis represents relative values with respect to nonstarved N2 cells . Average values ± SEM are plotted as bar graphs ( N = 4 ) . ( E ) Starved N2 cells were treated for 2 hr with 2 μM PMA . To detect the remaining intracellular mucin after PMA release , the secreted mucin was removed by DTT and trypsin treatment of the goblet cells prior to fixation ( experimental procedures ) . After fixation , cells were permeabilized and examined by immunofluorescence microscopy with DAPI and an anti-MUC5AC antibody . Secreted MUC5AC was detected by fixing the secreted mucus directly on the cells after PMA treatment , followed by immunofluorescence microscopy using an anti-MUC5AC specific antibody . DOI: http://dx . doi . org/10 . 7554/eLife . 00658 . 003 To quantitate MUC5AC secretion , starved goblet cells were treated for 2 hr with 2 µM PMA , followed by fixation and incubation with an anti-MUC5AC antibody . The secreted MUC5AC was monitored by chemiluminescence using secondary antibodies conjugated to HRP ( Figure 2A , B ) . The time course for PMA induced MUC5AC secretion shows a significant increase at 15 min and maximal MUC5AC secretion is observed at 2 hr post incubation with 2 µM PMA ( Figure 2—figure supplement 1 ) . 10 . 7554/eLife . 00658 . 004Figure 2 . Mucin secretion assay . ( A ) Illustration of the mucin secretion assay . Starved N2 cells are treated with PMA . Secreted MUC5AC is fixed on the cell surface and labeled with anti-MUC5AC antibodies followed by quantitative detection using HRP-conjugated secondary antibody . ( B ) Starved N2 cells were treated for 2 hr ± 2 μM PMA , fixed with formaldehyde and the amount of secreted MUC5AC bound to the cell surface was detected with anti-MUC5AC antibody and measured by chemiluminescence . The values were normalized to values obtained for—PMA treatment . Average values ± SEM are plotted as bar graphs ( N = 10 ) . ( C ) Starved N2 cells were pretreated for 30 min with BAPTA-AM , Latrunculin A or Jasplakinolide and incubated at 37°C . After 30 min , 2 μM PMA was added containing the respective drugs ( BAPTA-AM , Latrunculin A or Jasplakinolide ) and incubated for 2 hr at 37°C . Cells were fixed and secreted MUC5AC was detected by chemiluminescence . The values were normalized to values obtained for −PMA treatment . Average values ± SEM are plotted as bar graphs ( N = 10 ) . Compared datasets were considered as statistically significant when p<0 . 01 ( ** ) and p<0 . 0001 ( **** ) . ( D ) Starved N2 cells were preincubated for 15 min with 2 μg/ml ( +BFA ) and incubated at 37°C . After 30 min , increasing concentrations of PMA were added in the presence or absence of 2 μg/ml BFA ( +/− BFA ) and incubated for 2 hr at 37°C . Cells were fixed and secreted MUC5AC was detected by chemiluminescence . The values were normalized to values obtained for −PMA treatment . Average values ± SEM are plotted as bar graphs ( N = 5 ) . ( E ) Starved N2 cells were incubated for 45 min with or without 2 μg/ml BFA ( +/− BFA ) at 37°C . After 45 min cells were fixed , permeabilized and examined by immunofluorescence microscopy with an anti-MUC5AC antibody ( green ) , an anti-GRASP65 antibody ( red ) and DAPI ( blue ) . DOI: http://dx . doi . org/10 . 7554/eLife . 00658 . 00410 . 7554/eLife . 00658 . 005Figure 2—figure supplement 1 . Time course for MUC5AC secretion . Starved N2 cells were treated for the indicated times with 2 μM PMA , fixed with formaldehyde and the amount of secreted MUC5AC bound to the cell surface was detected with anti-MUC5AC antibody and measured by chemiluminescence . The values were normalized to values obtained for −PMA treatment . Average values ± SEM are plotted as bar graphs ( N = 6 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 00658 . 005 Secretion of mucins requires a dynamic actin cytoskeleton and Ca2+ ( Abdullah et al . , 1997; Ehre et al . , 2005; Wollman and Meyer , 2012 ) . We tested the effect of perturbing actin cytoskeleton and Ca2+ levels on the PMA-dependent secretion of MUC5AC from starved N2 cells . Starved N2 cells were treated with the drugs that affect actin filaments: Latrunculin A and Jasplakinolide . The cells were also treated with the membrane-permeant Ca2+ chelator BAPTA-AM . The extracellular levels of MUC5AC were measured with the chemiluminescence-based assay . Depolymerization of actin filaments by Latrunculin A had no effect on PMA-stimulated MUC5AC secretion , while BAPTA-AM and the actin-stabilizing agent Jasplakinolide severely affected MUC5AC secretion ( Figure 2C ) . The inhibitory effect of hyperstabilized actin filaments ( by Jasplakinolide treatment ) on MUC5AC secretion reveals that actin filaments likely act as a barrier to prevent premature fusion of MUC5AC-containing granules with the cell surface . Inhibition of MUC5AC secretion by BAPTA-AM treatment confirms the known requirement of Ca2+ in the events leading to mucin secretion . Befreldin A ( BFA ) is known to inhibit cargo export from the ER and causes Golgi membranes to fuse with the ER ( Lippincott-Schwartz et al . , 1989 ) . To test whether BFA affected the formation of secretory granules , starved N2 cells were incubated with or without 2 µg/ml BFA . After 45 min cells were fixed and examined by immunofluorescence analysis with an anti-MUC5AC antibody and an antibody to the Golgi membrane specific GRASP65 protein ( Figure 2E ) . The dispersal of GRASP65 with BFA treatment shows that our experimental conditions are effective in disrupting the Golgi apparatus . However , MUC5AC staining was unperturbed by BFA treatment ( Figure 2E ) . We then tested the effect of BFA on the constitutive secretion of newly synthesized proteins . Starved N2 cells were labeled with 35S-methionine and then chased in cold methionine-containing medium in the presence of BFA . Analysis of the medium revealed that BFA severely inhibited the secretion of newly synthesized proteins from the starved N2 cells ( Figure 3—figure supplement 1 ) . To test whether BFA affected the regulated secretion of the secretory granules , starved N2 cells were pretreated with 2 µg/ml BFA for 15 min and then treated with 2 µM PMA for 2 hr in the presence of BFA . MUC5AC was then measured in the extracellular medium by chemiluminescence ( Figure 2D ) . The results reveal that BFA treatment does not affect PMA-dependent MUC5AC secretion under the experimental conditions . Therefore , in our assay , we only measure the secretion of MUC5AC contained in the post-Golgi secretory carriers . This measurement is independent of MUC5AC synthesis , export from the ER to the late Golgi , and its sorting and packing into the secretory granules . N2 cells were starved for 6 days and transfected with siRNA oligos against each of the selected 7343 genes . A pool of four different siRNAs targeting the same component was used and every component was analyzed in triplicate . 3 days after transfection , the cells were treated with 2 μM PMA for 2 hr and analyzed by chemiluminescence-based detection of secreted MUC5AC ( Figures 2A and 3A ) . For the data analysis we assumed that the majority of the siRNAs will not affect the secretion of MUC5AC . Data points were normalized by the B-score and the triplicates were ranked according to the Ranking Product method ( Breitling et al . , 2004; Supplementary file 1 ) . The hits were plotted as median of the B-score and positive hits were selected above and below a B-score of ±1 . 5 . siRNAs that scored above 1 . 5 B-score were considered as hypersecretory phenotype and those below 1 . 5 B-score were considered as inhibitors of MUC5AC secretion ( hyposecretory phenotype ) ( Figure 3B ) . From this analysis we selected 413 components that upon knockdown resulted in hyposecretion and 534 that exhibited a hypersecretion of MUC5AC ( Figure 3C ) . The hits were analyzed by Ingenuity Pathway Analysis and categorized according to their intracellular localization and type . For further analysis we removed 678 proteins from this pool that included secreted proteins , nuclear proteins , proteins affecting protein modification , and those involved in basic metabolism . This narrowed the hits to 114 with hyposecretion and 155 with hypersecretion phenotype ( Supplementary file 1 ) . 10 . 7554/eLife . 00658 . 006Figure 3 . Identification of proteins required for MUC5AC secretion . ( A ) Screening procedure . N2 cells were serum starved for 6 days . The cells were then seeded into the wells of 96-well plates and transfected in triplicates on three sets of plates with siRNAs directed against 7343 components . 3 days after transfection , the cells were treated for 2 hr with 2 μM PMA and analyzed by an automated chemiluminescence assay for the detection of secreted MUC5AC . ( B ) The B-score of each gene product tested was calculated using the triplicate measurements of the chemiluminescence values . All siRNAs that altered secretion with a B-score less than −1 . 5 ( hyposecretory ) and higher than 1 . 5 ( hypersecretory ) were selected as positive hits for further analysis . −PMA ( lower red line ) and +PMA ( upper red line ) indicate mock treated controls without PMA ( −PMA ) and with 2 μM PMA ( +PMA ) . ( C ) Flowchart depicting screen for PIMS and validation process . From the initial collection of 7343 siRNAs , 413 were classified as hyposecretory and 534 as hypersecretory hits ( Supplementary file 1 ) . From this pool we further selected 114 hyposecretory and 155 hypersecretory components ( Supplementary file 1 ) . A second validation screen with siRNAs distinct from the primary screen confirmed 29 proteins that gave a MUC5AC hyposecretory phenotype and 5 with a hypersecretory secretory defect ( Table 1 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 00658 . 00610 . 7554/eLife . 00658 . 007Figure 3—figure supplement 1 . PIMS are not required for constitutive secretion of newly synthesized proteins . N2 cells were starved for 6 days and transfected with the corresponding siRNA on day 6 . The cells were grown for 3 additional days , pulsed with 35S-methionine and chased with cold methionine-containing medium . The medium was collected and analyzed by SDS-PAGE followed by autoradiography . Treatment with BFA during the pulse and chase was used as positive control to score effects on the secretion of newly synthesized proteins . DOI: http://dx . doi . org/10 . 7554/eLife . 00658 . 007 This collection of 269 hits was rescreened with another siRNA library composed of a pool of four different siRNAs targeting the same protein . The same procedure as described above was used to monitor the effect of siRNA on MUC5AC secretion . The secreted MUC5AC levels were normalized with the Z-score . For the hit analysis we assumed mainly positive hits affecting MUC5AC secretion . Therefore the cutoff was set according to mock transfected cells ±2 SD . With that setup , we identified 29 components exhibiting a hyposecretory phenotype and 5 with a hypersecretory phenotype ( Figure 3C and Table 1 ) . 10 . 7554/eLife . 00658 . 008Table 1 . Thirty-four proteins involved in MUC5AC secretionDOI: http://dx . doi . org/10 . 7554/eLife . 00658 . 008Gene IDSymbolMUC5AC secretionTypeKnown localizationExpression in N2 cells6833SUR1HyposecretoryTransporterPMDetected2900GRIK4HyposecretoryIon channelPMDetected29 , 850TRPM5HyposecretoryIon channelPMDetected6326SCN2AHyposecretoryIon channelPMNot detected6328SCN3AHyposecretoryIon channelPMNot detected6915TBXA2RHyposecretoryGPCRPMNot detected6753SSTR3HyposecretoryGPCRPMNot detected122042RXFP2HyposecretoryGPCRPMNot detected1394CRHR1HyposecretoryGPCRPMNot detected10 , 803CCR9HyposecretoryGPCRPMDetected524CX3CR1HyposecretoryGPCRPMNot detected8484GALR3HyposecretoryGPCRPMNot detected83 , 550GPR101HyposecretoryGPCRPMNot detected118442GPR62HyposecretoryGPCRPMDetected4914NTRK1HyposecretoryTM receptorPMNot detected9437NCR1HyposecretoryTM receptorPMNot detected55 , 824PAG1HyposecretoryOtherPMDetected259215LY6G6FHyposecretoryTM receptorPMNot detected56 , 140PCDHA8HyposecretoryCadherinPMNot detected225689MAPK15HyposecretoryKinaseCytosolDetected23 , 542MAPK8IP2HyposecretoryOtherCytosolNot detected10 , 454TAB1HyposecretoryScaffoldCytosolDetected6490SILVHyposecretoryMelanosome biogenesisMelanosomeDetected6494SIPA1HyposecretoryGTPase activatorCytosol , nucleusDetected30 , 818KCNIP3HyposecretoryCa2+ bindingPalmitoylated: membrane , cytosolDetected22 , 926ATF6HyposecretoryTranscriptionERDetected26 , 499PLEK2HyposecretoryActin bindingCytoskeletonDetected4790NFKB1HyposecretoryTranscriptionNucleus , cytosolDetected6720SREBF1HyposecretoryTranscriptionERDetected1238CCBP2HypersecretoryGPCRPMDetected84 , 033OBSCNHypersecretoryKinaseCytosolNT9373PLAAHypersecretoryOtherCytosolNT4843NOS2HypersecretoryOtherCytosolNT4893NRASHypersecretoryG-proteinCytosolNTER: endoplasmic reticulum; GPCR: G protein coupled receptor; PM: plasma membrane; TM receptor: transmembrane receptor; NT: not tested . It is important to test whether any of the proteins identified in our screening assay have a role in constitutive secretion of cargoes that do not enter the secretory granules . This could reveal the convergent function of PIMS in conventional and regulated protein secretion . N2 cells were starved for 6 days , transfected with siRNAs for the individual PIMS , and 3 days later were washed in methionine free medium for 20 min . The cells were then incubated with 35S-methionine containing medium for 15 min and subsequently cultured in methionine containing medium . After 3 hr , the medium was collected and the cells were lysed and measured for total 35S-methionine incorporation . As a control , BFA was added ( at 2 µg/ml ) during the whole pulse-chase procedure to inhibit general protein secretion . The secreted medium was normalized to the incorporation of 35S-methionine , corresponding volumes precipitated by TCA , and analysis done by SDS-PAGE . We did not detect any obvious changes in the secreted polypeptide pool in starved N2 cells depleted of the individual PIMS compared with control cells ( Figure 3—figure supplement 1 ) . As expected , BFA treatment inhibited general protein secretion ( Figure 3—figure supplement 1 ) . We tested whether the proteins selected for MUC5AC secretion are expressed in starved and unstarved cells , as their expression could depend on the differentiation state of the cells . Total RNA was extracted from starved and unstarved N2 cells and the expression of 34 components selected for MUC5AC secretion ( Table 1 ) was determined by reverse transcription of the mRNA and PCR with specific primers for each component . The PCR products were separated by agarose gel electrophoresis , quantified and normalized to the housekeeping gene GAPDH ( Figure 4A , B ) . Our findings reveal that 16 of the 34 hits can be detected in N2 cells ( Table 1 and Figure 4A ) . Of these , the mRNA levels of eight components remain unchanged upon starvation compared with unstarved cells ( Figure 4A , B ) . Interestingly , in differentiated N2 cells the levels of GRIK4 , KCNIP3 , SIPA1 , MAPK15 , ATF6 and CCR9 were upregulated , while the mRNA levels of SUR1 and SILV were downregulated ( Figure 4A , B ) . As expected , the MUC5AC mRNA was upregulated upon starvation whereas the mRNA levels for the housekeeping gene GAPDH remained unchanged ( Figure 4A , B ) . Based on the data presented thus far we suggest that these 16 proteins are required for MUC5AC secretion and have labelled them as PIMS ( Protein Involved in Mucin Secretion ) for the sake of simplicity ( Table 2 ) . 10 . 7554/eLife . 00658 . 009Figure 4 . Expression profile of PIMS . ( A ) Total RNA was extracted from nonstarved- and 5-day starved N2 cells . 1 μg of total RNA was used to generate cDNA by reverse transcription . PCR was performed on cDNA with primers specific for the indicated genes and PCR products were analyzed by agarose gel electrophoresis . ( B ) Results in ( A ) were quantified and values were normalized to the housekeeping gene GAPDH . The y-axis represents relative values of starved compared to nonstarved N2 cells . Average values ± SEM are plotted as bar graphs ( N = 3 ) . Compared datasets were considered as statistically significant when p<0 . 05 ( * ) , p<0 . 01 ( ** ) , p<0 . 001 ( *** ) and p<0 . 0001 ( **** ) . Abbreviations: R1: replicate 1; R2: replicate 2; R3: replicate 3; RT-: reverse transcription without reverse transcriptase . DOI: http://dx . doi . org/10 . 7554/eLife . 00658 . 00910 . 7554/eLife . 00658 . 010Table 2 . Identification of proteins involved in MUC5AC secretion ( PIMS ) DOI: http://dx . doi . org/10 . 7554/eLife . 00658 . 010GenePIMSMUC5AC secretionTypeKnown localizationExpression in starved N2 cellsDiseaseSUR11HyposecretoryTransporterPMUnchangedDiabetes mellitus , insulin secretion ( Aittoniemi et al . , 2009 ) GRIK42HyposecretoryIon channelPMUpregulatedTRPM53HyposecretoryIon channelPMUnchangedInsulin secretion ( Colsoul et al . , 2010 ) KCNIP34HyposecretoryCa2+ bindingPalmitoylated membrane , cytosolUpregulatedSIPA15HyposecretoryGTPase activatorCytosol , nucleusUnchangedPLEK26HyposecretoryActin bindingCytoskeletonUnchangedTAB17HyposecretoryScaffoldCytosolUnchangedMAPK158HyposecretoryKinaseCytosolUpregulatedSILV9HyposecretoryMelanosome biogenesisMelanosomeUnchangedSREBF110HyposecretoryTranscription regulatorERUnchangedATF611HyposecretoryTranscription regulatorERUnchangedNFKB112HyposecretoryTranscription regulatorCytosol , nucleusUnchangedMUC5AC biosynthesis ( Fujisawa et al . , 2009 ) ; asthma ( Hart et al . , 1998 ) ; COPD ( di Stefano et al . , 2002 ) PAG113HyposecretoryOtherPMUnchangedCCR914HyposecretoryGPCRPMUpregulatedInflammatory bowel disease ( Nishimura et al . , 2009 ) GPR6215HyposecretoryGPCRPMUnchangedCCBP216HypersecretoryGPCRPMUnchangedER: endoplasmic reticulum; GPCR: G protein-coupled receptor; PM: plasma membrane; TM receptor: transmembrane receptor . TRPM5 , a hit in our assay , is required for the regulated secretion of insulin ( Brixel et al . , 2010; Colsoul et al . , 2010 ) and this prompted us to further test its involvement in the regulated secretion of MUC5AC . TRPM5 is a Ca2+-activated monovalent cation-selective channel that responds to warm temperature and participates in the taste-receptor signaling pathway ( Perez et al . , 2002; Hofmann et al . , 2003; Zhang et al . , 2003; Talavera et al . , 2005 ) . We generated a shRNA-dependent stable N2 goblet cell line depleted of TRPM5 . This procedure resulted in greater than 80% reduction in the TRPM5 mRNA levels ( Figure 5A ) and a 50% reduction in the quantity of PMA dependent MUC5AC secreted by starved cells ( Figure 5B ) . The defect in secretion could reflect a reduction in the total intracellular protein pool of MUC5AC . Therefore , we probed the total cell lysate from TRPM5 knockdown and control cells with an anti-MUC5AC antibody by dot blot . The dot blot was quantified and revealed no difference in the total MUC5AC contents in the TRPM5 knockdown cells compared with control cells ( Figure 5C ) . We also tested whether the reduction in secretion is due to a defect in granule formation . Intracellular granules were visualized by immunofluorescence microscopy with an anti-MUC5AC antibody in starved TRPM5-depleted cells . GFP-positive cells indicate the expression of TRPM5 shRNA and show that secretory granules were formed in TRPM5-depleted cells upon starvation ( Figure 5D ) . As shown in Figure 2D , secretion of MUC5AC is BFA-independent and therefore from a post-Golgi pool . 10 . 7554/eLife . 00658 . 011Figure 5 . TRPM5 in mucin homeostasis . ( A ) Total RNA was extracted from control and TRPM5 stable knockdown ( TRPM5 KD ) cells and analyzed for knockdown efficiency of TRPM5 on mRNA level by quantitative real-time PCR using primers specific for TRPM5 . TRPM5 values were normalized to values of the housekeeping gene GAPDH . The knockdown of TRPM5 is represented as relative value compared to control cells . Results are means ± SEM . ( N = 5 ) . ( B ) Control and TRPM5 stable knockdown cells were starved for 6 days and seeded on 96-well plates . At day 9 , cells were treated with increasing concentrations of PMA for 30 min and analyzed by chemiluminescence using anti-MUC5AC antibody . After the mucin secretion assay these cells were stained with DAPI and imaged by fluorescence microscopy . Nuclei were counted using ImageJ and the chemiluminescence value for MUC5AC of each well was normalized to the number of nuclei per well . The y-axis represents relative values with respect to the values of control cells not treated with PMA . Average values ± SEM are plotted as bar graphs ( N = 5 ) . Datasets were considered as statistically significant when p<0 . 0001 ( **** ) . ( C ) N2 cells were starved for 6 days and seeded for dot blot analysis . At day 9 cells were lysed and analyzed by dot blot with an anti-MUC5AC and anti-actin antibody . The intensity of the spots was quantified using ImageJ . Intensities of MUC5AC spots were normalized to the intensity of actin spots . Results are means ± SEM . ( N = 6 ) . ( D ) Control and TRPM5 stable knockdown cells were differentiated by starvation . After starvation cells were fixed , permeabilized and analyzed by immunofluorescence microscopy with an anti-MUC5AC antibody ( red ) and DAPI ( blue ) . Cells shown in green represent expression of GFP , showing that these cells express shRNA specific for TRPM5 and are depleted of TRPM5 . DOI: http://dx . doi . org/10 . 7554/eLife . 00658 . 011 HA-tagged ( N-terminal ) TRPM5 was expressed in N2 cells and localized by confocal fluorescence microscopy in cells grown as polarized monolayers . HA-TRPM5 was localized to the apical surface of the polarized non-starved and starved N2 cells while apical MUC5AC staining was only evident in starved N2 cells ( Figure 6A ) . TRPM5 channel ( green , Figure 6A ) appears proximal to the nucleus in non-starved cells because these cells are shorter in height compared with the starved-differentiated cells . Next , we checked the presence of functional TRPM5 channels at the plasma membrane of starved N2 cells . For this purpose , we carried out whole-cell patch-clamp experiments . Dialyzing N2 cells with Ca2+-free intracellular pipette solutions resulted in negligibly low cationic current recorded; the presence of 1 μM Ca2+ in the pipette solution generated an outwardly rectifying TRPM5-like current ( Figure 6B ) . Replacement of extracellular Na+ by the non-permeant cation N-methyl-D-glucamine reduced the inward current without affecting the outward current ( Figure 6B ) . shRNA-mediated knockdown of TRPM5 abolished Ca2+-dependent cationic currents recorded in starved N2 cells ( Figure 6C , D ) . Together , these results are consistent with the participation of the TRPM5 channel in the generation of the plasma membrane , nonselective , Ca2+-dependent cationic current in N2 cells . 10 . 7554/eLife . 00658 . 012Figure 6 . TRPM5 localization and activity . ( A ) Nonstarved and starved N2 cells stably transfected with HA-TRPM5 were fixed and analyzed by immunofluorescence microscopy with an anti-HA antibody ( green ) and an anti-MUC5AC antibody ( red ) . The nuclear DNA was stained with DAPI ( blue ) to localize the position of the nucleus . ( B ) Ramp current-voltage relations of cationic currents recorded from a starved N2 cell dialyzed with internal solutions containing 1 μM Ca2+ and bathed in Na+-containing or Na+-free solutions . A ramp current obtained in a cell dialyzed with internal solutions containing 0 Ca2+ and bathed in Na+-containing is also shown . ( C ) Representative ramp current-voltage relations of cationic currents recorded from a control and TRPM5-depleted N2 cells dialyzed with internal solutions containing 1 μM Ca2+ and bathed in Na+-containing solutions . ( D ) Mean TRPM5-like current density recorded at ±100 mV from control ( n = 6 ) and TRPM5 KD cells ( n = 8 ) . * p<0 . 05 . DOI: http://dx . doi . org/10 . 7554/eLife . 00658 . 012 Is TRPM5 required for goblet cell response to physiological secretagogues such as ATP ? The induction of mucin secretion by ATP is linked to phospholipase C activation , IP3 and DAG generation , and Ca2+ release from the ER ( Abdullah et al . , 2003; Davis and Dickey , 2008 ) . To test the involvement of TRPM5 in ATP-induced MUC5AC secretion , wild-type and stable TRPM5-depleted N2 cells were differentiated and treated with ATP for 30 min . Knockdown of TRPM5 significantly reduced ATP-induced secretion of MUC5AC ( Figure 7A ) . These findings strongly support the significance of our procedure for the identification of proteins required for the regulated secretion of MUC5AC and confirm the role of TRPM5 in mucin secretion under physiological ( extracellular ATP ) conditions . 10 . 7554/eLife . 00658 . 013Figure 7 . TRPM5 modulates ATP-induced MUC5AC secretion . ( A ) Control and TRPM5 stable knockdown cells were starved and incubated for 30 min at 37°C with 100 μM ATP . Secreted MUC5AC was collected and processed for dot blot analysis with an anti-MUC5AC antibody . The dot blots were quantified and normalized to intracellular actin levels . The y-axis represents relative values with respect to the values of untreated control cells . Average values ± SEM are plotted as bar graphs ( N = 6 ) . Datasets were considered as statistically significant when p<0 . 01 ( * ) . ( B ) Starved N2 cells were incubated for 30 min at 37°C with 100 μM ATP in the presence ( 1 . 2 mM CaCl2 ) or absence ( 0 mM CaCl2 ) of extracellular Ca2+ . Secreted MUC5AC was collected and analyzed by dot blot with an anti-MUC5AC antibody . The dot blots were quantified and normalized to intracellular actin levels . The y-axis represents relative values with respect to the values of untreated N2 cells in the presence of 1 . 2 mM CaCl2 . Average values ± SEM are plotted as bar graphs ( N = 3 ) . Datasets were considered as statistically significant when p<0 . 001 ( *** ) . DOI: http://dx . doi . org/10 . 7554/eLife . 00658 . 01310 . 7554/eLife . 00658 . 014Figure 7—figure supplement 1 . Ca2+ dependent MUC5AC secretion . Starved N2 cells were treated for 30 min with the indicated concentrations of thapsigargin and ionomycin . Secreted MUC5AC was collected and processed for dot blot analysis with an anti-MUC5AC antibody . The dot blots were quantified and normalized to intracellular β-tubulin levels . The Y-axis represents relative values with respect to the values of untreated cells . Average values ± SEM are plotted as bar graphs ( N = 6 ) . Datasets were considered as statistically significant with the following p values: Untreatedt vs 1 μM thapsigargin , p<0 . 0001 ( **** ) ; untreated vs 1 μM ionomycin , p<0 . 0001 ( **** ) ; thapsigargin vs ionomycin , p<0 . 01 ( ** ) DOI: http://dx . doi . org/10 . 7554/eLife . 00658 . 014 Ca2+ entry in response to different secretagogues has been reported for several types of mucin-secreting cells , although the mechanism of Ca2+ influx is poorly understood ( Bou-Hanna et al . , 1994; Bertrand et al . , 2004; Lu et al . , 2011 ) . Since Ca2+ is essential for the regulated secretion of MUC5AC and TRPM5 channel activity is known to affect cell membrane potential , we reasoned that the role of TRPM5 in MUC5AC secretion is to modulate secretagogue-induced Ca2+ influx . We tested whether extracellular Ca2+ was required for ATP-dependent MUC5AC secretion . N2 cells were differentiated by starvation and washed; MUC5AC secretion was then induced in either Ca2+-containing or Ca2+-free medium for 30 min in the presence of 100 μM ATP . N2 cells cultured without Ca2+ in the extracellular medium did not secrete MUC5AC upon exposure to ATP . However , there was a sixfold increase in MUC5AC secretion by differentiated N2 cells in the presence of extracellular Ca2+ ( Figure 7B ) . To further investigate the source of Ca2+ in this process we measured MUC5AC secretion in cells treated with thapsigargin , which promotes the release of Ca2+ from the ER ( Cantero-Recasens et al . , 2010 ) or in response to the Ca2+ ionophore ionomycin , which mainly facilitates Ca2+ entry into the cytoplasm from the extracellular medium ( Figure 7—figure supplement 1 ) . MUC5AC secretion by ionomycin was significantly higher than treatment with thapsigargin , however , these reagents were less effective , compared with ATP , in eliciting MUC5AC secretion . Together , these results indicate that the contribution of Ca2+ release from internal stores is not as significant as Ca2+ entry from the extracellular medium for MUC5AC secretion by N2 cells . Consistent with the secretion data , measurements of intracellular Ca2+ levels in starved N2 cells loaded with the calcium dye Fura-2 showed that cells treated with ATP elicited an increase in intracellular Ca2+ concentration , which was markedly decreased when extracellular Ca2+ was removed ( Figure 8A ) . Mean increases in the peak Ca2+ signal are shown in the right panel . These data suggested a close link between secretagogue-induced Ca2+ entry and MUC5AC secretion . 10 . 7554/eLife . 00658 . 015Figure 8 . TRPM5 modulates ATP-induced Ca2+ entry . ( A ) Time course of mean Ca2+ responses ( Fura-2 ratio ) obtained in starved N2 cells treated with 100 μM ATP in the presence ( n = 138 ) or absence of 1 . 2 mM Ca2+ ( n = 118 ) in the extracellular solution . Right panel , average peak [Ca2+] increases obtained from traces shown in the right panel . *p<0 . 01 . ( B ) Time course of mean Ca2+ responses ( Fura-2 ratio ) obtained in starved control ( n = 179 ) and TRPM5 KD N2 cells ( n = 163 ) treated with 100 μM ATP . Right panel , average peak [Ca2+] increases obtained from traces shown in the right panel . *p<0 . 01 . ( C ) Time course of mean Ca2+responses ( Fura-2 ratio ) obtained in starved control ( n = 118 ) and TRPM5 KD N2 cells ( n = 89 ) treated with 100 μM ATP and bathed in Ca2+-free solutions . Right panel , average peak [Ca2+] increases obtained from traces shown in the right panel . *p<0 . 01 . DOI: http://dx . doi . org/10 . 7554/eLife . 00658 . 015 We then evaluated whether TRPM5 was required for the ATP-induced entry of extracellular Ca2+ and the subsequent regulation of MUC5AC secretion . Control starved N2 cells and N2 cells stably depleted of TRPM5 were treated with 100 µM ATP and intracellular Ca2+ levels were recorded . After ATP treatment , cells depleted of TRPM5 showed reduced increase in intracellular [Ca2+] compared with control cells ( Figure 8B ) . To test if the reduction in intracellular [Ca2+] upon purinergic receptor activation with ATP reflected a defect in Ca2+ influx from the extracellular medium , we measured the elevation in intracellular Ca2+ level by ATP treatment in N2 cells and TRPM5-depleted cells in the absence of extracellular Ca2+ ( Figure 8C ) . In the absence of extracellular Ca2+ there was no difference between control and TRPM5 depleted cells in ATP-induced increase of intracellular Ca2+ levels , suggesting that TRPM5 participation in ATP-mediated MUC5AC secretion is related to the regulation of the secretagogue-induced Ca2+ entry . TRPM5 might be involved in modulating Ca2+ influx by changing the cell membrane potential following the entry of monovalent cations . Positive modulation of Ca2+ entry by TRPM5-mediated membrane depolarization has been linked to the activation of voltage-gated Ca2+ channels ( Colsoul et al . , 2010; Shah et al . , 2012 ) . However , we detected neither voltage-gated whole-cell Ca2+ currents ( Figure 9—figure supplement 1A ) nor depolarization-induced Ca2+ signals ( Figure 9—figure supplement 1B ) in starved N2 cells . Accordingly , inhibitors of voltage-gated Ca2+ channels did not modify ATP-mediated Ca2+ signals ( Figure 9—figure supplement 1C ) . Therefore , we hypothesized that TRPM5-mediated Na+ entry was coupled to the functioning of a Na+/Ca2+ exchanger ( NCX ) in reverse mode , thereby promoting further Ca2+ entry . We investigated the participation of NCX in ATP-mediated MUC5AC secretion and Ca2+ signaling using KB-R9743 , an NCX inhibitor that preferentially blocks the reverse Ca2+ influx mode of the transporter ( Iwamoto et al . , 1996 ) . Control starved N2 cells and N2 cells stably depleted of TRPM5 were pretreated with 50 μM KB-R9743 for 15 min and then incubated with 100 μM ATP . ATP induced MUC5AC secretion was greatly reduced in the presence of the NCX inhibitor ( Figure 9A ) , which suggests that TRPM5- and Ca2+-dependent MUC5AC secretion involves the activity of an NCX . This hypothesis was further examined by measuring ATP-induced Ca2+ signals in the presence of the NCX inhibitor . ATP-induced Ca2+ signals were reduced by ∼ 50% in cells treated with the NCX inhibitor ( Figure 9B ) . Similar to the results obtained in the absence of extracellular Ca2+ ( Figure 8D ) , in the presence of the NCX inhibitor there was no difference in Ca2+ signals between control and TRPM5-depleted N2 cells ( Figure 9B ) . These results suggest that N2 cells exhibit an ATP-induced Ca2+ entry mechanism that is consistent with the operation of an NCX in reverse mode and this control mechanism is lost in N2 cells depleted of TRPM5 . 10 . 7554/eLife . 00658 . 016Figure 9 . Effect of inhibiting the NCX on MUC5AC secretion and Ca2+ entry . ( A ) Starved N2 cells were preincubated for 15 min with or without KB-R7943 ( 50 μM ) followed by incubation with 100 μM ATP in the presence or absence of KB-R7943 . Secreted MUC5AC was analyzed by dot blot with an anti-MUC5AC antibody . The dot blot was quantified and normalized to intracellular tubulin amount . The y-axis represents relative values with respect to values of untreated control cells . Average values ± SEM are plotted as bar graphs ( N = 6 ) . Datasets were considered as statistically significant when p<0 . 01 ( ** ) . ( B ) Time course of mean Ca2+ responses ( Fura-2 ratio ) obtained in starved control ( n = 84 ) and TRPM5 KD N2 cells ( n = 83 ) treated with 100 μM ATP in the presence of 50 μM KB-R7943 . Right panel , average peak [Ca2+] increases obtained from traces shown in the right panel . DOI: http://dx . doi . org/10 . 7554/eLife . 00658 . 01610 . 7554/eLife . 00658 . 017Figure 9—figure supplement 1 . Voltage-gated Ca2+ channels are not expressed or functional in N2 cells . ( A ) Current traces ( bottom ) obtained in a starved N2 cell held at −80 mV and pulsed from −60 mV to +60 mV in 5 mV voltage steps ( top ) . Traces are representative of six cells recorded under identical conditions . None of the recorded cells presented voltage-gated Ca2+ currents . ( B ) Lack of Ca2+ response ( Fura-2 ratio ) in starved control N2 cells exposed to high K+ solutions . Mean ± SEM , n = 43 . ( C ) Left , time course of mean Ca2+ responses ( Fura-2 ratio ) obtained in starved control N2 cells treated with 100 μM ATP in the absence ( n = 35 ) or presence ( n = 48 ) of the voltage-gated Ca2+ channel blocker nifedipine ( 100 μM ) . Right panel , peak [Ca2+] increases obtained from traces shown in the right panel . DOI: http://dx . doi . org/10 . 7554/eLife . 00658 . 017
The size and rheological properties of gel-forming mucins has hindered the development of a quantitative assay to monitor their secretion . Our antibody-based detection of secreted MUC5AC is relatively easy , quantitative , and highly accurate . It involves starvation-induced synthesis of MUC5AC , which is then released by treating the cells with PMA . It has recently been shown that secretion of total polymeric mucins from goblet-cell metaplastic human bronchial epithelial cultures is inhibited by BFA treatment ( Okada et al . , 2010 ) . This likely represents secretion of newly synthesized mucin that is secreted at some basal rate . PMA mediated MUC5AC secretion reported here is unaffected by BFA treatment ( Figure 2D , E ) . Our assay , therefore , measures release of MUC5AC from the post Golgi secretory storage granules . Based on our experimental data from a pool of 7343 gene products tested , we selected 16 proteins because their knockdown significantly affected MUC5AC secretion from the goblet cell line . These proteins ( PIMS ) are expressed in the goblet cells and not required for general protein secretion . PIMS include ion channels and regulatory molecules ( SUR1 , GRIK4 and TRPM5 ) ; GPCR’s ( CCR9 , GRP62 and CCBP2 ) , transcription regulators ( SREBF1 , ATF6 and NFKB1 ) , Ca2+ binding protein ( KCNIP3 ) , GTPase activator for Rap1 that controls actin dynamics ( SIPA1 ) , actin binding protein ( PLEK2 ) , scaffold for the MAPK ( TAB1 ) , MAPK15 , and a protein involved in melanosome biogenesis ( SILV ) . Actin dynamics are important for MUC5AC secretion and , as shown here , stablization of actin filaments but not their depolymerization inhibited MUC5AC secretion . The identification of SIPA1 and PLEK2 could help reveal the components involved in regulating Rap1 , which is known to regulate actin filament dynamics in the events leading to the docking/fusion of the MUC5AC-containing secretory granules . SILV is required for the early stages of melanosome biogenesis , and goblet cells express SILV but are not known to make melanosomes . It is reasonable to propose that SILV performs an analogous function in the maturation of MUC5AC granules in the goblet cells . TAB1 and MAPK15 are likely involved in stress response-mediated synthesis and secretion of MUC5AC . The cell-surface ion channels and the GPCRs are likely involved in signaling events that lead to the secretion of MUC5AC . Future analysis of these proteins will help reveal their significance in MUC5AC homeostasis . TRPM5 is a Ca2+-activated monovalent cation selective channel that responds to warm temperature and a key component of the bitter , sweet and umami taste-receptor signaling cascade ( Perez et al . , 2002; Zhang et al . , 2003 ) . Bitter , sweet and umami tastants are detected by GPCRs that signal through gustducin and PLCβ2 in order to produce DAG and IP3 . IP3 activates the release of Ca2+ from the ER , which then activates TRPM5 ( Hofmann et al . , 2003; Liu and Liman , 2003 ) . The activated TRPM5 affects membrane potential to control events leading to the sensation of bitterness . Interestingly , bitter taste receptors ( GPCRs ) are expressed in airway smooth muscle and GPCRs coupled to Gαq or Gαs control muscle contraction and relaxation , respectively ( Deshpande et al . , 2010 ) . A large number of GPCRs regulate taste signaling , but it is not clear which family members are expressed in the goblet cells and linked to the activation of TRPM5 . We have found several GPCRs and ion channels that are expressed in the goblet cells and required for MUC5AC secretion . It would be important to test whether any of these GPCRs are directly linked to TRPM5 activation in the goblet cells and therefore in the events leading MUC5AC secretion . The activity of TRPM5 , and its close relative TRPM4 , has been shown to couple intracellular Ca2+ to changes in membrane potential and either an activation of voltage-gated Ca2+ channels favouring further Ca2+ increase ( Earley et al . , 2004; Colsoul et al . , 2010; Uchida and Tominaga , 2011; Shah et al . , 2012 ) or a reduction in the electrochemical gradient for store-operated Ca2+ entry ( Vennekens et al . , 2007 ) . Our results provide evidence for a novel link between TRPM5 activity and the control of intracellular Ca2+ signaling via the functional coupling of TRPM5 and NCX . NCX at the plasma membrane is one of the major means to extrude Ca2+ ( in exchange for Na+ ) from the cells . However , this transporter can also function in the reverse mode to allow Ca2+ entry when intracellular Na+ concentration increases . Significant Na+ entry is expected to occur during activation of TRPM5 following IP3-mediated Ca2+ release . Sodium entering through the TRPM5 would accumulate intracellularly , thus promoting NCX to act in the reverse mode and permit Ca2+ entry in exchange for the exiting Na+ . Such a coupling between endogenous NCX and heterologously expressed Na+ and Ca2+ permeable TRPC3 channel has been reported ( Rosker et al . , 2004 ) . However , our observation describes for the first time the functional coupling between endogenous TRPM5 channel and NCX , and its physiological relevance in the context of mucin secretion . In conclusion , we have devised a quantitative assay to measure mucin secretion in human goblet cell lines . Our findings have revealed new components that control MUC5AC secretion . We have described the mechanism linking the activity of TRPM5 with the modulation of Ca2+ signals and MUC5AC secretion from the differentiated colonic goblet cell line . It is important now to test the expression of PIMS in cells of the respiratory and gastric lining; their involvement in secretion of MUC5AC and other secreted mucins such as MUC2 to understand the mechanisms of mucin homeostasis .
All chemicals were obtained from Sigma-Aldrich ( St . Louis , MO ) except BAPTA-AM ( Biomol International , Farmingdale , NY ) , Latrunculin A ( Calbiochem , Billerica , MA ) , Jasplakinolide ( Life Technologies , Carlsbad , CA ) and KB-R7943 ( Tocris Bioscience , Bristol , UK ) , anti-MUC5AC antibody clone 45M1 ( Neomarkers , Waltham , MA ) , anti-GRASP65 C-20 antibody from Santa Cruz Biotechnology ( Dallas , TX ) , anti-actin antibody clone AC-15 from Sigma-Aldrich anti-β tubulin antibody from Sigma-Aldrich and anti-HA antibody clone 3f10 from Roche ( Basel , Switzerland ) . Secondary antibodies for immunofluorescence microscopy and dot blotting were from Life Technologies . The shRNA targeting TRPM5 was constructed as follows: target sequence 5′- GTACTTCGCCTTCCTCTTC - 3′ , loop 5′ - TTCAAGAGA - 3′ , antisense 5′ - GAAGAGGAAGGCGAAGTAC - 3′ , terminator 5′ - TTTTTTC - 3′ . The sense and antisense shRNA oligos were annealed and cloned via HpaI and XhoI into the lentilox 3 . 1 vector containing GFP and sequence verified . HA-pcDNA3 . 1 plasmid was constructed as follows: primers encoding the HA-tag were designed ( forward primer 5′ - GGCCGCACCATGTACCCTTACGACGTTCCTGATTACGCATCCCTTGAATTCCCCGGGG - 3′ and reverse primer 5′ - GATCCCCCGGGGAATTCAAGGGATGCGTAATCAGGAACGTCGTAAGGGTACATGGTGC - 3′ ) , annealed , cloned via NotI and BamHI into pcDNA3 . 1 and sequence verified . TRPM5 cDNA was obtained from Imagenes ( UK ) ( accession number BC093787 ) and amplified using the following primers: forward primer 5′ - CGCATAGAATTCCAGGATGTCCAAGGCCCCCG - 3′ and reverse primer 5′ - TAAGGTACCTGTTCAGGTGTCCGAGGGAGGCTGGCC . The amplified PCR product was cloned into HA-pcDNA3 . 1 plasmid via EcoRI and KpnI and sequence verified . HA-TRPM5 was cloned via NotI and BclI into a lentiviral vector . All primers were obtained from Sigma-Aldrich . N2 cells were seeded in complete growth medium ( DMEM complemented with 10% FBS ) . The next day ( d0 ) medium was exchanged for PFHMII protein free medium ( Invitrogen , Carlsbad , CA ) and cells were grown for 3 days . At day 3 ( d3 ) the medium was replaced with fresh PFHMII medium and the cells were grown for 3 additional days ( total time 6 days ) . At day 6 ( d6 ) cells were seeded in complete growth medium for different experiments following replacement of complete growth medium for PFHMII medium at day 7 ( d7 ) and experimental procedure at day 9 ( d9 ) . N2 cells were differentiated for 6 days and then aliquoted into the wells of a 96-well plate . After 1 day of growth ( d7 ) , the medium was replaced by fresh PFHMII medium and the cells were grown for 2 more days ( total time 9 days ) . On d9 cells were treated with 2 μM PMA for 2 hr at 37°C . The secreted MUC5AC was fixed by adding paraformaldehyde ( PFA ) to the cells at a final concentration of 4% for 30 min at RT . Cells were subsequently washed with PBS and incubated with the anti-MUC5AC antibody ( clone 45M1 ) diluted 1:100 in 4% BSA/PBS for 1 hr at RT . The cells were then washed with PBS and incubated with a donkey anti-mouse HRP conjugated antibody ( Santa Cruz Biotechnology ) diluted 1:10 , 000 in 4% BSA/PBS for 1 hr at RT . After washing the cells in PBS , 100 μl ECL solution was added to the cells and luminescence was measured with a Tecan plate reader . N2 cells were differentiated for 6 days and then seeded into six-well plates . The cells were processed until d9 as described for the mucin secretion assay . On d9 cells were treated with 100 µM ATP for 30 min at 37°C . Supernatant was collected and centrifuged for 5 min at 800×g at 4°C . Cells were washed 2× in PBS and lysed in 1% Triton X-100/1 mM DTT/PBS for 1 hr at 4°C and centrifuged at 1000×g for 10 min . The supernatants and cell lysates were spotted on nitrocellulose membranes and membranes were incubated in blocking solution ( 4% BSA/0 . 1% Tween/PBS ) for 1 hr at room temperature . The blocking solution was removed and the membranes were incubated with the anti-MUC5AC antibody diluted 1:1000 or the anti-actin antibody at a dilution of 1:1000 in blocking solution . Membranes were washed in 0 . 1% Tween/PBS and secondary antibodies conjugated to HRP were incubated in blocking solution at a dilution of 1:10 , 000 for 1 hr at room temperature . For the detection of β-tubulin , cell lysates were separated on SDS-PAGE , transferred to nitrocellulose membranes and processed as described for the dot blot analysis using the anti-β-tubulin antibody at a dilution of 1:10 , 000 . Membranes were washed , incubated with ECL substrate and imaged with a Fujifilm LAS-3000 camera . Membranes were analyzed and quantitated in ImageJ ( version 1 . 44o; National Institutes of Health ) . N2 cells were differentiated for 6 days . On d6 , 4 × 104 cells were seeded into the wells of a 96-well plate and transfected in triplicates on three sets of plates with 150 nM siRNA ( provided by the high throughput screening facility at the Center for Genomic Regulation ) and Dharmafect 4 ( Dharmacon , Lafayette , CO ) according to manufacturer’s instructions . The cells grown on the plates were handled until d9 as described above . On d9 , cells were treated with 2 μM PMA for 2 hr at 37°C and processed for MUC5AC secretion as described in the Mucin secretion assay . The Mucin secretion assay was automated and performed on the Caliper LS staccato workstation . Each plate was normalized by the B-score method ( Brideau et al . , 2003 ) and positive hits were selected above B-score 1 . 5 and below B-Score −1 . 5 . The hits were classified using the ranking product method ( Breitling et al . , 2004 ) using the triplicates . The data was analyzed and automated by a script written with the statistical toolbox from Matlab ( Mathwork ) . The validation screen was performed exactly as described for the screen procedure . The ontarget PLUS siRNAs were obtained from Dharmacon ( Lafayette , CO ) . All the plates were normalized platewise by:z-score= ( xi−average ( xn ) ) /SD ( xn ) , xn = total population and xi = sample . Positive hits were selected 2 SD above and below mock treated samples . Undifferentiated and differentiated N2 cells were grown on coverslips . For the visualization of intracellular MUC5AC cells were fixed with 4% PFA/PBS for 30 min at RT . Cells were washed with PBS and permeabilized for 20 min with 0 . 2% Triton X-100 in 4% BSA/PBS . The anti-MUC5AC antibody was added to the cells at 1:1000 in 4% BSA/PBS for 1 hr . Cells were washed in PBS and incubated with a donkey anti-mouse Alexa 488 coupled antibody ( Invitrogen ) , diluted at 1:1000 in 4% BSA/PBS , and DAPI . Cells were washed in PBS and mounted in FluorSave Reagent ( Calbiochem , Billerica , MA ) . For the detection of secreted MUC5AC , differentiated N2 cells were treated with 2 µM PMA for 2 hr at 37°C . The secreted MUC5AC was fixed on the cells by adding PFA to the cells at a final concentration of 4% for 30 min at RT . The cells were then processed for immunofluorescence analysis ( as described before ) without the permeabilization step with Triton X-100 . For the removal of secreted MUC5AC , cells were incubated for 2 hr with 2 µM PMA at 37°C . The cells were then placed on ice and washed 2× with ice cold PBS . Subsequently , cells were incubated in 1 mM DTT/0 . 05% Trypsin-EDTA 1× ( Invitrogen ) /PBS for 10 min at 4°C , following four washes in ice-cold PBS and two washes in 4% BSA/PBS . The cells were then fixed in 4% PFA/PBS for 30 min at room temperature , permeabilized with 0 . 2% Triton X-100 in 4% BSA/PBS and processed for immunofluorescence as described before . Cells were imaged with a confocal microscope ( SP5; Leica ) using the 63× Plan Apo NA 1 . 4 objective . For detection , the following laser lines were applied: DAPI , 405 nm; and Alexa Fluor 488 , 488 nm; Alexa Fluor 568 , 561 nm . Pictures were acquired using the Leica software and converted to TIFF files in ImageJ ( version 1 . 44o; National Institutes of Health ) . Differentiated N2 cells grown on six-well plates were starved in methionine- and cystine-free DMEM ( Invitrogen , Carlsbad , CA ) for 20 min at 37°C . Cells were labeled with 100 μCi 35 S-methionine for 15 min and chased for 3 hr at 37°C in medium supplemented with 10 mM L-methionine . Brefeldin A ( BFA ) Sigma-Aldrich was added at a concentration of 2 µg/ml during starvation , pulse and chase . The supernatant was collected and centrifuged for 5 min at 800×g at 4°C . Cells were washed with PBS and lysed in 1% Triton X-100/PBS for 1 hr at 4°C , following centrifugation for 30 min at 4°C at 16 , 000×g . Lysates were measured for 35S-methionine incorporation with a beta-counter . Supernatants were normalized to incorporated 35S-methionine and precipitated by TCA . Samples were separated by SDS-PAGE and analyzed by autoradiography . Unstarved- and 5-day starved N2 cells were lysed and total RNA was extracted with the RNeasy extraction kit ( Qiagen , Netherlands ) . Total RNA was treated with Dnase I ( New England Labs , Ipswich , MA ) for 1 hr at 37°C and purified by phenol extraction . cDNA was synthesized with Superscript III ( Invitrogen ) . Primers for each gene ( sequence shown below , Table 3 ) were designed using Primer 3 v 0 . 4 . 0 ( Rozen and Skaletsky , 2000 ) , limiting the target size to 300 bp and the annealing temperature to 60°C . To determine expression levels of MUC5AC and TRPM5 , quantitative real-time PCR was performed with Light Cycler 480 SYBR Green I Master ( Roche , Switzerland ) according to manufacturer’s instructions . Expression of PIMS in unstarved and starved cells was determined by quantifying the PCR band intensities with ImageJ software . 10 . 7554/eLife . 00658 . 018Table 3 . Primer sequences used for detecting mRNA for the respective PIMS in N2 cellsDOI: http://dx . doi . org/10 . 7554/eLife . 00658 . 018GeneForward primer 5′–3′Reverse primer 5′–3′TRPM5GTGGCCATCTTCCTGTTCATCTTCATCATGCGCTCTACCACCR9GCCAGCCTTGGCCCTGTTGTTCCAGCAAGGCCTGCGCTTCPLEK2AGAACAGGCCAGTGGGTGGGTGCTCGCTCAGCCTTGCTGCTTAB1TCAATCATCGCAGCAATCTCGGCTACACGGACATTGACCTKCNIP3CCACCACCTATGCACACTTCCGTCGTAGAGATTAAAGGCCCACSILVGGGCTACAAAAGTACCCAGAAACCCTTGAGGGACACTTGACCACSIPA1CTCCTTTCTGCCACGTACCTTTTTTGGAGTTCCCTTAGGGTCTHPRT1TGACACTGGCAAAACAATGCAGGTCCTTTTCACCAGCAAGCTMUC5ACCAACCCCTCCTACTGCTACGCTGGTGCTGAAGAGGGTCATGPR62GGTGGTTTCCGTGGGGGCTCTGGGCCCAGACCGCAGGATTPAG1TGGACGGCAGCCATGCATCCACTGTTGGTGTGGGCAGCGGATF6AGGTGGGTAGCGGTTGGGAGGGCGGCACCTTACAGGCACCCSREBF1CCACGGCAGCCCCTGTAACGGGGACTGAGACCTGCCGCCTMAPK15TACAACAGGTCCCTCCCCGGCCCCAGTGCCGAGTGGCAGACSUR1GCCTTCGCAGACCGCACTGTCTGCACGGACGAAGGAGGCGNFKB1CGCCACCCGGCTTCAGAATGGGTATGGGCCATCTGCTGTTGGCACCBP2CGGCGGGCATGGGACCATTTAAGGCCACCACCAAGGCTGCGRIK4CGTGGCTCGTGATGGTCGCCGCCTCTCAGGAGCGCGGTTGGAPDHTGCACCACCAACTGCTTAGCGGCATGGACTGTGGTCATGAG Lentivirus was produced by co-tranfecting HEK293 cells with the plasmid , VSV . G and delta 8 . 9 by calcium phosphate . At 48 hr posttransfection the secreted lentivirus was collected , filtered and directly added to N2 cells . Stably infected cells were either selected by puromycine resistance or sorted for GFP positive signal by FACS . The whole-cell configuration of the patch-clamp technique was employed as previously describe to test for the functional expression of TRP channel activity ( Fernandes et al . , 2008 ) and voltage-gated calcium currents ( Serra et al . , 2010 ) . Pipettes with a resistance of 2–3 MΩ were used . To record voltage-gated calcium currents we employed an external solution containing ( in mM ) : 140 tetraethylammonium-Cl , 3 CsCl , 5 CaCl2 , 1 . 2 MgCl2 , 10 HEPES and 10 glucose ( pH 7 . 4 adjusted with Tris ) ; and an intracellular solution: 140 CsCl , 1 EGTA , 4 Na2ATP , 0 , 1 Na3GTP and 10 HEPES ( pH 7 . 2–7 . 3 adjusted with Tris; and 295–300 mosmoles/l ) . To record TRPM5 currents we used an external solution containing: 140 NaCl , 2 . 5 KCl , 0 . 5 MgCl2 , 5 Glucose , 10 Hepes , pH was adjusted to 7 . 4 with Tris , and 300–305 mosmoles/l; and internal solution: 140 Cs acetate , 0 . 85 CaCl2 , 2 MgCl2 , 10 HEPES , 2 ATP , 0 . 5 GTP , pH 7 . 2 adjusted with Tris . Free intracellular calcium concentration to record TRPM5 current was adjusted to either 1 μM or <50 nM ( 0 Ca solution ) with EGTA as calculated with WEBMAXC ( http://www . stanford . edu/~cpatton/webmaxcS . htm ) . Cells were plated in 35-mm plastic dishes and mounted on the stage of an Inverted Olympus IX70 microscope . Whole cell currents were recorded with an Axon200A amplifier or with a D-6100 Darmstadt amplifier , filtered at 1 kHz . Currents were acquired at 33 kHz . The pClamp8 software ( Axon Instruments , Foster City , CA ) was used for pulse generation , data acquisition and subsequent analysis . Cells were clamped at −80 mV and pulsed for 20 ms from −60 mV to +60 mV in 5 mV steps when recording voltage-gated Ca2+ currents or clamped at 0 mV and applying ramps from −100 mV to +100 mV ( 400 ms ) at 0 . 2 Hz to record TRPM5 currents . Cells were plated onto glass coverslips , loaded with 5 μM of Fura-2AM for 30 min at room temperature , washed out thoroughly and bathed in an isotonic solution containing ( in mM ) : 140 NaCl , 2 . 5 KCl , 1 . 2 CaCl2 , 0 . 5 MgCl2 , 5 glucose , 10 HEPES ( 305 mosmol/l , pH 7 . 4 adjusted with Tris ) . Ca2+-free solutions were obtained by replacing CaCl2 with equal amount of MgCl2 plus 0 . 5 mM EGTA . ATP was added to the bath solution as indicated in the figure legend . All experiments were carried out at room temperature as previously described ( Fernandes et al . , 2008 ) . AquaCosmos software ( Hamamatsu Photonics ) was used for capturing the fluorescence ratio at 505 nm obtained post-excitation at 340 and 380 nm . Images were computed every 5 s . | Goblet cells are specialized cells that produce proteins called mucins , which combine with water , salt and other proteins to form mucus , the slippery fluid that protects the respiratory and digestive tracts from bacteria , viruses and other pathogens . However , a defect in the production of one particular type of mucin—Mucin 5AC—can result in diseases such as cystic fibrosis , chronic obstructive pulmonary disease and Crohn’s disease , so there is a clear need to understand the production of mucus in detail . Before they are secreted , the mucins are packaged inside granules in the goblet cells . When a certain extracellular signal arrives at a goblet cell , these granules move through the cell , fuse with the cell membrane and release the mucins , which then expand their volume by a factor of up to a 1000 . Calcium ions ( Ca2+ ) have a critical role in the signal that leads to the secretion of mucins , but many details about the signalling and secretion processes are poorly understood . Now , Mitrovic et al . have used genetic methods to study 7343 gene products in goblet cells derived from a human colon . They identified 16 new proteins that are involved in the secretion of Mucin 5AC , including a channel protein called TRPM5 . This protein is activated when the concentration of Ca2+ inside the cell increases , and its activation allows sodium ( Na+ ) ions to enter the cells . These intracellular Na+ ions are then exchanged for Ca2+ ions from outside the cell , and these Ca2+ ions then couple to the molecular machinery that is responsible for the secretion of the mucins . By using electrophysiological and Ca2+ imaging approaches , Mitrovic et al . were able to visualize and measure TRPM5-mediated Na+ currents and the subsequent Ca2+ uptake by the cells , and confirmed that extracellular Ca2+ ions were responsible for stimulating the secretion of mucins . The next step is to determine how the other 15 genes are involved in mucin secretion and , in the longer term , explore how these insights might be translated into treatments for cystic fibrosis and other conditions associated with defective mucus secretion . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"cell",
"biology"
] | 2013 | TRPM5-mediated calcium uptake regulates mucin secretion from human colon goblet cells |
Brief experiences while a memory is consolidated may capture the consolidation , perhaps producing a maladaptive memory , or may interrupt the consolidation . Since consolidation occurs during sleep , even fleeting experiences when animals are awakened may produce maladaptive long-term memory , or may interrupt consolidation . In a learning paradigm affecting Aplysia feeding , when animals were trained after being awakened from sleep , interactions between new experiences and consolidation were prevented by blocking long-term memory arising from the new experiences . Inhibiting protein synthesis eliminated the block and allowed even a brief , generally ineffective training to produce long-term memory . Memory formation depended on consolidative proteins already expressed before training . After effective training , long term memory required subsequent transcription and translation . Memory formation during the sleep phase was correlated with increased CREB1 transcription , but not CREB2 transcription . Increased C/EBP transcription was a correlate of both effective and ineffective training and of treatments not producing memory .
New experiences that occur while a memory is being consolidated interact with the consolidation . The interactions can lead to problems in keeping apart related memories . This problem is partially solved by deferring some of the memory consolidation to sleep , a period in which new experiences are minimized . However , experiences when animals are wakened from sleep are likely to interact with the ongoing consolidation , thereby potentially causing maladaptive memories . We have found that during the sleep phase consolidation , new experiences produce a protein synthesis dependent process that blocks the formation of new memories , thereby preventing the formation of maladaptive memories . Some of the potential problems that arise from successive experiences have been known for many years . An experience that occurs while a memory is being consolidated may interfere with the consolidation , and partially stops it . Indeed , the process of memory consolidation was discovered over 100 years ago by interference of memory formation as a result of a new learning trial ( Lechner et al . , 1999 ) . Only much later was consolidation shown to depend on gene transcription and translation ( Montarolo et al . , 1986; Sweatt , 2010 ) , and dependence on these molecular processes has become the defining feature of consolidation . A second interaction between an experience and an ongoing consolidation has been more recently discovered . An experience that is generally ineffective in producing long-term memory can become effective when it occurs during the consolidation of a memory that arises from another , effective experience . This phenomenon was originally described on a synaptic level , and was called synaptic tagging and capture , since training at synapses that have undergone short-term plasticity produces local tags ( Frey and Morris , 1997; Martin et al . , 1997 ) . These tags can capture molecular products that are synthesized as a result of a consolidation process that arises from another training . Capture at the tags causes expression of long-term memory at synapses that would otherwise only display short-term memory . More recent studies have demonstrated behavioral tagging that is analogous to the synaptic tagging ( Ballarini et al . , 2009; de Carvalho Myskiw et al . , 2013; Moncada and Viola , 2007; Moncado et al . , 2015 ) . The problem of potentially non-adaptive interactions between memory consolidation and new experiences may be particularly acute when animals sleep . There is now abundant evidence that after completion of the early stages of consolidation which follow an experience , an additional stage of consolidation occurs while animals sleep ( Rasch and Born , 2013; Walker and Stickgold , 2006 ) , primarily during the night for diurnally active animals , or during the day for nocturnally active animals . Similar processes may govern consolidation that immediately follows training , and that during sleep . Memory consolidation and cortical plasticity during sleep are also accompanied by , or is dependent on protein synthesis ( Grønli et al . , 2013; Rasch and Born , 2013; Seibt et al . , 2012; Tudor et al . , 2016; Walker , 2005 ) . There is also a general increase in cerebral protein synthesis during sleep phases associated with memory consolidation ( Ramm and Smith , 1990 ) . In addition , molecular events associated with memory consolidation during sleep are similar to those induced during consolidation following training ( e . g . , Luo et al . , 2013 ) . Consolidation-like processes occur during sleep even in the absence of a previous training trial . Animals and humans sleep every day , even when no overt training occurred in the previous active phase . Proteins supporting consolidation will be synthesized during sleep , even when sleep does not follow overt training ( Ramm and Smith , 1990 ) . As in consolidation that follows training , new experiences during the sleep-phase may cause non-adaptive interaction with the consolidation . A transient experience that would not cause long-term memory could be amplified by the ongoing consolidation , and cause an unwanted long-term memory . The amplification of the experience could then in turn increase the possibility that the new experience will interfere with consolidation of the earlier memory during sleep . One way of preventing interference is to prevent new experiences by losing consciousness . Indeed , the earliest papers ( Jenkins and Dallenbach , 1924 ) showing that sleep improves memory posited that it does so passively , by preventing interference from new experiences . In recent years it has become clear that active processes also strengthen memories while animals sleep ( Rasch and Born , 2013; Walker and Stickgold , 2006 ) . However , the presence of active processes promoting consolidation during sleep strengthens the need to limit experiences , or the potential memories elicited by such experiences , while memories are consolidated . Animals and humans often wake from sleep , and experiences during such times should interfere with consolidation . However , paradoxically , training during the inactive phase , when animals are awakened from sleep , is often relatively ineffective in producing memory ( Lyons et al . , 2005; Michel and Lyons , 2014; Page , 2015; Rawashdeh et al . , 2007 ) . We explored the possibility that training during the sleep phase initiates an active block of learning , as a means of preventing such inappropriate learning from being consolidated , or of interfering with consolidation . If such a block is present , one would predict that if the block is removed , experiences adjacent to sleep should become more effective in initiating long-term memory . Under this condition , experiences that do not cause long-term memory when animals are active should cause long-term memory . To explore mechanisms by which new experiences during the period when animals generally sleep are prevented from interfering with consolidation , we utilized an associative learning paradigm in Aplysia in which animals learn that a tasty food cannot be swallowed ( Botzer et al . , 1998; Katzoff et al . , 2002; 2006 , Lyons et al . , 2005; Michel et al . , 2013; Susswein et al . , 1986 ) . Training produces a gradual decrease in responsiveness to food , and animals stop responding after 10–25 min of training . Training during the active phase of the day produces short , intermediate , long-term ( Botzer et al . , 1998; Michel et al . , 2013 ) , and persistent ( Levitan et al . , 2012; Schwarz et al . , 1991 ) memories , with the long-term memory dependent on protein synthesis adjacent to the training and persistent memory dependent on a somewhat later round of protein synthesis ( Levitan et al . , 2010 ) . Training during the inactive , sleep phase of the day is ineffective in producing long-term memory ( Lyons et al . , 2005 ) . We now report that training during the sleep phase can be made effective by treatment with a short-acting inhibitor of protein synthesis . This treatment permits even experiences that are too brief to cause long-term memory to become effective . Thus , protein synthesis following training has opposite effects on memory resulting from training during the active or the sleep phase , blocking the former while enabling the latter . Our findings indicate that training during the sleep phase initiates an active protein-synthesis dependent process that blocks the formation of new long-term memories . In addition , when the active block is removed , long-term memory formation is no longer dependent on protein synthesis at the time of the training . However , long-term memory formation is dependent on transcription and translation a number of hours after training . We examined a number of molecular correlates of long-term memory formation during the sleep phase . Changes were examined in the transcription of Aplysia CREB1 , CREB2 and C/EBP , molecules that are associated with memory formation ( Alberini , 2009 ) . We found no changes in CREB2 expression . Treatments that did and that did not cause memory formation led to increased C/EBP expression , indicating that C/EBP expression is not sufficient for memory formation . By contrast , only a treatment causing long-term memory caused a significant increase in the expression of CREB1 , indicating that this increase is a correlate of memory formation .
It was first important to show that Aplysia consolidate memory during sleep . Similar to almost all animals , Aplysia display a circadian rhythm of activity . They spend most of their inactive phase immobile , and relatively unresponsive to external stimuli ( Kupfermann , 1968; Strumwasser , 1971 ) . Recent work has shown that immobility during the inactive phase has characteristics of sleep ( Vorster et al . , 2014 ) . In mammals , memory consolidation occurs in part during the inactive phase , while animals sleep ( Rasch and Born , 2013; Walker and Stickgold , 2006 ) . Since consolidation generally depends on protein synthesis ( Montarolo et al . , 1986; Sweatt , 2010 ) , an effective way to examine whether consolidation occurs during sleep is to block protein synthesis at this time , and then determine whether this treatment blocks memory formation . To this end , Aplysia were trained with inedible food during the active period , and were treated with 10 µM of the protein synthesis inhibitor anisomycin 12 hr later , during the sleep phase . A previous study had already shown that injecting 10 µM anisomycin 10 min before training during the active phase blocked the expression of memory 24 hr later ( Levitan et al . , 2010 ) . Thus , consolidation occurs during the hours following training . However , 10 µM anisomycin applied several hours after training did not block 24 hr memory ( Levitan et al . , 2010 ) . Before treatment with anisomycin , we confirmed that all of the animals were immobile , and presumably were asleep . Treatment with 10 µM anisomycin , but not with the vehicle ( artificial seawater - ASW ) 12 hr after training blocked 24 hr memory ( Figure 1A ) . Thus , after a period in which protein synthesis is not required , 24 hr memory again depends on protein synthesis , during the inactive phase , when Aplysia sleep . This experiment confirms that active memory consolidation occurs during sleep in Aplysia , and that consolidation can be interrupted by blocking protein synthesis . 10 . 7554/eLife . 17769 . 003Figure 1 . Memory is consolidated , but training is ineffective , during the sleep phase . In this and in subsequent figures , values from measurements obtained during the inactive and active phases are respectively shaded and unshaded . In this and subsequent figures , upper bars show the time to stop responding . Lower bars show the percent change in the response ( - ( ( Train-Test ) /Train ) *100 ) , which is a measure of memory . In all figures standard errors are shown . Experiments in this figure were performed on the diurnally active A . californica . ( A ) Blocking protein synthesis during the sleep phase blocks 24 hr memory . Animals were trained during their active phase ( day ) , and were tested 24 hr later . 12 hr after training , during the sleep phase , animals were injected with ASW ( N = 9 ) or with 10 µM anisomycin ( N = 12 ) . ( 1 ) There is a significant decrease in the time to stop responding to the food during the 24 test ( p=0 . 02 , t = 2 . 76 , df = 8; two tailed paired t-test ) , indicating memory . ( 2 ) There was no significant difference between time to stop responding during the original training , and the test 24 hr later ( p=0 . 21 , t = 1 . 34 , df = 11; two tailed paired t-test ) , indicating that block of protein synthesis in the sleep phase following training blocks memory consolidation . ( 3 ) In addition to consolidation that follows training ( not shown ) , a second phase of consolidation occurs during sleep . This phase is blocked by the translation blocker anisomycin , preventing the expression of long-term memory . ( B ) Training during the sleep phase is ineffective in causing long-term memory . ( 1 ) Training animals during the sleep phase ( N = 5 ) did not lead to long-term memory , as shown by a lack of savings when animals were tested 24 hr later ( p=0 . 3 , t = 1 . 12 , df = 4; two-tailed paired t-test ) . In this experiment , animals were treated with ASW just before the training . ( 2 ) Memory after training during the active phase is expressed during the sleep phase ( N = 11 ) , as shown by a significant reduction in the time to stop 36 hr after training , when animals are tested during the inactive phase ( p=0 . 013 , t = 3 . 03 , df = 10; two-tailed paired t-test with Bonferroni correction ) . ( 3 ) A diagram showing that training during the sleep phase is ineffective in producing memory . ( C ) Effect of a brief recall during the sleep phase . ( 1 ) A 3 min training during the sleep phase ( N = 7 ) , which is an effective means of recalling a memory , leaves the memory intact ( p=0 . 001 , t ( 6 ) = 6 . 02 , two-tailed paired t-test ) . ( 2 ) A 3 min training paired with 10 µM anisomycin ( N = 6 ) rescues the memory that would have been blocked by the anisomycin alone ( p=0 . 004 , t ( 5 ) = 5 . 01 , two-tailed paired t-test ) . ( 3 ) A flow diagram showing that the anisomycin does not block memory formation when followed by a brief training . DOI: http://dx . doi . org/10 . 7554/eLife . 17769 . 00310 . 7554/eLife . 17769 . 004Figure 1—source data 1 . Memory is consolidated , but training is ineffective , during the sleep phase . DOI: http://dx . doi . org/10 . 7554/eLife . 17769 . 004 Previous studies examined the efficacy of training Aplysia during the inactive , sleep phase ( Fernandez et al . , 2003; Lyons et al . , 2006 ) . Using a number of learning paradigms , including learning with inedible food ( Lyons et al . , 2005 ) , training during the inactive sleep phase was ineffective in producing memory tested 24 hr later . We confirmed that training with inedible food during the sleep phase ( 3–6 hr after the change in lighting ) does not produce 24 hr memory ( Figure 1B1 ) . In this experiment , all animals were inactive and presumably asleep before being awakened and trained . The inability to detect 24 hr memory after training during the sleep period reflects an inability to form a new memory , rather than an inability to retrieve the memory . As shown previously ( Lyons et al . , 2005 ) , memory is readily retrieved and expressed during the inactive period , as demonstrated by the presence of memory at night 36 hr after training during the active period ( Figure 1B2 ) . This experiment indicates that a blocker of memory formation is active during the sleep phase . The blocker prevents experiences that would have given rise to memory during the active phase from doing so during the sleep phase ( Figure 1B1 ) . Ineffective experiences during consolidation can capture products synthesized by the consolidation process and become effective in producing long-term memory . Consolidation processes during sleep follows training with inedible food . We tested whether an abbreviated training that does not give rise to memory will do so when experienced during sleep phase consolidation . In a previous study , we showed that a brief 3 min training with inedible food is too short to elicit memory formation ( Levitan et al . , 2010 ) . However , a 3 min training is meaningful to the animals , in that it retrieves a memory . When the memory is retrieved after it has already been consolidated , the retrieval causes the memory to become labile , so that inhibition of protein synthesis can again block memory ( Dudai , 2004; Nader , 2003; Nader et al . , 2000 ) . The retrieval produces an additional consolidation-like state that has been termed reconsolidation ( Dudai , 2004 ) . Previous work showed that 24 hr after training , when a 3 min abbreviated training ( reminder ) is paired with 10 µM anisomycin , 48 hr memory is blocked ( Levitan et al . , 2010 ) . We tested whether a 3 min training during sleep phase consolidation could rescue a memory that would otherwise have been erased by treatment with 10 µM anisomycin . To examine this possibility , after a full training ( training until animals stop responding to food ) during the active phase , animals received a 3 min recall ( training with inedible food stopped after 3 min ) 12 hr later , during the sleep phase . Animals were injected with 10 µM anisomycin 10 min before the 3 min training . Controls were injected with ASW . Memory was retained after the 3 min training with ASW ( Figure 1C1 ) . In addition , the 3 min reminder training with 10 µM anisomycin was effective in protecting the memory ( Figure 1C2 ) , so that the anisomycin treatment no longer blocked the consolidation . Thus , the 3 min training rescued the memory that would otherwise have been blocked by the anisomycin treatment . This experiment could be explained in two ways . The two explanations are not mutually contradictory: ( 1 ) the brief training blocked the effect of the anisomycin , and therefore the memory that was initiated by the training during the active phase was retained; ( 2 ) the brief training initiated a de novo memory process . Although a brief training is insufficient to produce memory during the active phase , it produces memory during the sleep phase , in the presence of anisomycin . If this hypothesis is correct , the 3 min training may be able to produce long-term memory even if it is not preceded by a full training during the active phase . Consolidation-like activity occurs during sleep , even in the absence of an overt training during the previous active phase . Is the previous active phase training required for the brief training during the sleep phase with anisomycin to be effective in producing long-term memory ? We tested the possibility that a 3 min training during the sleep phase after treatment with 10 µM anisomycin produces long-term memory , even without a previous active phase training . Naïve animals were trained for 3 min . They were injected with 10 µM anisomycin 10 min before the training . Memory was tested 24 hr later . For comparison , memory after training was compared to that before and after a non-abbreviated training session during the active period . Animals tested 24 hr after the brief training with anisomycin during the sleep phase responded like animals that had been previously trained during the active phase ( Figure 2A , B ) . Thus , the combination of 3 factors which each alone is either ineffective in producing long-term memory ( training during the sleep phase , training with a protein synthesis inhibitor , training stopped after 3 min ) , caused memory when combined . 10 . 7554/eLife . 17769 . 005Figure 2 . 3 min training with anisomycin during the inactive phase produces memory . Experiments were on A . californica . ( A ) Time to stop responding during training in the active phase , and during a test of memory 24 hr later ( N = 10 ) . These data provide comparisons for data in B-E on the time to stop in a naïve , previously untrained animals , and during a test of memory after successful training . ( B ) Animals were trained for 3 min during the inactive phase just after treatment with anisomycin , and memory was tested 24 hr later ( N = 15 ) . ( C ) As a control , animals were trained for 3 min during the inactive phase just after treatment with ASW , and memory was tested 24 hr later ( N = 7 ) . ( D , E ) To test whether memory is expressed during the active phase , animals were trained for 3 min during the inactive phase just after treatment with anisomycin ( N = 9 D ) , or ASW ( N = 11 E ) , and memory was tested 12 hr later . There were significant differences between the six groups tested ( training and testing in part A , and the four tests of memory in parts B–E ) ( p=0 . 00005 , F ( 5 , 54 ) = 6 . 95; one-way analysis of variance ) . A post hoc- test ( Student-Newman-Keuls , α = 0 . 05 ) showed no significant difference between naïve animals trained during the day and either group of animals trained for 3 min with ASW ( marked by an n for behaving as if naive ) . Thus , a 3 min training during the inactive phase with ASW produces no memory . These 3 groups were significantly different from the other three groups ( marked by a t for behaving as if trained ) , which were not significantly different from one another . Thus , a 3 min training during the inactive phase after anisomycin treatment produced memory 12 and 24 hr later . ( F ) The data are explained by the effects of 10 µM anisomycin on a process initiated by sleep phase training that blocks memory . The anisomycin prevents the action of the blocker , allowing the formation of long-term memory . DOI: http://dx . doi . org/10 . 7554/eLife . 17769 . 00510 . 7554/eLife . 17769 . 006Figure 2—source data 1 . 3 min training with anisomycin during the inactive phase produces memory . DOI: http://dx . doi . org/10 . 7554/eLife . 17769 . 006 Would the 3 min training during the sleep phase produce long-term memory even without the treatment with 10 µM anisomycin , or is the long-term memory dependent on the anisomycin treatment ? Training animals for 3 min during the sleep phase after ASW treatment , rather than after 10 µM anisomycin , produced no memory ( Figure 2C ) . We also tested the possibility that the brief training with anisomycin caused memory that is only expressed during the sleep phase . Animals were tested 12 hr after the brief training with 10 µM anisomycin , or with ASW . The 3 min training produced memory 12 hr later only when training was with 10 µM anisomycin ( Figure 2D ) , but not with ASW ( Figure 2E ) . The findings suggest that training during the sleep phase initiates a protein synthesis dependent block of long-term memory formation . When the block is removed by the anisomycin treatment , learning is more effective during the sleep phase than during the active phase . Even an experience that is too brief to cause long-term memory during the active phase , a 3 min training , is effective during the sleep phase ( Figure 2F ) . In addition , memory is expressed earlier . It was previously shown ( Botzer et al . , 1998 ) that after effective training during the day , long-term memory is expressed only 24 hr after training , and no memory is expressed at 12 hr after the training . However , 12 hr after a 3 min training with 10 µM anisomycin during the inactive phase , memory is expressed . We have suggested that the 3 min training during sleep is effective in producing memory because of the ongoing consolidation during sleep , and the lack of a protein synthesis dependent blocker of memory formation . However , it is conceivable that the 10 µM anisomycin treatment per se makes the 3 min training effective , even without the consolidation process . Although the 3 min training alone during the active phase is ineffective in producing memory ( Levitan et al . , 2010 ) , the combination of a 3 min training and treatment with 10 µM anisomycin might produce memory , even during the active phase . We tested this possibility . However , we found that the 3 min training 10 min after 10 µM anisomycin treatment during the active phase was ineffective in producing long-term memory ( data not shown ) . Thus , this combination is effective in producing memory only during the sleep phase , presumably because of the ongoing consolidation process . Does the ongoing consolidation in the presence of 10 µM anisomycin facilitate only a brief training , or will it also facilitate a longer training , which is effective during the active phase , but not during the sleep phase ? We examined this possibility by training animals during the sleep phase until they stopped responding ( full training ) , and testing memory 24 hr later . They were treated 10 min before training with either 10 µM anisomycin ( Figure 3A ) , or with ASW ( Figure 3B ) . A third group was treated with 10 µM anisomycin , but was not trained ( Figure 3C ) . Only the animals that were trained after treatment with 10 µM anisomycin displayed 24 hr memory . This experiment shows that both an abbreviated and a longer training session are effective in producing memory after treatment with 10 µM anisomycin . In addition it shows that anisomycin alone during the sleep phase is ineffective in producing memory without training . 10 . 7554/eLife . 17769 . 007Figure 3 . Full training during the sleep phase with anisomycin causes long-term memory . Experiments were performed on A . californica . ( A ) Training animals until they stop responding during the inactive phase 10 min after treatment with anisomycin ( N = 7 ) leads to memory when animals are tested 24 hr later , as shown by a decrease in the time stop responding ( p=0 . 05 , t = 2 . 49 , df = 6; two-tailed paired t-test ) . ( B ) Training animals until they stop responding during the inactive phase 10 min after treatment with ASW ( N = 5 ) did not lead to long-term memory , as shown by a lack of savings when animals were tested 24 hr later ( p=0 . 3 , t = 1 . 12 , df = 4; two-tailed paired t-test ) . ( C ) Anisomycin treatment alone during the inactive phase ( N = 7 ) does not produce memory 24 hr later ( p=0 . 74 , t = 0 . 34 , df = 16 – data after anisomycin treatment were compared to data for naïve animals trained during the day in Figure 1B2 , which were trained along with these animals . DOI: http://dx . doi . org/10 . 7554/eLife . 17769 . 00710 . 7554/eLife . 17769 . 008Figure 3—source data 1 . Full training during the sleep phase with anisomycin causes long-term memory . DOI: http://dx . doi . org/10 . 7554/eLife . 17769 . 008 Injecting 10 µM anisomycin during the sleep phase allowed training procedures that would have been ineffective in producing long-term memory to become effective . Our interpretation is that a protein-synthesis dependent blocker of memory is inhibited by the anisomycin , and therefore is not present to suppress memory formation . However , consolidation leading to long-term memory generally requires protein synthesis . Even if it inhibited synthesis of the blocker , why was long-term memory formed when protein synthesis was blocked ? We suggest that the proteins having a role in memory consolidation are either specifically expressed during sleep , or are expressed more strongly during sleep . These proteins are expressed even in the absence of an explicit training during the previous active phase , and their presence substitutes for the proteins synthesized just after training during the active phase . These proteins are synthesized throughout the sleep phase , and in our experiments they were already expressed when animals were trained later in the sleep phase . Their expression explains why training can produce long-term memory with a protein synthesis inhibitor . This hypothesis suggests that temporally separated processes of protein synthesis explain our findings . Protein synthesis functioning in consolidation begins within the first hours of the sleep phase , and these proteins can enhance memory formation from new experiences during the sleep phase , explaining why even a brief training can produce long-term memory . However , the new experiences themselves initiate an additional process of protein synthesis that blocks the formation of new memories , explaining why training during the sleep phase is generally ineffective in causing long-term memory . Treatment with 10 µM anisomycin just before training blocks synthesis of the memory blocker , leaving intact the previously synthesized proteins associated with consolidation that enhance memory formation ( see Figure 5E ) . We tested this hypothesis by separately blocking the protein synthesis associated with consolidation , and the protein synthesis that blocks memory formation . We predict that blocking the protein synthesis associated with consolidation will block long-term memory formation during the sleep phase , even if the memory blocker induced by training is also not present . Since consolidation is likely to occur throughout the sleep phase , a short-acting blocker of protein synthesis that is applied early in the sleep phase may block the proteins that support consolidation , but will no longer be present at a later time , when animals are trained . In intact animals , the length of time that anisomycin effectively blocks protein synthesis depends on the concentration used . Smaller doses of anisomycin transiently block protein synthesis , whereas larger doses have longer-lasting effects ( Flood et al . , 1973 ) . To find a transiently active dosage of anisomycin , we examined the effects of different concentrations at different times before a full training session during the active phase . We found that 10 µM anisomycin was effective in blocking memory formation 1 hr before training ( Figure 4A ) , but was not effective 2 hr before training ( Figure 4B ) . By contrast , a three times larger dose ( 30 µM anisomycin ) was still effective in blocking memory formation when injected 2 . 5 hr before training ( Figure 4C , D ) . We also tested the effect of a dosage of anisomycin that is a third of that used previously ( 3 . 3 µM in place of 10 µM anisomycin ) , and found that this concentration injected 10 min before training is too low to inhibit long-term memory formation ( Figure 4E ) . Thus , 10 µM anisomycin is close to threshold in blocking protein synthesis , and its effect is relatively short-lasting . Injection of this concentration early in the sleep phase , 2 hr or more before training , will not affect synthesis of the proposed memory blocker , but will affect synthesis of proteins that support consolidation during sleep . 10 . 7554/eLife . 17769 . 009Figure 4 . Length of protein synthesis inhibition is concentration dependent . Experiments in parts A and B were performed on A . fasciata; Experiments in parts C-E were performed on A . californica . ( A ) 10 µM anisomycin injected 1 hr before training ( N = 5 ) blocks 24 hr memory , as shown by no significant change in the time to stop between the training session and the 24 hr test of memory ( p=0 . 66 , t = 0 . 48 , df = 4; two-tailed paired t-test ) ( B ) 10 µM anisomycin injected 2 hr before training ( N = 6 ) does not block 24 hr memory , as shown by a significant change in the time to stop between the training session and the 24 hr test of memory ( p=0 . 006 , t = 4 . 66 , df = 5; two-tailed paired t-test ) . Thus , the actions of 10 µM anisomycin are short-acting , and this concentration is no longer effective after 2 hr . ( C ) 30 µM anisomycin injected 2 . 5 hr before training ( N = 6 ) blocks memory , as shown by no significant change in the time to stop between the training session and the 24 hr test of memory ( p=0 . 37 , t = 0 . 99 , df = 5; two-tailed paired t-test ) . ( D ) In a group of animals collected and tested along with those in C , 10 µM anisomycin 2 . 5 hr before training ( N = 3 ) did not block memory ( p=0 . 3 , t = 5 . 45 , df = 2 ) . Thus , 30 µM anisomycin produces a longer-lasting inhibition than does 10 µM anisomycin . For parts C and D , note that the initial training time is shorter than in parts A and B . This is because the animals were young and collected from the ocean during the beginning of the season and were also trained and tested at a different time . Baseline values often change in different batches of animals collected at different times . ( E ) 3 . 3 µM anisomycin injected 10 min before training ( N = 7 ) is ineffective in blocking memory , as shown by a significant decrease in the time to stop between the training session and the 24 hr test of memory ( p=0 . 01 , t = 3 . 53 , df = 6 , two-tailed paired t-test ) . DOI: http://dx . doi . org/10 . 7554/eLife . 17769 . 00910 . 7554/eLife . 17769 . 010Figure 4—source data 1 . Length of protein synthesis inhibition is concentration dependent . DOI: http://dx . doi . org/10 . 7554/eLife . 17769 . 010 Aplysia were treated with 10 µM anisomycin 1 to 3 hr after the change in lighting that signaled the start of the sleep phase , but 2 hr before training began ( Figure 5A ) . A second group of animals was treated with 10 µM anisomycin 30 min before the training ( Figure 5B ) . Animals in both groups received a second injection of 10 µM anisomycin 10 min before a 3 min training . Our expectation is that the treatment 2 hr before training will block the synthesis of proteins associated with consolidation during sleep , without affecting the synthesis of the memory blocker . Thus , if sleep-phase proteins functioning in consolidation are required for long-term memory formation during the sleep phase , this treatment should block memory formation . By contrast , the injection 30 min before training will leave intact proteins that have already been synthesized in the early hours of the sleep phase , before the anisomycin was injected . 10 . 7554/eLife . 17769 . 011Figure 5 . Block of sleep-time expression of proteins blocks memory formation . These experiments were on A . fasciata . In addition to injecting animals with 10 µM anisomycin treatment 10 min before a 3 min training , the animals were also injected with 10 µM anisomycin A ) 2 hr before the training ( N = 8 ) , or B ) 0 . 5 hr before the training ( N = 8 ) . Values 24 hr after these procedures were compared to values in C ) 24 hr after a 3 min training with a single injection of 10 µM anisomycin 10 min before training ( N = 15 ) . A one-way analysis of variance between the three groups showed significant differences ( p=0 . 004 , F ( 2 , 28 ) = 6 . 90 ) . ( A ) Injecting 10 µM anisomycin 2 hr before training blocked memory formation ( p=0 . 003 , t = 3 . 62 , df = 21; two-tailed t-test with Bonferroni correction – comparison of A to C ) , presumably because proteins that accompany consolidation during sleep were blocked . ( B ) 10 µM anisomycin treatment 0 . 5 hr before a 3 min training plus 10 µM anisomycin did NOT block memory formation ( p=0 . 52 , t = 0 . 66 , df = 21; two-tailed t-test – comparison of B to C ) , presumably because sleep-phase proteins having a role in consolidation are still present at the time of the training . ( C ) As shown in the previous experiments , a 3 min training during the inactive phase10 min after 10 µM anisomycin produced 24 hr memory . ( D ) To provide a comparisons for data in B–D , the time to stop responding during the active phase in naïve , previously untrained animals was also measured ( N = 6 ) . ( E ) A diagram showing that translation relevant to consolidation during sleep is necessary for the expression of long-term memory after the brief training with anisomycin . DOI: http://dx . doi . org/10 . 7554/eLife . 17769 . 01110 . 7554/eLife . 17769 . 012Figure 5—source data 1 . Block of sleep-time expression of proteins blocks memory formation . DOI: http://dx . doi . org/10 . 7554/eLife . 17769 . 012 Treatment with 10 µM anisomycin 2 hr before training blocked the formation of long-term memory ( Figure 5A ) , as shown by a significant increase in the response time 24 hr later , with respect to animals that were treated with anisomycin only once , 10 min before training ( Figure 5C ) . By contrast , treatment with 10 µM anisomycin 30 min before training did not block the formation of long-term memory ( Figure 5B ) , as shown by a significant decrease in response , with respect to the response time of naïve animals trained during the active phase ( Figure 5D ) . These findings support the hypothesis that proteins synthesized during sleep substitute for the proteins that are generally synthesized as a result of long term training , and that presence of these proteins is required for long-term memory formation during the sleep phase ( Figure 5E ) . This experiment also indicates that the double injection of anisomycin per se did not block memory formation , since animals injected 30 min before training , and again 10 min before training , displayed long-term memory . These findings show that 10 µM anisomycin 2 hr before training has opposite effects during the active and inactive phases of activity . During the active phase , this treatment does not block long-term memory formation , presumably because its concentration within the animal decreases with time , and becomes too low to block the protein synthesis initiated by the training . By contrast , during the sleep phase it blocks long-term memory formation , presumably because it blocks the background expression of molecules that support memory consolidation during sleep . A hallmark of long-term memory is dependence on gene transcription and translation ( Montarolo et al . , 1986 ) . However , we have shown that memory formation at night only occurs when protein synthesis is blocked just preceding the training , although protein synthesis in the hours preceding training is necessary for long-term memory formation . Consolidation is a slow process , and is dependent on a number of waves of mRNA and protein synthesis ( Barzilai et al . , 1989 ) . The effects of 10 µM anisomycin are relatively short-lasting ( see Figure 4 ) . After training , one might still expect that long-term memory formation will depend on molecular signals initiating transcription of mRNAs , and their subsequent translation , after the effects of anisomycin present during or just after training have worn off . To determine whether memory formation is dependent on transcription , we examined the effects of the reversible transcription blocker 5 , 6-Dichlorobenzimidazole 1-β-D-ribofuranoside ( DRB ) on memory formation . After treating animals with anisomycin , they were trained 10 min later for 3 min , and then treated with DRB . Memory was then tested 24 hr later . DRB is not water soluble , and DMSO injection inhibits feeding in intact , behaving Aplysia ( unpublished observation ) . We therefore used ethanol to dissolve DRB . To rule out the possibility that ethanol , or ethanol with DRB , damages animals and affects the ability to respond to food , we first examined the effects of ethanol alone and of DRB dissolved in ethanol on learning 24 hr after the injections , when their effects had worn off , and on memory 48 hr after the injection . Neither ethanol ( Figure 6A1 ) nor ethanol plus DRB ( Figure 6A2 ) affected the ability to train animals 24 hr after the treatment , or to obtain memory 24 hr after the training . 10 . 7554/eLife . 17769 . 013Figure 6 . Learning during the inactive phase is transcription dependent . Experiments were on A . californica . ( A ) Ethanol is benign . Animals were treated with ( 1 ) ethanol alone ( N = 3 ) , or with ( 2 ) DRB dissolved in ethanol ( N = 4 ) , and were trained 24 hr later . Memory was then tested after 24 hr ( 48 hr after ethanol or DRB treatment ) . Neither treatment affected the ability of animals to learn or remember that food was inedible 24 hr later ( For animals tested with ethanol: p=0 . 009 , t = 10 . 22 , df = 2; for animals tested with DRB: p=0 . 03 , t = 3 . 57 , df = 3 ) . ( B ) DRB blocks memory . ( 1 ) Control: Animals were not trained . They were treated with anisomycin and then treated with DRB dissolved in ethanol , and then tested after 24 hr ( N = 6 ) . ( 2 ) Vehicle: Animals were treated with anisomycin , and then trained for 3 min . After training , the animals were injected with ethanol , with no DRB . Memory was then tested 24 hr later ( N = 7 ) . ( 3 ) DRB: Animals were treated with anisomycin , and then trained for 3 min , and then injected with DRB dissolved in ethanol . Memory was then tested 24 hr later ( N = 8 ) . There were significant differences between the 3 groups tested ( p=0 . 04 , F ( 2 , 18 ) = 3 . 73 ) . For the 2 groups of trained animals , there was a significant difference in the time to stop 24 hr after the training ( comparison of 2 and 3 – p=0 . 027; t = 2 . 50 , df = 13; two-tailed t-test ) . There was no significant difference between the time to stop in animals treated with DRB , and in control animals that had not been trained ( comparison of 1 and 3 p=0 . 49; t = 0 . 71 , df = 12; two-tailed t-test ) . Thus , the DRB blocked memory that would have been formed after training with anisomycin . ( C ) A diagram showing that after the brief training with 10 µM anisomycin , long-term memory depends on transcription and translation , The DRB blocks memory because it blocks transcription , and a 30 µM dose of anisomycin , which has a longer-lasting effect , blocks the transcription-dependent translation . DOI: http://dx . doi . org/10 . 7554/eLife . 17769 . 01310 . 7554/eLife . 17769 . 014Figure 6—source data 1 . Learning during the inactive phase is transcription dependent . DOI: http://dx . doi . org/10 . 7554/eLife . 17769 . 014 After a 3 min training at night with anisomycin , animals were injected with either ethanol ( Figure 6B2 ) , or DRB dissolved in ethanol ( Figure 6B3 ) . A control group also examined memory 24 hr after 10 µM anisomycin treatment at night , without subsequent training , followed by ethanol plus DRB ( Figure 6B1 ) . Significant memory was shown in animals treated with anisomycin alone , whereas no memory was shown in animals treated with anisomycin plus DRB . Thus , memory 24 hr after a 3 min training at night with anisomycin is dependent on transcription , and presumably also on later rounds of translation that occur after the effects of anisomycin have worn off ( Figure 6C ) . As an indicator of whether a later round of translation is required after training , we used a higher concentration of anisomycin ( 30 µM in place of 10 µM ) , and then trained animals for 3 min . The higher concentration presumably has a longer-lasting effect on protein synthesis ( Flood et al . , 1973 ) , accounting for a longer-lasting inhibition of memory formation when Aplysia were trained during the active phase ( see Figure 4C ) . Training with 30 µM anisomycin was ineffective in producing memory 24 hr later ( Figure 7A2 ) , presumably because it not only blocks protein synthesis transiently at the time of the training , but also blocks later rounds of synthesis that follow transcription ( e . g . , Barzilai et al . , 1989 ) ( Figure 6C ) . 10 . 7554/eLife . 17769 . 015Figure 7 . Dissociation between memory and C/EBP expression . Data are from A . californica . ( A ) Neither 30 µM ( N = 6 ) nor 3 . 3 µM ( N = 7 ) anisomycin injected 10 min before a brief training during the inactive phase produce 24 hr memory . The times to stop responding 24 hr after treatment with 30 µM anisomycin and 3 . 3 µM anisomycin are significantly longer than the time to stop 24 hr after 10 µM ( N = 5 ) anisomycin ( for 30 µM: p=0 . 005 , t = 5 . 98 , df = 9; for 3 . 3 µM: p=0 . 0002 , t = 6 . 42 , df = 10; two tailed t-tests with Bonferroni corrections ) . ( B ) C/EBP expression in the buccal ganglia 2 hr after injecting 10 µM anisomycin during the active ( N = 5 ) and inactive ( N = 6 ) phases increases , and after injecting 3 . 3 µM ( N = 6 ) anisomycin during the inactive phase . Data are normalized to C/EBP expression in animals that were injected with ASW ( not shown ) . Normalization was to the ASW treatment at the same time as the anisomycin treatment ( N = 6 during the active phase , N = 5 during the inactive phase ) . There were no significant differences in C/EBP expression after ASW treatment between active and inactive phases . Expression after ASW treatment was set at 100% . All treatments with anisomycin produced large ( X thousand percent ) increases in C/EBP expression , although only 10 µM anisomycin during the inactive phase produces long-term memory when paired with training ( For the 3 groups shown: p=0 . 139; F ( 2 , 14 ) = 2 . 27 . DOI: http://dx . doi . org/10 . 7554/eLife . 17769 . 01510 . 7554/eLife . 17769 . 016Figure 7—source data 1 . Dissociation between memory and C/EBP expression . DOI: http://dx . doi . org/10 . 7554/eLife . 17769 . 016 In addition to blocking protein synthesis , anisomycin also increases MAP-kinase activity , and thereby causes a large increase in CCAAT enhancer-binding protein ( C/EBP ) mRNA expression ( Alberini et al . , 1994; Cano et al . , 1994 ) . Increased C/EBP is an early step in the formation of memory in Aplysia ( Alberini et al . , 1994 ) , and in other animals ( Alberini , 2009 ) . In addition , training with inedible food during the active phase caused a significant ( 5–10 fold ) increase in C/EBP mRNA expression in the buccal ganglia ( Levitan et al . , 2008 ) , with respect to C/EBP mRNA expression in controls whose lips had been stimulated with food , a procedure that does not lead to memory formation ( Schwarz et al . , 1988 ) . In another Aplysia learning paradigm , a tyrosine phosphatase inhibitor was used to increase MAP-kinase activity , and thereby also presumably increase C/EBP mRNA expression . This treatment alone was not effective in allowing long-term memory to be formed after sleep phase training ( Lyons et al . , 2006 ) . However , when this treatment was combined with a histone deacetylation inhibitor , the combination of both was effective in allowing long-term memory to be formed by sleep phase training ( Lyons et al . , 2006 ) . These finding raise the possibility that anisomycin during the sleep phase permits memory formation because it increases C/EBP expression , rather than because it inhibits the synthesis of a protein that actively blocks memory formation . We tested this possibility . Using quantitative real time polymerase chain reaction ( qRT-PCR ) , we found that injection of 10 µM anisomycin produced a large increase in the expression of C/EBP mRNA ( see Figure 7B ) . As shown above ( Figure 3C ) , 10 µM anisomycin treatment alone , with no training , has no effect on subsequent memory , indicating that increased C/EBP mRNA expression per se does not produce long-term memory . However , it is possible that increased C/EBP mRNA expression underlies the ability of a brief training during the sleep phase to produce long-term memory , rather than the block of protein synthesis produced by 10 µM anisomycin . We examined this possibility . Injection of 10 µM anisomycin 10 min before training during the active phase blocks memory , but during the sleep phase it facilitates memory . If increased C/EBP is responsible for the facilitation of long-term memory during sleep , there should be a difference in the C/EBP expression in the buccal ganglia after 10 µM anisomycin treatment during the active and sleep phases . Using qRT-PCR , we measured C/EBP mRNA expression in the buccal ganglia 2 hr after anisomycin treatment during both the active and the sleep phases . Expression was compared to that in response to animals injected with ASW , during the active and the sleep phases respectively . During both the active and sleep phases , there were large ( over 100-fold ) increases in C/EBP mRNA expression in response to 10 µM anisomycin ( Figure 7B ) , with no significant difference in expression between the two phases . Thus , C/EBP is increased when anisomycin blocks memory , as well as when it is required to form long-term memory . We also tested the effects of a 3 . 3 µM dose of anisomycin . We found that 3 . 3 µM anisomycin injected 10 min before training during the active phase did not block memory formation ( see Figure 4E ) . In addition , it did not facilitate memory formation during the inactive phase ( Figure 7A3 ) , presumably because the dose is too low to block protein synthesis sufficiently . However , 3 . 3 µM anisomycin still caused a large ( over 80-fold ) increase in C/EBP expression ( Figure 7B ) . The difference between C/EBP production during the sleep phase in response to 10 µM or 3 . 3 µM anisomycin was not significant . Thus , there is a dissociation between the effect of anisomycin on protein synthesis , which is likely to be critical for its effects on memory , and the effect of anisomycin on C/EBP , whose increase is not systematically associated with the ability to form long-term memory . Although increases in C/EBP caused by anisomycin treatment could not explain the ability to produce long-term memory , it is possible that increases in C/EBP are nonetheless correlated with training procedures that are effective in producing long-term memory . If this is so , training with 10 µM anisomycin at night should produce an increase in C/EBP over that produced by anisomycin alone , whereas training with ASW at night should produce no increase in C/EBP expression over that produced by ASW alone . We tested this possibility by measuring C/EBP expression 2 hr after a 3 min training with ASW or with 10 µM anisomycin . A 3 min training with ASW causes no long-term memory ( Figure 2C ) , but nonetheless caused a 7-fold increase in C/EBP expression ( Figure 8A1 ) , which is similar to the increase in C/EBP expression caused by training during the day ( Levitan et al . , 2008 ) . A 3 min training after treatment with 10 µM anisomycin also caused a significant increase in C/EBP expression ( Figure 8A2 ) . Note that the increased C/EBP expression after anisomycin treatment is 2 orders of magnitude larger than that after ASW treatment , because of the effect of the anisomycin . However , in both conditions training caused an increase in C/EBP expression . These data indicate that increased C/EBP mRNA expression is not a necessary correlate of memory formation . 10 . 7554/eLife . 17769 . 017Figure 8 . Increased C/EBP expression is a correlate of training , not of memory formation . ( A ) Effect of 3 min training during the inactive phase on C/EBP expression . Data are from the buccal ganglia of A . californica . ( 1 ) Training 10 min after injection with ASW ( Trained: N = 8; Naïve: N = 9 ) produced a significant increase in C/EBP mRNA expression ( p=0 . 004 , t = 3 . 43 , df = 15 ) , even though this treatment did not lead to long-term memory . ( 2 ) Training after injection with anisomycin ( Trained: N = 8; Naïve: N = 8 ) also produced a significant increase in C/EBP ( p=0 . 008 , t = 3 . 11 , df = 14; both tests are two-tailed unpaired t-tests ) , over that caused by the injection of anisomycin in naïve controls ( see Figure 6 ) Note that values for all 4 treatments are normalized to the value for naïve animals treated with ASW . As shown above , anisomycin alone caused a large increase in expression . ( B ) Effects of training in isolation on C/EBP expression . Data are from the buccal ganglia of A . fasciata , in which training during the active phase when animals are maintained in isolation does not produce long-term memory . Note that C/EBP expression in all four groups was normalized to expression in isolated , naïve animals , in which the mean value was set as 100% . ( 1 ) Training in A . fasciata maintained in isolation ( Trained: N = 10; Naïve: N = 10 ) produced a significant increase in C/EBP expression ( p=0 . 0009 , t = 4 . 27 , df = 18 ) , as did ( 2 ) training in animals housed with conspecifics ( Trained: N = 6; Naïve: N = 6 ) ( p=0 . 004 , t = 4 . 13 , df = 10 ) . Note that maintenance with conspecifics itself produced a small increase in C/EBP expression ( p=0 . 05 , t = 2 . 52 , df = 14; all tests are two-tailed unpaired t-tests with Bonferroni correction ) . DOI: http://dx . doi . org/10 . 7554/eLife . 17769 . 01710 . 7554/eLife . 17769 . 018Figure 8—source data 1 . Increased C/EBP expression is a correlate of training , not of memory formation . DOI: http://dx . doi . org/10 . 7554/eLife . 17769 . 018 The data above suggest that the increase in C/EBP expression seen after training , either during the active phase ( Levitan et al . , 2008 ) , when training causes memory , or during the inactive phase in which training does not cause memory , is a correlate of the training procedure per se , rather than being a correlate of memory formation . Additional evidence also supported this suggestion . When Aplysia fasciata are trained during their active phase , but in isolation from conspecifics , no long-term memory is produced ( Schwarz et al . , 1998 ) . We found that both training in isolation , and training in the presence of conspecifics produced significant increases in C/EBP expression ( Figure 8B ) . Thus , increased C/EBP expression is a correlate of the training procedure , rather than being a correlate of memory formation . In addition to increasing C/EBP expression , anisomycin could increase the expression of additional genes associated with memory formation , and thereby induce memory during training in the sleep phase . We examined the possibility that the expression of two additional genes associated with memory formation , cAMP- response element-binding proteins 1 and 2 ( CREB1 and CREB2 ) ( Lee et al . , 2008 ) , are changed by doses of anisomycin that facilitate memory formation . Since CREB2 is a repressor of memory formation ( Bartsch et al . , 1995 ) , regulation of CREB2 might be associated with the block of memory during the sleep phase . As noted above , 3 . 3 µM anisomycin does not block memory formation during the active phase , or facilitate memory formation during the inactive phase , whereas 10 µM anisomycin is effective in blocking memory during the active phase , and enhancing memory formation during the inactive phase . If 10 µM anisomycin were to significantly change the expression of one of these genes , over that expressed after treatment with 3 . 3 µM anisomycin , this change could underlie the ability of training during the inactive phase to cause long-term memory . Unlike the effect of anisomycin on C/EBP , neither 3 . 3 µM nor 10 µM anisomycin during the inactive phase produced significant increases in the expression of CREB1 or of CREB2 , with respect to expression after treatment with ASW , although both concentrations produced moderate increases that approached significance ( Figure 9A , and legend – note that the bars are normalized to expression after ASW , which is 100% ) . There were no significant differences in expression between treatment with 3 . 3 µM and 10 µM anisomycin ( Figure 9A ) , effectively ruling out changed expression of either of these two genes as a result of anisomycin treatment alone from being the underlying cause of memory formation with anisomycin treatment during the inactive phase . 10 . 7554/eLife . 17769 . 019Figure 9 . Increased CREB1 expression after training is a correlate of memory . Data are from the buccal ganglia of A . californica . ( A ) Increases in CREB1 or CREB2 by anisomycin treatment alone cannot account for memory formation . A preliminary analysis examined whether either CREB1 or CREB2 expression is increased by either 3 . 3 or 10 µM anisomycin 2 hr after treatment , with respect to expression 2 hr after injection with ASW ( not shown ) . A one-way analyses of variance showed no significant differences in CREB1 expression between ASW ( N = 6 ) , 10 µM ( N = 8 ) and 3 . 3 µM ( N = 6 ) anisomycin ( p=0 . 09 , F ( 2 , 17 ) = 2 . 77 ) or of CREB2 expression between these three treatments ( ASW: N = 6; 10 µM anisomycin: N = 7; and 3 . 3 µM: N = 5 ) anisomycin ) ( p=0 . 07 , F ( 2 , 15 ) = 3 . 10 ) 2 hr after the treatments . In addition , there were no significant differences between expression after treatment with 3 . 3 and 10 µM anisomycin ( For CREB1: p=0 . 29 , t = 1 . 1 , df = 12; for CREB2 , p=0 . 42 , t = 0 . 84 , df = 10 ) . ( B ) CREB1 , but not CREB2 , is a correlate of memory formation . ( 1 ) There was a significant increase ( p=0 . 006 , t = 3 . 24 , df = 14 ) in CREB1 expression 2 hr after a 3 min training with 10 µM anisomycin during the inactive phase ( N = 8 ) , with respect to naïve animals treated with anisomycin ( N = 8 ) . However , there was no significant increase in expression in animals treated with ASW ( p=0 . 97 , t = 0 . 03 , df = 15 . N = 8 for the trained group , N = 9 for the naïve group ) . ( 2 ) CREB2 expression was unaffected by training after either treatment ( For animals treated with ASW: p=0 . 71 , t = 0 . 37 , df = 15 , N = 8 trained animals and 9 untrained animals; for animals treated with 10 µM anisomycin: p=0 . 50 , t = 0 . 69 , df = 12 , N = 7 trained and 7 untrained animals ) . Titles and brief legends for the source data files . DOI: http://dx . doi . org/10 . 7554/eLife . 17769 . 01910 . 7554/eLife . 17769 . 020Figure 9—source data 1 . Increased CREB1 expression after training is a correlate of memory . DOI: http://dx . doi . org/10 . 7554/eLife . 17769 . 020
We hypothesize that proteins which must be synthesized for memory formation as a result of training during the active phase are either already present during the sleep phase , or that other proteins present during the sleep phase substitute for proteins that are translated as a result of training . After briefly reviewing the evidence in favor of this hypothesis , other hypotheses that could explain long-term memory formation without protein synthesis will be considered . The general requirement for both transcription and translation for long-term memory formation suggests that training first induces transcription , and then proteins are translated from the transcribed mRNAs . However , aspects of memory formation also require the translation after training of proteins from pre-existing mRNAs ( e . g . , Ghirardi et al . , 1995; Parsons et al . , 2006 ) . A short-acting inhibitor of protein synthesis , such as 10 µM anisomycin , will preferentially block this synthesis , with weaker effects on later rounds of protein synthesis that arise when mRNAs induced by training are translated . We propose that proteins synthesized from pre-existing mRNAs during active-phase training do not need to be synthesized during sleep-phase training , because proteins that are already present substitute for them . This is consistent with findings in higher animals , which showed that protein synthesis , and translation initiation factors supporting protein synthesis , are increased during Slow Wave Sleep ( SWS ) ( Grønli et al . , 2013 ) , the sleep phase associated with consolidation of declarative memory ( Rasch and Born , 2013 ) . Why are proteins that support memory present during the Aplysia sleep phase ? Their presence is likely to reflect their functions in the consolidation during sleep of memories that result from previous experiences . Thus , inhibiting protein synthesis during sleep disrupts memory consolidation in Aplysia ( Figure 1A2 ) , and interrupts synaptic plasticity in mammals ( Seibt et al . , 2012 ) . However , in most of our experiments animals were not explicitly trained in the previous active phase , raising the question of why consolidation occurs in the absence of previous training . Processes related to consolidation may not occur only after an explicit training trial . Animals and humans sleep every night , even when no training occurred in the previous day , and sleep-phase consolidation products are likely to be synthesized even without overt training during the previous waking phase ( Ramm and Smith , 1990 ) . The consolidation may be relevant to the processing of the normal experiences of previous active phase , but even in the most boring environments in which there is little new to learn , animals sleep . The hypothesis that proteins present during sleep which function in consolidation substitute for proteins synthesized after training during the active phase is supported by a number of findings . As noted above , inhibiting protein synthesis during the sleep phase blocks memory formation after training during the previous active phase ( Figure 1A2 ) . In addition , anisomycin treatment during the first hours of the sleep phase , 2 hr before training , blocks memory formation ( Figure 5A ) , because the synthesis of proteins synthesized during sleep has been blocked , and they can no longer support memory formation when animals are subsequently trained . This treatment may be effective because protein synthesis was blocked before factors required for consolidation were synthesized . An alternate possibility is that during the sleep phase there is a rapid turnover of proteins supporting consolidation , and as previously synthesized proteins are removed , they are not replaced . The synthesis of sleep-phase consolidation products predicts that training when animals are awakened from sleep should be particularly effective in producing long-term memory . This prediction is consistent with the finding that when protein synthesis is blocked just before training , even a 3 min training produces long-term memory ( Figure 2B ) . During the active phase , a 3 min training is too brief to produce long-term memory ( Levitan et al . , 2010 ) . The finding that a training which is ineffective in producing memory becomes effective if it occurs during an ongoing process of memory consolidation has a number of precedents . In synaptic and behavioral tagging , even a weak stimulus that is insufficient to produce long-term memory captures products supporting memory formation generated as a result of a training at another site by a training protocol that does cause long-term memory ( Ballarini et al . , 2009; de Carvalho Myskiw et al . , 2013; Frey and Morris , 1997; Martin et al . , 1997; Moncado and Viola , 2007 ) . As in our experiments , generation of the memory promoting products requires translation , but generation of the tags that capture the memory promoting products does not require translation . In addition , molecular cascades initiated by brief training sessions become more effective in inducing memory when they are timed to training sessions that initiate other , perhaps longer-lasting molecular cascades ( Zhang et al . , 2011 ) . In our experiments , the translation dependent products that are required for memory formation are provided by molecular events that occur during sleep , rather than by an earlier training session .
Experiments were performed on Aplysia californica weighing 75–150 g that were purchased from either Marinus Scientific ( Garden Grove , CA ) or from South Coast Bio-Marine ( San Pedro , CA ) , and on Aplysia fasciata collected along the Mediterranean coasts of Israel . The animals were stored in 600 liter tanks of aerated , filtered Mediterranean seawater maintained at 17°C . Lighting was L:D 12:12 . Animals were fed 2–3 times weekly with Ulva lactuca , which was collected at various sites along the Mediterranean coast of Israel , or purchased from Seakura , Israel ( http://www . seakura . net/ ) , and then stored frozen . A . californica are diurnally active , whereas A . fasciata are nocturnally active ( Lyons et al . , 2005 ) . Training and testing during the respective active or inactive phases were opposite in the 2 Aplysia species . A . fasciata are found locally only during the summer months ( Gev et al . , 1984 ) . When A . fasciata were unavailable , which was most of the year , A . californica were used . Experiments that were started using one Aplysia species were completed using that species , with no mixing of individuals of the two species in the same experiments . The species used in each experiment is noted in the relevant Figure Legend . There were no differences in training and testing procedures between the two species , and data gathered from the two species were comparable . All experiments utilizing qPCR were on A . californica , except for that shown in Figure 8B , which is from an earlier series of experiments on A . fasciata . Many previous experiments have shown that statistically significant savings after training are readily obtained with samples of 6–10 individuals per treatment . For this reason , both behavioral and molecular experiments were performed using Ns of this size . In some cases , multiple replications of the same experimental procedure were combined , producing larger Ns . Training during the inactive phase commenced from 3 to 6 hr after the onset or offset of light . In the dark , animals were observed during training with a red LED ( Kingbright LED L53SRC-E , peak wavelength = 660 nm , dominant wavelength = 640 nm ) approximately 1 M from the animals . The Aplysia eye responds well to wavelengths from 400 to 600 nm , and the response then drops precipitously to higher wavelengths , responding poorly at the wavelengths that we used ( Waser , 1968 ) . Animals did not respond to the onset of the illumination . Before training during the sleep phase , animals were generally immobile , which is an external sign of sleeping ( Vorster et al . , 2014 ) . Training and testing during the active phase was from 3 to 9 hr after light onset or offset . As in numerous previous studies examining learning that food is inedible in Aplysia ( Botzer et al . , 1998; Katzoff et al . , 2002 , 2006; Levitan et al . , 2012 ) , 24 hr before being trained animals were transferred to 10 L experimental aquaria that were maintained at room temperature ( 23°C ) . They were kept two to an aquarium , with the two animals separated by a partition allowing the flow of water . As in previous studies ( Susswein et al . , 1986 ) , the animals were trained with inedible food , the seaweed Ulva wrapped in plastic net . The food induced biting , leading to food entering the buccal cavity , where it induced attempts to swallow . Netted food cannot be swallowed , and it produces repetitive failed swallows . When the unswallowed food subsequently leaves the buccal cavity , the experimenter continues holding it touching the lips , inducing further bites , entries into the buccal cavity , and failed swallows . As training proceeds many bites fail to cause entry of food into the mouth . When food does enter the mouth , it stays within for progressively shorter periods , eliciting fewer attempted swallows . In some experiments , training proceeded until the animals stopped responding to food , which was defined as a lack of entry of food into the mouth for 3 min . In experiments in which the initial training was continued until the animals stopped responding ( full training ) , data were included only from animals in which food was in the mouth eliciting failed attempts to swallow for at least 100 s , since previous experience ( Levitan et al . , 2012 ) showed that such animals almost always show long-term memory . Animals in which food was not in the mouth for 100 s during a full training were discarded . A full training session until animals stop responding to food requires 10–25 min of training . Animals that stopped responding in less than 5 min were discarded . Such a training session causes long-term memory measured after 24 hr . In other experiments , training was terminated 3 min after the first response to food . In these experiments , criterion for inclusion in the experiment was 50 s of food in the mouth . In all experiments , testing of memory was performed using a blind procedure . After training , animals were coded , and their positions changed by a person not involved in the experiments , who kept the code , and revealed the identity of the animals to the experimenter only after the conclusion of the experiment . Blind procedures sometimes required repeating control procedures with known results , simply to have extra groups of animals , to maintain the blind procedure . In many experiments , the protein synthesis inhibitor anisomycin was injected into animals at various times before training . Animals which showed stress as a result of the injection , and which inked profusely , were discarded . The specific time before training at which the anisomycin was injected was often part of the experimental design , and is always noted when the experiment is described . Animals were generally injected with a 1 cc solution of anisomycin at a concentration that caused a 10 µM concentration within the animals . The volume of the whole animal was measured by displacement in a beaker of seawater , and this volume was considered to be the volume of the solvent . This concentration blocks protein synthesis in ganglia ( Schwartz et al . , 1971 ) . Controls were injected with one cc artificial seawater ( ASW - NaCl 460 mM , KCl 10 mM , CaCl211 mM , MgCl255 mM and NaHCO35 mM ) . In some experiment , animals were injected with anisomycin so as to achieve concentrations within the animal of either 3 . 3 µM or 30 µM . In some experiments , the reversible transcription inhibitor 5 , 6-Dichlorobenzimidazole 1-β-D-ribofuranoside ( DRB ) was injected into animals within 2–3 min following training . DRB is not soluble in an aqueous solution , and in previous experiments in which it was used in dissected tissues it was dissolved in DMSO ( Raju et al . , 1991 ) . We have found that injecting even one cc of DMSO into behaving animals inhibits feeding . For this reason , ethanol was used as a solvent . DRB ( 3 mg ) was dissolved in 180 µl of absolute ethanol , and then was diluted to a volume of 3 ml with ASW , and injected into 100 g animals . Dosage was adjusted when smaller or larger animals were used . Quantitative real-time PCR ( qRT-PCR ) was used to examined whether training for 3 min with anisomycin or ASW at night increased the expression of Aplysia C/EBP , CREB1 and CREB2 ( ApC/EBP , ApCREB1 and ApCREB2 ) mRNAs . In most experiments , the expression of target mRNAs was normalized to the expression of Glyceraldehyde 3-phosphate dehydrogenase ( GAPDH ) mRNA , whose expression is not thought to be regulated by training , and which has been used extensively as a housekeeping control gene ( e . g . , Hu et al . , 2015 ) . In a single experiment ( that shown in Figure 8B ) that was performed much earlier than the rest of the experiments , histone H4 was used as the housekeeping gene . This gene was also used in previous work as a housekeeping gene ( Levitan et al . , 2008 ) . The value of C/EBP/GAPDH , CREB1/GAPDH or of CREB2/GAPDH obtained for each experimental or control animal was further normalized and expressed as a percentage of the mean value of the normalized gene expression in control , untrained animals run in the same experiment , which was set at 100% . Thus , each ganglion from an untrained animal has a different value , but the mean of all these values was set at 100% . Naïve animals , and animals that were to be trained , were handled identically . Individual ganglia were rapidly excised 120 min after training . Dissected tissues were maintained in RNA Save solution ( Biological Industries Israel Beit Haemek Ltd . ) at −80°C . Total RNA was extracted using EZ-RNA ( Biological Industries Israel Beit Haemek Ltd . ) . DNA contamination was eliminated using DNA-free DNAse ( Ambion ) . Total RNA concentration was evaluated using Thermo Scientific NanoDrop 2000c UV-Vis spectrophotometer . 200 ng of total RNA from each sample was reverse-transcribed to cDNA for qPCR analyze . Reverse transcriptase was applied using a high-capacity cDNA archive kit ( RevertAid H Minus First Strand cDNA synthesis kit , Thermo Scientific ) . Samples were analyzed in triplicate using an Applied Biosystems StepOnePlus Real-Time PCR Systems . If one of the 3 samples deviated from the other 2 by more than 0 . 12 cycles , the outlier was discarded . Real-time PCR was performed using ABsolute Blue qPCR SYBR Green ROX Mix ( Thermo Scientific ) with the following specific primers: ApC/EBP , forward primer , 5′-GCAACTCAGCAACGCAACAAATGC-3′; reverse primer , 5′-TTTAGCGGAGATGTGGCATGGAGT-3′ . ApCREB1 , forward primer , 5′-TGACAAACGCTAGTCCAACCTCAG-3′; reverse primer , 5′-CCTGACGTCATGACAACACCTTGA-3′ . CREB2 , forward primer , 5’-CTACGATGGAGCTGGACCTTTGG-3’; reverse primer , 5’- AGGGTTCCAACTTCAGTGTAGCG-3’ . H4 , forward Primer , 5’-GGTGGTGTGAAGCGTATTTCTGGT-3’; reverse primer , 5’-GGCCTTGACGTTTGAGAGCATAGA-3’ . ApGAPDH , forward primer , 5′-AAGGGCATCTTGGCCTACAC-3′; reverse primer , 5′-CGGCGTACATGTGCTTGATG-3′ . Analysis of mRNA levels was done using the comparative Ct method ( Livak and Schmittgen , 2001 ) . The Aplysia CREB1 gene is transcribed into 2 mRNA isomeres , CREB1α and CREB1β . ( Bartsch et al . , 1998 ) . The primers that we used span both the CREB1α and CREB1β sequences , and will react to both mRNAs . Analyses of mRNA expression were performed on buccal ganglia 2 hr after training . The sole exception were the data shown in Figure 8B , which were examined 15 min after training . | Throughout our waking lives we are exposed to a continuous stream of experiences . Some of these experiences trigger changes in the strength of connections between neurons in the brain and begin the process of forming memories . However , these initial memory traces are fragile and only a small number will become long-term memories with the potential to last a lifetime . For this transition to occur , the brain must stabilize the memory traces through a process called consolidation . During consolidation , the brain produces new proteins that strengthen the fragile memory traces . However , if a new experience occurs while an existing memory trace is being consolidated , the new experience could disrupt or even hijack the consolidation process . To avoid this problem , the brain performs most consolidation while we are asleep . But what happens if we wake up while consolidation is taking place ? How does the brain prevent events that occur just after awakening from disrupting consolidation ? Levy et al . have now answered this question using a seemingly unlikely subject , the sea slug Aplysia . Sea slugs are capable of basic forms of learning , and their simple nervous systems and large neurons make them convenient to study . Blocking the production of new proteins in sleeping sea slugs prevents the animals from forming long-term memories , confirming that , like us , they do consolidate memories during sleep . Levy et al . now show that exposing sea slugs to new stimuli immediately after they wake up does not trigger the formation of new memories . However , if the slugs were treated with a drug that blocks protein production just beforehand , the new stimuli could trigger memory formation . These findings show that proteins blocking the formation of new memories prevent an experience upon waking from being effective in producing memory . Removing this block – by inhibiting protein production – allows experiences just after waking to be encoded in memory . This even applies to experiences that are too brief to trigger memory formation in fully awake sea slugs . The next step following on from this work is to identify these memory blocking proteins and to work out how they prevent new memories from forming . A future challenge is to find out is whether the same proteins could ultimately be used to block unwanted memories . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"neuroscience"
] | 2016 | New learning while consolidating memory during sleep is actively blocked by a protein synthesis dependent process |
The biological players involved in angiogenesis are only partially defined . Here , we report that endothelial cells ( ECs ) express a novel isoform of the cell-surface adhesion molecule L1CAM , termed L1-ΔTM . The splicing factor NOVA2 , which binds directly to L1CAM pre-mRNA , is necessary and sufficient for the skipping of L1CAM transmembrane domain in ECs , leading to the release of soluble L1-ΔTM . The latter exerts high angiogenic function through both autocrine and paracrine activities . Mechanistically , L1-ΔTM-induced angiogenesis requires fibroblast growth factor receptor-1 signaling , implying a crosstalk between the two molecules . NOVA2 and L1-ΔTM are overexpressed in the vasculature of ovarian cancer , where L1-ΔTM levels correlate with tumor vascularization , supporting the involvement of NOVA2-mediated L1-ΔTM production in tumor angiogenesis . Finally , high NOVA2 expression is associated with poor outcome in ovarian cancer patients . Our results point to L1-ΔTM as a novel , EC-derived angiogenic factor which may represent a target for innovative antiangiogenic therapies .
L1CAM , also known as CD171 or L1 , is a cell adhesion molecule encoded by the gene L1CAM . It is a cell surface glycoprotein with an extracellular portion that comprises six Ig-like domains , five fibronectin type III repeats , a short transmembrane ( TM ) domain , and a conserved cytoplasmic tail . A disintegrin and metalloproteinase ( ADAM ) 10-mediated cleavage at membrane proximal site induces shedding of the L1CAM ectodomain ( Mechtersheimer et al . , 2001 ) , while intramembrane processing mediated by γ-secretase generates a cytosolic domain which can translocate to the nucleus and modulate gene expression ( Maretzky et al . , 2005 ) . L1CAM was initially identified in the nervous system and characterized for its important function in neural development and plasticity ( Maness and Schachner , 2007 ) . Further studies then showed that L1CAM is not restricted to the nervous system and , in particular , we reported its expression in the tumor vasculature of several cancer types , while no or very low L1CAM expression is detectable in normal vessels ( Maddaluno et al . , 2009; Magrini et al . , 2014 ) . L1CAM orchestrates different endothelial cell ( EC ) functions within tumor-associated vessels , such as permeability , pericyte coverage and polarity . Since these cellular processes influence tumor angiogenesis , cancer growth and metastasis , L1CAM has emerged as a potential target for tumor vascular-specific therapies ( Magrini et al . , 2014 ) . L1CAM occurs mainly in two alternatively spliced isoforms: while neurons typically express the full-length variant of L1CAM , non-neural cell types produce a shorter isoform which lacks exon 2 and exon 27 . Exon 2 is involved in the interaction with other neuronal proteins , and exon 27 facilitates endocytosis of L1CAM ( Schäfer and Altevogt , 2010 ) . Alternative splicing ( AS ) produces different mature transcripts ( mRNAs ) from a single primary pre-mRNA . AS decisions are modulated by a number of cis-acting motifs and splicing regulatory factors ( SRFs ) that function in a coordinate manner to promote or inhibit the inclusion of specific exons into the mRNA ( Fu and Ares , 2014; Nilsen and Graveley , 2010 ) . More than 90% of human protein-coding genes undergo AS ( Pan et al . , 2008; Wang et al . , 2008 ) giving rise to different protein isoforms with distinct structural and functional properties . Hence , AS represents an important mechanism to expand the coding potential of the human genome , thus contributing to generate the cellular complexity of different tissue types and to support key functional properties ( Chen and Manley , 2009; Baralle and Giudice , 2017 ) . Notably , several findings highlighted a direct role of AS in promoting cancer progression ( Anczuków and Krainer , 2016; Biamonti et al . , 2014; Pradella et al . , 2017 ) . In particular , it has been shown that mutations or altered expression of specific SRFs allow neoplastic cells to generate cancer-specific AS isoforms involved in tumor establishment , progression and resistance to therapeutic treatments ( Bonomi et al . , 2013a; Anczuków and Krainer , 2016; Biamonti et al . , 2014; Oltean and Bates , 2014 ) . These ‘oncogenic AS switches’ can be used to stratify patients according to tumor stage ( Stricker et al . , 2017; Inoue and Fry , 2015 ) , while their targeting represents a promising approach to improve the efficacy of anti-cancer treatments ( Bonomi et al . , 2013a; Agrawal et al . , 2018; Anczuków and Krainer , 2016 ) . However , in contrast to the established role of AS in tumor cells , it remains unclear whether this process is also relevant in tumor microenvironment and , in particular , in cancer vasculature . In fact , AS events specifically occurring in tumor-associated ECs have been described ( Neri and Bicknell , 2005 ) and proposed as potential targets for antiangiogenic therapies ( Steiner and Neri , 2011 ) . However , how such AS events impact on the pathophysiology of tumor vasculature remains elusive . Recently , we described the SRF Neuro-Oncological Ventral Antigen 2 ( NOVA2 ) as a prominent regulator of AS during vascular development ( Giampietro et al . , 2015 ) . NOVA2 was initially identified in neural cells where it controls AS of several genes involved in various neural developmental processes by binding to clusters of YCAY ( Y = C/U ) repeats within its pre-mRNA targets ( Licatalosi et al . , 2008; Ule et al . , 2003; Zhang et al . , 2010; Leggere et al . , 2016; Saito et al . , 2016 ) . Our study revealed that NOVA2 is also expressed in vascular endothelium and is regulated during angiogenesis ( Giampietro et al . , 2015 ) . NOVA2 controls at the post-transcriptional level the establishment of EC polarity , a process that is essential for vascular lumen formation and , hence , for angiogenesis ( Iruela-Arispe and Davis , 2009 ) . Accordingly , NOVA2 ablation causes defects in vascular lumen formation in vivo ( Giampietro et al . , 2015 ) . Here , we report a novel isoform of L1CAM expressed in ECs as the result of a NOVA2-induced AS event that removes the exon encoding the transmembrane domain of the protein . This gives rise to a soluble L1CAM variant , referred to as L1-ΔTM , that is released by ECs and is able to stimulate angiogenesis via autocrine/paracrine mechanisms . NOVA2 and L1-ΔTM are overexpressed in the vasculature of ovarian cancer and correlate with poor outcome and tumor vascularization , respectively . Our findings , therefore , implicate the novel NOVA2/L1-ΔTM axis in EC pathophysiology and in ovarian cancer aggressiveness .
We have recently reported the novel function of L1CAM in vascular endothelium ( Magrini et al . , 2014 ) . Since AS is known to influence the biological activities of cell-surface adhesion molecules ( Wang et al . , 2005 ) , it is possible that AS of L1CAM accounts for , or at least contributes to , its peculiar role in ECs . A bioinformatics analysis with the ExonMine program ( http://www . imm . fm . ul . pt/exonmine/ ) ( Mollet et al . , 2010 ) identified a human expressed sequence tag ( EST ) in which the L1CAM exon 25 ( a 135-nucleotide cassette exon ) is excluded from the mature mRNA ( Figure 1A ) . We then analyzed several normal human tissues and human ECs for the AS of human L1CAM exon 25 by RT–PCR ( Figure 1B ) . In addition , we also investigated the AS of this exon in the mouse . In the murine gene , this exon is annotated as exon 26 by UCSC and Ensembl , due to the presence of an additional non-coding exon upstream of exon 1 ( i . e . , the one containing the ATG codon ) . Nevertheless , based on its high homology to the human L1CAM exon 25 ( 89% identity ) , we refer to it as exon 25 also in mouse L1cam . The AS of this exon was examined in normal mouse tissues , mouse EC lines and freshly purified murine ECs . As shown in Figure 1B and C , in both human and mouse samples the skipping of exon 25 mainly occurred in ECs . Overall , these data suggest that ECs express a novel alternatively spliced isoform of L1CAM devoid of exon 25 . Skipping of exon 25 results in an in-frame deletion of a 44-amino acid sequence ( 45 in mouse ) that encompasses the entire transmembrane ( TM ) domain of L1CAM ( Figure 2A ) . This suggests that AS of exon 25 could affect L1CAM localization and , hence , its activity . To test this hypothesis , we selected an immortalized endothelial cell line of murine origin , moEC ( Lampugnani et al . , 2002; Taddei et al . , 2008 ) , because it expresses no or very little endogenous L1CAM ( Figure 2B ) and , therefore , is amenable to gain-of-function studies . As shown in Figure 1—figure supplement 1 , ECs express the non-neural isoform of L1CAM mRNA , which lacks exons 2 and 27 ( Schäfer and Altevogt , 2010 ) . Thus , exon 25 was deleted from non-neural mouse L1cam in order to obtain L1-ΔTM . MoEC were stably transduced with either full-length , non-neural L1CAM ( L1-FL ) or with L1-ΔTM . Immunoblotting of total extracts from L1-FL-expressing cells showed the expected doublet pattern ( Figure 2B ) , with the upper band representing the mature , fully glycosylated cell-surface form , and the lower band corresponding to the precursor form ( Moulding et al . , 2000; Zisch et al . , 1997 ) . In contrast , L1-ΔTM-expressing moEC showed a single band that migrated slightly faster than the lower band of L1-FL ( Figure 2B ) , consistent with the deletion of exon 25 . The lack of the TM domain might affect the subcellular localization of L1CAM . To address this question , we performed immunofluorescence staining for L1CAM on moEC expressing the two isoforms . Only L1-FL was prominently exposed on the cell surface of moEC , while cells expressing L1-ΔTM exhibited mainly a cytoplasmic localization ( Figure 2C ) . Based on the deletion of the TM domain and on the absence of membrane staining , we hypothesized that L1-ΔTM is released into the extracellular space and represents a novel soluble form of the protein . Indeed , conditioned medium ( CM ) from L1-ΔTM-expressing moEC contained high amounts of L1-ΔTM ( Figure 2D ) . Of note , the protein in the CM retained the cytoplasmic tail ( Figure 2—figure supplement 1 ) . This confirmed the release of the entire molecule as opposed to the proteolytic cleavage of full-length L1CAM , which results in the shedding of its extracellular portion ( Figure 2D; Mechtersheimer et al . , 2001 ) . Both the cytoplasmic localization of L1-ΔTM and its release into the CM were also confirmed in luEC , another murine EC line ( Figure 2—figure supplement 1 ) . Interestingly , L1-ΔTM was detected in the culture medium as a single band that migrated slower than the protein found in the cell lysate ( Figure 2D ) . We hypothesized that the different size was accounted for by the glycosylation state of the protein . This was confirmed by the forced expression of L1-ΔTM in N-acetylglucosaminyltransferase I-deficient HEK293 cells [GnTI ( - ) ] ( Reeves et al . , 2002 ) . The latter , indeed , released a form of L1-ΔTM that exhibited a lower molecular weight as compared to wild-type HEK293 cells ( Figure 2—figure supplement 1 ) . These findings support the notion that L1-ΔTM is released in its mature , fully glycosylated form . In order to confirm the release of endothelial L1-ΔTM in an endogenous system , we employed lu2EC , an immortalized mouse endothelial cell line that expresses relatively high levels of L1cam ( Figure 2—figure supplement 1 ) . These cells also express endogenous L1-ΔTM ( Figure 2—figure supplement 1 ) . When the lu2EC-derived CM was immunoblotted with the antibody against the cytoplasmic tail of L1CAM , we found high levels of L1-ΔTM ( Figure 2—figure supplement 1 ) , confirming its release into the extracellular space . To compare our findings in ECs with a non-endothelial cell type , we used the mouse melanoma cell line B16 , which also expresses high levels of endogenous L1CAM ( Linnemann and Bock , 1986; Magrini et al . , 2014 ) , but does not express L1-ΔTM ( Figure 2—figure supplement 1 ) . While we could detect high levels of cell-associated L1CAM , no C-terminus-containing L1CAM was detected in the CM of B16 cells ( Figure 2—figure supplement 1 ) , supporting the hypothesis that the AS of L1CAM results in the release of L1-ΔTM in ECs . Collectively , our results indicate that skipping of L1CAM exon 25 generates a novel isoform of the protein that is released in the extracellular compartment . To investigate the biological role of the L1-ΔTM isoform in ECs , we focused on their ability to form capillary-like tubes in three-dimensional matrices , which reflects their angiogenic potential ( Di Blasio et al . , 2014 ) . Therefore , we assayed control and L1-ΔTM-expressing ECs for tube formation on Matrigel . As shown in Figure 3—figure supplement 1 , Figure 3A and Video 1 , L1-ΔTM enhanced significantly the tube forming ability of moEC , thus suggesting that it is endowed with angiogenic properties . The direct role of L1-ΔTM in moEC tube formation was probed with 324 , a L1CAM-neutralizing antibody ( Appel et al . , 1993; Di Sciullo et al . , 1998 ) . As shown in Figure 3B , the antibody 324 , but not a control antibody , abolished the tube-forming potential of moEC . The results of this proof-of-concept experiment also support the neutralization of vascular L1-ΔTM as a potential strategy to interfere with the angiogenic process . Based on the substantial release of L1-ΔTM into the extracellular space , we asked whether the molecule could also exert its biological function as a soluble factor in a paracrine fashion . To address this question , parental moEC were subjected to tube formation assays in the presence of the CM from moEC expressing either L1-ΔTM or the control vector . ECs exposed to the CM from L1-ΔTM-expressing cells exhibited higher tube-forming activity than those exposed to control medium ( Figure 3C ) or to the CM from L1-FL-expressing cells ( not shown ) . Similar results were obtained by using CM from luEC expressing either L1-ΔTM or the control vector ( Figure 3—figure supplement 1 ) . To further verify the angiogenic activity of soluble L1-ΔTM , we treated parental moEC with a purified , recombinant version of the protein produced in mammalian cells ( Figure 3—figure supplement 1 ) . Indeed , recombinant soluble L1-ΔTM induced moEC tube formation in a dose-dependent manner ( Figure 3D ) , thus confirming its ability to stimulate EC remodeling and morphogenesis . In order to validate our results in an EC model with endogenous L1-ΔTM , we treated lu2EC with a morpholino oligonucleotide that selectively prevents the inclusion of L1cam exon 25 ( Figure 3E ) . As shown in Figure 3F , this resulted in increased expression and extracellular release of endogenous L1-ΔTM . Importantly , lu2EC exposed to the CM from morpholino-treated cells exhibited higher tube-forming activity than those exposed to control CM , thus confirming the functionality of endogenous L1-ΔTM ( Figure 3F ) . Finally , we aimed at validating our findings in an in vivo assay of angiogenesis . Mice underwent subcutaneous implantation of Matrigel plugs containing CM from either L1-ΔTM- or L1-FL-expressing ECs or from control cells . Neovascularization was markedly induced by L1-ΔTM-containing CM , while a weaker effect was observed with the CM from L1-FL-expressing cells ( Figure 3G ) . This strongly supports the angiogenic function of L1-ΔTM . Previous studies implicated fibroblast growth factor receptor ( FGFR ) signaling as an effector of L1CAM in different cellular contexts ( Díaz-Balzac et al . , 2015; Kulahin et al . , 2008; Mohanan et al . , 2013; Williams et al . , 1994; Zecchini et al . , 2008 ) . However , the L1CAM/FGFR interplay in ECs has not been investigated . Given the well-characterized role of FGFR function in vascular biology and angiogenesis ( Ronca et al . , 2015 ) , we hypothesized that the pro-angiogenic effect of L1-ΔTM was mediated by FGFR . Among the four FGFR family members , moEC express only FGFR1 ( data not shown ) ( Giampietro et al . , 2012 ) , as previously reported for other EC types ( Giacomini et al . , 2016; Javerzat et al . , 2002 ) . To determine if L1-ΔTM function could be mediated by FGFR1 , we first investigated whether soluble L1-ΔTM affects FGFR1 activation . As shown in Figure 3H , treating parental moEC with recombinant L1-ΔTM resulted in increased phospho-FGFR1 , consistent with the L1-ΔTM-induced activation of FGFR1 signaling . Moreover , when L1-ΔTM-expressing moEC were subjected to tube formation assay in the presence of the small-molecule FGFR1 inhibitor PD173074 ( Skaper et al . , 2000 ) , L1-ΔTM-dependent tube-forming activity was reduced to the level of control cells ( Figure 3I ) . Thus , our data implicate FGFR1 signaling as an effector of L1-ΔTM in ECs . To gain further insights into the molecular mechanisms regulating the AS of L1cam in endothelium , we analyzed the sequence of mouse L1cam exon 25 and its flanking intronic regions , using SFmap ( http://sfmap . technion . ac . il/ ) ( Paz et al . , 2010; Akerman et al . , 2009 ) to search for putative binding sites of RNA-binding proteins . We sorted the results based on: i ) the predicted ability of the RNA-binding protein to promote exon 25 skipping; ii ) the presence of clusters of putative binding sites for a given RNA-binding protein , which are expected to enhance binding affinity; iii ) the evolutionary conservation of the identified motifs; iv ) the known expression of the identified factor in ECs . This analysis resulted in the identification of clustered and evolutionarily conserved putative binding sites for NOVA2 , hnRNP A1 and SRSF3 ( Figure 4—figure supplement 1 ) , three factors previously reported to be expressed in ECs ( Giampietro et al . , 2015; Holly et al . , 2013; Lomnytska et al . , 2004 ) . To investigate the role of the identified candidate splicing factors in the AS of L1cam , we first performed a splicing assay in HeLa cells co-transfected with a minigene ( p-L1 ) encompassing exons 24 , 25 , and 26 of L1cam along with the flanking intron sequences ( Figure 4A ) and the candidate splicing factors or with the empty vector . As shown in Figure 4B and Figure 4—figure supplement 1 , skipping of L1cam exon 25 in the minigene was only observed upon overexpression of NOVA2 , a key regulator of AS in ECs ( Giampietro et al . , 2015 ) . In contrast , the overexpression of hnRNP A1 and SRSF3 had no effect on the skipping of exon 25 , suggesting that the latter is a NOVA2-specific effect . To support a direct and specific role of NOVA2 in controlling L1cam AS , we mutated YCAY ( Y = C/U ) motifs , which represent putative binding sites for NOVA proteins ( Ule et al . , 2006 ) , in L1cam exon 25 to ACAY , a sequence that reduces NOVA2 binding ( Jensen et al . , 2000 ) . We found that mutations in only three repeats ( Mut3 ) had a limited effect , whereas mutations in five repeats ( Mut5 ) decreased skipping of L1cam exon 25 caused by NOVA2 overexpression ( Figure 4C ) . These results are consistent with the dose-dependent binding of NOVA2 to its pre-mRNA targets ( Darnell , 2006; Leggere et al . , 2016 ) and further supported the involvement of NOVA2 in the AS regulation of L1cam exon 25 . Such a hypothesis was also sustained by the following observations: i ) the higher expression of NOVA2 in freshly purified ECs from mouse lung and in lu2EC as compared with total mouse lung or melanoma cell line B16 , respectively , was accompanied by the skipping of L1cam exon 25 ( Figure 4—figure supplement 2 and Figure 2—figure supplement 1 ) ; and ii ) L1cam exon 25 emerged as a novel NOVA2 target in ECs from the RNA-seq data in NOVA2-knockdown ECs ( Giampietro et al . , 2015 ) ( Supplementary file 2 ) ( see Materials and methods ) . To investigate the causal relationship between NOVA2 expression and AS of the endogenous L1cam , we performed gain- and loss-of function studies in moEC ( Figure 4D–F and Figure 4—figure supplement 3 ) . In particular , forced expression of NOVA2 increased skipping of L1cam exon 25 ( Figure 4E and Figure 4—figure supplement 3 ) . Conversely , in NOVA2-depleted moEC ( Figure 4—figure supplement 3 ) the skipping of L1cam exon 25 was markedly reduced ( Figure 4F ) . The NOVA2-mediated AS regulation of L1cam exon 25 was also confirmed in lu2EC , another murine EC line ( Figure 4—figure supplement 3 ) . Whether NOVA2 promotes exon skipping or inclusion depends on the location of its binding sites ( i . e . YCAY clusters ) in the pre-mRNA targets ( Ule et al . , 2003 ) . In particular , NOVA2 usually induces exon skipping when bound to the exonic or upstream intronic region , while it stimulates exon inclusion when interacting with downstream intronic region . In the case of L1cam exon 25 , the YCAY repeats are located within exon 25 ( Figure 4A ) , consistent with the NOVA2-induced exon skipping observed in mouse ECs . Notably , the YCAY cluster is conserved between mouse and human L1CAM exons 25 with six repeats present in the human sequence ( Figure 4—figure supplement 3 ) . Accordingly , NOVA2 overexpression promotes skipping of L1CAM exon 25 also in human ECs ( Figure 4—figure supplement 3 ) . To determine if NOVA2 directly regulates AS of the endogenous L1cam , we carried out UV crosslinking and immunoprecipitation ( CLIP ) , which allows to identify direct protein-RNA interactions in live cells ( Ule et al . , 2006 ) . RNA from UV cross-linked ECs was immunoprecipitated by using anti-NOVA2 or control antibodies and then analyzed by RT-qPCR with primers spanning the YCAY cluster within L1cam exon 25 . Primers that span either exon 26 or intron 26 were used as negative controls ( Figure 4G ) . As shown in Figure 4G , NOVA2 bound to the endogenous L1cam transcript at the level of exon 25 , while we observed no binding with either exon or intron 26 . These data indicated a direct and specific interaction of NOVA2 with L1cam exon 25 . NOVA2 has been implicated also in the inclusion of exons 2 and 27 in neural cells ( Mikulak et al . , 2012 ) . However , both RT-PCR and CLIP data showed no involvement of NOVA2 in the AS of these two exons in ECs ( Figure 4G and Figure 1—figure supplement 1 ) , further supporting the specific effect of NOVA2 on exon 25 in this cell type . Collectively , our results support the notion that NOVA2 promotes skipping of L1cam exon 25 by binding to the YCAY motifs located within this exon . We have recently described the expression of NOVA2 in vascular endothelium ( Giampietro et al . , 2015 ) . Furthermore , our earlier reports demonstrated that L1CAM is expressed in tumor-associated vasculature ( Maddaluno et al . , 2009; Magrini et al . , 2014 ) . Taken together with the data presented here , these findings raise the hypothesis that NOVA2 regulates AS of L1CAM in cancer vessels . To test this possibility , we selected human ovarian carcinoma ( OC ) as a suitable model system . In fact , we found a markedly higher number of NOVA2-positive vessels in OC ( identified via staining with the endothelial marker CD31 ) than in healthy ovaries ( Figure 5A and Figure 5—figure supplement 1 ) . The abundance and the vessel-restricted expression of NOVA2 in OC were also confirmed in tissue samples from the Human Protein Atlas project ( https://www . proteinatlas . org/ ) ( Uhlén et al . , 2015 ) ( Figure 5—figure supplement 1 ) . The percentage of L1CAM-positive vessels was also dramatically increased in OC samples as compared to normal ovary ( Figure 5A ) . In addition , NOVA2 was often co-expressed with L1CAM in OC vessels ( Figure 5B and Figure 5—figure supplement 2 ) . Thus , we applied RT-PCR to examine the AS of L1CAM in ECs isolated from OC ( HOC-EC ) . As shown in Figure 5C , the L1-ΔTM isoform was readily detected in HOC-EC from seven independent OC samples . To test whether vascular L1-ΔTM in OC is associated with tumor angiogenesis , we measured the vessel density in a small cohort of OC samples pre-classified as L1-ΔTM-positive or negative by RT-qPCR ( Figure 5—figure supplement 2 ) . A significantly higher vessel density was found in L1-ΔTM-positive tumors ( Figure 5—figure supplement 2 ) . Furthermore , among the tumors which exhibited L1-ΔTM expression , the levels of L1-ΔTM correlated with vessel density ( r = 0 . 7671; p<0 . 01 ) , measured by CD31 immunostaining ( Figure 5D ) . These findings imply that the AS of L1CAM correlates with the degree of OC vascularization , which is consistent with a proangiogenic function of L1-ΔTM in this tumor type . To further assess the clinical relevance of our findings , we investigated the prognostic value of NOVA2 in OC , profiting from the RNA sequencing analysis of 372 OC patients performed through The Cancer Genome Atlas ( TCGA ) program . As shown in Figure 5E , higher expression of NOVA2 correlated with shorter overall survival of the patients ( HR: 1 . 486; p=0 . 003 ) . Taken together , these results suggest that NOVA2 promotes AS of the L1CAM pre-mRNA in OC vessels , thus accounting for the vascular expression of L1-ΔTM , and highlight the proangiogenic role and the prognostic value of the NOVA2/L1-ΔTM axis in OC .
Our data implicated for the first time the splicing factor NOVA2 in the generation of a novel , EC-specific isoform of the cell adhesion molecule L1CAM , referred to as L1-ΔTM . Due to NOVA2-induced skipping of exon 25 that encodes the TM domain , L1-ΔTM is no longer associated to the cell surface and , hence , is released in the extracellular space . Consistent with the expression of NOVA2 in vascular ECs ( Giampietro et al . , 2015 ) , the latter express and release high levels of L1-ΔTM . We demonstrated that L1-ΔTM increases the ability of ECs to form tube-like structures in vitro and stimulates neovascularization in vivo . These data point to L1-ΔTM as a bona fide angiogenic factor which , however , belongs to a class of molecules highly divergent from the classic polypeptide growth factors that exert this function ( vascular endothelial growth factors , fibroblast growth factors , etc . ) . To our knowledge , L1-ΔTM provides the first example of an immunoglobulin-like cell adhesion molecule converted into a soluble angiogenic factor through AS removal of the exonic sequence that encodes the TM domain . AS of the TM domain occurs in a broad spectrum of human genes ( Xing et al . , 2003 ) , yet for most of them the biological significance and the functional outcome of such an event remain elusive . Based on our findings , it is conceivable that this post-transcriptional modification expands the repertoire of biologically active proteins that are released into the extracellular compartment . We discovered that ECs acquire angiogenic properties not only upon ectopic expression of L1-ΔTM , but also when exposed to exogenously added L1-ΔTM ( either as CM or as a recombinant protein ) . This supports the notion that extracellular L1-ΔTM can promote angiogenesis via both autocrine and paracrine stimulation . The latter mode of action , in particular , might represent a way to amplify the angiogenic signal provided by vascular L1CAM . Indeed , in most cases only a fraction of vessels within tumor and inflammatory tissues ( i . e . the most prominent conditions where vascular L1CAM is detected ) , exhibits L1CAM expression ( Issa et al . , 2009; Kaifi et al . , 2006; Maddaluno et al . , 2009; Magrini et al . , 2014 ) . Thus , the AS-mediated generation of L1-ΔTM , and hence its release , would contribute to spread the proangiogenic stimulus also to L1CAM-negative vessels . In different cellular contexts , such as neurons , L1CAM frequently engages in homophilic L1CAM-L1CAM interactions between adjacent cells , which underlie its function in cell-cell adhesion ( Maness and Schachner , 2007 ) . However , our experiments were conducted in two EC lines , moEC and luEC , that express no or very little endogenous L1CAM . This implies that , regardless of autocrine or paracrine activity , L1-ΔTM stimulates ECs by heterophilic interactions with different surface molecules . Accordingly , the extracellular portion of L1CAM has been reported to interact with a wide spectrum of membrane proteins , including different integrins , CD24 , NCAM , F11R , neuropilins , etc . ( Haspel and Grumet , 2003 ) . With regard to potential L1-ΔTM interactors , one of the most intriguing aspects of our study is related to the crosstalk between L1-ΔTM and FGFR1 . Following the pioneering studies of Doherty and Walsh in neuronal systems ( Williams et al . , 1994 ) , L1CAM has been proposed to interact with the FGFR signaling machinery in different experimental settings ( Mohanan et al . , 2013; Son et al . , 2011; Zecchini et al . , 2008 ) , possibly entailing a direct binding between the two molecules ( Kulahin et al . , 2008 ) . Our data show for the first time a functional interplay between L1-ΔTM and FGFR1 in ECs , and implicate it in L1-ΔTM-induced angiogenesis . Future studies should elucidate whether interfering with the L1-ΔTM/FGFR1 crosstalk will open novel perspectives for antiangiogenic therapy . In neural cell types , NOVA2 is required for the inclusion of the L1CAM exons 2 and 27 ( Mikulak et al . , 2012 ) . This is not the case in ECs , where even upon ectopic over-expression of NOVA2 exons 2 and 27 are not included in the L1CAM mRNA ( Figure 1—figure supplement 1 ) . Moreover , despite the high expression of NOVA2 in the nervous system ( Yano et al . , 2010; Giampietro et al . , 2015 ) , we did not detect L1-ΔTM in human or mouse brain ( Figure 1 ) , which implies that NOVA2 is not involved in the AS of exon 25 in neural cells . Interestingly , our CLIP data on ECs revealed the binding of NOVA2 to exon 25 , while no binding was detected with the regions flanking the neural-specific exons 2 and 27 ( Figure 4G ) . Overall , these observations point to differential and cell type-specific interactions of NOVA2 with the L1CAM pre-mRNA that , in turn , underlie distinct AS outcomes . Future studies should aim at identifying the molecular determinants of such a cell type specificity . NOVA2 is significantly up-regulated in OC vasculature as compared to vessels in healthy ovary . Of note , NOVA2 expression is detected only in the vascular endothelium of OC and , in particular , in the nucleus of ECs . It is remarkable that , in spite of such a restricted expression pattern , NOVA2 has a prognostic value in OC , given its association with shorter patients’ survival . To the best of our knowledge , NOVA2 is the only SRF reported to be upregulated in cancer vasculature , including OC ( this study ) and colorectal carcinoma ( Gallo et al . , 2018 ) . This implies that NOVA2-mediated AS might play a relevant , yet underappreciated , role in the phenotypic and functional aberrancies of the tumor vessels ( Carmeliet and Jain , 2011 ) . Consistent with the proposed role of the NOVA2/L1-ΔTM axis in tumor angiogenesis , increased NOVA2 levels are frequently accompanied by the expression of L1-ΔTM in tumor vessels . Furthermore , L1-ΔTM expression correlates with OC vascularization , although further studies should investigate the causal role of L1-ΔTM in tumor angiogenesis . These data , together with our observations on the role of vascular L1CAM in tumor angiogenesis and progression ( Magrini et al . , 2014 ) and on the proangiogenic activity of L1-ΔTM , point to NOVA2 as a potential driver of OC neovascularization . In addition , our results further strengthen the rationale for testing vascular L1CAM as a novel target for antiangiogenic strategies . Indeed , interfering with the function of cell surface-associated endothelial L1CAM ( Magrini et al . , 2014 ) and , at the same time , with that of extracellular L1-ΔTM might result in efficient repression of cancer angiogenesis . The translational implications of our findings extend beyond the therapeutic area . Based on its specific up-regulation in OC vessels ( and likely in other cancer types as well ) , we speculate that L1-ΔTM could serve as a new circulating biomarker of cancer-associated vasculature . In this case , it may become necessary to discriminate between vessel-derived L1-ΔTM and the ectodomain of transmembrane L1CAM which is released by various cell types , including tumor cells , upon proteolytic cleavage ( Kiefel et al . , 2012; Yu et al . , 2016 ) . Assays aimed at detecting the cytosolic domain of circulating L1CAM would serve this purpose . Circulating L1-ΔTM could be harnessed for diagnostic and prognostic purposes , ranging from the detection of primary and recurrent OC to monitoring the therapeutic response , all objectives that would improve dramatically the clinical management of OC patients . Along this line , the circulating levels of vessel-derived L1-ΔTM could offer a novel predictive tool to identify patients eligible for antiangiogenic therapies . In summary , we provide evidence that L1-ΔTM is a novel , soluble isoform of L1CAM generated in ECs through NOVA2-mediated AS . L1-ΔTM exhibits proangiogenic activity and is up-regulated in tumor vessels , which may have high translational and clinical relevance .
Human cervix carcinoma HeLa cells ( ATCC , CCL-2 ) were grown in Dulbecco’s modified Eagle’s medium ( DMEM , Euroclone ) supplemented with 10% fetal bovine serum ( FBS , Euroclone ) and 2 mM L-glutamine ( Lonza ) . Human embryonic kidney ( HEK ) 293 cells ( FreeStyle 293 F Cells , Gibco ) were grown in FreeStyle 293 Expression Medium ( Gibco ) . N-acetylglucosaminyltransferase I-deficient HEK293 cells [GnTI ( - ) ] ( Reeves et al . , 2002 ) were grown in FreeStyle 293 Expression Medium supplemented with 0 . 1% Pluronic F-127 ( Sigma-Aldrich ) and 1% FBS . Vascular endothelial ( VE ) cadherin-positive ECs , here referred to as moEC , have been described in Lampugnani et al . ( 2002 ) and Taddei et al . ( 2008 ) . An extensive characterization confirmed the endothelial nature of these cells ( Lampugnani et al . , 2003; Taddei et al . , 2008 ) . Mouse lung-derived luEC and lu2EC were described in Magrini et al . ( 2014 ) , and Bazzoni et al . ( 2005 ) , respectively . Primary EC were isolated from mouse lung with anti-CD31 immunomagnetic beads as described ( Malinverno et al . , 2017 ) . More than 99% isolated cells were positive for VE-cadherin and CD31 by immunofluorescence ( Figure 4—figure supplement 2 ) . Furthermore , isolated cells exhibited a marked enrichment in the mRNA for VE-cadherin and CD31 , while no or very little expression was detected for epithelial ( E ) -cadherin and alpha-smooth muscle actin ( Figure 4—figure supplement 2 ) . Mouse EC lines derived from whole embryo ( EmbEC ) , lu2EC and moEC were cultured in DMEM-High Glucose ( Lonza ) with 10% FBS , 2 mM L-glutamine , 100 U/L penicillin/streptomycin ( Sigma-Aldrich ) , 1 mM sodium pyruvate ( Sigma-Aldrich ) , 25 mM HEPES ( Sigma-Aldrich ) , 100 μg/ml heparin ( from porcine intestinal mucosa; Sigma-Aldrich ) and 50 μg/ml EC growth supplement ( ECGS from bovine pituitary gland; Sigma-Aldrich ) . LuEC were cultured in MCDB131 medium ( Sigma-Aldrich ) supplemented with 20% FBS , 2 mM L-glutamine , 1 mM sodium pyruvate ( Gibco ) , 100 μg/ml heparin and 50 μg/ml ECGS as previously described ( Magrini et al . , 2014 ) . To enhance EC adhesion , plates were coated with 0 . 1% porcine gelatin ( Difco ) and incubated overnight at 37°C before seeding . Human umbilical vein EC ( HUVEC , isolated as in ( Giampietro et al . , 2015 ) and HUVEC/TERT2 ( Evercyte ) were cultured on porcine gelatin-coated plates in MCDB131 containing 20% FBS and the same supplements as moEC and luEC . Human ovarian cancer-derived EC ( HOC-EC ) have been described previously . HOC-EC cultures were found to contain at least 99% endothelial marker-positive ( LDL uptake , CD31 , von Willebrand Factor ) cells , and <1% alpha-smooth muscle actin-positive cells ( Ghilardi et al . , 2015 ) . Human immortalized cerebral microvascular ECs ( hCMEC/D3 ) were provided by PO Couraud and cultured as described previously ( Weksler et al . , 2013 ) . hCMEC/D3 cells were grown on Collagen type I coated dishes ( Corning ) . Mouse melanoma B16F10 cells ( hereafter referred to as B16 ) were purchased from ATCC and cultured in DMEM , 10% FBS . All cells were routinely tested for mycoplasma with a PCR-based method . Whenever applicable , cell line authentication was performed with the GenePrint STR System typing kit ( Promega ) . The cDNAs encoding HA-tagged , full-length mouse L1cam ( deleted of the neural-specific exons 2 and 27 ) and its TM—deleted L1cam isoform ( deleted of exons 2 , 25 and 27 ) ( Figure 1—figure supplement 1 ) were generated by PCR-mediated mutagenesis of pcDNA3 . 1-Hygro/L1 construct ( Magrini et al . , 2014 ) and then cloned into the pcDNA3 . 1-Hygro ( - ) ( Invitrogen ) or in the pLenti-III-HA ( ABM Inc ) by using standard DNA cloning procedures . To generate the p-L1 WT minigene ( Figure 4 ) , the genomic mouse L1cam cassette was amplified with primers p-L1_F/R and then cloned into the EcoRI and XbaI restriction sites of pcDNA3 . 1 ( + ) ( Invitrogen ) . In p-L1 Mut minigenes ( Mut3 and Mut5 ) , the NOVA2-binding sites TCAT were replaced with ACAT by PCR-mediated mutagenesis of p-L1 WT minigene . HA-tagged human NOVA2 cDNA was PCR amplified with primers pcDNA-NOVA2_F/R and cloned into the BamH1 and XhoI restriction sites of the pcDNA3 . 1 ( + ) , whereas T7-tagged hnRNP A1 expression vector was generated as described previously ( Bonomi et al . , 2013b ) . The expression vector for T7-SRSF3 was kindly provided by J . Caceres ( MRC Human Genetics Unit , University of Edinburgh , UK ) . To produce recombinant L1-ΔTM protein ( see below ) , the L1-ΔTM cDNA was amplified with pUPE-L1_F/R primers from pcDNA Hygro-L1-ΔTM . The resulting amplicon , bearing in-frame 5'-BamHI and 3'-NotI restriction sites was then sub-cloned into pUPE . 06 . 45 mammalian expression vector ( U-Protein Express B . V . , Utrecht , The Netherlands ) , bearing a cystatin signal peptide followed by N-terminal 6xHis , 3xStrepII tags , and a specific cleavage site for Tobacco Etch Virus protease ( TEV ) preceding the 5'-BamHI restriction site . All PCR products were verified by sequencing , whereas all primers are listed in Supplementary file 1 . HeLa cells were transiently transfected with Lipofectamine 3000 ( Invitrogen ) , according to the manufacturer’s protocol . luEC were stably transfected with pcDNA-Hygro/L1-FL or pcDNA-Hygro/L1-ΔTM or the empty vector pcDNA-Hygro ( - ) ( see below ) , followed by the selection of positive clones as previously described ( Magrini et al . , 2014 ) . HEK293 and HEK293 GntI ( - ) cells were transfected using polyethyleneimine ( Polysciences , Germany ) as described ( Durocher et al . , 2000 ) . To obtain viruses expressing L1CAM isoforms , mouse L1cam cDNAs were cloned in the pLenti-III-HA vector ( ABM Inc , see above ) . Lentiviruses were produced using HEK293T as recipient cells ( ATCC , CRL-1573 ) . Cells were co-transfected with 10 μg of each packaging vector pMD2 . G ( Addgene , plasmid #12259 ) , pRSV-Rev ( Addgene , plasmid #12253 ) , pMDLg/pRRE ( Addgene , plasmid #12251 ) and 10 μg of L1CAM-expressing lentiviral vectors by calcium phosphate precipitation method . After 24 hr , the medium containing the lentiviruses was filtered , supplemented with 8 μg/ml of polybrene ( Sigma-Aldrich ) and used to infect moEC . MoEC were transduced with HA-NOVA2 or with shRNA vectors as described in ( Giampietro et al . , 2015 ) , using lentiviral vectors carrying human HA-tagged NOVA2 cDNA ( pLenti-GIII-CMVhumanNOVA2-HA , THP Medical Products ) or shRNA for the mouse Nova2 gene ( GIPZ shRNAs from Open Biosystems ) , respectively . After 48 hr of infection , the medium was refreshed and puromycin selection ( 3 μg/ml ) was started . Since NOVA2 expression is regulated by EC density ( Giampietro et al . , 2015 ) , for the analysis of L1CAM splicing NOVA2-knockdown moEC were used as confluent monolayers ( 500000 cells in 35 mm Petri dishes ) , whereas moEC overexpressing HA-tagged NOVA2 were tested at low density ( 500000 cells in 100 mm Petri dishes ) . hCMEC/D3 cells were transduced with lentiviral vectors carrying human HA-tagged NOVA2 cDNA ( pLenti-GIII-CMVhumanNOVA2-HA , THP Medical Products ) and after 48 hr , infected cells were selected with 3 μg/ml puromycin for 5 days . To ablate Nova2 in lu2EC , we used siRNA from Sigma-Aldrich ( MISSION siRNA ID SASI_Mm01_00094763 ) and the corresponding negative control ( MISSION siRNA Universal Negative Control #1 ) . Transfection was performed with Lipofectamine RNAiMax ( Invitrogen ) in accordance with the manufacturer’s instructions . Two subsequent transfections ( with 24 hr intervals ) were performed with 13 nM siRNA , and cells were collected 24 hr after the second transfection . To knock down NOVA2 in HUVEC/TERT , we used siRNA from Sigma-Aldrich ( MISSION siRNA ID SASI_Hs01_00220812 ) and the corresponding negative control ( MISSION siRNA Universal Negative Control #1 ) . Transfection was performed with Lipofectamine RNAiMax ( Invitrogen ) following the manufacturer’s instructions . Two subsequent transfections ( with 24 hr intervals ) were performed with 30 nM siRNA , and cells were collected 48 hr after the second transfection . Subconfluent ( 80–85% ) lu2EC cells were transfected with the MO-L1-SB oligonucleotide ( 5’- CCTGTACATTTTCTAGGTTACCTGA-3’; GENE-TOOLS ) at 15 μM plus 8 μM Endo-Porter PEG system ( GENE-TOOLS ) according to the manufacturer’s instructions . An irrelevant morpholino oligonucleotide ( Standard Control Oligo , GENE-TOOLS ) was used as control . After 16 hr , lu2EC were washed three times with PBS and starved for 24 hr in the following medium: DMEM High Glucose , 2 mM L-glutamine , 100 U/l penicillin streptomycin , 1 mM sodium pyruvate , 25 mM HEPES , 0 . 5% FBS . Conditioned media ( CM ) were collected and centrifuged for 10 min to remove cell debris . Total protein extraction and immunoblot were performed as previously described in ( Magrini et al . , 2014 ) . Briefly , total proteins were extracted by solubilizing cells in Laemmli buffer ( 4% SDS , 16% glycerol , 40 mM Tris-HCl pH 6 . 8 ) . To ensure an equal loading , lysates were quantified using Pierce BCA protein assay kit ( Thermo Fisher Scientific ) and 20 μg loaded on gel . The amount of EC-derived CM analyzed by immunoblotting was normalized against the number of producing cells . Lysates and CM were separated using SDS–PAGE and analyzed by western blotting . The following primary antibodies were used: anti-NOVA2 C-16 ( 1:200; Santa Cruz Biotechnology ) , anti-α-Tubulin ( 1:50 , 000; Sigma-Aldrich ) , anti-HA High Affinity ( 1:1000; Roche ) , anti-T7 tag ( 1:5000; Novagen ) , anti-Vinculin ( 1:5000 Millipore ) , anti-L1CAM [1 μg/ml; ( Magrini et al . , 2014 ) ] , anti-L1CAM cytoplasmic domain [1:8000; Ral1cd , kindly provided by V Lemmon; ( Schaefer et al . , 2002 ) ]; anti-FGFR1 ( 1:200; clone M2F12 , Santa Cruz Biotechnology ) and anti-phospho-FGFR1 ( Y653/654 ) ( 1:1000 Santa Cruz Biotechnology ) . The following secondary antibodies conjugated to horseradish peroxidase ( Jackson ImmunoResearch ) were used: anti-Mouse ( 1:5000 ) , anti-Goat ( 1:5000 ) , anti-Rat ( 1:5000 ) and anti-Rabbit ( 1:10000 ) . Immunostained bands were detected using the chemiluminescent method ( Clarity ECL Western Blotting Substrate , Bio-Rad ) and images were obtained by ChemiDoc Imaging Systems ( Bio-Rad ) . ECs were grown on gelatin-coated coverslips and then fixed with 4% PFA for 10 min at room temperature ( RT ) . Membrane permeabilization was obtained incubating coverslips in ice-cold PBS and 0 . 5% Triton X-100 for 3 min at 4°C . Cells were then incubated for 1 hr at RT in a humid chamber with blocking solution ( PBS , 2% BSA , 5% donkey serum , and 0 . 05% Triton X-100 ) . Samples were then incubated for 2 hr with primary antibodies ( anti-L1CAM , 1 μg/ml ) diluted in blocking buffer , followed by the incubation with Cy3 donkey anti-Rabbit ( 1:600 , Listarfish ) secondary antibodies ( 45 min at RT ) . Finally , samples were washed in PBS and nuclei counterstained with DAPI solution ( 0 . 2 μg/ml , Sigma-Aldrich ) . Confocal microscopy was performed with a Leica SP2 confocal microscope equipped with a motorized stage and violet ( 405 nm laser diode ) , blue ( 488 nm Argon ) , yellow ( 561 nm laser diode ) , and red ( 633 nm HeNe laser ) excitation laser lines . For quantification purposes , at least five fields for each condition were counted and the average of the positive/negative cells was calculated . The statistical difference among groups was determined as described in the Statistical analysis section . Fresh tissue samples were obtained upon informed consent from patients undergoing surgery at the Gynecology Division of the European Institute of Oncology ( Milan ) . Sample collection was performed under the protocol n . R789-IEO approved by the Ethics Committee of the European Institute of Oncology . The immunohistochemical analysis of L1CAM expression was carried out on a panel of high-grade serous ovarian carcinoma . Fresh samples were 4% PFA-fixed and paraffin embedded . After an overnight at 37°C , tumor sections ( 3 µm ) were deparaffinized using Leica ST5020 Multistainer . Tissue sections were treated with the antigen unmasking solution EDTA pH 8 in pre-warmed water bath at 95°C for 50 min and then endogenous peroxidases were blocked using 3% Hydrogen peroxide solution ( Carlo Erba ) . Tissue sections were incubated in blocking solution ( TBS , 4% BSA , 0 . 05% Triton X-100 ) for 1 hr . Samples were then incubated for 2 hr with primary antibodies ( anti-L1CAM , 1 μg/ml; anti-NOVA2 C-16 , 1:100 SantaCruz; anti-CD31 , 1:50 Abcam ) followed by secondary antibodies incubation ( Dako EnVision + System HRP Labelled Polymer or Goat-on-Rodent HRP Polymer from Biocare ) for 30 min at RT . Dako chromogen substrate ( Liquid DAB +Substrate Chromogen System Ref . K3468 ) or Vina Green Chromogen Kit ( Biocare Medical ) were used for signal detection . Samples were counterstained using Hematoxylin solution ( Leica ) . Double immunohistochemistry for NOVA2 and L1CAM ( using the same primary antibodies as above ) was performed as previously described ( Giampietro et al . , 2015 ) . Pictures of stained sections were acquired with the Aperio ScanScope XT instrument . For quantification of CD31 and NOVA2 staining , five different fields were counted for each section from five independent samples of normal ovaries or ovarian cancer . Vessel density was determined calculating the number of CD31-positive vessels per area unit ( mm2 ) using a proprietary tool of the Aperio ImageScope software ( Leica Biosystems Imaging ) . Total RNA was isolated from both cultured cells and paraffin-embedded samples . RNA from cultured cells was obtained by using the RNeasy Mini Kit ( QIAGEN ) , while tissue RNA was extracted with AllPrep DNA/RNA FFPE ( QIAGEN ) according to manufacturer’s instructions . cDNA was obtained starting from 500 to 1000 ng of total RNA with Superscript IV RT cDNA synthesis kit ( Invitrogen ) according to the manufacturer’s instructions and an aliquot ( 1–2 μl ) of cDNA was then PCR-amplified ( with GoTaq DNA Polymerase , Promega ) . The percentage of exon inclusion was calculated as the ratio between the intensity of the band of L1cam transcripts with the exon included and the total intensity of the L1cam bands . Splicing of L1CAM in human normal tissues ( brain , heart , liver , kidney , lung and trachea , Clontech ) , normal human ovary and breast samples ( Ambion ) and human EC lines ( HUVEC and hCMEC/D3 ) was analyzed with primers hL1E23_F and hL1E28_R ( Supplementary file 1 ) . RT-PCR analysis on FFPE samples of ovarian cancer tissue was performed with the primers hL1E24_F and hL1E26_R ( Supplementary file 1 ) after selection of suitable areas of the tumors by a trained pathologist ( GB ) . Splicing of L1cam in mouse tissues ( brain , cortex , testis , lung , tongue and liver ) , freshly purified ECs from mouse lung and EC lines was analyzed with primers mL1E23_F and mL1E28_R ( Supplementary file 1 ) . Mouse tissues were obtained from Karolinska Institutet ( Stockholm , Sweden ) and IRCCS San Raffaele Scientific Institute ( Milan , Italy ) , in accordance to Institutional Animal Care and Use Committees . Band intensity on agarose gel was quantified with the NIH Image J program ( version 1 . 50i ) . All PCR products were verified by sequencing . For RT-qPCR experiments , cDNA samples were amplified with QuantiTect SYBR Green PCR ( QIAGEN ) by using LightCycler 480 ( Roche ) . Target transcript levels were normalized to those of GAPDH or Ubb housekeeping genes . All primers used in RT-qPCR are listed in Supplementary file 1 . CLIP assay was performed as previously described ( Paronetto et al . , 2014; Paronetto et al . , 2011 ) . moEC were irradiated once with 150 mJ/cm2 in a Stratlinker 2400 at 254 nm . Cell suspension was centrifuged at 4000 rpm for 3 min , and pellet was incubated for 10 min on ice in lysis buffer [50 mM Tris-HCl , pH 7 . 4 , 100 mM NaCl , 1% Igepal CA-630 ( Sigma-Aldrich ) , 0 . 1% SDS , 0 . 5% sodium deoxycholate , 0 . 5 mM Na3VO4 , 1 mM DTT , protease inhibitor cocktail ( Sigma-Aldrich ) , and RNase inhibitor ( Promega ) ] . Samples were briefly sonicated and incubated with 10 μl of 1:1000 RNase I ( 100 U/µL , Ambion ) dilution and 2 μl of DNase ( 2 U/µL , Ambion ) for 3 min at 37°C shaking at 1100 rpm , and then centrifuged at 15 , 000 g for 10 min at 4°C . One milligram of extract was immunoprecipitated using anti-NOVA2 antibody ( C-16 , Santa Cruz Biotechnology ) or purified IgG ( negative control ) in the presence of protein A/G magnetic Dynabeads ( Life Technologies ) . Immunoprecipitates were incubated overnight at 4°C under constant rotation . After stringent washes with high salt buffer ( 50 mM Tris-HCl , pH 7 . 4 , 1 M NaCl , 1 mM EDTA , 1% Igepal CA-630 , 0 . 1% SDS , 0 . 5% sodium deoxycholate ) , beads were equilibrated with PK buffer ( 100 mM Tris-HCl , pH 7 . 4 , 50 mM NaCl , 10 mM EDTA ) . An aliquot ( 10% ) was kept as input lysate , while the rest was treated with 50 µg Proteinase K and incubated for 20 min at 37°C shaking at 1100 rpm . 7 M urea was added to the PK buffer and incubated for further 20 min at 37°C shaking at 1100 rpm . The solution was collected and phenol/CHCl3 ( Ambion ) was added . After incubation for 5 min at 30°C shaking at 1100 rpm , phases were separated by centrifuging for 5 min at 13000 rpm at RT . The aqueous layer was transferred into a new tube and precipitated by addition of 0 . 5 μl glycoblue ( Ambion ) , 3 M sodium acetate pH 5 . 5 and 100% ethanol . After mixing , the solution containing retained RNA was precipitated overnight at −20°C . RNA extracted from both the input material and the immunoprecipitates was then analyzed by RT-qPCR as described in the above paragraph . The binding was expressed as percentage of the input material . HeLa cells were transiently co-transfected by using Lipofectamine 3000 with 500 ng of p-L1 WT minigene and either 500 ng of protein expression vectors ( HA-NOVA2 , T7-hnRNPA1 and T7-SRSF3 ) or the empty vector . Five-hundred ng of each mutated minigene ( Mut3 and Mut5 ) were transfected with 250 ng of HA-NOVA2 expression vector plus 250 ng of the empty vector; as a control , we used 500 ng of the empty vector . Total RNAs were extracted from HeLa cells after 24 hr and analyzed by RT-PCR with primers p-L1_F and BGH_R annealing to mouse L1cam exon 25 and to the Bovine Growth Hormone ( BGH ) polyadenylation site , respectively . Primers are listed in Supplementary file 1 . For the production of secreted , recombinant L1-ΔTM ( Figure 2—figure supplement 1 ) , FreeStyle 293 F cells were cultured and transfected in FreeStyle medium . Four hours after transfection with pUPE . 06 . 45-L1-ΔTM , Primatone RL ( Sigma-Aldrich ) was added to the culture medium at the final concentration of 0 . 6% . The culture medium was harvested 7 days after transfection by 10 min centrifugation at 1000 g . The supernatant containing secreted L1-ΔTM was loaded at 0 . 5 ml/min on a 1 ml StrepTrap HP column ( GE Healthcare ) . The column was washed with 10 column volumes ( cv ) of buffer P , composed of 50 mM HEPES and 250 mM NaCl pH 7 . 5 . Elution was performed with 1 cv of the same buffer supplemented with 5 mM d-Desthiobiotin ( Sigma-Aldrich ) . After elution , protein was incubated overnight at 4°C with TEV protease to remove the N-terminal affinity tags . The resulting protein was concentrated by centrifugation using Vivaspin Turbo 15 , 100000 MWCO centrifugal filters ( Sartorius ) to a volume less than 500 μl and then loaded onto a Superdex 200 10/300 GL ( GE Healthcare ) gel filtration column equilibrated with buffer P . The elution peak corresponding to L1-ΔTM as judged by SDS-PAGE analysis was collected and concentrated to 1 mg/ml . The purified sample was then flash-frozen in liquid nitrogen and stored at −80°C until further usage . Confluent mouse ECs were washed three times with PBS 1x and cultured for 48 hr in the following medium: DMEM High Glucose , 2 mM L-glutamine , 100 U/l penicillin streptomycin , 1 mM sodium pyruvate , 25 mM HEPES , 0 . 5% FBS . Conditioned media ( CM ) were collected and centrifuged for 10 min to remove cell debris . A Matrigel-based tubulogenesis assay was performed to assess the ability of ECs to form an organized capillary-like network . Before proceeding with the assay , 96-well plate was coated with 50 μl/well of Growth Factor-Reduced Matrigel ( BD Biosciences ) and left for 1 hr at 37°C for gelification . To assess the cell-autonomous effect of L1-ΔTM , transduced moEC were plated on Matrigel-coated plates in complete growth medium . Tube-like structures were manually counted under the microscope after 24 hr . To assess the paracrine effect of L1-ΔTM , confluent moEC were cultured overnight in the same medium used for CM production . The day after , moEC were seeded on polymerized Matrigel-coated wells in the appropriate CM or , where indicated , were treated with different concentrations of recombinant , purified L1-ΔTM . After 8 hr of incubation at 37°C , tubes-like structures were counted under the microscope . To block L1-ΔTM activity , transduced moEC were pre-incubated for 1 hr at 37 ˚C with 10 µg/ml of an anti-L1CAM blocking antibody [clone 324 ( Di Sciullo et al . , 1998; Appel et al . , 1993 ) ] or with control rat IgG ( Sigma-Aldrich ) . Cells were then seeded on Matrigel-coated wells in complete growth medium containing 10 µg/ml of anti-L1CAM 324 ( or control rat IgG ) and tube-like structures were counted manually 8 hr later under the microscope . Where indicated , transduced moEC were pre-incubated for 1 hr at 37 ˚C with FGFR inhibitor PD173074 at a final concentration of 70 nM . Cells were then seeded on Matrigel-coated wells in normal growth medium containing 70 nM of PD173074 and tube-like structures were counted manually after 8 hr under the microscope . Both image acquisition and cell counts were performed using EVOS FL Imaging System . Cells were plated in triplicate ( technical replicates ) and the experiment was performed three times ( biological replicates ) . All animal studies were performed following a protocol approved by the fully authorized animal facility of European Institute of Oncology and by the Italian Ministry of Health ( as required by the Italian Law ) ( IACUC n . 1256/2015 ) and in accordance to EU directive 2010/63 . The sample size estimation was based on previous studies and pilot experiments . C57Bl/6 mice were injected into the right flank with 200 μl of CM derived from mouse ECs expressing either L1-ΔTM or L1-FL or from control ECs in a final volume of 600 μl of Growth Factor-Reduced Matrigel ( BD Biosciences ) . Matrigel containing 0 . 5 μg of FGF2 was used as positive control . Groups were composed by three mice for each construct . Plugs were removed 7 days after injection , fixed in 4% PFA and paraffin embedded . Sections from fixed plugs were stained for CD31 as described above . The number of CD31-positive vessels that invaded Matrigel plugs were evaluated by manual counting of five different fields per section using Axioskop two microscope ( Leica Biosystems ) . To determine the prognostic relevance of NOVA2 , overall survival curves of ovarian cancer patients were built analyzing the GDC TCGA dataset with the UCSC Xena web tool ( http://xena . ucsc . edu/ ) . Survival plots were drawn using the Kaplan-Meier method and patients were stratified according to NOVA2 expression using median as threshold value . The log-rank Mantel-Cox test was employed to determine any statistical difference between the survival curves of the cohorts . We previously performed RNA sequencing on NOVA2-knockdown moEC versus their parental control ( Giampietro et al . , 2015 ) . To capture additional NOVA2-regulated AS events in that experimental system , we first aligned each of the four samples independently ( two control and two knockdown replicates ) with the Vertebrate Alternative Splicing and Transcription Tools ( vast-tools ) ( Tapial et al . , 2017 ) ( https://github . com/vastgroup/vast-tools ) . Then both replicates of each experimental point were pooled with vast-tools merge using the default parameters . This increased markedly the read coverage at each of the exon-exon junctions per experimental condition . Then , we applied vast-tools compare to perform a differential splicing analysis between the pooled control and the pooled knockdown samples , using a cutoff of delta percent spliced in ( ΔPSI ) of 15 with default settings . Independent experiments were considered as biological replicates . When performed , technical replicates deriving from the same biological replicate were averaged . For in vivo experiments , each mouse represented one biological replicate . For staining of human tissues , each patient represented one biological replicate . Data are expressed as mean ±SEM , calculated from at least three independent experiments . Student’s two-tailed t test or ANOVA multiple comparison test , followed by Tukey’s post hoc analysis , were used to compare two or three or more groups , respectively , and to determine statistical significance ( GraphPad Prism 5 ) . The correlation between the expression of L1-ΔTM and tumor vessel density was assessed using the Spearman rank correlation coefficient . Differences were considered significant at p<0 . 05 . Asterisks correspond to p-value calculated by two-tailed , unpaired , t-test ( *p<0 . 05 , **p<0 . 01 , ***p<0 . 001 ) . The sample size estimation was based on previous studies and pilot experiments . | Growing tumors stimulate the formation of new blood vessels to supply the oxygen and nutrients the cancerous cells need to stay alive . Stopping tumors from forming the blood vessels could therefore help us to treat cancer . To do so , we need to understand how different proteins control when and how blood vessels develop . Cells make proteins by first ‘transcribing’ genes to form RNA molecules . In many cases , the RNA then goes through a process called alternative splicing . Proteins known as splicing factors cut out different segments of the RNA molecule and stick together the remaining segments to form templates for protein production . This enables a single gene to produce many different variants of a protein . Angiolini , Belloni , Giordano et al . have now studied mouse and human versions of the cells that line the blood vessels grown by tumors . This revealed that a splicing factor called NOVA2 targets a protein called L1CAM , which is normally responsible for gluing adjacent cells together . Angiolini et al . found that NOVA2 splices L1CAM into a form not seen before . Instead of remaining anchored to cell surfaces , the newly identified form of L1CAM is released into the blood circulation , where it stimulates new blood vessels to grow . Samples taken from the blood vessels of human ovarian tumors showed high levels of both NOVA2 and the modified form of L1CAM , while blood vessels in healthy tissue contain no , or very low levels of both proteins . Therefore , if the new form of L1CAM can be detected in the blood , it could be used to help cancer diagnosis , and to indicate which patients would benefit from treatments that restrict the growth of blood vessels in tumors . Further work is now needed to explore these possibilities . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"cell",
"biology",
"cancer",
"biology"
] | 2019 | A novel L1CAM isoform with angiogenic activity generated by NOVA2-mediated alternative splicing |
Aberrant activation of anaplastic lymphoma kinase ( ALK ) has been described in a range of human cancers , including non-small cell lung cancer and neuroblastoma ( Hallberg and Palmer , 2013 ) . Vertebrate ALK has been considered to be an orphan receptor and the identity of the ALK ligand ( s ) is a critical issue . Here we show that FAM150A and FAM150B are potent ligands for human ALK that bind to the extracellular domain of ALK and in addition to activation of wild-type ALK are able to drive 'superactivation' of activated ALK mutants from neuroblastoma . In conclusion , our data show that ALK is robustly activated by the FAM150A/B ligands and provide an opportunity to develop ALK-targeted therapies in situations where ALK is overexpressed/activated or mutated in the context of the full length receptor .
Activation of anaplastic lymphoma kinase ( ALK ) is commonly due to fusion of the ALK kinase domain with a dimerization partner that drives activation , however , ALK activation also occurs in the context of the full length receptor , for example , as activating point mutations in neuroblastoma ( Maris et al . , 2007; Carén et al . , 2008; Chen et al . , 2008; George et al . , 2008; Janoueix-Lerosey et al . , 2008; Mosseé et al . , 2008; Hallberg and Palmer , 2013 ) . In many additional tumor types ALK overexpression and activation has been described , and it is unclear whether this is dependent on the activity of a ligand ( Hallberg and Palmer 2013 ) . The ALK receptors in both Drosophila and Caenorhabditis elegans have well defined ligands – Jeb ( Englund et al . , 2003; Lee et al . , 2003; Stute et al . , 2004 ) and HEN-1 ( Ishihara et al . , 2002 ) , respectively . In contrast , vertebrate ALK has long been considered as an orphan receptor . The human ALK locus encodes a classical receptor tyrosine kinase ( RTK ) comprising a unique extracellular ligand-binding domain , a transmembrane domain and an intracellular tyrosine kinase domain ( Hallberg and Palmer , 2013 ) . The extracellular portion of ALK which contains two MAM domains ( named after meprin , A-5 protein and receptor protein tyrosine phosphatase μ ) , a glycine-rich region ( GR ) and a LDLa domain , is unique among the RTKs . ALK , and the related leukocyte tyrosine kinase ( LTK ) RTK , share kinase domain similarities as well as a GR in the membrane proximal portion of their extracellular domains ( ECDs ) ( Iwahara et al . , 1997; Morris et al . , 1997 ) . Recent screening of the extracellular proteome identified two novel secreted proteins as ligands for LTK – family with sequence similarity 150A ( FAM150A ) and family with sequence similarity 150B ( FAM150B ) . Both bind to the ECD of the receptor leading to activation of downstream signaling in cell culture models ( Zhang et al . , 2014 ) . FAM150A and FAM150B are unique , displaying homology only with one another but not with any other proteins in mammals ( Zhang et al . , 2014 ) . Furthermore , we found the reported strong expression of FAM150B in the human adrenal gland ( Zhang et al . , 2014 ) intriguing , given the role of ALK in neuroblastoma . Here we report the identification of FAM150A and FAM150B as potent ligands for human ALK . We investigated ALK activation by FAM150A and FAM150B proteins in PC12 cell neurite outgrowth assays where we observed a strong activation of ALK signaling . Conditioned medium containing either FAM150A or FAM150B was able to activate endogenous ALK signaling in neuroblastoma cells . We also employed the model organism Drosophila melanogaster as a readout for activation of ALK by FAM150A and FAM150B , showing that FAM150 proteins are able to robustly drive human ALK activation when ectopically coexpressed in the fly . FAM150A and FAM150B bind to the ECD of ALK and , in addition to activation of wild-type ALK , are able to drive ‘superactivation’ of activated ALK mutants from neuroblastoma . The GR of the ALK receptor ECD is important for FAM150 activation , and monoclonal antibodies ( mAb ) recognizing the GR of ALK are able to inhibit activation of ALK by FAM150A . In conclusion , our data show that ALK is robustly activated by FAM150A/B finally providing an answer to the identity of the elusive ligands for this RTK .
ALK and the related LTK share similarity in their membrane proximal ECD in the form of a glycine-rich domain that is ∼250 amino acids in length ( Figure 1A , GR depicted in grey ) . This domain contains multiple runs of up to eight glycine residues , and is unique to ALK and LTK within the human genome . The importance of the GR in ALK has been highlighted in Drosophila studies , where four independent point mutations leading to exchange of single glycine residues result in complete loss of function in vivo ( Englund et al . , 2003 ) ( Figure 1—figure supplement 1 ) . The similarity between ALK and LTK within the GR is ∼70% , with amino acid identity of 55% , containing a total of 51 conserved glycine residues ( Figure 1B ) . Given this similarity , and the important role of the glycine-rich domain for function in Drosophila , we hypothesized that FAM150A and FAM150B , which were recently reported as ligands for LTK ( Zhang et al . , 2014 ) may act as ligands for ALK . 10 . 7554/eLife . 09811 . 003Figure 1 . FAM150A and FAM150B activate ALK . ( A ) Schematic overview of human anaplastic lymphoma kinase ( ALK ) and leukocyte tyrosine kinase ( LTK ) protein domain structures . ALK and LTK share a membrane proximal extracellular glycine-rich region ( GR , grey ) , transmembrane and an intracellular tyrosine kinase domain ( red ) . In addition , the extracellular region of ALK contains two MAM domains ( purple ) and an LDLa-motif ( yellow ) . ( B ) Alignments of the GR of ALK and LTK , conserved runs of glycine residues are highlighted in bold with red asterisks . ( C ) Neurite outgrowth in PC12 cells expressing either vector control , FAM150A , FAM150B , ALK , FAM150A and ALK , FAM150B and ALK or ALK-F1174L quantified in ( D ) . Experiments were performed in triplicate and each sample within an experiment was performed in duplicate ( error bars indicate SD ) . ( E ) Whole cell lysates from PC12 cells expressing either vector control , FAM150A , ALK , FAM150A and ALK or ALK-F1174L were analyzed by immunoblot analysis of ALK , pALK-Y1604 ( arrowheads ) , FAM150A and ERK1/2 . Pan-ERK was employed for equal loading . ( F ) Whole cell lysates from PC12 cells expressing either vector control , FAM150B , ALK , FAM150B and ALK or ALK-F1174L were analyzed by immunoblot analysis of ALK , pALK-Y1604 ( arrowheads ) , HA ( FAM150B ) and pERK1/2 . Pan-ERK was employed for equal loading . DOI: http://dx . doi . org/10 . 7554/eLife . 09811 . 00310 . 7554/eLife . 09811 . 004Figure 1—figure supplement 1 . Glycine residues in the glycine-rich region of Drosophila ALK are critical for function . ( A ) Schematic overview of anaplastic lymphoma kinase ( ALK ) and leukocyte tyrosine kinase ( LTK ) protein domain structure . ALK and LTK share a membrane proximal extracellular glycine-rich region ( grey ) , transmembrane and an intracellular tyrosine kinase domain ( red ) . In addition , the extracellular region of ALK contains two MAM domains ( purple ) and an LDLa-motif ( yellow ) . ( B ) Alignment of the glycine-rich region of ALK and LTK , conserved runs of glycine residues are highlighted in bold with red asterisks . Highlighted in yellow are single glycine residues mutated to either aspartic acid or glutamate that result in complete loss of function in Drosophila ALK ( Englund et al . , 2003 ) , and the conserved residue in human ALK is indicated . DOI: http://dx . doi . org/10 . 7554/eLife . 09811 . 00410 . 7554/eLife . 09811 . 005Figure 1—figure supplement 2 . Conservation of phosphoepitopes in the intracellelular domains of ALK and LTK . ( A ) Schematic overview of anaplastic lymphoma kinase ( ALK ) and leukocyte tyrosine kinase ( LTK ) protein domain structure . ( B ) Alignment of the intracellular region of ALK and LTK , highlighting in yellow the Y1278 phosphoepitope in the activation loop which shares significant homology between ALK and LTK , and the Y1604 phosphoepitope of ALK which is not found in LTK . DOI: http://dx . doi . org/10 . 7554/eLife . 09811 . 005 In order to initially test whether FAM150A or FAM150B could activate ALK we assayed neurite outgrowth activity in PC12 cells . Expression of either FAM150A , or FAM150B or ALK alone , did not lead to activation . This was in contrast to the strong activation seen with the positive control ALK-F1174L which is a well characterized constitutively active ALK neuroblastoma mutation . However , coexpression with either FAM150A or FAM150B led to robust activation of ALK neurite outgrowth activity ( Figure 1C , quantified in 1D ) . This ALK activation was visualized by phosphorylation of Y1604 in the tail of the ALK intracellular domain ( Figure 1E , F; Figure 1—figure supplement 2 ) and was further associated with stimulation of downstream signaling such as the phosphorylation of the downstream target ERK1/2 ( Figure 1E , F ) . Thus in PC12 cells FAM150A and FAM150B act to stimulate signaling via the wild-type ALK receptor , in a manner that leads to phosphorylation of the ALK receptor . We next tested the ability of exogenously produced FAM150A or FAM150B to activate ALK , employing medium conditioned with FAM150A or FAM150B to activate PC12 cells expressing ALK . Addition of conditioned medium from cells expressing FAM150A or FAM150B led to strong activation of ALK in receiving cells that expressed ALK , as measured by neurite outgrowth and activation of downstream signaling ( Figure 2A , B ) . We then examined whether ALK signaling activity induced by either FAM150A or FAM150B was sensitive to ALK inhibition , employing the ALK inhibitor crizotinib ( Zou et al . , 2007 ) . We observed that ALK activation by FAM150A or FAM150B was inhibited by addition of 250 nM crizotinib ( Figure 2A , B ) . 10 . 7554/eLife . 09811 . 006Figure 2 . Conditioned medium containing either FAM150A or FAM150B activates endogenous ALK . ( A ) Neurite outgrowth in PC12 cells expressing either vector control or anaplastic lymphoma kinase ( ALK ) were cultured in medium from Human Embryonic Kidney ( HEK ) 293 cells transfected with either FAM150A or FAM150B , quantified below . Experiments were performed in triplicate and each sample within an experiment was performed in duplicate . Values represent mean ± SD from at least three independent experiments . ( B ) Whole cell lysates from PC12 cells expressing either vector control or ALK stimulated with medium from HEK293 cells transfected with vector control , FAM150A or FAM150B were analyzed by immunoblot . Analysis was carried out in the presence or absence of 250 nM crizotinib . Detection of ALK activation was visualized with pALK-Y1604 ( arrowheads ) , ALK and pERK1/2 in whole cell lysates . The presence of FAM150A in supernatants was confirmed with anti-FAM150A antibodies , while the presence of FAM150B-HA was confirmed with anti-HA antibodies . Pan-ERK was employed for equal loading . ( C ) IMR32 cells harboring a wild-type ALK receptor were stimulated for 20 min with medium from HEK293 cells transfected with either vector control , FAM150A or FAM150B prior to analysis by immunoblot . Analysis was carried out in the presence or absence of 250 nM crizotinib . Stimulation with the ALK activating antibody mAb46 was employed as positive control . Detection of ALK activation was visualized with ALK , pALK-Y1604 ( arrowheads ) and pERK1/2 . Pan-ERK was employed for equal loading . ( D ) IMR32 cells harboring a wild-type ALK receptor stimulated with increased amounts of recombinant His-tagged FAM150A purified from Sf21 cells . Detection of ALK activation was visualized with ALK , pALK-Y1278 ( arrowheads ) and pERK1/2 . Pan-ERK was employed for equal loading . ( E ) Time course of IMR32 cells stimulated with FAM150A conditioned medium . Stimulation with ALK activating antibody mAb46 was employed as positive control . Detection of ALK activation was visualized with ALK , pALK-Y1604 ( arrowheads ) , pERK5 and pERK1/2 . Pan-ERK was employed for equal loading . ( F ) Time course of IMR32 cells harboring a wild type ALK receptor stimulated with FAM150B conditioned medium . Stimulation with ALK activating antibody mAb46 was employed as positive control . Detection of ALK activation was visualized with ALK , pALK-Y1604 ( arrowheads ) , pERK5 and pERK1/2 . Pan-ERK was employed for equal loading . DOI: http://dx . doi . org/10 . 7554/eLife . 09811 . 00610 . 7554/eLife . 09811 . 007Figure 2—figure supplement 1 . FAM150 proteins interact with heparin . ( A ) Purified FAM150A-HIS was incubated with heparin-agarose overnight at 4º°C prior to extensive washing . Bound proteins were eluted , separated on 15% sodium dodecyl sulfate-polyacrylamide gel electrophoresis ( SDS-PAGE ) and analyzed by immunoblotting with anti-FAM150A . Purified FAM150A-His ( 5 µg ) was employed as input control . ( B ) Conditioned media from cells transfected with pcDNA3 control , pTT5-FAM150A-HA , or pcDNA3-FAM150B-HA was incubated with heparin-agarose for 3 hr at 4º°C prior to extensive washing . Bound proteins were eluted , separated on 13% SDS-PAGE and analyzed by immunoblotting with anti-HA . Conditioned media ( 50 µl ) was employed as input control . *blue asterisk indicates FAM150A-HA , *red asterisk indicates FAM150B-HA in conditioned medium input . DOI: http://dx . doi . org/10 . 7554/eLife . 09811 . 00710 . 7554/eLife . 09811 . 008Figure 2—figure supplement 2 . Investigation of the effect of Heparin and FAM150A on ALK activation . ( A ) IMR32 cells were treated with FAM150A alone in the presence or absence of 10 µg/ml heparin alone for the times indicated , prior to analysis by immunoblot . Anaplastic lymphoma kinase ( ALK ) signaling was visualized with pERK1/2 . Tubulin was employed for equal loading . ( B ) NB1 cells were treated with FAM150A alone or heparin alone or with FAM150A and heparin together at the indicated concentrations for 10 min , prior to analysis by immunoblot . Detection of ALK activation was visualized with ALK , pALK-Y1604 ( arrowhead ) and pERK1/2 . Pan-ERK was employed for equal loading . DOI: http://dx . doi . org/10 . 7554/eLife . 09811 . 008 To address whether endogenous levels of ALK responded to medium conditioned with either FAM150A or FAM150B , we employed the IMR-32 neuroblastoma cell line that expresses wild-type ALK . Once again , we observed robust activation of ALK by the addition of either FAM150A-containing or FAM150B-containing conditioned medium , employing pALK-Y1604 and pERK1/2 as readout ( Figure 2C ) . The activation of ALK and downstream signaling events in IMR-32 cells were effectively blocked by the addition of 250 nM crizotinib ( Figure 2C ) . The activation of exogenous ALK in PC12 cells or endogenous ALK in IMR-32 cells was reproduced with the addition of recombinant FAM150A purified from Sf21 cells ( Figure 2D ) . Unfortunately we could not produce sufficient amounts of recombinant FAM150B protein , precluding a similar analysis with FAM150B . More careful examination of ALK activation by FAM150A or FAM150B containing conditioned media in IMR-32 cells revealed activation of signaling at 2–5 min , peaking around 20 min before decreasing at later time points . This response was observed at the level of ALK receptor phosphorylation ( pALK-Y1604 ) , and with downstream pERK5 ( Umapathy et al . , 2014 ) and pERK1/2 ( Figure 2E , F ) . Here we employed antibodies recognizing pALK-Y1604 , allowing differentiation of ALK activation in IMR-32 cells from that of LTK ( Figure 1—figure supplement 2 ) . Thus we observed that exogenously produced FAM150A and FAM150B proteins stimulate ALK signaling in both PC12 cells and IMR-32 cells . We also examined potential modulation of ALK activation by FAM150A/B upon addition of heparin , which has recently been reported to activate ALK ( Murray et al . , 2015 ) . While we were able to see strong binding of purified FAM150A protein as well as FAM150A-HA and FAM150B-HA from conditioned medium to heparin-agarose , we did not observe any additional activation of ALK by FAM150 proteins in the presence of heparin ( Figure 2—figure supplement 1; Figure 2—figure supplement 2 ) . We next analyzed the ability of FAM150A or FAM150B to activate human ALK in a Drosophila model , which offers a clear readout . Neither the Drosophila Alk ligand Jeb ( Englund et al . , 2003; Lee et al . , 2003; Stute et al . , 2004 ) nor previously proposed vertebrate ligands , that is , human midkine ( MDK ) and pleiotrophin ( PTN ) are able to activate either mouse or human ALK ( Yang et al . , 2007; Hugosson et al . , 2014 ) . Expression of either FAM150A or FAM150B in the developing eye , using the GMR-Gal4 driver , resulted in normal eye morphology ( Figure 3A ) . In contrast , expression of constitutively active ALK-F1174S described in neuroblastoma patients results in a rough eye morphology ( Figure 3A ) ( Martinsson et al . , 2011 ) , while no eye phenotype was observed upon the expression of wild-type human ALK alone ( Figure 3A ) ( Martinsson et al . , 2011 , Schonherr , Ruuth et al . , 2011 , Schonherr , Ruuth et al . , 2011 , Chand et al . , 2013; Hugosson et al . , 2014 ) . This can be compared with coexpression of either FAM150A or FAM150B together with human ALK which led to a rough eye phenotype , proving that both FAM150A and FAM150B were able to activate human ALK in this in vivo system ( Figure 3A , B ) . 10 . 7554/eLife . 09811 . 009Figure 3 . Expression of either FAM150A or FAM150B is sufficient for the activation of wild-type ALK . ( A ) Ectopic expression of either FAM150A or FAM150B together with the wild-type human anaplastic lymphoma kinase ( ALK ) receptor in the Drosophila eye with the GMR-Gal4 driver disrupts the highly organized pattern of ommatidia of the Drosophila eye and generates a rough eye phenotype . Images of adult Drosophila eyes ectopically expressing wild-type ALK in the presence of either FAM150A or FAM150B are shown . Controls expressing either wild-type ALK , FAM150A or FAM150B alone do not display a rough eye phenotype . The constitutively active ALK-F1174S mutant was employed as positive control . ( B ) Phenotypes observed in adult flies upon GMR-Gal4 driven expression of either FAM150A or FAM150B together with the wild-type ALK at two different temperatures , 18°C and 25°C . DOI: http://dx . doi . org/10 . 7554/eLife . 09811 . 009 We next examined binding of the human ALK ECD with purified FAM150A by ELISA and Biacore surface plasmon resonance ( SPR ) analysis . Both ELISA and Biacore analysis showed that FAM150A binds specifically to the ALK ECD ( Figure 4A; Figure 4—figure supplement 1 ) . Purified human FAM150A ( Zhang et al . , 2014 ) binds to ALK-ECD-Fc with a dissociation constant of ∼20 nM ( Figure 4A ) . We also examined the ability of either FAM150A or FAM150B to bind human ALK by immunoprecipitation and immunofluorescence analysis ( Figure 4B–D ) . Here we observed that both FAM150A and FAM150B could be independently immunoprecipitated with the ALK receptor , and that ALK could be immunoprecipitated with either FAM150A or FAM150B in the reciprocal pull-down experiments ( Figure 4B , C; Figure 4—figure supplement 2 ) . Since FAM150A/B also bind LTK , we asked whether ALK and LTK interacted in the presence of FAM150A . Indeed , both ALK and FAM150A were found to immunoprecipitate with LTK ( Figure 4—figure supplement 3 ) . Immunofluorescence analysis of cells expressing human ALK confirmed these findings , with clear association of both HA-tagged FAM150A and HA-tagged FAM150B with ALK expressing cells observed ( Figure 4D ) . In these experiments we observed uptake of FAM150A and FAM150B into ALK-positive intracellular vesicles , suggesting that binding to ALK on the cell surface leads to uptake and internalization of the ALK-FAM150 complex ( Figure 4D ) . 10 . 7554/eLife . 09811 . 010Figure 4 . FAM150A and FAM150B bind to ALK and further activate signaling mediated by the R1275Q ALK neuroblastoma mutation . ( A ) Binding kinetics of purified FAM150A to extracellular domain of anaplastic lymphoma kinase ( ALK-ECD-Fc ) in a Biacore surface plasmon resonance ( SPR ) analysis . ( B ) FAM150A immunoprecipitates with human ALK . Immunoprecipitation with either anti-FLAG ( DYKDDDDK ) ( ALK ) or anti-HA ( FAM150A ) was performed and the resulting immunoprecipitates immunoblotted for the presence of ALK ( blue arrowheads ) and FAM150A ( red arrowheads ) , *indicates immunoglobulin light and heavy chains . ( C ) FAM150B immunoprecipitates with human ALK . Immunoprecipitation with either anti-FLAG ( ALK ) or anti-HA ( FAM150B ) was performed and the resulting immunoprecipitates immunoblotted for the presence of ALK ( blue arrowheads ) and FAM150B ( red arrowheads ) , *indicates immunoglobulin light and heavy chains . ( D ) Human Embryonic Kidney ( HEK ) 293 cells expressing ALK were incubated with either control or HA-tagged FAM150A or HA-tagged FAM150B conditioned medium prior to analysis by immunohistochemistry . Both HA-tagged FAM150A and HA-tagged FAM150B bind to ALK-expressing cells . Higher magnification panels indicate intracellular vesicles positive for both ALK and HA-tagged FAM150A/B . ( E ) NB1 and IMR32 neuroblastoma cells were treated with 2 µg/ml monoclonal antibodies ( mAB13 , mAb48 or mAb135 ) prior to stimulation with FAM150A . Detection of ALK activation was visualized with pALK-Y1604 ( arrowheads ) and pERK1/2 in whole cell lysates . Pan-ERK or tubulin were employed for equal loading . ( F ) Whole cell lysates from PC12 cells expressing either vector control or ALK-R1275Q were stimulated with medium from HEK293 cells transfected with vector control , FAM150A or FAM150B prior to analysis by immunoblot . Analysis was carried out in the presence or absence of 250 nM crizotinib . Detection of ALK activation was visualized with pALK-Y1604 ( arrowheads ) and pERK1/2 in whole cell lysates . Pan-ERK was employed for equal loading . Neurite outgrowth was performed in triplicate and each sample within an experiment was performed in duplicate ( error bars indicate SD ) . ( G ) CLB-GAR cells harboring the ALK-R1275Q mutant were stimulated for 30 min with medium from HEK293 cells transfected with either vector control , FAM150A or FAM150B prior to analysis by immunoblot . Analysis was carried out in the presence or absence of 250 nM crizotinib . Detection of ALK signaling activation was visualized with pERK1/2 . Pan-ERK was employed for equal loading . pERK1/2 intensity was analyzed from three independent experiments ( error bars indicate SD ) . DOI: http://dx . doi . org/10 . 7554/eLife . 09811 . 01010 . 7554/eLife . 09811 . 011Figure 4—figure supplement 1 . FAM150A binds to the extracellular domain of human ALK by ELISA . Purified FAM150A was coated on plates and binding of extracellular domain of ALK ( ALK-ECD-Fc ) was detected by anti-Fc HRP conjugate . LTK-ECD-Fc was used as a binding positive control . FAM150A binds specifically to the ECD of human ALK . DOI: http://dx . doi . org/10 . 7554/eLife . 09811 . 01110 . 7554/eLife . 09811 . 012Figure 4—figure supplement 2 . ALK interacts with both FAM150A and FAM150B . IgG control immunoprecipitations . ( A ) FAM150A coimmunoprecipitates with human anaplastic lymphoma kinase ( ALK ) . Immunoprecipitation with either anti-IgG control , anti-ALK or anti-HA ( FAM150A ) was performed and the resulting immonoprecipitates immunoblotted for the presence of ALK and FAM150A-HA . ( B ) FAM150B coimmunoprecipitates with human ALK . Immunoprecipitation with either anti-IgG control , anti-ALK or anti-HA ( FAM150B ) was performed and the resulting immonoprecipitates immunoblotted for the presence of ALK and FAM150B-HA . * indicates immunoglobulin light and heavy chains . DOI: http://dx . doi . org/10 . 7554/eLife . 09811 . 01210 . 7554/eLife . 09811 . 013Figure 4—figure supplement 3 . ALK and LTK interact in the presence of FAM150A . FAM150A and anaplastic lymphoma kinase ( ALK ) coimmunoprecipitate with leukocyte tyrosine kinase ( LTK ) . Immunoprecipitation with either anti-IgG control or anti-HA ( LTK ) was performed and the resulting immonoprecipitates immunoblotted for the presence of ALK ( mAb135 ) , LTK ( HA ) and FAM150A . *indicates immunoglobulin light and heavy chains . DOI: http://dx . doi . org/10 . 7554/eLife . 09811 . 01310 . 7554/eLife . 09811 . 014Figure 4—figure supplement 4 . Effect of deletion of the glycine rich domain of ALK on FAM150A and FAM150B binding . HA-tagged FAM150A coimmunoprecipitates both wild-type anaplastic lymphoma kinase ( ALK-WT ) and ALK lacking the glycine-rich region ( ALK-delGR ) . In contrast , HA-tagged FAM150B is able to immunoprecipitate ALK-WT , but not ALK-delGR . Input indicates 1/50th whole cell lysate , * indicates immunoglobulin light and heavy chains . DOI: http://dx . doi . org/10 . 7554/eLife . 09811 . 01410 . 7554/eLife . 09811 . 015Figure 4—figure supplement 5 . Effect of glycine mutations in the glycine-rich domain of ALK on FAM150A and FAM150B binding . ( A ) FAM150A coimmunoprecipitates with wild-type anaplastic lymphoma kinase ( ALK-WT ) and with ALK bearing mutations in conserved glycines 740 ( G740D ) , 823 ( G823D ) , 893 ( G893E ) and 934 ( G934D ) residing in the extracellular glycine-rich region ( GR ) of ALK . Immunoprecipitation with either anti-IgG control , anti-ALK ( D5F3 ) or anti-HA ( FAM150A ) was performed . ( B ) Control blot indicating that the various ALK glycine mutants are expressed . ( C ) FAM150B coimmunoprecipitates with ALK-WT but not with ALK bearing mutations in conserved glycines 740 ( G740D ) , 823 ( G823D ) , 893 ( G893E ) and 934 ( G934D ) residing in the extracellular GR of ALK . Immunoprecipitation with either anti-IgG control , anti-ALK ( D5F3 ) or anti-HA ( FAM150B ) was performed . ( D ) Control blot indicating that the various ALK glycine mutants are expressed . Immunoblots were analyzed for the presence of ALK ( D5F3 ) and FAM150A-HAor FAM150B-HA ( HA ) . * indicates immunoglobulin light and heavy chains . DOI: http://dx . doi . org/10 . 7554/eLife . 09811 . 01510 . 7554/eLife . 09811 . 016Figure 4—figure supplement 6 . Identification of monoclonal antibodies recognising the glycine-rich region of the ALK ECD . Monoclonal antibodies raised against anaplastic lymphoma kinase ( ALK ) ( ALK D5F3 , mAb13 , mAb48 and mAb135 ) were tested for their ability to recognize the glycine-rich region ( GR ) of ALK . Among those tested , mAb13 , mAb48 and mAb135 were found to recognize ALK-WT , but not ALK in which the GR had been deleted ( ALK-delGR ) . DOI: http://dx . doi . org/10 . 7554/eLife . 09811 . 01610 . 7554/eLife . 09811 . 017Figure 4—figure supplement 7 . FAM150A and FAM150B bind to ALK and further activate signaling mediated by the ALK-F1174L neuroblastoma mutant . Whole cell lysates from PC12 cells expressing either vector control or anaplastic lymphoma kinase ( ALK ) -F1174L were stimulated with medium from Human Embryonic Kidney ( HEK ) 293 cells transfected with vector control , FAM150A or FAM150B prior to analysis by immunoblot . Analysis was carried out in the presence or absence of 250 nM crizotinib . Detection of ALK activation was visualized with pALK-Y1604 ( arrowhead ) and pERK1/2 in whole cell lysates . Pan-ERK was employed for equal loading . Neurite outgrowth was performed in triplicate and each sample within an experiment was performed in duplicate ( error bars indicate SD ) DOI: http://dx . doi . org/10 . 7554/eLife . 09811 . 017 We further examined the role of the GR of ALK in FAM150A/B binding . Deletion of the GR in the ALK ECD did not affect FAM150A binding but led to loss of FAM150B binding ( Figure 4—figure supplement 4 ) . Given the importance of the GR of ALK in Drosophila ( Englund et al . , 2003 ) ( Figure 1—figure supplement 1 ) , we mutated these conserved residues in human ALK ( G740D , G823D , G893E and G934D ) and tested their effect on the FAM150-ALK interaction . Despite being well expressed , all four glycine mutations in the GR severely impaired binding of FAM150B to ALK . Consistent with our earlier results we were unable to see any effect of mutation of individual glycine on the FAM150A-ALK interaction ( Figure 4—figure supplement 5 ) . In previous work , we and others have generated antibodies to the ECD of ALK ( Moog-Lutz et al . , 2005 , Witek et al . , 2015 ) . We investigated whether any of these antibodies recognized the GR in the ALK ECD , and were able to identify three interesting candidates ( anti-ALK mAb13 , mAb48 and mAb135 , Figure 4—figure supplement 6 ) . These antibodies were next tested for their ability to modulate the activation of ALK by purified FAM150A . While mAb135 had no effect on FAM150A activation of ALK , we observed that mAb13 exhibited strong inhibition of FAM150A-induced ALK activation ( Figure 4E ) , while mAb48 robustly activated ALK in the absence of FAM150A ligand ( Figure 4E ) . Taken together , these data strongly support a role for the GR in the activation of ALK by FAM150 proteins . In cell lines as well as in the Drosophila model , we observed extremely high levels of ALK activation when compared even with highly active ALK mutants such as ALK-F1174S , suggesting the possibility that these already active mutants can be activated to a higher level in the presence of a potent ligand . Therefore , we examined the ability of either FAM150A or FAM150B to ‘superactivate’ already active ALK mutants as described in neuroblastoma . In PC12 cells we indeed observed activation of the ALK-R1275Q mutant to a higher level , as evidenced by increased pERK1/2 and in neurite outgrowth assays ( Figure 4F ) . A similar effect was observed with the ALK-F1174L mutant ( Figure 4—figure supplement 7 ) . This increased activity was also seen in CLB-GAR cells , which endogenously express the ALK-R1275Q mutant , with increased activation of ERK1/2 upon the addition of either FAM150A or FAM150B ( Figure 4G ) . In conclusion , our work identifies FAM150A and FAM150B as novel ligands that bind and activate the human ALK RTK . The identity of a ligand ( s ) for vertebrate ALK has been an evolving question for many years , with a number of reports over the past 15 years examining the role of the small heparin binding growth factors MDK and PTN as ligands for vertebrate ALK ( Stoica et al . , 2001; Stoica et al . , 2002 ) . However , a number of independent studies including our own work have presented convincing evidence that MDK and PTN do not activate ALK ( Motegi et al . , 2004; Moog-Lutz et al . , 2005; Mourali et al . , 2006; Mathivet et al . , 2007; Hugosson et al . , 2014; Murray et al . , 2015 ) . The recent identification of long chain heparins as activators of ALK in neuroblastoma cells ( Murray et al . , 2015 ) and the interplay of this with ALK activation by FAM150A and FAM150B will require further work . FAM150A and FAM150B are very basic proteins; FAM150A has a predicted pI of 10 . 6 , while FAM150B has one of 9 . 75 , and we show here that FAM150A exhibits heparin binding properties , however we have not observed a cooperative role of heparin and the FAM150 ligands in ALK activation as has been described for the fibroblast growth factor receptors and FGF . Our data also suggest that FAM150 ligands can bind an ALK-LTK heterodimer complex . While the physiological significance of such an interaction is currently unclear , ALK and LTK have been reported to be coexpressed in some tissues , such as the mouse hippocampus ( Weiss et al . , 2012 ) , where such an interaction may be important . FAM150 proteins do not seem to be conserved outside the vertebrates . In zebrafish there are three FAM150 proteins ( FAM150ba , FAM150bb and FAM150A ) , while in mouse and humans only FAM150A and FAM150B exist ( Zhang et al . , 2014 ) . In invertebrate model organisms the ALK ligands , Jeb in Drosophila and HEN-1 in C . elegans , do not appear to resemble the FAM150A/B ligands . The Drosophila Jeb ligand is unable to activate either mouse or human ALK ( Yang et al . , 2007 ) , and to date , no Jeb-like ligand for vertebrate ALKs has been reported . The evolution of the FAM150 proteins as ALK ligands is an interesting topic for further investigation . In the case of ALK mutations that are ligand dependent , as well as in situations where the ALK receptor is overexpressed , such as in neuroblastoma , the mechanisms by which ALK is activated is fundamental for understanding their role in tumor development . Here the identification of ALK ligands will allow a more rigorous interrogation of ALK signaling in these scenarios . The finding that FAM150A and FAM150B are not only able to activate the wild-type receptor , but also to ‘superactivate’ mutant ALK receptors , is important in this context . These findings , together with the observed expression pattern of FAM150A and FAM150B in adrenal gland and thyroid and the inhibition of ALK activation by FAM150A with monoclonal antibodies shown here , suggest that a future potential therapeutic arm may involve agents that interfere with FAM150A/B activation of ALK offering additional therapeutic approaches in patients with neuroblastoma for which single drug treatments targeting ALK have had limited effect to date .
The primary antibodies used were anti-pan-ERK ( 1:10 , 000; BD Transduction Laboratories ) , anti-ALK ( for immunofluorescence 1:1000; ab4061 , Abcam ) , anti-ALK ( D5F3 , 1:5000; Cell Signaling Technology ) , anti-ALK mAb135 ( 1:2000; [Witek et al . , 2015] ) , anti-pERK5 ( 1:1000; Cell Signaling Technology ) , anti-pALK-Y1278 ( 1:2000; Cell Signaling Technology ) , anti-pALK-Y1604 ( 1:2000; Cell Signaling Technology ) and anti-pERK1/2-T202/Y204 ( 1:2000; Cell Signaling Technology ) , anti-FAM150A ( 1:4000 , Atlas Antibodies ) , anti-HA ( 1:1000 for immunofluorescence , 1:6000 for immunoblotting; 16B12 , Covance ) . The activating monoclonal antibody mAb46 and ALK monoclonal antibodies mAb13 , mAb48 were a kind gift from M . Vigny and have been described previously ( Moog-Lutz et al . , 2005 ) . The ALK inhibitor crizotinib was purchased from Chem Express ( Shanghai ) . Human embryonic kidney ( HEK ) 293 cells were grown on collagen-coated cover slips in 24-well plates , prior to Lipofectamine transfection with either pcDNA3-ALK or pcDNA3 alone as control . Conditioned medium containing FAM150A-HA or FAM150B-HA was then incubated for 20 min prior to fixation and immunostaining . For immunostaining of HEK cells , cells were fixed with 4% paraformaldehyde/Dulbecco's Modified Eagle Medium ( DMEM ) and blocked with 50 mM NH4Cl/phosphate buffered saline ( PBS ) . After permeabilization with 0 . 3% Triton X-100 and 5% goat serum containing PBS , cells were incubated with primary antibody as indicated overnight at 4°C . For visualization cells were further incubated with fluorescence-labeled secondary antibody followed by analysis on LSM710 or LSM800 confocal microscopes ( Zeiss ) . PC12 cells cotransfected with wild-type ALK , together with either FAM150A or FAM150B , were incubated for 24 hr prior to starvation in serum-free DMEM for a further 24 hr . PC12 cells expressing human wild-type ALK were serum starved for 24 hr prior to stimulation with FAM150A or FAM150B conditioned medium for 30 min . IMR-32 cells were stimulated with either 0 . 85 µg/ml mAb46 as positive control ( Moog-Lutz et al . , 2005 ) , or with FAM150A or FAM150B conditioned medium for 20 min unless otherwise stated . For ALK inhibition , crizotinib was employed at a final concentration of 250 nM . Both PC12 and IMR-32 cells were lysed in 1 × sodium dodecyl sulfate ( SDS ) sample buffer , and precleared lysates were run on SDS-PAGE , followed by immunoblotting using the indicated antibodies . ALK downstream activation was detected by anti-pERK1/2 . Pan-ERK was used as loading control . ALK phosphorylation was checked by using pALK-Y1278 and pALK-Y1604 antibodies . Cell lysis and immunoblotting was performed according to protocols described previously ( Schonherr , Yang et al . , 2010 ) . To detect the interaction between ALK and FAM150A or FAM150B , HEK293 cells ( in 10 cm dishes , 80–90% confluent ) were transfected with 5 μg of pcDNA3-ALK-FLAG together with either 5 μg of pTT5-FAM150A-HA or pcDNA3-FAM150B-HA . As controls , cells were transfected with 5 μg of pcDNA3 vector together with 5 μg of pcDNA3-ALK-FLAG or pTT5-FAM150A-HA or pcDNA3-FAM150B-HA , correspondingly . Cells were lysed 16–20 hr after transfection , in 1 ml of lysis buffer ( 50 mM Tris-Cl , pH7 . 4 , 250 mM NaCl , 1% Triton X-100 and proteinase inhibitor cocktail ) on ice for 10 min prior to clarification by centrifugation at 14 , 000 rpm at 4º°C for 15 min . Supernatant ( 30 μl ) was taken and used as input control , and the remaining sample was incubated with either 20 μl of anti-FLAG M2 magnetic beads ( Sigma ) or 20 μl of anti-HA agarose beads ( ThermoScientific ) correspondingly for 2 hr at 4º°C . The beads were then washed five times with lysis buffer and boiled in 80 μl of 1 x SDS sample buffer . Both inputs and immunoprecipitation products were separated by 9% SDS-PAGE gel . ALK and FAM150A or ALK and FAM150B were detected in the same blot with anti-ALK mAb135 ( 1:2000 ) and anti-HA antibody ( 1:6000 , Covance ) , respectively . PC12 ( 2 x 106 ) cells were co-transfected with either 0 . 3 µg of empty pcDNA3 vector control , pcDNA3-ALK or pcDNA3-ALK-F1174L together with 0 . 5 µg pEGFPN1 ( Clontech ) and either 0 . 5 µg pTT5-FAM150A-HA or pcDNA3-FAM150B-HA as indicated by electroporation using Amaxa electroporator ( Amaxa Biosystems ) . Cells were resuspended in 100 µl Ingenio electroporation solution ( Mirus Bio LCC ) prior to transfection . After transfection , cells were kept in minimum essential medium ( MEM ) supplemented with 7% horse serum and 3% fetal bovine serum and seeded into 24-well plates . After 24 hr of incubation , the fraction of Green fluorescent protein ( GFP ) -positive and neurite carrying cells versus GFP-positive cells was observed under a Zeiss Axiovert 40 CFL microscope . PC12 cells incubated with FAM150A-conditioned or FAM150B-conditioned media were analyzed for neurite outgrowth 24 hr after addition of conditioned media . For ALK inhibition , crizotinib was employed at a final concentration of 250 nM . To be judged as a neurite carrying cell , the neurite of the cell should be at least twice the diameter of a normal cell body . Experiments were performed in triplicates , and each sample within an experiment was performed in duplicate . HEK293 cells , at ∼90% confluency in 10 cm dishes , were transfected with 8 . 0 µg of either pcDNA3 vector control , pTT5-FAM150A-HA or pcDNA3-FAM150B-HA with Lipofectamine 2000 ( Invitrogen ) . After 6 hr , medium was replaced with non-supplemented MEM and the subsequent conditioned medium was harvested 48 hr post transfection . For IMR-32 stimulation experiments 1 ml of either control , FAM150A-conditioned or FAM150B-conditioned medium was added to IMR-32 cells seeded in 6-well plates resulting in a total final volume of 3 ml . For PC12 stimulation experiments , medium was replaced with either control , FAM150A-conditioned or FAM150B-conditioned medium as indicated . DNA encoding FAM150A ( amino acids 21–129 ) was cloned into pFastBac-HBM TOPO ( Invitrogen ) and used to transform competent DH10Bac cells ( Invitrogen ) generating the recombinant FAM150A bacmid . Sf21 cells were transformed with the bacmid DNA using Cellfectin II reagent ( Invitrogen ) in serum-free Grace’s Medium ( Invitrogen ) . Supernatant from Sf21 cells containing recombinant FAM150A virus was collected after 72 hr and used for virus amplification . For protein expression , Sf21 cells were grown to a cell density of 1–1 . 5 × 106 cells/ml in Sf900-II medium ( Gibco ) containing 1% fetal bovine serum , penicillin and streptomycin , and infected with FAM150A recombinant viral stock . Medium was harvested 72 hr after infection and protein was affinity-purified on Ni-NTA agarose resin ( Qiagen ) . Wild-type ALK expressing neuroblastoma cell lines NB1 and IMR-32 were treated with either 2 μg/ml of FAM150A , 2 μg/ml of one of the following monoclonal antibodies: mAb13 , mAb48 , and mAb135 , or combinations of FAM150A and either of the three antibodies respectively for 20 min . In case of the combination of FAM150A and antibodies , antibodies were added to the cells 30 min prior to addition of FAM150A . The activation of ALK by FAM150A was visualized with pALK-Y1604 and pERK1/2 . Total ALK and tubulin were employed as internal controls . The Gal4-UAS expression system was used for the ectopic expression of human ALK in Drosophila eye ( Brand and Perrimon 1993 ) . cDNAs encoding either FAM150A or FAM150B were cloned into pUAST and verified by sequencing . Transgenic flies carrying pUAST-FAM150A or pUAST-FAM150B were generated by BestGene , Inc . , and crossed with pGMR-Gal4 , UAS-ALK-WT ( Martinsson et al . , 2011 ) , where hALK expression was driven by pGMR-Gal4 ( Bloomington Stock Center ) in the developing eye . UAS-ALK-F1174S ( Martinsson et al . , 2011 ) driven by pGMR-Gal4 was included as a positive control . Scale bar indicates 100 µm . Images were taken on a Zeiss AxioZoomV16 stereomicroscope . FAM150A was purified as described ( Zhang et al . , 2014 ) . For ELISA , 384-well white Maxisorp ( Nunc Nalge ) were coated overnight at 4º°C with 20 µl of 2 µg/ml FAM150A diluted in 100 mM sodium bicarbonate buffer , pH 9 . 6 . Subsequent steps were performed at room temperature . Plates were washed with PBST ( PBS , 0 . 05% Tween20 ) and blocked for 1 hr with blocking/dilution buffer ( PBST with 1% bovine serum albumin [BSA] ) . Plates were washed and 20 µl/well of ALK-Fc , LTK-Fc , and control-Fc diluted in blocking buffer added and incubated for 2 hr . Plates were washed and subsequently 20 µl/well of 1:10 , 000 dilution of horseradish peroxidase conjugated anti-human IgG ( Jackson Labs ) was added and incubated for 1 hr . After washing , 20 µl/well of ELISA Pico Chemiluminescent Substrate ( Thermo Fisher Pierce ) was added and incubated for 5 min before reading on a plate reader . Binding kinetics of FAM150A to LTK-ECD-Fc and ALK-ECD-Fc fusion proteins was determined using Biacore T100 SPR ( GE Healthcare Life Sciences , Piscataway , NJ ) . LTK-ECD-Fc and ALK-ECD-Fc were captured on a CM4 sensor chip immobilized with anti-human IgG antibody using the human antibody capture kit ( GE Healthcare Life Sciences , Piscataway , NJ ) . 4- ( 2-hydroxyethyl ) -1-piperazineethanesulfonic acid ( HEPES ) buffered saline10 mM , pH 7 . 4 with 0 . 05% Tween20 20 µg/ml heparin sodium salt ( HBS-P , GE Healthcare Life Sciences , Piscataway , NJ , Sigma Aldrich , St Louis , MO ) was used as the running and dilution buffer . Capture levels of the ECD-Fc fusions were adjusted to ∼300–500 resonance units ( RU ) . FAM150A was injected at eight concentrations ( 200 , 66 . 6 , 22 . 2 , 7 . 4 , 2 . 46 , 0 . 82 , 0 . 27 nM , and a 0 nM control ) for 120 s and dissociation was followed for 180 s . The association constant , dissociation constant , and affinity for FAM150A for LTK and ALK Fc-ECD fusions were calculated using the Biacore T100 evaluation software package using standard double referencing technique and the 1:1 binding model . | Cells have receptor proteins on their surface that enable them to detect changes in their environment and communicate with other cells . Signal molecules bind to a segment of the receptor called the extracellular domain that faces out from the cell . This can result in the activation of another domain in the receptor that is just inside the cell , which , in turn , activates signaling pathways that relay the information around the cell . However , these communication systems are often disrupted in cancer cells . This helps the cells to override the strict growth controls imposed upon them by other ( healthy ) cells in the body . The gene that encodes a receptor protein called Anaplastic Lymphoma Kinase ( or ALK for short ) is often mutated in some types of human cancer so that the protein is always active . However , we still do not know what signal molecules bind to the ALK protein to activate it in normal cells . Guan , Umapathy et al . used a variety of cell biology and biochemical techniques to study the role of ALK . The experiments show that when either of two proteins called FAM150A and FAM150B are produced in rat nerve cells alongside ALK , the nerve cells rapidly respond and form outgrowths . Experiments using cancer cells derived from human nerve cells also yielded similar results . Guan , Umapathy et al . found that the extracellular domain of ALK can physically interact with FAM150A and FAM150B . The eyes of fruit flies that had been genetically modified to produce the human ALK protein alongside either FAM150A or FAM150B grew more than normal , giving the eyes an abnormal "rough" appearance . Further experiments showed that FAM150A and FAM150B are also able to increase the level of activation of an ALK mutant protein that is already active . Therefore , in future , the development of drugs that stop FAM150A and FAM150B from binding to ALK may be useful for treating cancers that are driven by high levels of ALK activity . Many challenging questions lie ahead to better understand how FAM150A and FAM150B interact with ALK . | [
"Abstract",
"Introduction",
"Results",
"and",
"discussion",
"Materials",
"and",
"methods"
] | [
"short",
"report",
"cell",
"biology"
] | 2015 | FAM150A and FAM150B are activating ligands for anaplastic lymphoma kinase |
In mammalian cells three closely related cavin proteins cooperate with the scaffolding protein caveolin to form membrane invaginations known as caveolae . Here we have developed a novel single-molecule fluorescence approach to directly observe interactions and stoichiometries in protein complexes from cell extracts and from in vitro synthesized components . We show that up to 50 cavins associate on a caveola . However , rather than forming a single coat complex containing the three cavin family members , single-molecule analysis reveals an exquisite specificity of interactions between cavin1 , cavin2 and cavin3 . Changes in membrane tension can flatten the caveolae , causing the release of the cavin coat and its disassembly into separate cavin1-cavin2 and cavin1-cavin3 subcomplexes . Each of these subcomplexes contain 9 ± 2 cavin molecules and appear to be the building blocks of the caveolar coat . High resolution immunoelectron microscopy suggests a remarkable nanoscale organization of these separate subcomplexes , forming individual striations on the surface of caveolae .
Caveolae are an abundant feature of the plasma membrane of many vertebrate cells . The surface of adipocytes , endothelial cells , smooth muscle , skeletal muscle and many other cell types is characterized by a dense covering of these small invaginations with a characteristic striated coat , as viewed by electron microscopy , and by the presence of membrane proteins termed caveolins ( Peters et al . , 1985; Kurzchalia et al . , 1992; Rothberg et al . , 1992; Way and Parton , 1995; Scherer et al . , 1996; Parton and Del Pozo , 2013 ) . Three caveolins are present in mammalian cells with caveolin-1 ( CAV1 ) and caveolin-3 ( CAV3 ) essential for caveolar formation in nonmuscle and muscle cells respectively . Caveolins bind cholesterol and fatty acids and form homo-oligomers required for caveolar formation . Approximately 140 CAV1 molecules associate with a single caveola in mammalian cells ( Pelkmans and Zerial , 2005 ) and in a model prokaryotic system upon caveolin expression ( Walser et al . , 2012 ) . Genetic ablation of caveolins in mice has diverse cellular consequences with impact upon numerous signal transduction pathways and lipid dysregulation . Human patients lacking CAV1 show a severe lipodystrophy while CAV3 mutations , many of which disrupt caveola formation in muscle , are associated with a number of muscle diseases . Recent years have seen a dramatic increase in our understanding of caveolae with the characterization of a family of caveolar coat proteins ( Hill et al . , 2008; Bastiani et al . , 2009; Hansen et al . , 2009; McMahon et al . , 2009; Bastiani and Parton , 2010; Hansen and Nichols , 2010 ) . Cavin1/polymerase transcript release factor ( PTRF ) , cavin2/SDPR ( serum deprivation response protein ) , cavin3/SRBC ( serum deprivation response factor-related gene product that binds to C-kinase ) and cavin4/MURC ( muscle-restricted coiled-coil protein ) are cytoplasmic proteins characterized by conserved putative N-terminal coiled coil domains ( Hansen and Nichols , 2010 ) . Cavin1 was originally identified as a nuclear protein that can dissociate paused ternary transcription complexes ( Jansa et al . , 1998 ) , while cavin2 ( Gustincich and Schneider , 1993 ) and cavin3 ( Izumi et al . , 1997 ) were identified as protein kinase C ( PKC ) substrates and have been suggested to function in the targeting of PKC to caveolae . Cavin4 is predominantly expressed in cardiac and skeletal muscle ( Ogata et al . , 2008; Tagawa et al . , 2008 ) . The cavin proteins have been shown to co-associate and form a cytosolic complex ( es ) in cells lacking CAV1 ( or CAV3 ) , but are recruited to the cell surface to stabilize caveolae in cells expressing CAV1 . Ablation of cavin1 expression causes loss of caveolae with CAV1 being released into the bulk membrane whereas expression of cavin1 or cavin2 in cells lacking cavins but expressing endogenous CAV1 is sufficient to generate caveolae ( Hill et al . , 2008; Hansen et al . , 2009 ) . Like caveolins , cavins have also now been linked to many disease conditions including cardiomyopathies , lipodystrophy , and skeletal muscle disorders ( Parton and Del Pozo , 2013 ) . Caveolae possess a unique cytoplasmically facing striated coat enriched in CAV1 ( Rothberg et al . , 1992 ) . Given the recent identification of the cavin cytoplasmic coat , essential for the formation and regulation of caveolae , one could hypothesize that CAV1 would associate with the cavin complex and aid in generating the striated coat . Furthermore , this cavin coat may provide a possible mechanism for spatial and temporal regulation of caveola formation: caveolae form at the plasma membrane as caveolins and cavins associate , rather than earlier in the exocytic pathway ( Hill et al . , 2008; Hayer et al . , 2010 ) . In addition , the dissociation of the cavin coat complex could potentially provide a mechanism to disassemble caveolae . This may be crucial for caveolar function in setting membrane tension and in mechanosensing as increased membrane tension causes caveolar flattening and dissociation of cavin1 ( Sinha et al . , 2011 ) . However , the mechanisms underlying the formation of the cavin complex ( es ) , their stoichiometry and association with caveolae are all unknown . Here we have developed new methods that allow reconstitution of the cavin complex and performed a quantitative assessment of cavin complex formation . We use single-molecule fluorescence for its proven ability to directly observe multiple populations and quantify interactions in complex mixtures . These techniques are especially well suited for the study of coat proteins and have been used to study the mechanisms of clathrin assembly/disassembly ( Böcking et al . , 2011 ) . Those studies required labeling of recombinantly expressed purified proteins , which in the case of the cavin complex is difficult ( Hansen et al . , 2009 ) . We have taken an alternative approach to obtain the stoichiometry of the cavin complex and the interactions between the members of the cavin family ( cavin1 , cavin2 , cavin3 ) directly from cell extracts . We could observe fascinating behaviour of the cavin members , with exquisite segregation of interactors and defined stoichiometries in mixed oligomers . By combining these experiments with novel electron microscopy techniques we show the surprising existence of two distinct cavin complexes , cavin1-2 and cavin1-3 . Remarkably , the two complexes co-interact with individual caveolae but associate within distinct striated nanodomains .
Formation of higher order structures through oligomerization is a common behaviour of proteins involved in control of membrane dynamics . Here we set out to determine what role oligomerization of cavins plays in the biogenesis of caveolae . To this end we sought a technique that would allow rapid and quantitative analysis of homo- and hetero- interactions of proteins in vivo and in vitro . We chose single-molecule fluorescence spectroscopy as a way of directly assessing interactions between cavin proteins in complex mixtures . Fluorescence spectroscopy at the single-molecule level typically requires labeling with specific organic dyes , sufficiently bright for detection of individual molecules . The brightness of genetically encoded fluorophores such as the widely used eGFP and mCherry is considered too low to allow their individual detection , especially when proteins are freely diffusing and not immobilized on a surface . As a result , while the actual single-molecule experiment consumes typically only a few thousands of molecules , a typical sample preparation requires preparative protein expression , purification and labeling . This limits the throughput of the technology and biases its application to proteins that can be obtained in an active form through recombinant expression in Escherichia coli . This is typically not possible with cavin proteins , which show a tendency to degrade upon expression in a bacterial host ( Hansen et al . , 2009 ) . In order to resolve this bottleneck , we performed single-molecule analysis of proteins tagged with fluorescent domains such as eGFP and mCherry expressed in mammalian cells . To test this experimentally we expressed GFP- and Cherry-tagged cavin1 proteins in MCF-7 cells . As MCF-7 cells are cavin and caveola-deficient the expressed cavin1 should be mostly in the cytosol . The transfected cells were mechanically lysed , centrifuged to remove large membranous material and the supernatant was analyzed using a confocal microscope configured for single-molecule detection . The principle of the technique and the results obtained are schematized in Figure 1 . In brief , the brownian motion of freely diffusing proteins brings them in and out of the confocal detection volume created by two lasers that simultaneously excite the GFP and the Cherry fluorophores . When fluorescent proteins transit through the focal volume the emission of GFP and Cherry fluorophores is detected simultaneously on two separate single-photon counting detectors . The measurement generates a highly temporally resolved ( 100 ns ) fluorescence time trace , binned in 100 ms time intervals to obtain sufficient signal/noise ratio . As shown in Figure 1C the signal obtained for co-expressed cavin1-GFP and cavin1-Cherry shows coincident bursts of fluorescence with large amplitudes . The presence of the two fluorophores at the same time in the focal volume demonstrates that the cavin1 proteins self-interact . However , these data alone are not sufficient to distinguish between a dimer and higher order structures . In order to extract this information , each burst of fluorescence can be analysed for intensity of GFP and Cherry signals and duration of the burst . The former analysis is based on the assumption that multiple fluorescent proteins are brighter than individual GFP/Cherry . The detailed analysis of the amplitude of each burst should allow us to calculate the number of proteins in a complex . In the burst duration analysis we expect time-trajectories of proteins to be different for small and larger oligomers , as large protein complexes will diffuse slower and spend more time in the focal volume . This is quantified burst-by-burst by correlating the fluorescent signal intensity as done in Fluorescent Correlation Spectroscopy ( FCS ) ( ‘Materials and methods’ ) and allows extraction of residence times in the focal volume for proteins or protein oligomers . 10 . 7554/eLife . 01434 . 003Figure 1 . Principles of single-molecule detection and analysis of cavin1 . ( A ) Sample preparation for single-molecule analysis of fluorescently tagged proteins expressed in mammalian cells . Cells transfected with GFP and Cherry fusions of the cavin1 are mechanically lysed and centrifuged to remove cellular debris and the supernatant is placed in a chamber for single-molecule detection . ( B ) Schematic representation of single-molecule fluorescence experiment in which the proteins freely diffuse in and out of the focal volume created by two lasers simultaneously exciting the GFP and Cherry fluorophores . ( C ) An example of a single-molecule trace obtained for lysates of the cells co-expressing cavin1-GFP and cavin1-Cherry . The numbers of photons detected in green and red channels are plotted as a function of time . The trace shows simultaneous bursts in both GFP and Cherry channels that reflect formation of complexes containing both fluorophores . ( D ) Detailed analysis of a fluorescence burst from ( C ) . Each fluorescent burst is analysed for three parameters: ( 1 ) the coincidence between the GFP and Cherry fluorescence that reflects co-diffusion of at least two proteins with different tags; ( 2 ) the total brightness of the burst , indicating the number of proteins present in the oligomer; ( 3 ) the burst profile that is determined by the rate of diffusion and reflects the apparent size of the complex . ( E ) Histogram of the number of photons measured per burst for GFP expressed in MCF-7 cells . The distribution of bursts is consistent with the behaviour of a monomeric GFP . The data was used to calibrate the brightness profile and estimate the number of cavins-GFP molecules . ( F ) GFP residence time in the confocal volume plotted against the number of observations; ( G ) a histogram of data shown in ( F ) that reveals a very tight distribution of residence times in the focal volume ( or diffusion times ) around 110 μs . This value is used to convert the observed residence time into apparent size in the subsequent experiments . ( H ) A plot of burst size distribution of cavin1-GFP expressed in MCF-7 cells . The high quantum yield of the observed particles suggests that cavin1 forms oligomers approximately 50 times brighter than a single GFP molecule . ( I ) as in ( F ) but for cavin1-GFP . The duration of the burst for cavin1-GFP is also much higher than that of GFP indicating that the apparent size of the diffusing objects is at least 10-fold larger ( J ) as in ( G ) but X-axes represents size as calculated using residence time of GFP as a reference . The distribution of cavin1 sizes is broad and is centered on 60 nm . DOI: http://dx . doi . org/10 . 7554/eLife . 01434 . 003 In order to calibrate the brightness and duration of GFP/Cherry fluorescent bursts , we analysed the time trace obtained for GFP ( Figure 1E ) . Here we observed much smaller fluorescent bursts than for cavin1 . The diffusion of the molecule in the focal volume is random , and the optimal trajectory maximizing the numbers of photons emitted is exponentially rare . The burst profile decreases rapidly and GFP molecules yield a maximum of 100 photons ( see ‘Materials and methods’ for the explanation of the burst profile plots ) . We performed the same brightness analysis for Cherry monomers , and adjusted the excitation laser intensity to obtain the same burst profile as for GFP and a maximum of 100 photons per Cherry monomers . This ensures accurate ‘counting’ in both GFP and Cherry channels , correcting for the differences in extinction coefficients , quantum yields and detection efficiency of the two fluorophores . Using these brightness values as references we estimated that cavin1 forms complexes that contain 50 ± 5 proteins . We next analyse the apparent size of the cavin1 oligomer . As in the earlier case , the measurement is calibrated by GFP , monomeric at the concentration used ( 100 pM ) , which displays a well-defined residence time in the focal volume of 100 μs . The residence times measured for cavin1 oligomers are approximately 10 times slower than GFP . The structure of GFP is well known and the near-spherical β-barrel has a typical diameter of 5 nm . If we assume that the cavin1 oligomer forms an isotropic complex and if we simplify its shape to a sphere , we can estimate the diameter of the diffusing object to approximately 60 nm ( Figure 1J ) . These data demonstrate that cavin1 spontaneously assembles into a higher order structure of defined size even in the absence of the scaffolding membrane protein CAV1 . Although cavin1 and CAV1 appear to be sufficient to form caveolae , cells generally express 3 cavin family members that all associate with caveolae ( Bastiani et al . , 2009; Bastiani and Parton , 2010; Hansen and Nichols , 2010 ) . After the observation that cavin1 can form a large oligomeric structure , we next analyzed the ability of the other cavins to form homo- and hetero-oligomers . We performed dual colour co-transfection of MCF-7 cells and single-molecule coincidence analysis of all three cavins . As shown in Figure 2A , B , in the cytosol of MCF-7 cells cavin1 forms homo-complexes but also stable heteromeric complexes when expressed with cavin2 or cavin3 . Strikingly , we could not detect any mixed oligomers when cavin2 and cavin3 were co-expressed . Figure 2C demonstrates that co-expressions lead to two segregated populations of cavin2-cavin2 and cavin3-cavin3 oligomers . 10 . 7554/eLife . 01434 . 004Figure 2 . Single-molecule coincidence and ALPHAScreen mapping of interactions among cavin1 , 2 and 3 . ( A ) Histogram of single-molecule coincidence between cavin1-GFP and cavin2-Cherry co-transfected in MCF-7 cells . The coincidence is calculated as the ratio of intensity in the Cherry channel divided by the sum of the signals in the GFP and Cherry channels . GFP-only bursts show a coincidence at 0 and Cherry-only oligomers are located at coincidence = 1 . For the oligomers containing both fluorophores , the coincidence ratio is a measure of the stoichiometry of the assembly . ( B ) Same as ( A ) , coincidence between cavin1-GFP and cavin3-Cherry . ( C ) Same as ( A ) , coincidence between cavin2-GFP and cavin3-Cherry . ( D–F ) : using cell-free protein expression , three-dimensional histograms of single-molecule coincidence between GFP and Cherry cavin proteins at expression ratios spaning from 100% GFP to 100% Cherry . For each DNA ratio of cavin-GFP and cavin-Cherry , we collected >1000 bursts and plotted the corresponding histograms of coincidence . We varied the ratio of cavins and created a stack of 10 histograms representing the various stoichiometries . The histograms for individual pairs were then aggregated into 3D plots . ( D ) Evolution of mixed oligomers of cavin1-GFP and cavin2-Cherry revealing formation of oligomers with a full range of stoichiometries . ( E ) The cavin1 and cavin3 plot shows formation of cavin1-cavin3 oligomers with predominantly 3/1 composition . ( F ) The cavin2 and cavin3 plot shows that these proteins do not form mixed oligomers . ( G ) Schematic representation of ALPHAScreen principle . This bead–bead assay relies on transfer of singlet oxygen from a donor bead to a luminescent acceptor bead when protein–protein interactions bring the beads within 200 nm ( see ‘Materials and methods’ and SI for details ) . ( H ) A plot of ALPHAScreen signal across the concentrations of cavin proteins attached to donor and acceptor beads . The interactions for cavin2 and cavin3 in the presence or absence of cavin1 display amplitudes close to the background signal . ( I ) Values obtained for cavin1-GFP and cavin2-myc , cavin1-GFP and cavin3-myc , cavin2-GFP and cavin2-myc reveal robust interactions . However the curves obtained for cavin3-GFP and cavin2-myc , cavin2-GFP and cavin3-myc demonstrate that cavin2 and cavin3 cannot bind to each other . The triple co-expression of cavin2-GFP , cavin3-myc and untagged cavin1 results in no change in binding , suggesting that cavin1 cannot act as a bridge between cavin2 and cavin3 . DOI: http://dx . doi . org/10 . 7554/eLife . 01434 . 00410 . 7554/eLife . 01434 . 005Figure 2—figure supplement 1 . Single-molecule fluorescence trace of cavin1-GFP during expression in the cell-free system . The fluorescence trace shows the real-time oligomerization of cavin1 , occurring within 15 min . While all the GFP fluorophores are not yet folded and fluorescent , we can already detect bursts of large amplitude suggesting the formation of oligomers at very low concentration . We estimate that at 15 min , the concentration in the cell-free expression system has not reached 10 nM . We stopped translation after 15 min by diluting the sample and performed single-molecule analysis; the residence times and brightness obtained correspond to a population of oligomerized cavin1 . DOI: http://dx . doi . org/10 . 7554/eLife . 01434 . 00510 . 7554/eLife . 01434 . 006Figure 2—figure supplement 2 . Comparison of cavin1 oligomers observed in MCF-7 cells and expressed in the cell-free system . ( A ) Plot of the residence times as a function of number of experiments for cavin1-GFP in MCF-7 cells . ( B ) Histogram of sizes measured by single-molecule diffusion and calibrated by the diffusion of GFP , as shown in Figure 1 of the main text . ( C ) Plot of the residence times as a function of number of experiments for cavin1-GFP expressed in the cell-free system . ( D ) Histogram of apparent sizes of cavin1-GFP expressed in cell-free system . The distributions of residence times and apparent sizes obtained from cell extract and in the cell-free expression system match closely , suggesting that cavin1 has an intrinsic propensity to form oligomers . DOI: http://dx . doi . org/10 . 7554/eLife . 01434 . 00610 . 7554/eLife . 01434 . 007Figure 2—figure supplement 3 . Principle of the ALPHA screen . ( top panel ) The donor bead is coated with protein A . A laser pulse at 680 nm triggers the release of singlet oxygen from the donor bead . The singlet oxygen has a half-life of 4 μs and can diffuse over 200 nm . If an interaction occurs between protein A and protein B , the acceptor bead is brought into proximity of the donor bead . The singlet oxygen will react with thioxene derivatives encapsulated in the acceptor bead , resulting in luminescence , emitted between 520 and 620 nm . ( bottom panel ) The ‘hook’ effect is a signature of the Screen signal . This effect appears due to loading effects on the beads . If the concentration of interacting proteins is too low , the beads do not bind strongly and the luminescence signal is low . At optimal protein concentration , bead–bead contacts are maximized as all the beads are covered with proteins . When the system is overloaded with proteins , unbound proteins are competing with bead-immobilized proteins , diminishing the number of bead–bead interactions and lowering the AlphaScreen signal . DOI: http://dx . doi . org/10 . 7554/eLife . 01434 . 00710 . 7554/eLife . 01434 . 008Figure 2—figure supplement 4 . Pull-down analysis of the cavin complex formation in MCF-7 cells . MCF-7 cells were transfected with equal amounts of GFP and Cherry alone , GFP alone + Cavin 1-Cherry , GFP + Cavin 2-Cherry , GFP + Cavin 3-Cherry , Cavin 1-GFP and Cavin 2-Cherry , Cavin 1-GFP and Cavin 3-Cherry , Cavin 2-GFP and Cavin 3-Cherry and Cavin 1-Flag with Cavin 3-GFP and Cavin 2-Cherry . Post-nuclear supernatant ( soluble cytoplasmic fraction ) were prepared by extensive washing of cells in PBS . Cells were then scraped into PBS containing protease and phosphatase inhibitors followed by mechanic disruption by syringe lysis . Cells were then pelleted at 2000 rpm , 10 min at 4°C . 120th of the supernatant was retained as the starting material ( B ) . The remaining supernatant was mixed with 20 μl of prewashed GFP Trap beads for 30 min at 4°C on a rotating wheel . The beads were pelleted and washed three times in PBS supplemented with protease and phosphatase inhibitors and were boiled in 4X Sample buffer for 75°C for 2 . 5 min to preserve the fluorescence . Sample were separated by SDS-PAGE and the fluorescence corresponding to each of the overexpressed GFP or Cherry tagged Cavin construct was detected and quantification in the gel using the BioRad ChemiDoc MP Imaging System . DOI: http://dx . doi . org/10 . 7554/eLife . 01434 . 008 In order to analyse further the stoichiometries of association between cavin members , we needed to tune the co-expression levels of the different cavin proteins in a more controlled manner . We chose to produce them in vitro , using cell-free expression as it enables the precise control over the ratios of multiple co-expressed proteins via titration of DNA template concentrations . We used a recently developed Leishmania tarentolae–based cell-free system ( LTE ) as it is derived from a eukaryotic organism and is capable of producing complex eukaryotic proteins in functional form ( Mureev et al . , 2009; Kovtun et al . , 2011 ) . We first compared the properties of cavin1-oligomers produced in LTE cell-free system to the ones previously observed in MCF-7 cells . The open format of cell-free protein expression enables real time observation of protein production , folding and interactions . To obtain the time resolved picture of cavin complex assembly , LTE was loaded directly into the single-molecule observation chamber mounted on the confocal microscope and primed with DNA template coding for cavin1-GFP ( Figure 2—figure supplement 1 ) . After 2 hr of expression , the sizes and brightness of cavin1-GFP assemblies were very similar to the ones obtained for MCF-7 cells ( Figure 2—figure supplement 2 ) , indicating that we could use the cell-free system to reproduce the behaviour of cavin oligomerization . The time-trace of cavin production contains additional information . Surprisingly , we found that most of the cavin1 protein assembled into oligomers within the first 15 min while the concentration was still below 10 nM ( Figure 2—figure supplement 1 ) . This indicates that cavin oligomer assembly is a very rapid process that operates spontaneously at nanomolar concentrations . In the next step we used the in vitro expression system to analyse the interactions of cavin1 with cavin2 and cavin3 . We observed that the composition of oligomers formed by co-expression of cavin1 and 2 varied with levels of expression resulting in continuous transition from cavin-1 to cavin-2 homo-oligomer ( Figure 2D ) . On the contrary , we found that only a maximum of ca . 30% of cavin3 could be incorporated into cavin1 oligomers ( Figure 2E ) . Similarly to MCF-7 cells , we found complete segregation between cavin2 and cavin3 oligomers regardless of co-expression ratio ( Figure 2F ) . Available biological data strongly indicate that in vivo all three cavin members are present on a single caveola ( Bastiani and Parton , 2010 ) . The observed segregation between cavin2 and cavin3 was so unexpected that we decided to use an alternative biophysical method to analyse interactions between cavin2 and 3 . In order to detect even weak but potentially relevant biological interactions we employed ALPHAScreen ( Amplified Luminescent Proximity Homogeneous Assay Screen ) . This sensitive bead-based proximity assay ( Figure 2G ) is able to detect interactions in a wide range of affinities ( from pM to mM ) ( Eglen et al . , 2008 ) . When used for protein–protein interaction analysis the approach typically involves purified proteins ( Waller et al . , 2010; Mackie and Roman , 2011; Demeulemeester et al . , 2012 ) . In our case , we analysed interactions of protein pairs directly after their co-expression in the cell-free system and utilized GFP and myc tags for capture of the proteins on the reporter beads ( see ‘Materials and methods’ , Figure 2—figure supplement 3; Sierecki et al . , 2013 ) . As shown in Figure 2H , co-expression of cavin1 and 2 or cavin1 and 3 resulted in a very strong positive signal . However , no interaction between cavin2 and 3 was detected using this approach . This was confirmed by classical biochemistry and pull-downs from MCF-7 cells as shown in Figure 2—figure supplement 4 . Taken together these results indicate that cavin2 and 3 do not interact with each other , either in vitro or in vivo . This is a surprising finding that could potentially indicate that cavin1 operates as a bridging factor between cavin2 and 3 . To test this idea we repeated the ALPHAscreen assay where we measured the binding of cavin2 and 3 under conditions of cavin1 co-expression . Figure 2I shows that the presence of cavin1 did not enhance the interaction of cavin2 and 3 . To further corroborate this finding we performed single-molecule coincidence analysis where co-expression titration of unlabeled cavin1 to a fixed ratio of cavin2-GFP/cavin3-Cherry was monitored . The total fluorescence of GFP and Cherry cavin2/3 decreased as more cavin1 was co-expressed reflecting competition of the templates for the translational machinery . However no mixed oligomers containing both cavin2 and cavin3 were produced ( data not shown ) . To verify this result in mammalian MCF-7 cells , we next expressed cavin2-Cherry together with cavin3-GFP in the presence of non-fluorescent cavin1-FLAG . Supporting our in vitro findings we observed that the vast majority of the cavin2 and cavin3 proteins do not interact even when cavin1 is co-expressed ( Figure 3A ) . These findings suggest the existence of previously undetected segregation in cavin oligomers . 10 . 7554/eLife . 01434 . 009Figure 3 . Single-molecule counting and coincidence analysis of cavin oligomerization in MCF-7 cells in the presence of CAV1 . ( A ) Histogram of single-molecule coincidence between cavin2-GFP and cavin3-Cherry in MCF-7 cells co-expressing cavin1 . The lack of coincidence ( 0 . 25<coincidence<0 . 75 ) demonstrates that cavin2 and cavin3 do not interact even in the presence of cavin1 . ( B ) Histogram of single-molecule coincidence between CAV1-GFP and Cavin1-Cherry . The distribution shows a large peak of coincidence centered around C = 0 . 25 , indicating that cavin1 localizes on caveolae with approx . three CAV1 molecules for one molecule of cavin1 . ( C ) Distribution of burst brightness from MCF-7 cells transfected with CAV1-GFP ( shown in red ) or cavin1-GFP and CAV1 untagged ( in blue ) . The brightness of CAV1-GFP is typically 3 . 5 higher than the brightness of cavin1-GFP . ( D ) Histogram of single-molecule coincidence between cavin2-GFP and cavin3-Cherry in MCF-7 cells when both cavin1 and CAV1 are co-expressed . ( E ) Schematic representation of the co-localization of cavin2 and cavin3 when cavin1 and CAV1 are co-expressed , indicating that the two subcomplexes are assembled on the caveolae . ( F ) Histogram of apparent single-molecule sizes of cavin1-GFP co-expressed with CAV1 ( untagged ) . The size distribution is centered around the size of caveolae , suggesting that the cavin1 complex wraps around caveolae without significantly increasing its apparent size . DOI: http://dx . doi . org/10 . 7554/eLife . 01434 . 00910 . 7554/eLife . 01434 . 010Figure 3—figure supplement 1 . Effect of co-transfection of CAV1 in MCF-7 cells on the size and brightness of cavin1-GFP oligomers . ( A ) Plot of residence times obtained as a function of experiments for cavin1-GFP in MCF-7 cells that do not express CAV1 . ( B ) Plot of residence times observed for acvin1-GFP when the MCF-7 cells were co-transfected with untagged CAV1 . The residence times observed are centered on the same value but the values are more dispersed in the absence of CAV1 . This suggest that the scaffolding protein CAV1 limits the growth and structure of cavin1-GFP oligomers . ( C ) Distribution of burst sizes observed for cavin1-GFP in MCF-7 cells , with and without co-transfection of CAV1-untagged . The number of cavin1 involved in the cavin oligomer does not seem to change with the presence of CAV1 and caveolae , suggesting that the cavin1 oligomer intrinsically forms in a shape that resembles its caveolae-bound state . DOI: http://dx . doi . org/10 . 7554/eLife . 01434 . 01010 . 7554/eLife . 01434 . 011Figure 3—figure supplement 2 . Single-molecule coincidence analysis of cavin interactions in HeLa cells . Histogram of single-molecule coincidence between cavin2-GFP and cavin3-Cherry in HeLa cells; HeLa cells express endogenous levels of cavin1 and CAV1 . DOI: http://dx . doi . org/10 . 7554/eLife . 01434 . 01110 . 7554/eLife . 01434 . 012Figure 3—figure supplement 3 . Localization of Cavin 2-Cherry and Cavin 3-GFP in MDCK cells . MDCK cells were transiently transfected with Cavin2-Cherry and Cavin3-GFP . Cells were fixed in 4% PFA in PBS and were processed for immunofluorescence to localize Cavin2 and Cavin3 . A representative image of the localization of Cavin 2 and Cavin 3 in cavin and caveolin-1 expressing MDCK cells is shown . DOI: http://dx . doi . org/10 . 7554/eLife . 01434 . 012 The experiments described above revealed the intrinsic propensity to assemble into defined structures even in the absence of the scaffolding membrane . We next set out to characterize a more complete cellular system that included CAV1 protein to target cavins to the membrane . Hence , we expressed Cherry-tagged cavin1 in MCF-7 cells expressing CAV1-GFP . Microscopic analysis revealed that cavin1 was now located at the plasma membrane but a fraction of CAV1-labeled structures decorated with cavin1 proteins were found in the cytoplasm , presumably reflecting the dynamic nature of caveolae in cultured cells ( Pelkmans and Zerial , 2005; Boucrot et al . , 2011 ) . This allowed us to extend our single-molecule analysis of fluorescent caveolar proteins to analysis of caveolae in the post nuclear supernatant of transfected cells . We first characterized the behaviour of CAV1-GFP present in the cytoplasm . We observed intensely fluorescent puncta in the supernatant of CAV1-GFP transfected cells . As shown in Figure 3C , the intensity of these puncta reaches values that are 150 to 200-fold higher than the brightness of a single GFP . As caveolae contain up to 180 CAV1 proteins , this corresponds well to the expected brightness of fully-assembled caveola ( Pelkmans and Zerial , 2005 ) . As shown in Figure 3B , we performed dual-colour coincidence experiment between cavin1-Cherry and CAV1-GFP . The presence of a large coincident peak shows that most of the cavin1-Cherry co-diffuses with CAV1-GFP . As cavin1 does not associate with CAV1 in non-caveolar cellular compartments , such as when CAV1 is moving along the secretory pathway or if caveolae are disassembled ( Hill et al . , 2008; Hayer et al . , 2010 ) we are confident that these represent bona fide caveolae . The coincidence peak is centred between C = 0 . 25 and C = 0 . 3 , indicating that each caveola contains three to four times more CAV1 than cavin1 . This observation corroborates the brightness analysis that demonstrates that the CAV1 signal is 3 . 5 times brighter than that of cavin1 ( Figure 3B ) . Both methods suggest that the number of cavin1 per caveolae is estimated , again , between 40 and 50 . The number of cavin1 subunits per oligomer did not differ significantly between CAV1 positive and negative cells . However , while the apparent number of subunits per cavin1 assembly has not significantly changed , the presence of CAV1 had a pronounced effect on the distribution of particle sizes ( Figure 1J; Figure 3F , Figure 3—figure supplement 1 ) , narrowing the distribution to around 60 nm , which corresponds to the typical size of a caveola . To understand how the observed segregation of cavin subcomplexes translates into formation of native caveolae we repeated our cellular experiments with the full complement of caveolae-forming molecules . As in the experiments described above , we co-expressed cavin2-Cherry and cavin3-GFP together with CAV1 and cavin1-FLAG in MCF-7 cells . We observed coincidental diffusion of cavin2 and cavin3 indicating that they were part of the same assembly ( Figure 3D ) . The brightness and physical size of the cavin2-cavin3 complex correspond to a mixed coat of cavins formed on a single caveola . These data suggest that cavin1 , cavin2 and cavin3 are found together at the surface of the same caveolae , as depicted in Figure 3E . Identical results were obtained when cavin2-Cherry and cavin3-GFP were transfected into HeLa cells that express endogenous CAV1 and cavin1 ( Figure 3—figure supplement 2 ) . The co-binding of cavin2 and cavin3 on the same caveolae fits well with their colocalisation observed in MDCK cells , which also express endogenous CAV1 and cavin1 ( Figure 3—figure supplement 3 ) . The results described above suggest that the cavin1 proteins have an intrinsic propensity to assemble into a large complex in the absence of the CAV1/caveolae scaffold . A single complex made of only 50 cavin1 proteins would be sufficient to wrap around a caveola , possibly as a loose mesh . However it is possible that the cavin1 complex produced in the absence of their CAV1 membrane target represents an artificial dead-end product . We therefore took advantage of hypo-osmotic treatment to dissociate the caveolae-associated cavin1 complex from the plasma membrane and to study the properties of the released native cavin complex . We co-transfected MDCK cells with cavin1-GFP and cavin1-Cherry and performed cell swelling experiments by subjecting cells to hypo-osmotic treatment for 20 min . Cytoplasmic fractions were prepared from cells before and after the treatment and were subjected to single-molecule analysis . The dual-colour experiment showed that the cavin1 oligomer recovered in the cytosolic fraction diffuses much faster than the caveolae bound protein: the apparent size of the cavin1 complex is reduced to ‹ 30 nm ( calculated as diameter of a diffusing spherical object ) ( Figure 4A ) . Yet the protein remained oligomeric , as all fluorescent events register both fluorophores ( Figure 4B ) . The distribution of burst brightness showed that the observed oligomers are composed of 9 ± 2 cavin1 proteins ( Figure 4C ) . This suggests that the cavin coat is actually made of multiple stable subcomplexes that can be released from the caveolae without being disassembled into monomers . Because the cavins have such a high propensity to oligomerise , it is possible that the subcomplexes would form in the cytosol before reaching the CAV1-rich membrane , and that the subcomplexes can reversibly assemble on a caveola . 10 . 7554/eLife . 01434 . 013Figure 4 . Dissociation of the cavin coat upon cell swelling induced by hypo-osmotic treatment . ( A ) Histogram of apparent single-molecule sizes for cavin1-GFP before ( green ) and after ( grey ) membrane stretching induced by hypo-osmotic treatment . The size of observed cavin1 oligomers decreases from 60 nm to 20 nm . ( B ) Histogram of single-molecule coincidence for cavin1-GFP and cavin1-Cherry after hypo-osmotic treatment . We observe that solubilized cavin1 remains oligomeric as all bursts contain both GFP and Cherry fluorophores . ( C ) Burst brightness distribution for cavin1-GFP ( green ) and cavin1-Cherry ( red ) in oligomers released from the membrane upon hypo-osmotic treatment . Oligomers contain GFP and Cherry in equal amounts; the total fluorescence indicates that the sub-oligomers are typically made of 8–10 cavin1 proteins and hence significantly reduced in size compared to the isotonic conditions ( Figure 1 ) . ( D ) Histogram of single-molecule coincidence between cavin2 and cavin1 after hypotonic treatment . Data show that all cavin2 are bound to cavin1 but approximately 15% of the oligomers contain cavin1 only . ( E ) Histogram of single-molecule coincidence between cavin3 and cavin1 after membrane stretching . Data show that all cavin3 are bound to cavin1 and approx . 20% of cavin1-only oligomers are observed . ( F ) Histogram of single-molecule coincidence between cavin2 and cavin3 after membrane stretching . The absence of coincidence suggests that while cavin2 and cavin3 can localize to the same caveolae , their release from the membrane during stretch causes them to separate into two different subcomplexes . ( G ) Histogram of apparent single-molecule sizes obtained after hypo-osmotic treatment for cavin1-GFP , cavin2-GFP and cavin3-GFP , in the presence of co-expressed cavin1-Cherry . The measurements indicate that the cavin1-cavin1 and cavin1-cavin3 subcomplexes are very similar in size ( average of 20 nm ) , but the apparent size of the cavin1-cavin2 subcomplex is higher with an average of 30 nm . ( H ) Model of cavin1-cavin1 , cavin1-cavin2 and cavin1-cavin3 subcomplexes dissociation from CAV1 domains upon membrane stretching mediated by cell swelling . DOI: http://dx . doi . org/10 . 7554/eLife . 01434 . 01310 . 7554/eLife . 01434 . 014Figure 4—figure supplement 1 . Pull-down analysis of the effect of membrane stretching on the association between cavin members . Experiments are performed in MDCK cells , which have endogenous levels of cavin1 and CAV1 . MDCK cells were transfected with equal amounts of Cavin 1-GFP ( C1-GFP ) and Cavin2-Cherry ( C2-C ) , C1-GFP and Cavin 3-Cherry ( C3-C ) , C2-GFP and C3-C , Cavin 3-GFP ( C3-GFP ) and C2-C and Cavin1-Flag with C3-GFP and C2-C . Cells were then treated with 100% DMEM ( Iso-osmotic; Iso ) or with DMEM diluted in 90% sterile water ( Hypo-osmotic; Hypo ) for 20 min at 37°C . Post nuclear supernatant ( soluble cytoplasmic fraction ) were prepared by extensive washing of cells in PBS . Cells were then scraped into PBS containing protease and phosphatase inhibitors followed by mechanic disruption by syringe lysis . Cells were then pelleted at 2000 rpm , 10 min at 4°C . 120th of the supernatant was retained as the starting material ( B ) . The remaining supernatant was mixed with 20 μl of prewashed GFP Trap beads for 30 min at 4°C on a rotating wheel . The beads were pelleted and washed three times in PBS supplemented with protease and phosphatase inhibitors and were boiled in 4× Sample buffer for 75°C for 2 . 5 min to preserve the fluorescence . Sample were separated by SDS-PAGE and the fluorescence corresponding to each of the overexpressed GFP or Cherry tagged Cavin construct was detected and quantification in the gel using the BioRad ChemiDoc MP Imaging System . DOI: http://dx . doi . org/10 . 7554/eLife . 01434 . 014 The hypotonic treatment mediated release of cavin protein from the membrane provides an opportunity to test the cohesion of the observed cavin1 , 2 and 3 caveolar coat . Examination of the cavins released in the cytosol of hypo-osmotic cells showed that all cavins now existed as small oligomers . Remarkably we observed a complete segregation of cavin2 and cavin3 . We observed rare oligomers ( >15% ) made of cavin1-only ( red peak in histograms ) , but not oligomers made of cavin2-only or cavin3-only . The single-molecule coincidence histogram shows clearly that cavin3 remains bound to cavin1 , and cavin2 bound to cavin1 , and that these two subcomplexes detach separately from the membrane . This was confirmed by classical biochemistry and pull-downs as shown in Figure 4—figure supplement 1 . Interestingly , even the size of the two subcomplexes differs , with cavin1-cavin1 and cavin1-cavin3 forming oligomers of similar size ( 20 nm ) that are significantly smaller than the cavin1-cavin2 sub-complex ( 30 nm ) . The complete lack of co-diffusion indicates that subcomplexes do not form direct interactions in solution; it also suggests that on the membrane , they do not bind strongly to form stable mixed structures . The identification of two distinct cavin complexes raised the question of their assembly and spatial organization on the cytoplasmic face of caveolae . An immuno-EM method was developed using an ‘unroofing’ protocol modified from Heuser ( 2000 ) . 3T3-L1 fibroblasts transiently expressing cavin1-Cherry , cavin2-Cherry ( co-transfected with cavin1-FLAG ) or cavin3-mCherry ( co-transfected with cavin1-FLAG ) were unroofed and pre-embedding labeling was performed using a directly conjugated 3 nm-α-RFP gold ( α-RFP antibody is cross-reactive with the Cherry-tag ) as described in ‘Materials and methods’ . Cavin1 ( Figure 5A–top two rows ) , cavin2 ( Figure 5A–middle two rows ) and cavin3 ( Figure 5A–bottom two rows ) respectively demonstrated immunolabeling consistent with a striated pattern . Predicted three-dimensional orientations of the immunolabeling were generated from thin sections and are depicted in Figure 5 right hand columns . 10 . 7554/eLife . 01434 . 015Figure 5 . The cavins form part of the striated caveola coat and cavin2 and cavin3 localize to the same caveola while remaining spatially distinct . ( A ) High magnification images of immuno-EM labeled 3T3-L1 fibroblasts expressing cavin1 ( top row ) , cavin2 ( middle row ) and cavin3 ( lower row ) and their predicted three-dimensional orientation ( right hand columns , red line = predicted orientation based on a model of cytoplasmic caveolar striations , yellow box = field of view from electron micrograph of 60 nm section ) . Scale bars , 100 nm . ( B ) High magnification images of cavin1-Cherry ( 3 nm gold–colored red ) and cavin3-GFP ( 7 nm gold–colored green ) label the same caveolae and have significant overlap . Scale bars , 100 nm . ( C ) Cavin-2-GFP ( 7 nm gold ) and cavin-3-Cherry ( 3 nm gold ) demonstrates the spatial separation between both proteins within individual caveolae , consistent with the formation of separate striations . Black circles in the low magnification image denote the spatial limit of the individual caveola and its associated coat of caveolae that are positive for both cavin2 and cavin3 . Scale bars , 100 nm . ( D ) Bivariate clustering analysis of different plasma membrane lawns expressing various combinations of cavin proteins . Significantly higher co-clustering of cavin1-cavin3 was observed when compared to cavin2-cavin3 ( CI = confidence interval ) . Scale bars , 100 nm . ( E ) Model of association and disassociation of the cavin1-cavin2 and cavin1-cavin3 subcomplexes . DOI: http://dx . doi . org/10 . 7554/eLife . 01434 . 015 To determine if subcomplexes of cavin2 and cavin3 could be resolved on intact caveolae , immuno-EM was performed on the basolateral surface of the plasma membrane in cells expressing cavin2 and cavin3 . 3T3-L1 fibroblasts were transfected with either GFP- or Cherry-tagged cavin constructs and were ‘unroofed’ . Pre-embedding labeling was performed using with two different directly conjugated antibodies: 3 nm-α-RFP and 7 nm-α-GFP . In accordance with the single-molecule fluorescence data , cavin1-Cherry ( 3 nm gold–coloured red ) and cavin3-GFP ( 7 nm gold–coloured green ) demonstrated a close spatial association and co-labeled the same caveolae with no perceivable separation ( Figure 5B ) . Dual-labeling of cavin2-GFP ( 7 nm–coloured green ) and cavin3-Cherry ( 3 nm–coloured red ) , in cells expressing roughly equivalent amounts ( with co-expression of cavin1-FLAG ) , demonstrated that cavin2 and cavin3 were localized to the same caveola ( Figure 5C ) . However , within an individual caveola significant spatial separation between these proteins was observed , consistent with separate striations . In order to verify and quantify these observations , Ripley’s K function bivariate analysis of the co-clustering of cavin2-GFP and cavin3-Cherry to the co-clustering of cavin1-Cherry and cavin3-GFP was performed . There was a significantly lower level of co-clustering of cavin2-cavin3 as compared to cavin1-cavin3 ( Figure 5D ) . These results , together with single-molecule fluorescence studies , demonstrate that cavin1-cavin2 and cavin1-cavin3 complexes form sub-caveolar complexes on the cytoplasmic face of caveolae . These subcomplexes form distinct striations that we propose can shape caveolae and peel away as caveolae flatten ( see model Figure 5E ) .
Here we report the development of a novel single-molecule fluorescence method for analysis of self-assembly and caveolar association of cavin proteins in vivo and in vitro . We quantitatively describe the homo- and hetero-oligomerization properties of the three cavin family members and we demonstrate a specificity of interactions that generate two discrete subcomplexes . All three cavin members can associate with the same caveolae , but in two subcomplexes that can dissociate separately in response to membrane stretch . For each of the cavins the resulting assembly is of a defined size , estimated to be approximately 50 monomers . The consistent size of the cavin oligomers shows that the process is not random and that the cavins have an intrinsic property to limit their growth . One possible mechanism for this behaviour would be a structure regulated by curvature , driven by higher order interactions between the cavin subunits . The methodological toolbox developed in this study allowed us to quantitatively examine the cavin complexes in three different states: ( 1 ) in association with caveolae in cell extracts and on intact caveolae at the plasma membrane; ( 2 ) in the cytoplasm , in the absence of caveolin and caveolae; and ( 3 ) upon dissociation from membrane-bound caveolae in response to hypotonic medium . 1 ) Our data suggest that the cavin complex associated with caveolae in the cell extract is made of typically 50 cavin proteins . We believe that the contribution of endogenous proteins must be minor because of the striking consistency in the size of the complexes observed both for cavins and caveolins in our different systems ( Leishmania cell-free system , MCF-7 cells that lack endogenous cavin proteins , and MDCK cells that natively express cavins ) . The comparison of results obtained in in vitro and in vivo systems shows clearly that each of the cavin proteins has an intrinsic property for self-association to form stable oligomers . After synthesis , in vitro large oligomers are formed at very low concentration ( <10 nM , see Figure 2—figure supplement 1 ) , suggesting that cavins would form the same oligomers at endogenous levels . Most importantly , recent studies using biochemical approaches ( Ludwig et al . , 2013 ) show striking agreement with our observations . A ratio of CAV1 to cavin of 4:1 determined biochemically ( Ludwig et al . , 2013 ) is in good agreement with the brightness analysis and CAV1-cavin1 coincidence data presented here . Similarly , the ratio of cavin1 to cavin3 of ∼2:1 is in excellent agreement with the result obtained from quantitative pull-downs ( Ludwig et al . , 2013 ) . The same ratio of 2:1 of cavin1 to cavin2 observed in subcomplexes released from caveolae ( Figure 4D ) is similar to that reported in Ludwig et al . ( 2013 ) but notably in the absence of caveolae more variable ratios are observed ( Figure 2 in MCF-7 and cell-free systems ) . The studies indicate that the number of cavin proteins in the caveolar coat is relatively low , considering the number of CAV1 proteins within a caveola , and compared to the other protein coats such as clathrin ( made of 30 + triskelias , approximately 200 subunits ) . We find that when caveolae are disassembled by membrane stretch , cavins are released as subcomplexes of ∼9 cavin molecules , suggesting a typical caveola contains typically 5 cavin subcomplexes . Cavin1 is believed to form trimers , possibly through a coiled-coil domain ( Ludwig et al . , 2013 ) , suggesting that each sub-complex may be composed of three trimers . Our data clearly shows however , that cavin2 and cavin3 are not present in the same assemblies with cavin1: the sub-complexes are mutually exclusive for cavin2 or cavin3 . Furthermore , cavin2 and cavin3 also segregate spatially on the surface of caveolae as revealed by high resolution immunoelectron microscopy . This segregation is consistent with the formation of striations as shown in Figure 5 . Thus stable units of cavin1-cavin2 and cavin1-cavin3 can come together on the membrane to regulate the formation of the curved caveolar structure . 2 ) Even in the absence of the scaffolding element CAV1 , cavins can assemble in the cytoplasm forming a complex that has a similar size to a caveola assembly . But this does not mean that the cavin complex will be fully formed before interacting with caveolae . Expression in cells lacking caveolins may reflect an artificial situation as cavins and caveolins are generally expressed together in vivo . We believe the large complexes formed under these conditions may be rare in vivo , possibly explaining the slow association of expressed cavin with newly-arrived caveolin at the plasma membrane ( 25 min ) when fluorescently-tagged cavin is expressed in mammalian cells ( Hayer et al . , 2010 ) . The size of the cavin1 complex in the absence of CAV1 shows a broader size distribution and can be even larger than the size of a cavin1-CAV1 containing particle , where the presence of CAV1 appears to condense the complex ( Figure 3F vs Figure 1J , raw data presented in Figure 3—figure supplement 1 ) . This observation suggests that the 50-mer cavin1 complexes form a more heterogeneous population in terms of their size and shape , which become more tightly organized when bound to CAV1 . The fact that oligomers of cavin1 form so readily in the cell-free expression system ( Figure 2—figure supplement 1 ) suggests that the subcomplexes could be pre-assembled rapidly in the cytosol , before reaching the CAV1-rich domains at the membrane . We hypothesize that a combined interaction with membranes ( Hansen et al . , 2009 ) and caveolin itself will result in recruitment of the subcomplexes to the budding caveolae , where they can be assembled into the striated structures observed on intact caveolae ( this study and Ludwig et al . , 2013 ) . 3 ) The use of hypotonic treatment to swell cells allowed us to gain insights into the complexes that dissociate from membrane-bound caveolae in a reversible fashion ( Sinha et al . , 2011 ) . We show that the cavin coat dissociates into stable sub-elements , cavin1/cavin2 and cavin1/cavin3 subcomplexes with a well-defined size of 9 ± 2 cavin proteins . Interestingly , based on the observed residence times in the focal volume the complexes have slightly different sizes with cavin1-cavin1 20 nm , cavin1-cavin2 30 nm and cavin1-cavin3 20 nm . These sizes are calculated based on the assumption that the diffusing objects are spherical , which remains an oversimplification until further structural information is available . The cavin sub-oligomers are very stable and the proteins remain associated even after caveola disassembly , with few monomers detected in solution , as shown by the distribution of sizes; all cavin proteins are associated with objects of apparent size greater than 15 nm ( Figure 4G ) . By high resolution immuno-EM and quantitative spatial analysis we show for the first time that cavin2 and cavin3 label distinct subdomains of single caveolae . The preferential labeling of spatially-distinct linear elements raises the possibility that the cavin1/cavin2 and cavin1/cavin3 complexes actually make up the caveolar striations seen by high resolution SEM and deep etch electron microscopy ( Peters et al . , 1985; Rothberg et al . , 1992 ) consistent with EM observations with miniSOG-cavin fusion proteins ( Ludwig et al . , 2013 ) . These striated ‘nanodomains’ could be released from ( ‘peel off’ ) the caveolae/membrane in response to membrane stretch . We speculate that this could potentially release two distinct subcomplexes , each able to reach distinct cellular destinations or interact with different partners . These interactions could be further regulated by additional post-translational modifications to individual cavins . An additional possibility is that the properties of individual caveolae can be modulated by their ratio of cavin1 , 2 , and 3 subcomplexes to vary their individual properties , including stability or morphology . The mode of assembly of cytoplasmic caveolar coat proteins shown here is distinct from that of any other known coat complex . Cavins form characteristic homo- and hetero-oligomeric complexes , which associate to form striated nanodomains that can wrap around caveolae . In contrast to other coat proteins , such as clathrin , which form a very dense basket-like network in a cooperative and stepwise process through protein–protein interactions and binding to membrane components and then is disassembled into individual triskelia in an ATP-dependent process , we hypothesize that caveolar assembly relies on association of pre-oligomerized cavin subcomplexes that give rise to the unique caveolar striations . Further functional characterization of the cavin subcomplexes should prove illuminating in understanding the diverse roles of caveolae .
MCF-7 and MDCK cells were maintained as previously described ( Kirkham et al . , 2008 ) . MDCK cells were grown in DMEM/F-12 ( Life Technologies , Carlsbad , CA , USA ) supplemented with 10% FBS and 2 mM L-glutamine . MCF-7 cells were grown in DMEM supplemented with 10% FBS and 2 mM L-glutamine . Caveolin-1 , cavin1 , cavin2 and cavin3 constructs were cloned as described in Hill et al . ( 2008 ) . Tagged constructs were transfected using Lipofectamine 2000 reagent ( Life Technologies , CA , USA ) following the manufacturer’s instructions using a 1:3 ratio of DNA:Lipofectamine . 3T3-L1 fibroblasts were grown in DMEM with 10% fetal bovine serum ( FBS ) and 1 mM L-glutamine ( Life Technologies , CA , USA ) . All transient transfections were performed with Lipofectamine 2000 ( Life Technologies ) as per the manufacturer’s instruction . Cell-free expression of proteins for interaction mapping used the eukaryotic Leishmania cell-free system . Manufacture and supplementation of lysate from Leishmania tarentolae is as described in Kovtun et al . ( 2011 ) . DNA templates for the various ORFs used the Gateway cloning system , with ribosome engagement with T7 transcribed mRNA mediated by the species independent translation initiation site ( Mureev et al . , 2009 ) . Coupled transcription/translation occurred for 3 hr at 27°C unless described as otherwise . Single-molecule spectroscopy was performed based on ( Gambin et al . , 2009 , 2011 ) . 20 µl of samples are used for each experiment , placed into a custom-made silicone 192-well plate equipped with a 70 × 80 mm glass coverslip ( ProSciTech Australia ) . Plates were analysed on a Zeiss LSM710 microscope with a Confocor3 module , at room temperature . Two lasers ( 488 nm and 561 nm ) are co-focussed in solution using a 40 × 1 . 2 NA water immersion objective ( Zeiss , Germany ) ; fluorescence was collected and split into GFP- and Cherry-channel by a 560 nm dichroic mirror . The GFP emission was further filtered by a 505–540 nm band pass filter and the Cherry emission was filtered by a 580 nm long-pass filter . The single-molecule multicolour detection method is based on a simple principle: the two excitation lasers are focussed within the same point , creating a very small observation volume ( ∼1 fl ) . The proteins are tagged with genetically-encoded GFP and Cherry fluorophores . Proteins diffuse freely by Brownian motion , and they enter and exit the confocal volume constantly . To reach single-molecule detection , the samples were diluted to approximately 100 pM concentration , so that only single proteins or protein complexes are present in the confocal volume . For single-molecule coincidence , the time-trace of GFP and Cherry intensity was binned in 100 microseconds time ‘bins’ . That is , we calculate the number of photons collected in 100 ms . When the fluorescence is greater than the defined threshold ( 20 photons ) , we consider that fluorophores are present in the focal volume . When a burst lasts for more than one time bin , we ‘stitch’ the bursts together . That is , we calculate the total fluorescence emitted by the diffusing objects as long as they are in the focal volume . For single-molecule coincidence , the coincidence ratio is calculated as Mukhopadhyay et al . ( 2007 ) [Intensity ( Cherry ) −leakage from GFP channel]/total intensity ( Intensity ( Cherry ) +Intensity ( GFP ) ) ; the leakage from GFP to Cherry channel was experimentally determined at 10% . An average of 1000 events was accumulated for each single-molecule coincidence histogram . The brightness analysis is calibrated using GFP expressed in the same MCF-7 cells or cell-free extracts . Due to random diffusion , most of the events are the result of incomplete transfer through the focal volume , and few bursts represent the maximal number of photons that the GFP-tagged proteins can emit . The perfect trajectory , maximizing the time spent in the middle of the focal volume , is extremely rare , and the protein complex is more likely to exit the detection volume quickly . The escape from the detection volume can be modeled as a first order decay: the probability of a long residence in the focal volume is the combined effect of a succession of random events , each with a successful outcome . In this simple model , following the laws of probability , the longer bursts will be exponentially less frequent . The probability of detecting a burst of a given brightness ( plotted in Figures 1 and 3 ) seems to be described well by this model , as shown by the linear decrease in the log-scaled graphs . The behaviour of GFP is extremely well defined , as we obtained a profile of burst sizes that rapidly decreases , and no bursts were observed above 100 photons . In case of larger oligomerization of the GFP- labeled cavins , multiple proteins and their fluorescent labels pass the confocal volume at the same time , resulting in an even higher photon count . For single-molecule measurement of residence times , the time trace was acquired with a much higher frequency of 100 ns per time bin . We define a 1 s window before and after each burst and the intensity I ( t ) at short timescales is analysed for correlation as <G ( T ) ) >=<I ( t ) . I ( t + T ) >/<I ( t ) > . <I ( t + T ) > using the FCS ( Fluorescence Correlation Spectroscopy ) software ( Zeiss710 Confocor ) ( Ferreon et al . , 2009 ) . This single-molecule FCS gives direct information on the size of protein complexes . FCS typically leads to an average correlation curve and an average diffusion value , measured over thousands of proteins . On the contrary , the analysis burst-by-burst gives access to a distribution of sizes , where large aggregates can be separated from smaller oligomers . This dwell time measured at the single-molecule level for GFP monomers corresponds perfectly to the one measured at a 10 nM concentration using FCS . We did not attempt single-molecule FCS using Cherry fluorophores , as the FCS curves typically display an important contribution from triplet states with a characteristic time of 30 μs . This phenomenon interferes with the accurate determination of residence times . ALPHAScreen cMyc detection and Proxiplate-384 Plus 384-wells plates were purchased from Perkin Elmer ( MA , USA ) . Proteins ( one bearing a N-terminal eGFP tag , the other labeled with C-terminal mCherry-Myc ) were co-expressed in the cell-free system by adding mixed DNA vectors in 10 μl of the Leishmania tarentolae-based cell-free system ( to a final DNA concentration of 30 nM for the GFP-vector and 60 nM for the Cherry-vector ) The mixture was incubated for 3 . 5 hr at 27°C . Four serial dilutions of the proteins of ¼ were made in buffer A ( 25 mM HEPES , 50 mM NaCl ) . The AlphaScreen Assay was performed in 384-well plates . Per well , 0 . 4 μg of the Anti-Myc coated Acceptor Beads ( PerkinElmer , MA , USA ) was added in 12 . 5 μl reaction buffer B ( 25 mM HEPES , 50 mM NaCl , 0 . 001% NP40 , 0 . 001% casein ) . 2 μl of the diluted proteins and 2 μl of biotin labeled GFP-nanotrap ( diluted in reaction buffer A to a concentration of 45 nM ) were added to the acceptor beads , followed by incubation for 45 min at room temperature . Then 0 . 4 μg of Streptavidin coated donor beads diluted in 2 μl buffer A were added , followed by an incubation in the dark for 45 min at room temperature . ALPHAScreen signal was recorded on a PE Envision Multilabel Platereader , using the manufacturer’s recommended settings ( excitation: 680/30 nm for 0 . 18 s , emission: 570/100 nm after 37 ms ) . The signals were then averaged and normalized to background . For hypo-osmotic treatments , cells were treated with 100% DMEM ( iso-osmotic ) or with a dilution of DMEM ( 1:10 ) in sterile water . The duration of the hypo-osmotic treatment was 20 min at 37°C . MCF-7 or MDCK cells were transfected with equal amounts of cavin constructs as specified in the pertinent figure legends . The soluble cytoplasmic fractions were prepared by extensive washing of cells in PBS . Cells were then scraped into PBS containing protease ( Merck Pty Ltd , Kilsyth , Australia ) and phosphatase ( Roche Diagnostics Australia , Castle Hill , Australia ) inhibitors . Cells were mechanically disrupted by syringe lysis using a 1 mL syringe and a 25-gauge needle ( Terumo , Tokyo Japan ) . Cells were then pelleted at 2000 rpm , 10 min at 4°C . 1/120th of the supernatant was retained as the starting material . The remaining supernatant was mixed with 20 μl of prewashed sepharose conjugated antiGFP beads ( Kovtun et al . , 2011 ) for 30 min at 4°C on a rotating wheel . The beads were pelleted and washed three times in PBS supplemented with protease and phosphatase inhibitors and were boiled in 4 X SDS containing sample buffer at 75°C for 2 . 5 min to preserve the fluorescence of GFP . Sample were pelleted at maximal speed for 10 min at room temperature and the supernatant was separated by SDS-PAGE and the fluorescence was detected and quantified in the gel using the BioRad ChemiDoc MP Imaging System ( BioRad , Hercules , California , USA ) . Cells were grown as described above , plated onto 6 cm tissue culture dishes ( TRP ) , transfected overnight with the plasmid of interest , trypsinized ( TryplExpress–Gibco ) and replated onto 3 cm tissue culture dishes ( TRP ) that had been pre-coated with Poly-L-Lysine ( Sigma–Aldrich , St Louis , MO , USA ) . Cells were incubated for 4 hr to ensure attachment and flattening of cells to the coated dish , washed with PHEM buffer ( 80 mM PIPES , 25 mM HEPES , 10 mM EGTA and 2 mM MgCl2 at pH 6 . 9 ) and sonicated with a probe tip sonicator ( Vir Sonic Ultrasonic Cell Disruptor 100–setting 3 ) at an angle of 45° . Baso-lateral membranes were selected using fluorescence microscopy ( EVOSfl AMG-Life Technologies , CA , USA fitted with green , red and blue filters ) . Areas with fluorescent membranes but lacking nuclei were selected and washed repeatedly with PHEM buffer , blocked in 0 . 2% bovine serum albumin and 0 . 2% fish skin gelatin then incubated with α-GFP and α-RFP antibodies . Dual labeling was performed as follows: unroofed cells were labeled first with the large 7 nm α-GFP antibody for 30 min , washed with block and then re-labeled with the 3 nm α-RFP antibody for 30 min . Basal membranes were then fixed in 2 . 5% glutaraldehyde ( Electron Microscopy Sciences , PA , USA ) for 1 hr at room temperature and then post-fixed in 1% OsO4 for 1 hr , also at room temperature . Dishes were serially dehydrated in increasing percentages of ethanol in a BioWave microwave and LX-112 resin was infiltrated at ratios of 1:2 , 1:1 , 2:1 ( resin to ethanol ) and 100% resin twice in a BioWave microwave . Resin was polymerized at 60°C overnight . Ultrathin sections ( 55–65 nm ) were then cut , such that only the very first sections were collected ( i . e . , the most basal region of the cell ) and imaged in an JEOL1011 electron microscope at 80 kV equipped with a Morada Soft Imaging System 4K × 4K camera at twofold binning under the control of iTEM ( Olympus , Japan ) . Ripley’s bivariate K-function analysis was performed as described ( Prior et al . , 2003 ) . | If you could look closely enough at the surface of some animal cells , especially fat or muscle cells , you would see that they are covered with pocket-like indents called ‘caveolae’ . These structures are thought to help the cells communicate with the outside world , but they can also be used by viruses to gain entry into living cells . Examining these caveolae even closer would reveal that these pockets contain proteins called caveolins that bind to each other—and also to cholesterol and fatty acids—to form a scaffold that help to maintain the shape of the caveolae from inside the cell . Each caveolae in a mammalian cell typically contains over 100 caveolin proteins . Caveolar coat proteins , or cavins for short , are also important building blocks for caveolae: however , we know relatively little about the interactions between caveolins and cavins . Now , Gambin et al . have used powerful new single-molecule techniques to study these interactions . These experiments looked at the three main types of cavin proteins that associate with caveolae , and by tracking individual protein molecules they showed that cavin1 can interact with either cavin2 or cavin3 , but that cavin2 and cavin3 do not interact with each other . Furthermore , cavin2 and cavin3 exist in separate stripes on a caveolae . Gambin et al . also stretched the cell membrane by forcing cells to take in extra water , and showed that this caused the cavin coat to peel away from the caveolae and break down into distinct cavin1-cavin2 and cavin1-cavin3 building blocks . Faulty versions of caveolins and cavins have both been associated with several diseases in humans , including heart disease and muscle disorders . As such , an improved understanding of the formation and break down of caveolae may prove useful for developing treatments for these conditions . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"cell",
"biology",
"structural",
"biology",
"and",
"molecular",
"biophysics"
] | 2014 | Single-molecule analysis reveals self assembly and nanoscale segregation of two distinct cavin subcomplexes on caveolae |
The flexible control of sequential behavior is a fundamental aspect of speech , enabling endless reordering of a limited set of learned vocal elements ( syllables or words ) . Songbirds are phylogenetically distant from humans but share both the capacity for vocal learning and neural circuitry for vocal control that includes direct pallial-brainstem projections . Based on these similarities , we hypothesized that songbirds might likewise be able to learn flexible , moment-by-moment control over vocalizations . Here , we demonstrate that Bengalese finches ( Lonchura striata domestica ) , which sing variable syllable sequences , can learn to rapidly modify the probability of specific sequences ( e . g . ‘ab-c’ versus ‘ab-d’ ) in response to arbitrary visual cues . Moreover , once learned , this modulation of sequencing occurs immediately following changes in contextual cues and persists without external reinforcement . Our findings reveal a capacity in songbirds for learned contextual control over syllable sequencing that parallels human cognitive control over syllable sequencing in speech .
A crucial aspect of the evolution of human speech is the development of flexible control over learned vocalizations ( Ackermann et al . , 2014; Belyk and Brown , 2017 ) . Humans have unparalleled control over their vocal output , with a capacity to reorder a limited number of learned elements to produce an endless combination of vocal sequences that are appropriate for current contextual demands ( Hauser et al . , 2002 ) . This cognitive control over vocal production is thought to rely on the direct innervation of brainstem and midbrain vocal networks by executive control structures in the frontal cortex , which have become more elaborate over the course of primate evolution ( Hage and Nieder , 2016; Simonyan and Horwitz , 2011 ) . However , because of the comparatively limited flexibility of vocal production in nonhuman primates ( Nieder and Mooney , 2020 ) , the evolutionary and neural circuit mechanisms that have enabled the development of this flexibility remain poorly understood . Songbirds are phylogenetically distant from humans , but they have proven a powerful model for investigating neural mechanisms underlying learned vocal behavior . Song learning exhibits many parallels to human speech learning ( Doupe and Kuhl , 1999 ) ; in particular , juveniles need to hear an adult tutor during a sensitive period , followed by a period of highly variable sensory-motor exploration and practice , during which auditory feedback is used to arrive at a precise imitation of the tutor song ( Brainard and Doupe , 2002 ) . This capacity for vocal learning is subserved by a well-understood network of telencephalic song control nuclei . Moreover , as in humans , this vocal control network includes strong projections directly from cortical ( pallial ) to brainstem vocal control centers ( Doupe and Kuhl , 1999; Simonyan and Horwitz , 2011 ) . These shared behavioral features and neural specializations raise the question of whether songbirds might also share the capacity to learn flexible control over syllable sequencing . Contextual variation of song in natural settings , such as territorial counter-singing or female-directed courtship song , indicate that songbirds can rapidly alter aspects of their song , including syllable sequencing and selection of song types ( Chen et al . , 2016; Heinig et al . , 2014; King and McGregor , 2016; Sakata et al . , 2008; Searcy and Beecher , 2009; Trillo and Vehrencamp , 2005 ) . However , such modulation of song structure is often described as affectively controlled ( Berwick et al . , 2011; Nieder and Mooney , 2020 ) . For example , the presence of potential mates or rivals elicits a global and unlearned modulation of song intensity ( James et al . , 2018 ) related to the singer’s level of arousal or aggression ( Alcami et al . , 2021; Heinig et al . , 2014; Jaffe and Brainard , 2020 ) . Hence , while prior observations suggest that a variety of ethologically relevant factors can be integrated to influence song production in natural settings , it remains unclear whether song can be modified more flexibly by learned or cognitive factors . Here , we tested whether Bengalese finches can learn to alter specifically targeted vocal sequences within their songs in response to arbitrarily chosen visual cues , independent of social or other natural contexts . Each Bengalese finch song repertoire includes ~5–12 acoustically distinct elements ( ‘syllables’ ) that are strung together into sequences in variable but non-random order . For a given bird , the relative probabilities of specific transitions between syllables normally remain constant over time ( Okanoya , 2004; Warren et al . , 2012 ) , but previous work has shown that birds can gradually adjust the probabilities of alternative sequences in response to training that reinforces the production of some sequences over others . In this case , changes to syllable sequencing develop over a period of hours to days ( Warren et al . , 2012 ) . In contrast , we investigate here whether birds can learn to change syllable sequencing on a moment-by-moment basis in response to arbitrary visual cues that signal which sequences are adaptive at any given time . Our findings reveal that songbirds can learn to immediately , flexibly , and adaptively adjust the sequencing of selected vocal elements in response to learned contextual cues .
For each bird in the study , we first identified variably produced syllable sequences that could be gradually modified using a previously described aversive reinforcement protocol ( ‘single context training’; Tumer and Brainard , 2007; Warren et al . , 2012 ) . For example , a bird that normally transitioned from the fixed syllable sequence ‘ab’ to either ‘c’ or ‘d’ ( Figure 1A , B , sequence probability of ~36% for ‘ab-c’ and ~64% for ‘ab-d’ ) was exposed to an aversive burst of white noise ( WN ) feedback immediately after the ‘target sequence’ ‘ab-d’ was sung . In response , the bird learned over a period of days to gradually decrease the relative probability of that sequence in favor of the alternative sequence ‘ab-c’ ( Figure 1C ) . This change in sequence probabilities was adaptive in that it enabled the bird to escape from WN feedback . Likewise , when the sequence , ‘ab-c’ was targeted , the probability of ‘ab-d’ increased gradually over several days of training ( Figure 1D ) . These examples are consistent with prior work that showed such sequence modifications develop over a period of several days , with the slow time course suggesting a gradual updating of synaptic connections within syllable control networks in response to performance-related feedback ( Warren et al . , 2012 ) . In contrast , the ability to immediately and flexibly reorder vocal elements in speech must reflect mechanisms that enable contextual factors to exert moment-by-moment control over selection and sequencing of alternative vocal motor programs . Having identified sequences for each bird for which the probability of production could be gradually modified in this manner , we then tested whether birds could be trained to rapidly switch between those same sequences in a context-dependent manner . To determine whether Bengalese finches can learn to flexibly select syllable sequences on a moment-by-moment basis , we paired WN targeting of specific sequences with distinct contextual cues . In this context-dependent training protocol , WN was targeted to defined sequences in the bird’s song as before , but the specific target sequence varied across alternating blocks , signaled by different colored lights in the home cage ( see Materials and methods ) . Figure 1E shows an example experiment , with ‘ab-d’ targeted in yellow light , and ‘ab-c’ in green light . At baseline , without WN , switches between yellow and green contexts ( at random intervals of 0 . 5–1 . 5 hr ) did not lead to significant changes in the relative proportion of the target sequences , indicating that there was no inherent influence of the light cues on sequence probabilities ( Figure 1F , p ( ab-d ) in yellow vs . green context was 67 ± 1 . 6% vs . 64 ± 1 . 5% , p=0 . 17 , rank-sum test , n = 53 context blocks from baseline period ) . Training was then initiated in which WN was alternately targeted to each sequence , over blocks that were signaled by light cues . After 2 weeks of such context-specific training , significant sequencing differences developed between light contexts that were appropriate to reduce aversive feedback in each context ( Figure 1G , p ( ab-d ) in yellow vs . green context shifted to 36 . 5 ± 4 . 8% vs . 83 . 1 ± 3 . 5% , p<0 . 01 , rank-sum test , n = 22 context blocks , block duration between 1 and 2 . 5 hr ) . Likewise , for all birds trained on this protocol ( n = 8 ) , context-dependent sequencing differences developed in the appropriate direction over a period of weeks ( 27 ± 6% difference in probabilities between contexts after a mean of 33 days training , versus 1% ± 2% average difference in probabilities at baseline; p<0 . 01 , n = 8 , signed rank test , Figure 1H ) . Thus , Bengalese finches are able to learn context-specific modifications to syllable sequencing . Contextual differences between different blocks could arise through an immediate shift in sequence probabilities upon entry into a new context and/or by rapid learning within each block . We examined whether trained birds exhibited any immediate shifts in their syllable sequencing when entering a new light context by computing the average probability of target sequences across songs aligned with the switch between contexts ( Figure 2A , B , example experiment ) . This ‘switch-triggered average’ revealed that across all birds , switches to the yellow context were accompanied by an immediate decrease in the probability of the yellow target sequence , whereas switches out of the yellow context ( and into the green context ) led to an immediate increase in the yellow target sequence ( Figure 2C , D , p<0 . 05 , signed rank test comparing first and last song , n = 8 ) . To quantify the size of these immediate shifts , we calculated the difference in sequence probability from the last five songs in the previous context to the first five songs in the current context; this difference averaged 0 . 24 ± 0 . 06 for switches to green light and −0 . 22 ± 0 . 06 for switches to yellow light ( Figure 2E , F ) . These results indicate that birds could learn to immediately recall an acquired memory of context-appropriate sequencing upon entry into each context , even before having the chance to learn from reinforcing feedback within that context . We next asked whether training additionally led to an increased rate of learning within each context , which also might contribute to increased contextual differences over time . Indeed , such faster re-learning for consecutive encounters of the same training context , or ‘savings’ , is sometimes observed in contextual motor adaptation experiments ( Lee and Schweighofer , 2009 ) . To compare the magnitude of the immediate shift and the magnitude of within-block learning over the course of training , we plotted the switch-aligned sequence probabilities at different points in the training process . Figure 2G shows for the example bird that the magnitude of the shift ( computed between the first and last five songs across context switches ) gradually increased over 11 days of training . Figure 2H shows the switch-aligned sequence probability trajectories ( as in Figure 2A , B ) for this bird early in training ( red ) and late in training ( blue ) , binned into groups of seven context switches . Qualitatively , there was both an abrupt change in sequence probability at the onset of each block ( immediate shift at time point 0 ) and a gradual adjustment of sequence probability within each block ( within-block learning over the first 80 songs following light switch ) . Over the course of training , the immediate shift at the onset of each block got larger , while the gradual change within blocks stayed approximately the same ( learning trajectories remained parallel over training , Figure 2H ) . Linear fits to the sequence probabilities for each learning trajectory ( i . e . the right side of Figure 2H ) reveal that , indeed , the change in sequence probability at the onset of blocks ( i . e . intercepts ) increased over the training process ( Figure 2K ) , while the rate of change within blocks ( i . e . slopes ) stayed constant ( Figure 2I ) . To quantify this across birds , we measured the change over the course of learning in both the magnitude of immediate shifts ( estimated as the intercepts from linear fits ) and the rate of within-block learning ( estimated as the slopes from linear fits ) . As for the example bird , we found that the rate of learning within each block stayed constant over time for all five birds ( Figure 2L ) . In contrast , the magnitude of immediate shifts increased over time for all birds ( Figure 2L ) . These analyses indicate that adjustments to sequence probability reflect two dissociable processes , an immediate cue-dependent shift in sequence probability at the beginning of blocks , that increases with contextual training , and a gradual adaptation of sequence probability within blocks , that does not increase with contextual training . The ability of Bengalese finches to implement an immediate shift in sequencing on the first rendition in a block – and thus before they have a chance to learn from reinforcing feedback – argues that they can maintain context-specific motor memories and use contextual visual cues to anticipate correct sequencing in each context . To explicitly test whether birds can flexibly switch between sequencing appropriate for distinct contexts using only visual cues , we included short probe blocks which presented the same light cues without WN stimulation . Probe blocks were interspersed in the sequence of training blocks so that each switch between types of blocks was possible and , on average , every third switch was into a probe block ( see Materials and methods ) . Light switches into probe blocks were associated with similar magnitude shifts in sequence probability as switches into WN blocks of the corresponding color ( −0 . 22 ± 0 . 06 to both yellow WN and yellow probe blocks from green WN blocks , p=0 . 94 , signed rank test; 0 . 24 ± 0 . 06 to green WN and 0 . 23 ± 0 . 07 to green probe blocks from yellow WN blocks , p=0 . 64 , signed rank test ) . As the most direct test of whether light cues alone evoke adaptive sequencing changes , we compared songs immediately before and after switches between probe blocks without intervening WN training blocks ( probe-probe switches ) . Figure 3A , B shows song bouts for one example bird ( Bird 2 ) which were sung consecutively across a switch from yellow probe to green probe blocks . In the first song following the probe-probe switch , the yellow target sequence ( ‘f-ab’ ) was more prevalent , and the green target sequence ( ‘n-ab’ ) was less prevalent , and such an immediate effect was also apparent in the average sequence probabilities for this bird aligned to probe–probe switches ( Figure 3C , D ) . Similar immediate and appropriately directed shifts in sequencing at switches between probe blocks were observed for all eight birds ( Figure 3E , F , p<0 . 05 signed rank test , n = 8 ) , with average shifts in sequence probabilities of −0 . 21 ± 0 . 09 and 0 . 17 ± 0 . 08 ( Figure 3G , H ) . The presence of such changes in the first songs sung after probe–probe switches indicates that visual cues alone are sufficient to cause anticipatory , learned shifts between syllable sequences . A decrease in the probability of a target sequence in response to contextual cues must reflect changes in the probabilities of transitions leading up to the target sequence . However , such changes could be restricted to the transitions that immediately precede the target sequence , or alternatively could affect other transitions throughout the song . For example , for the experiment illustrated in Figure 1 , the prevalence of the target sequence ‘ab-d’ was appropriately decreased in the yellow context , in which it was targeted . The complete transition diagram and corresponding transition matrix for this bird ( Figure 4A , B ) reveal that there were four distinct branch points at which syllables were variably sequenced ( after ‘cr’ , ‘wr’ , ‘i’ , and ‘aab’ ) . Therefore , the decrease in the target sequence ‘ab-d’ could have resulted exclusively from an increase in the probability of the alternative transition ‘ab-c’ at the branch point following ‘aab’ . However , a reduction in the prevalence of the target sequence could also have been achieved by changes in the probability of transitions earlier in song such that the sequence ‘aab’ was sung less frequently . To investigate the extent to which contextual changes in probability were specific to transitions immediately preceding target sequences , we calculated the difference between transition matrices in the yellow and green probe contexts ( Figure 4C ) . This difference matrix indicates that changes to transition probabilities were highly specific to the branch point immediately preceding the target sequences ( specificity was defined as the proportion of total changes which could be attributed to the branch points immediately preceding target sequences; specificity for branch point ‘aab’ was 83 . 2% ) . Such specificity to branch points that immediately precede target sequences was typical across experiments , including cases in which different branch points preceded each target sequence ( Figure 4D–F , specificity 96 . 9% ) . Across all eight experiments , the median specificity of changes to the most proximal branch points was 84 . 95% , and only one bird , which was also the worst learner in the contextual training paradigm , had a specificity of less than 50% ( Figure 4G ) . Hence , contextual changes were specific to target sequences and did not reflect the kind of global sequencing changes that characterize innate social modulation of song structure ( Sakata et al . , 2008; Sossinka and Böhner , 1980 ) . Our experiments establish that birds can shift between two distinct sequencing states in response to contextual cues . In order to test whether birds were capable of learning to shift to these two states from a third neutral context , we trained a subset of three birds with three different color-cued contexts . For these birds , after completion of training with WN targeted to distinct sequences in yellow and green contexts ( as described above ) , we introduced interleaved blocks cued by white light in which there was no reinforcement . After this additional training , switches from the unreinforced context elicited changes in opposite directions for the green and yellow contexts ( example bird Figure 5A ) . All birds ( n = 3 ) showed adaptive sequencing changes for the first song bout in probe blocks ( Figure 5B , C ) as well as immediate shifts in the adaptive directions for all color contexts ( Figure 5D , 0 . 11 ± 0 . 04 and 0 . 19 ± 0 . 05 for switches to green WN and green probe blocks , respectively; −0 . 15 ± 0 . 06 and −0 . 09 ± 0 . 02 for switches to yellow WN and yellow probe blocks , respectively ) . While additional data would be required to establish the number of distinct associations between contexts and sequencing states that can be learned , these findings suggest that birds can maintain at least two distinct sequencing states separate from a ‘neutral’ state and use specific associations between cue colors and sequencing states to rapidly shift sequencing in distinct directions for each context .
Our demonstration of contextual control over the ordering of vocal elements in the songbird builds on previous work showing that a variety of animals can learn to emit or withhold innate vocalizations in response to environmental or experimentally imposed cues . For example , nonhuman primates and other animals can produce alarm calls that are innate in their acoustic structure , but that are deployed in a contextually appropriate fashion ( Nieder and Mooney , 2020; Suzuki and Zuberbühler , 2019; Wheeler and Fischer , 2012 ) . Similarly , animals , including birds , can be trained to control their vocalizations in an experimental setting , by reinforcing the production of innate vocalizations in response to arbitrary cues to obtain food or water rewards ( Brecht et al . , 2019; Hage and Nieder , 2013; Nieder and Mooney , 2020; Reichmuth and Casey , 2014 ) . In relation to these prior findings , our results demonstrate a capacity to flexibly reorganize the sequencing of learned vocal elements , rather than select from a fixed set of innate vocalizations , in response to arbitrary cues . This ability to contextually control the ordering , or syntax , of specifically targeted syllable transitions within the overall structure of learned song parallels the human capacity to differentially sequence a fixed set of syllables in speech . The ability to alter syllable sequencing in a flexible fashion also contrasts with prior studies that have demonstrated modulation of vocalizations in more naturalistic settings . For example , songs produced in the context of courtship and territorial or aggressive encounters ( ‘directed song’ ) differ in acoustic structure from songs produced in isolation ( ‘undirected song’ ) ( Sakata et al . , 2008; Searcy and Beecher , 2009 ) . This modulation of song structure by social context is characterized by global changes to the intensity of song production , with directed songs exhibiting faster tempo , and greater stereotypy of both syllable structure and syllable sequencing , than undirected songs ( Sakata et al . , 2008; Searcy and Beecher , 2009; Sossinka and Böhner , 1980 ) . This and other ethologically relevant modulation of song intensity may serve to communicate the singer’s affective state , such as level of arousal or aggression ( Alcami et al . , 2021; Hedley et al . , 2017; Heinig et al . , 2014 ) , and may largely reflect innate mechanisms ( James et al . , 2018; Kojima and Doupe , 2011 ) mediated by hypothalamic and neuromodulatory inputs to premotor regions ( Berwick et al . , 2011; Gadagkar et al . , 2019; James et al . , 2018; Nieder and Mooney , 2020 ) . In contrast , here we show that birds can learn to locally modulate specific features of their songs ( i . e . individually targeted syllable transitions ) in response to arbitrarily assigned contextual cues that have no prior ethological relevance . The capacity for moment-by-moment adjustment of vocalizations in response to arbitrary learned cues may depend on similar capacities that evolved to enable appropriate modulation of vocalizations in ethologically relevant natural contexts . For example , some species of songbirds preferentially sing different song types depending on factors such as time of day , location of the singer , or the presence of an audience ( Alcami et al . , 2021; Hedley et al . , 2017; King and McGregor , 2016; Searcy and Beecher , 2009; Trillo and Vehrencamp , 2005 ) . Even birds with only a single song type , such as Bengalese finches , vary parameters of their song depending on social context , including the specific identity of the listener ( Chen et al . , 2016; Heinig et al . , 2014; Sakata et al . , 2008 ) . The ability to contextually control vocalizations is also relevant for the customization of vocal signatures for purposes of individual and group recognition ( Vignal et al . , 2004 ) and to avoid overlap and enhance communication during vocal turn-taking and in response to environmental noises ( Benichov and Vallentin , 2020; Brumm and Zollinger , 2013 ) . Such capacities for vocal control likely reflect evolutionary advantages of incorporating sensory and contextual information about conspecifics and the environment in generating increasingly sophisticated vocal signaling . Our results indicate a latent capacity to integrate arbitrary sensory signals into the adaptive deployment of vocalizations in songbirds and suggest that some of the contextual control observed in natural settings may likewise rely on learned associations and other cognitive factors . Perhaps evolutionary pressures to develop nuanced social communication led to the elaboration of cortical ( pallial ) control over brainstem vocal circuitry ( Hage and Nieder , 2016 ) , and thereby established a conduit that facilitated the integration of progressively more abstract cues and internal states in that control . The ability of birds to switch between distinct motor programs using visual cues is reminiscent of contextual speech and motor control studies in humans . For example , human subjects in both laboratory studies and natural settings can learn multiple ‘states’ of vocal motor adaptation and rapidly switch between them using contextual information ( Houde and Jordan , 2002; Keough and Jones , 2011; Rochet-Capellan and Ostry , 2011 ) . Similarly , subjects can learn two separate states of motor adaptation for other motor skills , such as reaching , and switch between them using cues or other cognitive strategies ( Cunningham and Welch , 1994 ) . Models of such context-dependent motor adaptation frequently assume at least two parallel processes ( Abrahamse et al . , 2013; Ashe et al . , 2006; Green and Abutalebi , 2013; Hikosaka et al . , 1999; Lee and Schweighofer , 2009; McDougle et al . , 2016; Rochet-Capellan and Ostry , 2011; Wolpert et al . , 2011 ) , one that is more flexible , and sensitive to contextual information ( McDougle et al . , 2016 ) , and a second that cannot readily be associated with contextual cues and is only gradually updated during motor adaptation ( Howard et al . , 2013 ) . Specifically , in support of such a two-process model , Imamizu and Kawato , 2009 and Imamizu et al . , 2007 found that contextual information can drive rapid shifts in adaptation at the beginning of new blocks , without affecting the rate of adaptation within blocks . The similar separation in our study between rapid context-dependent shifts in sequence probability at the onset of blocks , and gradual adaptation within blocks that does not improve with training ( Figure 2G–L ) , suggests that such contextual sequence learning in the Bengalese finch may also be enabled by two distinct processes . Humans studies of two-process models suggest that slow adaptation occurs primarily within primary motor structures , while fast context-dependent state switches , including for cued switching between languages in bilinguals , engage more frontal areas involved in executive control ( Bialystok , 2017; Blanco-Elorrieta and Pylkkänen , 2016; De Baene et al . , 2015; Imamizu and Kawato , 2009 ) . In songbirds , the gradual adaptation of sequence probabilities within blocks might likewise be controlled by motor and premotor song control structures , while visual contextual cues could be processed in avian structures analogous to mammalian prefrontal cortex , outside the song system . For example , the association area nidopallium caudolaterale ( Güntürkün , 2005 ) , is activated by arbitrary visual cues that encode learned rules ( Veit and Nieder , 2013; Veit et al . , 2015 ) , and this or other avian association areas ( Jarvis et al . , 2013 ) may serve as an intermediate representation of the arbitrary contextual cues that can drive rapid learned shifts in syllable sequencing . At the level of song motor control , our results indicate a greater capacity for rapid and flexible adjustment of syllable transition probabilities than previously appreciated . Current models of song production include networks of neurons in the vocal premotor nucleus HVC responsible for the temporal control of individual syllables , which are linked together by activity in a recurrent loop through brainstem vocal centers ( Andalman et al . , 2011; Ashmore et al . , 2005; Cohen et al . , 2020; Hamaguchi et al . , 2016 ) . At branch points in songs with variable syllable sequencing , one influential model posits that which syllable follows a branch point is determined by stochastic processes that depend on the strength of the connections between alternative syllable production networks , and thus dynamics local to HVC ( Jin , 2009; Jin and Kozhevnikov , 2011; Troyer et al . , 2017; Zhang et al . , 2017 ) . Such models could account for a gradual adjustment of sequence probabilities over a period of hours or days ( Lipkind et al . , 2013; Warren et al . , 2012 ) through plasticity of motor control parameters , such as the strength of synaptic connections within HVC . However , our results demonstrate that there is not a single set of relatively fixed transition probabilities that undergo gradual adjustments , as could be captured in synaptic connectivity of branched syllable control networks . Rather , the song system has the capacity to maintain distinct representations of transition probabilities and can immediately switch between those in response to visual cues . HVC receives a variety of inputs that potentially could convey such visual or cognitive influences on sequencing ( Bischof and Engelage , 1985; Cynx , 1990; Seki et al . , 2008; Ullrich et al . , 2016; Wild , 1994 ) , and one of these inputs , Nif , has previously been shown to be relevant for sequencing ( Hosino and Okanoya , 2000; Vyssotski et al . , 2016 ) . It therefore is likely that the control of syllable sequence in Bengalese finches involves a mix of processes local to nuclei of the song motor pathway ( Basista et al . , 2014; Zhang et al . , 2017 ) as well as inputs that convey a variety of sensory feedback and contextual information . The well-understood circuitry of the avian song system makes this an attractive model to investigate how such top-down pathways orchestrate the kind of contextual control of vocalizations demonstrated in this study , and more broadly to uncover how differing cognitive demands can flexibly and adaptively reconfigure motor output .
The experiments were carried out on eight adult male Bengalese finches ( Lonchura striata ) obtained from the lab’s breeding colony ( age range 128–320 days post-hatch , median 178 days , at start of experiment ) . Birds were placed in individual sound-attenuating boxes with continuous monitoring and auditory recording of song . Song was recorded using an omnidirectional microphone above the cage . We used custom software for the online recognition of target syllables and real-time delivery of short 40 ms bursts of WN depending on the syllable sequence ( Tumer and Brainard , 2007; Warren et al . , 2012 ) . This LabView program , EvTAF , is included as an executable file with this submission , and further support is available from the corresponding authors upon request . All procedures were performed in accordance with animal care protocols approved by the University of California , San Francisco Institutional Animal Care and Use Committee ( IACUC ) . Bengalese finch song consists of a discrete number of vocal elements , called syllables , that are separated by periods of silence . At the start of each experiment , a template was generated to recognize a specific sequence of syllables ( the target sequence ) for each bird based on their unique spectral structure . In the context-dependent auditory feedback protocol , the target sequence that received aversive WN feedback switched between blocks of different light contexts . Colored LEDs ( superbrightleds . com , St . Louis , MO; green 520 nm , amber 600 nm ) produced two visually distinct environments ( green and yellow ) to serve as contextual cues to indicate which sequences would elicit WN and which would ‘escape’ ( i . e . not trigger WN ) . We wanted to test whether the birds would be able to associate song changes with any arbitrary visual stimulus; therefore , there was no reason to choose these specific colors , and the birds’ color perception in this range should not matter , as long as they were able to discriminate the colors . The entire day was used for data acquisition by alternating the two possible light contexts . We determined sensitivity and specificity of the template to the target sequence on a randomly selected set of 20 song bouts on which labels and delivery of WN was hand-checked . Template sensitivity was defined as follows: sensitivity = ( number of correct hits ) / ( total number of target sequences ) . The average template sensitivity across experiments was 91 . 3% ( range 75 . 2–100% ) . Template specificity was defined as: specificity = ( number of correct escapes ) / ( number of correct escapes plus number of false alarms ) , where correct escapes were defined as the number of target sequences of the currently inactive context that were not hit by WN , and false alarms were defined as any WN that was delivered either on the target sequence of the currently inactive context , or anywhere else in song . The average template specificity was 96 . 7% ( range 90 . 6–100% ) . At the start of each experiment , before WN training , songs were recorded during a baseline period in which cage illumination was switched between colors at random intervals . Songs from this baseline period were separately analyzed for each light color to confirm that there was no systematic , unlearned effect of light cues on sequencing before training . During initial training , cage illumination was alternatingly switched between colors at random intervals . Intervals were drawn from uniform distributions which differed between birds ( 60–150 min [four birds] , 10–30 min [two birds] , 60–240 min [one bird] , 30–150 min [one bird] ) . Different training schedules were assigned to birds arbitrarily and were not related to a bird’s performance . After an extended period of training ( average 33 days , range 12–79 days ) , probe blocks without WN were included , to test whether sequencing changes could be elicited by visual cues alone . During this period , probe blocks were interspersed with WN training blocks . Probe blocks made up approximately one third of total blocks ( 10 of 34 blocks in the sequence ) and 7–35% of total time , depending on the bird . The duration of probe blocks was typically shorter or equal to the duration of WN blocks ( 10–30 min for six birds , 30–120 min for one bird , 18–46 min for one bird ) . The total duration of the experiment , consisting of baseline , training , and probe periods , was on average 52 days . During this period , birds sang 226 ( range 66–356 ) bouts per day during baseline days and 258 ( range 171–368 ) bouts per day during the period of probe collection at the end of training ( 14% increase ) . The average duration of song bouts also changed little , with both the average number of target sequences per bout ( 8 . 7 during baseline , 7 . 7 during probes , 7% decrease ) and the average number of syllables per bout ( 74 during baseline , 71 during probes , 2% decrease ) decreasing slightly . In addition to the eight birds that completed this training paradigm , three birds were started on contextual training but never progressed to testing with probe blocks , because they did not exhibit single-context learning ( n = 1 ) ; because of technical issues with consistent targeting at branch points , ( n = 1 ) ; or because they lost sequence variability during initial stages of training ( n = 1 ) ; these birds are excluded from the results . Of the eight birds that completed training , three birds exhibited relatively small context-dependent changes in sequencing ( Figure 1H ) . We examined several variables to assess whether they could account for differences in the magnitude of learning across birds , including the bird’s age , overall transition entropy of the song ( Katahira et al . , 2013 ) , transition entropy at the targeted branch points ( Warren et al . , 2012 ) , as well as the distance between the WN target and the closest preceding branch point in the sequence . None of these variables were significantly correlated with the degree of contextual learning that birds expressed ( Figure 4—figure supplement 1 ) , and consequently , all birds were treated as a single group in analysis and reporting of results . In a subset of experiments ( n = 3 ) , after completing measurements with probe blocks , we added a third , neutral context ( Figure 5 ) , signaled by white light , in which there was no WN reinforcement . Syllable annotation for data analysis was performed offline . Each continuous period of singing that was separated from others by at least 2 s of silence was treated as an individual ‘song’ or ‘song bout’ . Song was bandpass filtered between 500 Hz and 10 , 000 Hz and segmented into syllables and gaps based on amplitude threshold and timing parameters determined manually for each bird . A small sample of songs ( approximately 20 song bouts ) was then annotated manually based on visual inspection of spectrograms . These data were used to train an offline autolabeler ( ‘hybrid-vocal-classifier’ , Nicholson , 2021 ) , which was then used to label the remaining song bouts . Autolabeled songs were processed further in a semi-automated way depending on each bird’s unique song , for example to separate or merge syllables that were not segmented correctly ( detected by their duration distributions ) , to deal with WN covering syllables ( detected by its amplitude ) , and to correct autolabeling errors detected based on the syllable sequence . A subset of songs was inspected manually for each bird to confirm correct labeling . Sequence probability was first calculated within each song bout as the frequency of the yellow target sequence relative to the total number of yellow and green target sequences: p=n ( target_Y ) ntarget_Y+n ( target_G ) . Note that this differs from transition probabilities at branch points in song in that it ignores possible additional syllable transitions at the branch point , and does not require the targeted sequences to be directly following the same branch point . For example for the experiment in Figure 3 , the target sequences were ‘n-ab’ and ‘f-ab’ , so the syllable covered by WN ( ‘b’ in both contexts ) was two to three syllables removed from the respective branch point in the syllable sequence ( ‘n-f’ vs . ‘n-a’ or ‘f-n’ vs . ‘f-a’ ) . Note also that units of sequence probability are in percent; therefore , reported changes in percentages ( e . g . Figures 1H and 2E , F ) describe absolute changes in sequence probability , which reflect the proportion of each target sequence , not percent changes . Song bouts that did not contain either of the two target sequences were discarded . In the plots of sequence probability over several days in Figure 1A–C , we calculated sequence probability for all bouts on a given day ( average n = 1854 renditions of both target sequences per day ) . We estimated 95% confidence intervals by approximation with a normal distribution as p±z*p* ( 1-p ) n with n=ntarget_Y+ntarget_G and z = 1 . 96 . Context switches were processed to include only switches between adjacent blocks during the same day , that is excluding overnight switches and treating blocks as separate contexts if one day started with the same color that had been the last color on the previous day . If a bird did not produce any song during one block , this block was merged with any neighboring block of the same color ( e . g . green probe without songs before green WN , where the context switch would not be noticeable for the bird ) . If the light color switched twice ( or more ) without any song bouts , those context switches were discarded . In order to reduce variability associated with changes across individual song bouts , shift magnitude was calculated as the difference between the first five song bouts in the new context and the last five song bouts in the old context . Only context switches with at least three song bouts in each adjacent block were included in analyses of shift magnitude . In plots showing songs aligned to context switches , the x-axis is limited to show only points for which at least half of the blocks contributed data ( i . e . in Figure 2D , half of the green probe blocks contained at least six songs ) . All statistical tests were performed with MATLAB . We used non-parametric tests to compare changes across birds ( Wilcoxon rank-sum test for unpaired data , Wilcoxon signed-rank test for paired data ) , because with only eight birds/data points , it is more conservative to assume that data are not Gaussian distributed . In order to investigate how context-dependent performance developed over training ( Figure 2G–L ) , we quantified changes to sequence probabilities across block switches for five birds for which we had a continuous record from the onset of training . Sequence probability curves ( e . g . Figure 2H ) for yellow switches were inverted so that both yellow and green switches were plotted in the same direction , aligned by the time of context switches , and were cut off at a time point relative to context switches where fewer than five switches contributed data . We then subtracted the mean pre-switch value from each sequence probability curve . For visual display of the example bird , sequence probability curves were smoothed with a nine bout boxcar window and displayed in bins of seven context switches . To calculate the slope of slopes and slope of intercepts ( Figure 2L ) , we calculated a linear fit to the post-switch parts of the unsmoothed sequence probability curve for each individual context switch . To calculate the specificity of the context difference to the targeted branch points in song , we generated transition diagrams for each bird . To simplify the diagrams , introductory notes were summarized into a single introductory state . Introductory notes were defined for each bird as up to three syllables occurring at the start of song bouts before the main motif , which tended to be quieter , more variable , with high probabilities to repeat and to transition to other introductory notes . Repeat phrases were also summarized into a single state . Motifs , or chunks , in the song with fixed order of syllables were identified by the stereotyped transitions and short gap durations between syllables in the motif ( Isola et al . , 2020; Suge and Okanoya , 2010 ) and were also summarized as a single state in the diagram . Sometimes , the same syllable can be part of several fixed chunks ( Katahira et al . , 2013 ) , in which case it may appear several times in the transition diagram . We then calculated the difference between the transition matrices for the two probe contexts at each transition that was a branch point ( defined as more than 3% and less than 97% transition probability ) . These context differences were split into ‘targeted branch points’ , i . e . , the branch point or branch points most closely preceding the target sequences in the two contexts , and ‘non-targeted branch points’ , i . e . , all other branch points in the song . We calculated the proportion of absolute contextual difference in the transition matrix that fell to the targeted branch points , for example for the matrix in Figure 4C ( 44 + 45 ) / ( 44 + 45 + 6+6 + 1+1 + 2+2 ) =83 . 2% . Typically , birds with clear contextual differences at the target sequence also had high specificity of sequence changes to the targeted branch points . To calculate the transition entropy of baseline song , we again summarized introductory notes into a single introductory state . In addition , the same syllables as part of multiple fixed motifs , or in multiple positions within the same fixed motif , were renamed as different syllables , so as not to count as sequence variability what was really a stereotyped sequence ( i . e . a-b 50% and b-c 50% in the fixed sequence ‘abbc’ ) . Transition entropy was then calculated as in Katahira et al . , 2013: with x denoting the preceding syllable and y denoting the current syllable , over all syllables in the song . | Human speech and birdsong share numerous parallels . Both humans and birds learn their vocalizations during critical phases early in life , and both learn by imitating adults . Moreover , both humans and songbirds possess specific circuits in the brain that connect the forebrain to midbrain vocal centers . Humans can flexibly control what they say and how by reordering a fixed set of syllables into endless combinations , an ability critical to human speech and language . Birdsongs also vary depending on their context , and melodies to seduce a mate will be different from aggressive songs to warn other males to stay away . However , so far it was unclear whether songbirds are also capable of modifying songs independent of social or other naturally relevant contexts . To test whether birds can control their songs in a purposeful way , Veit et al . trained adult male Bengalese finches to change the sequence of their songs in response to random colored lights that had no natural meaning to the birds . A specific computer program was used to detect different variations on a theme that the bird naturally produced ( for example , “ab-c” versus “ab-d” ) , and rewarded birds for singing one sequence when the light was yellow , and the other when it was green . Gradually , the finches learned to modify their songs and were able to switch between the appropriate sequences as soon as the light cues changed . This ability persisted for days , even without any further training . This suggests that songbirds can learn to flexibly and purposefully modify the way in which they sequence the notes in their songs , in a manner that parallels how humans control syllable sequencing in speech . Moreover , birds can learn to do this ‘on command’ in response to an arbitrarily chosen signal , even if it is not something that would impact their song in nature . Songbirds are an important model to study brain circuits involved in vocal learning . They are one of the few animals that , like humans , learn their vocalizations by imitating conspecifics . The finding that they can also flexibly control vocalizations may help shed light on the interactions between cognitive processing and sophisticated vocal learning abilities . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"neuroscience"
] | 2021 | Songbirds can learn flexible contextual control over syllable sequencing |
SNX6 is a ubiquitously expressed PX-BAR protein that plays important roles in retromer-mediated retrograde vesicular transport from endosomes . Here we report that CNS-specific Snx6 knockout mice exhibit deficits in spatial learning and memory , accompanied with loss of spines from distal dendrites of hippocampal CA1 pyramidal cells . SNX6 interacts with Homer1b/c , a postsynaptic scaffold protein crucial for the synaptic distribution of other postsynaptic density ( PSD ) proteins and structural integrity of dendritic spines . We show that SNX6 functions independently of retromer to regulate distribution of Homer1b/c in the dendritic shaft . We also find that Homer1b/c translocates from shaft to spines by protein diffusion , which does not require SNX6 . Ablation of SNX6 causes reduced distribution of Homer1b/c in distal dendrites , decrease in surface levels of AMPAR and impaired AMPAR-mediated synaptic transmission . These findings reveal a physiological role of SNX6 in CNS excitatory neurons .
SNX6 is a member of the sorting nexin ( SNX ) family that plays important roles in retromer-mediated , dynein−dynactin-driven retrograde vesicular transport from endosomes to the trans-Golgi network ( TGN ) . The retromer complex functions in endosomal protein sorting and trafficking . It is composed of the VPS26-VPS29-VPS35 core complex and a SNX subunit or subcomplex ( Gallon and Cullen , 2015 ) . In mammalian epithelial cells , SNX6 serves as dynein adaptor in retromer-mediated vesicular transport to regulate both cargo recognition and release via its interaction with the motor and the target membrane . SNX6 contains an amino-terminal Phox Homology ( PX ) domain that is evolutionarily conserved among SNXs and a carboxyl-terminal Bin/Amphiphysin/Rvs ( BAR ) domain that allows for dimerization with BAR domains of other proteins . It dimerizes with the SNX1 subunit of retromer through its BAR domain and binds to dynactin p150Glued through its PX domain , linking the dynein−dynactin motor complex to retromer-associated vesicular cargoes ( Hong et al . , 2009; Wassmer et al . , 2009 ) . Its PX domain also interacts with the TGN-enriched phospholipid PtdIns ( 4 ) P , which inhibits the interaction between SNX6 and p150Glued to facilitate dissociation of the retrograde motor from the retromer-associated cargo at the TGN ( Niu et al . , 2013 ) . Although retromer is involved in endosomal sorting and trafficking of amyloid precursor protein ( APP ) ( Fjorback et al . , 2012; Sullivan et al . , 2011 ) , and transport , surface expression and endocytic recycling of AMPAR ( Choy et al . , 2014; Munsie et al . , 2015; Zhang et al . , 2012 ) in neurons , the biological function of SNX6 in the CNS , whether retromer-dependent or not , remains to be explored . In most of the principal neurons in the central nervous system ( CNS ) , dendritic spines , the micron-sized membrane protrusions covering dendritic shaft , provide major sites of excitatory inputs . They are highly specialized postsynaptic structures containing transmembrane neurotransmitter receptors and proteins with signaling and scaffolding functions . Among them , scaffold proteins of the postsynaptic density ( PSD ) play crucial roles in glutamatergic neurotransmission by organizing glutamate receptors and signaling molecules at the postsynaptic terminal . One group of PSD scaffold proteins is the PSD95 membrane-associated guanylyl kinase ( MAGUK ) family proteins that anchor glutamate receptors to the PSD ( Elias and Nicoll , 2007 ) . Another group is the Homer and Shank family proteins . They interact with each other and form a high-order complex with a mesh-like network structure , which is believed to serve as a structural platform of the PSD essential for the structural integrity of dendritic spines ( Hayashi et al . , 2009 ) . The Homer family proteins also regulate trafficking and signaling of the group one metabotropic glutamate receptors ( mGluR1/5 ) and synaptic plasticity ( Ango et al . , 2001 , 2002; Gerstein et al . , 2012; Mao et al . , 2005; Roche et al . , 1999; Ronesi and Huber , 2008 ) . Moreover , Homer1b and 1c , the long isoforms encoded by the Homer1 gene that are differentiated by an insertion of 12 amino acid ( aa ) residues at aa 177 in Homer1c ( Xiao et al . , 1998 ) , regulate surface expression of α-amino-3-hydroxy-5-methyl-4-isoxazolepropionic acid receptors ( AMPAR ) at synaptic sites through endocytic recycling ( Lu et al . , 2007 ) . Although mechanisms underlying glutamate receptor trafficking to dendrites as well as their local trafficking into and out of synaptic sites have been intensively studied ( Anggono and Huganir , 2012; Hoerndli et al . , 2013; Horak et al . , 2014; Huganir and Nicoll , 2013; Ladépêche et al . , 2014; Setou et al . , 2000 , 2002 ) , the molecular basis for dendritic distribution and spine localization of most PSD scaffolding proteins including Homer remains largely unexplored . In this study , we investigated the physiological function ( s ) of SNX6 in mouse CNS neurons using multiple approaches including mouse genetics , behavior assays and electrophysiology , biochemistry and fluorescence imaging . Ablation of SNX6 in the CNS causes deficits in spatial learning and memory , decrease in spine density of the distal dendrites of hippocampal CA1 neurons and impairment of their AMPAR-mediated synaptic transmission , suggesting a role for SNX6 in synaptic structure and function . SNX6 interacts with Homer1b/c and loss of SNX6 leads to a reduction in its distribution in distal dendrites . Intriguingly , although SNX6 is required for the motility of a subpopulation of Homer1c on vesicles in dendritic shaft , live imaging and FRAP analyses indicate that Homer1c enters dendritic spines via protein diffusion but not SNX6-dependent active transport . Overexpression of SNX6 or Homer1c restores the spine density and AMPAR surface levels of Snx6-/- neurons . These findings uncover a physiological function for SNX6 in hippocampal CA1 excitatory neurons .
Immunoblotting analysis of tissue lysates indicated that SNX6 was ubiquitously expressed in mouse ( Figure 1—figure supplement 1A ) . In mouse brain , SNX6 was expressed in both the somatodendritic area and processes of neurons in the cortex and the hippocampal formation ( Figure 1—figure supplement 1B–E ) . Co-immunostaining with antibodies to SNX6 and axonal or dendritic markers revealed its distribution in a punctate pattern in both axon and dendrites of hippocampal neurons ( Figure 1—figure supplement 1F ) . Quantitative analysis of fluorescence signal intensity revealed that SNX6 was primarily located in dendrites ( Figure 1—figure supplement 1G ) . Moreover , SNX6 partially colocalized with PSD95 , a postsynaptic marker in dendritic spines but not synaptophysin , a presynaptic marker ( Figure 1—figure supplement 1H ) . To investigate physiological function ( s ) of SNX6 in the CNS , we generated a conditional allele by floxing exon 5 of the Snx6 gene ( Figure 1A–C ) , and obtained CNS-specific knockout mice ( CNS-Snx6 KO ) by mating Snx6 conditional KO mice ( Snx6fl/fl ) to Nestin promoter-driven Cre recombinase transgenic mice ( Nestin-Cre ) . Lack of SNX6 protein expression in the CNS of Nestin-Cre; Snx6fl/fl mice was confirmed by immunoblotting analysis of mouse brain homogenates ( Figure 1D ) . The CNS neurons of Snx6fl/fl and Nestin-Cre; Snx6fl/fl mice were hence referred to as Snx6+/+ and Snx6-/- neurons , respectively . CNS-Snx6 KO mice were born with the expected Mendelian ratio and appeared indistinguishable from wild-type littermates . Their brain size was comparable with that of wild-type ( Figure 1E ) , and no gross abnormalities in the structure of the cortex , hippocampus and cerebellum were observed by histological examination ( Figure 1F ) . 10 . 7554/eLife . 20991 . 003Figure 1 . Generation and characterization of Snx6 CNS-specific knockout mice . ( A ) Domain structure of SNX6 . ( B ) Schematic diagram of the Snx6 gene locus , the targeting vector , and the mutant alleles after homologous recombination . FRTtF/FRTtR and loxtF/loxtR: primer pairs used for genotyping . The XhoI and HpaI probes used for Southern blotting analysis are shown . Neo: the neomycin resistance cassette . ( C ) Southern blotting analysis of wild-type ( WT ) and two independent clones of targeted ES cells ( 9 hr and 5D ) . ( D ) Immunoblots of tissue lysates from mouse littermates , probed with antibodies to SNX6 . ( E ) Comparison of brain weight of Snx6fl/fl ( 15 ) and Nestin-Cre; Snx6fl/fl mice ( 12 ) . Data represent mean ± SEM for each group . ( F ) Nissl staining of sagittal sections of whole brain from Snx6fl/fl and Nestin-Cre; Snx6fl/fl mice . Also shown are magnification of the cerebellum ( middle panel ) and the hippocampus/cortex area ( right panel ) . Scale bar: 1 mm . DOI: http://dx . doi . org/10 . 7554/eLife . 20991 . 00310 . 7554/eLife . 20991 . 004Figure 1—figure supplement 1 . Expression and subcellular distribution of SNX6 in the CNS . ( A ) Western blotting of wild-type mouse tissue lysates . ( B ) A schematic showing the relative positions of coronal sections in ( D ) and ( E ) in sagittal view . ( C–E ) Immunohistochemical analysis of SNX6 expression in mouse brain . Coronal sections of wild-type mouse were fixed and stained with antibodies to SNX6 and counterstained with hematoxylin . ( F ) Mouse hippocampal neurons were cultured in vitro for 18 days , fixed and immunostained with antibodies to SNX6 and Tau1 or MAP2 . Shown are representative confocal microscopy images . ( G ) Background-subtracted mean intensity of SNX6 fluorescence in primary axon and dendrites . Measurement of fluorescence intensity is expressed in arbitrary units per square area in both axons and dendrites . All images ( 1024 × 1024 pixels , 16 bit ) were obtained in the same settings ( mean ± SEM , n = 15 ) . ( H ) Mouse hippocampal neurons were cultured in vitro for 18 days , fixed and immunostained with antibodies to SNX6 and synaptophysin ( SYP ) or PSD95 . Shown are representative confocal microscopy images . OB , olfactory bulb . Fi , fimbria . LD , lateral dorsal nucleum of thalamus . Sm , stria medullaris . MH , medial habenula . LH , lateral habenula . Bars: 100 μm in ( C ) , ( D ) , ( E ) and ( D-1 ) , 20 μm in ( C-1 – 4 ) , ( D-2 – 3 ) and ( E-1 – 8 ) , 2 μm in ( F ) and ( H ) . DOI: http://dx . doi . org/10 . 7554/eLife . 20991 . 004 Next we conducted behavioral analyses on Nestin-Cre; Snx6fl/fl mice and their wild-type littermates . No change in locomotor activity was detected by rotarod and open field assays ( Figure 2A , B ) , and the mood levels of CNS-Snx6 KO were also similar to that of wild-type mice in elevated plus maze , tail suspension and forced swimming tests ( Figure 2C–E ) . In the Three-Chamber test , the CNS-Snx6 KO mice showed no abnormality in sociability and social novelty ( Figure 2F ) , nor did they display repetitive behaviors ( Figure 2G ) . We then focused on their performance in learning and memory . Although Nestin-Cre; Snx6fl/fl mice performed as well as their littermates in Y maze and shuttle box ( Figure 2H , I ) , in the Morris water maze test , they were significantly retarded in spatial learning using latency traveled to reach the hidden platform as measures ( Figure 2J ) . A probe trial showed that they were also severely impaired in spatial memory ( Figure 2K ) . Moreover , these mice exhibited deficits in memory recall ( Figure 2L , M ) . As the hippocampal region participates in the processes of the encoding , storage , consolidation and retrieval of spatial memory ( Riedel et al . , 1999 ) , the behavioral phenotypes suggest that ablation of SNX6 affects synaptic function of hippocampal neurons . 10 . 7554/eLife . 20991 . 005Figure 2 . Impaired spatial learning and memory in Nestin-Cre; Snx6fl/fl mice . ( A–I ) No effects of SNX6 ablation on the performance in assays of rotarod ( A ) ( 13 Snx6fl/fl and 16 Nestin-Cre; Snx6fl/fl mice ) , open field ( B ) ( 23 Snx6fl/fl and 23 Nestin-Cre; Snx6fl/fl mice ) , elevated plus maze ( C ) ( 14 Snx6fl/fl and 13 Nestin-Cre; Snx6fl/fl mice ) , tail suspension ( D ) ( 14 Snx6fl/fl and 24 Nestin-Cre; Snx6fl/fl mice ) , forced swimming ( E ) ( 15 Snx6fl/fl and 25 Nestin-Cre; Snx6fl/fl mice ) , Three-Chamber test ( F ) ( 10 Snx6fl/fl and 9 Nestin-Cre; Snx6fl/fl mice ) , repetitive behaviors ( G ) ( 12 Snx6fl/fl and 10 Nestin-Cre; Snx6fl/fl mice ) , Y maze ( H ) ( 11 Snx6fl/fl and 15 Nestin-Cre; Snx6fl/fl mice ) and shuttle box ( I ) ( 20 Snx6fl/fl and 13 Nestin-Cre; Snx6fl/fl mice ) . The data represent mean ± SEM for each group . ( J–K ) Increased escape latency at acquisition learning ( J ) ( data represent mean ± SEM of four trials per day ) , decreased number of crossing and increased latency to first enter the 1 . 5x area at probe test ( K ) ( the data represent mean ± SEM for each group ) in Nestin-Cre; Snx6fl/fl mice in the Morris water maze . Subject numbers were 18 Snx6fl/fl and 22 Nestin-Cre; Snx6fl/fl mice . ( L ) After a 20-day rest , both Snx6fl/fl and Nestin-Cre; Snx6fl/fl mice exhibited memory extinguishment . ( M ) Decreased number of crossing and increased latency to first enter the 1 . 5x area at probe test in Nestin-Cre; Snx6fl/fl mice after one recall training . The data represent mean ± SEM . N = 3 independent experiments . DOI: http://dx . doi . org/10 . 7554/eLife . 20991 . 005 To investigate changes in synaptic function caused by SNX6 ablation at the cellular level , we examined neuronal morphology in the hippocampal region by crossing Snx6fl/fl and Nestin-Cre; Snx6fl/fl mice with Thy1-EGFP transgenic mice and analyzing brain sections by confocal microscopy ( Figure 3A ) . We focused on the morphology of CA1 and CA3 pyramidal cells for two reasons: first , neurons in the CA1 and CA3 region were sparsely labeled by EGFP and hence easily distinguishable from neighboring ones for the purpose of morphological assessment; second , changes in the morphology and density of dendritic spines have been linked to synaptic function and plasticity . For quantification of spine number and morphology , we imaged segments of dendrites that are easily distinguishable from those of neighboring neurons , i . e . , the oriens/distal branches of the basal and radiatum/thin branches of the apical dendrites of CA1 neurons , and secondary/tertiary branches of the basal and apical dendrites of CA3 neurons in stratum oriens and stratum radiatum , respectively ( Figure 3B ) . Quantitative analysis showed that , although spine morphology did not change in either CA1 or CA3 pyramidal cells ( Figure 3C–F ) , there was a decrease in the spine density of the distal portion of apical dendrites of Snx6-/- CA1 neurons ( Figure 3C , D ) . In contrast , no change in spine density was detected in Snx6-/- CA3 neurons ( Figure 3E , F ) . Consistently , ultrastructural analysis revealed a decrease in the number of asymmetric/excitatory synapses in the CA1 , but not in the CA3 region of Nestin-Cre; Snx6fl/fl mouse brain ( Figure 3G , H ) . Together , these data indicate that SNX6 is required for spine morphogenesis and/or maintenance of distal apical dendrites of CA1 pyramidal neurons . 10 . 7554/eLife . 20991 . 006Figure 3 . Decreases in spine density of hippocampal CA1 apical dendrites and number of excitatory synapses in the CA1 region in Nestin-Cre; Snx6fl/fl Mice . ( A ) Confocal images of coronal sections of hippocampi from Snx6fl/fl; Thy1-GFP and Nestin-Cre; Snx6fl/fl; Thy1-GFP . ( B ) Schematic of the location of the dendritic segments selected for morphological analysis . ( C ) Representative 3D-reconstructed confocal images of dendrites of CA1 pyramidal cells . The z-dimension position is color-coded according to the color scale bar . ( D ) Quantification of spine density ( n = 5 pairs of mice , apical/basal: 43/40 cells , 95/82 dendritic segments and 4979/3721 spines for Snx6fl/fl; Thy1-GFP; 40/38 cells , 96/78 dendritic segments and 3662/3341 spines for Nestin-Cre; Snx6fl/fl; Thy1-GFP ) and morphology ( n = 2 pairs , apical/basal: 16/14 cells , 1053/764 spines for Snx6fl/fl; Thy1-GFP; 18/17 cells , 1099/894 spines for Nestin-Cre; Snx6fl/fl; Thy1-GFP ) of CA1 dendrites . ( E ) Representative 3D-reconstructed confocal images of CA3 dendrites . ( F ) Quantification of spine density ( n = 5 pairs , apical/basal: 34/34 cells , 80/75 dendritic segments and 3155/2261 spines for Snx6fl/fl; Thy1-GFP; 34/34 cells , 77/73 dendritic segments and 3053/2216 spines for Nestin-Cre; Snx6fl/fl; Thy1-GFP ) and morphology ( n = 2 pairs , apical/basal: 13/12 cells , 822/595 spines for Snx6fl/fl; Thy1-GFP; 11/14 cells , 739/568 spines for Nestin-Cre; Snx6fl/fl; Thy1-GFP ) of CA3 dendrites . ( G ) Representative TEM images of hippocampal CA1 regions of adult animals . Yellow solid arrowheads indicate asymmetric ( excitatory ) synapses . Insets are representative higher magnification images of synapses in the boxed areas . Yellow empty arrowheads indicate mitochondria . Yellow arrows indicate lysosomes . ( H ) Quantification of synapse density ( n = 3 pairs , CA1: 1553 synapses for Snx6fl/fl and 1038 synapses for Nestin-Cre; Snx6fl/fl . CA3: 1102 synapses for Snx6fl/fl and 1069 synapses for Nestin-Cre; Snx6fl/fl ) . Data represent mean ± SEM . Bars: 200 μm in ( A ) , 2 μm in ( C ) and 500 nm in ( G ) . DOI: http://dx . doi . org/10 . 7554/eLife . 20991 . 006 That ablation of SNX6 causes a decrease in spine density of distal dendrites suggests that it functions in the formation/stabilization of dendritic spines , probably via regulating dendritic distribution of postsynaptic proteins such as PSD components and/or neurotransmitter receptors . As the first step to investigate its molecular function , we determined the subcellular distribution of SNX6 in dendrites by co-immunostaining of SNX6 and vesicular markers in cultured mature hippocampal neurons . Confocal microscopy revealed that the majority of SNX6 signals colocalized with EEA1 and Rab5B , the early endosome markers ( Figure 4A , B ) . SNX6 also partially colocalized with the late endosome marker Rab7 and Rab4 , marker for the fast recycling pathway , though to a lesser extent , but not Rab11 recycling endosomes ( Figure 4A , B ) . Intriguingly , although SNX6 signals did not colocalize with Golgin97 , a TGN resident protein , they overlapped partially with TGN46 ( Figure 4A , B ) , a protein involved in membrane traffic to and from the TGN ( Ponnambalam et al . , 1996 ) , suggesting that SNX6 associates with endosomes and transport carriers in the dendrite . 10 . 7554/eLife . 20991 . 007Figure 4 . SNX6 interacts with Homer1b/c and colocalizes with Homer1b/c on endosomes . ( A ) Hippocampal neurons were transfected with pLL3 . 7 . 1 on DIV14 to express DsRed as volume marker , fixed on DIV17 and immunostained with antibodies to SNX6 and vesicular markers . DsRed is pseudocolored for presentation . White arrowheads indicate overlapped signals . ( B ) Quantification of colocalization in ( A ) from 45 dendritic segments of 15 neurons ( mean ± SEM , N = 3 . Total length of dendrites: 1568 μm for EEA1; 1447 μm for Rab5; 1637 μm for Rab4; 1489 μm for Rab7; 1319 μm for Rab11; 1207 μm for Golgi97 and 1462 μm for TGN46 ) . ( C ) Mouse brain lysates were incubated with His-SNX1-N or His-SNX6-N immobilized on Ni-NTA agarose . Bound proteins were subjected to SDS-PAGE and mass spectrometry analysis . The table shows the number of Homer1b/c unique peptides identified by mass spec analysis and their sequence coverage . ( D ) Schematic representation of the domain structure of Homer1 isoforms and Homer1c fragments used in this study . ( E ) Upper panels: immunoblotting of bound proteins in ( C ) . Lower panel: coomassie brilliant blue ( CBB ) stained SDS-PAGE gel shows purified recombinant proteins . ( F ) Mapping of SNX6-Homer1b/c interaction sites by in vitro binding assay . ( G ) In vitro binding assay of SNX6 and Homer family members . ( H ) Lysates from HEK293 cells overexpressing Flag-SNX6 and mEmerald-Homer1c were subjected to co-IP with Flag M2 beads , followed by immunoblotting with antibodies to Flag and Homer1b/c . ( I ) Total lysates and membrane fractions from mouse brain lysates were subjected to IP and immunoisolation with antibodies to Homer1b/c or SNX6 , and antibodies to SNX6 coupled to Dynabeads Protein G , respectively . Shown are immunoblots probed with antibodies to SNX6 , p150Glued , DIC , GluN1 , GluN2A , GluN2B , Homer1b/c and Homer1a . ( J ) DIV18 neurons were immunostained with antibodies to Homer1b/c and SNX6 . ( K ) Quantification of colocalization in ( J ) from 45 dendritic segments of 15 neurons ( mean ± SEM , N = 3 independent experiments . Total length 1677 μm ) . ( L ) DIV18 neurons were immunostained with antibodies to Homer1b/c and EEA1 . ( M ) Quantification of colocalization in ( L ) from 45 dendritic segments of 15 neurons ( mean ± SEM , N = 3 . Total length 1459 μm ) . ( N ) DIV18 neurons were immunostained with antibodies to EEA1 , Homer1b/c , and SNX6 . Superresolution images were captured by structured illumination microscopy ( SIM ) . White arrowheads indicate overlaps of signals from different channels . Bars: 2 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 20991 . 00710 . 7554/eLife . 20991 . 008Figure 4—figure supplement 1 . Colocalization analysis of superresolution images of triple-stained neurons captured by 3D-SIM ( 15 neurons for each immunostaining experiment ) . Unlike colocalization analysis of 2D confocal images of double-stained neurons in other figures , we adopted the methodology developed by Fletcher et al . to measure voxel-based colocalization between three fluorophores in the 3D images ( Fletcher et al . , 2010 ) . ( A ) Quantification result of colocalization among EEA1 , SNX6 and Homer1b/c in Figure 4N . ( B ) Quantification result of colocalization among p150Glued/DIC , SNX6 and Homer1b/c in Figure 6D . Overlaps of signals from three channels are shown as voxel colocalization values ( % ) . The statistical significance of colocalization values was evaluated by estimation of those occur by chance in randomized images generated by Monte Carlo Simulation as described in ( Fletcher et al . , 2010 ) and the results are shown in Table 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 20991 . 008 Next we attempted to identify SNX6-interacting protein ( s ) in dendrite . We performed pull down experiment from mouse brain lysates using a His-tagged SNX6 N-terminus ( aa 1–181 , encompassing the PX domain ) immobilized on Ni-NTA agarose . Mass spectrometry analysis revealed six matching peptides with 24% sequence coverage corresponding to Homer1b/c , a postsynaptic scaffold protein ( Figure 4C ) . Immunoblotting analysis verified that Homer1b/c , not Homer1a , the shorter isoform encoded by the Homer1 gene , was pulled down by the N-terminus of SNX6 but not SNX1 ( Figure 4D , E ) . Moreover , in vitro binding assays showed that SNX6-N interacted directly with the coiled-coil domain of Homer1b/c ( Figure 4F ) , which is not present in Homer1a . In contrast , neither Homer2b nor Homer3 , the longer isoforms of other Homer family members , interacted with SNX6-N ( Figure 4G ) . Further , mEmerald-Homer1c co-immunoprecipitated with Flag-tagged SNX6 in transiently transfected HEK293T cells ( Figure 4H ) . Consistently , reciprocal co-immunoprecipitations of endogenous proteins from mouse brain lysates verified that SNX6 and Homer1b/c interact with each other ( Figure 4I , left and center panels ) . Moreover , immunoisolation of SNX6-positive vesicles from membrane fractions of mouse brain lysates detected Homer1b/c together with p150Glued and dynein intermediate chain ( DIC ) , subunits of the dynein−dynactin complex ( Figure 4I , right panel ) . In contrast , neither Homer1a nor subunits of the N-methyl-D-aspartate receptor ( NMDAR ) were detected on SNX6-positive vesicles ( Figure 4I , right panel ) . In dendrites , Homer1b/c not only colocalized with SNX6 on vesicular structures ( Figure 4J , K ) , but also colocalized with EEA1 ( Figure 4L , M ) . Both wide-field microscopy with deconvolution and superresolution fluorescence microscopy revealed colocalization of EEA1 , Homer1b/c and SNX6 in dendrites ( Figure 4N , Table 1 , Figure 4—figure supplement 1 and Video 1 ) , indicating that SNX6 associates with Homer1b/c on endosomes . 10 . 7554/eLife . 20991 . 009Table 1 . Quantitative analysis of colocalization of signals in superresolution images and statistical significance of colocalization ( related to Figures 4N and 6D , and Figure 4—figure supplement 1 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 20991 . 009 voxel colocolization values ( % ) / p value EEA1-SNX6-Homer1b/c EEA1 with SNX6 and Homer1b/c15 . 68/ 0 . 0007912 . 05/ 0 . 01117 . 48/ 0 . 008615 . 18/ 0 . 010216 . 92/ 0 . 0086416 . 48/ 0 . 012715 . 33/ 0 . 009218 . 86/ 0 . 006915 . 76/ 0 . 008911 . 96/ 0 . 014121 . 87/ 0 . 0005315 . 98/ 0 . 0006514 . 58/ 0 . 009616 . 76/ 0 . 00119 . 04/ 0 . 0005SNX6 with EEA1 and Homer1b/c18 . 56/ 0 . 0007913 . 12/ 0 . 01117 . 28/ 0 . 008613 . 96/ 0 . 010217 . 44/ 0 . 0086412 . 36/ 0 . 012715 . 92/ 0 . 009222 . 64/ 0 . 006915 . 08/ 0 . 008914 . 32/ 0 . 014116 . 34/ 0 . 0005319 . 92/ 0 . 0006515 . 12/ 0 . 009615 . 44/ 0 . 00117 . 56/ 0 . 0005Homer1b/c with EEA1 and SNX614 . 24/ 0 . 0007913 . 16/ 0 . 01113 . 64/ 0 . 008610 . 22/ 0 . 010216 . 28/ 0 . 0086416 . 84/ 0 . 012713 . 24/ 0 . 009213 . 92/ 0 . 006914 . 84/ 0 . 008914 . 52/ 0 . 014118 . 56/ 0 . 0005315 . 36/ 0 . 0006513 . 92/ 0 . 009613 . 44/ 0 . 00117 . 63/ 0 . 0005 p150Glued-SNX6-Homer1b/c p150Glued with SNX6 and Homer1b/c15 . 63/ 0 . 0048 . 13/ 0 . 0110 . 06/ 0 . 01510 . 06/ 0 . 010210 . 06/ 0 . 005310 . 06/ 0 . 004710 . 06/ 0 . 00310 . 06/ 0 . 006910 . 06/ 0 . 01877 . 62/ 0 . 020< p <0 . 046016 . 38/ 0 . 00319 . 35/ 0 . 00518 . 85/ 0 . 008< p <0 . 03409 . 97/ 0 . 001879 . 21/ 0 . 0071SNX6 with Homer1b/c and p150Glued14 . 57/ 0 . 00412 . 89/ 0 . 0110 . 41/ 0 . 01513 . 02/ 0 . 010210 . 18/ 0 . 005312 . 67/ 0 . 004718 . 49/ 0 . 00312 . 33/ 0 . 00699 . 51/ 0 . 01879 . 54/ 0 . 010< p <0 . 025023 . 55/ 0 . 003111 . 24/ 0 . 00519 . 39/ 0 . 008< p <0 . 03209 . 18/ 0 . 0018710 . 7/ 0 . 0071Homer1b/c with SNX6 and p150Glued10 . 32/ 0 . 0048 . 27/ 0 . 018 . 33/ 0 . 0157 . 15/ 0 . 01029 . 66/ 0 . 00539 . 8/ 0 . 004713 . 16/ 0 . 0038 . 49/ 0 . 00698 . 73/ 0 . 01878 . 69/ 0 . 0120< p <0 . 036615 . 07/ 0 . 00319 . 17/ 0 . 00518 . 24/ 0 . 0100< p <0 . 03708 . 43/ 0 . 001879 . 03/ 0 . 0071 DIC-SNX6-Homer1b/c DIC with SNX6 and Homer1b/c9 . 21/ 0 . 00679 . 65/ 0 . 006120 . 05/ 0 . 00497 . 9/ 0 . 00713 . 41/ 0 . 00615 . 25/ 0 . 00589 . 91/ 0 . 0018710 . 88/ 0 . 00729 . 41/ 0 . 00798 . 53/ 0 . 006110 . 61/ 0 . 00719 . 64/ 0 . 00839 . 78/ 0 . 008210 . 98/ 0 . 008616 . 12/ 0 . 0042SNX6 with DIC and Homer1b/c12 . 00/ 0 . 006713 . 19/ 0 . 006114 . 91/ 0 . 004912 . 02/ 0 . 00713 . 07/ 0 . 00615 . 49/ 0 . 005811 . 32/ 0 . 001879 . 81/ 0 . 007210 . 1/ 0 . 007916 . 07/ 0 . 006110 . 79/ 0 . 007112 . 48/ 0 . 008311 . 77/ 0 . 00829 . 86/ 0 . 008623 . 11/ 0 . 0042Homer1b/c with DIC and SNX611 . 43/ 0 . 00679 . 81/ 0 . 006113 . 78/ 0 . 00499 . 82/ 0 . 00710 . 71/ 0 . 00610 . 86/ 0 . 00589 . 77/ 0 . 001879 . 82/ 0 . 00729 . 22/ 0 . 007910 . 36/ 0 . 00619 . 75/ 0 . 00717 . 87/ 0 . 00839 . 73/ 0 . 008210 . 21/ 0 . 008617 . 17/ 0 . 004210 . 7554/eLife . 20991 . 010Video 1 . 3D-SIM movie of an enlarged region of interest from a hippocampal neuron dendrite shows the association of EEA1 ( red ) , SNX6 ( green ) and Homer1b/c ( blue ) . DOI: http://dx . doi . org/10 . 7554/eLife . 20991 . 010 Dendritic distribution of Homer1b/c is essential for its scaffolding and signaling functions at the PSD . To determine whether SNX6 regulates Homer1b/c distribution in dendrites , we examined Homer1b/c expression and subcellular distribution in Snx6-/- hippocampal neurons by immunofluorescence staining . Indeed , although no change in the protein levels of Homer1b/c was detected in the hippocampi of Nestin-Cre; Snx6fl/fl mice ( Figure 5—figure supplement 1 ) , quantitative analysis revealed not only a decrease in the number of Homer1b/c puncta in dendritic segments ( 30–120 μm from the cell body ) of Snx6-/- neurons as compared with that of Snx6+/+ , but also a significant reduction in the fluorescence intensity of Homer1b/c puncta in spines ( Figure 5A , B ) . Overexpression of EGFP-SNX6 rescued both puncta number and spine distribution of Homer1b/c in Snx6-/- neurons ( Figure 5A , B ) . In contrast , neither the number nor the fluorescence intensity of PSD95 puncta was significantly affected in Snx6-/- neurons ( Figure 5C , D ) . Further , quantification of the Homer1b/c signal intensity over distance from the cell body revealed a decrease in both shaft and spines of the distal dendrites and concurrent accumulation in the soma of Snx6-/- neurons , which was rescued by mCherry-SNX6 ( Figure 5E , F ) . Together these data indicate that SNX6 is required for Homer1b/c distribution in distal dendrites . 10 . 7554/eLife . 20991 . 011Figure 5 . Partial loss of Homer1b/c from distal dendrites of Snx6-/- neurons . ( A ) Neurons were co-transfected with pLL3 . 7 . 1 and EGFP or EGFP-SNX6 construct on DIV13 , fixed on DIV18 and immunostained with antibodies to Homer1b/c . Shown are representative confocal images of dendritic segments . ( B ) Quantification of puncta number per 100 μm dendrite length and mean intensity in spines for Homer1b/c ( mean ± SEM , n = 30 , N = 3 ) . ( C ) Neurons were transfected with pLL3 . 7 . 1 on DIV13 , fixed on DIV18 and immunostained with antibodies to PSD95 . ( D ) Quantification of PSD95 distribution in dendrites ( mean ± SEM , n = 30 , N = 3 ) . ( E ) DIV14 neurons were co-transfected with constructs overexpressing EGFP and mCherry or mCherry-SNX6 , fixed on DIV16 and immunostained with antibodies to Homer1b/c . shown are representative confocal images of transfected neurons . Dashed lines outline the cell bodies . ( F ) Quantification of Homer1b/c distribution in the cell body and dendrites , and its mean intensity in spines in ( E ) ( mean ± SEM , n = 30 , N = 3 ) . ( G ) DIV13 neurons were co-transfected with constructs expressing DsRed and EGFP , EGFP-SNX6 , mEmerald-Homer1c-FL or EGFP-Homer1c-C and fixed on DIV18 . ( H ) Quantification of spine density in ( G ) ( mean ± SEM , n = 30 , N = 3 ) . Bars: 20 μm in ( E ) , 2 μm in other panels . DOI: http://dx . doi . org/10 . 7554/eLife . 20991 . 01110 . 7554/eLife . 20991 . 012Figure 5—figure supplement 1 . Immunoblotting analysis of Homer1b/c and PSD95 in hippocampi from Snx6fl/fl and Nestin-Cre; Snx6fl/fl mice . β-actin serves as loading control . Shown are samples from three pairs of littermates . Data represent mean ± SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 20991 . 01210 . 7554/eLife . 20991 . 013Figure 5—figure supplement 2 . Ablation of SNX6 causes decrease in spine density of CA1 but not CA3 neurons . ( A ) Hippocampal neurons in dissociated culture were transfected with pLL3 . 7 . 1 on DIV14 , fixed on DIV18 and immunostained with antibodies to CTIP2 , a transcription factor specifically expressed in DG and CA1 , but not CA3 neurons . Compared with DG neurons , CA1/CA3 neurons have larger cell bodies and more primary dendrites . Shown are representative confocal images of CA1 and CA3 neurons . ( B ) Quantification of spine density in ( A ) ( mean ± SEM , n = 30 cells , N = 3 ) . Bar: 2 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 20991 . 013 Homer and Shank are among the most abundant postsynaptic scaffolding proteins that contribute to the structural and functional integrity of dendritic spines . Consistent with the in vivo data , the spine density of Snx6-/- neurons was lower than that of Snx6+/+ in dissociated cultures ( Figure 5G , H , and Figure 5—figure supplement 2 ) . Overexpression of EGFP-SNX6 or mEmerald-Homer1c but not a Homer1c fragment that is truncated of its mGluR1/5-binding EVH1 domain ( Homer1c-C ) ( Shiraishi-Yamaguchi and Furuichi , 2007 ) restored the spine density of Snx6-/- neurons ( Figure 5G , H ) , indicating that SNX6-dependent dendritic distribution of Homer1b/c contributes to spine formation/stabilization , and that the cellular function of Homer1b/c as postsynaptic scaffold protein is required to restore the number of spines . Given that SNX6 is a cargo adaptor for the microtubule-based dynein−dynactin motor , we reasoned that SNX6 might mediate transport of Homer1b/c in dendrites . To this end , we transfected hippocampal neurons with constructs expressing mEmerald-Homer1c and mCherry-SNX6 ( Figure 6—figure supplement 1 ) and performed live-cell imaging by total internal reflection fluorescence microscopy ( TIR-FM ) to monitor their movement in dendrites . Indeed , we observed movement of SNX6- , Homer1c-double positive puncta in the shaft of both proximal and distal dendrites ( Figure 6A–C , Figure 6—figure supplement 2A , B , Videos 2 and 3 ) . A retrospective staining of MAP2 right after live imaging verified dendrite identity of the distal branch ( Figure 6C ) . Quantitative analysis revealed that , similar to mobility characteristics of PSD95 clusters in dendrites ( Gerrow et al . , 2006 ) , the majority ( ~90% ) of Homer1c puncta ( 1022 out of 1128 puncta from 31 neurons ) were stationary . Of note , the majority of motile SNX6- , Homer1c-positive puncta were smaller than 0 . 3 μm2 in size ( <600 nm in apparent diameter ) , whereas most of the immotile ones were larger ( Figure 6—figure supplement 2C ) , suggesting that the moving structures were vesicles rather than large protein aggregates . The SNX6- , Homer1c-positive puncta moved bidirectionally in the dendritic shaft , with the mean velocity of 0 . 416 ± 0 . 037 μm/s , over distances ranging from 2 . 026 to 18 . 324 μm ( Figure 6—figure supplement 2D–F ) . Movement of SNX6- , Homer1b-double positive puncta in the dendritic shaft was also observed by live imaging ( Video 4 ) . In contrast , in neurons overexpressing EGFP fusion of the GluN1 subunit of NMDAR , no comovement of SNX6- and GluN1-positive vesicles in dendrite was observed ( Video 5 ) . Moreover , we performed live imaging of mEmerald-Homer1c in Snx6-/- neurons and found that compared with wild-type , there was a dramatic decrease in the fraction of motile Homer1c fluorescent puncta ( 29 motile puncta out of 311 from 10 Snx6+/+ neurons vs . 10 out of 1217 from 40 Snx6-/- neurons ) . 10 . 7554/eLife . 20991 . 014Figure 6 . SNX6 is required for motility of Homer1b/c vesicles in dendritic shaft and their association with dynein−dynactin . ( A–C ) Dynamic behavior of mCherry-SNX6 and mEmerald-Homer1c in distal dendrite . The last frame of motile SNX6- , Homer1c-positive puncta ( arrowhead ) in dendrite ( A ) with the respective kymograph of boxed area ( B ) is shown . A retrospective staining of MAP2 after live imaging ( C ) illuminates dendrite identity . ( D ) Superresolution images of DIV18 neurons immunostained with antibodies to SNX6 , Homer1b/c , and p150Glued or DIC . Shown are representative images of dendrites outlined with dashed lines . White arrowheads indicate overlapped signals . ( E ) DIV14 neurons were transfected with construct expressing EGFP , treated with DMSO or Ciliobrevin D on DIV16 for 2 hr and immunostained with antibodies to Homer1b/c . ( F ) Quantification of puncta number per 100 μm dendrite length and mean intensity in spines for Homer1b/c in ( E ) ( mean ± SEM , n = 30 , N = 3 ) . ( G ) Same as ( E ) , shown are representative confocal images of EGFP-expressing neurons . Dashed lines outline the cell bodies . ( H ) Quantification of Homer1b/c distribution in the cell body and dendrites in ( G ) ( mean ± SEM , n = 30 , N = 3 ) . ( I ) DIV14 neurons were transfected with construct overexpressing EGFP or p150Glued-N-EGFP , fixed on DIV16 and immunostained with antibodies to Homer1b/c . Dashed lines outline the cell bodies . ( J ) Quantification of Homer1b/c distribution in the cell body and dendrites in ( I ) ( mean ± SEM , n = 30 , N = 3 ) . ( K ) Membrane fractions from mouse brain lysates were subjected to immunoisolation with antibodies to p150Glued or DIC coupled to Dynabeads Protein G , respectively . Shown are immunoblots probed with antibodies to SNX6 , p150Glued , DIC , and Homer1b/c . ( L ) Hippocampal neurons cultured from Snx6fl/fl and Nestin-Cre; Snx6fl/fl mice were transfected with pLL3 . 7 . 1 on DIV14 , fixed on DIV17 and immunostained with antibodies to Homer1b/c and p150Glued , DIC or EEA1 . Shown are representative confocal images of dendritic segments . ( M ) Quantification of colocalization in ( L ) from 45 dendritic segments of 15 neurons ( mean ± SEM , N = 2 . Total length of dendrites: Snx6fl/fl/Nestin-Cre; Snx6fl/fl: 1247 μm/1058 μm for p150Glued; 1264 μm/1301 μm for DIC; 1244 μm/1291 μm for EEA1 ) . Bars , 2 μm in ( A ) , ( D ) , ( E ) and ( L ) , 20 μm in ( G ) and ( I ) . DOI: http://dx . doi . org/10 . 7554/eLife . 20991 . 01410 . 7554/eLife . 20991 . 015Figure 6—figure supplement 1 . Overexpressed mCherry-SNX6 and mEmerald-Homer1c recapitulate the distribution of endogenous proteins in hippocampal neurons . ( A ) Hippocampal neurons in dissociated culture were transfected with construct expressing SNX6 on DIV14 , digitonin extracted and fixed on DIV16 and immunostained with antibodies to EEA1 , Homer1b/c , or PSD95 . Shown are the representative confocal images . ( B ) Hippocampal neurons in dissociated culture were transfected with construct expressing mEmerald-Homer1c on DIV14 , digitonin extracted and fixed on DIV16 and immunostained with antibodies to EEA1 , SNX6 , or PSD95 . ( C ) Hippocampal neurons in dissociated culture were fixed and stained with antibodies to Homer1b/c and PSD95 on DIV16 , or transfected with construct expressing membrane-bound GFP ( mGFP ) on DIV14 , digitonin extracted and fixed on DIV16 and immunostained with antibodies to Homer1b/c . ( D ) Quantification of colocalization between Homer1 and PSD95 in ( B ) and ( C ) ( mEmerald-Homer1c , 26 dendritic segments of 13 neurons; endogenous Homer1b/c , 22 dendritic segments of 11 neurons ) . ( E ) Quantification of size distribution of PSD95-positive ( + ) or negative ( − ) Homer1 puncta in dendrites from ( B ) and ( C ) . Sizes of puncta: small ( 0 . 04–0 . 3 μm2 ) , medium ( 0 . 3–0 . 8 μm2 ) and large ( 0 . 8–2 . 0 μm2 ) . Numbers of puncta quantified: mEmerald-Homer1c , 672 from 13 neurons; endogenous Homer1b/c , 502 from 11 neurons . Bar: 2 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 20991 . 01510 . 7554/eLife . 20991 . 016Figure 6—figure supplement 2 . SNX6- , Homer1c-positive puncta move bidirectionally in the dendritic shaft . ( A ) SNX6- , Homer1c-positive puncta in proximal dendrite move away from the cell body . Shown are the first frames of motile SNX6- , Homer1c-positive puncta in boxed areas ( Left panel , black arrow indicates the relative orientation of the cell body ) in proximal dendrite and the respective kymographs ( right panels , black arrowheads indicate the initial position and red arrowheads indicate the trajectory of vesicles ) . ( B ) SNX6- , Homer1c-positive puncta in distal dendrite moves towards the cell body . Shown are the first frame of a motile SNX6- , Homer-positive structure ( top left panel ) in distal dendrite and the kymograph ( all other panels ) . Bars: 2 μm . ( C ) Quantification of the size distribution of motile and immotile SNX6- , Homer1c-positive puncta in dendrites ( 1128 puncta from 31 neurons ) . Shown are the percentage of small ( 0 . 14‒0 . 3 μm2 ) , medium ( 0 . 3‒0 . 8 μm2 ) and large puncta ( 0 . 8‒2 . 8 μm2 ) in the motile and immotile populations , respectively . ( D ) The positions of all motile SNX6 and Homer1c-positive puncta as a function of time ( the trajectory of each vesicle ) were aligned at their starting positions away from the cell body ( 106 vesicles from 31 neurons ) . ( E and F ) Histograms of frequency distribution for run lengths ( E ) and average velocities ( F ) of anterograde and retrograde motile SNX6- , Homer1b/c-positive puncta . DOI: http://dx . doi . org/10 . 7554/eLife . 20991 . 01610 . 7554/eLife . 20991 . 017Video 2 . Time-lapse live imaging showing movement of Homer1c-labeled and SNX6-labeled puncta in proximal dendrites . Hippocampal neurons co-transfected with Emerald-Homer1c and mCherry-SNX6 expressing constructs were imaged live by TIR-FM . The trajectories of two mobile Homer1c- , SNX6-positive puncta are indicated by white arrowheads . Images were acquired at 2 frames/s . Video plays at 50 frames/s . Bar: 2 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 20991 . 01710 . 7554/eLife . 20991 . 018Video 3 . Time-lapse live imaging showing movement of Homer1c-labeled and SNX6-labeled puncta in distal dendrites . Hippocampal neurons co-transfected with Emerald-Homer1c and mCherry-SNX6 expressing constructs were imaged live by TIR-FM . The trajectories of two mobile Homer1c- , SNX6-positive puncta are indicated by white and yellow arrowheads respectively . Images were acquired at 2 frames/s . Video plays at 50 frames/s . Bar: 2 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 20991 . 01810 . 7554/eLife . 20991 . 019Video 4 . Time-lapse live imaging showing movement of Homer1b-labeled and SNX6-labeled structures in a distal dendrite . Hippocampal neurons co-transfected with mEmerald-Homer1b and mCherry-SNX6 expressing constructs were imaged live by TIR-FM . The trajectory of a mobile Homer1b- , SNX6-positive structure is indicated by white arrowheads . Images were acquired at 2 frames/s . Video plays at 50 frames/s . Bar: 2 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 20991 . 01910 . 7554/eLife . 20991 . 020Video 5 . Time-lapse live imaging showing movement of GluN1-labeled vesicles in dendrite . Hippocampal neurons co-transfected with mCherry-SNX6 and GluN1-EGFP expressing construct were imaged live by TIR-FM . The trajectory of the mobile GluN1-positive , SNX6-negative vesicle is indicated by white arrowheads . Images were acquired at 2 frames/s . Video plays at 50 frames/s . Bar: 1 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 20991 . 020 To verify that Homer1b/c-associated vesicles are of early endosome origin , we performed live imaging on neurons expressing fluorescently tagged EEA1 . First , we imaged neurons coexpressing EEA1-YFP and mCherry-SNX6 and observed movement of EEA1- , SNX6-double positive vesicles in the dendritic shaft ( Video 6 ) . In neurons coexpressing EEA1-YFP and mCherry-Homer1c , more than 90% of EEA1 vesicles also contained Homer1c ( 461/482 = 95 . 6%from 20 Snx6+/+ and 796/872 = 91 . 3 % from 26 Snx6-/- neurons ) , whereas the majority of Homer1c puncta were also EEA1-positive ( 461/690 = 66 . 8 % from 20 Snx6+/+ and 796/1249 = 63 . 7% from 26 Snx6-/- neurons ) . Live imaging detected not only movement of EEA1- , Homer1c-double positive vesicles in the dendritic shaft ( Video 7 ) , but also a significant decrease in the fraction of motile EEA1-positive vesicles in Snx6-/- neurons ( total EEA1-positive vesicles: 58/482 motile = 12% from 20 Snx6+/+ neurons vs . 37/872 motile = 4 . 2% from 26 Snx6-/- neurons; EEA1- , Homer1c-double positive vesicles: 53/461 motile = 11 . 5% from 20 Snx6+/+ neurons vs . 25/796 motile = 3 . 1% from 26 Snx6-/- neurons ) . In contrast , in dendrites of neurons coexpressing Homer1c and the late endosome marker Rab7 , neither comovement of Rab7- and Homer1c-positive structures was observed ( Videos 8 and 9 ) , nor the fraction of motile Rab7-labeled structures changed significantly when SNX6 was ablated ( 331/601 motile = 55 . 1% from 27 Snx6+/+ neurons vs . 341/647 motile = 52 . 7% from 22 Snx6-/- neurons ) . Together , these data indicate that SNX6 is required for the motility of EEA1-positive early endosomes or early endosome-derived vesicles carrying Homer1b/c . 10 . 7554/eLife . 20991 . 021Video 6 . Time-lapse live imaging showing movement of EEA1- and SNX6-labeled vesicles in dendrite . Hippocampal neurons co-transfected with EEA1-YFP and mCherry-SNX6 expressing constructs were imaged live by TIR-FM . The trajectories of two mobile EEA1- , SNX6-positive vesicles are indicated by white arrowheads . Images were acquired at 2 frames/s . Video plays at 50 frames/s . Bar: 1 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 20991 . 02110 . 7554/eLife . 20991 . 022Video 7 . Time-lapse live imaging showing movement of EEA1- and Homer1c-labeled vesicles in dendrite . Hippocampal neurons co-transfected with EEA1-YFP and mCherry-Homer1c expressing constructs were imaged live by TIR-FM . Yellow arrowheads indicate the trajectory of an EEA1- , Homer1c-positive vesicle detaching and moving away from a large structure , suggesting fission and formation of transport carriers from early endosomes . White arrowheads indicate the trajectory of another vesicle moving in the dendritic shaft . Images were acquired at 2 frames/s . Video plays at 50 frames/s . Bar: 1 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 20991 . 02210 . 7554/eLife . 20991 . 023Video 8 . Time-lapse live imaging showing movement of Rab7- and Homer1c-labeled structures in the dendrite of a wild-type neuron . Snx6+/+ hippocampal neurons co-transfected with Rab7-RFP and mEmerald-Homer1c expressing constructs were imaged live by TIR-FM . Images were acquired at 2 frames/s . Video plays at 50 frames/s . Bar: 1 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 20991 . 02310 . 7554/eLife . 20991 . 024Video 9 . Time-lapse live imaging showing movement of Rab7- and Homer1c-labeled structures in the dendrite of a Snx6 KO neuron . Snx6-/- hippocampal neurons co-transfected with Rab7-RFP and mEmerald-Homer1c expressing constructs were imaged live by TIR-FM . Images were acquired at 2 frames/s . Video plays at 50 frames/s . Bar: 1 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 20991 . 024 Next we determined whether dynein−dynactin is involved in SNX6-regulated dendritic distribution of Homer1b/c . Consistent with the immunoisolation data ( Figure 4I ) , superresolution microscopy analysis revealed that a significant fraction of both p150Glued and DIC colocalized with SNX6 and Homer1b/c on vesicular structures ( Figure 6D , Table 1 , Figure 4—figure supplement 1 and Videos 10 and 11 ) . Because dynein is essential for cell viability , we transiently treated neurons with Ciliobrevin D , an inhibitor of dynein activity ( Firestone et al . , 2012 ) , and detected a decrease in both shaft and synaptic distribution of Homer1b/c in distal dendrites ( Figure 6E–H ) . Further , overexpression of p150Glued-N , a dominant negative mutant that disrupts the interaction between SNX6 and dynactin ( Hong et al . , 2009 ) , also caused a decrease in Homer1b/c signals in distal dendrites and concurrent increase in the soma ( Figure 6I , J ) , indicating a role for dynein−dynactin in the dendritic distribution of Homer1b/c . 10 . 7554/eLife . 20991 . 025Video 10 . 3D-SIM movie of an enlarged region of interest from a hippocampal neuron dendrite shows association of p150Glued ( red ) , SNX6 ( green ) and Homer1b/c ( blue ) . DOI: http://dx . doi . org/10 . 7554/eLife . 20991 . 02510 . 7554/eLife . 20991 . 026Video 11 . 3D-SIM movie of an enlarged region of interest from a hippocampal neuron dendrite shows the association of DIC ( red ) , SNX6 ( green ) and Homer1b/c ( blue ) . DOI: http://dx . doi . org/10 . 7554/eLife . 20991 . 026 Next we asked whether SNX6 serves as a linker between the vesicles carrying Homer1b/c and the dynein−dynactin motor complex . To determine whether association of Homer1b/c and the motor complex on vesicles requires SNX6 , we attempted immunoisolation of Homer1b/c and dynein−dynactin-associated vesicles from mouse brain . Although antibodies to Homer1b/c failed to isolate Homer1b/c-positive vesicles from membrane fractions of mouse brain lysates , western blotting of vesicles immunoisolated with antibodies to both p150Glued and DIC detected Homer1b/c signals in wild-type but not CNS-SNX6 KO mice ( Figure 6K ) . Moreover , we performed immunostaining and colocalization analysis of Homer1b/c on Snx6+/+ and Snx6-/- neurons . Quantitative analysis of confocal images showed that there was a decrease in colocalization of Homer1b/c with p150Glued and DIC , but not with EEA1 , in Snx6-/- neurons ( Figure 6L , M ) . Together , these data suggest that SNX6 mediates dynein−dynactin-driven transport of a fraction of Homer1b/c molecules in the dendritic shaft . Although SNX6 mediates association of Homer1b/c vesicles with dynein‒dynactin , the majority of Homer1b/c puncta were immotile in steady-state neurons , suggesting that SNX6 might regulate Homer1b/c levels in distal dendrites via mechanisms other than vesicular transport . Moreover , ablation of SNX6 did not cause complete loss of Homer1/c from distal dendrites , suggesting that SNX6-independent mechanism ( s ) is required for Homer1b/c distribution in dendritic regions far from the cell body . Previous studies have found that the postsynaptic scaffolding proteins PSD95 , guanylate kinase domain-associated protein ( GKAP ) and Shank are transported in a preformed protein complex in dendrite ( Gerrow et al . , 2006 ) . To determine whether Homer1b/c shares the same trafficking pathway with PSD95 , we performed live imaging of neurons expressing PSD95-RFP and determined association of SNX6 and Homer1b/c with motile PSD95 clusters by retrospective immunofluorescence staining . No SNX6/Homer1b/c signals were detected on motile PSD95 clusters ( seven motile PSD95 puncta from five cells , Figure 7A and Video 12 ) , indicating that Homer1b/c is not cotransported with PSD95 . 10 . 7554/eLife . 20991 . 027Figure 7 . The dendritic trafficking pathway of Homer1b/c is distinct from the PSD95 and secretory trafficking pathways . ( A ) TIR-FM of hippocampal neurons transfected with PSD95-RFP expressing construct . Left panel: still image of the last frame of time lapse imaging . Arrowheads mark the final positions of two motile puncta . Center panels: confocal images of retrospective staining of endogenous SNX6 and Homer1b/c after live imaging . Right panels: enlargement of arrowhead-indicated , numbered puncta in the center panels . 1 and 2: motile PSD95 puncta lacking both SNX6 and Homer1b/c; 3: vesicle containing endogenous SNX6 and Homer1b/c , but not PSD95; 4: an immobile PSD95 punctum that contacts with SNX6 signal . 5: an immobile PSD95 punctum that colocalizes with Homer1b/c . 6: an immobile PSD95 punctum that colocalizes with both Homer1b/c and SNX6 . ( B ) DIV14 hippocampal neurons were transfected with construct overexpressing PKD-KD and immunostained with antibodies to Homer1b/c on DIV16 . Shown are representative confocal images . ( C ) Quantification of the mean intensity of Homer1b/c signals in the cell body and dendrites in ( B ) ( mean ± SEM , n = 30 , N = 3 ) . ( D ) The effect of PKD-KD overexpression on the distribution of Homer1b/c in dendrites and spines . ( E ) Quantification of Homer1b/c distribution in ( D ) ( mean ± SEM , n = 30 , N = 3 ) . Bars: 1 μm in ( A ) , 20 μm in ( B ) , 2 μm in ( D ) . DOI: http://dx . doi . org/10 . 7554/eLife . 20991 . 02710 . 7554/eLife . 20991 . 028Video 12 . Time-lapse live imaging showing movement of PSD95-RFP-labeled puncta in dendrites . Hippocampal neurons transfected with PSD95-RFP expressing construct were imaged live by TIR-FM . White arrowheads indicate two mobile puncta followed by fixation . Images were acquired at 2 frames/s . Video plays at 50 frames/s . Bar: 1 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 20991 . 028 Next we asked whether the secretory trafficking pathway in dendrite contributes to Homer1b/c distribution in dendritic shaft and spines . To address this question , we transfected neurons with plasmid overexpressing a kinase-dead version of protein kinase D1 ( PKD1-K618N , or PKD-KD ) , which blocks secretory trafficking by preventing fission of transport carriers from the TGN ( Liljedahl et al . , 2001 ) . Expression of PKD-KD did not affect either dendritic distribution or spine localization of Homer1b/c ( Figure 7B–E ) , indicating that its dendritic trafficking and synaptic delivery does not rely on the secretory pathway . As localization of Homer1b/c to synaptic sites in dendritic spines is crucial for its function , next we asked the question whether its translocation from shaft to spines requires SNX6 and vesicular transport . First , we determined whether or not shaft-localized Homer1b/c puncta enter spines . In our live imaging experiments using mEmerald-Homer1c , we did not detect entry of Homer1c puncta into spines in either wild-type or Snx6 KO neurons . We also imaged dendritic segments of neurons expressing mCherry-Homer1c ( Snx6+/+: 30 cells , 50 dendritic segments , 791 spines; Snx6-/-: 30 cells , 50 dendritic segments , 547 spines ) . Under the experimental condition we used , most motile fluorescent puncta moved in shaft ( Figure 8A and Video 13 ) , only one event of Homer1c particle entry into spine was observed ( Video 14 ) . Since the plus ends of microtubules transiently invade dendritic spines ( Jaworski et al . , 2009 ) and dynein is a minus end-directed motor , it is conceivable that dynein-driven transport is not involved in transfer of Homer1b/c from shaft to spines . Moreover , we reasoned that if delivery of Homer1b/c from shaft to spines requires SNX6 , there would be a decrease in spine distribution of Homer1b/c signals in dendrites of Snx6 KO neurons . Quantitative analysis showed that , although both total and spine Homer1b/c signals decreased in distal dendrites ( Figure 5F ) , its spine:shaft ratio remained constant throughout the dendrite and did not change in Snx6-/- neurons ( Figure 8B ) , indicating that in steady-state neurons , SNX6 is not involved in local trafficking of Homer1b/c from shaft to spines . 10 . 7554/eLife . 20991 . 029Figure 8 . Homer1b/c enters spines by SNX6-independent protein diffusion . ( A ) Representative images from a time-lapse video ( Video 13 ) of a wild-type neuron co-expressing EGFP and mCherry-Homer1c . White solid lines indicate outline of the dendritic shaft and spines . White arrowheads indicate positions of a Homer1c-labeled structure moving in the shaft . Bar: 2 μm . ( B ) Spine:shaft ratios of Homer1b/c fluorescence intensity over distance from the cell body . DIV14 neurons from Snx6fl/fl and Nestin-Cre; Snx6fl/fl mice were transfected with construct overexpressing EGFP as volume marker , fixed on DIV16 and immunostained with antibodies to Homer1b/c . Shown are the ratios of the mean intensity of Homer1b/c in spines to that in the corresponding shaft ( mean ± SEM , 50 spines from 15 neurons/group , N = 2 independent experiments ) . ( C ) FRAP analysis of mEmerald-Homer1c in dendritic spines . Hippocampal neurons were co-transfected with constructs expressing mEmerald-Homer1c and DsRed on DIV13 . FRAP analysis was performed on DIV16 . Shown are examples of the fluorescence intensity of mEmerald-Homer1c before , immediately after , 10 s and 600 s after photobleaching of the spines indicated with white circles . Bar: 1 μm . ( D ) Averaged fluorescence recovery curves after photobleaching for mEmerald-Homer1c in spines of Snx6+/+ ( 31 spines , 10 cells ) and Snx6-/- ( 32 spines , 10 cells ) neurons . Data represent mean ± SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 20991 . 02910 . 7554/eLife . 20991 . 030Video 13 . A motile Homer1c-labeled structure in dendritic shaft did not enter spines . Snx6+/+ hippocampal neurons co-transfected with EGFP and mCherry-Homer1c expressing constructs were imaged live by TIR-FM . Images were acquired at 2 frames/s . Video plays at 10 frames/s . White solid lines indicate outline of the shaft and spines . White arrowheads indicate the trajectory of a Homer1c-labeled structure moving in the dendritic shaft . Bar: 1 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 20991 . 03010 . 7554/eLife . 20991 . 031Video 14 . A motile Homer1c-labeled structure in dendritic shaft entered a spine . Snx6+/+ hippocampal neurons co-transfected with EGFP and mCherry-Homer1c expressing constructs were imaged live by TIR-FM . Images were acquired at 2 frames/s . Video plays at 10 frames/s . White solid lines indicate outline of the shaft and spines . White arrowheads indicate the trajectory of a Homer1c-labeled structure entering a spine . Bar: 1 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 20991 . 031 As a fraction of overexpressed fluorescently tagged Homer1c is cytosolic , next we tested the possibility that Homer1c enters spines via protein diffusion by fluorescence recovery after photobleaching ( FRAP ) assay on neurons expressing mEmerald-Homer1c . Consistent with previous findings on Homer1c turnover rates measured by FRAP using EGFP-Homer1c ( Kuriu et al . , 2006 ) , following full synapse photobleaching of fluorescent signals , 7 ~ 9 % of recovery after 10 s was observed in both wild-type and Snx6 KO neurons ( 8 . 09 ± 0 . 91% in Snx6+/+ and 8 . 14 ± 0 . 73% in Snx6-/- neurons , Figure 8C , D ) , indicating that although the fraction of fast fluorescence recovery in spines attributable to entry of soluble cytosolic proteins ( Blanpied et al . , 2008; Kerr and Blanpied , 2012; Kuriu et al . , 2006 ) is minor , diffusion of Homer1c protein molecules into spines does not require SNX6 . After 10 min , about half of Homer1c fluorescence was recovered in both wild-type and Snx6 KO neurons with similar recovery half-time ( Recovery level Rfinal = 54 . 55 ± 4 . 03% , τ1/2 = 108 . 4 ± 9 . 4 s in Snx6+/+ and Rfinal = 52 . 63 ± 3 . 53% , τ1/2 = 107 . 2 ± 13 . 1 s in Snx6-/- , Figure 8C , D ) . Further , similar results were obtained by fitting a single-exponential recovery curve to the average recovery time trace ( τ1/2 = 117 . 6 ± 11 . 3 s , Rfinal = 51 . 63 ± 0 . 86% in Snx6+/+ and τ1/2 = 108 . 9 ± 9 . 1 s , Rfinal = 48 . 57 ± 0 . 73% in Snx6-/- ) . These results are in good agreement with previous findings that there are immobile and mobile fractions of Homer1c in spines ( Kuriu et al . , 2006 ) and indicate that the dynamic turnover of Homer1c in spines is not affected by ablation of SNX6 . Taken together , these data indicate that in steady-state neurons , neither SNX6 nor vesicular transport is required for recruitment of Homer1b/c from shaft to spines , and that synaptic Homer1b/c exchanges with the soluble protein pool that enters the spine by diffusion . As ablation of SNX6 causes decrease in Homer1b/c distribution in distal dendrites , given the role of Homer1b/c in synaptic structure and function , next we sought to determine whether synaptic transmission is impaired in Snx6-/- neurons by electrophysiological analysis . We eliminated the Snx6 gene in a small subset of hippocampal neurons by injection of organotypic hippocampal slice culture from Snx6fl/fl mouse with recombinant adeno-associated virus ( AAV ) coexpressing EGFP and the Cre recombinase ( Figure 9A and Figure 9—figure supplement 1 ) . By simultaneous recording the evoked EPSCs ( eEPSCs ) on infected and adjacent uninfected CA1 pyramidal neurons , we found that AMPAR-mediated eEPSCs were significantly impaired by about 50% with ablation of SNX6 ( Figure 9B ) , whereas NMDAR-mediated eEPSCs and the pair-pulse ratio of AMPAR eEPSCs were unaffected ( Figure 9C , D ) , indicating that the impairment of AMPAR eEPSCs is due to decrease in the number of AMPARs in the postsynaptic membrane but not reduction of presynaptic glutamate release . Indeed , no change in surface expression of the GluN2B subunit of NMDAR was detected in Snx6-/- neurons by immunostaining and quantitative analysis ( Figure 9E , F ) . In contrast , there was a decrease in the surface expression of GluA1 and GluA2 , components of AMPAR , which was fully rescued by overexpressing SNX6 or Homer1c ( Figure 9G–J ) . 10 . 7554/eLife . 20991 . 032Figure 9 . Impaired synaptic transmission and decreased surface AMPAR levels of Snx6-/- neurons . ( A ) Hippocampal slice culture from Snx6fl/fl mouse was partially infected with AAV-EGFP-2A-Cre . Lower panel shows a CA1 neuron infected with AAV-EGFP-2A-Cre and an adjacent control neuron ( indicated with red arrows ) that were chosen to be recorded simultaneously . ( B ) Dual recording analysis of AMPAR-mediated synaptic responses ( n = 18 pairs ) . Scatterplots show amplitudes of AMPAR eEPSCs ( absolute values ) for single pairs of neurons ( open circles ) and mean ± SEM ( filled circle ) across all neuron pairs collected . The current amplitudes of infected neurons were plotted on the ordinate and the current amplitudes of the control neurons were plotted on the abscissa . Inset shows sample current traces from an infected ( green ) and a control ( black ) neurons . Bar graph shows mean ± SEM of AMPAR amplitudes represented in the scatterplots . ( C ) NMDAR-mediated eEPSCs ( n = 12 pairs ) . ( D ) Paired-pulse recording of AMPAR eEPSCs ( n = 11 pairs ) . Two identical stimulus pulses were delivered in an interval of 50 ms and AMPAR eEPSCs were recorded at −70 mV . Left are sample traces of eEPSCs from a pair of infected ( green ) and control neurons . The paired-pulse ratio ( PPR ) was the enhancement of the second eEPSC relative to the first eEPSC . Bar graph shows mean ± SEM of PPRs . ( E ) Hippocampal neurons transfected with pLL3 . 7 . 1 were fixed on DIV18 and immunostained with antibodies to surface GluN2B . ( F ) Quantification of puncta number per 100 μm dendrite length and fluorescence mean intensity in spines for surface GluN2B ( mean ± SEM , n = 30 , N = 3 ) . ( G ) DIV13 neurons were co-transfected with constructs expressing DsRed and EGFP , EGFP-SNX6 , mEmerald-Homer1c-FL or EGFP-Homer1c-C , fixed on DIV18 and immunostained with antibodies to surface GluA1 . ( H ) Quantification of signal intensity and spine distribution of surface GluA1 ( mean ± SEM , n = 32 , N = 3 ) . ( I ) Same as ( G ) , except that neurons were immunostained with antibodies to surface GluA2 . ( J ) Quantification of surface GluA2 ( mean ± SEM , n = 33 , N = 3 ) . ( K ) Hippocampal neurons from Snx6fl/fl ( WT ) and Nestin-Cre; Snx6fl/fl ( KO ) littermates were cultured till DIV16 . Surface levels of mGluR5 , GluA1 and GluA2 were then measured by cleavable surface biotinylation followed by immunoblotting with antibodies to mGluR5 , GluA1 and GluA2 . SNX6 serves as a negative control for surface proteins . Shown are immunoblots from two pairs of Snx6fl/fl and Nestin-Cre; Snx6fl/fl littermates . ( L ) Antibody uptake assay was performed on neurons transfected with pLL3 . 7 . 1 . Shown are representative images of dendrites immunostained for internalized and surface AMPAR signals . ( M ) Quantification of AMPAR endocytosis rate ( mean ± SEM , 45 dendritic segments , n = 15 , N = 3 ) . ( N ) DIV14 neurons were treated with inverse agonists for mGluRs for 48 hr , fixed and immunostained with antibodies to surface GluA1 . ( O ) Quantification of surface GluA1 ( mean ± SEM , n = 30 , N = 3 ) . ( P ) Same as ( N ) , except that neurons were immunostained with antibodies to surface GluA2 . ( Q ) Quantification of surface GluA2 ( mean ± SEM , n = 30 , N = 3 ) . Bar: 2 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 20991 . 03210 . 7554/eLife . 20991 . 033Figure 9—figure supplement 1 . Cre-mediated knockout of SNX6 in cultured neurons . Hippocampal neurons in dissociated culture from Snx6fl/fl mice were infected with AAVs expressing pAOV-CaMKIIα-EGFP-2A-3Flag or pAOV-CaMKIIα-EGFP-2A-Cre on DIV12 , fixed on DIV18 and immunostained with antibodies to SNX6 . Shown are representative confocal images . Bar: 10 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 20991 . 03310 . 7554/eLife . 20991 . 034Figure 9—figure supplement 2 . Ablation of SNX6 causes an increase in surface expression of mGluR5 in hippocampal neurons , which is rescued by SNX6 or Homer1c . ( A ) DIV13 neurons were transfected with constructs expressing DsRed and EGFP , EGFP-SNX6 , mEmerald-Homer1c-FL or EGFP-Homer1c-C , fixed on DIV18 and immunostained with antibodies to surface-expressed mGluR5 . ( B ) Quantification of surface mGluR5 fluorescence intensity per μm2 ( mean ± SEM , n = 30 , N = 3 , the exact p values are indicated in the bar graph ) . Bar: 2 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 20991 . 03410 . 7554/eLife . 20991 . 035Figure 9—figure supplement 3 . Homer1a overexpression increases surface mGluR5 expression . This experiment serves as a control for Figure 9—figure supplement 2 and is in agreement with previous reports ( Ango et al . , 2002 ) . ( A ) Hippocampal neurons were co-transfected with pLL3 . 7 . 1 and EGFP-Homer1a or EGFP construct on DIV13 , and fixed on DIV18 . Surface mGluR5 expressions were immunostained with antibodies to mGluR5 . ( B ) Quantification of surface mGluR5 signal in ( A ) ( mean ± SEM , n = 30 cells , N = 3 ) . Bar: 2 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 20991 . 03510 . 7554/eLife . 20991 . 036Figure 9—figure supplement 4 . Ablation of SNX6 causes a decrease in the number of clathrin positive synapses . ( A ) Hippocampal neurons in dissociated culture from Snx6fl/fl and Nestin-Cre; Snx6fl/fl littermates were immunostained with antibodies to clathrin heavy chain and PSD95 . Dendritic spines are outlined for quantification . ( B ) Quantification of the number of synapses without clathrin signal ( mean ± SEM , n = 30 cells , N = 3 ) . Bar: 2 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 20991 . 036 Next we investigated mechanism ( s ) underlying reduced AMPAR surface expression in Snx6-/- neurons . The longer isoforms of Homer inhibit not only trafficking to synaptic membrane but also constitutive activities of mGluR1/5 ( Ango et al . , 2001 , 2002 ; Roche et al . , 1999 ) . mGluR activation down-regulates surface expression of AMPAR by increasing its endocytosis rate ( Snyder et al . , 2001 ) . Indeed , there was an increase in surface levels of mGluR5 in Snx6-/- neurons ( Figure 9K , Figure 9—figure supplements 2 and 3 ) , which was restored by overexpression of SNX6 or Homer1c but not Homer1c-C ( Figure 9—figure supplement 2 ) . Moreover , ratiometric analysis of internalized and surface AMPAR detected an increase in the rate of AMPAR endocytosis in Snx6-/- neurons ( Figure 9L , M ) . To test the possibility that loss of Homer1b/c leads to decrease in surface AMPAR via increased surface levels and/or constitutive activation of mGluRs , we treated neurons with inverse agonists for mGluR1/5 and determined surface levels of GluA1 and GluA2 by immunostaining . Quantitative analysis indicated that inhibition of mGluRs did rescue surface expression of AMPAR in Snx6-/- neurons ( Figure 9N–Q ) . Together , these data indicate that dendritic loss of Homer1b/c caused by ablation of SNX6 leads to increase in mGluR surface expression and mGluR-regulated AMPAR endocytic trafficking . Another role of Homer1b/c in maintaining AMPAR surface levels is via the positioning of the endocytic zone ( EZ ) near the PSD ( Lu et al . , 2007 ) . In the dendritic spine , an EZ adjacent to the PSD captures and retrieves AMPAR diffusing out of the synapse through local endocytic trafficking ( Blanpied et al . , 2002; Lu et al . , 2007 ) , which is crucial for the supply of a mobile pool of receptor molecules required for synaptic transmission and potentiation ( Lu et al . , 2007; Petrini et al . , 2009 ) . To determine whether ablation of SNX6 caused uncoupling of the EZ from the PSD , we performed co-immunostaining of PSD95 and clathrin heavy chain , a marker for the EZ . Indeed , ablation of SNX6 caused ~2 fold increase in the fraction of EZ ( clathrin ) -negative synapses ( Figure 9—figure supplement 4 ) , indicating that the coupling of the EZ to the PSD is impaired in Snx6-/- neurons , which might also contribute to lower surface levels of AMPAR in dendritic spines . Previous studies have established a role for SNX6 as cargo adaptor in retromer-mediated vesicular transport ( Hong et al . , 2009; Wassmer et al . , 2009 ) . To test whether retromer is involved in SNX6-regulated dendritic distribution of Homer1b/c , we depleted VPS35 , a core component of the retromer complex , by siRNA-mediated RNA interference in hippocampal neurons ( Figure 10A , B and Figure 10—figure supplement 1 ) . Immunofluorescence staining and quantitative analysis of confocal images indicated that depletion of VPS35 in mature neurons caused a decrease not only in surface GluA1 but also in the number of Homer1b/c puncta in dendrites ( Figure 10C–E ) , possibly resulted from reduced spine density as VPS35 has been found to promote spine formation and maturation ( Tian et al . , 2015; Wang et al . , 2012 ) . Nevertheless , compared with the phenotypes of Snx6-/- neurons , there was decrease in Homer1b/c signals throughout the cell body and dendrites , but no change in spine distribution of Homer1b/c or surface levels of mGluR5 when VPS35 was depleted ( Figure 10C and E–H ) , indicating that it serves distinct function ( s ) from SNX6 in dendritic distribution of postsynaptic proteins . Moreover , live imaging of hippocampal neurons coexpressing mEmerald-Homer1c and VPS35-mCherry did not detect comovement of Homer1c- and VPS35-labeled vesicles in dendrite ( Figure 10I , J and Videos 15 and 16 ) . Further , retrospective staining of MAP2 and VPS35 right after live imaging confirmed the dendrite identity and absence of endogenous VPS35 at the base of the spine where a Homer1c-positive structure stopped ( Figure 10J , rightmost panels ) . To further confirm that SNX6 does not cooperate with the retromer to regulate motility of Homer1b/c vesicles , we performed coimmunostaining of Homer1b/c and VPS35 on Snx6+/+ and Snx6-/- neurons . Quantitative analysis showed that colocalization between Homer1b/c and VPS35 was very little compared with that between Homer1b/c and SNX6 or EEA1 . Moreover , it was not affected by ablation of SNX6 ( Figure 10—figure supplement 2A , B ) . Conversely , no change in colocalization between Homer1b/c and SNX6 was detected in VPS35-depleted neurons either ( Figure 10—figure supplement 2C , D ) . Taken together , these data indicate that SNX6 functions independent of retromer to regulate dendritic distribution of Homer1b/c . 10 . 7554/eLife . 20991 . 037Figure 10 . The retromer core complex is not required for SNX6-regulated dendritic distribution of Homer1b/c . ( A ) Hippocampal neurons were transfected with lentiviral vector expressing siRNA along with GFP at DIV12 , fixed on DIV18 and immunostained with antibodies to VPS35 . ( B ) Quantification of VPS35 signal intensity and puncta number in neurons in ( A ) ( mean ± SEM , n = 10 , N = 3 ) . ( C ) Neurons transfected with siRNA constructs were immunostained with antibodies to surface GluA1 and Homer1b/c . ( D–E ) Quantification of surface GluA1 ( D ) ( mean ± SEM , scrambled: 33 neurons; VPS35 RNAi #1: 30 neurons; VPS35 RNAi #2: 30 neurons , N = 3 . ) or Homer1b/c distribution in spines ( E ) ( mean ± SEM , scrambled: 35 neurons; VPS35 RNAi #1: 32 neurons; VPS35 RNAi #2: 30 neurons . N = 3 ) . ( F ) Quantification of Homer1b/c distribution in the cell body and dendrites of hippocampal neurons expressing scrambled or VPS35-targeting siRNA ( mean ± SEM , n = 15 , N = 3 ) . The results show a decrease in signal intensity of Homer1b/c throughout the cell when VPS35 was depleted . ( G ) Same as ( C ) , except that neurons were immunostained with antibodies to surface mGluR5 . ( H ) Quantification of surface mGluR5 in ( G ) ( mean ± SEM , scrambled: 35 neurons; VPS35 RNAi #1: 44 neurons; VPS35 RNAi #2: 52 neurons . N = 3 ) . ( I–J ) TIR-FM of hippocampal neurons transfected with Homer1c and VPS35-expressing constructs . Shown in ( I ) is a VPS35-positive vesicle ( white arrow ) moving away from the cell body and bypassing a static Homer1c-positive structure ( yellow arrow ) with their respective kymographs to the right . Shown in ( J ) are still images of representative time points: a Homer1c-positive structure reached the base of spine and part of which entered the spine after a brief lag . White arrowheads indicate the mobile Homer1c structure . A retrospective staining of MAP2 and VPS35 after live imaging ( right panels ) illuminates the dendrite identity and the absence of endogenous VPS35 at the base of the spine . Yellow arrowhead indicates the position of the Homer1c-positive structure right before fixation . White circles indicate VPS35 puncta appearing in both retrospective staining and live imaging . Bars: 5 μm in ( A ) and ( C ) , 20 μm in ( G ) and 1 μm in ( I ) and ( J ) . DOI: http://dx . doi . org/10 . 7554/eLife . 20991 . 03710 . 7554/eLife . 20991 . 038Figure 10—figure supplement 1 . VPS35 is efficiently knocked down in HEK293 cells . HEK 293 cells were transfected with lentiviral vector expressing siRNA along with EGFP followed by immunoblotting with antibodies to VPS35 and β-actin after 72 hr . The relative amount of VPS35 was determined with NIH ImageJ ( N = 3 experiments ) . DOI: http://dx . doi . org/10 . 7554/eLife . 20991 . 03810 . 7554/eLife . 20991 . 039Figure 10—figure supplement 2 . Colocalization of Homer1b/c with VPS35 does not require SNX6 and vice versa . ( A ) Hippocampal neurons in dissociated culture from Snx6fl/fl and Nestin-Cre; Snx6fl/fl littermates were transfected with pLL3 . 7 . 1 on DIV14 , immunostained with antibodies to Homer1b/c and VPS35 on DIV17 . Shown are representative confocal images . ( B ) Quantification of colocalization in ( A ) from 45 dendritic segments of 15 neurons ( mean ± SEM ) . ( C ) Hippocampal neurons were transfected with lentiviral vector expressing siRNA along with EGFP on DIV12 , fixed on DIV18 and immunostained with antibodies to Homer1b/c and SNX6 . Shown are representative confocal images . ( D ) Quantification of colocalization in ( C ) from 30–45 dendritic segments of 15 neurons ( mean ± SEM ) . Bar: 2 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 20991 . 03910 . 7554/eLife . 20991 . 040Video 15 . Time-lapse live imaging showing movement of a VPS35 vesicle in dendrite . Hippocampal neurons co-transfected with mEmerald-Homer1c and VPS35-6G-mCherry expressing constructs were imaged live by TIR-FM . White arrowheads indicate the trajectory of a VPS35-positive vesicle moving away from the cell body and bypassing a static Homer1c-positive structure ( indicated with yellow arrowhead ) . Images were acquired at 2 frames/s . Video plays at 50 frames/s . Bar: 1 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 20991 . 04010 . 7554/eLife . 20991 . 041Video 16 . Time-lapse live imaging showing movement of Homer1c and VPS35 puncta in dendrites . Hippocampal neurons co-transfected with mEmerald-Homer1c and VPS35-6G-mCherry expressing construct were imaged live by TIR-FM . White arrowheads indicate the initial and pausal sites of a Homer1c-positive , VPS35-negative structure . Yellow arrowhead indicates the spine position . Images were acquired at 2 frames/s . Video plays at 50 frames/s . Bar: 1 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 20991 . 041
In this study , we demonstrate that ablation of SNX6 in the CNS causes deficits in spatial learning and memory , a hippocampal-dependent brain function . At the cellular level , loss of SNX6 causes a decrease in spine density in the distal apical dendrites of CA1 hippocampal cells and impairment of their AMPAR-mediated synaptic transmission , indicating that SNX6 is required for synaptic structure and function of these excitatory neurons . We also show that SNX6 directly interacts with Homer1b/c , a PSD scaffolding protein crucial for the structural and functional integrity of dendritic spines , and that there is decrease in Homer1b/c distribution in distal dendrites in Snx6-/- neurons . Moreover , the spine density and surface AMPAR level phenotypes of Snx6-/- neurons could be rescued by overexpressing Homer1b/c or SNX6 . These findings reveal an important physiological function of SNX6 in the CNS excitatory neurons . The long isoforms of Homer1 are implicated in learning and memory . Homer1 knockout mice exhibit deficits in spatial learning and a decrease in LTP in the CA1 region ( Gerstein et al . , 2012; Jaubert et al . , 2007 ) , which could be rescued by AAV-mediated expression of Homer1c in the hippocampus ( Gerstein et al . , 2012 ) . Although Homer1b/c is widely expressed in the brain , in the hippocampus it is predominantly distributed to the CA1 region ( Shiraishi et al . , 2004 ) , which partially explains why ablation of SNX6 function causes a decrease in spine density of CA1 distal apical dendrites and impairs hippocampal-dependent spatial learning and memory that mainly involves the Schaffer collateral-CA1 synapses . Whether there are other SNX6-interacting proteins that are also required for CA1 neuron function awaits further investigation . As spatial learning and memory involve not only the hippocampus but also other cortical areas such as the entorhinal cortex and the medial prefrontal cortex ( Jo et al . , 2007; Nagahara et al . , 1995; Nakazawa et al . , 2004; Steffenach et al . , 2005; Zhou et al . , 1998 ) , it also remains to be determined whether and how ablation of SNX6 affects the synaptic structure and function of neurons in other parts of the cortex . PSD scaffolding proteins interact , anchor and stabilize glutamate receptors . Change in their protein content in dendritic shaft and spines influences synaptic transmission through receptor localization and distribution at synaptic sites . Among the PSD scaffolding proteins , PSD95 has been intensively studied . Dendritic trafficking and synaptic targeting of PSD95 requires a C-terminal tyrosine-based signal , palmitoylation of the N-terminus and its interaction with both Myosin V and dynein through binding of GKAP to these molecular motors ( Craven and Bredt , 2000; El-Husseini et al . , 2000; Hruska et al . , 2015; Naisbitt et al . , 2000 ) . Synaptic localization of Shank also requires GKAP ( Naisbitt et al . , 1999; Sala et al . , 2001 ) , whilst recruitment of Homer1b to dendritic spines requires synaptically targeted Shank ( Sala et al . , 2001 ) . We found that the motility of Homer1c-associated vesicles in dendritic shaft requires SNX6 , and that ablation of SNX6 or inhibition of dynein-dynactin activity causes reduction in the amount of Homer1b/c in distal dendrites . Previously imaging assays and quantitative modeling have established that dynein-driven bidirectional transport contributes to the polarized targeting of dendrite-specific cargo ( Kapitein et al . , 2010 ) . Therefore , lack of dynein‒dynactin-driven transport in the dendritic shaft provides a possible mechanism for the Homer1b/c distribution phenotype in Snx6-/- neurons . However , since the majority of Homer1c puncta are immobile in dendrites of steady-state neurons , and little is known about the cellular functions of SNX6 apart from its role as dynein cargo adaptor , it is also possible that SNX6 regulates the distribution of Homer1b/c in dendrites via mechanism ( s ) distinct from dynein‒dynactin-driven transport . Moreover , ablation of SNX6 does not cause complete loss of Homer1b/c from distal dendrites , indicating that mechanism ( s ) other than SNX6-mediated transport contributes to its localization to dendritic shaft far from the cell body . Since disruption of the secretory pathway does not affect Homer1b/c localization to dendritic shaft and spines , alternative mechanisms for its distribution in dendrites include diffusion of free protein molecules , cotransport with protein ( s ) other than the PSD95-GKAP-Shank complex or vesicular transport mediated by different motor ( s ) and/or adaptor ( s ) . Notably , in Snx6-/- neurons , although there was a decrease in the amount of Homer1b/c in both shaft and spines of distal dendrites ( Figure 5F ) , the spine:shaft ratio of its signals remained constant throughout the dendrite ( Figure 8B ) , indicating that once in dendrite , Homer1b/c could enter the spines via SNX6-independent mechanism ( s ) . In dendrites , several mechanisms exist for the transfer of postsynaptic components from shaft to synaptic sites in spines , including cytosolic diffusion , exocytosis of transmembrane proteins at the plasma membrane and their lateral diffusion to synaptic sites , and active transport by molecular motors . The AMPARs enter dendritic spines via both lateral diffusion and actin-based , Myosin V-driven transport of recycling endosomes ( Adesnik et al . , 2005; Correia et al . , 2008; Makino and Malinow , 2009; Wang et al . , 2008; Yudowski et al . , 2007 ) . Since Homer1b/c is a scaffolding protein , its entry into spines might rely on diffusion of free molecules , possibly released from endosomal carriers or from the cytosolic pool in the shaft , or transport of the vesicular cargo directly from the shaft by a different motor . Our results from live imaging , FRAP and quantitative analyses show that direct spine entry of Homer1c puncta is an extremely rare event , and that the dynamic turnover of Homer1c in spines is not affected by ablation of SNX6 . Collectively , these data indicate that in steady-state neurons , Homer1b/c enters spines by cytosolic diffusion , and SNX6 is not required for its spine localization . In retromer-mediated vesicular transport of transmembrane proteins , although the VPS26−VPS29−VPS35 core complex of retromer has been shown to interact with cargo proteins in the endosomal membrane ( Arighi et al . , 2004; Fjorback et al . , 2012; Nothwehr et al . , 2000 ) and was hence termed the cargo selection complex ( CSC ) , the SNX subunits of the retromer also contribute to cargo recognition ( Harterink et al . , 2011; Strochlic et al . , 2007; Temkin et al . , 2011; Voos and Stevens , 1998; Zhang et al . , 2011 ) . Our study not only shows that SNX6 directly interacts with Homer1b/c , but also provides evidence that SNX6-regulated distribution of Homer1b/c in dendritic shaft is retromer-independent . The CSC is involved in retrieval of APP from endosomes to the TGN and plasma membrane delivery and endocytic recycling of the AMPAR in dendritic spines ( Choy et al . , 2014; Fjorback et al . , 2012; Munsie et al . , 2015 ) . SNX27 , another member of the SNX family and a component of retromer ( Temkin et al . , 2011 ) , interacts with both NMDAR and AMPAR and regulates their recycling to the cell surface ( Cai et al . , 2011; Hussain et al . , 2014; Loo et al . , 2014 ) . It remains to be determined whether or not SNX6 functions in retromer-mediated dendritic transport . Notably , SNX6 shares the highest sequence similarity ( 85% ) with SNX32 , another SNX with unknown function . The mild neurodevelopmental phenotype exhibited by Nestin-Cre; Snx6fl/fl mice suggests functional redundancy among the SNX family members . Further studies are needed to characterize the roles of evolutionarily conserved SNXs , including SNX6 and SNX32 , in sorting and trafficking of neuronal proteins and their functions in synaptic development and activity .
All experiments were performed in compliance with the guidelines of the Animal Care and Use Committee of the Institute of Genetics and Developmental Biology , Chinese Academy of Sciences . The Nestin-Cre transgenic C57BL/6 J mice were obtained from the Nanjing Biomedical Institution of Nanjing University ( Tronche et al . , 1999 ) . The Thy1-EGFP transgenic mice were obtained from the Jackson Laboratory ( Feng et al . , 2000 ) . The Snx6fl/fl mice were generated at the Nanjing Biomedical Institution of Nanjing University . SNX6 CNS-specific knockouts were generated by crossing Snx6fl/fl mice with Nestin-Cre mice . A 7 . 1 kb genomic fragment of Snx6 containing exon 5 , 6 and 7 was cloned from a bacterial artificial chromosome ( BAC ) clone ( Clone name: RP23-422H22 , BACPAC Resource Center , Oakland , CA , USA ) to construct targeting vector . The first loxP site was inserted 248 bp upstream of exon 5 , and a 2 . 1 kb FRT-neo-loxP-FRT cassette with the second loxP site was inserted 424 bp downstream of exon 5 . The targeting vector was linearized with NotI and electroporated into C57BL/6NTac embryonic stem ( ES ) cells . ES cells were selected in G418- and ganciclovir-containing medium as described previously ( Liu et al . , 2003 ) . Genomic DNA of 96 drug-resistant colonies were isolated , digested with XhoI or HpaI and analyzed by southern blotting . Two independent ES cell clones ( 9 hr and 5D ) were chosen to inject into C57BL/6 J blastocysts to obtain chimeric mouse . CNS-specific Snx6 knockout mice were obtained by crossing Snx6fl/fl with Nestin-Cre mice . The Snx6 frameshift mutation was generated by deleting exon five via Cre-loxp mediated site-specific recombination . Deletion of exon five resulted in premature stop in mRNA translation at nt 270 . Genotyping of mouse lines was performed by genomic PCR . PCR genotyping of tail prep DNA from offspring was performed with the following primer pairs: FRTtF/loxtR: 5'-TTTGGCATTATCAAAACGTTGTTGTA-3' 5'-GAGATGCTCAGCACACTTTCTCTAC-3' ( PCR primer locations are shown in Figure 1B resulting in a PCR product of 295 base pairs in Nestin-Cre; Snx6fl/fl mice and none in Snx6fl/fl mice ) . loxtF/loxtR: 5'-AGAGTCACTATCAGAGCCTCTTCAG-3' 5'-GAGATGCTCAGCACACTTTCTAC-3’ ( PCR primer locations are shown in Figure 1B resulting in a PCR product of 366 base pairs in Snx6fl/flmice and none in Nestin-Cre; Snx6fl/fl mice ) . The Nestin-Cre transgene was detected using the following primer pairs: 5'-TGCCACGACCAAGTGACAGCAATG-3' 5'-ACCAGAGAGACGGAAATCCATCGCTC-3' Mice with a sparsely distributed population of GFP-expressing neurons for analysis of single-cell morphology were generated by crossing Nestin-Cre; Snx6fl/flmice to Thy1-EGFP transgenic mice . Thy1-EGFP transgene was detected using the following primer pairs: 5'-TCTGAGTGGCAAAGGACCTTAGG-3' 5'-CGCTGAACTTGTGGCCGTTTACG-3' The mEmerald-Homer1c construct was a generous gift from Michael Davidson ( Addgene plasmid # 54120 , Addgene , Cambridge , MA ) . The PSD-95-pTagRFP was a generous gift from Johannes Hell ( Addgene plasmid # 52671 ) . The FUmGW construct expressing membrane bound GFP ( mGFP ) was a generous gift from Connie Cepko ( Addgene plasmid # 22479 ) . The mCherry-Homer1C construct was generated by cloning full-length Homer1c amplified from mEmerald-Homer1c into pmCherry-C2 ( Clontech Laboratories , Inc . , Mountain View , CA ) . The mEmerald-Homer1b construct was derived from mEmerald-Homer1c by deletion of its aa 177–188 . EGFP-Homer1C-N/C constructs were generated by cloning Homer1c N-terminus ( aa1-187 ) /C-terminus ( aa188-366 ) amplified from mEmerald-Homer1c into pEGFP-C2 ( Clontech Laboratories , Inc . ) . His-sumo fusion of full-length Homer1c , Homer1c-N ( aa1-111 ) , Homer1c-M1 ( aa111-177 ) , Homer1c-M2 ( aa111-187 ) and Homer1c-C ( aa188-366 ) were generated by cloning fragments amplified from mEmerald-Homer1C into pET28a-sumo . His-sumo-Homer2b ( NCBI Accession: AF093260 . 1 ) and Homer3 ( NCBI Accession: NM_001146153 . 1 ) constructs were generated by PCR amplification of full-length Homer2b and Homer3 from mouse brain cDNA and cloning into pET28a-sumo . The GluN1-EGFP ( NR1-EGFP ) construct was a generous gift from A . Kimberley McAllister ( Department of Neurology , University of California at Davis , USA ) . The EEA1-YFP construct was a generous gift from Li Yu ( Tsinghua University , China ) . The Rab7-RFP construct was a generous gift from Hong Tang ( Institute of Biophysics , Chinese Academy of Sciences , China ) . The pEGFP-PKD-K618N construct was a generous gift from Vivek Malhotra ( Center for Genomic Regulation , Spain ) . pEGFP-N3-p150Glued-N ( aa 1–1060 ) and constructs expressing full-length SNX1 , SNX2 , SNX5 , and SNX6 fused with EGFP , mCherry or FLAG tag were described previously ( Hong et al . , 2009 ) . To obtain expression constructs for His-SNX1-N , His-SNX6-N , GST-SNX1-N and GST-SNX6-N , N termini of SNX1 ( aa 1–271 ) and SNX6 ( aa 1–181 ) were PCR-amplified from the full-length constructs and subcloned into pET28a and pGEX-4T-1 , respectively . To avoid disrupting the subunit interactions of VPS35-VPS29-VPS26 ( Munsie et al . , 2015 ) , VPS35-6G-EGFP was constructed by PCR amplifying human VPS35 coding sequence from HeLa cDNA with a pair of primers in which nucleotides encoding six glycines were added to the reverse primer and inserting into pCMV-EGFPN3 . To construct VPS35-6G-mCherry , EGFP sequence of VPS35-6G-EGFP was replaced by mCherry coding sequence from pCMV-mCherryC1 . Viral particles of adeno-associated virus ( AAV ) carrying pAOV-CaMKIIα-EGFP-2A-Cre and the control construct pAOV-CaMKIIα-EGFP-2A-3FLAG were purchased from Obio Technology ( Shanghai ) Corp . Ltd . , ( Shanghai , China ) . Antibodies used in this study are: mouse anti-GluA1 ( MAB2263 ) , mouse anti-GluA2 ( MAB397 ) , mouse anti-Tau1 ( MAB3420 ) , and mouse anti-MAP2 ( MAB3418 ) from Millipore ( Billerica , MA ) ; mouse anti-SYP ( D-4 ) ( sc-17750 ) , mouse anti-PSD95 ( 6G6 ) ( sc-32291 ) , mouse anti-DIC ( 74-1 ) ( sc-13524 ) , goat anti-Homer1a ( M-13 ) ( sc-8922 ) , rabbit anti-Homer1b/c ( H-174 ) ( sc-20807 ) , mouse anti-SNX6 ( D-5 ) ( sc-365965 ) , goat anti-SNX6 ( N-19 ) ( sc-8679 ) , and rabbit anti-Rab5B ( A-20 ) ( sc-598 ) from Santa Cruz Biotechnology ( Santa Cruz , CA ) ; mouse anti-p150Glued ( 610474 ) , mouse anti-EEA1 ( 610457 ) from BD Biosciences ( San Diego , CA ) ; rabbit anti-clathrin ( ab21679 ) , rabbit anti-TGN46 ( ab50595 ) , rat anti-CTIP2 ( ab18465 ) , rabbit anti-SNX1 ( ab134126 ) and rabbit anti-Rab4 ( ab109009 ) from Abcam; rabbit anti-GluN1 ( D65B7 ) from Cell Signaling Technology ( Mississauga , ON , Canada ) ; rabbit anti-GluN2B ( AGC-003 ) and rabbit anti-mGluR5 ( AGC-007 ) from Alomone labs ( Jerusalem , Israel ) ; rabbit and mouse anti-GFP ( MBL598 , D153-3 ) , rabbit and mouse anti-RFP ( PM005 , M155-3 ) from Medical and Biological Laboratories ( Naka-kuNagoya , Japan ) ; mouse anti-β-actin ( A5441 ) ( Sigma-Aldrich , St . Louis , MO ) ; mouse anti-Golgi97 ( A21270 ) from Invitrogen ( Carlsbad , CA ) ; goat anti-VPS35 ( PAB7499 ) from Abnova ( Taipei , Taiwan ) and rabbit anti-GluN2A ( 612–401-D89 ) from Rockland Immunochemicals ( Limerick , PA ) ; rabbit anti-Rab7 ( #9367 ) from Cell Signaling Technology ( Mississauga , ON , Canada ) ; rabbit anti-Rab11 ( 3H18L , 700184 ) from Life Technologies ( Carlsbad , CA ) ; Rabbit anti-SNX6 was described previously ( Hong et al . , 2009 ) . For SNX6 immunohistochemistry , 10 weeks-old mice were anesthetized and pre-fixed by perfusion with 4% paraformaldehyde ( pH 7 . 4 , Sigma-Aldrich ) transcardially . Brains were removed , post-fixed overnight at 4°C followed by dehydration with gradient sucrose ( 30% , 40% , 50% ) , and sectioned at 25 μm on a LEIGA CM 1950 vibratome ( Leica Biosystems , Germany ) . Braine slices were pasted on glass slides coated with gelatin/chromium potassium sulfate solution ( gelatin , 3 g , and chromium potassium sulfate , 0 . 05 g from Sigma-Aldrich were dissolved in 200 ml sterile water ) and antigen retrieval was performed in water bath with Tri-Sodium citrate ( 10 mM , pH 6 . 0 , dihydrate , Sigma-Aldrich ) at above 90°C for 20 min . Immunohistochemistry was performed following instructions from the manufacturer of polinker-2 plus polymer HRP detection system ( GBI , Bothell , WA ) . Briefly , After permeabilization with 1% TritonX-100 for 15 min followed by an endogenous peroxidase activity quenching in 3% hydrogen peroxide , sections were rinsed with PBS , blocked in 1% BSA plus 5% normal goat serum , and incubated with anti-SNX6 primary antibody ( Santa Cruz Biotechnology , diluted in PBS containing 1% BSA , 1% normal goat serum ) overnight at room temperature ( RT ) . Sections were then rinsed and incubated with polymer helper for 15 min at RT , poly-HRP anti-goat IgG for 20 min at 37°C followed by DAB color development , hematoxylin dyeing for the nucleus , and xylene clearing . Samples were analyzed using a Nikon ECLIPSE TE2000-U microscope . All mice used for behavior analysis were 10 weeks old male with normal eating and movement in cages by eye observation . One day before test , mice were transferred into the room installed with test platform . The mouse was fed in a separate cage one day before the test and was gently placed in an open-field test chamber and allowed to freely explore for 10 min . The locomotor activity ( total distance traveled in the whole chamber ) and the emotionality ( the percentage of distance and time spent in the center area ) were monitored and analyzed by an automated system ( the Anilab System , AniLab Software and Instruments Co . , Ltd , Ningbo , China ) . On the first training day , the mouse was placed on the rotating rod with straight line acceleration of 9 . 9 rpm/s from minimal ( 10 rpm ) to maximal speed ( 30 rpm ) ( the acceleration process takes about 10 min ) followed by fixed speed at 30 rpm for another 5 min . On the second day , the mouse was placed on the rotating rod with fixed speed ( 30 rpm ) , the motor function and coordination were determined by the latency to fall off the rod . The mouse was placed on a platform consisting of four perpendicularly intersected arms ( two open arms without walls and two arms enclosed by walls ) 50 cm high from the ground and allowed to freely explore for 6 min after a 4 min pre-adaptation by allowing the mouse to move freely in an open chamber with high walls placed on the ground after leaving the feeding cage . The ration of retention time staying at and numbers entering the open arms to closed arms were monitored and analyzed by an automated system ( the Anilab System , AniLab Software and Instruments Co . , Ltd ) . The mouse was suspended by its tail above 50 cm high from the ground for 5 min . Due to innate aversion to this tail-up situation , the mouse will struggle until immobile after multiple failures . A camera was installed as closely as possible in order to obtain the highest possible resolution of the animals . The immobile time was quantified and used to evaluate the depression condition . On the first day , the mouse was gently and slowly placed into a round tank ( height: 27 cm , diameter: 18 . 5 cm ) of which two-thirds was filled with water to , typically , avoid the animal's head from being submerged under the water for 90 s for pre-adaptation . On the second day , the mouse was placed in the same round tank for 5 min . The immobile time was quantified and used to evaluate the depression condition . Subject mouse was placed into the middle chamber and habituated for 5 min . Then , the wall between the chambers was removed to allow the mouse access freely to explore the three chambers with two empty wire cup-like cup housing in both left and right chambers for the first 10 min . The duration of subject mouse stretching into a 5 cm circular area around the cup is monitored as an active contact within 10 min . Then ‘stranger 1’ mouse was placed into cup housing in the left chamber for a second 10 min . For a third 10 min , ‘stranger 2’ mouse was placed into cup housing in the right chamber . The mouse was observed for a 10 min period in 20 cm × 30 cm quadrate cage with bedding . The duration of each mouse spending in the following behaviors was measured: cage-lid flipping/jumping , rearing , grooming and digging . Testing was performed in a Y-shaped maze consisting of three radial arms at a 120° angle . The mouse was put in the center of the maze and allowed to explore freely its three arms for 6 min . Typically , the mouse prefers to explore a new arm rather than the one that was just visited . The percentage of alternation ( the number of trials/the number of arm entries ) is calculated for evaluating the spatial working memory . The mouse was placed in a two-compartment shuttle box for about 5 ~ 10 min to make it quiet . A trial constituted of 5 s tone on followed by 5 s footshock ( 0 . 39 ~ 0 . 4 mA ) was given to make mouse build the association between tone and footshock . The mouse can avoid to receive the shock by escaping to the opposite compartment during tone on ( active escape ) , or can receive a foot shock shorter than 5 s during shock on after tone off ( passive escape ) . Each day 25 trials were performed with 15 s interval for five days . The time latency of active or passive escape was monitored and analyzed by an automated system ( the Anilab System , AniLab Software and Instruments Co . , Ltd ) . The average time latency of 25 trials each day was used to evaluate conditioned memory . The Morris water maze procedure consists of hidden platform acquisition training , probe trial testing and recall training . The water tank is a 120 cm diameter circular pool with a circular goal platform submerged 0 . 5 cm below the water surface . Water temperature was about 22°C to 24°C . Cues with different shapes and colors were pasted on the wall of the tank above the water surface in four different directions . A circular black curtain around the tank eliminated competing environmental cues . The mouse trajectory in the pool was monitored and analyzed by an automated system ( Smart 3 . 0 , Panlab SMART video tracking system , Barcelona , Spain ) . The day before the experiment , the mouse was gently placed into the pool without a platform to freely swim for 90 s for pre-adaptation . Acquisition training was then performed for eight days and four trials per day with different water-entering the site ( at north , south , east , and west positions adjacent to the pool wall ) . During each trial , mouse must learn to use cues to navigate a path to the hidden platform within 90 s . If failed to find the platform , mouse will be led to the platform with a stick , and kept on it for 10 s . The escape latency ( the average value of the time duration from entering the water to finding the platform of four trials per day ) was calculated for each mouse . After acquisition training , the hidden platform was removed and probe testing was performed for five days and one trial each day at the distal water-entering site away from the platform . A 1 . 5x platform circle area where the platform was placed was monitored . The latency to first enter 1 . 5x area ( time duration from entering the water to first enter the 1 . 5x area ) and numbers of crossing 1 . 5x platform circle area of each mouse within 90 s were analyzed . For recall training , after a 20 day-rest , mouse was placed in the same pool without a platform to examine memory extinguishment . Similarly , the latency to first enter 1 . 5x area and numbers of crossing 1 . 5x area of each mouse within 90 s were analyzed . Afterward , the platform was placed back to the pool and recall training was performed for one day with four trials with different water-entering sites . The second day , the platform was removed again , and the mouse was placed in the pool at the farthest water-entering site away from the platform . The latency to first enter 1 . 5x area and numbers of crossing 1 . 5x area of each mouse was analyzed . The 100 μm-thick brain slices from Snx6fl/fl; Thy1-GFP andNestin-Cre; Snx6fl/fl; Thy1-GFP mice were prepared as described in Histology and immunohistochemistry and mounted onto slides . For quantitative analysis of the morphology and density of dendrite spines , only those dendrite segments located in similar branches of the dendritic tree ( oriens/distal or radiatum/thin branches for CA1 basal/apical dendrites ( Megías et al . , 2001 ) , and secondary and tertiary branches in stratum oriens or stratum radiatum for CA3 basal/apical dendrites ( Baker et al . , 2011 ) from GFP-expressing and relatively isolated dorsal hippocampal CA1 or CA3 pyramidal neurons were selected for imaging by z-stack sectioning with a 0 . 40 μm interval using a Nikon confocal microscopy ( EZ-C1 , 100x oil , 4x optical zoom ) with the same acquisition parameters . Three-dimensional reconstructions were performed and spine density was quantified using the Filament module of IMARIS software as described previously ( Shen et al . , 2008 ) . The parameters were: the minimums pine head diameter ( thinnest diameter ) was ≥ 0 . 1 μm , the ratio of branch length to trunk radius was ≥ 1 . 5 μm , and the branch length ≥ 0 . 5 μm . The spine numbers each segment was further verified by manual counting . Spine density was defined as spine numbers/3D-segment length . Tissue preparation and electron microscopy were conducted as described ( Chen et al . , 2008 ) with slight modification . Briefly , eight-week-old mice ( Snx6fl/fl and Nestin-Cre;Snx6fl/fl ) were anesthetized with 2% pentobarbital sodium and perfusion-fixed with cold phosphate buffer ( PB , 0 . 1 M , pH 7 . 4 ) containing 2 . 5% glutaraldehyde ( Electron Microscopy Sciences , Hatfield , PA ) and 1% PFA ( Electron Microscopy Sciences ) . Following removal of mouse brain , the hippocampal CA1 and CA3 regions were sliced into 1 mm-thickness sections transverse to its longitudinal axis . Sections were fixed in the same fixative overnight at 4°C , rinsed with PB and postfixed with 1% osmium tetroxide ( OsO4 ) for 1 hr at 4°C in dark . Sections were then rinsed with distilled water and dehydrated in an ascending series of acetone ( 50% , 70% , 80% , 90% and 100% , 15 min per dilution ) . Samples were embedded in Embed 812 ( Electron Microscopy Sciences ) and polymerized at 60°C for 48 hr . Ultrathin sections ( 60 nm ) were mounted on carbon-coated copper grids , stained with 2% uranyl acetate for 15 min , rinsed with distilled water and stained with lead citrate for 5 min , rinsed with distilled water and air dried . Sections were imaged on a JEM-1400 electron microscope ( JEOL ) at 80 kV . Electron micrographs were captured with a Gatan 832-CCD ( 4 k x 3 . 7 k pixels , Gatan Inc . , Pleasanton , CA ) at 30 , 000× magnification . All image analysis was conducted blind to the genotype . Hippocampi from C57BL/6 J mouse embryos ( E17 . 5 ) were removed and trypsinized ( 0 . 125% trypsin , 15 min at 37°C ) . Dissociated cells were suspended in Dulbecco’s modified Eagle medium ( DMEM , Hyclone , Logan , UT ) supplemented with 10% horse serum and 10% F12 , then plated on coverslips pre-coated with poly-D-lysine ( 100 μg/ml , Sigma-Aldrich ) in 24-well plates at a density of 2000 ~ 5000 cells/well . Four hours later , the medium was replaced with neuronal culture medium ( Neurobasal medium , 2% B27 , 1% Glutamax ) . Half of the media were changed every three days until use . Neurons were transfected at DIV12 ~14 using Lipofectamine 2000 ( Invitrogen ) or Lipofectamine LTX ( Invitrogen , Carlsbad , CA ) according to the manufacturer’s recommendations . For VPS35 knockdown , Lentiviral vectors coexpressing VPS35-targeting siRNA and EGFP were purchased from Applied Biological Materials Inc . , USA . Target sequences for VPS35: siRNA #1: 5’-GGTGTAAATGTGGAACGTTACAAACAGAT-3'; siRNA #2: 5’-AGCTGTTATGTGCTTAGTAATGTTCTGGA-3' . Scrambled ( non-targeting ) siRNA: 5’-GGGTGAACTCACGTCAGAA-3' . Neurons were fixed for immunostaining analysis six days after siRNA transfection . The procedures for immunofluorescence staining and confocal microscopy were performed essentially as previously described ( Hong et al . , 2009 ) . Briefly , hippocampal neurons were fixed at DIV 16 ~ 18 with 4% paraformaldehyde supplied with 4% sucrose in phosphate-buffered saline ( PBS ) for 10 min at RT , then permeabilized and blocked in PBS containing 1% BSA , 0 . 4% Triton X-100 for 15 min at RT . Then primary and secondary antibodies conjugated with Alexa Fluor 488/555/647 were used for detection . For goat anti-SNX6 and rabbit anti-Rab5B ( Santa Cruz Biotechnology ) , antigen retrieval was performed as described in Histology and Immunohistochemistry for optimal staining . Surface staining of GluA1 and mGluR5 was performed as previously described by Peebles et al . ( 2010 ) . Surface staining of GluN2B and GluA2 was performed as previously described by Swanger et al . ( 2013 ) . For digitonin extraction of cytosolic proteins before immunostaining , neurons were rinsed with KHM buffer ( 20 mM HEPES ( pH 7 . 4 ) , 110 mM CH3CO2K , and 2 mM Mg ( CH3COO ) 2 ) and then treated with KHM buffer containing 25 μg/ml digitonin for 5 min on ice . Neurons were then rinsed once with KHM buffer and fixed with 4% paraformaldehyde ( PFA ) in PBS for 10 min at RT , blocked with PBS containing 5 % BSA and 0 . 2% Triton X-100 for 15 min followed by overnight incubation with primary antibody . Images were acquired by confocal microscopy ( EZ-C1 , 100x oil , 4x optical zoom ) and analyzed with the NIS-Elements AR3 . 1 software . Some confocal images ( Figure 5A , C , E , G; Figure 6E , G and I; Figure 7B , D; Figure 9E , G , I , N , P; Figure 5—figure supplement 2A; Figure 9—figure supplement 2; Figure 9—figure supplement 3A; Figure 9—figure supplement 4 ) were collected using z-stack with a 0 . 40 μm interval and analyzed with ImageJ . Confocal imaging after applying 31 × 31 median followed by Costes' auto-threshold subtraction was done to quantify colocalization ( Mander’s colocalization coefficient ( MCC ) , values are %; as previously described [Dunn et al . , 2011] ) . Control and experimental group neurons which were to be directly compared were imaged with the same acquisition parameters . To reduce variability , only segments of the secondary and tertiary dendrites ( distance from the cell body: 30–120 μm; length: 30–40 μm /segment ) were imaged . Ten to fifteen neurons each group in each independent experiment , and 90–120 μm dendrites per neuron were analyzed . Two-dimensional , background-subtracted , maximum projection reconstructions of Z-stack images were used for morphologic analysis and quantification . To examine the size , number , and fluorescence intensity of signal puncta in shaft and spines , the EGFP- or DsRed-labeled dendrites or spines were outlined manually . Numbers of puncta were measured manually and the size of Homer1 puncta as well as the mean intensity of fluorescent signals in individual spines was measured using the ImageJ function ‘Analyze > measure’ . Quantification of spine:shaft ratios of Homer1b/c was conducted as described ( Smith et al . , 2014 ) . Briefly , the dendritic shaft values of Homer1b/c signals was calculated as the mean fluorescence intensity of three regions of shaft along the dendritic region within 1–2 μm of analyzed spines , and used with corresponding spine values to produce spine:shaft ratios . All image analyses were conducted blind to the genotype . For quantitative analysis of VPS35 knockdown in neurons , the fluorescence intensity or VPS35 puncta from EGFP-expressing neurons in each group was normalized by that of untransfected neurons in the same group . The normalized values in each VPS35 siRNA group were further presented as relative values to the scrambled siRNA group . For identification of SNX6-interacting proteins , mouse brain was homogenized with lysis buffer ( 20 mM Tris . HCl , 10 mM HEPES , pH 7 . 4 , 150 mM NaCl , 1 mM EDTA , 20 mM imidazole , 1% Triton ) supplemented with protease inhibitors . Lysates were centrifuged at 12 , 000 × g for 20 min . The supernatants were incubated with recombinant His-tagged SNX1-N ( aa 1–271 ) or SNX6-N ( aa 1–181 ) immobilized on Ni-NTA agarose overnight at 4°C . Beads were rinsed with wash buffer ( 150 mM NaCl , 20 mM Tris . HCl , 10 mM Hepes , 1 mM EDTA , 100 mM imidazole , 1% Triton ) for five times and bound proteins were resolved by SDS-PAGE . The SDS-PAGE gel containing the protein sample was cut into pieces and destained with 25 mM ammonium bicarbonate/50% acetonitrile . Proteins in the sliced gels were reduced with 10 mM DTT at 37°C for 1 hr , and then alkylated with 25 mM iodoacetamide at RT for 1 hr in the dark before digested with trypsin ( Sigma T1426; enzyme-to-substrate ratio 1:50 ) in 25 mM ammonium bicarbonate at 37°C overnight . Tryptic peptides were extracted from gel by sonication with a buffer containing 5% trifluoroacetic acid and 50% acetonitirile . The liquid was dried by SpeedVac , and peptides were resolubilized in 0 . 1% formic acid and filtered with 0 . 45 μm centrifugal filters before analysis with a TripleTOF 5600 mass spectrometer ( AB SCIEX , Canada ) coupled to an Eksigent nanoLC . Proteins were identified by searching the MS/MS spectra against the Mus musculus SwissProt database using the ProteinPilot 4 . 2 software . Carbamidomethylation of cysteine was set as the fixed modification . Trypsin was specified as the proteolytic enzyme with a maximum of 2 miss cleavages . Mass tolerance was set to 0 . 05 Da and the false discovery rates for both proteins and peptides were set at 1% . For immunoprecipitation , HEK293T cells transfected with constructs expressing Flag-SNX6 and mEmerad-Homer1c were washed with ice-cold PBS and lysed with lysis buffer 1 ( 0 . 5% [vol/vol] NP-40 , 50 mM Tris-HCl , pH 7 . 4 , 150 mM NaCl , 1 mM EDTA ) supplemented with protease inhibitors . The following steps were performed as previously described ( Hong et al . , 2009 ) . HEK293T ( ATCC , Manassas , VA ) used for protein overexpression and immunoprecipitation in this study were negative for mycoplasma . Cells were cultured in DMEM ( HyClone ) supplemented with 10% fetal bovine serum ( FBS ) ( HyClone ) , penicillin and streptomycin ( HyClone ) . For endogenous IP , whole brains of 10 week-old C57 mice were homogenized with 10 times volume of lysis buffer 2 ( 150 mM NaCl , 20 mM Tris-HCl , 10 mM Hepes , 1 mM EDTA , 1% TritonX-100 , pH 7 . 4 ) supplemented with protease inhibitors and rotated for 30 min at 4°C . Lysates were centrifuged at 12 , 000 × g for 20 min . The supernatants were collected and incubated with 5 μg rabbit IgG ( control ) or mouse anti-SNX6 antibody bound to 20 μL of pre-washed Protein A/G Sepharose beads overnight at 4°C . Beads were washed five times with lysis buffer and bound proteins were eluted with 2× loading buffer and subjected to SDS-PAGE and immunoblotting . For GST-pull down assays , 2 μg of recombinant His-sumo-Homer1c-FL , His-sumo-Homer2b , His-sumo-Homer3 , His-sumo-Homer1c-N , His-sumo-Homer1c-M1 , His-sumo-Homer1c-M2 or His-sumo-Homer1c-C was incubated with GST , GST-SNX1-N or GST-SNX6-N immobilized on glutathione-Sepharose beads overnight at 4°C and proceeded as described for endogenous IP . For membrane IP/immunoisolation assays , mouse brain was homogenized with lysis buffer ( 20 mM Tris-HCl , 10 mM HEPES , pH 7 . 4 , 150 mM NaCl , 1 mM EDTA , 0 . 25 M sucrose ) supplemented with protease inhibitors and centrifuged at 800 × g for 15 min . The supernatants were collected and subjected to high-speed centrifugation at 100 , 000 × g for 1 hr ( TLS-55 rotor , Optima MAX Ultracentrifuge , Beckman Coulter , USA ) . The pellets ( p100 , the membrane fraction ) were resuspended in lysis buffer and subjected to immunoisolation with Dynabeads Protein G ( Invitrogen , Carlsbad , CA , USA ) coupled with 3–5 μg of mouse anti-SNX6 antibody . Then the beads were eluted by boiling in 2× SDS gel loading buffer and bound proteins were subjected to SDS-PAGE and immunoblotting . 3D-SIM images of immunostained neurons were acquired as previously described ( Niu et al . , 2013 ) on the DeltaVision OMX V4 imaging system ( Applied Precision Inc , USA ) with a 100 × 1 . 4 oil objective ( Olympus UPlanSApo ) , solid state multimode lasers ( 488 , 593 and 642 nm ) and EMCCD cameras ( Evolve 512 × 512 , Photometrics , USA ) . For quantitative analysis of 3D-SIM images with three colors/triple channels , first raw images were committed to costes auto-threshold subtraction for unbiased background subtraction as previously described ( Dunn et al . , 2011 ) . Then Biobprob ImageJ plugin was used to measure the colocalilzation of triple channels ( Fletcher et al . , 2010 ) . The statistical significance of triple colocalization was determined by comparing the values of voxel colocalization of the original images with those of images generated by randomizing spatial locations of signals in the original ones ( i . e . , colocalization occurred by chance ) , which was illustrated with p values produced by Biobprob ImageJ plugin ( parameters: voxel size: 40 , 40 , 125 nm; confidence interval: 95; p<0 . 05 means significantly more colocalization than chance ) . Hippocampal neurons isolated from embryonic day 17 . 5 ( E17 . 5 ) or newborn ( P0 ) mice were transfected at DIV 10 ~ 13 with Lipofectamine LTX ( Invitrogen ) and imaged live by TIR-FM ( Nikon TE2000-E equipped with 488 and 561 nm solid laser , 473/543 filter , 60 × 1 . 49 oil objective ( Nikon CFI Apochromat TIRF ) and EMCCD camera ( iXon Ultra 897 , ANDOR , UK ) ) at DIV 11 ~ 16 at 37°C with 5% CO2based on the optimal fluorescent protein expression . Image acquisition ( 512 × 512 pixels , two frames/sec for 5–10 min ) was controlled by μManager software . Kymographs were prepared using NIH ImageJ functions: ‘reslice’ with one pixel Z-spacing ( pixels ) and ‘Z projection’ with ‘Standard deviation’ type . The distance between start position of each track in Kymographs and cell body was recorded and used for aligning these tracks of the motile SNX6- , Homer1c- positive vesicles along dendrites . Average velocities ( run length / [ total time- pause time] ) were acquired with ImageJ plugins ‘Macros > read velocities from tsp’ . For retrospective staining , cells were fixed with 4% PFA immediately after live imaging , and immunostained with antibodies to MAP2 or endogenous SNX6 , Homer1b/c and VPS35 as previously described ( Hong et al . , 2009 ) . For size measurement of fluorescent puncta , the shape of puncta was distinguished by the naked eye from the diffuse cytosolic signal and the area value was obtained by the ImageJ function ‘Analyze>measure’ . FRAP was performed as described by Kerr and Blanpied ( 2012 ) . Briefly , hippocampal neurons were transfected with constructs expressing mEmerald-Homer1c and pLL3 . 7 . 1-DsRed as spine volume marker on DIV13 and FRAP was performed on DIV15 ~ 16 . Photobleaching of entire spines with synaptic Homer1c signals was achieved using 50 ~ 60% 488 nm laser power , while the following acquisition immediately after bleaching was achieved using 5 ~ 6% 488 nm laser power . To analyze the recovery of fluorescence , raw images were background subtracted frame-by-frame . The bleached spine and an additional ‘control’ spine were targeted as ROI . The recovery rate was calculated as R = ( I ( t ) -I ( 0 ) ) / ( I ( before bleaching ) -I ( 0 ) ) , with I ( 0 ) being the intensity immediately after bleaching . After normalization using the ‘control’ spine , recovery trace of each bleached spine over time was drawn . The fluorescence recovery at the end of recording time was determined as the Rfinal . After averaging the intensity of five time points , the first time point at which the intensity recovered to the half of Rfinal was determined as the half-time . By averaging all individual recovery traces and fitting a single exponential recovery curve using ‘Graphpad Prism 5>analysis>fit>exponential>one-phase association>least squares fit’ , the value of Rfinal and half-time obtained through the exponential fitting curve are similar to the average values obtained from each individual recovery trace . Cultured hippocampal slices were prepared from P6-P8 Snx6fl/fl male mice as previously described ( Herring et al . , 2013 ) . After one day in culture ( DIV1 ) , the slices were injected with adeno-associated virus ( AAV ) carrying EGFP-2A-Cre driven by the CaMKIIα promoter . Infected slices were cultured for an additional two weeks before recording . For recording , cultured slices were perfused with artificial cerebrospinal fluid ( ACSF ) , containing ( in mM ) NaCl 119 , KCl 2 . 5 , NaHCO3 26 , NaH2PO4 1 , glucose 11 , CaCl2 4 , MgCl2 4 , 2-chloroadenosine 0 . 01 ( to dampen epileptiform activity ) and saturated with 95% O2/5% CO2 . Isolation of currents from glutamatergic ( AMPA and NMDA ) receptors was achieved by adding picrotoxin ( 0 . 1 mM ) to the ACSF to block GABAA receptors . CA1 pyramidal cells were visualized by infrared differential interference contrast microscopy . The intracellular solution contained ( in mM ) CsMeSO4 135 , NaCl 8 , HEPES 10 , Na3GTP 0 . 3 , MgATP 4 , EGTA 0 . 3 , QX-314 5 , and spermine 0 . 1 . Cells were recorded with 3–5 MΩ borosilicate glass pipettes , following stimulation of Schaffer collaterals with bipolar metal electrodes placed in the stratum radiatum of the CA1 region . Series resistance was monitored and not compensated , and cells in which series resistance varied by 25% during a recording session were discarded . Synaptic responses were collected with a Multiclamp 700B-amplifier ( Axon Instruments , Foster City , CA ) , filtered at 2 kHz and digitized at 10 kHz . GFP-positive neurons were identified by epifluorescence microscopy . All paired recordings involved simultaneous whole-cell recordings from one GFP-positive neuron and a neighboring GFP-negative neuron . The stimulus was adjusted to evoke a measurable , monosynaptic eEPSC in both cells . AMPAR eEPSCs were measured at a holding potential of −70 mV , and NMDAR eEPSCs were measured at +40 mV 150 ms after the stimulus , at which point the AMPAR eEPSC has completely decayed . All paired recording data were analyzed statistically with a Wilcoxon Sign Rank Test for paired data . A p value of <0 . 05 was considered statistically significant . Error bars represent standard error measurement . At DIV 16 , cultured hippocampal neurons were treated with 20 μM Ciliobrevin D ( EMD Chemicals , Gibbstown , NJ ) for 2 hr and fixed for immunostaining analysis . For mGluR inverse agonist treatment , experiments were performed as previously ( Hu et al . , 2010 ) . Briefly , DIV14 neurons were treated with Bay 36–7620 ( Bay , 10 μM , Sigma-Aldrich ) and MPEP ( 5 μM , Sigma-Aldrich ) dissolved in DMSO for 48 hr and fixed for the subsequent assay . Endocytosis assay of GluA1 in steady state neurons was performed essentially as previously described ( Lu et al . , 2007 ) . Briefly , DIV16 hippocampal neurons were pre-chilled on ice for 5 min and incubated with GluA1 N-terminal antibody ( 1:200 , Millipore , Billerica , MA ) for 30 min . After antibody washout , neurons were transferred to 37°C for 30 min before fixation in PFA-sucrose . Surface-bound GluA1 was immunostained by using Alexa Fluor 647-conjugated goat anti-mouse secondary antibodies ( 1:1000 , Invitrogen ) followed by blocking surface-remaining GluA1 with unconjugated goat anti-mouse secondary antibodies ( undiluted stock , Jackson ImmunoResearch , West Grove , PA ) . Then neurons were post-fixed with PFA and permeabilized with blocking buffer containing 0 . 4%Triton-100 . The internalized GluA1 was immunostained with Alexa Fluor 488-conjugated goat anti-mouse antibody ( 1:1000 , Invitrogen ) . Confocal images were captured and used for calculating the ratio of internalized GluR1 as follows after applying median and costes auto-threshold subtraction: mean intensity of internalized GluA1/ ( mean intensity of surface GluA1 + mean intensity of internalized GluA1 ) . Biotinylation assay was performed as previously described ( Fu et al . , 2011 ) . Briefly , DIV16 hippocampal neurons cultured from newborn ( P1 ) mice were washed twice with ice-cold PBS containing 1 mM CaCl2 and 0 . 5 mM MgCl2 ( PBS-Ca-Mg ) , then incubated with PBS-Ca-Mg supplemented with 0 . 25 mg/ml EZ-link Sulfo-NHS-LC-biotin ( Pierce , Thermo Fisher Scientific , Rockford , IL ) for 1 hr at RT with mild shaking . The biotinylation reaction was quenched and unbound biotin was removed by washing the cells twice with PBS-Ca-Mg containing 100 mM glycine for 15 min at 4°C . Neurons were then lysed in lysis buffer ( 50 mM Tris-Cl , 10 mM HEPES , pH 7 . 4 , 150 mM NaCl , 0 . 5 mM EDTA , 1% Triton ) supplemented with protease inhibitors . After centrifugation at 8000 × g at 4°C for 10 min , the supernatants were collected and incubated with streptavidin beads ( Thermo Fisher Scientific , Rockford , IL ) overnight at 4°C . Beads were washed five times with lysis buffer . Bound proteins were eluted with 2× loading buffer and subjected to SDS-PAGE and immunoblotting . All data are presented as mean ± SEM . GraphPad Prism 5 ( GraphPad Software ) was used for statistical analysis . The two-tailed unpaired t-test was used for statistical analysis of immunoblotting data and to evaluate statistical significance of two groups of samples . One-way analysis of variance with a Tukey post hoc test was used to evaluate statistical significance of three or more groups of samples . For TEM data , non-parametric two-sided statistical testing Mann–Whitney was used to avoid the restriction of sample sizes without a normal distribution ( Morris , 2000 ) . A p value of less than 0 . 05 was considered statistically significant . | Neurons are the building blocks of the nervous system . These cells generally consist of a round portion called the cell body and a long cable-like axon . The cell body bears numerous branches called dendrites , which are in turn covered in spines . Neurons communicate with one another at junctions – or synapses – that typically form between the end of the axon of one cell and a dendritic spine on another . Specialized proteins stabilize the dendritic spines and enable the cells to exchange messages across the synapse . However , it is the cell body – rather than the dendrites – that produces most of these proteins . Structures called molecular motors transport proteins to their destinations within the cell along fixed tracks , similar to how a freight train carries cargo over the rail network . One of the key molecular motors within neurons is called dynein‒dynactin . This in turn interacts with other proteins called adaptors , enabling it to transport specific types of cargo . Niu , Dai , Liu et al . have now examined the role of SNX6 , an adaptor protein for the dynein‒dynactin motor . Mice that have been genetically modified to lack SNX6 in their brains have fewer spines on their dendrites compared with normal mice . This was particularly true for dendrites that contain AMPAR , a protein that receives signals sent across synapses . Niu , Dai , Liu et al . showed that SNX6 interacts with another protein called Homer1b/c and is responsible for distributing this protein in dendrites far from the cell body . The Homer1b/c protein helps to stabilize dendritic spines and to regulate the number of AMPAR proteins within them . Mice that lack SNX6 therefore have less Homer1b/c in the dendrites furthest from the cell body , and fewer spines on these dendrites too . These mice also have fewer AMPAR proteins at their synapses than control mice . Mice that lack SNX6 show impaired learning and memory compared to control mice . This is consistent with the fact that changes in the strength of synapses that possess AMPAR proteins are thought to underlie learning and memory . Additional experiments are required to explore these relationships further , and to determine whether SNX6 helps to localize any other proteins that also contribute to changes in the strength of synapses . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"cell",
"biology",
"neuroscience"
] | 2017 | Ablation of SNX6 leads to defects in synaptic function of CA1 pyramidal neurons and spatial memory |
Supergene regions maintain alleles of multiple genes in tight linkage through suppressed recombination . Despite their importance in determining complex phenotypes , our empirical understanding of early supergene evolution is limited . Here we focus on the young ‘social’ supergene of fire ants , a powerful system for disentangling the effects of evolutionary antagonism and suppressed recombination . We hypothesize that gene degeneration and social antagonism shaped the evolution of the fire ant supergene , resulting in distinct patterns of gene expression . We test these ideas by identifying allelic differences between supergene variants , characterizing allelic expression across populations , castes and body parts , and contrasting allelic expression biases with differences in expression between social forms . We find strong signatures of gene degeneration and gene-specific dosage compensation . On this background , a small portion of the genes has the signature of adaptive responses to evolutionary antagonism between social forms .
Evolutionary antagonism can emerge when selection favors multiple phenotypic optima within a population . This process can lead to selection for reduced recombination between co-adapted alleles encoding the different phenotypes ( Charlesworth , 2016 ) . In turn , reduced recombination will sometimes favor the formation of supergene regions containing tightly linked alleles of up to hundreds of genes . Such regions enable the maintenance of genetic interactions over evolutionary time ( Darlington and Mather , 1949; Thompson and Jiggins , 2014 ) . We now know that supergenes controlling ecologically important traits are widespread . Examples include flower heterostyly in Primula ( Li et al . , 2016 ) , mating type in Mycrobotryum fungi ( Branco et al . , 2018 ) , Batesian mimicry in butterflies ( Joron et al . , 2011; Kunte et al . , 2014 ) , mating behavior in white-throated sparrows ( Zinzow-Kramer et al . , 2015; Tuttle et al . , 2016 ) and male sexual morphs in ruff sandpipers ( Küpper et al . , 2016; Lamichhaney et al . , 2016 ) . The non-recombining portions of sex chromosomes are extensively studied examples of supergenes ( Bergero and Charlesworth , 2009 ) . Their evolution is thought to have been shaped by antagonism over sexual phenotypes ( Zemp et al . , 2016; Vicoso et al . , 2013; Khil et al . , 2004; Parsch and Ellegren , 2013; Wright et al . , 2017; Mank , 2017 ) . However , studies focusing on young sex chromosome systems suggest otherwise ( Dufresnes et al . , 2015; Nozawa et al . , 2014; Alekseyenko et al . , 2013; Muyle et al . , 2012; Charlesworth et al . , 2005; Stöck et al . , 2011 ) . Indeed , sex chromosome evolution could be driven mostly by processes resulting from suppressed recombination rather than by antagonistic selection ( Branco et al . , 2018; Cavoto et al . , 2018; Branco et al . , 2017; Ma et al . , 2020 ) . The social supergene of the red fire ant Solenopsis invicta is an excellent model for comparing how evolutionary antagonism and the consequences of suppressed recombination shape early supergene evolution . This species includes single-queen and multiple-queen colonies , which differ in behavior , physiology and ecology . This social polymorphism is controlled by a pair of ‘social chromosomes’ , SB and Sb , that carry distinct supergene variants ( Wang et al . , 2013 ) . In single-queen colonies , workers and queens are SB/SB homozygotes . In multiple-queen colonies , egg-laying queens are SB/Sb heterozygotes , but workers can either be homozygous or heterozygous . Sb/Sb queens are rare and their offspring are unviable ( Gotzek and Ross , 2007 ) . Furthermore , because of at least three inversions ( Stolle et al . , 2019; Huang et al . , 2018 ) , recombination is severely repressed between SB and Sb ( Wang et al . , 2013; Pracana et al . , 2017a ) . Similar to a Y chromosome , Sb thus lacks recombination opportunities . Importantly , the fire ant supergene system could be as young as 175 , 000 generations . It thus offers a rare glimpse into the evolutionary forces shaping the early stages of supergene evolution ( Wang et al . , 2013 ) . Suppressed recombination reduces the efficacy of purifying selection through Hill-Robertson interference , which can lead to gene degeneration ( Charlesworth , 2016 ) . Accordingly , Sb has accumulated non-synonymous single nucleotide substitutions and repetitive elements ( Stolle et al . , 2019; Pracana et al . , 2017a ) . Despite Sb degeneration , gene content in SB and Sb is highly similar ( Wang et al . , 2013; Pracana et al . , 2017a ) , likely because of the system’s young age and purifying selection in haploid males ( Hall and Goodisman , 2012 ) . Modifications in the genomic sequence can lead to gene expression changes ( Denver et al . , 2005; Rifkin et al . , 2005 ) , the accumulation of mutations in Sb alleles of some genes could therefore result in SB/Sb expression bias . Changes in expression due to gene degeneration should result in specific expression patterns . For instance , if such mutations go to fixation because of inefficient purifying selection , they would likely result in consistent allelic bias across all tissues . Alternatively , if such mutations go to fixation because they are adaptive responses to antagonistic selection , their expression levels would be likely to require tissue-specific fine-tuning for particular functions . This process should result in tissue-specific allelic bias , as observed across human tissues ( GTEx Consortium , 2015 ) . The accumulation of deleterious mutations in Sb alleles should lead to lower expression levels due to impacts in regulatory sequences . The opposite effect seems to be less likely and may only involve highly specific mutations ( Loewe and Hill , 2010 ) . Additionally , selection against the expression of deleterious alleles should result in the downregulation of some Sb alleles ( Ma et al . , 2020; Xu et al . , 2019; Pucholt et al . , 2017 ) . If lower Sb expression occurs for dosage-sensitive genes , selection should favor upregulation of the corresponding SB alleles . Such dosage compensation can occur through different mechanisms . For instance , in male Drosophila melanogaster the entire X chromosome is expressed at roughly twice the level of autosomes to compensate for the lack of expression in the highly degenerated Y chromosome ( Conrad and Akhtar , 2012 ) . In species with much younger sex chromosomes such as Drosophila miranda , dosage compensation instead occurs only at a gene-by-gene level ( Alekseyenko et al . , 2013; Nozawa et al . , 2018 ) . Because the fire ant supergene is young , we hypothesize that individual genes for which the Sb allele has begun to degenerate would show evidence of dosage compensation . For such genes , the SB allele would be more highly expressed than the Sb allele , while their combined total expression should be similar between SB/Sb and SB/SB individuals . Because the supergene variants determine social forms in the fire ant , their evolution will also have been shaped by antagonistic selection . We thus expect that the supergene region is enriched for genes with alleles that are beneficial for one of the social forms but detrimental for the other ( Zemp et al . , 2016; Vicoso et al . , 2013; Khil et al . , 2004; Parsch and Ellegren , 2013 ) . Genes with such alleles are more likely to also show differences in expression between social forms , resulting in an enrichment of socially biased expression in the supergene . However , such enrichment could instead emerge from lowered expression associated with gene degeneration ( Ma et al . , 2020 ) , or the fixation of mutations with neutral phenotypic effects ( Harrison et al . , 2012 ) . Adaptation through lower expression of Sb alleles is indistinguishable from gene degeneration . In contrast , increased expression of Sb alleles is more likely to result from adaptation ( Harrison et al . , 2012; Pál et al . , 2001 ) . This line of reasoning has been used in the analysis of sex chromosomes , in which patterns of high sexually-biased expression for sex-chromosome linked genes is used as a proxy for benefit ( Mank , 2017; Mank et al . , 2013; Zhou and Bachtrog , 2012 ) . Following this logic and given the expected general pattern of Sb downregulation due to degeneration , genes with high expression of the Sb allele are more likely to be adaptive . Furthermore , because Sb should be enriched in alleles beneficial for multiple-queen colonies , we expect that genes with high expression in multiple-queen colonies will tend to show Sb bias . We disentangle these evolutionary processes by generating detailed genomic and transcriptomic data: we sequenced genomes of fire ants from their native South American range and combined these with existing genomes from the invasive North American range . This enables us to identify genes with fixed differences between the SB and Sb variants of the social chromosome . To detect differences in expression between SB and Sb alleles , we performed RNA sequencing ( RNA-seq ) from SB/Sb individuals from the South American range of the species . We used three body parts of queens ( head , thorax and abdomen ) and whole bodies of workers . We combined this data with published RNA-seq data from SB/Sb individuals from invasive North American and Taiwanese populations . We then compared these expression patterns with those obtained from comparisons between social forms . We find that most genes in the social chromosome show no strong allelic bias and that there is no clear pattern of supergene-wide expression bias towards either variant . Genes with biased expression tend to show patterns consistent with gene degeneration , such as lower Sb expression with increasing numbers of non-synonymous mutations . Additionally , we find that more genes are SB biased than expected given differences in expression between social forms , a pattern consistent with partial dosage compensation . Accordingly , the accumulation of non-synonymous mutations in Sb alleles correlates with an allelic bias towards SB , consistent with ongoing Sb degeneration . Despite these observations , we find an overrepresentation of Sb-biased genes among genes with higher expression in individuals from multiple-queen colonies . This result indicates that antagonistic selection has also shaped the expression patterns in the fire ant supergene . Given the observed impact of gene degeneration , our results highlight the importance of considering the genomic context of gene expression patterns before making inferences about functionality .
To identify differences between supergene variants , we obtained 408-fold genome coverage from 20 haploid SB males and 20 haploid Sb males . Thirteen within each group ( 65% ) were from the native South American range whereas the rest were from an invasive North American population ( Wang et al . , 2013 ) . By comparing the two groups of males we identified 2877 single nucleotide polymorphisms ( SNPs ) with one allele in all SB individuals and a different allele in all Sb individuals , affecting 352 genes ( Supplementary file 1 ) . Among the 3 . 4% of SNPs affecting coding sequence , almost half changed the amino-acid sequence , with one change to a premature stop codon ( 47 . 7% non-synonymous vs . 52 . 3% synonymous changes ) . The remaining SNPs were in intergenic ( 36 . 1% ) , intronic ( 58 . 0% ) or in untranslated regions ( 2 . 5% ) . Because the invasive North American population went through a severe bottleneck in the 20th century ( Ascunce et al . , 2011 ) , we repeated the analysis after separating populations . We found 252 additional SNPs with fixed differences between SB and Sb individuals in South America , and 23 , 022 additional fixed differences between SB and Sb in North America . The latter number is 4-fold higher than expected due to differences in sampling size alone and is in line with lower genetic diversity of both supergene variants in North America due to the invasion bottleneck . To understand the impacts of different evolutionary processes on the supergene , we compared the expression of SB alleles and Sb alleles for all genes in the region . For this , we generated RNA-seq data from whole bodies of SB/Sb workers and from abdomens , thoraces , and heads of SB/Sb virgin queens collected in South America . To compare with patterns in other fire ant populations , we additionally incorporated existing RNA-seq gene expression data from pools of whole bodies of SB/Sb queens collected in the USA and Taiwan ( Wurm et al . , 2011; Fontana et al . , 2020 ) . We summarize our hypotheses and results in Table 1 . Among the 352 genes with fixed differences between SB and Sb in this combined dataset , 122 had sufficient expression for analysis of differences between alleles . We found that seven of the genes ( 5 . 7% ) had consistent expression differences between variants across all populations ( linear mixed-effects model; Benjamini-Hochberg ( BH ) adjusted p<0 . 05 , Figure 1 ) . Expression bias went in both directions: the Sb variants of ‘pheromone-binding protein Gp-9/OBP3’ ( LOC105194481 ) , ‘retinol-binding protein pinta-like’ ( LOC105199327 ) and uncharacterized LOC105193135 were consistently more highly expressed . In contrast , the SB variants of ‘ejaculatory bulb-specific protein 3’ ( LOC105199531 ) , ‘carbohydrate sulfotransferase 11-like’ ( LOC105193134 ) , ‘calcium-independent phospholipase A2-gamma’ ( LOC105203065 ) and uncharacterized LOC105199756 were consistently more highly expressed . We repeated our analysis in a population-specific manner . Within each of our three populations , this approach independently confirmed most of the allelic biases we had previously detected ( Figure 1—figure supplement 1 ) . It additionally uncovered population-specific patterns of allelic bias within the supergene ( Figure 1—figure supplements 2–4 , Supplementary file 3 ) . If allelic bias evolved in response to antagonistic selection , we expect it to be fine-tuned for specific functions across different tissues . In contrast , if allelic bias resulted from the accumulation of neutral or deleterious mutations , we would expect a consistent bias across tissues . To test which scenario occurred in the fire ant , we used the South American RNA-seq data . We found no effect of body part or caste on allelic expression for any gene in the supergene region ( DESeq2’s Logarithmic Ratio Test and all pairwise Wald comparisons between interaction terms; all BH adjusted p>0 . 05 ) . This result was unlikely to be due to lack of power because we did find such differences for genes in normally recombining parts of the genome , despite having less power to do so ( see Materials and methods ) . The general lack of tissue-specific fine-tuning of allelic expression bias in the supergene suggests that most of this bias is due to the accumulation of neutral or deleterious mutations ( Stolle et al . , 2019; Pracana et al . , 2017a ) in Sb rather than being adaptive . We further tested this idea by comparing expression patterns between populations . If most changes in Sb were not adaptive , patterns of allelic bias should correlate with similarity of the supergene genomic sequence . We therefore expected a stronger positive correlation of allelic bias between the two closely related invasive populations than to the less closely related South American population ( Ascunce et al . , 2011 ) . For each pair of populations , we calculated correlations of the log2 ratios between the expression levels of SB and Sb alleles ( Figure 1—figure supplement 5 ) . Our findings were in line with our expectation: correlation was stronger between invasive populations ( Spearman’s r2 = 0 . 67 ) , than between either invasive and the native population ( Spearman’s r2 = 0 . 21 and 0 . 18 ) . This result was further supported by a linear mixed-effects model of all data: population ancestry has a stronger effect on allelic expression bias than geographic proximity ( respective interaction terms with the effect of gene: F = 3 . 42 , p<10−15 and F = 0 . 94 , p=0 . 65 ) . Together , these results support the idea that the effects of suppressed recombination on genomic architecture explain most of the allelic biases in the supergene region . Social-form specific selection should lead to an overrepresentation of socially biased gene expression in the supergene region . To test whether this pattern occurs , we compared gene expression between egg-laying queens from single-queen and from multiple-queen colonies . There were 293 socially biased genes with known chromosomal locations ( Supplementary file 4 ) . Such genes were indeed overrepresented in the supergene region ( Figure 2a , 33 out of 474 , 12 expected by chance , χ2 = 29 . 7 , p<10−7 ) . Next , we examined the direction of expression bias: we found that most socially biased genes had higher expression in multiple-queen colonies than in single-queen colonies ( 274 out of 293 , i . e . , 94%; binomial test , p<10−15 ) . However , this pattern was not specific to the supergene ( χ2=1 . 04 , p=0 . 3 , Figure 2b ) . In sum , more socially biased genes are present in the supergene than in the rest of the genome , but the direction of social bias is similar across the genome . Since the trend of social bias is genome-wide , it cannot be explained by Sb degeneration alone . Gene degeneration in Sb could lead to dosage compensation . To test whether this occurs , we compared the differences in expression levels between the SB and Sb alleles within heterozygous SB/Sb individuals from multiple-queen colonies to differences in expression between queens from single-queen ( SB/SB ) and multiple-queen colonies ( SB/Sb ) . Dosage compensation should lead to a pattern where higher expression of the SB allele does not result in differences in expression between social forms . We tested whether such a pattern occurs for 294 genes in the supergene region using North American data . We compared the proportion of SB allele expression ( PB ) for each gene , with its expression in multiple-queen colonies relative to single-queen colonies ( PMQ; Figure 3 ) . If Sb degeneration had led to its global downregulation or inactivation , we would have found a stark bias towards SB , irrespective of social bias . Instead , we find that the expression levels of SB and Sb are balanced for most genes without differences between social forms ( black line in Figure 3 , linear regression passes through the point 0 . 5 , 0 . 5 with a non-significant deviation of 0 . 0058; p=0 . 32 ) . At the extremes of the distribution , however , allelic bias is higher when social bias is higher . We therefore considered two models to explain this pattern . The null model assumes that differences in expression patterns between social forms are due solely to differences in baseline allelic expression ( purple line in Figure 3 ) . In this model , the PB and PMQ values change solely as a consequence of the relative expression of the Sb and the SB alleles ( see Materials and methods ) . This model ( the purple line ) fits the data poorly . The second model ( red line ) , additionally allows for the effect of gene-specific dosage compensation by increasing expression of B-alleles in SB/Sb individuals in multiple queen-colonies . This model fits the data much better and it is significantly different from the null model ( analysis of variance p<10−5 ) . This finding confirms that SB expression in SB/Sb individuals is higher than expected given the observed patterns of expression differences between social forms . Further in line with this , we found no relation between the relative expression levels of the SB and Sb alleles in queens from multiple queen-colonies to the total expression of the genes in the supergene in either social form ( Figure 3—figure supplement 1;Wilcoxon sum rank test p>0 . 05 ) . Thus , higher SB allele expression in heterozygous individuals does not imply higher expression in single-queen colonies . To complement these analyses of general trends , we focused on individual genes . Most genes had no significant allelic or socially biased expression ( 256 , i . e . , 87%; Figure 4a and Figure 4—figure supplement 1 ) . However , fifteen of the 294 genes showed allelic bias but no differences in gene expression between social forms ( 5%; Figure 4c ) . Most of these genes ( 12 ) had higher expression in SB , with only three being more highly expressed in Sb ( binomial test , p=0 . 03 ) . In line with this result , the median expression of SB alleles of these 15 genes was 5 . 9-fold higher than that of Sb alleles ( Wilcoxon signed rank test p=0 . 0008 against equal expression ) . The general and gene-specific patterns are hallmarks of dosage compensation: Lower expression of Sb alleles is compensated for by higher expression of SB alleles , resulting in the absence of expression differences between social forms . Importantly , the overall trends and gene-specific patterns both indicate that this dosage compensation occurs in gene-specific manners rather than being a response to global Sb-inactivation . Dosage compensation would arise if gene degeneration leads to decreased Sb expression . We tested this hypothesis by asking whether genes with higher coding sequence degeneration have a higher SB bias . We used the number of non-synonymous mutations per gene in the Sb allele as a proxy for gene degeneration and compared this measure against allelic bias ( Figure 4—figure supplement 2 ) . Indeed , as the number of non-synonymous mutations increases , so does the allelic bias towards SB ( coefficient = 0 . 052 , p=0 . 01 ) . This indicates that coding-sequence degeneration could lead to lower expression , or alternatively that some genes generally degenerate faster than others . Antagonistic selection should lead to an enrichment of genes in the supergene that are highly expressed in multiple-queen colonies and show allelic bias towards Sb . In contrast , without antagonistic selection all expression differences between Sb and SB alleles should be due to gene degeneration , and lead to lower Sb expression levels ( Ma et al . , 2020 ) . From the 294 genes analyzed in the previous section , eight ( 3% ) had both allele-biased and socially-biased expression ( Figure 4d ) . Their expression patterns were strongly directionally biased towards higher expression in Sb and in multiple-queen colonies ( 5 out 8 Sb biased genes were significantly more highly expressed in multiple-queen colonies Figure 4d; compared to 1 out of 15 for SB biased genes , χ2 = 5 . 8 , p=0 . 02; Figure 4c ) . This enrichment of multiple-queen biased genes in the Sb variant is consistent with antagonistic selection . Although unlikely , Sb-specific upregulation could lead to higher expression in multiple-queen colonies without affecting fitness . However , the broad expression patterns described in the previous section ( Figure 3 ) show that differences in expression between social forms are not due exclusively to changes in expression levels between Sb and SB alleles . We additionally showed that the fixation of non-synonymous mutations in Sb alleles correlates with lower expression levels ( Figure 4—figure supplement 2 ) . We therefore consider it unlikely that the trend of enrichment in high multiple-queen expression among Sb biased genes would have arisen neutrally . Consistent with the lack of direct correlation between allelic bias and social bias , only 3 of the seven genes with consistent allelic bias in all populations were also differentially expressed between social forms ( Supplementary file 2 ) : ‘pheromone-binding protein Gp-9’ ( LOC105194481 ) , ‘ejaculatory bulb-specific protein 3’ ( LOC105199531 ) and ‘retinol-binding protein pinta-like’ ( LOC105199327 ) . This narrow overlap between allelic and social bias makes these genes candidates for playing roles in phenotypic differences between social forms .
We found that a small proportion of the genes in the supergene region showed consistent allele-specific expression differences between the SB and Sb variants . It is tempting to conclude that such gene expression differences arose through selection , as a consequence of evolutionary antagonism between the single-queen and multiple-queen phenotypes . However , this interpretation may be too simplistic , as it ignores the impacts of supergene degeneration . Several studies have shown that Sb is degenerating ( Wang et al . , 2013; Stolle et al . , 2019; Pracana et al . , 2017a ) . Our observation that roughly half of the fixed differences in coding sequence between SB and Sb impact the protein sequence ( where they are likely to have a deleterious effect ) is also consistent with degeneration of the Sb variant . Such sequence-level degeneration is a symptom of reduced selection efficacy . By examining gene expression , we revealed three additional symptoms of degeneration and low selection efficacy . First , the absence of tissue-specificity in our study would not be expected if expression differences were adaptive . For some genes , the expression differences between Sb and SB are likely due to mutations that are completely neutral or deleterious . However , for other genes , expression differences could be partially adaptive , but low selection efficacy may have hindered the fine-tuning of their expression during the short timespan since the supergene’s emergence ( Wang et al . , 2013 ) . As a result , strong selection for a particular level of allele-specific expression in one body part ( e . g . , in antennae ) , could result in consistent allele-specific expression patterns across tissues , even if this has mildly deleterious effects ( e . g . , in the gut ) . Second , there was a strong correlation in allelic bias between the invasive North American and Taiwanese populations , despite the data from the two populations being from different studies . Both invasive populations have lower genetic diversity overall ( Ascunce et al . , 2011 ) , and in the supergene region in particular ( Pracana et al . , 2017a ) . The strong effect of ancestry on allelic bias indicates that genomic architecture , rather than gene function , defines most expression patterns within the supergene . Finally , gene degeneration can result in lower expression levels ( Xu et al . , 2019; Pucholt et al . , 2017 ) . For example , gene expression is reduced in genes with sequence-level signatures of degeneration in the mating-type chromosomes of the anther-smut fungus Microbotryum ( Ma et al . , 2020 ) . We similarly find that Sb alleles with more non-synonymous mutations tend to have lower expression than SB alleles . This pattern could result from the relative inefficacy of selection on Sb , leading to selection favoring the downregulation of alleles accumulating detrimental mutations . We find such patterns of degeneration despite the likely effects of antagonistic selection ( Huang and Wang , 2014 ) and the lack of evolutionary strata ( Pracana et al . , 2017a ) , both of which are likely to dampen the degeneration signal ( Ma et al . , 2020 ) . The reduction in expression of the Sb allele in several genes could result in detrimental fitness effects if the genes involved are dosage sensitive , as observed in many sex chromosomes ( Mank , 2013 ) . We tested this idea in the fire ant supergene and found that allelic expression bias is relatively balanced , with similar levels of SB and Sb bias across the supergene . However , far more genes have multiple-queen biased expression than single-queen biased expression . This pattern implies that much of the observed SB bias leads to no differences between social forms . Our findings are consistent with dosage compensation , where the higher SB expression effectively counteracts lower Sb expression . Some of this dosage compensation likely emerged through selection despite Hill-Robertson effects and a short time of divergence ( Pracana et al . , 2017a ) . However , some dosage compensation could instead occur automatically , whereby transcriptional machinery is less able to bind to degenerate regions of Sb and thus binds to SB instead ( Teufel et al . , 2019 ) . Additionally , we speculate that regulatory elements located outside the supergene region could have co-evolved with Sb degeneration , allowing for variant-specific expression regulation . These elements would not be affected by suppressed recombination , allowing for a quicker emergence of dosage controlling mechanism ( Lenormand et al . , 2020 ) . Regardless of the mechanisms which mediate gene-by-gene dosage compensation in the fire ant , we show that it can emerge over time scales as short as approximately 175 , 000 generations ( 1 million years ) ( Wang et al . , 2013; Pracana et al . , 2017a ) . Furthermore , this is to our knowledge only the second known instance of dosage compensation in a supergene that does not determine sex or mating type . Indeed , a 2–3 million-year-old supergene controlling color morphs of the white-throated sparrow also shows patterns consistent with dosage compensation ( Sun et al . , 2018 ) . Such findings support the idea that many of the patterns seen in sex chromosomes are representative of supergenes in general . Indeed , rapid evolution of dosage compensation similarly occurred in the 10 million-year-old sex chromosomes of the plant Silene latifolia ( Muyle et al . , 2012 ) and in two Drosophila species neo-sex chromosomes that are only a few million years old ( Nozawa et al . , 2014; Alekseyenko et al . , 2013; Nozawa et al . , 2018 ) . Most Sb-biased genes are also more highly expressed in multiple-queen colonies , supporting the idea that antagonistic alleles are present in the supergene . In the unlikely case that this pattern had no fitness effects , it could arise neutrally ( Harrison et al . , 2012 ) . We argue against this possibility because , as discussed above , expression patterns in the supergene do not follow what would be expected if differences between social forms were due exclusively to differences in allelic expression . Additionally , the patterns of expression differences between social forms are similar within the supergene and in the rest of the genome . This observation further suggests that expression differences between social forms are driven by factors unrelated to the genomic architecture of the supergene . Given the patterns of ongoing degeneration in Sb , we conclude that genes with Sb and multiple-queen bias were likely under antagonistic selection . Similarly , the neo-Y chromosome of Drosophila miranda is enriched in male-biased genes in the gonads ( Zhou and Bachtrog , 2012 ) . The approaches underpinning our analyses are unable to detect allelic differences in genes absent from the reference genome such as OBP-Z5 , a putative Odorant Binding Protein exclusive to Sb ( Pracana et al . , 2017b ) . However , our analysis does single out candidate genes that potentially contribute to the social polymorphism of the fire ant ( lists of all sequence and expression differences are in Supplementary files 1 , 2 , 3 , 4 ) . In particular , three genes stood out because they were differentially expressed between social forms and had variant-specific allele expression in all populations . For the first gene , ‘Pheromone-binding protein Gp-9’ ( LOC105194481 ) , also known as OBP-3 , the Sb allele was more highly expressed . For decades , this gene has been a candidate effector for social form differences ( Pracana et al . , 2017a; Keller and Ross , 1998 ) , yet its linkage to hundreds of other genes in the supergene led to doubts that its association to social form is any more than coincidental . We found five fixed differences between SB and Sb for this gene , four of which could affect protein efficiency ( consistent with previous findings [Krieger and Ross , 2002] ) . For the second gene , ‘Ejaculatory bulb-specific protein 3’ ( LOC105199531 ) , which also contains an insect odorant binding protein domain ( InterPro IPR005055 ) , the SB allele was more highly expressed . Orthologs of this gene are associated with mating ( Laturney and Billeter , 2014 ) in Drosophila melanogaster , sexual behavior in a moth ( Bohbot et al . , 1998 ) , subcaste differences in bumblebees ( Wolschin et al . , 2012 ) , venom production in social hornets ( Yoon et al . , 2015 ) and caste differences in the termite Reticulitermes flavipes ( Steller et al . , 2010 ) . Finally , LOC105199327 is likely a Pinta retinol-binding protein . Such proteins help pigment transport and vision in D . melanogaster and the butterfly Papilio xuthus ( Pelosi et al . , 2018 ) . In sum , all three candidate genes have putative functions related to environmental perception , in line with the complex social phenotype requiring subtle changes in environmental perception or signaling ( Favreau et al . , 2018 ) . We found differences in expression between the SB- and Sb-linked alleles of genes in the fire ant supergene across three populations . Such strong patterns can naively be assumed to be indicative of adaptive processes emerging from evolutionary antagonism between social forms . However , we show that the evolutionary forces shaping expression patterns in the supergene are complex and must be interpreted with care . In particular , genes with higher expression of the SB allele than the Sb allele tend to either lack expression differences between social forms or have lower expression in multiple-queen than in single-queen colonies . Both patterns are consistent with the idea that suppressed recombination leads to degeneration in Sb and thus lower Sb allele expression . In some cases , a dosage compensation mechanism through higher expression of the healthy SB allele leads to similar expression levels in both social forms . In cases where no dosage compensation occurs , overall expression is lower in multiple-queen colonies than in single-queen colonies . Conversely , we also show that genes with higher expression of the Sb allele than the SB allele are also biased towards higher expression in multiple-queen colonies . This pattern is consistent with evolutionary antagonism favoring the accumulation of beneficial alleles for multiple-queen individuals in Sb , the supergene variant found only in this social form . Our study shows that multiple complex evolutionary forces can simultaneously act on a young supergene system . It highlights that allele-specific expression patterns alone are insufficient for inferring whether they are adaptive , deleterious or phenotypically neutral . Instead , putting such expression differences into broader contexts is needed to draw reasonable conclusions . Applying this idea to our data highlights genes which have molecular roles that could affect perception or signaling of the social environment .
We used three published and one new RNA-seq gene expression datasets from fire ants . Wurm et al . , 2011 obtained whole-body RNA-seq data from six pools of 4 egg-laying SB/Sb queens , each from a multiple-queen colony from Georgia , USA . Morandin et al . , 2016 obtained whole-body RNA-seq data from six samples , each being a pool of 3 queens from a single-queen or a multiple-queen colony ( three replicates per social form ) from Texas , USA . All the queens were mature and egg-laying ( C . Morandin , personal communication ) , thus queens from multiple-queen colonies carried the SB/Sb genotype ( Keller and Ross , 1998 ) . Additionally , we manually checked the resulting RNA reads for heterozygous positions at key supergene markers using the IGV gene browser v2 . 4 . 19 ( Robinson et al . , 2011 ) . Fontana et al . , 2020 generated RNA-seq data from 4 samples of SB/Sb queens from multiple-queen colonies from Taiwan . Each sample is a pool of whole bodies from two virgin queens ( more details in Supplementary file 5 ) . All three published datasets are from pools of whole bodies from the invasive ranges of S . invicta . The Taiwanese invasive population of red fire ants is derived from that of North America ( Ascunce et al . , 2011 ) . Because comparisons of whole bodies can be confounded by allometric differences ( Johnson et al . , 2013 ) and genetic diversity is reduced among Sb haplotypes in the invasive populations ( Pracana et al . , 2017a ) , we generated a new gene expression dataset . We collected samples from the native South American range of this species to obtain molecular data ( ( collection and exportation permit numbers 007/15 , 282/2016 , 433/02101-0014449-4 and 25253/16 ) . To obtain RNAseq , we collected six multiple-queen colonies . We confirmed the social form of each colony ( Krieger and Ross , 2002 ) on a pool of DNA from 10 randomly chosen workers . Colonies spent six weeks under semi-controlled conditions before sampling ( natural light , room temperature , cricket , mealworm and honey water diet ) . From each colony , we snap-froze one worker and one unmated queen for gene expression analysis between 12:00 and 15:00 local time . To partly control for allometric differences between genotypes , we separated each queen into head , thorax and abdomen . This was done in petri dishes over dry ice using bleached tweezers . In total , we had 24 samples for RNA extraction: six whole bodies of workers and six replicates of three body parts from queens ( more details in Supplementary file 5 ) . We extracted RNA and DNA from each sample using a dual DNA/RNA Tri Reagent based protocol ( https://www . protocols . io/view/rna-dna-extraction-protocol-bi8fkhtn ) . We applied the Krieger and Ross , 2002 assay on the extracted DNA to identify only individuals with the SB/Sb genotype . Once RNA was extracted , we prepared Illumina sequencing libraries from total RNA using half volumes of the NEBNext Ultra II RNA Library Prep Kit . We checked RNA and library qualities on an Agilent Tapestation 2200; library insert size averaged 350 bp . An equimolar pool of the 24 libraries was sequenced on a single lane of Illumina HiSeq 4000 using 150 bp paired-end reads . This produced an average of 14 , 848 , 226 read pairs per sample ( maximum: 27 , 766 , 980; minimum: 6 , 015 , 662 . Raw RNA-seq reads for all samples are on NCBI SRA ( PRJNA542606 ) . For all datasets , we assessed read quality using fastQC ( v0 . 11 . 5; http://www . bioinformatics . babraham . ac . uk/projects/fastqc/ ) . Raw reads for all samples were of sufficient quality to be used in subsequent analysis . We removed low quality bases using fqtrim with default parameters ( v0 . 9 . 5; http://ccb . jhu . edu/software/fqtrim/ ) , and Illumina adapters using Cutadapt v1 . 13 ( Martin , 2011 ) . We then generated a STAR v2 . 5 . 3a ( Dobin et al . , 2013 ) index of the S . invicta reference genome ( version gnG; RefSeq GCF_000188075 . 1 [Wurm et al . , 2011] while providing geneset v000188075 . 1 in GFF format through the ‘sjdbGTFtagExonParentTranscript = Parent’ option . As recommended by the developers of STAR , we aligned each sample to the reference twice , using the ‘out . tab’ file for the second run , and set ‘sjdbOverhang’ to the maximum trimmed read length minus one , that is 74 for the Wurm et al . , 2011 data and Morandin et al . , 2016 data , 125 for the Fontana et al . , 2020 data and 149 for the South American data we generated here . Alignments were run using GNU Parallel v20150922 ( Tange , 2011 ) . All steps and downstream analyses were performed on the Queen Mary University of London’s Apocrita High Performance Computing Cluster ( King et al . , 2017 ) . We further assessed aligned reads ( i . e . , BAM files ) using MultiQC v1 . 5 ( Ewels et al . , 2016 ) and the BodyGene_coverage . py script of the RSeQC toolkit v2 . 6 . 4 ( Wang et al . , 2016 ) . We removed one sample from multiple-queen colonies in the Morandin et al . data from subsequent analyses due to poor alignment quality . None of the other BAM files showed markers of technical artefacts that could bias our results . To detect allele-specific differences between SB and Sb we first identified SNPs with fixed differences between the SB and Sb variants . Because the patterns of genetic diversity differ between the invasive and South American S . invicta populations ( Ross et al . , 2007; Ahrens et al . , 2005 ) , we estimated allele specific expression differences in the social chromosome independently for each population . For this we used haploid male ants because they can provide unambiguous genotypes . For the invasive populations , we identified fixed allelic differences between a group of 7 SB males and a group of 7 Sb males from North America ( NCBI SRP017317 ) ( Wang et al . , 2013 ) . For the South American population , we sequenced the genomes of 13 SB males and 13 Sb males sampled from across Argentina . For each individual , we extracted 1 µg of genomic DNA using a phenol-chloroform protocol . The extracted material was sheared to 350 bp fragments using a Covaris ( M220 ) . We constructed individually barcoded libraries using the Illumina TruSeq PCR-free kit . The libraries were quantified through qPCR ( NEB library quant kit ) . An equimolar pool of the 26 libraries was sequenced on a HiSeq4000 at 150 bp paired reads . This produced an average of 17 , 790 , 416 pairs of reads per sample , with a maximum of 38 , 823 , 285 and a minimum of 7 , 910 , 042 ( Supplementary file 6 , genomic reads of all samples deposited on NCBI SRA ( PRJNA542606 ) ) . For each dataset , we identified fixed allelic differences between the group of SB males and the group of Sb males . We first aligned the reads of each sample to the S . invicta reference genome ( Wurm et al . , 2011; gnG assembly; RefSeq GCF_000188075 . 1 ) using Bowtie2 v2 . 3 . 4 ( Langmead et al . , 2009 ) . We then used Freebayes v1 . 1 . 0 ( Garrison and Marth , 2012 ) to call variants across all individuals ( parameters: ploidy = 1 , min-alternate-count=1 , min-alternate-fraction=0 . 2 ) . We used BCFtools ( Li et al . , 2009 ) and VariantAnnotation ( Obenchain et al . , 2014 ) to only retain variant sites with single nucleotide polymorphisms ( SNPs ) , with quality value Q greater than or equal to 25 , and where all individuals had a minimum coverage of 1 . To avoid considering SNPs erroneously called from repetitive regions that are collapsed in the reference genome , we discarded any SNP with mean coverage greater than 16 for the North American samples or 12 for the South American samples or where any individuals had less than 60% reads supporting the called allele . This last filtering step also acts to remove SNPs called from reads with sequencing errors . We then extracted only the SNPs located within the supergene ( based on the genomic locations from Pracana et al . , 2017a ) and with fixed differences between SB and Sb . This step was performed independently for each population . The two resulting variant call files were inspected using VCFtools v0 . 1 . 15 ( Danecek et al . , 2011 ) and we manually ensured that all variants had the SB allele as reference and Sb allele as alternative . To test the effect of sample size differences between populations we downsampled the South American dataset to 7 pairs of SB and Sb males , matching the sample size in the North American dataset . We extracted SNPs shared between South and North American populations using BCFtools isec v1 . 9 ( Li et al . , 2009 ) . We then used SNPeff ( Cingolani et al . , 2012 ) to characterize the effects of individual SNPs . Because the reference genome for S . invicta is based on an SB individual , read mapping could be biased towards the SB variant in heterozygous individuals , resulting in false positive detection of allelic bias ( Castel et al . , 2015 ) . To overcome this potential artifact , we called BCFtools consensus v1 . 9 ( Li et al . , 2009 ) once using North American Sb males and once using South American Sb males . We then aligned the RNAseq reads from each sample to the regular reference genome ( version gnG; RefSeq GCF_000188075 . 1 ) and also , independently , to the most relevant of the modified references . For the reads from the Taiwanese population , we used the Sb reference using North American SNPs . For the alignment we used STAR with the same parameters as described above . We merged the two resulting BAM files from each sample using SAMtools v1 . 9 ( Li et al . , 2009 ) . We then used the ‘rmdup’ function from the WASP pipeline ( Soneson et al . , 2015 ) to generate reference-bias free alignment files . The resulting BAM files can be considered reference bias free alignments . We added a reading group ID to each reference-bias free BAM file using the ‘AddOrReplaceReadGroups’ tool from Picard ( v 2 . 7 . 0-SNAPSHOT; http://broadinstitute . github . io/picard/ ) . We then ran all BAM files through GATK’s ‘ASEReadCounter’ v 3 . 6–0-g89b7209 ( Wright et al . , 2017 ) with default options to obtain read counts for each allele . We performed this step once on each population independently . We then imported the resulting allele-specific SNP read counts per sample generated by GATK into R v3 . 4 . 4 ( R Development Core Team , 2017 ) . We used the R packages ‘GenomicRanges’ v1 . 26 . 4 ( Lawrence et al . , 2013 ) and ‘GenomicFeatures’ v1 . 26 . 3 ( Lawrence et al . , 2013 ) along with the NCBI protein-coding gene annotation for S . invicta to identify which SNPs are in which genes . We estimated the total expression level for a particular allele ( i . e . , the SB or Sb variant for any given gene ) as the median of all SNP-specific read counts per gene and per variant . For instance , consider a gene with three fixed SNPs between SB and Sb for which the SB variants have support from 12 , 15 and 18 reads , and the Sb variants from 5 , 8 and 6 reads . In this particular case , we would report that the SB variant for this gene has an expression level of 15 reads and the Sb variant , six reads . If instead of this approach , we randomly select one of the possible SNPs for every gene , we find qualitatively similar results to those reported . Additionally , to test whether we would be able to detect allele-specific expression changes across body parts and castes in the South American data , we calculated allele-specific expression in the whole genome as a positive control . We used the VCF file containing all SNPs in the 26 males collected from South America . We retained only SNPs with expression data in all samples and a median of at least 1 X RNA coverage in each allele across all samples . After filtering , 1096 SNPs remained for which we were able to test for allele-specific expression . We performed an allele specific expression analysis throughout the whole genome using body part and caste information from South American populations . Unlike the analysis of genes in the supergene region , in the whole genome analysis we cannot ensure that every individual is heterozygous for all SNPs . Indeed , the average frequency in the population for all the alleles analyzed was 0 . 41 with a standard deviation of ±0 . 2 . This implies that both alleles were not necessarily present in all samples . We therefore had far less power to detect allele-specific expression across body parts using data from the whole genome than using SNPs from the supergene region only . Despite this lack of power , we were able to detect significant ( Wald test BH adjusted p<0 . 05 ) allele-specific expression changes across body parts of queens and workers in 15 SNPs . These significant SNPs were distributed across nine genomic scaffolds . The significant differences in allele-specific expression were between a queen body part and whole bodies of workers . We imported the estimated read counts generated by Kallisto into R using Tximport v1 . 2 . 0 ( Soneson et al . , 2015 ) and DESeq2 v1 . 14 . 1 ( Love et al . , 2014 ) . For every sample , read counts for the SB alleles and for the Sb alleles come from the same sequencing library , thus standard normalization methods ( Dillies et al . , 2013 ) are not applicable . As recommended by the developers of DESeq2 ( Love , 2018 ) , we thus deactivated normalization by setting SizeFactors = 1 . For the North American and Taiwanese datasets ( Wurm et al . , 2011; Fontana et al . , 2020 ) , we only considered genes expressed in all samples for downstream analyses , whereas for the South American populations RNA dataset , we only analyzed genes expressed in all replicates of at least one body part . To have the strongest possible analysis of expression between the SB and Sb variants of the supergene region , we performed a joint analysis of RNAseq data from Taiwanese , South and North American populations . The South American dataset includes body part information , which is absent in the North American dataset . We applied a linear mixed effects model on the log2 of the expression ratios between SB and Sb across populations and body parts , using the R packages lme4 v1 . 1–18 . 1 ( Bates et al . , 2014 ) and lmerTest v3 . 1–0 ( Kuznetsova et al . , 2017 ) . We fitted the log2 expression ratios using a 0 intercept with gene , population and their interaction as fixed effects , and the interaction between gene and body part as random effects ( formula: log2 expression ratio ~0 + gene * population + ( 1|body part:gene ) ) . We also performed an additional linear mixed-effects model to test the effect of geography and ancestry on the allele-specific expression patterns within the supergene . We grouped the populations by geographical proximity ( North and South America vs . Taiwan ) or by phylogenetic proximity ( Taiwan and North America vs . South America ) . We then fitted the log2 expression ratios using a 0 intercept , the main effects of gene , ancestry and geographic proximity and the interactions between ancestry and gene and between geographic proximity and gene as fixed effects . As random effects we used again the interaction between gene and body part ( formula: log2 expression ratio ~0 + gene * ancestry + gene * geography + ( 1|body_part:gene ) ) . We then performed an analysis of variance on the model to estimate the size effects of each term . For both models , the log2 expression ratios were weighed by a function of the total read counts per gene to reduce the impacts of genes with low expression which have extremely high variance . Here we only report the results of the fixed effects per gene after adjustment of p values for multiple testing following the Benjamini-Hochberg approach ( Benjamini and Hochberg , 1995 ) . For this joint analysis we only used genes that had fixed differences between SB and Sb in all three populations . We additionally analyzed the allele-specific expression patterns between the SB and Sb variants of the supergene in each population independently using DESeq2 ( Love et al . , 2014 following Castel et al . , 2015 . The model formula used for the South American RNA-seq data used ‘body part’ and ‘colony of origin’ as blocking factors , and allele-specific expression , that is ‘variant effect’ , as variable of interest . This analysis allowed us to detect differences in expression between variants specific to body part . Preliminary analyses showed that the interaction between ‘variant effect’ and ‘body part’ had no significant effect in any of the genes , and consequently , only the main ‘variant effect’ was considered as the factor of interest for this analysis . The model formula for both the Wurm et al . , 2011 and the Fontana et al . , 2020 RNA-seq datasets included only whole bodies of queens . We thus used ‘sample’ as a blocking factor and ‘variant effect’ as variable of interest . In all analyses , we report gene differences between variants as log2 expression ratios between the SB and the Sb counts . That is , genes with expression biased towards SB will produce positive log2 expression ratios whereas those biased towards Sb will produce a negative value . To check whether there was an overall bias towards either variant , we tested the significance of the deviation from 0 for the median log2 expression ratios between SB and Sb via a Wilcoxon sum rank test . We determined the expression levels for all samples from the North American populations ( Morandin et al . , 2016 ) by using the count mode in Kallisto v0 . 44 . 0 ( Bray et al . , 2016 ) using S . invicta coding sequences . We imported the estimated counts into DESeq2 v1 . 14 . 1 ( Love et al . , 2014 ) using Tximport v1 . 2 . 0 ( Soneson et al . , 2015 ) . We compared the DESeq2 normalized expression levels between social forms , determining significance of differential expression using the default Wald test for pairwise comparisons between genes . We estimated the proportion of significantly differentially to non-differentially expressed genes within and outside the supergene region based on supergene region coordinates from Pracana et al . , 2017a . We then used the R packages GenomicRanges and GenomicFeatures ( Lawrence et al . , 2013 ) along with the annotations of S . invicta coding sequences to locate each gene with expression information in the genome . Our analyses are restricted to the 10 , 481 known S . invicta genes that can be reliably placed within or outside the supergene region; other genes are on scaffolds which lack chromosomal locations ( Pracana et al . , 2017a ) . We fitted a model to test whether there is a significant relationship between allele-specific expression differences between supergene variants in the Wurm et al . , 2011 dataset , and gene expression differences between social forms ( log2 expression ratios using the Morandin et al . , 2016 dataset ) . We examined the overall trend in allele-specific expression patterns within the supergene ( i . e . , any bias towards expression of either the SB or Sb allelic variant ) . We obtained relative expression levels using DESeq2 for both comparisons: single-queen vs . multiple-queen expression for each gene ( XSQ vs . XMQ ) from the Morandin et al . , 2016 dataset and expression of the SB allelic variant vs . the Sb ( XB vs . Xb ) within each gene from the Wurm et al . , 2011 dataset DESeq2 returned an estimate of log2 ( XB/Xb ) for the differences in expression between alleles and log2 ( XSQ/XMQ ) for the differences among colony types . We first generated a null model in which the PB and PMQ values change solely as a consequence of the relative expression ( r ) of the Sb allele ( xb ) and the SB alleles ( xB ) , such that xb = r xB , and therefore:PB=XbXb+XBandPMQ=Xb+XBXb+3XB Notice that in this case the minimum value of PMQ = ⅓ would occur when there was no expression of the b allele ( xb = 0 ) . For greater values of xb , we can solve the pair of equations to obtain the relationship:PMQ=12PB+1 The second model additionally allows for the effect of dosage compensation by increasing expression of B-alleles in SB/Sb individuals in multiple queen-colonies . To do this , PB is defined by the expression differences between social forms such that:PB=1−PMQ2PMQ Finally , we test whether the two models are significantly different using a standard analysis of variance test using the aov function in R . The linear regressions and statistical tests were performed in R v3 . 4 . 4 ( Lawrence et al . , 2013 ) . We also explored whether genes with low Sb allele expression had higher SB allele expression , resulting in similar expression between multiple-queen ( SB/Sb genotype ) and single-queen ( SB/SB genotype ) individuals . Such a pattern would be consistent with an ongoing process of dosage compensation . To do so , we excluded the nine genes with significant biases towards Sb and high SB-SB/Sb ratios ( i . e . , SB variant more highly expressed in SB/Sb than SB/SB individuals ) , since they are more likely to have been subjected to antagonistic selection . We also excluded genes with fewer than three read counts mapping to either allele to remove more noisy estimates . The rest of all analyzed genes were then grouped by relative SB/Sb expression . We then compared the overall expression levels between these groups in multiple-queen and single-queen individuals . We deposited the genomic and transcriptomic reads we generated from South American Solenopsis invicta on NCBI SRA ( PRJNA542606 ) . All analysis scripts used will be made available at https://github . com/wurmlab/2019-11-allelic_bias_in_fire_ant_supergene ( copy archived at https://github . com/elifesciences-publications/2019-11-allelic_bias_in_fire_ant_supergene; Martinez-Ruiz , 2020 ) . | Red fire ants ( Solenopsis invicta ) are native to South America , but the species has spread to North America , Australia and New Zealand where it can be an invasive pest . A reason for this species’ invasiveness types of colonies : one with a single egg-laying queen and another with several queens . However , it is not possible to simply add more queens to a colony with one queen . Instead , the number of queens in a colony is controlled genetically , by a chromosome known as the ‘social chromosome’ . Like many other animals , red fire ants are diploid: their cells have two copies of each chromosome , which can carry two different versions of each gene . The social chromosome is no different , and it comes in two variants , SB and Sb . Each ant can therefore have either two SB chromosomes , leading to a colony with a single queen; or one SB chromosome and one Sb chromosome , leading to a colony with multiple queens . Ants with two copies of the Sb variant die when they are young , so the Sb version is inherited in a similar way to how the Y chromosome is passed on in humans . However , the social chromosome in red fire ants appeared less than one million years ago , making it much younger than the human Y chromosome , which is 180 million years old . This makes the social chromosome a good candidate for examining the early evolution of special chromosome variants that are only inherited . How differences between the SB and the Sb chromosomes are evolving is an open question , however . Perhaps each version of the social chromosome has been optimised through natural selection to one colony type . Another suggestion is that the Sb chromosome has degenerated over time because its genes cannot be ‘reshuffled’ as they would be on normal chromosomes . Martinez-Ruiz et al . compared genetic variants on the SB and Sb chromosomes , along with their expression in different types of ant colonies . The analysis showed that the Sb variant is in fact breaking down because of the lack of gene shuffling . This loss is compensated by intact copies of the same genes found on the SB variant , which explains why ants with the Sb variant can only survive if they also carry the SB version . Only a handful of genes on the social chromosomes appear to have been optimised by natural selection . Therefore Martinez-Ruiz et al . concluded the differences between the two chromosomes that lead to different colony types are collateral effects of Sb’s inability to reshuffle its genes . This work reveals how a special chromosome similar to the Y chromosome in humans evolved . It also shows how multiple complex evolutionary forces can shape a species’ genetic makeup and social forms . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"chromosomes",
"and",
"gene",
"expression",
"evolutionary",
"biology"
] | 2020 | Genomic architecture and evolutionary antagonism drive allelic expression bias in the social supergene of red fire ants |
Septins are guanine nucleotide-binding proteins that polymerize into filamentous and higher-order structures . Cdc42 and its effector Gic1 are involved in septin recruitment , ring formation and dissociation . The regulatory mechanisms behind these processes are not well understood . Here , we have used electron microscopy and cryo electron tomography to elucidate the structural basis of the Gic1-septin and Gic1-Cdc42-septin interaction . We show that Gic1 acts as a scaffolding protein for septin filaments forming long and flexible filament cables . Cdc42 in its GTP-form binds to Gic1 , which ultimately leads to the dissociation of Gic1 from the filament cables . Surprisingly , Cdc42-GDP is not inactive , but in the absence of Gic1 directly interacts with septin filaments resulting in their disassembly . We suggest that this unanticipated dual function of Cdc42 is crucial for the cell cycle . Based on our results we propose a novel regulatory mechanism for septin filament formation and dissociation .
Septins are ubiquitous guanine nucleotide-binding proteins that have been implicated in many cellular processes such as cytokinesis , spindle positioning , morphogenesis , and exocytosis , and their mutation or overexpression is associated with neoplasia , neurodegenerative diseases and male infertility ( Hall and Russell , 2004 ) . In yeast , four essential septins ( Cdc3 , Cdc10 , Cdc11 and Cdc12 ) are found at the bud neck ( Haarer and Pringle , 1987; Ford and Pringle , 1991; Kim et al . , 1991 ) , where they form an ordered ring composed of membrane-adjacent filaments ( Hartwell , 1971; Byers and Goetsch , 1976 ) . In total seven different septins were identified in S . cerevisiae , where they form filaments of variable size and combinations . Whereas the human genome encodes thirteen septins , C . elegans has only two and plants are devoid of septin genes ( Hall and Russell , 2004; Ihara et al . , 2005; Kinoshita , 2006 ) . Despite the genetic variability , all septins share defined structural features . A recent crystallographic study on the human SEPT2-SEPT6-SEPT7 complex has shed light on the structural organization of human septins at the atomic level , which differs profoundly from that of other cytoskeletal structures ( Sirajuddin et al . , 2007 , 2009 ) . Septins interact via their central guanine nucleotide-binding domains ( G-domains ) and/or the N- and C-terminal extensions forming oligomers and non-helical filaments . The basic structural unit of the yeast septin complex is an octamer , composed of four subunits , namely Cdc10 , Cdc3 , Cdc12 and Cdc11 , arranged into two tetramers with two-fold rotational symmetry ( Bertin et al . , 2008 ) . Cdc42 has been identified as a central regulator of septin ring assembly and disassembly during different stages of the cell cycle ( Gladfelter et al . , 2002; Kozminski et al . , 2003 ) . Mutations that affect the GTPase activity of Cdc42 impair the initial assembly of septin rings , while after bud emergence , septin rings are maintained independently of Ccd42 ( Gladfelter et al . , 2002 ) . It was also reported that the activity of Cdc42’s guanine nucleotide exchange factor ( GEF ) and GTPase activating protein ( GAP ) are required for proper septin ring formation and localization , implying that one or more cycle ( s ) of nucleotide binding and hydrolysis are required for Cdc42 at the beginning of budding ( Gladfelter et al . , 2002; Caviston et al . , 2003 ) . Among the essential effectors of Cdc42 in yeast are the two structurally homologous proteins Gic1 and Gic2 , which are functional homologues of the human Borg protein ( Joberty et al . , 2001; Sheffield et al . , 2003 ) . It has been shown that Gic1 and Gic2 play an essential and overlapping role in cytoskeletal polarization ( Brown et al . , 1997; Hall and Russell , 2004 ) and septin recruitment ( Iwase et al . , 2006 ) . However , the complex interplay between Cdc42 , Gic1 and septins at the molecular level and its role during the cell cycle is not yet understood . In this study , we have used electron microscopy and cryo electron tomography ( cryo-ET ) to describe the structural basis for the direct interaction of Gic1 and Cdc42 with septin filaments . Gic1 interacts with Cdc10 subunits of adjacent septin filaments and cross-links them . Because of this scaffolding , septin filaments are stabilized and form long railroad-like ordered filament cables . Cdc42-GTP directly binds to Gic1 and at higher concentrations inhibits the Gic1 interaction with Cdc10 , resulting in the dissociation of the Gic1-septin complex . In its GDP-state , however , in absence of Gic1 Cdc42 interacts directly with Cdc10 and thereby mechanically disassembles septin filaments . Gic1 and Cdc42-GDP therefore compete for the same septin subunit . Finally , based on our results we propose a novel regulatory mechanism for septin ring formation and dissociation involving Cdc42 and Gic1 .
EGFP-labeled septins without Gic1 ( Cdc3-EGFP , Cdc10 , Cdc11 and Cdc12 ) form relatively short and straight filaments ( Figure 1A ) . Interestingly , when Gic1 is added during septin polymerization , long filaments that cluster together in large bundles are formed ( Figure 1B ) . Studying the same but not EGFP-labeled samples using electron microscopy ( EM ) , we found that in contrast to blank septin polymers that form long , often pairwise arranged filaments ( Figure 1C ) , Gic1-septin complexes display a regular railroad-like structure with many cross-linked filaments bundled together ( Figure 1D ) . Gic1 forms cross-bridges between at least two filaments , keeping them at a distance of about 20 nm ( Figure 1E ) . At each cross-bridge , Gic1 binds to at least two adjacent septin subunits on each filament ( Figure 1F ) , leaving a gap of six free subunits between individual Gic1 molecules ( Figure 1G ) . The structure of Gic1 is not well defined ( Figure 1F ) indicating that Gic1 is flexible and oriented differently at each cross-bridge . 10 . 7554/eLife . 01085 . 003Figure 1 . Gic1 scaffolds septin filaments resulting in long and flexible filament cables . ( A and B ) Yeast septin octamers containing Cdc3-EGFP polymerized by dialysis alone ( A ) or together with Gic1 ( B ) and imaged using fluorescence microscopy . Scale bar , 0 . 5 µm . ( C and D ) Representative EM image of negatively stained septin filaments ( C ) and septin-Gic1 complexes ( D ) without EGFP . Scale bar , 100 nm . ( E–G ) Representative class averages with focus on the overall structure of the septin-Gic1 complex ( E ) , the Gic1 cross-bridges ( F ) and the septin filaments ( G ) ; arrows indicate single septin proteins . Scale bars , 10 nm . DOI: http://dx . doi . org/10 . 7554/eLife . 01085 . 003 To determine which septin subunit interacts with Gic1 , we labeled septin-Gic1 complexes with antibodies against Cdc11 and observed that it sits exactly in the middle of a septin filament between two Gic1 cross-bridges ( Figure 2A , B ) . Based on the known sequential order of septin filaments , this suggests that Gic1 binds to Cdc10 ( Figure 2C ) . 10 . 7554/eLife . 01085 . 004Figure 2 . Gic1 binds specifically to septin Cdc10 . ( A and B ) Representative EM image ( A ) and class averages ( B ) of septin-Gic1 complexes labeled with antibody against Cdc11 . Arrow indicates the antibodies . The class average in ( B ) contains 15 single particles . Scale bars , 100 nm and 10 nm in ( A ) and ( B ) , respectively . ( C ) Model of the septin-Gic1 complex based on the known sequential order of septin filaments ( Bertin et al . , 2008 ) . The G- and the N/C-interfaces are indicated by straight and circular interfaces between circles , respectively . Antibodies are indicated as Y shapes . ( D and E ) Representative EM images of septin-Cdc10Δ filaments without ( D ) and with Gic1 ( E ) . ( F and G ) Representative EM images of septin-Cdc11Δ filaments ( F ) and septin-Cdc11Δ-Gic1 complexes ( G ) . ( H ) Representative class average of the septin-Cdc11Δ-Gic1 complex with focus on the septin filament . Arrows indicate single septin proteins . Scale bar , 10 nm . ( I ) Model of the septin-Cdc11Δ-Gic1 complex . ( J ) Yeast two-hybrid assay of different septin proteins , namely Cdc3 , Cdc10 , Cdc11 , and Cdc12 with Gic1 . DOI: http://dx . doi . org/10 . 7554/eLife . 01085 . 004 However , previous yeast two-hybrid , in vitro interaction and coimmunoprecipitation data indicated that Gic1 directly interacts with Cdc12 ( Iwase et al . , 2006 ) . Therefore , we performed additional experiments to further corroborate our findings and prepared septin-Gic1 complexes devoid of either Cdc10 or Cdc11 ( Figure 2D–I ) . As expected from the results of antibody labeling , septin-Cdc10Δ , which formed short filaments , did not bind to Gic1 ( Figure 2D–E ) . Septin-Cdc11Δ , however , which in accordance with previous studies ( Bertin et al . , 2010 ) does not polymerize ( Figure 2F ) produced a similar track-like structure as the wildtype when Gic1 was added ( Figure 2G ) . Thus , addition of Gic1 resulted not only in the binding of the protein to the septin complex but also induced septin polymerization . We then performed single particle analysis ( SPA ) of the septin filaments and observed that only four subunits were located between Gic1 cross-bridges of Cdc11Δ filaments , ruling out an interaction with Cdc12 ( Figure 2H–I ) . In addition , yeast two-hybrid studies supported our in vitro data , showing that Gic1 only interacts with Cdc10 ( Figure 2J ) . Notably , Cdc10 is mainly responsible for the specific interaction of septin filaments with PIP2 , localizing them to membranes and promoting filament polymerization and organization ( Bertin et al . , 2010 ) . For Gic2 , which is a homologue of Gic1 , a direct interaction through its polybasic region with PIP2 was also reported ( Orlando et al . , 2008 ) . Together with our findings , this suggests that the organization and polymerization of septin filaments is controlled by the interaction of Gics with Cdc10 and the interaction of both proteins with PIP2 . Bertin et al . reported that septin filaments are cross-linked by overlapping C-terminal extensions of Cdc3 and Cdc12 ( Bertin et al . , 2008 ) . Depending on the stain thickness we also observed these thin cross-links between bare septin filaments . However , in septin-Gic1 complexes , the large Gic1 cross-bridges are very prominent , causing stain to accumulate between the Gic1 cross-bridges . This makes it impossible to visualize the thin coiled-coils between Cdc3 and Cdc12 , although they are probably still there . To study the native three-dimensional structure of the filaments , we vitrified septin-Gic1 complexes and determined their structure using cryo electron tomography ( cryo-ET ) ( Figure 3; Videos 1–4 ) . We observed that Gic1 , instead of cross-bridging only two filaments , simultaneously interacted with up to six septin polymers , forming long filament cables ( Figure 3E–G ) . The organization of these cables is such that Gic1 forms a central scaffold to which septin filaments attach . Remarkably , individual septin filaments do not start at the same position and run along in parallel rigid lines . The septin-Gic1 cables are rather composed of many filaments of variable lengths that start at random positions , sometimes bypassing a Gic1 cross-bridge or changing place with another filament ( Figure 3H–I ) . Filaments are often only attached to adjacent filaments for several nanometers . This gives the septin-Gic1 cables a certain flexibility that on the one hand allows them to bend and adjust to membrane curvature ( Figure 3C ) . On the other hand it enables the interaction between cables , resulting in mesh-like structures ( Figure 3D ) . Because of the limited resolution of the reconstructions , the additional thin connections described by Bertin et al . ( Bertin et al . , 2008 ) are not visible . 10 . 7554/eLife . 01085 . 005Figure 3 . Cryo-ET of the septin-Gic1 complex . ( A ) Central slice of a cryo electron tomogram ( for full tomogram see Videos 1–3 ) . Arrow indicates a septin-Gic1 cable . Scale bar , 200 nm . ( B ) Segmentation of the tomograms . ( C and D ) Extracts from tomograms that show the flexibility and bending of the septin-Gic1 complex ( C ) as well as its ability to form mesh-like structures ( D ) . Scale bar , 20 nm . ( E–G ) Side ( E ) and top view ( F ) of a septin-Gic1 complex . The septin filaments and Gic1 cross-bridges are depicted in gold and green , respectively . Scale bar , 20 nm . ( G ) The crystal structure of the human SEPT2/6/7 complex ( PDB 2QAG ) was manually fit into the EM structure . Scale bar , 5 nm . ( H and I ) Side views of the septin-Gic1 3D models . To allow a better observation of the septin filament interaction with Gic1 , part of the septin filaments have been omitted . Scale bar , 20 nm . ( J ) Model of the septin-Gic1 complex based on the known sequential order of septin filaments ( Bertin et al . , 2008 ) . The G- and the N/C-interfaces are indicated by straight and circular interfaces between circles , respectively . DOI: http://dx . doi . org/10 . 7554/eLife . 01085 . 00510 . 7554/eLife . 01085 . 006Video 1 . Video through a cryo electron tomogram of the septin-Gic1 complex with bundles running perpendicular to the beam . DOI: http://dx . doi . org/10 . 7554/eLife . 01085 . 00610 . 7554/eLife . 01085 . 007Video 2 . Video through a cryo electron tomogram of the septin-Gic1 complex with bundles running parallel to the beam ( indicated by an arrow ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01085 . 00710 . 7554/eLife . 01085 . 008Video 3 . Close-up on a septin-Gic1 complex running parallel to the beam ( full tomogram see Video 2 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01085 . 00810 . 7554/eLife . 01085 . 009Video 4 . Video of the 3D reconstruction of the septin-Gic1 complex derived from tomograms of vitrified samples . DOI: http://dx . doi . org/10 . 7554/eLife . 01085 . 009 Both in three-dimensional reconstructions and two-dimensional class averages , the density corresponding to Gic1 is poorly defined , suggesting a flexible structure of Gic1 . Gic1 , which based on its sequence has a molecular weight of 23 kDa ( Gic1 ( 104-314 ) ) ( Figure 4A ) , elutes at about 49 kDa from a gel filtration column corresponding roughly to a dimer ( Figure 4B ) . In a typical Gic1 cross-bridge up to 12 Cdc10 subunits are involved ( two per filament ) ( Figure 3E ) . If we assume that each of them binds independently to a Gic1 dimer , we would expect that 12 Gic1 dimers assemble into a large cross-bridge of 600 kDa . Although we cannot exclude that the density corresponding to Gic1 appears more spread out due to missing wedge artifacts in our tomograms , the average large volume of the Gic1 density in the electron tomograms ( Figure 3E–I ) indicates that it must contain multiple Gic1 molecules . In addition , the heterogeneity of the Gic1 cross-bridges suggests that the number of Gic1 molecules varies between cross-bridges . 10 . 7554/eLife . 01085 . 010Figure 4 . Purification of Gic1 . ( A ) SDS-PAGE of the Gic1 ( 104-314 ) purification . ( 1 ) Flow through Ni-NTA , ( 2 ) wash Ni-NTA , ( 3 ) elution Ni-NTA , ( 4 ) elution from gel filtration after cleavage with prescission protease . ( B ) Gel filtration chromatography of purified Gic1 ( 104-314 ) . The calculated molecular weight of Gic1 ( 104-314 ) is 23 . 38 kDa . The protein elutes from the gel filtration chromatography at a volume that corresponds to a molecular weight of 49 kDa , suggesting that it forms dimers . Protein standards for gel filtration: 1 , ferritin ( 440 kDa ) ; 2 , aldolase ( 158 kDa ) ; 3 , conalbumin ( 75 kDa ) ; 4 , ovalbumin ( 43 kDa ) ; 5 , carbonic anhydrase ( 29 kDa ) ; 6 , RNase A ( 13 . 7 kDa ) ; 7 , aprotinin ( 6 . 5 kDa ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01085 . 010 It was previously reported that Gic1 and its homologue Gic2 are effectors of Cdc42 , which bind via their CRIB motif to the GTP-bound form of the Rho GTPase in cells ( Brown et al . , 1997; Chen et al . , 1997 ) . We made similar observations when studying the interaction of purified Gic1 and Cdc42 in vitro . Gic1 formed a stable complex with Cdc42-GppNHp ( non-hydrolysable GTP analogue ) , but did not bind to Cdc42 in the GDP-state ( Figure 5 ) . On producing septin-Gic1 complexes in the presence of Cdc42-GppNHp , we found the same railroad-pattern as for the septin-Gic1 complexes , however , in this case with much bulkier cross-bridges ( Figure 6A ) . 10 . 7554/eLife . 01085 . 011Figure 5 . Cdc42-GppNHp binds specifically to Gic1 . ( A and B ) Gel filtration chromatography and SDS-PAGE of Gic1 with Cdc42-GppNHp ( A ) and Cdc42-GDP ( B ) , respectively . DOI: http://dx . doi . org/10 . 7554/eLife . 01085 . 01110 . 7554/eLife . 01085 . 012Figure 6 . Cdc42-GppNHp binds specifically to Gic1 and dissociates septin-Gic1 complexes . ( A ) Representative EM image of negatively stained septin-Gic1-Cdc42-GppNHp complexes . The concentration of Gic1 and Cdc42-GppNHp used for filament preparation is 0 . 5 µM . Scale bar , 100 nm . ( B and C ) Representative class averages of negatively stained septin-Gic1-Cdc42-GppNHp complexes with focus on septin filaments and the cross-bridges , respectively . Arrows indicate single septin proteins . Scale bar , 10 nm . ( D ) Same as ( A ) but at 10× higher concentration of Cdc42-GppNHp . ( E–G ) Side ( E ) and top view ( F ) of a septin-Gic1-Cdc42-GppNHp complex . The septin filaments and Gic1-Cdc42-GppNHp cross-bridges are depicted in gold and green , respectively . Scale bar , 20 nm . ( G ) The crystal structure of the human SEPT2/6/7 complex ( PDB 2QAG ) was manually fit into the EM structure . Scale bar , 5 nm . ( H and I ) Side views of the septin-Gic1-Cdc42-GppNHp 3D structures . To allow a better observation of the septin filament interaction with Gic1 , part of the septin filaments have been omitted . Scale bar , 20 nm . ( J ) Model of the septin-Gic1-Cdc42-GppNHp complex . Cdc42-GppNHp is depicted as yellow ovals . DOI: http://dx . doi . org/10 . 7554/eLife . 01085 . 012 SPA and cryo-ET of the Gic1-Cdc42-septin complex revealed that the cross-bridges were almost one and a half times larger in size , spanning not only two but four septin subunits , leaving only four septins unoccupied ( Figure 6B–I , Video 5 ) . This suggests a direct interaction of Cdc42-GppNHp with Gic1 on septin filaments ( Figure 6J ) . 10 . 7554/eLife . 01085 . 013Video 5 . Video of the 3D reconstruction of the septin-Gic1-Cdc42-GppNHp complex derived from tomograms of vitrified samples . DOI: http://dx . doi . org/10 . 7554/eLife . 01085 . 013 Unexpectedly , an increase of the Cdc42-GppNHp concentration resulted in dissociation of the septin-Gic1-Cdc42 complex ( Figures 6D and 7 ) . Smaller in size , but still filamentous , the septin filament structure deteriorated somewhat as a result of this dissociation . Gic1 and Cdc42 , however , formed protein aggregates ( Figure 7C , F ) . A possible explanation for this observation is that full decoration of Gic1 with Cdc42 destabilizes the Gic1 interaction with septins , resulting in dissociation of the septin-Gic1-Cdc42 complex . In contrast , Cdc42-GDP did not bind to septin-Gic1 complexes even at 10 times higher concentrations ( Figure 8 ) . Similar behavior was reported for the functionally related human protein Borg ( Joberty et al . , 1999; Sirajuddin et al . , 2007 ) . Borg binds with its BD3 domain to SEPT7/SEPT6 dimers and induces the formation of long and thick septin filament fibers ( Joberty et al . , 2001; Sheffield et al . , 2003 ) . In accordance with our observations with Cdc42-GTP , Gic1 and septin , the human Cdc42-GTP negatively regulates this process and after binding to Borg inhibits the Borg–septin association . 10 . 7554/eLife . 01085 . 014Figure 7 . Time- and concentration-dependent interaction of Cdc42-GppNHp with the septin-Gic1 complex . ( A–C ) Representative EM images of the septin-Gic1 complex incubated for 60 min with ( A ) 0 . 25 μM , ( B ) 1 . 5 μM or ( C ) 5 μM of Cdc42-GppNHp . ( D–F ) Representative EM images of the septin-Gic1 complex incubated with 5 µM of Cdc42-GppNHp for ( D ) 2 min , ( E ) 15 min and ( F ) 60 min . Scale bar , 100 nm . DOI: http://dx . doi . org/10 . 7554/eLife . 01085 . 01410 . 7554/eLife . 01085 . 015Figure 8 . Cdc42-GDP does not interact with septin-Gic1 complexes . ( A ) Representative EM image of negatively stained septin-Gic1-Cdc42-GDP complexes . Scale bar , 100 nm . ( B ) Same as ( A ) but at 10× higher concentration of Cdc42-GDP . ( C and D ) Representative class averages of negatively stained septin-Gic1-Cdc42-GDP complexes with focus on septin filaments and the cross-bridges , respectively . Arrows indicate single septin proteins . Scale bar , 10 nm . DOI: http://dx . doi . org/10 . 7554/eLife . 01085 . 015 To determine whether Cdc42 interacts directly with septins in the absence of Gic1 , we produced septin filaments in the presence of Cdc42-GppHNp and Cdc42-GDP , respectively . Although Cdc42 in its GTP-state bound to septin filaments , no regular pattern as in the case of Gic1 was observed , indicating non-specific interaction ( Figure 9A ) . Surprisingly , Cdc42-GDP bound more specifically to septin filaments , and formed , in contrast to Cdc42-GppNHp , railroad track-like structures ( Figure 9B ) . Even more surprising was that Cdc42-GDP completely dissociated septin filaments at higher concentrations in less than an hour ( Figures 9C and 10 ) . We then analyzed the oligomers by size-exclusion chromatography , which showed that Cdc42-GDP binds to septin complexes ( Figure 9D ) . SPA of the dissociated septins clearly identified either octamers with an extra density at their ends ( 90% ) or hexamers without additional density ( 10% ) ( Figure 9E–F ) . Cdc3 sits at the ends of the hexamers , and at the second outer position of the octamers ( Figure 9G–H ) . Considering the sequential order of septin oligomers , this indicates that Cdc42-GDP directly binds to Cdc10 ( Figure 9I ) , which is also the binding partner of Gic1 ( Figure 2 ) , thereby causing dissociation of the filaments . Consequently , the resulting septin octamers differ from the ones observed under high-salt conditions ( Figure 11 ) . Instead of Cdc11 , which is now in the center , Cdc10 forms the cap of the octamers at both ends . In addition , binding of Cdc42-GDP to Cdc10 probably influences its interaction with Cdc3 at the G interface resulting in the formation of Cdc42-GDP-Cdc10 dimers and hexameric ( Cdc3-Cdc12-Cdc11 ) complexes . A direct interaction of a protein with Cdc42-GDP is unusual , since in most reported cases Cdc42 , like other small G proteins , binds to its effectors in its active , that is GTP-bound state . However , also other proteins such as Msb3 and Msb4 , which are GAP proteins , that are , like Gic1 , involved in cell polarization , were shown to directly interact with Cdc42-GDP ( Tcheperegine et al . , 2005 ) . 10 . 7554/eLife . 01085 . 016Figure 9 . Cdc42-GDP binds specifically to Cdc10 and dissociates septin filaments . ( A ) Representative EM image of negatively stained septin complexes incubated with Cdc42-GppNHp . Scale bar , 100 nm . ( B ) Representative EM image of negatively stained septin-Cdc42-GDP-complexes . ( C ) Same as ( B ) but at 10× higher concentration of Cdc42-GDP . ( D ) Gel filtration chromatography and SDS-PAGE of the septin-Cdc42-GDP complex . ( E and F ) Representative class averages of septin-Cdc42-GDP complexes . The asterisk indicates the additional density corresponding to Cdc42-GDP . Arrows indicate single septin proteins . Scale bar , 10 nm . ( G and H ) Representative class averages of the septin-Cdc42-GDP complex labeled with antibody against Cdc3 . The asterisk indicates the additional density corresponding to Cdc42-GDP . Arrows indicate single septin proteins . The triangle indicates the antibody . ( I ) Model of the septin-Cdc42-GDP complex . Cdc42-GDP is depicted as yellow circle . DOI: http://dx . doi . org/10 . 7554/eLife . 01085 . 01610 . 7554/eLife . 01085 . 017Figure 10 . Time- and concentration-dependent interaction of Cdc42-GDP with septin filaments . ( A–C ) Representative EM images of septin filaments incubated for 60 min with ( A ) 0 . 25 μM , ( B ) 1 . 5 μM or ( C ) 5 μM of Cdc42-GDP . ( D–E ) Representative EM images of septin filaments incubated with 5 µM of Cdc42-GDP for ( D ) 2 min , ( E ) 15 min and ( F ) 60 min . Scale bar , 100 nm . DOI: http://dx . doi . org/10 . 7554/eLife . 01085 . 01710 . 7554/eLife . 01085 . 018Figure 11 . Septin polymerization depends on the ionic strength . ( A and B ) Representative EM images and class averages of septin complexes at high-salt ( 500 mM NaCl ) labeled with antibody against Cdc3 ( C ) and low-salt ( 100 mM NaCl ) ( D ) conditions . Arrows indicate single septin proteins . Scale bars , B = 100 nm , D = 10 nm . ( E ) Model of the septin complex based on the known sequential order of septin filaments . The G- and the N/C-interfaces are indicated by straight and circular interfaces between circles , respectively . DOI: http://dx . doi . org/10 . 7554/eLife . 01085 . 018 To identify the Cdc42 binding site on Cdc10 , we calculated homology models of Cdc3 , Cdc10 , Cdc11 and Cdc12 using the SEPT2 structure ( PDB 2QA5 ) as a reference and mapped regions of conservation between the four structures on their surfaces . Examining the Cdc10-Cdc10 interface , which is an N/C interface , we found that the N-terminal region is the least conserved , and would therefore present an ideal site for specific Cdc42-GDP binding ( Figure 12A–B ) . We deleted the N-terminal 29 residues of Cdc10 and produced complexes whose Cdc10 was replaced by Cdc10 ( 30-322 ) . It was previously shown that deletion of the N-terminal residues of Cdc10 results in non-polymerizing septin complexes forming tetramers and inside-out octamers ( Bertin et al . , 2010 ) . We analyzed the particles by EM and SPA , which showed that most of the complexes were indeed tetramers or octamers , and that the polymerization of the complexes was clearly impaired , as we did not observe filaments at low-salt conditions ( Figure 13A–C ) . We then incubated the oligomers with high concentrations of Cdc42-GDP and looked for additional densities corresponding to Cdc42-GDP by EM and SPA . However , no additional protein was bound to the tetramers or octamers ( Figure 13D–F ) . Gel filtration experiments corroborated this result and showed that indeed Cdc42-GDP does not bind to Cdc10 ( 30-322 ) ( Figure 13G ) . This indicates that the N-terminal region of Cdc10 is essential for the binding of Cdc42-GDP to septin filaments . Interestingly , the same region on Cdc10 was shown to be important for the interaction of septin filaments with PIP2 ( Bertin et al . , 2010 ) . 10 . 7554/eLife . 01085 . 019Figure 12 . Conservation of the N/C interface between Cdc3 , Cdc10 , Cdc11 and Cdc12 . ( A and B ) Side ( A ) and end-on views ( B ) of the N/C interface between two Cdc10 septins . Homology models of Cdc3 , Cdc10 , Cdc11 and Cdc12 using the SEPT2 structure ( PDB 2QA5 ) as reference were calculated using Phyre2 ( Kelley and Sternberg , 2009 ) . The models were aligned in Chimera ( Pettersen et al . , 2004 ) using the SEPT2/6/7 complex ( PDB 2QAG ) as reference . The sequences of Cdc3 , Cdc10 , Cdc11 and Cdc12 were aligned using ClustalW ( Larkin et al . , 2007 ) and the conservation between the four septins was mapped on their surfaces . DOI: http://dx . doi . org/10 . 7554/eLife . 01085 . 01910 . 7554/eLife . 01085 . 020Figure 13 . Gic1 but not Cdc42-GDP binds to polymerization impaired septin complexes . ( A–C ) Representative EM image and class averages of polymerization-impaired septin complexes containing Cdc10 ( 30-322 ) . Scale bars , 100 nm in ( A ) and 10 nm in ( B ) . ( D–F ) Representative EM image and class averages of polymerization-impaired septin complexes containing Cdc10 ( 30-322 ) + Cdc42-GDP . ( G ) Gel filtration chromatography and SDS-PAGE of septin-Cdc10 ( 30-322 ) and Cdc42-GDP . ( H–J ) Representative EM image and class averages of septin-Cdc10 ( 30-322 ) -Gic1 complexes . The asterisk indicates the additional density corresponding to Gic1 . ( K ) Gel filtration chromatography and SDS-PAGE of the septin-Cdc10 ( 30-322 ) -Gic1 complex . ( L ) Model of the septin-Cdc42-GDP complex . Gic1 is depicted as blue oval . Arrows indicate single septin proteins . DOI: http://dx . doi . org/10 . 7554/eLife . 01085 . 020 From our previous experiments , we know that in septin-Gic1 complexes Cdc42-GDP binds neither to septins nor to Gic1 ( Figure 8 ) . Since both Gic1 and Cdc42-GDP bind to Cdc10 and very probably compete for the same binding site , Gic1 must have a higher affinity for Cdc10 than Cdc42-GDP . In order to prove this , we added Gic1 to septin-Cdc42-GDP complexes ( Figure 14A–B ) . As expected , we found long railroad track-like ordered filaments indicating that Gic1 replaced Cdc42-GDP to interact with Cdc10 , resulting in septin polymerization . 10 . 7554/eLife . 01085 . 021Figure 14 . Gic1 stabilizes septin filaments . ( A–D ) Representative EM images of septin-Cdc42-GDP complexes and septin octamers in high-salt buffer before ( A and C ) and after incubation with Gic1 ( B and D ) , respectively . Scale bar , 100 nm . ( E and F ) Representative EM images of the septin-Gic1 complex before ( E ) and after increasing the salt concentration ( F ) , respectively . DOI: http://dx . doi . org/10 . 7554/eLife . 01085 . 021 We also observed septin-Gic1 filament complexes when we mixed Gic1 with octamers under conditions that normally inhibit polymerization , such as high-salt buffer ( Figure 14C–D ) . This was surprising , since Gic1 binds and crosslinks Cdc10 proteins while high salt weakens the interaction at the N/C-interface between two Cdc11 proteins . Thus , the stabilizing effect of Gic1 must be stronger than the weakening effect of high salt . We confirmed this observation by performing the inverse experiment , that is septin-Gic1 complexes were dialyzed against a high-salt buffer . As expected , septin-Gic1 complexes were stable and did not dissociate ( Figure 14E–F ) . Even septin complexes completely lacking Cdc11 , which do not polymerize at low-salt conditions , form filaments when Gic1 is added ( Figure 2G ) . Taken together , we conclude that Gic1 not only scaffolds septin filaments , but also increases their stability . We then asked whether Gic1 would also stabilize septin filaments whose assembly is weakened at the Cdc10-Cdc10 instead of the Cdc11-Cdc11 interface . As shown above , complexes whose Cdc10 is replaced by Cdc10 ( 30-322 ) do not polymerize at all and form tetramers or inside-out octamers ( Figure 13A–C ) . Interestingly , Gic1 did not induce filament formation of such impaired septins ( Figure 13H ) , but still bound to Cdc10 ( 30-322 ) as indicated by an additional density attached to the septin oligomers and by size-exclusion chromatography ( Figure 13I–L ) . Our data ( see Figure 15 for overview ) provide important insights into the interaction of Gic1 , Cdc42 and septins . Although our analysis focuses on only three proteins that are important for septin recruitment , ring formation and disassembly , which are complex processes involving many proteins ( reviewed in Park and Bi , 2007 ) , we think that the new information provided by our study is coherent enough to suggest the following mechanism for the interplay between Gic1 , Cdc42 and septins during budding or the cell cycle . 10 . 7554/eLife . 01085 . 022Figure 15 . Schematic overview of all results . ( A ) At high ionic strength septin filaments disassemble into octamers ( Figure 11 ) . However , when Gic1 is added , septin-Gic1 filament cables are formed even at high-salt concentrations ( Figure 14 ) . ( B ) Gic1 binds specifically to Cdc10 and thereby scaffolds and stabilizes septin filaments forming long filament cables ( Figures 1 and 2 ) . Cdc42-GTP binds specifically to Gic1 resulting in a Cdc42-GTP-Gic1 complex ( Figures 5 and 6 ) . However , at higher Cdc42-GTP concentrations , the Gic1-septin interaction is negatively influenced and results in the dissociation of the complex ( Figure 6 ) . ( C ) Cdc42-GDP interacts with Cdc10 and binds specifically to septin filaments ( Figure 9 ) . However , at higher Cdc42-GDP concentrations , the complex dissociates into octamers with Cdc42-GDP bound to Cdc10 and Cdc10-less hexamers ( Figure 9 ) . Gic1 displaces Cdc42-GDP and septin-Gic1 filament cables are formed ( Figure 14 ) . ( D and E ) The polymerization of septin complexes containing Cdc10 ( 30-322 ) is impaired . ( D ) Gic1 still binds to Cdc10 , however , does not cross-bridge complexes and septins do not polymerize ( Figure 13 ) . ( E ) Cdc42-GDP does not bind to septin-Cdc10 ( 30-322 ) ( Figure 13 ) . ( F ) Gic1 does not bind to polymerization-impaired septin-Cdc10Δ complexes ( Figure 2 ) . ( G ) Gic1 binds to polymerization-impaired septin-Cdc11Δ complexes resulting in septin polymerization and formation of septin-Gic1 filament cables ( Figure 2 ) . Septins are depicted as orange rods . Green caps indicate Cdc11 . Gic1 is depicted as blue rectangle . Cdc42-GTP and Cdc42-GDP are depicted as red and yellow rectangles , respectively . The N-terminal truncation of Cdc10 is marked by a cross and the destabilization of the Cdc11-Cdc11 N/C interface is indicated by a black block . DOI: http://dx . doi . org/10 . 7554/eLife . 01085 . 022 Because Cdc42-GDP binds to Cdc10 in septin complexes and prevents their polymerization , we propose that Cdc42-GDP rather than Gic1 recruits septin octamers ( with Cdc10 at the caps ) to the bud site ( Figure 16A ) . At the bud site Cdc42 interacts with its GEF Cdc24 , which catalyzes the exchange of its nucleotide ( Figure 16B ) . As a result , Cdc42 dissociates from the septin complexes , which in turn polymerize . Besides activating many other proteins involved in septin recruitment and ring formation Cdc42-GTP recruits its effector Gic1 to the bud site ( Figure 16C ) . Gic1 , which has a high affinity for Cdc10 , binds , scaffolds and stabilizes septin filaments ( Figure 16D ) . Gic1-Cdc42-GTP-septin cables generated in this manner form a ring at the bud neck . Since our data show that the process of Gic1-septin cable formation does not necessarily require Cdc42-GTP , we propose that upon GAP-supported GTP hydrolysis , Cdc42-GDP , which has a much lower affinity to Gic1 than Cdc42-GTP , dissociates from the Gic1-septin complex ( Figure 16E ) and recruits more septin octamers to the bud site ( Figure 16F ) . 10 . 7554/eLife . 01085 . 023Figure 16 . Model for septin recruitment , ring formation and disassembly . ( A ) Cdc42-GDP recruits septin complexes to the bud site . ( B ) At the bud site Cdc24 catalyzes the nucleotide exchange of Cdc42 , which recruits its effector Gic1 . ( C ) Septins polymerize . ( D ) Gic1 scaffolds and stabilizes septin filaments and forms septin-Gic1-Cdc42-GTP filament cables that are used for building the septin ring . ( E ) Cdc42-GTP is not necessary for the stability of the filament cables and upon GTP hydrolysis dissociates from the septin-Gic1 complexes . ( F ) Cdc42-GDP recruits more septin complexes to the bud site . ( G ) During cell division the local concentration of Cdc42-GTP increases by the up-regulation of Cdc24 . This leads to a dissociation of the septin-Gic1-Cdc42-GTP filament cables and the septin ring disassembles . ( H ) After Cdc42-GAP catalyzed hydrolysis , Cdc42-GDP binds to the septin filaments and disassembles them to octamers . Septins are depicted as orange rods . Green caps indicate Cdc11 . Gic1 is depicted as a blue rectangle . Cdc42-GTP and Cdc42-GDP are depicted as red and yellow rectangles , respectively . DOI: http://dx . doi . org/10 . 7554/eLife . 01085 . 023 At later stages of the cell cycle the local Cdc42-GTP concentration increases at the bud neck , probably caused by an up-regulation of Cdc24 and/or identical spatial localization ( Butty et al . , 2002; Goryachev and Pokhilko , 2008 ) . As a result , Cdc42-GTP binds to Gic1-septin complexes and competes with Cdc10 for Gic1 , which finally results in the disassembly of the complex and thereby septin ring rearrangement ( Figure 16G ) . Gic1 will then be either degraded or relocated . After GTP hydrolysis , Cdc42-GDP in the absence of Gic1 specifically binds to septin filaments and dissociates them ( Figure 16H ) . Thus , septins can be reassembled and reused for the next cycle , as was observed by McMurray and Thorner when tracking labeled septins through several cell divisions ( McMurray and Thorner , 2009 ) . The model elegantly explains the cycles of GTP loading and hydrolysis by Cdc42 that were observed by Gladfelter et al . during budding ( Gladfelter et al . , 2002 ) . In conclusion , Gic1 and Cdc42 in combination with many other factors , such as the nucleotide state of septins , specific interaction with lipids and protein posttranslational modifications regulate septin recruitment , ring formation and disassembly . Most importantly , Cdc42 does not only act as a regulator , but seems to be also involved in septin recruitment .
Construction of yeast two-hybrid plasmids of septins were previously described ( Farkasovsky et al . , 2005 ) . In order to construct plasmids coding for the LexA and AD fusions with Gic1 fragments of different size , gic1 ( 310-942 ) ( primers GIC1-N3: CCAAGGATCCATGTTCAAAAAAAAGGACCTGTTGTCGAGG and GIC1-C1: CCAAGTCGACGGTATTTCGAGGAGTACTAGTTTC ) and gic1 ( 670-942 ) ( primers GIC1-N4: CCAAGGATCCGATTTGGAAATGACCTTGGAAGAC and GIC1-C1: CCAAGTCGACGGTATTTCGAGGAGTACTAGTTTC ) were amplified by using yeast chromosomal DNA as a template and the Expand High Fidelity PCR system ( Roche , Mannheim , Germany ) . The PCR products were digested with BamHI and SalI and fragments were introduced between the BamHI and SalI sites of pEG202 or pJG4-5 vectors . The same PCR products were also used in the construction of expression plasmids pFM812 ( gic1 ( 310-942 ) in pEGST with C-terminal His6-tag ) or pFM562 ( gic1 ( 670-942 ) in pGEX4T-3 ) . All plasmid constructs were confirmed by sequencing . The yeast septin complex was expressed and purified as described earlier ( Farkasovsky et al . , 2005 ) . In order to study the interaction of Gic1 with septin filaments in vitro , we first expressed Gic1 recombinantly in E . coli . Since the full-length protein aggregated during expression , we tested different constructs and could obtain sufficient amounts of stable non-aggregating protein only after deleting the N-terminal 103 amino acids ( Figure 4 ) . Gic1 ( 104-314 ) ( in the text referred to as Gic1 ) contains the CRIB domain , which is essential for its interaction with Cdc42 , and the C-terminus , which , because of its homology to the Borg BD3 domain , might be important for Gic1 binding to septins . For the bacterial expression of Gic1 ( 104-314 ) , the plasmid pFM812 was transformed into the E . coli strain BL21 ( DE3 ) Rosetta ( Merck KGaA , Darmstadt , Germany ) . The cells were grown in TB medium , supplemented with ampicillin and chloramphenicol at 37°C and induced by addition of 0 . 2 mM IPTG at an optical density of OD600 = 0 . 6 . After 8 hr at 28°C , cells were harvested by centrifugation , resuspended in isolation buffer IB1 ( 25 mM NaHPO4 pH 7 . 8 , 5% glycerol , 0 . 3 M NaCl , 1 mM MgCl2 , 5 mM ß-mercaptoethanol , 10 mM imidazole , complete protease inhibitors [Roche] , 0 . 2 mM PMSF ) and disrupted by using a microfluidizer ( Microfluidics Co . , Westwood , MA , USA ) . After high-speed centrifugation at 100000×g , the supernatant was incubated with 50 ml ( Vt ) Ni-NTA-Sepharose ( Qiagen , Hilden , Germany ) , washed with 300 ml buffer IB2 ( 25 mM NaHPO4 pH 7 . 8 , 5% glycerol , 0 . 5 M NaCl , 1 mM MgCl2 , 5 mM ß-mercaptoethanol , 50 mM imidazole , 5 mM ATP , complete protease inhibitors , 0 . 2 mM PMSF ) and with 100 ml of buffer IB3 ( 25 mM NaHPO4 pH 7 . 8 , 5% glycerol , 0 . 3 M NaCl , 1 mM MgCl2 , 5 mM ß-mercaptoethanol ) . Gic1 was eluted with 300 mM imidazole in buffer IB3 and the GST-tag was cleaved using thrombin at 4°C . Then , the GST and the undigested fusion protein were removed by glutathione sepharose column chromatography ( GE Healthcare , Buckinghamshire , UK ) . Gic1 was concentrated and further purified on a Superdex S200 column ( GE Healthcare , Buckinghamshire , UK ) . Expression and purification of Cdc42 ( G12V ) was performed as previously described ( Rudolph et al . , 1998 ) . Due its intrinsic GTPase activity , Cdc42 is usually GDP-bound after purification . In order to exchange GDP to GppNHp or to remove residual GTP , 5 mM EDTA ( 5 times the MgCl2 concentration ) and 20 times excess of the desired nucleotide over the protein were added to Cdc42 and incubated at room temperature for 2 hr . Subsequently , the protein was concentrated using Amicon Ultra-4 Centrifugal Filters with a cut-off of 10 kDa and washed with the gel filtration buffer devoid of EDTA and nucleotide ( 150 mM NaCl , 20 mM Tris-HCl pH 7 . 5 , 1 mM MgCl2 ) . For septin filament production , septin oligomers in a high-salt buffer ( 500 mM NaCl , 1 mM MgCl2 , 50 mM Tris-HCl pH 7 . 5 , 1 mM DTT ) at a final concentration of 0 . 3 µM were dialyzed overnight at 4°C against a low-salt buffer ( 100 mM NaCl , 20 mM Tris-HCl pH 7 . 5 , 1 mM DTT ) . In the case of Gic1-septin complexes , septin oligomers ( final concentration of 0 . 3 µM ) were mixed with Gic1 ( final concentration of 1 . 5 µM ) in a high-salt buffer and dialyzed as described above . For antibody decoration , 5 μl of polyclonal antibodies against Cdc11 ( Santa Cruz Biotech , Heidelberg , Germany ) and Cdc3 ( gift from Dr Michael Knop , ZMBH , Heidelberg ) ( 1:100 ) were added to 20 μl of a sample containing the filaments or the septin octamers and incubated overnight at 4°C . For studies involving Cdc42 , 0 . 1 μM of the septin octamer and 0 . 5 μM of Gic1 were used . Cdc42-GppNHp and Cdc42-GDP were used at the same concentration as Gic1 or at higher concentrations ( as indicated in the figures ) and incubated for different time intervals ( Figures 7 and 10 ) . For gel filtration analyses of binding between non-polymerizing septin , Gic1 and Cdc42 , 1 mg of each protein of the desired complex was mixed and incubated for 15 min at 4°C . Then , 500 μl of the solution was injected into Superdex S200 ( GE Healthcare , Buckinghamshire , UK ) column . The sample was run at 0 . 4 ml/min with a buffer containing 100 mM NaCl , 20 mM Tris-HCl pH 7 . 5 , 1 mM DTT and 1 mM MgCl2 . Two-hybrid studies were performed using the LexA-based system as described previously ( Sirajuddin et al . , 2009 ) . The yeast strain EGY48 was co-transformed with pEG202-based and pJG4-5-based plasmids . The reporter plasmid pSH18-34 was used for the quantitative ß-galactosidase assay . Three independent isolates of each strain were tested on minimal medium in absence of leucine or presence of X-gal , respectively . To obtain more details on the structure of septin-Gic1 complexes , we formed single railroad tracks by untangling the bundles on a lipid monolayer , which we then studied by EM and SPA . Since Cdc3 carries a His6-tag , the filaments can be adsorbed to a lipid monolayer containing Ni-NTA-lipids . To obtain single-stranded filaments , 30 µl of the samples were adsorbed to a lipid monolayer composed of 0 . 5 mg/ml of DOGS-NTA:DOPC at a molar ratio of 1:3 and incubated for 1 hr at 4°C . The monolayer was transferred to a carbon-coated grid and then negatively stained as described below . To prove that the His6-tags of the proteins are not responsible for the entangling of the filaments , we performed additional experiments . Both Gics and septins have been reported to interact strongly with PIP2 ( Orlando et al . , 2008; Bertin et al . , 2010 ) . We therefore immobilized septin-Gic1 filaments on lipid monolayers containing PIP2 instead of Ni-NTA lipids . The filaments adsorbed to the grid and bundles were ‘untangled’ comparably to that seen in the experiments with Ni-NTA lipids ( Figure 17 ) , indicating that the interaction of the septin-Gic1 filaments , respectively , is not His6-tag-induced . 10 . 7554/eLife . 01085 . 024Figure 17 . Septin-Gic1 complexes immobilized on a PIP2-containing lipid monolayer . ( A–C ) Representative EM image of negatively stained septin-Gic1 complexes immobilized on a PIP2-containing monolayer . Scale bar , 100 nm . DOI: http://dx . doi . org/10 . 7554/eLife . 01085 . 024 Conventional negative staining was performed as previously described ( Bröcker et al . , 2012 ) . In brief , samples were applied onto freshly glow-discharged , carbon-coated copper grids . The sample was left for 1 min on the grid before blotting and staining with uranyl formate ( 0 . 7% wt/vol ) . All images of negatively stained samples were taken with a JEOL JEM 1400 electron microscope equipped with a LaB6 filament at an acceleration voltage of 120 kV . Electron micrographs were taken in minimal dose mode at a magnification of 50 , 000× and a defocus of 1–2 µm . Negatives ( Kodak S0-163 film ) were scanned with a Heidelberg Tango drum scanner with 2419 dpi resolution yielding a pixel size of 4 . 5 Å on the specimen level . Alternatively , images were recorded with a 4k × 4k CMOS camera F-416 ( TVIPS ) at a calibrated magnification of 67 , 200× , resulting in a pixel size of 2 . 32 Å/pixel . Single particles were manually selected using boxer ( Ludtke et al . , 1999 ) . To analyze septins and Gic1 in their non-filamentous state , 4461 particles of septin octamers , 199 particles of septin octamers labeled with anti-Cdc11 antibody and 1282 and 2195 particles of non-polymerizing septin-Cdc10 ( 30-322 ) tetramers and octamers were selected . In the same way , 14 , 407 particles of septin-Cdc42-GDP , 451 particles of septin-Cdc42-GDP labeled with anti-Cdc3 antibody , 1152 particles of septin-Cdc42-GDP + GTP were selected . 3 , 169 particles of septin + GTP , 188 particles of septin + GTP labeled with anti-Cdc3 antibody , 281 particles of septin + GTP and anti-Cdc11 antibody were selected . To analyze septin and Gic1 in filamentous structures , two different sets of particles were selected . The first set focused on the septin filaments ( sf ) between two Gic1 cross-bridges and the second set in the middle of the Gic1 cross-bridge ( gc ) . 2 , 228 sf particles and 1471 gc particles of the septin-Gic1 complex , 1971 sf particles and 1561 gc particles of the Cdc11Δ mutant filaments , 3979 sf particles and 3837 gc particles of the septin-Gic1-Cdc42-GppNHp complex and 180 sf particles of the septin-Gic1 complex labeled with anti-Cdc11 antibody were selected , respectively . Single particles were aligned and classified using reference-free alignment and k-means classification procedures implemented in SPARX ( Hohn et al . , 2007 ) . Briefly , images were normalized to the same mean and standard deviation and band-pass filtered . Images were then centered , subjected to 2D reference-free rotational alignment ( sxali2d ) and k-means classification ( sxk_means ) , with approximately 100-150 images per class . The images were then further aligned and classified by several rounds of multireference alignment ( sxmref_ali2d ) , where only high quality classes were used as references , followed by k-means classification . Classification was performed within an elongated mask including the respective septin density , expanded by 5 pixels . For analysis of antibody binding , all members of class averages showing additional density were merged and subjected to further rounds of alignment and classification , with approximately 15-20 images per class . Shown are characteristic class averages with the lowest intra-class variance in the region of antibody binding . For cryo-ET of the septin-Gic1 and septin-Gic1-Cdc42-GppNHp complexes , 2 μM of septins and 10 μM of Gic1 with or without 10 μM Cdc42-GppNHp were used , respectively . The septin-Gic1 and septin-Gic1-Cdc42-GppNHp complexes were mixed with 5 nm colloidal gold particles . 4 μl aliquots of each preparation were applied to a glow discharged C-flat holey carbon grid ( Protochips Inc . ) and plunge-frozen in liquid ethane using a Cryoplunge3 ( Gatan Inc . ) . Images were collected with a JEOL JEM 3200FSC TEM equipped with an 8k × 8k pixel TVIPS CMOS camera ( F-816 ) at an acceleration voltage of 200 kV and a magnification of 85 , 470× . An in-column omega energy filter was used to improve image contrast by zero-loss filtering with a slit-width of 15 eV . Tilt series were collected at a defocus of ∼ 4-5 μm , covering the range of ±60 in 2° increments and a dosage of about 1e−/Å2 per image . Images were then reduced by 4 × 4 pixel averaging resulting in a pixel size of 7 . 3 Å . Data were processed using the IMOD software package ( Kremer et al . , 1996 ) . Gold particles were tracked as fiducial markers to align the stack of tilted images , and tomograms were reconstructed by weighted back-projection . Selected sub-tomograms were segmented using Amira ( Stalling et al . , 2005 ) and rendering was performed in Chimera ( Pettersen et al . , 2004 ) . The segmentation of tomographic reconstructions was performed by manually tracing structural features through sequential slices of the tomograms . Regions of high density between filaments were assigned to Gic1 or Gic1-Cdc42 respectively . After visual inspection of the first tomograms and subsequent careful segmentation , it became immediately clear that septin-Gic1 complexes are highly heterogeneous concerning their overall arrangement , diameters and even number of septin-filaments , which excluded the possibility of subtomogram averaging . Furthermore , the majority of septin cables reside in a preferred side-view orientation perpendicular to the beam direction ( Figure 3; Video 1 ) . Although such tomograms gave us clear hints about the overall arrangement of septin cables and revealed clear differences between septin-Gic1 and septin-Gic1-Cdc42-GppNHp , the missing wedge artifacts caused the broadening of the septin filaments , making their exact tracing difficult in all slice-directions . However , these initial attempts revealed bending as a characteristic feature of the septin-Gic1 cables ( Figure 3 ) . On grids with relatively thick ice we observed filaments , which changed their orientation and ran parallel to the beam direction at 0° tilt . In some cases , short cables were completely embedded in ice in a top view orientation ( Video 2 and 3 ) . Since the missing wedge artifact in this geometry causes only the elongation of the septin filaments , their exact tracing was straightforward , but the cables were often too short . We therefore scanned thousands of positions in several grids for both samples ( septin-Gic1 and septin-Gic1-Cdc42-GppNHp ) to find complexes of sufficient length and recorded two tomograms of such regions of septin-Gic1 and three tomograms of septin-Gic-Cdc42-GppNHp complexes . Each tomogram contained 7–10 septin-Gic1 or septin-Gic1-Cdc42-GppNHp cables running parallel to the beam at 0° tilt . We extracted all separate cables as subtomograms and processed them as described above . The quality of these raw subtomograms was extremely good and details that would otherwise require the technique of subtomogram averaging ( such as number and shape of filaments , or interaction between filaments ) were clearly discernable even without further processing ( Figure 18A–B ) . 10 . 7554/eLife . 01085 . 025Figure 18 . Processing of subtomograms . ( A and B ) Top and side view of a representative raw subtomogram filtered to 30 Å , respectively . ( C and D ) Segmentation in Amira ( Stalling et al . , 2005 ) . Top and side view of representative slices with selected densities , respectively . ( E and F ) Top and side view of three-dimensionally rendered segments in Amira , respectively . ( G and H ) Top and side view of masked raw densities filtered to 40 Å using the Amira derived segments as masks . DOI: http://dx . doi . org/10 . 7554/eLife . 01085 . 025 We segmented the subtomograms using Amira ( Figure 18C–F ) . The resulting segments were then binarized , expanded , gauss filtered and used as masks to extract the respective density from the raw subtomograms . The extracted densities were low-pass filtered to 40 Å ( Figure 18G–H ) . Therefore , the images shown in Figures 3 , 6 and 18G–H do not represent the typical renderings of manual segmentations , but the extracted raw densities using masks obtained by manual segmentations . To prove that the observed organization of septin-Gic1 or septin-Gic1-Cdc42 complexes is independent of the geometry during image recording , we also processed septin-Gic1-Cdc42 complexes running perpendicular to the beam and parallel to the tilt axis ( Figure 19A–C ) . As expected and described above the septin filaments are broadened and flat . However , the structure clearly shows that the septin-Gic1-Cdc42 complexes have the same organization as the complexes depicted in Figure 3E–G and Figure 6E–G , therefore ruling out that our observations are influenced by the geometry of the complexes . 10 . 7554/eLife . 01085 . 026Figure 19 . Tomography of septin-Gic1-Cdc42 complexes running parallel to the tilt-axis . ( A ) Central slice of representative tomogram of septin-Gic1-Cdc42 complexes . A railroad-like complex running parallel to the tilt axis was extracted ( see inset ) . Scale bars , 200 nm . ( B and C ) Top ( B ) and side view ( C ) of a septin-Gic1-Cdc42-GppNHp complex . The septin filaments and Gic1-Cdc42-GppNHp cross-bridges are depicted in gold and green , respectively . Scale bar , 20 nm . DOI: http://dx . doi . org/10 . 7554/eLife . 01085 . 026 To determine the effect of the missing wedge on our structure we performed tomography-simulations using an idealized railroad-like structure model . The idealized model of a septin-Gic1-Cdc42-GppNHp complex ( Figure 20B–C ) was obtained by fitting several copies of the crystal structure of the mammalian septin trimer ( PDBid: 2QAG ) into the density of septin filaments and placing of GROEL/GROES ( PDBid: 1AON ) into the density corresponding to Gic1-Cdc42-GppNHp cross-bridges in one of our reconstructions ( Figure 20A ) . 10 . 7554/eLife . 01085 . 027Figure 20 . Simulations of electron tomograms of septin-Gic1-Cdc42-GppNHp complexes . ( A ) Side view of a septin-Gic1-Cdc42-GppNHp subtomogram . The septin filaments and Gic1-Cdc42-GppNHp cross-bridges are depicted in gold and green , respectively . ( B ) Model of a septin-Gic1-Cdc42-GppNHp complex obtained by fitting the crystal structure of the mammalian septin trimer ( PDBid: 2QAG , gold ) and GROEL/GROES ( PDBid: 1AON , green ) into ( A ) , shown in three different orientations . ( C ) Simulated EM density map of ( B ) at a resolution of 45 Å . ( G–L ) Simulations of electron tomograms obtained by tilting the model shown in ( C ) in the range of ±60 in 2° increments with its long axis running parallel to the beam ( D ) , parallel ( G ) and perpendicular ( J ) to our microscope’s tilt axis , respectively . ( E , H , and K ) Corresponding projections at −60° , 0° , 60° and ( F , I , and L ) the resulting simulated tomograms , respectively . Note that the tomograms shown in ( I and L ) are obviously affected by missing wedge artifacts , whereas the tomogram in ( F ) ( long axis of the molecule parallel to the beam axis during tilting ) is only slightly stretched in comparison to the original model ( C ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01085 . 027 We then simulated electron tomograms by tilting the model shown in Figure 20C in the range of ±60 in 2° increments with its long axis running parallel to the beam , and parallel and perpendicular to our microscope’s tilt axis , respectively ( Figure 20D , G , J ) . All tomograms are affected by missing wedge artifacts . However , the complexes running parallel to the tilt axis are better resolved ( Figure 20E–F , H–I ) than the ones running perpendicular to the tilt axis ( Figure 20K–L ) . Because of the longitudinal appearance of the septin-Gic1-Cdc42 complexes , tomograms with complexes running parallel to the beam result in a reconstruction that is more similar to the original model ( Figure 20E–F ) compared to complexes running perpendicular to the beam ( Figure 20H–I ) . The filaments are fully comparable to the filaments of the original model . Only the densities of GROEL/GROES ( simulating Gic1-Cdc42 cross-bridges ) ( Figure 20F , shown in green ) show an increase in diameter along the z direction by ∼14% . Moreover , to simulate a more ‘close to reality’ situation , the same procedure was repeated , this time after adding noise and applying CTF to the projections ( Figure 21A–B ) . The volume was then filtered to 45 Å ( Figure 21C ) and compared again to the simulated model ( Figure 21D ) . Even without further processing of this simulated tomogram , both volumes show a high degree of similarity in all aspects with a cross-correlation coefficient of 0 . 95 , further suggesting that tomograms taken under similar conditions are fully sufficient to describe the basic architecture of the filaments . 10 . 7554/eLife . 01085 . 028Figure 21 . Electron tomogram simulation of septin-Gic1-Cdc42-GppNHp with CTF and noise added . ( A ) Projections of the model shown in ( B ) in the range of ±60° in 2° increments , after applying the CTF at a defocus of 4 µm and addition of noise . Note that at 0° , the long axis of the model was running in the z direction ( Figure 19D ) . ( B ) Original model used for generating the reprojections ( Figure 19F ) . ( C ) Reconstruction obtained by back-projecting the images shown in ( A ) . ( D ) Fitting of the simulated tomographic reconstruction ( yellow mesh ) into the original model ( cyan surface ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01085 . 028 For septin filament production , 0 . 3 µM of septin hetero-oligomers ( wild-type , EGFP-tagged , Cdc10Δ , Cdc11Δ or Cdc10 ( 30-322 ) mutants ) in a high-salt buffer ( 500 mM NaCl , 1 mM MgCl2 , 50 mM Tris-HCl pH 7 . 5 , 1 mM DTT ) were dialyzed overnight at 4°C against a low-salt buffer ( 100 mM NaCl , 20 mM Tris-HCl pH 7 . 5 , 1 mM DTT ) . In order to form the septin-Gic1 complexes , 1 . 5 µM of Gic1 was included during dialysis . Since Gic1 forms dimers and septin octamers contain two Cdc10 subunits , the molar ratio ( Cdc10:Gic1 ) is 1 . 25:1 . For antibody decoration , 5 μl of polyclonal antibodies against Cdc11 ( Santa Cruz Biotechnology Inc . ) and Cdc3 ( a gift from Michael Knop , DKFZ-ZMBH , Heidelberg , Germany ) ( diluted 1:100 ) were added to 20 μl of a sample containing either septin-Gic1 complex or the septin octamers ( generated by GTP , Cdc42-GDP or both ) and incubated overnight at 4°C to allow efficient binding . In order to evaluate the effect of Cdc42 on septin complexes , 0 . 1 μM of septin octamers and 0 . 5 μM of Gic1 were used . Cdc42-GppNHp and Cdc42-GDP were used at a concentration equal to Gic1 or at 10 times higher concentrations and incubated for different time intervals ( Figures 7 and 10 ) . To assess the effect of nucleotides on septin complexes , 2 . 4 mM of GMP , GDP , GTP or GppNHp were added to 0 . 5 μM of septin filaments dialyzed alone or with 2 . 5 μM of Gic1 and incubated for 16 hr at 4°C . The same sample was dialyzed to remove the residual GTP and the generated octamers were incubated with 25 μM of Cdc42-GDP and incubated at 4°C for 16 hr . For cryo-ET of the septin-Gic1 and septin-Gic1-Cdc42 complexes , 2 μM of YSC and 10 μM of Gic1 with or without 10 μM Cdc42-GppNHp were used , respectively . The coordinates for the EM structures have been deposited in the EM Data Bank under accession codes EMDB-2504 and EMDB-2505 . | Septins are proteins that provide structural support for cells as they divide . Yeast cells are known to have seven types of septins , which have been widely studied , and 13 different septins have been identified in human cells , although they all seem similar to those found in yeast . Mutations in the genes that carry the genetic code for septins lead to a range of debilitating conditions in humans , including neurodegenerative diseases and male infertility . An enzyme called Cdc42 is thought to have a key role in the formation of ring-like structures by septins before a cell divides , and in the subsequent dismantling of these rings after the cell has divided . A pair of proteins , called Gic1 and Gic2 , is known to be critical for the formation of the septin rings , but the details of the interactions between these two proteins , Cdc42 and the septins are sketchy . Now Sadian et al . have used two imaging approaches—electron microscopy and cryo-electron tomography—to scrutinise the role of Gic1 in greater detail in yeast cells . Gic1 interacts with specific subunits within adjacent septins , and these interactions have the effect of crosslinking the septins and stabilizing them in long filaments . However , high concentrations of the enzyme Cdc42 block the interaction between the Gic1 proteins and the subunits , causing the filaments to be dismantled . A future challenge will be to elucidate the interaction of these proteins in molecular detail using other techniques , in particular X-ray crystallography . | [
"Abstract",
"Introduction",
"Results",
"and",
"discussion",
"Materials",
"and",
"methods"
] | [
"biochemistry",
"and",
"chemical",
"biology",
"structural",
"biology",
"and",
"molecular",
"biophysics"
] | 2013 | The role of Cdc42 and Gic1 in the regulation of septin filament formation and dissociation |
Signal propagation from the cell membrane to a promoter can induce gene expression . To examine signal transmission through sub-cellular compartments and its effect on transcription levels in individual cells within a population , we used the Wnt/β-catenin signaling pathway as a model system . Wnt signaling orchestrates a response through nuclear accumulation of β-catenin in the cell population . However , quantitative live-cell measurements in individual cells showed variability in nuclear β-catenin accumulation , which could occur in two waves , followed by slow clearance . Nuclear accumulation dynamics were initially rapid , cell cycle independent and differed substantially from LiCl stimulation , presumed to mimic Wnt signaling . β-catenin levels increased simultaneously at adherens junctions and the centrosome , and a membrane-centrosome transport system was revealed . Correlating β-catenin nuclear dynamics to cyclin D1 transcriptional activation showed that the nuclear accumulation rate of change of the signaling factor , and not actual protein levels , correlated with the transcriptional output of the pathway .
Imaging of gene expression in individual cells using quantitative microscopy has become a central experimental approach for unraveling the dynamic aspects of mRNA transcription ( Darzacq et al . , 2009; Coulon et al . , 2013; Hager et al . , 2009 ) , and for examining various events of gene expression in real time ( Darzacq et al . , 2007; Huranová et al . , 2010; Brody et al . , 2011; Martins et al . , 2011; Yao et al . , 2006; Mueller et al . , 2010 ) . Cells govern specific transcriptional responses to various stimuli by use of signaling pathways and transducing factors that relay the signal to the promoters of induced target genes ( Purvis and Lahav , 2013; Carmo-Fonseca et al . , 2002 ) . Studies of transcription factor dynamics in single cells in response to signaling have revealed dynamic aspects of transcription factor nuclear translocation and modulation ( Kalo and Shav-Tal , 2013; Yissachar et al . , 2013; Lahav et al . , 2004; Loewer et al . , 2010; Nelson et al . , 2004; Vartiainen et al . , 2007 ) . This study centers on the dynamics of the Wnt/β-catenin signaling pathway and its control of cyclin D1 gene expression , as a model system for examining the dissemination of a signal in the cell and the transcriptional response it elicits . The Wnt/β-catenin canonical signaling pathway is activated by the binding of the Wnt ligand to plasma membrane receptors , thereby triggering downstream events that culminate in the accumulation of β-catenin in the cytoplasm and its translocation into the nucleus ( Clevers and Nusse , 2012; Krieghoff et al . , 2006; Jamieson et al . , 2011 ) . The interaction of β-catenin with transcription factors of the TCF/LEF family in the nucleus modifies gene expression of crucial genes , thus leading to changes in key cellular pathways , such as proliferation , migration and cell fate ( Cadigan and Waterman , 2012 ) . Mechanistically , in the absence of Wnt , cytoplasmic β-catenin protein is constantly degraded ( Stamos and Weis , 2013 ) via the ‘destruction complex’ and proteosomal degradation ( Aberle et al . , 1997; Salomon et al . , 1997; Orford et al . , 1997 ) , thus preventing β-catenin nuclear targeting . In many pathological cases β-catenin is not degraded but accumulates in the nucleus and activates genes , some of which are associated with cell proliferation , such as MYC and cyclin D1 ( Shtutman et al . , 1999; Tetsu and McCormick , 1999 ) . The cyclin D1 protein is a major player in the regulation of the cell cycle ( Johnson and Walker , 1999; Sherr , 1994 ) and its expression is regulated at several levels , including mRNA transcription ( Hosokawa and Arnold , 1998 ) via an elaborate promoter region ( Klein and Assoian , 2008 ) . Cyclin D1 levels were shown to be induced by the Wnt/β-catenin canonical signaling pathway ( Shtutman et al . , 1999; Tetsu and McCormick , 1999; Chocarro-Calvo et al . , 2013; Willert et al . , 2002; Lin et al . , 2000; Porfiri et al . , 1997; Yun et al . , 2005; Torre et al . , 2011 ) . The Wnt/β-catenin signaling pathway has received much experimental attention due to its centrality in gene expression patterning , and its involvement in many cancer types ( Klaus and Birchmeier , 2008 ) . While the endpoint of β-catenin protein stabilization by Wnt signaling has been well studied biochemically , the kinetic aspects of this signaling pathway in living cells , for the β-catenin protein and the target mRNA , remain under-studied . To address this topic we used a cell system for the in vivo visualization and analysis of the mammalian mRNA transcriptional kinetics of single alleles ( Yunger et al . , 2010 , 2013 ) . Whereas , we had previously followed transcription from a single cyclin D1 ( CCND1 ) gene in living human cells , we now set out to examine the real-time behavior of β-catenin during active signaling in a population of living cells , and the effect of signaling on the activity pattern of the target gene .
We previously generated a cell system in which a CCND1 gene was integrated as a single copy allele into human HEK293 cells using Flp-In recombination ( Yunger et al . , 2010 ) . Transcription kinetics on this gene were visualized and quantified using RNA FISH and live-cell imaging techniques . RNA tagging was achieved using a series of MS2 sequence repeats ( Bertrand et al . , 1998 ) inserted into the long 3’UTR of CCND1 . The MS2 repeats form stem-loop structures in the transcribed mRNA . By co-expressing a fluorescent coat protein termed MS2-CP-GFP that binds to the MS2 stem-loops , we obtained fluorescent tagging of the mRNAs produced from this gene , designated CCND1-MS2 ( Yunger et al . , 2010 , 2013 ) . This CCND1-MS2 allele is under the regulation of the endogenous cyclin D1 promoter ( Albanese et al . , 1995 ) and therefore serves as a candidate gene for activation by Wnt/β-catenin signaling ( Fu et al . , 2004 ) . Studying individual living cells , we found that the CCND1-MS2 gene transits between transcriptionally active and non-active states under steady-state conditions ( Yunger et al . , 2010 ) . At steady state , only around 40–50% of the cells were actively transcribing CCND1-MS2 . In order to verify that the Wnt signaling pathway activates the CCND1-MS2 gene we added Wnt3a conditioned medium to the cells and imaged the cells over time . Indeed , on the population level , after 75 min over 80% of cells had shown an actively transcribing CCND1-MS2 gene ( Figure 1a , b , Supplementary file 1a , Video 1 ) . 10 . 7554/eLife . 16748 . 003Figure 1 . Cell system for following β-catenin intra-cellular dynamics and CCND1 transcription in single living cells . ( a ) CCND1-MS2 HEK293 cells stably expressing MS2-GFP-CP were treated with Wnt3a and followed for 6 hr ( every 15 min ) . Several frames from Video 1 are presented . The number of cells exhibiting transcriptionally active CCND1-MS2 genes ( green dot in the nucleus , white arrow ) was counted over time . Scale bar , 10 µm . ( b ) Plots showing the percentage of cells in the population with actively transcribing CCND1-MS2 genes in Wnt3a-treated ( red , n = 98 ) and mock treated ( blue , n = 128 ) cells . Mean ± sd from three fields imaged on different days—see Supplementary file 1a for statistics . ( c ) Immunofluorescence showing that endogenous β-catenin ( green ) is prominent at the cell membrane in untreated HEK293 cells ( top ) and accumulates in the cytoplasm and nucleus following activation by Wnt3a for 2 hr ( bottom ) . Hoechst DNA stain is in blue . ( d ) Similar changes in the subcellular distribution following activation are seen in the YFP-β-catenin low-expressing clone . Bar = 10 μm . ( e ) Western blot time course of endogenous β-catenin and YFP-β-catenin protein accumulation following either Wnt3a ( top ) or LiCl ( bottom ) stimulation . Anti-β-catenin antibody was used for the detection of both β-catenin proteins . Tubulin was used as a loading control . Time 0 is the time point of activator addition . Blots are representative of 3 repeated experiments . The average quantification of 3 repeated experiments is presented in the plots below ( mean ± sd ) . There is no statistical difference between the endogenous and exogenous levels of β-catenin in the two plots . DOI: http://dx . doi . org/10 . 7554/eLife . 16748 . 00310 . 7554/eLife . 16748 . 004Figure 1—figure supplement 1 . Measuring the effect of YFP-β-catenin expression in HEK293 cells . ( a ) Luciferase assay showing the levels of cyclin D1 promoter activation following the transient transfection of YFP-β-catenin into HEK293 cells . p=0 . 003 . ( b ) Overexpression of YFP-β-catenin shows that overexpressed protein localization does not resemble endogenous β-catenin under non-activated conditions , since it is highly present in the nucleus prior to Wnt activation , and does not appear at the membrane . Bar = 10 µm . ( c ) Cell cycle analysis of HEK293 CCND1-MS2 cells with and without YFP-β-catenin . ( d ) Quantification of CCND1-MS2 nascent mRNAs ( left ) ( p=0 . 8 ) and cellular mRNAs ( right ) ( p=0 . 16 ) levels by RNA FISH in HEK293 CCND1-MS2 cell clones with or without YFP-β-catenin ( n = 18 , n = 26 respectively ) . **p<0 . 01 , n . s . = p>0 . 05 . ( e–h ) Example of single molecule mRNA FISH quantification procedure with a probe that hybridizes to the MS2 region in the 3’UTR of the CCND1-MS2 mRNA . ( e ) Raw 3D image ( 76 planes in z stack ) showing the active transcription site ( red ) and single mRNA molecules . Hoechst DNA stain is in blue . ( f ) Deconvolved 3D image . Boxes show the transcription site ( bottom ) and single cellular mRNAs ( top ) . ( g ) Identification of ‘spots’ of single mRNAs and transcription site ( green dots ) by Imaris . ( h ) Generation of a 3D shell for each spot to be taken for intensity measurements . Bar = 10 µm . Then the sum of intensity at the transcription site was divided by the frequent intensity value of a single mRNA . This ratio provided the number of mRNAs associated with the transcription unit , as explained in the Materials and methods section . DOI: http://dx . doi . org/10 . 7554/eLife . 16748 . 00410 . 7554/eLife . 16748 . 005Video 1 . Transcriptional activation of CCND1 in response to Wnt3a . HEK293 CCND1-MS2 cells stably expressing MS2-GFP ( green ) were treated with Wnt3a . The transcribed CCND1 mRNA on the active gene is seen as a bright green dot . The fluorescent signal on the active genes was enhanced using ImageJ 'Spot Enhancing Filter 2D' in order to clearly present the active sites in the movie . Cells were imaged every 15 min for 6 hr . DOI: http://dx . doi . org/10 . 7554/eLife . 16748 . 005 Since an imaging-based approach for studying signaling dynamics requires that relevant molecules be fluorescently tagged , we verified using a luciferase assay , that a YFP-tagged version of β-catenin ( Krieghoff et al . , 2006 ) activates the CCND1 promoter , and observed 2 . 3 fold activation after transient transfection of the protein into the HEK293 CCND1-MS2 cells ( Figure 1—figure supplement 1a ) . We note that HEK293 cells are known to have a low background of β-catenin activity ( Kang et al . , 2012 ) , and are not known to have mutations in proteins associated with Wnt signaling ( Tan et al . , 2012 ) . Immunofluorescence with an antibody to the endogenous β-catenin protein showed normal β-catenin localization at the cell membrane region ( a portion of β-catenin is located in adherens junctions and functions in cell adhesion [Harris and Tepass , 2010] ) , as well as low cytoplasmic levels under non-induced conditions , compared to a predominant increase in cytoplasmic and nuclear distribution after the addition of Wnt3a ( Figure 1c ) . In summary , this cell system enables the measurement of CCND1 transcription activation kinetics in single cells following Wnt signaling . To mimic endogenous β-catenin distribution using YFP-β-catenin , we generated a HEK293 CCND1-MS2 cell clone that stably expressed YFP-β-catenin . Since high overexpression conditions of YFP-β-catenin typically result in increased subcellular distribution and high accumulation in the nucleus prior to any signal ( Figure 1—figure supplement 1b ) , which is in stark contrast to the endogenous β-catenin protein that is observed mainly at the membrane ( Figure 1c ) , we screened and identified a clone that stably expressed low levels of YFP-β-catenin . The clone phenotypically resembled endogenous protein localization and distribution , namely , membrane localization in the non-induced state , and enhanced nuclear localization following Wnt stimulation ( Figure 1d ) . Characterization of endogenous β-catenin and YFP-β-catenin accumulation levels by Western blotting showed that YFP-β-catenin expression levels were ~80% of the endogenous β-catenin , thus doubling β-catenin levels in the cell , and that the accumulation dynamics of both proteins were identical ( Figure 1e ) . The time-scale of β-catenin induction is in agreement with other studies ( Hernández et al . , 2012; Lustig et al . , 2002 ) . The addition of YFP-β-catenin to the cell clone did not influence the cell cycle or CCND1 expression at steady state as quantified by single molecule RNA FISH ( Yunger et al . , 2010 , 2013 ) ( Figure 1—figure supplement 1c–h ) . To understand the intra-cellular dynamics of β-catenin in a cell population under living cell conditions , cells were imaged for over 12 hr . Rapid nuclear accumulation of β-catenin was observed in most cells that were stimulated with Wnt3a , compared to no change in β-catenin levels in control cells that received mock conditioned medium without Wnt3a ( Figure 2a , b , Video 2 ) . Rising levels of β-catenin in the cytoplasm and the nucleus were detected 15 min after Wnt3a addition , and the accumulation peak was observed 2–3 hr later ( Figure 2c ) , during which β-catenin levels increased 3-fold compared to the initial state . Recombinant Wnt3a ( 200 ng/ml ) showed the same dynamics ( data not shown ) . The rate at which β-catenin levels increased in the nucleus was faster than in the cytoplasm , leading to a higher nucleus/cytoplasm ( N/C ) protein ratio , whereas in the control cells there was no change ( Figure 2d ) . 10 . 7554/eLife . 16748 . 006Figure 2 . The dynamics of β-catenin accumulation following Wnt3a activation in cell populations . Frames from live-cell movies ( Video 2 ) showing YFP-β-catenin dynamics in cells treated with ( a ) mock conditioned medium or ( b ) Wnt3a for 12 hr . Red bordered frames compare between the 0 min and 120 min time points . Bar = 20 μm . ( c ) The relative average intensity of β-catenin measured in the cytoplasm ( n = 24 ) and nucleus ( n = 31 ) of cells treated with Wnt3a for 12 hr , compared to mock-treated control cells ( n = 13 ) . ( d ) Nucleus to cytoplasm ratio ( N/C ) of fluorescence intensities over 12 hr from c . The initial ratio was designated as 1 . Inset plot shows the statistical significance p values ( t test ) at each time point between the two treatments over the experiment time course . ( e ) The rate of change in β-catenin levels ( ΔI/Δt ) , during accumulation or degradation , in the cytoplasm and nucleus over time in cells from c . DOI: http://dx . doi . org/10 . 7554/eLife . 16748 . 00610 . 7554/eLife . 16748 . 007Figure 2—figure supplement 1 . FRAP and FLIP measurements of YFP-β-catenin import and export dynamics . ( a ) Frames showing one pre-bleach frame , the bleach of the YFP-β-catenin in the nucleus ( top ) or cytoplasm ( bottom ) of a Wnt3a-treated cell ( 2 hr , arrows point to bleached region ) , and frames following the recovery of signal over time ( frame every 1 . 5 s for 8 min ) . Bar = 10 μm . ( b ) Averaged data plot of FRAP recovery import curves from Wnt3a-treated cells ( n = 27 , red curve ) , and transiently overexpressing YFP-β-catenin cells ( n = 33 , blue curve ) . Pink curve shows the decline in YFP-β-catenin in the cytoplasm of Wnt3-treated cells concurrent with nuclear import ( red curve ) . ( c ) Averaged data plot of FRAP recovery export curves ( green ) from Wnt3a-treated cells ( n = 14 ) , compared to the import curve ( red ) . ( d ) FLIP curves for Wnt3a-treated cells photobleached continuously in the nucleus to show import rates from the cytoplasm ( n = 15 , red curve ) compared with cells photobleached continuously in the cytoplasm to show export rates from the nucleus ( n = 16 , blue curve ) . Statistics can be found in Supplementary file 1 . ( e ) Data from Figure 2c ( blue dots ) of nuclear YFP-β-catenin accumulation were fit with a two-phase exponential ( red curve ) . DOI: http://dx . doi . org/10 . 7554/eLife . 16748 . 00710 . 7554/eLife . 16748 . 008Video 2 . YFP-β-catenin dynamics at steady state and after Wnt3a activation . HEK293 CCND1-MS2 cells stably expressing YFP-β-catenin were treated with Wnt3a ( top ) and showed nuclear and cytoplasmic accumulation of YFP-β-catenin , followed by slow egress . No change in YFP-β-catenin levels was seen in mock-treated cells ( bottom ) . Right – The YFP signal is pseudo-colored using ImageJ ‘Royal’ look-up table to show YFP-β-catenin levels . Cells were imaged every 15 min for 510 min . DOI: http://dx . doi . org/10 . 7554/eLife . 16748 . 008 Analyzing the rate of change in β-catenin levels in the nucleus and cytoplasm over time ( ΔI/Δt ) showed that the accumulation was comprised of two phases; an initial rapid one , in which the peak of the change in accumulation was reached 60 min after induction , and a second accumulation phase in which cellular β-catenin continued to amass but at a declining rate up until 180 min ( Figure 2e ) . Subsequently , the rate of change turned negative , meaning that β-catenin levels were declining , probably due to degradation . In control cells , the rate of change in β-catenin remained unaltered . To examine whether the dynamics of nuclear entry of β-catenin were modified during Wnt activation and how they compared to β-catenin shuttling out of the nucleus , we used fluorescence recovery after photobleaching ( FRAP ) . Nuclei of cells showing nuclear β-catenin , either after 2 hr of Wnt3a activation or transiently overexpressing β-catenin , were photobleached , and nuclear import of β-catenin was monitored over time ( Figure 2—figure supplement 1a top ) . The dynamics were relatively slow , however , the import rate under Wnt3a conditions was more rapid than transient overexpression , showing the advantage of measurements performed at low expression conditions ( Figure 2—figure supplement 1a , b , Supplementary file 1b ) . The incomplete recovery of YFP-β-catenin during the FRAP time-course meant that a significant population of β-catenin molecules had already accumulated and had been retained in the nucleus prior to photobleaching . Next , we photobleached the cytoplasm and found that the rate of β-catenin shuttling out from the nucleus was slower than the import rate ( Figure 2—figure supplement 1a bottom , c , Supplementary file 1c ) . Similarly , fluorescence loss in photobleaching ( FLIP ) , either in the nucleus or in the cytoplasm , showed that β-catenin shuttling out of the nucleus was slower than its nuclear entry ( Figure 2—figure supplement 1d , Supplementary file 1d ) . Altogether , the data suggest that Wnt signaling causes a transient shift in the dynamic interplay between β-catenin stabilization and degradation processes , towards protein stabilization and accumulation . The averaged population data obtained from living cells presented above ( Figure 2 ) are in agreement with biochemical data as seen by Western blotting of protein extracts from large cell populations , showing the accumulation of β-catenin beginning from around 30 min after Wnt and peaking at 3 hrs ( Hernández et al . , 2012; Li et al . , 2012 ) . However , the averaged behavior of a population does not necessarily represent the actual dynamics in individual cells . Examining the dynamic behavior of β-catenin accumulation in the nucleus and cytoplasm of individual cells after Wnt3a showed that although an increase in β-catenin levels was initiated in most cells , the subsequent dynamics were variable ( Figure 3a , b , Video 3 ) . For instance , comparing cells 1 , 2 and 4 ( Figure 3a ) showed a major and rapid wave of β-catenin nuclear accumulation in cell 1 ( 30–165 min ) that subsided and then mildly rose again ( 465–585 min ) ; a similar range of events occurred in cell 2 but the two waves were less intense and the second wave occurred earlier compared to cell 1 ( first wave 30–150 min , second wave 330–435 min ) ; in contrast , cell four showed a longer accumulation period ( 30–240 min ) . Cells 3 and 6 showed slow nuclear accumulation , peaking late only after 825 min and 525 min , respectively , from Wnt3a stimulation . This analysis showed that the dynamic behavior of β-catenin in the cytoplasm and the nucleus was highly similar within the same cell , but that the time-frames of accumulation could be quite different between individual cells , some showing two cycles of nuclear accumulation . In these cases , the first cycle of accumulation lasted 360 min on average and the second cycle 180 min on average . 10 . 7554/eLife . 16748 . 009Figure 3 . Variability of β-catenin accumulation dynamics following Wnt3a activation in individual cells . ( a ) Frames from time-lapse Video 3 showing YFP-β-catenin accumulation in a population of cells . The YFP signal is pseudo-coloured using ImageJ ‘Green Fire Blue’ look-up table . White and yellow arrows point to cells in which β-catenin levels increase and decrease twice during the movie . The pink arrow points to centrosomal accumulation . Bar = 10 μm . ( b ) β-catenin levels in the nucleus ( left ) and cytoplasm ( right ) in individual cells ( as numbered in a ) are plotted in different colors . The grey background plots show the complete set of plots from all the cells . Maximum β-catenin intensity in each cell was normalized to 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 16748 . 00910 . 7554/eLife . 16748 . 010Figure 3—figure supplement 1 . β-catenin accumulation dynamics in response to LiCl activation in individual cells . ( a ) Frames from time-lapse Video 4 showing YFP-β-catenin accumulation in a population of cells . The YFP signal is pseudo-colored using the ImageJ ‘Green Fire Blue’ look-up table . Bar = 10 μm . ( b ) β-catenin levels in the nucleus ( top ) and cytoplasm ( bottom ) of individual cells ( as numbered in a ) are plotted in different colors . The grey background plots show the complete set of plots from all the cells . The maximum intensity of β-catenin in each cell was normalized to 1 . ( c ) The relative average intensity of β-catenin measured in the cytoplasm ( n = 17 ) and nucleus ( n = 18 ) of individual cells treated with LiCl for 12 hr , compared to Wnt3a-treated cells ( from Figure 2 ) . ( d ) The rate of change in β-catenin levels ( ΔI/Δt ) accumulation or degradation in the cytoplasm and nucleus over time in cells from c . ( e ) Frames from a time-lapse movie showing YFP-β-catenin accumulation in a population of Wnt3a+MG132-treated cells . The YFP signal is pseudo-colored using the ImageJ ‘Green Fire Blue’ look-up table . Bar = 10 μm . ( f ) The relative average intensity of β-catenin measured in the centrosome , membrane , cytoplasm and nucleus of LiCl-treated cells . DOI: http://dx . doi . org/10 . 7554/eLife . 16748 . 01010 . 7554/eLife . 16748 . 011Figure 3—figure supplement 2 . The relationship between YFP-β-catenin levels of accumulation and time of Wnt3a activation . ( a ) Frames from a time-lapse movie showing YFP-β-catenin accumulation in a population of cells in the field . The YFP signal is pseudo-colored using the ImageJ ‘Royal’ look-up table . Red arrows point to cells with very high β-catenin levels . Bar = 10 μm . Plots showing the relative maximal levels of β-catenin measured in the nuclei of ( b ) Wnt3a-treated ( n = 31 ) or ( c ) LiCl-treated cells ( n = 18 ) . Order of cells is according to increasing relative intensities . Plots showing the time from the addition of the activator until reaching the maximal levels of β-catenin in the same set of ( d ) Wnt3a-treated or ( e ) LiCl-treated cells . ( f , g ) The respective correlation plots and scores for a Pearson correlation analysis between the maximum intensity in each cell and the time to reach the highest accumulation . ( h ) The integral of the fluorescence values in the six cells ( from Figure 3 ) showing the total accumulation levels over time during the whole observation period ( left ) . The right-hand plot shows the differences between accumulation in the cells at earlier times . DOI: http://dx . doi . org/10 . 7554/eLife . 16748 . 01110 . 7554/eLife . 16748 . 012Video 3 . YFP-β-catenin dynamics in individual cells . HEK293 CCND1-MS2 cells stably expressing YFP-β-catenin were treated with Wnt3a , and the dynamics of the protein were observed in individual cells . The YFP signal is pseudo-colored using ImageJ ‘Green Fire Blue’ look-up table to show YFP-β-catenin levels . Cells were imaged every 15 min for 825 min . DOI: http://dx . doi . org/10 . 7554/eLife . 16748 . 012 The similar dynamics of decline in β-catenin levels in the nucleus and the cytoplasm suggests that β-catenin is not simply shuttling in and out of the nucleus , but rather reflects an enhanced activity of the degradation arm controlling β-catenin levels . To test this , we added lithium chloride ( LiCl , 20 mM ) , a glycogen synthase kinase-3β ( GSK3β ) inhibitor that mimics Wnt signaling ( Klein and Melton , 1996; Hedgepeth et al . , 1997 ) . Indeed , LiCl caused β-catenin nuclear and cytoplasmic accumulation , but the dynamics were completely different than Wnt3a ( Figure 3—figure supplement 1a , b , and Video 4 ) . β-catenin accumulation occurred synchronously and continuously throughout 10–11 hr in all cells , and only then did the accumulation cease . The increasing accumulation rate of change ( ΔI/Δt ) in the nucleus and cytoplasm continued for 10 hr , compared to 3 hr , in response to Wnt3a ( Figure 3—figure supplement 1c , d ) . The levels of β-catenin were 4-fold higher in LiCl treated cells compared to Wnt3a . Since LiCl prevents β-catenin degradation , we hypothesized that Wnt3a treatment together with the proteasome inhibitor MG132 , which stabilizes β-catenin , but not through GSK3β phosphorylation , should have a similar effect on β-catenin dynamics . Indeed , accumulation dynamics under Wnt3a+MG132 were similar to LiCl treatment ( Figure 3—figure supplement 1e ) . Treatment with MG132 without Wnt3a showed the same dynamics ( data not shown ) . When the curve describing the dynamics of β-catenin in response to Wnt3a ( Figure 2c ) was fitted with a two-phase exponential fit that describes production and degradation ( Figure 2—figure supplement 1e ) , we found linear accumulation in the first phase , showing that degradation was very low , as expected ( Hernández et al . , 2012 ) . β-catenin production rates did not change significantly during the accumulation and clearance phases , whereas , the degradation rate became predominant during the clearance phase . β-catenin degradation had a characteristic time of 2 . 75 hr . These data exemplify the difference between a signaling molecule and a chemical that target the same signaling pathway . While drug action is less influenced by endogenous molecules , a signaling molecule will relay a transient signaling effect , depending on the level of other signaling molecules that are present in the cell at the time of induction . 10 . 7554/eLife . 16748 . 013Video 4 . YFP-β-catenin dynamics in response to LiCl . HEK293 CCND1-MS2 cells stably expressing YFP-β-catenin were treated with LiCl and increased accumulation of the protein was observed . The YFP signal is pseudo-colored using ImageJ ‘Green Fire Blue’ look-up table to show YFP-β-catenin levels . Cells were imaged every 15 min for 825 min . DOI: http://dx . doi . org/10 . 7554/eLife . 16748 . 013 Since the maximum levels of β-catenin accumulation differed between cells in the population ( Figure 3—figure supplement 2a–e ) , and we could identify intense and prolonged accumulation in some Wnt3a-treated cells , we examined whether there was a correlation between the time to reach the maximum level and the peak of the response . However , a low correlation score ( 0 . 28 ) was observed for the Wnt3a-treated cells , and a more prominent correlation score ( 0 . 53 ) in LiCl-treated cells ( Figure 3—figure supplement 2f , g ) . The latter was expected due to the continuous accumulation over time . But for Wnt3a treatment , this meant that a longer Wnt3a signaling response did not necessarily result in higher levels of β-catenin accumulation . Moreover , calculating the integral of the fluorescence signal that accumulated over the whole observation period in a cell population ( from Figure 3 ) , showed that the total accumulation in most cells was similar ( Figure 3—figure supplement 2h ) , and that differences between single cells were pronounced mainly at earlier time points of the response . Cluster analysis of the dynamic behavior of β-catenin in individual living cells , shows the dramatic difference between Wnt signaling activation by Wnt3a compared to LiCl ( Figure 4a; membrane and centrosome will be discussed below ) . ~80% of the cells showed similar dynamics ( e . g . Figure 2c ) and ~20% portrayed different behavior patterns ( e . g . Figure 3 ) . In order to determine whether the variabilities in β-catenin dynamics in the cell population in response to Wnt3a , may be due to the cell cycle stage , we examined time-lapse movies in which cells had undergone mitosis , and in which daughter cells could be identified . For example , in the population of cells seen accumulating β-catenin in response to Wnt3a in Figure 4b ( Video 5 ) , there were two dividing cells at the beginning of the movie , both with low β-catenin levels prior to mitosis . In the daughter cells originating from the top dividing cell there was low β-catenin accumulation , whereas in the bottom dividing cell , one daughter cell responded rapidly and accumulated very high levels of β-catenin , while the other daughter cell responded later and accumulated to low levels ( Figure 4b–e ) . In summary , we could not detect a pattern of β-catenin accumulation in daughter cells . 10 . 7554/eLife . 16748 . 014Figure 4 . Variability of β-catenin dynamics in the cell population and during the cell cycle . ( a ) Heat map and cluster analysis of normalized β-catenin accumulation dynamics in sub-cellular compartments following Wnt3a ( top , n ( nucleus ) = 31 , n ( cytoplasm ) = 24 , n ( membrane ) = 21 , n ( centrosome ) = 11 ) or LiCl ( bottom , n ( nucleus ) = 18 , n ( cytoplasm ) = 17 , n ( membrane ) = 9 , n ( centrosome ) = 14 ) treatments . Data were taken from live-cell movies with each column representing one cell , and rows representing time from Wnt addition . Relative levels of β-catenin are depicted from low ( green ) to high ( red ) . Hierarchical cluster analysis depicted above the plots shows the homogenous behavior in LiCl-treated cells and heterogenous behavior in Wnt3a-treated cells . Most cells reach maximal levels of β-catenin within 2–3 hr . ( b ) ( Top ) Frames from time-lapse Video 5 showing YFP-β-catenin accumulation in a population of cells . The YFP signal is pseudo-colored using the ImageJ ‘Fire’ look-up table . Boxes denote cells that go through mitosis , and enlargements are shown below . Green arrows point to mother cells , and yellow and white arrows point to the daughter cells . Bar = 10 μm . Plots showing the relative intensity levels of YFP-β-catenin in the cytoplasm and nucleus of the ( c ) top and ( d ) bottom daughter cells of each cell division . ( e ) Plot comparing the relative intensity levels in the nuclei of the four daughter cells . DOI: http://dx . doi . org/10 . 7554/eLife . 16748 . 01410 . 7554/eLife . 16748 . 015Figure 4—figure supplement 1 . YFP-β-catenin dynamics during the cell cycle in Wnt3a induced cells . HEK293 CCND1-MS2 YFP-β-catenin cells were stably infected with the Fucci system ( mCherry-Cdt1 and AmCyan1-Geminin ) . Cdt1 levels peak during G1 ( red cells ) , and as cells transition into S , Cdt1 levels decline and Geminin levels rise ( cyan cells ) , remaining high from G2 onwards . ( a ) Frames from Video 6 . Before Wnt3a treatment ( time 0 min ) , β-catenin levels are low in the cytoplasm and the nucleus of all cells marked by arrowheads ( cells in G1 or G2 ) . At time 255 min after Wnt3a , the cells marked with blue , green and pink arrowheads show an increase in the β-catenin levels in response to Wnt signaling . The cell marked with a white arrowhead has not responded yet . At time 780 min , the cell marked with a pink arrowhead has gone through mitosis and the three cells marked by blue , green and white arrowheads have similar β-catenin levels . At time 1065 min , the cells marked by white and blue arrowheads are increasing further , while the cell marked with the green arrowhead is not changing . ( b ) Frames from Video 7 . Before Wnt3a treatment ( time 0 min ) , all the cells marked with arrowheads ( green , blue , white and pink ) are in G2 . At time 240 min after Wnt3a , the cell marked by a green arrowhead has gone through mitosis and the daughter cells have similar β-catenin levels . At time 420 min , all four cells have gone through mitosis . β-catenin levels in each of the two daughter cells in all four cases are similar to each other . At time 705 min , the daughter cells marked by blue , green and pink arrowheads are different from each other . The cells marked with white arrowheads have similar levels . Bottom rows are the same frames without Fucci labels . Bar = 10 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 16748 . 01510 . 7554/eLife . 16748 . 016Video 5 . YFP-β-catenin dynamics following Wnt3a activation during cell division . HEK293 CCND1-MS2 cells stably expressing YFP-β-catenin were treated with Wnt3a , and the dynamics of the protein in the nucleus were followed over time . Two cells that undergo mitosis were observed in the field . The levels of the protein in the daughter cells formed from the upper cell were low ( also note the appearance and division of the centrosome detected via YFP-β-catenin ) . In comparison , in the bottom mitotic cell , one daughter cell accumulated high YFP-β-catenin levels very rapidly , while the other responded slowly and had very low levels . The YFP signal is pseudo-colored using ImageJ ‘Fire’ look-up table to show YFP-β-catenin levels . Cells were imaged every 15 min for 225 min . DOI: http://dx . doi . org/10 . 7554/eLife . 16748 . 016 To examine the cell cycle and Wnt response more closely in a large population of living cells we used the Fucci system ( Videos 6 and 7 ) , which uses two fluorescent cell cycle markers to identify cell cycle phases ( Sakaue-Sawano et al . , 2008 ) . We introduced the Fucci molecules into the CCND1-MS2 cells containing YFP-β-catenin . The cells did not show any special pattern of YFP-β-catenin accumulation relative to the cell cycle stage ( Figure 4—figure supplement 1a ) , and cells passing through mitosis also exhibited different accumulation levels in the mother cell and between daughter cells ( Figure 4—figure supplement 1b ) . In summary , we did not identify a cell cycle dependent pattern of YFP-β-catenin levels in response to Wnt . 10 . 7554/eLife . 16748 . 017Video 6 . YFP-β-catenin dynamics following Wnt3a activation during the cell cycle . HEK293 CCND1-MS2 cells stably expressing YFP-β-catenin ( yellow ) and the Fucci markers for G1 ( red ) and G2 ( cyan ) , were treated with Wnt3a , and the dynamics of the protein were followed over time . Cells were imaged every 15 min for 1065 min . DOI: http://dx . doi . org/10 . 7554/eLife . 16748 . 01710 . 7554/eLife . 16748 . 018Video 7 . YFP-β-catenin dynamics following Wnt3a activation during cell division . HEK293 CCND1-MS2 cells stably expressing YFP-β-catenin ( yellow ) and the Fucci markers for G1 ( red ) and G2 ( cyan ) , were treated with Wnt3a , and the dynamics of the protein were followed over time in four cells that undergo mitosis . Cells were imaged every 15 min for 705 min . DOI: http://dx . doi . org/10 . 7554/eLife . 16748 . 018 β-catenin is normally present in the adherens junctions proximal to the cell membrane , and is bound to E-cadherin in the membrane and to α-catenin , which mediates the connection between the adherens junction and the actin cytoskeleton ( Yap et al . , 1997; Brembeck et al . , 2006 ) . Not much is known about the subcellular localization of this β-catenin population in response to Wnt . Before treatment , β-catenin was observed as a string of punctate sub-regions distributed along the cell outline only at cell-cell contacts ( Figure 3a , Video 3 ) . Since we could detect changes in the intensity of the puncta after Wnt , we followed the intensity of β-catenin at the membrane during Wnt activation and found an increase with similar dynamics to the cytoplasmic and nuclear sub-populations ( Figure 5a , Video 8 ) . There was no obvious reduction in the membrane levels even after many hours ( Figure 5b ) . However , the relative increase at the membrane was lower than the nucleus and the cytoplasm , and the rate of β-catenin accumulation ( ΔI/Δt ) at the membrane was less rapid than the nuclear accumulation rates ( Figure 5b , c ) . LiCl caused longer β-catenin accumulation times and significantly higher accumulation at the membrane ( Figure 5c , d ) . 10 . 7554/eLife . 16748 . 019Figure 5 . The dynamics of β-catenin accumulation at the membrane following Wnt3a activation . ( a ) Frames from time-lapse Video 8 showing YFP-β-catenin accumulation at the cell membrane . The YFP signal is pseudo-colored using the ImageJ ‘Green Fire Blue’ look-up table . Bar = 10 μm . ( b ) The relative average intensity of β-catenin measured in the membrane ( n = 21 ) , cytoplasm and nucleus ( from Figure 2 ) of Wnt3a-treated cells . ( c ) The rate of change in β-catenin levels ( ΔI/Δt ) accumulation or degradation in the membrane , cytoplasm and nucleus over time in Wnt3a- and LiCl-treated cells . ( d ) The relative average intensity of β-catenin measured in the membrane , cytoplasm and nucleus of LiCl-treated cells . DOI: http://dx . doi . org/10 . 7554/eLife . 16748 . 01910 . 7554/eLife . 16748 . 020Figure 5—figure supplement 1 . FRAP measurements of YFP-β-catenin dynamics at adherens junctions . ( a ) Frames showing one pre-bleach frame , the bleach of the YFP-β-catenin in the membrane region of a Wnt3a-treated cell , and frames following the recovery of the signal over time . Circle denotes the bleached region . Bar = 10 μm . ( b ) Averaged data plot of FRAP recovery curves from mock-treated ( control , n = 21 ) , Wnt3a-treated ( n = 32 ) and LiCl-treated cells ( n = 18 ) . Statistics can be found in Supplementary file 1e , f . DOI: http://dx . doi . org/10 . 7554/eLife . 16748 . 02010 . 7554/eLife . 16748 . 021Video 8 . YFP-β-catenin dynamics at the cell membrane following Wnt3a activation . HEK293 CCND1-MS2 cells stably expressing YFP-β-catenin were treated with Wnt3a , and the dynamics of the protein at the membrane were followed over time , and were similar to the nucleus and cytoplasm accumulation . The YFP signal is pseudo-colored using ImageJ ‘Green Fire Blue’ look-up table to show YFP-β-catenin levels . Cells were imaged every 15 min for 1065 min . DOI: http://dx . doi . org/10 . 7554/eLife . 16748 . 021 To examine if Wnt signaling changed the dynamics of β-catenin at the membrane we performed FRAP experiments on this region and found that the recovery dynamics were slow and indicative of slow exchange of β-catenin molecules at the membrane . Yet , similar recovery in unactivated , Wnt3a-treated and LiCl-treated cells was observed , meaning that there was no change in the dynamics of protein exchange but rather an increase in the number of β-catenin molecules in the membrane-bound fraction ( Figure 5—figure supplement 1 , Supplementary file 1e , f ) . In many of the Wnt-induced cells that were followed in the live-cell movies we noticed the appearance of β-catenin in a single prominent dot ( Figure 3a , Video 3 ) . β-catenin can localize at the centrosome during interphase and mitosis , and functions in centriolar cohesion ( Kaplan et al . , 2004; Hadjihannas et al . , 2010; Bahmanyar et al . , 2008 ) . Since the β-catenin dot was in proximity to the nucleus , and since the centrosome is juxtaposed to the nucleus , we examined if centrosomal accumulation of β-catenin was occurring . Indeed , movies of dividing cells demonstrated that each daughter cell received one β-catenin-labeled body after division , reminiscent of centrosome behavior ( Figure 6a , Video 5 and 9 ) . Immunofluorescence of pericentrin ( a centrosome marker ) , together with either endogenous β-catenin or YFP-β-catenin , showed accumulation of β-catenin at the centrosome following activation ( Figure 6b ) . 10 . 7554/eLife . 16748 . 022Figure 6 . Accumulation of β-catenin at the centrosome after Wnt3a activation . ( a ) Frames from time-lapse Video 9 showing YFP-β-catenin accumulation at the centrosome ( white arrowheads ) and after cell division . Bar = 10 μm . ( b ) The colocalization ( white arrowheads ) of YFP-β-catenin ( top ) or endogenous β-catenin ( bottom ) with the centrosomal marker pericentrin ( red immunofluorescence ) in untreated and LiCl-treated cells . Hoechst DNA stain is in blue , and DIC in grey . Boxes show enlarged centrosomal areas . Bar = 10 μm . ( c ) The relative average intensity of YFP-β-catenin measured in the centrosome ( n = 11 ) , membrane , cytoplasm and nucleus ( from Figure 2 and 5 ) of Wnt3a-treated cells . Correlation scores ( r ) between the nucleus ( n ) , cytoplasm ( c ) , membrane ( m ) and centrosome ( ce ) YFP-β-catenin levels are presented at the bottom . ( d ) The rate of change in YFP-β-catenin levels ( ΔI/Δt ) accumulation or degradation in the centrosome , membrane , cytoplasm and nucleus over time in Wnt3a- and LiCl-treated cells . ( e ) Plots of YFP-β-catenin levels in the sub-cellular compartments of individual cells ( from Figure 3 ) . Boxes show the correlation scores ( r ) between the nucleus ( n ) , cytoplasm ( c ) , membrane ( m ) and centrosome ( ce ) . DOI: http://dx . doi . org/10 . 7554/eLife . 16748 . 02210 . 7554/eLife . 16748 . 023Figure 6—figure supplement 1 . Detachment of membranal YFP-β-catenin puncta and movement towards the centrosome . ( a ) Frames from Video 10 showing the tracks of several YFP-β-catenin membranal puncta ( colored tracks ) moving from the membrane region towards the centrosome area ( red circle ) . Time is minutes after addition of Wnt3a . YFP signal is shown in negative greyscale colors . Bar = 10 μm . ( b ) Maximum time projections of movements of membranal YFP-β-catenin puncta ( arrows ) towards the centrosome region ( circles ) in four different cells . The YFP signal is pseudo-colored using the ImageJ ‘Green Fire Blue’ look-up table . Top row are Wnt3a-treated cells . Bottom row are Wnt3a + MG132-treated cells . Bar = 10 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 16748 . 02310 . 7554/eLife . 16748 . 024Figure 6—figure supplement 2 . Summary of FRAP measurements of YFP-β-catenin dynamics in subcellular compartments in response to Wnt3a treatment . ( a ) Frames showing one pre-bleach frame , the bleach of the YFP-β-catenin in the centrosome of a Wnt3a-treated cell , and frames following the recovery of signal over time . Circle denotes the bleached region . Bar = 10 μm . ( b ) Averaged data plot of FRAP recovery curves in the cytoplasm ( n = 24 ) , nucleus ( n = 25 ) , membrane ( n = 32 ) and centrosome ( n = 13 ) . Membrane recovery from Figure 5—figure supplement 1 and the import rate to nucleus from Figure 2—figure supplement 1 are also plotted . DOI: http://dx . doi . org/10 . 7554/eLife . 16748 . 02410 . 7554/eLife . 16748 . 025Video 9 . YFP-β-catenin accumulation at the centrosome following Wnt3a activation . HEK293 CCND1-MS2 cells stably expressing YFP-β-catenin were treated with Wnt3a , and the dynamics of the protein at the centrosome were observed in parallel to the accumulation in the nucleus and cytoplasm . The separation of the centrosome in a cell during division can be seen after the 960 time point . Cells were imaged every 15 min for 1005 min . DOI: http://dx . doi . org/10 . 7554/eLife . 16748 . 025 The accumulation dynamics of β-catenin at the centrosome occurred in parallel to the accumulation seen in the nucleus , cytoplasm and adherens junctions . However , centrosomal levels were significantly higher , five-fold higher compared to the initial state ( Figure 6c ) . The rates of change were the highest and most rapid of all measured cell compartments ( Figure 6d ) . LiCl also led to β-catenin localization at the centrosome , but here too with very different dynamics from Wnt3a ( Figure 6d; Figure 3—figure supplement 1f ) . To obtain a more general outlook of the changes in β-catenin levels in all four compartments , we performed a correlation analysis ( Figure 6c , e ) . As was seen in individual cells , the highest correlation in accumulation dynamics following Wnt3a , was observed between the cytoplasm and the nucleus , whereas the lowest correlation was between the centrosome and the membrane . Interestingly , in some cells we observed β-catenin puncta detaching from the membrane and traveling in the cell ( Videos 10 and 11 ) . When these structures were tracked during movement in the cell , they usually ended up at the centrosome ( Figure 6—figure supplement 1 ) . This phenomenon was frequently seen in cells treated with Wnt3a , LiCl and MG132 , and less frequently in unactivated cells . We did not observe a correlation with the timing of Wnt addition , and perhaps detection was easier after Wnt due to the increase of β-catenin at the membrane following stimulation . Tracking of the detached β-catenin puncta showed that they reached the centrosome between 30 to 90 min after detachment . To examine whether the residence times of β-catenin molecules at the centrosome resembled the membrane region , we performed FRAP analysis , which showed very rapid recovery kinetics at the centrosome , in comparison to all other cell regions ( Figure 6—figure supplement 2 ) . This implied that β-catenin duration at the centrosome is short-lived , with a half-time of fluorescence recovery ( t1/2 ) of 1 . 9 s , similar to other centrosomal components ( Hames et al . , 2005 ) . Altogether , this suggests that the molecular interactions of β-catenin at the membrane in adherens junctions are significantly more stable than at the centrosome , where the exchange of β-catenin molecules is highly rapid . 10 . 7554/eLife . 16748 . 026Video 10 . YFP-β-catenin puncta move from the membrane to the centrosome . HEK293 CCND1-MS2 cells stably expressing YFP-β-catenin were treated with Wnt3a . At the 300 min time point , a series of YFP-β-catenin puncta can be tracked ( track colors ) moving from the membrane to the centrosome . An inverted presentation of the movie shows the movie puncta ( black dots ) . Cell was imaged every 15 min for 1005 min . DOI: http://dx . doi . org/10 . 7554/eLife . 16748 . 02610 . 7554/eLife . 16748 . 027Video 11 . YFP-β-catenin puncta move from the membrane to the centrosome . HEK293 CCND1-MS2 cells stably expressing YFP-β-catenin were treated with Wnt3a and MG132 . At the 165 min time point , a series of YFP-β-catenin puncta can be tracked ( track colors ) moving from the membrane to the centrosome . The YFP signal is pseudo-colored using ImageJ ‘Green Fire Blue’ look-up table to show YFP-β-catenin levels . Cells were imaged every 15 min for 1065 min . DOI: http://dx . doi . org/10 . 7554/eLife . 16748 . 027 We next examined the influence of Wnt signaling dynamics on CCND1 gene activity . As shown ( Figure 1a ) , a significant increase in the percentage of cells actively transcribing CCND1-MS2 could be seen starting 15 min post-activation , and peaking after 75–90 min . Cells returned to steady state activity levels after 6 hr . We examined several parameters of the transcriptional response . First , we measured the time for an active CCND1-MS2 transcribing gene to appear in the population . In the control unstimulated population ( mock conditioned medium ) , after 120 min most cells had activated the gene once , whereas in Wnt3a-induced cells , gene activation in the population was reached more quickly , already after 60 min . The response time for CCND1 activation following Wnt3a was also short , ranging at 15 min ( Figure 7a–b ) . This meant that Wnt signaling increased the probability of CCND1 to initiate transcription . 10 . 7554/eLife . 16748 . 028Figure 7 . Measuring the transcriptional response of CCND1-MS2 to Wnt3a activation in living cells . ( a ) The percentage of cells in a population of either mock-treated ( blue ) or Wnt3a-activated cells ( red ) showing an active CCND1-MS2 transcribing gene , over time . ( b ) The promoter response time of CCND1-MS2 activation from the addition of Wnt3a ( n = 27 ) or in mock-treated conditions ( n = 22 ) . In the boxplots , the median is indicated by a red line , the box represents the interquartile range , the whiskers represent the maximum and minimum values , and red dots represent outliers . ( p=0 . 01 ) . ( c , d ) Periods of gene activity measured in mock-treated and Wnt3a-treated cells . The population was divided into cases where the gene was either not transcribing before the addition of Wnt3a or mock-treatment ( ‘off’ , n ( Wnt3a ) = 27 , n ( Con ) = 22 , p=0 . 01 ) or if the gene was already active ( ‘on’ , n ( Wnt3a ) = 37 , n ( Con ) = 52 , p=0 . 77 ) . *p<0 . 05 , n . s . = p>0 . 05 . ( e ) Frames from Video 12 showing the activation of the CCND1-MS2 gene detected by MS2-GFP mRNA tagging ( arrow ) following Wnt3a treatment . Bar = 10 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 16748 . 02810 . 7554/eLife . 16748 . 029Figure 7—figure supplement 1 . Wnt signaling causes shorter rest duration in addition to an increase in the gene burst duration . Plots of single cells demonstrate the active ( blue ) and inactive ( red ) state of CCND1-MS2 transcribing gene along 6 hr in ( a ) mock-treated cells ( Control , n = 74 ) and ( b ) Wnt3a-treated cells ( n = 64 ) . Data were taken from live-cell movies with each column representing one cell along 6 hr . Histograms showing the distribution of ( c ) active and ( d ) inactive state durations in Wnt3a-treated and mock-treated ( Control ) cells ( p=0 . 02 , p=0 . 0004 respectively ) . The curves are a fit to exponential distribution ( Golding et al . , 2005 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 16748 . 029 We next examined whether the periods of gene activity were altered after Wnt activation . When CCND1 was at first non-active and began to transcribe after Wnt3a , there was prolonged transcriptional activation for a time-frame of 180 min , compared to a shorter activity period of 65 min in unactivated cells ( Figure 7c , e , Video 12 ) . This meant that Wnt signaling increased the time-frame of CCND1 promoter activity . Surprisingly , if the gene was detected in an already active state , and Wnt3a was then added , there was no difference in the activity period compared to that in unactivated cells . Under both conditions , activation persisted for an average of 130 min ( Figure 7d ) , meaning that if the promoter was already activated then there was no Wnt-induced change in this time-frame . 10 . 7554/eLife . 16748 . 030Video 12 . Prolonged activation of CCND1 after Wnt activation . HEK293 CCND1-MS2 cells stably expressing MS2-GFP ( green ) were treated with Wnt3a . CCND1 mRNA transcription could be detected 15 min after Wnt3a ( green dot , transcription site ) and continued for 4 hr . Cells were imaged every 15 min for 270 min . DOI: http://dx . doi . org/10 . 7554/eLife . 16748 . 030 When we examined the levels of CCND1-MS2 activity after Wnt activation in living cells ( Figure 8—figure supplement 1a–c ) , we found that even if the gene was active before Wnt3a addition , the intensity of MS2-GFP fluorescence on the gene showed higher levels , indicative of higher expression levels due to signaling , meaning that the promoter could integrate additional signals ( Figure 8—figure supplement 1b ) . We measured a 1 . 5–1 . 7 increase in the maximum MS2-GFP intensity levels , and observed that the maximum intensity distribution for Wnt3a-treated cells shifted such that many more cells displayed higher levels of gene activity ( Figure 8a , b , and Figure 8—figure supplement 1d ) . The time required to reach the maximum point of activity did not seem to change when examining the whole population ( Figure 8—figure supplement 1e ) . However , this time was actually shortened from 170 min to 120 min in cells where the gene was initially inactive , and the distribution of cells shifted to shorter times to reach maximum levels of transcription ( Figure 8c ) . This time did not change in cells where the gene was initially active ( Figure 8d ) . When gene activity and gene inactivity patterns were further examined , not only was an expected increase in the duration of gene activity found , but also a reduction in the rest duration . This means that Wnt activation not only increases the duration time for gene activity , but also reduces periods of inactivity by increasing the frequency of promoter firing events ( Figure 7—figure supplement 1 ) . 10 . 7554/eLife . 16748 . 031Figure 8 . Quantification of CCND1 activity levels following Wnt activation in single fixed and living cells . ( a , b ) Boxplots showing the maximal MS2-GFP intensity levels reached on actively transcribing CCND1-MS2 genes during 6 hr in Wnt3a-treated and mock-treated ( Con ) cells , when ( a ) the gene was either not transcribing before the addition of Wnt3a or mock-treatment ( ‘off’ , n ( Wnt3a ) = 27 , n ( Con ) = 22 , p=0 . 0001 ) or ( b ) if the gene was already active ( ‘on’ , n ( Wnt3a ) = 37 , n ( Con ) = 52 , p=0 . 0006 ) . The median is indicated by a red line , the box represents the interquartile range , the whiskers represent the maximum and minimum values , and red dots represent outliers . ( c , d ) Boxplots showing the time required to reach the maximal intensity levels when ( c ) the gene was either not transcribing before the addition of Wnt3a ( ‘off’ , p=0 . 03 ) or ( d ) if the gene was already active ( ‘on’ , p=0 . 42 ) . ( e ) YFP-β-catenin ( yellow ) together with RNA FISH images obtained with a probe hybridizing to the MS2 region in the 3’UTR of the CCND1-MS2 mRNA ( cyan ) , showing CCND1 nascent mRNAs on active genes ( large dots ) and cellular mRNAs ( small dots ) in Wnt3a-treated cells ( 2 hr ) , in comparison to YFP-β-catenin levels . Nuclei are stained with Hoechst ( pseudo-colored red ) . Bottom row is the pseudo-colored YFP signal using the ImageJ ‘Royal’ look-up table . Cells are numbered . Bar = 10 μm . ( f ) Quantification of the number of cellular CCND1-MS2 mRNAs ( ordered from low to high ) compared to YFP-β-catenin levels . ( g ) Quantification of the number of nascent CCND1-MS2 mRNAs compared to YFP-β-catenin levels . ( h , i ) Correlation analysis between ( h ) the number of cellular CCND1-MS2 mRNAs and YFP-β-catenin levels and ( i ) between the number of nascent CCND1-MS2 mRNAs and YFP-β-catenin levels . Blue dots – subpopulation with low nuclear YFP-β-catenin levels and low numbers of cellular/nascent CCND1-MS2 mRNAs . Red dots – subpopulation with high nuclear YFP-β-catenin levels and high numbers of cellular/nascent CCND1-MS2 mRNAs . Total correlation score between the number of cellular/nascent CCND1-MS2 mRNAs and YFP-β-catenin levels is 0 . 88 and 0 . 59 , respectively . ( j ) The field from panel e demonstrating higher intensity of active CCND1-MS2 genes in cells with high nuclear YFP-β-catenin levels ( red arrows ) compared to cells with low nuclear YFP-β-catenin levels ( yellow arrows ) . Active genes are pseudo-colored using the ImageJ ‘Red Hot’ look-up table . The fluorescent signal of the active genes was enhanced using ImageJ 'Spot Enhancing Filter 2D' . This enhancement led to the reduced detectability of single mRNAs in this presentation of the image , in order to emphasize the difference in transcriptional activity between low and high levels of nuclear YFP-β-catenin . Bar = 10 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 16748 . 03110 . 7554/eLife . 16748 . 032Figure 8—figure supplement 1 . Transcription site intensity levels in living cells following Wnt3a activation . Plots showing the MS2-GFP average intensity levels measured on active CCND1-MS2 transcription sites during 6 hr in Wnt3a-treated and in mock-treated cells , when ( a ) the gene was either not transcribing before addition of Wnt3a or mock-treatment ( ‘off’ , n ( Wnt3a ) = 27 , n ( Con ) = 22 ) ( y axis is ‘relative intensity’ going from an ‘off’ state to an ‘on’ state' ) or ( b ) if the gene was already active to begin with ( ‘on’ , n ( Wnt3a ) = 37 , n ( Con ) = 52 ) ( y axis is the fold change compared to the beginning of the movie ) . Results were normalized to the intensity at time 0 . ( c ) Combined data from a and b . ( d ) Boxplot ( left ) showing the MS2-GFP maximal intensity levels reached on active CCND1-MS2 transcription sites during 6 hr in Wnt3a-treated and in mock-treated cells . In the boxplots , the median is indicated by a red line , the box represents the interquartile range , the whiskers represent the maximum and minimum values , and red dots represent outliers . The histograms ( right ) show the distribution of maximal intensity levels in these cells ( combined data from Figure 8a and Figure 8b [p=2 . 13e-06] ) . The histograms show normalized data such that the area of each bar is relative to the number of observations ( i . e . graph height is the probability density of the bar value , and graph area is equal to the probability of obtaining the bar value ) . The sum area of all bars is 1 . The data were fitted with a Gaussian curve . ( e ) Boxplot ( left ) showing the time required to reach the maximal intensity in Wnt3a-treated and in mock-treated cells . The histograms showing the distribution of this time in these cells ( combined data from Figure 8c and Figure 8d p=0 . 69 ) . ***p<0 . 001 , ns = p>0 . 05 . DOI: http://dx . doi . org/10 . 7554/eLife . 16748 . 032 These measurements suggested that Wnt3a signaling increases promoter firing events so that more CCND1 mRNAs are transcribed . To further examine this on the single mRNA level , we performed quantitative RNA FISH on CCND1-MS2 mRNA molecules in parallel to measuring β-catenin nuclear levels within the same single cell ( fixed cells ) . We counted the number of cellular and nascent CCND1-MS2 mRNAs in Wnt3a-treated cells ( Figure 8e ) and compared this value to the accumulation levels of nuclear β-catenin in the different cells . Cells that had accumulated β-catenin had significantly higher numbers of cellular CCND1-MS2 mRNAs ( 3-fold; Figure 8e , f ) and nascent CCND1-MS2 mRNAs ( 3 . 8-fold; Figure 8e , g , j ) , which correlated well with the transcription measurements in living cells ( Figure 8—figure supplement 1c ) . Correlating between cellular and nascent CCND1-MS2 mRNA numbers and β-catenin levels showed two sub-populations of high- and low-expressing CCND1-MS2 cells , in correlation with nuclear β-catenin accumulation , respectively ( Figure 8h , i ) . Regarding gene activation , altogether we find that Wnt signaling leads to increased promoter firing frequency , increased gene activity duration time , reduced gene rest time , and significantly higher numbers of mRNAs in the cell .
Signaling factors that translocate into the nucleus following signal transduction do so via different modes of shuttling . For instance , some factors display continuous nucleo-cytoplasmic oscillations ( p53 , mdm2 , NF-κB , ERK ) ( Purvis and Lahav , 2013; Kalo and Shav-Tal , 2013; Lahav et al . , 2004; Shankaran et al . , 2009 ) , while some show a rapid and limited pulse of nuclear build-up ( NFAT ) ( Yissachar et al . , 2013 ) , or a prolonged presence in the nucleus ( MAL ) ( Cui et al . , 2015 ) . These dynamics have been characterized using microscopy studies performed in single cells . Biochemical examination of these dynamics can give a true sense of the time-scales of the accumulation as seen by studying protein levels in Western blots ( Hernández et al . , 2012; Lustig et al . , 2002 ) . However , such approaches cannot provide an accurate temporal dynamic profile of the response as it unfolds within the cell , since they represent an average picture of the behavior of the whole cell population from which the proteins were extracted ( Levsky and Singer , 2003 ) . By characterizing β-catenin accumulation dynamics in several subcellular compartments within individual living cells , we could examine how varying responses in individual cells translate into a well-timed response of the cell population . Using a cell system we previously generated to follow CCND1 transcription in real-time on the single gene level ( Yunger et al . , 2010 ) , we now measured β-catenin sub-cellular dynamics , as well as characterized the transcriptional response of CCND1 to Wnt . Even though Wnt/β-catenin signaling has been highly studied , the basic propagation dynamics of this signal in single living cells have not been characterized . This is due to the lack of an appropriate system that would allow analysis of the behavior of a fluorescent β-catenin protein that resembles the endogenous protein ( Tan et al . , 2014 ) . Previous studies using transiently overexpressed β-catenin and photobleaching methods were important in establishing the characteristics of its intra-cellular mobility ( Krieghoff et al . , 2006; Jamieson et al . , 2011 ) . However , the subcellular distribution of transiently overexpressed fluorescent β-catenin is different than the endogenous protein , since the overexpressed protein is found throughout the whole cell including the nucleus ( even without a signal ) , membrane staining is lacking , and cytoskeletal organization is disrupted ( Krieghoff et al . , 2006; Jamieson et al . , 2011; Ligon et al . , 2001 ) . Even the use of nanobodies targeting endogenous β-catenin in living cells did not mimic the membrane localization of non-induced cells ( Traenkle et al . , 2015 ) . Hence , using our cell system in which YFP-β-catenin was stably expressed at relatively low levels ( 80% over the endogenous protein ) and was distributed similarly to the endogenous protein , we were able to follow the subcellular dynamics of β-catenin in real-time . Upon Wnt activation , β-catenin levels in the cell population portrayed a relatively rapid increase in the cells . The general time-scale of hours of β-catenin accumulation concurred with Western blotting experiments ( Hernández et al . , 2012; Lustig et al . , 2002 ) , and altogether portrayed an orchestrated response of the cell population to the Wnt signal . However , examination of the accumulation profiles in single cells showed response patterns deviating from the average behavior in at least 20% of the population; accumulation rates and levels varied , and in some cells additional but less intense waves of β-catenin nuclear accumulation were observed . We suggest that the balance between accumulation and degradation affects the outcome in β-catenin build-up in each cell . The Kirschner group has shown ( Hernández et al . , 2012 ) that Wnt does not completely abolish the activity of the destruction complex . We therefore suggest that if the total levels of accumulation are similar in most cells ( e . g . integral analysis ) , then the level of inhibition of the destruction complex is expected to vary in each cell and to determine the response . However , the fact that CCND1 transcriptional activation occurs within the same time frame as the main initial phase of β-catenin nuclear accumulation means that in most cells in the population , the CCND1 gene will become activated shortly after Wnt activation . Possibly , later phases of β-catenin nuclear accumulation could have an influence on prolonging CCND1 activity ( steady state activity levels return after 6 hr ) . Indeed , measurements of CCND1 activity in living cells following Wnt activation showed a positive change in several parameters relating to gene activation; not only did the frequency of CCND1 activation in the cell population rise and the time to activate CCND1 shorten , but the levels of CCND1 transcriptional output increased , the timeframe of gene activity became substantially longer , and gene resting periods were shortened . Overall , this means that Wnt signaling increases the number of CCND1 mRNAs generated , by increasing the frequency of RNA polymerase II recruitment to the promoter and by lengthening the time of promoter responsiveness . Interestingly , even when an already active CCND1 gene received the Wnt signal , the levels of gene activity increased . Although we were unable to examine YFP-β-catenin dynamics and CCND1-MS2 transcription activity simultaneously in living cells , by integrating the measurements of CCND1 transcriptional activity with the measured dynamics of β-catenin nuclear accumulation from the separate experiments , we found that the rate of change of nuclear β-catenin correlated well with transcription induction ( Figure 9a ) , specifically during the first rapid phase of nuclear β-catenin accumulation . This fits in well with a previous study demonstrating that the fold change in β-catenin nuclear levels is the element affecting target gene activity ( Goentoro and Kirschner , 2009 ) , and that the transcriptional machinery is capable of computing the fold change in β-catenin , thereby determining the transcriptional response ( Goentoro and Kirschner , 2009 ) . Similar behavior was observed for the ERK signaling pathway ( Cohen-Saidon et al . , 2009 ) . Hence , it is not the absolute number of β-catenin molecules in the nucleus that correlates with transcription rates , but the rate of change of β-catenin levels over time , and particularly the rapid change during the first phase of induction that elicits the transcriptional effect ( Figure 9b ) . The advantage of such a sensing mechanism would be its ability to buffer out cellular noise and variability in the cell population . 10 . 7554/eLife . 16748 . 033Figure 9 . Comparing the kinetics of CCND1 transcriptional activation to the dynamics of β-catenin nuclear accumulation rate of change following Wnt signaling in living cells . ( a ) Plots of the average transcriptional activation kinetics of CCND1-MS2 ( red ) following Wnt3a activation , compared to the plot of rate of change in β-catenin nuclear accumulation ( green ) . ( b ) Scheme of the dynamic changes occurring in the studied cell system following Wnt signaling . Top - from left to right: Levels of β-catenin ( yellow ) in the nucleus are normally low but after the addition of Wnt3a to the medium a significant and rapid increase in the nucleus is observed , peaking after 2–3 hr . β-catenin levels later decline in the nucleus and cytoplasm due to degradation . While this is the average behavior in the population ( e . g . cells 1 and 2 ) , when examining individual cells , different dynamics such as multiple pulsations ( e . g . cell 3 ) and rapid initial accumulation ( e . g . cell 4 ) are observed . β-catenin levels increase simultaneously at the membrane and at the centrosome . Bottom- β-catenin induces cyclin D1 transcriptional activity ( green dot ) , and modulation of the transcriptional reaction can be observed as the gene reaches higher levels of activity , for longer periods of time . The rate of change in β-catenin accumulation ( blue curve , top ) , rather than the actual levels of β-catenin in the nucleus , correlate with the kinetics of transcriptional activation . DOI: http://dx . doi . org/10 . 7554/eLife . 16748 . 033 Concurrent β-catenin accumulation the cell membrane and the centrosome were quantified . β-catenin demarcates the cell outline when there are cell-cell contacts due to its presence in adherens junctions ( Harris and Tepass , 2010 ) . Generally , while the nuclear accumulation of β-catenin has been the focus of Wnt signaling studies , the membrane region has not been considered a major target of the response . However , one study has shown localization of unphosphorylated β-catenin to the membrane upon Wnt activation within 30 min , in cells lacking E-cadherin , although the function was unclear ( Hendriksen et al . , 2008 ) . We found increased β-catenin levels in the membrane following Wnt activation . The punctate membranal pattern persisted during activation , suggesting that Wnt increases the recruitment of β-catenin to existing adherens junctions . Indeed , β-catenin dynamics in the membrane showed a relatively slow exchange both before and after Wnt activation , similar to a study conducted in LiCl induced cells ( Johnson et al . , 2009 ) . This implies long residence times of β-catenin in the membrane and that potential binding sites for β-catenin molecules at adherens junctions exist constantly , and only when the protein becomes abundant , do they fill up . Centrosomal localization of β-catenin is known ( Kaplan et al . , 2004; Hadjihannas et al . , 2010; Bahmanyar et al . , 2008 , 2010; Mbom et al . , 2014; Huang et al . , 2007; Vora and Phillips , 2015 ) . The exact function is not clear and it probably plays a role in regulation of cell separation . It has been suggested that Wnt signaling abolishes the phosphorylation of β-catenin and leads to centrosome splitting ( Hadjihannas et al . , 2010 ) . Our study shows for the first time , the highly rapid accumulation rates of β-catenin at the centrosome in real-time , following Wnt signaling . β-catenin at the centrosome is highly mobile as seen in our FRAP study and in another ( Bahmanyar et al . , 2008 ) . Interestingly , we identified a connection between the membranal and centrosomal β-catenin fractions . Puncta of membranal β-catenin were detected moving relatively slowly from the membrane region and ending up at the centrosome , sometimes several in parallel in the same cell . Since unphosphorylated β-catenin is found in the membrane after Wnt ( Hendriksen et al . , 2008 ) , we can postulate that the β-catenin fraction moving to the centrosome is unphosporylated , and may be involved in driving cell division . Notably , our study also provides a temporal view of β-catenin dynamics in single cells under conditions of LiCl activation . Although LiCl is considered a chemical that mimics Wnt activation and increases β-catenin levels in the nucleus , it is obvious that the dynamics , build-up rates and levels of β-catenin in all subcellular compartments were dramatically exaggerated and unregulated in comparison to Wnt activation . This should be taken into account when inferring information regarding Wnt signaling and β-catenin from LiCl treatment . The Wnt pathway has been implicated in cell cycle regulation , and levels of phosporylated β-catenin oscillate and increase towards mitosis ( Davidson and Niehrs , 2010; Hadjihannas et al . , 2012 ) . Examining cells that had undergone mitosis after Wnt activation , did not show a pattern of β-catenin levels in daughter cells , nor did Fucci labeling uncover a cell cycle pattern of β-catenin accumulation following Wnt . This suggests that Wnt-induced nuclear accumulation is not cell cycle dependent . The propagation of a signal from a membrane receptor to the gene promoter can follow different types of kinetics . Single-cell analysis revealed significant variability in the dynamics of β-catenin nuclear buildup , but also that most cells did finally accumulate the same total level of β-catenin over time . This behavior is quite different than the serum activation pathway that activates β-actin via MAL shuttling ( Kalo et al . , 2015 ) . β-actin transcriptional activation begins less than 5 min after serum addition , and β-actin alleles respond in the same manner and same time-frame; i . e . variability of the response in single cells is low . Hence , some signaling cascades must relay the information rapidly and tightly since this will lead to the translation of a highly required protein , e . g . β-actin , to generate a protein that is required for cell motility in response to environmental sensing ( Kislauskis et al . , 1994; Latham et al . , 1994 ) . Other pathways such as Wnt/β-catenin may also signal to activate gene expression , but their response emerges much later , probably since the required biological outcomes , such as cell proliferation , require more regulation points . The changes in β-catenin levels in response to Wnt , in several subcellular compartments , indicate that the signaling pathway does not only activate gene expression but is involved in additional processes . Further studies should reveal the exact roles of these subpopulations of β-catenin in response to signal transduction .
HEK293 Flp-in CCND1-MS2 cells ( Yunger et al . , 2010 ) were maintained in Dulbecco’s modified Eagle’s medium ( DMEM , Biological Industries , Israel ) containing 10% FBS ( HyClone Laboratories , Logan , UT ) and hygromycin selection ( 100 µg/ml; Sigma , Israel ) . Stable expression of MS2-GFP was obtained by co-transfection of the cells with MS2-GFP ( 10 µg ) and puromycin resistance ( 300 ng ) plasmids using calcium phosphate transfection , and selection with puromycin ( 1 μg/ml; Invivogen , San Diego , CA ) and hygromycin ( 100 µg/ml ) . Stable expression of YFP-β-catenin ( Krieghoff et al . , 2006 ) ( 10 µg ) was performed by calcium phosphate transfection , and selection with neomycin ( 500 μg/ml; Sigma ) and hygromycin ( 100 µg/ml ) . Cells with very low expression levels were collected by FACS ( FACSAria III , BD Biosciences ) . Transient expression of YFP-β-catenin was performed using PolyJET ( SignaGen , Israel ) . For generating Wnt3a conditioned medium ( CM ) and mock CM , L-Wnt-3A and L- mouse fibroblast cells were grown in DMEM and 10% FBS , and CM was prepared according to the American Tissue Culture Collection ( ATCC ) instructions ( Shibamoto et al . , 1998 ) . Wnt activation was performed with either Wnt3a-CM or with recombinant human Wnt3a ( 200 ng/ml; R& D Systems , Minneapolis , MN ) . Wnt3a-CM or mock-CM were added 1:1 to the volume of the cells medium . Cells were also treated with LiCl ( 20 mM; Sigma ) and MG132 ( 20 µM; Sigma ) . The Fucci system ( Clontech , Mountain View , CA ) was used for cell cycle phase detection . For G1 phase detection , the pRetroX-G1-Red vector ( mCherry-hCdt1 ) was used , and for S/G2/M phase the pRetroX-SG2M-Cyan vector ( AmCyan-hGeminin ) . The Fucci system , being a viral-based system first required the introduction of the mouse ecotropic retroviral receptor on the membrane surface of HEK293 CCND1-MS2 cells expressing YFP-β-catenin . Transient transfection was performed 24 hr prior to infection using PolyJet transfection with the pBABE ecotropic receptor plasmid ( Addgene #10687 , Cambridge , MA ) . This step was performed twice for each infection . After mCherry-hCdt1 infection , mCherry positive cells were collected by FACS and maintained in medium containing puromycin ( 1 μg/ml; Invivogen ) . Cells were then transfected with the pBABE ecotropic receptor plasmid and 24 hr post-transfection , the cells infected with AmCyan-hGeminin . Positive cells were collected by FACS and maintained in medium containing neomycin ( 500 μg/ml ) and puromycin ( 1 μg/ml ) . For infections , HEK293T cells were maintained in DMEM containing 10% FBS and used to package the Fucci retroviruses , which were collected over a period of three days before infecting the ecotropic HEK293 cells . SDS-PAGE and Western blotting were performed as previously described ( Aizer et al . , 2008 ) . Primary antibodies used were mouse anti-β-catenin ( BD Transduction Laboratories , cat# 610154 , San Jose , CA ) and rabbit anti-tubulin ( Abcam , Cambridge , MA ) . The secondary antibody was an HRP-conjugated goat anti-rabbit or anti-mouse IgG ( Sigma ) . Immunoreactive bands were detected by the Enhanced Chemiluminescence kit ( ECL , PierceThermo scientific , Waltham , MA ) . Experiments were performed three times . HEK293 CCND1-MS2 cells were co-transfected with the cyclin D1 promoter −1745CD1LUC Firefly luciefarse construct ( Albanese et al . , 1995 ) and either YFP-β-catenin or eYFP-C1 ( mock ) , together with a Renilla luciferase construct using PolyJet transfection . 50 ng of each plasmid were used . A luciferase assay was performed after 24 hr using the Dual-Glo Luciferase assay system ( Promega , Madison , WI ) . After standardization with Renilla luciferase activity , a relative luciferase activity was obtained and the mean and standard deviation from triplicate wells was calculated . Each experiment was performed three times . YFP-β-catenin ( Krieghoff et al . , 2006 ) was obtained from Jürgen Behrens ( University of Erlangen-Nürnberg ) . Cells were harvested and DNA quantification was performed using 5 μg/ml DAPI solution ( Sigma ) . The BD FACSAria III cell sorter was used . For quantifying DNA in fixed cells , we used a 405 nm laser for excitation and a 450/40 nm bandpass filter for detection . Data were processed and analyzed using FlowJo software . The average quantification of 3 repeated experiments is presented ( mean ± sd ) . Cells were grown on coverslips coated by Cell-Tak ( BD Biosciences ) , washed with PBS and fixed for 20 min in 4% PFA . Cells were then permeabilized in 0 . 5% Triton X-100 for 3 min . After blocking , cells were immunostained for 1 hr with a primary antibody , and after subsequent washes the cells were incubated for 1 hr with secondary fluorescent antibodies . Primary antibodies: mouse anti-β-catenin and rabbit anti-pericentrin ( Abcam , cat# ab4448 ) . Secondary antibodies: Alexa488-labeled goat anti-mouse IgG and Alexa594-labeled goat anti-rabbit ( Invitrogen , Carlsbad , CA ) . Nuclei were counterstained with Hoechst 33342 ( Sigma ) and coverslips were mounted in mounting medium . CCND1-MS2 cells were grown on coverslips coated by Cell-Tak ( BD Biosciences ) and fixed for 20 min in 4% paraformaldehyde , and overnight with 70% ethanol at 4°C . The next day cells were washed with 1x PBS and treated for 2 . 5 min with 0 . 5% Triton X-100 . Cells were washed with 1x PBS and incubated for 10 min in 40% formamide ( 4% SSC; Sigma ) . Cells were hybridized overnight at 37°C in 40% formamide with a specific fluorescently-labeled Cy3 DNA probe ( ~10 ng probe , 50 mer ) . The next day , cells were washed twice with 40% formamide for 15 min and then washed for two hours with 1X PBS . Nuclei were counterstained with Hoechst 33342 and coverslips were mounted in mounting medium . The probe for the MS2 binding site was: CTAGGCAATTAGGTACCTTAGGATCTAATGAACCCGGGAATACTGCAGAC . 3D stacks ( 0 . 2 µm steps , 76 or 51 planes ) of the total volume of the cells were collected from fixed CCND1-MS2 cells . The 3D stacks were deconvolved and the specific signals of mRNAs were identified ( Imaris , Bitplane ) . mRNA identification was performed in comparison to deconvolved stacks from cells not containing the MS2 integration , which therefore served as background levels of nonspecific fluorescence . No mRNAs were identified in control cells . The sum of intensity for each mRNA particle and active alleles was measured in the same cells using Imaris , as previously described ( Yunger et al . , 2010 , 2013 ) . The single mRNA intensities were pooled and the frequent value was calculated . The sum of intensity at the transcription site was divided by the frequent value of a single mRNA . This ratio provided the number of mRNAs associated with the transcription unit from the point of the MS2-region and onwards . As mRNAs should be associated with a polymerase , this number should reflect the maximum number of polymerases engaged with this region . Quantification and counting experiments were applied to experiments performed on different days . Wide-field fluorescence images were obtained using the Cell^R system based on an Olympus IX81 fully motorized inverted microscope ( 60X PlanApo objective , 1 . 42 NA ) fitted with an Orca-AG CCD camera ( Hamamatsu ) driven by the Cell^R software . Live-cell imaging was carried out using the Cell^R system with rapid wavelength switching . For time-lapse imaging , cells were plated on glass-bottomed tissue culture plates ( MatTek , Ashland , MA ) coated by Cell-Tak ( BD Biosciences ) in medium containing 10% FBS at 37°C . The microscope is equipped with an incubator that includes temperature and CO2 control ( Life Imaging Services , Reinach , Switzerland ) . For long-term imaging , several cell positions were chosen and recorded by a motorized stage ( Scan IM , Märzhäuser , Wetzlar-Steindorf , Germany ) . In these experiments , HEK293 Flp-in CCND1-MS2 expressing MS2-GFP cells were imaged in 3D ( 26 planes per time point ) every 15 min , at 0 . 26 µm steps for 6 hr . HEK293 Flp-in CCND1-MS2 cells expressing YFP-β-catenin were imaged in 3D ( 15 planes per time point ) at 0 . 7 µm steps , every 15 min , up to 18 hr . For presentation of the movies , the 4D image sequences were transformed into a time sequence using the maximum or sum projection options or manually selecting the in-focus plane using the ImageJ software . Time-lapse data were collected from single cells in several fields and on several days until reaching an appropriate sample size , and then all single-cell data were pooled and either averaged and presented as plots , or presented as single cell data . The intensity of the active transcription sites labeled with MS2-GFP fluorescence in time-lapse movies was corrected for photobleaching using ImageJ , and the 3D movies were transformed to 2D by choosing the in-focus plane in which the intensity of the transcription site is the highest . Movies were manually tracked and the intensity measured for each frame ( Is ) . Background from another location in the nucleus ( In ) was subtracted for each frame , and the final intensity was calculated using: I = Is ( t ) − In ( t ) and then normalized to the initial intensity . Measuring the intensity of the YFP-β-catenin signal in the subcellular compartments was performed manually using ImageJ , and background was subtracted from all measurements . When YFP-β-catenin levels were low , DIC images that were acquired in parallel were used for nucleus detection . For measurements of centrosome intensity , the intensity of the centrosome in each frame ( Ic ) was multiplied by the area occupied by the centrosome ( Ac ) : I = Ic ( t ) *Ac ( t ) . For membrane intensity , a sum projection of the 3D movies was used . Intensity was normalized either to the initial frame or to the highest intensity measured . The values of the nucleus/cytoplasm ( N/C ) ratio of YFP-β-catenin were obtained by division of the YFP-β-catenin intensity levels measured . Correlation coefficient values were calculated by comparing the intensity of β-catenin over time between all possible pairs of sub-cellular compartments , from Wnt activation onset . Values of rate of change ( ΔI/Δt ) in YFP-β-catenin in the sub-cellular compartments over time were obtained by measuring the intensity difference ( ΔI ) between two consecutive time points divided by the time difference ( Δt ) between the two time points:ΔIΔt ( t ) n+12=I ( t ) n+1−I ( t ) ntn+1−tn FRAP and FLIP experiments were performed using a 3D-FRAP system ( Photometrics ) built on an Olympus IX81 microscope ( 636 Plan-Apo , 1 . 4 NA ) equipped with an EM-CCD ( Quant-EM , Roper ) , 491 nm laser , Lambda DG-4 light source ( Sutter ) , XY and Z stages ( Prior ) , and driven by MetaMorph ( Molecular Devices ) . Experiments were performed at 37°C with 5% CO2 using a live-cell chamber system ( Tokai ) . For each acquisition , YFP-β-catenin was bleached using the 491 nm laser . Six pre-bleach images were acquired . In FRAP , post-bleach images were acquired every 0 . 8 s for 80 s in the cytoplasm and the nucleus , every 1 s for 2 min in adherens junctions , every 0 . 4 s for 40 s at the centrosome , and every 1 . 5 s for 8 min to measure nuclear import and export rates . In FLIP , images were acquired every 1 . 9 s for 280 s in the cytoplasm and the nucleus . The experiments were analyzed using ImageJ macros previously described ( Aizer et al . , 2008 ) . Data from at least 10 experiments for each cell line were collected and the averaged FRAP and FLIP measurements were fitted by Matlab with a double exponential model: I ( t ) =α1*exp ( −τ1*t ) +α2*exp ( −τ2*t ) +c Where t = 0 is the time immediately after photobleaching . t0 . 5 was defined as time where I ( t=t0 . 5 ) =I ( t=∞ ) 2 . We used a simple model for describing β-catenin concentration ( C ) dynamics in the nucleus based on the data presented in the plot from Figure 2c:dCdt=P ( t ) −α ( t ) C Where α is the time dependent degradation rate , and P is the time dependent production rate . Both rates are allowed to change when t=T:α ( t ) ={α1;fort≤Tα2;fort>TP ( t ) ={P1;fort≤TP2;fort>T The solution is:C ( t ) ={[C ( 0 ) −P1α1]∗e−α1∗t+P1α1; for t≤T [C ( T ) −P2α2]∗e−α2∗ ( t−T ) +P2α2; for t>T Where:C ( T ) =[ C ( 0 ) −P2α1 ]*e−α1*t+P2α1 We fit the model by minimizing the sum of the squares of the residuals with the function ‘fmincon’ in MATLAB using the ‘active-set’ algorithm . Two tailed t-test was performed in the following experiments: Quantitative FISH , Luciferase assay , the N/C ratio of YFP-β-catenin and live cell analysis . A Mann–Whitney test was performed in FRAP and FLIP experiments ( Supplementary file 1 ) . | Cells in an animal’s body must communicate with one another to coordinate many processes that are essential to life . One way that cells do this is by releasing molecules that bind to receptors located on the surface of others cells; this binding then triggers a signaling pathway in the receiving cell that passes information from the surface of the cell to its interior . The last stage of these pathways typically involves specific genes being activated , which changes the cell’s overall activity . Wnt is one protein that animal cells release to control how nearby cells grow and divide . One arm of the Wnt signaling pathway involves a protein called β-catenin . In the absence of a Wnt signal , there is little β-catenin in the cell . When Wnt binds to its receptor , β-catenin accumulates and enters the cell’s nucleus to activate its target genes . One of these genes , called cyclin D1 , controls cell division . However it was not understood how β-catenin builds up in response to a Wnt signal and influences the activity of genes . Using microscopy , Kafri et al . have now examined how the activities of β-catenin and the cyclin D1 gene change in living human cells . These analyses were initially performed in a population of cells , and confirmed that β-catenin rapidly accumulates after a Wnt signal and that the cyclin D1 gene becomes activated . Individual cells in a population can respond differently to signaling events . To assess whether human cells differ in their responses to Wnt , Kafri et al . examined the dynamics of β-catenin in single cells in real time . In most cells , β-catenin accumulated after Wnt activation . However , the time taken to accumulate β-catenin , and this protein’s levels , varied between individual cells . Most cells showed the “average” response , with one major wave of accumulation that peaked about two hours after the Wnt signal . Notably , in some cells , β-catenin accumulated in the cell’s nucleus in two waves; in other words , the levels in this compartment of the cell increased , dropped slightly and then increased again . So how does β-catenin in the nucleus activate target genes ? Kafri et al . saw that the absolute number of β-catenin molecules in the nucleus did not affect the activity of cyclin D1 . Instead , cells appeared to sense how quickly the amount of β-catenin in the nucleus changes over time , and this rate influences the activation of cyclin D1 . Importantly , problems with Wnt signaling have been linked to diseases in humans; and different components of the Wnt signaling pathway are mutated in many cancers . An important next challenge will be to uncover how the dynamics of this pathway change during disease . Furthermore , a better understanding of Wnt signaling may in future help efforts to develop new drugs that can target the altered pathway in cancer cells . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"chromosomes",
"and",
"gene",
"expression",
"cell",
"biology"
] | 2016 | Quantifying β-catenin subcellular dynamics and cyclin D1 mRNA transcription during Wnt signaling in single living cells |
Acute Myeloid Leukemia ( AML ) is an aggressive hematological malignancy with abnormal progenitor self-renewal and defective white blood cell differentiation . Its pathogenesis comprises subversion of transcriptional regulation , through mutation and by hijacking normal chromatin regulation . Kat2a is a histone acetyltransferase central to promoter activity , that we recently associated with stability of pluripotency networks , and identified as a genetic vulnerability in AML . Through combined chromatin profiling and single-cell transcriptomics of a conditional knockout mouse , we demonstrate that Kat2a contributes to leukemia propagation through preservation of leukemia stem-like cells . Kat2a loss impacts transcription factor binding and reduces transcriptional burst frequency in a subset of gene promoters , generating enhanced variability of transcript levels . Destabilization of target programs shifts leukemia cell fate out of self-renewal into differentiation . We propose that control of transcriptional variability is central to leukemia stem-like cell propagation , and establish a paradigm exploitable in different tumors and distinct stages of cancer evolution .
Acute Myeloid Leukemia ( AML ) is the most prevalent leukemia in adults with a dismal prognosis of less than 30% 5 year survival ( Döhner et al . , 2017 ) . It is a heterogeneous disease , clinically and pathologically , with common cellular themes of myeloid differentiation block , and recurrent molecular targeting of chromatin and transcriptional regulators . Effects on chromatin and transcription are reflected in the AML mutational spectrum ( Ley et al . , 2013 ) , as well as through the implication of general chromatin regulators in AML pathogenesis , in the absence of specific mutation events ( Roe and Vakoc , 2013 ) . Specific examples of AML dependence on unmutated chromatin regulators include BRD4 ( Dawson et al . , 2011; Zuber et al . , 2011 ) , LSD1 ( Harris et al . , 2012 ) or DOT1L ( Bernt et al . , 2011; Daigle et al . , 2011 ) , their importance highlighted by the fact that chemical inhibitors developed to target these regulators have progressed to clinical trials ( Gallipoli et al . , 2015 ) . More recently , TAF12 , a component of the basal transcription factor complex TFIID , was shown to be critical for AML cell maintenance through regulation of protein stability and activity of the transcription factor MYB ( Xu et al . , 2018 ) . In a recent CRISPR drop-out screen of genetic dependencies in AML , we have identified several members of the promoter-bound histone acetyl-transferase SAGA complex , including acetyl-transferase KAT2A , as being required for AML maintenance ( Tzelepis et al . , 2016 ) . KAT2A was suggested to impact cell survival and differentiation status , but its precise molecular mechanisms of action remain to be elucidated , and it is unclear whether it is required in AML initiation , as well as maintenance . Kat2a is a mammalian orthologue of yeast histone acetyl-transferase Gcn5 , and is required for H3K9 acetylation ( H3K9ac ) ( Jin et al . , 2014 ) , a modification that fine-tunes , rather than initiates , locus-specific transcriptional activity . Kat2a is required for specification of mesodermal derivatives during early embryonic development ( Lin et al . , 2007; Wang et al . , 2018 ) , and for survival of neural stem and progenitor cells ( Martínez-Cerdeño et al . , 2012 ) . Loss of Kat2a in the hematopoietic system from an early developmental stage did not grossly impact blood formation in vivo , but could promote terminal granulocyte differentiation in vitro , through relief of protein acetylation-dependent inactivation of transcription factor Cebpa ( Bararia et al . , 2016 ) . Nevertheless , detailed testing of Kat2a contribution to hematopoietic stem and progenitor cell function is still lacking . Yeast Gcn5 is a classical regulator of transcriptional noise ( Raser and O'Shea , 2004 ) , with deletion mutants enhancing cell-to-cell variability in gene expression measured across a range of locus fluorescence reporters ( Weinberger et al . , 2012 ) . Transcriptional noise reflects the variability in the number of mRNA molecules produced from a given locus through time; snapshot studies of gene expression capture the same phenomenon as cell-to-cell transcriptional heterogeneity ( Sanchez et al . , 2013 ) . Transcriptional noise can result from the bursting nature of gene expression ( Chubb and Liverpool , 2010 ) . Most if not all loci , undergo bursts of transcriptional activity with characteristic frequency and size: burst frequency corresponds to the rate at which promoters become engaged in active transcription; burst size measures the number of mRNA molecules produced during each transcriptional burst ( Cai et al . , 2006 ) . Both parameters contribute to mean gene expression , whereas transcriptional noise is more strictly dependent and shown to be anti-correlated with burst frequency ( Hornung et al . , 2012 ) . In yeast , size and frequency of bursts are increased through histone acetylation of gene bodies and promoters , respectively ( Weinberger et al . , 2012 ) . In functional terms , transcriptional noise has been directly implicated as a mechanism of cell fate choice in yeast ( Blake et al . , 2006 ) and bacteria ( Süel et al . , 2006 ) , and recurrently associated , albeit correlatively , with cell fate transitions in some mammalian systems ( Moris et al . , 2016 ) . We had previously shown that normal transitions into hematopoietic lineage specification associate with cell-to-cell transcriptional heterogeneity ( Pina et al . , 2012; Teles et al . , 2013 ) . More recently , we have inhibited the activity of Kat2a in mouse embryonic stem cells , and observed an increase in transcriptional heterogeneity that impacted the stability of pluripotency with rewiring of correlation gene regulatory networks ( GRNs ) ( Moris et al . , 2018 ) . Whilst we have not mechanistically linked enhanced heterogeneity with loss of pluripotency , we did observe propagation of variability of transcriptional levels through the GRNs , downstream of nodes with differential H3K9ac . Cancer , and in particular leukemia , can be perceived as a pathogenic imbalance of cell fate choices , with maintained self-renewal and reduced exit into differentiation . We postulated that enhancing transcriptional variability in AML cells , as may be achieved through Kat2a depletion , would enhance the probability of differentiation cell fate transitions . By using a retroviral-delivered MLL-AF9 model of AML in a conditional Kat2a knockout background , we show that loss of Kat2a depletes AML stem-like cells , imposing a mild delay to disease initiation and severely impairing AML propagation . At a molecular level , these changes are accompanied by specific loss of H3K9 acetylation at a subset of promoters , with reduced transcription factor binding and frequency of bursting , and associated variability in transcriptional levels . Affected loci encode for general mitochondria and nucleic acid metabolism , including translation , suggesting that they may contribute to cancer fate decisions more generally , rather than in a disease-specific manner .
In order to investigate the impact of Kat2a loss on self-renewal vs . differentiation of AML stem-like cells ( AML-SC ) , we generated an inducible conditional Kat2aFl/Fl KO mouse model and transformed Kat2a excised ( KO ) and non-excised ( WT ) bone marrow ( BM ) cells through retroviral delivery of an MLL-AF9 fusion transcript . This strategy allows for cellular and molecular investigation of Kat2a requirements during transformation , whilst minimizing putative confounding effects of acquired mutations on heterogeneity of transcription , such as might be observed in established human or mouse AML . The choice of a strong oncogenic event such as MLL-AF9 ( Ley et al . , 2013; Krivtsov et al . , 2006; Somervaille and Cleary , 2006 ) minimizes the need for cooperating genetic alterations . Similarly , the option for an inducible conditional knockout may reduce compensatory effects against Kat2a loss . Specifically , we crossed Kat2aFl/Fl C57Bl/6 mice ( Lin et al . , 2008 ) on the interferon response-inducible Mx1-Cre+/- background , and generated a stable mouse line homozygous for the Flox allele . Mx1-Cre-positive ( KO ) and Mx1-Cre-negative ( WT ) mice were compared across all experiments ( Figure 1—figure supplement 1A ) , with locus excision obtained through treatment of experimental and control mice with intra-peritoneal polyinosylic-polycytidylic ( pIpC ) acid ( Chan et al . , 2011 ) . Excision was tested 4–6 weeks after injection and consistently achieved values greater than 80% in stem and progenitor cell compartments ( Figure 1—figure supplement 1B ) , reflected in a profound loss of gene expression , including amongst myeloid-biased ( LMPP ) and committed ( GMP ) progenitors critical for AML initiation ( Goardon et al . , 2011; Figure 1—figure supplement 1C ) . Locus excision generates an in-frame product joining the first two and the last exons and is transcribed ( Figure 1—figure supplement 1D ) but does not code for acetyl-transferase activity or other functional domains ( Figure 1—figure supplement 1A ) . Kat2a was dispensable for normal mouse hematopoiesis , with minimal transient effects on the number ( Figure 1—figure supplement 1E ) and in vitro activity ( Figure 1—figure supplement 1F ) of KO hematopoietic stem cells ( HSC ) . These effects were not sustained through aging ( Figure 1—figure supplement 1G ) , and , crucially , did not affect HSC function in vivo , measured by long-term reconstitution of irradiated recipients through transplantation ( Figure 1—figure supplement 1H ) . Having verified that Kat2a deletion does not perturb normal hematopoiesis and thus preserves candidate progenitor cells-of-origin for leukemia transformation , we used a retroviral delivery system to express the MLL-AF9 leukemia fusion in progenitor-enriched , lineage-depleted ( Lin- ) WT and KO BM cells . Cells were transformed in vitro through serial re-plating in semi-solid medium-based colony-forming assays , with similar efficiency for both genotypes ( Figure 1A ) . Importantly , the level of locus excision was mildly increased during transformation ( Figure 1B ) , indicating that loss of Kat2a does not impede the initial selection of leukemia-transformed clones . Continued re-plating revealed that Kat2a KO affected the type of colonies produced , with a shift from compact or type I ( Johnson et al . , 2003 ) colonies ( Figure 1C–D ) with Kit+ progenitor-like cells ( Figure 1E and Figure 1—figure supplement 2A–B ) , to mixed , or type II ( Johnson et al . , 2003 ) colonies ( Figure 1F ) , with a characteristic halo ( Figure 1G and Figure 1—figure supplement 2C ) of more differentiated cells and a corresponding reduction in Kit-positive cells , which have higher levels of the differentiation marker Mac1/CD11b ( Figure 1H and Figure 1—figure supplement 2D ) . Compatible with the serial re-plating experiments , cell lines established from MLL-AF9-transformed cell lines of both Kat2a WT and KO genotypes exhibited a relative gain in mixed colonies ( Figure 1—figure supplement 2E ) , with higher levels of Mac1 ( Figure 1—figure supplement 2F ) . Taken together , the data suggest that Kat2a loss is permissive to MLL-AF9-driven transformation , but alters the balance between in vitro self-renewal and differentiation , favoring the latter . We tested these observations in vivo by monitoring leukemia development in mice that received WT and KO Lin- BM cells transduced for 2 days with retrovirus encoding the MLL-AF9 oncogenic fusion . Kat2a KO recipient animals had a moderate advantage in survival ( Figure 2A ) , suggesting a protracted development of leukemia . Leukemias were nevertheless depleted of Kat2a expression ( Figure 2B ) , thus excluding selection of escapee cells . WT and KO leukemic animals had similar disease burdens at the point of culling , as measured by organ infiltration ( Figure 2—figure supplement 1A ) and peripheral blood counts ( Figure 2—figure supplement 1B ) . In contrast with in vitro transformation , we did not observe gross changes to the balance between progenitor and differentiated cells in the resulting WT vs KO leukemias , which had preservation of phenotypic leukemia stem-like cells ( LSC ) , defined by the Lin-Kit+Sca1-CD34+CD16/32+ GMP surface phenotype ( L-GMP ) ( Figure 2C ) . We reckoned that loss of Kat2a affected the probability of leukemia development through dysregulation of transformed cells , and sought to probe this hypothesis by investigating the transcriptional programs of WT and KO leukemias at the single-cell level . To this end , we pooled samples from 4 to 5 primary leukemias of each genotype , sorted live GFP+ cells reporting the presence of the MLL-AF9 fusion , and successfully sequenced over 4000 cells Kat2a WT and KO cells each , for a total of 13166 transcripts , using a high-throughput 10X Genomics platform ( Zheng et al . , 2017 ) . Basic measures of gene alignment and quality control are summarized in Supplementary file 1 . In order to minimize the confounding effect of rarely or very low expression genes , we filtered out transcripts detected in less than 10% of all cells , and only included cells where a minimum of 500 different transcripts were detected . These filtering criteria match those recently used in analysis of heterogeneity of hematological malignancies using a similar 10X Genomics platform ( Pastore et al . , 2019 ) and were deliberately chosen for reproducibility . We thus focused the analysis on what we designated Robust gene set . These were a total of 2588 genes ( Supplementary file 2 ) , sampled from 7360 cells ( 3835 WT , 3525 KO ) , which captured general metabolic , biosynthetic and regulatory ontologies , as well as hematopoietic-specific categories , for example ‘myeloid cell differentiation’ , indicating broad representation of transcriptional programs within the dataset ( Supplementary file 3 ) . We assessed gene expression using the D3E algorithm ( Delmans and Hemberg , 2016 ) , which makes implicit use of the single-cell nature of the data to extract information about differential dynamics of transcription , in addition to determining changes in average gene expression . The algorithm views single-cell measurements as multiple observations of the same or closely-related cells and fits a two-state promoter model that interprets gene expression measurements in terms of frequency and size of bursts of promoter activity ( Figure 3—figure supplement 1A ) . We used a multiple linear regression model to verify the contribution of burst size and frequency to average gene expression and gene expression CV in our single-cell RNA-seq data ( Figure 3—figure supplement 1B ) . In line with other studies ( Hornung et al . , 2012 ) , we observed that both bursting parameters contributed to mean expression to similar extents , but there was a greater contribution of burst frequency to CV . Differential gene expression analysis between Kat2a WT and KO primary leukemic cells revealed mild albeit significant changes in average gene expression ( Figure 3A; median difference −0 . 03 ) , which were of down-regulation upon Kat2a excision , as might be anticipated from loss of a histone acetyl-transferase . Compatible with the proposed role of Kat2a in regulating transcriptional noise , we observed a significant increase in gene expression variability as measured by coefficient of variation ( CV = standard deviation/mean ) ( Figure 3B; median difference 0 . 24 ) in Kat2a KO cells which was apparent at all levels of gene expression ( Figure 3C ) . The increase in gene expression CV associates with greater cell-to-cell dispersion ( Mohammed et al . , 2017 ) , in other words reduced cell-to-cell correlation in transcript levels , amongst Kat2a KO leukemia cells ( Figure 3D ) . The increased dispersion of transcriptional variability can be attributable to a change in burst frequency ( Figure 3E; median difference −1 . 15 ) , but not burst size ( Figure 3F ) . This has also been observed for the Kat2a yeast orthologue Gcn5 in modulating transcriptional noise ( Weinberger et al . , 2012 ) . There have been several reports ( Mohammed et al . , 2017; Antolović et al . , 2017; Blake et al . , 2006; Chambers et al . , 2007; Chang et al . , 2008; Reynolds et al . , 2012 ) , including our own ( Pina et al . , 2012 ) , that suggest an association between variability in gene expression and probability of cell fate change , although most of the data remains correlative , at least in mammalian systems . Our recent analysis of Kat2a inhibition in mouse ES cells ( Moris et al . , 2018 ) is compatible with this view , suggesting that the observed enhanced transcriptional variability may promote exit from pluripotency through disruption of gene regulatory networks . In this light , we asked if its enhancement in Kat2a KO leukemias resulted in an imbalance of self-renewal vs differentiation states that could lead to the observed delay in leukemia progression ( Figure 2A ) . We made use of the RACE-ID algorithm ( Grün et al . , 2016 ) to cluster the combined 7360 WT and KO cells filtered as displaying Robust gene expression , and optimally separated them into 12 clusters on the basis of the 500 most highly variable genes in each genotype ( Figure 4A ) . Despite the occupancy of a broadly similar transcriptional space , the genotypes had differential cluster-association patterns , with some clusters , namely 7 , 11 and 12 , which together comprise 22 . 9% of WT cells , being relatively underrepresented amongst Kat2a KO cells ( respectively 0 . 5 , 0 . 2 and 0 . 5 of Kat2a WT ) ( Figure 4—figure supplement 1A ) . Other clusters were over-represented amongst Kat2a KO cells but to a lesser extent , with only 2 over 1 . 5-fold ( cluster 4–1 . 7-fold , 3 . 3% of WT cells; cluster 6–1 . 5-fold , 8 . 5% of WT cells ) , causing us to focus on the Kat2a KO-depleted clusters . Unsupervised alignment of the 12 clusters along a putative differentiation trajectory , shows that the underrepresented clusters lie at its undifferentiated end ( Figure 4B ) . This is enriched for gene expression signatures associated with MLL self-renewal ( Figure 4—figure supplement 1B ) , suggesting a depletion of cells with stem-like characteristics . Overall , gene expression programs at the start of the trajectory were similar between clusters 2 , 4 and 7 , with few genes unique to individual clusters thus preventing exploration of gene ontologies . There were nevertheless subtle differences in the functional categories enriched amongst differentially expressed genes between genotypes in each of the clusters , namely an association of categories associated with apoptosis in clusters 2 and 4 , which were absent from cluster 7 ( Supplementary file 4 ) . The STEM-ID algorithm builds on RACE-ID to define a ‘stemness’ score that has successfully identified previously elusive stem cell populations in mouse pancreas ( Grün et al . , 2016 ) and cellular hierarchies in human liver ( Aizarani et al . , 2019 ) . It postulates that stem cells exhibit a multitude of incipient lineage-affiliated programs ( high information entropy ) , which are shared ( high connectivity ) with more differentiated cells , and attributes a ‘stemness’ score to each cluster as the product of entropy and number of links ( Figure 4—figure supplement 1C ) . Indeed , cluster seven scores as the most stem-like transcriptional state , suggesting that Kat2a KO leukemias may be depleted of functional , if not surface phenotype ( Figure 2C ) , LSC . We tested this by pooling equal numbers of cells from all primary WT or all primary KO leukemias in Figure 2A , and transplanting them into secondary recipients at limiting cell doses . Despite the equivalent number of phenotypic LSC in the primary leukemias of both genotypes ( Figure 2C ) , recipients of Kat2a KO leukemia cells failed to develop leukemia at the lowest cell doses and had a dramatically reduced frequency of functional LSC ( Figure 4C–D ) , indicating that a requirement for Kat2a in leukemia self-renewal and/or propagation . Significantly , the level of Kat2a gene expression knockout was profound and retained in secondary leukemias , similar to the primary leukemias they originated from ( Figure 4—figure supplement 1D ) . In agreement , the frequency of Kat2a KO cells escaping the excision was negligible amongst phenotypically undifferentiated Lin-Kit+Sca1-CD16/32+ cells ( Figure 4—figure supplement 1E ) . This confirmation of a maintained and profound gene expression knockout with minimum contribution from undeleted cells is particularly important , as the persistence of a rearranged transcript fusing exons 1 , 2 and 18 ( Figure 1—figure supplement 1D ) prevents knockout quantification in the 3’-biased 10X Genomics platform . In contrast , the Taqman assay and the nested primers in the PCR analyses ( Figure 4—figure supplement 1D–E ) are directed at exons 6–7 and 11–13 , respectively , which capture the excised region . Thus , cellular and molecular analyses , including inspection of transcriptional variability , are likely minimally affected by the presence of cells escaping Kat2a gene deletion , and should truly reflect the effects of Kat2a loss . Secondary leukemias , like primary leukemias , did not have a reduction in surface phenotypic LSC ( L-GMP , Figure 4E ) , putatively highlighting a dissociation between primitive surface phenotype and function , or highlighting the existence of multiple stem-like states , as suggested by cluster-specific depletion at the undifferentiated end of the Kat2a KO leukemia trajectory . Nevertheless , secondary leukemias displayed an increased proportion of early ( Mac1+Gr1- ) , but not late ( Mac1+Gr1+ ) differentiated myelo-monocytic cells ( Figure 4E ) , suggesting that loss of Kat2a may perturb self-renewal and differentiation progression at multiple levels . Accordingly , we observed reduction of transcriptional burst frequency in KO cells to be prevalent across clusters , albeit slightly more marked at the undifferentiated end ( Figure 4—figure supplement 1F ) . The exception are the central clusters 11 and 12 , which have minimal representation in Kat2a KO leukemia ( Figure 4—figure supplement 1A ) . Also , separate inspection of the differentiation trajectories for Kat2a WT ( Figure 4—figure supplement 2A–B ) and Kat2a KO cells ( Figure 4—figure supplement 2C–D ) suggests that the almost linear developmental relationship between WT leukemia cells is replaced by multiple branching decisions upon Kat2a loss , an observation also captured by a second pseudo-time trajectory algorithm ( Figure 4—figure supplement 2E–F ) . In summary , loss of Kat2a depletes functional LSC and alters cellular hierarchies within MLL-AF9 leukemia . This is likely achieved through perturbation of promoter activity and we sought to define the genes and pathways directly regulated by Kat2a . We defined Kat2a regulatory targets by chromatin immunoprecipitation followed by next generation-sequencing ( ChIP-seq ) . Kat2a is a histone acetyl-transferase required for deposition of the activating H3K9 acetylation ( ac ) mark ( Jin et al . , 2014 ) and capable of catalyzing multiple acetyl-modifications ( Kuo and Andrews , 2013 ) at promoters and at enhancers ( Krebs et al . , 2011 ) . We identified promoters as H3K4 tri-methyl ( me3 ) peaks ( Figure 5—figure supplement 1A ) and enhancers with H3K4 mono-methyl ( me1 ) - enriched regions ( Figure 5—figure supplement 1B ) and inspected the respective pattern of distribution of H3K9ac in pooled MLL-AF9 primary leukemias initiated by Kat2a KO or WT cells . Experiments were performed in duplicate , with good overlap between replicates ( Figure 5—figure supplement 1C–D ) . We included analysis of another acetylation activation mark , H3K27ac , which associates with active enhancers ( Creyghton et al . , 2010 ) , but has not been specifically linked to Kat2a activity . Inspection of peak overlaps between methylation and acetylation marks indicated a specific depletion of H3K9ac binding at promoters in regions devoid of concomitant H3K27ac activation mark ( Figure 5A ) . Conversely , H3K9ac was mildly increased at candidate active enhancer regions marked by the presence of H3K27ac ( Figure 5—figure supplement 1E ) suggesting a possible pattern of imbalance of H3K9ac regulation between promoters and enhancers . Guided by these preliminary observations , we focused on those promoter peaks with unique loss of H3K9ac upon Kat2a depletion , and used the ENCODE database ( Auerbach et al . , 2013 ) to confirm enriched experimental binding of Kat2a ( aka GCN5 ) in other model systems ( Figure 5B and Supplementary file 5 ) . Genes associated with differentially-acetylated promoter peaks represented general nucleic acid and mitochondrial metabolic categories ( Supplementary file 6 ) . Somewhat surprisingly , the list of target promoters failed to include most MLL-AF9 ChIP targets ( Bernt et al . , 2011 ) with the leukemia self-renewal associated Hox gene signature , namely Hoxa9 , Hoxa10 and Meis1 , escaping Kat2a-dependent promoter acetylation control . This suggests that Kat2a-mediated maintenance of LSC may be achieved through general , rather than leukemia-specific , regulatory programs , with putative implications for other leukemic and non-leukemic malignancies . Having previously observed that loss of Kat2a associates with a decrease in frequency of transcriptional bursting ( Figure 3E ) , we asked if this was more evident for genes dependent on Kat2a for promoter H3K9ac . Global inspection of the Robust gene set across all cells showed a similar reduction in burst frequency for Kat2a acetylation targets and non-targets ( Figure 5C ) . However , H3K9ac target genes had uniquely preserved mean gene expression levels ( 95% CI KO-WT means = −0 . 1118 to 0 . 01303 , Welch t-test p=0 . 12 for targets vs . −0 . 0894 to −0 . 0016 , p=0 . 04 for non-targets ) , with increased CV upon Kat2a KO ( both groups p<0 . 0001 ) , suggesting a specific impact of Kat2a on variability of transcription . Nevertheless , global analysis of transcriptional burst frequency failed to produce a specific association with Kat2a acetylation targets , which could be due to a confounding effect of mixing cells at different points in the differentiation trajectory , for which locus-specific regulation might differ , or indeed a cascading of transcriptional consequences along gene regulatory networks , as suggested by our analysis of mouse ES cells where Kat2a activity was inhibited ( Moris et al . , 2018 ) . Alternatively , the lack of association could reflect the fact that our subset of acetylation targets captures absolute , but not relative losses of promoter H3K9ac upon Kat2a KO , as a result of the necessarily limited replication of ChIP analysis of primary leukemia cells . To avoid potential confounding effects of cellular heterogeneity on determination of transcriptional parameters , we repeated the analysis exclusively on stem-like cluster 7 cells . Indeed , by using a more homogeneous group of cells , we could observe that the reduction in frequency of bursting in Kat2a KO cells was significantly more marked for H3K9ac target genes ( Figure 5D; Mann-Whitney test for difference in burst frequency , Fburst , defined as ( FburstKO-FBurstWT ) /FburstWT p=0 . 0155 ) , indeed suggesting that Kat2a acts by regulating bursting activity at target promoters . Transcriptional variability has been associated with density of transcription factor binding . As Kat2a is known to interact with transcription factors , including the oncogene Myc ( Wang et al . , 2018; Hirsch et al . , 2015 ) , to regulate common downstream gene expression programs , we asked if loss of Kat2a reduced transcription factor binding at target H3K9ac peaks . Analysis of H3K9ac targets using the ENCODE ChIP-Seq significance tool suggested binding by Myc in other experimental systems ( Supplementary file 3 ) , in addition to general transcription and pause-release factors . It also showed increased binding by the respiratory chain regulator Gabpa ( Ongwijitwat and Wong-Riley , 2005 ) ( Nrf2 in the ENCODE database ) , which has been described as a regulator of normal hematopoietic and Chronic Myeloid Leukemia ( CML ) stem cells ( Manukjan et al . , 2016; Yang et al . , 2013 ) . We performed ChIP-qPCR analysis of Myc and Gabpa binding at promoter peaks depleted of H3K9ac binding in Kat2a KO leukemias . We selected candidate Myc and Gabpa target peaks based on experimental DNA occupancy by the transcription factors ( TF ) across different mouse cell types in the ENCODE database . Q-PCR primers were designed under the respective H3K9ac peak , for analysis of pooled mouse MLL-AF9 secondary leukemia samples of each genotype . We confirmed that the transcription factors ( TF ) Myc and Gabpa did indeed bind at most of the locations analyzed in MLL-AF9 leukemias ( Figure 5E–F and Figure 5—figure supplement 2A–B ) . Critically , we observed that binding of both TF at promoter regions dependent on Kat2a for H3K9ac was globally reduced in Kat2a KO leukemias ( Figure 5E–F ) as compared to WT ( Myc: 95% CI WT-KO enrichment 0 . 01273 to 1 . 589; 2-way ANOVA p<0 . 05 for genotype contribution; Gabpa: 95% CI WT-KO enrichment 0 . 2640 to 1 . 364; 2-way ANOVA p<0 . 01 for genotype contribution ) , thus suggesting that Kat2a may regulate binding of sequence-specific TF at the promoter regions it acetylates . Having established that Kat2a can control the H3K9ac status , and putatively TF binding , of a subset of promoters , with impact on their bursting activity , we focused on the nature of the genes showing reduced frequency of bursting in Kat2a KO leukemias . We found that these were preferentially associated with translation categories ( Supplementary file 7 ) , including ribosomal proteins and translation initiation and elongation factors , some of which we showed to be depleted of TF binding upon Kat2a loss . Additionally , we could observe an enrichment of translation-associated gene expression signatures amongst undifferentiated cells relatively depleted in Kat2a KO leukemias ( Figure 6A ) , suggesting that they may contribute to leukemia propagation and/or maintenance . We started by verifying that perturbed regulation by Kat2a had functional consequences for the translation machinery by performing polysomal profiling of MLL-AF9-carrying MOLM-13 AML cells in which KAT2A activity was inhibited by the MB-3 inhibitor ( Tzelepis et al . , 2016 ) . Reassuringly , we observed a dramatic reduction in polysomal content ( Figure 6B ) , indicating the functional impact of Kat2a transcriptional control , despite minimum changes in average gene expression . We then tested the effect on emergent protein synthesis in mouse MLL-AF9 leukemias by quantifying OP-Puro incorporation in Kat2a WT and KO phenotypic L-GMP . We saw qualitative and quantitative reductions in OP-Puro incorporation in Kat2a KO cells , with a bimodal distribution of low and high incorporating cells that was unique to Kat2a-depleted cells ( Figure 6C ) , and significantly lower levels protein synthesis within the high OP-Puro distribution ( Figure 6D ) . Altogether , the data suggest that phenotypic leukemia stem-like cells are less translationally active in the absence of Kat2a , which could explain their reduced functionality . We assessed functional impact in vitro by inhibiting the translation machinery of Kat2a WT and KO MLL-AF9 transformed cells with the S6K1 inhibitor PF4708671 ( Pearce et al . , 2010 ) . PF4708671 targets the TOR pathway and impedes translation initiation and elongation . Treatment of MLL-AF9 transformed primary mouse BM cultured cells with PF4708671 in colony-forming assays recaptured the imbalance between compact and mixed colonies ( Figure 6E ) observed upon in vitro transformation of Kat2a KO cells . Taken together , the data indicate that regulation of the translational machinery at least partially mediates the imbalance between self-renewal and differentiation observed in MLL-AF9-driven leukemia upon Kat2a loss . It further supports the notion that fine control of transcriptional activity leading to changes in variability of transcript levels influences cell fate transitions and can be exploited by malignant cells for cancer maintenance .
In this study , we combine functional assays with single-cell transcriptional analysis and identify a requirement for the histone acetyl-transferase Kat2a in sustaining leukemia stem-like cells ( LSC ) and their downstream leukemia cellular structure in MLL-AF9-initiated AML . Loss of Kat2a increases cell-to-cell variability in transcription at all levels of mean gene expression , which is reflected in poor coordination in gene expression programs in Kat2a KO leukemia cells . Perturbation of transcription associates with an increased trend towards differentiation in vitro and loss of long-term functional LSC in vivo . However , the differentiation routes followed by Kat2a KO leukemic cells are aberrant and seem to follow multiple alternative dead ends that deviate from the seemingly linear pathway of Kat2a WT MLL-AF9 cells . These characteristics are reminiscent of the perturbation of pluripotency we observed in Kat2a-inhibited mouse embryonic stem ( mES ) cells ( Moris et al . , 2018 ) , where cells accumulate at the exit from pluripotency and only progress to differentiation with slow kinetics . Indeed , that may be the expected outcome of constitutively enhancing variability in gene expression , or transcriptional noise , which may need to be buffered downstream of a differentiation transition for differentiation to proceed ( Ahrends et al . , 2014 ) . Although it is surprising that we do not observe changes to normal hematopoietic stem and progenitor cells , the same was reported by others using a tissue-specific developmental ( Vav-Cre ) knockout ( Bararia et al . , 2016 ) . Stem cell and developmental systems are typically robust to noise perturbations ( Urban and Johnston , 2018 ) , and may have compensatory systems that buffer the loss of Kat2a . A putative compensatory mechanism would be the up-regulation of the Kat2a orthologue Pcaf/Kat2b , which is normally absent in HSC and progenitors . Whilst we did not observe up-regulation of Kat2b in leukemia cells , we did not specifically look for it in normal hematopoiesis , and cannot exclude that it may explain the difference between healthy and leukemic blood cells . We did note a very early reduction in HSC and their multilineage colony-forming potential upon Kat2a loss , but this was not sustained and did not affect engraftment potential , a more accurate measure of HSC activity . Also , the Kat2a KO hematopoietic system did not exhibit increased sensitivity to prolonged treatment with the cytotoxic drug 5-fluorouracil ( data not shown ) , suggesting that KO and WT HSC are similarly functional . Although we cannot exclude that serial or competitive transplantation may reveal a defect in Kat2a KO HSC , our data are nevertheless supportive of a difference in the sensitivity of leukemia cells to Kat2a loss . It has been proposed that cancer cells exist in a state of enhanced transcriptional activity that is required to sustain their oncogenic self-renewal programs ( Lin et al . , 2012 ) . Amongst other factors , hyper-transcription has been associated with Myc , a known collaborator of Kat2a in transcriptional regulation ( Hirsch et al . , 2015 ) , which is broadly required in AML ( Delgado and León , 2010 ) , including in MLL fusion-driven disease ( Schreiner et al . , 2001 ) . Accordingly , we found Myc to be depleted in a subset of promoters targeted by Kat2a . However , transcription factor depletion at promoters was not exclusive to Myc , and we made similar and indeed more extensive observations with another broad metabolic regulator , Gabpa , which has been previously characterized as a regulator of CML self-renewal ( Yang et al . , 2013; Yu et al . , 2012 ) and associated with Kat2a-containing complexes ( Krebs et al . , 2011 ) , but for which a requirement in AML has not been established . Of interest , loss of both transcription factors analysed have severe consequences to normal hematopoiesis , which are of HSC depletion in the case of Gabpa ( Manukjan et al . , 2016; Yu et al . , 2011 ) and impaired differentiation with accumulation of defective phenotypic HSC in the case of Myc ( Wilson et al . , 2004 ) . Notably , none of these roles is phenocopied by Kat2a loss . We thus suggest that Kat2a acts through co-option of the transcriptional machinery present at target loci , rather than rely on a unique conserved transcription factor circuit , to exert its pleiotropic activating effects in leukemia cells . This said , we did not observe a monotonous reduction in transcription factor binding across all loci analysed , suggesting that the presence of Kat2a may facilitate transcription factor recruitment and/or binding in a probabilistic manner , a view compatible with a role for Kat2a in stabilizing rather than initiating transcription ( Jin et al . , 2014 ) . Live-cell imaging of transcription factor recruitment to individual loci ( Donovan et al . , 2019 ) and/or single-cell ChIP ( Ai et al . , 2019; Hainer et al . , 2019; Ku et al . , 2019 ) , currently undergoing significant development , will be central to definitively test this hypothesis . An alternative , albeit not mutually exclusive , explanation is that the role of Kat2a in transcription factor binding or recruitment is influenced by the chromatin context in which it acts , including the post-translational modifications it catalyses as well as the modifications introduced by other factors . Kat2a catalyzes acetylation of Lys nine in Histone 3 ( H3K9ac ) , a chromatin mark associated with maintenance , but not initiation , of locus transcription . Kat2a is able to catalyze acetylation of additional lysine residues in cell-free systems ( Kuo and Andrews , 2013 ) , but its loss in vivo more specifically affects H3K9ac , particularly so in the vicinity of transcriptional start sites ( Wang et al . , 2018 ) . Compatible with these observations , Kat2a loss in the context of MLL-AF9 leukemia impacted H3K9ac specifically at gene promoters . In addition , H3K9ac was specifically lost at gene loci that do not exhibit additional activating acetylation marks such as H3K27ac . Whilst the specific meaning of single vs . double-acetylated regions is unclear , one possibility is that the presence of H3K27ac marks genomic regions that also function as enhancers of more distant genes , suggesting that Kat2a may be strictly required at promoters . In agreement with this view , we did not observe loss of H3K9ac at H3K4me1-positive enhancers . Instead , we observed a gain in H3K27ac at both H3K4me3 promoters and H3K4me1 enhancers as a single acetylation mark , which may help explain the minimal consequences of Kat2a loss in terms of average gene expression , and highlight the specific role of Kat2a-dependent H3K9ac of promoters in stabilizing transcriptional activity . Gain of H3K27ac may reflect differentially regulated or compensatory acetylation by other histone acetyl-transferases in the absence of Kat2a . In light of the promoter vs . enhancer specificity of the changes observed , it will be interesting to investigate to what extent reprogramming of acetylation marks reflects proximal reconfiguration of enhancer-promoter interaction via Ctcf binding . Ctcf loss has been shown to increase gene expression CV with moderate or no differences in mean expression ( Ren et al . , 2017 ) , a pattern akin to our observations upon Kat2a loss . Whereas their study specifically investigated sequence-driven loss of Ctcf binding at proximal , intra-TAD enhancer regions , we observed that H3K9ac-depleted promoters in Kat2a KO leukemia cells had a significant association with experimental Ctcf binding in ENCODE experiments , and we speculate that Ctcf may be dislodged to enhancers and promote asymmetric distribution of histone acetylation marks , with dysregulation of locus control . Of note , too , is the fact that despite the almost complete knockout of Kat2a expression , the loss of H3K9ac , although specific in terms of chromatin context , is far from dramatic . This is similar to recent observations in embryoid bodies ( Wang et al . , 2018 ) and suggests that Kat2a requirement for H3K9ac is not absolute , although it may be locus-specific . Other acetyltransferases may , either normally or compensatorily , contribute to H3K9 acetylation in at least some locations , and it will be interesting to understand the parameters that determine specific dependency on Kat2a activity and its unique consequences to transcription . Variability in gene expression levels reflects regulation of locus activity , and whilst specific contribution from enhancers has been proposed ( Fukaya et al . , 2016 ) and remains an area of active investigation ( Larsson et al . , 2019 ) the role of promoter configuration and sequence has been more extensively characterized in multiple model systems ( Antolović et al . , 2017; Faure et al . , 2017; Zoller et al . , 2015 ) . In most if not all loci , transcriptional activity is discontinuous , with promoters cycling between active ( ON ) and inactive ( OFF ) states . Self-limited bursts of transcriptional activity are characterized by the burst frequency , reflecting the rate of OFF-to-ON transitions , and the burst size , which captures the number of mRNA molecules produced during each burst . In yeast , regulation of both burst parameters is dependent on H3K9ac at specific gene locations: gene body acetylation regulates burst size; promoter H3K9ac associates primarily with burst frequency ( Weinberger et al . , 2012 ) . Furthermore , in yeast , promoter H3K9ac is deposited by the Kat2a orthologue and founder histone acetyl-transferase Gcn5 , and removed by the Sin3a orthologue Rpd3 ( L ) deacetylase complex . Loss of Gcn5 decreases burst frequency across multiple yeast loci and has been modelled to increase intrinsic transcriptional noise , a finding we capture in mammalian cells in the present study . Whilst our study specifically links promoter H3K9ac to regulation of burst frequency in mammalian cells , recent work by the Naëf lab has shown that locus-specific manipulation of promoter , but not distal or enhancer , H3K27ac can also change transcriptional bursting frequency ( Nicolas et al . , 2018 ) . Although the Naëf study has not specifically manipulated H3K9ac levels , it did reveal an association between promoter H3K9ac and frequency of locus activation , which agrees with our own observations . Whether other residue-specific acetylations of promoters ( or indeed enhancers ) can produce the same effect remains to be determined , and this knowledge will undoubtedly deepen current understanding of transcriptional regulation . Moreover , Kat2a was recently shown to catalyze other acyl-modifications of lysine residues , namely succinylation ( Wang et al . , 2017 ) , which also associates with transcriptional activation ( Tong et al . , 2020 ) . Characterization of the exact mechanistic consequences of additional acylations and their interaction with the better characterized lysine acetylations is still lacking , not least due to lack of modification and residue-specific antibodies for the newly-identified marks . It is possible that their loss also contributes to the changes in transcriptional regulation seen upon Kat2a KO , and could for example explain why the reduction in frequency of bursting , although more strongly associated with loss of H3K9ac , is not exclusive to sites depleted of this modification . Additionally , it remains possible that loss of Kat2a may impact other residue-specific acylations more dramatically than its specific effect on H3K9ac and the effects of their combined loss more completely link all the effects observed . Our lab has recently developed a KAT2A-dCas9 fusion capable of catalyzing targeted acetylation events ( data not shown ) , which will be instrumental in the mechanistic understanding of individual acetylation events and specific sequences in regulating bursting activity . It should also allow us to probe other candidate acyl modifications to unveil their unique and combined effects on transcriptional bursting . Somewhat unexpectedly , we found that the genes regulated by Kat2a at the level of promoter acetylation , and which responded to Kat2a loss with decreased frequency of bursting , were specifically associated with ribosomal assembly and translation activity . Similar categories have been shown to be regulated by non-catalytic components of the Kat2a SAGA complex in controlling ESC pluripotency ( Seruggia et al . , 2019 ) , reinforcing the notion that Kat2a complexes impact general metabolic processes in multiple cell types , and that these general processes can specifically influence cell fate transitions . Importantly , we demonstrated that Kat2a-depleted leukemia stem-like cells ( phenotypic L-GMP ) have reduced protein synthesis activity , putatively due to a perturbation of polysomal assembly consequent to variability in levels of ribosomal proteins . Moreover , perturbation of the translational machinery could re-capture the enhanced in vitro differentiation of leukemia cells observed upon Kat2a depletion , suggesting that alterations in protein synthesis activity may indeed be central to exit from leukemia self-renewal . In agreement , Morrison and collaborators ( Signer et al . , 2014 ) had previously reported that impaired protein synthesis upon genetic depletion of the ribosomal protein machinery impedes leukemia self-renewal , whilst having non-linear dose-dependent effects on normal hematopoiesis , mimicking our own observations in the Kat2a KO setting . Future studies directing Kat2a catalytic activity to single or multiple loci will illuminate individual vs . global target gene contributions to the leukemia phenotype . However , it is tempting to speculate that the generic nature of the programs impacted by Kat2a at the level of transcriptional noise may configure an underlying propensity towards execution of cell fate transitions , which can be of a different nature in different biological contexts . Analysis of the impact of Kat2a target programs in other malignant and normal stem cell systems , or at different stages of leukemia progression will test this hypothesis . It will also be interesting to determine if other candidate regulators of transcriptional noise produce similar effects and can be exploited therapeutically in AML , as well as in other hematological and non-hematological malignancies .
Kat2aFl/Fl conditional knockout mice ( Lin et al . , 2008 ) ( MGI:3801321 ) were bred with Mx1-Cre +/- transgenic mice ( Kühn et al . , 1995 ) , in a C57Bl/6 background . Littermates were genotyped for Kat2a LoxP sites ( forward: CACAGAGCTTCTTGGAGACC; reverse: GGCTTGATTCCTGTACCTCC ) and for Mx1-Cre: ( forward: CGTACTGACGGTGGGAGAAT; reverse: TGCATGATCTCCGGTATTGA ) : Ear notch biopsies were digested using KAPA express extract ( Sigma Aldrich ) and KAPA2G ROBUST HS RM Master Mix ( 2x ) ( Sigma Aldrich ) . PCR cycling protocol: 95C , 3 min; 40x ( 95°C , 15 s; 60°C , 15 s; 72°C , 60 s ) ; 72°C , 60 s . DNA products were run on a 1% Agarose Gel in TAE ( 1x ) , at 100V and visualized using an AlphaImager UV transilluminator ( Protein Simple ) . Cre-mediated recombination was induced in 6–10 week-old mice by administration of 5 alternate-day intraperitoneal injections of poly ( I ) - poly ( C ) ( pIpC ) , 300 μg/dose . After pIpC treatment , animals were identified as Kat2a WT = Kat2 aFl/Fl * Mx1-Cre -/- and Kat2a KO = Kat2 aFl/Fl * Mx1-Cre +/- . Excision efficiency was determined by qPCR of genomic DNA ( gDNA ) from Peripheral Blood ( PB ) , Spleen ( Sp ) or Bone Marrow ( BM ) . gDNA was extracted using Blood and Tissue DNA easy Kit ( Qiagen ) and quantified by Nanodrop ( Thermo Scientific ) . qPCR analysis used Sybr Green Master Mix ( Applied Biosystems ) and two sets of primers ( Figure 1A ) : Kat2a-IN11 ( forward: CAACTTCCCCAAGGTATGGA; reverse: CGGGGACCTTAGACTTGTGA ) , within the excised region; Kat2a-OUT18 ( forward: AGTCTGGGCTGTTTCCATGT; reverse: GCCCGTTGTAGAATGTCTGG ) , distal to the second LoxP site . Expression levels were determined by the Pfaffl method following normalization to Kat2a-OUT . PB was collected by saphenous vein and differential blood cells counts were determined using a Vet abc automated counter ( Scil Animal Care , Viernheim , Germany ) . Mice were kept in an SPF animal facility , and all experimental work was carried out under UK Home Office regulations . Animal research was regulated under the Animals ( Scientific Procedures ) Act 1986 Amendment Regulations 2012 following ethical review by the University of Cambridge Animal Welfare and Ethical Review Body ( AWERB ) . BM was isolated from mouse long bones as described before ( Pina et al . , 2015 ) . Following red blood cell lysis , total BM suspension was depleted of differentiated cells using a cocktail of biotinylated lineage ( Lin ) antibodies ( Table 1 ) and streptavidin-labeled magnetic nanobeads ( Biolegend ) , according to manufacturers’ instructions . Cells were directly used in transplants , colony-forming assays or flow cytometry for analysis of normal hematopoiesis . For leukemia studies , cells were cultured overnight at 37°C 5% CO2 in RPMI supplemented with 20% Hi-FBS ( R20 ) , 2 mg/mL L-Glutamine , 1% PSA , 10 ng/mL of murine Interleukin 3 ( mIL3 ) , 10 ng/mL of murine Interleukin 6 ( mIL6 ) , and 20 ng/mL of murine Stem Cell Factor ( mSCF ) ( cytokines from Peprotech ) ( supplemented R20 ) , followed by retroviral transduction . For analysis of normal progenitors , sorted mouse BM cells were plated at a density of 200–400 cells/plate in duplicates , in MethoCult GF M3434 ( STEMCELL Technologies ) . Colonies were scored at 7–9 days . For analysis of MLL-AF9 leukemia , retroviral-transduced BM cells were plated in M3434 at an initial density of 10000 cells/condition and scored and re-plated every 6–7 days . Re-plating was performed up to passage 9 , with 4000 cells/condition used from plate 3 . CFC assays from mouse MLL-AF9 transformed lines were seeded in M3434 and scored 6–7 days later . RPS6K inhibition studies were set by adding 3 . 3 μL DMSO , either as vehicle or with a final concentration of 3 . 5 μM of PF4708671 ( Tocris ) , directly to the methylcellulose medium , with mixing prior to cell addition . For analysis of normal hematopoiesis , 106 Kat2a WT or Kat2a KO cKit+ cells were intravenously injected via tail vein into lethally irradiated ( 2*5 . 5Gy ) CD45 . 1 recipient mice . At the described time-points , BM and Sp were collected and processed into a single-cell suspension for surface marker staining and flow cytometry analysis . For AML induction , we transplanted 1 . 5 × 106 cKit+ Kat2a WT or Kat2a KO cells transduced with MSCV-MLL-AF9-IRES-YFP , intra-venous into lethally irradiated ( 2* 5 . 5Gy ) CD45 . 1 recipient mice . The number of recipients used was determined by the numbers of cells available post-retroviral transduction and the transduction efficiency estimated by flow cytometry at the point of injection , aiming at a minimum of 1 x 105 YFP+ cells/recipient and the same number of YFP+ cells delivered to all recipients . The investigators were blinded as to the group allocation , with the injections performed by an investigator not involved in sample preparation , or in subsequent animal follow-up and tissue collection . Upon signs of illness and following human end-point criteria , animals were culled , tissue samples collected for histology analysis , and BM and Sp processed into single-cell suspensions . Flow Cytometry analysis and DNA extraction were performed . Data collection was performed using general identification numbers with no reference to the experimental group . For limiting-dilution analysis , 5*102 - 5 × 104 cells from primary leukemia pooled BM of each genotype were transplanted into sub-lethally irradiated ( 1*5 . 5Gy ) CD45 . 1 recipient mice ( 3–4/dose and genotype ) . Numbers of animals used were contingent on availability on CD45 . 1 recipients aiming at no less than 10 recipients per genotype divided between 3 cell doses to allow for limiting dilution statistical analysis . Retroviral construct MSCV-MLL-AF9-IRES-YFP was previously described ( Fong et al . , 2015 ) . For viral particle production , Human Embryonic Kidney ( HEK ) 293 T cells were seeded at 2 . 5 × 106 cells/10 cm dish in DMEM supplemented with 10% Hi-FBS , 2 mg/mL L-Glutamine , 1% PSA and cultured overnight at 37°C 5% CO2 . The following day , a transfection mix [per plate: 47 . 5 µL of TranSIT ( Miros ) , 5 µg of packaging plasmid psi Eco vector ( 5 µg ) , retroviral vector ( 5 µg ) and 600 µL of Optimem Medium ( Gibco ) ] was prepared according to manufacturer’s instructions and added dropwise to cells followed by plate swirling and overnight culture at 37°C 5% CO2 . Medium was replaced with R20 the next day . At 24 and 48 hr after R20 replacement , medium was collected and filtered through a 0 . 45 μM syringe filter , and viral particle suspension medium was added to BM cells . BM cells from 6 to 10 week old Kat2a WT and Kat2a KO mice were collected and Lineage-depleted as described above ( Isolation of mouse BM stem and progenitor cells ) , and cultured overnight at 37°C 5% CO2 in supplemented R20 . For viral transduction , BM cells were briefly centrifuged at 400G , 5 min , and viral particle suspension medium supplemented with 10 ng/mL mIL3 , 10 ng/mL mIL6 , and 20 ng/mL mSCF added to a final density of 106 cells/mL . Cells were plated in 6-multiwell plates and centrifuged for 1 hr at 2000 rpm , 32°C . After , cells were incubated for 4 hr at 37°C 5% CO2 . A second round of viral transduction was performed , with post-centrifugation incubation performed overnight . Next day , cells were collected , pelleted and washed three times with PBS ( 2x ) and R20 ( 1x ) . YFP level was accessed by Flow Cytometry in a Gallios Analyser ( Beckman Coulter ) . MLL-AF9 clonal liquid cultures were set up using MLL-AF9 retrovirus-transduced primary BM cells ( see Retroviral Transduction section ) . Transformed cells enriched in vitro by 3 rounds of serial plating ( CFC assays ) were maintained in R20 supplemented on alternate days with mSCF , mIL3 and mIL6 , all at 20 ng/mL . Cells were cultured at 2*105 cells/ml and passaged when they reached a density of 1*106/ml . Cell surface analysis of BM and Sp was performed using a panel of antibodies marked with * described in Table 1 , as per the following sorting strategies: HSC - Lin– cKit+ Sca1+ CD34– Flt3-; MPP: Lin– cKit+ Sca1+ CD34+ Flt3-; LMPP: Lin– cKit+ Sca1+ CD34+ Flt3+; CMP: Lin– cKit+ Sca1- CD34+/low CD16/32low; GMP: Lin– cKit+ Sca1+ CD34+ CD16/32high; MEP: Lin– cKit+ Sca1+ CD34- CD16/32-; Lin-: CD3e- B220- Gr1- CD11b- Ter119- . Data were acquired on Gallios ( Beckman Coulter ) or LSRFortessa ( BD ) cytometers; data analysis used Kaluza software ( Beckman Coulter ) . Cell sorting was performed on Influx or AriaII instruments ( both from BD ) . Six million cells each from the spleens of 3 individual secondary leukemia samples per genotype were cultured overnight in supplemented R20 to allow recovery after thaw . O-propargyl-puromycin ( OP-Puro , Thermo Fisher Scientific ) was added directly to 80% of each culture at a final concentration of 50 μM and incubated for 1 hr at the end of the culture period; the remainder was treated with PBS and processed in parallel as a control . After incubation , cells were washed twice in ice-cold PBS without Ca2+ or Mg2+ ( Sigma ) , and resuspended in PBS/10%FBS for cell surface staining with Lineage markers ( Gr1/Mac1-AF700 ) , c-Kit-APC-ef780 , Sca1-PE-Cy7 , CD34-APC , CD16/32-PerCP-Cy5 . 5 ( see Table 1 ) for 30 min on ice . After washing , cells were fixed in 1% paraformaldehyde ( PFA ) in PBS for 15 min on ice , protected from light , washed , and then permeabilized in PBS/3% FBS/0 . 1% saponin ( permeabilization buffer ) at room temperature , in the dark , for 5 min . Cells were washed and used immediately in the azide-alkyne cyclo-addition reaction with Click-iT Cell Reaction Buffer Kit ( Thermo Fisher Scientific; C10269 ) and Alexa Fluor 555-Azide ( Thermo Fisher Scientific; A20012 ) with a master reaction solution freshly prepared as per manufacturer’s instructions for immediate use . Alexa Fluor 555-Azide was used at a final concentration of 5 μM . The reaction proceeded in the dark at room temperature for 30 min; cells were washed twice in permeabilization buffer and then counterstained with DAPI 3 . 3 μg/ml in PBS for 5 min prior to flow cytometry analysis . We did not observe any effect of cell cycle status on differential OP-Puro labeling . MOLM-13 cells ( ID: CVCL_2119 ) were grown to an approximate density of 1 × 106 cells/mL , treated with cycloheximide ( 100 μg/mL ) for 15 min , washed in ice-cold PBS and stored at −80°C . Cells were lysed in buffer A ( 20 mM HEPES pH 7 . 5 , 50 mM KCl , 10 mM ( CH3COO ) 2 Mg , EDTA-free protease inhibitors ( Roche ) , supplemented with cycloheximide 100 μg/mL , 1 mM PMSF , 100 U/mL RNase inhibitor ( Promega ) , 1% ( vol/vol ) sodium deoxycholate , and 0 . 4% ( vol/vol ) NP-40 ) at 108 cells/mL for 10 min on ice . Lysates were cleared by centrifugation ( 8000 g for 5 min at 4°C ) and 3 A254nm units loaded onto a 10–50% ( wt/vol ) sucrose gradient in buffer A in Polyallomer 14 × 95 mm centrifuge tubes ( Beckman ) . After centrifugation ( Beckman SW40Ti rotor ) at 260 900 g for 3 hr at 4°C , gradients were fractionated at 4°C using a Gilson Minipulse three peristaltic pump with continuous monitoring ( A254nm ) . Samples were analysed using a Brandel gradient fractionator , the polysome profiles were detected using a UV monitor ( UV-1 , Pharmacia ) at A254 , and 0 . 5 mL fractions were collected . The electronic outputs of the UV-1 monitor and fraction collector were fed into a Labjack U3-LV data acquisition device with an LJTick-InAmp preamplifier . Total RNA was extracted using Trizol Reagent ( Invitrogen ) . RNA from equal numbers of cells was reverse-transcribed using Superscript II ( Invitrogen ) , following manufactures’ instructions . Complementary ( c ) DNA was analyzed in duplicate or triplicate by qPCR using Taqman gene expression assays ( Ppia; Mm03024003_g1; Hprt: Mm01545399_m1; Kat2a: Mm00517402_m1 ) and Taqman Gene Expression Mastermix ( Applied Biosystems ) . Gene expression levels were determined by the Pfaffl method following normalization to Reference gene , as stated . For exon 2–18 in-frame products , qPCR using Sybr Green Master Mix ( Applied Biosystems ) was performed in triplicates . Primers used were: Kat2a Exon 1–2 ( forward: GTCTTCTCAGCTTGCAAGGCC , reverse: AAAGGGTGCTCACAGCTACG ) ; Kat2a Exon2-18 ( forward: GTAGCTGTGAGCACCCTTTGG , reverse: TTCGCTGTCTGGGGGATTGT ) ; Kat2a Exon18 ( forward: CTCATCGACAAGTAGCCCCC; reverse: GTCCCTGGCTGGAGTTTCTC ) . MLL-AF9 secondary leukemia cells from Kat2a WT and KO backgrounds were freshly thawed , stained with Lin cocktail , c-Kit-APC-ef780 , Sca1-PE-Cy7 , CD16/32-PerCP-Cy5 . 5 , CD34-AF700 and Streptavidin BV510 ( see Table 1 ) , plus dead cell exclusion with Hoechst 32558 ( Thermo Fisher Scientific ) , sorted as Lin-c-Kit+Sca1-CD16/32+ and single-cell deposited into 96-well plates containing 3 μl/well of 0 . 67% NP-40 and 2U of RNasin Plus ( Promega ) in RNase-free water . Plates were vortexed after deposition , centrifuged for lysate collection , and frozen immediately in dry ice . Plates were stored at −80°C . Upon thaw , cDNA was synthesized from the single-cell lysates using 5U Superscript II ( Invitrogen ) and 1 μM each of gene-specific primers for Hprt and Kat2a ( outer reverse , Table 2 ) in a 10 μl reaction mix as per the manufacturer’s protocol ( 42°C , 1 hr; 70°C inactivation , 15 min ) . The total cDNA reaction was used in a 50 μl first round PCR with duplexed outer forward primers for Hprt and Kat2a at a final concentration of 200 nM and 1 . 25U HotStar Taq ( Qiagen ) in a reaction mix as per manufacturer’s protocol . Cycling conditions: 95°C , 15 min; 40* ( 94°C , 1 min; 60°C , 1 min; 72°C , 2 min ) ; 72°C , 5 min; 25°C , 30 s . Two μl of the first-round product were amplified in each of two separate second round PCR using nested primers for the two individual genes ( Table 2 ) . Reaction mixes as per the first round PCR , but in a final volume of 25 μl . Cycling conditions: 95°C , 15 min; 40* ( 94°C , 30 s; 60°C , 1 min; 72°C , 1 min ) ; 72°C , 5 min; 25°C , 30 s . Second-round products were run in a 2% agarose TAE1x gel at 50V , 1 hr , stained for 45–60 min in SYBRSafe solution ( Thermo Fisher Scientific ) in double-distilled water , and DNA visualized on a BioRad Imager . Pools of total BM cells from MLL-AF9 Kat2a WT and Kat2a KO primary leukemia samples ( ChIP-sequencing ) and of total BM or spleen cells from MLL-AF9 Kat2a WT and Kat2a KO secondary leukemias ( ChIP-qPCR ) were crosslinked with 1% Formaldehyde Solution ( Sigma Aldrich ) for 10 min at room temperature ( RT ) , with gentle rotation ( 50 rpm ) . Fixation was stopped with Glycine , and cells incubated for 5 min , RT , with gentle rotation ( 50 rpm ) , followed by two washing steps in ice-cold PBS . Cell pellets were resuspended in Lysis buffer ( 20 mM Hepes pH 7 . 6 , 1% SDS and 1/100 Protease Inhibitors cocktail ( PIC , Sigma Aldrich ) followed by Nuclei preparation . Chromatin pellets were sheared in a Bioruptor Pico Plus ( Diagenode ) in TPX tubes , using 3 runs of 11 cycles ( Cycle: 30 s ON 30 s OFF ) on high setting . A short spin was performed between runs and samples were transferred to new TPX tubes . 2–10% of total sheared chromatin was kept for input reference . Immunoprecipitation was set up using Dilution Buffer ( 0 . 15% SDS , 1% Triton X-100 , 1 . 2 mM EDTA , 16 . 7 mM Tris pH8 and 167 mM NaCl ) , PIC , and the respective antibody ( Table 3 ) and the sheared chromatin incubated overnight at 4°C with rotation . On the following day , protein A/G magnetic beads were pre-cleared with Dilution Buffer supplemented with 0 . 15% SDS and 0 . 1%BSA , then mixed with immunoprecipitation mix and incubated for at least 4 hr at 4°C with rotation . Chromatin-Antibody-Beads mixes were sequentially washed with ChIP Wash1 ( 2 mM EDTA , 20 mM Tris pH8 , 1% Triton X-100 , 0 . 1% SDS and 150 mM NaCl ) , ChIP Wash2 ( 2 mM EDTA , 20 mM Tris pH8 , 1% Triton X-100 , 0 . 1% SDS and 500 mM NaCl ) , ChIP Wash3 ( 1 mM EDTA and 10 mM Tris pH8 ) and captured on a magnetic rack . Captured beads were incubated for 20 min with rotation in freshly prepared Elution Buffer ( 1% SDS and 0 . 1M NaHCO3 ) . Supernatants were collected and decrosslinking performed overnight . DNA was column-purified using DNA clean and concentrator TM 5 KIT ( Zymo Research ) , according to manufacturer’s instructions , using 20 μL Zymo Elution Buffer . DNA quality control was performed using DNA Qubit 2 . 0/3 . 0 ( Invitrogen ) ; DNA fragment size <500 bp ( typically 200–300 bp ) was confirmed by gel electrophoresis of input material on a 1 . 5% agarose/TAE gel . For ChIP-qPCR , eluted DNA samples were diluted 3–5 times and tested by SYBR green qPCR ( primer sequences in Table 4 ) using 2 μl diluted DNA per triplicate reaction . Peak enrichments were quantified by the −2ΔΔCt method relative to mouse IgG ( Gabpa ChIP ) or rabbit IgG ( Myc ChIP ) , using the intergenic region in mouse chromosome 1 ( mChr1 ) as a reference . Duplicate to quadruplicate ChIP experiments were analyzed . For ChIP-sequencing analysis of histone modifications , DNA from duplicate immunoprecipitation experiments and the respective input material underwent library preparation with the NextFlex Rapid DNA kit ( 12 cycles ) at the MRC/WT Cambridge Stem Cell Institute Genomics Core Facility . After quality control , libraries were sequenced on an Illumina HiSeq 4000 sequencer at the CRUK Cambridge Research Institute , using 50bp-single-end sequencing . Raw ChIPseq reads were analyzed on the Cancer Genomics Cloud ( CGC ) platform ( Lau et al . , 2017 ) . Reads were aligned to the mm10 mouse genome using the Burrows-Wheeler Aligner ( BWA ) . Peaks from aligned reads were obtained using MACS2 peak calling algorithm with a significance q-value of 0 . 05 . The deepTools bamCoverage command ( Ramírez et al . , 2016 ) was used to determine reads enrichment relative to input; only ChIP-seq samples with clear separation from control were retained , with exclusion of one H3K4me1 and one H3K27ac replicate; consensus peaks were used for duplicate samples . To analyze changes in acetylation patterns at promoters and enhancers , H3K4me3 and H3K4me1 peaks respectively , were crossed with H3K9ac-only , H3K27ac-only and dual acetylation peaks from the corresponding genotypes . H3K9ac-only peaks associated with me3 were used for further analysis . Genomic peaks were obtained for Kat2a WT and Kat2a KO genotypes separately using Bedtools intersect ( Quinlan and Hall , 2010 ) and H3K4me3 K9ac peaks exclusive to WT retained as putative Kat2a peaks . Peak locations were converted to fastq sequences using UCSC table browser tool ( Karolchik , 2004 ) . Genomic Regions Enrichment of Annotations Tool , GREAT ( McLean et al . , 2010 ) was used to assign gene identities to the fastq sequences associated with putative Kat2a peaks , with gene promoter peaks called within - 1 kb to + 500 bp of the transcription start site ( TSS ) . We used ENCODE ChIP-Seq Significance Tool ( Auerbach et al . , 2013 ) to obtain putative transcription factor binding , including identification of genes bound by GCN5/KAT2A , to confirm the identity of putative Kat2a targets . Terminal BM samples from Kat2a WT and Kat2a KO MLL-AF9 primary leukemia animals were collected ( WT - 5; KO - 4 ) and the individual cell samples stored at −150°C . Cells were thawed and pooled for library preparation . Specifically , 12K live cells per genotype pool were sorted on an Influx sorter ( BD ) on the basis of YFP expression ( reporting MLL-AF9 ) , Hoechst 32258 exclusion ( live cells ) and singlet configuration ( pulse width ) and used for library preparation with Chromium Next GEM Single Cell 3’GEM , Library and Gel Bead Kit v1 ( 10XGenomics ) aiming at 6K single cells per sample . Library preparation and single-end sequencing on a NextSeq 500 sequencer were done at CRUK Cambridge Research Institute . Raw single cell RNAseq fastq reads were analysed using Cellranger software ( v2 . 1 . 1 ) to obtain the cell-gene count-matrix . Seurat ( Butler et al . , 2018 ) was used for pre-processing the count-matrix data and obtaining differential gene expression between the two genotypes . We employed pairwise distance between gene correlations as a measure of cellular heterogeneity as described ( Mohammed et al . , 2017 ) , by identifying the top 500 highly variable genes in both Kat2a WT and KO genotypes based on distance-to-median ( DM ) and calculating Spearman correlation coefficients between all gene pairs . The correlation matrix was used to compute the pairwise distance measure . RaceID/StemID ( Grün et al . , 2016 ) algorithms were used for clustering using t-SNE and obtaining pseudo-temporal arrangement of clusters based on entropy information and cluster stem scores . Monocle version 2 . 6 . 4 ( Trapnell et al . , 2014 ) was used as an alternative pseudo-time analysis for Kat2a WT and Kat2a KO cells . The relative cell state ordering in Monocle was unsupervised , with leukemia self-renewal Hoxa9 expression employed to determine the directionality of the trajectory . Parameters for the stochastic gene expression were fitted to the two-state promoter model using the D3E algorithm ( Delmans and Hemberg , 2016 ) with the Bayesian method option for model fitting . Global normalized data were used in the Robust gene set analysis; cluster-specific parameter derivation included computation of cluster-specific normalization . A multiple linear regression of CV and average gene expression with respect to burst size and burst frequency for Kat2a WT and Kat2a KO cells were performed using lm ( ) function in R statistical package . P-values for significant coefficients were calculated as a output of the lm ( ) function . Scripts for integration of single-cell RNA-seq and ChIP-seq data were coded in R Language ( version 3 . 4 . 4 ) and are provided with this submission . Gene Ontology analysis was performed with the PANTHER online tool ( Mi et al . , 2019 ) , selecting binomial analysis with Bonferroni correction . Gene sets were obtained from the Molecular Signatures Database ( MSigDB ) ( Subramanian et al . , 2005 ) to plot gene signatures on tSNE plots . The self-renewal gene signature associated with MLL leukemia , GCM_MLL , was employed in Figure 4—figure supplement 1A . For Figures 6A and 4 gene sets ( MORF_EIF4E , MORF_EIF3S2 , MORF_EIF4A2 and MORF_EIF3S6 ) were pooled to represent a translation-associated gene signature . Statistical tests performed are specified in the figure legends . Differences were obtained with significant p-value<0 . 05 . Analyses were performed in statistical language R ( version 3 . 4 . 4 ) or using Prism version 8 . 1 . 2 ( GraphPad ) . | Less than 30% of patients with acute myeloid leukaemia – an aggressive cancer of the white blood cells – survive five years post-diagnosis . This disease disrupts the maturation of white blood cells , resulting in the accumulation of immature cells that multiply and survive but are incapable of completing their maturation process . Amongst these , a group of cancer cells known as leukemic stem cells is responsible for continually replenishing the leukaemia , thus perpetuating its growth . Cancers develop when cells in the body acquire changes or mutations to their genetic makeup . The mutations that lead to acute myeloid leukaemia often affect the activity of genes known as epigenetic regulators . These genes regulate which proteins and other molecules cells make by controlling the way in which cells ‘read’ their genetic instructions . The epigenetic regulator Kat2a is thought to ‘tune’ the frequency at which cells read their genetic instructions . This tuning mechanism decreases random fluctuations in the execution of the instructions cells receive to make proteins and other molecules . In turn , this helps to ensure that individual cells of the same type behave in a similar way , for example by keeping leukaemia cells in an immature state . Here , Domingues , Kulkarni et al . investigated whether interfering with Kat2a can make acute myeloid leukaemia less aggressive by allowing the immature white blood cells to mature . Domingues , Kulkarni et al . genetically engineered mice to remove Kat2a from blood cells on demand and then inserted a mutation that causes acute myeloid leukaemia . The experiments showed that the loss of Kat2a delayed the development of leukaemia in the mice and progressively depleted leukaemia stem cells , causing the disease to become less aggressive . The results also showed that loss of Kat2a caused more fluctuations in how the white blood cells read their genetic code , which resulted in more variability in the molecules they produced and increased the tendency of the cells to mature . These findings establish that loss of Kat2a causes leukaemia stem cells to mature and stop multiplying by untuning the frequency at which the cells read their genetic instructions . In the future , it may be possible to develop drugs that target human KAT2A to treat acute myeloid leukaemia . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"chromosomes",
"and",
"gene",
"expression",
"cancer",
"biology"
] | 2020 | Loss of Kat2a enhances transcriptional noise and depletes acute myeloid leukemia stem-like cells |
Many developing neurons transition through a multi-polar state with many competing neurites before assuming a unipolar state with one axon and multiple dendrites . Hallmarks of the multi-polar state are large fluctuations in microtubule-based transport into and outgrowth of different neurites , although what drives these fluctuations remains elusive . We show that actin waves , which stochastically migrate from the cell body towards neurite tips , direct microtubule-based transport during the multi-polar state . Our data argue for a mechanical control system whereby actin waves transiently widen the neurite shaft to allow increased microtubule polymerization to direct Kinesin-based transport and create bursts of neurite extension . Actin waves also require microtubule polymerization , arguing that positive feedback links these two components . We propose that actin waves create large stochastic fluctuations in microtubule-based transport and neurite outgrowth , promoting competition between neurites as they explore the environment until sufficient external cues can direct one to become the axon .
During development , hippocampal neurons transit through a multi-polar intermediate state in which neurons typically extend 4–5 immature neurites , which are each capable of becoming either an axon or a dendrite ( Barnes and Polleux , 2009; Dotti et al . , 1988 ) . These neurites stochastically retract and elongate for a period of hours to days before a single neurite is specified as the axon ( Figure 1a ) . This delayed axon specification is typically mediated by extracellular cues such as soluble growth factors , neighboring cells , and the extracellular matrix ( Arimura and Kaibuchi , 2007; Barnes and Polleux , 2009 ) or in vitro by stochastic extension and stabilization of the longest single neurite ( Dotti et al . , 1988 ) . Previous studies showed that axon specification is based on a competition between “axon-promoting” signals such as Ras , Phosphatidylinositol-4 , 5-bisphosphate 3-kinase ( PI3K ) , and Protein Kinase A ( PKA ) and “axon-inhibiting” signals such as Glycogen synthase kinase 3 beta ( GSK3β ) and Protein Kinase G ( PKG ) ( Barnes and Polleux , 2009; Shelly et al . , 2010 ) . It is believed that these signals are controlled by the selective accumulation of axon-promoting proteins in the nascent axon via microtubule-based transport involving one or more self-reinforcing positive feedback loops ( Figure 1a ) ( Cheng et al . , 2011; Fivaz et al . , 2008; Inagaki et al . , 2001; Shi et al . , 2003; Toriyama et al . , 2006 ) . During the multi-polar state , however , pro-axon components appear to dynamically shuttle collectively to and from different neurites through an unknown mechanism ( Figure 1a ) , a process that has been investigated using a constitutively active version of the microtubule motor Kinesin-1 ( a . a . 1–560 , CA-KIF5C ) ( Hammond et al . , 2010; Jacobson et al . , 2006; Konishi and Setou , 2009; Toriyama et al . , 2010 ) . Kinesin-1 carries numerous proteins known to promote axon formation such as CRMP2 ( Kimura et al . , 2005 ) and WAVE/Sra ( Kawano et al . , 2005 ) , and perturbing Kinesin-1 expression or localization inhibits single axon formation ( Konishi and Setou , 2009 ) . There is also evidence that the Kinesin-1 adaptor protein c-Jun N-terminal kinase-interacting protein-1 ( JIP1 ) ( Dajas-Bailador et al . , 2008 ) and the PI3K interactor Shootin1 ( Toriyama et al . , 2006 , 2010 ) also relocalizes from one neurite to another before settling in the nascent axon . A similar dynamic collective relocalization of this kinesin motor domain has also been observed in vivo in developing neurons expressing CA-KIF5C ( Randlett et al . , 2011 ) . What drives these fluctuations is a fundamental open question , as the number , orientation , or age of microtubules may play a role , and conflicting studies suggest that Kinesin-1 may preferentially bind microtubules that are stable ( Cai et al . , 2009; Hammond et al . , 2010; Konishi and Setou , 2009; Reed et al . , 2006 ) or newly polymerized ( Nakata et al . , 2011; Valesoq et al . , 1994 ) . These results regarding microtubule-based transport may also be related to an earlier observation that bulk cytoplasmic flow precedes axon specification ( Bradke and Dotti , 1997 ) . 10 . 7554/eLife . 12387 . 003Figure 1 . Stochastically-generated actin waves correlate with neurite extensions . ( a ) Schematic showing two stages of symmetry breaking . The multi-polar phase , where the neuron experiences fluctuating neurite outgrowth and retraction and fluctuating microtubule-based transport , is highlighted . ( b ) Timelapse images of a F-tractin-mCherry-expressing primary hippocampal neurite showing actin wave propagation . Images were taken every 2 min . Yellow arrowheads mark the actin wave , yellow asterisks mark the neurite tip , white arrow marks direction of actin wave progression . Scale bar = 5 μm . ( c ) Structured illumination images of a phalloidin-stained neuron showing an actin wave . Top image is a x-z projection of a z-stack of images taken every 0 . 125 μm , bottom image is a maxiumum intensity projection of the z-stack . White arrow marks direction of actin wave progression . ( d ) Kymograph generated from a timelapse of a F-tractin-mCherry expressing neurite . Source images were acquired every 5 min . ( e ) Timelapse images of a F-tractin-mCherry-expressing neurite undergoing a growth spurt as the actin wave impacts the growth cone . Images were acquired every 5 min . Yellow arrowheads mark actin waves , yellow asterisks mark neurite tips . Scale bar = 15 μm . ( f ) Actin waves are stochastically generated in different neurites over time . Actin wave generation was assessed by eye in all neurites of a single neuron over time . Horizontal bars mark individual neurites , white dashes mark actin waves . Source images were acquired every 5 min . DOI: http://dx . doi . org/10 . 7554/eLife . 12387 . 00310 . 7554/eLife . 12387 . 004Figure 1—figure supplement 1 . Frequency of actin waves . DOI: http://dx . doi . org/10 . 7554/eLife . 12387 . 00410 . 7554/eLife . 12387 . 005Figure 1—figure supplement 2 . Speed of actin waves . DOI: http://dx . doi . org/10 . 7554/eLife . 12387 . 005 Here we show that actin waves , growth-cone-like cytoskeletal structures , promote anterograde microtubule and kinesin-based transport during the multi-polar state of symmetry breaking . Previous observations of actin waves showed that they are triggered stochastically , linked to neurite outgrowth and axon specification , more prevalent in early neurons compared to mature neurons , and found in cultured organotypic slices as well as primary hippocampal neurons ( Flynn et al . , 2009; Katsuno et al . , 2015; Ruthel and Banker , 1998 , 1999 ) . It has been suggested that they constitute a transport mechanism that brings actin and actin associated proteins towards growth cones ( Flynn et al . , 2009; Katsuno et al . , 2015 ) . Our study shows that actin waves act in concert with microtubules to direct microtubule-based transport and that these discrete transport events are tightly linked to neurite outgrowth during the multi-polar state before an axon is specified .
We first hypothesized that actin waves may link to microtubules after observing an increase in neurite volume , or cross-sectional area , in and behind the actin wave ( Figure 2a ) , which was confirmed with averaged line scan analysis ( Figure 2b , Figure 2—figure supplement 1 ) . Immunofluorescence experiments staining for neuronal tubulin revealed an increase in microtubule intensity in and behind actin waves ( Figure 2c ) , which was confirmed by structured illumination microscopy ( SIM ) of single microtubules ( Figure 2e ) and using averaged line scans of lower-resolution immunofluorescence images ( Figure 2d ) . As a control , we confirmed using lower-resolution IF ( Figure 2f ) and SIM ( Figure 2—figure supplement 2 ) that the increases in tubulin levels behind versus in front of an actin wave are not the result of a gradual thickening of neurite shafts closer to the cell body ( microtubule number ratio in Figure 2—figure supplement 2 is statistically greater than control measurements in Figure 2f ) . 10 . 7554/eLife . 12387 . 009Figure 2 . Actin waves contain more polymerizing microtubules in widened neurites . ( a ) The volume marker cytoplasmic Turquoise shows an increase in volume in and behind the wave . Images acquired every 8 min . Yellow arrowheads mark actin waves . White arrow denotes direction of wave movement . Scale bar = 10 µm . ( b ) Averaged line scans show increased volume in and behind actin wave . Measurements are taken from cytoplasmic Turquoise and F-tractin-mCherry expressing cells . Gray arrow denotes direction of wave movement . Dashed line indicates alignment at half max of actin wave . All traces were normalized by mean intensity then smoothed before averaging . Error is standard deviation . N = 14 neurites . ( c ) Fixed hippocampal neurons stained with phalloidin ( actin ) and anti-βIII tubulin ( neuronal microtubules ) show enrichment of microtubules in and behind wave . White arrowheads mark actin waves , white asterisks mark neurite tips . Scale bar = 10 μm . ( d ) Quantification of ( c ) confirmed enrichment of microtubules in and behind wave . Averaged line scans of phalloidin signal and anti-βIII tubulin signal were obtained for neurites containing waves . See 2b for methodology . N = 27 neurites . ( e ) Structured illumination microscopy on phalloidin and anti-βIII tubulin-stained neurons shows enrichment of microtubules behind wave with single-microtubule resolution . White arrow marks direction of actin wave propagation . Scale bar = 10 μm . ( f ) Fold enrichment of phalloidin and anti-βIII tubulin intensity behind the wave was calculated by taking the ratio of intensities in an area behind the actin wave to an area in front of the actin wave ( depicted on left: region1/region2 ) . Fold enrichment was calculated for neurites containing waves ( “Wave” , n = 20 ) and neurites lacking waves ( “Ctr” , n = 12 ) . Tubulin enrichment was statistically higher in waves compared to the control ( two-sided Wilcoxon rank sum test ) . ( g ) Hippocampal neurons expressing F-tractin-mCherry and YFP-EB1 show enrichment of EB1 puncta in and behind actin wave . Yellow arrowheads mark front edge of actin waves . Yellow asterisks mark neurite tips . Scale bar = 10 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 12387 . 00910 . 7554/eLife . 12387 . 010Figure 2—figure supplement 1 . 2D line scan analysis method . DOI: http://dx . doi . org/10 . 7554/eLife . 12387 . 01010 . 7554/eLife . 12387 . 011Figure 2—figure supplement 2 . Single microtubule enrichment behind actin wave . DOI: http://dx . doi . org/10 . 7554/eLife . 12387 . 01110 . 7554/eLife . 12387 . 012Figure 2—figure supplement 3 . EB1 puncta move in an anterograde fashion . DOI: http://dx . doi . org/10 . 7554/eLife . 12387 . 012 We then set out to find the source of the increased microtubule number in and behind actin waves and tested the hypothesis that actin waves contain more polymerizing microtubules by imaging neurons that co-express F-tractin and fluorescently-tagged EB1 , which binds to the plus ends of growing microtubules . Indeed , live cell imaging of F-tractin and EB1 revealed an increase in the number of EB1 puncta within and behind actin waves , with much lower levels of EB1 puncta in front of the wave ( Figure 2g , Figure 9b ) . This enrichment in the number of polymerizing microtubules progresses down the neurite with the actin wave , creating a dual wave of polymerizing microtubules and polymerizing actin . Control experiments using a maximum intensity projection of time lapse images showed that single EB1 puncta moved in a persistent fashion outward , confirming that they track with anterograde polarized polymerizing microtubules ( Figure 2—figure supplement 3 ) , consistent with previous measurements showing that microtubules generally polymerize outward during the multi-polar phase ( Stepanova et al . , 2003 ) . This striking co-localization of polymerizing actin and increased microtubule polymerization prompted us to investigate the correlation between actin waves and microtubule-based transport by using the minimal Kinesin 1 motor domain construct ( CA-KIF5C-Venus; a . a . 1–560 , adapted from Jacobson et al . ( 2006 ) ) . The attachment and movement of this motor domain to microtubules is not regulated by conformational changes and is constitutively active ( Friedman and Vale , 1999 ) . The mechanisms controlling the dynamic localization pattern of CA-KIF5C during symmetry breaking have previously been investigated , and the dynamic , stochastic nature of actin wave entry into various neurites struck us as reminiscent of the dynamic localization of CA-KIF5C . We first confirmed that the Kinesin-1 motor , CA-KIF5C , is often enriched at the tip of one or sometimes more neurites ( Jacobson et al . , 2006 ) during the multipolar phase ( Figure 3a ) . We also confirmed dynamic switching of the localization of CA-KIF5C back and forth between one or more neurites via the cell body over a period of many hours until most of the construct ultimately localizes to the emerging axon ( Video 4 ) . 10 . 7554/eLife . 12387 . 013Figure 3 . Actin waves coordinate with pulsatile transport of Kinesin-1 motor domain . ( a ) Live cell images of a neuron expressing CA-KIF5C-Venus ( green ) and F-tractin-mCherry ( white ) exhibiting fluctuating CA-KIF5C localization and neurite lengths characteristic of the multi-polar stage . Images were acquired every hour . Scale bar = 30 μm . ( b ) For 45 visually-identified actin waves , total CA-KIF5C intensity in the neurite was measured before and after generation of the actin wave . A significant increase in CA-KIF5C is observed relative to a control . Control traces were obtained by randomly selecting 45 points in time and assessing CA-KIF5C intensity before and after each time point . The black line signifies the mean of the CA-KIF5C traces . For each time point the data is also represented with standard box plots with outliers not shown . Significance between the -10 and 10 min set of points was assessed using a two-sided Wilconox rank-sum test . ( c ) Increase in actin intensity precedes increase in CA-KIF5C intensity . Total intensities of actin and CA-KIF5C before , during , and after entry of CA-KIF5C was assessed and the delay between actin and CA-KIF5C intensity increase was noted . ( 28 entry events ) . Source images were acquired every 5 min . ( d ) Timelapse images of a CA-KIF5C-Venus and F-tractin-mCherry expressing neurite show that CA-KIF5C transports in pulses which coincide with actin waves . Images were taken every 10 min . White arrowheads mark position of actin waves . White asterisks mark neurite tips . Scale bar = 10 μm . ( e ) CA-KIF5C moves with an actin wave as illustrated by single cell successive line scans taken from a neurite with a traveling actin wave . Image data is in Figure 3—figure supplement 2 . Frames were acquired every 5 min . ( f ) Averaged line scans show enrichment of CA-KIF5C in and behind the actin wave . See 2b for methodology . ( g ) Kymographs generated from timelapse images of a F-tractin-mCherry and CA-KIF5C-Venus expressing neurite show that CA-KIF5C travels with actin waves and accumulates in the growth cone . Source images were acquired every 5 min . DOI: http://dx . doi . org/10 . 7554/eLife . 12387 . 01310 . 7554/eLife . 12387 . 014Figure 3—figure supplement 1 . CA-KIF5C dynamic localization is dependent on microtubules . DOI: http://dx . doi . org/10 . 7554/eLife . 12387 . 01410 . 7554/eLife . 12387 . 015Figure 3—figure supplement 2 . CA-KIF5C travels with actin waves . DOI: http://dx . doi . org/10 . 7554/eLife . 12387 . 01510 . 7554/eLife . 12387 . 016Figure 3—figure supplement 3 . Actin waves can reverse retrograde CA-KIF5C movement . DOI: http://dx . doi . org/10 . 7554/eLife . 12387 . 01610 . 7554/eLife . 12387 . 017Video 4 . CA-KIF5C switches between neurites before localizing into a single neurite . This movie shows timelapse images from a neuron expressing F-tractin-mCherry ( white ) and CA-KIF5C-Venus ( green ) . Images were collected every 15 min and the movie was generated at 5 frames per second . Scale bar = 30 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 12387 . 017 Next , we determined whether the stochastic appearance of actin waves in neurites is temporally correlated with that of CA-KIF5C . We marked the time when actin waves entered neurites by visual inspection and averaged the time course of change in total CA-KIF5C intensity in the neurite receiving the actin wave . This analysis showed in the same neurite a marked increase in CA-KIF5C over a period of over 10 min following an actin wave ( Figure 3b ) . Moreover , we found that when CA-KIF5C entered a neurite previously lacking CA-KIF5C , such entry events occurred in parallel with an actin wave ( in 34 out of 35 entry events ) . We then assessed the time lag ( if any ) between the increases of actin and CA-KIF5C signals during CA-KIF5C entry events and found that the increase in actin intensity preceded the increase in CA-KIF5C intensity by several minutes ( Figure 3c ) . To address the concern that the co-localization of actin waves and CA-KIF5C entry into neurites may be an artifact of a cytoplasmic fraction of CA-KIF5C , we imaged CA-KIF5C moving in and out of neurites in neurons also expressing a cytoplasmic Turquoise fluorescent protein . Taking the ratio of CA-KIF5C over the Turquoise volume marker shows that the movement of CA-KIF5C into and out of neurites is significant over changes in volume; moreover , addition of the microtubule polymerization inhibitor Nocodazole eliminates the movement of CA-KIF5C into and out of neurites relative to a volume marker , confirming that CA-KIF5C movement is dependent on microtubules ( Figure 3—figure supplement 1 ) . In a parallel analysis of the spatial correlation between actin polymerization and CA-KIF5C , we found that CA-KIF5C was transported in pulses , or waves , along the neurite shaft to the growth cone , coincident with actin waves ( Figure 3d , Figure 3—figure supplement 2 , Video 5 ) . We confirmed this correlation using single cell analysis ( Figure 3e ) , averaged line traces ( Figure 3f ) , and kymographs ( Figure 3g ) . Most of the pulses of CA-KIF5C arriving at growth cones resulted in an increase in CA-KIF5C concentration that persisted for tens of minutes ( Figure 3g ) . Moreover , the arrival of an actin wave can reverse the retrograde movement of CA-KIF5C out of the neurite ( Figure 3—figure supplement 3 ) . Taken as a whole , this data shows a strong connection between CA-KIF5C movement and actin waves . 10 . 7554/eLife . 12387 . 018Video 5 . Pulsatile CA-KIF5C transport coincides with moving actin waves . This movie shows timelapse images of the F-tractin-mCherry- and CA-KIF5C-Venus-expressing neurons displayed in Figure 3d . Images were collected every 5 min and the movie was generated at 5 frames per second . Scale bar = 30 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 12387 . 018 To further verify an increase in microtubule-based transport in actin waves , we expressed fluorescently-tagged Synaptophysin , a pre-synaptic vesicle marker , in order to image vesicular movement . Similar to CA-KIF5C localization , Synaptophysin-positive vesicles were highly enriched in and behind actin waves ( Figure 4a ) . Averaged line scans of intensity profiles of only moving vesicles ( Moving Vesicle Intensity ( MVI ) ) showed the same results ( Figure 4b see Materials and methods ) . Higher frequency imaging showed that Synaptophysin-positive vesicles enriched in the actin waves were not subject to the Brownian motion suggesting that they remain attached to microtubules ( Video 6 ) . Taken as a whole , this data suggests a strong correlation between actin waves , increased microtubule polymerization , and increased microtubule based transport . 10 . 7554/eLife . 12387 . 019Figure 4 . Actin waves contain Synaptophysin-positive vesicles . ( a ) Timelapse imaging of a neurite expressing F-tractin-mCherry and Citrine-Synaptophysin shows Synaptophysin positive vesicles enriched in and behind wave . Images displayed with inverted grayscale . Frames were taken every 4 min . Red and purple arrowheads mark front edge of actin waves . Red and purple asterisks mark neurite tips . Half asterisks mark neurites continuing out of frame . Scale bar = 5 μm . ( b ) Difference imaging of images acquired every 600 ms ( schema left ) taken as average line scans shows enrichment of mobile vesicles in and behind actin waves ( right ) . Grey arrow denotes direction of wave movement . Dashed line indicates alignment at half max of actin wave . All traces were normalized by mean intensity then smoothed before averaging . Error is standard deviation . N = 12 neurites . DOI: http://dx . doi . org/10 . 7554/eLife . 12387 . 01910 . 7554/eLife . 12387 . 020Video 6 . Synaptophysin-positive vesicles in actin wave do not experience Brownian motion . This movie shows timelapse images of an F-tractin-mCherry and Citrine-Synaptophysin expressing neurite with an actin wave . A single F-tractin-mCherry image was taken to identify the actin wave , followed by timelapse imaging of Citrine-Synaptophysin . Images were acquired every 600 ms and the movie was generated at 5 frames per sec . Scale bar = 10 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 12387 . 020 We next examined whether the striking connection between actin waves , microtubule polymerization and microtubule-based transport are causal . We first probed the potential role of actin waves in promoting microtubule polymerization and microtubule-based transport by examining both microtubule polymerization and microtubule-based transport after the addition of Jasplakinolide , an actin-stabilizing agent that stalls the progression of actin waves . We found that stalling actin waves halted the progression of CA-KIF5C ( Figure 5a , b ) as well as the progression of the wave of polymerizing microtubules ( Figure 5c , d ) , suggesting that progression of actin waves was necessary in order to continue the forward progression of the other two components . Notably , EB1 comets were still observed in the stalled actin wave , suggesting that although microtubules were still polymerizing in the actin waves , they were unable to polymerize beyond the structural bottleneck present at the front of the actin wave . This observation suggests that a mechanical mechanism may , at least partially , drive the increase in microtubule polymerization observed in the actin wave , and may explain the conundrum presented by the discrepant speeds of the components moving along with actin waves . More precisely , vesicular transport and Kinesin motors both advance at a speed of ~50 μm/min along microtubules while actin waves only advance at a speed of ~ 2 μm/min ( Figure 1—figure supplement 2 ) . This raises important mechanistic questions of why Kinesin motors and vesicles have a net progression that is much slower than their molecular transport rate , why motors and vesicles appear to be restrained behind the leading edge of actin waves , and why vesicular transport stays elevated in and behind an actin wave ( Figure 4a ) . 10 . 7554/eLife . 12387 . 021Figure 5 . Forward advance of microtubule polymerization and Kinesin-1 is dependent on actin wave progression . ( a ) Wave of CA-KIF5C does not advance independently of actin wave advancement . Addition of 10 nM Jasplakinolide stalls actin wave and movement of CA-KIF5C . Frames were acquired every 4 mins . White arrowheads mark actin waves . White half asterisks mark neurites continuing out of frame . Scale bar = 15 μm . ( b ) Quantification of ( a ) . Speeds of actin waves and CA-KIF5C waves were measured before and after Jasplakinolide addition . Error bars represent standard deviation . N = 6 neurites . Statistical significance assessed with a two-sided Wilcoxon rank-sum test . ( c ) Wave of polymerizing microtubules does not advance independently of actin wave . Addition of 50 nM Jasplakinolide freezes actin wave and prevents wave of EB1 puncta from moving forward . Images were taken every 4 min . Yellow arrowheads mark front edge of actin waves . Yellow asterisks mark neurite tips . Half asterisks mark neurites continuing out of frame . Scale bar = 10 μm . ( d ) Quantification of ( c ) . Speeds of actin waves and waves of EB1 puncta were measured before and after Jasplakinolide addition . Error bars representation standard deviation . N = 6 neurites . Statistical significance assess with a two-sided Wilcoxon rank-sum test . DOI: http://dx . doi . org/10 . 7554/eLife . 12387 . 021 To answer these discrepancies , we speculated that the bottleneck at the front of the actin wave restrains the movement of polymerizing microtubules and microtubule-based transport , and that the dilation of the neurite shaft caused by the actin wave may spatially de-restrict microtubules and allow more room for microtubules to polymerize towards the growth cone . In this hypothesis , the increase in the number of microtubules could enhance the flux of microtubule-based transport through the neurite . Another motivation for this hypothesis was that the speed of microtubule polymerization is ~10x faster than the speed of actin waves ( in neurite: 20 . 4 +/- 5 . 4 um/min ( 39 measurements ) in wave: 16 . 8 +/- 5 . 4 ( 23 measurements ) ) , which is consistent with previously measured microtubule polymerization rates ( Stepanova et al . , 2003 ) . Thus , in order for the wave of EB1 puncta to advance at the same speed as the actin wave , individual EB1 puncta must stall and disappear at the leading edge of the actin wave , similar to what we see when a polymerizing microtubule hits the membrane of the neurite tip ( Figure 6a , left ) . Indeed , kymograph analysis of EB1 comets moving through the bottleneck of the actin wave revealed that EB1 comets disappear at the bottleneck ( Figure 6a , right ) . As a separate experiment , we quantifiedthe flow of EB1 puncta through two windows placed right before and right after the bottleneck ( but within 6 μm of each other ) at time points 100 s apart which showed the same marked difference consistent with a loss of EB1 comets in between . In control experiments , we observed that the flow through each window does not significantly change over the time window measured , suggesting that microtubule polymerization is at a steady state over short time periods ( Figure 6b ) . Also , the difference between the flow of EB1 puncta through the two windows does not change over time ( Figure 6—figure supplement 1 ) . Thus , the larger flow of polymerizing microtubules through the window closer to the cell body does not reach the second window ahead of the actin wave , suggesting EB1 puncta disappear between the two windows . To understand this result , it is helpful to again consider that processive EB1 puncta should move 25–30 μm in approximately 100 s , much faster than the actin waves . We also measured EB1 puncta flow versus neurite width in a smooth neurite or within an actin wave which reveals a correlation between neurite width and the number of EB1 puncta ( Figure 6c ) . The lack of a pronounced higher number of EB1 puncta per unit area in actin waves relative to smooth neurites suggests that the mechanism by which actin waves promote microtubule polymerization is mainly steric , although this does not rule out other signaling cross-talks between the two cytoskeletal components . Furthermore , as for microtubule polymerization , the number of microtubules present also correlates with neurite width , as shown by SIM ( Figure 6d ) . 10 . 7554/eLife . 12387 . 022Figure 6 . Structural bottleneck provided by actin waves inhibits progression of polymerizing microtubules . ( a ) Image depicting EB1 puncta at a neurite tip ( top left ) and at a bottleneck provided by an actin wave ( top right ) . Accompanying kymographs illustrating EB1 puncta disappearing at the growth cone tip ( bottom left ) and the bottleneck of an actin wave ( bottom right ) are below . For better signal , the actin image of the neurite tip was constructed using a maxiumum intensity projection . Images were acquired every 2 sec for 2 min . Scale bars are both 3 μm , with the horizontal axis of the kymographs matching the spatial scale of the images . ( b ) Flow of EB1 puncta is restricted by a bottleneck . Flow ( number of EB1 puncta through a plane over 20 sec ) was assessed in two windows <6 μm apart ( w1 and w2 ) on each side of a structural bottleneck at two time points separated by 100 sec . Flow through each window does not significantly increase or decrease between t1 and t2 , however the number of EB1 puncta moving through w1 was significantly higher than the number moving through w2 at each time point . Different colors denote different neurites . Signicance was assessed using a two-sided sign test ( testing difference between measurement in t1 and t2 ( p = 1 for w1 and p = 0 . 6 for w2 ) and between w1 and w2 ( p = 0 . 03 , for both t1 and t2 ) ) . Colors represent distinct neurites . ( c ) Flow of EB1 puncta ( defined in ( b ) ) assessed in neurites with waves ( red ) and without waves ( blue ) of varying widths . Both sets ( red and blue ) display a linear correlation between neurite width and puncta flow . Flows per unit width for neurites bearing waves falls within the distribution for neurites lacking waves . Pearson’s correlation coefficients are 0 . 60 for wave case and 0 . 68 for smooth neurite case . ( d ) Analysis of individual number of microtubules ( visualized with SIM ) assessed in neurites of varying widths shows a positive correlation . Measurements made in 9 distinct neurites . DOI: http://dx . doi . org/10 . 7554/eLife . 12387 . 02210 . 7554/eLife . 12387 . 023Figure 6—figure supplement 1 . The difference in flow between windows 1 and 2 does not change over time . DOI: http://dx . doi . org/10 . 7554/eLife . 12387 . 02310 . 7554/eLife . 12387 . 024Figure 6—figure supplement 2 . LatA treatment can cause neurite widening . DOI: http://dx . doi . org/10 . 7554/eLife . 12387 . 024 Consistent with an expected role for Rac-regulated actin polymerization in generating actin waves , we used a Raichu FRET reporter ( Komatsu et al . , 2011 ) and found high Rac activity in and around actin waves . Interestingly , we also used a Cdc42 Raichu FRET reporter and observed that only Cdc42 exhibited a relatively higher activity in front of the actin wave , suggesting that the consistent anterograde direction of actin waves propagation may in part be directed by Cdc42 ( Figure 7a , b ) . To directly investigate the dependence of microtubule polymerization on actin waves , we used a photo-activatable Rac1 construct designed by the Hahn laboratory to test whether Rac activity is sufficient to generate actin waves ( Wu et al . , 2009 ) . This approach led to the successful generation of either fully processive ( 5 times ) or partially processive ( 5 times ) actin waves during successful Rac1 activations ( 18 times ) using selective photo-excitation at the base of a neurite on a scanning confocal microscope ( example in Figure 7c ) . The generated waves caused neurite widening ( Figure 7—figure supplement 1 ) and increased numbers EB1 puncta ( Figure 7d ) , consist with naturally-generated waves . This suggests that Rac1 activity is sufficient to initiate actin waves and to generate the enrichment of microtubule polymerization observed in actin wave . 10 . 7554/eLife . 12387 . 025Figure 7 . Rac1 activity is sufficient to generate actin waves enriched in polymerizing microtubules . ( a ) Actin waves are high in Cdc42 and Rac1 activity . Neurons are expressing F-tractin-mCherry ( left ) and FRET sensors for Cdc42 ( top ) and Rac1 activity ( bottom ) , images were taken every 5 min . Scale bar = 8 . 4 μm . ( b ) Averaged line scans show enrichment of Cdc42 activity in and in front of the wave ( top ) and enrichment of Rac1 activity in the wave ( bottom ) . Methodology in 2b . N = 22 neurites ( Cdc42 ) , n = 19 neurites ( Rac1 ) . ( c ) Neuron expressing F-tractin-mCherry and Cerulean-PA-Rac1 generates actin wave upon local excitation . White asterisk marks excitation area . Excitation protocol is described in Materials and Methods . Images were acquired every 4 min . Scale bar = 10 μm . ( d ) Neuron expressing F-tractin-mCherry , Cerulean-PA-Rac1 and YFP-EB1 shows stereotypical widening and increase in EB1 puncta upon excitation of actin wave . White asterisk marks excitation area . Scale bar = 10 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 12387 . 02510 . 7554/eLife . 12387 . 026Figure 7—figure supplement 1 . Activation of PA-Rac1 leads to neurite widening . DOI: http://dx . doi . org/10 . 7554/eLife . 12387 . 026 Our data thus far has shown that the increases in microtubule polymerization is dependent on actin waves . We next investigated the corollary question of whether microtubules may also have a role in the advancement of actin waves . Indeed , consistent with a previous study ( Ruthel and Banker , 1998 ) , we found that the addition of Nocodazole caused actin waves to dissolve ( Figure 8a , b ) . We also found that the addition of a high dose of Taxol ( 50 nM ) , a small molecule that stabilizes microtubules , leads to non-processive bursts of actin polymerization , interfering with normal actin wave formation ( Figure 8c ) . 50 nM Taxol also appears to affect microtubule polymerization ( Figure 8—figure supplement 1 ) , indicating that microtubule stabilization and/or an effect on microtubule polymerization may be affecting actin wave production . These drug studies argue for a co-dependence between actin waves and microtubules – one is needed to advance the other . 10 . 7554/eLife . 12387 . 027Figure 8 . Microtubules are necessary to drive actin wave progression . ( a ) Actin waves are dependent on polymerizing microtubules . Addition of 1 μM Nocodazole dissolves an actin wave . Frames were taken every 2 min . Yellow arrowheads mark front edge of actin waves . Yellow asterisks mark neurite tips . Half asterisks mark neurites continuing out of frame . Scale bar = 15 μm . ( b ) Quantification of ( a ) , showing that addition of Nocodazole dissolves a significantly greater proportion of actin waves than addition of a control . Data averaged from 3 experiments . Statistical significance assessed with 2 sample t-test . ( c ) Addition of 50 nM Taxol causes non-processive actin polymerization . Neurons were imaged for 69 frames before addition of DMSO or Taxol , then imaged for another 69 frames ( 5 min per frame ) . Neurons were expressing F-tractin-Citrine . Kymographs were constructed in Fiji . DOI: http://dx . doi . org/10 . 7554/eLife . 12387 . 02710 . 7554/eLife . 12387 . 028Figure 8—figure supplement 1 . 50 nM Taxol can affect microtubule polymerization . DOI: http://dx . doi . org/10 . 7554/eLife . 12387 . 02810 . 7554/eLife . 12387 . 029Figure 9 . Individual actin waves drive transient increases in neurite width , microtubule polymerization , and microtubule-based transport . ( a ) EB1 puncta observed in a single section of neurite shaft before , during , and after wave progression ( schema , top ) . Images show lingering EB1 puncta after wave has passed ( bottom ) . Scale bar = 2 μm . ( b ) Quantification of ( a ) . Fold change of EB1 puncta counts and actin intensity in the wave versus before the wave , and after the wave versus before the wave , show increased numbers of EB1 puncta in the wave and lingering enrichment of EB1 puncta after the wave has passed . N = 14 , a two-sided sign test was used to asses statistical significance of a set of ratios distinct from 1 . ( c ) 2D kymograph were generated from timelapse images of hippocampal neurons expressing F-tractin-mCherry and YFP-EB1 ( Video 7 ) . Width was calculated from segmenting summed actin and EB1 image . Region containing actin wave was marked with a dashed line and superimposed on width and EB1 kymographs . Each kymograph was normalized from 0 to 1 . ( d ) 2D Kymograph shows transient enrichment of Synaptophysin vesicles in actin waves . Kymograph generated from timelapse imaging data ( Video 8 ) . Region containing an actin wave was marked with a dashed line and superimposed on the Synaptophysin kymograph . Each kymograph was normalized from 0 to 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 12387 . 029 Finally , we measured how long the effect of enhanced microtubule polymerization and microtubule-based transport lasts . If such changes were long-lasting , we would expect that neurites would steadily thicken over time , which has not been previously observed . Also , our analysis of neurite growth over 10 hr time periods suggest that the growth promoting effect of an incoming actin waves is lost after tens of minutes ( Figure 1d ) . To determine whether the induced microtubule polymerization persists after an actin wave , we imaged the number of EB1 puncta in a neurite segment before , during , and after actin waves ( Figure 9a , top ) . Markedly , the EB1 puncta number increased along with an actin wave and persisted after the wave passed ( Figure 9a , bottom; Figure 9b ) , in agreement with our analysis of the persistence of the spatial profiles of Synaptophysin , volume , and microtubule enrichment . We then generated kymographs of actin intensity , EB1 intensity , and neurite width in a single section of neurite with multiple actin waves ( Figure 9c ) . The kymographs reveal simultaneous increases in EB1 intensity and neurite width that remained past the passage of the actin wave , but gradually decayed with variable timescales ( Figure 9c , Video 7 ) . The transient nature of the increases in microtubule polymerization and microtubule-based transport was confirmed using a kymograph analysis of Synaptophysin vesicles in a neurite containing actin waves ( Figure 9d , Video 8 ) . Thus , our data as a whole suggests that each actin wave , by creating a transient increase in neurite width , creates a burst of microtubule polymerization and microtubule-based transport that will eventually decay after approximately 30 min and neurites will start to retract ( Figure 1d , Figure 10a ) . The transient characteristic of the increase in transport explains the frequent neurite retractions and the necessity of frequent actin waves to continually deliver cargo to a single growth cone to maintain and elongate the neurite and ultimately allow one of the neurites to dominate and become the axon . 10 . 7554/eLife . 12387 . 030Figure 10 . Model illustrating the co-regulation of actin and microtubules that drive microtubule-based transport . ( a ) Schematic of changes caused by the actin wave . Actin waves cause transient increases in cargo delivery by increasing microtubule polymerization . We propose that actin waves aid microtubule polymerization by widening the neurite , allowing more space for microtubules to polymerize within the shaft and leading to an increase in microtubule-based transport . However , the changes are transient and will fade in time . ( b ) Flow chart depicts working model: positive feedback between actin waves and microtubule polymerization increases microtubule-based transport . DOI: http://dx . doi . org/10 . 7554/eLife . 12387 . 03010 . 7554/eLife . 12387 . 031Video 7 . Actin waves coincide with transiently increased microtubule polymerization and neurite width . This movie shows timelapse images of an F-tractin-mCherry and YFP-EB1 expressing neuron generating actin waves . As actin waves moves through , increases in neurite width and EB1 puncta number were observed . Spatially cropped images were used to generate Figure 9c kymograph . Images were acquired every min and the movie was generated at 5 frames per sec . Scale bar = 8 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 12387 . 03110 . 7554/eLife . 12387 . 032Video 8 . Actin waves coincide with transiently increased numbers of Synaptophysin-positive vesicles . This movie shows timelapse images of an F-tractin-mCherry and Citrine-Synaptophysin expressing neuron generating actin waves . Actin waves coincide with increased numbers of vesicles . Spatially cropped images were used to generate Figure 9d kymograph . Images were acquired every min and the movie was generated at 5 frames per sec . Scale bar = 10 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 12387 . 032
Our study argues that developing neurons employ an interlinked cytoskeletal system whereby actin waves cross-talk with microtubules to direct microtubule-based transport and drive neurite extension . More broadly , our study suggests that the stochastic nature of actin waves leads to the stochastic increase of microtubule-based transport paired with growth cone extension in different neurites thereby creating the dynamic multi-polar state that allows a search for external cues and ultimately enables single axon formation . In particular , we show that the fluctuating actin waves control the previously observed pulsatile anterograde Kinesin transport generated during the multi-polar phase of symmetry breaking . We discovered that the link from actin wave to Kinesin-mediated transport appears to be based on a mechanical control mechanism whereby actin waves widen the neurite shaft and create the space needed for more microtubules to polymerize . The increase in microtubule polymerization in turn increases microtubule number and vesicular microtubule-based transport . In addition , these morphological changes and consequent changes in microtubule-based transport are transient , resulting in the pulsatile delivery of cargo to the growth cone and necessitating the generation of frequent actin waves in order to continue to deliver axon-promoting factors into the neurite . Finally , the results of our pharmacological perturbation of microtubules suggest that , in fact , actin waves and microtubules co-regulate each other through a positive feedback mechanism , potentially mediated by cross-talk mechanisms that have been identified in other systems ( Akhshi et al . , 2014 ) . In combination with studies in other systems , our results suggest that mechanical regulation is a major avenue of crosstalk from the actin cytoskeleton to microtubule growth during neurite outgrowth . Consistent with this finding , it has also been shown that depolymerizing actin in growth cones spatially de-restricts microtubule polymerization , thus promoting neurite growth ( Bradke and Dotti , 1999 ) . A later study further found that ADF/cofilin , an actin severing enzyme , promotes neuritogenesis by disassembling the actin meshwork at the cortex , allowing for microtubule polymerization to promote a new protrusion ( Flynn et al . , 2012 ) . In conjunction with these previous studies , our results argue that the actin cortex structurally inhibits microtubule extension , and that a stent-like broadening of the neurite shaft or other mechanisms to “loosen” the actin cortex , by actin waves or other processes , relieves this inhibition to enable more microtubules to extend forward . In the context of neuron polarization , the stochastic nature of actin waves allows each neurite to repeatedly grow and retract thereby allowing the growth cone of each neurite to spatially explore its local environment to find external axon-promoting inputs ( e . g . Arimura and Kaibuchi , 2007 ) . When receiving sufficient input , signaling in one of the neurites is expected to be strengthened and the neurite is marked as the future axon . Such a specification of one neurite as the axon is known to subsequently inhibit axonal maturation of the other neurites and convert them to dendrites ( Esch et al . , 1999; Ménager et al . , 2004; Shelly et al . , 2010 ) . Together with these previous considerations , our study argues that actin waves have a dual function . First , they are the drivers for neurite outgrowth by directing microtubule-based transport to deliver axon-promoting factors to the growth cones at the neurite tip . Second , they promote competition between neurites by generating large stochastic fluctuations that allow each neurite to sample “winning” and “losing” states as they explore their surroundings for the cues that will ultimately determine axon specification . Our study does not directly address the role of actin waves in in vivo symmetry breaking and single axon formation . However , imaging of later stage polarized neurons in culture showed that actin waves still move through the nascent dendrites and the axon , but their effect on neurite outgrowth appears to be smaller ( Video 9 ) . This suggests that actin waves may become less important after polarization . With respect to their in vivo relevance , actin waves have been observed in hippocampal and cortical neurons in slice cultures ( Flynn et al . , 2009; Katsuno et al . , 2015 ) , although whether they promote microtubule polymerization and transport is more difficult to explore in this setting . Nevertheless , the results of our study suggest that actin waves could play an equally important role in neurite extension in vivo , to allow neurites to properly sense the polarized growth cues provided by the developing brain . 10 . 7554/eLife . 12387 . 033Video 9 . Actin waves are observed in post-polarized neurons . This movie shows timelapse images of a polarized Lyn-mCherry-expressing neuron producing actin waves . Neuron was imaged on DIV2 under CO2 in standard culturing Neurobasal Media . Images were collected every 5 min and the movie was generated at 5 frames per second . Scale bar = 50 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 12387 . 033 Together , our study provides a key missing link in our understanding of the axonal symmetry breaking process by demonstrating that the multi-polar state is characterized by actin waves driving microtubule-based transport , a link that we show is likely generated by an actin wave-triggered stent-like dilation of the neurite shaft . Our study further argues that the stochastic nature of actin waves creates the previously described stochastically fluctuating microtubule-based transport events during the multi-polar state . Thus actin waves may play a critical role in a stochastic search mechanism that allows a set of neurites to explore the surrounding space , integrate external signaling cues , and measure relative input differences to select the a “winning” axon .
Primary hippocampal neurons were cultured from Wistar rat E18 embryos . Hippocampi were removed from embryonic brains , placed in HBSS ( Gibco , Life Technologies , NY ) and incubated with 0 . 25% trypsin ( Life Technologies ) and 0 . 1% DNase ( Roche Applied Science , Penzberg , Germany ) for 15 min at 37°C . Hippocampi were then mechanically dissociated using gentle pipetting . Neuronal cells were either plated at 20 , 000–50 , 000 cell/cm2 or electroporated using the Nuclefector Amaxa system ( Lonza , Basal , Switzerland , Rat Neuron Nuclefector Kit , CN # VPG-1003 ) according to manufacturer’s instructions then plated at 50 , 000 cell/cm2 . Cells were plated in MEM + Glutamax ( Gibco ) containing 3% glucose and 10% serum and then switched to standard culturing media ( Neurobasal media ( Gibco ) with 1% Pen/Strep/Glut ( Life Technologies ) , Glutamic acid ( Life Technologies ) and 2% B27 ( Gibco ) or 2% SMI ( Stemcell Technologies , Vancouver , Canada ) ) 2–4 hr following plating . Cell were plated in LabTek II Chambered Coverglass chambers ( Nunc , Rochester , New York ) , 96-well glass bottomed plates ( In Vitro Scientific , Sunnyvale , CA ) or 96-well plastic bottomed plates ( Costar , Corning , New York ) coated with poly-l-lysine ( 0 . 1 mg/mL , MW 30 , 000–70 , 000 Sigma Aldrich ) or poly-l-lysine and laminin ( 2 μg/mL , Mouse Protein , Natural , Life Technologies ) . Jasplakinolide ( Santa Cruz Biotechnology , Dallas , Texas ) was used at 10 nM ( extracellular buffer ) or 50 nM ( CO2 independent media ) depending on which media was used . Nocodazole ( Sigma Aldrich , St . Louis , Missouri ) , was used at concentrations specified in the Figure Legends . Taxol ( Abcam , Cambridge , MA ) was used at 50 nM . The Rac1 and Cdc42 FRET sensors were gifts from Michiyuki Matsuda and used the optimized backbone described in Komatsu et al . , 2011 . All other constructs used were cloned to express under a CAG promoter . Original F-tractin , EB1 ( human ) , Synatophysin ( rat ) , and CA-KIF5C ( a . a . 1–560 of KIF5C ( rat ) ) constructs were gifts from Michael Schell , Clare Waterman , Craig Garner , and Gary Banker respectively . mCerulean-PA-Rac1 was ordered from Addgene ( Addgene #22030 ) . Lyn-mCherry was used as a membrane marker , with the N-terminal Lyn sequence used described in ( Inoue et al . , 2005 ) . The original GFP-KIF5C ( 1–560 ) vector was reassembled with Venus or Turquoise with the pCAGEN ( Addgene #11160 [Matsuda and Cepko , 2004] ) backbone using PCR assembly methods while maintaining the original linker sequence . For the remaining constructs , all final constructs were generated by Gateway cloning using the same methodology . First , a destination vector with the desired CAG promoter ( for higher , more even expression in primary cells [Qin et al . , 2010] ) was generated . pCAGEN-DEST was constructed by taking the pCAGEN backbone and using the Gateway Vector Conversion System . Briefly , a Gateway cassette ( Life Technologies ) containing attR recombination sites with the ccdB gene and a chloramphenicol-resistance gene was inserted into the EcoRV restriction site contained within pCAGEN . Full-length YFP-EB1 , F-tractin-mCherry ( Wollman and Meyer , 2012 ) , pCerulean-PA-Rac1 , and Citrine-Synaptophysin was PCRed with the TOPO-compatible tag CACC at the 5’ end and put into a pENTR vector using the pENTR/D-TOPO vector kit ( Life Technologies ) . The resulted entry clones were then subject to an LR reaction ( LR Clonase II Enyzme Mix , Life Technologies ) with pCAGEN-DEST . Hippocampal neurons were fixed for 20 min in 4% paraformaldehyde and 4% sucrose in PBS . 2x fixation solution was added to native media . For IF , neurons were blocked for 1 hr in blocking/staining solution ( 3% Normal Goat Serum , 0 . 5% BSA , 0 . 2% TX-100 in PBS ) , incubated in primary antibody for one hour in blocking/staining solution and incubated in secondary antibody for one hour in blocking/staining solution , with standard washes . Primary antibodies: Neuronal Class III β-tubulin ( TUJ1 , 1:1000 dilution , Covance , Princeton , New Jersey ) . Secondary antibodies: Alexa-Fluor 488 ( 1:1000 , Life Technologies ) . Dyes: Alex-Fluor 594 Phalloidin ( 1:400 , Life Technologies ) . Live cell imaging was conducted on a Zeiss Axiovert 200 M inverted epifluorescent microscope ( Zeiss , Oberkcochen , Germany ) equipped with a Nipkow spinning disc confocal and 488 , 514 and 594 nm lasers . Images were acquired on a CoolSnap HQ CCD camera ( Photometrics , Tucson , Arizona ) using 20x ( 0 . 75 NA ) , 63x water ( 1 . 2 NA ) , and 100x oil ( 1 . 4 NA ) Zeiss objectives . Images were acquired with Micro-Manager ( Edelstein et al . , 2010 ) and processed using methods described below . For some experiments , live cell and fixed cell images were acquired on the ImageXpress Micro XLS Widefield High Content Screening System ( Molecular Devices , Sunnyvale , CA ) using 20x ( 0 . 45 or 0 . 75 NA ) Nikon objectives . Photo-excitation studies were performed on a Leica SP8 scanning confocal microscope with a white-light laser and a 40x ( 1 . 3 NA ) objective . Photo-excitation was performed by locally exciting a region of the neuron with 480 nm light on the white-light laser ( 10 rounds of 5 excitations , 10 sec between rounds , 10% power on the FRAP user interface ) . Structured illumination microscopy was performed on a GE/Applied Precision OMX V4 ( GE Healthcare , Little Chalfont , UK ) at the Neuroscience Microscopy Service at Stanford . All live cell imaging was performed at 37C in the absence of CO2 unless otherwise noted . For imaging , CO2 independent media ( Gibco ) or custom-made extracellular imaging buffer was used with added Pen/Strep/Glut ( Life Technologies ) and B27 ( Gibco ) or SMI ( Stemcell Technologies ) as a vitamin supplement . Image analysis was conducted using basic image analysis tools available in Fiji ( Schindelin et al . , 2012 ) or custom-written MATLAB scripts . Total neurite intensity: analysis was performed by generating a binary mask of the summed timelapse set of images , and assessing total intensity within the mask for each timepoint . 2D line scans: analysis was conducted as follows 1 ) generate binary mask of neurite section of interest , 2 ) skeletonize mask , 3 ) select two endpoints on skeleton to generate “shortest path” aka a single path , 4 ) divide path into regular coordinates with desired spacing , 5 ) divide original binary mask into windows based on proximity to the closest coordinate . Using the window analysis , parameters such as mean signal intensity ( from live cell probes or fixed cell staining ) , max signal intensity , neurite width ( total number of pixels divided in a window divided by given spacing ) , puncta per window , and others can be calculated . In most cases , intensity measurements were determined by taking the mean or median of the top 20% of pixels in a given window . Trace alignment for averaging data between independent waves: traces were aligned to the half-maximum value of the front of the actin wave ( oriented towards the growth cone ) . Half maximum values were determined by manual identification of the maximum and minimum actin signals on the front half of the actin curve . Traces were then computationally aligned . EB1 puncta counting: to count EB1 puncta , puncta were manually identified and computationally assigned to their corresponding window . EB1 flow measurements: to calculate flow , the number of EB1 puncta passing through a perpendicular plane of the neurite was counted by eye for 20 s . 2D kymographs: to generate 2D kymographs , window analysis was used with coordinate spacing of 1–3 pixel ( s ) . For each window , the maximum signal from the various sensors was measured . For kymographs in Figure 6a and Figure 8c , the “Multi Kymograph” feature of Fiji was used . Mobile vesicle intensity measurement: to obtain a measure of mobile vesicles , fast image series of 600 m-1 s per image were gathered . Images were filtered using a tophat filter with a disc ( 2–3 pixel radius ) to remove background signal . Negative values were then set to zero and the difference between subsequent images was taken in order to remove stationary particles . The resulting subtracted images from successive timepoints were added together and the summed image was subjected to the window analysis to produce line traces . Single microtubule counting: maximum intensity projection of SIM images were taken and the “Plot Profile” function used to take a line scan perpendicular to the neurite at desired locations in Fiji . Peaks in the line scan were identified by eye and counted as single microtubules . This method likely undercounts microtubules because of the high density of microtubules present in the neurite . SIM images: raw SIM data was processed on the API DeltaVision OMX softWoRx image processing software available in the Stanford Neuroscience Microscopy facility . Generation of maxiumum intensity projections and 3D reconstructions were performed in Fiji . FRET analysis: CFP and FRET images were collected with a 20x ( NA = 0 . 75 ) Nikon objective . Regions of interest were identified and subjected to a flat background subtraction ( background calculated after removal of the object of interest ) , segmentation , and smoothing with a Gaussian filter before calculating FRET/CFP ratios . Custom written Matlab scripts used for 2D line scan analysis , mobile vesicle intensity measurements , and FRET analysis can be found at github . com/MeyerLab/AWinans_Elife_2016 . | Nerve cells ( also known as neurons ) connect with each other to form complex networks through which signals are carried around the body . Signals are received by branch-like projections called dendrites , pass through the cell body and then pass along a long projection called the axon before being transmitted to the dendrites of neighboring neurons . In animal embryos , immature neurons in part of the brain called the hippocampus – which is crucial for learning and forming memories – develop into mature neurons through a series of steps . In the early stages of development , an immature neuron sends out multiple projections that extend out in all directions from its cell body . These projections randomly retract and lengthen for a while before a single projection grows into an axon and the others become dendrites . It is believed that signal proteins inside the neuron that promote the formation of an axon selectively accumulate in a projection as it grows into an axon . These axon-promoting proteins are carried to the axons by a motor protein called kinesin , which moves along fibers called microtubules . In immature neurons , kinesin motors randomly move in and out of different projections , before settling in the projection that will grow into the axon . However , it is not clear what drives these fluctuations . To address this question , Winans et al . used microscopy to study the transport of axon-promoting proteins in hippocampal neurons . The experiments show that a protein called actin forms a mesh of filaments in a wave-like manner , starting in the cell body and moving outwards into the projections . When a wave of actin reaches a projection , the projection grows for a while and then stops until the next actin wave arrives . Furthermore , the actin waves promote the formation of more microtubule filaments . This work shows that actin waves make the projections wider to create space for more microtubules to form , which increases the transport of axon-promoting proteins to the projections . Winans et al . ’s findings suggest that actin waves direct axon-promoting proteins to axons and promote competition between the projections early on by generating random fluctuations that allow all the projections to grow and retract . This would allow each projection to explore its environment in the search for signals that promote axon growth . The next challenge is to understand how different signals select the “winning axon” . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"cell",
"biology"
] | 2016 | Waves of actin and microtubule polymerization drive microtubule-based transport and neurite growth before single axon formation |
The characterization of prostate epithelial hierarchy and lineage heterogeneity is critical to understand its regenerative properties and malignancies . Here , we report that the transcription factor RUNX1 marks a specific subpopulation of proximal luminal cells ( PLCs ) , enriched in the periurethral region of the developing and adult mouse prostate , and distinct from the previously identified NKX3 . 1+ luminal castration-resistant cells . Using scRNA-seq profiling and genetic lineage tracing , we show that RUNX1+ PLCs are unaffected by androgen deprivation , and do not contribute to the regeneration of the distal luminal compartments . Furthermore , we demonstrate that a transcriptionally similar RUNX1+ population emerges at the onset of embryonic prostate specification to populate the proximal region of the ducts . Collectively , our results reveal that RUNX1+ PLCs is an intrinsic castration-resistant and self-sustained lineage that emerges early during prostate development and provide new insights into the lineage relationships of the prostate epithelium .
The prostate is a glandular organ of the mammalian male reproductive system . In mice , prostate development starts during embryogenesis at embryonic day ( E ) 15 . 5–16 . 5 with the emergence of the first prostatic buds from the rostral end of the urogenital sinus ( UGS ) ( Bhatia-Gaur et al . , 1999; Georgas et al . , 2015; Keil et al . , 2012; Toivanen and Shen , 2017 ) . These initial buds grow into the surrounding mesenchyme to develop postnatally and through puberty into a branched ductal network organized in distinct pairs of lobes , known as the anterior prostate ( AP ) , dorsolateral prostate ( DLP ) , and ventral prostate ( VP ) ( Sugimura et al . , 1986a ) . Each lobe has distinct branching patterns , histopathological characteristics , and is thought to contribute differently to the physiological function of the prostate . The differentiated epithelium of the adult prostate gland is mainly composed of basal and luminal cells , interspersed with rare neuroendocrine cells ( Shen and Abate-Shen , 2010; Toivanen and Shen , 2017; Wang et al . , 2001 ) . Luminal cells form a layer of polarized tall columnar cells that depend on androgen signaling and produce the prostatic secretions . Basal cells act as a supportive layer located between the luminal cells and the surrounding stroma . Despite being mostly quiescent under homeostatic conditions , the prostate gland encompasses incredible plasticity . In mice , surgical castration-induced prostate involution has proven an invaluable tool to identify progenitor castration-resistant cell populations , characterized by their ability to survive in the absence of androgens , and to fully regenerate an intact adult prostate after re-administration of testosterone ( Barros-Silva et al . , 2018; Kwon et al . , 2016; McAuley et al . , 2019; Tsujimura et al . , 2002; Wang et al . , 2015; Wang et al . , 2009; Yoo et al . , 2016 ) . Such plasticity has also been shown in defined experimental conditions to stimulate regenerative properties of epithelial subpopulations , including transplantations ( Barros-Silva et al . , 2018; Burger et al . , 2005; Lawson et al . , 2007; Lukacs et al . , 2010; Richardson et al . , 2004; Wang et al . , 2009; Xin et al . , 2005; Yoo et al . , 2016 ) , injury repair ( Centonze et al . , 2020; Horton et al . , 2019; Kwon et al . , 2014; Toivanen et al . , 2016 ) , and organoid assays ( Chua et al . , 2014; Höfner et al . , 2015; Karthaus et al . , 2014 ) . In addition , several studies have proposed that progenitor populations with distinct physiological roles and regenerative capacity reside at different locations within the prostate ( Burger et al . , 2005; Crowell et al . , 2019; Goldstein et al . , 2008; Goto et al . , 2006; Kwon et al . , 2016; Leong et al . , 2008; McNeal , 1981; Tsujimura et al . , 2002 ) . However , the precise cellular hierarchy and how it is established during development remains controversial . RUNX transcription factors ( TF ) are master regulators of lineage commitment and cell fate ( Mevel et al . , 2019 ) . In particular , RUNX1 is essential for the ontogeny of the hematopoietic system and alterations of RUNX1 have been associated with a broad spectrum of hematological malignancies . Interestingly , increasing evidence implicates RUNX1 in the biology and pathology of hormone-associated epithelia ( Lie-A-Ling et al . , 2020; Riggio and Blyth , 2017; Scheitz and Tumbar , 2013 ) , including breast ( Browne et al . , 2015; Chimge et al . , 2016; Ferrari et al . , 2014; van Bragt et al . , 2014 ) , uterine ( Planagumà et al . , 2004; Planagumà et al . , 2006 ) , ovarian ( Keita et al . , 2013 ) , and prostate cancers ( Banach-Petrosky et al . , 2007; Scheitz et al . , 2012; Takayama et al . , 2015 ) . Despite the documented importance of RUNX TFs and reports of RUNX1 in PCa , its expression in the normal prostate gland during development , homeostasis , and regeneration has not been explored . In this study , we found that Runx1 marks a discrete subset of luminal cells located in the proximal region of the prostatic ducts . Using mouse models , combined with lobe-specific single-cell transcriptomic profiling of adult , castrated , and developing prostates , we show that RUNX1+proximal luminal cells represent a distinct lineage established at the onset of prostate development , displaying intrinsic castration-resistant and self-sustaining properties .
We initially sought to characterize the expression pattern of Runx1 in adult mouse prostate . While RUNX1 was detected in basal cells at multiple spatial locations , its expression was specifically high in a subset of luminal cells found in the proximal region of all three lobes , also known as periurethral ( Figure 1A , B; Figure 1—figure supplement 1A , B ) . Sections were co-stained with NKX3 . 1 , a master regulator of prostate development broadly expressed in luminal cells . Using quantitative image-based cytometry ( QBIC ) , we found that RUNX1 and NKX3 . 1 had a largely mutually exclusive expression pattern , with a sharp transition from RUNX1+ NKX3 . 1- to RUNX1- NKX3 . 1+ cells in the proximal region ( Figure 1A , B; Figure 1—figure supplement 1A , B ) . These proximal luminal cells had a unique histological profile , with a compact organization , intense nuclear hematoxylin staining , and increased nuclear-to-cytoplasmic ratio ( Figure 1—figure supplement 1C ) . In contrast , distal luminal cells had a large cytoplasm with intense pink eosin staining , likely reflecting their secretory function . These observations suggest that RUNX1 marks a subset of proximal luminal cells , distinct from the abundant NKX3 . 1+ luminal population lining the rest of the prostate epithelium . The proximal site of the prostate has been proposed to be enriched in cells with stem/progenitor properties ( Goldstein et al . , 2008; Kwon et al . , 2016; Tsujimura et al . , 2002; Yoo et al . , 2016 ) . In order to study the regenerative potential of Runx1-expressing cells ex vivo , we took advantage of isoform-specific fluorescent reporter mouse models of Runx1 ( Draper et al . , 2018; Sroczynska et al . , 2009 ) . Runx1 expression is controlled by two promoters , P1 and P2 , that respectively drive the expression of the Runx1c and the Runx1b isoform ( Mevel et al . , 2019 ) . We found that Runx1 expression in the prostate was exclusively mediated by the proximal P2 promoter , in up to 30% of all epithelial EPCAM+ prostate cells ( Figure 1—figure supplement 2A–C ) . Flow-cytometry profiling confirmed the enrichment of P2-Runx1:RFP in both basal ( EPCAM+ CD49fhigh ) and luminal ( EPCAM+ CD24high ) lineages of the proximal compared to the distal prostate ( Figure 1C , D; Figure 1—figure supplement 2D ) . Mirroring our QBIC spatial analysis ( Figure 1—figure supplement 1B ) , P2-Runx1:RFP was also detected in a large fraction of the VP epithelium ( Figure 1—figure supplement 2D ) . We therefore used the P2-Runx1:RFP mouse line to isolate Runx1 positive ( RFP+ ) and negative ( RFP- ) epithelial cells from the basal and luminal compartments of all three prostate lobes and evaluated their regenerative potential in organoid culture assays ( Drost et al . , 2016; Figure 1E ) . The proximal and distal regions of the AP were analyzed separately . In line with previous reports , basal cells were more efficient at forming organoids compared to all luminal fractions ( Drost et al . , 2016 , Kwon et al . , 2016 ) . Importantly , in the luminal fraction , proximal RFP+ luminal cells of the AP consistently displayed higher Organoid Formation Capacity ( OFC ) than the RFP- fraction ( Figure 1F ) . Luminal RFP+ sorted cells of the DLP and VP also had a greater OFC than RFP- cells ( Figure 1—figure supplement 3A ) . In contrast , no significant differences in OFC were observed between basal enriched subsets and distal luminal RFP+ and RFP- cells . Brightfield assessment revealed that virtually all organoids had a ‘solid’ aspect , except for the predominantly ‘hollow’ organoids derived from proximal RFP+ luminal cells ( Figure 1—figure supplement 3B ) . To further characterize their lineage potential , we classified organoids into three types based on the expression of specific lineage markers: unipotent ‘basal-like’ Keratin 5+ ( K5+ ) , unipotent ‘luminal-like’ Keratin 8+ ( K8+ ) , or multipotent K5+ K8+ ( Figure 1G , H; Figure 1—figure supplement 3C ) . Interestingly , AP proximal luminal RFP+-derived organoids were predominantly small unipotent K8+ , while the remainder fraction mainly gave larger unipotent K5+ organoids ( Figure 1H; Figure 1—figure supplement 3D–E ) . Few multipotent K5+ K8+ organoids were also identified in nearly all populations . Together , our results show that RUNX1 marks a specific subset of proximal luminal cells ( PLCs ) , and that its expression in the prostate is mediated by the P2 promoter . RUNX1+ PLCs have a particular predisposition to form unipotent K8+ hollow organoids , suggesting a lineage bias toward the luminal identity , and highlighting differences within the luminal compartment of proximal and distal regions . In mice , androgen-deprivation can be modeled by surgical castration which leads to prostate regression and enriches for castration-resistant cells ( Toivanen and Shen , 2017; Zhang et al . , 2018 ) . This process is accompanied by the death of luminal androgen-dependent cells and a small proportion of basal cells ( English et al . , 1987; Sugimura et al . , 1986b ) . To track changes in Runx1 expression following androgen withdrawal , we surgically castrated P2-Runx1:RFP mice and harvested tissue ≥4 weeks post-surgery ( Figure 2A ) . While intact prostates contained 22 . 8 ± 6 . 0% RFP+ epithelial cells , their frequency increased to 87 ± 6 . 0% following castration ( Figure 2B , C ) . High RUNX1 levels were no longer restricted to the proximal region , and RFP was detected in virtually all basal cells of the AP , DLP , and VP , as well as more than 75% of the luminal castration-resistant cells ( Figure 2D; Figure 2—figure supplement 1A ) . RUNX1-expressing cells often co-expressed TROP2 ( Figure 2E ) , known to be widely expressed in castrated prostate epithelium ( Goldstein et al . , 2008; Wang et al . , 2007 ) . Several castration-resistant luminal populations have been identified in mice ( Barros-Silva et al . , 2018; Kwon et al . , 2016; McAuley et al . , 2019; Tsujimura et al . , 2002; Wang et al . , 2015; Wang et al . , 2009; Yoo et al . , 2016 ) , including rare castration-resistant Nkx3-1-expressing cells ( CARNs ) . Accordingly , we observed low , but detectable , levels of NKX3 . 1 in some luminal cells , but only occasional RUNX1+ NKX3 . 1+ luminal cells in the distal regions of the castrated prostate ( Figure 2D; Figure 2—figure supplement 1B , C ) . Importantly , the clear transition from RUNX1+ to NKX3 . 1+ cells identified in the proximal luminal layer of intact mice was conserved after castration ( Figure 2D , ii ) . Together , these results show that RUNX1 is expressed in the majority of the castration-resistant cells . The RUNX1+ NKX3 . 1- subset identified in the proximal luminal epithelium of the intact prostate remain NKX3 . 1- following castration , supporting the notion that RUNX1+ PLCs constitute a distinct lineage from distal NKX3 . 1+ cells . To further characterize the RUNX1+ and RUNX1- fractions residing at different anatomical locations of the prostate , we performed droplet-based single cell ( sc ) RNA-seq . We sorted EPCAM+ RFP+ and RFP- cells from individually dissected lobes of intact and castrated prostates isolated from P2-Runx1:RFP reporter mice ( Figure 3A ) . Sorted populations were multiplexed using MULTI-seq lipid-tagged indices to minimize technical confounders such as doublets and batch effects ( McGinnis et al . , 2019b ) . We retrieved a total of 3825 prostate epithelial cells from all sorted populations , with a median of 2846 genes per cell ( see Materials and methods; Figure 3—figure supplements 1 and 2A–G ) . We identified nine in silico computed clusters expressing canonical epithelial , basal , and luminal markers ( Figure 3—figure supplement 2H–J ) . A large population of basal cells was annotated by merging three tightly connected subclusters broadly expressing Krt5 , Krt14 , and Trp63 ( Figure 3B–D; Figure 3—figure supplement 2E , J ) . Luminal populations expressed surprisingly heterogeneous levels of canonical luminal markers such as Cd26/Dpp4 , Cd24a , Krt8 , and Krt18 ( Figure 3—figure supplement 2I ) . We annotated those distinct clusters as Luminal-A ( Lum-A ) , Lum-B , Lum-C , Lum-D , Lum-E , and Lum-F ( Figure 3B ) . Differential gene expression analysis revealed genes strongly associated with each luminal subpopulation ( Figure 3C and D; Figure 3—figure supplement 3A; Supplementary file 2 ) . Initially , we sought to evaluate the effect of androgen withdrawal on lobe-specific cellular heterogeneity . Lum-A/B/C/D were largely enriched in luminal cells originating from intact prostates , whereas Lum-E/F contained mainly castrated luminal cells ( Figure 3E; Figure 3—figure supplement 3B ) . Interestingly , Lum-A/C/F mainly contained VP cells , while Lum-B/D/E had a majority of AP and DLP cells , indicating that the lobular identity of luminal cells in the intact prostate is conserved following castration ( Figure 3F; Figure 3—figure supplement 3C ) . These results suggest that a subset of intact Lum-A/C might undergo partial reprogramming during castration-induced regression and gives rise to the Lum-F cluster . Similarly , surviving Lum-B/D may predominantly reprogram into Lum-E cells upon castration . Alternatively , the small fraction of intact cells observed in Lum-E and Lum-F clusters might give rise to the expanded Lum-E/F clusters upon castration . In contrast to luminal cells , castrated basal cells were minimally affected by androgen-deprivation and clustered together with intact basal cells ( Figure 3E ) . Overall , these results highlight the dramatic changes occurring upon androgen deprivation in the representation of distinct luminal subpopulations . We next specifically focused our attention on RUNX1+ luminal cells . The Lum-D cluster predominantly consisted of AP-derived RFP+ cells , as well as a small number of RFP+ DLP and VP cells ( Figure 3F , H; Figure 3—figure supplements 2E , 3B and C ) . High Runx1 expression in Lum-D correlated with higher levels of Tacstd2/Trop2 , Ly6 family members as well as Runx2 ( Figure 3D , G; Figure 3—figure supplement 3D , E ) . In contrast , Runx1 was barely detected in clusters Lum-B/C which expressed high levels of Nkx3-1 while Lum-A cells expressed low levels of both Runx1 and Nkx3-1 . These results suggest that the Lum-D cluster corresponds to the distinct RUNX1+ luminal cells identified in the proximal region of all three prostate lobes ( Figure 1 ) . To further characterize the specificities of those populations , we performed gene ontology analysis . In line with the secretory role of distal luminal cells , clusters Lum-A/B/C were enriched in enzymatic activity and protein synthesis functions . In contrast , the Lum-D cluster was enriched in terms related to epithelial developmental processes , similar to Lum-E/F ( Figure 3—figure supplement 4A–I ) . This was supported by partition-based graph abstraction ( Wolf et al . , 2019 ) , which uncovered a strong degree of connectivity between the mainly intact Lum-D and castrated Lum-E population ( Figure 3B ) . Additionally , the Lum-D cluster contained a small , but defined , subpopulation of castrated epithelial cells , suggesting the preservation of its identity upon androgen deprivation ( Figure 3E , F ) . In this population , we found very few genes significantly differentially expressed between intact and castrated cells ( n = 103; Supplementary file 3 ) . As expected , androgen-regulated genes including Psca and Tspan1 were downregulated in the castrated subset , while strong contributors of the Lum-D identity such as Tacstd2/Trop2 , Krt4 and Runx1 did not vary ( Figure 3—figure supplement 4J ) . These observations further support the hypothesis that Lum-D/RUNX1+ PLCs maintain their identity following androgen-deprivation . Overall , our single-cell transcriptomic analysis highlighted a vast degree of heterogeneity within and between the luminal compartments of both intact and castrated mouse prostates . The tight transcriptional relationship observed between high Runx1 expressing clusters Lum-D and Lum-E/F suggest that the Lum-D population , which corresponds to PLCs , may contain intrinsically castration-resistant luminal cells . To determine if RUNX1+ PLCs were enriched in castration-resistant cells , we combined prostate regression-regeneration assays with genetic lineage tracing using Runx1mER-CRE-mERRosaLox-Stop-Lox-tdRFP mice ( Luche et al . , 2007; Samokhvalov et al . , 2007 ) , henceforth Runx1CreER Rosa26LSL-RFP ( Figure 4A ) . Using this model , we could genetically label an average of 4 . 70 ± 2 . 8% prostate epithelial Runx1-expressing cells with RFP upon tamoxifen injection ( Figure 4B , C; Figure 4—figure supplement 1A ) . This corresponded to 0 . 54 ± 0 . 2‰ of the total epithelium ( Figure 4E ) . Consistent with the expression pattern of Runx1 , the majority of labeled cells were located in the proximal region of the prostate ( Figure 4C ) , and co-expressed Keratin 4 ( K4 ) ( Figure 4—figure supplement 1D , E ) , previously found enriched in Lum-D cells ( Figure 3D ) . Following surgical castration , we found that the absolute number of RFP+ marked cells remained stable ( Figure 4—figure supplement 1C , D ) . However , the frequency of RFP+ cells in the epithelial compartment increased by ~4 . 3 fold ( Figure 4E; Figure 4—figure supplement 1B ) indicating that Runx1-expressing cells have an enhanced capacity to survive castration compared to Runx1-negative cells . Next , we investigated whether these intrinsically castration-resistant Runx1-expressing cells were involved in epithelial regeneration upon testosterone addback ( Figure 4B , bottom ) . Surprisingly , only 0 . 71 ± 0 . 2‰ RFP+ epithelial cells were found in the regenerated prostate , which was comparable to the intact state ( Figure 4E; Figure 4—figure supplement 1B–D ) . Although the majority of RFP+ clones consisted of single cells , we did observe a minor ~2-fold increase in the frequency of larger clones ( 2–4 cells ) after regeneration , highlighting a modest contribution of RFP labeled cells during prostate regeneration ( Figure 4F , G ) . We found that most RFP marked cells were luminal K8+ in intact , castrated , and regenerated prostates ( Figure 4F , H ) , with only a few basal K5+ RFP+ cells detected in distal areas ( Figure 4F ) . Strikingly , more than 90% of all RFP+ cells remained negative for NKX3 . 1 in all experimental arms ( Figure 4I ) . Thus , these results indicate that RFP+ cells , including PLCs , are mostly unaffected by fluctuations in androgen levels during regression-regeneration assays . RUNX1 expression marks intrinsically castration-resistant luminal cells that do not contribute substantially to the expansion of luminal NKX3 . 1+ cells during prostate regeneration . Given the singular identity of proximal luminal Runx1-expressing cells in the adult prostate , we then asked if this luminal lineage was already emerging during prostate development . At E18 . 5 , once the first prostate buds have emerged , RUNX1 was mainly found in the K8high inner layers of the stratified urogenital epithelium ( UGE ) ( Figure 5A ) . Interestingly , these cells also co-expressed K4 ( Figure 5—figure supplement 1A ) , previously found in the Lum D population ( Figure 3D ) , as well as LY6D , recently shown to mark a subset of adult luminal progenitors ( Barros-Silva et al . , 2018; Figure 5—figure supplement 1B ) . In contrast , RUNX1 expression was low in p63+ and K5+ cells , either lining the outer UGE or found in the tips of premature NKX3 . 1+ prostate buds ( Figure 5A–C ) . At postnatal day 14 ( P14 ) , a prepubescent stage when most of the initial branching events have already occurred ( Sugimura et al . , 1986a; Tika et al . , 2019 ) , RUNX1 was broadly expressed in the proximal region ( Figure 5D ) , mainly in K4+ luminal cells and in some K5+ or p63+ cells ( Figure 5—figure supplement 1C–E ) . Conversely , NKX3 . 1+ cells were found in distal locations , largely distinct from RUNX1+ cells . The specific spatial expression pattern of RUNX1 in proximal luminal cells , largely mutually exclusive with NKX3 . 1 , suggests that these two transcription factors already mark distinct cellular lineages during embryonic prostate organogenesis . To study the dynamic emergence of RUNX1+ cells during prostate development , we utilized an explant culture system ( Berman et al . , 2004; Doles et al . , 2005; Kruithof-de Julio et al . , 2013; Lopes et al . , 1996 ) . Dissected E15 . 5 UGS were cultured for up to 7 days in the presence of dihydrotestosterone ( Figure 5E , F ) . Bud formation was initiated within 2 days of culture ( Figure 5G ) and composed of a double positive K5+ K8+ stratified epithelium , partially diversifying by day 7 ( Figure 5—figure supplement 2A , B ) . On day 0 ( E15 . 5 ) , RUNX1 was detected at the rostral end of the UGE , particularly within the inner layers of the stratified epithelium . After 1 day in culture , NKX3 . 1 expression emerged in RUNX1+ cells located in the outer layers of the UGE , while defined budding was yet to be observed . On day 2 , NKX3 . 1+ prostate buds were evident and had reduced or absent RUNX1 expression . This pattern was conserved in the mature explant , in which distal tips were mainly NKX3 . 1+ , whereas the proximal area remained RUNX1+ ( Figure 5G , H ) , and co-expressed LY6D and K4 ( Figure 5—figure supplement 2C , D ) . Cellular proliferation marked by Ki67 was more substantial in distal regions , suggesting that most of the expansion did not take place in the RUNX1+ compartment ( Figure 5—figure supplement 2E ) . These results suggest that prostate budding originates from a subset of cells located in the outer layers of the stratified UGE , transiently marked by RUNX1 and NKX3 . 1 . During embryonic prostate development , Runx1 expression is already primarily confined to the proximal region of the prostatic ducts , in a distinct compartment from NKX3 . 1+ cells . The characterization by immunostainings of continuous developmental processes is generally constrained to a small number of markers at a time . To further study the specification of RUNX1 and NKX3 . 1 lineages , we performed scRNA-seq on UGS explant cultures collected at successive time points: E15 . 5 ( D0 ) , day 1 ( D1 ) , day 3 ( D3 ) , and day 6 ( D6 ) ( Figure 6A ) . After data processing , 3937 developing prostatic cells were retained , with a median of 3608 genes per cell ( see Materials and methods; Figure 6—figure supplement 1 ) . Visualization of the dataset using a force-directed layout highlighted the progressive cellular diversification taking place from D0 to D6 ( Figure 6B ) . Cellular populations were divided into nine clusters , annotated C0 to C8 ( Figure 6C–E ) . C0/C1 contained the majority of D0 and D1 derived cells , while C2-C8 emerged and expanded at later time points . Due to the primitive nature of the UGE at these time points , the classical basal and luminal lineages were not fully established yet ( Figure 6F; Figure 6—figure supplement 2A–E; Supplementary file 4 ) . Nevertheless , C4-C6 had a more pronounced ‘basal’ identity compared to the other clusters . Krt5/Krt14 marked mainly C4 , and additional basal markers including Trp63 , Dcn , Apoe , or Vcan were higher in C5/C6 . Overall , known regulators of prostate development ( Toivanen and Shen , 2017 ) displayed a variable expression pattern across the different clusters . For example , Foxa1 and Shh were strongly expressed in C0/C1 , Notch1 was higher in C3 , and Sox9 in C7 ( Figure 6—figure supplement 2C ) , highlighting the potential of this dataset to interrogate specific features of prostate development . Consistent with our previous results , Runx1 was highly expressed in clusters having lower Nkx3-1 levels , including C0 , C1 , C2 , and C4 ( Figure 6G ) . To determine how these clusters relate to differentiated prostate lineages , we interrogated population-specific gene signatures previously identified in the adult ( Figure 3 ) . The ‘Basal’ signature was enriched across all clusters , especially in C4/C6 ( Figure 6I; Figure 6—figure supplement 2F , G ) . Strikingly , the ‘Lum-D’ derived signature was highly enriched in C2 compared to all the other adult luminal population signatures , suggesting that the ‘Lum-D’ fate is determined early during prostate development . The singular identity of C2 was characterized by genes previously found highly expressed in the adult ‘Lum-D’ population , including Tacstd2/Trop2 , Krt4 , Psca , as well as Ly6d and Nupr1 ( Figure 6H; Figure 6—figure supplement 2A ) . Collectively , our scRNA-seq analysis show that adult ‘Lum-D’/PLCs share strong similarities with the unique C2 population identified in embryonic explant cultures . This suggests that the distinct proximal luminal lineage is established at the very onset of prostate specification . To trace the fate of RUNX1+ cells during embryonic prostate specification , we cultured UGS explants isolated from the Runx1CreER Rosa26LSL-RFP lineage-tracing model . We performed 2 pulses of tamoxifen treatment on day 0 and 1 of culture and analyzed the explants on day 2 and day 7 ( Figure 7A ) . The majority of the RFP labeled cells were in the most proximal RUNX1+ subset and rarely found in the distal area of the branches , where RUNX1- cells reside ( Figure 7B , C ) . Accordingly , the proportion of RFP+ RUNX1+ cells remained stable between days 2 and 7 ( Figure 7D ) . Also , the fraction of RFP+ cells co-expressing p63 remained unchanged throughout the culture ( Figure 7—figure supplement 1A–C ) , while a small fraction diversified into either K5+ or K8+ cells ( Figure 7—figure supplement 1D , E ) . The scattered RFP+ RUNX1- cells detected in distal branches by day 7 often co-expressed NKX3 . 1 ( Figure 7E , F ) . Overall , this indicates that Runx1-expressing cells only marginally contribute to the expansion of the NKX3 . 1 compartment ( Figure 7G ) . Finally , we wondered whether RUNX1+ cells contributed to the establishment of the proximal luminal lineage . We evaluated the proportion of RFP-labeled cells co-expressing K4 , previously identified as a marker of the developing C2 and adult Lum-D populations ( Figures 3D and 6H ) . Interestingly , the fraction of K4+ RFP labeled cells increased from 56 . 9 ± 10 . 6% to 74 . 1 ± 3 . 0% between day 2 and 7 ( Figure 7F , G ) . There was also an increase of RFP+ cells expressing Nupr1 , another marker of the C2 cluster ( Figure 7—figure supplement 1F–H ) . Taken together , these results show that only a small subset of Runx1-expressing cells contributes to the expansion of NKX3 . 1+ lineage , found in the distal region of the developing prostatic buds . Instead , the majority of Runx1-expressing cells preferentially remain in the proximal region of the premature buds , where the proximal luminal lineage is established .
In this study , we identified RUNX1 as a new marker of a luminal population enriched in the proximal region of the prostatic ducts . By combining scRNA-seq profiling and genetic lineage tracing of Runx1-expressing cells , we show that RUNX1+ PLCs present in the intact prostate constitute a developmentally distinct and intrinsically castration-resistant luminal lineage . We propose that proximal and distal lineages are separate luminal entities from the earliest stages of prostate development . As such , our study provides novel insights into the cellular composition and developmental hierarchy of the mouse prostate epithelium . Until the recent advances in single-cell technologies , the prostate epithelial hierarchy was mainly defined based on anatomical features of the basal and luminal layers , their histological characteristics and the expression of a small subset of markers ( Shen and Abate-Shen , 2010; Toivanen and Shen , 2017 ) . Here , we present two comprehensive scRNA-seq dataset covering both the adult and the developing prostate . To our knowledge , this constitutes the first comprehensive single-cell atlas covering both intact and castrated adult mouse prostates , annotated by their lobe of origin . These datasets can be browsed interactively at http://shiny . cruk . manchester . ac . uk/pscapp/ . In particular , our adult scRNA-seq dataset highlighted an extensive degree of cellular heterogeneity , in particular within the luminal epithelia . Several studies recently made similar observations either focusing on the AP ( Karthaus et al . , 2020 ) , the intact prostate ( Crowley et al . , 2020; Joseph et al . , 2020 ) , or both the intact and castrated prostates ( Guo et al . , 2020 ) . Integration of these multiple datasets will provide a more global view of the transcriptional landscape of the prostate epithelium . Although mainly known as a master regulator of hematopoiesis , RUNX1 is increasingly implicated in hormone-associated epithelia including malignant conditions such as prostate cancer ( Banach-Petrosky et al . , 2007; Lie-A-Ling et al . , 2020; Scheitz et al . , 2012; Takayama et al . , 2015 ) . Here , we identified a subset of RUNX1+ luminal cells located in the proximal region of the developing and adult prostate , referred to as RUNX1+ PLCs , and corresponding to the Lum-D cluster identified in our adult scRNA-seq dataset . Of note , this subset appears to be the equivalent of the ‘L2’ ( Karthaus et al . , 2020 ) or ‘LumP’ ( Crowley et al . , 2020 ) , or ‘Lum-C’ ( Guo et al . , 2020 ) clusters identified in recent studies . In light of the extensive contribution of RUNX transcription factors to developmental processes ( Mevel et al . , 2019 ) , our study suggests that Runx1 , but also Runx2 , may be involved in the development and maintenance of specific subpopulations of the prostate epithelium . Future work should therefore aim at characterizing the functional role played by RUNX factors in the prostate , in particular in PLCs . We demonstrate that these RUNX1+ PLCs exhibit a greater organoid forming potential compared to the remaining luminal fraction , consistent with previous reports isolating similar proximal populations using different markers such as SCA-1 , TROP2 or CD26 ( Crowley et al . , 2020; Goldstein et al . , 2008; Guo et al . , 2020; Karthaus et al . , 2020; Kwon et al . , 2016 ) . Furthermore , RUNX1+ PLCs predominantly formed unipotent K8+ hollow organoids demonstrating their preferential commitment to the luminal fate . The greater clonogenicity of RUNX1+ PLCs may in fact be linked to the gene expression profile of the corresponding Lum-D population , suggesting a more immature epithelial state , committed to the luminal lineage but not the secretory function of the prostate . Similar to the enhanced regenerative potential of glandular basal cells under specific regenerative conditions ( Centonze et al . , 2020 ) , it is tempting to speculate that these cells act as a latent niche of ‘facultative’ luminal stem cells ( Clevers and Watt , 2018 ) , primed to generate a structured prostatic epithelium under defined conditions . Further characterization of RUNX1 expression in prostate development revealed a consistent expression pattern with the adult . RUNX1+ luminal cells were restricted to the most proximal region of the developing prostate buds , both in embryos and UGS explant cultures . Our scRNA-seq of the developing prostate revealed a broad basal identity , supporting the presence of multipotent basal progenitors during embryonic development ( Ousset et al . , 2012; Pignon et al . , 2013 ) , switching to unipotency postnatally ( Tika et al . , 2019 ) . However , we observed a distinct cluster ( C2 ) that strongly resembled the adult Lum-D population , suggesting an early branching event towards the proximal luminal fate at the onset of prostate development . Subsequent lineage-tracing experiments indicated that Runx1-expressing cells preferentially populate the emerging proximal luminal identity . It would be interesting to determine if the adult Lum-A , Lum-B , and Lum-C derive from multipotent-basal progenitors or from any specific clusters identified in the developing prostate . This appears to be the case at least for the adult Lum-D/RUNX1+ PLCs which already emerges during embryonic specification . Our data also sheds a light on the regenerative potential of specific epithelial populations . Basal and luminal lineages have previously been shown to be largely self-sustained using generic basal and luminal Cre drivers ( Choi et al . , 2012; Ousset et al . , 2012 ) . However , whether distinct subpopulations of luminal cells contribute to the regeneration of the others remains poorly understood ( Wang et al . , 2009; Yoo et al . , 2016 ) . Our characterization of RUNX1+ PLCs and the detection of a wide variety of luminal populations in our adult prostate scRNA-seq data highlights the possible existence of multiple self-contained luminal populations . Indeed , Runx1-driven genetic-tracing experiments in regression-regeneration assays revealed that RUNX1+ PLCs did not contribute substantially to the regeneration of distal NKX3 . 1+ cells . It was however evident that RUNX1+ PLCs are intrinsically castration resistant and capable of sustaining their own lineage in the regenerated prostate . Recently , it was proposed that prostate epithelial regeneration is driven by almost all luminal cells persisting in castrated prostates ( Karthaus et al . , 2020 ) . Our results are compatible with this model , but we further demonstrate that not all luminal subsets retain the same in vivo regenerative potential in response to androgen stimulation . Thus , we suggest that the model of self-sustained basal and luminal populations might be extended to individual luminal subpopulations . This hypothesis should be tested in the future using a more specific Lum-D Cre driver ( e . g . Krt4/Psca ) . It will also be of interest to investigate the self-sustenance of other luminal compartments using Lum-A , Lum-B , and Lum-C specific Cre drivers . Finally , our study suggests that the emerging C2/Lum-D population retains a more embryonic-like program , which may relate to their intrinsic castration-resistant potential and have broader relevance to cancer treatment . Along these lines , recent work by Guo and colleagues indicates that Pten loss induced in Psca-expressing cells of the proximal prostate can initiate prostatic intraepithelial neoplasia ( Guo et al . , 2020 ) . These results warrant future investigation of this luminal subset in the context of cancer development , tumor aggressiveness and treatment responses . In conclusion , we characterized the expression pattern of Runx1 in the developing , normal and castrated mouse prostate . We observed that Runx1 marks proximal luminal cells , which is a distinct luminal lineage emerging early during prostate specification , displaying intrinsic castration-resistant and self-sustaining properties . Our results therefore reveal strong intrinsic lineage differences within the luminal compartment of the prostate epithelium .
Animal experiments were approved by the Animal Welfare and Ethics Review Body ( AWERB ) of the Cancer Research UK Manchester Institute and conducted according to the UK Home Office Project Licence ( PPL 70/8580 ) . Genetic lineage-tracing experiments were performed at the Beatson Biological Services Unit ( PPL 70/8645 and P5EE22AEE ) and approved by the University of Glasgow AWERB . Mice were maintained in purpose-built facility in a 12 hr light/dark cycle with continual access to food and water . Immunocompetent wild-type ICR ( CD-1 ) mice were purchased from Envigo . P1-Runx1:GFP and P2-Runx1:RFP have been described previously ( Draper et al . , 2018; Sroczynska et al . , 2009 ) . Colonies were maintained on a ICR ( CD-1 ) background . C57Bl/6J Runx1mER-CRE-mER ( Samokhvalov et al . , 2007 ) were provided by RIKEN ( Japan ) . C57Bl/6J Rosa26lox-stop-lox-tdRFP mice ( Luche et al . , 2007 ) were acquired from the European Mouse Mutant Archive ( EMMA ) . For all transgenic lines , routine genotyping was undertaken at weaning ( 3 weeks of age ) by automated PCR genotyping ( Transnetyx ) . For timed mating experiments , vaginal plug detection was considered as embryonic day ( E ) 0 . 5 . All animal procedures were performed on adult males at least 7 weeks of age . Surgical castration was carried out under aseptic conditions . For prostate regeneration assays , testosterone pellets ( Belma Technologies ) were implanted subcutaneously . For in vivo genetic lineage-tracing experiments , tamoxifen ( Sigma , T5648 ) was resuspended in ethanol and diluted in corn oil at a concentration of 10 mg/mL and administered via intra-peritoneal injections daily for 4 consecutive days using the following regimen: 3 mg , 2 mg , 2 mg , 2 mg . All dissections were performed under a stereo microscope in sterile PBS . Dissociated murine prostate cells were obtained by digesting pre-minced prostate tissue for 1 hr at 37°C in digestive medium prepared in prepared in ADMEM/F12 ( Gibco ) , and containing 1 mg/mL Collagenase Type I ( ThermoFischer Scientific , #17018029 ) , 1 mg/mL Dispase II ( ThermoFischer Scientific , #17105041 ) , 10% Fetal Bovine Serum ( Gibco ) , 1% Penicillin-Streptomycin-Glutamine ( Sigma ) , and 10 μM Y-27632 dyhydrochloride ( Chemdea , #CD0141 ) . For embryonic urogenital sinuses ( UGS ) , dissociation time was reduced to 30 min . Single cells were obtained after an additional 10 min incubation in TrypLE ( Gibco ) at 37°C before mechanical dissociation with a syringe and needle ( 25G ) . Cells were then filtered through a 50-μm cell strainer . Single cell suspensions were kept in Advanced DMEM/F-12 ( Gibco ) containing 5% FBS supplemented with 10 μM Y-27632 . Cells were incubated for 10 min using unconjugated anti-mouse CD16/32 antibody ( Biolegend , C93 , #101301 ) at 4 °C prior to labeling with specific fluorochrome-labeled antibodies . Details of FACS reagents and antibodies are listed in the Key Resources Table . Cells were filtered through a 50 μm cell strainer prior to acquisition . Hoechst 33258 or Sytox blue ( ThermoFischer Scientific ) were used as viability stains . Single-cell suspensions were analyzed on a Fortessa ( BD Biosciences ) and sorts were performed on a FACSAriaIII ( BD Biosciences ) . FACS data were analyzed using FlowJo software ( BD Life Sciences ) . In vitro organoid formation assays were performed as described in Drost et al . , 2016 . Single cells were resuspended in 40 μL drops of phenol red-free Cultrex RGF BME Type 2 ( BME 2 , Amsbio , #3533-005-02 ) , and seeded in CellCarrier-96 Ultra Microplates ( PerkinElmer , #6055302 ) . Defined organoid culture medium was prepared with Advanced DMEM/F-12 ( Gibco ) , supplemented with 10 mM Hepes ( Sigma ) , Gutamax ( Gibco ) , Penicillin/Streptomycin ( Sigma ) , B27 ( Life Technologies , 17504–044 ) , 50 mg/mL EGF ( PeproTech , #AF-100–15 ) , 500 ng/mL R-spondin 1 ( R and D Systems , #4645-RS ) , 100 ng/mL Noggin ( R and D Systems , #6057 NG ) , 10 mM Y-27632 dyhydrochloride ( Chemdea , #CD0141 ) , 200 nM A83-01 ( Tocris Bioscience , #2939 ) , 1 . 25 mM N-Acetylcystein ( Sigma ) , and 1 nM Dihydrotestosterone ( DHT , Sigma #730637 ) . Medium was refreshed every 2–3 days , and organoid cultures were scored after 7 days . UGS explant cultures were performed as described previously ( Kruithof-de Julio et al . , 2013 ) . Briefly , E15 . 5 embryos were obtained from timed matings . Urogenital sinuses ( UGS ) were isolated from the embryos and cultured using a Durapore Membrane Filter 0 . 65 μm ( #DVPP02500 ) placed on a stainless-steel mesh for up to 7 days in Ham’s F-12/DMEM ( Gibco ) supplemented with Insulin-Transferrin-Sodium Selenite Supplement ( Roche ) and 10 μM dihydrotestosterone ( Sigma ) . Media were renewed every 2–3 days . For lineage-tracing experiments , tamoxifen-induced labeling was performed using 0 . 5 μM 4-hydroxytamoxifen ( #T176 , Sigma ) . Prostate tissues were harvested and fixed in 10% buffered formalin for 24 hr . Fixed tissues were processed using standard procedures and embedded in paraffin . Formalin-fixed paraffin-embedded ( FFPE ) sections ( 4 μm ) were cut and dried overnight at 37°C . Multiplexed immunofluorescent stainings of FFPE sections were performed on an automated Leica BOND RX platform using the Opal multiplexing workflow ( PerkinElmer ) . In brief , sections were dewaxed , and rehydrated , and endogenous peroxidase activity was quenched by 10 min pre-treatment with 3% hydrogen peroxide diluted in TBS-T ( Tris-Buffered Saline 0 . 05% Tween-20 ) . Following on-board heat-induced epitope retrieval with citrate buffer ( pH 6 . 0 ) for 20 min , sections were incubated for 10 min with 10% Casein ( Vector Laboratories ) diluted in TBS-T . Each staining cycle included a primary antibody incubation for 30 min , followed by buffer washes , and 30 min incubation with HRP -labeled secondary antibodies ( Key Resources Table ) . After further washes , the Tyramide labeled with a fluorophore ( Opal 520 , Opal 570 or Opal 650 , PerkinElmer ) was added for a final 10 min . Subsequent antibody stainings were performed by repeating the same procedure , separated by heat-mediated antibody denaturation using citrate buffer ( pH 6 . 0 ) for 5 min at 95°C . Nuclei were counterstained with DAPI ( Sigma ) and slides were sealed using ProLong Gold Antifade Mountant ( ThermoFischer Scientific ) . In situ hybridization ( ISH ) to detect Nupr1 ( ACD , LS 2 . 5 Mm-Nupr1 #434818 ) was done using the Multiplex Fluorescent detection kit ( ACD ) on the automated Leica BOND RX platform following the manufacturer’s instructions . Pre-treatment was done using an EDTA based pH 9 . 0 epitope retrieval solution for 15 min at 88°C followed by 10 min protease incubation . After ISH , antibody staining was carried out using an anti-RFP antibody for 1 hr detected with EnVision HRP anti-rabbit secondary ( Agilent ) followed by incubation with Tyramide-conjugated Opal 570 ( PerkinElmer ) as described above . Anti-CDH1 antibody was applied for 1 hr and detected using an anti-goat Alexa Fluor 647 secondary antibody ( ThermoFischer Scientific , #A-21447 ) . Staining of frozen sections was performed as described previously ( Thambyrajah et al . , 2016 ) . The list of antibodies used is available in the Key Resources Table . Whole-slide images were acquired on an Olympus VS120 slide scanner . Images were analyzed using QuPath v0 . 2 ( Bankhead et al . , 2017 ) . Briefly , annotations were drawn manually to select areas of interest . Nuclear detection was achieved using the ‘cell detection’ module on the DAPI channel . A classifier was then trained for each batch of images using the random forest algorithm , to detect the epithelial layers based on either CDH1 or K5/K8 stainings . Single-cell intensity measurements were analyzed using R ( 3 . 6 . 3 ) . For Quantitative Imaged-Based Cytometry ( QBIC ) , single-cell intensity measurements were log10 transformed and plotted using the ‘geom_hex’ function of the ggplot2 R package . QuPath was used to extract representative high-quality raw images of selected areas from whole slide images using the ‘Send region to ImageJ’ tool . Images used for publication were processed with ImageJ ( NIH Image , Maryland , USA ) . Confocal images were acquired using a Leica TCS SP8 confocal microscope and LAS X Leica software . Images of whole UGS explant culture were taken using a Leica MZ FLIII microscope . Whole-mount staining was adapted from Yokomizo et al . , 2012 . Organoids were fixed directly in 96-well plates using 4% paraformaldehyde for 1 hr at 4°C . After three washes of 5 min in PBS , organoids were incubated in PBS-BST , containing PBS , 1% milk , 1% BSA , 10% goat serum ( Agilent , #X090710 ) , 0 . 4% Triton X-100 . Pre-conjugated primary antibodies , K5 Alexa Fluor 647 ( #ab193895 , Abcam ) and K8 Alexa Fluor 488 ( #ab192467 , Abcam ) were diluted at 1/400 in PBS-BST and incubated with the organoids overnight at 4°C on a rocking platform . After three washes of 1 hr in PBS-BST at 4°C , organoids were stained with DAPI at 2 μg/mL diluted in PBS-BST and incubated for another 30 min at 4°C on a rocking platform . Images were acquired on an Opera Phenix High Content Screening System using the 10x air and 20x water lenses . Quantitative analysis was performed using the Harmony software on maximum projection images . A detailed description of the samples , replicates , and the corresponding cellular populations used for each sequencing run is provided in Supplementary file 1 . For the adult mouse prostate dataset , AP , DLP , and VP lobes were micro dissected and pooled from P2-Runx1:RFP reporter mice after dissociation . Single live EPCAM+ cells from RFP+ and RFP- fractions of each lobes were sorted separately ( containing a mix of CD49fhigh basal and CD24high luminal cells ) . For the UGS explant culture dataset , the middle regions of the explants were micro dissected to enrich for prostatic branching events and pooled by time point after dissociation . Single live EPCAM+ cells were sorted for each independent time point . Individually sorted populations were multiplexed using the MULTI-seq protocol ( McGinnis et al . , 2019b ) . Reagents were kindly provided by Dr . Zev Gartner . In brief , after sorting , cells were washed once in cold serum- and BSA-free PBS . A lipid-modified DNA oligonucleotide and a specific sample barcode oligonucleotide were then mixed and added to the cells at a final concentration of 200 nM each , and incubated in cold PBS for 5 min . Each individual sample to be multiplexed received an independent sample barcode . Next , a common lipid-modified co-anchor was added at 200 nM to each sample to stabilize the membrane bound barcodes . After an additional 5 min incubation on ice , cells were washed two times with PBS containing 1% FBS 1% BSA in order to quench unbound barcodes . Samples were then pooled together and washed once with PBS 1% FBS 1% BSA . After cell counting , cells were loaded in a Chromium Single Cell 3ʹ GEM Library and Gel Bead Kit v3 ( 10x Genomics ) . scRNA-seq library preparation , sequencing and pre-processing . Gene expression ( cDNA ) libraries were prepared according to the manufacturer’s protocol . MULTI-seq barcode libraries were separated from the cDNA libraries during the first round of size selection , and PCR amplified prior to sequencing according to the MULTI-seq library preparation protocol ( McGinnis et al . , 2019b ) . For the adult mouse prostate dataset , cDNA libraries of ‘run 1’ and ‘run 2’ were sequenced on Illumina NovaSeq 6000 System , and ‘run 3’ was sequenced on Illumina HiSeq 2500 . The UGS mouse prostate explant run was also sequenced on Illumina HiSeq 2500 . Sequencing data of cDNA libraries were processed using Cellranger v3 . 1 . 0 and mapped onto mm10 mouse reference genome . Pre-processing of the MULTI-seq library fastq files was performed using the ‘deMULTIplex’ ( v1 . 0 . 2 ) R package ( https://github . com/chris-mcginnis-ucsf/MULTI-seq ) to generate a sample barcode UMI count matrix . Detailed quality control metrics of each sequencing run are provided in Supplementary file 1 . A similar strategy was applied for the analysis of the UGS explant culture dataset , with some alterations described below . Raw sequencing files and processed gene expression matrices have been deposited in the NCBI Gene Expression Omnibus under the accession number GSE151944 . The processed datasets for both mouse adult prostate and UGS prostate explant cultures can be accessed via a searchable R Shiny application available at http://shiny . cruk . manchester . ac . uk/pscapp/ . All codes used to process data and generate figures are available on a public GitHub repository at https://github . com/glacaud/prostate-scRNAseq ( Mevel , 2020 copy archived at swh:1:dir:c8a38de85e999a595715a4e0a41585fd6b94c44f ) . Statistical analyses were performed using Graphpad/Prism ( v8 . 4 . 2 ) . Data are represented as mean ± SD . Unless otherwise specified in the corresponding figure legend , two-tailed unpaired t-tests were used to compare means between two groups . Statistical significance was set at p<0 . 05 . For animal model studies , no statistical method was used to pre-determine the sample size . No randomization or blinding was used for in vivo studies . | The prostate is part of the reproductive organs in male mammals . Many of the cells lining the inside of the prostate – known as ‘luminal cells’ – need hormones to survive . Certain treatments for prostate cancer , including surgical and chemical castration , lead to fewer hormones reaching the prostate , which shrinks as luminal cells die . But some of these luminal cells are able to survive the damaging effects of castration , rebuilding the prostate upon treatment with hormones , which can lead to the cancer reappearing . It is unclear which type of luminal cells survive during periods without hormones and are responsible for regenerating the prostate . RUNX1 is a protein responsible for switching genes on and off , and is usually found in blood cells , which it helps to mature and perform their roles , but has also been detected in tissues that depend on hormones . Since the luminal cells of the prostate rely on hormones , could RUNX1 also be present in these cells ? To answer this question , Mével et al . used mice to determine where and when RUNX1 is found in prostate cells . Mével et al . detected high levels of RUNX1 in a patch of luminal cells at the base of the prostate . Samples of these cells were taken for further testing from developing mouse embryos , healthy adult mice and mice in which the prostate was regenerating after surgical castration . Mével et al . found that these cells were a distinct subtype of luminal cells that were able to resist the effects of castration – they survived without hormones . Though these cells were present during the early stages of prostate embryonic development and in healthy adult prostate tissue , they were not responsible for rebuilding the prostate after castration . Mével et al . ’s results indicate that , in mice , RUNX1 may act as a marker for a subset of luminal cells that can survive after castration . Further probing the roles of these castration-resistant luminal cells in normal and cancerous prostate tissue may improve the outcome of patients with prostate cancer treated with hormone deprivation therapy . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"stem",
"cells",
"and",
"regenerative",
"medicine",
"developmental",
"biology"
] | 2020 | RUNX1 marks a luminal castration-resistant lineage established at the onset of prostate development |
Mitochondrial dysfunction is associated with a spectrum of human disorders , ranging from rare , inborn errors of metabolism to common , age-associated diseases such as neurodegeneration . How these lesions give rise to diverse pathology is not well understood , partly because their proximal consequences have not been well-studied in mammalian cells . Here we provide two lines of evidence that mitochondrial respiratory chain dysfunction leads to alterations in one-carbon metabolism pathways . First , using hypothesis-generating metabolic , proteomic , and transcriptional profiling , followed by confirmatory experiments , we report that mitochondrial DNA depletion leads to an ATF4-mediated increase in serine biosynthesis and transsulfuration . Second , we show that lesioning the respiratory chain impairs mitochondrial production of formate from serine , and that in some cells , respiratory chain inhibition leads to growth defects upon serine withdrawal that are rescuable with purine or formate supplementation . Our work underscores the connection between the respiratory chain and one-carbon metabolism with implications for understanding mitochondrial pathogenesis .
Damaged mitochondrial respiratory chains play a key role in the pathogenesis of rare congenital metabolic disorders , as well as in a number of age-associated disorders such as diabetes ( Szendroedi et al . , 2012 ) , neurodegenerative disease ( Schapira et al . , 1989 ) , and aging . Mitochondrial respiratory chain components are encoded by two genomes; respiratory chain proteins encoded by mitochondrial DNA ( mtDNA ) are expressed in all tissues , yet inherited lesions within mtDNA can lead to varying tissue pathology ( Koopman et al . , 2012; Vafai and Mootha , 2012 ) , suggesting a complex interplay between the primary respiratory chain dysfunction and the compensatory adaptations to that dysfunction . Improved understanding of cellular responses to respiratory chain dysfunction promises to deepen our understanding of mitochondrial disease pathogenesis , nominate new biomarkers ( Suomalainen et al . , 2011 ) , and motivate new therapeutic strategies ( Zhang et al . , 2013 ) . Cellular responses to respiratory chain dysfunction , collectively known as the mitochondrial retrograde response , have been studied in a number of organisms ( Haynes et al . , 2013; Liu and Butow , 2006 ) . In yeast , early application of genomic profiling and genetic studies identified a transcriptional program ( Liu and Butow , 2006 ) that senses respiratory chain dysfunction , rewires metabolism to bypass a congested tricarboxylic acid ( TCA ) cycle , and promotes cellular survival . In worms , errors in mitochondrial biosynthesis trigger the mitochondrial unfolded protein response ( UPRmt ) to activate transcription of mitochondrial proteases and chaperones , proteins that reduce oxidative stress , and glycolytic enzymes ( Nargund et al . , 2012 ) . In flies , respiratory chain dysfunction signals to the nucleus and cytosol via reactive oxygen species ( ROS ) and altered energetics to JNK and AMPK , respectively , to place a checkpoint on cell division ( Owusu-Ansah et al . , 2008 ) . Induction of a mitochondrial retrograde response in flies has recently been shown to suppress age-dependent degradation of mitochondria ( Owusu-Ansah et al . , 2013 ) . Studies of mammalian cells have identified a small collection of cellular responses to respiratory chain dysfunction . Classic studies showed that loss of mtDNA gives rise to uridine and pyruvate dependency ( King and Attardi , 1989 ) . Altered mitochondrial calcium uptake stemming from collapse of the mitochondrial membrane potential triggers metabolic alterations ( Amuthan et al . , 2001 ) . Loss of the mtDNA induces expression of mitochondrial chaperones ( Martinus et al . , 1996 ) . Recently , a mouse model of mitochondrial respiratory chain disease has been shown to mount an ATF4-mediated , starvation-like transcriptional response ( Tyynismaa et al . , 2010 ) , though the functional relevance of this response remains unclear . At present , a systematic picture of how mammalian cells respond to respiratory chain dysfunction at the levels of both gene expression and metabolism is lacking; such a picture could help to unify the above observations , while also pointing to novel responsive and adaptive pathways . We modeled mitochondrial respiratory chain dysfunction in human HEK-293 cells by depleting them of mtDNA . In human cells , mtDNA is required for expression of 13 structural subunits of the oxidative phosphorylation ( OXPHOS ) complexes that make up the respiratory chain; human patients with mtDNA depletion exhibit severe multicomplex respiratory chain deficiency ( Shoffner , 2005 ) . To generate hypotheses about pathways remodeling in respons to mitochondrial dysfunction , we systematically characterized the response of HEK-293 cells to mtDNA depletion using complementary metabolite , RNA , and protein profiling . Integrated analysis of the profiles raised the hypothesis that one carbon metabolism , especially serine biosynthesis and transsulfuration , remodels in response to mitochondrial dysfunction . We confirmed this hypothesis with mechanistic follow-up experiments . We then explored alterations in serine metabolism in greater detail , and discovered that cells experience impairment of mitochondrial one-carbon synthesis upon respiratory chain dysfunction . In some cell lines , this leads to dependence on exogenously provided serine in the presence of mitochondrial dysfunction . Thus , our study identifies a new metabolic vulnerability of cells facing mitochondrial stress , with implications for our understanding and treatment of human mitochondrial disorders .
We stably transfected T-REx-293 cells , a HEK-293-derived cell line expressing the tet repressor , with a plasmid that expresses a dominant-negative mutant of DNA polymerase gamma ( POLGdn ) ( Jazayeri et al . , 2003 ) under tet repressor control . Doxycycline-triggered expression of POLGdn halts replication of mtDNA , which is then diluted as cells continue to divide ( Figure 1a ) . mtDNA-encoded OXPHOS complex components become depleted ( Figure 1a ) ( Jazayeri et al . , 2003 ) , cellular respiratory capacity becomes compromised ( Figure 1—figure supplement 2 ) , and after several days , cell growth also slows . Removal of doxycycline allows recovery of mtDNA content and cell growth ( Figure 1—figure supplement 1 ) . Some of the recovery could be due to selection of cells that lose tet-inducible POLGdn expression ( see Materials and methods ) , so we repeated these experiments using 100 ng/ml ethidium bromide treatment . Ethidium bromide directly inhibits mitochondrial DNA replication , and gave similar results as POLGdn expression ( Figure 1—figure supplement 1 ) . 10 . 7554/eLife . 10575 . 003Figure 1 . Integrated RNA , protein , and metabolite profiling of mtDNA depletion . ( a ) Experimental model . Doxycycline treatment induces PolGdn expression , mtDNA depletion , and reduction in oxidative phosphorylation complexes ( n = 1 for all data ) . Arrowheads indicate time points analyzed by metabolite , proteomic , and transcriptional profiling . ( b ) Results from hypothesis-generating integrated profiling ( n = 2 for metabolites; n = 1 for transcripts and proteins ) showing serine- and homocysteine-related measurements . Numbers in parentheses represent ranks of the respective measurements . ( c ) Serine biosynthesis and transsulfuration pathways . DOI: http://dx . doi . org/10 . 7554/eLife . 10575 . 00310 . 7554/eLife . 10575 . 004Figure 1—figure supplement 1 . mtDNA depletion time courses using PolGdn overexpression or EtBr treatment . ( a ) Relative mtDNA copy number . ( b ) Cell growth . n = 1 for all data . DOI: http://dx . doi . org/10 . 7554/eLife . 10575 . 00410 . 7554/eLife . 10575 . 005Figure 1—figure supplement 2 . Changes in oxygen consumption induced by mtDNA depletion . ( a ) oxygen consumption time courses showing basal and uncoupler-stimulated respiration . ( b ) changes in basal respiration rates with mtDNA depletion . n = 5 for each trace . DOI: http://dx . doi . org/10 . 7554/eLife . 10575 . 00510 . 7554/eLife . 10575 . 006Figure 1—figure supplement 3 . Metabolite profiling results . Spent media ( a ) and cell extract ( b ) metabolites with nominally significant changes after 2 or 3 d of mtDNA depletion , as compared to untreated cell samples ( n = 2 , nominal p<0 . 05 ) . Only metabolites detected in all three time points are shown . Metabolites are ordered by mean fold change with 3 d of mtDNA depletion , compared to untreated cell samples . DOI: http://dx . doi . org/10 . 7554/eLife . 10575 . 00610 . 7554/eLife . 10575 . 007Figure 1—figure supplement 4 . Spent media metabolite levels shown relative to levels in base media . Only metabolites present in the base media and significant changed with 2 or 3 d mtDNA depletion are shown . n = 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 10575 . 00710 . 7554/eLife . 10575 . 008Figure 1—figure supplement 5 . Transcriptional profiling results . ( a ) Heat map showing genes changed ( see Materials and methods ) during the course of mtDNA depletion and repletion . Time course of mtDNA copy number is shown below the heat map for comparison . n = 1 for each time point . ( b ) Most strongly upregulated and downregulated probesets in the microarray data . DOI: http://dx . doi . org/10 . 7554/eLife . 10575 . 00810 . 7554/eLife . 10575 . 009Figure 1—figure supplement 6 . Protein profiling results ( n = 1 ) at 2 d ( yellow ) and 5 d ( red ) of mtDNA depletion . The leftmost box shows fold changes for all proteins quantitated by mass spectrometry . Subsequent boxes show subsets of these data . Starred data points are mtDNA-encoded proteins . DOI: http://dx . doi . org/10 . 7554/eLife . 10575 . 009 To generate hypotheses as to what biochemical changes arise from POLGdn-induced mtDNA depletion , we performed initial studies using three profiling modalities . We performed metabolite profiling using targeted mass spectrometry ( control and days 2 and 3; Figure 1—figure supplement 3; Figure 1—figure supplement 4; and Supplementary file 1 ) , transcriptional profiling using microarrays ( selected days between 1 and 25 , with two untreated controls; Figure 1—figure supplement 5 and Supplementary file 2 ) , and protein profiling using mass spectrometry ( control and days 2 and 5; Figure 1—figure supplement 6 and Supplementary file 3 ) . We stress that these experiments were performed with limited numbers of replicates: one transcriptional profiling replicate for each of 18 time points; one protein profiling replicate for each of 3 time points; and two metabolite profiling replicates for each of 3 time points and 2 different sample types . The paucity of replicates prevents us from drawing definitive conclusions from the individual profiles . Any hypotheses posed by analyzing these profiles must be considered preliminary and subject to confirmation and validation . When jointly analyzed , the profiling methodologies suggest increases in transsulfuration and serine biosynthesis ( Figure 1b and Figure 1c ) . Serine itself is the most strongly increased metabolite in all of our analyses , and the serine biosynthesis enzymes are among the most strongly upregulated proteins in our proteomic data . Transcriptional profiling reveals elevation of the entire serine biosynthetic pathway and two high-affinity human serine transporters ( SLC1A4 and SLC1A5; Supplementary Table 2a ) . While our metabolite profiling did not measure cysteine , we did find that the precursor for transsulfuration , homocysteine , was the most strongly decreased metabolite in all of our analysis . Further , we observed increased taurine , a known product of cysteine breakdown , and increased α-hydroxybutyrate , a byproduct of transsulfuration , in metabolite profiling of spent media ( Figure 1—figure supplement 3 ) . Finally , we observed transcriptional activation of transsulfuration genes as well as SLC7A11 , the dominant cystine transporter of the plasma membrane . We next performed experiments to confirm the hypothesis that serine metabolism and transsulfuration are altered upon mtDNA depletion , and to understand the molecular basis of these alterations . To discover cis-regulatory motifs and factors that might be responsible for our observed transcriptional changes , we performed motifADE analysis ( Mootha et al . , 2004 ) on our transcriptional profiles . motifADE scores evolutionarily conserved cis-motifs for their relative enrichment within the vicinity of the transcription start sites of genes that are differentially expressed . Among all possible 6 , 7 , 8-mer , and gapped 9-mer motifs ( Figure 2a , Supplementary file 2 ) , the highest-scoring motif was the 8-mer 5’-TGATGCAA-3’ ( p ≈ 1 . 8×104 , Mann-Whitney U test , adjusted for number of motifs tested ) , which is strikingly similar to consensus ATF4-C/EBP binding site TGATGHAAH ( Kilberg et al . , 2009 ) . Indeed 26 of the 50 genes ( 52% ) most upregulated in response to mtDNA depletion contain this consensus ATF4-C/EBP binding site near their transcription start sites ( Supplementary file 2 ) , representing a strong enrichment over the 14% seen across the genome ( p<10–9 , binomial test ) . 10 . 7554/eLife . 10575 . 010Figure 2 . mtDNA depletion activates ATF4 . ( a ) Volcano plot of motifADE analysis results ( see Results ) . Δ median denotes the normalized rank of the median gene associated with each motif . ( b ) Fold changes of four of the most-upregulated genes in microarray data , in response to mtDNA depletion by ethidium bromide ( EtBr ) treatment ( 100 ng/ml , 9 d ) , and with expression of either GFP or GADD34ΔN . ( c ) Activation of serine and homocysteine biosynthesis genes , compared to that of genes at the 1st , 10th , 90th , and 99th percentiles in each transcriptional profiling timepoint . ( d ) Activation of serine and cysteine biosynthesis genes in response to mtDNA depletion , with and without GADD34ΔN expression . ( e ) Activation of ATF4 target genes in response to mitochondrial inhibitors . ( f ) Activation of ATF4 target genes in response to cytoplasmic redox imbanace elicited by lactate . n = 3 for b , d , e , and f . DOI: http://dx . doi . org/10 . 7554/eLife . 10575 . 01010 . 7554/eLife . 10575 . 011Figure 2—figure supplement 1 . Doxycycline control treatment data . ( a ) Activation of ATF4 target genes in response to PolGdn expression , and in response to 6 d doxycycline treatment ( 1 μg/ml ) in the absence of PolGdn induction in the parental T-REX-293 cell line . Left side , ATF4-responsive genes listed in ( Kilberg et al . , 2009 ) . Right , ATF4-responsive genes identified in this paper . For all genes except PSAT1 , the activation in the absence of PolGdn induction was less than 10% of the activation with PolGdn induction . ( b ) motifADE results from doxycycline control treatment data . n = 1 for data with PolGdn expression , and n = 3 for data without . DOI: http://dx . doi . org/10 . 7554/eLife . 10575 . 011 We confirmed ATF4 involvement in mtDNA depletion-elicited transcriptional changes by overexpressing an N terminally truncated GADD34 protein ( GADD34△N ) , which inhibits ATF4 activation ( Novoa et al . , 2001 ) , in the parent T-REx-293 cell line . We tested four genes that were among the most highly up-regulated in our microarray dataset . We confirmed that these genes were also highly upregulated upon mtDNA depletion with EtBr , and that GADD34△N overexpression inhibits this upregulation ( Figure 2b ) . We performed a similar analysis of serine and cysteine biosynthesis genes . In the preliminary transcriptional profiling dataset , five of these genes were upregulated in nearly all the early mtDNA depletion time points ( Figure 2c ) . GADD34△N overexpression blunted activation of all five genes in response to EtBr-induced mtDNA depletion ( Figure 2d ) . ATF4 has previously been shown to activate expression of serine biosynthesis genes ( Ye et al . , 2012 ) and transsulfuration genes ( Dickhout et al . , 2012 ) . Furthermore , chromatin immunoprecipitation experiments have shown direct ATF4 protein binding to the promoter regions of PHGDH , PSAT1 , and CTH ( Han et al . , 2013 ) . We used a series of mitochondrial inhibitors to determine what bioenergetic parameters might be upstream of ATF4 activation ( Figure 2e ) . Inhibition of OXPHOS complexes III and V using antimycin and oligomycin , respectively , gives the strongest activation of genes shown above to be ATF4-responsive . Inhibition of complex I using rotenone gives subtle activation . Membrane potential dissipation using the uncoupler carbonyl cyanide m-chlorophenyl hydrazone ( CCCP ) fails to elicit activation of these genes , and indeed partially reversed ATF4 activation from mtDNA depletion . These results suggest that mtDNA depletion-triggered activation of ATF4 arises from redox stress due to a stalled respiratory chain , and not from defects in ATP synthesis or membrane depolarization . To further investigate redox effects on ATF4 activation , we manipulated NAD+/NADH in cells by growing them in media with either 1 mM pyruvate or 1 mM lactate . Pyruvate and lactate are known to freely cross the plasma membrane , and altering the extracellular pyruvate:lactate ratio allows control over the cytoplasmic NAD+/NADH ratio through the action of lactate dehydrogenase ( Bücher et al . , 1972; Williamson et al . , 1967 ) . Cells grown in media for 24 hrs with 1 mM lactate showed significantly higher expression of ATF4 target genes than with 1 mM pyruvate ( Figure 2f ) . However , the magnitude of the changes is modest compared to those seen with either antimycin treatment or with mtDNA depletion . Therefore , while altered cellular redox balance appears to contribute to ATF4 activation , other factors , such as oxidative stress from stalling of the respiratory chain , may also be important when mtDNA is depleted . Because doxycycline can itself inhibit mitochondrial translation ( Ugalde et al . , 2004; Wang et al . , 2015 ) , we wondered whether direct doxycycline toxicity , and not doxycycline-induced POLGdn , was responsible for the transcriptional effects we observe in our profiling experiments . We used microarrays to measure the effect of 6 d doxycycline treatment on T-REx-293 cells that did not express POLGdn . ATF4 target genes showed much lower activation in doxycycline-treated T-REx-293 cells than in cells with POLGdn induction ( Figure 2—figure supplement 1a ) . When we performed motifADE on the doxycycline control data and considered Bonferroni corrected p-values , the ATF4 motif TTGCATCA was not significant ( Figure 2—figure supplement 1b ) . Transsulfuration is important for generating H2S , an emerging gaseous signaling molecule important for many aspects of mammalian physiology ( Kabil et al . , 2014 ) . Given our transcriptional and metabolic profiles suggest an activation of transsulfuration , we wondered if mtDNA depletion could also increase H2S production . To confirm that mtDNA depletion leads to decreased homocysteine abundance , we performed focused measurements of homocysteine levels in T-Rex-293 cells depleted of mtDNA using ethidium bromide ( Figure 3a ) . Next , we used two different methods to detect H2S . First , we reacted cell extracts with monobromobimane ( MBB ) and detected H2S adducts using HPLC ( Tokuda et al . , 2012 ) . Second , we detected sulfane sulfur species in cell extracts using SSP4 , a fluorescent probe ( Marutani et al . , 2014 ) . H2S degradation requires a functioning respiratory chain ( Hildebrandt and Grieshaber , 2008 ) , so we used acute antimycin A treatment as a control for changes in H2S degradation rate . Both methods revealed increased H2S accumulation as a consequence of mtDNA depletion , greater than the amount induced by acute treatment with antimycin A ( Figure 3b ) . These data support the notion that transcriptional changes elicited by mtDNA depletion have the effect of increasing cellular H2S production . 10 . 7554/eLife . 10575 . 012Figure 3 . Alterations in transsulfuration-associated metabolites upon mtDNA depletion . ( a ) Confirmation of decreased homocysteine abundance in spent media with EtBr-induced mtDNA depletion . n = 3 . ( b ) Hydrogen sulfide levels in cells measured either directly ( sulfide ) or indirectly by its sulfane products ( SSP4 ) . Acute antimycin treatment ( 1 hr ) was used to control for increased H2S levels arising from decreased H2S degradation due to loss of the respiratory chain . n = 6 for both plots . DOI: http://dx . doi . org/10 . 7554/eLife . 10575 . 012 We confirmed alterations in serine using EtBr-treated T-Rex-293 cells , ruling out possible side effects due to doxycycline toxicity as well as statistical anomalies arising from the small sample size used in the initial metabolite profiling ( Figure 4a ) . 10 . 7554/eLife . 10575 . 013Figure 4 . Respiratory chain dysfunction impairs mitochondrial 1C metabolism . ( a ) Confirmation of altered serine levels in spent media with EtBr-induced mtDNA depletion . n = 3 . ( b ) Tracing serine metabolism using 13C-labeled glucose . Serine synthesis rates are inferred by labeling cells for 30 min with 13C6-glucose and measuring the amount of 13C3-serine that emerges . Simultaneously , serine consumption rates can be inferred by the amount of unlabeled serine that disappears during the labeling . n = 3 . ( c ) Testing impairment of mitochondrial 1C metabolism using serine isotope scrambing in SHMT1 KO cells . Impairments in mitochondrial 1C metabolism downstream of 13CH2-THF are reflected in increased generation of 13C1-serine ( see Figure 4—figure supplement 1 ) . n = 2 . ( d ) Testing impairment of mitochondrial 1C metabolism by assaying formate production from serine using isolated mitochondria with acute RC inhibition . n = 3 . DOI: http://dx . doi . org/10 . 7554/eLife . 10575 . 01310 . 7554/eLife . 10575 . 014Figure 4—figure supplement 1 . Rationale of serine isotope scrambling assay . Fully labeled serine ( left ) is broken down to glycine and methylene-THF by SHMT2 . The labeled methylene-THF has two fates: oxidation by MTHFD2 to formyl-THF and recombination with unlabeled glycine , which is present in excess , to form singly labeled serine . Decreased MTHFD2 turnover is expected to give rise to increased synthesis of singly labeled serine . DOI: http://dx . doi . org/10 . 7554/eLife . 10575 . 01410 . 7554/eLife . 10575 . 015Figure 4—figure supplement 2 . Alterations in cellular NAD+/NADH ratio elicited by mitochondrial manipulations . Cellular NAD+ and NADH were determined using a commercially available assay kit ( Sullivan et al . , 2015 ) . n = 3 . DOI: http://dx . doi . org/10 . 7554/eLife . 10575 . 01510 . 7554/eLife . 10575 . 016Figure 4—figure supplement 3 . Serine isotope scrambling in other cell types . * , different from 0 day EtBr with p<0 . 05 ( n = 3 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 10575 . 01610 . 7554/eLife . 10575 . 017Figure 4—figure supplement 4 . Determination of homocysteine remethylation . Deuterium label was traced from labeled formate to labeled methionine , to determine the amount of homocysteine that is consumed by remethylation . n = 3 . DOI: http://dx . doi . org/10 . 7554/eLife . 10575 . 017 In the initial metabolite profiling data , serine was the most strongly consumed metabolite , by fold change compared to base media , in untreated cells ( Figure 1—figure supplement 4 ) . Therefore , increased endpoint serine in the spent media atop mtDNA-depleted cells could reflect either increased synthesis or decreased consumption . To determine the contribution of serine biosynthesis to our observed increase in serine abundance , we performed isotopic tracer analysis using 13C labeled glucose ( Chaneton et al . , 2012 ) . To avoid product inhibition of serine biosynthesis ( Fell and Snell , 1988 ) , we used a low ( 50 μM ) concentration of unlabeled serine in the labeled glucose media . The analysis ( Figure 4b ) shows that in response to mtDNA depletion , cells both produce more serine from glucose and take up less serine from the media . We estimate that the cell volume is at least 1000 fold smaller than the media volume in our experiments . Therefore , decreased serine uptake implies decreased overall serine consumption and not simply decreased serine accumulation in the cytoplasm . To explain why serine consumption decreases , we considered the metabolic roles of serine in mammalian cells . In addition to its role in protein synthesis , serine is also a precursor for phospholipid biosynthesis and a major source of one-carbon ( 1C ) units in folate metabolism , which supports purine and thymidylate synthesis as well as cellular methylation reactions ( Tibbetts and Appling , 2010 ) . Serine can allosterically activate PKM2 ( Chaneton et al . , 2012 ) and thereby regulate the exit of carbons from glycolysis ( Ye et al . , 2012 ) . Serine is also a precursor for cysteine and glutathione biosynthesis via transsulfuration , and HCT116 cells lacking p53 exhibit serine dependency that is rescuable with glutathione supplementation ( Maddocks et al . , 2013 ) . We became interested in the role of serine in 1C metabolism because mitochondria are the source of most of the 1C units used in cellular biosynthesis , and because consumption of serine for 1C metabolism involves an oxidation step that requires an NAD+ cofactor ( Tibbetts and Appling , 2010 ) . Thus , loss of NAD+ reoxidation due to the loss of the mitochondrial respiratory chain could conceivably lead to impairment of mitochondrial 1C metabolism , as theorized previously ( Desler et al . , 2010 ) . To test this model of 1C metabolism blockade in intact cells , we note that the NAD+-dependent step in mitochondrial metabolism is the oxidation of methylene-THF to formyl-THF by MTHFD2 . Our model thus predicts an increase in abundance of methylene-THF . We inferred increased methylene-THF abundance by determining rates of serine isotopic scrambling ( Figure 4—figure supplement 1 ) . Briefly , 13C3-labeled serine generates 13C2-glycine and a 13C labeled methylene-THF in mitochondria by the action of SHMT2 . This labeled methylene-THF can recombine with an unlabeled glycine – which is present in excess – by the reverse action of SHMT2 to generate 13C1-serine . The rate of emergence of 13C1-serine thus reads out methylene THF abundance . We performed these experiments in SHMT1 knockout cells ( see below ) to avoid interference from cytoplasmic folate reactions , and found increased emergence of 13C1-serine upon mtDNA depletion ( Figure 4c ) . We observed similarly increased serine scrambling by treatment with rotenone , which directly inhibits NADH reoxidation through respiratory complex I . Measurement of cellular NAD+:NADH ratios confirms that these are altered with both mtDNA depletion and rotenone treatment ( Figure 4—figure supplement 2 ) . Therefore , inhibition of mitochondrial respiration likely shifts the redox state of the mitochondrial folate pool in favor of reduced species such as methylene-THF . To determine whether the shift in the mitochondrial redox state could inhibit mitochondrial 1C metabolism , we performed formate synthesis assays on isolated mitochondria treated with respiratory chain poisons . As shown in Figure 4d , both rotenone and antimycin A partially inhibit mitochondrial synthesis of formate from serine . García-Martínez and Appling performed similar experiments with mitochondria isolated from rat liver ( García-Martínez and Appling , 1993 ) , and also observed partial inhibition of formate synthesis with rotenone treatment . Elevations in homocysteine often signify insufficient one carbon pools . Hence , an important question is the degree to which mitochondrial 1C compromise is compatible with our observation of decreased homocysteine ( Figure 1b ) . In mammals , homocysteine has two major fates: conversion to cysteine via transsulfuration , and conversion to methionine via remethylation; the latter process consumes cytoplasmic 1C units . Our hypothesized mitochondrial 1C compromise might therefore lead to increased homocysteine abundance and be in direct conflict with the strong observed decrease in homocysteine levels in both spent media and cell extract . To examine this issue more closely , we performed deuterium tracer experiments to determine the extent of homocysteine remethylation in T-REx-293 cells . As shown in Figure 4—figure supplement 4 , homocysteine remethylation using exogenously supplied , deuterated formate was nearly 1000-fold slower than synthesis of serine from deuterated formate . We conclude that remethylation is not a major fate for homocysteine in T-REx-293 cells , regardless of 1C availability . The observation of decreased homocysteine in our experiments , which we attribute to increased transsulfuraton , is therefore compatible with our model of mitochondrial 1C impairment . 1C units are required for cellular biosynthesis , but treating T-REx-293 cells with either rotenone or antimycin , both of which impair mitochondrial 1C generation ( Figure 5a ) , only gave subtle growth rates changes . To test the robustness of 1C metabolism in T-REx-293 cells undergoing mitochondrial dysfunction , we withdrew serine from cells treated with inhibitors . Antimycin- or rotenone-treated cells , but not control DMSO-treated cells , showed strongly impaired growth when serine was withdrawn ( Figure 5a , Figure 5—figure supplement 1a ) . We observed a similar effect with serine withdrawal in cells depleted of mtDNA using EtBr ( Figure 5b ) . A serine dose-response curve in the presence of antimycin ( Figure 5—figure supplement 1c ) shows optimal cell growth above about 100 μM serine in the media . 10 . 7554/eLife . 10575 . 018Figure 5 . Serine dependence in cells with compromised mitochondrial function . ( a ) Growth of T-REx-293 cells treated with OXPHOS inhibitors , with and without serine , and with serine replaced by formate or hypoxanthine . ( b ) Same , but of T-REx-293 cells depleted of mtDNA using EtBr . ( c ) Growth of T-REx-293 cells with knockouts of SHMT1 or SHMT2 , with and without serine , and with serine replaced by formate . ( d ) Growth of S . cerevisiae in nonfermentable media with 5 nM antimycin ( see Materials and methods ) , in the presence or absence of serine and adenine . n = 3 for panels a-c; n = 4 for panel d . DOI: http://dx . doi . org/10 . 7554/eLife . 10575 . 01810 . 7554/eLife . 10575 . 019Figure 5—figure supplement 1 . Additional data on RC inhibition-induced serine dependency in T-REx-293 cells . ( a ) Quantitated growth rates , with and without media serine , for T-REx-293 cells treated with different mitochondrial inhibitors . n = 3 . ( b ) Suppression of cellular growth in galactose , a nonfermentable sugar ( Gohil et al . , 2010 ) , by a dose of CCCP ( 5 μM ) that does not elicit serine dependency . n = 3 . ( c ) Serine dose response curve of T-REx-293 cell growth in the presence of antimycin . n = 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 10575 . 01910 . 7554/eLife . 10575 . 020Figure 5—figure supplement 2 . SHMT1 and SHMT2 single and double knockouts . ( a ) Western blots ( top ) and growth rates ( bottom ) for fourteen single cell clones derived from simultaneous CRISPR transfection for SHMT1 and SHMT2 knockout . All clones were expanded in the presence of hypoxanthine and thymidine ( HT ) . ( b ) Western blots showing SHMT1 and SHMT2 single and double knockouts . ( c ) Formate synthesis from mitochondria isolated from SHMT1 and SHMT2 knockout cell lines . ( d ) Replicate serine dependence data from single knockouts . ( e ) Growth of SHMT1/SHMT2 double KO cells ( two independent clones ) with different 1C-related supplements . ( f ) 1C-rescued double knockouts do not show serine dependence , even when treated with RC inhibitor . n = 1 for panel a; n = 3 for panels c–f . DOI: http://dx . doi . org/10 . 7554/eLife . 10575 . 02010 . 7554/eLife . 10575 . 021Figure 5—figure supplement 3 . RC inhibition-induced serine dependence in other cell types . n = 3 . DOI: http://dx . doi . org/10 . 7554/eLife . 10575 . 021 We note that this phenomenon is not a true auxotrophy , since cells are still able to synthesize serine . Indeed , if cells with mitochondrial dysfunction were truly serine auxotrophs , they would show increased serine consumption . Instead , we observed the opposite ( Figure 4b ) . We surmise that these cells can overcome blockade of mitochondrial serine metabolism by mass action , but only at high cytoplasmic serine concentrations . Such a bypass mechanism would give rise to serine dependency if cells lose the cytoplasmic serine that they synthesize by diffusion to the media when exogenous serine is withheld , as has been observed for many cell types ( Eagle and Piez , 1962 ) . We also observed serine dependency with oligomycin ( Figure 5—figure supplement 1a ) , which others ( Maddocks et al . , 2013 ) have reported in HCT116 cells and attributed to serine involvement in energy metabolism . To distinguish the two models of serine dependency , we tested the effect of the mitochondrial uncoupler CCCP , which impairs energy production ( Figure 5—figure supplement 1b ) but increases NADH oxidation ( Figure 4—figure supplement 2 ) . CCCP did not give rise to serine dependency , and indeed a low dose of CCCP could rescue both the growth defect and the serine dependency arising from oligomycin treatment ( Figure 5—figure supplement 1a ) . These data suggest that the mitochondrial dysfunction-induced serine dependency that we observe is related to the redox state of NAD cofactors , and not to energy impairment . To further implicate 1C metabolism in serine dependency , we rescued this dependency using 1C related metabolites . Knockouts of MTHFD2 and MTHFD1L , two enzymes involved in mitochondrial formate synthesis , are embryonic lethal in mammals ( Di Pietro et al . , 2002; Momb et al . , 2013 ) , but some knockout phenotypes are rescuable with formate or hypoxanthine supplementation ( Momb et al . , 2013; Patel et al . , 2003 ) . Exogenous formate bypasses the need for its mitochondrial synthesis , and hypoxanthine is an alternate source of purines , the de novo synthesis of which requires 1C units . Experiments on OXPHOS-inhibited T-REx-293 cells with serine withdrawal showed near-complete rescue of growth when media was supplement with either 30 µM hypoxanthine or 3 mM sodium formate ( Figure 5a ) . These concentrations of hypoxanthine and formate also gave strong rescue of growth defects in the absence of serine in mtDNA-depleted cells ( Figure 5b ) . These data further support the notion that serine dependence in the face of mitochondrial dysfunction is related to 1C metabolism . Parallel folate pathways exist in the cytoplasm and the mitochondria ( Tibbetts and Appling , 2010 ) . To further determine the compartment specificity of serine fluxes , we used CRISPR/Cas9 ( Ran et al . , 2013 ) to knock out serine hydroxymethyltransferase ( SHMT ) genes in T-REx-293 cells ( Figure 5—figure supplement 2 ) . SHMT catalyzes the first step in folate-mediated production of formate from serine . Two isoforms exist in mammals: SHMT1 , which is exclusively cytoplasmic , and SHMT2 , which is mostly mitochondrial but has a cytoplasmic splice variant ( Anderson and Stover , 2009 ) . As expected , knockout of SHMT2 , but not SHMT1 , abolished formate synthesis from serine in isolated mitochondria ( Figure 5—figure supplement 2c ) . In the presence of serine , neither SHMT1 KO nor SHMT2 KO cell lines required formate or hypoxanthine for growth , indicating that serine can be used to generate 1C units in either compartment . Without serine in the culturing media , SHMT2 KO cells showed strongly suppressed growth that could be rescued with formate , similarly to cells treated with RC inhibitors ( Figure 5c ) , whereas SHMT1 KO cells did not . We hypothesize that high levels of exogenous serine are required to drive conversion of serine to formyl-THF in the cytoplasm via SHMT1 , likely because MTHFD1 , the cytoplasmic enzyme that makes formate , is coupled to NADPH and thus favors formyl-THF consumption rather than generation ( Tibbetts and Appling , 2010 ) . This provides a possible explanation for why high levels of serine can rescue 1C synthesis in T-REx-293 cells with compromised RC function . Mammalian cells can also derive 1C units from breakdown of choline and glycine , which could in principle support growth in the absence of serine ( Tibbetts and Appling , 2010 ) . To investigate the roles of these other 1C precursors , we generated T-REx-293 cells with simultaneous knockout of both SHMT1 and SHMT2 genes . Supplementation with hypoxanthine and thymidine ( HT ) is required for cell growth in the absence of 1C metabolism ( Hakala , 1957 ) , since 1C units are used for the synthesis of purines and of thymidylate ( Tibbetts and Appling , 2010 ) . Therefore , we supplemented media with HT when we generated these double-KO ( DKO ) clones . DKO cells so generated were unable to grow on unsupplemented media , whereas most single knockouts and WT cells were able to grow ( Figure 5—figure supplement 2a ) . In all cases , formate ( 3 mM ) could replace HT to support DKO cell growth ( not shown ) . We conclude from this that other sources of 1C are unable to completely replace serine-derived 1C units . Interestingly , hypoxanthine alone was also able to partially rescue DKO cell growth ( Figure 5—figure supplement 2e ) , suggesting that an alternative pathway might be sufficient to supply 1C for thymidylate synthesis . Finally , DKO cells , when rescued using either formate or hypoxanthine , were no longer serine dependent even with antimycin treatment ( Figure 5—figure supplement 2f ) , indicating that no other functions of serine account for its requirement in cells undergoing RC inhibition . We performed serine scrambling assays in five cell lines – T-Rex293 , 293T , HeLa , C2C12 , and MCH58 – to determine whether mtDNA depletion could induce 1C deficits in each of them . Instead of using SHMT1 knockout cells , we sought to suppress cytoplasmic SHMT activity by using lower concentrations of serine in the labeling media . Except with MCH58 cells , where high variance in the untreated samples prevented us from reaching any conclusions , we consistently observed increased ratios of 13C1-serine to 13C3-serine , with at least one time point showing a statistically significant increase for all the other cell lines tested ( Figure 4—figure supplement 3 ) . However , impairment of 1C metabolism only led to serine dependence , manifesting as a proliferation defect upon serine withdrawal , in a small handful of cell lines: besides T-Rex293 , we only observed similar results in 293MSR and U251 cells ( Figure 5—figure supplement 3 ) ; a recent report ( Maddocks et al . , 2013 ) additionally shows serine dependence of HCT116 cells treated with oligomycin . We failed to observe serine dependence in many other cell lines , including 293T cells which , like T-Rex293 and 293MSR cells , are derived from HEK-293 cells . We surmise that expression of the large or small T antigen in 293T cells enables them to better cope with respiratory chain dysfunction-induced 1C compromise . Although few mammalian cells demonstrate serine dependence in the presence of respiratory chain inhibition , we have observed similar phenomena in S . cerevisiae grown in nonfermentable media with sub-lethal doses of antimycin: serine withdrawal exacerbates antimycin-induced growth defects in a manner rescuable with purine supplementation ( Figure 5d ) .
Classic inborn errors of metabolism are due to lesions within linear pathways , leading to accumulation of upstream substrates ( that can be toxic ) or depletion of essential downstream molecules . Therapeutic strategies for these diseases are often aimed at limiting substrate accumulation or replenishing of downstream factors . In respiratory chain dysfunction , however , the resultant metabolic derangements are manifold because of the numerous mitochondrial and cytoplasmic processes that are intimately coupled to the respiratory chain ( Vafai and Mootha , 2012 ) . Well-known metabolic alterations associated with mitochondrial disease include increased reliance on glycolysis as a source of ATP ( Robinson et al . , 1992 ) , dependence on pyruvate to support aspartate biosynthesis ( Sullivan et al . , 2015; Birsoy et al . , 2015; Cardaci et al . , 2015 ) , and inability to synthesize pyrimidines ( King and Attardi , 1989 ) . In this paper , we offer two lines of evidence to support the idea that cellular 1C metabolism is also altered upon respiratory chain dysfunction . First , generating hypotheses with metabolic , proteomic , and transcriptional profiling and following these observations up with focused validation experiments , we report that cells activate serine biosynthesis and transsulfuration in response to mtDNA depletion . Second , we show in separate experiments that lesioning the respiratory chain impairs mitochondrial production of formate from serine . In a small number of cell types , this leads to growth defects upon serine withdrawal that is rescuable with purine or formate supplementation . An important question is the degree to which our observations of one-carbon metabolism remodeling in proliferating cells are relevant to in vivo mitochondrial disease , which mostly affects post-proliferative tissue . We note that our observation of ATF4 activation and alterations in serine metabolism and transsulfuration is richly supported by recent studies of mitochondrial disease . These have shown ATF4 activation by mitochondrial dysfunction in human cells lines ( Martínez-Reyes et al . , 2012; Silva et al . , 2009 ) , rodent models ( Tyynismaa et al . , 2010 ) , and humans ( Crimi et al . , 2005 ) . Increased transsulfuration has been observed following ATF4 activation in general ( Dickhout et al . , 2012 ) , and mitochondrial dysfunction in particular ( Krug et al . , 2014 ) . One of the byproducts of transsulfuration , α-hydroxybutyrate , has been recently identified as a promising biomarker for mitochondrial disease in humans ( Thompson Legault et al . , 2015 ) and mice ( Jain et al . , 2016 ) . ATF4 is known to activate serine biosynthesis ( Ye et al . , 2012 ) , and increased serine has been observed in a number of mitochondrial disease settings , albeit with less consistency . Muscle from male mice ( but not female ones ) with mitochondrial myopathy show increased serine levels ( Tyynismaa et al . , 2010 ) . Serine was found to be highly elevated in urine from mitochondrial disease patients ( Smuts et al . , 2013 ) and proposed as a component of a novel biosignature for mitochondrial disease , but not found to be increased in the blood of a different cohort of mitochondrial disease patients ( Clarke et al . , 2013 ) . Serine levels are increased in a mitochondrial Parkinson’s disease model in Drosophila ( Tufi et al . , 2014 ) , but not in C . elegans genetic models of mitochondrial disease ( Falk et al . , 2008; Morgan et al . , 2015 ) . Furthermore , while serine dependency appears to be cell type specific , 1C compromise as evidenced by altered serine scrambling appears to be more general . We note that a recent report on the effect of mitochondrial inhibitors showed impaired serine uptake and formate release in differentiated C2C12 myotubes ( Xu et al . , 2011 ) . Therefore , 1C compromise does not appear to be limited to proliferating cells . More recently , studies of mouse mitochondrial disease models have also noted imbalanced nucleotide pools and altered 1C-related metabolites ( Nikkanen et al . , 2016 ) . Taken together , these data suggest that 1C metabolism is more efficient in the presence of a functioning respiratory chain in a variety of in vivo and in vitro contexts , and argues that the 1C compromise seen with respiratory chain dysfunction might be generalizable to whole body metabolism . Whether the ATF4 response is adaptive in this context has not been clear . Our study shows ATF4-dependent activation of serine synthesis ( Ye et al . , 2012 ) following respiratory chain blockade , and poses the hypothesis that the serine synthesis-activating aspect of ATF4 might be adaptive by helping maintain cellular 1C availability . Indeed , ATF4 has recently been shown to mediate purine synthesis induction by mTORC1 ( Ben-Sahra et al . , 2016 ) . Transsulfuration constitutes another potentially adaptive element of the ATF4 response program , since it supports synthesis of glutathione , a key cellular ROS scavenger . However , we have so far failed to observe deleterious effects of ablating the ATF4 response in T-REx-293 cells . Since the ATF4 integrated stress response program encompasses a number of different genes and functions , we hypothesize that it contains both adaptive and maladaptive components . In support of this notion , a recent paper showed that inhibition of ATF4 activation is protective in rodent models of intracerebral hemorrhage ( Karuppagounder et al . , 2016 ) . Our work may provide mechanistic insights into the folate deficiency in mitochondrial disease , with therapeutic implications . Mitochondrial disorders are occasionally associated with cerebral folate deficiency ( CFD ) , and some of these patients are known to respond to treatment with folinic acid ( Garcia-Cazorla et al . , 2008 ) . No proven mechanism yet connects mitochondrial disease to CFD , but given that brain expresses very little SHMT1 ( Girgis et al . , 1998 ) , our results pose the hypothesis that CFD may result from impaired mitochondrial formate synthesis ( Figure 4 ) . Formate supplementation has been shown to ameliorate defects associated with 1C gene knockout ( Momb et al . , 2013 ) , and a recent report demonstrated that the bioenergetics defects of a fly model of Parkinson’s disease could be reversed with supplementation of nucleotides ( Tufi et al . , 2014 ) . Mitochondrial disease patients are commonly given folic acid as part of the 'mito cocktail , ' with no proven efficacy . Our results suggests that these patients might lack the 1C units that are bound by the folate cofactors , and not the co-factors themselves . Our work raises the hypothesis that formate or nucleotide supplementation may be of benefit in some of respiratory chain diseases . One of the most puzzling characteristics of mitochondrial diseases is their phenotypic heterogeneity and tissue-specific pathology ( Vafai and Mootha , 2012 ) . Our identification of 1C metabolism and transsulfuration as metabolic alterations associated with mitochondrial disease presents a new set of hypotheses that could help explain this heterogeneity . Different organs in the body differ in which 1C donors they utilize , in their ability to metabolize those 1C units ( Yoshida and Kikuchi , 1973 ) , and in their transsulfuration activity ( Mudd et al . , 1965 ) . Because many of these one-carbon pathways are coupled to the respiratory chain , a lesion within the mitochondrion could in principle ripple out to 1C defects that could manifest in many different ways depending on the tissue and cell state . Rapidly proliferating cells might be unable to replicate their DNA due to nucleotide imbalances , whereas metabolically active cells in the liver and kidney might be deprived of essential purine-containing cofactors such as NAD+ . Tissue-specific variations in 1C metabolism may contribute to the remarkable phenotypic variation of mitochondrial disease .
T-REx-293 cells were obtained from Thermo Fisher ( Waltham , MA ) . POLGdn cells were generated as in Jazayeri et al . ( 2003 ) . U251 cells were obtained from the National Cancer Institute . 293T , HeLa , and C2C12 were obtained from ATCC . MCH58 cells were a generous gift from Eric Shoubridge . We did not authenticate cell lines . However , we were able to differentiate C2C12 myoblasts into myotubes , and were able to PCR the SV40 small and large T antigens from 293T cell genomic DNA . All cells were cultured with DMEM ( 11995 , Thermo Fisher ) supplemented with 10% FBS ( F6178 , Sigma-Aldrich , St . Louis , MO ) and grown at 37°C with 5% CO2 . DMEM without serine or glycine , but with 1 mM pyruvate and in every other respect identical to the DMEM obtained from Thermo Fisher , was custom-made ( U . S . Biological , Salem , MA ) and supplemented with 10% dialyzed serum ( F0392 , Sigma-Aldrich ) . All cell cultures were tested for mycoplasma contamination monthly and confirmed to be free of mycoplasma . DNA was extracted from cells and analyzed by multiplex real time quantitative PCR as in Baughman ( 2011 ) . Briefly , samples of 5×104 cells were pelleted , aspirated , and suspended in 50 μl lysis buffer ( 25 mM NaOH , 0 . 2 mM EDTA ) . Cell lysate was heated to 95°C for 15 min to hydrolyze protein and RNA , and neutralized with addition of 50 μl neutralization buffer ( 40 mM Tris-HCl ) . 5 μl of 1:50 diluted neutralized cell lysate was analyzed using multiplexed TaqMan real-time quantitative PCR in 20 μl reactions , with a custom-synthesized assay for the AluYb8 repeat element to quantitate nuclear DNA copy number and a custom-synthesized assay for MT-ND2 to quantitate mitochondrial DNA copy number . The custom-synthesized AluYb8 Taqman assay consisted of the forward primer 5’-CTTGCAGTGAGCCGAGATT-3’ , the reverse primer 5’-GAGACGGAGTC-TCGCTCTGTC-3’ , and the probe 5’-VIC-ACTGCAGTCCGCAGTCCGGCCT-MGBNFQ-3’ . The custom-synthesized MT-ND2 assay consisted of the forward primer 5’-TGTTGGTTATACCCTTCCCGTACTA-3’ , the reverse primer 5’-CCTGCAAAG-ATGGTAGAGTAGATGA-3’ , and the probe 5’-6FAM-CCCTGGCCCAACCC-MGBNFQ-3’ . To calibrate MT-ND2 PCR Ct values , we used dilution ladders of a chemically synthesized primer corresponding to the target of the quantitative PCR assay . To calibrate AluYb8 PCR Ct values , we used dilution ladders of total DNA extracted from human cells . PolGdn cells were split every 3 d . On each split , two 10 cm plates were seeded with 2 . 0 , 1 . 0 , and 0 . 5 M each for RNA and protein collection after 1 , 2 , and 3 d . RNA was extracted from cell plates using RNeasy columns ( Qiagen , Germany ) with DNase treatment to remove DNA contamination . For protein collection , cells were first trypsinized , suspended , and counted . Counts derived from protein collection were used for growth curve computation . 5×104 cell were used for mtDNA analysis . The remaining cells were then pelleted and lysed using RIPA buffer with cOmplete protease inhibitor ( Roche , Switzerland ) . Ethidium bromide growth curves were generated in similar fashion , without plates for RNA samples . We note that continuous doxycycline treatment also gave rise to mtDNA repletion , albeit more slowly than when doxycycline was removed , raising the possibility that the strong selective pressure against POLGdn expression could yield mutant cells that lost POLGdn expression . Oxygen consumption measurements were performed in a Seahorse XF24 Analyzer ( Agilent , Santa Clara , CA ) . Untreated and mtDNA-depleted POLGdn-expressing cells , suspended by trypsinization , were seeded at 105 per well into Seahorse cell plates pre-treated with Cell-Tak ( Becton Dickinson , Santa Clara , CA ) according to manufacturer’s recommendations . Cells were incubated in DMEM in Seahorse cell plates for 1 hr before oxygen consumption measurement . PolGdn-expressing cells were seeded to twelve 6-cm plates at 5×105 ea in 3 ml DMEM + 10% FBS . Four of these plates had 1 μg/ml doxycycline , and another four had 1 μg/ml doxycycline added 1 d after seeding . 3 d after seeding , media was collected from plates . Plates were washed once with ice-cold PBS and aspirated . Polar metabolites were extracted from two plates of each condition using 1 ml -80°C 80% methanol:20% water . Nonpolar metabolites were extracted from two plates of each condition using 1 ml -80°C isopropanol . A combination of three liquid chromatography tandem mass spectrometry ( LC-MS ) methoda was used to measure metabolites in cell extracts and spent media as described previously ( Townsend et al . , 2013 ) . Briefly , polar metabolites were profiled using a 4000 QTRAP triple quadrupole mass spectrometer ( SCIEX; Framingham , MA ) coupled to a 1200 Series pump ( Agilent Technologies; Santa Clara , CA ) and an HTS PAL autosampler ( Leap Technologies; Carrboro , NC ) . Media samples ( 10 µL ) were extracted using 90 µL of 74 . 9:24 . 9:0 . 2 v/v/v acetonitrile/methanol/formic acid containing stable isotope-labeled internal standards ( valine-d8 , Isotec; and phenylalanine-d8 , Cambridge Isotope Laboratories; Andover , MA ) and centrifuged ( 10 min , 9000 x g , 4°C ) . Cell extracts and media extraction supernatants ( 10 µL ) were injected directly onto a 150 x 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 . MS analyses were carried out using electrospray ionization and selective multiple reaction monitoring scans in the positive ion mode . Declustering potentials and collision energies were optimized for each metabolite by infusion of reference standards before sample analyses . The ion spray voltage was 4 . 5 kV and the source temperature was 450°C . Polar metabolite were profiling in the negative ion mode using a 5500 QTRAP triple quadrupole mass spectrometer ( SCIEX; Framingham , MA ) coupled to an ACQUITY UPLC ( Waters; Milford , MA ) . Media samples ( 30 µL ) were extracted using 120 µL of 80% methanol containing inosine-15N4 , thymine-d4 and glycocholate-d4 internal standards ( Cambridge Isotope Laboratories; Andover , MA ) and were centrifuged ( 10 min , 9000 x g , 4°C ) . Cell and media extract supernatents ( 10 µL ) were injected directly onto a 150 x 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 v/v acetonitrile/methanol ) followed by a 10 min linear gradient to 100% mobile phase A . The ion spray voltage was -4 . 5 kV and the source temperature was 500°C . Lipid profiling was performed in the positive ion mode using a 4000 QTRAP triple quadrupole mass spectrometer ( SCIEX; Framingham , MA ) coupled to a 1100 Series pump ( Agilent Technologies; Santa Clara , CA ) and an HTS PAL autosampler ( Leap Technologies; Carrboro , NC ) . Lipids were extracted from media ( 10 µL ) using 190 µL of isopropanol containing 1-dodecanoyl-2-tridecanoyl-sn-glycero-3-phosphocholine ( Avanti Polar Lipids; Alabaster , AL ) . Cell and media extracts ( 10 µL ) were injected directly onto a 150 x 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 . MS analyses were carried out using electrospray ionization and Q1 scans in the positive ion mode . Ion spray voltage was 5 . 0 kV and source temperature was 400°C . For each lipid analyte , the first number denotes the total number of carbons in the lipid acyl chain ( s ) and the second number ( after the colon ) denotes the total number of double bonds in the lipid acyl chain ( s ) . For each method , internal standard peak areas were monitored for quality control . MultiQuant software ( Version 1 . 1; AB SCIEX; Foster City , CA ) was used for automated peak integration and metabolite peaks were manually reviewed for quality of integration and compared against a known standard to confirm identity . Cell extract metabolite abundances were normalized to sample geometric means to adjust for cell number differences . We used a multi-step protocol to adjust spent media metabolite abundances . First , we measured a media volume change of 12 . 0% in 6 cm plates with 3 ml media over the course of 3 d , and re-scaled media metabolite measurements accordingly . Then , we used the log mean cell extract metabolite differences between the different samples to estimate the per-day growth rate difference due to doxycycline treatment . We then used the growth rate difference to estimate changes in the area-under-the-curve ( AUC ) exposure of media to cells and adjusted accordingly . While this adjustment entailed straightforward flux multiplication in the case of metabolites released to the media , metabolites strongly absorbed from the media were more difficult to treat because multiplication would give rise to cells absorbing more metabolite than was present in the base media . We therefore used an ad-hoc adjustment function for uptaken metabolites that satisfied three criteria: ( 1 ) its value and slope would match that for released metabolites at the zero flux boundary case; ( 2 ) it would be continuous for all uptake fluxes; ( 3 ) measurements of near-100% uptake would adjust to yield near-100% uptake . Our final adjustment function wasRadj={1+α ( R−1 ) , R>1Rα−R ( α−1 ) , 0<R<1 where R is the evaporation-adjusted abundance ratio compared to base media , α the AUC adjustment factor , and Radj the growth-adjusted metabolite abundance ratio used in Figure 1—figure supplement 3 and Figure 1—figure supplement 4 . RNA from 18 time points ( n = 1 for each time point ) were analyzed using Affymetrix ( Santa Clara , CA ) Human Genome U133 Plus 2 . 0 arrays . RNA sample processing , hybridization to Affymetrix U133 Plus 2 . 0 microarrays , and microarray imaging were performed according to manufacturer’s recommendations . Raw microarray data were analyzed using the RMA algorithm with manufacturer-provided probeset definitions . Probesets were filtered in an ad-hoc manner according to three criteria: ( a ) probeset standard deviation across all the timepoint samples were required to be at least 7 . 5% of the probeset mean , to remove unchanging probesets; ( b ) maximum values for probesets were required to be at least 50 , to remove probesets with insufficient signal; ( c ) total power in the second through sixth coefficients of the Fourier transform of the data were required to be at least as strong as the total power in the seventh through twelfth , to remove probesets with high frequency noise that we considered unlikely to be informative . Probesets that did not survive these filters were not considered for any downstream analysis . To estimate the number of transcripts with significant changes at day 6 , we pooled data from days 5–7 and compared these to data from one and zero days prior to initiation of doxycycyline treatment . Full microarray data have been deposited to GEO: http://www . ncbi . nlm . nih . gov/geo/query/acc . cgi ? acc=GSE55311 ( Mootha et al . , 2004 ) was performed using publicly available code ( Broad Institute ) . Genes were ordered by the ratio of the average value from days 1 through 10 of dox treatment , to the average of untreated samples . Sequences flanking transcription start sites , with repeat sequences masked , were obtained from the UCSC genome browser . Mouse-human homology was used to improve detection of conserved transcription factor binding sites . We looked for all ungapped motifs between 6 and 8 in length , as well as 9mer motifs with single gaps . We observed the ATF4 signal regardless of whether we looked for bidirectional occurrence of the motifs or considered each direction separately; the results shown in Figure 2a and Supplementary file 2 are for bidirectional searching . The N-terminal truncation of GADD34 reported in Novoa et al . ( 2001 ) was from CHO-K1 cells; we wished to obtain the analogous human version of the protein . We BLASTed the C . griseus GADD34 protein sequence against the human , and determined that the truncation point corresponded to position 308 in the H . sapiens GADD34 protein sequence ( NCBI reference sequence symbol PPP1R15A ) . We amplified this fragment from T-REx-293-derived complementary DNA using the PCR primers GATGAAGAGGAGGGTGAGGTCAAG and GCCACGCCTCCCACTGAGG , performed an additional 30 cycles of PCR using Gateway-flanked primers that also incorporated an ATG start codon , and recombined the resulting PCR product into the pDONR221 Gateway entry vector using BP Clonase II ( Thermo Fisher ) . We call the protein product of this insert GADD34△N . We encountered difficulties in selecting for lentivirally-transduced GADD34△N expression using puromycin resistance , and suspected that inhibition of the ATF4 response might make cells more susceptible to puromycin treatment . To work around this , we used restriction digestion to remove the puromycin resistance open reading frame from the pLX302 lentiviral expression vector ( Addgene , Cambridge , MA ) and used Gibson isothermal assembly ( Gibson et al . , 2009 ) to replace it with GFP . We recombined the GADD34△N insert into this new vector ( which we call pLX-GFP ) , made virus with psPAX2 ( Addgene ) and pCMV-VSVG ( Broad Institute ) helper plasmids , and infected T-Rex293 cells at MOI < 1 . We then selected for GFP expression with two rounds of fluorescence-activated cell sorting on a FACSAria II ( Becton Dickinson ) . Control infections were done similarly , except with an additional copy of GFP instead of GADD34△N inserted using Gateway recombination . We characterized protein changes ( n = 1 ) for two different time points , plus a control . PolGdn cells were grown in SILAC labeled DMEM ( Caisson Labs , Logan , UT ) with 10% dialyzed FBS ( Sigma ) . They were first seeded at 4 . 0×106 each into three sets of replicate 15-cm plates with heavy ( R10K8 ) , medium ( R6K4 ) , and light ( R0K0 ) SILAC media and grown for 3 d . They were then split into eight 15-cm plates for each condition . Doxycycline was added at 1 μg/ml 2 d before the split to heavy labeled cells , and 1 d after the split to medium labeled cells . 3 d after the split , cells were trypsinized , and cells from replicate plates pooled and counted . 7×106 cells from each condition were mixed , pelleted , aspirated , suspended in 8 M urea and sonicated . This sample was treated with dithiothreitol ( 1 mM final ) , and iodoacetamide ( 6 mM final ) and proteins fractionated by SDS-PAGE . Each gel lane was cut into ten slices of approximately equal staining intensity , in-gel digested with trypsin , and analyzed by liquid chromatography-mass spectrometry on a LTQ-Orbitrap ( Thermo , Inc . ) following the procedure in Sancak et al . ( 2013 ) . SILAC data were analyzed using MaxQuant ver . 1 . 0 . 13 . 13 according to the procedure described in Cox et al . ( 2009 ) . A minimum of two quantified peptides was required for each quantified protein . The FDR for protein and peptide identification was set at 0 . 01 . The IPI human database ver 3 . 65 was used and supplemental sequences from mitochondrial DNA and common contaminants such as keratins and serum proteins were included at search time . Raw data are publicly available at www . broadinstitute . org/proteomics Sulfide ( sum of H2S , HS- , and S2- ) levels in the T-REx-293 cells were measured using monobromobimane ( MBB ) -based high performance liquid chromatography ( HPLC ) analysis as reported previously ( Marutani et al . , 2012; Tokuda et al . , 2012 ) . Briefly , cells were washed with ice-cold Tris-HCl ( 100 mM , pH 9 . 5 , 0 . 1 mM DTPA ) buffer , scraped , transfered to an eppendorf tube , and centrifuged to obtain the supernatant . MBB ( 10 mM in acetonitrile , 50 µl ) was added to 100 µl of supernatant . After 30 min of incubation at room temperature in dark , 50 µl of 200 mM 5-sulfosalicylic acid ( SSA ) was added . After centrifugation , supernatant was analyzed by HPLC equipped with Agilent HPLC column C18 and Waters 2475 Multi λ fluorescence detector . Sulfane sulfur levels in T-REx-293 cells were measured using a sulfane sulfur-specific fluorescent probe SSP4 ( generous gift from Dr . Ming Xian , Washington State University ) as reported previously ( Marutani et al . , 2014 ) . Briefly , cells were washed with ice-cold Tris-HCl ( 100 mM , pH 9 . 5 , 0 . 1 mM DTPA ) buffer , scraped , transfered to an eppendorf tube , and centrifuged to obtain the supernatant . The supernatant was transferred into a 96 well-plate , incubated with SSP4 at 20 µM at 37°C for 30 min , and fluorescent intensity was read by a microplate reader ( SpectraMax M5 Microplate reader , Molecular Devices , Sunnyvalem CA ) . PolGdn cells , treated for 0 , 3 , and 6 d with doxycycline to induce mtDNA depletion , were in triplicate 35 mm plates at approximately 2×106 cells per plate . Each plate was washed with warm PBS and aspirated . To each plate was then added 1 ml of warm 13C glucose media , which was glucose- and serine-free DMEM + 10% dialyzed FBS , supplemented with 25 mM [U-13C]glucose ( Cambridge Isotope Labs ) and 50 μM unlabeled serine ( Sigma ) . We used this lower concentration of serine ( DMEM normally contains 400 μM ) to avoid potential product-inhibition of serine synthesis ( Fell and Snell , 1988 ) . Cells were incubated with labeling media at 37°C for 30 min , after which labeling media was removed , pelleted at 2000 g for 20 s , and supernatant taken for LC-MS/MS quantitaton ( see below ) . The cell pellet and cells on plates were trypsinized , pooled , resuspended , and counted to adjust serine fluxes for cell number . Cell medium samples ( previous paragraph ) were prepared using nine volumes ( 1:9 v/v ) of extraction solution containing 75% acetonitrile , 25% methanol , and 0 . 2% formic acid , and vortexed . Samples were then centrifuged at 11 , 000 rpm at 4ºC for 8 min , and the supernatant was injected into LC-MS/MS . A series of 12C serine ( BioUltra 99 . 5% , Sigma ) and 13C serine ( 13C3 , 99% , Cambridge Isotope Labs ) standard solutions at concentrations of 0 . 1 , 0 . 5 , 1 , 5 , 10 , 50 µM were prepared in serine-free base medium to generate a calibration curve . An Agilent 1260 HPLC system coupled with Q Exactive ( Thermo Fisher ) was used perform LC-MS/MS quantitation . LC Column was Atlantis HILIC Silica 2 . 1 × 150 mm column and particle size was 3 μ m . Mobile Phase A was 0 . 1% formic acid , 10 mM ammonium formate . Mobile Phase B was 0 . 1% formic acid in acetonitrile . Column was at room temperature . 10 uL volume of samples was injected into LC-MS/MS . The flow rate was 0 . 25 mL/min . The LC gradient was as follows: 95% B was held from 0 to 0 . 5 min , then from 0 . 5 to 10 . 5 min , 95% B was linearly changed to 40% . From 10 . 5 to 15 min , 40% B was held . From 15 min to 17 min 40% B was changed to 95% . From 17 to 32 min , 95% B was held to equilibrate the column . Serine was quantified using the targeted MS2 method . For 12C serine , the precursor ion was 106 . 0506 and the fragment ion 60 . 0454 was used for extraction ion chromatogram . For 13C serine , the parent ion was 109 . 0606 and the fragment ion 62 . 0521 was used for extraction ion chromatogram . The extraction ion mass window was 10 ppm . The NCE was 20 . To confirm that mtDNA depletion from EtBr treatment could also give rise to increased serine and decreased homocysteine , we seeded T-REx-293 cells to 6-cm plates at 5×105 ea in 3 ml DMEM + 10% FBS , with and without 100 ng/ml EtBr , and collected spent media samples after 3 d incubation . To measure total homocysteine ( i . e . , both oxidized and reduced ) , 200 μl spent media was reduced ( Magera et al . , 1999 ) by addition of 40 μl DTT ( 0 . 5 M ) , vortexed , centrifuged at 2000×g for 1 min to remove any cell debris , and the supernatant incubated at room temperature for 15 min . 120 μl of the supernatant was extracted by adding 200 μl 0 . 1% formic acid in ACN , vortexing for 15 s , incubating on ice for 30 min , and centrifuging at 13000×g for 20 min . The resulting supernatant was analyzed by LC-MS for homocysteine abundance as above . Homocysteine signal was acquired in SIM mode . To measure serine , 200 μl spent media was pelleted at 2000×g for 1 min to remove cell debris , and 100 μl the supernatant extracted by adding 900 μl 0 . 1% formic acid in 75% ACN: 25% methanol , vortexing for 15 s , incubating on ice for 30 min , and centrifuging at 13000×g for 20 min . The resulting supernatant was analyzed by LC-MS for serine abundance as above . Serine signal was acquired in full scan mode . For single SHMT knockouts , 6×105 T-Rex293 cells in 6-wells were transfected with 300 ng U6-sgRNA PCR product ( Ran et al . , 2013 ) and 1 μg pCas9_GFP ( Addgene #44719 ) using Roche X-tremeGENE 9 . We used the guide sequence CATCTGCAATCTTCCGTAGC for SHMT1 and GCAACCTCACGACCGGATCA for SHMT2 . 2–3 d after transfection , single cells from the top 5% of GFP expressors were deposited by FACS into 96-wells containing 50% spent media and 50% fresh media , supplemented with 200 μM serine and 1:500 normocin ( InvivoGen ) . Single cell colonies were analyzed for SHMT1 and SHMT2 expression by Western blotting; SHMT1 antibodies were Cell Signaling Tech ( Danvers , MA ) #12612 , and SHMT2 antibodies were Thermo Scientific #PA5-32228 . Double SHMT knockouts were generated likewise , except transfections were with 225 ng of each U6-sgRNA PCR product , and FACS sorted cells were grown in media further supplemented with 1x HT ( 100x solution from Thermo Fisher; 16 μM thymidine and 100 μM hypoxanthine final ) . We also repeated SHMT1 and SHMT2 single knockout generation with 1x HT supplementation , and none of the knockout lines so generated were hypoxanthine or formate auxotrophs . Untreated and EtBr-treated SHMT1 KO T-REx-293 cells were seeded in 35 mm plates at 106 and allowed to attach for 1 d . They were then washed with warm PBS and treated with labeling media ( DMEM with 400 uM 13C3-serine , 10% dialyzed FBS ) for 1 hr . Cells were then washed with ice-cold PBS , and metabolites extracted using 750 ul 80% acetonitrile: 20% water . 13C3-serine and 13C1-serine were quantitated using the same procedure as for serine synthesis determination . In follow-up studies , we used 50 uM 13C3-serine on wild type cells . T-REx-293 , 293T , and HeLa cells were seeded at 106; C2C12 cells at 2 . 5×105; MCH58 at 5×105 . Cellular NAD+ and NADH content were determined as previously described ( Sullivan et al . , 2015 ) . We used a hybrid cavitation chamber ( Kristián et al . , 2006 ) / homogenizer ( Mootha et al . , 1997 ) technique to disrupt cells , followed by differential centrifugation to isolate mitochondria ( Mootha et al . , 1997 ) . All steps for mitochondrial isolation were performed at 4°C . Briefly , 2–4 confluent 15 cm plates of T-REx-293 cells were washed with PBS ( room temp ) and scraped into a conical tube . These were pelleted at 600g for 10 min , aspirated , and resuspended in 11 ml IBc ( 200 mM sucrose , 10 mM Tris/MOPS , 1 mM EGTA/Tris , Roche cOmplete protease inhibitor , pH 7 . 4 ) ( Frezza et al . , 2007 ) . The cell suspension was pressurized to 800 psi with nitrogen in a pre-chilled cavitation chamber ( Parr Instruments , Moline , IL ) for 15 min , and rapidly decompressed into a Potter-Elvehjem homogenizer . The cell suspension was then homogenized with 5 strokes at 1000 rpm . This lysate was centrifuged at 600 g to remove nuclei and intact cells . Mitochondria in the supernatant were then pelleted at 8000 g , and the supertatant aspirated . Mitochondria were resuspended in IBc , and the differential centrifugation procedure repeated . After the second round , mitochondria were suspended in 300–400 μl IBc , and quantitated by BCA assay ( Thermo Fisher ) . Mitochondria were then pelleted and resuspended to 5 . 71 mg/ml in experiment buffer EB ( 137 mM KCl , 2 . 5 mM MgCl2 , 10 mM HEPES , 1 mg/ml BSA , pH 7 . 4 ) . Formate production assays were performed in 105 μl of EB with 150 μg mitochondria , 3 mM Pi , 1 mM serine , and 1 mM ADP , at 37°C for 20 min . Formate production was stopped by centrifuging the mixture at 8000 g for 10 min ( 4°C ) and removal of supernatant . Replicate wells of 40 ul supernatant were analyzed using a formate assay kit ( Sigma ) , where one well did not contain enzyme and was used as a background NAD ( P ) H control . Antimycin and rotenone were used at 1 μM , and the corresponding vehicle control was 0 . 1% ( w/v ) DMSO . When drug treatment was used , mitochondria were incubated with drugs for 2 min at 37°C prior to addition of serine , Pi , and ADP . T-REx-293 cells were seeded to 6 cm plates at 2×106 and allowed to attach and grow for 1 d . They were then washed with warm PBS and treated with serine-free DMEM supplemented with 3 mM D-formate ( Sigma ) and 10% dialyzed FBS . After 12 hr treatment , cells were aspirated , washed with ice-cold PBS , and metabolites extracted using 1 ml 80% methanol: 20% water . Deuterated serine and methionine were quantitated as described previously ( Mascanfroni et al . , 2015 ) . Briefly , cell extracts ( 10 μL ) were diluted 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 . LC-MS data were processed and visually inspected using TraceFinder 3 . 1 software ( Thermo Fisher Scientific; Waltham , MA ) . For each experimental condition , T-Rex293 cells were seeded in triplicate , either at 1 . 5×105/well in 12-well plates or at 3×105/well in 6-well plates , in DMEM + 10% dialyzed FBS , and split at 2 d or 3d intervals . Ethidium bromide was applied at 100 ng/ml , antimycin at 500 nM , rotenone at 100 nM , and oligomycin at 500 nM . CCCP was used at 5 μM when alone and 3 μM to rescue oligomycin treatment . On each split , cells were trypsinized , counted , and re-seeded to the original density . We used the cell counts to compute growth curves . Growth rates shown in Figure 5—figure supplement 1a and Figure 5—figure supplement 1c were computed by performing least-squares line fits to individual log-transformed growth traces from days 2–6 . We omitted growth between days 0 and 2 because the drug-induced growth effects did not appear fully developed before day 2 . 293MSR cells were grown identically . U251 cells were grown in triplicate in 6-cm plates at 1 . 2×105 per plate , and rotenone treatment on U251 was at 50 nM . U251 cells were split at 2 d intervals , and the growth rates shown in Figure 5—figure supplement 3 are from days 2–10 in the U251 growth curves and days 0–6 in the 293MSR growth curves . To revive the S . cerevisiae BY4741 strain , cells were plated onto YPD ( rich medium ) plates from frozen glycerol stocks . After 2 days , cells were taken from plates , re-suspended into liquid YPD , and counted . Next , an appropriate amount of cells were taken to inoculate a 3 mL culture of SD +2% Dextrose ( Sunrise Science , San Diego , CA ) at 1x106cells/ml . The resulting 3 mL culture was placed in a New Brunswick Scientific ( Edison , NJ ) model TC-7 roller drum on the fastest rotation until saturated ( 16 hr ) . The cells were then counted and diluted back to 1x106cells/ml in SD-serine-adenine+2% glycerol+2% ethanol with 5 nm Antimycin A . For the serine and adenine additions , 1X corresponds to 85 . 6 mg/L and 21 mg/L , respectively . To monitor growth in each condition , 150 uL of culture was placed in the wells of a 96-well plate and growth curves were done using the Bio Tek ( Winooski , VT ) Synergy H1 multi-mode plate reader . The growth conditions were 30°C with continuous low shaking . OD600 was measured every 15 min for 48 hr . All significance values were computed using Student’s t-test ( two-tailed ) unless otherwise specified . All error bars shown in figures denote standard errors of the mean . n values denote biological replicates , i . e . measurements performed on independent biological samples . | Mitochondria are found within virtually all of our body’s cells and are best known as their power plants . Damaged mitochondria cause many diseases in humans – from rare , inherited metabolic disorders that cause symptoms including muscle weakness and developmental problems , to age-related diseases such as diabetes and Parkinson’s disease . How does mitochondrial damage lead to such a variety of symptoms and conditions ? To answer this question , researchers must understand how cells respond to and compensate for such damage . To mimic mitochondrial failure , Bao et al . reduced the amount of DNA in the mitochondria of human cells and observed that this caused the cells to accumulate more of an amino acid called serine . Further investigation showed that this accumulation comes in part from cells producing more serine , and that a protein called Activating Transcription Factor 4 is responsible for increasing the expression of the genes needed to produce serine in the cells . Bao et al . also found that damaged mitochondria are less able to consume serine to produce a compound called formate , which is a precursor for DNA building blocks . If cells cannot acquire enough extra serine to compensate for this inefficiency , they cannot produce some of the building blocks required to make DNA and other critical compounds in the cell . Supplementing the cells with formate or the DNA building blocks enabled the cells to recover , which suggests that formate supplements may help to treat some mitochondrial disorders . At a higher level , these results suggest that the mitochondrion’s role as a major chemical factory in the cell , and not just as the power plant , may also contribute to disease when the mitochondria are broken . Further work is now needed to investigate how cells know to turn on Activating Transcription Factor 4 when their mitochondria are damaged . It also remains to be discovered whether this reduces or exacerbates the symptoms of mitochondrial disease . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"computational",
"and",
"systems",
"biology"
] | 2016 | Mitochondrial dysfunction remodels one-carbon metabolism in human cells |
Most neurogenesis in the mammalian brain is completed embryonically , but in certain areas the production of neurons continues throughout postnatal life . The functional properties of mature postnatally generated neurons often match those of their embryonically produced counterparts . However , we show here that in the olfactory bulb ( OB ) , embryonic and postnatal neurogenesis produce functionally distinct subpopulations of dopaminergic ( DA ) neurons . We define two subclasses of OB DA neuron by the presence or absence of a key subcellular specialisation: the axon initial segment ( AIS ) . Large AIS-positive axon-bearing DA neurons are exclusively produced during early embryonic stages , leaving small anaxonic AIS-negative cells as the only DA subtype generated via adult neurogenesis . These populations are functionally distinct: large DA cells are more excitable , yet display weaker and – for certain long-latency or inhibitory events – more broadly tuned responses to odorant stimuli . Embryonic and postnatal neurogenesis can therefore generate distinct neuronal subclasses , placing important constraints on the functional roles of adult-born neurons in sensory processing .
The adult central nervous system has been long believed to be incapable of self-regeneration . Five decades ago , however , pioneering studies revealed the existence of well-defined neurogenic niches in the brain of adult rodents ( Altman , 1962 ) . Such neurogenic zones are small , spatially defined and give rise to two broad neuronal populations: hippocampal dentate granule cells , and interneurons in the olfactory bulb ( OB ) ( Lledo et al . , 2006 ) . These newly generated cells are believed to bring unique properties to existing networks , largely by virtue of the specialised functional and plastic features associated with their transient immature status ( e . g . Carleton et al . , 2003; Ge et al . , 2007; Gu et al . , 2012; Livneh et al . , 2014; Marín-Burgin and Schinder , 2012; Nissant et al . , 2009; Schmidt-Hieber et al . , 2004 , but see Sailor et al . , 2016 ) . Once fully mature , though , the functional properties of adult-generated neurons in both the hippocampus and olfactory bulb often closely match those of their developmentally generated neighbours ( e . g . Carleton et al . , 2003; Grubb et al . , 2008; Laplagne et al . , 2006; Marín-Burgin and Schinder , 2012; Nissant et al . , 2009 , but see Valley et al . , 2013 ) . Does this mean , then , that within broad classes of neuron – for example , dentate granule cells , or OB granule cells – embryonic and postnatal neurogenesis always produce fundamentally similar cell types ? Among the wider population of adult-generated OB cells is a heterogeneous group of inhibitory neurons situated in the structure’s glomerular layer , whose main role is to modulate the earliest stages of sensory information processing ( Alonso et al . , 2012; Fukunaga et al . , 2014; Grubb et al . , 2008; Livneh et al . , 2014; Livneh and Mizrahi , 2012 ) . Different subclasses of glomerular layer interneuron can be identified by their specific expression of calcium-binding proteins ( Kosaka and Kosaka , 2011 ) , while another major subclass is identified by its ability to co-release both GABA and dopamine ( Borisovska et al . , 2013; Vaaga et al . , 2017 ) . These dopaminergic ( DA ) interneurons can be generated via adult neurogenesis ( Adam and Mizrahi , 2011; Bonzano et al . , 2016; De Marchis et al . , 2007 ) , and the survival of postnatally generated DA cells is activity-dependent ( Bonzano et al . , 2014 ) . Moreover , adult-born DA cells have been shown to contribute to specific olfactory behaviours ( Lazarini et al . , 2014 ) . In recent years , the manner in which resident and adult-generated DA neurons contribute to olfactory processing has been widely studied , and there is now an accumulating - and sometimes contrasting - body of evidence on the role played by DA cells in glomerular circuits . A wide spectrum of functions has been proposed for these cells , involving either local or broadly distributed actions within the glomerular layer , and roles as diverse as modulating release from olfactory sensory neuron terminals , signal normalisation , contrast enhancement , and temporal decorrelation ( Banerjee et al . , 2015; Cavarretta et al . , 2016; Economo et al . , 2016; Liu et al . , 2013; Mainland et al . , 2014; Pignatelli and Belluzzi , 2017; Roland et al . , 2016; Vaaga et al . , 2017 ) . The complexity and sometimes mutually exclusive nature of such functions – especially in relation to spatial connectivity – make it unlikely that a single class of interneuron could perform them all . Indeed , morphological variability has been demonstrated among OB DA neurons and has been linked to their time of birth ( Halász et al . , 1981; Kiyokage et al . , 2010; Kosaka and Kosaka , 2007; Kosaka et al . , 2008; Kosaka and Kosaka , 2009; Kosaka and Kosaka , 2011; Kosaka and Kosaka , 2016; McLean and Shipley , 1988; Pignatelli and Belluzzi , 2017; Pignatelli et al . , 2005 ) . However , no discrete features demarcating distinct OB DA subpopulations have yet been identified . More importantly , nothing is currently known regarding the functional properties of putative OB DA subtypes . Are there physiological differences between embryonically generated and adult-born DA cells ? And might such differences start to account for the various functional roles ascribed to this cell type in sensory processing ? Here , we build on previous work in vitro ( Chand et al . , 2015 ) , to show that different classes of OB DA neuron in vivo can be clearly distinguished based on the presence or absence of an axon and its key subcellular specialisation , the axon initial segment ( AIS ) . AIS-positive DA cells are larger , with broader dendritic arborisations , and are exclusively born in early embryonic development . Postnatally generated DA cells , in contrast , are all small and anaxonic . Crucially , these morphological and ontological distinctions also map onto clear functional differences in both cellular excitability and odorant response properties in vivo , strongly constraining the potential role of adult-born DA cells in sensory processing .
To investigate the presence of an axon initial segment ( AIS ) in DA cells we performed immunohistochemistry on fixed slices of the olfactory bulb of juvenile ( P28 ) wild-type C57/Bl6 mice . We identified DA cells by labelling them with an antibody against tyrosine hydroxylase ( TH ) , the rate-limiting enzyme in the biosynthesis of dopamine . For AIS identification we stained for ankyrin-G , the master AIS organising molecule ( AnkG , Figure 1A ) ( Hedstrom et al . , 2008; Jenkins et al . , 2015; Zhou et al . , 1998 ) . While for most TH-positive cells we could not detect AnkG label on any of their processes , we identified a subset of large DA neurons that possessed a clear AnkG-positive AIS ( Figure 1A , middle and right panel; Kosaka et al . , 2008 ) . Like their midbrain DA ( González-Cabrera et al . , 2017; Meza et al . , 2018 ) and cultured ( Chand et al . , 2015 ) counterparts , the AISs of OB DA neurons were often thin , short and located rather distally away from the soma . Along with the dense and highly interwoven meshwork of TH-positive processes in the glomerular layer ( GL ) , this made AIS-positive cells difficult to identify . Nevertheless , a lower-bound estimate ( see Materials and methods ) suggests they comprise at least 2 . 5% of the overall OB DA population . Soma size quantification revealed these AIS-containing DA neurons to be morphologically distinct . As shown in Figure 1B , the soma area distribution of the general bulbar TH+ population is clearly not unimodal ( blue distribution ) : most DA cells are relatively small ( peak 55 μm2 ) , but there is a distinct minority that are significantly larger ( peak 140 μm2 ) ( Kosaka and Kosaka , 2007; McLean and Shipley , 1988; Pignatelli et al . , 2005 ) . In contrast , performing a similar analysis solely for TH+/AnkG+ cells ( i . e . DA neurons with an AIS ) produced a unimodal distribution centred on the large-cell peak of the full population curve ( Figure 1B , magenta line; peak 137 μm2 ) . Large AIS-positive cells therefore represent a distinct sub-population of OB DA neurons . These large , AIS-positive DA neurons are also located in a specific sub-region of the GL . Dividing the GL into four sub-laminae ( Figure 1A; see Materials and methods ) revealed the overall TH-positive population to be concentrated in the mid-GL ( Figure 1C ) . AIS-positive DA neurons , however , were mostly found in the lower portions of the GL towards the external plexiform layer ( EPL ) border , with very little presence in the upper or mid-GL ( Figure 1C; Liberia et al . , 2012 ) ; effect of sub-lamina ×cell type in two-way repeated-measures ANOVA , F3 , 66 = 35 . 47 , p<0 . 0001; post-hoc Sidak’s test between cell types , upper-GL , p=0 . 014; mid-GL , p<0 . 0001; lower-GL , p<0 . 0001; EPL border , p=0 . 98; n = 24 slices from N = 3 mice ) . The AIS is crucial for the maintenance of axo-dendritic neuronal polarity ( Hedstrom et al . , 2008 ) , and is often employed as an indicator of axonal identity ( e . g . Watanabe et al . , 2012 ) , so does the absence of an AIS in the majority of small DA neurons mean that these cells do not possess an axon ? Addressing this question required us to be able to identify and follow all of a given cell’s individual processes . We therefore achieved sparse label of individual OB DA neurons , either by injecting floxed GFP-encoding viruses ( either AAV or RV::dio ) in embryos or neonates from VGAT-Cre or DAT-Cre reporter lines , or by electroporating GFP-encoding plasmid DNA in wild-type neonates ( see Materials and methods ) . The dopaminergic phenotype of the infected neurons was confirmed by immunohistochemical label for TH . We then adopted a dual strategy for axon identification . First – as a positive control – we confirmed that while the AnkG-positive processes of large AIS-containing DA cells co-localised with the axonal marker TRIM-46 ( Figure 2A;van Beuningen et al . , 2015 ) , this axonal marker was entirely absent from the processes of small OB DA neurons ( Figure 2B; n = 10 , N = 3 , average soma area 58 μm2 ) . Second – as a negative control – we analysed the expression of the dendritic marker MAP-2 ( Kosik and Finch , 1987; Rolls and Jegla , 2015; van Beuningen et al . , 2015 ) . DA cells with an AIS express MAP-2 in all processes , even in the proximal axon ( Figure 2C ) . However , as reported for other cell types ( Gumy et al . , 2017; van Beuningen et al . , 2015 ) , this proximal axonal MAP-2 expression fades where AnkG expression begins , and MAP2 is absent from the post-AnkG portion of the axon ( Figure 2C ) . Conversely , AIS-negative DA neurons express MAP-2 along the entire length of all their processes ( Figure 2D; n = 10 , N = 3 , average soma area 49 μm2 ) . These data strongly suggest that the presence of an AIS is indicative of axonal identity in OB DA cells , and that the small TH-positive neurons that lack an AIS are truly anaxonic . Sparse labelling of individual OB DA neurons also allowed us to investigate their dendritic morphology ( Figure 3A ) , and this again revealed clear differences between AIS-positive and AIS-negative subtypes . Small , anaxonic DA neurons had limited dendritic arborisations that ramified across a small region of the glomerular layer ( Figure 3B , C , E , F ) . By contrast , the dendrites of large , axon-bearing DA cells were much more broadly spread ( Figure 3D , E , F ) . Despite considerable cell-to-cell morphological variability within each sub-class ( Figure 3E , F ) , quantitative Sholl analysis ( see Materials and methods ) revealed highly significant cell-type differences on multiple dendritic parameters ( Table 1; Figure 3E; effect of cell type in mixed model ANOVA analysis of Sholl distributions , F1 , 36 = 5 . 30 , p=0 . 027 ) . This is all the more striking given the thin OB slices necessary for AnkG label , and the likely resulting underestimation of glomerular layer ramification by AIS-positive DA neurons . Glomerular layer interneurons in the OB , including TH-positive DA neurons , belong to the highly restricted group of neuronal types capable of regenerating throughout life via adult neurogenesis ( Betarbet et al . , 1996; Bonzano et al . , 2016; De Marchis et al . , 2007; Winner et al . , 2002 ) . This prolonged neurogenic capacity is often considered to be a universal and hallmark feature of these inhibitory interneurons ( e . g . Liu et al . , 2013 ) . However , data from birthdating experiments suggest that – at least on the basis of soma size – OB DA neurons are not homogeneous in their time of generation ( Kosaka and Kosaka , 2009 ) . This prompted us to ask whether the two morphological subtypes of AIS-positive and AIS-negative OB DA neuron also differ developmentally . To address this question , we performed classic pulse-chase birthdating experiments . We injected pregnant mice with the thymidine analogue bromodeoxyuridine ( BrdU ) at different gestational days starting from embryonic day ( E ) 11 , when the nascent olfactory bulb has begun to appear . We then collected tissue from the progeny once they reached one month of age , and from the mothers themselves to analyse adult-generated neurons ( Figure 4A ) . We also labelled neonatally generated bulbar interneurons via postnatal electroporation of GFP-encoding plasmid DNA injected into the lateral ventricles at P1 ( Figure 4A ) . Cells were then immunostained for BrdU or GFP , along with TH and – because of the histological processing necessary for BrdU detection – a more robust but as yet mysterious marker of the AIS: the unidentified microtubule-associated protein labelled by the ‘pIκBα’ antibody ( Buffington et al . , 2012 ) . For each injection time point , we analysed all cells that were both BrdU and TH positive , measuring soma area , and noting the presence or absence of an AIS . The results , presented in Figure 4B–H , clearly show that with increasing birth age the soma area distributions lose their right-end tail , indicating that the large DA neurons are mostly born during early development ( Kosaka and Kosaka , 2009 ) . Interestingly , soma area in adult-born OB DA neurons ( Figure 4H ) appeared small compared to overall DA population distributions at earlier ages , an effect that was not accounted for by either histological processing differences or cell maturation state , and which might indicate a potential specialization of this DA subpopulation that warrants investigation in future studies . The developmental distinction between AIS-positive and AIS-negative DA cells , however , was even clearer . AIS-positive DA neurons were exclusively born in embryonic development , with a clear peak in their generation at E11-12 ( Figure 4B , C , H ) and only very few being produced between E13 and E18 ( Figure 4D–F , H ) . We did not find a single neonatal- or adult-born OB DA neuron that possessed an AIS ( Figure 4G , H ) . To fully convince ourselves that adult-born DA neurons never possess an AIS , we had to rule out the possibility that these cells took longer than 1 . 5 months to fully mature . We therefore collected tissue from adult BrdU-injected mice after a prolonged chase period of 4 months ( Figure 5A ) . We found very few double-labelled BrdU+/TH +cells ( Figure 5B; n = 15 cells from N = 2 mice ) , and all were AIS-negative . The absence of an AIS in adult-generated cells is therefore unlikely due to insufficient maturation time , and instead likely reflects a fundamental characteristic of this OB DA subtype . These data strongly suggest that prolonged neurogenic capabilities are not a widespread property of bulbar DA neurons . Instead , adult neurogenesis is restricted to the axonless subpopulation , while large AIS-bearing cells are only born during early developmental stages . This finding raised two immediate questions concerning the production and longevity of AIS-positive neurons: 1 ) is there a preponderance of large AIS-positive DA cells in the neonate ? And 2 ) do embryonically generated AIS-positive neurons persist throughout life ? To address the first question , we collected tissue from newborn pups ( P0 ) and quantified the soma area of TH-labelled DA neurons ( Figure 6A ) . We indeed found a right-shifted distribution of larger DA neurons at this early postnatal timepoint ( Figure 6B , Kolmogorov-Smirnov test D = 0 . 3581 , p<0 . 01; McLean and Shipley , 1988 ) . We also obtained a slightly higher lower-bound estimate for the prevalence of AIS-positive DA cells at P0 ( Figure 6C; see Materials and methods ) , at ~6% of the overall TH-positive population . In addition , AIS-positive neurons at P0 already have large soma areas ( mean ± SEM 98 ± 5 μm2 , n = 16 , N = 2; Figure 6D ) . To address the second question concerning the longevity of AIS-positive DA neurons , we employed a prolonged pulse-chase birthdating protocol . Mice were injected with BrdU at E12 , then left until 6 months of age , when we collected tissue and looked for AIS-positive DA neurons ( Figure 7A ) . As shown in Figure 7B , we were still able to find DA neurons born at E12 in the bulb of these fully adult mice . All these neurons were large and the overwhelming majority carried an AIS . When we compared these adult animals to littermates that had also been injected at E12 but perfused at P28 , we found that E12-born DA neurons in adult animals were larger ( 91 ± 4 μm2; Figure 7C , D ) . Moreover , while in 1-month-old animals TH+/E12-BrdU+ cells were relatively abundant but only 25% of them were AIS-positive , in adult mice TH+/E12-BrdU+ cells were rare but over 90% of them possessed an AIS ( Figure 7E ) . These findings strongly suggest that , of the mixed pool of DA neurons born at E12 , the initially abundant AIS-negative cells are turned over at some point before 6 months of age , to be substituted by postnatallyborn AIS-negative neurons . In contrast , embryonicallygenerated , large AIS-positive cells can persist throughout adult life . Our morphological and developmental analyses revealed a clear distinction between - on the one hand – embryonically born , large and widely branching DA neurons with an axon and AIS , and – on the other – lifelong-generated , small and locally ramifying anaxonic DA cells that do not have an AIS . But do these marked ontological and structural differences also translate into functional heterogeneity ? To test this hypothesis , we first performed whole-cell current-clamp electrophysiological recordings on DA neurons in acute ex vivo OB slices . We visualised DA neurons by crossing the dopaminergic reporter line DAT-Cre ( Bäckman et al . , 2006 ) with a floxed tdTomato ( tdT ) reporter line ( Madisen et al . , 2010 ) . In the resulting DAT-tdT mice , the majority of TH+ DA cells also expressed tdT ( 90 ± 4% of all TH+ neurons were also tdT+; n = 369 , N = 3; Figure 8A ) . The rare TH+ neurons that lacked tdT fluorescence tended to be large ( Figure 8B ) , suggesting that while our genetic labelling strategy comprehensively identified small , anaxonic DA neurons , it under-represented the large , AIS-positive DA subtype . Indeed , co-label for AnkG revealed that only 67% of TH+/AIS+ DA neurons also expressed tdT ( Figure 8B , C ) . However , tdT expression did not appear to reveal any further subdivision amongst the large , AIS-positive DA cells , because we found no difference in soma size between TH+/AIS+/tdT+ and TH+/AIS+/tdT- neurons ( tdT+ , mean ± SEM 118 ± 10 µm2; tdT- , 126 ± 5 µm2; Welch’s corrected t21 . 12 = 0 . 66 , p=0 . 52 ) . So , although DAT-tdT mice do not comprehensively reveal all bulbar DA neurons , visually targeting tdT-positive cells for electrophysiological recordings ( Figure 8C ) still enables functional comparisons to be made between AIS-positive and AIS-negative DA cell types . Post-recording survival of bulbar DA cells for morphological or immunohistochemical analysis is notoriously difficult ( A . Pignatelli , personal communication ) ; this meant that we could not classify our recorded neurons as AIS-positive or AIS-negative on the basis of AnkG staining . Instead , we relied on a functional indicator of AIS presence: non-somatic action potential ( AP ) initiation . In phase-plane plots of single spikes fired in response to 10 ms somatic current injection , the site of AP generation can be inferred from the shape of the initial , rising component of the spike waveform ( Bean , 2007; Chand et al . , 2015; Coombs et al . , 1957; Jenerick , 1963; Khaliq et al . , 2003; Shu et al . , 2007 ) . While a smooth , monophasic phase plane plot is indicative of AP initiation at the somatic recording site , cells that initiate spikes at a distance from the electrode location – almost always at the AIS ( Bender and Trussell , 2012; Coombs et al . , 1957; Foust et al . , 2010; Kole et al . , 2007; Palmer and Stuart , 2006 ) - display a distinctive biphasic , or ‘double-bumped’ phase plane plot waveform ( Figure 8E ) . We therefore divided our recorded DAT-tdT+ cells into monophasic and biphasic groups ( see Materials and methods; Figure 8—figure supplement 1 ) , which should be largely representative of AIS-negative and AIS-positive DA neurons , respectively . Indeed , we found that biphasic neurons were significantly larger than their monophasic counterparts ( Figure 8F; Table 2; [Chand et al . , 2015] ) . We also identified several differences in intrinsic excitability between monophasic and biphasic DAT-tdT cells . Biphasic neurons generated single APs in response to lower amplitude somatic current injection , and initiated those APs more rapidly ( Figure 8F; Table 2 ) . When induced to fire repeatedly in response to longer lasting 500 ms somatic current injections of increasing intensity , biphasic cells displayed a linear input-output curve . Conversely , monophasic cells could not produce such a linear increase in spike number and soon reached a firing plateau ( Figure 8G ) . This resulted in monophasic cells having a significantly lower slope of their input-output curve , and a significantly lower maximum number of fired APs ( Figure 8H; Table 2 ) . While these differences in intrinsic excitability are certainly consistent with reported functional characteristics of AIS-positive versus AIS-negative neurons ( Chand et al . , 2015; Zhou et al . , 1998; Zonta et al . , 2011 ) , we cannot rule out contributions from other , non-AIS-dependent factors ( Baranauskas et al . , 2013; Eyal et al . , 2014; Pignatelli et al . , 2009 ) . Nevertheless , and regardless of their underlying cause , these physiological differences point to significantly greater intrinsic excitability in the biphasic , presumptive AIS-possessing DA subpopulation . Finally , we asked whether the above measures from our whole-cell recordings could be reliably used to classify DAT-tdT neurons as belonging to either the biphasic/AIS-positive or the monophasic/AIS-negative subtype . Applying principal component analysis ( PCA; see Materials and methods ) to the five variables that differed significantly between mono- and biphasic DA cells generated primary and secondary component scores for each neuron that , when plotted against each other , revealed clear clustering by cell type ( Figure 8I ) . Furthermore , using a k-means classification approach with the same data ( see Materials and methods ) we were able to assign our recorded cells to either the mono- or biphasic group with 85% accuracy . This suggests that , although there is considerable overlap in the functional properties of different subclasses of OB DA neuron , when taken together those properties reveal a significant distinction between putative AIS-positive and putative AIS-negative cell types . We next asked whether the morphological and physiological differences between the two subtypes of OB DA neuron are associated with distinct sensory response properties in vivo . Do different types of OB DA cell respond differently to olfactory stimuli ? To address this question , we employed a conditional mouse line in which the Cre-dependent Ca2+ indicator GCamP6s was selectively expressed in OB DA neurons under the control of the DAT promoter ( see Materials and methods ) . We could then characterise the sensory response properties of these cells by monitoring changes in GCaMP fluorescence while animals were presented with a panel of eight odour stimuli ( see Materials and methods; Figure 9A; Kapoor et al . , 2016 ) . Given the current lack of a reliable in vivo live AIS marker , we classified DA neurons in these experiments using the proxy measure of soma size – this can be readily measured in live neurons , and is consistently associated with AIS-positive or AIS-negative identity across multiple datasets ( Figures 1 , 3 and 8 ) . Our upper bound for cells classed as ‘small’ was 70 µm2 , taken from the mean soma size of confirmed AIS-positive DA cells ( Figure 1B , magenta distribution ) minus two standard deviations ( i . e . 136 . 7–2 × 33 . 2 µm2 ) . Under an assumption of normality , this cutoff excludes all but the smallest 2 . 5% of AIS-containing neurons . Similarly , our lower bound for cells classed as ‘big’ was 99 µm2 , taken from the mean soma area of confirmed AIS-negative DA neurons labelled via GFP electroporation at P1 ( Figure 4H ) plus two standard deviations ( i . e . 65 . 6 + 2×16 . 6 µm2 ) . Again assuming normality , this cutoff excludes all but the biggest 2 . 5% of AIS-lacking cells . We then analysed the odorant response properties of small/putative AIS-negative ( n = 594 ) and big/putative AIS-positive ( n = 622 ) GCaMP+ cells imaged in 13 mice . It immediately became apparent that different forms of odour-evoked responses could occur in these neurons . In many cases , a given odorant stimulus produced a relatively rapid increase in GCaMP fluorescence that then decayed back towards baseline – these responses , which we termed ‘early excitatory’ events , were the most prevalent form of odour-evoked signal in our DAT-GCaMP neurons ( at least one early excitatory response was observed in 631/1216 = 52% of cells ) . In other cases , stimuli produced an increase in GCaMP intensity that had a delayed onset and peaked late in a given recording sweep – these ‘late excitatory’ events were readily and objectively distinguishable from early excitatory events ( see Materials and methods; Figure 9C , D ) and were less frequent in our sample ( 245/1216 = 20% of cells had at least one late excitatory response ) . Finally , we observed reasonably common examples of decreased GCaMP fluorescence upon odorant presentation . These ‘inhibitory’ events ( at least one seen in 296/1216 = 24% of cells ) usually had delayed onset , and were perhaps detectable because of the characteristically high spontaneous activity levels in OB DA neurons ( Chand et al . , 2015; Pignatelli et al . , 2005; Puopolo et al . , 2005 ) . Although both late excitatory and inhibitory response types were unusual in their long peak latencies ( but see similar long-latency excitatory responses in OB DA cells in Banerjee et al . , 2015 ) , their Figure 2B ) , and although they were more variable than early excitatory responses ( Figure 9E , F , G ) , analysis of responses to individual stimulus presentations revealed them nevertheless to be reliable odour-evoked events . Peak latencies for all response types had coefficients of variation that were significantly less than one , indicative of non-random event timing across individual stimulus repeats ( Figure 9F; early excitatory , mean ±SEM 0 . 28 ± 0 . 0063 , Wilcoxon test vs 1 , W = −198733 , p<0 . 0001; late excitatory 0 . 40 ± 0 . 017 , W = −29773 , p<0 . 0001; inhibitory 0 . 38 ± 0 . 015 , W = −42085 , p<0 . 0001 ) . In addition , no response type was generated by spurious , one-off fluctuations in fluorescence – mean amplitudes for the two stimulus repeats that produced the weakest responses were still significantly greater than one standard deviation above baseline ( Figure 9G; early excitatory , mean ±SEM 5 . 38 ± 0 . 26 , Wilcoxon test vs 1 , W = 198036 , p<0 . 0001; late excitatory 2 . 29 ± 0 . 20 , W = 17963 , p<0 . 0001; inhibitory 1 . 28 ± 0 . 060 , W = 12954 , p<0 . 0001 ) . Crucially , neither of these measures of response reliability differed between small and big DA neurons , showing that late excitatory and inhibitory events were not only reliable per se , but were also just as reliable in both cell types ( peak time CV , late excitatory , fixed effect of cell-type in multilevel ANOVA , F1 , 232 = 1 . 89 , p=0 . 17; inhibitory , F1 , 289 = 0 . 12 , p=0 . 75; mean amplitude of 2 weakest responses , late excitatory , F1 , 212 = 0 . 83 , p=0 . 36; inhibitory , F1 , 177 = 0 . 19 , p=0 . 67 ) . Most response types occurred in isolation , although we did see some examples of combined excitatory-inhibitory responses ( at least one seen in 76/1216 = 6% of cells ) . Overall , 817/1216 = 67% of imaged DAT-GCaMP+ cells displayed at least one response type evoked by at least one of the eight odour stimuli we used . All forms of odorant-evoked GCaMP response were observed in both big and small OB DA cell types . There were , however , some significant differences in their relative prevalence in the two neuronal populations . Paired , within-animal comparisons for the three major response types across the 13 mice in our sample revealed no significant differences in the proportions of small vs big DA neurons that displayed at least one odorant-evoked fast excitatory ( Figure 9H; small cells , mean ±SEM 46 ± 7%; big cells 51 ± 6%; paired t-test , t12 = 1 . 10 , p=0 . 29 ) , or slow excitatory ( Figure 9H; small cells , 19 ± 3%; big cells 23 ± 3%; paired t-test , t12 = 1 . 15 , p=0 . 27 ) response . However , we did see a significantly higher proportion of inhibitory-responding neurons amongst the big cell population ( Figure 9H; small cells , 16 ± 4%; big cells 26 ± 6%; Wilcoxon test , W13 = 62 , p=0 . 012 ) . To interrogate sensory stimulus selectivity further , we calculated a simple ‘tuning index’ ( TI ) for each response type in each cell , from the sum of all stimuli producing a significant change in GCaMP fluorescence ( see Materials and methods ) . Cells with higher TI values responded to more odorants in our 8-stimulus panel . Although we acknowledge that this cannot represent a comprehensive description of tuning across all of odour space , this measure nevertheless allowed us to detect differences in response selectivity to a select group of odorant stimuli known to activate broad regions of the dorsal OB ( Livneh et al . , 2014; Rokni et al . , 2014 ) . In line with previous observations ( Banerjee et al . , 2015 ) , we observed broad representations of odours in the responses of OB DA neurons ( Figure 10A ) . The mean TI value for all excitatory responses ( early + late combined ) was 1 . 99 across all neurons in our sample , rising to 3 . 29 within the subset of neurons that displayed at least one excitatory response . Overall , this broad tuning was shared by both big and small OB DA sub-populations . However , we did observe significant cell-type-dependent differences in odour selectivity for particular response types . Importantly , we found only very weak correlations between TI measures calculated for the three major forms of odour-evoked response ( early excitatory vs late excitatory , Spearman r = 0 . 018 , p=0 . 53; early excitatory vs inhibitory , r = 0 . 13 , p<0 . 0001; late excitatory vs inhibitory , r = 0 . 030 , p=0 . 30; n = 1216 in all cases ) , suggesting that TI values for early excitatory , late excitatory and inhibitory events represent rather independent measures of tuning for distinct types of response produced by glomerular layer circuitry . To compare these TI measures between cell types , we needed powerful statistical tests that could leverage the large numbers of imaged neurons in our dataset whilst accounting for significant across-animal variability ( see Materials and methods; Figure 10B , C ) . We therefore employed multilevel ANOVA analyses , where TI values from individual cells were compared between small versus big cell populations nested in animal subjects ( see Materials and methods; [Aarts et al . , 2014] ) . Using this approach , we found no effect of cell type on early excitatory TI values ( Figure 10B; fixed effect of cell-type in multilevel ANOVA , F1 , 1180 = 2 . 04 , p=0 . 15 ) . For the most prevalent form of odour-evoked response , then , tuning was strikingly similar in small and big OB dopaminergic neurons . For both late excitatory and inhibitory response types , though , the effect of cell type on TI was significant ( Figure 10B; late excitatory , F1 , 1214 = 5 . 58 , p=0 . 018; inhibitory , F1 , 1143 = 6 . 92 , p=0 . 009 ) , with big cells possessing consistently larger TI values on a mouse-by-mouse basis . When responding to odorant stimuli with late excitatory or inhibitory events , therefore , big OB DA cells are significantly more broadly tuned than their small-soma neighbours . Could this broader tuning in big OB DA neurons be explained by larger , more readily detectable odour-evoked responses in this cell type ? Actually , measures of event amplitudes revealed the opposite to be the case: big cells had significantly weaker responses to odorant stimuli , a highly significant effect that held across all response types ( Figure 10C; early excitatory , fixed effect of cell-type in multilevel ANOVA , F1 , 626 = 6 . 58 , p=0 . 011; late excitatory , F1 , 244 = 4 . 50 , p=0 . 035; inhibitory , F1 , 295 = 32 . 69 , p<0 . 0001 ) . Despite their higher intrinsic excitability ( Figure 8 ) , big , putative AIS+ DA neurons therefore do not display stronger responses to sensory stimuli in vivo . This unexpected effect may be because of fundamental differences between sensory stimulation in vivo versus direct electrical stimulation in vitro , or it may be due to cell-type differences in synaptic connectivity , or in the modulation of intrinsic properties in the intact OB . Additionally , it could be related to another feature of in vivo GCaMP activity: baseline fluorescence . Resting fluorescence was significantly higher in big cells ( fixed effect of cell-type in multilevel ANOVA , F1 , 658 = 12 . 00 , p=0 . 001 ) , while baseline noise was significantly lower in this cell type ( F1 , 1211 = 41 . 84 , p<0 . 0001 ) , and both measures , especially noise , correlated strongly with all response amplitude measures ( baseline fluorescence vs early excitatory amplitude , Spearman r = −0 . 49 , p<0 . 0001 , n = 612; vs late excitatory amplitude , r = −0 . 44 , p<0 . 0001 , n = 235; vs inhibitory amplitude , r = −0 . 58 , p<0 . 0001 , n = 282; baseline noise vs early excitatory amplitude , r = 0 . 67 , p<0 . 0001 , n = 631; vs late excitatory amplitude , r = 0 . 72 , p<0 . 0001 , n = 245; vs inhibitory amplitude , r = 0 . 83 , p<0 . 0001 , n = 296 ) . The increased buffering capacity associated with higher resting GCaMP levels ( Svoboda et al . , 1999 ) could therefore lead to dampened response amplitudes in big cells . Additionally , lower spontaneous fluctuations in resting activity could allow big OB DA neurons to significantly respond to odorant stimuli with lower amplitude events . However , these cell-type distinctions in baseline activity cannot account for the differences in response selectivity between big and small cell populations ( Figure 10B ) . Not only did we see identical big versus small cell selectivity for early excitatory events when baseline differences might be expected to influence tuning across all response types , we also observed only weak and inconsistent correlations with the different TI measures for both baseline fluorescence and noise ( baseline fluorescence vs early excitatory TI , Spearman r = −0 . 05 , p=0 . 07; vs late excitatory TI , r = 0 . 15 , p<0 . 0001; vs inhibitory TI , r = 0 . 13 , p<0 . 0001; n = 1164 in all cases; baseline noise vs early excitatory TI , r = 0 . 075 , p=0 . 0085; vs late excitatory TI , r = −0 . 20 , p<0 . 0001; vs inhibitory TI , r = −0 . 17 , p<0 . 0001; n = 1216 in all cases ) . Finally , we wondered whether the prolonged timecourse and therefore truncated response profiles of late excitatory and inhibitory responses ( e . g . Figure 9C , Figure 10A ) might have contributed to the functional differences observed for these event types between big and small OB dopaminergic cells . However , accurate estimates of decay kinetics found no cell-type differences in either response type ( see Materials and methods , Figure 10—figure supplement 1 ) . Overall in terms of odorant response properties , therefore , big/putative AIS-positive and small/putative AIS-negative OB DA neurons differ significantly in: 1 ) the broader selectivity of big cells for specific odour-evoked response types; and 2 ) the higher resting fluorescence , lower baseline noise , and smaller response amplitudes of big cells . Moreover , these two major functional features appear to be largely independent of each other .
An absolutely crucial step in understanding information processing in any neuronal network is to build an accurate classification of its component parts ( Zeng and Sanes , 2017 ) . Cell-type identity – as determined by ontology , gene expression , morphology , connectivity , and/or physiology – is intimately linked to the functional role that any neuron can play in a given circuit . It is therefore no surprise that in recent attempts to model realistic network operations , a great deal of effort has been spent delineating just how many component parts those networks contain . In different regions of the mammalian brain , we now have comprehensive descriptions of cell-type diversity with regards to , for instance , gene expression ( e . g . Romanov et al . , 2017; Tasic et al . , 2016; Zeisel et al . , 2015 ) , neuronal morphology ( e . g . Cerminara et al . , 2015; Parekh and Ascoli , 2015 ) , synaptic connectivity ( e . g . Morgan et al . , 2016 ) , and sensory response properties ( e . g . Baden et al . , 2016 ) , as well as combinatorial cellular-level identification schemes that multiplex several levels of description ( e . g . Fuzik et al . , 2016; Markram et al . , 2015; Sanes and Masland , 2015 ) . These studies show that broad cell-type distinctions must be supplemented by fine-scale subdivisions within different cell types in order to fully understand network function . Such classification schemes are no less vital in our understanding of information processing in olfactory bulb circuits , where a uniquely modular topographic organisation of sensory inputs , coupled with the constant remodelling associated with both peripheral and central adult neurogenesis , promises novel insight into the way the brain interprets and adapts to the outside world . However , our current understanding of functional diversity amongst neuronal populations in the olfactory bulb is far from complete . In glomerular layer circuits – the first networks to process sensory information arriving from the periphery – there is at least broad consensus on the division of juxtaglomerular neurons into excitatory and inhibitory cell types: glutamatergic , vGlut-expressing external tufted cells are , on the whole , readily distinguished from their GABA-positive interneuron neighbours ( Hayar et al . , 2004 , but see Tatti et al . , 2014 ) . Furthermore , amongst those GABAergic interneurons are three neurochemically-distinct subpopulations , distinguishable ( at least in mouse ) by their non-overlapping expression of calretinin , calbindin , and tyrosine hydroxylase ( Kosaka and Kosaka , 2007 ) . However , although it has long been recognised that this latter group of TH-positive OB DA neurons are highly heterogeneous ( Davis and Macrides , 1983; Halász et al . , 1981; Kosaka and Kosaka , 2007; Pignatelli et al . , 2005 ) , there has been significant disagreement as to the precise nature of cell sub-type identity within this population . There is as yet no definitive classification of OB DA neurons , even though such a scheme is vital for our understanding of sensory processing functions in glomerular circuits . Two major differing approaches to classifying OB DA neurons are currently under dispute . In the first , morphological considerations , especially the fact that many OB DA neurons spread their processes across more than one glomerulus ( but see [Bywalez et al . , 2016] ) , were used to label all of these neurons as superficial ‘short-axon’ cells ( SACs; Kiyokage et al . , 2010 ) . This DA SAC population was then further subdivided into ‘oligoglomerular’ and ‘polyglomerular’ subtypes based on the extent of ramification across the glomerular layer ( Kiyokage et al . , 2010 ) . In contrast , the second approach argues that classic morphological descriptions of superficial SACs report a complete lack of glomerular arborisation , and that the term ‘SAC’ should not be used to describe any OB DA neurons ( Kosaka and Kosaka , 2011; Kosaka and Kosaka , 2016 ) . Instead , according to this scheme , small-soma DA neurons form a subset of true periglomerular cells ( DA-PGCs ) , while large-soma DA cells that project long distances across the glomerular layer are termed ‘inhibitory juxtaglomerular association neurons’ , or IJGAs ( Kosaka and Kosaka , 2011; Kosaka and Kosaka , 2016 ) . This lack of agreement has led to some studies simply grouping all DA neurons into a single neurochemically or genetically defined class ( e . g . Banerjee et al . , 2015 ) . We agree that the dopaminergic-GABAergic phenotype of these cells is one of their most striking characteristics , defining them as a distinct population of OB interneurons . Moreover , we found here that on many measures the sensory response characteristics of the overall bulbar dopaminergic population are rather homogeneous ( Figures 9 and 10 ) . However , failing to identify important DA subclasses can produce issues in the interpretation of their functional roles within OB networks . The division we observe here may help to clarify matters substantially , and actually appears to fit reasonably well with both of the alternative schemes already proposed . On the one hand , AIS-positive , large OB DA neurons share many features with the ‘polyglomerular’ ( Kiyokage et al . , 2010 ) and ‘IJGA’ ( Kosaka and Kosaka , 2011; Kosaka and Kosaka , 2016 ) classes . On the other hand , AIS-negative , small OB DA cells have much in common with the ‘oligoglomerular’ ( Kiyokage et al . , 2010 ) and ‘DA-PGC’ ( Kosaka and Kosaka , 2011; Kosaka and Kosaka , 2016 ) subtypes . The AIS-negative class also shares important morphological features with a population of DAT-expressing ‘clasping SACs’ identified by recent live imaging of intracellular fills in acute OB slices ( Bywalez et al . , 2016 ) , whose distinct dendritic architecture and predominantly juxtaglomerular arborisations would appear to separate them from classically-defined PGCs ( Kosaka and Kosaka , 2016; Pinching and Powell , 1971 ) . Most importantly , while soma size and dendritic spread are continuous variables that do not permit simple sub-group identification , the presence or absence of an axon is a discrete feature that should allow for cleaner classification . Indeed , segregating OB DA neurons based on axonal criteria has enabled important functional distinctions to be identified between subgroups ( Figure 8; [Chand et al . , 2015] ) that were not evident from previous divisions based on continuous measures ( Pignatelli and Belluzzi , 2017; Pignatelli et al . , 2005 ) . Finally , in terms of nomenclature , we certainly feel that ‘SAC’ is a misleading term for all OB DA neurons , unless it is acknowledged that in some cells the axon in question is so short as to be non-existent . Perhaps , a simple distinction between ‘axonic’ and ‘anaxonic’ OB DA neurons will prove both clear and useful , although whether those subgroups represent forms of classically-defined PGC , SAC or other cell types can remain a matter for debate . The existence of two distinct subgroups of DA neurons raises the obvious question: how might these two subpopulations contribute to sensory processing ? In the GL , inhibitory signalling can be either intraglomerular or interglomerular in nature – acting within the circuitry of an individual glomerulus , or acting between different glomeruli , respectively . Both AIS-positive and AIS-negative DA subtypes possess dendritic processes that ramify within the glomerular neuropil ( Figure 3 ) , so , assuming that the release of GABA and/or dopamine occurs from these dendrites in both cell types ( Borisovska et al . , 2013; Kiyokage et al . , 2017; Vaaga et al . , 2017 , but see Liberia et al . , 2012 ) , both subpopulations have the potential to contribute to intraglomerular inhibition . This includes GABA and/or dopamine inhibiting release probability at OSN presynaptic terminals via the activation of GABAB and D2 receptors , respectively ( Ennis et al . , 2001; Hsia et al . , 1999; Korshunov et al . , 2017; McGann , 2013; Vaaga et al . , 2017 ) . Local GABA release can also provide a brake on recurrent excitatory glomerular networks ( Gire and Schoppa , 2009; Murphy et al . , 2005; Najac et al . , 2011 ) , as well as effecting auto-disinhibition at high input strengths ( Parsa et al . , 2015 ) . By acting at the levels of both input terminals and projection neuron dendrites , intraglomerular inhibition produced by both subtypes of OB DA neuron may subserve highly local gain control , potentially acting as a high-pass temporal and contrast filter to facilitate the detection of strong odorant stimuli ( e . g . Banerjee et al . , 2015; Cavarretta et al . , 2016; Cleland and Sethupathy , 2006; Gire and Schoppa , 2009; Korshunov et al . , 2017 ) . Interglomerular inhibition , by contrast , would appear to be restricted solely to the subpopulation of large , deep-lying , highly excitable AIS-positive OB DA neurons . Quite simply , only this population of GABAergic GL interneurons has an axonal process that can signal sufficient distances between glomerular networks ( Figures 1 , 2 and 3; Kiyokage et al . , 2010; Kosaka and Kosaka , 2008 ) . Given the distribution of odorant information across glomeruli in a spatial map for odour identity ( Murthy , 2011 ) , such lateral inhibition between glomeruli has been predicted to enhance contrast between individual stimulus representations , and therefore aid odorant identification and/or discrimination ( e . g . Linster and Cleland , 2004; Uchida et al . , 2000; Urban , 2002 ) . This lateral signal could be distributed by AIS-positive DA neurons via well-described long-range GABAergic monosynaptic connections onto external tufted cells ( Banerjee et al . , 2015; Liu et al . , 2013; Whitesell et al . , 2013 ) . In addition , the distal interglomerular projections of ( AIS-positive ) OB DAs can induce rebound excitation via dopaminergic D1-receptor activation ( Liu et al . , 2013 ) . Together with these cells’ slightly more broadly-tuned late excitatory and delayed inhibitory responses to odorant stimuli ( Figure 10 ) , this delayed DA-mediated effect could contribute to the complex modulation of interglomerular dynamics , especially in the later stages of stimulus processing . Finally , there might also be a significant developmental component to the relative functional contributions of axonic vs anaxonic OB DA neurons . Early in postnatal development , when neuronal activity contributes to the refinement of both OSN terminals ( Yu et al . , 2004; Zou et al . , 2004 ) and projection neuron dendrites ( Lin et al . , 2000; Matsutani and Yamamoto , 2000 ) to individual glomeruli , the large , interglomerular-projecting AIS-positive DA cell type is relatively more numerous ( Figure 6B ) . Maybe these inhibitory interneurons play a crucial role in co-ordinating odour-evoked and/or spontaneous activity across the glomerular layer at these early ages , allowing distinct activity patterns to drive anatomical segregation at the individual glomerulus level . Perhaps the most remarkable difference between the two axonic and anaxonic DA subtypes is that only the latter is generated throughout adult life . This observation is in agreement with a more general trend of adult-born neurons in the olfactory bulb , where neither PGCs nor granule cells possess an axon ( Lledo et al . , 2006 ) . A recently observed small cohort of adult-born cortical neurons is also anaxonic ( Le Magueresse et al . , 2011 ) . In fact , with the notable exception of hippocampal dentate granule cells , it appears that all CNS neurons constitutively born during adulthood are anaxonic , contributing purely to local network activity by releasing neurotransmitter from their dendrites . Indeed , one may speculate that it is simpler for a newly generated neuron to insert itself in a pre-existing network without having to extend and connect a far-reaching axonal process . Accordingly , the large , axonic OB DA cells that do need to form such extensive connections are born only during early development at the same time that other projection neurons are populating the bulb ( Treloar et al . , 2010 ) . What is the evolutionary advantage for maintaining continuous neurogenesis of anaxonic local interneurons ? The answer to this question is highly dependent on understanding the exact role of these small local neurons in olfactory processing . Immature adult-born neurons are distinguished by their heightened potential for activity-dependent plasticity ( Livneh and Mizrahi , 2012 ) . We might hypothesise , then , that a local intraglomerular gain control mechanism that can be readily modified by experience allows for broader behavioural flexibility , and permits rapid adaptation to new conditions in the external world ( Lazarini et al . , 2014; Livneh and Mizrahi , 2012; Rochefort et al . , 2002 ) . Moreover , if adult neurogenesis can be seen as an extreme form of structural plasticity that only one subtype of DA cells is capable of performing , it is probably reasonable to assume that more standard and less dramatic forms of plasticity are also differentially expressed by anaxonic and axonic cells . By definition , cells that do not have an AIS cannot undergo AIS plasticity , while in vitro evidence suggests that large axonic OB DA cells are capable of regulating the length and position of their AISs in an activity-dependent manner ( Chand et al . , 2015 ) . But does this happen in vivo , and if so , does it have an impact on the cell’s processing of olfactory inputs ? Additionally , are other forms of activity-dependent plasticity in OB DA neurons ( Banerjee et al . , 2013; Bonzano et al . , 2016; Coppola , 2012; Hsia et al . , 1999; Mizrahi , 2007; Wang et al . , 2017 ) specific to individual axonic versus anaxonic subclasses ? Future studies will need to elucidate if other forms of neuronal plasticity can be induced in both DA cell subtypes in response to perturbations in sensory experience , and if so , how they impact on olfactory behaviour ( Taylor et al . , 2009; Tillerson et al . , 2006 ) .
Unless otherwise stated , we used mice of either gender , and housed them under a 12 hr light-dark cycle in an environmentally controlled room with free access to water and food . Wild-type C57/Bl6 mice ( Charles River ) were used either as experimental animals , or to back-cross each generation of transgenic animals . The founders of our transgenic mouse lines – DAT-Cre ( B6 . SJL-Slc6a3tm1 . 1 ( cre ) Bkmn/J , Jax stock 006660 ) , VGAT-Cre ( Slc32a1tm2 ( cre ) Lowl/J , Jax stock 016962 ) , flex-tdTomato ( B6 . Cg–Gt ( ROSA ) 26Sortm9 ( CAG-tdTomato ) Hze , Jax stock 007909 ) , and flex-GCaMP6s animals ( Ai96; B6;129S6-Gt ( ROSA ) 26Sortm96 ( CAG-GCaMP6s ) Hze/J , Jax stock 024106 ) – were purchased from Jackson Laboratories . If not stated otherwise , all experiments were performed at postnatal day ( P ) 28 . All experiments were performed under the auspices of UK Home Office personal and project licences held by the authors , or were within institutional ( Harvard University Institutional Animal Care and Use Committee ) and USA national guidelines . Mice were anaesthetised with an overdose of pentobarbital and then perfused with 20 mL PBS with heparin ( 20 units . mL−1 ) , followed by 20 mL of 1% paraformaldehyde ( PFA; in 3% sucrose , 60 mM PIPES , 25 mM HEPES , 5 mM EGTA , and 1 mM MgCl2 ) . The olfactory bulbs were dissected and post-fixed in 1% PFA for 2–7 d , then embedded in 5% agarose and sliced at 50 µm using a vibratome ( VT1000S , Leica ) . Free-floating slices were washed with PBS and incubated in 5% normal goat serum ( NGS ) in PBS/Triton/azide ( 0 . 25% triton , 0 . 02% azide ) for 2 hr at room temperature . They were then incubated in primary antibody solution ( in PBS/Triton/azide ) for 2 days at 4°C . The primary antibodies used and their respective concentrations are indicated in the key resources table . Slices were then washed three times for 5 min with PBS , before being incubated in secondary antibody solution ( species-appropriate Life Technologies Alexa Fluor-conjugated; 1:1000 in PBS/Triton/azide ) for 3 hr at room temperature . After washing in PBS , slices were incubated in 0 . 2% sudan black in 70% ethanol at room temperature for 3 min to minimise autofluorescence , and then mounted on glass slides ( Menzel-Gläser ) with MOWIOL-488 ( Calbiochem ) . Unless stated otherwise , all reagents were purchased from Sigma . To birth-date neurons , we injected mice with a saline-based solution containing 50 mM bromodeoxyuridine ( BrdU , Sigma ) and 17 . 5 mM NaOH . Pregnant C57/BL6 female mice received one single intraperitoneal injection of this solution ( 0 . 075 ml/g ) on the relevant gestational day; pregnancy start date ( E0 ) was investigated twice daily and confirmed by the presence of a vaginal plug . Injected mothers and offspring were transcardially perfused on the relevant day , as detailed above . To permit BrdU detection , slices were first incubated in 2 M HCl for 30 min at 37°C , washed thoroughly and then processed for immunohistochemistry as described above . Sparse morphological labelling was achieved by injecting 2 µl of AAV9 . EF1a . ChR2-YFP lox/lox virus ( AV-9-PV1522 , Penn Vector Core , USA ) in the lateral ventricle of P1-2 DATCre or VGATCre neonatal mice . A combination of birth-dating and sparse labelling was accomplished by either electroporating 2 μl of EGFP in the lateral ventricle of P1 C57/BL6 mice , or by injecting floxed rv::dio-GFPlox/lox retrovirus ( Ciceri et al . , 2013 ) in the lateral ventricle of E12 DATCre embryos . All invasive surgery was performed under isoflurane anaesthesia , with Fast Green ( 0 . 3 mg/ml ) co-injected to visually confirm positional accuracy . Injections in embryos were performed with an injector and a 30 . 5 ga needle through the uterine wall into one of the lateral ventricles of the embryos . The uterine horns were then returned into the abdominal cavity , the wall and the skin were sutured , and embryos were allowed to continue their normal development . Injections in neonates were performed in a semi-stereotaxic frame using a Hamilton syringe and a borosilicate glass capillary ( GC100-15 , Harvard Apparatus ) . For electroporation , after the injection of 2 µl of GFP in the lateral ventricle , five 50 ms-0 . 15 A electrical pulses were delivered at 1 Hz with plate electrodes ( 10 mm diameter , Nepagene , Japan ) oriented in such a way to drive the current dorso-ventrally . All images were acquired with a laser scanning confocal microscope ( Zeiss LSM 710 ) using appropriate excitation and emission filters , a pinhole of 1 AU and a 40x oil immersion objective . Laser power and gain were set to either prevent signal saturation in channels imaged for localisation analyses , or to permit clear delineation of neuronal processes in channels imaged for neurite identification ( e . g . TH , GFP ) . In ex vivo tissue , for branching patterns and reconstructions , images were taken with a 1x zoom ( 0 . 415 µm/pixel ) , 512 × 512 pixels , and in z-stacks with 1 µm steps . For AIS identification , images were taken with 3x zoom , 512 × 512 pixels ( 0 . 138 µm/pixel ) and in z-stacks with 0 . 45 µm steps . All quantitative analyses were performed with Fiji ( Image J ) . AISs were identified by confirming double labelling in 3D of TH along with a contiguous , elongated AnkG-positive stretch of neurite , and by then following the AnkG-positive process backwards to a clearly identifiable TH-positive cell body . Cell position in the glomerular layer ( GL; defined by the perimeter of TH immunofluorescence ) was classified as: a ) ‘upper’ if the soma bordered with both the GL and the olfactory nerve layer; b ) ‘middle’ if the soma was fully embedded in the GL; c ) ‘lower’ if the soma bordered with both the GL and external plexiform layer ( EPL ) ; and d ) ‘EPL border’ if the soma did not border at all with the GL . Soma area was measured at the cell’s maximum diameter , by experimenters blind to cell type identity . Independent soma area measurements by two experimenters on a subset of TH-positive cells revealed extremely high levels of agreement ( Pearson r = 0 . 97 , n = 58 ) . Soma-to-OSN layer distance was calculated from a straight perpendicular line connecting the cell body to the outer border of TH immunofluorescence . Morphological reconstructions of neurons that were sufficiently sparsely and brightly labelled were obtained with the auto-tracing Neuron 2 . 0 function in Vaa3D-3 . 20 . Sholl analysis was performed on traced images using the automated Image J function , with a fixed first circle radius of 10 µm , and 5 µm increments for the following concentric circles . The number and length of primary dendrites was manually calculated with the freehand drawing function of Image J . The density of AIS-positive DA neurons at P28 was calculated as follows . First , the average density of TH-positive cells per cubic millimetre of GL was calculated by manually counting cells and measuring GL area in medium-magnification stacks ( ~43 , 000 cells / mm3; 40x objective , zoom 1; n = 14 slices , N = 3 mice ) . Second , the total volume of tissue used to identify a total of 297 AIS- and TH-positive cells was estimated by manually drawing GL area profiles in low-magnification images ( ~0 . 28 mm3; 5x objective; n = 81 , N = 9 ) and multiplying by the slice thickness ( 50 µm ) . The estimated number of TH-positive cells in this volume ( ~11 , 801 ) was then calculated by multiplying GL volume by the previously defined average TH-positive density per cubic millimetre . Finally , the AIS-positive/TH-positive cell percentage was calculated by dividing the total number of identified AIS-positive/TH-positive cells ( 297 ) by the estimated number of TH-positive cells present in the analysed volume . In P0 tissue , the proportion of axon-bearing DA neurons was calculated by manually counting the number of TH-positive cells in 20 stacks ( n = 239 , N = 2 ) , and confirming with TRIM-46 co-label the subset of those with an axonal process ( n = 14 ) . P21-35 DATCre-tdTomato mice were decapitated under isoflurane anaesthesia , and the OB was removed and transferred into ice-cold slicing medium containing ( in mM ) : 240 sucrose , 5 KCl , 1 . 25 Na2HPO4 , 2 MgSO4 , 1 CaCl2 , 26 NaHCO3 and 10 D-Glucose , bubbled with 95%O2 and 5% CO2 . Horizontal slices ( 300 µm thick ) of the olfactory bulb were cut using a vibratome ( VT1000S , Leica ) and maintained in ACSF containing ( in mM ) : 124 NaCl , 5 KCl , 1 . 25 Na2HPO4 , 2 MgSO4 , 2 CaCl2 , 26 NaHCO3 and 20 D-Glucose , bubbled with 95% O2 and 5% CO2 for >1 hr before experiments began . Whole-cell patch-clamp recordings were performed using an Axopatch amplifier 700B ( Molecular Devices , San Jose , CA , USA ) at physiologically relevant temperature ( 32–34°C ) with an in-line heater ( TC-344B , Warner Instruments ) . Signals were digitised ( Digidata 1550 , Molecular Devices ) and Bessel-filtered at 3 kHz . Recordings were excluded if series ( RS ) or input ( RI ) resistances ( assessed by −10 mV voltage steps following each test pulse , acquisition rate 20 KHz ) were respectively bigger than 30 MΩ or smaller than 100 MΩ , or if they varied by >20% over the course of the experiment . Fast capacitance was compensated in the on-cell configuration and slow capacitance was compensated after rupture . Recording electrodes ( GT100T-10 , Harvard Apparatus ) were pulled with a vertical puller ( PC-10 , Narishige ) and filled with an intracellular solution containing ( in mM ) : 124 K-Gluconate , 9 KCl , 10 KOH , 4 NaCl , 10 HEPES , 28 . 5 Sucrose , 4 Na2ATP , 0 . 4 Na3GTP ( pH 7 . 25–7 . 35; 290 MOsm ) and Alexa 488 ( 1:150 ) . DA cells were visualised using an upright microscope ( FN1 , Nikon , Tokyo , Japan ) equipped with a 40X water immersion objective , and tdT/Alexa 488 fluorescence was revealed by LED ( CoolLED pE-100 ) excitation . Post-patch fill with Alexa 488 was used both to confirm tdT-positive cell identity , and to measure soma area ( ImageJ ) in live images captured via a SciCam camera ( Scientifica ) . In current-clamp mode , evoked spikes were measured with Vhold set to −60 ± 3 mV . For action potential waveform measures , we injected 10-ms-duration current steps of increasing amplitude until we reached the current threshold at which the neuron reliably fired an action potential ( Vm >0 mV; acquisition rate 200 KHz ) . For multiple spiking measures , we injected 500-ms-duration current steps from 0 pA of increasing amplitude ( Δ2pA ) until the neuron passed its maximum firing frequency ( acquisition rate 50 KHz ) . Exported traces were analysed using either ClampFit ( pClamp10 , Molecular Devices ) or custom-written routines in MATLAB ( Mathworks ) . Before differentiation for dV/dt and associated phase plane plot analyses , recordings at high temporal resolution ( 5 μs sample interval ) were smoothed using a 20 point ( 100 μs ) sliding filter . Monophasic versus biphasic phase plane plots were then visually determined independently by EG and MSG . We classified completely monotonic plots with continually increasing rate-of-rise as monophasic , and any plots showing a clear inflection in rate-of-rise over the initial rising phase as biphasic . Any discrepancies in classification were resolved by mutual agreement . We also corroborated our subjective classification using a quantitative measure of spike onset sharpness: the ratio of errors produced by linear and exponential fits to the perithreshold portion of the phase plane plot ( Baranauskas et al . , 2010; Volgushev et al . , 2008 ) . Fit error ratios were calculated with a custom Matlab script written by Maxim Volgushev , using variable initial portions of the phase plane plot between voltage threshold and 40% of maximum dV/dt ( Baranauskas et al . , 2010 ) , for single spikes fired in response to 10 ms current injection at current threshold and up to three subsequent suprathreshold sweeps . Using strict , established ( Baranauskas et al . , 2010 ) , but non-inclusive criteria for ‘steep’ ( ≈ biphasic; maximum fit error ratio >3 ) versus ‘smooth’ ( ≈ monophasic; maximum fit error ratio <1 ) spike onset , we were able to objectively classify phase plane plot shape in a smaller subset of our recorded DAT-tdT+ neurons . Of 13 neurons classified in this manner , only one was classified differently by subjective vs objective criteria . Importantly , the limited subset of objectively classified cells still displayed significant differences between mono- and biphasic OB dopaminergic neurons: as in the larger , subjectively-classified dataset , biphasic cells had bigger soma area , and were more excitable than their monophasic counterparts ( Figure 8—figure supplement 1 ) . For quantification of AP properties , voltage threshold was taken as the potential at which dV/dt first passed 10 V/s . Onset rapidness was taken from the slope of a linear fit to the phase plane plot at voltage threshold . Spike width was measured at the midpoint between voltage threshold and maximum voltage . Rheobase and afterhyperpolarisation values were both measured from responses to 500 ms current injection , the latter from the local voltage minimum after the first spike fired at rheobase . Input-output curves were constructed by simply counting the number of spikes fired at each level of injected current density . Thirteen DATCre-GCaMP6s mice ( either gender , age 4–10 months ) were anaesthetised with a mixture of ketamine ( 100 mg/kg ) and xylazine ( 10 mg/kg ) , placed in a stereotaxic apparatus and equipped with a cranial window over the olfactory bulbs using a sterile 3 mm biopsy punch ( Integra Miltex ) . A custom-built titanium head plate was secured to their skull with adhesive luting cement ( C and B Metabond , Parkell ) . A coverslip ( 3 mm , Warner Instruments ) was placed over the cranial window and tissue adhesive ( 3M Vetbond ) was used to secure the coverslip to the bone . The mice were allowed a minimum of a week to recover from surgery before the first imaging session . Prior to each imaging session the mice were newly anaesthetised with a mixture of ketamine ( 100 mg/kg ) and xylazine ( 10 mg/kg ) and secured in a custom-built microscope as described previously ( Kapoor et al . , 2016 ) . Mice were given a maximum of one booster injection of anaesthesia per session , and were never imaged on consecutive days . GCaMP was excited and imaged via a water immersion objective ( 20x , 0 . 95 NA , Olympus; sterile saline was used as the fluid for the immersion objective ) at 927 nm using a Ti:sapphire laser ( Chameleon Ultra , Coherent ) with 140 fs pulse width and 80 MHz repetition rate . Image acquisition , scanning , and stimulus delivery were controlled by custom-written software in LabVIEW ( National Instruments ) . Eight odors were individually delivered via a custom-built olfactometer ( Kapoor et al . , 2016 ) . The odour panel included: methyl propionate ( Sigma , 81988 ) , methyl butyrate ( Sigma , 246093 ) , ethyl valerate ( Sigma , 290866 ) , hexanal ( Sigma , 115606 ) , methyl tiglate ( Penta , 13–73400 ) , valeraldehyde ( Sigma , 110132 ) , propyl acetate ( Tokyo Chemical Industry , A0044 ) , and pentyl acetate ( Sigma , 109549 ) . All odours were diluted in diethyl phthalate solvent ( Sigma-Aldrich , St . Louis , Missouri , United States ) at 2% v/v . Single-plane images of 300 × 300 pixel fields of view were acquired at 4 Hz during odour stimulation trials . Trial temporal structure consisted of: 7 . 5 s baseline , 3 s odour delivery , 7 . 5 s post-odour acquisition ( 18 s total , with 10 s inter-trial interval ) . All eight odours were probed sequentially , and then the entire block was repeated two more times , for a total of 24 odour trials for each field of view ( three repetitions per odour ) . Cell soma selection was performed manually in Image J , using both the timecourse and the maximum intensity projections of each odour trial , and stored in a ROI mask for each field of view . Soma area and mean intensity values were then extracted for each ROI with a custom-written ImageJ macro , and saved as . xls and . txt files respectively . Data were then analysed with custom scripts in Matlab ( Mathworks ) . Mean response traces were calculated across three individual stimulus presentations of each odour for each cell from background-subtracted raw fluorescence values , before bleach correction was carried out by extrapolation and subtraction of a single exponential function fitted to the 7 . 5 s pre-stimulus baseline . Mean bleach-corrected baseline fluorescence over a 3 s window immediately preceding odorant presentation ( f ) was then used to generate Δf/f traces . In some analyses , comparing baseline fluorescence values across different animals and imaging sessions ( but not when calculating Δf/f values for all other measures ) , this baseline F value , averaged across all stimulus presentations , was normalised by the mean value for all small DA cells in a given field of view . The standard deviation of Δf/f values within the 3 s pre-stimulus period , averaged across all stimulus presentations , was taken as a measure of each cell’s baseline noise . For each cell and each odorant , we then detected the point of maximum ( for excitatory events ) and minimum ( for inhibitory events ) Δf/f after stimulus onset , and took response amplitude as mean Δf/f over a 3 s window centred on this peak timepoint . Responses were classed as significant if this amplitude value was ≥3 x the baseline noise for that trace . For analyses of response reliability ( Figure 9 ) , background subtraction , bleach correction and Δf/f calculation were carried out on individual response traces . Coefficients of variability for peak latencies were then calculated as sd/mean across three stimulus repeats . Z-scores were calculated on a repeat-by-repeat basis by dividing response amplitude ( as above , mean Δf/f over a 3 s window centred on the peak timepoint ) by the standard deviation of Δf/f values within the 3 s pre-stimulus period . We then calculated the mean of the two smallest amplitude z-scores across the three stimulus repeats . Excitatory responses displayed a clear bimodal distribution of peak timepoints ( see Figure 9D ) , so in an pre-analysed subsample of our data ( n = 408 cells from N = 4 mice ) we used unbiased k-means clustering on this parameter to set a threshold timepoint at 6 s after stimulus onset – all significant excitatory responses which peaked at or before this timepoint were classed as ‘early’ , and all significant excitatory responses which peaked after this timepoint were classed as ‘late’ . Tuning index ( TI ) values were calculated by summing the number of stimuli producing significant responses of a given type for each cell . These values were often zero , and many cells had zero TI values for all response types . These non-responding cells were included in all reported TI analyses , but results were identical in terms of significance if analyses were restricted only to those cells that displayed at least one significant response of any type to at least one odour . Amplitude measures were calculated for each cell as the mean across all odours that produced significant responses , but cell-type effects were also consistent if this was calculated as the maximum amplitude across all significant responses instead . To estimate response decay constants for late excitatory and inhibitory responses , we started by identifying a subset of cells whose long-latency responses returned to baseline within the timeframe of our 18 s imaging sweep ( late excitatory , n = 14; inhibitory , n = 44 ) . In these cells , we compared the decay constants produced by a single exponential fit to either the 2 . 5 s immediately following the response peak ( =10 timepoints at 4 Hz sample rate ) , or to the entire post-peak response profile . We found that fitting just the 2 . 5 s from the response peak produced accurate decay constant estimates – the differences between partial and full fits were small , and were not significantly different from zero for either response type ( Figure 10—figure supplement 1A , B; late excitatory mean ±SEM - 3 . 35 ± 2 . 26 s , Wilcoxon test vs 0 , W = −21 , p=0 . 50; inhibitory , −0 . 41 ± 0 . 72 s , W = 221 , p=0 . 17 ) . Importantly , there was also no difference in the accuracy of this estimation in big versus small OB DA cells ( late excitatory , fixed effect of cell type in multilevel ANOVA , F1 , 9 = 0 . 92 , p=0 . 36; inhibitory , F1 , 41 = 1 . 22 , p=0 . 28 ) . We were then able to estimate decay constants in a larger subset of cells whose long-latency responses peaked at least 2 . 5 s before the end of our imaging sweeps . This subset comprised 42% and 44% of all cells with late excitatory and inhibitory responses , respectively , proportions that did not differ between cell types ( late excitatory , small cells , 42/105; big cells , 60/140 , Fisher’s exact test , p=0 . 70; inhibitory , small cells 66/152 , big cells 63/144 , p>0 . 99 ) . We found slow decay constants of approximately 8–11 s for both late excitatory and inhibitory response types , but no difference in decay kinetics between big and small OB DA cells ( Figure 10—figure supplement 1C , D; late excitatory , fixed effect of cell type in multilevel ANOVA , F1 , 98 = 0 . 32 , p=0 . 57; inhibitory , F1 , 113 = 0 . 078 , p=0 . 78 ) . Importantly , although making up only ~40% of the total population of responding cells , the subsample of neurons for which we could accurately estimate decay kinetics was representative of our sample as a whole , with no significant differences observed between decay-estimated and decay-non-estimated cells on a wide range of measures ( t-test or Mann-Whitney as appropriate within small and big cell subpopulations , Bonferroni-corrected p>0 . 05 for soma area , baseline fluorescence , baseline noise , and response amplitude ) . Statistical analysis was carried out using Prism ( Graphpad ) , Matlab ( Mathworks ) or SPSS ( IBM ) . Sample distributions were assessed for normality with the D’Agostino and Pearson omnibus test , and parametric or non-parametric tests carried out accordingly . α values were set to 0 . 05 , and all comparisons were two-tailed . Principal component analysis ( PCA ) and k-means classification on electrophysiological data were performed ( Matlab functions ‘pca . m’ and ‘kmeans_lpo . m’ , respectively ) on the five variables that differed significantly between monophasic and biphasic DAT-tdT neurons . All were normally distributed except onset rapidness , which was rendered normal by logarithmic transform . Results of the k-means analysis were validated with a ‘leave-one-out’ protocol , which revealed cell-type classification to be robust to the removal of any one cell from the dataset . For multilevel analyses of in vivo GCaMP data , distributions of baseline noise and response amplitude measures were rendered normal by logarithmic transform , and outliers – defined as any value with an absolute z-score >3 – were removed ( Aarts et al . , 2014 ) ; a single outlier was removed from each dataset , representing <0 . 5% of each sample ) . These parameters were then analysed using linear mixed models ( SPSS ) with mouse as the subject variable . Tuning index data could not be rendered normal by any standard transforms , so were analysed using generalised linear mixed models with a negative binomial target distribution ( accounting for >90% of sample variance; SPSS ) and mouse as the subject variable . Dummy variable analysis revealed significant intracluster correlations in all cases , stressing the importance of nesting cell-by-cell data on individual mouse subjects ( Aarts et al . , 2014 ) . Due to the non-normal nature of tuning index distributions , and the rarity of observing multiple different response types in any given single neuron , PCA was not attempted on our GCaMP data . | Most of your brain cells were born before you were . But in mammals , including humans , some of these brain cells , also known as nerve cells or neurons , are created after birth . These later-generated neurons are often extremely similar to their counterparts produced in the womb , and also seem to perform a similar role once they are fully mature . However , it has not been entirely clear if the later-produced neurons may also have a specific purpose . Neurons are made of a cell body with a cable-like structure called axon that transmits information to more distant neurons , and dendrites , which are branches that receive information from other neurons . Neurons use different signalling molecules to communicate , one of which is called dopamine , and the neurons that use this specific signal are called dopaminergic neurons . Now , Galliano et al . wanted to test if neurons created in the womb , and neurons created after birth , are really so similar . To investigate this , they compared the dopaminergic neurons from mice found in the first part of the brain to process information about smell – the olfactory bulb . These specific neurons are known to have diverse properties and can also be produced after birth . Galliano et al . studied their development , form and purpose , and discovered that only neurons produced in the womb can possess an axon . Moreover , the axon-bearing cells had a different form and functional properties to their axon-less cousins , and also showed some subtle differences in their ability to respond to smell . This demonstrates that two very different types of dopaminergic neurons in the olfactory bulb are produced at different stages during the development . A better knowledge of such basic brain-developmental features is essential for the wider goal of understanding how the brain operates , and to discover ways to repair it when it is not working properly . Neurons created after birth in particular , might enable us to develop new treatment strategies; for example , adding new dopaminergic neurons to replace those lost in degenerative disorders such as Parkinson’s Disease . When developing such regenerative therapies , why not learn lessons from how the brain can achieve this naturally ? | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"neuroscience"
] | 2018 | Embryonic and postnatal neurogenesis produce functionally distinct subclasses of dopaminergic neuron |
Fundamental gaps remain in our understanding of how immunity to malaria develops . We used detailed clinical and entomological data from parallel cohort studies conducted across the malaria transmission spectrum in Uganda to quantify the development of immunity against symptomatic P . falciparum as a function of age and transmission intensity . We focus on: anti-parasite immunity ( i . e . ability to control parasite densities ) and anti-disease immunity ( i . e . ability to tolerate higher parasite densities without fever ) . Our findings suggest a strong effect of age on both types of immunity , not explained by cumulative-exposure . They also show an independent effect of exposure , where children living in moderate/high transmission settings develop immunity faster as transmission increases . Surprisingly , children in the lowest transmission setting appear to develop immunity more efficiently than those living in moderate transmission settings . Anti-parasite and anti-disease immunity develop in parallel , reducing the probability of experiencing symptomatic malaria upon each subsequent P . falciparum infection .
The last decades have seen substantial declines in malaria transmission in sub-Saharan Africa that are largely attributable to increased access to effective control measures , including insecticide-treated bednets , indoor residual spraying of insecticide and artemisinin-based combination therapy ( Bhatt et al . , 2015; World Health Organization , 2016 ) . In settings where transmission has been low , increased access to effective control interventions opens the possibility for malaria elimination . In highly endemic settings , however , there are concerns around the potential impact of failing to sustain interventions that reduce but do not stop transmission . Short-term decreases in malaria incidence due to reductions in transmission could be offset over time by reductions in population immunity to malaria resulting from lower exposure to parasites ( Filipe et al . , 2007; Smith et al . , 2001; Snow et al . , 1997 ) . Gradual acquisition of immunity against symptomatic malaria ( also referred to as clinical immunity ) is a key driver of the epidemiology of malaria in endemic settings , where the incidence of disease typically peaks in early childhood and then declines , while the prevalence of detectable asymptomatic parasitemia increases throughout childhood before declining in adulthood ( Griffin et al . , 2015; Reyburn et al . , 2005; Okiro et al . , 2009; Carneiro et al . , 2010; Idro et al . , 2006; Roca-Feltrer et al . , 2010; Rodriguez-Barraquer et al . , 2016 ) . While these epidemiologic patterns have been described across the transmission spectrum , there are still many fundamental gaps in our understanding of the factors driving the development of immunity , and of the independent roles of age and repeated infection . One reason it has been challenging to study immunity to malaria is that there are currently no agreed upon reliable and quantifiable immune correlates of protection that can be measured in epidemiological studies ( Valletta and Recker , 2017; Fowkes et al . , 2010 ) . In addition , there are few available datasets that include both detailed clinical data and independent metrics of exposure at the individual level . Here , we use data from three parallel cohort studies conducted across the spectrum of malaria transmission in Uganda to model and quantify the development of immunity against symptomatic malaria as a function of transmission intensity and age . A key strength of these studies is that they involved detailed clinical and entomological surveillance of all study households . We focus on two specific types of immunity: anti-parasite immunity ( i . e . the ability to control parasite densities upon infection ) and anti-disease immunity ( i . e . the ability to tolerate higher parasite infections without developing objective fever ) , as they have been described as independent components of clinical immunity ( Struik and Riley , 2004 ) .
To assess whether entomological metrics were a good indicator of individual exposure to P . falciparum , we correlated the measured annual EIRs ( aEIR ) for each household ( Figure 2a ) with estimates of the average individual hazard of infection ( Figure 2b ) . Individual hazards were estimated by fitting time-to-event models to the incidence data from each site . We found a significant correlation between these two independent metrics of exposure across sites ( R2 = 0 . 47 , p<0 . 001 ) . aEIR explained less of the variance between individuals within each site: Nagongera ( R2 = 0 . 03 , p=0 . 004 ) ; Kihihi ( R2 = 0 . 12 , p<0 . 001 ) ; Walukuba ( 0 . 01 , p=0 . 05 ) . Parasite densities developed upon infection decreased with increasing age in all settings and for both symptomatic ( passive detection ) and asymptomatic ( detected during routine visits ) infections . Despite the large variability in parasite densities recorded within and between individuals , this trend is evident in the raw data ( Figure 3a ) . A trend toward lower parasite densities was also observed among individuals living in settings with higher aEIRs ( Nagongera ) , as compared to settings with lower aEIR ( Kihihi and Walukuba ) . We considered multiple candidate models to describe the association between parasite density , age and aEIR ( Appendix 1 ) . Models allowing smooth ( non-linear ) relationships with aEIR best fit the data . Models allowing for two-way interactions between age and aEIR also outperformed models that did not include interactions . In moderate and high transmission settings ( households with aEIR >5 ) , increasing age and increasing exposure were independently and linearly associated with decreases in the parasite densities ( Table 2 ) . On average , parasite densities decreased by a factor of 0 . 76 ( 95%CI 0 . 75–0 . 77 ) for each additional year of age and by a factor of 0 . 73 ( 95%CI 0 . 70–0 . 76 ) for each two-fold increase in the aEIR . The relationship was less evident for the lower transmission households ( aEIR <5 ) . In these settings , there continued to be a decreasing ( although smaller ) association with age , but the expected parasite densities at any given age were equal or lower to those observed in the higher exposure ( aEIR >10 ) settings . Figures 4a and 5a present the predicted parasite densities , as a function of age and aEIR , according to the best fitting model . While an individual aged 1 year exposed to an aEIR of 10 is expected to develop a parasite density of 14 , 610 parasites/μL ( 95% CI 5924–36 , 031 parasites/μL ) upon infection , the expected parasite density goes down to 3237 parasites/μL ( 95% CI 1381–7586 parasites/μL ) by age 10 years . In contrast , the expected parasite density in an individual living in a setting with aEIR of 150 will be similar at age 1 year ( 13 , 071 parasites/μL ( 95% CI 5256–32 , 503 parasites/μL ) ) , but significantly lower by age 10 years ( 999 parasites/μL ( 95% CI 398–2508 parasites/μL ) ) . To test whether the observed associations with age could be explained by the cumulative exposure over a life time , we also fit models where , instead of adjusting for the aEIR , we adjusted for the cumulative number of infectious bites ( i . e . the product of age and aEIR ) ( Figure 5—figure supplement 2 ) . Results from these models are consistent with a smaller , yet independent effect of age on the development of anti-parasite immunity; for any given level of cumulative exposure , each additional year of life was associated with decreases in parasite densities by a factor of 0 . 82 ( 95%CI 0 . 81–0 . 85 ) . We define anti-disease immunity as the ability to tolerate a given parasite density without developing objective fever . Thus , we were interested in modeling temperatures recorded at specific parasite densities , as a function of age and aEIR . Consistent with models characterizing anti-parasite immunity , models including smooth effects and interactions fitted the data significantly better than simpler models . As expected , we found a strong association between parasite densities and objective temperature ( Figure 6—figure supplement 1 ) . Increases in parasite densities above 1000 parasites/μL were associated with higher expected temperatures across ages and transmission settings . In addition , we found a negative association between objective temperature at a given parasite density and age ( Figures 3b , 4b and 6 ) . In moderate and high transmission settings ( aEIR >5 ) , the objective temperature at a given parasite density decreased on average by 0 . 08°C ( 95% CI 0 . 07–0 . 10°C ) for each additional year of life ( Table 2 ) . Thus , while the expected temperature for a child aged 1 year living in a setting with aEIR of 10 with a parasite density of 40 , 000 would be 38 . 8°C ( 95% CI 38 . 5–39 . 2°C ) , the expected temperature would decrease to 37 . 6°C ( 95% CI 37 . 3–38 . 0°C ) if the same child experienced the infection at age 10 years ( Figures 4b and 6 ) . This association was similar even when adjusting for cumulative exposure and for the differences in incidence of non-malarial fever across age-groups ( Figure 5—figure supplement 5 ) . Similar to the anti-parasite immunity results described above , the observed association between exposure level and anti-disease immunity was less evident than the association with age ( Figures 3b , 4b and 6 ) . For moderate and high transmission settings ( aEIR 5 to 300 ) , there was a linear negative association between objective temperature at a given parasite density and aEIR . The objective temperature decreased by 0 . 07°C ( 95% CI 0 . 05–0 . 10°C ) for each two-fold increase in aEIR . However , the relationship did not follow this trend for lower transmission settings ( Table 2 ) . Children living in the lowest transmission settings ( aEIR 1 to 5 ) appeared to tolerate higher parasite densities than children living in moderate transmission settings ( aEIR 5 to 10 ) . As an alternative way to characterize anti-disease immunity , we used our best fitting model to predict the fever threshold , defined as the minimum parasite density associated with objective fever ( temperature >38°C ) , across levels of age and aEIR ( Figure 5b ) . This quantity is often referred to as the ‘pyrogenic density’ . Results from this analysis show that , for settings with moderate and high transmission ( aEIR >5 ) , the fever threshold increases both with age and increasing exposure . Thus , while a 1-year-old child living in a setting with aEIR of 10 presenting with a parasite density as low as 3747 parasites/μL ( 95% CI 777–11 , 129 parasites/μL ) will be expected to be febrile , children older than 6 years of age exposed to very high transmission ( aEIR 150 ) might be afebrile even with parasite densities higher than 60 , 000 parasites/μL . Finally , to characterize the association between age and aEIR on the overall risk of developing symptomatic malaria upon infection ( i . e . the combined effect of anti-parasite and anti-disease immunity ) , we fit a series of models where the outcome of each independent microscopically detectable infection ( i . e . symptomatic malaria or asymptomatic parasitemia ) was modeled as a function of age and aEIR . Models allowing smooth relationships , with or without two-way interactions , fit the data equally well . Results from this analysis are consistent with results from the anti-parasite and anti-disease models ( Figure 7 ) . While young children living in low transmission settings ( aEIR = 5 ) develop symptomatic malaria in most their infections , the probability that an infection results in symptomatic malaria decreases as a function of age and exposure . The expected probability of symptomatic disease for a child aged 1 year living in a setting with aEIR of 50 is 0 . 92 ( 95% CI 0 . 79–0 . 97 ) , but it decreases to 0 . 51 ( 95% CI 0 . 29–0 . 73 ) by age 10 years . To assess whether recent P . falciparum infection was associated with different levels of anti-parasite and anti-disease immunity , we used data on the recent malaria history of each individual to fit models adjusted for number of P . falciparum positive visits in the past 3 and 6 months . We found no association between the number of recent malaria infections and our outcomes of interest ( Appendix 2 , figure 5—figure supplement 3 ) . Models that included random effects at the individual and household levels outperformed models that assumed independence of observations , consistent with large heterogeneity between individuals in the development of anti-parasite , anti-disease and overall immunity against symptomatic malaria . To illustrate this heterogeneity , we used the best fitting model to predict the trajectories of a subset of individuals with respect to anti-parasite and anti-disease immunity , as a function of age and aEIR ( Figure 5—figure supplement 12 ) . Our main analyses include data from all visits regardless of their type ( routine vs passive case detection ) . Thus , the expected values modeled here may be biased by the frequency of active vs passive episodes detected . In particular , it is possible that we have under-sampled the instances of asymptomatic infection , and thus , our estimates of the expected parasite densities may be an over-estimate of those present in the population . Similarly , it is also possible that consecutive asymptomatic infections represent persistent , rather than new infections . To address these limitations , we performed sensitivity analyses where we ( a ) up-weighted the episodes of asymptomatic parasitemia , to account for potentially unobserved asymptomatic infections and ( b ) included only ‘incident’ asymptomatic infections , under the assumption that subsequent asymptomatic samples represented persistent ( rather than new ) infections . Results from these analyses were qualitatively identical to the main analysis reported here and are presented in the supplementary material ( Figure 5—figure supplements 6 and 7 ) . To explore whether differences in the prevalence of certain host genetic polymorphisms between sites could be driving some of our findings , we also performed sensitivity analyses limiting the dataset to those subjects without the sickle hemoglobin mutation ( β globin E6V ) , known to protect against malaria ( Lopera-Mesa et al . , 2015; Taylor et al . , 2012 ) . Even though the sample size of these analyses was smaller ( observations from 155/773 individuals were excluded ) , results were unchanged qualitatively ( Figure 5—figure supplement 8 ) . Similarly , restricting the dataset to children without two other known polymorphisms ( the α-thalassemia 3 . 7 kb deletion or glucose-6-phosphate dehydrogenase deficiency caused by the common African variant ( G6PD A- ) ) , had little effect on the results .
Our findings illustrate how anti-parasite and anti-disease immunity develop gradually and in parallel , complementing each other in reducing the probability of experiencing symptomatic disease upon infection with P . falciparum . While anti-parasite immunity acts to restrict the parasite densities that develop upon each subsequent infection , anti-disease immunity increases the tolerance to high parasite densities . Thus , older children are less likely to develop symptomatic malaria upon infection both because they tolerate parasite densities better without developing fever , and because they are less likely to develop high parasite densities . Our results indicate independent effects of age on the acquisition of both anti-parasite and anti-disease immunity . These independent age effects may reflect maturation of the immune system as well as other physiological changes that decrease the propensity to fever ( Struik and Riley , 2004; Baird , 1998 ) . Furthermore , our findings are consistent with independent effects of transmission intensity on the acquisition of these two types of immunity . While the results obtained for moderate and high transmission settings ( aEIR >5 ) are consistent and expected , and suggest that immunity develops faster in settings where individuals get infected by P . falciparum more often , the results obtained for the lowest transmission settings are harder to reconcile . These results were largely driven by observations collected in the Walukuba site , and as such it is possible that site-specific characteristics may have driven them . Walukuba was previously a relatively high transmission rural area , but substantial decreases in transmission intensity have been observed since 2011 , likely due to urbanization . While our sensitivity analyses suggested that differences in the prevalence of three well characterized host-genetic polymorphisms between sites do not explain these discrepant results , it is still possible that other unmeasured site-specific characteristics may have driven them . Lower complexity of infection coupled with lower parasite diversity in Walukuba , for example , could cause this difference , as developing an effective immune response against fewer parasite strains may be much easier than developing immunity against a more diverse pool ( Hviid , 1998; Bull et al . , 1998 ) . Testing this hypothesis would require careful characterization of the complexity and diversity of infections in each of our cohort settings . While site-specific characteristics may underlie the observed high levels of clinical immunity against malaria in the low transmission setting , it is also possible that this finding reflects biologically relevant differences in how immunity against malaria develops . For example , it has been hypothesized that immunity may develop optimally in individuals that are exposed at a low rate , and that more frequent infections may interfere with the development of robust immune responses ( Wipasa et al . , 2010; Langhorne et al . , 2008 ) . Answering this question will require further detailed studies across transmission settings , with careful characterization of both exposure and infection outcomes . There are several limitations to this study . With a study design including routine visits every 3 months , we are likely to have missed several asymptomatic infections , particularly in the moderate and high transmission settings . Moreover , since infections were detected using microscopy , we were unable to detect sub-patent infections , and we lack knowledge about the genetic complexity of each infection . While it is possible that the expected values modeled here ( expected parasite density and fever threshold ) were biased by these sources of measurement error , sensitivity analyses suggest that the relationships observed were robust . Secondly , while we found an independent association between the average household aEIR and both anti-parasite and anti-disease immunity , it is not clear that this is the most relevant metric of exposure for the development of clinical immunity to malaria . Alternative metrics such as the number of discrete infections , the number of ‘strains’ seen or the total parasite-positive time might be more relevant , but require the collection of additional data , including more frequent sampling . Finally , while this study provides very detailed insight into how two types of clinical immunity to malaria develop in endemic settings as a function of age and repeated exposure , it says nothing about the duration of immunity . Prior studies have tried to model the processes driving acquisition of clinical immunity against malaria . However , these models have been generally informed by aggregated epidemiological data ( age-incidence and age-prevalence ) which limits their capacity to isolate the contributions of age and repeated exposure ( Filipe et al . , 2007; Griffin et al . , 2015 , 2014 ) . Our results quantify how anti-parasite and anti-disease immunity develop in children across the malaria transmission spectrum , and they support strong independent effect of age and a perhaps paradoxical effect of exposure . The methods proposed here to model anti-parasite and anti-disease immunity may also provide a framework to select individuals with immune and non-immune phenotypes for evaluations of immune correlates of protection .
The study protocol was reviewed and approved by the Makerere University School of Medicine Research and Ethics Committee ( Identification numbers 2011–149 and 2011–167 , the London School of Hygiene and Tropical Medicine Ethics Committee ( Identification numbers 5943 and 5944 ) , the Durham University School of Biological and Biomedical Sciences Ethics Committee ( PRISM Entomology Uganda ) , the University of California , San Francisco , Committee on Human Research ( Identification numbers 11–05539 and 11–05995 ) and the Uganda National Council for Science and Technology ( Identification numbers HS-978 and HS-1019 ) . All parents/guardians were asked to provide written informed consent at the time of enrollment . We used data from three parallel cohort studies conducted in Uganda in sub-counties chosen to represent varied malaria transmission ( Kamya et al . , 2015 ) . Walukuba , in Jinja district , is a peri-urban area near Lake Victoria that has the lowest transmission among the three ( annual entomological inoculation rate ( aEIR ) estimated to be 2 . 8 [Kamya et al . , 2015] ) . Kihihi , in Kanungu district , is a rural area in southwestern Uganda characterized by moderate transmission ( aEIR = 32 ) . Nagongera , Tororo district , is a rural area in southeastern Uganda with the highest transmission ( aEIR = 310 ) ( Kamya et al . , 2015; Kilama et al . , 2014 ) . Details on how the study households and participants were selected has been described elsewhere ( Kamya et al . , 2015 ) . Briefly , all households were enumerated , and then approximately 100 households were selected at random from each site . Between August and September 2011 , all children from these households aged between 6 months and 10 years who met eligibility criteria were invited to participate . As the cohorts were dynamic , additional children from participating households were invited to participate if they became eligible while the study was ongoing . Unless participants were withdrawn from the study either voluntarily or because they failed to comply with study visits , they were followed-up until they reached 11 years of age . Children from 31 randomly selected additional households were enrolled between August and October 2013 to replace households in which all study participants had been withdrawn . For this analysis , we used data collected from visits between August 2011 and November 2014 . The studies included passive and active follow-up of participants . Parents/guardians were encouraged to bring their children to designated study clinics for any illness . All medical care was provided free of charge , and participants were reimbursed for transportation costs . All children who reported fever in the previous 24 hr or were febrile at the time of the visit ( tympanic temperature >38 . 0°C ) were tested for malaria infection with a thick blood smear . Light microscopy was performed by an experienced laboratory technician who was not involved in direct patient care and verified by a second technician . Parasite density was calculated by counting the number of asexual parasites per 200 leukocytes ( or per 500 leukocytes , if the count was <10 asexual parasites/200 leukocytes ) , assuming a leukocyte count of 8 , 000/μl . A blood smear was considered negative when no asexual parasites were found after examination of 100 high-power fields . If the smear was positive , the patient was diagnosed with symptomatic malaria and received treatment with artemether-lumefantrine ( AL ) , the recommended first-line treatment in Uganda . Episodes of complicated or recurrent malaria occurring within 14 days of therapy were treated with quinine . In addition , routine evaluations were performed every 3 months , including testing for asymptomatic parasitemia using thick blood smears . Entomological surveys were also conducted every month at all study households . During these surveys , mosquitoes were collected using miniature CDC light traps ( Model 512; John W . Hock Company ) . Established taxonomic keys were used to identify female Anopheles mosquitoes . Individual mosquitoes were tested for sporozoites using an ELISA technique ( Kilama et al . , 2014 ) . All female Anopheles mosquitoes captured in Walukuba and Kihihi were tested; in Nagongera testing was limited to 50 randomly selected female Anopheles mosquitoes per household per night due to the large numbers collected . Therefore , for each household and/or site it was possible to calculate multiple entomological metrics , including the average human biting rate ( average number of female Anopheles mosquitoes caught in a household per day ) , the average sporozoite rate ( the average proportion of mosquitos that tested positive for Plasmodium falciparum ) and the entomological inoculation rate ( EIR , the product of the household human biting rate and the site sporozoite rate ) . The purpose of these analyses was to model and quantify the development of immunity against symptomatic malaria , as a function of age and exposure , measured by the household EIR . We modeled two specific types of immunity that have been previously described as components of immunity to malaria . We defined anti-parasite immunity as the ability to control parasite densities upon infection and anti-disease immunity as the ability to tolerate parasite infections without developing objective fever . Thus , for models of anti-parasite immunity , the outcome of interest was the parasite density recorded ( using thick blood smear ) at each parasite-positive study visit . For models of anti-disease immunity , the outcome of interest was the objective temperature recorded during parasite positive visits , conditional on the parasite density . In addition , we also modeled overall immunity against symptomatic malaria . For these analyses , the outcome of interest was the probability of presenting with fever given infection ( parasite positivity ) . In order to model the association between the outcomes and covariates of interest we used generalized additive models ( gams ) . Gams provide a good framework , as they allow for smooth non-linear relationships . Details on the specific models explored are provided in the supplementary material ( Appendix 1 ) . In summary , the models followed the following form . ( 1 ) Anti-parasite immunityLog10 ( Parasite density ) ijk=f ( ageijk , Log2aEIRj ) +ui+γj ( 2 ) Anti-disease immunityTemperatureijk=f ( ageijk , Log2aEIRj , Log10Parasite densityijk ) +ui+γj ( 3 ) Overall immunity against symptomatic malariaP ( symptomatic malaria upon infection ) ijk=f ( ageijk , Log2aEIRj ) +ui+γjwhere i is an index for individuals , j for households and k for specific visits . Thus , ageijk represents the age of child i from household j during visit k , and aEIRj represents the average annual EIR recorded for household j . We included the EIR as an average ( time-invariant ) covariate , as we were interested in modeling the impact of the average exposure to malaria over time on the development of clinical immunity . Therefore , our model implicitly assumes that malaria transmission has been relatively stable at these three sites . To account for lack of independence , all models included random effects at the individual ( ui ) and household ( γi ) levels . All our primary analyses included the full dataset . However , since results were consistent with a non-monotonic relationship between aEIR and the outcomes of interest , we also fit models stratified by aEIR ( aEIR ≥5 vs . aEIR <5 ) . All models were fitted in the R statistical framework using package mgcv ( R Core Team , 2016 ) . Best fitting models were selected based on Akaike’s Informaiton Criterion , but changes in the percent deviance explained are also presented . All the data used for these analyses as well as the R code used to reproduce the main study findings are available at https://github . com/isabelrodbar/immunity ( Rodriguez-Barraquer , 2018; ( copy archived at https://github . com/elifesciences-publications/immunity ) . Complete data from the 3 cohort studies are available in the ClinEpiDB website ( https://clinepidb . org/ce/app ) . Confidence bounds are presented in Figure 5—figure supplement 1 . | Malaria kills around 500 , 000 children every year . The disease occurs when an infected mosquito bites a human and passes on a Plasmodium parasite . One parasite in particular , Plasmodium falciparum , is responsible for most malaria-related deaths across the globe . A person can be infected by P . falciparum many times throughout their life . However , after children have had multiple infections , they become less likely to develop symptoms of malaria , such as high fever . In other words , they gradually acquire immunity . This immunity to malaria can come in two forms: “anti-parasite immunity” , where the body fights the parasites and keeps their numbers low; and “anti-disease immunity” , where the body is more likely to tolerate an infection without developing a fever . To date , scientists do not fully understand how either kind of immunity arises in children . Is it because they have simply been exposed to more malaria ? Or does being older and having a more mature immune system also help ? Now , Rodriguez-Barraquer et al . have followed more than 1 , 000 children living in places with high , moderate and low rates of malaria infection in Uganda . Over three years , regular blood samples were taken to see if the children were infected with P . falciparum . Mosquitoes were also collected from their houses to estimate how often the children were being bitten and infected . Using this information , Rodriguez-Barraquer et al . studied the different factors that affect how children develop anti-parasite and anti-disease immunity . Both types of immunity develop differently in places with high , moderate and low rates of infection , so being infected multiple times is important . Yet , the findings also show that growing older itself contributes to the development of immunity regardless of how often a child is infected . Children who get infected most often – in other words , those living in houses with the most mosquitoes – develop immunity faster than those who get infected at a moderate rate . Unexpectedly , however , children living in places with low rates of infection also develop immunity faster than those living in places with moderate rates . Understanding how children acquire immunity to malaria is important for people trying to control the disease . These results suggest that reducing rates of infection to very low levels may not interfere with development of immunity and may even improve it . However , future research should see if these findings apply to other parts of the world as well , and , if so , why children develop immunity faster in places with lower rates of malaria infection . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"epidemiology",
"and",
"global",
"health"
] | 2018 | Quantification of anti-parasite and anti-disease immunity to malaria as a function of age and exposure |
There is great interest in understanding human olfactory experience from a principled and quantitative standpoint . The comparison is often made to color vision , where a solid framework with a three-dimensional perceptual space enabled a rigorous search for the underlying neural pathways , and the technological development of lifelike color display devices . A recent , highly publicized report claims that humans can discriminate at least 1 trillion odors , which exceeds by many orders of magnitude the known capabilities of color discrimination . This claim is wrong . I show that the failure lies in the mathematical method used to infer the size of odor space from a limited experimental sample . Further analysis focuses on establishing how many dimensions the perceptual odor space has . I explore the dimensionality of physical , neural , and perceptual spaces , drawing on results from bacteria to humans , and propose some experimental approaches to better estimate the number of discriminable odors .
The perceptual space for human color vision has three dimensions . Experimental proof of this dates to the 17th century , when it was found that every color sensation can be matched by mixing together three primary lights , but not if only two lights are available ( Mollon , 2003 ) . Therefore every color sensation can be fully characterized by three numbers , namely the intensity of the primaries that match it . We now know that color vision is based on three kinds of cone photoreceptors in the retina that differ in their sensitivity to the wavelength spectrum of light . Any pattern of excitation among cones can be matched by an appropriate mix of three primary lights , and this is the basis for RGB color display devices . Cone signals get processed by several circuits in the retina and beyond . This system ultimately has limited resolution , and in practice human subjects can distinguish about 1–2 million different colors ( Masaoka et al . , 2013 ) . Clearly , a quantitative understanding of color perception has both energized the search for the underlying neural circuits and made possible the design of image display technologies that mimic reality . One would like to achieve a similarly satisfying understanding of human smell . Human olfaction begins with the binding of odor molecules to olfactory receptors , of which there exist ∼400 types ( Malnic et al . , 2004 ) . It is believed that these receptor types all differ in their relative sensitivity to various odorants . Two odors can only be distinguished if they cause different patterns of activity among these types . Thus the neural space of odors at the very input to the olfactory system would seem to have 400 dimensions , many more than encountered in the color system . But what is the perceptual space for odors ? The exact analog of the early color mixing experiments has not been done , but from a quantitative analysis of perceptual similarities it has been argued that the space of odors is dominated by just one or two dimensions ( Secundo et al . , 2014 ) , much fewer than the 400 dimensions at the level of sensory receptors . Similarly the number of discriminable odors is thought to be only 10 , 000 , although this estimate is largely anecdotal ( Gilbert , 2008 ) . Clearly there is a paradox that remains to be resolved pitting the perceptual space of human smell against the receptor space . On this background , a recent article claims that ‘Humans can discriminate one trillion odors’ ( Bushdid et al . , 2014 ) . If correct , this would dramatically reorient thinking in this field . Not least , such a result would dash any realistic hopes for olfactory displays that can mix any odor sensation from a small number of primaries . The result was nominated for ‘Breakthrough of the Year 2014’ and heavily promoted to the popular press . By now many people ‘know’ that humans discriminate a trillion odors , and that our color vision system pales miserably in comparison . Remarkably , these claims were based on an experimental study in which humans discriminated successfully only 148 pairs of odors . Clearly some mathematical method was needed to extrapolate from that number to trillions . Here I show that this mathematical method fails , and as a result the claims are without any basis . We will see that the data on human olfaction are equally consistent with a trillion discriminable odors and with just 10 ( and anything in between ) . Moreover , if the same method were applied to human color vision , one would conclude that humans can distinguish at least 1027 colors , in dramatic conflict with experimental evidence . Beyond correcting the erroneous claims , the analysis of failure in the Bushdid et al . study yields useful insights about the nature of the olfactory code . I follow with some suggestions for an approach to estimate the number of distinguishable odors .
The reader is encouraged to read the original report by Bushdid et al . ( 2014 ) . The following is a brief summary of the procedures used in that study . The space of all possible odor stimuli is huge . There are likely several hundred thousand distinct chemicals that smell . Any mixture of those chemicals is a point in odor space . The goal is to find how many of those mixtures produce distinct sensations . What is the largest collection of odors such that any one is discriminable from all the others ? The reported experiments were limited to a subspace spanned by 128 substances . From these primary odors the authors made mixtures by drawing a number ( N ) of primary odors and mixing those in equal parts . Each primary odor is either present or absent in the mixture . Some of these mixtures are very similar to each other , for example if they share 29 of the 30 primary components . Others are very different , for example when they share none of the primary components . For any two mixtures the number of unshared components can be defined as their ‘distance’ . The authors assume that the ability of humans to discriminate two odors improves systematically with this distance . The next step is to determine the critical distance at which two odors become discriminable . The authors probe this systematically by making pairs of mixtures with N components of which M are unshared , and testing those for discrimination by human subjects . Indeed they find that the probability of discrimination increases with M ( Figure 3B of Bushdid et al . , 2014 ) . They define the critical distance D as that separation M at which 50% of the mixture pairs are discriminable . With these assumptions , two odors closer than D tend to smell the same , whereas two odors separated by more than D will smell different . To determine how many discriminable odors there are , the authors ask how many regions of diameter D can fit in the original 128-dimensional odor space . This is analogous to the problem of packing spheres in high-dimensional spaces and the authors compute the number of packable spheres of diameter D by methods of combinatorics ( Figure 3D of Bushdid et al . , 2014 ) . This yielded the ‘one trillion’ in the title of their paper . Given a novel analysis method , it can be instructive to test it on a ‘simplest possible’ model , for which the desired answer is known and the calculations are easy . In the present context we need a model of olfactory processing that treats mixtures of many components , can compare pairs of such mixtures , and exhibits a clear performance limit in their discrimination . Imagine a toy microbe that lives in an environment with many odors and has receptors to sense all of them . Pure odors are either attractants ( providing a sensory input of +1 ) or repellents ( −1 ) . Following the procedure of Bushdid et al . ( 2014 ) , we choose at random 128 of those as primary odors and combine them to make mixtures containing 30 odors . The microbe responds to a mixture by simply summing the sensory input from the component odors . If the sum is less than −2 it says ‘yuck’; if it is greater than +2 it says ‘yum’; and from −2 to +2 it says ‘meh’ . Two odor mixtures are discriminable if the microbe responds differently to them . Now we make odor mixtures that share a certain number of odors and plot the fraction of discriminable mixtures vs the odor overlap ( Figure 1A , compare to Figure 3B of Bushdid et al . , 2014 ) . We find that the critical value of 50% discriminability is reached when the mixtures share 15 of the odors . This leads to an estimate of ∼9·1011 discriminable odors ( Figure 1B , C , see Figure 3D of Bushdid et al . , 2014 ) . So by the standards of the proposed analysis , this toy microbe can also discriminate 1 trillion odors . Yet we know by construction that it recognizes only three classes of odors . Thus one can produce at most three odors such that each can be discriminated from all the others . 10 . 7554/eLife . 07865 . 004Figure 1 . Model of olfaction in a toy microbe . ( A ) This 3-state olfactory system counts how many odors in the mixture are attractants vs repellents , and converts the result into three response categories ( see text ) . Two odor mixtures are discriminable if they cause different responses . A numerical simulation of the response to many odor mixtures yields the fraction of discriminable mixtures as a function of the number of odors , O , that they share . Mean ± SD over 1000 repeats using different random assignments of the primary odors . Horizontal dashes: criterion for critical distance ( 50% discriminable pairs ) . Vertical dashes: critical distance D = 30 − O = 15 . ( B ) Points in odor space separated by a distance D cause different responses at least half the time . Counting how many such points exist in the space is like trying to pack spheres of diameter D to fill the space as efficiently as possible . ( C ) The number of such spheres in 128-dimensional space as a function of the discriminable overlap O among 30-odor mixtures , computed by the formula given in Bushdid et al . ( 2014 ) . The value O = 15 from panel ( A ) yields ∼9 × 1011 spheres . DOI: http://dx . doi . org/10 . 7554/eLife . 07865 . 004 One gets the sense that something is amiss with the analysis procedure of Bushdid et al . If it fails so dramatically on a crude toy model of odor integration , should one trust it on more complex sensory systems ? Nevertheless it is instructive to evaluate the procedure on a more realistic case . For example , the toy microbe actually has a name for each odor in the space ( one of three possible names ) whereas the human subjects were not asked to name the odor . Furthermore , for any given mixture pair the microbe's response is deterministic , whereas humans varied in their response . Thus I will now test the analysis on a system in which one can more exactly replicate the human psychophysics methods used in the reported odor discrimination tests . A prominent claim in Bushdid et al . ( 2014 ) is that olfaction vastly outperforms color vision in terms of discriminable stimuli . Specifically the authors compare their own estimate of a trillion discriminable odors to the literature's estimates of ∼1 million discriminable colors . However , studies of color vision used a very different procedure to determine the number of distinct sensations ( more on this below ) . Here I ask what number the authors' methods would produce . Fortunately we know enough about the rules and mechanisms of color perception to simulate this with great confidence , so there is no need to actually perform new color discrimination trials . The space of all colored lights has infinite dimensions . Each light is characterized by its wavelength spectrum S ( λ ) , which specifies how much power exists at each wavelength λ . Because the wavelength can take on any positive real value , the S ( λ ) are functions of a continuous variable , and thus have infinite dimensions . However , as introduced above , the sensory space of human color vision has only 3 dimensions ( Wandell , 1995 ) . For each of the three cone types in the retina , the excitation is determined by the rate of photon absorption by its visual pigment . In turn , that rate is a linear function of the spectrum of the light , namely the projection of the light spectrum onto the absorption spectrum of the pigment . If two lights are mixed together , the resulting excitation is the sum of the effects of the individual lights . Therefore human color vision begins by projecting the infinite space of lights down onto a subspace of just 3 dimensions , spanned by the three cone excitations . This subspace has been probed extensively in psychophysical experiments . Typically the subject is shown two lights side-by-side and asked whether they appear different . By systematic probing of the 3-dimensional space one finds there are upward of 1 million discriminable lights , in the sense that any two of them will look different in a pairwise comparison ( Masaoka et al . , 2013 ) . For the purpose of simulation , I will therefore consider a space with three axes: R , G , and B ( Figure 2A ) , corresponding to the three cone excitations . For simplicity , the three variables will range from 0 ( dark ) to 1 ( bright ) , so the space is a unit cube . Every physical light stimulus is projected onto a vector in this unit cube . The color vision system adds some noise to that vector along each of the 3 dimensions . As a result , two lights are discriminable if their vectors are separated by more than the noise . The level of noise is chosen so that a difference of 0 . 01 along any dimension is discriminable , which gives 1 million discriminable vectors in the unit cube . 10 . 7554/eLife . 07865 . 005Figure 2 . Model of human color vision . ( A ) The RGB color cube with three of the 128 primary colors represented by vectors from the origin . Tick marks represent just noticeable differences , for example , along the R-axis . ( B ) The fraction of discriminable 30-light mixtures as a function of their overlap . Mean ± SD over 1000 repeats using different random assignments of the 128 lights . 30 lights per mixture , 20 mixture pairs per class , 26 subjects per pair . Horizontal dashes: criterion for critical distance . Vertical dashes: critical distance . ( C ) As in panel B but with the mixture components drawn at random from all possible directions in the space rather than from a preselected set of 128 primaries . The results are almost identical . DOI: http://dx . doi . org/10 . 7554/eLife . 07865 . 005 Now one can implement the procedures of Bushdid et al . ( 2014 ) : Choose at random 128 primary lights in this space and make them of equal intensity . Then produce mixtures of 30 lights from those primaries; design different classes of mixtures that vary in the number of shared components . Within each mixture class , present pairs of stimuli to the model and ask whether they can be discriminated , following an ‘odd-man-out’ procedure . Plot the fraction of discriminable pairs against the mixture overlap . Find the overlap at which that fraction is 50% , and take the number of unshared components as the critical distance D for discrimination . Compute how many spheres of size D fit in the original 128-dimensional space . From this simulation , I find that all 30-light mixtures are discriminable if they differ by as little as one component ( Figure 2B ) . This is to be expected . Each of the primary lights has a vector length of 1/30 ( see ‘Materials and methods’ ) . This means that a single unshared component can separate the two mixture vectors by more than 1/30 in the RGB space . But a separation of just 1/100 along any axis is discriminable . So the critical distance D = 1 . From this one calculates there are more than 1027 discriminable colors in the 128-dimensional space probed here ( Figure 1C , see Figures 3D and S1B of Bushdid et al . , 2014 ) . Actually , the number of discriminable colors by this argument is much higher , even infinite . To drop down to the 50% discriminability criterion used by the authors in defining D , we have to make much larger mixtures of ∼60 lights that differ by just one component . The number of possible mixtures of that kind is about 1037 . Furthermore , there is no reason to limit the starting space to just 128 primary lights . There is an infinity of spectra that are physically achievable lights , so we could have started with a subspace of arbitrarily high dimensionality . One can simulate infinite dimensionality by choosing for each of the mixtures a different random set of 30 normalized vectors from the cube ( again maintaining the specified overlap among mixture pairs ) . As shown in Figure 2C , the critical distance under those conditions is still 1 . Thus the methods of Bushdid et al . ( 2014 ) would conclude that humans can discriminate an infinite number of colors . That is much bigger than 1 trillion , so color vision would still win over smell , at least by the internal logic of this analysis . However , the number is also much bigger than the known experimental result , a few million . We can conclude that the analysis method fails the ‘positive control test’ , namely the application to a related problem with known solution . Given that the analysis of Bushdid et al . can vastly overestimate the number of discriminable stimuli , one wonders whether the human odor discrimination data in that report could equally be explained by a much smaller number of odor percepts . That is indeed the case . Let us consider a simple model of odor processing in which there is a very large number of odor stimuli and a large number of associated odor receptors . However , I will suppose that the nervous system ultimately projects all those odors onto a neural representation with a single dimension , and we can use the vectors on the unit circle for that ( Figure 3A ) . So the 128 primary odors map onto 128 unit vectors with random angles . Furthermore , we will suppose that a mixture of odors gets mapped into the sum vector of all the components , normalized again to unit length . Finally , the angle of this vector gets corrupted by some perceptual noise . Two vectors around the circle will be discriminable if they are separated by more than the noise . 10 . 7554/eLife . 07865 . 003Figure 3 . A model simulation of the human smell experiments . ( A ) Left: Each primary odor gets mapped into a unit vector ( e . g . , red , green , blue ) . Mixtures of odors get mapped into the normalized sum vector ( gray ) . Right: When a subject sniffs an odor vial , the odor angle is corrupted by Gaussian noise and a value is drawn from that distribution . Here three vials were presented , two ( A and C ) containing the identical odor and a third ( B ) a different odor . This produced response variables xA , xB , and xC . On this trial , xB and xC are closest to each other , so the subject ( incorrectly ) identifies A as the odd odor . ( B ) Discriminability of odor mixtures under this model ( compare to Bushdid et al . , 2014 , Figure 2C ) . Mixtures were simulated according to the reported procedure with 10 , 20 , or 30 components and varying overlap . Each mixture pair was presented to 26 subjects , and the fraction of correct identification determined across subjects . Box-and-whisker plot shows the distribution of that fraction with percentiles 10 , 25 , 50 , 75 , 90 . Average over 1000 repeats of the procedure with different random numbers . Red dashes: chance performance . Black dashes: criterion for discriminability ( 14/26 correct ) . ( C ) Fraction of discriminable mixtures as a function of their overlap ( compare to Bushdid et al . , 2014 , Figure 3D ) . This is the fraction of mixture pairs in each class that exceeds 50% correct identification across subjects ( above the black line in panel B ) . Lines are mean ± SD . Symbols are data from Bushdid et al . ( 2014 ) . The model used Gaussian noise with a SD of 0 . 4 radians . DOI: http://dx . doi . org/10 . 7554/eLife . 07865 . 003 This encoding model can simulate the reported human subject experiments in detail ( Figure 3B , C ) : For every ‘odd-man-out’ trial , three mixtures are presented to the model , of which two are identical and the third is a different target mixture . The model maps all three onto the unit circle , adds noise to each , and asks which two are closest to each other . Then it reports the more distant one as the odd odor . If that corresponds to the target odor , it is a ‘correct’ decision ( Figure 3A ) . Qualitatively , mixtures that share a large fraction of components produce sum vectors at nearby angles , which makes them less discriminable because of the perceptual noise ( Figure 3B , compare to Figure 2C of Bushdid et al . , 2014 ) . Again one can compute the fraction of discriminable odor mixture pairs , namely those correctly identified in >50% of all trials . In Figure 3C , I plot the results of this simulation along with the published data ( from Figure 3B of Bushdid et al . , 2014 ) . The model has a single parameter , namely the amount of perceptual noise; nothing else is adjustable . With a noise value of 0 . 4 ( SD of the Gaussian noise measured in radians ) the fit to the data is quite good . Almost all the data points are within the 1 SD error bars . At noise = 0 . 3 the model does somewhat better than the humans , at noise = 0 . 5 somewhat worse . For a noise value of 0 . 4 , how many mutually discriminable stimuli are there ? We should use the standard of Bushdid et al . ( 2014 ) for discriminability of an odor pair: 50% correct identification in the ‘odd-man-out’ odor test . With that criterion , one can place at most 10 vectors around the unit circle such that each is discriminable from its neighbors . Therefore , the published measurements are consistent with a model in which humans can distinguish just 10 olfactory stimuli from each other . Many of us have experienced more than 10 different smells , so these experiments did not come close to exploring the richness of human olfactory experience . Why is the lower bound from this odor psychophysics study so weak ? Actually , the result is just about expected from the effort expended . To confirm that 10 odors are mutually discriminable by brute force one has to compare each odor to every other one . The authors did 260 pairwise comparisons , of which only about half were discriminable . So one expects to find evidence for about 148 distinct odors , close to what was obtained . The ring model of Figure 3 assumes that odor percepts lie in a 1-dimensional space . If one allows for higher numbers of dimensions , then the predicted number of discriminable odors increases , approximately exponentially with the dimensionality of the perceptual space ( L Abbott , E Schaffer , and R Axel , personal communication ) . Since we do not know the dimensionality of odor space , the results of Bushdid et al . ( 2014 ) are equally consistent with 10 or a trillion discriminable odors or anything in between . In other words , the experiments fail to distinguish alternative hypotheses for the number of discriminable odors , even absurd ones that are clearly incorrect . In the following discussion I will use the term ‘percept’ to strictly mean the internal state of the sensory system at the stage where discrimination decisions are made . For us humans two things have the same percept when they look or smell the same . I will extend the term to toy models , bacteria and mice , without presuming that those creatures engage in the more contemplative aspects of perception . Two stimuli may be physically distinct , yet cause the same percept , and thus become indiscriminable , as is the case for many color spectra . Two stimuli can even be distinct at the level of sensory receptors , and still produce the same percept , as in cases to be discussed below . The failure of the method in Bushdid et al . ( 2014 ) occurs after the sphere-packing estimate ( Figure 1B ) . It involves a step that is never mentioned but implicit in the procedure: the assignment of odor percepts to the different spheres in stimulus space . There are at least two problems that can lead to enormous overestimates of the number of discriminable odors , First , the authors assume that every one of the spheres packed into the space corresponds to a different odor percept . But this is unwarranted . The measurement of critical distance for discrimination only ensures that neighboring spheres correspond to different percepts , it says nothing about more distant spheres . Thus the same odor percept may recur over and over again for physically distinct odor mixtures . The situation can be understood already in 2 dimensions . Figure 4A shows a set of close-packed pennies on the desktop . Every penny is different in color from its neighbors . More importantly every penny in the whole space has a different color from every other penny . If the pennies correspond to odor percepts one could legitimately say that this organism discriminates as many odors as there are pennies . However , there are many other ways to color close-packed pennies so that no adjacent ones have the same color . In fact , three colors are sufficient ( Figure 4B ) , so just three percepts could account for an infinite number of spheres in a 2-dimensional odor space . 10 . 7554/eLife . 07865 . 006Figure 4 . Coloring close-packed spheres in 2 dimensions so no nearest neighbors have the same color . ( A ) All colors different . ( B ) Three colors suffice . ( C ) Progression of colors in one dimension only . DOI: http://dx . doi . org/10 . 7554/eLife . 07865 . 006 One might object that this is an esoteric arrangement: What kind of olfactory system would produce this strange periodic recurrence of the same percept ? However , there is a perfectly natural color progression that also provides enormous savings of colors . In the arrangement of Figure 4C the percept stays the same along one dimension of the space and varies along the orthogonal dimension , yet all neighboring spheres are distinct . The number of pennies one can pack this way goes as the square of the number of colors available . In 128-dimensional space , the number of hyperspheres one can pack this way will go as the 128th power of the number of percepts . So the key logical flaw is the assumption that all the close-packed hyperspheres produce different odor percepts . One can trace this back to the unstated assumption that the odors should systematically become more discriminable at greater distances in this space , and equally in every direction . Here I showed that there is a simple and natural way to violate that assumption , namely if the percept depends only on a single function of the coordinates . Then the odors become more discriminable with distance along one dimension , but remain indiscriminable along all other dimensions . More generally , if the percepts live in some low-dimensional space ( as is the case , e . g . , for color vision and for the hearing of pure tones ) and then one embeds that space in 128 dimensions , this will produce a similarly efficient labeling of the close-packed spheres . This is what causes the astronomical overestimates in the analysis of Bushdid et al . ( 2014 ) when applied to the three model simulations above . The above arguments cannot fully explain the results obtained with the model of the 3-state microbe . Here there are only three possible percepts ( ‘yum’ , ‘yuck’ , and ‘meh’ ) that can be used to paint the hyperspheres . Yet , in 128-dimensional space every sphere has at least 256 nearest neighbors . Clearly it is impossible to distinguish a sphere from all its nearest neighbors with just three colors . This brings us to a second weakness of the method: the definition of the critical distance D . This is taken as the distance at which 50% of odor mixture pairs are discriminable . In other words , two points separated by D in odor space only need to produce a different percept 50% of the time . So when assigning percepts to the spheres in 128-dimensional space , we merely have to ensure that each sphere is distinct from 50% of its neighbors . That can be done trivially with just two colors , painting alternating spheres white and black . That makes every sphere different from 50% of all its neighbors . The 3-percept microbe can do a bit better , which is why discriminability rises to almost 2/3 at large distances ( Figure 1A ) . Would it help to raise the criterion for discriminability to a higher percentage of mixture pairs ? If one raises it to 90% then 10 odor percepts are sufficient to color the entire space of spheres . A popular estimate for the number of discriminable odors ( though without solid scientific basis , as reviewed in Bushdid et al . , 2014 ) is around 10 , 000 . To ensure that a sphere-coloring system uses at least 10 , 000 odors , one would need to raise the criterion for discriminability in the definition of D to 99 . 99% . So there needs to be a data point in Figure 3B of Bushdid et al . ( 2014 ) with an ordinate of 99 . 99% . This would require human subjects to perform several tens of thousands of pairwise comparisons in just one mixture class , a truly extraordinary experimental effort .
The arguments above illustrate that the mathematical method for exploring sensory spaces advocated in Bushdid et al . ( 2014 ) does not work . Fundamentally the extrapolation presumes that the space of odor percepts has at least 128 dimensions . Furthermore , the 128 primary odors chosen must represent ‘orthogonal’ directions in that space , so that the percept varies with distance in the same manner and independently along each of these dimensions . Nothing in the report suggests why we should believe this , and the assumptions seem implausible a priori . Violation of those assumptions , for example if the true dimensionality of odor space is much less than 128 , leads to dramatic failure of the estimate . How can one do better ? To obtain the number of discriminable odors , we need to determine the largest set of stimuli such that every stimulus can be discriminated from every other one , not just from nearest neighbors . Classic studies of color vision and tone hearing have achieved that . For color vision , the answer is approximately n = 1 million . How can one come to that conclusion without doing n2 discrimination tests ? This is possible only by taking advantage of the systematic structure of the perceptual space . For example , if light A looks more red than light B and B looks more red than C , then one can trust that A will look more red than C . The same applies in tone hearing: If tone A sounds lower than tone B , and B sounds lower than C , then A will sound lower than C . This transitivity of discrimination means that it is sufficient to measure discrimination of neighboring stimuli to ensure mutual discrimination of all stimuli in the set , which reduces the experimental burden from order ( n2 ) down to order ( n ) . Effectively these studies work in a low-dimensional subspace of all stimuli ( R , G , B for lights; frequency and amplitude for pure tones ) , within which the progression of the percepts is monotonic , such that transitivity applies . So success of these percept-counting studies relied entirely on recognizing early on that the perceptual space is low-dimensional . How to apply those insights to olfaction ? Clearly one needs to first answer ‘how many dimensions span the perceptual space of odors’ ? The number of 400 human odorant receptor types certainly sets an upper bound on the dimensionality of the perceptual space . And one commonly hears the argument: ‘Why would Nature make so many receptors , unless it is to discriminate all possible patterns of activation’ ? However , there may be another explanation: We need this many receptors simply to sense the molecules of interest , not to disambiguate all possible mixtures of those molecules . The bacterium Escherichia coli illustrates this idea: It has five different chemoreceptor proteins with different ligand spectra . In principle , E . coli could therefore analyze odor mixtures in 5-dimensional space and react differently to every possible mixture . But it doesn't . The outputs of all five receptors readily converge on a single variable , namely the concentration of the CheY signaling molecule ( Grebe and Stock , 1998 ) . This single variable controls the bacterium's locomotor decisions , so we may identify it with the odor percept in the present use of the term . E . coli projects the 5-dimensional receptor space onto a 1-dimensional perceptual space of attraction/repulsion . Why then does it need five receptors , including different receptor proteins for aspartate and for serine , both amino acids ? Presumably it is difficult to make a generic amino acid receptor with high sensitivity . Both the amino and the carboxyl ends of the molecule vary in their charge distribution depending on pH , and a single binding pocket directed at these regions would not offer sufficient affinity under all conditions . Instead , the two receptor proteins focus on other more stable portions of the ligand , but those are also unique between serine and aspartate . The same arguments apply to odorants in the human nose . The molecules of interest are there at micromolar concentrations or less , in a mucus soup of other components at millimolar concentration . To sense the odorants separately from the mucus , a receptor needs to bind them with high affinity . That means many contact sites between binding pocket and ligand , which in turn leads to selectivity for the shape of the ligand . Even if the olfactory system just wanted to distinguish odors along one dimension ( attractive/repulsive ) , it would be impossible to make a receptor that is selective just for the attractants . Instead Nature makes many receptors that are each selective for small groups of related molecules and then combines their signals appropriately using the nervous system . In this picture the dimensionality of receptor space is determined by molecular principles involving the number of ligands of interest , their relevant concentrations , the energetics of ligand binding , and the design limitations of protein structures . The dimensionality of perceptual space , on the other hand , is governed by behavioral and ecological constraints: the nature of olfactory cues in the environment , the kinds of decisions the animal makes based on odorants , and the need to associate new odors with unusual events . There is no principled reason that this perceptual space should have the same dimensionality as the receptor space . And we have the neural circuits of the olfactory system to create an arbitrary map from one space onto the other . Another sensory system serves to illustrate this difference between receptor space and perceptual space: touch . Every hair follicle on our skin contains a sensitive mechanoreceptor , several million altogether ( Zimmerman et al . , 2014 ) . This makes us highly sensitive to touch: we can reliably detect the bending of an individual hair almost anywhere on the body . But clearly we cannot discriminate all patterns of bent hairs . Brushing across your head twice in a row feels very much the same , even though it is certain to cause two different patterns of activity among the touch receptors . As for E . coli chemotaxis , there is a benefit to detecting many possible sensory inputs , but no need to discriminate all possible patterns of those inputs . Beyond arguments by analogy , some recent studies of human olfaction suggest that the perceptual space for odors may have rather few dimensions . Furthermore , the dominant axes of this perceptual space can be related systematically to the physical characteristics of odorous molecules ( Secundo et al . , 2014 ) . Another relevant observation is that mixtures containing many ( >20 ) diverse odorants tend to smell alike , even if they don't share any molecular components ( Weiss et al . , 2012 ) , a phenomenon that has been termed ‘olfactory white’ , in analogy to the ‘white’ percept associated with a mix of many colored lights . This suggests that the dimensionality of odor percepts may be around 20 or less . Table 1 summarizes the number of dimensions of various spaces discussed here . 10 . 7554/eLife . 07865 . 007Table 1 . Number of dimensions of various spaces involved in sensory discriminationDOI: http://dx . doi . org/10 . 7554/eLife . 07865 . 007Toy microbeRing modelHuman colorE . coli smellHuman smellStimuli∞∞∞∞∞Receptors∞∞35∼400Percepts11311–20 ? The symbol ∞ stands for ‘very large or infinite’ . A useful experimental approach might then be to rigorously measure the dimensionality of perceptual space at least at one point . For example , choose N primary odors , and consider arbitrary mixtures of those as the odor space . Define the ‘white point’ as the mixture of all those odors at half concentration , that is , the odor vector w = ( 0 . 5 , … , 0 . 5 ) . How does odor perception vary as the stimulus deviates a little from this point ? To first order , the discriminability d between the white odor at w and an odor at w + x will vary as a quadratic form of the deviation vector x , namely: d = xT S x . We want to know the sensitivity matrix S . By definition it is positive definite and symmetric , and thus has N ( N + 1 ) /2 unknown components . So one needs to measure the just-discriminable-distance along N ( N + 1 ) /2 directions from the white point . Clearly this is an experimental challenge , but it seems plausible at least for N = 20 . If so , then the structure of the matrix S can reveal the dimensionality: In particular , if it has just a few large eigenvalues , those identify the relevant directions in odor space . By contrast , if all eigenvalues are comparable , then the perceptual space has dimensions higher than N . Animal studies can play an important role here . Mice are readily trained to distinguish odors , even closely related mixtures . More importantly , they offer an opportunity to stimulate the receptor neurons directly , by optogenetic activation of the olfactory bulb ( Spors et al . , 2012 ) . In a suitably engineered animal one could drive arbitrary activation patterns of the different olfactory receptor types by shining patterned light onto the glomeruli in the olfactory bulb . This approach promises several benefits in a study of odor dimensions: First , it does away with the tedium of olfactory stimulation , such as mixing dozens of vapors , switching valves , flushing tubes , and waiting for odors to dissipate . Using light , a different combination of receptors can be driven with millisecond precision and at high repetition rates . Also , this method allows patterns of stimulation that may not ever occur with natural odorants; one could then test if the perceptual space differs from receptor space , and is shaped to the ecology of real odors . In a way , one can think of the olfactory bulb surface as a retina for the smell system . Olfactory objects produce spatio-temporal patterns on this surface , and the downstream neural circuits are busy identifying , discriminating , or learning those spatio-temporal patterns . The optogenetic approach simply takes the analogy one step further by using light as a stimulus . At the same time , there is a parallel effort ongoing in vision science to determine the dimensionality of human pattern vision . It is clear already that the number of dimensions is much lower than the number of receptors on the retina . One can make pairs of visual images that have very different effects on the retina , but look the same to human subjects ( Freeman and Simoncelli , 2011 ) . And a systematic approach to measuring dimensionality of pattern vision is beginning to yield results ( JD Victor , personal communication ) . Based on these developments , I suggest that pursuing the analogy of smell to pattern vision will be more fruitful than the analogy to color vision . Regardless of approach though , determining the dimensionality of the space of odor percepts is a precondition to estimating the number of distinct percepts . The recognition that color space is three-dimensional has had enormous impact in science , art , and technology , as anyone reading this on a color monitor will confirm . The search for a similar basis set for odors has fascinated scientists , engineers , and perfumers for some time ( Gilbert , 2008 ) . Even proving that a low-dimensional basis does not exist would be a major advance .
All simulations and graphics were produced with Igor ( Wavemetrics , Tigard , OR ) . Annotated code is available as a supplement to this article ( Source Code 1 ) . Simulation of the three models ( Figures 1–3 ) followed the same process as the human odor tests performed in Bushdid et al . ( 2014 ) :Selection of the 128 primary stimuli . These were drawn at random from the stimulus space . Figure 1: binary distribution over ( −1 , +1 ) . Figure 2: Uniform distribution of R , G , and B over [0 , 1] , followed by normalization to a length of 1/30 . Figure 3: Uniform distribution on the unit circle . These are conservative choices , in that a random set of primaries does not cover the stimulus space particularly well , and thus will produce mixtures that occupy only a portion of the space . This will therefore underestimate the number of discriminable stimuli . By contrast Bushdid et al . chose odors that were ‘well distributed in both perceptual and physicochemical stimulus space’ . Creation of mixtures . Pairs of mixtures containing N primaries of which O are shared were created by choosing at random 2N-O components from the set of primaries , summing the first N to mixture 1 , and the last N to mixture 2 . For every class of mixtures ( i . e . , combination of N and O ) , 20 mixture pairs were created . The rules for combining the stimulus values in a mixture were as follows: Figure 1: Simple addition of the binary stimulus values . Figure 2: Addition of the primary vectors . Figure 3: Addition of the unit vectors followed by normalization to unit length . This normalization emulates the elimination of odor intensity cues in Bushdid et al . ( 2014 ) . Discrimination test . Every pair of mixtures was presented to the model to determine whether it was discriminable or not . Figure 1: The model classifies every mixture of N = 30 odors into the percepts ‘yum’ , ‘meh’ , or ‘yuck’ depending on whether the sum of stimulus values is >2 , from 2 to −2 , or <−2 respectively . Two mixtures that fall into different classes are discriminable . Figure 2: The model performs an odd-man-out discrimination test among three samples as done for human subjects in Bushdid et al . ( 2014 ) . Two of the samples contain mixture 1 and the third contains mixture 2 . Perceptual noise was simulated by adding a Gaussian random variable to each of the three coordinates of the mixture vectors . The resulting three sample vectors are inspected and the two with the smallest distance are declared to be the same . For every mixture , this is performed 26 times , with different draws from the noise distribution . A mixture pair is declared discriminable if the model gives the correct response on >50% ( 14 or more ) of those trials ( note chance performance is 1/3 ) . In Bushdid et al . ( 2014 ) these 26 trials were performed by 26 different human subjects . There was some indication that different subjects had different abilities , but the analysis merged them all . In my simulation , all trials were done with the same amount of perceptual noise . The noise magnitude was chosen so that stimuli separated by a distance of 0 . 01 are just discriminable . This leads to ∼1 million distinct percepts in the RGB stimulus space , a conservative choice , because the empirical estimates of that number are somewhat larger . Figure 3: This followed the same odd-man-out procedure as for Figure 2 . Perceptual noise was simulated by adding a Gaussian random variable to the angle of the mixture vector ( Figure 3A ) . Discriminability of mixture classes . For each class of mixtures I computed what fraction of the 20 were discriminable ( by the criteria in 3 ) , and plotted this against the mixture overlap ( Figures 1A , 2B , C , 3C ) . To estimate the reliability of the simulation , the entire procedure was repeated 1000 times ( with different random numbers ) and the plots show the mean and standard deviation of the outcome . Figure 3C shows my simulation along with the data from the Bushdid et al . 's human experiments . Estimating the number of discriminable stimuli . From the graph of discriminability vs overlap , the critical distance D was taken to be the number of unshared stimuli that allows 50% discriminability in that class of mixtures . The number of regions of diameter D that can fit into the stimulus space was computed using the formula presented in Bushdid et al . ( 2014 ) ( Figure 1C ) . If there are C primaries total and N primaries per mixture , then the number S of such regions is claimed to beS= ( CN ) ∑R=0D/2 ( NR ) ( C−NR ) . | Scientists are interested in the number of colors , sounds and smells we can distinguish because this information can shed light onto how our brains process these senses both in health and disease . It is relatively straightforward to determine how many colors we can see or sounds we can hear because these stimuli are well defined by physical properties such as wavelength . We know the range of wavelengths that the eye can see or the ear can hear , and we can also understand how two such stimuli ( e . g . , red and blue ) are arranged perceptually ( think of a color wheel ) . It is harder , however , to do the same for smell because most ‘olfactory stimuli’ consist of mixtures of different odor molecules . Moreover , we understand much less about how olfactory stimuli are arranged perceptually . In 2014 researchers at Rockefeller University reported that humans can distinguish more than one trillion smells from one another . To calculate this number the researchers tested the ability of human subjects to discriminate between mixtures of different odor molecules . Each mixture consisted of 10 , 20 or 30 molecules selected from a chemical library of 128 different odor molecules . Since each mixture of 10 molecules could contain any 10 of the 128 molecules , more than 200 trillion combinations were possible; the number of possible combinations for the 20- and 30-molecule mixtures were even higher . The aim of the experiment was to identify—by sampling from this very large number of combinations—the number of molecules that two mixtures could have in common and still be distinguishable to the typical person . The Rockefeller team used this number and a geometrical analogy to conclude that humans could discriminate at least 1 . 72 trillion odors , which was much higher than expected from previous reports and anecdotes . Now Meister reports that the claims made in the Rockefeller study are unsupported because of flaws in the design and analysis of the experiment . In particular , there are flaws in the mathematical methods used to infer the potential number of all smells that humans can discriminate from the numbers of experimental samples tested . Meister also applies the Rockefeller approach to a well-understood sensory system—the vision system—and finds that it predicts that humans should be able to discriminate an infinite number of colors: however , it is widely agreed that humans can only discriminate several million colors . In a separate paper Gerkin and Castro also report that the 1 . 72 trillion smells claim is unjustified . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"computational",
"and",
"systems",
"biology",
"neuroscience"
] | 2015 | On the dimensionality of odor space |
People who inject drugs ( PWID ) account for some of the most explosive human immunodeficiency virus ( HIV ) and hepatitis C virus ( HCV ) epidemics globally . While individual drivers of infection are well understood , less is known about network factors , with minimal data beyond direct ties . 2512 PWID in New Delhi , India were recruited in 2017–19 using a sociometric network design . Sampling was initiated with 10 indexes who recruited named injection partners ( people who they injected with in the prior month ) . Each recruit then recruited their named injection partners following the same process with cross-network linkages established by biometric data . Participants responded to a survey , including information on injection venues , and provided a blood sample . Factors associated with HIV/HCV infection were identified using logistic regression . The median age was 26; 99% were male . Baseline HIV prevalence was 37 . 0% and 46 . 8% were actively infected with HCV ( HCV RNA positive ) . The odds of prevalent HIV and active HCV infection decreased with each additional degree of separation from an infected alter ( HIV AOR: 0 . 87; HCV AOR: 0 . 90 ) and increased among those who injected at a specific venue ( HIV AOR: 1 . 50; HCV AOR: 1 . 69 ) independent of individual-level factors ( p<0 . 001 ) . In addition , sociometric factors , for example , network distance to an infected alter , were statistically significant predictors even when considering immediate egocentric ties . These data demonstrate an extremely high burden of HIV and HCV infection and a highly interconnected injection and spatial network structure . Incorporating network and spatial data into the design/implementation of interventions may help interrupt transmission while improving efficiency . National Institute on Drug Abuse and the Johns Hopkins University Center for AIDS Research .
People who inject drugs ( PWID ) bear a disproportionate burden of human immunodeficiency virus ( HIV ) and hepatitis C virus ( HCV ) infection and account for some of the fastest-growing epidemics globally . While there has been substantial progress in combating these epidemics , HIV and HCV prevalence and incidence among PWID remain high , especially in South , Southeast and Central Asia , and Eastern Europe ( DeHovitz et al . , 2014 ) . Individual-level factors for infection are well established , but less is known about network and spatial drivers of HIV and HCV among PWID , especially from low- and middle-income settings . Network-based interventions for HIV and HCV are increasingly being implemented; however , they are seldom informed by empirical sociometric network data and more often informed by models ( Zelenev et al . , 2018; Hellard et al . , 2015; Rolls et al . , 2013; Hellard et al . , 2014 ) . Existing data derive from small egocentric network studies of ‘indexes’ or ‘egos’ and their immediate connections ( first degree ‘alters’ ) ( Costenbader et al . , 2006; Latkin et al . , 2009; Latkin et al . , 2011; Latkin et al . , 2013; Latkin et al . , 2010 ) . Few studies have examined the broader sociometric network , which captures the alters of those first degree alters ( second degree alters of the index ) , and so on , providing a more complete representation of the underlying network ( Figure 1 ) . Even less is known about the overlap of these egocentric and sociometric networks in space . While spatial heterogeneity of HIV/HCV burden has been previously described in high-income settings ( Des Jarlais et al . , 2018 ) , less is known about whether transmission is driven more by injection partner connections versus spaces/venues people reside and/or inject within . Incorporating spatial data , specifically in the form of injection venues , can further inform whether independent sociometric injection networks overlap spatially to more comprehensively examine the distribution of HIV/HCV and assess the role of space in the diffusion of disease . This manuscript aims to characterize egocentric , sociometric , and spatial network structures in a community-based sample of 2512 PWID in New Delhi , India and examine the role of individual- and network-level correlates of HIV and HCV infection .
The ‘Spatial Network Study’ is an ongoing dynamic longitudinal cohort of PWID in New Delhi , India established to understand the role of networks in transmission of HIV and HCV among PWID . New Delhi , the capital city of India , is estimated to be home to ~86 , 000 PWID ( Ambekar et al . , 2019 ) with HIV prevalence ranging from 13 . 5% to 35 . 8% ( Mehta et al . , 2015; Lucas et al . , 2015 ) and HCV prevalence ranging from 42 . 4% to 90% ( Solomon et al . , 2015; Solomon et al . , 2019a; Solomon et al . , 2019b ) . With the exception of index participants , all participants were recruited via a name generator network referral methodology . Participants completed a baseline assessment and were invited to complete semi-annual follow-up visits . This manuscript presents baseline data from this cohort . Recruitment of the cohort was initiated with two indexes in November 2017—eight more indexes were included later to account for variability in type of drug injected , marital status , and zip code of residence/injection . All 10 indexes were selected from a cross-sectional sample of PWID in New Delhi accrued for an evaluation assessment of a cluster-randomized trial ( ClinicalTrials . gov Identifier: NCT01686750 ) ( Solomon et al . , 2019b ) . When a participant enrolled in the Spatial Network Study , whether the initial 10 indexes or subsequent recruits , they were asked to recall the names of people with whom they injected in the prior month ( regardless of whether they shared injection paraphernalia ) . In addition , they were asked to provide identifying information about each named network member ( e . g . , scar on left hand , one finger missing on right hand ) and a factoid about each named partner ( e . g . , ‘his wife’s name is Priyanka’ ) . Each participant was then provided with a referral card for each named injection partner and was asked to invite them to participate in the study . When these recruits visited the study site , their name and identifying information were compared against the information previously provided . If the information matched , they were enrolled and asked to name , describe , and recruit people with whom they injected in the prior month ( recruit’s egocentric network and the index’s sociometric network ) . Recruitment continued until the desired sample size ( ~2500 ) was reached . Biometric data ( fingerprint scans ) was used to identify duplicates and establish cross-network linkages ( if the same participant was recruited by two different participants ) . The fingerprint scans were converted to unique hexadecimal codes and stored as described previously ( Solomon et al . , 2019b ) —no images were stored . The eligibility criteria varied depending on if the participant was an index or a recruit . Index participants had to ( 1 ) be ≥18 years of age , ( 2 ) provide written informed consent , ( 3 ) report a history of injecting drugs for non-medicinal purposes in the prior 24 months; and ( 4 ) have consented to be recontacted from the prior cross-sectional sample in 2016 . The eligibility criteria for recruits were ( 1 ) ≥18 years of age , ( 2 ) provide written informed consent , ( 3 ) recruited to participate in the study via a network referral card , ( 4 ) match description provided by their recruiter , and ( 5 ) not identified as a duplicate participant by biometric ( fingerprint ) match . Participants under the age of 18 were excluded since the legal age of consent in India is 18 years . There were no exclusions based on gender or sexual identity . Baseline study visits began with informed consent and referral card validation that included matching the factoid/identifying characteristic provided by the recruiter , followed by biometric registration and identification of duplicates . Participants that were identified as duplicates , that is , previously enrolled in the cohort , were not enrolled again; however , these data were used to add additional edges ( injection partner connections ) to the network . Participants then completed the survey and blood draw , followed by rapid HIV and HCV antibody testing on-site with appropriate pre-test counseling and referrals , as applicable . Participants were provided with referral cards to recruit each of their named injection partners into the study . Participants received INR 300 ( USD 3 . 94 ) as compensation and could earn an incentive of INR 50 ( USD 0 . 66 ) per named network partner they referred who was eligible and completed study procedures . At baseline , participants completed an interviewer-administered electronic survey that captured information on sociodemographics , substance use and risk behavior , sexual risk behaviors and characteristics , social support , quality of life , and access to HIV and HCV services , among others . The survey also captured detailed information about their egocentric injection network and this data was used to generate referral cards . In addition to injection network data , participants were also asked to list venues where they had injected in the prior 6 months . A list of common injection venues ( latitude/longitude ) was pre-populated and available on maps of Delhi to select from—participants also had the option to add a new venue if they injected at a venue that was not listed . On-site rapid HIV antibody testing was performed in line with the current standard of care for HIV diagnosis in India using three different kits: Determine HIV-1/2 ( Alere Medical , USA ) ( Sensitivity: 99 . 9%; Specificity: 99 . 8% ) , First Response HIV card test 1-2-O ( Premier Medical , India ) ( Sensitivity: 100%; Specificity: 99 . 9% ) , and Signal HIV-1/2 ( Arkray Healthcare , India ) ( Sensitivity: 100%; Specificity: 100% ) . Rapid HCV antibody testing was performed using the Aspen HCV One Step Test Device ( Aspen Diagnostics , India ) ( Sensitivity: 99 . 8%; Specificity: 99 . 9% ) . All residual samples were shipped to the central lab in Chennai for RNA quantification and storage . HIV RNA was quantified in all HIV antibody-positive samples using the Abbott HIV-1 RNA Real-Time PCR ( Abbott Molecular Inc , Des Plaines , IL , USA ) with a lower limit of quantification ( LLOQ ) of 150 copies/mL . All HCV antibody-positive samples were tested for the presence of HCV RNA with the Real-Time HCV assay ( Abbott Molecular Inc , Des Plaines , IL , USA ) with an LLOQ of 30 IU/mL . Statistical analyses were carried out in Python ( v3 . 7 . 3 ) and R ( v3 . 5 . 1 ) . Individual and network variables were analyzed for an association with prevalent HIV and active HCV infection ( HCV RNA positive ) using univariable and multivariable logistic regression . The Boruta ( Kursa and Rudnicki , 2010 ) random forest feature selection algorithm was used to explore candidate factors . Variables were considered for inclusion in multivariable models if they held biological/epidemiological significance or had significant associations in univariable models or significant variable importance scores from random forest ( p<0 . 05 ) . Networks were constructed with Python using NetworkX ( Hagberg et al . , 2008 ) and network variables , that is , number of infected injection partners ( first degree alters ) and network distance from an infected alter/venue , were calculated from the sociometric network ( containing only person nodes ) . Network distance from an infected alter was calculated such that a distance of zero signifies a direct connection , a distance of one signifies one uninfected person along the shortest path between a given participant and infected alter . Similarly , for network distance from an injection venue , a distance of zero signifies a direct connection to the venue and a distance greater than zero signifies the number of person nodes along the shortest path between a participant and venue . Networks were visualized using Gephi ( https://gephi . org ) and interactive networks were created using Sigma . js ( http://sigmajs . org ) . Spatial nodes in the network were placed by GPS coordinates to be spatially congruent with their geographic position under a Mercator map projection . Person-nodes were placed using a degree-dependent force-directed algorithm . The study protocol was approved by Institutional Review Boards at Johns Hopkins Medicine ( IRB00110421 ) and the YR Gaitonde Centre for AIDS Research and Education in India ( YRG292 ) . All participants provided written informed consent .
2502 PWID were recruited by 10 indexes ( total n=2512 ) . A median one referral card was provided to each participant ( range: 0–6 ) and 75% ( 2437/3244 ) of referral cards were returned . As recruitment continued , the sociometric networks of 6 out of the 10 index participants merged into one larger network resulting in a total of 5 discrete sociometric networks ( Figure 2; Figure 2—figure supplement 1 ) . The median degree/egocentric network size was 2 ( interquartile range [IQR]: 1–3 ) . The sociometric network diameter was 39 , and the average path length was 14 . Participants identified a total of 181 unique injection venues , defined as any public space where two or more PWID report injecting drugs together in the prior 6 months , spanning a 20-km radius in New Delhi , India . The median number of venues participants reported injecting in the prior 6 months was 3 ( IQR: 2–6 ) . The five discrete sociometric networks depicted in Figure 2 merged into one sociospatial network ( Figure 3 ) when accounting for social and spatial ties between participants ( interactive version of figure available at https://github . com/sclipman/sociospatial-baseline , copy archived at swh:1:rev:f22127448e931699530d02475043b2279279d67f; Clipman , 2021 ) . The sociospatial network diameter was 8 , and the average path length was 3 . 3 , signifying a higher efficiency of network transmission when considering spaces . The median age of the 2512 participants was 26 years and 2489 ( 99% ) were male—20 cisgender women , and 3 transgender women were recruited ( Table 1 ) . A total of 218 participants ( 9% ) had at least high school education and 19% reported same-sex behavior . Buprenorphine and heroin were the most commonly injected drugs—2411 ( 96% ) and 1150 ( 46% ) reported ever injecting buprenorphine and heroin , respectively , and 1518 ( 60% ) reported ever sharing injection paraphernalia . The median duration of drug use in the sample was 5 years ( IQR: 2–10 ) . 2499 participants reported injecting at least once in the prior 6 months with a median injection frequency of 360 times in the prior 6 months ( IQR: 180–540 ) . The demographic and risk characteristics of the indexes are provided as a table in Supplementary file 1 . Baseline HIV prevalence was 37 . 0% ( 928/2506 ) , and 92% of these participants had detectable HIV RNA . Out of 928 HIV-positive participants at baseline , 65% were directly connected with at least one other HIV antibody-positive PWID ( Figure 2 ) ; median network distance to another HIV antibody-positive PWID was 0 ( range: 0–3 ) . At least one HIV-positive person reported injecting at 155 ( 86% ) of the 181 injection venues identified by participants ( Figure 3—figure supplement 1a ) ; all participants were directly connected to at least one venue containing an HIV-positive person . Venue #40 was the most frequented injection venue ( see Figure 3—figure supplement 2 for distribution ) —1219 ( 49% ) of all participants and 565 ( 60 . 0% ) of HIV-positive participants reported injecting at this venue . Participants who injected at venue #40 also reported , on average , 32% more injections in the prior 6 months than those who did not report injecting at this venue ( p<0 . 001 ) . Individual-level variables positively associated with prevalent HIV in multivariable logistic regression included younger age , lower education , experiencing homelessness , decreased sexual activity , sharing syringes , increased injection frequency , and injecting heroin and buprenorphine ( Table 2 ) . Network-level factors remained highly statistically significant even after adjusting for individual-level correlates . At the egocentric level , odds of prevalent HIV increased by 20% for each additional HIV-positive alter ( adjusted odds ratio [AOR]: 1 . 20; 95% confidence interval [CI]: 1 . 08–1 . 34 ) . At the sociometric level , likelihood of HIV infection decreased by 13% with each additional uninfected person between a participant and infected alter ( AOR: 0 . 87; 95% CI: 0 . 82–0 . 95 ) . Injecting at venue #40 , which represents a participant’s immediate spatial network , was positively associated with prevalent HIV ( AOR: 1 . 50; 95% CI: 1 . 24–1 . 82 ) after adjusting for individual- and network-level correlates . The sociospatial network parameter was also independently associated with HIV infection; odds of infection reduced by 14% for each additional person separating a participant from venue #40 ( AOR: 0 . 86; 95% CI: 0 . 82–0 . 91 ) . Sociometric and sociospatial network parameters were significantly associated with HIV even after accounting for the egocentric parameter . Baseline anti-HCV antibody prevalence was 65 . 1% ( 1634/2512 ) , and out of 1477 samples with HCV RNA data , 80% had active HCV infection ( detectable HCV RNA ) . The majority of participants were unaware of their HCV status , only 4% ( 104 ) reported ever being previously tested for HCV , and nine individuals reported ever testing positive ( all received or are currently taking treatment ) . Therefore , instances where a person had anti-HCV antibodies but no HCV RNA most likely represent natural clearance of HCV infection . A total of 897 ( 35 . 7% ) participants had evidence of HIV/HCV co-infection ( HIV and anti-HCV positive ) ; of these , 658 ( 73 . 4% ) were HCV RNA positive . Out of 1634 anti-HCV positive participants at baseline , 86% were directly connected with at least one other anti-HCV positive PWID , and out of 1178 HCV RNA positive participants at baseline , 74% were directly connected with at least one other HCV RNA positive participant . The mean network distance from a participant with active HCV infection ( HCV RNA positive ) differed significantly by HCV infection status ( one-way ANOVA; p<0 . 001 ) . Among persons with active HCV infection , the mean network distance to another participant with active HCV infection was 0 . 59 compared to 0 . 72 for anti-HCV positive persons with undetectable HCV RNA and 0 . 90 for anti-HCV negative participants . Betweenness centrality was 1 . 42 times higher on average among the 1178 HCV RNA positive participants compared to the 878 anti-HCV negatives ( two-sample t-test; p<0 . 01 ) . In addition , persons with active HCV infection had significantly higher degree of centrality ( p<0 . 01 ) . A total of 172 ( 95% ) injection venues contained at least one anti-HCV positive person ( Figure 3—figure supplement 1b ) ; all participants were directly connected to at least one venue containing an anti-HCV positive person . Out of 1219 participants who injected at venue #40 , 942 ( 77 . 3% ) were anti-HCV positive; HCV RNA testing was available on 868 , 79 . 7% of whom had detectable HCV RNA . Similar individual- and network-level correlates associated with prevalent HIV were associated with active HCV infection ( Table 3 ) . The odds of HCV RNA positivity increased by 21% with each additional HCV RNA positive first degree alters ( AOR: 1 . 21; 95% CI: 1 . 10–1 . 34 ) and decreased by 10% with each additional uninfected person along the shortest path to an HCV RNA positive participant ( AOR: 0 . 90; 95% CI: 0 . 82–0 . 99 ) . Injecting at venue #40 was the strongest correlate , increasing the odds of HCV RNA positivity by 69% ( AOR: 1 . 69; 95% CI: 1 . 40–2 . 03 ) , and each additional person between a participant and venue #40 reduced the odds of current HCV infection by 10% ( AOR: 0 . 90; 95% CI: 0 . 85–0 . 97 ) . This sociospatial parameter was statistically significant even after accounting for the egocentric parameter .
In this sample of PWID in New Delhi , India , we observed an extremely high burden of HIV and HCV infection and strong associations between HIV and HCV within not only an individual’s immediate egocentric network but also broader sociometric , spatial , and sociospatial networks that incorporate indirect ties . These data are among the first to elucidate sociometric and spatial injection networks of PWID from a low- and middle-income country and provide critical insights into the design of HIV and HCV programming . Empirical network data among PWID have often been limited to egocentric network data , which only capture information on individuals and direct contacts ( Costenbader et al . , 2006; Latkin et al . , 2009; Latkin et al . , 2011; Latkin et al . , 2013; Latkin et al . , 2010 ) . These studies have shown that network instability or turnover in PWID’s injection partners promote HIV transmission . Limited sociometric data that exist come primarily from non-PWID populations in developed country settings and support the importance of understanding network connections beyond direct ties . For example , a seminal study assessing HIV and STI transmission in Colorado Springs found that HIV risk appeared low based on individual-level or egocentric network data , but sociometric data revealed risk to be higher than anticipated , with most individuals being within a few steps of an HIV infected person ( Klovdahl et al . , 1994; Rothenberg et al . , 1995 ) . Other studies in the United States have shown that sociometric networks can propagate HIV , with core individuals of large network components serving as centers of high-risk behavior and pockets of infection that could be targeted by network-based interventions ( Friedman et al . , 1997; Young et al . , 2013 ) . The sociometric network presented here provides further support for transmission within large network components in a low- and middle-income country setting . For both prevalent HIV and active HCV infection , we found that while having direct ties with HIV/HCV infection was associated with prevalent HIV/HCV , sociometric factors such as network distance to an infected alter were also independently associated even after accounting for direct ties . Among HCV RNA positive persons , the average sociometric network distance to another HCV RNA positive participant was significantly shorter compared to those of anti-HCV positive persons with undetectable HCV RNA or anti-HCV negative persons , supporting that network proximity to PWID with HCV RNA infection indicates higher likelihood of reinfection . These analyses further contribute to available network data by overlaying sociometric network data with information on injection venues . Prior studies , including some among PWID , have shown that physical spaces play an important role in the spread of disease , but have not examined spread through sociometric injection networks or associations with indirect connections to spaces ( Zelenev et al . , 2016; Gesink et al . , 2014; Rudolph et al . , 2017; Logan et al . , 2016 ) . The incorporation of space accounts for undocumented connections between participants in a space as well as spatial factors themselves ( e . g . , access to harm reduction services ) . Logan et al . demonstrated this in a sample of 600 participants ( including 303 PWID ) from Winnipeg , Canada , defining a geographically and socially cohesive community through which infections spread and identifying key venues driving such spread ( Logan et al . , 2016 ) . This is one of the few studies that integrated space with injection network data but represents a small sample from a developed country setting where HIV burden was significantly lower and the social context of drug use is very different . The idea that the sociospatial risk network may be the most relevant for transmission of disease is particularly salient with the increased focus on network-based interventions to reduce HIV and HCV transmission and optimize care outcomes while conserving resources . These findings suggest that in some settings , the sociospatial network may explain the majority of disease spread and interventions targeted at key spaces have the potential to interrupt transmission across a network and impact an entire city . For example , in this sample , it could be hypothesized that blanket coverage of venue #40 with treatment and pre-risk exposure prophylaxis , could impact transmission across New Delhi given the strong association of venue #40 with prevalent HIV among PWID in New Delhi . Traditionally network-based interventions have relied heavily on social diffusion . For example , for HIV , network members are used to improve retention to antiretroviral therapy and improve viral suppression ( Klovdahl et al . , 1994 ) , and ‘deep chain’ respondent-driven sampling is being used to identify undiagnosed or out-of-care HIV-positive men who have sex with men in the United States ( Rothenberg et al . , 1995 ) . For HCV , egocentric network-based treatment approaches have been identified as the optimal approach to deliver therapy while minimizing reinfection ( Rolls et al . , 2013; Hellard et al . , 2014 ) . However , if networks are highly interconnected especially within other PWID at a venue as observed in New Delhi , failure to incorporate space in the consideration of a network could result in high rates of HCV reinfection . In addition to network-level factors , these findings further reinforce the importance of well-established individual-level factors such as needle sharing , injection frequency , and experiencing homelessness . These associations and the limited uptake of harm reduction in this sample support continued efforts to expand harm reduction in this population . A key challenge in delivering services to this population is the high prevalence persons experiencing of homelessness which has not been previously demonstrated among PWID in India . PWID reporting homelessness , unstable housing , and migration may experience unmet needs for services and further disease transmission through socially and spatially dynamic networks . A limitation of this cross-sectional analysis is that the reported networks and injection venues may not necessarily represent network members or venues where participants acquired HIV and/or HCV infection; however , this would likely bias observed associations toward the null and attenuated associations of sociometric and spatial factors . Further , the cross-sectional nature limits the ability to examine temporal associations , but the consistency of associations with active HCV infection suggests that these network factors may impact onward transmission . All responses related to drug use , network members , and spaces were self-reported and subject to social desirability and recall bias; to minimize bias , all interviewers were trained on optimal interviewing techniques . About 25% of referral coupons were not returned suggesting that the networks presented in these data may be incomplete; however , the response rate of 75% is higher than what has been seen in other network studies ( Kimani et al . , 2014; Johnston et al . , 2008 ) . PWID under the age of 18 were excluded from the study due to the legal age of consent in India; therefore , these individuals are not represented by the network topology . Statistical analyses assumed that observations are independent conditional on individual-level covariates . This assumption is likely to be violated to some extent , but violations are not expected to bias point estimates ( they would result in underestimated standard errors ) . Limitations notwithstanding , these data highlight the importance of networks on HIV and HCV burden in a community of PWID in New Delhi , India . Integrating strategies to intervene at sociometric- and spatial-levels in addition to individual-level interventions could improve the efficiency of prevention and treatment programming and may be critical to achieving epidemic control and elimination of HIV and HCV , respectively , while conserving resources . | Understanding the social and spatial relationships that connect people is a key element to stop the spread of infectious diseases . These networks are particularly relevant to combat epidemics among populations that are hard to reach with public health interventions . Network-based approaches , for example , can help to stop HIV or hepatitis C from spreading amongst populations that use injectable drugs . Yet how social and geographic connections such as acquaintances , injection partners , or preferred drug use places impact the risk of infection is still poorly mapped out . To address this question , Clipman et al . focused on people who inject drugs in New Delhi , India , a population heavily impacted by HIV and hepatitis C . Over 2500 people were recruited , each participant inviting their injection partners to also take part . The volunteers answered survey questions , including where they used drugs , and provided a blood sample to be tested . The results showed that , even after adjusting for individual risk factors , where people used drugs and with whom affected their risk of becoming infected with HIV and hepatitis C . In terms of social ties , the likelihood of HIV and hepatitis C infection decreased by about 13% for each person separating a given individual from an infected person . However , geographical networks also had a major impact . Injecting at a popular location respectively increased the odds of HIV and hepatitis C infection by 50% and 69% . In fact , even if the participant was not using drugs at these specific places , having an injection partner who did was enough to increase the risk for disease: for each person separating an individual from the location , the likelihood of being infected with HIV and hepatitis C decreased by respectively 14% and 10% . The results by Clipman et al . highlight how the relationships between physical spaces and social networks contribute to the spread of dangerous diseases amongst people who inject drugs . Ultimately , this knowledge may help to shape better public health interventions that would take into account the importance of geographical locations . | [
"Abstract",
"Introduction",
"Materials",
"and",
"methods",
"Results",
"Discussion"
] | [
"epidemiology",
"and",
"global",
"health",
"medicine"
] | 2021 | Role of direct and indirect social and spatial ties in the diffusion of HIV and HCV among people who inject drugs: a cross-sectional community-based network analysis in New Delhi, India |
Photoreceptors are the most numerous and metabolically demanding cells in the retina . Their primary nutrient source is the choriocapillaris , and both the choriocapillaris and photoreceptors require trophic and functional support from retinal pigment epithelium ( RPE ) cells . Defects in RPE , photoreceptors , and the choriocapillaris are characteristic of age-related macular degeneration ( AMD ) , a common vision-threatening disease . RPE dysfunction or death is a primary event in AMD , but the combination ( s ) of cellular stresses that affect the function and survival of RPE are incompletely understood . Here , using mouse models in which hypoxia can be genetically triggered in RPE , we show that hypoxia-induced metabolic stress alone leads to photoreceptor atrophy . Glucose and lipid metabolism are radically altered in hypoxic RPE cells; these changes impact nutrient availability for the sensory retina and promote progressive photoreceptor degeneration . Understanding the molecular pathways that control these responses may provide important clues about AMD pathogenesis and inform future therapies .
Uninterrupted blood flow and an intricate and architecturally optimized network of photoreceptors , interneurons , glia , and epithelial cells are required for vision . The primary blood supply for photoreceptors is the choriocapillaris , an extraretinal fenestrated capillary bed . A layer of extracellular matrix proteins , Bruch’s membrane , lies adjacent to the choriocapillaris , and a monolayer of retinal pigment epithelium ( RPE ) cells divides Bruch’s membrane from the photoreceptors . The choriocapillaris , Bruch’s membrane , RPE , and photoreceptors function as one unit , with the choriocapillaris providing fuel for phototransduction , and Bruch’s membrane and RPE cells filtering and regulating the reciprocal exchange of oxygen , nutrients , biomolecules , and metabolic waste products between the circulation and retina . RPE also provide critical support for both photoreceptors and the choriocapillaris ( Strauss , 2005 ) in large part by generating vascular endothelial growth factor ( VEGF ) , a cytokine required for choriocapillaris development and maintenance ( Kurihara et al . , 2012; Le et al . , 2010; Marneros et al . , 2005; Saint-Geniez et al . , 2009 ) . Defects in this unit , including reduced choriocapillaris density , the presence of sub-RPE lipid-rich deposits , and RPE/photoreceptor dysfunction , are characteristic of age-related macular degeneration ( AMD ) , a common vision-threatening disease whose prevalence is steadily increasing globally ( Friedman et al . , 2004; Wong et al . , 2014 ) . Several genetic and lifestyle risk factors have been identified but no cure exists ( Bird , 2010 ) . While current evidence suggests that AMD is a spectrum of closely related multifactorial polygenic diseases ( Bird et al . , 2014 ) , 10 year clinic-based data from the Age-Related Macular Degeneration Study ( AREDS; n = 4757 ) showed that the major risk factors include aging , severity of drusen ( sub-RPE deposits ) , and RPE abnormalities ( Chew et al . , 2014 ) . Early AMD is characterized by pigmentary changes and appearance of drusen ( Gass , 1973; Pauleikhoff et al . , 1990; Sarks , 1976; Wang et al . , 2010 ) . In most cases early AMD proceeds towards geographic atrophy ( or 'dry' AMD ) , a condition defined by focal photoreceptor , RPE , and choriocapillaris loss and thickening of Bruch’s membrane with immunomodulatory proteins and lipids ( Bird , 2010; Jager et al . , 2008; Zarbin , 1998 ) . While the primary defect could occur in Bruch’s membrane ( Pauleikhoff et al . , 1990; Bressler et al . , 1990; Mullins et al . , 2011 ) , the choriocapillaris ( Ramrattan et al . , 1994; Spraul et al . , 1996; Spraul et al . , 1999 ) , or photoreceptors ( Sarks , 1976; Hogan , 1972; Sarks et al . , 1988 ) , most evidence suggests that it probably occurs in RPE cells . Granules enriched with lipid-rich residues , lipofuscin , accumulate normally in aging RPE cells ( Bazan et al . , 1990 ) , but abnormal accretions are observed in patients with geographic atrophy in a band directly surrounding the lesion ( Feeney-Burns et al . , 1984; Holz et al . , 1999; von Ruckmann et al . , 1997 ) . Lipofuscin renders the RPE more sensitive to blue light-induced damage ( Rozanowska et al . , 1995; Schutt et al . , 2000; Sparrow et al . , 2000 ) , impairs RPE functions ( Holz et al . , 1999; Finnemann et al . , 2002; Lakkaraju et al . , 2007; Sparrow et al . , 1999 ) , and is potentially toxic for RPE cells ( Schutt et al . , 2000 ) . The downstream effects of RPE loss are catastrophic , and result in choriocapillaris attenuation and photoreceptor degeneration in late stage AMD patients ( Bhutto and Lutty , 2012; Coscas et al . , 2014; Jonas et al . , 2014; Lee et al . , 2013; Sohrab et al . , 2012; McLeod et al . , 2009 ) . Assuming , therefore , that the critical event of AMD pathogenesis occurs in RPE cells , how can the onset of the other early clinical manifestations of the disease in neighboring cells and structures be explained ? There is a growing body of evidence that choroidal and retinal blood flow is reduced in AMD ( Boltz et al . , 2010; Remsch et al . , 2000 ) . We hypothesize that hypoxia , a natural consequence of aging microenvironments ( that is exacerbated by obesity and smoking ) ( Blasiak et al . , 2014; Chiu and Taylor , 2011; Morgado et al . , 1994; Sagone et al . , 1973 ) , in RPE cells may be a central AMD risk factor based on the following lines of evidence: ( a ) RPE provide critical vasculotrophic support required for photoreceptor function ( Kurihara et al . , 2012 ) ; ( b ) Hypoxia alters lipid handling in other cell-types ( Glunde et al . , 2008; Santos et al . , 2012; Semenza , 2009 ) ; ( c ) at least 40% of lipids in drusen are secreted by RPE ( Cao et al . , 2013 ) ; ( d ) the RPE secretome is sensitive to stress ( Wang et al . , 2010 ) ; and ( e ) in drusen rich zones of AMD patient eyes , vascular density is significantly reduced ( Mullins et al . , 2011 ) . Therefore , hypoxia-mediated changes to the RPE lipidome and secretome could enhance lipofuscin accumulation and induce Bruch’s Membrane lipidization and thickening , thereby exacerbating RPE dysfunction , choriocapillaris drop-out , and photoreceptor dysfunction . This vicious cycle of events could accelerate progression of AMD . Hypoxia-inducible factor alpha subunits ( HIF-αs ) are the key transcription factors that mediate responses to hypoxia . Under normal conditions HIF-αs are constitutively expressed and targeted by von Hippel-Lindau protein ( VHL ) for ubiquitination and proteasomal degradation . VHL is inactivated at low oxygen tensions; this allows HIF-αs to translocate to the nucleus and activate a host of angiogenesis , glucose metabolism , erythropoiesis , and inflammation genes ( Semenza , 2011 ) . In this study we directly or indirectly hyperactivated HIF-αs in RPE by genetically perturbing components of the VHL/HIF/VEGF pathway using inducible and conditional gene ablation techniques . These manipulations altered lipid handling and glucose consumption of RPE cells , induced gross morphometric changes in RPE , reduced nutrient availability for the sensory retina , and promoted progressive photoreceptor atrophy . Understanding the effects of hypoxia on RPE metabolism , and learning how to control these effects , may provide insights for developing novel therapeutic strategies to treat retinal degenerative diseases .
Based on the hypothesis that choriocapillaris attenuation and prolonged hypoxia in RPE cells induces photoreceptor death/dysfunction , we set out to catalog the temporal and spatial manifestations of hypoxia in retinas of mice with severe choriocapillaris deficits ( VMD2-Cre;Vegfafl/fl ) and correlate these manifestations with any corresponding anatomical and functional changes in photoreceptors . Transgenic mice harboring human vitelliform macular dystrophy-2 promoter-directed cre ( VMD2-Cre ) ( Le et al . , 2008 ) were used to ablate Vegfa; severe choriocapillaris vasoconstriction is observed in adult Vegfa-cKO mice three days post induction ( dpi ) ( Kurihara et al . , 2012 ) . The first signs of RPE hypoxia , including nuclear HIF immunoreactivity ( Figure 1A ) , accumulation of the hypoxic probe pimonidazole in RPE ( Figure 1A&B; white arrows ) , and activation of a known panel of hypoxia-inducible target genes , were observed six months post induction ( mpi ) in the Vegfa mutants ( Figure 1—figure supplement 1&2 ) . However , we cannot exclude the possibility that low-grade hypoxia may occur in RPE or other retinal cells earlier at subthreshold levels of detection . Hypoxia in RPE induced several defects including severely distended basal infoldings , accumulation of lipid droplets within RPE cells ( Figure 1C; yellow arrows ) , RPE cell hypertrophy ( Figure 1D ) , and dramatic and progressive thickening of Bruch’s membrane beginning at nine months post induction ( Figure 1C red arrows; Figure 1E pseudo-colored blue ) . At 11 months post induction we also detected pigmentary abnormalities in fundus images ( Figure 2A ) and thinning of the photoreceptor cell layer ( Figure 2B; red line ) characteristic of photoreceptor degeneration . While RPE defects took months to manifest , defects in cone-driven pathways occurred within seven days of Vegfa ablation ( Kurihara et al . , 2012 ) and do not recover by 11 months post induction ( Figure 2C and D , photopic ) . Surprisingly , rod-driven pathway defects were not observed until 11 months post ablation ( Figure 2C and D , scotopic ) , suggesting that , for reasons that are unclear , rod photoreceptors are less sensitive to oxygen and nutrient deprivation than cones are . 10 . 7554/eLife . 14319 . 003Figure 1 . HIF-α accumulation precedes the induction of AMD-like features in Vegfa-cKO mice . ( A ) HIF-1α , HIF-2α , and pimonidazole ( Pim; green ) are detected 6 months post induction in flat-mounted RPE/choroids from Vegfa mutants but not in littermate controls . ZO-1 ( red ) labels the cell boundaries . ( B ) The hypoxia probe Pimonidazole ( Pim; green ) is detected specifically in the RPE ( white arrows ) of cross-sectioned Vegfa mutant retinas ( probe labeling is shown alone ( left ) and overlaid over brightfield images ( right ) to emphasize the retinal anatomy ) . ( C ) Electron micrographs of littermate control ( left ) and Vegfa-cKO RPE 11 months post induction ( right ) . Dashed squares in left panels are magnified in right panels . Note the absence of choriocapillaris in Vegfa mutants , accumulation of lipid droplets ( yellow arrows ) in the cytoplasm , and thickening of Bruch’s membrane ( red arrows ) . ( D ) Measured thickness values of the RPE of Vegfa mutant mice 11 months post induction ( n=4 ) ( see associated Figure 1—source data 1 ) . ( E ) Electron micrographs of RPE/Bruch’s membrane from control and Vegfa-cKO RPE taken 9 , 11 , and 18 months post induction . Note the progressive accumulation of material in Bruch’s membrane ( pseudo-colored light blue ) . Abbreviations: Pim=pimonidazole , INL=inner nuclear layer , ONL=outer nuclear layer ( photoreceptor cell bodies ) , OS=photoreceptor outer segments , BI=basal infoldings of RPE cells , BM=Bruch’s membrane , CC=choriocapillaris , mpi=months post [RPE-specific] induction . Scale bars=20 µm ( A ) , 50 µm ( B ) , 1 µm C&E . DOI: http://dx . doi . org/10 . 7554/eLife . 14319 . 00310 . 7554/eLife . 14319 . 004Figure 1—source data 1 . Source data for Figure 1D . DOI: http://dx . doi . org/10 . 7554/eLife . 14319 . 00410 . 7554/eLife . 14319 . 005Figure 1—figure supplement 1 . Recombination efficiency in VMD2-Cre mice . ( A ) RPE/choroid complexes from VMD2-Cre;ROSA26 mTomato/mGFP reporter mice were flatmounted 3 days post induction . ( B ) GFP expression , which occurs after cre-recombination , were counted and plotted . Error bars indicate mean s . e . m . ( n=3 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 14319 . 00510 . 7554/eLife . 14319 . 006Figure 1—figure supplement 2 . Upregulation of HIF target genes in Vegfa-cKO RPE . Relative mRNA expression values generated using qPCR gene-profiling analyses of 84 hypoxia-related genes in Vegfa-cKO RPE/Choroids 6 months post-induction . The upregulated genes ( P<0 . 05 or more than 1 . 5 fold-change ) are listed according to fold change ( highest to lowest ) ( n=4 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 14319 . 00610 . 7554/eLife . 14319 . 007Figure 2 . Photoreceptor atrophy and dysfunction is observed in late stage Vegfa-cKOs . ( A ) Color fundus images of littermate control and Vegfa-cKO mice 7 or 11 months post induction ( mpi ) . No obvious changes are seen after 7 months , but significant changes indicative of retinal degeneration are seen 11 months post induction . ( B ) Photoreceptor atrophy , determined by observing thinning of the outer nuclear layer ( ONL; red vertical line ) is seen in DAPI labeled cryosectioned Vegfa-cKO retinas 18 months post induction 600 μm from the optic nerve head compared with controls . ( C ) Full-field ERGs performed on controls and Vegfa-cKO mice 11 months post induction ( n=6 ) reveal rod dysfunction in both the a- and b-waves ( scotopic ) , and cone dysfunction in the b-wave and flicker response ( photopic flash & flicker ) in Vegfa-cKO mice . ( D ) Quantification of ERGs ( see associated Figure 2—source data 1 ) . *p<0 . 05 , **p<0 . 01 . Error bars indicate mean plus s . d . Scale bar=20 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 14319 . 00710 . 7554/eLife . 14319 . 008Figure 2—source data 1 . Source data for Figure 2D . DOI: http://dx . doi . org/10 . 7554/eLife . 14319 . 008 Theoretically , RPE should be able to respond to hypoxia and improve circulation in the choriocapillaris by increasing basal VEGF secretion . To induce hypoxia we maintained 3D cultures of primary human RPE in 3% oxygen in a controlled chamber and analyzed the molecular and metabolic changes . The HIF target gene Vegfa was upregulated after exposure to low oxygen ( Figure 3A ) or upon addition of dimethyloxalylglycine ( DMOG ) , an inhibitor of prolyl hydroxylase that leads to HIF stabilization ( Figure 3B ) . DMOG also significantly stimulated basal VEGF secretion in a dose dependent manner ( Figure 3C ) . To determine the physiological relevance of hypoxia-enhanced RPE-derived VEGF synthesis , we used VMD2-Cre to delete Vhl in RPE in vivo . Signs of [pseudo] hypoxia , i . e . HIF-α immunoreactivity and activation of known hypoxia-induced genes ( including Vegfa ) , were observed three days post induction in Vhl-cKO mice ( Figure 3D and Figure 3—figure supplement 1 ) . The increased VEGF production from Vhl-cKO RPE correlates with significant and progressive choriocapillaris vasodilation based on ultrastructural examinations ( Figure 3E and F ) . However , no vasodilation was observed in double ( Vhl/Hif1a-dKO or Vhl/Hif2a-dKO ) or triple ( Vhl/Hif1a/Hif2a-tKO ) knockout mice ( Figure 3G ) indicating that both RPE-derived HIF-1α and HIF-2α are mutually responsible for HIF/VEGF induced vasodilation . Collectively , these findings demonstrate that RPE can sense changes in oxygen/nutrient availability and respond by appropriately altering the vascular tone of the choriocapillaris . 10 . 7554/eLife . 14319 . 009Figure 3 . HIF-αs induce dilation of the choriocapillaris in Vhl-cKOs . ( A ) Vegfa is upregulated in hRPE exposed to 3% oxygen . Data are the mean plus s . e . m . ( n=5–6 ) . ( B–C ) Vegfa mRNA ( B ) and VEGF165 protein ( C—predominately from the basal surface when grown on transwells ) is upregulated in a dose-dependent manner by DMOG for 24 hr compared with DMSO controls . Data are the mean plus s . e . m . ( n=4–6 ) . ( D ) Immunohistochemistry analyses reveal that VHL ( green ) expression is lost 3 days post induction in Vhl-cKOs , and HIF-1α and HIF-2α are upregulated in the nucleus of Vhl-cKOs RPE , but not in controls . ( E ) Measurements of the choriocapillaris from electron micrographs 0–28 days post induction ( F ) in untreated and Vhl mutants revealed progressive choriocapillaris vasodilation ( n=4 ) . ( G ) Choriocapillaris thickness values of Vhl-cKO , Vhl/Hif1a-dKO , Vhl/Hif2a-dKO , Vhl/Hif1a/Hif2a-tKO mice measured 28 days post induction ( n=4 ) . ( See also associated Figure 3—source data 1 for panels A-C , E , and G . ) Scale bars=20 µm . Error bars represent mean plus s . d . DOI: http://dx . doi . org/10 . 7554/eLife . 14319 . 00910 . 7554/eLife . 14319 . 010Figure 3—source data 1 . Source data for Figure 3A–C , E , and G . DOI: http://dx . doi . org/10 . 7554/eLife . 14319 . 01010 . 7554/eLife . 14319 . 011Figure 3—figure supplement 1 . Upregulated hypoxia-related genes in Vhl-cKO RPE/Choroid . Relative mRNA expression values obtained using qPCR gene-profiling analyses of 84 hypoxia-related genes in Vhl-cKO RPE/Choroids 3 days post induction . The dysregulated genes ( p<0 . 05 or more than 1 . 5 fold-change ) are listed according to fold change values ( highest to lowest ) ( n=4 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 14319 . 011 Hypoxia-triggered choriocapillaris vasodilation may come at a significant cost for the retina . The accumulation of lipid droplets and material resembling glycogen is detectable within only three days post induction in the RPE of Vhl mutants ( Figure 4A ) . Severely distended basal infoldings , thickening of Bruch’s membrane , and numerous lipid droplets ( sometimes contiguous with subretinal extracellular spaces ) are observed 14 days post Vhl deletion ( Figure 4B; red ) . Measurements across the RPE from electron micrographs reveal significant hypertrophy ( Figure 4C ) . While the double deletion of Vhl/Hif1a did not prevent the RPE defects in the Vhl mutants , the RPE cells of Vhl/Hif2a or Vhl/Hif1a/Hif2a mutants appeared unremarkable and were not hypertrophic ( Figure 4D and E ) , suggesting that HIF-2α , as it is in other cell-types ( Qiu et al . , 2015; Zhao et al . , 2015 ) , is the pathological HIF isoform in hypoxic RPE . 10 . 7554/eLife . 14319 . 012Figure 4 . Dramatic and rapid-ensuing RPE defects observed in Vhl-cKO mice are dependent on Hif2a . ( A ) Electron micrographs of RPE cells from littermate control and Vhl-cKO mice 3 days post Vhl deletion . Regions marked with perforated white rectangles are in the lower panels . Note the intracellular accumulations of lipid droplets ( a; red ) and glycogen ( b ) . ( B ) Electron micrographs of RPE cells from littermate control and Vhl-cKO mice 14 days post Vhl deletion . Intracellular lipid droplets ( a ) , extrusion of lipid droplets into the subretinal space ( b ) , and lipids collecting along the basal laminar surface of Bruch’s membrane and between RPE basal infoldings ( c ) are observed . ( C ) Thickness measurements from electron micrographs and reveal that RPE hypertrophy occurs from 0–3 day post induction timepoints , and then plateaus from 3–28 timepoints in Vhl mutant mice ( n=5 ) . ( D ) Electron micrographs of RPE from Vhl/Hif1a ( left panels ) , Vhl/Hif2a ( upper middle panel ) , and Vhl/Hif1a/Hif2a ( bottom middle panel ) mutant mice 14 days post induction . Note that lipid droplets ( dark gray spheres , upper left panel ) and material resembling glycogen ( small punctate spots ) are observed in Vhl/Hif1a-dKO , but not in Vhl/Hif2a-dKO or Vhl/Hif1a/Hif2a-tKO RPE ) 14 days post induction . These data suggest Hif2a is responsible for the phenotype in Vhl mice . ( E ) Choriocapillaris thickness values of Vhl-cKO , Vhl/Hif1a , Vhl/Hif2a , Vhl/Hif1a/Hif2a mice measured 28 days post induction ( n=4 ) . ( See also associated Figure 4—source data 1 for panels C&E . ) Scale bars=5 µm ( A ) , 1 µm ( A’a & A’b ) , 2 µm ( B ) , 0 . 5 µm ( B’a , B’b , B’c ) , 5 µm ( D ) . Error bars represent mean plus s . d . DOI: http://dx . doi . org/10 . 7554/eLife . 14319 . 01210 . 7554/eLife . 14319 . 013Figure 4—source data 1 . Source data for Figure 4C&E . DOI: http://dx . doi . org/10 . 7554/eLife . 14319 . 013 Devastating secondary effects resulting from RPE hypoxia were observed in the sensory retina . Photoreceptor degeneration ( Figure 5A–C ) and significant functional impairments of both rod and cone driven-pathways ( Figure 5D—figure supplement 1A ) were observed in VMD-Cre;Vhl and VMD-Cre;Vhl/Hif1a mice 28 days post induction but not in in Vhl/Hif2a ( or Vhl/Hif1a/Hif2a mice; Figure 5C , E–G ) , ( or in any of the relevant controls; Figure 5—figure supplement 1B and C ) . In advanced stages of the phenotype ( >50 dpi ) , dramatic changes in RPE and the vasculature are observed consistent with retinal remodeling ( Figure 5—figure supplement 2 ) ( Marc et al . , 2003 ) . These findings imply that HIF-2α-mediated metabolic stress in RPE , which cannot be rescued even with a significantly dilated choriocapillaris , is enough to promote photoreceptor degeneration . 10 . 7554/eLife . 14319 . 014Figure 5 . Progressive and rapid photoreceptor degeneration observed in Vhl-cKO mice is dependent on Hif2a . ( A ) Thickness measurements from electron micrographs reveal progressive thinning of the outer nuclear layer ( ONL or photoreceptor cell bodies ) from 0–28 days post induction timepoints in Vhl mutant mice ( n=5 ) . ( B ) Cryosectioned DAPI stained retinas from Vhl-cKO mice prior to induction ( 0 dpi; left panel ) , 7 dpi , and 28 dpi . ( C ) Quantified thickness values measured 600 μm from the optic nerve head of the outer nuclear layer in Vhl-cKO , Vhl/Hif1a , Vhl/Hif2a , and Vhl/Hif1a/Hif2a 28 days post induction ( n=4 ) ( See associated Figure 5—source data 1 for panels A&C . ) . ( D ) Full-field ERGs performed on Vhl-cKO and control mice 28 days post induction . ( E ) ERGs from Vhl/Hif1a , ( F ) Vhl/Hif2a , and ( G ) Vhl/Hif1a/Hif2a mutant mice 28 days post induction . ERG analyses reveal that normal photoreceptor function is observed in Vhl/Hif2a ( F ) or Vhl/Hif1a/Hif2a ( G ) mutant mice . *p<0 . 05 , **p<0 . 01 . For all ERGs n=6–8 . Error bars indicate mean plus s . d . DOI: http://dx . doi . org/10 . 7554/eLife . 14319 . 01410 . 7554/eLife . 14319 . 015Figure 5—source data 1 . Source data for Figure 5A&C and Figure 5—figure supplement 1A–C . DOI: http://dx . doi . org/10 . 7554/eLife . 14319 . 01510 . 7554/eLife . 14319 . 016Figure 5—figure supplement 1 . Quantification of ERGs for Vhl cKO and lack of photoreceptor degeneration in relevant controls . ( A ) Quantification of ERGs from Figure 5D ( n=6–8 ) . ( B ) Thickness measurements from electron micrographs reveal no differences in ONL thickness values from multiple controls including VMD2-Cre;mGFP + Doxycycline , VMD2-Cre;Vhl no Doxycycline , VMD2-Cre;Vhl/Hif1a; no Doxycycline , VMD2-Cre;Vhl/Hif2a; no Doxycycline , and VMD2-Cre;Vhl/Hif1a/Hif2a; no Doxycycline 28 days post induction . VMD2-Cre/mGFP mice received doxycycline 28 days prior to examination ( n=4 ) . ( C ) ERGs measured in the same controls 28 days post injection as ( B ) ( n=6–8 ) . ( See associated Figure 5—source data 1 ) *p=0 . 05 , **p=0 . 01 . DOI: http://dx . doi . org/10 . 7554/eLife . 14319 . 01610 . 7554/eLife . 14319 . 017Figure 5—figure supplement 2 . RPE and choroid defects characteristic of retinal remodeling are observed in late stage Vhl mutants . ( A ) ( Left panel ) The basal infoldings of RPE are severely distended and cells are misshapen 50 days post induction . ( Right panel ) Pale nuclei are observed at 75 days post induction . Half of each micrograph is pseudo-colored blue to mark Bruch’s membrane . ( B ) Neovascularization is observed at very late stages in Vhl mutants ( 15 months post induction ) . Collagen IV labels blood vessels that are interdigitating with RPE cells ( white arrows ) . DOI: http://dx . doi . org/10 . 7554/eLife . 14319 . 017 We next set out to identify the molecular mechanisms driving the hypoxic-mediated metabolic stress in RPE cells . Based on histopathological evidence of impaired lipid handling in hypoxic RPE we performed gene profiling for fatty acid metabolism genes from RPE/choroid complexes of Vhl-cKO mice . Acyl-CoA synthetase and Acyl-CoA dehydrogenase family genes were downregulated ( Figure 6A and Figure 6—figure supplement 1A and B ) but not in Vhl/Hif2a or Vhl/Hif1a/Hif2a mutant RPE ( Figure 6B and Figure 6—figure supplement 1B ) . We also performed untargeted high-resolution mass spectrometry-based metabolomic analyses and observed abnormal levels of several long-chain saturated , unsaturated , and oxidized acylcarnitines in the Vhl-cKO mice ( Figure 6C and Figure 6—figure supplement 2 ) . Vhl/Hif1a mice exhibit dysregulated metabolomic profiles similar to Vhl mutants ( Figure 6—figure supplement 2A and B ) , but normal levels of acylcarnitines and other metabolites were observed in Vhl/Hif2a ( Figure 6—figure supplement 2C ) and Vhl/Hif1a/Hif2a mice ( Figure 6—figure supplement 2D ) . Collectively , these data strongly suggest that HIF-2α regulates lipid handling in RPE in vivo . 10 . 7554/eLife . 14319 . 018Figure 6 . Defects in lipid metabolism in Vhl mutant RPE . ( A ) Summary of gene-profiling experiments for lipid metabolism genes in RPE/choroid complexes from Vhl mutant mice 3 days post induction ( n=4 ) . ( B ) Downregulation of lipid metabolism genes was also seen in Vhl/Hif1a mutants , but nominally in Vhl/Hif2a , and no gross changes were seen in Vhl/Hif1a/Hif2a mutants 3 days post induction ( n=4 ) . ( C ) Untargeted high-resolution mass spectrometry-based metabolomic analyses revealed that a group of acylcarnitines ( AC ) was progressively dysregulated from 3 to 14 days post induction ( n=4–6 ) ( see also associated Figure 6—source data 1 ) . Box and whiskers plots are shown . Error bars represent maximum and minimum values . ( D ) Pre-treating hRPE with DMOG reduced the basal oxygen consumption rates ( initial OCR – OCR after injection of RAA ) when the cells were assayed in substrate limited media ( 2 . 5 mM glucose ) and provided BSA control or palmitate conjugated to BSA ( n=4 ) ( see also associated Figure 6—figure supplement 3B ) . Error bars are the maximum and minimum values in panel C and mean plus s . d . in panel D . DOI: http://dx . doi . org/10 . 7554/eLife . 14319 . 01810 . 7554/eLife . 14319 . 019Figure 6—source data 1 . Source data for Figure 6C , Figure 6—figure supplement 2A–D , Figure 6—figure supplement 3B . DOI: http://dx . doi . org/10 . 7554/eLife . 14319 . 01910 . 7554/eLife . 14319 . 020Figure 6—figure supplement 1 . Classification of lipid metabolism genes and gene-profiling in Vhl and combinatorial Vhl/Hif mutants . ( A ) Classifications of the genes from the Mouse Fatty Acid Metabolism PCR array ( PAMM-007; Qiagen ) . ( B ) Profiling of fatty acid metabolism genes in Vhl-cKO and Vhl-cKO , Vhl/Hif1a-dKO , Vhl/Hif2a-dKO , Vhl/Hif1a/Hif2a-tKO RPE/choroids 3 days post induction ( n=4 ) . Note that the significant dysregulations in Vhl-cKO mice are predominately rescued in Vhl/Hif2a-dKO and Vhl/Hif1a/Hif2a-tKO mice 3 days post induction ( n=4 ) . Vhl/Hif1a-dKO induces only partial rescue . Boxes are colored gray , blue , or yellow for fold change values with p<0 . 05 , p<0 . 01 , p<0 . 001 significance respectively . DOI: http://dx . doi . org/10 . 7554/eLife . 14319 . 02010 . 7554/eLife . 14319 . 021Figure 6—figure supplement 2 . MS-based metabolomic assays for Vhl-cKO , Vhl/Hif1a-dKO , Vhl/Hif2a-dKO , Vhl/Hif1a/Hif2a-tKO mice . ( A-D ) Untargeted high-resolution mass spectrometry-based metabolomic analyses revealed that multiple unique small molecule metabolites are dysregulated in Vhl-cKO and Vhl/Hif1a-dKO , mice but not in Vhl/Hif2a and Vhl/Hif1a/Hif2a-tKO mice 3 days post induction ( n=6 ) . ( See also associated Figure 6—source data 1 . ) The majority of the identifiable metabolites consisted largely of acylcarnitines . Error bars indicate maximum and minimum values . DOI: http://dx . doi . org/10 . 7554/eLife . 14319 . 02110 . 7554/eLife . 14319 . 022Figure 6—figure supplement 3 . DMOG inhibits lipid oxidation in RPE cells . ( A ) Basal oxygen consumption rates ( OCR ) of hRPE treated with BSA-Palmitate conjugates or with BSA alone after treatment with 40 μM Etomoxir , an inhibitor of CPT-1 , to validate the increased of OCR are due to fatty acid oxidation ( FAO ) . Data are the mean plus s . e . m . ( n=8 ) . ( B ) Seahorse Flux Analysis full OCR trace showing hRPE cells responding to pre-treatment with DMOG ( 10 μM ) for 24 hr in the presence and absence of BSA , palmitate and etomoxir ( see also associated Figure 6—source data 1 ) . Data are the mean plus s . d . ( n=6 ) DOI: http://dx . doi . org/10 . 7554/eLife . 14319 . 022 To test if HIF activation leads to altered lipid oxidation we monitored oxygen consumption in human RPE treated with a hypoxia mimetic , DMOG . Seahorse flux analysis revealed that human RPE cells in substrate-limited media oxidize exogenous lipids ( palmitate ) as an energy substrate ( BSA control vs . palmitate control , Figure 6D ) . The ability of palmitate to increase oxygen consumption rates ( OCRs ) was validated by the addition of an inhibitor of the carnitine transport , etomoxir , which reduced oxygen consumption to BSA control levels ( Figure 6—figure supplement 3A ) . Interestingly , treating the cells with low-dose DMOG for 48 hr prevented lipid oxidation ( palmitate control vs . palmitate DMOG , Figure 6D and Figure 6—figure supplement 3B ) . These data suggest that hypoxic RPE alter their lipid handling behaviors and begin storing lipids in droplets , rather than utilizing them as an energy substrate . The presence of visible glycogen stores in Vhl-cKO RPE indicates that glucose metabolism may also be dysregulated . Using PCR arrays for glucose metabolism we determined that glycogen degradation genes were downregulated in Vhl mutant mice ( summarized in Figure 7A ) . Furthermore , we observed that glycolysis-related genes were largely upregulated and TCA cycle genes were largely downregulated ( summarized in Figure 7B and Figure 7—figure supplement 1A and B ) . These data suggest that in vivo the RPE cells are reducing oxidative phosphorylation and meeting their energy demands by increasing glycolysis . To test if RPE cells reduce oxidative metabolism in response to hypoxia , we monitored oxygen consumption rates in cultured human RPE cells ( provided glucose , pyruvate and glutamine ) after the cells were exposed to hypoxic conditions ( 3% O2 ) for 72 hr compared to a parallel control plate maintained in normoxia . Basal and maximal oxygen consumption rates were greatly reduced in cells after hypoxia exposure ( Figure 7C and Figure 7—figure supplement 2A ) , suggesting mitochondrial respiration has been remodeled as a result of hypoxia . An internal ratio of the basal OCR divided by the proton leak ( oligomycin-insensitive OCR ) to normalize for potential plate-to-plate variation , showed a similar effect of hypoxia reducing oxidative capacity ( Figure 7—figure supplement 2B ) . Treating the human RPE cells with low-dose DMOG , a HIF activator , also significantly reduced basal and maximal oxidative capacities in a dose-dependent manner ( Figure 7D ) . In cells treated with higher doses of DMOG ( >250 uM ) oxidative capacity was completely lost ( Figure 7D ) , even though there were no outward signs of toxicity or cell death . Collectively these data suggest that the oxidative metabolism of RPE cells is very sensitive to hypoxia . 10 . 7554/eLife . 14319 . 023Figure 7 . Glucose consumption and metabolism is altered in the RPE of Vhl mutants . ( A ) Gene-profiling data revealed that glycogen degradation genes were significantly attenuated in Vhl mutants 3 days post induction ( n=4 ) . ( B ) Gene-profiling data for glucose metabolism genes were summarized by plotting changes along the glycolysis and TCA cycle pathways 3 days post induction ( n=4 ) ( red text=downregulated , blue text=upregulated ) . ( C ) Basal and maximal oxygen consumption rates ( OCR ) of hRPE after being exposed to 3% O2 for 72 hr or maintained at normoxia . Data are the mean plus s . e . m . ( n=7–10 ) . ( D ) Seahorse Flux Analysis OCR trace showing reduced OCR in hRPE cells treated with DMOG for 24 hr . Data points are the mean plus s . d . ( n=6 ) . ( E and F ) Changes in lactate ( E ) and glucose ( F ) levels of the media from hRPE cells , in transwells , after treatment with DMOG for 24 hr . Data are the mean plus s . e . m . ( n=4 ) . ( G ) Glucose levels are decreased in the sensory retina of Vhl-cKO mice 3 days post induction compared with littermate controls ( n=6–10 ) . ( H ) Relative glucose consumption is increased ( roughly two-fold ) in primary Vhl-cKO RPE compared with controls ( n=6 ) . ( I and J ) Relative expression level of Glut1 in hRPE cells after exposure to 3% 02 for 72 hr ( I ) or treatment with DMOG for 24 hr ( J ) . Data are the mean plus s . e . m . ( n=6 ) . ( K ) Changes in glutamine levels of the media from hRPE cells , in transwells , after treatment with DMOG for 24 hr . ( See also associated Figure 7—source data 1 for panels C-K . ) Data are the mean plus s . d . ( n=4 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 14319 . 02310 . 7554/eLife . 14319 . 024Figure 7—source data 1 . Source data for Figure 7C–K , Figure 7—figure supplement 2A . DOI: http://dx . doi . org/10 . 7554/eLife . 14319 . 02410 . 7554/eLife . 14319 . 025Figure 7—figure supplement 1 . Altered regulation of glycolytic and TCA genes in Vhl and Hif mutants . ( A ) Glycolysis-related gene expression changes ( fold change ) in Vhl-cKO and Vhl/Hif-dKO combination mutants 3 days post induction ( n=4 ) . Boxes are colored gray , blue , or yellow for fold change values with p<0 . 05 , p<0 . 01 , p<0 . 001 significance respectively . ( B ) TCA cycle-related gene expression changes ( fold change ) in Vhl and Vhl/Hif combination mutants 3 days post induction ( n=4 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 14319 . 02510 . 7554/eLife . 14319 . 026Figure 7—figure supplement 2 . Glucose metabolism is altered in Vhl mutant RPE and hRPE exposed to hypoxia . ( A ) Seahorse Flux Analysis full OCR trace showing hRPE cells exposed to 3% O2 for 72 hr or maintained at hypoxia , two plates combined onto one graph . Data are the mean pluss . e . m . ( n=10 ) ( See also associated Figure 7—source data 1 ) . ( B ) Ratio of basal OCR to OCR after addition of oligomycin in hypoxia and normoxia hRPE cells . Data are the mean plus s . e . m . ( n=10 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 14319 . 02610 . 7554/eLife . 14319 . 027Figure 8 . Summarizing the onset of the major phenotypes in Vhl and Vegfa mutant mice . ( A ) In normal conditions RPE deliver sufficient levels of glucose for photoreceptor homeostasis . During real ( Vegfa-cKO ) and pseudo hypoxia ( Vhl-cKO ) , Bruch’s membrane thickening and metabolic changes in RPE ( that induce lipid accumulation—dark gray circles ) limit glucose delivery to photoreceptors . This results in photoreceptor atrophy . ( B ) The effects of pseudo hypoxia in Vhl mutants and real hypoxia in Vegfa mutants present in a similar sequence but at radically different rates . The progression rate is accelerated in Vhl-cKOs since HIFs are dominant-stable . Changes in choriocapillaris density induce graded hypoxia in the Vegfa-cKO line , making it more physiologically relevant . DOI: http://dx . doi . org/10 . 7554/eLife . 14319 . 027 We found using 3D culture methods that polarized human RPE treated with DMOG increased lactate production ( Figure 7E ) further supporting the notion of a metabolic shift from oxidative phosphorylation toward glycolysis in RPE . This was coupled with evidence of increased glucose uptake from their apical surfaces ( Figure 7F ) , thereby potentially depleting glucose supplies allocated for the photoreceptors . In vivo , metabolic profiling analyses revealed that glucose levels in the sensory retina of Vhl mutants were significantly reduced ( Figure 7G ) . Similarly , primary RPE cells isolated from Vhl mutant mice took up more than double the amount glucose as controls ( Figure 7H ) . To allow increased glucose uptake , we found the glucose transporter 1 ( Glut1 or Slc2a1 ) was upregulated in vivo ( Vhl-cKO RPE; Figure 3—figure supplement 1 ) and in vitro in response to both hypoxia and DMOG treatment ( human RPE Figure 7I and J ) . These changes in glucose processing are similar to a HIF-dependent phenomenon observed in tumor cells known as the 'Warburg effect' ( Luo et al . , 2011; Takubo et al . , 2010; Warburg , 1956 ) . Measuring metabolite levels from human RPE differentiated in transwells highlights how changes in RPE metabolism may influence nutrient availability for the retina . In addition to DMOG reducing glucose ( and increasing lactate ) levels in the apical chamber , we observed that glutamine handling was also altered in response to hypoxia . Glutamine is an essential cellular metabolite that is both a key energy substrate and nitrogen source . DMOG reduced glutamine availability in the apical chamber ( Figure 7K ) . These data suggest that a confluence of hypoxia-mediated metabolic changes in RPE results in depleted energy substrates for photoreceptors that promotes retinal degeneration .
Photoreceptors are some of the most metabolically demanding cells in the body; this study reinforces how heavily photoreceptors rely on RPE for metabolic support . While it is widely accepted that death or dysfunction or RPE results in photoreceptor degeneration , the combination ( s ) of cellular stresses that impair RPE function and survival are incompletely understood . We present evidence here that a single causative factor , chronic HIF-2α-mediated metabolic stress in RPE cells , is enough to induce photoreceptor dysfunction/degeneration . We also provided the following mechanistic explanations: ( a ) hypoxia alters the RPE secretome and vectorial release of VEGF , lipids , and several metabolites including lactate , glutamine , and glucose; ( b ) hypoxic RPE shut down glucose and lipid oxidation and commit to glycolysis ( and store superfluous glucose and lipid ) ; ( c ) this metabolic shift allows RPE to obtain energy more quickly , but requires that they double their glucose intake due to low ATP yield . Since RPE are the major suppliers of glucose to the neurosensory retina ( Foulds , 1990 ) , these changes directly impact photoreceptor function and survival ( Linton et al . , 2010; Chertov et al . , 2011 ) . Aberrant lipid accumulation occurs in both human RPE and Bruch’s Membrane during aging and disease ( Pauleikhoff et al . , 1990; Holz et al . , 1994; Sheraidah et al . , 1993 ) . In fact lipids were the first identified molecules in human subretinal deposits ( Pauleikhoff et al . , 1992; Wolter and Falls , 1962 ) . Unlike retinosomes , lipid droplets that arise due to visual cycling defects in RPE ( Imanishi et al . , 2004 ) , the lipid inclusions observed in hypoxic RPE are likely derived from metabolic derangements . Similar inclusions ( and RPE hypertrophy and photoreceptor atrophy ) were observed in transgenic mice by conditionally inactivating OXPHOS in RPE ( Best1-Cre;Tfamfl/fl ) ( Zhao et al . , 2011 ) . Lipid accumulation in Bruch’s membrane is a common feature of AMD , and several lipid-processing genes have been associated with modifiable risk for AMD ( Chen et al . , 2010; Klaver et al . , 1998; Neale et al . , 2010; Souied et al . , 1998; Zerbib et al . , 2009 ) . Neutral lipids , likely derived from RPE , are abundant in AMD specific basal linear deposits ( Sarks et al . , 1988; Curcio and Millican , 1999; Rudolf et al . , 2008; Sarks et al . , 2007; Zweifel et al . , 2010 ) . Finally , thickening of Bruch’s membrane contributes to AMD pathogenesis ( Sarks , 1976; Sarks et al . , 1988 ) , probably by greatly limiting diffusion of aqueous based metabolites to the retina ( including energy substrates such as glucose ) ( Pauleikhoff et al . , 1990; Moore , 1995; Starita et al . , 1995; 1996; 1997 ) , by inducing inflammation , and by promoting choroidal neovascularization ( Baba et al . , 2010; Tamai et al . , 2002 ) . The identification of a broad class of acyl-carnitines in hypoxic RPE provides additional evidence of gross glucose metabolism and lipid-handling defects; identifying the combinations of perturbed acylcarnitines is informative for diagnosing fatty acid oxidation defects and other inborn errors of metabolism ( Rinaldo et al . , 2008 ) . Glucose availability correlates with retinal degeneration in both animal models and AMD patients . A comprehensive analysis of murine models of spontaneous degeneration revealed nutritional imbalances leading to cone photoreceptor degeneration , which could be delayed with insulin therapy ( Punzo et al . , 2009 ) . In addition RPE-specific inactivation of oxidative phosphorylation , which is induced by conditional deletion of mitochondrial transcription factor A , leads to increased glycolysis and photoreceptor degeneration . These phenotypes were prevented by inhibition of mTOR ( mammalian target of rapamycin ) ( Zhao et al . , 2011 ) , a positive regulator of HIF-αs ( Hudson et al . , 2002; Nayak et al . , 2012 ) . Furthermore , excessive glycemic load is a modifiable risk for AMD; and 20% of advanced AMD cases might be prevented by consuming less glucose-rich foods ( Chiu et al . , 2007 ) . Hypoxia in RPE can occur during aging and disease due to localized choriocapillaris dropout . Based on this study , morphometric changes to the choriocapillaris , which can be routinely and non-invasively examined ( Zhang et al . , 2015 ) , may be a powerful predictive tool for AMD disease progression . In addition , monitoring RPE/Bruch’s membrane changes and tracking the glycemic index of early AMD patients may improve case-specific treatment and risk progression prediction algorithms . Finally , abnormally high HIF-2α levels have been observed in the RPE of human aged donor eyes ( Sheridan et al . , 2009 ) . Therefore , anti-HIF therapies , which are being developed for treating other diseases ( Metelo et al . , 2015; Rini and Atkins , 2009; Rogers et al . , 2013; Scheuermann et al . , 2013 ) , may be effective for treating some AMD patients . In conclusion , our models may be employed to examine the progression of AMD-like features , as well as for developing novel preventive and therapeutic strategies for AMD and other vision-threatening diseases .
The TSRI Animal Care Committee approved all procedures involving animals and we adhered to all federal animal experimentation guidelines . Transgenic mice carrying the human vitelliform macular dystrophy-2 ( VMD2 ) promoter-directed reverse tetracycline-dependent transactivator ( rtTA ) and the tetracycline-responsive element ( TRE ) -directed Cre recombinase ( VMD2-Cre mice ) ( Le et al . , 2008 ) were mated with Vhlfl/fl mice ( Haase et al . , 2001 ) , Hif1afl/fl mice ( Ryan et al . , 2000 ) , Epas ( Hif2a ) fl/fl mice ( Gruber et al . , 2007 ) , or Vegfafl/fl mice ( Gerber et al . , 1999 ) . Control littermates harboring floxed alleles but no Cre-recombinase were utilized . Additional control experiments were also examined ( Figure 5—figure supplement 1B and C ) . To induce gene deletion , 80 μg/g body weight of doxycycline were injected daily into 6–8 week-old transgenic mice intraperitoneally for three days . Gene recombination was quantified and cre-mediated toxicity was examined in double transgenic VMD2-Cre with ROSA26 mTomato/mGFP reporter mice ( Muzumdar et al . , 2007 ) . We noted that recombination in VMD2-Cre occurs at either ~70% or ~10% , and either the high or low efficiency is inherited . All data shown in the main body of text was from the offspring of “high efficiency mice . ” Gene ablation in “low efficiency mice” yielded a similar , albeit much milder phenotype . Genotyping was performed at Transnetyx ( Memphis , TN ) . Mutations for rds ( retinal degeneration slow ) , rd1 , rd8 , and rd10 were examined and excluded from breeding pairs . RPE cells from VMD2-Cre;Vhlfl/fl and control Vhlfl/fl littermates were isolated as previously described ( Krohne et al . , 2012 ) . Cells were maintained at 37° and 5% CO2 in DMEM/F12 from Thermo Fisher Scientific ( Waltham , MA ) with 2% FBS from Jackson ImmunoResearch ( West Grove , PA ) . 2 μg/μl doxycycline was added to all cultures daily for three days to induce Vhl ablation . Human RPE cells ( Lonza ) were maintained either on plastic surfaces or on transwell filters ( Corning ) depending on the application in RtEGM Retinal Pigment Epithelial Cell Growth Medium fortified with RtEGM BulletKit from Lonza ( Basel , Switzerland ) . Glucose , lactate , and glutamine levels were measured from the media in apical and basal compartments using a 2900 Biochemistry Analyzer from YSI ( Yellow Springs , OH ) . A concentrated stock of DMOG from Cayman Chemical ( Ann Arbor , MI ) was made with DMSO , and added directly to the media in different concentrations to induce pseudo-hypoxia . RPE cells were maintained in low oxygen and metabolic changes were analyzed using Seahorse Flux Analysis ( see below ) in a Coy Dual Hypoxia Chamber from Coy Lab Products ( Grass Lake , MI ) ( Grassian et al . , 2014 ) . In vitro glucose consumption analyses ( GAHK20-1KT , Sigma-Aldrich; St . Louis , MO ) and in vivo glucose measurements ( SKU120003 , Eton Bioscience Inc . ; San Diego , CA ) were performed on the fourth day after doxycycline administration according to the manufacturer’s instructions . For in vivo glucose measurements , retinas were dissected out and homogenized in 80% ethanol . RPE were cultured on 96 well XF Microplates ( Agilent Technologies Inc . ; La Jolla , CA ) ) and maintained as described above . Prior to analysis human RPE were transferred to assay media ( modified Ringer’s solution lacking sodium bicarbonate ) . For glucose metabolism assays media was supplemented with 12 mM glucose , 2 mM pyruvate , 2 mM glutamine , and 10 mM HEPES , pH 7 . 4 . For fatty acid oxidation ( FAO ) assays the media contained 2 . 5 mM glucose , 0 . 5 mM carnitine and 10 mM HEPES , pH 7 . 4 . Oxygen consumption rates ( OCR ) were measured using a XFe96 device ( Agilent Technologies Inc . ) ) ; the following concentrations of drugs were used: 2 µM oligomycin , 0 . 5 µM carbonyl cyanide p-trifluoromethoxyphenylhydrazone ( FCCP ) , 1 µM antimycin A , and 1 µM rotenone . For measuring fatty acid oxidation higher concentrations of drugs were used: 2 µM oligomycin , 1 µM FCCP , 4 µM antimycin A , and 2 µM rotenone . For FAO , etomoxir ( 40 µM/well ) was added at least 30 min prior to analysis . BSA and BSA-conjugated palmitate ( Agilent Technologies Inc . ) were added to the cells immediately prior to reading . Basal OCR was calculated by subtracting the OCR in the presence of antimycin A and rotenone from that of the initial OCR prior to addition of drugs . Maximal OCR was calculated by subtracting OCR in the presence of antimycin A and rotenone from the OCR in the presence of FCCP . Cardiac perfusions were performed with freshly prepared saline , and then with 4% paraformaldehyde plus 1 . 5% gluteraldehyde in 0 . 1 M cacodylate buffer . Eyes were enucleated and fixed in the same mixture overnight at 4°C and then rinsing in 0 . 1 M Na cacodylate buffer for 1 hr . The eye cups were then postfixed in 1% OsO4 in 0 . 1 M cacodylate buffer for 2 hr , followed by another 1-hr wash and dehydration with graded ethanol solutions . Samples were incubated overnight in a 1:2 mixture of propylene oxide and Epon/Araldite ( Sigma-Aldrich ) and then placed in 100% resin followed by embedding . The blocks were sectioned and used for high-magnification electron microscopy analysis . Fully dark adapted mice were anesthetized under dim red light by intraperitoneal injection of 15 mg/kg ketamine and 7 mg/kg xylazine . Full-field ERGs were recorded from the corneal surface of each eye after pupil dilation ( 1% tropicamide and 2 . 5% phenylephrine ) using active contact lens electrodes ( Mayo; Inazawa , Japan ) placed on the cornea , a mouth reference , and tail ground electrode . A computerized system with an electronically controlled Ganzfeld dome was used ( Espion E2 with Colordome; Diagnosys; Westford , MA ) . In the dark-adapted state , we recorded rod and mixed cone/rod responses to a series of white flashes of increasing intensities ( 1 × 10–5 to 50 cd•s/m2 ) . In the light-adapted state , with a 30 cd/m2 background , cone responses to 1-Hz ( 0 . 63 to 20 cd•s/m2 ) and 30-Hz ( 3 . 98 , 10 , and 20 cd•s/m2 ) flicker stimuli were recorded . All ERG responses were filtered at 0 . 3–500 Hz , and signal averaging was applied . Retinas or RPE/choroid complexes were dissected and prepared for whole mounts or sectioning . For preparation of retinal cross-sections , dissected retinas were laid flat with 4 radial relaxing incisions , placed in 4% PFA , and incubated at 4°C overnight . Retinas were then placed in 20% sucrose at 4°C for 4 hr and embedded in Tissue-Tek OCT compound ( Sakura Finetek; Torrance , CA ) for cryosectioning . The entire retinas were sectioned , and central sections from at least three retinas from at least three mice were examined . Primary antibodies were used including anti- ZO-1 ( Life Sciences , Carlsbad , CA ) , HIF-1α ( Novus Biologicals; Littleton , CO ) , HIF-2α ( Novus Biologicals ) , and VHL ( Santa Cruz Biotechnology; Santa Cruz , CA ) . Fluorescent-conjugated isolectin Griffonia Simplicifolia IB-4 ( Lectin ) was also used ( Life Sciences ) . Images were obtained using a confocal microscope ( LSM710 , Carl Zeiss; Oberkochen , Germany ) Color fundus images were captured using a Micron III platform ( Phoenix Research Laboratories; Pleasanton , CA ) . Confocal scanning laser ophthalmoscopy ( cSLO ) detecting fundus autofluorescence , indocyanine green angiography , and optical coherence tomography ( OCT ) were performed using Spectralis ( Heidelberg Engineering; Heidelberg , Germany ) and Envisu ( Bioptigen; Durham , NC ) instruments . Retinal thickness values of the ONL , OS , RPE , and choriocapillaris were measured at 600 μm from the optic nerve head from electron micrographs using NIH ImageJ software . RPE thickness measurements were made by measuring the area between end of the basal infoldings and the outer segments of the photoreceptors . Proteomic arrays for angiogenesis-related proteins ( Angiogenesis Proteome Profiling Array , R&D Systems , Inc; Minneapolis , MN ) and ELISAs for human VEGF ( R&D Systems ) were performed according to the manufacturer’s instructions . For mRNA arrays , total RNA was prepared from RPE/choroid complexes using the RNeasy Plus Mini kit ( Qiagen; Hilden , Germany ) and was reverse transcribed using the RT² First Strand cDNA Kit ( Qiagen ) . mRNA PCR arrays for hypoxia signaling , glucose metabolism , and fatty acid metabolism ( RT² Profiler PCR Array for Mouse Hypoxia Signaling Pathway ( PAMM-032 ) , Mouse Glucose Metabolism ( PAMM-006 ) , and Mouse Fatty Acid Metabolism ( PAMM-007 ) , Qiagen ) were performed according to the manufacturer’s instructions . Quantitative PCR assays were performed on a real-time PCR System ( ABI 7900HT Fast; Thermo Fisher Scientific ) . After harvesting , eyes were flash frozen in LN2 and stored at -80°C prior to processing . Eyes were lyophilized overnight . The next day , 500 µL of 10% CHCl3 in MeOH was added to each tube . The eyes were subjected to 5 rounds of vortexing for 30s , frozen in LN2 and sonicated for 10 min at 50°C . The samples where then incubated for 1 hr at −20°C and centrifuged for 10 min @ 13 , 000 rpm , 4°C in a microcentrifuge . The supernatant was collected in a clean vial , an additoinal 500 µL of 10% CHCl3 in MeOH was added to each eye , and the extraction procedure was repeated . The supernatants were pooled and dried in a centrifugal concentrator . The resulting residue was reconstituted in 100 µL of 1:1 50 mM ammonium formate:MeCN and centrufuged at 13 , 000 rpm for 10 min in a 4°C microcentrifuge . The supernate was vialed and analyzed via LC-MS . 8 µL of each sample was analyzed on an Agilent 1200 capillary LC connected to an Agilent 6538 UHD-QToF mass spectrometer ( Agilent ) by injection on a Scherzo SM-18 150 x 2 mm column ( Imtakt; Portland , OR ) with a flow rate of 200 µL/min . For positive mode analysis , mobile phase A consisted of 0 . 1% formic acid in H2O and B was 0 . 1% formic acid in MeCN . The column was equilibrated in 100% A . The gradient was as follows: 0–5 min hold at 0% B , 0 to 100% B over 30 min , hold at 100% B for 10 min . MS data was collected from 80–1000 m/z across the entire chromatographic run , with a capillary voltage of 4000 V , a nebulizer gas flow of 11 L/min , and a pressure of 35 psig . Identical chromatographic conditions were used for metabolite identification via a targeted MS/MS analysis in which fragmentation was induced at 20 V and the product spectra collected from 50 to 1000 m/z . For negative mode analysis , mobile phase A consisted of H2O and B was 90% MeCN with 10% aqueous 50 mM ammonium formate . The column was equilibrated in 100% A . All other parameters were the same as the positive mode . XCMS was utilized to detect and align metabolic features , providing a matrix of retention time , m/z values , and intensities for each sample ( Tautenhahn et al . , 2012a ) . XCMS also provides the fold change and a P-value from an univariate t-test to determine which features are dysregulated . Features were identified by exact mass ( m/z ) , retention time , and MS/MS fragmentation by comparison against the METLIN ( http://metlin . scripps . edu/ ) database ( Tautenhahn et al . , 2012b ) . The mean , minimum , and maximum ion intensities were graphed for identified metabolites using Prism . Comparisons between the mean variables of two groups were performed by a two-tailed Student’s t-test for mRNA arrays ( RT² Profiler PCR Array Data Analysis Template v3 . 3; Qiagen ) , LC/MS ( XCMS; http://metlin . scripps . edu/xcms/ ) , and other results ( Excel; Microsoft ) . Pp <0 . 05 was considered to be statistically significant . | Cells use a sugar called glucose as fuel to provide energy for many essential processes . The light-sensing cells in the eye , known as photoreceptors , need tremendous amounts of glucose , which they receive from the blood with the help of neighboring cells called retinal pigment epithelium ( RPE ) cells . Without a reliable supply of this sugar , the photoreceptors die and vision is lost . As we age , we are at greater risk of vision loss because RPE cells become less efficient at transporting glucose and our blood vessels shrink so that the photoreceptors may become starved of glucose . To prevent age-related vision loss , we need new strategies to keep blood vessels and RPE cells healthy . However , it was not clear exactly how RPE cells supply photoreceptors with glucose , and what happens when blood supplies are reduced . To address this question , Kurihara , Westenskow et al . used genetically modified mice to investigate how cells in the eye respond to starvation . The experiments show that when nutrients are scarce the RPE cells essentially panic , radically change their diet , and become greedy . That is to say that they double in size and begin burning fuel faster while also stockpiling extra sugar and fat for later use . In turn , the photoreceptors don’t get the energy they need and so they slowly stop working and die . Kurihara , Westenskow et al . also show that there is a rapid change in the way in which sugar and fat are processed in the eye during starvation . Learning how to prevent these changes in patients with age-related vision loss could protect their photoreceptors from starvation and death . The next step following on from this research is to design drugs to improve the supply of glucose and nutrients to the photoreceptors by repairing aging blood vessels and/or preventing RPE cells from stockpiling glucose for themselves . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"cell",
"biology",
"neuroscience"
] | 2016 | Hypoxia-induced metabolic stress in retinal pigment epithelial cells is sufficient to induce photoreceptor degeneration |
Porphyrias are disorders of heme metabolism frequently characterized by extreme photosensitivity . This symptom results from accumulation of porphyrins , tetrapyrrole intermediates in heme biosynthesis that generate reactive oxygen species when exposed to light , in the skin of affected individuals . Here we report that in addition to producing an ommochrome body pigment , the planarian flatworm Schmidtea mediterranea generates porphyrins in its subepithelial pigment cells under physiological conditions , and that this leads to pigment cell loss when animals are exposed to intense visible light . Remarkably , porphyrin biosynthesis and light-induced depigmentation are enhanced by starvation , recapitulating a common feature of some porphyrias – decreased nutrient intake precipitates an acute manifestation of the disease . Our results establish planarians as an experimentally tractable animal model for research into the pathophysiology of acute porphyrias , and potentially for the identification of novel pharmacological interventions capable of alleviating porphyrin-mediated photosensitivity or decoupling dieting and fasting from disease pathogenesis .
The prosthetic group heme plays a key role in proteins ( hemoproteins ) with diverse cellular functions , including oxygen binding ( hemoglobin and myoglobin ) , electron transfer ( cytochromes ) , protection from oxidative stress ( catalases and peroxidases ) , and generation of second messengers ( nitric oxide synthase and guanylate cyclase ) ( Ponka , 1999 ) . With few exceptions ( Kořený et al . , 2012 ) , it is essential for viability in all 3 domains of life . Heme is composed of a heterocyclic tetrapyrrole called a porphyrin coordinated to a central iron atom ( some researchers classify the entire heme molecule as a porphyrin , but this term is reserved for the organic ring alone here ) . Although naturally occurring auxotrophs have been reported ( Rao et al . , 2005; Gruss et al . , 2012 ) , heme is usually synthesized from the precursor 5-aminolevulinic acid ( ALA ) through a series of enzyme-catalyzed reactions . In non-photosynthetic eukaryotes , ALA is produced in the mitochondria via condensation of glycine and succinyl CoA by ALA synthase ( ALAS ) . This is the first and typically rate-limiting step in the heme biosynthesis pathway , which proceeds in the cytoplasm before returning to the mitochondria ( Layer et al . , 2010 ) . Like ALAS , the enzymes catalyzing the 7 remaining reactions in the pathway are tightly regulated by ubiquitous and tissue-specific mechanisms ( Ponka , 1997; Balwani and Desnick , 2012 ) . Inherited , loss-of-function mutations in any of the heme biosynthesis enzymes downstream of ALAS ( or gain-of-function mutations in ALAS ) cause a group of diseases collectively referred to as porphyrias . These conditions are marked by characteristic clinical features – neurovisceral symptoms , skin lesions , or both ( Karim et al . , 2015 ) . While potentially attributable at least in part to heme deficiency , the primary etiology of these symptoms entails a buildup of pathway intermediates due to the bottleneck effect created by underlying mutations . ALA , and possibly also its pyrrole derivative porphobilinogen ( PBG ) , is neurotoxic ( Pierach and Edwards , 1978; Adhikari et al . , 2006; Felitsyn et al . , 2008; Bissell et al . , 2015 ) . Porphyrins cause cutaneous abnormalities by virtue of their photosensitizing properties . Though generated primarily in the liver or bone marrow , the major sites of heme biosynthesis , they are eventually deposited in other tissues including the dermis . There , they readily absorb light to enter an excited triplet state that reacts with oxygen to produce singlet oxygen . This in turn results in oxidative damage , such as lipid peroxidation and DNA damage , that can ultimately lead to cell death ( Poh-Fitzpatrick , 1986 ) . Thus , sunlight , or even bright indoor light , quickly damages exposed skin in many porphyria patients . Clinically , porphyrin-mediated phototoxicity manifests as edema , blistering skin lesions , and in extreme cases , disfiguring scarring and/or tissue loss ( Balwani and Desnick , 2012; Karim et al . , 2015 ) . A subset of porphyrias classified as ‘acute’ present with sudden and potentially life-threatening attacks characterized by severe abdominal pain and neurological symptoms ranging from anxiety and confusion to seizures or paralysis ( Balwani and Desnick , 2012; Karim et al . , 2015 ) . These episodes , which can last for weeks , are triggered by drugs ( e . g . , barbiturates ) , hormonal changes , dieting/fasting , and other factors that induce hepatic ALAS expression or activity . Accordingly , they are treated with therapies that effect ALAS downregulation . Carbohydrate loading reduces ALAS levels via insulin signaling and an antagonistic effect on its transcriptional coactivator PGC-1α ( peroxisome proliferator-activated receptor γ coactivator 1α ) ( Scassa et al . , 2004; Handschin et al . , 2005 ) . This is sometimes effective in ameliorating mild attacks . More severe cases are treated with intravenous heme , which downregulates ALAS expression through a feedback inhibition mechanism ( Bonkowsky et al . , 1971; Ponka , 1997 ) . A small interfering RNA therapy targeting ALAS decreased plasma ALA and PBG levels in a mouse model of acute porphyria ( Yasuda et al . , 2014 ) , and entered clinical trials in 2015 ( Alnylam Pharmaceuticals , 2016 , NCT02452372 ) . Given their toxic effects in porphyrias , it is interesting to note that porphyrins accumulate under physiological circumstances in some organisms . Porphyrins or their derivatives are pigments in the wing feathers of owls ( With , 1978; Weidensaul , 2011 ) , the brilliant crimson flight feathers of turacos ( With , 1957 ) , and numerous invertebrate lineages including earthworms , molluscs , and deep-sea medusae ( Kennedy , 1975 ) . It has been known for close to a century that high levels of porphyrins are present in the rodent Harderian gland ( Derrien and Turchini , 1924 ) , as well as in multiple tissues of the fox squirrel Sciurus niger ( Turner , 1937 ) , though the significance of these observations remains mysterious . Porphyrins are also produced by bacteria such as Propionibacterium acnes , a commensal skin microbe that contributes to the pathogenesis of acne ( Lee et al . , 1978 ) . In summary , porphyrins appear to have important , but often uncharacterized functions independent of their more well-known roles as precursors to their metal-coordinated counterparts ( e . g . , heme and the magnesium porphyrin chlorophyll ) . Here we extend previous biochemical studies documenting physiological porphyrin biosynthesis in the planarian Girardia ( formerly Dugesia ) dorotocephala ( MacRae , 1956; 1959; 1961; 1963 ) by showing the related planarian species Schmidtea mediterranea uses the first 3 enzymes in the heme biosynthesis pathway to generate porphyrins in its subepithelial pigment cells . We further demonstrate that porphyrins sensitize S . mediterranea to intense visible light , causing pigment cell loss in animals subjected to prolonged light exposure . This response is accelerated with starvation , echoing the connection between dieting or fasting and the symptomatic attacks experienced by acute porphyria patients . In addition to serving as a popular model organism for regeneration research ( Gentile et al . , 2011; Reddien , 2013; Adler and Sánchez Alvarado , 2015 ) , our observations establish planarians as an invertebrate model for in vivo studies of porphyrin photosensitization , as well as metabolic inputs into the pathophysiology of acute porphyrias .
This line of research originated with a serendipitous discovery in an undergraduate , general education course – sunlight exposure causes depigmentation of both regenerating and intact planarians ( Figure 1—figure supplement 1 ) . Infrared ( IR ) and ultraviolet B ( UVB ) radiation were neither necessary nor sufficient to induce depigmentation under conditions we tested in follow-up experiments ( Figure 1—figure supplement 2 ) . In contrast , we were able to reproduce this response with intense visible light ( Figure 1A , B and Figure 1—figure supplement 3 ) . Just over half of light-exposed animals ( 51%; n = 864 analyzed in 18 independent experiments ) developed one or more small tissue lesions on their dorsal surface; 4% lysed . Apart from these defects and their lack of bodily pigmentation , depigmented animals were indistinguishable from controls , exhibiting normal movement ( Video 1 ) , touch-responsiveness ( Video 2 ) , feeding behavior ( Figure 1—figure supplement 4A ) , and regenerative ability ( Figure 1—figure supplement 4B ) . Depigmented animals repigmented when light exposure stopped ( Figure 1—figure supplement 4C–E ) . 10 . 7554/eLife . 14175 . 003Figure 1 . Light-induced depigmentation in S . mediterranea . ( A ) Animals exposed to visible light ( incident intensity = 5000 lux; see figure supplement 3 for spectrum ) exhibit progressive loss of bodily pigmentation . Images show a single live animal photographed ( left to right ) at time 0 and immediately following each of a series of intermittent light exposure and recovery periods ( final timepoint = 10 days; see Materials and methods for details ) . Continuous exposure results in 100% lethality at this intensity . ( B ) Ventral surface of a representative depigmented animal . ( C–E ) Close-ups of control ( C ) and light-exposed ( D , E ) animals demonstrating non-uniform pigment loss ( photographs in D and E correspond to 4th and 5th images from left in A ) . Scale bars: A , B = 500 µm; C–E = 200 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 14175 . 00310 . 7554/eLife . 14175 . 004Figure 1—figure supplement 1 . Sunlight-induced depigmentation . To learn about the scientific method , non-science majors enrolled in a general education course analyzed the effects of environmental variables , including sunlight exposure , on regeneration in S . mediterranea . Control animals , maintained in a dark 20˚C incubator to mimic conditions in their natural environment ( Oviedo et al . , 2008 ) , are brown . Sunlight-exposed animals , placed on a windowsill receiving indirect sunlight for 4 hr per day and returned to a dark incubator between exposures , regenerated normally but unexpectedly lost almost all bodily pigmentation within 1 week . ( A ) Representative animals photographed immediately following amputation of cephalic and caudal tissue , and again after 7 days of regeneration . ( B ) Representative intact animals subjected to the same control and sunlight exposure conditions . Scale bars = 300 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 14175 . 00410 . 7554/eLife . 14175 . 005Figure 1—figure supplement 2 . IR and UVB radiation are neither necessary nor sufficient to induce depigmentation . ( A ) Animal photographed before ( left ) and 7 days after ( right ) a single day of sunlight exposure under IR- and UVB-blocking glass . ( B ) Animal photographed before ( left ) and after ( right ) 7 days in the dark at 28˚C ( average high temperature experienced by sunlight-exposed animals ) . Incubation at 30˚C resulted in 100% lethality within 72 hr , without apparent depigmentation . ( C ) Animal photographed before ( left ) and 7 days after ( right ) exposure to a UVB ( 302 nm ) lamp for 1 . 5 hr ( approximate LT50 exposure time; 8/16 animals died by 7 days post-exposure ) . All scale bars = 500 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 14175 . 00510 . 7554/eLife . 14175 . 006Figure 1—figure supplement 3 . Incident spectrum for visible light exposure . Relative light intensities ( total = 5000 lux ) measured under conditions experienced by animals subjected to visible light exposure . Applies to all experiments except those using a red LED ( see below ) . DOI: http://dx . doi . org/10 . 7554/eLife . 14175 . 00610 . 7554/eLife . 14175 . 007Figure 1—figure supplement 4 . Feeding , regeneration , and repigmentation in depigmented animals . ( A ) Depigmented animal fed with dyed calf liver; 21/26 depigmented animals and 25/26 controls consumed a visible amount of food . ( B ) Representative control ( left ) and depigmented ( right ) animals photographed immediately following amputation of cephalic and caudal tissue ( left in each panel ) , and again after 7 days of regeneration ( right in each panel ) . ( C ) Light-induced depigmentation is reversible . Images show a single live animal photographed ( left to right ) immediately after conclusion of light exposure , and again 7 and 21 days later . Animal was fed at 3 , 6 , and 9 days post-exposure . ( D , E ) Close-ups of a repigmenting animal photographed 3 ( D ) and 12 ( E ) days after conclusion of light exposure . Scale bars: A , B = 300 µm; C = 500 µm; D , E = 200 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 14175 . 00710 . 7554/eLife . 14175 . 008Figure 1—figure supplement 5 . Depigmentation is not due to direct photobleaching . Images show a single live animal photographed ( left to right ) at time 0 , after a 24-hr pulse of light exposure , and every 24 hr thereafter while maintained in a dark incubator . Note that depigmentation continues after light exposure ends . Scale bar = 500 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 14175 . 00810 . 7554/eLife . 14175 . 009Video 1 . Depigmented animals exhibit normal movement . Control animals maintained under standard laboratory conditions were filmed next to depigmented animals shortly after the conclusion of light exposure . DOI: http://dx . doi . org/10 . 7554/eLife . 14175 . 00910 . 7554/eLife . 14175 . 010Video 2 . Depigmented animals exhibit normal touch responsiveness . Like controls , depigmented animals change direction in response to the touch of a pipet tip . DOI: http://dx . doi . org/10 . 7554/eLife . 14175 . 010 Two observations indicated light-induced depigmentation was not a simple photobleaching effect . First , depigmentation was non-uniform ( Figure 1C–E ) , whereas direct photobleaching should result in relatively even pigment loss . Second , animals subjected to a 24-hr pulse of light exposure continued to depigment for several days after light exposure stopped ( Figure 1—figure supplement 5 ) . We conclude that prolonged exposure to intense visible light is sufficient to induce a reversible depigmentation response in S . mediterranea . To gain insight into the mechanisms of light-induced depigmentation , we sought to identify the S . mediterranea body pigment . The planarian Dugesia ryukyuensis , which also exhibits a brown color , produces a body pigment previously classified as an ommochrome on the basis of its absorption spectrum ( Hase et al . , 2006 ) . We found that a S . mediterranea body pigment co-purifying with RNA ( Materials and methods ) , like the D . ryukyuensis ommochrome , exhibited characteristic , local absorption maxima near 367 and 463 nm; a mock pigment purification from depigmented animals resulted in minimal absorbance ( Figure 2A ) . 10 . 7554/eLife . 14175 . 011Figure 2 . S . mediterranea produces an ommochrome body pigment . ( A ) Absorbance spectra of body pigment purified from control animals ( brown line ) or mock purified from depigmented animals ( grey line ) . Arrowheads denote local maxima at 367 and 463 nm , characteristic of ommochrome pigments . ( B ) Ommochrome biosynthesis pathway . Numbers in parentheses to the right of each enzyme denote the number of S . mediterranea homologs identified via reciprocal BLAST ( Materials and methods; source data 1 ) . Enzyme abbreviations are shown to the left . ( C ) Whole-mount in situ hybridizations for candidate ommochrome biosynthesis genes . Note absence of KMO-1-expressing cells from unpigmented regions of the photoreceptors ( arrowheads ) and lower numbers of these cells over the pharynx ( center ) , an area of reduced pigmentation . ( D ) RNAi phenotypes for candidate ommochrome biosynthesis genes ( results not shown for genes that did not generate a visible phenotype ) . Intact animals ( top ) were administered a total of 12 RNAi feedings over 3 . 5 weeks and photographed 3 days after the final feeding . Regenerated animals ( bottom ) were amputated at this timepoint to remove cephalic and caudal tissue . The resulting trunk fragments were allowed to regenerate for 2 weeks , administered 3 further RNAi feedings , and photographed at 21 days post-amputation . ( E ) Animals were placed in solutions containing the tryptophan 2 , 3-dioxygenase inhibitor 680C91 ( 0 . 7 µM final concentration ) or a vehicle ( ethanol ) control immediately after cephalic amputation , and photographed after 16 days of regeneration . Scale bars: C = 100 µm; D = 500 µm ( top ) , 200 µm ( bottom ) ; E = 200 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 14175 . 01110 . 7554/eLife . 14175 . 012Figure 2—source data 1 . S . mediterranea ommochrome biosynthesis enzymes . Predicted protein sequences for S . mediterranea genes identified and cloned by reciprocal BLAST and RT-PCR ( Materials and methods ) were used as queries in BLASTP searches against the non-redundant H . sapiens protein database ( NCBI ) . 1Smed Unigene transcripts are available at the Schmidtea mediterranea Genome Database ( Robb et al . , 2015 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 14175 . 012 Ommochromes are tryptophan-derived pigments produced in the eyes of many insect species , including Drosophila . Their biosynthesis involves an evolutionarily conserved pathway consisting of 4 steps ( Ryall and Howells , 1974 ) . We searched S . mediterranea genomic and EST databases ( Labbé et al . , 2012; Robb et al . , 2015; Zhu et al . , 2015 ) for candidate ommochrome biosynthesis enzymes using a reciprocal BLAST approach ( Materials and methods ) and identified a total of 7 genes corresponding to the first 3 steps in the pathway ( Figure 2B; Figure 2—source data 1 . Apparent absence of the terminal enzyme , phenoxazinone synthetase , is consistent with prior analyses of other animal genomes , including those of ommochrome-producing species ( e . g . , Croucher et al . , 2013 ) , and may reflect non-enzymatic oxidation of 3-hydroxykynurenine ( Wiley and Forrest , 1981 ) . We used whole-mount in situ hybridization ( WISH ) to analyze where these genes are expressed ( Figure 2C ) . Smed-kynurenine 3-monooxygenase-1 ( KMO-1 ) , 1 of 3 S . mediterranea KMO homologs , exhibited pigment cell-specific or enriched expression – staining was evident in cells distributed across the entire surface of the animal , but excluded from the unpigmented regions of the photoreceptors . Nearly all KMO-1-expressing cells were lost during light-induced depigmentation ( see below ) . To our knowledge , this represents the first marker for a planarian pigment cell outside of the melanin-producing cells of the eye cups ( Lapan and Reddien , 2011 ) . The second KMO homolog , Smed-kynurenine 3-monooxygenase-2 ( KMO-2 ) , also appeared to be expressed in body pigment cells ( in addition to the gut ) , while the third , KMO-3 , was expressed in a small group of cells of unknown function clustered anterior to the pharynx . We also characterized S . mediterranea homologs of 2 different enzymes capable of oxidizing tryptophan to N-formylkynurenine , the first reaction in ommochrome biosynthesis . Smed-tryptophan 2 , 3-dioxygenase ( TDO ) was expressed in the gut . Smed-indoleamine 2 , 3-dioxygenase ( IDO ) exhibited an expression pattern resembling that of KMO-1 , but we have not determined whether these transcripts are present in the same cell type . The 2 remaining pathway genes , Smed-kynurenine formamidase-1 and -2 ( KFM-1 and -2 ) , showed specific or enriched expression in the central nervous system and/or gut . We next used RNA interference ( RNAi ) to assess the functions of these genes . In multiple cases , this resulted in a noticeable change in body color ( Figure 2D ) . KMO-1 ( RNAi ) animals developed a yellow hue . This phenotype was also evident in KMO-3 ( RNAi ) animals; simultaneous knockdown of all 3 KMO homologs did not alter the color change . RNAi knockdown of TDO resulted in a charcoal grey color that was particularly apparent in the regeneration blastema , the mass of stem cell-derived tissue that forms at sites of amputation ( Figure 2D , bottom ) . Treatment with a pharmacological TDO inhibitor phenocopied the TDO ( RNAi ) effect ( Figure 2E ) . Although IDO catalyzes the same reaction as TDO , we detected at most a subtle color change in IDO ( RNAi ) animals , and did not observe an enhanced phenotype following double knockdown . No RNAi phenotypes were apparent for the 2 KFM homologs when targeted individually or in combination . These results are consistent with the prior identification of TDO ( vermilion ) and KMO ( cinnabar ) , but not KFM eye color mutants in Drosophila ( Searles and Voelker , 1986; Warren et al . , 1996 ) , as well as the potential for non-enzymatic generation of kynurenine from formylkynurenine ( Linzen , 1974 ) . It is not clear why inhibiting genes in this pathway resulted in color changes rather than depigmentation . This could be due to incomplete gene knockdown , the presence of another body pigment in analogy with the Drosophila eye , or accumulation of colored intermediates in ommochrome biosynthesis ( for instance , kynurenine imparts a yellow color to the eyes of deep-sea fish; Thorpe et al . , 1992 ) . Nevertheless , when taken together with the absorption spectrum of purified body pigment and the pigment cell expression of KMO-1 , the color changes in RNAi animals strongly suggest S . mediterranea produces an ommochrome body pigment . Ommochromes are not known to act as photosensitizers , but porphyrins , the reported body pigment in the planarian G . dorotocephala ( MacRae , 1956; 1959; 1961; 1963 ) , are . We therefore sought to determine whether porphyrins are produced in S . mediterranea , and if so , whether they might contribute to the depigmentation response we observed in light-exposed animals . A classic biochemical signature of porphyrins is their bright red/pink fluorescence under black light ( UVA ) . This trait is used to age owls on the basis of the porphyrin content of their flight feathers ( Weidensaul et al . , 2011 ) , and to detect excreted porphyrins in the urine of porphyria patients ( Balwani and Desnick , 2012 ) . We reproduced previous results showing intense red fluorescence in fixed G . dorotocephala at 400–440 nm excitation ( MacRae , 1961 ) , and found the related species Dugesia japonica was also highly fluorescent under these conditions ( Figure 3A ) . S . mediterranea exhibited minimal fluorescence by comparison; however , KMO-1 ( RNAi ) animals showed a dramatic increase in fluorescence relative to negative controls ( Figure 3B ) . To confirm this was due to porphyrins , we acid extracted whole-animal homogenates ( Materials and methods ) and determined their absorption spectra . Like G . dorotocephala and D . japonica extracts , KMO-1 ( RNAi ) extracts demonstrated bright red fluorescence under black light ( Figure 3C ) , as well as a characteristic porphyrin absorption spectrum ( Huang et al . , 2000 ) with a sharp peak ( the 'Soret' band ) around 400 nm ( Figure 3D ) . Smaller peaks ( 'Q' bands ) were evident within the visible region of the G . dorotocephala and D . japonica spectra , but could not be resolved for KMO-1 ( RNAi ) extracts . Importantly , faint fluorescence and a very small , yet reproducible Soret peak were apparent for S . mediterranea controls . We conclude that S . mediterranea makes porphyrins like G . dorotocephala , but at substantially lower levels , or primarily in a non-fluorescent ( e . g . , reduced or metal-chelate ) form . The effects of KMO-1 knockdown further suggest that porphyrin biosynthesis occurs in pigment cells and may be suppressed by ommochromes ( Discussion ) . 10 . 7554/eLife . 14175 . 013Figure 3 . Biochemical evidence of porphyrin biosynthesis in S . mediterranea . ( A ) Like G . dorotocephala , D . japonica exhibits bright red fluorescence under black light ( 400–440 nm excitation ) . S . mediterranea exhibits negligible fluorescence by comparison . ( B ) KMO-1 ( RNAi ) animals demonstrate strongly increased fluorescence relative to negative controls . The uniform fluorescence in the anterior ( top ) corresponds to recently regenerated tissue ( animals were photographed 3 . 5 weeks after cephalic amputation ) . Fluorescence does not appear to be restricted to pigment cells , particularly within regenerated tissue; this may reflect porphyrin movement across cell membranes ( Viljoen et al . , 1975 ) . ( C ) Whole-animal H2SO4 homogenates were photographed in plastic cuvettes over a black ( 400 nm ) LED . ( D ) Representative absorption spectra of whole-animal homogenates . Black arrowheads denote the Soret peak ( 405 nm ) ; a visible increase in the height of this peak was evident for 3/3 KMO-1 ( RNAi ) homogenates relative to Neg . Con . ( RNAi ) homogenates . White arrowheads denote Q bands ( 459 nm for G . dorotocephala; 459 , 549 , and 584 nm for D . japonica ) . All scale bars = 200 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 14175 . 013 Porphyrins are physiological intermediates in heme biosynthesis . To address the hypothesis that they are produced in S . mediterranea pigment cells , we identified , cloned , and characterized genes in the heme biosynthesis pathway ( Figure 4A; Figure 4—source data 1 ) , using the same basic approach as described above for the ommochrome pathway . Strikingly , S . mediterranea homologs of ALAS , ALA dehydratase ( ALAD ) , and PBG deaminase ( PBGD ) , enzymes catalyzing the first 3 reactions in metazoan heme biosynthesis , were highly expressed in pigment cells , while the 5 remaining genes in the pathway were expressed predominantly or exclusively in other cell types ( Figure 4B ) . In agreement with histological analyses of Dugesia gonocephala chromatophores ( Palladini et al . , 1979 ) , PBGD-1-expressing cells were located just beneath the epithelium ( Figure 4C ) and exhibited a dendritic morphology also characteristic of other pigment-producing cell types ( Figure 4D ) . Double fluorescent in situ hybridization ( dFISH ) confirmed Smed-PBGD-1 ( PBGD-1 ) was co-expressed with KMO-1 ( Figure 4E ) . 10 . 7554/eLife . 14175 . 014Figure 4 . Genetic evidence of porphyrin biosynthesis in S . mediterranea pigment cells . ( A ) Heme biosynthesis pathway . Numbers in parentheses to the right of each enzyme denote the number of S . mediterranea homologs identified via reciprocal BLAST ( Materials and methods; source data 1 ) . Enzyme abbreviations are shown to the left . ( B ) Whole-mount in situ hybridizations for porphyrin/heme biosynthesis genes . ALAS , ALAD-1 , and PBGD-1 , -2 , and -3 expression patterns resemble that of KMO-1 ( Figure 2C ) . PBGD-4 , UROD-1 , CPOX , and FECH-1 exhibit enriched expression in the gut . UROS is highly expressed in the brain . ( C ) 10 µm transverse section from the anterior of a PBGD-1-labeled animal , stained with nuclear fast red . PBGD-1 expression is subepithelial , higher on the dorsal surface ( matching that surface’s higher level of pigmentation ) , and excluded from the unpigmented regions just above the photoreceptors ( arrowheads ) . ( D ) PBGD-1 fluorescent in situ hybridization ( FISH ) , showing dendritic morphology of pigment cells . ( E ) Double FISH showing overlap in KMO-1 and PBGD-1 expression . Over 90% of KMO-1-positive cells were co-labeled with PBGD-1 and vice versa ( n = 11 animals analyzed by confocal microscopy ) . ( F ) RNAi phenotypes for porphyrin/heme biosynthesis genes ( results not shown for genes that did not generate a visible phenotype , or that generated phenotypes unrelated to pigmentation – see figure supplement 1 ) . Intact animals ( top ) were administered a total of 12 RNAi feedings over 3 . 5 weeks and photographed 3 days after the final feeding . Regenerated animals ( bottom ) were amputated at this timepoint to remove cephalic and caudal tissue . The resulting trunk fragments were allowed to regenerate for 2 weeks , administered 3 further RNAi feedings , and photographed at 21 days post-amputation . ( G ) Long-term RNAi feeding for PBGD-1 leads to complete loss of bodily pigmentation . This animal was from a group fed a total of 50 times over 6 months , with periodic amputation to increase numbers . Scale bars: B–D = 100 µm; E = 50 µm; F = 500 µm ( top ) , 200 µm ( bottom ) ; G = 500 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 14175 . 01410 . 7554/eLife . 14175 . 015Figure 4—source data 1 . S . mediterranea porphyrin/heme biosynthesis enzymes . Predicted protein sequences for S . mediterranea genes identified and cloned by reciprocal BLAST and RT-PCR ( Materials and methods ) were used as queries in BLASTP searches against the non-redundant H . sapiens protein database ( NCBI ) . 1Smed Unigene transcripts are available at the Schmidtea mediterranea Genome Database ( Robb et al . , 2015 ) . 2Not cloned ( no RT-PCR product ) . DOI: http://dx . doi . org/10 . 7554/eLife . 14175 . 01510 . 7554/eLife . 14175 . 016Figure 4—figure supplement 1 . Additional RNAi phenotypes for porphyrin/heme biosynthesis genes . ( A ) Animals were administered RNAi feedings every 2–3 days until less than 50% consumed a visible amount of food . This corresponded to a total of 8 ( ALAS ) or 9 ( PBGD-4 and UROD-1 , as well as negative control ) feedings . Representative animals were photographed as phenotypes arose ( 0–7 days after final RNAi feeding ) . Note that ALAS ( RNAi ) animals are slightly lighter than negative controls . Head regression and ventral curling are phenotypes indicative of stem cell dysfunction . Lesions are consistent with defects in one or more differentiated cell types . 100% of ALAS ( RNAi ) , PBGD-4 ( RNAi ) , and UROD-1 ( RNAi ) animals died within 2 weeks of the final RNAi feeding , compared to 0% of negative controls ( n > 30 animals per condition ) . ( B ) Absorbance spectra of body pigment purified from long-term Neg . Con . ( RNAi ) and PBGD-1 ( RNAi ) animals ( fed more than 50 times over more than 6 months , with periodic amputation to increase numbers ) . Arrowheads denote local maxima at 367 and 463 nm , characteristic of ommochrome pigments . All scale bars = 300 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 14175 . 016 ALAS , ALAD , and PBGD , but not downstream enzymes , exhibited RNAi phenotypes indicative of a role in pigment biosynthesis . Specifically , while ALAS ( RNAi ) animals developed morphological defects culminating in 100% lethality , they became lighter than negative controls prior to dying ( Figure 4—figure supplement 1A ) . PBGD-1 ( RNAi ) animals were viable and showed a more pronounced reduction in pigmentation ( Figure 4F , top ) , eventually turning completely white with sustained RNAi feeding ( Figure 4G ) . Biochemical analysis confirmed ommochrome loss ( Figure 4—figure supplement 1B ) . Following amputation , PBGD-1 ( RNAi ) animals failed to produce new body pigment in the blastema ( Figure 4F , bottom ) . ALAD-1 knockdown , while not effecting a pigmentation change in uncut animals , also disrupted pigment biosynthesis in newly regenerated tissue . No changes in pigmentation were evident in either intact or regenerating animals following knockdown of the remaining heme biosynthesis enzymes , though RNAi phenotypes unrelated to pigmentation and 100% lethality were observed for PBGD-4 and Smed-uroporphyrinogen decarboxylase-1 ( UROD-1 ) ( Figure 4—figure supplement 1A ) . These defects , as well as the lethality associated with ALAS knockdown , could in theory be due to heme deficiency . However , many organisms readily absorb heme from their diets ( Rao et al . , 2005; Shayeghi et al . , 2005 ) , making alternative explanations possible ( e . g . , noncanonical functions for pathway enzymes; Greenbaum et al . , 2003; Zhang et al . , 2014 ) . To summarize , we found the first 3 enzymes in heme biosynthesis are expressed in the body pigment cells of S . mediterranea and are required for pigmentation , while UROS and the remaining enzymes in the pathway are not expressed ( or are limiting ) in this cell type , and are dispensable for pigment biosynthesis . These results imply that S . mediterranea pigment cells share a metabolic bottleneck with the erythroid cells of individuals suffering from the rare , autosomal recessive disorder congenital erythropoietic porphyria ( CEP ) ( Xu et al . , 1996 ) . Patients with this condition , also known as Gunther’s disease , have reduction-of-function mutations in uroporphyrinogen III synthase ( UROS ) , the 4th enzyme in heme biosynthesis , and consequently produce high levels of porphyrins in their erythroid cells ( Discussion ) . Some of the porphyrin molecules generated by CEP patients accumulate in the skin , causing severe cutaneous photosensitivity usually beginning in infancy ( Xu et al . , 1996; Balwani and Desnick , 2012; Karim et al . , 2015 ) . Porphyrin-mediated photosensitization is observed to varying degrees in other porphyrias as well and has been clinically exploited in photodynamic therapy ( PDT ) . This technique entails administration of one or more photosensitizing compounds followed by irradiation with a wavelength of light absorbed by the sensitizer ( s ) . Red light is used in porphyrin-based PDT because it is transmitted by epithelial tissues but absorbed by porphyrins . The end result is production of singlet oxygen in targeted ( e . g . , tumor ) cells , leading to cell death ( Agostinis et al . , 2011 ) . Consistent with a porphyrin-based mechanism , red light ( 625 nm ) was sufficient to induce full bodily depigmentation ( Figure 5A ) , while the antioxidants dimethylthiourea ( DMTU ) and ascorbic acid exerted inhibitory effects ( Figure 5B and Figure 5—figure supplement 1A ) . Hypoxia was also protective ( Figure 5—figure supplement 1B ) . We further determined that depigmentation was due to pigment cell loss – light exposure eliminated KMO-1/PBGD-1-positive cells , with no apparent decrease in expression of these genes within remaining pigment cells at intermediate timepoints ( Figure 5C ) . Both white light and red light induced a significant increase in cell death , as measured by whole-mount TUNEL ( Figure 5D , E ) . Although we were unable to determine the cellular specificity of this response with existing methods , systemic induction of cell death by RNAi knockdown of a BCL-2 homolog results in large tissue lesions and lysis ( Pellettieri et al . , 2010 ) . These phenotypes were rarely observed following white light exposure and never observed with red light , suggesting cell death is not systemically induced in these contexts . 10 . 7554/eLife . 14175 . 017Figure 5 . Visible light exposure causes pigment cell loss . ( A ) Red light ( 625 nm LED ) is sufficient to induce full bodily depigmentation . Inset shows a magnified view of the dorsal surface , brightness , contrast , and gamma-enhanced to highlight remaining pigment cells . No animals developed lesions or lysed under these conditions ( n = 345 analyzed in 23 independent experiments ) . ( B ) DMTU inhibits light-induced depigmentation . Representative control and DMTU-treated animals were photographed before and after white light exposure ( left and right in each panel , respectively ) . DMTU treatment ( 10 mM final concentration ) was initiated 5 days prior to the start of light exposure and continued for the duration of the experiment . Total light exposure time = 72 hr ( 24- and 48-hr exposures were separated by a 24-hr dark recovery ) . ( C ) Light-induced depigmentation is due to pigment cell loss . Images show representative red light-exposed animals fixed at the indicated times and labeled with KMO-1 or PBGD-1 riboprobes . ( D ) Light-induced cell death visualized by whole-mount TUNEL . Light-exposed animals were fixed 12 hr after a 24-hr exposure . ( E ) Quantitative analysis of TUNEL results . The number of TUNEL-positive nuclei ( TPN ) /mm2 was averaged over 3 independent experiments ( n = total of 38 dark , 31 white and red light-exposed animals ) . Error bars = +/- s . e . m . *p-value <1 x 10–4 for two-tailed student’s t-test comparing light-exposed animals with controls . Scale bars: A = 300 µm; B = 500 µm; C , D = 100 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 14175 . 01710 . 7554/eLife . 14175 . 018Figure 5—figure supplement 1 . Antioxidants and hypoxia inhibit light-induced depigmentation . ( A ) Representative control and ascorbic acid-treated animals photographed before and after white light exposure ( left and right in each panel , respectively ) . Ascorbic acid treatment ( 10 mM final concentration , pH 7 . 0 ) was initiated 5 days prior to the start of light exposure and continued for the duration of the experiment . Total light exposure time = 60 hr ( 24- and 36-hr exposures were separated by a 24-hr dark recovery ) . ( B ) Animals were photographed before and after ( left and right images in each panel , respectively ) a series of intermittent periods of red light exposure under normoxic or hypoxic conditions . See Materials and methods for details . Scale bars = 200 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 14175 . 018 To directly address the hypothesis that porphyrins mediate light-induced depigmentation , we used RNAi to experimentally manipulate porphyrin levels in vivo . As noted above , KMO-1 ( RNAi ) animals exhibited a strong increase in porphyrin fluorescence relative to negative controls . Simultaneous PBGD-1 knockdown abrogated this effect ( Figure 6A ) ; it is unlikely this was due to decreased efficacy of KMO-1 knockdown , because a double RNAi control with Smed-ferrochelatase-1 ( FECH-1 ) did not alter fluorescence ( Figure 6—figure supplement 1 ) . As predicted , KMO-1 ( RNAi ) animals showed increased photosensitivity , with 100% failing to survive 48 hr of light exposure , less than 1/3 the exposure time used to achieve depigmentation in controls . Simultaneous PBGD-1 ( but not FECH-1 ) knockdown fully rescued viability ( n = 20 animals per condition analyzed in 2 independent experiments ) . We conclude that PBGD-1 is required for porphyrin biosynthesis in pigment cells . 10 . 7554/eLife . 14175 . 019Figure 6 . Porphyrins mediate light-induced pigment cell loss . ( A ) PBGD-1 knockdown suppresses the porphyrin fluorescence observed in KMO-1 ( RNAi ) animals . The difference in appearance of anterior tissues ( top ) between negative controls and KMO-1/PBGD-1 ( RNAi ) animals is a consequence of the latters’ failure to repigment newly regenerated tissue ( animals were amputated 3 . 5 weeks prior to photographing ) . See figure supplement 1 for additional controls . ( B ) PBGD-1 knockdown suppresses light-induced depigmentation . Animals were photographed before and after 48 hr of red light exposure . Note the greater pigmentation in PBGD-1 ( RNAi ) animals after exposure , despite their lower initial pigmentation . For reasons that are presently unclear , this effect was restricted to the posterior , which typically depigments at a lower rate than anterior tissues . ( C ) PBGD-1 knockdown suppresses light-induced pigment cell loss . A KMO-1 riboprobe was used to visualize pigment cells in animals fixed before and after 7 days of continuous red light exposure . Scale bars: A , B = 200 µm; C= 100 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 14175 . 01910 . 7554/eLife . 14175 . 020Figure 6—figure supplement 1 . Porphyrin fluorescence controls . Unlike PBGD-1 knockdown , FECH-1 knockdown did not suppress the porphyrin fluorescence observed in KMO-1 ( RNAi ) animals . The slightly elevated , uniform fluorescence in FECH-1 ( RNAi ) animals relative to negative controls is a reproducible effect that may reflect protoporphyrin IX accumulation in non-pigment cell types . Note that the images of Neg . Con . ( RNAi ) , KMO-1 ( RNAi ) , and KMO-1/PBGD-1 ( RNAi ) animals are the same as in Figure 6A ( all images are from a single experiment ) . All scale bars = 200 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 14175 . 020 Because PBGD-1 ( RNAi ) animals lose all bodily pigmentation in the absence of light exposure ( Figure 4G ) , we were unable to make a direct comparison of light-induced depigmentation rates between fully affected animals and controls . However , we did observe partial protection from light-induced depigmentation when PBGD-1 ( RNAi ) animals were light exposed prior to developing full phenotypic expressivity ( Figure 6B ) . Furthermore , by using a KMO-1 riboprobe to label pigment cells by in situ hybridization , we showed PBGD-1 knockdown conferred protection from light-induced pigment cell loss ( Figure 6C ) . We conclude that light-induced depigmentation in S . mediterranea is due to the photosensitizing action of porphyrins in its subepithelial pigment cells . Planarians can survive for months without feeding , undergoing an up to ~20-fold reduction in size through a tissue remodeling process involving increased cell death and decreased production of stem cell division progeny ( Pellettieri et al . , 2010; González-Estévez et al . , 2012 ) . During the course of this research , we noticed a correlation between how long animals were fasted prior to light exposure and the extent of photosensitivity . To document this relationship , we fed animals 4 times in 7 days , using dyed calf liver to verify that they had eaten during each feeding ( as in Figure 1—figure supplement 4A ) . These relatively well-fed animals were then starved for 1 , 7 , 14 , or 30 days before being light exposed . Depigmentation was strongly accelerated with starvation ( Figure 7A ) , and reversed by a single feeding 24 hr prior to initiation of light exposure ( Figure 7B ) . We considered the possibility that starvation might sensitize planarians to any inducer of cell death as a trivial explanation for these results . However , no difference in TUNEL staining was evident between 7 and 14 day-starved animals exposed to a sublethal dose of gamma irradiation ( Figure 7—figure supplement 1 ) , arguing against a nonspecific effect . 10 . 7554/eLife . 14175 . 021Figure 7 . Starvation induces porphyrin biosynthesis and acute photosensitivity . ( A ) Animals were fed 4 times in 1 week with dyed calf liver and then starved as indicated prior to 72 hr of red light exposure . Representative animals were photographed pre-exposure and 72 hr after the conclusion of light exposure . ( B ) Animals given a single feeding after 29 days of starvation and light exposed 24 hr later showed far less depigmentation than 30 day-starved animals . Post-exposure photographs were taken 24 hr after the conclusion of light exposure , as full depigmentation was already apparent in 30 day-starved animals . ( C ) qRT-PCR analysis of ALAS expression . The fold change relative to 24 hr-starved animals was averaged over 3 biological replicates . Depigmented animals showed reduced expression , as predicted based on the ALAS expression pattern ( Figure 4B ) . Error bars = +/- standard deviation . **p-value <0 . 001 for two-tailed student’s t-test in comparison with 24 hr starved; *p-value <0 . 01 . ( D ) Quantitative analysis of porphyrin fluorescence in D . japonica lysates , averaged over 10 biological replicates . AFU/mg = arbitrary fluorescence units/mg wet tissue weight . Error bars = +/- standard deviation . **p-value <1 x 10–7 for two-tailed student’s t-test; *p-value <0 . 001 . ( E ) Photoprotective effect of different food sources . Well-fed animals were fasted , light exposed , and photographed as in ( A ) , with a subset re-fed as indicated after 13 days . Light exposure was initiated for all groups at 24 hr post-feeding ( day 14 ) . Scale bars: A , B = 300 µm; E = 500 µm ( top ) , 200 µm ( bottom ) . DOI: http://dx . doi . org/10 . 7554/eLife . 14175 . 02110 . 7554/eLife . 14175 . 022Figure 7—figure supplement 1 . Starvation does not sensitize animals to induction of cell death by gamma radiation . ( A ) Cell death was visualized by whole-mount TUNEL in animals starved for 7 or 14 days , and exposed to a 1250 rad dose of gamma radiation as indicated . All animals were fixed 24 hr post-irradiation . ( B ) Quantitative analysis of TUNEL results . The number of TUNEL-positive nuclei ( TPN ) /mm2 was averaged over 5 independent experiments ( n > 40 animals per condition ) . Error bars = +/- s . e . m . NS = p-value >0 . 2 for two-tailed student’s t-test . The difference in animal size between conditions was also statistically insignificant ( p-value >0 . 3 ) . All scale bars = 100 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 14175 . 022 Reduced nutrient intake induces hepatic ALAS expression in mammals through the cAMP/CREB/PGC- 1α pathway ( Handschin et al . , 2005 ) . When this occurs in the presence of a heme biosynthesis bottleneck , as in acute porphyria patients , it can result in porphyrin accumulation . We found that ALAS expression was likewise elevated by starvation in S . mediterranea ( Figure 7C ) , and took advantage of the strong porphyrin fluorescence in D . japonica ( Figure 3A , C ) to show that fasting also leads to increased porphyrin levels in tissue homogenates ( Figure 7D ) . Because calf liver , the food source used in these experiments , is rich in heme , and heme exerts feedback inhibition of ALAS expression ( Bonkowsky et al . , 1971; Ponka , 1997 ) , we sought to determine whether this might account for the photoprotective effects of feeding . Egg yolk ( another laboratory food source ) was slightly less effective than liver in reversing starvation-induced photosensitivity ( Figure 7E ) , consistent with this possibility . However , addition of exogenous heme had no impact . While our results do not exclude a role for dietary heme in regulation of porphyrin biosynthesis in planarians , they do suggest porphyrin levels are at least partly influenced by metabolic regulation of ALAS expression , as in acute porphyrias .
The striking variety of pigment colors and patterns in the Platyhelminthes is one of the most conspicuous characteristics of this phylum , yet only a few pigments have been biochemically or genetically identified to date . The black color of the eye cups in species including S . mediterranea , D . ryukyuensis , and likely G . dorotocephala is due to melanins ( Ness et al . , 1996; Hase et al . , 2006; Lapan and Reddien , 2011; Lambrus et al . , 2015 ) , while the brown body color in D . ryukyuensis and G . dorotocephala has been attributed to ommochromes and porphyrins , respectively ( MacRae , 1956; 1959; 1961; 1963; Hase et al . , 2006 ) . It is not clear whether the latter observations reflect interspecies variation , or whether each of these species generates multiple body pigments as we propose here for S . mediterranea . Hemoglobin has been identified in free-living and endosymbiotic rhabdocoels as well as trematode parasites ( Young and Harris , 1973; Phillips , 1978; Jennings and Cannon , 1987; Kiger et al . , 1998 ) . Ommochromes and porphyrins have a wide range of colors in addition to brown and black . Named for their prevalence in the ommatidia of arthropod eyes , the former can also be red or yellow . Porphyrins and their derivatives not only confer the red color of hemoglobin and the green color of chlorophyll , but can sometimes be blue or purple ( ‘porphyrin’ is derived from the ancient Greek word for purple , ‘porphura’ ) . Thus , the growing evidence for production of these pigments in brown freshwater planarians also makes them logical candidates for contributing to the bright hues seen in many terrestrial flatworms and marine polyclads ( e . g . , Breugelmans et al . , 2012; Lapraz et al . , 2013; Noreña et al . , 2014 ) . Our initial characterization of the underlying biosynthetic pathways ( Figures 2 and 4 ) will facilitate future evo-devo studies exploring the evolutionary basis for this diversity , as well as the lack of pigmentation in some cave-dwelling species ( e . g . , de Souza et al . , 2015 ) . Additionally , the observation that light-exposed S . mediterranea repigment when returned to a dark environment ( Figure 1—figure supplement 4C–E ) , presumably through replacement of lost pigment cells ( Figure 5C ) , sets the stage for mechanistic analyses of pigment cell differentiation . The RNAi phenotypes of KMO-1 and PBGD-1 further imply biochemical or genetic interactions between ommochromes and porphyrins or their underlying biosynthetic pathways . KMO-1 knockdown resulted in increased porphyrin fluorescence ( Figure 3B , C ) and heightened photosensitivity , while PBGD-1 knockdown caused loss of all visible bodily pigmentation ( Figure 4G; Figure 4—figure supplement 1B ) . One possible explanation for these results is that ommochromes and porphyrins are jointly used as precursors in a downstream biosynthetic step to generate a single , brown body pigment , or that these molecules physically interact in a manner that quenches porphyrin fluorescence and is required for ommochrome maintenance . We are unaware of documented ommochrome-porphyrin interactions , but porphyrins can form complexes with melanins and this quenches their fluorescence in vitro ( Ito et al . , 1992; Losi et al . , 1993 ) . An alternative explanation for the effects of KMO-1 knockdown is that ommochromes maintain porphyrins in their non-fluorescent , reduced ( porphyrinogen ) state . This could account for the minimal porphyrin fluorescence observed in S . mediterranea ( Figure 3A ) . Ommochromes are known to have antioxidant activity in other species ( Ostrovsky et al . , 1987; Insausti et al . , 2013; Romero and Martínez , 2015 ) . Finally , our results are consistent with a negative feedback loop in which porphyrins induce ommochrome biosynthesis , with ommochromes in turn repressing porphyrin biosynthesis . In this regard , it is interesting to speculate that kynurenine or other tryptophan metabolites , rather than ommochromes themselves , might have signaling activities that impact porphyrin levels . Ommochromes and metalloporphyrins have diverse biological functions extending beyond pigmentation per se . Adaptive roles for porphyrins in the absence of metal coordination , while suggested by the ‘physiological porphyria’ in organisms such as fox squirrels , are not understood . Thus , we can only speculate as to possible body pigment function ( s ) in planarians . One possibility is photoreception . S . mediterranea displays strong negative phototaxis . This is partly dependent on the melanin-producing pigment cups of the photoreceptors , yet animals lacking eye pigment retain photophobic behavior ( Lambrus et al . , 2015 ) , and developing embryos display light avoidance prior to eye differentiation ( Sánchez Alvarado , 2003 ) . These observations point to the existence of one or more extraocular photoreceptors ( Paskin et al . , 2014 ) . The burning pain experienced by some porphyria patients upon sunlight exposure ( Balwani and Desnick , 2012 ) makes porphyrins logical candidates for further investigation . Planarians have long attracted the interest of biologists investigating regeneration ( Newmark and Sánchez Alvarado , 2002 ) , and have recently emerged as a useful invertebrate model for human disorders including Usher syndrome ( Lapan and Reddien , 2012 ) and cystic kidney disease ( Thi-Kim Vu et al . , 2015 ) . Our results add acute porphyrias to this growing list . The porphyrias are classified based on multiple criteria , including: 1 ) whether intermediates in heme biosynthesis are formed predominantly in the liver or bone marrow; 2 ) whether symptoms manifest primarily as skin lesions or neurovisceral episodes; and 3 ) the step at which heme biosynthesis is disrupted . Light-induced depigmentation in planarians combines aspects of multiple disease variants ( Figure 8 ) . UROS deficiency , the blockage in heme biosynthesis underlying CEP , is also the apparent determinant of physiological porphyrin biosynthesis in S . mediterranea . In the absence of this enzyme , its linear tetrapyrrole substrate hydroxymethylbilane undergoes spontaneous cyclization to form uroporphyrinogen I , a pathogenic isomer of uroporphyrinogen III that can in turn undergo spontaneous oxidation to form porphyrins . As in the human disease , our results implicate this biochemistry as the cause of severe cutaneous photosensitivity , but this is typically a chronic symptom in CEP , whereas it is induced by starvation in planarians . In the latter respect , S . mediterranea represents a unique animal model of acute porphyrias . The episodic nature of acute porphyria symptoms is attributed to hepatic induction of ALAS1 , 1 of 2 human isoforms of the gene , by endogenous and exogenous factors including dieting or fasting ( Balwani and Desnick , 2012; Karim et al . , 2015 ) . This can pose a weight-loss challenge for individuals carrying disease alleles and may be a complicating factor in bariatric surgery ( Lopes et al . , 2008 ) . ALAS and porphyrin levels are also increased in response to starvation in planarians ( Figure 7C , D ) . These results are formally consistent with an increase in pigment cell number relative to other cell types as animal size decreases , but we regard this as unlikely because planarians become lighter in color , not darker , during prolonged starvation ( Newmark and Sánchez Alvarado , 2002; Miller and Newmark , 2012 ) . Furthermore , an increase in pigment cell density might be predicted to slow the rate of light-induced depigmentation , rather than accelerate it , and the rapid reversal of porphyrin-dependent photosensitivity when starved animals are re-fed ( Figure 7B , E ) is more consistent with a direct metabolic effect than a change in pigment cell density . In mice , fasting-induced ALAS upregulation is dependent on PGC- 1α , and inhibited by insulin signaling ( Scassa et al . , 2004; Handschin et al . , 2005 ) . While we have been unable to identify a PGC- 1α homolog in S . mediterranea , insulin-like signaling has been described ( Miller and Newmark , 2012 ) . Future experiments , including RNAi knockdown of nutrient-sensing genes and the development of a defined diet , will be important for identifying the mechanisms linking metabolic cues to ALAS expression in planarians . Existing animal models of acute porphyrias commonly rely on chemical treatments to evoke phenotypes resembling human disease symptoms . The planarian model reported here complements these systems in several respects . First , off-target effects of porphyrogenic compounds can be avoided . Second , induction of photosensitization via starvation has the potential to reveal novel metabolic inputs into heme biosynthesis , as well as the toxic effects of pathway intermediates . Third , the fluorescence of fixed animals and easily inducible depigmentation response provide a convenient means of assessing porphyrin levels and their photosensitizing activity in vivo . This could prove especially useful in screening for drugs capable of reducing porphyrin biosynthesis ( constitutive or induced ) or alleviating porphyrin-based photosensitivity . The protective effect of DMTU against light exposure ( Figure 5B ) constitutes proof of principle for such an approach . Although porphyrias are usually manageable diseases , reliance on intravenous heme or liver transplantation to treat severe cases can result in significant complications ( Seth et al . , 2007 ) . There is no approved prophylactic treatment for patients who suffer recurrent attacks . We suggest planarians represent an experimentally tractable animal model in which to explore both physiological and pathological roles of porphyrins , and to seek new avenues for modulating their biosynthesis toward therapeutic benefit .
Asexual clonal populations of S . mediterranea ( strain CIW4 ) , G . dorotocephala , and D . japonica were maintained under standard laboratory conditions , as previously described ( Oviedo et al . , 2008 ) . Except where otherwise indicated , animals were fed homogenized calf liver 1–3 times per week and starved for 7 to 8 days prior to use in experiments . Designated animals in Figure 7E were fed boiled egg yolk from organic , free-range chickens , supplemented with 10% ( w/w ) hemin chloride ( Calbiochem ) as indicated . Animals were exposed to white light from a compact fluorescent bulb ( 14W , 2700K , EcoSmart; see Figure 1—figure supplement 3 for spectrum ) or red light from a light-emitting diode ( LED; 625 nm , BML Horticulture ) in 12-well plates , 1 animal per well . Continuous red light exposure typically resulted in full depigmentation of 1 week-starved S . mediterranea within 3–5 days ( smaller animals required less time on average ) . Continuous white light exposure resulted in 100% lethality prior to full depigmentation with a lamp height corresponding to 5 , 000 lux; lower intensities were less effective in inducing depigmentation . We empirically determined that exposure periods of 24 , 48 , 48 , and 48 hr , interspersed with 24-hr recovery periods in a dark , 20˚C incubator ( standard laboratory conditions ) , almost completely eliminated lethality while achieving full depigmentation . This 10-day , intermittent exposure regimen was used in all white light experiments , unless otherwise indicated . UV-blocking glass ( Tru Vue ) was placed between animals and the light source to guard against any possible cell damage from sustained exposure to trace amounts of near-visible UVA radiation . KG3 IR/UVB-blocking glass ( Omega Optical ) was placed directly above sunlight-exposed animals for the experiment in Figure 1—figure supplement 2A . Photographs of whole animals were obtained with an Olympus SZX16 microscope equipped with a DP72 digital camera . Photographs of tissue sections ( Figure 4C ) and FISH ( Figure 4D ) were obtained with an Olympus BX53 microscope using the same camera . dFISH results ( Figure 4E ) were photographed with a Quorum Spinning Disk Confocal 2 ( Olympus IX81 microscope and Hamamatsu C9100-13 EM-CCD camera ) . Raw images were captured using Olympus CellSens ( SZX16 and BX53 ) or Perkin Elmer Volocity ( confocal ) software . All photographs of live , light-exposed animals were acquired using identical camera settings and were processed only by placing images from different timepoints , different experimental conditions , etc . on a uniform , black background . Some variability in apparent animal color/pigmentation due to the position of the animal relative to the objective at the time of photographing , and position of the gooseneck lamps used for illumination was unavoidable . However , this variability was inconsequential compared to the pigmentation changes documented in the representative images shown in figures . Linear adjustments ( brightness and contrast ) were made for images of animals labeled by WISH , FISH , dFISH , and TUNEL , or fixed to reveal porphyrin fluorescence , in order to best represent actual results . These adjustments were identical within a given experiment where comparisons were drawn between conditions . The absorption spectra in Figures 2A and Figure 4—figure supplement 2B were attained for a black S . mediterranea body pigment co-purifying with RNA ( pigments like melanin commonly co-purify with nucleic acids in standard extraction procedures; Giambernardi et al . , 1998 ) . Briefly , experimental and control animals were decapitated to remove the melanin-producing eye cups , and equal amounts of remaining trunk/tail fragments ( 0 . 1 g and 0 . 05 g wet tissue weight per sample in Figures 2A and Figure 4—figure supplement 2B , respectively ) were homogenized in TRIzol reagent ( ThermoFisher Scientific ) . Following chloroform extraction , the aqueous phase of each extract was precipitated with isopropanol . The black precipitate ( not visible in the mock extraction from depigmented animals or PBGD-1 ( RNAi ) extracts ) was washed in ethanol , air dried , and resuspended in molecular biology grade water . Samples were treated with RNase I ( NEB ) for 1 hr at 37°C , after which TRIzol/chloroform extraction was repeated and the pigment and mock pigment samples were again resuspended in water . Degradation of RNA was confirmed by gel electrophoresis and absorption spectra were obtained on an HT Synergy plate reader ( BioTek ) . We tested a variety of acid extraction protocols for porphyrin purification . All resulted in intense fluorescence under black light and absorption spectra with a pronounced Soret peak for G . dorotocephala and D . japonica extracts; in contrast , we typically detected minimal fluorescence for S . mediterranea extracts and/or were unable to resolve a clear Soret band . Homogenization in 1M sulfuric acid produced visible fluorescence and a small , but reproducible Soret peak . It is possible this reflects some degree of oxidation of non-fluorescent porphyrinogens to their oxidized , fluorescent form , yet addition of reagents commonly used for this purpose , including hydrogen peroxide and iodine or Lugol’s solution ( Martásek et al . , 1982 ) , had no obvious effect . Light exposure , which can photooxidize porphyrinogens , similarly failed to alter S . mediterranea fluorescence , either in extracts or in whole animals . KMO-1 knockdown consistently enhanced extract fluorescence and resulted in a more prominent Soret peak , regardless of extraction method . Results in Figures 3C , D and 7D were obtained by centrifuging whole-animal H2SO4 homogenates for 10 min at 1452 x g , prior to analysis of absorbance/fluorescence . Equal amounts of tissue ( 30 mg wet weight ) were used for each species/RNAi condition in Figure 3C . Absorption spectra were determined using an HT Synergy plate reader ( BioTek ) . Fluorescence measurements ( Figure 7D ) were obtained using the same protocol and plate reader , with ~25 mg ( wet tissue weight ) D . japonica per sample . Readings were obtained with 400/30 nm ( excitation ) and 600/40 nm ( emission ) filters , and normalized for tissue weight . Porphyrin fluorescence was illustrated in Figure 3C by illuminating extracts in cuvettes with a black LED ( 400 nm , BML Horticulture ) . Animals in Figures 3A , B , 6A , and Figure 6—figure supplement 1 were flash-killed in 5% N-acetyl cysteine ( NAC ) in PBS , transferred to 95% ethanol ( MacRae , 1961 ) , mounted on glass slides in this solution , and photographed with an Olympus SZX16 microscope using a 400–440 nm excitation filter . Animals were exposed to TDO inhibitor 680C91 ( Focus Biomolecules ) , DMTU , and Ascorbic Acid ( Sigma-Aldrich ) as indicated in the relevant figure legends ( Figures 2E , 5B , and Figure 5—figure supplement 1A ) . Hypoxia ( Figure 5—figure supplement 1B ) was induced by placing animals in glass vials sealed with rubber stoppers and bubbling N2 into the water ( no air was bubbled into control vials ) . Dissolved O2 was reduced by approximately 85% using this approach . Animals were immediately subjected to a 5-hr pulse of red LED exposure ( in glass vials ) and then returned to normoxic conditions overnight ( sustained hypoxia was lethal ) . This process was repeated a total of 4 times over 5 days ( days 0 , 1 , 3 , and 4 ) , and depigmentation was assessed 3 days after the final light exposure . To identify candidate ommochrome and porphyrin biosynthesis enzymes ( Figures 2B and 4A ) , human and mouse protein sequences were used as queries in TBLASTN searches against S . mediterranea genomic and EST databases ( Labbé et al . , 2012; Robb et al . , 2015; Zhu et al . , 2015 ) . Top hits were used as queries in reciprocal BLASTX searches against the non-redundant protein database ( NCBI ) , and discarded if results did not match the identity of the original query . cDNA sequences were then cloned into a double-stranded RNA ( dsRNA ) expression vector ( pT4P; Rink et al . , 2009 ) by RT-PCR , using primers shown in Supplementary file 1 . We were unable to amplify ALAD-2 , UROD-2 , and FECH-2 by RT-PCR; there were also no ESTs matching these genomic ORFs in the S . mediterranea Genome Database ( Robb et al . , 2015 ) , raising the possibility they are pseudogenes . WISH , FISH , and dFISH were performed as previously described ( Pearson et al . , 2009; Zhu et al . , 2015; Currie et al . , 2016 ) , using riboprobes prepared from pT4P clones , and imaged as described above . Whole-mount TUNEL was also performed according to published protocols ( Pellettieri et al . , 2010 ) , and quantified using ImageJ software ( http://rsb . info . nih . gov/ij ) . Designated animals in Figure 7—figure supplement 1 were exposed to 1250 rad of gamma radiation , a sublethal dose that depletes cycling stem cells ( Wagner et al . , 2011 ) , using a Cs-137 source , 81-14R irradiator ( J . L . Shepherd & Associates ) . Irradiation was completed 24 hr prior to fixation . dsRNA-expressing E . coli cultures were prepared using pT4P clones , mixed with homogenized calf liver , and fed to animals as previously described ( Zhu et al . , 2015 ) . An RNAi vector with C . elegans unc-22 was used as a negative control . To evaluate pigmentation changes in intact animals as a result of existing pigment turnover ( Figures 2D and 4F , top panels ) , animals were fed RNAi food 12 times over 3 . 5 weeks ( days 0 , 2 , 4 , 7 , 9 , 11 , 14 , 16 , 18 , 21 , 23 , and 25 ) , discarding animals that did not eat after each feeding . Photographs were taken 3 days after the final feeding ( day 28 ) . For genes that generated phenotypes unrelated to pigmentation ( Figure 4—figure supplement 1A ) , feedings were discontinued when less than 50% of animals ate . The experiment in Figure 6B followed the same RNAi feeding schedule ( animals were light exposed 7 days after the final RNAi feeding ) . To evaluate pigmentation changes arising from defects in pigment biosynthesis during regeneration of new , initially unpigmented tissue ( Figures 2D and 4F , bottom panels ) , animals were amputated to form head , trunk , and tail fragments on day 28 . Regenerating trunk fragments were fed again on days 42 , 44 , and 46 , and anterior blastemas were photographed on day 49 , at 21 days post-amputation . In cases involving the potential for genetic redundancy ( e . g . , KFM-1 and -2 ) , equal amounts of dsRNA-expressing bacterial cultures were combined to prepare RNAi food ( Gurley et al . , 2008 ) . All other RNAi experiments involved 4 RNAi feedings ( days 0 , 2 , 4 , and 7 ) , followed by a single round of amputation on day 8 to promote tissue turnover , and another 4 RNAi feedings on the same schedule initiated 10 to 12 days post-amputation . Phenotypes were analyzed 7 days after the final RNAi feeding . Animals were rinsed and observed every 1–2 days . ALAS qRT-PCR was performed in triplicate as previously described ( Lin and Pearson , 2014 ) , using size-matched animals fed 4 times in 1 week with dyed calf liver at the start of the experiment . Animals were starved for 24 hr or 14 days , as indicated in Figure 7C , prior to preparation and reverse transcription of total RNA using TRIzol reagent ( ThermoFisher Scientific ) and SuperScript III reverse transcriptase ( Invitrogen ) . cDNA was amplified with LightCycler 480 SYBR Green I Master reaction mix ( Roche ) in a CFX96 Touch Real-Time PCR Detection System ( Bio-Rad ) , and results were normalized using the ubiquitously expressed GAPDH gene as a reference . Primers were designed to span exon-exon boundaries: ALAS: Forward - CAACGAGTGATTGTTAAGTCTGG; Reverse - GACAGACATTCATTTGGTTGCTC GAPDH: Forward - AGCTCCATTGGCGAAAGTTA; Reverse - CTTTTGCTGCACCAGTTGAA | Porphyrias are rare diseases that involve ring-shaped molecules called porphyrins accumulating in various parts of the body . Porphyrins are produced as part of the normal process that makes an important molecule called heme , which is required to transport oxygen . However , high levels of porphyrins can be toxic . For example , porphyrins deposited in the skin can cause swelling and blistering when the skin is exposed to bright light . Other disease symptoms include neurological issues ranging from anxiety and confusion to seizures or paralysis . It has been speculated that porphyrias may have affected several historical figures , including the artist Vincent van Gogh . In addition to their role in heme production , porphyrins also have other roles . For example , they are used as pigments in the wing feathers of some owls . Researchers are trying to understand more about how organisms regulate porphyrin production so that it might be possible to develop more effective treatments for porphyria in humans . Here , Stubenhaus et al . studied how a flatworm called Schmidtea mediterranea makes porphyrins . A group of undergraduate students noticed that these animals – which are normally brown in color – turned white when they were exposed to sunlight for several days . Stubenhaus et al . found that S . mediterranea makes porphyrins in the pigment cells of its skin using the same genes that make porphyrins in humans . Together with other molecules called ommochromes , the porphyrins give rise to the normal color of this flatworm . However , when the animals are exposed to intense light for extended periods of time , which is unlikely to occur in the wild , porphyrin production leads to loss of the pigment cells . The experiments also show that starvation increases the rate of pigment cell loss in light-exposed flatworms , which mirrors the worsening of disease symptoms some porphyria patients experience when they diet or fast . Stubenhaus et al . propose that flatworms are useful models in which to study the molecular processes that are responsible for porphyrias in humans . Further research is required to determine the exact chemical structure of the porphyrin and ommochrome molecules produced in different flatworm species . Stubenhaus et al . also plan to use flatworms to screen for drugs that could potentially be developed into new treatments for porphyria . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"biochemistry",
"and",
"chemical",
"biology",
"cell",
"biology"
] | 2016 | Light-induced depigmentation in planarians models the pathophysiology of acute porphyrias |
The sulfhydration of cysteine residues in proteins is an important mechanism involved in diverse biological processes . We have developed a proteomics approach to quantitatively profile the changes of sulfhydrated cysteines in biological systems . Bioinformatics analysis revealed that sulfhydrated cysteines are part of a wide range of biological functions . In pancreatic β cells exposed to endoplasmic reticulum ( ER ) stress , elevated H2S promotes the sulfhydration of enzymes in energy metabolism and stimulates glycolytic flux . We propose that transcriptional and translational reprogramming by the integrated stress response ( ISR ) in pancreatic β cells is coupled to metabolic alternations triggered by sulfhydration of key enzymes in intermediary metabolism .
Posttranslational modification is a fundamental mechanism in the regulation of structure and function of proteins . The covalent modification of specific amino acid residues influences diverse biological processes and cell physiology across species . Reactive cysteine residues in proteins have high nucleophilicity and low pKa values and serve as a major target for oxidative modifications , which can vary depending on the subcellular environment , including the type and intensity of intracellular or environmental cues . Oxidative environments cause different post-translational cysteine modifications , including disulfide bond formation ( -S-S- ) , sulfenylation ( -S-OH ) , nitrosylation ( -S-NO ) , glutathionylation ( -S-SG ) , and sulfhydration ( -S-SH ) ( also called persulfidation ) ( Finkel , 2012; Mishanina et al . , 2015 ) . In the latter , an oxidized cysteine residue included glutathionylated , sulfenylated , and nitrosylated on a protein reacts with the sulfide anion to form a cysteine persulfide . The reversible nature of this modification provides a mechanism to fine tune biological processes in different cellular redox states . Sulfhydration coordinates with other post-translational protein modifications such as phosphorylation and nitrosylation to regulate cellular functions ( Altaany et al . , 2014; Sen et al . , 2012 ) . Despite great progress in bioinformatics and advanced mass spectroscopic ( MS ) techniques , identification of different cysteine-based protein modifications has been slow compared to other post-translational modifications . In the case of sulfhydration , a small number of proteins have been identified , among them the glycolytic enzyme glyceraldehyde phosphate dehydrogenase , GAPDH ( Mustafa et al . , 2009 ) . Sulfhydrated GAPDH at Cys150 exhibits an increase in its catalytic activity , in contrast to the inhibitory effects of nitrosylation or glutathionylation of the same cysteine residue ( Mustafa et al . , 2009; Paul and Snyder , 2012 ) . The biological significance of the Cys150 modification by H2S is not well-studied , but H2S could serve as a biological switch for protein function acting via oxidative modification of specific cysteine residues in response to redox homeostasis ( Paul and Snyder , 2012 ) . Understanding the physiological significance of protein sulfhydration requires the development of genome-wide innovative experimental approaches . Current methodologies based on the modified biotin switch technique do not allow detection of a broad spectrum of sulfhydrated proteins ( Finkel , 2012 ) . Guided by a previously reported strategy ( Sen et al . , 2012 ) , we developed an experimental approach that allowed us to quantitatively evaluate the sulfhydrated proteome and the physiological consequences of H2S synthesis during chronic ER stress . The new methodology allows a quantitative , close-up view of the integrated cellular response to environmental and intracellular cues and is pertinent to our understanding of human disease development .
The ER is an organelle involved in the synthesis of proteins followed by various modifications . Disruption of this process results in the accumulation of misfolded proteins , causing ER stress ( Tabas and Ron , 2011; Walter and Ron , 2011 ) , which is associated with development of many diseases ranging from metabolic dysfunction to neurodegeneration ( Hetz , 2012 ) . ER stress induces transcriptional , translational , and metabolic reprogramming , all of which are interconnected through the transcription factor ATF4 . ATF4 increases expression of genes promoting adaptation to stress via their protein products . One such gene is the H2S-producing enzyme , cystathionine gamma-lyase ( CTH ) or γ-cystathionase , previously shown to be involved in the signaling pathway that negatively regulates the activity of the protein tyrosine phosphatase 1B ( PTP1B ) via sulfhydration ( Krishnan et al . , 2011 ) . We therefore hypothesized that low or even modest levels of reactive oxygen species ( ROS ) during ER stress may reprogram cellular metabolism via H2S-mediated protein sulfhydration ( Figure 1A ) . 10 . 7554/eLife . 10067 . 003Figure 1 . ER stress induces protein sulfhydration , a reversible cysteine-based post-translational modification . ( A ) Schematic overview of protein sulfhydration , which requires synthesis of H2S and low ROS levels . Pancreatic β cells ( MIN6 ) were treated with thapsigargin ( Tg ) for the indicated times , and the cellular levels of ROS ( B ) , total levels of GSH and GSH/GSSG ratios ( C ) and H2S levels ( E ) were evaluated . ( D ) RT-qPCR analysis of the mRNA levels for the H2S-producing enzyme CTH and the cystine/glutamate exchanger Slc7a11 in MIN6 cells treated with Tg or pancreatic islets as indicated . ( F ) Evaluation of GAPDH activity in cell extracts from MIN6 cells treated with Tg at the indicated times ( top ) , and GAPDH protein levels by Western blot analysis ( bottom ) . ( G ) . Time-dependent measurements of human recombinant GAPDH activities after exposure to H2O2 ( 50 μM , blue ) or H2O2 together with the H2S donor , NaHS ( 50 μM , red ) . ( H ) In vitro evaluation of the reversal of the inhibitory effect of glutathionylation on the activity of recombinant GAPDH treated for 15 min with either NaHS ( 20 mM ) , DTT ( reduced dithiothreitol , 20 mM ) , or NaCl ( 20 mM ) . ( I ) . Evaluation of GAPDH activities in MIN6 cell extracts either untreated or treated with Tg ( 18 hr ) with or without the CTH inhibitor , PAG ( 3 mM ) ( top ) . PAG was included for the last 3 . 5 hr of Tg-treatment . GAPDH protein levels were evaluated by Western blotting ( bottom ) . All quantifications are presented as mean ± S . E . M . of three independent experiments . CTH: γ-cystathionase; ER: endoplasmic reticulum; PAG: propargylglycine; ROS: reactive oxygen species . DOI: http://dx . doi . org/10 . 7554/eLife . 10067 . 00310 . 7554/eLife . 10067 . 004Figure 1—figure supplement 1 . ER stress induces the levels of the H2S-producing enzyme CTH but not CBS . Western blot analysis for the indicated proteins , of cell extracts isolated from Tg treated MIN6 cells for the indicated times . CTH: γ-cystathionase; ER: endoplasmic reticulum . DOI: http://dx . doi . org/10 . 7554/eLife . 10067 . 00410 . 7554/eLife . 10067 . 005Figure 1—figure supplement 2 . Regulation of gene expression in MIN6 cells , human , and mouse islets in response to ER stress . MIN6 cells ( A ) and human islets ( B ) were treated with Tg for the indicated times . Islets were isolated from WT and heterozygous Akita Ins2 c96y 6-week-old mice ( C ) . RNA was isolated from cells and islets and was tested by RT-qPCR for expression of the indicated genes . ER: endoplasmic reticulum; WT: wild type . DOI: http://dx . doi . org/10 . 7554/eLife . 10067 . 00510 . 7554/eLife . 10067 . 006Figure 1—figure supplement 3 . Activation of the integrated stress response leads to increased expression of CTH in wild-type mouse islets treated with Tg . Western blot analysis for the indicated proteins of cell extracts isolated from Tg treated islets for 24 hr . CTH: γ-cystathionase . DOI: http://dx . doi . org/10 . 7554/eLife . 10067 . 00610 . 7554/eLife . 10067 . 007Figure 1—figure supplement 4 . Glutamate uptake in MIN6 cells treated with Tg at the indicated times . DOI: http://dx . doi . org/10 . 7554/eLife . 10067 . 00710 . 7554/eLife . 10067 . 008Figure 1—figure supplement 5 . Analysis of S-glutathionylated GAPDH by LC-MS/MS . Recombinant GAPDH ( 20 μg ) was treated with GSSG ( 5 mM ) for 45 min , and then incubated with 50 mM of NEM . The sample was resolved by SDS-PAGE electrophoresis and analyzed by LC-MS/MS . Tandem mass spectrum of the active site peptide IISNASCTTNCLAPLAK of the protein with glutathionylated Cys150 , and Cys154 alkylated with NEM . Comparison with the same peptide both NEM alkylation , there are no mass changes of y series ions from y4 to y10 and b series ions from b2 to b4 , but a mass shift from y11 to y16 , and b8 , b9 ions , which strongly suggests the modification at Cys150 . @ represents the neutral loss of -129Da from a glutathione adduct . DOI: http://dx . doi . org/10 . 7554/eLife . 10067 . 008 We have previously shown that the insulin-producing mouse pancreatic β cells MIN6 , known for their high metabolic activity , are very susceptible to ER stress ( Guan et al . , 2014; Krokowski et al . , 2013 ) . We tested whether or not MIN6 cells expressed the essential components of H2S synthesis and protein sulfhydration in response to ER stress induced by thapsigargin ( Tg , Figure 1 ) . Upon Tg treatment , MIN6 cells exhibited higher levels of intracellular ROS , decreased GSH/GSSG ( reduced/oxidized glutathione ratios ) , and increased CTH protein levels via transcriptional activation ( Figure 1B–D , Figure 1—figure supplement 1 and 2A ) . This transcriptional reprogramming was also seen in mouse and human islets subjected to physiological or pharmacologically induced ER stress ( Figure 1D , Figure 1—figure supplement 2B , C and 3 ) . In agreement with increased CTH expression , H2S levels increased during the chronic phase of the stress response ( Figure 1E ) . Coordinated induction of expression of the gene encoding the glutamate/cystine exchanger , Slc7a11 , also was increased ( Figure 1D ) , and this induction was associated with increased glutamate/cystine flux ( Figure 1—figure supplement 4 ) . Slc7a11 mediates the exchange of oxidized extracellular cystine with intracellular glutamate . Once cystine is imported into the cells it gets reduced to cysteine , and then serves as a substrate for GSH and H2S synthesis . These data support the idea that both increased uptake of the CTH substrate and increased levels of CTH contribute to increased H2S levels in cells under ER stress . The functional significance of increased H2S synthesis was shown by measuring the catalytic activity of GAPDH ( Mustafa et al . , 2009 ) , which gradually increased in response to elevated H2S production in MIN6 cells during ER stress ( Figure 1F ) . This increase in activity was independent of GAPDH protein levels ( Figure 1F ) . As noted above , GAPDH has the unusual feature of being catalytically inactive when Cys150 is oxidatively modified , except when it undergoes sulfhydration which restores/increases its activity ( Mustafa et al . , 2009 ) . We therefore tested the protective effects of Cys150 sulfhydration by H2S on its catalytic activity in the presence of H2O2-induced oxidation . Recombinant GAPDH was incubated with H2O2 in the presence or absence of the H2S donor , NaHS . The inhibition of GAPDH activity by H2O2 was significantly reversed by H2S treatment ( Figure 1G ) . Furthermore , incubation of purified GAPDH with oxidized glutathione ( GSSG ) resulted in formation of inactive glutathionylated GAPDH ( Gao et al . , 2010 ) , which was significantly rescued by treatment with H2S as well as DTT reduction ( Figure 1H ) . HPLC-MS confirmed that recombinant GAPDH exposed to NaHS or GSSG was modified predominantly at Cys150 ( Figure 1—figure supplement 5 , Figure 2—figure supplement 7B , C ) . These data confirm that H2S is a positive regulator of GAPDH activity . Increased GAPDH activity is directly linked to H2S production , as shown by the loss of induction ( Figure 1I ) upon treatment with the CTH inhibitor , propargylglycine ( PAG ) . PAG inhibits H2S synthesis and therefore is expected to decrease Cys150 modification . Taken together , these results indicate that regulation of H2S synthesis during ER stress might regulate the catalytic activity of other metabolic pathway proteins . The latter is raising the possibility that the ATF4-mediated sulfhydration of proteins is part of the integrated stress response ( ISR ) , and has regulatory effects on cellular metabolism . ATF4 increases gene expression of CTH ( Dickhout et al . , 2012 ) , the cystine transporter Slc7a11 , as well as the ROS-producing enzyme Ero1α ( Han et al . , 2013; Tabas and Ron , 2011 ) . We hypothesized that during ER stress , this network of ATF4 target genes promotes protein sulfhydration ( Figure 2A ) . Knockdown of ATF4 in MIN6 cells during Tg-induced ER stress caused inhibition of H2S synthesis with a parallel loss of induction of CTH protein levels ( Figure 2B ) . In contrast , in the absence of stress , ATF4 overexpression increased CTH and Ero1α levels and H2S synthesis , in agreement with increased GAPDH activity ( Figure 2C–E , Figure 2—figure supplement 1 ) . The levels of GSH and the activity of the glutamate/cystine exchanger also increased with ATF4 overexpression ( Figure 2F , G ) . These data support the hypothesis that ATF4 is a master regulator of protein sulfhydration in pancreatic β cells during ER stress . 10 . 7554/eLife . 10067 . 009Figure 2 . ATF4-mediated transcriptional reprogramming during ER stress increases expression of a gene cohort involved in H2S synthesis and protein sulfhydration . ( A ) Schematic representation of the ATF4-induced cohort of genes leading to sulfhydration of proteins during ER stress . ( B ) Evaluation of intracellular H2S levels ( top ) or ATF4 and CTH protein levels ( bottom ) in MIN6 cells infected with either control shRNA or shRNA against ATF4 . Cells were untreated or treated with Tg for 18 hr . ( C ) Western blot analysis for the indicated proteins in MIN6 cells infected ( for 48 hr ) with either adenovirus mediated lacZ-expression as control , or ATF4-expression , at increasing viral particle concentrations . MIN6 cells infected with either control adenovirus or ATF4-expressing adenovirus were used to measure ( D ) H2S levels ( E ) GAPDH relative activities , ( F ) GSH levels , and ( G ) Glutamate ( Glu ) uptake by the cystine/glutamate exchanger . ( H ) Schematic representation of the novel Biotin-Thiol-Assay ( BTA ) , an experimental approach for the identification of sulfhydrated proteins in cell extracts . Highly reactive cysteine residues or sulfhydrated cysteine residues in proteins under native conditions were alkylated with low concentrations of maleimide-PEG2-biotin ( NM-Biotin ) . Subsequent avidin column purification and elution with DTT , which cleaved the disulfide bonds , leaving the biotin tag bound to the column , produced an eluate that was further analyzed either by western blotting or coupled with LC-MS/MS . ( I ) Identification via the BTA of sulfhydrated GAPDH in MIN6 cell extracts from Tg-treated cells in the presence or absence of ATF4 . ( J ) Identification of sulfhydrated GAPDH in MIN6 cells overexpressing ATF4 . ( K ) Determination of the effect of PAG on sulfhydrated GAPDH levels in MIN6 cells overexpressing ATF4 in the presence or absence of PAG . ER: endoplasmic reticulum; PAG: propargylglycine . DOI: http://dx . doi . org/10 . 7554/eLife . 10067 . 00910 . 7554/eLife . 10067 . 010Figure 2—source data 1 . Sulfhydrated proteins from MIN6 cells treated with Tg for 18 hr . DOI: http://dx . doi . org/10 . 7554/eLife . 10067 . 01010 . 7554/eLife . 10067 . 011Figure 2—figure supplement 1 . Expression of genes in MIN6 cells overexpressing the transcription factor ATF4 . RT-qPCR analysis of RNA isolated from cells infected with control adenovirus ( - ) or ATF4-expressing adenovirus ( + ) for 48 hr , for the indicated genes . The results are shown as the average of three independent determinations . DOI: http://dx . doi . org/10 . 7554/eLife . 10067 . 01110 . 7554/eLife . 10067 . 012Figure 2—figure supplement 2 . Schematic representation of the predicted proteins in the eluate of the BTA approach as a consequence of increasing concentrations of biotin conjugated maleimide ( NM-biotin , red ) . At low concentration ( top ) , the high reactive -SH groups ( orange ) including unmodified and sulfhydrated cysteines are discriminated for alkylation , leading to elution of sulfhydrated proteins from the avidin column by DTT . At the high concentration ( bottom ) , all -SH groups in cysteines are labeled , leading to the proteins permanently bound to the beads and unable to be eluted by DTT . BTA: Biotin thiol assay . DOI: http://dx . doi . org/10 . 7554/eLife . 10067 . 01210 . 7554/eLife . 10067 . 013Figure 2—figure supplement 3 . Increasing concentrations of NM-biotin in the BTA of cell extracts isolated from Tg treated for 18 hr MIN6 cells , inhibit the elution of sulfhydrated proteins . Eluted proteins were analysed by SDS–PAGE and stained by Coomassie blue . Western blot analysis for tubulin , of equal amount of MIN6 cell extracts before binding to the avidin column . BTA: biotin thiol assay . DOI: http://dx . doi . org/10 . 7554/eLife . 10067 . 01310 . 7554/eLife . 10067 . 014Figure 2—figure supplement 4 . H2S covalently modifies proteins via sulfhydration of cysteine residues . ( A ) Mouse liver lysates were subjected to NM-biotin , then divided into eight equal fractions , were bound on avidin columns and eluted by the addition of the indicated concentrations of DTT . The eluates were analysed by SDS–PAGE electrophoresis and stained by Coomassie blue . ( B ) Lysates from MIN6 cells treated with Tg for 18 hr , were pretreated with or without 20 mM DTT for 20 min , and after passing through a NAP-5 gel filtration column , were subjected to the BTA assay followed by SDS–PAGE electrophoresis and silver staining of the gels . Equal amount of extracts before loading on the avidin columns were analyzed by Western blotting for tubulin . DTT treatment of extracts before the avidin column , reversed global protein sulfhydration , implying a covalent cysteine-based modification . DOI: http://dx . doi . org/10 . 7554/eLife . 10067 . 01410 . 7554/eLife . 10067 . 015Figure 2—figure supplement 5 . Assessment of the specificity of BTA to identify reactive -S-SH groups of proteins via the use of recombinant GAPDH . ( A ) The purity of human recombinant GAPDH was evaluated by SDS–PAGE electrophoresis and was stained by Coomassie blue . *- indicates GST-tagged GAPDH . ( B ) Recombinant GAPDH ( 100 μg ) was incubated for 45 min with 50 μM NaCl , as the control , H2O2 ( mediating oxidation of cysteines ) or NaHS ( mediating sulfhydration of cysteines ) . Following desalting , samples were subjected to the BTA assay . Eluates were analyzed by Western blotting for GAPDH . Extracts from samples before loading to the avidin column were also analyzed by Western blotting for GAPDH ( input ) . BTA: biotin thiol assay . DOI: http://dx . doi . org/10 . 7554/eLife . 10067 . 01510 . 7554/eLife . 10067 . 016Figure 2—figure supplement 6 . Assessment of GAPDH sulfhydration by the red maleimide assay . Recombinant GAPDH ( 20 μg ) was exposed to NaHS ( 50 μM ) , or NaCl ( 50 μM ) as the control , for 45 min . The levels of sulfhydrated GAPDH were evaluated with the red maleimide assay and measured by the decreasing fluorescence intensity after β‑mercaptoethanol ( β-ME ) treatment . DOI: http://dx . doi . org/10 . 7554/eLife . 10067 . 01610 . 7554/eLife . 10067 . 017Figure 2—figure supplement 7 . BTA identifies sulfhydration of GAPDH at the catalytic cysteine , Cys150 . ( A ) Both wild-type ( WT ) and Cys150Ser recombinant GAPDH mutant ( 100 μg ) were incubated with 50 μM each , NaHS , H2O2 , or NaCl ( as the control ) , for 45 min . After desalting , samples were subjected to the BTA assay . Eluates were analyzed by Western blot analysis for GAPDH . Sulfhydration was observed only for the WT GAPDH . ( B ) WT recombinant GAPDH ( 20 μg ) was pretreated with DTT ( 20 mM ) for 45 min following desalting . The protein was subsequently incubated with or without NaHS ( 50 μM ) for 45 min , followed with treatment with NEM ( 50 mM ) . The latter treatment is expected to block all free -SH groups . The samples were resolved by SDS–PAGE electrophoresis followed by LC-MS/MS analysis . The NaHS-treated GAPDH showed a mass shift of Cys150 corresponding to the sulfinic acid due to sulfhydration ( bottom ) . The oxidation of the S-SH group of Cys150 to sulfinic acid is expected due to its high reactivity when exposed to O2 under the aerobic conditions of the experiment . BTA: biotin thiol assay . DOI: http://dx . doi . org/10 . 7554/eLife . 10067 . 01710 . 7554/eLife . 10067 . 018Figure 2—figure supplement 8 . The BTA assay shows that Tg-induced ER stress in MIN6 cells promotes global protein sulfhydration . ( A ) The BTA assay was performed to detect protein sulfhydration in extracts from MIN6 cells treated with Tg for 18 hr . ( B ) Western blot analysis confirmed that GAPDH was sulfhydrated in MIN6 cells during Tg treatment . ( C ) GAPDH activity in cell extracts from untreated or Tg-treated MIN6 cells for 18 hr . ( D ) LC-MS/MS of a subset of the sulfhydrated proteins in ( A ) , including GAPDH , actin , and β-tubulin ( Mustafa et al . , 2009 ) . BTA: biotin thiol assay . DOI: http://dx . doi . org/10 . 7554/eLife . 10067 . 01810 . 7554/eLife . 10067 . 019Figure 2—figure supplement 9 . LC-MS/MS spectrum of H2S-modified peptides purified by the BTA technique from MIN6 cells treated with Tg for 18 hr . DOI: http://dx . doi . org/10 . 7554/eLife . 10067 . 01910 . 7554/eLife . 10067 . 020Figure 2—figure supplement 10 . Western blot analysis of eluates from the BTA of sulfhydrated GAPDH and PHGDH in MIN6 cells treated with Tg in a time-dependent manner . a peak of protein sulfhydration occurred at 12 h of Tg treatment . BTA: biotin thiol assay . DOI: http://dx . doi . org/10 . 7554/eLife . 10067 . 02010 . 7554/eLife . 10067 . 021Figure 2—figure supplement 11 . Western blot analysis for CTH protein levels from MIN6 cells infected with either adenovirus-mediated expression of GFP as control , or shRNA against the CTH mRNA in a dose . ( A ) and time- ( B ) dependent manner . ( C ) Western blot analysis of the indicated proteins from cell extracts before applying to the column ( input ) or after the column ( eluate ) . The effects of CTH knockdown by shRNA on sulfhydrated GAPDH and PHGDH levels in MIN6 cells treated with or without Tg for 18 hr , is shown . DOI: http://dx . doi . org/10 . 7554/eLife . 10067 . 021 To profile genome-wide changes in protein sulfhydration by ER stress and the integrated stress response , we exploited the different reactivity of cysteine persulfides ( Cys-S-SH ) and thiols ( Cys-SH ) to alkylating agents ( Nishida et al . , 2012; Pan and Carroll , 2013; Paul and Snyder , 2012 ) , to develop a new thiol reactivity-based approach called BTA ( Biotin Thiol Assay , ( Figure 2H and Figure 2—figure supplement 2 ) . We employed the following steps ( Figure 2H ) : ( 1 ) labeling of reactive Cys-SH or Cys-S-SH groups by a biotin-conjugated maleimide ( maleimide-PEG2-biotin , NM-biotin ) ( Weerapana et al . , 2010 ) , ( 2 ) binding of the biotin-labeled proteins on an avidin column , ( 3 ) elution of the retained proteins that contain a persulfide bridge using DTT , and ( 4 ) analysis of the eluted proteins . The BTA approach was validated by testing the hypothesis that a lower concentration of the thiol-alkylating reagent ( NM-biotin ) will result in a selective labeling of highly reactive cysteine SH groups , and that this selectivity will be lost as the concentration of NM-biotin increases ( Weerapana et al . , 2010 ) . We tested this hypothesis in extracts isolated from Tg-treated MIN6 cells in the presence of increasing concentrations of NM-biotin ( 0 . 05–1 mM ) . The eluate containing the sulfhydrated proteins was analyzed with SDS–PAGE and the proteins were visualized by Coomassie blue staining . Indeed , increasing the concentration of NM-biotin beyond 0 . 5 mM decreased the levels of eluted proteins ( Figure 2—figure supplement 3 ) , consistent with cysteine residues having free -SH groups becoming alkylated at high concentrations of NM-biotin . No protein was eluted from the column with the addition of DTT at high concentrations of NM-biotin ( Figure 2—figure supplement 3 ) , confirming that the biotin was attached to proteins via a disulfide bond ( Figure 2—figure supplement 4A ) . This selectively labeling behavior not only relies on the probe concentration , but also is dependent on the protein structure conformations . We found that no protein was detected in the eluates if the BTA was performed on the denatured cell extracts ( data not shown ) . When Tg-treated MIN6 lysates were pretreated with DTT ( to decrease the level of sulfhydrated proteins ) , the signal was significantly reduced compared to untreated lysates ( Figure 2—figure supplement 4B ) , confirming that DTT reduces an intermolecular disulfide bond of cysteine persulfides labeled with biotin . These data show that sulfhydrated cysteine residues are the primary targets of NM-biotin at low concentrations , thus making the BTA a unique tool to identify the sulfhydrated proteome ( Figure 2—figure supplement 2 ) . We next assessed the BTA assay using recombinant GAPDH , which contains six cysteine residues including the redox-regulated Cys150 . GAPDH was incubated with either H2O2 or NaHS . The presence of sulfhydrated GAPDH was evaluated by the BTA method following Western blot analysis . We found that only the NaHS-treated GAPDH was eluted as an H2S-modified target , indicating that the assay distinguishes sulfhydration or free SH groups from other forms of cysteine modifications ( Figure 2—figure supplement 5B ) . This was independently confirmed with the red malemide assay ( Figure 2—figure supplement 6 ) , which discriminates between free -SH groups and sulfhydrated-SH ( Sen et al . , 2012 ) . We next tested if the BTA can identify modification of a specific cysteine on GAPDH by using recombinant wild-type and Cys150Ser GAPDH mutant , which were incubated with H2O2 or NaHS . We found that NaHS treatment induced sulfhydration in the wild-type GAPDH ( Figure 2—figure supplement 7A ) , a result that was also confirmed by high-resolution quadruple MS analysis ( Figure 2—figure supplement 7B–C ) , and that this modification was absent in the Cys150Ser mutant . Finally , a proteome-wide view of sulfhydration was obtained from Tg-treated MIN6 cell extracts subjected to BTA and further analyzed by LC-MS/MS ( Figure 2—figure supplement 8 ) . We identified 150 proteins , including several known targets for sulfhydration ( Figure 2—figure supplement 8D ) ( Mustafa et al . , 2009 ) . Similar results were obtained from analysis of mouse liver ( data not shown ) , a tissue known to exhibit high levels of H2S synthesis ( Kabil et al . , 2011 ) . Taken together , this shows that BTA discriminates between protein sulfhydration and other oxidative modifications . Next , the BTA methodology identified the sulfhydrated proteome downstream of the transcription factor ATF4 during ER stress ( Figure 2A ) . MIN6 cells treated with Tg increased GAPDH levels in the DTT eluate , that was abolished by knocking down ATF4 ( Figure 2I ) . In the absence of stress , ATF4 knockdown resulted in an increase in sulfhydrated GAPDH in MIN6 cells ( Figure 2I ) . Because ATF4-deficient cells have decreased levels of sulfur-containing amino acids ( Harding et al . , 2003 ) , and sulfur amino acid restriction is linked to an increase in the transulfuration pathway ( Hine et al . , 2015 ) , it is possible that the increase in GAPDH sulfhydration in ATF4-depleted cells is due to activation of CBS ( cystathionine-β-synthase ) ( Niu et al . , 2015 ) , the second cytosolic H2S-producing enzyme . Moreover , we found that in the absence of stress ATF4 overexpression induced GAPDH sulfhydration that is dependent on the CTH activity ( Figure 2J–K ) . The use of PAG decreased GAPDH sulfhydration ( Figure 2K ) . We conclude that sulfhydration of proteins during ER stress is part of the ISR and is controlled by the transcription factor , ATF4 . The BTA requires determining the concentration of NM-Biotin for selectively labeling proteins with reactive , sulfhydrated cysteines rather than the relatively high abundant and less reactive , unmodified ( with free SH groups ) cysteine residues . However , this labeling step has some limitations under certain circumstances . For example , if free Cys-SH groups are biotinylated on the same protein containing one or more Cys-S-SH groups ( Figure 2H and Figure 2—figure supplement 2 ) , then the protein will not be eluted with DTT and will not be identified as a target for sulfhydration . In addition , if a protein contains sulfhydrated cysteines with low reactivity for the probe , this protein will not be captured and identified as an H2S-modified target . Due to those limitations and in order to extend the capability of the BTA approach , we introduced a proteolytic digestion step before applying the avidin column step . This added step provided not only the isolation of cysteine-containing peptides with persulfide bonds , but also increased the identification of sulfhydrated proteins . The eluted peptides were then sequenced and identified by LC-MS/MS analysis , thereby identifying the modified cysteines on proteins . By using the modified BTA technique , we have identified over ~ 1000 novel sulfhydrated cysteines in MIN6 cells treated with Tg , corresponding to about 820 proteins ( Figure 2—figure supplement 9 and Figure 2—source data 1 ) , including GAPDH , wherein two cysteine-containing peptides were captured: Cys150 and Cys245 . Remarkably , the Cys150 active-site peptide was highly enriched as compared to the C-terminal of Cys245 , supporting prior mutagenesis studies that have shown Cys150 as the primary H2S-modified site on GAPDH , both in vitro and in vivo . One of novel targets including phosphoglycerate dehydrogenase ( PHGDH ) was also confirmed by Western blot analysis ( Figure 2—figure supplement 10 ) . CTH Knockdown mediated by shRNA decreased GAPDH and PHGDH sulfhydration levels , confirming that those proteins identified by the BTA method are bona-fide targets for sulfhydration in vivo ( Figure 2—figure supplement 11 ) . Sulfhydrated peptides do not reveal a consensus protein abundance and sulfhydrome ( Figure 3—figure supplement 6 , 7and Figure 3—source data 2 ) , supporting that lower concentrations of the NM-biotin labeling reveal reactivity of cysteine residues rather than protein abundance . In order to obtain genome-wide stress-induced changes in the sulfhydrome and individual cysteine residues within these proteins , we devised a modified BTA protocol ( Figure 3A ) by introducing a stable isotope-labeling step after the DTT elution step . This protocol uses ( 1 ) digestion of biotinylated cell extracts with trypsin before avidin capture , ( 2 ) labeling of free Cys-SH groups on peptides in the column eluent by mass-difference cysteine-alkylating reagents with either NEM-H5 ( Light ) or NEM-D5 ( Heavy ) , and ( 3 ) quantification by LC-MS/MS analysis of H/L ratios of the individual pair-labeled cysteines in the identified peptides based on a mass-difference labeling . In addition to quantifying changes in protein sulfhydration , this modified BTA approach allows the detection of additional proteins ( Figure 2—figure supplement 11 and Figure 2—source data 1 ) . We used as an experimental system the ATF4-expressing MIN6 cells to profile quantitatively the sulfhydrated proteome . As shown in Figure 2A , ATF4-mediated signaling triggers the cellular response , which leads to increased protein sulfhydration . Using the experimental plan in Figure 3A , we identified over 834 cysteine-containing peptides ( Figure 3B and Figure 3—source data 1 ) . Of these peptides , 771 exhibited pair-labeling with an overall average H/L ratio 2 . 6 , and 348 peptides ( 45% ) displayed high ratios ( H/L>2 ) . These findings confirmed that ATF4 drives global changes in protein sulfhydration in MIN6 cells . 10 . 7554/eLife . 10067 . 022Figure 3 . Quantitative and pathway analysis of sulfhydrated peptides in MIN6 cells overexpressing the transcription factor ATF4 . ( A ) Schematic representation of the BTA experimental approach combined with alkylation of free SH groups by either the stable isotope-labeled ( D5 , heavy ) or normal ( H5 , light ) maleimide . The relative levels of H5 and D5 labeled peptides were quantified by the LC-MS/MS method . ( B ) Distribution of peptides containing sulfhydrated cysteine residues relative to their H/L ratios as determined by the BTA analysis of cell extracts isolated from MIN6 cells overexpressing ATF4 ( indicated in A ) . Values of H/L ratios are plotted against the number of identified peptides . The red line marks the H/L ratio >2 , consisting of cysteine-containing peptides in proteins that exhibited higher reactivity with H2S under ATF4 overexpression . The black dots show redox sensitive cysteine peptides , which are common between the ones found in the RedoxDB database , and by the BTA assay . ( C ) Pie chart illustrating the percentage of cysteine-containing peptides ( from A ) that belong to known functional domains of proteins in the Uniprot database . ( D ) Heat map of H/L values obtained from experimental data in ( A ) , illustrating the profound differences in the reactivity with H2S of cysteine residues in different proteins . ( E ) Gene ontology biological pathways for peptides with H/L ratio >2 . H/L values were obtained from the experimental data in ( A ) . BTA: biotin thiol assay . DOI: http://dx . doi . org/10 . 7554/eLife . 10067 . 02210 . 7554/eLife . 10067 . 023Figure 3—source data 1Sulfhydrated proteins from ATF4 overexpressed MIN6 cells . DOI: http://dx . doi . org/10 . 7554/eLife . 10067 . 02310 . 7554/eLife . 10067 . 024Figure 3—source data 2 . Relative abundance of proteins in MIN6 cells treated with Tg for 18 hr . DOI: http://dx . doi . org/10 . 7554/eLife . 10067 . 02410 . 7554/eLife . 10067 . 025Figure 3—figure supplement 1 . Sulfhydrated cysteine-containing peptides are enriched in functional residues . Distribution of peptides containing sulfhydrated cysteines relative to their H/L ratios with functional annotations from the Uniprot database where active sites ( ACT_SITE ) , binding sites ( BINDING ) , calcium binding sites ( CA_BIND ) , disulfide bonds ( DISULFID ) , DNA binding sites ( DNA_BIND ) , specific domains ( DOMAIN ) , metal binding sites ( METAL ) , modified residues ( MOD_RES ) , motifs with biological activity ( MOTIF ) , nucleotide binding sites ( NP_BIND ) and zinc finger domains ( ZN_FING ) are shown as black dots . DOI: http://dx . doi . org/10 . 7554/eLife . 10067 . 02510 . 7554/eLife . 10067 . 026Figure 3—figure supplement 2 . Sulfhydrated proteins are potential targets for nitrosylation . A total of 827 H2S-modified and pair-labeled peptides were scored for their nitrosylation potential via the use of the GPS-SNO algorithm . We identified 303 peptides with predicted S-nitrosylation sites . Among them , only five peptides ( red dots ) had an H/L ratio below 2 ( 1 . 5% of the scored peptides for S-nitrosylation ) . DOI: http://dx . doi . org/10 . 7554/eLife . 10067 . 02610 . 7554/eLife . 10067 . 027Figure 3—figure supplement 3 . Quantitative profiling of proteins containing cysteines with different reactivity to H2S from ATF4 overexpressing MIN6 cells . MS1 profiles for multiple cysteine-containing peptides from GDI1 , PHGDH , VCP , and LAP3 , only one of which exhibits the highest reactivity with H2S , as shown by the H/L ratios . DOI: http://dx . doi . org/10 . 7554/eLife . 10067 . 02710 . 7554/eLife . 10067 . 028Figure 3—figure supplement 4 . ( A , B ) Sulfhydrated peptides do not reveal a consensus sequence motif , ( C , D ) but the modified cysteine residue is significantly accessible and preferentially positioned at the N-terminal of alpha helix . ( A ) A total of 739 pair-labeled peptide sequences were used to show that no primary sequence motif could be detected using the pLogo program . ( B ) A total of 333 peptides with H/L ratios over 2-fold showed significant enrichment for Arg residue found next to the modified cysteine . ( C , D ) the surface accessibility and secondary structure of the modified cysteine residues in peptides were annotated by comparing those peptides with H/L ratios greater than twofold and proteins with known structures in the PDB database ( 172 protein structures were employed ) . In the secondary structure motif , H= alpha helix; G=310-helix; E=beta sheet; T= helix turn; S=bend ( high curvature ) . DOI: http://dx . doi . org/10 . 7554/eLife . 10067 . 02810 . 7554/eLife . 10067 . 029Figure 3—figure supplement 5 . Gene ontology biological pathways enriched among all pair-labeled peptides in ATF4 overexpressed MIN6 cells . Seven hundred and thirty-nine pair-labeled cysteine-containing peptides were subjected to a pathway analysis carried by the Davide and IPA programs ( A , B ) . ( C , D ) the proteins associated with the H/L greater than twofold were selected and subjected for both pathway annotation calculations . Top five pathways are shown with their statistical significance ( Bonferroni correction for N=17 tests ) . DOI: http://dx . doi . org/10 . 7554/eLife . 10067 . 02910 . 7554/eLife . 10067 . 030Figure 3—figure supplement 6 . MS analysis of the full proteome from MIN6 cells treated with Tg for 18 hr . Cell extracts were resolved by reducing SDS–PAGE and stained with Coomassie blue . The entire gel lane was cut into 11 fractions for in-gel digestions . The peptides from each fraction were combined as indicated and submitted for LC-MS/MS analysis . A total of 2244 proteins were quantified by label-free , semi-quantitative MS approaches . DOI: http://dx . doi . org/10 . 7554/eLife . 10067 . 03010 . 7554/eLife . 10067 . 031Figure 3—figure supplement 7 . Protein sulfhydration does not correlate with their protein abundance . The relative abundance of the proteins in the full proteome was determined based on the peptide spectral counts , which were corrected by normalizing to both the total number of spectra and the length of the protein , and this value was expressed as normal spectra abundance factor ( NSAF ) . A comparison of the full protein abundance and sulfhydrome from Tg-treated ( A ) , or ATF4 overexpressing MIN6 ( B ) . ( C ) A comparison of sulfhydrated protein datasets for the full proteome reveals that a large fraction of medium abundant proteins are mainly targeted by H2S . ( D ) A comparison of sulfhydrated proteins with H/L ratios greater than twofold and the full proteome . DOI: http://dx . doi . org/10 . 7554/eLife . 10067 . 031 Sulfhydrated cysteine residues in proteins may contribute to their biological activities , especially when these modified cysteine residues reside within functional domains . We thus queried the Uniprot database to retrieve functional annotations for the aforementioned 827 peptides . The analysis revealed that 28% of the peptides were localized to protein regions whose structural and functional properties are known ( Figure 3C , Figure 3—figure supplement 1 ) . An additional 18% were found within functional regions of proteins with cysteine residues of unknown significance . Our finding of sulfhydration of specific cysteine residues within functional domains of proteins suggests that the cysteine modification influences the activity of proteins . In contrast , 4 . 2% of the peptides in question contained cysteine residues in experimentally proven active sites of enzymes or cysteine residues involved in disulfide bond formation ( Figure 3C ) . This percentage of active cysteine-containing peptides from the BTA assay is significantly larger than the 0 . 2% of all cysteines in the Uniprot database that have been assigned to experimentally characterized active cysteines ( Weerapana et al . , 2010 ) . Finally , 2 . 4% of the annotated peptides were found in known nucleotide binding domains of proteins , suggesting the potential for regulation of gene expression by H2S cysteine-modified proteins . None of these proteins have been reported as targets of sulfhydration . Although not found among the 827 peptides in this study , the NF-κB RelA transcription factor has been shown to be sulfhydrated in the DNA-binding domain , resulting in increased DNA-binding activity ( Sen et al . , 2012 ) . We also found by querying the redox modification databases ( RedoxDB and GPS-SNO ) that 11% of the 834 peptides from the RedoxDB and 36% from the GPS-SNO corresponded to cysteine residues that are known to be modified by nitrosylation or glutathionylation ( Figure 3B , Figure 3—figure supplement 2 ) . Taken together , these data indicate that protein sulfhydration not only influences the catalytic activity of enzymes like GAPDH , but also can potentially regulate a broad range of biological processes . We noted a wide range of H/L ratios in the identified peptides , reflecting a large difference in the reactivity of individual cysteines to sulfhydration in MIN6 cells that overexpress ATF4 ( Figure 3B–D ) . In addition , labeled cysteines on the same protein exhibited remarkably different ratios ( Figure 3—figure supplement 3 ) . For example , the protein , PHGDH , was labeled on two cysteines , Cys281 had an H/L ratio of 5 . 8 , whereas Cys254 displayed a ratio of 17 . To identify a sulfhydration motif in proteins , all cysteine-containing peptides ( Figure 3B ) were analysed by the pLogo motif analyzer program . This analysis did not reveal conserved residues surrounding the modified cysteine site ( Figure 3—figure supplement 4A ) , a result consistent with other enzyme-independent oxidative modifications of cysteines ( Weerapana et al . , 2010 ) ( Marino and Gladyshev , 2011 ) . However , when we selected only the peptides with high H/L ratios ( >2 ) , and searched for a sequence motif at the cysteine modification sites ( Figure 3—figure supplement 4B ) , an Arg residue was significantly enriched at the +1 position of the modified cysteine ( Figure 3—figure supplement 4B ) . Additionally , structure motif and surface accessibility analysis revealed that the modified cysteine is highly accessible and positioned at the N-terminal alpha-helices ( Figure 3—figure supplement 4C–D ) . This is consistent with previous reports suggesting that a reactive cysteine thiolate anion is stabilized by interaction with alpha-helix dipoles ( Kortemme and Creighton , 1995; Weerapana et al . , 2010 ) . Bioinformatics clustering with two pathway annotation programs DAVID and Ingenuity Pathway Analysis ( IPA ) revealed an enrichment of sulfhydrated proteins in glycolysis and mitochondrial oxidative metabolism ( Figure 3E , Figure 3—figure supplement 5 ) . Data from this analysis are in agreement with increased activity of the glycolytic enzyme , GAPDH , by sulfhydration , and prompted the question as to whether or not glycolytic flux is regulated by the H2S-dependent modification of enzymes involved in intermediary metabolism . Also in agreement with this hypothesis , we found that ATF4-overexpressing MIN6 cells had higher glycolytic rates as evaluated with a Seahorse analyzer ( Figure 4—figure supplement 1 ) . Furthermore , the activity of the rate-limiting glycolytic enzyme , pyruvate kinase 2 ( PKM2 ) , also was increased in ATF4 overexpressing MIN6 cells , in a manner dependent on CTH activity ( Figure 4—figure supplement 2 ) . We therefore returned to the induction of ER stress by Tg-treatment of MIN6 cells and evaluated glycolytic flux rates in the absence or presence of PAG . The advantage of using PAG instead of genetic manipulation such as gene knockdown is that the inhibitor could be added at the same time as labeled glucose , thus allowing assessment of the inhibitor’s effect on glycolytic flux rates . We directly measured changes in metabolic flux by utilizing stable isotope label incorporation and mass isotopomer analyses . MIN6 cells were treated with Tg for 18 hr; the growth media was changed to ( D-glucose-13C6 ) media in the presence or absence of PAG during the last 3 . 5 hr of treatment . Tg-treatment significantly augmented glycolytic flux as determined by the increase in the glycolytic intermediate , 3 phosphoglyceric acid ( 3PG ) , as well as lactate and alanine ( Figure 4A–C , Supplementary file 1 ) . The flux was consistent with an increase in the relative concentrations of 3PG and alanine ( Figure 4B , Supplementary file 1 ) . However , lactate levels were decreased significantly despite the increased flux . A decrease in cellular lactate levels supports the idea that there is high consumption of pyruvate by the mitochondria to generate oxaloacetate ( OAA ) . In contrast to the increased glycolytic rates , flux to tricarboxylic acid cycle ( TCA ) intermediates was significantly reduced by Tg-treatment as evidenced by the low 13C label incorporation of both acetyl-CoA and OAA moieties of citrate , fumarate , and malate ( Figure 4A–B , Supplementary file 1 ) . When the cells were exposed to PAG along with Tg-treatment , the increase in glycolytic flux was prevented , as shown by the decrease in 13C-labeling of 3PG , lactate , and alanine . In contrast , TCA cycle flux was restored , as exemplified by the increase in 13C-labeling of both OAA and acetyl-CoA moieties of citrate , succinate , fumarate , and malate , and a decrease in their concentrations , suggesting utilization . 13C-labeling of aspartate and glutamine also increased significantly , indicating increased cataplerosis of TCA cycle intermediates . Moreover , we determined the activity of PKM2 in MIN6 cells treated with Tg in the presence or absence of PAG . Tg-treatment increased PKM2 activity , but PAG addition inhibited the increase , without affecting PKM2 protein levels ( Figure 4—figure supplement 3 ) . These data suggest that ER stress , via H2S-mediated signaling , promotes glycolysis and decreases mitochondrial oxidative metabolism . 10 . 7554/eLife . 10067 . 032Figure 4 . H2S synthesis during ER stress modulates metabolism in MIN6 cells . ( A ) Measurement of 13C-glucose flux in metabolites , expressed as the molar percent enrichment [ ( ratio of labeled/sum ( labeled + unlabeled ) x 100% ) ] , in MIN6 cells treated with Tg for 18 hr or after addition of PAG for the last 3 . 5 hr of Tg treatment . [U-13C]-glucose replaced glucose in the media for the last 3 . 5 hr of treatments . ( B ) Evaluation of the concentration of metabolites and amino acids in the same samples described in ( A ) . All quantifications are presented as mean ± S . E . M . of technical duplicates and are represented four independent experiments . ( C ) Schematic representation of the major findings on metabolic flux from glucose during chronic ER stress . Chronic ER stress increased glycolytic flux and decreased forward TCA flux . Inhibition of CTH by PAG reversed the observed changes in glucose flux during ER stress . CTH: γ-cystathionase; ER: endoplasmic reticulum; TCA: tricarboxylic acid cycle . DOI: http://dx . doi . org/10 . 7554/eLife . 10067 . 03210 . 7554/eLife . 10067 . 033Figure 4—figure supplement 1 . ATF4 -overexpressing MIN6 cells exhibit significantly high glycolytic rates . The rates of extracellular acidification ( ECAR ) were determined by using the Seahorse analyzer from ATF4 expressed MIN6 cells , or GFP as the control after 48h of adenovirus infection . ECAR was normalized to cell numbers . The results are shown as a mean of four independent determinations . DOI: http://dx . doi . org/10 . 7554/eLife . 10067 . 03310 . 7554/eLife . 10067 . 034Figure 4—figure supplement 2 . Pharmacological inhibition of CTH activity represses PKM2 activation in ATF4 overexpressed MIN6 cells . Determination of the effect of PAG on the activity of PKM2 in MIN6 cells infected with either GFP as the control , or ATF4 adenovirus for 48 hr . The activities are shown as a mean of four independent determinations . CTH: γ-cystathionase; PAG: propargylglycine . DOI: http://dx . doi . org/10 . 7554/eLife . 10067 . 03410 . 7554/eLife . 10067 . 035Figure 4—figure supplement 3 . PKM2 activation dependents on CTH activity during ER stress in MIN6 cells . ( A ) ER stress does not affect PKM2 protein levels . MIN6 cells were incubated with Tg at the indicated times . The protein expression was evaluated by Western blot analysis . ( B ) Western blot analysis of the expression of PKM2 in mouse islets . ( C ) Time-dependent increase of PKM2 activity in MIN6 cells treated with Tg at the indicated times . ( D ) Determination of the effect of PAG on the activity of PKM2 in MIN6 cells treated with Tg in the presence or absence of PAG ( top ) and on protein levels evaluated by Western blot analysis ( bottom ) . For panels C and D the results are shown as a mean of four independent determinations . CTH: γ-cystathionase; ER: Endoplasmic reticulum; PAG: Propargylglycine . DOI: http://dx . doi . org/10 . 7554/eLife . 10067 . 035
In summary , sulfhydration of specific cysteines in proteins is a key function of H2S ( Kabil and Banerjee , 2010; Paul and Snyder , 2012; Szabo et al . , 2013 ) . Thus , the development of tools that can quantitatively measure genome-wide protein sulfhydration in physiological or pathological conditions is of central importance . However , a significant challenge in studies of the biological significance of protein sulfhydration is the lack of an approach to selectively detect sulfhydrated cysteines from other modifications ( disulfide bonds , glutathionylated thiols and sulfienic acids ) in complex biological samples . In this study , we introduced the BTA approach that allowed the quantitative assessment of changes in the sulfhydration of specific cysteines in the proteome and in individual proteins . BTA is superior to other reported methodologies that aimed to profile cysteine modifications , such as the most commonly used , a modified biotin switch technique ( BST ) . BST was originally designed to study protein nitrosylation and postulated to differentiate free thiols and persulfides ( Mustafa et al . , 2009 ) . A key advantage of BTA over the existing methodologies is that the experimental approach has steps to avoid false-positive and negative results , as target proteins for sulfhydration . BST is commonly generating such false targets for cysteine modifications ( Forrester et al . , 2009; Sen et al . , 2012 ) . Using mutiple validations , our data support the specificity and reliability of the BTA assay for analysis of protein sulfhydration both in vitro and in vivo . With this approach , we found that ATF4 is the master regulator of protein sulfhydration in pancreatic β cells during ER stress , by means of its function as a transcription factor . A large number of protein targets have been discovered to undergo sulfhydration in β cells by the BTA approach . Almost 1000 sulfhydrated cysteine-containing peptides were present in the cells under the chronic ER stress condition of treatment with Tg for 18 hr . Combined with the isotopic-labeling strategy , almost 820 peptides on more than 500 proteins were quantified in the cells overexpressing ATF4 . These data show the potential of the BTA method for further systematic studies of biological events . To our knowledge , the current dataset encompasses most known sulfhydrated cysteine residues in proteins in any organism . Our bioinformatics analyses revealed sulfhydrated cysteine residues located on a variety of structure–function domains , suggesting the possibility of regulatory mechanism ( s ) mediated by protein sulfhydration . Structure and sequence analysis revealed consensus motifs that favor sulfhydration; an arginine residue and alpha-helix dipoles are both contributing to stabilize sulfhydrated cysteine thiolates in the local environment . Pathway analyses showed that H2S-mediated sulfhydration of cysteine residues is the part of the ISR with the highest enrichment in proteins involved in energy metabolism . The metabolic flux revealed that H2S promotes aerobic glycolysis associated with decreased oxidative phosphorylation in the mitochondria during ER stress in β cells . The TCA cycle revolves by the action of the respiratory chain that requires oxygen to operate . In response to ER stress , mitochondrial function and cellular respiration are down-regulated to limit oxygen demand and to sustain mitochondria . When ATP production from the TCA cycle becomes limited and glycolytic flux increases , there is a risk of accumulation of lactate from pyruvate . One way to escape accumulation of lactate is the mitochondrial conversion of pyruvate to oxalacetic acid ( OAA ) by pyruvate carboxylase . This latter enzyme was found to be sulfhydrated , consistent with the notion that sulfhydration is linked to metabolic reprogramming toward glycolysis . The switch of energy production from mitochondria to glycolysis is known as a signature of hypoxic conditions . This metabolic switch has also been observed in many cancer cells characterized as the Warburg effect , which contributes to tumor growth . The Warburg effect provides advantages to cancer cell survival via the rapid ATP production through glycolysis , as well as the increasd conversion of glucose into anabolic biomolecules ( amino acid , nucleic acid , and lipid biosynthesis ) and reducing power ( NADPH ) for regeneration of antioxidants . This metabolic response of tumor cells contributes to tumor growth and metastasis ( Vander Heiden et al . , 2009 ) . By analogy , the aerobic glycolysis trigged by increased H2S production could give β cells the capability to acquire ATP and nutrients to adapt their cellular metabolism toward maintaining ATP levels in the ER ( Vishnu et al . , 2014 ) , increasing synthesis of glycerolphospholipids , glycoproteins and protein ( Krokowski et al . , 2013 ) , all important components of the ISR . Similar to hypoxic conditions , a phenotype associated with most tumors , the decreased mitochondria function in β cells during ER stress , can also be viewed as an adaptive response by limiting mitochondria ROS and mitochondria-mediated apoptosis . We therefore view that the H2S-mediated increase in glycolysis is an adaptive mechanism for survival of β cells to chronic ER stress , along with the improved ER function and insulin production and folding , both critical factors controlling hyperglycemia in diabetes . Future work should determine which are the key proteins targeted by H2S and thus contributing to metabolic reprogramming of β cells , and if and how insulin synthesis and secretion is affected by sulfhydration of these proteins during ER stress . Abnormal H2S metabolism has been reported to occur in various diseases , mostly through the deregulation of gene expression encoding for H2S-generating enzymes ( Wallace and Wang , 2015 ) . An increase of their levels by stimulants is expected to have similar effects on sulfhydration of proteins like the ATF4-induced CTH under conditions of ER stress . It is the levels of H2S under oxidative conditions that influence cellular functions . In the present study , ER stress in β cells induced elevated CTH levels , whereas CBS was unaffected . The deregulated oxidative modification at cysteine residues by H2S may be a major contributing factor to disease development . In this case , it would provide a rationale for the design of therapeutic agents that would modulate the activity of the involved enzymes .
Experimental protocols were approved by the Case Western Reserve University Institutional Animal Care and Use Committee . C57BL/6J and C57BL/6-Ins2+/Akita mice were used for experiments . Mice from the Jackson Laboratory were bred at the animal facilities at Case Western Reserve University and were fed standard lab chow ( LabDiet ) . Mice were housed under 12:12 hr light/dark cycle with free access to food and water at 23°C . Mouse pancreatic islets were isolated as described before ( Krokowski et al . , 2013 ) . Islets from 6 weeks old male C57BL/6-Ins2+/Akita ( n=6 ) and age and sex matched wild type ( n=6 ) were cultured for 2 hr in RPMI 1640 media supplemented with 10% FBS and 5 mM glucose before RNA isolation . For Tg treatment ( 1 μM ) , islets from wild-type mice ( n=6 ) were combined and cultured in RPMI 1640 medium supplemented with 10% FBS in atmosphere of 5% CO2 at 37°C for 24 hr . From each group 150–200 islets were manually picked and used for RNA isolation . Islets were treated with QIAshredder ( Qiagen GmbH , D-40724 Hiden , Germany ) , and RNA was purified using the RNeasy Plus Micro kit ( Qiagen GmbH , D-40724 Hiden , Germany ) . Institutional review board approval for research use of isolated human islets was obtained from the University of Michigan . Human islets were isolated from previously healthy , nondiabetic organ donors by the University of Chicago Transplant Center . The islets were divided into two groups , incubated in CMRL medium containing either 5 . 5 mM glucose with or without Tg ( 1 μM ) , for 24 hr . The islets were frozen at -80°C before analysis . RNA was isolated as described above from 200 islets/treatment . RNA was isolated from mouse MIN6 cells using TRIzol ( Invitrogen ) . cDNA was synthesized from total RNA isolated from islets or MIN6 cells using the SuperScript III First-Strand Synthesis Super Mix ( Invitrogen ) , and the abundance of cDNA isolated from each sample was quantified by qPCR using the VeriQuest SYBR Green qPCR Master Mix ( Affymetrix ) with the StepOnePlus Real-Time PCR System ( Applied Biosystems ) . MIN6 cells were cultured in high glucose DMEM supplemented with 10% FBS , 2 mM l-glutamine , 1 mM sodium pyruvate , 55 μM β-mercaptoethanol , 100 units/ml penicillin , and 100 mg/ml streptomycin at 37°C in atmosphere of 5% CO2 . β-mercaptoethanol was removed from the media 12 hr before experimentation . Rat INS1 cells were cultured in RPMI 1640 supplemented with 11 mM glucose , 10% heat inactive FBS , 2 mM l-glutamine , 1 mM sodium pyruvate , 100 units/ml penicillin , and 100 μg/ml streptomycin at 37°C in atmosphere of 5% CO2 . Tg ( Sigma-Aldrich ) was used at 400 nM and the CTH inhibitor - DL-propargylglycine ( PAG , Sigma Aldrich ) at 3 mM . Lentiviral particles were prepared in HEK293T as described before ( Saikia et al . , 2014 ) . Lentiviral vector expressing shRNA against ATF4 were obtained from Sigma-Aldrich ( TRCN0000301646 ) . Adenovirus mediated shRNA against mouse CTH ( shRNA sequence: CCGGCCATTACGATTACCCATCTTTCTCGAGAAAGATGGGTAATCGTAATGGTTTTTG ) was purchased from Vector Biolabs . MIN6 cells were infected in the presence of 10 μg/ml polybrene and selection with 2 . 5 μg/ml puromycin ( Life Technologies ) was conducted 24 hr post-infection for 5 days . Adenovirus particles for expression of β galactosidase ( β-Gal ) , GFP or mouse ATF4 protein were prepared in HEK293 cells and were used for infection as described before ( Guan et al . , 2014 ) . Human GST-tagged wild type or C150S GAPDH mutant ( Hara et al . , 2005 ) was expressed in the E . coli BL21 strain . Protein expression was induced by addition of IPTG ( 100 μM ) . When bacterial cultures reached OD600 of 0 . 8 at 37°C , IPTG was added for 4 hr incubation before lysis in a buffer containing 50 mM Tris–HCl ( pH 7 . 5 ) and 1 mM EDTA . Lysates were centrifuged and applied on a buffer-equilibrated GST-sepharose affinity spin column ( Pierce ) . After extensive washes to remove unbound protein , recombinant GAPDH was released by digestion with thrombin protease ( Sigma ) . The protein purity was determined on SDS–PAGE gels stained by Coomassie blue . The specific activity of GAPDH was determined as described before ( Hara et al . , 2005 ) . Recombinant protein ( 50 nM ) was used for the in vitro activity assays . To test the GAPDH activity in cell lysates , MIN6 cells were harvested in RIPA buffer ( 150 mM NaCl , 1 mM EDTA , 1% Triton X-100 , 0 . 5% deoxycholic acid , 50 mM Tris–HCl , pH 7 . 5 ) , sonicated on ice and centrifuged at 4°C . One to twenty microrams of protein lysate was used for the activity assays . The reaction mixture contained 100 mM Tris–HCl ( pH 7 . 5 ) , 5 mM MgCl2 , 3 mM 3-phosphoglycerate , 5 units/ml of Saccharomyces cerevisiae 3-phosphoglycerate kinase ( PGK , Sigma-Aldrich ) , 2 mM ATP and 0 . 25 mM NADH . Reactions were conducted in 200 μl volume at 25°C and monitored spectrophotometrically at 340 nm for 3 min using a M3 microplate reader ( Molecular Devices ) . PKM activity was tested in the cell extracts as described ( Anastasiou et al . , 2011 ) . MIN6 cells were collected by trypsinization and the cell pellets were washed twice with cold PBS . Cells were resuspended in RIPA buffer , sonicated on ice , and subjected to a quick centrifugation at 4°C . Reaction mixtures contained 50 mM Tris–HCl ( pH 7 . 5 ) , 100 mM KCl , 5 mM MgCl2 , 0 . 5 mM ADP , 0 . 2 mM NADH , 8 units of lactate dehydrogenase ( Sigma-Aldrich ) , and 1–10 μg of cell lysate . The enzymatic reaction was initiated by the addition of PEP ( phosphoenolpyruvic acid , 0 . 5 mM ) as the substrate . The oxidation of NADH was monitored at 340 nm for 3 min using a M3 microplate reader ( Molecular Devices ) . The activity of the amino acid transporter was tested as described before ( Krokowski et al . , 2013 ) . Uptake of Glu was tested in the absence of Na+ ions ( EBSS solution , NaCl replaced with choline chloride ) , with 100 μM Glu and 4 μCi/ml of 3H-Glu ( Parkin Elmer ) for 3 min at 37°C . MIN6 cells were washed twice with cold PBS then amino acids were extracted with ethanol and radioactivity was counted . The specific activity was normalized to protein content that was determined by the Lowery assay . MIN6 cells ( 6x104 cell per well ) were seeded into 96-well plates . After 48 hr the growth media was removed , the total glutathione contents and GSH/GSSG ratios were determined with the GSH/GSSG-Glo Tm assay from Promega . Total intracellular levels of ROS were quantified using dichlorofluorescein diacetate ( CM-H2DCFDA; 10 μM ) . MIN6 cells were seeded into 96-well plates . After 48 hr , cells were washed with warm PBS and incubated with the dye in phenol-red free DMEM without FBS . After 1 hr , the cells were washed with PBS to remove the dye and placed in phenol-red free DMEM . CM-H2DCFDA fluorescence was measured at excitation/emission wavelengths of 495/517 nm . Cells not exposed to the probe were used to test the background fluorescence . After subtraction of background fluorescence results were normalized to protein content determined by the BCA assay . Frozen cell pellets were lysed in 100 mM HEPES ( pH 7 . 4 ) , to obtain a lysate concentration of 100 mg/ml . H2S production was measured as described previously ( Kabil et al . , 2011 ) . Briefly , reactions containing cell lysate ( 200 μl ) , 10 mM cysteine and 100 mM HEPES ( pH 7 . 4 ) were prepared in 20-ml polypropylene syringes in a total reaction volume of 400 μl . Reactions were started with the addition of cysteine . Syringes were sealed and the headspace was flushed with nitrogen five times by using a three-way stopcock , and left in nitrogen in a final total volume ( aqueous + gas ) of 20 ml . Syringes were placed at 37°C in a shaker incubator ( 75 rpm ) for 20 min . Control reactions without cell lysates were prepared in parallel . Aliquots of 0 . 2 ml from the gas phase were collected through a septum attached to the stopcock , and injected in an HP 6890 gas chromatograph ( GC ) ( Hewlett Packard ) equipped with a DB-1 column ( 30 m×0 . 53 mm×1 . 0 μm ) . Flow rate of the carrier gas ( helium ) was 1 ml/min , and the temperature gradient ranged from 30°C to 110°C over a 20-min period . H2S was detected by a 355 sulfur chemiluminescence detector ( Agilent ) attached to the GC . H2S standard gas ( Cryogenic Gases , Detroit , MI ) with a stock concentration of 40 ppm ( 1 . 785 μM ) in nitrogen was used to generate a standard curve . The amount of H2S in the injected volume was calculated from the peak areas by using the calibration coefficient obtained from the standard curve . Ionized H2S concentration in the liquid phase was calculated for the pH of the reaction mixture ( pH 7 . 4 ) by using a pKa value of 6 . 8 for ionization of H2S . The resulting H2S concentration in the total reaction volume was then used to obtain the specific activity expressed as nmol H2S per mg protein per min . In order to detect and identify sulfhydrated proteins from MIN6 cells , cells were lysed with the RIPA buffer ( 150 mM NaCl , 1 mM EDTA , 0 . 5% Triton X-100 , 0 . 5% deoxycholic acid , and 100 mM Tris–HCl ( pH 7 . 5 ) containing protease and phosphatase inhibitor from Roche . Cells were sonicated on ice , lysates were clarified by centrifugation at 4°C and the protein concentrations were determined by the BCA assay ( BioRad ) . Equal amount ( 4 mg ) of proteins was incubated with 100 μM NM-biotin ( Pierce ) for 30 min with occasionally gentle mixing at room temperature and subsequently precipitated by cold acetone . After centrifugation , pellets of precipitated proteins were washed with 70% cold acetone twice , and then suspended in buffer ( 0 . 1% SDS , 150 mM NaCl , 1 mM EDTA and 0 . 5% Triton X-100 , 50 mM Tris–HCl , pH 7 . 5 ) mixed with Streptavidin-agarose resin ( Thermo Scientific ) and kept rotating overnight at 4°C . The beads were washed five times with wash buffer 1 ( 0 . 5% Triton x-100 , 150 mM NaCl , 50 mM Tris–HCl , pH 7 . 5 ) followed by five washes with wash buffer 2 ( 0 . 5% Triton X-100 , 600 mM NaCl , 50 mM Tris–HCl , pH 7 . 5 ) . Resin with bound proteins was incubated with 500 μl of the elution buffer with or without 20 mM DTT for 30 min at 25°C . Eluted proteins were concentrated to a final volume of 25–40 μl with utilization of Amicon Ultracel 10K ( Millipore ) and used for gel electrophoresis followed by either western blot or MS analysis . The assay was previously described ( Sen et al . , 2012 ) and modified in order to adapt to our experimental needs . Purified recombinant GAPDH was next treated with either 50 μM NaCl as control , 50 μM NaHS or 50 μM H2O2 for 45 min at room temperature . After desalting through a spin column ( Pierce ) , the samples were incubated with 1 μM red malemide probe ( Alexa Fluor 680 C2 Maleimide , Molecular Probes ) for 20 min at room temperature . After the incubation , these samples were treated with or without 10 mM β-mercaptoethanol and the reaction was stopped by the addition of 100 mM iodoacetamide . Samples were suspended in sample buffer for non-reducing gel electrophoresis . After electrophoretic separation , the gel was scanned with the Li-COR Odyssey system . The intensity of red fluorescence of GAPDH was quantified with the Odyssey system software . Subsequently proteins from the gel were transferred on a PVDF membrane and subjected to Western blot analysis for GAPDH . In order to identify H2S-modified cysteine-containing peptides from cell lysates , proteins were extracted and biotinylated as described above . Biotinylated proteins were precipitated with ice cold acetone , resuspended in denaturation buffer ( 30 mM Tris–HCl ( pH 7 . 5 ) , 8 M urea and 1 mM MgSO4 ) as described ( Morisse et al . , 2014 ) , diluted with 10 volumes of the buffer ( 30 mM Tris–HCl ( pH 7 . 5 ) , 1 mM EDTA and 200 mM NaCl ) and incubated with modified porcine trypsin ( Promega ) with occasionally mixing for 18 hr at 30°C . The ratio of the enzyme to substrate was 1:80 ( w/w ) . After digestion , trypsin was inactivated by incubation at 95°C for 10 min then reactions were mixed with the streptavidin-agarose beads ( 500 μl ) and incubated at 4°C for 18 hr following extensive washes in the presence of 0 . 1% SDS as described above . Peptides were eluted with 20 mM ammonium bicarbonate supplemented with 10 mM DTT after 25 min incubation at room temperature . DTT was removed with utilization of a C-18 column ( Pierce ) . Peptides were eluted from the desalting column with acetonitrile , dried under vacuum and suspended in buffer ( 30 mM Tris–HCl ( pH 7 . 5 ) , 1mM EDTA and 150 mM NaCl ) . Free -SH groups were alkylated by NEM ( either deuterium or hydrogen containing ) at final concentrations of 40 mM . The alkylated peptides were concentrated with a C-18 column ( Pierce ) for LC-MS/MS analysis . LC-MS/MS analysis was performed on an LTQ-Orbitrap Elite mass spectrometer ( Thermo-Fisher ) coupled to an Ultimate 3000 high-performance liquid chromatography system . Protein digests were loaded onto a 75 µm desalting column packed with 2 cm of Acclaim PepMap C18 reverse phase resin ( Dionex ) . The peptides were then eluted onto a Dionex 15 cm x 75 µm id Acclaim Pepmap C18 , 2μm , 100 Å reversed- phase capillary chromatography column using a gradient of 2–80% buffer B in buffer A ( buffer A: 0 . 1% formic acid; buffer B: 5% water , 95% acetonitrile , 0 . 1% formic acid ) . The peptides were then eluted from the C18 column into the mass spectrometer at a flow rate of 300 nl/ min and the spray voltage was set to 1 . 9 kV . One full MS scan ( FTMS ) ( 300–2 , 000 MW ) was followed by 20 data dependent scans ( ITMS ) of the nth most intense ions with dynamic exclusion enabled . Peak lists were extracted from Xcaliber RAW files using Proteome Discoverer 1 . 4 . These peak lists were searched Sequest HT and Mascot ( 2 . 3 ) search engines . The data was searched against the mouse reference sequence database which contains 77 , 807 entries using a precursor ion tolerance of 10 ppm and a fragment ion tolerance of 0 . 6 Da . These searches included differential modification of +125 . 047679 and +130 . 079062 on cysteine to account for NEM and d5-NEM alkylation and +15 . 994915 Da to account for oxidation on methionine residues . Peptide identification was validated with the Percolator node on the basis of q-values which are estimated from target-decoy searches . The false discovery rate ( FDR ) for these searches was set to 1% at the peptide level . In addition , peptides were also required to be fully tryptic and have Xcorr scores > 1 . 5 ( +1 ) , 2 . 0 ( +2 ) , 2 . 25 ( +3 ) , and 2 . 5 ( +4 ) . Quantification of light/heavy ratios ( d5-NEM/NEM ) was performed using two algorithms of Proteome Discoverer , the event detector and Precursor Ions Quantifier . The event detector applied a 2 ppm mass variability and 0 . 2 min chromatographic window for the generation of extracted ion chromatograms . The Peptide ratio was calculated from the summed extracted ion chromatograms of all isotopes for the NEM and d5-NEM containing peptides . All missing ions were assigned a value equivalent to the minimum intensity , only unique peptides were quantified , and since this included quantitation at the peptide level , single channel was used . The H/L ratios of approximately 25% of the quantified peptides were manually validated . For functional annotation: Protein sequences from the FTP site of the Uniprot Protein database for mouse ( Proteome_ID/Tax_ID: UP000000589/10090 ) , rat ( Proteome_ID/Tax_ID: UP000002494/10116 ) , and human ( Proteome_ID/Tax_ID: UP000005640/9606 ) release current as of May 23 2015 . Sequence annotation in the feature fields ( ACT_SITE , BINDING , CA_BIND , DISULFID , DNA_BIND , DOMAIN , METAL , MOD_RES , MOTIF , NP_BIND , SITE , ZN_FING ) of the Uniprot entry was searched and any annotation corresponding to the labeled cysteine peptides was collected . For redox cysteine annotation: each peptide identified by MS , all exact matches in any of the RedoxDB databases on any oxidative modification cysteine sequences ( fasta or additional_fasta ) were collected . For motif search: The lager data set of putative modification cysteine sites and their vicinity sequences were submitted to the pLogo program ( www . plogo . uconn . edu , version v1 . 2 . 0 ) ( O'Shea et al . , 2013 ) to identify linear motif . For prediction of candidate peptides for nitrosylation: the peptide sequences with H/L ratio >2 were submitted for use in predicting nitrosylation sites under the medium threshold condition using the batch prediction tool of the GPS-SNO 1 . 0 software ( Xue et al . , 2010 ) . The predicted nitrosylation sites of sequences were extracted for further analysis . For determination of surface accessibility and secondary structural motif: we turned to DSSP's ( Kabsch and Sander , 1983; Touw et al . , 2015 ) annotations of the PDB ( Berman et al . , 2000 ) . We downloaded a total of 108355 DSSP-annotated PDB files from ( rsync . cmbi . ru . nl/dssp/ ) on Sep . 9th 2015 . Each peptide with H/L greater than 2-fold was aligned on all matching DSSP profiles , from which the 10-state structural context and accessibility were extracted . When an exact match is not found , then all matches with 1 mismatch are considered . The structural context of a peptide is defined as the context that reoccur most frequently among the hits , while the accessibility is the average across the hits , log2-normalized by the median of accessibility considering the amino acid type; positive log2 values means that the amino acid embedded in the 3D structure of the protein is more accessible that the mode ( median ) . For pathway annotation: the canonical pathways were scored based on the total sulfhydrated peptides and the peptides with H/L ratios greater than 2 by using DAVID ( www . david . ncifcrf . gov ) and Ingenuity Pathway Analysis ( IPA , www . qiagen . com/ingenuity ) programs . Statistical significant of pathways are calculated , and pathways are ranked by the p-values based on those tests . The tests measure the likelihood that the association between proteins measured in our experiments and a pathway is due to random chance . The smaller the p-value the less likely that the association is due to random chance . Top scoring pathways are presented . MIN6 cells were plated onto 10-cm plates in triplicates and cultured in the cell growth medium . After 48 hr treatment with Tg , metabolic labeling was performed . Cells were washed with warm PBS and incubated for 3 hr in the DMEM medium containing 10% heat inactive FBS , 2 mM glutamine , and 25 mM glucose consisting of a mixture of 12 . 5 mM d-glucose plus 12 . 5 mM of d-[U-13C] glucose . After incubation , cells were washed with PBS , followed by trypsinization . Cells were pelleted by centrifuging at 4°C for 5 min at 650 x g , and pellets were stored at -80°C until extraction of metabolites . Glucose isotopic enrichment was determined following ( van Dijk et al . , 2001 ) with modifications . Briefly , glucose was extracted by the addition of 500 µl of ice-cold ethanol to 50 µl of media . Samples were mixed and incubated on ice for 30 min . Samples were centrifuged at 4°C for 10 min at 14 , 000 rpm and ethanol was transferred to a GC/MS vial and evaporated to dryness in a SpeedVac evaporator . Glucose was converted to its pentaacetate derivative by the reaction with 150 µl of acetic anhydride in pyridine ( 2:1 , v/v ) at 60°C for 30 min . Samples were evaporated to dryness and glucose derivatives were reconstituted in 80 µl of ethyl acetate and transferred to GC/MS inserts . Samples were injected in duplicate and masses 331–337 , containing M0…M+5 isotopomers were monitored . Enrichment was determined as a ratio of ( M+5 ) / ( ΣM0-M5 ) . Metabolites were extracted following ( Yang et al . , 2008 ) with modifications . Briefly , cellular pellets in Eppendorf tubes were homogenized and frozen in 600 µl of Folch solution ( chloroform:methanol , 2:1 , vol . /vol . ) on dry ice . After addition of 0 . 4 volumes of ice-cold water , cells were homogenized again and let sit on ice for 30 min . Homogenates were centrifuged at 4°C for 10 min at 14 , 000 rpm . The upper methanol/water layer was removed to GC/MS vial . To the bottom chloroform layer , 120 µl of water and 200 µl of methanol were added and extraction steps from above were repeated . Combined methanol/water layers were evaporated to dryness in Speedvac evaporator at 4°C . Metabolites were derivatized using two-step derivatization . First , keto- and aldehyde groups were protected by the reaction with MOX ( methoxylamine-HCl in pyridine , 1:2 ) overnight at room temperature . Then excess derivatizing agent was evaporated and dry residue was converted to MOX-TMS ( trimethylsilyl ) derivative by reacting with bis ( trimethylsilyl ) trifluoroacetamide with 10% trimethylchlorosilane ( Regisil ) at 60°C for 20 min . Resulting MOX-TMS derivatives were run in GC-MS . Analyses were carried out on an Agilent 5973 mass spectrometer equipped with 6890 Gas Chromatograph . A DB17-MS capillary column ( 30 m × 0 . 25 mm × 0 . 25 μm ) was used in all assays with a helium flow of 1 ml/min . Samples were analyzed in Selected Ion Monitoring ( SIM ) mode using electron impact ionization ( EI ) . Ion dwell time was set to 10 ms . The following metabolites were monitored: Glycerol 3 phosphate ( G3P ) , 3 Phosphoglycerate ( 3PG ) , Lactate , Alanine , Citrate , Succinate , Fumarate , Malate , Glycine , and Serine . MIN6 cells were diluted to 80 000 cells/well in a Seahorse tissue culture system in the presence of either GFP or ATF4 overexpression . Cells were plated 2 days prior to experimentation . The cells were washed with warm PBS and then incubated for 30 min at 37°C and ambient CO2 in HCO3-free DMEM containing 25 mM glucose , 2 mM glutamine , 1 mM pyruvate ( pH 7 . 4 ) . Cells were then treated sequentially with oligomycin ( 0 . 2 μg/ml ) , carbonyl cyanide 4- ( trifluoromethoxy ) phenylhydrazone ( FCCP , 1 μM ) and rotenone ( 1 μM ) . The rates of mitochondrial respiration and cellular acidification were determined by using the Seahorse extracellular flux analyzer ( Seahorse Bioscience , North Billerica , MA ) . Corrected oxygen consumption rate ( OCR ) and extracellular acidification rate ( ECAR ) values were normalized to cell number . MIN6 cells extracts for protein immunodetection were obtained after cell lysis in RIPA buffer as described before ( Krokowski et al . , 2013 ) . Protein content was determined by the BCA assay ( BioRad ) . Mouse islet protein extracts were extracted in RIPA buffer . From approximately 100 islets the same amount of extracts determined by measuring DNA content with the Quant-iT dsDNA assay kit ( Molecular Probes ) and equal DNA amount was used for immunodetection . Western blotting was performed as described before ( Krokowski et al . , 2013 ) . Anti-Actin ( ab 3280 ) antibodies were from Abcam . Anti-CTH ( H00001491 ) and anti-CBS ( H00000875 ) were from Abnova . Antibodies against: PERK ( 3192 ) , phospho PERK ( 3179 ) and PKM2 ( 4053 ) were purchased from Cell Signaling . Anti-ATF4 ( sc-200 ) , anti-GAPDH ( sc-32233 ) , anti-eIF2α ( sc-133227 ) , and XBP1 ( sc-7160 ) were from Santa Cruz Biotechnology . Antibodies against phosphorylated eIF2α ( NB 110–56949 ) and Ero1α ( NB 100–2525 ) were obtained from Novus and anti-tubulin ( T9026 ) serum was from Sigma-Aldrich . | Proteins play essential roles in almost every aspect of a cell’s life , and also contribute to the structure and function of body tissues and organs . Cells and tissues adapt to their continuously changing environments by regulating the activity of their proteins . For example , proteins that are not fully active immediately after they are built instead require further ‘posttranslational’ modifications to become active . Amino acids are the building blocks of proteins , and cysteine amino acids are frequent sites of posttranslational modifications because they are particularly chemically reactive . Under certain conditions inside the cell , the sulfur atom in a cysteine can bond with chemical group containing a second sulfur atom plus a hydrogen atom . This process , which is known as sulfhydration , can be triggered by the presence of the gas molecule , hydrogen sulfide ( H2S ) . The levels of hydrogen sulfide are highly regulated in the body , and it has been suggested to play a role in aging , environmental stress and many diseases . However , it is not clear whether sulfhydration plays a major role in disease conditions by modifying protein activity . Efforts to address this question have been limited by a lack of methods that can measure the extent of sulfhydration of proteins . However , Gao et al . have now devised such a method . The approach takes steps to avoid false-positive and false-negative results , and can identify changes in the sulfhydration of cysteines across the entire complement of proteins produced by a cell , tissue or organ . Gao et al . then used this new method to show that a master regulator of transcription ( i . e . a protein that regulates the expression of many genes ) causes large-scale changes in cysteine sulfhydration . These large-scale changes resulted in the reprogramming of the cell’s energy metabolism , and further experiments showed that hydrogen sulfide accumulation influences sulfhydration , protein activity and signaling pathways . The development of this new method may now lead to additional discoveries into the role of hydrogen sulfide as a signaling molecule . | [
"Abstract",
"Introduction",
"Discussion",
"Materials",
"and",
"methods"
] | [
"biochemistry",
"and",
"chemical",
"biology",
"cell",
"biology"
] | 2015 | Quantitative H2S-mediated protein sulfhydration reveals metabolic reprogramming during the integrated stress response |
Influenza viruses undergo continual antigenic evolution allowing mutant viruses to evade host immunity acquired to previous virus strains . Antigenic phenotype is often assessed through pairwise measurement of cross-reactivity between influenza strains using the hemagglutination inhibition ( HI ) assay . Here , we extend previous approaches to antigenic cartography , and simultaneously characterize antigenic and genetic evolution by modeling the diffusion of antigenic phenotype over a shared virus phylogeny . Using HI data from influenza lineages A/H3N2 , A/H1N1 , B/Victoria and B/Yamagata , we determine patterns of antigenic drift across viral lineages , showing that A/H3N2 evolves faster and in a more punctuated fashion than other influenza lineages . We also show that year-to-year antigenic drift appears to drive incidence patterns within each influenza lineage . This work makes possible substantial future advances in investigating the dynamics of influenza and other antigenically-variable pathogens by providing a model that intimately combines molecular and antigenic evolution .
Seasonal influenza infects between 10% and 20% of the human population every year , causing an estimated 250 , 000 to 500 , 000 deaths annually ( Influenza Fact sheet , 2009 ) . Although individuals develop long-lasting immunity to particular influenza strains after infection , antigenic mutations to the influenza virus genome result in proteins that are recognized to a lesser degree by the human immune system , leaving individuals susceptible to future infection . The influenza virus population continually evolves in antigenic phenotype in a process known as antigenic drift . A large proportion of the disease burden of influenza stems from antigenic drift , which allows individuals to be infected multiple times throughout their lives . Although influenza vaccines may lack efficacy for a variety of reasons ( Osterholm et al . , 2012 ) , antigenic drift causes efficacy of a fixed vaccine formulation against circulating viruses to decline over time . A thorough understanding of the process of antigenic drift is essential to public health efforts to control mortality and morbidity through the use of a seasonal influenza vaccine . Before 2009 , there were four major lineages of influenza circulating within the human population: the H3N2 and H1N1 subtypes of influenza A , and the Victoria and Yamagata lineages of influenza B . In the case of influenza A , subtypes A/H3N2 and A/H1N1 refer to the genes , hemagglutinin ( H or HA ) and neuraminidase ( N or NA ) , that are primarily responsible for the antigenic character of a strain . In the case of influenza B , Victoria ( B/Vic ) and Yamagata ( B/Yam ) refer to antigenically distinct lineages which diverged from a single lineage prior to 1980 ( Rota et al . , 1990 ) . Mutations to the HA1 region of the hemagglutinin protein are thought to drive the majority of antigenic drift in the influenza virus ( Wiley et al . , 1981; Nelson and Holmes , 2007 ) . Experimental characterization of antigenic phenotype is possible through the hemagglutination inhibition ( HI ) assay ( Hirst , 1943 ) , which measures the cross-reactivity of one virus strain to serum raised against another strain through challenge or vaccination . Sera raised against older strains react poorly to more recent viruses resulting in new strains having a selective advantage over previously established strains . The results of many HI assays across a multitude of viruses of a single subtype can be combined to yield a two-dimensional map , quantifying antigenic similarity and distance ( Smith et al . , 2004 ) . The antigenic map of influenza A/H3N2 has shown substantial evolution of the influenza virus population since its emergence in 1968 . Evolution of antigenic phenotype appears punctuated with episodes of more rapid innovation interspersed by periods of relative stasis , whereas genetic evolution appears more continuous ( Smith et al . , 2004 ) , suggesting that a relatively small number of genetic changes or combinations of genetic changes may drive changes in antigenic phenotype ( Koel et al . , 2013 ) . The process of antigenic drift results in the rapid turnover of the virus population , so that despite mutation , genetic diversity among contemporaneous viruses remains low . Such population turnover is supported by phylogenetic analysis that shows a characteristically ‘spindly’ tree with a single predominant trunk lineage and transitory side branches that persist for only 1–5 years ( Fitch et al . , 1997 ) . Previously , the antigenic and genetic patterns of influenza evolution have been analyzed essentially in isolation . An antigenic map is constructed from a panel of HI measurements , and a phylogenetic tree is constructed from sequence data . However , the opportunity for a combined approach exists as both the antigenic map and the phylogenetic tree often contain many of the same isolates . Here , we implement a flexible Bayesian approach to jointly characterize the antigenic and genetic evolution of the influenza virus population . We apply this approach to investigate the dynamics of A/H3N2 , A/H1N1 , B/Vic and B/Yam viruses , and , for the first time , present detailed reconstructions of the antigenic dynamics of all four circulating influenza lineages .
To assess patterns of antigenic evolution among influenza strains , we implemented a Bayesian probabilistic analog of multidimensional scaling ( MDS ) , referred to here as BMDS ( see ‘Materials and methods’ ) . In this model , viruses and sera are given N-dimensional locations , thus specifying an ‘antigenic map’ , such that distances between viruses and sera in this space are inversely proportional to cross-reactivity . In the BMDS model , a map distance of one antigenic unit translates to an expectation of a twofold drop in HI titer between virus and sera . Maps that produce pairwise distances most congruent with the observed titers will have a high likelihood and will be favored by the BMDS model . We integrate over sources of uncertainty , such as antigenic locations , in a flexible Bayesian fashion . We apply this model to HI measurements of virus isolates against post-infection ferret antisera for influenza A/H3N2 , A/H1N1 , B/Vic and B/Yam . We begin with Bayesian analogs of the models used by Smith et al . ( 2004 ) , in which viruses and sera are represented as N-dimensional locations as described in the ‘Antigenic cartography’ section of ‘Materials and methods’ . In this case , ‘serum potencies’ are fixed to the maximum titers exhibited by particular ferret sera and give the baseline expectation for titer when virus and serum are antigenically identical . Potency differs between serum isolates due to experimental noise ( e . g . , variation in serum concentration ) , but also due to differential ferret immune responses , causing some serum isolates to inhibit hemagglutination at generally higher titers than other isolates . In this model , virus and serum locations follow an uninformative diffuse normal prior . After comparing models of differing dimensions , Smith et al . ( 2004 ) arrive at a 2D model as the preferred model for their data . Smith et al . ( 2004 ) implement a form of MDS , seeking to optimize virus and serum locations such that the sum of squared errors between expected and observed titers is minimized ( Equation 3 of ‘Materials and methods’ ) . Here , in implementing BMDS , we provide a likelihood function for the probability of observing HI data given virus and serum locations ( Equation 8 of ‘Materials and methods’ ) and seek to estimate model parameters through Bayesian inference using Markov Chain Monte Carlo ( MCMC ) . However , the basic antigenic model describing drop in HI titer as proportional to Euclidean distance between virus and serum locations is identical between these methods . We test model performance by constructing training datasets representing 90% of the HI measurements for each of the four influenza lineages and test datasets representing the remaining 10% of the measurements for each lineage . By fitting the BMDS model to the training dataset , we are able to predict HI titers in the test dataset and compare these predicted titers to observed titers . We find that a two-dimensional model has better predictive power than models of lower or higher dimension in all four influenza lineages ( models 1–5; Table 1 ) . We find that this 2D model performs well , yielding an average absolute predictive error of between 0 . 78 and 0 . 91 log2 HI titers across influenza lineages ( model 2; Table 1 ) , in line with the results of Smith et al . ( 2004 ) . Consequently , we specify a two-dimensional model in all subsequent analyses . The finding of a low-dimensional map across influenza lineages extends previous studies in A/H3N2 ( Smith et al . , 2004 ) and remains an interesting and fundamental empirical observation . 10 . 7554/eLife . 01914 . 003Table 1 . Average absolute prediction error of log2 HI titer for test data across models and datasetsDOI: http://dx . doi . org/10 . 7554/eLife . 01914 . 003Test errorModelDataDimenLocation priorSerum potencyVirus avidityA/H3N2A/H1N1B/VicB/Yam1HI1DUninformedFixedNone1 . 350 . 940 . 901 . 082HI2DUninformedFixedNone0 . 910 . 780 . 820 . 903HI3DUninformedFixedNone0 . 930 . 800 . 850 . 924HI4DUninformedFixedNone0 . 980 . 840 . 900 . 975HI5DUninformedFixedNone1 . 040 . 890 . 981 . 046HI/year2DDriftFixedNone0 . 910 . 750 . 770 . 837HI/year/seq2DDiffusion/DriftFixedNone0 . 890 . 740 . 740 . 838HI/year/seq2DDiffusion/DriftEstimatedNone0 . 770 . 730 . 660 . 759HI/year/seq2DDiffusion/DriftFixedEstimated0 . 800 . 720 . 690 . 7510HI/year/seq2DDiffusion/DriftEstimatedEstimated0 . 760 . 710 . 640 . 72 Previous work on influenza antigenic and genetic evolution has shown that antigenic distance accumulates with increasing genetic distance ( Hay and Gregory , 2001; Smith et al . , 2004; Russell et al . , 2008 ) . Here , we examine pairwise relationships between viruses and observe a correlation between amino acid mutations and antigenic distance ( Figure 1 ) and a similar correlation between phylogenetic distance , measured in years , and antigenic distance ( Figure 1 ) . Thus , genetic relationships between viruses provide some predictive power to estimate antigenic distances in the absence of HI data . However , the magnitudes of the coefficients of determination R2are low ( Figure 1 ) , suggesting that genetic relationships alone will not completely resolve antigenic distances . 10 . 7554/eLife . 01914 . 004Figure 1 . Pairwise correlations between genetic distance , measured as amino acid mutations or as phylogenetic distance , and antigenic distance for influenza A/H3N2 , A/H1N1 , B/Vic , and B/Yam . The top row shows correlations between number of amino acid mutations in HA1 and average antigenic distance between 10 , 000 random pairs of viruses . The bottom row shows correlations between average phylogenetic distance , measured in terms of years , and average antigenic distance between 10 , 000 random pairs of viruses . Dashed lines show linear model fits , with R2 and slope noted , while solid lines show LOESS fits . Antigenic distances derive from model 2 of Table 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 01914 . 004 Consequently , we seek to flexibly incorporate genetic data by modeling antigenic phenotype as an evolutionary diffusion ( Lemey et al . , 2010 ) , wherein a virus’s antigenic character state evolves along branches of the phylogenetic tree according to a Brownian motion process ( see ‘Materials and methods’ ) . The phylogenetic diffusion process acts as a prior on virus locations , so that genetically similar viruses are expected to share similar antigenic locations . The antigenic diffusion process includes both systematic drift with time and covariance induced by phylogenetic proximity . We examine the effects of including only systematic drift ( model 6; Table 1 ) and systematic drift plus phylogenetic diffusion ( model 7; Table 1 ) , finding a small increase in predictive accuracy of between 0 . 02 and 0 . 08 log2 HI titers when both processes are included . The systematic drift process informs virus and serum locations by dates of isolation and the phylogenetic diffusion process informs virus locations by genetic sequences . Thus , in these models , antigenic locations are inferred using both genetic data and HI data and will differ from locations inferred from HI data alone . If HI data is rich , then we expect minor differences in antigenic locations with the inclusion of genetic data ( as may be the case for A/H3N2 ) , while if HI data is spare , then we expect genetic data to play a larger role in determining antigenic locations ( as may be the case for B/Vic and B/Yam ) . We further extend the model by estimating serum potencies , rather than fixing serum potencies at maximum titers . Serum potencies differ across isolates due to experimental variation in serum extraction and processing or due to variation in ferret immune response . Serum potency determines the baseline expectation of titer when virus and serum have identical antigenic locations . However , if serum potency is fixed to the serum’s maximum titer , this will often not be the case , as the virus giving the maximum titer may be antigenically distinct from the serum . Thus , fixing serum potencies will tend to under-estimate effect size; we observe a mean effect of 10 . 42 log2 HI titers for A/H3N2 when fixing serum potencies and a mean of 10 . 94 when estimating serum potencies . We find that estimating serum potencies improves test error further ( model 8 , Table 1 ) , with improvements of between 0 . 01 and 0 . 12 log2 HI titers . Additionally , we include and estimate ‘virus avidity’ in an analogous fashion , which is intended to represent differences in overall HI reactivity between viruses . Experimental work has demonstrated that influenza variants exist that differ in HA binding activity for cell surface glycan receptors , and that these high-avidity variants may arise in the presence of antibody pressure ( Hensley et al . , 2009 ) . The presence of differential virus avidity has been previously shown to distort antigenic maps constructed from a model that disregards avidity effects ( Li et al . , 2013 ) . Here , with virus avidities estimated , baseline titer derives from both the virus and the serum used in the HI reaction . We find that including virus avidities further improves test error , either with fixed serum potencies ( model 9 , Table 1 ) or with estimated serum potencies ( model 10 , Table 1 ) . With fixed serum potencies , the inclusion of virus avidities results in improvements of between 0 . 02 and 0 . 09 log2 HI titers and with estimated serum potencies , the inclusion of virus avidities results in improvements of between 0 . 01 and 0 . 05 log2 HI titers . We find that the average absolute error in predicted log2 HI titer is nearly constant with antigenic distance ( Pearson correlation , r = 0 . 098 ) , thus supporting our model assumption that the drop in log2 titer is proportional to the Euclidean distance separating viruses and sera on the antigenic map . Additionally , we find that the absolute error in predicted titer is nearly constant with time ( Pearson correlation , r = −0 . 007 ) . Antigenic locations inferred by the model are well resolved; estimates of antigenic distance between pairs of viruses show relatively little variation across the posterior . We estimate that virus distances have , on average , a 50% credible interval of ±0 . 45 antigenic units for A/H3N2 , ±0 . 57 units for A/H1N1 , ±0 . 76 units for B/Vic , and ±0 . 65 units for B/Yam . We find strong correspondence between our results and previous results by Smith et al . ( 2004 ) , with equivalent models producing globally consistent antigenic maps and other models producing locally consistent maps with a small degree of global inconsistency ( see ‘Methods’ ) . When implementing the same underlying model , differences in the MDS and BMDS approaches reflect greater philosophical differences between maximum-likelihood and Bayesian statistical approaches , with the former seeking the single most likely explanation for the data , and the latter seeking to fully characterize model uncertainty . Additionally , the BMDS method improves flexibility , allowing extensions to the basic cartographic model , such as the incorporation of virus avidities and evolutionary priors , that improve fit and add biological interpretability . Through our analysis , we reveal the antigenic , as well as evolutionary , relationships among viruses in influenza A/H3N2 , A/H1N1 , B/Vic and B/Yam , quantifying both antigenic and evolutionary distances between strains ( Figure 2 , Figure 2—source data 1 ) . Over the time period of 1968 to 2011 , influenza A/H3N2 shows substantially more antigenic evolution than is exhibited by A/H1N1 over the course of 1977 to 2009 or B/Vic and B/Yam over the course of 1986 to 2011 . We observe prominent antigenic clusters in A/H3N2 and A/H1N1 , but less prominent , though still apparent , clustering in B/Vic and B/Yam . Antigenic clusters show high genetic similarity , so that we observe very few mutation events leading to each cluster , rather than the repeated emergence of clusters . This analysis makes the fate of antigenic clusters obvious , with two clusters in A/H3N2 ( Victoria/75 and Beijing/89 ) appearing to be evolutionary dead-ends . Labeling of prominent antigenic clusters in Figure 2 is intended as a rough guide for orientation and not as exhaustive catalog of antigenic variation . 10 . 7554/eLife . 01914 . 005Figure 2 . Antigenic locations of A/H3N2 , A/H1N1 , B/Vic , and B/Yam viruses showing evolutionary relationships between virus samples . Circles represent a posterior sample of virus locations and have been shaded based on year of isolation . Antigenic units represent twofold dilutions of the HI assay . Absolute positioning of lineages , for example A/H3N2 and A/H1N1 , is arbitrary . Lines represent mean posterior diffusion paths when virus locations are fixed . Prominent antigenic clusters are labeled after vaccine strains present within clusters , and are abbreviated from Hong Kong/68 , England/72 , Victoria/75 , Bangkok/79 , Sichuan/87 , Beijing/89 , Beijing/92 , Wuhan/95 , Sydney/97 , Fujian/02 , California/04 , Wisconsin/05 , Brisbane/07 , Perth/09 ( A/H3N2 ) , USSR/77 , Singapore/86 , Beijing/95 , New Caledonia/99 , Solomon Islands/06 ( H1N1 ) , Victoria/87 , Hong Kong/01 , Malaysia/04 , Brisbane/08 ( Vic ) , Yamagata/88 , Shanghai/02 , Florida/06 , Wisconsin/10 ( Yam ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01914 . 00510 . 7554/eLife . 01914 . 006Figure 2—source data 1 . This tab-delimited text file lists every virus in Figures 2 and 3 , including lineage , strain name , year of isolation , and coordinates in antigenic dimensions 1 and 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 01914 . 006 HI assays lack sensitivity beyond a certain point , so that for A/H3N2 , cross-reactive measurements only exist between strains sampled at most 14 years apart , leaving only threshold titers , for example ‘<40’ , in more temporally distant comparisons . Because of the threshold of sensitivity of the HI assay , it is difficult to distinguish a linear trajectory in 2D antigenic space from a slightly curved trajectory ( see ‘Materials and Methods’ ) . To solve this problem of identifiability , we assumed a weak prior that favors linear movement in the 2D antigenic space ( present in models 6 through 9; Table 1 ) , with the slope of the linear relationship and the precision of the relationship incorporated into the Bayesian model ( see ‘Materials and methods’ ) . Because of this , we interpret map locations locally rather than globally , and assess rates of antigenic movement without making strong statements about the larger configuration under which the movement occurs . We find that influenza A/H3N2 evolved along antigenic dimension 1 at an estimated rate of 1 . 01 antigenic units per year ( Figure 3 , Figure 3—source data 1; Table 2 ) . However , we observe occasional large jumps in antigenic phenotype ( Figure 3 ) , corresponding to cluster transitions identified by Smith et al . ( 2004 ) . Most variation is contained within the first antigenic dimension , but dimension 2 occasionally shows variation when two antigenically distinct lineages emerge and transiently coexist ( Figure 2 ) , as is the case with the previously identified Beijing/89 and Beijing/92 clusters . 10 . 7554/eLife . 01914 . 007Figure 3 . Antigenic drift of A/H3N2 , A/H1N1 , B/Vic and B/Yam viruses showing evolutionary relationships between virus samples . Antigenic drift is shown in terms of change of location in the first antigenic dimension through time . Circles represent a posterior sample of virus locations and have been shaded based on year of isolation . Antigenic units represent twofold dilutions of the HI assay . Relative positioning of lineages , for example A/H3N2 and A/H1N1 , in the vertical axis is arbitrary . Lines represent mean posterior diffusion paths when virus locations are fixed . Prominent antigenic clusters are labeled after vaccine strains present within clusters , and are abbreviated from Hong Kong/68 , England/72 , Victoria/75 , Bangkok/79 , Sichuan/87 , Beijing/89 , Beijing/92 , Wuhan/95 , Sydney/97 , Fujian/02 , California/04 , Wisconsin/05 , Brisbane/07 , Perth/09 ( A/H3N2 ) , USSR/77 , Singapore/86 , Beijing/95 , New Caledonia/99 , Solomon Islands/06 ( H1N1 ) , Victoria/87 , Hong Kong/01 , Malaysia/04 , Brisbane/08 ( Vic ) , Yamagata/88 , Shanghai/02 , Florida/06 , Wisconsin/10 ( Yam ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01914 . 00710 . 7554/eLife . 01914 . 008Figure 3—source data 1 . This tab-delimited text file lists every virus in Figures 2 and 3 , including lineage , strain name , year of isolation , and coordinates in antigenic dimensions 1 and 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 01914 . 00810 . 7554/eLife . 01914 . 009Table 2 . Estimates of drift rate μ ( in units per year ) , diffusion volatility σx2 ( in units2 per year ) and scaled effective population size Neτ ( in years ) for influenza A/H3N2 , A/H1N1 , B/Vic and B/Yam including posterior means and 95% highest posterior density intervalsDOI: http://dx . doi . org/10 . 7554/eLife . 01914 . 009LineageDrift μVolatility σx2Effective pop size NeτA/H3N21 . 01 ( 0 . 98–1 . 04 ) 1 . 25 ( 0 . 98–2 . 35 ) 5 . 03 ( 4 . 42–5 . 73 ) A/H1N10 . 62 ( 0 . 56–0 . 67 ) 0 . 92 ( 0 . 65–1 . 56 ) 6 . 38 ( 4 . 99–8 . 12 ) B/Vic0 . 42 ( 0 . 32–0 . 51 ) 1 . 22 ( 0 . 85–2 . 25 ) 10 . 40 ( 8 . 42–12 . 80 ) B/Yam0 . 32 ( 0 . 25–0 . 39 ) 0 . 71 ( 0 . 46–1 . 36 ) 9 . 48 ( 7 . 76–11 . 50 ) We find that other lineages of influenza evolved in antigenic phenotype substantially slower than A/H3N2 ( Figure 3 , Table 2 ) . Influenza A/H1N1 evolved at a rate of 0 . 62 units per year , but showed a similar pattern of punctuated antigenic evolution with occasional larger jumps in phenotype , such as the emergence of the Solomon Islands/06 cluster . Influenza B/Victoria and B/Yamagata evolved slower still , with mean estimated rates 0 . 42 units per year and 0 . 32 units per year , respectively . Punctuated evolution is less obvious in B/Yam and B/Vic compared to A/H3N2 and A/H1N1 , but antigenic clusters are still apparent , with recent transitions to the Brisbane/08 cluster in B/Vic ( Barr et al . , 2010 ) and to the Wisconsin/10 cluster in B/Yam ( Klimov et al . , 2012 ) . Interestingly , a minor lineage of B/Vic , denoted B/Hubei-Songzi/51/2008 ( Barr et al . , 2010 ) , has persisted through 2011 , while remaining antigenically distinct from B/Brisbane/60/2008 viruses ( Figure 3 ) . Although we observe significantly different drift rates between lineages , we observe less variation in diffusion volatility ( Table 2 ) . This is reflected in Figure 3 , where all four lineages exhibit similar levels of standing antigenic variation , despite A/H3N2 drifting more quickly in antigenic phenotype . These patterns of antigenic drift influence the corresponding virus phylogenies ( Figure 4 ) . Influenza A/H3N2 has a characteristically spindly tree showing rapid turnover of the virus population , while A/H1N1 and B have trees that show greater degrees of viral coexistance ( Figure 4 ) . The scaled effective population size Neτ measures the timescale of coalescence of a phylogeny and quantifies the visual distinction between a ‘spindly’ tree and a ‘bushy’ tree ( Bedford et al . , 2011 ) . In this case , Ne represents the number of concurrent infections in a panmictic population with generation interval τ ( time separating infections up the genealogical tree ) , so that Neτ is measured in terms of years and gives the expected waiting time for two randomly chosen lineages to coalesce in the genealogical tree . We see that Neτ broadly correlates with the rate of antigenic drift ( Table 2 ) , with A/H3N2 showing fast drift and reduced effective population size as expected from basic epidemiological models ( Bedford et al . , 2012 ) . Antigenic drift results in the replacement of antigenically primitive lineages by antigenically advanced lineages , thereby reducing genealogical diversity . 10 . 7554/eLife . 01914 . 010Figure 4 . Time-resolved phylogenetic trees of A/H3N2 , A/H1N1 , B/Vic and B/Yam viruses . The maximium-clade credibility ( MCC ) tree is shown for each virus . These trees show genealogical relationships , so that branches are measured in terms of years rather than substitutions . DOI: http://dx . doi . org/10 . 7554/eLife . 01914 . 010 Thus , we observe a faster rate of antigenic drift in influenza A/H3N2 than in A/H1N1 or either lineage of influenza B . Previous work using general epidemiological models has suggested that rates of antigenic drift may be influenced by both the fundamental reproductive number R0 and the rate at which mutation decreases cross-immunity ( Gog and Grenfell , 2002; Lin et al . , 2003 ) . Correspondingly , models specific to influenza evolution ascribe differences in the rate of antigenic drift of A/H3N2 relative to A/H1N1 and influenza B to either greater R0 or greater mutation rate ( Ferguson et al . , 2003; Bedford et al . , 2012 ) . Without more detailed epidemiological modeling , the present study cannot conclusively distinguish between these causal possibilities . We sought to summarize year-to-year patterns of antigenic drift by calculating the difference in mean virus location between consecutive years ( Figure 5A ) . We estimate year-to-year antigenic drift for years 1992 to 2011 by calculating the average location along dimension 1 of phylogenetic lineages present in the tree at year i and comparing this location to the average location of phylogenetic lineages present in the tree at year i − 1 . There may often be large discontinuities in virus locations across the population; our use of difference in mean location is meant to capture both the distance between antigenic clusters and also the change in cluster frequency over consecutive years . We observe greater heterogeneity in year-to-year antigenic drift in type A than in type B lineages ( Figure 5B ) , with standard deviation of year-to-year antigenic drift equal to 0 . 97 units in A/H3N2 , 0 . 66 units in A/H1N1 , 0 . 46 units in B/Vic and 0 . 26 units in B/Yam . This analysis classifies drift only to the level of consecutive years; some coarse-graining of the timings of transition events will necessarily occur . 10 . 7554/eLife . 01914 . 011Figure 5 . Year-to-year antigenic drift between 1992 and 2011 in A/H3N2 , A/H1N1 , B/Vic and B/Yam viruses . ( A ) Timeseries of year-to-year antigenic drift between 1992 and 2011 in A/H3N2 , A/H1N1 , B/Vic and B/Yam viruses . Colored lines represent year-to-year antigenic drift , where drift for year i is measured as the mean of antigenic dimension 1 of phylogenetic lineages in year i compared to the mean of antigenic dimension 1 of phylogenetic lineages from the previous year i − 1 . For example , 2000 represents difference in antigenic dimension 1 between viruses from 1999 to 2000 . Error bars represent 50% Bayesian credible intervals of year-to-year drift . Gray dotted lines represent lineage-specific seasonal incidence in the USA taken as average influenza-like illness ( ILI ) multiplied by proportion of viruses attributable to a lineage for each season . Here , 2000 represents the 2000/2001 influenza season . ( B ) Distribution of year-to-year antigenic drift between 1992 and 2011 in A/H3N2 , A/H1N1 , B/Vic and B/Yam viruses . DOI: http://dx . doi . org/10 . 7554/eLife . 01914 . 011 We investigate the relationship between rates of antigenic drift and seasonal incidence in the USA in A/H3N2 , A/H1N1 , B/Vic and B/Yam . We measure seasonal incidence from USA CDC influenza surveillance reports for each virus lineage ( A/H3N2 , A/H1N1 , B/Vic , B/Yam ) by taking the average influenza-like ( ILI ) percentage in a season and multiplying this by the relative proportion of virus isolations attributable to a particular influenza lineage in a season ( see ‘Materials and methods’ ) . This measure of incidence has previously been shown to have have predictive power in the analysis of seasonal influenza trends ( Goldstein et al . , 2011 ) . We analyze incidence from the 1998/1999 to the 2008/2009 seasons to avoid possible complications from the 2009 pandemic . We begin by comparing overall rates of antigenic drift ( Figure 3 , Table 2 ) to overall levels of seasonal incidence across influenza lineages , finding a significant correlation between rate of antigenic drift and relative incidence across the four lineages ( Pearson correlation , r = 0 . 97 , p = 0 . 041 ) . We follow-up this analysis with a more detailed analysis of year-to-year variation in antigenic drift and lineage-specific incidence , comparing incidence in a season to antigenic drift of viruses coming into this season ( Figure 5A ) . For example , we compare antigenic drift of viruses from 2000 to 2001 to incidence in the 2001/2002 season . Within each virus lineage , we find that years with pronounced antigenic drift tend to show increased incidence ( Figure 6 ) , finding Pearson correlation coefficients of 0 . 51 , 0 . 29 , 0 . 44 and 0 . 14 for A/H3N2 , A/H1N1 , B/Vic and B/Yam respectively . We calculated significance using bootstrap permutation tests finding p-values of 0 . 056 , 0 . 201 , 0 . 097 , and 0 . 341 respectively . We applied a similar bootstrap permutation test to calculate the significance of finding the observed degree of correlation across all four lineages , arriving at a p-value of 0 . 018 . The fact that we observe periods of pronounced antigenic drift preceding increased incidence in each of the four influenza lineages suggests a causal relationship , in which antigenic evolution drives increased incidence . 10 . 7554/eLife . 01914 . 012Figure 6 . Relationship between antigenic drift and seasonal incidence for years 1998 to 2009 in influenza A/H3N2 , A/H1N1 , B/Vic and B/Yam . Antigenic drift from year i − 1 to year i is compared to incidence in the season i/i + 1 . For example , year-to-year antigenic drift from 2000 to 2001 is measured against incidence in the 2001/2002 season . DOI: http://dx . doi . org/10 . 7554/eLife . 01914 . 012 However , it is possible that if sampling count influences cartographic estimates then a spurious correlation could arise in which years with greater incidence have higher sample counts and artifactually high estimates of drift . We controlled for this possibility by testing to see if viral isolate count influences estimates of year-to-year drift by correlating drift between years i and i − 1 against the ratio of the number of isolates from year i to the number of isolates from year i − 1 . We found little correlation when combining data across lineages ( Figure 7 , Pearson’s r = −0 . 01 ) . We tested significance following the same bootstrap procedure we used to assess the correlation between drift and incidence , finding a p-value of 0 . 514 . Separating lineages gave p-values of 0 . 717 , 0 . 246 , 0 . 337 , 0 . 504 for A/H3N2 , A/H1N1 , B/Vic and B/Yam , respectively . These findings suggest our results to be unbiased with regard to sample count . 10 . 7554/eLife . 01914 . 013Figure 7 . Relationship between antigenic drift and sample counts for years 1998 to 2011 in influenza A/H3N2 , A/H1N1 , B/Vic and B/Yam . Antigenic drift from year i − 1 to year i is compared to the ratio of sample counts in year i to counts in year i − 1 . Only comparisons which had one or more samples in years i − 1 and i were retained , leaving 11 A/H3N2 , 7 A/H1N1 , 9 B/Vic and 10 B/Yam comparisons . Points are colored according to influenza lineage based on the color scheme in Figure 6 . DOI: http://dx . doi . org/10 . 7554/eLife . 01914 . 013 Understanding antigenic evolution in seasonal influenza is crucial to our efforts of surveillance and control . Cartographic methods allow complex HI datasets to be compressed to more approachable location-based summaries that quantify antigenic relationships between strains , including relationships not directly measured via HI . In this study , we provide a foundation for evolutionary antigenic cartography , which seeks to simultaneously assess antigenic phenotype and antigenic evolution . We use this approach to characterize competitive dynamics across influenza lineages A/H3N2 , A/H1N1 , B/Vic and B/Yam and show that antigenic evolution within each lineage drives strain replacement and contributes to seasonal incidence patterns . We find that influenza A/H3N2 evolves faster in antigenic phenotype than A/H1N1 , which in turn evolves faster than B/Vic or B/Yam . Consequently , the influenza A/H3N2 virus population turns over more quickly than A/H1N1 or influenza B , exhibiting a smaller effective population size and a ‘spindlier’ phylogenetic tree . Furthermore , we observe a correlation between antigenic drift and viral incidence both across and within influenza lineages . The finding that antigenic evolution correlates with subsequent increased incidence within a lineage suggests a causal role for antigenic drift driving influenza incidence patterns . The correlation between incidence and drift further suggests the possibility of using HI data at the start of an influenza season to predict which lineage will subsequently predominate . The statistical framework presented here represents a baseline to which further advancements in modeling antigenic phenotype and evolution may be made . For example , our likelihood-based model facilitates the inclusion of possible covariates affecting immunological titer , which could include experimental factors such as red blood cell type used in the HI assay ( Lin et al . , 2012 ) and whether oseltamivir is included in the HI reaction ( Lin et al . , 2010 ) . Additionally , this framework should be ideally suited to uncovering genetic determinants of antigenic change , as both the sequence state and antigenic location of internal nodes in the phylogeny may be estimated . In this fashion , it should be possible to correlate sequence substitutions directly to antigenic diffusion . Identifying viruses that will come to predominate in the global virus population while they are still at low frequency remains an enormous challenge . However , combining evolutionary and antigenic information may eventually prove useful in identifying low-frequency , but expanding , lineages of antigenically novel viruses that represent ideal targets for vaccine strain selection .
Antigenic characteristics of viral strains are often assessed through immunological assays such as the hemagglutination inhibition ( HI ) assay ( Hirst , 1943 ) . At heart , these assays compare the reactivity of one virus strain to antibodies raised against another virus strain via challenge or vaccination . In the case of HI , the measurement of cross-reactivity takes the form of a titer representing the dilution factor at which serum raised against a particular virus ceases to be effective at inhibiting the binding of another virus to red blood cells . These factors are commonly assessed by serial dilution , so that HI titers will form a log series , 40 , 80 , 160 , etc… . Because experimental HI titers typically differ by factors of two , we find it convenient to work in log2 space and represent the titer of virus i against serum j as Hij=log2 ( HI titer ) , that is a titer of 160 has Hij = 7 . 32 . Due to experimental constraints , most comparisons cannot be made , leading to a sparse observation matrix H={Hij} . Further , measurements are usually interval and truncated , for example inhibition may cease somewhere between the serial titers of 160 and 320 , or inhibition may be absent at all titers assayed , suggesting a threshold somewhere between 0 and 40 . Previous work ( Smith et al . , 2004; Cai et al . , 2010 ) has used multidimensional scaling ( MDS ) to place viruses and sera on an ‘antigenic map’ . These methods heuristically optimize locations of viruses and sera by seeking to minimize the sum of squared errors between titers predicted by map locations and observed titers . Antigenic maps produced by these methods have proven useful in categorizing virus phenotypes ( Smith et al . , 2004 ) , but the extension of these methods to integrate genetic data remains notably lacking . Here , we follow previous models in representing antigenic locations as points in a low P-dimensional antigenic map . One of our initial goals is to find an optimal projection of the high-dimensional distance matrix H into this lower dimensional space . We conduct this projection using Bayesian multidimensional scaling ( BMDS ) ( Oh and Raftery , 2001 ) in which we construct a probabilistic model to quantify the fit of a particular configuration of cartographic locations to the observed matrix of serological measurements . Typically , P = 2 , but higher or lower dimensions may better reflect the data . Let xi∈RP represent the cartographic location of virus i for i = 1 , … , n , so that xi= ( xi1 , xi2 ) ′ for P = 2 . Similarly , let yj represent the cartographic location of serum j for j = 1 , … , k , so that yj= ( yj1 , yj2 ) ′ for P = 2 . For notational compactness , we collect together all virus coordinates into an n × P matrix x= ( x1 , … , xn ) ′ and all serum coordinates into an k × P matrix y= ( y1 , … , yk ) ′ . Virus and serum may be isolated from/raised against the same strain and have different cartographic locations , and separate serum isolates raised against the same strain may also have different cartographic locations . This gives a set of distances between virus and serum cartographic locations . ( 1 ) δij=||xi−yj||2 , where , ||⋅||2 is an L2 norm . Traditional approaches to antigenic cartography ( Smith et al . , 2004 ) begin by defining immunological distance as ( 2 ) dij=sj−Hij , where , Hij is the log2 titer of virus i against serum j and serum potency sj=max ( H1j , … , Hnj ) is fixed . In following multidimensional scaling , these approaches attempt to optimize over unknown X and Y such that ( 3 ) ∑ ( i , j ) ∈I ( δij−dij ) 2is minimized , where , I={ ( i , j ) :Hij is measured} . In the case of threshold measurements , this error function is modified slightly; see Smith et al . ( 2004 ) for further details . Here , we instead assume a probabilistic interpretation in which an observed titer is normally distributed around its cartographic expectation with variance φ2 , ( 4 ) Hij∼N ( sj−δij , φ2 ) . Consequently , the likelihood of observing an exact titer given the placement of antigenic locations is ( 5 ) f| ( Hij ) =ϕ ( Hij+δij−sjφ ) , where , ϕ ( ⋅ ) represents the standard normal probability density function ( PDF ) . Previous BMDS has employed a sampling density truncated to strictly positive quantities since dij are directly observed , non-negative quantities . In the antigenic setting , these remain random and can be negative since neither sj is known nor is Hij observed with much precision . HI assays sometimes show no inhibition at all measured titrations , for example a measurement can be reported as ‘<40’ . In this case , the likelihood of observing the threshold measurement follows the cumulative density of the lower tail of the normal distribution . ( 6 ) f⌟ ( Hij ) =Φ ( Hij+δij−sjφ ) , where , Φ ( ⋅ ) represents the standard normal cumulative distribution function ( CDF ) . Although it is simplest to assume that immunological measurements represent point estimates , it seems more natural to assume that the threshold for inhibition occurs between two titers , for example we observe inhibition at 1:160 dilution and no inhibition at 1:320 dilution . Rather than taking the HI titer as 160 , we can instead treat this as an interval measurement , assuming that the exact titer for inhibition would occur somewhere between 160 and 320 . HI titers are usually reported as the highest titer that successfully inhibits virus binding , so that in this case , we calculate the likelihood of an interval measurement as ( 7 ) f⊔ ( Hij ) =Φ ( Hij+δij−sj+1φ ) −Φ ( Hij+δij−sjφ ) . These likelihoods are illustrated in Figure 8 . Throughout our analyses , we use interval likelihoods f⊔ rather than point likelihoods f| unless otherwise noted . 10 . 7554/eLife . 01914 . 014Figure 8 . Likelihood of HI titers in the BMDS model . Here , we show the likelihoods of observing three different outcomes given δij=4 , φ=0 . 95 , and sj=log2 ( 1280 ) =10 . 32 . The likelihood of observing a threshold titer of ‘<40’ is equal to the lower tail of the probability density function f⌟ ( 5 . 32 ) =0 . 146 . The likelihood of observing a point measurement with an exact inhibiting titer of ‘90 . 5’ is equal to the density function f| ( 6 . 5 ) =0 . 413 . The likelihood of observing an interval measurement with an inhibiting titer somewhere between ‘160’ and ‘320’ is equal to f⊔ ( 7 . 32 ) =0 . 129 . DOI: http://dx . doi . org/10 . 7554/eLife . 01914 . 014 We calculate the overall likelihood by multiplying probabilities of individual measurements ( 8 ) L ( x , y ) =∏ ( i , j ) ∈If ( Hij ) , using probability functions f| , f⌟ and f⊔ as appropriate . We begin by assuming independent , diffuse normal priors on virus and serum locations . xi∼N ( m , Σ ) ( 9 ) yj∼N ( m , Σ ) , where , m= ( 0 , … , 0 ) ′ and Σ is a diagonal matrix with diagonal elements all equal to 10 , 000 . The preceding model represents immunological distance as a drop in titer against the most reactive comparison for a particular serum . However , this model may be biased in some circumstances . In one example , if a particular serum j is only measured against distant viruses , its maximum titer will be artificially low and the likelihoods concerning this serum will appear poor . To address this issue , we relax the assumption of fixed sj values and treat the expected log2 titer when δij=0 as a random variable . In this case , Hij still follows Equation 4 with expectation sj−δij , but the vector of ‘serum potencies’ s = ( s1 , … , sk ) is random and estimated rather than fixed . We assume that sj values are hierarchically distributed according to a normal distribution . We take an Empirical Bayesian approach in specifying the mean and variance of this distribution , set to the empirical mean and empirical variance of the set of maximum titers across sera {max ( H1j , … , Hnj ) :j=1 , … , k} . This formulation assumes that particular sera are more reactive in general than other sera . Additionally , we follow the same logic and assume that some virus isolates are more reactive than other virus isolates and include a parameter for ‘virus avidity’ vi representing the general level of reactivity across HI assays . With virus avidity included , observed titers follow ( 10 ) Hij∼N ( vi+sj2−δij , φ2 ) , and the vector of virus avidities vi for i = 1 , … , n is estimated in an analogous hierarchical fashion , with v normally distributed with mean and variance equal to the empirical mean and variance of the set of maximum titers across viruses {max ( Hi1 , … , Hik ) :i=1 , … , n} . As presented , multiple configurations of virus and serum locations X and Y will give the same likelihood of an observed data matrix H . An example of this phenomenon is shown in Figure 9 . In this case , it is impossible to determine from the HI data at hand whether the blue and yellow viruses are antigenically similar ( Figure 9A ) or antigenically divergent ( Figure 9B ) . This presents an issue of model identifiability , where absolute , as opposed to relative , antigenic locations cannot be determined from observing the serological data alone . Thus , in order to achieve a more interpretable model we impose a weak prior on global locations . In influenza , it’s clear that antigenic distance between strains increases with time ( Smith et al . , 2004; Cai et al . , 2010 ) . To capture this , we replace our previous diffuse prior with an informed prior in which the expected location of viruses and sera increases with date of sampling along dimension one , and each virus and serum location follows an independent normal distribution centered around this temporal expectation , so thatxi1∼μ ti+N ( 0 , σx2 ) ( 11 ) yj1∼μ tj+N ( 0 , σy2 ) , where , t is the difference between the date of the indexed virus or serum and the date of the earliest sampled virus or serum , and other dimensions follow xim∼N ( 0 , σx2 ) and yjm∼N ( 0 , σy2 ) for m ≥ 2 . Thus , this model assumes that virus and serum locations drift in a line across the antigenic map at rate μ . The parameter σx determines the breadth of the cloud of virus locations at each point in time , while σy determines the breadth of the cloud of serum locations . 10 . 7554/eLife . 01914 . 015Figure 9 . Schematic antigenic map with three viruses and two sera . ( A ) Map with virus 1 and virus 3 antigenically similar . ( B ) Map with virus 1 and virus 3 antigenically divergent . Virus 1 is shown in blue , virus 2 is shown in red and virus 3 is shown in yellow . Virus isolates are represented by filled circles , sera raised against viruses are shown as open circles and map distances δij are shown as solid lines connecting viruses and sera . Sera from virus 1 is compared against viruses 1 and 2 , while sera from virus 2 is compared against viruses 2 and 3 . Configurations ( A ) and ( B ) represent cartographic models that would give equal likelihoods to a set of serological data {H11 , H21 , H22 , H32} . DOI: http://dx . doi . org/10 . 7554/eLife . 01914 . 015 We simultaneously model antigenic locations and genetic relatedness by assuming that virus locations are influenced by evolution following a Brownian motion process ( Lemey et al . , 2010 ) . To do this , we replace the previous prior specifying independent virus locations with a prior that incorporates covariance based on shared evolutionary history . ( 12 ) X∼ ( μ t10⋮⋮μ tn0 ) + Evolutionary Brownian Process ( σx , τ ) for P = 2 , where , σx is the volatility parameter of the Brownian motion over virus locations and τ is a phylogeny specifying tree topology and branch lengths . Thus , viruses that are genetically similar are induced to have prior locations close to one another on the antigenic map . In the evolutionary Brownian process , the tips of the phylogeny τ correspond to the set of virus locations ( x1 , … , xn ) , and the probability of observing tip locations depends on the locations of internal nodes ( xn+1 , … , x2n−2 ) and on the location of the root node x2n−2 . This process assumes that a virus location xi follows from the location of its parent virus xf ( i ) , and with the addition of drift along dimension 1 , is distributed as ( 13 ) xi∼ ( μ di , 0 ) ′+N ( xf ( i ) , di Σ ) for P = 2 , where , f ( i ) is a function that maps nodes to parental nodes , di is the length of the branch connecting virus i to parent virus f ( i ) , and Σ is a diagonal matrix with diagonal elements all equal to σx2 . The root virus location x2n−1 is assumed to follow a normal distribution with expectation ( μ t2n−1 , 0 ) ′ for P = 2 and variance determined by the diffusion volatility σx ( Lemey et al . , 2010 ) . The probability of virus locations p ( x|μ , σx , τ ) is determined through analytical integration across internal states following the methods introduced in ( Lemey et al . , 2010 ) . This formulation corresponds to a Wiener process with drift , in which the drift term μ only influences the expected states of nodes along the phylogeny , but does not influence the covariance structure among these nodes , which remains the same as it does in a standard Wiener process ( Borodin and Salminen , 2002 ) . This allows the separation in Equation 12 between drift terms affecting only expectations and the evolutionary Brownian process that includes covariance among virus locations ( x1 , … , xn ) . In this study , the phylogenetic tree τ is estimated using sequence data for viruses 1 , … , n according to well-established methods implemented in the software package BEAST ( Drummond et al . , 2012 ) . Top-level priors for 1/φ2 , μ , 1/σx2 , and 1/σy2 are assumed to follow diffuse Gamma ( a , b ) distributions with a = 0 . 001 and b = 0 . 001 . These diffuse priors were chosen to be non-informative and provide little-to-no weight on the resulting posterior distributions . Under the full model , the posterior probability of observing virus and serum locations given immunological data is factored . ( 14 ) p ( x , y|H ) ∝p ( H|x , y , s , v , φ ) p ( x|μ , σx , τ ) p ( y|μ , σy ) p ( s , v , φ , μ , σx , σy , τ ) . We sample from this posterior distribution using the MCMC procedures implemented in the software package BEAST ( Drummond et al . , 2012 ) . Metropolis–Hastings proposals include transition kernels that translate individual virus and serum locations xi and yj and individual virus avidities vi and serum potencies sj , and other transition kernels that scale the entire set of virus and serum locations X and Y and that scale parameters φ , μ , σx and σy . For the present analysis , a two-step approach was taken to sample phylogenies , where a posterior sample of phylogenies was gathered using sequence data and then , in the cartographic analysis , trees from this set were randomly proposed and accepted following the Metropolis–Hastings algorithm ( Pagel et al . , 2004 ) . We compiled an antigenic dataset of hemagglutination inhibition ( HI ) measurements of virus isolates against post-infection ferret sera for influenza A/H3N2 by collecting data from previous publications ( Hay and Gregory , 2001; Smith et al . , 2004; Russell et al . , 2008; Barr et al . , 2010 ) , NIMR vaccine strain selection reports for 2002 and 2008–2012 ( Hay et al . , 2002; 2008a , 2009a; McCauley et al . , 2010a; 2010b , 2011b , 2012 ) and the February 2011 VRBPAC report ( Cox , 2011 ) . We queried the Influenza Research Database ( Squires et al . , 2012 ) and the EpiFlu Database ( Bogner et al . , 2006 ) for HA nucleotide sequences by matching strain names , for example A/HongKong/1/1968 , and only strains for which sequence was present were retained . If a strain had multiple sequences in the databases , we preferentially kept the IRD sequence and preferentially kept the longest sequence in IRD . Many strains had full length HA sequences , while other strains only possessed HA1 sequences . Sequences were aligned using MUSCLE v3 . 7 under default parameters ( Edgar , 2004 ) . This dataset had 2051 influenza isolates ( present as either virus or serum in HI comparisons ) dating from 1968 to 2011 . However , the majority of isolates were present from 2002 to 2007 . Because we are interested in longer-term antigenic evolution , we subsampled the data to have at most 20 virus isolates per year , preferentially keeping those isolates with more antigenic comparisons . We then kept only those serum isolates that are relatively informative to the antigenic placement of viruses , dropping serum isolates that are compared to four or fewer different virus isolates . This censoring left 402 virus isolates , 519 serum isolates and 10 , 059 HI measurements . Each virus isolate was compared to an average of 21 . 9 serum isolates , and each serum isolate was compared to an average of 18 . 0 virus isolates . Antigenic data for influenza A/H1N1 was collected from previous publications ( Kendal et al . , 1978; Nakajima et al . , 1979; Webster et al . , 1979; Nakajima et al . , 1981; Chakraverty et al . , 1982; Pereira and Chakraverty , 1982; Cox et al . , 1983; Daniels et al . , 1985; Chakraverty et al . , 1986; Raymond et al . , 1986; Stevens et al . , 1987; Donatelli et al . , 1993; Hay and Gregory , 2001; Daum et al . , 2002; McDonald et al . , 2007; Barr et al . , 2010 ) and NIMR vaccine strain selection reports for 2002–2010 ( Hay et al . , 2002 , 2008a , 2009a; McCauley et al . , 2010; Hay et al . , 2003 , 2004 , 2005a , 2005b , 2006a , 2006b , 2007a , 2007b , 2008b ) . The same procedure that was followed for A/H3N2 was repeated to match sequence data and to subsample antigenic comparisons . This procedure yielded 115 virus isolates , 77 serum isolates , and 1882 HI measurements over the course of 1977–2009 . Each virus isolate was compared to an average of 10 . 0 serum isolates , and each serum isolate was compared to an average of 16 . 2 virus isolates . Antigenic comparisons for influenza B/Victoria were collated from previous publications ( Rota et al . , 1990; Hay and Gregory , 2001; Muyanga et al . , 2001; Shaw et al . , 2002; Ansaldi et al . , 2004; Puzelli et al . , 2004; Xu et al . , 2004; Barr et al . , 2006; Daum et al . , 2006; Lin et al . , 2007 ) and vaccine strain selection reports for 2002–2012 ( Hay et al . , 2002 , 2008a , 2009a; McCauley et al . , 2010a; 2010b , 2011b , 2012; Hay et al . , 2003 , 2004 , 2005a , 2005b , 2006a , 2006b , 2007a , 2007b; Gust et al . , 2006; Hay et al . , 2009a; McCauley et al . , 2011b ) . Here , the sequence matching and subsampling procedure yielded 179 virus isolates , 70 serum isolates and 2003 HI measurements over the course of 1986–2011 . Each virus isolate was compared to an average of 6 . 5 serum isolates , and each serum isolate was compared to an average of 16 . 7 virus isolates . Antigenic comparisons for influenza B/Yamagata were collected from previous publications ( Kanegae et al . , 1990; Rota et al . , 1990; Nakajima et al . , 1992; Nerome et al . , 1998; Hay and Gregory , 2001; Muyanga et al . , 2001; Nakagawa et al . , 2002; Shaw et al . , 2002; Abed et al . , 2003; Ansaldi et al . , 2003 , 2004; Matsuzaki et al . , 2004; Puzelli et al . , 2004; Xu et al . , 2004; Barr et al . , 2006; Daum et al . , 2006; Lin et al . , 2007 ) and vaccine strain selection reports for 2002–2012 ( Hay et al . , 2002 , 2008a , 2009a; McCauley et al . , 2010a; 2010b , 2011b , 2012; Hay et al . , 2003 , 2004 , 2005a , 2005b , 2006a , 2006b , 2007a , 2007b; Gust et al . , 2006; Hay et al . , 2009a; McCauley et al . , 2011b ) . For B/Yamagata , the matching and subsampling procedure resulted in 174 virus isolates , 69 serum isolates and 1962 HI measurements over the course of 1987–2011 . Each virus isolate was compared to an average of 6 . 9 serum isolates , and each serum isolate was compared to an average of 17 . 3 virus isolates . Surveillance data was obtained from the Centers of Disease Control and Prevention FluView Influenza Reports from the yearly summaries of influenza seasons 1997–1998 to 2010–2011 ( Centers for Disease Control and Prevention , 2012 ) . As an example , one report states ‘collaborating laboratories in the United States tested 195 , 744 respiratory specimens for influenza viruses , 27 , 682 ( 14% ) of which were positive . Of these , 18 , 175 ( 66% ) were positive for influenza A viruses , and 9507 ( 34% ) were positive for influenza B viruses . Of the 18 , 175 specimens positive for influenza A viruses , 7631 ( 42% ) were subtyped; 6762 ( 87% ) of these were seasonal influenza A ( H1N1 ) viruses , and 869 ( 13% ) were influenza A ( H3N2 ) viruses’ . In this case , we estimate the relative proportion of A/H3N2 of the four lineages as 0 . 66 × 0 . 13 = 0 . 09 . Similar calculations were performed for A/H1N1 , B/Vic and B/Yam . Phylogenetic trees were estimated for A/H3N2 , A/H1N1 , B/Vic , and B/Yam using BEAST ( Drummond et al . , 2012 ) and incorporated the SRD06 nucleotide substitution model ( Shapiro et al . , 2006 ) , a coalescent demographic model with constant effective population size and a strict molecular clock across branches . MCMC was run for 60 million steps and trees were sampled every 50 , 000 steps after allowing a burn-in of 10 million steps , yielding a total sample of 2000 trees . These trees were treated as a discrete set of possibilities when subsequently sampled in the BMDS analysis ( Pagel et al . , 2004 ) . However , it would be possible to jointly sample from sequence data and serological data using these methods . MCMC was used to sample virus locations X , serum locations Y , virus avidities v , serum potencies s , MDS precision 1/φ2 , antigenic drift rate μ , virus location precision 1/σx2 , serum location precision 1/σy2 , and phylogenetic tree τ . MCMC chains were run for 500 million steps and parameter values sampled every 200 , 000 steps after a burn-in of 100 million steps , yielding a total of 2000 MCMC samples . In all cases , when drift parameter μ was included the MCMC chain mixed well and arrived at the same estimated posterior distribution from different starting points . However , without drift parameter μ , maps for A/H3N2 showed some degree of metastability , where some chains would converge on one solution and other chains would converge on a different solution . We favor models that include μ , because its inclusion , in addition to correcting most identifiability issues , yields much improved mixing of antigenic locations . There is some difficulty in summarizing posterior cartographic samples , as sampled virus and serum locations represent only relative quantities , and because of this , over the course of the MCMC , virus locations may shift . Our prior on virus and serum locations remove much of this issue , orienting the antigenic map along dimension 1 and fixing it to begin at the origin . However , local isometries are often still a problem . For example , in A/H3N2 the HK/68 , EN/72 and VI/75 clusters may rotate in relation to other clusters . Consequently , it may be difficult to fully align MCMC samples using Procrustes analysis . For the present study , we take a simple approach and sample a single MCMC step and visualize the antigenic locations at this state ( Figure 2 , Figure 3 ) . Then , for specific quantities of interest , like rate of antigenic drift and rate of diffusion at different points along the phylogeny , we calculate the quantity across MCMC samples to yield an expectation and a credible interval . This approach accurately characterizes uncertainty that may be hidden in an analysis of a single antigenic map . We summarize diffusion paths of viral lineages ( Figure 2 , Figure 3 ) by taking each virus and reconstructing x and y locations along antigenic dimensions 1 and 2 backward through time . We use MCMC to sample tip locations , but when outputting trees sample internal node locations using a peeling algorithm as described in Pybus et al . ( 2012 ) . Thus , after the MCMC is finished , we have a posterior sample of 2000 trees each tagged with estimated tip locations and internal node locations . We post-processed each posterior tree by conducting a linear interpolation between parent–child node locations to arrive at x and y values at intervals of 0 . 05 years for each virus . Then , for each interval , x and y values are averaged across the sample of posterior trees . We draw lines between these locations to approximate mean posterior diffusion paths . As virus lineages coalesce backwards through time down the phylogeny these diffusion paths will also coalesce . Here , we attempt to compare antigenic locations inferred by our BMDS model to antigenic locations previously inferred by the error minimization methods of Smith et al . ( 2004 ) , referred to here as antigenic cartography by MDS . For this comparison , we use exactly the same HI data used to produce the results in Smith et al . ( 2004 ) , consisting of 273 virus isolates , 79 serum isolates and a total of 4252 HI measurements taken between 1968 and 2003 . We begin with a BMDS analog of the antigenic model used in Smith et al . ( 2004 ) , where serum potencies are taken as the maximum titer of a particular ferret serum and the expected log2 drop in HI titer is proportional to Euclidean distance between virus and serum locations . To bring models into further alignment , we use a Uniform ( −100 , 100 ) distribution over virus locations and serum locations . Unsurprisingly , we find that this BMDS model produces results that are strongly congruent with MDS results ( Figure 10 , Figure 10—source data 1 ) . Antigenic cluster locations are consistent between methods ( Figure 10A–B ) and antigenic distances between pairs of viruses are consistent between temporally similar and temporally divergent viruses ( Figure 10C–E ) , suggesting that the resulting maps are consistent at both local and global scales . Credible intervals of antigenic distances for the BMDS model remain narrow across the temporal spectrum ( Figure 10C–E ) , implying a fair degree of rigidity to the map . 10 . 7554/eLife . 01914 . 016Figure 10 . Comparison of A/H3N2 antigenic locations estimated by Smith et al . ( 2004 ) using MDS and an equivalent BMDS model . ( A ) MDS antigenic locations , reoriented so that the primary dimension lies on the x-axis rather than on the y-axis as in Figure 1 of Smith et al . ( 2004 ) . ( B ) A posterior sample of antigenic locations from an equivalent BMDS model . In ( A ) and ( B ) , viruses are shown as colored circles , with color denoting antigenic cluster inferred by ( Smith et al . , 2004 ) , and sera are shown as gray points . ( C ) Antigenic distances between A/Bilthoven/16 , 190/1968 and all other viruses determined for both methods . ( D ) Antigenic distances between A/Fujian/411/2002 and all other viruses determined for both methods . ( E ) Antigenic distances between 750 random pairs of viruses determined for both methods . In ( C ) , ( D ) and ( E ) red points show distances for the MDS model and gray bars show the 95% credible interval of distances for the BMDS model , while the red dashed line shows a LOESS regression to MDS distances , and the black dashed line shows a LOESS regression to the BMDS distances . The BMDS model has a Uniform ( −100 , 100 ) prior on antigenic locations and serum potencies fixed at maximum titer values . DOI: http://dx . doi . org/10 . 7554/eLife . 01914 . 01610 . 7554/eLife . 01914 . 017Figure 10—source data 1 . This tab-delimited text file lists every virus and serum in Figure 10 , including strain name , year of isolation , coordinates in antigenic dimensions 1 and 2 , and potency for sera . DOI: http://dx . doi . org/10 . 7554/eLife . 01914 . 017 Smith et al . 2004 show that there exist at least two solutions in their assignment of antigenic locations , involving the rotation of clusters HK/68 , EN/72 and VI/75 ( shown in Figure S2 of [Smith et al . , 2004] ) . We observe the same metastable behavior in our analysis; some MCMC chains converge on the solution shown in Figure 10B , while other MCMC chains converge on the alternative solution shown in Figure 11B . The distribution of likelihood values appears highly similar between these two solutions , suggesting that they represent global optima . The rotation of the HK/68 , EN/72 , and VI/75 clusters creates a map that bends slightly , so that temporally distant viruses appear closer in the rotated solution than in the original solution ( Figure 11C–E , Figure 11—source data 1 ) . In this case , it is clear that the solutions are locally consistent between viruses up to ∼15 years divergent , even if there is some degree of global flexibility . 10 . 7554/eLife . 01914 . 018Figure 11 . Comparison of A/H3N2 antigenic locations estimated by Smith et al . ( 2004 ) using MDS and an equivalent BMDS model under an alternative solution . ( A ) MDS antigenic locations , reoriented so that the primary dimension lies on the x-axis rather than on the y-axis as in Figure 1 of Smith et al . ( 2004 ) . ( B ) A posterior sample of antigenic locations from an equivalent BMDS model that has converged on the alternative solution . In ( A ) and ( B ) , viruses are shown as colored circles , with color denoting antigenic cluster inferred by Smith et al . ( 2004 ) , and sera are shown as gray points . ( C ) Antigenic distances between A/Bilthoven/16 , 190/1968 and all other viruses determined for both methods . ( D ) Antigenic distances between A/Fujian/411/2002 and all other viruses determined for both methods . ( E ) Antigenic distances between 750 random pairs of viruses determined for both methods . In ( C ) , ( D ) and ( E ) red points show distances for the MDS model and gray bars show the 95% credible interval of distances for the BMDS model , while the red dashed line shows a LOESS regression to MDS distances , and the black dashed line shows a LOESS regression to the BMDS distances . The BMDS model has a Uniform ( −100 , 100 ) prior on antigenic locations and serum potencies fixed at maximum titer values . DOI: http://dx . doi . org/10 . 7554/eLife . 01914 . 01810 . 7554/eLife . 01914 . 019Figure 11—source data 1 . This tab-delimited text file lists every virus and serum in Figure 11 , including strain name , year of isolation , coordinates in antigenic dimensions 1 and 2 , and potency for sera . DOI: http://dx . doi . org/10 . 7554/eLife . 01914 . 019 As discussed in the main text , the presence of multiple optima with different degrees of 2D curvature implies an issue of identifiability; the HI likelihood model alone cannot distinguish between these possibilities . Because of this issue , and to more easily estimate rates of antigenic drift , we include a model of systematic drift in antigenic location that favors linear movement in the antigenic map . We find that including this drift prior on antigenic locations removes the problem of identifiability . Antigenic locations produced by this model remain locally consistent with MDS results between viruses ∼15 years divergent , but global comparisons show that this BMDS model has partitioned more variance to the first antigenic dimension ( Figure 12 , Figure 12—source data 1 ) . We additionally find that including the drift prior on antigenic locations often results in greater predictive power , with a slight improvement of test error for the A/H1N1 , B/Vic , and B/Yam datasets ( Table 1 ) . 10 . 7554/eLife . 01914 . 020Figure 12 . Comparison of A/H3N2 antigenic locations estimated by Smith et al . ( 2004 ) using MDS and an extended BMDS model that includes date-informed priors on antigenic locations . ( A ) MDS antigenic locations , reoriented so that the primary dimension lies on the x-axis rather than on the y-axis as in Figure 1 of Smith et al . ( 2004 ) . ( B ) A posterior sample of antigenic locations from a BMDS model that includes date-informed priors on antigenic locations . In ( A ) and ( B ) , viruses are shown as colored circles , with color denoting antigenic cluster inferred by Smith et al . ( 2004 ) , and sera are shown as gray points . ( C ) Antigenic distances between A/Bilthoven/16 , 190/1968 and all other viruses determined for both methods . ( D ) Antigenic distances between A/Fujian/411/2002 and all other viruses determined for both methods . ( E ) Antigenic distances between 750 random pairs of viruses determined for both methods . In ( C ) , ( D ) and ( E ) red points show distances for the MDS model and gray bars show the 95% credible interval of distances for the BMDS model , while the red dashed line shows a LOESS regression to MDS distances , and the black dashed line shows a LOESS regression to the BMDS distances . The BMDS model has a date-informed prior on antigenic locations and serum potencies fixed at maximum titer values . DOI: http://dx . doi . org/10 . 7554/eLife . 01914 . 02010 . 7554/eLife . 01914 . 021Figure 12—source data 1 . This tab-delimited text file lists every virus and serum in Figure 12 , including strain name , year of isolation , coordinates in antigenic dimensions 1 and 2 , and potency for sera . DOI: http://dx . doi . org/10 . 7554/eLife . 01914 . 021 Our final BMDS model ( model 9 , Table 1 ) differs from antigenic model used by Smith et al . ( 2004 ) in including temporally- and phylogenetically-informed priors on antigenic locations and also in estimating serum and virus avidities . Here , we investigate the impact on antigenic locations of estimating virus avidity and serum potency in the BMDS model . To isolate this difference , we use a Uniform ( −100 , 100 ) prior on antigenic locations . Surprisingly , estimating virus avidity and serum potency results in a more linear antigenic map ( Figure 13 , Figure 13—source data 1 ) , resembling the appearance of the map incorporating the antigenic drift prior , while preserving local consistency . We generally observe congruence between MDS and BMDS antigenic locations for viruses less than ∼10 years divergent ( Figure 13E ) . However , specific viruses may be affected , for instance A/Bilthoven/16 , 190/1968 ( Figure 13C ) , which appears more distant from all other viruses when serum and virus avidities are included . 10 . 7554/eLife . 01914 . 022Figure 13 . Comparison of A/H3N2 antigenic locations estimated by Smith et al . ( 2004 ) using MDS and an extended BMDS model that estimates serum and virus avidities . ( A ) MDS antigenic locations , reoriented so that the primary dimension lies on the x-axis rather than on the y-axis as in Figure 1 of Smith et al . ( 2004 ) . ( B ) A posterior sample of antigenic locations from a BMDS model that estimates virus avidity and serum potency . In ( A ) and ( B ) , viruses are shown as colored circles , with color denoting antigenic cluster inferred by Smith et al . ( 2004 ) , and sera are shown as gray points . ( C ) Antigenic distances between A/Bilthoven/16 , 190/1968 and all other viruses determined for both methods . ( D ) Antigenic distances between A/Fujian/411/2002 and all other viruses determined for both methods . ( E ) Antigenic distances between 750 random pairs of viruses determined for both methods . In ( C ) , ( D ) , and ( E ) red points show distances for the MDS model and gray bars show the 95% credible interval of distances for the BMDS model , while the red dashed line shows a LOESS regression to MDS distances , and the black dashed line shows a LOESS regression to the BMDS distances . The BMDS model has a Uniform ( −100 , 100 ) prior on antigenic locations and virus avidities and serum potencies estimated in a hierarchical Bayesian fashion . DOI: http://dx . doi . org/10 . 7554/eLife . 01914 . 02210 . 7554/eLife . 01914 . 023Figure 13—source data 1 . This tab-delimited text file lists every virus and serum in Figure 13 , including strain name , year of isolation , coordinates in antigenic dimensions 1 and 2 , potency for sera and avidity for viruses . DOI: http://dx . doi . org/10 . 7554/eLife . 01914 . 023 In this dataset , viruses 15 or more years divergent always yield threshold titers , and hence , their relative locations must be indirectly inferred rather than through direct comparison . This may explain why we observe local consistency between models at scales less than ∼15 years , but some degree of global inconsistency . Still , these results suggest that , when making local comparisons , such as those used to calculate year-to-year antigenic drift ( Figure 3 ) , outcomes are expected to be robust to many model particulars . Source code implementing the cartographic models has been made fully available as part of the software package BEAST ( Drummond et al . , 2012 ) , and can be downloaded from its Google code repository ( http://code . google . com/p/beast-mcmc/ ) . More details on implementing these models can be found at https://github . com/trvrb/flux/tree/master/example-xmls . Incidence data and HI data used in this analysis is archived with Dryad ( doi: 10 . 5061/dryad . rc515 ) . | Every year , seasonal influenza , commonly called flu , infects up to one in five people around the world , and causes up to half a million deaths . Even though the human immune system can detect and destroy the virus that causes influenza , people can catch flu many times throughout their lifetimes because the virus keeps evolving in an effort to avoid the immune system . This antigenic drift—so-called because the antigens displayed by the virus keep changing—also explains why influenza vaccines become less effective over time and need to be reformulated every year . It is possible to determine which antigens are displayed by a new strain of the virus by observing how blood samples that respond to known strains respond to the new strain . This information about the “antigenic phenotype” of the virus can be plotted on an antigenic map in which strains with similar antigens cluster together . Gene sequencing has shown that there are four subtypes of the flu virus that commonly infect people; but the relationship between changes in antigenic phenotype and changes in gene sequences of the influenza virus is poorly understood . Bedford et al . have now developed an approach to combine antigenic maps with genetic information about the four subtypes of the human flu virus . This revealed that the antigenic phenotype of H3N2—a subtype that is becoming increasingly common—evolved faster than the other three subtypes . Further , a correlation was observed between antigenic drift and the number of new influenza cases per year for each flu strain . This suggests that knowing which antigenic phenotypes are present at the start of flu season could help predict which strains of the virus will predominate later on . The work of Bedford et al . provides a useful framework to study influenza , and could help to pinpoint which changes in viral genes cause the changes in antigens . This information could potentially speed up the development of new flu vaccines for each flu season . | [
"Abstract",
"Introduction",
"Results",
"and",
"discussion",
"Materials",
"and",
"methods"
] | [
"evolutionary",
"biology",
"microbiology",
"and",
"infectious",
"disease"
] | 2014 | Integrating influenza antigenic dynamics with molecular evolution |
The Notch signalling pathway plays fundamental roles in diverse developmental processes in metazoans , where it is important in driving cell fate and directing differentiation of various cell types . However , we still have limited knowledge about the role of Notch in early preimplantation stages of mammalian development , or how it interacts with other signalling pathways active at these stages such as Hippo . By using genetic and pharmacological tools in vivo , together with image analysis of single embryos and pluripotent cell culture , we have found that Notch is active from the 4-cell stage . Transcriptomic analysis in single morula identified novel Notch targets , such as early naïve pluripotency markers or transcriptional repressors such as TLE4 . Our results reveal a previously undescribed role for Notch in driving transitions during the gradual loss of potency that takes place in the early mouse embryo prior to the first lineage decisions .
The totipotent mammalian zygote has the self-organising capacity of generating embryonic and extraembryonic structures to build a complete organism ( Wennekamp et al . , 2013 ) . This undifferentiated cell will proliferate and its descendants will take lineage decisions that entail a progressive loss of potency . The first differentiation event that leads to distinct lineages takes place during preimplantation development at the morula to blastocyst transition , resulting in the formation of the trophectoderm ( TE , extraembryonic population ) and the inner cell mass ( ICM , that gives raise to the embryonic population and the extraembryonic yolk sac endoderm ) . How the establishment of these early lineages is achieved has been widely studied and we now know that a combination of morphogenetic cues breaks the symmetry in the embryo ( Cockburn and Rossant , 2010; Menchero et al . , 2018; Sasaki , 2015 ) . The first morphological sign of differentiation is evident in the compacting morula , 2 . 5 days after fertilization ( embryonic day E2 . 5 ) , when blastomeres increase their intercellular interactions and outer cells acquire an apical-basal polarity . These polarized cells on the surface enclose an inner group of apolar cells ( Johnson and Ziomek , 1981; Ziomek and Johnson , 1980 ) . The outer versus inner position of the blastomeres correlates with their fate , becoming TE or ICM respectively , although cells can change their position within the embryo ( Anani et al . , 2014; Tarkowski and Wróblewska , 1967; Watanabe et al . , 2014 ) . Prior to compaction , blastomeres appear morphologically equivalent . However , transcriptional differences among blastomeres have been described as early as in the 4 cell embryo ( Burton et al . , 2013; Goolam et al . , 2016; Torres-Padilla et al . , 2007 ) . Although cells at this stage are not committed to a specific fate , these early heterogeneities correlate with specific fate biases before lineage commitment . However , how these early heterogeneities arise and their implications in cell plasticity are still unclear ( Chen et al . , 2018 ) . Once the embryo compacts , differences in cell membrane contractility and the activity of signalling pathways orchestrate the lineage-commitment of cell populations ( Kono et al . , 2014; Korotkevich et al . , 2017; Maître et al . , 2016; Mihajlović and Bruce , 2016; Nishioka et al . , 2009; Nissen et al . , 2017; Rayon et al . , 2014 ) . The initial stochastic expression of the main lineage-specific transcription factors ( such as CDX2 or GATA3 for the TE , and OCT4 or NANOG for the ICM ) is gradually restricted to their definitive domains ( Dietrich and Hiiragi , 2007; Posfai et al . , 2017 ) . The Hippo pathway has been shown to act as a readout of cell polarity and therefore , differential intercellular distribution of its components and thus differential activity in polar or apolar cells , will dictate fate ( Cockburn et al . , 2013; Hirate et al . , 2013; Leung and Zernicka-Goetz , 2013; Wicklow et al . , 2014 ) . In outer cells , the pathway is switched off and the transcriptional coactivator YAP is translocated to the nucleus where it will interact with TEAD4 , the effector of the pathway , to promote the expression of key TE genes such as Cdx2 and Gata3 ( Nishioka et al . , 2009; Ralston et al . , 2010 ) . We have previously shown that Notch signalling also has a role in the regulation of Cdx2 . It is specifically active in the TE , where the intracellular domain of the Notch receptor ( NICD ) is translocated into the nucleus where it binds to the transcription factor RBPJ to promote target gene expression . Both Notch and Hippo converge on the TEE , an enhancer upstream of Cdx2 ( Rayon et al . , 2014 ) . YAP/TEAD and NICD/RBPJ transcriptional complexes interact with the chromatin modifier SBNO1 to favour the induction of Cdx2 ( Watanabe et al . , 2017 ) . Nevertheless , we still do not understand how these two signalling pathways interact to regulate Cdx2 in the embryo , if there is crosstalk between them , if they are acting in parallel during development or otherwise . Furthermore , Notch signalling could have other unexplored roles at early stages of mouse development . In this study , we show that Hippo and Notch pathways are largely independent , but that Notch is active earlier , before compaction , and that differences in Notch levels contribute to cell fate acquisition in the blastocyst . Single-embryo RNA-seq points at repressors that block early naïve pluripotency markers as Notch targets . We propose that Notch coordinates the triggering of initial differentiation events within the embryo and regulates the early specification of the trophectoderm .
Previously , we have described how Notch and Hippo pathways converge to regulate Cdx2 expression , and that different allelic combinations for Rbpj and Tead4 lead to a significantly reduced expression of CDX2 ( Rayon et al . , 2014 ) . Notably , we failed to recover double mutant embryos at the blastocyst stage ( E3 . 5 ) , suggesting that the lack of both factors caused lethality before the blastocyst stage . We therefore decided to investigate embryos at the earlier morula stage ( E2 . 5 ) , where we recovered double mutant embryos at Mendelian ratios ( Figure 1—figure supplement 1A ) . CDX2 levels were apparently lower in Rbpj-/-;Tead4+/- and Rbpj+/-;Tead4-/- morulae , as previously observed in blastocysts ( Rayon et al . , 2014 ) . Interestingly , this effect was exacerbated in double mutant embryos ( Rbpj-/-;Tead4-/- ) in which we did not detect any CDX2 expression in all embryos of this genotype analysed ( Figure 1A , Figure 1—figure supplement 1B ) . Compaction of blastomeres and polarization of outer cells are critical morphological events that take place at the morula stage and are linked to the onset of CDX2 expression ( Ralston and Rossant , 2008; Wu et al . , 2010 ) . We therefore decided to investigate if these processes were affected in double mutant morulae . We examined the expression of E-cadherin and phospho-ERM , as markers of cell-cell adhesion and apical polarity . No differences in the distribution or intensity of these markers was observed in any of the allelic combinations examined , including double mutants for Rbpj and Tead4 ( Figure 1—figure supplement 1C ) . Therefore , disruption of Notch and Hippo signalling does not alter cellular and morphological events prior to lineage specification , but does result in a dramatic downregulation of Cdx2 at this stage . To better understand the contributions of each of the Notch and Hippo pathways to CDX2 expression , we performed correlations in single cells between Notch , YAP and CDX2 . We used a transgenic mouse line carrying CBF1-VENUS as a reporter of Notch activity ( Nowotschin et al . , 2013 ) , and we performed immunostaining to detect YAP and CDX2 in morulae and blastocysts from that reporter line . In the blastocyst , the three markers were restricted to nuclei of the TE , while in the morula their expression was more heterogeneous ( Figure 1B ) . Nuclear YAP was detected preferentially in outer cells , presumably polarized blastomeres , whereas CBF1-VENUS and CDX2 were detected in both inner and outer cells of the morula . We quantified nuclear intensity levels using a Matlab based segmentation tool , MINS ( Lou et al . , 2014 ) , and found that CBF1-VENUS and YAP both correlated positively with CDX2 at early morula ( 8–16 cells ) , late morula ( 17–32 cells ) and blastocyst stages ( Figure 1C ) . This correlation was low in the earliest embryos but gradually increased . Interestingly , there was no correlation between CBF1-VENUS and YAP in the early or late morula , suggesting that the two pathways are activated independently at this stage ( Figure 1C ) . By the blastocyst stage , these markers did show a positive correlation , albeit weaker than the correlation of either marker with CDX2 ( Figure 1C ) . If all three components were taken into account simultaneously , the coefficient of correlation increases in early and late morulae and blastocyst ( Figure 1—figure supplement 2 ) , indicating that the combination of Notch and Hippo pathways better accounted for CDX2 levels than any of them individually . In most cases , individual nuclei from morulae were positive for the three markers . However , we did find a few cases in which nuclei were positive for CBF1-VENUS and CDX2 but negative for YAP ( Figure 1B , arrowhead ) . We therefore analysed all morulae to determine the distribution of cells positive for each combination of markers . We found that Notch was active in most of the cells at this stage and that the majority of blastomeres were positive for all three of the markers ( 295 blastomeres , 72 . 3%; Figure 1D ) . Another noteworthy population was represented by cells that were only positive for CBF1-VENUS and CDX2 ( 85 blastomeres , 20 . 8% ) . However , we rarely found cells expressing YAP and CDX2 but not CBF1-VENUS at the morula stage ( Figure 1D ) . Together , this set of experiments shows that Notch and Hippo correlate with CDX2 expression at the morula stage , and suggests that they could be acting independently from each other in its regulation . The correlation analysis between CBF1-VENUS and YAP expressing blastomeres indicated possible independent roles for Notch and Hippo in the regulation of CDX2 expression ( Figure 1C ) . Furthermore , our previous results showed how these two pathways acted in parallel to transcriptionally regulate Cdx2 through a distal enhancer element ( Rayon et al . , 2014 ) . To further study the interaction between these pathways , we examined TEAD4 and YAP expression in Rbpj-/- ( Figure 1—figure supplement 3A , B ) and Notch1-/- ( Figure 1—figure supplement 3C , D ) blastocysts . We did not detect any differences in levels or pattern of expression of TEAD4 and YAP either in Rbpj-/- or in Notch1-/- embryos as compared to wildtype embryos . We also studied the reverse situation , crossing the CBF1-VENUS mouse line as a reporter of Notch pathway activity into the Tead4 null background . We detected VENUS fluorescent protein in both wildtype and Tead4-/- embryos ( Figure 1—figure supplement 4 ) . Interestingly , CBF1-VENUS expression was maintained in outer cells although the Tead4-/- embryos do not form a proper blastocyst ( Nishioka et al . , 2008 ) , in line with previous results showing that some degree of outer identity still is present in Tead4-/- embryos ( Yagi et al . , 2007; Nishioka et al . , 2008; Frum et al . , 2018 ) . These results confirm that Notch is not required for proper deployment of the transcriptional effectors of the Hippo pathway , and vice versa , that activation of the Notch pathway can occur in embryos deleted for Tead4 , one of these effectors . To better understand how parallel signalling pathways drive Cdx2 expression , we determined if the temporal expression of Cdx2 was regulated differentially by Notch and Hippo . To do so , we took advantage of pharmacological compounds that allow inhibition of these pathways in a time-controlled manner . We used RO4929097 ( RO ) to inhibit the Notch pathway ( Münch et al . , 2013 ) and Verteporfin to block the YAP-TEAD4 interaction ( Liu-Chittenden et al . , 2012 ) . We treated wildtype embryos in two different time-windows: from the two-cell up to morula stage , and from morula to blastocyst . As a control , we treated embryos with DMSO , the solvent used for diluting both inhibitors . This was important , as high doses of DMSO result in embryo lethality , what also limits the concentration of inhibitors to be used . We confirmed that treatment with RO in both time-windows was affecting Notch signalling as expected , as it visibly reduced CBF-VENUS activity in embryos ( Figure 2—figure supplement 1A , B ) . Efficacy of Verteporfin in interfering with YAP-TEAD4 activity was inferred by its effect on known targets , as we have previously shown in blastocyst for Cdx2 ( Rayon et al . , 2014 ) . After treatment , gene expression in embryos was analysed by RT-qPCR . In the early time window , from two-cell to morula , we observed that Cdx2 was downregulated when Notch was inhibited , while there was no change when Hippo pathway activity was altered ( Figure 2A ) . Interestingly , the opposite was found when we modulated the pathways from the morula onwards . Cdx2 expression was only affected when YAP-TEAD4 activity was blocked ( Figure 2B ) . These results show that , although both pathways cooperate in the regulation of Cdx2 , they act sequentially to regulate Cdx2 levels in a stage specific manner rather than being redundant . Interestingly , Gata3 , a known target of YAP-TEAD4 independent of Cdx2 ( Ralston et al . , 2010 ) is downregulated by Verteporfin in embryos treated from the two-cell to morula stage ( Figure 1A ) . This suggests that certainly Hippo signalling acts differently on its various targets , as are Cdx2 or Gata3 . Oct4 and Nanog were not significantly changed after Notch or YAP inhibition in any of the time windows . Next , we wished to confirm these observations in morula stage embryos using genetic loss of function models . We recovered early ( 8–16 cells ) and late ( 17–32 cells ) morulae and analysed CDX2 expression in wildtype and Rbpj-/- embryos ( Figure 2C–E ) . We found that Rbpj-/- early morulae had a significantly lower level of nuclear CDX2 expression ( Figure 2D ) and number of CDX2 positive cells compared to control littermates ( Figure 2E ) . In contrast , we did not observe differences at the late morula stage ( Figure 2F–H ) . The same observations were obtained when we analysed embryos from another mutant for the pathway , Notch1-/-: early morulae ( 8–16 cells ) showed lower CDX2 expression and a decrease in the number of CDX2 positive cells ( Figure 2—figure supplement 1A–C ) , but late ( 17–32 cells ) morulae did not ( Figure 2—figure supplement 1D–F ) . This result is interesting , as it suggests that Notch1 is the main receptor acting upstream of RBPJ during preimplantation development as its loss is enough to recapitulate Rbpj loss of function effects . These results indicate that there is an earlier requirement for Notch than for Hippo in the regulation of Cdx2 , and that both pathways exert non-redundant roles . Our observations are suggestive of a model where Notch regulates the onset of Cdx2 expression , and the Hippo pathway subsequently maintains its expression . In light of the above findings revealing a requirement of the Notch pathway for the early stages of mouse preimplantation development , we decided to investigate when Notch is first active , using the CBF1-VENUS reporter line as a transcriptional readout of the pathway . We recovered embryos from the CBF1-VENUS line and found that the reporter was first active in 4 cell embryos , albeit at lower levels than at later stages ( Figure 3A ) . The number of VENUS positive cells was variable among embryos , with at least a third of embryos examined having no positive cells ( 7 out of 20; Figure 3—figure supplement 1 ) . This strongly suggests that the onset of Notch pathway activation is indeed occurring at this stage . As a general rule , the number of positive blastomeres increased with the total number of cells per embryo ( Figure 3B; Figure 3—figure supplement 1 ) . In the compacted morula , most of the cells were positive , but the activity of the reporter was quickly restricted to the outer TE cells once the blastocyst formed ( Figure 1B ) . In order to follow the dynamics of the reporter and determine how restriction of Notch activity is achieved during development , we performed live imaging for up to 24 hr of embryos from the compacted morula ( 16 cell ) to the early blastocyst stage ( Video 1 , Figure 3C ) . After tracking of the cells in each embryo ( n = 7; Figure 3—figure supplement 2A ) , we used a Matlab based tool to analyse the behaviour of each individual cell and its progeny within the embryo . With this tool , we were able to reconstruct the embryo in each time point and assign an initial position ( inner or outer ) to each blastomere as well as its final location in the TE/out or the ICM/in ( Video 2 , Figure 3D ) . We first generated a lineage tree so that each lineage or family includes a cell in the time frame 0 and all their descendant cells . We next classified families according to the position of the cells in the first and final time points . This allowed us to divide the cells in four groups: ‘IN-ICM’ ( cells that began in an inner position and their descendants remained in an inner position ) , ‘IN-TE +ICM’ ( cells that began in an inner position and at least one of their descendants ended up in an inner position but other/s in an outer position ) , ‘OUT-TE’ ( cells that began in an outer position and their descendants remained in an outer position ) , and ‘OUT-TE +ICM’ ( cells that began in an outer position and at least one of their descendants ended up in an outer position but other/s ended up in an inner position ) . Confirming previous findings ( Anani et al . , 2014; McDole and Zheng , 2012; Posfai et al . , 2017; Toyooka et al . , 2016; Watanabe et al . , 2014 ) , although most of the cells of the blastocyst retain the position of their predecessor cell in the compacted morula , a small percentage change their location ( Figure 3—figure supplement 2B ) . We next measured intensity levels of the reporter in all cells within families , and determined if it correlated with their position during the time lapse . Notch activity levels were variable among families and embryos , but we detected higher and increasing levels in OUT-TE families while IN-ICM families generally showed lower and decreasing levels ( Figure 3—figure supplement 2C–D ) . The intensity levels in families that contributed both inner and outer cells did not follow a clear pattern ( Figure 3—figure supplement 2E ) . When we analysed the mean intensity for each group , we saw that VENUS levels were significantly lower in the families that were always inside as compared to the families that were always outside ( Figure 3E ) . Interestingly , this difference was already manifest when we measured the initial intensity in the first time point ( Figure 3F ) . In the families whose cells end up in both inner and outer position , VENUS levels were intermediate ( Figure 3E , F ) . Therefore , the analysis of the CBF1-VENUS line showed that the reporter is active before the first lineage decision is taken , and that differences in the levels of pathway activation in inner or outer cells of the compacted morula correlate with the final position of their descendants in the blastocyst . We have previously shown that increasing the activity of the Notch signalling pathway leads to a preferential allocation of cells to the outer trophectoderm of the blastocyst ( Rayon et al . , 2014 ) . However , we had not tested the onset of this effect and whether blocking Notch would have an effect in early embryos . To address these questions , we used a genetic mosaic line ( iChr‐Notch‐Mosaic ) that allowed us to generate cells with different Notch activity levels within the same embryo ( Pontes-Quero et al . , 2017 ) . The construct consists of three different cassettes preceded by a specific LoxP site . The first cassette is a H2B-CHERRY fluorescent protein and generates wildtype cells . The second cassette contains a dominant-negative version of Mastermind-like 1 ( DN-MAML1 ) , a transcriptional co-activator of the Notch pathway , linked to a H2B-eGFP by a cleavable 2A peptide , whose expression leads to the loss of function ( LOF ) of the pathway , while the third is a gain of function ( GOF ) cassette through the expression of a constitutively active NICD linked to an HA-H2B-Cerulean ( Figure 4A ) . The specific LoxP sites are mutually exclusive , so in any unique cell there will be only one possible outcome as the result of Cre-mediated recombination . We used a Polr2aCreERT2 driver which is ubiquitously expressed and inducible by tamoxifen ( Guerra et al . , 2003 ) . We induced recombination by adding 4OH-Tx ( 4-hydroxy-tamoxifen ) from the 2- to the 4-cell stage , aiming to achieve a situation where cells expressing each cassette derive from a single recombined blastomere , and we evaluated recombination in the late morula ( <32 cells ) or in the blastocyst ( Figure 4A ) . We performed immunofluorescent assays with three antibodies to distinguish the three cassettes . The wildtype cassette was detected by an anti-RFP antibody , the LOF by an anti-GFP antibody , and the GOF an anti-HA antibody . However , GOF cells were triple positive because of cross-reactivity between antibodies and the HA-H2B-Cerulean protein ( Figure 4A , B ) . To validate this strategy , we induced recombination in a iChr‐Notch‐Mosaic ES cell line , and sorted cells according to their fluorescence . Testing the antibodies described above in these populations confirmed that they correctly identified cells derived from each different recombination event ( Figure 4—figure supplement 1A ) . Confirmation that we were modifying Notch signalling as predicted with the LOF and GOF cassettes came from analysing nuclear CDX2 levels , as a readout of Notch activity . Notch LOF blastomeres showed lower CDX2 levels than wildtype and GOF , and these higher levels than wildtypes ( Figure 4—figure supplement 1B ) . We selected embryos in which at least two recombination events leading to LOF and GOF cells had occurred , and analysed the percentage of unrecombined cells ( 36% ) , and cells expressing the control , LOF or GOF cassette . Although the probabilities of recombination are higher when the LoxP sites are closer to one another ( the control recombination event in this case ) , we found that most of the recombined cells were Notch GOF ( 36% of total cells ) while only a small proportion ( 11% ) were Notch LOF , ( Figure 4C; Figure 4—figure supplement 2A , B ) . If we used a control line ( iChr-Control-Mosaic; Pontes-Quero et al . , 2017 ) carrying the same construct and reporters but not the LOF or GOF cassettes ( Figure 4—figure supplement 3A ) , we observed a similar proportion of unrecombined cells ( 34% ) but in this case the most probable event ( red cells ) was the most abundant ( 34% ) , as expected ( Figure 4—figure supplement 3B ) . These results suggest that Notch activity could differentially affect cell proliferation or cell loss in the embryo . Next , we determined the proportion of cells from each population that were in an inner or outer position . Approximately 60% of unrecombined or wild type ( red ) cells , both being controls , were located at outer positions in both morula and blastocyst stage . However , Notch-LOF cells ( green ) were enriched at inner positons of the morula or more clearly in the inner cell mas of the blastocyst , while Notch-GOF cells ( blue ) tended to occupy outer positions ( Figure 4D , E ) . These differences were not observed when we analysed embryos from the iChr-Control-Mosaic line ( Figure 4—figure supplement 3C ) . These experiments show how manipulating levels of Notch pathway activity as early as the 4-cell stage instructs cells to adopt an inner or outer position at later stages . Results described above show that the Notch pathway plays an early role in mouse development , non-redundant with that of the Hippo pathway , in regulating Cdx2 gene expression and in determining the position of cells to inner or outer locations . To gain further insight into how Notch is acting during preimplantation development , we carried out RNA-sequencing ( RNA-seq ) in control and Rbpj-/- single morulae . 2028 genes were differentially expressed ( Figure 5—source data 1 ) , close to 70% of which were downregulated suggesting that Rbpj is mainly activating gene expression at the morula stage . Among the downregulated genes we found Cdx2 and other TE associated genes such as Gata3 or Fgfr2 ( Haffner-Krausz et al . , 1999; Home et al . , 2017; Home et al . , 2009; Ralston et al . , 2010 ) ; genes related with the Hippo pathway ( Nf2 , Amotl2 , Lats2 ) and , interestingly , also genes related with the embryonic pluripotency network including Sall1 , Sall4 , Tbx3 or Sox21 ( Goolam et al . , 2016; Han et al . , 2010; Karantzali et al . , 2011; Niwa et al . , 2009; Yang et al . , 2010 ) ( Figure 5A , Figure 5—figure supplement 1A ) . Among the upregulated genes , we found Dppa3 ( Stella ) and Prdm14 , which have been characterised as naïve pluripotency markers ( Hayashi et al . , 2008; Yamaji et al . , 2013 ) . In addition , a large set of chromatin modifiers were differentially expressed ( Figure 5—figure supplement 1B ) . Important chromatin dynamics have been reported during preimplantation development ( Burton and Torres-Padilla , 2014 ) , which could fit in with the broad mis-regulation of transcription in the mutant embryos . Remarkably , some of the downregulated modifiers like Dnmt3b or Kdm6a have been shown to be enriched in TE conversely to Prdm14 ( Burton et al . , 2013 ) . Overall , the transcriptome profiling suggests that embryos lacking Rbpj do not properly trigger trophectoderm differentiation programs , and that they also affect pluripotency related genes . To identify direct targets of Notch signalling at this stage , we searched for putative RBPJ binding sites in the vicinity of differentially expressed genes . We established an arbitrary window of 10 Kb surrounding each gene to perform the analysis and found RBPJ binding motifs in 1487 genes . We then examined how many of these putative binding sites were located in regions of open chromatin , a hallmark for active regulatory elements . For this , we took advantage of ATAC-seq profiles from published datasets of 8 cell mouse embryos ( Wu et al . , 2016 ) , and reduced our list to 282 genes ( Figure 5B; Figure 5—source data 2 ) . Among these was Cdx2 , where the predicted RBPJ sites and ATAC-seq open chromatin signature mapped to the TEE enhancer we had previously characterised ( Rayon et al . , 2014 ) , thus validating this approach . We selected two genes as putative Notch targets , that were downregulated in Rbpj-/- morulae and had been previously associated with exit from pluripotency in mouse ES cells: those coding for the Groucho-family transcriptional repressor TLE4 ( Laing et al . , 2015 ) , and the T-box family transcription factor TBX3 ( Russell et al . , 2015; Waghray et al . , 2015 ) . Both genes are heterogeneously expressed in ES cells and repress naïve pluripotency genes . We hypothesized that Tle4 and Tbx3 could be direct targets of Notch , and that their downregulation could in part explain the blockade in differentiation that we observe in the RNA-seq . We independently confirmed downregulation of their expression after blocking the Notch pathway by treating wildtype embryos with the RO4929097 inhibitor from 2 cell to morula stage ( Figure 5C ) . An RBPJ motif search within ATAC-seq peaks in the vicinity of the genes identified two potential candidate regions located 1 . 3 Kb upstream of Tle4 ( Tle4-up; Figure 5D ) and in the seventh intron of Tbx3 ( Tbx3-i7; Figure 5—figure supplement 2A ) , respectively . By means of transient transgenic assays ( Rayon et al . , 2014 ) , we proved that these regions could act as transcriptional enhancers driving H2B-mRFP reporter expression in the morula ( 32% positive embryos for the 700 bp Tle4-up element , Figure 5E , F; and 56% for the 600 bp Tbx3-i7 element; Figure 5—figure supplement 2B , C ) . To test if Notch was directly involved , we mutated the RBPJ motif inside these regions and found that the activity of the Tle4-up mutRBPJ fragment was significantly diminished ( from 32% to 13% positive embryos , Figure 5E , F ) while the Tbx3-i7 mutRBPJ fragment was not affected ( 60% positive embryos , Figure 5—figure supplement 2B–C ) . Finally , to examine whether these enhancers were necessary for the expression of their putative target genes , we deleted the regions within the enhancers that contained the RBPJ motif by CRISPR/Cas9 mediated genome editing ( Ran et al . , 2013 ) , and analysed gene expression by qPCR on individually edited morulae . For this we used two guide-RNAs flanking the regions that contained the putative RBPJ binding site , what generated deletions of approximately 150 bp ( Figure 5—figure supplement 2E , F ) . We observed a significant decrease in Tle4 expression in edited embryos ( deleted , n = 10 ) as compared to injected embryos that had been partially ( mosaic , n = 9 ) or not ( wildtype , n = 14 ) edited ( Figure 5G ) . However , Tbx3 expression did not change when the RBPJ motif from the seventh intron was deleted ( Figure 5—figure supplement 2D ) . These assays provide evidence that these genomic regions act as cis-regulatory elements and , in the case of Tle4 , are directly regulated by RBPJ and necessary for correct expression . The transcriptomic profiling carried out in Rbpj-/- embryos identified genes related with naïve pluripotency among the upregulated genes . Naïve pluripotency corresponds to a state in which cells are not prone to differentiate , in contrast to primed pluripotency ( Kalkan and Smith , 2014 ) . These pluripotent states as well as the transition between them have been extensively studied in ES cells and EpiLCs , in vitro counterparts of the epiblast of the blastocyst stage preimplantation embryo and the postimplantation pre-gastrulating epiblast respectively ( Hackett and Surani , 2014 ) . Interestingly , some of these naïve markers such as Prdm14 are initially expressed at the 2- and 4-cell stage , switched off in the morula and re-expressed in the ICM of the blastocyst ( Burton et al . , 2013 ) . Analysis of published single-cell RNA-seq data ( Goolam et al . , 2016 ) confirmed that Prdm14 decreased dramatically from the 4- to 8-cell stage , and expression of Dppa3 also decreases from the 2- to the 4-cell stage ( Figure 6—figure supplement 1 ) . In contrast , Tle4 and Tbx3 levels increased from the 4- to 8-cell stage ( Figure 6—figure supplement 1 ) . Our data from Rbpj-/- morulae suggests that embryos do not switch off Prdm14 and Dppa3 , and inhibiting Notch with RO4929097 from the 2- to 4-cell stage confirmed the effect on Prdm14 , whose levels were significantly increased after the treatment ( Figure 6—figure supplement 2 ) . We wondered if the effect of Notch guiding differentiation programs that we had seen in the embryo was also occurring in ES cells . We used iChr-Notch-Mosaic ES cells ( Pontes-Quero et al . , 2017 ) to confront populations with different Notch levels using the same strategy than we had previously used in the embryo ( Figure 6A ) . After recombination by transfection with Cre , ES cells were sorted according to the fluorescent reporter cassette they expressed ( Figure 6B ) . We measured expression levels of naïve pluripotency markers by qPCR , and found that levels of Prdm14 and Dppa3 correlated negatively with Notch activity but other markers such as Nanog or Esrrb were not altered ( Figure 6C ) . We next asked how Notch would affect the differentiation potential of pluripotent cells using this system . For that , we allowed sorted iChr-Notch-Mosaic ES cells grown in serum +LIF to differentiate for 48 hr after LIF removal and analysed the expression of genes related to early differentiation at different time points ( Figure 6D ) . On the one hand , we observed that the peak of expression of Tle4 , and the early epiblast markers Fgf5 and Pou3f1 occurred earlier and remained at higher levels in Notch GOF than in wildtype ES cells . On the other hand , Notch LOF cells never reached normal levels of Tbx3 or Fgf5 during the differentiation process ( Figure 6E ) . These results suggest that Notch is not only sufficient to drive expression of some differentiation markers such as Tle4 , but also necessary to achieve proper levels of others such as Tbx3 . However , modulation of Notch levels is not enough to change expression of pluripotency markers once ES cells have started the differentiation process ( Figure 6—figure supplement 3 ) . If we carried out the experiment but using ES cells maintained under naïve conditions ( 2i + LIF ) , we observed similar dynamics for Tle4 and Tbx3 and other early epiblast markers , but no changes in later differentiating genes ( Figure 6—figure supplement 4 ) . Overall , our results suggest that Notch is involved in coordinating exit from pluripotency and promoting cell differentiation in ES cells , mirroring its role in the early embryo .
During the first three days of mouse embryonic development , cells lose their totipotent capacity as they form the first differentiated population , the trophectoderm ( TE ) . In this study , we show that Notch signalling regulates the early expression of Cdx2 , a key element in TE specification , and that this is later reinforced by the input of Hippo signalling through YAP and TEAD4 . Hippo has been shown to act as a readout of cell polarity ( Anani et al . , 2014; Hirate et al . , 2015 ) and it activates Cdx2 in cells that have established an apical domain . However , the initial triggering of Cdx2 both in inner and outer cells ( Dietrich and Hiiragi , 2007; Posfai et al . , 2017 ) suggested that inputs other than Hippo would initially be acting because its expression could not be explained only by YAP/TEAD4 activity . In fact , previous reports have described that although in most Tead4-/- blastocysts CDX2 is not detected , earlier Tead4-/- morulae retain CDX2 expression ( Nishioka et al . , 2008 ) . Similarly , double Wwtr1;Yap1 mutant embryos show some residual CDX2 expression ( Frum et al . , 2018 ) . In agreement with these observations , we found blastomeres in the morula that express CDX2 but do not have nuclear YAP . In this situation , expression of CDX2 is likely due to Notch activity as the CBF1-VENUS reporter , used as a proxy for activity of the pathway ( Nowotschin et al . , 2013 ) , is present in those cells . The analysis of Rbpj and Notch1 mutants in early and late morulae , as well as pharmacological treatments of preimplantation embryos , further support the notion that the input provided by Notch is necessary for the early phases of Cdx2 expression . These results show that Notch and Hippo have non-redundant but partially overlapping roles in early and late phases of Cdx2 expression , respectively . Furthermore , only double knockout morulae for Rbpj and Tead4 completely lack CDX2 , and all CDX2 positive cells have at least one of the two pathways active . These findings support a model whereby overlapping or complementary inputs from different signalling pathways may provide robustness in the system , buffering any disturbances and ensuring proper development ( Menchero et al . , 2017 ) . In such a model , Notch and Hippo would ensure the correct specification and maintenance of the TE respectively ( Rayon et al . , 2014 ) . The crosstalk between YAP and Notch has been studied in different cellular contexts ( Totaro et al . , 2018 ) . YAP acts upstream of Notch in controlling epidermal stem cell fate or liver cell fate ( Totaro et al . , 2017; Yimlamai et al . , 2014 ) while Notch is upstream of YAP in the corneal epithelium during chronic inflammation ( Nowell et al . , 2016 ) . Also , YAP and Notch can cooperate to control the onset of oscillations in the segmentation clock ( Hubaud et al . , 2017 ) and they interact to promote the expression of Jag1 in smooth muscle cells ( Manderfield et al . , 2015 ) . During TE establishment , YAP and Notch have also been shown to interact through SBNO1 , and act synergistically to regulate Cdx2 ( Watanabe et al . , 2017 ) . In this context , our results show that both pathways are acting in parallel since there is no correlation among YAP and CBF1-VENUS expression levels in single blastomeres in morula stage embryos . In addition , loss of the NOTCH1 receptor or RBPJ does not affect YAP/TEAD4 localisation and vice versa , Tead4 knockout does not alter CBF-VENUS expression in the blastocyst . Nevertheless , several components of the Hippo pathway are downregulated in Rbpj-/- morulae , so we cannot rule out the possibility of cross-transcriptional regulation between the pathways . The role of Notch signalling in the specification of cell fates during development has been widely studied ( Koch et al . , 2013 ) . Notch promotes heterogeneities and reinforces differences between neighbouring cells , explaining the segregation of cell fates in multiple processes and in different species ( Artavanis-Tsakonas et al . , 1999 ) . The heterogeneous activity of CBF-VENUS in the 4-cell stage coincides with the loss of cell equivalence and emergence of differences among blastomeres . Other factors have been shown to be differentially expressed among blastomeres of the 4 cell mouse embryo ( Burton et al . , 2013; Goolam et al . , 2016 ) , suggesting that this is the moment when cells lose their homogeneous state to start desynchronizing and differentiating . Interestingly , Prdm14 , one of these factors , and Notch show divergent patterns of expression during development . Prdm14 is first expressed at the 2- and 4-cell stage , to be turned off and then re-expressed in the ICM of the blastocyst and later in the primordial germ cells ( Burton et al . , 2013; Yamaji et al . , 2008 ) . In contrast , the Notch pathway , as revealed by the CBF-Venus reporter , begins to be active at the 4-cell stage , it is active in most of the cells of the morula , and is later restricted to the TE of the blastocyst . After implantation , Notch activity is detected throughout the epiblast ( Nowotschin et al . , 2013 ) . It has been suggested that Prdm14 expression coincides with conditions where groups of cells show an undetermined state , while Notch is activated when cells transition towards their next developmental phase . Our results suggest that Notch would be regulating these transitions by downregulating Prdm14 expression . In line with the upregulation of Prdm14 in embryos that lack Notch activity , we observed in the RNA-seq data from Rbpj-/- morulae a downregulation of Fgf receptors ( Fgfr1 , 2 and 3 ) and DNA methyltransferases ( Dnmt3b , Dnmt1 ) , which are known to be repressed by PRDM14 ( Grabole et al . , 2013; Yamaji et al . , 2013 ) . It is also interesting to note that in our mosaic ES cell experiments , Notch levels correlate with those of Prdm14 and Dppa3 , but not with other pluripotency markers such as Nanog or Esrrb . Therefore , Notch is not simply turning off the general pluripotency network to promote differentiation , but acting on a subset of early naïve pluripotency markers . Interplay between Notch and chromatin remodellers has been reported in several situations ( Schwanbeck , 2015 ) . Expression changes in chromatin modifiers precede the action of transcription factors that consolidate lineage choices during preimplantation development ( Burton et al . , 2013 ) . Therefore , these alterations suggest that Rbpj-/- embryos do not established correct epigenetic landscapes , do not switch off early markers such as Prdm14 or Dppa3 and are not able to properly trigger differentiation programs leading to a delay in the expression of lineage specifiers such as Cdx2 . In this regard , it is interesting to note that Rbpj mutant morulae downregulate Chaf1a , which encodes the large subunit of the histone-chaperone CAF-1 . Loss of CAF-1 promotes ES cells to transit to an earlier , totipotent 2-cell-like state ( Ishiuchi et al . , 2015 ) , and acts as a barrier for reprogramming ( Cheloufi et al . , 2015 ) . Furthermore , knockout of Chaf1a leads to developmental arrest at the 16-cell stage and a loss of heterochromatin ( Houlard et al . , 2006 ) . Thus , CAF-1 acts as a driver of differentiation in pluripotent cells . Interestingly , studies in Drosophila have shown that CAF-1 mediates downstream effect of the Notch pathway ( Yu et al . , 2013 ) . On the other hand , Asf1a , which encodes another histone chaperone , is among the few genes observed to be upregulated in Rbpj-/- embryos . Forced expression of Asf1a promotes reprogramming of human ES cells ( Gonzalez-Muñoz et al . , 2014 ) , revealing a critical role in maintaining pluripotency . Furthermore , Suv39h1 , a regulator of H3K9me3-heterochromatin that restrict cell plasticity and stemness ( Yadav et al . , 2018 ) , is also downregulated in Notch loss-of-function morulae . In conclusion , we observed that during preimplantation development , Notch regulates critical epigenetic components that mediate transitions along the progressive restriction of potency that occurs in the early embryo . In this study , we have also identified novel putative targets positively regulated by the Notch pathway , such as Tle4 and Tbx3 whose role in the exit from pluripotency has been described in ES cells ( Laing et al . , 2015; Russell et al . , 2015; Waghray et al . , 2015 ) . Their increase in expression from 2 cell to morula supports their possible role in promoting early differentiation in vivo as well . TLE4 does not bind directly to DNA , but associates with other proteins to act as a transcriptional corepressor ( Kaul et al . , 2015 ) . It will be of great interest to identify its transcriptional partners during preimplantation development and elucidate the mechanism by which it allows cell differentiation in this context . The role of TBX3 is more complex since , in addition to promoting differentiation , it has also been associated with pluripotency maintenance ( Han et al . , 2010; Niwa et al . , 2009 ) . Furthermore , in vivo TBX3 is detected in most of the cells of the morula but it is later restricted to the ICM ( Russell et al . , 2015 ) , following a complementary pattern to Notch . Thus , Tbx3 regulation must involve Notch-dependant and Notch-independent inputs , what could explain why the mutation or deletion of the RBPJ motif present in the intronic Tbx3 regulatory element did not disrupt enhancer activity or endogenous expression . The role of Notch in ES cells had already been explored in the context of neural differentiation ( Lowell et al . , 2006 ) . Blocking Notch signalling prevents ES cells from adopting a neural fate while its overexpression increases the frequency of neural specification . Our results suggest that Notch might have a more general role in promoting early differentiation , with a more specific function in neural specification at later stages ( Lowell et al . , 2006 ) . In summary , our findings suggest that Notch acts by promoting the gradual loss of potency in the early embryo , which is subsequently reinforced by additional mechanisms , such as heterochromatin formation before the morula stage , or differential activation of the Hippo pathway at the morula-to-blastocyst transition . Therefore , in order to correctly specify a given lineage , such as the trophectoderm , Notch is simultaneously activating fate choice markers such as Cdx2 and inducing a differentiation-prone state by lowering levels of naïve markers .
The following mouse lines were used in this work: CBF1-VENUS ( Nowotschin et al . , 2013 ) , Rbpj null ( Oka et al . , 1995 ) , Tead4 null ( Nishioka et al . , 2008 ) , Notch1 null ( Conlon et al . , 1995 ) , iChr-Notch-Mosaic ( Pontes-Quero et al . , 2017 ) , iChr-Control-Mosaic ( Pontes-Quero et al . , 2017 ) , Polr2aCreERT2 ( Guerra et al . , 2003 ) . All the lines were maintained in heterozygosis in an outbred background . Adults were genotyped by PCR of tail-tip DNA using primers and conditions previously described for each line . For preimplantation embryos , genotyping was performed directly on individually isolated embryos after recovery , culture or antibody staining . Mice were housed and maintained in the animal facility at the Centro Nacional de Investigaciones Cardiovasculares ( Madrid , Spain ) in accordance with national and European Legislation . Procedures were approved by the CNIC Animal Welfare Ethics Committee and by the Area of Animal Protection of the Regional Government of Madrid ( ref . PROEX 196/14 ) . Females from the different mouse lines or outbred CD1 were superovulated as previously described ( Behringer et al . , 2014 ) , except in the case of embryos to be used for RNA-seq . For embryo culture , zygotes were collected from oviducts , treated with hyaluronidase ( Sigma ) to remove cumulus cells and cultured until the desired stage at 37 . 5°C , 5% CO2 , in M16 medium ( Sigma ) covered with mineral oil ( NidOil , EMB ) . For experiments that did not require culture , embryos were collected at morula or blastocyst stage by flushing the oviduct or the uterus with M2 medium ( Sigma ) and fixed . Immunofluorescence was performed as previously described ( Dietrich and Hiiragi , 2007 ) . The following antibodies and dilutions were used: monoclonal mouse anti-CDX2 ( MU392-UC , BioGenex ) 1:200 , rabbit monoclonal anti-CDX2 ( ab76541 , Abcam ) 1:200 , mouse monoclonal anti-YAP ( sc-101199 , Santa Cruz Biotechnology ) 1:200 , rabbit polyclonal anti-pERM ( 3141 , Cell Signalling ) 1:250 , rat monoclonal anti-E-Cadherin ( U3254 , Sigma ) 1:250 , mouse monoclonal anti-TEAD4 ( ab58310 , Abcam ) 1:100 , rabbit polyclonal anti-DsRed ( 632496 living colors Clontech ) 1:400 , goat polyclonal anti-GFP ( R1091P , Acris , Origene ) 1:200 , rat monoclonal anti-HA ( 11867423001 , Sigma ) 1:200 . Secondary Alexa Fluor conjugated antibodies ( Life Technologies ) were used at 1:1000 . Nuclei were visualized by incubating embryos in DAPI at 1 μg/ml . Images of antibody-stained embryos were acquired on glass-bottomed dishes ( Ibidi or MatTek ) with a Leica SP5 , Leica SP8 or Zeiss LSM880 laser scanning confocal microscopes . The same parameters were used for imaging each experiment . Semi-automated 3D nuclear segmentation for quantification of fluorescence intensity was carried out using MINS , a MATLAB-based algorithm ( http://katlab-tools . org/ ) ( Lou et al . , 2014 ) , and analysed as previously described ( Saiz et al . , 2016 ) . To correct z-associated attenuation , intensity levels were fit to a linear model . Mitotic and pyknotic nuclei were excluded from the analysis . For defining cells as positive or negative for a given nuclear marker , we ordered cells by intensity levels and established a threshold for each experiment based on manual verification of the point where nuclear and cytoplasmic signals were equal . This process was repeated independently for each set of embryos processed and imaged in parallel , to overcome inter-experimental variability . For live imaging , embryos were cultured in microdrops of KSOM on glass-bottomed dishes ( MatTek ) in an environmental chamber as described previously ( Xenopoulos et al . , 2015 ) . Images were acquired with a Zeiss LSM880 laser scanning confocal microscope system using a 40x objective . An optical section interval of 1 . 5 μm was acquired per z-stack , every 15 min . Cell tracking of 3D-movies was carried out using a TrackMate plugin in Fiji ( Fernández-de-Manúel et al . , 2017; Schindelin et al . , 2012; Tinevez et al . , 2017 ) . The 3D reconstruction of the embryos and position of the cells was done using MatLab . The shape of the embryos was fitted into an ellipse and the coordinates in X , Y , Z for each blastomere were normalised to the centroid of the ellipse . The algorithm assigned an inner or outer position to each blastomere according to an established threshold , and they were manually verified . The intensity levels of VENUS fluorescent protein in each cell and time point were normalised according to the Z-position to correct the decay of signal intensity due to the distance with the objective ( Saiz et al . , 2016 ) . The frequencies of the intensity levels for each embryo followed a Gaussian distribution . In order to compare different embryos , intensity levels were normalised so that the mean was 0 and the standard deviation was 1 . Two-cell or morula stage embryos were cultured in drops of M16 medium ( Sigma ) covered with mineral oil ( NidOil , EMB ) at 37°C , 5% CO2 , containing the corresponding pharmacological inhibitor or only DMSO as control until the corresponding stage . The following inhibitors and concentrations were used: 10 or 20 μM of the γ-secretase inhibitor RO4929097 ( S1575 , Selleckchem ) ( Münch et al . , 2013 ) and 10 μM of the TEAD/YAP inhibitor Verteporfin ( SML0534 , Sigma ) ( Liu-Chittenden et al . , 2012 ) . RNA from pools of 25–30 embryos ( for pharmacological inhibitor experiments ) or from single embryos ( for CRISPR/Cas9 editing ) was isolated using the Arcturus PicoPure RNA Isolation Kit ( Applied Biosystems ) and reverse transcribed using the Quantitect Kit ( Qiagen ) . RNA was isolated from ES cells with the RNeasy Mini Kit ( Qiagen ) and reverse transcribed using the High Capacity cDNA Reverse Transcription Kit ( Applied Biosystems ) . cDNA was used for quantitative-PCR ( qPCR ) with Power SYBR Green ( Applied Biosystems ) in a 7900HT Fast Real-Time PCR System ( Applied Biosystems ) . Expression of each gene was normalized to the expression of the housekeeping genes Actin ( in mESC or pools of embryos ) or 18S rRNA ( in single embryos ) . Primers used are detailed in Supplementary file 1 . RNA-seq was performed on single morulae . cDNA synthesis was performed using SMART-Seq Ultra Low Input RNA Kit ( Clontech ) . Library preparation and sequencing was performed by the CNIC Genomics Unit using the Illumina HiSeq 2500 sequencer . Gene expression analysis was performed by the CNIC Bioinformatics Unit . Reads were mapped against the mouse transcriptome ( GRCm38 assembly , Ensembl release 76 ) and quantified using RSEM v1 . 2 . 20 ( Li and Dewey , 2011 ) . Raw expression counts were then processed with an analysis pipeline that used Bioconductor packages EdgeR ( Robinson et al . , 2010 ) for normalisation ( using TMM method ) and differential expression testing . Expression data of Rbpj and Neo were used to genotype the samples . Three mutant and three control ( two wildtype and one heterozygote ) embryos were selected for analysis . Changes in gene expression were considered significant if associated to Benjamini and Hochberg adjusted p-value<0 . 05 . RBPJ binding motifs were located according to the consensus motif from CIS-BP database ( M6499_1 . 02 motif ) using FIMO ( Grant et al . , 2011 ) . Association of RBPJ motifs to DEG was performed using BEDTOOLS ( Quinlan and Hall , 2010 ) using a 10 Kb window surrounding the transcriptional start site of genes . ATAC-seq data from 8-cell stage embryos ( Wu et al . , 2016 ) was mapped to the GRCm38 assembly and integrated with the coordinates of RBPJ motifs previously detailed . Sequencing data have been deposited at GEO under accession number GSE121979 . For the generation of transient transgenics , F1 ( C57Bl/6 x CBA ) females were superovulated to obtain fertilized oocytes as described ( Behringer et al . , 2014 ) . Each construct was microinjected into the pronucleus of fertilized oocytes at E0 . 5 at a concentration of 2 ng/μl . Microinjected oocytes were cultured in microdrops of M16 medium ( Sigma ) covered with mineral oil ( NidOil , EMB ) at 37°C , 5% CO2 until the morula stage . Each fragment to be tested was amplified from mouse genomic DNA using NEBuilder HiFi DNA Assembly kit ( New England Biolabs ) and cloned into a modified pBluescript vector ( Yee and Rigby , 1993 ) containing a H2BmRFP reporter gene under the control of the human beta-globin minimal promoter and including an SV40 polyadenylation signal . Primers for amplifying and cloning the 700 bp Tle4-up region are ctatagggcgaattggagctcTTCTTTAGAGGCACCAGTC and ggatccactagttctagagcggccgcATAAAGCCATTTTGCTTAACTG . Primers to amplify and clone the 600 bp Tbx3-i7 region are ctatagggcgaattggagctcCAAGCCAGCCTCAGTCCC and ggatccactagttctagagcggccgcCACACAAGCTTGCCAGCC . Lower case indicates sequence annealing to the plasmid and capital letters indicates sequence annealing to the genome . Constructs were linearized and plasmid sequences removed before microinjection . For H2BmRFP detection , embryos were fixed in 4% paraformaldehyde for 10 min at room temperature and immunostained . Any embryo showing at least one cell expressing the reporter was scored as positive . Due to mosaicism and variability in the amount of transgene present per cell , signal intensity of the reporter cannot be used as a reliable measure of enhancer activity in these assays . When using an empty vector containing only the minimal promoter and the reporter as a negative control , we routinely obtained H2BmRFP expression in approximately 10% of embryos ( Rayon et al . , 2014 ) . Mutated version of Tle4-up ( Tle4-up mutRBPJ ) was generated by site-directed mutagenesis ( Mutagenex Inc ) , changing the TGTGGGAAA binding motif to TGTccGAAA . Mutated version of Tbx3-i7 ( Tbx3-i7 mutRBPJ ) was generated using QuickChange II XL Site-Directed Mutagenesis Kit ( Agilent Technologies ) changing CGTGGGAAA to CGTccGAAA . Lower case indicates the altered residues . Changes that abolish RBPJ binding were based on previously described mutated versions of the binding site ( Tun et al . , 1994 ) . Two guide-RNAs at 60 ng/µl were incubated with tracRNA ( Sigma ) at 240 ng/µl for 5 min at 95°C . The hybridised gRNAs were then incubated with the Cas9 protein ( PNA bio ) at 30 ng/µl for 15 min at RT and microinjected into the pronuclei of ( CBAxC57 ) F1 zygotes . sgRNAs were designed using the CRISPOR tool ( http://crispor . tefor . net/ ) ( Haeussler et al . , 2016 ) . The following guide RNAs were used: Tle4 , TTAGCCTGCACTTCGAGTTA and CCCAATTCAAGGCGTTCTGT; Tbx3 , TAACCCTTTAGAGATAGGCT and TACCAGAGAGGTTTCCTACT . Embryos were recovered at E2 . 5 and lysed in 50 µl extraction buffer from the Arcturus PicoPure RNA Isolation Kit ( Applied Biosystems ) . Aliquots of 10 µl were used for DNA extraction for PCR genotyping . Mosaic embryos were those where we detected both the deleted and the wildtype allele . The remaining 40 µl were used for RNA extraction for RT-qPCR . To characterize the deletions generated , after PCR genotyping deleted bands of some embryos were gel purified , cloned and sequenced . iChr-Notch-Mosaic ES cells were generated by Rui Benedito at CNIC and have been tested negative for Mycoplasma by the CNIC Cell Culture Facility . Cells were cultured in standard ESC media ( DMEM , Gibco ) supplemented with 15% foetal bovine serum ( HyClone ) , 1% Glutamine , 1% NEAA ( Hyclone ) , 0 . 1% ß-mercaptoethanol ( Sigma ) and LIF ( produced in-house ) or 2i ( CHIR-99021 , Selleckchem; and PD0325901 , Axon ) in dishes seeded with a feeder layer of mouse embryonic fibroblasts ( MEFs ) . Cells were transfected with a Cre expressing plasmid to induce recombination using Lipofectamine 2000 ( Invitrogen ) for 24 hr . After recombination , cells were sorted using a Becton Dickinson FACS Aria Cell Sorter . To promote spontaneous differentiation , cells were cultured on gelatine-covered dishes for 48–72 hr after LIF or 2i + LIF removal in DMEM ( Gibco ) supplemented with 20% serum , 1% Glutamine and 0 . 1% ß-mercaptoethanol ( Sigma ) . Statistical analyses were performed with GraphPad Prism seven or R studio . Data are presented as means ± s . d . or means ± s . e . m . as indicted in each case . Differences were considered statistically significant at p-value<0 . 05 . Tests used to calculate p-value are detailed in the figure legends . Student’s t-test was used to compare two groups . ANOVA with Fisher or Bonferroni post-test was used to compare several groups . Fisher’s exact test was used to compare distributions . | We start life as a single cell , which immediately begins to divide to form an embryo that will eventually contain all the different kinds of cells found in the adult body . During the first few rounds of cell division , embryonic cells can become any type of adult cells , but also form the placenta , the organ that sustains the embryo while in the womb . As cells keep on dividing , they lose this ability , called potency , and they take on more specific and inflexible roles . The first choice embryonic cells must make is whether to become part of the placenta or part of the future body . These types of decisions are controlled by molecular cascades known as signalling pathways , which relay information from the cells surface to its control centre . There , specific genes get turned on or off in response to an outside signal . Previous research showed that two signalling pathways , Hippo and Notch , help separate placenta cells from those that will form the rest of the body . However , it was not known whether the two pathways worked independently , or if they were overlapping . Menchero et al . therefore wanted to find out when exactly the Notch pathway started to be active , and examine how it helped cells to either become the placenta or part of the future body . Experiments with developing mouse embryos showed that the Notch pathway was activated after the very first two cell divisions , when the embryo consists of only four cells . Genetic manipulations combined with drug treatments that changed the activity of the Notch pathway confirmed that Notch and Hippo acted independently at this stage . Further , larger-scale analysis of gene activity in these embryos also revealed that Notch signalling was working in a previously unknown way: it turned off the genes that maintain potency , pushing the cells to become more specialised . Ultimately , identifying this new mode of action for the Notch pathway in the early embryo may help to understand how the signalling cascade acts in other types of processes . This knowledge could be useful , for example , to push embryonic cells grown in the laboratory towards a desired fate . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"chromosomes",
"and",
"gene",
"expression",
"developmental",
"biology"
] | 2019 | Transitions in cell potency during early mouse development are driven by Notch |
Many eukaryotic regulatory proteins adopt distinct bound and unbound conformations , and use this structural flexibility to bind specifically to multiple partners . However , we lack an understanding of how an interface can select some ligands , but not others . Here , we present a molecular dynamics approach to identify and quantitatively evaluate the interactions responsible for this selective promiscuity . We apply this approach to the anticancer target PD-1 and its ligands PD-L1 and PD-L2 . We discover that while unbound PD-1 exhibits a hard-to-drug hydrophilic interface , conserved specific triggers encoded in the cognate ligands activate a promiscuous binding pathway that reveals a flexible hydrophobic binding cavity . Specificity is then established by additional contacts that stabilize the PD-1 cavity into distinct bound-like modes . Collectively , our studies provide insight into the structural basis and evolution of multiple binding partners , and also suggest a biophysical approach to exploit innate binding pathways to drug seemingly undruggable targets .
Structural and proteomic research over the past decade has supplanted the traditional structure-function paradigm by establishing the functional relevance of protein dynamics ( Wright and Dyson , 1999; Dunker et al . , 2000; Haynes et al . , 2006; Romero et al . , 2006; Ward et al . , 2004; Xie et al . , 2007 ) . In particular , eukaryotic regulatory and signaling proteins are skewed toward notably higher degrees of flexibility when compared to other functional categories ( Liu et al . , 2009; Iakoucheva et al . , 2002 ) . Regulatory proteins also tend toward comparatively higher degrees of binding promiscuity , and we have previously shown thermodynamically how the entropy associated with their flexibility can relate to their specificity toward multiple binding partners ( Liu et al . , 2009 ) . However , a structural understanding of how this selective promiscuity is achieved is still lacking . Flexible human regulatory proteins such as MDM2 and PD-1 usually only crystallize when ligand-bound . Although nuclear magnetic resonance ( NMR ) can occasionally resolve unbound ( apo ) structures of these proteins , it is noteworthy that their apo NMR ensembles often deviate from their bound crystal structures ( Schon et al . , 2002; Lo Conte et al . , 1999; Betts and Sternberg , 1999; Cheng et al . , 2013; Zak et al . , 2015; Lázár-Molnár et al . , 2008; Lin et al . , 2008 ) . Thus , for many such proteins , available structural data do not capture the full binding dynamics , and the pathway from the apo , non-bound-like state to the bound-like state is unclear . This lack of data obscures the mechanistic connection between interface flexibility , binding promiscuity , and ligand specificity . Moreover , given that many regulatory proteins are promising drug targets , this missing puzzle piece often spells failure for drug design efforts that only target the bound-like state , assuming that this state is available in the apo ensemble . Rational approaches to target flexible proteins will thus benefit from new methods that can reveal the binding pathways connecting the non-bound-like to the bound-like states . Binding to flexible receptors is traditionally described by conformational selection ( Ma et al . , 1999; Tsai et al . , 1999 ) or induced fit ( Koshland , 1958 ) mechanisms , and NMR techniques are often used to distinguish between these two ( Figure 1 ) . Generally speaking , one assumes a conformational selection scenario if the apo protein ensemble samples bound-like states ( apoBL ) ( Boehr et al . , 2009; Hoang and Prosser , 2014 ) . If not , one assumes induced fit ( Schon et al . , 2002 ) . In reality , whether a protein-protein interaction occurs via conformational selection or induced fit depends on the flux of the system through the two alternate pathways from the non-bound-like apo state ( apoNBL ) to the bound-like encounter complex ( ECBL ) ( Hammes et al . , 2009 ) . Flux through the conformational selection pathway is limited by the free -energy difference between the apoBL and apoNBL states , ΔGBLapo , which determines the fractional population of the bound-like state and thus restricts when selection-association with the ligand can occur . On the other hand , flux through the induced fit pathway is for the most part independent of ΔGBLapo , as the ligand is presumed to be able to associate with all apo receptor microstates . Instead , flux through this pathway is limited by the free-energy difference between the ECBL and the non-bound-like encounter complex ( ECNBL ) , ΔGBLEC , which is a function of specific interactions between receptor and ligand , and the energy barrier between these states . Both pathways terminate via a ubiquitous optimization step in which minor structural rearrangements at the ECBL interface lead to the high-affinity complex . 10 . 7554/eLife . 22889 . 003Figure 1 . General mechanism for ligand binding to flexible receptor . In the conformational selection pathway , the ligand docks to the bound-like ( BL ) form of the apo receptor ( apoBL ) to form the bound-like encounter complex ( ECBL ) . In the induced fit pathway , the ligand docks to the non-bound-like ( NBL ) form of the apo receptor ( apoNBL ) to form the non-bound-like encounter complex ( ECNBL ) . Intermolecular interactions then drive structural transitions to the ECBL . Both pathways end with a final induced fit step that optimizes interface side chains , transitioning to the high-affinity complex ( HAC ) . The binding mechanism also highlights an anchor residue often found to be important in molecular recognition ( Rajamani et al . , 2004 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 22889 . 003 To shed light on the structural basis of selective promiscuity in the aforementioned class of flexible-interface multi-ligand proteins , we study the binding mechanism of PD-1 to its cognate ligands PD-L1 and PD-L2 . Human PD-1 is a T-cell receptor and immune response regulator that has recently emerged as a breakthrough anticancer target ( Dömling and Holak , 2014; Couzin-Frankel , 2013 ) . NMR and crystallographic studies have revealed the flexibility of the PD-1 interface by showing that its apo and bound conformations are very different ( Cheng et al . , 2013; Zak et al . , 2015; Lázár-Molnár et al . , 2008; Lin et al . , 2008 ) ( Figure 2 , Figure 2—figure supplement 1 ) , suggestive of an induced fit mechanism . Specifically , while the apo PD-1 interface shows a polar surface around Asn66 with an unmatched NH2 ( Figure 2a ) , in complex this NH2 group forms two hydrogen bonds , with the PD-L1–bound interface exhibiting a hydrophobic patch around Ile126 ( Figure 2b ) , and the PD-L2–bound interface forming a large hydrophobic cavity flanked by Ile126 and Ile134 ( Figure 2c , Figure 2—figure supplement 2 ) . 10 . 7554/eLife . 22889 . 004Figure 2 . Flexibility of the PD-1 binding interface . ( a ) The apo PD-1 binding interface ( Cheng et al . , 2013 ) , showing a flat , polar , core binding interface . Surface residues that shape the core binding interface are labelled . ( b ) The core PD-1 ( green ) - PD-L1 ( yellow ) binding interface , showing a flat hydrophobic receptor surface ( Zak et al . , 2015 ) . White dashed lines indicate hydrogen bonds between PD-L1 side chains . ( c ) The core PD-1 ( cyan ) – PD-L2 ( orange ) binding interface , showing a large hydrophobic receptor cavity ( Lázár-Molnár et al . , 2008 ) . White dashed lines indicate hydrogen bonds between PD-L2 side chains . Note that the conserved anchor residue Tyr123/112 is present in both ( b ) and ( c ) . ( d ) Fractional occlusion of each bound-like Trp110 and Tyr123/112 atom position in the NMR ensemble of apo PD-1 . Numerical values at each atom position denote the fraction of NMR frames that overlap , or ‘occlude’ , that position ( see Materials and methods for full details of how fractional occlusion is calculated ) . Aside from the Cβ , the Trp110 pocket is mostly occluded in the apo PD-1 ensemble , whereas the Tyr123/112 anchor pocket is largely open . ( e ) Overlay of apo , PD-L1–bound , and PD-L2–bound structures of PD-1 defining the ‘open’ and ‘closed’ states of PD-1 residues Asn66 and Ile126 in relation to the open and closed states of the Trp110 binding pocket . DOI: http://dx . doi . org/10 . 7554/eLife . 22889 . 00410 . 7554/eLife . 22889 . 005Figure 2—figure supplement 1 . The cognate ligands of PD-1 . Cocrystal structures of the extracellular domain of PD-1 bound to the Ig-like V-type domains of its two cognate ligands: ( a ) PD-L1 ( Zak et al . , 2015 ) , and ( b ) PD-L2 ( Lázár-Molnár et al . , 2008 ) . Dashed lines indicate hydrogen bonds between the PD-L1/2 anchor and PD-1 residue Glu136 . DOI: http://dx . doi . org/10 . 7554/eLife . 22889 . 00510 . 7554/eLife . 22889 . 006Figure 2—figure supplement 2 . Modulation of PD-1’s flexible interface cavity . Aligned structures of the apo ( white ) ( Cheng et al . , 2013 ) , PD-L1–bound ( green ) ( Zak et al . , 2015 ) , and PD-L2–bound ( cyan ) ( Lázár-Molnár et al . , 2008 ) PD-1 interfaces . Key PD-1 interface residues that line the cavity are shown as small sticks and labelled , with Asn66 and Ile126 shown as large sticks as in Figure 2c . The interface cavity volume of each structure is indicated by the transparent surface of matching color . PD-L2 interface residues Trp110 and the conserved Tyr112 anchor are shown as small orange sticks , for reference . The anchor pocket is unstructured in all three receptor states , but only the PD-L2 bound state accommodates Trp110 . DOI: http://dx . doi . org/10 . 7554/eLife . 22889 . 006 To date , no small molecular weight PD-1 inhibitors have been reported in the literature despite the importance of this blockbuster target ( Dömling and Holak , 2014; Couzin-Frankel , 2013; Zarganes-Tzitzikas et al . , 2016 ) . This was somewhat surprising , since the Trp110 binding site observed in the PD-L2–bound cocrystal ( Figure 2c ) displays two key characteristics known to be favorable for ligand binding: concavity ( Laskowski et al . , 1996; Liang et al . , 1998 ) and hydrophobicity ( Cheng et al . , 2007 ) . It is reasonable to assume that the flexibility of the Trp110 pocket , and the fact that in the apo state it is largely occluded by the unmatched , polar NH2 group of Asn66 ( Figure 2a , d ) , would present significant obstacles to traditional structure-based drug-design methods attempting to target this cavity ( Cozzini et al . , 2008 ) . Thus , efforts to model the binding mechanism of PD-1 would not only shed light on nature’s design principles for flexible and promiscuous protein-protein interfaces , but they may also offer novel avenues for pursuing rational drug design against this and other high-impact targets . To study the mechanism of PD-1 binding , we use molecular dynamics simulations ( MDs ) to identify and quantify the effects of intermolecular interactions on the PD-1 binding interface . We first estimate �GBLapo for the free receptor and demonstrate that apoBL states are exceedingly rare . We then estimate �GBLEC for PD-1 interacting with various peptide constructs that mimic distinct subsets of ligand interface motifs ( Figure 3 ) and identify the critical features that trigger shifts in the PD-1 conformational ensemble toward the bound-like states . By quantifying the energetic contribution of each triggering contact in the ECNBL , we rationalize how PD-1 uses flexibility to simultaneously achieve both promiscuity , that is , binding to multiple ligands , and specificity . We show that a conserved set of three contacts in the PD-1 encounter complexes with PD-L1/2 progressively lowers the free energy of bound-like receptor states with respect to the non-bound-like state . These molecular triggers reshape the non-bound-like hydrophilic interface around Asn66 into a bound-like hydrophobic surface . A fourth contact that differs by a single atom stabilizes this surface into either a shallow patch that interacts with Ala121 in PD-L1 , or a deep cavity that buries Trp110 in PD-L2 . 10 . 7554/eLife . 22889 . 007Figure 3 . Structures of PD-L1/2 – mimicking peptides used to probe PD-1 interface dynamics . Left: core interface binding residues of ( a ) PD-L1 and ( b ) PD-L2 in their bound-like conformations . Right: peptides that were simulated in the presence of apo PD-1 in order to identify the triggers of induced fit interface deformations: ( c ) Y , ( d ) DY , ( e ) GGG , ( f ) GGY , ( g ) GDG , ( h ) ADG , ( i ) GDY , ( j ) ADY , and ( k ) mGDV . DOI: http://dx . doi . org/10 . 7554/eLife . 22889 . 007 We find that these triggers , which include the anchor Tyr123/112 in PD-L1/PD-L2 ( Figure 2b , c , d ) ( Rajamani et al . , 2004 ) , are highly conserved across species ( Lázár-Molnár et al . , 2008 ) and drive quantitatively similar , kinetically efficient downhill binding pathways . The importance of these triggers is underscored by the PD-1 – targeting , anticancer antibody pembrolizumab , which evolved via a distinct evolutionary pathway yet , as we show , exploits some of the same triggering machinery as PD-1’s natural ligands . Finally , we suggest how these induced-fit triggers could be used in rational , small-molecule drug discovery by studying the binding mode of a potent macrocyclic PD-1 inhibitor . Collectively , our findings demonstrate how nature exploits structural flexibility to achieve selective binding promiscuity in regulatory proteins .
Analysis of aligned PD-1 structures ( Figure 2 ) led us to classify the bound-like and non-bound-like conformational states using two binary order parameters defined by the ‘open’ or ‘closed’ states of Asn66 and Ile126 . Namely , for a non-bound-like interface Asn66 is closed and Ile126 is open; for the PD-L1-specific bound-like state Asn66 is open and Ile126 is closed; and for the PD-L2-specific bound-like state both Asn66 and Ile126 are open ( Figure 2e ) . In the PD-L1–bound state , the interface exhibits a large hydrophobic patch that interacts with the side chain of ligand interface residue Ala121 ( Figure 2b ) . In the PD-L2–bound state , the interface exhibits a deep hydrophobic cavity that buries ligand residue Trp110 ( Figure 2c ) . Neither this hydrophobic patch nor deep cavity is sampled in the apo PD-1 NMR ensemble , where , instead , the closed state of Asn66 blocks the Trp110-binding pocket by exposing its NH2 group ( Figure 2a , e , Figure 2—figure supplement 2 ) , making a hydrophilic site . MDs of apo PD-1 confirm that Asn66 remains closed ( Figure 4a ) , stabilized by a hydrogen bond with Lys78 that is also present in NMR structures ( Figure 5a ) . These findings suggest that specific interactions between apo PD-1 and a nearby ligand might be required to open Asn66 and reshape the hydrophilic interface into its hydrophobic bound-like states . 10 . 7554/eLife . 22889 . 008Figure 4 . Dynamics of PD-1 binding interface in the presence of different ligands . ( a ) Rolling averages of distance between Trp110_NE1 ( from bound PD-L2 ) and Asn66_ND2 from MDs of apo PD-1 ( blue ) alone and interacting with GGG ( maroon ) and GDG ( red ) peptides . Only GDG peptide sequesters Asn66 away from Trp110 binding pocket . ( b ) Rolling averages of PD-1 binding cavity volume from simulations of apo PD-1 alone ( blue ) and interacting with GDG ( red ) and GDY ( orange ) peptides shows that only GDY stabilizes an open cavity . ( c ) Ile126 X1 rotamer angle from MDs of apo PD-1 interacting with GDG ( red ) , GDY ( orange ) , and ADY ( yellow ) peptides . Peptide ADY and GDY position Ile126 in the closed and open states , respectively ( as in Figure 2e ) . Replicate trajectories for panels a , b , and c are shown in Figure 4—figure supplement 2 . ( d ) Fractional occlusion of each bound-like Trp110 atom position in simulations of PD-1 interacting with the GDY peptide show an open Trp110-binding pocket . The fractional occlusion of a Trp110 atom position is defined as the percentage of simulation frames in which a PD-1 atom overlaps , or ‘occludes’ , that position ( see Materials and methods for full details of how fractional occlusion is calculated ) . ( e ) Fractional occlusion of each bound-like Trp110 atom position in simulations of PD-1 interacting with the ADY peptide show a closed Trp110-binding pocket . DOI: http://dx . doi . org/10 . 7554/eLife . 22889 . 00810 . 7554/eLife . 22889 . 009Figure 4—source data 1 . Excel workbook containing all the simulation trajectory data plotted in Figure 4 , Figure 4—figure supplement 2 , and Figure 4—figure supplement 3 . The first three sheets of the workbook , titled ‘Asn66 position , ’ ‘cavity volume , ’ and ‘Ile126 dihedral’ contain the time-series data plotted in Figure 4a , b and c , respectively . They also contain the data for the replicate simulations shown in Figure 4—figure supplement 2 . The columns labeled with a ‘1’ ( apo1 , GDY1 , etc… ) correspond to the data from the first replicate simulation , i . e . the data shown directly in Figure 4 . Columns labeled with ‘2’ and ‘3’ correspond to data plotted in Figure 4—figure supplement 2 . The final sheet of the workbook is titled ‘alternate anchors , ’ and contains the time series data plotted in Figure 4—figure supplement 3 . DOI: http://dx . doi . org/10 . 7554/eLife . 22889 . 00910 . 7554/eLife . 22889 . 010Figure 4—figure supplement 1 . Apo PD-1 interactions with GDY peptide opens a hydrophobic cavity . Panels ( a ) and ( b ) illustrate the PD-1 interface cavity volume which is plotted in Figure 4b . ( a ) Snapshot from simulation of human PD-1 interacting with the GDG peptide . The PD-1 interface cavity volume is shown in red surface . Although Asn66 is in the open state , the cavity is closed by the closed state of I126 . ( b ) Snapshot from simulation of human PD-1 interacting with the GDY peptide . The PD-1 interface cavity volume is shown in orange surface . The Y anchor side chain positions Ile134 to pull Ile126 out of the pocket via hydrophobic interaction , leaving a large open cavity . DOI: http://dx . doi . org/10 . 7554/eLife . 22889 . 01010 . 7554/eLife . 22889 . 011Figure 4—figure supplement 2 . Replicate trajectories from Figure 4a , b , c . Top: Rolling averages of distance between Trp110_NE1 ( from bound PD-L2 ) and Asn66_ND2 from MDS of apo PD-1 ( blue ) alone and interacting with GGG ( maroon ) and GDG ( red ) peptides . Middle: Rolling averages of PD-1 binding cavity volume from simulations of apo PD-1 alone ( blue ) and interacting with GDG ( red ) and GDY ( orange ) peptides . Bottom: ( f ) Ile126 X1 rotamer angle from MDS of apo PD-1 interacting with GDG ( red ) , GDY ( orange ) , and ADY ( yellow ) peptides . DOI: http://dx . doi . org/10 . 7554/eLife . 22889 . 01110 . 7554/eLife . 22889 . 012Figure 4—figure supplement 3 . Dynamics of PD-1 binding cavity in the presence of different anchor substitutes . Ile126 X1 rotamer angle from MDs of apo PD-1 interacting with GDF ( orange ) , GDK ( grey ) , and GDH ( blue ) peptides . GDF produces a mostly open interface cavity , while GDK and GDH stabilize the closed surface . DOI: http://dx . doi . org/10 . 7554/eLife . 22889 . 01210 . 7554/eLife . 22889 . 013Figure 5 . Hydrogen bond network of PD-1 Asn66 in different contexts . ( a ) NMR structure of the dominant apo , non-bound-like state of the human PD-1 interface ( Cheng et al . , 2013 ) . Asn66 is in the closed state , forming a single hydrogen bond with Lys78 . ( b ) Cocrystal structure of the human PD-1 – PD-L1 complex ( Zak et al . , 2015 ) . PD-1 bound-like interface shows Asn66 in the open state , forming two hydrogen bonds with the ligand Ala121 backbone and the neighboring Tyr68 . For clarity , only relevant ligand atoms are shown . ( c ) Cocrystal structure of the murine PD-1–PD-L2 complex ( Lázár-Molnár et al . , 2008 ) . PD-1 bound-like interface shows Asn66 is in the open state , forming two hydrogen bonds with the ligand Trp110 backbone and a crystal water stabilized by neighboring residue Asn68 . ( d ) Simulation snapshot of human PD-1 interacting with the GDG peptide , showing the same hydrogen bond network as in ( b ) . ( e ) Simulation snapshot of human PD-1 Y68N mutant interacting with the GDG peptide , showing the same water-mediated hydrogen bond network as in ( c ) . DOI: http://dx . doi . org/10 . 7554/eLife . 22889 . 013 For both induced fit and conformational selection , the association of the apo receptor and ligand is driven mainly by diffusion ( DeLisi , 1980; Northrup and Erickson , 1992 ) . It has been shown that often protein-protein interactions stabilize the initial encounter complex through the burial of a bound-like anchor motif on the ligand ( Rajamani et al . , 2004 ) , which allows subsequent , longer timescale intermolecular interactions to take shape . Co-crystal structures , MDs and docking studies of PD-L1/2 suggest that the homologous interface residues Tyr123/112 ( see Figure 2b , c ) may serve as anchors . Specifically , MDs of apo PD-L1/2 show that Tyr123/112 remain within 0 . 5 Å heavy atom RMSD of their bound-like conformations 88 ± 16% of the time . Furthermore , the Tyr123/112-binding pocket is unobstructed in the apo PD-1 NMR ensemble ( Figure 2d ) , facilitating immediate burial of the side chain upon association . Docking exercises also point to the stabilizing role of the Tyr anchor . Namely , ClusPro ( Comeau et al . , 2004 ) successfully re-docked the wild-type human PD-1 – PD-L1 co-crystal ( Zak et al . , 2015 ) , but it failed for single-residue PD-L1 mutants Y123G and Y123A ( Table 1 ) . Collectively , these results suggest an anchor role for Tyr123/112 that facilitates molecular recognition between non-bound-like apo PD-1 and its ligands ( as sketched in Figure 1 ) . 10 . 7554/eLife . 22889 . 014Table 1 . Anchor Tyr123 is key determinant of bound-like docked conformations . Backbone RMSD of top 10 ClusPro ( Comeau et al . , 2004 ) predicted PD-L1 binding modes to the human PD-1–PD-L1 cocrystal ( PDB: 4ZQK ) . RMSDs shown for docked wild type human PD-L1 ( WT ) and for docked PD-L1 anchor mutants Y123G and Y123A . DOI: http://dx . doi . org/10 . 7554/eLife . 22889 . 014Docked PD-L1 backbone RMSD ( Å ) to 4ZQK PD-L1ClusPro modelWTY123GY123A04 . 658 . 849 . 7154 . 038 . 249 . 1249 . 549 . 139 . 2347 . 540 . 440 . 4439 . 449 . 448 . 5548 . 040 . 0753 . 2645 . 853 . 249 . 5740 . 646 . 548 . 1848 . 647 . 847 . 6950 . 748 . 750 . 4 Co-crystal structures of bound PD-1 exhibit an open Asn66 that forms two hydrogen bonds: the first with the backbone oxygen of homologous PD-L1/2 Ala121/Trp110 and the second with either PD-1 Tyr68 ( human PD-1 - PD-L1 complex ) or a crystal water ( murine PD-1 – PD-L2 complex ) ( Figure 5b , c ) . MDs of PD-1 in complex with a GGG peptide positioned to mimic the backbone of PD-L1/2 residues ADY123 and WDY112 , respectively , show that Asn66 fluctuates back and forth between a bound-like open state , where it makes the aforementioned backbone hydrogen bond to the GGG peptide , and the non-bound-like closed state , where it is bonded to PD-1 Lys78 ( Figure 4a ) . On the other hand , simulations with a GDG peptide show that the Asp122/111 mimic forms a hydrogen bond to the Tyr68 OH group , stabilizing a Tyr68 rotamer that can simultaneously hydrogen bond to the NH2 of Asn66 ( Figure 5d ) . Together , this Asn66 – Tyr68 hydrogen bond and the aforementioned Asn66 – backbone hydrogen bond stabilize the bound-like open state of Asn66 ( Figure 4a ) . The robust , four-membered hydrogen bond network between the Ala121/Trp110 backbone mimic , Asn66 , Tyr68 , and the Asp122/111 mimic that we observe in GDG MDs is fully consistent with all available structures and mutagenesis experiments . Namely , the hydrogen bonds rationalize the conservation of Asp122/111 in all known PD-L1/2 sequences and explain PD-L2 mutagenesis studies showing that the D111A mutation abolishes binding to PD-1 ( Lázár-Molnár et al . , 2008 ) . MDs of apo PD-L1/2 further support the importance of Asp122/111 interactions in the encounter complex by showing that this side chain remains within 0 . 4 Å RMSD of its bound-like conformation 82 ± 25% of the time . The stabilization of the bound-like Asp122/111 side chain in simulation is achieved via hydrogen bonds with the neighboring Lys124/113 , bonds which are also observed in bound cocrystal structures of PD-1 ( Figure 2b , c ) . The importance of this stabilizing interaction is underscored by the fact that the K124S and K113A point mutations in PD-L1 and PD-L2 , respectively , both abolish binding to PD-1 ( Lázár-Molnár et al . , 2008; Lin et al . , 2008 ) . PD-1 ligands open Asn66 by offering two novel hydrogen bonds ( with the Ala121/Trp110 backbone and Tyr68 ) that out-compete the single Lys78 hydrogen bond that stabilizes the closed state . Interestingly , the one known PD-1 sequence that diverges at the Tyr68 position is murine PD-1 , which has a Y68N mutation . The murine PD-1–PD-L2 co-crystal shows that although the shorter Asn68 side chain cannot hydrogen bond directly to Asn66 or Asp111 , it hydrogen bonds to a crystal water molecule that forms the same hydrogen bond network as Tyr68 ( Figure 5c ) . MDs of a human Y68N PD-1 mutant and the GDG peptide suggest a functional equivalence of Asn68 to Tyr68: the Asn68 side chain spontaneously recruits a stable water to the co-crystal position that then opens Asn66 via a specific hydrogen bond network analogous to that formed by Tyr68 ( Figure 5e ) . While GDG MDs show an open Asn66 ( Figure 4a ) that exposes a hydrophobic surface , this surface remains flexible and fluctuates between a deep open cavity and closed shallow patch ( Figure 4b ) . Contrary to the GGG MDs that exhibited open-closed fluctuations of Asn66 ( Figure 4a ) , the pocket instability observed in GDG MDs is caused by open-closed fluctuations of PD-1 residue Ile126 ( Figure 4c ) . In contrast , MDs show that the GDY peptide stabilizes the open states of both Asn66 and Ile126 and maintains the open hydrophobic interface cavity seen in the PD-L2 bound-like state of PD-1 ( Figure 4b , c , d ) . Comparison of the GDG and GDY MDs reveal that the Tyr side chain serves as a ‘wedge’ that stabilizes the flexible loop surrounding Ile134 into a bound-like configuration that is observed in both the PD-L1 and PD-L2 co-crystal structures ( Figure 6 ) . In the presence of the GDY peptide , the bound-like Ile134 makes a hydrophobic contact with the long arm of Ile126 , which pulls the latter residue out of the pocket and stabilizes its open state ( Figure 4—figure supplement 1 ) . 10 . 7554/eLife . 22889 . 015Figure 6 . Stabilization of bound-like Ile134 by conserved tyrosine ( Y ) anchor . Average and standard deviation heavy atom RMSD of PD-1 Ile134 to the PD-L1/2 bound-like state ( measured from human PD-1 – PD-L1 cocrystal , 4ZQK; Ile134 has <0 . 2 Å heavy atom RMSD between 4ZQK and the PD-L2 cocrystal 3BP5 ) . Data are shown for three 200ns replicate simulations for each system , including apo human PD-1 and PD-1 interacting with various peptides . DOI: http://dx . doi . org/10 . 7554/eLife . 22889 . 01510 . 7554/eLife . 22889 . 016Figure 6—source data 1 . Excel workbook with a single sheet containing the numerical RMSD data shown in the Figure 6 bar chart . DOI: http://dx . doi . org/10 . 7554/eLife . 22889 . 016 Although the PD-L1 interface exhibits the GDY scaffold , Ile126 is closed in the PD-L1-specific ECBL state , suggesting that an additional ligand motif not contained in the GDY scaffold is responsible for closing the pocket . MDs with an ADY peptide that mimics Ala121 show that the extra Cβ carbon of the Ala side chain out-competes Ile134 for the long arm of Ile126 , stabilizing its closed state ( Figure 4c , e ) . Interestingly , MDs with GDG and ADG peptides both show similarly unstable open-closed fluctuation of Ile126 ( see Figure 7 below ) , which suggests that the effect of the Ala121 Cβ carbon on Ile126 dynamics only emerges in the presence of the anchor Tyr123/112 . Thus , in addition to facilitating molecular recognition , stabilization of the Ile134 loop by the burial of Tyr123/112 is shown to enable ligand-specific induced fit responses by the PD-1 interface . 10 . 7554/eLife . 22889 . 017Figure 7 . Downhill binding pathways of PD-1 triggers of induced fit for each cognate ligand . Points on the plot represent average and standard deviation equilibrium free energy differences ( from three replicate simulations ) between the open and closed states of receptor residues Asn66 and Ile126 for apo PD-1 and PD-1 interacting with nine distinct ligand-mimicking peptides . The corresponding numerical values can be found in Table 2 . Yellow and orange lines represent the ligand-specific induced fit binding pathways from the apo receptor ensemble to the PD-L1 and PD-L2 bound-like ensembles , respectively . DOI: http://dx . doi . org/10 . 7554/eLife . 22889 . 01710 . 7554/eLife . 22889 . 018Figure 7—source data 1 . Excel workbook with a single sheet containing the numerical ΔGopen data plotted in Figure 7 . DOI: http://dx . doi . org/10 . 7554/eLife . 22889 . 018 We applied Maxwell-Boltzmann statistics to our peptide simulations ( see Materials and methods ) to quantify the role played by each trigger in the structural transitions at the PD-1 interface . We evaluate ΔGopen , that is , the free energy differences between the open and closed states of Asn66/Ile126 for PD-1 in isolation and PD-1 interacting with nine different peptides representing distinct PD-L1/2 interface motifs ( Figure 3 , Table 2; note that ΔGopen and ΔGBL are trivially related ) . These ΔGopen values are plotted in Figure 7 . Remarkably , the ADY and GDY motifs , respectively , shift the ratio of our predefined bound-like to non-bound-like states from 1: 44 ± 24 ( based on ΔGopenapo ( Asn66 ) ) to 7 . 4 ± 2 . 8: one for the PD-L1 bound-like state ( based on ΔGopenADY ( Ile126 ) ) and 12 ± 9 . 6: one for the PD-L2 bound-like state ( based on ΔGopenGDY ( Ile126 ) ) . More importantly , we show that each triggering contact monotonically lowers the relative free energy of ligand-specific bound-like states starting from no contacts ( apo ) , to the first , conserved contact with the anchor ( Y ) , to the second , conserved contact with Asp122/111 ( DY ) , to the final , unconserved contact with the backbone O of A/G in the complete triggering motifs ( ADY/GDY ) ( Figure 7 ) . The fact that these downhill binding pathways do not encounter energy barriers strongly suggests that the PD-1 binding mechanism is primarily one of induced fit ( see Figure 1 ) . 10 . 7554/eLife . 22889 . 019Table 2 . Free energy difference between the non-bound-like and bound-like states of PD-1 interface residues Asn66 and Ile126 in various systems . Listed values show the average and standard deviation of �GBL ( from three replicate simulations ) for Asn66 and Ile126 in the different PD-1 systems . Since the bound-like state of Ile126 is closed when PD-L1 – bound and open when PD-L2 - bound , the ΔGBL values for this residue take opposite signs . The trivial relationship between ΔGBL and ΔGopen are indicated for each column . Values shown are in units of kBT , with T = 300 K . DOI: http://dx . doi . org/10 . 7554/eLife . 22889 . 019Pd-l1 / PD-L2PD-L1PD-L2PD-1 SimulationΔGBL ( Asn66 ) ΔGopen ( Asn66 ) ( kBT ) ΔGBL ( Ile126 ) −ΔGopen ( Ile126 ) ( kBT ) ΔGBL ( Ile126 ) ΔGopen ( Ile126 ) ( kBT ) apo ( ΔGBLapo ) 3 . 6 ± 0 . 604 . 4 ± 1 . 2−4 . 4 ± 1 . 2Y ( ΔGBLY ) 1 . 7 ± 0 . 611 . 3 ± 0 . 3−1 . 3 ± 0 . 3DY ( ΔGBLDY ) 0 . 11 ± 0 . 44−0 . 21 ± 0 . 740 . 21 ± 0 . 74GGG ( ΔGBLGGG ) −0 . 28 ± 0 . 520 . 7 ± 0 . 8−0 . 7 ± 0 . 8GGY ( ΔGBLGGY ) 0 . 02 ± 0 . 170 . 58 ± 0 . 22−0 . 58 ± 0 . 22GDG ( ΔGBLGDG ) −2 . 3 ± 0 . 32−0 . 46 ± 0 . 360 . 46 ± 0 . 36ADG ( ΔGBLADG ) −3 . 0 ± 0 . 450 . 35 ± 0 . 60−0 . 35 ± 0 . 60GDY ( ΔGBLGDY ) −3 . 0 ± 0 . 72 . 3 ± 0 . 72−2 . 3 ± 0 . 72ADY ( ΔGBLADY ) −2 . 7 ± 0 . 44−2 . 0 ± 0 . 362 . 0 ± 0 . 36mGDV ( ΔGBLmGDV ) −3 . 0 ± 0 . 47−1 . 8 ± 0 . 481 . 8 ± 0 . 48 In the apo simulation Asn66 is closed ( ΔGopenapo ( Asn66 ) ≈ 3 . 6 kBT ) , repelling Ile126 into an open conformation ( ΔGopenapo ( Ile126 ) ≈ −4 . 4 kBT ) . Docking of the Tyr anchor ( Y ) and formation of the encounter complex destabilizes the non-bound-like apo PD-1 interface , causing increased open-closed fluctuations in both Asn66 and Ile126 . The subsequent docking of Asp122/111 ( DY ) allows Tyr68 to compete with Lys78 to form one hydrogen bond with Asn66 , causing it to swap back and forth between open and closed ( ΔGopenDY ( Asn66 ) ≈ 0 ) . Fluctuations of Asn66 correlate with simultaneous fluctuations of Ile126 ( ΔGopenDY ( Ile126 ) ≈ 0 ) . Adding the adjacent Ala121/Trp110 backbone from PD-L1/2 ( ADY/GDY ) provides the second hydrogen bond for the NH2 of Asn66 that fully stabilizes its open state ( ΔGopenGDY/ADY ( Asn66 ) ≈ −3 . 0 kBT ) . With Asn66 open , the Cβ atom of Ala121 modulates Ile126 dynamics . When present ( ADY ) , the Cβ hydrophobically recruits Ile126 into the closed pocket state ( ΔGopenADY ( Ile126 ) ≈ 2 . 0 kBT ) . Without Cβ ( GDY ) , Ile126 remains open ( ΔGopenGDY ( Ile126 ) ≈ −2 . 3 kBT ) . Our ΔGopen calculations also quantify the critical role of the anchor residue Tyr123/112 in ensuring the ligand specificity of PD-1 interface deformations . This is demonstrated by the fact that GDY and ADY peptides impose clear differential influence on the dominant rotamer state of Ile126 , while for both GDG and ADG , Ile126 fluctuates about evenly between the open and closed state ( ΔGopenGDG/ADG ( Ile126 ) ≈0 ) ( Figure 7 ) . We ran MDs of the PD-L1/2 encounter complexes starting from docked poses of apo PD-1 and the interacting domains of PD-L1/2 that anchored Tyr123/Y112 ( see Supporting Materials and methods ) . Encounter complex MDs recapitulated the triggering mechanisms we identified in our peptide simulations and their resulting PD-1 interface transitions from the ECNBL to the ligand-specific ECBL states . The chronology for these interactions ( Table 3 ) is the same for both ligands . Consistently , the first interaction to take place after docking the conserved anchor is the formation of the hydrogen bond between receptor residue Tyr68 and ligand residue Asp122/111 . This is followed by stabilization of Asn66 in the open pocket state via hydrogen bonds with neighboring Tyr68 and the ligand Ala121/Trp110 backbone . The Ala121/Trp110 side chains then proceed to stabilize a closed/open hydrophobic pocket . Note that the Trp in the WDY motif of PD-L2 readily fills the hydrophobic pocket as the XDY motif latches and opens Asn66 ( Figure 8 ) . Consistent with a downhill free energy induced fit mechanism , the realization of these four contacts takes less than 10 ns total . On a longer timescale , encounter complex simulations demonstrate the formation of secondary hydrogen bonds at the interface periphery that are also observed in co-crystal structures of human and murine PD-1 . These secondary hydrogen bonds , including the bond from PD-1 Lys78 to PD-L1/2 Phe19/21 and from Gln75 to Arg125/Tyr114 ( Figure 9 ) , were consistently observed to form approximately 10 nanoseconds after the aforementioned Asn66 and Tyr68 hydrogen bonds ( Table 3 ) , suggesting that ECBL contacts shaped by the triggers of induced fit are enough to drive the subsequent transition to the HAC . 10 . 7554/eLife . 22889 . 020Figure 8 . Modulation of the PD-1 interface binding cavity in encounter complex simulations with PD-L1 and PD-L2 . Plot shoes the ( rolling average ) number of atoms in the bound-like Trp110 side chain reference that are occluded by the PD-1 interface throughout encounter complex simulations with PD-L1/2 ( see Materials and methods for full details of how occlusion is calculated ) . Both encounter complexes begin with a closed Trp110 pocket , as this is the dominant state of apo PD-1 . The PD-L2 trigger then stabilizes the hydrophobic cavity ( no overlap ) , while the PD-L1 trigger stabilizes the hydrophobic patch ( significant overlap ) . DOI: http://dx . doi . org/10 . 7554/eLife . 22889 . 02010 . 7554/eLife . 22889 . 021Figure 8—source data 1 . Excel workbook with a single sheet containing the time-series Trp110 atom overlap data from the encounter complex simulations plotted in Figure 8 . DOI: http://dx . doi . org/10 . 7554/eLife . 22889 . 02110 . 7554/eLife . 22889 . 022Figure 9 . Secondary , non-triggering contacts in PD-1 encounter complexes . Specific hydrogen bonds observed in the PD-1 – PD-L1 ( a ) ( Zak et al . , 2015 ) and PD-1 – PD-L2 ( b ) ( Lázár-Molnár et al . , 2008 ) cocrystal structures . In simulation , these contacts form approximately 10 ns after triggering interactions and their resulting induced fit deformations of the receptor ( Table 3 ) . Note also that the conserved Tyr123/112 anchor forms identical hydrogen bonds with Glu136 in the PD-L1– and PD-L2–bound states . DOI: http://dx . doi . org/10 . 7554/eLife . 22889 . 02210 . 7554/eLife . 22889 . 023Table 3 . Chronology of the formation of intermolecular interactions between PD-1 and PD-L1/2 in encounter complex simulations . Listed values show the average and standard deviation time to formation ( from three replicate simulations ) of various inter- and intramolecular hydrogen bonds following the burial of the ligand anchor and formation of the key Tyr68–Asp122/111 hydrogen bond . DOI: http://dx . doi . org/10 . 7554/eLife . 22889 . 023Δt ( ns ) after Tyr68 – Asp122/111 hydrogen bond formationHydrogen bondPD-1 - PD-L1 Encounter ComplexPD-1 - PD-L2 Encounter ComplexAsn66 – Ala121/Trp1106 . 3 ± 2 . 96 . 7 ± 7 . 2Asn66 – Tyr685 . 0 ± 1 . 78 . 3 ± 7 . 5Gln75 – Arg125/Tyr11415 ± 7 . 817 ± 11Lys79 – Phe19/2113 ± 1515 ± 20 Recently , two FDA-approved PD-1–targeting antibodies have emerged as part of a new generation of anticancer immune checkpoint inhibitors . Published crystal structures of one of these antibodies , pembrolizumab , bound to extracellular PD-1 show a hydrophobic receptor binding surface that overlaps that which binds PD-L1/2 ( Figure 10b ) ( Horita et al . , 2016; Lee et al . , 2016; Na et al . , 2017 ) . Comparison of the pembrolizumab – PD-1 interface to the PD-L1 – PD-1 interface using the FastContact server ( Champ and Camacho , 2007 ) highlights several differences in the main contacts that characterize the two binding modes ( Figure 10a , Tables 4 and 5 ) . Remarkably , the pembrolizumab-bound crystal structures reveal that the antibody stabilizes the same open state of Asn66 as PD-L1/2 using an analogous hydrogen bond network ( Figure 10c ) . The fact that this antibody , designed via a distinct evolutionary pathway , shares PD-L1/2’s mechanism for opening Asn66 and revealing a hydrophobic binding surface ( Figure 2a , b , c , Figure 10b ) underscores the role of this specific interaction in PD-1 interface remodeling . 10 . 7554/eLife . 22889 . 024Figure 10 . Pembrolizumab–bound PD-1 interface resembles PD-L1–bound interface with a closed Ile134 . ( a ) Alignment of crystal structures of the pembrolizumab antibody ( Ab ) ( Horita et al . , 2016 ) and PD-L1 ( Zak et al . , 2015 ) binding modes , showing distinct but partially overlapping binding interfaces on PD-1 . The light chain of the Ab is shown in magenta and the heavy chain is shown in purple . ( b ) Detailed comparison of the aligned Ab–bound ( grey ) and PD-L1–bound ( green ) PD-1 interfaces . Most receptor interface residues exhibit near-identical conformations , except Ile134 which is open when bound to PD-L1 but closed when bound to pembrolizumab . Heavy chain Ab interface residues are shown in purple . ( c ) Detail of the Ab–PD-1 interface , highlighting the hydrogen bond ( hydrogen bond ) network that stabilizes the open state of Asn66 . This hydrogen bond network is functionally analogous to those observed in the PD-L1 and PD-L2–bound cocrystals ( Figure 5 ) , although the OD1 and ND2 atoms of Asn66 are flipped . ( d ) Simulation snapshot of human PD-1 interacting with the mGDV motif from Bristol-Myers Squibb macrocyclic PD-1 inhibitor , highlighting the canonical hydrogen bond network that opens Asn66 . DOI: http://dx . doi . org/10 . 7554/eLife . 22889 . 02410 . 7554/eLife . 22889 . 025Figure 10—figure supplement 1 . Model of potent Brystol-Myers-Squibb macrocyclic PD-1 inhibitor . Predicted macrocycle binding mode is shown brown , with certain side chains omitted for clarity ( see Figure 10—figure supplement 2 for full macrocycle structure ) . Key PD-L1 ( yellow ) ( Zak et al . , 2015 ) and pembrolizumab ( purple and magenta ) ( Horita et al . , 2016 ) interface residues from their bound cocrystal structures are shown to highlight predicted native-like contacts . Inset: the mGDV segment of the macrocycle aligned to PD-L1’s ADY trigger and pembrolizumab’s corresponding interface residues . Green and red spheres represent hydrophobic and polar pharmacophores matched by both pembrolizumab and the mGDV macrocycle motif . DOI: http://dx . doi . org/10 . 7554/eLife . 22889 . 02510 . 7554/eLife . 22889 . 026Figure 10—figure supplement 2 . Predicted interactions of Brystol-Myers-Squibb macrocyclic PD-1 inhibitor . Figure shows the 2D structure of the patented Brystol-Myers-Squibb macrocycle with the mGDV sequence highlighted in magenta . Dashed lines indicate the specific interactions between the macrocycle and the PD-1 interface ( green circles ) that are observed in our binding model . Amino-acid components of the macrocycle are labeled , and analogous PD-L1 cocrystal ( Zak et al . , 2015 ) residues that participate in the same interactions are indicated in parenthesis . Our binding model recapitulates most native-like contacts present in the human PD-1–PD-L1 cocrystal . DOI: http://dx . doi . org/10 . 7554/eLife . 22889 . 02610 . 7554/eLife . 22889 . 027Table 4 . Top 5 PD-1 residues contributing to electrostatic energy when binding to PD-L1 and pembrolizumab . Binding energies were calculated using the FastContact web server ( Champ and Camacho , 2007 ) and cocrystal structures of PD-1 bound to PD-L1 ( Zak et al . , 2015 ) and pembrolizumab ( Horita et al . , 2016 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 22889 . 027PD-L1–boundPembrolizumab–boundResidueEnergy ( kcal/mol ) ResidueEnergy ( kcal/mol ) Glu136*−11 . 531Asp85‡−8 . 367Asp77−5 . 073Ser87−3 . 629Lys78† −4 . 266Asp77−2 . 417Gln75−4 . 027Tyr68−2 . 156Glu84−3 . 119Glu136−2 . 096*The E136A mutation abolishes binding of PD-1 to PD-L1 and greatly reduces binding to PD-L2 ( Lázár-Molnár et al . , 2008 ) . †The K78A mutation abolishes binding of PD-1 to PD-L1 and greatly reduces binding to PD-L2 ( Lázár-Molnár et al . , 2008 ) . ‡The D85G mutation abolishes binding of PD-1 to pembrolizumab ( Na et al . , 2017 ) . 10 . 7554/eLife . 22889 . 028Table 5 . Top 5 PD-1 residues contributing to desolvation energy when binding to PD-L1 and pembrolizumab . Binding energies were calculated using the FastContact web server ( Champ and Camacho , 2007 ) and cocrystal structures of PD-1 bound to PD-L1 ( Zak et al . , 2015 ) and pembrolizumab ( Horita et al . , 2016 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 22889 . 028PD-L1–boundPembrolizumab–boundResidueEnergy ( kcal/mol ) ResidueEnergy ( kcal/mol ) Ile126*−1 . 853Leu128†−2 . 886Leu128†−1 . 673Pro89−2 . 486Ile134‡−1 . 361Val64−1 . 721Val64−0 . 463Pro130−1 . 586Ala132−0 . 37Pro83−1 . 131*The I126A mutation greatly reduces binding of PD-1 to both PD-L1 and PD-L2 ( Lázár-Molnár et al . , 2008 ) . †The L128A mutation abolishes binding of PD-1 to PD-L1 and partially reduces binding to PD-L2 ( Lázár-Molnár et al . , 2008 ) . ‡The I134A mutation abolishes binding of PD-1 t oPD-L1 and greatly reduces binding to PD-L2 ( Lázár-Molnár et al . , 2008 ) . Although pembrolizumab’s interaction with Asn66 mimics the native-like contacts of PD-L1/2 , the antibody-bound receptor exhibits a novel configuration of Ile134 , with both Ile126 and Ile134 in inward-flipped , ‘closed’ states ( Figure 10b ) . The result is a large hydrophobic surface where , like in the PD-L1–bound state , the closed Ile126 occludes the Trp110-binding pocket , but where , unlike the PD-L1/2–bound states , a closed Ile134 partially fills the Tyr/123/112 anchor cavity . In fact , pembrolizumab has no anchor analog . Instead , the Arg102 side chain extends along the PD-1 interface such that the CZ carbon overlaps the Cγ position of Tyr123/112 ( Figure 10—figure supplement 1 ) , and the NH1/2 groups hydrogen bond to a crystal water above the receptor interface ( Figure 10b ) . In this configuration , the hydrophobic carbon chain of Arg102 forms a ‘cap’ above the closed Ile126 and Ile134 , desolvating their hydrophobic interactions with each other and the neighboring Gly124 and stabilizing a flat hydrophobic surface ( Figure 10b ) . A similar closed conformation of Ile134 is observed in our MDs of PD-1 interacting with the GDG peptide ( Figure 4—figure supplement 1 ) . This is unsurprising: like pembrolizumab , the GDG peptide has the necessary machinery to trigger the opening of Asn66 , but lacks an anchor ‘wedge’ that prevents the resulting inward collapse of Ile134 . Results of the GDG MDs thus rationalize the pembrolizumab binding mode and suggest an anchor-independent induced fit PD-1 binding pathway: one in which the antibody opens Asn66 using the canonical hydrogen bond network and stabilizes the resulting flat hydrophobic interface by ‘capping’ the closed states of Ile126/134 with the carbon chain of Arg102 . Although two PD-1-targeting antibodies already exist on the market , there are no small-molecule PD-1 inhibitors in clinical trial , despite the enormous interest in this blockbuster immunotherapy target ( Dömling and Holak , 2014; Couzin-Frankel , 2013; Zarganes-Tzitzikas et al . , 2016 ) . Given that ligand-binding sites tend to be concave ( Laskowski et al . , 1996; Liang et al . , 1998 ) and largely hydrophobic ( Cheng et al . , 2007 ) , the undruggability of PD-1 might be due to the closed Asn66 and the resulting flat polar interface in the apo form ( Figure 2a ) . However , the highly specific hydrogen bond network presented by PD-L1/2 and pembrolizumab strongly suggests a path to open Asn66 and transform the hard to drug hydrophilic patch into a hydrophobic one . Interestingly , Brystol-Myers-Squibb recently patented a 1 . 03 nM macrocyclic inhibitor of the PD-1–PD-L1 interaction ( Miller et al . , 2014 ) . Although no mechanism of action has been described , the macrocycle includes a peptidic mGDV motif that is structurally similar to the aforementioned ADY induced fit trigger , with an N-methylated Gly and an Asp side chain that resemble PD-L1’s Ala121 and Asp122 , respectively ( Figure 10—figure supplements 1 and 2 ) . This alignment puts the mGDV motif’s short Val side chain at the position of the much longer Tyr123 anchor , where it aligns with the C∆ side chain carbon of pembrolizumab residue Arg112 ( Figure 10—figure supplement 1 ) . Given the resemblance of the mGDV motif to the interface residues of both PD-L1 and pembrolizumab , we used our MDs method to evaluate whether this motif was capable of remodeling the apo , non-bound-like PD-1 interface into a bound-like state . We observed that mGDV opened Asn66 using a native-like hydrogen bond network analogous to those seen in previous simulations ( Figures 5 , 7 and 10d ) . However , Ile126 and Ile134 dynamics mirrored those seen in the pembrolizumab cocrystal , with both sidechains favoring inward-flipped ‘closed’ configurations ( Figure 11 ) . Simulation trajectories showed that the short Val side chain of the mGDV motif , unlike the cognate Tyr123/112 anchors , did not penetrate deep enough into the PD-1 interface to be a ‘wedge’ stabilizing an open Ile134 . Instead , like the carbon chain of pembrolizumab residue Arg102 , the Val ‘capped’ stable hydrophobic interactions between a closed Ile134 , a closed Ile126 , and the neighboring Gly124 . 10 . 7554/eLife . 22889 . 029Figure 11 . Macrocyclic mGDV motif induces structural changes in the PD-1 interface towards the pembrolizumab–bound state . Heat maps show the distributions of PD-1’s Ile126 and Ile134 X1 rotamer angles in MDs of the receptor interacting with the ADY PD-L1 trigger ( left ) , the BMS macrocycle mGDV motif ( center ) , and the GDV peptide . Data for each ligand were gathered from three distinct 200ns simulations . White dots on the plots indicate the rotamer angles of the same two residues in the pembrolizumab ( Ab ) –bound ( Horita et al . , 2016 ) and PD-L1–bound ( Zak et al . , 2015 ) cocrystal structures . DOI: http://dx . doi . org/10 . 7554/eLife . 22889 . 02910 . 7554/eLife . 22889 . 030Figure 11—source data 1 . Excel workbook with a single sheet containing the 2D histogram data for the heatmaps shown in Figure 11 . DOI: http://dx . doi . org/10 . 7554/eLife . 22889 . 030 Our GDG , ADG , GDY and ADY simulations demonstrated that precise regulation of the closed/open states Ile126 via the Ala121 Cß is realized only when the Tyr123/112 anchor is buried ( Figure 7 ) . Thus , given that mGDV lacks an anchor , a natural question to ask is whether a Ile126 would be opened by a GDV peptide without the N-methyl group . Interestingly , MDs of PD-1 interacting with a GDV peptide revealed identical Ile126 and Ile134 dynamics to mGDV simulations ( Figure 11 ) , indicating that the N-methyl group was not recruiting Ile126 into the closed state in the style of Ala121 Cß . These results help to further illuminate the role of the conserved anchor Tyr123/112 , which in its absence does not wedge Ile134 into the open state , disabling the capability of PD-1 to stabilize an open Ile126 and form a hydrophobic cavity at that site . Compared to GDG simulations in which Ile126 fluctuated between open and closed ( Figure 4b , c ) , in GDV simulations it remained closed , suggesting a stabilizing role for the Val side chain . The overlap of ( m ) GDV’s Val with the carbon chain of pembrolizumab’s Arg102 ( Figure 10—figure supplement 1 ) and the similarity between the ( m ) GDV-induced PD-1 interface and the pembrolizumab–bound interface supports the ‘capping’ role of Arg102 in stabilizing the flat hydrophobic surface of PD-1 . This mechanism is also consistent with models of macrocycle conformations generated by Balloon ( Vainio and Johnson , 2007 ) docked to PD-1 , which readily identify poses that align the mGDV motif to corresponding PD-L1 and pembrolizumab interface residues ( Figure 10—figure supplements 1 and 2 ) , rationalizing the potency and specificity of the compound .
Our studies show that apo PD-1 does not sample bound-like hydrophobic interface conformations , but instead presents a non-bound-like hydrophilic patch around Asn66 at the core of its binding interface ( Figure 2 ) . By mapping the effect of specific ligand contacts on the apo PD-1 interface , we identify a highly conserved subset of PD-L1/2 motifs responsible for coordinating Asn66 and triggering the transition from the hydrophilic to hydrophobic interface . Namely , Asp122/111 and the backbone O of PD-L1/2 Ala121/Trp110 form a robust , four-membered hydrogen bond network with Tyr68 and Asn66 that neutralizes the latter residue into a bound-like open state . Simultaneously , the conserved anchor Tyr123/112 stabilizes Ile134 into a bound-like state that , with Asn66 open , creates a hydrophobic surface that fluctuates between a patch and a cavity modulated by Ile126 . These three linear ligand motifs ( XDY ) , shared by both PD-L1/2 , comprise the molecular key that unlocks the promiscuity of PD-1 by revealing a flexible hydrophobic binding surface ( Figure 4 ) . With XDY triggering the transition to the flexible hydrophobic surface , specificity toward the two PD-1 ligands is actualized by the formation of the hydrophobic patch when binding PD-L1 vs . the formation of hydrophobic cavity when binding PD-L2 . These two states can be distinguished by the conformation of Ile126 ( Figure 2e ) . For PD-L1 , we show that the ADY motif is sufficient to stabilize the hydrophobic patch ( Figure 4c ) . Specifically , the Ala121 Cβ atom , which does not overlap with PD-1 apo NMR structures ( Figure 2d ) , recruits Ile126 into the closed ( patch ) state . On the other hand , in the absence of Cβ , the GDY trigger stabilizes the open state of Ile126 , producing a large hydrophobic interface cavity consistent with the pocket that buries PD-L2 Trp110 . Note that the Trp in the WDY motif of PD-L2 readily fills the hydrophobic pocket as the XDY motif latches and opens Asn66 ( Figure 8 ) . The pre-arrangement of PD-L1/2 motifs XDY in bound-like conformations in the absence of the receptor is important for efficient ligand recognition and binding . Docking studies and peptide MDs highlight a critical role for the conserved Tyr123/112 anchor both in both molecular recognition and in modulating Ile134 during induced fit , both of which require the Tyr side chain to maintain a stable bound-like rotamer . Furthermore , simulations demonstrate that peptides such as GDG , mGDV , and GDV , which either lack or have a modified anchor analogue , cannot stabilize an open state of Ile126 , highlighting an allosteric role for Tyr123/112 in splitting the PD-1 induced fit binding pathway . Several anchors substitutes were tested in simulation starting in bound-like configurations similar to the cognate Tyr112/123 . These MDs produced three broad types of PD-1 interface dynamics ( Figure 4—figure supplement 3 ) : ( 1 ) aromatic substitutions XDF and XDW stabilized either an open ( X=G ) or closed ( X=A ) pocket like the cognate XDY motif . ( 2 ) Polar substitutions XDH , XDR , and XDK were not accommodated in the hydrophobic anchor pocket and their side chains laid along the receptor surface , consistent with pembrolizumab’s bound Arg102 ( Figure 10b ) , producing a closed pocket like that of ( m ) GDV . ( 3 ) XDG or XDA resulted in open-closed fluctuations of both Ile134 and Ile126 ( Figure 4b , c ) . These observations suggest that certain anchor mutations are tolerated by PD-1 and are consistent with mutagenesis studies showing that the Y112A PD-L2 point mutation slightly reduces , but does not abolish , binding to PD-1 ( Lázár-Molnár et al . , 2008 ) . However , the observed conservation of Tyr123/112 in mammalian species ( Lázár-Molnár et al . , 2008 ) might suggest specific kinetic constraints on ligand recognition arising from hydrophobic contacts with Ile134 and the hydrogen bond with Glu136 ( Figure 9 ) , which are not shared by other sidechains . In addition to the anchor residue , our peptide MDs also suggest an essential role for the conserved Asp122/111 in erecting a stable hydrogen bond network that opens PD-1 Asn66 , which can only be achieved by a bound-like Asp side chain . The primacy of these intermolecular interactions to PD-1 binding is reinforced by our MDs of apo PD-L1/2 , which reveal that Tyr123/112 and Asp122/111 all remain in bound-like conformations in the absence of the receptor , primed to interact immediately upon interface association . Equally important is the fact that apo PD-1 structures all accommodate ( i . e . do not block ) any of contacts of the XDY scaffold , ensuring a rapid recognition process that facilitates subsequent induced fit transitions . Our MDs demonstrate that the set of consecutive intermolecular interactions triggered by ADY and GDY peptides lead to energetically downhill binding pathways with no opposing energy barriers . These pathways strongly suggest that PD-1 occurs mostly by induced fit ( Figure 1 ) . Specifically , simulations and estimated ΔGopen values show that apoBL states of PD-1 are rare , which undermines a conformational selection mechanism . On the other hand , ligand-specific triggers are shown to efficiently shift the PD-1 interface conformational ensemble from a non-bound-like: bound-like ratio of roughly 44: 1 ( in the apo ensemble ) to roughly 1: 7 ( in the encounter complex ensemble ) ( Figure 7 ) . Unconstrained MDs of PD-L1/2 encounter complexes show that the geometry and chronology of triggering contacts is highly optimized , driving the transition from the non-bound-like to the bound-like states in less than 10 ns . This time scale promotes rapid recognition and ensures fast activation of this important T-cell checkpoint . Although regulatory proteins are promiscuous in that they bind multiple targets , they must also be specific so as to limit binding to just those targets . Our analysis of the binding mechanism to PD-1 reveals how these two seemingly contradictory requirements can be simultaneously achieved . Here , we show that apo PD-1 samples an ensemble of non-bound-like conformations that present an obstructive Asn66 on its interface , which likely prevents non-specific binding . The apo PDL1/2 interfaces feature a conserved , bound-like , XDY binding motif that holds the key to opening Asn66 and forming a flexible hydrophobic surface , which completes the first binding step . In the second step , the ligands then attune the flexible interface via specificity-determinant contacts ( X=A for PD-L1 , X=W for PD-L2 ) that modulate Ile126 , splitting the binding pathway and stabilizing either a hydrophobic patch or a binding pocket ( Figures 2 , 4 and 7 ) . The key structural properties in this pathway are: ( a ) a flexible , non-bound-like apo receptor interface ensemble that presents an unfavorable binding surface , ( b ) a core subset of shared ligand binding motifs clustered about an anchor residue that latch the receptor interface but allow it to remain flexible , and ( e ) ligand-specific motifs that split the binding pathway by stabilizing different conformations of the flexible interface . Bound cocrystal structures of the PD-1–targeting antibody pembrolizumab reveal that it exploits an evolutionarily designed induced fit trigger: the four-membered hydrogen bond network that opens Asn66 and makes the receptor interface hydrophobic . This same principle can be applied to design smaller molecular weight PD-1 inhibitors . We have shown that the mGDV motif of the Brystol-Myers-Squibb PD-1 inhibitor combines key pharmacophore features of both PD-L1 and pembrolizumab interfaces: the backbone O of the Gly resembles that of PD-L1’s Ala121 , the Asp side chain resembles PD-L1’s Asp122 , and the Val side chain resembles pembrolizumab’s Arg102 . Simulations suggest that this structural resemblance produces functionally similar dynamics by displacing receptor residue Asn66 ( Figure 10d ) and stabilizing a bound-like , flat hydrophobic surface formed by closed Ile126 and Ile134 ( Figures 7 and 11 ) . Docked conformations of the full inhibitor recapitulate most secondary native-like contacts in addition to the core triggering interactions ( Figure 10—figure supplements 1 and 2 ) . Taken together , these results support the idea that nature’s mechanisms for modulating receptor surfaces might be exploited to design novel chemistries capable of transforming hard to drug targets into more druggable candidates . Promiscuous regulatory proteins must optimize binding kinetics for multiple ligands by exploiting structural flexibility . Given nature’s general mechanisms for flexibility-mediated binding ( Figure 1 ) , specificity toward multiple ligands could , in principle , be conferred either through conformational selection , by evolving the receptor to intrinsically sample different ligand-specific apoBL states , or by induced fit , by evolving interface interactions that efficiently drive transitions to the ligand-specific ECBL states . If conformational selection is used to achieve multi-ligand specificity , the binding pathway flux will de facto be limited by ΔGBLapo , the free energy difference between each ligand-specific bound-like states and other states in the apo ensemble . In this scenario , a natural bottleneck would emerge as an increasing number of ligands would lead to lower association rates . On the other hand , if selective promiscuity is conferred through induced fit , binding pathway flux will not depend on the fractional populations of apo ensemble microstates , but instead will be determined by the ligand-specific triggering mechanisms . We show here that induced fit can efficiently reshape the shallow polar interface of a flexible receptor into a hydrophobic interface amenable to binding multiple ligands by co-evolving a common set of intermolecular contacts . From an evolutionary perspective , this is an efficient approach to spawning novel protein interactions , since these core contacts can be designed just once . Selectivity to novel ligands can then be achieved by evolving relatively small sequence modifications around these core contacts . Perhaps more importantly , we note that contrary to conformational selection , the induced fit approach to selective promiscuity is in principle not limited by the total number of ligands . It is interesting to note that many well-characterized eukaryotic regulatory domains ( Pawson and Scott , 1997 ) bind to several linear binding sequences that share common motifs around an anchor residue and differ in other nearby regions . This trend suggests that the selective promiscuity via induced fit mechanism proposed here for PD-1 might apply elsewhere in nature . This possibility is currently being studied by analyzing the triggers of induced fit in other systems .
Molecular dynamics simulations ( MDs ) of the extracellular domain of PD-1 were run in triplicate using the first three solution NMR structures of apo human PD-1 ( PDB ID: 2M2D [Cheng et al . , 2013] ) . Before simulating specific receptor-ligand interactions , MDs of apo PD-1 were evaluated to ensure that the resultant dynamics are consistent with the experimentally derived apo NMR ensemble . As shown in Figure 12 , apo MDs stabilize within about 2 . 0 Å backbone RMSD of their respective NMR starting points , suggesting that we can successfully sample native-like unbound states . 10 . 7554/eLife . 22889 . 031Figure 12 . Stability of apo PD-1 simulations . Backbone RMSD of apo PD-1 to the first three NMR models ( shown in blue , red , and yellow , respectively ) . Data are shown for six simulations: two replicates ( a , b ) starting from each of the first three NMR models ( 1 , 2 , 3 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 22889 . 03110 . 7554/eLife . 22889 . 032Figure 12—source data 1 . Excel workbook with a single sheet containing the time-series RMSD-to-unbound data from the apo PD-1 simulations plotted in Figure 12 . DOI: http://dx . doi . org/10 . 7554/eLife . 22889 . 032 Available co-crystal structures of human PD-1/human PD-L1 ( PDB ID: 4ZQK [Zak et al . , 2015] ) , murine PD-1/human PD-L1 ( PDB ID: 3BIK [Lin et al . , 2008] ) and murine PD-1/murine PD-L2 ( PDB ID: 3BP5 [Lázár-Molnár et al . , 2008] ) complexes were used as templates for placement of peptides in bound-like loci at the receptor interface , and the dynamics of the PD-1 binding interface in response to interactions with different structural motifs on the ligands were analyzed . We focus on interactions relevant for the opening and closing of the pocket around Asn66 . Based on co-crystals , we noticed that the core interacting residues of PD-L1 ( Ala121 , Asp122 , Tyr123 ) and the homologous residues on PD-L2 ( Trp110 , Asp111 , Tyr112 ) form critical hydrogen bonds ( hydrogen bonds ) shaping this pocket . Thus , to dissect the contribution of each contact , we simulate the effects of the receptor interacting with a diverse set of peptide derivatives of these specific ligand residues . Ten distinct PD-1 systems were simulated in order to dissect the ligand groups that trigger induced fit interface deformations on the receptor . These systems included the apo receptor in isolation and in complex with nine different peptides that mimic cognate ligand backbone and side chain interactions with the receptor ( Figure 3 ) : the anchor residue Tyr , the backbone peptide GGG , five peptides to probe role of ligand side chain contacts DY , GGY , GDG , ADG , GDY , the PD-L1 peptide ADY , and the mGDV peptide , which mimics a patented PD-1 inhibitor . To generate initial structures for our receptor-peptide MDs , NMR models 1–3 of the human PD-1 were backbone aligned to the murine receptor co-crystal ( Lin et al . , 2008 ) and peptides were modeled after the corresponding human PD-L1 interface residues Ala121–Tyr123 , homologous to PD-L2 interface residues W110-Tyr112 . Systems are simulated for 200 ns , resulting in three replicate MDs per system ( including the apo PD-1 system , which does not include any peptide ) , and receptor interface dynamics are compared across systems to identify the ligand motifs and interactions responsible for structural transitions toward the bound-like receptor state . Harmonic restraints ( 100 . 0 kcal/mol ) on all heavy atoms of ligand-mimicking peptides were used in simulation to prevent dissociation of the peptide from the receptor interface . In peptide MDs , harmonic restraints ( 100 . 0 kcal/mol ) were also placed on backbone atoms of non-interface PD-1 beta sheets residues 50–55 , 80–81 , 96–98 , 106–109 and 120–122 . These residues exhibit <0 . 35 Å backbone RMSD in the apo NMR ensemble , and previous studies have also shown that the conformational changes induced by ligand binding do not propagate through the major fold of PD-1 ( Cheng et al . , 2013 ) . Hence , these restraints should not prevent our ability to sample the native-like binding dynamics of the receptor interface in biological conditions . The resolved portion of the N-terminal tail of PD-1 ( residues 33–36 ) , which in NMR models has <0 . 65 Å backbone RMSD , was also restrained so as to limit artificial mobility that might result from the fact that residues 1–32 were missing from simulation . Human PD-1–PD-L1 and PD-1–PD-L2 encounter complexes were modeled and then simulated in triplicate to probe induced fit trajectories and determine the chronology of inter-molecular interactions and specific interface deformations . We modeled encounter complexes by rigid body docking the extracellular domain of the apo receptor and the Ig-like V-type domains of the apo ligands , allowing no structural overlaps . Docked models of PD-L1 had an average backbone RMSD of 5 . 7 ± 1 . 2 to the human PD-1–PD-L1 cocrystal . Docked models of PD-L2 had an average backbone RMSD of 4 . 8 ± 1 . 8 to the murine PD-1–PD-L2 cocrystal ( no human cocrystal is currently available for the PD-1–PD-L2 complex ) . Structural models of apo human PD-L1 and PD-L2 that we used when building encounter complexes were generated by simulating the ligands in solution for 400 ns , using a VMD ( Humphrey et al . , 1996 ) clustering plugin ( https://github . com/luisico/clustering ) to cluster frames by backbone RMSD using a 3 Å cutoff , and taking the centroid frame of the largest cluster for each ligand . The initial structure for the PD-L1 clustering MDs was taken as the structure of the bound human ligand from the co-crystal complex with murine PD-1 ( PDB ID: 3BIK ) . As there are currently no available crystal structures of human PD-L2 , a homology model was built as a starting point for the clustering simulation by manually mutating the bound structure of murine PD-L2 ( PDB ID: 3BP5 ) and minimizing the resulting structure . We used the ClusPro protein-protein docking server ( Comeau et al . , 2004 ) to dock the top apo PD-L1 and PD-L2 centroid structures from their respective MDs to the first three NMR structures of apo human PD-1 ( all three receptor structures are non-bound-like ) . Three bound-like candidate models for the PD-1–PD-L1/2 encounter complexes that correctly anchored Tyr123/112 were chosen from the ClusPro output . We then simulated these encounter complexes for 400 ns to probe the dynamics of the induced fit binding pathway . We ran MDs using AMBER14’s ( Case et al . , 2014 ) pmemd . cuda module ( Götz et al . , 2012 ) and the AMBER ff12SB force field . The cutoff for non-bonded interactions was set at 10 Å . Systems were simulated in an octahedral TIP3P water box with periodic boundary conditions and a 12 Å buffer around the solute . Cl ions were added to the solvent to neutralize the charge of the systems . We minimized each system twice and then equilibrated them before beginning production runs . In the first minimization , solute atoms were held fixed through 500 steps of steepest descent and 500 steps of conjugate gradient minimization . In the second minimization , only the solute backbone atoms were restrained through 2000 steps of steepest descent and 3000 steps of conjugate gradient . After minimization , system temperatures were raised to 300 K over the course of a 200 ps constant volume simulation ( with an integration step of 2 fs ) during which the solute was fixed with weak ( 10 . 0 kcal/mol ) restraints . Bonds involving hydrogens were held at constant length . For the production MDs , the 200–400 ns simulations were held at 300 K under constant pressure with the constraints as listed above for each system and an integration step size of 2 fs . The PyMOL Molecular Graphics System v1 . 7 . 4 . 0 was used for structure preparation and analysis ( Schrödinger , 2010 ) . Trajectories were analyzed using VMDv1 . 9 . 2 ( Humphrey et al . , 1996 ) and the MDpocket software package v2 . 0 ( Le Guilloux et al . , 2009; Schmidtke et al . , 2010 ) for cavity detection and volume/surface area measurement . Measurements of PD-1 binding pocket occlusion , shown in Figures 2d and 4d , e , were calculated from molecular dynamics simulations of PD-1 using a Python script ( available at https://github . com/npabon/md_pocket_occlusion; a copy is archived at https://github . com/elifesciences-publications/md_pocket_occlusion [Pabon , 2016] ) . Briefly , the script takes a molecular dynamics trajectory and a set of static reference atoms and identifies which reference atoms are overlapped in each frame of the simulation . Overlap occurs when any simulated atom crosses the ‘clash radius’ of a reference atom , the clash radius being equal to the sum of the van der Waals radii of the two atoms . The output of the script is the fractional occlusion of each reference atom position , equal to the percentage of simulation frames in which that reference atom is overlapped by simulated atoms . This script was used to evaluate the extent to which the Trp110 and Tyr112/123 binding cavities are open in simulations of PD-1 interaction with various peptides , simulations of apo PD-1 , and the apo NMR ensemble of PD-1 . We classified PD-1 interface conformations using two binary order parameters that define whether interface residues Asn66 and Ile126 are in their ‘open’ or ‘closed’ rotamer states . These parameters are used to distinguish the non-bound-like interface , where Asn66 is closed and Ile126 is open , from the PD-L1-specific bound-like state , where Asn66 is open and Ile126 is closed , and the PD-L2-specific bound-like state , where both Asn66 and Ile126 are open ( Figure 2e ) . We estimated the energy differences ΔGBLapo and ΔGBLEC ( Figure 1 ) using Maxewell-Boltzmann statistics by assessing the bound-like ( BL ) and non-bound-like ( NBL ) state population distributions in the apo and encounter complex ( EC ) receptor ensembles: ( 1 ) ⟨nBLapo/EC⟩⟨nNBLapo/EC⟩=e−ΔGBLapo/ECkBT In the above equations , ⟨nBLapo/EC⟩ and ⟨nNBLapo/EC⟩ denote fractional equilibrium populations of the apo / encounter complex receptor ensembles in the bound-like and non-bound-like macrostates , and kBT is the product of the Boltzmann constant and temperature . We used MDs to generate the equilibrium ensembles of receptor conformations and analyzed the trajectories to calculate ⟨nBL/NBLapo/EC⟩ values . MDs trajectories were analyzed as follows . Reference structures for the open and closed states of Asn66 were defined using its side chain configuration in the first apo NMR model and PD-L1-bound human cocrystal , respectively ( Asn66 has <0 . 2 Å heavy atom RMSD between PD-L1 and PD-L2 cocrystals 4ZQK and 3BP5 ) . Each frame of the MDs trajectory is labeled with the state to which the simulated Asn66 had the smaller side chain RMSD to the reference structure . Reference structures for the open and closed states of Ile126 were defined using its X1 rotamer angle in the murine PD-L2 and human PD-L1 cocrystals , respectively , this angle being the main distinguishing feature between the two different ligand-bound interfaces ( Figure 2e ) . Each frame of the MDs was labeled with the state to which the simulated Ile126 had the closest rotamer angle . The free energy changes of opening Asn66 and Ile126 are calculated using Equation ( 1 ) and then compared across different simulations in order to identify triggers of interface deformations . | Many proteins need to interact with other proteins to carry out their various tasks in cells . Such interactions are essential for almost all biological processes and are often disrupted in disease . Cells have thousands of different types of proteins and each has a unique shape that determines which other proteins it can bind to . It was previously thought that two proteins bind to each other in a manner similar to that of a lock and a key , in which the rigid shape of one protein meshes perfectly with the rigid shape of its partner . However , many proteins are flexible and adopt different shapes depending on whether they are attached to their partner , or not . Moreover , an individual protein may also bind to several different partners , each requiring that protein to adopt several different shapes . These observations have challenged the lock and key model and suggest that flexibility in the structure of a protein plays a key role in its binding to other proteins . However , it is not clear how structural flexibility enables a protein to bind to several different partners while being selective enough to prevent the protein from binding to the wrong ones . A protein called PD-1 is involved in immune responses in humans and is an emerging target for drugs to treat cancer . Pabon and Camacho used computer simulations to model PD-1’s structural flexibility and to find out how this enables the protein to form different shapes when it binds to different partners . The experiments show that the region of PD-1 that binds to other proteins adopts a different shape in the absence and presence of its partners . The binding partners make initial contact with PD-1 via specific features that they share in common . This causes PD-1 to change shape , uncovering a surface of PD-1 that is flexible and is able to accommodate a variety of partners . After this , the binding partners form additional contacts with PD-1 that are specific to each partner . These findings suggest that the ability of a protein to bind to several different partners is unlocked by certain structures that are present in the binding partners . These structures are found in proteins produced by many different organisms , suggesting that this mechanism is likely to be widespread in nature . This work may open up new avenues for designing drugs to target PD-1 and other proteins that contribute to disease but have so far been impossible to target with drugs . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"physics",
"of",
"living",
"systems",
"computational",
"and",
"systems",
"biology"
] | 2017 | Probing protein flexibility reveals a mechanism for selective promiscuity |
Gene expression noise is an evolvable property of biological systems that describes differences in expression among genetically identical cells in the same environment . Prior work has shown that expression noise is heritable and can be shaped by selection , but the impact of variation in expression noise on organismal fitness has proven difficult to measure . Here , we quantify the fitness effects of altering expression noise for the TDH3 gene in Saccharomyces cerevisiae . We show that increases in expression noise can be deleterious or beneficial depending on the difference between the average expression level of a genotype and the expression level maximizing fitness . We also show that a simple model relating single-cell expression levels to population growth produces patterns consistent with our empirical data . We use this model to explore a broad range of average expression levels and expression noise , providing additional insight into the fitness effects of variation in expression noise .
Gene expression is a dynamic process that results from a succession of stochastic biochemical events , including availability of transcription factors , binding of transcription factors to promoter sequences , recruitment of transcriptional machinery , transcriptional elongation , mRNA degradation , protein synthesis , and proteolysis . These events cause the expression level of a gene product to differ even among genetically identical cells grown in the same environment ( Elowitz et al . , 2002; Chong et al . , 2015 ) . This variability in gene expression is known as ‘expression noise’ and is under genetic control ( Raser et al . , 2004; Sanchez and Golding , 2013 ) , with heritable variation causing differences in noise among genes ( Newman et al . , 2006 ) and genotypes ( Murphy et al . , 2007; Hornung et al . , 2012; Fehrmann et al . , 2013; Sharon et al . , 2014; Liu et al . , 2015 ) . Because gene expression noise is heritable and variable within populations , it can evolve in response to natural selection if it affects fitness . Indeed , prior studies have suggested that expression noise can be either beneficial or deleterious depending on the context ( reviewed in Viney and Reece , 2013; Richard and Yvert , 2014; Liu et al . , 2016 ) . For example , Metzger et al . ( 2015 ) provides evidence that increased expression noise can be selected against in natural populations , presumably because elevated noise increases the probability that a given cell produces a suboptimal level of protein expression ( Wang and Zhang , 2011; Duveau et al . , 2017a ) . Consistent with this hypothesis , a negative correlation exists at the genomic scale between the expression noise of genes and their dosage sensitivity ( Fraser et al . , 2004; Batada and Hurst , 2007; Lehner , 2008; Keren et al . , 2016 ) . However , because the optimal level of gene expression can vary among environments , high gene expression noise has been suggested to be beneficial if it can produce individuals with phenotypes that are better adapted to a new environment than individuals produced with low gene expression noise . For instance , noise in gene expression can allow a small fraction of cells to survive when confronted with stressful environmental conditions ( Blake et al . , 2006; Fraser and Kaern , 2009; Ito et al . , 2009; Levy et al . , 2012; Viney and Reece , 2013; Liu et al . , 2015; Wolf et al . , 2015 ) . Consistent with this idea , a genomic screen in yeast found that plasma-membrane transporters involved in cell-environment interactions displayed elevated expression noise in yeast ( Zhang et al . , 2009 ) . Theoretical work also suggests the existence of cost-benefit tradeoffs that can make expression noise either beneficial or deleterious under different circumstances ( Tănase-Nicola and ten Wolde , 2008 ) . Despite a growing body of evidence that selection has acted on expression noise for many genes , direct measurements of how changes in expression noise impact fitness remain scarce ( Liu et al . , 2016 ) . A major reason for this scarcity is that most mutations that alter gene expression noise also alter average expression level ( Newman et al . , 2006; Hornung et al . , 2012; Carey et al . , 2013; Sharon et al . , 2014 ) , making it difficult to disentangle the fitness effects of changing expression noise and average expression level . Here , we directly estimate the effects of changing expression noise on fitness independently from changes in average expression level for the TDH3 gene of Saccharomyces cerevisiae . TDH3 encodes an isozyme of the yeast glyceraldehyde-3-phosphate dehydrogenase ( GAPDH ) involved in glycolysis and gluconeogenesis ( McAlister and Holland , 1985 ) as well as transcriptional silencing ( Ringel et al . , 2013 ) , RNA-binding ( Shen et al . , 2014 ) and possibly antimicrobial defense ( Branco et al . , 2014 ) . Variation in this gene’s promoter affecting expression noise has previously been shown to be a target of selection in natural populations ( Metzger et al . , 2015 ) . To assess the impact of changes in expression noise on fitness at different expression levels , we generated mutant alleles of the TDH3 promoter that covered a broad range of average expression levels and expression noise . We find that increases in expression noise are detrimental when the average expression level of a genotype is close to the fitness optimum , but beneficial when the average expression level of a genotype is further from this optimum . This pattern was reproduced by a simple computational model that links expression in single cells to their doubling time to predict population fitness . We used this individual-based model to explore the fitness effects of a broader combination of average expression levels and expression noise than were explored empirically , showing that not only do the fitness effects of changing expression noise depend on the average expression level , but that the fitness effects of changing average expression level also depend upon the amount of expression noise .
To disentangle the effects of changes in average expression level and expression noise on fitness , we examined a set of TDH3 promoter ( PTDH3 ) alleles with a broad range of activities . For each allele , we measured the average expression level and expression noise by cloning the allele upstream of a yellow fluorescent protein ( YFP ) coding sequence , integrating this reporter gene ( PTDH3-YFP ) into the HO locus , and quantifying fluorescence in living cells using flow cytometry in six replicate populations per genotype ( Figure 1A ) . The fluorescence value of each cell was transformed into an estimated mRNA level ( Figure 1A ) based on the relationship between fluorescence and YFP mRNA abundance ( Figure 1B , C ) . The average expression level of a genotype was then calculated by averaging the median values from the six replicates ( Figure 1A ) and expressing this value as a percent change from the wild type allele . Expression noise was calculated for each replicate as the variance divided by the median expression among cells , a measure of noise strength similar to the Fano factor ( Sanchez and Golding , 2013 ) . The expression noise of each genotype was then calculated by averaging the noise strength from the six replicate populations , and this value was expressed as a percent change from the wild type allele . The main conclusions of this study are robust to the choice of noise metric , as shown in supplementary figures using three alternative metrics of noise . Effects of 236 point mutations in the TDH3 promoter , including mutations in RAP1 and GCR1 transcription factor binding sites ( TFBS ) , have previously been described that cause a wide range of average expression levels and expression noise values ( Metzger et al . , 2015 ) . But average expression level and expression noise strongly co-vary among these alleles ( Metzger et al . , 2015 ) , making them insufficient for separating the effects of changes in average expression level and expression noise on fitness . We therefore sought to construct additional promoter alleles that showed a different relationship between average expression level and expression noise . First , we inserted a recognition motif for the GCN4 transcription factor at ten different positions in the TDH3 promoter because this TFBS was previously found to affect the relationship between expression level and expression noise ( Sharon et al . , 2014 ) . However , the insertion of GCN4 binding sites into PTDH3 did not show the expected departure from the relationship between expression level and expression noise observed for mutations in GCR1 and RAP1 TFBS ( Figure 1—figure supplement 1 ) . We next mutated the PTDH3 TATA box because previous studies showed that TATA box mutations confer lower expression noise for a given expression level when compared to other types of promoter alterations ( Blake et al . , 2006; Mogno et al . , 2010; Hornung et al . , 2012 ) . We generated 112 alleles of the TDH3 promoter that had between one and five random mutations in the TATA box sequence , which caused the expected lower levels of expression noise than TFBS mutant alleles with similar average expression levels ( Figure 1D ) . We then combined mutations in the TATA box , GCR1 TFBS and/or RAP1 TFBS to further increase the range of expression phenotypes . Finally , we constructed alleles containing two tandem copies of the PTDH3-YFP reporter gene with or without mutations in the PTDH3 sequence to sample expression levels closer to and above the wild-type allele . These mutant alleles captured a much greater range of mean expression and expression noise than TDH3 promoter alleles segregating in natural populations ( Metzger et al . , 2015; Duveau et al . , 2017a ) and allowed us to more fully explore the relationship between mean , noise and fitness than would be possible using naturally occurring variation alone . From this collection of 171 TDH3 promoter alleles ( Figure 1—figure supplement 1 , Figure 1—figure supplement 1—source data 1 ) , we selected 43 alleles ( Source data 1 ) to study the fitness effects of changes in average expression level and expression noise of the native TDH3 gene . The average expression level conferred by these 43 PTDH3 alleles ( including the wild type allele of PTDH3 ) ranged from 0% to 207% of the wild type allele and the expression noise ranged from 3% to 371% of the wild type allele ( Figure 1D ) . Most importantly , this set of alleles showed variation in expression noise independent of expression level at expression levels between 0% and 125% of the wild type allele ( Figure 1D ) . To measure the fitness effects of changing TDH3 expression , we introduced each of these 43 PTDH3 alleles upstream of the TDH3 coding sequence at the native TDH3 locus and performed competitive growth assays similar to those described in Duveau et al . ( 2017a ) ( Figure 2A ) . For each of the eight PTDH3 alleles that contained a duplication of the PTDH3-YFP reporter gene , we created a duplication of the entire TDH3 gene with the two corresponding PTDH3 alleles . We also included a strain with a deletion of the promoter and coding sequence of TDH3 to sample a TDH3 expression level of 0% . Prior studies have found that deletion of TDH3 causes a moderate decrease in fitness in glucose-based media: −5% in Pierce et al . ( 2007 ) and −6 . 8% in Duveau et al . ( 2017a ) . Each strain tested was marked with YFP and grown competitively for 30 hr ( ~21 generations ) with a reference genotype marked with a green fluorescent protein ( GFP ) ( Figure 2A ) . Competitive fitness was determined from the rate of change in genotype frequencies over time and averaged across at least six replicate populations for each genotype tested ( Figure 2B ) . The relative fitness of each strain was then calculated by dividing the competitive fitness of that strain by the competitive fitness of the strain with the wild type allele of TDH3 ( Source data 1 ) . This protocol provided a measure of fitness with an average 95% confidence interval of 0 . 2% . We then related these measures of relative fitness to the expression of the reporter gene driven by these PTDH3 alleles at the HO locus . Expression of this reporter gene provided a reliable readout of average expression level and expression noise driven by the same PTDH3 promoters at the native TDH3 locus , as measured using Tdh3-YFP fusion proteins ( Figure 2—figure supplement 1A , B ) . These fusion protein alleles were not used for comparing fitness effects among TDH3 promoter alleles because the YFP fusion reduced fitness by 2 . 5% relative to a strain expressing TDH3 and YFP from independent promoters ( Figure 2—figure supplement 1C ) . A local regression ( LOESS ) of average expression level on fitness for the 43 TDH3 alleles and the TDH3 deletion showed a non-linear relationship with a plateau of maximal fitness near the wild type expression level ( Figure 2C ) similar to that described in Duveau et al . ( 2017a ) . Deletion of TDH3 ( expression level of 0% in Figure 2C ) caused a statistically significant decrease in fitness of 6 . 1% relative to the wild type allele ( t-test , p=6 . 4×10−10 ) . The minimum change in TDH3 expression level that significantly impacted fitness was a 14 . 6% decrease in average expression relative to wild type , which reduced fitness by 0 . 19% ( t-test , p=0 . 00045 ) . Overexpressing TDH3 up to 175% did not significantly impact growth rate , but the 207% expression level of the strain carrying a duplication of the wild type TDH3 gene caused a 0 . 92% reduction in fitness ( Figure 2C; t-test , p=1 . 4×10−7 ) . Notably , none of the 42 mutant alleles of TDH3 conferred a significantly higher fitness than the wild type allele ( one-sided t-tests , p>0 . 05 ) , indicating that the wild type expression level of TDH3 is near an optimum for growth in the environment assayed . We expect these differences in fitness among genotypes with different TDH3 promoter alleles to arise primarily from differences in TDH3 expression; however , differences in pleiotropy among promoter alleles might also contribute to differences in fitness . Residual variation was observed around the LOESS fitted line relating expression level to fitness ( Figure 2C ) that we hypothesized might be explained by differences in noise among genotypes . To examine the effects of differences in expression noise on fitness independent of differences in average expression level , we used the residuals from a local regression of expression noise on expression level for the alleles with average expression levels between 0% and 125% of the wild type allele to define a metric called ΔNoise ( Figure 3A; Figure 3—figure supplement 1A–D ) . This metric was not significantly correlated with expression level ( Figure 3—figure supplement 2 ) . TDH3 alleles with positive ΔNoise values had a higher level of noise than expected based on their expression level and were classified as ‘high noise’ , whereas TDH3 alleles with negative ΔNoise values had lower levels of noise than expected given their expression level and were classified as ‘low noise’ . We then compared the relationship between expression level and fitness for genotypes in the high noise and low noise classes ( Figure 3B ) . We found that promoter alleles with positive ΔNoise tended to show a higher fitness than strains with negative ΔNoise ( Figure 3B , Figure 3—figure supplement 1E–H ) . This beneficial effect of noise on fitness was surprising given prior evidence that selection favored alleles of PTDH3 with low expression noise in natural populations ( Metzger et al . , 2015 ) . We noticed , however , that the fitness benefit of increasing expression noise was limited to a particular range of average expression levels . Specifically , positive ΔNoise was associated with higher fitness only for average expression levels between 2% and 80% of the wild type expression level ( Figure 3B ) . Above 80% of expression , no clear difference in fitness was detected between strains with positive and negative ΔNoise ( Figure 3B ) . These same trends were also observed for the other metrics of noise ( Figure 3—figure supplement 1E–H ) . Based on these observations and prior theoretical work ( Tănase-Nicola and ten Wolde , 2008 ) , we hypothesized that the distance between the average expression level of a PTDH3 allele and the optimal level of TDH3 expression influenced how a change in expression noise impacted fitness . To test this hypothesis , the 43 promoter alleles were split into two groups depending on the distance of their average expression level from the optimal expression level of TDH3 . Using a local regression of fitness on average expression level , we inferred the value of average expression that would confer a fitness reduction of 0 . 5% from maximal fitness . Promoter alleles for which the median activity was below this threshold were considered to be ‘far from optimum’ and promoter alleles with median activity above the threshold were considered to be ‘close to optimum’ ( Figure 3C ) . A metric called ΔFitness , corresponding to the residuals of a local regression of fitness on average expression , was calculated to remove the confounding effect of average expression levels on fitness ( Figure 3C , Figure 3—figure supplement 1I–L , Figure 3—figure supplement 3 ) . We found that changes in noise ( ΔNoise ) and changes in fitness ( ΔFitness ) were positively correlated for genotypes classified as far from the optimum ( Pearson correlation coefficient: r = 0 . 74 , p=9 . 36×10−7 , Figure 3D , Figure 3—figure supplement 4A–D ) , but not for genotypes classified as close to the optimum ( r = −0 . 08 , p=0 . 84 , Figure 3E , Figure 3—figure supplement 4E–H ) . This result was robust to variation in the choice of the smoothing parameter used for the local regression of noise on average expression , the choice of the smoothing parameter used for the local regression of fitness on average expression , and the fitness threshold used to separate strains with expression levels close and far from optimum ( Figure 3—figure supplements 5 , 6 and 7 ) . We note , however , that the smaller number of genotypes with mean expression close to the optimum provided less power to detect a significant relationship than genotypes with mean expression far from the optimum . As an alternative way to test for the impact of expression noise on fitness , we compared ΔFitness for genotypes with positive and negative values of ΔNoise . Permutation tests were used to assess the significance of differences in ΔFitness by randomly shuffling values of ΔNoise among genotypes . Consistent with the correlation analyses , genotypes with positive ΔNoise showed a significantly greater median value of ΔFitness than genotypes with negative ΔNoise at expression levels far from optimum ( 105 permutations , PΔNoise ≤ 10−5; Figure 3F ) . Among genotypes with average expression close to optimum , no significant difference in median ΔFitness was detected between the positive and negative ΔNoise groups ( 105 permutations , PΔNoise = 0 . 6442 ) ( Figure 3F ) . The same pattern was observed for all metrics of noise and was not driven by differences in average expression levels between the two ΔNoise groups ( Figure 3—figure supplement 8 ) . The results presented in the preceding section provide strong evidence that variation in TDH3 expression can directly affect fitness , but the methods used have at least two limitations . First , ΔNoise and ΔFitness values can be influenced by the set of PTDH3 alleles included in the analyses since they are regression residuals . Second , comparisons of fitness among PTDH3 genotypes rely upon the assumption that fitness effects are transitive , i . e . that differences in fitness between two strains are accurately reported by competitive growth against a third reference strain . Even though such transitivity has often been verified ( de Visser and Lenski , 2002; Elena and Lenski , 2003; Gallet et al . , 2012 ) , intransitive competition has been observed in several organisms , including yeast ( Paquin and Adams , 1983 ) . To test whether differences in TDH3 expression noise affect fitness without calculating regression residuals and without assuming transitivity , we performed direct competition assays between strains with PTDH3 promoter alleles that showed similar average expression levels but different levels of expression noise . Five pairs of TDH3 alleles for which ( i ) the median level of activity was similar between the two promoters of each pair , ( ii ) the level of noise was different between the two promoters of each pair , and ( iii ) the median level of activity varied among different pairs were chosen from the full set of 171 alleles described above ( Figure 4A; Figure 4—source data 1 ) . The promoter variants of four of these pairs were included in the indirect competition assays and showed the general pattern of increased fitness with increased expression noise when average expression was far from optimum and no significant difference in fitness despite differences in expression noise when average expression was close to optimum ( Figure 4B ) . Promoters in the fifth pair were not among the 43 alleles included in the indirect competition experiment but were selected for the direct competition assays because they showed variation in expression noise at average expression levels close to wild type ( purple points in Figure 4A ) . For each of the five pairs , the low noise genotype and the high noise genotype were directly competed against each other under the same conditions used in the competitive growth fitness assay described above except that we doubled the number of generations and the number of replicates to increase the sensitivity of our fitness estimates . In addition , we used pyrosequencing ( Neve et al . , 2002 ) instead of flow cytometry to determine the relative frequency of the two genotypes at each time point because the two strains competed against each other could not be distinguished based on fluorescence . Relative fitness of the high and low noise genotypes in each pair was calculated based on the changes in relative allele frequency during competitive growth . For the three pairs of genotypes with an average expression level far from optimum ( 12% , 19% , and 59% average expression relative to wild type ) , fitness of the high noise genotype relative to the low noise genotype was significantly greater than 1 ( Figure 4C ) , indicating that the high noise genotype grew faster than the low noise genotype . This result was consistent with the differences in fitness measured from the indirect competition assays ( Figure 4B ) . By contrast , both pairs of strains with an average expression level closer to the fitness optimum ( 93% and 102% relative to wild type expression levels ) showed slightly but significantly lower fitness of the high noise genotype than the low noise genotype ( Figure 4C ) . In these cases , higher expression noise resulted in a ~ 0 . 1% decrease in fitness relative to lower noise . This difference was detectable with the direct competition assay because the average span of the 95% confidence intervals of fitness estimates was 0 . 1% , which is half of the 0 . 2% average 95% confidence intervals from the indirect competition assay described above . Taken together , our empirical measures of relative fitness show that higher expression noise for TDH3 is beneficial when average expression level is far from the optimum , but deleterious when average expression is close to the optimum . An intuitive explanation of this phenomenon is that when the average expression level is close to the optimum , increasing expression noise can result in enough cells with suboptimal expression to decrease fitness of the population . Conversely , when the average expression level is far from the optimum , increasing expression noise can result in enough cells with expression closer to the optimum to increase fitness of the population . These effects of expression noise on population fitness can result from differences in expression level among cells causing differences in the cell division rate ( a . k . a . doubling time ) among cells ( Kiviet et al . , 2014 ) . To better understand the interplay among average expression level , expression noise , and fitness , we developed a simple computational model that allowed us to ( 1 ) vary the expression mean and noise independently while holding all other parameters constant , ( 2 ) track the resulting single cell growth dynamics , and ( 3 ) evaluate the consequences for population fitness . To further investigate how the distribution of expression levels among genetically identical cells influences population fitness , we modeled the growth of clonal cell populations that differed in the mean expression level and expression noise for a single gene . In this model , we specified a function defining the relationship between the expression level of a cell and the doubling time of that cell . Following each cell division , the expression levels of mother and daughter cells were sampled independently from an expression distribution characterized by its mean and noise ( Figure 5A ) . This independent sampling ignores any inheritance of expression noise , which is a conservative choice for detecting differences in fitness among genotypes due to differences in noise . The doubling time of each cell was then calculated from its expression level ( Figure 5B ) , and each clonal population was allowed to expand for the same amount of time , increasing in size at a rate determined by the doubling times of the cells sampled ( Figure 5C ) . Empirical measures of single-cell division rates were consistent with these elements of the model , showing more variable cell division times in genotypes with greater TDH3 expression noise and shorter cell division times in genotypes with mean TDH3 expression closer to the fitness optimum ( Figure 5—figure supplement 1 ) . Competitive fitness was ultimately determined in the model by comparing the population size obtained at the end of each simulation experiment to the population size obtain for a constant ‘wild type’ competitor ( Figure 5D , Figure 6—source data 1 ) . 100 independent simulations were performed for each unique combination of mean expression level and expression noise . Three metrics of expression noise were used for this work: noise strength ( similar to Fano factor , Figure 6 ) , standard deviation ( Figure 6—figure supplement 1A , C ) and coefficient of variation ( Figure 6—figure supplement 1B , D ) . To calculate doubling times from single cell expression levels , we first used a linear function akin to directional selection in which increases in expression level resulted in shorter doubling times ( faster growth ) ( Figure 6A ) . With this relationship , higher levels of expression noise conferred higher population fitness for a given mean expression level ( Figure 6B ) , a pattern more pronounced for high values of mean expression and observed for all metrics of noise ( Figure 6—figure supplement 1A , B ) . This finding is consistent with prior work demonstrating that an increased variability of doubling time among individual cells is sufficient to increase fitness at the population level ( Tănase-Nicola and ten Wolde , 2008; Cerulus et al . , 2016; Hashimoto et al . , 2016; Nozoe et al . , 2017 ) . This is because the doubling time of a population tends to be dominated by the doubling time of the fastest dividing cells in the population , i . e . population doubling time is higher than the mean doubling time among all cells in the population . Next , we used a Gaussian function akin to stabilizing selection in which an intermediate expression level produced the shortest doubling time ( faster growth ) , while lower or higher expression than this optimum would increase doubling time ( slower growth ) ( Figure 6C ) . With this function , we found that the fitness effects of increasing expression noise depended on the mean expression level . Specifically , increasing expression noise increased fitness when the average expression level was far from the optimal expression level and it decreased fitness when the average expression level was close to the optimum ( Figure 6D ) , similar to the pattern we observed with our empirical fitness data and in agreement with theoretical work by Tănase-Nicola and ten Wolde ( 2008 ) . This result was observed for all three metrics of noise , suggesting it is robust to the different scaling relationships between the mean expression level and variability around the mean captured by different metrics of noise ( Figure 6—figure supplement 1C , D ) . These in silico analyses not only provide a plausible mechanistic explanation for our empirical finding that increasing noise can be both beneficial and deleterious in a single environment but they also show that increasing expression noise can alter the effects of changes in mean expression level on fitness . Specifically , when expression noise is high ( red lines on Figure 6D and Figure 6—figure supplement 1C , D ) , changes in mean expression level are predicted to have much smaller impacts on fitness than equivalent changes when expression noise is low ( blue lines on Figure 6 and Figure 6—figure supplement 1C , D ) . This pattern is also readily apparent when changes in expression noise , instead of changes in mean expression level , are plotted as a function of population fitness ( Figure 6—figure supplement 2 ) . These observations are consistent with a previously published population genetic model showing that increasing expression noise can reduce the efficacy of natural selection acting on mean expression level ( Wang and Zhang , 2011 ) . Despite many studies providing evidence that natural selection can ( Tănase-Nicola and ten Wolde , 2008; Wang and Zhang , 2011; Barroso et al . , 2018 ) and has ( Fraser et al . , 2004; Lehner , 2008; Zhang et al . , 2009; Metzger et al . , 2015 ) acted on expression noise , the precise effects of expression noise on fitness have proven difficult to measure empirically . This difficulty arises from the facts that ( 1 ) most mutations that alter expression noise also alter mean expression in a correlated fashion , making it difficult to isolate the effects of changes in expression noise on fitness ( Hornung et al . , 2012; Keren et al . , 2016; Liu et al . , 2016 ) , and ( 2 ) the magnitude of fitness effects resulting from changes in expression noise is expected to be smaller than that resulting from changes in mean expression level ( Zhang et al . , 2009 ) . In this study , we overcame these challenges by surveying a broad range of mutant promoter alleles for their effects on mean expression level and expression noise , measuring the fitness effects of a subset of these alleles with reduced dependency between effects on mean expression level and expression noise , and using an assay for fitness with power to detect changes as small as 0 . 1% . We found that the fitness effects of changes in expression noise are indeed generally much smaller than changes in expression level , although they are large enough to be acted on by natural selection in wild populations of S . cerevisiae ( Wagner , 2005; Metzger et al . , 2015 ) . We also show that changes in expression noise can be beneficial or deleterious depending on the distance between the mean expression level and the expression level conferring optimal fitness in the environment examined , with increases in expression noise deleterious near the optimal expression level , consistent with data for TDH3 in Metzger et al . ( 2015 ) . Although our empirical work focused solely on the TDH3 gene , the small number of parameters in our simulation model producing the same pattern as these empirical data suggests that the observed relationship among fitness , average expression level and expression noise are likely generalizable to other genes . That said , the precise relationship between expression noise and fitness at the population level is expected to be shaped by the relationship between average expression level and doubling time of single cells as well as the temporal dynamics of expression in single cells ( Blake et al . , 2006; Tănase-Nicola and ten Wolde , 2008 ) . We provide some experimental measures of single-cell division rates here ( Figure 5—figure supplement 1 ) , but studies that more directly compare expression levels and division times in individual cells are needed to fully address this issue . Assuming that the average expression level of a population is near the fitness optimum in a stable environment , but further from the optimum following a change in the environment , our results unify studies showing that increasing expression noise tends to be deleterious in a constant environment but beneficial in a fluctuating one ( Fraser et al . , 2004; Blake et al . , 2006; Batada and Hurst , 2007; Lehner , 2008; Tănase-Nicola and ten Wolde , 2008; Zhang et al . , 2009; Fraser and Kaern , 2009; Ito et al . , 2009; Wang and Zhang , 2011; Levy et al . , 2012; Wolf et al . , 2015; Liu et al . , 2015; Keren et al . , 2016 ) . Expression noise may be particularly important in the early phase of adaptation to a fluctuating environment , when a new expression optimum makes an increase in noise beneficial and before expression plasticity evolves as a more optimal strategy ( Wolf et al . , 2015 ) . Such plasticity in expression level seems to have already evolved for TDH3 ( Duveau et al . , 2017b ) . Our data suggest that high levels of expression noise can also be beneficial in a stable environment when the mean expression level is far from optimal . For example , if an allele driving suboptimally low expression were to be fixed in a population , selection should initially favor alleles that increase mean expression and/or expression noise . After alleles driving mean expression close to the optimum are fixed , selection should then favor alleles with lower levels of expression noise . The relative frequency by which evolution proceeds through these two paths will depend on both the relative frequency of alleles that increase mean expression and expression noise , as well as the fitness differences between these alleles . We note , however , that the often correlated effects of promoter mutations on mean expression level and expression noise ( Hornung et al . , 2012; Carey et al . , 2013; Sharon et al . , 2014; Vallania et al . , 2014 ) may limit the ability of natural selection to optimize both mean expression level and expression noise . Future work investigating the effects of other types of mutations on mean expression level , expression noise , and fitness in multiple environments is needed to more fully define the range of variation affecting gene expression upon which natural selection can act .
All strains used in this work were haploids with similar genetic backgrounds that were derived from crosses between BY4724 , BY4722 , BY4730 and BY4742 ( Brachmann et al . , 1998 ) and carry the alleles RME1 ( ins-308A ) ; TAO3 ( 1493Q ) from Deutschbauer and Davis ( 2005 ) and SAL1; CAT5 ( 91M ) ; MIP1 ( 661T ) from Dimitrov et al . ( 2009 ) that contribute to increased sporulation efficiency and decreased petite frequency relative to the alleles of the laboratory S288c strain . The construction of this genetic background is described in more detail in Metzger et al . ( 2016 ) . Strains used to assay transcriptional activity and fitness ( described in detail below ) had different mating types and drug resistance markers , but these differences did not significantly affect PTDH3 transcriptional activity ( Figure 2—figure supplement 2A , B ) . Transcriptional activity ( average expression level and expression noise ) was assayed for 171 PTDH3 alleles in S . cerevisiae strains carrying a fluorescent reporter construct inserted at the HO locus on chromosome IV in MATα cells ( Metzger et al . , 2016 ) . From these alleles ( Figure 1—figure supplements 1—source data 1 ) , 43 were selected for assaying fitness effects of changing TDH3 expression ( Source data 1 ) . 36 of the final 43 PTDH3 alleles carried a single copy of a reporter construct consisting of the TDH3 promoter followed by the Venus YFP coding sequence , the CYC1 terminator and an independently transcribed KanMX4 drug resistance cassette Metzger et al . ( 2016 ) . 7 of the final 43 PTDH3 alleles variants consist of two copies of the PTDH3-YFP-TCYC1 construct in tandem separated by a URA3 cassette . The different PTDH3 alleles contain mutations located either in the known binding sites for GCR1 and RAP1 transcription factors , in the TATA box or in combinations of both , as described below . The wild type allele of PTDH3 consists of the 678 bp sequence located upstream of the TDH3 start codon in the genome of the laboratory strain S288c , with a single nucleotide substitution that occurred during the construction of the PTDH3-YFP-TCYC1 construct ( A - > G located 293 bp upstream of the start codon ) . This substitution is present in all PTDH3 alleles used in this study . The effect of this mutation on PTDH3 activity in YPD medium was previously described ( Metzger et al . , 2015 ) . A set of 236 point mutations corresponding to almost all C - > T and G - > A substitutions in the TDH3 promoter was previously inserted upstream of a YFP reporter gene on chromosome I in the BY4724 genetic background ( Metzger et al . , 2015 ) . From these , we selected seven PTDH3 alleles for which the transcriptional activity spanned a broad range of median fluorescence levels when cells were grown in glucose medium ( 25% to 90% relative to wild type expression level ) . These seven promoters carried mutations either in the GCR1 or RAP1 transcription factor binding sites ( TFBS ) previously characterized in the TDH3 promoter ( Yagi et al . , 1994 ) . Each PTDH3 allele was inserted upstream of YFP at the HO locus using the dellitto perfetto approach ( Stuckey et al . , 2011 ) . Briefly , in the reference strain YPW1002 carrying the wild-type PTDH3-YFP-TCYC1 construct at HO ( Metzger et al . , 2016 ) , we replaced PTDH3 with a CORE-UH cassette ( COunterselectable REporter URA3-HphMX4 amplified from plasmid pCORE-UH using oligonucleotides 1951 and 1926 in Supplementary file 2 ) to create strain YPW1784 . Then , each of the seven PTDH3 alleles was amplified by PCR using oligonucleotides 2276 and 2277 ( Supplementary file 2 ) and transformed into YPW1784 to replace the CORE-UH cassette and allow expression of YFP ( Metzger et al . , 2015 ) . Correct insertion of PTDH3 alleles was verified by Sanger sequencing of PCR amplicons obtained with primers 2425 and 1208 ( Supplementary file 2 ) . To sample average expression levels less than 25% of wild type , we created and measured the activity of 12 PTDH3 alleles containing mutations in two different TFBS . We then selected seven of these alleles to be included in the final set of 43 PTDH3 alleles ( Source data 1 ) . Point mutations from different alleles were combined on the same DNA fragment using PCR SOEing ( Splicing by Overlap Extension ) . First , left fragments of PTDH3 were amplified from genomic DNA of strains carrying the most upstream TFBS mutations . These PCRs used a common forward primer ( 2425 in Supplementary file 2 ) and a reverse primer containing the most downstream TFBS mutation to be inserted ( P4E8 , P4E5 , P4G8 or P4G7 in Supplementary file 2 ) . In parallel , right fragments of PTDH3 were amplified from YPW1002 gDNA using forward primers containing the most downstream TFBS mutations ( P1E8 , P1E5 , P1G8 or P1G7 in Supplementary file 2 ) and a common reverse primer ( 104 in Supplementary file 2 ) . Then , equimolar amounts of the overlapping upstream and downstream fragments of PTDH3 were mixed and 25 PCR cycles were performed to fuse both fragments together and to reconstitute the full promoter . Finally , the fused fragments were further amplified for 35 cycles using oligonucleotides 2425 and 1305 ( Supplementary file 2 ) and the final products were transformed in strain YPW1784 . The presence of desired mutations in PTDH3 was confirmed by Sanger sequencing of amplicons obtained with primers 1891 and 1208 ( Supplementary file 2 ) . To try to create variation in noise independent of the median expression level , we inserted GCN4 binding sites at several locations in the TDH3 promoter because GCN4 binding sites in synthetic promoters were shown to increase expression noise ( CV2 ) relative to average expression level ( Sharon et al . , 2014 ) . We introduced substitutions in PTDH3 to create the GCN4 binding motif TGACTCA at 10 different locations ( −121 , –152 , −184 , –253 , −270 , –284 , −323 , –371 , −407 and −495 upstream of start codon ) that originally differed by one , two or three nucleotides from this motif . Targeted mutagenesis was performed using the same PCR SOEing approach as described in Metzger et al . ( 2015 ) ( see Supplementary file 2 for the list of primers used to insert GCN4 binding sites ) and the resulting PCR products were transformed into strain YPW1784 . Correct insertion of the TGACTCA motif was confirmed by Sanger sequencing . However , none of the 10 alleles of PTDH3 with GCN4 binding sites showed the expected increase in expression noise when cells were grown in glucose ( Figure 1—figure supplement 1 ) . This could be due to the genomic context being different from the synthetic library used in Sharon et al . ( 2014 ) or to the fact that PTDH3 is one of the most highly active promoters in the yeast genome . None of these 10 alleles were included in the set of 43 PTDH3 alleles used for fitness assays . A second strategy we employed to create variation in expression noise independent of median expression was to mutate the TATA box in the TDH3 promoter because the presence of a canonical TATA box in yeast promoters has been associated with elevated expression noise ( Newman et al . , 2006 ) . Mutations in the TATA box were also shown to have a clearly distinct impact on expression noise compared to other types of cis-regulatory mutations ( Blake et al . , 2006; Hornung et al . , 2012 ) . We used a random mutagenesis approach to create a large number of alleles with one or several mutations in the PTDH3 TATA box . Variants were obtained using PCR SOEing as described above , except that one of the internal overlapping oligonucleotides ( primer 2478 , Supplementary file 2 ) used to amplify the downstream fragment of PTDH3 contained a degenerate version of the wild type TATA box ( TATATAAA at position −141 upstream of start codon ) . This oligonucleotide was synthesized by Integrated DNA Technologies using hand-mixed nucleotides at the eight bases of the TATA box with a proportion of 73% of the wild type nucleotide and 9% of each of the three alternative nucleotides . At this level of degeneracy , ~10% of the DNA molecules should carry no mutation , ~25% should carry a single mutation in the TATA box , ~35% two mutations , ~20% three mutations and ~10% four mutations or more . The degenerate primer was used with oligonucleotide 104 to amplify the right fragment of PTDH3 , and the overlapping primer 2479 was used with oligo 2425 to amplify the left fragment ( Supplementary file 2 ) . Then , these fragments were fused and amplified as described above for the TFBS mutants . Six independent transformations of the fragments containing random mutations in the TATA box were performed in strain YPW1784 to obtain a large number of colonies . After growth on selective medium ( Synthetic Complete medium with 0 . 9 g/L 5-FluoroOrotic Acid ) , 244 colonies selected regardless of their fluorescence level were streaked on fresh plates ( again SC +5 FOA medium ) and then replica plated on YPD +Hygromycin B ( 10 g/L Yeast extract , 20 g/L Peptone , 20 g/L Dextrose and 300 mg/L Hygromycin B ) for negative screening . 106 of the resulting strains turned out not to be fluorescent , among which the vast majority were resistant to Hygromycin B , suggesting they were false positive transformants . The remaining 138 strains were all fluorescent and sensitive to Hygromycin B , as expected from true positive transformants . We then tried to amplify PTDH3 in all 244 strains using oligonucleotides 1891 and 1208 ( Supplementary file 2 ) and we only observed a band of correct size after electrophoresis for the 138 fluorescent strains . After Sanger sequencing of the PCR products for the 138 positive strains , the type and frequency of mutations observed in the TATA box were found to be very close to expectation ( Figure 1—figure supplements 1—source data 1 ) . Average expression level and expression noise were measured for all 138 strains as described below . This set of alleles showed broad variation in average expression level ( Figure 1—figure supplement 1 ) and had a lower expression noise than TFBS mutations with comparable average expression levels . We selected seven TATA box variants ( Source data 1 ) with expression levels ranging from 20% to 75% of wild type to be included in the final set of 43 PTDH3 alleles . One of the random TATA box mutants contained a PCR-induced mutation in the GCR1 . 1 TFBS and was also included in the final set ( Var23 in Source data 1 ) . To obtain variation in expression noise at expression levels below 20% , we combined mutations in TFBS with mutations in the TATA box in 12 additional PTDH3 alleles ( Figure 1—figure supplements 1—source data 1 ) . Two TATA box variants with 25% and 50% median fluorescence levels were each combined with six different TFBS variants for which median expression ranged from 4% to 45% relative to wild type . The 12 variants were created by PCR SOEing as described above for the double TFBS mutants , except in this case oligonucleotides 2425 and 2788 were used to amplify the upstream PTDH3 fragments and oligonucleotides 2787 and 104 were used to amplify the downstream fragments ( Supplementary file 2 ) . All 12 variants were transformed in strain YPW1784 and confirmed by Sanger sequencing . To create variation in average expression level and expression noise for expression levels more than 75% of wild type , we constructed 13 alleles with two copies of the whole PTDH3-YFP-TCYC1 construct inserted in tandem at the HO locus ( Figure 1—figure supplements 1—source data 1 ) . One of these constructs carried two copies of the wild type TDH3 promoter , while the others carried mutated versions of PTDH3 . We reasoned that the presence of a second copy of the construct would lead to overexpression of YFP , as shown previously ( Kafri et al . , 2016 ) , while differences in noise between the different alleles should be conserved . To construct these alleles , we first fused the selectable marker URA3 upstream of the PTDH3-YFP-TCYC1 allele located at the right end of each of the final constructs ( ‘CONSTRUCT . 2’ in Source data 1 ) using PCR SOEing . URA3 was amplified from the pCORE-UH plasmid using oligonucleotides 2688 and 2686 and the 13 PTDH3-YFP-TCYC1 constructs were amplified from the strains carrying the corresponding PTDH3 alleles using oligonucleotides 2687 and 1893 ( Supplementary file 2 ) . URA3 and PTDH3-YFP-TCYC1 were then fused by overlap extension and the resulting fragments were amplified with oligonucleotides 2684 and 2683 ( Supplementary file 2 ) . Finally , each of the 13 different URA3-PTDH3-YFP-TCYC1 PCR products was transformed in the strain carrying the desired allele of PTDH3-YFP-TCYC1 ( strain carrying ‘CONSTRUCT . 1’ in Source data 1 ) . During these transformations , the KanMX4 drug resistance cassette was replaced with URA3-PTDH3-YFP-TCYC1 by homologous recombination so that the final constructs were ho::PTDH3-YFP-TCYC1-URA3-PTDH3-YFP-TCYC1 . To control for the impact of the URA3 marker on the activity of the TDH3 promoter , we constructed strain YPW2675 ( ho::PTDH3-YFP-TCYC1-URA3 ) by replacing the KanMX4 cassette with URA3 amplified using primers 2684 and 2685 ( Supplementary file 2 ) . YPW2675 was used as the reference when reporting the expression phenotypes ( median and noise ) of the alleles with two copies of PTDH3-YFP-TCYC1 . To validate the sequence of the full ( 5 . 2 kb ) constructs , we performed two overlapping PCRs using oligonucleotides 2480 and 1499 , and 1872 and 2635 ( Supplementary file 2 ) . PCR products were sequenced using primers 2480 , 1499 , 1204 , 1872 , 2635 , 2686 , 1305 and 601 in Supplementary file 2 ) to confirm they contained the correct PTDH3 alleles . However , using this PCR approach , insertion of more than two tandem copies of PTDH3-YFP-TCYC1 would remain undetected . Therefore , we used quantitative pyrosequencing to determine the exact number of copies inserted in the 13 strains . We took advantage of the fact that all PTDH3 alleles inserted at HO carried the mutation A293g upstream of the start codon , while the endogenous TDH3 promoter did not . This allowed us to determine the total number of PTDH3 copies at the HO locus by quantifying the relative frequency of A and G nucleotides at position −293 across all copies of the TDH3 promoter in the genome . For instance , if only one copy of PTDH3 is present at HO , then the frequency of G at position −293 is expected to be 0 . 5 , since there is one copy of the G allele at the HO locus and one copy of the A allele at the endogenous TDH3 locus . If two copies are present at HO , a frequency of 2/3 is expected for G , and if three copies are present , a frequency of 0 . 75 is expected . To determine these allele frequencies , we amplified PTDH3 in five replicates from all strains carrying two copies of the construct as well as from YPW2675 carrying a single copy using oligonucleotides 2268 and 3094 ( Supplementary file 2 ) . PCR products were denatured and purified using a PyroMark Q96 Vacuum Workstation ( Qiagen ) and pyrosequencing was performed on a PyroMark Q96 ID instrument using oligonucleotide 2270 for sequencing ( Supplementary file 2 ) . Allele frequencies were determined from the relative heights of the peaks corresponding to the A and G alleles on the pyrograms , with the typical correction factor of 0 . 86 applied to A peaks . Using this method , a small but significant bias toward the G allele was detected , as the observed frequency of G in strain YPW2675 was 0 . 55 instead of 0 . 5 . This could be caused by PCR bias due to the fact that the A and G alleles are located at different genomic positions . We applied the linear correction y = x * ( 0 . 5/0 . 45 ) –0 . 111 to remove the effect of this PCR bias when calculating the frequency of G alleles . Overall , we found that six strains had a frequency of G significantly higher than 2/3 ( t-test , p<0 . 05 ) . This suggested that these strains carried more than two copies of the PTDH3-YFP-TCYC1 construct and they were therefore removed from all subsequent analyses ( except Var42 for reasons explained below ) . Sanger sequencing revealed that a substantial fraction of all PTDH3 alleles constructed ( ~25% of sequenced strains ) carried an indel of one nucleotide in one of the homopolymer runs present in the promoter ( Source data 1 ) . These mutations probably result from polymerase slippage during PCR amplification . For some PTDH3 alleles , we were able to isolate independent clones that differed only by the presence or absence of these homopolymer mutations , giving us the opportunity to test the impact of homopolymer length variation on transcriptional activity . Using the fluorescence assay described below , we found that del431A , del54T and ins432A had no detectable effect on median expression level or expression noise ( Figure 2—figure supplement 2C , D ) . Therefore , strains carrying these mutations were included in the expression and fitness assays . The strains described above all carried the ho::PTDH3-YFP-TCYC1 reporter construct , allowing sensitive quantification of the transcriptional activity of different PTDH3 alleles . In these strains , the endogenous promoter driving expression of the native TDH3 protein was left unaltered . To measure how variation in TDH3 protein levels induced by mutations in the TDH3 promoter could impact cell growth , we inserted the final set of 43 PTDH3 alleles described above upstream of the endogenous TDH3 coding sequence . PTDH3 variants were integrated in the genetic background of strain YPW1001 , which is almost identical to the reference strain YPW1002 used for the expression assays , except that the mating type of YPW1001 is MATa and it carries a PTDH3-YFP-TCYC1-NatMX4 construct at HO conferring resistance to Nourseothricin . The reporter construct served a dual purpose: it ensured that the strains used in the expression and fitness assays carried the same number of copies of TDH3 promoter in their genomes and it allowed high-throughput counting of yellow-fluorescent cells carrying PTDH3 variants in the competition experiments described below . Importantly , we did not detect any difference in fluorescence levels between strains YPW1002 and YPW1001 ( Figure 2—figure supplement 2A , B ) , indicating that the few genetic differences between the background of the strains used in the expression and fitness assays did not significantly affect the activity of the TDH3 promoter . To insert the 35 alleles containing a single copy of PTDH3 at the native TDH3 locus , we first replaced the endogenous TDH3 promoter of strain YPW1001 with a CORE-UK cassette ( URA3-KanMX4 ) amplified with oligonucleotides 1909 and 1910 ( Supplementary file 2 ) to create strain YPW1121 . Then , the 35 PTDH3 alleles were amplified from the HO locus in the strains previously constructed ( Source data 1 ) using oligonucleotides 2425 and 1305 ( Supplementary file 2 ) . PCR products were purified using a DNA Clean and Concentrator kit ( Zymo Research ) , amplified using primers 1914 and 1900 ( Supplementary file 2 ) to attach appropriate homology tails and transformed in strain YPW1121 . In addition , because all the PTDH3 variants inserted at HO carried the PCR-induced mutation A293g , we created the strain YPW1189 that carried mutation A293g in the endogenous TDH3 promoter . YPW1189 served as the reference strain when calculating relative fitness . In all these strains , the presence of the correct mutations in PTDH3 at the native locus was confirmed by Sanger sequencing of PCR products obtained with oligonucleotides 1345 and 1342 ( Supplementary file 2 ) . To measure the impact on fitness of overexpression of the native TDH3 protein , we created seven tandem duplications of the whole TDH3 locus ( TDH3::PTDH3-TDH3-URA3- PTDH3-TDH3 ) that contained the same combinations of promoter alleles as those inserted at HO ( Source data 1 ) . Duplications of TDH3 were built in a similar way as the double-copy constructs inserted at HO . First , URA3 was amplified from the pCORE-UH plasmid using oligonucleotides 2688 and 2686 and the TDH3 variants corresponding to the copy located on the right in the final constructs ( ‘CONSTRUCT . 2’ in Source data 1 ) were amplified using oligonucleotides 2687 and 1893 ( Supplementary file 2 ) . URA3 and PTDH3-TDH3 PCR products were then fused by overlap extension and the resulting fragments were amplified with oligonucleotides 2696 and 2693 ( Supplementary file 2 ) . Finally , each of the seven different URA3-PTDH3-TDH3 products was transformed in the strain carrying the desired allele for the left PTDH3-TDH3 copy ( ‘CONSTRUCT . 1’ in Source data 1 ) . To control for the impact of URA3 expression on fitness , we constructed strain YPW2682 ( TDH3::PTDH3-TDH3-URA3 ) by transforming a URA3 cassette amplified from plasmid pCORE-UH with oligonucleotides 2696 and 2697 in strain YPW1189 . YPW2675 was used as the reference when reporting the relative fitness of the seven strains carrying two copies of TDH3 . To sequence the full TDH3 duplications ( 5 . 5 kb ) , we performed four overlapping PCRs using oligonucleotides 1345 and 1499 , 2694 and 1911 , 2670 and 1342 , 601 and 2695 and sequenced them with oligonucleotides 1345 , 1499 , 601 , 2691 , 2053 , 2670 , 1342 , 601 , 2695 ( Supplementary file 2 ) . As described for the double-copy constructs at HO , we used quantitative pyrosequencing to determine the exact number of TDH3 copies inserted in the seven strains . However , we could not directly quantify the frequency of mutation A293g in these strains , because all copies of TDH3 promoters present in their genomes carry the G mutation . Therefore , we first crossed all seven strains to YPW1139 ( Metzger et al . , 2016 ) , a strain that contains the A allele at position −293 of the native TDH3 promoter . In the resulting diploids , the frequency of G allele at the native TDH3 locus is expected to be 0 . 5 if the original haploid strain carried a single copy of TDH3 at the native locus , 2/3 if it carried two copies of TDH3 at the native locus and 3/4 if it carried three copies . To determine allele frequency at position −293 of PTDH3 for the native TDH3 locus only , we amplified the promoter using primers 2268 and 3095 specific to the native locus ( Supplementary file 2 ) and then used pyrosequencing as described above . We found that one strain carried three copies of TDH3 at the native locus instead of two ( Figure 4—figure supplement 1—source data 1 ) . However , we did not exclude the corresponding variant ( Var42 ) from subsequent analyses , because it also integrated three copies of the reporter construct at HO . Finally , during growth rate assays , cells carrying a tandem duplication of TDH3 could potentially lose a copy of TDH3 through intrachromosomal homologous recombination , which could affect fitness estimates . In strains carrying TDH3::TDH3-URA3-TDH3 constructs , the loss of a TDH3 copy by recombination should be accompanied by the deletion of the URA3 marker . To estimate how frequently such recombination events might occur , we quantified the frequency of Ura- cells in strain YPW2679 ( TDH3::TDH3-URA3-TDH3 ) at four time points over the course of 50 generations of growth in similar conditions as used in competition growth assays . Four replicate cultures of YPW2679 were grown to saturation in SC - Ura medium at 30°C . Then , 0 . 1 ml of each culture was plated on SC +5 FOA medium and each culture was diluted to a density of 104 cells/ml in YPD rich medium . Dilution to 104 cells/ml in YPD was repeated every 12 hr for 72 hr and plating on SC +5 FOA was repeated every 24 hr . After three days of incubation at 30°C , colony-forming units were counted on all SC +5 FOA plates , allowing the estimation of the frequency of Ura- cells every ~17 generations for a total of ~50 generations . The frequency of Ura- cells was found to increase during the first 34 generations of growth before reaching a plateau representing a state of mutation-selection balance . At this stage , the average frequency of Ura- cells was about 5 . 2 × 10−5 . Therefore , even if spontaneous loss of one TDH3 copy occurred in a fraction of cells , these events were found to be too rare to have a significant impact on fitness estimates . Data used to estimate the frequency of intrachromosomal recombination can be found in Supplementary file 1 – Dataset 6 . We deleted the native TDH3 locus in the genetic background of strain YPW1001 to create strain YPW1177 . To do this , we amplified a region of 171 bp immediately upstream of the TDH3 promoter using oligonucleotides 1345 and 1962 ( Supplementary file 2 ) . Oligonucleotide 1962 is composed of a 5’ sequence of 22 nucleotides priming directly upstream of the TDH3 promoter fused to a 3’ sequence of 38 nucleotides homologous to the 3’UTR sequence immediately downstream of TDH3 coding sequence . Therefore , transformation of the PCR product in strain YPW1121 ( tdh3::URA3-KanMX4-TDH3 ) led to the deletion of the URA3-KanMX4 cassette and of the TDH3 coding sequence . In this strain , both the TDH3 promoter and the TDH3 coding sequence are deleted , and the coding sequence of the upstream gene PDX1 is fused to the terminator sequence of TDH3 , so that PDX1 would remain functional . Correct deletion of TDH3 was confirmed by Sanger sequencing of the region amplified with oligonucleotides 1345 and 2444 ( Supplementary file 2 ) in strain YPW1177 . To measure how variation in TDH3 expression affected growth rate , the strains described above were all grown competitively against a common strain , YPW1160 , which carried a PTDH3-GFP-TCYC1-KanMX4 construct inserted at the HO locus in the same genetic background as the other strains . The expression of Green Fluorescent Protein in YPW1160 cells allowed for highly efficient discrimination from cells expressing the Yellow Fluorescent Protein using flow cytometry . To construct strain YPW1160 , the GFP-TCYC1 sequence was amplified from strain YPW3 ( swh1::PTDH3-GFP-TCYC1 , obtained from Barry Williams ) using oligonucleotides 601 and 2049 ( Supplementary file 2 ) . In parallel , KanMX4 was amplified from strain YPW1002 using oligonucleotides 2050 and 1890 ( Supplementary file 2 ) . The two fragments were fused by PCR SOEing and the product was amplified using oligonucleotides 601 and 1890 ( Supplementary file 2 ) before transformation in strain YPW1001 ( ho::PTDH3-YFP-TCYC1-NatMX4 ) . Selection on G418 allowed the recovery of cells that switched the YFP-TCYC1-NatMX4 cassette for the GFP-TCYC1-KanMX4 cassette . The fluorescence emission detected on the flow cytometer was consistent with expression of GFP . The relationship between the average expression level of TDH3 and fitness is not expected to follow a simple mathematical function . Therefore , we used LOESS regression to describe the relationship between median expression and fitness from the data collected with the set of 43 PTDH3 alleles , using the R function loess with a span of 2/3 . Next , we tested the impact of expression noise on fitness , which was complicated by the fact that expression noise is correlated with median expression and by the fact that median expression is expected to have a larger impact on fitness than expression noise . To disentangle the effects of median expression and noise on fitness , we first calculated the residuals ( ΔNoise ) from a LOESS regression ( span = 2/3 ) of expression noise on median expression . Next , we used a similar approach to calculate the residuals ( ΔFitness ) from a LOESS regression ( span = 2/3 ) of fitness on median expression . ΔFitness is the variation in fitness that cannot be explained by a difference in median expression in our dataset . To test whether ΔFitness could be at least partially explained by variation in expression noise , we calculated the Pearson’s correlation coefficient between ΔNoise and ΔFitness and used the R function cor . test to test for significance of this correlation . We excluded the two strains that showed a median expression level above 125% , because the number of samples with high expression was too low for meaningful interpretation of ΔNoise and ΔFitness in this range of expression levels . In addition , we compared the correlations between ΔNoise and ΔFitness for two different classes of promoter variants determined based on their expression levels . First , we determined the maximum fitness from the LOESS regression of fitness on median expression . Next , we estimated the median expression value that would lead to a 0 . 005 reduction in fitness relative to the maximum . This expression value was used as a threshold to determine which strains had an expression close to the optimum or far from it . Three quantitative parameters were determined arbitrarily in these analyses: the span of the two LOESS regressions and the reduction in fitness used to determine the expression threshold . To test the robustness of the results to variation in these parameters , we calculated the correlations between ΔNoise and ΔFitness for 100 combinations of parameters where the span of the LOESS regressions took one of five values ( 2/6 , 3/6 , 4/6 , 5/6 and 1 ) and the reduction in fitness took one of four values ( 0 . 0025 , 0 . 005 , 0 . 0075 and 0 . 01 ) . In addition , we used permutation tests to compare median expression , ΔNoise values and ΔFitness values between two groups of genotypes: those with ΔNoise values below −1 ( low noise ) and those with ΔNoise values above 1 ( high noise ) . For each parameter considered ( median expression , ΔNoise and ΔFitness ) , the observed values were randomly shuffled between the two groups 100 , 000 times . P-values were then calculated as the proportion of shuffled groups for which the absolute difference of median was greater than the observed difference of median between the groups before shuffling . All analyses were repeated for the four different metrics of noise mentioned above ( Noise strength , SD , CV* and log ( CV* ) ) . One important assumption in our analyses of the relationship between TDH3 expression and fitness is that the median and noise of expression measured using the fluorescent reporter constructs inserted at HO are representative of the expression level of the TDH3 protein when the promoter variants are introduced at the native TDH3 locus . To test whether the effects of mutations in the TDH3 promoter were the same when introduced at HO or at the native TDH3 locus , we constructed a TDH3-YFP fusion gene at the TDH3 locus and then introduced 20 different PTDH3 alleles upstream of this reporter gene , including eight TFBS and four TATA box variants that were present in the competition assays ( Figure 2—figure supplements 1—source data 1 ) . To fuse the coding sequences of TDH3 and YFP , we amplified the YFP-TCYC1-KanMX4 construct from strain YPW1002 using primers 3415 and 3416 and transformed the PCR product in the non-fluorescent strain YPW978 . Primer 3415 was designed to remove the stop codon of TDH3 and the start codon of YFP and to insert a 30 bp linker between the coding sequences of the two genes ( Huh et al . , 2003 ) . Then , the TDH3 promoter was replaced with a CORE-UH cassette ( URA3-HphMX4 ) amplified with oligonucleotides 1909 and 1910 ( Supplementary file 2 ) to create strain YPW1618 . The 20 PTDH3 alleles were amplified from the native locus in the strains previously constructed ( Figure 2—figure supplements 1—source data 1 ) using oligonucleotides 1344 and 1342 ( Supplementary file 2 ) and transformed into YPW1618 to replace the CORE-UH cassette . The presence of the expected mutations was confirmed by sequencing PCR products obtained with primers 1345 and 1952 ( Supplementary file 2 ) . The fluorescence level of the strains expressing the fusion proteins was measured in parallel to the fluorescence of strains carrying the same PTDH3 alleles at the HO locus . Four replicate samples of each strain were analyzed by flow cytometry after growth in YPD medium as described above . The expression of the reporter gene at the HO locus was found to be a strong predictor of the expression of the gene fusion at the native TDH3 locus , both for median expression level ( Figure 2—figure supplement 1A , R2 = 0 . 99 ) and for expression noise ( Figure 2—figure supplement 1B , R2 = 0 . 76 ) . These expression data are available in Supplementary file 1 – Dataset 4 and Figure 2—figure supplements 1—source data 1 . Flow data ( FCS files ) used to compare the effects of the PTDH3 alleles when inserted at the HO locus and at the native TDH3 locus are available in the FlowRepository ( flowrepository . org ) under experiment ID FR-FCM-ZYJX . In addition , the impact of fusing YFP to TDH3 on fitness was quantified by comparing the competitive growth rate of strain YPW1002 expressing YFP from the HO locus to the growth rate of strain YPW1964 expressing the TDH3-YFP protein fusion . The expression of the fusion protein was found to cause a 2 . 5% reduction in fitness ( Figure 2—figure supplement 1C ) , which could be caused by altered function and/or stability of the TDH3 protein when fused with YFP . For this reason , we decided not to use protein fusions to measure the fitness associated with different levels of TDH3 expression . To understand how cell-to-cell variability in gene expression level could contribute to population fitness , we performed individual-based stochastic simulations of the growth of clonal populations of cells covering a wide range of mean expression and expression noise values of a single gene . All simulations were run as short experiments of fixed duration ( 1 , 000 minutes ) where variability in expression level impacting single cell division rate was the only determinant of population growth rate . The behavior of the population was determined by: a ) a normal distribution 𝒩E of expression levels for the focal genotype described by its mean μE and variance σE2 , and b ) a function DT=fE relating single cell expression level E to the time in minutes separating two consecutive cell divisions , or doubling time DT . Single cell expression levels sampled from the expression distribution defined the doubling time for a given cell . Two different functions relating expression level to DT were explored: 1 ) a linear function ( DT=-40 × E+160 ) and 2 ) an inverted Gaussian function ( DT=-160 ×exp-E-12/0 . 18+240 ) . In each run of the simulation , a population of cells was tracked by recording information on the current expression level of each cell , the current DT derived from that expression level , and the amount of time remaining before the end of the experiment . For simplicity , the expression level of each mother and daughter cell was drawn from the normal distribution 𝒩E at each cell division and this expression level directly determined the DT value for the cell . To seed a starting population , 103 cells were sampled from the expression distribution and their expression level was transformed into DT . To desynchronize the founding population , the initial values of DT were scaled by a random value between 0 and 1 to randomize the time to first division and a complete simulation was run . 103 cells were drawn randomly at the end of the seed experiment and used to found a population for which growth rate was quantified . In the body of the simulation , each single cell was evaluated to determine if the current DT was greater than the remaining time in the experiment assessed for that cell , and if so , the cell divided , at which point new expression levels were drawn randomly from the normal distribution 𝒩E and independently for the mother and daughter cells . After cell division , the time remaining in the experiment for both mother and daughter cells decreased by the amount of the last DT , the new expression levels were translated into new values of DT , and the process repeated until DT values for all cells were greater than the remaining time in the experiment . Competitive fitness was calculated from the ratio of total number of cells Ni at the end of the experiment and the total number of cells Nref obtained from simulating the growth of a reference genotype with mean expression μref=1 and noise νref=0 . 1 , as follows:Fitness= explnNiNrefT Mean expression μE of experimental genotypes were explored in the interval [0 , 2] . Noise values νE were explored in the interval [0 , 3] where noise was specified separately as standard deviation , coefficient of variation , and Fano factor . Experiment duration T was set at 1000 minutes for ease of computation . 100 replicates of each stochastic simulation were run to estimate 95% confidence intervals on fitness estimates . Simulations were coded in MATLAB R2015 ( Supplementary file 5 ) . We performed time-lapse imaging of cells grown in microfluidic devices to compare the distributions of doubling time among four strains chosen for their differences in TDH3 mean expression level and expression noise ( strains YPW2879 , YPW2868 , YPW3064 and YPW3047 in Figure 4—source data 1 ) . The four strains were assayed on four consecutive days using the same procedure . First , cells were grown to saturation in liquid YPD medium at 30°C for ~16 hr . Then , 100 μl of culture was transferred in 5 ml of fresh YPD and grown for another 4 hr at 30°C until it reached an optical density at 660 nm comprised between 0 . 2 and 0 . 3 ( ~3×106 cells/ml ) . At this point , ~100 μl of cell culture was injected in a microfluidic chip using a 1 ml syringe . Microfluidic devices consisted of a PDMS ( Polydymethylsiloxane ) chip mounted on a 24 mm x 60 mm coverslip , as described in Llamosi et al . , 2016 . Each device contained five imaging chambers of 200 × 200×3 . 7 μm where a monolayer of cells could be grown . These chambers were connected on two sides to wide flow channels of 100 μm height where YPD medium was allowed to flow at 120 ul/min using an Ismatec IPC tubing pump . Cells were imaged using an inverted microscope ( Olympus IX83 ) equipped with a CoolLED pE-300 illumination system , a Zyla sCMOS camera ( Andor ) and an IX3-ZDC2 system for autofocus . The temperature of the entire microfluidic system was maintained at 30°C in a Plexiglass chamber covering the microscope ( Life Imaging Services ) . After 60 min of acclimation to growth in the microfluidic device , one bright field image and one fluorescence image were recorded once every six minutes for twenty hours at five positions centered on each of the five imaging chambers using a 60x oil immersion objective ( Olympus PlanApo N 60x ) . Only images obtained during the first eight hours ( 80 frames ) were analyzed , because tracking was not reliable after this time because of high cell densities . Fluorescence images captured expression from the PTDH3-YFP reporter gene in each strain with a wild-type TDH3 promoter that was used for cell counting in the fitness assays . We were unable to reliably track individual cells and to correctly assign buds to their mother cells with a software for automated image analysis ( ilastik v1 . 3 . 0 ) using this cytoplasmic YFP expression , thus we measured doubling times by analyzing the bright field images manually with Fiji ( Schindelin et al . , 2012 ) . Raw bright field and fluorescence images , as well as bright field images where cell division events were annotated , are available on Zenodo ( https://zenodo . org ) with DOI 10 . 5281/zenodo . 1327545 . For each movie , we randomly selected eight cells on the first frame and determined the doubling times of all cells produced by these eight starting cells . The doubling time of a cell was defined as the time separating the appearance of two consecutive buds ( Figure 5—figure supplement 1 ) . Following this procedure , we quantified the doubling time of at least 362 cells for each of the four genotypes . We then compared the mean doubling time and the standard deviation of doubling time for pairs of genotypes using permutation tests in R: doubling time values of the two genotypes were pooled and resampled 105 times without replacement in two groups of same size as the number of cells analyzed for the two genotypes . P-values were calculated as the proportion of permutations for which the absolute difference of mean doubling time ( or standard deviation ) between the two groups was greater than the observed absolute difference between the two genotypes . Custom R scripts containing the code used to process and analyze data as described above are provided as Supplementary file 3 . Input files necessary to run the R scripts are available as . zip files in Supplementary file 4 . Matlab code used to model population growth is provided as Supplementary file 5 . | Single-celled organisms that reproduce by dividing , like yeast , can create whole populations of genetically identical cells . However , some differences will exist among such cells , even when they have all experienced the same environment . These differences are known as “noise” . By definition , noise is not caused by differences in DNA sequence , but some DNA sequences are noisier than others ( i . e . they cause more differences among cells ) . Because the amount of noise can be under genetic control , noise could evolve due to natural selection . Scientists often study noise at the level of gene expression – in other words , how many RNA or protein molecules are produced from each gene within each cell . Prior work has suggested that this type of noise can affect how often individual cells divide in a population , which is a component of that population’s fitness . Yet directly measuring these effects has proven challenging . Different studies have in the past reached opposite conclusions about whether a change in gene expression noise would increase or decrease fitness . One major reason for the lack of clear results is that most mutations that alter gene expression noise also alter the average level of expression of that gene . To find DNA sequences that produced the same average amount of protein but different levels of expression noise , Duveau et al . compared the effects of hundreds of mutations in the DNA sequence regulating the expression of a gene in baker’s yeast . Experiments focused on 43 DNA sequences then showed that increased expression noise could either speed up or slow down the growth of the population by affecting how long it took each cell to divide . More specifically , the effects of increasing expression noise depended on the average amount of protein produced among the cells in the population . If the average expression level was close to the optimum amount at which cells divided as fast as possible , increasing expression noise reduced the growth of the whole population . If , however , the average protein level caused cells to divide slower than their maximum rate , increasing expression noise resulted in faster growth of the population as a whole . Duveau et al . explain their results as follows: more expression noise in a population that is already making the optimal amount of protein can reduce fitness because it increases the fraction of that population making a suboptimal amount of the protein . However , when the average expression level is not optimal , more expression noise would mean more cells producing an amount of protein that is closer to the optimum and thus having higher fitness . These findings provide conceptual tools needed to understand how genetic variation affecting expression noise evolves . They could also help understand how expression noise might contribute to biological processes that depend upon cell division , such as diseases like cancer . | [
"Abstract",
"Introduction",
"Results",
"and",
"discussion",
"Materials",
"and",
"methods"
] | [
"evolutionary",
"biology"
] | 2018 | Fitness effects of altering gene expression noise in Saccharomyces cerevisiae |
Suction is widely used by animals for strong controllable underwater adhesion but is less well understood than adhesion of terrestrial climbing animals . Here we investigate the attachment of aquatic insect larvae ( Blephariceridae ) , which cling to rocks in torrential streams using the only known muscle-actuated suction organs in insects . We measured their attachment forces on well-defined rough substrates and found that their adhesion was less reduced by micro-roughness than that of terrestrial climbing insects . In vivo visualisation of the suction organs in contact with microstructured substrates revealed that they can mould around large asperities to form a seal . We have shown that the ventral surface of the suction disc is covered by dense arrays of microtrichia , which are stiff spine-like cuticular structures that only make tip contact . Our results demonstrate the impressive performance and versatility of blepharicerid suction organs and highlight their potential as a study system to explore biological suction mechanisms .
Of the approximately one million known species of insects , only 325 attach using muscle-controlled suction organs ( Stork , 2018; Roskov et al . , 2020 ) . These species belong to a single dipteran family , the Blephariceridae , and their larvae and pupae develop on rocks in torrential alpine streams where flow rates can exceed 3 ms–1 ( Frutiger and Buergisser , 2002; Zwick , 2004; Figure 1a & b; Video 1 ) . Each blepharicerid larva has six ventral suction organs to attach to biofilm-covered rock surfaces , where it feeds on diatoms . Using its suction organs , the larva can locomote relatively quickly and possibly over long distances: blepharicerid larvae migrate from one stone to another to find the swiftest regions of the stream ( Frutiger , 1998; Mannheims , 1935 ) . Once development is complete , the winged adult emerges from its pupa , floats to the water surface , and immediately flies away to mate and lay eggs to begin the cycle anew ( Oosterbroek and Courtney , 1995; Craig , 1966 ) . The remarkable morphology of blepharicerid suction organs is well described ( Mannheims , 1935; Rietschel , 1961; Kang et al . , 2019; Komárek and Wimmer , 1922 ) . The organ superficially resembles a synthetic piston pump , with a suction disc that interacts with the surface and creates a seal , a central piston , and powerful piston muscles to manipulate the pressure , and a suction chamber with a thick cuticular wall to withstand low pressures during attachments . There are spine-like microstructures called microtrichia on the suction disc that contact glass surfaces and may increase resistance to shear forces . In addition , we have shown that a dedicated active detachment system allows the larva to rapidly detach its suction organ during locomotion ( Kang et al . , 2019 ) . While much is known about their morphology , the mechanisms involved in blepharicerid suction attachment are less well understood . Two studies to date have measured the attachment performance of blepharicerid larvae ( Frutiger , 2002; Liu et al . , 2020 ) , yet neither of them offers mechanistic insights into how their suction organs cope with different surface conditions to generate strong underwater attachments . Suction is one of the main strategies for strong and controllable underwater adhesion . Biological suction organs can adhere with high strength , rapid controllability , and reusability on smooth , rough , and biofilm-covered surfaces ( Ditsche and Summers , 2014b ) . This is in remarkable contrast to artificial suction devices widely used in technical applications , which allow only slow control and are limited to clean and smooth surfaces . Despite their potential for bio-inspiration , there are only a few well-studied animals ( namely , remora fish , clingfish , octopus , and leeches ) , for which the function of specific structures in biological suction attachments has been experimentally demonstrated ( Beckert et al . , 2015; Fulcher and Motta , 2006; Arita , 1967; Wainwright et al . , 2013; Kampowski et al . , 2016; Kampowski et al . , 2020; Ditsche et al . , 2014c; Kier and Smith , 1990; Kier and Smith , 2002; Smith , 1996 ) . To date , mechanistic studies on biological adhesion have focused primarily on terrestrial climbing animals such as geckos , tree frogs , insects , and spiders ( Lengerer and Ladurner , 2018; Federle and Labonte , 2019 ) , and have greatly expanded our knowledge on how to achieve and control adhesion in air . Likewise , mechanistic studies on biological suction are needed to identify new strategies for generating and controlling underwater adhesion in different surface conditions . Here we have investigated the mechanisms underlying suction attachments of blepharicerid larvae . We first conducted a detailed morphological study of Hapalothrix lugubris ( Blephariceridae ) to provide new insights into structures that are relevant for suction attachments . To understand how well blepharicerid larvae attach to different surfaces , we quantified their performance on smooth , micro-rough , and coarse-rough surfaces . We compared blepharicerid suction performance with that of a model terrestrial insect to investigate how two fundamentally distinct adhesive systems cope with surface roughness . Finally , we examined the function of spine-like microtrichia through in vivo visualisation of the contact zone during attachments on smooth and microstructured substrates .
H . lugubris larvae have six ventromedian suction organs , with each organ comprising a suction disc , a central opening and a piston , a suction chamber surrounded by a thick-walled cuticular cuff , and a V-notch ( Figure 1c-f and Video 2 ) . The suction disc contacts the surface for attachment , and the piston and underlying piston muscles ( Figure 1d & e ) actively lower the pressure inside the suction chamber . Two apodemes attaching to the V-notch in H . lugubris mediate its muscle-controlled opening for rapid detachment of the suction organ ( Figure 1f; see also Kang et al . , 2019 ) . The ventral disc surface of H . lugubris is covered in a dense array of microtrichia ( Figure 2a-e ) . The suction disc-sealing rim , which seals the disc for suction attachment , closely resembles that of Liponeura cinerascens ( Kang et al . , 2019 ) and comprises a dense array of upright rim microtrichia ( Figure 2b ) . This is different from Liponeura cordata , which has a distinct rim made up of a single row of horizontally flat rim microtrichia ( Kang et al . , 2019 ) . Going from the rim to the centre of the disc , the short rim microtrichia transition into longer spine-like microtrichia ( 6 . 7 ± 0 . 5 µm in length and 0 . 56 ± 0 . 01 µm in mid-length diameter; mean of means ± standard error of the mean; measured from scanning electron microscopy ( SEM ) images of n = 2 individuals ) , and then again to shorter microtrichia in the centre . The following imaging techniques were used to gain insights into the ultrastructure and internal organisation of the suction disc: freeze-fracture SEM , 3D models using computed microtomography ( micro-CT ) data , and in vivo transmitted light microscopy ( Figure 2 ) . While internal fan-fibre networks underneath the outer regions of the suction disc have been mentioned previously ( Rietschel , 1961 ) , we discovered that each internal fibre leads to a single microtrichium ( Figure 2b-d ) . Moreover , all the microtrichia that were fractured during sample preparation appeared to be solid ( in-filled ) cuticular structures ( Figure 2e ) . The small internal fibres leading into the microtrichia branch out from thicker trunks originating from the ventral side of the outer radial beams ( Figure 2b-e ) . Another notable ultrastructural feature is the radial beams , which are solid cuticular structures that alternate between a wide and a narrow beam ( Figure 2c ) . The beams originate from the palisades , a radial zone consisting of dorsoventral cuticular rods ( Figure 2f & g ) . There are 72 radial beams in a 90° segment of the disc , corresponding to 288 beams per disc ( assuming no interruption from the V-notch ) and a centre-to-centre spacing of around 4 µm or 1 . 3° .
Blepharicerid larvae possess some of the most powerful ( in terms of body weight ) and complex suction organs among animals . The three species of blepharicerid larvae studied here ( H . lugubris , L . cordata , and L . cinerascens ) produced extreme shear forces on smooth surfaces with averages that ranged from 320 to 1120 times their own body weight . In terms of weight-specific attachment performance , the larvae performed better than all terrestrial insects measured using comparable methods ( ie , whole-animal detachment experiments ) ( Federle et al . , 2000; Grohmann et al . , 2014 ) . For example , the weight-specific shear attachment of blepharicerid larvae on smooth surfaces was 3–11 times greater than that of stick insects measured in this study . To achieve this extreme shear attachment , blepharicerid suction organs must come in close contact by generating an effective seal . Based on our in vivo visualisations of H . lugubris attaching to smooth glass underwater , the microtrichia make close contact with the surface , helping to both seal the organ and generate friction . This corroborates our previous findings on the suction disc contact behaviour with L . cinerascens and L . cordata ( Kang et al . , 2019 ) . Likewise , the soft adhesive pads of stick insects make close contact on smooth surfaces , and while the weight-specific attachment forces are not as high as in blepharicerid larvae , they can withstand forces close to 100 times their body weight . In contrast , the attachment of stick insects on micro-rough surfaces is significantly different to that of blepharicerid larvae: for stick insects , there was a 16-fold decrease in performance compared to smooth substrates , while for blepharicerid larvae , the decrease was only twofold . This difference in the impact of micro-roughness can be attributed to the two fundamentally different mechanisms of attachment: on micro-rough surfaces , neither the soft adhesive pads nor the tarsal claws of stick insects function properly ( Bullock and Federle , 2011 ) . This is in part due to the reduced effective contact area ( the adhesive pads cannot mould sufficiently to the asperities ) and also due to the reduced friction from tarsal claws ( the claws cannot interlock with the small asperities ) . On the other hand , blepharicerid suction organs are still able to seal on micro-rough surfaces and microtrichia can interact with the asperities , which likely explains why their performance was not as diminished as in the stick insects . In addition , blepharicerid suction organs may adhere better to micro-rough surfaces because during partial contact , the gaps between the detached regions and the substrate are filled with water , whereas detached regions of stick insect pads are filled with air . As water is effectively incompressible and approximately 50 times more viscous than air , the water-filled contact zone can provide a much stronger resistance against detachment even under conditions of partial contact as on micro-rough substrates . While blepharicerid larvae attached more strongly than stick insects on micro-rough surfaces , the opposite was found on coarse-rough surfaces: for stick insects , there was no difference in performance between coarse-rough and smooth surfaces , whereas blepharicerid larvae attachment decreased 11-fold . It is likely that both blepharicerid suction organs and stick insect adhesive pads are unable to cope with coarse surface roughness . The adhesive pads of both insects may be unable to fully mould to the large asperities , and the length of the microtrichia may be insufficient to reach the lower regions of the surface profile ( Figure 6 ) . Stick insects , however , have large pretarsal claws that can interlock with large asperities for strong attachment . Previous studies on dock beetles ( Gastrophysa viridula ) and stick insects ( C . morosus ) have reported that both beetle and stick insect attachments on coarse-rough surfaces decrease significantly when the claws are removed ( Bullock and Federle , 2011; Scholz et al . , 2010 ) . This means that , although stick insects and dock beetles use two distinct adhesive systems ( smooth versus hairy pads ) , the combination of the claws and the adhesive pads produces the same trend: both insects attach strongly to smooth and coarse-rough surfaces but poorly to micro-rough surfaces . In contrast , blepharicerid larvae do not have claw-like appendages and rely on suction organs for attachment . Consequently , these aquatic larvae do not follow the same trend as terrestrial climbing insects and perform the worst on coarse-rough substrates . A similar result was reported by Liu et al . , 2020 using Blepharicera sp . , where the larval attachment performance decreased with increasing surface roughness , although no quantitative information on attachment forces can be extracted from their study ( the study also used a centrifuge , but only reported the rotation speed but not the insects’ mass and position; it is also unclear whether the larvae were wetted prior to the tests ) . Although attachment forces per body weight help to assess the performance from a biological perspective , the attachment force per contact area , or the stress , is needed to make comparisons between animals that adhere to surfaces . In blepharicerid larvae , the shear stress on smooth substrates ranged from 39 to 117 kPa , and from 24 to 120 kPa for normal stress ( the range in values arises from the assumptions used to calculate the contact area and the number of organs remaining in contact immediately prior to detachment ) . These values are similar to those reported in the literature for suction attachments of other animals on smooth substrates: the remora fish can withstand 93 kPa in shear and 38 kPa in the normal direction ( Fulcher and Motta , 2006 ) ; octopus can resist normal stresses of up to 271 kPa , squids up to 830 kPa , and lumpsucker fish up to 102 kPa ( Smith , 1996; Davenport and Thorsteinsson , 1990 ) . Like the cephalopods , blepharicerid suction stress can surpass 101 kPa ( 1 atm at standard sea level and temperature ) , with one L . cordata reaching 228 kPa ( realistic estimate; 77 kPa if based on conservative assumptions , as outlined in ‘Results’ section ) . Although a pressure difference of 1 atm is considered the upper threshold for suction attachments in air , this is not the case if the contact zone is completely wet and bubble-free , as the strong cohesion of water allows suction stresses to exceed 1 atm ( Smith , 1991; Smith , 1996; Wang et al . , 2019; Wang et al . , 2020 ) . Even if they do not reach 1 atm , blepharicerid larvae can generate sufficient attachment force to resist the fast flow rates in their natural habitat . Recent work on cupped microstructures ( which resemble microscopic suction cups ) has revealed the mechanisms of failure in underwater suction attachments: ( 1 ) under sustained tensile stress , the rim slides inwards and the rim diameter contracts by ~30%; ( 2 ) immediately prior to detachment , sections of the rim buckle inwards , leading to adhesive failure ( Wang et al . , 2019 ) . Similar failure modes were also reported for macroscopic suction cups ( Ditsche and Summers , 2019 ) . Two structural features of the blepharicerid suction organs could represent adaptations to counter the aforementioned failure mechanisms seen in cupped microstructures . First , the internal radial beams can provide structural support to reduce inward sliding and buckling of the suction disc rim . Similar to how the flexible membrane of an umbrella is stiffened by radial spokes , these stiff cuticular radial beams can stabilise the suction disc when the organ is under high tensile stress . Bones within the clingfish suction organ may also prevent inward sliding ( Wang et al . , 2020 ) . While we have yet to visualise blepharicerid suction organs fail under extreme forces , their powerful attachments suggest that they possess mechanisms to counter suction cup failure . The second morphological feature that may reduce inward sliding of the rim is the numerous microtrichia in the contact zone of the blepharicerid suction disc . The microtrichia can interlock with surface asperities and minimise inward sliding . This has been reported for the remora fish , which have stiff posterior-facing structures called spinules within their suction pads that passively engage with asperities during high-drag conditions ( Beckert et al . , 2015; Fulcher and Motta , 2006 ) , as well as for the clingfish , where a hierarchical system of rods and filaments near the periphery of the suction organ may increase friction on rough surfaces ( Ditsche and Summers , 2019 ) . Similarly , blepharicerid microtrichia are naturally angled ( ~40° to 50° relative to the horizontal ) and point towards the centre of the suction disc ( Kang et al . , 2019 ) . Hence , inward sliding would passively lead to additional interlocking of the microtrichia tips with rough surfaces , thereby loading the microtrichia along their axis . To interlock effectively , structures like the remora spinules need to be stiff and strong ( Dai et al . , 2002; Wang et al . , 2017 ) . The blepharicerid microtrichia are indeed likely to be stiff structures since ( 1 ) they are made of solid cuticles ( a composite material of chitin fibres embedded in a matrix of proteins ) , and the dense , sclerotised cuticle can reach high elastic moduli [Parle et al . , 2017]; ( 2 ) we never observed any microtrichia in side contacts , even on micro-structured surfaces when the disc was pressed into contact . From the observation that microtrichia only make tip contact , and assuming that they are loaded equally , we can give a conservative lower estimate for the elastic modulus of the microtrichia cuticle by following Goss and Chaouki , 2016 and modelling the microtrichium as a cylindrical beam loaded at an angle ( Figure 7; see ‘Materials and methods’ section for details ) . Based on this model , the microtrichia cuticle must have a stiffness of at least 0 . 3–0 . 4 GPa , similar to the stiffness of wood and bone ( Vincent and Wegst , 2004 ) , to prevent any side contact ( for an angle to the disc surface of 40° to 50°; see Figure 7 ) . It is not unlikely that the microtrichia cuticle is even stiffer: sclerotised cuticles can have an elastic modulus of up to 20 GPa ( Vincent and Wegst , 2004; Wegst and Ashby , 2004; Sykes et al . , 2019 ) . This supports our idea that microtrichia can maintain tip contact during interactions with rough surfaces and could serve a similar function to stiff remora spinules . It is worth mentioning that , while stiff interlocking structures play an important role in increasing friction on rough surfaces , an effective seal is crucial for attachments on both smooth and rough surfaces . In remora fish and clingfish , a soft rim helps the organ to mould to the surface and form a seal ( Wang et al . , 2020 ) . In bio-inspired suction devices as well , it is the combination of stiff structures and a soft sealing rim that generates the strongest attachments to smooth and rough surfaces ( Ditsche and Summers , 2019; Wang et al . , 2017 ) . The detailed biomechanics of how the dense array of microtrichia produces a tight seal on rough surfaces and interlocks with small substrate asperities is beyond the scope of this study and remains to be explored in future work . We have demonstrated that blepharicerid suction organs can attach with extreme strength to both smooth and rough surfaces . Despite the potential strength of each attachment , the larvae are surprisingly mobile in their natural habitat ( Frutiger , 1998; Federle and Labonte , 2019 ) . Both attachment force and mobility are required for blepharicerid larvae to survive in their challenging habitats , which include raging alpine torrents and areas near the base of waterfalls . Currently , blepharicerid suction organs are the only examples of piston-driven suction organs in insects . While the circular setae of male diving beetles ( Dytiscidae ) are also considered to be suction organs , the two systems have markedly different morphologies: ( 1 ) the ventral surfaces of circular setae are comparatively smooth; ( 2 ) circular setae lack a muscle-driven central piston; ( 3 ) there are no known mechanisms for rapid detachment in dytiscid circular setae ( Chen et al . , 2014; Karlsson Green et al . , 2013; Nachtigall , 1974 ) . As male dytiscid beetles use their suction organs to attach to the smooth sections of the female’s pronotum and elytra , their attachment works best on smooth surfaces and their performance declines strongly on rough surfaces ( Karlsson Green et al . , 2013 ) . Researchers have suggested that male and female dytiscid beetles are engaged in an evolutionary arms race driven by sexual conflict , as female cuticular surfaces are modified to hinder male attachment with suction discs ( Karlsson Green et al . , 2013; Bergsten et al . , 2001 ) . Interestingly , it appears that male beetles did not evolve friction-enhancing structures on their circular setae to facilitate adhesion to rougher regions of the female cuticle . In contrast , blepharicerid suction discs are densely covered in microtrichia that likely enhance the grip on rough surfaces in high-drag conditions . This difference in morphology may be based on function or phylogenetic constraints: male diving beetles may not need to attach to rough elytra if their setae can generate sufficient attachment forces on smooth cuticle alone; alternatively , beetle setae may be more limited in the structures that can be developed from them , compared to the blepharicerid organ , which is highly complex and multicellular . A more comparable suction-based attachment system can be found in remora fish . Remora fish use suction pads—highly modified dorsal fin spines—to attach to sharks , whales , and manta rays ( Beckert et al . , 2015; Fulcher and Motta , 2006 ) . Recent studies on the functional morphology of remora suction pads have greatly expanded our understanding on the mechanisms underlying their impressive performance ( Beckert et al . , 2015; Wang et al . , 2017; Gamel et al . , 2019 ) . A remora suction pad comprises a soft fleshy outer rim and rows of lamellae topped with spinules . The pitch of the lamellae is muscle-controlled to facilitate spinule contact with rough surfaces , such as shark skin . When engaged , the tips of these stiff spinules interlock with surface asperities and increase friction , thereby increasing shear resistance . Moreover , the angled posterior-facing lamellae and spinules promote passive engagement when subjected to shear forces from a swimming host ( Beckert et al . , 2015; Fulcher and Motta , 2006 ) . The similarities between the remora suction pad and the blepharicerid suction organ may be based on overlapping functional requirements , as both animals have to cope with high shear forces ( fast-swimming hosts for the remora and torrential rivers for blepharicerid larvae ) . Since blepharicerid larvae attach to rocks underwater and feed on epilithic algae , their suction organs will in most cases contact the biofilm , yet the details of this interaction are unknown ( Frutiger and Buergisser , 2002 ) . As hypothesised previously , it is possible that the stiff microtrichia penetrate the biofilm layer ( Rietschel , 1961; Nachtigall , 1974 ) . This may allow the microtrichia to directly interlock with asperities on the rock surface or to generate additional friction from embedding numerous microtrichia into the biofilm . Mayfly larvae , which also inhabit fast-flowing watercourses , have friction-enhancing hairs that benefit from interacting with the biofilm ( Ditsche et al . , 2014a ) . It was found that a higher proportion of mayfly larvae can withstand fast flow rates on smooth hard ( epoxy ) substrates when the biofilm is present . Moreover , the setae and spine-like acanthae on the ventral surfaces of mayfly larvae can generate friction forces on clean rough substrates ( Ditsche-Kuru et al . , 2010 ) . Additional experiments with blepharicerid larvae are underway to investigate the interaction between stiff microtrichia and soft substrates . To conclude , we have shown that blepharicerid larvae use their suction organs to generate extreme attachment to diverse surfaces . The suction organ morphology is conserved between Hapalothrix and Liponeura larvae , and consists of a suction disc that contacts the substrate , dense arrays of microtrichia on the disc surface , muscles to control the piston , and the V-notch detachment system . We characterised the suction disc ultrastructure , which includes internal radial beam structures that could help to stabilise the suction disc when subjected to high stress , and fan-fibres that connect individual microtrichia to the radial beams . In terms of attachment performance , blepharicerid larvae withstand extreme shear forces equivalent to 320–1120 times their body weight on smooth substrates , depending on the species . H . lugubris performed the best overall , reaching shear forces equivalent to 1430 times the body weight , as well as an estimated shear stress of 117 kPa and normal stress of 24 kPa . Although their attachment decreased with increasing surface roughness , blepharicerid suction organs performed better than the smooth adhesive pads of stick insects on micro-rough surfaces . We confirmed that blepharicerid suction organs can mould to large surface asperities and that microtrichia come into close contact between the asperities . These microtrichia are stiff spine-like structures that are specialised for maintaining tip contact with the surface for interlocking with asperities . Our study provides new insights into the function of a highly adapted insect adhesive organ and expands our understanding of the function of biological suction organs .
L . cinerascens ( Loew , 1845 ) larvae were collected from fast-flowing alpine rivers near Meiringen , Switzerland ( GPS location 46° 44' 05 . 6" N , 8° 06' 55 . 4" E , in May 2018 ) , and close to Grinzens , Tirol , Austria ( 47° 12' 41 . 4" N , 11° 15' 28 . 1" E , in September 2018 ) . At the latter site , L . cordata ( Vimmer , 1916 ) and H . lugubris ( Loew , 1876 ) were also collected . For all the species , we collected third and fourth instar larvae that were large enough to be handled for experiments . Wearing fishing waders and diving gloves , we removed rocks from the most turbulent areas of the river and brought them to the riverbank for specimen collection . Although it was previously noted that the larvae can attach so firmly that they are torn upon detachment ( Komárek , 1914 ) , we found that a gentle nudge using soft-touch tweezers can elicit an evasive response from them , whereupon they could be easily picked up using tweezers and placed in specially prepared 50 ml Falcon tubes . All larvae were kept in these tubes in an ice box during collection and transport . Rocks were returned to their approximate locations after collection . For long-term maintenance of the larvae , an aquarium tank was set up with water and small rocks from the collection site . A filter unit with two outlets for a small water cascade was used to filter the water and to simulate the natural environment . Multiple air pumps were also placed close to the aquarium walls to provide ample oxygenation and additional regions with turbulent flow . To promote algal growth , an over-tank light-emitting diode ( LED ) light was set to a 12-hr day-night cycle . The aquarium was kept in a 4°C climate room ( mean temperature of 3 . 2°C ± 0 . 9°C; mean ± SD ) to replicate alpine stream temperatures . SEM was used to image fourth instar H . lugubris larvae as described previously ( Kang et al . , 2019 ) . In brief , samples fixed in 70% ethanol ( v/v ) were flash-frozen in liquid ethane cooled with liquid nitrogen and freeze-fractured immediately afterwards with a double-edged razor blade on a cooled aluminium block to obtain longitudinal views . Samples were freeze-dried overnight , then carefully mounted on SEM aluminium stubs using carbon tape and silver paint . They were then sputter-coated with 15 nm of iridium and imaged using a field-emission SEM ( FEI Verios 460 ) . One H . lugubris fourth instar larva was fixed in 2% paraformaldehyde and 2% glutaraldehyde ( v/v ) in 0 . 05 M sodium cacodylate buffer ( pH 7 . 4 ) for 7 days at 4°C . The larva was then dissected into six pieces—each containing one suction organ—and fixed for an additional day . The samples were then rinsed multiple times in 0 . 05 M sodium cacodylate buffer followed by deionised water before dehydration through a graded ethanol series: 50% , 75% , 95% , 100% ( v/v ) , and 100% dry ethanol . The dehydrated samples were critical-point dried using four flushes of liquid CO2 in a Quorum E3100 . One critical-point-dried suction organ was used for imaging via micro-CT . The sample was mounted on a standard dressmaker’s pin using ultraviolet ( UV ) -curable glue , then imaged using a lab-based Zeiss Xradia Versa 520 ( Carl Zeiss XRM , Pleasanton , CA , USA ) x-ray microscope . The sample was scanned at 0 . 325 µm/pixel with an accelerating X-ray tube voltage of 50 kV and a tube current of 90 µA . A total of 2401 projections collected at 20 s exposure intervals were used to perform reconstruction using a Zeiss commercial software package ( XMReconstructor , Carl Zeiss ) , utilising a cone-beam reconstruction algorithm based on filtered back-projection . Subsequent 3D volume rendering and segmentations were carried out using Dragonfly v4 . 0 ( Object Research Systems Inc , Montreal , Canada ) and Drishti v2 . 6 . 5 and v2 . 7 ( Limaye and Stock , 2012 ) . Insect attachment forces were measured using a custom centrifuge set up described previously ( Federle et al . , 2000 ) . The centrifuge operated on the following principle: a platform with the test substrates and the insect was driven by a brushless motor , and a light barrier sensor was triggered per rotation . This signal was used to synchronise image acquisition from a USB camera ( DMK 23UP1300; The Imaging Source Europe GmbH , Bremen , Germany ) , and image frames and their corresponding times were recorded using the StreamPix4 software ( NorPix Inc , Montreal , Canada ) . For safety reasons , the maximum centrifugation speed was limited to approximately 75 rotations per second ( rps ) . Some of the blepharicerid larvae could not be detached even at the maximum speed ( n = 14 out of 136 measurements ) ; in such cases , we used the maximum acceleration of a successfully detached individual from the given species . Effect of surface roughness on the peak shear force of blepharicerid larvae was measured on the following substrates: smooth ( clean polyester films ) , micro-rough ( polishing films with a nominal asperity size of 0 . 05 µm; Ultra Tec , CA , US ) , and coarse-rough ( 30 µm polishing films; Ultra Tec ) . The same substrate types were used to measure the normal force ( substrates mounted vertically in the centrifuge ) , but a polished polymethyl methacrylate ( PMMA ) surface was used as the smooth substrate . Surface characteristics ( average roughness ( mean height deviation ) Ra , root-mean-squared roughness Rq , and maximum peak-to-valley height ( PV ) ) of the micro-rough substrates were obtained using white-light interferometry with a scan area of 0 . 14 × 0 . 10 mm ( Zygo NewView 200; Zygo Corporation , CT , USA ) . Micro-rough substrates were sputter-coated with 5 nm of iridium prior to scanning to improve the surface reflectivity . As the coarse-rough substrate could not be adequately imaged via white-light interferometry , we used a Z-stack image focal-depth analysis technique as described elsewhere ( Sarmiento-Ponce et al . , 2018 ) with a scan area of 0 . 44 × 0 . 58 mm . For both surfaces , three regions were selected at random and imaged . Interferometry images were analysed using MetroPro software ( Zygo ) , and a custom MATLAB script was used to reconstruct the surface profile from the Z-stack images ( The MathWorks Inc , MA , United States ) . Since L . cinerascens and L . cordata were difficult to maintain in laboratory conditions , these two species were tested only on smooth horizontal surfaces ( n = 43 and n = 10 , respectively ) . The full range of tests on smooth , micro-rough , and coarse-rough surfaces was conducted for H . lugubris ( n = 9–15 for shear tests; n = 10 for all tests in normal direction ) . Prior to the experiments , individuals were selected from the laboratory aquarium and placed inside specially prepared 50 ml Falcon tubes . This tube was kept on ice for the duration of the experiment . For each run , a larva was carefully removed from the tube and placed on the test surface . A droplet of water ( taken from the aquarium ) was used to wash excess debris from the insect , and lab tissue paper was used to wick away excess water without removing all moisture from the larva; the contact zone of the suction discs was still completely wetted under these conditions , as confirmed by IRM observations . These steps were necessary to prime the larvae for the centrifugation trials as they often displayed defensive behaviour while being handled . The larvae adhered and remained still once primed , and between two and four repetitions were performed for each larva . All centrifuge trials were conducted within 7 days of collection . After the trials , all the larvae were blot-dried on filter paper and weighed using an analytical balance ( 1712 MP8; Sartorius GmbH , Göttingen , Germany ) . Statistical analyses were conducted on log10-transformed values using R v3 . 6 . 2 run in RStudio v 1 . 2 . 5033 ( R Development Core Team , 2019; Team RS , 2019 ) . Normal ( adhesive ) stress and shear stress ( defined as the peak attachment force divided by the contact area ) were calculated using suction disc areas measured for L . cordata and H . lugubris larvae . Larvae were placed on microscope slides so that the suction organs fully contacted the glass and imaged with a stereomicroscope . Every tested L . cordata and H . lugubris specimen was imaged . A representative organ was selected from each L . cordata and H . lugubris larva , and the contact area calculated by fitting a circle inclusive of the outer fringe layer using FIJI ( Schindelin et al . , 2012 ) ( https://imagej . net/Fiji ) . The peak attachment force was then divided by this contact area to determine the peak stress for each larva . C . morosus ( Sinéty , 1901 ) stick insects were used as a model for terrestrial insect adhesion , and their attachment on surfaces with varying roughness was measured to compare against blepharicerid larval attachment . Second instar nymphs with undamaged legs and tarsi were selected for centrifuge experiments using smooth , micro-rough , and coarse-rough surfaces ( n = 10 per surface ) . No normal forces could be measured on the micro-rough surface as stick insects failed to hold their body weight during preliminary tests . Before each trial , we checked that the specimen was in contact with the surface using all six legs and that the surface was uncontaminated . Stick insects were oriented with the head facing out , and each individual was tested twice and weighed afterwards . The higher attachment force per individual was used as the peak attachment force . In order to examine blepharicerid larvae locomoting for extended periods of time , a custom flow chamber was built to imitate the fast-flow conditions of their natural environments ( Figure 8 ) . Two aluminium plates ( approximately 60 × 100 mm in height × width ) , each with a rectangular window , were used to sandwich an inner chamber made out of polydimethylsiloxane ( PDMS; Sylgard 184 , Dow Corning , MI , USA ) . This inner chamber had a lemon-shaped chamber to serve as the observation arena , and an inlet and an outlet for water circulation . Two microscope coverslips ( 0 . 16–0 . 19 mm thickness; Agar Scientific , Stansted , UK ) were used to encase the inner chamber . Two to five larvae were placed on the bottom coverslip , and once the top coverslip was placed over the arena , four clamps were used to squeeze the aluminium plates and coverslips against the PDMS . The soft PDMS moulded closely to the plates and created a water-tight seal . Aquarium water ( kept cool in an ice bath ) was pumped via a micro-pump ( M200S-V; TCS Micropumps Ltd , UK ) , and the input voltage was controlled by a microprocessor . The flow rate was controlled by setting an appropriate pump-operating voltage . With this flow chamber , we recorded H . lugubris larvae locomotion and the attachment/detachment of suction organs on smooth glass surfaces via IRM , which has been previously used to investigate the contact between animal adhesive organs and the substrate ( Federle et al . , 2006; Federle et al . , 2002 ) . Videos were recorded using a USB camera ( DMK 23UP1300 ) and the IC Capture software ( v2 . 4 . 642 . 2631; The Imaging Source GmbH ) at 30 frames per second ( FPS ) . To observe how suction organs respond to surface roughness , we used transparent micro-structured surfaces with well-defined micro-ridges and grooves fabricated by photolithography and nanoimprinting ( Zhou et al . , 2014 ) . In brief , a master surface was first produced using photolithography , and a PDMS mould of this master was used to cast the final surface out of epoxy . Three micro-ridge geometries were used in our experiments: ( 1 ) 3 × 3 × 2 µm ( ridge width × groove width × ridge height ) ; ( 2 ) 3 × 3 × 4 µm; and ( 3 ) 10 × 10 × 2 µm . As the ridge height is only approximately controlled through the spin-coating of photoresist when producing the master , we measured it from the epoxy replicas using white-light interferometry as mentioned above ( see Table 1 ) . Four to five regions from each uncoated substrate were imaged and only regions without artefacts were used in the final calculation . For simplicity , when referring to the substrates , the depths of the grooves were reported to one significant figure ( ie , 3 × 3 × 2 µm , 3 × 3 × 4 µm , and 10 × 10 × 2 µm , for widths of ridges , grooves , and ridge height ) . Note that as these surfaces could not be used in combination with the flow chamber , a H . lugubris larva was placed on the substrate , wetted with a droplet of aquarium water , gently motivated with soft-touch forceps , and recorded as they moved around on the substrate . One H . lugubris specimen was used for both the 3 × 3 × 2 µm and 10 × 10 × 2 µm substrates , and a different larva was used for the 3 × 3 × 4 µm surface . Based on our observation that the microtrichia never showed any side contact , even when the suction discs were in contact with micro-structured substrates , we estimated the minimum elastic modulus of the microtrichia cuticle . We estimated the maximum force F on one microtrichium , perpendicular to the surface , as 56 nN . This was obtained based on the following assumptions: ( 1 ) the suction discs are loaded with a peak normal force of 11 . 6 mN , ie , 1 . 9 mN per sucker ( from centrifuge measurements of H . lugubris ) ; ( 2 ) equal loading of ca 34 , 000 spine-like microtrichia in tip contact ( based on the area of the suction disc bearing spine-like microtrichia of ca 49 , 000 µm2 , with the average microtrichia tip density of 0 . 7 per µm2 ) . The microtrichia were assumed to be cylindrical , with a length L=6 . 7±0 . 5μm and diameter D=0 . 56±0 . 01μm ( mean of means ± standard error of the mean; n = 2 H . lugubris ) , and the angle between the unloaded microtrichia and the surface was estimated as α=45±3° ( mean ± SD measured from five microtrichia of H . lugubris ) . The local adhesion and friction force of the microtrichium were assumed to be negligible . Following Goss and Chaouki , 2016 , the elastic modulus below which a cylindrical beam loaded at an angle α ( see ϕa definition ) would exhibit side contact is , E=1Kp2-Fϕap22FL2I where Kp2 and Fϕap2 are the complete and incomplete elliptic integrals of the first kind , respectively , p2=1/2 is the elliptic modulus , ϕa=sin-1sinα/2p is the elliptic amplitude , and I=D4π/64 is the second moment of area . See Figure 6d–ii for a schematic of a hypothetical scenario where microtrichia make side contacts on a smooth surface . | Suction cups are widely used to attach objects to surfaces in bathrooms and kitchens . They work well on tiles and other smooth surfaces , but do not stick well to rougher materials like brick or wood because they are unable to form an air-tight seal . Researchers have been searching for ways to improve these cups by studying how octopuses , remora fish and other sea animals use muscle-powered suction organs to stick to wet and rough surfaces . However , the experiments needed to understand the detailed mechanics of suction organs are difficult to perform on living specimens of these animals . The aquatic larvae of a family of insects known as the net-winged midges also have suction organs that are powered by muscles . These insects survive in fast flowing mountain streams where they use their suction organs to stick to rocks underwater . However , it remained unclear how these suction organs work . Here , Kang et al . found that net-winged midge larvae attach extremely well to a variety of surfaces . The larvae were able to withstand forces over one thousand times their body weight when attached to smooth surfaces . Even on rough materials , where human-made suction cups attach poorly , the larvae were able to withstand forces up to 240-times their body weight . Further experiments using several microscopy approaches revealed that the suction organs of the larvae are covered in multiple spine-like structures called microtrichia that interlock with bumps and dips on a surface to help the organ remain in place . Similar structures have previously been found on the suction organs of remora fish , but are not as tightly packed together . These findings demonstrate that net-winged midge larvae may be useful model systems to study how natural suction organs operate . Furthermore , they provide a new source of inspiration for scientists and engineers to design and manufacture suction cups capable of attaching to a wider variety of surfaces . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"ecology",
"physics",
"of",
"living",
"systems"
] | 2021 | Extreme suction attachment performance from specialised insects living in mountain streams (Diptera: Blephariceridae) |
The extensive use of mollusc shell as a versatile raw material is testament to its importance in prehistoric times . The consistent choice of certain species for different purposes , including the making of ornaments , is a direct representation of how humans viewed and exploited their environment . The necessary taxonomic information , however , is often impossible to obtain from objects that are small , heavily worked or degraded . Here we propose a novel biogeochemical approach to track the biological origin of prehistoric mollusc shell . We conducted an in-depth study of archaeological ornaments using microstructural , geochemical and biomolecular analyses , including ‘palaeoshellomics’ , the first application of palaeoproteomics to mollusc shells ( and indeed to any invertebrate calcified tissue ) . We reveal the consistent use of locally-sourced freshwater mother-of-pearl for the standardized manufacture of ‘double-buttons’ . This craft is found throughout Europe between 4200–3800 BCE , highlighting the ornament-makers’ profound knowledge of the biogeosphere and the existence of cross-cultural traditions .
Two modern marine mollusc shells , O . edulis and M . modiolus , were collected in northern Jutland ( Denmark ) by Søren H . Andersen and were selected for the following reasons: O . edulis shells had been suggested as the potential raw material for the Hornstaad-Hörnle IA assemblage ( Heumüller , 2010 ) and are very abundant at the shell midden site of Havnø; M . modiolus is a thick-shelled mussel with a nacreous layer , therefore a suitable raw material for the Havnø ornaments ( Appendix 1 , section 2 ) . Furthermore , close relatives of both species are present in public sequence databases , which is important for palaeoproteomics: O . edulis belongs to family Ostreidae ( genomes available for Crassostrea gigas and C . virginica ) and M . modiolus to family Mytilidae ( genome available for Mytilus galloprovincialis ) . With regard to the freshwater species ( order Unionoida ) , U . pictorum and U . crassus belong to family Unionidae , P . auricularius and M . margaritifera to family Margaritiferidae . Modern U . pictorum shells were collected in a stream close to Izeure ( Burgundy ) by Frédéric Marin and modern M . margaritifera was collected in northern Jutland by Søren H . Andersen . The morphological determination of both taxa was carried out by Frédéric Marin . U . crassus and P . auricularius are archaeological shell specimens from the sites of Peştera Ungurească and Isorella ( Neolithic , Po Plain , Italy [Starnini et al . , 2018] ) . The determination of U . cf . crassus had been carried out by Alberto Girod on the basis of morphological observations of the whole shell valves and comparison to extant specimens from the area ( Girod , 2010a ) . Archaeological P . auricularius was used as this species is critically endangered ( Altaba , 1990 ) and extant populations rare ( Appendix 1—figure 1 ) . The morphological determination of the species had been carried out by Alberto Girod using comparative specimens from museum collections ( Biddittu and Girod , 2003 ) . An advantage of including archaeological shells as reference materials is that we were able to assess the extent of molecular preservation in shells that are contemporary to the double-buttons . Furthermore , whole and fragmented valves of U . cf . crassus had been recovered from all the archaeological layers that yielded the double-buttons at Peştera Ungurească ( Girod , 2010a ) . Therefore , in this case , potential raw material and finished product have experienced the same post-depositional conditions . None of the Unionoida species is well-represented in public sequence databases , especially with regard to proteins related to shell biomineralization ( see Methods section ) .
The double-buttons described here are circular , with a groove in the middle of the body and no perforation . The main body is shiny but all of them have a thin matte layer on one of the two surfaces ( Figures 1 and 3 ) . These two layers , both aragonitic ( as shown by infrared spectroscopy , see Appendix 1 , section 3 . 2 ) , are the nacre and prisms of a mollusc shell . Scanning electron microscopy showed the presence of the ‘brickwall’ microstructure of nacre ( sheet nacre ) juxtaposed with the thin layer of prisms , the latter having elongation axes perpendicular to the nacre plane ( Figure 3 ) . No secondary calcite was observed and there was no sign of the occurrence of diagenetic recrystallization . The overall nacre appearance is typical of bivalves and not of gastropods . The combination of nacroprismatic microstructure and aragonitic mineralogy is observed in freshwater unionoid mussels but also in marine trigonioids and anomalodesmatans ( Appendix 1 , section 2 ) . Stable isotope analyses for all of the samples yielded average δ18O and δ13C values of −5 . 3 ± 0 . 4 and −11 . 1 ± 0 . 6 ‰ for Havnø , −6 . 1 ± 1 . 0 and −11 . 9 ± 1 . 7 ‰ for Peştera Ungurească and −9 . 3 ± 0 . 5 and −10 . 6 ± 1 . 6 ‰ for Hornstaad , respectively ( Appendix 1 , section 3 . 3 ) . The consistently low δ18O and δ13C values of shells from Peştera Ungurească and Hornstaad indicate a local freshwater origin for the shells ( Keith et al . , 1964; Leng and Lewis , 2016 ) , whereas the δ18O values at Havnø suggest some mixing of marine water or changes in the atmospheric circulation , with precipitations slightly enriched in 18O compared to present day over the region . Our interpretations are broadly supported by the average annual δ18O values of modern local precipitations for the sites ( WaterIsotopes . org , 2018 ) . The isotope data therefore suggest that the shells were locally sourced ( Keith et al . , 1964 ) . The absence of recrystallization observed by SEM was consistent with the concentration , composition and relatively low D/L values for all the amino acids analysed ( e . g . alanine D/L ~0 . 1 for Peştera Ungueraşca , ~0 . 2 for Havnø and Hornstaad ) , except for sample HorA . This supports a non-fossil origin for the shells used to make the double-buttons , that is the makers used ‘fresh’ or recently dead mollusc shells . However , the extent of degradation for HorA was significantly higher and both D/L and concentration values showed a clear ‘burning’ signal ( Crisp , 2013; Demarchi et al . , 2011 ) . The amino acid composition was similar to that of freshwater bivalves ( Unio , Margaritifera ) present in the reference database of Demarchi et al . ( 2014 ) , although HorA and PesB appeared to be rather different from the other double-buttons ( Appendix 1 , section 3 . 4 ) . We characterized the proteomes preserved within the seven double-buttons and performed bioinformatic searches ( PEAKS 8 . 5 , Bioinformatics Solutions Inc [Ma et al . , 2003] ) of the product ion spectra against both a Protein sequences and an Expressed Sequence Tags ( ESTs ) database , restricting the taxonomy to Mollusca ( see Methods section ) . This resulted in the identification of 1973 and 3233 peptide sequences , respectively , which represent the 3 . 5% and 5 . 1% of the total number of sequences generated by the de novo algorithm of the software ( excluding contaminant sequences ) . For comparison purposes , we also performed shotgun proteomics on the shell matrices extracted from the six reference mollusc species ( Appendix 1 , section 3 . 5 ) and analysed the data using the same databases and parametres ( see Table 2—source data 1 for full results of the palaeoproteomic analyses ) . Table 2 shows the top-scoring proteins from the seven double-buttons: the numbers indicate the peptides supporting each protein identification , while protein coverage ( i . e . the percentage of sequence for which we could detect peptides ) is represented by different colours . Additionally , on the right hand side of the table we indicate if each of the double-button proteins also occurred in the reference shell proteomes ( the list of all shell proteins identified in both the double-buttons and the reference shells can be found in Table 2—source data 1 ) . The main protein sequences from the double-buttons were identified as belonging to the pearl-producing triangle sail mussel Hyriopsis cumingii ( Bai et al . , 2013 ) . Protein Hic74 ( GenBank: ARG42316 . 1 ) was found in all of the archaeological samples , except HorA . The percentage coverage for the Hic74 sequence was highest for the Havnø beads ( 35–55% ) , where it was supported by 132 , 153 , 255 unique peptides in HavA , HavB and HavC , respectively ( Appendix 1 , section 3 . 5 ) . This protein was also securely identified in all of the freshwater unionoid reference shells ( coverage varying from 34% in M . margaritifera , supported by 67 peptides , to 50% in U . crassus , supported by 203 peptides ) . Hic74 is an acidic , Ala- and Gly-rich shell matrix protein ( Liu et al . , 2017a ) . Consisting of 19 poly-A blocks , GA repeats , short acidic motifs ( that probably bind to the mineral ) and a GS-rich domain at the C-terminus ( which resembles that of lustrin-A ) , this silk fibroin-like protein is likely to play a structural role in nacre formation and in enhancing its mechanical properties ( Liu et al . , 2017a ) . Protein Hic52 ( GenBank: ARH52598 . 1 ) was identified in all the reference unionoid shells and in the Havnø samples , but only when less stringent parameters were used for the identification ( i . e . number of unique peptides ≥ 1 ( instead of 2 ) and protein score −10lgP ≥ 20 ( instead of 40 ) ) . Hic52 is a very basic ( theoretical pI > 10 ) , Gly- and Gln-rich protein , with few poly-Q and poly-G blocks and several degenerate G-rich repeats of different lengths along the sequence . It possesses a collagen-like structure which suggests a structural role in nacre formation ( Liu et al . , 2017b ) . Silkmapin ( GenBank: AIZ03589 . 1 , and its isoforms nasilin 1 and 2 ) are Gly-rich non-acidic proteins with a structural function , probably related to the formation of both nacreous and prismatic layers ( Liu et al . , 2015; Marie et al . , 2017 ) . Present in the shell matrix of all the unionoids , these proteins were also detected in the Havnø samples and in one of the Hornstaad beads ( HorB ) . Finally , we also identified protein sequences from marine mollusc genera ( mainly Pinctada , Crassostrea and Mytilus ) , but all these ‘marine’ sequences only displayed repeated low-complexity ( RLC ) domains ( typically consisting of Ala and Gly-rich repeats and/or poly-Ala blocks ) . RLC-containing peptides are not sufficient for distinguishing between freshwater and marine shells . On the contrary , in double-buttons and in unionoid reference shells , the top-scoring protein Hic74 was supported by remarkably high ( for shell proteins ) coverages , and , together with Hic52 and silkmapin/nasilin , showed a number of specific peptides that do not exhibit RLC domains . These proteins showed no homologues with any other shell proteins of marine origin currently present in the NCBI database ( BLASTp search ) , being unique to H . cumingii and suggesting their specificity to freshwater unionoid shells . We argue that their presence ( where identified as the major shell matrix proteins , supported not only by RLC domains ) is specific to Unionoida , freshwater mother-of-pearl shells , which in combination with the isotopic data , and supported by the microstructural and amino acid results , excludes a marine origin for the raw material used to make the double-buttons . We performed a search of the product ion spectra from the double-buttons and the six reference shells against the redundant EST database , so that we could recover complementary information from non-annotated sequences . For example , a search of Hyriopsis cumingii on NCBI will retrieve 246 protein sequences but 10156 EST sequences . The dataset was used to explore the similarities between double-buttons and molluscan shell proteomes , presented as a circular plot in Figure 4 . This output was derived from an adjacency matrix , showing which proteins ( EST sequence identifiers ) occurred in two or more samples ( R code and data files can be found in Figure 4—source data 1 ) . From this , the subset of unique identifiers , present both in the double-buttons and in any of the reference shells , was associated with its sequence coverage ( % ) in the archaeological samples . This information is represented in the right-hand side of Figure 4: the length of the circular segment for each molluscan species is proportional to the number of sequences that the shell shares with the archaeological samples , scored on the basis of the coverage . On the left-hand side of the graph , each double-button is represented by a circular segment , which is proportional to the number of unique peptides that supports each shared protein sequence . The similarity between double-buttons and mollusc taxa can be visualised through the thickness of the connecting bands . Overall , the data showed that the EST sequences shared between the ornaments and the shells were mainly from the unionoids , consistent with the results obtained by searching the annotated protein database ( Table 2 ) . From all of the archaeological samples , the Havnø set showed the best match to the freshwater unionoids , owing to better-preserved proteins , that is with high coverage and number of supporting peptides . The LC-MS/MS analysis of PesB produced a number of tandem mass spectra comparable to the other samples , but a lower number of sequences were identified . Sample HorA ( burnt ) , from which no proteins had been identified using the annotated protein database , yielded some matches to EST sequences , most of which shared with U . pictorum . In order to provide further , independent , evidence for the origin of the raw material , we developed an in-house tool ( in C language , available in Figure 5—source code 1 ) for ‘proteome comparison’ , using all the peptide sequences generated by the de novo algorithm of the software PEAKS , that is before performing any database search ( Appendix 1 , section 3 . 6 ) . The tool was able to provide a score for the sequence similarity between two lists of peptides and to generate a similarity matrix from all pairwise comparisons , which was then converted to a distance matrix . Multidimensional scaling ( MDS; Gower , 1966 ) was used to visualise the similarity of observations ( Figure 5 ) and confirmed that the Havnø set and the freshwater reference unionoids display the higher degree of proteome similarity , while the samples from Hornstaad and PesB fall in a different area of the plot from each other and from the marine reference shells . The results were also in accordance with those obtained from another database-independent approach Appendix 1—figure 25 , based on direct product ion spectra comparison ( Rieder et al . , 2017 ) , which was adapted for this study ( Appendix 1 , section 3 . 6 ) . Overall , our study , which represents one of the few that attempts to compare molluscan proteomes within the same clade ( genus , family or order ) , shows that unionoid shells exhibit very similar proteome profiles , sharing many sequences between species Appendix 1—figure 23 . This may suggest that this group has a rather conserved , homogeneous and recognisable proteomic signature , a conclusion that is completely congruent with earlier findings ( Marie et al . , 2017 ) . Furthermore , all analyses showed that the three sets of archaeological ornaments have similar proteome profiles ( Table 2 , Figures 4 and 5 , Appendix 1—figure 24 , but do not exhibit a simple correspondence to a molluscan species , at least among the Unionoida considered here , further highlighting the complexity of molluscan shell proteomes . We examined the sequence of the top-scoring protein , Hic74 , recovered from the reference shells and ornaments , with the aim of assessing the presence and frequency of any amino acid substitutions , which could potentially yield taxonomic resolution within Unionoida . Figure 6 shows the alignment ( performed using the software Geneious Prime 2019 . 1 . 1 ) of these incomplete sequences to the reference ( Hic74 from Hyriopsis cumingii ) . The sequence coverage of each sample was obtained from the ‘Spider’ output of PEAKS 8 . 5 . The Spider algorithm takes into account potential amino acid substitutions , as well as a large number of in vivo , laboratory-induced ( e . g . carbamidomethylation ) , and diagenetically-relevant ( e . g . deamidation ) modifications , therefore it is especially useful in highlighting possible mutation sites . In our sequence reconstruction we only considered peptides displaying typical sample preparation-induced or diagenesis-induced modifications ( Figure 6—figure supplement 1 , Figure 6—figure supplement 2 , Figure 6—figure supplement 3 , Figure 6—figure supplement 4 , Figure 6—figure supplement 5 , Figure 6—figure supplement 6 , Figure 6—figure supplement 7 , Figure 6—figure supplement 8 , Figure 6—figure supplement 9 , Figure 6—figure supplement 10 ) . The potential amino acid substitutions and their positions are summarised in Table 3 ( supporting product ion spectra can be found in Figure 6—source data 1 ) . In the best-case scenario , only around half of the Hic74 ( Hyriopsis cummingii ) reference sequence was covered ( Table 2 , Figure 6 ) . This may be due to: genuine sequence differences between Hyriopsis and all the other Unionoida examined here; low susceptibility of low-complexity domains to enzymatic cleavage; selective post mortem ( or laboratory-induced ) degradation of half of the sequence; errors in the transcriptome assembly of the protein . It is likely that a combination of all of these factors is responsible for this , particularly as the Hic74 regions not covered in the samples are mainly low-complexity domains , sometimes highly polar and thus prone to hydrolysis ( Ser and Asp-rich ) . Within the limits due to the incomplete coverage of the mutation sites in reference shells , it is interesting to note that the Hic74 sequence of P . auricularius diverges significantly from those of U . pictorum , U . crassus and M . margaritifera , which appear to have a higher degree of sequence similarity ( Table 3 ) . While this is in contrast with a recent taxonomy reassessment of Unionida based on mitochondrial DNA ( Lopes-Lima et al . , 2018 ) , it has been noted before that shell proteomes do not follow a simple phylogenetic signal ( Jackson et al . , 2010 ) . The sequence coverage was insufficient to attempt any further consideration for the Hornstaad and PesB samples , but the better-preserved Havnø double-buttons shared the same amino acid substitutions , supporting the hypothesis that the same species was used to make these three ornaments ( Table 3 ) . Furthermore , Table 3 shows that this taxon was unlikely to be P . auricularius , and more likely to be Unio or Margaritifera sp . The raw material used to manufacture the seven double-buttons can be firmly and consistently identified as Unionoida , freshwater shells with a thick mother-of-pearl layer , on the basis of morphological , microstructural , mineralogical , geochemical and biomolecular data . From a microstructural viewpoint alone , the combination of aragonitic prisms and sheet nacre ( ‘brickwall type’ ) structures is restricted to three bivalvian orders: the Unionoida , the Trigonioida and the Anomalodesmata , the first two belonging to sub-class Palaeoheterodonta ( Taylor et al . , 1973 ) . The Trigonioida relic order could be ruled out , since it is represented nowadays by a single genus , Neotrigonia , with very small shells and living exclusively on the Australian and Tasmanian coasts . Geochemical data , that is stable isotope values of carbon and oxygen , overall indicated that all biominerals studied here were formed in freshwater environments , and δ18O values in double-buttons tracked the average annual δ18O values of local precipitations . This excluded Anomalodesmata as potential candidates , since this order of enigmatic , rare and specialised bivalves are strictly marine ( Taylor et al . , 1973 ) . This finally left only one possibility , Unionoida , the representatives of which are all freshwater bivalves . Biomolecular analyses showed that the proteome similarity is highest between the double-buttons and the unionoid reference shells . The identification of proteins Hic74 , Hic52 and silkmapin in almost all of the archaeological samples confirms the freshwater nacre ( Unionoida ) origins of the double-buttons . With our current knowledge on shell proteins , these sequences probably represent taxon-specific adaptations for the biomineralization of nacroprismatic structures in unionoid shells: they do not bear any homologues with other shell proteins and are not found in the proteomes of other non-nacreous shell structures characterized here ( O . edulis ) . Furthermore , the analysis of the amino acid substitutions on the Hic74 sequence , recovered from both the Unionoida shells and the ornaments , indicates that Unio or M . margaritifera ( not Pseudunio ) had been used for making the Havnø ornaments . Except technical bias , inherent to standard proteomics per se and discussed elsewhere ( Marin et al . , 2016 ) , we identified three potential sources of bias that may hamper , limit or confound the current use of ‘palaeoshellomics’: 1 ) the intrinsic peculiarities of several shell matrix proteins; 2 ) the completeness of the dataset used for identification searches; 3 ) the diagenetic degradation of shell proteins . While diagenesis may represent a true source of bias , we were able to accurately evaluate its effects , and we found that the extent of protein degradation ( racemization , deamidation ) was consistent with the age and burial history of the samples . More specifically , we observed that samples from Havnø were the best preserved - the coverage of the main proteins was high , especially for Hic74 ( up to 55% ) . Surprisingly , we found that the coverage of this protein in Havnø samples is similar to that of modern U . pictorum and Neolithic U . crassus and P . auricularius - indeed , the number of Hic74 peptides in Havnø even surpasses that of the reference unionoid shells , despite the fact that the sample size for the archaeological double-buttons was at least 100 times smaller ( Appendix 1 , section 3 ) . We assume that this effect is due to early diagenetic changes ( such as protein unfolding , loss of linked sugars ) that render the protein backbone more accessible to proteolytic enzymes , thus increasing the chance of releasing and identifying peptides . Interestingly , we observed a similar phenomenon in other mineralised systems , for example ostrich eggshell ( Demarchi et al . , 2016 ) . Samples from Hornstaad yielded significantly lower coverages ( 7% and 12% respectively for HorB and HorC ) with fewer supporting peptides ( 6 and 11; Table 2 ) . In one instance ( sample HorA ) , no proteins were identified . This is consistent with the results of the chiral amino acid analysis , which had flagged this sample as compromised and probably burnt , as well as with archaeological evidence for widespread fire destruction of the settlement ( Heumüller , 2012 ) . However , the same sample yielded a high number of unidentified peptide sequences ( ~6000 ) , some of which appeared to be highly acidic and reminiscent of biogenic carbonate-associated proteins ( Marin and Luquet , 2008 ) . Furthermore , neither the microstructure nor the mineralogy of the double-buttons from Hornstaad showed any apparent sign of recrystallization to secondary calcite . We therefore hypothesize that the exposure to high temperatures had been relatively moderate , sufficient for inducing protein degradation and/or modification ( Asp and Ser decomposition , amino acid racemization , Appendix 1—figure 9 ) but not high enough to induce mineral conversion , which starts to occur around 300 °C ( Yoshioka and Kitano , 1985 ) . Double-button PesB yielded low D/L values but only a modest number of peptides were identified ( ~100 , much less compared to the Havnø samples , where at least 200–400 peptides were matched to known shell protein sequences ) ; this suggests that the sample had not been diagenetically compromised ( also supported by the amino acid data , Appendix 1—figure 7 , Appendix 1—figure 9 ) . We found that mother-of-pearl of freshwater origin ( Unionoida ) was used in three European sites over a wide geographical range but relatively short time span ( ~4200–3800 BCE ) . Crucially , the crafters manufacturing such highly-standardized ornaments belong to different cultural groups: Late Mesolithic , Neolithic and Copper Age . Our results settle the ‘marine vs freshwater’ debate ( Heumüller , 2012 ) for the double-buttons from Hornstaad ( Borrello and Girod , 2008 ) , and confirm previous identifications for the Peştera Ungurească examples ( Girod , 2010a ) . The use of freshwater nacre ( Unio or Margaritifera ) comes as a surprise for the Havnø material , a coastal shell midden with a dominance of marine resource exploitation and rich in marine shells perfectly suitable for the purpose of making beads , including the horsemussel M . modiolus . Therefore , this finding suggests that the importance of freshwater mother-of-pearl be re-evaluated . Unio sp . shells were probably selected to make ‘disc beads’ in the Epipaleolithic of the Levant , at Eynan ( Natufian , 10 , 000–8 , 000 BCE; Bar-Yosef Mayer , 2013 ) , and in Europe the presence of Unio sp . beads has been recorded 259 times according to the dataset gathered from the literature by Rigaud et al . ( 2015 ) , mainly from Neolithic sites . Despite this relative frequency of ( presumed ) freshwater mollusc ornaments in prehistoric Europe , a systematic study of their exploitation as raw materials is almost completely lacking . This is especially surprising since it is known that unionoid shells were exploited for mother-of-pearl until the Middle Ages ( Bertin , 2015 ) . Indeed , North American freshwater mussels were the basis for the ‘pearl rush’ during the 19th century , and their overexploitation for pearl harvesting , for making nuclei to be inserted in Pinctada pearl oysters as well as button-making on an industrial scale , almost drove a high number of species to extinction ( Haag , 2012 ) . The lack of comprehensive archaeological studies on freshwater molluscs can be explained by two main factors . The first is methodological: the typical chaîne opératoire of bead-making involves several steps that obliterate most of the anatomical features ( e . g . hinge apophysis ) that are usable for taxonomic identification . These include: cutting and abrading small pieces until they take a circular shape; perforating the disc ( Gurova and Bonsall , 2017 ) or , in the case of the double-buttons , working the side ( with an abrasive wire ? ) to shape the central groove ( Bertin , 2015; Borrello and Girod , 2008 ) . Our work provides a series of analytical tools for overcoming this issue and for determining the biological origin of the raw material . A second , perhaps more relevant , factor is the long-standing perception that freshwater molluscs are inherently less ‘prestigious’ than marine species , because of their presumed local origin . However , marine and freshwater molluscs are used side-by-side in a number of instances , for example the high-status burials at Mulhouse-Est ( Bonnardin , 2009 ) , or complex parures from the Swiss Early Bronze Age ( Borrello and Girod , 2008 ) . This clearly demonstrates that both were held in the same ‘esteem’ by the craftsman and that her/his choice was dictated by reasons other than the ‘exoticism’ of the material . The use of freshwater mother-of-pearl at Hornstaad and Peştera Ungurească , two sites with large procurement networks of exotic raw materials and at which there are clear signs of specialised production of ornaments ( including gold at Peştera Ungurească; Biagi and Voytek , 2006; Heumüller , 2012 ) , also confirms that freshwater pearl mussels were seen as prized materials , locally available . Furthermore , the use of freshwater molluscs for the manufacture of the doppelknöpfe recovered from Havnø ( together with unworked fragments of the shells , Appendix 1—figure 2 ) shows that the manufacture of these ornaments was consistently associated with the use of freshwater mother-of-pearl , even in marine settings . Therefore , the Late Mesolithic people of Jutland and the Neolithic people of central Europe were either exchanging the finished products/raw materials , or the knowledge that the manufacture of the double-buttons required the use of unionoid shells . It is clear that mother-of-pearl ( nacre ) from freshwater molluscs was a prime material of choice for the manufacture of shell double-buttons . Further investigation of other types of shell ornaments may reveal that this raw material was more frequently selected than previously thought , but in the meanwhile it is necessary to consider the reasons behind this choice . Unionoids inhabit clean flowing waters ( they are occasionally also found in lakes ) and are dependent upon the presence of sufficient salmonid fish to carry the larval glochidial stage of the pearl mussel life cycle ( IUCN: International Union for Conservation of Nature , 2019 ) . It is highly likely that freshwater mussels were collected near the site ( as supported by the ‘local’ isotope signatures in this study ) , and that the procurement of the mussels was not especially difficult ( for example , M . margaritifera lives at depths of up to two meters ) nor too time-consuming . Therefore , the choice of this material must have been linked to reasons other than its long-distance provenance , the skills involved in procurement , or its rarity; rather , it is more likely a result of the characteristics of the raw material per se ( mechanical properties and aesthetic qualities ) and its connection to other things , be these in the sensory world ( the river and its water , the landscape ) or in the symbolic . Mother-of-pearl is exceptionally hard - a thousand times more resistant to fractures than its mineral alone ( Currey , 1977; Jackson et al . , 1990 ) - and unionoid shells ( Unio and Margaritifera sp . , but not Anodonta ) typically have a rather thick layer of nacre , while in some of the marine molluscs ( particularly those occurring in European waters , such as Modiolus sp . ) , the ratio between nacre and prisms shifts , favouring the latter , where the nacre only partly covers the inner surface of the shell . The preservation of the prismatic matte layer may indicate that the coloured periostracum , which can give an appealing effect of chromatic contrast , was deliberately kept , for aesthetic reasons . Alternatively , if the periostracum was removed by mild abrasion , this would have resulted in fully white ornaments , showing both the brilliance of nacre and the dullness of prisms . The white colour of the ornaments may have been associated to wellbeing , peace and fertility ( Trubitt , 2003 ) . White was certainly a sought-after effect , so much so that red-purple Spondylus shells were often worked in order to remove the striking hue and reveal the white underneath ( Borrello and Micheli , 2011 ) . At the same time , the gloss of mother-of-pearl has been linked , in historical periods , with spirituality , life , royalty , and pearl fishing is a tradition that dates back to the same period considered here , around or a few centuries before 5500 BCE , in the Arabian Peninsula ( Charpentier et al . , 2012 ) . The choice of the raw material could also be a reflection of the role of freshwater environments: the Neolithic is the period in which water , together with plants and animals , is ‘domesticated’ ( Garfinkel et al . , 2006; Mithen , 2010 ) . Rivers provided fast access routes to Central , Western and Northern Europe for hunter-gatherers during the Palaeolithic and Mesolithic and , later on , for agriculturalists coming from the East ( Rowley-Conwy , 2011 ) . Despite their ‘fluidity’ , rivers and lakes were meaningful and persistent places in the prehistoric landscape . In summary , the streams , rivers and lakes near occupation sites were inhabited by organisms that provided the crafters with exceptional-quality raw material , easy to procure and which could be worked following a well-established chaîne opératoire in order to obtain a standardized result . The small white double-buttons could then be threaded using the central groove or pressed into the fabric or leather ( Kannegaard , 2013 ) . Our work thus highlights an interpretative bias whereby exoticism is considered the primary reason for choice of raw materials , and suggest that local environments held an equally important place in the mind of prehistoric people . The first application of ‘palaeoshellomics’ has demonstrated that it is possible to recover and identify ancient proteins sequences from mollusc shell , despite significant analytical challenges due to the combined effects of several factors , including low protein concentrations , small samples sizes , diagenesis and database insufficiency ( Table 2 ) . Our molecular data showed that molluscan proteins are similar across the four freshwater taxa we examined ( U . pictorum , U . crassus , M . margaritifera , P . auricularius ) and differ significantly from the two marine species ( O . edulis , M . modiolus; Table 2 , Figures 4 and 5 ) . We confirmed that freshwater molluscan matrix proteins are characterized by highly repetitive low complexity domains ( RLCs ) . This is consistent with results obtained on other shell taxa , and improves our understanding of the biomineralization mechanisms within these invertebrate systems . The archaeological double-buttons examined here were all confidently identified as Unionoida , freshwater shells with a thick layer of mother-of-pearl , using a combination of mineralogical , geochemical and biochemical techniques ( SEM , FTIR-ATR , oxygen and carbon isotopes , chiral amino acid analyses , palaeoshellomics ) . The analysis of the sequence of the shell matrix protein Hic74 supports the use of Unio or Margaritifera as the raw material for the three Havnø ornaments ( excluding Pseudunio ) , but lack of coverage of most amino acid modification sites in the reference samples hampered identification to a lower taxonomic level ( Figure 6 , Table 3 ) . The high degree of standardization of the ornaments ( Figures 1 and 2 ) , as well as the consistent choice of freshwater mother-of-pearl as raw material indicate that , in Europe , between ~ 4200 and~3800 BCE , there was a common notion of the manufacture of the doppelknöpfe , which was shared by different cultural groups: Late Mesolithic ( Ertebølle ) , early Late Neolithic ( Hornstaad Group ) , and Copper Age ( Toarte Pastilate/Coţofeni ) . Our in-depth study therefore puts into question the most commonly accepted interpretations , which privilege the preponderant use of exotic marine shells as prestigious raw materials for the manufacture of prehistoric shell ornaments .
Whole beads and fragments of the reference shells were observed using an environmental Scanning Electron Microscope ( Hitachi TM1000 Tabletop Microscope ) . The mineralogy of the beads was identified by infrared spectroscopy in attenuated total reflectance ( ATR ) mode ( FTIR-ATR ) ( Appendix 1 , sections 3 . 1 and 3 . 2 ) . Isotopic analysis was carried out on biogenic carbonate to obtain bulk δ13C and δ18O values for the double-buttons . Small amounts of cleaned samples ( bleached using concentrated NaOCl ( 12% w/v ) for 48 hr ) were analysed using a Delta V Plus mass spectrometer coupled with a Kiel IV carbonate device ( ThermoFisher ) . All steps are detailed in Appendix 1 , section 3 . 3 . All reference shells and beads were powdered using a clean mortar and pestle and accurately weighed . All the mass spectrometry proteomics data have been deposited in the ProteomeXchange Consortium ( http://proteomecentral . proteomexchange . org ) via the PRIDE partner repository ( Vizcaíno et al . , 2013 ) with the data set identifier PXD011985 . | Just like people do today , prehistoric humans liked to adorn themselves with beautiful objects . Shells , from creatures like clams and snails , were used to decorate clothing or worn as jewelry at least as far back as 100 , 000 years ago . Later people used shells as the raw materials to make beads or bracelets . Learning where the shells came from may help scientists understand why prehistoric people chose certain shells and not others . It may also offer clues about how they used natural resources and the cultural significance of these objects . But identifying the shells is difficult because they lose many of their original distinctive features when worked into ornaments . New tools that use DNA or proteins to identify the raw materials used to craft ancient artifacts have emerged that may help . So far , scientists have mostly used these genomic and proteomic tools to identify the source of materials made from animal hide , ivory or bone – where collagen is the most abundant protein molecule . Yet it is more challenging to extract and characterize proteins or genetic material from mollusc shells . This is partly because the amount of proteins in shells is at least 300 times lower than in bone , and also because the makeup of proteins in shells is not as well-known as in collagen . Sakalauskaite et al . have now overcome these issues by combining the analytical tools used to study the proteins and mineral content of modern shells with those of ancient protein research . They then used this approach , which they named palaeoshellomics , to extract proteins from seven “double-buttons” – pearl-like ornaments crafted by prehistoric people in Europe . The double-buttons were made between 4200 and 3800 BC and found at archeological sites in Denmark , Germany and Romania . Comparing the extracted proteins to those from various mollusc shells showed that the double-buttons were made from freshwater mussels belonging to a group known as the Unionoida . The discovery helps settle a decade-long debate in archeology about the origin of the shells used to make double-buttons in prehistoric Europe . Ancient people often crafted ornaments from marine shells , because they were exotic and considered more prestigious . But the results on the double-buttons suggest instead that mother-of-pearl from fresh water shells was valued and used by groups throughout Europe , even those living in coastal areas . The palaeoshellomics technique used by Sakalauskaite et al . may now help identify the origins of shells from archeological and palaeontological sites . | [
"Abstract",
"Introduction",
"Results",
"and",
"discussion",
"Methods"
] | [
"evolutionary",
"biology",
"biochemistry",
"and",
"chemical",
"biology"
] | 2019 | 'Palaeoshellomics’ reveals the use of freshwater mother-of-pearl in prehistory |
Neurons in sensory cortex are tuned to diverse features in natural scenes . But what determines which features neurons become selective to ? Here we explore the idea that neuronal selectivity is optimized to represent features in the recent sensory past that best predict immediate future inputs . We tested this hypothesis using simple feedforward neural networks , which were trained to predict the next few moments of video or audio in clips of natural scenes . The networks developed receptive fields that closely matched those of real cortical neurons in different mammalian species , including the oriented spatial tuning of primary visual cortex , the frequency selectivity of primary auditory cortex and , most notably , their temporal tuning properties . Furthermore , the better a network predicted future inputs the more closely its receptive fields resembled those in the brain . This suggests that sensory processing is optimized to extract those features with the most capacity to predict future input .
Sensory inputs guide actions , but such actions necessarily lag behind these inputs due to delays caused by sensory transduction , axonal conduction , synaptic transmission , and muscle activation . To strike a cricket ball , for example , one must estimate its future location , not where it is now ( Nijhawan , 1994 ) . Prediction has other fundamental theoretical advantages: a system that parsimoniously predicts future inputs from their past , and that generalizes well to new inputs , is likely to contain representations that reflect their underlying causes ( Bialek et al . , 2001 ) . This is important because ultimately , we are interested in these causes ( e . g . flying cricket balls ) , not the raw images or sound waves incident on the sensory receptors . Furthermore , much of sensory processing involves discarding irrelevant information , such as that which is not predictive of the future , to arrive at a representation of what is important in the environment for guiding action ( Bialek et al . , 2001 ) . Previous theoretical studies have suggested that many neural representations can be understood in terms of efficient coding of natural stimuli in a short time window at or just before the present ( Attneave , 1954; Barlow , 1959; Olshausen and Field , 1996 , Olshausen and Field , 1997 ) . Such studies generally built a network model of the brain , which was trained to represent stimuli subject to some set of constraints . One pioneering such study trained a network to efficiently represent static natural images using a sparse , generative model ( Olshausen and Field , 1996 , Olshausen and Field , 1997 ) . More recent studies have used related ideas to model the representation of moving ( rather than static ) images ( van Hateren and Ruderman , 1998a; Berkes and Wiskott , 2005; Berkes et al . , 2009 ) and other sensory stimuli ( Klein et al . , 2003; Carlson et al . , 2012; Zhao and Zhaoping , 2011; Kozlov and Gentner , 2016; Cusack and Carlyon , 2004 ) . In contrast , we built a network model that was optimized not for efficient representation of the recent past , but for efficient prediction of the immediate future of the stimulus , which we will refer to as the temporal prediction model . The timescale of prediction considered for our model is in the range of tens to hundreds of milliseconds . Conduction delays to cortex and very fast motor responses are on this timescale ( Bixler et al . , 1967; Yeomans and Frankland , 1995; Bizley et al . , 2005 ) . The idea that prediction is an important component of perception dates at least as far back as Helmholtz ( Helmholtz , 1962; Sutton and Barton , 1981 ) , although what is meant by prediction and the purpose it serves is quite varied between models incorporating it ( Chalk et al . , 2018; Salisbury and Palmer , 2016 ) . With regards to perception and prediction , two contrasting but interrelated frameworks have been distinguished ( Chalk et al . , 2018; Salisbury and Palmer , 2016 ) . In the ‘predictive coding’ framework ( Huang and Rao , 2011; Rao and Ballard , 1999; Friston , 2003 ) , prediction is used to remove statistical redundancy in order to provide an efficient representation of the entire stimulus . Some models of this type use prediction as a term for estimation of the current or a static input ( such as images ) from latent variables ( Rao and Ballard , 1999 ) , whereas other have also considered the temporal dimension of the input ( Rao and Ballard , 1997; Rao , 1999; Srinivasan et al . , 1982 ) . Sparse coding models ( Olshausen and Field , 1996 , Olshausen and Field , 1997 ) can be related to this framework ( Huang and Rao , 2011 ) . In contrast , the ‘predictive information’ framework ( Bialek et al . , 2001; Salisbury and Palmer , 2016; Palmer et al . , 2015; Heeger , 2017 ) , which our approach relates to more closely , involves selective encoding of those features of the stimulus that predict future input . A related idea to predictive information is the encoding of slowly varying features ( Berkes and Wiskott , 2005; Creutzig and Sprekeler , 2008; Kayser et al . , 2001; Hyvärinen et al . , 2003 ) , which are one kind of predictive feature . Hence , the predictive coding approach seeks to find a compressed representation of the entire input , whereas the predictive information approach selectivity encodes only predictive features ( Chalk et al . , 2018; Salisbury and Palmer , 2016 ) . Our model relates to the predictive information approach in that it is optimized to predict the future from the past , but it has a combination of characteristics , such a non-linear encoder and sparse weight regularization , which have not previously been explored for such an approach . To evaluate the representations produced by these normative theoretical models , they can be optimized for natural stimuli , and the tuning properties of their units compared to the receptive fields of real neurons . A useful and commonly used definition of a neuron’s receptive field ( RF ) is the stimulus that maximally linearly drives the neuron ( Adelson and Bergen , 1985; Aertsen et al . , 1981; Aertsen and Johannesma , 1981; Reid et al . , 1987; deCharms et al . , 1998; Harper et al . , 2016 ) . In mammalian primary visual cortex ( V1 ) , neurons typically respond strongly to oriented edge-like structures moving over a particular retinal location ( Hubel and Wiesel , 1959; Jones and Palmer , 1987; DeAngelis et al . , 1993; Ringach , 2002 ) . In mammalian primary auditory cortex ( A1 ) , most neurons respond strongly to changes in the amplitude of sounds within a certain frequency range ( deCharms et al . , 1998 ) . The temporal prediction model provides a principled approach to understanding the temporal aspects of RFs . Previous models , based on sparsity or slowness related principles , were successful in accounting for many spatial aspects of V1 RF structure ( Olshausen and Field , 1996 , Olshausen and Field , 1997; van Hateren and Ruderman , 1998a; Berkes and Wiskott , 2005; Berkes et al . , 2009; van Hateren and van der Schaaf , 1998b ) , and had some success in accounting for spectral aspects of A1 RF structure ( Klein et al . , 2003; Carlson et al . , 2012; Zhao and Zhaoping , 2011; Cusack and Carlyon , 2004 ) . However , these models do not account well for the temporal structure of V1 or A1 RFs . Notably , for both vision ( Ringach , 2002 ) and audition ( deCharms et al . , 1998 ) , the envelopes of real neuronal RFs tend to be asymmetric in time , with greater sensitivity to very recent inputs compared to inputs further in the past . In contrast , the RFs predicted by previous models ( van Hateren and Ruderman , 1998a; Klein et al . , 2003; Carlson et al . , 2012; Kozlov and Gentner , 2016; Cusack and Carlyon , 2004 ) typically show symmetrical temporal envelopes , with either approximately flat envelopes over time or a balanced falloff of the envelope over time either side of a peak . They also lack the greater sensitivity to very recent inputs . Here we show using qualitative and quantitative comparisons that , for both V1 and A1 RFs , these shortcomings are largely overcome by the temporal prediction approach . This suggests that neural sensitivity at early levels of the cortical hierarchy may be organized to facilitate a rapid and efficient prediction of what the environment will look like in the next fraction of a second .
To determine what type of sensory RF structures would facilitate predictions of the imminent future , we built a feedforward network model with a single layer of nonlinear hidden units , mapping the inputs to the outputs through weighted connections ( Figure 1 ) . Each hidden unit’s output results from a linear mapping ( by input weights ) from the past input , followed by a monotonic nonlinearity , much like the classic linear-nonlinear model of sensory neurons ( Klein et al . , 2003; Carlson et al . , 2012; Zhao and Zhaoping , 2011 ) . The model then generates a prediction of the future from a linear mapping ( by output weights ) from the hidden units’ outputs . This is consistent with the observation that decoding from the neural response is often well approximated by a linear transformation ( Eliasmith and Anderson , 2003 ) . We trained the temporal prediction model on extensive corpora , either of soundscapes or silent movies , modelling A1 ( Figure 1a ) or V1 ( Figure 1b ) neurons , respectively . In each case , the networks were trained by optimizing their synaptic weights to most accurately predict the immediate future of the stimulus from its very recent past . For vision , the inputs were patches of videos of animals moving in natural settings , and we trained the network to predict the pixel values for one movie frame ( 40 ms ) into the future , based on the seven most recent frames ( 280 ms ) . For audition , we trained the network to predict the next three time steps ( 15 ms ) of cochleagrams of natural sounds based on the 40 most recent time steps ( 200 ms ) . Cochleagrams resemble spectrograms but are adjusted to approximate the auditory nerve representation of sounds ( see Materials and methods ) . During training we used sparse , L1 weight regularization ( see Equation 3 in Materials and methods ) to constrain the network to predict future stimuli in a parsimonious fashion , forcing the network to use as few weights as possible while maintaining an accurate prediction . This constraint can be viewed as an assumption about the sparse nature of causal dependencies underlying the sensory input , or alternatively as analogous to the energy and space restrictions of neural connectivity . It also prevents our network model from overfitting to its inputs . Note that this sparsity constraint differs from that used in sparse coding models , in that it is applied to the weights rather than the activity of the units , being more like a constraint on the wiring between neurons than a constraint on their firing rates . To compare with the model , we recorded responses of 114 auditory neurons ( including 76 single units ) in A1 and the anterior auditory field ( AAF ) of 5 anesthetized ferrets ( Willmore et al . , 2016 ) and measured their spectrotemporal RFs ( see Materials and methods ) . Ferrets are commonly used for auditory research , because they are readily trained in a range of sound detection , discrimination or localization tasks ( Nodal and King , 2014 ) , the frequency range of their hearing ( approximately 40 Hz–40 kHz [Kavanagh and Kelly , 1988] ) overlaps well with ( and extends beyond ) the human range , and most of their auditory cortex is not buried in a sulcus and hence easily accessible for electrophysiological or optical measurements . The A1 RFs we recorded are diverse ( Figure 2a ) ; their frequency tuning can be narrowband or broadband , and sometimes showing flanking inhibition . Some may also be more complex in frequency tuning , lack clear order , or be selective for the direction of frequency modulation ( Carlin and Elhilali , 2013 ) . In their temporal tuning , A1 RFs tend to weight recent inputs more heavily , with a temporally asymmetric power profile , involving excitation near the present followed by lagging inhibition of a longer duration ( deCharms et al . , 1998 ) . The temporal prediction model RFs ( Figure 2b ) are similarly diverse , showing all of the RF types seen in vivo ( including examples of localized , narrowband , broadband , complex , disordered and directional RFs ) and are well matched in scale and form to those measured in A1 . This includes having greater power ( mean square ) near the present , with brief excitation followed by longer lagging inhibition , producing an asymmetric power profile . This stands in contrast to previous attempts to model RFs based on efficient coding , sparsecoding and slow feature hypotheses , which either did not capture the diversity of RFs ( Zhao and Zhaoping , 2011 ) , or lacked temporal asymmetry , punctate structure , or appropriate time scale ( Klein et al . , 2003; Carlson et al . , 2012; Kozlov and Gentner , 2016; Cusack and Carlyon , 2004; Carlin and Elhilali , 2013; Brito and Gerstner , 2016 ) . By eye , substantial similarities were also apparent when we compared the temporal prediction model’s RFs trained using visual inputs ( Figure 1b ) with the 3D ( space-space-time ) and 2D ( space-time ) spatiotemporal RFs of real V1 simple cells , which were obtained from Ohzawa et al ( Ohzawa et al . , 1996 ) . Simple cells ( Hubel and Wiesel , 1959 ) have stereotyped RFs containing parallel , spatially localized excitatory and inhibitory regions , with each cell having a particular preferred orientation and spatial frequency ( Jones and Palmer , 1987; DeAngelis et al . , 1993; Ringach , 2002 ) ( Figure 2c ) . These features are also clearly apparent in the model RFs ( Figure 2d ) . Unlike previous models ( van Hateren and Ruderman , 1998a; Hyvärinen et al . , 2003; Olshausen , 2003 ) , the temporal prediction model captures the temporal asymmetry of real RFs . The RF power is highest near the present and decays into the past ( Figure 2d ) , as observed in real neurons ( Ohzawa et al . , 1996 ) ( Figure 2c ) . Furthermore , simple cell RFs have two types of spatiotemporal structure: space-time separable RFs ( Figure 2cI ) , whose optimal stimulus resembles a flashing or slowly ramping grating , and space-time inseparable RFs , whose optimal stimulus is a drifting grating ( DeAngelis et al . , 1993 ) ( Figure 2cII ) . Our model captures this diversity ( Figure 2dI–III separable , Figure 2dIV–VI inseparable ) . We also examined linear aspects of the tuning of the output units for the visual temporal prediction model using a response-weighted average to white noise input , and found punctate non-oriented RFs that decay into the past . For comparison , we trained a sparse coding model ( Olshausen and Field , 1996 , Olshausen and Field , 1997; Carlson et al . , 2012 ) ( https://github . com/zayd/sparsenet ) using our dataset . We would expect such a model to perform less well in the temporal domain , because unlike the temporal prediction model , the direction of time is not explicitly accounted for . The sparse coding model was chosen because it has set the standard for normative models of visual RFs ( Olshausen and Field , 1996 , Olshausen , 2003; Olshausen and Field , 1997 ) , and the same model has also been applied for auditory RFs ( Carlson et al . , 2012; Brito and Gerstner , 2016; Młynarski and McDermott , 2017; Blättler et al . , 2011 ) . Past studies ( Olshausen and Field , 1996 , Olshausen and Field , 1997; Carlson et al . , 2012 ) have largely analysed the basis functions produced by the sparse coding model and compared their properties to neuronal RFs . To be consistent with these studies we have done the same , and to have a common term , refer to the basis functions as RFs ( although strictly , they are projective fields ) . We can visually compare the large set of RFs recorded from A1 neurons ( Figure 3 ) to the full set of RFs obtained from the temporal prediction model when trained on auditory inputs ( Figure 4 ) and those of the sparse coding model ( Figure 5 ) when trained on the same auditory inputs . A range of RFs were produced by the sparse coding model , some of which show characteristics reminiscent of A1 RFs , particularly in the frequency domain . However , the temporal properties of A1 neurons are not well captured by these RFs . While some RFs display excitation followed by lagging inhibition , very few , if any , show distinct brief excitation followed by extended inhibition . Instead , RFs that show both excitation and inhibition tend to have a symmetric envelope and these features are randomly localized in time , and many RFs display temporally elongated structures that are not found in A1 neurons . We also trained the sparse coding model on the dataset of visual inputs to serve as a control for the temporal prediction model trained on these same inputs . We compared the full population of spatial and 2D spatiotemporal visual RFs of the temporal prediction model ( Figure 4—figure supplements 2–3 ) and the sparse coding model ( Figure 5—figure supplements 1–2 ) . As shown in previous studies ( Olshausen and Field , 1996 , Olshausen and Field , 1997; van Hateren and Ruderman , 1998a; van Hateren and van der Schaaf , 1998b ) , the sparse coding model produces RFs whose spatial structure resembles that of V1 simple cells ( Figure 5—figure supplements 1–2 ) , but does not capture the asymmetric nature of the temporal tuning of V1 neurons . Furthermore , while it does produce examples of both separable and inseparable spatiotemporal RFs , those that are separable tend to be completely stationary over time , resembling immobile rather than flashing gratings ( Figure 5—figure supplement 2 ) . We compared the RFs generated by both models to the RFs of the population of real A1 neurons we recorded . We first compared the RFs in a non-parametric manner by measuring the Euclidean distances between the coefficient values of the RFs , and then used multi-dimensional scaling to embed these distances in a two-dimensional space ( Figure 6a ) . The RFs of the sparse coding model span a much larger region than the real A1 and temporal prediction model RFs . Furthermore , the A1 and temporal prediction model RFs occupy a similar region of the space , indicating their greater similarity to each other relative to those of the sparse coding model . We then examined specific attributes of the RFs to determine points of similarity and difference between each of the models and the recorded data . We first considered the temporal properties of the RFs and found that for the data and the temporal prediction model , most of the power is contained in the most recent time-steps ( Figures 2a–b , 3–4 and 6b , and Figure 4—figure supplement 1 ) . Given that the direction of time is not explicitly accounted for in the sparse coding model , as expected , it does not show this feature ( Figures 5 and 6b ) . Next , we examined the tuning widths of the RFs in each population for both time and frequency , looking at excitation and inhibition separately . In the time domain , the real data tend to show leading excitation followed by lagging inhibition of longer duration ( Figures 2a , 3 and 6c–e ) . The temporal prediction model also shows many RFs with this temporal structure , with lagging inhibition of longer duration than the leading excitation ( Figures 2b , 4 and 6c–e , and Figure 4—figure supplement 1 ) . This is not the case with the sparse coding model , where units tend to show either excitation and inhibition having the same duration or an elongated temporal structure that does not show such stereotyped polarity changes ( Figures 5 and 6c–e ) . It is also the case that the absolute timescales of excitation and inhibition match the data more closely in the case of the temporal prediction model ( Figure 6c–e ) , although a few units display inhibition of a longer duration than is seen in the data ( Figure 6c ) . The sparse coding model shows a wide range of temporal spans of excitation and inhibition , in keeping with previous studies ( Carlson et al . , 2012; Carlin and Elhilali , 2013 ) . Regarding the spectral properties of real neuronal RFs , the spans of inhibition and excitation over sound frequency tend to be similar ( Figure 6f–h ) . This is also seen in the temporal prediction model , albeit with slightly more variation ( Figure 6f–h ) . The sparse coding model shows more extensive variation in frequency spans than either the data or our model ( Figure 6f–h ) . We also compared the spatiotemporal RFs derived from the temporal prediction and sparse coding models with restricted published datasets summarizing RF characteristics of V1 neurons ( Ringach , 2002 ) and a small number of full spatiotemporal visual RFs acquired from Ohzawa et al ( Ohzawa et al . , 1996 ) . We assessed the orientation and spatial frequency tuning properties of the models’ RFs by fitting Gabor functions to them ( see Materials and methods ) . We compared temporal properties of the RFs from the neural data and the temporal prediction model . In both cases , most power ( mean over space and neurons of squared values ) is in the most recent time steps ( Figure 7a ) . Previous normative models of spatiotemporal RFs ( van Hateren and Ruderman , 1998a; Hyvärinen et al . , 2003; Olshausen , 2003 ) ( Figure 7—figure supplement 1c–d ) do not show this property , being either invariant over time or localized , but with a symmetric profile that is not restricted to the recent past . We also measured the space-time separability of the RFs of the temporal prediction model ( see Materials and methods ) ; substantial numbers of both space-time separable and inseparable units were apparent ( 631 separable , 969 inseparable; Figure 4—figure supplement 3 ) . In addition to this , we measured the tilt direction index ( TDI ) of the model units from their 2D spatiotemporal RFs . This index indicates spatiotemporal asymmetry in space-time RFs and correlates with direction selectivity ( DeAngelis et al . , 1993; Pack et al . , 2006; Anzai et al . , 2001; Baker , 2001; Livingstone and Conway , 2007 ) . The mean TDI for the population was 0 . 34 ( 0 . 29 SD ) , comparable with the ranges in the neural data ( mean 0 . 16; 0 . 12 SD in cat area 17/18 ( Baker , 2001 ) , mean 0 . 51; 0 . 30 SD in macaque V1 [Livingstone and Conway , 2007] ) . Finally , we observed an inverse correlation ( r2 = −0 . 33 , p<10−9 , n = 1205 ) between temporal and spatial frequency tuning ( See Materials and methods ) , which is also a property of real V1 RFs ( DeAngelis et al . , 1993 ) and is seen in a sparse-coding-related model ( van Hateren and Ruderman , 1998a ) . The spatial tuning characteristics of the temporal prediction model’s RFs displayed a wide range of orientation and spatial frequency preferences , consistent with the neural data ( DeAngelis et al . , 1993; Kreile et al . , 2011 ) ( Figure 4—figure supplement 2 ) . Both model and real RFs ( Kreile et al . , 2011 ) show a preference for spatial orientations along the horizontal and vertical axes , although this orientation bias is seen to a greater extent in the temporal prediction model than in the data . The orientation and frequency tuning characteristics are also well captured by sparse coding related models of spatiotemporal RFs ( van Hateren and Ruderman , 1998a; Olshausen , 2003 ) ( Figure 7—figure supplement 1e-f ) . Furthermore , the widths and lengths of the RFs of the temporal prediction model , relative to the period of their oscillation , also match the neural data well ( Figure 7d ) . The distribution of units extends along a curve from blob-like RFs , which lie close to the origin in this plot , to stretched RFs with several subfields , which lie further from the origin . Although this property is again fairly well captured by previous models ( Olshausen and Field , 1996 , Olshausen and Field , 1997; Berkes et al . , 2009; Ringach , 2002; van Hateren and van der Schaaf , 1998b ) ( Figure 7—figure supplement 1g ) , only the temporal prediction model seems to be able to capture the blob-like RFs that form a sizeable proportion of the neural data ( Ringach , 2002 ) ( Figure 7d where nx and ny < ~0 . 25 , Figure 4—figure supplement 2 ) . A small proportion of the population have RFs with several short subfields , forming a wing from the main curve in Figure 7d . Under our hypothesis of temporal prediction , we would expect that the better the temporal prediction model network is at predicting the future , the more the RFs of the network should resemble those of real neurons . To examine this hypothesis , we plotted the prediction error of the network as a function of two hyperparameters; the regularization strength and the number of hidden units ( Figure 8a ) . Then , we plotted the similarity between the auditory RFs of real A1 neurons and those of the temporal prediction model ( Figure 8b ) , as measured by the mean KS distances of the temporal and frequency span distributions ( Figure 6d–e , g–h , Materials and methods ) . The set of hyperparameter settings that give good predictions are also those where the temporal prediction model produces RFs that are most similar to those recorded in A1 ( r2 = 0 . 8 , p<10−9 , n = 55 ) . This result argues that cortical neurons are indeed optimized for temporal prediction . When the similarity measure was examined as a function of the same hyperparameters for the sparse coding model ( Figure 8—figure supplement 1 ) , and this was compared to that model’s stimulus reconstruction capacity as a function of the same hyperparameters , a monotonic relationship between stimulus reconstruction capacity and similarity of real RFs was not found ( Figure 8—figure supplement 1; r2 = −0 . 05 , p=0 . 69 , n = 50 ) . In previous studies in which comparisons have been made between normative models and real data , the model hyperparameters have been selected to maximize the similarity between the real and model RFs . In contrast , the temporal prediction model provides an independent criterion , the prediction error , to perform hyperparameter selection . To our knowledge , no such effective , measurable , independent criterion for hyperparameter selection has been proposed for other normative models of RFs . The change in the qualitative structure of the RFs as a function of the number of hidden units and L1 regularization strength , for both the visual and auditory models , can be seen in the interactive supplementary figures ( Figure 8—figure supplements 2–3; https://yossing . github . io/temporal_prediction_model/figures/interactive_supplementary_figures . html ) The main effect of the regularization is to restrict the RFs in space for the visual case and in frequency and time for the auditory case . When the regularization is non-existent or substantially weaker than the optimum for prediction , the visual RFs become less localized in space with more elongated bars . The auditory RFs become more disordered , losing clear structure in most cases . When the regularization is made stronger than the optimum , the RFs become more punctate , for both the visual and auditory models . When the regularization strength is at the optimum for prediction , the auditory and visual model RFs qualitatively most closely resemble those of A1 neurons and V1 simple cells , respectively . This is consistent with what we found quantitatively in the previous section for the auditory model . The temporal prediction model and the sparse coding model both produce oriented Gabor-like RFs when trained on visual inputs . This raises the possibility that optimization for prediction implicitly optimizes for a sparse response distribution , and hence leads to oriented RFs . To test for this , we measured the sparsity of the visual temporal prediction model’s hidden unit activities ( by the Vinje-Gallant measure [Baker , 2001] ) in response to the natural image validation set . Examining the relationship between predictive capacity and sparsity , over the range of L1 weight regularization strength and hidden units explored , we did not find a clear monotonic relationship . Indeed , in both the auditory and visual cases , the hidden unit and L1 regularization combination with the best prediction had intermediate sparsity . For the visual case , the best-predicting model had sparsity 0 . 25 , and other models within the grid search had sparsity ranging from 0 . 16 to 0 . 57 . For the auditory case , the best-predicting model had sparsity 0 . 58 , and other models had sparsity ranging from 0 . 42 to 0 . 69 . We also varied other characteristics of the temporal prediction model to understand their influence . For both the auditory and visual models , when a different hidden unit nonlinearity ( tanh or rectified linear ) was used , the networks had similar predictive capacity and produced comparable RFs . However , when the temporal prediction model had linear hidden units , it no longer predicted as well and produced RFs that were less like real neurons in their structure . For the auditory model , the linear model RFs generally became more narrowband in frequency with temporally extended excitation , instead of extended lagging inhibition ( Figure 4—figure supplement 4 ) . For the visual model , the linear model RFs also showed substantially less similarity to the V1 data . At low regularization ( the best predicting case ) , the RFs formed full-field grid-like structures . At higher regularization , they were more punctate , with some units having oriented RFs with short subfields . The RFs also did not change form or polarity over time , but simply decayed into the past . The temporal prediction model and sparse coding model results shown in the main figures of this paper were trained on inputs with added Gaussian noise ( 6 dB SNR ) , mimicking inherent noise in the nervous system . To determine the effect of adding this noise , all models were also trained without noise , producing similar results ( Figure 4—figure supplements 5–7; Figure 5—figure supplements 3–5; Figure 6—figure supplement 1; Figure 7—figure supplements 2–3 ) . The results were also robust to changes in the duration of the temporal window being predicted . We trained the auditory model to predict a span of either 1 , 3 , 6 , or 9 time steps into the future and the visual model to predict 1 , 3 or 6 time steps into the future . For the auditory case , we found that increasing the number of time steps being predicted had little effect on the RF structure , both qualitatively and by the KS measure of similarity to the real data . In the visual case , Gabor-like units were present in all cases . Increasing the number of time steps made the RFs more restricted in space and increased the proportion of blob-like RFs .
A number of principles , often acting together , have been proposed to explain the form and diversity of sensory RFs . These include efficient coding ( Barlow , 1959; Olshausen and Field , 1996 , Olshausen and Field , 1997; Carlson et al . , 2012; Zhao and Zhaoping , 2011; Srinivasan et al . , 1982; Brito and Gerstner , 2016; Olshausen , 2003; Attneave , 1954 ) , sparseness ( Olshausen and Field , 1996 , Olshausen and Field , 1997; Carlson et al . , 2012; Kozlov and Gentner , 2016; Brito and Gerstner , 2016; Olshausen , 2003 ) , and slowness ( Hyvärinen et al . , 2003; Carlin and Elhilali , 2013 ) . Efficient coding indicates that neurons should encode maximal information about sensory input given certain constraints , such as spike count or energy costs . Sparseness posits that only a small proportion of neurons in the population should be active for a given input . Finally , slowness means that neurons should be sensitive to features that change slowly over time . The temporal prediction principle we describe here provides another unsupervised objective of sensory coding . It has been described in a very general manner by the information bottleneck concept ( Bialek et al . , 2001; Salisbury and Palmer , 2016; Palmer et al . , 2015 ) . We have instantiated a specific version of this idea , with linear-nonlinear encoding of the input , followed by a linear transform from the encoding units’ output to the prediction . In the following discussion , we describe previous normative models that infer RFs with temporal structure from auditory or movie input and relate them to spectrotemporal RFs in A1 or simple cell spatiotemporal RFs in V1 , respectively . For focus , other normative models of less directly relevant areas , such as spatial receptive fields without a temporal component ( Olshausen and Field , 1996 , Olshausen and Field , 1997 ) , complex cells ( Berkes and Wiskott , 2005 ) , retinal receptive fields ( Huang and Rao , 2011; Srinivasan et al . , 1982 ) , or auditory nerve impulse responses ( Smith and Lewicki , 2006 ) , will not be examined . A number of coding objectives have been explored in normative models of A1 spectrotemporal RFs . One approach ( Zhao and Zhaoping , 2011 ) found analytically that the optimal typical spectrotemporal RF for efficient coding was spectrally localized with lagging and flanking inhibition , and showed an asymmetric temporal envelope . However , the resulting RF also showed substantially more flanking inhibition , more ringing over time and frequency , and operated over a much shorter timescale ( ~10 ms ) than seen in A1 RFs ( Figure 3 ) . Moreover , this approach produced a single generic RF , rather than capturing the diversity of the population . Other models have produced a diverse range of spectrotemporal RFs . In the sparse coding approach ( Carlson et al . , 2012; Brito and Gerstner , 2016; Młynarski and McDermott , 2017; Blättler et al . , 2011 ) , a spectrogram snippet is reconstructed from a sum of basis functions ( a linear generative model ) , each weighted by its unit’s activity , with a constraint to have few active units . This approach is the same as the sparse coding model we used as a control ( Figure 5 ) . A challenge with many sparse generative models is that the activity of the units is found by a recurrent iterative process that needs to find a steady state; this is fine for static stimuli such as images , but for dynamic stimuli like sounds it is questionable whether the nervous system would have sufficient time to settle on appropriate activities before the stimulus had changed . Related work also used a sparsity objective , but rather than minimizing stimulus reconstruction error , forced high dispersal ( Kozlov and Gentner , 2016 ) or decorrelation ( Klein et al . , 2003; Carlin and Elhilali , 2013 ) of neural responses . Although lacking some of the useful probabilistic interpretations of sparse generative models , this approach does not require a settling process for inference . An alternative to sparseness is temporal slowness , which can be measured by temporal coherence ( Carlin and Elhilali , 2013 ) . Here the linear transform from sequential spectrogram snippets to unit activity is optimized to maximize the correlation of each unit’s response over a certain time window , while maintaining decorrelation between the units’ activities . Although the frequency tuning derived with these models can resemble that found in the midbrain or cortex ( Klein et al . , 2003; Carlson et al . , 2012; Kozlov and Gentner , 2016; Carlin and Elhilali , 2013; Brito and Gerstner , 2016; Młynarski and McDermott , 2017; Blättler et al . , 2011 ) ( Figure 5 ) , the resulting RFs lack the distinct asymmetric temporal profile and lagging inhibition seen in real midbrain or A1 RFs . Furthermore , they often have envelopes that are too elongated over time , often spanning the full temporal width of the spectrotemporal RF . This is related to the fact that the time window to be encoded by the model is set arbitrarily , and every time point within that window is given equal importance , that is , the direction of time is not accounted for . This is in contrast to the temporal prediction model , which naturally gives greater weighting to time-points near the present than to those in the past due to their greater predictive capacity . The earliest normative model of spatiotemporal RFs of simple cells used independent component analysis ( ICA ) ( van Hateren and Ruderman , 1998a ) , which is practically equivalent for visual or auditory data to the critically complete case of the sparse coding model ( Olshausen and Field , 1996 , Olshausen and Field , 1997 ) we used as a control ( Figure 5—figure supplements 1–2 and Figure 7—figure supplement 1 ) . The RFs produced by this model and the control model reproduced fairly well the spatial aspects of simple cell RFs . However , in contrast to the temporal prediction model ( Figure 7d ) , the subset of more ‘blob-like’ RFs seen in the data are not well captured by our control sparse coding model ( Figure 7—figure supplement 1g ) . In the temporal domain , again unlike the temporal prediction model and real V1 simple cells , the RFs of the ICA and sparse coding models are not pressed up against the present with an asymmetrical temporal envelope , but instead show a symmetrical envelope or span the entire range of times examined . A related model ( Olshausen , 2003 ) assumes that a longer sequence of frames is generated by convolving each basis function with a time-varying sparse coefficient and summing the result , so that each basis function is applied at each point in time . The resulting spatiotemporal RFs are similar to those produced by ICA ( van Hateren and Ruderman , 1998a ) , or our control model ( Figure 5—figure supplement 2 and Figure 7—figure supplement 1c ) . Although they tend not to span the entire range of times examined , they do show a symmetrical envelope , and require an iterative inference procedure , as described above for audition . Temporal slowness constraints have also been used to model the spatiotemporal RFs of simple cells . The bubbles ( Hyvärinen et al . , 2003 ) approach combines sparse and temporal coherence constraints with reconstruction . The resulting RFs show similar spatial and temporal properties to those found using ICA . A related framework is slow feature analysis ( SFA ) ( Berkes and Wiskott , 2005; Wiskott and Sejnowski , 2002 ) , which enforces temporal smoothness by minimizing the derivative of unit responses over time , while maximizing decorrelation between units . SFA has been used to model complex cell spatiotemporal RFs ( over only two time steps , Berkes and Wiskott , 2005 ) , and a modified version has been used to model spatial ( not spatiotemporal ) RFs of simple cells ( Berkes et al . , 2009 ) . These results are not directly comparable with our results or the spatiotemporal RFs of simple cells . In the slowness framework , the features found are those that persist over time; the presence of such a feature in the recent past predicts that the same feature will be present in the near future . This is also the case for our predictive approach , which , additionally , can capture features in the past that predict features in the future that are subtly or radically different from themselves . The temporal prediction principle will also give different weighting to features , as it values predictive capacity rather than temporal slowness ( Creutzig and Sprekeler , 2008 ) . In addition , although slowness models can be extended to model RFs over more than one time step ( Berkes and Wiskott , 2005; Hyvärinen et al . , 2003; Carlin and Elhilali , 2013 ) , capturing temporal structure , they do not inherently give more weighting to information in the most recent past and therefore do not give rise to asymmetric temporal profiles in RFs . There is one study that has directly examined temporal prediction as an objective for visual RFs in a manner similar to ours ( Palm , 2012 ) . Here , as in our model , a single hidden layer feedforward neural network was used to predict the immediate future frame of a movie patch from its past frames . However , only two frames of the past were used in this study , so a detailed exploration of the temporal profile of the spatiotemporal RFs was not possible . Nevertheless , some similarities and differences in the spatial RFs between the two frames were noted , and some units had oriented RFs . In contrast to our model , however , many RFs were noisy and did not resemble those of simple cells . Potential reasons for this difference include the use of L2 rather than L1 regularization on the weights , an output nonlinearity not present in our model , the optimization algorithm used , network size , or the dataset . Another very recent related study ( Chalk et al . , 2018 ) also implemented a somewhat different form of temporal prediction , with a linear ( rather than linear-nonlinear ) encoder , and linear decoder . When applied to visual scenes , oriented receptive fields were produced , but they were spatio-temporally separable and hence not direction selective . Temporal prediction has several strengths as an objective function for sensory processing . First , it can capture underlying features in the world ( Bialek et al . , 2001 ) ; this is also the case with sparseness ( Olshausen and Field , 1996 , Olshausen and Field , 1997 ) and slowness ( Wiskott and Sejnowski , 2002 ) , but temporal prediction will prioritize different features . Second , it can predict future inputs , which is very important for guiding action , especially given internal processing delays . Third , objectives such as efficient or sparse reconstruction retain everything about the stimulus , whereas an important part of neural information processing is the selective elimination of irrelevant information ( Marzen and DeDeo , 2017 ) . Prediction provides a good initial criterion for eliminating potentially unwanted information . Fourth , prediction provides a natural method to determine the hyperparameters of the model ( such as regularization strength , number of hidden units , activation function and temporal window size ) . Other models select their hyperparameters depending on what best reproduces the neural data , whereas we have an independent criterion – the capacity of the network to predict the future . One notable hyperparameter is how many time-steps of past input to encode . As described above , this is naturally decided by our model because only time-steps that help predict the future have significant weighting . Fifth , the temporal prediction model computes neuronal activity without needing to settle to a steady state , unlike some other models ( Olshausen and Field , 1996 , Olshausen and Field , 1997; Carlson et al . , 2012; Brito and Gerstner , 2016; Młynarski and McDermott , 2017 ) . For dynamic stimuli , a model that requires settling may not reach equilibrium in time to be useful . Sixth , and most importantly , temporal prediction successfully models many aspects of the RFs of primary cortical neurons . In addition to accounting for spatial and spectral tuning in V1 and A1 , respectively , at least as well as other normative models , it reproduces the temporal properties of RFs , particularly the asymmetry of the envelopes of RFs , something few previous models have attempted to explain . Although the temporal prediction model’s ability to describe neuronal RFs is high , the match with real neurons is not perfect . For example , the span of frequency tuning of our modelled auditory RFs is narrower than in A1 ( Figure 6g–h ) . We also found an overrepresentation of vertical and horizontal orientations compared to real V1 data ( Figure 7b–c ) . Some of these differences could be a consequence of the data used to train the model . Although the statistics of natural stimuli are broadly conserved ( Field , 1987 ) , there is still variation ( Torralba and Oliva , 2003 ) , and the dataset used to train the network may not match the sensory world of the animal experienced during development and over the course of evolution . In future work , it would be valuable to explore the influence of natural datasets with different statistics , and also to match those datasets more precisely to the evolutionary context and individual experience of the animals examined . Furthermore , a comparison of the model with neural data from different species , at different ages , and reared in different environments would be useful . Another cause of differences between the model and neural RFs may be the recording location of the RFs and how they are characterized . We used the primary sensory cortices as regions for comparison , because we performed transformations on the input data that are similar to the preprocessing that takes place in afferent subcortical structures . We spatially filtered the visual data in a similar way to the retina ( Olshausen and Field , 1996 , Olshausen and Field , 1997 ) , and spectrally decomposed the auditory data as in the inner ear , and then used time bins ( 5 ms ) which are coarser than , but close to , the maximum amplitude modulation period that can be tracked by auditory midbrain neurons ( Rees and Møller , 1983 ) . However , primary cortex is not a homogenous structure , with neurons in different layers displaying certain differences in their response properties ( Harris and Mrsic-Flogel , 2013 ) . Furthermore , the methods by which neurons are sampled from the cortex may not provide a representative sample . For example , multi-electrode arrays tend to favour larger and more active neurons . In addition , the method and stimuli used to construct RFs from the data can bias their structure somewhat ( Willmore et al . , 2016 ) . The model presented here is based on a simple feedforward network with one layer of hidden units . This limits its ability to predict features of the future input , and to account for RFs with nonlinear tuning . More complex networks , with additional layers or recurrency may allow the model to account for more complex tuning properties , including those found beyond the primary sensory cortices . Careful , principled adjustment of the preprocessing , or different regularization methods ( such as sparseness or slowness applied to the units’ activities ) , may also help . There is an open question as to whether the current model may eliminate some information that is useful for reconstruction of the past input or for prediction of higher order statistical properties of the future input , which might bring it into conflict with the principle of least commitment ( Marr , 1976 ) . It is an empirical question how much organisms preserve information that is not predictive of the future , although there are theoretical arguments against such preservation ( Bialek et al . , 2001 ) . Such conflict might be remedied , and the model improved , by adding feedback from higher areas or by adding an objective to reconstruct the past or present ( Barlow , 1959; Olshausen and Field , 1996 , Olshausen and Field , 1997; Attneave , 1954 ) in addition to predicting the future . To determine whether the model could help explain neuronal responses in higher areas , it would be useful to develop a hierarchical version of the temporal prediction model , applying the same model again to the activity of the hidden units rather than to the input . Another useful extension would be to see if the features learnt by the temporal prediction model could be used to accelerate learning of useful tasks such as speech or object recognition , by providing input or initialization for a supervised or reinforcement learning network . Indeed , temporal predictive principles have been shown to be useful for unsupervised training of networks used in visual object recognition ( Srivastava et al . , 2015; Ranzato , 2016; Lotter et al . , 2016; Oh et al . , 2015 ) . Finally , it is interesting to consider possible more explicit biological bases for our model . We envisage the input units of the model as thalamic input , and the hidden units as primary cortical neurons . Although the function of the output units could be seen as just a method to optimize the hidden units to find the most predictive code given sensory input statistics , they may also have a physiological analogue . Current evidence ( Dahmen and King , 2007; Huberman et al . , 2008; Kiorpes , 2015 ) suggests that while primary cortical RFs are to an extent hard-wired in form by natural selection , their tuning is also refined by individual sensory experience . This refinement process may require a predictive learning mechanism in the animal’s brain , at least at some stage of development and perhaps also into adulthood . Hence , one might expect to find a subpopulation of neurons that represent the prediction ( analogous to the output units of the model ) or the prediction error ( analogous to the difference between the output unit activity and the target ) . Indeed , signals relating to sensory prediction error have been found in A1 ( Rubin et al . , 2016 ) , though they may also be located in other regions of the brain . Finally , it is important to note that , although the biological plausibility of backpropagation has long been questioned , recent progress has been made in developing trainable networks that perform similarly to artificial neural networks trained with backpropagation , but with more biologically plausible characteristics ( Bengio et al . , 2015 ) , for example , by having spikes or avoiding the weight transport problem ( Lillicrap et al . , 2016 ) . We have shown that a simple principle - predicting the imminent future of a sensory scene from its recent past - explains many features of the RFs of neurons in both primary visual and auditory cortex . This principle may also account for neural tuning in other sensory systems , and may prove useful for the study of higher sensory processing and aspects of neural development and learning . While the importance of temporal prediction is increasingly widely recognized , it is perhaps surprising nonetheless that many basic tuning properties of sensory neurons , which we have known about for decades , appear , in fact , to be a direct consequence of the brain’s need to efficiently predict what will happen next .
The sparse coding model was used as a control for both visual and auditory cases . The Python implementation of this model ( https://github . com/zayd/sparsenet ) was trained using the same visual and auditory inputs used to train the predictive model . The training data were divided into mini-batches which were shuffled and the model optimized for one full pass through the data . Inference was performed using the Fast Iterative Shrinkage and Thresholding ( FISTA ) algorithm . A sparse L1 prior with strength λ was applied to the unit activities , providing activity regularization . A range of λ-values and unit numbers were tried ( Figure 8—figure supplement 1 ) . The learning rate and batch size were also varied until reasonable values were found . As there was no independent criterion by which to determine the ‘best’ settings , we chose the network that produced basis functions whose receptive fields were most similar to those of real neurons . In the auditory case , this was determined using the mean KS measure of similarity ( Figure 8—figure supplement 1 ) . In the visual case , as a similarity measure was not performed , this was done by inspection . In both cases , the model configurations chosen were restricted to those trained in an overcomplete condition ( having more units than the number of input variables ) in order to remain consistent with previous instantiations of this model ( Olshausen and Field , 1996; Olshausen and Field , 1997; Carlson et al . , 2012 ) . In this manner , we selected a sparse coding network with 1600 units , λ = 100 . 5 , learning rate = 0 . 01 and 100 mini-batches in the auditory case ( Figures 5–6 ) . In the visual case , the network selected was trained with 3200 units , λ = 100 . 5 , learning rate = 0 . 05 and 100 mini-batches ( Figure 5—figure supplements 1–2 and Figure 7—figure supplement 1 ) . Although the sparse coding basis functions are projective fields , they tend to be similar in structure to receptive fields ( Olshausen and Field , 1996; Olshausen and Field , 1997 ) , and , for simplicity , are referred to as RFs . All custom code used in this study was implemented in MATLAB and Python . We have uploaded the code to a public Github repository ( Singer , 2018; copy archived at https://github . com/elifesciences-publications/temporal_prediction_model ) . The raw auditory experimental data is available at https://osf . io/ayw2p/ . The movies and sounds used for training the models are all publicly available at the websites detailed in the Materials and methods . | A large part of our brain is devoted to processing the sensory inputs that we receive from the world . This allows us to tell , for example , whether we are looking at a cat or a dog , and if we are hearing a bark or a meow . Neurons in the sensory cortex respond to these stimuli by generating spikes of activity . Within each sensory area , neurons respond best to stimuli with precise properties: those in the primary visual cortex prefer edge-like structures that move in a certain direction at a given speed , while neurons in the primary auditory cortex favour sounds that change in loudness over a particular range of frequencies . Singer et al . sought to understand why neurons respond to the particular features of stimuli that they do . Why do visual neurons react more to moving edges than to , say , rotating hexagons ? And why do auditory neurons respond more to certain changing sounds than to , say , constant tones ? One leading idea is that the brain tries to use as few spikes as possible to represent real-world stimuli . Known as sparse coding , this principle can account for much of the behaviour of sensory neurons . Another possibility is that sensory areas respond the way they do because it enables them to best predict future sensory input . To test this idea , Singer et al . used a computer to simulate a network of neurons and trained this network to predict the next few frames of video clips using the previous few frames . When the network had learned this task , Singer et al . examined the neurons’ preferred stimuli . Like neurons in primary visual cortex , the simulated neurons typically responded most to edges that moved over time . The same network was also trained in a similar way , but this time using sound . As for neurons in primary auditory cortex , the simulated neurons preferred sounds that changed in loudness at particular frequencies . Notably , for both vision and audition , the simulated neurons favoured recent inputs over those further into the past . In this way and others , they were more similar to real neurons than simulated neurons that used sparse coding . Both artificial networks trained to foretell sensory input and the brain therefore favour the same types of stimuli: the ones that are good at helping to grasp future information . This suggests that the brain represents the sensory world so as to be able to best predict the future . Knowing how the brain handles information from our senses may help to understand disorders associated with sensory processing , such as dyslexia and tinnitus . It may also inspire approaches for training machines to process sensory inputs , improving artificial intelligence . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"neuroscience"
] | 2018 | Sensory cortex is optimized for prediction of future input |
Concentrative nucleoside transporters ( CNTs ) are responsible for cellular entry of nucleosides , which serve as precursors to nucleic acids and act as signaling molecules . CNTs also play a crucial role in the uptake of nucleoside-derived drugs , including anticancer and antiviral agents . Understanding how CNTs recognize and import their substrates could not only lead to a better understanding of nucleoside-related biological processes but also the design of nucleoside-derived drugs that can better reach their targets . Here , we present a combination of X-ray crystallographic and equilibrium-binding studies probing the molecular origins of nucleoside and nucleoside drug selectivity of a CNT from Vibrio cholerae . We then used this information in chemically modifying an anticancer drug so that it is better transported by and selective for a single human CNT subtype . This work provides proof of principle for utilizing transporter structural and functional information for the design of compounds that enter cells more efficiently and selectively .
Nucleosides play critical roles in biology as precursors to nucleic acids and the energy currency of the cell and also serve as signaling molecules ( King et al . , 2006; Rose and Coe , 2008; Molina-Arcas et al . , 2009 ) . Furthermore , nucleoside analogs have clinical applications as anticancer and antiviral drugs ( Damaraju et al . , 2003; Jordheim and Dumontet , 2007 ) . Because of their immense biological and clinical importance , efficient entry of nucleosides and their analogs into the cell is crucial to human health and disease . Cellular entry is accomplished by a class of membrane proteins known as nucleoside transporters ( NTs ) . There are two types of NTs in humans: concentrative nucleoside transporters ( CNTs ) and equilibrative nucleoside transporters ( ENTs ) . CNTs utilize the energy of ion gradients to actively transport nucleosides into the cell against their concentration gradients while ENTs transport nucleosides down their chemical gradients without the requirement of any additional energy source ( Gray et al . , 2004b ) . In addition to nucleosides , NTs are responsible for the transport of a wide range of nucleoside-derived anticancer ( e . g . , gemcitabine and 5-fluorouridine ) and antiviral ( e . g . , ribavirin ) drugs ( Farre et al . , 2004; Marechal et al . , 2009; Rabascio et al . , 2010; Bhutia et al . , 2011; Doehring et al . , 2011; Fukao et al . , 2011; Rau et al . , 2013 ) . Both NT families possess subtype-dependent nucleoside specificities and tissue distributions , while CNTs are more highly subtype-specific for their substrates and distributions than ENTs ( Gray et al . , 2004b; Paproski et al . , 2013 ) . As a result , different NT subtypes are responsible for the transport of different types of nucleosides and nucleoside drugs , and expression levels of different NTs can predict how patients with certain types of cancer and viral infection will respond to nucleoside-drug treatment ( Mackey et al . , 1998; Farre et al . , 2004; Spratlin et al . , 2004; Gray et al . , 2004a; Damaraju et al . , 2009; Marechal et al . , 2009; Rabascio et al . , 2010; Bhutia et al . , 2011; Doehring et al . , 2011; Fukao et al . , 2011; Rau et al . , 2013 ) . Since there are three different isoforms of human CNTs ( hCNT1-3 ) that possess differing nucleoside and nucleoside-drug specificities and tissue distributions , greater knowledge of the molecular origins of nucleoside selectivity by CNTs could potentially lead to better-tailored nucleoside drug delivery as well as a better understanding of CNT-mediated physiological processes . CNTs belong to the solute carrier ( SLC ) superfamily , constituting the family SLC28 . The SLC superfamily , composed of 52 families , is responsible for the transport of ions , metabolites , neurotransmitters , and drugs in humans . Several SLC families are of particular clinical interest because of their roles in drug absorption , distribution , metabolism , and excretion ( ADME ) ( Schlessinger et al . , 2013 ) . The recent determination of the structures of several SLC transporters has advanced our understanding of the inner workings of these transporters and expanded the applicability of structure-based ligand discovery using computational methods ( Gao et al . , 2009; He et al . , 2010; Hu et al . , 2011; Newstead et al . , 2011; Johnson et al . , 2012; Pedersen et al . , 2013 ) . Although this computational method of ligand discovery is a valuable approach , it cannot accurately predict the energetically important interactions between ligands and transporters , and therefore experimental approaches should be pursued to understand the principles of ligand and drug selectivity by these transporters . The crystal structure of a CNT from Vibrio cholerae ( vcCNT ) presents the first opportunity to examine specific nucleoside recognition by CNTs from a structural perspective ( Johnson et al . , 2012 ) . vcCNT is an excellent model system to study hCNTs: it utilizes a Na+ gradient for nucleoside transport like hCNTs and shares high sequence identity ( 36–39% ) with hCNTs with particularly high sequence identity for the nucleoside-binding site ( 64% with hCNT1 , 73% with hCNT2 , 91% with hCNT3 ) . For these reasons , vcCNT has been identified as an optimal candidate for structure-based ligand discovery using computational methods ( Schlessinger et al . , 2013 ) . Here , we have exploited a combination of X-ray crystallographic studies and equilibrium-binding measurements of vcCNT to understand the structural basis of CNT selectivity . We have discovered that CNTs use a unique mode of nucleoside recognition that is suitable for its function as a transporter . Using the insights gained from these studies , we have chemically modified the anticancer drug gemcitabine and found that its binding affinity for vcCNT is greatly enhanced . Furthermore , the modified compound now possesses subtype-specific transport among human CNTs . Follow-up structural and mutational studies revealed the origin of subtype-specificity of the modified compound . Not only do our studies illuminate the structural basis of nucleoside selectivity by CNTs but they also provide proof of principle for utilizing membrane transporter structures for the design of drugs with more selective delivery ( Han and Amidon , 2000; Majumdar et al . , 2004 ) .
vcCNT forms a homotrimer with each protomer possessing its own nucleoside-binding site and permeation pathway ( Figure 1A ) . The protomer adopts a new fold and is divided into two domains: the scaffold domain that is responsible for trimerization and maintaining the overall architecture of the transporter ( light blue , Figure 1B ) , and the transport domain where nucleoside binding and transport occur ( other colors , Figure 1B ) . The nucleoside-binding site , facing the trimer axis , is formed at the center of the transport domain between the tips of two helical hairpins ( HP1 and HP2 ) and two partially unwound transmembrane helices ( TM4 and TM7 ) ( Johnson et al . , 2012 ) . 10 . 7554/eLife . 03604 . 003Figure 1 . The nucleoside-binding site of vcCNT and fluorescence-anisotropy-based competition assay . ( A ) The vcCNT-7C8C trimer viewed from within the plane of the membrane . The location of the membrane is marked by rectangles . The scaffold domain of one protomer is colored light blue , and the transport domain is colored red , blue , orange , cyan , wheat , and brown . The other two protomers are colored gray . Uridine is shown bound to each protomer in stick representation . The nucleoside-binding site is delineated with dashed lines . vcCNT-7C8C functions similarly to wild type ( Figure 1—figure supplement 1 ) . ( B ) The vcCNT-7C8C protomer . The structure is rotated 120° about the trimer axis relative to A , zoomed in , and the other two protomers have been removed for clarity . ( C ) Nucleoside-binding site . Amino acid residues that interact with the uridine are labeled and shown in stick representation and were used for sequence identity calculation with hCNTs . Hydrogen bonds are shown as dashed lines . The uracil base is marked with a blue box , and the ribose is marked with a gray box . For a stereo view of the electron density in the nucleoside-binding site , see Figure 1—figure supplement 2 . ( D ) Fluorescence titration of vcCNT with uridine . Uridine was titrated into solution containing vcCNT and the fluorescent nucleoside pyrrolo-cytidine , anisotropy was measured , and data were fit to a single-site competitive binding model to obtain a KD of 36 ± 3 μM ( mean ± SEM , n = 3 measurements ) . DOI: http://dx . doi . org/10 . 7554/eLife . 03604 . 00310 . 7554/eLife . 03604 . 004Figure 1—figure supplement 1 . vcCNT-7C8C maintains nucleoside transport activity . Wild-type ( WT ) and 7C8C vcCNT were reconstituted into lipid vesicles and assessed for their ability to support the uptake of 2 μM of 3H-labeled nucleosides as described ( Johnson et al . , 2012 ) . The data shown represent 2 min timepoints . Empty vesicles were included as a negative control . DOI: http://dx . doi . org/10 . 7554/eLife . 03604 . 00410 . 7554/eLife . 03604 . 005Figure 1—figure supplement 2 . Electron density at the nucleoside-binding site of vcCNT-7C8C-uridine . The nucleoside-binding site of vcCNT-7C8C bound to uridine is shown in stereo in stick representation in the same orientation and colored as in Figure 1C . Uridine is yellow and the red spheres are water molecules . The resolution is 2 . 1 Å and density shown is from a 2Fo–Fc electron density map contoured at 1σ . DOI: http://dx . doi . org/10 . 7554/eLife . 03604 . 005 We solved the structure of a uridine-bound double mutant of vcCNT ( Leu 7 to Cys and Ile 8 to Cys , termed 7C8C ) at 2 . 1 Å , which is higher resolution than the 2 . 4-Å wild-type structure . This mutant was originally designed to introduce binding sites for hydrophobic mercury compounds for heavy-atom phasing . The high-resolution mutant structure revealed another water in the binding site that bridges the 4-carbonyl of the uracil base with Glu 156 but was otherwise identical to the wild-type structure ( Figure 1C , Figure 1—figure supplement 2; Table 1 ) . The mutations do not affect transporter function significantly ( Figure 1—figure supplement 1 ) . 10 . 7554/eLife . 03604 . 006Table 1 . Data collection and refinement statisticsDOI: http://dx . doi . org/10 . 7554/eLife . 03604 . 006vcCNT-7C8C-uridinevcCNT-zebularinevcCNT-7C8C-adenosinevcCNT-7C8C-ribavirinData collection Space groupP63P63P63P63 Cell dimensions a , b , c ( Å ) 119 . 7 , 119 . 7 , 83 . 1119 . 8 , 119 . 8 , 82 . 7120 . 0 , 120 . 0 , 83 . 5119 . 7 , 119 . 7 , 83 . 6 α , β , γ ( ° ) 90 , 90 , 12090 , 90 , 12090 , 90 , 12090 , 90 , 120 Resolution ( Å ) 2 . 08 ( 2 . 12–2 . 08 ) *2 . 90 ( 2 . 95–2 . 90 ) 3 . 10 ( 3 . 15–3 . 10 ) 2 . 80 ( 2 . 85–2 . 80 ) Rsym or Rmerge0 . 052 ( 0 . 554 ) 0 . 141 ( 0 . 766 ) 0 . 104 ( 0 . 567 ) 0 . 114 ( 0 . 758 ) I/σI42 . 0 ( 2 . 3 ) 13 . 5 ( 1 . 9 ) 14 . 1 ( 1 . 3 ) 20 . 2 ( 1 . 8 ) Completeness ( % ) 99 . 3 ( 93 . 0 ) 99 . 9 ( 100 . 0 ) 99 . 0 ( 96 . 5 ) 99 . 8 ( 100 . 0 ) Redundancy5 . 9 ( 4 . 3 ) 5 . 8 ( 5 . 1 ) 4 . 2 ( 3 . 4 ) 7 . 0 ( 6 . 3 ) Refinement Resolution ( Å ) 2 . 08 ( 2 . 13–2 . 08 ) 2 . 91 ( 3 . 13–2 . 91 ) 3 . 10 ( 3 . 41–3 . 10 ) 2 . 80 ( 2 . 98–2 . 80 ) No . reflections40368 ( 2381 ) 14932 ( 2798 ) 12429 ( 2887 ) 16840 ( 2635 ) Rwork/Rfree ( % ) 20 . 2/23 . 321 . 1/24 . 722 . 4/26 . 922 . 2/25 . 9 No . atoms Protein2834286829252947 Ligand/ion17/116/119/117/1 Water/detergent130/333/333/332/33 B-factors Protein44 . 145 . 672 . 158 . 7 Ligand/ion34 . 2/31 . 325 . 6/47 . 661 . 9/71 . 054 . 2/64 . 5 Water/detergent54 . 4/62 . 629 . 3/56 . 454 . 6/89 . 048 . 1/72 . 3 R . m . s deviations Bond lengths ( Å ) 0 . 0050 . 0030 . 0020 . 003 Bond angles ( ° ) 0 . 7820 . 6760 . 6170 . 709vcCNT-7C8C-5-fluorouridinevcCNT-7C8C cytidinevcCNT-7C8C-pyrrolo-cytidinevcCNT-7C8C-gemcitabineData collection Space groupP63P63P63P63 Cell dimensions a , b , c ( Å ) 119 . 8 , 119 . 8 , 83 . 2120 . 0 , 120 . 0 , 82 . 5119 . 6 , 119 . 6 , 83 . 1119 . 0 , 119 . 0 , 82 . 3 α , β , γ ( ° ) 90 , 90 , 12090 , 90 , 12090 , 90 , 12090 , 90 , 120 Resolution ( Å ) 2 . 30 ( 2 . 34–2 . 30 ) 2 . 60 ( 2 . 64–2 . 60 ) 2 . 75 ( 2 . 80–2 . 75 ) 2 . 90 ( 2 . 97–2 . 90 ) Rsym or Rmerge0 . 067 ( 0 . 500 ) 0 . 094 ( 0 . 665 ) 0 . 083 ( 0 . 656 ) 0 . 061 ( 0 . 554 ) I/σI28 . 0 ( 1 . 6 ) 24 . 0 ( 2 . 0 ) 22 . 6 ( 1 . 9 ) 30 . 8 ( 1 . 9 ) Completeness ( % ) 97 . 3 ( 79 . 2 ) 99 . 3 ( 98 . 7 ) 99 . 9 ( 100 . 0 ) 99 . 6 ( 99 . 0 ) Redundancy4 . 0 ( 2 . 4 ) 6 . 4 ( 6 . 0 ) 5 . 8 ( 5 . 5 ) 5 . 7 ( 4 . 6 ) Refinement Resolution ( Å ) 2 . 30 ( 2 . 38–2 . 30 ) 2 . 61 ( 2 . 75–2 . 61 ) 2 . 75 ( 2 . 92–2 . 75 ) 2 . 91 ( 3 . 13–2 . 91 ) No . reflections29380 ( 2089 ) 20600 ( 2754 ) 17698 ( 2784 ) 14682 ( 2751 ) Rwork/ Rfree ( % ) 20 . 0/21 . 621 . 4/23 . 920 . 2/24 . 722 . 7/25 . 8 No . atoms Protein2937289829212883 Ligand/ion18/117/120/118/1 Water/detergent58/3328/333/339/33 B-factors Protein54 . 459 . 160 . 976 . 7 Ligand/ion41 . 4/47 . 246 . 1/48 . 648 . 4/54 . 159 . 6/70 . 1 Water/detergent59 . 1/67 . 960 . 3/62 . 647 . 9/72 . 758 . 8/101 . 4 R . m . s deviations Bond lengths ( Å ) 0 . 0020 . 0020 . 0030 . 004 Bond angles ( ° ) 0 . 6600 . 6090 . 6460 . 664*Highest resolution shell is shown in parenthesis . The structure of vcCNT bound to uridine revealed that the interactions can be divided into two groups: those that involve the ribose moiety and those with the nitrogenous base ( Figure 1C ) . To determine the energetic contributions of each of the interactions between nucleoside and vcCNT , we developed a fluorescence-anisotropy-based competition assay for measuring the equilibrium dissociation constants ( KDs ) for a variety of nucleosides and nucleoside analogs using the fluorescent cytidine analog pyrrolo-cytidine ( Table 2; Damaraju et al . , 2011 ) . We calculated the KD for uridine to be 36 μM ( Figure 1D ) , which is similar to the reported Km values of uridine for hCNTs ( Km = 22–80 μM ) , further suggesting that vcCNT is a good model system to study hCNTs ( Molina-Arcas et al . , 2009 ) . 10 . 7554/eLife . 03604 . 007Table 2 . KD values for nucleosides and nucleoside analog drugs calculated from fluorescence titrationsDOI: http://dx . doi . org/10 . 7554/eLife . 03604 . 00710 . 7554/eLife . 03604 . 008Table 2—source data 1 . Fluorescence data for KD calculations . For pyrrolo-cytidine and pyrrolo-gemcitabine , individual solutions with fixed nucleoside concentrations and increasing concentrations of vcCNT were prepared , the fluorescence anisotropy was measured , and the data were fit globally to a single-site binding model accounting for ligand depletion . For all other nucleosides , the nucleoside of interest was titrated into solution containing vcCNT and pyrrolo-cytidine 5 μl at a time , the fluorescence anisotropy was measured , and the data were fit globally to a single-site competitive binding model accounting for ligand depletion . All experiments were performed at least three times . DOI: http://dx . doi . org/10 . 7554/eLife . 03604 . 008CompoundKD ( μM ) *uridine36 ± 3cytidine61 ± 5adenosine470 ± 100gemcitabine1 , 370 ± 430ribavirin1 , 530 ± 350zebularine120 ± 53-methyluridine520 ± 805-fluorouridine16 ± 15-chlorouridine14 ± 15-iodouridine58 ± 55-methyluridine61 ± 72′-deoxyuridine170 ± 103′-deoxyuridine>2 , 8005'-deoxyuridine>2 , 800cytarabine>3 , 000pyrrolo-cytidine0 . 94 ± 0 . 17pyrrolo-gemcitabine23 . 5 ± 0 . 2*Titrations were performed in triplicate and data were fit globally . All values are given as means ± SEM . See Table 2—source data 1 for fluorescence data used in calculating KD values . Human CNTs have differing nucleoside-base preferences: hCNT1 mainly transports pyrimidines , hCNT2 prefers purines , and hCNT3 is broadly selective for both pyrimidines and purines ( Gray et al . , 2004b; Molina-Arcas et al . , 2009 ) . The uracil base interacts with residues on HP1 ( Gln 154 directly and Thr 155 and Glu 156 through water molecules ) and TM4 ( Val 188 via van der Waals interactions ) . To examine the energetics of these interactions , we measured KDs of vcCNT for uridine analogs with modifications to the uracil base as well as other nucleosides . The anticancer drug zebularine is a uridine analog with no substituent at the C4 position of the pyrimidine base . Zebularine exhibits a ∼threefold loss of binding affinity ( KD = 120 μM ) relative to uridine . To deduce the structural basis of the reduced binding affinity , we solved the crystal structure of vcCNT bound to zebularine ( Figure 2A , Figure 2—figure supplement 1 ) . The crystal structure shows that the side chain of Glu 156 adopts a different rotamer position probably because it is unable to form the water-mediated interaction with the C4-carbonyl of the uracil base , consistent with the loss of binding affinity . 10 . 7554/eLife . 03604 . 009Figure 2 . Structural basis of nucleobase recognition by vcCNT . ( A ) The crystal structure of vcCNT bound to zebularine . ( B ) Chemical structure of 3-methyluridine . ( C ) The crystal structure of vcCNT-7C8C bound to adenosine . ( D ) The crystal structure of vcCNT-7C8C bound to ribavirin . ( E ) The crystal structure of vcCNT-7C8C bound to 5-fluorouridine . Fluorine is colored cyan . All electron density maps represent Fo–Fc SA-OMIT maps for the nucleoside contoured at 3σ . Uridine is shown in the center of the figure for reference . For stereo views of the electron density in the nucleoside-binding site for each of these structures , see Figure 2—figure supplements 1–4 . DOI: http://dx . doi . org/10 . 7554/eLife . 03604 . 00910 . 7554/eLife . 03604 . 010Figure 2—figure supplement 1 . Electron density at the nucleoside-binding site of vcCNT-zebularine . The nucleoside-binding site of vcCNT bound to zebularine is shown in stereo in stick representation in the same orientation and colored as in Figure 2A . Zebularine is yellow and the red spheres are water molecules . The resolution is 2 . 9 Å and density shown is from a 2Fo–Fc electron density map contoured at 1σ . DOI: http://dx . doi . org/10 . 7554/eLife . 03604 . 01010 . 7554/eLife . 03604 . 011Figure 2—figure supplement 2 . Electron density at the nucleoside-binding site of vcCNT-7C8C-adenosine . The nucleoside-binding site of vcCNT-7C8C bound to adenosine is shown in stereo in stick representation in the same orientation and colored as in Figure 2C . Adenosine is yellow and the red sphere is a water molecule . The resolution is 3 . 1 Å and density shown is from a 2Fo–Fc electron density map contoured at 1σ . DOI: http://dx . doi . org/10 . 7554/eLife . 03604 . 01110 . 7554/eLife . 03604 . 012Figure 2—figure supplement 3 . Electron density at the nucleoside-binding site of vcCNT-7C8C-ribavirin . The nucleoside-binding site of vcCNT-7C8C bound to ribavirin is shown in stereo in stick representation in the same orientation and colored as in Figure 2D . Ribavirin is yellow and the red sphere is a water molecule . The resolution is 2 . 8 Å and density shown is from a 2Fo–Fc electron density map contoured at 1σ . DOI: http://dx . doi . org/10 . 7554/eLife . 03604 . 01210 . 7554/eLife . 03604 . 013Figure 2—figure supplement 4 . Electron density at the nucleoside-binding site of vcCNT-7C8C-5-fluorouridine . The nucleoside-binding site of vcCNT-7C8C bound to 5-fluorouridine is shown in stereo in stick representation in the same orientation and colored as in Figure 2E . 5-fluorouridine is yellow and the red spheres are water molecules . The resolution is 2 . 3 Å and density shown is from a 2Fo–Fc electron density map contoured at 1σ . DOI: http://dx . doi . org/10 . 7554/eLife . 03604 . 013 N3 of the uracil base interacts with both Thr 155 and Glu 156 through a single water molecule that is coordinated by both residues . We measured the affinity of vcCNT for 3-methyluridine , which contains a methyl group at this position that blocks the water-mediated interaction , and we found that it significantly decreased the binding affinity ( KD = 520 μM , Figure 2B , Figure 2—figure supplement 2 ) . The purine nucleoside adenosine and the antiviral guanosine analog ribavirin also possess similarly weaker binding affinity ( KD = 470 and 1530 μM , respectively ) . To examine the structural basis of the reduced affinity , we solved the crystal structures of vcCNT-7C8C bound to adenosine ( Figure 2C , Figure 2—figure supplement 3 ) and ribavirin ( Figure 2D , Figure 2—figure supplement 4 ) . The structures reveal that the bulky purine base displaces the water observed in the uridine structure while maintaining similar modes of ribose binding , corroborating the idea that the reduced affinities of adenosine and ribavirin are due to the loss of the water-mediated interactions and highlighting the importance of the water coordinated by Thr 155 and Glu 156 in nucleoside recognition by CNTs . To test the energetic contribution of substituents at the C5 position of the pyrimidine ring , we measured the KD for the anticancer drug 5-fluorouridine to be 16 μM , which is ∼twofold lower than the KD for uridine . We also solved the crystal structure of vcCNT-7C8C bound to 5-fluorouridine ( Figure 2E ) . The structure shows that the modified uracil base still fits into the nucleoside-binding site without any structural rearrangements of the protein . To understand the origin of enhanced affinity by the addition of fluorine at the C5 position , we compared the binding affinity of 5-fluorouridine to other C5-substituted uridine analogs with differing electronegativities and atomic radii ( Table 3 ) . Fluorine is highly electronegative ( 3 . 98 on the Pauling scale ) and relatively small ( 1 . 47 Å radius ) . When another highly electronegative but slightly larger halogen , chlorine ( 3 . 16 , 1 . 75 Å ) , is substituted at the C5 position , a similar KD is observed ( 14 μM ) . However , when the large , weakly electronegative substituent iodine ( 2 . 66 , 1 . 98 Å ) or a methyl group ( 2 . 55 , 2 . 00 Å ) is added , we observe an increase in KD ( ∼60 μM ) . In short , we observed an increase in affinity with smaller , highly electronegative substituents but a decrease in affinity with larger , less electronegative substituents . What structural feature could account for the differing affinities of these compounds ? 10 . 7554/eLife . 03604 . 014Table 3 . Properties of substituents of 5-substituted uridines and their binding affinities for vcCNTDOI: http://dx . doi . org/10 . 7554/eLife . 03604 . 014SubstituentRadius ( Å ) ElectronegativityKD ( μM ) fluorine1 . 473 . 9816 ± 1chlorine1 . 753 . 1614 ± 4iodine1 . 982 . 6658 ± 5methyl2 . 002 . 5561 ± 7 These differences may result from interactions of the nucleosides with Phe 366 . In the structure of vcCNT-7C8C bound to uridine , Phe 366 appears to interact with both the uracil base and the ribose of the nucleoside . Phe 366 forms an offset π–π interaction with the aromatic pyrimidine ring and also forms CH–π interactions with the ribose ( Figure 1C , Figure 3A , B ) . The addition of a small , highly electronegative substituent at the C5 position could strengthen the interaction between the pyrimidine ring and Phe 366 , as is the case with many π–π interactions ( Hunter and Sanders , 1990; Ringer et al . , 2006 ) . 10 . 7554/eLife . 03604 . 015Figure 3 . Phe 366 is crucial for nucleoside binding by vcCNT . ( A ) The nucleoside-binding site of vcCNT-7C8C bound to uridine is shown viewed from the cytoplasm . Phe 366 interacts with the uracil base via π–π interactions . The other epimeric 2′ position is marked with an arrow . ( B ) Another view of the interaction between Phe 366 and uridine . Phe 366 interacts with the ribose via CH–π interactions ( dashed lines ) . ( C–F ) Isothermal titration calorimetry of uridine binding to wild-type vcCNT and Phe 366 mutants . KD = 45 ± 8 μM and ΔHo = −2970 ± 330 cal/mol for WT , KD = 1630 ± 120 μM and ΔHo = −2200 ± 190 cal/mol for F366A , KD = 920 ± 170 μM and ΔHo = −1600 ± 440 cal/mol for F366Y , and KD = 1470 ± 90 μM and ΔHo = −3190 ± 130 cal/mol for F366W ( means ± SEM , n = 3 measurements ) . Note that the KD for F366A could not be reliably measured due to the low heat associated with binding . Each of the F366 mutants is biochemically stable as evidenced by a single , sharp peak at the expected trimer size when subjected to size-exclusion chromatography ( Figure 3—figure supplement 1 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 03604 . 01510 . 7554/eLife . 03604 . 016Figure 3—figure supplement 1 . F366 mutants are biochemically stable . 1 liter of each construct was expressed and purified in parallel without the addition of any exogenous nucleosides as described in the ‘Materials and methods’ . Each sample was initially purified by size-exclusion chromatography using a Superdex 200 10/300 GL column to remove MBP , and then the peak fractions were collected , concentrated , and re-ran to generate the figures . Each individual chromatogram is shown in the top panels with absolute absorbance at 280 nm on the y-axis , and all of the normalized chromatograms are shown overlaid in the bottom panel . DOI: http://dx . doi . org/10 . 7554/eLife . 03604 . 016 Notably , Phe 366 is universally conserved between members of the CNT family , and its functional importance has never been tested . To examine the role of Phe 366 in nucleoside recognition , we mutated this residue and performed isothermal titration calorimetry ( ITC ) experiments . We found that the requirement for Phe at this position is strict , as mutation of this residue to alanine or tyrosine or tryptophan results in significantly decreased binding affinity for uridine while not affecting the stability significantly ( Figure 3C–F , Figure 3—figure supplement 1 ) . Therefore , we suggest that Phe 366 plays a critical role in recognition of the nucleoside by CNTs . Amino acid residues that interact with the ribose in the vcCNT structure ( Glu 332 , Asn 368 , and Ser 371 ) are invariant between hCNTs and vcCNT . To probe these interactions , we measured KDs for deoxyuridines from which each of the ribose hydroxyls have been removed . Both 3′- and 5′-deoxyuridine yielded KD values greater than 2800 μM ( Figure 4A ) , while the binding affinity for 2′-deoxyuridine is not as drastically affected ( KD = 170 μM ) . Several anticancer nucleoside analog drugs contain modifications at the C2′ position of the ribose . Gemcitabine , for example , is a cytidine analog with two fluorine atoms bonded to C2′ . The measured KD of gemcitabine for vcCNT ( KD = 1370 μM ) is ∼22-fold higher than that for cytidine ( KD = 61 μM ) ( Figure 4B , Figure 4—figure supplement 1 ) . To understand the structural basis of this significant reduction in binding affinity associated with the fluorine substitutions , we solved the crystal structure of vcCNT-7C8C bound to gemcitabine ( Figure 4C , Figure 4—figure supplement 2 ) . In the uridine-bound structure , a CH–π interaction was observed between the C2′ hydrogen and Phe 366 ( Figure 3B ) . In order for vcCNT to accommodate for the bulkier fluorine atom on the other epimeric position of C2′ , which creates steric interference and electrostatic repulsion with the π electrons of Phe 366 , both Phe 366 and TM7b ( including Ser 371 ) move slightly away from the nucleoside with respect to the uridine-bound structure ( Figure 4D ) . Furthermore , the ribose of the gemcitabine is reoriented with respect to the other nucleoside-bound structures ( Figure 4E ) . In addition to the steric and electrostatic disruption of the ribose-binding site , fluorine is a poor substitute for a hydroxyl as a hydrogen-bond acceptor and thus provides a less favorable interaction of the ribose with Ser 371 ( Howard et al . , 1996; Dunitz and Taylor , 1997 ) . To further test the importance of the epimeric position of the 2′-hydroxyl group of the ribose , we attempted to measure the KD of another anticancer cytidine analog known as cytarabine ( cytosine arabinoside ) which has its 2′-hydroxyl flipped up above the ribose ring ( Figure 4B ) . Cytarabine displayed no measurable binding when titrated into vcCNT ( KD > 3000 μM ) . Interestingly , consistent with our observation with vcCNT , hCNTs show no significant binding and transport of cytarabine ( Clarke et al . , 2006 ) . Taken together , these results reveal that the interactions of both the nucleobase and the ribose with Phe 366 and the interactions of the ribose with TM7 are critical for nucleoside recognition by CNTs , which explains the intolerance of CNTs for substituents at the other epimeric position of C2′ of nucleosides . 10 . 7554/eLife . 03604 . 017Figure 4 . Structural basis of ribose recognition by vcCNT . ( A ) Dissociation constants for deoxyuridines . ( B ) Chemical structures and KDs of cytidine , gemcitabine , and cytarabine . The cytidine is from the crystal structure of vcCNT-7C8C bound to cytidine ( for a stereo view of the electron density in the nucleoside-binding site of this structure , see Figure 4—figure supplement 1 ) , and the other nucleosides are simply chemical structures in the same orientation as cytidine . Fluorine atoms in gemcitabine are colored cyan . ( C ) Crystal structure of vcCNT-7C8C bound to gemcitabine . Density shown is from an Fo–Fc SA-OMIT map contoured at 3σ . For a stereo view of the electron density in the nucleoside-binding site , see Figure 4—figure supplement 2 . ( D ) Alignment of uridine-bound and gemcitabine-bound vcCNT structures . Structures were aligned by Cα using PyMOL . TM7 was not used for the alignment . Cα traces and interacting amino acid residues are shown . The uridine-bound vcCNT structure ( PDB ID: 3TIJ ) is gray and the gemcitabine-bound vcCNT-7C8C structure is deep purple . ( E ) Alignment of vcCNT and vcCNT-7C8C structures bound to uridine ( PDB ID: 3TIJ ) , zebularine , cytidine , pyrrolo-cytidine , 5-fluorouridine , and gemcitabine . Alignments were performed in the same manner as D . vcCNT-7C8C-gemcitabine is shown in hot pink and all other structures are shown in gray . DOI: http://dx . doi . org/10 . 7554/eLife . 03604 . 01710 . 7554/eLife . 03604 . 018Figure 4—figure supplement 1 . Electron density at the nucleoside-binding site of vcCNT-7C8C-cytidine . The nucleoside-binding site of vcCNT-7C8C bound to cytidine is shown in stereo in stick representation in the same orientation and colored as the structures in Figure 2 . Cytidine is yellow and the red spheres are water molecules . The resolution is 2 . 6 Å and density shown is from a 2Fo–Fc electron density map contoured at 1σ . DOI: http://dx . doi . org/10 . 7554/eLife . 03604 . 01810 . 7554/eLife . 03604 . 019Figure 4—figure supplement 2 . Electron density at the nucleoside-binding site of vcCNT-7C8C-gemcitabine . The nucleoside-binding site of vcCNT-7C8C bound to gemcitabine is shown in stereo in stick representation in the same orientation and colored as in Figure 4C . Gemcitabine is yellow and the red sphere is a water molecule . The resolution is 2 . 9 Å and density shown is from a 2Fo–Fc electron density map contoured at 1σ . DOI: http://dx . doi . org/10 . 7554/eLife . 03604 . 019 Of all of the nucleosides and nucleoside analogs studied , the fluorescence probe pyrrolo-cytidine had the strongest binding affinity for vcCNT ( KD = 0 . 9 μM , Figure 5A ) . We solved the crystal structure of vcCNT-7C8C bound to pyrrolo-cytidine and found that the methylpyrrole ring fits neatly into a pocket formed by TM4 and TM6 ( Figure 5B , Figure 5—figure supplement 1 ) . None of the other nucleosides in this study have moieties that can exploit this ‘nucleo-pocket’ , and therefore this could be the root of the added strength of binding for pyrrolo-cytidine . 10 . 7554/eLife . 03604 . 020Figure 5 . Design of pyrrolo-gemcitabine and its transportability by hCNTs . ( A ) Chemical structures and KDs of gemcitabine and the pyrrolo-nucleosides . ( B ) The crystal structure of vcCNT-7C8C in complex with pyrrolo-cytidine is shown in surface representation with pyrrolo-cytidine shown in stick representation . The additional three carbons that comprise the methylpyrrole ring of pyrrolo-cytidine are colored green . The vcCNT-7C8C nucleo-pocket , formed mainly by G187 ( TM4 ) , V188 ( TM4 ) , and L259 ( TM6 ) , is delineated with a dotted line . The location of the cytoplasm , adjacent to the nucleo-pocket , is shown . For a stereo view of the electron density in the nucleoside-binding site , see Figure 5—figure supplement 1 . ( C ) hCNT3 transports gemcitabine but not pyrrolo-gemcitabine . Na+ currents were elicited by the addition of nucleoside to Xenopus oocytes expressing hCNT3 , and currents were measured by two-electrode voltage-clamp . An example current trace is shown . For each individual oocyte , the area under each current peak was measured to calculate total charge ( Q ) transported during application of nucleoside . The ratio of total charge co-transported with gemcitabine to that with pyrrolo-gemcitabine or gemcitabine total charge to gemcitabine + pyrrolo-gemcitabine total charge was calculated for each oocyte experiment ( means ± SEM , n = 14 oocytes ) . For Gem + Pyr-Gem , 200 μM of each nucleoside was added simultaneously . ( D ) hCNT1 transports pyrrolo-gemcitabine better than gemcitabine . Same experiment as in C but hCNT1-expressing oocytes were used and the ratio of total charge for pyrrolo-gemcitabine to gemcitabine is shown ( means ± SEM , n = 11 oocytes ) . Neither gemcitabine nor pyrrolo-gemcitabine elicited currents in water-injected oocytes ( Figure 5—figure supplement 2 ) See Figure 5—source data 1 for total charge source data . DOI: http://dx . doi . org/10 . 7554/eLife . 03604 . 02010 . 7554/eLife . 03604 . 021Figure 5—source data 1 . Total charge data . Total charge transported ( μC ) was calculated as the area under the curve of the current trace after application of the indicated nucleoside to hCNT-expressing Xenopus oocytes . Each row represents a separate experiment done using a different oocyte . The ratios of the total charge transported for the nucleosides are shown in the last columns . DOI: http://dx . doi . org/10 . 7554/eLife . 03604 . 02110 . 7554/eLife . 03604 . 022Figure 5—figure supplement 1 . Electron density at the nucleoside-binding site of vcCNT-7C8C-pyrrolo-cytidine . The nucleoside-binding site of vcCNT-7C8C bound to pyrrolo-cytidine is shown in stereo in stick representation in the same orientation and colored as the structures in Figure 2 . Pyrrolo-cytidine is yellow , and the red spheres are water molecules . The resolution is 2 . 8 Å and density shown is from a 2Fo–Fc electron density map contoured at 1σ . DOI: http://dx . doi . org/10 . 7554/eLife . 03604 . 02210 . 7554/eLife . 03604 . 023Figure 5—figure supplement 2 . Water-injected oocytes do not respond to gemcitabine or pyrrolo-gemcitabine treatment . Current traces are shown for three different oocyte experiments . Experiments were performed in the same manner as those in 5C and 5D except water-injected oocytes were used . DOI: http://dx . doi . org/10 . 7554/eLife . 03604 . 023 From our structural and binding studies , we learned that ribose interactions are important for CNT binding but nucleobase interactions are less stringent and can even be modified to improve binding . Because many nucleoside drugs contain modifications at the ribose ( e . g . , gemcitabine , AZT , and cytarabine ) , their apparent affinities for hCNTs are low ( Graham et al . , 2000; Clarke et al . , 2006 ) . We wondered whether this loss of affinity due to ribose modification could be compensated for by modification of the nucleobase . We synthesized a gemcitabine analog with the fluorescent nucleobase of pyrrolo-cytidine , which we now refer to as pyrrolo-gemcitabine ( Figure 5A ) . We measured its binding affinity for vcCNT using the fluorescence-anisotropy assay and found that the KD decreased by ∼60-fold to 24 μM , suggesting that nucleobase interactions and ribose interactions are additive . We next sought to test how these results translated to transportability by hCNTs . It is known that hCNT1 , hCNT3 , and hENT1 are the main NTs that transport gemcitabine ( Marechal et al . , 2009; Bhutia et al . , 2011; Damaraju et al . , 2012; Paproski et al . , 2013 ) . To detect nucleoside transport by hCNTs , we turned to two-electrode voltage-clamp electrophysiological recording . Because hCNTs are Na+-nucleoside symporters , one can measure the current generated by the Na+ transport that is coupled with nucleoside transport . We injected Xenopus oocytes with mRNA coding for each of the hCNTs and measured inward Na+ currents elicited by the addition of different nucleosides to the extracellular side . Addition of 200 μM gemcitabine to hCNT3-expressing oocytes induced Na+ currents ( Figure 5C , Figure 5—figure supplement 2 ) . However , the same amount of pyrrolo-gemcitabine had almost no effect . The lack of Na+ current upon addition of pyrrolo-gemcitabine can either mean that hCNT3 is unable to bind the modified compound or hCNT3 binds but cannot transport the modified compound . To resolve this issue , a mixture of equal concentrations of both compounds was added and a reduction of Na+ current was observed , suggesting that hCNT3 binds to but does not transport pyrrolo-gemcitabine . In contrast , 200 μM pyrrolo-gemcitabine elicited ∼sevenfold higher total charge uptake than gemcitabine for hCNT1 ( Figure 5D ) . hCNT2 transported neither compound ( data not shown ) . Although both hCNT1 and hCNT3 can transport pyrimidines , by modifying the pyrimidine nucleobase of gemcitabine we have created a subtype-specific nucleoside analog with enhanced transportability by hCNT1 . The nucleo-pocket of vcCNT differs by only one amino acid ( Gly 187 to Ser 374 ) from hCNT3 and three amino acids ( Gly 187 to Ser 352 , Val 188 to Leu 353 , and Leu 259 to Val 424 ) from hCNT1 . Since the nucleoside-binding site of vcCNT is highly homologous to hCNTs , we generated models of the hCNT1 and hCNT3 nucleo-pockets by swapping out these residues in the vcCNT-pyrrolo-cytidine structure . The structural models of hCNT1 and hCNT3 suggest changes in the overall structure of the nucleo-pocket ( Figure 6A , B ) . In particular , the nucleo-pocket of the hCNT1 model does not have a large enough cavity to accommodate the pyrrole ring , and the opening to the intracellular solution is larger due to the smaller side chain on TM6 . In the paradigm of the alternating-access mechanism of sodium-coupled symporters ( Krishnamurthy et al . , 2009 ) , substrate release is achieved by the transition from the inward-facing occluded to the inward-facing open state ( Figure 6C ) . Because the structure of vcCNT adopts an inward-facing occluded conformation where TM6 , including Leu 259 , serves as part of the gate ( Johnson et al . , 2012 ) , the hCNT1 model suggests that the additional pyrrole group may destabilize the inward-occluded state and facilitate the transition to the inward-open state . In contrast , the additional pyrrole group may stabilize the inward-occluded state of hCNT3 and slow the transition into the inward-open state . Our hypothesis predicts that changing the structure of the nucleo-pocket in the inward-facing state would affect nucleoside transport by hCNTs . Consistent with our prediction , mutation of Val 375 and Leu 446 of hCNT3 to mimic the nucleo-pocket of hCNT1 leads to an increase in transport of pyrrolo-gemcitabine ( Figure 6D ) . Furthermore , mutation of Leu 353 and Val 424 of hCNT1 to mimic the nucleo-pocket of hCNT3 leads to a decrease in transport of pyrrolo-gemcitabine ( Figure 6E ) . 10 . 7554/eLife . 03604 . 024Figure 6 . Structural basis of the subtype selectivity of pyrrolo-gemcitabine . ( A ) Model of hCNT3 nucleo-pocket . The structure of vcCNT-7C8C bound to pyrrolo-cytidine was used to generate a model of the hCNT3 nucleo-pocket by mutating the appropriate residues in PyMOL and selecting the rotamer that yielded the lowest amount of steric clash . ( B ) Model of hCNT1 nucleo-pocket . The model was generated in the same manner as A . Note that the methylpyrrole ring ( green ) will clash with the hCNT1 nucleo-pocket if it maintains the nucleoside-binding mode observed in the vcCNT structure . ( C ) Hypothetical alternating-access mechanism of vcCNT . A cartoon representation of the different conformational states along the transport cycle is depicted . The transport domain ( including HP1 , HP2 , TM4b , and TM7b ) and TM6 are shown as cylinders . Uridine is shown in stick representation . The nucleo-pocket in the inward-occluded conformation ( bottom right ) is located between TM6 and TM4 and is marked with a green star . The inward-occluded conformation is derived from the crystal structures of vcCNT . All other conformations are purely hypothetical . The transition between inward- and outward-facing conformations has been proposed to be achieved by a rigid-body movement of the transport domain across TM6 ( Johnson et al . , 2012 ) . Extracellular and intracellular gating likely involves slight rearrangements of HP2/TM4b and HP1/TM7b , respectively . ( D ) hCNT3 ( 375L/446V ) is capable of transporting both gemcitabine and pyrrolo-gemcitabine . Same experiment as Figure 5C , D but hCNT3 ( 375L/446V ) -expressing oocytes were used and the ratio of total charge ( Q ) co-transported with gemcitabine to pyrrolo-gemcitabine is shown ( means ± SEM , n = 11 oocytes ) . ( E ) hCNT1 ( 353V/424L ) transports pyrrolo-gemcitabine less efficiently than gemcitabine ( means ± SEM , n = 8 oocytes ) . Note that a higher nucleoside concentration was needed for D and E than the wild-type experiments due to lower transporter activity and/or expression . See Figure 6—source data 1 for total charge source data . DOI: http://dx . doi . org/10 . 7554/eLife . 03604 . 02410 . 7554/eLife . 03604 . 025Figure 6—source data 1 . Total charge data . Total charge transported ( μC ) was calculated as the area under the curve of the current trace after application of the indicated nucleoside to hCNT-expressing Xenopus oocytes . Each row represents a separate experiment done using a different oocyte . The ratios of the total charge transported for the nucleosides are shown in the last columns . DOI: http://dx . doi . org/10 . 7554/eLife . 03604 . 025
Our structural and equilibrium-binding studies of vcCNT have allowed us to better understand the design principles of nucleoside recognition by CNTs . Two helical hairpins ( HP1 and HP2 ) and two unwound helices ( TM4 and TM7 ) , related by twofold pseudo-symmetry , create a bowl-shaped nucleoside-binding site at the center of the transport domain of vcCNT ( Figure 1B , C ) . The interactions with the nucleobase are mainly formed with HP1 and TM4b , and the interactions with the ribose are mainly formed with HP2 and TM7b ( Figure 1C ) . At the base of the bowl-shaped nucleoside-binding site , Phe 366 interacts with both the nucleobase and the ribose through π–π and CH–π interactions , respectively ( Figure 3A , B ) . Three features make the architecture of the nucleoside-binding site of vcCNT particularly interesting: ( 1 ) the twofold pseudo-symmetry of the binding site that is divided in nucleobase and ribose binding on either side of the symmetry axis; ( 2 ) an aromatic ring at the base of the bowl that interacts with both the nucleobase and the ribose; ( 3 ) the localization of most of the protein–nucleoside interactions to one side of the nucleoside . These features of the nucleoside-binding site lead to the following questions: What are the energetics of the nucleobase and ribose interactions ? What is the role of Phe 366 ? Why are the interactions with the nucleoside localized to the concave side of the bowl ? With regard to binding energetics , our equilibrium-binding studies have shown that the ribose interactions are energetically important consistent with previous non-equilibrium studies with hCNTs ( Clarke et al . , 2006 ) . However , while disruption of the nucleobase interactions can have significant effects , the nucleobase can also be modified to improve binding . It is worth noting that our equilibrium-binding studies of vcCNT translate well to hCNT function . For example , we showed that the interaction between the C3′ ribose hydroxyl group and Ser 371 on TM7b is important in vcCNT ( Figures 1C and 4A ) . Consistent with our observation , the hCNT1S546P variant is non-functional ( Ser 546 in hCNT1 is equivalent to Ser 371 in vcCNT ) ( Cano-Soldado et al . , 2012 ) . Another important finding of our studies is that interactions with the nucleobase and ribose can be additive , and thus the loss of binding energy from modification of part of the nucleoside can be compensated for by the gain of energy from modification of another part of the nucleoside . Therefore , if a nucleoside drug contains a chemical modification necessary for its pharmacological function that hampers its recognition by CNTs , a compensatory chemical modification can be made so that the drug can still be recognized by CNTs . Our structural and equilibrium-binding studies highlighted the importance of Phe 366 in nucleoside recognition . The nucleoside-bound structures of vcCNT help to shed light upon the structural basis of the importance of this residue . Notably , Phe 366 is the only residue within the nucleoside-binding site that forms interactions with both portions of the nucleoside . Furthermore , all of the other binding-site residues on HP1 , HP2 , TM4b , and TM7b interact from the same side of the nucleoside as Phe 366 , forming the shape of a bowl around the nucleoside with Phe 366 serving as its base . As a result , the nucleoside rests on the face of Phe 366 , likely helping to orient the nucleoside so that it may form all of the other interactions within the binding site . Although π–π and CH–π interactions are generally weak , they can provide significant interaction energies depending on the circumstance ( Waters , 2002; Nishio , 2011 ) . The importance of Phe 366 is demonstrated by the changes in affinity of C5-substituted uridine analogs for vcCNT by altering the π–π interaction with the nucleobase and the significant reduction of affinity of C2′-substituted nucleoside drugs via disruption of the CH–π interaction with the ribose . Furthermore , ITC experiments with Phe 366 mutants revealed the stringent requirement for phenylalanine at this position , as even replacement with tyrosine resulted in a significant loss of binding affinity for uridine . Taken together , we propose that the role of Phe 366 is to position the nucleoside for effective binding thus serving as a ‘selectivity ring’ . What is the molecular basis of having a bowl-shaped nucleoside-binding site ? Several other nucleoside-binding proteins bind to their substrates by sandwiching the nucleobase between aromatic residues ( Suzuki et al . , 2004; Monecke et al . , 2014 ) . The utilization of a bowl-shaped binding site for a transporter makes practical sense as the substrate must be able to bind to and dissociate readily from the binding site in order for efficient transport to occur . We previously proposed that the transport domain undergoes a rigid-body motion while the scaffold domain , including TM6 , remains static during the transition from the outward-occluded to the inward-occluded conformational state ( Figure 6C; Johnson et al . , 2012 ) . In the inward-occluded conformation , TM6 ( including Leu 259 ) is located on top of the bowl , serving as part of the intracellular gate and partially occluding the nucleoside from dissociating into the cytoplasm . Portions of HP1 and TM7b form the rest of the intracellular gate and may move slightly during the transition from the inward-occluded to the inward-open conformational state ( Figure 6C ) , allowing the nucleoside to exit from the top of the bowl . This rigid-body conformational change also suggests that most of the interactions between transporter and nucleoside that are present in the inward-occluded structure are maintained in the outward-facing conformation , as observed with other sodium-coupled transporters ( Reyes et al . , 2009; Zhou et al . , 2014 ) . One exception is the interaction between the methylpyrrole ring of pyrrolo-cytidine with the nucleo-pocket , which is likely to preferentially interact with the inward-occluded conformation because of the involvement of part of the scaffold domain ( TM6 , Figures 5B and 6C ) , which we expect to be immobile during the conformational change . While both conformations should be represented when the transporter is solubilized in detergent micelles , we anticipate that our structural and equilibrium-binding data revealed most of the amino acid residues that are important for nucleoside recognition . Consistent with our hypothesis , amino acid residues that were shown to be important from our inward-conformation-based binding studies in detergent have also been shown to be important for the binding and transport of nucleosides by human CNTs in several cell-based mutational studies ( Loewen et al . , 1999; Zhang et al . , 2003 , 2005; Yao et al . , 2007; Slugoski et al . , 2009 ) . Further structural and biophysical studies probing the outward-facing state will help us to develop a complete understanding of the principles of nucleoside recognition by CNTs . Despite the recent progress in the area of structural biology of transporters , most SLC transporter structures determined to date are not highly homologous to human transporters , and it is often the case that drug interactions with these non-human transporters are different from their human counterparts , rendering structural studies of drug–transporter interactions technically challenging and necessitating substantial engineering to mimic the behavior of the human transporters ( Singh et al . , 2007; Wang et al . , 2013 ) . Although our CNT is prokaryotic in origin , it is an excellent model system to study human CNTs and offers us the opportunity to conduct structural studies of nucleoside and nucleoside-drug selectivity by hCNTs . Understanding the structural principles of nucleoside and nucleoside drug recognition by vcCNT not only allowed us to understand why a certain class of drugs ( e . g . , gemcitabine and cytarabine ) are not well recognized and transported by hCNTs , but also offered us an opportunity to modify an existing drug to improve its affinity for vcCNT . Furthermore , our electrophysiological studies show that it is not only transported more efficiently by hCNT1 , but it is also selectively transported by hCNT1 . Our structural models of hCNT1 and hCNT3 allowed us to hypothesize that the subtype-specific differences in the structures of the nucleo-pocket in the inward-facing-occluded conformation give rise to the subtype-selectivity of the modified compound . The results of the mutational studies of the nucleo-pockets in hCNTs are consistent with this hypothesis . Although we do not know whether the modification of gemcitabine affects the outward conformation due to the lack of a structure of the outward-facing conformation , our studies suggest that destabilization of the inward-facing-occluded step would facilitate the release of the pyrrolo-gemcitabine . Conversely , stabilization of the outward-facing-occluded step would facilitate the capture of the substrate . Taken together , if one has knowledge of both the outward- and inward-facing conformations of the transporter , it might be possible , at least in principle , to modify a compound to bind to both conformations with differential affinities , which may improve its transportability and selectivity . Prior to our studies , the presence of the nucleo-pocket structure was unknown . Although our study with pyrrolo-gemcitabine serves merely as proof of concept , it is conceivable that the nucleo-pocket structure can be utilized in the design of nucleoside-derived drugs or prodrugs that can be specifically targeted only to cell types that express hCNT1 since expression levels of hCNT1 are closely related to the responsiveness of many different types of normal or cancer cells to chemotherapy treatment ( Lane et al . , 2010; Naito et al . , 2010; Rabascio et al . , 2010; Bhutia et al . , 2011; Choi , 2012 ) . Finally , our studies provide another valuable concept: even though two transporter subtypes may share substrate and drug specificity , as is the case with hCNT1 and hCNT3 , it is still possible to use structural differences ( with the help of modeling ) between the two subtypes to design or modify a drug that can be selectively transported . This concept has broad applications to many SLC transporters that are involved in ADME since many families such as SLC21 , SLC22 , and SLC29 possess subtype-dependent drug specificities and/or tissue distributions ( Baldwin et al . , 2004; Koepsell and Endou , 2004; Hagenbuch and Stieger , 2013 ) . These results can also have an impact on basic scientific research . Since some transporter subtypes show significant changes in expression levels between normal and pathological conditions ( i . e . , cancer ) , and these changes in transporter expression are usually important for the pathological conditions to persist , a fluorescent compound that is subtype-specific ( e . g . , pyrrolo-gemcitabine ) for a certain transporter family can be a valuable tool to study the role of transporter subtypes in human health and disease through live cell imaging ( Farre et al . , 2004; Zhang et al . , 2006; Bhutia et al . , 2011; Perez-Torras et al . , 2013 ) . Taken together , this work not only represents an structural study of substrate and drug selectivity by membrane transporters , but our results also provide proof of principle for using this type of structure-function study for modifying drugs so that they are recognized and taken up into the cell by their cognate transporters more efficiently and selectively ( Han and Amidon , 2000; Majumdar et al . , 2004 ) .
Wild-type vcCNT and vcCNT-7C8C were expressed and purified as described ( Johnson et al . , 2012 ) in the absence of any added nucleoside . Briefly , protein was expressed as a His10-MBP fusion in C41 ( DE3 ) cells , cells were lysed by homogenizer ( AVESTIN , Ottawa , ON ) , protein was extracted from crude lysate using 30 mM dodecyl-maltoside , lysates were spun down to remove the insoluble fraction , and the supernatant was applied to a Co2+-affinity column for purification . The His10-MBP was cleaved by overnight digestion by PreScission Protease , and vcCNT was separated from His10-MBP by gel filtration using a Superdex 200 10/300 GL column in the presence of 5 mM decyl-maltoside . After gel filtration , protein was concentrated to ∼10 mg/ml and nucleoside was added to 1 mM ( uridine ) , 2 mM ( cytidine , adenosine ) , or 10 mM ( ribavirin , gemcitabine , 5-fluorouridine , pyrrolo-cytidine , zebularine ) . Crystals were grown in the presence of 100 mM CaCl2 , 37–42% PEG400 , and 100 mM buffer: HEPES pH 7 . 5 ( ribavirin ) , Tris–HCl pH 8 . 0–8 . 5 ( adenosine , cytidine , gemcitabine , pyrrolo-cytidine ) , or glycine pH 9 . 5 ( uridine , 5-fluorouridine , zebularine ) . Crystals were grown using the microbatch-under-oil technique . Crystals were harvested after 10–14 days , transferred to cryo solution containing 32 . 5% PEG400 , and flash frozen in liquid nitrogen . X-ray data were collected at beamlines 22-ID-D and 24-ID-C at the Advanced Photon Source at Argonne National Laboratory . Data were processed using HKL-2000 . The uridine , cytidine , zebularine , 5-fluorouridine , ribavirin , and pyrrolo-cytidine complex structures were refined using PHENIX with the original vcCNT structure ( PDB ID 3TIJ ) as the input model . The adenosine and gemcitabine complex structures were solved by molecular replacement with the original vcCNT structure as the search model using PHASER and refined using PHENIX . To measure the binding affinity of vcCNT for fluorescent nucleoside analogs , individual 500-μl solutions were prepared containing varying concentrations of vcCNT in 5 mM DM and either 5 μM pyrrolo-cytidine or 2 μM pyrrolo-gemcitabine . The fluorescence anisotropy of each solution at λex = 340 nm and λem = 467 nm was measured using a Cary Eclipse Fluorescence Spectrophotometer ( Agilent , Santa Clara , CA ) with automated polarizers . Each titration was performed at least three times . The data for the three titrations were simultaneously fit to a single-site binding model based off of Morrison's quadratic equation using nonlinear least-squares analysis in GraphPad Prism to obtain a dissociation constant and standard error . To measure the binding affinity of vcCNT for other nucleosides and nucleoside analogs , the nucleoside of interest was titrated 5 μl at a time into 500 μl of solution initially containing 5 μM of vcCNT in 5 mM DM and 1–2 μM pyrrolo-cytidine . The fluorescence anisotropy after each addition was measured . Each titration was performed at least three times . The data for the three titrations were simultaneously fit to a one-site competitive binding model based off of Wang's method ( Wang , 1995 ) using nonlinear least-squares analysis in GraphPad Prism to obtain a dissociation constant and standard error . vcCNT mutants were prepared in the same manner as wild-type vcCNT . 15–30 mM of uridine was titrated 5 μl at a time into 25–40 μM of vcCNT solubilized in 5 mM DM using a MicroCal VP-ITC system ( GE Healthcare , Pittsburgh , PA ) . The total heat exchanged during each injection was fit to a single-site binding isotherm with KD and ΔHo as independent parameters . See Supplementary file 1 for a full description of the pyrrolo-gemcitabine synthesis . Briefly , the Sonogashira coupling of known 2′-deoxy-2′ , 2′-difluoro-5-iodo-uridine ( Quintiliani et al . , 2011 ) with propyne followed by Cu ( I ) -mediated cyclization provided the corresponding furano-gemcitabine in 64% for 2 steps . Treatment of furano-gemcitabine with NH4OH and CH3OH completed the synthesis of the desired pyrrolo-gemcitabine in 75% . The genes coding for the three hCNTs and vcCNT were cloned into a pGEM-HE vector . Plasmids were linearized using either SphI or NheI , and mRNA was transcribed using the mMESSAGE mMACHINE T7 Transcription Kit ( Ambion , Grand Island , NY ) . Defolliculated Xenopus laevis oocytes were purchased from Ecocyte Bioscience ( Austin , TX ) . Individual oocytes were injected with 40 ng of mRNA using a 10-μl microdispenser ( Drummond Scientific , Broomall , PA ) fitted with a tapered glass pipette tip and incubated at 17 °C for 4–5 days in ND96 buffer ( 96 mM NaCl , 2 mM KCl , 1 mM MgCl2 , 1 . 8 mM CaCl2 , and 5 mM HEPES pH 7 . 5 ) with 0 . 1% penicillin and streptomycin before recording . The oocyte-recording chamber was gravity-perfused with ND96 buffer at a rate of 2 ml/min . Membrane currents were measured using an Oocyte Clamp ( OC-725C; Warner Instruments , Hamden , CT ) . Individual oocytes were penetrated with two microelectrodes filled with 3 M KCl ( 0 . 5–1 . 0 MΩ ) . All electrophysiological experiments were conducted at room temperature . The OC-725C Oocyte Clamp was computer-interfaced via an Axon Digidata 1550 and controlled by Axoscope software ( Molecular Devices , Sunnyvale , CA ) . The current signals were filtered at 20 Hz and sampled at intervals of 20 ms . The signals were filtered at 0 . 5 Hz by use of pCLAMP 10 . 4 software for data presentation . Ooctyes were impaled with the electrode filled with 3 M KCl , and then membrane potentials were observed for 10 min . Cells were discarded if resting membrane potentials were unstable or more positive than −30 mV . Oocyte membrane potentials were clamped at −90 mV for holding potentials to measure transporter-generated currents . All data are shown as means ± SEM . | DNA molecules are made from four bases—often named ‘G’ , ‘A’ , ‘C’ , and ‘T’—that are arranged along a backbone made of sugars and phosphate groups . Chemicals called nucleosides are essentially the same as these four building blocks of DNA ( and other similar molecules ) but without the phosphate groups . Proteins called nucleoside transporters are found in the membranes that surround cells and can pump nucleosides into the cell . These transporters also allow drugs that are made from modified nucleosides to enter cells; however , it was previously unclear how different transporters recognized and imported specific nucleosides . Like other proteins , nucleoside transporters are basically strings of amino acids that have folded into a specific three-dimensional shape . A protein's shape is often important for defining what that protein can do , as often other molecules must bind to proteins—much like a key fitting into a lock . Johnson et al . have now revealed the three-dimensional structure of one nucleoside transporter protein bound to different nucleosides and nucleoside-derived chemicals , including three anti-cancer drugs and one anti-viral drug . Some of these chemicals were shown to bind more strongly to the transporter protein than others , and examining the three-dimensional structures revealed that the different chemicals interacted with slightly different amino acids in the transporter protein . Johnson et al . then used this information to chemically modify an anticancer drug so that it is transported more easily into cells and is imported by only one of the subtypes of nucleoside transporters that are found in humans . This provides proof of principle that information about the structure and function of a transporter protein can help to redesign chemicals such that they can enter cells more efficiently , and to tailor them for transport by specific transporters . A similar approach may in the future allow researchers to design new nucleoside-derived drugs that are better at getting inside specific cells and , as such , provide effective treatments against cancers and viral infections . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"biochemistry",
"and",
"chemical",
"biology",
"structural",
"biology",
"and",
"molecular",
"biophysics"
] | 2014 | Structural basis of nucleoside and nucleoside drug selectivity by concentrative nucleoside transporters |
Any organism faces sensory and cognitive limitations which may result in maladaptive decisions . Such limitations are prominent in the context of groups where the relevant information at the individual level may not coincide with collective requirements . Here , we study the navigational decisions exhibited by Paratrechina longicornis ants as they cooperatively transport a large food item . These decisions hinge on the perception of individuals which often restricts them from providing the group with reliable directional information . We find that , to achieve efficient navigation despite partial and even misleading information , these ants employ a locally-blazed trail . This trail significantly deviates from the classical notion of an ant trail: First , instead of systematically marking the full path , ants mark short segments originating at the load . Second , the carrying team constantly loses the guiding trail . We experimentally and theoretically show that the locally-blazed trail optimally and robustly exploits useful knowledge while avoiding the pitfalls of misleading information .
Decision-making using noisy sensory information is a well-studied topic ( Kochenderfer , 2015; Chittka et al . , 2009; Kepecs et al . , 2008 ) . However , even in the absence of noise , cognitive or sensory limitations filter those aspects of the environment that are perceived by an organism ( Von Uexkull , 1957 ) . Relying on partial knowledge may result in maladaptive decisions ( Schlaepfer et al . , 2002 ) , where even a single bad decision somewhere among a long sequence of good ones has the potential to lead along an undesirable path ( von Herrath et al . , 1994; Davies , 2000 ) . This difficulty is pronounced in the cooperative contexts of multi-cellular systems or animal groups ( Czaczkes et al . , 2016; Schmidt et al . , 2006; Beckers et al . , 1990; Braess , 1968 ) which collectively require information that may be inaccessible to their individual members ( Berdahl et al . , 2013; Robinson et al . , 2014 ) . Cooperative transport by ants is a behavior in which ants join to collectively carry an object that is too heavy for any single one of them ( McCreery and Breed , 2014; Czaczkes and Ratnieks , 2013; Gelblum et al . , 2015 ) . The inherent scale differences that characterize this behavior hold the potential of introducing conflicts between the knowledge available to individual ants to that required for the load’s navigation . This system provides a unique opportunity for studying the consequences and possible resolution of such conflicts due to two main reasons . The first is experimental tangibility: In the context of cooperative transport , abstract decision making processes are manifested as collective motions; Sequences of decisions are evident as the load’s direction changes wherein bad turns eventually lead toward dead-ends . Second , cooperative transport is highly efficient as a general transportation solution . Ants manage to maneuver a wide range of loads while quickly navigating through complex terrains and back to their nest ( Czaczkes and Ratnieks , 2013; Gelblum et al . , 2015 ) . Here , we study the cooperative transport by P . longicornis ants ( McCreery and Breed , 2014; Czaczkes and Ratnieks , 2013; Gelblum et al . , 2015 ) , an invasive species with a broad worldwide distribution ( Wetterer , 2008 ) . Similar to other ant species , P . longircornis employ a pheromonal recruitment mechanism ( Czaczkes et al . , 2013b ) . Here , we demonstrate , for the first time in any ant species , the use of scent marks that assist navigation during the retrieval stage of cooperative transport . Tracking these scent marks provides us with a direct glance into the relations between the information that individual ants present to the group and the subsequent navigational choices on the collective scale . We find that while the information conveyed by scent marks is typically precise , it is occasionally misleading . This is because individual ants appear to be ignorant of the accessible paths from the perspective of the much larger load . When grouped together , the scent marks collectively left by the ants form a locally-blazed trail that is characterized by short range markings and highly stochastic following . We discuss the traits of this trail in the context of previously described trails and suggest that it should be considered as a new kind of ant trail . We further use an algorithmic approach to show how the unique characteristics of the locally-blazed trail are suited to the distinct navigational challenges of cooperative transport . To support our theoretical and numerical findings , we present experimental evidence for efficiency in the ants' collective motion along with experimental verification of some of the model's predictions .
In P . longicornis , an ant that finds a food item that is too heavy to carry alone , returns to the nest and recruits help ( Figure 1A , Figure 1—figure supplement 1 ) ( Stanley and Robinson , 2007; Trager , 1984; Kenne et al . , 2005 ) . This recruitment trip spans the full distance to the nest ( N = 12 out of 12 ants at six meters , 2000 folds the body length of the ants ) and is characterized by a highly punctuated motion ( Figure 1B ) that was previously associated with pheromone laying behavior ( Holldobler and Wilson , 1990; Beckers et al . , 1992a; Hölldobler et al . , 1978 ) . Using a side-view camera we could indeed relate dips in the ant’s speed with marking events in which the ant touches the tip of its gaster to the surface , a known property of marking behavior in various ant species ( Beckers et al . , 1992a; Hölldobler et al . , 1978; Czaczkes et al . , 2013a; Holldobler , 1981 ) ( Figure 1C , Video 1 ) . Chemical analysis of the marked surfaces ascertains that the identified events are accompanied by the deposition of undecane , a prominent P . longicornis trail pheromone ( Morgan et al . , 2005; Witte et al . , 2007a; Witte et al . , 2007b ) ( Figure 1D–E , Materials and methods ) . The fast evaporation of this short hydrocarbon dictates an extremely short lifetime ( Czaczkes and Ratnieks , 2013; Fujiwara-Tsujii et al . , 2006 ) in comparison to most other pheromones found in the trails of other ant species ( Morgan , 2009; Jackson et al . , 2006; Beugnon and Déjean , 1992 ) and even to trails constructed by P . longicornis outside the context of cooperative transport ( Witte et al . , 2007a ) . In fact , the high volatility of undecane makes it a prevalent ant alarm pheromone ( Fujiwara-Tsujii et al . , 2006; Blum , 1969; Lenz et al . , 2013; Regnier and Wilson , 1968 ) . The behavioral response of P . longicornis to this pheromone has , accordingly , been demonstrated to elicit short lived attractive responses , on the order of one minute ( Witte et al . , 2007b ) . 10 . 7554/eLife . 20185 . 003Figure 1 . Scent mark detection . ( A ) A typical recruitment run . The first recruiter deposits a sequence of pheromone marks ( blue dots ) that span an entire path between the food ( ring ) and the nest . Briefly thereafter , other ants are recruited ( see Figure 1—figure supplement 1 ) either from the surrounding area or from the nest , showing high attraction to the original scent trail . The trajctories of several such recruited ants , as they move towards the food item , are depicted by black lines . ( B ) The speed time-series of the recruiting ant depicted in panel ( A ) . Blue dots indicate marking events . ( C ) A side view of a typical marking event characterized by both a lowering of the gaster ( shortening of the vertical yellow line ) and reversal in the ant’s speed ( magenta ) . See also Figure 1—figure supplement 2 and Video 2 . ( D ) Normalized ion count Gas-Chromatography-mass-Spectrometry in the region of undecane for marking ants , control ants and blank . ( E ) A box-plot representation of the area beneath the undecane peak displayed in ( D ) . Asterisks denote p-value<0 . 0001 Kolmogorov-Smirnov test for non-equal areas between transport samples ( N = 6 experiments ) and grouped blank and control samples ( N = 12 experiments ) . ( F ) Marking positions during cooperative transport . The marks were produced by multiple ants over a 1200 cm2 area during 265 s . Full line denotes the trajectory of the load ( ring ) during the same time period . Color codes for elapsed time since the beginning of motion . Red bars denote 2 cm . DOI: http://dx . doi . org/10 . 7554/eLife . 20185 . 00310 . 7554/eLife . 20185 . 004Figure 1—figure supplement 1 . Recruitment . The number of ants in a 10 × 10 cm square around a food item before and during the progression of recruitment events ( N = 10 , mean±SEM ) . All data sets are aligned with the first marking event of the first recruiter ant ( green line ) and to the moment of transport initiation ( blue line ) . This allows a better overlay of recruitment events that naturally differ in their time of initiation and recruitment rate ( e . g . , due to different initial distance from the nest ) , thus revealing the characteristic profile of the recruitment process . DOI: http://dx . doi . org/10 . 7554/eLife . 20185 . 00410 . 7554/eLife . 20185 . 005Figure 1—figure supplement 2 . Verification of marking identification . The marking behavior was simultaneously recorded from both the top- and the side-view . The detection of the marking behavior , based on the ants speed profile extracted from the top-view images , was then verified against the side-view video by observing the ant touches the ground with the tip of its gaster . DOI: http://dx . doi . org/10 . 7554/eLife . 20185 . 00510 . 7554/eLife . 20185 . 006Video 1 . Marking behavior . A side close-up view of a recruiter ant as she lays three pheromone marks . Movie slowed down four times . DOI: http://dx . doi . org/10 . 7554/eLife . 20185 . 006 The association between a specific speed signature and the lowering of the gaster enables the identification of marking events using a top view camera ( 93% true positives and a false-detection-rate of 13 . 1% , see Figure 1—figure supplement 2 , Video 2 and Video 3 ) . This holds a significant advantage since discerning marking events by use of a top-view , rather than a side-view , increases the area over which this behavior can be tracked from a narrow one-dimensional bridge ( Beckers et al . , 1992; Aron et al . , 1989; Beckers et al . , 1993 ) to a large two-dimensional surface . Using this technique , we pinpoint the details of the ants’ dynamical scent map over the large length scales relevant for navigation by P . longicornis ants ( Figure 1F , Video 4 and Video 5 ) . 10 . 7554/eLife . 20185 . 007Video 2 . Verification of marking identification . Ants were recorded from both top ( a ) and side ( b ) view simultaneously . The detection of the marking behavior , based on the ants speed profile ( c ) that was extracted from the top-view video , was then verified by observing frames at which the tip of the ant’s gaster touches the surface as visible in the side view video . DOI: http://dx . doi . org/10 . 7554/eLife . 20185 . 00710 . 7554/eLife . 20185 . 008Video 3 . Verification of marking identification including speed signature . Speed profiles that were determined as candidates for marking events were further confirmed by a human observer against the raw video , where more subtle movements that are lost in the automatic tracking can be detected . This video shows a single marking event slowed down three times . DOI: http://dx . doi . org/10 . 7554/eLife . 20185 . 00810 . 7554/eLife . 20185 . 009Video 4 . Dynamic scent map of first recruiting ant . This video depicts the first ant that found the food and initiated recruitment back to the nest . Purple discs pinpoint the times and locations of the scent marks deposited by this ant . The radius of the discs is proportional to the time that had passed since the marking event . This video is slowed down four times . DOI: http://dx . doi . org/10 . 7554/eLife . 20185 . 00910 . 7554/eLife . 20185 . 010Video 5 . Dynamic scent map during recruitment and cooperative transport . An animation , based on actual scent-marks data , of a recruitment phase followed by cooperative transport . For clarity , this video depicts the load and laid marks ( as in Video 4 ) but not the participating ants . As in all other recruitment occurrences , the recruiting ant marked all the way towards the nest in a relatively straight line . In this video a mobile camera was used , and the frame was restricted by the field of view of the camera , showing an area of 20 cm × 15 cm around the load , at any given moment . All marks by all ants within this field of view are depicted in the animation , with the radius of the disc specifying the elapsed since the marking event . Conversely , any marking behavior that occurred outside the frame is not represented . The video is sped up twelve times . DOI: http://dx . doi . org/10 . 7554/eLife . 20185 . 010 A prominent characteristic of cooperative transport in P . longicornis ants is that a large fraction of ants accompany the load rather than physically carry it ( Czaczkes et al . , 2013b ) . Our detection method reveals that many of these ants engage in marking behavior ( Figure 1F ) , collectively laying scent marks with a mean rate of 1 . 4 marks per second ( T = 785 s ) in the vicinity of the load . We found that when ants were prevented from obtaining information about the location of the nest ( see the Experimental setup subsection of the Materials and methods ) , marking rates were significantly ( p<10-4 unpaired T-test , T = 978 s ) lower at 0 . 001 marks per second . This suggests that scent marks convey navigational information . Indeed , we find that , in obstacle free environments , markings hold accurate information ( 1 . 39 ± 0 . 02 bits per mark , N = 1395 , see Materials and methods ) regarding the direction to the nest ( Figure 2A ) . 10 . 7554/eLife . 20185 . 011Figure 2 . Following scent marks . ( A-C ) . Distribution of marking events ( N = 408 ) during a specific example of 113 cm of cooperative transport . Upper panels show the spatial distribution of scent marks ( purple dots ) locations , upon marking , in a moving frame of reference that is attached to the center of the transported load . The x-axis of this reference frame points towards the nest ( a ) or in the direction of the load movement in the 2 s that immediately proceed ( b ) or follow ( c ) the time this mark was deposited . Purple lines indicate quartile polytopes . Bottom panels: Angular distribution of the same data points . ( D ) Cooperative transport of a large prey item in a natural environment . Green dots denote scent marks . Red bars denote 2 cm . DOI: http://dx . doi . org/10 . 7554/eLife . 20185 . 011 As may be expected , scent marks influence the load’s motion: We find that , on average , each new mark transfers 0 . 35 ± 0 . 02 bits of directional information to the team of carriers , information that causes the group to divert the load’s motion towards the mark ( Figure 2B–C , see Materials and methods ) . To obtain a rough quantitative estimate on the rate of information transfer we can attribute ten such scent marks ( that occur , on average , within about 10marks/1 . 4 ( marks/second ) = 7 . 1 s ) with 10∙0 . 35 = 3 . 5 bits . Since each bit reduces uncertainty by a factor of two; 3 . 5 bits could be enough to increase the precision of the load's direction of motion from complete lack of orientation ( 360° ) to an accuracy of 360°/23 . 5=32° . Similarly , twenty marks occurring on a much shorter time scale than the duration of retrieval ( which typically lasts many minutes ) convey enough information for a high directional accuracy of under 5° . While the high directional accuracy of scent marks is typically advantageous , it may occasionally play a negative role in ragged environments which naturally contain obstacles . Near an obstacle , markings that are directed towards the nest may form misleading trails that are inaccessible to large loads ( Figure 2D ) . This could arise from the inherent gap between the perspective of an ant ( Von Uexkull , 1957 ) and the relevant information that is required by the group ( Video 6 ) . In this case , high fidelity to the directionally accurate yet navigationally misleading information could summon trap-like conditions and deadlocks ( Schmidt et al . , 2006; Beckers et al . , 1990 ) similar to those observed in classical , mass recruitment ant trails ( Czaczkes et al . , 2016; Planqué et al . , 2010; Beckers et al . , 1990 ) . 10 . 7554/eLife . 20185 . 012Video 6 . Scale differences between ants and load . Ants that cooperatively carry a large seed encounter a fallen leaf . The ants can easily pass over and under the obstacle while the carried load cannot . DOI: http://dx . doi . org/10 . 7554/eLife . 20185 . 012 Surprisingly , despite the fact that the ant-load system generally follows the scent trail ( Figure 2B–C ) , we observe that it readily abandons it ( Figure 3A ) on a length scale that is as short as 10 cm ( Figure 3B ) . This behavior significantly deviates from classical trail following behavior as exhibited by numerous ant species as well as P . longicornis itself in other scenarios ( see Figure 3B , Figure 3—figure supplement 1 ) . Indeed , ants wander off both the classical trail ( Beekman and Dussutour , 2007; Deneubourg et al . , 1983 ) and the locally-blazed trail . However , Figure 3B demonstrates the stark difference between these two cases: Of the P . longicornis ants walking along a 'classical trail' , at most 10% of them lose the trail which remains stable over time ( up to several days ) and over long distances . Conversely , in the locally-blazed trail the probability that the moving load approaches a scent mark continuously diminishes with distance so that segments of the trail marked even slightly away of the object become practically irrelevant to subsequent movement . We note that this observation is in agreement with the fast evaporation of undecane ( Witte et al . , 2007b ) . At first glance , abandoning the trail , especially the initial well-trodden recruitment trail which marks the direct path to the nest , after such short distances , seems to be counter intuitive and even mal-adaptive . 10 . 7554/eLife . 20185 . 013Figure 3 . A new kind of ant trail . ( A ) In an obstacle-free environment , the load loses the scent trail which then reforms in front of it . Green dots indicate the position of scent marks produced before the load reached the position indicated on the image ( ring ) . Blue dots indicate scent marks laid after this time . Solid line marks the load’s trajectory . ( B ) The probability that the load eventually approaches ( to less than 3 cm ) a scent mark as a function of the distance between them at the moment of marking . For comparison , the green line depicts the corresponding curve for a classical ant trail ( see Figure 3—figure supplement 1 ) ( C ) Cooperative transport while bypassing an obstacle ( thick red lines ) with a slit . Load position and marking colors as in panel ( a ) . ( D ) Distribution of single ant marking bout lengths defined as the distance between the load and the furthest mark in a marking sequence . The inset shows a typical marking bout of nine marks ( discs ) . Furthest mark is denoted in orange . Red bars denote 2 cm . DOI: http://dx . doi . org/10 . 7554/eLife . 20185 . 01310 . 7554/eLife . 20185 . 014Figure 3—figure supplement 1 . Classical trail . A 2D density plot of a 42 cm segment from a classical trail of the polydomous P . longicornis connecting two nests . The ants show high fidelity to the marked trail . A 3 cm wide stripe , bounded by green lines , around the center of trail ( yellow line ) was used in order to estimate a lower bound to the deviation of carrier ants from marks deposited in various distances ( the green line in Figure 3B ) . DOI: http://dx . doi . org/10 . 7554/eLife . 20185 . 01410 . 7554/eLife . 20185 . 015Figure 3—figure supplement 2 . Distance of first mark in bout from load . A distribution of the distance of the first mark in bout from the load ( N = 91 ) showing that the ants originate their marking sequence near the load ( mean = 3 . 1 cm , SD = 1 . 9 cm ) . DOI: http://dx . doi . org/10 . 7554/eLife . 20185 . 015 A possible hypothesis that could resolve this puzzling behavior is that the partial fidelity of the carrying team to scent marks is useful for avoiding obstacle-imposed deadlocks , as described above . To test this hypothesis , we simulated the presence of natural obstacles ( Figure 2D , Video 6 ) by constructing a barrier which has a slit at its center that easily allows for passage of ants but is too narrow for the load . This configuration results in markings that guide the load towards the dead-end ( Figure 3C , green marks ) . We find that rather than being caught in a deadlock , the transported load leaves the misleading scent marks and commences a perpendicular motion along , and eventually around , the obstacle ( Figure 3C ) . This effect of stochasticity is , in some sense , reminiscent of the positive role held by individual ants that wander off classical ant trails ( Beekman and Dussutour , 2007; Deneubourg et al . , 1983 ) ( more about this point in the Discussion below ) . The fact that the load is prone to leave the scent trail reduces the probability that it will approach scent marks laid far from it ( Figure 3B ) . In light of this unpredictable motion , investing in the construction of a long trail that connects the moving load to the nest appears to be wasteful . Indeed , we find that the vast majority of individual ants deposit markings in bouts which originate near the load ( Figure 3—figure supplement 2 ) , are directed towards the nest ( Figure 2A ) and terminate a short distance away ( Figure 3D ) much before they get anywhere near the nest ( which is several meters away ) . This locality of scent laying , with a median bout length of less than 10 cm ( 95% confidence interval is [8 . 6 cm , 10 . 3 cm] as estimated by a non-parametric Binomial-based method over N = 735 marking bouts ) , stands in stark contrast to the 600 cm trail laid by the first recruiter ant ( see the section titled Identification of pheromone deposition events , above ) ( p<10−6 probability that first ant marking distances come from the distribution depicted in Figure 3D ) . Whereas the function of the latter is long range recruitment ( Figure 1—figure supplement 1 ) the former is required both for orientation ( Figure 2A–C ) and , possibly , local recruitment of nearby ants ( Czaczkes et al . , 2013b; Hölldobler et al . , 1978; Holldobler , 1971; Schatz et al . , 1997 ) . Macroscopically , this individual marking and collective following behaviors lead to the formation of the locally-blazed trail; this trail significantly differs , in several aspects , from other previously described ant trails . First , rather than designating the whole route between the food and the nest ( Hölldobler et al . , 1978; Cammaerts and Cammaerts , 1980 ) ( or any other two stable points that are , in no sense , required to be proximal [Latty et al . , 2011] ) ; the locally-blazed trail is segmented and continuously formed in the vicinity of the object such that it implies only the next step to be taken . This locality holds on the collective level that takes into account all marking ants and is very different from the classical long-range punctuated or intermittent trail marking by individual ants as previously described ( Beckers et al . , 1992; Aron et al . , 1989; Cammaerts and Cammaerts , 1980; Hangartner , 1969 ) . Second , as the load occasionally deviates from the trail another trail dynamically forms in front of it . This happens in open space ( Figure 3A ) as well as near an obstacle ( Figure 3C ) . Interestingly , the length scales of the deviation from the scent trail and the marking bout length both match and are on the order of 10 cm , possibly reflecting an evolutionary dependency between the two . It is well known that greedy algorithms , i . e . algorithms that tend to optimize the next local step rather than its future consequences , often fail by converging onto a local rather than global solution . In some cases , such problems can be resolved by actively adding a random component to the algorithmic rules ( Fonseca and Fleming , 1993; Vermorel and Mohri , 2005; Kirkpatrick et al . , 1983 ) . In this section , we explore the possibility that the locally-blazed ant trail is one of these cases . To gain insight on the efficiency of the locally-blazed ant trail , we developed an abstract routing model ( see Supplementary file 1-1 , 2 ) . To model the terrain we discretize it into a graph . We use the freedom in determining the edge lengths of the graph such that they correspond to a reasonable step size . We naturally choose this scale to coincide with the motion’s persistence length , i . e . the length scale over which the motion can be approximated by a straight line , in the absence of scent marks , which was previously measured to be on the order of 10 cm ( Gelblum et al . , 2015 ) . It is the goal of a memoryless agent that has no sense of orientation to advance in a given direction as fast as possible . Inspired by the ants’ behavior , our model includes two intertwined components: Marking the graph with pointers ( advice ) that are biased towards the desired direction and utilizing this advice to progress the navigating agent . To model the fact that the trail is composed of broken segments ( see Figure 3D ) we anchor an advice pointer to each node . The length of a pointer is taken to be the experimentally measured median marking distance . Since this distance agrees with the grid spacing ( 10 cm ) , advice pointers can be understood to connect each node to one of its nearest neighbors . To model the unreliability of scent marks resulting from complex and unpredictable topologies , we assume that , although advice pointers typically point to the correct neighboring node , they sometimes point at an arbitrary direction . The second part of the model describes the partial responsiveness of the load to scent marks ( Figure 3A–C ) . This is modeled by the Probabilistic-Following algorithm: When reaching a node , the agent follows the local advice with some fixed probability and performs a random walk otherwise . These following rules have the potential of using some of the information held by the advice , while avoiding possible deadlocks in areas with misleading directions . Adapting results from the theory of Random Walks in Random Environments ( Drewitz et al . , 2014; Sznitman , 2002 ) , we prove that our model allows for asymptotically optimal ( i . e . linear speed ) travel times in 2D graphs ( which are a good description of the world our ants live in ) provided that advice error rates are sufficiently small ( see Supplementary file 1-4 ) . In other words , on these graphs and in the presence of occasionally misleading advice , adding noise to the advice following rules is not only necessary but also sufficient for efficient navigation . We further prove that similar results hold on line-graphs ( Supplementary file 1-3 and Figure 4A and Figure 4—figure supplement 1 ) and grids of all dimensions ( Supplementary file 1-4 ) . The question whether such results can be obtained for the general graph case seems to be beyond the scope of currently known techniques ( see Remark 23 in Supplementary file 1-4 ) . It remains open for further theoretical study . 10 . 7554/eLife . 20185 . 016Figure 4 . Performance and robustness of Probabilistic Following – simulation results . ( A ) Median first passage times of an agent performing Probabilistic Following on a line using a following parameter of 0 . 8 . Different plots signify different reliability of advice . When advice is all or mostly correct ( blue , purple , and magenta curves ) , the agent achieves fast , linear passage times ( for a linear scale plot see Figure 4—figure supplement 1 ) . When all advice points in the wrong direction passage times are exponential ( linear green line in this logarithmic plot ) . ( B ) Stretch ( see Materials and methods ) of paths on a 2D grid for the full range of environmental advice reliability and the agent's probability of following the advice . The colored zone signifies stretch values that are under 3 . 2 . Following probabilities between 0 . 5 and 0 . 9 ( dashed red lines ) are adequate for a large range of environmental values . ( C ) Stretch values for an advice reliability of 0 . 7 as a function of the agent's probability of following the advice . The local maxima near the edges indicate the poor performance of both a random walk ( left-hand side ) and a perfect following ( right-hand side ) strategy . The shallow minimum in between these indicates the lack of need for fine tuning to achieve near-optimal navigation times . ( D ) The stretch of an agent with a probability of 0 . 8 of following advice at a particular value of advice reliability ( x-axis ) normalized by the stretch of an agent which uses the optimal following probability for this particular environment . The value of one at an environmental reliability of 0 . 9 indicates that the agent with 0 . 8 following probability is the optimal at this environment . This same agent exhibits near-optimal performance ( i . e . deviates from the optimum by less than a factor of 1 . 5 , red line ) for the full range of environmental reliability values over 0 . 4 . DOI: http://dx . doi . org/10 . 7554/eLife . 20185 . 01610 . 7554/eLife . 20185 . 017Figure 4—figure supplement 1 . Median first passage times of an agent performing Probabilistic Following on a line . The data is identical to that presented in Figure 4A but , here , the y-axis is on a linear scale . Thin lines represent linear fits . DOI: http://dx . doi . org/10 . 7554/eLife . 20185 . 017 The navigational scheme described above is highly efficient at extracting useful information present in the environment even when it is masked by occasional misleading advice . Accordingly , Probabilistic-Following is expected to fail when most of the information is misleading . In fact , using this navigation scheme to cross large areas that contain only misleading information would take an exponentially long time ( see Supplementary file 1-3 and Figure 4A ) . In this case , completely ignoring advice and performing a simple random walk instead is a better strategy that yields quadratic crossing times . In an environment in which all advice is precise it is clear that the best navigational policy is to strictly follow this advice . However , in an environment which contains areas of misleading advice , this policy quickly breaks down and can lead to infinite deadlocks . In the previous section , we have shown how Probabilistic-Following can restore linearity in this case . It is therefore clear that an optimal navigational policy must take into account the statistics of the relevant environment . In this section , we discuss the degree to which Probabilistic-Following must be tuned to environmental statistics . We prove , in Supplementary file 1-4 , that to achieve linear passage times on the line graph there is no requirement for any sort of fine tuning . In fact , the fraction of times at which Probabilistic-Following ignores advice can be chosen from a large range of parameters without lengthening the travel time beyond linear . This suggests a degree of robustness in the performance of Probabilistic-Following . While linearity of travel time in distance traveled is crucial it is important to verify that the coefficient of this linear relation is not too high . We define the stretch of a path on a graph as its length normalized by the length of the shortest possible path between its two end-points . We used simulations to calculate the stretch of path generated by Probabilistic-Following on the 2D grid . These simulations include two parameters: the fraction of correct advice ( advice reliability ) and the following probability of the navigational algorithm . We find that the stretch is very low ( under 3 ) over a large fraction of the parameter space which spans following probabilities between 0 . 5 and 0 . 9 ( Figure 4B ) . This implies that for a given environment , there is a large range of following probabilities that perform almost equally as well as the best possible following probability ( Figure 4C ) . Furthermore , fixing the following probability to a single value within this range yields near optimal performance for almost every value of the advice reliability ( see Figure 4D ) . To summarize , Probabilistic-Following is highly robust in the sense that it achieves near-optimal performance with hardly any requirement to fine-tune its parameters to the statistics of the environment . Such robustness and the generality which it implies are biologically appealing . The ants’ navigation algorithm shares a number of traits with our model . It is therefore interesting to check whether the navigational optimality implied by the model indeed holds in the case of ant cooperative transport . The main prediction of the model is that travel times are linear and therefore near-optimal even in the presence of obstacles and misleading information . In the absence of obstacles , we have previously shown that , indeed , the load travels almost ballistically ( Gelblum et al . , 2015 ) . To check if this efficiency carries over to more complex environments , we studied trail formation and load motion around an obstacle with a slit located at its center ( Figures 3C and 5A ) . In such cases , scent marks near the slit are misleading while those closer to the obstacle's edges are beneficial ( Figure 5B ) . We further observe that the motion of the load around the obstacle exhibits partial responsiveness to the underlying scent marks ( see the trajectory as depicted in Figure 5B ) thus validating the assumptions of our model . 10 . 7554/eLife . 20185 . 018Figure 5 . Efficiency and limitations of the locally-blazed trail . ( A-B ) A trajectory of a transported load ( ring ) while it moves around an 80 cm block with a slit at its center . Color code depicts temporal progression , these panels are spatially aligned along the horizontal axis . ( A ) 3205 marks ( discs ) identified during this transport , show a marking preference towards the nest ( purple arrow ) or along the walls . Short red scale bar denotes 2 cm . ( B ) X coordinate of the load as a function of time . Polar histograms present the scent marks density towards the nest ( red ) , to the left ( magenta ) or to the right ( blue ) , during three phases along the trajectory: ( i ) While the load was close to the slit , ( ii ) While moving to the right until the reversal point and ( iii ) in the last 5 cm just before the load reaches the same turning point . ( C ) The mean time to first reach a given distance from the slit along the obstacle walls is shown for both open- ( Light blue ) and enclosed-configurations ( Green ) with their corresponding best fit . The two obstacle configurations are illustrated in Figure 5—figure supplement 1 . The log transformation of the data for the enclosed block , given in Figure 5—figure supplement 2 , reveals exponential dynamics . ( D ) The marking rates along the obstacle walls are given for the two obstacle configurations . Asterisks denote p-value<10-6 . DOI: http://dx . doi . org/10 . 7554/eLife . 20185 . 01810 . 7554/eLife . 20185 . 019Figure 5—figure supplement 1 . Two obstacle configurations . Both obstacles are flat and have a slit at their center which was pointed in the direction of the nest . The slit is large enough to allow for free motion of ants but not the load . The ‘open’ configuration ( A , blue ) allows for ant passage through the slit but also via the longer detour from either left and right . The ‘enclosed’ configuration ( B , green ) allows for ant passage through the slit only . This configuration we used to collect the data presented in Figure 5C–D ( using a matching color code ) . DOI: http://dx . doi . org/10 . 7554/eLife . 20185 . 01910 . 7554/eLife . 20185 . 020Figure 5—figure supplement 2 . First passage time in collective motion directed against the scent marks . This figure presents the ‘enclosed’ configuration data as in Figure 5C with in a semi-logarithmic plot . DOI: http://dx . doi . org/10 . 7554/eLife . 20185 . 020 The partial responsiveness of the ant-group to the scent marks entails two different effects: On the one hand as discussed above , despite misleading marks near the slit , partial responsiveness allows the group to avoid a deadlock ( Figure 5B , left side ) . The inevitable cost of this is that as the load follows marks that are correctly directed to the right , it temporarily switches its direction against the useful advice ( Figure 5B , right side ) . Taken together , the time saved near the slit outweighs the time lost on the non-productive detour and the load rapidly circumvents the obstacle . As predicted by our model , this motion is near-optimal in the sense that the mean passage time scales linearly with obstacle width ( Figure 5C ) . The previous section provided experimental support for the main prediction of our theoretical model . Here , we test further model predictions , both qualitative and quantitative , against the ants' cooperative transport behavior . An important qualitative prediction of any model are the boundaries beyond which it is expected to fail . For Probabilistic-Following these are environments with large ( compared to the persistence length of the collective motion ) areas that contain mainly misleading information . As noted above , in such cases our theoretical model predicts slow traversal times that scale exponentially with the distance . To test this prediction , we challenged the carrying team with an 80 cm obstacle similar to that depicted in Figure 5A but modified such that the only path free for ant passage is through the slit and any bypass from the left or the right is blocked ( see enclosed-configuration Figure 5—figure supplement 1 ) . This confined obstacle is useful since it almost completely eliminates marking bouts in these perpendicular directions ( see Figure 5D ) such that , practically , all scent marks mislead the carrying ants towards the slit . We find that , in this case , in agreement with the model's predictions , the time it takes the group to travel away from the slit scales exponentially with distance ( Figure 5C and Figure 5—figure supplement 2 ) . As predicted by the model , there is a stark contrast between linear traversal times when scent marks guide the ants around the obstacle and the exponential times when they guide it to the slit . This observation provides strong support for the importance of scent marks in the ants' obstacle circumvention behavior . This result has further consequences on the effect of direct interactions between the ants with the carried load and the obstacle . Such interactions are inevitable and include , at the least , the physical interaction between the load and the obstacle wall . While these interactions may induce some perpendicular motion ( see Figure 3C ) our result shows that , in general , this is insufficient to efficiently guide the group around the obstacle . Rather , scent marking and scent following are crucial components in this process . The qualitative agreement between experiment and theory demonstrates the usefulness of introducing a general , intuitive , and analytically solvable abstract model . Although we do not claim that the collective motion can be fully described by our model , we provide quantitative evidence that it can serve as a good first approximation . We do this by comparing obstacle circumvention in the presence ( open configuration depicted by blue Figure 5—figure supplement 1 ) or absence ( enclosed configuration depicted by green in Figure 5—figure supplement 1 ) of useful scent marks . First , we estimate the rate at which the ants turn against pheromonal advice in the open-configuration experiment ( see direction changes in Figure 5B ) to be , on average , once per 37 . 5 cm ( as measured over 15 trials and a total length of 636 cm ) . We can then use this rate to estimate the power of the exponential in Figure 5C . These two values are related since , in the enclosed-configuration experiment , all scent marks lead towards the slit and any movement away from it must result from instances in which the group moves against pheromonal markings . We find ( see calculation in Supplementary file 1-5 ) that the correct relation between these values holds for a persistence length of 7 cm , in agreement with the 10 cm scale mentioned above , and an 80% probability of following advice , a value that lies well within the model's high efficiency zone ( see Figure 4B ) .
To date , it was unknown whether ants employ scent marks during the retrieval phase of cooperative transport . Establishing an image analysis based method for the detection of scent laying events , we find that cooperative transport is indeed accompanied by significant pheromonal marking behavior ( Figure 1 ) . The context in which these scent marks appear significantly differs from that of ants foraging on small food items , such as seeds , that are amenable for individual transport . A first difference is a direct consequence of the presence or absence of scale discrepancies between individual ants and the load they carry . During the transport of small items marking ants and carrying ants can , to a large extent , traverse the same paths ( it is often the case that laden ants concurrently mark the path ) . In this case , the information communicated by trail laying ants is directly valuable to those that retrieve the food . This is not the case when large food items are involved . Simply put , in the context of cooperative transport , a passage that is open for a marking ant may be inaccessible for the larger load . Accordingly , we find that while scent marks typically point the group in the correct direction ( Figure 2A–C ) they may , at certain topological circumstances , be systematically misleading ( Figures 2D , 3C and 5A–B ) . A second difference between the two contexts involves the degree of dynamics that characterizes the navigational challenge . Scent marks deposited during the transport of small food items assist traffic between two stationary locations ( e . g . the pile of seeds and the nest ) ( Franks et al . , 1991; Bruce and Burd , 2012; Plowes et al . , 2013; Bottinelli et al . , 2015; Buhl et al . , 2009 ) . Conversely , during cooperative transport , scent marks bridge between a fixed location ( the nest ) and a dynamically moving object ( the carried load ) . In light of these differences – we find that , during cooperative transport , ants employ a trail that is morphologically distinct from previous descriptions of ant trails . First , this trail is locally-blazed , i . e . it is composed of short broken segments that are dynamically laid in the near vicinity of the moving load . Unlike classical ant trails where individual ants lay scent marks over long distances that are on the length scale of the distance to the nest ( Cammaerts and Cammaerts , 1980; Beckers et al . , 1992b ) , here , marking ants indicate only the next step to be taken ( Figure 3D ) . This highly transient role of the scent marks is reflected in their chemical makeup where the major component is the highly volatile pheromone undecane . Second , classical ant trails are typically clearly bounded and narrow . This is a result of the relatively strict manner in which ants follow the scent marks ( Bruce and Burd , 2012; Eidmann , 1927; Carthy , 1951 ) which dominates over a secondary stochastic process where some ants distributively and individually wander off the trail ( Beekman and Dussutour , 2007; Deneubourg et al . , 1983; Perna et al . , 2012; Gordon , 1995; Goss et al . , 1989; Reid et al . , 2011 ) . The situation is very different for the locally-blazed trail which does not have well defined boundaries . This is a result of repeated occasions in which the entire group of carrying ants collectively and simultaneously loses the trail ( Figure 3B ) . In turn , the trail is then reassembled from another location ( Figure 3A , C ) . It is an interesting question whether the unique characteristics of the locally-blazed ant trail are compatible with the navigational challenges that are specific to cooperative transport . A first challenge is the highly dynamic nature of the cooperative transport process wherein the constant motion of one of the trail’s end-points ( i . e . the food load ) leads to low predictability . Above , we have discussed how the local nature of marking behavior is compatible with these circumstances . A second main challenge that must be overcome during cooperative transport is the high propensity for deadlocks . Here , the challenge faced by the ants is not only to overcome isolated deadlocks but , rather , to successfully go through a long sequence of decisions while avoiding the accumulations of errors that may rise due to maladaptive choices . Below , we discuss the efficiency of stochastic following in this context . In general , it is theoretically established that stochastic components may sometimes grant search algorithms with the ability to escape from local minima ( Vermorel and Mohri , 2005; Selman et al . , 1998; Kirkpatrick et al . , 1983 ) . In particular , it has been shown that this principle is utilized by ants on classic foraging trails . In this case , noisy trail following can resolve possible deadlocks that may occur when a high quality food source is introduced to ants that have already established a trail to a mediocre food source ( Beekman and Dussutour , 2007; Deneubourg et al . , 1983 ) . In a similar manner , we predicted that the stochastic following behavior in the locally-blazed ant trail ( Figure 3B ) plays an important role in resolving deadlocks . We then validated this prediction experimentally ( Figure 3C ) . While stochasticity can help groups to collectively escape from isolated deadlocks it is by no means a given that it is a sufficient component for an efficient navigational strategy which , in the context of cooperative transport , must employ a sequence of highly dependent decisions while overcoming misleading information . Indeed , the addition of noise cannot be considered as a silver-bullet that increases the efficiency of any search algorithm and there exist scenarios in which it simply does not work ( Selman et al . , 1998; Reingold , 2005; Patrascu and Thorup , 2007 ) . Understanding how the locally-blazed trail allows ants to achieve near-optimal navigation performances ( Figure 4A–C ) is the subject of the theoretical aspects of this work . We conducted an algorithmic study to obtain a more fundamental understanding regarding the efficiency of the locally-blazed trail as a navigational strategy . Specifically , we studied an ant-inspired algorithm which includes imperfect ( stochastic ) responsiveness to local , occasionally misleading , advice . An intuitive version of this problem involves driving a car in an unknown country that is in the aftermath of a major hurricane which has randomly flipped a certain small fraction of the road-signs . A driver's goal is to reach her destination as quickly as possible . If the driver chooses to follow all signs , she can be trapped in an infinite loop . Altogether ignoring signs is also inefficient since this fails to utilize the useful information that is contained in unflipped signs . We analyze the algorithm in which the driver follows any road-sign she encounters with some fixed probability and takes a random direction otherwise . Adapting results from the mathematical field of Random Walks in Random Environments ( Drewitz et al . , 2014; Sznitman , 2002 ) , we have shown that this ant-inspired algorithm yields near-optimal navigation in topologies that are relevant to ant cooperative transport . Furthermore , the driver need not know exactly what fraction of the road-signs have been flipped . The robustness of the algorithm ( see Figure 4 ) implies that the time it takes to reach the destination does not strongly depend on the probability at which the driver chooses to ignore the road-signs . The lack of a requirement for fine-tuning is a favorable trait when modeling biological systems which are inherently noisy and must function in contexts of unforeseen external challenges . In the context of the intuitive example presented above , following each road-sign with some fixed probability works to help the driver reach her destination quickly , at least on 2D grids . We propose this mechanism as a simple strategy that may prove its efficiency in multiple contexts in which an agent needs to act in the presence of typically correct but , at places , systematically misleading advice . We note , however , that a rigorous proof would highly depend on the particular given model and may become extremely hard mathematically . In fact , even proving that this strategy holds with respect to navigation in general graph structures seems to be beyond the scope of currently known techniques ( see Supplementary file 1-4 ) , and remains a challenging mathematical question for future theoretical investigations . Above we have compared the locally blazed trail to previously described ant trails . It is further interesting to discuss this form of collective navigation in the context of collective decision making in other animal groups . To remain as a cohesive group , social animals have to take decisions collectively ( Conradt et al . , 2007 ) . A common strategy for reaching quality decisions is to follow the majority ( Ame et al . , 2006 ) . Majority based averaging mechanisms have been shown to help the group reach accurate decisions even when its members' individual perception is inaccurate and noisy ( Dell'Ariccia et al . , 2008; Simons , 2004 ) . Majority based decision making has also been demonstrated to be helpful even in the context of conflict ( Couzin et al . , 2011 ) : It is often the case that information perceived by each individual group member encompasses only a small fragment of the complete environmental state . This may result in a case wherein different group members drive the group towards different , often disjoint , collective choices ( Franks et al . , 2002; Seeley et al . , 2012; Biro et al . , 2006 ) . However , if individuals sense their environment in a reliable manner then favorable choices will be supported by a larger number of individuals than less favorable ones . In this case majority boosting techniques such as opinion polling ( Franks et al . , 2002; Seeley et al . , 2012 ) , quorum sensing ( Franks et al . , 2002; Seeley et al . , 2012 ) , and collective motion patterns ( Couzin et al . , 2005; Strandburg-Peshkin et al . , 2015 ) can assure that the group as a whole followed the best alternative . Cooperative transport through natural environments presents a different sort of challenge . While the majority opinion is usually reliable , at certain locations scale differences between an individual and the carried item can cause it to be plain wrong . Accordingly the locally-blazed trail includes two different movement components . The first stands in agreement with the examples presented above: namely the group follows the majority opinion as evident in the scent mark pattern ( see Figure 5B ) . The second component ignores the scent marks or majority opinion and moves in a different direction . Our model shows that randomly switching between these two behavioral patterns yields near-optimal navigation times . In a dynamic environment , majority opinions may grow to be outdated ( Rieucau and Giraldeau , 2009 , 2011; Warner , 1988 ) . Animal groups have been shown to produce innovation even when this contradicts the natural tendencies to conform to the majority ( Parrish , 1991; Reader and Laland , 2001 ) . The capacity for innovation has been observed in a diverse array of group living animals including fish ( Brown and Laland , 2002 ) , ants ( Beekman and Dussutour , 2007; Deneubourg et al . , 1983; Czaczkes and Heinze , 2015 ) , mammals ( Reader and Laland , 2001 , 2012 ) , and birds ( Rieucau and Giraldeau , 2011 ) that rely on mechanisms such as preference of individual information to social information ( Rieucau and Giraldeau , 2011 ) , the addition of random components to individual behaviors ( Beekman and Dussutour , 2007; Deneubourg et al . , 1983 ) , reduction of social signals ( Czaczkes and Heinze , 2015 ) , and deliberate individual exploration ( Brown and Laland , 2002 ) . It has been shown how this capacity allows the group to break away from suboptimal behavior ( i . e . local minimum ) ( Beekman and Dussutour , 2007; Deneubourg et al . , 1983; Reader and Laland , 2001; Day et al . , 2001 ) . The noisy component of the locally blazed ant trail falls within this group of collective phenomenon . An interesting distinction is that while previous observations often focused on noise and exploration in the context of a single choice we demonstrate its usefulness in a more complex situation involving a string of strongly dependent decisions . We show both theoretically and experimentally ( Figure 5 ) how the locally blazed trail suffices to efficiently navigate the group over long distances . Our work suggests that the usefulness of noise and local exploration in animal groups extends beyond short time scales and single decisions and may actually be efficient for the long time scales and multiple decisions that characterize any biological system . Finally , an ant colony is a hierarchical structure ( Wilson and Hölldobler , 1988 ) that can be studied on various scales of organization . In this work , we focused on the scale of the team ( Anderson , 2001 ) of carrying ants and studied the relations between its collective motion and the pheromone marking . It is , of course , interesting to understand how these relate to the single ant level . Although it is beyond the scope of this paper , below , we provide some hypotheses and directions for future research . One aspect of this concerns the mechanism by which individuals decide to lay scent marks . Such decisions may involve the ant's general knowledge of the environment ( Collett and Collett , 2002 ) , her recent history ( Wystrach et al . , 2013 ) , and her interaction with obstacles or passageways . The differences in scent mark laying behavior according to the large-scale structure of an obstacle , as presented in Figure 5D , suggest an experimental scheme for testing the interplay between these factors . The decision when and where to lay scent marks may further be influenced by the presence of previously laid pheromones . Note however , that a positive feedback between marking and further marking is not required for the cohesiveness of the trail itself; Marks may be in close proximity simply because they all originate at the load and point in roughly the same direction ( towards the nest ) . This may break down in the presence of obstacles where several alternative routes to the nest are available . Indeed , in this case , we observe that not all scent marks agree in their direction ( see Figure 5B ) such that no strict consensus is enforced . Importantly , such lack of consensus does not hold significant implications to our theoretical analysis ( see Supplementary file 1-1 ) . It is further interesting to study how the relationship between load movement and the locally blazed trail emerge from the interactions of individual ants with individual pheromone marks . Previous work showed that , in P . longicornis , steering of the carried load is accomplished by the pull exerted by newly attached ants ( Gelblum et al . , 2015 ) . We hypothesize that non-attached ants interact with the scent marks and that this affects their pulling direction once they attach to the load . Under this hypothesis , the response of the carrying group to the scent marks is not direct but , rather , mediated by these newly attached ants . Verifying this hypothesis is the subject of future work . To summarize , cooperative transport is an experimentally tangible phenomenon that allowed us to probe the inherent conflicts between the different organizational scales . These conflicts lead to persisting errors in the navigational instructions that individual ants provide to the group . We have shown how random following behavior with respect to short marking trails works to resolve these conflicts and allow for efficient collective performance .
Data was collected from six ant nests in Rehovot , Israel . Tests were carried out , in the field , during the summer when P . longicornis ants display cooperative transport behavior . In general , a 100 cm × 70 cm laminated paper sheet was placed near the ants’ nest on which either a Cheerio or a 1 . 5 mm thick ring-shaped piece of silicon ( incubated over-night in either Royal canin , Aimargues , France , or Happycat , Wehringen , Germany , brands cat food ) was introduced . In some trials obstacles were introduced between the location of the food and the nest entrance . Perspex obstacles of width 20 , 60 , or 80 cm were coated with Fluon ( Sorpol ) to prevent ants from climbing over them , and had a 5 mm slit at their center such that ants could easily pass but the object cannot . Side-view movies were filmed as the ants were transporting a load while walking on an elevated surface covered by millimeter paper . Two cameras , filming at the same frame rate , were placed above and to the side of the load's trajectory and were manually synchronized according to the ants locations on the millimeter paper . A scenario in which ants do not hold information about the direction to the nest was created by gently picking up the load together with the ants that were connected to it and laying it on a 40 cm × 30 cm clean paper sheet , with no freely moving ants . The lack of directional information was evident by the load’s random walk motion rather than the ballistic motion observed without the manipulation ( for details see Gelblum et al . , 2015 ) . The data were filmed using either a Canon EOS 550D camera or a Panasonic HC-VX870K camcorder . A designated image processing program , developed using MATLAB programming language and described in our previous study ( Gelblum et al . , 2015 ) , enabled high-quality tracking of the ants' location as well as the location and orientation of the transported load . We developed a method that allows us to pinpoint the pheromone marks laid by individual ants by using their velocity profile . Typically , the speed signature of a marking event lasts about 0 . 2 s and involves a sharp deceleration followed by high acceleration phase ( Figure 1B and C and Video 3 ) . Candidate marking events were found by screening the speed profiles of the non-carrying ants with a time window of 0 . 25 s . The selection criteria included the existence of a local speed minimum lower than 3 . 5 cm/s and absolute value of deceleration/acceleration around the minimum that is greater than 30 cm/s2 indicating a sharp stop-and-go episode . These criteria capture more than 97% of the true marking events as well as multiple false positives ( non-marking episodes ) whose percentage vary considerably due , for example , to variations in the number of non-marking ants present in the vicinity of the load . To rid the dataset of these false hits , the automated stage was followed by a second manual stage . Here , suspect marking events automatically extracted from the speed profile , were further examined by a human observer who reviewed the corresponding raw video footage , where more subtle typical movements ( e . g . , slight distortion of the gaster shape , ant shadow location relative to the gaster , and the speed reversal described in Figure 1C ) are visible ( see close-up visualization in Video 3 ) . This allowed differentiating true marking events from other stopping behavior showing similar velocity profiles . To test the reliability of the entire marking detection procedure , we simultaneously filmed ants from both the top and side view ( Figure 1—figure supplement 2 and Video 2 ) . Marking behavior was detected , as described above , from the top-view video and then verified against the corresponding side view video . We find that our method is highly reliable ( Identification rates are 93% of true positives and a positive likelihood ratio of 6 . 75 ) . Trail pheromone samples were collected in the ants’ natural environment at the same site used for obstacle bypass assays . The pheromones were collected on thin layer chromatography ( TLC ) plates of silica modified with covalently bonded octadecyl on a glass support ( Analtech , Newark , DE ) . The plates were precut to a size of 5 × 20 cm2 and thoroughly cleaned using ethyl acetate , hexane and acetone . The plates were stored in a closed polypropylene box until arrival to the site . Pheromone samples were collected by placing a TLC plate between the nest and an immobilized object which the ants attempted to carry to their nest . The ants were allowed to freely mark the plates for 5 min after which the plates were placed on dry ice to minimize pheromone evaporation . Control experiments were done by creating a scenario were ants walk on the TLC plates without marking it with trail pheromones . To do this , TLC plates were placed on poles such that ants could not reach them and they could not form a trail to this area . Ants were continuously collected from the floor and placed on the TLC plates for 10 min after which the plates were placed on dry ice . To prepare the samples for analysis , silica was scraped of the surface of the glass and immediately transferred to clean glass vials containing 3 ml hexane . The samples were sonicated for 20 min after which the supernatant was collected and transferred to new vials . Excess solvent was evaporated under a nitrogen stream to a total volume of 70 µl . Samples were analyzed on a 7890 Agilent gas chromatograph coupled to time of flight mass spectrometer equipped with a Gerstel cooled injection inlet and a fused silica column ( DB5-MS , 30 m × 0 . 25 mm , Agilent ) and compared against a linear hydrocarbon mix standard . The large volume injection mode was used to increase sensitivity with an injected volume of 50 µl . Inlet temperature was set to −21°C , the vent flow was set to 260 mL/min at 7 . 5 PSI and the injection speed was set to 1 . 04 µl/sec . The inlet temperature was kept for 1 min after which it was heated to 260°C at 720 degrees/min . The oven program started at 30 where it stayed for 3 min after which it was raised to 310°C at eight degrees/min where it stayed for 10 min . The instrument was operated at constant flow of 1 ml/min . We estimated the amount of information that a single scent mark conveys about the direction between the load and the nest using 1395 scent marks recorded over several meters of collective motion , and gathered in several different occasions . We considered each scent mark at the moment it appears . We measured the angle between the line that connects the center of the load to the nest and the line that connects the center of the load to the scent mark . The resulting collection of angles can be regarded as messages conveyed by the ants to the load regarding the direction to the nest . The uncertainty in these messages was quantified by grouping these measurements into 36 bins of 10o each ( bottom panel of Figure 2A ) and estimating the entropy of the resulting normalized probability histogram , pi: Habsolute=−∑pi⋅log2 ( pi ) , where the sum is taken over all 36 bins . This uncertainty could be compared to the case in which no message was received such that any of the 36 possible directions is equally probable: Hnomessage=−∑136⋅log2 ( 136 ) =log2 ( 36 ) . The average information , in bits per mark , is estimated as the reduction of entropy between these two cases: Iabsolute=Hnomessage−Habsolute . We used the same dataset to estimate the amount of directional information that the load gains , on average , from a single mark . To do this , we compared two probability histograms . The first , pibefore , is a histogram of the angles , θbefore , between the line connecting the center of the load to the scent mark ( at the time of marking ) and the tangent to the load’s trajectory in the two seconds that precede the appearance of the mark ( bottom panel of Figure 2B ) . The second histogram , piafter , is similar but is calculated over angles , θafter , between the line connecting the center of the load to the scent mark and the tangent to the load’s trajectory in the 2 s that follow the marking event ( bottom panel of Figure 2C ) . While both of these histograms are centered around zero , the second one is narrower . This indicates that the object adjusts its direction of motion by turning towards the mark in the two seconds that follow its appearance . To quantify this reduction in histogram width we calculate the entropies of the distributions before and after the marking: Hbefore/after=−∑pibefore/after⋅log2 ( pibefore/after ) . The amount of information , in average bits per mark , that is gained by the load from each such event was estimated by the reduction in entropy in the seconds that follow the marking event: Iload=Hbefore−Hafter . To measure the distance of the first mark in a bout from the object we used the video footage to identify N = 735 marking bouts and measure the distance of the first mark from the center of the load . We observed that marking bout distances can vary between several meters ( e . g . for the recruiting ant ) and several centimeters ( Figure 3D ) . To measure the distribution of bout lengths , we used a Disto-D3 distance meter ( Leica Geosystems; Heerbrugg , Switzerland ) . Ants were induced to carry an object at distances of 5–7 m from the nest during which we identified ants that commence marking near the object ( this is justified by our measurement indicating that marking bouts commence in the close proximity of the carried load Figure 3—figure supplement 2 , ) . A person holding the distance meter marked the location of the object at the beginning of the bout using a small marker while a second person tracked the marking ant . Once the ant marks the furthest mark from the load , a distance measurement was taken . Note that using video techniques to track fast moving 3 mm ants over many meters in the field is impractical . Simulation of routing in grids and lines was performed under the following conditions . The network was obtained by placing N uniformly-spaced nodes in the form of a grid of dimensions √N × √N , or in the form of a line of length N , respectively . For the studied graphs , we considered an unreliable advice model in which each node with probability p had its advice chosen to point along a shortest path towards the target ( with ties between different possible shortest paths on the grid broken so that the selected paths towards the target always formed a tree structure ) , whereas with probability ( 1-p ) this advice was chosen to point to a neighboring node , chosen uniformly at random . The routing process was subsequently performed without any updates to the advice . A routing protocol which performs shortest path routing solely by following the advice would therefore be likely to be stuck at some intermediary node of the considered environment . All simulation results were obtained through a Monte Carlo simulation . For the grid , we tested the performance of the PF algorithm for different values of probability of following advice in the range [0 , 1] and a fixed number of nodes N = 50 × 50 of the network . Each data point corresponds to a random sample of 1000 graphs under the studied model of unreliable advice , in which the routing process was repeated 50 times for different source nodes , assuming that a node located next to the center of the grid was the target . For the line , we varied the number of nodes in the range N ∈[2 200] , starting from the leftmost node and routing towards the rightmost node . For each length of the line , we sampled between 104 and 106 advice configurations under the studied advice model . The quality of a specific execution of the PF routing process between a source-target pair in a graph is represented by its stretch , defined as the ratio between the number of hops traversed when performing PF routing and the number of hops of the actual shortest source-target path in the updated graph . For the grid , the aggregated parameter presented in Figure 4B–C , called effective stretch , represents the average value of stretch in each scenario , taken over 90% of the obtained data-set after discarding executions with the largest stretch . Such a definition of effective stretch ensures the meaningfulness of simulation results , considering that the expected value of the studied random variables may , in some settings , be potentially unbounded . In Figure 4D , this effective stretch parameter is subsequently normalized with respect to the effective stretch of the optimal following probability . For the line , the parameter presented in Figure 4A is the median stretch over all runs . Box plots representations ( Figure 1D ) denote the median values inter-quartile range ( box ) and the lowest datum still within 1 . 5 times the inter-quartile range ( whiskers ) . Error bars for all information content measurements were estimated using a bootstrap method ( technical replication ) . For each non-normalized histogram , we generated 10 , 000 noisy histograms in which each bin i takes the value qi’=qi + βi·√qi where qi is the value of the bin in the corresponding original histogram and βi is a random noise term with values in the interval [1 , 1] . These histograms were then normalized to obtain probability histograms ( pi ) from which were used to generate a distribution of entropies ( using the exact same procedure as described in the ‘Calculation of scent marks information content’ section ) . This procedure ensures that bins with more measurements are associated with smaller errors in the estimation of the corresponding probabilities . The standard deviations of these distributions were taken as the errors of the entropy measurements , and thus of the information . Error bars in Figure 3C signify the SEM ( Standard error of the mean ) using , for each point , a sample size that equals the number of measurements it includes and assuming , again for each bin , a binomial distribution with a success probability corresponding to the y-axis measurement . The SEM is estimated as the standard deviation of this for this binomial distribution divided by the square root of the sample size . The probability that the marking bout distances of the first recruiter ant can be considered as samples of the distribution characterizing the locally-blazed ant trail as depicted in Figure 3D was calculated by the probability of obtaining twelve independent samples of over six meters from this distribution . The probability of obtaining a single such sample is the number of measurements of six meters and over the total number of measurements which is p=16/735 ≈0 . 022 . The probability of twelve consecutive such measurements is p12 < 10−6 . | Ants forage to find food and bring it back to the colony . If they come across food items that are too large or heavy for a single individual to carry , some species are able to form teams to cooperatively carry these items to the nest . This collective navigation process hinges on the navigational abilities of the individual ants . However , in natural terrains , the routes that are available to an individual ant are often inaccessible for a large group carrying a bulky item . So how do the ants manage to navigate together ? Fonio et al . studied how longhorn crazy ants cooperate to move large items . The experiments show that nearby ants not currently engaged in carrying the item mark the ground with chemical scents . Fonio et al . devised an automated method of detecting scent marking events and this has provided some of the first real time documentation of ant scent trails as they form . This shows that when cooperating to move large objects , the ants use scent marks to form a new type of trail that is highly dynamic . Unlike other ant trails that mark the whole path between the food and the nest , these new trails only direct the next step of the movement . Furthermore , the team of ants carrying the item only follows these local directions in a loose manner and often ignores them . Fonio et al . then used a mathematical model and further experiments to show that this new type of trail effectively solves the problems of collective navigation during cooperative transport . Essentially , the locality of the trail and the loose way in which the group follows it tune the degree to which the collective motion depends on the directions provided by individual ants . This allows the group to benefit from the useful information available to individuals while avoiding local traps that may occur when these individuals wrongly direct them towards dead ends . The next step following on from this work is to understand the mechanisms behind this newly discovered trail , and in particular , understand how the collective motion results from the actions of individual ants that react to single drops of scent . Another challenge for future research would be to find technological applications for this newly discovered strategy , such as routing over communication networks . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"ecology",
"computational",
"and",
"systems",
"biology"
] | 2016 | A locally-blazed ant trail achieves efficient collective navigation despite limited information |
Categorization has been associated with distributed networks of the primate brain , including the prefrontal cortex ( PFC ) and posterior parietal cortex ( PPC ) . Although category-selective spiking in PFC and PPC has been established , the frequency-dependent dynamic interactions of frontoparietal networks are largely unexplored . We trained monkeys to perform a delayed-match-to-spatial-category task while recording spikes and local field potentials from the PFC and PPC with multiple electrodes . We found category-selective beta- and delta-band synchrony between and within the areas . However , in addition to the categories , delta synchrony and spiking activity also reflected irrelevant stimulus dimensions . By contrast , beta synchrony only conveyed information about the task-relevant categories . Further , category-selective PFC neurons were synchronized with PPC beta oscillations , while neurons that carried irrelevant information were not . These results suggest that long-range beta-band synchrony could act as a filter that only supports neural representations of the variables relevant to the task at hand .
Executive brain functions are closely linked with a network of cortical areas in the prefrontal and posterior parietal cortices . Human imaging studies have shown widespread frontal and parietal cortex activation during a broad range of cognitive demands ( Duncan , 2010; Fedorenko et al . , 2013 ) . Likewise , there is an intermixing of neurons throughout frontal and parietal cortex that have neural correlates of different cognitive functions: attention , working memory , decision-making , rule-coding , and categorization ( Andersen and Cui , 2009; Buschman and Miller , 2007; Crowe et al . , 2013; Freedman and Assad , 2006; Goodwin et al . , 2012; Hayden and Pasternak , 2013; Hussar and Pasternak , 2009; Jacob and Nieder , 2014; Merchant et al . , 2011; Rishel et al . , 2013; Salazar et al . , 2012; Vallentin et al . , 2012 ) . These two regions not only have remarkable similarity in their patterns of neural activity , but they are also functionally interdependent . Deactivating one decreases neural activity in the other ( Chafee and Goldman-Rakic , 2000 ) and the temporal dynamics of neural information in one can mirror that of the other ( Chafee and Goldman-Rakic , 1998; Crowe et al . , 2013; Merchant et al . , 2011 ) . Thus , it is becoming increasingly clear that an understanding of executive functions will depend on further insight into how the circuits within and between prefrontal and parietal cortices interact . Insight can be gained by examining synchrony in the rhythms of their activity , which can reflect network properties ( Miller and Buschman , 2013; Roberts et al . , 2013; Thut et al . , 2012 ) . There is already evidence that frontoparietal rhythmic synchrony underlies at least two cognitive functions . Frequency-dependent increases in their synchrony have been seen during shifts of attention ( Buschman and Miller , 2007 ) and during visual working memory ( Salazar et al . , 2012 ) , functions long associated with both regions . More recently , single-neuron studies indicate that both areas also make a major contribution to visual categorization ( Antzoulatos and Miller , 2011; Cromer et al . , 2010; Crowe et al . , 2013; Freedman et al . , 2001; Freedman and Assad , 2006; Goodwin et al . , 2012; Rishel et al . , 2013; Roy et al . , 2010; Swaminathan and Freedman , 2012; Tsutsui et al . , 2016 ) . Individual neurons in both areas carve up different types of sensory information ( i . e . , motion , shape , location ) according to learned category boundaries . Further , the category information carried by neuron spiking in one area evolves in lock-step with the other ( Crowe et al . , 2013; Goodwin et al . , 2012 ) . Yet , rhythmic synchrony during visual categorization has not been examined . Thus , we recorded local field potentials ( LFPs ) from both regions while monkeys made decisions about abstract spatial categories . We used a delayed-match-to-category task in which monkeys were required to categorize visual cues along an abstract spatial dimension and then make a match/non-match decision about whether subsequent test stimuli belonged in the same spatial category as the sample cue . This type of match/non-match category decision has been shown to engage PFC and PPC neurons ( Cromer et al . , 2010; Crowe et al . , 2013; Freedman et al . , 2001; Goodwin et al . , 2012; Roy et al . , 2010; Swaminathan and Freedman , 2012 ) . We found increases in synchrony between and within frontoparietal networks that depended on whether monkeys categorized a stimulus as ‘Above' or 'Below’ . Moreover , we found that spiking activity that reflected the category distinction was better synchronized to frontoparietal LFP category networks than spiking that encoded information irrelevant to the task .
Two rhesus macaque monkeys were trained to perform a Go/No-go category matching task . The trial began when they held a bar and achieved fixation of a central target ( Figure 1A ) . The initial Above vs . Below category boundaries were indicated by two horizontal lines on each side of fixation , which divided the top from bottom halves of the screen . After the monkeys initiated the trial , the boundaries disappeared and a sample cue appeared briefly at a location chosen randomly from a set of 16 locations ( ± a small jitter ) on the right or left side of the screen ( Figure 1B; 144 possible locations in total ) . After a 500 ms post-sample delay , the two horizontal lines reappeared at shifted positions . On half of the trials , the lines shifted clockwise ( the left shifted up and the right shifted down by 4 degrees ) . On the other half of the trials , they shifted counter clockwise ( the opposite ) . After an additional 1 s delay , a test stimulus appeared randomly at one of 624 possible locations ( Figure 1B ) . This test stimulus was a category match if it appeared on the same side of the ( shifted ) horizontal line ( Above or Below ) as the sample had appeared relative to the original horizontal line , irrespective of its exact location . If the test stimulus was a match , the monkey had to release the bar within 1 s ( Go ) to receive a juice reward . If the test was a non-match , the animals continued holding the bar ( No-go ) until another test stimulus ( always a match ) appeared on the screen . Note that the category of the initial sample cue did not change with the boundary shift . The boundary shift only changed which test stimulus locations would qualify as a category match or non-match . Thus , the purpose of the boundary shift was to ensure that the monkeys were encoding an abstract category representation , one that was not solely based on retinotopic coordinates . The monkeys also could not make a decision based solely on the proximity between sample and test stimuli because they could often be quite far apart ( e . g . , on the right and left of the screen; see Figure 1A ) , yet still belong to the same category . Both monkeys performed the task quite well , with 92 . 77% of all trials ( excluding fixation breaks ) correct . 10 . 7554/eLife . 17822 . 003Figure 1 . The task design . ( A ) The trial started when the animal held a bar and maintained visual fixation on the red central dot for 1 s . During fixation , in order to help the animal re-calibrate the Above/Below categorization boundary after the last trial , the 2 hemi-boundaries were also displayed . The sample stimulus appeared either above or below the horizontal meridian . To reduce the possibility of perceptual binding between the sample stimulus and the horizontal boundaries , there was a brief , 200 ms time gap between the two displays ( pre-sample epoch ) . After the post-sample delay epoch , 2 hemi-boundaries were displayed at new locations , according to a CW ( as shown ) or CCW shift . The animal had to adjust the decision criterion based on the location of the new boundaries . When a test stimulus appeared , it had to be compared to the sample stimulus in a boundary-referenced , rather than a retinotopic , spatial frame . If a non-match , the first test would be followed by a second test , always a match to the sample category . Until display of a match test , the animal had to maintain visual fixation and contact with the bar . Upon a match test display , the animal had to release the bar for liquid reward . ( B ) Sample stimuli could appear randomly at any one of the 144 locations shown ( top ) , with equal probability in the category Above or Below ( DVA: degrees of visual angle ) . Test stimuli ( equal probability for categories above and below the new boundaries ) could appear at any one of the 624 locations shown ( bottom ) . The continuous red horizontal lines in the test space indicate the position of hemi-boundaries after CW shift , and the dashed lines ( dashed only for illustration ) indicate the CCW shift . Each trial would test one of 89 , 856 possible sample-test combinations , equally distributed between match and non-match types . After the boundary shift , the Above/Below categories spanned a visual area that overlapped with the corresponding area of the pre-shift categories by 50% . ( C ) The probability to categorize a sample stimulus as ‘Above' , 'Below' , 'Left’ , or 'Right’ , by spatial location . The animals would choose Left or Right with equal probability , because the Left/Right dimension was task-irrelevant . See also Figure 1—figure supplement 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 17822 . 00310 . 7554/eLife . 17822 . 004Figure 1—figure supplement 1 . Performance on the categorization task . The animals’ performance on the categorization task remained above 85% correct , regardless of the eccentricity of the sample stimuli ( A ) , their distance from the vertical and horizontal meridians ( B ) , the horizontal or vertical hemifield of their display ( C ) , or the visual quadrant ( D ) . Both the monkeys’ accuracy ( E ) and reaction time ( F ) were similar across a range of distance between the sample and test stimuli . ( G ) For those samples that were proximal to the boundary ( at y = 2 or −2 degrees ) , accuracy remained above 85% even when they flipped to the other side of the new category boundaries ( as in the example shown in Figure 1A ) . DOI: http://dx . doi . org/10 . 7554/eLife . 17822 . 004 To further quantify the monkeys' behavior , we calculated the probability of categorizing a sample as Above . This ( PABOVE ) was defined as the probability of a Go response to a test stimulus above the shifted boundary lines or a No-go response to test stimuli below the shifted lines ( the probability of categorizing a sample as Below was complementary , PBELOW = 1 – PABOVE ) . This allowed us to quantify the monkeys’ category judgments as a function of the exact location of the sample cue . Figure 1C indicates that the monkeys’ categorization of Above ( or Below ) was flat across the wide range of sample locations ( top ) . Overall , PABOVE was at least 0 . 83 if the sample was also Above ( above the horizontal meridian ) and at most 0 . 13 if the sample was actually Below . We further demonstrated the monkeys’ ability to generalize within the same category and distinguish between the categories in several ways . First , because the sample and test stimuli could appear in separate visual hemifields , yet still belong in the same category , we computed probability of the monkeys mistakenly categorizing the sample and test as being in different categories if they appeared in opposite hemifields and categorizing them as being from the same category just because they appeared in the same hemifield ( i . e . , categorization based on Left vs . Right ) . Thus , just as we did for Above vs . Below , we computed the probability of the monkeys categorizing the sample as Left . This ranged from 0 . 39 to 0 . 57 ( Figure 1C ) , which is what we would expect if the monkeys were disregarding Left vs . Right ( irrelevant dimension ) and categorizing by Above vs . Below ( relevant dimension ) . Thus , the monkeys were generalizing across diverse locations within the same category and not mistakenly categorizing the sample and test as being in different categories simply because they were far apart . Likewise , Above vs . Below category performance remained high ( at least 85% correct ) , regardless of sample cue eccentricity , sample distance from the horizontal and vertical meridians , the visual quadrant or hemifield of sample display , and the distance between the sample and test stimuli ( Figure 1—figure supplement 1 ) . Finally , performance remained high ( at least 85% correct ) on trials in which the retinotopic location that was occupied by the sample stimulus changed categories after the shift in boundaries and that location became a category non-match ( Figure 1—figure supplement 1 ) . Thus , the monkeys’ behavior showed the hallmark of categorization: sharp discrimination across the boundary and flat generalization on either side of a boundary . The monkeys maintained a category representation that was uncoupled from the particular spatial location of the sample stimulus . We recorded from multiple electrodes simultaneously advanced in the lateral PFC ( lPFC , areas 46 and 45 ) , caudal PFC ( cPFC , area 8A and the FEF ) , and the anterior intraparietal area ( AIP; see Figure 2—figure supplement 1 ) . The AIP subregion of the posterior parietal cortex was selected because it is strongly coupled to premotor areas and plays a role in visuospatial transformations that guide motor movements ( Verhoef et al . , 2015 ) . Data from subareas of the lPFC ( as well as subareas of the cPFC ) were combined because they yielded similar results . To examine interactions between the areas , we analyzed simultaneously recorded LFPs from a total of 159 electrodes in AIP , 111 electrodes in cPFC , and 129 electrodes in lPFC . From the LFP we first removed stimulus-evoked components ( to isolate the task-induced oscillations ) , and decomposed it in the frequency domain using the Morlet wavelet . The LFP was decomposed in six octaves ( from 2–128 Hz ) , at a resolution of 0 . 1 octave . From the amplitude of the wave components we then computed the LFP time- and frequency-dependent power and normalized it to 1/frequency ( in order to correct the power-law decay of the LFP power spectrum ) . Removing the LFP components that are phase-locked to the stimulus may conceal high-frequency rhythms time-locked to stimulus presentation , but it is necessary to avoid spurious synchrony due to simultaneous stimulus-evoked responses in the LFPs . This analysis revealed elevated power of the 16–32 Hz octave ( beta band ) in all three areas ( Figure 2A ) . Beta-band power was especially prominent at the end of the sample presentation and in the early part of the immediately following delay . This is around the time that we would expect the monkey to categorize the location of the sample . Beta power was also evident just before presentation of the test stimulus , after the boundaries switched . 10 . 7554/eLife . 17822 . 005Figure 2 . Brain rhythms of the frontoparietal network . ( A ) Average power ( mV2 ) of the LFPs recorded from AIP ( left ) , cPFC ( middle ) , and lPFC ( right ) electrodes in time-frequency space . Strong oscillations are seen at the beta band ( 16–32 Hz; dotted rectangle in all panels ) in all three areas . For these analyses , the evoked component has been removed from the LFPs , and the power has been normalized to 1/f . Spectral analyses were performed at six octaves ( 2–128 Hz ) , at a 0 . 1 octave resolution ( i . e . , 10 frequency bins per octave ) . Vertical white lines demarcate the Category and Shift epochs , and dashed white lines indicate the end of sample display . ( B ) Strong synchrony ( Pairwise Phase Consistency ) of beta-band oscillations ( 16–32 Hz ) is observed across pairs of electrodes within AIP ( left ) , cPFC ( middle ) and lPFC ( right ) . ( C ) As also seen with power ( A ) and intrinsic synchrony in the three areas ( B ) , strong synchrony of beta-band rhythms is also seen in the extrinsic pairs of electrodes across the three areas: AIP-cPFC ( left ) , cPFC-lPFC ( middle ) , and AIP-lPFC ( right ) . Recording sites appear in Figure 2—figure supplement 1 . The corresponding figure before removal of the evoked component appears in Figure 2—figure supplement 3 and the cross-area synchrony for match vs . non-match trials appears in Figure 2—figure supplement 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 17822 . 00510 . 7554/eLife . 17822 . 006Figure 2—figure supplement 1 . Electrophysiological recording sites . The MRI-guided recording sites , projected to a single sagittal ( left ) and a single horizontal ( right ) slice from animal P ( top ) and animal G ( bottom ) . The intraparietal sulcus ( IPS ) , arcuate sulcus ( AS ) and principal sulcus ( PS ) are indicated on the top left panel . The data from posterior parietal recordings came from the anterior intraparietal area , located on the rostral lateral bank of IPS ( area AIP , green marks ) . Data from prefrontal recordings were distinguished between the sites around the AS ( area 8A , white marks ) , which we label as caudal PFC ( cPFC ) in the text , and the area around the PS ( red marks ) , which included areas 46 and 45 and we label as lateral PFC ( lPFC ) in the text . DOI: http://dx . doi . org/10 . 7554/eLife . 17822 . 00610 . 7554/eLife . 17822 . 007Figure 2—figure supplement 2 . Match vs . non-match trials . Frontoparietal synchrony following the Shift epoch separately for match ( A ) and non-match trials ( B ) . Strong synchrony of low frequency rhythms was observed upon presentation of the test stimulus that matched the sample category ( at 2 . 3 s following sample onset for match trials and at 3 . 8 s for non-match trials ) . This effect may be related to the bar-release motor movement , the oculomotor movement that typically follows the release from required visual fixation , or the reward . Vertical lines demarcate the onset of the test stimulus display ( 2 . 3 s ) , the delay interval following ( at 3 . 3 s ) , and the onset of the second test at non-match trials ( 3 . 8 s ) . Dotted vertical lines mark the average reaction time in match ( 299 . 06 ms ) and non-match trials ( 181 . 86 ms ) , relative to the display onset of the corresponding match test stimulus . DOI: http://dx . doi . org/10 . 7554/eLife . 17822 . 00710 . 7554/eLife . 17822 . 008Figure 2—figure supplement 3 . Power and synchrony of LFPs before removal of the evoked component . The results are very similar to those of Figure 2 , except for some stimulus-locked effects at low frequencies . DOI: http://dx . doi . org/10 . 7554/eLife . 17822 . 008 To compare beta power across areas and time , we divided the trial into two main epochs . The Category epoch lasted from the sample display onset up to the boundary shift ( time 0–1 . 3 s ) . This is presumably when the monkeys categorized the sample as Above or Below . The Shift epoch started after the boundary shift and lasted up to the test stimulus display ( time 1 . 3–2 . 3 s ) . At this time , the monkeys had to maintain in working memory the category of the sample despite the fact that the retinotopic location at which the sample had appeared might have become a category non-match after the boundary shift . A 2-way ANOVA compared beta-band power among the three areas and two trial epochs . As suggested by Figure 2A , there was significantly greater power during the Category epoch ( average: 0 . 06 mV2 ± SEM: 0 . 002 ) than the Shift epoch ( 0 . 04 ± 0 . 002; p=4 . 7×10–10 ) . The strongest beta power appeared in AIP , followed by cPFC ( p=3 . 4×10−23; posthoc comparisons: AIP ( 0 . 065 ± 0 . 002 ) > cPFC ( 0 . 05 ± 0 . 002 ) , p=7×10–10; cPFC > lPFC ( 0 . 03 ± 0 . 002 ) , p=2 . 6×10–3; AIP>lPFC , p=4 . 7×10–12 ) . We next compared LFP synchrony within and between areas using the pairwise phase consistency metric ( PPC ) . Synchrony was bias-corrected by subtracting the chance-level PPC , as estimated from a surrogate dataset . For every pair of electrodes , the surrogate dataset was created by randomly shuffling the trials of each electrode independently of the other electrode 200 times without replacement . The PPC that was computed across these random permutations was averaged and provided the baseline-level synchrony that would be expected by chance . The pattern of synchrony largely paralleled the spectral power results . Beta band synchrony was evident within the areas ( Figure 2B; AIP-AIP: n = 340 pairs of electrodes; cPFC-cPFC: n = 205 pairs; lPFC-lPFC: n = 290 pairs ) and between areas ( Figure 2C; AIP-cPFC: n = 474 pairs; cPFC-lPFC: n = 323 pairs; AIP-lPFC: n = 535 pairs ) . A 2-way ANOVA compared synchrony as a function of area pairing and trial epoch . Like the power analysis , there was significantly greater beta synchrony in the Category epoch ( 0 . 09 ± 0 . 002 ) than the Shift epoch ( 0 . 07 ± 0 . 002; p=2 . 7×10–10 ) . There were also significant differences in synchrony between areas . The strongest was within AIP and weakest between AIP and lPFC . ( Posthoc comparisons , corrected p<0 . 01: AIP-AIP ( 0 . 18 ± 0 . 003 ) > cPFC-cPFC ( 0 . 13 ± 0 . 004 ) > lPFC-lPFC ( 0 . 08 ± 0 . 003 ) > cPFC-lPFC ( 0 . 04 ± 0 . 003 ) > AIP-lPFC ( 0 . 02 ± 0 . 003 ) ; AIP-cPFC ( 0 . 03 ± 0 . 003 ) not different from AIP-lPFC and cPFC-lPFC ) . The low-frequency synchrony that was observed late in the trial , during the test stimulus ( Figure 2B and C ) , may be related to the motor response to a category match ( in 50% of the trials ) , the reward that followed correct trials , or the eye movement that typically follows the release from visual fixation requirement . Figure 2—figure supplement 2 illustrates the cross-area synchrony during the Shift and Test epochs separately for match and non-match trials , indicating that the strong low-frequency synchrony coincided with the animals’ response . Previous studies have shown rule and category-selective beta synchrony within the PFC and between the PFC and striatum ( Antzoulatos and Miller , 2014; Buschman et al . , 2012 ) , wherein different pairs of electrodes showed increased beta synchrony for one or the other category/rule . We found the same results for frontoparietal synchrony and spatial categories . An example of category-selective synchrony of beta oscillations between the frontal and parietal cortex from one recording session is shown in Figure 3 . It shows the filtered beta synchrony between a cPFC electrode ( F09 ) and two different AIP electrodes ( P07 and P10 ) for trials of category Above vs . Below . The cPFC ( F09 ) electrode had stronger beta synchrony with one AIP electrode ( P07 ) for Above ( red and black traces , left ) but stronger beta synchrony with the other AIP electrode ( P10 ) for Below ( red and green traces , right ) . 10 . 7554/eLife . 17822 . 009Figure 3 . Example traces of beta rhythms with category-selective synchrony . During a trial of category Above ( left ) , a cPFC electrode ( F09; black trace ) displays stronger phase and amplitude coupling with one AIP electrode ( P07; red ) than with another ( P10; green ) . The strength of coupling among the same electrodes is reversed during a trial of category Below ( right ) . The cPFC electrode now couples more weakly with the P07 than with the P10 electrode . Scale bars: 0 . 4 mV and 100 ms . DOI: http://dx . doi . org/10 . 7554/eLife . 17822 . 009 We quantified the degree of LFP synchrony for Above vs . Below trials using pairwise phase consistency ( PPC ) , a measure of synchrony that is corrected for small and unequal sets of trials ( Vinck et al . , 2010 ) . The absolute difference in PPC between the trials of the two categories was transformed into a z-score based on 200 random permutations of the trials between the two categories . This revealed category-selective synchrony for Above/Below categories both between and within the areas ( Figure 4A and C respectively ) , predominantly in two octaves: 16–32 Hz ( beta ) , and 2–4 Hz ( delta ) . The contours ( black lines ) in Figure 4 indicate regions of time/frequency space where category selectivity was significantly greater than zero ( p<0 . 05 at each bin for at least 60 ms and three consecutive frequencies ) . Category-selective synchrony between AIP and cPFC was mostly concentrated in the beta band during the Category epoch and in the delta band during the Shift epoch ( Figure 4A left panel ) . Category selective synchrony between cPFC and lPFC electrodes ( Figure 4A middle panel ) was similar . By contrast , lPFC and AIP category selective synchrony was mostly in the low-frequency oscillations ( Figure 4A right panel ) . Within AIP , category-selective synchrony was longer-lasting and covered a wider range of frequencies ( Figure 4C ) . A similar pattern of results was found using the percentage of electrode pairs with category selective synchrony as a measure , instead of the average strength ( Figure 4—figure supplement 1 ) . In addition , the observed category-selective synchrony did not arise from a difference in the pattern of network dynamics between the categories . Both the synchrony between and within the frontal and parietal areas displayed similar time-frequency dependence for the category Above as for the Below ( Figure 4—figure supplement 2 ) . 10 . 7554/eLife . 17822 . 010Figure 4 . Category-selective synchrony of frontoparietal rhythms . ( A ) and ( B ) Population average of PPC selectivity of all simultaneously recorded pairs of AIP and cPFC ( left ) , cPFC and lPFC ( middle ) , lPFC and AIP ( right ) electrodes along the task-relevant , Above/Below , dimension ( A ) and task-irrelevant , Right/Left , dimension ( B ) . ( C ) and ( D ) PPC selectivity of intrinsic pairs of electrodes within AIP ( left ) , cPFC ( middle ) and lPFC ( right ) along the Above/Below ( C ) and Right/Left ( D ) dimensions . In all panels , contours identify regions of time-frequency space with statistically significant population selectivity . Similar time-frequency dependent patterns were seen in the percent of sites with significantly selective synchrony ( see Figure 4—figure supplement 1 ) . The time-frequency dynamics of PPC were similar for both the Above and Below categories ( see Figure 4—figure supplement 2 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 17822 . 01010 . 7554/eLife . 17822 . 011Figure 4—figure supplement 1 . Prevalence of category-selective synchrony of frontoparietal rhythms . Same as Figure 4 , but for the percent of pairs of electrodes with statistically significant category selectivity in PPC . DOI: http://dx . doi . org/10 . 7554/eLife . 17822 . 01110 . 7554/eLife . 17822 . 012Figure 4—figure supplement 2 . Category-specific synchrony of oscillations in the frontoparietal network . The time-frequency dynamics of synchrony are very similar for both the Above and the Below categories in the between-area pairs of electrodes ( A and B ) as well as the within-area pairs of electrodes ( C and D ) . DOI: http://dx . doi . org/10 . 7554/eLife . 17822 . 01210 . 7554/eLife . 17822 . 013Figure 4—figure supplement 3 . Frontoparietal synchrony for preferred/non-preferred categories vs . pre-trial baseline . Average PPC between AIP and cPFC electrodes in the beta band ( left ) and the delta band ( right ) during the Category epoch ( top ) and the Shift epoch ( bottom ) , during the pre-trial baseline and the corresponding trial epoch . Black lines correspond to trials of the preferred category ( Above or Below ) , and red lines correspond to trials of the non-preferred category . Each comparison to baseline was statistically significant ( p<0 . 001; 2-tailed t-test for paired samples ) . DOI: http://dx . doi . org/10 . 7554/eLife . 17822 . 013 We examined whether the category selectivity of frontoparietal synchrony arose from a selective loss or gain of synchrony for the preferred/non-preferred categories relative to baseline . To address this , we averaged the synchrony observed in the beta and delta bands during the Category and Shift epochs , separately for trials of the Above and Below categories . Every electrode pair’s category preference was determined from the polarity of the difference Above-Below . For every pair of frontoparietal electrodes we compared the epoch- and frequency-specific synchrony to the corresponding synchrony observed during the fixation epoch ( the last 750 ms of the fixation window ) , separately for the preferred and non-preferred categories . Figure 4—figure supplement 3 shows results only for the AIP-cPFC pairs of electrodes because they showed overall stronger category-selective synchrony than the AIP-lPFC pairs ( Figure 4 ) . The directional changes from baseline were mixed ( although all were statistically significant ) , but the trend was for the beta-band synchrony during the Category epoch ( i . e . , the epoch that indicated stronger selectivity in beta synchrony; Figure 4 ) to increase from baseline for the preferred category and to decrease from baseline for the non-preferred category . The same trend was observed for the delta-band synchrony during the Shift epoch ( the epoch with stronger selectivity of delta synchrony ) . In the other two epochs ( when synchrony selectivity was weaker; Figure 4 ) synchrony of both the beta and delta rhythms was decreased from baseline , albeit to a smaller extent for their preferred than for their non-preferred category . It was possible that the category-selective synchrony described above did not reflect the Above vs . Below categories per se but , instead , differences in the retinotopic location of the sample cue that were unrelated to the categorization task . If so , we would expect to see differences in synchrony between sample cues appearing in the left vs . right hemifields . After all , receptive fields in both the frontal and parietal cortex are strongly biased to the contralateral field . However , when we re-sorted the trials based on whether the sample cue appeared in the left vs . right hemifields , there was virtually no selectivity in beta synchrony ( Figure 4B and C ) . Thus , category-selective synchrony mirrored the demands of the task: To encode the Above vs . Below category of the sample , irrespective of Right vs . Left . We collapsed the frequency and time dimensions by taking the average synchrony across each of the two trial epochs and two frequency bands , beta and delta . We performed a 4-way ANOVA on z-transformed PPC values ( area x epoch x category boundary x frequency band ) , which included six area combinations ( AIP-cPFC , cPFC-lPFC , lPFC-AIP , AIP-AIP , cPFC-cPFC , and lPFC-lPFC ) , two trial epochs ( Category and Shift ) , two types of category boundaries ( relevant: Above/Below and irrelevant: Right/Left ) , and two frequency bands ( beta and delta ) . This analysis confirmed what was seen in the time-frequency analysis ( Figure 4 ) . It indicated significant differences among the areas ( p=2 . 4×10–25 ) , significantly greater selectivity for the Above/Below than the Right/Left boundary ( p=1 . 97×10–168 ) , and significantly greater selectivity of beta synchrony during the Category epoch and of delta synchrony during the Shift epoch ( band x epoch interaction: p=1 . 2×10–32 ) . Figure 5 illustrates in more detail the interactions among the experimental variables area , band , category boundary , and trial epoch , as estimated from the ANOVA post-hoc comparisons . It plots the mean synchrony of each group along with its 95% confidence interval ( CI ) after the Bonferroni correction for multiple comparisons . As seen at the upper half of Figure 5 , both beta and delta rhythms displayed synchrony that was significantly selective for Above vs . Below ( i . e . , the corrected 95% CIs were greater than zero ) . None of the area combinations showed beta synchrony selective for Right vs . Left ( bottom quadrant of Figure 5 ) . Delta synchrony was not selective for Right vs . Left within any area but was selective in the cross-area interactions ( i . e . , AIP-cPFC , cPFC-lPFC , and AIP-lPFC ) and only during the Shift epoch . Thus , pure category-selective synchrony was restricted to the beta band . There was more delta selectivity for Above vs . Below but also some delta selectivity for Right vs . Left . 10 . 7554/eLife . 17822 . 014Figure 5 . Group averages of category-selective synchrony of LFPs . Filled dots represent group averages and horizontal lines the 95% confidence intervals ( CI ) after correction for multiple comparisons . The upper half of the graph illustrates selectivity for the task-relevant category boundary ( Above/Below ) and the lower half the selectivity for the task-irrelevant , Right/Left boundary . Within each boundary , the upper halves show the category-selective synchrony of delta rhythms and the lower halves the selective synchrony of beta rhythms . Within each quadrant of the graph , the rows correspond to the 2 trial epochs ( Category and Shift ) for each of the six combinations of the areas studied . Groups with no overlapping confidence intervals are significantly different ( e . g . , the Above/Below selectivity of AIP-AIP delta in the Shift epoch is significantly greater than the selectivity of most other groups ) . Category selectivity is significantly greater than zero ( dashed vertical line ) in all groups with green markers , and not significantly different from zero in groups with red markers ( i . e . , depending on whether a CI includes zero ) . Synchrony of beta rhythms displays significant selectivity for the Above/Below but not for the Right/Left boundary . In turn , synchrony of delta rhythms displays significant Above/Below selectivity , while Right/Left selectivity is area and epoch-dependent . The data of the figure can also be seen in Figure 5—source data 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 17822 . 01410 . 7554/eLife . 17822 . 015Figure 5—source data 1 . Group averages ( and SEM ) for the post-hoc comparisons plotted in Figure 5 . DOI: http://dx . doi . org/10 . 7554/eLife . 17822 . 015 From the same electrodes we isolated the spiking activity of 1078 single neurons ( 296 from cPFC , 387 from lPFC , and 395 from AIP ) . As previously reported ( Goodwin et al . , 2012 ) , spiking activity in these areas showed selectivity for the learned spatial categories ( Above vs . Below ) . This is illustrated in Figure 6A , which shows the average selectivity ( assessed as percent explained variance or PEV ) across the entire population of all isolated neurons from each of the three areas as a function of time . The colored bars at the top of the figure show when the PEV values of each area were significantly greater than zero ( t-test , corrected p<0 . 05 ) . Note that selectivity for the Above vs . Below categories was evident shortly after the onset of the sample cue and continued throughout the remainder of the trial , growing stronger in the delay following the sample . This figure also shows that category selectivity in spiking activity was strongest in cPFC , next in lPFC , and weakest in AIP ( effect of area from 2-way ANOVA: p=7 . 9×10–22; post-hoc comparisons with corrected p<0 . 05: cPFC ( 1 . 24 ± 0 . 07 ) > lPFC ( 0 . 62 ± 0 . 06 ) > AIP ( 0 . 39 ± 0 . 06 ) ) . Figure 6B shows similar results with an alternative measure , i . e . the percentage of neurons in each area that showed a significant effect of category as a function of time . 10 . 7554/eLife . 17822 . 016Figure 6 . Selectivity of neural spiking activity . ( A ) Average selectivity ( ±SEM ) of neural spiking for the task-relevant category of sample location ( Above/Below ) in each of the three neural populations as a function of time from sample onset . Horizontal bars indicate time bins of significant selectivity ( p<0 . 05 lasting for at least 100 ms ) in the corresponding populations ( blue: cPFC; green: AIP; red: lPFC ) . ( B ) Percent of each population with significant selectivity ( p<0 . 01 ) for the Above/Below categorization . ( C ) Same as A for the task-irrelevant dimension of sample location ( Right/Left ) . ( D ) Same as B for the Right/Left dimension . Overall , all three populations showed stronger selectivity for Above/Below than for Right/Left ( note difference in y scales between left and right panels ) , and different time courses between Right/Left selectivity ( which gradually decayed after sample display ) and Above/Below selectivity ( which tended to grow stronger after the end of sample display ) . The results are similar if selectivity is analyzed on the pooled spikes from all neurons isolated from each electrode ( see Figure 6—figure supplement 1 ) . Synchrony of LFP oscillations between two electrodes showed similar selectivity whether the corresponding spiking had the same or different category preferences ( see Figure 6—figure supplement 2 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 17822 . 01610 . 7554/eLife . 17822 . 017Figure 6—figure supplement 1 . Selectivity of pooled multi-unit activity . ( A ) Average selectivity ( ±SEM ) of pooled neural spiking for the task-relevant category of sample location ( Above/Below ) in each of the three neural populations as a function of time from sample onset . Horizontal bars indicate time bins of significant selectivity ( p<0 . 05 lasting at least 100 ms ) in the corresponding populations ( blue: cPFC; green: AIP; red: lPFC ) . ( B ) Percent of each population with significant selectivity ( p<0 . 01 ) for the Above/Below categorization . ( C ) Same as A for the task-irrelevant dimension of sample location ( Right/Left ) . ( D ) Same as B for the Right/Left dimension . Analyses of spiking selectivity result to similar patterns whether they are performed on a pooled spiking activity per electrode or on the spikes of individual neurons ( Figure 6 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 17822 . 01710 . 7554/eLife . 17822 . 018Figure 6—figure supplement 2 . LFP synchrony in relation to individual site preference . ( A ) Average PPC selectivity between pairs of electrodes in AIP , cPFC and lPFC that showed the same ( top ) or different ( bottom ) category preference in spiking activity along the Above/Below dimension . ( B ) The same for category preferences along the Right/Left dimension . DOI: http://dx . doi . org/10 . 7554/eLife . 17822 . 018 For our LFP analyses ( discussed above ) , the beta band was of particular interest because beta oscillations have been associated with a variety of cognitive functions and previous studies have shown category and rule selectivity in this band ( Antzoulatos and Miller , 2014; Buschman et al . , 2012 ) . As noted above , there was category-selective beta-band synchrony within and between each area . However , there was virtually no selective beta synchrony for the ( task-irrelevant ) Right vs . Left hemifield location of the sample cue . By contrast , the spiking of single neurons was sensitive to the Right vs . Left location of the sample . This is shown in Figure 6C and D . The average population activity of all three areas showed significant selectivity for Right vs . Left early on during sample display , albeit weaker than the learned Above vs . Below categories . Right vs . Left selectivity was strongest in the prefrontal cortex ( cPFC and lPFC ) and weak in AIP ( effect of area from 2-way ANOVA: p=3 . 1×10–8; post-hoc comparisons with corrected p<0 . 05: AIP ( 0 . 1 ± 0 . 02 ) < cPFC ( 0 . 22 ± 0 . 02 ) and lPFC ( 0 . 27 ± 0 . 02 ) , cPFC not different from lPFC ) . Thus , information about the task-irrelevant Right vs . Left location of the sample ‘leaked through’ in spiking activity but did not seem to do so for the beta synchrony ( but , as noted , did for the delta synchrony ) . As shown in Figure 6—figure supplement 1 , these analyses yielded similar results when performed at the level of population spiking activity ( i . e . , on the pooled spikes of all neurons isolated from each electrode , rather than the activity of each individual neuron ) . Another notable difference between the patterns of results for synchrony vs . spike rate was in the strength of effects across areas . As noted above , beta power and synchrony was strongest within AIP and between AIP and the other areas . Beta category-selectivity was also strongest within AIP . However , as Figure 6 shows , AIP showed the weakest category selectivity in the spiking rate of its neurons . This is not because there were fewer isolated neurons in AIP vs . other areas ( see above for numbers of neurons from each area ) . The average number of isolated neurons per electrode was also similar across areas ( AIP: 2 . 48 neurons/electrode; cPFC: 2 . 67; lPFC: 3; note that electrodes with zero neurons were not included in any of the LFP analyses ) . LFP synchrony displayed similar dynamics between pairs of electrodes that had the same category preference in their spiking activity as those that had different category preferences ( Figure 6—figure supplement 2 ) . The above results indicate category-selective patterns of LFP synchrony within and between the parietal and prefrontal cortices . Here , we examine synchrony between spikes in one area and LFPs in another . Synchrony between spikes in area A ( the neuronal output from A ) and LFPs in area B ( presumably reflecting the synaptic inputs to B ) , but not the other way around , may suggest a unidirectional influence from A to B . We started with pairs of electrodes from different areas that displayed Above/Below category-selective beta synchrony . For each pair of electrodes we first averaged the category selectivity ( i . e . , the data plotted in Figure 4 ) of LFP synchrony separately and the Category and Shift epochs . Then , we identified the electrode pairs that showed the strongest 10% of selectivity in beta synchrony in each epoch . We computed spike-field synchrony between them , using the same method that we previously used to examine frontostriatal interactions ( Antzoulatos and Miller , 2014 ) . A phase-locking value ( PLV ) between spikes and LFPs ( see Materials and methods ) indicated whether spikes in one area are more likely to occur at specific LFP phases from an electrode in other areas . Because this metric can show spurious synchrony if the number of spikes is small ( Antzoulatos and Miller , 2014 ) , we took two mitigating measures . First , we pooled together all the spikes from isolated neurons that were recorded from each electrode , thus maximizing the number of spikes in each electrode . Second , we transformed the PLV to a z score based on 200 random permutations of the LFP . By keeping each trial’s series of spikes and permuting only the LFP trace we were able to evaluate the levels of spike-LFP PLV that arose solely from the number and temporal pattern of the spikes . We first focused on the Category epoch because that is where we observed the strongest category-selective synchrony of beta oscillations ( Figure 4 ) . This analysis revealed significant spike-LFP synchrony between cPFC spikes and AIP LFP beta oscillations . An example is shown in Figure 7A , where the red trace corresponds to the 25–35 Hz oscillations of an AIP LFP , and the vertical black bars indicate spike times simultaneously recorded from a cPFC electrode . As the superposition indicates , the cPFC spikes tended to appear at the same phase of the LFP oscillation , namely the downstroke of the LFP cycle . Average results from all pairs of spikesand LFPs during the Category epoch are shown in panels B-D of Figure 7 for each combination of areas ( i . e . , cPFC-AIP: n = 47 pairs , lPFC-AIP: n = 54 , and cPFC-lPFC: n = 32 ) . The dots indicate when the PLV between spikes in one area and LFPs in the other were significantly greater than the other way around ( 2-tailed t-test , p<0 . 05 ) . As can be seen from this figure , significant spike-LFP synchrony was between cPFC spikes and AIP LFPs , with a peak at 30 Hz ( Figure 7B ) . This synchrony was asymmetric: there was significantly stronger phase-coupling of cPFC spikes to AIP LFP oscillations in the 22–32 Hz band ( 0 . 006 ± 0 . 002 ) than the other way around ( −0 . 001 ± 0 . 001; phase-coupling of cPFC oscillations to AIP spikes , t-test: p=0 . 004 ) . This asymmetry was more pronounced for the category Below than Above ( Figure 7—figure supplement 1 ) . No frequency-dependent spike-LFP synchrony was observed between lPFC spikes and AIP LFPs or the reverse ( Figure 7C ) . There was only modest synchrony between cPFC spikes and lPFC LFPs in the beta band , but this was not significantly unidirectional ( Figure 7D ) . 10 . 7554/eLife . 17822 . 019Figure 7 . Frequency-dependent synchrony of spiking activity to LFPs during the Category epoch . ( A ) Example LFP ( red trace; passband filtered at 25–35 Hz ) from an AIP electrode , simultaneously recorded with spiking activity from a cPFC electrode ( black bars ) . Prefrontal spikes are phase-locked to the downstroke of the parietal LFP . Scale bars: 100 ms and 0 . 2 mV . ( B ) Average ( ±SEM ) synchrony ( PLV ) between spikes in cPFC and LFPs in AIP ( black trace ) , or the reverse ( red trace ) as a function of frequency , for pairs of electrodes with category-selective synchrony of beta oscillations ( Figure 4 ) . Dots indicate significantly different spike-field synchrony ( p<0 . 05 ) between the two directions . ( C ) Same as in B , for pairs of electrodes between lPFC and AIP . ( D ) Same as in B and C , for pairs of electrodes between cPFC and lPFC . See also Figure 7—figure supplement 2 for spike-field synchrony during the Shift epoch , as well as spike-field synchrony between pairs of electrodes with category-selective delta synchrony . The asymmetry in cPFC-AIP vs . AIP-cPFC spike-LFP synchrony is stronger in the category Below than Above ( see Figure 7—figure supplement 1 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 17822 . 01910 . 7554/eLife . 17822 . 020Figure 7—figure supplement 1 . Category-specific spike-field synchrony between cPFC spikes and AIP LFPs . Synchrony between cPFC spikes and AIP LFPs ( black lines ) or the reverse ( red lines ) , separately for category Above ( top ) and Below ( bottom panel ) . DOI: http://dx . doi . org/10 . 7554/eLife . 17822 . 02010 . 7554/eLife . 17822 . 021Figure 7—figure supplement 2 . Spike-LFP synchrony . Frequency-dependent average ( ±SEM ) spike-field synchrony ( PLV ) between cPFC and AIP ( top row of panels ) , between lPFC and AIP ( middle row ) , and between cPFC and lPFC ( bottom row ) . ( A ) Same as panels B–D of Figure 7 , but for the Shift epoch . ( B ) Same analyses as ( A ) and Figure 7 , but the pairs of electrodes were selected on the basis of their category-selective synchrony of delta , rather than beta , oscillations . DOI: http://dx . doi . org/10 . 7554/eLife . 17822 . 02110 . 7554/eLife . 17822 . 022Figure 7—figure supplement 3 . Category-selective synchrony between cPFC-AIP LFPs . Only pairs of electrodes with strong ( top 10% ) spike-LFP synchrony ( left panels ) or without ( bottom 10% ) spike-LFP synchrony ( right panels ) were used in this analysis of Above/Below category selectivity between cPFC and AIP LFPs . Top panels show pairs of electrodes with spikes in cPFC and LFPs in AIP and bottom panels show the reverse . cPFC-AIP category selectivity of beta synchrony is more pronounced for electrodes that also have strong spike-LFP coupling than pairs of electrodes without spike-LFP coupling . DOI: http://dx . doi . org/10 . 7554/eLife . 17822 . 022 There was only modest spike-LFP synchrony during the Shift epoch ( Figure 7—figure supplement 2 ) , with modest but significant directionality between the two prefrontal subregions ( coupling of cPFC spikes to lPFC 12–16 Hz oscillations ( 0 . 002 ± 9 . 5×10–4 ) was stronger than the reverse ( −0 . 002 ± 9 . 01×10–4 ) , p=0 . 004 ) . Finally , we performed the same analyses on select pairs of electrodes that showed strong category-selective synchrony in the delta band . We only saw modest synchrony between cPFC spikes and AIP beta oscillations ( Figure 7—figure supplement 2 ) . These results indicate that the pairs of frontoparietal sites that displayed category-selective beta synchrony also showed strong spike-LFP synchrony in the beta band . Conversely , pairs of frontoparietal sites that displayed strong spike-LFP synchrony displayed more pronounced category-selective synchrony of beta rhythms ( Figure 7—figure supplement 3 ) . We next asked whether spike-LFP synchrony across areas ( evaluated as above ) was stronger when spiking activity carried task-relevant information . Indeed , we found that cPFC electrodes with spiking selectivity for the relevant category distinction ( Above vs . Below ) showed stronger synchrony with AIP beta oscillations than the cPFC electrodes with spiking selectivity for the irrelevant dimension ( Right vs . Left ) . From each neural population ( cPFC , lPFC , and AIP ) we first averaged the selectivity of each electrode’s spiking activity within the Category and Shift epochs ( i . e . , the PEV metric shown in Figure 6—figure supplement 1 ) . We then selected the electrodes at the top 10% of spiking selectivity for the Above/Below dimension and the electrodes at the top 10% for the Right/Left dimension ( group statistics appear in Table 1 ) . This resulted in the same number of electrodes for each dimension within each population ( cPFC: n = 11 electrodes for Above/Below selectivity and same number but different electrodes for Right/Left selectivity; lPFC: n = 13; AIP n = 16 ) . The selection of the top 10% was necessary because the Above/Below selectivity was more prevalent than the Right/Left selectivity ( Figure 6—figure supplement 1 , panels B and D ) . Selecting the top 10% eliminated this bias . We subsequently evaluated the synchrony of spikes from each of these electrodes with all the simultaneously recorded LFPs from other electrodes , regardless of the category selectivity in the synchrony between the corresponding pair of LFPs . 10 . 7554/eLife . 17822 . 023Table 1 . Electrodes with spiking at the top 10% of selectivity . DOI: http://dx . doi . org/10 . 7554/eLife . 17822 . 023Neural population cPFC lPFC AIP Above/Below selectivityCategory epochMean11 . 743 . 573 . 96SEM2 . 140 . 490 . 30Shift epochMean9 . 126 . 544 . 53SEM1 . 310 . 750 . 43Right/Left selectivityCategory epochMean2 . 693 . 261 . 64SEM0 . 410 . 420 . 21Shift epochMean1 . 301 . 930 . 88SEM0 . 260 . 310 . 11Average ( ±SEM ) selectivity during the Category and Shift epochs for the Above/Below and Right/Left dimensions of sample location is provided for each neural population of the analyses in Figure 8 . The results of the cPFC-AIP analyses appear in Figure 8 . In the Category epoch , we found significantly stronger synchrony of AIP LFPs with cPFC spikes that had Above/Below selectivity than with cPFC spikes that had Right/Left selectivity ( peak at 32 Hz , Figure 8A , t-test: p=9 . 3×10–7 ) . In fact , synchrony of cPFC spikes to AIP LFPs was significant for spikes with Above/Below selectivity ( n = 54 spike-LFP pairs , 0 . 01 ± 0 . 002 , t-test comparison to zero , p=1 . 7×10–6 ) , but not different from zero for spikes with Right/Left selectivity ( n = 44 spike-LFP pairs , −0 . 003 ± 0 . 002 , p=0 . 98 ) . By contrast , synchrony between AIP spikes and cPFC LFPs was significant ( Figure 8B; Above/Below: n = 50 , 0 . 006 ± 0 . 002 , p=3 . 7×10–4; Right vs . Left: n = 52 , 0 . 004 ± 0 . 001 , p=0 . 001 ) but there was no difference between AIP spikes with Above/Below vs . Right/Left selectivity ( p=0 . 32 ) . The difference between cPFC spike rates with Above/Below selectivity vs . spike rates with Right/Left selectivity was similar for both the Above and Below categories ( Figure 8—figure supplement 1 ) . Similarly , spike rates from electrodes without any Above/Below selectivity ( the bottom 10% of PEV ) also displayed weaker spike-LFP beta synchrony than the electrodes with spike rates with strong Above/Below selectivity ( Figure 8—figure supplement 2 ) . By contrast , there was little or no cPFC-AIP spike-LFP synchrony in the Shift epoch for either Above/Below or Right/Left selective spikes ( Figure 8C , D ) . Finally , there was only modest spike-LFP synchrony between lPFC and AIP , or between cPFC and lPFC , in either direction , regardless of Above/Below or Right/Left selectivity ( Figure 8—figure supplement 3 ) . 10 . 7554/eLife . 17822 . 024Figure 8 . Spike-LFP synchrony for selective spiking . ( A ) Average ( ± SEM ) PLV between cPFC spikes and AIP LFPs as a function of frequency , separately for electrodes with strong Above/Below selectivity in their spiking ( black trace ) and electrodes with strong Right/Left selectivity in their spiking ( red trace ) during the Category epoch of the categorization trials . Dots mark frequencies of significant difference between the red and black traces ( 2-tailed t-test; p<0 . 05 ) . ( B ) Same as ( A ) , but in reverse direction: synchrony between AIP spikes and cPFC LFPs . ( C ) and ( D ) Same as ( A ) and ( B ) respectively , but for the Shift epoch of the categorization trials . The group statistics of the electrodes used in these analyses ( electrodes at the top 10% of spiking selectivity ) appear in Table 1 . See also Figure 8—figure supplement 3 for the corresponding analyses on the other combinations of spike-LFPs . See Figure 8—figure supplement 2 for a comparison of spikes with vs . without Above/Below selectivity . Figure 8—figure supplement 1 illustrates cPFC-AIP spike-LFP synchrony separately for each category . DOI: http://dx . doi . org/10 . 7554/eLife . 17822 . 02410 . 7554/eLife . 17822 . 025Figure 8—figure supplement 1 . Spike-field synchrony of cPFC-AIP pairs separated by category . Top panels for PLV evaluated from trials of category Above and bottom panels for trials of category Below . Left for spikes from cPFC and LFPs from AIP , and right panels for the reverse . For each of the categories , the dynamics of spike-LFP synchrony are similar to those in Figure 8 . DOI: http://dx . doi . org/10 . 7554/eLife . 17822 . 02510 . 7554/eLife . 17822 . 026Figure 8—figure supplement 2 . Spike-field synchrony for category-selective vs . non-selective spiking . As in Figure 8 , ‘Above/Below selective spikes’ ( black line ) refers to electrodes with the top 10% of spiking selectivity for Above/Below categories . ‘Non-selective spikes’ ( red line ) refers to electrodes with the bottom 10% of spiking selectivity for Above/Below categories . DOI: http://dx . doi . org/10 . 7554/eLife . 17822 . 02610 . 7554/eLife . 17822 . 027Figure 8—figure supplement 3 . Spike-field synchrony for selective spikes . Same analyses as in Figure 8 , but for spikes in lPFC and LFPs in AIP ( top panels in A ) , spikes in cPFC and LFPs in lPFC ( top panels in B ) , and the respective reverse directions ( bottom panels in A and B ) . DOI: http://dx . doi . org/10 . 7554/eLife . 17822 . 027 These results suggest a relationship between the magnitude of category selectivity a cPFC electrode had in its spike rate and the synchrony of that spiking to the parietal beta oscillations . Indeed , as Figure 9 illustrates , there was significant linear ( but not rank ) correlation between the Above/Below spike rate selectivity of cPFC electrodes and their synchrony to the AIP beta rhythms ( averaged across all simultaneously recorded AIP electrodes: Pearson’s r=0 . 25 , p=0 . 009; Spearman’s rank r=0 . 16 , p=0 . 09 ) . By contrast , there was no linear ( or rank ) correlation between the Right/Left spike rate selectivity of the same cPFC electrodes and their synchrony to parietal beta rhythms ( Pearson’s r=−0 . 11 , p=0 . 28; Spearman’s rank r=−0 . 04 , p=0 . 70 ) . 10 . 7554/eLife . 17822 . 028Figure 9 . Correlation of spike-LFP synchrony with spiking selectivity . Synchrony of cPFC spikes to AIP LFP beta-band oscillations , during the Category epoch . Above/Below selectivity in spiking ( top ) is positively correlated with the spike synchrony ( averaged across all their electrode pairings ) to parietal beta rhythms ( r=0 . 25 , p=0 . 009 ) , whereas Right/Left selectivity ( bottom ) is not ( r=−0 . 11 , p=0 . 28 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 17822 . 028
We found that during a rule-based spatial categorization task , there was an increase in beta-band oscillatory power and synchrony within and between the prefrontal cortex and parietal cortex . Different pairs of recording sites , both within and between the prefrontal cortex and parietal cortex , showed category selectivity , that is , increased LFP synchrony for one or the other category , primarily in the beta but also in the delta band . Thus , each category was accompanied by different network patterns of synchrony . Prefrontal spikes were synchronized to parietal LFP beta rhythms , but parietal spikes were not synchronized to prefrontal rhythms . This suggests a unidirectional influence with the PFC influencing the parietal cortex . Finally , we found that prefrontal neural spiking selective for the task-relevant Above vs . Below category dimension was synchronized with the parietal beta rhythms , but spiking that reflected the task-irrelevant Right vs Left dimension was not . Oscillatory synchrony has been suggested to play a role in establishing functional networks ( Engel et al . , 2001; Fries , 2005; Miller and Buschman , 2013; Uhlhaas et al . , 2009 ) . Our finding of category-selective patterns of increased synchrony between recording sites in the frontoparietal cortex supports this . It adds to previous reports of content-specific patterns of oscillatory synchrony in these circuits ( and beyond ) . For example , there is object-selective synchrony between the prefrontal and parietal cortex for different objects held in working memory ( Salazar et al . , 2012 ) and within the prefrontal cortex for different behavioral rules ( Buschman et al . , 2012 ) . Further , category-selective patterns of synchrony develop between the prefrontal cortex and striatum in parallel with category learning ( Antzoulatos and Miller , 2014 ) . These ‘rhythmic ensembles’ do not seem to be limited to circuits involving the prefrontal cortex . Selective synchrony of gamma rhythms between V4 and V1 has been reported to guide attentional selection in primates ( Bosman et al . , 2012 ) . A recent study in humans suggests functional networks across the cortex are formed by harmonic patterns constrained by known anatomy ( Atasoy et al . , 2016 ) . In other words , it seems as if anatomy ( the connectome ) may be the infrastructure that determines which functional circuits can potentially form . Which circuits do form from moment to moment may be regulated , at least in part , by patterns of synchrony . To use an analogy: Anatomy is the roads , activity is the traffic , and rhythmic synchrony is the traffic lights . In prior reports , as well as in this study , these ‘rhythmic ensembles’ tend to center on the beta band ( Antzoulatos and Miller , 2014; Buschman et al . , 2012; Haegens et al . , 2011; Micheli et al . , 2015; Salazar et al . , 2012; Siegel et al . , 2009 ) . This is noteworthy because there is increasing evidence that beta-band synchrony may play a role in sustained top-down , or feedback , signaling in the cortex ( Bastos et al . , 2015; Engel and Fries , 2010 ) . In line with this is our observation that category selectivity was more ‘pure’ in synchrony ( especially in the beta band ) than for spiking activity . While beta synchrony was selective for the task relevant categories , we could not detect any ( irrelevant ) information about the left vs . right hemifield location of the sample cue in beta synchrony . By contrast , left vs . right was detectable in neural spike rates . Of course , selectivity of spiking rate and LFP synchrony are two different phenomena , which rely on different measures and analyses . Thus , this difference could merely reflect differences in the ability to detect information in them . But it is also possible that beta synchrony is more selective . Beta synchrony may provide an additional level of filtering beyond that seen in spiking activity . Because of its putative role in feedback processing in the cortex , it may only pass back top-down information to other cortical areas . Support for this also comes from our finding that only the prefrontal spiking activity that conveyed task-relevant information was synchronized to the parietal beta rhythms . By contrast , for pairs of recording sites with category-selective beta synchrony , we did not detect any synchrony between parietal spikes with prefrontal oscillations . This all suggests a top-down influence from the PFC to parietal cortex . Our results are in line with other studies that indicate prefrontal and parietal cortex involvement in visual categorization ( Braunlich et al . , 2015; Crowe et al . , 2013; Goodwin et al . , 2012; Swaminathan and Freedman , 2012 ) . Neurons in both areas show correlates of a variety of abstract cognitive factors such as number , shape and spatial categories ( Goodwin et al . , 2012; Murata et al . , 2000; Tudusciuc and Nieder , 2009 ) . Chafee and colleagues and Merchant and colleagues have previously reported single-neuron correlates of spatial categories . By analyzing the fluctuations of the time course of spiking from both areas , they also concluded that category information was transmitted in a top-down fashion from the PFC to parietal cortex ( Crowe et al . , 2013; Merchant et al . , 2011 ) . In sum , our results suggest that the spatial categories may be computed in prefrontal circuits , where this information intermingles with bottom-up information ( in this case , the task-irrelevant distinction between the hemifields ) . But the prefrontal cortex selectively broadcasts via beta rhythms , and thus strengthens , only the relevant information back to the parietal cortex . Long-range frontoparietal networks supported by synchronized beta rhythms , then , permit the most relevant representation to dominate and guide the animal’s category decision .
Data were collected from two adult male rhesus macaque monkeys ( Macaca mulatta ) , weighing 10–12 kg . The animals were taken care of in accordance with the National Institutes of Health guidelines and the policies of the Massachusetts Institute of Technology Committee for Animal Care . Both animals were trained on the rule-based categorization task until they reached similar levels of proficiency . Each of them was surgically implanted with one titanium headbolt and two titanium recording chambers , one of which was placed above the left principal sulcus and the other above the left intraparietal sulcus . Placement of the chambers as well as estimation of the electrode recording sites were guided by structural MRI scans and computed with Matlab ( Mathworks , Natick MA ) . Neural data were collected from 39 experiments ( 22 experiments from animal G and 17 from animal P ) . Experimental control was implemented via Cortex ( NIMH , Laboratory of Neuropsychology ) and infrared eye-tracking , sampled at 500 Hz , via Eyelink 1000 ( SR Research Ltd , Mississauga , Canada ) . The animals were seated in customized primate chairs and placed inside sound-attenuating chambers . Visual stimuli were presented at full contrast on a CRT monitor ( at a distance of approx . 50 cm from the animal ) , refreshing at 100 Hz . The task design is shown in Figure 1A . Trials began when the animal held a bar and maintained visual fixation on a central target for 1 s . The fixation target was a red square ( 0 . 25-degree side ) , flanked on the right and left side with 2 white horizontal lines of 0 . 2-degree thickness and 6-degree length . These lines , each of which had a 2-degree gap from the fixation target ( the allowable fixation window had a 1 . 5-degree radius ) , were only displayed for the 1 s duration of the fixation window , in order to help the animals re-calibrate the division between upper/lower visual hemifields after the previous trial . The sample stimulus was displayed on the screen 200 ms after the lines had been turned off ( the fixation target stayed on the screen for the duration of the trial and the animals were required to maintain fixation throughout ) . Sample stimuli were circular white dots ( 0 . 2-degree radius ) , the location of which was selected randomly from 16 possible locations ( four in each visual quadrant ) , plus a random jitter of ±0 . 5 degree on each axis . Overall , the sample could appear in any one of 144 distinct locations , spanning an area of 170 degrees2 . No samples were displayed inside the animals’ spatial window of fixation . Sample display lasted for 800 ms and was followed by a 500 ms delay interval , during which only the fixation target appeared on the screen . Up to this point of the trial the animals were required to determine whether the sample had appeared in the upper or lower visual hemifield ( i . e . , category Above or Below ) and hold the sample category in working memory . At the end of the delay interval , the two white horizontal lines ( hemi-boundaries ) re-appeared at new locations . A clockwise ( CW ) or counter-clockwise ( CCW ) shift was randomly selected on each trial . A CW shift meant that the left hemi-boundary appeared 4 degrees above the horizontal meridian and the right hemi-boundary appeared 4 degrees below the horizontal meridian . A CCW shift was implemented in the reverse way . During the 1 s interval following the new boundary display , the animals were required to adjust the category decision criterion . Note that the new location of the boundaries did not change the sample category . Rather , it only changed the spatial area that corresponded to that category . After the boundary shift , the new retinotopic spatial area of the Above/Below categories overlapped with the pre-shift retinotopic area by only 50% . Because of this , test stimuli that were very close to the sample location could potentially be a non-match to the sample category , and test stimuli far away from the sample location could potentially be a match to the sample category . At the end of this delay interval , a test stimulus , identical in form to the sample , was displayed for 1 s . The location of the test stimulus was selected randomly from one of the 624 locations shown in Figure 1B . If the test stimulus matched the category of the sample stimulus ( Above or Below the hemi-boundaries ) , the animals had to release the bar before the end of its display for liquid reward . If it did not match the sample category , the animals were required to maintain contact with the bar and visual fixation for another 0 . 5 s following the test , when a second test would appear on the screen . The second test would always be a match and the animals would release the bar for reward . All neurophysiological recordings were performed with the Cerebus Neural Processing System ( Blackrock Microsystems , Salt Lake City , Utah ) . Simultaneous multi-electrode recordings were made from the left frontal cortex and the left posterior parietal cortex , using two custom-made multi-electrode arrays ( each array with 8–25 tungsten electrodes; FHC , Bowdoin , ME ) . Electrodes from both arrays were acutely lowered every day of recording , either individually or in pairs , at depths that were guided both by the MRI images ( to target the areas of interest ) and the success in isolating neural spiking activity . This method of multi-electrode/multi-area recording allowed us to: ( a ) sample neural signals from relatively large surface areas ( each array covered up to 250 mm2 , with electrodes that could be no closer than 1 mm from each other ) , ( b ) sample neural signals from different sets of locations every day of recording , ( c ) evaluate the real-time cross-area dynamic interactions in the frequency and time domains , and ( d ) make cross-area comparisons of neural processing under identical experimental conditions . Neural data from the posterior parietal cortex came from the rostral part of the lateral bank of the intraparietal sulcus ( anterior intraparietal area; AIP ) . Our analyses of prefrontal neural data distinguished between the regions of the principal sulcus ( lateral PFC; lPFC ) and the more posterior area 8A ( caudal PFC; cPFC ) , which lies at the junction between the prefrontal and premotor cortices . Area 8A , extending from the caudal end of the principal sulcus up to the arcuate sulcus , also includes the frontal eye field ( FEF ) . Results from the FEF in isolation were similar to the results from the rest of area 8A , and were pooled together in the cPFC group . In addition , results from area 45 were similar to those from area 46 and pooled together in the lPFC group . Additional recordings were made from various adjacent frontal and parietal areas , but in this report we are focusing on the three areas that gave us the bulk of the data from both monkeys . Electrode recordings were first fed to a unity-gain headstage and were referenced to ground . The recorded signals were band-pass filtered to separate spikes from LFPs ( 0 . 3–500 Hz for LFPs and 250–7500 Hz for spikes ) . LFPs were digitized at 1 KHz sampling rate and spikes were digitized at 30 KHz . To ensure that only signals from active regions of the brain were collected , LFPs were recorded only from sites that also displayed spiking activity . Spikes were sorted into clusters of putative single-neuron activity using Offline Sorter v3 ( Plexon; Dallas , TX ) , based on principal component analysis . Firing rates ( in spikes/second ) were computed from spike counts over contiguous 20 ms bins and convolved with a 200 ms Gaussian window . For selectivity analyses , the firing rates were z-transformed based on the mean and variance of firing rate during a 3 s pre-trial baseline . Neural selectivity of spiking activity was quantified in Matlab as the bias-corrected , omega-variant of percent explained variance ( ωPEV; Brincat and Miller , 2015; Buschman et al . , 2011 ) . LFPs were bi-directionally bandstop filtered ( 59–61 Hz ) , using a 10th order Butterworth filter , to minimize the component of 60-cycle noise . The LFPs were subsequently downsampled at 333 Hz , and centered on the cross-trial mean . The latter processing step reduced the amplitude of stimulus-evoked potentials and helped isolate the trial-induced oscillations . The LFPs were then decomposed to their spectral components using a Matlab-based wavelet analysis toolbox ( Torrence and Compo , 1998; offered at the URL: http://atoc . colorado . edu/research/wavelets/ ) , through their convolution with a Morlet wavelet , at six octaves ( from 2 Hz to 128 Hz ) at a 0 . 1- octave resolution . Quantification of LFP-LFP synchrony ( pairwise phase consistency ) relied on the Matlab-based toolbox FieldTrip ( Oostenveld et al . , 2011 ) . Finally , spike-LFP synchrony was quantified as the single-trial phase-locking value ( PLV ) , which measures the angular concentration of the frequency-specific instantaneous phases of the LFP at the time of spikes ( Antzoulatos and Miller , 2014; Siegel et al . , 2009 ) . For illustration purposes , the group-averaged data were smoothed with kernels of FWHM less than 0 . 5 octave and/or 100 ms . All statistical analyses were performed on unsmoothed data using Matlab , and corrections for multiple comparisons followed the Bonferroni method . Unless noted , group data are reported as average ± SEM . | A brain that could store only exact experiences would bog us down with details . We have instead evolved to be able to detect the common elements in different experiences and group them into meaningful categories . This imbues the world with meaning . We can recognize and respond appropriately to objects , situations and expressions even if we have never encountered those exact examples before . Without this ability , experiences would be fragmented and unrelated . Things would seem strange and unfamiliar if they differed even trivially from previous examples . This situation describes many of the characteristics of neuropsychiatric disorders such as autism and schizophrenia . Most studies have focused on how single brain areas or single neurons categorize experiences . The brain , however , is composed of many interacting networks of neurons that extend across several different areas of the brain . Repetitive rhythms , or waves , of electrical activity generated by the neurons seem to play a major role in network interactions . These rhythms are given different names depending on how rapidly they oscillate . In order for our brain to work successfully , these rhythms synchronize across the relevant brain areas . Antzoulatos and Miller trained monkeys to categorize stimuli as “Above” or “Below” depending on where dots and lines appeared on a screen . The activity of the neurons in two regions of the monkey’s brain – called the prefrontal cortex and the parietal cortex – was recorded as each monkey performed the task . The recordings revealed that synchronized rhythms between the prefrontal and parietal cortices supported the monkey’s ability to categorize the stimuli . Changes in how strongly the rhythmic electrical activity of the neurons was synchronized – particularly for a type of wave called a beta wave – conveyed information about the category of stimuli ( i . e . , whether they counted as Above or Below ) . Single neurons also conveyed this categorization , but unlike rhythms , they also carried irrelevant information . Therefore the synchronized beta waves could act as a filter for the features of an object or experience that are relevant to the task at hand . The prefrontal cortex and the parietal cortex are only two of many brain areas involved in categorization . Much more territory remains to explore . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"neuroscience"
] | 2016 | Synchronous beta rhythms of frontoparietal networks support only behaviorally relevant representations |
Adequate responses to noxious stimuli causing tissue damages are essential for organismal survival . Class IV neurons in Drosophila larvae are polymodal nociceptors responsible for thermal , mechanical , and light sensation . Importantly , activation of Class IV provoked distinct avoidance behaviors , depending on the inputs . We found that noxious thermal stimuli , but not blue light stimulation , caused a unique pattern of Class IV , which were composed of pauses after high-frequency spike trains and a large Ca2+ rise in the dendrite ( the Ca2+ transient ) . Both these responses depended on two TRPA channels and the L-type voltage-gated calcium channel ( L-VGCC ) , showing that the thermosensation provokes Ca2+ influx . The precipitous fluctuation of firing rate in Class IV neurons enhanced the robust heat avoidance . We hypothesize that the Ca2+ influx can be a key signal encoding a specific modality .
Animals can sense diverse sensory inputs such as optical stimuli , plumes of volatile chemicals , acoustic waves , and atmospheric temperature by using a variety of specific sensory organs . Among these , polymodal nociceptors are defined as sensory neurons that respond to aversive stimuli with distinct physical properties . For example , mammalian C-fiber nociceptors detect harsh touch , high temperature , acid stimulus , and a number of toxic chemicals ( Abraira and Ginty , 2013; Delmas et al . , 2011; Perl , 1996 ) . Ion channel proteins that are expressed in polymodal nociceptors of nematodes , insects , and mammals have been discovered , and nowadays , molecular basis of thermal or mechanical nociceptive sensory transduction has primarily converged on three protein superfamilies: the TRP channels , the PIEZO channels , and the DEG/ENaC channels , all of which are well-conserved across many species ( Basbaum et al . , 2009; Geffeney and Goodman , 2012; Lumpkin et al . , 2010 ) . TRP channels are nonspecific cation channels and several members of this family have been implicated in thermal nociception ( Lumpkin and Caterina , 2007 ) . For example , dTrpA1 is an essential component of both thermal and mechanical nociception in Drosophila ( Zhong et al . , 2012 ) . PIEZO and DEG/ENaC channels are required specifically for the mechanical nociception ( Chatzigeorgiou et al . , 2010; Coste et al . , 2010; Kim et al . , 2012b ) . Exposure to short-wavelength light is also one of the hazardous stimuli for terrestrial animals such as soil-dwelling nematodes and a number of boring insects including the larval forms of holometabolous insects ( Hori et al . 2014 ) . Both the lite-1 gene in C . elegans and Gustatory receptor 28b ( Gr28b ) gene in Drosophila encode the light-activated G-protein-coupled receptors and are essential for the noxious light sensation eliciting stereotypical light-avoidance behaviors ( Liu et al . , 2010; Xiang et al . , 2010; Yamanaka et al . , 2013 ) . In at least two noteworthy cases , organisms can react to different modes of stimuli through a single type of nociceptor: the C . elegans PVD neuron responds to a harsh poke and cold temperatures ( Chatzigeorgiou et al . , 2010 ) , whereas one subclass of multidenritic neurons in the Drosophila larva perceives a sharp poke , noxious high temperatures , and UV or blue light ( Im and Galko , 2012; Xiang et al . , 2010 ) . Intriguingly , Drosophila larvae exhibit distinct escape behavior outputs in response to the thermal or mechanical stimuli versus blue light ( see below ) . To unveil the mechanism of this type of polymodal sensation and processing , we addressed how the distinct sensory inputs are transduced and encoded into differential firing patterns . Drosophila dendritic arborization ( da ) neurons constitute a subfamily of sensory neurons that elaborate their dendritic arbors two-dimensionally on the basal surface of the epidermis ( Grueber et al . , 2002; Han et al . , 2012; Im and Galko , 2012; Jan and Jan , 2010; Kim et al . , 2012a; Shimono et al . , 2009; Yasunaga et al . , 2010 ) . da neurons in the larval abdominal hemisegment are classified into four morphological categories , Classes I-IV , in order of increasing territory size and/or branching complexity ( Grueber et al . , 2002; Jan and Jan , 2010 ) . A series of studies have shown that Class IV neurons are polymodal nociceptors and relevant receptors are identified as described below: Touching the dorsal cuticle of larvae with a metal probe hotter than 45°C evokes a stereotypic corkscrew-like escape rolling behavior ( Babcock et al . , 2009; Tracey et al . , 2003 ) ; and activation of the Class IV neurons is necessary and sufficient for this output ( Hwang et al . , 2007; Ohyama et al . , 2013 ) . This thermal nociception behavior is mediated by thermoTRPs genes including Drosophila TRPA1 ( dTrpA1 ) , Painless , and Pyrexia ( Babcock et al . , 2011; Hwang et al . , 2012; Lee et al . , 2005; Neely et al . , 2010; 2011; Tracey et al . , 2003; Zhong et al . , 2012 ) . In addition to the noxious heat , strong mechanical stimuli , such as a sharp poke , elicit the same nocifensive behavior ( Hwang et al . , 2007 ) . Moreover , Class IV neurons are extra-ocular photoreceptors that respond to short-wavelength light stimuli and evoke a directional shift of locomotion , which requires Gr28b ( Berni et al . , 2012; Xiang et al . , 2010; Yamanaka et al . , 2013 ) . Therefore , to achieve such a discriminative processing , Class IV neurons are supposed to integrate multiple modes of sensory inputs into distinct firing patterns in a highly sophisticated manner . Recently , extracellular single-unit recordings have been developed to monitor physiological activities of Class I-IV dendritic arborization neurons upon naturalistic stimulation , such as light and gentle touch ( Tsubouchi et al . , 2012; Xiang et al . , 2010; Yan et al . , 2013; Zhang et al . , 2013 ) . However , there have been few extensive analyses of firing responses of Class I–IV neurons upon spatially restricted and temporally controlled noxious heat stimulations . In this study , we built a new measurement system that combines three components: an infrared ( IR ) laser that allows quantitatively controlled step-like and local heating , neuronal calcium imaging , and extracellular single-unit recording to monitor firing patterns . In response to the local heating over 45°C , Class IV neurons generated unique firing patterns that were composed of intermittent pause periods after high-frequency firing trains . This 'burst and pause' firing pattern was associated with a large rise in Ca2+ in the entire dendritic tree , which we designated as the Ca2+ transient . We showed that both of the unique firing pattern and the Ca2+ transient are physiological responses of Class IV neurons , characterized their underlying mechanisms by using pharmacological and genetic approaches , and compared the response to the noxious heat with that to short-wavelength light to address the basis of the polymodal sensation . Furthermore , we pursued which characteristics of the 'burst and pause' firing pattern contributes to the induction of the rolling behavior , and showed that the multiple fluctuations of firing rate enhanced robust output by using optogenetic manipulation . Altogether , we propose that the low-frequency continuous firing train transmits the noxious light sensation; in contrast , the multiple peaks of firing rate fluctuation facilitate the perception of a noxious heat stimulus .
To understand how Class IV neurons respond to noxious thermal stimuli , we built a new system employing a 1462-nm infrared ( IR ) laser as a step-like and local heating device ( Figure 1A ) . The IR laser has been applied for the precise control of temperature with high spatial and temporal resolution , such as the biophysical analyses of the temperature dependency of thermoTRP channels ( Yao et al . , 2009 ) and heat-shock-mediated expression of transgenes in targeted single cells by IR-laser irradiation through a microscope objective ( Kamei et al . , 2008 ) . 10 . 7554/eLife . 12959 . 003Figure 1 . ThermoTRP-dependent responses of Class IV neurons to IR-laser irradiation . ( A ) A schematic diagram of the heat-stimulation system with the IR-laser irradiation . The IR-laser beam passed through an objective and targeted Class IV neurons in whole- or fillet-mounted larvae . The wavelength of IR matches the combination of symmetric and anti-symmetric OH stretching modes of water and can heat water efficiently ( Palmer and Williams , 1974 ) . We measured firing responses of the neuronal somata using an extracellular recording electrode in fillet preparations , and acquired ratiometric data of neuronal Ca2+ fluctuations in whole-mount or fillet preparations . ( B and C ) Temporal and spatial profiles of the IR-laser induced temperature changes of the microenvironment . A specimen was irradiated with the IR-laser beam for 1 s ( orange horizontal bar ) and temperature changes at the laser focus were calculated on the basis of changes of electrical resistances of a solution in a glass microelectrode ( see details in Methods ) . ( B ) Traces represent temporal changes of temperature at corresponding laser powers ( 10 , 20 , 30 , and 40 mW ) . The inset shows a plot of temperature increments at 700 ms at each power , and the red line indicates a linear fit to the data ( slope = +0 . 6°C/mW ) . ( C ) Pseudo-color map of spatiotemporal temperature changes around the laser focus ( the zero point along Y-axis ) in response to 40-mW laser irradiation and subsequent shut-off . The position of the glass microelectrode was shifted along the Y-axis at 5 μm intervals to monitor temporal changes at individual locations . At right is the color scale for the temperature change . ( D–G ) Extracellular single-unit recordings of Class IV neurons that were irradiated at various output powers ( D and E ) or at 40 mW ( F and G ) for 1 s ( orange horizontal bar in D and F ) . A location in the proximal dendritic arbor was targeted . ( D ) Spike trains and time courses of temperature changes at IR foci ( bottom ) . ( E ) Power-dependent increment in the maximum firing rate ( defined in 'Materials and methods' ) of the wild type ( circle ) , dTrpA11 mutant ( triangle ) , and pain3 mutant ( square ) . ( F ) Raster plots of firing ( top ) and time course of firing rate ( bottom ) of the wild type and the mutants ( bin size 50 ms ) . ( G ) Quantification of timing of maximum excitation ( 'Time to Max . Firing Rate' as defined in 'Materials and methods' ) of Class IV neuron of each genotype . Numbers of neurons tested were 9 ( the wild type ) , 10 ( dTrpA11 ) , and 7 ( pain3 ) , and data are presented as mean ± s . e . m . ( E–G ) . * p < 0 . 05 , ** p < 0 . 01 , *** p < 0 . 001 versus wild type by unpaired t-test with Bonferroni correction ( E and G ) . IR , infrared . DOI: http://dx . doi . org/10 . 7554/eLife . 12959 . 00310 . 7554/eLife . 12959 . 004Figure 1—source data 1 . The maximum firing rate of the wild type , dTrpA11 mutant , and pain3 mutant ( Columns A–F ) and Quantification of timing of maximum excitation of Class IV neuron of each genotype ( Columns H–J ) . DOI: http://dx . doi . org/10 . 7554/eLife . 12959 . 00410 . 7554/eLife . 12959 . 005Figure 1—figure supplement 1 . ThermoTRP-dependent responses of Class IV neurons to IR-laser irradiation . ( A–C ) Extracellular single-unit recordings of Class IV neurons that were irradiated at 40 mW for 1 s ( orange horizontal bar ) . A location in the proximal dendritic arbor was targeted . ( A ) Raster plots of firing ( top ) and time course of firing rate ( bottom ) of dTrpA1 heterozygote and pain null homozygotes ( bin size 50 ms ) . Numbers of neurons tested were 10 ( dTrpA11/ins ) , 8 ( painpc ) , and 8 ( painpf ) , and data are presented as mean ± s . e . m . dTrpA1 mutant ( dTrpA11/ins compound heterozygotes ) and pain mutants ( painpc and painpf homozygotes ) exhibited normal basal firing in the neurons ( data not shown ) , but showed significantly lower firing rates , compared to WT , during the IR-laser irradiation ( B ) [painpc] 43 . 1 ± 6 . 6 Hz , n = 8 cells; [painpf] 48 . 1 ± 6 . 4 Hz , n = 8; [dTrpA11/ins] 24 . 3 ± 10 . 2 Hz , n = 10 cells ) . Interestingly , dTrpA11/ins mutants displayed a significantly longer latency time between the onset of the IR irradiation and the increase in firing rate when compared to the wild-type , painpc and painpf larvae ( C ) Time to Max . Firing Rate: [painpc] 159 . 8 ± 84 . 4 ms; [painpf] 46 . 6 ± 11 . 0 ms; [dTrpA11/ins] 632 . 8 ± 96 . 3 ms ) . * p < 0 . 05 , ** p < 0 . 01 , *** p < 0 . 001 versus wild type ( Figure 1E–1G ) by unpaired t-test with Bonferroni correction . Genotype: ( painlesspc ) painpc/painpc; 3×[ppk-TN-XXL] ( attP2 ) / 3×[ppk-TN-XXL] ( attP2 ) ( painlesspf ) painpf/painpf; 3×[ppk-TN-XXL] ( attP2 ) / 3×[ppk-TN-XXL] ( attP2 ) ( dTrpA11/ins ) +/+; 3×[ppk-TN-XXL] ( attP2 ) dTrpA11/ dTrpA1insDOI: http://dx . doi . org/10 . 7554/eLife . 12959 . 00510 . 7554/eLife . 12959 . 006Figure 1—figure supplement 2 . Assessment of effects of IR-laser irradiation on Class IV neurons . IR laser irradiations had no apparent effect on the integrity of actin filaments ( A ) and the morphology of mitochondria ( B ) in dendritic arbors of the neurons . ( A ) Effects of IR-laser irradiation ( 42 mW , 1 s ) targeting a Class IV neuron expressing Lifeact:GFP ( marker for F-actins ) . ( Left ) Shown are GFP fluorescence images of dendrites before irradiation ( before ) , in the middle of irradiation ( middle ) , and after irradiation ( after ) . The red dashed circle represents the focus of the IR-laser beam . ( Right ) The time course of the fluorescent brightness change in the region indicated by the blue rectangle in 'before . ' The magenta shades and the red bar indicate timings of IR-laser irradiations . The cyan line represents the estimated photobleaching curve of GFP fluorescence . We assume that the transient decreases of fluorescence are due to heat-dependent attenuations of fluorescence during IR irradiations ( Kamei et al . , 2009 ) . Note that the brightness recovered to the estimated fluorescent level after the irradiations and that no apparent gap was found in the fluorescence image . ( B ) Effects of IR-laser irradiation onto a soma ( top ) or dendrites ( bottom ) of a Class IV neuron expressing mito:GFP ( marker for mitochondria ) . The red dashed circle represents the focus of the IR-laser beam ( soma: 38 mW , 1 s; dendrite: 42 mW , 1 s × 3 ) . No obvious changes in mitochondrial signals were detected . Genotype: ( A ) UAS-Lifeact-GFP/+; ppk-Gal4/+ , ( B ) UAS-mito-HA-GFP/ppk-Gal4 . IR , infrared . DOI: http://dx . doi . org/10 . 7554/eLife . 12959 . 006 We first examined the in vitro heating profiles of the IR laser: the kinetics of temperature changes , controllability of heating , and spatial distributions of heat around the laser focus . To measure the microenvironmental temperature , we exploited the temperature dependence of the electrical resistance of a glass microelectrode ( Palmer and Williams , 1974; Shapiro et al . , 2012; Yao et al . , 2009 ) . The temperature within a radius of approximately 50 μm from the laser focus rapidly increased from an ambient temperature ( 25°C ) to target temperatures ( 30–50°C ) within 100 ms , and stayed almost constant at the maximum temperature throughout the 1 s irradiation ( Figure 1B and C ) . Moreover , the temperature maximum was directly proportional to the power output of the laser ( Figure 1B ) . Compared to the perfusion protocol involving a preheated solution that is introduced into the recording chamber ( Xiang et al . , 2010 ) as well as the direct heating of the chamber ( Liu et al . , 2003 ) , the IR-laser system allowed us to reproducibly deliver more spatially restricted , and more temporally controlled stimuli . Next , we stimulated Class IV neurons by means of this system and measured the cellular physiological responses . For this purpose , we dissected third instar Drosophila larvae , prepared fillet preparations , and performed extracellular single-unit recordings . The foci of the IR-laser irradiation were targeted onto the proximal dendritic arbors ( the path length of branches between the IR-laser foci and soma was no longer than 100 µm ) . IR-laser irradiation of more than 30-mW laser output power evoked high-frequency firings in Class IV neurons ( Figure 1D and E: 30 mW , 40 . 6 ± 4 . 8 Hz; 40 mW , 83 . 9 ± 7 . 3 Hz; n = 9 cells ) . From our calibration of the IR laser , an irradiation of 1-mW raised the temperature by an increment of ~0 . 6°C in the focused region ( inset of Figure 1B ) ; thus , the estimated temperature was ~43°C for the 30-mW laser output power . This threshold temperature was close to that of previous experiments , where the jumps of the firing rate were detected at around 40°C with perfusion of the preheated solution into the recording chamber ( Xiang et al . , 2010 ) . We were intrigued by the firing patterns in response to IR irradiation of 40-mW laser power , which typically contained pauses that followed high-frequency spikes ( see 40-mW in Figure 1D ) ; and we characterized the underlying mechanistic details as described later . Our control or 'wild-type' larvae in the electrophysiological experiments carried a transgene of a Ca2+ probe that was expressed selectively in Class IV neurons , unless described otherwise . We investigated how the IR-laser irradiations evoked high-frequency firings in Class IV neurons by extracellular recording . We focused on the roles of two members of the TRPA channel family , dTrpA1 and Painless . Previous studies had shown that the activation of both of the two channels in the neurons was required to induce nocifensive escape behaviors ( Babcock et al . , 2011; Zhong et al . , 2012 ) , and that the expression of either of the channels was sufficient to elicit thermocurrents in cultured cell lines ( Sokabe et al . , 2008; Wang et al . , 2013 ) . Therefore , we expected that these molecules served as primary heat-activated cation channels in the neurons . dTrpA1 mutant ( dTrpA11 homozygotes ) and pain mutant ( pain3 homozygotes ) exhibited normal basal firing in the neurons ( data not shown ) , but showed significantly lower firing rates during the IR-laser irradiation ( Figure 1E: [dTrpA11] 30 mW: 0 . 5 ± 0 . 3 Hz , 40 mW: 43 . 8 ± 7 . 9 Hz , n = 10 cells; [pain3] 30 mW: 10 . 0 ± 3 . 0 Hz , 40 mW: 54 . 6 ± 4 . 8 Hz , n = 7 cells ) . Interestingly , dTrpA11 mutants displayed significantly longer latency time between the onset of the IR irradiation and the increase in firing rate when compared to the wild-type or pain3 larvae ( Figure 1F and G; Time to Max . Firing Rate in Figure 1G: [wild type] 45 . 0 ± 7 . 1 ms , [dTrpA11] 698 ± 89 . 5 ms; [pain3] 61 . 6 ± 16 . 4 ms , p < 5 . 7×10-6 , t-test with Bonferroni correction; our results of other allelic combinations are described in Figure 1—figure supplement 1 ) . The temporal pattern of the wild-type firing was not a simple summation of the two TRPA mutants; thus , we infer that the two heat-activated channels should function in a coordinated manner . Our results also strongly suggested that the firing responses of the wild-type neurons reflected intrinsic neuronal activities , dependent on the heat-activated channels and were not induced by photo damage ( see also Figure 1—figure supplement 2 ) . Although the occurrences of intracellular Ca2+ rises are often taken to be a hallmark of neural activity , it was not clear that the Ca2+ rises were faithfully indicative of the activation of Drosophila sensory neurons . We therefore decided to confirm the functional relationship between the Ca2+ rises and the neuronal firing responses . We first examined the dynamics of changes of cytoplasmic Ca2+ concentration during IR-laser irradiations . For this purpose , transgenic strains were generated that highly and specifically expressed an intramolecular FRET-based Ca2+ probe , TN-XXL ( Mank et al . , 2008 ) , in Class IV neurons ( see Methods and Figure 2—figure supplement 1 for details ) . Using ratiometric imaging , we measured the Ca2+ changes in the neurons in whole-mount larvae ( Figure 2A–E ) or fillet preparations ( Figure 2F ) . 10 . 7554/eLife . 12959 . 007Figure 2 . L-type VGCC-dependent global Ca2+ transients occur in Class IV neurons upon noxious thermal stimulation . Ca2+ responses of Class IV neuron ddaC expressing TN-XXL in whole-mount larvae , except for F ( fillet preparations ) , upon IR irradiation . Throughout this figure , somata were targeted ( red dotted circle in A ) . The output power of the laser was 38 mW except for C . Red squares above traces and magenta shadings in A , B , E , and F indicate the 1 s irradiations . ( A ) ( Left ) Time-projected image that is constructed by multiplying CFP and YFP images at every time point . Rectangles i–vii indicate the regions of interest ( ROIs ) . ( Right ) Global Ca2+ transients were detected in individual ROIs ( i–vii ) . The transient was also detected when we drew a ROI around the entire dendritic arbor excluding the soma ( whole dendrite ) . Ca2+ transient measured in somata displayed slower fluctuation compared to the dendritic ones ( soma ) . ( B ) Time courses of Ca2+ transients in whole dendrites . Gray lines indicate dendritic Ca2+ transients of 14 cells , and the blue line represents the averaged amplitude . The data from cells that did not generate Ca2+ transients were excluded . ( C ) Occurrence rate of dendritic Ca2+ transients when stimulated with different IR-laser output powers . We performed one trial per cell and the number of cells examined is displayed at the top of each bar . ( D ) Occurrence rate of dendritic Ca2+ transients was dramatically reduced in Class IV neurons overexpressing Kir2 . 1 ( ppk-Gal4 UAS-Kir2 . 1 ) or in neurons of larvae with mutations of the L-type VGCC gene ( Ca-α1DX10/AR66 and Ca-α1DX7/AR66 ) . *** p < 0 . 001 versus wild type by Fisher’s exact test . ( E ) Ca2+ fluctuations in dendrites of Ca-α1D mutant neurons ( Ca-α1DX10/AR66 ) . Gray lines indicate Ca2+ responses of 17 cells , and the blue line represents the averaged amplitude . ( F ) Ca2+ fluctuations in dendrites of fillet preparations when treated with 5 μM Nimodipine . Gray lines indicate Ca2+ responses of 11 cells , and the blue line represents the averaged amplitude . In E and F , we excluded data where ratiometric signals could not be continuously recorded due to movements of mounted larvae . DOI: http://dx . doi . org/10 . 7554/eLife . 12959 . 00710 . 7554/eLife . 12959 . 008Figure 2—source data 1 . A sample file of FRET imaging . DOI: http://dx . doi . org/10 . 7554/eLife . 12959 . 00810 . 7554/eLife . 12959 . 009Figure 2—figure supplement 1 . Configuration of the region-of-interest ( ROI ) in Ca2+ FRET imaging . To improve the fluorescent S/N ratio , original CFP time series were multiplied by the YFP series ( orange shading ) . Then the multiplied images ( CFP*YFP time series ) were projected on the time-axis ( CFP*YFP time projection image ) . Next the backgrounds of the image were equalized and the connected components ( such as dendrites ) were weighted by using the 'morphology filter' ( MATLAB , MathWorks ) , and then the image was binarized to make the 'binary mask image' . The ROIs were indicated by rectangles on the original CFP or YFP time series , and the regions of proper signals and of background noise were determined by using the 'binary mask image' within such ROIs . Genotype: 3×[ppk-TN-XXL] ( attP40 ) /+ . DOI: http://dx . doi . org/10 . 7554/eLife . 12959 . 00910 . 7554/eLife . 12959 . 010Figure 2—figure supplement 2 . Ca2+ transients evoked by IR-laser irradiation of dendrites . Ca2+ responses of Class IV neuron ddaC expressing TN-XXL in whole-mount larvae upon IR irradiation . Throughout this figure , dendrites were targeted ( e . g . red dotted circle in A ) . The output power of the laser was 42 mW . ( A ) ( Left ) Time-projected image that was constructed by multiplying CFP and YFP images at every time point . Rectangles i-viii indicate the regions of interest ( ROIs ) . ( Right ) Global Ca2+ transients were detected in individual ROIs ( i–viii ) . The transient was also detected when we drew a ROI around the entire dendritic arbor and excluded the soma ( whole dendrite ) . Ca2+ transient measured in the dendrites including the IR-laser focus ( trace labeled as 'viii' ) displayed a slower decay compared to the other dendritic regions ( i–vii ) . A red square above traces and a magenta shading indicate duration of 1 s irradiation . ( B ) Occurrence rate of dendritic Ca2+ transients when stimulated at dendrites located at different root distances from somata . We performed one trial per cell and the number of cells examined is displayed above each bar . Genotype: 3×[ppk-TN-XXL] ( attP2 ) /+ . DOI: http://dx . doi . org/10 . 7554/eLife . 12959 . 01010 . 7554/eLife . 12959 . 011Figure 2—figure supplement 3 . Effects of genotypes and pharmacological treatments on occurrence rates of dendritic Ca2+ transients , peak amplitudes of the transients , and/or slow Ca2+ rises in somata , when somata were irradiated . ( A ) Occurrence rates of dendritic Ca2+ transients in whole-mount preparations of the wild type , dTrpA1 , and painless mutants , which were evoked by IR-laser irradiations ( 38 mW , 1 s ) onto somata . We performed one trial per cell , and the number of cells of each genotype examined are indicated above each bar . ( B ) Peak amplitudes in somata of individual trials are plotted . Note that Ca2+ rises in somata were slow in contrast to the sharp dendritic Ca2+ transients when somata were irradiated ( cf . Figure 2A; 'soma' ) . The magenta dots in B indicate occurrences of dendritic Ca2+ transients; the gray dots represent those with no apparent Ca2+ transient . The peak amplitudes of these slow Ca2+ rises were significantly decreased in dTrpA1 null mutants . This result was suggestive of a contribution of a dTrpA1-mediated physiological process , although an exact mechanism underlying such slow fluctuations remains unclear . ( C–E ) Occurrences of dendritic Ca2+ transients in the presence of TTX or Thapsigargin treatment , which depletes the cytoplasmic Ca2+ store . We focused the IR-laser ( 38 mW , 1 s ) onto the somata of Class IV neurons in the fillet-mounted larvae , which were pretreated with TTX and Thapsigargin ( C and D , respectively ) . Blue traces indicate the amplitudes of Ca2+ transients of controls , whereas magenta traces indicate those with the drugs . The semitransparent traces represent amplitudes of Ca2+ transients of individual cells , and the solid traces show the average values of each group . Error bars indicate the standard deviations of the peak amplitudes of Ca2+ transients . The TTX-treatment had no effect on the occurrence and amplitude of Ca2+ transients ( C ) , whereas the Thapsigargin-treatment only decreased the amplitude of Ca2+ transients ( D ) . The effects of TTX was confirmed by loss of spikes . The application of TTX is indicated by a green bar above the trace ( E ) . Fisher’s exact test ( A ) , ANOVA and Dunnet post-hoc analysis ( B ) , and Student’s unpaired t-test ( C and D ) were performed and statistical significances were assigned , *p < 0 . 05 , **p < 0 . 01 , ***p < 0 . 001 . Genotype: 3×[ppk-TNXXL] ( attP40 ) /+ . To further address the contribution of Ca2+-induced Ca2+-release ( CICR ) from the cytoplasmic Ca2+ store to the formation of Ca2+ transients , we examined mutants for the inositol triphosphate receptor gene ( Itp-r83A05616/90B . 0 ) ( Venkatesh and Hasan , 1997 ) or the Ryanodine receptor gene ( RyRK04913/16 ) ( Casas-Tinto et al . , 2011; Gao et al . , 2013 ) . Both mutants showed Ca2+ transients with occurrence and amplitude that were not significantly different from those in the wild type ( data not shown ) , showing that the role of CICR was negligible . DOI: http://dx . doi . org/10 . 7554/eLife . 12959 . 011 When the soma was irradiated with the IR laser , a large increase in Ca2+ was observed over the entire dendritic tree , which we designated as dendritic Ca2+ transients or simply Ca2+ transients ( Figure 2A , B and Video 1 ) . The rise of the Ca2+ transients emerged simultaneously in both proximal and distal dendrites , and decayed exponentially . Dendritic Ca2+ transients were also visualized by using GCaMP5G ( Akerboom et al . , 2012 ) as an indicator ( Video 2 ) , but we used TN-XXL throughout this study for monitoring Ca2+ dynamics in both whole-mount larvae and fillet preparations , which allowed more robust quantitative analysis in the presence of perturbations along Z-axis motions by body wall muscles . 10 . 7554/eLife . 12959 . 012Video 1 . Dendritic Ca2+ transients ( 1 ) . Related to Figure 2 . Ratiometric pseudocolor time-lapse images of TN-XXL-expressing Class IV neuron ddaC in a whole-mount preparation of a larva . The time course is indicated in the upper left corner of each image , and the duration of IR irradiation ( 1 s at 38 mW ) is displayed by red squares in upper right corner of relevant images . Genotype: 3×[ppk-TN-XXL] ( attP2 ) / 3×[ppk-TN-XXL] ( attP2 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 12959 . 01210 . 7554/eLife . 12959 . 013Video 2 . Dendritic Ca2+ transients ( 2 ) . Related to Figure 2 . Intensiometric unicolor time-lapse images of GCaMP5G-expressing Class IV neuron ddaC in a whole-mount preparation of a larva . The time course is indicated in the upper left corner of each image , and the duration of IR irradiation ( 1 s at 38 mW ) is displayed by red squares in upper right corner of relevant images . Genotype: UAS-GCaMP5G/+; ppk-Gal4/+ . DOI: http://dx . doi . org/10 . 7554/eLife . 12959 . 013 The Ca2+ transients occurred in all-or-none fashion upon IR-laser irradiations; suggesting that the magnitude of the thermal stimuli was converted into another value of information ( e . g . dendritic membrane potential ) , and the value higher than a presumptive threshold evoked Ca2+ transients . The probabilities of the occurrence of Ca2+ transients were dependent on the output powers of the IR-laser , which arose abruptly at around 36 mW ( Figure 2C ) , so this high occurrence is one advantage of irradiation to the soma . Furthermore , the probabilities were significantly lowered in the neurons of dTrpA1 or pain mutants ( Figure 2—figure supplement 3A ) , indicating that Ca2+ transients also reflect normal neuronal activities that depend on the heat-activated channels . Irradiation to the dendritic arbor also evoked Ca2+ transients , although this required higher IR-laser power compared to the stimulation of the soma ( Figure 2—figure supplement 2 ) . Once a Class IV neuron elicited a transient , that neuron then responded more weakly to the repetitive IR-laser irradiations onto the same location in the cell ( data not shown ) . In contrast to the sharp Ca2+ rises throughout dendrites upon IR irradiations to somata , Ca2+ fluctuations within the somata were relatively slow ( Figure 2A; 'soma' and Figure 2—figure supplement 3B ) . We next explored the molecular basis of the dendritic Ca2+ transient . Excitability of Class IV neurons was prerequisite for this Ca2+ transient , as shown by the fact that the occurrence of the transients was completely blocked by overexpression of the inward-rectifier potassium channel Kir2 . 1 ( Figure 2D ) , which hyperpolarizes neurons and decreases their resting membrane potential ( Hardie et al . , 2001 ) . This raised the possibility that the transient was caused by Ca2+ influx from extracellular space ( see another possibility discussed in Figure 2—figure supplement 3 and its legend ) and that some of the voltage-gated Ca2+ channels ( VGCCs ) were involved . To verify these possibilities , we screened genes encoding α subunits of VGCCs for those necessary for the occurrence of Ca2+ transients; specifically , we recorded and analyzed probabilities and amplitudes of the Ca2+ transients in various VGCC loss-of-function mutants . A dramatic phenotype was found in mutants of the L-type VGCC gene ( Ca-α1DX7/AR66 and Ca-α1DX10/AR66 ) ( Eberl et al . , 1998 ) ; normal Ca2+ transients hardly occurred in those mutants ( Figure 2D and E ) . A similar defect was observed in the wild type when a selective inhibitor of L-type VGCC , Nimodipine ( Gielow et al . , 1995; Grey and Burrell , 2010 ) , was applied ( Figure 2F ) . These results showed that L-type VGCC-mediated excitability is required for the occurrence of Ca2+ transients in Class IV neurons ( see our negative results of other types of the Drosophila VGCC family in 'Materials and methods' ) . As for mammalian neurons , it has been well known that there is a link between changes in Ca2+ levels , which are regulated by L-type VGCC , and cell-specific gene expression ( Greer and Greenberg , 2008 ) . Therefore , the expression level of dTrpA1 and/or pain could have been lowered in the L-type VGCC mutants . To test this possibility , we performed real-time PCR to quantitate the mRNA levels of dTrpA1 and pain in neural tissues of Ca-α1D mutants , but found no significant differences between the Ca-α1D mutant and the wild type ( data not shown ) . The Ca2+ transient in Class IV neurons was reminiscent of that in rat neocortical pyramidal neurons , which are also detectable throughout the entire dendritic arbor ( Schiller et al . , 1995 ) . The occurrence of this transient in the pyramidal neurons requires voltage-gated Na+ channels ( VGSCs ) that are distributed along dendritic branches . In our system , however , the addition of Tetrodotoxin ( TTX ) had no effect on the dendritic Ca2+ transients in Class IV neurons ( Figure 2—figure supplement 3C and E ) , indicating that VGSCs are required neither for the occurrence of Ca2+ transients nor for the regenerative propagation of membrane potential along the dendritic branches . To elucidate the relationship between Ca2+ transients and neuronal activities , we performed extracellular single-unit recording and the ratiometric Ca2+ imaging simultaneously ( Figure 3A ) . Throughout our simultaneous recordings in this study , the IR stimulation was given to the proximal dendrite . Under 40-mW IR-laser irradiation , peak amplitudes of Ca2+ transients did not display a strong correlation with maximum firing rates ( Figure 3B; p > 0 . 46 , ρ = 0 . 17 , Spearman’s rank correlation test; n = 20 cells ) . In our analyses , the firing rates were computed by sliding a rectangular window function along the spike train with Δt = 400 ms ( see 'Materials' and methods for details ) . Even when we increased the width of the sliding window ( Δt ) from 10 to 1000 ms in steps of 10 ms , there were no statistically significant correlations between the estimated maximum firing rates and the peak amplitudes of Ca2+ transients ( p ≥ 0 . 1; Figure 3—figure supplement 1A ) . 10 . 7554/eLife . 12959 . 014Figure 3 . Simultaneous recordings of firing responses and dendritic Ca2+ transients in fillet preparations of larvae upon noxious thermal stimulation . Data obtained from fillet preparations of control larvae ( A–F ) , those from L-type VGCC-blocked larva ( G and H ) , and their comparisons ( I–L ) . In each trial , 40-mW IR laser was focused onto a proximal dendritic branch for 1 s ( indicated by red bars above traces , and magenta shadings in ( A , D , G , I and H ) and the Ca2+ transient was detected from a ROI that was set on a distal branch 100 μm away from the focus of the IR-laser beam . ( A and G ) Representative recordings of a control larva ( A ) and a larva treated with 5 μM Nimodipine ( G ) . Dendritic Ca2+ fluctuations for 20 s ( top ) and an enlarged trace of an extracellular recording for 1 . 675 s including the 1 s duration of the IR-laser irradiation ( middle ) . The time derivative of firing rate fluctuation based on the spike density estimation with the Gaussian kernel ( σ = 25 ms ) and the normalized threshold settings ( bottom ) . The peak of firing rate fluctuation is defined as in the text and marked with red circles . ( B ) Peak amplitudes of Ca2+ transients are plotted against maximum firing rates ( defined in Materials and methods ) . ΔRpeak is defined as the averaged ratio of 400–700 ms after the cessation of IR-laser stimulation , which is employed as a representative value of each trial . See also the legend of D . A significant correlation was not observed ( p > 0 . 46 , ρ = 0 . 17; Spearman's rank correlation test; n = 20 cells ) . ( C ) The definition of 'unconventional spikes' ( US ) . Signatures of US were extracted according to the following two conditions: ( 1 ) the first interspike interval ( ISI ) of three sequential spikes was less than 9 ms , and ( 2 ) the second ISI was longer than 20 ms . Then , a pair of the first and second spikes within each sorted triplet was designated as 'US' . ( D and H ) Raster plots of firing ( left ) and magnitudes of ΔRpeak ( right ) . ( D ) All trials in panel ( B ) , where control larvae were used , are sorted in a descending order of magnitude of ΔRpeak ( right gray bars ) . Red raster lines indicate US . Six trials that elicited no US are labeled by gray color in B . ( H ) When fillet preparations were treated with 5 μM Nimodipine , none of the 11 trials elicited US during IR irradiation . ( E ) Amplitudes of ΔRpeak are plotted against total US numbers for each trial . Short horizontal red bars indicate the averages of ΔRpeak , and the red line is a linear regression of plotted data ( p < 1 . 9 × 10–8 , ρ = 0 . 91; Spearman’s rank correlation test; n = 20 cells ) . Our original definition of US in fact displayed a suboptimal correlation to the peak amplitude . We redefined USs using different parameter sets ( lengths of the first and second interspike intervals ) and calculated Spearman’s correlation coefficients to estimate how well the total US number is correlated with the peak amplitude . As a result , our original definition in C in fact displayed a suboptimal correlation to the peak amplitude ( data not shown ) . ( F ) Comparative analysis of ISI between US and between non-US events . For trials with at least one US , the minimum length of ISI between US in each trial was compared to that between non-US ( p < 5 . 2 × 10–6; paired t-test; n = 13 cells ) . ( I–L ) Temporal patterns of firing rates ( I ) and histograms of the total US number ( J ) or the total peak number ( L ) in control and Nimodipine-treated larva ( gray and magenta , respectively ) . See also the legend of A . ( K ) A plot of total peak numbers versus total US numbers for each trial in the control . Data in I are presented as mean ± s . e . m . DOI: http://dx . doi . org/10 . 7554/eLife . 12959 . 01410 . 7554/eLife . 12959 . 015Figure 3—figure supplement 1 . Simultaneous recordings of firing responses and dendritic Ca2+ transients in fillet preparations of larvae upon noxious thermal stimulation . ( A ) As shown in Figure 3B , peak amplitudes of Ca2+ transients did not display a strong correlation with maximum firing rates , which were defined by one window size ( the sliding window/∆t = 400 ms ) , when 40-mW IR-laser irradiation was given to proximal dendrites . We increased ∆t between 10 ms and 1000 ms in steps of 10 ms; and showed that there were no statistically significant correlations between the estimated maximum firing rates and the peak amplitudes of Ca2+ transients ( Spearman rank correlation test ) . The red line indicates p = 0 . 1 . ( B ) In two out of 20 datasets , extracellular displayed irregular waveforms representing the fluctuation of membrane conductance during the US , suggesting the presence of the excitatory currents . The red arrows indicate the spikes identified as the components of US . This low success rate of detecting the irregular waveform was possibly due to the spatially restricted membrane area of the soma where the recording electrode can detect the minute change of cellular conductance . We hypothesized that these differential patterns may encode command signals and that those signals should be transmitted to and decoded by target neuron ( s ) ( see 'Discussion' ) . ( C ) The spike density representation of the firing patterns , which are indicated by raster plots in Figure 3D . Traces of spike densities were estimated using a Gaussian filter kernel with σ = 12 ms . All traces are sorted in descending order of magnitude of ΔRpeak . The magenta traces indicate that the amplitudes of ΔRpeak are greater than 10%; the blue traces indicate those below the threshold . The red bar above the traces represents the timing of IR-laser irradiation . Genotype: 3×[ppk-TNXXL] ( attP2 ) / 3×[ppk-TNXXL] ( attP2 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 12959 . 01510 . 7554/eLife . 12959 . 016Figure 3—figure supplement 2 . Simultaneous recordings of firing responses and dendritic Ca2+ transients in fillet preparations of larvae expressing the lowest-threshold form of dTrpA1 . ( A–D ) Simultaneous recordings of dendritic Ca2+ transients and firing responses that were obtained from a single dTrpA1-A-expressing Class IV neuron . One second IR-laser irradiation ( red bars above traces in A ) was targeted to the proximal dendritic arbors , and the output power was raised from 10 to 18 mW in steps of 2 mW . ( A ) Dendritic Ca2+fluctuations ( top ) and an enlarged trace of extracellular recording for 1 . 4 s ( from –0 . 2 to 1 . 2 s ) including the duration of IR-laser irradiation ( bottom ) . Colors represent trials at the indicated power settings ( e . g . the green traces are from the same trials with 14-mW IR-laser power ) . The red arrows indicate the occurrences of US . The following three measurements of two neurons are plotted in ( B–D ) : the maximum firing rates ( Max . Firing Rate [Hz]; B ) , the peak amplitudes of ΔR ( ΔRpeak [%]; C ) , and the number of occurrences of US ( US number; D ) . The firing rate was almost linearly correlated to the laser power ( B ) ; on the other hand , the peak amplitude of the Ca2+ transient and the number of accompanied USs abruptly rose at power settings above 14 mW ( n = 3 cells; C and D ) . These occurrences of both the 'burst and pause' firing pattern and the Ca2+ transient above a certain threshold in the TrpA1-A+ neurons is comparable to the responses of the wild-type neurons ( compare Figure 3—figure supplement 2A with Figure 1D , and Figure 3—figure supplement 2B with Figure 1E ) , except for the fact that the TrpA1-A+ neuron continued to respond to the repetitive IR-laser irradiations onto the identical location within the dendritic tree . ( E ) Effects of Nimodipine on the firing responses of neurons ectopically expressing dTrpA1-A , tested at three different output powers . The firing responses are presented as the spike density estimation computed by using a Gaussian filter kernel ( σ = 12 ms ) as in Figure 3—figure supplement 1C . The peaks and troughs of the spike density fluctuations clearly show the occurrences of USs clearly in the control neurons ( left , blue traces ) . By contrast , the traces from Nimodipine-treated neurons display much flatter transitions ( right , magenta traces ) . The red bars above the traces and the magenta shading indicate the timings of IR-laser irradiations . Genotype: w; 3×[ppk-TNXXL] ( attP40 ) /+; UAS-dTrpA1-A/ppk-Gal4 . DOI: http://dx . doi . org/10 . 7554/eLife . 12959 . 01610 . 7554/eLife . 12959 . 017Figure 3—figure supplement 3 . Dual recordings of firing responses from the soma and the axon bundle . We performed dual extracellular recordings from the soma of a Class IV neuron and its axon bundle ( black waveforms and magenta waveforms in ( A–D ) , respectively ) . ( A ) ( Left ) A schematic diagram of a subset of sensory neurons in a larval abdominal hemisegment , and pipets for the extracellular recording from the soma of a Class IV neuron ( gray electrode ) and the axon bundle ( magenta electrode ) . ( Right ) An example of dual recordings of spontaneous activities . The upper black trace shows the recording from a soma; the lower magenta one displays that from the axon bundle . Spikes of variable amplitudes were observed in the trace from the axon bundle , which were presumably evoked in sensory neurons other than the Class IV . Genotype throughout this figure: w; 3×[ppk-TNXXL] ( attP40 ) /+; UAS-dTrpA1-A/ppk-Gal4 . ( B ) The superposition of multiple waveforms from the dual recordings . The traces are aligned by the peak position of somatic spikes ( black ) . Voltage traces from individual axon bundles ( magenta ) always contained spikes lagging those from the somata by an invariant latency . ( C ) An example of firing responses dually recorded from the soma ( black ) and the axon bundle ( magenta ) upon IR-laser irradiation of the proximal dendritic arbors ( 18 mW , 1 s; red bar above the traces ) . In the extracellular recording from the soma , the red asterisk identifies the second spike of the first US ( cf . Figure 3C ) . We focused on voltage traces right after US to see whether US from soma were followed by pauses in firing from both the soma and the axon or not . Although it was difficult to detect such coinstantaneous pauses , due to the interference of predominant spikes of sensory neurons other than Class IV , we did find several instances where firing was paused from both an axon bundle and the corresponding soma . ( D ) The averaged waveforms of a part of the dual extracellular recordings that include the second spike of the US in C ( red asterisks in C and D ) . The invariant dendritic location of the same cell was irradiated repeatedly by IR-laser as in C . The recordings from the soma of the same neuron were aligned at the spikes ( red asterisk ) , and then the averaged values were computed . ( Bottom ) The solid magenta line indicates the average and the magenta shading represents the standard deviations of trials ( n = 14 trials ) . When all of the voltage traces of axon bundles were aligned relative to the peak amplitudes of the first US from somata , the mean change of the voltage traces disclosed no additional spikes behind the US for 50 ms . DOI: http://dx . doi . org/10 . 7554/eLife . 12959 . 017 Upon closer inspection of the firing patterns with Ca2+ transients , we noticed characteristic pauses in firing that followed the high-frequency spikes ( Figure 3A ) . Thus , we defined these spikes as a pair of unconventional spikes ( US ) based on the interspike interval ( Figure 3C ) , and designated the firing pattern composed of the pause and US tentatively as a 'burst and pause' pattern . When we sorted datasets of the firing patterns in a descending order of the peak amplitude of Ca2+ transients , the occurrence of US showed a high correlation with the peak amplitude ( Figure 3D and E; p < 1 . 9×10–8 , ρ = 0 . 91; Spearman’s rank correlation test; n = 20 cells ) . We also performed a post hoc parameter fitting analysis on the basis of our finding that the occurrence of the US showed a high correlation with the peak amplitude of Ca2+ transients ( Figure 3E; see also its legend ) . More importantly , in each dataset , the minimal interspike interval within the US was invariably shorter than that within non-US firings ( Figure 3F; p < 5 . 2×10–6; paired Student’s t-test; n = 13 cells ) , indicating that large excitatory currents existed just prior to the pause in firing ( see also Figure 3—figure supplement 1 ) . To address how inhibition of L-type VGCC affects the firing rate and the occurrence of Ca2+ transients and US , we recorded Class IV neurons in the presence of Nimodipine . Ca2+ transients and US were hardly detected in the spike trains ( Figure 3G , H , and J; [Control] 1 . 6 ± 0 . 32 , [Nimodipine] 0 ± 0; p = 0 . 001 , Wilcoxon’s rank-sum test; n = 20 cells [Control] and n = 11 cells [Nimodipine] ) ; in contrast , the firing rate was not decreased compared to that of the control and showed its peak around 100 Hz ( Figure 3I ) . In other words , firing patterns were modified as if the 'burst and pause' pattern had been replaced with conventional high-frequency spike trains , indicating that the L-type VGCC is essential for the generation of US . We also found that firing rates decreased for short periods following the pauses ( data not shown ) ; therefore , we assumed that an after-hyperpolarization current is generated during the pauses ( see 'Discussion' ) . Collectively , our findings strongly suggest that dendritic Ca2+ transients we observed in response to heat produced by the IR laser are generated by accumulations of L-type VGCC-dependent Ca2+ spikes in Class IV neurons . The ectopic expression of thermoTRP has been widely used as a tool to achieve conditional controls of neuronal activity in vivo by modulating the ambient temperature ( Bernstein et al . , 2012 ) . This is probably also the case with the ectopic expression of the lowest-threshold isoform of dTrpA1 , one of the thermoTRPs described above , in Class IV neurons . dTrpA1 encodes four distinct isoforms that display different temperature sensitivities ( Zhong et al . , 2012 ) ; among them , the dTrpA1-A isoform ( Viswanath et al . , 2003 ) ( also designated as dTrpA1 . K ) ( Hamada et al . , 2008 ) has the lowest threshold ( ~29°C ) and shows little desensitization by repetitive heat-activations ( Pulver et al . , 2009 ) . Forced expression of dTrpA1-A in Class IV neurons allows larvae to respond to a 30°C heat probe and elicit the nocifensive rolling behavior ( Babcock et al . , 2011; Zhong et al . , 2012 ) , possibly by endowing the neurons with higher sensitivity to the moderate-temperature stimulus . Consistent with this speculation , our recordings demonstrated that TrpA1-A-expressing neurons ( TrpA1-A+ neurons ) evoke both Ca2+ transients and the unique firing pattern below 20 mW ( Figure 3—figure supplement 2A ) , to which the wild-type neurons were insensitive ( Figure 1D ) , and all of other results ( Figure 3—figure supplement 2B–E ) strengthened the possibility that the cellular responses of Class IV neurons to IR irradiation reflected the intrinsic membrane properties . Are the spikes detected in somata by extracellular single-unit recordings faithfully propagated through their axons ? This had to be verified because of concerns in previous studies . For example , when neuronal activities with Ca2+ spikes are recorded extracellularly , the intracellularly detected spikes sometimes disappear in the voltage traces of extracellular single-unit recordings of axons ( Schonewille et al . , 2006 ) . Another instance shows that intracellularly detected spikes are not always propagated along the axon ( Monsivais et al . , 2005 ) . To explore these possibilities , we performed extracellular single-unit recordings from both the soma of a Class IV neuron and the axon bundle , including its own axon , and analyzed the correspondence between the firing patterns from somata , which coincided with US , and those from axon bundles ( Figure 3—figure supplement 3A ) . Spikes from a single axon bundle can include those of not only Class IV but also other sensory neurons . Nonetheless , our findings indicate that the spikes recorded extracellularly from Class IV soma propagate faithfully through its axon ( Figure 3—figure supplement 3B–D , see also the legends ) . We hypothesized that the unique firing patterns accompanied by Ca2+ influx should be output signals provoking the robust nocifensive rolling behavior . To verify this hypothesis , we conducted a series of physiological and behavioral assays of Ca-α1D mutant or knockdown larvae ( Figure 4 and Figure 4—figure supplement 1 ) . We first examined the physiological responses of Ca-α1D-knockdown Class IV neurons . Both the occurrence of the Ca2+ transient and that of the unique firing pattern were suppressed in the neurons ( Figure 4A and B ) without a dramatic decrease in firing rate ( Figure 4—figure supplement 1 ) , as was seen in the case of pharmacological blockade of L-type VGCC ( Figure 3G–I ) . We also observed that the total US number was significantly decreased in Ca-α1D-knockdown neurons compared to controls ( Figure 4C; [white RNAi GL00094] 1 . 55 ± 0 . 20 total US number , [Ca-α1D RNAi HMS00294] 1 . 00 ± 0 . 00; p = 0 . 0055; Wilcoxon’s rank-sum test ) . 10 . 7554/eLife . 12959 . 018Figure 4 . Responses of Class IV neurons with a knockdown of Ca-α1D and behavioral analysis of the knockdown larvae upon noxious thermal stimulation . ( A–D ) Responses of control Class IV neurons ( Control , n = 18 and white RNAi GL00094 , n = 11 ) and Ca-α1D knockdown neurons ( Ca-α1D RNAi JF01848 , n = 10 and Ca-α1D RNAi HMS00294 , n = 9 ) to 1 s irradiation of a 48-mW IR laser . ( A ) Raster plots of firing ( left ) and magnitudes of ΔRpeak corresponding to Ca2+ transients ( right ) . Trials are sorted in a descending order of the magnitude of ΔRpeak ( right bars ) . Red bar indicates the 1 s irradiation and red raster lines are US . ( B ) Occurrence rate of dendritic Ca2+ transients in each genotype ( [Control] 72 . 2 ± 17 . 6% , [white RNAi GL00094] 81 . 8 ± 19 . 4% , [Ca-α1D RNAi JF01848] 30 . 0 ± 24 . 2% , [Ca-α1D RNAi HMS00294] 0 . 0 ± 0 . 0%; mean ± 95% confidence interval [Clopper-Pearson method] ) . ( C and D ) Histograms of the total US number ( C ) or the total peak number ( D ) during the irradiation in the control and the knockdown neurons . The total peak number of firing rate fluctuation is defined as in the text . See also Figure 3A and its legend . ( E and F ) The distribution of nocifensive escape locomotion latency for wandering third-instar larvae of the wild type and the Ca-α1D mutant that were stimulated with a 47°C probe . The numbers of larvae tested are indicated and data are presented as percentage . ( E ) The distributions of the wild-type and Ca-α1D mutant larvae ( mean latency: [w1118] 1 . 78 s , [Ca-α1D AR66/X10] 6 . 12 s; Wilcoxon rank-sum test ) . ( F ) The distributions of control larvae and larvae with Class IV neuron-specific white or Ca-α1D knockdown . The Ca-α1D mutant and the Ca-α1D knockdown larvae showed significant delayed nocifensive responses in comparison to the wild type ( E ) and the controls ( Control and white RNAi GL00094 ) ( F ) , respectively . p values in F are those of the test between white RNAi GL00094 and each of the Ca-α1D knockdowns ( mean latency: [Control] 3 . 87 s , [white RNAi GL00094] 3 . 39 s , [Ca-α1D RNAi JF01848] 5 . 21 s , [Ca-α1D RNAi HMS00294] 5 . 25 s; Wilcoxon rank-sum test with Bonferroni correction ) . ( G ) Percentage of larvae avoiding blue light ( 457–487 nm; 0 . 72 mW/mm2; [Control] 59 . 6 ± 8 . 0% , [Ca-α1D RNAi JF01848] 53 . 9 ± 10 . 3% , [Ca-α1D RNAi HMS00294] 51 . 0 ± 8 . 5%; mean ± 95% confidence interval [Clopper-Pearson method] ) . We employed GMR-hid to ablate Bolwig’s organs in the photo-avoidance assay . p values are indicated ( two-tailed Fisher’s exact test ) . Sixty-five to 104 larvae were tested for each condition . DOI: http://dx . doi . org/10 . 7554/eLife . 12959 . 01810 . 7554/eLife . 12959 . 019Figure 4—source data 1 . The distributions of nocifensive escape locomotion latency for wandering third-instar larvae of the wild-type and Ca-α1D mutant larvae ( Columns A , B ) and The distributions of control larvae and larvae with Class IV neuron-specific white or Ca-α1D knockdown ( Column D–G ) . DOI: http://dx . doi . org/10 . 7554/eLife . 12959 . 01910 . 7554/eLife . 12959 . 020Figure 4—figure supplement 1 . Simultaneous recordings of firing responses and dendritic Ca2+ transients in fillet preparations of Ca-α1D knockdown larvae upon noxious thermal stimulation . ( A and B ) Responses of control Class IV neurons ( Control and white RNAi GL00094 ) and Ca-α1D knockdown neurons ( Ca-α1D RNAi JF01848 and Ca-α1D RNAi HMS00294 ) to 1 s irradiation of a 48-mW IR laser . ( A ) Time courses of firing rates of the wild type and the Ca-α1D knockdown larvae upon noxious thermal stimulations . ( B ) The max . firing rate ( defined in Methods ) of the Control , white RNAi GL00094 , Ca-α1D RNAi JF01848 , and Ca-α1D RNAi HMS00294 showed normal firing rates during the IR-laser irradiation ( [Control] 96 . 2 ± 5 . 1 Hz , n = 18 cells; [white RNAi GL00094] 100 . 0 ± 8 . 4 Hz , n = 11 cells; [Ca-α1D RNAi JF01848] 85 . 5 ± 4 . 1 Hz , n = 10; [Ca-α1D RNAi HMS00294] 82 . 5 ± 7 . 7 Hz , n = 9 ) . ( C ) Quantification of timing of maximum excitation ( 'Time to Max . Firing Rate' as defined in Methods ) of Class IV neuron of each genotype . ( Time to Max . Firing Rate: [Control] 47 . 5 ± 16 . 8 ms , [white RNAi GL00094] 37 . 8 ± 11 . 9 ms; [Ca-α1D RNAi JF01848] 21 . 1 ± 3 . 3 ms; [Ca-α1D RNAi HMS00294] 23 . 5 ± 1 . 6 ms ) . Data are presented as mean ± s . e . m ( A–C ) . No statistically significant difference was observed between the Ca-α1D RNAi neurons and wild type or white RNAi GL00094 by unpaired t-test with Bonferroni correction ( B and C ) . Genotypes: ( Control ) w1118/+; ppk-Gal4/+; 3×[ppk-TNXXL] ( attP2 ) /+ ( white RNAi GL00094 ) ppk-Gal4/+; 3×[ppk-TN-XXL] ( attP2 ) /UAS-white RNAi GL00094 ( Ca-α1D RNAi JF01848 ) ppk-Gal4/+; 3×[ppk-TNXXL] ( attP2 ) /UAS-Ca-α1D RNAiJF01848 ( Ca-α1DRNAi HMS00294 ) ppk-Gal4/+; 3×[ppk-TNXXL] ( attP2 ) /UAS-Ca-α1DRNAiHMS00294DOI: http://dx . doi . org/10 . 7554/eLife . 12959 . 020 We then examined how Ca-α1D mutant larvae ( Ca-α1DX10/AR66 ) responded to the heat probe and found that they displayed a significantly delayed response ( Figure 4E ) , almost identical to dTrpA1 mutants ( Babcock et al . , 2011; Neely et al . , 2011 ) ( our data not shown ) . However it was reported that synaptic transmission at the neuromuscular junction ( NMJ ) is impaired in the Ca-α1D mutant larvae due to abrogation of the Ca2+ influx in both muscles and motor neurons ( Ren et al . , 1998; Worrell and Levine , 2008 ) , so it was difficult to attribute the behavioral defect entirely to the role of L-type VGCC in Class IV neurons . Therefore , we knocked down expression of L-type VGCC selectively in Class IV neurons as was previously done ( Kanamori et al . , 2013 ) . The phenotypic defect of the knockdown larvae was less dramatic than that of the whole-body mutant; nonetheless , the knockdown larvae displayed significantly delayed response in the rolling behavior than the control larvae ( Figure 4F ) . These data support our hypothesis that L-type VGCC-dependent unique firing patterns in Class IV neurons are necessary for the highly penetrant induction of the larval nocifensive rolling behavior . To explore the possibility of causal connections between the 'burst and pause' firing pattern and the rolling behavior , we performed a series of optogenetic activation experiments ( Figure 5 ) . It has been shown that larvae expressing channelrhodopsin-2 ( ChR2 ) in Class IV neurons mimic the rolling behavior when ChR2 is activated by continuous light exposure ( Hwang et al . , 2007 ) . We have established our assay of the induced rolling by expressing a faster kinetic variant of ChR2 , ChIEF in Class IV neurons ( Mattis et al . , 2012 ) . To address whether ChIEF-dependent intermittent firing of Class IV neurons is sufficient to elicit the rolling behavior , whereas continuous firing is not , we tried to recapitulate the naturalistic 'burst and pause' firings in the background of a Class IV-specific Ca-α1D-knockdown by activating ChIEF using square pulses of light . On and off durations of light pulses were set to be comparable to 'burst' and 'pause' timings ( 100 ms and 100 ms , respectively; Figure 5A top ) . 10 . 7554/eLife . 12959 . 021Figure 5 . Optogenetic activation of Class IV neurons . ( A ) Data obtained from fillet preparations of larvae where Ca-α1D gene was knocked down specifically in Class IV neurons . ( Top ) Raster plots of firing . A faster kinetic variant of ChR2 , ChIEF , was expressed selectively in the neurons and activated with a blue LED continuously ( 'Continuous activation'; n = 9 ) or in a pulsatile fashion ( 'Intermittent activation'; n = 9 ) . ( Bottom ) The time derivative of firing rates under the continuous activation ( gray ) or intermittent activation ( orange ) . The peak of firing rate fluctuation is defined as in the text and marked with red circles . ( B ) Percentage of larvae rolling their bodies upon optogenetic activation ( [Continuous] 65 . 9 ± 11 . 5% , n = 75; [Intermittent] 88 . 5 ± 7 . 6% , n = 95; mean ± 95% confidence interval [Clopper-Pearson method] ) . p-Value is indicated ( two-tailed Fisher’s exact test ) . ( C ) A histogram of the total peak number during the continuous activation ( gray ) or intermittent activation ( orange ) . DOI: http://dx . doi . org/10 . 7554/eLife . 12959 . 021 With these illumination parameters , either continuous or intermittent firing patterns were successfully evoked in the knockdown neurons of fillet preparations ( top of Figure 5A ) . In live animals , Ca-α1D-knockdown larvae displayed rolling when the continuous firing pattern was evoked , and the rolling occurred more efficiently by the pulsatile optogenetic activation ( Figure 5B; [Continuous] 65 . 3 ± 9 . 2% , n = 75 , [Intermittent] 86 . 3 ± 5 . 9% , n = 95 , mean ± 95% confidence interval based on Clopper-Pearson method; p = 0 . 0017 , Fisher’s exact test ) . Although the probability of rolling was rather high in the continuous activation condition ( see 'Discussion' ) , this result is , nevertheless , consistent with the idea that the intermittent firing pattern facilitates the rolling behavior . However , it should be stressed that the intermittent pattern produced under our optogenetic experiment was distinct from the pattern of the wild-type neuron upon noxious thermal stimulations according to two quantitative features . First , the maximum firing rate of the optogenetic pattern was 40~50 Hz ( data not shown ) and lower than that of wild-type neurons ( ~80 Hz; Figure 1E ) . Second , none of the spike trains generated in our optogenetic condition fully met the definition of US ( Figure 3C ) . Therefore , we reconsidered our idea that the occurrence of the US is a reliable metric for the artificial firing pattern , and we searched for other metrics that would explain all of the results of our three experimental settings: the effect of Nimodipine application ( Figure 3D and H ) , the Ca-α1D-knockdown phenotypes ( Figure 4A ) , and the optogenetics ( the top of Figure 5A ) . We noticed that fluctuations of firing rate were prominent around the time when each US occurred ( the recording trace of Δ Firing rate of Figure 3A; see also Figure 3D ) . To express such fluctuations quantitatively , we first speculated that the number of the peaks of the firing rate during IR stimuli might be an appropriate metric and computed the firing rate by using the Gaussian kernel method ( Figure 3—figure supplement 1C; see also below ) . Although we defined the peak of the firing rate using different parameter sets , we did not find statistically significant differences in the peak number between the controls and the samples in the above three experimental settings ( data not shown ) . Next , we focused our attention on the local steepness of the firing rate fluctuation during heat stimulation because a precipitous change in the firing rate is one essential component of burst firing . To quantify 'up' and 'down' of the firing rate fluctuation , we calculated the time derivative of the firing rate fluctuation during IR irradiation , defined peaks of firing rate fluctuation , and scored the total number of the peaks ( 'total peak number' ) as follows: first , the firing rate estimation was computed by using the Gaussian kernel method ( σ = 25 ms ) ; second , a threshold was configured as 0 . 5 of the primal peak value; finally , peaks of the time derivative above the threshold were specified ( red circles in the bottom graphs of Figure 3A , G and 5A ) . With this analysis , the 'total peak number' showed a significant correlation with the total US number in control neurons ( p = 3 . 17 x 10–5 , ρ = 0 . 79; Spearman’s rank correlation test; n = 20 cells; Figure 3K ) . We then showed a significant decrease in the total peak number in Nimodipine-applied neurons ( Figure 3L; [Control] 2 . 00 ± 0 . 21 total peak number , [Nimodipine] 1 . 09 ± 0 . 091; p = 0 . 0034; Wilcoxon’s rank-sum test ) . Furthermore , the total peak number was significantly decreased in Ca-α1D-knockdown neurons compared to controls ( Figure 4D; [white RNAi GL00094] 1 . 55 ± 0 . 20 total peak number , [Ca-α1D RNAi HMS00294] 1 . 00 ± 0 . 00; p = 0 . 027 ) . As is intuitively obvious in the optogenetics experiment , the total peak number under the continuous activation was significantly different from that of the intermittent one ( Figure 5C; [Continuous] 1 . 11 ± 0 . 11 , [Intermittent] 5 . 00 ± 0 . 00; p = 4 . 11 x 10-5 ) . To conclude , we consider the 'total peak number' is a reliable metric that can represent the complex burst and pause patterns and designate the peak of the time derivative of firing rate fluctuation as the peak of firing rate fluctuation for short . These results suggest that the total number of peaks of firing rate fluctuation in Class IV neurons plays a key role in increasing the likelihood of rolling locomotion upon noxious heat input . We designate this hypothesis as the 'burst number-coding' model because the burst ( i . e . , high-frequency firing ) is accompanied by the peak of firing rate fluctuation . To explore another potential impact of the 'burst and pause' firing pattern on escape behaviors , we tested whether it plays an important role in evoking the 'fast crawling' after noxious heat stimulation ( Ohyama et al . 2013; 2015 ) . To evaluate a causal link between the firing patterns and the fast crawling , we performed a set of optogenetic experiments as follows: First , we confirmed that ChIEF-expressing Class IV neurons could be activated with lower-power blue light ( 0 . 056 mW/mm2; 470 nm ) and displayed lower-frequency firings ( 10~20 Hz ) with no 'burst and pause' pattern in Ca-α1D-knockdown larvae ( data not shown ) . Then , we observed that such a firing pattern triggered the fast crawling , even when no rolling behavior was evoked ( Maximum speed of crawling stride: [before activation] 6 . 37 ± 0 . 29 mm/s , [after activation] 10 . 51 ± 0 . 66 mm/s , mean ± s . e . m . , n = 24 larvae; p = 0 . 0269 , Wilcoxon signed-rank test; see 'Materials and methods' for details ) . These findings indicate that lower frequency firing is sufficient for eliciting the fast crawling , whereas the occurrence of 'burst and pause' firing pattern is not required . Finally we addressed two questions that are relevant to the polymodality of Class IV neurons ( Xiang et al . , 2010 ) . The first one was whether the L-type VGCC is also required for the avoidance of the short-wavelength light stimuli ( the directional shift of locomotion ) , for which dTrpA1-dependent neural activity of Class IV neurons was essential . Ca-α1D-knockdown larvae avoided blue light , as did the control larvae , when they were illuminated ( Figure 4G ) . This finding indicates that not only L-type VGCC is dispensable for the photo-avoidance behavior , but also the synaptic transmission at the NMJ is not severely affected under our Ca-α1D knockdown condition . The second question was whether the wild-type Class IV evokes the 'burst and pause' firing pattern in response to the blue light as it did upon the noxious thermal stimuli . We illuminated larvae expressing TN-XXL in Class IV with short-wavelength light of four different intensities , and found the following ( Figure 6 ) : ( 1 ) The firing change and the Max firing rate increased with the light intensity , but the firing rate was 32 . 1 Hz even at the strongest intensity ( 182 . 3 mW/mm2; Figure 6A–D ) , which exceeds the level that larvae normally experience in natural environments . This firing rate was far smaller than that in response to the thermal stimuli ( 83 . 9 ± 7 . 3 Hz; Figure 1E ) . ( 2 ) 'US-like spikes' ( compare its definition in the inset of Figure 6E with that of US in Figure 3C ) were detected only under the strongest light condition ( Figure 6E ) . ( 3 ) Ca2+ transients were not observed at a light intensity ( 146 . 7 mW/mm2; data not shown ) . These observations indicate that the wild-type Class IV did not generate L-type VGCC-dependent US in response to the blue light . 10 . 7554/eLife . 12959 . 022Figure 6 . Firing responses of Class IV neurons upon illumination with blue light . Extracellular single-unit recordings of blue light ( 460–495 nm ) -stimulated Class IV neurons in fillet-mounted larvae . ( A ) The blue bar above the traces indicates the 5 s duration of blue light illumination . Power densities of illumination are indicated on the left sides of individual traces ( 9 . 9–182 . 3 mW/mm2 ) . The red asterisk indicates a redefined unconventional spike ( see US in the inset of E and the definition below ) . ( B ) Raster plots of firing with various power densities of illumination ( n = 7~13 ) . Red tick marks in the responses to 182 . 3 mW/mm2 indicate redefined US ( see also D ) . ( C ) Firing frequency changes ( average frequency of 5 s before blue light exposure subtracted from average frequency during 5 s of blue light exposure [Xiang et al . , 2010] ) are plotted against power density . ( D ) Maximum firing rates are plotted against intensity of blue light . Data are presented as mean ± s . e . m . ( E ) Occurrences of redefined US upon blue light illumination . US is redefined as follows: ( 1 ) the first interspike interval ( ISI ) of three sequential spikes is less than 15 ms , and ( 2 ) the second ISI is longer than 20 ms . Then , a pair of the first and second spikes within each sorted triplet is designated as 'redefined US' ( inset ) . DOI: http://dx . doi . org/10 . 7554/eLife . 12959 . 02210 . 7554/eLife . 12959 . 023Figure 6—figure supplement 1 . A model of the information processing: The polymodal encoding of thermo- and photo-nociception and decoding . ( Top ) The nociceptive receptors in Class IV neurons . Painless and dTrpA1 channels generate cation influx upon noxious heat stimuli in a synergistic manner ( orange box ) , whereas Gr28b and dTrpA1 mediate the formation of receptor potential upon noxious light exposure ( blue box ) . ( Bottom ) A model of sensory transduction of thermal and optical nociception in Class IV neurons . Thermal stimuli induce the relatively larger generator currents ( orange line ) ; in contrast , noxious light inputs evoke the smaller currents ( blue line ) . L-type VGCC has a relatively high threshold of activation ( green line ) . When the input exceeds the threshold , the neuron evokes the 'burst and pause' firing pattern . DOI: http://dx . doi . org/10 . 7554/eLife . 12959 . 023 Our results , together with previous studies , indicate that molecular mechanisms in Class IV underlying avoidance to two distinct stimuli only partially overlap . Furthermore we propose the 'burst number-coding' hypothesis that , in Class IV neurons , the number of burst firings is employed as the facilitating signal for the heat stimuli and they instruct target neuron ( s ) to execute one out of the two possible behavioral outputs: the rolling behavior rather than the photo-avoidance ( Figure 6—figure supplement 1; see 'Discussion' ) .
It is accepted that the phototransduction machinery exists in the dendrite of Class IV neurons ( Xiang et al . , 2010 ) . However , previous protocols for delivering noxious thermal or mechanical stimuli did not allow us to address whether the dendrite possesses transduction machineries for those stimuli . Our IR laser irradiation-measurement device is a noninvasive system , which is applicable to whole-mount animals and able to induce a rapid and local rise in the ambient temperature , mimicking the noxious heat input . With the resolution of our experimental conditions , the IR irradiation had no apparent effect on the integrity of organelles and evoked physiological responses of Class IV neurons . With the help of this system , we showed that the thermotransduction machinery is present in the dendrite . The hypothetical thermotransduction machinery includes dTrpA1 and Painless of the TRPA channel family , which are required for the stereotype avoidance behavior to noxious heat stimuli ( Neely et al . , 2011; Tracey et al . , 2003 ) . Class IV neurons in the dTrpA1 mutant displayed a much longer latency time prior to the increase in the firing rate compared to the wild type . On the other hand , the rising profile of the firing rate in the pain mutant was comparable to that of the wild type , but the rate reached only 65% of that of the wild-type neurons . Therefore the temporal pattern of the wild-type firing response was not a simple summation of those of the two TRPA mutants . If dTrpA1 and Painless are the unique cation channels for the conductance of the putative thermocurrent , the time course of the firing in painless mutants may represent that of dTrpA1 activation , and vice versa . If this inference is the case , our results can be interpreted in a way that dTrpA1 can become activated to some extent by itself and that it is prerequisite for Painless to be properly activated . A previous electrophysiological analysis of Painless showed that Painless is a noxious-heat activated , Ca2+-requiring , Ca2+-permeable channel ( Sokabe et al . , 2008 ) . Therefore , we suppose that noxious heat stimulation activates dTrpA1 and causes local Ca2+ entry in the dendrite , followed by the robust increase in Ca2+ conductance of Painless , leading to large depolarization . This results in gating of L-type VGCC and generation of the Ca2+ transient , most likely by accumulated Ca2+ spikes , throughout the dendrites of Class IV neurons . Although little is known about subcellular localization of dTrpA1 and Painless in Class IV neurons ( Hwang et al . , 2012 ) , they may distribute along the entire dendritic arbors . Among the subunits presumably composing the functional VGCC in Drosophila , Straightjacket ( Stj ) of the α2δ family is expressed in da neurons and required for the heat avoidance ( Neely et al . , 2010 ) . Thus , Stj may coexist with Ca-α1D in the dendrite . It remains to be elucidated how the burst is followed by the pause . We assumed that any of the Ca2+-activated K+ channels ( KCa channels ) and/or other channels might be engaged in generating the pause through production of an after-hyperpolarization current . In general , specialized sense organs such as nociceptors have thresholds at or near noxious levels , and increase activity with stronger noxious stimuli ( Perl , 2007 ) . Likewise , the probability of the Ca2+ transient occurrence in Class IV neurons arose abruptly at around an IR-power that was estimated to raise the temperature close to 47°C , and the neuron increased the magnitude of the responses at the suprathreshold temperature . What would be the potential advantages for Class IV to set the threshold by employing Ca2+ spikes ? One advantage would be the improvement of the signal-to-noise ratio of the thermosensation . The cellular threshold of the response to the noxious heat could become precipitous by multiplying a physical property of the high-threshold channel ( dTrpA1 ) by the probability of the Ca2+ spike generation . Another advantage would be to utilize the generation of Ca2+ spikes to winnow a heterogeneous modality in a selective manner . Our observations highlighted two features of the neuronal response to the different noxious stimuli: ( 1 ) the presence or the absence of multiple peaks of firing rate fluctuation and ( 2 ) the high or low absolute value of maximum firing rate . We hypothesize that the Ca2+ spike can be a key signal encoding a specific modality: the total number of peaks of firing rate fluctuation transmits the noxious heat input by way of Ca2+ spikes; in contrast , the continuous firing train at a lower rate encodes the noxious light sensation ( e . g . see Figure 1E and 6D ) . This means that the thermal stimuli induce the high-threshold and larger generator currents ( orange line; Figure 6—figure supplement 1 bottom ) , while blue light inputs evoke the low-threshold and smaller currents ( blue line; Figure 6—figure supplement 1 bottom ) . We showed that the heat-induced unique firing patterns of Class IV were conducted along the axon , which projects to the putative central pattern generator ( s ) driving the thermal nocifensive responses . It has been shown that Channelrhodopsin2-mediated activation of Class IV provokes the thermal nocifensive behavior ( Hwang et al . , 2007 ) and the photo-avoidance ( Xiang et al . , 2010 ) . We predict that inward currents would be larger in the former . Our optogenetic activation experiment shows that the multiple peaks of firing rate fluctuation is not essential for the nocifensive escape locomotion ( rolling ) , but enhances its occurrence rate . Therefore , we speculate that such unique firing rate fluctuation may be decoded as a facilitating signal of rolling ( the 'burst number-coding' hypothesis ) . It has also been shown in olfactory behaviors of C . elegans that specific activity patterns may modulate the probability of action selection ( Gordus et al . , 2015 ) . Recently , it has been reported that combining mechanosensory and nociceptive cues synergistically enhances the selection of rolling behavior in Drosophila larvae ( Ohyama et al . , 2015 ) . Altogether , the likelihood of rolling locomotion upon noxious heat stimulation can be potentiated by the input ( s ) from other sensory modalities and multiple peaks of firing rate fluctuation . We also noticed that the probability of rolling was rather high in the continuous activation condition ( 66 . 0%; gray bar in Figure 5B ) . We assume that almost all of the Class IV neurons were activated under our optogenetic conditions , so that such a high behavioral reactivity was elicited . On the other hand , lower probabilities of rolling were detected in the initial periods of the heat probe assay of Ca-α1D knockdown larvae ( ~20% , <2 s; Ca-α1D RNAi in Figure 4F ) . This low reactivity can be explained by local heat-stimulation that may activate a small number of neurons ( Robertson et al . , 2013 ) . The neural circuits in the central nervous system must read out the activity of Class IV neurons and distinguish one firing pattern from another to decode the polymodal sensory input . How are the two types of firing patterns in response to the distinct noxious inputs decoded and transformed into differential escape behaviors ? Previous studies have shown that the burst coding is a behaviorally relevant mechanism ( Krahe and Gabbiani , 2004 ) . For instance , the bursts in an ultrasound-sensitive acoustic sensory neuron of the field cricket Teleogryllus oceanicus code for a conspicuous increase in amplitude of an ultrasound stimulus ( Sabourin and Pollack , 2009 ) . It is well known that in the cerebellar cortex , climbing fiber inputs evoke characteristic complex spikes in Purkinje neurons , which induce the dendritic Ca2+ entry through Ca2+ spikes and a subsequent pause in the spike train ( Davie et al . , 2008; Kitamura and Häusser , 2011; Llinas and Sugimori , 1980; Mathews et al . , 2012 ) . The firing response of Purkinje neurons is remarkably similar to that of Class IV neurons in terms of the pattern and the underlying mechanism . It should be noted that the pause of firing in Purkinje neurons evokes the high-frequency firing response in the target neurons of the deep cerebellar nuclei via the post-inhibitory rebound ( PIR ) mechanism ( De Zeeuw et al . , 2011 ) . Moreover , it has been proposed that the pauses of firings per se convey signals ( the ‘pause coding’ theory , De Schutter and Steuber , 2009; Hong and Optican , 2008 ) . Another circuit evoking the nocifensive light-avoidance behavior may receive direct excitatory inputs from Class IV neurons; this would be consistent with our hypothesis that the dual codes of the polymodal sensory information are decoded and segregated as distinctive command signals in the downstream circuit . Recently , the candidate synaptic partner ( s ) of Class IV neurons were identified through functional and anatomical assays ( Vogelstein et al . , 2014; Ohyama et al . , 2015 ) ; thus , our hypothesis can be tested by investigating the neuronal responses of these candidate neurons to noxious thermal stimulations of Class IV neurons .
The following fly strains were generated and/or used: ( 1 ) The overall structure of our construct for expressing the FRET-based Ca2+ indicator TN-XXL ( Mank et al . , 2008 ) in Class IV neurons was as follows: The TN-XXL ORF was flanked by 1 . 9 kb pickpocket ( ppk ) promoter ( Grueber et al . , 2003 ) and a SV40 polyA signal sequence . Three copies of this cassette with intervening gypsy insulators ( Ni et al . , 2009 ) were subcloned into a pCasper4-derived plasmid that contains a PhiC 31 attB site ( Groth et al . , 2004 ) . The resulting construct ( 3×[ppk-TN-XXL] ) was integrated into either attP40 or attP2 . ( 2 ) Mutants of thermoTRP genes were pain1 and pain3 ( Sakai et al . , 2009; Tracey et al . , 2003 ) from T Sakai , and dTrpA11 ( Kwon et al . , 2008 ) and dTrpA1ins ( Hamada et al . , 2008 ) from P . Garrity . ( 3 ) Mutants of the L-type VGCC α subunit gene were Ca-α1DAR66 from C Duch , and Ca-α1DX10 and Ca-α1DX7 ( Eberl et al . , 1998 ) from Bloomington Stock Center . ( 4 ) A mutant of the T-type VGCC α subunit gene was Ca-α1Tdel ( Ryglewski et al . , 2012 ) from Bloomington Stock Center . ( 5 ) Mutants of the P/Q-type VGCC α subunit gene were cacophonyS ( cacS[Smith et al . , 1998] ) from RW Ordway and cacts4 ( Rieckhof et al . , 2003 ) from JT Littleton . We examined the peak amplitude of thermo-dependent Ca2+ transients in the above T-type and P/Q-type VGCC mutants , and found that there was no statistically significant difference in the amplitude ( data not shown ) . ( 6 ) Stocks related to CICR component genes were TRiPJF03381 ( dsRNA of Rya-r44F ) from Bloomington Stock Center , and Itp-r83A90B . 0 and Itp-r83A05616 ( Venkatesh and Hasan , 1997 ) from Bloomington Stock Center . ( 7 ) Other transgenic lines were UAS-dTrpA1 . K ( attP2 ) , UAS-Ca-α1D RNAi JF01848 , UAS-Ca-α1D RNAi HMS00294 , UAS-white RNAi GL00094 , UAS-Lifeact:GFP , UAS-mito:HA:GFP , UAS-Kir2 . 1 , UAS-GCaMP5G , GMR-Hid and UAS-Dcr-2 from Bloomington Stock Center , and UAS-ChIEF-tdTomatoC3-3 ( Wang et al . , 2011 ) from Z Wang . Exact genotypes of individual animals used in figures are described below: Figure 1 ( D ) 3×[ppk-TN-XXL] ( attP2 ) / 3×[ppk-TN-XXL] ( attP2 ) ( E-G ) 3×[ppk-TN-XXL] ( attP2 ) / 3×[ppk-TN-XXL] ( attP2 ) ( WT ) 3×[ppk-TN-XXL] ( attP40 ) /+; dTrpA11/ dTrpA11 pain3/pain3; 3×[ppk-TN-XXL] ( attP2 ) /+ Figure 2 ( A–C , F ) 3×[ppk-TN-XXL] ( attP40 ) /+ ( D ) 3×[ppk-TN-XXL] ( attP40 ) /+ ( WT ) 3×[ppk-TN-XXL] ( attP40 ) /+; ppk-Gal4/UAS-Kir2 . 1 Ca-α1DX10/AR66; 3×[ppk-TN-XXL] ( attP2 ) /+ Ca-α1DX7/AR66; 3×[ppk-TN-XXL] ( attP2 ) /+ ( E ) Ca-α1DX10/AR66; 3×[ppk-TN-XXL] ( attP2 ) /+ Figure 3 ( A , B , D–J ) 3×[ppk-TN-XXL] ( attP2 ) / 3×[ppk-TN-XXL] ( attP2 ) Figure 4 ( A–D , F ) w1118/+; ppk-Gal4/+; 3×[ppk-TN-XXL] ( attP2 ) /+ ( Control ) ppk-Gal4/+; 3×[ppk-TN-XXL] ( attP2 ) /UAS-white RNAi GL00094 ppk-Gal4/+; 3×[ppk-TN-XXL] ( attP2 ) /UAS-Ca-α1D RNAi JF01848 ppk-Gal4/+; 3×[ppk-TN-XXL] ( attP2 ) /UAS-Ca-α1D RNAi HMS00294 ( E ) w1118/w1118 or Y ( WT ) Ca-α1DX10/AR66 ( G ) ppk-Gal4/GMR-Hid; 3×[ppk-TN-XXL] ( attP2 ) /+ ( Control ) ppk-Gal4/GMR-Hid; 3×[ppk-TN-XXL] ( attP2 ) /UAS-Ca-α1D RNAi JF01848 ppk-Gal4/GMR-Hid; 3×[ppk-TN-XXL] ( attP2 ) /UAS-Ca-α1D RNAi HMS00294 Figure 5 ( A–C ) UAS-ChIEF-tdTomatoC3-3/UAS-Dcr-2; ppk-Gal4/ UAS-Ca-α1D RNAi HMS00294 Figure 6 ( A–E ) 3×[ppk-TN-XXL] ( attP2 ) / 3×[ppk-TN-XXL] ( attP2 ) To measure the temperature of microenvironment , we employed electrical resistance thermometry using a glass microelectrode ( Palmer and Williams , 1974; Shapiro et al . , 2012; Yao et al . , 2009 ) . To elucidate the relationship between the electrical resistance of an electrode and the temperature of the external saline solution ( Xiang et al . , 2010 ) , we monitored the electrical resistances at various values of temperature of the solution . First , we heated the saline solution in a petridish up to 50°C with an inline heater ( SF-28 , Warner Instruments , Hamden , CT ) that was regulated by a thermo-controller ( TC-324B , Warner Instruments ) . Then , we turned off the heater and recorded the electrical resistances ( R ) of the electrode and the temperatures ( T ) of the solution simultaneously during natural cooling . The electrical resistances of glass microelectrodes were 5–10 MΩ at ambient temperature ( ~25°C ) . We measured the electrical resistances by giving square pulses ( 50 ms , 10 mV; 1 Hz ) in the voltage clamp mode . The reciprocal of temperature ( 1/T ) was plotted against the log of the electrical resistance ( log R ) , so that the Arrhenius equations were estimated as a R-T transformation formula by linear regression . Fillets were made from wandering third instar larvae with the cuticle facing down in the external saline solution ( Xiang et al . , 2010 ) ( 120 mM NaCl , 3 mM KCl , 4 mM MgCl2 , 10 mM NaHCO3 , 10 mM trehalose , 10 mM glucose , 5 mM TES , 10 mM sucrose , 10 mM HEPES . pH adjusted to 7 . 25 with NaOH ) . To immobilize larvae , all of the segmental nerves were cut , and then central nervous systems were removed . Muscles covering the neurons of interest were gently digested by infusion of Protease Type XIV ( 0 . 5% w/v , Sigma-Aldrich , St . Louis , MO ) through a glass micropipette ( ~1 . 2 MΩ ) mounted on a PatchStar micromanipulator ( Scientifica , East Sussex , UK ) . After the enzymatic treatment , a couple of perfusions of external saline solution were carried out to remove excess enzyme , hemocytes , and other cellular debris . No detectable difference in dendrite morphology of Class IV dendritic arborization neurons was observed before and after muscle digestion , indicating the neurons were intact . Class IV neurons were identified by the fluorescence of TN-XXL driven by the ppk promoter ( see details in Ca2+ imaging ) . Recording pipettes were pulled with a P-1000 puller ( Sutter Instruments , Novato , CA ) from thin wall borosilicate glass ( B150-110-10 , Sutter Instruments ) , filled with external saline solution , with a tip opening of 5 μm ( 800 kΩ–1MΩ ) . Gentle negative pressure was delivered to suction the soma to get good signal-to-noise ratios of recording traces . We mostly recorded activity of v’ada of Class IV neurons . Recordings were performed with a Multiclamp 700B amplifier ( Molecular Devices , Sunnyvale , CA ) , and data were acquired with Digidata 1440A ( Molecular Devices ) and Clampex 10 . 0 software ( Molecular Devices ) . Extracellular single-unit recordings of action potentials were obtained in current clamp mode , with an 8 kHz low-pass filter and sampled at 10 kHz or 20 kHz . Data were analyzed by Clampfit 10 . 3 . 1 . 5 software ( Molecular Devices ) and custom programs written in MATLAB ( MathWorks , Natick , MA ) . Extracellular spikes were detected by first band-pass filtering ( the high-pass filter: RC monopole , fc = 100 Hz; the low-pass filter: Butterworth 8-pole , fc = 2 kHz ) and then thresholding the extracellular voltage trace with the 'Threshold Search' plugin of Clampfit software . The Maximum firing rates were computed by sliding a rectangular window function along the spike train with Δt = 400 ms . The peaks of the time derivative of firing rate fluctuation were assigned as follows: first , the spike density estimation was computed by using the Gaussian kernel method ( σ = 25 ms; Shimazaki and Shinomoto , 2010 ) ; second , the temporal differences of firing rate were calculated ( time derivative of firing rate fluctuation; time interval = 0 . 1 ms ) ; third , a threshold was configured as 0 . 5 of the primal peak value; finally , the peaks of the time derivative above the threshold were specified . 'Time to Max . Firing Rate' was defined as the left edge of the sliding rectangular time window that gave the earliest maximum firing rate in each spike train . To perform simultaneous recording from the soma and the axon bundle , we first obtained the extracellular recording from the soma , and then the cutting edge of the axon bundle in the same hemisegment was suctioned into the recording electrode with a resistance of 700–800 kΩ . A 200-mW continuous wave oscillation IR-laser light source was connected to the microscope through the IR-LEGO-200 system ( Kamei et al . , 2008 ) ( Sigma Koki , Tokyo , Japan ) to provide IR-laser stimulation through a dry objective ( 40× UApo/340 NA 0 . 90 , Olympus , Tokyo , Japan ) , and the IR-laser beam was focused onto the center of the field of view . The source of IR laser was a continuous wave semiconductor laser ( SLD-1462-200-C; FiberLabs , Fujimino , Japan ) . Precise focus of the IR-laser was localized by the spotted heat evaporation of indelible marker ink ( K-177N , Shachihata , Nagoya , Japan ) . All of the power values in the text and figures are settings of the laser driver . IR-laser power density was measured at the focal plane of the objective using a radiometric sensor head ( UP17P-6S-H5-DO , Gentec-EO , Quebec , Canada ) coupled with a power meter ( UNO , Gentec-EO ) . The measured output power was ~48 . 2% of the preset power . Custom programs in Digidata 1440A controlled the duration and timing of IR-laser irradiation by opening a TTL-triggered shutter ( Shutter Controller F77-6 , Suruga Seiki , Shizuoka , Japan ) in the IR-LEGO system . A 100-W mercury light source ( U-LH100HGAPO , Olympus ) was connected to the microscope with a light guide to provide short-wave light stimulation through a filter ( BP460-495 , Olympus ) and the objective , yielding an evenly illuminated light spot , which covered the entire Class IV dendritic arborization neuron . Light intensity was measured as in IR-laser irradiation . Custom programs in Digidata 1440A controlled the duration and timing of light illumination by opening a TTL-triggered shutter ( SSH-R , Sigma Koki ) in the light guide . To administrate 3 μM Tetrodotoxin ( Nacalai , Kyoto , Japan ) or 5 μM Nimodipine ( Sigma-Aldrich ) , fillets were incubated in the chamber for 30 min before recording . 10 μM Thapsigargin ( Nacalai ) was employed to deplete the cytoplasmic Ca2+ store . Ca2+ imaging of TN-XXL-expressing Class IV neurons was performed on fillet or whole-mount preparations . Fillet preparations were made essentially as in extracellular recording , except that the protease treatment was omitted when only Ca2+ dynamics were recorded . For whole-mount imaging , larvae were anesthetized with isoflurane and mounted in halocarbon oil on a microscope slide . Data were collected on an IX-71 ( Olympus ) equipped with an objective ( 40× UApo/340 NA 0 . 90 , Olympus ) , Nipkow disk confocal system ( CSU-X1 , Yokogawa Electric , Tokyo , Japan ) and two EMCCD imagers ( iXon X3 DU897-BV , Andor Technology , Belfast , UK ) . The custom-made camera mount ( Olympus ) of each EMCCD imager was equipped to adjust either the rotation angle around the optical axis ( for CFP imaging ) or the displacements along X- and Y-axes ( for YFP imaging ) . Using the adjustment devices , the two imagers’ fields of view were aligned with minimal distortion in the rotation angle and minimal displacements less than one pixel along the X- and Y-axes . TN-XXL was excited with a 445-nm diode laser ( CUBE 445-40C , Coherent , Santa Clara , CA ) . Images were acquired at 512×512 pixels or 256×256 pixels with 2x2 binning , in 14-bit dynamic range , and with 100 or 33 ms exposure time . CFP and YFP fluorescence signals were captured synchronously and separately by the two imagers with 10 or 30 Hz , through 490/40 and 578/105 band-pass filters ( Semrock , Lake Forest , IL ) , respectively . In simultaneous acquisitions of extracellular single-unit recordings and Ca2+ imaging , two imagers were controlled by Solis 4 . 19 software ( Andor Technology ) and the exposure timings were recorded by Digidata 1440A; these timing data were used for the offline processing as synchronization cues . In the case of the Ca2+ image acquisition only , two imagers were controlled by iQ 1 . 10 . 3 software ( Andor Technology ) . Data were analyzed by ImageJ ( NIH ) and custom programs written in MATLAB ( MathWorks ) . The setting of regions of interest ( ROI ) was guided with the binary mask image generated by the dendritic morphology . The background signals were subtracted from CFP or YFP images; then the FRET ratios were calculated ( see Figure 2—figure supplement 1 ) . In response to IR stimulations , Ca2+ fluctuations with ΔRpeak larger than 10% were designated as Ca2+ transients . The thermal nociception behavioral tests were performed essentially as described ( Caldwell and Tracey , 2010; Hwang et al . , 2007 ) , with slight modifications . Animals were raised at 25°C in an incubator with 12 hr light/dark cycles and humidity was manually controlled ( 75–80% ) . Third instar larvae were gently picked up from the vial , washed twice with deionized water , and transferred to a 140×100-mm petridish with fresh 2% agarose . Excessive water was removed from the animals . For acclimation , animals were allowed to rest on the plate for at least 5 min before testing . The response latency was measured as the time interval from the point at which the larva was first contacted by the probe until it completed the first 360° roll . The light nociception behavioral tests were performed essentially as described ( Xiang et al . , 2010 ) , with slight modifications . Animals were raised and treated as in the thermal nociception test . The assay was carried out with a Macro Zoom Microscope system ( MVX-10 , Olympus ) . Light was delivered from a 100-W mercury lamp ( U-LH100HGAPO , Olympus ) through a MV PLAPO 1X objective ( Olympus ) at 2 . 5X magnification , yielding a light spot of 1 . 5 mm in diameter . Other devices we used were a shutter ( SSH-R1X , Sigma Koki ) in the light guide , a filter for the background light ( 30 . 5S-R64 , Olympus ) , and a lens for a CCD camera ( GE60 , Library , Tokyo , Japan ) . A filter ( GFP-3035D , Olympus ) was placed into the above microscope to illuminate blue light ( 472 . 5 nm with 30 nm band width ) . The light intensity at 2 . 5X magnification was measured by a radiometric sensor head ( UP17P-6S-H5-DO , Gentec-EO , Quebec , Canada ) coupled with a power meter ( UNO , Gentec-EO ) and it was determined to be 0 . 72 mW/mm2 in the light spot . Sixty-five to 104 animals were tested in each condition and the percentage of positive responses was calculated . Larvae for optogenetic activation experiments , harboring the UAS-ChIEF-tdTomato transgene ( Wang et al . , 2011 ) , were grown in the dark at 25°C for 4 days on fly food containing all trans-retinal ( R2500; Sigma-Aldrich ) at a final concentration of 0 . 5 mM . For behavioral experiments , one larva at a time was separated from food and placed into the center of a 9 . 5 x 13 . 5 cm square plastic dish filled with 2% agar ( Bacto Agar; Becton Dickinson , Franclin Lakes , NJ ) . After 5 min of acclimation , the dish with larva was placed into the behavior rig . 5 cycles of 100 ms pulses followed by 100 ms pause intervals or a single 1 s long pulse of blue light ( 0 . 234 mW/mm2; Figure 5B ) , or a single 1 s long pulse ( 0 . 056 mW/mm2 for the crawling assay below ) were applied by using a collimated LED light source ( M470L3-C1; ThorLabs , Newton , NJ ) with an emission peak at around 470 nm . A data acquisition system ( USB-6212 BNC; National Instruments , Austin , TX ) provided the triggering TTL signals , with timing controlled by a custom LabVIEW program ( National Instruments ) . We recorded videos of larval behavior by using a GE60 CCD image sensor ( 640 x 480 pixels; Library ) at 30 frames per second . For the analyses of crawling stride speeds , larval locomotions were traced in videos by using 'Manual Tracking' Fiji plugin ( ImageJ 1 . 46J , NIH , Bethesda , MD ) , and then the maximum stride speeds for 5 s before and after optogenetic activation were calculated . For electrophysiological recordings , 5 cycles of 100 ms pulses with 100 ms pause intervals or a single 1 s long pulse of blue light ( 0 . 294 mW/mm2 for Figure 5A ) , or a single 1 s long pulse ( 0 . 056 mW/mm2 ) was applied from above through external saline solution . The Digidata 1440A digitizer ( Molecular Devices ) generated the triggering TTL signals , with timing controlled by a custom Clampex protocol ( Molecular Devices ) . RNA was purified from four larval CNS complexes for each genotype using an RNeasy kit ( QIAGEN , Hilden , Germany ) . ReverTra Ace and Thunderbird ( TOYOBO , Osaka , Japan ) were used according to the manufacturers’ instructions . Sequences of the primers were: 5’-GCT TGT GCC CAG GGA GC-3’ and 5’-AGG ACA CAA TGT CCG GAT GAT CG-3’ ( Probe 1 ) , and 5’-GCG ACC AGA ATG GCG ACT TTA ATG-3’ and 5’-CCC GTA TCC ACT GGG ATG GAC-3’ ( Probe 2 ) for dTrpA1; 5’-GCG ACA CCC AAG TTA TTA AGG GTC-3’ and 5’-GTT CAT CAA ACG TTG GCA GAT GC-3’ ( Probe 1 ) , and 5’-TGC TGA CAG GCG AGT TTG AC-3’ and 5’-GCC TGA GCC TTA ATA ACT TGG GTG-3’ ( Probe 2 ) for pain; and 5’-GCT AAG CTG TCG CAC AAA TG-3’ and 5’-GTT CGA TCC GTA ACC GAT GT-3’ for rp49 used as a reference . Data were analyzed using the comparative CT method on a StepOne Real-Time PCR System ( Applied Biosystems , Foster City , CA ) . | Animals often need to get away quickly from dangers in their environment , such as temperatures that are hot enough to damage their tissues . As such , an animal’s brain often encodes automatic ‘avoidance responses’ to signs of danger , which help the animal move away from harm . The nervous system of a fruit fly larva , for example , contains a distinct class of neurons ( known as class IV neurons ) that respond specifically to high temperatures and ultraviolet or blue light . Both of these stimuli are potentially harmful , but the larvae escape from heat by rolling with a corkscrew-like motion , yet they turn their heads away from a source of ultraviolet or blue light . So , how does the same set of neurons orchestrate these two different types of behavior ? To answer this question , Terada , Matsubara , Onodera et al . measured the activity in the class IV neurons in two different ways . First , the levels of calcium ions in the neurons , which play a key role in neurons’ activity , were imaged using a calcium-sensitive biosensor . Second , electrodes were used to directly on the class IV neurons to record changes in their electrical activity . The experiments showed that class IV neurons responded to heat by producing a characteristic burst of electrical activity followed by a pause , and that this pattern of electrical activity was accompanied by a large rise in the calcium signal . In contrast , the same neurons did not show this ‘burst and pause’ pattern of activity when the fruit fly larvae were exposed to ultraviolet/blue light . Instead , these conditions triggered much smaller changes in electrical activity . Further experiments then confirmed that the characteristic ‘burst and pause’ pattern of electrical activity was linked to the rolling motion observed when the larvae try to escape from heat . These findings show how differing patterns of activity in the same neurons can be used to differentiate between different types of stimuli . Further work is now needed to explain how these two different patterns of activity in one set of neurons is translated by the fruit fly’s brain into different patterns of behavior . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"neuroscience"
] | 2016 | Neuronal processing of noxious thermal stimuli mediated by dendritic Ca2+ influx in Drosophila somatosensory neurons |
The first point of our body’s contact with tactile stimuli ( innocuous and noxious ) is the epidermis , the outermost layer of skin that is largely composed of keratinocytes . Here , we sought to define the role that keratinocytes play in touch sensation in vivo and ex vivo . We show that optogenetic inhibition of keratinocytes decreases behavioral and cellular mechanosensitivity . These processes are inherently mediated by ATP signaling , as demonstrated by complementary cutaneous ATP release and degradation experiments . Specific deletion of P2X4 receptors in sensory neurons markedly decreases behavioral and primary afferent mechanical sensitivity , thus positioning keratinocyte-released ATP to sensory neuron P2X4 signaling as a critical component of baseline mammalian tactile sensation . These experiments lay a vital foundation for subsequent studies into the dysfunctional signaling that occurs in cutaneous pain and itch disorders , and ultimately , the development of novel topical therapeutics for these conditions .
Peripheral sensory neurons detect external stimuli and transmit this information to spinal cord and brainstem circuits . Despite their location below the epidermal surface , convention proposes that cutaneous sensory nerve terminals are the exclusive transducers of mechanical stimuli . This concept has recently been challenged by data that demonstrate epidermal Merkel cells’ responsiveness to mechanical stimuli and subsequent signaling to sensory neurons , two processes that are essential for two-point touch discrimination ( Maksimovic et al . , 2014; Woo et al . , 2014 ) . Notably , Merkel cells constitute only a small portion ( 3–6% ) of total skin cells ( Moll et al . , 1986; Fradette et al . , 2003; Halata et al . , 2003 ) , whereas keratinocytes , which have traditionally been known for their roles in barrier formation and protection rather than sensory transduction , comprise 94–97% of the epidermis ( Fuchs , 1995 ) . However , keratinocytes are closely apposed to sensory nerve terminals ( Löken et al . , 2009 ) and are constantly exposed to external mechanical forces in the environment like brushing and pressure from stimuli like clothing , objects and other living organisms . Previously , it was demonstrated that isolated keratinocytes directly respond to mechanical probing by increasing intracellular calcium concentrations ( Koizumi et al . , 2004; Tsutsumi et al . , 2009; Goto et al . , 2010 ) . Furthermore , keratinocytes can release many neuroactive substances including ATP , calcitonin gene-related peptide β ( CGRPβ ) , acetylcholine , glutamate , epinephrine , neurotrophic growth factors , and cytokines among others ( Barr et al . , 2013; Hou et al . , 2011; Lumpkin and Caterina , 2007; Shi et al . , 2013 ) . In co-cultures of keratinocytes and dorsal root ganglia ( DRG ) neurons , mechanical stimulation of keratinocytes evokes inward currents in adjacent sensory neurons , presumably through release of one of the aforementioned metabolites ( Klusch et al . , 2013 ) . Taken together , these data suggest that sensory neurons may not be the sole transducers of mechanical stimuli , but rather may collaborate with other cell types such as keratinocytes to initiate or amplify somatosensory signals to sensory neurons . Here , we sought to define the role of keratinocytes in mechanotransduction by utilizing cell-specific optogenetic approaches during evoked and non-evoked behavioral testing . We found that these epidermal cells are critical for innocuous and noxious touch detection . Using ex vivo sensory fiber recording techniques and pharmacological approaches , we identified ATP as a key signaling molecule released by keratinocytes in response to mechanical stimulation . Finally , we used novel sensory neuron-specific knockout mice to demonstrate that mechanically induced ATP release is functionally coupled to the activation of P2X4 receptors on sensory neurons . These data are the first to identify purinergic signaling as a critical component of innocuous and noxious skin mechanotransduction , specifically in the context of non-neuronal to neuronal cellular communication .
We first sought to determine whether keratinocytes have a functional role in sensing touch . Keratinocytes were isolated from the glabrous hindpaw skin of transgenic mice that express tdTomato in Keratin14 ( K14 ) -positive ( epidermal ) cells , the vast majority ( ~94–97% ) of which are keratinocytes ( Byrne et al . , 1994; Dassule et al . , 2000; Wang et al . , 1997 ) , with a small percentage ( 3–6% ) being Merkel cells ( Moll et al . , 1986; Fradette et al . , 2003; Halata et al . , 2003 ) . Individual keratinocytes were visually identified and then subjected to focal mechanical stimulation or ‘poke’ ( Wu et al . , 2017 ) under current clamp conditions . Increasing indentation revealed a stimulus-dependent depolarization that returned to baseline between each stimulation ( Figure 1A ) . We hypothesized that this depolarization may induce release of keratinocyte-derived factors that subsequently signal to sensory neurons , and therefore we aimed to utilize optogenetic approaches to manipulate this process . A previous study that used optogenetic methods demonstrated that keratinocytes can modulate the responses of cutaneous sensory neurons in ex vivo skin nerve recordings ( Baumbauer et al . , 2015 ) . However , this investigation stopped short of investigating the contributions of keratinocytes to tactile behavioral responses in vivo . Therefore , we created a mouse line that selectively expresses GFP-tagged Archaerhodopsin-3 ( Arch ) in K14-expressing epidermal cells ( K14Cre+ Arch/Arch ( Arch-K14Cre+ ) and K14Cre- Arch/Arch ( Arch-K14Cre- ) littermate controls ) and tested whether keratinocytes have a functional role in sensing innocuous or noxious touch in vivo . When Arch is activated by amber light ( peak photocurrent between 550 and 600 nm ) , it pumps protons out of the membrane , thereby hyperpolarizing the cell ( Chow et al . , 2010 ) . Here , we activated Arch via transdermal light stimulation to inhibit epidermal cells in vivo . To confirm that Arch expression was restricted primarily to epidermal cells , we evaluated GFP expression patterns in glabrous hindpaw skin sections . As expected , GFP ( Figure 1C , F ) overlapped substantially with K14-positive epidermal cells ( Figure 1B , E ) in Arch-K14Cre+ skin ( Figure 1G ) , but not in Arch-K14Cre- skin ( Figure 1D ) . Because keratinocytes migrate from the basal to superficial epidermal layers in a temporal fashion , GFP expression was found throughout all layers , and was not restricted only to the basal keratinocyte layer where K14 expression is found . We next assessed whether the Arch expressed in keratinocytes was functional . Whole cell current clamp recordings were performed on keratinocytes isolated from glabrous hindpaw skin of Arch-K14Cre+ and Arch-K14Cre- mice in order to measure amber light ( 590 nm ) -evoked changes in membrane potential . During light stimulation , Arch-K14Cre+ keratinocytes exhibited hyperpolarized membrane potentials , as compared to no light stimulation ( Figure 1H ) . Light stimulation had no effect on the membrane potential of keratinocytes from Arch-K14Cre- animals ( Figure 1H ) . To determine whether optogenetic inhibition affects the mechanical responsiveness of keratinocytes , we recorded membrane voltage in Arch-K14Cre+ cells during focal stimulation and light exposure . Analogous to Figure 1A , the membrane voltage of Arch-K14Cre+ keratinocytes depolarized in a graded manner upon mechanical stimulation , and 590 nm light significantly reduced the overall level of depolarization at each force ( Figure 1I ) . Hyperpolarization of keratinocytes significantly lowered the membrane potential even with mechanical stimulation , thus indicating that hyperpolarization can inhibit evoked signaling processes in keratinocytes . To determine whether inhibition of K14-expressing cells affects animals’ behavioral sensitivity to tactile stimuli , the glabrous hindpaw skin was briefly exposed to 590 nm light before ( 1 min ) and during mechanical stimulation . Thresholds for tactile detection were measured using von Frey filaments ( Dixon , 1980; Chaplan et al . , 1994 ) . Keratinocyte inhibition significantly elevated mechanical paw withdrawal thresholds in Arch-K14Cre+ animals compared to Arch-K14Cre- controls ( Figure 1J ) , reflecting decreased tactile sensitivity . Similar exposure to 490 nm light , a wavelength incapable of activating Arch , had no effect on mechanical thresholds ( Figure 1J ) . Responses to repeated suprathreshold tactile stimuli were tested by applying a 3 . 61 mN filament 10 times to the plantar hindpaw and quantifying frequency ( % ) of withdrawal responses . The 590 nm light caused Arch-K14Cre+ animals to be less responsive to repeated probing than their Arch-K14Cre- littermates ( Figure 1K ) . Control light ( 490 nm ) had no effect on the mechanical responsiveness of the Arch-K14Cre+ or Arch-K14Cre- animals ( Figure 1K ) . Although light pretreatments were given for 1 min before application of evoked stimuli for ease of stimuli administration , light pretreatment was not necessary to induce the behavioral mechanical inhibition , as 590 nm light treatment delivered simultaneously with the mechanical stimuli elicited the full effect of inhibition that was observed with pretreatment ( compare Figure 1—figure supplement 1A and B to Figure 1J and K ) . We next asked whether inhibition of K14-expressing cells also affects responses to noxious mechanical stimuli . The tip of a spinal needle was used to poke the hindpaw 10 times; responses were characterized as normal ( innocuous , simple paw withdrawal ) , noxious ( licking , flicking and elevating the paw for extended periods of time ) , or null ( no response ) ( Hogan et al . , 2004; Moehring et al . , 2016 ) . Arch-K14Cre+ animals exposed to 590 nm light exhibited significantly fewer noxious and innocuous responses , and a concomitant increase in null responses to the needle poke compared to Arch-K14Cre-controls ( Figure 1L ) . Exposure to the control 490 nm light had no effect on the responses subtype distribution in either genotype ( Figure 1L ) . Importantly , the behavioral changes observed in response to the 590 nm light were not a result of temperature changes in the hindpaw skin as neither 590 nm nor 490 nm light altered the temperature in the hindpaw skin ( Figure 1—figure supplement 1C and D ) . Next , to determine whether keratinocyte inhibition affects ongoing behaviors , the Arch-K14 cohorts were tested in a non-evoked place preference assay . Animals were allowed to freely explore a two-chamber box , where the floor of one chamber floor was illuminated with 595 nm light and the other was illuminated with 460 nm light . Neither Arch-K14Cre+ nor Arch-K14Cre- animals preferred either chamber when the lights were on ( Figure 1—figure supplement 1E ) , suggesting that the inhibition of keratinocytes and other epidermal cells alone does not evoke aversive or pleasant sensations in the animals . Together , these results demonstrate that epidermal K14-expressing cells play a key role in detecting evoked innocuous and noxious mechanical stimuli . Since light-induced inhibition of epidermal cells reduced the animals’ baseline sensitivity to evoked mechanical stimuli , we performed complementary experiments to determine the effects of light-induced activation of K14-expressing epidermal cells using light-sensitive Channelrhodopsin 2 ( ChR ) , which depolarizes cells when activated by 450 – 500 nm light ( Nagel et al . , 2003 ) . We generated a mouse line that expresses eYFP-tagged ChR in K14-expressing cells ( K14Cre+ ChR/ChR ( ChR-K14Cre+ ) and K14Cre- ChR/ChR ( ChR-K14Cre- ) littermate controls ) . eYFP expression was absent in ChR-K14Cre- skin ( Figure 2A–C ) . Similar to the pattern of expression of Arch in Arch-K14Cre+ sections , ChR expression was present ( Figure 2E ) throughout the keratinocyte layers , extensively overlapping with K14 immunoreactivity ( Figure 2D ) in ChR-K14Cre+ animals ( Figure 2F ) . To determine whether the ChR expressed in keratinocytes was functional , keratinocytes were isolated from the glabrous hindpaw skin of adult ChR-K14Cre+ and ChR-K14Cre- mice and voltage clamped . All keratinocytes from ChR-K14Cre+ animals responded with a sustained inward current during a 30 second 490 nm light stimulation ( Figure 2G–I ) , whereas none of the ChR-K14Cre- keratinocytes responded ( Figure 2I ) ; only a small leak current was present in some cells . These data indicate that ChR is expressed and functional in adult mouse keratinocytes . To determine if light-evoked activation of epidermal cells elicits behavioral responses , one hindpaw was focally exposed to 473 nm light . The time to initial response was measured and the type of response was categorized . On average , ChR-K14Cre+ animals responded to hindpaw light stimulation within 46 seconds , whereas most ChR-K14Cre- controls did not respond to the 473 nm laser stimulation during the 6 min test ( Figure 2J ) . The majority of mice from either genotype did not respond to the 589 nm control laser ( Figure 2—figure supplement 1A ) . Additionally , ChR-K14Cre+ animals spent significantly more time attending to their 473 nm light-stimulated hindpaw ( Figure 2—video 1 ) than ChR-K14Cre- littermates ( Figure 2—video 2 ) ; most of the ChR-K14Cre- mice were unresponsive to the stimulus for the 6 min duration of the test ( Figure 2—figure supplement 1B ) . Once again , these results could not be accounted for by increases in skin temperature , as no difference was noted in skin temperature during light stimulation between genotypes and because either the 473 nm or 589 nm laser elicited the same small temperature increase in both genotypes ( Figure 2—figure supplement 1C and D ) . Together , these data indicate that light-induced depolarization of K14-expressing cells elicits attending responses in vivo . To determine if the light induced responses in K14-ChR animals displayed in Figure 2J and Figure 2—figure supplement 1B were aversive , we utilized a non-evoked real-time place preference assay ( Figure 2K ) . ChR-K14Cre- and ChR-K14Cre+ animals spent the same amount of time on the 595 nm and 460 nm-paired sides during light on and light off conditions ( Figure 2K ) . Upon closer inspection of the animal behaviors on the two sides , it became apparent that the 460 nm light evoked significantly more ‘grooming’ and attending behaviors in the ChR-K14Cre+ than in their littermate controls ( Figure 2L ) . The aversive/grooming behaviors noted were repeated face ‘wiping’ with the forepaws , biting both fore- and hindpaws , and shaking of the tail . Taken together , these data make clear that keratinocyte signaling is necessary for naive behavioral responses to both innocuous and noxious mechanical stimuli , and that activation of keratinocytes and other epidermal cells alone is sufficient to elicit behavioral responses . Because we and others have shown the importance of keratinocyte signaling to sensory neurons ( Baumbauer et al . , 2015; Pang et al . , 2015 ) , we next sought to determine the signaling molecule ( s ) that mediate keratinocyte to sensory neuron communication . We investigated ATP because we recently showed that mice with a deficit in tactile sensitivity have decreased mechanically evoked ATP release from skin ( Zappia et al . , 2016 ) . To assess levels of mechanically evoked ATP release from skin , ATP-sensing enzymatic probes were inserted into isolated glabrous hindpaw skin from naive wild-type mice ( Figure 3A ) . The skin was then probed with forces ranging from 1 . 6 to 84 . 8 mN . Transient ATP release was detected during all of the epidermal stimulations ( Figure 3B ) . To determine whether ATP release was graded according to stimulus intensity , increasing mechanical forces were applied to the glabrous skin with von Frey filaments . Increasing forces resulted in graded , increased ATP release ( Figure 3C ) . Further , repeated stimulation with a single force ( 20 . 1 mN ) elicited reproducible ATP release ( Figure 3D ) . These data indicate that ATP is released in a reproducible and graded manner by mechanical stimulation of isolated glabrous hindpaw skin . Next , we asked whether keratinocytes are the key source of ATP release in the skin . To do this , we performed a ‘cell sniff’ assay ( Lalo et al . , 2014; 2007 ) . HEK-293 cells were transfected with P2X2 receptors tagged with a C-terminal GFP . P2X2 transfected HEK-293 cells were co-cultured with K14Cre+/tdTomato+ keratinocytes obtained from adult mouse glabrous hindpaw skin . The P2X2 transfected HEK-293 cell-line ( GFP+ ) was used to detect ( ‘sniff’ ) ATP release during mechanical stimulation of a keratinocyte ( tdTomato+ ) . P2X2-mediated currents in HEK-293 cells were monitored via whole cell voltage clamp while an adjacent keratinocyte was mechanically stimulated with a glass probe ( Figure 3E and F ) . P2X2-GFP+ HEK-293 cells , but not GFP- cells , showed robust inward currents when a nearby keratinocyte was mechanically stimulated as shown in current trace examples ( Figure 3G ) and current density ( Figure 3H ) . Further , the magnitude of evoked current in GFP+ HEK cells increased with increasing indentation of the nearby keratinocyte ( Figure 3H ) . These results demonstrate that keratinocytes from naive adult mouse skin are capable of rapidly releasing ATP in response to mechanical probing and that the amount of ATP released is indentation-dependent . We next asked whether optogenetic manipulations of isolated keratinocytes could alter mechanically evoked ATP release . We used the same cell sniff assay as in Figure 3H , except that the P2X2-GFP+ HEK-293 cells were co-cultured with Arch-K14-Cre+ keratinocytes . Keratinocytes were mechanically stimulated in the presence or absence of 590 nm light , and inward currents in the P2X2-GFP+ HEK-293 cells were recorded . Exposure to the 590 nm light blunted the amplitude of the mechanically evoked currents in the sniffer cells compared to the light off current amplitudes ( Figure 3I ) . In addition to optogenetic inhibition , we also tested the converse experiment by co-culturing ChR-K14Cre+ keratinocytes with P2X2-GFP+ HEK-293 cells . To determine if optogenetic-induced depolarization was sufficient to cause inward currents in the P2X2-GFP+ HEK-293 cells , 490 nm light with one of three different intensities ( 0 . 2 , 2 or 20 mW ) , or a 590 nm ( 5 mW ) control light , was shone on the ChR-K14Cre+ keratinocytes while inward currents in the P2X2-GFP+ HEK-293 cells were measured . Light-induced 490 nm depolarization was sufficient to elicit inward currents in an intensity-dependent manner , whereas no depolarization occurred with 590 nm light ( Figure 3J ) . Collectively , these data indicate that ATP release from keratinocytes is largely voltage-dependent and that its release is amenable to optogenetic manipulations We next asked whether ATP release in skin is required for normal behavioral responses to tactile stimuli . Apyrase , an enzyme that catalyzes ATP hydrolysis ( Palygin et al . , 2015 , Palygin et al . , 2017 ) , was injected subcutaneously into the hindpaw and sensitivity to tactile stimuli was tested . Apyrase injection significantly elevated paw withdrawal thresholds compared with vehicle-injected animals ( Figure 4A ) . Apyrase-treated animals also responded significantly fewer times to repeated suprathreshold stimulation than vehicle-treated animals ( Figure 4B ) . Further , Apyrase-injected animals showed a significant decrease in both the number of noxious and innocuous responses to needle stimulation and a concomitant increase in the number of null responses ( Figure 4C ) . Therefore , acute hydrolysis of ATP in hindpaw skin reduces the responsiveness to both noxious and innocuous mechanical stimuli under baseline conditions . Importantly , investigation of apyrase at the cellular level through utilizing the cell sniff assay demonstrated that the apyrase used in this study does indeed degrade ATP , as the inward currents in P2X2-GFP+ HEK-293 cells were attenuated when apyrase was added to the extracellular buffer while nearby keratinocytes were mechanically stimulated evoked ( Figure 4D ) . Since in vivo hydrolysis of peripheral ATP reduced behavioral tactile and noxious mechanical sensitivity , we next asked whether mechanically evoked action potential firing of peripheral sensory neurons depends on cutaneous ATP release . To explore this question , we used the tibial nerve ex vivo preparation , which mirrors the anatomical location of mechanical stimuli applied in the behavioral assays . We first recorded from C-fibers because C-fiber terminals are closely apposed to keratinocytes and non-peptidergic C-fibers project most superficially in the epidermis ( Zylka et al . , 2005 ) . The isolated glabrous skin of the tibial nerve preparation was exposed to apyrase or PBS in the bath . Overall , C-fibers treated with apyrase fired significantly fewer action potentials in response to a series of increasing forces than those treated with PBS ( Figure 4E ) . Examples of C-fiber action potentials in the presence of apyrase or PBS are shown in Figure 4F . Apyrase had no effect on the mechanical thresholds as evaluated by von Frey filaments ( PBS: 5 . 88 ± 5 . 88 mN , apyrase: 5 . 88 ± 5 . 88 mN; median ± interquartile range; Mann-Whitney U-test , p=0 . 60 ) or conduction velocity ( PBS: 0 . 55 ± 0 . 22 m/s , apyrase: 0 . 53 ± 0 . 24 m/s; mean ± SEM; unpaired t-test , p=0 . 79 ) of C-fibers . We next tested myelinated slowly adapting ( SA ) Aδ and Aβ-fibers , which are typically found slightly deeper within the keratinocyte layers of the epidermis . Apyrase significantly reduced the number of mechanically evoked action potentials fired in both SA-Aδ and SA-Aβ-fibers ( Figure 4G and I ) . Example traces of Aδ and Aβ-fiber action potentials show that apyrase decreases the number of action potentials fired in response to increasing forces ( Figure 4H and J ) . Like C-fibers , apyrase had no effect on the mechanical thresholds of Aδ-fibers ( PBS: 4 . 00 ± 2 . 82 mN , apyrase: 4 . 00 ± 2 . 82 mN; median ± interquartile range; Mann-Whitney U-test , p=0 . 70 ) or Aβ-fibers ( PBS: 4 . 00 ± 2 . 82 mN , apyrase: 4 . 00 ± 5 . 19 mN; median ± interquartile range; Mann-Whitney U-test , p=0 . 80 ) as measured by von Frey filaments , or the conduction velocities of Aδ-fibers ( PBS: 7 . 28 ± 0 . 47 m/s , apyrase: 6 . 60 ± 0 . 55 m/s; mean ± SEM; unpaired t-test , p=0 . 36 ) or Aβ-fibers ( PBS: 17 . 71 ± 2 . 87 m/s , apyrase: 14 . 31 ± 0 . 81 m/s; mean ± SEM; unpaired t-test , p=0 . 25 ) . Furthermore , in order to ensure that apyrase would not directly alter membrane excitability of sensory neurons , lumbar dorsal root ganglia ( DRG ) neurons from naive animals were isolated and current clamped in the presence of a high concentration of apyrase or vehicle . Apyrase treatment did not alter the rheobase ( current required to elicit one action potential ) values as compared to the PBS control ( Figure 4—figure supplement 1A ) . In addition , apyrase treatment did not alter the resting membrane potential of the neurons ( Figure 4—figure supplement 1B ) . Together , these data demonstrate that cutaneous ATP signaling is essential for normal behavioral responses to innocuous and noxious mechanical stimuli and is required at cutaneous terminals for afferent firing elicited by mechanical stimuli . We have thus far discovered that: ( 1 ) inhibition of keratinocytes decreases innocuous and noxious touch responses , ( 2 ) mechanical stimulation of keratinocytes releases ATP , and ( 3 ) degradation of ATP in skin decreases mechanical sensitivity at the behavioral and afferent levels . We next asked whether keratinocytes are the major source of ATP released from the skin upon mechanical stimulation . The combined apyrase treatment and 590 nm light inhibition of K14-expressing epidermal cells had no additive effect on behavioral mechanical thresholds ( Figure 5A ) , or on responses to repeated suprathreshold stimulation ( Figure 5B ) when compared to vehicle with 590 nm light . This suggests that K14-expressing cells are the major source of ATP in skin . Importantly , the mechanical thresholds and suprathreshold responses in apyrase-treated animals ( in every genotype and light condition ) were not different from the Arch-K14Cre+ PBS treated 590 nm light condition ( Figure 5A and B ) . As expected , apyrase treatment alone markedly decreased the animals’ mechanical sensitivity as is evident by the increased paw withdrawal thresholds ( Figure 5A ) and decreased responses in the suprathreshold assay ( Figure 5B ) . Further , the non-specific 490 nm light had no effect in either Arch-K14Cre+ or Arch-K14Cre- cohorts ( Figure 5A and B ) . In the noxious needle assay , the effects of apyrase treatment were not significantly different from that of both optogenetic inhibition and apyrase treatment , again suggesting that K14-expressing cells are a major source of the ATP that is required for noxious mechanosensation ( Figure 5C ) . The control 490 nm light had no effect on either Arch-K14Cre+ or Arch-K14Cre- cohorts ( Figure 5—figure supplement 1A ) . Taken together , these findings indicate that other non-K14-expressing cells are not a significant source of mechanically evoked ATP release in skin . Furthermore , because there was no additive effect of apyrase together with optogenetic inhibition of keratinocytes , the data make clear that ATP is the major signaling molecule released from keratinocytes in response to innocuous and noxious mechanical stimulation of normal skin . Next , we assessed whether the light-evoked increase in behavioral responses in ChR-K14Cre+ mice was specifically due to ATP release from keratinocytes by injecting apyrase into hindpaw skin . Apyrase-treated ChR-K14Cre+ animals exhibited their first response to the 473 nm light significantly later than the ChR-K14Cre+ animals treated with PBS ( Figure 5D ) . Furthermore , the response times of apyrase-treated ChR-K14Cre+ mice were similar to those of either apyrase- or PBS-treated ChR-K14Cre- animals ( Figure 5D ) . When animals did respond to the 473 nm light , apyrase treatment had no effect on the type of response in either ChR-K14Cre+ or ChR-K14Cre- animals ( Figure 5—figure supplement 1B ) . Exposure to 589 nm laser stimulation failed to initiate a response in either genotype or treatment group ( Figure 5—figure supplement 1C ) . These data indicate that light-induced activation of keratinocytes is sufficient to evoke attending behavioral responses and that ATP release from skin is an essential signaling molecule involved in these attending responses . Taken together , these results show that ATP is a major signaling molecule released from keratinocytes especially in response to mechanical stimulation . Next , we asked which receptor on sensory nerve terminals responds to the ATP released from keratinocytes . Although there are many P2X family members , P2X4 receptors were of particular interest due to their high abundance and relatively equal expression in C and A-fiber neurons ( Kobayashi et al . , 2013 ) . To pharmacologically inhibit P2X4 in the periphery , the selective P2X4 inhibitor 5-BDBD ( 5- ( 3-Bromophenyl ) -1 , 3-dihydro-2H-benzofuro[3 , 2-e]-1 , 4-diazepin-2-one ) was injected into one plantar hindpaw and mechanical sensitivity was tested . 5-BDBD significantly increased the mechanical thresholds in naive animals ( Figure 6A ) and significantly decreased the responsiveness to repeated mechanical probing compared to vehicle ( Figure 6B ) . Because P2X2 and P2X3 have been shown to be involved in various pain states ( Novakovic et al . , 1999; Cockayne et al . , 2000; North , 2004; Bernier et al . , 2017 ) and because they are also highly expressed on sensory neurons ( Kobayashi et al . , 2013 ) we tested if P2X2 and P2X3 receptors could play a role in baseline mechanical sensation . To pharmacologically inhibit P2X3 and P2X2/3 receptors , two concentrations of NF 110 were injected subcutaneously into the plantar hindpaw of naive animals . At a low concentration ( 500 nM ) , NF 110 inhibits P2X3 , but at a high concentration ( 5 mM ) , NF 110 inhibits both P2X2 and P2X3 receptors ( Hausmann et al . , 2006 ) . Neither concentration affected baseline mechanical sensitivity as measured in either the Up-Down mechanical threshold or in the suprathreshold assay , in these mice 60 min after subcutaneous injection ( Figure 6C and D ) . These data demonstrate that mechanically induced release of ATP from keratinocytes is most likely acting through P2X4 receptors . Conversely , to determine if P2X4 activation sensitizes animals to mechanical stimulation , ivermectin , an anti-parasitic drug that potentiates P2X4 currents induced by ATP binding ( Priel and Silberberg , 2004 ) , was used . Subcutaneous ivermectin administration significantly decreased mechanical withdrawal thresholds ( Figure 6E ) and significantly increased responsivity to repeated suprathreshold stimulation ( Figure 6F ) . Pretreatment with the P2X4 inhibitor 5-BDBD prevented the ivermectin-induced mechanical sensitization in both the mechanical threshold ( Figure 6E ) and repeated force assays ( Figure 6F ) , indicating that the effects of ivermectin were due to P2X4 activation . Once again , these behavioral results could not be accounted for by altered membrane excitability of sensory neurons due to high concentrations of 5-BDBD or ivermectin as neither drug altered the rheobase ( Figure 6—figure supplement 1A and C ) or resting membrane potential of sensory neurons ( Figure 6—figure supplement 1B and D ) compared to their vehicle controls . To determine whether P2X4 receptor activation was specific to receptors on cutaneous sensory nerve terminals , we generated mice with a selective deletion of P2rx4 in sensory neurons . Mice expressing Cre under the control of the sensory neuron-Advillin promoter ( da Silva et al . , 2011; Zappia et al . , 2017 ) were crossed with mice carrying a P2rx4 loxP conditional knockout allele ( Yang et al . , 2014 ) to produce AdvillinCre+P2X4fl/fl ( P2X4-AdvCre+ ) and AdvillinCre-P2X4fl/fl ( P2X4-AdvCre- ) littermate controls . ( Yang et al . , 2014 ) . To confirm that P2rx4 is decreased in DRG , we performed qRT-PCR on whole DRG homogenates . As expected , there was a significant decrease in P2rx4 mRNA in lumbar DRG of P2X4-AdvCre+ animals ( Figure 7—figure supplement 1 ) . P2X4 sensory neuron mutants ( P2X4-AdvCre+ ) displayed significantly higher paw withdrawal thresholds ( Figure 7A ) and significantly fewer responses to repeated mechanical probing when compared with control littermates ( Figure 7B ) . P2X4-AdvCre+ animals were also less responsive to noxious mechanical stimulation . These animals had significantly fewer noxious and innocuous responses and more null responses during the noxious needle assay than littermate controls ( Figure 7C ) . Furthermore , ivermectin injection had no effect on mechanical withdrawal thresholds ( Figure 7D ) or responsiveness to repeated suprathreshold stimulation ( Figure 7E ) in P2X4-AdvCre+ animals . Finally , to determine whether other P2X or P2Y channels besides P2X4 were activated by the mechanically released ATP , hindpaws of sensory neuron P2X4 mutants were injected with apyrase or PBS . P2X4-AdvCre- animals injected with apyrase had significantly higher paw withdrawal thresholds than P2X4-AdvCre- animals injected with PBS ( Figure 7F ) . However , apyrase treatment had no additional effect on mechanical withdrawal thresholds in P2X4-AdvCre+ animals; apyrase treatment had the same effect in both P2X4 expressing and P2X4 sensory neuron mutant mice ( Figure 7F ) . Similarly , apyrase had no effect on P2X4-AdvCre+ animal responsiveness to repeated suprathreshold stimulations ( Figure 7G ) or the noxious needle assay ( Figure 7H ) , but did reduce the responses in P2X4-AdvCre- littermate controls ( Figure 7G and H ) . In conclusion , these data demonstrate that either pharmacological inhibition or genetic deletion of P2X4 channels specifically from sensory nerve terminals is sufficient to fully decrease the baseline mechanical sensitivity to both innocuous and noxious force . Since the Advillin-driven knockout of P2X4 reduced innocuous and noxious mechanical sensitivity , we next asked whether mechanically evoked action potential firing of peripheral sensory neurons was also affected . Single nerve recordings from the ex vivo tibial nerve preparation revealed that C-fibers from P2X4-AdvCre+ animals had overall significantly reduced firing rates as compared to P2X4-AdvCre- controls ( Figure 8A and B ) . Furthermore , the action potential firing rate of slowly adapting myelinated Aδ and Aβ-fibers in response to increasing mechanical stimuli was also decreased overall ( Figure 8C and E ) . Example traces for Aδ and Aβ-fibers are shown in Figure 8D and F . In contrast to the apyrase studies where mechanical thresholds of fibers were measured via traditional von Frey filaments , here we utilize a newly designed custom feedback-controlled mechanical stimulator which applied a force ramp of 0 to 100 mN to determine mechanical thresholds . Remarkably , C-fiber , Aδ-fiber and Aβ-fibers from P2X4-AdvCre+ animals had significantly higher action potential thresholds than their wild type littermates ( Figure 8G–I ) . Of note , we have never before found such a major shift in von Frey threshold at the single fiber level for any knockout or injury model in skin nerve recordings by using von Frey filaments . These data indicate that P2X4 on sensory neuron terminals participates in setting the threshold for mechanical firing of multiple classes of cutaneous sensory neurons . Conduction velocities were not altered by the knockdown of P2X4 in any fiber type; C-fibers ( P2X4-AdvCre-: 0 . 72 ± 0 . 28 m/s , P2X4-AdvCre+: 0 . 76 ± 0 . 27 m/s mean ± SEM; unpaired t-test p=0 . 565 ) , Aδ-fibers ( P2X4-AdvCre-: 3 . 64 ± 1 . 91 m/s , P2X4-AdvCre+: 3 . 88 ± 2 . 29 m/s; mean ± SEM; unpaired t-test p=0 . 661 ) or Aβ-fibers ( P2X4-AdvCre-: 12 . 87 ± 2 . 67 m/s , P2X4-AdvCre+: 12 . 76 ± 2 . 98 m/s; mean ± SEM; unpaired t-tests , p=0 . 913 ) . Together , these data identify sensory neuron P2X4 as a key target of the ATP released from keratinocytes in response to both noxious and innocuous touch and show that sensory neuron-expressed P2X4 is essential for setting both the threshold for initiating mechanical firing and for regulating the firing frequency to suprathreshold sustained stimuli in several classes of slowly adapting sensory neurons in the skin .
Our study builds on the elegant ex vivo findings of Bambauer and colleagues ( 2015 ) to demonstrate the tactile function of keratinocytes in the awake , behaving animal . Specific optogenetic inhibition of epidermal cells in Arch-K14Cre+ mice elevated the behavioral tactile thresholds , dampened the responses to suprathreshold innocuous force , and blunted the responses to noxious pin prick , indicating that keratinocytes have a major role in conveying a broad range of innocuous and noxious mechanical information to the CNS . Although this optogenetic silencing probably inhibited Merkel cells , which also express Keratin14 , the contribution from this cell type to the overall behavioral observations is likely small since Merkel cells comprise only a very small portion ( 3–6% ) of epidermal cells located in the glabrous or hairy skin ( Moll et al . , 1986; Fradette et al . , 2003; Halata et al . , 2003 ) . Therefore , the K14 epidermal cell inhibition in vivo is largely mediated by keratinocytes . Complementary experiments , where keratinocytes were selectively activated by Channelrhodopsin in ChR-K14Cre+ mice , showed that specific keratinocyte activation caused animals to attend or ‘groom’ in response to the light-illuminated body regions by wiping of the face or front paws in both the evoked behavior and non-evoked place preference assay . While both assays reveal attending and grooming-like behaviors , there are also major differences in the types of behaviors observed . The light-evoked behavior from focal hindpaw stimulation appears to be nocifensive ( Pang et al . , 2015 ) , with behaviors mimicking those observed in experiments utilizing optogenetic activation of TRPV1+ sensory neurons , which causes nocifensive behaviors as well as place aversion ( Beaudry et al . , 2017 ) . However , in our hands , the place-preference assay did not cause a place aversion in the ChR-K14Cre+ animals , and the attending behaviors elicited by the 460 nm floor light are most reminiscent of paresthesia-like behaviors ( Kahn Safdar et al . , 2012 ) . Further experiments must be done to determine the repertoire of sensations elicited during these interesting behaviors . Since the same set of epidermal cells are activated by 460 nm light in both assays , we believe the discrepancies in these two behavioral assays are due to different light power intensities and consequently , different levels of keratinocyte depolarization ( 75 . 2 μW LED floor in the place preference assay vs . 25 mW laser in the hindpaw light-evoked assay , a 333-fold difference in light intensity ) . This hypothesis is further supported by findings in our ChR cell sniff assay where P2X2 expressing HEK-293 cells exhibit intensity-dependent increases in inward currents in response to light stimulation of ChR-expressing keratinocytes , indicating an intensity-dependent release of ATP by keratinocytes . These data show that keratinocyte depolarization alone is sufficient to cause attending behaviors in freely moving animals . It may be surprising that keratinocyte function in vivo can be modulated by optogenetic manipulation of the membrane because these cells do not fire action potentials . However , we show here that primary keratinocytes from adult glabrous skin do depolarize upon mechanical stimulation in an indentation-dependent manner . Moreover , Arch inhibition during focal mechanical stimulation of keratinocytes significantly reduced the overall level of depolarization at each indentation . Conversely , we show that ChR activation directly elicits inward currents in P2X2 expressing HEK-293 cells , thereby showing that the depolarization of keratinocytes is sufficient to release ATP . Other reports using keratinocyte cell lines show that keratinocytes can depolarize or hyperpolarize in response to changes in extracellular ionic gradients ( Wohlrab et al . , 2000 ) , exhibit increased intracellular Ca2+ in response to mechanical stimuli ( Koizumi et al . , 2004; Tsutsumi et al . , 2009; Goto et al . , 2010 ) , and express voltage-gated sodium and calcium channels , as well as Transient Receptor Potential channels ( Denda et al . , 2006; Zhao et al . , 2008; Caterina and Pang , 2016 ) . These prior reports suggest that keratinocytes possess the functional ion channels required for producing rapid changes in membrane excitability , and our new data reveal that naive adult mouse keratinocytes depolarize in response to mechanical stimuli . We next sought to determine how keratinocytes must be communicating with sensory nerve terminals . This process likely occurs via a chemical signaling pathway based on evidence of synapse-like structures between keratinocytes and sensory nerve terminals ( Hilliges et al . , 1995; Chateau and Misery , 2004; Château et al . , 2007; Klusch et al . , 2013; Roggenkamp et al . , 2013 ) and the fact that keratinocytes have been shown to contain and release a variety of neurotransmitter molecules ( Burrell et al . , 2005; Lumpkin and Caterina , 2007; Barr et al . , 2013; Hou et al . , 2011Shi et al . , 2013 ) . We found at the cellular , tissue and behavioral levels evidence that ATP is released from keratinocytes in response to mechanical stimulation of the skin . Experiments utilizing ex vivo glabrous hindpaw skin show that mechanical stimulation elicits reproducible and graded ATP release , and experiments employing a cell sniff assay verify that keratinocytes are specifically responsible for the ATP release , which occurs in an indentation- and voltage-dependent fashion . To complement the ATP release studies , we tested the converse by degrading ATP using the enzyme apyrase ( Palygin et al . , 2015 , Palygin et al . , 2017 ) . Apyrase decreased innocuous and noxious mechanical responses in evoked behavior assays in vivo , reduced mechanically evoked action potential frequencies for all primary afferent fiber types ( C , Aδ and Aβ-fibers ) tested in skin-nerve recordings , and diminished inward currents in P2X2 expressing HEK cells during mechanical stimulation of keratinocytes . Collectively , these data suggest that graded ATP signaling is essential for the transmission of mechanically relevant information between keratinocytes and sensory neurons . While platelets , fibroblasts and neurons also release ATP and are present in the skin ( Fukami and Salganicoff , 1977; Grierson and Meldolesi , 1995; Lazarowski et al . , 2003; Abbracchio et al . , 2009 ) , it is unlikely that ATP stores from these cells are critical for mechanotransduction since in vivo apyrase treatment did not amplify the effects of specific optogenetic inhibition of keratinocytes . Together , these data indicate that during mechanical stimulation ( 1 ) ATP is a key signaling molecule released from keratinocytes , and ( 2 ) keratinocytes are the predominant source of ATP that is released . Additionally , our findings in the Archaerhodopsin cell sniff assay coupled with the data showing that keratinocytes depolarize in an indentation-dependent manner in response to mechanical stimulation , argue that the release of ATP from keratinocytes in response to mechanical stimuli is largely voltage-dependent , thereby , pointing towards a vesicular release mechanism for ATP . Other studies utilizing normal human epidermal keratinocytes suggest that that ATP can be released via a calcium-dependent vesicular mechanism and/or non-vesicular mechanism via connexin hemichannels ( Harden and Lazarowski , 1999; Lazarowski et al . , 2003; Inoue et al . , 2014; Barr et al . , 2013 ) . Further studies are needed to identify the exact ATP release mechanism via mechanical stimulation of skin and primary mouse keratinocytes . The keratinocyte-released ATP must be acting through a specific receptor or set of receptors on sensory nerve terminals in order to convey the innocuous and noxious touch signal ( s ) to the spinal cord . Our studies estimate the amount of ATP release to be in the micromolar range , which is an amount sufficient to activate most P2X receptors ( Jacobson et al . , 2002 ) ; however , the levels of ATP we measured are likely underestimations given that both keratinocytes and sensory neurons express ectonucleotidases ( Lazarowski et al . , 2000; Zylka et al . , 2008; Sowa et al . , 2010a; Sowa et al . , 2010b ) . Additionally , it is possible that ATP is focally released in high concentration pocket ‘domains’ between the sensory neuron and keratinocyte cell membranes , and therefore , the ATP concentrations that occur in those localized signaling regions might be much higher than the generalized levels we measured in our assays . Although there are a number of P2X channels that have been shown to be expressed by sensory neurons ( Kobayashi et al . , 2005 ) , we chose to investigate P2X4 because it is the most highly expressed P2X receptor on sensory neurons and because of its relatively equal expression on both light touch and nociceptive neurons ( Kobayashi et al . , 2005 ) . Indeed , both pharmacological inhibition and genetic ablation of P2X4 in sensory neurons reduced tactile thresholds , blunted suprathreshold responses , and dampened responses to noxious pin prick . Accordingly , activation of P2X4 via a positive allosteric modulator decreased mechanical thresholds , and increased responses to suprathreshold stimuli , but had no effect in P2X4-deficient mice . Furthermore , the ablation of P2X4 in sensory neurons decreased mechanical responsiveness of primary afferent fibers in the ex vivo skin nerve preparation , as reflected by both significantly elevated thresholds and decreased afferent firing , especially at the higher intensity stimulus for all slowly adapting fiber types tested . These data indicate that the ATP released from keratinocytes is most likely signaling to P2X4 receptors on sensory neurons . Peripheral inhibition of P2X2 and P2X3 receptors , which are also highly expressed on sensory neurons ( Kobayashi et al . , 2005 ) and have been shown to be involved in various pain states ( Novakovic et al . , 1999; Cockayne et al . , 2000; North , 2004; Bernier et al . , 2017 ) , had no effect on baseline tactile thresholds or responses to a suprathreshold stimulus . Furthermore , in vivo degradation of ATP had no additional behavioral effect in sensory neuron-specific P2X4 knockout mice , therefore indicating that ATP signaling occurs mainly through P2X4 receptors on sensory neurons . It may be surprising that P2X4 was identified in our study as the key target of mechanically released ATP . However , purinergic signaling appears to be more multifaceted than would be expected by simply determining the probability of ATP binding via EC50 values because ( 1 ) receptors can also exist in heteromeric confirmations and ( 2 ) P2 receptors have been shown to have complex response patterns , where rather than having distinct individual roles , different P2 receptors have been shown to work in concert through having both additive and inhibitory interactions ( Xing et al . , 2016 ) . Sensory neurons are intrinsically capable of sensing mechanical stimuli via the activation of mechanosensitive ion channels found on their terminals . Studies have identified Piezo2 and Transient Receptor Potential Ankyrin 1 ( TRPA1 ) as key mechanosensitive ion channels found on sensory neurons ( Coste et al . , 2010; Kwan et al . , 2009 ) ; Vilceanu and Stucky , 2010 ) ; Woo et al . , 2015 ) . ATP is likely a potentiator , rather than an initiator , of action potential firing in response to mechanical stimulation of the skin because we observed blunting of behavioral and afferent mechanical responses and decreased sensitivity thresholds in the various assays instead of finding a complete lack of responsiveness to force . Furthermore , in addition to ATP , keratinocytes can release a variety of chemical factors such as CGRPβ , β-endorphins , endothelin-1 , neurotrophins , and cytokines ( Lumpkin and Caterina , 2007; Hou et al . , 2011; Shi et al . , 2013 ) , all of which can activate receptors on sensory nerve terminals . However , our data utilizing Arch-inhibition in combination with apyrase shows that ATP most likely is the major molecule released from keratinocytes upon innocuous and noxious touch at baseline in non-injured skin . However , after skin injury or disease , it is possible that one or more of these factors potentiates signals after injury in addition to ATP . If true , this injury-induced augmentation of keratinocyte communication could provide sensory neurons with numerous opportunities for initiation and amplification of signaling mechanisms that underlie pain , itch or dysesthesia in the setting of disease . Through the use of a glabrous skin-tibial nerve preparation , we show that either ATP hydrolysis or genetic knockdown of P2X4 in sensory neurons diminishes mechanically evoked activity in all primary afferent fiber types tested . It should be noted that the dampening of the afferent firing rate was much more prominent in the genetic P2X4 mutant model than in the experiment where apyrase was applied to the receptive fields via bath exposure . Moreover , effects on mechanical thresholds of fibers were observed in the genetic P2X4 mutants but not in the apyrase experiments . We believe that the lesser effect in the apyrase experiment is most likely due to the apyrase enzyme , which is diluted in aqueous buffer , not being able to penetrate and fully distribute within the tissue to the receptive terminals of sensory neurons . Nonetheless , in both the P2X4 mutant and apyrase teased fiber experiments , there were effects on all fiber types studied including slowly adapting Aβ , Aδ and C-fibers . These data indicate that keratinocytes are not only communicating with the more superficial non-peptidergic C-fibers ( i . e . MrgD+/IB4- binding C-fibers ) , but also signaling to the deeper-projecting peptidergic C-fibers , Aδ-fiber nociceptors and slowly adapting Aβ-fibers that mediate light touch , all of which are closer to the dermal-epidermal border in the skin ( Zylka et al . , 2005; Basbaum et al . , 2009; Abraira and Ginty , 2013; Le Pichon and Chesler , 2014 ) . These data are supported by previous studies , which also showed that optogenetic stimulation of K14-expressing cells activated slowly adapting Aδ and Aβ-fibers ( Baumbauer et al . , 2015 ) as well as C-fibers ( Baumbauer et al . , 2015; Pang et al . , 2015 ) . Consequently , these data together indicate that keratinocyte signaling is essential in potentiating signals of many fiber subtypes and that keratinocyte-sensory neuron communication is not a fiber-type-specific phenomenon . Another interesting and novel finding is that all fiber types tested in P2X4 sensory neuron knockout animals also show elevated action potential thresholds , indicating decreased mechanical sensitivity at the terminal receptive field of single fibers . We have never before observed significant , potentially biologically relevant changes in fiber mechanical thresholds in any genetic mutant or injury study . The most parsimonious reason for this novel difference is that instead of using von Frey filaments as we have always done in the past , we utilized a new custom-designed feedback-controlled mechanical stimulator probe to exert a continuous force ramp from 0 to 100 mN perpendicularly to each fiber’s skin receptive field . The surface area of the new stimulator probe utilized to exert this force is flat , circular and larger ( 0 . 8 mm diameter; approximately 2–4 times larger ) than the tip of a typical von Frey filament which is small and pointed ( 4 mN: 0 . 19 mm diameter; 6 mN: 0 . 25 mm diameter ) . Thus , the new probe used for the ramp may stimulate the receptive field more evenly and consistently than the von Frey filament tip which may have sharper edges that deliver more punctate stimuli to the receptive field and activate the fiber at lower thresholds . This idea is generally supported by previous evidence that cutaneous sensory terminals are finely tuned to detect and encode the edges of objects touched ( Wheat and Goodwin , 2001 ) . This is the first study that establishes a clear role for keratinocyte-initiated purinergic signaling in mechanotransduction at the cellular , tissue and behavioral levels . It has been shown that diseases such as complex regional pain syndrome and post-herpetic neuralgia are accompanied by increased epidermal ATP release which can lead to excessive activation of P2X receptors on sensory neurons ( Zhao et al . , 2008 ) . Many other skin disorders , such as dermatitis and psoriasis , are characterized by altered keratinocyte function and signaling , and also share cutaneous pain as a common debilitating symptom that leads to severely decreased quality of life in affected patients ( Man , 2011 ) . If mechanical allodynia and hyperalgesia can effectively be treated at the site of pain ( i . e . the skin , via interfering with keratinocyte-sensory neuron communication ) , it would allow for easy , non-invasive treatment options that avoid the central nervous system-mediated side effects of most current pain treatments , including opioid analgesic drugs . Our current study encourages further exploration into ATP and P2X4 as valuable targets for novel topical analgesics and antipruritics .
Adult male C57BL/6J mice from Jackson Laboratories ( Jackson stock number 000664; Bar Harbor , Maine ) of at least 8 weeks of age were used for pharmacological studies ( apyrase , 5-BDBD , ivermectin with 5-BDBD , and NF 110 ) , ATP biosensor , and skin nerve experiments . For all other studies male and female mice aged 7–16 weeks were used . Male and female mice were analyzed separately , however , since no sex differences were observed , the data for all studies show combined results of both sexes . Keratin14 ( K14 ) is expressed in all keratinocytes as early as E9 . 5 ( Byrne et al . , 1994; Dassule et al . , 2000; Wang et al . , 1997 ) . To create the mouse line that selectively expresses GFP-tagged Archaerhodopsin-3 in K14-positive cells , Ai35D ( B6;129S-Gt ( ROSA ) 26Sortm35 . 1 ( CAG-aop3/GFP ) Hze/J ) Archaerhodopsin ( Jackson stock number 012735 ) and B6N . Cg-Tg ( KRT14-cre ) 1Amc ( Jackson stock number 018964 ) lines were mated . Offspring were genotyped as either K14Cre+ Arch/Arch ( Arch-K14Cre+ ) or K14Cre- Arch/Arch ( Arch-K14Cre- ) . To create a complementary line that selectively expresses eYFP-tagged Channelrhodopsin-2 in K14-positive cells , Ai32 ( B6;129S-Gt ( ROSA ) 26Sortm32 ( CAG-COP4+H134R/EYFP ) Hze/J ) enhanced Channelrhodopsin 2 ( Jackson stock number 012569 ) and B6N . Cg-Tg ( KRT14-cre ) 1Amc ( Jackson stock number 018964 ) lines were mated . Offspring were genotyped as either K14Cre+ChR/ChR ( ChR-K14Cre+ ) or K14Cre- ChR/ChR ( ChR-K14Cre- ) . To create mice that selectively express tdTomato in Keratin14-expressing cells were created by Ai14; B6 . Cg-Gt ( ROSA ) 26Sortm14 ( CAG-TDTomato ) Hze/J ( Jackson stock number 007914 ) with B6N . Cg-Tg ( KRT14-cre ) 1Amc ( Jackson stock number 018964 ) . Conditional P2rx4 knockout animals were generated by mating Wang Advillin Cre mice ( previously described in da Silva et al . , 2011 ) with P2rx4 animals , which were generously provided to us by Dr . Bruce Liang ( Yang et al . , 2014 ) . Offspring were genotyped as either AdvillinCre+P2X4fl/fl ( P2X4-AdvCre+ ) and AdvillinCre-P2X4fl/fl ( P2X4-AdvCre- ) . As a note , Advillin has also been found to be expressed in Merkel cells of the glabrous epidermis , and has been uses as to create knockout models to study Merkel cells ( Ranade et al . , 2014 ) . All genotypes were confirmed by PCR . Animals were housed in a climate-controlled room with a 14:10 light:dark cycle , on Sani-Chips an aspen wood chip bedding ( P . J . Murphy Forest and Products , New Jersey ) with Enviro-dri nesting material ( Shepherd Specialty Papers , Michigan ) and ad libitum access to food and water . All animals were group housed . Animal procedures adhered to the NIH Guide for the Care and Use of Laboratory animals , and were performed in accordance with the Institutional Animal Care and Use Committee at the Medical College of Wisconsin ( approval #: 0383 ) . Animals were randomly assigned different treatment groups . Glabrous skin of the mouse hindpaw was dissected and tissue was fixed in 4% paraformaldehyde . After 30% sucrose cryoprotection , skin was processed into 6 μm sections using a ThermoFisher Scientific Microm HM550 . Sections were blocked in 10% normal donkey serum , 0 . 3% Triton X-100 in PBS , then incubated overnight at 4˚C in rabbit anti-keratin 14 primary antibody ( 1:500; AB_2616894; Biolegend , San Diego , CA ) . After rinsing , sections were incubated in donkey anti-rabbit AlexaFluor 594 ( 2 drops/1 mL; R37119; Invitrogen ) for 30 min . Immunofluorescent images were taken on a Nikon Eclipse E800 confocal microscope equipped Nikon EZ-C1 software ( Nikon Instruments , Melville , NY ) . The experimenter was blinded to genotype throughout the staining and imaging procedure and at least 3 animals of each genotype were used . Images were assessed in ImageJ ( National Institutes of Health , Bethesda , MD ) and Adobe Illustrator software ( Adobe Systems Inc . , San Jose , CA ) . Glabrous skin was isolated from the hindpaw as described above and incubated in 10 mg/mL dispase ( Gibco , ThermoFisher Scientific , Waltham , MA ) for 45 min at room temperature . Epidermal sheets were peeled from the dermis , then incubated in 50% EDTA ( Sigma-Aldrich ) in Hanks’ Balanced Salt Solution without calcium chloride , magnesium chloride and magnesium sulfate ( Gibco ) for 27 min at room temperature . Sheets were exposed to 15% heat inactivated fetal bovine serum ( ThermoFisher Scientific , Carlsbad , CA ) then rubbed against the base of a petri dish to separate the keratinocytes from the epidermal sheets . The mixture was then centrifuged , the supernatant removed and the pellet was then re-suspended and cells were grown in Epilife media ( Gibco ) supplemented with 1% human keratinocyte growth supplement ( Gibco ) , 0 . 2% GibcoAmphotericin B ( 250 µg/mL of Amphotericin B and 205 µg/mL sodium deoxycholate , Gibco ) and 0 . 25% penicillin-streptomycin ( Gibco ) on laminin coated coverslips . Plates were incubated and grown in 37°C and 5% CO2 conditions . Growth media was exchanged every 2 days . Keratinocytes were used 3 days after plating . For primary keratinocyte cell cultures , a mixture of male and female mice were utilized , although no significant sex differences were observed and therefore , the data for all studies show combined results from both sexes . DRG neurons after overnight culture and keratinocytes 3 days after plating were viewed on a Nikon Eclipse TE200 inverted microscope . Cells were continuously superfused with extracellular normal HEPES solution containing ( in mM ) : 140 NaCl , KCl , 2 CaCl2 , 1 MgCl2 , 10 HEPES , and 10 glucose , pH 7 . 4 ± 0 . 05 , and 310 ± 3 mOsm . GFP+ HEK293 cells ( holding voltage −40 mV ) , keratinocytes ( holding voltage −50 mV ) or DRG neurons ( holding voltage −70 mV ) were patched in voltage clamp mode with borosilicate glass pipettes ( Sutter Instrument Company , Novato , CA ) filled with intracellular normal HEPES solution containing ( in mM ) : 135 KCl , 10 NaCl , 1 MgCl2 , 1 EGTA , 0 . 2 NaGTP , 2 . 5 ATPNa2 , and 10 HEPES , pH 7 . 20 ± 0 . 05 , and 290 ± 3 mOsm . Cell capacitance and series resistance were kept below 10 MΩ for all cell types . Mechanical stimulation of cells occurred at a rate of 106 . 25 μm/ms by a second borosilicate glass pipette that was driven by a piezo stack actuator ( PA25 , PiezoSystemjena , Jena , Germany ) . Cells were stimulated with increasing displacements of 1 . 7 μm/Volt over 200 ms . To avoid sensitization and/or desensitization of keratinocytes , 3 min were given between each displacement . Recordings were made in Pulse software via an EPC9 amplifier ( HEKA Electronics , Holliston , MA ) . Cells were included in the study if the leak current stayed below 200 pA for at least three stimulations . Data were analyzed using FitMaster ( HEKA Electronics ) . Keratinocytes were used on day 3 days after plating . Keratinocytes from K14-Arch animals were patched in current clamp mode ( held at 0 pA ) ; resting membrane potentials were measured at baseline , during 1 min of 590 nm LED light ( 4 mW and 5 mW , Thorlabs Inc . Newton , NJ ) exposure , and 1 min after 490 nm light cessation . Keratinocytes from K14-ChR animals were patched in voltage clamp mode ( held at −50 mV ) using the method described above . Cells were exposed to 490 nm LED light ( 490 nm; 0 . 2 , 2 , 3 , and 20 mW , Thorlabs ) in 30 second increments; peak and sustained currents were recorded . In all assays , animals were randomly assigned to treatment groups; the experimenter was blinded to genotype and/or treatment . All assays were performed between 8 am and 2 pm , and mice were acclimated to their surroundings and the experimenter for at least 1 hr prior to testing . Adult male C57BL/6J mice from Jackson Laboratories of at least 8 weeks of age were used for pharmacological studies ( apyrase , 5-BDBD , ivermectin with 5-BDBD , and NF 110 ) , apyrase skin nerve and ATP biosensor experiments . Approximately equal numbers of male and female mice were used for ( K14-Arch , K14-ChR and P2X4-AdvCre ) experiments , and because no significant differences were noted between the sexes , data from both sexes were combined . To determine if significant heating of the paw occurred in response to either the LED or laser stimulation , an implantable thermocouple microprobe was inserted into the glabrous skin ( Physitemp; Clifton , NJ ) . For this procedure , animals were anesthetized with 1 . 5% isoflurane; body temperature was measured throughout the procedure and maintained with a heating pad . Laser coupled fiber optics and LEDs were held at the appropriate distance from the hindpaw and temperature measurements were made over a 6 min time window . To assess changes in the presence or absence of apyrase of primary afferent firing , we utilized tibial skin-nerve preparations , as described ( Reeh , 1988 ) . We chose to use the tibial nerve because it innervates a majority of the glabrous skin of the mouse hindpaw , which was tested in all behavior assays . Animals were briefly anesthetized and then sacrificed via cervical dislocation . The leg of the animal was then shaved with commercial clippers , and the glabrous skin and tibial nerve was quickly removed and placed in a heated ( 32 +- 0 . 5˚C ) , oxygenated bath consisting of ( in mM ) : 123 NaCl , 3 . 5 KCl , 2 . 0 CaCl2 , 0 . 7 MgSO4 , 1 . 7 NaH2PO4 , 5 . 5 glucose , 7 . 5 sucrose 9 . 5 sodium gluconate and 10 HEPES . The buffer pH was then adjusted to a pH of 7 . 45 +- 0 . 05 . Either PBS or 40 units of apyrase were added to the bath where the skin was kept ( experimenter was blinded to the treatment group ) . To keep the skin in place it was pinned down with insect needles , and the tibial nerve was placed in a chamber with a mirror plate . The nerve end was kept on the mirror plate surrounded by mineral oil while it was being teased into small fascicles . These small bundles were then placed on the recording electrode and a blunt glass probe was used to mechanically stimulate the preparation in order to find receptive fields of single afferent fibers . Fibers were characterized based on their shape and conduction velocities: C-fibers < 1 . 2 m/s; Aδ-fibers 1 . 2–10 m/s; and Aβ-fibers for conduction velocities over 10 m/s ( Koltzenburg et al . , 1997 ) . Of note , in the glabrous skin-tibial nerve preparation , all Aδ fibers we encountered were slowly adapting , and the majority of Aβ fibers we encountered were slowly adapting . Thus , only slowly adapting afferents were included in this study . To determine action potential firing thresholds , von Frey filaments were utilized in apyrase experiments . However , action potential thresholds in P2X4-AdvCre experiments were determined using a continuous force ramp from 0 to 100 mN utilizing a new custom designed feedback-controlled mechanical stimulator . Once a fiber was identified a baseline recording of their firing activity was recorded for 2 min . Next , a feedback-controlled mechanical stimulator was placed over the receptive field to stimulate it with increasing forces . For apyrase studies , the receptive field was stimulated with 20 , 40 , 75 , and 110 mN for 12 seconds for C-fibers and for Aδ-fibers and Aβ-fibers with 25 , 40 , 75 , 110 , and 150 mN for 12 seconds . For P2X4-AdvCre experiments a new custom designed closed-loop feedback-controlled mechanical stimulator was used , which consists of three motorized and linear stages ( T-LSM200A , Zaber Technologies Inc . , Vancouver , BC , Canada ) configured as a Cartesian ( x , y , z ) gantry . Using an ultra low force transducer ( F30 , Harvard Apparatus , Holliston , MA ) mounted to the vertical , z-axis of this gantry , mechanical stimulations ( 2 , 5 , 10 , 20 , 40 , 100 and 150 mN for 10 seconds ) of the receptive field were performed . To prevent sensitization and desensitization of the fiber , a 1 min interval was given between the different forces . Data was recorded via Labchart ( ADInstruments; Colorado Springs , CO ) . In order to develop a cell line that over expresses P2rx2 , 1 million HEK293 ( ATCC ) cells were transfected with 500 ng of a plasmid expressing a C-terminal GFP-tagged P2rx2 plasmid ( RG216207 Origene , Rockville , MD ) using the Lonza 4D nucleofector ( Basel , Switzerland ) . P2X2 receptors were chosen due to their favorable ion channel kinetics ( Coddou et al . , 2011 ) . Following transfection , cells were sorted on a FACSARIA cell sorter ( San Jose , CA ) to select for GFP-expressing cells 72 hr post transfection . To maintain the cells , once every other week , cells were sorted for GFP expression to establish a cell line that has stable GFP-tagged P2rx2 overexpression . GFP- cells in the culture were used as internal controls for the naive cell sniff assay . Primary mouse keratinocytes from K14-tdTomato Cre+ animals were cultured for at least 3 days before being utilized in the cell sniff assay . HEK293 ( ATCC ) cells transfected with P2X2 GFP construct ( Origene ) in close proximity to the keratinocytes . The experimenter was blinded to which transfected HEK cell line was utilized in the experiment . Round coverslips containing mixed cultures were viewed on a Nikon Eclipse TE200 inverted microscope . Keratinocytes in close proximity ( 1–15 μm ) to patched P2X2-GFP+ and GFP- HEK293 were mechanically stimulated as described above . When optogenetic mouse lines were utilized , either a 590 nm ( 5 mW ) or 490 nm ( 0 . 2–20 mW ) LED was mounted on top of the microscope and turned on during recordings . Mice were euthanized and sensory neurons were isolated from bilateral lumbar 1–6 ganglia . Isolated DRG were incubated with 1 mg/mL collagenase type IV ( Sigma , St . Louis , MO ) for 40 min at 37°C and 5% CO2 , followed by a 45 min incubation with 0 . 05% trypsin ( Sigma ) . The DRG were then mechanically dissociated and plated onto laminin-coated glass coverslips . Two hours after plating , neurons were fed with media containing Dulbecco’s modified Eagle’s medium/Ham’s F12 medium , supplemented with 10% heat-inactivated horse serum , 2 mM L-glutamine , 1% glucose , 100 units/ml penicillin , and 100 µg/ml streptomycin . The media contained no exogenous growth factors . RNA was obtained from DRG samples by manually homogenizing tissue in Trizol then extracting nucleic acids using the Purelink RNA Micro Scale kit ( ThermoFisher Scientific ) . All RNA samples were reverse transcribed into cDNA using the Superscript Variable Input Linear Output ( VILO ) cDNA Synthesis Kit ( ThermoFisher Scientific ) . qRT-PCR was performed on a Bio-Rad C1000 Touch CFX96 Real-Time-System Thermal Cycler ( Bio-Rad Laboratories Inc . , Hercules , CA ) using the following TaqMan primers and probes ( LifeTechnologies ) for mouse P2X4 and mouse GAPDH . Paw withdrawal thresholds and suprathreshold stimulus responses were compared between two groups using non-parametrcic Mann-Whitney U-test , and between three groups using a Kruskal-Wallis test . For groups that had a two by two set up , a two-way ANOVA with Tukey post-hoc test was used . For datasets that had a two by two by two set up , a three-way ANOVA with Tukey post-hoc test was used . Skin nerve recordings were analyzed using a repeated measures two-way ANOVA with Sidak post-hoc test . Types of responses to laser stimulation were analyzed using Chi square test with Fisher’s exact for groupsof two , and Dirichlet-multinomial regression using the FMM procedure in SAS 9 . 4 ( SAS Insitute , Cary , NC ) for multiple groups , such as after apyrase/PBS treatment . Gene expression was analyzed using a two-tailed parametric t-test . Rheobase values were analyzed using a Mann-Whitney U-test , and resting membrane potentials were analyzed using a two-tailed parametric t-test . Summarized data are reported as mean ± SEM except for apyrase and PBS von Frey thresholds for the different fiber types which are reported as median ± interquartile range in the text . For all behavior experiments ‘n’ corresponds to the number of animals . For patch clamp studies , skin nerve recordings or ATP measurements at least n = 3 animals were utilized in each group analyzed , and the n on the graph corresponds to the number of cells , fibers , or repetitions . All data analyses were performed using Prism 7 software ( GraphPad , La Jolla , CA ) , with an alpha value of 0 . 05 set a priori . *p<0 . 05 **p<0 . 01 , ***p<0 . 001 , ****p<0 . 0001 , n . s . denotes a non-significant comparison . | The skin is the largest sensory organ of the body , and the first point of contact with the outside world . Whether it is being pinched or caressed , the skin’s sense of touch informs organisms about their surroundings and allows them to react appropriately . Nerve cells present in the skin capture information about touch and transmit it to the brain where it is decoded . However , there are many other types of cells in the skin besides nerve cells . The role that these other skin cells play in perceiving non-painful and painful touch is still unclear . Moehring et al . now report how the skin cells that form 95% of the most outer layer of the skin are involved in detecting touch . In mutant mice whose cells can be ‘switched off’ by a certain light , artificially deactivating these cells makes the animals less able to respond to tactile stimuli . Further experiments show that when pressure is applied onto the skin , the surface skin cells release a chemical messenger , which then binds specifically to the nerve cells . When the messaging molecule is experimentally destroyed or prevented from attaching to the nerve cell , the mice react less to non-painful and painful touch . This means the cells at the surface of the skin detect tactile signals from the environment and then communicate this information to the nerve cells , where it is taken to the brain . Disrupted communication between the cells in the outer layer of the skin and the nerve cells is found in painful and itchy skin conditions such as eczema and psoriasis . Knowing how these two types of cells normally work together may help with finding new pain and itch treatments for these skin disorders . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"neuroscience"
] | 2018 | Keratinocytes mediate innocuous and noxious touch via ATP-P2X4 signaling |
Mutualism is of fundamental importance in ecosystems . Which factors help to keep the relationship mutually beneficial and evolutionarily successful is a central question . We addressed this issue for one of the most significant mutualistic interactions on Earth , which associates plants of the leguminosae family and hundreds of nitrogen ( N2 ) -fixing bacterial species . Here we analyze the spatio-temporal dynamics of fixers and non-fixers along the symbiotic process in the Cupriavidus taiwanensis–Mimosa pudica system . N2-fixing symbionts progressively outcompete isogenic non-fixers within root nodules , where N2-fixation occurs , even when they share the same nodule . Numerical simulations , supported by experimental validation , predict that rare fixers will invade a population dominated by non-fixing bacteria during serial nodulation cycles with a probability that is function of initial inoculum , plant population size and nodulation cycle length . Our findings provide insights into the selective forces and ecological factors that may have driven the spread of the N2-fixation mutualistic trait .
The evolutionary dynamics of mutualistic interactions between higher organisms and microbes depends to a large extent on the transmission mode of microbial symbionts . Vertical transmission is expected to promote fitness alignment of obligate symbionts and their partners ( Herre et al . , 1999 ) . In contrast , horizontal transmission generates more complex ecological cycles for facultative symbionts . When going through these cycles , microbes are subjected to several trade-offs regarding host range ( specialist vs . generalist ) and investment in the mutualism ( good or bad cooperator , life in the host vs . outside the host ) . The large number of possible strategies to maximize fitness , and the ability to segregate in a population of genetically variable partners , often entails conflicts of interests between symbionts and their hosts ( Bever et al . , 2009; Sachs et al . , 2010; Porter and Simms , 2014; Jones et al . , 2015 ) that may result in the classic Tragedy of the Commons ( Hardin , 1968 ) . The emergence and stability of mutualism thus requires that proliferation of symbionts is allowed but restricted to appropriate spaces and times and that beneficial partners are ultimately favored over uncooperative ones ( Vigneron et al . , 2014; Visick and McFall-Ngai , 2000; Koch et al . , 2014 ) . The theoretical aspects of the evolution and maintenance of mutualistic interactions have been extensively discussed ( Archetti et al . , 2011; Akcay , 2015 ) . Yet experimental assessment is scarce and the impact of ecological factors , such as population size of hosts and symbionts or the duration of the interaction , has been under-explored , although they are an essential component of the evolutionary potential of symbiotic systems . Rhizobia , the N2-fixing symbionts of legumes , induce the formation of and massively colonize nodules , where intracellular bacteria fix atmospheric nitrogen for the benefit of the plant in exchange for photosynthates . When the nodule senesces , nodule bacteria are released to the soil where they can return to free-living lifestyle and/or colonize a new host ( Thies et al . , 1995 ) . During evolution , symbiosis modules carrying genes essential for the symbiotic process have spread to many different taxa so that extant rhizobia are distributed in hundreds of species in 14 genera of α- and β-proteobacteria ( Remigi et al . , 2016 ) . Acquisition of symbiotic genes may not be sufficient to create an effective symbiont and may lead to bacteria exhibiting various levels of symbiotic capacities ( Nandasena et al . , 2006; Nandasena et al . , 2007; Marchetti et al . , 2010 ) that can be further optimized and maintained under legume selection pressure ( Marchetti et al . , 2017; Marchetti et al . , 2014 ) . It has been established that bacteria better able to form and infect nodules are selected by a partner choice mechanism involving the specific recognition of bacterial molecular signals by plant receptors ( Kawaharada et al . , 2015; Radutoiu et al . , 2003 ) . Bacterial features that are recognized by the plant include Nod factors that initiate rhizobial entry and nodule formation ( Perret et al . , 2000; Broghammer et al . , 2012 ) , and lipo/exopolysaccharides critical for root infection and bacterial release inside the plant cell ( Kawaharada et al . , 2015 ) , as well as an array of bacterial effectors that refine host specificity ( Deakin and Broughton , 2009 ) . Nitrogen fixation however is uncoupled from nodulation and infection , and legumes can be nodulated and infected by ineffective symbiotic partners ( Gehlot et al . , 2013; Gourion et al . , 2015 ) . The emergence of mutualism in populations resulting from the transfer of symbiosis modules , and its maintenance over evolutionary timescales ( Werner et al . , 2014 ) indicates that the cooperative behaviour of the bacterial symbionts is controlled at the infection and/or post-infection levels by one or a combination of mechanisms . Partner choice is the selection of appropriate symbionts at the ( pre- ) infection stage based on signal recognition while post-infection sanctions rely on the ability to discriminate between low- and high-quality cooperators during an established interaction and to punish or reward them accordingly ( Kiers and Denison , 2008; Frederickson , 2013 ) . Partner-fidelity feedback ( PFF ) ensures positive assortment of symbionts during long lasting or repeated interactions in spatially structured environments independently from any recognition process or conditional response ( Sachs et al . , 2004 ) . These different control mechanisms have been proposed to affect the dynamics of mutualistic traits , particularly in the context of the nitrogen-fixing symbiosis ( Kiers et al . , 2003; Oono et al . , 2009 ) . Here we evaluate how selective forces and ecological factors act on the dissemination of the nitrogen fixation mutualistic trait on the Cupriavidus taiwanensis-Mimosa pudica mutualistic interaction . Specifically we evaluated the spatio-temporal dynamics of N2-fixing and non-fixing bacterial subpopulations to model the spread of the N2-fixation trait across plant generations .
During the symbiotic process , most rhizobia enter the legume root via infection threads that ensure colonization of the forming nodule and ultimately release bacteria into nodule cells where differentiated forms called bacteroids fix nitrogen ( Batut et al . , 2004 ) . Although they induce the formation of indeterminate nodules , it is noteworthy that Cupriavidus taiwanensis symbionts of Mimosa spp . are not terminally differentiated and ca . 20% of bacteroids recovered from nodules , together with bacteria present in infection threads , can resume growth ( Marchetti et al . , 2011 ) . To evaluate the specific fates of mutualists and non mutualists in plants infected by a mixed population , we monitored the fitness of total nodule bacteria over time following co-inoculation of Mimosa pudica seedlings with a mixture ( 1/1 ratio , 106 total bacteria/plant ) of isogenic N2-fixing and non-fixing strains of C . taiwanensis . Fix+ and Fix- strains only differed by the presence of the nifH gene , encoding the nitrogenase reductase subunit of the nitrogenase enzyme , and of constitutively expressed GFP or mCherry fluorescent genes . For technical reasons ( see Materials and methods ) , nodules were only collected from 14 dpi . Importantly , each nodule was individually analyzed for bacterial fitness by plating , allowing analysis at the nodule and plant individual levels . In these experimental conditions 97% of the nodules were infected by either Fix+ or Fix- bacteria . We observed a marked difference in the reproductive fitness of Fix+ and Fix- bacteria from the same plant over time , which significantly differed from 21 days post-infection ( dpi ) and up to 28 fold on average ( Figure 1A and Figure 1—figure supplement 1 ) , perhaps because of plant control mechanisms , including sanctions ( Kiers et al . , 2003 ) and possibly PFF . A significant difference was also obtained from 28 dpi when analyzing control plants singly-infected with either Fix+ or Fix- strains ( Figure 2A ) . Non-fixers did not proliferate better than fixers even at 14 dpi ( Figure 1A ) possibly because the metabolic cost paid by bacteria to fix nitrogen in terms of ATP and reducing power is too low to be detected in our experimental conditions , or because plant sanctions/PFF and the metabolic cost of nitrogen fixation equilibrate until sanctions become prominent . The resulting net fitness cost of cooperation , which is the weighted metabolic cost of nitrogen fixation by any form of plant control , thus appeared to be zero or negative , enabling mutualism to spread . The differential fitness was not due to a better nodulation competitveness of Fix+ bacteria . The number of nodules formed by each strain was indeed proportional to the inoculum ratio ( 1/1 ) throughout the time course ( Figure 3 ) , confirming that bacterial nitrogen-fixing ability is not selected at the root entry level ( Hahn and Studer , 1986; Westhoek et al . , 2017 ) . Yet the number of nodules in nitrogen-starved non-fixing plants ( infected with 99% or 100% Fix- ) constantly increased over a 42 day period , while this number reached a plateau at ca . 20 dpi in healthy N2-fixing plants ( infected with 50% or 100% Fix+ ) ( Figure 4 ) , indicative of a mechanism of autoregulation of nodulation acting at the whole-plant level ( Ferguson et al . , 2010 ) and depending on the nitrogen status of the plant ( Malik et al . , 1987; van Noorden et al . , 2016 ) . This difference in time course increases the chance that a rare Fix+ among a Fix- population will form a nodule . To identify the spatial level at which selection applies we first analyzed double occupancy nodules , which were obtained in significant proportion by modifying the plant culture system and increasing the inoculum density by four logs ( see Materials and methods ) . Co-infected nodules contained a similar number of Fix+ and Fix- bacteria at 14 dpi , but on average ca . 80 times more N2-fixing bacteria than non-fixing bacteria at 35 dpi ( Figure 1B ) , indicating that the control occurs at the nodule scale . Previous studies established that bacteroids do not persist in nodule cells of nitrogen-starved plants infected only by non-fixers , leading to premature nodule senescence ( Berrabah et al . , 2015; Hirsch and Smith , 1987 ) , while they persist in healthy plants singly-infected with fixers . We therefore then analyzed the viability of bacteroids on sections of singly-occupied or double-occupied nodules collected from co-inoculation experiments using propidium iodide ( PI ) , which stains dead cells ( Virta et al . , 1998 ) . Bacteroid viability in Fix+-occupied nodules remained stable from 14 to 35 dpi ( Figure 5D ) . By contrast , bacteroids in the nitrogen-fixing zone of Fix--occupied nodules started losing viability at 16–21 dpi and were all dead ( PI-stained ) at 35 dpi ( Figure 5E ) . Electron microscopy confirmed signs of nodule cell and bacterial degeneration in Fix--occupied nodules at 19 dpi ( Figure 6 ) . Co-infected nodules showed clear sectoring , with infected plant cells in one part filled with Fix+ strains and in the other part filled with Fix- strains ( Figure 5FGHI ) . We never observed co-infected nodule cells . While at 14 dpi both strains were alive ( Figure 5G ) , at 35 dpi only Fix- bacteroids were PI-stained confirming that Fix+ and Fix- intracellular bacteria have distinct fates within the same nodule ( Figure 5HI ) . The ca . 5 × 106 bacteria recovered at 35 dpi from nodules infected with only Fix- bacteria may thus be bacteria colonizing the infection threads and the infection zone . In conclusion we provide evidence for differential spatio-temporal dynamics of N2-fixing and non-fixing partners during the symbiotic process , highlighting the importance of considering temporal variations when studying the evolution of cooperative interactions ( Barker and Bronstein , 2016 ) . We established that the control of mutualism ( i ) acts at the nodule cell scale , ( ii ) occurs relatively early , ca . 16–21 days after inoculation when the wild-type nitrogenase is fully active in Fix+ bacteria ( Figure 7 ) and ( iii ) leads to up a ca . 80 fold relative increase in mutualistic partners . Next , we addressed the question of whether mutualism control will allow a minority Fix+ subpopulation to invade the symbiotic population . We first used our experimental data to develop a stochastic mathematical model qualitatively simulating the fate of C . taiwanensis populations during nodulation in M . pudica plants . The two key components of this model are ( i ) the kinetics of nodule formation from bacteria randomly chosen from the rhizospheric population and ( ii ) bacterial multiplication within nodules , according to bacterial genotype ( see Materials and methods and Table 1 for details on model construction and parameterization ) . While the model is developed as a proof-of-concept , instead of a simple deterministic model we chose to include stochasticity in the nodulation process in order to reflect the variability observed in the experimental data . In order to test our model , we first simulated the reproductive fitness of nodule bacteria following single-inoculation with either Fix- or Fix+ bacteria over a 49 day-period , and compared this simulation to the kinetics experimentally observed ( Figure 2BC ) . We then both simulated and experimentally determined the relative proportion of Fix+ bacteria recovered from plants co-inoculated with a minor subpopulation of Fix+ ( 1% ) and a major subpopulation of Fix- ( 99% ) bacteria over 49 days ( Figure 8 ) . Simulation outcomes qualitatively matched the dynamics of bacterial populations observed experimentally ( Figures 2BC and 8 ) , indicating that the experimentally measured and inferred model parameters are appropriate for studying the evolutionary dynamics of C . taiwanensis populations in different ecological conditions . We then used this model to explore how plant population size and the length of inoculation cycles impact on the dynamics of C . taiwanensis populations during serial cycles of inoculation of M . pudica plants and re-isolation of bacteria from nodules . Starting with a fixed proportion of Fix+ bacteria ( 1% or 0 . 1% ) in the inoculum , we varied the number of inoculated plants from 1 to 100 ( or 1 to 1000 ) and the length of nodulation cycles ( time from plant inoculation to nodule bacteria harvesting ) from 14 to 49 days , which is shorter than the lifespan of a nodule in nature . We found that larger plant pools and longer cycles progressively reduced extinction probabilities and increased the proportion of Fix+ in the nodule bacterial population ( Figure 9A and Figure 9—figure supplement 1 ) . For example , the model predicted that using an initial inoculum of 1% Fix+ , 4 cycles of 42 days with pools of 20 plants were sufficient to yield more than 85% of Fix+ bacteria in all replicates where Fix+ populations avoided extinction ( 89 times out of 100 replicates in Figure 9A ) . Smaller plant pools or shorter cycles all yielded higher probabilities of extinction and decreased proportions of Fix+ bacteria . An initially lower Fix+ proportion ( 0 . 1% ) could be compensated for by a higher plant population size and/or a longer cycle length ( Figure 9—figure supplement 1 ) . We analyzed in detail the dynamics of Fix+ subpopulations over 10 cycles in a situation where the cycle length had a major impact on the evolutionary outcome ( 20 plants ) ( Figure 9A ) and plotted the proportion of Fix+ bacteria recovered after each cycle , for cycles ranging from 14 to 49 days ( Figure 9B ) . We observed that , in the vast majority of cases , the fate of Fix+ populations is already determined after the first cycle: these populations are either bound to extinction ( with a probability indicated in Figure 9A ) or to a gradual increase in frequency that ultimately leads to fixation . This result holds true for all cycle lengths except 14 days , where population dynamics is dominated by drift due to the equivalent fitness of Fix- and Fix+ clones ( Figure 1A ) . A key factor controlling the early bifurcation between extinction and fixation of Fix+ population is the probability that a Fix+ bacterium forms a nodule during the first cycle , which depends on both the size of plant pools and the length of nodulation cycles . Understanding the influence of plant pool size is straightforward . Very few nodules are produced on each plant , creating a bottleneck in bacterial population size at each nodulation cycle . Whatever the cycle length , larger numbers of plants per pool increase the likelihood that at least one Fix+ clone is sampled from the rhizospheric population , giving Fix+ subpopulations an opportunity to increase in frequency and avoid extinction in the next cycle ( Figure 9A and Figure 9—figure supplement 1 ) . Under longer cycles , extinction probability decreases ( Figure 9AB ) since more nodules are produced ( Figure 4 ) and the size of Fix+ populations increases at a faster rate ( Figure 9B ) as a result of a decrease in Fix- fitness in older nodules ( Figure 1A ) . The combined action of these two factors act on the inoculum for next cycle , generating an eco-evolutionary feedback . To assess the predictions of the model experimentally , we performed serial inoculation-nodulation cycles of 21 or 35 days using 20 M . pudica plants and an initial inoculum of 5 × 103 Fix+/5 × 105 Fix- C . taiwanensis per plant . In each 35 day-cycle the nitrogen-fixing subpopulation increased and it reached nearly 100% of the population after four cycles ( Figure 10A ) , similar to what observed with the model . Under 21 day-cycles , both simulations and experiments lead to a slower progression of Fix+ subpopulations ( Figure 10B ) . It is worth noting that an increase in frequency of the best cooperators among natural strains was also observed after three consecutive nodulation cycles between Medicago truncatula and Sinorhizobium meliloti ( Heath and Tiffin , 2009 ) , indicating that the selective advantage of the best N2-fixing strains seems to be robust to the natural diversity of symbiotic associations .
Identifying the selective forces and ecological factors that shape mutualism is central to predicting its maintenance and dissemination over evolutionary scales . Here we provide conclusive evidence that nitrogen fixation per se , the ultimate trait that turns a parasitic rhizobium-legume association into a mutualistic one , determines the in planta spatio-temporal fate of endosymbiotic bacteria . Non-N2-fixing symbionts do not persist within cells of indeterminate nodules even when they share a nodule with N2-fixing symbionts , indicative of a cell autonomous senescence program as recently shown for determinate nodules ( Regus et al . , 2017 ) . This results in the progressive and selective in planta expansion of fixers during the symbiotic process . The most likely explanation is that the plant exerts a post-infection control of N2-fixation that overcomes the metabolic cost of nitrogen fixation paid by mutualistic bacteria . Sanctions could occur as defense responses and/or by decreasing nutrient supply to non-fixing bacteroids . Given that Fix- and Fix+ bacteria are spatially segregated within nodules , the latter case could also result from the local degeneration of nodule cells , and be interpreted as an example of Partner Fidelity-Feedback mechanism occurring at the level of individual cells ( Shou , 2015 ) . Since control mechanisms prevent social dilemma –i . e . the possibility that one partner increases its own fitness by decreasing its investment in mutualism- and help cooperation persist ( Kiers and Denison , 2008; Frederickson , 2013; Sachs et al . , 2004 ) , non-fixers do not threaten mutualism in our system . Yet the fate of strains able to fix intermediate levels of nitrogen fixation may be different . Monitoring the fitness of strains varying in their nitrogen fixation capacity would provide a more complete picture of mutualism control . Nevertheless , our results provide an additional example supporting the emerging idea that low quality rhizobial partners rarely benefit from low investment in mutualism ( Jones et al . , 2015; Friesen , 2012 ) . Plant sanctions resulting in bacterial fitness reduction were demonstrated in some rhizobium-legume systems by simulating N2 deficiency via gas manipulation around nodules ( Kiers et al . , 2003; Oono et al . , 2011 ) , although not seen in other systems ( Marco et al . , 2009; Ling et al . , 2013 ) . That different plants may rely on different control mechanisms would not be surprising given the variety of mechanisms that lead to symbiosis with legumes ( Masson-Boivin et al . , 2009 ) . Experimental investigations can fuel a theoretical framework able to reframe general evolutionary questions in an ecological context ( Hoek et al . , 2016 ) . Our qualitative model of the eco-evolutionary dynamics of mutualistic and non-mutualistic populations includes serial inoculation-nodulation cycles . This regime mimics an experimental set up of horizontal transmission of rhizobia across plant generations albeit on an accelerated basis . A general outcome of the model is that rare fixers will invade a population dominated by non-fixing bacteria , above a threshold combination of plant and bacterial population sizes and cycle lengths . The model helps explore further combinations of number of cycles , cycle lengths and plant pool sizes to hypothesize the evolutionary trajectory of the Fix+ genotype . While the selective advantage of the Fix+ phenotype is expected to ensure its fixation in a deterministic manner , strong population bottlenecks occurring at the nodulation step introduce a source of stochasticity in these dynamics and may thus prevent the action of directional selection . The effect of stochasticity has been shown to be of immense evolutionary consequence in related models of host parasite coevolution ( Papkou et al . , 2016 ) . Another characteristic of our system is that , when the Fix+ populations increase in abundance then so does their proliferation , leading to a quick increase of Fix+ over successive nodulation cycles ( Figure 9B ) . This interaction between the demographic composition of the population and the evolutionary success of one of the traits is an example of the eco-evolutionary feedback present in this system . Although the selective and ecological forces at play in the lab and in field conditions may differ significantly , our results predict that both forces have played a major role in the evolution of the rhizobium-legume mutualism by favoring the fixation of emerging N2-fixing sub-populations among uncooperative symbiotic populations as well as their evolutionary maintenance . Yet the uncooperative population does not become extinct within nodules , likely because sanctions mainly target bacteroids of the nitrogen fixation zone . Releasing non-fixing bacteria may allow progenitors to meet appropriate hosts or to evolve new symbiotic traits . This loose selection process helps maintain genetically diverse rhizobial communities in the soil and shape the ecology and evolution of rhizobia . More generally , acknowledging the existence of non-cooperators as an integral component of the ecological and evolutionary dynamics of mutualistic interactions may provide a better understanding of the long-term persistence of bacterial lineages ( Heath and Tiffin , 2009; Heath and Stinchcombe , 2014; Tarnita , 2017; Fiegna et al . , 2006; Hammerschmidt et al . , 2014 ) . An emerging trend in fundamental and applied plant microbiology is to select upon microbes indirectly through the host ( Mueller and Sachs , 2015 ) . This engineering approach , called host-mediated selection , involves selection of microbial traits that are not selectable in vitro . Modelling the eco-evolutionary scenarios provides predictions to guide experimental evolution studies aiming at designing beneficial microbes ( Marchetti et al . , 2010; Marchetti et al . , 2017 ) and microbiomes ( Mueller and Sachs , 2015; Johns et al . , 2016 ) .
Strains and plasmids used in this study are listed in Table 2 . C . taiwanensis strains were grown at 28°C on TY medium supplemented with 6 mM CaCl2 and 200 µg/ml streptomycin . E . coli strains were grown at 37°C on LB medium and antibiotics were used at the following concentrations: kanamycin 25 µg/ml , trimethoprim 100 µg/ml , tetracycline 10 µg/ml . For in vitro competition experiments , strains were pre-cultured in TY medium , mixed in equal proportion then co-inoculated to a 100 ml culture in TY medium . Bacteria were plated every 2 hr during the exponential phase , at the entry of stationary phase and 15 hr after the entry into the stationary phase . Plated bacteria were grown for 48 hr at 28°C then green and red bacteria were counted using a fluorescence stereo zoom microscope ( Axiozoom V16 , Zeiss ) . Mutant and labeled strains of C . taiwanensis were constructed using the mutagenesis system developed by Flannagan et al . ( Flannagan et al . , 2008 ) involving the suicide plasmid pGPI-SceI carrying an I-SceI recognition site and the pDAI-SceI replicative plasmid expressing the I-SceI nuclease . To construct the unmarked C . taiwanensis nifH mutant , regions upstream and downstream nifH were amplified with the oCBM1821-oCBM2362 and oCBM1822-oCBM2363 primer pairs using GoTaq DNA polymerase ( Promega ) . The two PCR products were digested with XbaI-BamHI and BamHI-EcoRI respectively and cloned into the pGPI-SceI plasmid digested by XbaI and EcoRI . Ligation products were transformed into a DH5α λpir E . coli strain . The resulting plasmid was transferred into C . taiwanensis CBM832 by triparental mating using pRK2013 as helper plasmid . Transconjugants that have integrated the plasmid by single crossing over were selected on streptomycin and trimethoprim and verified by PCR using the oCBM1824-oCBM2363 and oCBM1825-oCBM2362 primer pairs . Then we introduced the pDAI-SceI replicative plasmid into these strains by conjugation and selection on tetracyclin . Expression of the I-SceI nuclease causes a double strand break into the inserted plasmid and promotes DNA recombination . Mutants deleted in nifH were screened by trimethoprim sensitivity and verified by PCR using the oCBM1824-oCBM1825 pair of primers . Mutants were then cultivated on unselective TY medium . Tetracycline sensitive colonies which have lost the pDAI-SceI plasmid were selected . The Pps-GFP and Pps-mCherry fusions were inserted into the wild-type and nifH mutant of C . taiwanensis at the same chromosomal locus , i . e . in the intergenic region between the glmS and RALTA_A0206 genes using the same pGPI-SceI/pDAI-SceI mutagenesis system . Flanking regions of the insertion site were amplified by PCR using the Phusion DNA polymerase ( ThermoFisher Scientific ) and the oCBM2619-oCBM2620 and oCBM2621-oCBM2622 primer pairs . PCR products were digested by XbaI and Acc65I or Acc65I and EcoRI respectively and cloned into the pGPI-SceI plasmid digested by XbaI and EcoRI . The two fusions Pps-GFP and Pps-mCherry were obtained by digesting the pRCK-Pps-GFP and pRCK-Pps-mCherry by AvrII and SpeI and cloned into the pGPI-SceI carrying the intergenic region glmS-RALTA_A0206 digested by the same enzymes . The resulting pCBM161 and pCBM162 were first transformed into a DH5α λpir E . coli strain then transferred into C . taiwanensis by triparental mating with the pRK2013 helper plasmid . Integration of the fusions by double crossing over was carried out using the pDAI-SceI plasmid as described above . CBM2700 ( Fix+ , GFP ) and CBM2707 ( Fix- , mCherry ) had the same plating efficiency in in vitro competition experiments , indicating that these genetic modifications did not noticeably affect bacterial growth rate . Oligonucleotide sequences used for genetic constructions are provided in Supplementary file 1 . Mimosa pudica seeds were of Australian origin ( B and T World Seed , Paguignan , France ) and were sterilized as described ( Chen et al . , 2003 ) . Seedlings were cultivated in Gibson tubes ( 2 M . pudica plantlets/tube ) as previously described ( Marchetti et al . , 2014 ) . To increase the frequency of co-infection , plants were grown on 12 cm2 plates ( three plants per plate ) containing slanting nitrogen-free Fahraeus agar medium for 3 days at 28°C . Roots were covered with a sterile , gas-permeable , and transparent plastic film ( BioFolie 25; Sartorius AG , Vivascience , Bedminster , NJ , U . S . A . ) . For single-strain inoculation experiments , each plant in Gibson tubes was inoculated with 5 . 105 bacteria either CBM832 ( wild-type ) or its isogenic nifH mutant , CBM2568 . For co-inoculation experiments in Gibson tubes , plants were inoculated with the two isogenic strains CBM2700 ( wild-type , GFP labeled ) and CBM2707 ( nifH , mCherry labeled ) at ratio 1/1 ( 5 . 105 bacteria of each strain per plant ) or 1/100 ( 5 . 103 bacteria of CBM2700 and 5 . 105 bacteria of CBM2707 per plant ) . For co-inoculation experiments in plates , plants were inoculated with 1010 bacteria of each strain per plant . To measure the number of nodule bacteria over time , all nodules from 5 to 10 individual plants , except very small nodules , were individually collected with at least 2 mm of root left on both sides of nodules and treated at each time point . We did not collect very small nodules since there was a risk that the sterilization agents penetrate these nodules . In the same line we did not collect nodules before 14 dpi since most nodules were very small at that stage . Nodules were surface sterilized for 15 min in a 2 . 5% sodium hypochlorite solution , rinsed with water and crushed . Each nodule crush was diluted and plated using an easy spiral automatic plater ( Interscience ) . Colonies were counted after 2 day-incubation at 28°C , under a fluorescence stereo zoom microscope ( Axiozoom V16 , Zeiss ) when appropriate . For nodulation kinetics , the number of nodules formed on 20 plants grown in Gibson tubes was counted daily for 6 weeks . For serial inoculation-nodulation cycles on M . pudica plants , 10 Gibson tubes of plants were inoculated with CBM2700 and CBM2707 in 1/100 ratio as described above . 35 days after inoculation , all nodules were collected , surface-sterilized and crushed together . The nodule crush was used to inoculate a new set of 10 tubes of plants with 50 µl of a 1/10 dilution of the nodule crush per plant . At each cycle , dilutions of the nodule crush were spread on plates , incubated 2 days at 28°C and colonies were counted under a fluorescence stereo zoom microscope . The viability of nodule bacteria was estimated using propidium iodide staining at a concentration of 20 mM in DMSO ( Molecular Probes , Fisher scientific , Oregon ) on 55/58 µm nodule sections . For each experiment , a dozen nodules were individually analyzed at 14 , 16 , 17 , 21 , 28 and 35 dpi . For electron microscopy analysis , nodules were fixed in glutaraldehyde ( 2 . 5% in phosphate buffer 0 . 1 M [pH 7 . 4] ) , osmium treated , dehydrated in an alcohol series , and embedded in Epon 812 . Semithin nodule sections were observed by brightfield microscopy after staining in 0 . 1% aqueous toluidine blue solution and observed under a Zeiss Axiophot light microscope . Ultrathin sections were stained with uranyl acetate and observed with a TEM Hitachi HT7700 . M . pudica plants were inoculated with the wild-type strain of C . taiwanensis CBM832 . At different time points , plants were removed from the culture Gibson tube and placed in an airtight tube and incubated with 1 ml of acetylene for 4 hr . 100 µl of gas were then injected into a gas chromatograph ( Agilent GC7820 ) . The area of the ethylene peak was measured and compared to an ethylene standard of known concentration . Ethylene background was estimated by analyzing empty tubes incubated with the same amount of acetylene . The model aimed at simulating nodulation dynamics during single or repeated inoculation-nodulation cycles . First we parameterized the population dynamics during the symbiosis process . Then we simulated repeated nodulation cycles varying the following parameters: ( i ) the Fix+/Fix- ratio in the initial inoculum , ( ii ) the number of inoculated plants , and ( iii ) the cycle length . The model ran on a pool of plants ( of given , variable size ) from which nodules were collected and mixed together after each inoculation cycle . For each time-step ( 1 day ) after inoculation , the number of new nodules formed on each plant was randomly drawn from a Poisson distribution of parameter λ ( t , nod+t ) , which is itself a function of time t and of the number of nodules already present on the plant nod+t at time t . The maximal number of nodules that could potentially be formed per day per plant was set to λmax . Changing the value of parameter λ depending on the number of Fix+nodules already present on the plant simulated the autoregulation of nodulation process; this was done by subtracting the factor a1 × nod+t from λmax . Lastly , to allow for some ‘aging’ process that would decrease the rate of nodulation with time ( even for plants inoculated only with Fix- bacteria ) , we incorporated a time-decay coefficient: a2 × ( t- a3 ) , meaning that a reduction in the rate of nodulation occurred at a rate a2 when t > a3 . This time-decay factor was set to 0 when t < a3 . Therefore , the parameter of the Poisson distribution controlling the rate at which new nodules are formed was given by: λ ( t , nod+t ) = λmax - a1 × nod+t for t < a3 and by: λ ( t , nod+t ) = λmax - a1 × nod+t - a2 × ( t- a3 ) for t > a3 . Since nodules are persistent once formed , we further set: λ ( t , nod+t ) ≥0 . Experimental evidence indicated that the number of inoculated bacteria did not affect nodulation kinetics as long as the total inoculum remains above 103 bacteria per plant . These conditions were met in all experiments described in this work . Therefore , we did not explicitly take inoculum size into consideration in the simulations , and restricted the applicability of our model to cases where inoculum was above this threshold value . The second module of the model dealt with bacterial multiplication within plant nodules . Within each nodule we assumed a logistic growth model for the bacteria given by: X ( t + 1 ) = ( r-c –suds ( t ) ) ×X ( t ) × ( 1-X ( t ) /K ) , where r was the growth rate , c the net fitness cost of nitrogen fixation in Fix+ bacteria , suds ( t ) the additional plant sanctions against Fix- bacteria occurring in the later phase of the interaction , X ( t ) the bacterial population at time t and K the nodule carrying capacity . In our simulations , we set c = 0 since we experimentally did not detect any difference in the populations of Fix- or Fix+ nodule bacteria at 14 dpi . We emphasize that a net fitness cost of 0 does not necessarily imply that nitrogen fixation does not impose a metabolic burden on the bacteria ( referred to as ‘metabolic cost’ in the results section ) . Instead , this burden , if significant during the early steps of the interaction , may be compensated for by plant control mechanisms acting at a basal level . Beyond this time point , additional plant sanctions ( possibly including partner fidelity-feedback ) were given by suds ( t ) , taking the value s of plant sanctions indicated in Table 1 as long as the age of the nodule was higher than ds days ( denoted by the step function uds ( t ) =0 if t < ds or uds ( t ) =1 if t > ds ) . Parameters values were estimated by computing the minimal root mean square error ( RMSE ) of experimental data ( nodulation kinetics and bacterial multiplication within nodules ) versus model outputs calculated for a range of parameter values . Parameter values selected to minimize RMSE are indicated in Table 1 . Simulations were implemented in R ( R Core Team , 2014 ) and code is available in the Source code file 1 . | Rhizobia are soil bacteria that are able to form a symbiotic relationship with legumes – plants that include peas , beans and lentils . The bacteria move into cells in the roots of the plant and cause new organs called nodules to form . Inside the nodules the bacteria multiply before being released to the soil again . Also while in the nodules , the bacteria receive carbon-containing compounds from the plant . In return many of the bacteria convert ( or “fix” ) nitrogen from the air into compounds that the plant can use to build molecules such as DNA and proteins . Yet , some of the bacteria are “non-fixers” that provide little or no benefit to the host plant . Evidence suggests that legumes select against non-fixer bacteria , though it was not clear when or how this selection process occurs . Daubech , Remigi et al . have now followed the number and viability of two variants of a bacteria species called Cupriavidus taiwanensis as they form a symbiotic interaction with Mimosa pudica , a member of the pea family . The two types of bacteria differed only by whether or not they were able to fix nitrogen . At first fixers and non-fixers entered nodules and multiplied at equal rates . Later , the fixers progressively outcompeted the non-fixers . Then , around 20 days after the bacteria entered the plant , nodule cells that contained non-fixers degenerated . This indicates that the nodule cells help to control bacterial proliferation based on the benefits they receive in return . Further experiments and mathematical modeling also showed that over repeated cycles of root nodule formation , nitrogen fixers can invade a bacterial population dominated by non-fixer bacteria . The likelihood that this invasion will be successful increases as three other factors increase: the proportion of fixer bacteria in the initial population , the number of available plants , and the length of time the bacteria spend in the nodules . This mechanism ensures the maintenance and spread of nitrogen-fixing traits in the bacterial population . Improving the processes of biological nitrogen fixation could help to reduce the amount of fertilizers required to grow crops . This in the future could help make agricultural ecosystems more sustainable . The results presented by Daubech , Remigi et al . provide guidelines that could be used to select nitrogen-fixing bacteria on legume crops or on nitrogen-fixing cereals that may be engineered in the future . Further work is now needed to understand in more detail the molecular mechanisms that lead to the death of non-fixer bacteria . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"evolutionary",
"biology"
] | 2017 | Spatio-temporal control of mutualism in legumes helps spread symbiotic nitrogen fixation |
Understanding how patterning influences cell behaviors to generate three dimensional morphologies is a central goal of developmental biology . Additionally , comparing these regulatory mechanisms among morphologically diverse tissues allows for rigorous testing of evolutionary hypotheses . Zebrafish skin is endowed with a coat of precisely patterned bony scales . We use in-toto live imaging during scale development and manipulations of cell signaling activity to elucidate core features of scale patterning and morphogenesis . These analyses show that scale development requires the concerted activity of Wnt/β-catenin , Ectodysplasin ( Eda ) and Fibroblast growth factor ( Fgf ) signaling . This regulatory module coordinates Hedgehog ( HH ) dependent collective cell migration during epidermal invagination , a cell behavior not previously implicated in skin appendage morphogenesis . Our analyses demonstrate the utility of zebrafish scale development as a tractable system in which to elucidate mechanisms of developmental patterning and morphogenesis , and suggest a single , ancient origin of skin appendage patterning mechanisms in vertebrates .
Developmental patterning generates distinct gene expression regimes that regulate morphogenetic cell behaviors . Identifying core regulatory modules , elucidating the specific interactions they comprise , and how these activities are translated into discrete morphological outcomes are central goals of modern developmental biology . To these ends , considerable progress has been made at embryonic stages , yet patterning and morphogenesis at post-embryonic stages remain poorly understood . Skin appendages are classic developmental model systems that have been leveraged to generate insights into how mesenchymal–epithelial signaling interactions pattern tissues and affect morphogenesis ( Lai and Chuong , 2016 ) . These structures , including the hair , teeth , mammary and eccrine glands of mammals , feathers and scales of birds , and scales or scutes of reptiles are among the most conspicuous features of the adult form . In addition to being developmental biology models with clear biomedical relevance , skin appendages have been of longstanding interest for their evolutionary significance . Although amniote skin appendages such as hairs and feathers have widely varied morphologies , recent work suggests that all of these appendages derive from a common progenitor that was present in stem amniotes ( Di-Poï and Milinkovitch , 2016; Wu et al . , 2018 ) . Another skin appendage is the scale of fishes . In many extant teleosts , scales comprise thin , overlapping plates of dentin-like calcified extracellular matrix embedded in the dermis ( Sire et al . , 2009 ) . Similar to amniote skin appendages , fish scales develop relatively late in ontogeny and are distributed across the skin in a tight , hexagonal grid pattern in the adult . Scales have been the object of excellent histological and ultrastructural studies that have characterized developmental anatomy , and genetic analyses that have identified phenotypes associated with Ectosyslasin-A ( Eda ) and Fgf mutations ( Harris et al . , 2008; Rohner et al . , 2009; Sire et al . , 1997 ) . Nevertheless , these serially reiterated , highly accessible organs have yet to be exploited as a model for understanding skin appendage development at a cellular level . There is also uncertainty as to whether fish scales and amniote skin appendages are homologous , that is , derived from a single archetype organ in a common ancestor ( Paul , 1972; Sharpe , 2001 ) . If all vertebrate skin appendages are homologous , we would expect that common developmental regulatory mechanisms underlie teleost scales , avian feathers , mammalian hair and other appendage types . Here , we use conditional-genetic manipulations , live imaging and gene expression assays optimized for post-embryonic fish to show that multiple signaling pathways , including Wnt/β-catenin , Eda , Fgf and Shh regulate scale patterning and morphogenesis . These analyses show that scale development relies on signaling interactions similar to interactions that regulate the patterning and morphogenesis of amniote skin appendages such as hair and feathers , and support a model in which diverse skin appendages of vertebrates arose from a common archetype . Additionally , we uncover a novel process of HH-dependent collective cell migration that is necessary for epidermal invagination during skin appendage morphogenesis . Together , our analyses establish the fundamental parameters that govern scale initiation and morphogenesis and lay the groundwork for exploiting zebrafish scale development as a system in which to discern general principles of developmental patterning , regulation of morphogenetic cell behaviors , and the evolution of genetic regulatory mechanisms .
Histological , ultrastructural and fate mapping studies showed that scale morphogenesis begins with the formation of a bi-layered dermal papilla immediately beneath the epidermis ( Mongera and Nüsslein-Volhard , 2013; Shimada et al . , 2013; Sire et al . , 1997 ) . Yet these studies did not unambiguously identify the cell types involved . We hypothesized that , due to their calcified composition , scales will be formed by osteoblast-like cells . To test this , we analyzed the distribution of osteoblasts using fish transgenic for a previously described reporter with osteoblast-specific expression , sp7:EGFP ( DeLaurier et al . , 2010 ) . sp7 ( formerly osterix ) encodes a zinc finger transcription factor that is necessary and sufficient for osteoblast differentiation from committed progenitors ( Zhang , 2012 ) . Simultaneously , we monitored the distribution of calcified extracellular matrix using Alizarin Red S ( ARS ) vital dye ( Adkins , 1965; DeLaurier et al . , 2010 ) . Weak sp7:EGFP expression was first detected in clusters of cells in the skin , immediately followed by detection of calcified matrix ( Figure 1A d1–d2 and Figure 1—figure supplement 1A and B ) . Throughout subsequent scale development , the distribution of calcified matrix correlated with the distribution of sp7:EGFP+ cells ( Figure 1A , d2–d6; Figure 1—figure supplement 1B ) suggesting that sp7:EGFP labeled the dermal cell population that deposits calcified extracellular matrix comprising the scale plate . Hereafter , we refer to this cell population as dermal Scale Forming Cells ( dSFCs ) . Early sp7:EGFP+ dSFC clusters that lacked detectable calcified matrix showed a bi-layer papilla corresponding with structures described in histological and ultrastructural studies ( Figure 1—figure supplement 1C ) . Therefore , sp7:EGFP labels developing scales from very early in morphogenesis and likely earlier than other scale osteoblast markers ( Iwasaki et al . , 2018 ) . Following their initial appearance , scales grew centripetally from the scale focus with a posterior bias leading to polarized extension through the addition of dSFCs and calcified matrix ( Figure 1—figure supplement 1D , E ) . In fully formed scales , sp7:EGFP+ dSFCs were arranged in a monolayer along the deep aspect of the calcified matrix but also looped around to cover the superficial surface , overlapping with the intensely ARS-labeled limiting layer ( Figure 1A , Figure 1—figure supplement 1F–H ) . Skin appendage primordia in amniotes spread sequentially across the skin during development ( Chuong et al . , 2013; Dalle Nogare and Chitnis , 2017; Painter et al . , 2012; Sick et al . , 2006 ) . In-toto repeated live imaging of individual zebrafish revealed similar sequential addition of scale primordia ( Figure 1B–1F , Video 1 ) . The first scales were found in a row on the caudal peduncle followed by a second row that formed on the flank above the ribs ( Figure 1—figure supplement 2A , B ) . Within one day , two complete rows of scales formed ( Figure 1B ) . Additional rows were added dorsally and ventrally , and columns were added anteriorly and posteriorly ( Figure 1B–1D; Figure 1—figure supplement 2F–I ) . Finally , a third scale origin appeared just posterior to the mandible and spread along the ventral surface to meet the ventral row scales just posterior to the pectoral fin insertion ( Figure 1E; Figure 1—figure supplement 2J ) . These events lead to a complete coat of scales arranged in a half-offset , hexagonal grid on the juvenile fish ( Figure 1F ) . Under optimized conditions , this process took ~13 days ( Figure 1—figure supplement 2C–E ) . Identical scale patterning can be visualized using sequential calcium vital dye labelling ( Figure 1—figure supplement 2K–O ) . There are presently no staging conventions for squamation in zebrafish . Based on our live imaging , we propose a staging system for this process . Two rows ( 2R ) represents the initial appearance of scales up to two complete scale rows . Four rows ( 4R ) are fish with four complete scale rows , one dorsal and one ventral to the original two scale rows . This stage corresponds with the stage of grossly apparent posterior squamation ( SP ) defined in ( Parichy et al . , 2009 ) . Finally , beard ( B ) stage fish have scales along the dorsal anterior , as well as the initiation of post-mandibular scales , corresponding with the anterior squamation ( SA ) stage defined in ( Parichy et al . , 2009 ) . Our observations of scale development indicate that sequential addition of skin appendage primordia , leading to a tightly packed hexagonal grid , is a feature of epidermal appendage development common to both amniotes and anamniote fishes . In turn , this suggests the hypothesis that these diverse appendage types are patterned by common mechanisms . To evaluate this hypothesis we next tested requirements for specific signaling activities during scale development . If amniote skin appendages and teleost scales arose from a common ancestral organ , we predicted that Wnt/β-catenin signaling should be necessary for scale development as it is for ectodermal appendages of amniotes ( Andl et al . , 2002; Dhouailly et al . , 2017 ) . To inhibit Wnt/β-catenin signaling during scale development , we used the hs:dkk1 transgenic line that allows conditional expression of a potent and selective Wnt/β-catenin signaling inhibitor ( Glinka et al . , 1998; Stoick-Cooper et al . , 2007 ) . As predicted , inhibiting Wnt/β-catenin beginning prior to the appearance of sp7:EGFP+ papillae prevented scale formation ( Figure 2A and C ) . This early treatment also blocked localized expression of the Wnt/β-catenin signaling activity reporter 7xTCF:mCherry ( Figure 2B ) ( Moro et al . , 2012 ) . Beyond scale phenotypes , Wnt/β-catenin inhibition prevented fin outgrowth and formation of pelvic fin rays ( Figure 2A and D ) , demonstrating a role for this pathway in fin development , in addition to previously documented functions in fin regeneration ( Kawakami et al . , 2006; Stoick-Cooper et al . , 2007; Wehner et al . , 2014 ) . Scale and fin phenotypes were not caused by a generalized retardation of development , as Dkk1-overexpressing fish achieved sizes similar to those of non-transgenic , heat-shocked sibling controls ( Figure 2E ) . If Wnt/β-catenin signaling is necessary for initiating scale development , signaling activity must precede overt scale morphogenesis . Live imaging of fish doubly transgenic for 7xTCF:mCherry and sp7:EGFP showed that , as predicted , Wnt/β-catenin signaling activity was evident before sp7:EGFP+ scale papilla had formed ( Figure 2F ) . After papilla formation , Wnt/β-catenin signaling became polarized toward the posterior scale margin , where it remained throughout scale growth . Because scales develop in an invariant sequence we were also able to analyze expression of conserved Wnt/β-catenin targets lef1 and axin2 in fixed specimens ( Cadigan and Waterman , 2012; Hovanes et al . , 2001; Jho et al . , 2002; Ramakrishnan and Cadigan , 2017 ) . In agreement with the 7xTCF:mCherry reporter , lef1 and axin2 , as well as nuclear localized β-catenin , were found in skin patches presaging the appearance of scales ( Figure 2—figure supplement 1A , B , pre scale and papilla; Figure 2—figure supplement 1E , F ) . Expression subsequently became restricted to a posteriorly biased ring around the circumference of the developing scale ( Figure 2—figure supplement 1A , B , extension; Figure 2—figure supplement 1E , G , H ) . Although expression dynamics of these Wnt/β-catenin activity markers were broadly similar , axin2 was expressed exclusively in dermal cells and was absent from the epidermis , whereas lef1 was expressed in both dermis and epidermis ( Figure 2—figure supplement 1A , B ) . As expected for Wnt target genes , early Dkk1 induction prevented patterned expression lef1 and axin2 ( Figure 2—figure supplement 1C , D , early hs:dkk1 ) . However , Dkk1-mediated Wnt repression initiated after scales had formed attenuated expression of axin2 , but not lef1 ( Figure 2—figure supplement 1C , D , late hs:dkk1 ) , suggesting a mechanism independent of Wnt/β-catenin signaling for maintaining lef1 expression during scale outgrowth . Taken together , expression of Wnt/β-catenin target genes prior to papilla appearance and lack of scales in Dkk1 overexpressing fish demonstrates that Wnt/β-catenin signaling is necessary for initiation of scale development . We next sought to address how Wnt/β-catenin signaling interacts with other signaling pathways during scale development . The phenotypes of fish inhibited for Wnt/β-catenin signaling—abrogated scale and fin formation —were similar to phenotypes of fish lacking Ectodysplasin-A ( Eda ) signaling ( Harris et al . , 2008; Kondo et al . , 2001 ) . We therefore hypothesized that Wnt/β-catenin and Eda signaling interact during scale development . To elucidate regulatory linkages between these pathways we used the zebrafish nkt mutant line , harboring a presumptive eda loss of function allele ( Harris et al . , 2008 ) . nkt mutants lacked overt indications of scale morphogenesis and sp7:EGFP+ dSFCs in the skin , in addition to having impaired fin outgrowth ( Figure 3—figure supplement 1A ) . Yet , live imaging of sp7:EGFP; 7xTCF:mCherry transgene expression and in situ hybridization against lef1 showed that Wnt/β-catenin activity was patterned into spots in the epidermis of scale-less nkt mutants ( Figure 3A–3D ) . Wnt/β-catenin targets were initially expressed in scale-appropriate patterns in nkt mutants , yet further rows were not formed and expression domains did not polarize . These findings indicate that initiation and patterning of Wnt activity in the skin is Eda-independent , whereas later refinement and reiteration of Wnt activity requires Eda signaling . The absence of overt scale development despite initiation of Wnt signaling , also demonstrates that Wnt signaling alone is not sufficient to induce scales in the absence of Eda . Conversely , we asked whether Eda signaling requires Wnt/β-catenin activity . We first examined the expression of genes encoding Eda and its receptor Edar during normal scale development . Immediately prior to scale formation , eda expression disappeared from epidermal cells above forming scale papillae where edar expression was first detected ( Figure 3—figure supplement 1B , C pre-scale ) . Subsequently , eda was expressed broadly in dSFCs whereas edar expression became localized principally to the posterior margin of the scale epidermis ( Figure 3—figure supplement 1B , C papilla , extension ) in the vicinity of Wnt/β-catenin target genes ( Figure 2F and Figure 2—figure supplement 1 ) . These similar expression patterns ( and common phenotypes of pathway blockade ) suggested the hypothesis that edar is a Wnt/β-catenin signaling target during scale formation , whereas eda expression might be independent of Wnt/β-catenin . We therefore induced hs:dkk1 following the appearance of initial scales ( 2R stage ) and assayed expression of eda and edar . As predicted , edar expression was strongly attenuated in Dkk1-overexpressing , Wnt/β-catenin inhibited fish , whereas eda expression persisted ( Figure 3E , F ) . Lack of edar expression in Wnt/β-catenin inhibited fish , taken together with residual Wnt activity in nkt mutants that nevertheless formed no scales , suggested that a lack of Eda pathway signaling is the primary reason that scales do not develop in Wnt inhibited fish . To test this , we conditionally misexpressed Eda and Edar while manipulating Wnt/β-catenin signaling . In fish with normal Wnt/β-catenin signaling , misexpression of either Eda or Edar initiated scale development , as revealed by accumulations of sp7:EGFP+ dSFCs in regions where scales do not normally form at these stages ( Figure 3G , H , control ) . Misexpression of Eda in either epidermis or dermis initiated ectopic scale development ( Figure 3—figure supplement 2A , B ) , whereas Edar did so only when misexpressed in the epidermis ( Figure 3—figure supplement 2C , D ) , suggesting that competence to respond to Eda signaling is unique to epidermis . However , Eda/Edar driven ectopic scale induction was completely blocked in the absence of Wnt/β-catenin signaling ( Figure 3G–3H , hs:dkk1 ) . Therefore , neither Wnt/β-catenin nor Eda signaling is sufficient to trigger scale development in the absence of the other . Fgf signaling has been implicated in skin appendage development of amniotes ( Huh et al . , 2013; Mandler and Neubüser , 2004; Petiot et al . , 2003 ) and in scale development of teleosts: zebrafish harboring mutations simultaneously in fgfr1a and fgf20a have scales that are smaller than normal , whereas fish mutant for fgfr1a alone have scales that are larger than normal . It remains unclear whether Fgf signaling is required for scale initiation , outgrowth or both ( Daane et al . , 2016; Rohner et al . , 2009 ) . To circumvent potential functional redundancies and to test the necessity of Fgf signaling for scale development , we employed the pan-Fgf receptor inhibitor BGJ398 that has been shown to specifically block activity of Fgf receptor kinase but not other closely related kinases ( Guagnano et al . , 2011 ) . This treatment led to an immediate and complete arrest of squamation and scale growth without affecting overall somatic growth of the fish ( Figure 4A , B; Figure 4—figure supplement 1A–C ) , demonstrating that Fgf signaling is necessary for both scale initiation and outgrowth . To determine which Fgf receptors are involved in scale development , we generated cDNA from 4R stage skin and assayed by RT-PCR the expression of each of the five Fgf receptor genes present in the zebrafish genome . Of these , amplicons for fgfr1a , fgfr1b and fgfr2 were recovered . While riboprobes against fgfr1b did not yield tissue specific staining , consistent with previous reports ( Rohner et al . , 2009 ) , fgfr1a was detected in dSFCs and fgfr2 in the epidermal posterior margin during scale development ( Figure 4—figure supplement 1D , F ) . To determine how Fgf signaling integrates with Wnt and Eda signaling during scale development , we used hagoromo ( hag ) mutants that overexpress Fgf8a in the skin post-embryonically due to a viral insertion upstream of fgf8a ( Amsterdam et al . , 2009; Kawakami et al . , 2000 ) . We found that hag/+ fish develop large , disorganized sheets of sp7:EGFP+ dSFCs ( Figure 4C ) , a previously overlooked phenotype . In hag/+ fish with repressed Wnt/β-catenin signaling , neither sheets nor foci of sp7:EGFP+ dSFCs developed in response to fgf8a overexpression ( Figure 4D ) . By contrast , hag/+ fish simultaneously homozygous for nkt—and so lacking Eda but retaining residual Wnt signaling activity ( Figure 3A–3D ) —formed broad sheets of sp7:EGFP+ dSFCs ( Figure 4F ) . Taken together these results show that Fgf-mediated differentiation of dSFCs requires Wnt/β-catenin , but not Eda signaling . The Wnt-dependence of Fgf signaling is not likely due to modulation of Fgfr gene expression ( Figure 4—figure supplement 1E , G ) but could reflect direct or indirect regulation of other Fgf pathway components yet to be identified . HH signaling is necessary for morphogenesis of amniote skin appendages ( Bitgood and McMahon , 1995; Dassule and McMahon , 1998; St-Jacques et al . , 1998 ) , and shha transcript has been detected in developing zebrafish scales ( Harris et al . , 2008; Iwasaki et al . , 2018; Sire and Akimenko , 2004 ) . To elucidate the role of HH signaling , we used a heat-shock inducible dominant repressor form of Gli2 ( DR-Gli2 ) and treatment with the Smo antagonist cyclopamine ( Gould and Missailidis , 2011; Shen et al . , 2013 ) . These treatments did not affect induction or patterning of scales , but did alter scale morphogenesis . In normal scale development , the epidermis folds around the growing posterior margin of the developing scale ( Figure 5A ) . Strikingly , epidermal folding was completely blocked by HH repression ( Figure 5B ) . Correspondence of DR-Gli2 and cyclopamine phenotypes ( Figure 5—figure supplement 1A ) and repression of HH pathway targets ( Figure 5—figure supplement 1B ) confirmed specificity and efficacy . These results also demonstrated that epidermal morphogenesis during scale development is an active , HH-dependent process and not simply a passive consequence of scale plate growth , indeed , epidermal folding was completely absent even when underlying scale plates overlapped ( Figure 5—figure supplement 1A ) . We next investigated the expression dynamics of HH ligand-encoding shha , the conserved HH transcriptional target hhip , and a transgenic reporter of HH signaling , gli2-D:mCherry ( Chuang and McMahon , 1999; Lum and Beachy , 2004; Schwend et al . , 2010 ) . Since HH repression did not interfere with scale induction we predicted that expression of HH pathway genes and signaling activation would appear later in scale morphogenesis . We found that—unlike Wnt/β-catenin , Eda and Fgf signaling—HH signaling began only after papilla morphogenesis ( Figure 5C ) . shha was initially expressed in the epidermis overlying the scale papilla and was later restricted to a column of two to three cells at the posterior scale margin ( Figure 5—figure supplement 1D ) . gli2-D:mcherry and hhip were detected in a population of cells beneath the sp7:EGFP+ dSFCs , where expression persisted throughout scale morphogenesis ( Figure 5C; Figure 5—figure supplement 1E ) . To further investigate the requirement for HH signaling in epidermal morphogenesis , we imaged fish doubly transgenic for the epidermal transgene cldnb:EGFP and gli2-D:mCherry throughout scale development . This revealed cldnb:EGFP+ epidermal cells invaginating into the underlying dermis in close association with the HH-responding population ( Figure 5D ) . Live imaging of cells at the leading edge of invaginating epidermis revealed hallmarks of active invasive migration including broad lamellipodia and long cellular extensions ( Figure 5E ) . If HH signaling triggers epidermal invagination by promoting invasion , we reasoned that Shha overexpression might promote excessive or ectopic invagination ( Shen et al . , 2013 ) . Unfortunately , Shha overexpression in fish transgenic for hs:shha-GFP led to rapid kyphosis , failure to feed and developmental arrest , precluding analyses of scale morphogenesis ( Figure 5F ) . Strikingly , however , epidermal cells of these fish rapidly moved from the surface of the fish to the glass coverslip ( Figure 5G ) . Cells at the leading edge of this sheet were cldnb:EGFP+ and displayed numerous long filopodia , consistent with active migration . Similar phenotypes have been observed in early larval skin and are associated with impaired epidermal cohesion and increased invasiveness ( Boggetti et al . , 2012; Carney et al . , 2007 ) . Because expression of shha in the epidermal posterior margin overlaps with expression of Wnt/β-catenin target genes and with edar ( Figure 2F , Figure 2—figure supplement 1A , B , E–H; Figure 3—figure supplement 1C ) , we hypothesized that shha expression might require Wnt/β-catenin signaling , Eda signaling , or both . We therefore assayed expression of HH signaling components while manipulating Wnt/β-catenin and Eda pathways . shha , hhip and gli-D:mCherry expression were strongly attenuated in both Wnt-repressed and nkt mutant fish ( Figure 5—figure supplement 1F–H ) . Since nkt mutants retain epidermal Wnt signaling activity ( Figure 3B ) but lack Eda , these data suggest that shha expression is regulated by Eda signaling during scale development . During normal life , fish often lose scales owing to interactions with other species and the environment , and scales have long been recognized for their regenerative ability ( Smith , 1931 ) . To further test roles for early ( Wnt/β-catenin ) and late ( HH ) signaling pathways , and whether functions are conserved in both ontogenetic and regenerative contexts , we removed all scales from the caudal peduncle of adult transgenic fish . One day after scale removal , patterned 7xTCF:mCherry expression revealed Wnt/β-catenin signaling in the epidermis , presaging the appearance of regenerated scales ( Figure 6A ) . To test the requirement for this activity , we repressed Wnt/β-catenin signaling during scale regeneration by Dkk1 induction . This resulted in the regeneration of significantly fewer , slower growing scales as compared to controls ( Figure 6B–6D ) . Finally , to test the role of HH signaling during regeneration we expressed DR-Gli2 following scale removal . As during scale ontogeny , HH-repression completely blocked epidermal morphogenesis ( Figure 6E ) . Therefore , Wnt/β-catenin and HH signaling play similar roles during scale development and regeneration .
Our analysis , based on live imaging of individual fish , has revealed the developmental anatomy of squamation and individual scale development . We find that zebrafish squamation proceeds through an invariant sequence of row and column addition ( Figure 7A ) , this differs from previous reports based on fixed specimens that inferred a sequential spread of scales from posterior to anterior ( Sire et al . , 1997 ) . The sequence we documented resembles that of goldfish ( Carassius auratus ) and medaka ( Oryzias latipes ) ( Iwamatsu , 2014; Li et al . , 2015 ) . Since the lineages leading to zebrafish and medaka are thought to have diverged over 300 million years ago ( Near et al . , 2012 ) , the squamation sequence presented here potentially represents a conserved , ancestral character of teleosts . Zebrafish squamation is also strikingly reminiscent of the ordered addition of chicken feather and reptile scale anlagen , suggesting this mode of patterning is conserved throughout vertebrates ( Di-Poï and Milinkovitch , 2016; Jung et al . , 1998 ) . Previous histological and ultrastructural studies have characterized the basic anatomy of scale development ( Lippitsch , 1990; Sire et al . , 1997; Sire and Akimenko , 2004 ) . We infer that sp7:EGFP+ osteoblast like cells ( dSFCs ) deposit the calcified scale plate . Based on the morphology and distribution of these cells , they likely correspond to dermal cells identified in previous fate-mapping ( Lee et al . , 2013; Mongera and Nüsslein-Volhard , 2013; Shimada et al . , 2013 ) . Our analyses of dSFC distribution in live animals showed that these cells first appear as a bi-layered condensation , corresponding to the dermal papilla identified in histological analyses ( Sire et al . , 1997 ) . Since dermal papilla formation represents the very first overt indication of scale morphogenesis , sp7:EGFP proved to be a useful marker for scale initiation . By labelling epidermis and using vital dyes that reveal calcified matrix , we were able to monitor key cell populations involved in scale formation . Combining these assays with visualization of signaling pathway activity revealed the distribution of signaling event during scale morphogenesis . For example , we find that a sheet of HH-responding cells develop beneath nascent scales and presumably coordinate scale extension ( Iwasaki et al . , 2018 ) with epidermal invagination ( Figure 7B ) , through intermediary mechanisms that are currently not known . Forward genetic screens have implicated Eda and Fgf signaling in scale development ( Harris et al . , 2008; Kondo et al . , 2001; Rohner et al . , 2009 ) . The discovery that Eda signaling is necessary for development of both amniote skin appendages and fish scales suggested a single origin of skin appendages in ancient fishes ( Di-Poï and Milinkovitch , 2016; Harris et al . , 2008; Houghton et al . , 2005; Kondo et al . , 2001; Zhang et al . , 2009 ) . Yet , it remains debated whether involvement of Eda signaling implies homology between amniote and fish skin appendages or rather a general functional requirement of Eda for epithelial–mesenchymal signaling interactions ( Sharpe , 2001 ) . We found that Eda and Wnt/β-catenin signaling integrate similarly during fish scale and amniote skin appendage development: ( i ) Wnt/β-catenin signaling was initiated in the absence of Eda signaling yet pattern refinement and reiteration were Eda-dependent ( Figure 3A–D ) ; and ( ii ) Wnt/β-catenin signaling was necessary for the expression of the Eda receptor , edar ( Figure 3F; Houghton et al . , 2005; Zhang et al . , 2009 ) . Simultaneous manipulation of multiple signaling activities revealed that while Eda and Fgf signaling are sufficient to drive ectopic differentiation of dSFCs and scale development , these pathways were able to do so only in the presence of functional Wnt/β-catenin signaling . Since dSFCs do not differentiate in Wnt-inhibited fish with simultaneous upregulation of Eda or Fgf , there are likely other , as yet unidentified , parallel mechanisms by which Wnt regulates scale development ( Figure 7C ) . These interactions are again similar to signaling interactions previously described for hair and feather patterning in amniotes ( Andl et al . , 2002; Houghton et al . , 2005; Huh et al . , 2013; Liu et al . , 2008; Mandler and Neubüser , 2004; Petiot et al . , 2003; Sick et al . , 2006 ) . During late steps in scale development , we found that Wnt/β-catenin–Eda dependent HH signaling regulates epidermal morphogenesis . Since HH signaling is also necessary for epidermal morphogenesis during hair and feather development ( McKinnell et al . , 2004; St-Jacques et al . , 1998 ) , it is possible that some functional outputs of skin appendage signaling networks are conserved and anciently evolved . In turn , this suggests a previously unappreciated role for invasive migration in epidermal morphogenesis that may be of general relevance to understanding cellular mechanisms underlying skin appendage development and regeneration ( Armstrong et al . , 2017 ) . It will be interesting to learn how modifications to these terminal processes have contributed to diversity in skin appendage morphologies across vertebrates . Taken together , our analyses of teleost scale developmental genetics using zebrafish are consistent with a single origin of skin patterning mechanisms in ancient fishes that has been conserved in extant vertebrates , even as the final adult morphology of feathers , hairs and scales appear wildly divergent . Importantly , the fossil record indicates that early tetrapods were endowed with fish-like calcified dermal scales ( Coates , 1996; Jarvik , 1996 ) , with a progressive loss of calcified matrix over geological time ( Mondéjar-Fernández et al . , 2014 ) . In light of conserved developmental regulatory architecture , this suggests a scenario in which skin appendages lost dermal calcified matrix and gained epidermal keratinization ultimately leading to the skin appendages of extant amniotes . Conservation of molecular mechanisms that regulate skin appendage patterning and early morphogenesis enables the use of zebrafish scale development as a model system for understanding vertebrate skin patterning and morphogenesis with exceptional opportunities for live imaging and forward genetic analysis , complementing existing chicken and mouse models .
Fish were maintained in the WT ( ABb ) background at 28 . 5°C . Lines used were Tg ( sp7:EGFP ) b1212 abbreviated sp7:EGFP ( DeLaurier et al . , 2010 ) ; Tg ( hsp70l:dkk1b-GFP ) w32T abbreviated hs:dkk1 ( Stoick-Cooper et al . , 2007 ) ; Tg ( 7xTCF-Xla . Siam:nlsmCherry ) ia5 abbreviated 7xTCF:mcherry ( Moro et al . , 2012 ) . Tg ( Mmu . Foxa2-cryaa:mCherry ) nu15Tg abbreviated gli2-D:mCherry ( Schwend et al . , 2010 ) ; Tg ( hsp70l:gli2aDR-EGFP ) umz33Tg abbreviated hs:gli2-DR ( Shen et al . , 2013 ) ; Tg ( hsp70l:shha-EGFP ) umz30Tg abbreviated hs:shha ( Shen et al . , 2013 ) ; Tg ( −8 . 0cldnb:LY-EGFP ) zf106Tg abbreviated cldnb:EGFP ( Haas and Gilmour , 2006 ) ; nacktdt1261 abbreviated nkt ( Harris et al . , 2008 ) ; hagoromo abbreviated hag ( Kawakami et al . , 2000 ) . For regeneration experiments , scales were removed using forceps . All coding sequences and in-situ probe templates were amplified using Primestar-GXL ( Takara ) from cDNA prepared with SSIII ( ThermoFisher ) and cloned into pJet1 . 2/blunt ( ThermoFisher ) with the following primers: lef1 5’tgtagggtgaggaggactttca , 5’cctgtagctgctgtctttgctt; axin2 5’agggataatattaagcgtcagcag , 5’ggcccttttgaagaagtatctgta; eda 5’agaggacgaggaagttcggtat; 5’gtgcatgtgttcaggtttggta; edar 5’ttacggcactaaagacgatgatta; 5’ggattagtgcagttctgtgttcc; fgfr1a 5’tcagaaagtgctgatgtcctagtc , 5’cataagtctgcacacacacacact , fgfr2 5’aattcgctgtctgctctttttct , 5’gtctcagtgtttttgagaactgga; shha 5’acaacgagaaaccctgctagac; 5’gtctctctctcactctcgctctct; hhip 5’tcagcagtcctgtttatttctgag , 5’gtaacattgccaaatggtgaagag . hsp70l:eda-2A-nls-mCherry abbreviated Eda-2A-mCherry , and hsp70l:edar-2A-nls-mCherry abbreviated Eda-2A-mCherry , were made using the tol2 Gateway Kit ( Kwan et al . , 2007 ) , and injected together with tol2 mRNA ( Kawakami , 2004 ) . In-situ probes and tissue were prepared as described previously ( Quigley et al . , 2004 ) . Probes were hybridized for 24 hr at 66°C . Post-hybridization washes were performed using a BioLane HTI 16Vx ( Intavis Bioanalytical Instruments ) , with the following parameters: 2x SSCT 3 × 5 min , 11 × 10 min at 66°C; 0 . 2x SSCT 10 × 10 min; blocking solution [5% normal goat serum ( Invitrogen ) , 2 mg/mL BSA ( RPI ) in PBST] for 24 hr at 4°C; anti-Dig-AP , Fab fragments ( 1:5000 in blocking solution , Millipore-Sigma ) for 24 hr at 4°C; PBST 59 × 20 min . AP staining was performed as described ( Quigley et al . , 2004 ) . Tissue for sectioning was equilibrated into 5% gelatin ( 300-bloom , type-A , Sigma ) , post-fixed in 4% PFA/PBS overnight at 4°C and sectioned on a Vibratome 1500 ( Harvard Apparatus ) . Tissue was fixed in freshly prepared 4% PFA/PBS ( EMS ) for 1 hr at 4°C , blocked using blocking solution ( described above ) for 24 hr at 4°C and incubated with rabbit anti-beta-catenin antibody ( 1:1000 in blocking solution , Sigma PLA0230 ) for 24 hr at 4°C . Secondary antibody ( Alexa Fluor 568 Goat anti-rabbit; Life-Technologies ALL036 ) was applied at 1:400 in blocking solution . 3 µM DAPI and 130 nM Alexa Fluor-488 Phalloidin ( ThermoFisher ) were used as counterstains . 12 × 20 min PBST washes were performed following both primary and secondary antibodies . All heatshocks were performed in a 10 gallon glass aquarium equipped with a 1000 watt submersible heater and a programmable temperature controller ( Process Technology ) . Larvae were given 6 × 1 hr 39°C heat shocks per day . Early hs:dkk1 and hs:gli2-DR: PR stage larvae ( Parichy et al . , 2009 ) were selected and heat-shocked for 10 d . Late hs:dkk1: 2R stage larvae ( Figure 1B ) were selected and heat-shocked for 3 d . hs:Eda-2A-mCherry and hs:Edar-2A-mCherry: PR stage larvae were selected and heat-shocked for 5 d . For scale regeneration , heatshock induction began 16 hr prior to scale removal and was maintained throughout regeneration timecourse . BGJ398: 2R stage larvae ( Figure 1B ) were selected and incubated daily in 1 µM BGJ398 ( Selleckchem ) ( Guagnano et al . , 2011 ) or 0 . 1% DMSO for 1 hr over 5 d . Cyclopamine: PR stage larvae were selected and incubated daily in 40 µM cyclopamine ( Selleckchem [Gould and Missailidis , 2011] ) or 0 . 25% ethanol for 2 hr over 12 d . All larvae were housed individually in 100 mL system water and fed freshly-hatched Artemia ( Aquacave ) between treatments . Vital dyes were dissolved in fish system water and titrated to pH 7 . 5 with sodium bicarbonate . Fish were incubated in 0 . 4% Alizarin Red-S ( ARS ) or 0 . 2% calcein ( Sigma ) for 1 hr at 28 . 5°C and washed for at least 1 hr in fresh system water . Live imaging: larvae were immobilized using 0 . 2% Tricane Methanesulfonate ( Western Chemical ) and imaged on an inverted Zeiss AxioObserver microscope equipped with Yokogawa CSU-X1M5000 laser spinning disk . In-situs: Zeiss AxioObserver or AxioZoom V16 uprigt stereomicroscope . Immunostainings: Zeiss LSM-800 scanning confocal microscope . Brightness and contrast were adjusted using Adobe Photoshop when necessary . Quantifications were made using ImageJ and analyzed using GraphPad Prism . All tests of statistical significance used two-tailed , unpaired t-tests with Welch's correction . | Hair , feathers or scales cover the skin of most land animals . Despite their apparent diversity , these appendages share many features: they are mainly formed of the protein keratin , are produced by the topmost layer of the skin and they start to form with skin cells moving inwards to form a pit . Across species , the same genes are also involved in controlling the development of these structures . This suggests that they have all evolved from a shared ancestral appendage , which may have been fish scales . However , scales in fish are formed of bones , not keratin , and they come from a different skin layer . Here , Aman et al . explore the molecular mechanisms that control how zebrafish scales form and get their shape , which is a little-studied area of research . Cells at the surface of the fish were imaged live on the animal as they were developing and creating scales . The experiments involved manipulating the genetic information of these cells to tease out the molecular mechanisms that drive the creation of scales . This revealed that the genes that control the formation of the scales and of the appendages of land animals are the same and interact in similar ways . In particular , scales also require the skin to form a pit to develop , and the same genes direct this process in zebrafish and in furred or feathered creatures . The work by Aman et al . suggests that all skin appendages , regardless of being sported by fish , birds or mammals , descend from the same structure . It also puts forward the zebrafish and its scales as a good model for scientists to study so they can understand better how certain hair and teeth disorders arise in humans . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"developmental",
"biology",
"evolutionary",
"biology"
] | 2018 | Wnt/β-catenin regulates an ancient signaling network during zebrafish scale development |
Neurons recorded in behaving animals often do not discernibly respond to sensory input and are not overtly task-modulated . These non-classically responsive neurons are difficult to interpret and are typically neglected from analysis , confounding attempts to connect neural activity to perception and behavior . Here , we describe a trial-by-trial , spike-timing-based algorithm to reveal the coding capacities of these neurons in auditory and frontal cortex of behaving rats . Classically responsive and non-classically responsive cells contained significant information about sensory stimuli and behavioral decisions . Stimulus category was more accurately represented in frontal cortex than auditory cortex , via ensembles of non-classically responsive cells coordinating the behavioral meaning of spike timings on correct but not error trials . This unbiased approach allows the contribution of all recorded neurons – particularly those without obvious task-related , trial-averaged firing rate modulation – to be assessed for behavioral relevance on single trials .
Spike trains recorded from the cerebral cortex of behaving animals can be complex , highly variable from trial-to-trial , and therefore challenging to interpret . A fraction of recorded cells typically exhibit trial-averaged firing rates with obvious task-related features and can be considered ‘classically responsive’ , such as neurons with tonal frequency tuning in the auditory cortex or orientation tuning in the visual cortex . Another population of responsive cells are modulated by multiple task parameters ( ‘mixed selectivity cells’ ) and have recently been shown to have computational advantages necessary for flexible behavior ( Rigotti et al . , 2013 ) . However , a substantial number of cells have variable responses that fail to demonstrate firing rates with any obvious trial-averaged relationship to task parameters ( Jaramillo and Zador , 2011; Olshausen and Field , 2006; Raposo et al . , 2014; Rodgers and DeWeese , 2014 ) . These ‘non-classically responsive’ neurons are especially prevalent in frontal cortical regions but can also be found throughout the brain , including primary sensory cortex ( Hromádka et al . , 2008; Jaramillo and Zador , 2011; Rodgers and DeWeese , 2014 ) . These response categories are not fixed but can be dynamic , with some cells apparently becoming non-classically responsive during task engagement without impairing behavioral performance ( Carcea et al . , 2017; Kuchibhotla et al . , 2017; Otazu et al . , 2009 ) . The potential contribution of these cells to behavior remains to a large extent unknown and represents a major conceptual challenge to the field ( Olshausen and Field , 2006 ) . How do these non-classically responsive cells relate to behavioral task variables on single trials ? While there are sophisticated approaches for dissecting the precise correlations between classically responsive cells and task structure ( Erlich et al . , 2011; Jaramillo and Zador , 2011; Kiani and Shadlen , 2009; Murakami et al . , 2014; Raposo et al . , 2014 ) , there is still a need for complementary and straightforward analytical tools for understanding any and all activity patterns encountered ( Jaramillo and Zador , 2011; Raposo et al . , 2014; Rigotti et al . , 2013 ) . Moreover , most behavioral tasks produce dynamic activity patterns throughout multiple neural circuits , but we lack unified methods to compare activity across different regions , and to determine to what extent these neurons might individually or collectively perform task-relevant computations . To address these limitations , we devised a novel trial-to-trial analysis using Bayesian inference that evaluates the extent to which relative spike timing in single-unit and ensemble responses encode behavioral task variables .
We trained 15 rats on an audiomotor frequency recognition go/no-go task ( Carcea et al . , 2017; Froemke et al . , 2013; King et al . , 2016; Martins and Froemke , 2015 ) that required them to nose poke to a single target tone for food reward and withhold from responding to other non-target tones ( Figure 1A ) . Tones were 100 ms in duration presented sequentially once every 5–8 s at 70 dB sound pressure level ( SPL ) ; the target tone was 4 kHz and non-target tones ranged from 0 . 5 to 32 kHz separated by one octave intervals . After a few weeks of training , rats had high hit rates to target tones and low false alarm rates to non-targets , leading to high d' values ( mean performance shown in Figure 1B; each individual rat included in this study shown in Figure 1—figure supplement 1 ) . To correctly perform this task , animals must first recognize the stimulus and then execute an appropriate motor response . We hypothesized that two brain regions important for this behavior are the auditory cortex ( AC ) and frontal cortical area 2 ( FR2 ) . Many but not all auditory cortical neurons respond to pure tones with reliable , short-latency phasic responses ( Hromádka et al . , 2008; Hubel et al . , 1959; Kadia and Wang , 2003; Merzenich et al . , 1975; Polley et al . , 2007; Wehr and Zador , 2003; Yaron et al . , 2012 ) . These neurons can process sound in a dynamic and context-sensitive manner , and AC cells are also modulated by expectation , attention , and reward structure , strongly suggesting that AC responses are important for auditory perception and cognition ( David et al . , 2012; Fritz et al . , 2003; Hubel et al . , 1959; Jaramillo and Zador , 2011; Weinberger , 2007 ) . Previously , we found that the go/no-go tone recognition task used here is sensitive to AC neuromodulation and plasticity ( Froemke et al . , 2013 ) . In contrast , FR2 is not thought to be part of the canonical central auditory pathway but is connected to many other cortical regions including AC ( Nelson and Mooney , 2016; Schneider et al . , 2018; Schneider et al . , 2014 ) . This region has recently been shown to be involved in orienting responses , categorization of perceptual stimuli , and in suppressing AC responses during movement ( Erlich et al . , 2011; Hanks et al . , 2015; Schneider et al . , 2014 ) . These characteristics suggest that FR2 may be important for goal-oriented behavior . We first asked if activity in AC or FR2 is required for animals to successfully perform this audiomotor task . We implanted cannulas into AC or FR2 ( Figure 1—figure supplement 2 ) , and infused the GABA agonist muscimol bilaterally into AC or FR2 , to inactivate either region prior to testing behavioral performance . We found that task performance was impaired if either of these regions was inactivated , although general motor functions , including motivation or ability to feed were not impaired ( Figure 1—figure supplement 3; for AC p=0 . 03; for FR2 p=0 . 009 Student’s paired two-tailed t-test ) . Thus activity in both AC and FR2 may be important , perhaps in different ways , for successful performance on this task . We note that a previously published study ( Gimenez et al . , 2015 ) observed a more modest effect of muscimol-based inactivation of auditory cortex ( although we used a separate task and higher dose of muscimol than that study which might contribute to this difference ) . Once animals reached behavioral criteria ( hit rates ≥ 70% and d’ values ≥ 1 . 5 ) , they were implanted with tetrode arrays in either AC or FR2 ( Figure 1—figure supplement 4 ) . After recovery , we made single-unit recordings from individual neurons or small ensembles of two to eight cells during task performance . The trial-averaged responses of some cells exhibited obvious task-related features: neuronal activity was tone-modulated compared to inter-trial baseline activity ( Figure 1C ) or gradually changed over the course of the trial as measured by a ramping index ( Figure 1D; hereafter referred to as ‘ramping activity’ ) . However , 60% of recorded cells were non-classically responsive in that they were neither tone modulated nor ramping according to statistical criteria ( Figure 1E and F; Figure 1—figure supplement 5; 64/103 AC cells and 43/74 FR2 cells from 15 animals had neither significant tone-modulated activity or ramping activity; pre and post-stimulus mean activity compared via subsampled bootstrapping and considered significant when p<0 . 05; ramping activity measured with linear regression and considered significant via subsampled bootstrapping when p<0 . 05 and r > 0 . 5; for overall population statistics see Figure 1—figure supplement 6 ) . While the fraction of non-classically responsive AC neurons observed is consistent with previous studies that use different auditory stimuli or behavioral paradigms ( Jaramillo and Zador , 2011; Rodgers and DeWeese , 2014 ) , this definition does not preclude the possibility that non-classically responsive cells can be driven by other acoustic stimuli or behavioral paradigms . Given that the majority of our recordings were from non-classically responsive cells , we developed a general method for interpreting neural responses even when trial-averaged responses were not obviously task-modulated which allowed us to compare coding schemes across different brain regions ( here , AC and FR2 ) . The algorithm is agnostic to the putative function of neurons as well as the task variable of interest ( here , stimulus category or behavioral choice ) . Our algorithm empirically estimates the interspike interval ( ISI ) distribution of individual neurons to decode the stimulus category ( target or non-target ) or behavioral choice ( go or no-go ) on each trial via Bayesian inference . The ISI was chosen because its distribution could vary between task conditions even without changes in the firing rate – building on previous work demonstrating that the ISI distribution contains complementary information to the firing rate ( Lundstrom and Fairhall , 2006; Reich et al . , 2000; Zuo et al . , 2015 ) . The distinction between the ISI distribution and trial-averaged firing rate is subtle , yet important . While the ISI is obviously closely related to the instantaneous firing rate , decoding with the ISI distribution is not simply a proxy for using the time-varying , trial-averaged rate . To demonstrate this , we constructed three model cells: a stimulus-evoked cell with distinct target and non-target ISI distributions ( Figure 2A ) , a stimulus-evoked cell with identical ISI distributions ( Figure 2B ) , and a non-classically responsive cell with distinct target and non-target ISI distributions ( Figure 2C ) . These models clearly demonstrate that trial-averaged rate modulation can occur with or without corresponding differences in the ISI distributions and cells without apparent trial-averaged rate-modulation can nevertheless have distinct ISI distributions . Taken together , these examples demonstrate that the ISI distribution and trial-averaged firing rate capture different spike train statistics . This has important implications for decoding non-classically responsive cells that by definition do not exhibit large firing rate modulations but nevertheless may contain information latent in their ISI distributions . For each recorded neuron , we built a library of ISIs observed during target trials and a library for non-target trials from a set of ‘training trials’ . Two different cells from AC are shown in Figure 3A and Figure 3—figure supplement 1A–D , and another cell from FR2 is shown in Figure 3—figure supplement 1E–H . These libraries were used to infer the probability of observing an ISI during a particular trial type ( Figure 3B , C; Figure 3—figure supplement 1C , G; left panels show target in red and non-target in blue ) . These conditional probabilities were inferred using non-parametric statistical methods to minimize assumptions about the underlying process generating the ISI distribution and better capture the heterogeneity of the observed ISI distributions ( Figure 3B; Figure 3—figure supplement 1C , G ) . We verified that our observed distributions were better modeled by non-parametric methods rather than standard parametric methods ( e . g . rate-modulated Poisson process; Figure 3—figure supplement 2 ) . Specifically , we found the distributions using Kernel Density Estimation where the kernel bandwidth for each distribution was set using 10-fold cross-validation . To accommodate any non-stationarity , these ISI distributions were calculated in 1 s long sliding windows recalculated every 100 ms over the course of the trial . We then used these training set probability functions to decode a spike train from a previously unexamined individual trial from the set of remaining ‘test trials’ . This process was repeated 124 times using 10-fold cross-validation with randomly generated folds . Importantly , while the probabilities of observing particular ISIs on target and non-target trials were similar ( Figure 3B; Figure 3—figure supplement 1C , G ) , small differences between the curves carried sufficient information to allow for decoding . To characterize these differences , we used the weighted log likelihood ratio ( W . LLR; Figure 3C; Figure 3—figure supplement 1C , G ) to clearly represent which ISIs suggested target ( W . LLR > 0 ) or non-target ( W . LLR < 0 ) stimulus categories . Our algorithm relies only on statistical differences between task conditions; therefore , the W . LLR summarizes all spike timing information necessary for decoding . Similar ISI libraries were also computed for behavioral choice categories ( Figure 3B , C; Figure 3—figure supplement 1C; right panels show go decision in green and no-go in purple ) . These examples clearly illustrate that the relationship between the ISIs and task variables cannot simply be approximated by an ISI or firing rate threshold where short ISIs imply one task variable and longer ISIs imply another: in the cell shown in Figure 3 , short ISIs ( ISI < 50 ms ) indicated non-target , medium ISIs ( 50 ms <ISI < 100 ms ) indicated target , and longer ISIs indicated non-target ( 100 ms <ISI ) . The algorithm uses the statistical prevalence of certain ISI values under particular task conditions ( in this case the ISIs accompanying stimulus category or behavioral choice ) , to infer the task condition for each trial . Each trial begins with equally uncertain probabilities about the stimulus categories ( i . e . p ( target ) =p ( non-target ) =50% ) . As each ISI is observed sequentially within the trial , the algorithm applies Bayes’ rule to update p ( target|ISI ) and p ( non-target|ISI ) using the likelihood of the ISI under each stimulus category ( p ( ISI|target ) and p ( ISI|non-target ) ( Figure 3B–D ) . As these functions were estimated in 1 s long sliding windows , each ISI was assessed using the distribution that placed the final spike closest to the center of the sliding window . As shown for one trial of the example cell in Figure 3D , ISIs observed between 0 and 1 . 0 s consistently suggested the presence of the target tone , whereas ISIs observed between 1 . 0 and 1 . 4 s suggested the non-target category thereby also necessarily reducing the belief that a target tone was played ( Figure 3D , top trace ) . These ISI likelihood functions consider each ISI to be independent of previous ISIs and therefore ignore correlations between ISIs . After this process was completed for all ISIs in the particular trial , we obtained the probability of a non-target tone and a target tone as a function of time during the trial ( Figure 3D ) . Because it is particularly challenging to dissociate choice from motor execution or preparatory motor activity in this task paradigm , the prediction for the entire trial p ( target|ISI ) is evaluated at the end of the trial ( in the example trial , p ( target|ISI ) =61%; Figure 3D ) . This process is repeated for the behavioral choice ( Figure 3B–D; right panels; trials separated according to go , no-go; probabilities of ISIs in each condition generated; conditional probabilities used as likelihood function to predict behavioral choice on a given trial ) . The single-trial decoding performance of each neuron is then averaged over all trials as a measure of the overall ability of each neuron to distinguish behavioral conditions ( Figure 4A ) . Note that this measure not only takes into account whether the algorithm was correct on individual trials ( i . e . target vs . non-target ) , but also its prediction certainty . Can we uncover task information from non-classically responsive cells ? We found that non-classically responsive cells in both AC and FR2 provided significant spike-timing-based information about each task variable ( Figure 4A , B , red; Figure 4—figure supplement 1 ) . The ability to decode was poorly explained by the average firing rate ( Figure 4—figure supplements 2A–F and 0 . 30 < r < 0 . 46 ) , z-score ( Figure 4—figure supplements 2G–I , –0 . 05 < r < 0 . 05 ) , and ramping activity ( Figure 4—figure supplements 2J , –0 . 02 < r < 0 . 28 ) . Stimulus decoding performance was also independent of receptive field properties including best frequency and tuning curve bandwidth for AC neurons ( Figure 4—figure supplement 3 ) . We also observed that task information was distributed across both AC and FR2 , and neural spike trains from individual units were multiplexed in that they often encoded information about both stimulus category and choice simultaneously ( Figure 4B , Table 1 ) . Given the strong correlation between stimulus and choice variables in the task design , it is difficult to fully separate information about one variable from information about the other . To establish that multiplexing was not simply a byproduct of this correlation , an independent measure of multiplexing relying on multiple regression was applied ( Figure 4—figure supplement 4 ) . This analysis confirmed that the information revealed by our algorithm about a behavioral variable was primarily a reflection of that variable and not simply an indirect measure of the other , correlated variable . This analysis establishes that a certain degree of separability is possible and demonstrates that the multiplexing observed in our decoding results is unlikely to be a trivial byproduct of correlations in the task variables . Despite the broad sharing of information about behavioral conditions , there were notable systematic differences between AC and FR2 . Surprisingly , neurons in FR2 were more informative about stimulus category than AC , and AC neurons were more informative about choice than stimulus category ( Figure 4A , pAC = 0 . 016 , pstim = 0 . 0013 , Mann-Whitney U test , two-sided ) . Both of these observations would not have been detected at the level of the PSTH , as most cells in AC were non-classically responsive for behavioral choice ( no ramping activity , 91/103 ) , yet our decoder revealed that these same cells were as informative as choice classically responsive cells ( Figure 4C , p=0 . 32 Mann-Whitney U test , two-sided; red circles indicate cells non-classically responsive for both variables , dark-red cells are choice non-classically responsive , and black cells are classically responsive ) . Similarly , most cells in FR2 were sensory non-classically responsive ( not tone modulated , 60/74 ) , yet contained comparable stimulus information to sensory classically responsive cells ( Figure 4D , p=0 . 29 Mann-Whitney U test , two-sided; red cells are non-classically responsive for both variables , dark-red cells are sensory non-classically responsive , black cells are classically responsive ) . To assess the statistical significance of these results , we tested our algorithm on two shuffled data sets . First , we ran our analysis using synthetically-generated trials that preserved trial length but randomly sampled ISIs with replacement from those observed during a session without regard to condition ( Figure 4E ) . Second , we left trial activity intact but permuted the stimulus category and choice for each trial ( Figure 4F ) . We restricted analysis to cells with decoding performance significantly different from synthetic spike trains ( all cells in Figure 4A–D significantly different from synthetic condition shown in Figure 4E , p<0 . 05 , bootstrapped 1240 times ) . To directly assess the extent to which information captured by the ISI distributions in our data set was distinct from the time-varying rate , we compared the performance from our ISI-based decoder to a conventional rate-modulated ( inhomogeneous ) Poisson decoder ( Rieke et al . , 1999 ) , which assumes that spikes are produced randomly with an instantaneous probability equal to the time-varying firing rate . As our model cells illustrate ( Figure 2 ) , it is possible to decode using the ISI distributions even when firing rates are uninformative ( Figure 5A ) . When applied to our dataset , the ISI-based decoder generally outperformed this conventional rate-based decoder confirming that ISIs capture information distinct from that of the firing rate ( Figure 5B; Overall stimulus decoding performance: pAC = 0 . 0001 , pFR2 = 8 × 10−6; Overall choice decoding performance pAC = 0 . 0057 , pFR2 = 0 . 02 , Mann-Whitney U test , two-sided ) . Moreover , comparing single trial decoding outcomes demonstrated weak to no correlations between the ISI-based decoder and the conventional rate decoder , further underscoring that these two methods rely on different features of the spike train to decode ( Figure 5C; stimulus medians: AC = 0 . 10 FR2 = 0 . 11; choice medians: AC = 0 . 07 , FR2 = 0 . 08 ) . We hypothesize that ISI-based decoding is biologically plausible . Short-term synaptic plasticity and synaptic integration provide powerful mechanisms for differential and specific spike-timing-based coding . We illustrated this capacity by making whole-cell recordings from AC neurons in vivo and in brain slices ( Figure 5—figure supplement 1A , B ) , as well as in FR2 brain slices ( Figure 5—figure supplement 1C ) . In each case , different cells could have distinct response profiles to the same input pattern , with similar overall rates but different spike timings . Moreover , we note that this type of coding scheme requires few assumptions about implementation , and does not require additional separate integrative processes to compute rates or form generative models . Thus , ISI-based decoding coding could be generally applicable across brain areas , as demonstrated here for AC and FR2 . To further demonstrate the generalizability and utility of our approach , we applied our decoding algorithm to neurons that were found to be non-classically responsive in a previously published study ( Rodgers and DeWeese , 2014 ) . In this study , rats were trained on a novel auditory stimulus selection task where depending on the context animals had to respond to one of two cues while ignoring the other . Rats were presented with two simultaneous sounds ( a white noise burst and a warble ) . In the ‘localization’ context , the animal was trained to ignore the warble and respond to the location of the white noise burst and in the ‘pitch’ context it was trained to ignore the location of the white noise burst and respond to the pitch of the warble ( Figure 6A ) . Using our algorithm , we found significant stimulus and choice-related information in the activity of non-classically responsive cells that displayed no stimulus modulation nor ramping activity in the firing rate ( Figure 6B–D ) . The main finding of the study is that the pre-stimulus activity in both primary auditory cortex and prefrontal cortex encodes the selection rule ( i . e . activity reflects whether the animal is in the localization or pitch context ) . This conclusion was entirely based on a difference in pre-stimulus firing rate between the two contexts . The authors reported , but did not further analyze , cells that did not modulate their pre-stimulus firing rate . In our nomenclature these cells are ‘non-classically responsive for the selection rule’ . Using our algorithm , we found that the ISI distributions of these cells encoded the selection rule and were significantly more informative than the classically responsive cells ( Figure 6E , pAC = 5 × 10−6 , pPFC <0 . 0002 , Mann-Whitney U test , two-sided ) . This surprising result demonstrates that our algorithm generalizes to novel datasets , and may be used to uncover coding for cognitive variables beyond those apparent from conventional trial-averaged , rate-based analyses . Furthermore , these results indicate that as task complexity increases non-classically responsive cells are differentially recruited for successful task execution . Downstream brain regions must integrate the activity of many neurons and this ISI-based approach naturally extends to simultaneously recorded ensembles . We therefore asked whether using small ensembles would change or improve decoding . To decode from ensembles , likelihood functions from each cell were calculated independently as before but were used to simultaneously update the task condition probabilities ( p ( target | ISI ) and p ( go | ISI ) ) on each trial ( Figure 7A ) . Analyzing ensembles of two to eight neurons in AC and FR2 significantly improved decoding for both variables in FR2 and stimulus decoding in AC ( Figure 7B , pAC stim=0 . 04 , pFR2 stim=1×10−5 , pAC = 0 . 29 , pFR2 choice=7×10−5 , Mann-Whitney U test , two-sided ) . This was not a trivial consequence of using more neurons , as the information provided by individual ISIs on single trials can be contradictory ( e . g . compare LLR functions in Figure 3C and Figure 3—figure supplement 1C for 50 ms <ISIs < 120 ms ) . For ensemble decoding to improve upon single neuron decoding , the ISIs of each member of the ensemble must indicate the same task variable . Can our decoding method predict errors on a trial-by-trial basis ? In general , trial-averaged PSTHs did not reveal systematic differences between correct and error trials ( Figure 7—figure supplement 1 ) . However , when we examined single-trial performance with our algorithm , ensembles of neurons in AC and FR2 predicted behavioral errors ( Figure 7C ) . In general , ensembles in AC predicted behavioral errors significantly better than those in FR2 ( Figure 7C , for three-member ensembles: p=1 . 2 × 10−5 , for four-member ensembles: p=0 . 03 , Mann-Whitney U test , two-sided ) . Interestingly , decoding with an increasing number of non-classically responsive cells improved error prediction in both AC and FR2 ( Figure 7D , ( pAC = 0 . 013 , pFR2 = 0 . 046 , Welch’s t-test ) . While improvements were seen in decoding performance with increasing ensemble size , the ISI distributions/ISI-based likelihood functions were highly variable across individual ensemble members . Thus , we wondered if there was task-related structure in the timing of population activity that evolved over the course of the trial to instantiate behavior . To answer this question , we examined whether local ensembles share the same representation of task variables over the course of the trial . Do they ‘reach consensus’ on how to represent task variables using the ISI ( Figure 8A ) ? Without consensus , a downstream area would need to interpret ensemble activity using multiple disparate representations rather than one unified code ( Figure 8B ) . The firing rates and ISI distributions of simultaneously-recorded units were generally variable across cells requiring an exploratory approach to answer this question ( Figure 8C , example three-member ensemble with heterogeneous conditional ISI distributions ) . Therefore , we examined changes in the distributions of ISIs across task conditions , asking how the moment-to-moment changes in the log-likelihood ratio ( LLR ) of each cell were coordinated to encode task variables ( Figure 8C ) . We focused on the LLR because it quantifies how the ISI represents task variables for a given cell and summarizes all spike timing information needed by our algorithm ( or a hypothetical downstream cell ) to decode . We examined how ensembles coordinate their activity moment-to-moment over the course of the trial by quantifying the similarity of the LLRs across cells in a sliding window . Similarity was assessed by summing the LLRs of ensemble members , calculating the total area underneath the resulting curve , and normalizing this value by the sum of the areas of each individual LLR . We refer to this quantified similarity as ‘consensus’; a high consensus value indicates that ensemble members have similar LLRs and therefore have a similar representation of task variables ( Figure 8D ) . We should emphasize that successful ensemble decoding ( Figure 7 ) does not require the LLRs of ensemble members to be related in any way; therefore , structured LLR dynamics ( Figure 8 ) are not simply a consequence of how our algorithm is constructed . While the conventional trial-averaged PSTH of non-classically responsive ensembles recorded in AC and FR2 showed no task-related modulation , our analysis revealed structured temporal dynamics of the LLRs ( captured by the consensus value ) . On correct trials , we observe a trajectory of increasing consensus at specific moments during the trial signifying a dynamically created , shared ISI representation of task variables . In FR2 , sensory non-classically responsive ensembles ( ensembles in which at least two out of three cells were not tone-modulated ) encode stimulus information using temporally-precise stimulus-related dynamics on correct trials . The stimulus representation of sensory non-classically responsive ensembles reached consensus rapidly after stimulus onset followed by divergence ( Figure 8E , stimulus-aligned , solid line , Δconsensus , t = 0 to 0 . 42 s , pSNR = 3 . 9 × 10−4 Wilcoxon test with Bonferroni correction , two-sided ) . Sensory classically responsive ensembles in AC increased consensus beyond stimulus presentation , reaching a maximum ~750 ms after tone onset on correct trials ( Figure 8E stimulus-aligned , dotted line , Δconsensus , t = 0 to 0 . 81 s , pSR = 0 . 14 Wilcoxon test with Bonferroni correction , two-sided ) . For choice-related activity , choice non-classically responsive ensembles in both regions as well as choice classically responsive ensembles in FR2 each reached consensus within 500 ms of the behavioral response ( Figure 8E , response-aligned , Δconsensus , t = −1 . 0 to 0 . 0 s , pCNR = 2 . 0 × 10−5 , pCR = 0 . 12 Wilcoxon test with Bonferroni correction , two-sided ) . Importantly , this temporally precise pattern of consensus building is not present on error trials . On error trials , stimulus consensus dynamics decreased over the course of the trial whereas choice dynamics did not display a systematic increase with the exception of choice non-classically responsive ensembles in AC which remained systematically lower than correct trials ( Figure 8F , Δconsensus , correct trials vs . error trials , stimulus: pSNR = 0 . 007 , pSR = 0 . 065 , choice: pCNR = 0 . 0048 , pCR = 0 . 065 Mann-Whitney U test , two-sided , Δconsensus on error trials , t = 0 to 0 . 42 s , pSNR = 1 . 3 × 10−33 , t = −1 . 0 to 0 s , pCNR = 0 . 032 , pCR = 0 . 14 Wilcoxon test with Bonferroni correction , two-sided ) . The observed increases in ensemble consensus on correct trials ( while failing do so on error trials ) suggests that achieving a shared ISI representation of task variables may be relevant for successful task execution . These results reveal that consensus-building and divergence occur at key moments during the trial for successful execution of behavior in a manner that is invisible at the level of the PSTH . As sensory and choice non-classically responsive ensembles participated in these dynamics , changes in the consensus value cannot simply be a byproduct of correlated firing rate modulation due to tone-evoked responses or ramping . While consensus-building can only indicate a shared representation , divergence can indicate one of two things: ( 1 ) the LLRs of each cell within an ensemble are completely dissimilar or ( 2 ) they are ‘out of phase’ with one another – the LLRs partition the ISIs the same way ( Figure 8D , dotted lines ) , but the same ISIs code for opposite behavioral variables . This distinction is important because ( 2 ) implies coordinated structure of ensemble activity ( the partitions of the ISI align ) whereas ( 1 ) does not . To distinguish between these two possibilities , we used the ‘unsigned consensus’ , a second measure sensitive to the ISI partitions but insensitive to the sign of the LLR . Both ‘in phase’ and perfectly ‘out of phase’ LLRs would produce an unsigned consensus of 1 , whereas unrelated LLRs would be closer to 0 ( Figure 8D ) . For example , in the second row of Figure 8D , both cells agree that ISIs < 100 ms indicate one stimulus category and ISIs > 100 ms indicate another , but they disagree about which set of ISIs mean target and which mean non-target . This results in a consensus value of 0 ( out of phase ) but an unsigned consensus value of 1 . Using this metric , we found that the unsigned consensus pattern for non-classically responsive ensembles ( ensembles with two or more non-classically responsive members ) were shared between AC and FR2 – increasing until ~750 ms after tone onset on correct trials ( Figure 8G , stimulus-aligned , Δ consensus , t = 0 to 0 . 89 s , p=1 . 7 × 10−5 Wilcoxon test , two-sided ) . Non-classically responsive ensembles in AC and FR2 also increased their unsigned consensus immediately before behavioral response ( although values in AC were lower overall; Figure 8G , response-aligned , Δconsensus , t = −1 . 0 to 0 . 0 s , p=0 . 0011 Wilcoxon test , two-sided ) . This pattern of consensus-building was only present on correct trials . On error trials unsigned consensus values did not systematically increase ( Figure 8H , Δconsensus compared to error trials , p=1 . 9 × 10−9 Mann-Whitney U test , two-sided ) suggesting that behavioral errors might result from a general lack of consensus between ensemble members . In summary , we have shown that cells which appear unmodulated during behavior do not encode task information independently , but do so by synchronizing their representation of behavioral variables dynamically during the trial .
Using a straightforward , single-trial , ISI decoding algorithm that makes few assumptions about the proper model for neural activity , we found task-specific information extensively represented by non-classically responsive neurons in both AC and FR2 that lacked conventional task-related , trial-averaged firing rate modulation . The complexity of single-trial spiking patterns and the apparent variability between trials led to the development of this novel decoding method . Furthermore , the heterogeneity in the observed ISI distributions within and across brain regions precluded a straightforward interpretation of these distributions and instead suggested an approach which focused on whether and when these distributions are shared in local ensembles via consensus-building . The degree to which single neurons were task-modulated was uncorrelated with conventional response properties including frequency tuning . AC and FR2 each represent both task-variables; furthermore , in both regions , we identified many multiplexed neurons that simultaneously represented the sensory input and the upcoming behavioral choice including non-classically responsive cells . This highlights that the cortical circuits that generate behavior exist in a distributed network – blurring the traditional modular view of sensory and frontal cortical regions . Most notably , FR2 has a better representation of task-relevant auditory stimuli than AC . The prevalence of stimulus information in FR2 might be surprising given that AC reliably responds to pure tones in untrained animals; however , when tones take on behavioral significance , this information is encoded more robustly in frontal cortex , suggesting that this region is critical for identifying the appropriate sensory-motor association . Furthermore , the stark improvement in stimulus encoding for small ensembles in FR2 suggests that task-relevant stimulus information is reflected more homogeneously in local firing activity across FR2 ( perhaps through large-scale ensemble consensus-building ) while this information is reflected in a more complex and distributed manner throughout AC . We have identified task-informative non-classically responsive neurons recorded while animals performed a frequency recognition task or a task-switching paradigm . This does not preclude the possibility that these cells are driven by other acoustic stimuli or in other behavioral contexts; however , determining the significance of non-classically responsive activity must ultimately be considered in the specific behavioral context in question , as their role may be dynamic and context dependent . The finding that the ISI-based approach of our algorithm is not reducible to rate despite their close mathematical relationship raises the question of how downstream regions could respond preferentially to specific ISIs . Our whole-cell recordings from both AC and FR2 demonstrate that different postsynaptic cells can respond differently to the same input pattern with a fixed overall rate , emphasizing the importance of considering a code sensitive to precise spike-timing perhaps via mechanisms of differential short-term plasticity such as depression and facilitation ( Figure 5—figure supplement 1 ) . Furthermore , this is supported by experimental and theoretical work showing that single neurons can act as resonators tuned to a certain periodicity of firing input ( Izhikevich , 2000 ) . This view could also be expanded to larger neuronal populations comprised of feedback loops that would resonate in response to particular ISIs . In this case , cholinergic neuromodulation could offer a mechanism for adjusting the sensitivities of such a network during behavior on short time-scales by providing rapid phasic signals ( Hangya et al . , 2015 ) . Our consensus results reveal dynamic changes in the relationship between the LLRs of ensemble members . How might such a downstream resonator interpret a given ISI in the context of these dynamics ? Our consensus analysis provides one possible answer: downstream neurons may be attuned to the ISIs specified by the consensus LLR of an ensemble . In such a model , an ensemble would have the strongest influence on downstream activity when they reach high consensus . We additionally hypothesize that mechanisms of long-term synaptic plasticity such as spike-timing-dependent plasticity can redistribute synaptic efficacy , essentially changing the dynamics of short-term plasticity independent from overall changes in amplitudes ( Markram and Tsodyks , 1996 ) . Thus , after training , downstream neurons do not need to continually change the readout mechanism- rather , the upstream and downstream components might be modified together by cortical plasticity during initial phases of behavioral training . This would set the ISI distributions appropriate for firing of task-relevant downstream neurons , which would ensure that ensemble consensus is reached for correct sensory processing in highly trained animals . It is still unclear what the relevant timescales of decoding might be in relation to phenomena such as membrane time constants , periods of oscillatory activity , and behavioral timescales . Given that our ISI-based decoder and conventional rate-modulated decoders reveal distinct information , future approaches might hybridize these rate-based and temporal-based decoding methods to span multiple timescales . Other recent studies have also contributed to our understanding of non-classically responsive activity , by evaluating firing rates or responses from calcium imaging to demonstrate how correlations with classically responsive activity may contribute to the linear separability of ensemble responses ( Leavitt et al . , 2017; Zylberberg , 2018 ) . We have shown that underlying the task-relevant information encoded by each ensemble is a rich set of consensus-building dynamics that is invisible at the level of the PSTH . Ensembles in both FR2 and AC underwent stimulus and choice-related consensus building that was only observed when the animal correctly executed the task . Moreover , non-classically responsive cells demonstrated temporal dynamics synchronized across regions which were distinct from classically responsive ensembles . These results underscore the importance of measuring neural activity in behaving animals and using unbiased and generally applicable analytical methods , as the response properties of cortical neurons in a behavioral context become complex in ways that challenge our conventional assumptions ( Carcea et al . , 2017; Fritz et al . , 2010; Kuchibhotla et al . , 2017; Otazu et al . , 2009 ) .
All animal procedures were performed in accordance with National Institutes of Health standards and were conducted under a protocol approved by the New York University School of Medicine Institutional Animal Care and Use Committee . We used 23 adult Sprague-Dawley male and female rats ( Charles River ) in the behavioral studies . Animals were food restricted and kept at 85% of their initial body weight , and maintained at a 12 hr light/12 hr dark cycle . Animals were trained on a go/no-go audiomotor task ( Carcea et al . , 2017; Froemke et al . , 2013 ) . Operant conditioning was performed within 12’ L x 10’ W x 10 . 5’ H test chambers with stainless steel floors and clear polycarbonate walls ( Med Associates ) , enclosed in a sound attenuation cubicle and lined with soundproofing acoustic foam ( Med Associates ) . The nose and reward ports were both arranged on one of the walls with the speaker on the opposite wall . The nose port , reward port , and the speaker were controlled and monitored with a custom-programmed microcontroller . Nose port entries were detected with an infrared beam break detector . Auditory stimuli were delivered through an electromagnetic dynamic speaker ( Med Associates ) calibrated using a pressure field microphone ( ACO Pacific ) . Animals were rewarded with food for nose poking within 2 . 5 s of presentation of the target tone ( 4 kHz ) and given a short 7 s time-out for incorrectly responding to non-target tones ( 0 . 5 , 1 , 2 , 8 , 16 , 32 kHz ) . Incorrect responses include either failure to enter the nose port after target tone presentation ( miss trials ) or entering the nose port after non-target tone presentation ( false alarms ) . Tones were 100 ms in duration and sound intensity was set to 70 dB SPL . Tones were presented randomly with equal probability such that each stimulus category was presented . The inter-trial interval delays used were 5 , 6 , 7 , or 8 s . For experiments involving muscimol , we implanted bilateral cannulas in either FR2 ( +2 . 0 to +4 . 0 mm AP , ±1 . 3 mm ML from Bregma ) of seven animals or AC ( -5 . 0 to -5 . 8 mm AP , 6 . 5-7 . 0 mm ML from Bregma ) of three animals . We infused 1 μL of muscimol per side into FR2 or infused 2 μL of muscimol per side into AC , at a concentration of 1 mg/mL . For saline controls , equivalent volumes of saline were infused in each region . Behavioral testing was performed 30-60 min after infusions . Power analysis was performed to determine sample size for statistical significance with a power of β: 0 . 8; these studies required at least three animals , satisfied in the experiments of Figure 1-figure supplement 3B , E . For motor control study , animals could freely nose poke for food reward without presentation of auditory stimuli after muscimol and saline infusion . Animals were implanted with microdrive arrays ( Versadrive-8 Neuralynx ) in either AC ( eight animals ) or FR2 ( seven animals ) after reaching behavioral criteria of d’≥1 . 0 . For surgery , animals were anesthetized with ketamine ( 40 mg/kg ) and dexmedetomidine ( 0 . 125 mg/kg ) . Stainless steel screws and dental cement were used to secure the microdrive to the skull , and one screw was used as ground . Each drive consisted of eight independently adjustable tetrodes . The tetrodes were made by twisting and fusing four polyimide-coated nichrome wires ( Sandvik Kanthal HP Reid Precision Fine Tetrode Wire; wire diameter 12 . 5 μm ) . The tip of each tetrode was gold-plated to an impedance of 300–400 kOhms at 1 kHz ( NanoZ , Neuralynx ) . Recordings in behaving rats were performed as previously described ( Carcea et al . , 2017 ) . After the animal recovered from surgery ( ~7 days ) recordings began once performance returned to pre-surgery levels . Tetrodes were advanced ~60 μm 12 hr prior to each recording session , to a maximum of 2 . 5 mm ( for FR2 ) or 2 . 0 mm ( for AC ) from the pial surface . For recording , signals were first amplified onboard using a small 16-bit unity-gain preamplifier array ( CerePlex M , Blackrock Microsystems ) before reaching the acquisition system . Spikes were sampled at 30 kS/sec and bandpass filtered between 250 Hz and 5 kHz . Data were digitized and all above-threshold events with signal-to-noise ratios > 3:1 were stored for offline spike sorting . Single-units were identified on each tetrode using OfflineSorter ( Plexon Inc ) by manually classifying spikes projected as points in 2D or 3D feature space . The parameters used for sorting included the waveforms projection onto the first two principal components , energy , and nonlinear energy . Artifacts were rejected based on refractory period violations ( <1 msec ) . Clustering quality was assessed based on the Isolation Distance and Lratio sorting quality metrics . To be initially included for analysis , cells had to have >3 spikes per trial for 80% of trials to ensure that there were enough ISIs to reliably estimate the ISI probability density functions . We used two positive statistical tests for non-classical responsiveness: one to establish a lack of tone-modulation , the other to establish a lack of ramping activity . To accommodate the possibility of tone onset and offset responses , we performed our tone-modulation test on a 100 ms long tone presentation window as well as the 100 ms window immediately after tone presentation . The test compared the number of spikes during each of these windows to inter-trial baseline activity as measured by three sequential 100 ms windows preceding tone onset . Three windows were chosen to account for variability in spontaneous spike counts . Given that spike counts are discrete , bounded , and non-normal , we used subsampled bootstrapping to evaluate whether the mean change in spikes during tone presentation was sufficiently close to zero ( in our case 0 . 1 spikes ) . We subsampled 90% of the spike count changes from baseline , calculated the mean of these values , and repeated this process 5000 times to construct a distribution of means . If 95% of the subsampled means values were between −0 . 1 and 0 . 1 , we considered the cell sensory non-classically responsive ( p<0 . 05 ) . The range of mean values from −0 . 1 to 0 . 1 were included to account for both tone-evoked ( increases in spike count ) and tone-suppressed ( decreases in spike count ) activity . The value of 0 . 1 spikes was chosen to be conservative as it is equivalent to an expected change of 1 spike every 10 trials . This is a conservative , rigorous method for establishing sensory non-classical responsiveness that is commensurate with more standard approaches for establishing tone responsiveness such as the z-score . To quantify the observed sustained increase or decrease in firing rate preceding the behavioral response , a ramp index was calculated adapted from the ‘build-up rate’ used in previous literature31 . First , the trial averaged firing rate was determined in 50 ms bins leading up to the behavioral response . We then calculated the slope of a linear regression in a 500 ms long sliding window beginning 850 ms before behavioral response . The maximum value of these slopes was used as the ‘ramp index’ for each cell . Cells were classified as choice non-classically responsive if the ramp index did not indicate an appreciable change in the firing rate ( less than 50% change ) established via subsampled bootstrapping . Cells that were shown to be both sensory and choice non-classically responsive were considered non-classically responsive overall ( Figure 4A , B , red circles ) . Spontaneous average firing rate was established by averaging spikes in a 100 ms time window immediately prior to tone onset on each trial . To quantify tone modulated responses observed during stimulus presentation , we calculated z-scores of changes in spike count from 100 ms before tone onset to 100 ms during tone presentation:z= μσwhere μ is the mean change in spike count and σ is the standard deviation of the change in spike count . Receptive fields were constructed by calculating the average change in firing rate from 50 ms before tone onset to 50 ms during tone presentation . The window used during tone presentation was identical to that used to calculate the z-score . Best frequency was defined as the frequency where the largest positive deviation in the evoked firing rate was observed . Tuning curve bandwidth was determined by calculating the width of the tuning curve measured at the mean of the maximum and minimum observed evoked firing rates . Sprague-Dawley rats 3–5 months old were anesthetized with pentobarbital . Experiments were carried out in a sound-attenuating chamber . Series of pure tones ( 70 dB SPL , 0 . 5–32 kHz , 50 ms , 3 ms cosine on/off ramps , inter-tone intervals between 50 and 500 ms ) were delivered in pseudo-random sequence . Primary AC location was determined by mapping multiunit responses 500–700 µm below the surface using tungsten electrodes . In vivo whole-cell voltage-clamp recordings were then obtained from neurons located 400–1100 µm below the pial surface . Recordings were made with an AxoClamp 2B ( Molecular Devices ) . Whole-cell pipettes ( 5–9 MΩ ) contained ( in mM ) : 125 Cs-gluconate , 5 TEACl , 4 MgATP , 0 . 3 GTP , 10 phosphocreatine , 10 HEPES , 0 . 5 EGTA , 3 . 5 QX-314 , 2 CsCl , pH 7 . 2 . Data were filtered at 2 kHz , digitized at 10 kHz , and analyzed with Clampfit 10 ( Molecular Devices ) . Tone-evoked excitatory postsynaptic currents were recorded at –70 mV . Acute brain slices of AC or FR2 were prepared from 2 to 5 month old Sprague-Dawley rats . Animals were deeply anesthetized with a 1:1 ketamine/xylazine cocktail and decapitated . The brain was rapidly placed in ice-cold dissection buffer containing ( in mM ) : 87 NaCl , 75 sucrose , 2 . 5 KCl , 1 . 25 NaH2PO4 , 0 . 5 CaCl2 , 7 MgCl2 , 25 NaHCO3 , 1 . 3 ascorbic acid , and 10 dextrose , bubbled with 95%/5% O2/CO2 ( pH 7 . 4 ) . Slices ( 300–400 µm thick ) were prepared with a vibratome ( Leica ) , placed in warm dissection buffer ( 32–35°C ) for 10 min , then transferred to a holding chamber containing artificial cerebrospinal fluid at room temperature ( ACSF , in mM: 124 NaCl , 2 . 5 KCl , 1 . 5 MgSO4 , 1 . 25 NaH2PO4 , 2 . 5 CaCl2 , and 26 NaHCO3 , ) . Slices were kept at room temperature ( 22–24°C ) for at least 30 min before use . For experiments , slices were transferred to the recording chamber and perfused ( 2–2 . 5 ml min−1 ) with oxygenated ACSF at 33°C . Somatic whole-cell current-clamp recordings were made from layer five pyramidal cells with a Multiclamp 700B amplifier ( Molecular Devices ) using IR-DIC video microscopy ( Olympus ) . Patch pipettes ( 3–8 MΩ ) were filled with intracellular solution containing ( in mM ) : 120 K-gluconate , 5 NaCl , 10 HEPES , 5 MgATP , 10 phosphocreatine , and 0 . 3 GTP . Data were filtered at 2 kHz , digitized at 10 kHz , and analyzed with Clampfit 10 ( Molecular Devices ) . Focal extracellular stimulation was applied with a bipolar glass electrode ( AMPI Master-9 , stimulation strengths of 0 . 1–10 V for 0 . 3 msec ) . Spike trains recorded from AC and FR2 units during behavior were then divided into 150–1000 msec fragments , and used as extracellular input patterns for these recordings . Our decoding method was motivated by the following general principles: First , single-trial spike timing is one of the only variables available to downstream neurons . Any observations about trial-averaged activity must ultimately be useful for single-trial decoding , in order to have behavioral significance . Second , there may not be obvious structure in the trial-averaged activity to suggest how non-classically responsive cells participate in behaviorally-important computations . This consideration distinguishes our method from other approaches that rely explicitly or implicitly on the PSTH for interpretation or decoding ( Churchland et al . , 2008; Erlich et al . , 2011; Jaramillo et al . , 2014; Jaramillo and Zador , 2011; Murakami et al . , 2014; Wiener and Richmond , 2003 ) . Third , we required a unified approach capable of decoding from both classically responsive and non-classically responsive cells in sensory and frontal areas with potentially different response profiles . Fourth , our model should contain as few parameters as possible to account for all relevant behavioral variables ( stimulus category and behavioral choice ) . This model-free approach also distinguishes our method from others that rely on parametric models of neural activity . These requirements motivated our use of ISIs to characterize neuronal activity . For non-classically responsive cells with PSTHs that displayed no systematic changes over trials or between task conditions , the ISI distributions can be variable . The ISI defines spike timing relative to the previous spike and thus does not require reference to an external task variable such as tone onset or behavioral response . In modeling the distribution of ISIs , we use a non-parametric Kernel Density Estimator that avoids assumptions about whether or not firing occurs according to a Poisson ( or another ) parameterized distribution . We used 10-fold cross validation to estimate the bandwidth of the Gaussian kernel in a data-driven manner . Finally , the use of the ISI was also motivated by previous work demonstrating that the ISI can encode sensory information ( Lundstrom and Fairhall , 2006; Reich et al . , 2000; Zuo et al . , 2015 ) and that precise spike timing has been shown to be important for sensory processing in rat auditory cortex ( DeWeese et al . , 2003; Lu and Wang , 2004 ) . Our data-driven method combines ( 1 ) non-parametric statistical procedures ( Kernel Density Estimation ) , ( 2 ) use of the ISI as the response variable of interest ( rather than an estimate of the instantaneous firing rate locked to an external task variable ) , and ( 3 ) single-trial decoding via Bayesian inference rendering it a novel decoder capable of decoding responsive as well as non-classically responsive activity from any brain region . To test the null hypothesis that the ISI-based single-trial Bayesian decoder performance was indistinguishable from chance , synthetic spike trains were constructed for each trial of a given unit by randomly sampling with replacement from the set of all observed ISIs regardless of the original task variable values ( synthetic spike trains , Figure 4E ) . In principle under this condition , ISIs should no longer bear any relationship to the task variables and decoding performance should be close to 50% . For single-unit responses , this randomization was completed 1240 times . Significance from the null was assessed by a direct comparison to the 124 bootstrapped values observed from the true data to the 1240 values observed under the null hypotheses . The p-value was determined as the probability of finding a value from this synthetic condition that produced better decoding performance than the values actually observed as in a standard permutation test . As a secondary control , we used a traditional permutation test whereby observed spike trains were left intact , but the task variables that correspond to each spike train were randomly permuted ( condition permutation , Figure 4F ) . This process was completed 1240 times . To decode using the trial-averaged firing rate , we implemented a standard method ( Rieke et al . , 1999 ) which uses the probability of observing a set of n spikes at times t1 , … , tn assuming those spikes were generated by a rate-modulated Poisson process ( Figure 4—figure supplement 4 ) . Just as with this ISI-based decoder , we decoded activity from the entire trial . First , we use a training set comprising 90% of trials to estimate the time-varying firing rate for each condition from the PSTH ( rtargett , rnon-targett , rgot , rno-got ) by Kernel Density Estimation with 10-fold cross-validation . The remaining 10% of spike trains are then decoded using the probability of observing each spike train on each condition assuming they were generated according to a rate-modulated Poisson process:pti target ) =1N ! ( rtargett1 rtargett2…rtargettn ) exp-∫TiTfrtargett dt , where Ti and Tf are the beginning and end of the trial respectively . This likelihood function is straightforward to interpret: the first product is the probability of observing spikes the spikes at the times they were observed ( where the 1/N ! term serves to divide out by the number of permutations of spike labels ) and the exponential term represents the probability of silence in the periods between spikes . For comparison with our method , we can reformulate this equation using interspike intervals , if we first break up the exponential integral into domains that span the observed interspike intervals . p ( {ti}|target ) =1N ! ( rtarget ( t1 ) exp ( −∫Tit1rtarget ( t ) dt ) ) × ( rtarget ( t2 ) exp ( −∫t1t2rtarget ( t ) dt ) ) …× ( exp ( −∫tnTfrtarget ( t ) dt ) ) . Collecting the first and last terms relating to trial start and trial end asLi ( t1 , Ti ) ≡rtargett1exp-∫Tit1rtargett dtLftn , Tf≡exp-∫tnTfrtargett dt , this becomespti target ) =1N ! Li∏i=1n-1rtargetti+Δti exp-∫titi+Δtirtargett dtLf , where Δti is the time difference between spikes ti and ti+1 . The interpretation of each term in the product is straightforward: it is the infinitesimal probability of observing a spike a time Δt after a spike at time t multiplied by the probability of observing no spikes in the intervening time . In other words , it is simply pISI target , t ) , the probability of observing an ISI conditioned on observing the first spike at time t , as predicted by the assumption of a rate-modulated Poisson process . We can easily verify that this term is normalized which allows us to write , pISI target , t ) =rtargett+ISI exp-∫tt+ISIrtargett dt . With the exception of the terms relating to trial start and end , we can then view the likelihood of a spike train as resulting from the likelihood of the individual ISIs ( just as with our ISI-decoder ) , pti target ) =1N ! Li Lf∏i=1n-1pISIi target , ti ) , with the key difference that these ISI probabilities are inferred from the firing rate rather than estimated directly using non-parametric methods . To compare the ISI distribution inferred using non-parametric methods to one predicted by a rate-modulated Poisson process , we use the relationship above to calculate the predicted probability of observing an ISI of given length within the 1 s window used for our non-parametric estimates . The formula above assumes a spike has already occurred at time t , so we multiply by the probability of observing a spike at time t , p ( t | target ) =rtarget ( t ) , to obtain the total probability of finding an ISI at any given point in the trial . p ( ISI , t| target ) =p ( ISI | target , t ) p ( t | target ) = rtarget ( t ) rtarget ( t+ISI ) exp ( −∫tt+ISIrtarget ( t ) dt ) . In other words , the probability of observing an ISI beginning at time t is simply the probability of observing spikes at times t and t + ISI with silence in between . The probability of observing an ISI at any time within a time window spanning wi to wf is simply the integral of this ISI probability as a function of time across the window . To ensure the final spike occurs before wf the integral spans wi to ( wf - ISI ) , p ( ISI|wI , wf , target ) =C−1∫wiwf−ISIp ( ISI , t|target ) dtwhere C is a normalization constant which ensures p ( ISI | wi , wf , target ) integrates to 1 , C=∫0wf−wi ( ∫wiwf−ISIp ( ISI , t|target ) dt ) dISI . For each cell , we fit a Logit model for both the stimulus and choice decoding probabilities on individual trials with the true stimulus category and behavioral choice as regressors . We then calculated the extent to which the stimulus decoding probability was determined by true stimulus category by subtracting the regression coefficient for stimulus from that of choice ( Figure 4—figure supplement 3A , x-axis , stimulus selectivity index ) ; when this number is positive it indicates that stimulus was a stronger predictor of stimulus decoding on a trial-by-trial basis . The same process was repeated for choice ( Figure 4—figure supplement 3A , y-axis , choice selectivity index ) . According to this analysis , we took multiplexed cells to be those that were positive for both measures ( Figure 4—figure supplement 3A , orange symbols , 19/90 cells ) . In other words , multiplexed cells were cells for which stimulus decoding probabilities were primarily a result of true stimulus category and choice decoding probabilities were primarily a result of true behavioral choice . Given the moderate negative correlation for these indices , we projected each of these points onto their linear regression to create a one-dimensional regression-based uniplexing index . Cells with a value near zero are the multiplexed cells described above and cells with positive or negative values are primarily stimulus or choice selective ( Figure 4—figure supplement 3A ) . We compared the uniplexing values produced by this regression method to those produced by examining only the average decoding performance for stimulus and choice ( Figure 4—figure supplement 3B ) . A decoding-based uniplexing index was defined as the difference between average stimulus and choice decoding for each cell . When these two values are comparable , this measure returns a value close to zero and the cell is considered multiplexed; moreover , cells that are uniplexed for stimulus or choice receive positive and negative values , respectively , just as with the regression-based measure . While the overall magnitude of these two measures need not be related , both measures of multi/uniplexing rank cells on a one-dimensional axis from choice uniplexed to multiplexed to stimulus uniplexed centered on zero . The log likelihood ratio ( LLR ) was calculated by first calculating the conditional ISI probabilities and then taking the difference of the logarithm of these distributions . For stimulus , LLRstimulusISI= log2pISItarget-log2pISInon-target , and for choice , LLRchoiceISI= log2pISIgo-log2pISIno-go . The weighted LLR weights the LLR according to the prevalence of a given ISI . For stimulus , W . LLRstimulusISI= p ( ISI ) log2pISItarget-log2pISInon-target , and for choice , W . LLRchoiceISI= p ( ISI ) log2pISIgo-log2pISIno-go . The consensus value evaluates the extent to which the LLR ( or weighted LLR ) is shared across an ensemble . It is the norm of the sum of the LLRs ( W . LLRs ) divided by the sum of the norms . In principle , the functional norm can be anything but in this case we used the l1 norm ( the absolute area under the curve ) , f1≡∫f ( x ) dx . The for an n-member ensemble , the consensus is thenConsensus≡‖∑i=1nLLRi‖1∑i=1n‖LLRi‖1 . For the unsigned consensus , we first generate every permutation of the LLRs used and their inverses , -LLR , up to an overall sign . For example , for a pair of LLRs there are only two options , ( LLR1 , LLR2 ) or ( LLR1 , −LLR2 ) , and for three LLRs there are four options , ( LLR1 , LLR2 , LLR3 ) , ( −LLR1 , LLR2 , LLR3 ) , ( LLR1 , −LLR2 , LLR3 ) , or ( LLR1 , LLR2 , −LLR3 ) The consensus is then calculated over each these sets and the maximum value is taken to be the value of the unsigned consensus . To generate the consensus curves in Figure 8 , LLRs are calculated using a 750 ms sliding window recalculated every 100 ms . The resulting consensus value is assigned to the center of the 750 ms window . For visual clarity , these values were interpolated by a third-degree univariate spline calculated using the python package scipy . interpolate . InterpolatedUnivariateSpline ( this technique is guaranteed to intercept the measured values ) . Using our novel ISI-based decoding algorithm , we analyzed cells found to be non-classically responsive in a previously published study ( Rodgers and DeWeese , 2014 ) . Briefly , rats were trained on a novel auditory stimulus selection task where animals had to respond to one of two cues while ignoring the other depending on the context . Rats held their nose in a center port for 250 to 350 ms and were then presented with two simultaneous sounds ( a white noise burst played from only the left or right speaker and a high- or low-pitched warble played from both speakers ) . In the ‘localization’ context , animals were trained to ignore the warble and respond to the location of the white noise burst and in the ‘pitch’ context they were trained to ignore the location of the white noise burst and respond to the pitch of the warble . Cells recorded from both primary auditory cortex and prefrontal cortex ( prelimbic region ) were shown to be classically responsive to the selection rule during the pre-stimulus period ( i . e . firing rates differed between the two contexts ) . Non-classically responsive cells were reported but not further analyzed . We established that cells were non-classically responsive for the stimulus location or pitch using our own positive statistical criteria for non-classical responsiveness ( described above ) by comparing the average spiking activity in the 250 ms stimulus period and the 250 ms following stimulus to inter-trial baseline activity . Cells were also determined to be non-classically responsive for ramping using the same criteria as with our own data . We confirmed that cells were non-classically responsive for the selection rule by comparing their average spiking activity in the 100 ms immediately preceding stimulus onset across contexts . To determine whether non-classically responsive cells also encoded task information ( stimulus location , stimulus pitch , behavioral choice , and the selection rule ) , we decoded each variable on single-trials using our ISI-based decoding algorithm . Selection rule information was only assessed in the pre-stimulus hold period , whereas stimulus and choice information was assessed in the period after stimulus onset prior to behavioral response ( as with our own data ) . Cells shown in Figure 5B were deemed statistically significant when compared to the decoding performance of a control using synthetically generated data ( p<0 . 05 ) . All statistical analyses were performed in Python , MATLAB , or GraphPad Prism 6 . Datasets were tested for normality , and appropriate statistical tests applied as described in the text ( e . g . Student’s paired t-test for normally distributed data , Mann-Whitney U test for unmatched non-parametric data , and Wilcoxon matched-pairs signed rank test for matched non-parametric data ) . https://github . com/badralbanna/Insanally2017 ( Albanna , 2019; copy archived at https://github . com/elifesciences-publications/Insanally2017 ) . | Neurons encode information in the form of electrical signals called spikes . Certain neurons increase the rate at which they produce spikes under specific circumstances , e . g . , whenever an animal hears a particular sound . These neurons are said to be 'classically responsive' . But not all neurons behave in this way . Others produce spikes at a variable rate that does not obviously relate to the animal's behavior . These neurons are said to be 'non-classically responsive' . They are often omitted from analyses , despite typically outnumbering their classically responsive counterparts . So , what are these neurons doing ? To find out , Insanally et al . trained rats to respond to sounds . The animals learned to poke their nose into a window whenever they heard a specific tone , and to avoid responding whenever they heard any other tone . As the rats performed the task , Insanally et al . recorded from neurons in two areas of the brain , the frontal cortex and the auditory cortex . A computer then analyzed the activity of individual neurons during each trial . As expected , the firing rate of non-classically responsive cells did not relate to the animals' behavior . But the timing of this firing did . The interval between spikes contained information about which tone the animals had heard and/or how they had responded . The cells worked together in groups to encode this information . Over the course of each trial , every neuron in the group varied the interval between its spikes . Eventually , the group reached a consensus , with all neurons using the same interval to represent information relevant to the task . Groups of neurons in the frontal cortex encoded more information about the category of the tone than those in the auditory cortex . By including all neurons – both classically and non-classically responsive – this model offers a more comprehensive view of how neural activity relates to behavior . This may in turn help us understand the variable and complex neural activity seen in people with sensory and cognitive disorders . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"computational",
"and",
"systems",
"biology",
"neuroscience"
] | 2019 | Spike-timing-dependent ensemble encoding by non-classically responsive cortical neurons |
Neurogenesis in Drosophila occurs in two phases , embryonic and post-embryonic , in which the same set of neuroblasts give rise to the distinct larval and adult nervous systems , respectively . Here , we identified the embryonic neuroblast origin of the adult neuronal lineages in the ventral nervous system via lineage-specific GAL4 lines and molecular markers . Our lineage mapping revealed that neurons born late in the embryonic phase show axonal morphology and transcription factor profiles that are similar to the neurons born post-embryonically from the same neuroblast . Moreover , we identified three thorax-specific neuroblasts not previously characterized and show that HOX genes confine them to the thoracic segments . Two of these , NB2-3 and NB3-4 , generate leg motor neurons . The other neuroblast is novel and appears to have arisen recently during insect evolution . Our findings provide a comprehensive view of neurogenesis and show how proliferation of individual neuroblasts is dictated by temporal and spatial cues .
The embryonic ventral nerve cord ( VNC ) of Drosophila has been used as a model system for over three decades to understand how a small number of neuronal stem cells , called neuroblasts ( NBs ) , generate a highly complex but organized tissue in which almost all cells adopt unique fates ( Jimenez and Campos-Ortega , 1979; Cabrera et al . , 1987; Doe , CQ 1992; Skeath and Carroll , 1992; Bossing et al . , 1996; Schmidt et al . , 1997; 1999; Rickert et al . , 2011 ) . Patterning of the neural ectoderm is the first step in promoting neuronal diversity . The orthogonal interaction of segment-polarity genes [e . g . , runt ( run ) , wingless ( wg ) and gooseberry ( gsb ) ] and columnar genes [e . g . , ventral nervous system defective ( vnd ) , intermediate neuroblasts defective ( ind ) , and muscle specific homeobox ( msh; also referred to as Drop ) ] divides the neuroectoderm into a Cartesian grid system , in which each NB acquires a unique identity based on its position within the grid ( reviewed in Skeath , 1999 ) . About 30 distinct NBs form in a segmentally repeated bilateral pattern through most of the VNC segments , although the number of NBs is reduced in the anterior gnathal and terminal abdominal segments ( Bossing et al . , 1996; Schmidt et al . , 1997; 1999; Technau et al . , 2014; Birkholz et al . , 2013 ) . Each NB undergoes multiple rounds of asymmetric cell division . During each division it renews itself and generates a secondary precursor cell , called a ganglion mother cell ( GMC ) , which terminally divides to generate a pair of neurons or glia ( Campos-Ortega , 1993; Goodman and Doe , 1993; Rhyu et al . , 1994; Spana et al . , 1995 ) . Through successive cell divisions , the number of which depends on the NB identity , each NB produces unique and highly diverse progeny ( Bossing et al . , 1996; Schmidt et al . , 1997; 1999 ) . Recent studies have shown that many NBs in the embryonic VNC undergo the following temporal changes of the transcription factor expression: Hunchback → Kruppel→ Pdm→ Castor ( Kambadur et al . , 1998; Brody and Odenwald 2000; Isshiki et al . , 2001; Pearson and Doe 2003; Grosskortenhaus et al . 2005 ) . Each of these factors defines a temporal identity window for the NB , and each is maintained in the GMC , establishing different transcriptional states . The GMC then divides via Notch-mediated asymmetric cell division to produce two sibling cells with distinct identities: the Notch-ON “A” cell and the Notch-OFF “B” cell ( reviewed in Jan and Jan , 2000 ) . Consequently , diversity within a NB lineage is produced through two main mechanisms: transcriptional changes in the NB that occur as the stem cell divides and Notch mediated asymmetric cell fates of the daughters of the GMC . Towards the end of embryogenesis , most NBs in the thoracic and gnathal segments enter a mitotically quiescent state , whereas most NBs in the abdominal segments and a few in the thoracic segments die through apoptosis ( Peterson et al . , 2002; Cenci and Gould , 2005; Baumgardt et al . , 2009 ) . The quiescent NBs re-enter the cell cycle at the beginning of the second larval instar stage and continue to generate progeny from the Castor ( Cas ) window ( Tsuji et al . , 2008; Maurange et al . , 2008 ) . This quiescent state divides the neurogenesis of Drosophila , and other insects that undergo complete metamorphosis , into two phases: an embryonic phase , which generates the neurons of the larval nervous system , and a postembryonic phase , which generates adult-specific neurons . Because of the NB quiescence and the anatomical changes that occur in the CNS during late embryogenesis , it has been difficult to establish the correspondence between the embryonic and postembryonic lineages . Recently , using the technique of Flybow , Birkholz et al . ( 2015 ) reported the correspondence between the embryonic and postembryonic lineages . We attempted a similar linkage using a suite of genetic and molecular tools to identify individual NB lineages in the embryo and then used these tools to bridge the postembryonic lineages to their embryonic origins . While we have concordance with most of the findings of Birkholz et al . ( 2015 ) , we differ on eight of the lineages . We also find that within a lineage , the postembryonically born neurons show significant similarities to neurons that are born in the embryonic Cas window in terms of axonal projection and transcription-factor expression . Moreover , our findings complete previous work on identifying embryonic and postembryonic progeny of NBs by characterizing thorax-specific NBs , NB2-3 and NB3-4 , which produce leg motor neurons and identifying a novel NB , NB5-7 . Our complete lineage map , and the reagents we generated to follow individual lineages throughout development , lay the groundwork for investigating how neural patterning and NB identity in the embryonic CNS direct the formation of neural circuits in the adult .
We linked the embryonic NBs to their postembryonic progeny via a recently developed technique , which irreversibly marks the complete progeny of a NB after the onset of GAL4 expression ( Figure 1A , B; Awasaki et al . , 2014 ) . To use this technique ( which we call here “reporter immortalization” ) , we visually screened publicly available databases ( Manning et al . , 2012; Kvon et al . , 2014 ) to identify GAL4 lines whose reporter expression is restricted to one or a few embryonic NBs . We identified over 100 such GAL4 lines . Only a few of them marked an individual NB , while most marked a few NBs with or without their progeny ( Table 1 ) . To identify which NB lineages are marked by these GAL4 lines , we generated random lineage clones for each line ( Lacin et al . , 2009; Nern et al . , 2015 ) and compared their morphology and molecular-marker expression to previously published embryonic neuronal lineages ( Bossing et al . , 1996; Schmidt et al . , 1997; 1999; Birkholz et al . , 2013 and references therein ) . With this information , we also intersected split GAL4 combinations to restrict overlapping expression patterns of different drivers to individual NBs and their progeny . 10 . 7554/eLife . 13399 . 003Figure 1 . Tracing individual NB lineages identifies postembryonic progeny of NBs . ( A ) Schematic illustration of the strategies used in this study to trace NB lineages . Since expression of GAL4 lines is usually not maintained throughout the development , GAL4 expression was immortalized in progeny of the NBs to identify their postembryonic progeny . The “reporter immortalization” technique requires several steps of transcriptions and recombinations ( Awasaki et al . , 2014 ) . Thus , cells that are born from initial divisions ( marked by asterisk ) after the GAL4 presence are not labeled with this technique . ( B ) Schematic representations of embryonic NBs ( left ) and their corresponding postembryonic lineages ( right ) for the T2 segment shown . 30 bilaterally symmetric NBs and 1 medial NBs generate 26 postembryonic lineages . For segment specific differences , see Figure 1—figure supplement 1 . Row identity of NBs shown in a color code . Three thoracic specific NB lineages are outlined with a thick line . Thick crosses depict NBs , which are eliminated by apoptosis; thin crosses depict NBs , which are present at early stage embryos , but not detected at stage 17 embryos . Dashed line indicates the midline . DOI: http://dx . doi . org/10 . 7554/eLife . 13399 . 00310 . 7554/eLife . 13399 . 004Figure 1—figure supplement 1 . Schematic representations of postembryonic lineages in different segments of the nerve cord shown . T2 and T3 segments virtually have the same lineages with the exception of lineage 11 ( asterisk ) , which is dramatically reduced in cell number in the T3 segment . Among A2-A7 segments , NB6-2 survives postembryonically and gets reactivated only in the A2 segment ( Birkholz et al . , 2015 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 13399 . 00410 . 7554/eLife . 13399 . 005Figure 2 . Sample GAL4 lines that mark NBs and their progeny . ( A-K ) Expression patterns of selected GAL4 lines in the nerve cords of late-stage embryos were visualized by driving mCD8-GFP ( green ) . Only the T2-A2 segments are shown . ( A-H ) GAL4 lines uniquely label individual NBs and a subset of their progeny: NB1-2 ( A ) , NB2-1 ( B ) , NB3-5 ( C ) , NB4-2 ( D ) , NB6-1 ( E ) , NB6-2 ( F ) , NB7-2 ( G ) , and NB7-4 ( H ) . Although NB4-2 generates progeny in both thoracic and abdominal segments , R81C12AD-R42F01DBD marks the NB4-2 lineage only in thoracic segments ( D ) . ( I-K ) Expression of GAL4 lines that sparsely label a few NBs and their progeny . Color-coded arrowheads indicate the location of the NB lineages . See Figures 2–4 for the presence of NBs , which are revealed by Dpn staining in some of these GAL4 lines . FasII+BP102 ( blue ) visualizes embryonic neuronal architecture; anterior is up . DOI: http://dx . doi . org/10 . 7554/eLife . 13399 . 00510 . 7554/eLife . 13399 . 006Table 1 . GAL4 lines used to mark NBs and for reporter immortalization . DOI: http://dx . doi . org/10 . 7554/eLife . 13399 . 006LineNB expressionImmortalization*FigureR16A05AD-R28H10DBDNB1-2lin1 ( 30/30 ) Figure 2 , Figure 4R70D06AD-R28H10DBDNB2-1lin 2 ( 18/ 30 ) , lin10 ( 7/ 30 ) Figure 2 , Figure 4—figure supplement 1R19H09AD-R28H10DBDNB2-2lin10 ( 16/30 ) , lin7 ( 4/30 ) Figure 4—figure supplement 1R10C12NB2-3lin15 ( 30/30 ) , lin25 ( 7/30 ) Figure 5R65G02NB2-4lin 18 ( 21/30 ) , lin20-22 ( 9/30 ) , lin5 ( 5/30 ) Figure 35172J-Gal4NB3-1lin4 ( 12/18 ) Figure 4—figure supplement 1R21E09AD-R16H11DBDNB3-2 , NB4-2lin7 ( 18/24 ) , lin13 ( 8/24 ) Figure 4 , Figure 4—figure supplement 1R21E09AD-R28H10DBDNB1-2 , NB3-2lin1 ( 15/24 ) , lin7 ( 8/24 ) ems-Gal4NB2-2 , NB3-3 , NB3-5lin10 ( 10/30 ) , lin8 ( 15/30 ) , lin9 ( 20/30 ) Figure 3 , Figure 4—figure supplement 1R59E09NB3-5lin9 ( 30/30 ) Figure 2 , Figure 3R77B09AD-R28H10DBDNB2-3 , NB3-4 , NB2-1lin15 ( 30/30 ) , lin25 ( 24/30 ) ; lin2 ( 20/30 ) Figure 1 , Figure 4VT0048571NB4-1 , NB7-2lin14 ( 34/36 ) , lin11 ( 13/24 ) Figure 4—figure supplement 1R81C12AD-R42F01DBDNB4-2Lin13 ( 30/30 ) Figure 2 , Figure 4R19B03AD-R16H11DBDNB4-3 , NB3-4 , NB2-3lin21 ( 30/30 ) , lin25 ( 19/30 ) , lin15 ( 8/30 ) , Figure 2 , Figure 4—figure supplement 1VT0041296NB4-4Lin24 ( 20/24 ) ; lin18 ( 5/18 ) Figure 4—figure supplement 1R54B10NB5-3 , NB5-6lin5 ( 17/30 ) Figure 4—figure supplement 1R19B03AD-R45D04DBDNB5-4 , NB5-7 , NB2-3lin20 ( 30/30 ) , lin22 ( 30/30 ) , lin15 ( 10/30 ) Figure 2 , Figure 6R24C10NB5-7lin20 ( 23/30 ) , lin18 ( 6/24 ) Figure 6R81F01NB6-1lin12 ( 13/30 ) , lin18 ( 5/30 ) R70D06AD-R42F01DBDNB6-1lin12 ( 12/24 ) , lin13 ( 7/24 ) Figure 2 , Figure 4R76D11NB6-2lin19 ( 16/30 ) Figure 2; Figure 3R51B04NB7-1 , NB6-2lin3 ( 22/30 ) , lin19 ( 5/30 ) Figure 4—figure supplement 1R35B12AD-R28H10DBDNB7-2lin11 ( 12/30 ) Figure 2 , Figure 4R35B12NB7-1 , NB6-2 , NB7-2lin3 ( 10/30 ) , lin19 ( 6/30 ) , lin11 ( 8/24 ) R19B03AD-R18F07DBDNB7-4lin23 ( 26/30 ) Figure 2 , Figure 4—figure supplement 1R19B03**NB2-5 , NB2-4lin17 ( 20/24 ) , lin18 ( 26/30 ) Figure 3R13G03MNBlin0 ( 10/18 ) Figure 4—figure supplement 1lbe-K-GAL4***NB5-6lin5-6 ( 20/20 ) Figure 8R45D04NB5-2 , NB5-3 , NB5-4 , NB5-7 , NB6-2lin6 ( 30/30 ) , lin5 ( 30/30 ) , lin20 ( 30/30 ) , lin22 ( 30/30 ) , lin19 ( 18/30 ) eg-GAL4NB2-4 , NB3-3 , NB3-4 , NB6-4 , NB7-3lin18 ( 19/30 ) , lin8 ( 12/30 ) lin25 ( 16/30 ) Figure 5 , Figure 4—figure supplement 1* The number of immortalized lineage per hemisegment shown in paranthesis . Corresponding NBs and lineages are color matched . ** Only dorsal part VNC scored***NB5-6 generate postembryonic progeny only in S3 segments Lineages marked less than 15% of the time are not included . “lin” refers to postembryonic lineage . Ultimately , the GAL4 and split-GAL4 lines characterized during this study mark 28 out of 31 previously documented NBs individually ( 12 NBs ) or in combination with a small number of other lineages ( Table 1 , examples in Figure 2 ) . Many lines also drive reporter expression in the progeny of the marked NB , and their expression pattern is maintained into early larval stages . Due to the large number of lineages and the repetitive nature of the lineage-tracing method , we will discuss only a few NBs in detail to illustrate how we identified their postembryonic progeny . We will also focus on where our results differ from Birkholz et al . ( 2015 ) . All of the driver lines and molecular markers that were used to link NBs to their postembryonic progeny can be found in Tables 1 and 2 . 10 . 7554/eLife . 13399 . 007Table 2 . Expression profile of transcription factors in NBs and their corresponding embryonic and postembryonic progeny . DOI: http://dx . doi . org/10 . 7554/eLife . 13399 . 007NBslinEmbryonic progeny*Postembryonic lin**NB Marker***MNBlin 0En , FoxD , VgEnEn , unpg-LaczNB1-1lin16Hb9 , Lim3 , Isl , D , Eve , Hb9 , Lim3mirr-LaczNB1-2lin1Msh , Nmr1 , Hb9 , Nkx6Nmr1 , Mshmirr-LaczNB2-1lin2Toymirr-LaczNB2-2lin10Hb9 , Lim3 , Nkx6 , Hb9 , Lim3 , Nkx6mirr-Lacz , RunNB2-3lin15Lim3 , Nkx6 , Isl , Isl , Lim3 , Nkx6mirr-Lacz , Msh , RunNB2-4lin18Unc-4 , Eg , Toy , MshUnc-4mirr-Lacz , MshNB2-5lin17Unc-4 , IslUnc-4 , Islmirr-LaczNB3-1lin4Hb9 , MshHb9Nkx6 , RunNB3-2lin7Hb9 , Toy , Barh , Unc-4Unc-4Ey , DbxNB3-3lin8Toy , Lim3 , Ems , Acj6 , Eg , EveLim3 , Ems , Acj6 , Toy , EyEms , RunNB3-4lin25Toy , Ey , Msh , EgToy , Nkx6Msh , Run , Ey , Eg-Gal4NB3-5lin9Ems , Msh , Islet , Unc-4Ems , Msh , IsletEmsNB4-1lin14Msh , Unc-4Mshunpg-LaczNB4-2lin13Dbx , D , Vg , Ey , EveDbx , D , VgEyNB4-3lin21Msh , EyMsh , EyEy , MshNB4-4lin24Ems , ToyEms , ToyEms , Ey , NB5-1-gsb-LaczNB5-2lin6Toy , En , Vg , Hb9Toy , En , Vggsb-Lacz , RunNB5-3lin5Vg , Toy , Ey , EnVg , Toy , gsb-Lacz , Ey , RunNB5-4lin22BarHBarHgsb-Lacz , MshNB5-5-unpg-LaczNB5-6-EyA , ToyEyA ( S3 segments ) gsb-Lacz , lbe-Gal4NB5-7lin20-gsb-Lacz , MshNB6-1lin12Unc-4 , Nmr1 , DbxDbx , Unc-4 , Nmr1gsb-Lacz , En , DbxNB6-2lin19Unc-4 , DbxDbx , Unc-4gsb-Lacz , En , DNB6-4-Eg , Toy , Mshgsb-Lacz , Eg-laczNB7-1lin3Unc-4 , Dbx , EveDbx , Nkx6 , gsb-Lacz , EnNB7-2lin11Unc-4 , Nkx6Unc-4 , Nkx6 , EveEn , unpg-Lacz , DbxNB7-3-Hb9 , Isl , Eg , EyEn , eg-GAL4NB7-4lin23Unc-4 , Acj6Unc-4 , Acj6En , Msh , D* Transcription factors that are also expressed in the postembryonic neurons are highlighted in bold . ** We failed to detect transcription factors in red in the corresponding embryonic progeny . *** Expression of these markers is maintained from embryonic to postembryonic stages . “lin” refers to postembryonic lineage . The NBs of insects with incomplete metamorphosis , like grasshoppers , generate all of their progeny during embryogenesis ( Bate , 1976; Shepherd and Laurent , 1992 ) . With the evolution of metamorphosis , this single phase of neurogenesis was split into two , with the embryonic phase producing the neurons of the larva and the postembryonic phase dedicated to making adult neurons . We wanted to know if NB arrest late in embryogenesis simply interrupts the temporal progression of neural types or if the arrest somehow “reprograms” the NB so that cell classes produced after the arrest bear no resemblance to those produced before . The segmental NB array in Drosophila embryos has been the focus of many studies to investigate neurogenesis . Individual NBs have been identified based on their large size and location in the array , by molecular marker expression including genetic handles such as lacZ and GAL4 lines , or by the morphology of their progeny . Only a few NBs have been studied comprehensively including all these features [e . g . , NB5-6 ( Baumgardt et al . , 2009 ) . Here , we have used a suite of molecular markers and a library of GAL4 lines , many of which are specific to individual NBs , to characterize the entire set of thoracic NBs , including their embryonic and postembryonic progeny . We found that 26 of 30 ( +1 medial ) thoracic NBs survive into larval stages to generate neurons for the adult nervous system . We found that three of these NBs appear to be confined to the thorax , and two of them ( NB2-3 and NB3-4 ) are dedicated to producing motor neurons for the leg . The third NB , NB5-7 , had not been previously described and is unique in that it appears to have only a postembryonic neurogenic phase . Although the segmental NB array has been highly conserved through insect evolution ( Thomas et al . , 1984; Truman and Bate , 1988 ) , we think that NB5-7 is a recent addition , likely by a duplication of NB5-4 . This two-phase pattern of neurogenesis evolved from a simpler scheme , such as seen in grasshoppers , in which all neurons were generated during an extended embryonic phase ( Bate 1976; Shepherd and Bate , 1990 ) . The similarity of the neurons just before and after quiescence in flies suggests that the insect NBs dealt with metamorphosis by simply suspending their ongoing fate determination through the arrest period and then resuming from that point once neurogenesis was restarted . In other words , metamorphosis did not likely require the resetting of a fate determination clock on the development of a novel postembryonic system .
We used the following fly strains with indicated genotypes during this study Canton-S as wild type , gsbn-lacZ . 4z1 ( Li and Noll , 1994 ) , lbe ( K ) -GAL4 ( Baumgardt et al . , 2009 ) , mirror-lacZ ( Broadus et al . , 1995 ) , unpg-lacZ ( Doe , CQ 1992 ) , 5172J-GAL4 ( Lacin et al . , 2014a ) , elav-GAL4 ( DiAntonio et al . , 2001 ) , en- GAL4 ( Brand and Perrimon , 1993 ) , ems-GAL4 ( Estacio-Gomez et al . , 2013 ) , eg-kinesin-lacZ ( Higashijima et al . , 1996 ) , eg-GAL4 ( Ito et al . , 1995 ) , antp25 ( Abbott and Kaufman , 1986 ) , Df ( 3L ) H99 ( Abbott and Lengyel , 1991 ) , Df ( 3R ) Ubx109 ( Lewis , 1980 ) , cas24/TM6b ( Cui and Doe , 1992 ) , pJFRC19-13XLexAop2-IVS-myr::GFP , pJFRC7-20XUAS-IVS-mCD8::GFP ( Pfeiffer et al . 2010 ) . Cell death was detected by TUNEL labeling ( Roche , in situ cell death detection kit , TMR red ) . Embryos were first labeled with primary and secondary antibodies , and then treated with proteinase K ( 2 ug/ml ) for 5 min at 37°C . TMR labeling was performed according to manufacturer’s instructions . To mark all cell born postembryonically , newly hatched first-instar larvae were fed yeast paste containing 300 μM EdU . At the wandering larval stage , larvae were dissected , fixed and stained as described previously ( Lacin et al . , 2014b ) . Tissue fixation and staining were performed as described by Patel ( 1994 ) . To visualize major axonal tracts we used AMCA-conjugated HRP or combination of BP102 and FasII in embryos , Alexa Fluor 568-conjugated Phalloidin or BP104 in larvae , and NC82 in adult flies . Secondary antibodies were obtained from Jackson Immunoresearch and Life Technologies and used at 1/200 dilution . Tissues were mounted in Vectashield ( Vector labs ) . To get better resolution , some samples were cleared with xylene and mounted in DPX ( Truman et al . , 2010 ) . The following antibodies were used at indicated dilutions: Rabbit Runt ( 1/1000; E . Wieschaus ) , rabbit Vg ( 1/50; Williams et al . , 1991 ) , guinea pig Toy ( 1/500; U . Walldorf ) , guinea pig Ems ( 1/300; U . Walldorf ) , rabbit Msh ( 1/1000; Isshiki et al . , 1997 ) , rabbit Ey ( 1/1000; U . Walldorf ) , guinea pig Dbx ( 1/1000; Lacin et al . , 2009 ) , rabbit Hb9 ( 1/1000; Broihier and Skeath , 2002 ) , guinea pig Hb9 ( 1/1000; Broihier and Skeath , 2002 ) , rat Islet ( 1/500; Broihier and Skeath , 2002 ) , guinea pig Lim3 ( 1/100; Broihier and Skeath , 2002 ) , rat Nkx6 ( 1/500; Broihier et al . , 2004 ) , rabbit Unc-4 ( 1/1000; Lacin et al . , 2014b ) , guinea pig Dpn ( 1/1000; J . Skeath ) , rabbit Dichaete ( 1/1000; Nambu and Nambu , 1996 ) , rabbit Nmr1 ( 1/1000; Leal et al . , 2009 ) , rabbit anti-Eagle ( 1/500; Dittrich et al . , 1997 ) , rabbit Ubx ( 1/500; Marin et al . , 2012 ) chicken GFP ( 1/500; # A10262 , Life Tech . ) , Alexa Fluor 568 Phalloidin ( 1/250; # A12380 , Life Tech . ) , rabbit HA ( 1/500; # 3724S , Cell Sig . ) , rat HA ( 1/500; 3f10 , Roche ) , rat Flag ( 1/200; # NBP1-06712 , Novus B . ) , Chicken V5 ( 1/500; # ab9113 , Abcam ) , mouse B-gal ( 1/1000 ) , rabbit B-gal ( 1/1000; # A11132 , Life Tech . ) , Goat AMCA-HRP ( 1/200; # 123-155-021 , Jackson L . ) . The following mouse primary antibodies were obtained from Developmental Studies Hybridoma Bank: Engrailed-4D9 ( 1/5 ) , BP104 ( 1/40 ) , BP102 ( 1/100 ) , FasII-D4 ( 1/100 ) , NC82 ( 1/100 ) , Antp-8C11 ( 1/20 ) . A Zeiss LSM 710 was used to collect confocal images . Z projections were performed via FIJI ( http://fiji . sc/Fiji ) : To the show presence of NB of interest in the clone , we projected Dpn channel by including only the sections that displayed this NB . Similarly , for channels in which neuronal architecture was shown by markers such HRP or BP102 , we projected only the sections where the major bundles were located . In Figure 10A , a contaminating from a different segment was removed manually . | Fruit flies undergo a process called metamorphosis in which they change from a maggot or larva into an adult fly . These two life stages look and behave differently and appear to have strikingly different nervous systems . The relationship between the two nervous systems has been most extensively studied in the ventral nerve cord ( which is the equivalent to the spinal cord in humans ) . Although the ventral nerve cords of a larva and an adult fly look quite different , they are generated by the same set of stem cells known as neuroblasts . This is made possible because the neuroblasts proliferate in two separate phases: the first phase occurs in the embryo to generate the neurons of the larval nervous system , and the second phase occurs in the larva to generate neurons for the adult’s nervous system . Now , Lacin and Truman have paired each of the neurons in the adult fruit fly’s nerve cord with their corresponding neurons in the nerve cords of fruit fly larvae . This involved identifying the original neuroblasts that gave rise to each of the neurons in both larval and adult fruit flies . The results suggest that most neurons that arise from a given neuroblast produce a similar set of molecules and extend similar nerve fibers , even though they work in two different nervous systems . Since neuroblasts in non-metamorphosing insects proliferate continuously , these findings also suggest that , when metamorphosis evolved , a pause was introduced to create the two separate phases of proliferation without a big effect on the types of neurons generated . Lacin and Truman then went on to discover three neuroblasts that appear to be unique to the middle ( or thoracic ) segments of a fruit fly . The experiments reveal that the presence of these neuroblasts depended on specific genes that control the development of animal body plans . Two of these neuroblasts generate the so-called motor neurons that control the movement of a fly’s legs . Flies only have legs on their thoracic segments , so this indicates that the development of new neurons is coordinated with the development of the body plan at the stem cell level . The third neuroblast generates neurons that connect with the leg motor neurons , and Lacin and Truman propose that this neuroblast arose from a copy of a neighboring stem cell . The resulting extra neurons may have enabled finer control over the leg movements required for activities such as walking and grooming . Following on from this work , it is now possible to investigate how molecular events that occur from the embryonic to the adult stages of a fruit fly’s life control the formation and function of its nervous system . | [
"Abstract",
"Introduction",
"Results",
"and",
"discussion",
"Materials",
"and",
"methods"
] | [
"developmental",
"biology",
"neuroscience"
] | 2016 | Lineage mapping identifies molecular and architectural similarities between the larval and adult Drosophila central nervous system |
The gene regulatory network ( GRN ) is the central decision‐making module of the cell . We have developed a theory called Buffered Qualitative Stability ( BQS ) based on the hypothesis that GRNs are organised so that they remain robust in the face of unpredictable environmental and evolutionary changes . BQS makes strong and diverse predictions about the network features that allow stable responses under arbitrary perturbations , including the random addition of new connections . We show that the GRNs of E . coli , M . tuberculosis , P . aeruginosa , yeast , mouse , and human all verify the predictions of BQS . BQS explains many of the small- and large‐scale properties of GRNs , provides conditions for evolvable robustness , and highlights general features of transcriptional response . BQS is severely compromised in a human cancer cell line , suggesting that loss of BQS might underlie the phenotypic plasticity of cancer cells , and highlighting a possible sequence of GRN alterations concomitant with cancer initiation .
At every level of organisation , biological entities , such as genes , proteins and cells , function as ensembles . Interaction networks are therefore a fundamental feature of biological systems , and a vast amount of analysis exploring the organisation of biological networks has been performed ( Milo et al . , 2002; Barabasi and Oltvai , 2004; Alon , 2006; Buchanan et al . , 2010 ) . This analysis has provided interesting insights into the features of these networks ( Barabasi and Oltvai , 2004; Brock et al . , 2009; Tyson and Novák , 2010; Ferrell et al . , 2011; Liu et al . , 2011; Cowan et al . , 2012 ) , and has led to new methodologies for characterizing their topologies . However , one might argue that this work has had less impact on our understanding of the reasons underlying the network topologies observed and on the possible selective pressures leading to the emergence of common network features . Here we present a simple theory , Buffered Qualitative Stability ( BQS ) , motivated by biological robustness , which has strong explanatory power and provides a number of hard , readily verifiable predictions for the topological structure of interaction networks , at both global and local scales . Besides leading to new predictions—that are consistently verified—BQS provides a theoretical justification for the ubiquitousness of network features already observed . BQS is therefore an important step in providing a general mechanistic explanation for the overall structure of GRNs at different scales and in shedding new light on previous observations . Robustness is a remarkable feature of living organisms allowing them to tolerate a wide variety of contingencies , such as DNA damage , limitations in nutrient availability , or exposure to toxins ( Lopez-Maury et al . , 2008; MacNeil and Walhout , 2011 ) . Although much is now known about how cells respond to particular stresses or environmental cues , little is known about how cells remain stable and respond appropriately whatever the contingency . Over evolutionary time it is also advantageous for organisms to be robust to genetic changes , including those that occur as a consequence of the shuffling of genes during sexual reproduction . In order for cells to be fully robust , changes to any of the thousands of individual quantitative parameters—for example the concentration of a transcription factor or its affinity for its cognate DNA sequence—cannot be critical because contingencies may cause these to change . We propose that the robustness of a biological system should therefore depend on qualitative , not quantitative , features of its response to perturbation . Robustness is a complex and fundamental feature that can be formalised in many ways ( Jen , 2003; Silva-Rocha and de Lorenzo , 2010 ) . Features commonly associated with robustness include resistance to noise , redundancy and error-correction . Here we will focus on an important component of robustness: the ability of a system at equilibrium to respond to a perturbation by returning to its equilibrium state . Such a feature , generally called ‘stability’ , is essential to allow a system to properly operate in noisy conditions and withstand unexpected environmental challenges . This type of robustness has been studied before ( Quirk and Ruppert , 1965; Puccia and Levins , 1985 ) and has been applied to economics ( Quirk and Ruppert , 1965; Hale et al . , 1999 ) , ecology ( May , 1973a; Puccia and Levins , 1985 ) and chemistry ( Tyson , 1975 ) . However , this notion has never been used to predict network features beyond simple topological properties required by the ‘rules’ that allow such stringent robustness and has not previously been applied to molecular cell biology , or the evolutionary pressures shaping the behaviours of living organisms . Transcriptional regulation plays a central role in the behaviour of cells in response to environmental cues and is aberrant in many diseases ( Lee and Young , 2013 ) . Moreover , networks representing transcriptional interactions have been derived for diverse organisms . These networks , termed ‘gene regulatory networks’ ( GRNs ) , comprise directed links between pairs of genes . For a given pair of linked genes , one encodes a transcription factor ( TF ) that regulates the expression of the other ( Buchanan et al . , 2010 ) . Figure 1 shows the GRN of Escherichia coli , with TFs coloured red and the arrows colour-coded according to the number of genes regulated by the source TF . Systematic network analysis of GRNs is possible because comprehensive and high quality GRN datasets are available for different organisms ( Lee et al . , 2002; Harbison et al . , 2004; Luscombe et al . , 2004; MacIsaac et al . , 2006; Galan-Vasquez et al . , 2011; Sanz et al . , 2011; Garber et al . , 2012; Gerstein et al . , 2012; Salgado et al . , 2012 ) . 10 . 7554/eLife . 02863 . 003Figure 1 . The E . coli GRN . The E . coli GRN derived from Salgado et al . ( 2012 ) using two evidence codes . Genes that are reported to regulate transcriptionally at least one other gene , that is transcription factors ( TFs ) , are represented as red circles; the other genes are represented by blue circles . Arrows indicate a transcriptional interaction from the TF to the target gene . The arrows are colour-coded according to the number of genes regulated by the source TF . Note the logarithmic scale in the colour coding . DOI: http://dx . doi . org/10 . 7554/eLife . 02863 . 003 Here we consider the hypothesis that to confer robustness and promote evolvability , GRNs must be stable to changes in interaction parameters and also stable to the addition of new regulatory links , that is to changes in the structure of the GRN itself . The type of robustness that we hypothesize ensures that the transcriptional state of a cell remains largely stable in response to random perturbations . We show that published GRNs , including those of E . coli , M . tuberculosis , P . aeruginosa , S . cerevisiae , mouse and humans are robust in this way , a property we term ‘Buffered Qualitative Stability’ ( BQS ) . Remarkably , the only published GRN of a cancer cell line deviates strongly from BQS , suggesting that the loss of BQS may play an important role in cancer .
Interaction networks are ubiquitous in biology , and robustness in their response to perturbation is a desirable property in many circumstances . A mathematical theory called ‘Qualitative Stability’ has determined how the topological structure of a network is related to robustness ( Quirk and Ruppert , 1965 ) . This theory , discussed in ‘Materials and methods’ , shows that certain network topologies remain stable even if the strength of any of the network interactions is varied in an arbitrary way . Qualitatively Stable GRNs would be robust , for example , to changes in the concentration of a transcription factor or its affinity for its cognate DNA sequence . A primary requirement for Qualitative Stability is the absence of long feedback loops ( meaning , in the case of GRNs , feedback loops involving three or more genes ) regardless of whether the connections comprising the loops are stimulatory or inhibitory . In addition , 2-node feedback loops can be Qualitatively Stable depending on the precise nature of their interactions ( see ‘Materials and methods’ ) . The danger inherent in feedback loops was first analysed by James Clerk Maxwell who showed that mechanical governors regulating the output of steam engines can fail if the input changes faster than the system response , causing ‘an oscillating and jerking motion , increasing in violence till it reaches the limit of action of the governor’ ( Maxwell , 1868 ) . Because there is an inevitable time lag between a transcription factor ( TF ) binding to the promoter of a gene and the production of the protein product of that gene , this form of instability can occur if GRNs contain feedback loops consisting of three or more TFs . This concept is supported by the behaviour of the repressilator , a well-known gene circuit consisting of a 3-gene feedback loop , which has been shown to produce oscillations of increasing intensity in vivo ( Elowitz and Leibler , 2000 ) . We therefore examined the structure of organisms for which system-wide GRNs have been published , including three bacteria—E . coli ( Salgado et al . , 2012 ) , M . tuberculosis ( Sanz et al . , 2011 ) , P . aeruginosa ( Galan-Vasquez et al . , 2011 ) —the yeast S . cerevisiae ( Harbison et al . , 2004 ) , and human ( represented by the GM12878 cell line ) ( Gerstein et al . , 2012 ) . The main Figures present data from E . coli , S . cerevisiae and human , whilst analysis of M . tuberculosis and P . aeruginosa plus additional confidence levels for E . coli and S . cerevisiae , and additional yeast datasets ( Lee et al . , 2002; Luscombe et al . , 2004; MacIsaac et al . , 2006 ) are reported in the Figure Supplements and confirm our main results . A review of available datasets and the rationale for our selection is given in ‘Materials and methods’ . On studying feedback loops in the GRNs of these organisms ( Figure 2A–C , Figure 2—figure supplement 1A , B , lightly shaded bars ) , we find that P . aeurginosa , S . cerevisiae and human GRNs have no feedback loops comprising three or more genes . The E . coli GRN has no feedback loops comprising four or more genes , and only two 3-gene feedback loops . M . tuberculosis has two 3-gene feedback loops and one 4-gene feedback loop . Notably , all the 3-gene feedback loops observed in real GRNs share the same peculiar structure , with implications discussed below . In contrast , when networks of the size and connectivity of the biological GRNs are constructed with randomly placed links , they display an exponential increase in feedback loops consisting of three or more genes , which number in the thousands ( Figure 2A–C , Figure 2—figure supplement 1A , B , heavily shaded bars , and Figure 2—figure supplement 4B ) . Each of 1000 randomly simulated E . coli networks had at least one long feedback loop . The vastly different abundances of feedback loops clearly demonstrate the profound difference in topologies between real and random networks . Statistical analyses suggest that there is an extremely small probability ( <10−6 ) that the absence of long feedback loops with >3 genes in E . coli is a chance event ( Figure 2—figure supplement 2A ) . Similar results hold for S . cerevisiae ( Figure 2—figure supplement 2B ) and human ( Figure 2—figure supplement 2C ) . These results are robust to variations in the confidence levels of the E . coli and S . cerevisiae GRNs ( Figure 2—figure supplement 3E , J ) , despite large variations in other properties of the GRNs ( Figure 2—figure supplement 3A–D , F–I ) , and remain valid when different random models are considered ( Figure 2—figure supplement 4B ) . 10 . 7554/eLife . 02863 . 004Figure 2 . Feedback loops in real and simulated GRNs . Number of feedback loops is provided on a logarithmic scale for E . coli ( A ) , S . cerevisiae ( B ) , and the human GM12878 cell line ( C ) and in each case is compared with a randomly simulated network containing the same number of genes , TFs , and connections . For the random networks , each graph reports the mean and standard deviation . ( D ) Schematic illustrations of feedback loops of length 2–6 are plotted in black . DOI: http://dx . doi . org/10 . 7554/eLife . 02863 . 00410 . 7554/eLife . 02863 . 005Figure 2—figure supplement 1 . Feedback loops in M . tuberculosis , P . aeruginosa and other yeast datasets . Number of feedback loops is provided on a logarithmic scale for M . tuberculosis ( A ) , P . aeruginosa ( B ) , the yeast dataset derived from Lee et al . ( C ) , the yeast dataset derived from Luscombe et al . ( D ) , and the yeast dataset derived from MacIsaac et al . ( E ) . In each case the real dataset is compared with a randomly simulated network containing the same number of genes , TFs and connections . For the random networks , each graph reports the mean and standard deviation . Note that Luscombe et al . ( 2004 ) was mainly concerned with the dynamics of yeast GRN and built the full GRN in such a way to better stress differences under different growth conditions . The dataset was constructed by adding to the network derived by Lee et al . ( 2002 ) new interactions derived under different experimental conditions , and the absence of a measure of confidence for the interactions makes it difficult to assess the strength of the experimental support for the interactions that form the feedback loops . The above considerations suggest that a moderate rate of false positives may be present in the dataset . The sensitivity of loop count to noise in the data therefore suggests that loop counting may not provide a good indication of the role of BQS by itself in this case . Moreover , MacIsaac et al . ( 2006 ) takes a computational approach to the construction of the GRN of S . cerevisiae using the experimental data provided by Harbison et al . ( 2004 ) as a starting point . Therefore , noisy data should be expected and caution should be used in interpreting the data . The network used was the most statistically restrictive . Although a limited number of long feedback loops is present in the dataset ( E ) , comparison with random data indicates that this number is much smaller than expected by chance , still suggesting a selective pressure against instability . Note how the number of long feedback loops is extremely limited for the dataset derived from direct biological experimentation ( A–C ) . DOI: http://dx . doi . org/10 . 7554/eLife . 02863 . 00510 . 7554/eLife . 02863 . 006Figure 2—figure supplement 2 . p-Value estimation for the number of long feedback loops . The theoretical probability density function of the number of feedback loops of a given length is not well characterized . However , calling nloops the number of feedback loops of a given length in a single simulation , the transformation y=log ( nloops+1 ) appears to produce normally distributed data in all the datasets analysed . The normality of the distribution was supported by the Shapiro–Wilk normality test . Using this transformation it is possible to estimate the probability of having no feedback loops of length 14 in a random graph . ( A–C ) . The data for E . coli , yeast and human ( black line ) support quite well the fitted log-normal distribution ( red line ) , as expected from the very low p-value of the Shapiro–Wilk normality test . Analysing the number of loops of a specific length was preferred to performing an analysis combining the information on the number of loops of different length . This choice was made due to the non-independence of the number of loops of different length . DOI: http://dx . doi . org/10 . 7554/eLife . 02863 . 00610 . 7554/eLife . 02863 . 007Figure 2—figure supplement 3 . General properties and number of feedback loops in the RegulonDB ( E . coli ) and Harbison et al . ( yeast ) datasets under different statistical conditions . RegulonDB reports , for each interaction , a list of evidence codes indicating which experimental techniques reported a particular interaction . Some of these codes ( called `strong’ ) are associated with particularly reliable biological techniques ( Salgado et al . , 2012 ) . The information provided by RegulonDB can be used in different ways to assess the statistical validity of our analysis . A first approach is to use the number of evidence codes as an indicator of the confidence level of an interaction: the larger the number of evidence codes , the greater the confidence . A second approach is to restrict the network only to those interactions supported by strong evidence codes . We decided to study the effect of varying the number of general or strong evidence codes on the results discussed . ‘kEc’ indicates the network constructed using only those interactions supported by at least k evidence codes , while ‘kSEc’ indicates the network constructed using only those interactions supported by at least k strong evidence codes . The number of genes ( A ) , TFs ( B ) , and interactions of the networks ( C ) vary with the number of evidence codes . However , the edge density ( D ) is rather stable , suggesting that the global properties of the GRN are mostly preserved in the different types of network . The number of feedback loops is very limited regardless of the number of evidence codes used ( E ) . 2Ec and 2SEc have similar edge density ( D ) and number of illegal feedback loops ( E ) , thus suggesting that taking two evidence codes gives a network similar to the one obtained by considering only strong biological evidence . In the yeast dataset derived by Harbison et al . ( 2004 ) each interaction is associated with a p-value that measures the probability that such an interaction has been detected due to experimental error . Selecting a large threshold p-value increases the probability of false positives , but decreases the probability of false negatives . Lee et al . ( 2002 ) reported that a threshold p-value of 10–3 provides a good trade-off for this kind of data . However , a reliable theory would be expected to display a limited sensitivity to a small variation of the threshold . The number of genes ( F ) , TFs ( G ) , and interactions of the network ( H ) change with different thresholds . Nevertheless , the edge density is quite stable ( I ) . As with the RegulonDB data ( A–D ) , these results suggest that the main characteristics of the network are mostly preserved under different statistical constrains . The number of feedback loops remains low for reasonable variations around the selected threshold ( J ) , and the increase in number is compatible with the expected increase in the rate of false positives . Interestingly , looking at feedback loop number , a ‘phase transition’ becomes apparent for large thresholds . This behaviour suggests the use of great care when performing this type of analysis on very noisy data . DOI: http://dx . doi . org/10 . 7554/eLife . 02863 . 00710 . 7554/eLife . 02863 . 008Figure 2—figure supplement 4 . General properties and number of feedback loops under different random model for RegulonDB ( E . coli ) . The different constraints ( and algorithms ) disused in ‘Materials and methods’ result in different networks properties . ( A ) Number of TFs , number of regulatory connections among TFs and number of unregulated TFs . ( B ) Despite the different properties ( A ) , all the random models considered display a number of feedback loops strikingly different from the real data . DOI: http://dx . doi . org/10 . 7554/eLife . 02863 . 008 Qualitative Stability of GRNs prevents the catastrophe described by Maxwell from occurring , even when any of the myriad quantitative system parameters ( TF abundance , promoter availability , the rate of transcription , etc . ) are altered . This stability provides a type of ‘damping’ , which will tend to restore the state of the GRN when challenged with contingencies that might otherwise induce chaotic or unpredictable behaviour . Notably , the damped oscillatory response expected from a stable system has been observed in single cell experiments ( Tay et al . , 2010 ) . In principle the Qualitative Stability observed in GRNs might be easy to break by addition of another link to the network . For example , a long feedback loop can be created by the addition of a feedback connection from a TF lower down in a network path to a TF higher up in that path . This could occur , for example , through a mutation in the promoter of the target gene allowing it to be bound by a new TF . It is also possible that stress conditions could cause TFs to act inappropriately at promoters they do not normally regulate . In this way Qualitative Stability could be lost , and GRNs could become unstable . Thus , we predict that if long feedback loops are detrimental because of their instability , then GRNs would be configured to minimize destabilization via the addition of new connections . We call network paths that can be transformed into loops by the addition of a single new link ‘incomplete feedback loops’ ( Figure 3D–E ) . The abundance of incomplete feedback loops in the GRNs of E . coli , S . cerevisiae and human is shown in Figure 3A–C ( lightly shaded bars ) . Data for M . tuberculosis and P . aeruginosa are shown in Figure 3—figure supplement 1A , B . For each of these GRNs there are <2000 incomplete feedback loops and they tend to be of a relatively small size . A similar empirical observation has been made regarding transcriptional cascades ( Rosenfeld and Alon , 2003; The modENCODE Consortium et al . , 2010 ) . This is in stark contrast to comparable random networks ( Figure 3A–C , Figure 3—figure supplement 1A , B , heavily shaded bars , and Figure 3—figure supplement 4 ) , which on average have a thousand-fold greater number of incomplete feedback loops ( >105 ) of a significantly larger size . Statistical analyses suggest that there is an extremely small probability ( <10−19 ) that the absence of long incomplete feedback loops is a chance event in the three organisms considered ( Figure 3—figure supplement 2A–C ) . These results are relatively robust to variations in the confidence levels of the E . coli and S . cerevisiae GRN ( Figure 3—figure supplement 3A , B ) , and remain valid when different random models are considered ( Figure 3—figure supplement 4 ) . Note that the distribution of incomplete feedback loops is indicative of the different topological structures that can be observed in the network , and is not necessarily monotonically decreasing ( Figure 3—figure supplement 5 ) . 10 . 7554/eLife . 02863 . 009Figure 3 . Incomplete feedback loops in real and simulated GRNs . Number of incomplete feedback loops is provided on a logarithmic scale for E . coli ( A ) , S . cerevisiae ( B ) , and the human GM12878 cell line ( C ) and in each case is compared with a randomly simulated network containing the same number of genes , TFs , and connections . For the random networks , each graph reports the mean and standard deviation . ( D ) Schematic illustrations of incomplete feedback loops of length 2–6 are plotted in black , with the edge whose additions would result in additional long feedback loops plotted in red . ( E ) The number of long loops ( >2 nodes ) that can be created by the addition of a single new connection to incomplete loops of the specified length ( numerical values also given in red text in panel D ) . DOI: http://dx . doi . org/10 . 7554/eLife . 02863 . 00910 . 7554/eLife . 02863 . 010Figure 3—figure supplement 1 . Incomplete feedback loops in P . aeruginosa , M . tuberculosis and other yeast datasets . Number of incomplete feedback loops is provided on a logarithmic scale for M . tuberculosis ( A ) , P . aeruginosa ( B ) , the yeast dataset derived from Lee et al . ( C ) , the yeast dataset derived from Luscombe et al . ( D ) , and the yeast dataset derived from MacIsaac et al . ( E ) . In each case the real dataset is compared with a randomly simulated network containing the same number of nodes and connections . For the random networks , each graph reports the mean and standard deviation . Note how the number of incomplete feedback loops decreases rapidly for the more validated datasets ( A–C ) . DOI: http://dx . doi . org/10 . 7554/eLife . 02863 . 01010 . 7554/eLife . 02863 . 011Figure 3—figure supplement 2 . p-Value estimation for the number of long incomplete feedback loops . The distribution of the number of long incomplete feedback loops in random graphs has the same structure as the number of feedback loops and the same arguments apply ( See the caption of Figure 2—figure supplement 2 ) . The distributions in E . coli ( A ) , yeast ( B ) , and human ( C ) are plotted using the same conventions of Figure 2—figure supplement 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 02863 . 01110 . 7554/eLife . 02863 . 012Figure 3—figure supplement 3 . Number of incomplete feedback loops in the RegulonDB ( E . coli ) and Harbison et al . ( yeast ) datasets under different statistical conditions . The number of incomplete feedback loops in E . coli ( A ) and yeast ( B ) remain stable under reasonable variation of the conditions used to construct the GRNs . See the caption of Figure 2—figure supplement 3 for the description of the different conditions associated with the network reconstruction . Note the logarithmic scale and the same phase shift observed in the feedback loop distribution ( Figure 2—figure supplement 3 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 02863 . 01210 . 7554/eLife . 02863 . 013Figure 3—figure supplement 4 . Number of feedback loops under different random models for RegulonDB ( E . coli ) . The different random models lead to different numbers of incomplete feedback loops . However , the same exponential growth is observable in all the models , in striking contrast to the real data . Note the logarithmic scale . DOI: http://dx . doi . org/10 . 7554/eLife . 02863 . 01310 . 7554/eLife . 02863 . 014Figure 3—figure supplement 5 . Incomplete feedback loop distribution in different networks . The number of incomplete feedback loops is reported for different networks with the same number of edges . All the incomplete feedback loops of length two and three are reported for the network considered with matching colours . Note how the number of three-edge incomplete feedback loops is smaller , equal or larger than the number of two-edge incomplete feedback loops depending on the structure of the network considered . DOI: http://dx . doi . org/10 . 7554/eLife . 02863 . 014 The striking difference between real and random networks in the preponderance of both the number of long feedback loops and incomplete feedback loops strongly suggests a profound selective pressure on living organisms to adopt GRN topologies that are stable under all parameter regimes . In addition , these topologies efficiently prevent random mutations from introducing possible sources of destabilization . Note that feedback loops involving more than three genes are also highly susceptible to the creation of additional long feedback loops making them strongly disadvantageous to stability . We say that networks configured to minimize the number of real and incomplete long feedback loops possess Buffered Qualitative Stability ( BQS ) . Networks having this property are stable to perturbations and are also buffered against the potentially destabilising effects that occur when new links are added . If BQS is really a fundamental design principle of GRNs , as our data seem to suggest , they should display a range of other properties , which we describe and examine henceforth . An important global network property constrained by BQS is the degree of cross-regulation between TFs . Since a TF must be both regulated and regulating to take part in a feedback loop , one way that GRNs could satisfy BQS and minimise the risk of unstable loops being formed , is by having a high proportion of TFs that are not regulated by other TFs . Consistent with this prediction , the percentage of unregulated TFs in E . coli , S . cerevisiae , M . tuberculosis , P . aeruginosa , human and other yeast datasets is very high ( Figure 4A , Figure 4—figure supplement 1A–E ) . Comparison with random networks indicates that the probability of obtaining this proportion of unregulated TFs by chance is between 10−68 and 10−39 ( Figure 4—figure supplement 2A–C ) . Similar results hold for M . tuberculosis ( Figure 4—figure supplement 1A ) , P . aeruginosa ( Figure 4—figure supplement 1B ) , and other yeast datasets ( Figure 4—figure supplement 1C–E ) . These results are robust to variations in the confidence levels of the E . coli and S . cerevisiae GRNs ( Figure 4—figure supplement 4A , B ) , and remain valid when different random models are considered ( Figure 2—figure supplement 4A ) . 10 . 7554/eLife . 02863 . 015Figure 4 . Evidence for BQS from TF regulation . ( A ) For each organism , the percentage of TFs which are not regulated by any other TF is shown . Comparisons are made to randomly simulated networks containing the same number of genes , TFs , and connections . ( B ) Each of the 154 TFs in the E . coli GRN is plotted in the space of incoming regulatory connections ( number of regulatory links from other TFs ) and outgoing regulatory connections ( number of regulatory links to other TFs ) . Solid lines indicate isoclines of relative probability ( normalized to unity at reference point RP ) for creating a 3-gene feedback loop under random addition of a link . Percentages and colours indicate the fraction of TFs within each band demarcated by isoclines . ( C ) Each of the 154 TFs in the E . coli GRN is plotted in the space of incoming regulatory connections ( number of regulatory links from other TFs ) and outgoing regulatory connections of regulated TFs ( number of regulatory links originating from a TF that is regulated by the selected TF ) . In ( B and C ) , schematic motifs are provided; black solid arrows indicate the motif , while red dashed arrows indicate potential arrows whose addition results in the formation of long feedback loops . For each motif , the number of actual arrows is indicated in black and the number of potential destabilising arrows is indicated in red . DOI: http://dx . doi . org/10 . 7554/eLife . 02863 . 01510 . 7554/eLife . 02863 . 016Figure 4—figure supplement 1 . Unregulated TFs in P . aeruginosa , M . tuberculosis and other yeast datasets . Percentage of unregulated TFs for M . tuberculosis ( A ) , P . aeruginosa ( B ) , the yeast dataset derived from Lee et al . ( 2002 ) ( C ) , the yeast dataset derived from Luscombe et al . ( 2004 ) ( D ) , and the yeast dataset derived from MacIsaac et al . ( 2006 ) ( E ) . In each case the real dataset is compared with a randomly simulated network containing the same number of genes , TFs and connections . For the random networks , each graph reports the mean and standard deviation . Note the statistically significant difference of real and random data in all the dataset considered . DOI: http://dx . doi . org/10 . 7554/eLife . 02863 . 01610 . 7554/eLife . 02863 . 017Figure 4—figure supplement 2 . p-Value estimation for the number of unregulated TFs . The distribution of the number of unregulated TFs in random graphs follows a normal distribution in E . coli ( A ) , yeast ( B ) and human ( C ) . See Figure 2—figure supplement 2 , for the conventions used in the plots . Note the very small p-vals . DOI: http://dx . doi . org/10 . 7554/eLife . 02863 . 01710 . 7554/eLife . 02863 . 018Figure 4—figure supplement 3 . Cross regulation and second-order cross regulation in yeast and human . ( A and B ) TF cross regulation in yeast and the human GM12878 cell lines . ( C and D ) Second-order cross regulation control for the same datasets . See the caption of Figure 4 for the conventions used in the plots . DOI: http://dx . doi . org/10 . 7554/eLife . 02863 . 01810 . 7554/eLife . 02863 . 019Figure 4—figure supplement 4 . Number of unregulated TFs in the RegulonDB ( E . coli ) and Harbison et al . ( 2004 ) ( yeast ) dataset under different statistical conditions . The number of unregulated TFs in E . coli ( A ) and yeast ( B ) remain very high regardless of the conditions used to construct the GRNs , supporting the strong embedding of this feature in the GRNs . See the caption of Figure 2—figure supplement 3 for the description of the different conditions associated with the network reconstruction . DOI: http://dx . doi . org/10 . 7554/eLife . 02863 . 019 Some TFs , however , must be regulated by other TFs in order for the GRN to be able to combine information from multiple pathways and change state depending on different circumstances . In order to satisfy BQS , highly connected TFs should either be regulated by a large number of other TFs or should themselves regulate a large number of target TFs , but not both ( since otherwise the TF in question is significantly more susceptible to becoming part of a 3-gene feedback loop after addition of a link ) ; in other words , highly centralised control is disallowed . This prediction of BQS is indeed verified in E . coli ( Figure 4B ) , S . cerevisiae ( Figure 4—figure supplement 3A ) and human ( Figure 4—figure supplement 3B ) . The E . coli TF with the largest number of ‘outgoing regulatory connections’ regulates 38 other TF genes , but is itself regulated by only one TF; the TF with the largest number of ‘incoming regulatory connections’ is regulated by nine other TFs but itself regulates only one TF gene . There are no E . coli TFs that are both highly regulated and highly regulating ( in fact , there are no TFs regulating >2 other TFs that are themselves regulated by >2 TFs ) . BQS does not even allow central control to be split by connecting highly regulated TFs to highly regulating TFs , as this would create a large number of incomplete loops . Consistent with this idea , no E . coli or human TFs with in-degree >2 regulate TFs with out-degree >2 ( Figure 4C , Figure 4—figure supplement 3D ) . In S . cerevisiae there are several TFs that exceed these limits , but this represents only a small minority of TFs ( Figure 4—figure supplement 3C ) . These results show that BQS strongly favours distributed control over central control , and provide another example of BQS being a key determinant of the topology of GRNs . Our discussion so far has focused on the effect of BQS on large- and intermediate-scale global properties of GRNs . BQS also makes strong predictions about the small-scale local structure of GRNs . To investigate this , we dissected each of the GRNs into a series of small motifs comprising three or four genes ( Alon , 2006; Milo et al . , 2002 ) . A single motif can , in principle , break Qualitative Stability by forming a feedback loop composed of three or more genes . As we have shown above , such motifs are essentially absent from real GRNs . However , motifs may be susceptible to feedback loop formation through the addition of a link , and we can therefore speak of ‘buffered motifs’ as motifs that are resilient to this , and therefore enhance BQS locally . Note that , to prevent possible biases introduced by the large number of non-TF genes , only motifs completely formed by TFs were considered . Using symmetry arguments , we grouped 3- and 4-gene motifs into buffered and non-buffered categories , which are equi-probable in a random network ( confirmed by Figure 5—figure supplement 4B , E , F , H , K , L ) . Figure 5A–F show that in the real GRNs of E . coli , S . cerevisiae and human , buffered motifs ( blue ) are much more abundant than would be expected by chance , while unbuffered ( green and violet ) motifs are much less abundant; and indeed , unbuffered motifs which are particularly susceptible to breaking BQS ( violet ) are rare . Similar results hold for other confidence levels of E . coli ( Figure 5—figure supplement 3A–L ) , other confidence levels of S . cerevisiae ( Figure 5—figure supplement 3M–X ) , M . tuberculosis ( Figure 5—figure supplement 2A , B ) , P . aeruginosa ( Figure 5—figure supplement 2 , D ) , and other yeast datasets ( Figure 5—figure supplement 2E–J ) . Note that the IDs used in Figure 5—figure supplements 2–4 are described by Figure 5—figure supplement 1 . 10 . 7554/eLife . 02863 . 020Figure 5 . BQS in selected 3- and 4-gene motifs . Count of 3- and 4-gene motifs in the GRNs of E . coli ( A and D ) , S . cerevisiae ( B and E ) , and human GM12878 cell line ( C and F ) are classified into different categories . The y-axis reports the number of motifs plotted under each bar using blue ( buffered ) , green ( mono-unbuffered ) or violet ( poly-unbuffered ) solid arrows . Red dashed arrows are not part of the motifs but indicate the additional links that would create a 3- or 4-gene feedback loop . The motifs selected have an equal probability of occurrence in a random network ( See also Figure 5—figure supplement 4B , E , F , H , K , L ) . Horizontal black dotted lines in panels A–F indicate the motif abundance expected in random networks . ( G ) In the network schematic indicated by black arrows , adding a new connection can affect BQS; certain arrows will create a long feedback loop ( red dashed arrow ) , while others will not ( green dashed arrow ) . ( H ) Probability that a random edge addition between TFs would result in the creation of a long feedback loop is reported compared to random simulations . The different values observed in the simulations are due to the different number of TFs and regulatory connections in the three organisms . DOI: http://dx . doi . org/10 . 7554/eLife . 02863 . 02010 . 7554/eLife . 02863 . 021Figure 5—figure supplement 1 . Labelling conventions for Figure Supplements . To simplify the structure of Figure 5—figure supplements 2–4 , the motifs are indicated by an IDs instead of being depicted graphically . These IDs are described here . DOI: http://dx . doi . org/10 . 7554/eLife . 02863 . 02110 . 7554/eLife . 02863 . 022Figure 5—figure supplement 2 . 3- and 4-gene motifs in P . aeruginosa , M . tuberculosis and other yeast datasets . Number of selected 3- and 4-gene motifs for M . tuberculosis ( A and B ) , P . aeruginosa ( C and D ) , the yeast dataset derived from Lee et al . ( 2002 ) ( E and F ) , the yeast dataset derived from Luscombe et al . ( 2004 ) ( G and H ) , and the yeast dataset derived from MacIsaac et al . ( 2006 ) ( I and J ) . Note how the predictions of BQS are verified in all the datasets . DOI: http://dx . doi . org/10 . 7554/eLife . 02863 . 02210 . 7554/eLife . 02863 . 023Figure 5—figure supplement 3 . 3- and 4-gene motifs in the RegulonDB ( E . coli ) and Harbison et al . ( 2004 ) ( yeast ) datasets under different statistical conditions . The number of selected 3- and 4-gene motifs in E . coli ( A–L ) and yeast ( M–X ) follows the distribution predicted by BQS under all the conditions considered . This result supports the strong robustness of this analysis to noise . DOI: http://dx . doi . org/10 . 7554/eLife . 02863 . 02310 . 7554/eLife . 02863 . 024Figure 5—figure supplement 4 . 3- and 4-gene motifs under different random models for RegulonDB ( E . coli ) . As expected from theoretical considerations , when the number of TFs is fixed , the 3- and 4-gene motifs considered are equi-probable ( B , E , F , H , K , and L ) . This equal probability is broken when the number of TFs is allowed to vary freely , as a consequence of the variable size of the TF-only network . However , this type of randomness limits , rather than increases , the number of buffered stable motifs , in striking contrast to the real data ( A , C , D , G , I , and J ) . DOI: http://dx . doi . org/10 . 7554/eLife . 02863 . 024 These findings , besides confirming the role of BQS , provide additional support and potential explanations for the prevalence of well-studied motifs such as the feedforward loop ( a stable motif ) and the bi-fan ( a buffered stable motif ) as building blocks of networks ( Alon , 2006; Milo et al . , 2002 ) . Indeed , the most buffered 3- and 4-gene motifs studied here are , respectively , latent feed-forward loops and latent bi-fans . When a network uses only a subset of the possible edges in a motif , new connections can be added to expand the network ( Figure 5G ) . In general , some of these connections will create long feedback loops ( red dashed arrow ) , while others will not ( green dashed arrow ) . As we have shown , robustness appears to exert a strong selective pressure on the topology of GRNs . To assess the extent of this pressure we used an extensive computational approach to estimate the probability that a random edge addition between two TFs creates a long feedback loop in the GRNs . All the possible edge insertions were tested in the real GRNs of E . coli , S . cerevisiae and human , and the values were compared with those estimated in the corresponding random networks . The extent of buffered stability is quite remarkable: 4363 new interactions can be added to the human GRN , but only 48 of them will create long feedback loops ( Supplementary file 1 ) . By contrast , addition of single extra links to random networks would lead to the creation of approximately 2000 different feedback loops on average: a hit rate of approximately 50% . The probability that real GRNs will gain long feedback loops by random edge additions is low ( Figure 5H , lightly shaded bars ) , and is significantly smaller than that found for comparable random networks ( Figure 5H , heavily shaded bars ) . Nevertheless , these probabilities are non-zero , indicating a trade-off between stability and the need for cells to regulate gene expression . Qualitative Stability is compromised to a very small degree in the GRN of E . coli by the two small , but ‘illegal’ feedback loops shown in Figure 6A , B and highlighted in Figure 6—figure supplement 1 . Three aspects are noteworthy . Firstly , of the seven 2-node feedback loops in the E . coli GRN , four are embedded into the two potentially unstable motifs depicted in Figure 6A , B , whilst the other three are isolated from other feedback loops and so are either stable or likely to act as switches . Secondly , the genes comprising the two illegal motifs are involved in drug resistance ( Ariza et al . , 1995; Martin and Rosner , 2002; Nishino et al . , 2008; Keseler et al . , 2011; Figure 6C ) and/or acid resistance ( Sayed et al . , 2007; Keseler et al . , 2011; Figure 6C ) . This raises the possibility that limited instances of deviation from BQS may arise through evolution as a short-term expediency allowing survival in a changing environment . Thirdly , both loops share a remarkably similar sub-structure: two linked 2-gene feedback loops connected by one link into a 3-node feedback loop . It has been previously shown in chemical networks that such a configuration can display chaotic behaviour ( Sensse and Eiswirth , 2005 ) . It is tempting to speculate that these illegal motifs act as localized sources of chaos , allowing a cell population to quickly explore very diverse gene expression levels , thus accelerating the emergence of resistant phenotypes ( Lopez-Maury et al . , 2008 ) . Moreover , compatible with the idea that chaotic behaviour should be tightly controlled , most of the genes depicted in Figure 6A , B are highly regulated ( Figure 6D ) . 10 . 7554/eLife . 02863 . 025Figure 6 . The two ‘illegal’ feedback loops in E . coli . The figure shows connections between TFs in the two motifs in E . coli that contain feedback loops with >2 links ( A and B ) . Note the common feature of two adjacent 2-gene feedback loops . ( C ) Colours indicate if a gene is implicated in drug resistance ( red ) , acid resistance ( blue ) , or both ( violet ) . ( D ) The number of regulatory connections for each TF in ( A and B ) . DOI: http://dx . doi . org/10 . 7554/eLife . 02863 . 02510 . 7554/eLife . 02863 . 026Figure 6—figure supplement 1 . The transcription factor GRN in E . coli with long feedback loops highlighted . The TFs implicated in the formation of the two motifs illustrated in Figure 6 are coloured in black and their interactions in red . All the other TF are coloured in grey and their interactions in blue . DOI: http://dx . doi . org/10 . 7554/eLife . 02863 . 02610 . 7554/eLife . 02863 . 027Figure 6—figure supplement 2 . Long feedback loops in M . tuberculosis . Only three long feedback loops are observed in M . tuberculosis ( A–C ) . Note the presence of the same 3-node motifs observed in E . coli ( Figure 6A , B ) and how these two motifs are entangled in the four node feedback loop ( C ) . DOI: http://dx . doi . org/10 . 7554/eLife . 02863 . 02710 . 7554/eLife . 02863 . 028Figure 6—figure supplement 3 . Long feedback loops in Lee et al . ( yeast ) . Only one long feedback loop is observed in the yeast dataset derived by Lee et al . ( A ) . Since Harbison et al . ( 2004 ) uses a technology similar to Lee et al . ( 2002 ) and can be considered to have provided an updated version of the network , we analysed what caused the disappearance of the invalid loop reported by Lee et al . ( 2002 ) . Five of the six interactions have comparable p-values in the two datasets ( A and B ) . However , the p-value for the interaction between ROX1 and YAP6 changes by three orders of magnitude , thus transforming the feedback loop into a feedforward loop that is consistent with BQS . This observation indicates how the strong predictions of BQS can be used to identify potential inaccuracies and problems in GRN data . DOI: http://dx . doi . org/10 . 7554/eLife . 02863 . 028 These ideas are also supported by the M . tuberculosis GRN: all the four genes involved in the formation of illegal motifs are implicated in stress responses ( He et al . , 2006; Rodriguez et al . , 2002 ) , and the two 3-gene feedback loops share the same topology observed in E . coli ( Figure 6—figure supplement 2A , B ) . In addition , of the six 2-node feedback loops observed in the M . tuberculosis GRN , three are isolated from other feedback loops and the other three are embedded into the two potentially chaotic motifs . Finally , it is noteworthy that the 4-gene feedback loop in the M . tuberculosis GRN is formed by joining the two 3-gene feedback loops ( Figure 6—figure supplement 2C ) , consistent with our earlier observation that long feedback loops are susceptible to the formation of additional embedded feedback loops . P . aeruginosa has no long ( >3 genes ) feedback loops , and of its seven 2-node feedback loops five are isolated , and so are either stable or likely to act as switches , whilst the other two are linked but are of a form ( positive-negative ) that makes them Qualitatively Stable . Both of the S . cerevisiae 2-node feedback loops are isolated , whilst the human GM12878 cell line has only a single 2-node feedback loop which is therefore isolated . Curiously , the only long feedback loop observed in the yeast GRN derived by Lee et al . ( 2002 ) presents the same potentially chaotic topology discussed above . However , this illegal motif is not present in the more recent GRN derived by Harbison et al . ( 2004 ) ( Figure 6—figure supplement 3 ) . Cancer cells have a dysregulated behaviour , breaking the ‘social contract’ necessary for maintenance of a healthy multicellular organism . Even within an individual tumour a wide range of cellular phenotypes is often observed ( Marusyk et al . , 2012 ) . To some degree , this is likely to be a consequence of the genotypic heterogeneity of tumours . However , cancer cells also appear to be phenotypically less stable than normal cells ( Brock et al . , 2009; Gupta et al . , 2011 ) . Might this phenotypic instability result from a breakdown of BQS in cancer cells ? There is currently only a single cancer cell line , the human leukaemia cell line K562 , for which a high quality system-wide GRN has been derived ( Gerstein et al . , 2012 ) . We therefore investigated the topological differences between the GRNs of K562 and the human non-cancer cell line GM12878 . Figure 7 and Figure 7—figure supplement 1 show that the feedback loop distribution in the GRN of the leukaemia cell line is strikingly different from that of the non-cancer cell line . In K562 , significantly more moderately long feedback loops of 4–8 genes are present ( Figure 7A , B ) , and the number of loops formed by 3–5 genes is comparable to the number expected in a random network ( Figure 7A ) . The number of incomplete feedback loops is also significantly larger in K562 ( Figure 7C , D ) . In addition , poorly buffered motifs are abundant ( Figure 7E , F ) and the TFs cross regulation is less extreme ( Figure 7—figure supplement 2A , B ) . 10 . 7554/eLife . 02863 . 029Figure 7 . Broken BQS in a human cancer cell line . ( A and C ) Number of feedback loops and incomplete feedback loops in the GRN of the human cancer cell line K562 , compared to the corresponding random network . Data is provided on a logarithmic scale . ( B and D ) Number of feedback loops and incomplete feedback loops in the GRN of K562 , compared to the human non-cancer cell line GM12878 . Note the use of a linear scale in ( B ) . ( E and F ) Motif analysis of BQS for K562 for 3- and 4-gene motifs , using the same convention as described in Figure 5 . ( G ) Genes implicated in the formation of long feedback loops in K562 are reported with the number and percentage of loop created , the length of the longest incomplete feedback loops involving the gene in GM12878 are also indicated . ( H ) The probability that a random edge addition would result in the creation of a long feedback loop is reported for both GM12878 vs K562 and K562 vs randomly simulated networks containing the same number of nodes and connections . Note that the GRNs of GM12878 and K562 have similar link densities . They comprise 4071 and 4049 genes respectively , and 8466 and 11 , 707 regulatory connections respectively . DOI: http://dx . doi . org/10 . 7554/eLife . 02863 . 02910 . 7554/eLife . 02863 . 030Figure 7—figure supplement 1 . The transcription factor GRNs for GM12878 and K562 . TFs and their interactions are plotted for the non-cancer cell line GM12878 ( A ) and the leukaemia cell line K562 ( B ) . TF names are indicated in black , transcriptional interactions implicated in the formation of long feedback loops are coloured in red , while all the other interactions are coloured in blue . DOI: http://dx . doi . org/10 . 7554/eLife . 02863 . 03010 . 7554/eLife . 02863 . 031Figure 7—figure supplement 2 . Cross regulation and second-order cross regulation in cancer . TF cross regulation ( A ) and second-order cross regulation ( B ) display in cancer similar overall trend when compared to E . coli , yeast and the non-cancer cell line ( Figure 4 , Figure 4—figure supplement 3 ) . However , a slightly less stringent distribution can be observed , compatible with a diminished stability . DOI: http://dx . doi . org/10 . 7554/eLife . 02863 . 03110 . 7554/eLife . 02863 . 032Figure 7—figure supplement 3 . Longest incomplete feedback loops in the GM12878 GRN and their destabilization potential . The smallest subset of genes containing the longest incomplete feedback loops of the GRN of the human cell line GM12878 are illustrated ( A–D ) . Along with the smallest subset of genes containing all longest incomplete feedback loops of the GM12878 GRN ( E ) and the simplest 5-gene incomplete feedback loop ( F ) . The probability that inserting a random edge will introduce a long feedback loop in the human transcription factor networks ( green bar ) , in the sub networks identified by the single incomplete feedback loops ( blue bars ) , in the sub network identified by all the incomplete feedback loops ( red bar ) , and in a pure incomplete feedback loop ( black bar ) is also displayed ( G ) . Note how the probability increases drastically as incomplete feedback loops are considered . DOI: http://dx . doi . org/10 . 7554/eLife . 02863 . 032 Interestingly , only a limited number of TFs contribute to the formation of the 59 long loops in the K562 GRN ( Figure 7G ) and it can be substantially ‘stabilized’—that is most of the long feedback loops can be removed—by the removal of single pairs of TFs: FOSL1 and JUNB , JUNB and EGR1 , or EGR1 and CEBPB . JUNB and FOSL1 are proto-oncogenes that are components of the AP-1 transcription complex ( Kouzarides and Ziff , 1989 ) and have been reported to be part of a long feedback loop in ovarian cancer ( Stelniec-Klotz et al . , 2012 ) ; EGR1 is a regulator of tumour suppressor genes and an oncogene itself ( Baron et al . , 2005 ) ; and the TFs of the C/EBP family have been described as both tumour promoters and suppressors ( Nerlov , 2007 ) . These genes are also active in the non-cancer cell line GM12878 , yet their destabilizing role in cancer arises from a rewiring of the network . GM12878 satisfies BQS whilst K562 comparatively does not , and yet the two GRNs have similar numbers of genes and regulatory connections ( in fact , the K562 TF network has a lower link density than that of GM12878 ) . Finally , the probability of introducing long feedback loops by random edge additions is significantly larger in the cancer cell line than in non-cancer cells , but still lower than the value expected in a random network ( Figure 7H ) . These findings suggest that cancer cells break BQS and have a vastly less stable GRN than normal cells , which is however less unstable than expected in a random network . Certain features of GRNs , such as rare , long incomplete feedback loops , make them more susceptible to the formation of long feedback loops when new regulatory connections are added ( Figure 7—figure supplement 3A–G ) . Therefore , we investigated how many of the long feedback loops observed in K562 cancer cells could have originated from these relatively unbuffered network structures in non-cancerous GM12878 cells . We find that the three genes that contribute the most to the formation of long feedback loops in the cancer cell line ( EGR1 , FOSL1 , and JUNB ) are all involved in the formation of the longest incomplete feedback loops observed in the non-cancer cell line ( Figure 7G ) . As highlighted earlier ( Supplementary file 1 ) , single insertions of a new connection into the non-cancer human cell line can create 48 different long feedback loops . Of these 48 potentially destabilising interactions of the non-cancer cell line , three are actually observed in the cancer cell line , namely JUNB-BCLAF1 , BATF-EBF1 , and JUNB-EBF1 . The likelihood of these potentially destabilising interactions occurring in the cancer cell line is 3/48 = 0 . 063 . In comparison , there are a total of 4363 additional links that could be made between the TFs of the non-cancer cell line . Of these potential links , 56 are observed in the cancer cell line . Hence , the links observed represent 56/4363 ( 0 . 013 ) of all the possible ones . Therefore , the destabilizing interactions are five times more abundant than would be expected by chance ( 0 . 063 vs 0 . 013 ) , consistent with the idea that destabilizing interactions were positively selected for during the microevolution process underlying cancer progression . This suggests that , despite the diverse biological histories of these two cell lines ( the networks of GM12878 and K562 share only four common links between TFs ) , the progression of K562 into a cancerous state has involved changes to regions of the GRN in the progenitor normal cells that displayed weaker BQS . Such genetic factors are therefore likely to play a pivotal role in the process of cancer progression in other cell types . The eukaryotic GRNs analysed so far are static ‘snapshots’ of potential transcriptional interactions of a population of cells under rich media growth . As discussed in ‘Materials and methods’ , such conditions are ideal to minimize cell heterogeneity and to obtain high quality equilibrium networks that are ideally suited for theoretical analysis . However , during the lifetime of an organism GRNs are dynamic and the set of actual transcriptional regulations can change ( Luscombe et al . , 2004 ) . GRNs transitioning from one transcriptional program to another are unlikely to be at steady state and , under these circumstances , transcriptional robustness may be less important . We decided to test if BQS could provide new insights into the role of robustness during such structural changes . To this end , we used the data presented in Garber et al . ( 2012 ) to reconstruct the GRN of murine dendritic cells at four different time points after stimulation by pathogens: at the time of stimulus application ( marked as ‘0 hr’ ) and after the cells have been exposed to the stimulus for 30 min ( ‘0 . 5hr’ ) , 1 hr ( ‘1 hr’ ) , and 2 hr ( ‘2 hr’ ) . Since the number of TFs studied in these networks is much smaller than the number considered previously , a different simulation algorithm has been used ( see ‘Materials and methods’ ) . The ‘0 hr’ GRN was obtained under conditions comparable with the organisms discussed previously . Figure 8 shows that all the predictions of BQS are met at this time . As indicated by Figure 8A , the GRN has a very limited number of long feedback loops ( only three with three or more genes ) , in striking contrast with the hundreds observed in randomly generated networks of the same link density . Interestingly , all of the long feedback loops depend on the transcriptional interaction between SFPI1 and E2F1 ( Figure 8—figure supplement 1A–C ) . Notably E2F1 plays a crucial role in the cell cycle and is only transiently activated at commitment to cell division at the end of G1 . Therefore , all of the long feedback loops detected are likely to be transient . Similarly , the number of incomplete feedback loops is very small , and much lower than would be expected in random networks ( Figure 8B ) . Motif analysis is also consistent with BQS: there is a much higher proportion of unregulated transcription factors than would be expected by chance ( Figure 8E ) , and the proportion of stable 3- and 4-node motifs is heavily biased towards the buffered stable forms that enhance BQS ( Figure 8C , D ) . Additionally , the mode of cross regulation—with transcription factors tending to be either highly regulating or highly regulated ( Figure 8F ) —also follows the distribution predicted by BQS . Finally , the probability of creating additional long feedback loops in the network by randomly inserting a new regulatory connection is only 0 . 18 , much lower than the value of 0 . 74 expected in a comparable random network . 10 . 7554/eLife . 02863 . 033Figure 8 . BQS in homeostatic murine dendritic cells . ( A and B ) Number of feedback loops and incomplete feedback loops compared to the corresponding random network . Data is provided on a logarithmic scale . ( C and D ) Motif analysis of BQS for 3- and 4-gene motifs , using the same convention as described in Figure 5 . ( E ) Percentage of TFs which do not regulate other TFs compared to the corresponding random network . ( F ) Relation between the number of incoming and outgoing regulatory connections; the same conventions of Figure 4B are used . DOI: http://dx . doi . org/10 . 7554/eLife . 02863 . 03310 . 7554/eLife . 02863 . 034Figure 8—figure supplement 1 . Long feedback loops in the GRN of homeostatic dendritic cells . Three long feedback loops can be found in the GRN of homeostatic dendritic cells ( A–C ) . However , all of them depend critically on the transcriptional interaction between Sfpi1 and E2f1 . The interaction between Sfpi1 and E2f1 is highlighted in red for clarity . Note that such an interaction is the only one that is fundamental for the formation of all the long feedback loops . DOI: http://dx . doi . org/10 . 7554/eLife . 02863 . 034 Since a full eukaryotic transcriptional response typically requires >1 hr , we expect the ‘0 . 5 hr’ network to be similar to the ‘0 hr’ one . Indeed , the ‘0 . 5 hr’ network still satisfies all the predictions of BQS and strongly resembles the ‘0 hr’ network ( Figure 9A , B , G ) . In contrast , at 1 and 2 hr after the stimulus , a marked deviation from BQS is observed . A significant number of new long loops are created , peaking at 1 hr and declining slightly by 2 hr ( Figure 9C , E ) . 22 long feedback loops remain at 2 hr , but interestingly all depend on the transcriptional interaction between RUNX1 and CEBPB . The probability of creating additional long feedback loops is noticeably larger at 1 and 2 hr than in the previous networks ( Figure 9G ) . There is also a significant increase in the number of 4-node unbuffered motifs at 1 hr , though this does not persist in the ‘2 hr’ GRN . However , some components of BQS remain unchanged in the stimulus response . Incomplete feedback loops still remain preferentially short ( Figure 9—figure supplement 1A , E , I ) , TF cross regulation remains essentially unchanged ( Figure 9—figure supplement 1D , H , L ) and the numbers of unbuffered 3-node motifs ( Figure 9—figure supplement 1B , F , J ) and unregulated TFs ( Figure 9—figure supplement 1C , G , K ) remain low . Taken together , these observations show that in response to a stimulus that changes the transcriptional profile , there is a modest and probably transient loss of BQS as the GRN moves into a new configuration . 10 . 7554/eLife . 02863 . 035Figure 9 . BQS during a transcriptional program activation in dendritic cells . ( A , C and E ) Number of feedback loops and incomplete feedback loops in the GRN of dendritic cells at different times after a stimulus compared to the corresponding random network . Data is provided on a logarithmic scale . ( B , D and F ) Motif analysis of BQS for 4-gene motifs , using the same convention as described in Figure 5 . ( G ) Probability of creating additional long feedback loops by the addition of a random link between TFs in the GRN of dendritic cells at different times after a stimulus and the corresponding random network . DOI: http://dx . doi . org/10 . 7554/eLife . 02863 . 03510 . 7554/eLife . 02863 . 036Figure 9—figure supplement 1 . BQS in the GRN of dendritic cells . As discussed in the main text , various degrees of stability can be observed in the GRN of dendritic cells when they are in the process of activating a new transcriptional program . The GRN is very stable 0 . 5 hr after the new transcriptional program has been initiated ( A–D ) , it is unstable after 1 hr ( E–H ) , but partially stabilizes after 2 hr ( I–L ) . The comparison of ( D , H and L ) indicates that a change in the balance between the number of inputs and outputs of a few TFs plays a strong role in controlling the stability of the GRN . To provide a better understanding of the dynamics , in ( D , H and L ) , the TFs with the most unbalanced input/output are labelled . Looking at the labelled TFs it is possible to notice how they tend to move towards the centre of the plot in panel H and move back to the sides in ( L ) . DOI: http://dx . doi . org/10 . 7554/eLife . 02863 . 036
It is widely believed that feedback plays a major role in biological control ( Harris and Levine , 2005; Tsang et al . , 2007; Peter et al . , 2012 ) . As we have demonstrated here , BQS demands that GRNs are free of long feedback loops , even under the addition of new links . The question remains , to what extent feedback operates in post-transcriptional regulation rather than the purely transcriptional networks we have examined here . It was not possible for us to perform a similar analysis of post-transcriptional regulation because strongly validated system-wide data describing such interactions are not currently available . However , the biological literature provides some clues . Motif analysis of post-transcriptional networks reveals that BQS-compliant feedforward loops are overrepresented , while BQS-breaking feedback loops are not ( Gerstein et al . , 2010; The modENCODE Consortium et al . , 2010; Cheng et al . , 2011; Joshi et al . , 2012 ) . This suggests that Qualitative Stability may still be an important principle governing the topology of these networks . Additionally , it has been noted that stable motifs are more common than unstable ones even in post-transcriptional signal transduction ( Milo et al . , 2002 ) , suggesting again a role for Qualitative Stability in these networks . Nevertheless , the existence of feedback loops in post-transcriptional networks ( Harris and Levine , 2005; Tsang et al . , 2007 ) supports the idea that the severe constraints of BQS are ‘loosened’ to allow more responsive dynamic functionalities in post-transcriptional regulation . Consistent with these expectations , short feedback loops involving two or three genes appear to be present in some developmentally regulated gene networks ( Peter et al . , 2012 ) . This raises the hypothesis that the different levels of gene regulation ( transcriptional vs post-transcriptional ) provide a way of segregating control modules with different robustness properties . The robustness provided by BQS allows a transcriptional network to filter out internal or external disturbances . Such robustness is desirable under normal conditions , but can be detrimental during a transcriptional response that requires effective and fast changes in the set of transcribed genes . Consistent with this idea , our analysis indicates that a certain level of instability builds up during a response to a stimulus that produces a transcriptional response . Quite remarkably our results also suggest that two hours after such a stimulus the GRNs considered are ‘on the verge of stability’ , as the deactivation of just one transcriptional interaction will make the network robust according to the BQS rules . It is worth pointing out , however , that each cell in a population is responding to a stimulus independently and potentially at a different pace . Therefore , the apparent post-stimulus instability observed may be the result of sampling transcriptional occupancy of cells at different stages of the transcriptional response . Future experimentation focused on the robustness property of cells and single cell GRN reconstruction will likely clarify which interpretation is correct . The molecular bases of evolutionary innovations are complex and poorly understood ( Wagner , 2011 ) . It is notable that the only instances where the E . coli and M . tuberculosis GRNs deviate from BQS are found in genes functionally related to stress responses . This might provide controlled instability allowing bacteria to explore new gene expression levels in response to environmental stresses , thus achieving a short-term evolutionary advantage . While other mechanisms are likely to be in play to achieve longer-term adaptations , therapeutic targeting of unstable motifs may provide a novel systems approach to drug discovery . The deviation of the human cancer cell line K562 from BQS is also very striking . This deviation allows a cancer cell , in principle , to readily change its phenotype in response to internal or external stresses , so it can explore different phenotypic states , which might help its proliferation in otherwise challenging tissue environments , or even to survive drug treatment . Nevertheless , the cancer cell line still possesses a degree of BQS far greater than that observed in a random network . This is consistent with the role of the unstable motifs in E . coli and M . tuberculosis and supports the idea that a small breakdown of BQS might be a hallmark of recent or rapid selection pressure . We note that single-cell experiments report large changes in protein abundance occurring in individual cancer cells after drug treatment , consistent with our theory ( Cohen et al . , 2008 ) . Taken together , our results indicate that BQS adds a powerful new weapon to the arsenal of network medicine ( Barabasi et al . , 2011; Noh et al . , 2013; Pe’er and Hacohen , 2011 ) . The discovery that the transcriptional interactions most critically involved in the formation of long feedback loops in the cancer cell line K562 are also present in the longest incomplete feedback loops in the non-cancer cell line GM12878 may provide clues to mechanisms of cancer initiation and progression . Our theory suggests that random perturbations to the GRN of normal cells will probably leave its stability unchanged . Destabilization of the GRN is likely to involve those genes that are on the edge of stability , for example genes participating in long incomplete loops . It is striking that the two gene families with the highest destabilization potential in the non-cancer cell line ( JUN and FOSL ) are actually found to be the two gene families with the highest destabilization role in the cancer cell line . These observations are consistent with the notion of cancer as a ‘systems disease’ that involves changes that lead to destabilisation of the GRN . If network instability is a general feature of cancer cell GRNs , analysis of this kind can potentially be used to design new anti-cancer strategies that exploit the unique weaknesses of cells lacking BQS .
The study of stability in qualitative networks of interacting entities was introduced in the seminal work of Quirk and Ruppert in economics ( Quirk and Ruppert , 1965 ) . Since then , the idea has been applied to different disciplines ( May , 1973b; Tyson , 1975 ) and the theory has been slightly enhanced ( Jeffries , 1974 ) . It has also been studied within the broader field of qualitative matrix theory ( Maybee and Quirk , 1969; Hale et al . , 1999 ) . While the theory developed by Quirk and Ruppert is probably the most well-known tool to study Qualitative Stability , a similar formalism is provided by the work of Puccia and Levins on loop analysis ( Puccia and Levins , 1985 ) . The theory deals with systems at equilibrium and considers the effect of small perturbations . Stable systems are characterized by the ability to react to these perturbations by returning to their original equilibrium state . In qualitative networks , each node is associated with a quantity or concentration and the presence of an arrow from node A to node B indicates that changing the concentration of A has an effect on B . The sign of the arrow indicates whether increasing A increases ( positive arrow ) or decreases ( negative arrow ) B . The absence of an arrow indicates that increasing the concentration of A does not directly affect the concentration of B . Note that self-regulation , represented by an arrow from a note to itself , is also possible . In its original form , the theory states four conditions for qualitative stability:Absence of positive self-regulationAbsence of double positive or double negative two-node feedback loopsAbsence of feedback loops longer that twoInvertibility of the sign matrix Condition 1 prevents unlimited autocatalysis . Condition 2 prevents unlimited ‘collaborative autocatalysis’ ( double positive ) and switches ( double negative ) , but note that positive/negative two-node feedback loops are allowed . Condition 3 is less intuitive , and disallows feedback systems that may go ‘out-of-sync’ for example as a consequence of non-linearities in the interactions . Condition 4 forbids the presence of two or more nodes that are being affected , or affect , the same nodes in the same way . While conditions 1 , 2 and 4 are very important from a theoretical point of view , they are of limited relevance to our analysis . For autocatalytic positive self-regulation , saturating effects on transcription ( for example due to limiting numbers of RNA polymerases ) along with protein degradation means that autocatalytic TFs will reach a stable steady-state rather than increasing to arbitrarily high values . Therefore condition 1 is of questionable relevance to GRNs . This argument of finite resources obviating positive self-regulation has also been applied to the study of Qualitative Stability in ecosystems ( May , 1973a ) . GRNs breaking condition 4 have no biologically plausible incarnation . In order for this condition to be violated , two different TFs must act on the network in the same way , that is , they need to regulate each other and promote and inhibit exactly the same set of genes , as this would make the columns of the sign matrix linearly dependent and therefore the matrix non-invertible . If two TFs act in the way described , they would be biochemically indistinguishable . Due to the methodology used to derive GRNs , two such genes would be collapsed into the same node . Condition 4 is therefore trivially verified . The case of double positive and double negative feedback loops is more complex . Under biological constraints relevant for GRNs , isolated double negative feedback loops ( i . e . , not connected with other feedback loops formed by two or more nodes ) are unstable only to the extent that they form switches that can exist in one of two stable states . Double positive feedback loops are potentially capable of a more complex behaviour . However , when considered in isolation with negative self-regulation under constraints relevant to GRNs , they are likely to display a switch-like behaviour ( Banerjee and Bose , 2008 ) . Remarkably , in both GRNs for which sign information is available these conditions are verified: in P . aeruginosa the only double positive feedback loop is isolated and in E . coli all non-isolated two-gene feedback loops are part of the potentially chaotic motifs discussed above , and are therefore not relevant for these particular 2-node stability arguments . Hence , the only scenario in which condition 2 potentially threatens Qualitative Stability in GRNs is a ‘daisy chain’ of linked 2-node feedback loops; in the single case where such a daisy chain is observed ( in P . aeruginosa ) both 2-node feedback loops are of the positive/negative type and so result in a form compatible with Qualitative Stability . Taken together , the above considerations indicate that condition 3 is the most significant in a rigorous comparative study on the GRNs . The introduction of a time delay into a system generally leads to an increased dimensionality . It is therefore not surprising that a delay introduced into feedback loops can lead to oscillations and instability , and this is one additional reason contributing to feedback loops containing >2 nodes not being Qualitatively Stable . Gene expression is a complex multi-step process and the role of regulatory mechanisms as potential sources of delay is an active area of research ( Gorgoni et al . , 2014 ) . The complexity of the problem , and the difficulty in obtaining quantitative information , limit our ability to assess the role of delays on our model . A delay that is fast compared to transcriptional regulation can be ignored due to the separation of the time-scales . When this is not the case , delays are potential sources of instability , especially in the presence of feedbacks of any size . Delays may be stronger in eukaryotes than in prokaryotes , as transcription and translation are spatially separated . This may partially explain why the GRNs of yeast and GM12878 have a remarkably limited number of feedback loops of any size . A final aspect of the theory is worth noting . The theory of Qualitative Stability has been developed by means of differential equations and the conditions discussed above properly apply only to deterministic autonomous systems . This results in our model being a simplification when compared to biological networks such as GRNs . It has been observed that cells limit the noise in gene expression ( Raj and van Oudenaarden , 2008 ) , suggesting the existence of biological mechanisms that reduce the extent of stochasticity . Therefore , whilst our model is likely to be largely compatible with the biological behaviour , noise may play a role in triggering GRN reorganization in response to strong stimuli . It also is worth stressing that while stability is generally of pivotal importance , particular situations may require a fast , rather than stable , response . Therefore , when GRNs must be able to undergo changes of state , for example during development ( Peter et al . , 2012 ) , or must be able to respond to external stimuli , for example to mount an immune response ( Ciofani et al . , 2012 ) , the role of stability may be more limited . The reconstruction of the GRN of an organism is a challenging task , and an active field of research ( Kim and Park , 2011 ) . Different methodologies have been developed and each of them presents advantages and disadvantages . Nevertheless , certain techniques , such as PCR and ChIP-Seq , are generally regarded as more reliable . Besides the technical problems , additional complications arise from the intrinsic working of gene interactions . The GRN is dynamic and changes according to external and internal conditions ( Harbison et al . , 2004; Luscombe et al . , 2004 ) . The sources of this variability are probably diverse and additional work will be needed to assess the cellular mechanisms that shape the GRN . Moreover , the different molecular responses observed at a single-cell level ( Cohen et al . , 2008; Tay et al . , 2010 ) suggest that the different stages of the cell cycle and different stress conditions may result in structurally different GRNs . Single-cell GRN reconstruction is currently beyond experimental reach , and stress is known to promote intra-population diversity ( Cohen et al . , 2008; Lopez-Maury et al . , 2008 ) . Therefore , where possible , we preferred to analyse GRNs obtained under rich media growth . These conditions result in a low cellular stress and thus are more likely to promote homogeneity in transcriptional response . This homogeneity minimizes the errors in GRN reconstruction due to superpositions of possibly different GRNs . No GRN currently available should be expected to be a completely faithful representation of the real interactions among the genes . However , it is reasonable to assume that networks obtained with direct biological methodologies under controlled conditions are not biased towards certain topological features , and therefore provide a good representation of the topology of the real GRN . Given these premises , it is not surprising that different GRNs are available in the literature for the same organism , and a choice had to be made to determine the datasets better suited for our analysis . We tried to select high quality datasets characterized by a statistical assessment of the interactions , a low rate of false positives , and public availability of the data . The Escherichia coli RegulonDB dataset ( Salgado et al . , 2012 ) is probably the most validated GRN available in the literature . This dataset is regularly updated to incorporate new data , and is consistently used as a basis for theoretical studies ( Alm and Arkin , 2003; Milo et al . , 2004 ) . In its current state the dataset does not report the environmental conditions associated with each interaction . Therefore , it is likely that under any specific conditions only a subset of the interactions is active . Recently , the GRNs of two other prokaryotes have been published: Pseudomonas aeruginosa ( Galan-Vasquez et al . , 2011 ) and Mycobacterium tuberculosis ( Sanz et al . , 2011 ) . Since these organisms are less well studied than E . coli , their GRNs should be expected to be less complete . Data on human GRNs are limited and the recent work by the ENCODE consortium ( Gerstein et al . , 2012 ) provided a unique opportunity to compare the GRNs of a non-cancerous and a cancerous cell line , studied under similar experimental conditions . Moreover , the methodology used to construct these networks ( ChIP-Seq ) and the carefully engineered protocol suggest a high degree of biological reality . The situation for yeast is more complex . After the work by Lee et al . ( 2002 ) , other datasets have been made available . To the authors’ knowledge , the work by Harbison et al . ( 2004 ) is the most exhaustive GRN derived from direct biological methods under stable conditions ( rich media ) , and therefore the ideal workbench for a topological analysis . Other yeast GRNs have been published and used for different types of studies about the genetic bases of yeast behaviour; in this context , Luscombe et al . ( 2004 ) and MacIsaac et al . ( 2006 ) are interesting examples . Luscombe et al . ( 2004 ) extended Lee et al . ( 2002 ) by introducing additional interactions obtained under different environmental conditions , but the data available do not provide a statistical assessment for the interactions . Moreover , the data from Garber et al . ( 2012 ) raise the possibility that the GRN of yeast is different under different conditions , suggesting that the network derived by Luscombe et al . ( 2004 ) may not provide a faithful representation of an equilibrium yeast GRN . MacIsaac et al . ( 2006 ) used the data provided by Harbison et al . ( 2004 ) to construct a regulatory network encompassing different Saccharomyces species , and derived the more conserved interactions . To the authors’ knowledge , the dataset discussed in Garber et al . ( 2012 ) is the only one available where the author used ChIP-Seq to study the dynamics of transcriptional response . Other authors either make use of gene expression data—thus removing the purely transcriptional nature of the network—or consider only a handful of transcription factors—thus making the statistical analysis discussed here inappropriate . As remarked above , the heterogeneity of the transcriptional response in a population of cells is likely to contribute to the experimental error in this this type of data , and additional experimentation is important to verify our conclusions . The E . coli GRN was constructed using version 8 of the RegulonDB ( Salgado et al . , 2012 ) available at http://regulondb . ccg . unam . mx/ . The network was restricted to those interactions supported by at least two evidence codes . The validity of our approach with a different number of evidence codes is assessed in Figure 2—figure supplement 3A–E , Figure 3—figure supplement 3A , Figure 4—figure supplement 4A , and Figure 5—figure supplement 3A–L . The S . cerevisiae GRN was constructed using the interactions reported by Harbison et al . ( 2004 ) under rich media growth ( http://younglab . wi . mit . edu/regulatory_code ) . The network was restricted to those interactions with a p-value lower than 10−3 . The validity of our approach with different p-values is assessed in Figure 2—figure supplement 3F–J , Figure 3—figure supplement 3B , Figure 4—figure supplement 4B , and Figure 5—figure supplement 3M–X . The human non-cancer and cancer cell GRNs were constructed using the filtered data constructed by Gerstein et al . ( 2012 ) for the GM12878 and K562 cell lines respectively . As previously observed , cells from different tissues generally display different GRNs ( Pe’er and Hacohen , 2011; Bensimon et al . , 2012 ) . Therefore , we analysed the GRNs from the two cell lines separately . These networks are encoded by the files enets8 . GM_proximal_filtered_network . txt and enets7 . K562_proximal_filtered_network . txt respectively . The files are available at http://encodenets . gersteinlab . org/ . The murine dendritic cell GRNs were constructed using the interactions reported by Garber et al . ( 2012 ) available at http://www . weizmann . ac . il/immunology/AmitLab/data-and-method/iChIP/data . Only the interactions between transcription factors were considered . An edge is inserted from gene A to gene B at time T ifThe protein product of A binds to a promoter area of BThe combined score for the binding is larger that 26 . 9The score for the binding at time T is larger than 26 . 9 Note that the threshold value of 26 . 9 was extracted from the experimental procedure of Garber et al . ( 2012 ) . The P . aeruginosa GRN was constructed from the dataset provided by Galan-Vasquez et al . ( 2011 ) . No filtering was applied and all the interactions were considered; therefore a perceivable level of false positives and negatives is to be expected . The M . tuberculosis GRN was constructed from the dataset provided by Sanz et al . ( 2011 ) . The dataset includes a list of evidence codes for each interaction . However , most interactions are supported by only one evidence code . Therefore , no filtering was applied and all the interactions were considered . Similarly to P . aeruginosa , a perceivable level of false positives and negatives is to be expected . The yeast dataset provided by Luscombe et al . ( 2004 ) does not include any systemic information on the statistical validity of the interactions . Therefore no filtering was applied and a perceivable level of false positives and negatives is to be expected . The yeast dataset provided by Lee et al . ( 2002 ) was treated in the same way as Harbison et al . ( 2004 ) and only interactions supported by a p-value lower than 10−3 were considered for the general analysis . To provide compatibility with the statistical conditions used in Harbison et al . ( 2004 ) and Lee et al . ( 2002 ) , the yeast dataset provided by MacIsaac et al . ( 2006 ) was constructed using the file orfs_by_factor_p0 . 001_cons2 . txt . This file is available at http://fraenkel . mit . edu/improved_map/ . Table 1 reports the number of genes , the number of transcriptional interactions and the network density for the full networks , identified by the ( F ) , and for the networks composed by the transcription factor and the interaction among themselves , identified by ( T ) , for all genome-wide datasets used in the article . 10 . 7554/eLife . 02863 . 038Table 1 . Basic network properties for all the genome-wide GRNs used in the articleDOI: http://dx . doi . org/10 . 7554/eLife . 02863 . 038OrganismNodes ( F ) Edges ( F ) Nodes ( T ) Edges ( T ) Density ( F ) Density ( T ) E . coli147029091541400 . 00130 . 0059M . tuberculosis1624316982850 . 00120 . 013P . Aeruginosa69299185810 . 00210 . 011Yeast ( Lee ) 24174348106960 . 000740 . 0086Yeast ( Harbison ) 293361521691880 . 000710 . 0066Yeast ( Luscombe ) 345970531422540 . 000590 . 013Yeast ( MacIsaac ) 207940971161340 . 000950 . 010GM12878404911 , 70782840 . 000710 . 013K5624071846667590 . 000510 . 013Except when stated otherwise , random networks were generated preserving the number of transcription factors , genes , and interactions for E . coli , P . aeruginosa , M . tuberculosis , yeast , and the human cell lines . Random networks generated by preserving additional topological properties were also considered ( See the subection ‘Effect of different constraints on the generation of random networks’ ) and confirm the results discussed . For murine dendritic cells , random networks were constructed by preserving the number of transcription factors and interactions among them . Note that consequently the number of feedback loops and incomplete feedback loops is an underestimate with respect to the data considered in the other random networks . Self-regulating interactions were ignored . Since genes not encoding for TFs do not regulate other genes , a full-network analysis would artificially increase the stability property of the network . Therefore , to limit the bias introduced by the limited number of transcription factors , the number of incomplete loops , the motif abundance , the number of regulatory connections , and the probability of adding additional long feedback loops when a new regulatory connection is inserted were computed considering only the interactions among transcription factors . For each different type of random network , 1000 instances were generated for the data presented in the main Figures , while 100 instances were generated for the data presented in the Figure Supplements . Feedback loops and incomplete feedback loops were computed by counting the number of sub-isomorphisms from the feedback or incomplete feedback loops for each GRN under consideration . For feedback loops this value was divided by the length of the loop to account for automorphisms . Note that , due to the nature of the analysis used ( sub-isomorphism count ) , all the possible ways in which a feedback loop or incomplete feedback loop can be created in the network are considered separately . Therefore , an edge or node is generally counted multiple times . The analysis is indicative of the general structure of the network , but can lead to counter-intuitive results . Graphical motifs were computed in the usual way ( Milo et al . , 2002 ) , but due to the different theoretical approach , no direct comparison with random networks was performed . The probability of feedback creation by random addition of an edge in real GRNs was computed by trying all the possible edge insertions between the TFs and taking the ratio of insertions that form long feedback loops over the total number of insertions . This approach was computationally unfeasible for random networks and a sampling procedure was used: for each random network , 100 , 000 independent random insertions of one connection from two randomly selected transcription factors were tried and the probability of interest was estimated by considering the value:NS100000 , where NS is the number of insertions that resulted in the creation of long feedback loops . Additional details on the comparison of GM12878 and K562 are presented in Supplementary file 1 . Simulations and analyses were performed using R version 2 . 15 . 1 ( R Core Team , 2012 ) and the ‘igraph’ package version 0 . 6-2 ( Csardi and Nepusz , 2006 ) . It is common practice in statistics to use random simulations as a null model to test whether a feature can emerge with a high probability by chance . However , selecting the right type of randomness can be problematic . Network theory is no exception in this regard . Different constraints can be built into a simulation to obtain different types of random networks . To assess the role of different types of randomness on the properties of Qualitative Stability , we generated different types of random networks with different constraints using a variable number of characteristics of the E . coli network discussed in the main article . We focused our attention on four characteristics of the real networks:The number of genes ( the number of vertices of the random network ) The number of interactions among the genes ( the number of edges of the random network ) The number of transcription factors ( the number of vertices allowing an out-degree larger than zero ) The absence of isolated genes ( the absence of non-connected vertices ) Enforcing the number of vertices and edges of random graph is a common feature of random graphs: these random graphs are called Erdős–Rényi random graphs ( Bollobás , 2001 ) . However , it is less common in the literature to enforce additional characteristics . We stress that including additional constraints results in a network that is less ‘random’ . The types of random networks that we considered are detailed in Table 2 . The algorithms used to generate the different types of random networks have been implemented in R and are available as Source code . The functions used are listed by Table 3 . Note that TFFixedIGNot AllowedV1 and TFFixedIGNot AllowedV2 use different algorithms to generate the networks:TFFixedIGNot AllowedV1 generates an initial random network constructed by connecting each gene to one TF ( selected at random ) . This ensures that no isolated genes are present . Then , if the number of edges used is less than the number required , edges are added at random until the expected number of edges is reached . TFFixedIGNot AllowedV1 randomly adds edges between TFs and genes until a network with no isolated genes has been generated . At this point , if the number of edges used is less than the number required , edges are added at random until the expected number of edges is reached . Alternatively , if the number of edges is more than the number required , edges are removed at random , in such a way as not to create isolated genes , until the expected number of edges is reached . 10 . 7554/eLife . 02863 . 039Table 2 . Properties of the different random models usedDOI: http://dx . doi . org/10 . 7554/eLife . 02863 . 039Model name# of genes# of interactions# of TFsIsolated genesTFVariableIGAllowedFixedFixedVariableAllowedTFFixedIGAllowedFixedFixedFixedAllowedTFVariableIGNot AllowedFixedFixedVariableNot allowedTFFixedIGNot AllowedV1FixedFixedFixedNot allowedTFFixedIGNot AllowedV2FixedFixedFixedNot allowed10 . 7554/eLife . 02863 . 040Table 3 . Functions used to produce the different random networksDOI: http://dx . doi . org/10 . 7554/eLife . 02863 . 040Model nameFunction usedTFVariableIGAllowedGenerateRandomNetwork . IG . Allowed ( ) TFFixedIGAllowedGenerateRandomNetwork . IG . Allowed ( ) TFVariableIGNot AllowedGenerateRandomNetwork . IG . NotAllowed . V1 ( ) TFFixedIGNot AllowedV1GenerateRandomNetwork . IG . NotAllowed . V1 ( ) TFFixedIGNot AllowedV2GenerateRandomNetwork . IG . NotAllowed . V2 ( ) TFFixedIGNot AllowedV1 should be expected to be less biased , but is computationally extremely intensive , as a large number of edges usually needs to be removed . TFFixedIGNot AllowedV2 uses a more biased procedure , but is more computationally tractable . As indicated by Figure 2—figure supplement 4A , B , Figure 3—figure supplement 4 , and Figure 5—figure supplement 4A–L , the conclusions of the main text remain valid under all the conditions considered . In our analysis we focused on a minimal number of constraints , in such a way to be able to assess the selective pressure of robustness at all scales . A detailed analysis of the adherence to BQS of different types of random networks is beyond the scope of this article and will be the subject of future investigations . However , given the existence of widely used more advanced models , it is important to justify our choice . BQS provides predictions on GRNs at many scales , including degree distribution and motifs abundance , and the clear compliance of GRNs to such predictions indicates that robustness has left striking and detectable signatures . Therefore , using a method designed to preserve features such as degree distribution is likely to result in a network carrying the seed of robustness . To test this hypothesis , we assessed the effect of the degree-preserving model commonly used in motif analysis ( Milo et al . , 2002 ) for the E . coli GRN . This model , implemented by the function ‘rewire’ of igraph , reshuffles the original network in such a way to completely preserve the original in-degree and out-degree for each single node . Our results ( not shown ) indicate that feedback loops are relatively uncommon in degree-preserving random networks , regardless of the number of rounds of rewiring . However , incomplete feedback loops were more abundant and longer when compared to the real GRNs , even though to a lesser extent than purely random networks . Finally , the distributions of 3- and 4-node motifs were perturbed in such a way as to decrease the number of buffered motifs and to increase the number of unbuffered ones . However , the degree-preserving algorithm was unable to generate networks displaying an equal number of the 3- and 4-node motifs for the classes highlighted in the article , in stinking difference from a purely random model , even after 70 , 000 rounds of rewiring . Taken together , these observations support the idea that GRNs carry a strong signature of BQS and that random models constrained to be similar to GRNs will inherit , at least in part , many features of BQS . Our results also support the idea that purely random models may be the ideal null models to explore and highlight pervasive features of networks . | The genomes of living organisms consist of thousands of genes , which produce proteins that perform many essential functions . Cells receive signals from both their internal and external environments , and respond by changing how they express their genes . This allows a cell to make the right amount of different proteins when needed . The proteins that a cell produces can then , in turn , influence how the cell's genes are expressed . This set of interactions between genes and proteins is called a gene regulatory network , and is akin to a computer program that the cell runs to define its behaviour . At present , we understand very little about why these networks take on the forms seen in living cells . A remarkable feature of living organisms is their ability to withstand an extremely wide variety of predicaments , such as DNA damage , physical trauma or exposure to toxins . This ability , generally called robustness , requires a cell to rapidly activate different gene sets and maintain their activity for as long as necessary . However , very little is known about how cells are programmed to respond appropriately , whatever happens , and keep themselves in a stable state . Albergante et al . propose that a fully robust gene regulatory network should be able to stabilize itself . This means that the robustness of a gene regulatory network should only depend on how it is wired up , and not on quantitative changes to any features that may change unpredictably—for example the concentration of a protein . By analysing data that is already available about gene regulatory networks in a wide selection of organisms ranging from bacteria to humans , Albergante et al . show that all known gene regulatory networks are wired up in a way that any quantitative change to the network will not cause the state of network to change . In addition , gene regulatory networks tend to remain stable even if new regulatory links are randomly added . Albergante et al . call this property Buffered Qualitative Stability ( BQS ) : the network is qualitatively stable because its state does not change when the activity of particular regulatory links in the network changes , and it is buffered against its stability being compromised by the random addition of new links . Albergante et al . also found that the gene regulatory network of a cancer cell does not match up with the predictions of BQS , suggesting that the robustness of the network is compromised in these cells . This could explain why cancer cells are able to easily change their characteristics in response to changes in the environment . In addition , using BQS to analyse the gene regulatory network of bacteria such as E . coli reveals points in the network that , if disrupted , would make the network unstable , potentially harming the cell . Therefore , in the future , an understanding of BQS could help efforts to design new drugs to treat a range of infections and diseases . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"genetics",
"and",
"genomics"
] | 2014 | Buffered Qualitative Stability explains the robustness and evolvability of transcriptional networks |
Many animal species , including insects , are capable of acoustic duetting , a complex social behavior in which males and females tightly control the rate and timing of their courtship song syllables relative to each other . The mechanisms underlying duetting remain largely unknown across model systems . Most studies of duetting focus exclusively on acoustic interactions , but the use of multisensory cues should aid in coordinating behavior between individuals . To test this hypothesis , we develop Drosophila virilis as a new model for studies of duetting . By combining sensory manipulations , quantitative behavioral assays , and statistical modeling , we show that virilis females combine precisely timed auditory and tactile cues to drive song production and duetting . Tactile cues delivered to the abdomen and genitalia play the larger role in females , as even headless females continue to coordinate song production with courting males . These data , therefore , reveal a novel , non-acoustic , mechanism for acoustic duetting . Finally , our results indicate that female-duetting circuits are not sexually differentiated , as males can also produce ‘female-like’ duets in a context-dependent manner .
Studies of acoustic communication focus on the production of acoustic signals by males and the arbitration of mating decisions by females . However , for many species of primates ( Haimoff , 1986 ) , birds ( Hall , 2004 ) , frogs ( Tobias et al . , 1998 ) , and insects ( Bailey , 2003 ) , females also produce songs , and duets are common; moreover , recent studies suggest that female song production may be ancestral ( Wiens , 2001; Odom et al . , 2014 ) . Animal duets involve predictable response latencies between the calls of males and females; that is , males and females do not sing simultaneously , as human duetters do , but rather interchange acoustic signals ( Bailey and Hammond , 2003 ) . Duetting species can be grouped into two classes: those that answer their partner's song without fine-scale coordination of song syllables or elements ( polyphonal duetters ) and those that synchronize within a song bout ( antiphonal duetters ) ( Hall , 2009 ) . Regardless of duet type , each individual must adjust the rate and timing of his/her courtship songs relative to each other . Therefore , duetting requires speed and accuracy in both detection of a partner's signal and the production of a response . Latencies between male and female songs are reported to be as short as tens of milliseconds for some birds ( Logue et al . , 2008; Fortune et al . , 2011 ) and insects ( Heller and von Helversen , 1986; Rheinlaender et al . , 1986 ) . For example , the antiphonal duets of plain-tailed wrens are so rapid that they sound as if produced by a single animal ( Mann et al . , 2006 ) . Both male and female wrens display differences in inter-syllable intervals when singing alone vs with a partner , indicating that sensory perception plays an ongoing role in shaping song timing ( Fortune et al . , 2011 ) . In support of this , recordings from pre-motor areas of the wren brain ( HVC ) reveal auditory responses that are tuned to duets ( Fortune et al . , 2011 ) ; that is , neural centers that drive song production also directly integrate auditory information . For polyphonal duetting bushcrickets , which like many insects produce sounds via their wings , auditory information is transmitted not only to the brain but also directly to wing-controlling neural ganglia; thus , similar to birds , this suggests that song pattern-generating circuits ( housed in these ganglia ) directly integrate auditory information to shape duets ( Rheinlaender et al . , 1986 ) . While auditory perception appears to play an important role in producing duets , no studies to date have investigated a role for non-acoustic cues . However , for many other behaviors , the use of multisensory cues is proposed to improve signal detection ( Rowe , 1999; Mirjany et al . , 2011; McMeniman et al . , 2014 ) , synchronization ( Elliott et al . , 2010 ) , or decision-making abilities ( Kulahci et al . , 2008; Raposo et al . , 2012 ) . For example , mosquitos locating human hosts rely on both thermal and olfactory cues; this dual requirement increases the fidelity of host-seeking ( McMeniman et al . , 2014 ) . Similarly , bees have been shown to make faster and more accurate-foraging decisions when choosing between flowers that differ in multiple modalities ( vs a single modality ) ( Kulahci et al . , 2008 ) . These studies ( and others ) lend support to the idea that acoustic-duetting behaviors , which are under tight temporal control , might also benefit from the use of multimodal sensory processing . Moreover , because many species produce songs that are not stereotyped ( Kovach et al . , 2014 ) , the use of multiple sensory cues should aid in response timing . To test this hypothesis , we developed Drosophila virilis as a new model system for the study of acoustic coordination . Song production ( via wing vibration ) is an integral component of the courtship ritual among Drosophilid flies ( Ewing and Bennet-Clark , 1968 ) , but typically , as in Drosophila melanogaster , only males produce song ( Dickson , 2008 ) . The presence of female song has been reported in only a handful of Drosophila species ( Donegan and Ewing , 1980 ) , and it is not known if any of these species duet . Here , by combining a battery of specific sensory manipulations , quantification of a large number of courtship songs , and statistical modeling , we not only show that D . virilis males and females duet but also uncover the underlying mechanisms . We find that females integrate both auditory ( male song ) and tactile ( male contact with her abdomen and genitalia ) cues for song production and coordination . Moreover , the precise timing of male tactile cues predicts female song timing , revealing a novel , non-acoustic , mechanism for acoustic coordination . Finally , we also address the importance of male and female song production for courtship success , and demonstrate , by comparing duets produced in male–male pairings , that acoustic-duetting behavior in females is not sexually differentiated .
Previous studies have documented the presence of female song in the D . virilis group of species ( Satokangas et al . , 1994 ) ; however , it is not known if males and females coordinate their song production into duets nor have the mechanisms underlying female song production been studied . By combining video and audio recordings , we matched D . virilis male and female wing movements with acoustic signals to accurately classify male and female song ( Figure 1A–C and Video 1 ) . We recorded song in a multi-channel recording apparatus ( Arthur et al . , 2013 ) . We chose virilis strain 15010–1051 . 47 , because males and females of this strain sang robustly in our courtship chambers ( Figure 1—figure supplement 1 ) . Consistent with previous reports ( Huttunen et al . , 2008 ) , we found that virilis males ( when paired with a virgin female ) produce highly stereotyped bouts of pulses via unilateral wing vibration ( Figure 1A , D ) . We define a bout as a stretch of song from either a male or female , which is separated from the next stretch of song by more than 150 ms . Each male bout contains 6 . 9 ± 1 . 2 pulses with an inter-pulse interval ( IPI ) of 21 . 2 ± 1 . 9 ms ( Figure 1E ) . Females ( when paired with a virgin male ) , in contrast , use bilateral wing vibration to generate variable-length bouts of pulses separated by longer ( and more variable ) IPIs ( 7 . 2 ± 6 . 2 pulses per bout , with IPIs of 55 . 2 ± 26 . 3 ms ) ( Figure 1—figure supplement 2 and Figure 1E ) . The frequency spectra of individual male pulses show a peak at higher frequencies relative to individual female pulses ( Figure 1F ) . Although there are instances of song overlap between males and females , this behavior is extremely rare ( Figure 1C and Figure 1—figure supplement 3E [population median of overlaps = 0%] ) . Based on these characteristics , we modified our software for automated segmentation of D . melanogaster song ( Arthur et al . , 2013 ) to segment virilis male and female pulses ( Figure 1—figure supplement 3 ) . 10 . 7554/eLife . 07277 . 003Figure 1 . Drosophila virilis males and females produce distinct courtship songs . Schematic of the wing movement during the production of courtship song in Drosophila virilis male ( A ) and female ( B ) . ( C ) Examples of acoustic duets between male ( blue ) and female ( red ) in five wild-type ( WT ) courtships . Regions of song overlap ( arrow ) are shaded in green . ( D ) Detailed view of male ( blue ) and female ( red ) song produced during courtship . Song parameters described in the ‘Materials and methods’ and used in all analyses are indicated . IBI = inter-bout interval and IPI = inter-pulse interval . ( E ) The median ( per individual ) IPI for male ( blue ) and female ( red ) song; male and female pulses were identified by matching acoustic and video recordings ( Video 1 ) . Black squares show population mean and black bars standard deviation ( n = 13 courtships , Student's t-test ***p < 0 . 001 ) . ( F ) Power spectral density for pulses from males ( n = 3854 , blue ) and females ( n = 497 , red ) . Based on the differences in IPI and pulse frequency between male and female song , we created a semi-automated software pipeline to segment virilis song ( Figure 1—figure supplement 2 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 07277 . 00310 . 7554/eLife . 07277 . 004Figure 1—figure supplement 1 . Assessment of song production in D . virilis strains . We first chose a virilis strain to work with based on male bout ( blue ) and female pulse ( red ) rates , comparing five different D . virilis strains ( n = 10 recorded courtships each , see ‘Materials and methods’ ) that produced song in our behavioral chambers . Black squares represent population median and black bars the IQR . DOI: http://dx . doi . org/10 . 7554/eLife . 07277 . 00410 . 7554/eLife . 07277 . 005Figure 1—figure supplement 2 . Bout durations of male and female courtship songs . Histogram of male ( A , n = 5995 bouts ) and female ( B , n = 1224 bouts ) bout durations in 67 pairings . DOI: http://dx . doi . org/10 . 7554/eLife . 07277 . 00510 . 7554/eLife . 07277 . 006Figure 1—figure supplement 3 . Development of song segmentation software for Drosophila virilis . ( A ) Diagram of FlySongSegmenter-Virilis , a MatLab-based software for automated song segmentation . The raw microphone signal is transformed by a wavelet basis scaled at frequencies from 100 to 900 Hz . The frequency profile of detected pulses is compared to frequency templates for male ( blue ) and female ( red ) pulses , overlapping regions ( green ) , and noise ( black ) . Likelihood values determine the group assignment of each pulse . The sensitivity ( B ) , positive predictive value ( PPV , C ) , and the harmonic mean ( D ) for male bout ( blue ) and female pulse ( red ) calls ( n = 15 courtships ) , calculated by comparing manual versus automated segmentation ( see ‘Materials and methods’ ) . ( E ) The percent overlap ( see ‘Materials and methods’ ) in WT courtships ( n = 89 courtships ) . Black squares represent population median and black bars the IQR . DOI: http://dx . doi . org/10 . 7554/eLife . 07277 . 00610 . 7554/eLife . 07277 . 007Video 1 . D . virilis duetting behavior . Representative example of courtship between a naive D . virilis male and female pair . Shown in the video is one channel/chamber of a multi-channel song recording system . Video is acquired at 15 Hz . DOI: http://dx . doi . org/10 . 7554/eLife . 07277 . 007 Acoustic duets are characterized by predictable response latencies between male and female calls ( Bailey , 2003 ) . To determine if virilis males and females duet during their courtship ritual , we characterized response times between male bouts ( which are highly stereotyped , see Figure 1 ) and female pulses and vice versa . The distribution of female response times ( relative to the onset of a male bout ) is peaked at 409 ms and is significantly different from the distribution of response times calculated from randomized versions of the data set in which either male bout times or female pulse times were shuffled ( Figure 2A and see ‘Materials and methods’ ) . Thus , female song is temporally coordinated with male song . Male response times to female song have not been examined previously in any insect . We calculated the delay between the onset of male bouts and the center of each pulse in the preceding female bout ( Figure 2—figure supplement 1 ) . Male response times to the first , second , penultimate , and last pulse in a female bout were all significantly different from response times from randomized data , with responses to the last pulse showing the largest difference ( Figure 2—figure supplement 1C ) . We , therefore , defined male response times as the latency between the onset of the male bout and the previous female pulse ( Figure 2B ) and found that the distribution of male response times is peaked at short delays ( 110 ms ) , compared with female response times ( even when accounting for differences in how response times are calculated for males and females ) . To the best of our knowledge , these data represent the first quantitative evidence of acoustic duetting in a Drosophilid species . 10 . 7554/eLife . 07277 . 008Figure 2 . Drosophila virilis courting pairs coordinate song production into an acoustic duet . ( A ) Female response times were calculated as the time between male bout onset and the following female pulse ( inset ) . ( B ) Male response times were calculated as the time between the onset of the male bout and the preceding female pulse ( inset ) . Normalized distributions of female ( red , A , C , and E ) and male ( blue , B , D , and F ) response times for: WT females paired with WT males ( A , n = 970 response times and B , n = 1031 response times , n = 89 courtships ) , WT females paired with arista-cut ( AC , deaf ) males ( C , n = 1173 response times and D , n = 1206 response times , n = 124 courtships ) , and AC ( deaf ) females paired with WT males ( E , n = 3391 response times and F , n = 3948 response times , n = 34 courtships ) . Response times were also calculated after shuffling either male bout ( blue dashed ) or female pulse ( red dashed ) intervals . Comparisons with these null distributions reveal that only removing male hearing impacts response timing ( Kolmogorov–Smirnov ( K-S ) Test ***p < 0 . 001 **p < 0 . 01 ) . Basic song structure remains stable for the AC manipulations ( Figure 1—figure supplement 2 ) . Transparent blue bar ( A , C , and E ) indicates average male bout duration . Shading around each response time distribution represents bootstrapped 95% confidence intervals . DOI: http://dx . doi . org/10 . 7554/eLife . 07277 . 00810 . 7554/eLife . 07277 . 009Figure 2—figure supplement 1 . Quantification of male response times . ( A ) Representative female pulse train preceding a male bout , where male response times were calculated from the first pulse forward ( cool color scheme , from the same data set , n = 803 , 749 , 622 , and 588 , respectively , n = 83 courtships ) and the last pulse backward ( warm color scheme , from the same data set , n = 1031 , 935 , 821 , 721 , respectively , n = 83 courtships ) . ( B ) Normalized male response time distributions to the various female pulses within a female pulse train ( *p < 0 . 05 compared to null distribution , Kolmogorov–Smirnov Test ) . ( C ) Kullback–Leibler ( K-L ) divergence values for comparisons between the response time distributions to the first , second , penultimate , and last female pulse within a pulse train , and the same response times following shuffling of male bout intervals . Larger K-L divergence values indicate that response time distributions are more distinct than the corresponding null distribution . DOI: http://dx . doi . org/10 . 7554/eLife . 07277 . 00910 . 7554/eLife . 07277 . 010Figure 2—figure supplement 2 . Song parameters of AC flies . Median individual male IBI ( A ) and female IPI ( B ) from songs from WT ( WT , n = 83 courtships ) or AC males ( n = 124 courtships ) and females ( n = 34 courtships , *p < 0 . 05 Kruskall–Wallis test ) . Black squares represent population median and black bars the IQR . The power spectral density of male ( blue ) and female ( red ) pulses from AC males paired with WT females ( C ) and AC females paired with WT males ( D ) . DOI: http://dx . doi . org/10 . 7554/eLife . 07277 . 010 To determine if response timing relies on hearing the song of the partner , we rendered either the male or female deaf by removing the arista , a feathery appendage of the antenna that serves as the acoustic receiver ( Gopfert and Robert , 2002 ) . Intact females maintained their response timing in pairings with arista-cut ( AC ) males ( Figure 2C , peak at 407 ms ) ; however , song bouts from AC males no longer followed female pulses with a predictable response latency ( Figure 2D ) . Instead , AC male response times were not significantly different from response times from randomized versions of the data ( p = 0 . 24 ) . This effect was specific to response timing , because AC males maintained wild-type ( WT ) levels of song production ( Figure 4B , manipulation 3 ) , and showed no change in inter-bout interval ( IBI ) or pulse frequency ( Figure 2—figure supplement 2 ) . Thus , males rely on hearing their partner's song for acoustic coordination . In contrast , AC females paired with intact males maintained coordination ( Figure 2E , peak at 408 ms ) , even though this manipulation caused an up-regulation in song production ( Figure 4E , manipulation 3 ) and a shift in IPI and pulse frequency ( Figure 2—figure supplement 2 ) . This result suggests that females may use a non-auditory cue to coordinate song production with their partner . To determine the role of song in mating success , we removed male or female acoustic cues by either amputation of the wings ( rendering the fly mute ) or the aristae ( rendering the fly deaf ) . Removal of either appendage did not reduce courtship rates ( Figure 3—figure supplement 1 ) . However , removing either the production of male song ( in pairings between WT females and wing-cut ( WC ) males ) or the detection of male song ( in pairings between WT males and AC females ) dramatically reduced the percentage of copulating pairs ( Figure 3A ) . These results indicate that male song is required for mating success , consistent with data from other closely related species ( Aspi and Hoikkala , 1993 ) . Because females increase song production rates ( a parameter distinct from IPI , see ‘Materials and methods’ ) in the absence of male song ( Figure 3—figure supplement 2 ) , increased female song production might cause the observed decrease in copulation rates . Regardless , both interpretations support a role for male song , either directly or indirectly , in mating success . 10 . 7554/eLife . 07277 . 013Figure 3 . Testing the role of male and female song in courtship success . ( A and B ) Percent copulation over the 30-min observation period for manipulated flies ( WC , wing-cut [mute]; AC , dark , lights off [no visual cues] ) . Each manipulated fly was paired with a WT fly of the opposite sex ( solid lines ) . Latency to mating was compared to WT controls prepared under similar conditions ( dashed lines ) . The numbers of individuals in each experiment were: WC male n = 20 , WC male control n = 22 , AC female n = 21 , AC female control n = 23 , WC female n = 24 , WC female control n = 23 , AC male n = 20 , AC male control n = 24 , AC male dark n = 41 , AC male control dark n = 37 ( Mantel–Cox test , ***p < 0 . 001 ) . Experimental manipulations did not affect courtship rates ( Figure 3—figure supplement 1 ) . ( C ) Percent copulation over the 30-min observation period for mated females paired with virgin males ( solid black , n = 28 ) vs virgin females paired with virgin males ( dashed black , n = 28 , Mantel–Cox test , **p < 0 . 01 ) . ( D ) Median female pulse ( red ) or male bout ( blue ) rate ( see ‘Materials and methods’ ) dependent on whether the female is virgin ( v , n = 83 ) or mated ( m , n = 26 ) , independent of copulation success ( Figure 3—figure supplement 2 ) . Black squares are the population median and black bars the interquartile range ( IQR ) ( generalized linear model [GLM] t statistic **p < 0 . 01 ) . ( E ) Mean female pulse number ( normalized ) in 10-s bins prior to the moment of copulation ( n = 8 courtships ) . Shading represents s . e . m . DOI: http://dx . doi . org/10 . 7554/eLife . 07277 . 01310 . 7554/eLife . 07277 . 014Figure 3—figure supplement 1 . Courtship activity of manipulated flies in latency to mating assays . Percent of pairs performing courtship behaviors in latency to mating assays in Figure 3: WT—wild type , AC—arista-cut , WC—wing-cut , v—virgin , m—mated , ( + ) —lights on , ( − ) —lights off . DOI: http://dx . doi . org/10 . 7554/eLife . 07277 . 01410 . 7554/eLife . 07277 . 015Figure 3—figure supplement 2 . Female pulse rates in the absence of male acoustic cues . Female pulse rate in the absence of male acoustic cues whether the male is WC ( n = 10 ) or the female is AC ( n = 34 , ***p < 0 . 001 , Kruskal–Wallis Test ) . Black squares represent population median and black bars the IQR . DOI: http://dx . doi . org/10 . 7554/eLife . 07277 . 01510 . 7554/eLife . 07277 . 016Figure 3—figure supplement 3 . Song rates in successful versus unsuccessful courtships . WT male bout ( A ) and female pulse ( B ) rate ( see ‘Materials and methods’ ) dependent on whether the courtship ( within 20 minutes ) resulted in copulation ( cop , n = 33 courtships , filled circle ) or was unsuccessful ( no cop , n = 44 courtships , open circle , ***p < 0 . 001 *p < 0 . 05 Wilcoxon rank-sum test ) . Black squares represent population median and black bars the IQR . DOI: http://dx . doi . org/10 . 7554/eLife . 07277 . 016 To test the role of female song in mating , we first examined WT males paired with mute ( WC ) females and found a significant reduction in copulation ( Figure 3B , yellow solid vs yellow dashed control ) . However , deaf males , which cannot detect female song , still copulated at WT rates when paired with WT females ( Figure 3B , light blue solid vs light blue dashed control ) . Because virilis females indicate their receptivity via wing spreading ( Vuoristo et al . , 1996 ) , deaf males may use this visual signal ( produced by intact females ) to drive mating in the absence of acoustic cues . To prevent males from observing female wing spreading , we allowed deaf males to court WT females in the dark and subsequently observed a 29% reduction in copulation ( Figure 3B , black solid vs black dashed control , p = 0 . 08 ) . The lack of significance of the reduction in courtship success may be due to the ability of males to still detect female wing spreading in the dark ( e . g . , using mechanosensation ) . Because AC males do not coordinate their song production with females ( Figure 2D ) but still maintain high-copulation success ( Figure 3B ) , our data also suggest that females don't require coordinated duetting for mating when paired with a single male . Duetting , however , may be important for mate selection in the context of competition as has been suggested for other insects and some species of birds ( Bailey , 2003; Hall , 2004 ) . A role for female song in mating would nonetheless be indicated by a correlation between female song production and her receptivity state . To test this hypothesis , we compared courtship success and song production rates ( see ‘Materials and methods’ ) of virgin females and mated females . Mated D . melanogaster females show a reduction in copulation rates following mating; this is due to the effects of a sex peptide transferred from males to females during copulation ( Villella and Hall , 2008; Yapici et al . , 2008 ) . Similarly , we found that mated virilis females show significantly lower copulation rates ( Figure 3C ) . We , therefore , investigated the association between song production and state of the female . Both virgin virilis males and females sang more if they copulated within the 30-min experimental window vs if they did not copulate ( Figure 3—figure supplement 3 ) . After correcting for this difference ( see ‘Materials and methods’ ) , we still observed a significant reduction in song production in mated virilis females compared with virgin females ( Figure 3D ) ; this reduction was specific to female song , as males who courted mated vs virgin females produced similar amounts of song . Finally , we observed that virilis females , on average , increased song production over the 400 s leading to copulation ( Figure 3E ) . These data collectively argue that female song production ( and therefore duetting ) is not required for mating , but rather is a positive cue that promotes mating . In order to duet , both males and females must regulate their song production rates relative to each other . We next examined the sensory stimuli and pathways that influence male bout and female pulse rates in D . virilis . We examined the rate of male bouts and female pulses , because male bouts are highly stereotyped while female bouts are not ( Figure 1—figure supplement 2 ) ; see ‘Materials and methods’ for a description of how song rates were calculated . We manipulated sensory structures relevant for courtship ( Figure 4A and see ‘Materials and methods’ ) and used statistical modeling to predict song rates from data sets containing all possible combinations of sensory organ manipulations to the male or female ( each paired with a WT partner ) . Based on the interactions we observed in videos of virilis courtship ( Video 1 ) , we manipulated the following sensory structures: ( i ) aristae ( to block the detection of acoustic signals [Gopfert and Robert , 2002] ) , ( ii ) antennae ( to block the detection of volatile pheromones [Benton , 2007] ) , ( iii ) tarsi ( to block the detection of contact pheromones [Thistle et al . , 2012] ) , and ( iv ) female genitalia ( to block male contact with the lower abdomen , which is known to be important for D . virilis courtship [Vuoristo et al . , 1996]; see ‘Materials and methods’ ) . To block visual cues , we placed flies in the dark . We used generalized linear models ( GLMs ) to predict male or female song rates based on the sensory manipulations ( see ‘Materials and methods’ ) . The GLM weights or coefficients reveal the contribution of each sensory channel to song production . We first fit the GLM to data from pairings between manipulated males and WT females and found that bilateral removal of the male tarsi was the only significant predictor of male bout rate ( Figure 4B and Figure 4—figure supplement 1 ) . This result contrasts with results from D . melanogaster , where tarsal removal enhances courtship activity in males ( Fan et al . , 2013 ) . We also found that sensory organ manipulation of males has similar effects on female song production ( Figure 4C and Figure 4—figure supplement 1 ) , suggesting that either female song production is dependent on the presence of male song or that some male manipulations also impact sensory perception in females ( for example , removal of male tarsi may block a mechanosensory cue to the female; see below ) . 10 . 7554/eLife . 07277 . 011Figure 4 . Multiple sensory channels influence song production rates in virilis males and females . ( A ) Schematic of sensory organ manipulations . Orange shading indicates the region of organ modification ( amputation or blocking ) . Male bout ( blue , B ) and female pulse ( red , C ) rate in pairings between sensory organ manipulated males and WT females . Sensory organ manipulation for each experiment is indicated with a black box . Each colored bar represents the median bout ( blue , B ) or pulse rate ( red , C ) and error bars the IQR . The number of pairs examined is listed below the plots . A GLM was used to predict the song rate based on the presence/absence of different sensory channels . GLM coefficients indicate the relative importance of each sensory channel in determining male song production ( light blue ) and female song production ( pink ) ; error bars represent bootstrap estimate of the s . e . m . ( ***p < 0 . 001 , *p < 0 . 05 ) . Similarly , male bout ( blue , D ) and female pulse ( red , E ) rate , along with GLM coefficients ( light blue and light pink , respectively ) , for pairings between WT males and sensory organ manipulated females . DOI: http://dx . doi . org/10 . 7554/eLife . 07277 . 01110 . 7554/eLife . 07277 . 012Figure 4—figure supplement 1 . GLM statistics for sensory organ manipulations . GLM statistics for male manipulations and predictions of female and male song rate and for female manipulations and predictions of female and male song rate . DOI: http://dx . doi . org/10 . 7554/eLife . 07277 . 012 We next performed a similar panel of sensory manipulations on the female . We found that none of the female manipulations ( either alone or in combination ) impacted male song production rates ( Figure 4D ) ; consequently none of the GLM coefficients were significantly different from zero ( Figure 4—figure supplement 1 , p = 0 . 68 ) . Thus , male song production is unaffected by the state of the female . However , our GLM analysis revealed that all of the evaluated female sensory manipulations impacted female pulse rate ( Figure 4E and Figure 4—figure supplement 1 ) . Interestingly , inputs to the female aristae and genitalia have the largest effects on female song production , but act in opposite directions ( Figure 4E ) . Removal of the aristae up-regulates female pulse rate ( as does removing male song [Figure 3—figure supplement 2] ) , while blocking the female genitalia down-regulates pulse rate . These results establish a role for multisensory cues in female song rate and indicate that information sensed via the female genitalia and aristae is combined to fine-tune how much song the female produces . We , therefore , next focused on these two sensory pathways to test their role in regulating female song timing relative to her partner . Data from AC females indicated that male auditory cues are not necessary for female song response timing ( Figure 2E ) . To further investigate the role of auditory cues in duetting , we examined female acoustic coordination with playback of recorded male song ( see ‘Materials and methods’ ) . Previous studies of the D . virilis species group showed that females respond to the playback of male song by spreading their wings ( Ritchie et al . , 1998; Isoherranen et al . , 1999 ) , but these studies did not characterize female song production . We were unable to detect any female song in response to either playback alone or to playback in the presence of a decapitated male that does not interact with the female ( data not shown ) . However , females produced abundant song in the presence of a WC male ( Figure 5A ) . For these pairings , we calculated female response times relative to the playback stimulus and found the distribution to be significantly different from response times from shuffled data ( Figure 5B , p < 0 . 001 ) . However , the distribution of female response times to playback does not resemble the WT distribution ( compare with Figure 2A ) . Due to issues ( in only the playback experiments ) with identifying female pulses that overlap with the male playback ( see ‘Materials and methods’ ) , we also calculated female response times from the end of the male bout ( Figure 5B inset ) . We found that this distribution more closely matched the distribution of response times from shuffled data ( although they are still significantly different , p < 0 . 05 ) . Additionally , we performed the same experiment with AC females and saw similar trends in response time curve shape and significance values ( Figure 5C ) . These results , therefore , suggest that auditory cues , when uncoupled with the male behavior , have little influence on female response timing ( vs their influence on female song rate [Figure 4] ) . Any relationship between female song and the acoustic playback must be due to indirect effects on the WC male . We , therefore , hypothesized that another cue provided by the WC male influenced female song timing . To test the role of contact cues provided by the male , we next examined pairings between intact males and headless females . 10 . 7554/eLife . 07277 . 017Figure 5 . Testing the sufficiency of auditory cues for female song production and timing . ( A ) Female pulses ( red ) produced in response to playback of male song ( gray ) and in the presence of a WC male . Normalized WT ( B , red , n = 451 , n = 13 courtships ) and AC ( C , red , n = 522 , n = 16 courtships ) female response time distributions to playback compared with response time distributions from shuffled data ( red dashed ) . Inset: the same data , but female response times are calculated from the end of the male bout , therefore , ignoring overlaps ( K-S test ***p < 0 . 001 , *p < 0 . 05 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 07277 . 017 Previous studies showed that headless D . virilis females reject male courtship attempts ( Spieth , 1966 ) , so we were surprised to observe that these females continued to produce song ( Figure 6A ) . Furthermore , we observed song production from headless females only in the presence of a male and while he was courting ( Video 2 ) . Headless females produced pulses with longer IPI but similar frequency spectra , relative to intact females ( compare Figure 6B , C with Figure 1E , F ) . In addition , headless females produced as much song as intact females; song production rates were significantly reduced only when we additionally blocked the genitalia of headless females ( Figure 6D ) . Importantly , headless females still coordinated acoustic signal production with their male partner ( Figure 6E ) , although the peak of the response time distribution was broader ( more variable ) and shifted to the right ( longer latency ) relative to the intact female distribution ( compare with Figure 2A ) . 10 . 7554/eLife . 07277 . 018Figure 6 . Testing the sufficiency of abdominal sensory inputs for female song production and timing . ( A ) Headless female song ( orange ) and WT male song ( blue ) . ( B ) Median IPI of headless female song ( orange , n = 33 ) compared with IPI from WT females ( red , n = 83 ) . Black squares indicate population mean and black bars s . d . ( Student's t-test , ***p < 0 . 001 ) . ( C ) Power spectral density of pulses from headless females ( orange , n = 21 , 481 pulses ) or their WT male courtship partner ( blue , n = 11 , 136 pulses ) . ( D ) Median pulse rate for headless females ( orange , n = 33 ) and headless and genitalia blocked females ( dark orange , n = 31 ) , compared to WT females ( red , n = 83 ) . Black squares indicate population median and black bars IQR ( Kruskall–Wallis test , ***p < 0 . 001 ) . ( E ) Normalized female response time distribution for headless females ( orange , n = 1430 , n = 33 courtships ) , compared with distributions from shuffled male bout ( blue dashed ) and female pulse ( orange dashed ) intervals ( ***p < 0 . 001 K-S test ) . Shading around each response time distribution represents bootstrapped 95% confidence intervals . DOI: http://dx . doi . org/10 . 7554/eLife . 07277 . 01810 . 7554/eLife . 07277 . 019Video 2 . D . virilis headless female duetting behavior . Representative example of a naive male paired with a naive decapitated female in one channel/chamber of the multi-channel song recording system . Video is acquired at 15 Hz . DOI: http://dx . doi . org/10 . 7554/eLife . 07277 . 019 From these data , we conclude that the neural circuitry that controls female duetting behavior lies largely within the ventral nerve cord and does not require descending activation signals from the brain . However , while male contact with the female genitalia is sufficient to drive the female song circuit , inputs from the brain contribute to response timing , IPI , and the decision to mate ( in our experiments , headless females did not perform the characteristic wing-spreading behavior indicative of female acceptance and never copulated with WT males [data not shown] ) . While song production in decapitated D . melanogaster males has been observed ( Clyne and Miesenbock , 2008 ) , it was elicited only upon optogenetic neural activation of song circuits . The song produced by headless D . virilis females , in contrast , does not require experimental stimulation , but simply the natural sensory cues provided by the male partner . In contrast with D . melanogaster , D . virilis males continually contact the female during courtship ( Vedenina et al . , 2013 ) ; however , the temporal pattern of male contact with the female has not been characterized . From higher speed videos taken of the ventral side of the courting flies ( Video 3 ) , we observed that males rub their tarsi on ventral tergites A3–A5 of the female abdomen while also licking ( with their extended proboscis ) the female genitalia ( Figure 7A ) . We , therefore , quantified the temporal dynamics of these motor programs ( see ‘Materials and methods’ and Figure 7B ) to determine if either behavior predicted male and female song timing . We again used GLMs to predict female pulse and male bout times from male behaviors . More specifically , we predicted the presence/absence of female or male song based on the temporal pattern of behaviors preceding each time point during courtship interactions . The resulting set of coefficients for each behavior combine to produce temporal filters for predicting song ( see ‘Materials and methods’ ) . Since these behaviors are correlated in time and with each other , we used a sparse prior on the GLM coefficients that penalize non-predictive filter components ( Coen et al . , 2014 ) . 10 . 7554/eLife . 07277 . 020Video 3 . High-speed video of D . virilis duetting behavior . Representative example of courtship between a naive D . virilis male and female pair in a 1 × 1 × 0 . 5 cm clear plastic chamber , with microphone placed adjacent to the chamber . Video is acquired at 60 Hz . DOI: http://dx . doi . org/10 . 7554/eLife . 07277 . 02010 . 7554/eLife . 07277 . 021Figure 7 . Male contact with the female abdomen and genitalia predicts song timing . ( A ) Schematic of annotated male contact behaviors ( see also Video 3 ) . ( B ) Example of a duet between a WT male ( blue ) and female ( red ) , accompanied by annotation of male tarsal contact with the abdomen ( tarsal vibration ( gray ) and tarsal pause ( black ) ; tarsal pause denotes male tarsal contact with the female abdomen , but without vibration ) and proboscis licking of the genitalia ( purple ) . GLM analysis was used to predict the presence/absence of male or female song based on the time course of annotated behaviors during courtship interactions . Model performance ( indicated by relative deviance reduction ) reveals the behaviors most predictive of female song ( C ) and male song ( E ) ( see ‘Materials and methods’ ) . GLM filters reveal the times at which each behavior is most predictive of female song ( D ) or male song ( F , n = 7 courtships with approximately 200 instances of both male and female song ) . Error bars and shading represent bootstrapped s . e . m . estimates . ( G ) Females with their upper abdomens ( tergites A3–A5 ) blocked produce very little song ( n = 23 courtships ) , compared with females whose aristae were additionally cut ( n = 29 courtships , ***p < 0 . 001 Wilcoxon rank-sum test ) . Black squares represent median and black bars the IQR . ( H ) Normalized female response time distributions for AC females ( red , reproduced from Figure 1H ) , AC-upper abdomen blocked females ( dark red , n = 1892 , n = 29 courtships , Kullback–Leibler ( K-L ) Divergence = 0 . 12 between AC and AC-upper abdomen block curves ) and AC-genitalia blocked females ( black , n = 483 , n = 18 courtships , K-L Divergence = 0 . 36 between AC and AC genitalia block curves ) ; all females were paired with WT males . Shading around each response time distribution represents bootstrapped 95% confidence intervals . DOI: http://dx . doi . org/10 . 7554/eLife . 07277 . 02110 . 7554/eLife . 07277 . 022Figure 7—figure supplement 1 . Two-variable GLM for male–female pairings . Relative deviance reduction and linear filter values for prediction of female song ( A and B ) or male song ( C and D ) . Error bars and shaded regions around linear filters represent bootstrapped s . e . m . estimates . Relative deviance reduction of best single variable model is indicated by dashed horizontal line ( associated with Figure 7C , E , respectively ) . DOI: http://dx . doi . org/10 . 7554/eLife . 07277 . 022 Looking at individual male features predicting female song , we found that tarsal vibrations and proboscis licking were most predictive of the occurrence of female song ( Figure 7C ) . The associated filters had their peaks within 16 milliseconds of female song , indicating low-latency coupling of male cues with female song ( Figure 7D ) . Predictability was not substantially enhanced with a two-variable model ( Figure 7—figure supplement 1 ) . Notably , male song was much less predictive of female song , further suggesting that acoustic cues play a subordinate role in female song timing . Interestingly , we found that the male behaviors predicting female song also predict male song . That is , male tarsal vibration and proboscis licking were most predictive of the presence/absence of male song ( Figure 7E ) . However , the associated filters peaked 200 and 50 ms preceding male song , respectively ( Figure 7F ) . This implies that male courtship behavior forms a behavioral sequence that starts with tarsal vibration , transitions into proboscis licking , and ends with male song . These results suggest that both male tarsal contact with the female abdomen and male proboscis contact with the female genitalia are independently predictive of female response timing . Above , we showed that blocking just the female genitalia reduces female song production rates ( Figures 4E , 6D ) . Given the GLM results , we next blocked segments A3–A5 of the female abdomen , while leaving the genitalia unblocked , and discovered that this manipulation also significantly reduced female song production ( Figure 7G ) . However , additionally removing the aristae restored song production rates to WT levels ( compare with Figure 3—figure supplement 2 ) , indicating that auditory and abdominal sensory inputs are combined to drive song production during courtship . We then compared female response time distributions for AC females with block to either abdominal segments A3–A5 or the genitalia ( Figure 7H ) . We found a larger shift in the response time distribution when blocking the genitalia ( KL divergence between AC distribution and AC genitalia block distribution = 0 . 36 ) vs when blocking the upper abdomen ( KL divergence between AC distribution and AC upper abdomen block distribution = 0 . 12 ) . We , therefore , propose that while both forms of male contact are required for normal song rates , female song timing relies more strongly on the timing of male proboscis contact with her genitalia . This represents a novel mechanism for coordinating duets . We hypothesize that male contact cues ( both tarsal vibration and proboscis licking ) are detected via mechanosensory neurons , as the female abdomen and genitalia are covered in mechanosensory machrochete bristles ( Fabre et al . , 2008 ) . In contrast , gustatory receptors are expressed in multidendritic neurons that tile the abdominal body wall ( Park and Kwon , 2011 ) , but none of these neurons detect external chemical cues . Our videos , however , did not reveal which subsets of bristles are likely to be responsible for detecting the male contact cues . In D . melanogaster , song production circuits are sexually differentiated or dimorphic ( von Philipsborn et al . , 2011 ) ; only males of this species produce song . Females are typically silent but can be forced to produce ( aberrant ) song via artificial activation of fruitless-expressing neurons ( Clyne and Miesenbock , 2008 ) . This sexual dimorphism relies on male-specific isoforms of both the fruitless and doublesex genes ( Demir and Dickson , 2005; Manoli et al . , 2005; Rideout et al . , 2010 ) , the regulation of which are conserved in virilis ( Yamamoto et al . , 2004; Usui-Aoki et al . , 2005 ) . We , therefore , expected that the disparate male and female song behaviors in D . virilis should also be sexually dimorphic . To test this hypothesis , we paired either two females or two males in behavioral chambers . We did not observe any song production in pairings between two females ( data not shown ) . However , similar to studies in D . melanogaster ( Villella et al . , 1997; Pan and Baker , 2014 ) , we found that males will court another male . We observed two types of song in these interactions ( Figure 8A ) : the courting male produced male-typical songs ( blue ) , while the male being courted produced a secondary song that appeared more female-like ( green ) , with longer IPIs ( Figure 8B , compare to Figure 1E ) . However , the fundamental frequency of the pulses of the courted male's song remained male-like ( Figure 8C ) . This secondary song ( Suvanto et al . , 1994 ) , similar to female song , was generated by bilateral wing vibration ( Video 4 ) . Strikingly , we found response times of the courted male ( secondary ) song relative to the courting male ( primary ) song to match the distribution of female response times ( compare Figure 8D with Figure 2A ) . This implies that the male nervous system contains separable circuits for song production , which are each activated in a context-dependent fashion . When the male is courting a target , he produces ‘male’ or primary song bouts , but when he is being courted , he produces ‘female’ or secondary song at the appropriate ( female-like ) response delay ( see Video 5 e . g . , of males alternating between courting and being courted ) . 10 . 7554/eLife . 07277 . 023Figure 8 . When courted by another male , males can produce female-like duets . ( A ) Song produced in pairings between two WT males . Combining video and acoustic recording ( Video 4 ) reveals that male primary song is produced by the courting male ( blue ) , and male secondary song is produced by the courted male ( green ) . ( B ) IPI of male secondary song ( green , n = 8 ) compared with the IPI of male primary song ( blue , n = 13 , Wilcoxon rank-sum Test , ***p < 0 . 001 ) . ( C ) Power spectral density of pulses from male secondary song ( n = 6349 , green ) and male primary song ( n = 1309 bouts , dark blue ) . ( D ) Normalized distribution of courted male response times to male primary song ( green , n = 766 , n = 53 courtships ) compared to response times from shuffled data ( K-S Test , ***p < 0 . 001 ) . ( E ) Example of a male–male duet , accompanied with video annotation of tarsal and proboscis movements of the courting male . GLMs were used to predict the presence/absence of male secondary song and male primary song from the temporal pattern of annotated behaviors preceding each time point during courtship interactions . GLM performance indicates the male behaviors most predictive about male secondary song ( F ) and male primary song ( H ) . GLM filters reveal the times at which each behavior is most predictive of male secondary song ( G ) or male primary song ( I ) . ( n = 7 courtships with 107 and 127 instances of male secondary and primary song , respectively ) . Error bars and shading in F-I represent bootstrap estimates of the s . e . m . DOI: http://dx . doi . org/10 . 7554/eLife . 07277 . 02310 . 7554/eLife . 07277 . 024Video 4 . High-speed video of male–male behavior . Representative example of an interaction between two naive D . virilis males in a 1 × 1 × 0 . 5 cm clear plastic chamber , with microphone placed adjacent to the chamber . Video is acquired at 60 Hz . DOI: http://dx . doi . org/10 . 7554/eLife . 07277 . 02410 . 7554/eLife . 07277 . 025Video 5 . High-speed video of male–male behavior . Another example of an interaction between two naive D . virilis males , showing role-switching behavior between courting and being courted . DOI: http://dx . doi . org/10 . 7554/eLife . 07277 . 025 To determine how the response timing of male secondary song is regulated , we again combined higher speed video and acoustic recordings . Similar to male–female courtship , tarsal vibration and proboscis licking are most predictive of the occurrence of secondary song ( Figure 8F ) , with the GLM filters peaking immediately prior to secondary song ( Figure 8G ) . Likewise , these behaviors are also predictive of male primary song ( Figure 8H ) , and the shapes of filters were similar for male–male vs male–female interactions ( compare Figure 8I with Figure 7F ) . Predictability was reduced when compared with models for predicting song in male–female interactions ( e . g . , compare Figure 7C with Figure 8F ) and was not substantially enhanced with a two-variable model ( data not shown ) . We conclude that male secondary song timing is best correlated with , similar to female song timing , both male tarsal and proboscis contact with the abdomen and genitalia . Therefore , we propose that sexually monomorphic ( or undifferentiated ) sensory bristles and neural circuits drive female duetting behavior in D . virilis .
In this study , we combine targeted sensory manipulations , high-throughput song recording and analysis , and statistical modeling to provide the first quantitative characterization of acoustic duetting in a Drosophilid species . This is significant because , as a Drosophilid , virilis shares developmental similarities with melanogaster ( Kuntz and Eisen , 2014 ) and in addition has had its genome fully sequenced and annotated ( Drosophila 12 Genomes Consortium et al . , 2007 ) . This makes feasible the development of genetic ( Bassett and Liu , 2014 ) and neural circuit tools ( Simpson , 2009 ) to resolve the mechanisms underlying duetting . Such tools will allow us to determine , for example , if genes such as fruitless and doublesex , known to be important for the establishment of sexually dimorphic and courtship-related behaviors in melanogaster ( Kimura et al . , 2008 ) , play different roles in a species in which both males and females are capable of song production . Our study also uncovered the sensory cues and putative neural mechanisms that orchestrate duetting in virilis . We can address these mechanisms by , for example , targeting neurons ( via genetic methods ) in virilis that are homologous to recently mapped song pathway neurons in melanogaster ( von Philipsborn et al . , 2011 ) . These experiments should reveal for the first time how the function and modulation of the song motor pathway differs between males and females in a duetting species . Because virilis is separated from melanogaster by >40 million years of evolutionary time ( Drosophila 12 Genomes Consortium et al . , 2007 ) , the development of these tools should additionally provide insight into the evolution of new behaviors such as female song production . We also establish with this study that duetting in D . virilis relies on multisensory cues . Our experiments using song playback ( to test the sufficiency of auditory cues ) and headless females ( to test the sufficiency of tactile cues ) showed that neither auditory nor tactile cues alone fully recapitulated female song responses ( e . g . , response time distributions did not wholly match those of intact females interacting with males ) . In addition , we showed that interfering with song production ( e . g . , blocking male tactile inputs to the female abdomen and genitalia ) could be overcome by additional removal of the auditory receiver ( arista ) . These results taken together suggest that auditory and tactile information are integrated to drive song production and coordination in females . This may be similar to the integration of multimodal signals observed in frogs during territorial behaviors ( Narins et al . , 2005 ) . Experiments that target the virilis song pathway for neural recordings will reveal if this pathway is directly sensitive to both auditory and tactile information ( both types of stimuli can be easily applied in a preparation fixed for in vivo recordings [Murthy and Turner , 2010] ) . These experiments should also reveal if auditory information is combined linearly ( simple integration ) vs nonlinearly with tactile cues . Other studies in insects have found a role for tactile cues in some social ( but non-courtship ) behaviors ( Rogers et al . , 2003; Ramdya et al . , 2014 ) ; however , ours is the first demonstration of a role for tactile cues in acoustic duetting in any animal species . Moreover , our GLM analysis revealed that it is the precise timing of the male's contact with the female abdomen and genitalia that is the biggest predictor of her song timing . These experiments and analysis not only expose a completely novel mechanism for coordinating acoustic behavior between individuals but raises the question of what advantage incorporating tactile information provides to the female . Because removing any individual sensory cue to the virilis female ( e . g . , tactile cues sensed either via the genitalia or the upper abdomen ) disrupts female song production , it seems unlikely that a male provides multiple cues in the event that one of his signals fails—rather , females may select for males that can reliably provide multiple timing cues . This idea is consistent with studies showing that generating multiple cues is energetically costly for the sender ( Partan and Marler , 2005 ) . Moreover , studies of decision-making in higher systems show that different sensory channels provide independent sources of information for directing behavior ( Fetsch et al . , 2012 ) . If this is true also for virilis , females may be better able to perform the duet by relying on multiple channels . The only sensory manipulation performed in this study that disrupted duetting while leaving song production rates intact was deafening the male ( Figure 2 ) . However , this manipulation did not affect time to copulation , which suggests that neither do males need to hear the female's song nor do females require a properly timed duet for mating . On the other hand , our experiments on virgin vs mated females suggest that female song production is a positive cue , because it is associated with her receptivity state . It is possible that our experimental conditions ( pairing one male with one female ) did not uncover the role duetting normally plays in mating . Studies in other model systems suggest the female song production evolved in species that experience high levels of reproductive competition ( Clutton-Brock , 2007 ) . Examining interactions between either two males and one female or two females and one male , for example , could reveal the relationship between duetting and mate selection in virilis; such experiments are feasible using Drosophila behavioral chambers with multiple microphones to separate the signals of multiple singers ( Coen et al . , 2014 ) . Finally , because humans can integrate both auditory and tactile cues during speech perception ( Gick and Derrick , 2009 ) , we propose that , more broadly , tactile cues may be critical for acoustic communication across the animal kingdom . Mapping the virilis circuits that detect tactile information , and then relay this information to the song production pathway , should therefore provide general insights into how precisely timed behaviors , such as song and speech production , are regulated . Finally , our data suggest that neural circuits for female duetting ( from circuits that detect tactile cues all the way to circuits that direct the production of song pulses ) are not sexually dimorphic or differentiated between males and females; this result sharply contrasts with the observed sexual dimorphism of song production circuits in D . melanogaster ( Clyne and Miesenbock , 2008; von Philipsborn et al . , 2011 ) . However , we did not find conditions under which females produce male-like songs , suggesting that virilis male primary song production circuits are sexually differentiated . This discovery should aid in identifying the circuits for duetting in virilis ( in other words , some song circuit elements should be present in both males and females , while others should only be present in males ) . Moreover , during male–male courtship , virilis males can switch between the production of female-like secondary or male-like primary song based on their role as courtee or courter , respectively ( Video 5 ) . While some insects participate in sexual mimicry ( imitating the song of the partner ) during male–female courtship ( Luo and Wei , 2015 ) , ours is , to the best of our knowledge , the first demonstration of context-dependence of duetting behavior in any animal species . Therefore , dissecting the neural circuit basis for song production in D . virilis promises to reveal novel insights into how animals rapidly modulate behavior in response to changing sensory feedback .
We tested D . virilis strains 15010-1051 . 47 , 51 , 09 , 48 , 49 , and 52 ( UCSD Stock Center , San Diego , CA ) in our song-recording chambers . We chose strain 15010-1051 . 47 , derived from Hangchow , China , for all experiments in this study , because both males and females of this strain produced the most song during courtship . Flies were maintained on standard medium at 20°C , on a 16-hr light:8-hr dark cycle ( Suvanto et al . , 1999 ) . For behavioral assays , virgin males were housed individually , while virgin females were group housed . All flies were aged 10–20 days; this is the time required to reach sexual maturity ( Isoherranen et al . , 1999 ) . Behavioral assays were performed between 0–4 hr ZT and at approximately 22°C . Aristae and antennae were removed by plucking with tweezers ( Fine Science Tools , Foster City , CA ) . Distal fore-tarsi were removed by cutting with scissors ( Fine Science Tools , Foster City , CA ) . To block contact with the female genitalia , we covered the genitalia and a portion of the A6 tergite with a low-melting temperature paraffin wax . We confirmed that results were similar whether we used wax or a UV-curable glue . To block contact with the female upper abdomen , we covered only tergites A3–A5 . Visual cues were removed by turning off the lights in a windowless room and covering the recording system in blackout curtains . In this condition , red LED lighting was used for observations . We performed all physical manipulations on CO2-anesthetized flies a minimum of 12 hr prior to behavioral assays . The only exception was for decapitation , which was performed on females at least 30 min prior to recording . Courtship songs were recorded at 10 kHz on a 32-channel apparatus ( Arthur et al . , 2013 ) . Mature flies were mouth aspirated into plastic chambers modified from the design in ( Arthur et al . , 2013 ) . Modified chambers were larger ( 2-cm diameter ) and raised , to prevent direct contact with the microphones . These changes were made to accommodate virilis flies , which are larger and sing louder than melanogaster flies . To encourage two males to interact with each other , a small piece of paper was added to the chamber to reduce the available space . Acoustic behaviors were recorded for 20 min , and it was noted if copulation occurred during this period . We wrote annotation software ( FlySongSegmenter-Virilis [FSS-V] ) that performs a continuous wavelet transformation ( using the frequency B-spline wavelet ) on the raw song trace to extract the spectral features ( between 100 and 900 Hz ) of the song as a function of time . See Figure 1—figure supplement 3 . For each time point in the wavelet transform , we computed the likelihood that it was generated from one of four previously assembled template distributions . These template distributions were created from manually annotated data for male , female , and overlapping pulses . The wavelet transforms corresponding to the pulse locations for each of these groups were then collated and rotated into a new orthogonal coordinate frame via principal components analysis . The distributions of projections along the modes were fit to a Gaussian mixture model through an expectation maximization algorithm . Thus , each template consists of a mean wavelet spectrum , an orthogonal basis set , and a set of distributions along each of these newly created axes . An identical procedure was performed to create a noise ( N ) template as well . However , for this case , we created a new noise template from the low-amplitude fluctuations in each data set separately , allowing us to account for differing noise conditions . Given these templates , we defined the likelihood that a wavelet was drawn from one of these templates , T , by transforming it into the appropriate template coordinate system and calculating p ( xi|T ) for each coordinate value , xi . The log-likelihood is simply the sum of the logarithms of these values . The end result of this process was to have a set of 4 log-likelihoods as a function of time ( male [M , blue] , female [F , red] , overlap [O , green] , and noise [N , black] ) . If we assume a uniform prior for the template distributions , then , p ( T|x ( t ) ) is simply proportional to the likelihood value , so we can then assign a probability to each of the templates as a function of time . We first eliminated all noise from the time series , zeroing out all segments where p ( N|x ( t ) ) > 1/2 . We then looked at each connected segment of non-noise data separately . If p ( M|x ( t ) ) + P ( O|x ( t ) ) > p ( F|x ( t ) ) and the segment were at least 100 ms in length , we assigned that segment to M . All remaining segments at least 15 ms in length were assigned to F . Lastly , we looked within each of the M regions for overlapping pulses , where both the male and the female were singing . If p ( F|x ( t ) ) + P ( O|x ( t ) ) > 0 . 8 over a region at least 15 ms in length , we labeled that as a male region with a female pulse . The automated code is available on github ( https://github . com/murthylab/virilisSongSegmenter ) . Lastly , these pulse calls were scanned by visual inspection , correcting mislabeled pulses by hand . Song segmented by this method was compared to manually segmented song from a WT data set that was not previously used to inform the likelihood models . Sensitivity , positive predictive value , and the harmonic mean ( F ) were calculated as previously described ( Arthur et al . , 2013 ) . To correlate song production with a particular singer ( e . g . , male vs female ) , we used a USB2 . 0 CMOS 1280 × 1024 Monochrome Camera ( Thor Labs , New Jersey , USA ) with 5-mm EFL lens , and uc480Viewer recording software was used to record single-courting pairs at ∼15 Hz . Audio and video data were synced using iMovie . To record fly behavior ( in particular , male contact with the female abdomen ) during duetting , we used a Point Grey Monochrome Camera with a 5-mm EFL lens to record single-courting pairs at 60 Hz in a 1 cm × 1 cm × 0 . 5 cm clear plastic chamber . Audio and video data were synced using iMovie . These videos were annotated for onset and offset of three behavioral features ( tarsal contact with vibration , tarsal contact without vibration ( pausing ) , and proboscis contact ) using ANVIL ( www . anvil-software . org ) ; we annotated only video segments during which both males and females produced song . Male and female songs were annotated in MatLab ( Mathworks Inc ) . Synthetic male courtship song ( a recorded male bout was smoothed using a 300-Hz Butterworth low-pass filter; a single stimulus contained this bout repeated at 6× at 1 . 2-s intervals; stimuli were delivered every 30 s for 10 min ) was delivered to females , paired with WC males , in a modified courtship chamber ( a 2-cm diameter hole was cut into the top of the plastic chamber and replaced with mesh ) . Song was delivered via Koss earbud speakers , and earbuds were mounted above each chamber and oriented at a 45° angle toward the arena . Song intensity was calibrated to match song recorded in the same chambers ( between an intact male and female pair ) . We were unable to score most female song that occurred at the same time as the playback stimulus ( due to differences in intensity between the playback and the female song ) . Therefore , we also calculated female response time from the offset of the male artificial bout , which ignores potential overlaps ( see inset in Figure 5B , C ) . This is the only instance in all analyses in this study where overlaps were ignored . Male/female pairs were aspirated into clear plastic chambers , each 1 × 1 × 0 . 5 cm . Pairs were manually observed for 30 min , and copulations were recorded every minute . Manipulated flies ( e . g . , AC ) and their corresponding controls ( e . g . , flies with intact aristae , but held on the CO2 pad for the same amount of time ) were observed simultaneously . Experiments that required observation in the dark were performed in a dark room under red LED lighting . Only pairs for which there was visible male wing vibration were scored . To generate mated females , females were paired with males until copulation occurred , at least 24 hr prior to pairing with virgin males . All statistical analysis was performed in Matlab ( Mathworks , Inc . ) . A male bout must contain at least four concurrent male pulses , with IPIs of less than 25 ms . Female bouts consist of successive pulses with IPIs <100 ms . IPI values were calculated as the time between pulses with a threshold of 100 ms for males and 500 ms for females and reported as a median per individual . Male bout and female pulse rates were calculated as the number of bouts or pulses divided by total courtship time in seconds ( Hz ) , where total courtship time is the time between the first pulse in the recording ( male or female ) and the last pulse ( male or female ) . Because there can be long stretches of silence during a recording , song rates report how much singing occurs overall within a recording , whereas IPI reports the rate of pulsing when singing occurs . We chose to quantify female pulse rates ( as opposed to bout rates ) due to the highly variable structure of female bouts . Recordings with less than 20 s of either male or female song were assigned a male bout rate or female pulse rate of 0 Hz . The square roots of female pulse rates are plotted in all figures ( to temper outliers ) , but all statistics were performed on raw data . A GLM ( for details see next section ) was used to determine the significance of song rate differences between virgin and mated females , independent of courtship success . Instances of overlaps between male and female song are reported as a percent ( number of overlaps/number of male bouts ) for each courtship pairing . For response times , we calculated the delay from the onset of a male bout to the previous female pulse within 1 . 5 s ( male response time ) or the delay from the onset of a male bout to the following female pulse within 1 . 5 s ( female response time ) . This analysis includes regions of male and female overlap . We also randomly shuffled either female IPIs or male IBIs and then calculated male and female response times . Confidence intervals were generated with 500 bootstrapped permutations . The two-sample Kolmogorov–Smirnov test was used to determine if the difference between response time curves was statistically significant . We report median values and interquartile range ( IQR ) for non-normally distributed data and mean values and standard deviation for normally distributed data . Male secondary song elicited during male–male interactions was analyzed in the same manner as female song above . To determine the sensory pathways influencing song production rate in males and females , we fit a GLM to the behavioral data . Since pulse/bout rates follow Poisson statistics ( rates are bounded between 0 and infinity ) , we used an exponential link function in the model . The presence/absence of a sensory channel was coded as 1/−1 . To ensure that the GLM coefficients reflect the relative impact of removing a sensory channel , we first z-scored the data . Fitting was performed using Matlab's fitglm function , and statistics and errorbars were extracted from the output of that function . The same approach was used to demonstrate differences in song rates dependent on the female mating state . For each pair , the state of the female ( virgin/mated ) and the success of the courtship ( no copulation/copulation ) were coded in a two-variable matrix ( represented as −1/1 , respectively ) . Statistical output of the fitglm function ( from MatLab ) is reported . To determine the behaviors that control the timing of song , we fit GLMs , as in Coen et al . ( 2014 ) , to predict the presence/absence of male song or female song from annotated behaviors during a courtship interaction . To learn about when a behavior is predictive about song , we used the temporal pattern of behaviors for prediction . For each time point in male or female song , we used the temporal pattern of behaviors in a window of ∼1 s ( 64 frames at 60 fps ) preceding that time point for predicting song . This yielded GLM filters—a sequence of weights whose magnitude indicates the importance of each time point in the behavioral history . Since the predicted variable is binary ( song/no song ) , we used a logistic link function for the GLM . The annotated behaviors are strongly correlated both in time ( autocorrelation ) as well as with each other ( cross-correlation ) . We , therefore , used a GLM with a sparse prior that penalized non-predictive weights , which would have large magnitudes merely due to correlations in the data ( Mineault et al . , 2009; Coen et al . , 2014 ) . Model performance ( relative deviance reduction ) was evaluated using cross-validation; that is , model parameters were fitted from 80% of the data , and the model was tested with the remaining 20% of the data ( random subsampling ) . To obtain a bootstrap estimate of the standard error of the mean for the GLM filters and GLM performance , we re-ran this cross-validation procedure 1000 times with a randomly selected 75% of all data . We started by fitting single-variable models ( e . g . , using only tarsal contact as a predictor ) and determined the most predictive features . To rule out that including more parameters substantially improved performance , we took the best predictor and added any of the remaining three features to the model . However , this never increased performance by more than 14% . Code for sparse GLM analysis is available on GitHub ( https://github . com/murthylab/GLMvirilis ) . | When performing a duet , human singers listen to each other and also use other cues like tapping and nodding to keep in time . Duets are also found in the courtship rituals of some animals , but unlike human duetters—who tend to sing at the same time—the male and female animals take turns singing back and forth in quick succession . For most species , the male starts the duet and the female responds . It is thought that animals rely on their sense of hearing to perform their duets successfully , but it is not clear whether they also rely on other cues , or what these might be . Fruit flies produce songs during courtship displays by vibrating their wings . Typically , only the male fly in a pair will sing , while the female is silent and makes the mating decisions . However , both the male and female sing in the courtship rituals of a fruit fly species called Drosophila virilis . Now , LaRue et al . have studied the songs and other behaviors produced by this fruit fly during courtship displays . The experiments show that D . virilis males and females perform duets; in other words , there is a predictable timing between male and female songs . For the female fly to produce the right amount of song , she needs to detect sound cues from the male fly and also be in physical contact with him . This contact involves the male tapping the female's abdomen and licking her genitalia , and the precise timing of these behaviors predicts when she sings . This behavior was also observed in duets produced by male-only pairs , so it is not gender-specific . These findings show that D . virilis flies use both sound and other types of cues to coordinate singing duets during courtship . LaRue et al . suggest that female flies may choose to mate with males that provide multiple timing cues during the duet . A future challenge is to understand how information provided by the different cues is combined in the brain of the female to drive this courtship behavior . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"neuroscience"
] | 2015 | Acoustic duetting in Drosophila virilis relies on the integration of auditory and tactile signals |
Postsynaptic compartments can be specifically modulated during various forms of synaptic plasticity , but it is unclear whether this precision is shared at presynaptic terminals . Presynaptic homeostatic plasticity ( PHP ) stabilizes neurotransmission at the Drosophila neuromuscular junction , where a retrograde enhancement of presynaptic neurotransmitter release compensates for diminished postsynaptic receptor functionality . To test the specificity of PHP induction and expression , we have developed a genetic manipulation to reduce postsynaptic receptor expression at one of the two muscles innervated by a single motor neuron . We find that PHP can be induced and expressed at a subset of synapses , over both acute and chronic time scales , without influencing transmission at adjacent release sites . Further , homeostatic modulations to CaMKII , vesicle pools , and functional release sites are compartmentalized and do not spread to neighboring pre- or post-synaptic structures . Thus , both PHP induction and expression mechanisms are locally transmitted and restricted to specific synaptic compartments .
Synaptic strength can be modulated with a remarkable degree of specificity to enable the flexibility necessary for learning and memory , where compartmentalized changes in dendritic spines tune responses to neurotransmitter release during information transfer in the nervous system . Such plasticity mechanisms require compartmentalized trafficking and insertion of glutamate receptors ( GluRs ) into postsynaptic densities at specific locations in response to correlated activity ( Herring and Nicoll , 2016; Malinow , 2003 ) . However , these processes of Hebbian plasticity are inherently destabilizing , and homeostatic mechanisms have been proposed to adaptively counteract such forces to maintain synaptic strength within physiologically stable levels ( Davis , 2013; Pozo and Goda , 2010; Turrigiano , 2012 ) . Although the induction and expression of various forms of plasticity can clearly be restricted to individual postsynaptic compartments , it is less certain that such plasticity can be similarly compartmentalized at the presynaptic terminals of a single neuron . There is evidence for heterogeneity in presynaptic efficacy ( Branco et al . , 2008; Dobrunz and Stevens , 1997; Holderith et al . , 2012; Trommershäuser et al . , 2003; Vitureira et al . , 2011 ) , and release probability can vary considerably along a single axon ( Murthy et al . , 1997; Paul et al . , 2015; Peled and Isacoff , 2011 ) . Further , target-specific differences in presynaptic function ( Frank , 1973; Katz et al . , 1993; Scanziani et al . , 1998 ) and homeostatic plasticity ( Davis and Goodman , 1998 ) have been demonstrated . However , how presynaptic terminals are modulated by Hebbian and homeostatic forces and whether these adaptations can occur without ‘spreading’ to adjacent synapses remains enigmatic . The Drosophila neuromuscular junction ( NMJ ) is a powerful model system to interrogate the mechanisms governing homeostatic synaptic plasticity . At this glutamatergic synapse , genetic and pharmacological perturbations to postsynaptic receptors initiates a retrograde , trans-synaptic signaling system that homeostatically increases presynaptic neurotransmitter release to maintain stable levels of synaptic strength ( Davis and Müller , 2015; Frank , 2014 ) . This form of plasticity is achieved through an increase in presynaptic efficacy and is therefore referred to as presynaptic homeostatic potentiation ( PHP ) . Parallel forms of homeostatic regulation are conserved at NMJs of rodents ( Wang et al . , 2016b ) and humans ( Cull-Candy et al . , 1980 ) . PHP initiates a single retrograde signaling system that triggers two key expression mechanisms to enhance presynaptic glutamate release: increases in both presynaptic calcium influx and the number of synaptic vesicles participating in the readily releasable pool ( Goel et al . , 2017; Kiragasi et al . , 2017; Müller and Davis , 2012; Weyhersmüller et al . , 2011 ) . Several genes necessary for PHP expression have been identified that function as putative retrograde signals ( Orr et al . , 2017; Wang et al . , 2014 ) and presynaptic effectors in the motor neuron ( Bruckner et al . , 2017; Dickman and Davis , 2009; Dickman et al . , 2012; Kiragasi et al . , 2017; Müller et al . , 2015; Müller et al . , 2012; Tsurudome et al . , 2010; Younger et al . , 2013 ) , but the postsynaptic induction mechanisms that initiate PHP signaling remain enigmatic ( Chen and Dickman , 2017; Goel et al . , 2017 ) . Indeed , PHP induction and expression at a subset of synapses within a single motor neuron has never been demonstrated at the Drosophila NMJ . PHP is expressed exclusively at one of two motor neuron subtypes that innervate most muscles at the Drosophila NMJ , Type Is and Type Ib . Type Is motor inputs exhibit smaller boutons , less subsynaptic reticulum ( SSR ) , higher basal release probability , and do not participate in PHP adaptation over chronic time scales ( Lnenicka and Keshishian , 2000; Newman et al . , 2017 ) . In contrast , Type Ib motor neurons have larger boutons , more elaborate SSR , and lower basal release probability , which is enhanced after loss of postsynaptic GluRs ( Newman et al . , 2017 ) , demonstrating that PHP is expressed exclusively at Type Ib synapses . Further , a reduction in phosphorylated ( active ) levels of CaMKII , presumably related to PHP inductive signaling , occurs specifically in the postsynaptic density of Ib boutons ( Goel et al . , 2017; Newman et al . , 2017 ) , suggesting a possible mechanism for the specificity of retrograde PHP signaling to the Ib motor neuron . Despite these insights , it is not known whether PHP can be expressed at a subset of Type Ib boutons within a single motor terminal , nor whether PHP modulations at individual boutons influence neighboring synapses . We have developed a genetic manipulation that enables the reduction of postsynaptic GluR expression on one of the two muscles innervated by a single Type Ib motor neuron at the Drosophila NMJ . We have used this system to test whether PHP signaling is synapse specific and to determine to what extent the postsynaptic induction and presynaptic expression of PHP is compartmentalized . This analysis has revealed highly specific and compartmentalized PHP adaptations that are restricted and target specific without influencing neurotransmission at neighboring synapses within the same motor neuron .
The postsynaptic response to glutamate release at the Drosophila NMJ is mediated by two types of GluRs . Both types contain the essential subunits GluRIIC , GluRIID , and GluRIIE , but differ in containing either GluRIIA or GluRIIB subunits ( DiAntonio et al . , 1999; Featherstone et al . , 2005; Han et al . , 2015; Marrus et al . , 2004 ) . Although null mutations in the GluRIIA subunit have been studied for decades ( Petersen et al . , 1997 ) , RNAi knock-down of this receptor has not been reported or characterized . We obtained an RNAi transgene targeting the GluRIIA subunit ( see Mateials and methods ) and compared the impact of postsynaptic knock-down of GluRIIA by the muscle driver G14-Gal4 ( G14 >GluRIIARNAi ) to GluRIIA null mutants ( Figure 1 ) . First , we immunostained the NMJ of wild type , GluRIIA mutants , and G14 >GluRIIARNAi with antibodies against the GluRIIA subunit as well as the common subunits GluRIIC and GluRIID ( Figure 1A ) . This revealed an absence of GluRIIA signals from GluRIIA mutants , as expected , with signals from GluRIIC and GluRIID persisting due to the remaining GluRIIB-containing receptors ( Figure 1A , B ) . Similarly , GluRIIA expression is almost completely absent in G14 >GluRIIARNAi , with no significant difference in fluorescence intensity compared to GluRIIA mutants ( Figure 1A , B ) . Indeed , quantitative PCR analysis revealed a dramatic reduction in the level of transcripts encoding the GluRIIA subunit in G14 >GluRIIARNAi , while levels of the other four subunits were not significantly changed ( Figure 1C ) . In addition , we quantified synaptic growth in these genotypes , finding a small reduction in bouton number in both GluRIIA mutants and G14 >GluRIIARNAi compared to controls ( Figure 1—figure supplement 1 ) , as reported previously ( Choi et al . , 2014; Schmid et al . , 2006; Sigrist et al . , 2002 ) . Finally , we examined synaptic physiology , which revealed a large reduction in mEPSP amplitude in both GluRIIA mutants and G14 >GluRIIARNAi compared to wild type and G14-Gal4/+ ( Figure 1D , E and Supplementary file 1 ) , while EPSP amplitudes were not significantly changed between these genotypes because of a homeostatic enhancement in presynaptic glutamate release ( quantal content; Figure 1D , H ) . Together , this demonstrates that postsynaptic knock down of the GluRIIA subunit effectively phenocopies GluRIIA mutants and induces the robust expression of PHP . Next , we sought to specifically knock down the GluRIIA subunit on one of the two muscles innervated by a single Ib motor neuron at the NMJ by selectively biasing expression of Gal4 . At the muscle 6/7 NMJ , a single Ib and a single Is motor neuron bifurcates to innervate both muscle 6 and 7 , with ~60% of the boutons from each motor neuron subtype innervating the larger muscle 6 , and ~40% innervating the smaller muscle 7 ( Figure 2—figure supplement 1A , B , C ) . To bias Gal4 expression selectively on muscle 6 , we modified a genetic manipulation using the H94-Gal4 driver , which expresses transiently on muscle six early in development ( Davis and Goodman , 1998 ) . Gal4 expression on muscle six is amplified and converted into constitutive expression by utilizing a cassette in which a flippase is co-expressed to excise a stop codon between the strong and ubiquitous Tubulin-promotor and Gal4 ( see Mateials and methods; [Choi et al . , 2014; Roy et al . , 2007] ) . Thus , this manipulation enables Gal4 to be strongly and consistently expressed specifically on muscle 6 ( Choi et al . , 2014 ) . We validated this approach by visualizing UAS-GFP selectively on muscle 6 , which demonstrated strong expression on muscle six and no detectable expression in either the motor neuron or the adjacent muscle 7 ( Figure 2—figure supplement 1A ) . Importantly , we confirmed that M6 > Gal4 driving a control RNAi ( M6 > mCherryRNAi ) did not have any significant impact on muscle surface area , synaptic growth , active zone numbers , or synaptic physiology compared to wild type ( Figure 2—figure supplement 1A–H ) . Thus , M6 > Gal4 enables strong and biased expression of Gal4 on muscle six without impacting synaptic growth or function . Finally , we evaluated this M6 >Gal4 system to determine whether the GluRIIA subunit could be specifically knocked down on muscle 6 , as observed using pan-muscle knock down in Figure 1 ( see schematic in Figure 2A ) . First , we performed immunocytochemistry at the larval NMJ with antibodies that label the neuronal membrane ( HRP ) and the GluRIIA subunit in wild type and following knock down of GluRIIA on muscle 6 ( M6 > GluRIIARNAi ) . We observed a near-absence of the GluRIIA signal specifically on muscle 6 , while GluRIIA expression on the adjacent muscle seven was unperturbed ( Figure 2B , C ) . Quantification of synaptic growth on muscles 6 and 7 in M6 > GluRIIARNAi revealed no significant change on muscle 7 , while a small but significant reduction was observed on muscle 6 , as expected ( Figure 2D ) . Finally , quantification of GluRIIA and GluRIID fluorescence levels at muscles 6 and 7 confirmed a large reduction of GluRIIA expression on muscle six and no significant change on muscle seven in M6 > GluRIIARNAi ( Figure 2C , E ) . Thus , M6 > GluRIIARNAi effectively and specifically eliminates GluRIIA expression on muscle six without altering glutamate receptor expression on the adjacent muscle 7 . A single Ib motor neuron ( RP3 ) bifurcates to innervate both muscles 6 and 7 at the Drosophila NMJ ( Broadie and Bate , 1993 ) ( Figure 3A ) . Having established strong and selective knock down of the GluRIIA subunit on muscle six in M6 > GluRIIARNAi , we next characterized synaptic function and homeostatic plasticity . As expected , mEPSP amplitudes on muscle 6 were diminished at M6 > GluRIIARNAi NMJs , while mEPSP amplitudes were not affected on the adjacent muscle 7 ( Figure 3A–C ) . We considered three possibilities for how presynaptic neurotransmitter release sites may be modulated within the single Ib terminal in response to GluRIIA knock down exclusively on muscle 6 . First , if PHP signaling is communicated to synapses in the Ib motor neuron innervating muscle 6 , but the entire motor neuron undergoes PHP adaptations , then quantal content would be enhanced on both muscles 6 and 7 . Second , PHP signaling may be communicated to synapses innervating only muscle 6 , but PHP expression may be occluded without simultaneous signaling also received from muscle 7 , leading to no change in quantal content on either muscle . Finally , if PHP signaling is target-specific and compartmentalized , then quantal content should be selectively enhanced on synapses innervating muscle 6 in response to reduced GluRIIA expression , while synaptic function at synapses innervating the adjacent muscle seven would be unchanged . Results of electrophysiological recordings were consistent with this last model: EPSP amplitude was similar on both muscles 6 and 7 in M6 > GluRIIARNAi and not significantly different compared to wild type ( Figure 3A , D ) . Indeed , quantal content was selectively enhanced only at synapses innervating muscle 6 , while quantal content at synapses innervating muscle seven was unaffected ( Figure 3B , E ) . Thus , PHP can be induced and expressed exclusively at a subset of synapses within the same motor neuron without influencing neurotransmitter release at neighboring sites . The results above suggest that while PHP was chronically induced and expressed specifically on synapses innervating muscle 6 , the adjacent synapses within the same motor neuron that innervate muscle seven were apparently not affected . One possibility is that PHP adaptations are induced throughout the entire motor neuron innervating both muscles 6 and 7 , but that negative regulators are active that repress or occlude the expression of PHP on synapses innervating muscle 7 . Indeed , such a model has been proposed ( Müller et al . , 2011 ) . In this case , PHP should not be capable of being induced or expressed at the synapses innervating muscle seven in M6 > GluRIIARNAi . PHP can be acutely induced and expressed through a pharmacological blockade of the postsynaptic GluRs using a 10-min incubation in the presence of the antagonist philanthotoxin-433 ( PhTx; [Frank et al . , 2006] ) . This results in an acute reduction in mEPSP amplitude due to blockade of GluRIIA-containing receptors in the postsynaptic muscle , but EPSP amplitudes are maintained because of PHP expression . We reasoned that acute application of PhTx to M6 > GluRIIARNAi NMJs would enable us to determine whether PHP could be induced and expressed at synapses innervating muscle 7 following chronic expression of PHP on the adjacent synapses innervating muscle 6 . We therefore applied PhTx to wild type and M6 > GluRIIARNAi synapses . As expected , this caused a large reduction in mEPSP amplitudes at wild-type muscles 6 and 7 , as well as muscle 7 in M6 > GluRIIARNAi , while a small reduction in mEPSP amplitude was observed at muscle 6 ( Figure 4A–C ) . Interestingly , EPSP amplitudes at both muscles 6 and 7 in wild type and M6 > GluRIIARNAi NMJs were maintained at similar levels ( Figure 4A , D ) due to the homeostatic enhancement of quantal content ( Figure 4B , E ) . Together , this demonstrates that PHP can be acutely induced and chronically expressed at distinct presynaptic release sites within the same neuron according to the state of GluR functionality at postsynaptic compartments opposing these sites . One important presynaptic expression mechanism that enables the enhanced efficacy necessary for PHP expression is an increase in the readily releasable synaptic vesicle pool ( RRP; [Kiragasi et al . , 2017; Müller et al . , 2015; Weyhersmüller et al . , 2011] ) . The RRP is defined as the pool of vesicles that are primed and available for immediate release upon strong synaptic stimulation ( Rosenmund and Stevens , 1996 ) . Although PHP expression appears to be compartmentalized , it is unknown whether conventional homeostatic modulations to the presynaptic terminal are similarly compartmentalized , or rather whether novel mechanisms are utilized in M6 > GluRIIARNAi . In particular , synaptic vesicles are highly mobile and can rapidly traffic between adjacent boutons at presynaptic terminals ( Darcy et al . , 2006; Kahms and Klingauf , 2018 ) , and vesicle pools can span multiple presynaptic terminals ( Staras et al . , 2010 ) . Hence , it is possible that changes in the RRP following PHP expression at a subset of presynaptic terminals may influence vesicle pools at neighboring synapses that do not directly experience local PHP signaling . We therefore measured the size of RRP in wild type and M6 > GluRIIARNAi synapses separately innervating muscles 6 and 7 . To determine RRP size , we performed two electrode voltage clamp ( TEVC ) measurements using high-frequency stimulation ( 60 Hz ) in elevated external calcium concentrations ( 3 mM ) and measured the cumulative EPSC ( Figure 5A–D; ( Kiragasi et al . , 2017; Müller et al . , 2015; Weyhersmüller et al . , 2011 ) . We observed a ~ 65% increase in the estimated RRP size that was restricted to boutons innervating muscle 6 in M6 > GluRIIARNAi , similar in magnitude to what has been reported for muscle 6 synapses in which PHP is expressed following PhTx application ( Müller et al . , 2015 ) and by loss of GluRIIA ( Kiragasi et al . , 2017; Weyhersmüller et al . , 2011 ) . However , no significant change in RRP size was observed at synapses innervating the adjacent muscle 7 ( Figure 5A , E ) . Thus , the homeostatic modulation of the RRP is restricted to presynaptic terminals that oppose postsynaptic compartments with reduced GluR functionality , and does not ‘spread’ to influence vesicle pools at adjacent release sites . Next , we examined whether a change in the number of functional release sites ( N ) accompanies the compartmentalized expression of PHP . N is defined as the number of functional release sites and is one of the three basic parameters used to describe synaptic transmission , where quantal content ( QC ) is the product of N , P ( release probability ) , and Q ( quantal size ) . Although there is no major difference in the anatomical number of active zones at NMJs of wild-type and GluRIIA mutants ( Frank et al . , 2006; Goel et al . , 2017; Penney et al . , 2012; Schmid et al . , 2006 ) , an increase in the fraction of active zones that participate in release has been reported following the expression of PHP ( Davis and Müller , 2015; Newman et al . , 2017; Penney et al . , 2016 ) . Indeed , the value of N is significantly increased following PhTx application and in GluRIIA-mutant synapses ( Müller et al . , 2012; Weyhersmüller et al . , 2011 ) . We determined the number of functional release sites at NMJs of muscles 7 and 6 in wild type , GluRIIA mutants , and M6 >GluRIIARNAi using a variance-mean plot analysis ( Böhme et al . , 2016; Clements and Silver , 2000 ) . We performed TEVC recordings over a range of increasing extracellular calcium concentrations , from 0 . 5 mM to 6 mM ( Figure 6B , F ) . The variance in the amplitude of repeated evoked responses fluctuates across different calcium concentrations in relation to the proportion of total release sites that participate in these responses . At low-calcium concentrations , the number of release sites that participate in the evoked response is low , and the variance is therefore small in this condition . As the extracellular calcium concentration is elevated , the variance then increases with increasing release during repeated evoked responses due to an increase in the number of release sites that participate in synaptic transmission . At very high extracellular calcium concentrations , the variance is then reduced due to saturation of the total number of releasable sites . We plotted the variance of EPSC responses across increasing calcium conditions against the mean of the EPSC amplitude at each individual calcium conditions from recordings at both muscle 7 and muscle 6 ( Figure 6C , G ) . This analysis resulted in a parabolic behavior of variance-mean plot due to the binomial nature of the fluctuation ( Clements and Silver , 2000; Weyhersmüller et al . , 2011 ) . The number of functional release sites ( N ) was determined by fitting a parabola to the variance-mean plot ( see Mateials and methods ) . Based on this result , the value of N increased in GluRIIA-mutant NMJs compared to wild type at both muscles 7 and 6 ( Figure 6D , H ) , consistent with what was reported following acute PHP expression and indicating that chronic PHP expression requires the recruitment of additional functional release sites to participate in presynaptic neurotransmitter release . Further , a similar increase in N was observed at muscle 6 in M6 > GluRIIARNAi ( Figure 6H ) , while no significant change was found at muscle 7 ( Figure 6D ) , suggesting that the biased induction of PHP results in the compartmentalized expression of the same mechanisms observed in GluRIIA mutants ( Figure 6A , E ) . Therefore , retrograde PHP expression is achieved by elevating the RRP and recruiting additional functional release sites to participate in transmission , with specificity according to the excitability state of their postsynaptic partners . The postsynaptic induction mechanisms that drive PHP retrograde signaling are unclear . However , reductions in postsynaptic CaMKII activity have been proposed to mediate the induction of retrograde PHP signaling ( Haghighi et al . , 2003 ) . Indeed , modulations to postsynaptic CaMKII phosphorylation has been demonstrated to occur in a synapse-specific and activity-dependent manner at the Drosophila NMJ ( Hodge et al . , 2006 ) . Consistent with this idea , reductions in the level of phosphorylated ( active ) CaMKII were observed specifically at postsynaptic densities of Type Ib boutons following PhTx application and in GluRIIA mutants ( Goel et al . , 2017; Newman et al . , 2017 ) . To determine whether this same signaling system mediates PHP induction and is restricted to specific targets at M6 > GluRIIARNAi NMJs , we examined pCaMKII levels at postsynaptic densities of Is and Ib boutons on both muscles 6 and 7 . Ib and Is boutons were distinguished by differential areas and intensity signals of the postsynaptic scaffold Discs large ( Dlg ) . We observed a significant reduction in the mean intensity of pCaMKII in Ib postsynaptic densities only on muscle 6 of M6 > GluRIIARNAi , while no significant difference was observed at Ib synapses in the adjacent muscle 7 , nor were any changes found at Is synapses on either muscle ( Figure 7A–E ) . Thus , postsynaptic pCaMKII levels are diminished and compartmentalized at Ib boutons specifically on muscle 6 without impacting pCaMKII levels on neighboring Is boutons or at the adjacent muscle 7 in M6 > GluRIIARNAi . Finally , we tested whether a reduction in postsynaptic CaMKII activity was required for retrograde PHP signaling . In particular , if PHP signaling is induced at Ib postsynaptic densities through diminished pCaMKII levels , as suggested by immunostaining , then postsynaptic overexpression of a constitutively active , phospho-mimetic form of CaMKII , CaMKIIT287D , should inhibit or occlude PHP induction and expression ( Haghighi et al . , 2003 ) . We first expressed CaMKIIT287D alone on muscle 6 ( M6>CaMKIIT287D ) to determine if baseline synaptic function was influenced by constitutively active CaMKII ( Figure 8A ) . We found no significant difference in synaptic physiology on muscles 6 or 7 in this condition ( Figure 8A–D ) . Next , we expressed constitutively active CaMKII in combination with GluRIIA knock down on muscle 6 ( M6 > GluRIIARNAi + CaMKIIT287D ) . If reduced CaMKII activity were functionally required for retrograde PHP signaling , and not just a compartmentalized biomarker of GluRIIA levels and/or activity at Ib boutons , then constitutively active CaMKII should disrupt PHP expression at M6 >GluRIIARNAi synapses . Indeed , EPSP amplitude was not maintained at baseline levels due to a failure to homeostatically increase quantal content in M6 > GluRIIARNAi + CaMKIIT287D ( Figure 8E–H ) . In this condition , quantal content was significantly increased ( 134 ± 8% increase for M6 > GluRIIARNAi + CaMKIIT287D compared to M6>CaMKIIT287D ) but was below the level necessary to maintain synaptic strength ( 166 ± 9% increase in quantal content for M6 > GluRIIARNAi compared to wild type ) . This data is consistent with reduced CaMKII activity , compartmentalized at Ib postsynaptic densities , being required for retrograde PHP signaling . We propose a model schematized in Figure 8E .
GluRs are dynamically trafficked in postsynaptic compartments where they mediate the synapse-specific expression of Hebbian plasticity such as LTP ( Herring and Nicoll , 2016; Matsuzaki et al . , 2004 ) and homeostatic plasticity , including receptor scaling ( Hou et al . , 2008; Pozo and Goda , 2010; Sutton et al . , 2006 ) . In contrast , homeostatic plasticity at the human , mouse , and fly NMJ is expressed through a presynaptic enhancement in neurotransmitter release , but is induced through a diminishment of postsynaptic neurotransmitter receptor functionality ( Cull-Candy et al . , 1980; Frank et al . , 2006; Petersen et al . , 1997; Wang et al . , 2016b ) . Using biased expression of Gal4 to reduce GluR levels on only one of the two muscle targets innervated by a single motor neuron , we demonstrate that the inductive signaling underlying PHP is compartmentalized at the postsynaptic density , and does not influence activity at synapses innervating the adjacent muscle . Postsynaptic changes in CaMKII function and activity have been associated with PHP retrograde signaling ( Goel et al . , 2017; Haghighi et al . , 2003; Newman et al . , 2017 ) . Consistent with this compartmentalized inductive signaling , we observed pCaMKII levels to be specifically reduced at postsynaptic densities of Ib boutons in which GluR expression is perturbed , while pCaMKII was unchanged at postsynaptic compartments opposite to Is boutons and at NMJs in the adjacent muscle with normal GluR expression . Further , postsynaptic overexpression of the constitutively active CaMKII occludes the expression of PHP . Similar synapse-specific control of postsynaptic CaMKII phosphorylation , modulated by activity , has been previously observed ( Hodge et al . , 2006 ) . As noted in other studies ( Goel et al . , 2017; Newman et al . , 2017 ) , this localized reduction in pCaMKII provides a plausible mechanism for the inductive PHP signaling restricted to and compartmentalized at Ib synapses . How does a perturbation to GluR function lead to a reduction in CaMKII activity that is restricted to postsynaptic densities opposing Type Ib boutons ? Recent evidence suggests that distinct mechanisms regulate pCaMKII levels during retrograde PHP signaling depending on pharmacologic or genetic perturbation to glutamate receptors and the role of protein synthesis ( Goel et al . , 2017 ) . Scaffolds at postsynaptic densities are associated in complexes with GluRs and CaMKII ( Gillespie and Hodge , 2013; Hodge et al . , 2006; Koh et al . , 1999; Lu et al . , 2003; Mullasseril et al . , 2007 ) . Intriguingly , the scaffold dCASK is capable of modulating CaMKII activity at specific densities in an activity-dependent fashion ( Hodge et al . , 2006; Malik et al . , 2013 ) . Further , CaMKII activity can regulate plasticity with specificity at subsets of synapses in Drosophila and other systems ( Griffith , 2004; Hodge et al . , 2006; Merrill et al . , 2005 ) . Although we cannot rule out intra-cellular cross talk between Is and Ib boutons , as GluRIIA is reduced at postsynaptic sites of both neuronal subtypes , it is striking that reductions in pCaMKII are restricted to Ib postsynaptic compartments . An attractive model , therefore , is that the postsynaptic density isolates calcium signaling over chronic time scales to compartmentalize PHP induction . The membranous complexity and geometry of the SSR at the Drosophila NMJ may be the key to restricting calcium signaling at these sites , as this structure can have major impacts on ionic signaling during synaptic transmission ( Nguyen and Stewart , 2016; Teodoro et al . , 2013 ) . These properties , in turn , may lead to local modulation of CaMKII function ( Goel et al . , 2017; Griffith , 2004; Haghighi et al . , 2003; Newman et al . , 2017 ) . Interestingly , Drosophila mutants with defective SSR elaboration and complexity have been associated with defects in PHP expression ( Koles et al . , 2015 ) . In the mammalian central nervous system , it is well established that dendritic spines function as biochemical compartments that isolate calcium signaling while enabling propagation of voltage changes ( Svoboda et al . , 1997; Yuste and Denk , 1995 ) , and it is tempting to speculate that the SSR may subserve similar functions at the Drosophila NMJ to enable synapse-specific retrograde signaling . The homeostatic modulation of presynaptic neurotransmitter release is compartmentalized at the terminals of Type Ib motor neurons . It was previously known that PHP can be acutely induced and expressed without any information from the cell body of motor neurons ( Frank et al . , 2006 ) . Our data suggests that the signaling necessary for PHP expression is even further restricted to specific postsynaptic densities and presynaptic boutons , demonstrated through several lines of evidence . First , quantal content is specifically enhanced at boutons innervating muscle 6 in M6 > GluRIIARNAi without measurably impacting transmission on the neighboring boutons innervating muscle 7 . In addition , PHP can be acutely induced at synapses innervating muscle 7 despite PHP having been chronically expressed at muscle 6 . Finally , the homeostatic modulation of the RRP and enhancement of the functional number of release sites is fully expressed regardless of whether PHP is induced at all Type Ib boutons or only a subset . Thus , PHP signaling is orchestrated at specific boutons according to the state of GluR functionality of their synaptic partners and does not influence neighboring boutons within the same motor neuron . Although the compartmentalized expression of PHP was not unexpected , there was precedent to suspect inter-bouton crosstalk during homeostatic signaling . In the dynamic propagation of action potentials along the axon , the waveform could , in principle , change following PHP expression to globally modulate neurotransmission at all release sites in the same neuron . However , voltage imaging did not identify any change in the action potential waveform at individual boutons following PHP signaling ( Ford and Davis , 2014; Gaviño et al . , 2015 ) , and we did not observe any impact on neighboring boutons despite PHP being induced at a subset of synapses in the same motor neuron . Further , mobilization of an enhanced readily releasable synaptic vesicle pool is necessary for the expression of PHP ( Davis and Müller , 2015; Kiragasi et al . , 2017; Müller et al . , 2015; Weyhersmüller et al . , 2011 ) , and synaptic vesicles and pools are highly mobile within and between presynaptic compartments ( Darcy et al . , 2006; Kahms and Klingauf , 2018; Staras et al . , 2010 ) . Hence , it was conceivable that a mobilized RRP , induced at some presynaptic compartments , may be promiscuously shared between other boutons . However , while we observed a large enhancement in the RRP at synapses innervating muscle 6 in M6 > GluRIIARNAi , this adaptation had no impact on the RRP at adjacent presynaptic compartments innervating muscle 7 . Thus , PHP signaling is constrained to boutons innervating one of two postsynaptic targets and does not ‘spread’ to synapses innervating the adjacent target despite sharing common cytosol , voltage , and synaptic vesicles . What molecular mechanisms mediate the remarkable specificity of PHP expression at presynaptic compartments ? One attractive possibility is that active zones themselves are fundamental units and act as substrates for the homeostatic modulation of presynaptic function . The active zone scaffold BRP remodels during both acute and chronic PHP expression ( Goel et al . , 2017; Weyhersmüller et al . , 2011 ) , and other active zone proteins are likely to participate in this remodeling ( Gratz et al . , 2018 ) . Indeed , many genes encoding active zone components are required for PHP expression , including the calcium channel cac ( Frank et al . , 2006 ) and auxiliary subunit α2-δ ( Wang et al . , 2016a ) , the piccolo homolog fife ( Bruckner et al . , 2017 ) , the scaffolds RIM ( Rab3-interacting Molecule; [Müller et al . , 2012] ) and RIM-binding protein ( RBP; [Müller et al . , 2015] ) , and the kainite receptor DKaiR1D ( Kiragasi et al . , 2017 ) . If individual active zones can undergo the adaptations necessary and sufficient for PHP expression , this would imply that PHP can be induced and expressed with specificity at individual active zones . Indeed , the BRP cytomatrix stabilizes calcium channel levels at the active zone ( Kittel et al . , 2006 ) , and also controls the size of the RRP ( Matkovic et al . , 2013 ) , two key presynaptic expression mechanisms that drive PHP . Further , we and others have observed the recruitment of new functional release sites following both chronic and acute PHP expression ( Davis and Müller , 2015; Newman et al . , 2017; Weyhersmüller et al . , 2011 ) , suggesting that previously silent active zones become ‘awakened’ and utilized to potentiate presynaptic neurotransmitter release . Interestingly , presynaptic GluRs , localized near active zones , are necessary for PHP expression and have the capacity to modulate release with specificity at individual active zones ( Kiragasi et al . , 2017 ) . Thus , active zones can remodel with both the specificity and precision necessary and sufficient for compartmentalized PHP expression . If each active zone operates as an independent homeostat to adjust release efficacy in response to target-specific changes , how is information transfer at individual sites integrated to ensure stable and stereotypic global levels of neurotransmission ? One speculative possibility is that active zones at terminals of each neuron are endowed with a total abundance of material that is tightly controlled and sets stable global levels of presynaptic neurotransmitter release . Such active zone material may be sculpted with considerable heterogeneity within presynaptic terminals , varying in number , size , and density . Consistent with such a possibility , mutations in the synaptic vesicle component Rab3 exhibit extreme changes in active zone size , number , and density , but stable global levels of neurotransmission ( Graf et al . , 2009 ) . Within this global context , plasticity mechanisms may operate at individual active zones , superimposed as independent homeostats to adaptively modulate synaptic strength . In addition , there is intriguing evidence for the existence of ‘nanocolumns’ between presynaptic active zones and postsynaptic GluRs that form structural and functional signaling complexes ( Biederer et al . , 2017; Tang et al . , 2016 ) . One particularly appealing possibility , therefore , is that a dialogue traversing synaptic nanocolumns functions to convey the retrograde signaling and active zone remodeling necessary for PHP expression at individual release sites . Studies in mammalian neurons have revealed parallel links between the functional plasticity of active zones , including their structure and size , and the homeostatic modulation of neurotransmitter release ( Glebov et al . , 2017; Matz et al . , 2010; Murthy et al . , 2001; Schikorski and Stevens , 2001 ) . Such intercellular signaling systems are likely to modify synaptic structure and function to not only establish precise pre- and post-synaptic apposition during development , but also to maintain the plasticity necessary for synapses to persist with the flexibility and stability to last a lifetime .
Drosophila stocks were raised at 25°C on standard molasses food . The w1118 strain is used as the wild-type control unless otherwise noted , as this is the genetic background of the transgenic lines and other genotypes used in this study . The following fly stocks were used: G14-Gal4 ( Aberle et al . , 2002 ) ; GluRIIARNAi ( p{TRiP . JF02647}attP2}; Bloomington Drosophila Stock Center ( BDSC ) ) ; UAS-CaMKIIT287D ( BDSC ) ; mCherryRNAi ( p{VALIUM20-mCherry}attP2}; BDSC ) ; GluRIIASP16 ( Petersen et al . , 1997 ) ; M6-Gal4 ( Tub-FRT-STOP-FRT-Gal4 , UAS-FLP , UAS-CD8-GFP; H94-Gal4 , nSyb-Gal80 ) ( Choi et al . , 2014 ) . Third-instar larvae were dissected in ice cold 0 Ca2+ modified HL-3 saline and immunostained as described ( Kikuma et al . , 2017 ) . Briefly , larvae were fixed in Bouin's fixative ( Sigma , St . Louis , MO; HT10132-1L ) for 2 min and washed with PBS containing 0 . 1% Triton X-100 ( PBST ) for 30 min , then blocked for 1 hr in 5% Normal Donkey Serum ( NDS ) . Following overnight incubation in primary antibodies at 4°C , preparations were washed in PBST , incubated in secondary antibodies for 2 hr , washed and mounted in VectaShield ( Vector Laboratories ) . The following antibodies were used: guinea pig anti-vGlut ( 1:2000; ( Chen et al . , 2017 ) ; rabbit anti-DLG ( 1:5000; [Pielage et al . , 2005] ) ; mouse anti-GluRIIA ( 8B4D2; 1:100; Developmental Studies Hybridoma Bank ( DSHB ) ) ; rabbit anti-GluRIIC ( 1:1000; [Marrus et al . , 2004] ) ; guinea pig anti-GluRIID ( 1:1000; [Perry et al . , 2017] ) ; mouse anti-pCaMKII ( 1:100; MA1-047; Invitrogen ) ; mouse anti-GFP ( 8H11; 1:100; DSHB ) ; Tetramethylrhodamine ( TRITC ) -conjugated phalloidin ( R415; Thermo Fisher ) ; Alexa Fluor 647-conjugated goat anti-HRP ( 1:200; Jackson ImmunoResearch ) . Donkey anti-mouse , anti-guinea pig , and anti-rabbit Alexa Fluor 488- , DyLight 405- , and Cyanine 3 ( Cy3 ) -conjugated secondary antibodies ( Jackson ImmunoResearch ) were used at 1:400 . Samples were imaged using a Nikon A1R Resonant Scanning Confocal microscope equipped with NIS Elements software and a 100x APO 1 . 4NA oil immersion objective using separate channels with four laser lines ( 405 nm , 488 nm , 561 nm , and 637 nm ) as described ( Chen et al . , 2017 ) . All genotypes were immunostained in the same tube with identical reagents , then mounted and imaged in the same session . z-stacks were obtained using identical settings for all genotypes with z-axis spacing between 0 . 15 µm to 0 . 2 µm within an experiment and optimized for detection without saturation of the signal . Both Type Ib and Is boutons were counted using vGlut and HRP-stained NMJ terminals on muscle 6/7 of segment A3 , considering each vGlut puncta to be a bouton . Muscle surface area was calculated by creating a mask around the phalloidin channel that labels the entire muscle . The general analysis toolkit in the NIS Elements software was used to quantify GluRIIA/C/D and pCamKII intensity levels by applying intensity thresholds and filters to binary layers for each the channels of the maximum intensity projection images . To quantify GluR intensity levels , the total fluorescence intensity of each GluRIIA , GluRIIC , or GluRIID puncta was averaged over the NMJ area covered by muscle 6 or muscle seven separately to determine mean fluorescence intensity; this value was then normalized as a percentage of wild type for the corresponding muscle . For analysis of pCaMKII levels , Ib and Is regions were identified using DLG and HRP on muscle 6/7 of segment A2 and A3 , and only the pCamKII signal that co-localized with DLG was summated and divided by the bouton area under consideration to obtain average pCamKII . quantitative PCR: quantitative PCR ( qPCR ) was performed using the Luna Universal One-Step RT-qPCR Kit ( NEB , E3005S ) according to the manufacturer’s instructions . RNA was isolated and prepared from body wall tissue as described previously ( Chen and Dickman , 2017 ) . 20 ng of total RNA was used as the template for each reaction . Three biological replicates were performed for each sample and the comparative Ct method was used for qPCR data analysis . The following primers were used ( fwd; rev: 5’−3’ ) : GluRIIA: TCCTCAACTTGGAACTGGAAAG; CGTACTTTTCCCTGCCTCTG . GluRIIB: GCGAATACAGATGAATGGGATG; TGCATGAAGGGTACAGTGAAG . GluRIIC: CGGAAAACTGGACAAGGAAAC; AGCTGCATAAAGGGCACTG . GluRIID: CCCAAGCTGTCAACTTCAATG; CCATAACCCTGGAACTGATTGT . GluRIIE: CGGTGCAAAGAAAACTGGATC; GTCTTAACTCGATTCACTCCCTC . αTub84D ( control ) : CTACAACTCCATCCTAACCACG; CAGGTTAGTGTAAGTGGGTCG . All dissections and recordings were performed in modified HL-3 saline ( Dickman et al . , 2005; Kiragasi et al . , 2017; Stewart et al . , 1994 ) at room temperature containing ( in mM ) : 70 NaCl , 5 KCl , 10 MgCl2 , 10 NaHCO3 , 115 Sucrose , 5 Trehelose , 5 HEPES , and 0 . 4 CaCl2 ( unless otherwise specified ) , pH 7 . 2 . Neuromuscular junction sharp electrode recordings were performed on muscles 6 or 7 of abdominal segments A2 or A3 in wandering third-instar larvae . Biased Gal4 expression was verified by verifying GFP fluorescence on the particular muscle before experimentation , and recordings were performed at only the GFP-positive muscle 6 and the adjacent muscle 7 . Muscle input resistance ( Rin ) and resting membrane potential ( Vrest ) were monitored during each experiment ( Supplementary file 1 ) . To acutely block postsynaptic receptors , larvae were incubated with or without philanthotoxin-433 ( 20 μM; Sigma ) and resuspended in HL-3 for 10 mins , as described ( Dickman and Davis , 2009; Frank et al . , 2006 ) . The readily releasable pool ( RRP ) size was estimated by analyzing cumulative EPSC amplitudes while recording using a two-electrode voltage clamp ( TEVC ) configuration . Muscles were clamped to −70 mV and EPSCs were evoked with a 60 Hz , 60 stimulus train while recording in HL-3 supplemented with 3 mM Ca2+ . A line fit to the linear phase ( stimuli # 18–30 ) of the cumulative EPSC data was back-extrapolated to time 0 . The RRP value was estimated by determining the extrapolated EPSC value at time 0 and dividing this value by the average mEPSC amplitude . More details of the RRP size analysis can be found at Bio-protocol ( Goel et al . , 2019 ) . Data used in the variance-mean plot was obtained from TEVC recordings using an initial 0 . 5 mM Ca2+ concentration , which was subsequently increased to 1 . 5 , 3 . 0 , and 6 . 0 mM through saline exchange using a peristaltic pump ( Langer Instruments , BT100-2J ) . EPSC amplitudes were monitored during the exchange , and 30 EPSC ( 0 . 5 Hz stimulation rate ) recordings were performed in each calcium condition . To obtain the variance-mean plot , the variance ( squared standard deviation ) and mean ( averaged evoked amplitude ) were calculated from the 30 EPSCs at each individual Ca2+ concentration . The variance was then plotted against the mean for each specific calcium condition using MATLAB software ( MathWorks , USA ) . One additional data point , in which variance and mean are both theoretically at 0 , was used for Ca2+-free saline . Data from these five conditions were fit with a standard parabola ( variance = Q*Ī -Ī2/N ) , where Q is the quantal size , Ī is the mean evoked amplitude ( x-axis ) , and N is the functional number of release sites . N , as a parameter of the standard parabola , was directly calculated for each cell by best parabolic fit . All data are presented as mean ± SEM . Data was compared using either a one-way ANOVA followed by Tukey’s multiple comparison test , or using a Student’s t-test ( where specified ) . Data was analyzed using Graphpad Prism or Microsoft Excel software , with varying levels of significance assessed as p<0 . 05 ( * ) , p<0 . 01 ( ** ) , p<0 . 001 ( *** ) , p<0 . 0001 ( **** ) , ns: not significant . M6 >GluRIIARNAi results were compared to M6 > mCherryRNAi ( Figure 2—figure supplement 1 ) and G14/+ , in addition to w1118; in no case did the control change the statistically significant result . See Supplementary file 1 for further statistical details and values . | Everything we think and do is the result of communication between neurons . This communication takes place at junctions called synapses . When two nerve cells or neurons communicate at a synapse , the output terminal of the first cell releases a chemical called a neurotransmitter . This binds to receiver proteins , or receptors , on the second cell . When this communication is interrupted , synapses can adapt to maintain a stable dialogue between them . This can occur in two ways . Either the first neuron starts to release more neurotransmitter from its output terminal , or the second neuron produces extra receptors with which to detect the neurotransmitter . But how specific are these changes ? The brain contains far more synapses than neurons because each neuron can form synapses with many other cells . Can a neuron adjust how much of the neurotransmitter it releases at some of its synapses while leaving the others unchanged ? Li et al . have now addressed this question by studying a special type of synapse that forms between neurons and muscles , known as a neuromuscular junction . At one particular neuromuscular junction in fruit flies , a single neuron splits into two output terminals , each of which forms a synapse with a different muscle . Li et al . show that when the number of neurotransmitter receptors in one of the muscles is artificially reduced , the associated output terminal compensates by increasing its neurotransmitter release . By contrast , the other output terminal remains unaffected . This suggests that a neuron can induce remarkably specific changes in a subset of its synapses . This discovery paves the way towards identifying the smallest possible unit of change that can occur in the neurons’ ability to communicate . This unit may in turn be the smallest change that can support learning . Such knowledge will help us understand how the nervous system processes and stabilizes information transfer , both in health and after injury or disease . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"neuroscience"
] | 2018 | Synapse-specific and compartmentalized expression of presynaptic homeostatic potentiation |
Bacterial Xer site-specific recombinases play an essential genome maintenance role by unlinking chromosome multimers , but their mechanism of action has remained structurally uncharacterized . Here , we present two high-resolution structures of Helicobacter pylori XerH with its recombination site DNA difH , representing pre-cleavage and post-cleavage synaptic intermediates in the recombination pathway . The structures reveal that activation of DNA strand cleavage and rejoining involves large conformational changes and DNA bending , suggesting how interaction with the cell division protein FtsK may license recombination at the septum . Together with biochemical and in vivo analysis , our structures also reveal how a small sequence asymmetry in difH defines protein conformation in the synaptic complex and orchestrates the order of DNA strand exchanges . Our results provide insights into the catalytic mechanism of Xer recombination and a model for regulation of recombination activity during cell division .
In organisms with circular genomes , homologous recombination-mediated repair behind a stalled replication fork can join the two nascent daughter chromosomes , resulting in a chromosome dimer ( Barre et al . , 2001 ) . Dimer formation prohibits proper segregation of the genetic information at cell division ( Figure 1A ) , and must be repaired to produce viable progeny . In bacteria and archaea , chromosome dimers are monomerized by members of a large family of tyrosine recombinases , the Xer recombinases ( Blakely et al . , 1991 , 1993 ) . These enzymes act by promoting recombination between two DNA sites , called dif . The dif site is normally present in a single copy at the replication terminus ( Carnoy and Roten , 2009; Kuempel et al . , 1991 ) , but it is duplicated in chromosome dimers , so that intramolecular recombination results in separation ( ‘resolution’ ) of the two chromosome copies ( Figure 1A ) . Removal of the xer genes or the dif site results in increased DNA content , activation of the SOS response , cell filamentation , and cell death ( Britton and Grossman , 1999; Debowski et al . , 2012a; Hendricks et al . , 2000; Pérals et al . , 2000; Val et al . , 2008 ) . Besides chromosome dimer resolution , Xer recombinases can support plasmid resolution and mobilization of the cholera toxin phage CTXϕ and pathogenicity islands ( Das et al . , 2013; Fischer et al . , 2010 ) . 10 . 7554/eLife . 19706 . 003Figure 1 . Xer recombination . ( A ) The role of Xer recombination in the maintenance of bacterial chromosomes . Homologous recombination behind a stalled replication fork can result in a chromosome dimer . Xer recombinases monomerize these to rescue healthy genome segregation . The absence of Xer leads to cell division arrest and cell death . ( B ) Schematic representation of tyrosine site-specific recombination . Two recombinase monomers ( beige ovals ) bind one specific DNA site ( grey ) and two such sites are aligned in antiparallel in a tetrameric synaptic complex ( i ) . The catalytic tyrosines of two symmetry-related protomers ( red star ) cleave one strand of each DNA , creating a covalent 3’ phosphotyrosyl bond and a free 5’ hydroxyl group ( ii ) . The latter then attacks the phosphotyrosyl bond of the partner DNA , forming the HJ intermediate ( iii ) . Following an isomerization step , the second pair of protomers becomes catalytically active ( iv ) , leading to a reciprocal set of cleavage and strand exchange reactions that resolve the HJ and generate the recombined DNA products ( v–vi ) . Only two protein subunits are active in the tetramer at a time ( ‘half-of-the-sites reactivity’ ) , strictly ordering the chemical steps to ensure faithful progression of the recombination reaction to the desired products . Note that the DNA substrates are drawn with the strand going 3’ to 5’ on the top . DOI: http://dx . doi . org/10 . 7554/eLife . 19706 . 003 In E . coli , Xer recombination is carried out by cooperation of two similar enzymes XerC and XerD ( 37% identity ) that act together to bind and recombine dif sites ( Blakely et al . , 1993 ) . Many other organisms ( including Lactococcus , Helicobacter , Campylobacter spp . and archaea ) employ a single Xer recombinase system ( Carnoy and Roten , 2009; Cortez et al . , 2010; Debowski et al . , 2012a; Le Bourgeois et al . , 2007; Leroux et al . , 2013 ) , with a prime example , XerH/difH , found in Helicobacter pylori , a gastric pathogen implicated in peptic ulcer disease and gastric cancer . Xer recombinases are members of the tyrosine site-specific recombinase superfamily , a large group of enzymes that catalyze DNA breakage and rejoining using a conserved tyrosine nucleophile ( Grindley et al . , 2006; Guo et al . , 1997; Midonet and Barre , 2014; Nunes-Düby et al . , 1998 ) . Tyrosine recombinases promote various programmed DNA rearrangements including the monomerization of phage , plasmid and chromosome multimers , resolution of hairpin telomeres , and the movement of virulence and antibiotic resistance carrying integrative mobile genetic elements ( including phages and transposons ) ( Grindley et al . , 2006; Jayaram et al . , 2015; Midonet and Barre , 2014 ) . In addition , tyrosine recombinases ( as exemplified by Cre and Flp ) provide powerful genetic engineering tools that are widely used to carry out mutagenesis and DNA insertion in eukaryotic chromosomes ( Nagy , 2000; Turan et al . , 2011 ) . Tyrosine recombinases share a common chemical mechanism that involves step-wise breakage and exchange of four DNA strands in pairs , proceeding through a characteristic four-way Holliday junction ( HJ ) DNA intermediate ( Figure 1B ) ( Gopaul and Duyne , 1999; Grindley et al . , 2006; Holliday , 2007 ) . They cut each DNA strand with a polarity creating a covalent 3’ phosphotyrosyl protein-DNA linkage and a free 5’ hydroxyl group . The DNA ends then go on to join with the complementary ends of the partner DNA strand generating the recombined products . All DNA cleavage and rejoining reactions take place in an ordered protein-DNA synaptic complex , comprising four recombinase molecules holding the two recombination partner DNA molecules together . An unusual feature of Xer recombination at dif is that it requires an accessory factor , FtsK ( Aussel et al . , 2002; Debowski et al . , 2012a; Le Bourgeois et al . , 2007; Leroux et al . , 2013; Nolivos et al . , 2010; Steiner et al . , 1999 ) . This DNA motor protein localizes to the bacterial cell division septum and contributes to segregating the sister chromosomes into the daughter cells by translocating towards their replication termini . On chromosome dimers , FtsK stops at the Xer-bound dif sites and activates recombination , triggering resolution of the dimers to monomers ( Aussel et al . , 2002; Grainge et al . , 2011; May et al . , 2015 ) . Without FtsK , Xer-dif synaptic complexes are formed , but do not lead to final recombination products ( Aussel et al . , 2002; Diagne et al . , 2014; Grainge et al . , 2011; Zawadzki et al . , 2013 ) . Regulation by FtsK is critical to ensure that Xer recombination takes place only in the correct spatial and temporal context – at the division septum , when genome replication has been completed – thereby ensuring faithful genome segregation . Here , we present two crystal structures of the H . pylori XerH recombinase in complex with its recombination site difH . Together with associated biochemical and in vivo data , these first Xer-DNA complex structures shed light on potential regulatory mechanisms of the recombination pathway . Remarkably , the overall shape and DNA conformation of initially formed XerH-difH synaptic complexes is considerably different from those of other tyrosine recombinases , such as Cre-loxP that has served as a model system for the family . The unanticipated conformation of the pre-cleavage synaptic complex suggests a possible model for why Xer proteins require external activation , and in comparison with the post-cleavage complex structure provides clues for how FtsK might stimulate recombination activity . Our structures provide a resource to construct models for other Xer synaptic complexes , including that of the heterotetrameric E . coli XerC/D system .
The current mechanistic model of Xer recombination conforms to the tyrosine recombinase paradigm , which is supported by extensive biochemical and structural studies on Xer and other systems ( reviewed in: [Grindley et al . , 2006; Midonet and Barre , 2014; Van Duyne , 2001] ) . This model proposes a step-wise process that starts with two Xer monomers binding to each dif site , which then interact to form a synaptic complex ( Figure 1B , i ) ( Blakely et al . , 1993 ) . Here , two Xer protomers each cleave one strand of one dif site ( Figure 1B ii ) ( Blakely et al . , 1997; Gopaul and Duyne , 1999; Guo et al . , 1997 ) , and the broken strands are exchanged and rejoined creating the HJ intermediate ( Figure 1B , iii ) ( Gopaul et al . , 1998 ) . Then the second Xer pair performs cleavage and strand exchange on the other strand pair , completing recombination ( Figure 1B , v–vi ) . Based on available synaptic complex structures of other tyrosine recombinases , it was hypothesized that the dif DNA is bent upon synapsis , creating a square planar DNA arrangement that is then maintained throughout the recombination reaction to allow energetically inexpensive exchange of DNA strands ( Gopaul et al . , 1998; Guo et al . , 1997 , 1999 ) . However , in the absence of direct structural data the exact architecture of Xer-DNA complexes has remained unknown . Previous DNA-free crystal structures of Xer recombinases ( Jo et al . , 2016; Serre et al . , 2013; Subramanya et al . , 1997 ) showed a domain arrangement that is incompatible with DNA binding . Another puzzling aspect of the mechanism concerned activation by FtsK . In early work , it was noted that in the absence of E . coli FtsK , HJ formation was catalysed by XerC , leading to the hypothesis that XerC/D-dif synaptic complexes assemble preferentially with XerC in an active conformation ( Barre et al . , 2000 ) . Subsequently , FtsK was shown to interact directly with the C-terminal domain of XerD ( Keller et al . , 2016; Yates et al . , 2003 , 2006 ) , and this interaction was hypothesized to promote reassembly of the ‘XerC-active’ synaptic complex into a conformation in which XerD is active and performs cleavage of the first DNA strand pair ( Aussel et al . , 2002; Grainge et al . , 2011; Zawadzki et al . , 2013 ) . More recently , single-molecule Fluorescence Resonance Energy Transfer ( smFRET ) studies indicated that the majority of XerC/D-dif synaptic complexes formed in the absence of FtsK are in a conformation where XerD is catalytically inactive , but ready for direct activation by FtsK without major dis- and reassembly of the complex ( Zawadzki et al . , 2013 ) . A small proportion of XerC active complexes nevertheless form and create unproductive HJs , but the pre-active XerD synaptic complexes are dominant and are the ones activated by FtsK . However , none of the previous structures or biochemical data have provided an explanation for why Xer recombinases require activation , or what the nature of the proposed conformational rearrangements might be . To begin to shed light on these issues , we determined the structure of Helicobacter pylori XerH in complex with its difH DNA substrate . difH ( Figure 2A ) was previously predicted by comparative genome analysis and consists of two XerH-binding arms separated by a 6 bp central region ( Carnoy and Roten , 2009 ) . We confirmed XerH binding to difH by DNase I footprinting ( Figure 2—figure supplement 1A ) and analytical size exclusion chromatography ( SEC ) ( Figure 2—figure supplement 1B ) . We then co-crystallized full-length wild-type XerH with a 30 bp difH DNA duplex . The resulting 2 . 1 Å resolution structure ( Figure 2B and Table 1 ) shows a synaptic complex with two difH molecules and four XerH subunits ( Figure 2B , left panel ) . Each difH site interacts with two XerH molecules , one molecule ( molecule A ) binding to the left arm and one ( molecule B ) to the right arm . The complex has overall two-fold symmetry relating the two DNA molecules and the XerH molecules bound to them ( i . e . molecules A and B to A’ and B’; Figure 2B ) . The overall fold of each protein subunit resembles that of other tyrosine recombinases , comprising two mostly helical domains connected by a 7-aa linker ( Figure 2C ) . Notably , the C-terminal helix αO protrudes from the body of each catalytic domain into a cleft on the surface of an adjacent subunit in a cyclic fashion , creating a cyclic domain-swapped arrangement similar to the ones observed in the structures of Cre ( Figure 2B , right panel ) and λ integrase synaptic complexes ( Biswas et al . , 2005; Guo et al . , 1997 ) . 10 . 7554/eLife . 19706 . 004Figure 2 . Structure of the XerH-difH complex . ( A ) Sequence of the H . pylori difH site . The sequence is written in the 3’ to 5’ direction to maintain consistency with the historical nomenclature of the arms , the structural figures , and the schematic models . The two XerH-binding arms are shown in gold and blue and the central region and the terminal base-pairs in gray . The inserted base-pair in the left arm is shown in brown . Red arrows indicate XerH cleavage sites . Binding assays confirming the site are shown in Figure 2—figure supplement 1 . ( B ) The XerH-difH synaptic complex structure ( left ) , compared to the previously solved Cre-loxP synaptic complex ( PDB: 4CRX; right ) . XerH molecules are shown in cartoon representation , colored as their bound difH arms in A ( molecule A in gold , molecule B in blue ) . Red arrows indicate the cleavage sites on the DNA . Side arrows mark the synaptic interface . The dyad symmetry axis of the tetramer runs along the midline of the synaptic interface . See Figure 2—figure supplement 2 for a snapshot of the electron density map . ( C ) Close-up of the left arm-bound XerH ( molecule A , gold ) . The N-terminal domain ( residues 1–163 ) consists of six α-helices ( αA to αF , cylinders ) ; helices αB and αC are XerH-specific . The C-terminal catalytic domain ( residues 171–362 ) is also mostly helical ( αG-αO ) , with a single β-sheet containing three antiparallel β-strands ( brown arrows ) . The interdomain linker could not be located in the structure ( dashed line ) . The catalytic tyrosine is shown in red . Figure 2—figure supplement 3 shows a comparison with the previously solved DNA-free structure of XerD . ( D ) Sequence-specific interactions of XerH ( side chains shown as sticks with atomic coloring ) and DNA . Hydrogen bonds ( <3 . 5 Å ) are shown as dashed black lines . See Figure 2—figure supplement 4 for a comprehensive overview of the protein-DNA interactions and their biochemical validation . DOI: http://dx . doi . org/10 . 7554/eLife . 19706 . 00410 . 7554/eLife . 19706 . 005Figure 2—figure supplement 1 . XerH binding to the predicted difH site . ( A ) DNase I footprinting assay mapping the XerH-protected region of difH . A difH-containing bottom strand-labeled DNA substrate was incubated with XerH and then digested with the indicated amounts of DNase I to reveal the XerH-protected region of the substrate . Sequencing ladders ( G , A , T , and C ) were run alongside the reactions . The sequence of the protected region is shown , with the two difH arms shaded gray . ( B ) Analysis of complex formation between XerH and a 33 bp difH substrate by size-exclusion chromatography ( SEC ) detected at 260/280 nm . Retention volumes indicate formation of a complex with a larger size upon combining protein and DNA . DOI: http://dx . doi . org/10 . 7554/eLife . 19706 . 00510 . 7554/eLife . 19706 . 006Figure 2—figure supplement 2 . A cross-eyed stereo image of the bias-minimized 2Fo-Fc composite omit electron density map of the pre-cleavage XerH-difH complex structure . The map ( black mesh ) shows the region of the active site of the left arm-bound subunit contoured at 1 . 2 sigma level . Protein ( gold ) and DNA ( gray ) residues are shown as sticks with atomic coloring; water molecules are shown as red spheres . DOI: http://dx . doi . org/10 . 7554/eLife . 19706 . 00610 . 7554/eLife . 19706 . 007Figure 2—figure supplement 3 . Comparison of the DNA-bound XerH ( gold ) and DNA-free XerD ( PDB: 1A0P; green ) structures . The insert shows how a rotation of the N-terminal domain ( in rainbow coloring from blue for N-terminus to red for C-terminus ) can lead to the opening of the XerD clamp in order to accommodate a DNA duplex without clashes . DOI: http://dx . doi . org/10 . 7554/eLife . 19706 . 00710 . 7554/eLife . 19706 . 008Figure 2—figure supplement 4 . Interactions between XerH and the difH site . ( A ) Schematic view of the interactions observed in the XerH-difH synaptic complex crystal structure . Amino acids of the XerH subunits bound to the left and right difH arms are shown in gold and blue , respectively . The residues of the N-terminal domain are shown with a darker color , and the catalytic residues are shown in red . Dashed lines indicate hydrogen bonds . Bases directly contacted by XerH are colored gold or blue . Interacting water molecules are shown in grey . Red stars mark the cleavage sites . The diagram is based on protein-DNA interaction analysis performed with NUCPLOT ( Luscombe et al . , 1997 ) . ( B ) Mutational analysis of XerH-difH interactions . The left-arm palindrome difH ( difHLP ) substrate is shown on the left with the left arm sequences in gold . Note that the sequence is written in the 3’ to 5’ direction as in the main figure . Arrows beneath the sequence indicate the palindromic region , and triangles mark the introduced nicks . On the right , in vitro XerH cleavage assays with various substrate variants . The DNA mutations numbered as in ( A ) are indicated above the gel . In this assay , upon cleavage XerH becomes covalently attached to the cleaved DNA strand via a phosphotyrosyl bond , which is trapped by the use of double-nicked ‘suicide’ substrates ( see schematic representation in Figure 3C ) . The XerH-DNA covalent intermediate and unmodified XerH can be separated on SDS-PAGE . DOI: http://dx . doi . org/10 . 7554/eLife . 19706 . 00810 . 7554/eLife . 19706 . 009Table 1 . X-ray diffraction data collection and refinement statistics . DOI: http://dx . doi . org/10 . 7554/eLife . 19706 . 009XerH-difH XerH-difHLP nativeXerH-difHLP SeCrystal properties Space groupP 21 21 21I 2 2 2I 2 2 2Unit cell: a , b , c ( Å ) 79 . 28 , 153 . 2 , 169 . 3986 . 38 , 115 . 22 , 235 . 285 . 79 , 115 . 73 , 235 . 29Unit cell: α , β , γ ( ° ) 90 , 90 , 9090 , 90 , 9090 , 90 , 90Data collection BeamlineI04-1 ( DLS ) ID29 ( ESRF ) ID29 ( ESRF ) Wavelength ( Å ) 0 . 928190 . 979080 . 97908Resolution range ( Å ) 48 . 89–2 . 1 ( 2 . 18–2 . 1 ) 46 . 96–2 . 4 ( 2 . 49–2 . 4 ) 47 . 04–3 . 15 ( 3 . 2–3 . 15 ) Total reflections793843372682100854Unique reflections1206594621637960Multiplicity6 . 6 ( 6 . 0 ) 8 . 1 ( 8 . 2 ) 2 . 7 ( 2 . 6 ) Completeness ( % ) 99 . 87 ( 99 . 81 ) 99 . 86 ( 99 . 50 ) 97 . 2 ( 96 . 2 ) R-meas ( % ) 11 . 28 ( 82 . 4 ) 10 . 32 ( 140 . 1 ) 11 . 9 ( 71 . 0 ) R-sym ( % ) 10 . 4 ( 75 . 4 ) 9 . 7 ( 131 . 4 ) 9 . 7 ( 58 . 2 ) I/σI12 . 53 ( 2 . 25 ) 14 . 90 ( 1 . 51 ) 11 . 17 ( 1 . 89 ) CC1/20 . 998 ( 0 . 754 ) 0 . 999 ( 0 . 648 ) 0 . 994 ( 0 . 727 ) Wilson B-factor31 . 5556 . 9759 . 75Refinement R-work0 . 19130 . 1949R-free0 . 22330 . 2203Number of non-hydrogen atoms146706984Protein residues1508769RMS ( bonds ) 0 . 0030 . 002RMS ( angles ) 0 . 530 . 55Ramachandran favored ( % ) 9997Ramachandran outliers ( % ) 00Clashscore1 . 922 . 99Average B-factor37 . 9072 . 20 Whereas the structure of individual XerH subunits is similar to that of other tyrosine recombinases , the arrangement of the subunits within the synaptic complex differs from that observed previously , forming a considerably less compact synaptic interface with a wide open central channel ( illustrated by comparison with the ‘paradigm’ Cre-loxP structure in Figure 2B ) ( Guo et al . , 1999 ) . Furthermore , the DNA molecules assume a near-straight conformation , in contrast to all previously reported tyrosine recombinase synaptic complexes . All four subunits of the XerH tetramer interact extensively with DNA . The two domains of each subunit form a tight , C-shaped clamp around one arm of the difH site ( Figure 2C ) . The N-terminal domain contacts the DNA using a four-helix bundle ( αA-αD , aa 12–88 ) and helix αF , both of which insert into the major groove ( Figure 2C ) . These protein segments contribute most of the sequence-specific interactions to the DNA bases , while the catalytic domain contacts mainly the DNA backbone ( Figure 2—figure supplement 4A ) . Many interactions of the catalytic domain involve helix αK ( aa 285–299 ) that is inserted into the major groove at the outer part of difH , narrows the groove ( 17 Å as opposed to 22 Å in typical B-DNA ) and induces a slight DNA bend ( Figure 2C ) . In addition , several contacts dispersed along the DNA backbone help to stabilize the position of the catalytic domain . XerH covers 11 bp of the left difH arm and 10 bp of the right arm ( Figure 2—figure supplement 4A ) . The inner 5 bp of each arm ( positions 7–11 and 20–24 , Figure 2A ) are recognized sequence-specifically ( Figure 2D ) , while the outer nucleotides ( positions 2–4 and 26–28 ) are mostly contacted at the phosphate backbone . The central region is only involved in sparse backbone interactions ( Figure 2—figure supplement 4A ) . To confirm the importance of the difH DNA sequence , we performed cleavage assays with ‘suicide’ difH substrates ( Figure 2—figure supplement 4B ) . These substrates contain a nick in the DNA backbone of each strand , one nucleotide downstream of the cleavage position . Upon XerH-mediated cleavage , the resulting covalent XerH-difH intermediate is trapped and can be detected by SDS-PAGE ( see also Figure 3C ) . As predicted by the structure , mutations of the inner base-pairs of the arms ( where XerH makes base-specific contacts ) resulted in abolished or greatly reduced cleavage activity , whereas mutations of the neighboring base-pairs did not affect the activity . 10 . 7554/eLife . 19706 . 010Figure 3 . Differential recognition of the two difH arms . ( A ) XerH binding to difH and derivatives containing a single difH arm ( difHL or difHR ) flanked by a random sequence , as quantified from EMSA ( see original gels in Figure 3—figure supplement 1 and numerical quantification data in Figure 3—source data 1 ) . Random DNA was used as a control . ( B ) SEC of XerH complexes with palindromic difH substrates ( left-arm palindrome difHLP , right-arm palindrome difHRP ) . XerH alone migrates as a monomer . A shift in retention is observed with difHLP , indicating stable complex formation . No corresponding shift is seen with difHRP . ( C ) The design of the in vitro cleavage assay using nicked ‘suicide’ substrates: upon cleavage by XerH , a single nucleotide diffuses away trapping the covalent phosphotyrosyl intermediate . ( D ) In vitro DNA cleavage assays , showing that the left arm of difH is required for XerH activity . The covalent protein-DNA intermediate is detected by SDS-PAGE . ( E ) Left arm is required for XerH-mediated recombination in E . coli . Intramolecular recombination rates were measured with plasmids containing two difH sites ( WT+WT ) , one difH site and one difHLP site ( WT+LP ) , two difHLP sites ( LP+LP ) , one difH site and one difHRP site ( WT+RP ) , two difHRP sites ( RP+RP ) , two difH sites with G3 and T4 mutated to CA , or two difH sites with G3 and T4 mutated to CC . G3 was mutated together with T4 , because it may interact with K290 and complement the role of T4 . XerH catalytic mutant R213K XerH is shown as a control . Bars indicate standard deviation determined from three independent experiments ( n = 3 ) . **p<0 . 05 ( Student’s test ) . Colony counts and recombination rates are tabulated and their statistical analysis is shown in Figure 3—source data 2 . ( F ) XerH recognizes the left ( gold ) and right ( blue ) difH arms differently . Superimposition of two adjacent XerH subunits shows differences in the DNA bending , in the positions of the C-terminal domains , and in the protruding helices αO . Red insert: Interaction of K290 and T4 ( dashed lines: hydrogen bonds ) at the left arm ( left panel ) is absent at the right arm ( right panel ) . Black insert: Interactions with the three outermost base-pairs of the left ( top ) and the right ( bottom ) difH arms are remarkably similar despite the shifted DNA sequence . Functional characterization of these interactions and the role of the specific features of the left difH arm are shown in Figure 3—figure supplement 2 . ( G ) Active site conformations at the left ( gold ) and right ( blue ) difH arm . Catalytic residues ( sticks ) are incompletely assembled around the scissile phosphates ( orange spheres ) . The red sphere denotes a bound water molecule . DOI: http://dx . doi . org/10 . 7554/eLife . 19706 . 01010 . 7554/eLife . 19706 . 011Figure 3—source data 1 . Quantification of XerH binding to difH variants based on EMSA experiments . The original gels are shown in Figure 3—figure supplement 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 19706 . 01110 . 7554/eLife . 19706 . 012Figure 3—source data 2 . Results of the in vivo recombination assays . Colony counts and recombination rates are tabulated and their statistical analysis by Student’s t-test is shown . DOI: http://dx . doi . org/10 . 7554/eLife . 19706 . 01210 . 7554/eLife . 19706 . 013Figure 3—figure supplement 1 . Electrophoretic mobility shift assay ( EMSA ) gels showing XerH binding to difH and derivatives . EMSA was performed with 50 bp substrates containing wild-type difH ( A ) , left arm of difH accompanied by a random sequence ( B ) , right arm of difH preceded by a random sequence ( C ) , or a fully randomized sequence with matched GC content ( D ) . The resulting bands correspond to free DNA , DNA bound to one XerH monomer ( gray oval ) , or DNA bound to two XerH monomers , as indicated by the schematics on the right hand side of the gels . These complex stoichiometries are derived from analytical SEC and supported by EMSA with half-site difH DNA ( data not shown ) . DOI: http://dx . doi . org/10 . 7554/eLife . 19706 . 01310 . 7554/eLife . 19706 . 014Figure 3—figure supplement 2 . Functional characterization of the role of the specific features of the left difH arm . ( A ) Cleavage assays with difHLP derivatives modified at the inserted base-pair ( positions 6/25’ , brown ) . The figure illustrates the unmodified substrate ( written in the 3’ to 5’ direction ) , nicked at the two positions marked by black wedges . Upon cleavage , XerH becomes covalently attached to DNA via a phosphotyrosyl bond , which is separated from unmodified XerH on SDS-PAGE ( see also Figure 3C ) . ( B ) Mutations of the inserted base in the left arm do not affect XerH-mediated recombination in E . coli . galK reporter plasmids contained two wild-type difH sites ( A–WT ) or two difH sites with the left arm mutated at position 6/25’ ( A6 mutated to G , C , or T as indicated ) . Bars indicate standard deviation ( n = 3 ) . None of the mutations showed significant reduction of recombination ( p>0 . 05; Student's test ) . Exact colony counts , recombination rates , and their statistical analysis are shown in Figure 3—source data 2 . ( C ) K290S mutation does not affect difH binding ( WT: wild-type XerH control ) . Binding curves were quantified from EMSA ( see numerical data in Figure 3—source data 1 ) . ( D ) K290S mutation reduces XerH recombination in E . coli by 50% ( WT: control assay with wild-type XerH ) . Reporter plasmids contained two wild-type difH sites . Statistical analysis is as in ( B ) . **p<0 . 05 ( Student’s test ) . Colony counts , recombination rates , and statistical analysis are included in Figure 3—source data 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 19706 . 014 In contrast to the previously described DNA-free structure of XerD ( Subramanya et al . , 1997 ) , the N-terminal and catalytic domains of XerH are positioned such that the cleft formed between their inner surfaces readily accommodates the difH DNA in a conformation that resembles DNA-bound structures of other tyrosine recombinases ( Aihara et al . , 2003; Chen et al . , 2000; Guo et al . , 1997 ) . A simple rotation of the N-terminal domain pivoting at the interdomain linker is sufficient to transition from the ‘closed’ XerD conformation to the ‘open’ DNA-bound XerH conformation ( Figure 2—figure supplement 3 ) , consistent with previous proposals that Xer recombinases may undergo a major conformational change upon DNA binding ( Jo et al . , 2016; Serre et al . , 2013; Subramanya et al . , 1997 ) . One of the hallmarks of tyrosine recombination is that the DNA strands are exchanged step-wise in a strictly ordered manner . In several studied examples , this ordering is believed to be linked to asymmetry in the DNA sequences , but the exact mechanisms vary ( Blakely et al . , 1993; Ennifar et al . , 2003; Nolivos et al . , 2010 ) . To determine how the XerH-difH system achieves ordering of the recombination steps , we first tested the roles of the two distinct protein-binding arms of difH ( Figure 2A ) in XerH binding and activity . DNA binding assays ( EMSA ) with substrates containing the sequences of one or both arms ( Figure 3A and Figure 3—figure supplement 1 ) showed that two XerH monomers bind cooperatively ( see especially Figure 3—figure supplement 1A ) . The affinity of a difHL substrate containing only the left arm sequence , with a random sequence in place of the right arm , was higher ( dissociation constant , KD = 180 nM; similar to wild-type difH KD = 150 nM ) than the affinity of a similar right arm-containing substrate difHR ( KD = 230 nM ) . Similarly , a palindromic substrate containing two difH left arms ( difHLP ) made a stable complex with XerH ( observed by SEC ) , whereas the equivalent right arm substrate ( difHRP ) did not ( Figure 3B ) . DNA cleavage assays also revealed different efficiencies at the two arms: no cleavage product was detected with difHR or difHRP , whereas difHL and difHLP substrates were cleaved as efficiently as the native difH site ( Figure 3C and D ) . An assay that assesses the efficiency of intramolecular recombination between two plasmid-borne sites in E . coli also indicated differential activity at the two difH arms in vivo: the wild-type recombination rate ( 9 . 9% recombination-positive colonies ) increased when both difH sites were replaced with difHLP ( 29 . 4%; p=0 . 0016 , t-test ) , but decreased to barely detectable levels with difHRP ( 1 . 3%; p=0 . 0021 , t-test ) ( Figure 3E ) . Our observation that XerH binds and preferentially cleaves the left arm of difH led us to ask how the arms are distinguished . Remarkably , the sequences of the two arms are identical apart from a single base-pair insertion in the left arm ( Figure 2A ) . In the crystal structure , the respective XerH subunits ( A and B ) make very similar interactions with the two difH arms ( Figure 2—figure supplement 4A ) . The inserted base-pair ( A6/T25’ , brown in Figure 2A ) is not recognized directly . Instead , the main difference in the recognition of the two arms involves a conserved thymine base ( T4/T5' ) in the outer parts of the arms: In the left arm , T4 forms hydrophobic contacts with XerH αK and makes a specific hydrogen bond with lysine K290 ( Figure 3F , red insert ) , whereas in the right arm T5' is closer to the center of difH and cannot make these interactions . Surprisingly , the three outermost base-pairs of each difH arm make very similar backbone interactions with their respective XerH subunits , despite the fact that they are shifted in space due to the insertion in the left arm ( altering the ‘helical phase’ of the sequence by ~35°; Figure 3F , black insert; Figure 2—figure supplement 4A ) . This is possible because the conformation of the DNA is asymmetric – the left arm is more bent than the right arm – and the catalytic domains of the protein subunits are rotated by ~10° relative to each other ( Figure 3F ) . Together , the observed differences result in stronger interaction of XerH with the left difH arm ( ΔG = −21 . 3 kcal/mol , estimated by PISA [Krissinel and Henrick , 2007] ) than with the right arm ( ΔG = −15 . 4 kcal/mol ) . A further difference between the XerH subunits bound to the left and right arms of difH concerns the C-terminal helices . While the αN-αO segment of molecule A is fully ordered , the linker and αO are partially disordered in molecule B . Also , helices αO point in different directions ( ~100° rotation; Figure 3F ) . In agreement with the structure , difH variants containing mutations at the inserted base-pair showed no or only moderate decreases in cleavage in vitro ( Figure 3—figure supplement 2A ) and recombination in vivo ( Figure 3—figure supplement 2B ) . In contrast , substitutions of T4 practically abolished recombination in E . coli ( Figure 3E ) and in H . pylori ( Debowski et al . , 2012a ) ; and mutation of K290 ( K290S ) decreased recombination activity by about half ( Figure 3—figure supplement 2D ) . Insertion of an additional base-pair next to A6/T25’ in the left arm , further shifting the positions of the three outermost base-pairs , also abolished cleavage completely ( Figure 3—figure supplement 2A ) . Like the arms , the central region of difH is also asymmetric , with different nucleotides flanking the left and right cleavage sites ( positions C13 and G13’ ) . However , XerH does not contact these nucleotides in our structure , and cleavage assays with difH variants carrying mutations at these positions revealed no reduction in activity ( Figure 2—figure supplement 4B ) , suggesting that the identity of these nucleotides does not contribute to XerH binding and cleavage asymmetry . Together , these data demonstrate that XerH binds preferentially to the left arm of difH thanks to favorable interactions with its outer sequence ( including T4 ) . Asymmetric interactions with the two arms of difH also appear to dictate distinct protein conformations in the synaptic complex , including differential positioning of the αN-αO segment that carries the nucleophilic tyrosine , perhaps helping to define which arm is cleaved first . Perhaps the most unexpected feature of the XerH-difH synaptic complex structure is the DNA conformation , which is nearly straight . Both DNA molecules in the synapse resemble B-form DNA , with a wide angle ( 165° ) between the two difH arms ( Figure 2B , left panel ) . This is in sharp contrast to other currently available DNA-bound structures of tyrosine recombinases , all of which contain strongly bent DNA . For example , the loxP DNA is bent to ~100° in the analogous pre-cleavage structures of Cre-loxP complexes ( Figure 2B , right panel ) . This bent DNA conformation is maintained throughout the recombination reaction , allowing easy transition between the reaction intermediates ( Ennifar et al . , 2003; Gopaul et al . , 1998; Guo et al . , 1999 ) ( see also Figure 1B ) . In the active site of each XerH subunit , several conserved residues including the catalytic tyrosine Y344 , two arginine ( R213 and R312 ) , and two histidine ( H309 and H335 ) residues are assembled around the scissile phosphate , together forming a catalytic pocket characteristic of tyrosine recombinases . The active sites of the subunits are very similar , except for the R213 side chain which points in different directions in molecules A and B ( Figure 3G ) . Electron density for another essential catalytic residue , K239 , could not be observed in either of the subunits , so we presume that it is disordered . The catalytic tyrosine Y344 is far away ( 5 . 2–5 . 8 Å ) from the scissile phosphate in all subunits , so the structure represents a catalytically inactive synaptic complex , implying a conformational change is required prior to catalysis . To capture XerH in a post-cleavage synaptic complex , we used ‘suicide’ difH substrates ( containing a nick in each DNA strand , as in the in vitro cleavage assays described above , Figure 3C ) . Attempts to co-crystallize XerH with suicide versions of the native difH sequence were unsuccessful , but palindromic difHLP suicide substrates gave us crystals that diffracted to 2 . 4 Å ( Table 1 ) . The resulting structure ( Figure 4A ) differs considerably from the pre-cleavage structure . Each difHLP site interacts with two XerH molecules in a tetrameric synaptic complex , but in this structure , the two halves of the tetramer are related by crystallographic two-fold symmetry ( relating molecule A to A’ and B to B’; Figure 4A ) . While the structures of the individual XerH subunits and their interactions with the respective difH DNA arms in this structure and in the pre-cleavage structure are similar , the overall arrangement of the complex is different . Most strikingly , the difH DNA is now strongly bent ( Figure 4B , Video 1 ) . The bend mainly originates from a single distortion within the central region of difH: bases A17’ and G18’ are un-stacked with a 90° tilt , which introduces a kink resulting in asymmetric bending of the DNA ( insert in Figure 4B ) . The total angle of 120° between the difH arms is only slightly wider than the angles observed in previous structures of various tyrosine recombinases ( e . g . 109° in the post-cleavage structure of λ integrase ( Biswas et al . , 2005 ) and 100° in Cre-loxP ( Guo et al . , 1997 ) ; see also Figure 2B ) . Notably , the kink in difH is at an equivalent position to the kink in loxP ( Ennifar et al . , 2003 ) . However , whereas in Cre the nucleotides involved in the kink interact with three arginines within a tight pocket ( Figure 4C ) ( Ennifar et al . , 2003; Guo et al . , 1997 ) , XerH forms only a single interaction with these nucleotides ( R129-A17’; Figure 4C; Figure 2—figure supplement 4A ) and encircles the kink less tightly . Notably , the nucleotides of the central region of difH remain fully base-paired after cleavage , unlike the analogous structure of Cre-loxP , where several of these are unpaired ( Guo et al . , 1997 ) . The extensive Cre contacts can thus actively bend loxP . In contrast , weaker XerH-difH interactions appear to be insufficient to promote sharp DNA bending , but nicks in both DNA strands of the difH substrate facilitate bending . When we tried to re-build the DNA strands in the post-cleavage structure to model a bent but un-nicked substrate , we observed a clash with the protein near the tyrosine nucleophile-bearing helix αN , suggesting that the nicks might favor the XerH conformation adopted on the bent sites . 10 . 7554/eLife . 19706 . 015Figure 4 . The post-cleavage XerH-difHLP synaptic complex structure . ( A ) Overall view of the XerH-difHLP structure in cartoon representation . Sequence of the difHLP substrate is shown above ( written in the 3’ to 5’ direction , with the arms in gold , central region in gray ) the nick positions are marked by triangles . Figure 4—figure supplement 1 shows a snapshot of the electron density map . ( B ) Compared to the pre-cleavage XerH-difH complex , both subunits ( A , gold; B , blue surface ) have rotated ~22° towards each other , concomitant with DNA bending . Insert: Close-up of DNA bending: A17’ and G18’ ( red ) are unstacked , with a 90° kink . ( C ) XerH ( left ) interacts with the DNA kink differently than Cre ( PDB: 1NZB; right ) . The electrostatic surface potential is shown in red ( negative ) and blue ( positive ) . ( D ) Superimposition of the two subunits ( A , gold and B , blue ) bound to the same difHLP site illustrates their different conformations . Left insert: Helices αN and αO are repositioned relative to the pre-cleavage structure ( grey ) , including rotations ( marked by arrows ) and shortening of the helices ( red dashed lines showing the distances between Cα of L338 and Cα of W359 ) . Right insert: repositioning of the β2-β3 loop upon activation enables the catalytic K239 to interact with the DNA . For clarity , only the DNA associated with the golden monomer is shown . ( E ) Active site conformations of the distinct XerH subunits . In the active subunit ( molecule A , left ) , R213 , H309 , R312 , and H335 make hydrogen bonds with the scissile phosphate ( orange ) , K239 contacts the base of the adjacent nucleotide A12 , and Y344 is covalently attached to the DNA . In the inactive subunit ( right ) the catalytic tyrosine is 5 . 5 Å away from the scissile phosphate , R213 and H335 point away , and the K239 side chain is disordered . Red sphere – bound water molecule; dashed lines – hydrogen bonds . DOI: http://dx . doi . org/10 . 7554/eLife . 19706 . 01510 . 7554/eLife . 19706 . 016Figure 4—figure supplement 1 . Cross-eyed stereo image of the bias-minimized 2Fo-Fc composite omit electron density map of the post-cleavage XerH-difH complex structure . The map ( black mesh ) shows the region of the active site of the active subunit contoured at 1 . 2 sigma level . Protein ( gold ) and DNA ( gray ) residues are shown as sticks with atomic coloring; water molecules are shown as red spheres . DOI: http://dx . doi . org/10 . 7554/eLife . 19706 . 01610 . 7554/eLife . 19706 . 017Video 1 . Activation of the XerH-difH synaptic complex . Morphing of the XerH-difH synaptic complex in pre-cleavage conformation into the post-cleavage conformation . Activation involves major conformational rearrangement of the synaptic complex , involving rotation of all XerH subunits and sharp DNA bending . DOI: http://dx . doi . org/10 . 7554/eLife . 19706 . 017 The arrangement of the XerH subunits is also markedly different in the two structures . Each subunit is rotated by ~22° in the post-cleavage structure ( Figure 4B , Video 1 ) , resulting in major rewiring of the inter-subunit interactions and an overall compaction of the complex . Also , the synaptic interface is considerably larger ( 3926 Å2 , calculated by PISA [Krissinel and Henrick , 2007] ) than in the pre-cleavage structure ( 3040 Å2 ) . The previously disordered parts of the αO helices of molecule B are now fully visible , ordered , and more compact ( F338 - W359 distance reduced from 43 Å to 34 Å; Figure 4D , left insert ) . At the same time , helices αN and αO of molecule A are rotated by ~30° , bringing the catalytic Y344 to a suitable position to attack one strand on each difHLP DNA forming a covalent phosphotyrosyl bond ( Figure 4E , left ) . Interestingly , despite the symmetry of the palindromic difHLP substrate , the two bound XerH molecules ( A and B ) are in different conformations ( Figure 4D ) . Their overall structures are similar ( r . m . s . deviation 0 . 78 Å for Cα carbons ) , but important protein segments that contain catalytic residues ( the β2-β3 turn with K239 , and αN-αO carrying Y344 ) are in different conformations ( Figure 4D ) . Consequently , subunit A assumes a fully active conformation , while subunit B is in an inactive conformation with the catalytic tyrosine distant from the scissile phosphate ( Figure 4E ) . This simultaneous presence of both active and inactive XerH subunit conformations within the synaptic complex is consistent with the ‘half-of-the-sites reactivity’ mechanism observed for other tyrosine recombinases ( Figure 1B ) . The marked difference between the XerH subunits bound to the two difHLP arms ( comparable to the asymmetry seen in the pre-cleavage complex with native difH ) is particularly striking considering that the artificially introduced symmetry of the substrate could be expected to mask some asymmetric features of the synaptic complex . The asymmetry is presumably an intrinsic property of the system , essential to the mechanism of synapsis and catalysis . Due to high structural conservation within the Xer family ( Subramanya et al . , 1997 ) ( Figure 5—figure supplement 1A ) , our XerH structures can provide insights into the mechanisms of other Xer recombinases that have eluded structural studies so far . We have used our structures to model the heterotetrameric synaptic complex of E . coli XerC/D in both pre- and post-cleavage states ( Figure 5A ) . We modeled DNA-bound XerC and XerD using the DNA-free XerD structure ( PDB: 1A0P; ( Subramanya et al . , 1997 ) ) and a homology model for XerC . These were then superimposed onto the XerH-difH structures to assemble heterotetramers . Our models place XerC on the right arm of difH and XerD on the left arm . We cannot exclude the alternative assignment , but our observation that XerH preferentially cleaves at the left arm of difH first ( Figure 3D ) supports the hypothesis that the XerH subunit on the left arm corresponds to XerD whereas the subunit on the right arm corresponds to XerC . 10 . 7554/eLife . 19706 . 018Figure 5 . XerC/D-dif synaptic complex models reveal common features of Xer recombination . ( A ) Cartoon representation of heterotetrameric XerC/D-dif synaptic complexes modeled based on the pre-cleavage ( left ) or the post-cleavage ( right ) XerH-difH complexes . XerD ( green ) and XerC ( yellow ) monomers are arranged compatibly in the tetramers . Figure 5—figure supplement 1 shows the structure-based sequence alignment and a comparison of protein-DNA interactions . ( B ) Superimposition of XerD and XerC monomers in the pre-cleavage structure model . ( C ) Conserved active site residues of XerD and XerC ( sticks with atomic coloring ) are assembled around the scissile phosphate ( orange ) in the pre-cleavage XerC/D-dif synaptic complex . DOI: http://dx . doi . org/10 . 7554/eLife . 19706 . 01810 . 7554/eLife . 19706 . 019Figure 5—figure supplement 1 . Modeling of XerC/D-dif synaptic complexes . ( A ) Secondary structure-based alignment of XerH , XerD , and XerC . α-helices and β strands are marked above , conserved catalytic residues are in red . ( B ) Schematic view of the interactions of XerH , XerD , and XerC with the outer base-pairs of their respective binding sites . Dashes indicate hydrogen bonds . DOI: http://dx . doi . org/10 . 7554/eLife . 19706 . 019 In the models , the XerC and XerD monomers each form C-shaped clamps around one arm of the dif site ( Figure 5B ) , and make multiple sequence-specific and backbone interactions with the DNA . The overall fold and conformation of the two proteins are very similar; only their C-terminal helices assume different relative orientations , as is also seen in the XerH structures . The five conserved catalytic residues ( RHRHY ) assemble around the catalytic pocket in all XerC and XerD subunits ( Figure 5C ) . In the post-cleavage complex , the XerD subunits are in an active conformation , with their nucleophilic tyrosines approaching the scissile phosphates , whereas the XerC subunits are in an inactive conformation . As in the XerH structures , the three outer base-pairs of each dif arm are contacted extensively by the proteins ( Figure 5—figure supplement 1B ) , and the bases corresponding to T4 in difH interact with lysine K222 of XerC or histidine H225 of XerD , both counterparts of XerH K290 ( Figure 5—figure supplement 1B ) . These interactions might contribute to the arm specificity of XerC and XerD , as well as to ordering recombination , as we inferred for XerH . The interactions involved in DNA kinking in Cre are absent in the XerC/D-dif complex , suggesting that these enzymes , like XerH , might be unable to independently initiate sharp DNA bending . The protein arrangements and interfaces observed in our two XerH-difH structures are well accommodated by XerC and XerD , the pre-cleavage model showing nearly straight DNA and the post-cleavage model containing bent DNA and tightened intersubunit interactions ( Figure 5A ) . This suggests that the XerC/D and XerH complexes might undergo similar conformational rearrangements upon activation . In agreement with this idea , two previous smFRET studies ( Diagne et al . , 2014; Zawadzki et al . , 2013 ) reported that pre-formed XerC/D-dif synaptic complexes undergo conformational change upon activation by FtsK , leading to an increase of the distance between the two dif termini from about 53 to 67 Å across the synapse ( as predicted from a change in FRET efficiencies; [Zawadzki et al . , 2013] ) . Our XerC/D-dif models predict a similar change from 46 Å in the pre-cleavage state to 60 Å in the post-cleavage state .
In this work , we investigated the structural and mechanistic bases of Xer site-specific recombination . Using the homomeric XerH-difH system from H . pylori as a model system , we report the first high-resolution crystal structures of Xer-dif synaptic complexes . These structures demonstrate that Xer proteins follow the established chemical pathway of tyrosine recombinases and reveal how the reaction steps can be choreographed by small differences in the arms of the difH recombination sites . Together with associated biochemical data , the structures also show that XerH-difH synaptic complexes initially assemble in an inactive state with straight DNA , which must undergo a major conformational change for catalytic activation , as previously observed in single molecule fluorescence studies of XerC/D-dif recombination ( Diagne et al . , 2014; Zawadzki et al . , 2013 ) . Our post-cleavage XerH-difH synaptic complex structure elucidates the structural nature of this conformational change , which involves major rearrangement of the protein-protein interfaces and DNA bending ( Video 1 ) . With molecular modeling , we extend our structural and mechanistic findings to the prototypical E . coli XerC/D-dif system , demonstrating that our structures provide a resource for understanding the mechanism of other Xer recombinases . The thoroughly characterized Cre site-specific recombination reaction has long been considered a paradigm for tyrosine recombinase-based DNA rearrangements ( Gopaul and Duyne , 1999; Grindley et al . , 2006; Guo et al . , 1997; Van Duyne , 2015 ) . Unusually , Xer recombinases require an external factor ( generally the cell division protein FtsK ) for catalytic activation . This feature is essential for Xer’s chromosome maintenance function , but does not have any analogy in the Cre system . Therefore , the structural basis of Xer recombination and its activation have remained debated . From our structural , biochemical , and microbiological data on H . pylori XerH-difH recombination we can now propose a mechanistic model for XerH recombination and the regulatory function of FtsK , as follows ( Figure 6 ) :10 . 7554/eLife . 19706 . 020Figure 6 . Model for XerH recombination activation . ( A–C ) Differential binding affinities of XerH to the left and right arms of difH DNA trigger an asymmetric DNA conformation and arrange the two protein subunits in distinct conformations ( A ) . Two difH sites are brought together in a tetrameric synaptic complex ( B ) stabilized by a cyclic arrangement of the protruding αO helices . Initially formed XerH-difH complexes are inactive ( indicated by gray nucleophilic tyrosine ) as seen in our pre-cleavage structure . Figure 6—figure supplement 1 shows that XerH alone is unable to cleave an intact DNA substrate . Catalytic activation then involves DNA bending and a major rearrangement of the protein subunits ( as in our post-cleavage structure ) , presumably enabled by a direct interaction with FtsK ( C ) . As a result , helices αO become ordered across the synapse , while helices αN and αO along one difH molecule rotate , bringing the nucleophilic Y344 into the active conformation ( yellow ) that enables DNA cleavage . Note that the DNA substrates are drawn with the strand going 3’ to 5’ on the top as in Figure 1B . DOI: http://dx . doi . org/10 . 7554/eLife . 19706 . 02010 . 7554/eLife . 19706 . 021Figure 6—figure supplement 1 . XerH activity on intact and nicked difH substrates . XerH is unable to cleave intact difH substrates , as assessed by DNA cleavage assays on SDS-PAGE . Wild-type difH with or without ‘suicide’ nicks was used as indicated . ( - ) denotes no-DNA control . DOI: http://dx . doi . org/10 . 7554/eLife . 19706 . 021 The recombination process starts with XerH binding to the difH site on the H . pylori chromosome ( Figure 6A ) . Two XerH subunits bind cooperatively to one difH site ( Figure 3—figure supplement 1; Figure 6A ) , each forming a clamp around one arm of difH ( Figure 2C ) . The difH left arm has higher affinity for XerH than the right arm ( Figure 3A , B ) , leading to the second XerH molecule binding to the right arm more often than vice versa . Cooperative binding of XerH involves extensive contacts between the two subunits bound to each difH , as revealed by our structures ( 1650 . 6 Å2 surface area with ΔG = −12 . 6 kcal/mol and 1515 . 5 Å2 with ΔG = −14 . 5 kcal/mol between molecules A-B and A’-B’ respectively in the pre-cleavage structure; and 1379 . 8 Å2 surface area with ΔG of −13 . 8 kcal/mol in the post-cleavage structure ) . Despite the near-perfect dyad symmetry of the difH arms , the bound XerH subunits are structurally distinct . The left arm subunit assumes a conformation primed for DNA cleavage , whereas the right arm subunit is forced in an inactive state . This asymmetric configuration defines the order of all subsequent cleavage and strand transfer events . Two XerH-bound difH sites then interact to create a tetrameric synaptic complex , as we see in our crystal structures ( Figure 2B; Figure 6B ) . The sites align in antiparallel , and intersubunit interactions across the synaptic interface anchor the αO helices of the right arm-bound XerH subunits on their left arm-bound partner , creating a ‘circular’ domain-swapped arrangement , similar to the ones described for other tyrosine recombinases such as Cre and λ integrase ( Biswas et al . , 2005; Guo et al . , 1997 ) . The pre-cleavage synaptic complex crystal structure contains nearly straight difH DNA ( Figure 2B; Figure 6B ) , and XerH is in a catalytically inactive conformation ( Figure 3G ) . Correspondingly , we were unable to observe difH recombination or XerH-mediated cleavage of intact difH DNA substrates in vitro ( Figure 6—figure supplement 1 , and data not shown ) . We cannot exclude the possibility that XerH does still catalyze transient strand cleavage of these substrates , followed by rapid efficient re-ligation making cleavage undetectable , but there is no evidence to support this idea . In contrast , nicked difH suicide substrates are cleaved efficiently , and our crystal structure in which XerH-mediated cleavage of nicked difH has occurred showed a sharply bent DNA conformation , consistent with a direct link between DNA bending and cleavage activity . The crystals giving our two structures were grown under different conditions ( see Materials and methods ) , but XerH was active ( on nicked difH substrates ) in both conditions ( data not shown ) , supporting our view that both structures are biologically relevant . We hypothesize that our in vitro systems lack a stimulatory factor that is required for normal recombination between intact difH sites . Septum-borne FtsK was shown to be required for XerH recombination in H . pylori ( Debowski et al . , 2012a ) , so we propose that this is the missing factor for XerH activation in vitro . FtsK-promoted rearrangement of the XerH-difH synaptic complex might bring the catalytic tyrosines of the left arm-bound subunits into their active positions close to the scissile phosphates and sharply bend the DNA as seen in our structures ( Figures 4B and 6C; Video 1 ) . Another possibility is that FtsK mediates formation of a different synaptic complex with the right arm-bound subunits activated for cleavage . However , this would be inconsistent with smFRET studies of XerC/D complexes showing that FtsK activates pre-formed synaptic complexes without major remodeling ( Zawadzki et al . , 2013 ) . Analogy with the XerC/D system suggests that FtsK might directly interact with the C-terminal ~20 amino acids of XerH ( Yates et al . , 2006 ) or the back of its C-terminal domain ( Keller et al . , 2016 ) . However , the role of FtsK in the H . pylori Xer system still requires further substantial investigation , and it remains possible that the bent-DNA configuration required for DNA cleavage can be reached from the pre-cleavage structure without the intervention of FtsK . Following activation , XerH-catalysed DNA strand cleavage and rejoining presumably follow the conserved tyrosine recombination pathway ( Figure 1B ) , with the XerH subunits bound to the left arms performing the first strand exchanges . The resulting HJ is then resolved by a reciprocal set of chemical reactions catalysed by the XerH subunits bound to the right arms . DNA bending is a prerequisite for activity of many molecular machines , including transcription factors , topoisomerases and recombinases ( Dong and Berger , 2007; Kim et al . , 1993; Lee et al . , 2013; Yang and Steitz , 1995 ) . In most of these cases , DNA bending and the downstream function are performed by the same protein or complex , or DNA bending depends on ubiquitously available factors such as IHF . For tyrosine recombinases , the paradigm suggests that sharp DNA bending that is required for DNA cleavage and energetically inexpensive strand exchange , is introduced by the recombinase itself concomitant with DNA binding and synapsis . Here , we show that this is not the case for XerH , which binds and assembles its recombination substrates in an almost straight conformation . The required DNA bending then presumably occurs upon activation by an external factor that is present only in a particular spatial and temporal context , providing an elegant regulatory mechanism to ensure faithful chromosome segregation . Our modeling of XerC/D-dif indicates that the architectures of the pre- and post- cleavage synaptic complexes are probably conserved between XerC/D and XerH ( Figure 5B ) . The DNA conformational changes that we observe for XerH-difH are also consistent with smFRET studies of XerC/D-dif synaptic complexes ( Diagne et al . , 2014; Zawadzki et al . , 2013 ) , which imply a conformational change upon FtsK-mediated activation consistent with the differences between our models of the pre-cleavage and post-cleavage complexes . Thus , we expect that the mechanism of catalytic activation proposed here for XerH – preassembly of a synaptic complex with almost straight DNA , followed by FtsK-induced protein and DNA rearrangements activating the complex for catalysis – may be conserved in the XerC/D system . In summary , our structures of XerH-difH complexes demonstrate that catalysis by the Xer family of site-specific recombination systems follows the established tyrosine recombinase pathway , with the reaction steps being choreographed by small differences in the arms of the dif recombination sites . Contrary to previous assumption , bending of the dif sites does not occur concomitantly with synaptic complex assembly but at a post-synaptic step , when the accessory factor FtsK might be needed to license the reaction by promoting a conformational change . We also provide structural insight into the conformational change required for activation ( Video 1 ) , and extend our findings to other Xer recombinases through modeling . Our structural insights help us to better understand how Xer proteins function and how they have adapted the tyrosine recombinase machinery for their unique genome maintenance function . In the long term , our data can also help to improve XerH-based genetic engineering tools that have been recently introduced for markerless gene deletions in H . pylori ( Debowski et al . , 2012b ) .
Full-length xerH from H . pylori strain 26995 ( NCBI: HP0675 ) was synthesized with codon-optimization ( MrGene , Regensburg , Germany ) for over-expression in E . coli and cloned into vector pETM-28 ( PepCore , EMBL ) using BamHI/XhoI restriction sites . XerH Y344F , R213K , S161A , and K290S constructs were prepared by site-directed mutagenesis using primers listed in Supplementary file 1 . H . pylori XerH and its mutants were overexpressed in E . coli BL21 ( DE3 ) as N-terminal fusions with hexahistidine and Small Ubiquitin-like Modifier ( SUMO ) tags . The constructs were expressed in E . coli BL21 ( DE3 ) for 16 h at 15°C after addition of 0 . 5 mM isopropyl β-D-1-thiogalactopyranoside ( IPTG ) . Cells were lysed by sonication in purification buffer ( 1x phosphate buffered saline ( PBS ) , 1 M NaCl , 5% glycerol , and 0 . 2 mM tris ( 2-carboxyethyl ) phosphine ( TCEP ) , pH 7 . 5 ) supplemented with protease inhibitors ( cOmplete Protease Inhibitor Cocktail , Roche Diagnostics , Mannheim , Germany; 1 . 5 mM phenylmethanesulfonylfluoride [PMSF] ) . The lysate was cleared by centrifugation at 40 , 000 g . The protein was purified from the soluble fraction by Ni-affinity chromatography on a HisTrap column ( GE Healthcare , Munich , Germany ) , followed by tag cleavage with SenP2 protease , tag removal on a Ni-affinity column , and size exclusion chromatography on Superdex 200 column ( GE Heathcare ) . The seleno-methionine derivative of XerH was expressed in BL21 ( DE3 ) in M9 growth medium supplemented with the essential amino acids , with seleno-methionine replacing methionine , and was purified as above . The XerH-difH complexes were formed in activity buffer ( 25 mM HEPES , pH 7 . 5 , 100 mM NaCl , 10 mM MgCl2 , 5% glycerol , 1 mM DTT , and 1 mM EDTA ) . 120 bp substrates containing the left or right TnPZ end were PCR-amplified from H . pylori 26695 genomic DNA using 5’-32P-labelled primers shown in Supplementary file 1 . Samples were incubated with various amounts of DNase I for 1 min after addition of 5 mM CaCl2 and 10 mM MgCl2 . The reactions were stopped with 120 mM NaCl , 18 mM EDTA , and 0 . 6% SDS , DNA was purified by ethanol precipitation and analyzed by PAGE on Urea-TBE 12% gel . Sequencing ladders were prepared with DNA Cycle Sequencing Kit ( Jena Bioscience , Jena , Germany ) using 5’-32P-labeled primers . DNA oligonucleotides were synthesized and HPLC-purified by IDT ( Leuven , Belgium ) , then annealed in TE buffer by heating to 98°C and slow cooling to room temperature or 4°C . XerH-difH complexes were formed by mixing XerH with DNA ( sequences shown in Supplementary file 1 ) at a 1 . 2:1 molar ratio in purification buffer , dialyzed in three steps to crystallization buffer containing 25 mM sodium acetate buffer ( pH 5 . 5 ) , 200 mM NaCl , 5% glycerol , and 1 mM DTT , and concentrated to 5 mg/ml . The crystallization conditions were screened using The Classics Suite ( QIAGEN , Hilden , Germany ) , Index ( Hampton Research , Aliso Viejo , CA , USA ) , and JCSG+ ( Page et al . , 2003 ) sparse matrix crystallization screens by sitting drop vapor diffusion . The resulting hits were scaled-up and optimized in hanging-drop vapor diffusion setup . 2 μl of complex solution were mixed with 2 μl of well solution and equilibrated against 0 . 5 ml well solution for two weeks at 6°C . The pre-cleavage complex was crystallized using well solutions containing 0 . 2 M NaCl and 20–25% ( w/v ) polyethylene glycol ( PEG ) 3350 , while for the post-cleavage complex 0 . 1 M HEPES , pH 7–8 , 0 . 2 M MgCl2 , and 25–35% ( v/v ) PEG 400 was used . For data collection , the crystals were harvested and flash-frozen in liquid nitrogen . The pre-cleavage complex crystals were cryoprotected by the addition of 20% v/v glycerol . The data were collected on beamline ID-29 at the European Synchrotron Radiation Facility ( ESRF ) and on beamline I04-1 at the Diamond Light Source ( DLS ) . The diffraction data were processed with XDS ( Kabsch , 2010 ) . The structure of the post-cleavage complex was determined by SAD phasing in AutoSHARP ( Vonrhein et al . , 2007 ) , using anomalous data from a seleno-methionine derivative crystal . The phases were extended onto a native dataset , and the initial model was built in AutoBuild in Phenix ( Adams et al . , 2010; Terwilliger et al . , 2008 ) . The asymmetric unit contained two XerH molecules and one nicked difHLP DNA , with two-fold crystallographic symmetry generating a tetramer . The structure of the pre-cleavage complex was solved by molecular replacement in Phaser ( McCoy et al . , 2007 ) , using an XerH monomer as a search model . This structure showed four XerH molecules and two difH DNA per asymmetric unit . The final models were obtained by alternating model building in Coot ( Emsley et al . , 2010 ) and simulated annealing , restrained positional and B-factor refinement with Phenix ( Afonine et al . , 2012 ) . The data collection and refinement statistics are given in Table 1 . Protein interfaces were analyzed in PISA ( Krissinel and Henrick , 2007 ) , protein-DNA contacts in NUCPLOT ( Luscombe et al . , 1997 ) , and DNA topology in W3DNA ( Zheng et al . , 2009 ) . All structural figures were made in PyMOL ( Version 1 . 5 . 0 . 4; Schrödinger , LLC , New York , NY , USA ) and animations in Chimera ( Pettersen et al . , 2004 ) . For electrophoretic mobility shift assays ( EMSA ) , XerH was incubated with 50 bp difH DNA or its derivatives ( Supplementary file 1 ) in the activity buffer as for DNase I footprinting for 1 h at 30°C . 20 nM 5’-32P-labelled DNA was titrated with increasing amounts of XerH ( 0–360 nM ) , and complex formation was assessed on native 12% polyacrylamide TBE gels . Gels were imaged in a Typhoon FLA 7000 Phosphoimager , and images were quantified and analyzed with ImageQuant ( GE Healthcare ) . Dissociation constants ( KD ) were calculated for the sum of all complex species . XerH-difH complexes were also analyzed by analytical size exclusion chromatography on Superdex 200 3 . 2/30 column on Äkta systems ( GE Healthcare ) . The XerH-DNA complexes were prepared as for crystallization . Cleavage assays were carried out in activity buffer with 25 μM XerH and 25 μM difH substrates . ‘Suicide’ difH substrates containing nicks in the backbone of both DNA strands 1 nt 3’ to the cleavage position were used . These substrates trap the covalent XerH-difH reaction intermediates ( Figure 3C ) as seen for Cre recombinase and λ integrase ( Guo et al . , 1997; Pargellis et al . , 1988 ) . Oligonucleotide sequences are shown in Supplementary file 1 . DNA substrates were annealed by slow cooling from 98°C to 4°C . The reactions were incubated for 1 h at 37°C , and analyzed by SDS-PAGE . XerH-mediated intramolecular recombination was assessed in E . coli by a galK marker-based assay , similar to the one described by ( Arnold et al . , 1999 ) . The difH cassette was PCR-amplified from H . pylori 26695 genome using primers shown in Supplementary file 1 and inserted in a direct repeat orientation at two positions of plasmid pMS183Δ using NheI/BsrGI or EcoRI/KpnI restriction sites . The mutated variants of the difH cassette were obtained by site-directed mutagenesis of the constructed plasmids using primers shown in Supplementary file 1 . The xerH gene was cloned from the expression plasmids used for XerH production into plasmid pBAD/MCS ( PepCore , EMBL ) using NcoI/XhoI restriction sites . GalK-deficient E . coli strain DS941 was sequentially transformed first with the XerH expression plasmid , and then with the reporter plasmid . Transformants were plated on 4% ( w/v ) MacConkey agar ( DifcoTM , Becton and Dickinson , Heidelberg , Germany ) supplemented with 1% ( w/v ) galactose , 50 μg/ml kanamycin , and 100 μg/ml ampicillin . Upon XerH expression , recombination between the two difH sites leads to loss of galK . After overnight growth the plates were scraped to resuspend the cells in 1 ml of LB , and an overnight culture was set up with 1 μl of the scraped cells . E . coli DS941 cells were transformed with the plasmid DNA extracted from the overnight cultures , and the cells were plated again on MacConkey agar supplemented with 1% ( w/v ) galactose and 50 μg/ml kanamycin . Red ( indicating no recombination ) and white ( indicating recombination deleting the galK gene from the plasmid ) colonies were counted and recombination efficiency was calculated as the number of white colonies divided by the total number of colonies . Representative plasmids from white colonies were analyzed by agarose gel electrophoresis and sequencing to confirm that reciprocal recombination between difH sites had occurred as predicted . First , DNA-bound XerC and XerD were modeled using the DNA-free XerD structure ( PDB: 1 A0P ) ( Subramanya et al . , 1997 ) and an XerC homology model created using I-TASSER ( Yang et al . , 2015 ) . XerC/D tetramers were assembled by superposing individual protein domains onto our XerH-difH structures by rigid structural alignment using FATCAT ( Li et al . , 2006 ) with XerD placed onto the left difH arm ( i . e . superposed onto subunit A in the cleavage-competent conformation ) . Flexible parts that could not be aligned ( αM-αO and β2-β3 hairpin ) were isolated and modeled by threading with Phyre2 ( Karaca and Bonvin , 2011; Kelley et al . , 2015 ) . dif DNA was modeled with mode-RNA server ( Rother et al . , 2011 ) using difH as a template and refined in HADDOCK ( van Zundert et al . , 2016 ) . Both pre- and post-cleavage complexes were modeled and the assembled models were refined in HADDOCK to optimize molecular geometry . Coordinates and structure factors have been deposited in the Protein Data Bank under accession codes 5JK0 ( XerH-difH pre-cleavage synaptic complex ) and 5JJV ( XerH-difHLP post-cleavage synaptic complex ) . | Similar to humans , bacteria store their genetic material in the form of DNA and arrange it into structures called chromosomes . In fact , most bacteria have a single circular chromosome . Bacteria multiply by simply dividing in two , and before that happens they must replicate their DNA so that each of the newly formed cells receives one copy of the chromosome . Occasionally , mistakes during the DNA replication process can cause the two chromosomes to become tangled with each other; this prevents them from separating into the newly formed cells . For instance , the chromosomes can become physically connected like links in a chain , or merge into one long string . This kind of tangling can result in cell death , so bacteria encode enzymes called Xer recombinases that can untangle chromosomes . These enzymes separate the chromosomes by cutting and rejoining the DNA strands in a process known as Xer recombination . Although Xer recombinases have been studied in quite some detail , many questions remain unanswered about how they work . How do Xer recombinases interact with DNA ? How do they ensure they only work on tangled chromosomes ? And how does a protein called FtsK ensure that Xer recombination takes place at the correct time and place ? Bebel et al . have now studied the Xer recombinase from a bacterium called Helicobacter pylori , which causes stomach ulcers , using a technique called X-ray crystallography . This enabled the three-dimensional structure of the Xer recombinase to be visualized as it interacted with DNA to form a Xer-DNA complex . Structures of the enzyme before and after it cut the DNA show that Xer-DNA complexes first assemble in an inactive state and are then activated by large conformational changes that make the DNA bend . Bebel et al . propose that the FtsK protein might trigger these changes and help to bend the DNA as it activates Xer recombination . Further work showed that the structures could be used to model and understand Xer recombinases from other species of bacteria . The next step is to analyze how FtsK activates Xer recombinases and to see if this process is universal amongst bacteria . Understanding how this process can be interrupted could help to develop new drugs that can kill harmful bacteria . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"chromosomes",
"and",
"gene",
"expression",
"structural",
"biology",
"and",
"molecular",
"biophysics"
] | 2016 | Structural snapshots of Xer recombination reveal activation by synaptic complex remodeling and DNA bending |
How cancer cells globally struggle with a chemotherapeutic insult before succumbing to apoptosis is largely unknown . Here we use an integrated systems-level examination of transcription , translation , and proteolysis to understand these events central to cancer treatment . As a model we study myeloma cells exposed to the proteasome inhibitor bortezomib , a first-line therapy . Despite robust transcriptional changes , unbiased quantitative proteomics detects production of only a few critical anti-apoptotic proteins against a background of general translation inhibition . Simultaneous ribosome profiling further reveals potential translational regulation of stress response genes . Once the apoptotic machinery is engaged , degradation by caspases is largely independent of upstream bortezomib effects . Moreover , previously uncharacterized non-caspase proteolytic events also participate in cellular deconstruction . Our systems-level data also support co-targeting the anti-apoptotic regulator HSF1 to promote cell death by bortezomib . This integrated approach offers unique , in-depth insight into apoptotic dynamics that may prove important to preclinical evaluation of any anti-cancer compound .
Most cytotoxic chemotherapeutics eliminate tumor cells by activating the intrinsic apoptotic pathway ( Kaufmann and Earnshaw , 2000 ) . The final stages leading to mitochondrial pore formation and caspase activation have been well-described at the molecular level ( Gonzalvez and Ashkenazi , 2010; Spencer and Sorger , 2011; Parrish et al . , 2013 ) . However , prior to these terminal stages of apoptosis , it is becoming clear that cells fight to resist the pharmacological insult in ways that appear unique to the treatment applied ( Geva-Zatorsky et al . , 2010 ) . Existing genome-wide studies of cellular response to chemotherapeutic treatment have primarily relied on changes at the transcriptional level ( Lamb et al . , 2006 ) . Elegant single cell studies have tracked a subset of proteins in response to various chemotherapeutics ( Cohen et al . , 2008; Geva-Zatorsky et al . , 2010 ) . Recent work has identified hundreds of caspase cleavage substrates in apoptotic cells treated with chemotherapeutics ( Shimbo et al . , 2012 ) . However , each of these approaches only captures a segment of the functional reaction to a chemotherapeutic insult and does not tell the full story of how cancer responds to apoptosis-inducing drug treatment . It is known that cells undergoing apoptosis show strong suppression of protein translation ( Bushell et al . , 2004 ) . While a few specific transcripts are known to escape this translational suppression ( Spriggs et al . , 2010 ) , the general link between transcriptional and translational changes during apoptosis is not well understood . Furthermore , different chemotherapeutics produce distinct quantitative signatures of caspase cleavage substrates following apoptosis ( Shimbo et al . , 2012 ) , and it is unclear how the cellular response to chemotherapy prior to apoptotic induction may influence the later deconstruction of cellular protein contents . Here we examine in depth the response of a multiple myeloma cell culture line ( MM1 . S ) exposed to the proteasome inhibitor bortezomib , a clinically relevant model system . Multiple myeloma , one of the most common blood cancers , is an aggressive malignancy of clonal plasma cells . Proteasome inhibition has become an effective first-line therapy for this disease , though myeloma currently has no known cure ( Lonial et al . , 2011 ) . We use a suite of emerging systems-level approaches to globally examine the dynamic interplay between transcription , translation , and proteolytic degradation during chemotherapy-induced apoptosis in this system ( Figure 1A ) . Via ribosome profiling , we identify preferential translation and translational regulation of genes expected to reduce unfolded protein stress after bortezomib treatment . Surprisingly , despite these changes in translational control , during rapid apoptosis only a few critical pro-survival proteins are detectably increased in proteomic studies . Furthermore , we find that cleavage patterns by both caspase and non-caspase proteases during apoptosis are largely independent of drug response at the transcriptional and translational level prior to cell death . We also use this integrated data to examine the potential of small molecule therapeutics to work in concert with bortezomib in myeloma . Such a global examination of how cells struggle at all levels with a chemotherapeutic insult provides important insights into mechanisms of therapeutic resistance and novel methods to assess tumor response to chemotherapy . 10 . 7554/eLife . 01236 . 003Figure 1 . Experimental design and paired mRNA-seq/ribosome profiling data . ( A ) MM1 . S cells were prepared in six separate flasks each containing 3 × 108 cells . All time points past 0 hr were exposed to 20 nM bortezomib . Cells were harvested and analyzed using four systems-level technologies shown . ( B ) Cells undergo rapid apoptosis with initial increase in caspase activation seen at 6 hr and <10% cell viability by 12 hr ( duplicate , mean ± SD ) . Total RNA and mRNA were isolated and measured by spectrophotometry; total protein was measured by BCA assay ( measured in duplicate , isolated from 106 cells; mean ± SD ) . G ( t ) , describing global mRNA degradation , is derived from a sigmoid fit to the total mRNA data . ( C ) Heat map showing log2 expression compared to 0 hr for normalized mRNA and ribosome footprint read density ( reads per kilobase million , RPKM ) for 5680 well-expressed transcripts ( Figure 1—source data 1 ) . Also shown is calculated translational efficiency ( TE ) relative to 0 hr , calculated as the ratio of ( footprint RPKM ) / ( mRNA RPKM ) per transcript at each time point . We used unsupervised hierarchical clustering to define five broad groups of transcripts with relation to mRNA , footprint , and translational efficiency changes . DOI: http://dx . doi . org/10 . 7554/eLife . 01236 . 00310 . 7554/eLife . 01236 . 004Figure 1—source data 1 . Deep sequencing data , translational efficiency , and gene clusters . ( A ) mRNA-seq and ( B ) ribosome profiling deep sequencing data with assigned UCSC hg19 identifier , associated UniProt accession number , and gene identifier as defined by kgXref feature in UCSC genome browser . For samples at each time point , we display raw reads that uniquely mapped to canonical transcript as well as mapped reads normalized to total read number and transcript length ( reads per kilobase million , RPKM ) for comparison across samples . These 5680 tracked transcripts had RPKM ≥1 across all 12 samples ( six mRNA-seq , six ribosome footprint ) . ( C ) Translational efficiency ( TE ) for all tracked transcripts as measured by ( RPKM of footprint sample/RPKM of mRNA-seq sample ) at each time point . Table is ordered by descending TE at 0 hr . The histone genes are artifactually elevated as these mRNA's do not contain poly ( A ) tails and are therefore under-represented in the mRNA-seq data , as previously noted in Ingolia et al . ( 2011 ) . ( D ) List of transcripts included in each Cluster ( as in Figure 1C ) as uploaded to Ingenuity Pathway Analysis server for analysis ( results in Table 1 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01236 . 00410 . 7554/eLife . 01236 . 005Figure 1—figure supplement 1 . Sample comparison and biological subgroups . ( A ) mRNA and ( B ) footprint samples prepared and sequenced independently at early time points show strong correlation in read density , demonstrating consistency in experimental methods . In footprint samples the mean difference between read density at 1 . 5 hr and 0 hr across transcripts = 28% , reflecting both experimental error and biological variation . ( C ) Translational efficiency spans >10-fold across all tracked transcripts without any large shifts across time points . DOI: http://dx . doi . org/10 . 7554/eLife . 01236 . 00510 . 7554/eLife . 01236 . 006Figure 1—figure supplement 2 . Immunoblots . Western blotting for proteins related to ER stress/UPR ( ATF4 , CHOP , XBP1s [‘s’ for spliced , active form] ) and apoptosis ( Bid , XIAP ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01236 . 00610 . 7554/eLife . 01236 . 007Figure 1—figure supplement 3 . Biological subgroups . Genes curated from the literature related to ER stress ( left ) and apoptosis ( right ) that were found in among the ∼5700 tracked transcripts . Genes highlighted with ribosome footprint analysis are highlighted with red arrows ( DDIT3 = CHOP; DDIT4 = REDD1; PPP1R15A = GADD34 ) . All data are shown as Log2 expression ratio vs 0 hr . DOI: http://dx . doi . org/10 . 7554/eLife . 01236 . 007
We designed our experiment to monitor four systems-level processes simultaneously in the same cellular population , from transcription through protein production and proteolysis ( Figure 1A ) . Deep sequencing of mRNA ( mRNA-seq ) examines the detailed transcriptional response of myeloma cells to bortezomib-induced stress . We pair this data with ribosome profiling , the deep sequencing of ribosome-protected mRNA fragments during active translation; this approach offers significant insight into proteome dynamics and translational regulation not available by monitoring transcript expression alone ( Ingolia et al . , 2009 , 2011; Stern-Ginossar et al . , 2012 ) . To directly measure changes at the protein level after bortezomib perturbation , we use isobaric labeling by iTRAQ ( Mertins et al . , 2012 ) and tandem mass spectrometry to measure in an unbiased fashion the relative abundance of ∼2600 proteins . Finally , using enzymatically-driven labeling of free protein α-amines ( N-terminomics ) ( Mahrus et al . , 2008; Agard et al . , 2012 ) , we develop a quantitative mass spectrometry assay to monitor to the relative kinetics of both caspase and non-caspase cleavage events during apoptotic cellular deconstruction . To begin , we prepared separate flasks of MM1 . S cells and harvested cells at five time points after induction of apoptosis with 20 nM bortezomib . Similar to previous studies ( Shimbo et al . , 2012 ) , caspase activity was first detected at 6 hr and the majority of cells were non-viable by 12 hr ( Figure 1B ) . mRNA levels were also markedly decreased prior to loss of cell viability , as previously seen across a range of apoptotic conditions ( Del Prete et al . , 2002; Bushell et al . , 2004 ) . Total RNA levels , primarily reflecting ribosomal RNA abundance , and total protein levels were relatively stable . To understand how myeloma cells were struggling after bortezomib exposure , we determined the relative changes in transcription and translation within the remaining mRNA pool at each time point . We normalized mRNA-seq and ribosome profiling sample using reads per kilobase million ( RPKM ) and focused our analysis on 5680 well-expressed transcripts ( Figure 1C ) . We compared log2 expression vs the untreated sample and organized the data using unsupervised hierarchical clustering . This approach allowed us to center on relative expression changes found in multiple independent samples . We generally found that changes in ribosome footprints tracked with changes in transcript abundance . The transcripts could be sub-divided into five clusters ( Figure 1C ) . Clusters ‘Upreg’ and ‘Downreg’ encompass transcripts that are generally increased or decreased , respectively , at both the level of mRNA and footprint reads ( Figure 2A , B ) . Cluster ‘Stable’ includes transcripts that demonstrate very mild changes at both the mRNA and footprint level . We calculated the translational efficiency ( TE ) from the ratio of ribosome footprint to mRNA-seq read density for each transcript ( Ingolia et al . , 2011 ) . This enabled us to characterize two other groups that are particularly interesting , Cluster ‘TE Up’ and ‘TE Down’ . These groups showed little change in the level of mRNA , but large increases or decreases , respectively , in relative ribosome footprint density . This suggests that genes in these two clusters are regulated at the level of translation . 10 . 7554/eLife . 01236 . 008Figure 2 . Translational efficiency ( TE ) during bortezomib-induced apoptosis . ( A and B ) Paired mRNA and ribosome footprint reads for HSPB1 ( Cluster Upreg ) and HMGB1 ( Cluster Downreg ) showing that protein translation generally tracks with transcript abundance . Displayed read counts are median-normalized to total aligned reads in 0 hr sample for comparison across the time course . Thick green bars = protein coding sequence ( CDS ) ; medium bars = untranslated regions ( UTR ) ; thin lines = intronic or intergenic regions . Read counts are inverted depending on coding direction . ( C ) ATF4 shows increased footprint read density in the absence of mRNA increase at 6 hr . ( D ) SEC61B , a downstream target of XBP1 , shows increased footprints and decreased mRNA reads , leading to increased TE . ( E ) Many ribosome structural proteins , including RPL7A , are found in Cluster TE down . ( F ) Log2 changes in transcript TE from Cluster TE up and Cluster TE down at 12 hr vs 0 hr demonstrate statistically significant changes in the overall TE distribution ( p<0 . 0001 , Mann-Whitney test ) . ( G ) Mean ( ±SEM ) changes in TE during the apoptotic time course generally do not exceed twofold even at later time points . DOI: http://dx . doi . org/10 . 7554/eLife . 01236 . 00810 . 7554/eLife . 01236 . 009Figure 2—figure supplement 1 . Subgroup analysis from biological pathways showing changes in translation efficiency from Ingenuity Pathway Analysis . Genes included in IPA as XBP1 downstream targets ( left ) and Genes included as part of the eIF2 signaling pathway in IPA ( right ) . Green boxed area indicates numerous ribosome structural proteins with decreased footprint reads relative to mRNA . SEC61B and RPL7A ( shown in Figure 2 ) are highlighted with red arrows , as is PERK ( EIF2AK3 ) . All data are shown as Log2 expression ratio vs 0 hr . DOI: http://dx . doi . org/10 . 7554/eLife . 01236 . 009 Bortezomib is known to be an inducer of the unfolded protein response ( UPR ) and endoplasmic reticulum ( ER ) stress ( Aronson and Davies , 2012 ) . Myeloma plasma cells are particularly sensitive to proteasome inhibition in vivo due to extremely high rates of immunoglobulin production , potentially leading to apoptosis via the UPR ( Obeng et al . , 2006; Walter and Ron , 2011 ) . In Figure 2C we display reads for one of the critical factors in the UPR , ATF4 , a transcription factor known to be under translational control during ER stress ( Lu et al . , 2004 ) . Consistent with this prior work , we find that ATF4 does not increase in transcript abundance but shows a nearly threefold increase of ribosome occupancy ( see heat map in Figure 1—figure supplement 3 ) . To better assess the biological implications of these results we turned to Ingenuity Pathway Analysis ( IPA ) ( Ingenuity Systems , www . ingenuity . com ) ( Table 1 ) . Cluster Upreg is enriched for genes related to protein ubiquitination ( p=1 . 50 × 10−34 ) , protein degradation ( 9 . 63 × 10−9 ) , chaperones ( 4 . 15 × 10−11 ) , and hypoxic response ( 2 . 90 × 10−10 ) . Cluster Downreg includes genes important in cellular proliferation ( p=3 . 61 × 10−11 ) and DNA repair ( 3 . 81 × 10−9 ) . These findings are consistent with mRNA microarray studies of bortezomib response ( Mitsiades et al . , 2002 ) . We examined in more detail a subset of genes related to cellular apoptosis and both ER stress and hypoxic response ( Figure 1—figure supplement 3 ) . Notably , we found little change in expression or translation of canonical apoptosis players . 10 . 7554/eLife . 01236 . 010Table 1 . Biological relevance of findings from Ingenuity Pathway Analysis ( IPA ) DOI: http://dx . doi . org/10 . 7554/eLife . 01236 . 010ClusterMolecular and cellular functionp valueCanonical pathwayp valueUpstream regulatorp valueCluster UpregPost-translational modification4 . 15 e-11Protein ubiquitination pathway1 . 50e-34NFE2L2 ( NRF2 ) 7 . 77e-16Protein Folding4 . 15 e-11NRF2-mediated oxidative stress response2 . 90e-10HSF11 . 70e-14Cell death and survival2 . 47 e-10EIF2AK3 ( PERK ) 6 . 47e-10Protein synthesis8 . 82 e-09Protein degradation9 . 63 e-09Cluster StableRNA post-transcriptional modification1 . 44e-09––Molecular transport8 . 87 e-09Protein trafficking8 . 87 e-09Cluster TE Up–Mitochondrial dysfunction6 . 04e-10XBP18 . 79e-10Cluster DownregCell death and survival7 . 56e-12–TP532 . 67e-15Cellular growth and proliferation3 . 61e-11MYC1 . 53e-10DNA replication , recombination and repair3 . 81e-09XBP11 . 65e-10RNA post-transcriptional modification8 . 91e-08INSR5 . 94e-09E2F46 . 20e-09Cluster TE DownRNA post-transcriptional modification8 . 88e-11EIF2 signaling2 . 35e-17–Gene expression3 . 24e-09Regulation of eIF4 and p70S6K signaling1 . 40e-085% transcripts highest TE at 0 hr–Mitchondrial dysfunction1 . 52e-26–Increased 5′ UTR translation–EIF2 signaling2 . 45 e-17MYC1 . 09e-08HSF18 . 11e-08Decreased 5′ UTR translation–––Clusters defined by hierarchical clustering of mRNA , footprint , and translational efficiency data as shown in Figure 1B ( included genes listed in Figure 1—source data 1 ) . We also analyzed the top 5% of transcripts by translational efficiency at 0 hr as calculated by ( footprint RPKM ) / ( mRNA RPKM ) per transcript . p values are as calculated using Fisher′s exact test by Ingenuity Pathway Analysis software . We report findings here with a p<10−7 . Results from each category ( Molecular and cellular function; Canonical pathway; Upstream regulator ) are listed in order of decreasing p , independent of the other categories . Surprisingly , IPA showed that Cluster TE Up included many genes regulated by XBP1 ( p=8 . 79 × 10−10 ) ( Table 1 ) , an important component of the UPR ( Walter and Ron , 2011 ) . The only known function of XBP1 is as a transcription factor ( Acosta-Alvear et al . , 2007 ) without known direct effects on translation . Although we do see a slight increase in the active XBP1 protein ( Figure 1—figure supplement 2 ) , most downstream XBP1 targets do not respond by increases in mRNA abundance . Instead we see increased translation efficiency ( Figure 2—figure supplement 1 ) . Notably , transcripts related to the ER transport machinery ( including SEC61B , Figure 2D ) demonstrate 1 . 5- to 2-fold increases in translation vs transcription . Therefore , changes in translation of XBP1downstream targets may reflect the activity of a parallel UPR mechanism to favor adaptation to cellular stress . Cluster TE Down includes many transcripts involved in the translational machinery itself , particularly ribosomal structural proteins ( Figure 2E , Figure 2—figure supplement 1 ) of which the most significant pathway affected involved eIF2 signaling ( p=2 . 35 × 10−17 ) ( Table 1 ) . In parallel we examined the 5% of genes with the highest translational efficiency in the untreated sample . The pathway ‘mitochondrial dysfunction’ was highly enriched ( p=1 . 52 × 10−26 ) ( Table 1 ) , including numerous cytochrome c oxidase and NADH dehydrogenase subunits ( Figure 1—source data 1 ) . This finding suggests a favoring of translation of aerobic metabolism components in myeloma cells at baseline . In Figure 2F we display the distribution of log2-fold changes in translational efficiency for Clusters TE Up and TE Down , which both show a significant difference between 12 hr vs 1 . 5 hr ( p<0 . 0001 , Mann-Whitney test ) . These changes contrast with the large majority of transcripts which show little change in TE ( Figure 2G ) . Taking advantage of the nucleotide resolution of ribosome profiling , we examined whether there were any large-scale changes in ribosome occupancy along mRNA during apoptosis . We performed a metagene analysis , where all footprint profiles are averaged and then aligned based on the midpoint of the protected reads ( Figure 3A ) . We find a strong peak of ribosome occupancy at both the 5′ and 3′ ends of annotated coding sequence ( CDS ) and a 3-nucleotide offset resulting from position of the ribosome P site , as described previously ( Ingolia et al . , 2009 ) . We also note peaks appearing every three nucleotides across the averaged reads , consistent with the triplet nucleotide coding sequence . Averaged across all transcripts , we did not find any large changes in footprint read distribution across mRNAs at different time points . 10 . 7554/eLife . 01236 . 011Figure 3 . Nucleotide resolution of ribosome profiling reveals changes in 5′ UTR translation . ( A ) Metagene analysis , the alignment of footprint reads across all transcripts , to both the 5′ CDS start site and 3′ termination site ( Ingolia et al . , 2009 ) . Reads counts are median-normalized to total aligned reads per sample . No significant changes are noted in overall read alignments during apoptosis . ( B ) Relative translation of 5′ UTR measured by the ratio of 5′ UTR reads to CDS reads per transcript . The log2 change in this ratio is compared to the 5′ UTR/CDS ratio at 0 hr . ( C and D ) Examples of genes with most increased or decreased relative 5′ UTR translation include HSPA8 ( Hsc70 protein ) and CYCS ( cytochrome c ) , respectively . ( E–H ) Genes known to be involved in ER stress response show increased read density at 5′ UTR upstream open reading frames ( uORFs ) . ATF4 shows density at known uORFs as well as a single upstream AUG codon . Putative uORF translation with noted initiation codon is also identified based on areas of increased read density for DDIT4 ( REDD1 ) , PPP1R15B , and TXNIP ( see detailed sequence information in Figure 3—figure supplement 2 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01236 . 01110 . 7554/eLife . 01236 . 012Figure 3—source data 1 . 5′ UTR translation . ( A ) RPKM values of reads uniquely mapped to either 5′ UTR or CDS of transcripts ( same list of transcripts as in Figure 1—source data 1 ) . Sequence and length of 5′ UTR and CDS ( for RPKM calculation ) are defined based on the canonical transcript isoform annotated in hg19 . ( B ) Transcripts with 5′ UTR/CDS translation ratio increased ≥twofold at a minimum of three time points compared to transcript 5′ UTR/CDS translation ratio at 0 hr . ( C ) as in B but for transcripts with decreased 5′ UTR to CDS translation ratio . DOI: http://dx . doi . org/10 . 7554/eLife . 01236 . 01210 . 7554/eLife . 01236 . 013Figure 3—figure supplement 1 . 5′ UTR length . For genes found to have most increased ( 274 genes ) and most decreased ( 219 genes ) relative 5′ UTR translation during the time course , there is no significant difference in the distribution of annotated 5′ UTR length ( p=0 . 94 , Mann-Whitney U test ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01236 . 01310 . 7554/eLife . 01236 . 014Figure 3—figure supplement 2 . Ribosome profiling demonstrates strong footprint read density in known uORFs of DDIT3 ( CHOP ) and PPP1R15A ( GADD34 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01236 . 01410 . 7554/eLife . 01236 . 015Figure 3—figure supplement 3 . Sequence data indicating uORFs suggested by increased footprint read density in REDD1 , TXNIP , and PPP1R15B 5′ UTRs . Putative REDD1 uORFs are translated out of frame with the CDS; TXNIP uORFs 1 and 2 are out of frame , while uORF3 is in frame; both PPP1R15B uORFs are in frame with the CDS . DOI: http://dx . doi . org/10 . 7554/eLife . 01236 . 015 We next investigated whether there were general changes in proteome control by differing ribosome occupancy of the mRNA 5′ untranslated region ( UTR ) . Ribosome occupancy in this region may indicate translation of short regulatory polypeptides in upstream open reading frames ( uORFs ) or production of alternate N-terminal isoforms of canonically translated proteins ( Ingolia et al . , 2011; Lee et al . , 2012; Stern-Ginossar et al . , 2012 ) . Others have shown that yeast responding to oxidative stress produce large increases in 5′ UTR translation ( Gerashchenko et al . , 2012 ) . A general decrease in 5′ UTR translation was also seen in differentiating mouse embryonic stem cells ( Ingolia et al . , 2011 ) . In contrast , we identified no overall trend toward altered 5′ UTR translation relative to CDS translation in our system ( Figure 3B ) . However , these frequency distributions are broad and some individual transcripts do have large changes in the relative translation of the 5′ UTR during apoptosis . We examined transcripts with >twofold change in 5′ UTR translation relative to CDS translation when compared to untreated sample in at least three time points after drug exposure ( Figure 3—source data 1 ) . By χ2 analysis , genes in Cluster Upreg were significantly over-represented among the 274 genes in the increased UTR translation group ( p=0 . 033 ) . Both Clusters Upreg and Downreg were over-represented among the 219 genes in the decreased UTR translation group ( p=0 . 0006 and p=0 . 014 , respectively ) . No other Clusters showed significant over- or under-representation . We did not find any difference in 5′ UTR length between the groups with most increased and most decreased relative 5′ UTR translation ( Figure 3—figure supplement 1 ) . IPA ( Table 1 ) and inspection of genes in the increased UTR translation group revealed numerous translation elongation and initiation factors , ribosome structural proteins , chaperones ( i . e . , DNAJA1 , HSP90AB1 , CCT5 , HSPA8; Figure 3C ) , and proteasomal subunits . The decreased UTR translation group does not show as strong a biological pattern although we do find ubiquitin ( UBB ) , the co-chaperone BAG3 , and cytochrome c ( CYCS; Figure 3D ) , that are relevant to bortezomib-induced apoptosis . The production of ATF4 is thought to be governed by the relative translation of two uORFs in the 5′ UTR ( Lu et al . , 2004 ) . We find that changes in the footprint density across both of these uORFs correspond with increased translation of the ATF4 CDS ( Figure 1—figure supplement 3 ) and detection of the protein product by immunoblot ( Figure 1—figure supplement 2 ) . We do also note strong footprint read density of a single amino acid uORF 22 nucleotides from the 5′ end of the ATF4 transcript ( Figure 3E ) . Ribosome occupancy of this single methionine is of unclear significance , though it may also play an uncharacterized role in regulating ATF4 translation in myeloma . We note that in these samples translation elongation was inhibited by cycloheximide . This treatment does not identify uORFs with as high precision as inhibition of initiation complexes by harringtonine or lactimidomycin ( Ingolia et al . , 2011; Lee et al . , 2012; Stern-Ginossar et al . , 2012 ) . Thus , we cannot confidently assign 5′ UTR read density across all genes to specific start codons , particularly non-AUG uORFs . However , we did focus on a subset of genes important in the ER stress response . We first looked at DDIT3 ( CHOP ) and PPP1R15A ( GADD34 ) , where in both genes we found strong footprint read density in uORFs known to regulate CDS translation ( Figure 3—figure supplement 2 ) ( Jousse et al . , 2001; Lee et al . , 2009 ) . We next examined other genes with a role in the ER stress response but without known uORF translation . These include DDIT4 ( REDD1 ) , a repressor of mTOR signaling potentially under control of ATF4 ( Whitney et al . , 2009 ) ; TXNIP , recently found to be a key mediator of apoptosis under the UPR ( Lerner et al . , 2012; Oslowski et al . , 2012 ) ; and PPP1R15B , which also plays a role in the dephosphorylation of eIF2α ( Harding et al . , 2009 ) . Intriguingly , all three genes showed apparent uORF translation , identified as areas of increased 5′ UTR read density with a peak corresponding to AUG or near-AUG initiation codons ( Ingolia et al . , 2011 ) and bounded by a stop codon on the 3′ end ( Figure 3F–H , Figure 3—figure supplement 3 ) . We next explored how the changes at the level of transcription and translation compare to the changes in protein levels during bortezomib-induced apoptosis . At each time point we isolated total protein and employed 6-plex iTRAQ labeling with a different isobaric mass tag ( Mertins et al . , 2012 ) so that data from all time points can be compared simultaneously in a single mass spectrometry ( MS ) experiment . After analysis on two different mass spectrometers , a total of 2686 proteins were identified with at least two unique peptides and quantifiable iTRAQ reporter intensities . Typical MS spectra are shown in Figure 4A . 10 . 7554/eLife . 01236 . 016Figure 4 . Unbiased and targeted quantitative proteomics reveals abundance changes in a small subset of proteins . ( A ) Example mass spectra demonstrating both m/z peaks used for peptide sequencing and iTRAQ reporter ion signal to measure relative abundance across time points . ( B ) Hierarchical clustering heat map of paired mRNA and ribosome footprint relative read density vs 0 hr with relative iTRAQ protein abundance for 2572 proteins . ( C ) Inset of heat map for proteins increased in relative abundance by >50% at ≥2 time points shows few proteins are measurably produced during bortezomib-induced apoptosis . ( D ) Targeted selected reaction monitoring ( SRM ) assays orthogonally validate iTRAQ data for 152 proteins . In this representative data , each colored trace monitors the intensity of a given parent and fragment ion pair , as demarcated in the peptide sequence; multiple co-eluting peaks positively identify a targeted peptide . ( E ) Protein abundance measured by both SRM and iTRAQ demonstrate strong correlation across time points ( r = 0 . 80 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01236 . 01610 . 7554/eLife . 01236 . 017Figure 4—source data 1 . Proteomic data . iTRAQ proteomic data organized at the protein ( A ) and peptide ( B ) level for 2686 proteins . All proteins included had a minimum of two unique peptides mapping to the protein; all peptides showed minimum iTRAQ reporter ion intensity of 300 cps at 0 hr . ( C ) iTRAQ proteomic data matched by UniProt Accession number to genes tracked in mRNA and footprint data across the time course . ( D ) List of proteins identified by iTRAQ proteomics but not present in analyzed genes from RNA deep sequencing . ( E ) List of parent and fragment ion transitions used for all peptides in SRM validation of iTRAQ data . ( F ) Protein and ( G ) peptide intensity data for targets tracked in SRM assay . DOI: http://dx . doi . org/10 . 7554/eLife . 01236 . 01710 . 7554/eLife . 01236 . 018Figure 4—figure supplement 1 . Protein abundance comparison . Frequency distribution of protein abundance monitored by iTRAQ proteomics ( ∼2700 proteins ) and ribosome profiling ( ∼5700 proteins ) compared to average cellular protein abundance as estimated in PaxDB ( www . pax-db . org ) . We note that both distributions examine similar portions of the proteome , though the iTRAQ demonstrates greater coverage of higher abundance proteins . DOI: http://dx . doi . org/10 . 7554/eLife . 01236 . 01810 . 7554/eLife . 01236 . 019Figure 4—figure supplement 2 . Comparison of baseline ( 0 hr ) read density of mRNA transcripts vs number of identified peptides mapping to each protein in iTRAQ proteomics . We find a weak but positive correlation ( R = 0 . 27 , p<0 . 0001 ) . This distribution is comparable to the relationship between absolute transcript and protein copies at baseline in Figure 5—figure supplement 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 01236 . 01910 . 7554/eLife . 01236 . 020Figure 4—figure supplement 3 . Relative 5′ UTR translation across the time course for upregulated genes . Relative 5′ UTR translation ( measured as normalized footprint read density in 5′ UTR vs normalized footprint read density in coding sequence ) over the time course for selected genes ( A ) upregulated at the mRNA and footprint level and also increased by iTRAQ proteomics and ( B ) upregulated at the mRNA and footprint level with no detected change by iTRAQ . We note that for both sets of genes relative 5′ UTR translation stays relatively constant during bortezomib-induced apoptosis , consistent with genome-wide analysis in Figure 3B . DOI: http://dx . doi . org/10 . 7554/eLife . 01236 . 02010 . 7554/eLife . 01236 . 021Figure 4—figure supplement 4 . Unbiased and targeted proteomics comparison to deep sequencing data . Data from 150 proteins targeted in SRM assay to validate iTRAQ data . Comparison heat map of mRNA and footprint read density with iTRAQ and SRM relative protein abundance . DOI: http://dx . doi . org/10 . 7554/eLife . 01236 . 021 We compared the proteins identified by MS as well as those tracked by ribosome profiling to a database of cellular protein abundance ( www . pax-db . org; Figure 4—figure supplement 1; Wang et al . , 2012 ) . There was significant overlap between the protein abundance range monitored both by ribosome profiling and iTRAQ proteomics , demonstrating that these two techniques are probing a similar portion of the proteome . We concentrated on 2572 of these proteins that also had corresponding mRNA and ribosome footprint data ( Figure 4—source data 1 ) . We noted a slight positive correlation ( R = 0 . 27 ) between baseline transcript read density and number of peptides identified by iTRAQ proteomics ( Figure 4—figure supplement 2 ) . Strikingly , for the vast majority of proteins there is no distinct change in relative protein abundance while large changes are observed in relative mRNA and footprint reads ( Figure 4B ) . Nonetheless , we did find a subset of 12 proteins ( 0 . 47% of total ) with >50% increase in protein abundance across at least two time points ( Figure 4C ) . The majority of these proteins show strong upregulation at the mRNA and footprint level and have potential functions mediating either the folding ( HSPA1A , DNAJB1 , BAG3 , HSPB1 , SERPINH1 , CLU ) or degradation ( SQSTM1 ) of unfolded proteins . These proteins are key players in an anti-apoptotic response that reduce cellular stress after proteasome inhibition by bortezomib . Interestingly , we also find ferritin light chain ( FTL ) increased ( Figure 4C ) . Another group has recently found that depleting ferritin light chain enhances myeloma cell sensitivity to bortezomib ( Campanella et al . , 2012 ) , suggesting it also plays an anti-apoptotic role here . Of note , we did not observe any distinct changes in relative 5′ UTR translation for the few proteins increased by iTRAQ ( Figure 4—figure supplement 3 ) . To further validate the iTRAQ findings , we employed an orthogonal method of label-free , targeted , quantitative mass spectrometry termed Selected reaction monitoring ( SRM ) . This method is analogous to Western blotting without the need for dozens of antibodies and potential improvement in performance characteristics ( Maiolica et al . , 2012 ) . We prepared an independent biological time course of MM1 . S cells treated with 20 nM bortezomib and isolated tryptic peptides from total cellular lysate . We tracked the relative abundance of one to three peptides from 152 proteins . The SRM peak area intensity data was compared to the iTRAQ reporter signal ( e . g . , in Figure 4A–D , respectively; heat map in Figure 4—figure supplement 4 ) . Comparing across all time points in Figure 4E we found a strong correlation between SRM and iTRAQ data ( R = 0 . 80 ) . For the few increased proteins the abundance changes detected by SRM are greater than those found by iTRAQ , which is consistent with the greater suppression of iTRAQ reporter ions often seen in complex samples ( Ow et al . , 2009 ) . The label-free SRM data do not suffer from this same limitation and more likely reflect the true change in protein abundance . This independent MS assay confirms that only a small subset of proteins show measurable increases in abundance during apoptosis . Our quantitative proteomic results are strongly consistent with polysome analysis ( Figure 5—figure supplement 1 ) showing a significant decrease in translation well before any loss in cell viability . We wished to better understand the mechanism leading to this translational shutdown during apoptosis . During the UPR , it is well-known that phosphorylation of the initiation factor eIF2α by the ER-resident sensor kinase PERK is associated with general translation inhibition ( Walter and Ron , 2011 ) . At early time points we find that PERK appears phosphorylated , as previously reported ( Atkins et al . , 2013 ) , and at later time points it undergoes proteolysis ( Figure 5A ) . In contrast to other drugs that induce ER stress , we find levels of eIF2α phosphorylation actually decrease after bortezomib exposure ( Figure 5A ) . This finding suggests that there must be other drivers of translation inhibition in this system . 10 . 7554/eLife . 01236 . 022Figure 5 . Translational shutdown during apoptosis and quantitative modeling reconciles deep sequencing and proteomic data . ( A ) Western blotting tracked UPR proteins important in modulating translation via ER stress . We detect the known caspase cleavage of PERK and 4E-BP1 . ( B ) Differential rate equations describing protein production . R ( t ) = mRNA abundance; P ( t ) = protein abundance; ksp = protein translation rate constant; kdp = protein degradation rate constant . G ( t ) describes global mRNA degradation ( Figure 1B ) . ( C ) Absolute copies per cell at 0 hr of both mRNA ( ∼5700 transcripts ) and protein measured by iBAQ ( ∼3400 proteins ) demonstrates distributions of same order of magnitude as those seen in mouse NIH3T3 cells ( Schwanhausser et al . , 2011 ) . ( D ) Extrapolation of absolute mRNA copy numbers per cell based on RPKM , not accounting for mRNA degradation , is well-fit by either a sigmoid or quadratic function to define R ( t ) . ( E ) Absolute protein copies per cell were extrapolated from iBAQ results based on SRM assay intensity . The modified mass-action model incorporating mRNA degradation was able to well-fit the subset of proteins that were detectably increased and also explained why others increased at the transcriptional level did not show protein increases . DOI: http://dx . doi . org/10 . 7554/eLife . 01236 . 02210 . 7554/eLife . 01236 . 023Figure 5—source data 1 . mRNA and protein absolute abundance . ( A ) Absolute copies of mRNA and ( B ) protein per MM1 . S cell at 0 hr as determined using mRNA read density and iBAQ calculation , respectively , as described in the ‘Materials and methods’ . ( C ) Relative 5′ UTR translation at baseline vs absolute protein copies per cell by iBAQ . ( D ) Extrapolated mRNA and protein copies per cell for 13 genes as used for model fitting as shown in Figure 5—figure supplement 4 . DOI: http://dx . doi . org/10 . 7554/eLife . 01236 . 02310 . 7554/eLife . 01236 . 024Figure 5—figure supplement 1 . Polysome profiles . Polysome profiles by sucrose gradient during the apoptotic time course show decrease in polysome peaks over time during during apoptosis . Translation is largely suppressed well before any loss in cell viability . DOI: http://dx . doi . org/10 . 7554/eLife . 01236 . 02410 . 7554/eLife . 01236 . 025Figure 5—figure supplement 2 . Comparison between deep sequencing and proteomic data . Correlation of mRNA amount ( A ) and ribosome footprint ( B ) read density ( in RPKM ) with absolute protein abundance by iBAQ MS analysis at 0 hr in MM1 . S cells demonstrates limited correlation . ( C ) Strong correlation of ribosome footprint and mRNA read density at 0 hr . All correlations in A–C with p<0 . 0001 . ( D ) Relative 5′ UTR translation at baseline vs protein abundance . We note a weak but non-significant correlation ( p=0 . 35 ) suggesting that increased 5′ UTR translation may relate to lower steady-state protein abundance . DOI: http://dx . doi . org/10 . 7554/eLife . 01236 . 02510 . 7554/eLife . 01236 . 026Figure 5—figure supplement 3 . mRNA-seq data scaled by global mRNA degradation . Relative mRNA expression data from Figure 4B scaled to reflect global mRNA degradation at later time points during apoptosis . By this scaling ( scaling factors for mRNA based on fit G ( t ) function = 1 . 0 at 0 hr , 0 . 93 at 1 . 5 hr , 0 . 91 at 3 hr , 0 . 48 at 6 hr , 0 . 17 at 9 hr , 0 . 12 at 12 hr ) , assuming that all mRNA are degraded at the same rate , the absolute mRNA abundance can be estimated across all cells in the population . Upper inset demonstrate that transcripts with absolute mRNA abundance changes across the entire time course are more likely to have increases in protein level , whereas those with transient increases in mRNA generally do not result in detectable protein production . DOI: http://dx . doi . org/10 . 7554/eLife . 01236 . 02610 . 7554/eLife . 01236 . 027Figure 5—figure supplement 4 . Model fits to 13 selected genes . We used our quantitative model to describe a subset of genes increased at both the transcript and protein level ( A and B ) and those where relative transcript levels increased but protein levels were unchanged ( C and D ) . Solid lines in ( A and C ) represent quadratic fits ( FTL , SQSTM1 , CTSD ) or four-parameter sigmoid fits ( all others ) to relative mRNA abundance data to describe R ( t ) , as in Figure 5E . Solid lines in ( B and D ) represent fits to modified mass-action to describe P ( t ) as shown in Figure 5F . DOI: http://dx . doi . org/10 . 7554/eLife . 01236 . 02710 . 7554/eLife . 01236 . 028Figure 5—figure supplement 5 . Predicted protein changes from mRNA changes . Model predictions demonstrating that even during apoptosis , for the same increase in relative transcript levels low-abundance proteins ( 500 copies per cell at baseline ) will have substantial relative protein increases , whereas high abundance proteins ( 100 , 000 copies per cell at baseline ) will have much smaller changes in relative protein abundance . This is even true when the absolute transcript abundance change is much greater for the high abundance protein and including ksp , the rate constant describing proteins produced per mRNA per hour , as greater for the high abundance protein based on the results of Schwanhausser et al . 2011 . DOI: http://dx . doi . org/10 . 7554/eLife . 01236 . 028 Phosphorylation and dephosphorylation of proteins in the eIF4 complex may also inhibit mRNA cap-dependent translation , including hypophosphorylation of 4E-BP1 ( Spriggs et al . , 2010 ) . Indeed , we observed 4E-BP1 dephosphorylation at the 6 hr time point . This is followed by rapid caspase degradation of 4E-BP1 during cell death ( Figure 5A ) . It is also known that caspases target other proteins in the eIF4 complex ( Bushell et al . , 2004 ) . Using N-terminomics ( described in more detail below ) we found that proteolysis of initiation factors begins at the 6 hr time point and accelerates later in apoptosis ( Figure 6—figure supplement 1 ) . Another potentially important mechanism is the global degradation of mRNA during apoptosis ( Del Prete et al . , 2002 ) ( Figure 1B ) , depriving ribosomes of substrates for translation . Together , these results demonstrate that well before the loss of cellular viability , translation is inhibited both by shutting down translation initiation and destruction of mRNA . We sought a quantitative explanation for the limited changes in relative protein level we observed despite large changes in relative transcript and ribosome footprints . Recently developed systems-level technologies , as we use here , enable genome-wide quantitative assessment of the efficiency for decoding of mRNA to protein ( Vogel and Marcotte , 2012 ) . In this apoptotic system , proteins are relatively stable due to several factors: at 20 nM bortezomib in MM1 . S cells , proteasomal activity is almost completely inhibited ( Berkers et al . , 2005 ) . The proteasome is also extensively cut by caspases leading to loss of activity ( Gray et al . , 2010; Figure 6—figure supplement 1 ) . Moreover , we find that endoproteolysis during apoptosis cuts fewer than 20% of the proteins in the cell and often only once or twice per protein leading to stable domains ( Dix et al . , 2008; Mahrus et al . , 2008; Crawford et al . , 2013 ) . These domains would be indistinguishable from intact proteins by iTRAQ . Therefore , we expect that detectable protein degradation is extremely limited and changes by iTRAQ largely reflect protein production alone . In contrast , mRNAs with very different stabilities ( steady-state half-lives on the order of minutes to hours ) are rapidly degraded with similar kinetics under various apoptotic inducers even prior to cell death ( Del Prete et al . , 2002 ) . Mass-action models , as shown in Figure 5B , can describe the dynamic relationship between protein and mRNA but require absolute abundance measurements per cell . As described in ( Schwanhausser et al . , 2011 ) , we used measurements of total mRNA combined with mRNA-seq read density to estimate the absolute abundance of each transcript per cell in untreated MM1 . S cells . For absolute protein abundance we used a recently described method of label-free quantitation termed intensity based absolute quantitation ( iBAQ ) ( Schwanhausser et al . , 2011 ) . From untreated MM1 . S cells we generated absolute abundance estimates for 3369 proteins ( Figure 5—source data 1 ) . Compared to baseline protein abundance , we found a slightly stronger correlation for footprint read density than mRNA ( Figure 5—figure supplement 2 ) . Interestingly , we also found that increased 5′ UTR translation at baseline may correlate with decreased steady-state protein abundance ( Figure 5—figure supplement 2 ) . Overall distributions of protein and mRNA copies per MM1 . S cell , with a median of 11 transcript copies and ∼16 , 000 protein copies ( Figure 5C ) , were similar to those previously found in mouse NIH3T3 cells ( Schwanhausser et al . , 2011 ) . We used these absolute abundance measurements to prepare an approximate quantitative model in our system . We built a mass-action model similar to that in ( Schwanhausser et al . , 2011 ) but modified to incorporate a term to account for global mRNA degradation ( Figures 1B and 5B ) ( see ‘Materials and methods’ for details of modeling ) . This model describes the change in protein copies per cell as a function of transcript copies per cell , the rate of translation , and the rate of protein degradation . We analyzed a subset of 13 genes with transcriptional increase but non-uniform changes in protein levels ( Figure 5D , Figure 5—figure supplement 4 ) . This model was able to fit the protein abundance data , primarily by varying the term ksp , describing the number of proteins produced per transcript per hour ( Figure 5E , Figure 5—figure supplement 4 ) . ksp ranged from 10 to 270 , consistent with the range found previously ( Schwanhausser et al . , 2011 ) . Based on this model , we can begin to reconcile our deep sequencing and proteomic data . During the early stages of apoptosis , the group of transcripts with multi-fold increases in relative abundance do not all lead to measurable protein increases . This is because the absolute number of these transcripts per cell in the sample remains fairly stable as transcription and mRNA degradation offset . This is illustrated by scaling the relative mRNA data shown in Figure 4A by a factor reflecting global mRNA degradation ( Figure 5—figure supplement 3 ) . Notably , the few proteins that are increased tend to demonstrate increases in absolute transcript level at later time points . Our model further shows that with the same fold-increase in transcript level , proteins with low baseline abundance will have more dramatic increases in relative protein concentration when compared to proteins with high baseline abundance ( Figure 5—figure supplement 5 ) . It is unknown how the cellular response to chemotherapeutic treatment influences cellular deconstruction by proteases during apoptosis , including both caspase and non-caspase proteases ( Pop and Salvesen , 2009; Moffitt et al . , 2010 ) . Our laboratory has developed an N-terminomics technology using the engineered enzyme subtiligase to specifically label and enrich for free protein N-termini generated by proteolysis in complex biological samples ( Mahrus et al . , 2008 ) . Others have also identified caspase cleavage substrates after treatment with chemotherapeutics with different N-terminomics approaches ( Gausdal et al . , 2008; Impens et al . , 2008 ) . By MS analysis in a variety of cell lines under both apoptotic and non-apoptotic conditions our laboratory has identified >8000 protein cleavage sites , including >1700 putative caspase substrates with an Asp immediately N-terminal to cleavage site ( P1 position ) . We have compiled these cleavage sites into the ‘DegraBase’ ( Crawford et al . , 2013 ) ; wellslab . ucsf . edu/degrabase ) . We have previously found that different apoptotic inducers result in different patterns of caspase substrate cleavage , though the reasons for these differences remain unknown ( Shimbo et al . , 2012 ) . We can now directly compare whether the cellular response to chemotherapy at the transcriptional and translational level influences the cleavage of substrates during terminal apoptosis . For each Cluster ( Figure 1C ) we queried the DegraBase to investigate if a potential substrate was identified in the DegraBase , if it was cut by caspases , and , if so , the relative frequency of caspase cleavage . If the cellular response to bortezomib induced common anti-apoptotic pathways targeted by the caspases , it was possible that genes in Cluster Upreg would be more prominently featured as substrates in the DegraBase . However , we found no evidence that genes in any Cluster were significantly enriched ( or de-enriched ) among identified caspase substrates ( Figure 6A ) . 10 . 7554/eLife . 01236 . 029Figure 6 . N-terminomics tracks proteolytic cleavage substrates . ( A ) We compared defined hierarchical clusters ( Figure 1C ) to all proteolytic cleavage substrates present in the DegraBase , a database of over 8000 cleavage events ( Crawford et al . , 2013 ) . D at P1 suggests a putative caspase substrate . We find no significant variation in levels of proteolysis by χ2 analysis of each cluster compared to the set of all substrates . ( B ) Representative SRM intensity for appearance of proteolytically-processed peptides , both caspase- and non-caspase-cleaved ( Figure 6—source data 1 ) . Each trace represents a single parent ion/fragment ion transition; co-elution of multiple targeted transitions confirms peptide . ( C ) Heat map showing relative SRM intensity of 252 proteolytic peptides monitored in this system compared to 0 hr baseline . Five groups defined by unsupervised hierarchical clustering representing relative cleavage kinetics show caspase proteolysis generally precedes other apoptotic proteolytic events . DOI: http://dx . doi . org/10 . 7554/eLife . 01236 . 02910 . 7554/eLife . 01236 . 030Figure 6—source data 1 . Degradomics SRM assay . ( A ) SRM intensity data for peptides included in targeted degradomics assay . P1 residue refers to residue immediately N-terminal to identified proteolytic cleavage site based on protein sequence . ( B ) List of parent and fragment ion transitions tracked for targeted degradomic SRM assay . DOI: http://dx . doi . org/10 . 7554/eLife . 01236 . 03010 . 7554/eLife . 01236 . 031Figure 6—figure supplement 1 . N-terminomics biological data subgroups . Log2 expression of N-terminomics SRM data across apoptotic time course . N-terminomics data divided by biological function show that caspase proteolysis proceeds with similar rates across biological subdivisions . Proteins increased at the transcriptional and ribosome footprint level after bortezomib treatment are primarily included in the ‘chaperones and cell stress’ and ‘proteasome and ubiquitination’ groups . The ‘translation’ group demonstrates cleavage of many translation initiation factors . DOI: http://dx . doi . org/10 . 7554/eLife . 01236 . 03110 . 7554/eLife . 01236 . 032Figure 6—figure supplement 2 . N-terminomics data with caspase and non-caspase cleavage sites in same substrate . These data demonstrate that even in the same protein substrates , caspase cleavage tends to proceed more rapidly during apoptosis than non-caspase cleavage . DOI: http://dx . doi . org/10 . 7554/eLife . 01236 . 032 These findings suggest that protease substrates are not targeted based on changes in their abundance . We further explored this hypothesis by monitoring the kinetic appearance of proteolytic substrates with a quantitative SRM assay ( Agard et al . , 2012; Shimbo et al . , 2012 ) . We included both non-caspase and caspase-cleaved peptides in our SRM assay with a preference for those found in multiple experiments under apoptotic conditions . Interestingly , many of the chaperone and co-chaperone proteins , relatively increased at the transcriptional and translational levels , had only non-caspase proteolytic peptides identified in the DegraBase . These included Hsp70-1 , Hsp40 , Hsp27 , BiP ( HSPA5 ) , Hsc70 ( HSPA8 ) , and BAG3 . The highly abundant and constitutively expressed chaperone Hsp90-β had numerous non-caspase cleavages and a single caspase-cleaved peptide . We used this SRM assay to explore whether proteins upregulated in response to bortezomib were more rapidly cleaved during apoptotic proteolysis . In Figure 6B we show example SRM intensity data while Figure 6C displays a heat map of similar SRM intensity data compared to the untreated baseline across all 252 peptides targeted . Caspase-cleaved peptides begin to appear at the 6 hr time point , when caspase activity was also detected biochemically ( Figure 1B ) . In Figure 6—figure supplement 1 we subdivide the monitored peptides based on biological function . We do not find that stress response proteins are cleaved more rapidly than those in other groups . However , we note rapid cleavage of ATF4 ( as seen before , [Shimbo et al . , 2012] ) , BAG3 , and Hsp90-β . It is possible that these particular proteins represent important nodes in deconstructing the cellular stress response after bortezomib . Generally , our results suggest that drug-specific response at the transcriptional and translational level does not broadly affect protease dynamics once apoptosis ensues . Our prior studies examining quantitative appearance of proteolytic substrates only focused on caspase substrates ( Agard et al . , 2012; Shimbo et al . , 2012 ) . Here , we also monitored the kinetics of appearance of non-caspase proteolytic events during apoptosis . We tracked 163 caspase-cleaved peptides ( D at P1 ) and 89 peptides derived from non-caspase proteolysis . The list includes 52 proteins that were cleaved by both caspase- and non-caspase proteases to examine the relative kinetics of different proteolytic events in the same substrate ( Figure 6—figure supplement 2 ) . Unsupervised hierarchical clustering was used to rank the rates of proteolytic cleavage events ( Figure 6C ) . We defined five broad groups ranging from the Fastest cleavages , with robust substrate appearance at 6 hr , to Slowest/No Change , with no increase in cleavage even at late apoptosis . Interestingly , the large majority of non-caspase-cleaved substrates appear in the Slow and Slowest/No Change groups . While this only represents a small portion of the >3 , 000 non-caspase endoproteolytic cleavages in the DegraBase , this finding suggests that most non-caspase proteolytic events in apoptotic cells are a consequence of and not a driver of apoptosis itself . However , we did find 16 non-caspase-cleaved peptides appearing in either the Fastest , Faster , or Mid categories ( Table 2 ) . Of the 89 non-caspase peptides monitored , 24 were tryptic-like ( K or R at P1 ) . Yet none of the 16 most rapidly cleaved peptides had a tryptic-like cleavage and only one has been previously annotated in MEROPS , the largest current database of proteolytic cleavage ( Rawlings et al . , 2012 ) . These 16 peptides were cleaved with kinetics similar to caspases , suggesting that they may play roles in driving apoptosis . Among these we noted a relative preponderance of Pro ( 3 peptides ) , Gln ( 4 peptides ) , and Glu ( 4 peptides ) residues at the P1 site ( Table 2 ) . While caspases may occasionally cleave at Glu ( Krippner-Heidenreich et al . , 2001 ) , none of the other classic proteases thought to be active during MM1 . S cell apoptosis , including the cathepsins or calpains , have a preference for Pro , Gln , or Glu at the P1 site ( as described in MEROPS ) . Our findings suggest that , at least in a limited set of substrates , non-Asp cleavages may be mediated by uncharacterized proteases acting with similar kinetics as the caspases during apoptosis . 10 . 7554/eLife . 01236 . 033Table 2 . Comparison of caspase- and non-caspase cleavages during apoptotic time course as monitored by N-terminomics and SRMDOI: http://dx . doi . org/10 . 7554/eLife . 01236 . 033Total D = P1 peptides163Total non-D = P1 peptides89Tryptic-like ( K or R at P1 ) 24Non-tryptic ( all other residues at P1 ) 65ClusterD at P1Non-D at P1Slowest/no change636Slow2328Mid5010Faster372Fastest384Non-D peptides in fastest three clustersUniProtGeneP1Sequence PositionMEROPS annotated ? Fastest ( C ) GQVAAAAAAQPPASHGPERO95817BAG3C151N ( S ) AVGFNEMEAPTTAYKP14317HCLS1S208N ( P ) GHGSGWAETPRO75533SF3B1P304N ( Q ) VLTVPATDIAEETVISEEPPAKRQ06547GABP1Q306NFast ( Q ) ALKEEPQTVPEMPGETPPLSPIDMESQERP05412JUNQ223N ( P ) AVNGATGHSSSLDARQ07817BCL2L1P63NMid ( E ) AAGATGDAIEPAPPSQGAEAKP49006MARCKSL1E56N ( Q ) AASGDVQTYQIRP16220CREB1Q243N ( E ) GGIDMDAFQERP29083GTF2E1E297N ( A ) SIFGGAKPVDTAARP23588EIF4BA358Y–meprin alpha ( Becker-Pauly et al . , 2011 ) ( E ) AIQNFSFRO75122CLASP2E946N ( P ) HFEPVVPLPDKIEVKP49792RANBP2P1170N ( N ) SWFENAEEDLTDPVRQ13813SPTAN1N2104N ( L ) AFSEQEEHELPVLSRO75995SASH3L128N ( Q ) AIMEMGAVAADKGKKO95817BAG3Q522N ( E ) AILEDEQTQRQ9P2E9RRBP1E1298NBy Fisher′s exact test there is a significant absence of tryptic-like cleavages in the 16 rapidly cleaved peptides ( p=0 . 005 ) . Genes upregulated at both the transcriptional , translational , and protein level may represent the cellular protective response to avoid apoptosis . We thus hypothesized that by inhibiting these ‘first responders’ to the chemotherapeutic treatment we may be able to drive the malignant cell toward apoptosis and reduce therapeutic resistance . To this end we identified the most prominent upstream regulators of the genes in Cluster Upreg ( Table 1 ) . These include NRF2 , a transcription factor regulating the hypoxic response; HSF1 , a transcription factor regulating the heat shock response; and PERK , the ER-resident kinase . HSF1 in particular regulates the transcription of many genes we also found increased by proteomics experiments ( Mendillo et al . , 2012 ) . We therefore explored whether HSF1 inhibition would enhance cell death by bortezomib in MM1 . S cells . We used a recently described small molecule inhibitor of HSF1 , KRIBB11 ( Yoon et al . , 2011 ) . This drug was added with and without a low dose of 2 . 5 nM bortezomib , which alone led to ∼50% cell death over 48 hr . At the 24 hr time point , the combination of KRIBB11 and bortezomib caused significant cell death , while neither compound alone decreased cell viability ( Figure 7A ) . 10 . 7554/eLife . 01236 . 034Figure 7 . Systems-level analysis suggests potential chemotherapeutic combinations with bortezomib in myeloma . ( A ) Bort = bortezomib . HSF1 inhibitor used was KRIBB11 ( Yoon et al . , 2011 ) and shows additive effect with bortezomib . ( B ) PERK inhibitor was GSK2606414 and shows additive effect with bortezomib ( Axten et al . , 2012 ) . All measurements in 24-well plates in triplicate; mean ± SD shown . DOI: http://dx . doi . org/10 . 7554/eLife . 01236 . 034 There are no known small molecules which specifically inhibit NRF2 , but we could test whether PERK inhibition could sensitize cells to death using a newly described small molecule PERK inhibitor , GSK2606414 ( Axten et al . , 2012 ) . Indeed , PERK inhibition did sensitize cells to death by bortezomib ( Figure 7B ) . For both the HSF1 and PERK inhibitors the most robust effects were seen at concentrations in excess of the in vitro IC50 ( ∼1 . 2 µM for KRIBB11 , ∼1 nM for GSK2606414 ) . Notwithstanding potential off-target effects , these results would support the notion that integrated systems-level data can inform potential chemotherapeutic combinations .
Here we present a systems-level study of how malignant cells respond to apoptosis induced by a clinically relevant chemotherapeutic . Our results show how myeloma cells initially struggle to adapt to the proteasome inhibitor bortezomib . A general inhibition of translation coupled with mRNA degradation leads to surprisingly little change in overall protein levels despite relative transcriptional increases in many genes . Furthermore , global protein levels are maintained due to inhibition of proteasomal activity by bortezomib and extensive cleavage of proteasomal subunits by caspases ( Gray et al . , 2010 ) . The cell also modulates translational control of stress responsive proteins related to bortezomib treatment . Co-targeting key regulators of these processes such as HSF1 and PERK can further accelerate apoptosis . Myeloma cells ultimately succumb to the pharmacologic insult , leading to activation of the caspases as well as one or more non-caspase proteases that have yet to be characterized . While pieces of this story have been told by other experiments , the approach we present here , examining the integrated regulation of the transcriptional , translational , and proteolytic machinery , gives a global and in-depth view of the apoptotic process . Translational regulation during apoptosis has only been examined in a limited context . Bushell and coworkers examined the relative changes in translational efficiency during early TRAIL-induced apoptosis by polysome profiling ( microarray analysis of mRNA associated with translating ribosomes ) ( Bushell et al . , 2006 ) . Genes with the greatest degree of decrease in TE included many translation-related proteins , similar to our findings . Those with the largest increase , however , did not overlap with our results , suggesting that positive changes in TE may be specific to the apoptotic inducer or induction of the intrinsic vs extrinsic apoptotic pathways . Importantly , after proteasomal inhibition in myeloma cells we found that translational reprogramming favors proteins which help the cell adapt to unfolded protein stress . Possible mechanisms for the alteration of translation efficiency involve preferential translation via internal ribosome entry sites ( IRES ) ( Bushell et al . , 2006 ) , partitioning of transcripts toward ER-bound ribosomes under stress ( Reid and Nicchitta , 2012 ) , or other uncharacterized regulatory programs governing ribosomal-transcript association . We also identified uORF translation in genes related to the ER stress response , TXNIP , PPP1R15B , and DDIT3 ( REDD1 ) , that had not previously been characterized . The exact interplay of uORF and CDS translation for these genes remains to be investigated , but it is possible there is a regulatory function similar to that already known for ATF4 , CHOP , and GADD34 . We also found that differential 5′ UTR translation appears common in the cellular response to bortezomib but relates variably to increased CDS translation . It will be interesting to compare the effects on translation regulation in different cell types exposed to chemotherapeutics with different mechanisms of action to find whether these changes are unique to bortezomib . Incorporation of translation initiation inhibitors in ribosome profiling and paired-end reads in mRNA-seq can also provide novel insight into translation regulation through generation of custom proteomic databases ( Stern-Ginossar et al . , 2012; Menschaert et al . , 2013; Sheynkman et al . , 2013 ) . Surprisingly , we found that eIF2α phosphorylation did not play a major role in translation inhibition in this apoptotic system involving the UPR . Others have also observed decreased eIF2α phosphorylation in MM1 . S cells after proteasome inhibition ( Parlati et al . , 2009 ) . However , eIF2α phosphorylation is thought to be necessary for increased ATF4 translation ( Lu et al . , 2004 ) . The mechanism leading to the induction of ATF4 in this system ( Figure 1—figure supplement 2 ) is therefore unclear , but may relate to translation of the additional uORF we identified ( Figure 3E ) . Our paired transcriptomic and proteomic data stands in stark contrast to studies of heat and osmolar shock in yeast , where increases at the transcript level are well-correlated with increases in protein ( Lee et al . , 2011; Lackner et al . , 2012 ) . We used a quantitative model incorporating global mRNA degradation to explain this contrast with non-apoptotic systems . While our proteomic analysis is quite extensive , this mass-action model suggests that low abundance proteins not detected by MS , such as CHOP or XBP1 ( Figure 1—figure supplement 2 ) , are more likely to be increased in relative protein abundance after transcriptional changes . Future advances in proteomic techniques may allow us to detect more biologically relevant protein changes . The mass-action model also suggests the kinetics of protein production from mRNA transcripts largely governs protein abundance ( Schwanhausser et al . , 2011 ) . In a system undergoing less rapid apoptosis , such as those explored at the single cell level ( Cohen et al . , 2008; Geva-Zatorsky et al . , 2010 ) , there may be greater opportunity for proteins to accumulate and reveal relationships between ribosome footprint occupancy and protein production . Previous work showed that caspase substrates were largely conserved at the identification level after different chemotherapeutic treatments , though different quantitative signatures of caspase proteolysis could be observed ( Shimbo et al . , 2012 ) . Here we find that in the setting of rapid apoptosis , proteins with changes at the transcriptional and translational level do not appear to be targeted more efficiently by proteases during apoptosis . Under conditions of more prolonged apoptosis , however , changes in the proteome may influence the appearance of proteolytic substrates and this remains to be investigated . Furthermore , our identification of non-caspase proteolytic events occurring with similar kinetics to caspase cleavage suggests other proteases are being activated during apoptosis . The biological role of these proteases during apoptosis awaits exploration . Based on our systems-level analyses , we identified potential small molecules which may be effective in combination with bortezomib in myeloma therapy . Inhibition of HSF1 and PERK have also recently been proposed by others as promising therapeutic targets in myeloma ( Heimberger et al . , 2012; Atkins et al . , 2013 ) and represents a proof of principle for our approach . Our deep sequencing and proteomic results suggest in particular that the combination of bortezomib with HSF1 inhibition could be beneficial in myeloma therapy . We hypothesize that using a similar systems-based approach with less well-studied therapeutics will reveal novel means by which cells attempt to evade apoptosis , not just at the transcriptional level but also at the translational and post-translational levels . This analysis will likely complement information gained by genome-wide RNA knockdown approaches ( Zhu et al . , 2011 ) . By targeting these early response pathways identified at multiple systems levels cancer cell death can be maximized and resistance avoided . Furthermore , changes in specific proteins revealed by these analyses may be monitored after treatment to establish novel biomarkers of chemotherapeutic efficacy or drug resistance . We anticipate that integrated , systems-level approaches , such as that presented here , will help to form the future basis of evaluating preclinical chemotherapeutic effects .
MM1 . S cells ( obtained from ATCC , Manassas , VA , USA ) were grown in suspension to 1 × 106 cells/ml in RPMI-1640 media with 10% FBS . Bortezomib ( LC Laboratories , Woburn , MA , USA ) 20 µM stock solution in sterile-filtered phosphate buffered saline ( PBS ) was simultaneously added to a final concentration of 20 nM to flasks each containing 300 × 106 cells ( PBS only added to control sample ) . At the indicated time point cells were separated into aliquots for each experimental approach ( 15 × 106 cells in duplicate for each of mRNAseq , iTRAQ proteomics , and ribosome footprinting; 200 × 106 cells for N-terminomics ) . Cells for ribosome footprinting alone were incubated at 37°C for 1 min with 100 µg/mL cycloheximide ( Sigma-Aldrich , St . Louis , MO , USA ) from 50 mg/ml stock in 100% EtOH . All cells were washed in PBS ( PBS + 100 µg/ml cycloheximide for ribosome footprint samples ) , pelleted by centrifugation , and flash frozen in liquid N2 , then stored at −80°C . Cell viability and caspase activity were assessed by Cell-Titer Glo and Caspase-Glo ( Promega , Madison , WI , USA ) assays per manufacturer protocol , respectively . KRIBB11 and GSK2606414 were purchased from EMD Millipore ( Billerica , MA , USA ) and 10 mM stock solutions prepared in DMSO and stored in single-use aliquots at −80°C . These drugs and bortezomib were added as indicated in triplicate in 24-well plates at an MM1 . S cell density of 0 . 5 × 106 cells/ml . Duplicate samples of 4 × 106 MM1 . S cells at each time point after 20 nM bortezomib treatment were pelleted at 800×g for 5 min , supernatant aspirated , resuspended in ice-cold PBS , pelleted again , and supernatant aspirated . Total RNA was extracted using Trizol reagent ( Invitrogen , Carlsbad , CA , USA ) following manufacturer’s protocol; lysis was performed by passing 20x through a 22½ gauge needle . Total RNA was resuspended in 100 µl RNAse-free water and further purified using RNAeasy kit ( Qiagen , Germantown , MD ) with on-column DNAse I treatment per manufacturer’s protocol . Purified RNA pellet was resuspended in 100 µl RNAse-free water . Total RNA concentration ( as shown in Figure 1B ) was measured spectrophotometrically using a NanoDrop ND-1000 UV-Vis spectrophotometer ( Thermo Fisher , Waltham , MA , USA ) . mRNA was further purified by poly ( A ) separation using Oligo ( dT ) 25 Magnetic Beads kit ( New England BioLabs , Ipswich , MA , USA ) . Total RNA samples were diluted with 500 µl of kit Lysis/Binding buffer and bound to of equilibrated beads ( 100 µl bead slurry used ) . Beads were washed once in 500 µl Wash Buffer one followed by one wash in 500 µl Wash Buffer two . mRNA was eluted in 40 µl Elution Buffer and concentration measured by NanoDrop . A separate time course of MM1 . S cells was prepared and treated with 20 nM bortezomib as described above 10 × 106 MM1 . S cells harvested at indicated time points were lysed in RIPA buffer ( 50 mM Tris-HCl , 150 mM NaCl , 1% NP-40 , 0 . 5% sodium deoxycholate , 0 . 1% SDS , pH 8 . 0 ) supplemented with 5 mM EDTA and 1x protease inhibitor cocktail ( Cell Signaling Technology , Danvers , MA , USA ) . Protein concentration from cell lysates was measured by BCA assay ( Thermo Fisher , Waltham , MA , USA ) . ∼25 µg of total protein at each time point was separated on NuPAGE 4–12% Bis-Tris polyacrylmide gels ( Invitrogen ) and transferred to 0 . 45 µM PVDF membrane . Primary antibodies used for immunoblotting were obtained from Cell Signaling Technology ( α-CHOP , α-XIAP , α-Bid , α-PERK , α-4E-BP1 , α-phospho ( Thr37/46 ) -4E-BP1 , α-eIF2α , α-phospho ( Ser51 ) -eIF2α ) , Santa Cruz Biotechnology ( Santa Cruz , CA , USA ) ( α-CREB2 [ATF4] ) , BioLegend ( San Diego , CA , USA ) ( α-XBP1s ) , and Sigma ( α-β-actin ) . Harvested cell pellets for ribosome profiling were suspended and lysed in 500 µl ice-cold polysome lysis buffer ( 20 mM Tris , pH 7 . 4 , 250 mM NaCl , 15 mM MgCl2 , 1 mM dithiothreitol , 0 . 5% Triton X-100 , 24 U/ml Turbo DNase ( Ambion , Austin , TX , USA ) , and 100 µg/ml cycloheximide ) by repeated pipetting . Lysate was clarified by centrifugation for 10 min at 20 , 000×g at 4°C . 3 µl of RNase I 100 U/μl ( Ambion ) was added and sample incubated for 45 min at room temperature . The digestion was stopped by the addition of 10 μl SuperaseIn 20 U/μl ( Ambion ) . Digested samples was then loaded onto a 1 M sucrose cushion , prepared in polysome buffer plus 0 . 1 U/μl SuperaseIn . Ribosomes were pelleted by centrifugation for 4 hr at 70 , 000 rpm , 4°C in a TLA-110 rotor . The liquid was removed and the pellet was resuspended in 600 μl 10 mM Tris ( pH 7 ) , followed by the immediate addition of 40 μl 20% SDS . The sample was heated to 65°C and RNA was extracted using two rounds of acid phenol/chloroform followed by chloroform alone . RNA was precipitated from the aqueous phase by adding sodium acetate to a final concentration of 300 mM followed by at least one volume of isopropanol . Precipitation was carried out at −20°C overnight and RNA was then pelleted by centrifugation for 30 min at 20 , 000×g , 4°C . The supernatant was discarded , the pellet was air-dried , and the RNA was resuspended in 25 μl Tris ( pH 7 ) . RNA was separated by gel electrophoresis on a 15% TBE-Urea gel ( Invitrogen ) and gel fragments extracted corresponding to ∼25–35 nt in size . RNA was extracted from gel as in Ingolia et al . ( 2011 ) . Harvested cell pellets for mRNAseq were lysed by repeated pipetting in Trizol ( Invitrogen ) and total RNA isolated per manufacturer protocol . Poly ( A ) mRNA was purified from the total RNA sample using ( dT ) 25 DynaBeads ( Invitrogen ) per manufacturer protocol . mRNA was fragmented in high pH buffer ( 50 mM NaCO3 , pH 9 . 2 ) for 20 min at 95°C , then precipitated and separated by gel electrophoresis as above . mRNA fragments of 50–90 nt were extracted . Footprint and mRNA sample purity and fragment size were verified by Bioanalyzer 2100 ( Agilent , Santa Clara , CA , USA ) . RNA samples were dephosphorylated , ligated to linker , and separated by gel electrophoresis as described previously ( Ingolia et al . , 2011 ) . Ribosome footprint samples were enriched by subtractive hybridization of contaminating rRNA sequences . We used biotin-modified oligonucletoides and capture on streptavidin-coated beads; oligonucleotide sequences complimentary to rRNA were identical to Stern-Ginossar et al . ( 2012 ) . Reverse transcription and cDNA library preparation were completed as in Ingolia et al . ( 2011 ) . Sequencing was performed on an Illumina HiSeq 2000 using single end , 50-bp reads . Before alignment , linker and poly ( A ) sequences were computationally removed from the 3′ ends of raw sequencing reads . Bowtie v0 . 12 . 8 ( Langmead et al . , 2009 ) ( allowing up to two mismatches ) was used to perform the alignments with up to two mismatches allowed . For footprint data only reads of length 26–36 nt ( footprint length with cycloheximide ( Ingolia et al . , 2011 ) ) were used for alignment . Reads were first aligned vs human rRNA and tRNA sequences; aligned reads were discarded . All remaining reads were next aligned to known canonical transcripts in hg19 ( downloaded from genome . ucsc . edu 23 May 2012 ) . Remaining unaligned reads were aligned to the hg19 genome . We used in-house C++ scripts to assign and count unique reads mapping to canonical hg19 transcripts . Only uniquely mapping reads were used for further analysis . The midpoint of footprint reads was used to assign a unique nucleotide location for that read . mRNA-seq reads were assigned a unique nucleotide position at the 5′ end of the read . mRNA and footprint read density were calculated in units of reads per kilobase million ( RPKM ) to normalize for gene length and total reads per sequencing run . Raw sequencing data is available in the GEO repository with accession number GSE48785 . Selected sequencing reads were visualized through Integrative Genomics Viewer 2 . 0 ( Robinson et al . , 2011 ) . Metagene analysis was performed as in Ingolia et al . ( 2009 ) . 5′ UTR read assignments were based on canonical transcript splice isoform as in hg19 . For UTR analysis normalized read counts ( in RPKM ) were used to compare relative 5′ UTR vs CDS translation across time points . For Figure 1C , unsupervised hierarchical clustering was performed using centroid linkage across mRNA , footprint , and translation efficiency data with Spearman-Rank correlation in Cluster 3 . 0 and visualized in TreeView . All aligned transcripts were initially assigned to a unique UCSC gene ID . Conversion tables available from UCSC through the kgXref function were used to convert UCSC gene ID to HUGO gene nomenclature and to UniProt accession number for further analysis . Where indicated , gene lists were analyzed by Ingenuity Pathway Analysis ( Ingenuity Systems , Redwood City , CA , USA ) using default settings for included genes and interaction networks . iTRAQ sample preparation was performed in a similar manner to Mertins et al . ( 2012 ) with some modifications . 15 × 106 cells harvested at each time point ( as above ) were thawed and lysed by sonication in 250 µl lysis buffer ( 7M Guanidine HCl , 75 mM NaCl , 100 mM bicine pH 8 . 0 ) supplemented with protease inhibitors ( 500 µM 4- ( 2-Aminoethyl ) benzenesulfonyl fluoride HCl , 1 mM E-64 , 1 mM phenylmethylsulfonyl fluoride , and 1 mM EDTA ) . Lysates were cleared by centrifugation at 16 , 500×g for 10 min and protein concentration was measured by BCA assay . Lysate containing 250 µg of protein was diluted to 100 µl with lysis buffer . Disulfide bonds were reduced with 5 mM dithiothreitol and cysteines alkylated with 10 mM iodoacetimide . Lysate was diluted 1:8 with trypsin dilution buffer ( 100 mM bicine pH 8 . 0 , 1 mM CaCl2 , 75 mM NaCl ) . Sequencing grade modified trypsin ( Promega ) was added at an enzyme-to-substrate ratio of 1:25 . Trypsin digestion took place for 16 hr with agitation at room temperature . Samples were acidified with trifluoroacetic acid to a final concentration of 0 . 5% to halt digestion . Tryptic peptides were desalted on Waters SepPak C18 ( Milford , MA , USA ) columns and evaporated to dryness on a vacuum concentrator . Peptide samples were resuspended in 40 µl 0 . 5 M TEAB pH 8 . 5 . Peptide concentration was determined by BCA assay . Labeling with 8-plex iTRAQ reagent was performed per manufacturer instructions ( AB Sciex , Framingham , MA , USA ) : briefly , 100 µg of peptides at each time point was diluted to 30 µl with 0 . 5 M TEAB; 6-plex iTRAQ reagent was resuspended in 70 µl EtOH . Peptides were mixed with reagent and incubated for 1 hr at room temperature . The reaction was halted using 50 mM Tris/HCl pH 7 . 5 . The labeled reaction products for each sample were combined and evaporated to dryness . Combined labeled peptides were desalted using Waters SepPak C18 cartridge . 0 hr samples were labeled with 113 mass tag; 1 . 5 hr , 115; 3 hr , 116; 6 hr , 117; 9 hr , 118; 12 hr , 119 . Peptides were resuspended in 0 . 1% formic acid and then separated into 30 fractions via reverse phase high-pH fractionation using an XBridge C18 column ( 1 . 0 × 100 mm , 3 . 5 μm; Waters ) on a Waters 2796 BioSeparations Module HPLC . We used a 70 min gradient with a linear increase from 2% to 38% acetonitrile in water and constant 10% ammonium bicarbonate , pH 10 . 5; flow rate was 50 µl/min . Fractions were evaporated to dryness and stored at −80°C , then resuspended in 0 . 1% formic acid prior to mass spectrometry analysis . The combined iTRAQ sample was then analyzed using two separate mass spectrometers . The first was a QSTAR Elite QqTOF mass spectrometer ( AB Sciex ) coupled in-line to an Eksigent NLC-1DV-500 HPLC and in-house packed 75-µm × 15-cm C18 column . For each fraction in-line LC was performed in Buffer A ( 0 . 1% formic acid in water ) and Buffer B ( 0 . 1% formic acid in acetonitrile ) at a flow rate of 250 nl/min . The full method extended over 125 min with a linear gradient from 3% to 40% Buffer B over 90 min , increased to 75% B over 5 min and held for 15 min , then re-equilibration at 3% B for 15 min . Data-dependent acquisition was performed using Analyst v . 2 . 0 software ( AB Sciex ) across an m/z range of 385–1200 . We used an isolation window of 100 mDa for ions selected for MS/MS with a dynamic exclusion window of 60 s after acquisition . The two most intense ions in each MS1 scan were chosen for CID fragmentation and sequencing . The ‘iTRAQ’ option was chosen within the instrument software to enhance MS2 collision energy and increase intensity of iTRAQ reporter signal . Each fraction of the combined iTRAQ sample was also analyzed on an LTQ Orbitrap Velos mass spectrometer ( Thermo Fisher Scientific ) coupled in-line to a nanoAcquity UPLC system ( Waters ) . Injected samples were trapped on a Symmetry C18 Column ( 0 . 18 × 20 mm , 5 μm; Waters ) for 5 min at 1% of Buffer B before starting the gradient; an analytical BEH130 C18 column ( 0 . 075 × 200 mm column , 1 . 7 μm; Waters ) was used with a flow rate of 600 nl/min . A linear gradient to 38% Buffer B was run over 100 min , then increased to 75% B over 10 min , increased to 90% B over 5 min , then decreased to 2% B and re-equilibrated for 15 min ( total method 130 min ) . Data-dependent acquisition was performed using the Xcalibur 2 . 1 software in positive ion mode at a spray voltage of 2 . 5 kV . MS1 survey spectra were acquired in the Orbitrap with a resolution of 60 , 000 and a mass range from 300 to 1400 m/z . All iTRAQ data was acquired by fragmenting the eight most intense ions per cycle in higher-energy collisional dissociation ( HCD ) mode . Collision energy was set to 45 , maximum inject time was 250 ms and maximum ion count was 1 × 105 counts . We used an isolation window of 2 . 3 Th for ions selected for MS/MS . Ions selected for MS/MS were dynamically excluded for 60 s after acquisition . Peptide identification was performed using Protein Prospector ( v . 5 . 9 ) ( University of California , San Francisco ) . All spectra were searched vs the full human SwissProt database ( downloaded 21 March 2012 ) with reverse sequence database as implemented in Protein Prospector for decoy matches . Search parameters included: Fixed modifications carbamidomethyl ( cysteine ) , iTRAQ-8plex ( N-terminus ) , iTRAQ-8plex ( lysine ) ; Variable modifications methione-loss ( N-terminus ) and methionine oxidation; two missed tryptic cleavages allowed; for Orbitrap Velos data , parent and fragment mass tolerance of 20 ppm was used; for QSTAR Elite data , parent mass tolerance was 50 ppm and fragment mass tolerance was 150 ppm . Expectation value thresholds for peptide identification were modified to maintain a false discovery rate <1% based on the number of spectra matching to reverse decoy sequences . iTRAQ signal was quantified using the Search Compare function of Protein Prospector , which extracts the peak intensity of the MS2 signal in the iTRAQ label range . Only peptides with an iTRAQ signal >300 cps at the 0 hr time point were included for further analysis to reduce noise in quantification . Only peptides uniquely matching to a single entry in the human SwissProt database were included . Proteins were only included with a minimum of two unique peptides matched . Protein quantification was performed by summation of iTRAQ signal across all peptides assigned to that protein at each time point . Total log2 iTRAQ signal intensity across all peptides at a time point was median-normalized for comparison across time points . We attempted to develop orthogonal , targeted , label-free Selected reaction monitoring ( SRM ) assays for 250 proteins that were identified in iTRAQ experiments . We primarily concentrated on proteins with relative increases in mRNA expression as found in deep sequencing experiments . Based on prior yeast studies ( Lee et al . , 2011 ) , these would be expected to most likely result in protein changes after perturbation . We used the open-source software Skyline ( v1 . 3 ) ( MacLean et al . , 2010 ) to first build an unscheduled SRM method . Building these methods requires a library of MS/MS spectra for inclusion of the most intense transition pairs ( m/z data for parent ion and sequence fragment ion ) for a given peptide . We used a spectral database of tryptic peptides matched to >8 , 000 proteins from human cancer cell lines , as identified by extensive fractionation and analysis in HCD mode on the LTQ Orbitrap Velos ( AU , J Oses-Prieto , ALB , unpublished data ) . We selected a minimum of two and up to the seven most intense peptides present in the spectral library for each protein target . For each peptide we used the most intense seven transitions ( y- and b-series ions ) for initial measurement by unscheduled SRM . Selected peptides were chosen independently of those found by iTRAQ analysis . All SRM analysis was carried out on an AB Sciex QTRAP 5500 triple quadrupole mass spectrometer interfaced in-line with a nanoAcquity UPLC system ( Waters ) identical to that on the LTQ Orbitrap Velos ( Trapping column: Symmetry C18 Column ( 0 . 18 × 20 mm , 5 μm; Waters ) ; Analytical column: BEH130 C18 column ( 0 . 075 × 200 mm column , 1 . 7 μm; Waters ) ) . We injected ∼500 ng of tryptic peptides from MM1 . S cells on to the mass spectrometer with the following conditions: Trapping for 4 min in 1% buffer B at 5 µl/min , then after injection a linear gradient from 3%–35% B over 80 min , an increase to 90% B over 5 min , then held for 5 min , then a decrease to 3% B for 10 min ( total run time 100 min ) . Unit resolution was used at Q1 and Q3 . A three second cycle time was used for all runs . For unscheduled runs a 10 ms acquisition time was used per transition . Multiple injections were used to test for all targeted peptides . Using data analysis in SkyLine software , peptides were selected for further method development based on ( 1 ) the signal detection ( above baseline ) of at least 5 of 7 co-eluting transitions and ( 2 ) a retention time within 7 min of that acquired in the initial spectral library ( acquired under the same chromatographic conditions ) . Peptides chosen for further development were then limited to the four most intense transitions as found in unscheduled runs . A scheduled SRM method was developed with a retention time window of ±5 min . The four most intense transitions were first subject to collision energy ( CE ) optimization in ±4 V steps to enhance intensity of detected transitions . CE-optimization data from QTRAP 5500 on the same samples was again analyzed in SkyLine . Peptides were confidently assigned based on the identification of four out of four targeted transitions co-eluting at the appropriate retention time . Ultimately , our final SRM method included from one to three confidently-assigned peptides matching to 152 proteins ( all transitions listed in Figure 4—source data 1 ) with the CE for each transition used which maximized detected intensity . We prepared an independent time course of MM1 . S cells exposed to 20 nM bortezomib and harvested at 0 hr , 1 . 5 hr , 3 hr , 6 hr , 9 hr , and 12 hr . Cell viability and caspase activity assays were similar to that shown in Figure 1B . We prepared tryptic peptides from 15 × 106 cells at each time point ( as described above prior to iTRAQ labeling ) . ∼500 ng of tryptic peptides were injected in triplicate and analyzed using the final scheduled SRM assay . Peptide intensity in each sample was measured as the sum of all transition peak areas for that peptide ( as measured by analysis in SkyLine ) . For peptides not clearly identified at all time points , a retention time window was used identical to that in other injections with background signal integrated and used for analysis . To normalize peptide concentration across samples , we used peptides derived from a set of high abundance proteins not expected to significantly change during the time course ( actin , tubulin , and filamin-A ) . We derived an index based on the geometric mean intensity of peptides from these ‘housekeeping’ proteins and scaled SRM intensity of all peptides in each sample based on the median value of this index . Corrected peptide intensity was averaged across injections for each sample . Protein intensity was measured as the sum of peptide intensity for each sample . For mRNA copy number estimation we used an approach similar to that used in Schwannhausser et al . ( 2011 ) , described initially by Mortazavi et al . ( 2008 ) . Total mRNA measured in duplicate by NanoDrop in the untreated MM1 . S sample ( 0 . 88 ± 0 . 09 µg ) was divided by 4 × 106 cells to estimate the mRNA yield per cell ( 2 . 19 × 10−13 g ) . mRNA sequencing reads were approximately evenly distributed across A:C:G:T; the average molecular weight of a RNA monophosphate nucleotide ( averaged across AMP , CMP , GMP , and UMP ) is 339 . 5 g/mol . We can thus calculate the total number of mRNA nucleotides per cell ( T; 6 . 40 × 10−15 mol/cell ) . The copy numbers of individual mRNAs ( c ) can be calculated by the total number of sequencing reads which mapped uniquely to a given transcript ( rtranscript ) , the total reads ( rtotal; 8 , 744 , 024 reads uniquely aligning to hg19 transcripts for the 0 hr MM1 . S sample ) , and the transcript length ( L ) by the equation:rtranscriptrtotal=c×LT . For protein copy number estimation , we harvested and prepared tryptic peptides ( as described above prior to iTRAQ labeling ) from 15 × 106 untreated MM1 . S cells . Peptides were separated into 20 fractions by reverse-phase high pH chromatography ( same protocol as above for iTRAQ peptides ) . Each fraction was analyzed separately in HCD mode on the LTQ Orbitrap Velos Mass Spectrometer and nanoAcquity UPLC as above ( gradient: 60 min linear gradient from 3% to 30% buffer B , 5 min gradient to 90% B , held 10 min , re-equilibrate 15 min at 3% B; mass spectrometer settings same as for iTRAQ analysis except collision energy at 35% and ion accumulation at 1 × 106 ) . We analyzed the MS data using the MaxQuant software package ( v . 1 . 3 . 0 . 5 ) ( Cox and Mann , 2008 ) with Carbamidomethyl cysteine as a fixed modification; Methionine oxidation and N-terminal acetylation as variable modifications; initial parent and fragment MS/MS tolerance of 20 ppm; a minimum peptide length of seven; one missed tryptic site; and razor peptides used for iBAQ quantification . Peptides were searched against the full Uniprot Human Proteome database ( downloaded November 10 , 2012 ) and contaminants enabled in MaxQuant; maximum allowable peptide and protein false discovery rate was 1% as searched against reverse sequence library based on Uniprot database . In MaxQuant peptide matches were assigned to protein groups ( a cluster of a base protein plus additional proteins matching to a subset of the same peptides ) . Protein groups matching the reverse database or contaminants were discarded . iBAQ quantification in MaxQuant evaluates the intensity of each protein in the sample as the sum of all the peptide intensities in the MS1 scan for peptides which matched to that protein group . iBAQ analysis and database searching resulted in assignment of intensity to 3390 protein groups with a minimum of one assigned peptide . We assigned each protein group to an individual protein based on the primary UniProt ID associated with the analysis output . We used the total iBAQ signal in the sample ( Itotal ) , across all matched peptides , as a measure of the total cellular protein signal as measured on the mass spectrometer . We divided the measurement of total protein in the sample ( from BCA assay ) by 10 × 106 cells to obtain an estimate of total protein per cell ( gtotal; 5 . 28 × 10−11 g ) . We then calculated the grams of protein in the cell comprised of an individual protein ( gprotein ) based on the iBAQ signal assigned to that protein ( Iprotein ) as a fraction of total iBAQ signal . gproteingtotal=IproteinItotalUsing the grams of an individual protein per cell , the molecular weight of the assigned protein in g/mol , and Avagadro’s number , we assign an estimated number of copies per protein per cell . These measures of absolute copies of mRNA transcripts and proteins per cell are clearly rough approximations . In an important initial comparison , the median of the distributions for both mRNA and protein copy per cell are of the same order of magnitude as , though somewhat lower than , those found in Schwanhausser et al . ( 2013 ) . This may reflect differences in the cell line ( mouse NIH3T3 vs human MM1 . S ) as well as differences in measurement . However , the overall similarity of the distributions found here to those found by Schwanhausser et al . ( 2013 ) indicates that our data are sufficient as a guide for our mass-action model ( below ) . We further emphasize that our data are not intended as a definitive measure of mRNA or protein copy numbers in MM1 . S cells . Even if there are systematic errors in the quantitation ( i . e . , if true protein copies per cell are all threefold higher ) , the fitting parameters for our mass-action model ( see below ) will be uniformly scaled but the overall interpretation of the model will be unchanged . This issue of scaling was illustrated in the corrigendum to the original manuscript by Schwanhausser et al . ( 2013 ) . The original paper ( Schwanhausser et al . , 2011 ) used the incorrect standard for protein abundance estimation , leading to threefold in changes in the protein distribution; the corrected scale led to an increase in the fit parameter ksp but no changes to the model or conclusions of the paper . To model the conversion of mRNA to protein in this system we used a mass-action model of translation , as originally described in Hargrove and Schmidt ( 1989 ) , and further used by Schwanhausser et al . ( 2011 ) , as an ordinary differential equation:dPdt=kspR ( t ) −kdpP ( t ) Here the number of protein copies ( P ( t ) ) changes over time as a function of the number of mRNA transcripts ( R ( t ) ) , the translation rate constant ksp , and the protein degradation rate constant kdp . In our system , we first modeled R ( t ) using the normalized sequencing reads ( RPKM ) at later time points and taking the ratio to the RPKM at 0 hr . We then multiplied this RPKM ratio at each time point by the absolute number of mRNA transcripts at 0 hr as measured above . We focused on analyzing only a subset of 13 transcripts with increased relative mRNA abundance but with variable detected changes at the protein level ( i . e . , some with increased protein , some with no protein change ) . We plot this data in Figure 5D , Figure 5—figure supplement 4A , C . We found that for this subset of transcripts , this mRNA transcript data could be well-approximated by fitting either a quadratic ( FTL , SQSTM1 , CTSD ) or 4-parameter sigmoid ( all others ) . This fit function ( fitting performed in GraphPad Prism software ) describes R ( t ) . However , as discussed in the main text , this function R ( t ) is based on relative transcript expression at different time points . This would be an effective measure of mRNA transcripts per cell in the setting of constant total mRNA concentration . In this system , though , we find that mRNA is significantly degraded at later time points ( Figure 1B ) . Therefore , we fit an additional function G ( t ) to this mRNA data as a four-parameter sigmoid fit to describe this global mRNA degradation . By incorporating G ( t ) into the mass-action model above , assuming similar degradation rates of each transcript during apoptosis ( as described by Del Prete et al . ( 2002 ) ) , the combined term R ( t ) G ( t ) now represents the absolute number of mRNA transcripts per cell in our system:dPdt=kspR ( t ) G ( t ) −kdpP ( t ) . To estimate absolute protein copies per cell at later time points , we used the label-free SRM intensity data ( as this does not suffer from compression ratio artifacts as seen in iTRAQ data ) . We took the ratio of SRM intensity at later time points to that at 0 hr , then multiplied by the absolute copies protein copies per untreated MM1 . S cell as measured by iBAQ . We scaled the 12 hr SRM data to reflect the decreased protein concentration in the sample . We used Mathematica ( v . 9 ) software to numerically solve the differential equation above for each of the 13 genes chosen for analysis . We fit this differential equation to the plotted protein data ( as shown in Figure 5E , Figure 5—figure supplement 4B , D ) . This equation has three free fitting parameters to describe the six protein data points: protein copies at 0 hr , ksp , and kdp . Protein copies at 0 hr was varied a maximum of 10% from the measured protein copies per cell by iBAQ . kdp was assumed to be relatively constant for all proteins in the background of proteasomal blockade by bortezomib . We only varied the range between 0 . 010–0 . 015 hr−1 , similar to the corresponding average protein half-life found in Schwanhausser et al . ( 2011 ) . Therefore , we primarily varied the translation rate constant ksp in order to fit this model to the protein data . We performed the fits by manually minimizing the least-squares difference between model fit and protein data in Mathematica . We found that the model could indeed well-describe the protein data; in addition , the fit ksp parameter ranged from 10 to 270 proteins/transcript/hr , consistent with the range in Schwanhausser et al . ( 2011 ) . As described in Shimbo et al . ( 2012 ) , cell pellets were lysed by sonication in 4 . 0% SDS , 400 mM Bicine ( pH8 . 0 ) , and supplemented with protease inhibitors 0 . 1 mM z-VAD-fmk , 0 . 1 mM E-64 , 1 mM AEBSF , 1 mM PMSF , and 5 mM EDTA . Spike-in internal standards were added to each sample after lysis: 50 µg each of bovine catalase and yeast alcohol dehydrogenase ( Sigma ) . Proteins were reduced by TCEP , cysteines alklyated by iodoacetamide , and free N-termini biotinylated via the reaction of TEVest4 ester and subtiligase enzyme as previously described ( Shimbo et al . , 2012 ) . Biotinylated peptides were precipitated , resuspended in 5 . 3 M Guanidine HCl and captured on NeutraAvidin agarose beads ( Thermo ) . The beads were extensively washed in 5 M Guanidine HCl followed by on-bead trypsin digestion overnight at room temperature with agitation . As previously described , captured peptides were released from beads by incubation with TEV protease ( Shimbo et al . , 2012 ) . Peptides were desalted using C18 ZipTips ( Millipore ) , evaporated to dryness , and stored at −80°C for analysis . SRM assays were developed in a similar manner as described above , again using SkyLine software . In this case , however , targeted parent/fragment ion transitions were extracted from MS/MS spectra in the DegraBase . This publicly available database lists N-terminally labeled peptides found in human cell culture under both apoptotic and non-apoptotic conditions ( wellslab . ucsf . edu/degrabase ) . b-ion fragment m/z was modified as necessary to reflect the Abu- tag at the N-terminus of the peptide ( Shimbo et al . , 2012 ) . We initially chose 400 peptides for development of targeted SRM assays in SkyLine . We used the same chromatography and instrument settings as described for the earlier SRM assays . We initially examined seven transitions per peptide; we moved forward with peptides with a minimum of five co-eluting transitions and a retention time within 7 min of that predicted based on the peptide retention time prediction algorithm ( SSRCalc 3 . 0 ) implemented in SkyLine . Ultimately , we included 252 proteolytic peptides in the final SRM assay . We corrected samples for labeling efficiency and overall peptide concentration using five proteolytic peptides derived from the spike-in standards . Corrected peptide intensities were averaged across duplicate injections at each time point in the 20 nM bortezomib time course . These mean intensity values were used for analysis . | Many cancer treatments work by causing cancer cells to enter an advanced stage of a process known as programmed cell death or apoptosis . When a cell begins apoptosis , it takes a series of metabolic steps–such as fragmenting its DNA or reducing its volume–that eventually kills it . The cancer cells in tumours are able to grow because they are able to avoid apoptosis . When cancer cells are treated with cytotoxic drugs they do not die immediately but try to stave off the effect of the drug . However , we still know relatively little about what happens at the molecular levels as cancer cells struggle to avoid apoptosis . Now Wiita et al . have combined two methods for studying cancer cells–deep sequencing of RNA and quantitative proteomics–to simultaneously observe a variety of processes , including the transcription of genes to produce messenger RNA ( mRNA ) molecules , the translation of these mRNA molecules to produce proteins , and the proteolysis ( or breakdown ) of these proteins when the cells were subjected to chemotherapy . Wiita et al . studied how human myeloma cells responded to bortezomib , a drug that is used to treat various blood cancers , and found that ribosomes–the complex molecular machines that perform the translation step– reacted to the chemotherapy by preferentially translating certain mRNA molecules in order to produce a set of proteins that protect the cell . Developing drugs to inhibit the effects of these stress-response proteins could make the cancer cells more responsive to existing anticancer drugs . When this effort to stay alive is ultimately unsuccessful , the destruction of proteins appears surprisingly unrelated to the previous attempts that were made to protect the cell . With further work the “global cellular response” approach developed by Wiita et al . could lead to the discovery of new drug targets , improve our understanding of drug resistance in chemotherapy , and provide new ways to monitor how patients respond to treatment . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"biochemistry",
"and",
"chemical",
"biology"
] | 2013 | Global cellular response to chemotherapy-induced apoptosis |
Adolescents are especially prone to drug addiction , but the underlying biological basis of their increased vulnerability remains unknown . We reveal that translational control by phosphorylation of the translation initiation factor eIF2α ( p-eIF2α ) accounts for adolescent hypersensitivity to cocaine . In adolescent ( but not adult ) mice , a low dose of cocaine reduced p-eIF2α in the ventral tegmental area ( VTA ) , potentiated synaptic inputs to VTA dopaminergic neurons , and induced drug-reinforced behavior . Like adolescents , adult mice with reduced p-eIF2α-mediated translational control were more susceptible to cocaine-induced synaptic potentiation and behavior . Conversely , like adults , adolescent mice with increased p-eIF2α became more resistant to cocaine's effects . Accordingly , metabotropic glutamate receptor-mediated long-term depression ( mGluR-LTD ) —whose disruption is postulated to increase vulnerability to drug addiction—was impaired in both adolescent mice and adult mice with reduced p-eIF2α mediated translation . Thus , during addiction , cocaine hijacks translational control by p-eIF2α , initiating synaptic potentiation and addiction-related behaviors . These insights may hold promise for new treatments for addiction .
In humans , adolescence is a period of heightened susceptibility to drug addiction ( Chambers et al . , 2003; Kandel et al . , 1992 ) . Although some molecular and cellular adaptations associated with drug use have been identified ( Bowers et al . , 2010; Lüscher and Malenka , 2011 ) , the biological basis of heightened vulnerability to substance abuse during adolescence remains elusive . Converging evidence supports the notion that addictive drugs hijack the cellular and molecular mechanisms underlying long-term changes in synaptic strength in the mesocorticolimbic dopamine ( DA ) system ( including the ventral tegmental area ( VTA ) , a key brain reward area implicated in the development of addiction ( Kauer , 2004 ) ) in a way that reinforces drug-seeking behavior ( Bowers et al . , 2010; Kauer and Malenka , 2007; Hyman et al . , 2006 ) . Addiction has both initiation and maintenance phases ( Lüscher and Malenka , 2011; Wise and Koob , 2014 ) . Here we focus on the molecular mechanisms underpinning the initial neuronal circuit adaptations caused by addictive drugs because they represent important targets for therapeutic interventions ( Lüscher and Malenka , 2011 ) and are thought to contribute to the development of drug addiction ( Lüscher and Malenka , 2011; Kauer and Malenka , 2007; Kalivas et al . , 2009; Lammel et al . , 2014 ) . For instance , drugs of abuse ( including cocaine , amphetamine , nicotine , ethanol , and morphine ) all induce long-term potentiation ( LTP ) of excitatory synapses on VTA DA neurons that lasts for several days after exposure ( Bowers et al . , 2010; Lammel et al . , 2014; Ungless et al . , 2001; Saal et al . , 2003 ) . This LTP , resulting from the insertion of α-amino-3-hydroxy-5-methyl-4-isoxazolepropionic acid receptors ( AMPARs ) in the postsynaptic membrane , is measured by recording glutamatergic synaptic currents ( EPSCs ) at positive holding potentials , and is manifested as an increase in the AMPAR/N-methyl D-aspartate receptor ( NMDAR ) ratio ( Ungless et al . , 2001 ) . Furthermore , metabotropic glutamate receptor-mediated long-term depression ( mGluR-LTD ) , resulting from the removal of postsynaptic AMPARs , blocks cocaine-induced LTP in VTA DA neurons ( Bellone and Lüscher , 2006 ) . Thus , it has been postulated that impaired mGluR-LTD increases vulnerability to drugs of abuse ( Bellone and Lüscher , 2006; Lüscher and Huber , 2010; Loweth et al . , 2013 ) . In VTA DA neurons , protein synthesis is required for cocaine-induced LTP ( Argilli et al . , 2008; Yuan et al . , 2013 ) and mGluR-LTD ( Mameli et al . , 2007 ) . In addition , protein synthesis is also required for cocaine-induced behaviors ( Sorg and Ulibarri , 1995; Kuo et al . , 2007 ) . Protein synthesis encompasses three steps: initiation , elongation , and termination . Initiation is the rate limiting step and a major target for translational control ( Sonenberg and Hinnebusch , 2009; Buffington et al . , 2014 ) . There are two main mechanisms by which translation initiation is controlled . The first is by regulation of the eIF4F complex via the mechanistic target of rapamycin complex 1 ( mTORC1 ) . The second mechanism is by regulating ternary complex formation via phosphorylation of the translation initiation factor eIF2α . Phosphorylation of eIF2α blocks general translation , but also results in translational up-regulation of a small subset of select mRNAs that contain upstream open reading frames ( uORFs ) in their 5’ untranslated region ( 5’UTR ) ( Sonenberg and Hinnebusch , 2009; Buffington et al . , 2014 ) . Here we report a new mechanism underlying adolescent hypersensitivity to the synaptic and behavioral effects of cocaine . In particular , we show that drugs of abuse selectively hijack the translational program controlled by phosphorylation of eIF2α in the VTA , thus potentiating synaptic inputs to VTA DA neurons and drug-induced behaviors .
To examine the nature of the adolescent hypersensitivity to drugs of abuse , we first studied cocaine-induced LTP in the VTA . To this end , we recorded glutamate-mediated excitatory postsynaptic currents ( EPSCs ) from VTA dopaminergic ( DA ) neurons in midbrain slices ( Figure 1—figure supplement 1 ) from adolescent ( 5 weeks old ) and adult ( 3-5 months old ) mice 24 hr after a single intraperitoneal ( i . p . ) injection of saline or cocaine ( 1–20 mg/kg; Figure 1—figure supplement 2 ) . We used the peak amplitudes ( at +40 mV ) of the AMPAR and NMDAR-mediated components of the EPSCs ( isolated as described ( Ungless et al . , 2001 ) and Methods ) to calculate the AMPAR/NMDAR ratio , an index of the efficacy of synaptic transmission mediated by AMPARs . In adolescent ) mice , a relatively low dose of cocaine ( 5 mg/kg i . p . ) elicited LTP , manifested by an increase in the AMPAR/NMDAR ratio ( Figure 1a and Figure 1—figure supplement 2 ) . By contrast , only higher doses of cocaine ( 10 and 20 mg/kg ) induced LTP in VTA DA neurons from adult mice ( Figure 1b and Figure 1—figure supplement 2 ) . Thus , cocaine-induced LTP in VTA DA neurons is facilitated in slices from adolescent mice . 10 . 7554/eLife . 12052 . 003Figure 1 . Enhanced susceptibility of adolescent mice to cocaine-induced synaptic potentiation and behavior . ( a–b ) Left , Representative traces of AMPAR and NMDAR EPSCs recorded in VTA DA neurons 24 hr after a single i . p . injection of saline or cocaine . A low dose of cocaine ( 5 mg/kg ) induced LTP , as determined by the increase in the AMPAR/NMDAR ratio ( a , Right , p<0 . 001 , n=11/10 saline/cocaine , t19=8 . 09 ) as well as CPP ( c , p<0 . 0001 , n=11 , t20=7 . 487 ) in adolescent mice ( 5 weeks old ) , but not in adult mice ( 3–5 months old , b , Right , p=0 . 951 , n=8/9/7 saline/5 mg/kg cocaine/10 mg/kg cocaine , F2 , 22=27 . 20; c , p=0 . 3289 , n=9 , t16=1 . 007 ) . A higher dose of cocaine ( 10 mg/kg ) induced LTP in VTA DA neurons ( b , Right , p<0 . 01 vs . saline or 5 mg/kg cocaine , n=8/9/7 saline/5 mg/kg cocaine/10 mg/kg cocaine , F2 , 22=27 . 20 ) and CPP in adult mice ( d , p<0 . 0001 , n=15 , t28=5 . 750 ) . ( e ) DHPG ( 100 μM , 5 min ) evoked LTD in VTA DA neurons of adult mice ( p<0 . 001 , n=6 , t10=19 . 38 ) , but not in adolescent mice ( p=0 . 10 , n=7 , t12=1 . 76 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 12052 . 00310 . 7554/eLife . 12052 . 004Figure 1—figure supplement 1 . Identification of lateral VTA DA neurons in mouse midbrain slices . ( a ) Stable pacemaker firing at 1–5 Hz was recorded from neurons in the lateral VTA in cell-attached mode . ( b ) At Vh=-55 mV , spike width was measured from the start of the inward deflection to the outward peak . Cells with spike widths >1 . 0 ms were taken as dopaminergic . ( c ) Cells only in the ventrolateral VTA with a large ( >150 pA ) hyperpolarization-activated current ( Ih ) , and a large ( >150 pA ) leak current were studied . DOI: http://dx . doi . org/10 . 7554/eLife . 12052 . 00410 . 7554/eLife . 12052 . 005Figure 1—figure supplement 2 . Adolescent mice are more susceptible than adult mice to cocaine-induced LTP in VTA DA neurons . Adolescent ( 5 weeks old , n=6-11 per group ) or adult mice ( 3–5 months old , n=6-9 per group ) were i . p-injected with saline or cocaine at indicated doses and whole-cell recording were performed in VTA DA neurons . LTP , manifested by an increase in AMPAR/NMDAR ratio , was induced at a lower dose of cocaine ( 5 mg/kg , F5 , 77=22 . 15 , p<0 . 001 vs . saline ) in adolescent mice than in adults ( 10 mg/kg , F5 , 77=22 . 15 , p<0 . 01 vs . saline or 5 mg/kg cocaine ) . DOI: http://dx . doi . org/10 . 7554/eLife . 12052 . 00510 . 7554/eLife . 12052 . 006Figure 1—figure supplement 3 . VTA slices from adolescent mice more susceptible to cocaine-induced LTP in vitro . ( a ) Scheme of experimental procedure ( b ) Direct application of a low concentration of cocaine ( 1 μM ) increased AMPAR/NMDAR ratio 3–5 hr post-treatment in VTA DA neurons of adolescent mice , as compared to adult mice ( n=5-11 per group , F1 , 32=6 . 56 , p>0 . 01 Eif2s1S/A vs . wild-type control ) . DOI: http://dx . doi . org/10 . 7554/eLife . 12052 . 00610 . 7554/eLife . 12052 . 007Figure 1—figure supplement 4 . Basal p-eIF2α phosphorylation levels are similar in the VTA of adult and adolescent mice . Western blots are shown on top and quantification of eIF2α levels is shown below ( n=4 , p>0 . 05 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 12052 . 007 To examine whether the cocaine-induced LTP was linked to drug-related behavior , we performed conditioned place preference ( CPP ) tests in adolescent and adult mice . In this task , mice were first presented with either cocaine or saline in different environments . The amount of time spent in the environment previously associated with cocaine versus saline was subsequently recorded . Strikingly , we found that enhanced LTP in the VTA was mirrored in the behavior of adolescent mice: low doses of cocaine ( 5 mg/kg ) elicited CPP only in adolescents , but not in adult mice ( Figure 1c ) . Further mirroring the LTP results , higher doses of cocaine ( 10 mg/kg ) were required to induce CPP in adult mice ( Figure 1d ) . Taken together , these data indicate that adolescent mice are more sensitive to the effects of cocaine with regard to both synaptic transmission and behavior . To further support these findings and rule out potential differences in cocaine metabolism between age groups , we applied cocaine in vitro to midbrain slices from adult and adolescent mice and conducted whole-cell recordings ( Figure 1—figure supplement 3a ) , as previously described ( Argilli et al . , 2008 ) . In slices from adolescent mice , a relatively low concentration of cocaine ( 1 μM ) elicited LTP in VTA DA neurons ( Figure 1—figure supplement 3b ) , whereas in slices from adult mice only a higher concentration of cocaine ( 5 μM ) induced LTP ( Figure 1—figure supplement 3c ) . Taken together with previous reports that the concentrations of cocaine in both blood and brain are similar in adolescent and adult mice ( Zombeck et al . , 2009 ) , these data support the notion that cocaine-induced LTP in VTA DA neurons has a lower threshold in adolescent mice . Since mGluR-LTD blocks cocaine-induced LTP in VTA DA neurons ( Bellone and Lüscher , 2006 ) and its disruption has been postulated to enhance vulnerability to drug addiction ( Lüscher and Huber , 2010 ) , we predicted that mGluR-LTD in the VTA would be impaired in adolescent mice . In agreement with this prediction , a brief-application of DHPG—a selective mGluR1/5 agonist—induced LTD in VTA DA neurons from adult ( Figure 1e ) , but not adolescent mice ( Figure 1e ) . Protein synthesis is required for both cocaine-induced LTP ( Argilli et al . , 2008 ) and mGluR-LTD ( Mameli et al . , 2007 ) in VTA DA neurons , as well as cocaine-induced changes in behavior ( Sorg and Ulibarri , 1995; Kuo et al . , 2007 ) . Given that translation rates in the brain decrease significantly with age ( Vargas and Castaneda , 1983 ) , we examined whether a low dose of cocaine is sufficient to trigger LTP in VTA DA neurons and induce CPP in adolescent mice ( Figure 1a and 1c ) also activates a particular translational control program in the VTA of these mice . To this end , we measured the activity of key signaling pathways impinging on translation initiation ( Buffington et al . , 2014 ) . We found that a low dose of cocaine ( 5 mg/kg ) reduced the amount of phosphorylated eIF2α ( p-eIF2α ) only in VTA slices from adolescent mice ( Figure 2a and 2b ) . By contrast , a higher dose of cocaine ( 10 mg/kg ) was required to decrease p-eIF2α levels in VTA slices from adult mice ( Figure 2a and 2b ) . Importantly , a signle injection of cocaine failed to alter p-eIF2α levels in the nucleus accumbens ( Figure 2—figure supplement 1 ) , another brain region involved in addiction ( Kalivas and Volkow , 2011 ) . Moreover , the lack of effect on other translational signaling pathways in adolescent VTA neurons by the same low dose of cocaine ( Figure 2d–f ) highlights the selective involvement of p-eIF2α-mediated translational control during the period of heightened adolescent vulnerability to cocaine addiction . Thus , eIF2α is a newly identified effector of cocaine action . 10 . 7554/eLife . 12052 . 008Figure 2 . A low dose of cocaine selectively reduces p-eIF2α in the VTA of adolescent mice . ( a–b ) A low dose of cocaine ( 5 mg/kg ) reduced p-eIF2α in the VTA of adolescent ( p<0 . 05 , n=5 per group , t8=3 . 029 ) , but not adult mice ( p=0 . 329 , n=3 per group , t4=1 . 110 ) . A higher dose of cocaine ( 10 mg/kg ) was needed to reduce p-eIF2α in VTA of adult mice ( p<0 . 001 , n=6 per group , t10=4 . 640 ) . ( c ) Schematic of mTORC1- and eIF4E-mediated translation . In abilescnt mice , a low dose of cocaine ( 5 mg/kg ) did not significantly alter phosphorylation of S6K at Thr-389 ( d ) , 4E-BP1 at Thr-37 and Thr-46 ( e ) and eIF4E at Ser209 ( f ) . Western blots are shown on top and quantification for each phospho-protein/total-protein is shown at the bottom ( n=3/3 saline/cocaine; S6K , p=0 . 3467 , t4=01 . 066a; 4E-BP1 , p=0 . 5031 , t4=0 . 7351; eIF4E , p=0 . 5669 , t4=0 . 6233 ) . Plots are mean ± s . e . m . DOI: http://dx . doi . org/10 . 7554/eLife . 12052 . 00810 . 7554/eLife . 12052 . 009Figure 2—figure supplement 1 . Doses of cocaine which lower p-eIF2α in the VTA have no effect in nucleus accumbens ( NAc ) . ( a ) Scheme of the experimental procedure ( b ) A low dose of cocaine ( 5mg/kg ) or a higher dose of cocaine ( 10 mg/kg ) had no effect on p-eIF2 in the NAc of adolescent ( p=0 . 678 , n=3 per group , t4=0 . 4 ) or adult mice ( p=0 . 18 , n=3 per group , t4=1 . 6 ) , respectively . Plots are mean ± s . e . m . DOI: http://dx . doi . org/10 . 7554/eLife . 12052 . 009 If reduced eIF2α phosphorylation enhances the susceptibility of adolescents to the effects of cocaine , then decreasing its phosphorylation in adults should increase their vulnerability . To test this idea , we injected a low dose of cocaine ( 5 mg/kg ) into adult wild-type ( WT ) Eif2s1S/S mice and Eif2s1S/A heterozygous knock-in mice ( where a single phosphorylation site at serine 51 is replaced by alanine ) ( Scheuner et al . , 2001 ) . In Eif2s1S/A mutant mice , eIF2α phosphorylation was significantly reduced in the VTA ( Figure 3—figure supplement 1 ) . As predicted , a low dose of cocaine ( 5 mg/kg ) induced LTP in VTA DA neurons from Eif2s1S/A mutant mice but not in VTA DA neurons from WT littermates ( Figure 3a ) . In addition , application of a low concentration of cocaine ( 1 μM ) in vitro was sufficient to induce LTP in VTA DA neurons from Eif2s1S/A mutant mice , but not in those from adult WT controls ( Figure 3—figure supplement 2 ) . Consistent with the LTP results , low doses of cocaine ( 5 mg/kg ) induced CPP only in Eif2s1S/A mutant mice ( Figure 3b ) . Hence , like adolescent mice , adult mice with reduced p-eIF2α are more susceptible to cocaine-evoked LTP ( both in vivo and in vitro ) and drug-induced behavior . 10 . 7554/eLife . 12052 . 010Figure 3 . Decreasing p-eIF2α makes adult mice more susceptible to cocaine-induced LTP and behavior . ( a–b ) A low dose of cocaine ( 5 mg/kg ) induced both LTP in VTA DA neurons ( a , p<0 . 05 , n=5 , t8=4 . 193 ) and CPP in adult Eif2s1S/A mice ( b , p<0 . 01 , n=7 , t12=3 . 411 ) compared to Eif2s1S/S mice ( a , p=0 . 89 , n=5 , t8=0 . 14; b , p=0 . 2170 , n=7 , t12=1 . 303 ) . ( c–d ) A low dose of cocaine ( 5 mg/kg ) elicited LTP ( c , p<0 . 001 , n=6 , t10=3 . 43 ) and CPP ( d , p=0 . 1761 , n=8 vehicle+cocaine , t14=1 . 425; p<0 . 0001 , n=16 ISRIB+cocaine , t30=2 . 433 ) in ISRIB-injected adult mice compared to vehicle-injected mice . ( e–f ) DHPG ( 100 μM , 5 min ) induced LTD in WT adult VTA DA neurons ( e , p<0 . 001 , n=5 , t8=20 . 3 ) and vehicle-injected WT adult mice ( f , p<0 . 001 , n=5 , t8=5 . 17 ) , but not in Eif2s1S/A mice ( e , p=0 . 26 , n=7 , t12=1 . 2 ) and ISRIB-injected mice ( f , p=0 . 42 , n=4 , t6=0 . 86 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 12052 . 01010 . 7554/eLife . 12052 . 011Figure 3—figure supplement 1 . eIF2α phosphorylation is reduced in VTA from adult Eif2s1S/A mice . Western blots ( top ) show reduction in p-eIF2α in Eif2s1S/A mutant mice compared to wild-type littermates ( Eif2s1S/S ) . Quantification of eIF2α phosphorylation vs . total-eIF2α is shown below ( p<0 . 01 , n=3 per group , t4=6 . 67 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 12052 . 01110 . 7554/eLife . 12052 . 012Figure 3—figure supplement 2 . Decreasing p-eIF2α makes VTA slices from adult mice more susceptible to cocaine-induced LTP in vitro . Direct application of a low concentration of cocaine ( 1 μM ) increased AMPAR/NMDAR ratio 3–5 hr post-treatment in VTA DA neurons of Eif2s1S/A mice , as compared to wild-type controls ( n=5-11 per group , F1 , 32=6 . 56 , p<0 . 01 Eif2s1S/A vs . wild-type control ) . DOI: http://dx . doi . org/10 . 7554/eLife . 12052 . 01210 . 7554/eLife . 12052 . 013Figure 3—figure supplement 3 . In adult mice , systemic administration of ISRIB alone failed to induce LTP in VTA DA neurons and CPP . a , b . i . p . injection of ISRIB ( 2 . 5 mg/kg ) alone did not induce LTP ( a , p=0 . 79 , n=6/3 ISRIB/vehicle , t7=0 . 28 ) or CPP ( b , p=0 . 329 , n=9 , t16=1 . 008 ) , as indicated by the lack of potentiation of VTA DA neurons and difference between average pre- and post-test preference scores , respectively . DOI: http://dx . doi . org/10 . 7554/eLife . 12052 . 013 We next asked whether acute pharmacological inhibition of p-eIF2α-mediated translation in adult mice renders animals more susceptible to a low dose of cocaine . Phosphorylation of eIF2α inhibits general protein synthesis rates by binding to and inhibiting the guanine nucleotide exchange factor ( GEF ) eIF2B that is required for eIF2 activation . We therefore used a recently-discovered small molecule inhibitor ISRIB ( Sidrauski et al . , 2013 ) , which potently blocks p-eIF2α-mediated translational effects by promoting eIF2B activity ( Sekine et al . , 2015; Sidrauski et al . , 2015 ) . Adult WT mice injected with ISRIB ( 2 . 5 mg/kg ) and a low dose of cocaine ( 5 mg/kg ) showed both LTP in VTA ( Figure 3c ) and drug-induced CPP ( Figure 3d ) . Note that neither cocaine nor ISRIB alone triggered LTP or CPP in adult WT mice ( Figure 3—figure supplement 3 ) . These results support the notion that reduced p-eIF2α–mediated translational control renders adult mice more susceptible to the synaptic and behavioral effects of cocaine . Given that mGluR-LTD in the VTA of adolescent mice is impaired ( Figure 1e ) and adult mice with reduced p-eIF2α–mediated translational control resemble adolescent mice in their susceptibility to cocaine-induced synaptic potentiation and behavior ( Figure 1 and 3 ) , we predicted that mGluR-LTD might be deficient in adult mice with reduced p-eIF2α–mediated translational control . Consistent with this prediction , mGluR-LTD was impaired in VTA DA neurons from mice with reduced p-eIF2α ( Eif2s1S/A mice ) and adult WT mice injected with ISRIB ( Figure 3e–f ) . Thus , our parallel genetic and pharmacological experiments provide strong evidence that reducing p-eIF2α-mediated translational control in the VTA of adult mice makes them more like adolescents with respect to mGluR-LTD , and cocaine-evoked LTP and CPP . To examine whether increasing p-eIF2α in the VTA of adolescent mice is sufficient to confer resistance to low doses of cocaine , we administered Sal003 , an inhibitor of eIF2α phosphatases ( Boyce et al . , 2005; Robert et al . , 2006 ) ( Figure 4a and 4b ) , directly into the VTA of young mice to promote eIF2α phosphorylation locally ( Figure 4c and Figure 4—figure supplement 1 ) . As expected , a low dose of cocaine ( 5 mg/kg ) induced LTP in adolescent mice locally infused with vehicle ( Figure 4b ) , but not in VTA DA neurons from adolescent mice infused with Sal003 ( Figure 4b ) . The Sal003-mediated increase in p-eIF2α also blocked the LTP evoked by cocaine in vitro in brain slices ( Figure 4d–e ) , further supporting the in vivo experiments . 10 . 7554/eLife . 12052 . 014Figure 4 . Increasing p-eIF2α in young mice blocks cocaine-induced LTP and behavior . ( a ) Schematic showing Sal003 mechanism of action . ( b–c ) Infusion of Sal003 into the VTA blocked cocaine-induced potentiation ( c , p<0 . 001 , n=5 per group , t8=3 . 81 ) and increased p-eIF2α in the VTA ( p<0 . 01 , n=7/6 vehicle/Sal003 , t11=3 . 172 ) . ( d ) Schematic of experimental design . ( e ) Direct application of cocaine ( 1 μM ) induced LTP 3–5 hr post-treatment ( p<0 . 05 , n=11/6 vehicle/cocaine , F2 , 20=7 . 48 ) , whereas Sal003 prevented it ( p<0 . 05 , n=11/6 , vehicle/cocaine+Sal003 , F2 , 20=7 . 48 , cocaine vs . cocaine+Sal003 ) . Representative traces of AMPAR and NMDAR EPSCs ( top ) . ( f ) Infusion of Sal003 into the VTA blocked CPP ( p<0 . 05 , n=7 vehicle+cocaine , t12=2 . 592; p=0 . 1147 , n=10 Sal003+cocaine , t18=1 . 892 ) in adolescent mice . ( g ) Application of Sal003 ( 20 μM , 10 min ) , a selective inhibitor of eIF2α phosphatases , induced LTD in VTA DA neurons from adolescent mice ( p<0 . 001 , n=6 , t10=9 . 517 ) . Plots are mean ± s . e . m . DOI: http://dx . doi . org/10 . 7554/eLife . 12052 . 01410 . 7554/eLife . 12052 . 015Figure 4—figure supplement 1 . Sites of Sal003 infusions into VTA at seven rostrocaudal planes and corresponding increase in p-eIF2α . Coordinates are posterior to bregma and cannula tip placements are from mice infused with Sal003 ( 1 μl; 20 μM ) and vehicle ( 1 μl ) . DOI: http://dx . doi . org/10 . 7554/eLife . 12052 . 015 To determine whether changes in phosphorylation of eIF2α in the VTA and the behavioral susceptibility to cocaine are causally related , we assessed CPP in adolescent mice infused with vehicle or Sal003 directly into the VTA . Low doses of cocaine elicited CPP in vehicle-infused but not in Sal003-infused adolescent mice ( Figure 4f ) . Moreover , as expected , Sal003 induced mGluR-LTD in VTA slices from adolescent mice ( Figure 4g ) . Hence , a direct increase in p-eIF2α in the VTA of adolescent mice blocks the susceptibility to cocaine-induced LTP and behavior by promoting an opposing LTD in VTA DA neurons . We next studied the mechanism by which reduced p-eIF2α-mediated translation renders adult mice more susceptible to the effects of cocaine . Given that a ) eIF2α phosphorylation-mediated translational control is both necessary and sufficient for mGluR-LTD in the hippocampus ( Di Prisco et al . , 2014 ) and VTA ( Figures 3e , f and 4g ) , b ) eIF2α phosphorylation blocks general translation but selectively triggers translation of a few select mRNAs during mGluR-LTD ( including oligrophrenin-1 ( Ophn1 ) mRNA ) ( Di Prisco et al . , 2014 ) , and c ) translation of Ophn1 mRNA is required for mGluR-LTD ( Di Prisco et al . , 2014; Nadif Kasri et al . , 2011 ) , we predicted that in adult mice with reduced OPHN1 levels in VTA , a low dose of cocaine ( 5 mg/kg ) would induce LTP and CPP . As anticipated , in adult mice injected with a specific shRNA against Ophn1 ( Ophn1-shRNA ) in the VTA , this low dose of cocaine ( 5 mg/kg ) triggered LTP in the VTA in vivo ( Figure 5a ) and induced CPP ( Figure 5c ) . However , the same low dose of cocaine failed to do so in mice injected with a control ( scrambled ) shRNA ( Figure 5a and 5b ) . Hence , like adolescent mice ( Figure 1a and 1c ) or adult mice with reduced p-eIF2α–mediated translation ( Figure 3a and 3b ) , adult mice with reduced OPHN1 levels in the VTA are more sensitive to the effects of cocaine . 10 . 7554/eLife . 12052 . 016Figure 5 . Decreasing OPHN1 levels in VTA DA neurons makes adult mice more susceptible to cocaine-induced LTP . ( a ) A low dose of cocaine ( 5 mg/kg ) induced LTP in adult Ophn1-shRNA injected VTA DA neurons ( a , Right , p<0 . 01 , n=5 , t8=5 . 464 ) ; above representative traces of AMPAR and NMDAR EPSCs ( top ) . ( b ) Low doses of cocaine ( 5 mg/kg ) induced CPP in mice locally injected with Ophn1-shRNA ( p<0 . 01 , n=14 , t26=3 . 600 ) , but not in control mice injected with scrambled shRNA ( p=0 . 7829 , n=4 , t6=0 . 2882 ) . ( c ) Sal003 ( 20 μM ) blocked the cocaine-induced LTP in the VTA of control shRNA-injected mice ( p<0 . 01 , n=6/6/7 vehicle/cocaine/cocaine+Sal003 , F2 , 16=13 . 03 ) , but failed to do so in Ophn1-shRNA VTA DA neurons ( p=0 . 29 , n=6/6/11 , vehicle/cocaine/cocaine+Sal003 , F2 , 20=4 . 29 , cocaine vs . cocaine+Sal003; p<0 . 05 vehicle vs . cocaine or cocaine+Sal003 ) . ( d ) Representative sample traces of AMPAR EPSCs . ( e–f ) I-V plots . ( g ) Cocaine increased the rectification index in control-shRNA injected VTA neurons while Sal003 blocked it ( p<0 . 001 , n=6/6/7 vehicle/cocaine/cocaine+Sal003 , F2 , 16=30 . 30 , cocaine vs . vehicle or cocaine vs . cocaine+Sal003 ) , whereas both cocaine and cocaine+Sal003 increased the rectification index in VTA DA neurons from Ophn1-shRNA-injected mice ( p<0 . 05 , n=6/6/11 vehicle/cocaine/cocaine+Sal003 , F2 , 20=3 . 92 , vehicle vs . cocaine or cocaine+Sal003; p=0 . 80 cocaine vs . cocaine+Sal003 ) . Plots are mean ± s . e . m . DOI: http://dx . doi . org/10 . 7554/eLife . 12052 . 016 To assess the causal relationship between p-eIF2α and OPHN1 during cocaine-evoked LTP , we directly applied cocaine ( 5 μM ) and Sal003 ( 20 μM ) to brain slices in vitro and recorded LTP 3–5 hr after exposure , as previously described ( Argilli et al . , 2008 ) . Remarkably , Sal003 blocked cocaine-induced LTP in control VTA DA neurons , but not in VTA DA neurons in which OPHN1 was reduced by Ophn1-shRNA ( Figure 5c ) . In vivo injections of cocaine are known to induce LTP in the VTA by replacing postsynaptic AMPARs containing the GluR2 subunit with calcium-permeable AMPARs lacking the GluR2 subunit ( Bellone and Lüscher , 2006 ) . In order to investigate whether cocaine-induced LTP in vitro involves a similar process , we measured rectification ( manifested as lower amplitude AMPAR EPSCs measured at positive holding potentials vs . those measured at negative potentials ) , a hallmark of GluR2-lacking AMPARs ( Liu and Zukin , 2007 ) . We recorded EPSCs at -70 , 0 and +40 mV to calculate the rectification index and found that Sal003 blocked the increased inward rectification induced by in vitro application of cocaine in WT slices , but not in slices from OPHN1-deficient mice ( Figure 5d–g ) . Collectively , these data indicate that OPHN1 is a specific target by which eIF2α phosphorylation regulates plasticity changes in VTA DA neurons . Through their actions on distinct receptors , different drugs of abuse induce LTP in VTA DA neurons , thus reinforcing drug-seeking behavior ( Bowers et al . , 2010; Lüscher and Malenka , 2011 ) . To test the effects of other addictive drugs on the phosphorylation of eIF2α in VTA , we treated mice with methamphetamine , nicotine , and ethanol at doses known to evoke LTP in VTA DA neurons ( Saal et al . , 2003 ) . We found that , like cocaine , these drugs with very different mechanisms of action , all reduced p-eIF2α in VTA of adult mice ( Figure 6 ) . 10 . 7554/eLife . 12052 . 017Figure 6 . Multiple drugs of abuse reduce p-eIF2α in VTA of adult mice . ( a ) i . p . injection of nicotine ( 1 mg/kg ) , ethanol ( 2 g/kg ) , or methamphetamine ( 1 mg/kg ) reduces p-eIF2α in VTA ( n=5 per group; Saline vs . nicotine , p<0 . 05 , t8=2 . 879; ethanol , p<0 . 001 , t8=6 . 278 methamphetamine , p<0 . 001 , t8=5 . 449 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 12052 . 017
Adolescence is a period of heightened susceptibility to drug addiction ( Chambers et al . , 2003; Kandel et al . , 1992 ) , but little is known about the underlying biological mechanisms . Changes in gene expression in key reward areas have been shown to play a critical role in drug-induced changes in synaptic potentiation and reward-related behavior ( Hyman et al . , 2006; Robison and Nestler , 2011 ) . Until now , most of the research in this area has focused on how transcriptional control ( via key transcription factors , such as △FosB and CREB ) or epigenetic mechanisms ( via histone acetylation and methylation , DNA methylation , and non-coding RNAs ) contribute to addiction-related behavior ( Robison and Nestler , 2011 ) . Our focus on translational control was based on several key observations . Compared to the brains of adolescent mice—which are more vulnerable to drugs of abuse—protein synthesis is reduced in the brains of adult mice ( Vargas and Castaneda , 1983; Hovda et al . , 2006; Sun et al . , 1995 ) . Furthermore , translational control of gene expression is the ultimate step in the functional output of gene expression ( Sonenberg and Hinnebusch , 2009 ) and neurons could regulate protein synthesis without altering mRNA synthesis or transport , allowing for local control of protein synthesis at synapses ( Holt and Schuman , 2013 ) . Moreover , protein synthesis is required for cocaine-evoked LTP in VTA DA neurons ( Argilli et al . , 2008; Yuan et al . , 2013 ) , as well as cocaine-induced behavior ( Sorg and Ulibarri , 1995; Kuo et al . , 2007 ) . Our findings reveal a critical role for p-eIF2α-mediated translational control in the heightened susceptibility of adolescent mice to the initial synaptic and behavioral adaptations induced by cocaine . In adolescent , but not in adult mice , a low dose of cocaine selectively lowers eIF2α phosphorylation in the VTA ( Figure 2a–b ) , thereby eliciting LTP and drug-induced behavior . It is noteworthy that we could not detect a significant difference in the baseline phosphorylation of eIF2α between adolescent and adult mice ( Figure 1—figure supplement 4 ) . While the precise mechanism by which drugs of abuse lower phosphorylation of eIF2α in the VTA is currently under investigation , we hypothesize that drugs of abuse either inhibit the activity and/or expression of one of the eIF2α kinases or stimulate the activity of eIF2α phosphatases to lower p-eIF2α levels in the VTA . We suspect that adolescents are more prone to either ( or both ) of these processes . Genetic reduction of eIF2α phosphorylation or pharmacologically blunting its effects with ISRIB in adult mice enhances susceptibility to cocaine , mimicking the increased vulnerability of adolescent animals ( Figure 3 ) . By contrast , a local increase in p-eIF2α in the VTA blocked cocaine-evoked LTP and cocaine-induced behavior in adolescent mice ( Figure 4 ) . These effects converge on p-eIF2α–mediated translation of OPHN1 , which is selectively synthesized in neurons under conditions where eIF2α is phosphorylated ( Di Prisco et al . , 2014 ) and leads to endocytic down-regulation of post-synaptic AMPARs ( Nadif Kasri et al . , 2011 ) . Collectively , our data indicate that reduced p-eIF2α-mediated translation of Ophn1 mRNA accounts for the adolescent hypersensitivity to cocaine . Is p-eIF2α-mediated mGluR-LTD in VTA a defense mechanism that limits the early synaptic adaptations required for the initiation of addiction ? A delicate interplay between LTP and mGluR-LTD in VTA DA neurons is believed to modulate cocaine’s synaptic and behavioral effects ( Loweth et al . , 2013; Luscher , 2013 ) . Cocaine induces LTP at VTA DA synapses by inserting new AMPARs into the postsynaptic membrane ( Ungless et al . , 2001 ) . This process is reversed either by pharmacological or synaptic induction of mGluR-LTD in VTA by removing AMPARs from the post-synaptic DA neurons ( Bellone and Lüscher , 2006 ) . Conversely , reduced mGluR-LTD in the midbrain is postulated to enhance vulnerability to drug addiction ( Lüscher and Huber , 2010 ) . Our data provide the first compelling mechanistic evidence that p-eIF2α-mediated translational control of OPHN1 synthesis is a key mechanism underlying the reversal of cocaine-evoked LTP by mGluR-LTD . Thus , our findings provide a unifying model that explains how the two opposing forms of plasticity ( cocaine-induced LTP and mGluR-LTD ) are regulated by a single translational control mechanism . Given that blocking mGluR function in the VTA renders cocaine-evoked LTP more long-lasting ( Mameli et al . , 2009 ) , it will be interesting to determine whether decreasing p-eIF2α also leads to a persistent LTP in the VTA . These drug-induced persistent changes on excitatory afferents onto dopamine neurons in the VTA are particularly relevant since they may represent the cellular processes driving the progression from recreational use to chronic drug seeking ( Chen et al . , 2008 ) . In addition , eIF2α phosphorylation may also be required for subsequent synaptic adaptation in other mesolimbic reward areas , such as the NAc , in response to chronic exposure to cocaine or after withdrawal . The identification of a single common downstream mechanism of action of different drugs of abuse has been challenging . It is therefore significant that drugs of abuse that act on distinct receptors ( cocaine , methamphetamine , nicotine and ethanol ) , and induce LTP ( Bowers et al . , 2010; Lüscher and Malenka , 2011 ) in VTA DA neurons , all reduce p-eIF2α in the VTA ( Figure 6 ) . Taken together with our findings that reduced p-eIF2α-mediated translational control increases the susceptibility of adolescent mice to cocaine , these observations raise an intriguing possibility that polymorphisms in the eIF2α signaling pathway could be associated with drug use in humans . Indeed , in an accompanying paper ( Placzek et al . , 2016 ) , we provide evidence that eIF2α phosphorylation also controls adolescent hypersensitivity to nicotine-evoked synaptic potentiation and identify a polymorphism in the promoter of the Eif2s1 gene ( encoding eIF2α ) that is associated with changes in reward-related activity in human smokers , as measured by functional magnetic resonance imaging . Thus , since eIF2α phosphorylation is reduced by a variety of addictive drugs , agents that selectively alter eIF2α phosphorylation-mediated translational control in key reward areas in the brain could be useful for the treatment of a broad range of addictive behaviors .
All experiments were conducted using male and female mice from the C57Bl/6 background . Eif2s1S/A and Eif2s1A/A;ftg mice were previously described ( Di Prisco et al . , 2014 ) . Mice were kept on a 12h/12h light/dark cycle ( lights on at 7:00 am ) and had access to food and water ad libitum . During mid-adolescence ( postnatal day 35–40 ) ( Spear , 2000; Laviola et al . , 2003 ) , mice show characteristic behavior patterns , including impulsiveness and risk-taking ( Laviola et al . , 2003 ) . We therefore selected five week-old mice ( 35–42 postnatal days ) as adolescents , and 3–5 month old mice as adults . Animal care and experimental procedures were approved by the institutional animal care and use committee ( IACUC ) at Baylor College of Medicine , according to NIH Guidelines . No statistical methods were used to predetermine sample sizes . All sample sizes meet the criteria for corresponding statistical tests—our sample sizes are similar to those reported in previous publications ( Ungless et al . , 2001; Saal et al . , 2003; Bellone and Lüscher , 2006; Argilli et al . , 2008; Koo et al . , 2012 ) . For behavioral and biochemical studies , mice were arbitrarily assigned to control and treatment groups . These experiments were performed and analyzed blind to treatment conditions and/or genotype . All drugs of abuse were dissolved in 0 . 9% saline and injected in a volume of 5 ml/kg . Cocaine hydrochloride , ( − ) -nicotine hydrogen tartrate , and USP-grade 95% ethanol were obtained from Sigma-Aldrich ( St . Louis , MO ) . Racemic methamphetamine hydrochloride was a kind gift from Dr . Kristen Horner ( Mercer University School of Medicine ) . ISRIB ( P . Walter ) was dissolved in DMSO and further diluted in PEG-400 ( 1:1 ratio ) as previously described ( Di Prisco et al . , 2014 ) . For both electrophysiological and behavioral experiments , ISRIB ( 2 . 5 mg/kg ) or vehicle ( DMSO/PEG-400 , 2 ml/kg ) was injected 90 min before cocaine or saline injection , respectively . Sal003 ( Tocris Biosciences , R&D Systems , Minneapolis , MN ) was dissolved in DMSO and further diluted in 0 . 9% saline . Sal003 ( 20 μM ) or vehicle ( 0 . 4% DMSO in saline ) was infused bilaterally into the VTA as summarized in the 'Cannulation and Sal003 infusion' section . Electrophysiological recordings were performed as previously described ( Ungless et al . , 2001; Di Prisco et al . , 2014 ) and the investigators remained blind to genotypes . Each electrophysiological experiment was replicated at least three times . Briefly , mice were anesthetized with a mixture of ketamine ( 100 mg/kg ) , xylazine ( 10 mg/kg ) , and acepromazine ( 3 mg/kg ) . Mice were transcardially perfused with an ice-cold , oxygenated solution containing ( in mM ) NaCl , 120; NaHCO3 , 25; KCl , 3 . 3; NaH2PO4 , 1 . 2; MgCl2 , 4; CaCl2 , 1; dextrose , 10; sucrose , 20 . Horizontal slices ( 225–300 mm thick ) containing the VTA were cut from the brains of adolescent ( 5 weeks old ) or adult ( 3–5 months old ) C57BL/6J mice with a vibrating tissue slicer ( VF-100 Compresstome , Precisionary Instruments , San Jose , CA , or Leica VT 1000S , Leica Microsystems , Buffalo Grove , IL ) , incubated at 34°C for 40 min , kept at room temperature for at least 30 min prior to recording before they were transferred to a recording chamber where they were continuously perfused with artificial cerebrospinal fluid ( ACSF ) at 32°C and a flow rate of 2–3 ml/min . The recording ACSF was different from the cutting solution in the concentration of MgCl2 ( 1 mM ) and CaCl2 ( 2 mM ) . Recording pipettes were made from thin-walled borosilicate glass ( TW150F-4 , WPI , Sarasota , FL ) . After filling with intracellular solution ( in mM ) : 117 CsMeSO3; 0 . 4 EGTA; 20 HEPES; 2 . 8 NaCl , 2 . 5 ATP-Mg 2 . 0; 0 . 25 GTP-Na; 5 TEA-Cl , adjusted to pH 7 . 3 with CsOH and 290 mOsmol/l , they had a resistance of 3–5 MΩ . For studies of AMPAR current rectification , spermine ( 100 μM ) was added to the internal solution , which blocks GluR2-lacking receptors at depolarized potentials . Data were obtained with a MultiClamp 700B amplifier , digitized at 20 kHz with a Digidata 1440A , recorded by Clampex 10 and analyzed with Clampfit 10 software ( Molecular Devices ) . Recordings were filtered online at 4 kHz with a Bessel low-pass filter . A 2 mV hyperpolarizing pulse was applied before each EPSC to evaluate the input and access resistance ( Ra ) . Data were discarded when Ra was either unstable or greater than 25MΩ , holding current was >200 pA , input resistance dropped >20% during the recording , or EPSCs baseline changed by >10% . Traces illustrated in Figures are averages of 10–15 consecutive traces . After establishing a gigaohm seal ( >2GΩ ) and recording stable spontaneous firing in cell-attached , voltage clamp mode ( -70 mV holding potential ) , cell phenotype was determined by measuring the width of the cell-attached action potential ( Figure 1—figure supplement 1 ) . AMPAR/NMDAR ratios were calculated as previously described ( Ungless et al . , 2001 ) . Briefly , neurons were voltage-clamped at +40 mV until the holding current stabilized ( at <200 pA ) . Monosynaptic EPSCs were evoked at 0 . 05 Hz with a bipolar stimulating electrode placed 50–150 μm rostral to the lateral VTA . Picrotoxin ( 100 μM ) was added to the recording ACSF to block GABAAR-mediated IPSCs . After recording the dual-component EPSC , DL-AP5 ( 100 μM ) was bath-applied for 10 min to remove the NMDAR component , which was then obtained by offline subtraction of the remaining AMPAR component from the original EPSC . The peak amplitudes of the isolated components were used to calculate the AMPAR/NMDAR ratios . Rectification indices were calculated as the ratio of the chord conductance of evoked EPSCs at a negative holding potential ( -70 mV ) to the chord conductance at a positive holding potential ( +40 mV ) obtained in the presence of 100 μM DL-AP5 , as previously described ( Bellone et al . , 2011 ) . Picrotoxin and DL-AP5 were purchased from Tocris Bioscience and all other reagents and experimental compounds were obtained from Sigma-Aldrich . Experiments applying drugs in vitro were performed as previously described ( Argilli et al . , 2008 ) with slight modifications . Briefly , slices were incubated with cocaine ( 1 μM or 5 μM ) and Sal003 ( 20 μM ) for 15 and 20 min , respectively , as shown in Figure 3B . After treatment , slices were transferred ( twice ) to a 35 mm polycarbonate dish containing regular ACSF for complete drug washout and allowed to recover for 2–4 hr . Whole-cell recordings were then conducted 3–5 hr after the end of drug exposure . AAV5-Cre ( Titer: 1 . 0e13GC/ml ) was purchased from Vector Biolabs ( Cat#7012 , Philadelphia , PA ) ; Lentiviral constructs expressing Ophn1 shRNA and scrambled shRNA were generously provided by Dr . Linda van Aelst ( Nadif Kasri et al . , 2011 ) ( Cold Spring Harbor Laboratory ) and viruses were produced by Gene Vector Core Laboratory ( Baylor College of Medicine ) . Viral injections were performed as previously described ( Di Prisco et al . , 2014 ) . Briefly , mice were anaesthetized with isoflurane ( 2–3% ) and viruses ( 1–2 μl/site ) were injected bilaterally at the rate of 0 . 1 μl/min , and an additional 10 min to allow for diffusion of viral particles . Injection coordinates , targeting VTA , were as follows ( with reference to bregma ) : -2 . 50 AP , ± 0 . 45 ML , −4 . 50 DV . The incision was sutured after injection and mice were returned to home cages . Mouse body weight and signs of illness were monitored until full recovery from surgery ( ~1 week ) . Drug treatment and experiments were all performed at least three weeks after viral injection . The investigators were blind to the genotypes for the behavioral tests . CPP , performed as previously described ( Koo et al . , 2012 ) , was assessed over 6 days using an unbiased procedure and a standard two-chamber CPP apparatus ( Ugo Basile , Varese , Italy ) . Animal behavior was videotaped with an overhead camera and analyzed by ANY-maze software ( Stoelting , Wood Dale , IL ) . The difference in the time spent in cocaine-paired side versus saline-paired side was calculated as the CPP score . On day 1 , a mouse was placed in the chamber with the doors removed for a 30 min pre-training test and the baseline preference was calculated . Mice with strong pre-training preference to any chamber ( CPP score >540 s ) were excluded from the experiment ( <10% of all mice tested ) . On the following four days , training sessions were performed once a day . On alternate days , mice were given injections of cocaine ( 5 mg/kg or 10 mg/kg , i . p . ) or 0 . 9% saline ( 5 ml/kg , i . p . ) immediately before being confined to the cocaine-paired or saline-paired chamber for 30 min and then returned to their home cages . On day 6 , a test session identical to the pre-training test was conducted to determine the CPP scores . Mice were anesthetized with isoflurane ( 2–3% ) and mounted on a stereotaxic frame . Cannulae ( 26 gauge ) were implanted bilaterally to target the VTA region at an angle of 15° from the midline at these coordinates: -3 . 16 mm AP , ± 0 . 63 mm ML , -3 . 72 mm DV ( as determined from the Paxinos & Franklin atlas ) . Two jewelry screws were inserted into the skull and the cannulae were held in place by acrylic cement . A 33 gauge dummy probe was inserted into the guide to prevent clogging by tissue debris . Bilateral infusions [0 . 5 μl of vehicle ( 0 . 4% DMSO in saline ) or Sal003 ( 20 μM ) ] were made via the implanted cannulae in freely-moving mice 30 min before cocaine or saline injection , driven by a motorized syringe pump ( KdScientific ) at the rate of 0 . 1 μl/min . After 5 min of infusion , the injector remained in the cannulae for an additional minute to allow diffusion of the solution . Cannula placements were visually confirmed in subsequent brain sections . For electrophysiological and behavioral experiments , mice were killed after all tests for histological confirmation . Brains were fixed in 4% paraformaldehyde and 80 μm sections cut for Nissl-staining to identify cannula placement . Only mice with correct bilateral placements were included in the analyses . Cannulae and infusion accessories were custom-made by Plastics One ( Roanoke , VA ) . VTA samples were micro-dissected from 1 mm coronal sections obtained using an acrylic mouse brain matrix ( Stoelting ) . Briefly , mice were killed by an overdose of isoflurane and their brains were quickly removed and placed in ice-cold PBS-immersed brain matrix for sectioning . The section containing most of the VTA ( typically 8th from rostral to caudal ) was then transferred for microdissection of the VTA using scalpels over an ice-cold petri-dish . Samples were collected in pre-chilled microcentrifuge tubes and lysed in homogenizing buffer [200 mM HEPES , 50 mM NaCl , 10% Glycerol , 1% Triton X-100 , 1 mM EDTA , 50 mM NaF , 2 mM Na3VO4 , 25 mM β-glycerophosphate , and EDTA-free complete ULTRA tablets ( Roche , Indianapolis , IN ) ] . Western blotting was performed as previously described ( Huang et al . , 2013 ) . Primary antibodies for Western blotting were rabbit anti-p-eIF2α ( Ser51 ) ( 1:1000 , Cell Signaling Technology Laboratories , Danver , MA ) , mouse anti-total eIF2α ( 1:1000 , Cell Signaling Technology Laboratories , Danver , MA ) , and mouse anti-β-actin ( 1:10 , 000 , EMD Millipore , Billerica , MA ) . All data are presented as mean ± s . e . m . Statistical analyses were performed using SigmaPlot ( Systat Software ) . Data distribution normality and homogeneity of variance were assessed using the Shapiro-Wilk and Levene tests , respectively . The statistics were based on the two-sided Student’s t test , or one- or two-way ANOVA with Tukey’s HSD ( or HSD for unequal sample sizes where appropriate ) to correct for multiple post hoc comparisons . Within-groups variation is indicated by standard errors of the mean of each distribution , which are depicted in the graphs as error bars . P<0 . 05 was considered significant ( *P<0 . 05 , **P<0 . 01 , ***P<0 . 001 , ****P<0 . 0001 ) . | Drug addiction a is major mental health problem that presents a huge financial , social and legal burden worldwide . Adolescents are notoriously prone to drug abuse and addicts typically begin using drugs at a young age . However , an explanation for why young people are particularly vulnerable to the effects of addictive substances remains elusive . Addictive drugs change how the brain works , in particular by strengthening the connections ( synapses ) between brain cells ( neurons ) and making it easier for neurons to communicate with each other . Such strengthening of synaptic connections , which can be observed when the activity of the neurons is recorded with microelectrodes , relies on new proteins being made in the brain . Since adolescents have a greater capacity than adults to make new proteins , Huang et al . hypothesized that changes in synaptic strength might occur more easily in the brain of adolescents , explaining why they are more likely to become addicted to drugs than adults . A protein called eIF2α plays a key role in regulating the production of new proteins . Huang et al . discovered that reduced eIF2α activity accounts for why adolescents are particularly vulnerable to the synaptic and behavioral effects of cocaine . Giving adolescent mice a low dose of cocaine reduced the activity of eIF2α , caused an increase in the strength of synaptic connections in a part of the brain that processes pleasurable feelings , and promoted drug-reinforced behavior . This did not occur in adult mice . Reducing the activity of eIF2α using either genetics or pharmacological methods caused adult mice to become as vulnerable as adolescents to cocaine-induced changes in synaptic strength and addiction-related behavior . Conversely , increasing the activity of eIF2α made adolescent mice more resistant to cocaine’s effects; in other words , adolescents responded to cocaine more like adults . Huang et al . also found that other drugs of abuse , including alcohol , methamphetamine and nicotine , all reduce eIF2α activity , suggesting that eIF2α is a common target of different drugs of abuse . In a related study , Placzek et al . investigated the role of eIF2α in nicotine addiction in mice and humans . These findings raise several intriguing questions . How do cocaine and other drugs of abuse reduce eIF2α activity ? Could variations in the activity of eIF2α or other components of the eIF2α pathway in the brain explain why some people are more likely to abuse drugs ? Finally , could compounds that regulate the activity of eIF2α be useful for treating addiction ? | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"neuroscience"
] | 2016 | Translational control by eIF2α phosphorylation regulates vulnerability to the synaptic and behavioral effects of cocaine |
Mechanoelectrical transduction by hair cells commences with hair-bundle deflection , which is postulated to tense filamentous tip links connected to transduction channels . Because direct mechanical stimulation of tip links has not been experimentally possible , this hypothesis has not been tested . We have engineered DNA tethers that link superparamagnetic beads to tip links and exert mechanical forces on the links when exposed to a magnetic-field gradient . By pulling directly on tip links of the bullfrog's sacculus we have evoked transduction currents from hair cells , confirming the hypothesis that tension in the tip links opens transduction channels . This demonstration of direct mechanical access to tip links additionally lays a foundation for experiments probing the mechanics of individual channels .
Hair cells occur universally in the auditory and vestibular systems of vertebrates , where they transduce mechanical stimuli into electrical responses and thus initiate the perception of sounds and accelerations ( Hudspeth , 2008 , 2014 ) . The apical surface of a hair cell bears a hair bundle comprising actin-filled stereocilia whose gradation in height forms a bevel ( Figure 1A ) . The tip of each stereocilium is connected to the side of the tallest adjacent stereocilium through a protein filament , the tip link ( Pickles et al . , 1984 ) . The upper two-thirds of this link comprises a parallel dimer of cadherin 23 ( CDH23 ) molecules whereas the lower third encompasses a parallel dimer of protocadherin 15 ( PCDH15 ) molecules ( Kazmierczak et al . , 2007 ) . 10 . 7554/eLife . 16041 . 003Figure 1 . Experimental configuration and control experiments . ( A ) In diagrams of a hair cell at rest ( top ) and during excitatory stimulation ( bottom ) , the magnified images show how hair-bundle deflection is thought to raise the tension in tip links and thus open transduction channels . ( B ) A diagram , not to scale , portrays the molecular assembly that tethers a superparamagnetic bead to a tip link . ( C ) In overlaid fluorescence and differential-interference-contrast images of the apical surface of a bullfrog's sacculus , anti-CDH23 immunolabels the tops of hair bundles . ( D ) Labeling is absent under otherwise identical conditions when the primary antibody is omitted . ( E ) A differential-interference-contrast micrograph of a sacculus shows clusters of DNA-tethered superparamagnetic beads atop many hair bundles . ( F ) No beads are present in a control preparation lacking the primary antibody . DOI: http://dx . doi . org/10 . 7554/eLife . 16041 . 003 Deflection of a hair bundle toward its tall edge opens mechanoelectrical-transduction channels that allow an influx of cations , predominantly K+ but also Ca2+ , and thus depolarizes the hair cell . Electrophysiological and mechanical evidence suggests that deflection increases the tension in an elastic structure , the gating spring , that communicates force to the transduction channels ( Corey and Hudspeth , 1983a ) . Several lines of circumstantial evidence support the hypothesis that the tip link constitutes at least a portion of the gating spring: the stereociliary tips are the site of transduction ( Hudspeth , 1982; Beurg et al . , 2009 ) ; the orientation of the tip links corresponds to the hair bundle's axis of maximal sensitivity ( Shotwell et al . , 1981 ) ; and the responsiveness vanishes when the tip links are disrupted ( Assad et al . , 1991 ) . Nevertheless , the critical role of tip-link tension in gating mechanoelectrical-transduction channels has never been tested directly . Because they occur in the narrow spaces between the tips of contiguous stereocilia , tip links are not readily accessible to experimental manipulation . As a consequence , the stimulation of transduction channels has heretofore been limited to the deflection of entire hair bundles . We have developed a method of applying mechanical force specifically to tip links , allowing us to test directly the hypothesis that tip-link tension gates the transduction channels .
To apply tension at a defined position on each of a large number of tip links , we raised a polyclonal antiserum against a specific epitope near the carboxyl terminal of the bullfrog's CDH23 , and thus at the links' upper ends . Because each tip link resides in a confined space at the hair bundle's top , it is impractical to ligate the antibodies directly to a probe particle for use with optical or magnetic tweezers . We therefore created a mechanical tether for the tip link by interposing a 3 kb , double-stranded DNA molecule between the attached antibody and a 1 μm-diameter superparamagnetic bead ( Figure 1B ) . The DNA molecule was obtained by PCR of a plasmid with primers that incorporated biotin and fluorescein moieties at the molecule's opposite ends . The biotinylated terminus bound through a streptavidin molecule to a biotinylated anti-rabbit IgG , which in turn bound the rabbit anti-CDH23 . The fluorescein-modified end of the DNA adhered to a superparamagnetic bead coated with anti-fluorescein . The tether provided sufficient separation between the tip link and the relatively large bead to overcome the problem of steric inaccessibility . Control experiments confirmed that the DNA tethers functioned as intended . First , the anti-CDH23 antibodies labeled the stereociliary tips in living hair cells of the bullfrog's sacculus ( Figure 1C , D ) . The antibodies' binding to cadherin isoforms was also confirmed for both saccular and brain tissue by Western blot analysis . Following synthesis of the complete tether , superparamagnetic beads localized to the regions of the tip links but were absent if the primary antibody , secondary antibody , or streptavidin was omitted during the process of tether synthesis ( Figure 1E , F ) . Finally , as expected for DNA tethers , treatment of the preparation with DNAse liberated the superparamagnetic beads . The otolithic membrane that covered the hair bundles was removed enzymatically , exposing the tip links to the components of the DNA tethers . After completing tether synthesis we placed the chamber beneath an electromagnet and used magnetic force to pull the tethered beads directly upward , parallel to the stereocilia . If an increase in tip-link tension opens mechanoelectrical-transduction channels , pulling the links upward should elicit ionic currents into the hair cells and thus produce a transepithelial electrical response termed the microphonic potential ( Corey and Hudspeth , 1983b ) . Detected with electrodes in the two compartments of a recording chamber , such a signal reflects the total transduction current traversing the saccular epithelium ( Figure 2A ) . 10 . 7554/eLife . 16041 . 004Figure 2 . Electrical responses to magnetic stimulation . ( A ) The diagram shows a sacculus mounted in a two-compartment chamber beneath an electromagnet . The otolithic membrane has been removed and superparamagnetic beads ( red dots ) are attached to the tip links through DNA tethers . ( B ) Brief magnetic stimuli of opposite polarities , which are denoted by the voltages applied to the magnet , elicit transepithelial electrical responses . Exposure of the preparation to 1 mM gentamicin blocks these physiological responses , leaving only artifacts that are eliminated by averaging the signals obtained from stimuli of opposite polarity . ( C ) Longer stimuli demonstrate the persistence of transduction currents without the extensive adaptation observed during conventional stimulation . The stimulus traces shown produced the largest response; successively smaller responses were generated by lowering the voltage supplied to the magnet in steps of 0 . 2 V . ( D ) To eliminate the possibility that magnetic stimulation pulls hair bundles in the excitatory direction , most of the beads can be removed and the preparation tilted by 10° . ( E ) Brown superparamagnetic beads initially cover the apical surface of a mounted sacculus ( left ) . Scraping away most of the beads ( center ) leaves only the beads attached to hair bundles of similar orientations ( arrowheads ) , sensitive to stimulation away from the saccular nerve ( N ) . Further manipulation removes nearly all the beads ( right ) . The dashed line circumscribes the region in which hair cells occur . ( F ) Although smaller than the control response ( top ) , an electrical response nevertheless persists after removal of most beads ( middle ) . With the beads localized to one region with similar hair-bundle orientations , magnetic force pulls those bundles upward and toward their short edges . Removing the remaining beads eliminates the response ( bottom ) . The calibration bars in ( C ) apply to this panel as well . ( G ) Addition to the artificial perilymph in the top chamber of 1 mM EGTA , a Ca2+ chelator that dissociates tip links , abolished the electrical response . The calibration bars in ( C ) apply to this panel as well . ( H ) When a superparamagnetic bead is tethered by DNA to a glass surface through DNA , thermal motion causes lateral deflections that are partially suppressed by the magnetic force F . DOI: http://dx . doi . org/10 . 7554/eLife . 16041 . 004 Upon the application of current through the electromagnet in either direction , we observed electrical responses whose polarity was consistent with inward cationic currents through the apical surfaces of the hair cells ( Figure 2B ) . To confirm that these signals originated primarily from transduction channels , we conducted control experiments in the presence of 1 mM gentamicin , a blocker of transduction channels ( Kroese et al . , 1989 ) . We then detected only small , magnetically induced artifacts during stimulation of either polarity . Moreover , no response was observed when the primary antiserum was omitted from the preparative procedure . As a result of the electromotive force induced by the changing magnetic field , we expected our recordings to include stimulus artifacts . Because the pulling force on a superparamagnetic bead does not depend on the direction of the current through the electromagnet , the physiological responses to stimuli of opposite polarities should have been identical . The artifacts , however , should have reversed signs in accordance with Faraday’s law . To separate the transduction responses from artifacts , we therefore applied to the magnet successive voltage pulses of a constant magnitude but opposite polarities . By averaging the responses to these two stimuli we removed the artifacts and preserved only the responses owing to mechanoelectrical transduction ( Figure 2B ) . The peak response in nine experiments was 38 . 7 ± 9 . 8 μV ( mean ± standard deviation ) , roughly one-tenth the maximal value expected if every tip link were stimulated . That the responses to currents of opposite polarity were largely the same also indicated that the signals originated primarily from mechanoelectrical transduction rather than electromagnetic interference . As expected , gentamicin treatment revealed that the residual artifacts were equal and opposite and summed to zero . Although the signals elicited by conventional hair-bundle deflection adapt rapidly to a plateau ( Eatock et al . , 1987; Shepherd and Corey , 1994 ) , the responses elicited by protracted magnetic stimulation declined gradually over several seconds ( Figure 2C ) . Conventional stimulation of a hair bundle is thought to tense each tip link and open the associated channels . Aided by the effect of Ca2+ on the myosin molecules that secure the upper ends of the tip links , this tension then declines , resulting in adaptation of the electrical response ( Howard and Hudspeth , 1987; Assad et al . , 1989; Hacohen et al . , 1989 ) . Magnetic stimulation , in contrast , should pull the top ends of the tip links upward . This force not only opened transduction channels but evidently restrained the insertional plaques from descending as well . Recovery from the vertical offset of the plaques might explain why the electrical response relaxed far more slowly than during conventional adaptation . Because the magnitude of the response was sensitive to the strength of magnetic stimulation , the long plateau observed during protracted magnetic stimulation did not stem from saturation of the response . Imperfect positioning of the saccular epithelium in the experimental chamber might have created a tilt between the preparation and the direction of the magnetic-field gradient . If this were the case , any component of magnetic force directed toward the tall edges of hair bundles would have excited a response by the conventional mechanism of bundle deflection . Although that form of stimulation would be difficult to reconcile with the protracted adaptation observed during extended magnetic stimulation , we nevertheless performed an additional experiment to eliminate this possibility and ensure that the measured potentials resulted from upward tension in the tip links . The preparation was positioned in a sample holder with a 10° tilt between the saccular epithelium and the plane of the magnet's pole piece ( Figure 2D ) . After tethering superparamagnetic beads as usual , we scraped away most of the beads with an eyelash , leaving beads and intact hair bundles only in the segment of the epithelium opposite the saccular nerve ( Figure 2E ) . Throughout that region the hair bundles display a common direction of excitability away from the nerve . In the tilted configuration any magnetic force tangent to the epithelial surface would have pulled the hair bundles toward their short edges and would therefore have countered the conventional electrical response . Even under these conditions we observed a transduction current of the usual polarity but of diminished magnitude owing to the removal of most tethered beads ( Figure 2F ) . As expected , the response vanished when the remaining beads had been scraped away . Attachment of the magnetic tethers to cadherins on hair-cell surfaces rather than to fully formed tip links might elicit extraneous responses independent of the transduction process . To eliminate this possibility , we dissociated tip links by treatment with 1 mM EGTA . Under these conditions the summed response fell to zero ( Figure 2G ) , indicating that the responses stemmed from traction on tip links . In order to quantify the stimulus , we calibrated the force exerted by the magnet on a superparamagnetic bead attached to a glass surface through a DNA tether ( Gosse and Croquette , 2002 ) . The lateral Brownian fluctuations of the tethered bead were measured at 250 frames per second with a video camera ( Redlake MotionScope 2000S , DEL Imaging Systems ) . When the magnetic field was applied with the pole piece 4 . 1 mm distant , the axial force on the bead partially suppressed its lateral Brownian fluctuations . The equipartition theorem relates the axial pulling force F and the magnitude of lateral fluctuations of the bead's position ⟨δX2⟩: ( 1 ) F=kT ( L⟨δX2⟩ ) , in which k is the Boltzmann constant , T the temperature , and L the time-averaged end-to-end length of the DNA molecule ( Figure 2H ) . The wormlike-chain model for the extension of DNA under a load then relates F to L ( Bustamante et al . , 1994 ) : ( 2 ) F= ( kTP ) [ 14 ( 1−L/L0 ) 2+LL0−14 ] , in which P is the DNA molecule's persistence length of 50 nm and L0 is its contour length of 3000 base pairs or 1 . 02 μm . By measuring the lateral displacement ⟨δX2⟩ and numerically solving Equations ( 1 ) and ( 2 ) for 16 determinations , we estimated a magnetic force at the position of the specimen of 0 . 48 ± 0 . 07 pN ( mean ± standard deviation ) .
Because the inaccessibility of tip links obstructs their direct manipulation , mechanical stimulation of hair cells has heretofore been restricted to deflections of entire hair bundles . By overcoming this limitation , our experiments have demonstrated that upward force on tip links generates transduction currents independent of hair-bundle deflection . As expected , these responses vanish in the presence of a channel blocker . The observed responses accord well with the values expected on the basis of the calibrated magnetic force on tethered beads and the gating-spring theory for transduction by hair cells . More than three decades after the postulation that tension in gating springs within the hair bundle opens transduction channels ( Corey and Hudspeth , 1983a ) and the subsequent suggestion that tip links constitute those elements ( Pickles et al . , 1984 ) , the present observations provide direct support for both hypotheses . The technical approach introduced here additionally affords a means of investigating mechanoelectrical transduction at the scale of individual tip links and channels .
Wishing to develop an antiserum against an epitope of the bullfrog's CDH23 near the upper insertion of the tip link , we selected the region corresponding to the Ca2+-binding linker between extracellular cadherin domains EC25 and EC26 of the murine protein ( UniProtKB/Swiss-Prot: Q99PF4 . 2 ) . This region was selected on the basis of its probable accessibility and antigenicity and because it occurs near the uppermost of CDH23's 27 extracellular cadherin domains . Using primers based on the DNA sequence for the zebrafish's cdh23 ( NM_214809 ) , we obtained a product by RT-PCR of the RNA isolated from a single bullfrog's sacculus ( Superscript III One-Step RT-PCR System , Invitrogen ) . This cDNA was inserted into the vector pCR2 . 1-TOPO and sequenced to confirm its identity . The corresponding polypeptide DDNEPIFVRPPRGA was synthesized with the addition of a cysteine residue at its amino terminal and with amidation of its carboxyl terminal . Following the immunization of two rabbits ( Covance ) , the resultant serum ( RU1793 ) was affinity-purified against the immunizing peptide . After 250 μL of 1 μm-diameter carboxylate-functionalized superparamagnetic beads ( MyOne , Life Technologies ) had been washed twice in 15 mM 2- ( N-morpholino ) ethanesulfonate at pH 6 , they were resuspended in 45 μL of the same solution . Following the addition of 5 μL of 52 mM 1-ethyl-3- ( 3-dimethylaminopropyl ) carbodiimide ( Thermo Scientific ) the beads were tumbled for 30 min . The beads were then sedimented with a magnet and the solution was decanted . After 75 μL of rabbit anti-fluorescein ( Thermo Scientific ) had been diluted into 100 μL of the foregoing buffer solution , 100 μL of the product was added to the beads and tumbling was continued for 16 hr . Following two rinses in Ca2+- and glucose-free artificial perilymph , the beads were resuspended in 125 μL of that solution . Double-stranded , 3 kb DNA molecules modified at their opposite ends with biotin and fluorescein were produced by PCRs from the plasmid PGBKT7; the primers were 5′-biotin·dT-TTCTGGCAACCAAACCCATACATCG-3’ and 5′-fluorescein·dT-TTCTATGAAAGGTTGGGCTTCGGA-3’ ( IDT DNA ) . Each such DNA molecule has a contour length of about 1 μm but displays a radius of gyration about one-third that great when unloaded . After the DNA had been eluted into 10 mM tris ( hydroxymethyl ) aminomethane at pH 8 and concentrated to 700 ng/μL , 4 μL of the resultant solution was mixed with an equal volume of 1 mg/mL streptavidin ( Sigma ) in Ca2+- and glucose-free artificial perilymph and incubated at room temperature for 30 min . Following the addition of 20 μL of functionalized superparamagnetic beads to this mixture and tumbling for 30 min , the beads were sedimented with a magnet , washed five times with blocking solution consisting of 10% normal goat serum ( Jackson Research ) in artificial perilymph , and resuspended in 80 μL of blocking solution . Pipetting was conducted gently to prevent rupture of the DNA tethers by excessive shear . After the apical surface of a saccular macula mounted in a two-compartment chamber had been exposed to blocking solution for 10 min , the solution in the top chamber was replaced for 10 min with anti-CDH23 diluted 1:100 in blocking solution . That chamber was rinsed four times with blocking solution and biotinylated goat anti-rabbit secondary antibody ( Life Technology ) was added for 10 min at a 1:100 dilution . The top chamber was then rinsed five times with blocking solution and 40 μL of DNA-tethered superparamagnetic beads was gently added to the chamber . Following incubation for 20 min , excess beads were gently washed away with blocking solution . All procedures were approved by the University's Institutional Animal Care and Use Committee . Sacculi were dissected from adult bullfrogs ( Rana catesbeiana ) of both sexes into oxygenated artificial perilymph containing 114 mM Na+ , 2 mM K+ , 0 . 25 mM Ca2+ , 116 mM Cl− , 5 mM HEPES , and 3 mM D-glucose . After the removal of otoconia , each saccular macula was sealed with n-butyl cyanoacrylate ( Vetbond , 3M ) over a 1 mm-diameter hole in a 12 mm-diameter plastic disk to form the partition in a two-compartment recording chamber . The apical surface was exposed to 67 µg/mL of protease ( type XXIV; Sigma ) for 30 min at 22°C to loosen the otolithic membrane , which was removed with an eyelash . Both compartments of the recording chamber were then filled with artificial perilymph . The recording chamber was centered 4 . 1 mm below the pole piece of a solenoidal electromagnet ( E-40-300-15 , Magnetic Sensor Systems ) that was actuated by a power amplifier ( PA-119 , Labworks ) . Experimental control , stimulation , and recording were effected with a data-acquisition card ( PCIe-6353 , National Instruments ) and data were analyzed with MATLAB . The potential difference across the tissue was measured with a directly coupled extracellular amplifier ( EXT-02F , NPI Electronic ) with a gain of 200X and a low-pass cutoff frequency of 1 . 3 kHz . The response was recorded differentially and is presented as the potential in the upper chamber referenced to that in the lower . Each record represents the average of eight repetitions of each stimulus polarity and was smoothed to a maximal frequency of 1 kHz by rolling-window averaging . A cover glass was washed with HCl , coated for 30 min with 1 mg/mL concanavalin A , and rinsed with water . The treated cover glass was sequentially exposed to anti-CDH23 diluted 1:100 in PBS , washed with blocking solution , treated with biotinylated secondary antibody diluted 1:100 in blocking solution , washed with blocking solution , treated with DNA functionalized at either end with biotin and fluorescein at a concentration of 20 pM , washed with blocking solution , treated with superparamagnetic beads functionalized with anti-fluorescein diluted 1:10 in blocking solution , and finally washed with blocking solution . We compared our calibration of magnetic force with ab initio expectations , based on the nominal magnetic properties of the bead and our measurements of the strength and gradient of the magnetic field . In the linear range of magnetization of a bead of volume V owing to a flux density B , for which the magnetization M is related to the field strength H by the volume susceptibility χ , the induced magnetic dipole moment m is given by ( 3 ) m=VM=VχH= ( Vχμ0 ) B . Because the potential energy U of a magnetic dipole is U=−m⋅B/2 , the vertical component of force FZ on the dipole is ( 4 ) FZ=−∇→ZU=12∇→Z ( Vχμ0|B|2 ) = ( Vχμ0 ) |B|∂|B|∂Z . Here the positive value reflects the fact that , regardless of the direction of current through the electromagnet , a superparamagnetic bead is pulled into an increasing magnetic field , in this instance upward . Using a Hall-effect magnetometer ( OHS3150U; TT Electronics/Optek Technology ) , we measured a magnetic field strength of 49 mT ( 490 G ) at a distance of 4 . 1 mm from the magnet's pole piece and along its center axis . Measurements made after slight deflections of the detector yielded an estimated gradient of 1 . 6 T·m−1 . Substituting these values and the manufacturer's stated volume susceptibility of 1 . 4 into the foregoing relation leads to an estimate for the vertical force on each bead of 0 . 05 pN , lower than our calibrated value of 0 . 48 pN . Ab initio calculations provide only a lower limit to the magnetic force , however , and require scaling by an empirically determined factor to accord with measured values ( Lipfert et al . , 2009; Shevkoplyas et al . , 2007 ) . The gating-spring theory of transduction ( Corey and Hudspeth , 1983a; Hudspeth , 1992; Markin and Hudspeth , 1995 ) asserts that the change in energy ΔE associated with the opening of an individual transduction channel is ( 5 ) ΔEC→O=ΔE∅−κd ( γX+xC−d/2 ) , in which ΔE∅ is the intrinsic component of gating energy in the absence of tip-link tension , k the gating-spring stiffness , d the swing of a transduction channel's gate , γ the bundle's geometrical gain , and xC the elongation of the gating spring when attached to a closed channel in a resting bundle . For an undisturbed hair bundle , the displacement X is zero , so each channel's open probability P0 is ( 6 ) PO=11+eΔEC→O/kT=11+e[E∅−κd ( xC−d/2 ) ]/kT , in which k is the Boltzmann constant and T the temperature . If the gating-spring tension is increased , for example by a magnetically induced force F , the expected change in open probability is ( 7 ) dPOdxC=PO ( 1−PO ) ( κdkT ) . In the saline solution used for our experiments , P0≈0 . 2 , k≈500μN⋅m−1 , and d≈5nm ( Corey and Hudspeth , 1983b; Howard and Hudspeth , 1988; Martin et al . , 2000 ) , so we anticipate a sensitivity of dP0/dxC≈108m−1 . For a tip link of the stated stiffness , this value represents a responsiveness to force of dP0/dF≈2⋅1011N−1 . According to our force calibrations , magnetic stimulation induces a force of ΔF=0 . 48pN , which would thus be expected to enhance the open probability by ΔP0≈0 . 10 . The ab initio calculation estimates ΔF=0 . 05pN , which would imply ΔP0≈0 . 01 . There are two ways to estimate the maximal transepithelial current that might be evoked by activating all the transduction channels . First , a cluster of hair bundles representing about one-seventh of the saccular complement produces a peak transepithelial response near 500 μV during conventional stimulation ( Corey and Hudspeth , 1983b ) , so activating every cell in the sensory epithelium would yield about 3500 μV . The second approach is to consider that an individual hair cell from the frog's sacculus typically produces a peak transduction current of 200 pA ( Gillespie and Hudspeth , 1993; Cheung and Corey , 2006 ) , so the 2000 or so cells in a medium-sized sacculus ( Jacobs and Hudspeth , 1990 ) would yield a total current of about 400 nA . Returning across an epithelium of resistance 10 kΩ , that current would yield a transepithelial signal of 4000 μV , a value in good agreement with the first calculation . If magnetic stimulation were to increase the open probability of all the preparation's transduction channels by 10% as calculated from our calibration of the magnetic force , the magnetic stimuli would be expected to produce signals of 350–400 μV at most . The value would be approximately one-tenth as great on the basis of ab initio calculations . Because only a small fraction of the tip links are attached to beads , however , the actual responses should be still smaller . | In animals with backbones , the inner ear is responsible for both hearing and balance . Sound waves and head movements apply a mechanical force to hair cells inside the inner ear . This causes the cells to produce electrical signals that ultimately communicate information about the sound or movement to the brain . The apparatus that converts mechanical forces into electrical signals is called the hair bundle , which is an upright cluster of small rods called stereocilia that protrude from the hair cell's flattened top surface . Fine filaments called tip links connect the stereocilia within a hair bundle to one another . It is thought that the mechanical deflection of a hair bundle tenses the tip links and opens ion channels – molecular pores through which ions can pass – that are attached to the tip links . The resultant flow of ions across the hair cell's membrane would then cause a voltage change that in turn triggers the cell’s electrical response . It has not been possible to test this hypothesis , however , because the position of the tip links within a hair bundle prevents them from being stimulated directly in experiments . Basu et al . have now used specific antibody molecules to attach tip links to magnetic beads using a strand of DNA . The DNA acted as a string that penetrated into the hair bundles , connecting the tip links to magnetic beads outside the bundles . This meant that moving the bead by applying a magnetic force to it pulled upon the tip links , and the investigators observed that this activated the associated ion channels . The resultant electrical signals confirmed that tip links play a role in the responses of hair cells . Although there are methods that allow the electrical activity from a single ion channel to be recorded , the new approach provides an opportunity for studying the mechanical activity of a channel as well . Future studies could therefore investigate the mechanical and electrical signals associated with individual tip links and the ion channels to which they attach in order to investigate the specific role they play in hearing . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"short",
"report",
"physics",
"of",
"living",
"systems",
"neuroscience"
] | 2016 | Direct mechanical stimulation of tip links in hair cells through DNA tethers |
Small molecule inhibitors of site-specific O-glycosylation by the polypeptide N-acetylgalactosaminyltransferase ( ppGalNAc-T ) family are currently unavailable but hold promise as therapeutics , especially if selective against individual ppGalNAc-T isozymes . To identify a compound targeting the ppGalNAc-T3 isozyme , we screened libraries to find compounds that act on a cell-based fluorescence sensor of ppGalNAc-T3 but not on a sensor of ppGalNAc-T2 . This identified a hit that subsequent in vitro analysis showed directly binds and inhibits purified ppGalNAc-T3 with no detectable activity against either ppGalNAc-T2 or ppGalNAc-T6 . Remarkably , the inhibitor was active in two medically relevant contexts . In cell culture , it opposed increased cancer cell invasiveness driven by upregulated ppGalNAc-T3 suggesting the inhibitor might be anti-metastatic . In cells and mice , it blocked ppGalNAc-T3-mediated glycan-masking of FGF23 thereby increasing its cleavage , a possible treatment of chronic kidney disease . These findings establish a pharmacological approach for the ppGalNAc-transferase family and suggest that targeting specific ppGalNAc-transferases will yield new therapeutics .
Glycosylation , which fine-tunes the function of proteins , is the most abundant and diverse posttranslational modification ( Schjoldager and Clausen , 2012; van der Post et al . , 2013; Moremen et al . , 2012 ) . Despite many documented roles in health and disease , the enzymes that mediate mucin-type O-glycosylation in the Golgi apparatus have yet to be discovered as druggable targets . The initiating enzymes , a family of 20 ppGalNAc-transferase isozymes , determine which substrates are modified and at which sites . Significant questions remain regarding their specificity , regulation , targets and functions and the lack of a pharmacological approach has been a critical limitation . The only confirmed inhibitor of mucin-type O-glycosylation , benzyl-N-acetyl-α-galactosaminide , blocks elongation rather than initiation and requires millimolar concentrations that can be toxic ( Kuan et al . , 1989; Patsos et al . , 2009 ) . Not only are pan-specific modulators of the ppGalNAc-transferases lacking , there is nothing isoform-specific . The latter will be essential to restrict effects to particular pathways and could possibly lead to a new basis for therapeutics conceptually related to the widespread use of drugs targeting individual protein kinases . ppGalNAc-T3 serves as an important test case as it is implicated in at least two medically important pathways: cancer metastasis and stabilization of FGF23 ( Chefetz and Sprecher , 2009; Schjoldager et al . , 2011; Kato et al . , 2006; Peng et al . , 2012; Kohsaki et al . , 2000; Kitada et al . , 2013; Gao et al . , 2013; Brockhausen , 2006; Brooks et al . , 2007 ) . ppGalNAc-T3 is overexpressed in cancerous tissue often correlating with shorter survival ( Kitada et al . , 2013; Brockhausen , 2006; Harada et al . , 2016; Mochizuki et al . , 2013 ) . Knockdown of ppGalNAc-T3 expression in cultured ovarian cancer cells inhibits their invasive capacities arguing that ppGalNAc-T3 has potential as a therapeutic target ( Wang et al . , 2014 ) . ppGalNAc-T3 mediates glycan-masking of FGF23 in bone as part of a control mechanism determining the form of FGF23 that is secreted ( Kato et al . , 2006; Tagliabracci et al . , 2014 ) . When present , the added O-glycan blocks FGF23 cleavage by the furin protease resulting in secretion of intact FGF23 that activates FGF23 receptor complexes at the kidney and intestine ( Rowe , 2015 ) . In contrast , non-glycosylated FGF23 is cleaved and the cleaved C-terminal product competitively blocks these same receptors ( Goetz et al . , 2010 ) . Significantly , elevated intact FGF23 occurs in chronic kidney disease and upon kidney transplant where it is directly linked to poor prognosis due to its effects on renal phosphate reabsorption and 1 , 25-dihydroxyvitamin D biosynthesis ( Eckardt and Kasiske , 2009 ) . Many studies have concluded that therapeutic control over FGF23 would be transformative in the clinic ( Degirolamo et al . , 2016; Fukumoto , 2016; Isakova et al . , 2015; Smith , 2014 ) . As a step toward testing whether biologic discoveries and new therapeutics will result from developing small molecule modulators that control the activity of specific ppGalNAc-transferases we initiated high throughput cell-based screening of compound libraries . An inhibitor of ppGalNAc-T3 was identified and found to block breast cancer cell invasiveness as well as secretion of intact FGF23 promising new therapeutic approaches for cancer and chronic kidney disease , respectively .
HEK cell lines were engineered to express fluorescent sensors that are specific to ppGalNAc-T2 or ppGalNAc-T3 activity ( Song et al . , 2014 ) . For each sensor , glycosylation of its isozyme-specific target site prevents furin protease from removing a blocking domain ( Figure 1A ) . Thus , fluorescence increases upon ppGalNAc-transferase inhibition because removal of the blocking domain allows dimerization of a fluorogen activating protein domain so that it binds and activates the fluorescence of malachite green . They are ratiometric because the sensor backbone contains a green fluorescent protein as an internal control for expression . Each sensor showed clear activation after mutation of its glycan acceptor sites and these mutated constructs served as positive controls in the screen ( Figure 1B , ∆glycan ) . The T3 sensor exhibited a background level of activation due to incomplete glycosylation ( Song et al . , 2014 ) but this was considered advantageous for the possible identification of enzyme activators along with the desired inhibitors . Consistent with our previous report ( Song et al . , 2014 ) , sensor expression in HEK cell lines depleted of either ppGalNAc-T2 or T3 via zinc finger nuclease editing resulted in specific activation of the corresponding sensor confirming their isozyme selectivity ( Figure 1B , HEK∆T2 , HEK∆T3 ) . Our screen included compounds based on structural diversity ( 21 , 710 compounds in total ) with 6 hr treatments at 10 µM prior to flow cytometry to assay MG and GFP fluorescence on a cell-by-cell basis ( Figure 1C ) . Each compound was tested in duplicate and against both sensors ( Figure 1—source data 1 ) . Because each sensor requires essentially identical cellular reactions- the only difference being which ppGalNAc-transferase isoform modifies the sensor- most off-target hits ( such as sugar nucleotide transporters , extending enzymes , or furin ) will alter both sensors , whereas directly acting , isoform-specific candidates will be sensor-specific . Using cut-off parameters for the MG/GFP ratios that excluded >99% of the compounds ( Q ≥ 3 or Q ≤ −2 . 5 ) , the screen yielded 72 sensor-specific hits with 18 increasing and 35 decreasing the T2 sensor fluorescence and 11 increasing and 8 decreasing the T3 sensor fluorescence ( Figure 1D ) . 10 . 7554/eLife . 24051 . 003Figure 1 . Screen for modulators of ppGalNAc-T2/T3 . ( A ) Diagram showing sensor design and the linker sequences used . O-glycosylation of the linker masks the furin site but if an inhibitor blocks the ppGalNAc-transferase then furin cleaves the linker releasing the blocking domain ( BD ) allowing fluorescent activating protein ( FAP ) dimerization and dye activation . Linker furin sites are underlined and sites of glycosylation or mutation are in bold . ( B ) HEK cell lines with or without ppGalNAc-T2 or T3 stably expressing the WT or ∆glycan T2 or T3 sensor constructs ( see linkers in A ) were imaged in the presence of 110 nM of the dye MG11p ( MG ) to detect GFP or MG . Bar = 20 µm . ( C ) Schematic showing cell plating , drug treatment , cell release , fluorescence measurement and parallel analysis using both T2 and T3 sensors . Hits that activate both may be pan-specific or act on off-target pathways common to both sensors whereas sensor specific hits are likely acting directly on the corresponding ppGalNAc-transferase . ( D ) The plot shows Q values ( Q= ( R-RNeg ) /SDNeg ) for each compound ( treatment at 10 µM for 6 hr ) using the average of duplicate MG/GFP ratios for the compound ( R ) , the vehicle-only control ( RNeg ) , and the standard deviation of the vehicle-only controls ( SDNeg ) . The cut-off values of +3 and −2 . 5 are indicated ( * ) . Also indicated are the values for the positive controls ( T2∆glycan and T3∆glycan ) and the structure of the indicated T3-specific hit ( inset ) . ( E ) Values ( % enzyme activity relative to vehicle-only controls ) in the in vitro assay using purified ppGalNAc-T2 or T3 as a secondary screen are shown for 20 hits from the primary screen . Compounds were present at 50 µM . Compound 1614 is T3Inh-1 . DOI: http://dx . doi . org/10 . 7554/eLife . 24051 . 00310 . 7554/eLife . 24051 . 004Figure 1—source data 1 . Primary screen data for HEK cells expressing T2 or T3 sensors . The accompanying spreadsheet shows calculated Q values ( see Methods ) for each compound tested . Note that autofluorescent compounds are left blank . DOI: http://dx . doi . org/10 . 7554/eLife . 24051 . 00410 . 7554/eLife . 24051 . 005Figure 1—source data 2 . Secondary screen data ( in vitro enzyme assays ) . DOI: http://dx . doi . org/10 . 7554/eLife . 24051 . 00510 . 7554/eLife . 24051 . 006Figure 1—figure supplement 1 . Cell growth at various T3Inh-1 exposures . Identical numbers of HEK cells were plated and grown in the continuous presence of the indicated concentrations of T3Inh-1 and then at 24 , 48 , or 72 hr they were released and counted . Averages are shown normalized using the untreated sample at 72 hr ( n = 3 ± SEM ) . DOI: http://dx . doi . org/10 . 7554/eLife . 24051 . 00610 . 7554/eLife . 24051 . 007Figure 1—figure supplement 1—source data 1 . Cell counts at differing time points and T3Inh-1 concentrations . DOI: http://dx . doi . org/10 . 7554/eLife . 24051 . 00710 . 7554/eLife . 24051 . 008Figure 1—figure supplement 2 . General N- and O-glycosylation are unaffected . ( A , C , E , G ) Representative fluorescent emission spectra from 510–550 nm of lysates obtained from HeLa cells treated with T3Inh-1 for 24 hr at 0 , 10 , or 20 µM . The cells were stained with the indicated lectin for 30 min just prior to analysis . ( B , D , F , H ) Quantified average staining values for the indicated lectins ( at 520 nm emission ) and T3Inh-1 treatments ( n = 3 ± SEM ) . DOI: http://dx . doi . org/10 . 7554/eLife . 24051 . 00810 . 7554/eLife . 24051 . 009Figure 1—figure supplement 2—source data 1 . Fluorescent lectin staining of cells at differing T3Inh-1 concentrations . DOI: http://dx . doi . org/10 . 7554/eLife . 24051 . 00910 . 7554/eLife . 24051 . 010Figure 1—figure supplement 3 . ppGalNAc-transferase levels are unaffected . ( A ) Representative images of untreated or T3Inh-1 treated ( 6 hr , 10 µM ) HeLa cells after fixation and staining with antibodies against the indicated ppGalNAc-transferase and the Golgi marker GPP130 . Bar = 10 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 24051 . 010 To determine which of these directly acted on the targeted enzymes , we carried out in vitro glycosylation assays in which the purified lumenal domains of ppGalNAc-T2 or ppGalNAc-T3 ( containing catalytic and lectin domains ) were incubated with peptide and UDP-GalNAc substrates in the presence of 50 µM of each compound . A second stage reaction ( UDP-Glo ) then converted the accumulated UDP product to ATP and then , via luciferase , to light . This resulted in 20 candidates that either reduced or increased the luminescence by a factor ≥50% relative to vehicle-only controls ( Figure 1E ) . Of these , one compound ( #1614 ) stood out as a strong and selective inhibitor of ppGalNAc-T3 and became the focus of this phase of the study . The compound is a quinoline of no known activity that we now refer to as ppGalNAc-T3 Inhibitor 1 or T3Inh-1 ( Figure 1D , inset ) . Importantly , T3Inh-1 exhibited no toxicity and did not affect HEK cell proliferation ( Figure 1—figure supplement 1 ) . Also , a 24 hr treatment did not affect staining intensity by the lectins Concanavalin A ( ConA ) , Wheat germ agglutinin ( WGA ) , Sambucus Nigra ( SNA ) or Vicia Villosa ( VVA ) , which bind branched alpha-mannose , N-acetylglucosamine , sialic acid , or terminal GalNAc , respectively ( Figure 1—figure supplement 2 ) . This implies that the enzymes contributing to the abundant glycans of N- and O-glycosylation detected by these lectins were unaffected and is consistent with ppGalNAc-T3 modifying a relatively limited number of substrates ( Schjoldager et al . , 2015 ) . Finally , there was no change in localization or expression level of ppGalNAc-T3 , ppGalNAc-T2 or any other Golgi marker tested ( Figure 1—figure supplement 3 ) . To determine the effective concentration of T3Inh-1 in cells and in vitro , it was retested at various doses against the sensors and against the purified enzymes . T3Inh-1 activated the T3 sensor with an apparent IC50 of 12 µM and showed little or no activity towards the T2 sensor ( Figure 2A ) . Similarly , T3Inh-1 was a potent and selective direct inhibitor of ppGalNAc-T3 ( Figure 2B ) . Inhibition of ppGalNAc-T3 occurred with an IC50 of 7 µM and was undetectable against ppGalNAc-T2 . T3Inh-1 also lacked activity against ppGalNAc-T6 ( Figure 2B ) , which is the isozyme considered most closely related to ppGalNAc-T3 ( Bennett et al . , 2012 ; Revoredo et al . , 2016 ) . 10 . 7554/eLife . 24051 . 011Figure 2 . T3Inh-1 is a direct mixed-mode inhibitor of ppGalNAc-T3 . ( A ) Comparison of T2 and T3 sensor activation at the indicated concentrations of T3Inh-1 ( n = 3 ± SEM ) . MG/GFP ratio was determined for 20 , 000 cells by FACS and average value is plotted as percent of the positive control ( i . e . the ∆glycan version of each sensor ) . ( B ) Comparison of effect of the indicated concentrations of T3Inh-1 on in vitro glycosylation mediated by purified ppGalNAc-T2 , ppGalNAc-T3 , or ppGalNAc-T6 . Values are averages expressed as percentage of the control ‘vehicle-only’ reactions ( n = 6 ± SEM for ppGalNAc-T3 , n = 3 ± SEM for others ) . ( C–D ) The in vitro assay was carried out in the presence of 0 , 7 . 5 , or 15 µM T3Inh-1 at the indicated concentrations of peptide or UDP-GalNAc substrate . Values are averages expressed as percent of the control reactions with no inhibitor and saturating substrates ( n = 3 ± SEM ) . ( E ) Representative fluorescence spectra are shown for T3Inh-1 alone or for purified ppGalNAc-T3 in the presence of the indicated concentrations of T3Inh-1 . Note dose-dependent quenching of tryptophan fluorescence indicating direct binding . ( F ) Fluorescence quenching was quantified at each concentration using the peak value at 324 nm ( n = 3 ± SEM ) . Note that all graphs have error bars but some are too small to be apparent . DOI: http://dx . doi . org/10 . 7554/eLife . 24051 . 01110 . 7554/eLife . 24051 . 012Figure 2—source data 1 . Panel A: sensor signals versus T3Inh-1 concentration . DOI: http://dx . doi . org/10 . 7554/eLife . 24051 . 01210 . 7554/eLife . 24051 . 013Figure 2—source data 2 . Panel B: enzyme activity versus T3Inh-1 concentration . DOI: http://dx . doi . org/10 . 7554/eLife . 24051 . 01310 . 7554/eLife . 24051 . 014Figure 2—source data 3 . Panel C: inhibitor effect versus peptide concentration . DOI: http://dx . doi . org/10 . 7554/eLife . 24051 . 01410 . 7554/eLife . 24051 . 015Figure 2—source data 4 . Panel D: inhibitor effect versus UDP-GalNAc concentration . DOI: http://dx . doi . org/10 . 7554/eLife . 24051 . 01510 . 7554/eLife . 24051 . 016Figure 2—source data 5 . Panel E and F: Fluorescence change versus T3Inh-1 concentration . DOI: http://dx . doi . org/10 . 7554/eLife . 24051 . 016 Towards characterizing the mechanism of inhibition we used the in vitro assay with purified ppGalNAc-T3 and individually varied both peptide and UDP-GalNAc substrate concentrations in the presence of 0 , 7 . 5 or 15 µM T3Inh-1 . The results were similar for both substrates ( Figure 2C–D ) , where T3Inh-1 decreased the Vmax and increased the Km ( Table 1 ) indicating a mixed-mode of inhibition in which the inhibitor most likely binds both free enzyme to reduce substrate binding and enzyme-substrate complexes to reduce turnover . Implied in this model of action is direct binding to the enzyme typically at an allosteric site . To test for direct binding , intrinsic tryptophan fluorescence of ppGalNAc-T3 was determined in the presence of increasing T3Inh-1 concentrations . At all concentrations , the compound itself yielded miniscule signals , whereas the compound had a profound and dose-dependent effect on the ppGalNAc-T3 emission spectrum ( Figure 2E ) . These results confirmed direct binding with an apparent Kd of 17 µM ( Figure 1F ) . The similarity in concentration dependence of sensor activation in cells , in vitro inhibition and direct binding argues that T3Inh-1 acts directly on cellular ppGalNAc-T3 and inhibits its activity . 10 . 7554/eLife . 24051 . 017Table 1 . Inhibition by T3Inh-1 at varying substrate concentrations . Values shown were determined from the data in Figure 2 using Prism ( see Materials and methods ) . DOI: http://dx . doi . org/10 . 7554/eLife . 24051 . 017SubstrateParameter0 µM7 . 5 µM15 µMPeptide ( EA2 ) Vmax100%82%36% Km ( µM ) 173 . 7208 . 4210 . 3 Ki ( µM ) 9 . 9UDP-GalNAcVmax100%71%56% Km ( µM ) 74 . 9153 . 4448 . 3 Ki ( µM ) 2 . 9 Given its validation as a direct inhibitor of ppGalNAc-T3 without obvious off-target effects we turned to biologically relevant tests of T3Inh-1 . As mentioned , overexpression of ppGalNAc-T3 is linked to cancer cell invasiveness as well as poor outcomes in patients ( Kitada et al . , 2013; Brockhausen , 2006; Harada et al . , 2016; Mochizuki et al . , 2013 ) . Although no linkage to breast cancer has been reported , our analysis of publically available data for 1117 breast cancer patients ( Szász et al . , 2016 ) using Kaplan-Meier survival plots shows that high expression of ppGalNAc-T3 correlates with poor patient overall and metastasis-free survival ( Figure 3—figure supplement 1A–B ) . Therefore , we carried out migration and invasion assays with the breast cancer cell line MDA-MB231 , which expresses a relatively high level of ppGalNAc-T3 ( Figure 3—figure supplement 1C ) , in the absence or presence of 5 µM T3Inh-1 . Cells were cultured on uncoated ( to assay migration ) or Matrigel-coated ( to assay invasiveness ) Bioboat filters for 24 or 48 hr and those cells that moved to the underside of the filters were imaged and quantified . T3Inh-1 was strikingly effective , inhibiting migration by >80% ( Figure 3A ) and invasion by 98% ( Figure 3B ) while causing no discernable effect on cell proliferation ( Figure 3C ) . To confirm that the effect was due to ppGalNAc-T3 , the same experiment was carried out using MCF7 cells , which is a breast cancer cell line that expresses relatively low levels of ppGalNAc-T3 ( Figure 3—figure supplement 1C ) . Critically , invasion by MCF7 cells was significantly increased by transfection with ppGalNAc-T3 and this increase was strongly blocked by T3Inh-1 ( Figure 3D ) . Although O-glycosylation has been connected to cancer migration and invasion there is little evidence that it could be targetable for clinical purposes . Our results provide a ‘proof of concept’ and a strong starting point for developing the necessary tools . 10 . 7554/eLife . 24051 . 018Figure 3 . T3Inh-1 inhibits cell invasion . ( A ) Cell migration through uncoated filters was determined for the MDA-MB231 breast cancer cell line grown in the absence or presence of 5 µM T3Inh-1 . The raw image of the filter shows both cells and the filter holes whereas a size-cut off was used in the thresholded image to specifically visualize the cells . Results were quantified by counting cells that migrated to the underside of the filter and each experiment was normalized using the average determined for controls at 24 hr ( n = 3 ± SEM ) . ( B ) Identical analysis except that the filters were pre-coated with Matrigel so that the assay measures invasion not just migration and the 48 hr control was used for normalization ( n = 3 ± SEM ) . ( C ) MDA-MB231 proliferation was determined for cells grown in the presence or absence of 5 µM T3Inh-1 by cell counting at 24 or 48 hr . Representative images before and after thresholding ( no size cutoff ) are shown as well as quantification normalized by the value determined for untreated cells at 48 hr ( n = 3 ± SEM ) . ( D ) Mock and ppGalNAc-T3 transfected MCF7 cells were plated on Matrigel-coated filters in the absence or presence of 5 µM T3Inh-1 for 24 hr . Thresholded images show cells on underside of filters . Cell counts are shown relative to untreated controls after normalization using the total number of cells ( determined using parallel wells 24 hr post-plating ) . For all panels , asterisks denote p<0 . 05 ( two-tailed Student’s t test ) for untreated to T3Inh-1 comparison . DOI: http://dx . doi . org/10 . 7554/eLife . 24051 . 01810 . 7554/eLife . 24051 . 019Figure 3—source data 1 . Panel A: Cell counts in migration assay . DOI: http://dx . doi . org/10 . 7554/eLife . 24051 . 01910 . 7554/eLife . 24051 . 020Figure 3—source data 2 . Panel B: Cell counts in invasion assay . DOI: http://dx . doi . org/10 . 7554/eLife . 24051 . 02010 . 7554/eLife . 24051 . 021Figure 3—source data 3 . Panel C: Cell counts in proliferation assay . DOI: http://dx . doi . org/10 . 7554/eLife . 24051 . 02110 . 7554/eLife . 24051 . 022Figure 3—source data 4 . Panel D: Cell counts in MCF7 invasion assay . DOI: http://dx . doi . org/10 . 7554/eLife . 24051 . 02210 . 7554/eLife . 24051 . 023Figure 3—figure supplement 1 . Breast cancer survival as a function of ppGalNAc-T3 expression and ppGalNAc-T3 expression in cultured breast cancer cell lines . ( A–B ) Kaplan-Meier curves compare overall survival ( A ) and metastasis-free survival ( B ) in patients with breast cancer between groups with high or low expression of ppGalNAc-T3 . ( C ) Immunoblots of cell lysates from the indicated cell types [HEK , HEK∆T3 ( edited to lack ppGalNAc-T3 expression ) , MDA-MB231 , MCF7 , and MCF7-T3 ( transfected to overexpress ppGalNAc-T3 ) ] using anti-ppGalNAc-T3 and anti-tubulin antibodies . DOI: http://dx . doi . org/10 . 7554/eLife . 24051 . 023 As the relevant target ( s ) of ppGalNAc-T3 that drive metastatic-like cell behavior remain to be identified , we next tested whether T3Inh-1 could inhibit glycan masking of FGF23 , a known ppGalNAc-T3 target . If so , we expected reduced secretion of intact FGF23 . HEK cells co-expressing transfected FGF23 and ppGalNAc-T3 were treated with increasing concentrations of T3Inh-1 and secreted FGF23 was assayed by immunoblot . There was a clear dose-dependent loss of intact FGF23 ( Figure 4A ) and an increase in the ratio of cleaved/intact yielding a half-max of 14 µM for this effect ( Figure 4B ) . For an unknown reason , perhaps related to its instability in media ( Kato et al . , 2006 ) , the cleaved fragment did not show a corresponding increase . Rather , its recovery varied with the average over three experiments yielding a relatively small increase ( Figure 4—figure supplement 1A–B ) . As expected , intact and cleaved FGF23 showed no change in cell lysates even for cells with lysosomal degradation inhibited by chloroquine ( Figure 4—figure supplement 1C–E ) arguing that T3Inh-1 affected cleavage just prior to secretion and did not affect FGF23 expression or cause intracellular routing to lysosomes . Importantly , we also tested the effect of T3Inh-1 on cleavage of ANGPTL3 , which is controlled by ppGalNAc-T2-mediated glycan masking ( Schjoldager et al . , 2012 , 2010 ) . Secreted intact ANGPTL3 remained high at all concentrations , confirming the selectivity of T3Inh-1 towards ppGalNAc-T3 ( Figure 4A–B ) . Reasoning that we might be able to see a similar effect on secreted FGF23 in an animal model , mice were injected intraperitoneally with T3Inh-1 and serum levels of cleaved FGF23 were determined . Three groups ( 0 , 25 and 50 mg/kg T3Inh-1 ) of mice received either one or two injections separated by 24 hr , followed by blood collection after another 24 hr . There were no apparent ill effects on animal health . An ELISA assay with antibodies against the N- and C-terminal portions of FGF23 was used to determine the ratio of cleaved/intact FGF23 in the blood ( Sun et al . , 2015; Bai et al . , 2016 ) . Remarkably , T3Inh-1 caused a robust and statistically significant increase in this ratio at the tested 25 and 50 mg/kg concentrations ( Figure 4C ) . These findings support the further development of T3Inh-1 toward mitigating the effects of elevated FGF23 signaling in chronic kidney disease patients . 10 . 7554/eLife . 24051 . 024Figure 4 . T3Inh-1 increases cleavage of FGF23 . ( A ) Immunoblot of media collected from cells after a 6 hr period in the presence of the indicated concentrations of T3Inh-1 . HEK cells were transfected with FLAG-FGF23 and ppGalNAc-T3 or Myc-ANGPTL3 and anti-FLAG and anti-Myc antibodies were used to assay intact ( * ) and cleaved ( < ) FGF23 and ANGPTL3 , respectively . The identity and origin of the unmarked band ( at approximately 25kD ) is unknown and its presence was variable . ( B ) Quantified results showing the percent ratio change of cleaved/intact FGF23 or ANGPTL3 normalized to the amount present in untreated controls ( n ≥ 3 ± SEM ) . ( C ) Serum ELISA assay results showing ratio of cleaved/intact FGF23 in mouse sera collected 24 hr after either 1 or 2 ( consecutive day ) intraperitoneal injections of the indicated amount of T3Inh-1 ( averages of 4 animals ±SEM ) . P-values are from two-tailed Student’s t test . DOI: http://dx . doi . org/10 . 7554/eLife . 24051 . 02410 . 7554/eLife . 24051 . 025Figure 4—source data 1 . Panel B: Cleaved/intact FGF23 and ANGPTL3 secreted at differing T3Inh-1 concentrations . DOI: http://dx . doi . org/10 . 7554/eLife . 24051 . 02510 . 7554/eLife . 24051 . 026Figure 4—source data 2 . Panel C: Cleaved/intact FGF23 in mouse serum after one two T3Inh-1 injections . DOI: http://dx . doi . org/10 . 7554/eLife . 24051 . 02610 . 7554/eLife . 24051 . 027Figure 4—figure supplement 1 . Secreted and cellular FGF23 after T3Inh-1 treatment . ( A ) Shown is an additional representative blot ( see Figure 4A ) of recovery of intact ( * ) and cleaved ( > ) FGF23 in the media collected from cells after a 6 hr period in the presence of the indicated concentrations of T3Inh-1 . Unmarked bands here and in panels C and D were considered background because they were absent for untransfected cells . ( B ) Quantified results for FGF23 in the media showing the relative amounts of each band ( intact and cleaved ) as a percent of the total ( intact+cleaved for the untreated controls ) . Values are averages ( n = 3 ± SEM ) . ( C–D ) These panels are identical to panels A and B except that cell extracts were analyzed rather than cell media . ( E ) Recovery of intact ( * ) and cleaved ( > ) FGF23 in HEK cell extracts from cells treated with the indicated concentrations of T3Inh-1 for 6 hr in the absence or presence of 100 μM chloroquine . DOI: http://dx . doi . org/10 . 7554/eLife . 24051 . 02710 . 7554/eLife . 24051 . 028Figure 4—figure supplement 1—source data 1 . Cleaved versus intact FGF23 secreted at differing T3Inh-1 concentrations . DOI: http://dx . doi . org/10 . 7554/eLife . 24051 . 028 This study identifies an isozyme-selective inhibitor targeting ppGalNAc-T3 . The compound binds directly conferring a mixed-mode of inhibition and is equally active in vitro and in cells . Its discovery paves the way for structural studies that will contribute to our understanding of the enzyme reaction mechanism and guide rational design of modified versions of T3Inh-1 to improve its binding affinity and efficacy . Forthcoming tests of disease models , possibly employing higher affinity versions , may strengthen the case for therapeutic uses of T3Inh-1 . It is difficult to predict possible side effects because a full list of ppGalNAc-T3 substrates is not yet available . However , the known effects of its knockout in the mouse model are all attributed to FGF23 processing ( Ichikawa et al . , 2014 ) . Thus , use of T3Inh-1 to reduce intact FGF23 ( and increase the inhibitory cleaved fragment ) to treat chronic kidney disease may have limited side effects . Clearly , the issue of multiple substrates has not been a major concern in successful therapies targeting protein kinases . To conclude , we anticipate rational design aided by T3Inh-1 , as well as further screening using isozyme-specific sensors , to result in a panel of both isozyme- and pan-specific modulators targeting ppGalNAc-transferases . As individual ppGalNAc-transferase isozymes are associated with unique diseases ( Schjoldager and Clausen , 2012; van der Post et al . , 2013; Moremen et al . , 2012 ) , the result would be a new class of therapeutics capable of treating an array of differing diseases .
HEK cell lines were previously described ( Song et al . , 2014 ) . HeLa ( Cat#ATCC-CCL-2 , CVCL_0030 ) , MDA-MB231 ( Cat#ATCC-HTB-26 , CVCL_0062 ) , and MCF7 ( Cat#ATCC-HTB-22 , CVCL_0031 ) were purchased from ATCC ( Manassas , VA ) . All cell lines were verified mycoplasma free every two months using Hoechst staining . Antibodies used were monoclonal antibodies 4C4 against ppGalNAc-T2 and UH5 against ppGalNAc-T3 ( 8 , 31 , 36 ) , monoclonal 9e10 against the myc epitope ( Evan et al . , 1985; Jesch et al . , 2001 ) , a polyclonal against the FLAG epitope ( Bethyl Labs , Cat#A190-102B , AB_1944186 ) , a polyclonal against GPP130 ( Puri et al . , 2002 ) , a purchased anti-ppGalNAc-T3 antibody ( ThermoFisher , Cat#PA5-25217 , AB_2542717 ) , an anti-α-tubulin antibody ( Biolegend , Clone TU-01 , Cat#625902 , AB_2210041 ) , Alexa 488 anti-mouse ( Cat#A28175 , AB_2536161 ) and Alexa 555 anti-rabbit ( Cat#A27039 , AB_2536100 ) from Thermo Fisher ( Pittsburgh , PA ) , and horse radish peroxidase-conjugated goat anti-mouse ( Cat#170–6516 , AB_11125547 ) and goat anti-rabbit ( Cat#170–6515 , AB_11125142 ) antibodies from Sigma-Aldrich ( St . Louis , MO ) . HEK cells stably expressing the T2 sensor ( containing ANGPTL3 linker sequence with T225G modification ( Song et al . , 2014 ) ) or the T3 sensor ( containing FGF23-based linker [Song et al . , 2014] ) were cultured in MEM ( Corning , NY , Cat#10–010-CV ) with 10% fetal bovine serum ( FBS , Atlanta Biologicals , Flowery Branch , GA , Cat#S111150 ) and 100 IU/ml penicillin-streptomycin ( Sigma-Aldrich , Cat#P4333 ) at 37°C , 5% CO2 . Positive controls were cell lines expressing matched sensors with the glycosylation site mutated ( ∆glycan ) , specifically T225G/T226G and T178G for the T2 and T3 sensors , respectively ( Song et al . , 2014 ) . Cells ( 50 , 000/well ) were seeded in flat bottom 96-well-plates ( Corning , NY , Cat#3997 ) and grown for 24 hr . Compounds ( the diversity set from ChemBridge Corporation [Chicago , IL] and the approved oncology drugs set V and the diversity set II from the National Cancer Institute Developmental Therapeutics Program ) were then added to achieve a 10 µM final concentration . After 6 hr , the medium was aspirated and the cells were released by adding 100 µl 5 mM EDTA/PBS containing 110 nM MG dye ( Sharp Edge Labs , Pittsburgh , PA ) for 5 min at 37°C . The plates were then transferred to an Accuri C6 flow cytometer ( BD Biosciences ) where GFP and MG fluorescence was measured using 488 nm and 640 nm for 10 , 000 cells per well . Data analysis used FlowJo software ( www . flowjo . com , SCR_008520 ) . For each well the geometric means of the MG and GFP signals were used to compute the MG/GFP ratio . Each compound was analyzed in two wells and the average of the two resulting ratios ( R ) was recorded . Each daily run included at least 16 wells of vehicle-only controls ( sensor-expressing cells treated with a matching DMSO concentration [Fisher Scientific , Cat#BP231-100] ) and a similar analysis was used to calculate their average MG/GFP ratio ( RNeg ) and its standard deviation ( SDNeg ) . The Q value of each compound ( as well as the untreated ∆glycan positive controls ) was calculated by using the following equation: Q= ( R-RNeg ) /SDNeg . Background fell within the range −2 . 5 ≤ Q ≤ 3 and the average Q∆glycan was 135 and 38 for the T2 and T3 sensors , respectively . Glycosylation assays using recombinant ppGalNAc-T2 and ppGalNAc-T3 were carried out using the UDP-Glo Glycosyltransferase assay kit ( Promega , Madison , WI , Cat#V6962 ) , according to the manufacturer’s recommendation . The reaction ( 25 µl ) included 2 . 5 ng/ µl purified enzyme , 25 µM UDP-GalNAc ( Sigma-Aldrich , Cat#U5252 ) , 12 . 5 µM EA2 peptide ( AnaSpec , Mucin 10 , AA153-165 , PTTDSTTPAPTTK , Cat#AS-63841 ) , 25 mM Tris-HCl ( pH7 . 5 ) ( Fisher Scientific , Cat#77-86-1 ) , 5 mM MnCl2 ( Fisher Scientific , Cat#M87-500 ) , 2 . 5 mM CaCl2 ( Fisher Scientific , Cat#C70-500 ) and 50 µM compound . The negative control was vehicle only ( same reaction mixture with a matched percentage of DMSO instead of compound ) , whereas background was from the reaction carried out without enzyme and DMSO instead of compound . All reactions were incubated at 37°C in a water bath for 30 min and then cooled to room temperature . Aliquots ( 5 µl ) were then added to a 384-well plate ( Thermo Scientific , Waltham , MA , Cat#164610 ) to which 5 µl of UDP Detection Reagent was also added . Duplicate measures were made for all reactions . After 1 hr at room temperature the luminescent signals were determined using a Tecan Infinite M1000 ( Tecan Group Ltd . , Männedorf , Switzerland ) with integration time set to 1000 ms . The background-subtracted average for each compound was expressed as a percentage of the negative control ( taken as 100% ) . Compounds with ≥50% effect were considered direct modulators . For the sensor assay , about 200 , 000 cells expressing either sensor were seeded into Greiner Bio-One 24-well plates ( Sigma-Aldrich , Cat#662160 ) . After 24 hr , the cells were incubated for another 6 hr in the presence of 0–50 µM compound . To release the cells the medium was replaced with 200 µl 5 mM EDTA/PBS containing 110 nM MG dye . After 5 min at 37°C , fluorescence measurements ( 20 , 000 cells/well ) were carried out as described above . For the biochemical assay , the assay conditions were identical except for variations in the compound or substrate concentrations as indicated in the figure legends . Data analysis was using Prism ( Graph Pad Prism Inc . , SCR_002798 ) . The purified lumenal domain of ppGalNAc-T3 ( 30 ng/ µl ) was incubated with 0–500 µM compound at room temperature for 10 min and 200 µl aliquots were transferred to a Greiner Bio-One 96-well glass-bottom plate ( Sigma-Aldrich , Cat#655892 ) and the fluorescent emission was scanned ( 300–450 nm ) using a Tecan Infinite M1000 with excitation at 290 nm , gain set to 150 , number of flashes at 50 and flash frequency at 400 Hz . The value at the peak of emission at 324 nm was used for the binding curve analysis by Prism . For determination of sensor activation , spinning-disk confocal microscopy was used exactly as described ( Song et al . , 2014 ) . To assess possible effects of compounds on Golgi markers , including ppGalNAc-T2 and ppGalNAc-T3 , immunofluorescence was carried on HeLa cells treated with 10 µM compound for 6 hr . Briefly , the cells were grown on 12 mm diameter coverslips ( Fisher Scientific , Cat#12-545-81 ) for 48 hr , treated with the compounds , washed with PBS and fixed with 3% paraformaldehyde ( Sigma-Aldrich , Cat#P6148 ) for 15 min . Blocking , Triton X-100 permeabilization , antibody incubations and image capture by spinning-disk confocal were as described ( Mukhopadhyay et al . , 2010 ) . Monoclonal antibodies against ppGalNAc-T2 and T3 were used undiluted and the polyclonal against GPP130 was used at 1:2000 . All corresponding images were acquired and adjusted using identical parameters . HeLa cells ( treated with T3Inh-1 for 24 hr at the indicated concentrations ) were washed twice in PBS containing 0 . 5% FBS then stained for 30 min with fluorescent lectins ConA ( Cat#FL-1001 ) , WGA ( Cat#FL-1021 ) , SNA ( Cat#FL-1301 ) or VVA ( Cat#FL-1231 ) ( Vector Laboratories , Burlingame , CA ) at 1:100 dilutions ( except WGA used at 1:1000 ) in the wash buffer . After staining , the cells were washed twice , lysed with 0 . 2% Triton X-100 ( Fisher Scientific , Cat#BP151-100 ) for 15 min at 4°C , centrifuged at 14 , 000 x g for 15 min at 4°C , and the supernatants were read in a Greiner Bio-One 96-well glass-bottom plate using a Tecan Infinite M1000 with excitation at 495 nm and emission from 510–550 nm . Three independent trials were carried out and the peak value ( 520 nm ) was used for quantification . Equal numbers of HEK or MDA-MB231 cells were plated in Greiner Bio-One 24-well dishes in growth medium containing 0–50 µM T3Inh-1 . After 24 , 48 or 72 hr , the cells were released using trypsin and counted twice using a hemocytometer for three independent trials . Breast cancer MDA-MB231 cells were grown in DMEM medium ( Corning , NY , Cat#10–013-CV ) with 10% FBS and 100 IU/ml penicillin-streptomycin at 37°C and 5% CO2 and then plated at a density of 1 . 32 × 104 in 0 . 3 ml of DMEM medium without FBS into the upper chamber of a Bioboat insert fitted with a 8 . 0 µm PET membrane ( Corning , NY , Cat#354578 ) . For migration assays , the filter was uncoated . For invasion assays , it was pre-coated with 100 µl Matrigel ( BD Biosciences , Cat#356234 ) at concentration of 272 µg /ml for 1–2 hr at room temperature . Medium containing 10% FBS ( 0 . 6 ml ) was placed into the lower chamber as a chemo attractant . The compound ( final concentration of 5 µM ) or a matching amount of DMSO was added to both chambers . After 24 hr or 48 hr the cells were fixed with −20°C methanol for 15 min and then stained with Trypan blue for 5 min . Cells on the upper surface were removed using cotton swabs . Cells present on the underside of the membrane were photographed using an EVOS FL Cell Imaging System ( Invitrogen , CA ) and the images were thresholded for presentation and counting using Image J ( National Institutes of Health , Bethesda , MD , SCR_003070 ) . The assays involving MCF7 cells were identical except that they were performed 24 hr post transfection . The FLAG-tagged FGF23 , Myc-tagged ANGPTL3 with the T225G modification ( Song et al . , 2014 ) , and untagged ppGalNAc-T3 ( from [Kato et al . , 2006] and cloned into PCDNA 3 . 0 using BamH1 sites ) were transfected into HEK cells using the JetPEI transfection reagent ( VWR International , Radnor , PA , Cat#101–40N ) according to the manufacturer’s instructions . After 24 hr , the medium was replaced with serum-free MEM containing the compound for 6 hr . The medium and cells were then collected and , after trichloroacetic acid precipitation of the medium , analyzed by immunoblot using anti-FLAG antibody at 1:1000 or anti-Myc antibody at 1:2000 and then the peroxidase-coupled secondary antibodies . Emission was captured and quantified using a ChemiDoc Touch Imaging System with Image Lab Software ( BioRad , SCR_014210 ) . For ppGalNAc-T3 determinations in different cell lines , cells were collected and lysed with 100 µl buffer ( 10 mM Tris-HCl ( pH8 . 0 ) , 1 mM EDTA ( ACROS ORGANICS , Cat#446085000 ) , 1% Triton X-100 , 0 . 1% sodium deoxycholate ( Fisher Scientific , Cat#BP349-100 ) , 0 . 1% SDS ( Fisher Scientific , Cat#BP166-500 ) , 140 mM NaCl ( Fisher Scientific , Cat#S271-3 ) , 1 mM PMSF ( Sigma-Aldrich , Cat#PMSF-RO ) . Then 15 µl of each lysate was analyzed by immunoblotting using the purchased anti-ppGalNAc-T3 and anti-α-tubulin antibodies . Wild-type C57BL/6 six to eight week old mice were purchased from Charles River Laboratories international Inc . ( Wilmington , MA ) . Protocols , handling , and care of the mice conformed to protocols approved by the Institutional Animal Care and Use Committee of Carnegie Mellon University ( CMU IACUC protocol AS16-005 ) . The compound was dissolved in DMSO at 25 and 50 mg/ml then further diluted with PEG400 ( Hampton Research , CA , USA , HR2-603 ) to create 5 and 10 mg/ml stocks for injection . Control ( vehicle only: 20% DMSO , 80% PEG400 ) and experimental ( 25 or 50 mg/kg compound ) animals received either single or double ( separated by 24 hr ) intraperitoneal injections and , 24 hr after the last injection , a cardiac blood draw was carried out . The cleaved/intact FGF23 ratio was determined using ELISA kits from Immunotopics ( Carlsbad , VA , USA , Cat#60–6800 , Cat#60–6300 ) with cleaved equaling total minus intact . | Complex cascades of interactions between different molecules regulate every process in the body . Enzymes are critical for this , because they act as ‘catalysts’ to speed up chemical reactions in a cell . Each type of enzyme has a specific role . The enzyme ppGalNAc-T3 , for example , attaches sugar molecules onto certain proteins via a process called glycosylation . This modification fine-tunes the activity of the proteins . The enzyme ppGalNAc-T3 is implicated in at least two medically important pathways . It increases the amount of the hormone FGF23 , which regulates phosphate levels in the bloodstream . Hormones are messenger molecules that regulate most processes that are crucial for life , and too much or too little of a hormone can lead to diseases . High levels of FGF23 , for example , can cause serious and often fatal problems in patients with chronic kidney disease . The enzyme ppGalNAc-T3 is also known to encourage and stimulate some cancer cells to spread from their original location to other parts of the body – a process known as metastasis – which ultimately leads to the vast majority of cancer deaths every year . Thus , a drug or molecule that blocks ppGalNAc-T3 could be used to lower FGF23 levels in patients with kidney disease and potentially prevent cancer cells from spreading . However , until now , it was unknown if such a molecule existed . To identify a compound that specifically regulates ppGalNAc-T3 , Song and Linstedt engineered human cells grown in the laboratory to become fluorescent when ppGalNAc-T3 was blocked . Out of the over 20 , 000 compounds screened , one compound named T3Inh-1 selectively blocked ppGalNAc-T3 . Further experiments showed that the T3Inh-1 compound reduced FGF23 hormone levels in both tissue cells grown in the laboratory and mice , without causing any toxic side effects . It also prevented breast cancer cells grown in the laboratory from spreading . The results demonstrated that T3Inh-1 is the first drug-like inhibitor that can target this kind of enzyme . An important next step will be to test the compound in animal models for chronic kidney disease and cancer metastasis . | [
"Abstract",
"Introduction",
"Results",
"and",
"discussion",
"Materials",
"and",
"methods"
] | [
"short",
"report",
"cell",
"biology",
"cancer",
"biology"
] | 2017 | Inhibitor of ppGalNAc-T3-mediated O-glycosylation blocks cancer cell invasiveness and lowers FGF23 levels |
G-quadruplexes ( G4 ) are extremely stable secondary structures forming stacks of guanine tetrads . DNA G4 structures have been extensively studied , however , less is known about G4 motifs in mRNAs , especially in their coding sequences . Herein , we show that Aven stimulates the mRNA translation of the mixed lineage leukemia ( MLL ) proto-oncogene in an arginine methylation-dependent manner . The Aven RGG/RG motif bound G4 structures within the coding regions of the MLL1 and MLL4 mRNAs increasing their polysomal association and translation , resulting in the induction of transcription of leukemic genes . The DHX36 RNA helicase associated with the Aven complex and was required for optimal translation of G4 mRNAs . Depletion of Aven led to a decrease in synthesis of MLL1 and MLL4 proteins resulting in reduced proliferation of leukemic cells . These findings identify an Aven-centered complex that stimulates the translation of G4 harboring mRNAs , thereby promoting survival of leukemic cells .
RNA-binding proteins ( RBPs ) coordinate many steps of RNA metabolism ranging from splicing , RNA processing , RNA transport , mRNA translation , and RNA degradation ( Glisovic et al . , 2008 ) . RBPs associate with specific RNA motifs and/or secondary structures within coding , untranslated regions , and non-coding RNAs in functional units called ribonucleoprotein ( RNP ) complexes ( Mitchell and Parker , 2014 ) . Defects in RBPs have been associated with many complex diseases ranging from neurological disorders to cancer ( Lukong et al . , 2008; Cooper et al . , 2009; Castello et al . , 2013; Ramaswami et al . , 2013 ) . RBPs are predominantly defined by the presence of RNA-binding domains within their sequences ( Chen and Varani , 2013 ) . Recently , several ‘interactome capture’ strategies have been performed to identify RBPs genome-wide . In addition to identifying the known RBPs , these approaches have identified numerous mammalian proteins that do not possess a canonical RNA-binding domain ( Baltz et al . , 2012; Castello et al . , 2012; Kwon et al . , 2013 ) . Interestingly , RBPs that harbor repeated sequences including YGG and RGG motifs were identified ( Castello et al . , 2012 ) . The RGG/RG motif is enriched in proteins associated with RNA and is a known RNA-binding interface ( reviewed in Thandapani et al . , 2013 ) . The RGG/RG motif , also called RGG box , was shown to bind RNA ( Kiledjian and Dreyfuss , 1992 ) . Subsequently , the RGG/RG motifs of Nucleolin , FMRP , FUS , and EWS were also shown to bind guanine-rich sequences that are potential G-quadruplexes ( Darnell et al . , 2001; Takahama et al . , 2011 , 2013; Haeusler et al . , 2014 ) . RGG/RG motifs also mediate protein–protein interactions . Notably , the RGG/RG motif of yeast Scd6 mediates interactions with eIF-4G , which leads to stress granule formation and inhibition of cap-dependent translation ( Rajyaguru et al . , 2012 ) . Despite these recent advances , little is known about the role of RGG/RG motifs that bind both RNA and proteins . G-quadruplexes ( G4 ) are planar structures of stacks of guanine tetrads stabilized by monovalent potassium or sodium ions . G-quadruplexes have been shown to regulate DNA replication , DNA repair , gene expression , and telomeres ( Bates et al . , 2007 ) . Less is known about G4 structures found in RNA . There are >1500 potential G4s ( PG4s ) in 5′-UTR of mRNAs alone ( Beaudoin and Perreault , 2010 ) , but not all PG4s form stable G-quadruplexes , which are influenced by the numbers of G-quartets , the possibility of bulge formation , the length of the loops , and the presence of alternative Watson–Crick base pair-based stable structure ( Burge et al . , 2006; Mukundan and Phan , 2013; Jodoin et al . , 2014 ) . Some PG4s in the 5′-UTR of mRNAs form bona fide G-quadruplexes and inhibit cap-dependent translation ( Kumari et al . , 2007; Beaudoin and Perreault , 2010; Bugaut and Balasubramanian , 2012 ) . Recently , inhibitors of DEAD box RNA helicase eIF4A or eIF4A1 depletion have been shown to selectively inhibit translation of mRNAs with G-quadruplexes in their 5′ UTR ( Wolfe et al . , 2014; Modelska et al . , 2015 ) . However , presence of G-quadruplexes in 5′ UTRs does not appear to be sufficient to render translation of mRNAs sensitive to changes in eIF4A activity ( Rubio et al . , 2014 ) . In addition to the incomplete understanding of the role of 5′ UTR G-quadruplexes in translation control , little is known about how G4 structures in open-reading frames ( ORFs ) affect translation . Arginine residues within RGG/RG motifs are preferred substrates for methylation by protein arginine methyltransferases ( PRMTs ) ( Thandapani et al . , 2013 ) . Arginine methylation is known to regulate many cellular processes including signal transduction , transcription , pre-mRNA splicing , and DNA repair ( Bedford and Richard , 2005; Bedford and Clarke , 2009; Xu et al . , 2013 ) . PRMT1 generates >85% of asymmetric dimethylarginines found in cells with preference for RGG/RG motif containing proteins ( Bedford and Clarke , 2009 ) . PRMT1 is known for its nuclear roles in regulating gene expression and DNA damage ( Strahl et al . , 2001; Wang et al . , 2001; An et al . , 2004; Boisvert et al . , 2005 ) , however , less is known about its cytoplasmic roles . PRMT1-deficient mice die at E6 . 5 and the absolute removal of PRMT1 in mouse embryo fibroblasts ( MEFs ) leads to cell death ( Pawlak et al . , 2000; Yu et al . , 2009 ) . To identify other biological processes regulated by arginine methylation , we performed a bioinformatics approach to identify proteins harboring RGG/RG motifs and one such protein we identified was Aven ( Thandapani et al . , 2013 ) . Aven is a predominantly cytoplasmic protein required for cell survival and it has been shown to function as an apoptotic inhibitor by interaction with and stabilizing the pro-survival protein Bcl-xL , as well as inhibiting the function of Apaf-1 ( Chau et al . , 2000 ) . It was proposed that the proteolytic cleavage of Aven by Cathepsin D is required for its anti-apoptotic activity ( Melzer et al . , 2012 ) . Furthermore , Aven is required for ataxia telangiectasia-mutated ( ATM ) activation in Xenopus oocytes and HeLa cells ( Guo et al . , 2008 ) and ataxia telangiectasia-related activation following DNA damage in osteosarcoma cells ( Baranski et al . , 2015 ) . High Aven expression correlates with poor survival in metastatic patients with osteosarcomas ( Baranski et al . , 2015 ) . The elevated Aven expression is also frequently observed in acute myeloid leukemia and acute lymphoblastic leukemia ( T-ALL ) and is associated with poor prognosis ( Paydas et al . , 2003; Choi et al . , 2006 ) . A transgenic mouse model with T cell-specific overexpression of Aven showed that its expression enhanced T-cell lymphomagenesis in the absence of p53 ( Eismann et al . , 2013 ) . The mechanism by which Aven promotes hematological malignancies is yet to be understood . Herein , we report that the methylation of the RGG/RG motif of Aven functions in the translational control of mRNAs harboring G4 structures in their ORFs . The association of Aven with polysomes was dependent on the arginine methylation of its RGG/RG motif and on the methyl-dependent interactions with the Tudor domains of SMN and TDRD3 , previously shown to be associated with polysomes ( Goulet et al . , 2008; Sanchez et al . , 2013 ) . We identify Aven to be an RBP , as its RGG/RG motif bound G4 motifs in the ORFs of mRNAs encoding the mixed lineage leukemia ( MLL ) family proteins MLL1 and MLL4 . The RGG/RG motif of Aven also associated with the G4 RNA helicase , DHX36 , and this helicase was required for optimal translation of Aven-regulated mRNAs . Furthermore , Aven-deficient T-ALL cell lines had reduced MLL1 and MLL4 protein levels , but not mRNA levels , which were paralleled by proliferation defects . These findings define a hitherto unknown mechanism of action for arginine methylation in regulating translation of a subset of mRNAs including those encoding pivotal leukemogenic transcriptional regulators MLL1 and MLL4 .
Aven harbors an N-terminal RGG/RG motif ( Thandapani et al . , 2013 ) , a nuclear export sequence ( Esmaili et al . , 2010 ) , and a predicted BH3 domain ( Hawley et al . , 2012 ) ( Figure 1A ) . To define the function of the Aven RGG/RG motif , we initially investigated whether the motif was methylated by protein arginine methyltransferase 1 ( PRMT1 ) . An in vitro methylation assay was performed using a glutathione-S-transferase ( GST ) -Aven RGG/RG fusion protein incubated with recombinant GST-PRMT1 in the presence of ( methyl-3H ) -S-adenosyl-L-methionine . The proteins were separated by SDS-PAGE , stained with Coomassie Blue to visualize loading , and the methylated proteins were observed by fluorography . The GST-AvenRGG/RG fusion protein migrated as a doublet and was methylated by PRMT1 , while GST alone was not methylated ( Figure 1B , lanes 5 , 6 ) . 10 . 7554/eLife . 06234 . 003Figure 1 . Aven is a substrate of PRMT1 . ( A ) Schematic diagram of Aven with its RGG/RG motif , putative BH3 domain , and nuclear export sequence ( NES ) . ( B ) In vitro methylation assay with GST-PRMT1 and GST-AvenRGG/RG with ( 3H ) -S-adenosyl-L-methionine as the methyl donor ( n > 4 ) . Proteins were resolved by SDS-PAGE , stained with Coomassie Blue ( left ) , and analyzed by fluorography ( right ) . The migration of the molecular mass markers is shown on the left in kDa and the migration of the GST-PRMT1 , glutathione-S-transferase ( GST ) , GST-AvenRGG/RG proteins is indicated with arrows . The asterisk ( * ) denotes degraded proteins from the GST-PRMT1 preparation . ( C ) HEK293T cells transfected with empty vector pcDNA3 . 1 ( lanes 1 , 3 , 5 , 7 ) or Myc-Aven-pcDNA3 . 1 ( lanes 2 , 4 , 6 , 8 ) were lysed and immunoprecipitated ( IP ) with anti-Myc antibodies ( lanes 3 , 4 , and 7 , 8 ) . The total cell lysates ( TCL ) and the bound proteins were resolved by SDS-PAGE and immunoblotted with anti-PRMT1 ( lanes 1–4 ) or −Myc antibodies ( 5–8 ) . TCL denotes input TCL and IgG represents the heavy chain of immunoglobulin G . ( D ) HEK293T cells were cotransfected with siGFP ( -siPRMT1 ) or siPRMT1 ( +siPRMT1 ) and pcDNA3 . 1 or Myc-Aven plasmids . After 48 hr , the cells were lysed and IP with anti-Myc antibodies , the proteins were resolved by SDS-PAGE and immunoblotted with anti-ASYM25b ( lanes 1–4 ) , and anti-Myc antibodies ( lanes 5–8 ) . TCL were immunoblotted with anti-Myc , anti-PRMT1 , and anti-rpS6 antibodies . The latter was to control for equal loading . ( E ) PRMT1FL/−;CreERT mouse embryo fibroblasts ( MEFs ) treated with 4-hydroxytamoxifen ( OHT ) for 6 days or left untreated were transfected with Myc-Aven followed by anti-Myc antibody immunoprecipitations and the methylation monitored by immunoblotting with ASYM25b ( lanes 1–4 ) or anti-Myc antibodies ( lanes 5–8 ) . TCL were immunoblotted with anti-PRMT1 , anti-Aven , and anti-tubulin antibodies as indicated . ( F ) PRMT1FL/−;CreERT MEFs treated with OHT for 6 days or left untreated were lysed and IP with anti-Aven ( ab77014 ) or IgG antibodies . Immunoprecipitates were blotted with ASYM25b . TCL were immunoblotted with anti-PRMT1 , anti-Aven , and anti-tubulin antibodies as indicated . DOI: http://dx . doi . org/10 . 7554/eLife . 06234 . 00310 . 7554/eLife . 06234 . 004Figure 1—figure supplement 1 . Aven harbors dimethylarginines within its RGG/RG motif . ( A ) Immunoprecipitates of Myc-Aven transfected HEK293T cells resolved by SDS-PAGE . The molecular mass markers are indicated in kDa . Mass spectrometry profile of Aven . LC-MS/MS analysis of the excised Myc-Aven band . The sequence of Aven from residues 63 to 73 is shown . LC-MS/MS analysis revealed the presence of a modified peptide RGGRGGGGAPR containing dimethylated R63 and R66 . Analysis of the Aven peptide from residues 33 to 50 is shown as well as the dimethylation of R37 . Similar analysis identified R8 , R50 , and R11 to be dimethylated ( not shown ) . ( B ) The alignment of Aven N terminus from various eukaryotic species . The mono- ( * ) and di-methylated ( ** ) arginine residues identified by LC-MS/MS analysis and conserved across various eukaryotic species are indicated with blue boxes . DOI: http://dx . doi . org/10 . 7554/eLife . 06234 . 004 We next examined whether Aven and PRMT1 associated in vivo . HEK293T cells were transfected with Myc-epitope tagged Aven or empty vector ( pcDNA3 . 1 ) and cell extracts were immunoprecipitated ( IP ) with anti-Myc antibodies . The IP proteins were separated by SDS-PAGE and immunoblotted with anti-PRMT1 antibodies ( Figure 1C ) . PRMT1 was present in anti-Myc immunoprecipitates of the Myc-Aven-transfected cells , but not in the empty vector-transfected cells ( Figure 1C , lanes 3 , 4 ) . Immunoblotting with anti-Myc confirmed the presence of Myc-Aven ( Figure 1C , lanes 7 , 8 ) . To examine whether Aven is a substrate of PRMT1 in vivo , we depleted HEK293T cells of PRMT1 using siRNA . The cells were also transfected with empty plasmid ( pcDNA3 . 1 ) or an expression vector encoding 5 tags of the Myc epitope linked to Aven ( Myc-Aven ) . Cellular lysates were IP with immunoglobulin ( IgG ) control or anti-Myc antibodies , resolved by SDS-PAGE , and immunoblotted with ASYM25b , an asymmetric dimethylarginine-specific antibody . We observed that Myc-Aven was arginine methylated in PRMT1-proficient , but not in PRMT1-deficient cells ( Figure 1D , compare lanes 3 and 4 ) . An anti-Myc immunoblot confirmed the immunoprecipitations ( Figure 1D , lanes 7 and 8 ) . Immunoblots of total cell lysates confirmed the myc-Aven expression and the PRMT1 knockdown and rpS6 was used as a loading control ( Figure 1D , lower panels ) . A similar experiment was performed using conditional PRMT1FL/−;CreERT MEFs ( Yu et al . , 2009 ) . Ablation of PRMT1 was achieved by treating the cells with 4-hydroxytamoxifen ( OHT ) for 6 days . An expression vector encoding Myc-Aven was transfected into control ( PRMT1FL/−;CreERT; −OHT ) and PRMT1-deficient ( PRMT1FL/−;CreERT; +OHT ) MEFs . Myc-Aven was arginine methylated in PRMT1-proficient , but not in PRMT1-deficient cells ( Figure 1E , compare lanes 1 and 2 ) . We next proceeded to show that endogenous Aven is arginine methylated . Conditional PRMT1FL/−;CreERT MEFs were used for Aven immunoprecipitations and indeed endogenous Aven was asymmetrically dimethylated by PRMT1 ( Figure 1F ) . Endogenous Aven migrates at ∼50 kDa ( Figure 1F ) , while myc-Aven was generated with multiple Myc tags to avoid overlap with heavy chain of IgG during immunoprecipitation and migrated at ∼70 kDa ( Figure 1C–E ) . To identify the arginines within the RGG/RG motif that are methylated , immunopurified Myc-Aven was subjected to mass spectrometry analysis . We identified the dimethylation of Aven R37 , R63 , and R66 ( Figure 1—figure supplement 1 ) . Moreover , we also identified the dimethylation of R8 , R11 , R50 , as well as the monomethylation of R5 , R8 , R28 , and R37 . Arginines R5 , R8 , R11 , R37 , R50 , and R63 are also conserved in murines ( Figure 1—figure supplement 1 ) . Taken together , these findings demonstrate that the Aven RGG/RG motif is methylated by PRMT1 on conserved arginines . To define the role of the Aven RGG/RG motif , we generated a mutant that lacks the motif by deleting amino acids 1 to 73 . AvenΔRGG was not recognized by ASYM25b , confirming that all the methylarginines reside in the N-terminus of Aven ( Figure 2A ) . AvenΔRGG was able to activate ATM , comparably to wild-type Aven and deletion of the RGG/RG motif did not interfere with its ability to oligomerize ( Figure 2—figure supplement 1 ) . Aven is predominantly localized in the cytoplasm with some weak nuclear staining ( Figure 2—figure supplement 1 ) , as reported previously ( Chau et al . , 2000 ) , and FLAG-AvenΔRGG had the same cellular localization pattern as wild-type Aven ( Figure 2—figure supplement 1 ) . 10 . 7554/eLife . 06234 . 005Figure 2 . Aven binds G4 RNA sequences in an arginine methylation independent manner . ( A ) U2OS cells transfected with pcDNA3 . 1 , or expression vectors encoding FLAG-Aven or FLAG-AvenΔRGG were IP with anti-FLAG antibodies and immunoblotted with ASYM25b or anti-FLAG antibodies as indicated . The molecular mass markers are shown on the left in kDa and the migration of FLAG-Aven and FLAG-AvenΔRGG is shown . The asterisks ( * ) denote unknown arginine methylated proteins . ( B ) Biotinylated sc1 G4 bound to Streptavidin were incubated with HEK293T cell lysates . The bound proteins were washed with increasing NaCl ( mM ) as indicated and visualized by SDS-PAGE followed by immunoblotting with anti-Aven antibodies . ( C ) Biotinylated sc1 G4 or ( G4m ) bound to Streptavidin was incubated with cellular lysates expressing FLAG-Aven and FLAG-AvenΔRGG and detected as in panel B . ( D ) Biotinylated sc1 G4 RNA bound to Streptavidin beads was used to pull-down Aven from PRMT1-depleted HEK293T cells . Aven binding was performed as in panel B . PRMT1 depletion was confirmed by immunoblotting . ( E ) Biotinylated methylated and unmethylated Aven TriRG peptides were pre-bound on Streptavidin plates and were incubated with fluorescein-labeled sc1 G4 RNA . The bound RNA was quantified by measuring fluorescence at 521 nm . The experiment was performed twice in triplicates . DOI: http://dx . doi . org/10 . 7554/eLife . 06234 . 00510 . 7554/eLife . 06234 . 006Figure 2—figure supplement 1 . Aven RGG/RG motif binds RNA and does not regulate ATM activation , nor Aven cellular localization . ( A ) To investigate the role of RGG/RG motif in ataxia telangiectasia-mutated ( ATM ) activation , U2OS cells were transfected with FLAG-Aven and FLAG-AvenΔRGG . Transfected cells were treated with etoposide ( 50 ng/ml ) for 30 min . Lysates collected at various time points post-treatment were separated by SDS-PAGE and immunoblotted with anti-pATM S1981 , anti-FLAG , anti-pCHK2T68 , anti-CHK2 , and anti-tubulin antibodies . The experiment was performed three times and a representative experiment is shown . ( B ) U2OS cells were co-transfected with Myc-Aven and either FLAG-Aven or FLAG-AvenΔRGG . Immunoprecipitation was performed with anti-FLAG agarose beads and the membranes were immunoblotted with anti-FLAG and anti-Myc antibodies . 10% of the lysates were shown in the bottom panel to confirm the expression of the transfected constructs . The experiment was performed twice . ( C ) U2OS cells were transfected with FLAG-Aven and FLAG-AvenΔRGG . The cells were fixed and labeled for immunofluorescence with anti-FLAG antibodies . The experiment was performed three times . ( D ) Biotinylated methylated and unmethylated Aven DiRGG peptides were pre-bound on Streptavidin plates and were incubated with fluorescein-labeled sc1 G4 RNA . The bound RNA was quantified by measuring measuring fluorescence at 521 nm . DOI: http://dx . doi . org/10 . 7554/eLife . 06234 . 006 RGG/RG motifs are enriched amongst RBPs and they possess inherent RNA-binding activity ( reviewed in Thandapani et al . , 2013 ) . A high-affinity RNA sequence that forms a G4 , termed sc1 , binds RGG/RG sequences ( Phan et al . , 2011 ) . To test whether the RGG/RG motif of Aven binds G4 sequences , we performed binding assays using sc1 . Biotinylated RNA sequences of sc1 were generated , heated , and slowly cooled in the presence of K+ to favor formation of the G4 RNA structure . The RNA was used in an affinity ‘pull-down’ assay with HEK293T cell lysates . The presence of Aven after different washes of sodium chloride was monitored by immunoblotting following separation of the bound proteins by SDS-PAGE . Aven bound the sc1 wild-type G4 RNA , but not a mutant sc1 RNA sequence that is predicted not to form the G4 structure ( Figure 2B ) . To determine whether the Aven binding to the G4 RNA structures was mediated by the RGG/RG motif , binding assays were performed in HEK293T cells expressing FLAG-Aven and the FLAG-AvenΔRGG . FLAG-Aven , but not FLAG-AvenΔRGG , bound the sc1 G4 structure indicating that the RGG/RG motif is necessary for binding ( Figure 2C ) . We subsequently investigated whether methylation of the RGG/RG motif influences binding to the sc1 G4 structure . Hypomethylated Aven was obtained by depleting HEK293T cells of PRMT1 with siRNAs . Aven bound equally well to biotinylated sc1 G4 RNA from wild-type or PRMT1-depleted cells ( Figure 2D ) . A biotinylated Aven RGG/RG peptide denoted as TriRG was synthesized with or without asymmetric dimethylarginines , and RNA binding was measured using fluorescently labeled RNA . RGG/RG motif whether harboring arginine or asymmetric dimethylarginines bound a fluoresceinated sc1 G4 RNA , induced to fold into a G4 structure prior to binding , with same relative affinities with a Kd of ∼80–90 nM ( Figure 2E ) . Another set of peptides spanning the DiRGG motif of Aven with and without asymmetric dimethylarginines also bound the fluoresceinated sc1 G4 RNA with lower affinity than the TriRG peptides ( Kd of ∼175–200 nM , Figure 2—figure supplement 1 ) , suggesting that several RGG/RG motifs in Aven are able to bind RNA . These finding show that Aven is an RBP that interacts with G4 RNA sequences via its RGG/RG motifs independent of arginine methylation . Arginine methylation is known to regulate protein–protein interactions with Tudor domain-containing proteins ( Chen et al . , 2011 ) . Although there are many Tudor domain-containing proteins , methylated RGG/RG motifs are known to interact specifically with the Tudor domains of TDRD3 , SMN , and SPF30 ( Selenko et al . , 2001; Cote and Richard , 2005 ) . Since arginine methylation did not influence the ability of Aven to bind RNA , we next investigated whether Aven interacts with TDRD3 , SMN , and SPF30 in a methyl-dependent manner . U2OS cellular lysates expressing FLAG-Aven or FLAG-AvenΔRGG were incubated with the GST-TDRD3 , GST-SMN , and GST-SPF30 bound to Sepharose beads . The presence of bound Aven was detected by SDS-PAGE followed with anti-FLAG immunoblotting . FLAG-Aven , but not FLAG-AvenΔRGG , interacted with the GST Tudor domains of TDRD3 and SMN , suggesting that TDRD3 and SMN Tudor domains interact with the Aven RGG/RG motif ( Figure 3A ) . GST-SPF30 Tudor domain had a weak interaction with FLAG-Aven , but not with FLAG-AvenΔRGG ( Figure 3A ) . To verify whether Aven interacts with TDRD3 and SMN in vivo , co-immunoprecipitations were performed . Endogenous TDRD3 and SMN co-IP with FLAG-Aven , but not with FLAG-AvenΔRGG ( Figure 3B , C ) . To verify whether FLAG-Aven interacts with TDRD3 and SMN in an arginine methylation-dependent manner , co-immunoprecipitations were performed in PRMT1-depleted and control HEK293T cells . Indeed , cells depleted of PRMT1 showed reduced interaction between FLAG-Aven and SMN and TDRD3 , as compared to the control ( Figure 3D ) . To confirm the interaction of endogenous Aven with SMN and TDRD3 , conditional PRMT1FL/−;CreERT MEFs were treated with OHT or a vehicle , and the cellular lysates were IP with anti-Aven antibodies . Immunoprecipitates were resolved by SDS-PAGE and immunoblotted with anti-SMN and anti-TDRD3 antibodies . The ability of SMN and TDRD3 to co-immunoprecipitate with endogenous Aven was lost in PRMT1-deficient cells ( +OHT; Figure 3E ) . These findings confirm that methylation of the RGG/RG motif is required for interaction between the Aven/TDRD3 and Aven/SMN complexes . 10 . 7554/eLife . 06234 . 007Figure 3 . Tudor domains of TDRD3 and SMN recognize methylated Aven . ( A ) Recombinant Tudor domains of TDRD3 , SPF30 , and SMN were fused GST and used in ‘pull-down’ assays with HEK293T lysates expressing pcDNA3 . 1 ( control ) , FLAG-Aven , or FLAG-AvenΔRGG . The bound proteins were separated by SDS-PAGE and immunoblotted with anti-FLAG antibodies . ( B , C ) Lysates from HEK293T lysates expressing pcDNA3 . 1 ( control ) , FLAG-Aven , or FLAG-AvenΔRGG were IP with anti-FLAG antibodies . Co-immunoprecipitation of endogenous TDRD3 and SMN was detected by immunoblotting . ( D ) Aven interaction with TDRD3 and SMN was reduced in cells deficient for PRMT1 using siRNAs . FLAG-Aven was co-expressed with either siControl or siPRMT1 in U2OS cells . Anti-FLAG antibody immunoprecipitations were performed and the presence of endogenous TDRD3 and SMN monitored by immunoblotting following separation by SDS-PAGE . ( E ) PRMT1FL/−;CreERT MEFs treated with OHT for 6 days or left untreated were lysed and IP with anti-Aven antibodies . Co-immunoprecipitation of endogenous TDRD3 and SMN was detected by immunoblotting ( upper panels ) . DOI: http://dx . doi . org/10 . 7554/eLife . 06234 . 007 To identify the interactome of the RGG/RG motif of Aven , we used a stable isotope labeling by amino acids in cell culture ( SILAC ) approach to quantify protein complexes differentially associated with Aven and AvenΔRGG ( Blagoev et al . , 2003 ) . U2OS cells transfected with pcDNA3 . 1 , FLAG-Aven , or FLAG-AvenΔRGG were light ( L ) , medium ( M ) , or heavy ( H ) SILAC labeled , respectively , and IP with anti-FLAG antibodies . The data were expressed as fold-enrichment of Aven over control ( M/L ) and AvenΔRGG over control ( H/L ) . Aven was identified and quantified with high M/L and H/L ratios , while PRMT1 was only found with a high ratio M/L showing as expected that it associates with its RGG-containing substrates ( Figure 4—source data 1 ) . Of the 146 proteins enriched with Aven , but not AvenΔRGG , ∼23% were ribosomal proteins and ∼10% were RBPs , their association is likely RNA-dependent ( Figure 4—source data 1 ) . These data suggested that Aven , but not AvenΔRGG , associates with ribosomal proteins and/or RNP complexes , thereby implying that Aven associates with ribosomes in an RGG/RG motif-dependent manner . To address this question , we first assessed whether endogenous Aven and PRMT1 associate with ribosomes by sedimenting cytoplasmic extracts on 5–50% sucrose gradients by ultracentrifugation ( Figure 4A ) . Subsequently , sucrose gradients were fractionated to separate cytoplasmic mRNPs , ribosomal subunits , monosomes , and polysomes ( Gandin et al . , 2014 ) , and the amount of Aven , PRMT1 , rpS6 ( ribosomal protein S6 ) , and β-tubulin in each fraction was determined by immunoblotting . Aven and PRMT1 co-sedimented with the heavier polysomal fractions with rpS6 , while β-tubulin was restricted to the lighter fractions of the sucrose gradient corresponding to cytoplasmic mRNP fractions ( Figure 4B , fractions 11 to 16 ) . To investigate whether the RGG/RG motif and its methylation regulates the recruitment of Aven to polysomes , we performed polysomal fractionation in HEK293T cells transfected with FLAG-Aven or FLAG-AvenΔRGG ( Figure 4—figure supplement 1 ) . FLAG-Aven co-sedimented with the heavier polysomal fractions , while FLAG-AvenΔRGG was more shifted towards the lighter fractions ( Figure 4C , D ) and this was quantified ( Figure 4—figure supplement 1 ) . These findings suggest that RGG/RG motif of Aven is required for its recruitment to polysomes . To further investigate the role of RGG/RG motif methylation in polysomal localization of Aven , the FLAG-Aven was co-transfected with siPRMT1 , which reduced PRMT1 expression by ∼2 . 7-fold ( Figure 4—figure supplement 1 ) . Similar to what has been reported in fission yeast ( Bachand and Silver , 2004 ) , PRMT1 depletion did not have a major effect on the monosome/polysome ratio , thus , indicating that PRMT1 does not exert a major impact on global protein synthesis ( Figure 4—figure supplement 1 ) . Nonetheless , PRMT1 depletion shifted FLAG-Aven into the lighter fractions , as compared to a control ( compare Figure 4E and Figure 4G , quantified in Figure 4—figure supplement 1 ) . We confirmed that Aven does not co-sediment with mRNPs other than polysomes by showing that puromycin , an aminonucleoside antibiotic that dissociates polysomes ( Blobel and Sabatini , 1971 ) , leads to redistribution of both FLAG-Aven and rpS6 towards the lighter fractions corresponding to free ribosomal subunits , monosomes , and cytoplasmic mRNPs ( Figure 4F ) . Taken together , our findings suggest that the arginine methylation of the Aven RGG/RG motif by PRMT1 is required for the association of Aven with polysomes . 10 . 7554/eLife . 06234 . 008Figure 4 . Methylation of Aven and its association with TDRD3 and SMN is required for polysomal localization . ( A ) Cytoplasmic extracts from HEK293T cells were sedimented by centrifugation on a 5–50% sucrose gradient . Polysome profiles were obtained by continuous monitoring of UV absorbance at 254 nm . 40S , 60S , and 80S indicate the positions of the respective ribosomal subunits and the monomer on the gradient . ( B ) The distribution of endogenous Aven and PRMT1 across the gradient of panel A was monitored by immunoblotting . Ribosomal protein rpS6 was used as a loading control , whereas β-tubulin served as a cytoplasmic marker . ( C–H ) The distribution of FLAG-Aven or FLAG-AvenΔRGG across the gradient was monitored by immunoblotting as well as FLAG-Aven in siControl , siPRMT1 , siSMNsiTDRD3 , or with puromycin treatment . Both short ( 5 s , panels C–H ) and long exposures ( 30 s , panels C–H ) are shown . rpS6 was used as a loading control . The exposure time was determined using a standard curve with increasing amounts of lysates expressing FLAG-Aven immunoblotted with anti-FLAG antibodies for various times . Each polysomal profile experiment was performed independently three times . DOI: http://dx . doi . org/10 . 7554/eLife . 06234 . 00810 . 7554/eLife . 06234 . 009Figure 4—source data 1 . Quantitative mass spectrometry of Aven and AvenΔRGG interacting proteins . DOI: http://dx . doi . org/10 . 7554/eLife . 06234 . 00910 . 7554/eLife . 06234 . 010Figure 4—figure supplement 1 . Polysomal profiles of siRNA-treated cells and quantification of FLAG-Aven and FLAG-AVENΔRGG in polysomal fractions . ( A ) Immunoblots confirming the knockdown of PRMT1 and β-Tubulin was used as a loading control . The knockdown was ∼2 . 7-fold , as assessed by densitometric scanning . ( B ) Overlap of polysome profiles of cells overexpressing FLAG-Aven , FLAG-AvenΔRGG , FLAG-AVEN; siPRMT1 , FLAG-AVEN; puromycin treated . Cytoplasmic extracts from the indicated cells were sedimented by centrifugation on a 5–50% sucrose gradient , shown as fraction numbers 5 to 16 . Polysome profiles were obtained by continuous monitoring of UV absorbance at 254 nm . 40S , 60S , and 80S indicate the positions of the respective ribosomal subunits and the monomer on the gradient . ( C ) Overlap of polysome profiles of cells overexpressing FLAG-Aven; siGFP ( siCTRL ) or FLAG-AVEN; siSMNsiTDRD3 . ( D–H ) Quantification of FLAG-Aven in each fraction of two representative polysome experiments using densitometry . Relative FLAG-Aven expression levels were normalized to rps6 , and values expressed in terms of arbitrary densitometric units were shown as fold change of the most abundant fraction . Dashed lines indicate the densitometric units of the most abundant fraction in polysomes . DOI: http://dx . doi . org/10 . 7554/eLife . 06234 . 010 TDRD3 and SMN are known to be polysome-bound ( Goulet et al . , 2008; Sanchez et al . , 2013 ) . Thus to determine their requirement for polysomal localization of Aven , FLAG-Aven was transfected in HEK293T cells depleted of SMN or TDRD3 or both using siRNAs . The depletion of TDRD3 ( sixfold ) and SMN ( twofold ) was confirmed by immunoblotting ( compare Figure 4G and Figure 4H ) . FLAG-Aven co-sedimented in the polysome fractions with endogenous TDRD3 , SMN , and rpS6 in siCTRL ( Figure 4G , fractions 11 to 16 ) . In contrast , depletion of both TDRD3 and SMN , but not single TDRD3 or SMN depletion ( data not shown ) reduced polysomal association of Aven , as compared to the control ( Figure 4H ) . These findings suggest that TDRD3 and SMN are required for the recruitment of Aven to polysomes , whereby these proteins likely play a redundant role . Next , we reasoned that since RGG/RG motifs are encoded by G-rich sequences ( codons: Gly GGN; Arg CGN or AGA/G ) , it is likely that certain mRNAs encoding RGG/RG motif-containing proteins may occasionally comprise a PG4 sequences ( Figure 5A ) . To identify these RGG/RG encoding sequences and to identify all PG4 sequences in the coding sequences , we performed a bioinformatic search for PG4 sequences in mRNA-coding regions ( Gx-N1-7-Gx-N1-7-Gx-N1-7-Gx , where x ≥ 3 and N is any of the nucleotides [A , C , G , or U] ) , and it revealed ∼1600 PG4 in human ORFs ( Figure 5—source data 1 ) . We also provide a cG/cC score ( Figure 5—source data 1 ) , where >2 predicts higher G4 formation over Watson–Crick base pairing with neighboring sequences ( Beaudoin et al . , 2014 ) . In addition , we also use RNAfold v2 . 1 . 7 , which ( Figure 5—source data 1 ) is a new scoring system to identify RNA G4-folding ( Lorenz et al . , 2013 ) . RNAfold v2 . 1 . 7 was less efficient in predicting G4 formation than the cG/cC ratio ( Beaudoin et al . , 2014 ) . 10 . 7554/eLife . 06234 . 011Figure 5 . Aven RGG/RG motif binds G4 RNA structures of MLL1 and MLL4 . ( A , B ) RNA sequences of the RGG/RG motifs and the PG4 motifs of MLL1 and MLL4 . ( C , D ) Biotinylated MLL1 G4 or a mutant sequence ( G4m ) , biotinylated MLL4 G4 or a mutant sequence ( G4m ) bound to Streptavidin beads were incubated with HEK293T cell lysates . The bound proteins were washed with increasing concentrations of NaCl and visualized by SDS-PAGE followed by immunoblotting with anti-Aven antibodies . ( E , F ) HEK293T cells expressing FLAG-Aven and FLAG-AvenΔRGG were processed as in panel C , D except the bound proteins were visualized by immunoblotting with anti-FLAG antibodies . ( G ) Photocrosslinking IP experiments were performed using anti-FLAG antibodies . The bound RNA was analyzed in triplicate from two biological replicates by RT-qPCR with the primers spanning the PG4 sequence or a sequence ∼300 nucleotides downstream . The levels of bound RNA in immunoprecipitates were normalized to the levels of the total RNA in the input . Mean values are expressed as fold enrichment over pcDNA3 . 1 . Error bars represent ±SEM . *p < 0 . 05 , **p < 0 . 001 , n . s . non-significant . The experiment was performed twice . ( H ) Photocrosslinking IP experiments were performed on HEK293T cells using anti-Aven antibodies . The bound RNA was analyzed in triplicates by real-time RT-PCR with the primers spanning the PG4 sequence , as indicated in panel G . The level of bound RNA in immunoprecipitates was normalized to the levels of the total RNA in the input . Mean values are expressed as fold enrichment over IgG . Error bars represent ±SEM . *p < 0 . 05 , **p < 0 . 001 , n . s . non-significant . The experiment was performed twice . ( I ) In-line probing of MLL1 and MLL4 PG4 . The nucleotide sequence of the mixed lineage leukemias ( MLLs ) PG4 is shown below , the boxed guanines represent the predicted G-tracks . K+/Li+ ratios of the band intensities of the MLLs G4 ( black ) and G/A-mutant ( white ) for each nucleotide are shown . Error bars represent ± standard deviation . The experiment was performed twice . The dashed line represents a twofold change , an arbitrary set threshold that indicates G4 formation when exceeded . DOI: http://dx . doi . org/10 . 7554/eLife . 06234 . 01110 . 7554/eLife . 06234 . 012Figure 5—source data 1 . G4 sequences in coding regions of mRNAs . DOI: http://dx . doi . org/10 . 7554/eLife . 06234 . 01210 . 7554/eLife . 06234 . 013Figure 5—figure supplement 1 . Sequence conservation of the MLL1 and MLL4 PG4 sequences . DOI: http://dx . doi . org/10 . 7554/eLife . 06234 . 013 We observed a preference for amino acids G >> R/A/P > E/L consistent with the frequency of guanine in each codon . Some PG4 sequences identified encoded RG-rich sequences , as well as glycine-rich sequences . As Aven is a known survival protein and its depletion decreases the proliferation of leukemic cells ( Eismann et al . , 2013 ) , we searched for mRNAs harboring PG4 sequences involved in leukemic cell survival . We identified human KMT2A , KMT2B , and KMT2D ( Figure 5—source data 1 ) , the family of MLL histone methyltransferases known to be mutated and/or part of fusion proteins , as the result of chromosomal translocations in leukemia ( Liedtke and Cleary , 2009; Smith et al . , 2011 ) . KMT2A ( MLL1 ) has a PG4 sequence ( nucleotide 223–253 ) that encodes a glycine-rich sequence ( AGSSGAGVPGG , Figure 5B; cG/cC score of 2 . 975 , Figure 5—source data 1 ) , KMT2B ( MLL4 ) has 3 PG4 sequences ( nucleotides 262–287; 3139–3166; 6274–6301 ) that encode an RGG/RG motif ( RVQRGRGRG , Figure 5B; cG/cC score of 2 . 796 , Figure 5—source data 1 ) , a glycine-rich ( RGAGAGGPRE; cG/cC score of 1 . 887 , Figure 5—source data 1 ) , an alanine-glycine-rich ( RAGVLGAAGD; cG/cC score of 2 . 468 , Figure 5—source data 1 ) sequences , and KMT2D has a PG4 from nucleotide 961 to 986 that encodes RVCRACGAG ( cG/cC score of 2 . 029 , Figure 5—source data 1 ) . We examined whether Aven associated with the conserved PG4 RNA sequences ( Figure 5—figure supplement 1 ) of KMT2A ( MLL1 ) and KMT2B ( MLL4 ) near the initiator ATG ( i . e . , 223–253 and 262–287 ) . Indeed , Aven bound either MLL1 or MLL4 PG4 RNA sequences in the RGG/RG motif-dependent manner , but not to those harboring guanine to adenine mutants ( Figure 5C–F ) . We next examined whether FLAG-Aven binds the endogenous MLL1 and MLL4 PG4 sequences in vivo using photocrosslinking with 4-thiouridine followed by immunoprecipitations ( Cahill et al . , 2002 ) and treatment with RNase I to digest RNAs into fragments of 50–300 nucleotides in length ( Huppertz et al . , 2014 ) . HEK293T transfected with pcDNA3 . 1 , FLAG-Aven , or FLAG-AvenΔRGG were prepared for CLIP , as described in ‘Materials and methods’ and IP with anti-FLAG antibodies ( Figure 5G ) . Anti-FLAG immunoprecipitations of FLAG-Aven expressing cells , but not pcDNA3 . 1 or FLAG-AvenΔRGG-transfected cells , enriched ∼30-fold for the MLL4 G4 sequence , whereas an RNA region 300 nucleotides downstream was not enriched ( noG4; Figure 5G , right panel ) . MLL1 G4 sequence was also enriched ( ∼fivefold ) in FLAG-Aven immunoprecipitations , but not a region without G4 motifs ( noG4 ) ( Figure 5G , left panel ) . We investigated whether endogenous Aven could associate with the PG4s of MLL1 and MLL4 mRNAs . Photocrosslinking immunoprecipitations assays confirmed that MLL1 and MLL4 PG4 sequences associated with IP endogenous Aven , but not immunoglobulin G control albeit with a lower affinity ( ∼fourfold to sixfold , Figure 5H ) than with overexpressing FLAG-Aven ( Figure 5G ) and that ultraviolet light was required for this association ( Figure 5H ) . The fold induction observed between Aven and the G4 sequences is likely under represented due to the fact that the crosslinks impede the reverse transcription reactions ( Huppertz et al . , 2014 ) . These findings demonstrate that Aven is associated in vivo with the PG4s of MLL1 and MLL4 . We next performed in-line probing experiments , to determine whether the MLL PG4 sequences formed bona fide G4 structures . This assay compares the cleavage pattern in two conditions: in presence of K+ , which support G4 formation , and in presence of Li+ , which does not . G4 folding leads to an increased cleavage for the nucleotides within the loop regions , since they bulge out ( Beaudoin and Perreault , 2010 , 2013 ) . Using MLL1 and MLL4 PG4 , we observed increased cleavage in the predicted loops of the formed G4 , when incubated in K+ compared to Li+ . However , no such difference was observed for a mutant RNA , where guanines were replaced with adenines ( Figure 5I ) , confirming that the PG4 of MLL1 and MLL4 form G4 RNA structures . Aven is required for the proliferation of T-ALL cells ( Eismann et al . , 2013 ) . We generated Aven-deficient T-ALL cells , MOLT4 and CCRF-CEM using a lentivirus that expresses an shRNA against Aven and we achieved >80% knockdown ( Figure 6A , B ) . We next monitored the levels of MLL1 and MLL4 protein by immunoblotting and their mRNAs by RT-qPCR . Both MLL1 and MLL4 protein levels were reduced in Aven-deficient MOLT4 and CCRF-CEM cells ( Figure 6A , B ) , whereas the levels of corresponding mRNAs remained unchanged ( Figure 6C , D ) . Next , we investigated whether the reduced protein levels of MLL1 and MLL4 in Aven-depleted cells are associated with the reduced expression of HOX genes , which are well-established transcriptional targets of MLLs ( Krivtsov and Armstrong , 2007 ) . Aven-deficient cells exhibited reduced expression of several key HOX genes such as HOXA9 , HOXA7 , HOXA1 , and MEIS1 ( Figure 6E , F ) . These findings suggest that Aven regulates the translation of MLL1 and MLL4 mRNAs thereby leading to an increase in MLL1 and MLL4 protein levels and an increase in the transcription of leukemic genes . 10 . 7554/eLife . 06234 . 014Figure 6 . Aven regulates MLL1 and MLL4 protein expression required for leukemic cell survival . ( A , B ) Cellular lysates from MOLT-4 and CCRF-CEM cells stabling expressing shCTRL or shAven were separated by SDS-PAGE and immunoblotted with anti-MLL1 , anti-MLL4 , anti-Aven , and β-actin antibodies . The experiment was performed three times . ( C–F ) RT-qPCR of the indicated mRNAs was performed from RNA isolated from shCTRL and shAven MOLT-4 and CCRF-CEM cells and expressed as a relative fold change normalized to rpS6 levels . Error bars ±SEM is shown . The significance was measured by the Student's t-test and defined as *p < 0 . 05 , **p < 0 . 001 , n = 3 . ( G , H ) Proliferation curves for shControl ( CTRL ) , shPRMT1 , and shAven MOLT-4 and CCRF-CEM cells are shown . Immunoblots confirm the depletion of PRMT1 in MOLT-4 and CCRF-CEM cells . Error bars ± standard deviation is shown . The data were analyzed using ANOVA ( Analysis of Variance ) followed by post hoc comparison using Tukey test . *p < 0 . 05 , **p < 0 . 001 , n = 3 . DOI: http://dx . doi . org/10 . 7554/eLife . 06234 . 014 Aven-deficient MOLT4 and CCRF-CEM had decreased proliferation rates consistent with Aven being a survival protein ( Figure 6G , H ) . Interestingly , the depletion of PRMT1 with shRNAs phenocopied Aven depletion , that is , also had reduced proliferation rates ( Figure 6G , H ) . These findings show that Aven and PRMT1 are required for the proliferation of T-ALL cells . We generated Aven−/− HEK293T cells ( clone #2 ) using CRISPR/Cas9 technology ( Mali et al . , 2013 ) and this was confirmed by immunoblotting ( Figure 7A ) . A minor Aven ∼34 kDa fragment ( Figure 7A , denoted by asterisk ) was observed in HEK293T cells and likely represents a Cathepsin D cleaved fragment reported previously ( Melzer et al . , 2012 ) . Aven depletion did not influence the number of ribosomes involved in polysomes , as compared to control ( Figure 7B ) , suggesting that general mRNA translation was not affected by Aven . However , a specific subset of mRNAs may still be particularly sensitive to changes in Aven levels . To investigate this possibility , we monitored the distribution of MLL1 , MLL4 , and β-actin mRNA in polysomal fractions from Aven-proficient and Aven-deficient cells . Polysomal fractions were isolated and MLL1 , MLL4 mRNAs , as well as β-actin mRNA , were quantified by RT-qPCR . Loss of Aven expression reduced the amounts of MLL1 mRNA in the heavy polysomal fraction ( fractions 12–15 ) , with concomitant increase in lighter polysomal fractions ( fractions 6 and 7; Figure 7C ) . Similarly , MLL4 mRNA was reduced in the heavy polysomal fraction ( fractions 12–14 ) and a shift towards the light polysomal and pre-polysomal ( fractions 7–8 and 10–11 ) ( fractions 7–8 ) . As a control , we monitored the levels of mRNAs encoding β-actin and depletion of Aven did not have a major effect on its distribution , inasmuch as the most of β-actin mRNA was associated with heavy polysomes ( Figure 7E ) . These findings suggest that Aven selectively regulates the polysomal association of MLL1 and MLL4 mRNAs . 10 . 7554/eLife . 06234 . 015Figure 7 . Aven regulates polysomal association of MLL1 and MLL4 , but not β-actin mRNA . ( A ) Aven-deficient HEK293T cells were generated by CRISPR/Cas9 . Stable clones were obtained Aven+/+ ( clone #7 ) and Aven−/− ( clone #2 ) . Anti-Aven , anti-rpS6 , and anti-β-actin immunoblots of TCL are shown . The asterisks denote a minor Aven species of lower molecular mass . The band at ∼37 kDa is a non-specific band recognized by the anti-Aven antibody . n = 3 . ( B ) Polysome profiles of Aven+/+ and Aven−/− HEK293T cells are shown . Cytoplasmic extracts from HEK293T cells were sedimented by centrifugation on a 5–50% sucrose gradient , shown as fraction numbers 5–15 . Polysome profiles were obtained by continuous monitoring of UV absorbance at 254 nm . 40S , 60S , and 80S indicate the positions of the respective ribosomal subunits and the monomer on the gradient . ( C–E ) The indicated polysomal fractions were isolated , the RNA purified , and the presence of MLL1 , MLL4 , or β-actin was quantified by qRT-PCR . mRNAs in each fraction are represented as the percentage of input . Error bars represent ±SEM , n = 5 . DOI: http://dx . doi . org/10 . 7554/eLife . 06234 . 015 Since PRMT1 regulates the ability of Aven to associate with heavy polysomes , we investigated whether PRMT1-depleted cells also had reduced MLL1 and MLL4 mRNAs in heavy polysome fractions . Similarly to Aven , PRMT1 depletion ( Figure 8A ) did not influence the number of ribosomes involved in polysomes , as compared to a control , suggesting that PRMT1 does not affect global mRNA translation ( Figure 8B ) . HEK293T cells were transfected with siGFP ( siCTRL ) or siPRMT1 for 72 hr , and the distribution of MLL1 and MLL4 mRNAs in polysomal fractions was monitored by RT-qPCR . Comparably to Aven , PRMT1 depletion had a striking effect on the distribution of MLL1 and MLL4 mRNAs in polysomes , as illustrated by their dramatic shift towards ligher fractions as compared to a control ( Figure 8C , D ) . In contrast , depletion of PRMT1 resulted in a modest shift in β-actin mRNA , whereby the majority of β-actin mRNA remained in heavy polysome fractions . ( Figure 8E , fractions 11–15 ) . These findings suggest that Aven arginine methylation by PRMT1 regulates polysomal association of MLL1 and MLL4 , but not β-actin mRNAs . 10 . 7554/eLife . 06234 . 016Figure 8 . PRMT1 is required for the polysomal association of MLL1 and MLL4 , but not β-actin mRNA . ( A ) PRMT1 was depleted by siRNA and cell extracts were immunoblotted with anti-PRMT1 or anti-β-actin antibodies . ( B ) Polysome profiles of siGFP ( siCTRL ) or siPRMT1 HEK293T cells are shown . Cytoplasmic extracts from HEK293T cells were sedimented by centrifugation on a 5–50% sucrose gradient , shown as fraction numbers 5–15 . Polysome profiles were obtained by continuous monitoring of UV absorbance at 254 nm . 40S , 60S , and 80S indicate the positions of the respective ribosomal subunits and the monomer on the gradient . ( C–E ) The indicated polysomal fractions were isolated , the RNA purified , and the presence of MLL1 , MLL4 , or β-actin was quantified by qRT-PCR . mRNAs in each fraction are represented as the percentage of input . Error bars represent ±SEM , n = 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 06234 . 016 We closely examined the SILAC data for non-ribosomal proteins that are enriched in FLAG-Aven , but not with FLAG-AvenΔRGG immunoprecipitates that may function in unwinding G4 structures . The ATP-dependent RNA helicase DHX36 , also known as G4 resolvase I , had an M/H ratio of 2 . 69 ( Figure 4—source data 1 ) . DHX36 has been reported to unwind G4 RNA structures ( Creacy et al . , 2008; Lattmann et al . , 2010; Booy et al . , 2012 ) . We postulated that Aven could recruit the RNA helicases DHX36 to resolve G4 structures to facilitate protein synthesis . First , to validate whether the Aven RGG/RG motif is essential for interaction with the RNA helicase DHX36 , we performed co-immunoprecipitations from cellular lysates expressing FLAG-Aven or FLAG-AvenΔRGG . The bound proteins were immunoblotted with anti-DHX36 antibodies . Indeed , DHX36 co-IP with FLAG-Aven , but not with FLAG-AvenΔRGG ( Figure 9A ) . Moreover , we observed that DHX36 localized in the fast-sedimenting , heavier polysomal fractions with the control rpS6 ( Figure 9B , fractions 12–15 ) . To examine whether DHX36 influences the polysomal localization of MLL1 and MLL4 mRNAs , we monitored their mRNAs in polysomal fractions . Consistently with the observations for Aven and PRMT1 , DHX36 depletion in HEK293T cells did not have a major effect on polysome absorbance profiles , thus , indicating that DHX36 does not affect global protein synthesis ( Figure 9C ) . However , both MLL1 and MLL4 , but not β-actin mRNA , shifted toward lighter polysomal fractions in cells depleted of DHX36 , as compared to the control ( Figure 9D–F ) . These findings suggest that similarly to PRMT1 and Aven , DHX36 regulates translation of MLL1 and MLL4 mRNAs . 10 . 7554/eLife . 06234 . 017Figure 9 . DHX36 is required for MLL1 and MLL4 mRNA polysomal association . ( A ) HEK293T cells expressing FLAG-Aven and the FLAG-AvenΔRGG were IP with anti-FLAG agarose beads and the bound proteins were immunoblotted with anti-DHX36 antibodies . TCL were immunoblotted with anti-DHX36 and anti-FLAG antibodies as indicated . ( B ) Proteins from the polysomal fractions isolated from HEK293T cells were TCA precipitated , separated by SDS-PAGE , and immunoblotted with anti-DHX36 and anti-rpS6 antibodies . The experiment was performed n = 4 times and a typical polysomal profile is shown . ( C ) Immunoblots of TCL from siGFP ( CTRL ) and siDHX36-transfected HEK293T cells are shown . Polysome profiles siCTRL and siDHX36-transfected HEK293T cells . Cytoplasmic extracts from HEK293T cells were sedimented by centrifugation on a 5–50% sucrose gradient , shown as fraction numbers 5 to 15 . Polysome profiles were obtained by continuous monitoring of UV absorbance at 254 nm . 40S , 60S , and 80S indicate the positions of the respective ribosomal subunits and the monomer on the gradient . ( D–F ) The indicated polysomal fractions were isolated , total RNA isolated , and the presence of MLL1 , MLL4 , or β-actin was quantified by RT-qPCR . mRNAs in each fraction are represented as the percentage of input . Error bars represent ±SEM , n = 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 06234 . 017 We next determined whether the polysomal association of MLL4 mediated by Aven requires an intact G4 structure . Reporter mRNAs harboring the G4 of MLL4 or a mutated G4 motif was inserted in-frame with the ORF of luciferase ( Figure 10A ) . We examined whether Aven could promote the translation of the luciferase reporter protein in a G4-dependent manner and whether this was rescued by Aven re-expression in Aven-deficient cells . pGL3 , pGL3-MLL4 , or pGL3-MLL4 G4 mutant were transfected in Aven−/− HEK293T cells with pRenilla , as a control for transfection efficiency , along with either pcDNA3 . 1 , FLAG-Aven , FLAG-AvenΔRGG , or FLAG-AvenR-K . The FLAG-AvenR-K protein was generated where the arginines in the RGG/RG motif were substituted for lysines to maintain the charge of the N-terminus of Aven . 24 hr post-transfection , the cells were harvested for dual luciferase assay . The presence of the MLL4 G4 sequence inhibited the relative luciferase activity by >75% in Aven−/− HEK293T ( Figure 10B , compare white and black bars labeled pcDNA3 . 1 ) . The inhibition caused by the presence of the MLL4 G4 sequence was relieved by the transfection of FLAG-Aven , but not FLAG-AvenΔRGG nor FLAG-AvenR-K ( Figure 10B , black bars ) . The presence of Aven did not have any significant effects on the luciferase expressed from pGL3 or pGL3-MLL4:G4mutant ( Figure 10B ) . We next examined arginine methylation by PRMT1 and DHX36 were required for the stimulation of translation by FLAG-Aven . HEK293T cells were transfected with FLAG-Aven and pGL3-MLL4:G4 in the presence of siCTRL , siPRMT1 , or siDHX36 . The absence of PRMT1 or DHX36 blocked FLAG-Aven from stimulating translation ( Figure 10C , D ) , suggesting that both arginine methylation of Aven and DHX36 helicase activity are required to regulate the translation of mRNAs with G4 structures within their ORFs . 10 . 7554/eLife . 06234 . 018Figure 10 . PRMT1 and Aven RGG/RG motif required for optimal translation of MLL4 G4 sequence . ( A ) Schematic of the luciferase reporter plasmid pGL3 , as well as the chimeric pGL3-MLL4-G4 and pGL3-MLL4-G4mutant . pGL3-MLL4-G4 harbors the human MLL4 G4 sequence nucleotide 262 to 318 inserted in-frame at the N-terminus of luciferase , while pGL3-MLL4-G4mutant contains glycine to alanine mutations that disrupts the G4 structure . ( B ) Aven−/− HEK293T cells were transfected with the following reporter genes pGL3 , pGL3-MLL4-G4 , or pGL3-MLL4-G4mutant and pRenilla as well as pcDNA3 . 1 , FLAG-Aven , FLAG-AvenΔRGG , or FLAG-AvenR-K . The cells were harvested 24 hr post-transfection and dual luciferase assays were performed . The relative luciferase/Renilla ratio was normalized to 1 . 0 in pGL3 pcDNA3 . 1 transfected cells . Extracts were immunoblotted with anti-FLAG antibodies to confirm Aven , AvenΔRGG , or AvenR-K expression . Error bars represent standard deviation values . The experiments were performed three independent times ( n = 3 ) and each independent experiment was performed in technical triplicates . The significance was measured by ANOVA followed by post hoc comparison using Tukey test . *p < 0 . 05 . ( C ) HEK293T cells were co-transfected with FLAG-Aven and either siGFP ( siCTRL ) or siPRMT1 along with the following reporter genes pGL3 , pGL3-MLL4-G4 , or pGL3-MLL4-G4mutant and pRenilla . The cells were harvested 24 hr post-transfection and dual luciferase assays were performed . The relative luciferase/Renilla ratio was normalized to 1 . 0 in pGL3 siCTRL transfected cells . Extracts were immunoblotted with anti-PRMT1 or anti-β-actin antibodies , as indicated . Error bars represent standard deviation values . The experiments were performed three independent times ( n = 3 ) and each independent experiment was performed in technical triplicates . The significance was measured by ANOVA followed by post hoc comparison using Tukey test . *p < 0 . 05 , n . s . non-significant . ( D ) HEK293T cells were co-transfected with FLAG-Aven and either siGFP ( siCTRL ) or siDHX36 along with the following reporter genes pGL3 , pGL3-MLL4-G4 , or pGL3-MLL4-G4mutant and pRenilla . The cells were harvested 24 hr post-transfection and dual luciferase assays were performed . The relative luciferase/Renilla ratio was normalized to 1 . 0 in pGL3 siCTRL transfected cells . Extracts were immunoblotted with anti-DHX36 or anti-β-actin antibodies , as indicated . The error bars represent ± the standard deviation . Experiments were performed three times ( n = 3 ) and each experiment was analyzed in triplicates . The significance was measured by ANOVA followed by post hoc comparison using Tukey test . *p < 0 . 05 , n . s . non-significant . DOI: http://dx . doi . org/10 . 7554/eLife . 06234 . 018
The MLL proto-oncogene encodes a histone methyltransferase implicated in epigenetic modifications , regulating gene expression for embryonic development and hematopoiesis ( Liedtke and Cleary , 2009; Smith et al . , 2011 ) . MLL is a recurrent site of DNA translocations resulting in an MLL fusion protein where the N-terminus of the MLL is fused to a variety of proteins ( Liedtke and Cleary , 2009; Smith et al . , 2011 ) . In the present manuscript , we identify , within the mRNA-coding regions of MLL1 and MLL4 , RNA elements that regulate its polysomal association and protein synthesis . These RNA elements are located between 200 and 300 nucleotides downstream of the initiator methionine ATG and encode protein sequences rich in glycines and arginine–glycine repeats in MLL1 and MLL4 . The function of these N-terminal repeats is unknown . We show that Aven binds the MLL1 and MLL4 G4 RNA structures in vitro and in vivo with its RGG/RG motif . Aven was required for the translational regulation of MLL1 and MLL4 , as Aven-deficient T-ALL cells exhibited decreased MLL1 and MLL4 protein expression and consequently decreased the expression of their downstream targets including , the HOX genes . The association of Aven with polysomes required the methylation of its RGG/RG motif by PRMT1 and interaction with methyl-binding proteins , TDRD3 and SMN . The Aven interaction with TDRD3 and SMN may require other protein or RNA components in the complex for enhanced association . Deficiency of Aven or PRMT1 in acute leukemic cell lines led to decreased cell proliferation . Taken together , our studies suggest that Aven regulates the translation of MLL1 and MLL4 required for cancer survival and that targeting this pathway may have therapeutic potential . RGG/RG motifs have the biochemical properties to bind both RNA and proteins to fulfill their emerging roles in assembly of RNP complexes and translational control ( Rajyaguru and Parker , 2012; Thandapani et al . , 2013 ) . The RGG/RG motif of yeast proteins Scd6 , Npl3 , and Sbp1 was shown to interact with the translational initiation factor eIF4G and repress translation by preventing the formation of pre-initiation complex ( Rajyaguru et al . , 2012 ) . In trypanosomes , the RGG/RG motif of SCD6 is involved in regulating the type and number of RNP granules ( Krüger et al . , 2013 ) . Amyloid-like fibers were formed when the RGG/RG motif of FUS was incubated with RNA ( Schwartz et al . , 2013 ) . These fibers are characterized by the reversible transformation from soluble to polymeric state ( Han et al . , 2012; Kato et al . , 2012 ) . Although many proteins have an RGG/RG motif , the Aven RGG/RG motif may be more accessible , as it is located at the N-terminus and may protrude outwards . Since the Aven RGG/RG motif is not required for self-association ( Figure 2—figure supplement 1 ) , this suggests that an Aven dimer has 2 protruding RGG/RG motifs that can each mediate their own interactions . Therefore , we speculate that Aven functions as a scaffolding protein to assemble translationally competent RNPs for certain mRNAs containing G4 motifs ( Figure 11 ) . 10 . 7554/eLife . 06234 . 019Figure 11 . Model denoting the role of arginine methylated Aven by PRMT1 and DHX36 in the translation of G4 harboring MLL1 and MLL4 proteins . DOI: http://dx . doi . org/10 . 7554/eLife . 06234 . 019 DHX36 is a DEAH ( aspartic acid , glutamic acid , alanine , histidine ) -box helicase and it is the only G4 RNA resolvase known and is a major DNA G4 resolvase ( Creacy et al . , 2008; Lattmann et al . , 2010 ) . Aven associated with DHX36 to regulate translation of mRNAs with G4 structures . DHX36 knockdown increased the expression of PITX1 protein without changes in mRNA , suggesting that it functions in translational control ( Booy et al . , 2014 ) . Ribosomal footprinting studies have led to the proposal that elongating ribosomes likely use accessory RNA helicases ( Rouskin et al . , 2014 ) , and our data suggest that DHX36 may be such an accessory helicase . DHX36 null mice are embryonic lethal and deletion in the hematopoietic system using Vav1-Cre causes hemolytic anemia and defects at the proerythroblast stage with deregulation of genes with G4 motifs in their promoters ( Lai et al . , 2012 ) , however , a role DHX36 in translational control was not examined . Many PRMT1 substrates are RBPs with RGG/RG motifs ( Bedford and Clarke , 2009 ) and some have been shown to associate with RNAs with G4 motifs such as Nucleolin , FUS , EWS , and FMRP ( Thandapani et al . , 2013 ) . This suggests that several RBPs likely function in a similar manner to Aven in regulating accessibility of mRNPs with polysomes . It has been shown that the RGG/RG motif of FMRP is required for its polysomal association ( Blackwell et al . , 2010 ) , however , whether arginine methylation by PRMT1 regulates association is unknown . Our findings show for the first time that arginine methylation by PRMT1 regulates translational control . It is known , however , that the yeast homolog of PRMT3 ( RMT3 ) methylates rpS2 , regulating the balance between the small and large ribosomal subunits ( Bachand and Silver , 2004 ) . However , mammalian PRMT3 did not influence ribosomal assembly or polysomal formation ( Swiercz et al . , 2004 ) . It is known that secondary RNA structure including G-quadruplex structures within mRNAs hinder mRNA translation ( Koromilas et al . , 1992; Sonenberg and Hinnebusch , 2009 ) . Stable RNA secondary structures within the 5′-UTRs of mRNAs reduce cap-dependent translation by preventing assembly of the translational initiation machinery at the 5′-cap and also impair the scanning of the start site AUG by the initiation complex ( Beaudoin and Perreault , 2010; Bugaut and Balasubramanian , 2012 ) . Secondary structure in the 5′ UTRs including G4 motifs has been shown to require eIF4A for optimal translation output ( Wolfe et al . , 2014 ) . The 5′-UTR of NRAS and Zic-1 , which harbor G4 structures , reduce translation of a reporter luciferase ( Kumari et al . , 2007; Arora et al . , 2008 ) . G-quadruplex structures within ORFs of the virally encoded EBNA1 transcript were shown to hinder translational elongation by either ribosomal pausing or ribosomal dissociation ( Murat et al . , 2014 ) . We now extend these observations and identify a mechanism regulated by arginine methylation that leads to the positive regulation of mRNAs with G4 structures within their coding region . Aven is overexpressed in acute leukemia and was proposed to be a prognostic factor in acute childhood lymphoblastic leukemia for poor outcome ( Choi et al . , 2006 ) . Aven is a well-established cell survival protein or inhibitor of apoptosis that prevents apoptosis by stabilizing pro-survival protein Bcl-xL and inhibiting the function of pro-apoptotic protein Apaf-1 ( Chau et al . , 2000; Kutuk et al . , 2010 ) . It was reported that an N-terminal deleted fragment of Aven cleaved by cathepsin D harbors its anti-apoptotic function ( Melzer et al . , 2012 ) , however , such a ∼30 kDa Aven species was not visible in MOLT4 and CCRF-CEM cells ( Figure 6A , B ) and was faintly observed in HEK293T ( Figure 7A ) as previously described ( Melzer et al . , 2012 ) . Thus , cathepsin D mediated cleavage of Aven is unlikely involved in the regulation of translational control described herein . In addition to its pro-survival functions , Aven was identified to be essential for progression of acute leukemia in mice ( Eismann et al . , 2013 ) . The regions required for association with Bcl-xL and Apaf-1 reside C-terminal of the RGG/RG motif . Taken together with our findings , this suggests that Aven uses several mechanisms to increase cell survival , ( 1 ) preventing apoptosis via Bcl-xL and Apaf-1 , and ( 2 ) favoring the translation of mRNAs , including those encoding MLL1 and MLL4 required for cell survival . PRMT1 was shown to be essential for MLL by the MLL-EEN gene fusion protein ( Cheung et al . , 2007 ) . The EEN fusion partner leads to the recruitment of PRMT1 to methylate histones and lead to gene activation ( Cheung et al . , 2007 ) . Our findings identify a new role for PRMT1 in the cytoplasm that is required for cancer cell survival . This pathway is amenable to therapeutic intervention with future PRMT1 inhibitors and specific RNA G-quadruplex ligands .
HEK293T , U2OS , MOLT-4 , and CCRF-CEM were from the American Type Culture Collection ( Manassas , VA ) . PRMT1FL/−;CreERT MEFs were described previously ( Yu et al . , 2009 ) . Protein A-Sepharose , 4-hydroxytamoxifen ( OHT ) , anti-FLAG ( M2 ) antibody-coupled agarose beads , mouse anti-FLAG ( M2 ) , anti-Myc , anti-MLL4 ( WH0009757M2 ) , and anti-α-tubulin were purchased from Sigma–Aldrich ( St . Louis , MO ) . Mouse anti-Aven ( ab77014 , Abcam , UK ) was used for immunoprecipitations , and rabbit anti-Aven ( ProSci 2413 , ProScience , San Diego , CA ) was used for immunoblotting . Mouse anti-rpS6 was from Santa Cruz Biotechnology ( Santa Cruz , CA ) . Rabbit anti-PRMT1 and ASYM25b antibodies were purchased from Millipore ( Billerica , MA ) . Anti-SMN was from Transduction Laboratories ( Lexingtong , KY ) . Anti-TDRD3 was a kind gift from Mark Bedford ( Smithville , TX ) . Anti-MLL1 antibodies ( A300-086A ) were from Bethyl Laboratories ( Montgomery , TX ) and anti-DHX36 ( ab70269 ) was from Abcam . The full-length human FLAG-Aven and FLAG-AvenΔRGG lacking amino acids ( 1–73 ) were cloned in pcDNA3 . 1 . The full-length human Aven was also cloned in pcDNA3 . 1 with 5 epitope tags of Myc at the N-terminus between the BamHI and XhoI sites . GST Aven RGG/RG including amino acids ( 1–73 ) was cloned in pGEX5x-1 plasmid . FLAG-AvenR-K was assembled using G-blocks purchased from IDT . Aven arginines R5 , R8 , R11 , R13 , R14 , R17 , R19 , R24 , R28 , R37 , R50 , R51 , R53 , R55 , R57 , R60 , R63 , and R66 were replaced with lysines . pGL3-MLL4:G4 and pGL3-MLL4mutant was generated by cloning MLL4 nucleotide sequence ( 262–318 ) ( 5′-CGG GTC CAG CGG GGC CGG GGA CGG GGT CGG GGC CGG GGC TGG GGC CCG AGT CGA GGC-3′ ) or ( 5′-CGC GTC CAG CGC GCC CGC GCC CGC GCC CGC GCC CGC GCC TGG GCC CCG AGT CGA GGC-3′ ) in fusion with the luciferase coding region in the pGL3 basic plasmid at SacI and NcoI sites . HEK293T cells expressing either pcDNA3 . 1 , FLAG-Aven , and FLAG-AvenΔRGG were incubated with 100 μM 4-thiouridine ( 4SU ) for 8 hr prior to crosslinking . The cells were washed with ice-cold PBS ( phosphate buffered saline ) and irradiated with 0 . 15 J/cm2 of 365-nm UV light at 4°C . The cells were collected by centrifugation at 514×g for 1 min at 4°C . The cell pellets were resuspended in lysis buffer supplemented with protease inhibitors ( Roche , Germany ) and 0 . 5 U/ml RNasin ( Promega , Madison , WI ) ( Cahill et al . , 2002 ) . 10 μl of 1:250 dilution of RNase I ( Life Technologies , Carlsbad , CA ) and 2 μl Turbo DNase ( Life Technologies ) were added to the lysate while shaking at 37°C for 3 min . The lysates were then cleared and IP with 25 μl anti-FLAG M2 affinity beads ( Sigma ) . The beads were washed twice with high-salt buffer , twice with the lysis buffer and incubated with proteinase K buffer ( containing 1 . 2 mg/ml Proteinase K ) for 20 min at 37°C . RNA then was isolated through TRIzol reagent and subjected to RT-qPCR . For endogenous Aven , UV-crosslinked lysates were processed as described above expect they were incubated with rabbit IgG or anti-Aven ( Proscience ) . The reverse primers were used for the reverse transcription reaction: for MLL1 G4 structure , the RT reverse primer is 5′-GAG GAG GCT GCT GAG GCG GC-3′; for MLL1 negative control region ( ∼300 bp downstream of G4 ) , the RT reverse primer is 5′-TCT TCT TGA TCT TAT CTC CA-3′; for MLL4 G4 structure , the RT reverse primer is 5′-CTC TCC TCC TCC GGC ACG CAG C-3′; for MLL4 negative control region ( ∼300 bp downstream of G4 ) , the RT reverse primer is 5′-GAT TGT CAC AGC TGC TTC TGC-3′ qPCR was performed with MLL1 ( G4 structure , forward ) 5′-GGC GGG AAG CAG CGG GGC TG-3′ and MLL1 ( G4 structure , reverse ) 5′-CTG AGG CGG CGG CCG CTC CC-3′; MLL1 ( negative control , forward ) 5′-CAT CTG TGT TTT CCC CTC TA-3′ and MLL1 ( negative control , reverse ) 5′-CTT ATC TCC AGA TTT GGT CT-3′; MLL4 ( G4 structure , forward ) 5′-GCG CCG GCT CCG CCG CCT GT-3′ and MLL4 ( G4 structure , reverse ) 5′-GCA CGC AGC CTC GAC TCG GG-3′; MLL4 ( negative control , forward ) 5′-TGC AGG AGG AAG CAG CAA GC-3′ and MLL4 ( negative control , reverse ) 5′-CTG CTT CTG CCA CCA CTA CT-3′ . Polysome profiling has been performed as described in detail ( Gandin et al . , 2014 ) . Briefly , HEK293T cells in 150-mm plates were transfected with the indicated expression plasmids using Lipofectamine 2000 . Approximately 70% confluent cells were treated with 100 µg/ml cycloheximide for 5 min to ‘freeze’ mRNA translation . The cells were washed twice with ice cold-PBS and lysed in hypotonic lysis buffer containing 50 mM Tris-HCl ( pH 7 . 5 ) , 2 . 5 mM MgCl2 , 1 . 5 mM KCl , 100 μg/ml cycloheximide , 2 mM DTT , 0 . 5% Triton X-100 , and 0 . 5% sodium deoxycholate . The lysates were spun at 13 , 000 rpm for 10 min at 4°C and layered onto a 5–50% sucrose gradient as previously described ( Gandin et al . , 2013 ) . The gradients were formed using a SW40 rotor ( Beckman , Pasadena , CA ) at 36 , 000 rpm for 2 hr at 4°C . One ml fractions were collected by upward displacement with 60% sucrose and absorbance was continuously recorded at 254 nm using ISCO fractionator ( Teledyne , ISCO ) . Collected fractions were precipitated with 10% TCA , separated by SDS-PAGE and proteins visualized by immunoblotting . For RNA analysis , 800 µl TRIZOL was added to the 1 ml fractions and RNA was isolated using standard procedures . Isolated RNA was quantified using RT-qPCR . The cDNA samples were serially diluted and the efficiency and Cq values were used to generate a standard curve ( Piques et al . , 2009 ) . One standard curve was generated for each primer pair . All standard curves had R2 value higher than 0 . 99 , with a slope between −3 . 58 and −3 . 10 . Each data point for each fraction was plotted against the standard curve to calculate the percentage of input . Biotinylated RNAs were purchased from IDT ( Coralville , IA ) . The RNAs were dissolved in binding buffer ( 10 mM Tris-acetate , pH 7 . 7 , 200 mM potassium acetate , 5 mM magnesium acetate ) , heated to 75°C for 10 min , and allowed to renature at 21°C for 5 min ( Phan et al . , 2011 ) . For the RNA binding assays , 100 nM final concentration of biotinylated RNA was incubated with cellular lysates containing 2 mg/ml heparin for 1 hr on ice . Then , 25 µl of 50% Streptavidin agarose slurry was added and incubated at 4°C for 30 min with constant end-over-end mixing . The beads were then washed 2× with cell lysis buffer with increasing salt concentration and once with PBS . The samples were then boiled with 25 µl of 2× Laemmli buffer , resolved by SDS-PAGE and visualized by immunoblotting . Fluorescein-labelled RNA was dissolved in binding buffer ( 10 mM Tris-acetate , pH 7 . 7 , 200 mM potassium acetate , 5 mM magnesium acetate ) , heated to 75°C for 10 min , and allowed to renature at 21°C for 5 min ( Phan et al . , 2011 ) . The biotinylated peptides ( 20 pmol in 50 µl ddH2O ) were allowed to bind the streptavidin-coated high-capacity binding plates ( Pierce #15 , 503 , Rockford , IL ) overnight at 4°C or 2 hr at room temperature . The peptides were removed and the plates were washed four times with binding buffer . Different concentrations of the fluoresceinated RNA ( IDT ) was allowed to bind the peptides for 1 hr at room temperature . The unbound RNAs were removed and the plates were washed four times with binding buffer . Fluorescence was measured at 521 nm on a Synergy H4 instrument ( BioTek , Winooski , VT ) . The peptides were purchased from Epicypher Inc and their sequence are Aven50-74 biotin-RRGRGRGRGFRGARGGRGGGGAPRG , termed DiRGG; Aven50-74 ( Me2a ) biotin-RRGRGRGRGFRGAR ( Me2a ) GGR ( Me2a ) GGGGAPRG , termed DiRGGme; Aven2-26: biotin-QAERGARGGRGRRPGRGRPGGDRHS , termed TriRG; Aven2-26 ( Me2a ) : biotin-QAER ( Me2a ) GAR ( Me2a ) GGR ( Me2a ) GRRPGR ( Me2a ) GRPGGDRHS , termed TriRGme . GenePrimerSequence ( 5′ → 3′ ) MLL1ForwardGAGGACCCCGGATTAAACATReverseGGAGCAAGAGGTTCAGCATCMLL4ForwardCAGACCCGGCAGACAGATGAGReverseAGATGTTACGTAGTCAAGGCACArpS6ForwardAATGGAAGGGTTATGTGGTCCGReverseCCCCTTACTCAGTAGCAGGCHOXA9ForwardTCAAAAGGATAGCGCTGCCAReverseTGCATTACCAGAGAGCCGTGMEIS1ForwardACCGTTTGCGACTTGGTACTReverseTGCTCACAACCAGACAGCTCActinForwardACCACACCTTCTACAATGAGCReverseGATAGCACAGCCTGGATAGCHOXA1ForwardACCAAGAAGCCTGTCGCTCReverseACTTTCCCTGTTTTGGGAGGGHOXC10ForwardACCACAGGAAATTGGCTGACReverseGATCCGATTCTCTCGGTTCAAvenForwardCTCTGCCTCCGACTCAACReverseCCTTGCCATCATCAGTTCTCHOXA7ForwardAAGCTGCCGGACAACAAATCReverseGAAGCCCCCGCCGTATATTTGAPDHForwardAATCCCATCACCATCTTCCAReverseTGAGTCCTTCCACGATACCA CCRF-CEM and MOLT-4 cells maintained in RPMI ( Roswell Park Memorial Institute medium ) with 10% FBS ( Fetal Bovine Serum ) were transduced with lentiviruses harboring either shRNA targeting AVEN ( 5′-CCG GGA GAA TGA TGA ACA GGG AAA TCT CGA ATT TCC CTG TTC ATC ATT CTC TTT TTT G-3′ ) , PRMT1 ( 5′-CGG GTG TTC CAG TAT CTC TGA TTA CTC GAG TAA TCA GAG ATA CTG GAA CAC TTT TTG-3′ ) , or control shRNA in the vector pLKO . 1 . The lentiviruses were generated in HEK293T cells following recommended manufacturer's protocol with modifications in transfection as follows ( 9 μg psPAX2-; 4 μg vsv-g; 9 μg shRNA ) per 10-cm plate . The shRNAs were purchased from the shRNA library from Dharmacon ( Lafayette , CO ) . Post-infection , bulk populations of stably infected cells were selected with 2 µg/ml puromycin . Small interfering RNAs ( siRNAs; Dharmacon Inc . ) were transfected using Lipofectamine RNAi MAX ( Invitrogen , Carlsbad , CA ) as per the manufacturer's protocol . The final concentration of the siRNA was 40 nM and the cells were lysed 72 hr post-transfection . The siRNA target sequence for PRMT1 was 5′-CGU CAA AGC CAA CAA GUU A-3′ . The siRNA target sequences for Aven were siAven 5′-GAG GAG AAA GAA UGG GAU AUU-3′ . For SMN , TDRD3 , and DHX36 siRNAs , SMARTpools were purchased from Dharmacon Inc . PRMT1FL/−;CreERT MEFs or HEK293T cells were transfected using Lipofectamine 2000 ( Invitrogen ) as per manufacturer's instructions . After 24 to 48 hr , the cells were lysed with cell lysis buffer ( 20 mM Tris pH 7 . 4 , 150 mM NaCl , 1 mM EDTA ( Ethylenediaminetetraacetic acid ) , 1 mM EGTA ( Ethylene glycol tetraacetic acid ) , 1% Triton X-100 ) . For immunoprecipitations , cell lysates were incubated with the primary antibody for 1 hr on ice . Then , 25 µl of 50% protein A-Sepharose slurry was added and incubated at 4°C for 45 min with constant end-over-end mixing . The beads were then washed twice with cell lysis buffer and once with 1× PBS . The samples were then boiled with 25 µl of 2× Laemmli buffer , resolved by SDS-PAGE , transferred to nitrocellulose membranes and the proteins visualized by immunoblotting . U2OS cells transfected with FLAG-Aven full length or FLAG-AvenΔRGG were lysed in lysis buffer 48 hr after transfection . Cell lysates were prepared and incubated for 1 hr at 4°C with 20 µl of 50% slurry of the purified GST-Tudor proteins bound to the glutathione agarose . 1 µg of GST protein was used for each pull-down . Following the incubation , the beads were washed three times with lysis buffer and the proteins eluted in 1× Laemmli buffer . The bound proteins were analyzed by SDS-PAGE and visualized by immunoblotting . Aven−/− cells seeded in 24-well plates were transfected with 500 ng of either pGL3 control , pGL3 pGL3-MLL4:G4 , or pGL3-MLL4:G4mutant along with 1 μg of the different FLAG-Aven constructs and 5 ng pRenilla per well . 24 hr post-transfection , the cells were lysed with passive lysis buffer and the renilla and firefly luciferase activities were measured using Dual-luciferase reporter assay kit from Promega . Aven−/− cells were transfected with either siControl , siPRMT1 , or siDHX36 . 24 hr post-transfection , the cells were seeded in 24-well plates . The following day , the cells were transfected with 500 ng of either pGL3 control , pGL3-MLL4:G4 , or pGL3-MLL4:G4mutant along with 1 μg FLAG-Aven and 5 ng pRenilla per well . 24 hr post-transfection , the cells were lysed with passive lysis buffer and the renilla and firefly luciferase activities were measured using the Dual-luciferase reporter assay kit . In-line probing assays were performed , as previously described ( Beaudoin et al . , 2013 ) . Briefly , trace amounts of labelled RNA ( 50 , 000 cpm ) were heated at 70°C for 5 min and then slow-cooled to room temperature over 1 hr in buffer containing 100 mM Tris-HCl ( pH 7 . 5 ) and 100 mM LiCl or KCl of 10 µl . Following this incubation , the final volume of each sample was adjusted to 20 µl such that the final concentrations were 50 mM Tris-HCl ( pH 7 . 5 ) , 20 mM MgCl2 , and 100 mM LiCl or KCl . The reactions were then incubated for 40 hr at room temperature , ethanol-precipitated and the RNAs dissolved in formamide dye loading buffer ( 95% formamide , 10 mM EDTA , and 0 . 025% bromophenol blue ) . The radioactivity of the in-line probing samples was measured , and equal amounts in terms of desintegrations per minute of all conditions of each candidate were fractionated on denaturing ( 8 M urea ) 10% polyacrylamide gels . The SAFA software was used to quantify each band . The intensity of the band incubated in KCl was then divided by the intensity of the corresponding band incubated with LiCl . G4 formation is confirmed when this value exceed 2 ( Beaudoin and Perreault , 2010 , 2013 ) . Histograms show the mean result and standard deviation of two separate experiments , that is , two different RNA transcription , labeling , and in-line probing . The sequences used were for wild-type MLL1 5′-GGC CGC GGC GGC GGC GGC GGG AAG CAG CGG GGC UGG GGU UCC AGG GGG AGC GGC CGC CGC CUC-3′ and for the G/A mutant 5′-GGC CGC GGC GGC GGC GGC GAG AAG CAG CGA AGC UGA AGU UCC AGA AAG AGC GGC CGC CGC CUC-3′ . The sequences used were for wild-type MLL4 5′-GGC CCG CGG GUC CAG CGG GGC CGG GGA CGG GGU CGG GGC CGG GGC UGG GGC CCG AGU CGA GGC UG-3′ and for the G/A mutant 5′-GGC CCG CGG GUC CAG CGA AGC CGA AGA CGA AGU CGA AGC CGA AGC UGA AGC CCG AGU CGA GGC UG-3′ . The following RNA sequences were purchased from IDT . sc1 ( G4 WT ) 5′-GCUGC gg UGU gg AA gg AGU gg UC gg GUUGCGCAGCG-biotin-3′; sc1 ( G4m ) 5′-GCUGC aa UGU gg AA aa AGU gg UC gg GUUGCGCAGCG-biotin-3′; MLL1 ( G4; 220–258 ) 5′-CGGC ggg AAGCAGC gggg CT gggg TTCCA gggg GAGCGG biotin-3′; MLL1 ( G4m ) 5′-CGGC aaa AAGCAGC gggg CT aaaa TTCCA gggg GAGCGG-biotin-3′; MLL4 ( G4; 267–310 ) 5′-CCAGC gggg CC gggg AC gggg UC gggg CC gggg CU gggg CCCGA-biotin-3′; MLL4 ( G4m ) 5′-CCAGC aaaa CC gggg AC gggg UC aaaa CC gggg CU gggg CCCGA-biotin-3′ . U2OS cells grown on coverslips were washed twice with 1× PBS and fixed with 4% paraformaldehyde for 10 min at room temperature . The cells were then washed twice with 1× PBS and permeabilized with 0 . 5% Triton X-100 for 10 min . Permeabilization was followed by three washes with 1× PBS and cells were blocked with 10% FBS in PBS and labeled with primary antibody diluted in PBS containing 5% FBS . After three washes , the cells were labeled with Alexa Fluor 540 conjugated secondary antibody . DNA was counterstained with 4 . 6-diamidino-2-phenylindole ( DAPI ) . The coverslips were washed three times with 1× PBS and mounted on slides using Immuno-Mount ( Thermo Scientific , Waltham , MA ) . Images were captured using a Zeiss ( Germany ) M1 microscope with epifluorescence optics . Myc-Aven expressing HEK293T cells were lyzed and IP with anti-Myc epitope tagged antibodies . The IP proteins were resolved by SDS-PAGE , visualized by Coomassie Blue ( SimplyBlue Safestain , Life Technologies ) , excised from the gel and digested with trypsin and subjected to LC/MS/MS analysis , as previously described ( Boisvert et al . , 2012 ) . U2OS cells were cultured in DMEM ( Dulbeccos's Modified Eagle Medium ) depleted of arginine and lysine , as described previously ( Boisvert et al . , 2012 ) . DMEM was supplemented with 10% of dialyzed FBS . Arginine and lysine were substituted with light ( Arg0 , Lys0 ) , medium ( Arg6 , Lys4 ) , or heavy ( Arg10 , Lys8 ) amino acids ( Cambridge Isotope Laboratories , Inc . , UK ) . The cells were cultured in the labeled medium for 6 passages for metabolic incorporation of the labeled amino acids . The light , medium , and heavy labeled cells were transfected with empty vector pcDNA3 . 1 , FLAG-Aven , and FLAG-AvenΔRGG , respectively . Cell lysates were IP with anti-FLAG agarose beads and the IP complexes were mixed prior to mass spectrometry analysis . Trypsin-digested peptides were separated using a Dionex Ultimate 3000 nanoHPLC system . 10 μl of sample ( a total of 2 μg ) in 1% ( vol/vol ) formic acid was loaded with a constant flow of 4 μl/min onto an Acclaim PepMap100 C18 column ( 0 . 3 mm id × 5 mm , Dionex Corporation , Sunnyvale , CA ) . After trap enrichment , peptides were eluted off onto a PepMap C18 nano column ( 75 μm × 50 cm , Dionex Corporation ) with a linear gradient of 5–35% solvent B ( 90% acetonitrile with 0 . 1% formic acid ) over 240 min with a constant flow of 200 nl/min . The HPLC ( High-performance liquid chromatography ) system was coupled to a QExactive mass spectrometer ( Thermo Fisher Scientific Inc ) via a EasySpray source . The spray voltage was set to 2 . 0 kV and the temperature of the column was set to 40°C . Full-scan MS survey spectra ( m/z 350–1600 ) in profile mode were acquired in the Orbitrap with a resolution of 70 , 000 after accumulation of 1 , 000 , 000 ions . The ten most intense peptide ions from the preview scan in the Orbitrap were fragmented by collision-induced dissociation ( normalized collision energy 35% and resolution of 17 , 500 ) after the accumulation of 50 , 000 ions . Maximal filling times were 250 ms for the full scans and 60 ms for the MS/MS scans . Precursor ion charge state screening was enabled and all unassigned charge states as well as singly , 7 , and 8 charged species were rejected . The dynamic exclusion list was restricted to a maximum of 500 entries with a maximum retention period of 40 s and a relative mass window of 10 ppm . The lock mass option was enabled for survey scans to improve mass accuracy . Data were acquired using the Xcalibur software . Data were processed , searched , and quantified using the MaxQuant software package version 1 . 4 . 1 . 2 as described previously ( Cox and Mann , 2008 ) employing the Human Uniprot database ( 16/07/2013 ) . The settings used for the MaxQuant analysis were 2 miscleavage were allowed; fixed modification was N-ethylmaleimide on cysteine; enzymes were Trypsin ( K/R not before P ) ; variable modifications included in the analysis were methionine oxidation and protein N-terminal acetylation . A mass tolerance of 7 ppm was used for precursor ions and a tolerance of 20 ppm was used for fragment ions . A maximum false positive rate of 1% was allowed for both peptide and protein identification . GST-AvenRGG/RG was incubated with GST-tagged PRMT1 with 0 . 55 µCi of [methyl-3H] S-adenosyl-L-methionine in the presence of 25 mM Tris-HCl ( pH 7 . 5 ) for 1 hr at 37°C in a final volume of 25 µl . Reactions were stopped by adding 25 µl of 2× Laemmli buffer followed by boiling for 10 min . The proteins were separated by SDS-PAGE , stained with Coomassie Blue , and the destained gel was soaked in EN3HANCE ( Perkin Elmer Life Sciences , Waltham , MA ) , as per manufacturer's instructions and visualized by fluorography . HEK293T were co-transfected with a pEGFP-C1 ( Clontech , Palo Alto , CA ) , a Cas9 expression vector ( Addgene , Cambridge , MA ) and expression plasmids encoding the following gRNAs; 5′-GGG GCC AGC GCG CCG GTA AGA GG-3′ and 5′-GCA GCG GCG GTA GCC AGA GGC GG-3′ targeting Aven exon 1 . The gRNAs expression plasmids were synthesized by IDT ( Coralville , IA ) , as described ( Mali et al . , 2013 ) . Single cells expressing GFP were sorted using fluorescence-activated cell sorting several days after transfection and individual clones were expanded and screened by genomic PCR and by immunoblotting . | To make a protein , the DNA sequence that encodes it is first copied to make a molecule of messenger RNA ( or mRNA for short ) . The mRNA is then used as a set of instructions to assemble a protein in a process called translation . Both DNA and RNA molecules can fold into particular shapes . One such structure is known as a G-quartet and involves the DNA or RNA folding back on itself to form a highly stable planar structure . Stacks of G-quartets can form structures known as G-quadruplexes , but little is known about the G-quadruplexes that form in mRNA molecules . Leukemia affects cells in the bone marrow and causes blood cells to develop abnormally . A protein called Aven is often found in increased amounts in certain types of leukemic cells , but it was not clear how Aven affects how leukemia develops . Thandapani , Song et al . have now found that in leukemic cells , Aven binds to G-quadruplexes found in two mRNA molecules that encode proteins that are linked to leukemia . This binding increases the translation of these mRNAs , with translation occurring most efficiently when a particular enzyme called a helicase—which remodels RNA—also bound to Aven . Reducing the amount of Aven in cells caused fewer of the leukemic proteins to be produced , which also reduced the growth and multiplcation of leukemic cells . These findings raise the possibility that drugs that disrupt how Aven works could form part of treatments for leukemia . The next challenge will be to identify the signaling pathways that communicate with Aven and to define all the G-quadruplex mRNAs that are regulated by Aven . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"biochemistry",
"and",
"chemical",
"biology",
"cell",
"biology"
] | 2015 | Aven recognition of RNA G-quadruplexes regulates translation of the mixed lineage leukemia protooncogenes |
The flat Golgi cisterna is a highly conserved feature of eukaryotic cells , but how is this morphology achieved and is it related to its function in cargo sorting and export ? A physical model of cisterna morphology led us to propose that sphingomyelin ( SM ) metabolism at the trans-Golgi membranes in mammalian cells essentially controls the structural features of a Golgi cisterna by regulating its association to curvature-generating proteins . An experimental test of this hypothesis revealed that affecting SM homeostasis converted flat cisternae into highly curled membranes with a concomitant dissociation of membrane curvature-generating proteins . These data lend support to our hypothesis that SM metabolism controls the structural organization of a Golgi cisterna . Together with our previously presented role of SM in controlling the location of proteins involved in glycosylation and vesicle formation , our data reveal the significance of SM metabolism in the structural organization and function of Golgi cisternae .
The Golgi complex plays a central role in protein processing , sorting and transport ( Emr et al . , 2009 ) . In higher eukaryotes the Golgi complex consists of multiple stacks of polarized flattened cisternae ( Klumperman , 2011 ) . Cisternae polarization allows for the directional transport and sequential processing of newly synthesized proteins arriving at the cis-face of the Golgi complex from the endoplasmic reticulum ( Glick and Luini , 2011; Stanley , 2011 ) . At the trans-Golgi network ( TGN ) , fully processed proteins are sorted and exported to other compartments of the secretory pathway or for secretion ( Guo et al . , 2014 ) . Remarkably , despite the large influx and efflux of membrane-bound transport carriers , the overall morphology of the Golgi complex remains essentially unaltered . How is the shape of the Golgi cisternae maintained and how does it relate to the function of this organelle ? Golgi cisternae are characterized by having a relatively large area-to-volume ratio to accommodate the large numbers of incoming and outgoing transport carriers and also to efficiently regulate the enzymatic reactions occurring at the Golgi membranes ( Klumperman , 2011 ) . Moreover , a Golgi cisterna consists of two geometrically distinct regions with very different membrane curvatures: the central cisterna part , which is almost flat with the seldom presence of fenestrations or pores; and the highly bent rim of the cisterna . How the different functions of the Golgi membranes ( namely , protein processing and transport ) are organized between these two regions is not yet fully understood . We previously reported that disruption of sphingomyelin ( SM ) organization specifically at the Golgi membranes –by SM synthase-mediated formation of short-chain SM at the trans-Golgi membranes– leads to inhibition of transport carrier formation ( Duran et al . , 2012 ) and also to defects in transmembrane protein glycosylation and localization ( van Galen et al . , 2014 ) . Interestingly , we showed that these effects occur concomitantly with an overall reduction in the lateral lipid order of the Golgi membranes ( Duran et al . , 2012 ) as well as with striking alterations in the morphology of Golgi cisternae , which abandon their typical flat morphology and become highly curled ( van Galen et al . , 2014 ) . Based on our findings , we suggested that short-chain SM might not be able to generate liquid-ordered nanodomains at the Golgi membranes ( Duran et al . , 2012 ) . However , it is still unclear whether there is any causal relation between the ability of SM to control lateral Golgi membrane organization and the observed changes in the morphology of the Golgi cisternae . Motivated by these experimental observations , we decided to investigate the physical mechanisms by which SM metabolism controls Golgi cisternae morphology , with a general aim at understanding whether the shape and the function of the Golgi complex are two sides of the same coin and how they relate to each other . Curling of a flat Golgi cisterna has , from a physical point of view , severe consequences . A flat cisterna has a large surface area at its rim with a very high membrane curvature thereby bearing large elastic stresses ( Shibata et al . , 2009 ) . Hence , cisterna curling is accompanied by a change in the distribution of the membrane elastic stresses . The quantitative analysis of the extent of these stresses and how they can be sustained within the overall cisterna morphology requires a physical description of the Golgi cisterna free energy . Here we present a biophysical model that describes the free energy of a Golgi cisterna to elucidate the mechanisms by which SM homeostasis mechanically regulates the shape of the Golgi complex and therefore its function . In the following , we describe the model , the results derived from it and the experimental validation of the model's predictions .
DAG is a key regulator involved in transport carrier biogenesis at the Golgi membranes ( Baron and Malhotra , 2002; Bard and Malhotra , 2006; Fernández-Ulibarri et al . , 2007; Malhotra and Campelo , 2011 ) . On the one hand , DAG is a conical lipid , which has the ability on its own to generate negative ( positive ) membrane curvature if asymmetrically enriched in the cytosolic ( luminal ) leaflet of a membrane ( Carrasco and Mérida , 2007; Leikin et al . , 1996; Szule et al . , 2002 ) . On the other hand , protein kinase D ( PKD ) , a protein that controls the fission of TGN-to-cell surface transport carriers ( Malhotra and Campelo , 2011; Campelo and Malhotra , 2012 ) , is directly recruited to the TGN by binding to DAG ( Maeda et al . , 2001; Liljedahl et al . , 2001; Baron and Malhotra , 2002 ) . It has been reported that low levels of DAG leads to defects in protein export from the TGN ( Baron and Malhotra , 2002 ) and to abnormal Golgi morphology ( Litvak et al . , 2005 ) . In contrast , local increase in DAG levels leads to the activation of PKD ( Malhotra and Campelo , 2011 ) . Active PKD phosphorylates a number of substrates at the TGN membranes , including the lipid kinase PI4KIIIβ ( Hausser et al . , 2005 ) , the lipid transport proteins CERT ( Fugmann et al . , 2007 ) , and OSBP ( Nhek et al . , 2010 ) , thus regulating the local lipid homeostasis . The theoretical results presented here ( Figure 5 ) indicate that changes in the DAG levels do not directly promote a flat-to-curled Golgi cisternae transition by redistributing along the membrane and changing its curvature . Importantly , it has been recently shown that a local burst in DAG levels at the Golgi membranes caused by increased sphingolipid metabolic flow leads to a peak of activation of PKD , which in turn , through a downstream signaling cascade , results in the consumption of PI ( 4 ) P and the consequent release of PI ( 4 ) P-binding proteins ( Capasso et al . , 2017 ) . Phosphoinositides are mostly localized away from SM-rich rigid nanodomains ( Arumugam and Bassereau , 2015 ) , thus suggesting that this signaling event occurring at the cytosolic side of the membrane is spatially uncoupled from SM-rich rigid nanodomains in the lumenal leaflet . Based on these results and on our results showing that clathrin coats are released from the Golgi membranes prior to short-chain ceramide-mediated flat-to-curled Golgi cisterna transition , we propose that SM metabolism , through the by-product DAG , indirectly controls Golgi morphology by means of a downstream PKD-dependent signaling cascade rather than by playing a direct mechanical part in membrane bending . One of the main results of our model is that the release of membrane curvature-generating proteins leads to the destabilization of the flat Golgi cisterna configuration , triggering a morphological transition towards a curled configuration . Being the central hub of the secretory pathway , the Golgi complex recruits a number of different curvature-generating proteins to efficiently sustain transport carrier formation ( Cruz-Garcia et al . , 2013; Bonazzi et al . , 2005; Campelo and Malhotra , 2012 ) . Our results suggest that the role of such proteins is twofold . First , they induce membrane curvature to accommodate the secretory cargoes into nascent budding carriers prior to their fission from the Golgi membranes . Second , this dynamic series of budding and scission events serves to stabilize the highly bent rims of the Golgi cisternae . As such , we propose that the shape and function of Golgi membranes are maintained by membrane curvature generators via a positive feedback loop where the highly bent Golgi rims provide optimal nucleation sites for the budding of transport carriers . At the same time , the machinery involved in this process maintains and stabilizes a functionally optimal Golgi cisternae morphology . Does the presence of clathrin coats on the Golgi membranes represent the main driving force for flat cisternae stabilization ? Or are there other curvature-inducing proteins involved ? Altered SM metabolic flow at the Golgi membranes did not affect the localization of COPI components ( Capasso et al . , 2017 ) to those membranes . However , other curvature generators might be released in addition to clathrin due to the defects in SM metabolism . It has been recently reported that knockdown of the two Golgi-localized PI ( 4 ) kinases in Atg5 knockout cells induces curling of the Golgi cisternae ( Yamaguchi et al . , 2016 ) . Moreover , it has been shown that components of the COPI machinery are released from the Golgi membranes in HeLa cells incubated at 15°C , a situation that parallels curling of the Golgi cisternae towards the cis side of the stack ( Martínez-Alonso et al . , 2005 ) . Altogether , we propose that maintenance of the flat Golgi cisternae morphology requires a combined effort of different classes of curvature generating proteins and that the release of a subset of these proteins can lead to the destabilization of the flat configuration . We suggest that the breaking of the stack symmetry upon cisternae curling is driven by the initial release of rim stabilizers from a few cisternae ( trans-cisternae in our experiments , cis-cisternae in [Martínez-Alonso et al . , 2005] ) , which is then followed by the other cisternae . From a theoretical perspective , the effect of including multiple cisternae with different levels of budding effectors is analyzed and discussed in the Appendix . In brief , our results indicate that Golgi curling must parallel some level of release of curvature generators from all Golgi cisternae . Further experiments are needed to test this . Our model predicts the existence of shape bistability within a certain range of values of the membrane spontaneous curvature and of the membrane area fraction covered by rigid nanodomains ( Figure 2A ) . Within this region of the parameter space , both flat and curled cisterna configurations correspond to locally stable shapes ( Figure 2B ) . This means that the system can be kinetically trapped in a metastable configuration , which corresponds to a local but not global minimum of the cisterna free energy before it relaxes to the globally stable configuration . Such transition from a metastable to a stable configuration needs to overcome an energy barrier . If the value of the energy barrier is relatively small , transition to the stable configuration can be overcome by thermal fluctuations of the Golgi cisterna shape . Such thermally-triggered transitions follow Arrhenius kinetics , according to which an average transition time , τ , can be estimated as τ=t0eΔF/kBT , where t0 is a characteristic time scale of Golgi cisterna fluctuations and ΔF is the height of the energy barrier ( Hänggi et al . , 1990; Morlot et al . , 2012 ) . As mentioned above , we can estimate t0≈1 ms from hydrodynamic arguments as t0=ηR3/κ ( Allain et al . , 2004 ) , where η is the cytosol viscosity , R is a typical length scale of a Golgi cisterna , and κ is the bending rigidity of the membrane . We can qualitatively compare this to our experimental results on the curled-to-flat Golgi cisterna transition during short-chain ceramide washout ( Figure 6 ) . Those results indicated that 14 hr after the recovery of normal clathrin levels at the Golgi membranes , about 50% of Golgi stacks were composed of flat cisternae ( Figure 6D ) . According to the aforementioned kinetics , such shape transition time corresponds to an energy barrier for the curled-to-flat transition of ΔF≈18 kBT . If we compare this estimation for the curled-to-flat transition to the numerical results of our model ( Figure 2A ) , we find that the spontaneous curvature at the Golgi cisterna rims in the fully recovered state should be of the order of Js≈0 . 0225 nm−1 . Following the relationship between the membrane spontaneous curvature and the area fraction covered by the curvature generators ( Campelo et al . , 2008 ) , the predicted cisterna spontaneous curvature corresponds to a membrane area fraction covered by curvature generators of 5−10% , which is a physiologically reasonable estimation . Moreover , we showed that the time of recovery of Golgi protein localization after short-chain ceramide treatment depends on the duration of the treatment ( Figure 6E ) . These different recovery times , we suggest , can be explained by the fact that the longer the treatment with short-chain ceramide , the lower the levels of long-chain SM at the Golgi membranes . Hence , a longer time would be required for the Golgi membranes to recover their normal levels of SM-rich nanodomains and , according to our hypothesis , of Golgi protein recovery . Likewise , our model predicts that the Golgi cisterna shape transition is , in thermodynamic terms , a first-order transition because the transition is discontinuous in the shape parameter ( in our case , the distance between the center of the cisterna rim and the axis of symmetry , rgap ) ( Figure 3D ) . This indicates that once the transition from a flat cisterna ( rgap=500 nm ) to a highly curled cisterna ( 40 nm < rgap < 100 nm ) is triggered , the transition is abrupt because no cisternae of intermediate curling correspond to a locally stable configuration ( Figure 3D ) . Although it is hard to extract quantitative information of the curled cisterna gap opening size , rgap , from the ultrathin sections ( Figure 6C ) to compare with the theoretical predictions , our ultrastructural analysis of the Golgi morphology qualitatively showed that Golgi curling indeed occurs in an abrupt manner ( Figure 6C ) . In summary , we have presented a theoretical biophysical model of Golgi cisterna morphology , which describes the existence of stable flat and curled Golgi cisternae for different values of the membrane spontaneous curvature . We experimentally validated some of the model's predictions . In particular , our model helps explaining the mechanisms by which a reversible flat-to-curled Golgi cisternae transition is induced upon disruption of SM homeostasis by short-chain ceramide treatment . Flat Golgi cisternae in untreated HeLa cells have stationary levels of different curvature-inducing proteins , such as components of the clathrin-coated vesicle machinery ( Figure 7 , top left cartoon ) . Moreover , a certain amount of small , dynamic , SM-enriched rigid nanodomains might be present in the membrane , and slightly enriched in the central flat part of the cisterna ( Figure 7 , top left cartoon ) . Treatment of cells with D-cer-C6 has a twofold effect on the Golgi membrane properties: it causes the release of clathrin from the membranes ( Figure 6A , B ) and decreases the lateral order of the Golgi membranes ( Duran et al . , 2012 ) ( Figure 7 , top right cartoon ) . The results of our model show that the decrease in membrane spontaneous curvature ( through the release of curvature generating proteins such as clathrin ) but not a reduction in the number of rigid nanodomains alters the cisterna free energy profile to a situation where the flat cisterna configuration is unstable and hence a fast flat-to-curled cisterna transition occurs ( Figure 7 , right ) . Short-chain ceramide washout leads to the recruitment of clathrin back to the Golgi membranes ( Figure 6A , B ) and , we suggest , also leads to the recovery of the initial levels of SM-enriched rigid nanodomains ( Figure 7 , bottom left ) . Under these conditions the system free energy profile presents shape bistability , so the Golgi cisternae are kinetically trapped in the curled configuration . Hence , the curled-to-flat cisterna transition is slow because it requires the energy barrier to be overcome by thermal fluctuations ( Figure 7 , left ) . Overall , the model presented in here together with some of its experimental validation underscore the crucial role of SM metabolism in regulating the structural morphology and function of the Golgi cisternae . We foresee that future experimental work along these lines will strengthen our predictions and will help to understand better the different factors governing the shape and function of the Golgi complex .
N-hexanoyl-D-erythro-sphingosine ( D-cer-C6 ) was obtained from Matreya and dissolved in pure ethanol ( Merck ) to a 10 mM stock solution . Sheep anti–human TGN46 was obtained from AbD Serotec ( Bio-Rad / AbD Serotec Cat# AHP500 , RRID:AB_324049 ) . Mouse anti-p230 was obtained from BD ( BD Biosciences Cat# 611280 , RRID:AB_398808 ) . Goat anti-Clathrin-HC antibody was from Santa Cruz ( Santa Cruz Biotechnology Cat# sc-6579 , RRID:AB_2083170 ) . Alexa Fluor–labeled secondary antibodies were obtained from Invitrogen . HeLa cells , obtained from ATCC , were cultured in DMEM ( Lonza ) containing 10% FCS . HeLa cells stably expressing the plasmid encoding the first 100 amino acids of rat mannosidase-II in the pEGFP-N1 vector ( HeLa-MannII-GFP cells ) were described previously ( Sütterlin et al . , 2005; van Galen et al . , 2014 ) . All cell lines were periodically checked for mycoplasma contamination . For clathrin-HC immunostaining , samples were fixed and permeabilized in methanol for 6 min at −20°C . For p230 immunostaining , samples were fixed with 4% formaldehyde in PBS for 20 min and permeabilized with 0 . 2% Triton X-100 in PBS for 30 min . Fixed cells were then blocked in 2% BSA in PBS for 30 min before antibody staining . Cells were then sequentially incubated for 1 hr at room temperature first with primary and then with secondary antibodies diluted in blocking buffer . Samples were analyzed with a confocal system ( TCS SP5 II CW STED; Leica ) in confocal mode using a 100x , 1 . 4 NA objective and HyD detectors ( Leica ) . Alexa Fluor 488– , 568- , 594-conjugated secondary antibodies were used . Images were acquired using the Leica software and converted to TIFF files using ImageJ ( version 1 . 43; National Institutes of Health ) . Two-channel colocalization analysis was performed using ImageJ , and the Pearson’s correlation coefficient was calculated using the Manders’ coefficients plugin developed at the Wright Cell Imaging Facility ( Toronto , Ontario , Canada ) . The samples were fixed and prepared using standard methods , essentially as described previously ( Rizzo et al . , 2013 ) . In brief , the cells were fixed with 2% paraformaldehyde and 0 . 2% gluteraldehyde in PBS , for 2 hr at room temperature . The cells were then washed with PBS/0 . 02 M glycine , scraped in 12% gelatin in PBS , and then embedded in the same solution . The cells embedded in gelatin were cut in 1 mm blocks and infiltrated with 2 . 3 M sucrose at 4°C , mounted on aluminum pins , and frozen in liquid nitrogen . The samples were then sectioned and the ultrathin cryosections were picked up in a mixture of 50% sucrose and 50% methylcellulose and incubated with antibodies to antigen of interest ( anti-GFP and anti-p230 ) followed by protein A gold . The samples were observed in the FEI Tecnai-12 electron microscope . In this section we formulate in mathematical terms the physical model we used to describe how Golgi cisterna morphology is controlled by variations in SM homeostasis and metabolism . In the two following subsections we describe , respectively , ( i ) the system geometry , that is , the possible geometrical configurations of the Golgi cisterna; and ( ii ) the system free energy . The derivation of total bending energy of a Golgi cisterna is detailed here , taking separately the contributions from the central part of the cisterna and of the cisterna rim . In the Appendix we also present simplified analytical expressions for the free energy of the cisterna , obtained under certain approximations . To obtain the local stable shapes of Golgi cisternae as a function of the spontaneous curvature and of the total area fraction of nanodomains , our strategy is to compute the membrane free energy Equation ( 13 ) as a function of the gap aperture , rgap , and the nanodomain area fraction at the cisternae rim , Φrim . Then , for all values of the gap aperture , rrim < rgap≤rflat , we found the optimal distribution of nanodomains at the cisternae rim , Φrim∗ ( rgap ) , by minimization of the free energy with respect to this parameter , ( 20 ) F ( rgap , Φrim∗rgap ) =minΦrim∈[0 , 1]{F ( rgap , Φrim ( rgap ) ) } , Equation ( 20 ) represents the partitioning-optimized free energy as a function of the gap aperture length ( that is , as a function of the cisterna morphology ) . Depending on the parameter values , the free energy of the cisterna Equation ( 20 ) has one or two local minima , which correspond to curled or flat cisternae morphologies . Moreover , to compute the energy barriers , we used that Fmax=maxrgap∈[rcurl , rflat][F ( rgap , Φrim∗ ( rgap ) ) ] , Fcurl=F ( rcurl , Φrim∗ ( rcurl ) ) , and Fflat=F ( rflat , Φrim∗ ( rflat ) ) . The free energy of a Golgi cisterna , including the bending energy term taking into account a different distribution of DAG molecules between the top , bottom and rim regions of both the luminal and cytosolic monolayers ( see Figure 5A ) , as well as the associated entropic free energy penalty of a non-homogeneous distribution of DAG molecules reads as , ( 21 ) F=κ2∑i∈{top , bottom , rim}∫ ( Js−12 ( ϕDAG , cyti−ϕDAG , lumi ) ζDAG−Js , i ) 2dAi+kBTaDAG∑j∈{lum , cyt}∑i∈{top , bottom , rim}[ϕDAG , jilnϕDAG , ji+ ( 1−ϕDAG , ji ) ln ( 1−ϕDAG , ji ) ]Ai , where Js , i are the spontaneous curvatures of the top , bottom , and rim bilayers ( index i ) , aDAG≈0 . 6 nm2 is the area per DAG molecule ( Shemesh et al . , 2003 ) , and ϕDAG , ji is the DAG area fraction in the cytosolic or luminal monolayers ( index j ) of the top , bottom , and rim bilayers ( index i ) . Considering that the total amount of DAG is symmetrically distributed between the luminal and cytosolic monolayers of the Golgi membrane , these quantities are related to the total membrane DAG area fraction , ϕDAG , as ( 22 ) ∑i∈{ top , bottom , rim }ϕDAG , lumiAi=∑i∈{ top , bottom , rim }ϕDAG , cytiAi=ϕDAGA . | The Golgi complex is a hub inside cells that transports many proteins to various parts of the cell . It also receives freshly made proteins and modifies them to help them mature into their final active forms . The complex is made up of a stack of disc-like membrane structures called cisternae . Are the shapes of the cisternae important for the Golgi complex to work properly ? Membranes are made of mixtures of molecules known as lipids and proteins . Previous experiments show that when the mixture of lipids in the Golgi membranes changes in a specific manner , the cisternae curl into an onion-like shape and the Golgi cannot process or send out proteins anymore . Campelo et al . used mathematics and experimental approaches to investigate what causes the Golgi to change shape when the lipid mixture of the cisternae changes . A mathematical description of the shape of the Golgi predicted that some proteins keep the cisternae flat by holding the membrane rim that connects the two faces of a cisterna . To test this prediction , Campelo et al . performed experiments in human cells , which showed that when the mixture of lipids in the Golgi membranes changes , certain proteins jump from the rim , causing the cisternae to curl . These same proteins are also needed to transport cargo proteins out of the Golgi , meaning that there is a connection between the shape of the Golgi and the tasks it carries out . The shape of the Golgi complex is altered in Alzheimer’s disease and many other neurodegenerative diseases . The next challenges are to understand how these shape changes happen , how this affects cells , and if it could be possible to develop drugs that prevent these changes from occurring in patients . | [
"Abstract",
"Introduction",
"Discussion",
"Materials",
"and",
"methods"
] | [
"cell",
"biology"
] | 2017 | Sphingomyelin metabolism controls the shape and function of the Golgi cisternae |
Voltage-dependent Ca2+ channels ( Cavs ) are indispensable for coupling action potentials with Ca2+ signaling in living organisms . The structure of Cavs is similar to that of voltage-dependent Na+ channels ( Navs ) . It is known that prokaryotic Navs can obtain Ca2+ selectivity by negative charge mutations of the selectivity filter , but native prokaryotic Cavs had not yet been identified . We report the first identification of a native prokaryotic Cav , CavMr , whose selectivity filter contains a smaller number of negatively charged residues than that of artificial prokaryotic Cavs . A relative mutant whose selectivity filter was replaced with that of CavMr exhibits high Ca2+ selectivity . Mutational analyses revealed that the glycine residue of the CavMr selectivity filter is a determinant for Ca2+ selectivity . This glycine residue is well conserved among subdomains I and III of eukaryotic Cavs . These findings provide new insight into the Ca2+ selectivity mechanism that is conserved from prokaryotes to eukaryotes .
Voltage-dependent Ca2+ channels ( Cavs ) , which couple the membrane voltage with Ca2+ signaling , regulate some important physiological functions , such as neurotransmission and muscle contraction ( Hille , 2001 ) . The channel subunits of both mammalian Cavs and mammalian voltage-dependent Na+ channels ( Navs ) have 24 transmembrane helices ( 24TM ) ( Catterall , 2000 ) , and comprise four homologous subdomains with six transmembrane helices that correspond to one subunit of homo-tetrameric channels , such as voltage-dependent K+ channels and prokaryotic Navs ( BacNavs ) . Comparison of the sequences of Navs and Cavs indicate that Navs derived from Cavs . Their two pairs of subdomains , domains I and III , and domains II and IV , are evolutionarily close to each other ( Rahman et al . , 2014; Strong et al . , 1993 ) . Therefore , the 24TM-type of Cavs are thought to have evolved from the single-domain type of Cavs with two domain-duplication events . Although prokaryotes are expected to have such ancestor-like channels , native prokaryotic Cavs have not yet been identified . The lack of ancestor-like prokaryotic Cavs is a missing link in the evolution of voltage-dependent cation channels . In contrast to the lack of prokaryotic Cavs , many BacNavs have been characterized ( Irie et al . , 2010; Ito et al . , 2004; Koishi et al . , 2004; Lee et al . , 2012; Nagura et al . , 2010; Payandeh et al . , 2011; Ren et al . , 2001; Shimomura et al . , 2016; Shimomura et al . , 2011 ) . The simple structure of BacNavs has helped to elucidate the molecular mechanisms of Navs ( Irie et al . , 2018; Irie et al . , 2012; Tsai et al . , 2013; Yue et al . , 2002 ) . In addition , BacNavs have been used as a genetic tool for manipulating neuronal excitability in vivo ( Bando et al . , 2016; Kamiya et al . , 2019; Lin et al . , 2010 ) . The acquisition of Ca2+ selectivity by BacNavs can be engineered by the introduction of several negatively charged amino acids into the selectivity filter ( Tang et al . , 2014; Yue et al . , 2002 ) . A mutant channel NavAb ( a BacNav from Arcobacter butzleri ) produced in this way , showed high Ca2+ selectivity , and the structural basis of Ca2+ selectivity has been discussed on the basis of its crystal structures ( Tang et al . , 2016; Tang et al . , 2014 ) . However , the selectivity filter sequences of CavAb , which were made by mutation of NavAb and contain a large number of aspartates , are quite different from those of the original mammalian Cavs . The evolutional analysis also indicated that BacNavs acquired sodium selectivity independent from that of 24TM Navs ( Liebeskind et al . , 2013 ) . From these points of view , ancestor-like prokaryotic Cavs could be expected to help us to understand the structural and functional relationship between BacNavs and 24TM channels . Here , we newly characterized two BacNav homologs , CavMr from Meiothermus ruber and NavPp from Plesiocystis pacifica . These two channels are evolutionarily distant from the previously reported canonical BacNavs . We confirmed that CavMr has clear Ca2+ selectivity , and that NavPp has Na+ selectivity with Ca2+-dependent inhibition . The discovery of these channels suggests the possible importance of voltage-regulated Ca2+ signaling in prokaryotes and may be a new genetic tool for controlling Ca2+ signaling . Furthermore , mutational analyses indicate that the glycine residue of the CavMr selectivity filter is important for Ca2+ selectivity . This glycine residue is also well conserved in the selectivity filter of subdomains I and III of mammalian Cavs . On the basis of these observations , we propose that CavMr is an ancestor-like prokaryotic Cav with a Ca2+ selectivity mechanism that is different from that in artificial CavAb . Further phylogenetical analyses indicated that CavMr and NavPp homologs form a wide-spread group in prokaryote and archaea , which is different from canonical BacNavs . Therefore , they are expected to advance our understanding of Ca2+ recognition and the evolution of voltage-dependent cation channels .
We searched for the primary sequences of candidate prokaryotic Cavs in the GenBank database . In mammalian and prokaryotic Navs and Cavs , a larger number of negative charges in the filter increases Ca2+ selectivity ( Figure 1a ) ( Heinemann et al . , 1992; Tang et al . , 2014; Yue et al . , 2002 ) . Several BLAST search rounds using the pore regions ( S5–S6 ) of NaChBac ( or NavBh; a BacNav from Bacillus halodurans ) as templates revealed a series of candidate prokaryotic Cavs whose selectivity filters are similar to the ‘TLESW’ motif , but which contain more negatively charged residues like the filter sequence of CavAb ( Figure 1a ) . Phylogenetic analysis of these channel genes revealed that they apparently belong to a different branch of the tree than that of canonical BacNavs , namely a Bacillus group and a NavAb-like group ( Figure 1b; Figure 1—source data 1 ) . The selectivity filter sequences of these channels are similar to that of the ancestral BacNav channel predicted previously ( Liebeskind et al . , 2013 ) . Therefore , we named these channels ancestor-like BacNavs ( AnclNavs ) . AnclNavs are widely distributed in multiple bacterial phyla and even in archaea ( Figure 1—figure supplement 1; Figure 1—source data 1 ) . In some cases , one phylum , such as proteobacteria , contains both a NavAb-like BacNav and an AnclNav gene . Even in those cases , the NavAb-like BacNav and the AnclNav were included separately in their respective groups . The bacillus BacNav group is a different group located phylogenetically between the NavAb-like and the AnclNav groups . In addition , the firmicutes phylum , which includes Bacillus species , contains neither homologs of NavAb-like BacNavs nor homologs of AnclNavs . These observations suggest that our identified candidate prokaryotic Cavs , AnclNavs , are homologs rather than orthologs of canonical BacNavs and compose a distinct group . In addition , analyses that include some eukaryotic channels , such as each subdomain of 24TM-type of Navs/Cavs , CatSper and EukCatA , a group of eukaryotic non-selective homotetrameric channels , put AnclNavs closest to a Bacillus group ( Figure 1—figure supplement 2; Figure 1—figure supplement 2—source data 1 ) . We identified four AnclNavs genes and measured their channel activity: ZP_04038264 from M . ruber , ZP_01909854 from P . pacifica , YP 003896792_from Halomonas elongata , and YP_003073405 from Teredinibacter turnerae ( Figure 1b and c ) . When attempting to express prokaryotic channels transgenetically , insect cells are often better than mammalian cells for generating large current amplitudes ( Irie et al . , 2018 ) . We therefore transfected Sf9 cells with these four channel genes and measured the resulting whole-cell currents . The cells that were transfected with genes from M . ruber showed currents in response to a depolarizing stimulus from a −140 mV holding potential ( Figure 2a ) . To estimate the Ca2+ permeability , we measured their current-voltage relationships . The M . ruber channel clearly had larger currents in the high-Ca2+ solution than in the high-Na+ solution , and no obvious outward current was observed in a high-Ca2+ bath solution , even at very high membrane potential ( 100 mV ) ( Figure 2a and b; Figure 2—source data 1 ) . These current-voltage relationships suggest that the M . ruber channel has a preference for Ca2+ , and that other cations inside the cells ( sodium and cesium ) hardly permeated the activated channel . Therefore , the newly identified channel from M . ruber is abbreviated as CavMr , based on its ion selectivity and species name . We evaluated the voltage-dependent activation of CavMr by measuring deactivation tail currents ( Figure 2c ) . A Boltzmann fit of the averaged activation curve yielded an activating potential of 50% activation ( V1/2 ) of −51 . 7 ± 1 . 1 mV ( Figure 2d; Figure 2—source data 2 ) . To compare clearly the positions of the residues in the selectivity filter in each channel , we renumbered the seven residues comprising the selectivity filter . For example , the seven residues of the CavMr selectivity filter are 183-TLEGWVD-189 , and thus Thr183 and Asp189 were renumbered as Thr1 and Asp7 ( Figure 1a ) . Notably , the amino acid sequence of the selectivity filter in CavMr is similar to the conserved features of domains I/III in mammalian Cavs , with a glycine at position 4 and a polar or negatively charged residue at position 7 , which are not observed in the canonical BacNav family . In addition , the CavMr selectivity filter sequence is quite similar to that of the human Cav subdomain I , or even the same as Cav3 . 1 and 3 . 2 ( Figure 1a ) . In the following experiments , to evaluate the reversal potential for the ion selectivity analysis accurately , we introduced a single mutation that resulted in large and long-lasting channel currents . T220A and G229A mutations in NaChBac led to slower inactivation and provided a larger current , indicating suppression of the transition to the inactivated state ( Irie et al . , 2010; Shimomura et al . , 2016 ) . We introduced a G240A mutation to CavMr , corresponding to the NaChBac mutations of G229A . Th is mutant channels stably showed larger and more measurable currents than the wild-type channel , even after they were administrated multiple depolarizing stimuli ( Figure 2e and f ) . We accurately quantified the selectivity of CavMr for Na+ and Ca2+ ( PCa/PNa ) by measuring the reversal potential ( Erev ) under bi-ionic conditions , in which the Ca2+ concentration in the bath solution was changed to 4 , 10 , 20 , and 40 mM while the intracellular Na+ concentration was held constant at 150 mM ( Figure 3a and b; Figure 3—figure supplement 1a–d; Figure 3—source data 1 ) . The plot of the reversal potentials as a function of [Ca2+]out had a slope of 39 . 89 ± 3 . 31 mV/decade ( n = 4 ) . It was higher than the Nernst prediction for a divalent cation ( 29 mV ) . We think that this deviation came from incorporation of the measurements made with the 40 mM Ca2+ bath solution , in which no obvious outward current was observed ( Figure 3—figure supplement 1d ) , resulting in a higher value of Erev . The well-fitted plot of the reversal potentials in 4 , 10 and 20 mM Ca2+ bath solutions as a function of [Ca2+]outdid indeed have a slope of 32 . 38 ± 4 . 10 mV/decade ( n = 4 ) ( Figure 3b dashed line ) , a value close to the Nernst equation for Ca2+ ( Figure 3b ) , and indicated that CavMr had a PCa/PNa of 218 ± 38 ( Figure 3e , Table 1; Figure 3—source data 2 ) . This high PCa/PNa value is comparable to that of CavAb . Among several species of cations , including Sr2+ , K+ , and Cs+ , Ca2+ had the highest permeability relative to Na+ ( Figure 3c , d and f , Table 1; Figure 3—source data 3 ) . On the basis of these results , CavMr was confirmed to be a native prokaryotic Cav with high Ca2+ selectivity . We also investigated whether CavMr shows the typical anomalous mole fraction effect ( Almers and McCleskey , 1984 ) and the non-monotonic mole fraction effect observed in NaChBac ( Finol-Urdaneta et al . , 2014 ) . CavMr did not allow Na+ permeation under Ca2+-free ( 0 mM CaCl2 and 1 mM EGTA ) conditions ( Figure 4a and b; Figure 4—source data 1 ) . Also , in contrast to the recording of NaChBac currents in a solution containing Na+ and K+ , CavMr had an apparently monotonic current increase depending on the Ca2+ mole fraction to Na+ ( Figure 4c and d; Figure 4—source data 2 ) . Studies of an artificial Cav , CavAb , revealed that Ca2+ selectivity depends on the presence of a large number of aspartates in the filter sequence ( Tang et al . , 2014 ) . The high Ca2+ selectivity in CavMr was unexpected because the filter sequence contained only one aspartate residue ( Figure 1c ) . Furthermore , CavMr-D7M , which has only one negatively charged residue in the selectivity filter ‘TLEGWVM’ , still had high Ca2+ selectivity , comparable to that of wild-type CavMr ( PCa/PNa = 144 ± 12; Figure 4e-g and Table 1; Figure 4—source data 3 ) . These findings indicate that CavMr and artificial CavAb have different Ca2+-selection mechanisms . The currents derived from the P . pacifica channel became large with increases in the bath Na+ concentration and significantly decreased when the Na+ solution was replaced with a high Ca2+ solution ( Figure 5a and b; Figure 5—source data 1 ) . Because the reversal potential fit well to the Na+ equilibrium potential in the high-Na+ solution ( Figure 5b ) , we abbreviated this channel as NavPp on the basis of its ion selectivity and species name . Interestingly , NavPp , despite having one more aspartate in the selectivity filter than CavMr , exhibited larger currents in Na+ solutions than in Ca2+ solutions ( Figure 1c and Figure 5a ) . This observation indicates that Ca2+ selection is not achieved simply by increasing negative charges in the filter sequence of the AnclNav group , which again suggests the existence of an alternative ion-selectivity mechanism . Recordings in bath solution containing both Na+ and Ca2+ demonstrated that the increment of the extracellular Ca2+ decreased the current in NavPp and led to a positive shift in the voltage dependence , suggesting that a higher concentration of Ca2+ inhibited the gating and ionic permeation of NavPp ( Figure 5b ) . We tried to measure the voltage-dependent activation of NavPp , but the wild-type channel showed very fast deactivation and no tail current was observed ( Figure 5c ) . By introducing a T232A mutation , corresponding to the NaChBac mutations of T220A ( Shimomura et al . , 2016 ) , we were able to observe the tail current at −60 mV ( Figure 5d and e ) . A Boltzmann fit of the averaged activation curve of NavPp T232A yielded a V1/2 of −17 . 11 ± 1 . 8 mV ( Figure 5f; Figure 5—source data 2 ) . To characterize the effect of extracellular Ca2+ on the NavPp channel , the voltage dependence of activation of NavPp was measured under various bath Ca2+ concentrations ( Figure 6a–c and Figure 6—figure supplement 1a; Figure 6—source data 1 and Figure 6—source data 2 ) . The increments of the extracellular Ca2+ raised the value of the reversal potential , indicating that extracellular calcium ions can also permeate NavPp ( Figure 6b ) . However , in higher extracellular Ca2+ concentrations , the current amplitude of NavPp became smaller , even in the voltage at which NavPp opens fully , and the voltage dependence of the activation shifted more positively ( Figure 6b–c ) . These results indicated that calcium ions can permeate NavPp but disturb the gating and ionic permeation of NavPp . We then compared the relative permeability of various cations with that of Na+ in NavPp . The reversal potential was obtained under an extracellular solution containing Na+ ions , despite a partial Ca2+- or Sr2+-induced block ( Figure 6—figure supplement 1b and c ) . The selectivity of NavPp was higher for Na+ than for Ca2+ , Sr2+ , K+ , or Cs+ ( Figure 6d and e; Figure 6—source data 3 ) . The PCa/PNa was 0 . 308 ± 0 . 028 in a bath solution containing both Ca2+ and Na+ , suggesting that a larger fraction of Ca2+ is allowed to permeate with outside Na+ ions through NavPp than through canonical BacNavs . Similar to Ca2+ , Sr2+ also blocked the NavPp current , but may also permeate the channel along with Na+ ions ( Figure 6—figure supplement 1c ) . These findings demonstrate a unique feature of NavPp , a low-affinity Ca2+ block , which is not reported in canonical BacNavs . Interestingly , the filter sequence of NavPp , ‘TLEDWTD’ , has three negatively charged residues , similar to the filter sequences of the artificial Ca2+-selective BacNav mutants ( the ‘TLEDWSD’ mutant of NavAb and the 'TLEDWAD’ mutant of NaChBac ) ( Tang et al . , 2014; Yue et al . , 2002 ) . NavPp does not show high Ca2+ permeability , but rather a Ca2+ block . We also investigated NavPp mutants that have the same filter sequences as the artificial Cavs . NavPp-T6S ‘TLEDWSD’ exhibited Ca2+-blocked currents similar to those exhibited by wild-type NavPp ( Figure 6e: upper ) . Further , NavPp-T6A ‘TLEDWAD’ showed no inward current in bath solutions containing divalent cations , suggesting that the Ca2+-induced block was enforced ( Figure 6e: bottom ) . Therefore , both of the selectivity filter sequences that provide Ca2+ selectivity to canonical BacNavs failed to generate Ca2+-permeable NavPp , indicating that the cation-permeable mechanism of NavPp differs from that of canonical BacNavs , as well as that of CavMr . On the other hand , the cells transfected with genes from H . elongata and T . turnerae failed to show any detectable currents , while these genes code selectivity filter sequences that are similar to that of NavPp ( Figure 1c ) . To search for the determinants of Ca2+ selectivity in CavMr , we investigated a series of mutants in which the filter regions were swapped between CavMr and NavPp ( Figure 7a and b ) , which exhibited channel activity ( Figure 7—figure supplements 1–2 ) . A NavPp mutant whose selectivity filter was replaced with that of CavMr , named NavPp-Mr , exhibited much higher Ca2+ selectivity ( PCa/PNa = 215 ± 33 ) as well as high Sr2+ selectivity , comparable to that of CavMr ( Figure 7c; Figure 7—source data 1 ) . In addition , NavPp-Mr excluded Na+ and K+ similar to CavMr , but weakly allowed Cs+ permeation in contrast to CavMr . On the other hand , a CavMr mutant whose selectivity filter was replaced with that of NavPp ( CavMr-Pp ) almost lost its Ca2+ selectivity ( PCa/PNa = 13 . 8 ± 2 . 0 ) , and was less able to discriminate Cs+ and K+ from Na+ ( Figure 7d; Figure 7—source data 2 ) . That is , CavMr-Pp was a more non-selective channel than the wild-type CavMr , rather than a Na+-selective channel . The Ca2+ selectivity ( from NavPp to CavMr ) was almost transferable , but the Na+ selectivity was not . We also investigated the full swapping of the filter sequences between CavMr and NavAb ( Figure 7a ) , but neither swapped mutants of CavMr nor NavAb had detectable currents . This finding suggested that CavMr and NavAb achieve cation selectivity using different structural backbones and mechanisms . Positions 4 and/or 6 of the filter sequences are thought to be important for Ca2+-selective permeation through NavPp-Mr and CavMr , because only these two positions were mutated in the swapping experiments . We investigated which of the mutations in positions 4 and 6 had greater effects on the loss and acquisition of Ca2+ selectivity in CavMr and NavPp , respectively . In CavMr , two single mutants , CavMr-G4D and CavMr-V6T , both decreased Ca2+ selectivity and allowed K+ and Cs+ permeation ( Figure 8a;Table 1; Figure 8—figure supplements 1–2; Figure 8—source data 1 ) . The mutational effect was greater in CavMr-G4D , whose PCa/PNa was less than 10 ( 7 . 73 ± 2 . 24 ) . CavMr-G4S , in which Gly4 was replaced with the Ser4 of NavAb , also exhibited lower Ca2+ selectivity ( PCa/PNa = 11 . 9 ± 1 . 5 ) and was also K+ and Cs+ permeable , indicating that a minor substitution by serine allowed the channel to retain a little calcium selectivity , but the monovalent cation selectivity had completely disappeared ( Figure 8b; Table 1; Figure 8—figure supplement 3; Figure 8—source data 2 ) . In the case of NavPp , NavPp-D4G acquired divalent cation , Ca2+ and Sr2+ , selectivity over Na+ , and also showed a greater PK/PNa and PCs/PNa than wild-type NavPp ( Figure 8c; Table 1; Figure 8—figure supplements 4–5; Figure 8—source data 3 ) . By contrast , NavPp-T6V failed to acquire the high Ca2+ selectivity ( PCa/PNa = 1 . 72 ± 1 . 09 ) and also allowed K+ and Cs+ permeation , while it had relatively high Sr2+ selectivity . These results indicate that , in both CavMr and NavPp , a glycine residue at position 4 is a key determinant for Ca2+ selectivity . It is noteworthy that the glycine is a conserved residue at position 4 of subdomains I and III in all subtypes of mammalian Cavs ( Figure 1b ) .
In this study , we newly characterized two prokaryotic voltage-dependent cation channels , CavMr and NavPp . CavMr is the first native prokaryotic Cavs reported , and NavPp could be inhibited by high concentrations of extracellular Ca2+ . The PCa/PNa of CavMr was more than 200 ( Figure 3e and Table 1 ) , comparable to that of CavAb , an artificial Ca2+ channel . Anomalous mole fraction effects were not observed in CavMr ( Figure 4a and b ) , suggesting that CavMr has a very high affinity for Ca2+ . In addition to providing new insights about general Ca2+-selective mechanisms , CavMr has the potential to be a new genetic tool for upregulating calcium signaling , as BacNavs are useful genetic tools for increasing action potential firing in mice ( Bando et al . , 2016; Kamiya et al . , 2019; Lin et al . , 2010 ) . Phylogenetic analysis demonstrated that CavMr and NavPp are similar to each other , but distant from canonical BacNavs ( Figure 1b ) . The high Ca2+ selectivity of CavMr was transferable to NavPp . Intriguingly , two pairs of mutants with the same selectivity filter ( CavMr-G4D and NavPp-T6V , CavMr-V6T and NavPp-D4G ) showed a very similar tendency with regard to both the order and extent of cation selectivity ( Figure 8a and c ) . Therefore , the basic overall architecture of the NavPp selectivity filter could be similar to that of CavMr . On the other hand , the Ca2+-selectivity mechanism of CavMr completely differs from that of CavAb . Structural comparison of NavAb and CavAb showed that the aspartate mutations did not alter the main chain trace , and simply introduced the negative charges around the ion pathway to increase Ca2+ permeability ( Figure 9a and b ) ( Tang et al . , 2014 ) . By contrast , in the case of CavMr , two non-charged residues ( Gly4 and Val6 ) are required for the high Ca2+ selectivity ( Figures 7d and 8a ) , whereas Asp7 is not necessary ( Figure 4g ) . A no-charge mutation at position 7 , CavMr-D7M ‘TLEGWVM’ , is an outstanding example demonstrating that high Ca2+ selectivity can be achieved in the absence of any aspartates in its filter region . Furthermore , the introduction of a negative charge into the selectivity filter ( G4D mutation ) had an effect on the Ca2+ selectivity of CavMr that was opposite to the effect seen in NavAb and NaChBac ( Tang et al . , 2014; Yue et al . , 2002 ) . Moreover , the decreased selectivity of monovalent and divalent cations in G4S also indicates that the glycine at position 4 plays a crucial role in Ca2+ selectivity in CavMr ( Figure 8b ) . The flexibility and/or small size of the glycine at position 4 in CavMr might be critical . These findings are inconsistent with those derived from the Ca2+-selective mutants of NavAb and NaChBac , and therefore the native structure of the selectivity filter and the molecular mechanism of ion selectivity of CavMr are thought to differ from those of CavAb . The structure of CavMr is not yet available , but we are able to speculate on the structure of the selectivity filter of CavMr , based on the structure of human Cav1 . 1 subdomains I and III ( Wu et al . , 2016 ) ( Figure 9c ) , whose selectivity filter sequences are very similar to that of CavMr . In the selectivity filter of Cav1 . 1 subdomains I and III , the side chain of the residue at position 7 is shifted outward . The position 4 glycine residue widens the entrance of the selectivity filter , facilitating the entry of hydrated cations into the ion pore and possibly increasing Ca2+ selectivity . Prokaryotes have a number of putative Ca2+-binding proteins , such as EF-hand proteins , P-type Ca2+ pumps , and Ca2+ transporters ( Domínguez et al . , 2015 ) . The intracellular Ca2+ concentration is kept low and changes in response to mechanical and chemical stimuli ( Dominguez , 2004 ) . These features imply that prokaryotic Ca2+ signaling is similar to that of eukaryotes . The strong ability of CavMr to exclude Na+ and K+ along with Ca2+ permeation suggests that its primary physiological role is Ca2+ intake in response to a voltage change ( Figure 3e and f ) . In some bacteria , the direction of flagellar rotation and chemotaxis changes depending on the internal Ca2+ concentration ( Ordal , 1977; Tisa et al . , 1993; Tisa and Adler , 1995 ) . M . ruber was isolated from hot springs , and therefore a sufficient amount of Ca2+ is likely to exist in its native environment ( Loginova et al . , 1984 ) . CavMr activation by a voltage change , which could vary depending on the environmental ionic conditions , might lead to any response that allows adaptation to the new environment , such as flagellar rotation . These characteristics indicate the existence of signal coupling between the membrane voltage and Ca2+ , even in the early stages of life , which might be the origin of the corresponding functions in eukaryotes , such as muscle contraction . NavPp permeates more Na+ than Ca2+ , but its selectivity is modest ( Figure 6d and Table 1 ) . Notably , P . pacifica is a marine myxobacterium that requires NaCl for its growth ( Iizuka et al . , 2003 ) . As mentioned above , the basic architecture of the CavMr/NavPp group is thought to produce a preference for Ca2+ . P . pacifica might modify this channel architecture to acquire a Na+ intake pathway , which would probably result in the remaining feature of low-affinity Ca2+ inhibition in NavPp . This flexible usage of homo-tetrameric channels to allow different cations to permeate is also reported in another bacterium , Bacillus alkalophilus ( DeCaen et al . , 2014 ) . NsvBa from B . alkalophilus is a non-selective channel whose selectivity filter is changed from ‘TLESWAS’ , a typical Na+-selective sequence in alkaliphilic bacillus , to ‘TLDSWGS’ , possibly as an adaptation to its ionic environment . Recently , an early eukaryote , diatom , was found to have another homo-tetrameric channel with no selectivity , namely EukCat , which has an important role in electrical signaling in this species ( Helliwell et al . , 2019 ) . These findings suggest that the cation selectivity of the homo-tetrameric channel family can be flexibly tuned to realize the required roles specific to its original species . Aspartate residues are generally observed in the Ca2+ permeation pathway in ion channels , as well as in many Ca2+-binding proteins ( Halling et al . , 2016; Yan et al . , 2015; Zalk et al . , 2015 ) . Actually , NavAb and NaChBac were successfully transformed into Ca2+-selective channels with the aspartate-introduced filter sequences ‘TLDDW ( S/A ) D’ ( Tang et al . , 2014; Yue et al . , 2002 ) . But , our results elucidate that this strategy is not the only way to achieve high Ca2+ selectivity . Human Cav subdomains possess , at most , two aspartate residues in their selectivity filters in a part other than position 3 . In addition , the negatively charged residue at position 3 , which is thought to be the most critical for cation selectivity in both Navs and Cavs , is not aspartate but glutamate in most of the human Cav subdomains ( Yu and Catterall , 2004 ) . CavAb has 12 aspartates in the selectivity filter of its channel tetramer , while there are four aspartates in CavMr . The net negative charge is 5~7 in mammalian Cavs , 8 in CavMr , and 12 in CavAb . As shown in mammalian Cavs , Ca2+ selectivity can be achieved with even fewer negative charges in the selectivity filter than is the case in CavAb , which suggest that the calcium-selective mechanism requires a specific backbone structure of the pore domain depending on its selectivity filter charges . The members of the AnclNav group can be found in a variety of bacterial phyla and in archaea ( Figure 1—figure supplement 1 ) . In particular , the Deinococcus-Thermus phylum , in which M . ruber is included , is considered to be relatively close to the universal ancestor of life ( Hug et al . , 2016 ) . It is also notable that the filter sequence of NavPp is completely the same as that of the ancestral BacNav predicted in the previous report [TLED ( or S in equal probability ) WTD] ( Liebeskind et al . , 2013 ) . These pieces of evidence suggest that the AnclNav group preserves the feature of an ancestral BacNav . In our comprehensive phylogenetic analysis , all of the one-domain type of channel groups , including the AnclNav group , are almost equally distant from the root of the divided subdomains of eukaryotic 24TM Cavs/Navs ( Figure 1—figure supplement 2 ) . Again , this information is insufficient to allow us to deduce any conclusion for a eukaryotic ancestor of 24TM channels just before subdomain duplication . It is noteworthy , however , that the selectivity filter sequence of CavMr is very similar to those of human Cav subdomains I and III , both of which possess a glycine at position 4 ( Figure 9c ) . In particular , the Cav3 . 1 and 3 . 2 subdomains I have the same sequence as the evolutionally distant CavMr . These sequence similarities of the glycine residue at position 4 are also found in CatSper , the sperm calcium permeable channels ( Darszon et al . , 2011 ) , which branches close from the convergent point of four subdomains . The channel region of CatSper is formed by four different subunits ( CatSper1–4 ) . The selectivity filters of CatSper 3 ‘TVDGWTD’ and CatSper 4 ‘TQDGWVD’ are similar to that of CavMr . Taken together , these findings suggest that the selectivity filter of eukaryotic ancestor of 24TM channels might have been similar to those of AnclNavs , especially to that of CavMr . In the future , information about the structure of these homo-tetrameric channels could help us to gain a deeper understanding of channel evolution , and further investigation of the detailed structure of CavMr may help us to elucidate the principles and origin underlying Ca2+ selectivity .
The NaChBac amino acid sequence ( NP_242367 ) was used as the query for a BLASTP search against the Microbial Genomic database at NCBI . The identified primary sequence data were obtained from Entrez at NCBI ( Meiothermus ruber as ZP_04038264 , Plesiocystis pacifica as ZP_01909854 , Halomonas elongata as YP_003896792 and Teredinibacter turnerae as YP_003073405 ) . These DNAs were synthesized by Genscript Inc and subcloned into the pCI vector ( Promega ) using the EcoRI and SalI sites and the pBiEX vector ( Novagen ) using the NcoI and BamHI sites , respectively . Site-directed mutagenesis was achieved by polymerase chain reaction ( PCR ) of the full-length plasmid containing the Nav gene using PrimeSTARMAX DNA Polymerase ( Takara Bio . ) . All clones were confirmed by DNA sequencing . Phylogenetic analyses were performed using the Molecular Evolutionary Genetics Analysis ( MEGA X ) software ( Kumar et al . , 2016 ) . Protein sequences of putative BacNavs were collected by reference to a previous report ( Liebeskind et al . , 2013 ) and using multi-round searches of the NCBI database using NaChBac , NavAb , CavMr and NavPp as templates . Multiple sequence alignment was generated with MUSCLE contained in MEGA X . To generate the phylogenetic tree of BacNavs , the targeted sequences to be analyzed were selected using GBLOCKS0 . 91B ( Castresana , 2000 ) . The least stringent parameters selected 167 amino acids that cover most of transmembrane domains and the S4/S5 linker . For comparison with the eukaryotic Navs , Cavs , CatSper and EukCatA , the sequences of these proteins were collected by reference to the previous studies of Nav/Cav evolution ( Cai and Clapham , 2008; Gur Barzilai et al . , 2012; Helliwell et al . , 2019; Liebeskind et al . , 2011 ) . The sequences of four-domain type of Navs and Cavs were divided to each subdomain . These divided subdomain sequences were aligned with EukCatA , CatSper and BacNavs using MUSCLE , and then the targeted sequences , which correspond to the selected amino acids in the analysis of BacNavs , were extracted . Regions that are poorly conserved between BacNavs and eukaryotic channels were manually removed , leaving 162 amino acids . Maximum likelihood trees were generated using MEGA X . The model validation was performed and determined to be a LG+G+F model in both the analyses , with and without eukaryotic channels . The sequence logos that indicate the frequencies at which different amino acids appear in the selectivity filter were generated with WebLogo ( Crooks et al . , 2004 ) . For the recordings related to mole fraction effects ( Figure 4a–d ) , currents were recorded from Chinese hamster ovary ( CHO ) -K1 cells ( ATCC catalog number CCL-61 ) that expressed the channels . The recordings were performed as described previously ( Tateyama and Kubo , 2018 ) . Cells were transfected with channel DNAs using the LipofectAMINE 2000 ( Invitrogen ) and plated onto cover slips . Currents were recorded 24–36 hr after transfection . Current recording by the whole-cell patch-clamp technique was performed using Axopatch 200B amplifiers , Digidata1332A , and pClamp nine software ( Molecular Devices ) . The pipette solution contained 130 mM KCl , 5 mM Na2-ATP , 3 mM EGTA , 0 . 1 mM CaCl2 , 4 mM MgCl2 and 10 mM HEPES ( pH 7 . 2 adjusted with KOH ) . The bath solution contained 135 mM NaCl , 4 mM KCl , 1 mM CaCl2 , 5 mM MgCl2 and 10 mM HEPES ( pH 7 . 4 adjusted with NaOH ) . For the measurement of mole fraction effects , bath solutions containing different ratios of NaCl/CaCl2 ( 135/0 , 108/18 , 81/36 , 54/54 , 27/82 and 0/90 mM ) were used . A Ca2+-free solution was achieved by using a solution containing 135 mM NaCl , 1 mM EGTA and 0 mM CaCl2 . Recordings other than those for mole fraction effects were performed using SF-9 cells . SF-9 cells ( ATCC catalog number CRL-1711 ) were grown in Sf-900 III medium ( Gibco ) complemented with 0 . 5% 100 × antibiotic antimycotic ( Gibco ) at 27°C . Cells were transfected with target channel-cloned pBiEX vectors and enhanced green fluorescent protein ( EGFP ) -cloned pBiEX vectors using Fugene HD transfection reagent ( Promega ) . The channel-cloned vector ( 2 μg ) was mixed with 0 . 5 μg of the EGFP-cloned vector in 100 μL of the culture medium . Next , 3 μL Fugene HD reagent was added and the mixture was incubated for 10 min before the transfection mixture was gently dropped onto cultured cells . After incubation for 16–48 hr , the cells were used for electrophysiological measurements . In the measurement of I–V relation curves , the pipette solution contains 75 mM NaF , 40 mM CsF , 35 mM CsCl , 10 mM EGTA , and 10 mM HEPES ( pH 7 . 4 adjusted by CsOH ) . For evaluation of ion selectivity , a high-Na+ pipette solution [115 mM NaF , 35 mM NaCl , 10 mM EGTA , and 10 mM HEPES ( pH 7 . 4 adjusted by NaOH ) ] was used . For the evaluation of Ca2+ , Sr+ , K+ and Cs+ selectivity , Ca2+ solution [100 mM CaCl2 , 10 mM HEPES ( pH 7 . 4 adjusted by Ca ( OH ) 2 ) , and 10 mM glucose] , Sr2+ solution [100 mM SrCl2 , 10 mM HEPES ( pH 7 . 4 adjusted by Sr ( OH ) 2 ) , and 10 mM glucose] , K+ solution [150 mM KCl , 2 mM CaCl2 , 10 mM HEPES ( pH 7 . 4 adjusted by KOH ) , and 10 mM glucose] , and Cs+ solution [150 mM CsCl , 2 mM CaCl2 , 10 mM HEPES ( pH 7 . 4 adjusted by CsOH ) , and 10 mM glucose] , respectively , were used as the bath solution . Erev of highly Ca2+-selective channels were measured under three external solutions containing: 144 mM NMDG-Cl and 4 mM CaCl2; 135 mM NMDG-Cl and 10 mM CaCl2; and 120 mM NMDG-Cl and 20 mM CaCl2 ( 10 mM HEPES pH 7 . 4 adjusted with HCl ) . Erev of highly Ca2+-selective channels for the calculation of PK/PCa and PCs/PCa were measured under external solutions containing 135 mM NMDG-Cl and 10 mM CaCl2 ( 10 mM HEPES pH 7 . 4 adjusted with HCl ) with high-K+ pipette solution [115 mM KF , 35 mM KCl , 10 mM EGTA , and 10 mM HEPES ( pH 7 . 4 adjusted by KOH ) ] and high-Cs+ pipette solution [115 mM CsF , 35 mM CsCl , 10 mM EGTA , and 10 mM HEPES ( pH 7 . 4 adjusted by CsOH ) ] , respectively . Erev of NavPp for the calculation of PCa/PNa or PSr/PNa were measured in an external solution containing 50 mM NMDG-Cl , 40 mM NaCl , 40 mM CaCl2 or SrCl2 and 10 mM HEPES ( pH 7 . 4 ) adjusted with NaOH . Erev of NavPp for the calculation of PCs/PNa was measured using a high-Cs+ pipette solution and an external solution containing 110 mM NMDG-Cl , 40 mM NaCl , 3 mM CaCl2 and 10 mM HEPES ( pH 7 . 4 ) adjusted with NaOH . As the pipette solution for measurement of the Ca2+ block effect in NavPp , low-Na+ pipette solution [140 mM CsF , 10 mM NaCl , 10 mM EGTA , and 10 mM HEPES ( pH 7 . 4 adjusted by CsOH ) ] and high-Na+ pipette solution were used for inward and outward current measurement , respectively . As a bath solution , Ca2+ blocking solution [30 mM NaCl , 120 mM NMDG-Cl , 1 . 5 mM CaCl2 , 10 mM HEPES ( pH 7 . 4 adjusted by NaOH ) and 10 mM glucose] was used for the 1 . 5 mM Ca2+ blocking condition . In 10 mM Ca2+ blocking condition , a bath solution contains 30 mM NaCl , 105 mM NMDG-Cl , 10 mM CaCl2 , 10 mM HEPES ( pH 7 . 4 adjusted by NaOH ) and 10 mM glucose . In each Ca2+ blocking condition , 15 mM NMDG-Cl was replaced by 10 mM CaCl2 . Cancelation of the capacitance transients and leak subtraction was performed using a programmed P/10 protocol delivered at −140 mV . The bath solution was changed using the Dynaflow Resolve system . All experiments were conducted at 25 ± 2°C . Cells that have a leak current smaller than 0 . 5nA were used for measurement . When any outliers were encountered , these outliers were excluded if any abnormalities were found in other measurement environments , but were included if no abnormalities were found . All results are presented as mean ± standard error . To determine the ion selectivity of each channel , the intracellular solution and extracellular solution were arbitrarily set and the reversal potential at each concentration was measured by giving the ramp pulse of membrane potential . The applied ramp pulse was set to include the reversal potential . In addition , a depolarization stimulus of 2–10 ms was inserted to check whether the behavior of the cell changed for each measurement . As a result , PCa/PNa was calculated by substituting the obtained reversal potential ( Erev ) into the expression derived from the GHK equation ( Frazier et al . , 2000 ) ;PCa/PNa=- ( [Na+]in-[Na+]oute-ErevF/RT ) ( 1-e-2ErevF/RT ) 4 ( [Ca2+]in-[Ca2+]oute-2ErevF/RT ) ( 1-e-ErevF/RT ) where F is Faraday’s constant , R is the Gas constant , and T is 298 . 1 ( K ) . The same expression was used for Sr2+ . The Sr2+ selectivity ( PSr/PNa ) was measured in the same way . Na+ selectivity against monovalent cations ( PM/PNa ) was calculated by substituting the obtained reversal potential and PCa/PNa into the expression derived from the GHK equation ( Lopin et al . , 2012 ) :PM/PNa=[-4 ( [Ca2+]in-[Ca2+]oute-2ErevF/RT ) ( 1-e-ErevF/RT ) ( [Na+]in-[Na+]oute-ErevF/RT ) ( 1-e-2ErevF/RT ) ∙ ( PCa/PNa ) -1][ ( [Na+]in-[Na+]oute-ErevF/RT ) ( [M+]in-[M+]oute-ErevF/RT ) ] | Electrical signals in the brain and muscles allow animals – including humans – to think , make memories and move around . Cells generate these signals by enabling charged particles known as ions to pass through the physical barrier that surrounds all cells , the cell membrane , at certain times and in certain locations . The ions pass through pores made by various channel proteins , which generally have so-called “selectivity filters” that only allow particular types of ions to fit through . For example , the selectivity filters of a family of channels in mammals known as the Cavs only allow calcium ions to pass through . Another family of ion channels in mammals are similar in structure to the Cavs but their selectivity filters only allow sodium ions to pass through instead of calcium ions . Ion channels are found in all living cells including in bacteria . It is thought that the Cavs and sodium-selective channels may have both evolved from Cav-like channels in an ancient lifeform that was the common ancestor of modern bacteria and animals . Previous studies in bacteria found that modifying the selectivity filters of some sodium-selective channels known as BacNavs allowed calcium ions to pass through the mutant channels instead of sodium ions . However , no Cav channels had been identified in bacteria so far , representing a missing link in the evolutionary history of ion channels . Shimomura et al . have now found a Cav-like channel in a bacterium known as Meiothermus ruber . Like all proteins , ion channels are made from amino acids and comparing the selectivity filter of the M . ruber Cav with those of mammalian Cavs and the calcium-selective BacNav mutants from previous studies revealed one amino acid that plays a particularly important role . This amino acid is a glycine that helps select which ions may pass through the pore and is also present in the selectivity filters of many Cavs in mammals . Together these findings suggest that the Cav channel from M . ruber is similar to the mammal Cav channels and may more closely resemble the Cav-like channels thought to have existed in the common ancestor of bacteria and animals . Since other channel proteins from bacteria are useful genetic tools for studies in human and other animal cells , the Cav channel from M . ruber has the potential to be used to stimulate calcium signaling in experiments . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"biochemistry",
"and",
"chemical",
"biology",
"structural",
"biology",
"and",
"molecular",
"biophysics"
] | 2020 | A native prokaryotic voltage-dependent calcium channel with a novel selectivity filter sequence |
The amylase gene ( AMY ) , which codes for a starch-digesting enzyme in animals , underwent several gene copy number gains in humans ( Perry et al . , 2007 ) , dogs ( Axelsson et al . , 2013 ) , and mice ( Schibler et al . , 1982 ) , possibly along with increased starch consumption during the evolution of these species . Here , we present comprehensive evidence for AMY copy number expansions that independently occurred in several mammalian species which consume diets rich in starch . We also provide correlative evidence that AMY gene duplications may be an essential first step for amylase to be expressed in saliva . Our findings underscore the overall importance of gene copy number amplification as a flexible and fast evolutionary mechanism that can independently occur in different branches of the phylogeny .
Diet has been a significant adaptive force in shaping human and nonhuman primate variation ( Hardy et al . , 2015; Milton , 1981; Zhang et al . , 2002 ) . One of the best-described examples of diet-related adaptation is the expansion of the copy number of the amylase gene in concordance with the increase of starch consumption in the human lineage ( Perry et al . , 2007 ) , likely postdating the human Neanderthal split ( Inchley et al . , 2016 ) . A gene duplication in the ancestor of Old World monkeys and great apes initially led to the formation of two amylase genes ( AMY2A and AMY2B ) with pancreas-specific expression ( Samuelson et al . , 1990 ) . Then , a subsequent gene duplication in the ancestor of great apes led to the formation of AMY1 which gained salivary gland-specific expression ( Meisler and Ting , 1993 ) . In the human lineage , further gene copy number gains of AMY1 led to increased expression of the AMY1 enzyme in human saliva ( Perry et al . , 2007 ) . Gene copy numbers of both AMY1 and AMY2 vary in different human populations ( Carpenter et al . , 2015; Usher et al . , 2015 ) , the former correlating with the extent of traditional starch consumption in these communities ( Perry et al . , 2007 ) . While the evolution of the amylase locus in the human lineage is well described , its evolution in other mammals is less well understood . Some studies have produced intriguing findings . For example , it was shown that mice , rats , and pigs express substantial levels of amylase in their saliva ( Boehlke et al . , 2015; Janiak , 2016 ) In addition , the amylase locus has been shown to be evolving under positive selection in dogs and in house mice ( Reiter et al . , 2016; Staubach et al . , 2012 ) . However , a comprehensive analysis of the evolutionary dynamics shaping the amylase locus across mammals is missing . Within this context , one interesting question is how amylase has evolved in animals who live in a commensal relationship with humans . Recent studies , for instance , have shown that dogs gained multiple copies of the amylase gene after their split from the wolf 30 , 000–40 , 000 years ago ( Skoglund et al . , 2015 ) . This might likely be a result of their domestication , which exposed them to human food leftovers rich in starch ( Axelsson et al . , 2013; Botigué et al . , 2017; Ollivier et al . , 2016; Reiter et al . , 2016 ) . Thus , the evolution of amylase in other domesticated or human commensal mammals remains an alluring area of inquiry . Similarly , our understanding of the evolution of the amylase locus within the primate lineage remains limited . For instance , it is not known why some Old World monkeys have substantial amylase enzymatic activity in their saliva , despite missing the amylase duplication found in great apes ( Janiak , 2016 ) . Here , we address three areas of inquiry with regard to the evolution of the amylase locus in mammals: ( i ) Can the link between diet and amylase evolution , suggested in the human lineage , be generalized to other mammals ? ( ii ) What are the evolutionary forces that shape amylase copy numbers in mammals ? ( iii ) What are the genetic mechanisms in different mammals leading to expression of amylase in salivary glands ? To answer these questions , we pursued a comprehensive investigation of amylase gene copy numbers and salivary expression across multiple mammalian lineages .
The human lineage-specific amylase gene duplications were initially thought to represent a unique case of evolutionary adaptation to increased starch consumption in humans ( Perry et al . , 2007 ) . Up to five more haploid copies of the amylase gene can be found in humans than in chimpanzees . Therefore , the recent revelation that a similar increase in amylase gene copy number occurred in dogs ( Axelsson et al . , 2013; Ollivier et al . , 2016 ) is remarkable since it shows that the same gene underwent bursts of gene copy number gains in two separate species independently . Copy number variation was also noted in pigs ( Paudel et al . , 2013 ) . To comprehensively investigate amylase gene copy number gains in other mammalian lineages , we conducted a digital droplet polymerase chain reaction ( ddPCR ) -based analysis of amylase gene copy numbers from 204 DNA samples across 46 species encompassing all major branches of the mammalian phylogeny . In addition to humans and dogs , we discovered similar bursts ( i . e . gains of more than one copy ) of amylase gene copy number in mice , rats , pigs , and boars ( Figure 1—figure supplement 1 , Supplementary file 1 ) . We hypothesized that the elevated gene copy numbers observed in different branches of the mammalian phylogeny ( Figure 1 ) result from independent duplication events . An alternative explanation would be that the ancestor of placental mammals had multiple copies of the amylase gene , which were subsequently lost in certain mammalian lineages . To distinguish between these two scenarios , we constructed from available reference genomes a maximum likelihood tree of amylase coding sequences ( Figure 2A , see Figure 2—figure supplement 1 for a more comprehensive tree with outgroups included ) . Our results showed that amylase genes within a given species are more similar to each other than they are to those of other species . One explanation for this observation could be that duplications of the amylase gene might have occurred in each lineage independently . Yet another explanation could be that lineage-specific gene conversion events occurred among ancestral amylase copies . Indeed , inter- and intra-chromosomal crossover has been shown in the amylase locus of humans ( Groot et al . , 1990; Gumucio et al . , 1988 ) . Such a process , if it occurred frequently enough , could potentially generate high similarity among amylase gene copies in any given mammalian lineage . These two scenarios , independent lineage-specific duplication events and gene conversion among existing gene copies , are difficult to distinguish using phylogenetic analysis alone ( Mendes et al . , 2018 ) . To solve this conundrum in humans , Samuelson et al . searched for lineage-specific signatures associated with individual gene copies in the amylase locus . They identified a retrotransposon ( HERV_a_int ) inserted upstream of a new amylase gene duplicate ( AMY1 ) in the ancestor of great apes ( Samuelson et al . , 1990 ) . This retrotransposon was found to be associated with all the additional AMY1 copies detected in humans ( Perry et al . , 2007 ) . This finding strongly supported the notion that the duplications of AMY1 occurred after the human-chimpanzee phylogenetic split ( Perry et al . , 2007; Samuelson et al . , 1990 ) . Based on that , we asked if similar retrotransposons or other genomic signatures could help us determine whether the amylase gene copy number bursts in other mammalian genomes occurred independently . We first interrogated the mouse reference genome , as it is adequately complete for such an analysis . Indeed , we found a mouse lineage-specific retrotransposon ( L1Md_T ) in the upstream region of five out of the seven mouse amylase gene copies . The presence of this retrotransposon along with the duplicated copies parallels the situation in humans ( Figure 2B ) . By ddPCR analysis , we found 9–13 diploid copies of the amylase gene in brown rats and 6–7 copies in black rats and wood rats ( Supplementary file 1 ) . Considering the close phylogenetic relationship of rats and mice , we expected that the high copy number of amylase had evolved in their rodent ancestor . However , the L1Md_T retrotransposon was not found in rats . When we investigated the amylase locus in the rat genome , we found a rat-specific retrotransposon ( L1_Rat3 ) inserted upstream of two out of the three rat amylase copies in the assembled rat reference genome ( Figure 2B ) . Therefore , amylase gene duplications in rats likely occurred independently from the ones in mice . The amylase gene copy numbers in rats found by ddPCR do not match those annotated in the rat reference genome . This inconsistency could be due to major sequence gaps in the assembled amylase locus ( indicated by line breaks in Figure 2B ) . As an additional complication , we discovered that one of the gene copies , amy1 in mouse ( discussed in the next section within the context of salivary expression , see Figure 2—figure supplement 2 ) , is shared with other rodent species including rats . Nevertheless , our results support that most amylase gene copy number gains occurred in mouse and rat lineages independently . We also conducted similar analyses of the amylase locus in dogs and pigs ( Figure 2B ) . Despite the fact that the assemblies of their genomes are not as complete as human and mouse reference genomes , we were able to investigate the genomic signatures in contigs that harbor distinct copies of amylase genes . In dogs , we found a canid-specific L1 element ( L1_Canid ) in all four amylase gene copies assembled across three different contigs . In pigs , we found an older lineage-specific L1 element ( L1M3 ) downstream of all six amylase copies assembled across three contigs . Overall , we found lineage-specific retrotransposons located in similar proximity ( 0 . 8–4 . 2 kb upstream or downstream ) to multiple amylase gene copies in human , mouse , rat , pig , and dog reference genomes ( Figure 2B ) . In each of these cases , the respective retrotransposons are in identical positions relative to the amylase gene within each species . We surmised that these retrotransposons inserted in proximity to an ancestral amylase copy in each species independently and were subsequently duplicated along with further gained amylase gene copies . These findings do not rule out that gene conversion as well as other mechanisms ( incomplete lineage sorting , crossover events , and ancestral gene duplication polymorphisms ) might have shaped variation in this locus . However , the fact that lineage-specific retrotransposons accompany amylase gene copy number gains in humans , mice , rats , dogs , and pigs , clearly points to lineage-specific duplications as a major driver of amylase gene copy number bursts in these species . The extent to which retrotransposons affect mutational or functional dynamics in the amylase locus remains an important area for future research . Ancestrally , amylase was a pancreatic enzyme in mammals . However , in certain mammalian species , amylase became expressed also in saliva ( Chauncey et al . , 1963 ) . In humans , this acquisition of salivary gland-specific expression has been well explained ( Ting et al . , 1992 ) . It has been shown that the above-described retrotransposon insertion along with the AMY1 duplicate in the ancestor of great apes was responsible for tissue-specific expression of this gene in salivary glands ( Samuelson et al . , 1990 ) . Previous studies hypothesized that a similar , but independent gene duplication event led to the expression of amylase also in the saliva of mice ( Meisler and Ting , 1993 ) . It remained unresolved whether the mechanism that enabled the expression of amylase in mouse saliva is similar to that determined for humans . Moreover , even though some reports noted the expression of amylase in the saliva of various other mammalian species ( Janiak , 2016 ) , a comprehensive analysis of its expression across the mammalian phylogeny is still missing . Another unanswered question from an evolutionary perspective is what a potential adaptive benefit of expressing amylase in saliva could be . Even though amylase is a digestive enzyme , it is clear that in most mammals starch digestion primarily occurs through the activity of the pancreatic enzyme in the intestines rather than in the mouth ( Fernández and Wiley , 2017 ) . To address these questions , we performed a screen across the mammalian phylogeny to investigate which lineages express amylase activity in saliva . We used a two-pronged approach , comprising a starch lysis plate assay ( Figure 3A ) and a high-sensitivity in-solution fluorescence-based amylase assay ( Figure 3B ) . Currently , our study provides the most comprehensive documentation of salivary amylase activity in mammals , encompassing 127 saliva samples across 22 species ( Supplementary file 1 ) . This is a significant contribution given that previous studies varied considerably in sample preparation , methods of analysis , and assay sensitivity ( Janiak , 2016 ) . Our results showed that amylase activity in saliva is more widespread among mammals than previously thought ( Figure 3B ) . In addition to species that were already known to express amylase in their saliva , we observed salivary amylase activity in some New World monkeys , boars , dogs , deer mice , woodrats , and giant African pouched rats ( Supplementary file 1 ) . It is important to note here that our findings also suggest that amylase activity in dog saliva varies from breed to breed ( Supplementary file 1 ) . It remains to be determined to what degree this variable expression of amylase in the saliva of different dog breeds might have been the result of older adaptive forces , perhaps related to dogs becoming companions of humans thereby exposed to a human diet , rather than merely occurring as a byproduct of recent intentional breeding for other traits . To explain the expression of amylase in the saliva of some mammalian lineages , but not others , we considered two scenarios . First , it is possible that salivary expression of amylase could be an ancestral trait that was subsequently lost in most species . Second , it is possible that salivary expression may have evolved independently in different lineages . To distinguish between these two scenarios , we asked whether orthologous copies of the amylase gene are expressed in humans and mice . Based on previous work showing that AMY1 copies in humans are expressed in salivary glands , we asked which amylase copy is expressed in mouse salivary glands . By mapping parotid salivary gland RNA-Seq data ( Gluck et al . , 2016 ) to the mouse reference genome ( mm9 ) , we found that the copy annotated as mouse amy1 is expressed in salivary glands , while the other amylase gene duplicates have a negligible expression in that tissue ( Figure 2—figure supplement 2 ) . However , mouse amy1 is not orthologous to human AMY1 ( Meisler and Ting , 1993 ) . Furthermore , its amino acid sequence is distinct from those of other amylase copies in the mouse genome . This distinct amy1 sequence is shared with rat and other rodents including deer mouse , vole , Mongolian gerbil , and golden hamster . This suggests that the duplication event that led to the formation of mouse amy1 , the copy that is expressed in mouse salivary glands , likely occurred in an ancestor of muroidea . Interestingly , the more recently acquired , mouse and rat-specific amylase gene duplications are not expressed in mouse salivary glands ( Figure 2—figure supplement 2 ) . We could not find a mammalian species that underwent a ‘burst’ of amylase gene copy number that did not show concurrent salivary amylase activity ( Figure 3C ) . Also , we found no species with less than four diploid amylase copies that showed any measurable amylase activity in saliva . It is important to note here that the relationship between amylase gene copy number and salivary amylase activity cannot be explained by linear correlation . For example , rhesus macaques and baboons have relatively low amylase gene copy numbers ( 5–6 diploid copies ) but show high amylase activity in their saliva ( 514–1 , 652 units per mg of total protein ) . In contrast , dogs , which among mammals showed some of the highest gene copy numbers of amylase ( 9–19 diploid copies ) , express very low amounts of amylase in their saliva ( 0–9 units/mg ) . Overall , these results suggest that a gene duplication may be a necessary condition for amylase to become expressed in saliva . Subsequently , however , different regulatory architectures are likely responsible for the differences in amylase activity observed in saliva across different mammalian species . For humans , it has been postulated that starch consumption has driven the increased gene copy number of the amylase gene through positive selection ( Perry et al . , 2007 ) . For dogs , the rapid amylase gene copy number increase as compared to wolves ( 3–8 haploid copy number gain ) has been associated with a transition during their domestication from a primarily meat-based diet to a diet enriched in starch ( Axelsson et al . , 2013 ) . Based on these previous findings , we hypothesized that in other mammalian species , gains in gene copy number and the associated gain of amylase expression in saliva are likely driven by starch being a dietary component . Unfortunately , a systematic survey describing the amounts of starch consumption across mammals is lacking . Moreover , the amount of starch in the diet varies among subspecies , and sometimes even among geographically distinct populations of the same species ( Pineda-Munoz and Alroy , 2014 ) . Thus , to test the above hypotheses , we categorized mammals into those that consume specialized diets ( strict carnivores or strict herbivores ) and those consuming broad-range diets ( including different amounts of starch ) . Based on extensive literature review ( see Materials and methods ) , we subdivided mammals consuming broad-ranged diets further into those that consume diets containing high amounts of starch ( humans , mice , brown and black rats , dogs , pigs , and boars ) and those that consume low amounts of starch . We then conducted a comparative analysis of amylase copy number and salivary enzymatic activity among these categories ( Figures 4A , B and C ) . We found that species consuming a broad-range diet generally harbor significantly higher copy numbers of the amylase gene ( Figure 4A ) . Specifically , the mean diploid amylase copy number among animals consuming specialized diets is 2 . 4 , while it is 7 . 0 for those consuming broad-range diets . Corrected for phylogenetic dependence , a broad-range diet is significantly associated with increased amylase gene copy number ( p =~9×10−6 ) . Among species that consume a broad-range diet , we found that those who over recent evolutionary time gained access to abundant starch-rich foods — either through domestication ( as in the case of dogs and pigs ) or through dietary commensalism with humans ( as in the case of house mice or brown and black rats ) — harbor considerably higher copy numbers of the amylase gene ( Figure 4A ) . The mean diploid amylase gene copy number is 11 . 6 for species consuming a starch-rich diet , while it is 5 . 0 for those that consume lower amounts of starch in their diet . When corrected for phylogenetic dependence , starch consumption ( categorized into high- and low-starch content in diet ) is significantly correlated with amylase gene copy number ( p =~10−4 ) . We found that mammals consuming a broad-range diet also express on average a thousand-fold higher amylase activity in their saliva ( ~243 . 6 units/mg of total salivary protein ) than those consuming specialized diets ( ~0 . 1 units/mg ) ( Figure 4B ) . When corrected for phylogenetic dependence , it appeared that consumption of a broad-range diet is significantly associated with the enzymatic activity of amylase in saliva ( p =~10−4 ) . Salivary expression of amylase , however , turned out to not be associated with the amount of starch consumption per se ( Figure 4B ) . Rather , salivary expression of amylase appeared to be associated with the consumption of starch , regardless of the amount . Previous work in humans and in rats showed that salivary amylase is linked to the perception of starch ( Mandel et al . , 2010; Sclafani et al . , 1987 ) . Based on this , one possible evolutionary explanation could be that the ability to enzymatically liberate sugar from long-chain starch molecules for the perception of sweet taste might have provided a metabolic fitness advantage to mammalian species consuming a broad range diet . We like to point to two outliers in our dataset of primate species consuming broad-range diets . We found that marmosets and lemurs express negligible amylase activity in their saliva ( ~0 . 2 and ~0 . 009 units/mg , respectively ) . However , for them starch is not a primary component in their otherwise diverse diet . Marmosets were shown to consume tree gum while lemurs consume tamarind fruit , reportedly rich in non-starch polysaccharides ( for further information on starch consumption by mammalian species see Materials and methods ) . Next , to investigate the influence of human commensalism , we conducted a comparative investigation of amylase gene copy number and enzymatic activity in saliva between mammalian species interacting with humans and their closest evolutionary relatives in the wild . In dogs , that due to their commensalism with humans consume a considerable amount of starch , we found an increase , not only in amylase gene copy number ( Axelsson et al . , 2013 ) but also in enzymatic activity of amylase in saliva as compared to the carnivorous wolf from which dogs diverged approximately 30 , 000–40 , 000 years ago ( Skoglund et al . , 2015 ) ( Figure 3A ) . A less substantial increase was found in species that already consumed starch in their ancestral state ( e . g . mice and rats which diverged from granivorous ancestors ) . Along the same lines , we found no difference of amylase gene copy numbers and salivary enzymatic activity between domesticated pigs and wild boars . This could be explained because boars , the ancestral species , already consumed starch in amounts comparable to those of pigs , their domestic counterparts . In fact , previous observations showed that boars and humans have similar starch-rich ancestral diets due to their consumption of underground starch-containing storage stem tissues known as tubers ( Hatley and Kappelman , 1980 ) . In primates , we could conduct a finer resolution analysis of amylase evolution and its relationship to diet than in non-primate mammals because for primates we have access to more data and samples . Specifically , we could investigate the amylase gene locus in 14 different primate species ( 53 DNA samples ) for gene copy number , and from 8 of these species , we also could obtain saliva ( 26 saliva samples ) for measuring enzymatic activity ( Figure 5 ) . We confirmed previous studies which documented a duplication of the amylase gene in the ancestral population of the catarrhini and an additional duplication in the ancestral population of the great apes ( Meisler and Ting , 1993 ) . Among Old World monkeys , we found further amylase gene copies in rhesus macaques , baboons , and vervets , all species which consume a broad-range diet . In contrast , we found that leaf-eating Old World monkeys ( colobus , snub-nose , and proboscis monkeys ) ( Hohmann , 2009 ) possess only one copy of the amylase gene , indicating a potential loss of a copy in this lineage . Most New World monkey genomes that we tested carry two haploid amylase copies . Assuming that the ancestral state of this lineage had one haploid copy , our results suggest yet another occurrence of gene copy number gain in the ancestor of New World monkeys . Moreover , we found an additional amylase gene copy in capuchins , which generally consume more starch than other New World monkeys ( Galetti and Pedroni , 1994; Rowe and Myers , 2016 ) . Next , we investigated lemurs , an outgroup primate species to monkeys and great apes , and found that they indeed only harbor one haploid copy of the amylase gene ( Figure 5 ) . Our result in the lemur lineage , combined with prior reports that ancestors of simians have a single amylase gene copy ( Samuelson et al . , 1996; Samuelson et al . , 1990 ) , suggests that primate ancestors possessed only one haploid copy of the amylase gene . Still , other scenarios , involving multiple amylase copies in the primate ancestor followed by gene loss , may also explain the observed copy number variation across the primate phylogeny , and cannot be dismissed at the present stage , although they are more complex in their assumptions . Clearly , more studies are needed to fully resolve the evolutionary history of the variation in this locus in primates , which , in addition to gene duplications , may have been shaped by gene conversion and complex rearrangements as well as incomplete lineage sorting . Next , we investigated whether variation in amylase gene copy numbers among primates translates into salivary expression in a similar way as we had shown for non-primate mammals . We found that several species of Old World monkeys , including rhesus macaques and baboons , express high enzymatic activity of amylase in their saliva ( Figure 5 ) ( Mau et al . , 2010 ) . These primates , belonging to the subfamily cercopithecinae , are known for their cheek pouches in which they store food for prolonged oral predigestion ( Lambert , 2017; Rahaman et al . , 1975 ) . Thus , this primate subfamily could be an exception in that salivary amylase may substantially participate in oral digestion of starch . Most New World monkeys do not consume starch in their regular diets . For example , marmosets primarily consume insects and plant exudate ( Rylands and Faria , 1993 ) , while owl monkeys consume flowers , insects , nectar , and leaves ( Rowe and Myers , 2016; Wright , 1994 ) . In agreement with their dietary habits , we found little to no salivary activity of amylase in these New World monkeys . Capuchin monkeys are an exception because they consume fruits , bulbs , and seeds ( Galetti and Pedroni , 1994; Rowe and Myers , 2016 ) . Accordingly , we discovered enzymatic activity levels of salivary amylase in capuchins that reach levels found in pigs and boars ( Figures 3C and 5 ) . Our results in primates document two additional instances ( cercopithecinae and capuchins ) where duplications of the amylase gene coincide with salivary expression . Combined , our results suggest that the evolution of the amylase locus in primates followed the same general trends observed for all mammals in that dietary strategies coincide both with amylase gene copy number and salivary expression in a lineage-specific manner . Our results reveal a staggering diversity of amylase gene copy numbers across extant mammals that consume starch . We report multiple bursts of amylase copy number gains that occurred independently in different branches of the mammalian phylogeny . Our results showed that each of these bursts coincided with expression of amylase in saliva . Our results also showed that phylogenetically distant species living in different habitats and consuming different diets have arrived at astonishingly similar amylase gene copy numbers , which correlate with the level of starch in their diet . Building on earlier models of the locus’ evolution ( Axelsson et al . , 2013; Perry et al . , 2007; Samuelson et al . , 1990 ) and using our own data , we deduce a model of how the amylase gene locus might have evolved across mammals ( Figure 6 ) . We posit here that the amylase locus evolved under the influence of the dietary context driven by the functional importance of amylase enzymatic activity in two digestive gland systems , namely the salivary glands , located at the entrance to the gastrointestinal tract , and the pancreas located further distally in the digestive continuum . Most evolutionary models agree that the ancestral mammalian amylase gene was expressed in the pancreas . It has been suggested that increase in gene copy number leads to higher amylase expression in the pancreas , which in turn allows rapid and effective intestinal digestion of starch in species consuming a higher amount of that food component in their diet ( Axelsson et al . , 2013 ) . Based on our findings , we propose here that at least one gene duplication is required for amylase to gain expression in salivary glands . We argue that this is a form of neofunctionalization ( Ohno S , 1970 ) where an otherwise intact gene copy acquires mutations in its regulatory architecture , thus leading to expression in a new tissue , in this case the salivary gland system . We hypothesize that one or more gene duplication events of this copy are needed to gain a level of enzymatic activity in saliva optimally suited to accommodate starch predigestion in the mouth environment . As illustrated in Figure 6 , we further propose that there are different ways to achieve a given level of salivary amylase activity , depending on the strength of the regulatory element associated with the salivary gland specific gene copy . For instance , muroidea and cercopithecinae possess only a single salivary-gland-specific gene copy but arrive at high levels of amylase activity in their saliva . In this case , we surmise that a strong regulatory element is associated with that single copy . In addition to that copy , muroidea possess multiple other amylase gene copies that are not expressed in salivary gland tissue . It can be assumed that these gene copies are expressed in pancreatic tissue and may increase enzymatic activity there in the more distal parts of the digestive tract . Dogs and pigs have numbers of amylase gene copies comparable to humans and some muroidea but show much less amylase activity in their saliva than these species . Given the data , we are not able to say which of these copies are expressed in pancreas and which in salivary glands . Regardless , those copies that might be expressed in salivary glands must be associated with weak regulatory architectures . In humans , the number of amylase gene copy numbers was shown to correlate with levels of amylase expression in saliva ( Bank et al . , 1992; Perry et al . , 2007 ) . Our data show that across all mammalian species such a simple correlation does not hold true . However , within species possessing multiple and variable copy numbers there might be a correlation of copy number with salivary expression . In this regard , we could show a correlation of amylase gene copy number and salivary enzyme activity for dog ( R2 = 0 . 45 ) and pig ( R2 = 0 . 69 ) . A comparative study investigating expression of individual amylase gene copies in pancreas versus salivary gland system will be the logical next step to delineate the mechanisms through which gland-specificity of amylase has evolved in mammals . It is of particular interest to simultaneously elucidate the role of lineage-specific retrotransposons or other regulatory elements in modulating tissue-specific expression of amylase . From a broader ecological perspective , we showed that amylase gene copy numbers generally correlate with a broad-range diet and with high versus low-starch consumption . However , salivary amylase activity is only correlated with broad-range diet but not with the amount of starch . As such , a simple explanation solely based on the digestive function of amylase cannot fully explain why some mammals , including humans , express so much amylase in their saliva . As a matter of fact , most mammalian species do not keep food long enough in their mouth for salivary amylase to substantially participate in starch digestion . We argue here that in most mammalian species the role of amylase in saliva may be to liberate oligomeric sugar molecules from polymeric starch chains that can then be perceived by sweet-taste receptors in the oral cavity . Indeed , studies found links between salivary amylase and taste perception ( Mandel et al . , 2010; Sclafani et al . , 1987 ) . Being able to perceive otherwise tasteless starch in their diet might confer an adaptive advantage to species consuming broad-range diets by enabling them to detect high caloric ( i . e . starch-containing ) , food components . Lastly , putatively adaptive benefits of amylase expression in saliva depend on the ecological and behavioral context for any given species . An exceptional example are the cheek-pouched cercopithecinae where putative fitness advantage of salivary amylase expression goes beyond taste perception . In these species , which conduct almost half of their starch digestion in the oral cavity ( Janiak , 2016 ) , salivary amylase may have evolved to substantially participate in the overall digestion of dietary starch , a role executed by pancreatic amylase in most other species . In addition to gustatory and digestive functions , salivary amylase may also be involved in the regulation of metabolic glucose homeostasis ( Peyrot des Gachons and Breslin , 2016 ) as well as associated with bacterial composition in the oral cavity ( Davenport , 2017; Scannapieco et al . , 1989 ) or in the gut ( Poole et al . , 2019 ) . In that regard it will be interesting to find out whether different amylase gene copies encode proteins of slightly different functional activity due to differences in DNA sequence , differential RNA splicing , and post-translational modifications , including glycosylation . Overall our present study highlights the potential role of amylase enzymatic activity in saliva in shaping food preference and niche partitioning among omnivorous starch-consuming mammals , possibly in coevolution with the oral microbiome .
We chose our panel of mammalian species based on their phylogeny , diet preference ( broad vs . specialized ) , domestication status , and commensal relationship with humans . Overall , we compiled 204 DNA samples from 46 different species and 127 saliva samples from 22 different species . Detailed information about the samples used in this study and their sources can be found in Supplementary file 1 . Briefly , DNA from various animals was collected using buccal swabs ( PurFlock , Puritan Medical Products ) , saliva samples , or museum specimens ( dried blood and tissues , Museum of Southwestern Biology , Division of Genomics Resources ) . Saliva samples were collected by suction using commercially available devices containing absorbent sponges in a syringe-like receptacle ( Super-SAL and Micro-SAL , Oasis Diagnostics , Vancouver , WA ) unless otherwise specified . DNA was extracted from swabs using a commercially available kit ( ChargeSwitch gDNA Buccal Cell Kit , Invitrogen ) . For saliva samples , we used a commercially available extraction kit ( BioWorld , Dublin , OH ) . Detailed information about sampling strategies employed for each species can be found in Collection of saliva samples section below . Digital droplet PCR ( ddPCR ) was used to experimentally determine amylase gene copy numbers . If reference genomes were available for a given species , primers were designed specifically for use in these species . For species where reference genomes were unavailable , amylase coding sequences were chosen for primer design that were confirmed to be conserved in the two most closely related species . Further details about primer design and strategy are described in Primer design for digital PCR section below . The primer sets used for each species are listed in Supplementary file 2 . Translated amino acid sequences of the amylase gene copies were downloaded from NCBI reference genomes . Sequences were aligned and a phylogenetic output was generated using a custom Python code as described previously ( https://github . com/duoduoo/VCFtoTree ) ( Xu et al . , 2017; Pajic et al . , 2016 ) . We constructed a phylogenetic tree from the protein sequences by Randomized Axelerated Maximum Likelihood ( RAxML ) ( Stamatakis , 2014 ) using the LG substitution model ( Le and Gascuel , 2008 ) , bootstrapping it with 1000 replicates for branch support . Visualization was performed using the FigTree software ( Rambaut , 2012 ) . Previous work utilized lineage-specific retrotransposons to estimate the timing of amylase gene duplications and to distinguish between salivary and pancreatic amylase genes in humans ( Samuelson et al . , 1990 ) . Using this approach , the salivary AMY1 gene could be traced back to a great ape ancestor ( Samuelson et al . , 1990 ) . Later studies used the sequence of the great-ape specific retrotransposon to label AMY1 gene copies in humans by fiber-FISH ( Perry et al . , 2007 ) . Building upon these findings , we searched 5 kb upstream and downstream of the amylase copies in mouse , rat , pig , and dog reference genomes for the existence of lineage-specific retrotransposition markers . Specifically , we searched for relatively recent L1 elements having sW scores of more than 1000 and being located at nearly identical distances to the 5’ or 3’ ends of amylase gene copies . Using this approach , we detected a relatively small number ( less than 10 ) of distinct L1 retrotransposons in each species ( Figure 2B ) . To ensure that these retrotransposons were duplicated specifically in the amylase locus , we conducted a BLAST analysis to search for the existence of these same retrotransposons outside of the amylase locus . We found that the retrotransposons within the amylase locus are highly similar to each other ( >90% ) and we found no similarly close matches for these retrotransposons elsewhere in the reference genomes of these species ( Supplementary file 3 ) . Next , we verified that these retrotransposons were indeed lineage-specific by showing that there were no close matches in reference genomes of other species . The most parsimonious explanation for our observations is that these L1 elements inserted into the proximity of an amylase gene copy and then duplicated along with additional copies of that gene , thereby suggesting lineage-specificity of duplication events . We used two different methods to measure enzyme activity of amylase in saliva . First , we conducted a direct estimate of enzyme activity using a traditional starch lysis agar plate assay following a previously described protocol ( Kilian and Nyvad , 1990 ) . In brief , holes were punched in a starch-containing agar and filled with saliva . After 24 hr incubation at 37°C , the undigested starch remaining in the agar was stained with iodine and the diameters of the lysed clear rings were measured . Enzymatic activity was extrapolated from serial standard dilutions of purified α-amylase from human saliva ( Sigma ) measured in the same assay . In parallel , we measured the samples using a high-sensitivity ( detection limit 2 mU/ml ) colorimetric in-solution assay ( EnzCheck Ultra Amylase Assay Kit , Invitrogen ) following the manufacturer’s protocol with the same human α-amylase as the standard . Concentrations of total protein in saliva were determined by the bicinchoninic acid ( BCA ) assay ( micro-BCA , BioRad ) using bovine serum albumin as the standard . Optical density measurements were performed using a Nanodrop 2000 spectrophotometer ( Thermo Fisher ) . Amylase activities were calculated as units of enzymatic activity normalized per mg of total salivary protein . All input data used for creating the main figures are provided in Supplementary file 1 . Information about the dietary preferences of individual species was acquired from extensive literature research presented in Categorization of starch consumption section below . All figures were produced using the R statistical package ( https://www . r-project . org/ ) . For calculating the independent phylogenetic contrasts shown in Figure 4C , we used the approach outlined by Felsenstein ( 1985 ) . For this analysis , we used the subset of species available through the Hg19 100way conservation alignment ( http://genomewiki . ucsc . edu/index . php/Hg19_100way_conservation_alignment ) . Using the phylogenetic distance provided in this dataset , we first normalized the differences in amylase gene copy number and salivary enzyme activity between any two species by the square roots of the phylogenetic distance between them . Using these normalized values and applying the non-parametric Kolmogorov–Smirnov test , we tested the null hypothesis that the phylogenetic contrasts between species consuming a specialized diet is not different from the phylogenetic contrasts between species consuming the different types of diet . Among the species consuming a broad-range diet , we further tested that the phylogenetic contrasts between high and moderate starch consuming species are not different from those between species consuming moderate levels of starch . Saliva samples and buccal swabs from deer mice ( Peromyscus spp . ) were provided by Danielle Garneau ( SUNY Plattsburgh ) . Mice were trapped in the wild by Sherman live traps ( Garneau et al . , 2012 ) . After restraint by scruffing behind the neck , a glass capillary tube was introduced to the animal's mouth and was moved about the lower lip and cheeks to collect saliva . The tube was introduced at an angle such that gravity would help draw down the sample into the tube . The capillary tube was placed in an Eppendorf tube and a pipet pump was used to force air to drive the rest of the sample from the capillary tube into the Eppendorf tube for storage at −20°C and shipment on dry ice . Saliva from house mice ( laboratory strain C57BL10/SNJ ) was kindly provided by Jill Kramer ( University at Buffalo ) using a collection procedure as previously described ( Kiripolsky et al . , 2017 ) . Saliva from woodrats was kindly provided by Michelle Skopec ( Weber State University ) . To collect saliva , woodrats were scruffed and Micro-Sal collection ( Oasis Diagnostics , Vancouver , WA ) devices were placed in their mouths . The woodrats were allowed to chew on the absorbent sponge part of the device and , then , their tongues and cheeks were swabbed to retrieve residual saliva . Collection devices were centrifuged and saliva samples were stored at −20°C before shipment on dry ice . Saliva from Long Evans hooded rats was kindly provided by Ann-Marie Torregrossa ( University at Buffalo ) . As described previously ( Martin et al . , 2018; Torregrossa et al . , 2014 ) , rats were conditioned to salivate when a pipette was inserted into the mouth and saliva was collected . Approximately 50 µl saliva was retrieved by suction from below and around the tongue where it pools naturally . Saliva from dogs , cows , sheep , goats , horses , pigs , and giant African pouched rats was provided by Erin Daugherity and Luce E . Guanzini ( Cornell University ) . Animals were not allowed to eat or drink prior to the collection to ensure the oral cavity was free of food and other debris . Saliva from giant African pouched rats was collected opportunistically while animals were anesthetized for an unrelated clinical procedure . The collection was performed using a commercially available device ( Micro-Sal , Oasis Diagnostics ) . Large animals were gently restrained and a larger collection device ( Super-Sal , Oasis Diagnostics ) was placed under the tongue for up to three minutes , or until fully soaked . Devices were stored at −20°C before shipping on dry ice . Saliva from female wild boars and castrated domestic pigs was provided by Anja Globig ( Friedrich-Loeffler-Institut , Insel Riems - Greifswald , Germany ) . For the collection of saliva a commercial collection device , consisting of an absorbent cotton swab in a tube , was used ( Salivette , Sarstedt , Nümbrecht , Germany ) . The swab was inserted in the animal’s mouth and fixated with a forceps until it was drenched with saliva . After placing the swab back in the tube , saliva was extracted by centrifugation . Samples were lyophilized before international shipping . Saliva from wolves was kindly provided by Karen Davis ( Wolf Park , Battle Town , IN ) . The wolves housed in this facility are well socialized , which allowed saliva collection by inserting Super-Sal ( Oasis Diagnostics ) devices into the mouths of adult wolves willing to participate . Swabs were kept in the animals’ mouths as long as they would tolerate it or until fully soaked . Samples from juvenile wolves could be collected while they were resting by inserting the swabs into their mouths . Samples were stored at −20°C before shipment on dry ice . Saliva from dogs was kindly provided by Barbara McCabe ( Buffalo , NY ) . Samples were obtained from diverse breeds of dogs including Boxers , Pitbulls , Golden Retrievers , and Labradors , along with several mixed breeds ( see Supplementary file 1 for details ) . Super-Sal devices ( Oasis Diagnostics ) were placed in the mouth of dogs for 1–5 min , or until swab was damp . The swabs were stored at −20°C until transfer to our laboratory . Saliva from Ring-tailed Lemur samples was kindly provided by Erin Ehmke ( Duke Lemur Center ) . Samples were collected using commercially available absorbent strips ( SalivaBio Children's Swabs , Salimetrics , Carlsbad , CA ) . Saliva-soaked swabs were immediately centrifuged and the collected saliva was frozen at −80°C and shipped on dry ice . Saliva from humans was collected by passive drooling following the protocol approved by the University at Buffalo Human Subjects IRB board ( study # 030–505616 ) . Informed consent was obtained from all human participants . Saliva from chimpanzees and gorillas was collected in a noninvasive manner following the protocol approved by the University at Buffalo IACUC committee ( IACUC ID# AR201800024 ) . Chimpanzees were trained by the caretaker to voluntarily expectorate into a plastic cup . Gorilla ( Western lowland gorilla ) saliva was collected by the animal caretakers with a soft disposable plastic Pasteur pipette ( VWR , Radnor , PA ) from individuals who were previously trained to open their mouth upon request . Saliva from Rhesus macaques was provided by the Southwest National Primate Research Center , San Antonio , TX , and by the Yerkes National Primate Research Center , Atlanta , GA . All samples were immediately transferred into a polypropylene tube chilled on ice . Aliquots were stored at −80°C and shipped on dry ice . All input data used for creating the main figures are provided in Supplementary file 1 . Information about the dietary preferences of individual species was acquired from the Michigan Animal Diversity Web ( https://animaldiversity . org/ ) , unless other studies were cited . With regard to starch consumption , the literature was limited . Thus , we undertook the following steps to construct a categorization of starch consumption among the species that we used in our analysis presented in Figure 4A and B . Based on the information available on Michigan Animal Diversity Web , we first identified animals with specialized diets ( carnivores: cat , polar bear , cougar , and wolf; herbivores: sheep , bison , snow sheep , cow , goat , horse , bighorn sheep , ibex , yak , wild goat , zebra , sheep , and donkey ) . We assumed that starch makes up a negligible percentage of these animals’ diet . For the animals with broad-range diets , which presumably have considerable starch content in their diet , we conducted a wider literature research . Most information about starch consumption is available for present-day human populations ( Bright-See and Jazmaji , 1991 ) , and it has been suggested that humans consume a higher percentage of starch in their diet than great apes ( Perry et al . , 2007; Zohary et al . , 2012 ) . Indeed , chimpanzees and bonobos primarily consume ripe fruits ( poor in starch ) , and in the scarcity of ripe fruits , they prefer piths ( also poor in starch ) ( Hohmann , 2009; Wrangham et al . , 1998 ) . Orangutans and gorillas primarily consume ripe fruit and leaves ( Hohmann , 2009 ) . However , it was suggested that they also consume seeds and cambium , an observation that led Janiak ( Janiak , 2016 ) to argue that gorillas and orangutans have a relatively higher starch content in their diets than chimpanzees and bonobos . Old World monkeys show remarkable diversity in their diets . Specifically , cercopithecines ( represented in our dataset by baboons , rhesus macaques , vervet monkeys , green monkeys , and Allen’s swamp monkeys ) consume considerable amounts of unripened fruit ( higher starch content [Lambert , 1998] ) , especially when no ripe fruit is available ( Mau et al . , 2010; Wrangham et al . , 1998 ) . In contrast , most colobinae ( represented by the colobus monkeys in our dataset ) are primarily leaf-eating and , thus , likely have little starch in their diet ( Oates , 1994 ) . New World monkeys ( represented by capuchin monkeys , owl monkeys , and marmosets in our dataset ) also show diversity in their likely starch consumption . Capuchin and owl monkeys primarily consume fruits , even though they supplement their diets with flower foraging and insects ( Lambert , 2017; Kinzey , 1997 ) , which likely indicates starch consumption similar to chimpanzee and bonobos . Marmosets differ in their dietary habits as their primary food intake comprises gum and other exudates ( which are not starch sources ) from various trees and vines , and scarcely involve fruits ( Soini , 1982 ) . Lemurs ( represented by ring-tailed lemur in our dataset ) have been reported to primarily consume Tamarind fruit , which is rich in non-starch polysaccharides ( Gould et al . , 2003 ) . Overall , primates depend on a wide variety of food sources , including starch-based foods . However , humans are the only primate species consuming a diet unusually high in starch content . As for rodents , we first considered species which are primarily human-commensal ( represented by house mice as well as by brown and black rats in our dataset ) . These species have considerable variation in their diets depending on the ecological context they are living in . For example in the wild , house mice have been reported to eat primarily insects and seeds , the latter containing significant amounts of starch ( Badan , 1986; Roux et al . , 2002 ) . However , house mice normally live in human-influenced habitats and consume agricultural grains and other starch-rich human-produced food and human food leftovers ( Clark , 1982; Gardner-Santana et al . , 2009; Hulme-Beaman et al . , 2016; Pocock et al . , 2004; Schein and Orgain , 1953; Singleton et al . , 2003 ) . Other rodents with less human-commensal interactions ( represented by birch mouse , deer mouse , pouched rat , and woodrat in our study ) also consume diverse diets , including insects , seeds , grains , and flowers ( Ajayi , 1977; Baker , 1991; Everett et al . , 1978; Juskaitis , 2000 ) . However , their access to starch-rich grains and seeds is seasonal , and these foods are not necessarily their primary caloric source . The other mammals with broad-range diet in our dataset were dogs , pigs , boars , and bears . Brown and black bears eat a wide range of foods including leaves , fruits , grains , as well as meat from hunting or scavenging ( Bojarska and Selva , 2012; Graber and White , 1983; Torgersen et al . , 2001 ) . However , the same studies documented that starch-containing foods , such as grains , make up only a small portion of the bears’ diet . Boars also have a diverse preference in their diet , including mushrooms , roots , fruits , and insects . However , unlike bears , boar diets include substantial amounts of roots and tubers ( Baubet et al . , 2004; Massei and Genov , 2004 ) and , if available , human agricultural crops ( Herrero et al . , 2006 ) . A close relative of the boar , the domesticated pig thrives primarily on human-produced starch-rich food sources , such as corn or potatoes . Overall , pigs and boars have higher starch content in their diets than bears . In fact , their diet can be comparable to early human diets ( Hatley and Kappelman , 1980; Miller and Ullrey , 1987 ) . Another mammal consuming a broad-range diet that was included in our study was the dog , which has been discussed within the context of recent adaptation to human-derived starch-rich diets ( Arendt et al . , 2016; Axelsson et al . , 2013 ) . Based on this literature , we presume that dogs , pigs , and boars have higher starch content in their diet , while starch makes up a smaller portion of the diet of the bears . Overall , all the mammals consuming a broad-range diet mostly have considerable levels of starch in their diet . However , our literature search indicates that humans , pigs , boars , dogs , mice , and rats ( both brown and black ) stand out in that their diet predominantly depends on starch-rich foods ( grains , roots , and tubers ) . Thus , we grouped them under the ‘higher starch’ consuming category whereas we grouped the other species under the ‘lower starch’ consuming category ( Figure 4A and B ) . For digital droplet PCR experiments , we used two primer/probe sets . One targeted the amylase copies and the other targeted a conserved ‘reference’ sequence , which was found to be a single haploid copy in mammals with known reference genomes ( SRSF7 gene ) . For the reference sequence , we used a primer/probe set that targets one of the exons of the SRSF7 gene . The sequence is highly conserved across species and unique ( i . e . a single haploid copy ) in all mammalian reference genomes we investigated . We have checked that the sequence is 100% conserved in species that we considered and for which reference genomes were available from the UCSC genome portal . To capture as many amylase gene copies as possible , we carefully designed primers and probes for each species where a reference genome was available to match ( 100% as assessed by BLAST alignment ) all of the reference amylase copies . Primer and probe sequences are listed in Supplementary file 2 . It is possible that we underestimated or missed some of the amylase copies that are not represented in the reference genomes . However , digital PCR is robust to 1–2 mismatches and in most species , ddPCR results were highly concordant with copy number estimations based on BLASTx and BLASTp analysis ( Figure 1—figure supplement 1 , Supplementary file 2 ) . Therefore , we surmise that the main trends we observed in this study are reliable . To decide which primer/probe sets to use for species where no reference genome was available , we designed primer/probe sets that work in the phylogenetically most closely related species for which reference genomes were available . For example , for zebra , we used a primer/probe set designed for the horse reference genome . To ensure that this primer/probe set was appropriate , we first made sure that it also worked in the donkey reference genome . As such , we surmised that our approach should work unless there is rapid , species-specific sequence divergence in zebra as compared to horse and donkey . An analogous approach was used for all the other species for which reference genomes were not available ( Figure 1—figure supplement 1 , Supplementary file 2 , for substitute species genomes chosen ) . Of course , this approach might be prone to undercalling the number of amylase gene copies in species where reference genomes are not available . Although we are confident about these estimates , none of the major conclusions of this study depends on data from such species . | Many mammals can digest starch by using an enzyme called amylase , but different species eat different amounts of starchy foods . Amylase is released by the pancreas , and in certain species such as humans , it is also created by the glands that produce saliva , allowing the enzyme to be present in the mouth . There , amylase can start to break down starch , releasing a sweet taste that helps the animal to detect starchy foods . Curiously , humans have multiple copies of the gene that codes for the enzyme , but the exact number varies between people . Previous research has found that populations with more copies also eat more starch; if this correlation also existed in other species , it could help to understand how diets influence and shape genetic information . In addition , it is unclear how amylase came to be present in saliva , as the ancestors of mammals only produced the protein in the pancreas . Pajic et al . analyzed the genomes of a range of mammals and found that the more starch a species had in its diet , the more amylase gene copies it harbored in its genome . In fact , unrelated mammals living in different habitats and eating different types of food have similar numbers of amylase gene copies if they have the same level of starch in their diet . In addition , Pajic et al . discovered that animals such as mice , rats , pigs and dogs , which have lived in close contact with people for thousands of years , quickly adapted to the large amount of starch present in human food . In each of these species , a mechanism called gene duplication independently created new copies of the amylase gene . This could represent the first step towards some of these copies becoming active in the glands that release saliva . In people , having fewer copies of the amylase gene could mean they have a higher risk for diabetes; this number is also tied to the composition of the collection of bacteria that live in the mouth and the gut . Understanding how the copy number of the amylase gene affects biology will help to grasp how it also affects health and wellbeing , in humans and in our four-legged companions . | [
"Abstract",
"Introduction",
"Results",
"and",
"discussion",
"Materials",
"and",
"methods"
] | [
"evolutionary",
"biology",
"genetics",
"and",
"genomics"
] | 2019 | Independent amylase gene copy number bursts correlate with dietary preferences in mammals |
Many spinal circuits dedicated to locomotor control have been identified in the developing zebrafish . How these circuits operate together to generate the various swimming movements during development remains to be clarified . In this study , we iteratively built models of developing zebrafish spinal circuits coupled to simplified musculoskeletal models that reproduce coiling and swimming movements . The neurons of the models were based upon morphologically or genetically identified populations in the developing zebrafish spinal cord . We simulated intact spinal circuits as well as circuits with silenced neurons or altered synaptic transmission to better understand the role of specific spinal neurons . Analysis of firing patterns and phase relationships helped to identify possible mechanisms underlying the locomotor movements of developing zebrafish . Notably , our simulations demonstrated how the site and the operation of rhythm generation could transition between coiling and swimming . The simulations also underlined the importance of contralateral excitation to multiple tail beats . They allowed us to estimate the sensitivity of spinal locomotor networks to motor command amplitude , synaptic weights , length of ascending and descending axons , and firing behavior . These models will serve as valuable tools to test and further understand the operation of spinal circuits for locomotion .
Movements made in the early stages of development can be critical for the survival of many species . The escape response seen in various fish and amphibians is one such example of a vital movement present at early developmental stages ( Domenici and Hale , 2019 ) . However , the nervous system’s control of movement does not come fully formed but matures as the nervous system develops ( Favero et al . , 2014 ) . This maturation enables a broader repertoire of movements to arise . During this process , new neurons are born and subsequently integrated into neural circuits that are newly formed or refined , presumably leading to the emergence of progressively more coordinated and skillful maneuvers . Determining how the assembly of new circuits leads to the emergence of new movements can provide valuable insights into the role of distinct neurons or circuits in motor control . The maturation of swimming in developing zebrafish has been well described at both the ethological and the cellular levels ( Drapeau et al . , 2002; McLean and Fetcho , 2009 ) . Single strong body bends on one side of the body , also known as coils , emerge during the first day of development at around 17 hr post-fertilization ( hpf ) as the earliest locomotor behavior ( Saint-Amant and Drapeau , 1998 ) . Single coils are quickly followed by double coils ( i . e . , two successive coils , one for each side of the body ) at around 24 hpf ( Knogler et al . , 2014 ) . Touch-evoked swimming appears around 27 hpf as coiling begins to subside . Spontaneous swimming movements emerge around 2–3 days post-fertilization ( dpf ) ( Saint-Amant , 2010 ) . The first swimming movement zebrafish exhibit is burst swimming characterized by long ( 1 s long ) but infrequent episodes of tail beats . Burst swimming is then replaced by beat-and-glide swimming characterized by shorter ( several hundreds of milliseconds long ) but more frequent episodes . In both cases , swim episodes consist of repetitive left-right alternating , low-amplitude tail beats that propagate from the rostral toward the caudal end of the fish body and are generated at 20 to 80 Hz ( Budick and O'Malley , 2000; Buss and Drapeau , 2001 ) . During this rapid series of transitions between locomotor maneuvers , populations of spinal neurons are progressively generated , starting with primary motoneurons ( MNs ) at about 9 hpf . Subsequently , spinal MNs and interneurons ( INs ) are generated in stereotyped spatiotemporal birth orders ( Kimmel et al . , 1994; Myers et al . , 1986; Satou et al . , 2012 ) . Two successive waves of axogenesis occur in the embryonic spinal cord ( Bernhardt et al . , 1990 ) . The first wave occurs around 16–17 hpf . It includes axon growth in primary MNs that innervate red and white muscle fibers at early developmental stages ( Buss and Drapeau , 2000 ) . Primary MNs enable coiling and escape movements ( Kimmel et al . , 1995; Saint-Amant and Drapeau , 2000 ) . Several spinal INs that are also important for early movements extend their axons along with primary MNs . These include Ipsilateral Caudal ( IC ) INs that are thought to play an essential role in driving the rhythm of early locomotor behavior due to their endogenous bursting activity ( Tong and McDearmid , 2012 ) . The second wave of axon growth occurs at around 23–25 hpf . It involves axon growth in secondary MNs involved with slower movements ( Liu and Westerfield , 1988 ) and spinal IN populations that include excitatory and inhibitory , ipsilaterally and contralaterally , and ascending and descending projecting subtypes ( Bernhardt et al . , 1990; Higashijima et al . , 2004 ) . The progressive generation of new populations of spinal neurons and continued axonal growth coincides with the expansion of the zebrafish locomotor repertoire . This timing suggests that incorporating spinal circuits into existing locomotor circuits underlies the acquisition of novel locomotor maneuvers . We have recently provided evidence that the maturation from coiling to later stages of swimming is accompanied by an operational switch in how spinal locomotor circuits generate the rhythm underlying tail beats . Specifically , we demonstrated that spinal circuits transitioned from relying upon pacemakers with endogenous bursting properties during coiling toward depending upon network oscillators whose rhythm is driven by excitatory and inhibitory synapses ( Roussel et al . , 2020 ) . In light of these and earlier findings describing the composition and maturation of spinal locomotor circuits , we sought to generate computational models that replicate developmental locomotor movements of the zebrafish . We iteratively constructed models for several locomotor movements by incorporating specific spinal populations , shifts in relative connection strength , and changes in the firing behavior of neurons . While computational modeling has generated invaluable insights into the function and mechanisms of spinal locomotor circuits of several species ( Ausborn et al . , 2019; Bicanski et al . , 2013; Danner et al . , 2019; Ferrario et al . , 2018; Hull et al . , 2016; Kozlov et al . , 2014; Sautois et al . , 2007 ) , there is to our knowledge no such model for the developing zebrafish spinal cord . Here , we build some of the first computational models of the zebrafish spinal locomotor circuit that can accurately reproduce predominant locomotor behaviors during early zebrafish development . In the process , we test theories about the possible contributions of specific neural circuits and spinal populations to locomotor movements in zebrafish and identify untested hypotheses on the operation of spinal locomotor networks in developing zebrafish .
Coiling , which is already observed at 1 dpf , is characterized by a single strong , slow ( hundreds of milliseconds in duration ) tail beat on one side of the body followed by a return to resting position ( Saint-Amant and Drapeau , 1998 ) . Coiling events are relatively infrequent , reaching a maximum frequency of 1 Hz around 20 hpf ( Saint-Amant and Drapeau , 1998 ) . Previous studies have established that this behavior is generated by a spinal circuit relying primarily on gap junctions ( i . e . , electrical synapses ) ( Saint-Amant and Drapeau , 2001 ) . It has been proposed that rostrally located IC pacemaker spinal neurons ( Tong and McDearmid , 2012 ) drive periodic depolarizations ( PDs ) of ipsilateral MNs via electrical synapses ( Drapeau et al . , 2002; Saint-Amant and Drapeau , 2001 ) . Glycinergic synaptic bursts ( SBs ) are observed in MNs during contralateral coiling events ( Saint-Amant and Drapeau , 2001 ) . These SBs have been proposed to arise from contralaterally projecting glycinergic neurons ( Saint-Amant and Drapeau , 2000 ) but are not responsible for any action potential firings or coiling movements ( Saint-Amant and Drapeau , 2001 ) . Applying a gap junction blocker , heptanol , but not glutamatergic and glycinergic antagonists , suppressed spinal activity responsible for coiling ( Saint-Amant and Drapeau , 2000 ) . After single coils appear , double coils emerge as a transitory locomotor behavior at around 24 hpf , coexisting with the single coiling behavior ( Knogler et al . , 2014 ) . Double coiling is characterized by two successive coils , one on each side of the body , and lasts about 1 s ( Knogler et al . , 2014 ) . Eventually , double coiling becomes the predominant coiling behavior . Double coiling can represent nearly three-quarters of all coiling events at its peak frequency , with the rest mainly being single coils ( Knogler et al . , 2014 ) . At the stage when double coiling appears ( 24 dpf ) , the previous electrical scaffold for single coils seems to be supplemented with chemical glutamatergic synapses to form a hybrid electrical-chemical circuit ( Knogler et al . , 2014 ) . Blocking glutamatergic transmission precludes double coils while sparing single coils ( Knogler et al . , 2014 ) . In contrast , blocking glycinergic synapses led to triple or even quadruple coils ( Knogler et al . , 2014 ) . These experimental observations suggest that synaptic excitation is required for successive coils after a first coil . Glycinergic transmission seems to prevent the generation of more than two successive coils . Patch-clamp recordings of MNs at this developmental stage exhibit the same isolated PDs and SBs from earlier developmental stages and show mixed events in which a PD event immediately follows an SB or vice-versa ( Knogler et al . , 2014 ) . Interestingly , the application of CNQX eliminates mixed PD-SB events but not single isolated SBs , suggesting that the coupling of PD and SB in mixed events is glutamatergic ( Knogler et al . , 2014 ) . Therefore , we aimed to generate a model with the following characteristics: ( 1 ) double coils lasting about 1 s in duration accounting for over half of the coiling events , ( 2 ) a dependence of double coiling upon excitatory synaptic transmission , ( 3 ) an increase in multiple coiling events in the absence of inhibitory synaptic transmission , and ( 4 ) the presence of mixed PD-SB events with similar sensitivity to the blockade of excitatory synaptic transmission as double coils . Around 2 or 3 dpf , zebrafish transition from coiling movements to swimming ( Drapeau et al . , 2002; Saint-Amant and Drapeau , 1998 ) . This transition entails two fundamental changes in locomotor movements: first , long , slow coils are replaced by quick , short tail beats; and second , the number of consecutive tail beats are increased from the two side-to-side coilings seen in double coils to multiple consecutive tail beats that compose each swimming episode . One of the emerging swimming movements is beat-and-glide swimming , characterized by short swimming episodes lasting several hundreds of milliseconds separated by gliding pauses and lasting several hundreds of milliseconds ( Budick and O'Malley , 2000; Buss and Drapeau , 2001 ) . Swim episodes consist of repetitive left-right alternating , low-amplitude tail beats that propagate from the rostral toward the caudal end of the fish body and are generated approximately at 20–65 Hz ( Budick and O'Malley , 2000; Buss and Drapeau , 2001 ) . Beat-and-glide swimming can be produced in isolated larval zebrafish spinal cord preparations by NMDA application ( Lambert et al . , 2012; McDearmid and Drapeau , 2006; Wiggin et al . , 2012 ) or by optogenetic stimulation of excitatory spinal neurons ( Wahlstrom-Helgren et al . , 2019 ) . This capacity suggests that the transition from coiling to swimming involves a delegation of rhythm generation from ICs to spinal locomotor circuits ( Roussel et al . , 2020 ) . Therefore , we sought to model a spinal network that could generate beat-and-glide swimming activity hallmarks—swim episodes lasting about 200–300 ms with repeated left-right alternating low-amplitude tail beats at around 20–65 Hz—without relying on pacemaker cells . Recent experimental studies have also started to delineate the contributions of specific populations of spinal neurons to swimming . Ablation of ipsilaterally projecting , excitatory neurons in the V2a population eliminates swimming activity ( Eklöf-Ljunggren et al . , 2012 ) . Genetic ablation of ipsilaterally projecting , inhibitory neurons in the V1 population affects swim vigor but has no effects on the patterning of swimming ( Kimura and Higashijima , 2019 ) . Genetic ablation of a subset of commissural inhibitory neurons in the dI6 population reduces left-right alternation ( Satou et al . , 2020 ) . We sought to replicate the role of these neurons in our model .
The earliest locomotor behaviors in zebrafish , namely single and multiple coilings , require global recruitments of neurons to synchronously contract all ipsilateral muscles ( Warp et al . , 2012 ) . Electrical coupling , which lacks the delays inherent with chemical neurotransmission , enables these types of ballistic movements . Early locomotor behavior in zebrafish seems to rely on this architecture , as demonstrated by the necessity of electrical but not chemical synapses ( Saint-Amant and Drapeau , 2000; Saint-Amant and Drapeau , 2001 ) . The rapid and multidirectional current transmission supported by electrical synapses is a perfect solution for en masse activation of a neural circuit ( Bennett and Zukin , 2004 ) . However , synchronous activation of an ensemble of neurons does not accommodate rhythmic activity , which requires more precise timing and connection strength . For example , the emergence of double coiling in our model was generated by chemical synaptic excitation of contralateral pacemaker neurons that had to be sufficiently delayed to enable the first coil to complete before initiating the second contralateral coil . Commissural glycinergic transmission was also required to tamper down coiling events with more than two successive coils . Suppose multiple coiling is a preparatory stage toward the emergence of repetitive , left-right alternating tail beats . In that case , the possible importance of contralateral excitation and inhibition at this stage presages the establishment of similar operational mechanisms to the generation of swimming . To generate swimming , we delegated the generation of the rhythm driving tail beats to network oscillators distributed along the length of the spinal cord . Spinal locomotor circuits may transition away from pacemakers as the source of the rhythm to prevent being vulnerable to any flaws in the function of a small population of neurons . Also , there may be multiple local rhythms that control body oscillations along the developing zebrafish's length . Indeed , locomotor output has proven to be very robust to the sectioning of the spinal cord , leading to the suggestion that redundant rhythm-generating circuits must be present within the spinal cord ( McDearmid and Drapeau , 2006; Wiggin et al . , 2012; Wiggin et al . , 2014 ) . Experimental evidence from our lab further suggests that a transition from a rhythm driven by a pacemaker kernel to a rhythm driven by local network oscillators occurs progressively from the caudal toward the rostral end of the body ( Roussel et al . , 2020 ) . The V2as are well recognized as the neural engine that drives swimming activity in zebrafish spinal circuits ( Eklöf-Ljunggren et al . , 2012; Ljunggren et al . , 2014 ) . While some V2a INs have shown intrinsic burst firing in the adult zebrafish ( Song et al . , 2018; Song et al . , 2020 ) , V2a INs in developing zebrafish show either tonic or modestly spike adapting firing ( Menelaou and McLean , 2019 ) . We thus sought to generate beat-and-glide swimming with tonically firing V2as . Successive left-right alternating tail beats were generated by combining contralateral excitation from bursting commissural excitatory neurons to initiate alternating tail beats and contralateral inhibition to prevent co-contraction of both sides . In fact , a simulation with only tonic firing neurons could also generate beat-and-glide swimming over several seconds . Thus , V2as could very well drive rhythmic tail beats in larval zebrafish while firing tonically . If this is the case , then the central role of V2as depends less on their ability to produce a bursting rhythm . Instead , the pivotal role of V2as in enabling swimming activity would be to coordinate the many spinal IN populations that generate the patterns of repetitive , left-right alternating tail beats seen in developing zebrafish swimming ( Saint-Amant , 2010 ) . The observation that in our beat-and-glide simulation , V2a neuron firing phasically precedes firing of all the other intrasegmental spinal INs and MNs reinforces the central role of these neurons in driving rhythmic tail beats . We did find that in simulations where there were only tonic firing neurons , the stability of swimming episode durations started degrading after about 6000 ms . Therefore , burst firing neurons may help to promote the stability of the beat-and-glide pattern . Whether or not this is the case remains to be tested experimentally . Neuromodulation may serve as a mechanism that permits V2as , or other spinal neurons , to toggle between tonic and burst firing through neuromodulation . It is well established that neuromodulators shape the activity of spinal locomotor circuits , likely by regulating intrinsic properties of spinal neurons and through modulation of synaptic weighting and other mechanisms . Blocking D4 dopamine receptors at 3 dpf prevents the transition from burst to beat-and-glide swimming ( Lambert et al . , 2012 ) , suggesting that dopamine from supraspinal sources plays a role in setting the beat-and-glide phenotype by shortening swimming episode duration . Paired recordings of diencephalospinal dopaminergic neurons and spinal MNs during swimming show that these two populations often burst together ( Jay et al . , 2015 ) . Later in development at 6–7 dpf , activation of D1 dopamine receptors increases the recruitment of slow MNs to increase swimming speed ( Jha and Thirumalai , 2020 ) . The neuromodulator serotonin ( 5-HT ) has been found to either increase motor output by decreasing inter-episode intervals in intact larval zebrafish ( Brustein and Drapeau , 2005; Brustein et al . , 2003 ) or decrease swimming frequency or burst firing in spinalized larvae and adult zebrafish ( Gabriel et al . , 2009; Montgomery et al . , 2018 ) . In the adult zebrafish , serotonin strengthens inhibition to MNs between tail beats and slows down the onset of the depolarization that initiates each successive tail beat ( Gabriel et al . , 2009 ) . Our model could identify possible targets within the spinal cord for specific neuromodulators of locomotor function in zebrafish . Our sensitivity analysis suggests that the neuromodulation of intrinsic properties that affect the membrane potential dynamics of spinal neurons could easily modulate locomotor output . The behavior of our models was also sensitive to a lesser degree to increasing variability in descending drive , synaptic weighting , and rostrocaudal extent of connections . Variability in these parameters could change the proportions of coiling types or the values of the characteristics of swimming output measured ( e . g . episode duration and inter-episode interval ) . Model parameter variability sometimes increased the variability of motor output ( e . g . , Figure 10 ) , perhaps indicating a breakdown of the model . However , variability in both model parameters and motor output should not necessarily be considered weaknesses of the model but may instead reflect true biological variability ( Marder and Taylor , 2011 ) . For instance , recordings of swimming characteristics such as episode duration and inter-episode intervals in larval zebrafish show appreciable variation ( Brustein and Drapeau , 2005; Buss and Drapeau , 2001 ) . Quantifying heterogeneity within and between animals may guide the appropriate levels of parameter variability to include in future iterations of our models . Many computational models have already been made of spinal circuits for swimming in species that use undulatory movements spreading from head to tail . These include models for Xenopus ( Ferrario et al . , 2018; Hull et al . , 2016 ) , lamprey ( Kozlov et al . , 2009; Kozlov et al . , 2014; Messina et al . , 2017 ) , and salamanders ( Bicanski et al . , 2013; Ijspeert et al . , 2007 ) . These models have become detailed enough to include many neurons forming circuits distributed across the hindbrain and the spinal cord . Some models incorporate specific intrinsic and ligand-gated currents with known roles in rhythmogenesis in their respective species ( Ferrario et al . , 2018; Kozlov et al . , 2009; Kozlov et al . , 2014 ) . Simulations of the models have been used to test aspects of swimming control , including steering commands from descending commands to spinal networks ( Kozlov et al . , 2014 ) , the integration of sensory triggers of locomotion ( Ferrario et al . , 2018; Ijspeert et al . , 2007 ) , the coupling of axial and limb central pattern generators ( Ijspeert et al . , 2007 ) , and the role of left-right coupling in rhythm generation ( Messina et al . , 2017 ) . Our model could be used to identify possible similarities or differences in how these aspects of motor control are controlled in the zebrafish . To the best of our knowledge , this is the first model to generate several forms of locomotor movements in developing zebrafish based upon previously described neurons and their connectivity patterns . The analysis of the simulations generated yielded several predictions about possible connections between spinal neurons , firing properties of neurons , and roles for neurons in specific locomotor movements . For instance , the single coiling model predicts that IC and V0d are coupled together to facilitate the activation of V0ds , which are responsible for the glycinergic SBs observed in spinal neurons at this stage ( Saint-Amant and Drapeau , 2001; Tong and McDearmid , 2012 ) . Our modeling study also predicts that the generation of double and even multiple coils depend on untested connections between V2a and V0v neurons and between V0v and IC neurons . The latter connection would be needed to initiate consecutive left-right alternating coils through the activation of contralateral IC neurons , while the former connection would be needed to activate the ipsilateral V0v responsible for the activation of contralateral ICs . The V2a to V0v connection could be deemed unnecessary in light of possible gap junction coupling between ipsilateral IC and V0vs . However , our modeling suggests that delayed activation of V0vs would allow the ipsilateral coil to complete before activating the contralateral coil . This delay would not be possible with gap junction mediated excitation of V0vs by ipsilateral ICs . Our double-coiling model also predicts that contralateral inhibition of ICs by V0ds prevents the generation of multiple coilings . Several of these predictions are supported by pharmacological experiments suggesting that blocking glutamatergic transmission in embryonic zebrafish precludes double coiling while blocking glycinergic transmission at that stage promotes multiple coilings ( Knogler et al . , 2014 ) . The beat-and-glide model also proposes a prominent role of delayed contralateral excitation in ensuring repetitive left-right alternating tail beats during swimming . Whether delayed contralateral excitation is a conserved mechanism of operation in double coiling and swimming remains to be tested experimentally . While V0v neurons are the likely candidate to mediate the activation of contralateral movements , different subgroups of V0v neurons are probably involved in coiling versus swimming ( Björnfors and El Manira , 2016; Jay and McLean , 2019 ) considering the two different targets of contralateral excitation involved , namely ICs in coiling and V2as in swimming . The continued presence of left-right tail beats in simulations where the dI6 population of commissural inhibitory neurons was silenced or in simulations with blockade of glycinergic transmission further underscores the need to test the contributions of V0v neurons to left-right alternation . Finally , the ability of our model to generate beat-and-glide swimming with or without burst firing neurons suggests a possible degeneracy in the operation of spinal swimming circuits of the developing zebrafish . This possibility would be consistent with the well-characterized degeneracy of the nervous system , as reinforced by modeling studies where combinations of intrinsic properties or connectivity can generate the same motor output ( Goldman et al . , 2001; Taylor et al . , 2009 ) . Many rhythmogenic currents ( e . g . , NMDA , calcium-dependent potassium currents , and persistent sodium currents ) have been implicated in the operation of locomotor circuits of zebrafish ( Song et al . , 2020 ) and other invertebrate and vertebrate rhythm-generating circuits ( Anderson et al . , 2012; Golowasch and Marder , 1992; Ryczko et al . , 2010; Tazerart et al . , 2007; Zhong et al . , 2007 ) . In addition , while some motor systems rely upon pacemaker neurons , other rhythmic motor systems could also rely on network-based mechanisms ( Del Negro et al . , 2010 ) , further demonstrating the diversity of means by which the nervous system generates rhythmic activity . Whether the spinal circuits for swimming are degenerate or degeneracy is only exhibited in our modeling remains to be tested experimentally . The operation of the spinal swimming circuit in zebrafish may exhibit degeneracy dependent upon specific environmental or physiological conditions ( Vogelstein et al . , 2014 ) and their resulting neuromodulatory states . Our models will require integrating additional cell populations and circuitry to capture the full range of locomotor movements of developing zebrafish . The beat-and-glide model only generates swimming within a narrow frequency range . The generation of a broader range of swimming frequency ( McLean and Fetcho , 2009 ) will require expanding each cell population into subgroups with different intrinsic properties ( Menelaou and McLean , 2012; Song et al . , 2018 ) , rostrocaudal projection patterns , and specific connectivity patterns between subgroups and between cell populations ( Ampatzis et al . , 2014; Bagnall and McLean , 2014; Kimura and Higashijima , 2019; Menelaou and McLean , 2019; Sengupta et al . , 2021; Song et al . , 2020 ) . These subgroups , which may arise from different birth dates ( McLean and Fetcho , 2009; Satou et al . , 2012 ) , are active at specific swimming frequencies ( McLean et al . , 2007; McLean et al . , 2008; McLean and Fetcho , 2009 ) . There seem to be modules consisting of neurons within each cell population that are active at specific swim frequencies ( Ampatzis et al . , 2014; Menelaou and McLean , 2019; Song et al . , 2018; Song et al . , 2020 ) . Indeed , previous studies in zebrafish have shown that MNs and V2a neurons are organized in three different modules ( linked to slow , medium , and fast MNs ) that are differentially recruited according to swim frequency ( Ampatzis et al . , 2014; Song et al . , 2020 ) . Swim frequency modules likely include commissural excitatory V0v INs ( Björnfors and El Manira , 2016; McLean et al . , 2008 ) and commissural inhibitory INs belonging to either the dI6 or V0d populations ( Satou et al . , 2020 ) . The modeling of additional subgroups , especially in the context of swim-frequency modules , will need to take into account the high specificity of connectivity between subgroups within a cell population ( Menelaou and McLean , 2019; Song et al . , 2020 ) and subgroups belonging to different spinal populations within swim frequency-modules ( Ampatzis et al . , 2014; Bagnall and McLean , 2014; Menelaou and McLean , 2019; Song et al . , 2020 ) . Subgroups within cell populations are not necessarily restricted to those belonging to different swim frequency modules but may also exist between neurons involved in rhythm versus vigor of movement . Subgroups for vigor seem to be present within the V2a ( Menelaou and McLean , 2019; Song et al . , 2018 ) and V0v ( Björnfors and El Manira , 2016; Jay and McLean , 2019; McLean et al . , 2007 ) populations . Furthermore , the implementation of circuitry for swimming vigor is likely to necessitate adding the ipsilaterally projecting , inhibitory V2b population ( Callahan et al . , 2019 ) . The circuits for frequency and vigor are likely to interact , as seen by the swimming frequency-dependent action of V1 neurons ( Kimura and Higashijima , 2019 ) . Frequency and vigor are also likely to be shaped by sensory information . Incorporating spinal neurons that integrate sensory information ( Liu and Hale , 2017 ) provided by peripherally located and spinally located sensory neurons ( Böhm et al . , 2016; Picton et al . , 2021 ) will provide a more accurate representation of swimming control at the level of the spinal cord . Finally , the undefined role of specific spinal neuron populations could be studied after being integrated into the model following further characterization . For example , ventral V3 neurons in mouse spinal locomotor networks have been studied using modeling . Those studies suggest an important role for these neurons in left-right coordination in mouse locomotion ( Danner et al . , 2019 ) . Similar computational studies using our model could reveal testable predictions of the role of these neurons ( England et al . , 2011; Yang et al . , 2010 ) in zebrafish swimming . Our models simulate several developmental milestones of the zebrafish locomotor behavior . Iterative changes were made to each model to successively transition from single coiling to double coiling and then to beat-and-glide swimming . This iterative process could be further developed to obtain a higher resolution understanding of the maturation of locomotion in zebrafish . Further transitory models could be built to fill the gaps between our current models ( e . g . , a model for burst swimming that precedes beat-and-glide swimming ) . The generation of these additional transitory models could be coupled with experimental data studying the mechanisms that drive the transition from one milestone to the other ( Brustein and Drapeau , 2005; Knogler et al . , 2014; Lambert et al . , 2012; Roussel et al . , 2020 ) to identify specific underlying changes in intrinsic and network properties . Thus , the models presented herein offer invaluable tools to investigate further the mechanisms by which spinal circuits control facets of swimming , including speed , direction , and intensity through interactions within the spinal cord and with supraspinal command centers , as well as the developmental dynamics that ensure proper maturation of movement during development .
Modeling was performed using Python 3 . 6 . 3 64 bits ( RRID:SCR_008394 ) . We did not analyze the early parts of simulations ( up to 200 ms ) to allow the effects of initial conditions to dissipate . We modeled neurons using a single compartment , simple spiking neuron model developed by Izhikevich , 2007 . The following general differential equations govern the dynamics of the membrane potential: ( 1 ) CV′=k ( V−Vr ) ( V−Vt ) −u+Isynu′=a ( b ( V−Vr ) −u ) ifV=Vmax , thenV←c , u←u+d Specific active conductances are not included in these models . Instead , values of the parameters a , b , c , d , and Vmax ( which respectively represent the time scale of the recovery variable u , the sensitivity of u to the subthreshold variation of V , the reset value of V after a spike , the reset value of u , and the action potential peak ) , as well as values of the parameters k , C , Vr , and Vt ( coefficient for the approximation of the subthreshold part of the fast component of the current-voltage relationship of the neuron , cell capacitance , resting membrane potential , and threshold of action potential firing ) can be selected to model a wide range of firing behaviors including bursting ( or chattering ) pacemaker , tonic firing , phasic spiking neurons , or firing rate adaptation neurons ( Table 4 ) . Isyn represents the sum of the synaptic and gap junction currents received by the neuron . For all models , the Euler method was used for solving ordinary differential equations with a time step of 0 . 1 ms . We modeled all electrical synapses ( i . e . , gap junctions ) as ideal resistors following Ohm’s Law: ( 2 ) Igap:pre , post=VpreGpre , post With Igap:pre , post representing the synaptic current flowing to the postsynaptic neuron from the presynaptic neuron through gap junctions and Gpre , post representing the total conductance of gap junctions between the presynaptic and postsynaptic neurons ( Table 1 ) . Synaptic conductances of chemical synapses were modeled as a sum of two exponentials weighted by a synaptic weight based upon the general equation: ( 3 ) Ipre , post= ( Vpost−Erev ) ( e−t−t0τr−e−t−t0τf ) Wpre , postifVpre>Vthrwhere Ipre , post is the synaptic current received by the postsynaptic neuron from neurotransmitter release by the presynaptic neuron if the presynaptic neuron membrane potential , Vpre , crosses a voltage threshold , Vthr , at the synapse . Vpost is the membrane potential of the postsynaptic neuron , Erev is the reversal potential , τr and τf are the rise and fall time constants , respectively , t0 is the time at which Vpre crossed Vthr , and Wpre , post is the synaptic weight between the presynaptic and postsynaptic neurons ( Table 3 ) . Ipre , post is equal to 0 if Vpre is below Vthr . We implemented two types of chemical synapses: glutamatergic and glycinergic synapses . The former differs from the latter by the respective reversal potential values Erev of glutamatergic and glycinergic synapses and the time constant values τr and τf ( Table 3 ) . Values of the glycinergic Erev are depolarized at early developmental stages ( Saint-Amant and Drapeau , 2000; Saint-Amant and Drapeau , 2001 ) , and this reversal potential becomes gradually hyperpolarized ( Ben-Ari , 2002 ) as the equilibrium potential of chloride hyperpolarizes in the zebrafish nervous system ( Zhang et al . , 2010 ) . All chemical synapses were turned off in the initial 50 ms of every simulation to allow initial conditions to dissipate . A key feature of our modeling approach was to assign spatial coordinates ( x , y ) to point-like neurons ( i . e . , neurons have no spatial dimension , but they have a position in space ) , giving the spatial distribution of neurons a central place in our model computing process . We used the Euclidean distance to calculate the distance between each neuron and to approximate axon length . Distance unit is arbitrary and was set so that one model somite was 1 . 6 arbitrary distance units ( a . d . u . ) long . Time delays for each synaptic connection were computed as a function of the distance between neurons and were used to calculate delayed synaptic current: ( 4 ) Idelayed:pre , post ( t ) =Ipre , post ( t−Dpre , postcv ) With Dpre , post as the Euclidean distance between the presynaptic and postsynaptic neurons and cv as the transmission speed in arbitrary distance units per second ( a . d . u . /s ) . This distance and the neuron position were also used to apply conditions on synaptic weights of neurons ( e . g . , limits as to how far descending neurons project ) . For the single coiling model , cv was set to 4 . 0 a . d . u/s . For the multiple coiling model , cv was set to 1 . 0 a . d . u/s . These values were obtained through trial-and-error and may reflect changes in myelination and body size of the developing zebrafish . For the beat-and-glide swimming model , cv was set to 0 . 8 a . d . u/s , which led to intersegmental transmission delays in the range of 3 . 0–4 . 0 ms , closely matching the 1 . 6 ms intersomitic delay previously reported ( McDearmid and Drapeau , 2006 ) , assuming that each model somite represents two biological somites at this developmental stage ( see Musculoskeletal model below ) . Spinal locomotor circuits were distributed across two columns , one for each side of the body , giving the network a nearly one-dimensional organization along the rostrocaudal axis . Therefore , we used the x-axis as the rostrocaudal axis , whereas the y-axis was only used to partition neurons from the left and right sides ( assigning the coordinate y=1 a . d . u . for the right side and y=−1 a . d . u . for the left side ) . We scaled key parameters to Gaussian noise to test the robustness of our three base models ( single coiling , multiple coiling , and beat-and-glide swimming ) to parameter variability . Sensitivity to noise of the base models was tested by scaling the parameters that set the tonic motor command drive's amplitude , the rostrocaudal length of neuron projections , the membrane potential dynamics ( Izhikevich model ) , and synaptic weighting . These four sets of parameters were randomized by multiplying the parameters with a random number picked from a Gaussian distribution with mean=1 , and standard deviations , σd , σl , σp , and σw , respectively . The amplitude of the motor command drive was randomized at each time point . The parameters for the membrane potential dynamics , rostrocaudal length of axons , and synaptic weights were randomized at the start of each simulation and did not change during the simulations . We implemented a musculoskeletal model of the fish body to convert the output of the spinal circuit model into changes in body angles and frequency of locomotor movements . Each MN output along the fish body was inputted into a muscle cell ( Figure 1 ) . The membrane potential of the muscle ( V ) was modeled as a simple passive RC circuit ( R and C being the muscle cell membrane resistance and capacitance , respectively ) , described by the following equation: ( 5 ) Vmuscle'=VRC+IsynC For muscle cells , values of R were 25 ( single coiling ) , 50 ( double coiling ) , and 1 ( beat-and-glide ) ; values of C were 10 ( single coiling ) , 5 ( double coiling ) , and 3 ( beat-and-glide ) . These values were chosen to produce kinematics representative of those seen experimentally . To reduce computational load , we modeled one muscle cell as representing three somites of the body in the base model for coiling and two somites of the body in the base model for swimming . The whole body of the fish was modeled as a chain of uncoupled damped pendulums . We computed local body angles according to the difference in activity between the local left and right muscle cells . The deflection angle θi of the ith muscle cell was computed according to the following differential equation . ( 6 ) θi''+2ζω0θi'+ω02θi=α ( 1-0 . 2R ) ( VRmuscle , i-VLmuscle , i ) With VRmuscle , i and VLmuscle , i being the solution of the equation ( Equation 5 ) for the ith muscle on the right and left side of the body , respectively ( Figure 1D ) . α is the conversion coefficient from an electric drive of the muscle cells to a mechanical contraction of the same cells . The midline of the body can be computed at any given time as ( x , y ) coordinates using trigonometric identities from θi ( Figure 1E ) . Specifically , ( 7 ) xi=xi−1+l⋅sin ( θi ) yi=yi−1−l⋅sin ( θi ) where ( xi , yi ) are the spatial coordinates of the ith somite , and l is its length . We set ( x0 , y0 ) to ( 0 , 0 ) and applied the previous set of equations ( Equation 7 ) for i≥1 . Thus , heat-maps of local body angle ( θi ) variation through time provide comprehensive information about the network output ( Figure 1F ) . The integrated motor output of the model ( e . g . , Figure 5B ) was calculated as the sum of the muscle output at all muscle cells on both sides of the body , followed by a convolution of this sum with a 50 ms square wave . Left-right alternation at the ith somite was analyzed using cross-correlation of VRmuscle , i and VLmuscle , i at that somite . The minimum coefficient in the range of time delays between −20 and 20 ms was calculated to estimate left-right alternation . A value of 0 indicates left-right out-of-phase alternation , while a value of 1 suggests complete in-phase synchrony . To calculate the duration of swimming episodes , we summated the muscle activity across all somites from both sides of the body . This muscle activity was then convoluted , and a threshold of 0 . 5 arbitrary units was set to detect the start and end of each swimming episode of most simulations . In a few simulations where motor output was very large , the threshold was adjusted to detect episodes . To estimate the tail beat frequency , we determined when the most caudal somite crossed the midline of the body of the musculoskeletal model ( a threshold of 0 . 5 arbitrary units from the center was used to detect crossing to a side of the body ) . The reciprocal of the interval between consecutive left-to-right or right-to-left crossing was used to calculate the instantaneous tail beat frequency . Any interval greater than 100 ms was considered to be between episodes rather than within an episode and discarded from the calculation of instantaneous tail beat frequency . To calculate the phase delay between pairs of neurons in the beat-and-glide swimming model , we first calculated the autocorrelation of the reference neuron . The time delay at which the peak autocorrelation occurred was used to estimate the period of the reference neuron cycle . The cross-correlation between the reference and test neuron was then calculated , and the phase delay was calculated as the time delay at which the peak of the cross-correlation occurred divided by the cycle period of the reference neuron in radians . In the coiling models , the cycle period is 1000–2000 ms ( single coiling ) or 10 , 000–20 , 000 ms ( double coiling ) due to longer inter-coiling intervals . Normalizing phase shifts by this cycle period makes the phase delays very small . Therefore , for the coiling models , the period of the reference neuron cycle was estimated by the average duration of single coiling ( 1074 ms ) or double coiling events ( 1891 ms ) . Note that this procedure does not change the polarity of the phase delay but better separates the various phase delays on a polar plot . Statistical analysis was performed using the SciPy Python library . Statistical tests consisted of one-factor ANOVA tests followed by two-tailed Student’s t-tests . A p-value<0 . 05 was used to determine statistical significance , and all tests were corrected for multiple comparisons ( Bonferroni correction for multiple t-tests ) . The code for the models and for the figures can be accessed on GitHub , copy archived at swh:1:rev:dd36ce928a2775eeed45149444962a422cb16446 ( Roussel et al . , 2021 ) . Updates and revisions to the models will also be made available at this site . The simulation data that was used for the figures can be accessed at the Federated Research Data Repository at the following DOI: https://doi . org/10 . 20383/102 . 0498 . | The spinal cord is a column of nerve tissue that connects the brain to the rest of the body in vertebrate animals . Nerve cells in the spinal cord , called neurons , help to control and coordinate the body’s movements . As the spinal cord develops , new neurons are born and new connections are made between neurons and muscles , resulting in more coordinated and skillful movements as time goes on . Zebrafish , for example , display body-bending maneuvers called coils within 24 hours of the egg being fertilized . Next , bursts of swimming movements emerge , which are driven by sporadic tail beats . These tail maneuvers become more consistent as the fish develops , and eventually result in smooth movements called beat-and-glide swimming . The groups of spinal cord neurons that appear at each stage of zebrafish development have been characterized , but it remains unclear how newly formed circuits ( groups of neurons recently connected to each other ) work together to produce swimming maneuvers . To answer this question , Roussel et al . simulated changes in the spinal cord that help zebrafish acquire new swimming movements as they grow . The computer models encoded neural circuits based on cell populations identified in experimental studies , and replicated swimming behaviors that emerge during the first few days of zebrafish development . Simulations tested how specific neural circuits generate the characteristic swimming movements that represent key developmental milestones in zebrafish . The results showed that adding new neurons and more cell-to-cell connections led to increasingly sophisticated swimming maneuvers . As the zebrafish spinal cord matured , the fish were better able to control the pace and duration of their swimming movements . Roussel et al . also identified specific patterns of neural activity linked to particular maneuvers . For example , tail beats switch direction when neurons on one side of the spinal cord excite neurons on the opposite side . This activity , which becomes more rhythmic , also needs to be exquisitely timed to produce and coordinate the right motion . Roussel et al . ’s modelling of developmental milestones in growing zebrafish provides insights into how neural networks control movement . The computer models are among the first to accurately reproduce swimming behaviors in developing zebrafish . More experimental data could be added to the models to capture the full range of early zebrafish movements , and to further investigate how maturing spinal cord circuits control swimming . Since zebrafish and mammals have many spinal neurons in common , further research may aid our understanding of movement disorders in humans . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"neuroscience"
] | 2021 | Modeling spinal locomotor circuits for movements in developing zebrafish |
hIAPP fibrils are associated with Type-II Diabetes , but the link of hIAPP structure to islet cell death remains elusive . Here we observe that hIAPP fibrils are cytotoxic to cultured pancreatic β-cells , leading us to determine the structure and cytotoxicity of protein segments composing the amyloid spine of hIAPP . Using the cryoEM method MicroED , we discover that one segment , 19–29 S20G , forms pairs of β-sheets mated by a dry interface that share structural features with and are similarly cytotoxic to full-length hIAPP fibrils . In contrast , a second segment , 15–25 WT , forms non-toxic labile β-sheets . These segments possess different structures and cytotoxic effects , however , both can seed full-length hIAPP , and cause hIAPP to take on the cytotoxic and structural features of that segment . These results suggest that protein segment structures represent polymorphs of their parent protein and that segment 19–29 S20G may serve as a model for the toxic spine of hIAPP .
Amyloid fibrils are associated with more than 25 diseases , including Alzheimer’s disease , Parkinson’s disease , and Type-II Diabetes ( T2D ) ( Eisenberg and Jucker , 2012 ) . The fibrils observed in each disease are composed of a particular protein; in T2D , amyloid fibrils are composed of human islet amyloid polypeptide ( hIAPP ) ( Westermark et al . , 1987; Cooper et al . , 1988 ) . hIAPP is a 37 residue polypeptide hormone that is co-secreted with insulin to modulate glucose levels ( Roberts et al . , 1989; Westermark et al . , 2011 ) . Researchers have accumulated substantial evidence for a correlation between hIAPP aggregation and pancreatic β-cell death in the course of the disease , T2D . Approximately 90% of pancreatic tissue samples taken post-mortem from T2D patients contain islet amyloid primarily composed of hIAPP ( Höppener et al . , 2000 ) . The extent of islet amyloid positively correlates with pancreatic β-cell loss and insulin dependence ( Maloy et al . , 1981; Esapa et al . , 2005; Jurgens et al . , 2011 ) . Additional support for a link comes from comparison of human and mouse IAPP: mouse IAPP differs from human IAPP by only six residues , 3 of which are β-strand breaking prolines . Consequently , mouse IAPP does not aggregate ( Nishi et al . , 1989; Westermark et al . , 1990 ) . Moreover , mice can be induced to develop islet amyloid and T2D when they are engineered to express human IAPP and fed a high fat diet ( Verchere et al . , 1996; Westermark et al . , 2000 ) . Perhaps the strongest support for a link is the mutation in hIAPP , hIAPP-S20G; segments that contain this mutation aggregate more quickly , contribute to increased pancreatic β-cell apoptosis , and are associated with early onset T2D in families who carry this lesion ( Sakagashira et al . , 2000; Cao et al . , 2012; Meier et al . , 2016; Sakagashira et al . , 1996; Lee et al . , 2001; Morita et al . , 2011 ) . Although a link between hIAPP aggregation and pancreatic β-cell death is well established , precisely which type of hIAPP aggregate contributes to pancreatic β-cell death and insulin dependence remains undetermined . Using mostly in vitro studies , researchers have presented evidence for toxicity of multiple types of hIAPP aggregates . Early studies suggest that amyloid fibrils are the primary cytotoxic species because preparations that contain fibrillar hIAPP were more cytotoxic than soluble preparations of the protein ( Lorenzo et al . , 1994; Lorenzo and Yankner , 1994; Schubert et al . , 1995; Kapurniotu , 2001 ) . Using cells and transgenic rodents as disease models , other studies found hIAPP fibrils to be associated with apoptosis , β-cell loss , and T2D severity ( O'Brien et al . , 1995; Hiddinga and Eberhardt , 1999; Janson et al . , 1996; Hull et al . , 2005a , 2005b; Pilkington et al . , 2016 ) . In contrast , some studies show that the process of hIAPP fibril formation , not the amyloid fibrils themselves , is the source of toxicity ( Schlamadinger and Miranker , 2014; Oskarsson et al . , 2015 ) . However , most current research studies suggest soluble pre-fibrillar oligomers are the primary type of toxic aggregate . Support for oligomers as the primary cytotoxic species comes from the observation of oligomers associated with caspase activity and ER stress , which precede the formation of extracellular amyloid fibrils ( Meier et al . , 2006; Ritzel et al . , 2007; Bram et al . , 2014; Mukherjee et al . , 2015; Lin et al . , 2007; Huang et al . , 2007; Haataja et al . , 2008; Abedini et al . , 2016 ) . Several recent studies show that hIAPP fibrils are relatively inert and do not exert obvious toxicity . Despite these extensive in vitro studies , it is not clear that the toxic aggregates they describe also elicit toxicity in vivo . In closer agreement with earlier studies , we find that hIAPP preparations that contain fibrils are cytotoxic to a rat pancreatic β-cell line , thus motivating us to determine the structure of the spine of hIAPP fibrils . If fibrils are a bona fide type of toxic aggregate in vivo , then determining the atomic structure of the spine of hIAPP fibrils is a logical approach for advancing our understanding of disease-relevant targets ( Wiltzius et al . , 2008 , 2009a; Soriaga et al . , 2016 ) . Furthermore , we can utilize atomic structures as templates for structure-based design of novel therapeutics that may protect against pancreatic β-cell death . Although full-length amyloid proteins have so far been resistant to crystallization , select protein segments that form the spines of amyloid fibrils do form crystals ( Nelson et al . , 2005; Sawaya et al . , 2007; Rodriguez et al . , 2015 ) . Indeed , the atomic structures of nearly 90 amyloid spines have been revealed in this manner . Other studies have taken an alternative approach: they employed solid-state NMR to gain detailed structural insights into hIAPP fibril structure ( Luca et al . , 2007; Weirich et al . , 2016 ) ; some of these structures have spurred successful inhibitor designs ( Mirecka et al . , 2016 ) . Here , we use the cryoEM method MicroED to determine the atomic structure of two 11-residue segments , termed spine segments , that span the amyloid spine of hIAPP .
To compare the cytotoxic effects of oligomeric and fibrillar hIAPP , we generated hIAPP preparations that contained either amyloid oligomers or fibrils . We did this by aging the same concentration of hIAPP for 0 and 24 h time periods . Aging hIAPP for 24 h yielded amyloid fibrils and no detectable oligomers as assessed by Thioflavin-T ( ThT ) binding , negative-stain transmission electron microscopy ( TEM ) , and a dot blot assay using the fibrillar oligomer-sensitive antibody , LOC ( Figure 1A ) . Aging hIAPP for 0 h , which is a freshly dissolved hIAPP sample , yielded oligomers as assessed by a dot blot assay using LOC , and no amyloid fibrils ( Figure 1—figure supplement 1A ) . Of note , we probed both hIAPP preparations with 25 different conformational antibodies that are known to bind soluble oligomers , but only LOC showed binding to any of our preparations . Although LOC was raised against hIAPP fibrils ( Kayed et al . , 2007 ) , studies show that it also recognizes fibrillar oligomers ( Wu et al . , 2010 ) , which share structural epitopes with amyloid fibrils and are structurally distinct from A11-positive pre-fibrillar oligomers . 10 . 7554/eLife . 19273 . 003Figure 1 . Preparations of hIAPP that contain amyloid fibrils are cytotoxic to a rat pancreatic β-cell line . ( A ) Human IAPP ( hIAPP ) aged for 24 h contains amyloid fibrils and no detectable oligomers . Amyloid fibrils were observed using ThT binding and TEM . Oligomers were detected using a dot bot assay with the polyclonal anti-oligomer antibody , LOC . hIAPP oligomers were used as the positive control for LOC binding . The dashed line on the ThT binding graph indicates ThT fluorescence of vehicle alone . ( B ) and ( C ) hIAPP aged for 24 h is significantly more cytotoxic than hIAPP aged for 0 h . In these experiments , 50 μM human and mouse IAPP were aged for the designated time periods and then they were applied to cells at 5 μM final concentration . Mouse IAPP ( mIAPP ) , which does not form amyloid fibrils , is not cytotoxic regardless of time period of aging . Black horizontal bars indicate the median ( n = 12–15 across 4–5 biological replicates , each with three technical replicates ) . ( B ) Rin5F cells treated with hIAPP aged for 24 h reduce significantly less MTT dye than Rin5F cells treated with hIAPP aged for 0 h ( ns=not significant; ****p<0 . 0001 using an unpaired t-test with equal standard deviations ) . ( C ) Rin5F cells treated with hIAPP aged for 24 h exhibit significantly higher caspase-3/7 activation than Rin5F cells treated with hIAPP aged for 0 h . Additionally , Rin5F cells treated with hIAPP aged for 24 h exhibit significantly higher caspase-3/7 activation than vehicle-treated cells ( ***p=0 . 0008 using an ordinary one-way ANOVA ) , but Rin5F cells treated with hIAPP aged for 0 h do not ( p=0 . 4286 using an ordinary one-way ANOVA ) ( ns=not significant; **p=0 . 0011 using an unpaired t-test with equal standard deviations ) . ( D ) The insoluble fraction of hIAPP aged 24 h , which contains amyloid fibrils and no detectable oligomers , contains the cytotoxic species . Cytotoxicity was measured using MTT dye reduction and detection of caspase-3/7 activation ( ****p<0 . 0001; **p<0 . 0013; n = 9 across three biological replicates , each with three technical replicates ) . ( E ) Amino acid sequences of human IAPP and mouse IAPP . The location of the early onset familial mutation , S20G , is shown below the human sequence . Red residues in the mouse sequence differ from the human sequence . The amyloid spine of human IAPP and the corresponding region in the mouse sequence is enclosed in the gray box . ( F ) . Schematic of protein segments that span the amyloid spine , hereon referred to as spine segments , targeted for characterization . ( G ) Fibrils of spine segments seed hIAPP fibril formation , suggesting that spine segments embody structural characteristics of full-length hIAPP fibrils . 10 μM hIAPP was seeded with 10% ( v/v ) monomer equivalent of pre-formed , unsonicated seed of each spine segment . mIAPP , which does not contain amyloid fibrils , does not seed hIAPP fibril formation . Curves show average of 4 technical replicates . DOI: http://dx . doi . org/10 . 7554/eLife . 19273 . 00310 . 7554/eLife . 19273 . 004Figure 1—figure supplement 1 . Characterization of hIAPP aged for 0 h and the soluble and insoluble fractions of hIAPP aged for 24 h . ( A ) hIAPP aged for 0 h contains oligomers and no detectable amyloid fibrils as assessed by ThT binding , TEM , and a dot blot assay using the anti-oligomer antibody , LOC . hIAPP oligomers were used as the positive control for LOC binding . ( B ) The insoluble fraction of hIAPP aged for 24 h , which contains the cytotoxic species , is composed of amyloid fibrils and no detectable oligomers . The soluble fraction , which is not cytotoxic , contains no detectable amyloid fibrils or oligomers . The dashed line on the ThT binding graphs indicates ThT fluorescence of vehicle alone . DOI: http://dx . doi . org/10 . 7554/eLife . 19273 . 00410 . 7554/eLife . 19273 . 005Figure 1—figure supplement 2 . All spine segments form amyloid fibrils or 3D crystals only a few hundred nanometers thick , as observed using TEM . Fibrils and 3D crystals were formed by dissolving lyophilized protein segments at 1 mM in PBS and 1% DMSO and incubating them for one week at room temperature under quiescent conditions . Fibril and crystal formation occurred as quickly as a few hours ( 19–29 S20G ) to as long as overnight ( 15–25 WT and 15–25 S20G ) . ( A ) 15–25 WT forms striated ribbons . ( B ) 15–25 S20G forms striated ribbons . ( C ) 19–29 WT forms both striated ribbons and twisted fibrils of varying widths . ( D ) 19–29 S20G forms 3D crystals only a few hundred nanometers thick . Right panel scale bar is 1 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 19273 . 00510 . 7554/eLife . 19273 . 006Figure 1—figure supplement 3 . Technical replicates and control samples for ThT assay in Figure 1G . ( A ) Fibrils of spine segments seed hIAPP fibril formation . All four technical replicates performed in the experiment in Figure 1G are shown . ( B ) Seeds of spine segments ( 1 μM ) do not bind ThT . The graph shows mean ThT fluorescence across four technical replicates . DOI: http://dx . doi . org/10 . 7554/eLife . 19273 . 006 We observe that hIAPP preparations that contain fibrils are significantly more cytotoxic to rat pancreatic β-cells than hIAPP preparations that contain oligomers but no detectable fibrils ( Figure 1B and C ) . We assayed the cytotoxicity of the hIAPP preparations to Rin5F cells , a rat pancreatic β-cell line ( Gazdar et al . , 1980 ) using two metrics: 3- ( 4 , 5-dimethylthiazol-2-yl ) −2 , 5-diphenyltetrazolium bromide ( MTT ) dye reduction , an indicator of good metabolic health ( Mosmann , 1983; Liu et al . , 1997 ) , and activation of caspase-3/7 , an indicator of apoptosis ( Budihardjo et al . , 1999 ) . Furthermore , the insoluble fraction of the hIAPP 24 h sample , which contains fibrils ( Figure 1—figure supplement 1B ) , is cytotoxic , while the soluble fraction is not ( Figure 1D ) , further suggesting that fibrils are the toxic aggregate in our studies . Although we do not detect oligomers in the 24 h sample , we cannot rule out the possibility that it may contain some undetectable population of slowly forming , yet highly toxic oligomers that associate with fibrils . Despite this possibility , we chose to focus on studying fibrillar structures of hIAPP . Given that hIAPP fibrils are cytotoxic , we sought to identify the residues that compose their amyloid spine . We identified residues 15–29 as the amyloid spine based on several lines of evidence and previous work by others ( Westermark et al . , 1990; Moriarty and Raleigh , 1999; Goldsbury et al . , 2000; Tenidis et al . , 2000 ) . First , the sequence of mouse IAPP ( mIAPP ) , which is non-amyloidogenic , differs from human IAPP only within this region ( Figure 1E ) . Second , the only known familial disease mutation in hIAPP , hIAPP-S20G , also occurs within this region ( Figure 1E ) ( Sakagashira et al . , 2000; Cao et al . , 2012 ) . Third , previous work by our laboratory has shown that Phe15 may be part of the amyloid spine because it is required for stabilizing an on-pathway α-helical dimer and mutating this residue can delay fibril formation ( Wiltzius et al . , 2009b ) . For these reasons , we chose to focus on two overlapping 11-residue segments within this region of the sequence: residues 19–29 and residues 15–25 . We chose to study the WT and early onset S20G mutation segments ( Figure 1F ) . All four spine segments form amyloid fibrils or crystals ( Figure 1—figure supplement 2 ) that seed full-length hIAPP fibril formation ( Figure 1G , Figure 1—figure supplement 3 ) , suggesting that the spine segments embody structural characteristics of full-length hIAPP fibrils . To determine the structure of segment 19–29 S20G , we used Micro-Electron Diffraction ( MicroED ) . MicroED employs a standard cryo electron microscope ( cryoEM ) in diffraction mode for data collection from 3D crystals only a few hundred nanometers thick ( Figure 2A; Figure 3A ) ( Shi et al . , 2013; Nannenga et al . , 2014a , 2014b; Hattne et al . , 2015; Liu et al . , 2016 ) . Such thin crystals are capable of producing measurable Bragg peaks because electrons interact with matter more strongly than X-rays . Indeed , we found that the nano-sized 3D crystals used for MicroED produced higher resolution diffraction than relatively larger crystals suited for structure determination at a microfocus X-ray beamline ( Figure 2A ) . Evidently , micron-thick needle crystals are sufficient for X-ray structure determination with six or seven residue peptides , but not for 11-residue peptides . These experiences closely mirrored those in the determination of the atomic structure of the toxic core of α-synuclein ( Rodriguez et al . , 2015 ) , an 11-residue segment that forms the spine of amyloid fibrils associated with Parkinson’s disease . 10 . 7554/eLife . 19273 . 007Figure 2 . Bragg peaks produced by MicroED from 3D crystals only a few hundred nanometers thick are observed at higher resolution than peaks produced by X-ray diffraction at a microfocus beamline from microcrystals 10 , 000 times larger . ( A ) 3D crystals of 19–29 S20G ( right , inset ) diffract to 1 . 6 Å using MicroED , a whole angstrom better resolution than the microcrystals of 19–29 S20G ( left , inset ) . ( B ) 3D crystals of 15–25 WT ( right , inset ) diffract to 1 . 4 Å using MicroED , whereas microcrystals of 15–25 WT diffract to 2 . 2 Å using Microfocus X-rays ( left , inset ) . DOI: http://dx . doi . org/10 . 7554/eLife . 19273 . 00710 . 7554/eLife . 19273 . 008Figure 3 . The MicroED atomic structure of segment 19–29 S20G reveals pairs of β-sheets mated by a dry interface . ( A ) Electron micrograph of 3D crystals used for data collection . The red circle represents the area of the crystal used for diffraction . ( B ) Pairs of β-sheets are oriented face-to-face and they are tightly mated by a dry interface that excludes water . The dry interface is formed by tightly packed , interdigitating side-chains . This panel shows 5 β-strands or layers along the ‘a’ dimension of the unit cell; the average crystal used for data collection is 10 , 400 layers long in the ‘a’ dimension . ( C ) Orthogonal view of the steric-zipper formed by the dry interface . ( D ) The similarity between the fiber diffraction pattern calculated from the structure shown in Panel C and the fiber diffraction observed from full-length hIAPP fibrils supports the dry interface as a model for the amyloid spine of full-length hIAPP fibrils . Along the meridian ( left panel ) , the dry interface and full-length hIAPP fibrils share reflections at 4 . 7 Å and and 2 . 4 Å ( black arrows ) . Additionally , along the off-meridonal , the diffraction patterns share a reflection at 3 . 7 Å . It is difficult to see the reflection at 2 . 4 Å in the full-length hIAPP fiber diffraction image , but the reflection is clearly visible in the radial profile in Figure 3—figure supplement 1 . Along the equator ( right panel ) , the dry interface and full-length hIAPP fibrils share reflections at 10 . 0 Å and 5 . 0 Å ( black arrows ) . The right panel is magnified 2X to more clearly show the low-resolution reflections along the equator . DOI: http://dx . doi . org/10 . 7554/eLife . 19273 . 00810 . 7554/eLife . 19273 . 009Figure 3—figure supplement 1 . The crystal packing of segment 19–29 S20G reveals a second interface , termed the ‘Back-to-Back’ or wet interface , which does not form the amyloid spine . The wet interface does not form the amyloid spine because ( 1 ) the fiber diffraction pattern calculated from this interface does not match the fiber diffraction pattern collected from full-length hIAPP fibrils , ( 2 ) it contains waters , and ( 3 ) it possesses less side-chain interdigitation than the dry interface . ( A ) View of crystal packing down the ‘a’ dimension of the unit cell reveals the two different interfaces . The unit cell is outlined in red; waters are shown in cyan . ( B ) Orthogonal view of the wet interface . The wet interface buries only 153 Å2 of surface area per strand and it possesses a shape complementarity of 0 . 64 . DOI: http://dx . doi . org/10 . 7554/eLife . 19273 . 00910 . 7554/eLife . 19273 . 010Figure 3—figure supplement 2 . Scatter plot of sheet RMSD from planarity values for all hIAPP protein segment structures determined to date . The values for the 19–29 S20G and 15–25 WT atomic structures are highlighted in green and purple , respectively . DOI: http://dx . doi . org/10 . 7554/eLife . 19273 . 01010 . 7554/eLife . 19273 . 011Figure 3—figure supplement 3 . 19-29 WT and S20G have similar fibrillar structures . ( A ) Side-by-side comparison of X-ray diffraction from 19–29 WT fibrils ( left ) and 19–29 S20G fibrils ( right ) . ( B ) . Overlaid radial profiles calculated from X-ray fiber diffraction in panel A . 19–29 WT ( black ) and S20G ( gray ) fibrils share strong reflections at 4 . 6 Å , 8 . 4–9 Å , and 34 . 7 Å . DOI: http://dx . doi . org/10 . 7554/eLife . 19273 . 01110 . 7554/eLife . 19273 . 012Figure 3—figure supplement 4 . Radial profile calculated from the X-ray diffraction pattern given by cytotoxic full-length hIAPP fibrils . Cytotoxic full-length hIAPP fibrils were prepared under the same conditions as in the cytotoxicity assays . Next , the fibrils were pelleted by centrifugation , washed with water to remove salt , and then applied between two glass capillary tubes as described by Sunde and co-workers ( Sunde et al . , 1997 ) . The first peak on the left , which occurs at 37 . 3 Å , may represent a legitimate feature of full-length hIAPP fibrils , but it is too close to the beam stop to definitively makes this conclusion . DOI: http://dx . doi . org/10 . 7554/eLife . 19273 . 012 The MicroED atomic structure of segment 19–29 S20G reveals pairs of parallel in-register β-sheets mated face-to-face by interdigitation of side-chains and exclusion of water molecules ( Figure 3B and C;Table 1 ) . This arrangement is termed a class I steric-zipper . Such features are observed for amyloid spines of other proteins and have been associated with pathology ( Nelson et al . , 2005; Sawaya et al . , 2007; Ivanova et al . , 2009; Colletier et al . , 2011; Liu et al . , 2011 ) . This zipper contains a tightly packed hydrophobic core consisting of Phe23 , Ala25 , and Ile27 . Phe23 is the central and largest contributor the hydrophobic core , consistent with multiple other experiments ( Tenidis et al . , 2000; Griffiths et al . , 1995; Jack et al . , 2006; Madine et al . , 2008 ) . The dry interface buries 265 Å2 of surface area per strand , which equates to 24 Å2 per residue . This interface is one of the largest and most complementary of any structurally determined steric-zipper interface ( Supplementary file 1 ) ; it has a shape complementary of 0 . 85 . The dry interface is nearly as large as the toxic core of α-synuclein ( Rodriguez et al . , 2015 ) , but with higher shape complementarity . 10 . 7554/eLife . 19273 . 013Table 1 . Statistics of MicroED data collection and atomic refinement . DOI: http://dx . doi . org/10 . 7554/eLife . 19273 . 013Sample19–29 S20G 15–25 WTExcitation Voltage ( kV ) 200200Electron Sourcefield emission gunfield emission gunWavelength ( Å ) 0 . 02510 . 0251Total dose per crystal ( e−/ Å2 ) 3 . 42 . 9Frame rate ( frame/s ) 0 . 3–0 . 5 0 . 3–0 . 5 Rotation rate ( °/s ) 0 . 30 . 3# crystals used66Total angular rotation collected ( ° ) 6868Merging Statistics 19–29 S20G 15–25 WTspace groupP212121P1Unit cell dimensionsa , b , c ( Å ) 4 . 78 , 18 . 6 , 70 . 811 . 68 , 18 . 18 , 19 . 93α , β , γ ( ° ) 90 , 90 , 9062 . 8 , 88 . 9 , 87 . 6Resolution ( Å ) 1 . 91 . 4Rmerge10 . 6% ( 15 . 0% ) 19 . 9% ( 50% ) # of reflections1380 ( 221 ) 9014 ( 153 ) Unique reflections548 ( 115 ) 2180 ( 84 ) Completeness83% ( 65% ) 75% ( 35 . 3% ) Multiplicity2 . 5 ( 1 . 9 ) 4 . 1 ( 1 . 8 ) I/σ5 . 65 ( 3 . 65 ) 4 . 33 ( 1 . 10 ) CC1/2 ( Diederichs , 2013 ) 98 . 9% 98 . 5% Refinement Statistics 19–29 S20G 15–25 WTReflections in working set5462177Reflections in test set53218Rwork†22 . 75% 22 . 47% Rfree27 . 49% 25 . 90% RMSD bonds ( Å ) 0 . 010 . 008RMSD angles ( ° ) 1 . 21 . 2Ramachandran ( % ) ‡Favored100100Allowed00Outliers00PDB ID code5KNZ5KO0EMDB ID codeEMD-8272EMD-8273*Highest resolution shell shown in parenthesis . †Rfactor=100x∑||Fobs|−|Fcalc||/∑|Fobs| Fcalc and Fobs are the calculated and observed structure factor amplitudes , respectively . Rwork refers to the Rfactor for the data utilized in the refinement and Rfree refers to the Rfactor for 10% of the reflections randomly chosen that were excluded from the refinement . ‡Percentage of residues in Ramachandran plot regions were determined using Molprobity ( Chen et al . , 2010 ) . The β-sheets of the 19–29 S20G atomic structure possess a curvature that is not common in shorter hIAPP protein segments ( Wiltzius et al . , 2008 , 2009a; Soriaga et al . , 2015 ) . To assess β-sheet curvature , we compared the root mean square deviations ( RMSD’s ) of sheets from planarity across all hIAPP protein segment atomic structures determined to date ( Supplementary file 1 ) . The 19–29 S20G structure ranks in the upper half of the list ( Figure 3—figure supplement 2 ) , containing both sheet curvature and a sharp kink . Most of the shorter peptides are nearly flat , but some have sharp kinks . The significance of deviation from planarity is not yet clear . The similarity between the fiber diffraction pattern calculated from this steric-zipper and the fiber diffraction pattern collected from full-length hIAPP fibrils tends to validate the 19–29 S20G atomic structure as a model for the amyloid spine of full-length hIAPP ( Figure 3D ) . The diffraction patterns share several key features , including reflections at 4 . 7 Å and 2 . 4 Å along the meridian , a reflection at 3 . 7 Å along the off-meridian ( left panel ) , and reflections at 10 . 0 Å and 5 . 0 Å along the equator ( right panel ) . Structural studies performed here and elsewhere by others suggest that 19–29 WT can form a similar dry interface to the one observed in the 19–29 S20G atomic structure . Radial profiles calculated from X-ray fiber diffraction of 19–29 WT and 19–29 S20G fibrils show strong reflections in common at 4 . 6 Å , 8 . 4 Å and 8 . 7 Å , and 34 . 7 Å , indicative of interstrand , intersheet , and proto-filament spacing , respectively ( Figure 3—figure supplement 3 ) . A previous study of 20–29 WT fiber diffraction revealed comparable reflections , which the authors used to formulate a fibril model of 20–29 WT that roughly agrees with our 19–29 S20G atomic structure ( Madine et al . , 2008 ) . Our atomic structure and their model differ by a small shift in registration between sheets , allowing for tighter packing in the atomic structure . These results are consistent with earlier findings by Cao and co-workers , who observed that hIAPP-WT fibrils seed hIAPP-S20G fibril formation , thus suggesting a shared fibrillar structure ( Cao et al . , 2012 ) . Although the WT and mutant segments likely form similar structures , the structure of the mutant segment may be more stable . The stability of the mutant segment may stem from the early onset Gly20 mutation , which adopts an unusual geometry ( φ = −101 . 7° and ψ = 107 . 5° ) that creates a kink in the peptide backbone . To investigate this hypothesis , we generated a model of 19–29 WT consisting of a mated pair of ten-stranded sheets . The model was identical to the 19–29 S20G atomic structure with the exception that we adjusted the backbone torsion angles of Ser20 to comply with the allowed regions of the Ramachandran plot for a non-glycine residue . We compared the energies of the WT and S20G structures after minimization with FoldIt ( Cooper et al . , 2010 ) . The dry interfaces are nearly identical between the two segments , except near Asn21 , where the altered backbone torsion angles break the canonical Asn ladder hydrogen bonding interactions with neighboring Asn21 residues within the sheet and instead , form hydrogen bonds with Ser29 from the opposing sheet . The alteration separates the pair of sheets by approximately 1 . 5 Å in this region , and therefore the 19–29 S20G structure has a slightly lower energy than 19–29 WT ( −590 REU vs . −535 REU ) . The atomic structure of segment 15–25 WT , also determined using MicroED ( Figure 2B; Figure 4A ) , shows an arrangement of unmated β-sheets composed of anti-parallel out-of-register β-strands that is uncharacteristic of pathogenic amyloid fibrils ( Figure 4B;Table 1 ) . Most pathogenic amyloid fibrils are composed of β-strands that stack perpendicular to the sheet-long axis , but the β-strands in out-of-register structures stack at an angle . Deviation of strands from the fibril perpendicular is a natural consequence of the registration shift implied by out-of-register structures . The out-of-register β-strands are stabilized by extensive hydrogen bonding . Within each sheet , the β-strands form two distinct , unequal interfaces: a stronger interface with twelve hydrogen bonds , and a weaker interface with eight hydrogen bonds ( Figure 4B ) . This inequality between interfaces has been observed in previous examples of out-of-register sheets ( Soriaga et al . , 2015; Laganowsky et al . , 2012; Liu et al . , 2012; Yu et al . , 2015 ) . A view down the ‘proto-fibril axis’ of the crystal shows that the faces of adjacent sheets are wet and overlap only partially ( Figure 4C ) ; the asymmetric unit contains density for seven ordered water molecules and one thiocyanate molecule . The area buried between adjacent sheets is small ( 10 . 7 Å2 per residue ) compared to the average steric-zipper ( 20 . 1 Å2 per residue ) . Hence , there is no dry interface between adjacent sheets in the crystal , and the structure seems labile compared to that of 19–29 S20G . 10 . 7554/eLife . 19273 . 014Figure 4 . Segment 15–25 WT forms an arrangement of unmated β-sheets that is uncharacteristic of pathogenic amyloid fibrils . ( A ) Electron micrograph of 3D crystals used for data collection . The red circle represents the area of the crystal used for diffraction . ( B ) A single β-sheet contains anti-parallel out-of-register β-strands stabilized by two distinct , unequal interfaces: a stronger interface with twelve hydrogen bonds , and a weaker interface with eight hydrogen bonds . The β-strands are out-of-register by two residues because Leu16 on the first β-strand is directly above His18 on the third β-strand . ( C ) The view down the proto-fibril axis reveals hydrated interfaces between partially overlapping β-sheets . Notice that adjacent β-sheets lack side-chain interdigitation . Water molecules are shown as cyan spheres . The thiocyanate molecule is highlighted in gold in the central β-sheet and colored gray in the peripheral β-sheets . DOI: http://dx . doi . org/10 . 7554/eLife . 19273 . 01410 . 7554/eLife . 19273 . 015Figure 4—figure supplement 1 . 15-25 WT fibrils are relatively weak and reversible compared to 19–29 S20G fibrils . Equimolar concentrations of 15–25 WT and 19–29 S20G fibrils were treated with increasing amounts of SDS and then heated at 55°C for 20 min . ( A ) Turbidity measurements of the fibrils treated with heat and increasing amounts of SDS reveal that 15–25 WT fibrils disaggregate more readily than 19–29 S20G fibrils . Turbidity measurements were obtained by recording absorbance at 340 nm . ( B ) Negative-stain electron micrographs corroborate the results observed in the turbidity measurements . Scale bars are 1 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 19273 . 01510 . 7554/eLife . 19273 . 016Figure 4—figure supplement 2 . Scatter plot of sheet RMSD from planarity values for all hIAPP protein segment structures determined to date . The values for the 19–29 S20G and 15–25 WT atomic structures are highlighted in green and purple , respectively . DOI: http://dx . doi . org/10 . 7554/eLife . 19273 . 01610 . 7554/eLife . 19273 . 017Figure 4—figure supplement 3 . 15-25 WT and S20G have similar fibrillar structures . ( A ) Side-by-side comparison of X-ray diffraction from 15–25 WT fibrils ( left ) and 15–25 S20G fibrils ( right ) . ( B ) Overlaid radial profiles calculated from X-ray fiber diffraction in panel A . 15–25 WT ( black ) and S20G ( gray ) fibrils display strong reflections at 4 . 7 Å , 9 . 3–9 . 8 Å , 18 . 2 Å , and 37 . 0 Å . DOI: http://dx . doi . org/10 . 7554/eLife . 19273 . 017 Consistent with our observation of unmated β-sheets in the atomic structure , we observe that 15–25 WT fibrils are relatively weak and reversible compared to 19–29 S20G fibrils , which possess a canonical pathogenic amyloid fibril architecture . Turbidity readings followed by negative-stain EM reveal that 15–25 WT fibrils completely disaggregate in the presence of heat and 1% SDS , but 19–29 S20G fibrils remain intact in up to 2% SDS ( Figure 4—figure supplement 1 ) . Similar to 19–29 S20G , the 15–25 WT atomic structure reveals curved β-sheets . The sheets possess one of the highest RMSD’s of sheets from planarity for any hIAPP protein segment structure determined to date ( Supplementary file 1 , Figure 4—figure supplement 2 ) . X-ray fiber diffraction and radial profile analysis of 15–25 WT and 15–25 S20G fibrils indicate they form structures similar to each other ( Figure 4—figure supplement 3 ) . Taken together with the X-ray fiber diffraction data from the 19–29 segments , we conclude that the early onset S20G mutation does not confer a fibril morphology distinguishable from wild-type . Next we investigated the cytotoxic effects of the spine segments in order to determine if any of them were similarly cytotoxic to full-length hIAPP preparations . Although the cytotoxic mechanism of hIAPP is not fully understood , several reports show hIAPP induces mitochondrial dysfunction , alters cell metabolism , and initiates activation of pro-apoptotic machinery ( Butler et al . , 2003; Mulder and Ling , 2009; Zraika et al . , 2010; Magzoub and Miranker , 2012; Tomasello et al . , 2014 ) . Based on these findings , we tested the cytotoxicity of the spine segments using MTT dye reduction ( Mosmann , 1983; Liu and Schubert , 1997 ) and a FRET-based biosensor to assay altered metabolism and pro-apoptotic machinery activation ( Paulsson et al . , 2008 ) , respectively . Using MTT dye reduction , we observe that the labile 15–25 fibrils are not cytotoxic to HEK293 cells ( Figure 5A ) , whereas 19–29 S20G fibrils have comparable cytotoxicity to full-length hIAPP fibrils ( Figure 5B ) . To verify the cytotoxic effects of each sample , we examined the morphology of the treated cells under a light microscope . Additionally , in the context of residues 19–29 , the S20G segment is significantly more cytotoxic than the WT segment , consistent with parent full-length hIAPP ( Sakagashira et al . , 2000; Meier et al . , 2016 ) ( Figure 5B ) . We did not detect any oligomers present in the 15–25 WT or 19–29 S20G fibril samples using the LOC antibody ( Figure 5—figure supplement 1 ) . 10 . 7554/eLife . 19273 . 018Figure 5 . Segment 19–29 S20G forms the toxic core of hIAPP and segments 15–25 are not toxic . ( A and B ) Fibrils were formed by incubating the spine segments overnight under quiescent conditions , the same conditions used to prepared full-length hIAPP fibrils . Next , the samples were applied to HEK293 cells at the specified concentrationsand then cell viability was quantified using MTT dye reduction . Bars show median cell viability; dashed lines show median cell viability from 10 μM mIAPP and hIAPP . ( A ) 15–25 WT and 15–25 S20G fibrils are not toxic compared to full-length hIAPP fibrils ( n = 12 across four biological replicates , each with three technical replicates ) . ( B ) 19–29 WT fibrils are mildly cytotoxic and 19–29 S20G fibrils are significantly more cytotoxic than 19–29 WT fibrils ( ****p<0 . 0001 using a Mann-Whitney U test; n = 12 across four biological replicates , each with three technical replicates ) . 19–29 S20G fibrils ( 10 μM ) are similarly cytotoxic to full-length hIAPP fibrils at the same concentration ( lower dashed line ) ( p=0 . 09 using an unpaired t-test with equal standard deviations ) . ( C ) The insoluble fraction of the 50 μM 19–29 S20G cytotoxic preparation contains the cytotoxic species . 19–29 S20G fibrils were formed overnight at room temperature and then pelleted by centrifugation . The soluble fraction was carefully removed and then filtered to ensure it contained no insoluble material . The insoluble material was resuspended in its original volume . Each sample was applied to HEK293 cells and then cell viability was quantified with MTT dye reduction ( ***p<0 . 0002 using an ordinary one-way ANOVA; n = 3 technical replicates ) ( D ) and ( E ) Using a FRET-based biosensor assay for monitoring caspase-3 activity in real-time , 19–29 S20G fibrils induce the most caspase-3 activity , whereas segments 15–25 did not induce caspase-3 activity , consistent with the MTT dye reduction assay results . 50 μM of each spine segment seeded with 166 nM seeds was applied to stably transfected CHO cells . ( D ) Fold difference was recorded over 24 h . Datapoints represent average fold difference . The dashed line represents the 16 h mark . ( E ) Average levels of caspase-3 activation after a 16 h incubation relative to untreated cells ( ***p<0 . 0002; ****p<0 . 0001 using an ordinary one-way ANOVA , Bonferroni correction; n = 5 technical replicates ) . DOI: http://dx . doi . org/10 . 7554/eLife . 19273 . 01810 . 7554/eLife . 19273 . 019Figure 5—figure supplement 1 . Fibrillar samples of 15–25 WT and 19–29 S20G do not contain detectable amyloid oligomers . Oligomers were probed using a dot bot assay with the polyclonal anti-oligomer antibody , LOC . hIAPP oligomers were used as the positive control for LOC binding . DOI: http://dx . doi . org/10 . 7554/eLife . 19273 . 01910 . 7554/eLife . 19273 . 020Figure 5—figure supplement 2 . ( − ) -epigallocatechin gallate ( EGCG ) , a flavanol known to mitigate full-length hIAPP cytotoxicity by preventing it from forming fibrils , likewise mitigates 19–29 S20G cytotoxicity by preventing it from forming fibrils . ( A ) Chemical structure of EGCG . ( B ) Negative-stain electron micrographs reveal that EGCG mitigates 19–29 S20G fibril formation . 19–29 S20G was incubated overnight at room temperature under quiescent conditions in buffer alone or with equimolar concentration of ECGC . Next , the samples were spotted onto carbon-coated copper grids for negative-stain EM analysis . ( C ) EGCG mitigates 19–29 S20G cytotoxicity . Samples were generated as described in panel B and then applied to HEK293 cells . Cell viability was quantified using MTT dye reduction . Columns indicate median cell viability . Different symbols correspond to values observed in each independent experiment ( ***p=0 . 0004 using a unpaired t-test with Welch’s correction for unequal variances; n = 9 across three biological replicates , each with three technical replicates ) . ( D ) Negative-stain EM reveals that EGCG does not mitigate fibril formation of 15–25 WT , a spine segment that does not possess a hydrophobic core . 15–25 WT was incubated for five days under shaking conditions with equimolar concentrations of EGCG . Next , the samples were spotted onto carbon-coated copper grids for negative-stain EM . DOI: http://dx . doi . org/10 . 7554/eLife . 19273 . 020 Based on our examination of the insoluble and soluble fractions of the cytotoxic 19–29 S20G sample , we determine that the cytotoxicity of 19–29 S20G mainly resides in its fibrillar form . We tested the cytotoxicity of the total , insoluble and soluble fractions of the 19–29 S20G sample to HEK293 cells using MTT dye reduction . We observe that the insoluble fraction , which contains amyloid fibrils , is similarly cytotoxic to the total ( Figure 5C ) , just as we observed with full-length hIAPP ( Figure 1D , Figure 1—figure supplement 1B ) . These results suggest that 19–29 S20G may form the toxic spine of full-length hIAPP . Further evidence that 19–29 S20G may form the toxic spine of full-length hIAPP comes from our observation that ( − ) -epigallocatechin gallate ( EGCG ) , a flavanol known to mitigate full-length hIAPP cytotoxicity by preventing hIAPP from forming fibrils ( Meng et al . , 2010 ) , also mitigates 19–29 S20G cytotoxicity by preventing it from forming fibrils ( Figure 5—figure supplement 2B and C ) . We hypothesize that EGCG may mitigate fibril formation of full-length hIAPP and 19–29 S20G by binding to a common site , such as the dry interface of the amyloid spine . A previous study suggested EGCG may mitigate hIAPP fibril formation by binding hIAPP via hydrophobic interactions ( Young et al . , 2015 ) . Indeed , EGCG does not prevent fibril formation of 15–25 WT , which does not possess a dry hydrophobic interface ( Figure 5—figure supplement 2D ) . In addition , these results further support our conclusion that preparations of segment 19–29 S20G that contain fibrils are cytotoxic . Next we tested whether the spine segments activate pro-apoptotic machinery using a FRET-based biosensor assay for monitoring caspase-3 activity in real-time ( Paulsson et al . , 2008 ) . In this assay , CHO cells are stably transfected with a construct containing enhanced cyan fluorescent protein ( ECFP ) and enhanced yellow fluorescent protein ( EYFP ) fused by a DEVD linker . FRET signal is observed by exciting ECFP at 440 nm . In cells undergoing apoptosis , active caspase-3-like proteases target and cleave the DEVD linker , resulting in loss of FRET signal . Cell viability is measured by monitoring the ratio of 540 nm/480 nm , which reports loss of FRET signal and increased caspase-3 activity . Using this system , we observe that segment 19–29 S20G elicits the most caspase-dependent cytotoxicity of the spine segments and segments 15–25 are not cytotoxic ( Figure 5D and E ) . Segment 19–29 S20G is not as cytotoxic as full-length hIAPP in this assay , possibly because hIAPP interaction with heparan sulfate proteoglycans ( HSPG ) is important for apoptosis induction ( Oskarsson et al . , 2015 ) , and residues 1–8 , which are missing in all of the spine segments , are required for hIAPP binding to HSPG’s . Given that the spine segments seed full-length hIAPP fibril formation and that 19–29 S20G and 15–25 WT fibrils elicit different cytotoxic effects , we investigated whether seeding with either of the spine segments alters hIAPP cytotoxicity . To do this , we prepared seeded hIAPP at 10 μM with 10% monomer equivalent of pre-formed seeds , the same conditions used in the ThT assay in Figure 1G . For all cytotoxicity assays , we dilute samples 1 to 10 to the concentration specified in culture medium containing pre-plated cells . Thus , we tested the cytotoxicity of seeded hIAPP at 1 μM in order to preserve the conditions of the ThT assay . Using MTT dye reduction , we observe that hIAPP seeded with non-toxic 15–25 WT fibrils is less cytotoxic than hIAPP alone , but hIAPP seeded with stable , toxic 19–29 S20G fibrils is similarly cytotoxic to hIAPP alone ( Figure 6 ) . Likewise , hIAPP seeded with stable , toxic 19–29 S20G fibrils is significantly more cytotoxic than hIAPP seeded with labile , non-toxic 15–25 WT fibrils ( Figure 6 ) . Seeds alone are not cytotoxic , indicating the cytotoxic effects we observe originate from the interaction of each seed with hIAPP and not the seed alone . 10 . 7554/eLife . 19273 . 021Figure 6 . Fibril seeds of 15–25 WT reduce the cytotoxicity of full-length hIAPP . In this experiment , we incubated 10 μM hIAPP with or without 10% monomer equivalent of pre-formed seeds overnight under quiescent conditions , the same conditions used to seed full-length hIAPP fibril formation in Figure 1 . Next , we diluted the samples 1 to 10 in culture media containing pre-plated Rin5F cells . Note: the concentration of IAPP used in this experiment is less than the IAPP concentrations used in the cytotoxicity assays in Figures 1 and 5 . hIAPP seeded with stable , toxic 19–29 S20G fibrils is more cytotoxic to Rin5F cells than hIAPP seeded with labile , non-toxic 15–25 WT fibrils . Columns indicate median cell viability ( ns = not significant; **p=0 . 006; ****p<0 . 0001 using an unpaired t-test with equal standard deviations , n = 9 across three biological replicates , each with three technical replicates ) . 19–29 S20G seeds and 15–25 WT seeds ( 100 nM each ) are not cytotoxic to Rin5F cells . DOI: http://dx . doi . org/10 . 7554/eLife . 19273 . 02110 . 7554/eLife . 19273 . 022Figure 6—figure supplement 1 . hIAPP fibrils made by seeding with each spine segment have slightly different structural features . ( A ) Negative-stain electron micrographs reveal fibrils made by seeding with each spine segment do not have markedly different morphologies . Unseeded hIAPP forms pre-dominantly striated ribbons of uniform width that bundle together; some fibrils have twists ( black arrow head ) . hIAPP fibrils made by seeding with stable , toxic 19–29 S20G fibrils form striated ribbons of varying widths and some twisted fibrils ( black arrow head ) . hIAPP fibrils made by seeding with labile , non-toxic 15–25 WT fibrils form striated ribbons that appear slightly thinner than unseeded hIAPP fibrils . ( B ) and ( C ) X-ray fiber diffraction and radial profile analysis suggest hIAPP fibrils made by seeding with toxic and non-toxic segments have slightly different structures . hIAPP fibrils made by seeding with stable , toxic 19–29 S20G fibrils display shorter Bragg spacings compared to hIAPP fibrils made by seeding with labile , non-toxic 15–25 WT fibrils . The shorter spacings suggest tighter fibril packing . DOI: http://dx . doi . org/10 . 7554/eLife . 19273 . 022 There are two possible explanations for the reduced cytotoxicity of the seeded 15–25 WT sample . First , the 15–25 WT seeds may seed a non-toxic species of full-length hIAPP , or second , the 15–25 WT seeds may interact with full-length hIAPP in some way that reduces its cytotoxicity . X-ray fiber diffraction and radial profile analysis of the hIAPP fibrils used in the cytotoxicity assay reveal that fibrils formed by seeding with stable , toxic 19–29 S20G fibrils have a slightly tighter packing than fibrils formed by seeding with labile , non-toxic fibrils . hIAPP fibrils formed by seeding with stable , toxic 19–29 S20G fibrils exhibit reflections indicative of shorter equatorial Bragg spacings than hIAPP seeded with labile , non-toxic 15–25 WT fibrils ( 9 . 0 Å versus 10 . 0 Å ) ( Figure 6—figure supplement 1 ) . The tighter packing of these fibrils may explain their enhanced cytotoxicity . Fiber diffraction could not be detected from seeds alone prepared under the same conditions .
In 1901 , when Dr . Eugene Opie first observed islet amyloid in post-mortem pancreata of T2D patients , he proposed a link between the islet amyloid and T2D ( Opie , 1901 ) . Over a century later , multiple studies have shown an unequivocal link between hIAPP aggregation and T2D , but uncertainty remains about which type of hIAPP aggregate contributes to pancreatic β-cell death . Although most recent in vitro studies suggest soluble oligomers are the primary type of toxic aggregate , here , we find hIAPP samples that contain fibrils alter pancreatic β-cell metabolism and activate pro-apoptotic caspases . These findings motivated us to determine the structure of the spine of hIAPP fibrils and elucidate structural features important for hIAPP cytotoxicity . To improve our likelihood of crystallization and structure determination , we selected four protein segments that span the spine . We discovered that segment 19–29 S20G forms a pair of β-sheets mated at a dry interface , a structure that shares key features with full-length hIAPP fibrils as described in the following paragraph . What’s more , the fibrillar form of 19–29 S20G is cytotoxic . In contrast , segment 15–25 WT forms an unusual arrangement of single , out-of-register β-sheets that are not cytotoxic . The divergence in structure and cytotoxicity of segments 19–29 S20G and 15–25 WT suggests that strong , stable intermolecular interactions are important features of cytotoxic amyloid proteins . The experiments of this study show that the 19–29 S20G atomic structure recapitulates many of the structural features and cytotoxic properties of hIAPP . First , preparations of 19–29 S20G that contain fibrils are cytotoxic , as is the case for full-length hIAPP . Second , X-ray fiber diffraction calculated from the dry interface of the 19–29 S20G atomic structure shares key features with fiber diffraction collected from full-length hIAPP fibrils . Third , segment 19–29 S20G elicits cytotoxicity by altering cell metabolism and activating pro-apoptotic machinery , mechanisms by which full-length hIAPP fibrils are thought to contribute to pancreatic β-cell death during T2D . Fourth , the early onset S20G mutation confers greater cytotoxicity within segment 19–29 and within full-length hIAPP . Last , EGCG , a flavanol that mitigates full-length hIAPP fibril formation and cytotoxicity , likewise mitigates 19–29 S20G fibril formation and cytotoxicity . These results , taken together with the canonical pathogenic amyloid fibril architecture of segment 19–29 S20G , suggest it represents the toxic amyloid spine of hIAPP . Our studies begin to provide a framework for understanding which hIAPP fibril polymorphs may contribute to pancreatic β-cell death during T2D . Previous structural studies of hIAPP protein segments ( Wiltzius et al . , 2008 , 2009a; Soriaga et al . , 2015 ) and full-length hIAPP ( Luca et al . , 2007; Weirich et al . , 2016; Goldsbury et al . , 1997; Kajava et al . , 2005; Bedrood et al . , 2012; Wineman-Fisher et al . , 2015 ) identified an array of structures with diverse side-chain and sheet arrangements; the 15 hIAPP protein segment structures that overlap the hIAPP amyloid spine belong to six different steric-zipper classes ( Figure 7 ) . These multiple diverse structures suggest there is significant polymorphism within the hIAPP amyloid spine , but exactly which of these polymorphs elicit cytotoxicity was not known . By studying the structures and cytotoxic effects of protein segments in parallel , we identify a cytotoxic hIAPP fibril structure that may contribute to pancreatic β-cell death during T2D . Additionally , our studies suggest that not all hIAPP fibril structures are cytotoxic . 10 . 7554/eLife . 19273 . 023Figure 7 . Schematic of structural features of all hIAPP protein segment structures determined to date . Parallel ( // ) or Anti-parallel ( A// ) refers to the orientation of β-strands within β-sheets . Registry refers to the translational offset of β-sheets perpendicular to the fiber axis . DOI: http://dx . doi . org/10 . 7554/eLife . 19273 . 023 Both atomic structures presented here reveal a new and potentially important observation: curved β-sheets . In the dry interface of the 19–29 S20G atomic structure , the curved β-sheets accommodate the tightly packed hydrophobic core , which includes a bulky phenylalanine , while maintaining high shape complementarity and large buried surface area . Paradoxically , in the 15–25 WT atomic structure , the curved β-sheets appear to have an opposite effect: the curved β-sheets appear to prevent adjacent sheets from associating to form a canonical pathogenic amyloid fibril architecture . In both atomic structures , the effect of the curved β-sheets is dictated by the registry of adjacent β-sheets ( Supplementary file 1 , Figure 7 ) . The 15–25 WT atomic structure joins the recently discovered class of out-of-register protein segment structures , which exert disparate cytotoxic effects . Here , we show that 15–25 WT is not cytotoxic but in contrast , the out-of-register protein segment KDWSFY from β2-microglobulin elicits mild cytotoxicity ( Liu et al . , 2012 ) . One notable difference between the two structures is that the 15–25 WT structure is formed of single sheets , while the KDWSFY structure is formed of sheets mated by a dry interface . The dry interface of the KDWSFY atomic structure results in a higher solvation energy per strand compared to the 15–25 WT atomic structure ( 122 cal/mol/strand vs . 19 cal/mol/strand ) . Given that cytotoxic structures like 19–29 S20G have relatively high solvation energies per strand ( 279 cal/mol/strand; Supplementary file 1 ) , this difference may explain the disparate cytotoxic effects of the two out-of-register structures . However , we need more studies of out-of-register protein structures and their cognate cytotoxic effects to definitively make this conclusion . The disparate cytotoxic effects within this structure class lead us to believe that the nature of cytotoxicity is not simply conferred by in-register or out-of-register structures . As many studies have suggested , there may be more than one mechanism of amyloid-related toxicity and the different mechanisms may be catalyzed by different architectures . Alternatively , maybe if additional residues were included , the anti-parallel out-of-register fiber could be stabilized , thereby increasing its toxicity . Although the 19–29 WT fibrils prepared in this study appear morphologically similar to 19–29 S20G fibrils , the 19–29 WT fibrils are likely polymorphic and may contain some fraction of fibrils that are structurally similar to non-toxic 15–25 WT fibrils . Previous structural studies of segment 20–29 WT fibrils show that it forms an array of polymorphs , some of which are similar to the 15–25 WT atomic structure ( Griffiths et al . , 1995; Jack et al . , 2006; Madine et al . , 2008; Ashburn et al . , 1992; Nielsen et al . , 2009 ) . Structural polymorphism of 19–29 WT fibrils may explain their lower cytotoxicity than 19–29 S20G fibrils , which are homogenous in structure . These findings , expedited by MicroED , may inform our understanding of hIAPP fibril structures that contribute to pancreatic β-cell death in Type-II Diabetes patients . Going forward , we can use our toxic amyloid spine model as a template for structure-based design in the effort to develop much needed therapeutics that protect against pancreatic β-cell death and disease progression ( Sievers et al . , 2011; Kahn et al . , 2014 ) . In addition , if hIAPP fibrils truly are a major type of toxic aggregate that contributes to T2D , then raising antibodies against hIAPP fibrils may represent a promising strategy for therapeutic development , especially in light of the recent success of preliminary studies with antibodies raised against amyloid-β ( Sevigny et al . , 2016 ) .
Human IAPP ( 1–37 ) -NH2 wild-type and mouse IAPP ( 1–37 ) NH2 wild-type were synthesized by Innopep ( San Diego , CA ) and CS Bio ( Menlo Park , CA ) and purified to greater than 98% purity . Human and mouse IAPP were prepared by dissolving the lyophilized proteins at 0 . 25–1 mM in 100% HFIP and leaving them to dissolve for several hours to ensure complete solubility . Next , the HFIP was removed with a CentriVap Concentrator ( Labconco , Kansas City , MO ) . After removal of the HFIP , the peptides were dissolved at 1 mM , 5 mM , or 10 mM in 100% DMSO . The DMSO peptide stocks were diluted 100-fold in filter-sterilized Dulbecco’s PBS ( Cat . # 14200–075 , Life Technologies , Carlsbad , CA ) . Samples were incubated at room temperature for the designated time periods . All four spine segments were synthesized by GenScript ( Piscataway , NJ ) and purified to greater than 98% purity . Fibrils were formed by dissolving lyophilized peptide at 1 mM in PBS and 1% DMSO . 15-FLVHSSNNFGA-25 ( 15–25 WT ) . 15–25 WT was dissolved at 20 mg/ml in ice-cold , nano-pure water and then spin-filtered . Crystals were grown using the hanging drop vapor diffusion method at 4°C in 0 . 35 M NaSCN and 35% MPD . Crystals grew within several hours and reached maximum size in a week . 3D crystals only a few hundred nanometers thick grew alongside microcrystals in the same drops . 19-SGNNFGAILSS-29 ( 19–29 S20G ) . Microcrystals were grown using the hanging drop vapor diffusion method at 30°C in 0 . 2M acetate salts and 40% MPD . 3D crystals only a few hundred nanometers thick were grown in batch by dissolving lyophilized peptide at 1 mM in PBS and 1% DMSO without seeding . Crystals grew on the bench top at room temperature in several hours . The procedures for MicroED data collection and processing largely follow published procedures ( Hattne et al . , 2015; Shi et al . , 2016 ) . Briefly , a 2–3 μL drop of crystals in suspension was deposited onto a Quantifoil holey-carbon EM grid then blotted and vitrified by plunging into liquid ethane using a Vitrobot Mark IV ( FEI , Hillsboro , OR ) . Blotting times and forces were optimized to keep a desired concentration of crystals on the grid and to avoid damaging the crystals . Frozen grids were then either immediately transferred to liquid nitrogen for storage or placed into a Gatan 626 cryo-holder for imaging . Images and diffraction patterns were collected from crystals using an FEG-equipped FEI Tecnai F20 TEM operating at 200 kV and fitted with a bottom mount TVIPS TemCam-F416 CMOS-based camera . Diffraction patterns were recorded by operating the detector in a movie mode termed ‘rolling shutter’ with 2×2 pixel binning ( Nannenga et al . , 2014b ) . Exposure times for these images were either 2 or 3 s per frame . During each exposure , crystals were continuously unidirectionally rotated within the electron beam at a fixed rate of 0 . 3 degrees per second , corresponding to a fixed angular wedge of 0 . 6 or 0 . 9 degrees per frame . Crystals that appeared visually undistorted and that were 100–300 nm thick produced the best diffraction . Datasets from individual crystals were merged to improve completeness and redundancy . Each crystal dataset spanned a wedge of reciprocal space ranging from 40–80° . We used a selected area aperture of approximately 1 μm . The geometry detailed above equates to an electron dose rate of less than 0 . 01 e−/Å2 per second being deposited onto our crystals . Measured diffraction images were converted from TVIPS format into SMV crystallographic format , using in-house software ( available for download at http://cryoem . janelia . org/downloads ) ( Hattne et al . , 2015 ) . We used XDS to index and integrate the diffraction images and XSCALE ( Kabsch , 2010 ) for merging and scaling together datasets originating from different crystals . For 19–29 S20G , data from six crystals were merged to assemble the dataset used for molecular replacement . Of note , the resolution was cut off at 1 . 9 Å to facilitate subsequent rounds of structure refinement . For , 15–25 WT , data from six crystals were merged to assemble the dataset used for molecular replacement . Of note , the diffraction pattern from the 15-25 WT crystals diffracted with MicroED reveal a pseudo two-fold symmetry . In line with this observation , we indexed and integrated the diffraction images with space group C2 , but the datasets had relatively poor statistics compared to the P1 datasets and our attempts at refining molecular replacement solutions from the C2 datasets failed . 19–29 S20G . We determined the structure using molecular replacement . An idealized 7-residue poly-alanine strand led us to our atomic model . The solution was identified using Phaser ( McCoy , 2007 ) . A dataset merged from six crystals was used to identify the initial model , but subsequent rounds of model building and refinement were carried out using a dataset from a single crystal . Free R flags were copied over from the dataset merged from six crystals to the single crystal dataset . Subsequent rounds of model building and refinement were carried out using COOT and Phenix , respectively ( Emsley and Cowtan , 2004; McCoy et al . , 2005 ) . Electron scattering factors were used for refinement . 15–25 WT . We determined the structure using molecular replacement . Dozens of search models were used , but an out-of-register β-strand model led us to our solution . The solution was identified using Phaser ( McCoy , 2007 ) . Subsequent rounds of model building and refinement were carried out using COOT and Phenix , respectively ( Emsley and Cowtan , 2004; McCoy et al . , 2005 ) . Electron scattering factors were used for refinement . To aid in model building , we used a feature enhanced map ( FEM ) , which sharpens B factors at high resolution ( Afonine et al . , 2015 ) . Calculations of the area buried and shape complementarity ( SC ) were performed with AREAIMOL ( Lee and Richards , 1971; Collaborative Computational Project , Number 4 , 1994 ) and SC ( Connolly , 1983; Richards , 1977; Lawrence and Colman , 1993 ) , respectively . 30 μL of human and mouse IAPP preparations used in the cytotoxicity assays in Figure 1 were pipetted into a black-wall 384-well plate and then mixed with 3 μL of 1 mM Thioflavin-T ( ThT ) . Fluorescence was recorded with an excitation wavelength of 444 nm and an emission wavelength of 482 nm . 1 μL of each sample generated for cytotoxicity assays in Figure 1 and Figure 5A and B was applied to a nitrocellulose membrane ( Cat . # 162–0146 , BioRad , Hercules , CA ) . Next , the membrane was blocked in 5% ( w/v ) nonfat dry milk in PBS-T ( T = 0 . 1% ( v/v ) Tween-20 ( Cat . #BP337-500 , Fisher ) ) for 1 hr at room temperature . After blocking , the membrane was incubated with a 1:100 dilution of LOC polyclonal rabbit serum ( Pacific Immunology , Ramona , CA ) in 5% ( w/v ) milk in PBS-T at 4°C overnight . The membrane was washed in PBS-T for 10 min three times , and then incubated with anti-rabbit secondary antibody ( RRID:AB_2307391; Cat . #111-035-144 , Jackson ImmunoResearch , West Grove , PA ) diluted 1:10 , 000 in PBS-T for 1 hr at RT . The membrane was washed three more times , and then the signal was developed with Clarity Western ECL Substrate ( Cat . #170–5061 , BioRad ) and documented with a CCD camera . Exposures ranging from 5 s to 5 min were collected , but the 5 min exposure was used in all figures . Samples were spotted onto grids ( holey or non-holey ) and allowed to settle on the grid for 160 to 180 s . Remaining liquid was wicked off and grids were left to dry before analyzing . Sample grids were analyzed on the TF20 Electron Microscope ( FEI , Hillsboro , OR ) . Images were collected at 3500 or 6000x magnification with an additional 1 . 4x post-column magnification and recorded using a TIETZ F415MP 16 megapixel CCD camera . Samples for negative-stain EM were spotted on non-holey carbon-coated grids andallowed to settle on the grid for 160 to 180 s . Remaining liquid was wicked off and then 2% uranyl acetate was applied to the grid . After 1 min , the uranyl acetate was wicked off . The grids were left to dry before analyzing on the T12 Electron Microscope ( FEI ) . Images were collected at 3 , 200 or 15 , 000x magnification and recorded using a Gatan 2kX2k CCD camera . Rin5F cells were purchased from ATCC ( RRID:CVCL_2177; Cat . # CRL-2058 , Manassas , VA ) . Cells were cultured in RPMI media ( ATCC , Cat . # 30–2001 ) plus 10% heat-inactivated fetal bovine serum . Cells were cultured at 37°C in a 5% CO2 incubator . They tested negative for mycoplasma using a MycoAlert PLUS Detection Kit ( Cat . #: LT07-701 , Lonza , Switzerland ) and they were authenticated using Cytochrome C Oxidase 1 ( COX1 ) gene analysis by Laragen ( Culver City , CA ) . HEK293 c18 cells ( hereon referred to as HEK293 ) were a gift from Carol Eng in the laboratory of Arnold J . Berk at UCLA , but they were originally purchased from ATCC ( RRID:CVCL_6974 ) . Cells were cultured in DMEM media ( Cat . # 11965–092 , Life Technologies ) plus 10% heat-inactivated fetal bovine serum and 1% pen-strep ( Life Technologies ) . Cells were cultured at 37°C in a 5% CO2 incubator . They tested negative for mycoplasma using a MycoAlert PLUS Detection Kit and they were authenticated using STR profiling ( Laragen ) . CHO cells were purchased from ATCC ( RRID:CVCL_0214; Cat . #: CCL-61 ) . Cells were cultured in RPMI 1640 with 11 mM glucose ( Sigma ) with 10% FBS , and 1% pen-strep . Cells were cultured at 37°C in a 5% CO2 incubator . They tested negative for mycoplasma using a PCR-based method and they were authenticated using mRNA analysis . Spine segments were dissolved at 1 mM in PBS with 1% DMSO . Samples were incubated at room temperature for 15 hr or up to one week under quiescent conditions to form fibrils . The presence of fibrils was confirmed with electron microscopy . Fibril samples were diluted appropriately for cell viability assays and fibril formation assays . Fibrils were spun down and washed with water three times to remove any salt . Fibrils of spine segments were spun down using a tabletop microfuge . Full-length hIAPP fibrils and spine segment seeds were spun down using an Airfuge Ultracentrifuge set at 75 , 000 rpm for 1 hr ( Beckman-Coulter , Brea , CA ) . The samples were concentrated 10x in water and applied between two capillary ends and then the samples were left to dry overnight . Dried fibrils of spine segments and full-length hIAPP in Figure 3D were analyzed with a RIGAKU R-AXIS HTC imaging plate detector using Cu K ( alpha ) radiation from a FRE+ rotating anode generator with VARIMAX HR confocal optics ( Rigaku , Tokyo , Japan ) . Fiber diffraction from full-length hIAPP fibrils used in Figure 6 was recorded by an ADSC Q315 CCD detector at the Advanced Photon Source 24-ID-E beamline ( Argonne , IL ) . Radial profiles were calculated using a program written in-house . The program calculates the average intensity as a function of distance from the beam center . Thioflavin-T ( ThT ) assays were performed in black 96-well plates ( Nunc , Rochester , NY ) sealed with UV optical tape . hIAPP and mIAPP were dissolved at 1 mM in 100% HFIP . The peptides were then diluted 100-fold in 20 mM sodium acetate pH 6 . 5 and 10 μM ThT . Unsonicated fibril seeds were added at 1 μM monomer equivalent concentration ( 10% v/v ) . ThT fluorescence was recorded with excitation and emission of 444 nm and 482 nm , respectively , using a SpectraMax M5 ( Molecular Devices , Sunnyvale , CA ) . Experiments were performed in quadruplicate and readings were recorded every 3 min . To investigate whether 19–29 WT could form a similar structure to 19–29 S20G , we modeled a serine at position 20 in the 19–29 S20G atomic structure . We adjusted the backbone torsion angles so that they fell within the ‘allowed’ regions of the Ramachandran plot for a non-glycine residue ( Emsley and Cowtan , 2004 ) . We performed energy minimization using FoldIt ( RRID:SCR_003788 ) ( Cooper et al . , 2010 ) and compared the energies of the resulting models of 19–29 WT and 19–29 S20G . HEK293 cells and Rin5F cells were plated at 10 , 000 and 27 , 000 cells per well in 90 μL , respectively , in 96-well plates ( Cat . # 3596 , Costar , Tewksbury , MA ) . Cells were allowed to adhere to the plate for 20–24 hr . For the assay in Figures 1 and 50 μM full-length IAPP was aged in vitro for the designated incubation times . To generate the soluble and insoluble fractions , the ‘hIAPP 24 h’ preparation was centrifuged at 21 , 000xg for 45 min and then the supernatant , which is the soluble fraction , was carefully removed and transferred to a 0 . 1 μm spin filter tube . Next , the supernatant was filtered and the pelleted material , which is the insoluble fraction , was resuspended in the original total volume . For the assays in Figure 5 and Figure 5—figure supplement 2 , 1 mM spine segment and 100 μM full-length IAPP samples were generated by preparing the samples as described previously and then incubating them for 15 hr at room temperature under quiescent conditions . After the incubation period , the spine segments were diluted appropriately . For all assays , 10 μL of sample was added to cells . By doing this , samples were diluted 1/10 from in vitro stocks . Experiments were done in triplicate . The appropriate statistical test for significance was determined by assessing whether ( 1 ) The sample sets had a Gaussian distribution using a D’Agostino-Pearson omnibus normality test and ( 2 ) The sample sets had equal variance using a Bartlett’s test or F test . For samples with Gaussian distributions and equal variances , we employed an unpaired t-test with equal standard deviations . For samples with Gaussian distributions , but unequal variances , we employed an unpaired t-test with Welch’s correction . For samples with non-Gaussian distributions and unequal variances , we employed a Mann-Whitney U-test . After a 24 hr incubation of samples with cells , 20 μL of Thiazolyl Blue Tetrazolium Bromide MTT dye ( Sigma , St . Louis , MO ) was added to each well and incubated for 3 . 5 h at 37°C under sterile conditions . The MTT dye stock is 5 mg/mL in Dulbecco’s PBS . Next , the plate was removed from the incubator and 100 μL of MTT stop solution ( Cat . #4101 , Promega , Madison , WI ) was added to each well . We ensured the MTT crystals were fully dissolved by placing the plates on an orbital shaker ( slow speed ) for about an hour prior to taking measurements . Alternatively , the MTT assay was stopped by carefully aspirating off the culture media and adding 100 μL of 100% DMSO to each well . Absorbance was measured at 570 nm using a SpectraMax M5 . A background reading was recorded at 700 nm and subsequently subtracted from the 570 nm value . Cells treated with vehicle alone ( PBS + 0 . 1% DMSO ) were designated at 100% viable , and cell viability of all other treatments was calculated accordingly . For the MTT reduction assay in Figure 6 , a single data point from the mIAPP sample set was deemed an outlier based on 2 lines of evidence: ( 1 ) The data point was identified as an outlier using a Grubb’s test ( α = 0 . 1 ) for outliers using the n = 9 sample set , and ( 2 ) When the sample set was pooled with more data collected for different experiments ( n = 42 ) , the data point was identified as an outlier using a more stringent Grubb’s test ( α = 0 . 01 ) . We used the caspase-3/7 GLO assay ( Cat . # G8091 , Promega , Madison , WI ) to detect caspase-3/7 activation . For this assay , Rin5F cells were plated as previously described in white-walled 96-well plates ( Cat . # 3917 , Costar , Tewksbury , MA ) . After the designated aging period of each hIAPP preparation , 10 μL of sample was added to cells and thus diluted 1/10 from in vitro stocks . Experiments were performed in triplicate . Samples were incubated with cells for 24 h . Next , cell culture media , caspase-3/7 reagent , and the cells were brought to room temperature . All media was aspirated from wells and then replaced with 25 μL of media and 25 μL of caspase-3/7 reagent and mixed thoroughly . The plate was incubated at room temperature for 30 min and then luminescence was measured using a SpectraMax M5 . Experimental points were normalized to vehicle-treated cells , which were designated as 100% . Cells treated with 2 μM staurosporine were used as a positive control to ensure the assay kit worked correctly . CHO cells were stably transfected with a vector producing EYFP and ECFP connected via a short linker containing the Asp-Glu-Val-Asp ( DEVD ) sequence targeted by activated caspase-3 . The short linker allows fluorescence energy transfer ( FRET ) to occur between the two fluorophores . During apoptosis activated caspase-3 cleaves the linker resulting in a loss of FRET measured as a reduced 540 nm/480 nm emission ratio . Cells were plated at 25 , 000 cells per well in black 96-well optical bottom plates ( Nunc , Grand Island , NY ) and the assay was performed in Krebs-Ringer ( 120 mM NaCl , 4 . 7 mM KCl , 2 . 5 mM CaCl2 , 1 . 2 mM MgSO4 , 0 . 5 mM KH2PO4 , pH 7 . 4 ) supplemented with 20 mM HEPES and 2 mM glucose ( KRHG ) . hIAPP peptides ( 1–37 , 15–25 WT and S20G , 19–29 WT and S20G ) ( final peptide concentration 50 μM in 1% DMSO ) were mixed with sonicated , preformed fibrils ( seeds ) made of the same peptide ( corresponding to 166 nM of monomers ) and immediately added to the plated cells . FRET was monitored in real-time by measuring emission at 480 nm and 540 nm with 440 nm excitation in a FLUOstar Omega microplate reader ( BMG Labtech ) over 24 hr at 37°C . Fibrils of 15–25 WT and 19–29 S20G at monomer equivalent concentrations were allowed to form for one week to ensure complete fibril formation . The samples were homogenized with vortexing , and then aliquoted to 0 . 5 mL tubes with equal volumes . Each fibril sample was treated with water or increasing amounts of SDS , and then heated at 55°C for 20 min . Next , an aliquot of each sample was transferred to a 384-well plate and turbidity was measured by recording absorbance at 340 nm . Each fibril sample was spotted onto a grid for negative-stain EM to analyze fibril abundance . The experiment was repeated twice , but the results of 1 experiment are shown in Figure 4—figure supplement 1 . | In Type-II Diabetes , an individual’s cells fail to respond correctly to the hormone insulin , leaving them unable to counteract high levels of sugar in the blood . Another hormone , human islet amyloid polypeptide ( hIAPP ) , works with insulin to regulate blood sugar levels . hIAPP is an amyloid protein , which means that it can lose its normal structure and form fibrils . Fibrils are difficult for cells to break down and are often associated with disease . Indeed , fibrils of hIAPP often form in the pancreas as part of Type-II Diabetes . Some studies have shown that hIAPP fibrils are toxic to pancreatic cells and worsen the symptoms of Type-II Diabetes . Others suggest that it is the process of fibril formation that is toxic , not the fibrils themselves . Although the structures of the fibrils have been described , whether these structures cause cell toxicity has not been investigated . Krotee et al . have now explored the structures of two overlapping segments of hIAPP using a new cryo electron microscopy method called MicroED that is ideal for studying such segments . One segment , called 19-29 S20G , forms a standard amyloid fibril structure that is similar to the structure of full-length hIAPP fibrils . Adding these segments to human cells causes similar levels of toxicity as the full-length hIAPP fibrils . The second segment , called 15-25 WT , forms a non-toxic structure that is less stable than standard amyloid fibrils . The results presented by Krotee et al . support the view that standard amyloid fibril structures are toxic to cells and suggest that 19-29 S20G may be a good model to use when studying how full-length hIAPP fibrils behave . The structure of 19-29 S20G may also be useful as a template for designing molecules that block amyloid fibril growth . If amyloid fibrils cause cell toxicity in the pancreas , then these molecules could be used to treat Type-II Diabetes . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"biochemistry",
"and",
"chemical",
"biology",
"structural",
"biology",
"and",
"molecular",
"biophysics"
] | 2017 | Atomic structures of fibrillar segments of hIAPP suggest tightly mated β-sheets are important for cytotoxicity |
Morphogenesis of hierarchical vascular networks depends on the integration of multiple biomechanical signals by endothelial cells , the cells lining the interior of blood vessels . Expansion of vascular networks arises through sprouting angiogenesis , a process involving extensive cell rearrangements and collective cell migration . Yet , the mechanisms controlling angiogenic collective behavior remain poorly understood . Here , we show this collective cell behavior is regulated by non-canonical Wnt signaling . We identify that Wnt5a specifically activates Cdc42 at cell junctions downstream of ROR2 to reinforce coupling between adherens junctions and the actin cytoskeleton . We show that Wnt5a signaling stabilizes vinculin binding to alpha-catenin , and abrogation of vinculin in vivo and in vitro leads to uncoordinated polarity and deficient sprouting angiogenesis in Mus musculus . Our findings highlight how non-canonical Wnt signaling coordinates collective cell behavior during vascular morphogenesis by fine-tuning junctional mechanocoupling between endothelial cells .
Morphogenesis is driven by coordinated and dynamic cell movements , which are regulated by a combination of chemical and physical cues ( Jaalouk and Lammerding , 2009 ) . Morphogenic cues are sensed and read at the single cell-level , yet biomechanical information is relayed to and integrated by neighboring cells leading to tissue-level collective cell behaviors . These emergent collective behaviors arise by mechanically coupling cadherin-based adhesion and actomyosin-based contraction , allowing propagation of cell-cell interactions across large cell populations ( Friedl and Mayor , 2017; Lecuit and Yap , 2015; Yap et al . , 2018 ) . One of such morphogenic processes is the formation of blood vessels . Vascular morphogenesis occurs mainly through sprouting angiogenesis , a process where endothelial tip cells lead the vascular sprout , migrate and invade into avascular tissues in response to pro-angiogenic molecules . Endothelial stalk cells follow tip cells contributing to sprout elongation and branch formation through proliferation and migration ( Potente and Mäkinen , 2017 ) . Although sprouting angiogenesis is considered a collective cell migration process ( Friedl and Gilmour , 2009; Vitorino and Meyer , 2008 ) , little is known about the mechanisms regulating this collective behavior . Recently , endothelial cell front-rear polarity has emerged as a crucial regulator of collective behavior in sprouting angiogenesis . In fact , endothelial cell ( EC ) -specific deletion of NCK1/2 and Cdc42 impairs cell polarity , which correlates with decreased sprouting efficiency ( Dubrac et al . , 2016; Laviña et al . , 2018 ) . However , the mechanisms controlling and coordinating polarity patterns of endothelial cells during sprouting angiogenesis remain elusive . Recent reports showed that non-canonical Wnt signaling , a known regulator of cell migration and cell polarity in key morphogenic events such as gastrulation , neural tube closure , fur orientation , and ureteric bud formation ( Gray et al . , 2011; Yang and Mlodzik , 2015 ) , also controls sprouting angiogenesis and vascular remodeling ( Franco et al . , 2016; Korn et al . , 2014 ) . Non-canonical Wnt signaling was shown to control vascular remodeling by blocking excessive vessel regression in a flow-dependent manner ( Franco et al . , 2016; Korn et al . , 2014 ) . In this context , non-canonical Wnt signaling modulates the threshold for flow-dependent EC polarization , inducing premature vessel regression , and leading to a decrease in vessel density ( Franco et al . , 2016 ) . In parallel , abrogation of endothelial non-canonical Wnt ligands also leads to reduce sprouting efficiency ( Franco et al . , 2016; Korn et al . , 2014 ) . Yet , it remains unresolved how mechanistically non-canonical Wnt signaling regulates sprouting angiogenesis . Here , we have established a simple assay to measure endothelial collective cell behavior in vivo and in vitro using axial polarity histograms . Using this assay , we uncovered a novel Wnt5a pathway that stabilizes the binding of vinculin to ɑ-catenin at adherens junctions , and consequently the efficient coupling between adherens junctions and the actin cytoskeleton in endothelial cells . We showed that vinculin loss-of-function impairs collective polarity in vivo and in vitro , leading to deficient sprouting angiogenesis . Overall , we propose that non-canonical Wnt signaling coordinates collective cell behavior during vascular morphogenesis by fine-tuning junctional mechanocoupling between endothelial cells .
Non-canonical Wnt signaling deficiency leads to impaired sprouting angiogenesis , a process that requires extensive cell migration ( Franco et al . , 2016; Korn et al . , 2014 ) . To investigate the role of non-canonical Wnt ligands in endothelial cell migration , we used a well-characterized model of collective cell migration , the scratch-wound assay ( Tambe et al . , 2011 ) . Wnt5a is the major non-canonical Wnt ligand operating in vivo ( Franco et al . , 2016; Korn et al . , 2014 ) and in vitro ( Figure 1—figure supplement 1A ) . In the scratch-wound assay , siRNA-mediated knockdown ( KD ) of Wnt5a , hereafter siWNT5a , significantly impaired wound closure and straightness of cell migration without affecting cell velocity in human umbilical vein endothelial cells ( HUVECs ) ( Figure 1A ) . Accordingly , single cell tracking highlighted coordinated collective behavior in control siRNA ( siControl ) cells , whereas siWNT5a cells showed uncoordinated migration paths ( Figure 1B ) . The correlation length calculated from particle image velocimetry ( PIV ) analysis ( Ng et al . , 2012; Petitjean et al . , 2010 ) , confirmed loss of coordinated cell migration in siWNT5a cells , although not to the same extent as in cells treated with siRNA against alpha-E-catenin ( ɑ-catenin/CTNNA1 ) , a crucial component of adherens junctions and indispensable for collective cell migration ( Figure 1C ) ( Bazellières et al . , 2015 ) . Axial polarity correlates with the direction of migration in endothelial cells in vivo and in vitro ( Franco et al . , 2015; Kwon et al . , 2016 ) . Taking advantage of this feature , we generated a simplified method , compared to PIV analysis , to quantify the degree of coordination between cells by measuring the front-rear cell polarity ( nucleus-to-Golgi apparatus axis ) at the population level . The angular histogram of axial polarities relative to the wound-edge displays the distribution of cell polarities in the monolayer relative to the wound-edge ( Figure 1D ) . As a measure of collective polarization , we defined a polarity index ( PI , see Materials and methods ) , which ranges from 1 ( strongly polarized ) to 0 ( random distribution ) ( Figure 1D ) . The PI represents the length of the mean resultant vector ( Berens , 2009 ) . Using this approach , we measured PIs in consecutive 50 µm-wide areas from the leading edge towards the monolayer ( details in Materials and methods ) . As expected , siCTNNA1 led to a generalized poor collective coordination of polarities demonstrated by low PIs throughout the monolayer ( Figure 1E ) . According to the PI equation , perfect randomization should give a PI = 0 . However , ɑ-catenin KD cells shows PI >0 , which highlights a polarity bias caused by geometrical constraints that are generated by the free space-cell monolayer interface . Therefore , we used the polarity patterns of siCTNNA1 cells to define the threshold of PI that defines uncoordinated behavior . We established this PI threshold by determining the mean ± SD of the results obtained from the siCTNNA1 experiments across the monolayer . For the calculation of the mean value , we excluded the first row of cells , as these were strongly affected by wound-monolayer asymmetry , leading to a stronger polarity towards the wound . Taking these rules , we defined the PI threshold for uncoordinated migration at PI = 0 . 14 ( corresponding to the upper limit of the mean ± SD , PI = 0 . 1 ± 0 . 04 , in ɑ-catenin KD experiments ( Figure 1E ) . SiControl cells showed coordination of cell polarities up to ~300 µm from the leading edge ( Figure 1E ) . Remarkably , siWNT5a cells showed uncoordinated polarity starting at ~150 µm from the leading edge ( Figure 1E ) . In the wound assay , coordinated migration emerges because leader cells , localized at the edge of the monolayer , are polarized due to the presence of a free edge , and instruct follower cells’ directionality of migration through force transmission at adherens junctions ( Etienne-Manneville and Hall , 2001; Friedl and Mayor , 2017 ) . To understand the extent to which the polarization patterns of leaders and followers were affected in their polarization patterns , we measured the PI for leaders ( 1st row of cells ) and followers ( 2-5th row of cells ) separately ( Figure 1F ) . Leader cells showed polarization towards the leading edge above random in all three groups: siControl ( PI = 0 . 638 ) , siWNT5a ( PI = 0 . 493 ) and siCTNNA1 cells ( PI = 0 . 358 ) ( Figure 1G , H; Figure 1—figure supplement 1B , C ) . However , siWNT5a ( PI = 0 . 104 ) and siCTNNA1 ( PI = 0 . 101 ) follower cells showed randomized polarity patterns whilst siControl follower cells displayed coordinated polarity patterns ( PI = 0 . 345 ) ( Figure 1I , J; Figure 1—figure supplement 1B , C ) . Defects in collective polarity in siWNT5a follower cells were rescued by re-expression of exogenous WNT5a ( Figure 1J ) . Cryptic lamellipodia in follower cells have been associated with collective cell migration ( Das et al . , 2015 ) . Thus , we examined if WNT5a plays a role in the formation of these pro-migratory structures . We observed that WNT5a deficiency did not compromise the formation of cryptic lamellipodia but it affected their orientation toward the leading edge ( Figure 1—figure supplement 1D , E ) . Taken together , these results indicate that WNT5a signaling is necessary to coordinate the behavior of follower cells at the population-level . In vivo , endothelial tip cells lead the vascular sprout , whilst endothelial stalk cells follow tip cells and contribute to sprout elongation ( Potente et al . , 2011 ) . In order to evaluate if Wnt5a also regulates collective cell polarity in vivo , we calculated PIs for endothelial cells at the vascular sprouting front in control and non-canonical Wnt signaling-deficient mouse retinas ( Figure 1K , L ) . Remarkably , we observed a significant decrease in polarity patterns of mutant retinas compared to WT retinas , similar to the effect in the in vitro experiments . We observed collective polarization ( PI = 0 . 187 ) in control retinas demonstrating that the PI is able to capture collective behavior during sprouting angiogenesis . Whilst , non-canonical Wnt ligand-deficient showed a PI close to randomization ( PI = 0 . 094 ) ( Figure 1M , N ) . Thus , endothelial-derived non-canonical Wnt signaling is required for the coordination of collective cell polarity in vitro and in vivo . To understand how Wnt5a mechanistically controls collective behavior , we analyzed its effects on the adherens junction complex , a key mediator of collective cell migration ( Tambe et al . , 2011 ) . We first characterized the different junctional arrangements in endothelial cells , which are associated with low or high junctional tension ( Huveneers et al . , 2012 ) . We observed that siWNT5a cells had a significant decrease in the frequency of high-force serrated junctions , and a concomitant increase in the frequency of low-force reticular junctions ( Figure 2A , B ) . Reduction in the number of high-force junctions correlated with a decrease in the association between VE-cadherin and actin stress fibers ( Figure 2C , D ) , suggesting that Wnt5a depletion might negatively impact on force transmission through adherens junctions . To test this hypothesis , we used atomic force microscopy to probe mechanical strength of cell-cell interactions ( Figure 3A ) . Control-control cell interactions required on average 1 . 0 fJ ( 1 . 0 × 10−15 J ) of work ( energy ) for complete cell-cell detachment ( Figure 3B ) . siWNT5a-siWNT5a cell interactions required significant less work ( 0 . 5 fJ; p<0 . 0001 ) for complete cell separation ( Figure 3B ) . EGTA treatment , which chelates extracellular calcium and abolishes cadherin-dependent interactions , significantly reduced the strength of interactions between siControl cells , and canceled the differences between siControl and siWNT5a conditions ( Figure 3B ) . VE-cadherin-depleted cells showed a very similar strength of interaction as EGTA-treated cells ( Figure 3B ) . A detailed analysis of the frequency of detachment force of each cell-cell interactions in siVE-cadherin condition highlights that the majority of strong cell-cell contacts are mediated by VE-cadherin homophilic interactions ( Figure 3C ) . These are significantly reduced in siWNT5a cells ( Figure 3C–F ) , suggesting that Wnt5a signaling increases the strength of cell-cell interactions through adherens junctions . Strength of adhesion at adherens junctions relies on efficient coupling between the cytoplasmic VE-cadherin C-terminus tail and the actin cytoskeleton ( Gumbiner , 2005 ) . To confirm that WNT5a regulates tension in VE-cadherin , we used previously characterized FRET-based VE-cadherin tension sensors ( Conway et al . , 2013 ) . In Wnt5a-depleted cells , VE-cadherin FRET efficiency was significantly higher than in siControl cells , implying lower level of junctional tension ( Figure 3G , H ) . Force-insensitive VE-cadherin FRET sensors showed similar levels between siControl and siWNT5a cells ( Figure 3I ) . Taken together , these data demonstrate that Wnt5a signaling promotes high tension at the VE-cadherin intracellular domain and strengthens cell-cell interactions . Next , we investigated why loss of siWNT5a results in decreased coupling between adherens junctions and the actin cytoskeleton . First , we quantified the expression levels of key junctional proteins . We confirmed that levels of VE-cadherin , ß-catenin , ɑ-catenin , or vinculin were unaltered between control and Wnt5a-deficient cells ( Figure 4A , B ) . Next , we assessed the spatial distribution of components of the VE-cadherin complex by co-localization experiments ( Figure 4C , D ) . Interestingly , we observed a significant decrease of VE-cadherin co-localization with vinculin in siWNT5a cells but no change in co-localization with other junctional proteins ( Figure 4D ) . We further confirmed a specific decrease in vinculin recruitment to VE-cadherin in siWNT5a cells by proximity ligation assay ( PLA ) and co-immunoprecipitation ( co-IP ) in wounded monolayers ( Figure 4E–H ) . Altogether , these results indicate that Wnt5a is important to recruit and/or to stabilize vinculin binding to adherens junctions , which in turn is necessary for efficient collective cell polarity . Vinculin binds adherens junctions via ɑ-catenin . It has also been proposed that a conformational change in ɑ-catenin promotes vinculin recruitment and binding to adherens junctions ( le Duc et al . , 2010; Yao et al . , 2014; Yonemura et al . , 2010 ) . To test whether the impaired vinculin co-localization with VE-cadherin arises from defective ɑ-catenin conformational change or from the inability to recruit vinculin once opened , we used a specific antibody that recognizes ɑ-catenin in its open conformation ( ɑ18 antibody ) ( Yonemura et al . , 2010 ) . ɑ18 antibody-VE-cadherin co-localization showed a significant but mild decrease ( ~15% ) in Wnt5a-depleted cells . Yet , the decrease in vinculin-ɑ18 antibody co-localization was stronger ( ~32% ) in these same cells ( Figure 5A , B ) , suggesting a possible defect in vinculin junctional localization even when ɑ-catenin is in its open conformation . To clarify whether WNT5a affects recruitment or stabilization of vinculin to junctions , we quantified the dynamics of vinculin recruitment to adherens junctions at newly formed cell-cell junctions by performing a calcium-switch experiment in siControl and siWNT5a cells . Remarkably , the initial dynamics of vinculin recruitment were similar between siControl and siWNT5a cells . However , a significant decrease of VE-cadherin-vinculin co-localization in siWNT5a cells was observed 30 min after junction reassembly ( Figure 5C ) . This suggests that rather than controlling its initial recruitment , Wnt5a signaling regulates vinculin stabilization at junctions . The role of vinculin in adherens junctions’ mechanical coupling between cells , and in the regulation of collective behavior have been recently established in in vitro studies ( Bazellières et al . , 2015; Seddiki et al . , 2018 ) . Accordingly , vinculin loss-of-function ( LOF ) in the scratch-wound assay results in impaired collective cell polarity and migration in vitro , as reflected by the decrease in the closure rate ( Figure 6A–C ) . In contrast , the role of vinculin in collective cell migration in vivo remains controversial ( Alatortsev et al . , 1997; Han et al . , 2017 ) . Thus , we next evaluated the relevance of vinculin in collective polarity in vivo , using the mouse retina model of angiogenesis . We crossed the Vinculin floxed mouse ( Zemljic-Harpf et al . , 2007 ) together with the Pdgfb-iCre mouse ( Claxton et al . , 2008 ) to genetically abrogate vinculin expression in endothelial cells in post-natal mice . Vinculin endothelial-specific KO ( EC-KO ) mice showed decreased radial expansion , decreased vessel density ( Figure 6D , E ) , and a significant increase in the number of vessel regression profiles ( Figure 6F ) . Strikingly , analysis of polarity patterns of endothelial cells at the sprouting front demonstrated that Vinculin EC-KO have a significant decrease in PI when compared with control littermates ( Figure 6G–I ) . Altogether , these results indicate that Vinculin is necessary for efficient collective cell polarity in endothelial cells in vitro and in vivo . Remarkably , the Vinculin EC-KO phenotype shows strong similarities with the one reported for non-canonical Wnt signaling EC-KO not only in terms of radial expansion , vessel density and regression profiles ( Franco et al . , 2016 ) , but also in terms of polarity patterns ( Figure 1M , N ) , suggesting that Vinculin might participate in a pathway regulated by non-canonical Wnt signaling . Our cumulative observations place junctional vinculin as the main mediator of Wnt5a signaling in collective cell behavior . This prompted us to test whether reinstating junctional vinculin activity would rescue Wnt5a deficiency . To this end , we overexpressed either full-length chicken vinculin ( Vinc-FL ) or chicken vinculin T12 ( Vinc-T12 ) in siControl and siWNT5a cells . Vinc-T12 carries four amino acid mutations in its protein sequence which weaken the affinity of the auto-inhibitory head-to-tail interaction by 100-fold ( Cohen et al . , 2005 ) . Thus , Vinc-T12 is considered to be a constitutively active vinculin . We confirmed that both constructs were able to efficiently rescue polarity defects of siVinculin cells ( Figure 7—figure supplement 1 ) . Overexpression of either form of vinculin did not affect significantly the strength of polarity of control cells ( Figure 7A , B ) . Remarkably , Vinc-T12 but not Vinc-FL rescued impaired polarity of Wnt5a KD cells ( Figure 7A , B , and Figure 7—figure supplement 2 ) . Furthermore , overexpression of Vinc-T12 but not Vinc-FL led to a rescue in the organization of junctions in siWNT5a cells , promoting the formation of serrated high-tension junctions with the concomitant decrease in reticular junctions ( Figure 7C , D ) . To confirm if vinculin’s actin binding properties are required downstream of Wnt5a signaling pathway , we overexpressed a fusion protein containing the β-catenin binding domain of ɑ-catenin and the actin-binding domain of vinculin ( Figure 8A ) ( Maddugoda et al . , 2007 ) . ɑ-catenin-vinculin ( ɑCat-Vinc ) fusion protein strongly localizes to adherens junctions ( Figure 8B ) . ɑCat-Vinc overexpression did not significantly affect the overall PI of control cells , whilst it completely rescued collective cell polarity defects in siWNT5a cells ( Figure 8C , D and Figure 8—figure supplement 1 ) . Moreover , ɑCat-Vinc overexpression was sufficient to rescue cell migration straightness , the ratio of displacement to trajectory length , in siWNT5a cells ( Figure 8E ) . Altogether , these observations are highly indicative that Wnt5a signaling leads to the activation of vinculin at adherens junctions to promote stable interactions between ɑ-catenin and the actin cytoskeleton . To investigate how Wnt5a signaling leads to vinculin activity at the junctions , we screened for cell polarity defects upon downregulation of several known receptors for non-canonical Wnt ligands . Of all receptors tested , siROR2 was the only one phenocopying WNT5a depletion ( Figure 9A , B and Figure 9—figure supplement 1 ) . Moreover , siROR2 cells also showed a significant decrease in VE-cadherin-vinculin co-localization ( Figure 9C ) . ROR2 is a tyrosine kinase receptor and it has been shown to activate JNK , Rac1 and Cdc42 pathways downstream of Wnt5a stimulation ( Green et al . , 2014; Lee and Heur , 2014; Schambony and Wedlich , 2007; Stricker et al . , 2017 ) . Inhibition of Rac1 or JNK did not affect collective cell polarity ( Figure 9D ) . However , inhibition or siCdc42 impaired collective polarity of endothelial cells ( Figure 9D–F ) . In accordance , siCdc42 impaired vinculin co-localization with VE-cadherin ( Figure 9G ) . Analogous to siWNT5a , siCdc42 showed a significant decrease in the number of high-force serrated junctions ( Figure 9H , I ) , and a significant reduction in the association between actin stress fibers and VE-cadherin ( Figure 9J ) . PAK1-PBD-mediated pull-down of active GTP-bound Cdc42 confirmed that Wnt5a activates Cdc42 via ROR2 ( Figure 9K ) . Moreover , using a FRET sensor of active Cdc42 ( Cdc42-2G ) ( Martin et al . , 2016 ) , we observed activation of Cdc42 at cell-cell boundaries in siControl cells ( Figure 9L , M , Videos 1 and 2 ) . Interestingly , siWNT5a cells showed a significant decrease in the number of Cdc42-activation peaks at cell junctions between leader-follower or follower-follower cells when compared to siControl cells , whilst activation at the leading edge of leader cells was comparable between conditions ( Figure 9N and Videos 3 and 4 ) . To test whether Cdc42 regulates collective cell polarity during sprouting angiogenesis in vivo ( Laviña et al . , 2018 ) , we inhibited Cdc42 activity in postnatal mouse pups , as previously reported ( Fantin et al . , 2015 ) , and quantified collective polarity of endothelial cells at the vascular sprouting front . Remarkably , inhibition of Cdc42 led to a specific and significant randomization of endothelial cell polarity at the angiogenic sprouting front in vivo ( Figure 9O–Q ) . Thus , our results confirm that Cdc42 regulates collective cell polarity during sprouting angiogenesis in vivo . Taken together , we propose that Wnt5a signaling , through ROR2-Cdc42 activity , stabilizes vinculin at adherens junctions to reinforce its connection to the actin cytoskeleton . In this context , non-canonical Wnt signaling reinforces mechanocoupling between endothelial cells , which is essential for collective cell polarity in sprouting angiogenesis .
Sprouting angiogenesis requires efficient coordination of cell specification , cell proliferation , cell migration , and cell rearrangements . Previous work has elucidated the basic cellular and molecular mechanisms leading to endothelial tip/stalk cell specification and proliferation ( Potente and Mäkinen , 2017 ) . Yet , the mechanisms controlling collective cell polarity , migration and cell rearrangements at the vascular sprouting front are still poorly understood . Here , we identify a novel signaling pathway that reinforces mechanocoupling between endothelial cells to coordinate collective cell polarity and migration during sprouting angiogenesis . We uncover that Wnt5a , through ROR2 , activates Cdc42 at adherens junctions , which is necessary for stable binding of vinculin to ɑ-catenin , and efficient mechanocoupling between endothelial cells ( Figure 10 ) . Low non-canonical Wnt signaling weakens adherens junctions , impairs force propagation , and disrupts collective behavior of endothelial cells , which in turn affects angiogenic sprouting efficiency . We identify that Cdc42 plays an important role downstream of Wnt5a-ROR2 signaling in the regulation of vinculin’s stabilization and activation at adherens junctions . Cdc42 is a well-known regulator of cell polarity , playing important roles in yeast budding , epithelial polarity , migratory polarity and fate specification during cell division ( Heasman and Ridley , 2008 ) . In this context , Cdc42 frequently interacts with the PAR complex ( PAR6–PAR3–aPKC ) to mediate both front-rear polarity and apical-basal polarity ( Etienne-Manneville and Hall , 2001; Etienne-Manneville and Hall , 2003; Wu et al . , 2007 ) . In endothelial cells , Cdc42 was previously implicated in filopodia formation ( Barry et al . , 2015; Fantin et al . , 2015; Wakayama et al . , 2015 ) , adherence , junction stability ( Broman et al . , 2006 ) , cell migration ( Vitorino and Meyer , 2008; Wakayama et al . , 2015 ) , and more recently on collective polarity ( Laviña et al . , 2018 ) . Yet , Cdc42 seems to be dispensable for apical-basal but essential for front-rear polarization ( Laviña et al . , 2018 ) . Interestingly , non-canonical Wnt pathway was shown to cooperate with Cdc42/PAR complex to regulate front-rear polarity in migrating fibroblasts at the leading edge ( Schlessinger et al . , 2007 ) , evoking two parallel mechanisms regulating polarity of leader cells . This fits with our own results , as leader cells were mildly affected by Wnt5a KD . In endothelial cells , we show that Wnt5a regulates Cdc42 activity at cell-cell boundaries but not at the leading edge of leader cells . This signaling spatial regulation could explain why leader cells are less affected by deficient Wnt5a signaling . Indeed , Cdc42 inhibition or LOF in vitro or in vivo gives to a stronger polarity phenotype than non-canonical Wnt signaling LOF experiments ( Laviña et al . , 2018 ) . This suggests that Cdc42 is regulated by multiple inputs to control cell polarity , and that Wnt5a signaling fine-tunes Cdc42 activity at cell-cell junctions . Our data further shows that the role of non-canonical Wnt signaling on mechanocoupling relies on vinculin stabilization at adherens junctions . The biological function of vinculin at adherens junctions has been a theme of controversy . Despite being present at high-tension junctions in several model organisms , vinculin is dispensable for zebrafish and fruitfly normal development ( Alatortsev et al . , 1997; Han et al . , 2017 ) . However , its absence during mouse embryonic development results in lethal cardiovascular and neuronal defects ( Xu et al . , 1998 ) . To explain these differences , it has been proposed that mechanical and molecular properties of proteins from the adherens junctions might have diverged during evolution ( Han et al . , 2017 ) . For instance , zebrafish ɑ-catenin is monomeric and can form a complex with β-catenin and F-actin simultaneously , whilst the murine orthologue forms dimers and cannot bind simultaneously to F-actin and β-catenin in solution ( Buckley et al . , 2014; Miller et al . , 2013 ) . Thus , vinculin is required to promote efficient coupling between ɑ-catenin and F-actin in mouse . However , the factors that would regulate these interactions are so far elusive . Our results are compatible with the idea that a Wnt5a-ROR2-Cdc42 signaling axis could have evolved in mammals to enhance cadherin mechanoproperties through vinculin . Moreover , the ability to rescue the collective cell polarity defects on Wnt5a-deficient cells by re-expression of Vinc-T12 or ɑCat-Vinc fusion protein further suggests that Wnt5a modulates mechanocoupling efficiency by regulating vinculin’s actin-binding properties . How Wnt5a affects the dynamics or affinity of vinculin to actin filaments shall be investigated in future work . In addition , our results strongly suggest that Wnt5a acts as a permissive rather than an instructive cue regarding cell polarity . The ability to rescue the collective polarity phenotype of siWNT5a cells by re-expression of either Vinc-T12 or ɑCat-Vinc fusion protein implies that the polarity cue organizing collective cell polarity does not depend on Wnt5a . In this context , Wnt5a seems to be mainly necessary to potentiate mechanocoupling between cells via vinculin activation , a condition sufficient to propagate the external polarity cue in the system . This is also concordant with our previous observations that overexpression of Wnt5a in endothelial cells in vivo led to normal vascular sprouting and remodeling phenotypes ( Franco et al . , 2016 ) . Interestingly , a similar debate regarding a permissive or instructive role involves non-canonical Wnt signaling in planar cell polarity ( PCP ) establishment ( Humphries and Mlodzik , 2018 ) , where conflicting evidences exists in favor of each role . Our results suggest that non-canonical Wnt signaling plays a role in force transmission within cell populations . As mechanical cues were shown to play a contributing role in PCP establishment ( Humphries and Mlodzik , 2018 ) , a mechanobiology perspective into the function of non-canonical Wnt signaling in PCP could in part conciliate the possibility that non-canonical Wnt signaling can be seen as instructive or permissive , depending on the experimental setting . Non-canonical Wnt signaling was previously implicated in the regulation of vessel regression ( Franco et al . , 2016; Korn et al . , 2014 ) . Intriguingly , Vinculin EC-KO shows a very similar phenotype , with an increase in vessel regression , and a decrease in vessel density and radial expansion . It was suggested that non-canonical Wnt signaling regulates vessel regression by controlling a mechanosensitive threshold , based on wall shear stress , that induces endothelial cell polarization and migration ( Franco et al . , 2016 ) . The mechanisms controlling this threshold are still unclear . Given the well-known mechanoresponsive properties of vinculin , it is tempting to speculate that vinculin could also play a relevant role in establishing this threshold . Further work is necessary to clarify this question . Nevertheless , it is relevant to note that Wnt5a and vinculin regulates a different mechanosensitive pathway in flow-independent conditions . This also raises the question of how shear stress and junctional mechanotransduction pathways are regulated and coordinated by non-canonical Wnt signaling in space and time within the vascular network . Taken together , our results show that Wnt5a signaling fine-tunes junctional mechanocoupling between endothelial cells to promote collective cell behavior during vascular morphogenesis .
For the Cdc42 inhibition experiment , C57BL/6J mice were maintained at the Instituto de Medicina Molecular ( iMM ) under standard husbandry conditions and under national regulations . ML-141 ( SML0407 , Sigma , Germany ) was injected twice ( morning and evening ) intraperitoneally ( IP ) ( 20 ml/g of 1 mg/mL solution ) at postnatal day 5 ( P5 ) before eyes were collected at P6 . Vinculin floxed mouse ( Zemljic-Harpf et al . , 2007 ) was obtained from Robert S . Ross Pdgfb-iCreERT2 ( Claxton et al . , 2008 ) to generate a new Vinculin fl/fl::Pdgfb-iCreERT2 mouse line . Mice were maintained at the Instituto de Medicina Molecular ( iMM ) under standard husbandry conditions and under national regulations . Animal procedures were performed under the DGAV project license 0421/000/000/2016 . Tamoxifen ( Sigma , Germany ) was injected intraperitoneally ( IP ) ( 20 ml/g of 1 mg/mL solution ) at postnatal day 1 ( P1 ) and P3 before eyes were collected at P6 . For Figure 1 , we re-used mouse retinas previously collected ( Franco et al . , 2016 ) . For clarity , we transcribe the specificities of the breedings and experimental conditions . The following mouse strains were previously used: Pdgfb-iCreERT2 ( Claxton et al . , 2008 ) ; Wnt5a floxed ( Miyoshi et al . , 2012 ) ; Wnt11 null ( Majumdar et al . , 2003 ) . Mice were maintained at the London Research Institute under standard husbandry conditions . Tamoxifen ( Sigma , Germany ) was injected intraperitoneally ( IP ) ( 20 ml/g of 1 mg/mL solution ) at postnatal day 2 ( P2 ) before eyes were collected at P5 onwards . Animal procedures were performed in accordance with the Home Office Animal Act 1986 under the authority of project license PPL 80/2391 . Eyes were collected at P6 and fixed with 2% PFA in PBS for 5 hr at 4°C , thereafter retinas were dissected in PBS . Blocking/permeabilisation was performed using Claudio’s Blocking Buffer ( CBB ) ( Franco et al . , 2013 ) , consisting of 1% FBS ( Thermo Fisher Scientific ) , 3% BSA ( Nzytech ) , 0 . 5% triton X100 ( Sigma ) , 0 . 01% Na deoxycholate ( Sigma ) , 0 , 02% Na Azide ( Sigma ) in PBS pH = 7 . 4 for 2 hr in a rocking platform . Primary antibodies ( Anti-CD102 and Anti-Erg ) were incubated at the desired concentration ( see Key Resources Table ) in 1:1 CBB:PBS at 4°C overnight in a rocking platform and afterwards washed 3 × 60 min in PBS-T . Then , retinas were incubated in 1:1 CBB:PBS solution containing the secondary fluorophore conjugated antibodies at 4°C overnight in the dark . Next , and due to the fact that we are using same species primary antibodies , retinas were incubated with AffiniPureF ( ab’ ) 2 fragments Donkey anti-rabbit IgG ( see Key Resources Table ) for 2 hr at RT , followed by 3 washes of 30 min in PBS-T . Retinas were fixed with 4%PFA in PBS at RT and blocked using CBB and primary antibody ( Anti-GOLPH4 ) was incubated ( see Key Resources Table ) in 1:1 CBB:PBS at 4°C overnight in a rocking platform . Secondary antibody was done as previously described . Retinas were mounted on slides using Vectashield mounting medium ( Vector Labs , H-1000 , Burlingame , California , USA ) . For polarity quantification , a tile-scan spanning the sprouting front was acquired on a Zeiss Cell Observer Spinning Disk microscope , equipped with the Zen software with a Plan-Apochromat 40x/1 . 4 Oil DIC M27 objective . Human umbilical vein endothelial cells ( HUVECs ) were routinely cultured following the manufacturer’s guidelines , in filter-cap T75 flasks Nunclon ∆ surface treatment ( VWR international , LLC ) and cultured at 37°C and 5% CO2 to ensure a stable environment for optimal cell growth . HUVECs ( C2519A , Lonza ) were cultured with complete medium EGM-2 Bulletkit ( CC-3162 , Lonza ) supplemented with 1% penicillin/streptomycin ( #15140122 , Gibco ) . When passaging cells for experiments , cells were washed twice in sterile PBS ( 137 mM NaCl , 2 . 7 mM KCl , 4 . 3 mM Na2HPO4 , 1 . 47 mM KH2PO4 , pH7 . 4 ) . Then , cells were incubated for 3–5 min in trypsin/EDTA ( #15400054 , Gibco ) or in TrypLE Express ( #12605–028 , Gibco ) at 37°C , 5% CO2 . When 95% of the cells detached , complete medium was added to each flask to inhibit the activity of the trypsin/EDTA or TrypLE Express and the cell suspension was transferred to a falcon tube . To maximize the amount of cells collected , all flasks were washed again with complete medium , which was added to the cell suspension gathered previously . HUVECs were then centrifuged at 115 g for 5 min at room temperature . The pellet was re-suspended in fresh complete medium . The cell concentration present in the suspension was determined using a Neubauer Chamber Cell Counting ( Hirschmann EM Techcolor ) . All cells were then seeded on the desired culture vessels at 200 . 000–300 . 000 cells/mL and placed in the incubator . All experiments with HUVECs were performed between passages 3 and 6 . In order to silence the expression of genes of interest , a set of ON-TARGET human siRNAs were purchased from Dharmacon ( see Table 1 ) . Briefly , HUVECs were seeded the day before the transfection to reach 60–70% confluence and were then transfected with 25 nM of siRNA using the DharmaFECT one reagent ( Dharmacon , GE Healthcare ) following the Dharmacon siRNA Transfection Protocol . 24 hr after transfection the culture medium was replaced by fresh complete medium and cells were kept under culture conditions up until 72 hr post-transfection and then processed for further experiments . siRNA efficiencies were measured by qPCR and by WB when antibodies were available ( Figure 11 ) . RNA extraction was performed from HUVECs seeded on 12-well plates using the RNeasy Mini Kit ( Qiagen ) and the GeneJet RNA Purification Kit ( Thermo Scientific ) as described by the manufacturer’s protocol . RNA concentration was quantified using NanoDrop 1000 ( Thermo Scientific ) and adjusted equally , followed by DNase I digestion ( Thermo Scientific ) and cDNA synthesis ( Superscript IV First-Strand Synthesis System , Invitrogen ) . cDNA samples were then diluted in RNAse/DNAse-free water for the subsequent quantitative real-time PCR ( RT-qPCR ) reactions . RT-qPCR was performed using a 7500 Fast Real-Time PCR System ( Applied Biosystems ) with Power SYBR Green PCR Master Mix ( Applied Biosystems ) following the standard program of the system previously mentioned . For each reaction , 5 µL of cDNA was combined with 10 µL of Power SYBR Green PCR Master Mix , 4 . 5 µL of RNAse/DNAse free water and 0 . 5 µL of 4 µM primers pool ( Forward +Reverse ) ( see Table 2 ) in a MicroAmp Fast Optical 96-well Reaction Plate ( Applied Biosystems ) . The expression levels of each sample duplicate were then normalized to GAPDH and the 2-ΔΔT method was used to calculate relative alterations in gene expression ( Figure 11 ) . Protein extraction was performed from HUVECs seeded on 6-well plates which were lysed in 120 µL of RIPA buffer ( 50 mM Tris/HCl pH7 . 5 , 1% NP-40 , 150 mM NaCl , 0 . 5% sodium deoxycholate , 0 . 1% SDS in H2O ) supplemented with phosphatase and proteinase inhibitors cocktail ( 1:100 , #10085973 Fischer Scientific ) . Adherent cells were then detached from the plate with a cell scrapper and the cell lysates were gathered and transferred into an ice cold eppendorf tube . The cell lysates were then centrifuged at maximum speed for 10 min at 4°C and the supernatants collected into a new eppendorf tube . Protein concentration was quantified using the BCA protein assay kit ( Pierce ) following the guidelines recommended by the manufacturer . The Multimode microplate reader , Infinite M200 ( Tecan ) , was used for spectrophotometric measurement of protein with the i-control software . For Western Blotting protein samples were normalized up to 25 µL and combined with a mixture of 2x Laemmli Sample Buffer ( #161–0747 , Bio-rad Laboratories ) with 450 mM DTT ( D0632 , Sigma-Aldrich ) and incubated at 70°C in a Dry Block Thermostat ( Grant Instruments , Ltd ) for 10 min ( or 95°C for 5 min ) . Protein samples were loaded and separated on a 4–15% Mini-PROTEAN TGX Gel ( #456–1084 , BioRad ) along with 5 µL of protein ladder ( Full-Range RPN800E , GE Healthcare Rainbow Molecular Weight Markers ) , first at 50V for 5 min and then at 100–130V for 1–2 hr in SDS-PAGE running buffer ( 10x SDS-PAGE: 250 mM Tris , 1 . 92M Glycine , 1% SDS , pH8 . 3 ) . Gels were then transferred either onto a nitrocellulose membrane ( iBlot Transfer Stack Regular/Mini size , #IB3010-01/−02 , Invitrogen ) with iBlot Dry Blotting System ( Invitrogen ) for 4–7 min; or onto a Polyvinylidene Difluoride ( PVDF ) membrane ( #IPVH00010 , Merck Milipore ) with Mini Trans-Blot Electrophoretic Transfer Cell ( Biorad ) following the manufacturer’s guidelines . After transfer , blotted membranes were incubated in Ponceau Red to assess transfer quality , and then washed in TBS-T ( 50 mM Tris/HCl , 150 mM NaCl , 0 . 1% Tween-20 , pH7 . 5 ) . Then , membranes were incubated in blocking buffer containing 3% BSA ( Bovine Serum Albumin , MB04602 , Nzytech ) in TBS-T for 1 hr at RT , followed by an overnight incubation at 4°C with the primary antibodies diluted in the same blocking buffer ( see Key Resources Table ) . On the following day membranes were washed three times in TBS-T and incubated in blocking buffer containing the secondary horseradish peroxidase ( HRP ) conjugated antibodies for 1 hr at RT ( see Key Resources Table ) . Before revelation membranes were washed again three times in TBS-T for 5 min and then incubated in ECL Western Blotting Detection Reagent ( RPN2209 , GE Healthcare ) following the manufacturer’s protocol . Protein bands were visualized in Chemidoc XRS + and relative protein quantities were measured using the Image Lab software , both from Bio-Rad Laboratories . All results were normalized to tubulin levels . Active Cdc42 pulldown was performed from HUVECs cultured in 10 cm plates non-stimulated or stimulated with recombinant human Wnt5a protein ( 645-WN , R and D Systems , 200 ng/mL ) for 15 min using the Cdc42 Pull-down Activation Assay Biochem Kit ( Cytoskeleton ) as described by the manufacturer’s protocol . Briefly , after stimulation , cells were washed with ice cold PBS , scrapped and lysed in lysis buffer containing protease and phosphatase inhibitors . After lysate clarification , inputs from all the samples were gathered and the remaining lysate was used for the pulldown reaction . 10 µg of PAK-PBD beads were added to equivalent protein amounts of cell lysates ( 300 µg ) for each condition . The mixture was then incubated for 1 hr at 4°C with gentle rotation . After the pulldown reaction , beads were washed three times in washing buffer and the bound protein complexes were eluted in sample buffer with DTT by placing the beads for 5 min at 95°C . Samples were then blotted on SDS-PAGE following standard protocols . VE-cadherin and vinculin immunoprecipitation was performed from HUVECs cultured in 10 cm plates . After the scratch-wound assay , cells were incubated with PBS supplemented with 1 mM CaCl2 and 0 . 5 mM DSP ( #22585 , Thermo Scientific ) for 20 min at RT . Afterwards they were washed twice with ice cold PBS and then four times with ice cold quenching buffer ( 10 mM Tris/HCl , pH 7 . 5 , in PBS ) . Then , cells were scrapped and lysed in lysis buffer ( 25 mM Tris/HCl , pH 7 . 5 , 1% NP-40 , 1% deoxycholic acid , 150 mM NaCl ) containing protease and phosphatase inhibitors . Cell lysates were centrifuged at 16 , 100 g for 10 min at 4°C and the pellet digested in SDS IP buffer ( 15 mM Tris/HCl , pH 7 . 5 , 5 mM EDTA , 2 . 5 mM EGTA , 1% SDS ) . Samples were then incubated for 10 min at 100°C and diluted in lysis buffer . At this point , inputs from all the samples were gathered and the remaining lysate was used for immunoprecipitation . Pre-washed Pierce G-protein agarose beads ( #22851 , Thermo Scientific ) were added to equivalent protein amounts of cell lysates ( 100–200 μg ) for each condition , containing either 2 μg of anti-vinculin ( V9264 , Sigma-Aldrich ) or anti-VE-cadherin ( sc-9989 , Santa Cruz Biotechnologies ) antibody . The mixture was then incubated overnight at 4°C with gentle rotation . After immunoprecipitation , beads were washed four times in ice cold lysis buffer and the bound protein complexes were eluted in sample buffer with DTT by placing the beads for 10 min at 100°C . Samples were then blotted on SDS-PAGE following standard protocols . For immunofluorescence of in vitro cultured HUVECs , cells were seeded on 24-well plates with glass coverslips , or in 8-well Ibidi slides ( 80826 , Ibidi ) previously coated with 0 . 2% Gelatin in sterile water ( G1393 , Sigma-Aldrich ) or with Fibronectin in PBS ( F1141 , Sigma-Aldrich ) , respectively . After the scratch-wound assay ( described above ) , HUVECs were fixed in 1% Paraformaldehyde ( PFA ) supplemented with 1M MgCl2 and 1M CaCl2 ( 1 µL/2 mL ) in PBS for 30 min at RT . Cells were then washed with PBS to remove the remaining PFA and the immunostaining protocol initiated . When the PBS was removed , HUVECs were blocked and permeabilized with blocking solution containing 3% BSA in PBS-T ( PBS with 0 . 1% Triton X-100 ) for 30 min at RT . Then cells were incubated for 2 hr at RT with the primary antibodies diluted in the blocking solution ( see Key Resources Table ) and washed 3 × 15 min washes in PBS-T . Afterwards , cells were incubated in blocking solution containing the secondary fluorophore conjugated antibodies for 1 hr at RT in the dark , followed again by 3 washes of 15 min in PBS-T . Finally , HUVECs were incubated with 1x DAPI ( D1306 , Molecular Probes by Life Technologies ) diluted in PBS-T for 5 min in the dark . Coverslips were then mounted on microscopy glass slides using Mowiol DABCO ( Sigma-Aldrich ) , while for the 8-well Ibidi slides 50 µL of Mowiol DABCO was added to each well . To quantify co-localization of junctional molecules , high-resolution Z-stack images at multiple positions on the scratch front were acquired on a confocal Laser Point-Scanning Microscope 880 ( Zeiss ) equipped with the Zen black software with a Plan Apochromat 63x NA 1 . 40 oil DIC M27 objective . For polarity quantification , a tile-scan spanning the entire region of the scratch was acquired on a motorized inverted widefield fluorescence microscope , Zeiss Axiovert 200M ( Carl Zeiss ) equipped with the Metamorph software with an EC Plan-NeoFluar 40x NA 0 . 75 dry objective . For co-localization analysis , high-resolution Z-stack confocal images of HUVECs stained for junctional proteins ( VE-Cadherin , Vinculin , α-catenin , β-catenin and p120-catenin ) were imported and analyzed in MATLAB using a custom written code ( Source code 1 ) . An object-based co-localization approach was used . Briefly , each channel was segmented and a binary mask was generated . The masks were combined and the fraction of pixels with overlapping signals was quantified . Confluent HUVECs seeded on 24-well plates were subjected to the scratch-wound assay and then incubated for 15 min in Ca2 +free HBSS , followed by DMEM ( #41966–029 , Gibco ) supplemented with 1% penicillin/streptomycin ( #15140122 , Gibco ) , 10% fetal bovine serum ( FBS ) ( #10500–064 , Gibco ) and 2 mM Ca2+ from 1 up to 30 min at 37°C , 5% CO2 . Afterwards , cells were immediately fixed in 1% Paraformaldehyde ( PFA ) supplemented with 1M MgCl2 and 1M CaCl2 ( 1 µL/2 mL ) and processed for immunostaining . Confluent HUVECs seeded on 24-well plates were subjected to the scratch-wound assay and then processed for PLA using the Duolink In Situ Red Mouse/Rabbit Starter Kit ( DUO92101-1KT , Sigma-Aldrich ) as described by the manufacturer’s protocol . To probe interactions between vinculin and VE-cadherin , cells were incubated with an anti-vinculin antibody raised in rabbit ( V4139 , Sigma-Aldrich ) and an anti-VE-cadherin antibody raised in mouse ( sc-9989 , Santa Cruz Biotechnologies ) . In parallel , cells were also incubated with an anti-VE-cadherin antibody raised in goat ( AF938 , R and D Systems ) and subsequently with an anti-Goat fluorescent-conjugated secondary antibody ( A21447 , Thermo Fisher Scientific ) to label adherens junctions . To quantify co-localization of PLA signal at adherens junctions , high-resolution Z-stack images at multiple positions on the wound edge were acquired on a confocal Laser Point-Scanning Microscope 880 ( Zeiss ) equipped with the Zen black software with a Plan Apochromat 63x NA 1 . 40 oil DIC M27 objective . Briefly , PLA dots were quantified only at adherens junctions , using a similar approach to the co-localization studies described in the section ‘Immunostaining co-localization analysis’ , using the VE-cadherin immunofluorescence staining to detect overlapping pixels between junctions and PLA signals . Replication-incompetent lentiviruses were produced by transient transfection of HEK293T of pLX303 lentiviral expression vector co-transfected with the viral packaging vector Δ8 . 9 and the viral envelope vector VSVG . Medium was replaced with fresh culture medium 5–6 hr post transfection . 48 hr after medium replacement , lentiviral particles were concentrated from supernatant by ultracentrifugation at 90000 g for 1h30 and re-suspended in 0 . 1% BSA PBS . Seeded HUVECs were transduced 24 hr post-transfection with varying concentrations of lentiviral plasmids containing VE-Cad-TS , VE-Cad-TL , Cdc42-2G , Vinculin-Full-Length-GFP , Vinculin-T12-mutant-GFP and αCat-Vinc-HA fusion protein sequences ( see Key Resources Table ) . 24 hr after viral transduction the culture medium was replaced by fresh complete medium and cells were kept in culture conditions up until 48 hr post-transduction and then processed for immunofluorescence or imaging . In the analysis , we used a mix population containing transduced and non-transduced cells , selecting areas where high transduction efficiencies were observed . To assess functional collective cell behavior properties ( i . e . , polarity and migration ) , as well as morphological features of in vitro cultured HUVECs , we used the scratch-wound assay . The wound was created by scratching the surface of a well-plate or a microscopy glass slide containing a monolayer of adherent HUVECs with a 200 µL pipette tip . The culture medium was then replaced by fresh complete medium and HUVECs were allowed to migrate under optimal physiological conditions . When appropriate , drugs of interest were added to the medium . ( see Key Resources Table ) . Cells migrated for 5 hr after the wound , were fixed and then stained for immunofluorescence experiments . For live imaging experiments HUVECs migration was followed up to 16 hr . Imaging was performed using a Zeiss Cell Observer SD ( Carl Zeiss ) equipped with an EC Plan-Neofluar 10x NA 0 . 3 PH1 . To track individual cells within the monolayer more efficiently using the cell nuclei as reference , HUVECs were incubated in 1x Hoechst for 15 min at 37°C before the onset of the time lapse . Images of the scratch front were acquired at multiple positions every 10 min . Analysis of migration , including wound closure , cell speed and straightness was performed using FIJI TrackMate plug in and the Chemotaxis and Migration Tool ( free software from Ibidi ) . The velocity field of the moving cell sheet was calculated in Matlab using cell image velocimetry ( CIV ) ( Milde et al . , 2012 ) software . Interrogation windows were set to 64 × 64 pxls with a 50% overlap . Velocity spatial correlation was calculated in Matlab using the x-component of the velocity as in Petitjean et al . ( 2010 ) . Correlation length was determined from exponential fitting of correlation curves . HUVECs were re-plated onto 35 mm Petri dishes ( TPP ) 48 hr post-transfection from 6-well plates ( on a ratio of 1 6-well plate to 2 35 mm Petri dishes per condition ) to attain a confluence of 60–70% . On the following day , 1 hr before starting the cell-cell adhesion measurements , the culture medium was replaced by PBS in one of the 35 mm Petri dish replicates , to ensure cell detachment . 5 min before the experiment , the culture medium of the other 35 mm Petri dish replicate was replaced with serum-free culture medium . An atomic force microscope NanoWizard II ( JPK Instruments , Berlin , Germany ) mounted on the top of an Axiovert 200 inverted microscope ( Carl Zeiss , Jena , Germany ) was used for the cell-cell adhesion measurements . For these experiments , tipless arrow TL1 cantilevers ( Nanoworld , Neuchatel , Switzerland ) were used , with a nominal spring constant of 0 . 03 N m−1 , as described previously ( Ribeiro et al . , 2016 ) . Cantilevers were cleaned for 15 min with UV light and coated with poly-D-lysine ( 50 μg ml−1 ) for at least 30 min . Cantilevers were stored in poly-D-lysine solution until use . After that , a set of adherent cells from the other Petri dish were selected to perform the cell-cell adhesion measurements , composed of 5 force-distance curves performed on each cell , with a cell-cell contact time of 5 s and a 5 s pause between them . Cell–cell contact was established with an applied force of 300 pN , at a constant height and in closed-loop mode . The AFM tip resonant frequency was maintained at 2 Hz , with a z-range displacement of 50 μm . For the internal negative controls , we used 4 mM EGTA , a Ca2+ chelating agent that is able to sequestrate calcium ions from cadherins and render them inactive and unresponsive to force transmission . EGTA was added to the serum-free culture medium of the Petri dish containing the adherent cells at the time of the recordings . FRET images were obtained using a confocal Laser Point-Scanning Microscope 880 ( Zeiss ) equipped with a Plan-Apochromat 63x , NA 1 . 40 , oil immersion , DIC M27 objective and an argon laser featuring 405 , 458 and 514nm laser lines . For FRET experiments , it was used a MBS 458/514 beam splitter in combination with the following filters: mTFP1 GaAsP , band-pass 461–520; Venus/FRET , band-pass 525–575 . Acceptor photobleaching experiments were analyzed using a custom written Matlab script ( Source code 2 ) . A Gaussian filter with standard deviation of 0 . 75 was applied to the images before analysis . The intensity in the region of interest was measured before and after bleaching . FRET efficiency was calculated as EF=Ipost-IpreIpost where Ipost and Ipre are the intensity of the donor channel after and before bleaching respectively . The CDC42-2G FRET biosensor activity was obtained using a widefield fluorescence microscope Axio Observer ( Zeiss ) equipped with a Plan-Apochromat 63x , NA 1 . 40 , oil immersion , DIC M27 objective . For ratiometric FRET experiments we used the following excitations and emission filters: ET436/20 and ET480/40 for ECFP; ET500/20 and ET535/30 for EYFP ( Chroma Technology Corp ) and the images for each condition were acquired during 5 min with 1 s time interval . FRET experiments were performed as described by Louis Hodgson . Analysis of ratiometric FRET biosensor was performed in Matlab and the preprocessing was performed using the Biosensor Processing 2 . 1 software package from the Danuser lab ( Hodgson et al . , 2010 ) . The resulting images showing the localized activation of CDC42 were further processed to retrieve quantitative information from such maps . Briefly , junctional or free-edges regions were selected from each time-lapse image and the differential of the intensity vs time traces was calculated . For each image a region where no activation was detected was also selected to determine the level of background signal . The local maxima for each curve above background level were determined . Maxima found within three frames from each other were assumed to correspond to the same activation event . To quantify cell polarity , tile-scan images of HUVECs stained with Golgi ( Golph4 ) and nuclear ( DAPI ) markers were processed on Adobe Photoshop to separate leader cells , identified as the first row of cell directly in contact with the scratch , from follower cells , comprising the second to fourth rows of cells away from the scratch . Afterwards , each set of images was imported and analyzed in MATLAB using a modified version of a polarity analysis script kindly provided by Anne-Clémence Vion and Holger Gerhardt . Briefly , after segmenting each channel corresponding to the Golgi and nuclear staining , the centroid of each organelle was determined and a vector connecting the center of the nucleus to the center of its corresponding Golgi apparatus was drawn . The Golgi-nucleus assignment was done automatically minimizing the distance between all the possible couples . The polarity of each cell was defined as the angle between the vector and the scratch line . An angular histogram showing the angle distribution was then generated . Circular statistic was performed using the Circular Statistic Toolbox . To test for circular uniformity , we applied the polarity index ( PI ) , calculated as the length of mean resultant vector for a given angular distribution ( Figure 1D ) . PI=1N∑1Ncosα2+1N∑1Nsinα2 PI corresponds to length of the mean resultant vector , previously described in Berens ( 2009 ) . The PI varies between 0 and 1 . The closer to one the more the data are concentrated around the mean direction , while values close to 0 corresponds to random distribution . PI indicates the collective orientation strength of the cell monolayer . Box plots were generated by using every single PI calculated for images of each biological replica , which show the biological variability of the system . This data is used to calculate the significance of differences between experimental conditions . To obtain a global description of a given experimental condition , we pooled together all the different biological replicates in one single file and calculate a global PI and mean angle ( angular histograms and values below in the main text ) . This representation provides information on the general distribution of polarities in all experiments , and provides a mean angle of polarity , which is important to understand directionality . To calculate the PI as a function of distance , each image was divided starting from the wound edge in slices 50 μm apart . The cell polarity within each slice was extracted and represented as angular histogram and the corresponding PI was calculated . For Figure 1E , a global polarity index was calculated merging together the results from different images from the same experimental conditions . N = 9 for siControl , N = 10 for siWNT5a and N = 4 for siCtnna1 . All statistical analysis was performed using GraphPad Prism 7 and Matlab ( Mathworks ) . Statistical details of experiments are reported in the figures and figure legends . Sample size is reported in the figure legends and no statistical test was used to determine sample size . The biological replicate is defined as the number of cells , images , animals , as stated in the figure legends . No inclusion/exclusion or randomization criteria were used and all analyzed samples are included . Comparisons between two experimental groups were analyzed with unpaired parametric t test , while multiple comparisons between more than two experimental groups were assessed with one-way ANOVA . We considered a result significant when p<0 . 05 . Box plots for polarity indexes represent min to max , central line represents mean . | When a new blood vessel is created , a leader cell branches out from the lining of an existing vessel before being joined by other cells moving together in the same direction . A protein called Wnt5a regulates this process by helping the cells to orient themselves and finely coordinating their migration , but the exact details of this mechanism are still unclear . One way that cells can communicate is by touching and physically exerting forces on each other . This is made possible by structures called cellular junctions , which are present at the interface between cells . These can transmit forces within a tissue because they are connected with elements that form the cells’ internal skeletons . A protein known as vinculin is involved in these connections . To find out what role Wnt5a plays in cell migration , Carvalho et al . prevented blood vessel cells from creating the protein . The results showed that Wnt5a helps cells to move together by stabilizing vinculin at cell junctions . This strengthens the physical communication between cells and allows them to efficiently coordinate their movements . Indeed , in the mouse retina , deleting vinculin from cells that make blood cells impaired the formation of new blood vessels . Problems in the way that blood vessels grow are very common in the human population . In addition , Wnt5a is linked to cancer progression , which also relies on coordinated movement of cells . A better grasp of the role of this protein could therefore be relevant to understand how blood vessels are formed , but also how certain cancers invade surrounding tissues . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"developmental",
"biology",
"cell",
"biology"
] | 2019 | Non-canonical Wnt signaling regulates junctional mechanocoupling during angiogenic collective cell migration |
Co-opting the cellular machinery for protein production is a compulsory requirement for viruses . The Cricket Paralysis Virus employs an Internal Ribosomal Entry Site ( CrPV-IRES ) to express its structural genes in the late stage of infection . Ribosome hijacking is achieved by a sophisticated use of molecular mimicry to tRNA and mRNA , employed to manipulate intrinsically dynamic components of the ribosome . Binding and translocation through the ribosome is required for this IRES to initiate translation . We report two structures , solved by single particle electron cryo-microscopy ( cryoEM ) , of a double translocated CrPV-IRES with aminoacyl-tRNA in the peptidyl site ( P site ) of the ribosome . CrPV-IRES adopts a previously unseen conformation , mimicking the acceptor stem of a canonical E site tRNA . The structures suggest a mechanism for the positioning of the first aminoacyl-tRNA shared with the distantly related Hepatitis C Virus IRES .
Translation initiation is the most complex and highly regulated step of protein synthesis ( Schmeing and Ramakrishnan , 2009 ) . Canonical initiation results in the formation of an elongation-competent ribosome with an aminoacyl-tRNA base paired with messenger RNA ( mRNA ) at the peptidyl site ( P site ) of the ribosome . Translation initiation in eukaryotes is achieved by a highly sophisticated mechanism ( Jackson et al . , 2010 ) . Most eukaryotic mRNAs contain a unique nucleotide structure at their 5’ end , known as the cap structure . The multi-subunit initiation factor eIF4F binds the cap structure and recruits the 43S complex consisting of the small ribosomal subunit ( 40S ) , eIF2/GTP/Met-tRNAMet ternary complex , eIF3 , eIF1 , eIF1A , and eIF5 ( Jackson et al . , 2010 ) . The attached 43S complex scans the 5’-untranslated region of the mRNA downstream to the initiation codon , where it forms the 48S initiation complex with the established codon-anticodon base-pairing in the ribosomal P site . Finally , eIF5B , the eukaryotic ortholog of the bacterial initiation factor 2 ( IF2 ) , promotes the recruitment of the large ribosomal subunit ( 60S ) and the formation of the elongation-competent 80S ribosome ( Fernández et al . , 2013 ) . Eukaryotic viruses have evolved refined molecular strategies to interfere with canonical initiation factors , leading to the hijacking of host ribosomes to produce viral proteins ( Jackson et al . , 2010; Hertz and Thompson , 2011 ) . A common strategy used by different types of viruses relies on structured RNA sequences at the ends of their mRNAs ( Filbin and Kieft , 2009 ) . These sequences are called Internal Ribosomal Entry Sites ( IRES ) and form specific three-dimensional structures able to manipulate and co-opt the host translational machinery ( Yamamoto et al . , 2017 ) . IRES sequences are classified according to the subset of factors they require for initiation ( Filbin and Kieft , 2009 ) . Type IV IRES sequences , including the Cricket Paralysis Virus IRES ( CrPV-IRES ) and the Taura Syndrome Virus IRES ( TSV-IRES ) do not require initiation factors and are the best studied IRESs . Biochemical and structural studies have provided a detailed view on how these approximately 200-nucleotide-long sequences interact with and manipulate the ribosome ( Wilson et al . , 2000; Spahn et al . , 2004; Petrov et al . , 2016 ) . A modular architecture of three pseudoknots ( PKI , PKII and PKIII , Figure 1A ) is crucial for these IRESs to establish a balance between structural flexibility and rigidity , essential for interaction with the ribosome and with two elongation factors ( eEF2 and eEF1A ) required for IRES translocation through the ribosome ( Jan et al . , 2003 ) . PKI mimics an anti-codon stem loop ( ASL ) of a tRNA interacting with its cognate mRNA codon , and plays an essential role in setting up the correct reading frame in the aminoacyl site ( A site ) of the ribosome ( Costantino et al . , 2008 ) . A ribosome primed with a type IV IRES alternates between rotated and non-rotated configurations of the small ribosomal subunit ( Fernández et al . , 2014; Koh et al . , 2014 ) . Similarly , after peptidyl-transfer in canonical translocation , the ribosome alternates between rotated and non-rotated configurations of the small ribosomal subunit with respect to the large ribosomal subunit ( Voorhees and Ramakrishnan , 2013 ) . This pre-translocation stage of the ribosome is recognized by a protein translocation factor ( EF-G in Bacteria , eEF2 in Eukarya ) , which in GTP-bound form induces an additional rotation of the small subunit and blocks the A site of the ribosome ( Tourigny et al . , 2013 ) . Translocation proceeds forward by a back rotation of the small subunit to recover a canonical configuration of the ribosome . This is accomplished by a swiveling movement of the head of the small subunit in an orthogonal direction respect to that of the rotation of the small subunit ( Ratje et al . , 2010 ) . The back-rotation of the small subunit accompanied by the swiveling of the head is performed while EF-G/eEF2 are still bound ( Ramrath et al . , 2013 ) . Once the translocation factor EF-G/eEF2 leaves the ribosome , the head of the small subunit returns to its non-swiveled configuration , rendering a ribosome primed with a peptidyl-tRNA in the P site , a deacylated tRNA in the E site and a vacant A site ready to accept the next aminoacyl-tRNA ( Noller et al . , 2017 ) . The L1 stalk , a component of the large ribosomal subunit , also contribute to the vectorial movement of tRNAs , offering additional anchoring points to the leaving deacylated-tRNA in the E site ( Fei et al . , 2008 ) . The movements of the L1 stalk are coordinated with those of the small subunit . Type IV IRESs require two translocation events to place the first aminoacyl-tRNA in the P site making use of the intrinsic dynamic elements of the ribosome involved in canonical translocation ( Murray et al . , 2016; Abeyrathne et al . , 2016 ) . However , less is known about the second translocation event , required for the first aminoacyl-tRNA to enter the P site , ending the unusual initiation pathway followed by this type of IRESs . We report the visualization by means of single-particle electron cryo-microscopy ( cryoEM ) of two related states of the mammalian ribosome with a double translocated CrPV-IRES and P site aminoacyl-tRNA at 3 . 2 and 4 . 75 Ångstroms resolution . The head swiveling of the small ribosomal subunit plays a fundamental role in the late step of this translocation event , inducing a remarkable conformational change on the PKI of the CrPV-IRES , which becomes disassembled , to mimic the acceptor stem of a E site tRNA .
To analyze the integrity and stability of complexes for cryoEM , we assembled ribosomal complexes with a double translocated CrPV-IRES in a mammalian reconstituted system from individual components in the presence of GTP ( Figure 1B and C ) . The translocation efficiency was monitored by toe-printing . A pre-translocation complex , assembled by mixing of CrPV-IRES with 40S and 60S subunits , results in a +14–15 nt toe-print signal from the CCU ( sequence belonging to the PKI ) and thus indicate the presence of PKI in the ribosomal A site ( Figure 1C , lane 3 ) . The addition of elongation factors leads to a 5–6 nucleotides toe-print shift showing the double translocation event ( Figure 1C , lane 4 ) . However , the similar intensity of several bands , with difference in one nucleotide , suggests instability of the double translocated IRES or frame ambiguity . We reasoned this ambiguity could be explained by the absence of an A site ligand , what would allow partial back translocation of the IRES even in the presence of a translocated aminoacyl-tRNA in the P site . Similarly , in the single translocated CrPV-IRES cryoEM reconstruction , it was necessary the addition of an A site ligand ( Muhs et al . , 2015 ) . In this report , the mutation of the first sense codon to a stop codon and the addition of the release factor 1 ( eRF1 ) were applied to prevent spontaneous back-translocation of the IRES . Thus , to stabilize a double translocated complex , we mutated the second sense codon to a UAG stop codon ( Figure 1A ) . The supplementation of the reaction with a mutated and catalytically inactive version of the release factor 1 ( eRF1* , AGQ mutation ) causes a +20 nt toe-print signal from the first sense codon ( leucine CUU codon ) in the P site , which is in a good agreement with previous report ( Muhs et al . , 2015 ) , indicating proper binding of eRF1* ( Figure 1C , lane 5 ) . The simultaneous decrease of intensity for the 5–6 nucleotides toe-print suggests a more homogeneous complex , suitable for structural studies . Given the conformation of the P site tRNA is completely compatible with that of a translating ribosome , we believe the addition of eRF1* in the present sample does not significantly affect the conformation of the CrPV-IRES . Maximum likelihood particle sorting methods implemented in RELION ( Scheres , 2012 ) were applied to a large cryoEM dataset at two different stages ( Figure 1D ) . An initial classification in two dimensions allowed for the separation of full ribosome ( 80S ) from small subunit ( 40S ) particles . The two sorted subgroups were further classified using masking methods with signal subtraction with focus in the inter-subunit space , where this type of IRES binds as well as canonical translation factors . The L1 stalk was also included in the masked area to allow for a wider sampling . This strategy revealed , in a single classification step , several sub-populations , reflecting the heterogeneity of the sample . A binary 80S/CrPV-IRES complex in a pre-translocation conformation could be identified as well as two sub-populations with CrPV-IRES in a double translocated state . In the double translocated reconstructions , the ribosome adopts a non-rotated configuration of the small subunit , with clear density for aminoacyl-tRNA in the P site and a either eEF2 or eRF1* in the A site ( Figure 1D , Figure 2—figure supplement 1 , Figure 3—figure supplement 1 and Table 1 ) . Recent studies by single-molecule FRET ( smFRET ) have characterized the kinetics of the translocation events required for the CrPV-IRES to deliver an aminoacyl-tRNA to the P site of the ribosome ( Petrov et al . , 2016; Zhang et al . , 2016 ) . Slow movements of the CrPV-IRES compared with canonical translocation of tRNAs are evident from this data and explain the capturing of a late-stage intermediate of translocation with eEF2 in our dataset ( Figure 2A-C ) . Our cryoEM reconstruction reveals that the conformation of the IRES in this intermediate state is similar to the conformation reported for the single translocated state ( Muhs et al . , 2015 ) , with the SL-IV and SL-V detached from the 40S and exposed to the solvent . PKI is in an intermediate position between the P and E sites of the small subunit as well as the aminoacyl-tRNA is in an intermediate position between the A and P site of the 40S ( Figure 2B ) . Domain IV of eEF2 occupies the A site of the small ribosomal subunit . This configuration is maintained by a distinctive swiveled configuration of the 40S head , resembling one of the late stages recently reported for the first translocation event ( Figure 2C ) ( Abeyrathne et al . , 2016 ) . The most populated class of particles represent a double translocated CrPV-IRES with aminoacyl-tRNA in a canonical configuration in the P site and eRF1* in the A site ( Figure 3 ) . Both aminoacyl-tRNA and eRF1* populate conformations recently described , with the characteristic bent of the mRNA at the stop codon ( Figure 3—figure supplement 1D; Brown et al . , 2015 ) . The small subunit in this reconstruction is in a non-rotated configuration and the 40S head is not tilted or swiveled ( Figure 3 ) . CrPV-IRES has undergone a conformational change that mainly affects PKI , but also the relative orientation of PKII and PKIII . In the pre-translocated as well as in the single translocated conformation of the IRES , PKII and PKIII interact by a network of non-covalent interactions involving sugar–sugar stacking interactions as well as A-minor interactions ( Figure 4—figure supplement 1 , ( Murray et al . , 2016; Pfingsten et al . , 2006 ) . This compact configuration adopted due to the physical proximity of PKII and PKIII seems to be a requirement for the initial binding to the ribosome as well as for the first translocation event ( Murray et al . , 2016; Muhs et al . , 2015 ) . In the state described here , these interactions are no longer established and a pronounced gap could be observed between both pseudoknots ( Figure 4B , bottom right and Figure 4—figure supplement 1 ) . Upon back-swiveling of the 40S following eEF2 departure , the aminoacyl-tRNA is placed in its final canonical position in the P site ( Figure 3B ) . This event triggers the disassembly of the CrPV-IRES PKI . Although the mRNA-like part remains placed in the E site of the 40S , the ASL-like segment experiences a pronounced displacement to occupy the E site of the 60S , now mimicking the acceptor stem of a canonical E site tRNA ( Figure 3B C ) . The L1 . 1 part of CrPV-IRES remains attached to the L1 stalk along this process whose position relative to the 60S is similar to the one described in the complex with eEF2 and a non-hydrolyzable GTP analog or after the first translocation ( Figure 4A ) ( Murray et al . , 2016; Muhs et al . , 2015 ) . The back-swiveling of the 40S head upon eEF2 departure is also involved in a new relative orientation of PKII and PKIII ( Figure 3B C ) . The swiveled configuration ( Figure 4B , left ) , SL-IV and SL-V , components of PKII , are exposed to the solvent , in a similar position described for the single translocated CrPV-IRES ( Muhs et al . , 2015 ) . The eukaryotic specific protein eS25 , a key element of the small subunit involved in early recruitment of the CrPV-IRES as well as in the positioning of the IRES in the pre-translocation stage ( Schüler et al . , 2006 ) , is not interacting with the IRES ( Figure 4B , left arrow ) . Upon back swiveling of the 40S head , a new interaction is established between the CrPV-IRES and eS25 ( Figure 4B right , arrow and ( C ) not involving SL-V like in the pre-translocated complex . The α-helix of eS25 comprising residues 52 to 65 could be observed in interacting distance with a helical segment of the CrPV-IRES formed by residues 6123–6127 and 6159–6164 ( Figure 4C , top right ) . This new interaction stabilizes the CrPV-IRES in a distinctive conformation , with PKII and PKIII assembled but with a wider relative orientation ( Figure 4C ) and the PKI disassembled with residues 6175 to 6200 ( Figure 3C ) corresponding to the ASL mimicking part of PKI , populating a space corresponding to the acceptor stem of a canonical E site tRNA ( Figure 4C , bottom right and Video 1 ) .
The structures provide a structural view of the archetypical CrPV-IRES in the final stage of initiation , after transitioning through the ribosome . Combining the structures with published biochemical and smFRET data allows us to propose a comprehensive working model for how the CrPV-IRES ( and type IV IRES in general ) recruits , manipulates and redirects host ribosomes for the synthesis of its own proteins ( Figure 5 ) . As suggested by classic cross-linking experiments ( Pestova et al . , 2004 ) , smFRET data ( Petrov et al . , 2016 ) and cryoEM reconstructions ( Spahn et al . , 2004; Murray et al . , 2016 ) , CrPV-IRES initially assembles a binary 80S/CrPV-IRES complex by either directly recruiting empty 80S or by a step-wise pathway in which CrPV-IRES first recruits the 40S subunit and then the 60S subunit ( Figure 5 , bottom left ) . Once the binary 80S/CrPV-IRES is assembled , the 40S oscillates between rotated and non-rotated states , with PKI inserted in the A site and minimum changes in the overall conformation of the IRES ( Fernández et al . , 2014; Koh et al . , 2014 ) . These movements are coupled to oscillations of the L1 stalk . The rotated state is the substrate of eEF2 , which , in its GTP-bound form , induces an additional rotation of the small subunit and additional displacement of the L1 stalk , to facilitate the translocation of the PKI from the A to the P site ( Figure 5A , top left ) . Back rotation and back swiveling of the 40S , combined with ribosome-induced GTP hydrolysis by eEF2 results in the first translocation event of the CrPV-IRES , positioning PKI in the P site , mimicking a translocated , canonical aminoacyl-tRNA . This intermediate is unstable and prone to back-translocation ( Muhs et al . , 2015 ) , unless a cognate aminoacyl-tRNA , delivered to the ribosome in complex with eEF1A and GTP , captures the frame in the A site of the ribosome ( Petrov et al . , 2016 ) . Formation of this complex is a rate-limiting step in this kinetically driven process ( Petrov et al . , 2016 ) . In the single translocated IRES state , SL-IV and SL-V , which are initially attached to the ribosome , are solvent exposed , the PKI occupy the P site of the 40S and an aminoacyl-tRNA occupies the A site . It is reasonable to assume this state will oscillate between rotated and non-rotated configurations of the small subunit as a canonical pre-translocation complex with tRNAs ( Budkevich et al . , 2011 ) . The second translocation step is required to place the first aminoacyl-tRNA in the P site and thus finish the initiation phase of translation ( Figure 5A , bottom right ) . At this stage , CrPV-IRES translocation should be coupled with the movement of the aminoacyl-tRNA occupying the A site , and this seems to happen with a conformation of the IRES similar to the one reported for the first translocation ( Muhs et al . , 2015 ) . This conformation is maintained until the very last moment as the intermediate captured here with eEF2 presents a back-rotated configuration of the 40S ( Abeyrathne et al . , 2016 ) . However , a pronounced swiveling of the 40S head is in place , probably induced by the presence of eEF2 ( Abeyrathne et al . , 2016 ) . Once eEF2 leaves , the back-swiveling movement of the 40S head triggers a dramatic conformational change in the CrPV-IRES: PKI is disassembled resulting in the ASL-like segment relocating to mimic the acceptor stem of a canonical E site tRNA . The mRNA-like element of the disassembled PKI remains in the E site of the 40S . These conformational changes in the PKI of the CrPV-IRES upon back swiveling are combined with a reconfiguration of the relative positioning of PKII and PKIII . This new conformation is stabilized by a newly reported IRES/40S interaction with the ribosomal protein eS25 , which is also involved in the early recruitment of the IRES to the 40S ( Murray et al . , 2016 ) . The conformational change described here for the CrPV-IRES following translocation through the ribosome unexpectedly resembles the transition observed for the Hepatitis C Virus ( HCV ) IRES upon aminoacyl-tRNA delivery to the P site ( Figure 6 ) ( Yamamoto et al . , 2014; Yamamoto et al . , 2015 ) . The HCV-IRES belongs to a different class of IRES , due to its requirement of some canonical factors to initiate translation ( Filbin and Kieft , 2009; Yamamoto et al . , 2017 ) . It also interacts with the ribosome in a different manner ( Quade et al . , 2015 ) . However , a large stem ( Figure 6 , domain II , blue ) reaches the E site of the 40S and is maintained base paired with the mRNA-like part of this IRES by a tilted configuration of the 40S head ( Figure 6B ) ( Yamamoto et al . , 2015 ) . Upon delivery of initiator tRNA to the P site , the head recovers its non-tilted configuration resulting in the repositioning of the domain II to occupy a similar space as the CrPV IRES in the E site of the 60S ( Figure 6 , right ) . Therefore , to assemble translationally competent ribosomes , distantly related IRESs have converged on a similar mechanism to regulate the placement of the first aminoacyl-tRNA in the P site of the ribosome , by resembling endogenous tRNA states .
Expression vector for His-tagged eRF1* ( AGQ mutant ) ( Frolova et al . , 1999 ) and transcription vector for Leu-tRNA have been previously described ( Pisarev et al . , 2010 ) . Transcription vector for CrPV-Leu-STOP was constructed inserting a T7 promoter sequence upstream of CrPV IGR IRES sequence followed by the two first coding triplets and an EcoRI site , using pUC19 as a scaffold vector . Site-directed mutagenesis was employed to change the first coding triplet to CUU encoding leucine and the second coding triplet to a stop ( UAG ) codon , rendering the CrPV-Leu-STOP construct . CrPV-Leu-STOP RNA and Leu-tRNA were transcribed using T7 RNA polymerase . Native 40S and 60S subunits , eEF2 , rabbit aminoacyl-tRNA synthetases ( Alkalaeva et al . , 2006 ) , and eEF1A ( Carvalho et al . , 1984 ) were prepared as previously described . Recombinant eRF1* was purified according to a previously described protocol ( Alkalaeva et al . , 2006 ) . In vitro transcribed Leu-tRNA was aminoacylated with leucine in the presence of rabbit aminoacyl-tRNA synthetases as previously described ( Pisarev et al . , 2010 ) . To reconstitute different ribosomal complexes , we incubated 1 . 8 pmol 40S ribosomal subunits with 2 pmol CrPV-Leu-STOP RNA in a 20 μl reaction mixture containing buffer A ( 20 mM Tris-HCl , pH 7 . 5 , 100 mM KCl , 2 . 5 mM MgCl2 , 0 . 1 mM EDTA , 1 mM DTT ) with 0 . 4 mM GTP for 5 min at 37 . Then , the reaction mixture was supplemented with 2 . 5 pmol 60S ribosomal subunits and additionally incubated for 5 min at 37 . Next , we added 10 pmol eEF1A , 3 pmol eEF2 , and 0 . 4 Leu , and incubated for 5 min at 37 . Finally , ribosomal complexes were incubated with 20 pmol eRF1 ( AGQ ) for 5 min at 37 . We analyzed the assembled ribosomal complexes via a toe-printing assay essentially as described ( Pestova and Hellen , 2005 ) . Aliquots of 3 μl of assembled ribosomal complexes at concentration of 80–100 nM were incubated for 30 s on glow-discharged holey gold grids ( UltrAuFoil R1 . 2/1 . 3 ( Russo and Passmore , 2016 ) ) , on which a home-made continuous carbon film ( estimated to be 50Åthick ) had previously been deposited . Grids were blotted for 2 . 5 s and flash cooled in liquid ethane using an FEI Vitrobot . Grids were transferred to an FEI Titan Krios microscope equipped with an energy filter ( slits aperture 20 eV ) and a GatanK2 detector operated at 300 kV . Data were recorded in counting mode at a magnification of 130 , 000 corresponding to a calibrated pixel size of 1 . 08 Å . Defocus values ranged from 1 . 6 to 3 . 6 μm . Images were recorded in automatic mode using the Leginon ( Carragher et al . , 2000 ) software and frames were aligned with Motioncor2 ( Zheng et al . , 2017 ) and checked on the fly using APPION ( Lander et al . , 2009 ) . Contrast transfer function parameters were estimated using GCTF ( Zhang , 2016 ) and particle picking was performed using GAUTOMACH without the use of templates and with a diameter value of 260 pixels . All 2D and 3D classifications and refinements were performed using RELION ( Scheres , 2012 ) . An initial 2D classification with a four times binned dataset identified all ribosome particles . A second 2D classification step with two times binned data was employed to separate 80S from 40S particles . A consensus reconstruction with all 80S particles was computed using the AutoRefine tool of RELION whose resulting map was used to build a mask containing the inter-subunit space and the L1 stalk . 3D classification with signal subtraction using the previously described mask and a T value of 10 allowed for the identification of several population of ligands inside the mask , namely empty ribosomes , pre-translocated CrPV IRES and double translocated populations with aminoacyl-tRNA and eEF2 or eRF1* . Final refinements with unbinned data for the classes selected yielded high resolution maps with density features in agreement with the reported resolution . Local resolution was computed with RESMAP ( Kucukelbir et al . , 2014 ) . Models for the mammalian ribosome , Leu-tRNALeu , eEF2 and eRF1* were docked into the maps using CHIMERA ( Pettersen et al . , 2004 ) and COOT ( Emsley and Cowtan , 2004 ) was used to manually adjust the L1 stalk and rebuild CrPV IRES using our previous model as initial step . An initial round of refinement was performed in Phenix using real space refinement with secondary structure restrains ( Adams et al . , 2011 ) . A final step of reciprocal-space refinement using REFMAC was performed ( Murshudov et al . , 1997 ) for the eRF1* complex . The fit of the model to the map density was quantified using FSCaverage and Cref . | Viruses cannot replicate themselves , but instead depend on components of the host cell for their own survival . Once a virus successfully enters a cell , it must use part of the cell’s machinery – specifically the ribosomes – to produce its own proteins . Ribosomes normally make the cell’s proteins by reading instructions written in molecules known as messenger RNAs ( or mRNAs for short ) . Viruses hijack ribosomes using structured RNA segments in its mRNAs that can mimic natural components of the cell’s protein-producing machinery . These RNA sequences , known as IRESs , feature a refined balance between rigidity and flexibility . Their flexible nature has made them difficult to study in the past , though the latest advances in electron cryo-microscopy mean that IRESs can now be directly observed in complex with ribosomes . Pisareva et al . sought to image a prototypical IRES sequence from the Cricket Paralysis Virus as it is transitioned through the ribosome . The idea was to characterize the late stages of ribosome hijacking . First , all the essential components were purified , mixed in the laboratory , and then imaged via electron cryo-microscopy . Image processing and sorting algorithms were then used to visualize the process at a high level of detail . Unexpectedly , this showed that the IRES changes shape dramatically to mimic part of another RNA molecule , a tRNA , when it reaches the so-called exit site of the ribosome . Short for transfer RNAs , tRNAs are molecules that bring the building blocks of proteins ( called amino acids ) to the ribosome , ready to be linked together . The shape change in the IRES is coupled with the placement of the first amino acid-loaded tRNA in a site on the ribosome that commits it to producing the viral protein . These results illustrate the remarkable ability of RNA molecules , in general , and IRES sequences , in particular , to adopt distinctive and context-specific shapes . These features seem to be widely conserved among diverse virus families as a similar shape change has been see in the IRES of the distantly related Hepatits C Virus . Together these new insights could lead to new strategies to interfere with viral replication and further studies that deepen our understanding of how ribosome and RNA-based mechanisms work generally inside cells . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"biochemistry",
"and",
"chemical",
"biology",
"structural",
"biology",
"and",
"molecular",
"biophysics"
] | 2018 | Dual tRNA mimicry in the Cricket Paralysis Virus IRES uncovers an unexpected similarity with the Hepatitis C Virus IRES |
In budding yeast , a single cenH3 ( Cse4 ) nucleosome occupies the ∼120-bp functional centromere , however conflicting structural models for the particle have been proposed . To resolve this controversy , we have applied H4S47C-anchored cleavage mapping , which reveals the precise position of histone H4 in every nucleosome in the genome . We find that cleavage patterns at centromeres are unique within the genome and are incompatible with symmetrical structures , including octameric nucleosomes and ( Cse4/H4 ) 2 tetrasomes . Centromere cleavage patterns are compatible with a precisely positioned core structure , one in which each of the 16 yeast centromeres is occupied by oppositely oriented Cse4/H4/H2A/H2B hemisomes in two rotational phases within the population . Centromere-specific hemisomes are also inferred from distances observed between closely-spaced H4 cleavages , as predicted from structural modeling . Our results indicate that the orientation and rotational position of the stable hemisome at each yeast centromere is not specified by the functional centromere sequence .
Centromeres are the genetic loci that organize the proteinaceous kinetochore , which attaches to spindle microtubules to pull the chromosomes to the poles in both mitosis and meiosis . There is general agreement in the centromere field that the central determinant of centromere identity and propagation is the special centromeric nucleosome containing the cenH3 ( CENP-A in mammals and Cse4 in budding yeast ) histone variant ( Quenet and Dalal , 2012 ) . CenH3 nucleosomes have been shown to occupy the centromeres of nearly all eukaryotes studied , and to be necessary for kinetochore formation . Despite the central importance of this nucleosome , its composition and structure have been the subject of controversy . In vitro and in vivo studies have led to proposals for several mutually exclusive models , including conventional octameric ( cenH3/H4/H2B/H2A ) 2 nucleosomes ( ‘octasomes’ ) ( Camahort et al . , 2009 ) , cenH3/H4/H2B/H2A half-nucleosomes ( ‘hemisomes’ ) ( Dalal et al . , 2007 ) , ( cenH3/H4 ) 2 tetrasomes ( Aravamudhan et al . , 2013 ) , mixed ( cenH3/H3/H42/H2B2/H2A2 ) octasomes ( Lochmann and Ivanov , 2012 ) and ( cenH3/H4/Scm3 ) 2 hexasomes ( Mizuguchi et al . , 2007 ) , where Scm3 is a cenH3-specific histone chaperone . Evidence for each of these conflicting models has been presented for budding yeast , where the centromere is genetically defined by an ∼120-bp functional sequence on each of the 16 chromosomes . The functional centromere has a tripartite organization: the 8 bp CDEI sequence is a binding site for the Cbf1 protein , the Cse4 nucleosome maps to the 78–86 bp CDEII sequence , and the 26 bp CDEIII sequence is a binding site for the Cbf3 complex ( http://www . yeastgenome . org ) . Our previous native chromatin immunoprecipitation ( ChIP ) study has resolved all three particles at base-pair resolution , confirming that the Cse4-containing nucleosome is confined to CDEII ( Krassovsky et al . , 2012 ) . The implied single wrap of DNA around the Cse4-containing histone core is consistent with previous evidence that the Cse4 nucleosome wraps DNA in a right-handed orientation in vivo ( Furuyama and Henikoff , 2009; Huang et al . , 2011 ) , opposite to the left-handed wrap of conventional nucleosomes ( Tachiwana et al . , 2011 ) . However , conflicting structural interpretations have continued to appear , with some authors arguing for partially unwrapped octasomes ( Dunleavy et al . , 2013; Hasson et al . , 2013; Miell et al . , 2013; Padeganeh et al . , 2013 ) , others for cenH3/H4 octasomes ( Lochmann and Ivanov , 2012 ) , others for tetrasomes ( Aravamudhan et al . , 2013 ) , and others for hemisomes throughout the cell cycle but octasomes at anaphase ( Shivaraju et al . , 2012 ) . Although there have been suggestions of more than one Cse4 nucleosome per budding yeast centromere based on fluorescent microscopy ( Coffman et al . , 2011; Lawrimore et al . , 2011 ) , more recent evidence confirms that there is only one particle per centromere ( Henikoff and Henikoff , 2012; Shivaraju et al . , 2012; Haase et al . , 2013 ) . As budding yeast is the only model organism where there is a 1:1 relationship between the cenH3 nucleosome and the microtubule attachment site ( Furuyama and Biggins , 2007 ) , any conclusion concerning its composition and structure has an unequivocal functional interpretation . To definitively settle the controversy over budding yeast centromeric nucleosome composition and structure , we have turned to an in vivo mapping method that individually characterizes every nucleosome in the genome . H4S47C-anchored cleavage mapping determines the precise position and orientation of all histone H4 molecules in an unbiased manner ( Brogaard et al . , 2012a ) . By mapping the obligate H4 partner of Cse4 , we avoid potential complications arising from the need for antibodies , tags , nucleases or fluorescence . As H4 is the obligate partner of every H3 in the genome , this method provides ∼75 , 000 positive control H3 nucleosomes to compare against the 16 Cse4 nucleosomes at centromeres . There are important advantages to this approach to mapping yeast centromeres over previous mapping methods , such as Micrococcal Nuclease with sequencing ( MNase-seq ) and native ChIP . Unlike MNase , an endo-exonuclease that preferentially cleaves linkers between nucleosomes and so provides a map of regions protected from cleavage , H4S47C-anchored cleavage mapping determines base-pair positions within the particle . This means that mapping is not complicated by exonucleolytic ‘nibbling’ and internal cleavages that can lead to uncertainty as to the true size of a particle . In addition , CDEIIs are >90% A+T , and MNase prefers AT-rich regions , resulting in preferential exonucleolytic digestion and loss of centromeric DNA , whereas H4S47C-anchored cleavage is exclusively endonucleolytic and has no sequence bias . When combined with paired-end deep sequencing , H4S47C-anchored cleavage mapping provides precise center-to-center distances between adjacent particles that can be used to infer interactions between neighboring nucleosomes and to probe higher-order structural properties without chromatin solubilization . Most importantly for the problem being addressed in this study , H4S47C-anchored cleavage mapping predicts very different cleavage patterns for octasomes and tetrasomes , which are symmetrical , than for hemisomes , which are asymmetrical . Using H4S47C-anchored cleavage mapping we show that both the overall pattern and the distances between cleavages are incompatible with all proposed octasome or tetrasome particles . Rather , the pattern and cleavage distances indicate mutually exclusive occupancy of CDEII by oppositely oriented hemisomes in two rotational phases at similar frequencies . Our findings reveal surprising flexibility in orientation and phasing for a nucleosome particle that is tightly confined within an asymmetric ∼80-bp DNA loop .
In the original description of H4S47C-anchored cleavage mapping , nucleosome centers were determined as clusters of cleavages around the dyad axes of highly occupied and phased nucleosomes throughout the budding yeast genome ( Brogaard et al . , 2012b ) . It appeared that cleavages at centromeres were different from cleavages at other nucleosomes , although the empirical model used to interpret cleavage patterns did not permit further inferences concerning the structure and composition of the particle over centromeric DNA . Specifically , cleavages at the two closely spaced H4S47C residues can potentially result in <10-bp DNA duplexes that cannot be uniquely mapped in the genome , making the empirical approach unsuitable for discriminating cleavage patterns generated by single H4S47C residues in Cse4 hemisomes from those generated by two H4S47C residues in a canonical nucleosome . Therefore , we applied molecular dynamics simulations based on the high-resolution structure of the nucleosome core particle ( Davey et al . , 2002 ) with a single phenanthroline group attached by derivatizing the cysteine sulfide with N ( 1 , 10-phenanthroline-5-yl ) iodoacetamide ( Figure 1A ) . The phenanthroline moiety chelates a copper ion , and when it contacts the C-1 hydrogen ( C1H ) of deoxyribose through a hydroxyl radical , initiates a series of elimination reactions that result in strand cleavages , releasing the deoxyribonucleotide and leaving 5′- and 3′- phosphate ends ( Sutton et al . , 1993 ) . The simulations sampled possible conformations of a single H4S47C-phenanthroline-Cu+ in the context of the nucleosome , while the DNA and the rest of the protein were held static ( Video 1 ) . From these simulations , we observed that the copper can be within 4 Å of the C1H atoms at −2 , −3 , and −4 on the Watson strand and −5 , −6 on the Crick strand with respect to the nucleosome dyad axis ( Figure 1B ) . Contacts on both strands imply that a single H4S47C-phenanthroline-Cu+ can catalyze double-strand cleavages . 10 . 7554/eLife . 01861 . 003Figure 1 . Structural model for DNA cleavage by H4S47C-phenanthroline-Cu+ . ( A ) Snapshot from a molecular dynamics simulation showing a copper ion ( blue dot ) bound to phenanthroline within 4 Å ( dotted line ) of the C1H atom of the deoxyribonucleotide at the Dyad-3 position ( red ) . Phenanthroline ( shown in stick representation ) docks in the minor groove of the nucleosomal DNA ( shown in surface representation ) . ( B ) Cleavage positions with respect to the dyad axis on the Watson ( W ) and Crick ( C ) strands . The degree of red shading corresponds to the probability of predicted contact . ( C ) H4S47C-anchored cleavage and processing . The steps involved in H4S47C-anchored cleavage mapping are illustrated for an instance in which H4S47C-Phenanthroline-Cu cleavages occur on both sides of the dyad . The cleaved positions are indicated as red squares marked with crosses . In this instance , the W-C distance would be +12 . Note that that right end of a fragment marks the cleavage position on the Crick strand , and the left end marks the cleavage position on the Watson strand . See also Figure 1—figure supplement 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 01861 . 00310 . 7554/eLife . 01861 . 004Figure 1—figure supplement 1 . Determination of cleavage positions by in vitro cleavage mapping . A 147-bp duplex with the 601 nucleosome positioning sequence was used to reconstitute H3 octasomes and subjected to in vitro labeling and cleavage reactions . Sequencing libraries were constructed and cleavages were mapped to one orientation of the 601 sequence , where the dyad axis is the 74th base pair from either end . The sequenced positions are graphed , where most of the strong peaks observed are among those predicted by our structural model ( see Figure 5—figure supplement 1 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01861 . 00410 . 7554/eLife . 01861 . 005Video 1 . Molecular dynamics simulation of DNA contacts by a copper ion chelated to H4S47C-anchored 1 , 10-orthophenanthroline . See the legend to Figure 1A . DOI: http://dx . doi . org/10 . 7554/eLife . 01861 . 005 In an octasome , H4S47C-phenanthroline-Cu+ contacts made by the first H4 correspond to contacts at +5 and +6 on the Watson strand and +2 , +3 and +4 on the Crick strand made by the second H4 . We interpret the contacts made by both H4s as predictive of single-strand cleavages creating 5′ and 3′ ends on either side of the deoxyribonucleotide under attack ( e . g . , Figure 1C ) . Importantly , these predicted cleavages on both sides of the dyad axis are among those deduced by Brogaard et al . from their empirical in vivo data ( Brogaard et al . , 2012b ) , and confirmed by cleavage mapping of reconstituted octasomes ( Figure 1—figure supplement 1 ) . Previous high-resolution MNase and native-ChIP mapping of Saccharomyces cerevisiae centromeres revealed that the distance between the midpoint of the Cse4 nucleosome occupying CDEII to the midpoint of immediately flanking nucleosomes is ∼200–250 bp ( Cole et al . , 2011; Krassovsky et al . , 2012 ) . In the original study using H4S47C-anchored cleavage , fragments were size-selected to be mostly 125–200 bp , which resulted in very low coverage of individual centromeres ( Brogaard et al . , 2012b ) . To avoid such size-related biases , we followed essentially the same protocol in generating cleavage products , but applied a modified Solexa paired-end DNA sequence library preparation protocol that does not include a size-selection step ( Henikoff et al . , 2011 ) . In both biological and technical replicates using several variations in the protocol ( ‘Materials and methods’ ) , we obtained virtually identical fragment length distributions ( Figure 2A ) , indicating a high degree of repeatability and robustness in the basic cleavage mapping protocol . Centromere function is normal in the H4S47C mutant cell line , as we observed virtually identical doubling times of 90 min and a CEN plasmid retention rate of 99% per generation for the H4S47C strain ( Figure 2—figure supplement 1 ) . 10 . 7554/eLife . 01861 . 006Figure 2 . Pairs of cleavage clusters occupy CDEs . ( A ) Fragment length distributions for three experiments . For each length in base pairs , the percentage of the total number of mapped fragments is plotted . Experiment 1 ( using Phusion DNA polymerase ) : 27 , 871 , 803 fragments; Experiment 2 ( using KAPA polymerase ) : 62 , 926 , 977 fragments; Experiment 3: 81 , 473 , 515 fragments ( biological replicate using KAPA ) . ( B–K ) Examples of cleavage profiles around centromeres , where tracks represent the total number of fragment ends within successive 10-bp windows . ( B ) Chromosome 3 profile , where the position of the centromere and of the most frequently cleaved nucleotide position on the chromosome are indicated . ( C ) Expansion of the region indicated by the bracket in ( B ) . ( D ) Expansion of the region indicated by the bracket in ( C ) , where the extent of Cen3 is indicated . ( E ) Expansion of the region around the most frequently cleaved nucleotide position at the same scale as in ( D ) . ( F ) and ( G ) Same as ( D ) and ( E ) , respectively , for Chromosome 1 . ( H ) and ( I ) Same as ( D ) and ( E ) , respectively , for Chromosome 2 . ( J ) and ( K ) Same as ( D ) and ( E ) , respectively , for Chromosome 4 . ( See also Figure 2—figure supplement 1 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01861 . 00610 . 7554/eLife . 01861 . 007Figure 2—figure supplement 1 . The log-phase doubling rate of the H4S47C strain is normal . ( A ) Mid-log phase cultures were held at 4°C and used to inoculate flask cultures 1:10 in YPD medium . Cultures were shaken at 30°C , and aliquots were removed for counting , using a Millipore ( Billerica , MA ) Vi-cell automated cell counter . At the cell densities used in our experiments ( ∼1–2 × 107 , blue arrow ) both the parent strain ( BY4741 ) and the mutant strain ( H4S47C in a BY4741 background ) doubled every 90 min and were 97–98% viable ( B ) . ( C ) H4S47C mutant cells grow normally at elevated temperatures . We attribute the growth and temperature-sensitivity phenotypes previously reported for the H4S47C strain ( Brogaard et al . , 2012a ) to its failure to enter a quiescent state upon nutrient depletion . This failure results in cultures growing to higher densities , loss of viability >1 day after reaching stationary phase , smaller colonies , and 8% larger cell volumes . The H4S47C strain shows 99% CEN plasmid retention compared to 99 . 99% plasmid retention for BY4741 in plasmid loss assays ( Koshland et al . , 1987 ) ( data not shown ) . This phenotypic syndrome is likely attributable to the replacement of the two histone H4 genes , HHF1 and HHF2 with a single mutant HHF1 gene , resulting in reduced H4 dosage: Deleting HHF2 causes reduced starvation resistance ( Davey et al . , 2012 ) , and partial histone depletion in general is associated with reduced longevity ( Feser et al . , 2010 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01861 . 007 We mapped paired-end reads to the budding yeast genome , and except as noted , pooled all 172 , 272 , 295 mapped fragments for the analyses below . As nucleosome center-to-center distances should be ≥147 bp , we excluded fragments <147 bp , which resulted in a cleavage map ( Figure 2B , C ) that is very similar to that described previously ( Brogaard et al . , 2012b ) . This confirms that cleavage clusters correspond to nucleosomes throughout the genome . However , the cleavage pattern over centromeres is different from the large majority of cleavages on chromosome arms . As previously noted for all 16 aligned centromeres ( Brogaard et al . , 2012b ) , individual centromeres display two cleavage clusters spaced ∼40 bp apart centered over the middle of each CDEII ( Figure 2D , F , H , J ) . MNase-seq and Cse4 native ChIP studies have shown that the Cse4 nucleosome is almost perfectly aligned over CDEII of each functional centromere , and we wondered whether the double peak pattern of cleavage clusters over each centromere is seen at other highly occupied and phased nucleosomes . To test this possibility , we manually examined the region around the most frequently cleaved base pair on each chromosome ( e . g . , Figure 2E , G , I , K ) but detected no double peak pattern similar to that seen over centromeres . We next exhaustively tested the uniqueness of the centromere pattern . A profile was constructed from an ungapped alignment of centromeres of similar length ( 117–120 bp ) , and the average count at every position in the profile was used to scan every ungapped alignment in the genome ( median 764 , 931 alignments per chromosome ) . We scored alignments as the Pearson correlation ( r ) between the profile and the chromosome . For each chromosome , a centromere alignment to the profile ( median r = 0 . 76 ) scored above the best non-centromere alignment ( median r = 0 . 40 ) ( Table 1 ) . Removing each centromere from the model and rescanning ( ‘delete-one jackknife’ ) reduced their scores accordingly , but the median correlation ( r = 0 . 52 ) still scored above the median correlation for the best non-centromeric alignment ( r = 0 . 41 ) . We conclude that the conspicuously different cleavage pattern at centromeres is unique to centromeres . 10 . 7554/eLife . 01861 . 008Table 1 . Profile scanning for the centromere-specific cleavage pattern*DOI: http://dx . doi . org/10 . 7554/eLife . 01861 . 00815 centromere ProfileDelete-one jackknife†Chr# aligned# pass filtersFirst CenFirst FP# pass filtersFirst CenFirst FP1230 , 10760 . 778‡0 . 31750 . 7450 . 3032813 , 073340 . 8490 . 40123–0 . 4203316 , 50980 . 822‡0 . 38870 . 5710 . 4264§1 , 531 , 822–––31–0 . 3945576 , 763510 . 8070 . 611¶48–0 . 625¶6270 , 050140 . 726‡0 . 389130 . 7230 . 30671 , 090 , 829330 . 756‡0 . 450320 . 7070 . 4668562 , 53270 . 459‡0 . 3464–0 . 2939439 , 777110 . 4930 . 3752–−0 . 09210745 , 640220 . 626‡0 . 402190 . 5790 . 35711666 , 705230 . 634‡0 . 40715–0 . 326121 , 078 , 066480 . 742‡0 . 430470 . 6980 . 45313924 , 320480 . 873‡0 . 491530 . 8470 . 49814784 , 222230 . 778‡0 . 398160 . 7480 . 384151 , 091 , 180210 . 7960 . 45711–0 . 42016947 , 955320 . 764‡0 . 455230 . 4640 . 478Median**764 , 93122 . 50 . 7600 . 40217 . 50 . 5180 . 407*Alignments to the profile with more than three positions greater than three standard deviations from the mean of the profile position or with a maximum position less than the smallest maximum position within the profile ( 186 ) were excluded ( filters ) . Pearson correlation coefficients are shown . †Only jackknife results for the centromere deleted from the model are shown . ‡Multiple high-scoring centromere hits above the first false positive ( FP ) one or two base pairs apart . §Cen4 ( 111 bp ) was not included in the model . ¶Single base-pair cleavage peak at a site of anomalously low nucleosome occupancy . **Medians are based on all alignments for all 16 chromosomes , whether or not they passed the filters . Each double-strand cleavage should generate two fragments , and if they are recovered with equal probability , then the expected frequency of cleavage at a site for the left paired end should be the same as that for the right paired end . However , at all 16 centromeres , we observed a strong bias against recovery of the right-end fragment for the cleavage cluster on the left ( on the CDEI side of CDEII ) and against recovery of the left-end fragment for the cleavage cluster on the right ( on the CDEIII side of CDEII ) ( Figure 3A , C ) . In contrast , there was no consistent recovery bias seen for the single peaks of the most frequently cleaved base pair on every chromosome when similarly aligned ( Figure 3B , D ) . We suspected that the presence of ∼40 bp more of the >90% AT-rich CDEII sequence on a fragment caused it to be strongly discriminated against during some step in the sequencing pipeline . Consistent with this possibility , we found that when fragments are aligned around the mid-Cen and mapped as normalized counts , in addition to the strong depletion directly over CDEII , there is a gradient of decreasing depletion with distance from the centromere ( Figure 3E , red line ) . This gradual decrease with distance is expected if the fragments that are preferentially lost are those that span most of CDEII . 10 . 7554/eLife . 01861 . 009Figure 3 . Fragments spanning the mid-Cen are depleted due to high A + T content . ( A–D ) Histograms of cleavage counts in successive 10-bp intervals are plotted separately for left ( blue above ) and right ( magenta below ) ends of mapped fragments . ( A ) The 1-kb interval around the mid-Cen of Cen3 . Dotted line marks the mid-Cen position . ( B ) The 1-kb interval around the most frequently cleaved nucleotide position on Chromosome 3 . ( C ) Same as ( A ) for all 16 aligned mid-Cens . ( D ) Same as ( B ) for all 16 aligned most frequently cleaved nucleosome positions . ( E ) Fragments were stacked over aligned mid-Cens , and normalized counts for each base-pair position were calculated ( red line ) , where a value of 1 represents the random expectation . For comparison , normalized count plots are superimposed for the 40 , 83-bp ≥90 . 5% AT-rich intervals ( dots ) , where data for intervals 90–95% AT-rich are plotted in blue and intervals >95% AT-rich are plotted in green . Strong depletion is seen directly over the middle of aligned CDEIIs , gradually approaching 1 with distance from the mid-Cen . Similarly , strong depletion with gradual approach to 1 is also seen for all 83-bp AT-rich control sequences . ( F ) The percentage of expected number of cleavage sites within the 16 CDEII sequences , comprising a median of 83 bp , and that of the 46 sequences ≥90% A+T are plotted as a function of AT-richness . Depletion relative to expectation is seen for CDEIIs , indicating that the bias against recovery can be accounted for by high AT-richness . DOI: http://dx . doi . org/10 . 7554/eLife . 01861 . 009 To ascertain whether the preferential loss of CDEII-spanning fragments is attributable to the well-known bias against the most AT-rich sequences in Solexa sequencing ( Bartfai et al . , 2010; Lopez-Barragan et al . , 2010 ) , we compared the recovery of CDEII-containing fragments to the recovery of non-centromeric fragments of similar AT-richness and length . There are 46 segments in the budding yeast genome that are at least 90% AT-rich over 83-bp , the median length of CDEII , and all of them are under-represented relative to expectation in our cleavage datasets ( Figure 3E ) . Importantly , the degree of representation decreases significantly with increasing AT-richness between 90–100% ( r = 0 . 52 ) , showing depletions similar to what is seen for CDEII-containing fragments ( Figure 3F ) . We therefore attribute the preferential loss of CDEII-containing fragments to their AT-richness , and not to any other feature of centromeric DNA . We observed similar losses for a Solexa library produced using Phusion DNA Polymerase , which has been reported to bias against such strongly AT-rich sequences ( Bartfai et al . , 2010; Lopez-Barragan et al . , 2010 ) , and libraries produced using KAPA DNA Polymerase , which has been engineered to be exceptionally processive on sequences that are strongly compositionally biased ( Quail et al . , 2012 ) . It is more likely that an inherent feature of the Illumina platform , such as PCR-based cluster generation , is responsible for discrimination against sequencing templates that are >90% AT-rich ( Bartfai et al . , 2010 ) . When centromere cleavage maps were displayed at base-pair resolution , we observed that each of the two cleavage clusters resolved into cleavage pairs separated by ∼10 bp ( Figure 4A–F ) . A composite alignment of all 16 centromeres centered over the midpoint of the functional centromere ( mid-Cen ) revealed that the inner peak maxima were 33-bp apart and the outer peak maxima were 53-bp apart ( Figure 4A ) . No such distinct spacings were seen for the most frequently cleaved base pairs on each chromosome ( Figure 4G ) . Because centromeres range in length from 111 bp for Cen4 to 120 bp for Cen12 , we pooled aligned centromere cleavage data within length classes . For centromere lengths of 117 , 118 , 119 and 120 , we found that the distance between the inner peak maxima was 33 bp and between the outer peak maxima was 53 bp ( Figure 4C–F ) . However , for the 111 bp Cen4 , the inner peak maxima were ∼25 bp apart and the outer peak maxima were ∼45 bp apart ( Figure 4B ) . Similar peak maxima were seen for gel-purified OP-labeled Cse4/H4S47C/H2A/H2B hemisomes reconstituted with a 78-bp Cen4 CDEII DNA duplex ( Furuyama et al . , 2013; Codomo et al . , 2014 ) that had been subjected to in vitro cleavage reactions ( Figure 4—figure supplement 1 ) . 10 . 7554/eLife . 01861 . 010Figure 4 . Pairs of 10-bp cleavage sites are symmetrically offset around the mid-Cen . ( A ) Composite histogram of cleavage counts for all fragments in successive 1-bp intervals for all 16 centromeres . Blue ( left end ) and magenta ( right end ) bars are stacked , such that the overall profile represents the sum at each base-pair position . ( B–F ) Same as ( A ) grouped by centromere size class as indicated . ( G ) Same as ( A ) for the most frequently cleaved nucleotide position on each of the 16 chromosomes . ( H ) Expansion of ( G ) showing the 6-bp average spacing between observed cleavages . ( I ) Expansion of ( A ) showing the 10-bp/33-bp/10-bp average spacing between cleavages observed over centromeres . See also Figure 4—figure supplement 1 and Figure 4—figure supplement 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 01861 . 01010 . 7554/eLife . 01861 . 011Figure 4—figure supplement 1 . Cleavage mapping of Cse4 hemisomes reconstituted on Cen4 CDEII DNA in vitro . Top: sequence of the 78-bp Cen4 CDEII sequence used for in vitro H4S47C-anchored cleavage mapping and map of inferred cleavage peaks . After purification of gel-shifted hemisomes showing a strong FRET signal and the unshifted control band , samples were subjected to cleavage reactions and analyzed on a sequencing gel . The FRET and unshifted lanes were scanned for Alexa488 ( top panel ) and Cy3 ( middle panel ) , and two Maxam-Gilbert ladder lanes ( Maxam and Gilbert , 1977 ) ( bottom panel ) were scanned for Cy3 . The panels are vertically aligned such that the base assignments in the top and middle panels correspond to base-pair positions in the sequencing ladders run alongside on the gel . DOI: http://dx . doi . org/10 . 7554/eLife . 01861 . 01110 . 7554/eLife . 01861 . 012Figure 4—figure supplement 2 . Cen4 spacing around the mid-Cen position is anomalously short . Cleavage counts for Cen3- to Cen4-spanning fragments >147 bp show that when aligned with the left edge of CDEI , the left peak pairs line up precisely between Cen3 and Cen4 for all three experiments . Histogram bars for three experiments are stacked at each base-pair position . When aligned with the left end of Cen3 or Cen4 , the right peak pair of Cen4 ( 111 bp ) is seen to be 7–8 bp closer to CDEI than for other centromeres ( 117–120 bp ) , suggesting that cleavage occurs preferentially at fixed positions from the nearer CDE junction . DOI: http://dx . doi . org/10 . 7554/eLife . 01861 . 012 Similar 10-bp peak-to-peak distances within cleavage clusters were observed for all 16 centromeres , both for clusters on the left and on the right of the mid-Cen position . As explained below , our structural model cannot account for 10-bp distances between single H4S47C-anchored cleavages , which implies that these cleavages represent independent particles that are rotated by 10-bp relative to one another . Such 10-bp spacings between H4S47C-anchored cleavages have previously been interpreted as differences in rotational phasing ( Brogaard et al . , 2012b ) . We can use these measurements to distinguish among the various structural models that have been proposed for the Cse4 nucleosome . The high-resolution crystal structure of the cenH3 octasome shows an arrangement of H4 residues , including S47 , that is virtually identical to that for the conventional H3 octasome ( Tachiwana et al . , 2011 ) , and so would predict a similar cleavage pattern over the centromere as is seen for highly occupied and phased nucleosomes genome-wide . This is clearly not the case at any yeast centromere , as even the minimal spacing of the inner cleavages is much larger than can be explained by our molecular dynamics model for a single nucleosome ( Figure 1 ) or is observed for highly occupied and phased H3 nucleosomes throughout the genome ( Figure 4H ) . Likewise , Cse4/H3 octasomes and ( Cse4/H4 ) 2 tetrasomes also predict the same mirror-image symmetrical arrangement of two H4s as canonical octasomes , and so our data also exclude those models . This leaves two other models , a version of the ( Cse4/H4/Scm3 ) 2 particle and the hemisome , as potentially explaining the cleavage patterns that we observe . The presence of Scm3 as an integral part of the Cse4 nucleosome has been excluded based on subsequent in vivo mapping ( Shivaraju et al . , 2011 ) and in vitro reconstitution studies ( Cho and Harrison , 2011; Xiao et al . , 2011 ) , which leaves the hemisome as the only model that can account for the cleavage patterns that we observe . Specifically , a model in which independent hemisomes align around the mid-Cen position in either orientation and in two rotational phases can account for all of the cleavage peaks that we observe over centromeres . Although 15 centromeres show nearly precise alignment to the functional centromere , with ∼10-bp/33-bp/10-bp spacings between peak maxima , Cen4 is an exception in showing ∼10-bp/25-bp/10-bp spacings ( Figure 4—figure supplement 2 ) . This ∼8-bp difference in cleavage pattern corresponds to the fact that , as pointed out above , Cen4 is 6–9 bp shorter than the other 15 centromeres . A parsimonious explanation of this correspondence is that the position of H4 residue 47 in the Cse4 core particle is fixed relative to the nearer junction regardless of whether the junction is CDEI-II or CDEII-III . Thus , cleavages on the left side of the mid-Cen position would preferentially occur a fixed distance from CDEI-II , whereas cleavages on the right side of the mid-Cen position would preferentially occur a fixed distance from the CDEII-III junction . In both cases , the 10-bp spacing between cleavage maxima implies that the rotational phase of the Cse4 nucleosome is established by where the nearer sequence-specific DNA-binding protein binds , whether it is Cbf1 at CDEI or the CBF3 complex at CDEIII . We next compared experimental cleavage distances to our simulated cleavage model , taking into account the Solexa library end-polishing procedure , which removes 3′ overhangs and fills in the complement to 5′ overhangs ( Figure 1C ) . At least one cleavage on the Watson strand and one on the Crick strand are required to observe a fragment by sequencing . Cu+ attack on DNA results in base-loss , and we would observe the adjacent position when the cleaved products are sequenced after Solexa end-polishing . We took an unbiased approach and asked what are the preferred distances between left and right ends of fragments observed experimentally . The left end of a fragment marks the position of a cleavage and base loss on the Watson strand and the right end of a fragment marks the position of a cleavage and base loss on the Crick strand . The distances between left and right ends thus reflect preferred cleavage locations . We observed peaks at −2 , +5 and +12 in the Watson-Crick ( W-C ) distributions ( Figure 5A , red curve ) , which are all explained by the DNA positions shown to be accessible to H4S47C-phenanthroline-Cu+ in our structural model ( Figure 5—figure supplement 1 ) . Additionally , the structural model helps us distinguish between W-C peaks due to cleavages by H4S47C on one side of the dyad and W-C peaks due to cleavages on two H4S47Cs across the dyad . The +5 W-C peak results from cleavages by the same H4 , whereas the −2 and +12 W-C peaks result from cleavages occurring on both sides of the dyad . 10 . 7554/eLife . 01861 . 013Figure 5 . Centromere cleavage distances are predicted by a single-H4 model . Distances between fragment ends are plotted for all fragments ( Genome-wide ) and for fragments mapping to within 125 bp of the center of CDEII of all chromosomes ( Centromeres ) . Distribution of the distances between ( A ) right and left fragment ends ( W–C ) ; ( B ) left fragment ends ( W–W′ ) ; ( C ) right fragment ends ( C–C′ ) . Gray dashed lines mark the peaks observed in the genome-wide distributions . See also Figure 5—figure supplement 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 01861 . 01310 . 7554/eLife . 01861 . 014Figure 5—figure supplement 1 . Examples of predicted cleavage distances . The peaks observed in W–C , W–W′ and C–C′ distributions arise from cleavages at specific positions predicted by the structural model . For each peak , an example is shown of how that distance is inferred . DOI: http://dx . doi . org/10 . 7554/eLife . 01861 . 014 An alternative way to distinguish between cleavages by a single H4 and cleavages by two H4s across a dyad is to measure distances between the left ends of fragments , W–W’ , and the right ends , C–C’ . Their genome-wide distributions revealed strong peaks at +1 and +7 ( Figure 5B , C red curves ) . The peak at +1 reflects cleavages due to the same H4 molecule on the same side of the dyad , while the peak at +7 reflects cleavages due to adjacent H4 molecules across the nucleosome dyad . Thus , the predicted cleavage sites imply that the fragment spacings observed genome-wide resulted from a combination of cleavages by H4 on both sides of the nucleosome dyad axis . In order to determine whether our simulated cleavage model can explain the patterns at centromeres , we generated W-C distributions for reads that are within 125 base-pairs of the center of CDEII ( Figure 5A , black curve ) . Whereas we observed a peak at +5 , which can arise from the presence of a single H4 , the peaks at −2 and +12 , which require two H4s on either side of the dyad axis , were not seen . It might be argued that we do not observe −2 and +12 peaks in the W-C distributions near CDEII because of preferential loss of AT-rich fragments spanning CDEII ( Figure 3E , F ) . However , in that case , we would also not see the +5 peak , suggesting that the loss of −2 and +12 peaks is not due to an AT-rich sequencing bias . To verify this interpretation , we examined same-strand ( W–W’ and C–C’ ) distributions , which obviates the need to recover fragments spanning CDEII . For the W–W’ and C–C’ distributions we again observed a strong peak at +1 ( Figure 5B , C , black curves ) , which can arise from a single H4 , whereas the peak at +7 , which results from cleavage across a dyad axis , was absent . Rather , the 10-bp peak spacing evident in the centromere maps ( Figure 4 ) and attributable to rotational phasing was a prominent feature in both W–W’ and C–C’ for centromeres , but not genome-wide . Taken together , the absence of peaks representing H4S47C-anchored cleavages across the dyad axis but the presence of peaks representing a single H4 at centromeres provides independent confirmation based on mapping of cleavage positions that single H4-containing particles predominate at centromeres . The simplest explanation for the cleavage patterns around the mid-Cen is that oppositely oriented hemisomes occupy different members of the cell population . However , the cleavage data alone do not exclude a model in which two oppositely oriented hemisomes flank the mid-Cen . To examine this possibility , we superimposed the H4S47C cleavage data with MNase-seq and Cse4 ChIP/input profiles , using a variation of the ‘V-plot’ method that we introduced to facilitate the precise determination of particle position and size for MNase-seq and other paired-end sequencing data ( Henikoff et al . , 2011 ) . In our original implementation , a V-plot consists of a dotplot representation in which a dot is placed on the X-axis for the position of a fragment midpoint relative to a fixed position and on the Y-axis at a position representing the length of the fragment . For H4S47C-anchored cleavage mapping , we are interested in plotting the precise left and right end positions , and so we generate two V-plots , one in which the X-axis position corresponds to the left end of a fragment and one in which it corresponds to the right end . When plotted in this way , with the mid-Cen as the fixed position on the X-axis and increasing fragment length on the Y-axis , we observe vertical lines of dots in pairs and corresponding pairs of diagonal lines that represent preferred cleavage sites at the same cleavage position on fragments of increasing length ( Figure 6 , Figure 6—figure supplement 1 ) . 10 . 7554/eLife . 01861 . 015Figure 6 . Paired cleavage sites on opposite sides of the Cse4 nucleosome over CDEII . Left-right V-plots with MNase and Cse4 ChIP/Input profiles superimposed for all centromeres aligned over the mid-Cen . The X-axis position for each blue dot corresponds to the left end of a mapped fragment and the X-axis position for each red dot corresponds to the right end . With the mid-Cen as the fixed position on the X-axis and increasing fragment length on the Y-axis , we observe vertical lines of dots in pairs representing preferred cleavages within CDEII . The blue and red diagonals respectively correspond to ( mostly background ) cleavages on the left and right sides of the centromere . See also Figure 6—figure supplements 1 and 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 01861 . 01510 . 7554/eLife . 01861 . 016Figure 6—figure supplement 1 . Left-right V-plots with MNase and Cse4 ChIP/Input profiles superimposed for all 16 centromeres . See legend to Figure 6 . DOI: http://dx . doi . org/10 . 7554/eLife . 01861 . 01610 . 7554/eLife . 01861 . 017Figure 6—figure supplement 2 . Left-right V-plot representation of likely in vivo cleavages over all 16 centromeres in a wild-type strain . ( A ) AluI and hydroxyl radical background cleavage data shown in Figure 6A with the average nucleosome occupancy ( black curve ) and Cse4 ChIP profiles ( green curve ) . ( B ) DNA was purified directly from cells subjected to immediate alkaline lysis and cleaved with AluI prior to library construction . See legend to Figure 6 . DOI: http://dx . doi . org/10 . 7554/eLife . 01861 . 017 Several notable features are revealed by the left-right V-plots . First , the two double vertical lines of dots seen on either side of the mid-Cen mark the location of two 10-bp cleavage pairs observed using conventional cleavage density histograms . Second , the fact that the vertical lines are densely populated up to ∼400 bp excludes the possibility that the deficiency of the AT-rich mid-CDEII-containing fragments is attributable in part to being on average 40-bp longer than their sister fragments and so discriminated against based on size . Third , clusters of left and right fragments on either side of the centromere , representing neighboring nucleosomes , align closely with MNase-seq nucleosomal profiles for all 16 centromeres , and show centromere-to-centromere variability as previously observed ( Cole et al . , 2011; Krassovsky et al . , 2012 ) . Fourth , each left end must pair with a right end , and the fact that for each left-end vertical there is a corresponding dense right-end diagonal and vice-versa implies that there are dense cleavages on either side of the centromere in addition to those expected from cleavages centered over neighboring nucleosomes , a point that we will return to in the next section . Finally , when log2 ( Cse4/Input ) ChIP-seq profiles are superimposed over the V-plots , we see that the midpoints of the vertical lines of dots marking the preferred cleavages are precisely centered over the position of maximum occupancy of the Cse4 nucleosome . To confirm that cleavage features we observed depended on the H4S47C mutation , we subjected wild-type cells to OP-labeling and cleavage reactions . The cleavage frequency in a wild-type strain is too low to obtain a library comparable to that of the H4S47C strain , which requires that two cleavages are sufficiently close to generate <500-bp fragments for paired-end sequencing . Therefore , purified DNA was cleaved to completion with AluI , which cleaves the sequence AG^CT and leaves blunt ends . The very low level of cleavages observed over centromeres ( Figure 6—figure supplement 2A ) confirms that centromere cleavage patterns are specific for H4S47C and not for centromere-specific DNA-binding proteins . As an additional control , we identified background cleavages by isolating DNA from untreated cells and digesting with AluI before library preparation . This revealed that the level of background cleavage over centromeres is similarly low regardless of whether or not cells were subjected to OP-labeling and cleavage reactions , as expected if centromeres are fully occupied and protected during the cleavage reaction ( Figure 6—figure supplement 2B ) . CenH3 nucleosomes are hypersensitive to MNase digestion in vivo ( Takahashi et al . , 1992; Dalal et al . , 2007; Krassovsky et al . , 2012 ) , and some authors have suggested that hemisomes might represent unstable intermediates in the assembly or disassembly of octasomes ( Black and Cleveland , 2011; Dunleavy et al . , 2013 ) . H4S47C-anchored cleavage mapping relies on nucleosomes remaining stably wrapped during the multistep nuclear labeling and washing procedures , which includes an overnight incubation with OP-labeling reagent that is performed prior to the actual Cu-dependent cleavage reaction . Therefore , our mapping of Cse4 hemisomes over centromeres at levels expected for such high AT-rich sequences ( Figure 3E ) implies that they are stable particles in vivo as we previously showed for hemisomes wrapped by CDEII in vitro ( Furuyama et al . , 2013 ) . Nevertheless , it remained formally possible that the hemisomes that we mapped at centromeres were generated by dissociation of Cse4 octasomes during the OP-labeling step . To test whether OP-labeling conditions promote the dissociation of octasomes , we reconstituted octasomes with either H3-H4 or Cse4-H4S47C and with either the 147-bp 601 positioning sequence or with a 147-bp Cen3-containing DNA segment . Using a gel-shift assay , we observed no changes in particle migration associated with the OP-labeling procedure for any of the reconstituted octasomes ( Figure 7A ) . To confirm that the particles undergoing a gel-shift remained octasomes during this procedure , we gel-purified reconstituted particles with or without the OP-labeling treatment and performed atomic force microscopy ( AFM ) . We found that OP-treated and untreated Cse4 and H3 octasomes are similar in height to one another but are ∼40–50% taller than Cse4-H4S47C/CDEII hemisomes reconstituted and incubated in parallel ( Figure 7B ) , consistent with previous studies ( Furuyama et al . , 2013; Codomo et al . , 2014; Walkiewicz et al . , 2014 ) . We conclude that the OP-labeling procedure does not alter the form of nucleosomes reconstituted with centromeric DNA . 10 . 7554/eLife . 01861 . 018Figure 7 . OP-labeling does not disrupt reconstituted or native nucleosomes . ( A ) Octasomes were reconstituted by salt dialysis on 147-bp DNAs , subjected to native PAGE ( left ) , and gel-shifted bands were excised ( dotted green lines ) and extracted as described ( Codomo et al . , 2014 ) . OP-treated and untreated gel-purified particles were subjected to native PAGE ( right ) . No instability of Cse4/Cen3 octamers ( Dechassa et al . , 2011 ) was observed during low-temperature incubation during storage at 4°C ( Xiao et al . , 2011; Furuyama et al . , 2013 ) . Using the intensity ratio of the gel-shifted band to the free DNA band as a measure of octasome stability , stability in the presence or absence of OP reagent was similar in each case ( Fraction +OP/-OP: H3/601: 0 . 8; H3/Cen3: 1 . 0; Cse4/Cen3: 1 . 1; Cse4/Cen3; 1 . 0 ) , based on the average of two determinations . ( B ) AFM analysis of reconstituted particles ± OP treatment . Three representative particles from each sample are shown on the left at the same magnification and dynamic range , where the height was set at 0 . 6 nm below the mean height of free DNA imaged in the same scan . The median heights for 1 min trypsinized OP-treated and untreated H3 and Cse4 octasomes and Cse4 hemisomes are similar to what we previously reported using the same protocol ( 2 . 10 nm for Cse4/Cen3 octasomes and 1 . 59 nm for Cse4/CDEII hemisomes [Codomo et al . , 2014] ) . ( C ) MNase-seq was performed on wildtype ( WT ) and H4S47C mutant cells ± OP labeling . The mid-CDE position of each of the 16 centromeres was aligned at zero on the X-axis position . A blue dot corresponds to the midpoint of each mapped fragment on the X-axis and the fragment length on the Y-axis . The sharp vertex located over the mid-Cen marks the X-axis position of the minimally protected fragment and its Y-axis position ( red line ) indicates its length . See also Figure 7—figure supplement 1–5 . DOI: http://dx . doi . org/10 . 7554/eLife . 01861 . 01810 . 7554/eLife . 01861 . 019Figure 7—figure supplement 1 . MIdpoint V-plot representations of MNase-seq generated fragments from insoluble chromatin over all 16 centromeres . See the legend to Figure 7C . The dotted line extensions of the diagonals show precise positioning of protected particles over the centromere , and the solid red lines mark the length of the minimally protected fragment . DOI: http://dx . doi . org/10 . 7554/eLife . 01861 . 01910 . 7554/eLife . 01861 . 020Figure 7—figure supplement 2 . MIdpoint V-plot representations of Cse4 ChIP fragments over all 16 centromeres . See the legends to Figure 7C and Figure 7—figure supplement 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 01861 . 02010 . 7554/eLife . 01861 . 021Figure 7—figure supplement 3 . The fragile nucleosome over the Gal1-10 UASg is not detected by H4S47C-anchored cleavage mapping . ( A ) MNase-seq normalized count data from stacked paired-end reads showing that the nucleosome over the UASg is present at high occupancy after 2 . 5 min MNase digestion , but is strongly depleted after 20 min digestion ( Bartfai et al . , 2010 ) . ( B ) Left-right V-plot of the same region showing H4S47C-anchored cleavages , where the dotted lines indicate the UASg . ( C ) Same as ( B ) , except that labeling and cleavage reactions were performed using a wild-type strain , and DNA was cleaved to completion with AluI after purification . Green arrows indicate the location of AluI sites . The high density of dots on the diagonal over the UASg confirms that cleavages seen over the UASg in the H4S47C-anchored strain reflect background cleavages in a nucleosome depleted region . ( D ) V-plot diagonals are not seen in DNA from untreated cells . DNA was purified directly from cells subjected to immediate alkaline lysis and cleaved with AluI prior to library construction . Points that fell between −2 and +2 bp relative to the AluI cleavage position were excluded for clarity as these most likely represent AluI cleavages and occasional imperfect end-polishing . Green arrows indicate the location of AluI sites . These background cleavages would appear to have occurred in vivo . DOI: http://dx . doi . org/10 . 7554/eLife . 01861 . 02110 . 7554/eLife . 01861 . 022Figure 7—figure supplement 4 . Fragile nucleosome are not detected by H4S47C-anchored cleavage mapping . See the legend to Figure 7—figure supplement 4 . ( A-F ) Examples from Chromosome 2 are shown in which nucleosome-sized fragments ( >140 bp ) displayed fragility as defined by a conspicuous drop in occupancy during MNase digestion relative to well-phased nucleosomes on either side ( Codomo et al . , 2014 ) . Unlike the Gal4 UAS , which has a high G+C content and so is inherently resistant to MNase digestion ( Chung et al . , 2010 ) , the G+C content of each fragile nucleosome is near average or low relative to the entire genome . DOI: http://dx . doi . org/10 . 7554/eLife . 01861 . 02210 . 7554/eLife . 01861 . 023Figure 7—figure supplement 5 . Nucleosome-depleted regions show preferential background cleavages . Density left-right V-plots for pooled left and right ends of H4S47C-anchored cleavages over aligned binding sites of Reb1 ( left ) and Abf1 ( right ) , transcription factors that strongly phase nucleosomes on either side ( Bartfai et al . , 2010 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01861 . 023 To confirm that OP labeling does not alter centromeric nucleosomes in vivo , we profiled wildtype and H4S47C cells genome-wide before and after the OP-labeling treatment . Using modified MNase-seq ( Henikoff et al . , 2011 ) , we determined that the position of the Cse4 nucleosome is the same in H4S47C and wild-type and in OP-treated and untreated cells ( Figure 7C ) . To confirm that centromeric chromatin is intact after OP-labeling , we also profiled MNase protection of insoluble chromatin ( Figure 7—figure supplement 1 ) , which is ∼100-fold enriched for centromeric chromatin ( Krassovsky et al . , 2012 ) , and performed ChIP-seq on the soluble fraction using a Cse4-specific antibody ( Figure 7—figure supplement 2 ) . In both cases , midpoint V-plots reveal that CDEI , CDEII and CDEIII are strongly MNase-protected relative to flanking regions in both strains with or without OP treatment in both soluble and insoluble chromatin fractions . This extent of Cse4 particle occupancy is half that necessary to accommodate two spaced hemisomes , because if the cleavage pairs were attributable to two hemisomes particles wrapping ∼60 bp with spacings of 33–53 bp between cleavage sites , then the total span would be 153–173 bp , or about twice the ∼80 bp that the Cse4-containing particle occupies . This leaves the single hemisome per centromere model as the only one that can account for our cleavage data . To further test this conclusion , we investigated the basis for cleavages around the centromere that do not correspond to neighboring nucleosomes . That is , the fact that there are fairly uniform diagonals extending from ∼50 bp up to >400 bp is not consistent with cleavages occurring only at the center of flanking nucleosomes , but rather suggests that these are caused by mostly background cleavages on either side of the centromere . As the hydroxy radical cleavage protocol that we use was originally developed for mapping non-specific cleavages within linker DNA ( Cartwright and Elgin , 1982 ) , these diagonals indicate a moderate level of background cleavage between flanking highly occupied and phased nucleosomes . We also observed background cleavages at sites of ‘fragile’ nucleosomes , which are thought to be unstable based on hypersensitivity to MNase ( Xi et al . , 2011 ) . The best-studied fragile nucleosome is over the Gal4 UAS in the Gal1-10 regulatory region , when yeast are grown in glucose ( Floer et al . , 2010 ) ( Figure 7—figure supplement 3A ) . We therefore expected to observe high frequency cleavage by this nucleosome in our cleavage datasets . However , there was no evidence of nucleosome-directed cleavage , but rather especially strong cleavage diagonals directly over the Gal4 UAS ( UASg , Figure 7—figure supplement 3B ) . We observed similarly strong Xs over other fragile nucleosomes ( Figure 7—figure supplement 4 ) . The ‘X’ pattern of depletion seen at these sites resembles that for known nucleosome depleted regions , such as those around binding sites for the Reb1 and Abf1 transcription factors ( Figure 7—figure supplement 5 ) . We further confirmed this interpretation by mapping AluI-generated fragments of DNA in the Gal1-Gal10 region from wildtype cells subjected to OP-labeling and cleavage reactions . We observed a strong diagonal representing the left end of fragments cleaved by AluI at a site on the right side of the Gal4 UAS , but only weakly extending beyond the edge of the well-occupied next nucleosome to the left ( Figure 7—figure supplement 3C , D ) . In contrast , centromeres showed very little cleavage relative to regions on either side ( Figure 6—figure supplement 2A ) , indicating protection from background cleavages and confirming that Cse4 nucleosome-directed cleavages depend on the H4S47C mutation . The stability of the Cse4 centromeric nucleosome and protection from background cleavages under these conditions confirm that Cse4 hemisomes are immobile at centromeres , as expected from the stability of Cse4 hemisomes on short DNA fragments , including a fragment consisting of CDEII alone ( Figure 7B; Furuyama et al . , 2013 ) .
We have used directed chemical cleavage mapping to determine the precise position of the single Cse4 nucleosome at each budding yeast centromere . Using molecular dynamics simulation of H4S47C-anchored cleavages , and verification of the resulting model based on cleavage patterns at highly occupied and phased nucleosomes , we find that the pattern over all 16 centromeres , and the distance between cleavage sites , is profoundly different from what is seen elsewhere in the genome . As octasome and tetrasome structures for the Cse4 nucleosome position two H4S47 residues immediately on either side of the dyad axis , just as for conventional H3 octasomes and tetrasomes , our cleavage data rule out these structures as making a substantial contribution to Cse4 occupancy at centromeres ( Figure 8 ) . Indeed , the only proposed structure that fits our data is a hemisome in one of four mutually exclusive orientations occupying each centromere within the population of cells . Thus , the Cse4 nucleosome shows remarkable rotational flexibility within the confines of the 78–86 bp CDEII sequence , where it can assume opposite orientations and two preferred rotational phasing patterns . Such flexibility might be important for brief hemisome–octasome transitions reported for both budding yeast and human cenH3s ( Bui et al . , 2012; Shivaraju et al . , 2012 ) . 10 . 7554/eLife . 01861 . 024Figure 8 . Preferred cleavage positions for proposed structural models of the CDEII nucleosome . ( A ) H4S47C-anchored cleavage reactions within an octasome ( left ) or an ( H3/H4 ) 2 tetrasome ( middle ) centered over CDEII would give rise to mapped sites that are ∼5 bp apart ( right ) . ( B ) Observed cleavage positions ( right ) are explained by a structural model of a hemisome occupying either of two rotational positions ( compare left vs middle models ) in either of two reflectional orientations ( compare upper vs lower models ) . The ∼7 bp closer spacing of Cen4 cleavage pairs ( Figure 4—figure supplement 2 ) implies that positioning is determined by the distance of H4S47C from the closer end of CDEII . Histone H4 is indicated in orange and preferred cleavage sites are indicated in green . DOI: http://dx . doi . org/10 . 7554/eLife . 01861 . 024 The flexibility in orientation and phasing that we observed for such a precisely positioned particle is especially remarkable considering that yeast centromeres are directional at the DNA level , with the >90% AT-rich CDEII sequence oriented between Cbf1-bound CDEI and CBF3-bound CDEIII . We attribute this rotational and reflectional flexibility to the fact that a 78-bp CDEII duplex is sufficient to stably wrap a hemisome in vitro ( Furuyama et al . , 2013 ) , and that AT-richness alone is sufficient for full in vivo function of CDEII , as some random 92% AT-rich sequences substituted for natural CDEII supported faithful segregation of mini-chromosomes ( Baker and Rogers , 2005 ) . This interpretation helps to clarify the roles of Cbf1 and the CBF3 complex in maintaining centromeres . Both proteins tightly bend DNA and thus delimit an ∼80-bp sequence for assembly of the Cse4 nucleosome ( Henikoff and Furuyama , 2012 ) . CBF3 is essential for recruiting Cse4 , where the Ndc10 subunit interacts with the Scm3 Cse4-specific chaperone ( Shivaraju et al . , 2011 ) . In addition , our findings suggest that Cbf1 acts as a barrier on the left that sets the sharp rotational positions of the Cse4 nucleosomes that are asymmetrically oriented with H4S47 closest to CDEI . This implied role of Cbf1 in Cse4 hemisome positioning is consistent with our previous analysis of MNase-seq data in a cbf1Δ mutant strain showing that loss of Cbf1 is accompanied by a shift in position of the Cse4 nucleosome ( Krassovsky et al . , 2012 ) . The rotational and reflectional flexibility of the Cse4 nucleosome implied by our study argues against specific roles for either Cbf1 or CBF3 in orienting the histone core within the single DNA wrap ( Xiao et al . , 2011 ) . Although budding yeast are especially suitable for application of directed chemical cleavage mapping , because replacing endogenous histones with the mutant H4S47C version is straightforward , one might envision applying this strategy to complex genomes . Complete replacement of the replication-coupled histone gene cluster has been accomplished in Drosophila ( Gunesdogan et al . , 2010 ) , and conceivably other strategies for introducing cysteine-substituted histones might be considered for other model organisms . The application of this technology to the study of cenH3 nucleosomes more generally has the potential of resolving current debates as to the generality of hemisomes vs octasomes at centromeres . Current evidence suggests that both types of particles are present in animal genomes ( Bui et al . , 2012 ) , although which are at functional centromeres and which are on chromosome arms has been unclear . MNase-seq and ChIP-seq have been applied to the problem , but particle size has been ambiguous because of the inherent sensitivity of hemisomes to MNase digestion ( Dalal et al . , 2007; Krassovsky et al . , 2012 ) and ambiguities caused by partial unwrapping of octasomes ( Tachiwana et al . , 2011 ) and internal cleavages and nibbling ( Hasson et al . , 2013 ) . Together with high-resolution native ChIP , directed cleavage mapping has the potential of unambiguously determining particle composition and structure in complex genomes .
We constructed a molecular model of cysteine-orthophenanthroline ( OP ) using Avogadro ( Hanwell et al . , 2012 ) . We replaced the serine at position 47 of H4 in the yeast nucleosome structure ( PDB ID:1ID3 ) with cysteine-OP . Our simulation system consisted of H3-H4 ( 47C-OP ) bound to 10 bp of DNA in the same conformation as in the nucleosome structure ( corresponding to −2 to −12 positions with respect to the dyad ) . The DNA , H3 and H4 were held static , except for amino acid residues that were within 10 Å of H4S47C-OP . The simulation system was minimized and ∼30 , 000 conformations of H4S47C-OP were generated using Chiron ( Ramachandran et al . , 2011 ) . Yeast strain Sby3 ( MATa bar1-1 ura3-1 leu2 , 3-112 his3-11 trp1-1 can1-100 ade2-1 ) was used as control cells for cleavage mapping , MNase-seq and ChIP as in our previous centromeric chromatin analysis ( Krassovsky et al . , 2012 ) . Construction of H4S47C from BY4741 ( MATa his3Δ1 leu2Δ0 met15Δ0 ura3Δ0 ) was previously described ( Brogaard et al . , 2012b ) . The two histone H4 genes ( HHF1 and HHF2 ) had been deleted and replaced by a single HHF1 gene with a cysteine codon replacing the serine codon at position 47 . Growth and viability measurements were performed using a Vi-Cell automated cell counter ( Beckman–Coulter , Brea , CA ) . Plasmid loss assays were performed as described by Koshland et al . ( 1987 ) . As previously reported ( Brogaard et al . , 2012a ) , H4S47C cells show growth anomalies relative to BY4741 , which we have determined are caused by failure to achieve a quiescent state upon reaching stationary phase ( Figure 2—figure supplement 1 ) . This failure is likely attributable to deletion of HHF2 , which results in reduced starvation resistance ( Davey et al . , 2012 ) and is associated with reduced lifespan ( Feser et al . , 2010 ) . To map H4S47C-anchored cleavages within a reconstituted H3 octasome , octamers were prepared according to Luger et al . ( 1999 ) and OP-labeled by addition of a 10-fold molar excess of tris ( 2-carboxyethyl ) phosphine ( TCEP ) for 10 min , then a 30-fold molar excess of OP reagent in DMSO in the dark . Samples were incubated at room temperature 2 hr and at 4°C overnight , quenched with 1/350 1 . 4M β-mercaptoethanol , and excess OP-reagent removed with a Bio-Spin P-30 column ( Bio-rad , Hercules , CA ) . Reconstitutions were performed with 601 DNA according to Luger et al . ( 1999 ) . Cleavage reactions proceded by addition of 1 vol 5 mM NaCl , 100 mM Tris pH 7 . 5 , 300 μM CuCl2 , mercaptopropionic acid to 6 mM , and H2O2 to 6 mM . After 20 min at room temperature , reactions were quenched with neocuproine in DMSO to 2 . 5 mM , and DNA was extracted using standard methods . SOLID sequencing was performed on the resulting fragments as previously described ( Brogaard et al . , 2012b ) . To map H4S47C-anchored cleavages within Cse4 hemisomes , a 78-bp oligonucleotide with the sequence shown in Figure 4—figure supplement 1 , that was 5′ end-labeled with 5′-Cy3 and 3′ end-labeled with 3′-AmC7-Q+Alexa488 , was annealed with its unlabeled complementary unlabeled oligonucleotide . This 5′-Cy3-CDEII-Alexa488-3′ duplex was mixed with a fourfold excess of unlabeled CDEII duplex and used for octasome reconstitution and hemisome splitting To prepare OP-labeled hemisomes , Cse4/H4S47C/H2A/H2B octamers were lightly trypsinized as described ( Furuyama et al . , 2013 ) then subjected to in vitro OP-labeling and P-30 clean-up as described above , followed by reconstitution of pseudo-octasomes with CDEII DNA in 2M NaCl ( Furuyama et al . , 2013 ) . After dialysis vs 4M urea to split pseudo-octasomes into two hemisomes , samples were electrophoresed on a 7% Tris-acetate ( no EDTA ) PAGE gel and bands were excised and eluted as described ( Codomo et al . , 2014 ) , then subjected to in vitro cleavage reactions as described above . Reaction products from ( 1 ) the shifted band , which showed a strong gelFRET signal indicative of tight wrapping around a hemisome , ( 2 ) the unshifted ( control ) band and ( 3 ) G+A and C+T Maxam-Gilbert ladders produced from the 5′-Cy3-CDEII-Alexa488-3′ labeled oligonucleotide duplex , were electrophoresed on a 15% sequencing gel ( Sambrook et al . , 1989 ) . The gel was scanned for Alexa488 and Cy3 using a Typhoon Trio and images were processed using ImageJ . To ascertain the stability of reconstituted Cse4 octasomes and hemisomes , octamers were prepared as described ( Furuyama et al . , 2013 ) and reconstituted with 601 , Cen3 or CDEII DNA by salt dialysis at 37°C as described by Xiao et al . ( 2011 ) . Mutant and wildtype S . cerevisiae strains were grown to log phase at 30°C in YPD medium , harvested and used for labeling and cleavage reactions as described ( Brogaard et al . , 2012a ) . Briefly , the cell pellet was resuspended in spheroplasting buffer in 1M sorbitol , partially spheroplasted with lyticase , washed with 1M sorbitol 0 . 1% NP-40 and brought up in labeling buffer ( 1M sorbitol , 50 mM NaCl , 10 mM Tris–HCl pH 7 . 5 , 5 mM MgCl2 , 0 . 5 mM spermidine , 0 . 15 mM spermine , 0 . 1% NP-40 and 0 . 1 mM EDTA ) . Labeling was performed by addition of 7 mM OP reagent in DMSO to 20% volume and incubation for 2 hr at room temperature followed by overnight incubation at 4 °C . Cell pellets were washed with sorbitol/NP-40 , incubated with CuCl2 , washed with a sorbitol/NP-40-containing buffer , subjected to oxidative cleavage at room temperature for 20 min using 6 mM H2O2 and 6 mM 2-methylpropionic acid , and the reaction was quenched by addition of neocuproine to 0 . 28 mM . We varied the protocol in several ways , yet obtained comparable results ( Figure 2A ) . In some experiments 15–30 min cleavage reactions were performed 3–5 times by resuspending the cell pellet in the original volume of mapping buffer , repeating addition of methylpropionic acid and hydrogen peroxide followed by re-centrifugation . To further reduce background , in some experiments we reduced the concentration of lyticase 1:10 , the concentration of OP reagent 1:10 , and/or increased the concentration of NP-40 twofold . In some experiments we also used a modified DNA extraction protocol ( Zentner et al . , 2013 ) , repeating the RNAse A digestion , phenol-chloroform/chloroform extraction and ethanol precipitation to remove residual RNA before Illumina Tru-Seq paired-end library preparation as described ( Zentner et al . , 2013 ) . In some experiments KAPA HiFi DNA polymerase ( KAPABiosystems , Woburn , MA ) was used in place of Phusion Polymerase . AluI ( New England Biolabs , Ipswitch , MA ) cleavage reactions were performed according to the manufacturer’s instructions on wildtype DNA either subjected to mock hydroxy radical cleavage or directly purified from cells using the Epicentre ( Madison , WI ) MasterPure Yeast DNA Purification Kit . Wild-type and H4S47C cells were cultured , harvested , lyticased and washed twice in 1M sorbitol 0 . 1% NP-40 as described for cleavage mapping ( Brogaard et al . , 2012a ) . Samples derived from 500 ml of cells at 2 × 107/ml were split in half , and one half of each was washed with 1M sorbitol 0 . 1% NP-40 for OP labeling , while the other other half was suspended in 4 ml fresh MNase buffer ( 1 M sorbitol; 10 mM Tris-HCl , pH 7 . 5; 50 mM NaCl; 5 mM MgCl2; 2 mM CaCl2; 1 mM β-mercaptoethanol , 1 mM phenylmethanesulfonyl fluoride , + 1 protease inhibitor tablet [Roche , Nutley , NJ #04693159001] per 10 ml ) . Lightly lyticase-treated cells were digested with MNase for 10 min at 37°C using 4U MNase ( Sigma–Aldrich , St . Louis , MO ) . OP labeling ( 2 hr at room temperature and overnight incubation at 4°C ) was followed by three 1M sorbitol 0 . 1% NP-40 washes , then a sample of OP-treated cells was suspended in fresh MNase buffer and digested with MNase . MNase reactions were stopped by addition of EDTA to 10 mM and for MNase-seq , DNA was extracted . For soluble and insoluble chromatin isolation and ChIP , the remaining samples were snap-frozen for storage , thawed on ice , and soluble chromatin was extracted and ChIP was performed as described ( Krassovsky et al . , 2012 ) . An anti-Cse4 rabbit antibody was obtained from Sue Biggins and used at 5 μl per sample . Glycogen was added to ChIP samples before ethanol precipitation and Solexa library preparation . MNase fully penetrates into zymolyase-generated spheroplasts used for MNase-seq and ChIP , but hardly at all into the mildly lyticased cells used to allow penetration by OP-reagent for chemical cleavage . This resulted in >80% excess full-length DNA in the chromatin samples used for MNase-seq , which was removed prior to Solexa library preparation using a PCR cleanup kit ( Clontech , Mountain View , CA ) . Illumina Tru-Seq libraries were prepared as described ( Zentner et al . , 2013 ) . Paired-end sequencing data were processed and aligned with Novoalign ( Novocraft; http://www . novocraft . com ) as described ( Henikoff et al . , 2011; Krassovsky et al . , 2012 ) to Version 64 of the genome build from the Saccharomyces Genomic Database ( SGD , http://www . yeastgenome . org ) , which corresponds to UC Santa Cruz SacCer3 . Centromere CDEI-II-III coordinates are from SGD . Midpoint V-plots were constructed as described ( Krassovsky et al . , 2012 ) . Fragments with midpoints within ± 1 kb of mid-centromeres were randomly sampled and plotted to equalize coverage to the least populated sample . For analysis of cleavage positions , separate V-plots of each left and right fragment end rather than the fragment mid-point were used . For H4S47C-anchored cleavage experiments two biological replicates ( using KAPA DNA polymerase for library amplification ) and one technical replicate ( using Phusion DNA polymerase ) gave virtually identical results and so fragment counts were combined for data analysis and presentation . A profile of 15 centromeres ( length 117–120 bp ) comprising 111 base-pair positions was constructed by aligning at the left-most peak , excluding Cen4 ( length 111 bp ) . The mean , standard deviation and minimum and maximum values of the centromeres at each position were computed . Alignments of the profile along the chromosome were filtered as follows: ( 1 ) The absolute value of z-score at each position was allowed to be ≥3 at no more than 3/111 positions . ( 2 ) The alignment was required to contain at least one position ≥ the smallest maximum of the profile , which was 186 ( for Cen8 ) . The Pearson correlation coefficient was computed for alignments passing the two filters . The same two filters were used for the delete-one jackknife , which in at least some cases resulted in exclusion of the deleted centromere prior to alignment . W-C distributions were generated by determining for the left end of each fragment the number of fragments whose right ends were at a given distance . This analysis was performed for a distance range of −40 to +40 and normalized by total number of left-right combinations within the same distance range . W–W’ distributions were generated by determining for the left end of each sequencing fragment the number of sequencing fragments whose left ends were at a distance in the range of 1–40 and normalized by the total number of left–left combinations within the same range . C–C’ distributions were generated similarly for right ends . We reconstituted H3/H4 and Cse4/H4S47C ( Cse4 ) octasomes with 147-bp duplexes of either the 601 positioning sequence or Cen3 following a 1 min room temperature trypsinization to minimize aggregation , as previously described ( Codomo et al . , 2014 ) . After dialysis from 2M NaCl to 0 . 25 mM HEPES pH7 . 5 and 6% native PAGE , we excised the gel-shifted bands and extracted the nucleosomes as described ( Codomo et al . , 2014 ) , except that extraction was done in OP labeling buffer rather than in water . An aliquot of each sample was incubated at 4°C for 15 hr with ¼ volume of OP-reagent to mimic the in vivo labeling procedure , and samples with or without treatment were resolved by 6% native PAGE . For AFM , reconstituted H3/Cen3 and Cse4/Cen3 octasomes and Cse4/CDEII ( 78-bp ) hemisomes in 0 . 25 mM HEPES were dialyzed vs OP-labeling buffer in a 5-step gradient , and one-half was incubated at 4°C for ∼15 hr before gel purification . Bands were excised and particles were extracted into water and confirmed to be intact by native PAGE . Eluted samples were cross-linked in 0 . 6% glutaraldehyde for 25′ and imaged as described ( Codomo et al . , 2014 ) . Sequencing data generated in this publication have been deposited with GEO ( GSE51949 , Henikoff et al . , 2013 ) . | DNA is tightly packaged in cells for a variety of reasons—to allow it to fit inside the nucleus , to protect it from damage , and to help control the production of proteins from genes . The basic unit of packaged DNA is called a nucleosome , which consists of DNA wrapped around a structure formed by two pairs of four different proteins . These proteins , which are called histones , have a role that extends beyond providing structural support for DNA . When cells divide , for example , pairs of ‘sister chromosomes’ are pulled apart to ensure that the two daughter cells both have the same chromosomes as the original cell . The sister chromosomes are pulled apart from a single position called a centromere , and the nucleosomes at this position contain a histone that is different from the histones found everywhere else in the cell . However , until recently it was not clear if the nucleosomes that contained these special cenH3 histones had the same structure as other nucleosomes . Now Henikoff et al . have used a method called H4S47C-anchored cleavage mapping to study every nucleosome in the genome of the yeast S . cerevisiae . This mapping technique uses DNA sequencing to measure the precise distances between fixed points on the DNA in the nucleosome . Knowing these distances tells researchers a great deal about the number and position of the histones within each nucleosome in the genome . Using this approach , Henikoff et al . found that nucleosomes at centromeres are different from other nucleosomes in histone number and arrangement . In particular , the nucleosome at each yeast centromere contains only one each of the four different histones in an asymmetrical orientation , in contrast to all other yeast nucleosomes , which contain two sets of four histones in a symmetrical arrangement . Furthermore , each nucleosome at a centromere can adopt one of two orientations: these orientations are mirror images of each other , and they occur with equal probability . It should also be possible to use the mapping technique developed by Henikoff et al . to study the larger and more complex centromeres found in other organisms , including humans . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"chromosomes",
"and",
"gene",
"expression"
] | 2014 | The budding yeast Centromere DNA Element II wraps a stable Cse4 hemisome in either orientation in vivo |
In the early days of HIV treatment , drug resistance occurred rapidly and predictably in all patients , but under modern treatments , resistance arises slowly , if at all . The probability of resistance should be controlled by the rate of generation of resistance mutations . If many adaptive mutations arise simultaneously , then adaptation proceeds by soft selective sweeps in which multiple adaptive mutations spread concomitantly , but if adaptive mutations occur rarely in the population , then a single adaptive mutation should spread alone in a hard selective sweep . Here , we use 6717 HIV-1 consensus sequences from patients treated with first-line therapies between 1989 and 2013 to confirm that the transition from fast to slow evolution of drug resistance was indeed accompanied with the expected transition from soft to hard selective sweeps . This suggests more generally that evolution proceeds via hard sweeps if resistance is unlikely and via soft sweeps if it is likely .
In the first two decades of the HIV epidemic , HIV became a prime example of fast evolutionary change , especially because of the evolution of drug resistance quickly after initiation of treatment . Nowadays , HIV treatments are more clinically effective and the evolution of drug resistance has become much slower and often does not occur for years if at all . The rate at which evolution occurs has been the subject of considerable recent interest in the evolutionary biology community . Although traditionally evolution was thought to be slow ( Darwin , 1859 ) , there are a growing number of examples of fast evolution to selective pressures such as pesticides ( Lopes et al . , 2008; Daborn et al . , 2001; Karasov et al . , 2010; Palumbi , 2001 ) , industrialization ( Cook et al . , 2012 ) , or antibiotics ( Laehnemann et al . , 2014; Nair et al . , 2007 ) . HIV represents an interesting case because its evolutionary speed in treated patients has changed drastically over time . Population genetic theory suggests that whether populations evolve slowly or quickly is driven by the availability of adaptive mutations . In a large population with a high mutation rate , mutations may be available as standing genetic variation ( pre-existing variation ) or be generated anew every generation , allowing the population to adapt to its environment rapidly . If adaptive mutations are rare , because the population is small , the mutation rate is low , or only few specific mutations ( or combinations of mutations ) can help a population adapt , the population will likely adapt to its environment much more slowly . The availability of adaptive mutations does not only change the rate of adaptation , it also changes how adaptation affects genetic diversity in a population . If adaptive mutations are rare , i . e . , less than one adaptive mutation occurs per generation in the population , the first successful mutation is likely to rise to high frequency before any subsequent adaptive mutations reach appreciable frequencies ( see Figure 1A ) . This results in a hard selective sweep , in which the single adaptive mutation and the nearby linked mutations becomes fixed in the population ( Figure 1B ) . Hard selective sweeps sharply reduce genetic diversity in the population ( Figure 1C ) ( Smith and Haigh , 1974; Kaplan et al . , 1989 ) in a similar manner to a strong genetic bottleneck . 10 . 7554/eLife . 10670 . 003Figure 1 . Prediction of drug resistance acquisition with more and less effective treatments . Among patients treated with more effective treatments ( top ) , we predict HIV populations to have a lower probability of acquiring resistance per generation . As a result , the population must wait a long time for a beneficial genotype , so when resistance does occur , it will spread through the population in a hard selective sweep before other resistant genotypes emerge ( A ) . Since resistance only occurs on a single genetic background ( background mutations in grey ) , all sequences with resistance will be similar ( B ) and diversity following this type of selective sweep will be reduced ( C ) . We can use the reduction of diversity to determine that a selective sweep is hard . In patients treated with less effective treatments ( bottom ) , we predict HIV populations should have a higher probability of acquiring resistance per generation , so resistance will be acquired more quickly and selective sweeps of drug resistance mutations will be soft ( D ) . We can detect these soft selective sweeps , because diversity remains high when resistance mutations on different genetic backgrounds rise in frequency simultaneously ( E , F ) . DOI: http://dx . doi . org/10 . 7554/eLife . 10670 . 003 In contrast , when adaptive mutations are common , i . e . , more than one occurs per generation in the population , the same adaptive mutation may occur several times in a very short time span on different genetic backgrounds . These adaptive mutations can increase in frequency virtually simultaneously ( Figure 1D ) ( Pennings and Hermisson , 2006 ) and multiple genetic backgrounds are therefore expected to reach substantial frequencies with no single genetic background dominating the population ( Figure 1E ) . This pattern is known as a soft selective sweep and is expected to lead to almost no reduction of genetic diversity ( Figure 1F ) ( Pennings and Hermisson , 2006 ) , comparable to a mild bottleneck . In HIV , the evolution of drug resistance was fast in patients on early anti-retroviral therapies ( Larder et al . , 1989 ) , but current multi-drug regimens have substantially slowed the rate of evolution of resistance ( Martin et al . , 2008 ) . Although clinically effective drugs have decreased the rate of emergence of drug resistance , it is not clear what effect that they have had on the evolutionary dynamics of within-patient HIV populations . Specifically , population genetic theory predicts that populations should evolve primarily by soft sweeps when resistance is likely and by hard sweeps when resistance is rare . In essence , soft sweeps should mark cases of resistance that arise deterministically through many origins while hard sweeps mark cases of rare , 'unlucky' resistance . However , these predictions have not been tested in HIV or , in fact , in any natural population . In this study , we ask whether the transition from fast to slow evolution of drug resistance was indeed associated with a transition from soft to hard selective sweeps . If true , then in general , treatments in which resistance arises by soft sweeps might be predicted to have high rates of failure even before the rates of failure can be measured explicitly . To test whether a transition from soft to hard sweeps has occurred , we look at the relationship between fixed drug resistance mutations ( DRMs ) and genetic diversity across 29 common anti-retroviral drug regimens . The expectation is that when hard selective sweeps predominate , we will find a negative correlation between the number of DRMs and genetic diversity in a population . On the other hand , when soft selective sweeps predominate , we expect to find no such correlation . We use 6717 HIV sequences from the same number of patients from the Stanford HIV Drug Resistance Database ( Rhee et al . , 2003 , https://hivdb . stanford . edu/ ) . These sequences contain information about the number of DRMs and , as we will further explain in the next paragraph , they also contain information about genetic diversity in the viral population . Most sequencing of HIV populations in patients is done with the intent to discover DRMs for diagnostic and therapeutic reasons ( Dunn et al . , 2011 ) . As such , in standard clinical practice , a sample from a patient’s entire HIV population is amplified via PCR and then sequenced using the traditional Sanger method resulting in a single consensus sequence . Genetic diversity may result in ambiguous calls ( also referred to as mixtures ) in the reported sequence , so that a signal of within-patient genetic diversity is retained even though this sequencing approach generates only a single sequence per patient . We use the ambiguous calls to quantify within-patient genetic diversity ( see Figure 1B , E , grey box ) , following several other studies ( Kouyos et al . , 2011; Zheng et al . , 2013; Li et al . , 2012; Poon et al . , 2007 ) . Although ambiguous calls are an imperfect measure of diversity , it has been shown that the signal from ambiguous calls can be reproduced between laboratories ( Shafer et al . , 2001 ) . By using ambiguous calls as a proxy for diversity , we are able to take advantage of a large number of HIV-1 sequences , allowing us to study the evolutionary dynamics of HIV drug resistance evolution in a historical perspective ( Rhee et al . , 2003 ) . Through examining HIV sequences of 6717 patients over the past two and a half decades in the presence of many different drug regimens , all sequenced using Sanger sequencing technology , we leverage ambiguous sequence calls to understand how the fixation of drug resistance mutations affects diversity . We find that , across all sequences , the presence of drug resistance mutations is associated with lower within-patient genetic diversity , marking the occurrence of selective sweeps . Second , we find that the extent of diversity reduction associated with drug resistance mutations varies with the clinical effectiveness of the treatment - effective drug treatments with low rates of virologic failure ( such as NNRTI-based and boosted PI-based regimens ) show strong reductions in diversity associated with each additional resistance mutation , a pattern more consistent with hard selective sweeps , whereas treatments that fail more often ( such as regimens based only on NRTIs ) show no reduction in diversity , a pattern consistent with soft selective sweeps . Although our results do not explain mechanistically how effective treatments lead to harder sweeps of drug resistance mutations , they suggest a more general principle: a lower rate of the production of adaptive mutations should be accompanied by harder sweeps .
We collected sequences of reverse transcriptase and/or protease genes from 6717 patients from the Stanford HIV Drug Resistance Database ( Rhee et al . , 2003 ) . The sequences come from 120 different studies that were performed between 1989 and 2013 . The 6717 patients represent all individuals in the database who were treated with exactly one drug regimen , usually comprising a combination therapy of multiple drugs ( see Materials and methods: Data collection & filtering ) . The patients’ viral populations were sequenced after at least some period of treatment , although treatment may or may not have ceased at the time of sequencing and treatment may or may not have failed . This virus from a patient was amplified via PCR , sequenced using the Sanger method and then was reported to the database as a single nucleotide sequence . We call this dataset the D-PCR dataset , for direct PCR . All 6717 patients received some type of therapy ( between 1 and 4 drugs ) , with the majority ( 77% ) receiving a regimen of three drugs . Nearly all patients received one or two nucleoside reverse transcriptase inhibitors ( NRTI ) , usually paired with a non-nucleoside reverse transcriptase inhibitor ( NNRTI ) or a protease inhibitor ( PI ) , which was boosted with a low dose of ritonavir in some patients . HIV subtypes were varied , with the majority being B ( 36% ) , C ( 34% ) or CRF01_AE ( 13% ) . None of the remaining subtypes contributed more than 5% of the total sample . An additional dataset , which we call the clonal dataset , consisted of 11 , 653 sequences from 740 patients with multiple sequences per patient isolated through clonal amplification and subsequent Sanger sequencing . The clonal dataset was used for validation purposes only . We are interested in the effect of drug resistance evolution on within-patient genetic diversity , but in our main dataset , we only have one sequence per patient . To use this large dataset for our purposes , we therefore use ambiguous nucleotide calls as a proxy for within-patient genetic diversity . Although results from previous studies suggest that this approach is valid ( Kouyos et al . , 2011; Shafer et al . , 2001 ) , we independently validate this measure through comparing the D-PCR and clonal datasets . Using the clonal sequences , within-patient diversity ( π ) can be computed directly , giving an estimate of genetic diversity per site that does not rely on ambiguous calls . We compared the proportion of sequences with ambiguous calls at a site in the D-PCR dataset to the within-patient diversity ( π ) at that site in the clonal dataset ( see Materials and methods: Validation of ambiguous calls as genetic diversity measure ) . We find that clonal within-host nucleotide diversity has a high positive correlation with the percentage of nucleotide calls ambiguous in the D-PCR dataset ( r = 0 . 91 , p<2 . 2 × 10−16 ) . A similar pattern holds at the amino acid level ( r = 0 . 85 , p<2 . 2 × 10−16 ) . We therefore conclude that ambiguous calls are a reliable proxy for within-patient genetic diversity . We ask whether across all sequences , the presence of a drug resistance mutation ( DRM ) is associated with lower within-patient diversity , the classical signature of a selective sweep ( Smith and Haigh , 1974 ) . For each sequence , we therefore count the number of DRMs present that are relevant for the treatment the patient was taking ( Bennett et al . , 2009 ) ( i . e . , a mutation that confers resistance to a particular class of drugs was counted as a DRM only if the patient was actually being treated with that drug; see Materials and methods: Sequence processing for more information ) . First , for the most common reverse transcriptase and protease DRMs , we compare sequences that have exactly one DRM with sequences that have zero DRMs ( i . e . , ancestral state at all possible DRM sites ) . We plot the difference in within-patient diversity between the two groups with 95% confidence intervals in Figure 2A . Among reverse transcriptase and protease DRMs , sequences with the DRM have lower diversity than those with the ancestral state in 14 of 16 cases , with 7 of the 16 being significantly lower at the 95% confidence level . This reduction in diversity is consistent with expectations after a selective sweep ( Smith and Haigh , 1974 ) . 10 . 7554/eLife . 10670 . 004Figure 2 . Effect of DRMs on sequence diversity . ( A ) For the most common reverse transcriptase and protease mutations , 95% confidence intervals are drawn for the difference in diversity associated with a single derived mutation . For each DRM , the mean diversity among patients with a fixed ancestral state at the focal locus is compared to those patients with that fixed non-ambiguous DRM . All sequences have no additional DRMs . All DRMs occurring at least 5 times with these specifications are included . ( B ) The effect of multiple DRMs on diversity is shown as the average diversity level of sequences decreases conditional on number of fixed drug resistance mutations present . Means ± SE are plotted among all patients in the D-PCR dataset . DOI: http://dx . doi . org/10 . 7554/eLife . 10670 . 00410 . 7554/eLife . 10670 . 005Figure 2—figure supplement 1 . Diversity and the number of drug resistance mutations by treatment categories . The relationship between diversity and the number of drug resistance mutations is plotted separated out by the major treatment categories included in our analysis: ( A ) NRTI ( s ) without any other drugs , ( B ) NRTIs with unboosted PI , ( C ) NRTIs and an NNRTI and ( D ) NRTIs and PI/r . Data plotted is just the abundant treatment dataset . DOI: http://dx . doi . org/10 . 7554/eLife . 10670 . 00510 . 7554/eLife . 10670 . 006Figure 2—figure supplement 2 . Effect of multiple DRMs on sequence diversity separated by subtype . Average diversity level of sequences are plotted conditioned on number of fixed drug resistance mutations present separately by all the subtypes with more than 100 associated HIV populations . Means ± SE are plotted among all the D-PCR dataset . DOI: http://dx . doi . org/10 . 7554/eLife . 10670 . 006 Second , we looked at the effect of multiple DRMs on within-patient diversity , as we hypothesize that multiple fixed DRMs may decrease diversity even more than a single DRM . This could result from sequential selective sweeps of single DRMs each reducing diversity , or from a single selective sweep that fixes multiple DRMs . Indeed , we find that for sequences that have between 0 and 4 DRMs , additional DRMs are associated with reduced genetic diversity ( Figure 2B , p-value for t-test between diversity among sequences with 0 versus 1 is 7 . 2 × 10−4 , between 1 and 2 is 1 . 6 × 10−5 , between 2 and 3 is 1 . 1× 10−2 , between 3 and 4 is 2 . 5 × 10−3 ) . After 4 DRMs , subsequent DRMs do not significantly reduce diversity further . The observed pattern of DRMs associated with reduced diversity is mainly driven by the patients receiving NNRTI or boosted PI-based treatments , as can be seen when separating the above analysis by drug treatment category ( Figure 2—figure supplement 1C , D ) . Among patients treated with NRTIs alone or with unboosted PIs , this pattern is much less clear ( Figure 2—figure supplement 1A , B ) . The observed pattern holds across the each of the most common subtypes separately ( Figure 2—figure supplement 2 ) . We have now shown that in general , each additional DRM is associated with reduced diversity , which is consistent with expectations of selective sweeps . We want to assess how this effect depends on the clinical effectiveness of the treatment . For the most common drugs in our dataset , we assess clinical drug treatment effectiveness categorically and quantitatively . As a categorical approach , we separated regimens based on general clinical HIV-treatment recommendations where NNRTI-based treatments are preferred to NRTI-based treatments , and treatments based on ritonavir-boosted PIs ( PI/r ) are preferred to treatments based on unboosted PIs . These more and less effective groupings are the basis of comparisons in our parametric approach described below . To measure effectiveness quantitatively , we conducted a literature search to determine the percentage of patients who have remained virologically suppressed after one year of treatment ( see Materials and methods: Clinical effectiveness of antiretroviral treatments , Supplementary files 1–3 ) for 21 different treatments with at least 50 sequences per treatment in our D-PCR dataset ( see description of abundant treatment dataset in Materials and methods: Data collection & filtering for more information . ) This quantitative measure was used as the basis for our non-parametric approach described below . The two measures ( categorical and quantitative ) correspond well and clinical treatment effectiveness ranged widely , from very low effectiveness ( 5% of patients virologically suppressed after one year of treatment on AZT monotherapy ) to very high effectiveness ( 100% of patients virologically suppressed after one year of treatment on 3TC+AZT+LPV/r ) ( Figure 3A , B ) . 10 . 7554/eLife . 10670 . 007Figure 3 . Drug resistance mutations are correlated with diversity reduction differently in different types of treatments . Treatment effectiveness from literature review ( percentage of patients with virologic suppression after ~48 weeks ) showed positive correspondence with clinical recommendation among RTI regimens ( A ) and PI+RTI regimens ( B ) . Diversity reduction accompanying a DRM ( ΔDRM from the GLMM ) lower among the more effective and clinically recommended treatments among RTI treatments ( C ) and RTI+PI treatments ( D ) . 95% confidence intervals are plotted by excluding the highest 2 . 5% and lowest 2 . 5% of GLMM random effect fits of the 1000 subsampled datasets and treatments are ordered by mean ΔDRM within treatment categories . Generalized linear model fits show significantly different slopes for NNRTI treatments versus NRTI treatments ( E ) and PI/r treatments versus PI treatments ( F ) . Confidence intervals are plotted by excluding the highest 2 . 5% and lowest 2 . 5% of GLM fits to 1000 subsampled datasets . DOI: http://dx . doi . org/10 . 7554/eLife . 10670 . 00710 . 7554/eLife . 10670 . 008Figure 3—figure supplement 1 . Drug resistance mutations are correlated with diversity reduction differently in different types of treatments when all years are included . This figure is analogous with Figure 3 from the main text , but in this case , the data include 45 sequences sampled before 1995 . The inclusion of these sequences substantially changes the effect of the subsampling procedure , which explains the difference in scales of these figures ( see Materials and methods: p-thinning to adjust for the effect of year ) . The figure caption is otherwise shared with Figure 3 . DOI: http://dx . doi . org/10 . 7554/eLife . 10670 . 00810 . 7554/eLife . 10670 . 009Figure 3—figure supplement 2 . Drug resistance mutations are correlated with diversity reduction differently in different types of treatments with un-truncated data . This figure is analogous with Figure 3 from the main text , but in this case , the data are not truncated to only include the patients with 4 or fewer DRMs . The figure caption is otherwise shared with Figure 3 . DOI: http://dx . doi . org/10 . 7554/eLife . 10670 . 009 We hypothesize that effective treatments ( such as those containing an NNRTI or boosted PI ) likely make adaptation in viral populations limited by the generation of mutations and these populations should thus experience harder selective sweeps leading to a sharp reduction in diversity accompanying each additional DRM . Less effective treatments on the other hand ( such as those containing only NRTIs or unboosted PIs ) likely allow replication of fairly large HIV populations so that adaptation is not limited by the generation of mutations . They should thus experience soft selective sweeps and little or no reduction of diversity with each additional DRM . Below we test this hypothesis by assessing the reduction of diversity associated with the presence of a DRM among treatments that vary in clinical effectiveness . Before we are able to test this hypothesis , we have to deal with a peculiarity of our data . We found that even for sequences that carried no resistance mutations , more ambiguous calls were reported over time . This is likely due to increased awareness of genetic diversity in the HIV community , and not because diversity actually increased . We therefore employed a p-thinning routine to repeatedly subsample the data so that diversity measures would be comparable across years ( see Materials and methods: p-thinning to adjust for the effect of year ) . To determine whether the effect of DRMs on within-patient diversity depends on clinical treatment effectiveness , we first fit a generalized linear mixed model ( GLMM ) using the number of DRMs and sequence length to predict diversity as measured by the number of ambiguous calls . Because we found the subsampled number of ambiguous calls to be distributed according to a negative binomial distribution , we used a negative binomial error distribution ( see Materials and Methods: Quantifying the relationship between clinical effectiveness and diversity reduction ) . For each treatment , we report the total effect of the number of DRMs on diversity ( ΔDRM ) as the sum of the treatment-specific random effect plus the overall fixed effect from the GLMM ( ΔDRM , t+ΔDRM , all ) . These ΔDRM coefficients are plotted in Figure 3C , D . We included only treatments that had at least 15 sequences and a sufficient number of observed patients with different numbers of DRMs ( see description of abundant treatment dataset in Materials and methods: Data collection & filtering for more information ) . We also excluded 45 sequences sampled before 1995 which had an extreme influence on the p-thinning routine ( see Materials and methods: p-thinning to adjust for the effect of year for details ) . The analysis that includes these sequences leads to qualitatively similar results ( see the supplement ) . We also used only sequences with at most 4 DRMs . This captures the initial change in diversity due to the fixation of DRMs , and allows the ΔDRM measure not to be driven by few patients with many DRMs . The same analysis with all sequences is repeated in the supplement and yields qualitatively similar results . Lower ΔDRM values correspond to a bigger decrease in diversity associated with each DRM - a pattern more consistent with hard selective sweeps . We find that most of our ΔDRM estimates are qualitatively consistent with expectations: effective treatments have lower ΔDRM values than less effective treatments . Most NNRTI-based treatments are associated with a reduction of diversity per DRM . In 9 of 11 NNRTI-based treatment regimens , ΔDRM is significantly below 0 ( Figure 3C ) . This pattern suggests the presence of hard sweeps , although there is variation in effect size . Less effective treatments containing only NRTIs were generally associated with a smaller or no reduction in diversity per DRM . In some cases , such as DDI or AZT monotherapy , there was even an increase of diversity associated with DRMs ( a significantly positive ΔDRM value , see Figure 3C ) . This pattern is suggestive of soft sweeps . We found a slightly positive value of ΔDRM for the NRTI-based regimen 3TC+ABC+TDF , a treatment which is known to often lead to rapid treatment failure ( Gallant et al . , 2005; Khanlou et al . , 2005 ) . Among the most negative ΔDRM values of the NNRTI-based treatments were the treatments 3TC+ABC+EFV and EFV+FTC+TDF , which have long been on the list of recommended treatments in the USA ( Department of Health and Human Services , 2015 ) ( until their recent replacement with INSTI-based treatments which are not in our dataset ) . Among treatments containing PIs , both of the effective boosted-PI treatments had ΔDRM significantly below 0 ( Figure 3D ) . The less effective unboosted PI treatments had ΔDRM values on average closer to 0 , and two of five unboosted treatments had a ΔDRM value above 0 . As expected , we find that the LPV/r treatment in our dataset has a much lower ΔDRM value than the NFV treatments , consistent with the recommendations that LPV/r is preferable to NFV in relation to drug resistance prevention ( Walmsley et al . , 2002 ) . When the analysis is done without truncating the number of DRMs ( Figure 3—figure supplement 1 ) , results are qualitatively similar , but the values of specific treatments have shifted due to the effects of sequences with varying numbers of DRMs . When the 45 sequences before 1995 are included ( Figure 3—figure supplement 2 ) , results are also qualitatively similar . The fixed effects from the three different GLMM fits can be found in Table 1 . 10 . 7554/eLife . 10670 . 010Table 1 . Model fits for the fixed effects from GLMMs fit to subsampled data . See Materials and methods: Quantifying the relationship between clinical effectiveness and diversity reduction for further explanations of coefficients . Means of model fits for 1000 independent subsamples are reported for the three different subsampling and model-fitting regimes . 95% confidence intervals ( excluding the top 2 . 5% and bottom 2 . 5% of coefficent fits ) are given in parentheses . DOI: http://dx . doi . org/10 . 7554/eLife . 10670 . 010αall ( Intercept ) Δ ( Number of DRMs ) γ ( Length ) 1995+ , ≤ 4 DRMs-0 . 78-0 . 160 . 0030 ( -0 . 91 , -0 . 65 ) ( -0 . 17 , -0 . 14 ) ( 0 . 0029 , 0 . 0032 ) 1995+ , All DRMs-0 . 89-0 . 0970 . 0030 ( -1 , -0 . 77 ) ( -0 . 11 , -0 . 088 ) ( 0 . 0029 , 0 . 0032 ) 1989+ , ≤ 4 DRMs-1 . 20-0 . 150 . 0030 ( -1 . 4 , -1 . 1 ) ( -0 . 17 , -0 . 14 ) ( 0 . 0028 , 0 . 0032 ) To quantify and further test the observation that more clinically effective treatments lead to the greater diversity reduction per fixed DRM , we use two primary approaches , one parametric and the other non-parametric .
Treatment for HIV-1 represents an enormous success of modern medicine . Whereas early antiretroviral treatments were associated with fast evolution of drug resistance and high rates of treatment failure , there are currently many combinations of drugs that are successful at keeping HIV-1 at low or undetectable levels for many years , preventing the evolution of resistance and the progression to AIDS . Indeed , the evolution of drug resistance has become fairly uncommon ( Lee et al . , 2014 ) . In this paper , we have shown that this shift from fast to slow evolution of drug resistance has been accompanied by a corresponding shift from soft selective sweeps ( in which the same DRM occurs on multiple genetic backgrounds ) to hard selective sweeps ( in which a DRM occurs only a single time ) . This suggests that modern treatments have brought HIV into a regime where the viral population must wait until the correct mutation or combination of mutations is generated . This also means that for any given patient the acquisition of drug resistance has become at least partly an unlucky occurrence , in sharp contrast to the early days of HIV treatment in which all patients predictably failed treatment . Harder sweeps within well-treated patients are also consistent with the overall decrease in the rate of resistance . We want to study how the evolutionary dynamics of selective sweeps have changed in the evolution of drug resistance over the past two and a half decades . Because it would be unethical to give subpar treatment to HIV infected patients , we can only investigate this question using historical data . The only type of data that is available for a wide range of treatments and time points is Sanger sequencing data , due to its importance in HIV research and diagnostic testing . Although we often have only one sequence per patient , it is important to note that there is still information present in these sequences about genetic diversity and selective sweeps . First , we used ambiguous nucleotide calls as a proxy for genetic diversity . Although we are not the first study to do so ( Kouyos et al . , 2011; Li et al . , 2012; Poon et al . , 2007; Zheng et al . , 2013 ) , as far as we are aware , we are the first to use clonal sequences to validate its accuracy as a measure . Second , we can use the number of fixed drug resistance mutations to determine how much adaptation to treatment has taken place . We used a fairly conservative list of DRMs that are very unlikely to fix in the absence of the drug ( Mesplède et al . , 2013; Gonzalez et al . , 2004; Cong et al . , 2007 ) . Therefore , DRMs must have fixed as a result of strong positive selective pressure imposed by the drug , and are indicative of recent selective sweeps . We can then look at the correlation between the number of drug resistance mutations and genetic diversity and see if that relationship has changed across treatments and time . Because this approach relies only on widely-available Sanger sequences , we were able to compare 29 different treatments , from AZT monotherapy to treatments based on boosted PIs , sampled across more than two decades ( 1989–2013 ) . Examining the relationship between DRMs and diversity recapitulates expected results: we first find that across the entire dataset , sequences with a single DRM have lower genetic diversity than sequences without any DRMs . This result confirms a finding from a previous much smaller study looking at patients on NNRTI based treatments ( Pennings et al . , 2014 ) . In addition , we find that having more DRMs is associated with a greater reduction in diversity . This pattern could be generated by successive selective sweeps , in which DRMs are fixed one by one and each selective sweep lowers the diversity further . Alternatively , multiple DRMs may have fixed simultaneously in a single selective sweep . The key result of this paper ( illustrated in Figure 1 ) is that drug resistance mutations are associated with reduced diversity in patients on effective treatments , whereas this pattern is not seen among patients on older treatments with low clinical effectiveness . For example , among patients given treatments with 50% effectiveness , sequences with 3 DRMs are predicted to have approximately the same number of ambiguous calls as those with 0 DRMs . In contrast , among those patients given treatments with 80% effectiveness , sequences with 3 DRMs are predicted to have over 50% fewer ambiguous calls than those with 0 DRMs , a substantial decrease in genetic diversity . Thus , the higher the treatment effectiveness , the more DRMs are associated with low genetic diversity . This is consistent with drug resistance evolution dominated by soft selective sweeps when failure rates were high , transitioning over time to evolution dominated by hard selective sweeps as treatments improved and failure rates became much lower . Clinically effective treatments are thus characterized by a more frequent occurrence of hard selective sweeps . It is of interest to compare our new results to a previous study by one of us ( Pennings et al . , 2014 ) . Sequences in that study came from patients who were mostly treated with EFV + IDV , a combination that never became common and is not represented here , but which has an estimated effectiveness of 75% ( Staszewski et al . , 1999 ) . The study examined selective sweeps in those patients by looking at the fixation of a particular DRM , at amino acid 103 in RT , which changes from Lysine ( K , wild type ) to Asparagine ( N , resistant ) . K103N is special because it can be caused by two different mutations , as the wild type codon AAA can mutate to AAT or AAC , both of which encode Asparagine . When focusing on patients whose virus acquired the K103N mutation , the study found that in some patients , both the AAT and AAC codons were found ( which is clear evidence that a soft sweep has happened , see Figure 1 in the original paper ) , whereas in other cases only one of the two was there ( which suggests that a hard sweep may have happened , see Figure 2 in the original paper ) . Because of the detailed data available for these patients , it was shown that both soft and hard sweeps were occurring almost equally often . Placing the 75% effectiveness of EFV+IDV in the context of our above results , this is also what we would have predicted . Now we know that this result ( hard and soft selective sweeps occur ) is not necessarily something that will be generally true for HIV , but rather it is a function of the effectiveness of the treatment . Had Pennings et al . had data from a much worse or much better treatment , they might have concluded that hard sweeps or soft sweeps were the rule in HIV . The transition to highly effective treatments and hard selective sweeps was not abrupt . As visible in Figure 3 , treatment effectiveness and ΔDRM do not cluster into distinct groups based solely on the number and type of component drugs . The incremental changes in effectiveness and the evolutionary dynamics are worth noting because a simplified narrative sometimes suggests that solving the drug resistance problem in HIV was achieved simply by using three drugs instead of two ( Stearns and Koella , 2007 ) . According to this narrative , HIV can always easily evolve resistance when treatment is with one or two drugs , but it is virtually impossible for the virus to become resistant to three drugs . In truth , only some specific combinations seem to lead to more favorable evolutionary dynamics for patients ( as seen in Figure 3A , B ) . Several potential mechanisms could drive the observation that more effective drug combinations drive hard sweeps in within-patient populations of HIV-1 . Better drugs may allow for a faster collapse of population size , decreasing the probability that one or more ‘escape’ DRMs occurs ( [Alexander et al . , 2014] , although see [Moreno-Gamez et al . , 2015] ) . Alternatively , suppressed HIV populations may continue replicating at small numbers , and better drugs may cause this replicating population to be smaller than among patients given inferior drugs . Similarly , a treated patient may retain a reservoir of HIV unreachable by the treatment , and better drugs may make this reservoir smaller . If newer drugs have fewer side effects and therefore improve adherence among patients , this too could result in a smaller within-patient population size among patients treated with better drugs , and contribute to the decreased production of resistant genotypes . The effect could also be driven by standing genetic variation , if worse drugs would allow pre-existing mutations to establish , whereas better drugs make this less likely , for example , if a pre-existing mutation could only establish if it occurred in a small and specific compartment ( Moreno-Gamez et al . , 2015 ) . Finally , more effective drug combinations may simply require more mutations on the same background ( i . e . , a higher genetic barrier to resistance [Tang and Shafer , 2012] ) , so the effective mutation rate to a fully resistant genotype is lower . Our data do not give us sufficient resolution to distinguish between these hypotheses , and the true dynamics may be a combination of all of these factors . However , all these potential mechanisms work to reduce the rate of production of resistant genotypes and the principle is more general: if the probability of acquiring resistance is low , rare resistance should be generated by hard selective sweeps . It is notable that we can detect differences in relative changes in diversity , given the constraints of our data . Our data are cross-sectional , from different patients treated with different regimens from many studies over more than two decades . We have sequences from only one or two genes , and so it is possible that there are DRMs or other selected mutations driving population dynamics outside of the sequenced regions ( particularly among patients treated with protease inhibitors ) . The DRMs that are observed may be associated with variable selection coefficients and may therefore differ in their rate of expansion in the population and thus the associated signature of the selective sweep . Additionally , DRMs occurring later in the course of an infection may have smaller beneficial effects as compared to early DRMs . Different patients may have varying profiles of diversity at the time of diagnosis , which could also effect whether sweeps will be hard or soft . All of these items could serve to obscure the relationship between DRM fixation and changes in diversity , but our results appear robust despite the substantial noise . Although our results appear robust , certain systematic biases could also affect our results . First , we observe an increase in the number of ambiguous calls through our study period . This is likely due to a greater awareness of within-patient diversity and improvements in sequencing technology . Having more ambiguous reads called in later years gives us more power to detect large decreases in diversity as compared to earlier years . As effective treatments come from later in the study period , this could lead to more observed hard sweeps among effective treatments . However , we believe that our conclusion is robust to this systematic bias as we used a conservative p-thinning procedure that explictly equalized the power to detect hard and soft sweeps across years . Time can also factor into our analysis in other ways that are harder to control . Improved disease monitoring may mean that modern patients have a shorter time between the appearance of a DRM and its sequencing and discovery . This could result in hard selective sweep signatures that are more pronounced in later years because they are detected before they have had time to erode . However , Pennings et al . estimate that sweep signatures require a substantial amount of time to erode ( on the order of months ) ( Pennings et al . , 2014 ) , so we do not believe that this effect alone can explain the pattern . Alternatively , if DRMs acquired in response to earlier treatments have lower selection coefficients than treatments given more recently , early selective sweeps would take longer . Recombination could move DRMs to multiple genetic backgrounds before the mutation fixes in the population . This would also also result in muted changes in diversity among early treatments that could not be readily distinguished from the signatures of soft sweeps . However , there is no evidence that selection coefficients of DRMs among early treatments are smaller . Finally , because our primary measure to understand the effect of the fixation of a drug resistance mutation is an internal comparison with other patients on the same treatment , we do not believe that our results could be generated solely by data heterogeneity . Further , that our results are robust to different measurements and filtering approaches suggests DRMs do indeed sweep differently in HIV populations subject to more or less effective treatments . Drug resistance evolution is no longer the threat to HIV patients it once was and treatments exist that almost never lead to resistance . However , our observation that effective drug combinations lead to hard selective sweeps could be useful for improving treatments for other pathogens . Even among relatively small samples , if patients who acquire drug resistance mutations do not have a significant decrease in diversity relative to those who did not acquire drug resistance mutations , this might suggest that the fixing mutations are occurring via soft selective sweeps , and the treatment brings patients into a dangerous regime for drug resistance . Alternatively , if patients acquire drug resistance mutations , but those patients also have very little diversity , it may be that this is a safer treatment and that the emergence and fixation of drug resistance mutations was a relatively uncommon occurrence . Looking at changes in diversity following a sweep in order to assess the mode of adaptation could be particularly well-suited to looking at evolution of cancer . While single cell sequences isolated from tumors have yielded promising insights about evolutionary dynamics , the process is invasive and relatively difficult . Sequencing tumor-free cancerous cells circulating in the blood provides less information , but can be done serially and provides a good measure of tumor heterogeneity . Applying a method such as ours to monitor changes in cell-free DNA diversity over time may allow us to determine if certain treatments reproducibly lead to soft sweeps and thus are very likely to fail in general . Comparing treatment effectiveness with the occurrence of soft and hard selective sweeps may also provide supplementary information about additional risk factors among patients . In the case of HIV , we find that high effectiveness is associated with hard selective sweeps , which suggests that the the virus has a hard time evolving drug resistance , and the patients in whom resistance evolved are merely the unlucky ones . However , there may be cases where failure is rare , but associated with soft selective sweeps . Such a situation thus reflects a discordance between what happens within patients and what we see at a population level . This , in turn , may be indicative of a behavioral , genetic or virologic difference among the groups of patients and efforts should be made to find out how failing patients are different from non-failing patients . In conclusion , we find that the study of diversity in viral populations with resistance can show differences in evolutionary pathways of adapting pathogenic populations and provides a concrete example of how population genetics theory can make substantive predictions about medically relevant problems . Next generation and single molecule sequencing have the capacity to bring much more precision in determining the dynamics of within-patient populations . However , we also urge researchers and clinicians to report more information concerning the diversity of pathogen populations , even in the form of minor allele frequency cut offs for calling ambiguous calls or raw sequencing data , as this might allow new insight from data that might be otherwise overlooked .
Sequences from reverse transcriptase and protease were analyzed to determine the number of ambiguous calls and the number of drug resistance mutations per sequence . We called a nucleotide non-ambiguous if it read A , T , C or G , and grouped lowercase ( and less confident ) a , t , c and g calls with their capital counterparts . Nucleotides called as W , S , M , K , R , Y ( ambiguity between two nucleotides ) and B , D , H , V ( ambiguity between three nucleotides ) and their lowercase counterparts were included as ambiguous calls . Ns and Xs ( indicating no information about the identity of the position ) were excluded . We also examined ambiguities on the amino acid level by using nucleotide level information . If , for example , a nucleotide triplet was recorded as AAM , where M indicates an adenine/cytosine ambiguity , the amino acid at that position was ambiguous between AAA ( Lysine , K ) and AAC ( Asparagine , N ) . The amino acid for that position would then be recorded K/N . All ambiguous calls at the nucleotide level were translated into ambiguous calls at the amino acid level , including if the ambiguous call reflected synonymous encodings ( i . e . , AAA and AAG are both Lysine , and the amino acid would be encoded K/K ) . The number of ambiguous amino acids was recorded for each sequence . We determine whether mutations are associated with drug resistance based on the 2009 update of DRMs for the surveillance of transmitted HIV-1 drug resistance adopted by the World Health Organization ( Bennett et al . , 2009 ) , which lists drug resistance mutations that are indicative of selective pressure . For each patient , we determine the number of mutations that confer drug resistance at the amino acid level to any of the classes of drugs the patient is receiving ( i . e . , NRTI , NNRTI , PI ) . These drug resistance mutations are counted as fixed if they are either non-ambiguously the resistant type , or they are an ambiguous call with all possible states as the resistant type ( i . e . , N/N in the example above ) . 5163 patients in the D-PCR dataset had a sequenced protease gene available in the database ( 77% of patients ) . Not all patients had entire reverse transcriptase and protease genes sequenced , but only 1% of sequences had fewer than 500 basepairs sequenced of 1680 possible in reverse transcriptase and only 1% of available protease sequences had fewer than 210 of 300 bases in protease . In order to validate the appropriateness of ambiguous calls as a proxy for genetic diversity , we computed diversity using an ambiguous call measure and compared it to a diversity measure that did not use the ambiguous call data . For each site , we computed genetic diversity using the ambiguous calls as the proportion of all D-PCR HIV-1 sequences that had an ambiguous basepair call at that site . This approximates the percentage of patients with within-patient diversity by site . In order to test if this measure of diversity based on ambiguous calls correlates well with other measures of diversity that don’t depend on ambiguous calls , we used the clonal dataset to compute the average sitewise π . For a site with nucleotides A , T , C and G at within-patient frequencies pA , pT , pC and pG , π is computed asπ=1-∑i∈{A , T , C , G}pi2 π is equal to zero when every sequence has the same basepair call ( i . e . , all As ) and is maximized when multiple categories are at intermediate frequencies ( an even split between A , T , C and G ) . Within-patient diversity was measured by first computing π at each site separately within each patient and then averaging over all patients . The π calculation at a site for a particular patient was only included if the patient had at least two sequences calling non-N identity at that site . We computed diversity in the same way at the amino acid level to validate that our signals persisted when looking at codons . We expect that clinical effectiveness of an HIV treatment affects the probability of the virus undergoing a soft or hard selective sweep . All HIV treatment regimens that occurred at least 50 times in the D-PCR dataset were evaluated for treatment effectiveness based on a literature review . As a measure for treatment effectiveness , we recorded the proportion of patients whose treatment was still successful after a year of treatment , as indicated by a viral load of ≤ 50 copies of HIV-1 RNA/mL or less after 48 or 52 weeks in an on-treatment analysis . Our literature review was mostly based on the papers reviewed in Lee et al ( Lee et al . , 2014 ) . Because this review did not include review information for several older treatment regimens , we supplemented our analysis with additional studies . A full description of how clinical treatment effectiveness was calculated by study can be found in the supplement ( Supplementary files 1 and 2: Determining treatment effectiveness ) . A second researcher randomly chose 5 studies and independently followed the protocol to determine treatment effectiveness for these studies , providing confirmation of our method . Because we believe the thus collected information may be useful to other researchers , we provide our estimates in the supplement ( Supplementary file 3 ) . Because the drug regimens systematically changed over time ( Figure 5—figure supplement 2A ) , it is essential to ensure that our estimates of diversity are not confounded by the changes in sequencing practices over the same time . While all samples were Sanger sequenced , we do find that the number of recorded ambiguous calls increased over time ( Figure 5—figure supplement 2C ) , possibly because the cut-off of calling a read as ambiguous became lower . This effect , taken on aggregate across all sequences was not significant ( p=0 . 09 , linear regression with year predicting the number of ambiguous calls ) . However , when examining only sequences with 0 DRMs , a strong positive correlation emerges , with each year associated with 0 . 56 more ambiguous calls per sequence ( p=4 . 9 × 10−8 ) . That the difference in effect can be seen in the zero DRM class but not among sequences with any number of DRMs underscores the strong interaction between low diversity and multiple DRMs among modern treatments . Because the probability of calling a nucleotide as ambiguous has increased over time , we have potentially greater power to detect changes in the number of ambiguous reads with increasing number of DRMs . To account for this change in power , we used a p-thinning procedure ( Grandell , 1997 ) to control for the effect of year on the number of ambiguous calls . This allowed us to attribute differences in the relationship between the number of DRMs and diversity to treatment and not to increased probability of calling reads as ambiguous in later years . We first measured the effect of year on diversity by fitting a linear model to the number of ambiguous calls in each year among all sequences with zero observed DRMs . By limiting our analysis to only sequences without DRMs , we remove the hypothesized change in diversity expected following a sweep . We observed the following relationship between year and the number of ambiguous calls: ( 1 ) ( Number of ambiguous reads ) =f ( Year ) =−505 . 92+0 . 26∗Year According to this relationship , we computed probabilities of calling ambiguous reads in different years relative to the first year sampled ( here , 1989 ) . We subsampled our data using these probabilities so that late and early years had comparable numbers of ambiguous calls . For example , in 2000 , nearly twice as many ambiguous calls were reported as compared to 1989 . Therefore , for each ambiguous call observed in a sequence from 2000 , we include it in our sample with probability ∼1/2 . Intuitively , this translates observations from later years into units of ambiguous calls that are comparable to those observed in 1989 . The subsample effect for year i with respect to 1989 ( p1989 , i ) is calculated as P1989 , i=f ( 1989 ) /f ( i ) ( see Equation 1 ) . The fit of Equation 1 to the year means and the thinning effect Pi are shown in Figure 5—figure supplement 3A in black . Because very few ambiguous calls were made in 1989 , p-thinning all read counts to be comparable to sequences from 1989 lowers the resolution of our data . We therefore also performed subsampling using only data after and including 1995 , which retained more ambiguous calls . This particular cutoff was chosen for two reasons . First , ambiguous calls were not reliably recorded until several landmark papers studying within-patient diversity were published in 1993 ( Larder et al . , 1993; Piatak et al . , 1993 ) . Second , before 1995 , we have <20 sequences/year , so we have much lower resolution to determine the rates of ambiguous read calling . This excluded only 45 sequences , and changed the linear model fit only slightly ( see Equation 2 and Figure 5—figure supplement 3A ) . ( 2 ) ( Number of ambiguous reads ) =f1995 ( Year ) =−465 . 96+0 . 24∗Year Using data only after and including 1995 , the subsample effect for year i ( P1995 , i ) is calculated as P1995 , i=f1995 ( 1995 ) /f1995 ( i ) . Although the linear fits are similar , the effect of the subsampling is less severe ( see Figure 5—figure supplement 3B ) . This is because later years are rescaled to be comparable to 1995 observations as opposed to 1989 observations . Observations from 1995 have a greater number of ambiguous calls . In including each ambiguous call with probability relative to the subsample effect , for each sequence j sampled in year i , the number of ambiguous calls for sequence j is distributed as Poisson ( ( Number of ambiguous calls on sequence j ) ×Pi ) . Note , samples taken from the reference year ( i . e . , 1989 or 1995 ) were also re-drawn according to a Poisson distribution with λ = 1 . This process is known as p-thinning ( Grandell , 1997 ) , and is similar to standard bootstrapping and m out of n resampling ( see [Politis et al . , 1999] ) . These subsampled counts for the number of ambiguous calls are used throughout the analysis below . We estimate the relationship between the number of DRMs and genetic diversity by fitting a generalized linear mixed model ( GLMM ) with a negative binomial error distribution for our 29 abundant treatments . We found that the subsampled data visually fit a negative binomial distribution much more closely than a Poisson distribution , which is often used for count data ( See Figure 5—figure supplement 4 and Figure 5—figure supplement 5 ) . In this model , length , the number of DRMs and an intercept term are fit as fixed effects , and the number of DRMs by treatment is fit as a random effect . This allows us to assess the relationship between diversity and the number of ambiguous reads separately for each treatment . The models were fit using the glmmADMB package in R ( Fournier et al . , 2012 ) . ( 3 ) Subsampled number of ambiguous reads~ΔDRM , all ( numDRM ) +αall+γ ( Sequence Length ) + ( αt+ΔDRM , t ( numDRM ) |Regimen ) This model was fit to 1000 datasets created by p-thinning the number of ambiguous calls . The overall effect of a DRM on diversity was fit by the ΔDRM , all term , but the effect of a DRM on diversity by treatment t is fit by the random effect term , Δt . The full effect of a DRM on diversity for a given treatment was called ΔDRM and was computed by combining the treatment-specific random effect and the overall fixed effect of the model ( ΔDRM=ΔDRM , t+ΔDRM , all ) . Confidence intervals were generated by excluding the highest and lowest 2 . 5% of estimates of ΔDRM among the subsamples . We performed the above procedure three times: our main analysis used the p1995 , i procedure for p-thinning sequences from year i and included sequences with 4 or fewer DRMs . We performed this same analysis using p1995 , i-thinning and including all numbers of DRMs and using p1989 , i-thinning and including only sequences with 4 or fewer DRMs . To quantify the effect of treatment effectiveness on ΔDRM , we used both a parametric and a non-parametric approach . | In the early days of HIV therapy , the strains of the virus that infected patients frequently evolved drug resistance and the therapies would often eventually fail . These treatments generally involved using a single anti-viral drug . Nowadays , better therapies involving combinations of several anti-viral drugs are available and drug resistance in HIV is a much rarer occurrence . This means that now a particular therapy may be an effective treatment for an HIV-infected individual over much longer periods of time . A theory of population genetics predicts that when it is easy for a population to acquire a beneficial genetic mutation – like one that provides drug resistance – multiple versions of that mutation may spread in the population at the same time . This is called a soft selective sweep . However , when beneficial mutations occur only rarely , it is expected that only one version of that mutation will take over in a population , which is known as a hard selective sweep . Here , Feder et al . test this theory using data from 6717 patients with HIV who were treated between 1989 and 2013 using a variety of different drug therapies . The experiments aimed to find out whether the transition from the older drug therapies –where the virus frequently acquired resistance – to the newer , more effective drugs was associated with a transition from soft to hard sweeps . Feder et al . find that HIV more often evolved drug resistance via soft sweeps in patients treated with the less effective drug combinations ( like those given in the early days of HIV treatment ) , while hard sweeps were more common with the more effective drug combinations . This suggests that good drug combinations may allow fewer drug resistance mutations to occur in the HIV population within a patient . This may be because there are fewer virus particles in these patients , or because the specific combinations of mutations that provide resistance occur less often . Feder et al . ’s findings are a step towards understanding why modern HIV treatments work so well , which will ultimately help us find better treatments for other infectious diseases . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"evolutionary",
"biology"
] | 2016 | More effective drugs lead to harder selective sweeps in the evolution of drug resistance in HIV-1 |
Salmonella PhoQ is a histidine kinase with a periplasmic sensor domain ( PD ) that promotes virulence by detecting the macrophage phagosome . PhoQ activity is repressed by divalent cations and induced in environments of acidic pH , limited divalent cations , and cationic antimicrobial peptides ( CAMP ) . Previously , it was unclear which signals are sensed by salmonellae to promote PhoQ-mediated virulence . We defined conformational changes produced in the PhoQ PD on exposure to acidic pH that indicate structural flexibility is induced in α-helices 4 and 5 , suggesting this region contributes to pH sensing . Therefore , we engineered a disulfide bond between W104C and A128C in the PhoQ PD that restrains conformational flexibility in α-helices 4 and 5 . PhoQW104C-A128C is responsive to CAMP , but is inhibited for activation by acidic pH and divalent cation limitation . phoQW104C-A128C Salmonella enterica Typhimurium is virulent in mice , indicating that acidic pH and divalent cation sensing by PhoQ are dispensable for virulence .
Salmonellae are Gram-negative bacterial pathogens that cause severe gastroenteritis and systemic disease in animals and humans . Critical for salmonellae virulence is their ability to survive and replicate within host cells ( Fields et al . , 1986 ) . Following phagocytosis by macrophages , salmonellae are contained within a phagosomal environment containing a diversity of antimicrobial factors including proteases , reactive oxygen and nitrogen species , acidic pH and cationic antimicrobial peptides ( CAMP ) ( Flannagan et al . , 2009 ) . Salmonellae have multiple mechanisms , including the PhoQ sensor , to sense the phagosomal milieu and respond by increasing their resistance to host antimicrobial factors ( Haraga et al . , 2008; Chen and Groisman , 2013; Dalebroux and Miller , 2014 ) . PhoQ is the sensor kinase component of the PhoPQ two-component regulatory system that governs the phosphorylated state of the response regulator PhoP ( Groisman et al . , 1989; Miller et al . , 1989 ) . PhoQ exists as a dimer within the inner membrane and has a periplasmic sensor domain ( PD ) that transduces signals across the inner membrane to the cytoplasmic histidine kinase domain . Following activation of PhoQ by the phagosomal environment , PhoP is phosphorylated and transcriptionally controls a large network of genes ( >300 ) , many of which are involved in virulence ( Fields et al . , 1989; Behlau and Miller , 1993; Belden and Miller , 1994; Gunn and Miller , 1996; Guo et al . , 1997; Bearson et al . , 1998; Guo et al . , 1998; Adams et al . , 2001; Bader et al . , 2003; Dalebroux et al . , 2014 ) . Precise PhoPQ-mediated gene regulation is essential for salmonellae infection as strains with null or constitutively active mutations in PhoPQ are highly attenuated for virulence in animals and humans ( Fields et al . , 1989; Galán and Curtiss , 1989; Miller et al . , 1989; Miller and Mekalanos , 1990 ) . The PhoQ PD is a member of the PAS-fold and PDC-fold domain families ( Cho et al . , 2006; Cheung et al . , 2008; Cheung and Hendrickson , 2010 ) . Unlike other PDC-sensors , which bind small ligands in a defined binding pocket or PhoQ PD homologs found in environmental bacteria , the PhoQ PD from bacteria that primarily interact with animals has no apparent binding pocket due to an occluding structural element: α-helices 4 and 5 ( Cho et al . , 2006; Prost et al . , 2007; Cheung et al . , 2008; Prost et al . , 2008 ) . Acidic residues on α4 and α5 and β-strands 5 and 6 in the PhoQ PD form a structural scaffold for binding antimicrobial peptides , as well as the divalent cations Mg2+ , Mn2+ , and Ca2+ ( Waldburger and Sauer , 1996; Bader et al . , 2005; Cho et al . , 2006; Prost et al . , 2008 ) . PhoQ kinase activity is repressed and phosphatase activity is dominant at millimolar or greater concentrations of divalent cations ( Garcia Vescovi et al . , 1996; Castelli et al . , 2000; Montagne et al . , 2001 ) , presumably due to divalent cation salt-bridges formed between the PD acidic patch and inner membrane phospholipids ( Cho et al . , 2006 ) . Additionally , PhoQ activity is repressed by feedback inhibition involving the small inner membrane protein , MgrB ( Lippa and Goulian , 2009 ) . Conversely , bacterial growth in sub-millimolar divalent cation conditions results in PhoQ activation and increased kinase activity ( Garcia Vescovi et al . , 1996 ) , presumably due to disruption of salt-bridges between the PhoQ PD and inner membrane . However , the macrophage phagosome has a magnesium concentration of approximately one millimolar and a calcium concentration of approximately 500 micromolar , suggesting PhoQ is not activated by divalent cation limitation during intracellular infection ( Christensen et al . , 2002; Martin-Orozco et al . , 2006 ) . At one millimolar divalent cation concentration , PhoQ can be activated by exposure to pH 5 . 5 or sub-inhibitory concentrations of CAMP ( Bader et al . , 2005; Prost et al . , 2007 ) . These are relevant host signals as the macrophage phagosome acidifies to approximately pH 5 . 5 and contains CAMP ( Alpuche Aranda et al . , 1992; Rathman et al . , 1996; Rosenberger et al . , 2004; Martin-Orozco et al . , 2006 ) . Furthermore , neutralization of acidified macrophage intracellular compartments with chemical inhibitors results in decreased PhoQ-mediated gene expression during infection ( Alpuche Aranda et al . , 1992; Martin-Orozco et al . , 2006 ) . Combined , these findings suggested a model in which acidic pH and CAMP activate PhoQ within the macrophage phagosome; however , the individual contribution of these signals to PhoQ-mediated virulence remained unknown . Acidic pH and CAMP additively activate PhoQ suggesting that the PD has distinct sensing mechanisms for these stimuli ( Prost et al . , 2007 ) . A variety of experimental data indicate that CAMP directly compete with divalent cations for binding sites within the PhoQ PD acidic patch , leading to a model in which CAMP activates PhoQ by disrupting salt-bridges with the inner membrane ( Bader et al . , 2005 ) . The mechanism by which PhoQ is activated by acidic pH appears to be distinct from CAMP and involves perturbations to a network of residues surrounding H157 within the α/β-core of the PD ( Prost et al . , 2007 ) . In this study , we defined conformational changes that occur within the PhoQ PD on exposure to acidic pH . Characterization of the conformational changes induced by acidic pH inspired the construction of PhoQ variants which are impaired for acidic pH and divalent cation sensing , but retain their ability to respond to CAMP . Prior to this study , it was unclear which signals were important for PhoQ-mediated virulence . Utilizing these PhoQ variants , we have now established that acidic pH and divalent cation sensing are dispensable signals for PhoQ-mediated systemic virulence of Salmonella enterica Typhimurium , suggesting that CAMP or other host molecules facilitate PhoQ-dependent pathogenesis .
PhoQ is activated in acidic conditions in vitro and within the acidified environment of the Salmonella-containing vacuole ( SCV ) after phagocytosis ( Alpuche Aranda et al . , 1992; Martin-Orozco et al . , 2006; Prost et al . , 2007 ) . However , the mechanism by which the PhoQ PD senses acidic pH is not well characterized . Previously , we reported that the ( 1H , 15N ) -HSQC-NMR spectrum of the S . enterica Typhimurium PhoQ PD is highly sensitive to changes in pH ( Prost et al . , 2007 ) . Therefore , to further understand PhoQ dynamics during activation by acidic pH we collected a series of ( 1H , 15N ) -HSQC-NMR spectra of the PhoQ PD as a function of pH ( Figure 1A ) . To extract residue information from the spectra , resonance assignments were determined for the PhoQ PD at pH 3 . 5 , the condition that yielded the greatest number of observable resonances . Of the 138 residues that can yield HSQC signals , 120 resonances could be assigned in the spectrum at pH 3 . 5 ( Figure 1B ) . The remarkably well dispersed spectrum indicates that the PD remains stably folded , even at pH of 3 . 5 . 10 . 7554/eLife . 06792 . 003Figure 1 . The annotated PhoQ PD ( 1H , 15N ) -HSQC-NMR spectrum reveals significant peak shifting and broadening during pH titration . ( A ) ( 1H , 15N ) -HSQC-NMR spectra of neutral to acidic pH-titration of the PhoQ PD . The pH-titration is represented as a magenta ( pH 6 . 5 ) to black ( pH 3 . 5 ) color gradient . The pH-titration spectra include pH 6 . 5 , 6 . 0 , 5 . 5 , 4 . 9 , 4 . 1 , and 3 . 5 . ( B ) The assigned ( 1H , 15N ) -HSQC-NMR spectra of the S . enterica Typhimurium PhoQ PD at pH 3 . 5 . Residue numbers are labeled proximal to their corresponding peak . DOI: http://dx . doi . org/10 . 7554/eLife . 06792 . 003 In the absence of other pH-dependent processes , resonances that arise from residues that undergo a protonation/deprotonation event will shift in a continuous manner . Such processes will appear in the so-called ‘fast-exchange’ NMR regime due to the rapid on/off rate of protons . Many resonances in the PD spectra exhibited pH-dependent fast-exchange behavior , consistent with ionization of the many histidine and acidic residues . In addition , some resonances broadened and disappeared from the spectrum as a function of pH . This behavior corresponds to intermediate-to-slow exchange and is indicative of a conformational change or the existence of multiple states that interconvert slowly . Thus , pH-dependent changes in the PhoQ PD HSQC spectra reveal regions of the domain that experience changes in functional group ionization and conformational dynamics . Spectra , collected at pH 3 . 5 and pH 6 . 5 , were compared to identify regions in the PhoQ PD that are sensitive to changes in pH ( Figure 2A ) . Resonances that experienced significant pH-dependent chemical shift perturbations ( CSPs > 0 . 08 ppm ) or broadened beyond detection , localize to regions of the protein that contain ionizable functional groups and/or experience conformational dynamics; thereby defining pH-responsive regions in the domain . Of the 120 assigned residues in the PhoQ PD , resonances from 42 residues were affected by transition from pH 6 . 5 to 3 . 5 ( Figure 2B ) . Due to resonance overlap and broadening , it is difficult to partition the two spectroscopic effects throughout the comparison . Approximately , 20 affected resonances broadened beyond detection at pH 6 . 5 , consistent with pH-dependent conformational dynamics in the PhoQ PD . Furthermore , 66 resonances were relatively unaffected , indicating that the PhoQ PD has pH-insensitive regions . 10 . 7554/eLife . 06792 . 004Figure 2 . The PhoQ PD experiences significant pH-dependent perturbations which map to α4 and α5 and the α/β-core . ( A ) Comparison of ( 1H , 15N ) -HSQC-NMR spectra of the PhoQ PD at pH 6 . 5 ( magenta ) and pH 3 . 5 ( black ) . ( B ) Residues that experience CSPs >0 . 08 ppm and/or peak broadening determined from the spectral comparison in panel A are mapped onto the S . enterica Typhimurium PhoQ PD ( residues 45–188 ) primary and secondary structures ( pH-sensitive residues , magenta; pH-insensitive residues , teal; ambiguous or non-assigned residues , no color ) . The locations of activating mutations from Figure 2—figure supplement 1 are indicated with asterisks . ( C ) pH-sensitive residues from panel A mapped onto the PhoQ PD structure ( PDB 1YAX ) ; pH-sensitive residues ( magenta ) , pH-insensitive residues ( teal ) , and ambiguous or non-assigned residues ( black ) . pH-sensitive secondary structural features are labeled with yellow circles ( NT , N-termini; CT , C-termini ) . ( D ) Continuous surface representation ( 1 . 4 Å probe ) of pH-sensitive ( magenta ) and pH-insensitive ( teal ) residues from panel C mapped onto the PhoQ PD crystal structure . DOI: http://dx . doi . org/10 . 7554/eLife . 06792 . 00410 . 7554/eLife . 06792 . 005Figure 2—figure supplement 1 . Residues involved in PhoQ activation and repression form a buried network connecting α4 and α5 to the α/β-core . ( A ) Mutations identified by random and site-directed mutagenesis confer increased PhoQ-dependent phoN::TnphoA alkaline phosphatase activity when grown in N-mm supplemented with 10 mM MgCl2 . The data shown are representatives from at least three independent experiments performed in duplicate and presented as the mean ± SD . ( B ) Activating mutations from panel A ( magenta ) mapped onto the S . enterica Typhimurium PhoQ PD primary and secondary structures ( residues 45–188 ) . ( C ) The locations of activating mutations from panel A ( magenta sticks ) mapped onto the PhoQ PD structure ( PDB 1YAX ) . Secondary structural features with activating mutations are labeled with yellow circles ( NT , N-termini; CT , C-termini ) . ( D ) Continuous surface representation ( 1 . 4 Å probe ) of activating mutations from panel C mapped onto the PhoQ PD crystal structure . DOI: http://dx . doi . org/10 . 7554/eLife . 06792 . 005 Assignments for the HSQC spectrum allowed us to identify residues in the PhoQ PD structure that experience pH-dependent changes ( Figure 2C ) . A majority of residues affected by pH localize to α1 , α2 , α4 , and α5 and proximal regions , including β5 , β6 , and β7 . As an independent approach , we randomly mutagenized the PD to identify mutations that activate PhoQ in the presence of repressing concentrations of divalent cations ( Figure 2—figure supplement 1 ) . pH-sensitive regions identified in the NMR experiments overlap or are proximal to many of the mutations identified in our screen for activating mutations in the PhoQ PD ( Figure 2B , asterisks ) . Similar to the activating mutations , a majority of the pH-sensitive residues form an interconnected network which spans α4 and α5 and the α/β-core ( Figure 2D ) . These data suggested that PhoQ PD residues and structural features important for activation and repression undergo conformational change during pH titration . A majority of the residues that were unaffected by changes in pH mapped distally to α4 and α5 , providing support for the hypothesis that the detection and response to pH is contained within localized structural elements of the PD . Altogether , these observations are consistent with a model where fluctuations in pH promote local conformational dynamics between α4 and α5 and the α/β-core as part of the pH sensing mechanism . The NMR and mutagenesis data suggested that α4 and α5 are dynamic and that the relationship between the two helices and the α/β-core of the PD likely plays a role in activation . To test this hypothesis , we engineered a PhoQ PD mutant that contains an internal disulfide bond predicted to restrict conformational dynamics between α4 and α5 and the α/β-core . Cysteine residues were introduced at positions W104 ( on α2 ) and A128 ( on α4 ) based on their side-chain surface exposure , relative geometries , and Cβ distance ( ∼6 Å ) observed in the S . enterica Typhimurium PhoQ PD structure ( PDB 1YAX ) . Non-reducing SDS-PAGE and western blotting of membranes harvested from phoQW104C-A128C S . enterica Typhimurium revealed a faster migrating PhoQ species relative to wild type , suggesting W104C and A128C form an intramolecular disulfide bond when expressed in bacteria ( Figure 3—figure supplement 1A ) . When treated with sample buffer supplemented with β-mercaptoethanol to reduce the disulfide bond , PhoQW104C A128C migrated similarly to the wild-type protein . Membranes harvested from phoQW104C-A128C S . enterica Typhimurium grown in N-minimal media ( N-mm ) at pH 7 . 5 or pH 5 . 5 , supplemented with 10 micromolar or 1 millimolar MgCl2 , or CAMP showed no observable differences in PhoQW104C-A128C disulfide bond formation by SDS-PAGE , suggesting formation of the W104C-A128C disulfide bond is not dependent on growth conditions or PhoQ activation state ( data not shown ) . These data indicated that the disulfide bond formed between W104C-A128C is stably maintained within the S . enterica Typhimurium periplasm . The PhoQW104C-A128C disulfide mutant was designed to inhibit motion between α2 and α4 , allowing us to determine whether the dynamics of α4 and α5 play a critical role in activation . When exposed to acidic pH or low divalent cation growth media , activation of the PhoQ-dependent phoN::TnphoA reporter in S . enterica Typhimurium was significantly reduced in phoQW104C-A128C relative to wild type ( Figure 3A ) . Additionally , the previously identified T48I activating mutation in the T48 D179 K186 ( TDK ) network in the PhoQ PD ( Miller and Mekalanos , 1990; Garcia Vescovi et al . , 1996; Sanowar et al . , 2003; Cho et al . , 2006 ) was suppressed by the W104C-A128C disulfide bond , supporting the hypothesis that α4 and α5 and the TDK network are an interconnected signaling element ( Figure 3B ) . Interestingly , the T48I mutation potentiates CAMP activation in the phoQT48I W104C-A128C background by an unknown mechanism . Importantly , CAMP still activated the phoN::TnphoA reporter in phoQW104C-A128C and phoQT48I W104C-A128C S . enterica Typhimurium at or above wild-type levels , indicating that these mutant proteins are functional ( Figure 3A , B ) . Chromosomal phoQW104C-A128C had a similar phenotype to phoQW104C-A128C expressed from the pBAD24 vector , indicating the phenotype is not an artifact of expression in trans ( Figure 3—figure supplement 1B ) . Furthermore , the phoQW104C-A128C phenotype does not appear to be exclusive to phoN as other PhoQ-regulated genes—pagD , pagO , and phoP—are significantly reduced for induction by acidic pH and divalent cation limitation , but are induced by exposure to CAMP , similar to wild-type bacteria ( Figure 3—figure supplement 2 ) . Serine substitutions at W104 and A128 did not recapitulate the phenotype observed for phoQW104C-A128C , but rather resulted in increased phoN::TnphoA reporter activity relative to wild type ( Figure 3—figure supplement 1C ) . Additionally , neither single cysteine nor single serine substitutions at W104 or A128 recapitulated the phoQW104C-A128C phenotype ( Figure 3—figure supplement 1D ) . These results confirmed that a disulfide bond is required for the phoQW104C-A128C phenotype . Similar to residues identified in our screen for activating mutations , the replacement of partially buried , hydrophobic residues at positions W104 and A128 with smaller , polar side-chains promoted activation . Altogether , these data confirmed that restricting conformational flexibility or movement of α4 and α5 inhibits activation by acidic pH , divalent cation limitation , and activating mutations in the TDK network . These data suggest that CAMP activates PhoQ by a mechanism that is distinct and separable from the mechanism by which acidic pH or divalent cation limitation activate PhoQ . 10 . 7554/eLife . 06792 . 006Figure 3 . A disulfide bond between α-helices 2 and 4 inhibits PhoQ activation by acidic pH and divalent cation limitation , but does not inhibit activation by CAMP . PhoQ-dependent phoN::TnphoA alkaline phosphatase activity of ( A ) wild-type and phoQW104C-A128C or ( B ) phoQT48I and phoQT48I W104C-A128C S . enterica Typhimurium strains grown in basal ( pH 7 . 5 ) or activating ( pH 5 . 5 , 10 µM MgCl2 , or CAMP ) N-mm . ( A and B ) The data shown are representatives from at least three independent experiments performed in duplicate and presented as the mean ± SD . Unpaired Student's t-test was performed between wild type and phoQW104C-A128C or phoQT48I and phoQT48I W104C-A128C for all conditions; ( * ) p ≤ 0 . 05 , ( NS ) not significantly different . DOI: http://dx . doi . org/10 . 7554/eLife . 06792 . 00610 . 7554/eLife . 06792 . 007Figure 3—figure supplement 1 . The PhoQW104C-A128C disulfide forms in the Salmonella periplasm and individual mutations at W104 or A128 do not inhibit activation by acidic pH or divalent cation limitation . ( A ) Non-reducing SDS-PAGE and Western blotting of wild-type and phoQW104C-A128C membranes with an anti-PhoQ PD antibody . Membranes were treated with or without sample buffer containing β-mercaptoethanol to show the effect of disulfide reduction on migration rate . ( B , C , and D ) PhoQ-dependent phoN::TnphoA alkaline phosphatase activity of S . enterica Typhimurium strains grown in basal ( pH 7 . 5 ) or activating ( pH 5 . 5 , 10 µM MgCl2 , CAMP ) N-mm . ( B ) Chromosomal wild-type and phoQW104C-A128C S . enterica Typhimurium . ( C ) Wild-type and phoQW104S-A128S S . enterica Typhimurium . ( B and C ) Unpaired Student's t-test was performed between wild type and phoQW104C-A128C or phoQW104S-A128S for all conditions; ( * ) p ≤ 0 . 05 . ( D ) Single cysteine or serine mutations at position W104 and A128 in PhoQ . The data shown are representatives from at least three independent experiments performed in duplicate and presented as the mean ± SD . DOI: http://dx . doi . org/10 . 7554/eLife . 06792 . 00710 . 7554/eLife . 06792 . 008Figure 3—figure supplement 2 . Multiple PhoQ-dependent genes in phoQW104C-A128C Salmonella are induced by CAMP , but not by acidic pH or divalent cation limitation . PhoQ-dependent gene expression from S . enterica Typhimurium strains grown in basal ( pH 7 . 5 ) or activating ( pH 5 . 5 , 10 µM MgCl2 , or CAMP ) N-mm . Gene expression was normalized to rpoD and represented as fold-induction relative to ΔphoQ . The data shown are representatives from at least three independent experiments performed in duplicate and presented as the mean ± SD . DOI: http://dx . doi . org/10 . 7554/eLife . 06792 . 008 A disulfide bond spanning helices α2 and α4 inhibits activation of PhoQW104C-A128C by acidic pH and divalent cation limitation . Given the remarkable phenotype of this mutant , we sought to ascertain whether the PhoQW104C-A128C PD maintains a similar structure to the wild-type PD . A crystal structure of the S . enterica Typhimurium PhoQW104C-A128C PD ( PDB 4UEY ) was solved at 1 . 9 Å resolution ( Figure 4A and Table 1 ) . As predicted , the PhoQW104C-A128C PD formed an intramolecular disulfide bond between W104C and A128C , covalently linking α2 and α4 ( Figure 4A , inset ) . The protomers in the PhoQW104C-A128C PD structure are highly similar to each other , with an average root mean squared deviation ( r . m . s . d . ) of 0 . 3 . Furthermore , the disulfide mutant structure is similar to previously solved structures of wild-type S . enterica Typhimurium ( PDB 1YAX ) and Escherichia coli ( PDB 3BQ8 ) PhoQ PD , with an average r . m . s . d . value of 1 . 07 Å ( Figure 4B ) . These data further demonstrate that the PhoQW104C-A128C PD forms an intramolecular disulfide bond and a structure similar to the wild-type PhoQ PD . 10 . 7554/eLife . 06792 . 009Figure 4 . The PhoQW104C-A128C PD is structurally similar to wild type and has increased stability . ( A ) 1 . 9 Å crystal structure of the S . enterica Typhimurium PhoQW104C-A128C PD ( PDB 4UEY ) . The W104C-A128C disulfide bond ( inset ) is located between α2 and α4 . Secondary structural features are annotated with yellow circles ( NT , N-termini; CT , C-termini ) . ( B ) Structural comparison of the PhoQW104C-A128C PD ( blue ) , the wild-type S . enterica Typhimurium PhoQ PD ( PDB 1YAX , teal ) , and the wild-type E . coli PhoQ PD ( PDB 3BQ8 , purple ) . ( C ) Thermal denaturation of wild-type S . enterica Typhimurium PhoQ PD and PhoQW104C-A128C PD treated with or without TCEP reducing agent monitored by CD spectroscopy at 212 nm . Raw data were normalized to give the fraction unfolded protein assuming a two-state denaturation process . A sigmoidal curve was fit to the processed data . The data shown are representatives from three independent experiments . DOI: http://dx . doi . org/10 . 7554/eLife . 06792 . 00910 . 7554/eLife . 06792 . 010Figure 4—figure supplement 1 . Wild-type and PhoQW104C-A128C PD have similar secondary structure . ( A ) Non-reducing SDS-PAGE of purified wild-type S . enterica Typhimurium PhoQ PD and PhoQW104C-A128C PD treated with or without TCEP reducing agent . ( B ) CD wavelength scan of the wild-type PhoQ PD , PhoQW104C-A128C PD , and PhoQW104C-A128C PD treated with TCEP at 25°C buffered to pH 5 . 5 . DOI: http://dx . doi . org/10 . 7554/eLife . 06792 . 01010 . 7554/eLife . 06792 . 011Table 1 . Crystallographic data collection and refinementDOI: http://dx . doi . org/10 . 7554/eLife . 06792 . 011PhoQW104C-A128C PDData collection Space groupC2 Cell dimensions a , b , c ( Å ) 128 . 04 , 45 . 37 , 81 . 37 α , β , γ ( ° ) 90 , 102 . 53 , 90 Resolution ( Å ) 31 . 3–1 . 9 ( 2 . 01–1 . 90 ) * Rsym or Rmerge0 . 05 ( 0 . 51 ) I/σI12 . 9 ( 1 . 6 ) Completeness ( % ) 97 . 8 ( 93 . 8 ) Redundancy3 . 6 ( 3 . 2 ) Refinement Resolution ( Å ) 31 . 3–1 . 90 ( 1 . 95–1 . 90 ) * No . reflections35 , 633 Rwork/Rfree0 . 23/0 . 26 ( 0 . 38/0 . 44 ) No . atoms ( all ) Protein3391 Water138 Ca2+– B-factors Protein44 . 8 Water40 . 6 R . m . s . deviations Bond lengths ( Å ) 0 . 007 Bond angles ( ° ) 1 . 2Ramachandran statistics Residues in favored region no ( % ) 409 ( 98 . 3 ) Residues in allowed region no ( % ) 7 ( 1 . 7 ) Residues in outlier region no ( % ) 0 ( 0 ) PDB-entry4UEYCrystallization conditions0 . 1 M Bis-Tris pH 6 . 5 , 200 mM MgCl2 , 25% Peg3350*Values in parentheses are for highest-resolution shell . We hypothesize that the W104C-A128C disulfide may stabilize conformational dynamics between α4 and α5 and the α/β-core , preventing acidic pH from promoting a flexible , active state . This hypothesis was tested by performing thermal melts on purified PhoQ PD and PhoQW104C-A128C PD at pH 5 . 5 by following the CD signal of each protein as a function of temperature . We first confirmed that purified PhoQW104C-A128C PD forms a disulfide as visualized as a shift in SDS-PAGE migration rate relative to PhoQ PD and TCEP-reduced PhoQW104C-A128C ( Figure 4—figure supplement 1A ) . The CD spectra revealed that the PhoQ PD and PhoQW104C-A128C PD with or without TCEP are folded and have relatively similar secondary structure at pH 5 . 5 and 25°C ( Figure 4—figure supplement 1B ) . Thermal denaturation of the PhoQ PD at pH 5 . 5 proved to be irreversible . Therefore , we reported the apparent transition temperatures ( Tmapp ) . While the wild-type PhoQ PD unfolded with a Tmapp of 56°C in the presence and absence of TCEP ( Figure 4C ) , the PhoQW104C-A128C PD had a significantly increased Tmapp of 75°C . When reduced with TCEP , the PhoQW104C-A128C PD was slightly destabilized relative to the wild type , with a Tmapp 52°C . Therefore , the W104C-A128C disulfide increased the intrinsic stability of the PD at pH 5 . 5 relative to wild type . Furthermore , the observations that reduced PhoQW104C A128C PD is less stable than wild type and that phoQW104S A128S bacteria had increased PhoQ-dependent gene reporter activity relative to wild type ( Figure 3—figure supplement 1C ) suggests that substituting small polar side-chains at these positions in the PD results in decreased stability and increased PhoQ activity . Combined , these results suggest that the mechanism by which the W104C-A128C disulfide bond inhibits PhoQ activation by acidic pH and divalent cation limitation involves a loss of conformational flexibility between α4 and α5 and the α/β-core . Prior to this study , the contribution of specific stimuli to PhoQ-mediated bacterial virulence was difficult to ascertain as mutants that only respond to individual signals were not available . With the construction of phoQW104C-A128C S . enterica Typhimurium , the significance of acidic pH and divalent cation sensing by PhoQ to virulence could be directly determined independently of CAMP sensing . Thus , BALB/c mice were infected by the intraperitoneal ( IP ) route with wild-type , phoQW104C-A128C , or phoQ null ( ΔphoQ ) bacteria and splenic bacterial burden was determined at 48- and 96-hpi ( Figure 5A , solid lines ) . Similar to infection with wild-type bacteria , mice infected with phoQW104C-A128C bacteria had increased splenic bacterial burden relative to those infected with ΔphoQ bacteria . The equivalent experiment was performed in resistant A/J mice to determine if the virulence phenotype observed for mice infected with phoQW104C-A128C bacteria was due to the susceptible BALB/c mouse genetic background and to determine whether a subtle fitness defect would be exposed on infection of a relatively resistant inbred mice . Infecting A/J mice revealed the same relative phenotypes for wild-type , phoQW104C-A128C , and ΔphoQ bacteria , although , as expected , bacterial burden was lower compared to infected BALB/c mice ( Figure 5A , dotted lines ) . These results indicate that PhoQ sensing of acidic pH and divalent cation limitation are dispensable for systemic virulence of S . enterica Typhimurium in susceptible and relatively resistant inbred mice . 10 . 7554/eLife . 06792 . 012Figure 5 . phoQW104C-A128C Salmonella survive within host organisms and exhibits PhoQ-dependent gene expression within macrophage . ( A ) Individual S . enterica Typhimurium strains administered IP to BALB/c ( solid lines ) or A/J ( dotted lines ) mice . The inoculum is shown at T = 0 hpi . Spleens were harvested and bacterial burden quantified . ( B ) Competition between S . enterica Typhimurium strains administered IP to BALB/c mice . Spleens were harvested , bacteria quantified 48-hpi and CI determined . ( A and B ) The data shown are representatives from at least three independent experiments performed in quintuplet and presented as the mean ± SD . ( C ) BALB/c BMMΦ infected with strains of S . enterica Typhimurium . Bacteria were harvested and quantified at the indicated time-points . The inoculum is shown at T = 0 hpi . The data shown are representatives from at least three independent experiments performed in triplicate and presented as the mean ± SD . ( D ) PhoQ-dependent gene expression from S . enterica Typhimurium strains within BALB/c BMMΦ 4-hpi . Gene expression was normalized to rpoD and presented as fold-induction relative to ΔphoQ . The data shown are representatives from at least three independent experiments and presented as the mean ± SD . ( A , B , C , and D ) Unpaired Student's t-test was performed between all strains ( bar ) for each time-point or gene . Symbols for significant difference; ( ¤ ) wild type and phoQW104C-A128C are not significantly different from each other ( p ≥ 0 . 05 ) , but are significantly different from ΔphoQ ( p ≤ 0 . 05 ) , ( * ) all strains are significantly different from each other ( p ≤ 0 . 05 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 06792 . 01210 . 7554/eLife . 06792 . 013Figure 5—figure supplement 1 . Acidic pH and divalent cation sensing by PhoQ are dispensable for PO systemic competition of S . enterica Typhimurium . Competition between S . enterica Typhimurium strains administered PO to BALB/c mice . Spleens were harvested , bacteria quantified 96-hpi and CI determined . The data shown are from three independent experiments and presented as the mean ± SD . Data points on the x-axis represent samples with a CI of zero . DOI: http://dx . doi . org/10 . 7554/eLife . 06792 . 01310 . 7554/eLife . 06792 . 014Figure 5—figure supplement 2 . The in vitro growth rate of wild-type Salmonella is decreased relative to phoQW104C-A128C and ΔphoQ when grown at pH 5 . 5 . S . enterica Typhimurium strains were grown in N-mm buffered to pH 7 . 5 ( closed symbols ) or pH 5 . 5 ( open symbols ) supplemented with 1 mM MgCl2 . Bacterial growth was monitored by OD600 at the indicated time-points . The data shown are representatives from at least three independent experiments performed in duplicate and presented as the mean ± SD . DOI: http://dx . doi . org/10 . 7554/eLife . 06792 . 014 The importance of PhoQ activation by acidic pH and divalent cation limitation for systemic infection was also assessed by competing phoQW104C-A128C bacteria with wild-type S . enterica Typhimurium in IP or peroral ( PO ) infections of BALB/c mice . The splenic bacterial competitive index ( CI ) for wild-type , phoQW104C-A128C , and ΔphoQ bacteria was determined for both IP and PO infections at 48-hpi or 96-hpi , respectively ( Figure 5B and Figure 5—figure supplement 1 ) . Consistent with the single strain infections , phoQW104C-A128C demonstrated no reduction in CI and , in contrast , was more competitive than wild type . ΔphoQ showed the expected reduction in CI . Altogether , these data indicate that PhoQ activation by acidic pH and divalent cation limitation are dispensable for S . enterica Typhimurium to out compete strains with these capabilities during systemic infection of susceptible mice . Furthermore , the observation that IP and PO administered phoQW104C-A128C S . enterica Typhimurium have similar competitive indices suggests that acidic pH and divalent cation sensing by PhoQ are not required for survival or spread from the gastrointestinal tract to deep tissue sites . S . enterica Typhimurium is growth restricted in cultured fibroblasts and nonphagocytic stromal cells in the murine lamina propria via PhoPQ-dependent processes ( Cano et al . , 2001; Nunez-Hernandez et al . , 2013 ) . Thus , it was plausible that the competitive advantage observed for phoQW104C-A128C relative to wild-type bacteria was due to an increased growth rate resulting from the loss of acidic pH sensing by PhoQ . When grown in N-mm pH 5 . 5 , wild-type bacterial growth rate was decreased relative to phoQW104C-A128C and ΔphoQ ( Figure 5—figure supplement 2 ) . Conversely , wild type , phoQW104C-A128C , and ΔphoQ grown in N-mm pH 7 . 5 had similar growth kinetics . These data provide evidence that acidic pH sensing by PhoQ reduces S . enterica Typhimurium growth rate in vitro and correlates with the in vivo competitive advantage that was observed for phoQW104C-A128C bacteria within mice spleens . The contribution of acidic pH and divalent cation limitation as signals for PhoQ-mediated bacterial intracellular survival within macrophages was evaluated by measuring S . enterica Typhimurium survival after infection of bone-marrow derived macrophages ( BMMΦ ) from BALB/c mice . BMMΦ were infected with-wild type , phoQW104C-A128C , or ΔphoQ S . enterica Typhimurium strains and bacterial burden was determined at 2- , 4- , 8- , and 24-hpi ( Figure 5C ) . No difference in bacterial burden was observed between wild type , phoQW104C-A128C , and ΔphoQ at 2- or 4-hpi . At 8- and 24-hpi , bacterial burden for ΔphoQ was decreased relative to wild type and phoQW104C-A128C . Importantly , bacteria with the phoQW104C-A128C allele maintained at or above wild-type bacterial levels throughout infection . These data indicate that activation of PhoQ by acidic pH and divalent cation limitation are dispensable for S . enterica Typhimurium survival within BMMΦ from inbred mice . The discovery of the phoQW104C-A128C phenotype allowed for the unique opportunity to examine the contribution of acidic pH and divalent cation sensing to PhoQ-dependent gene expression during infection of macrophages . Therefore , BMMΦ from BALB/c mice were infected with wild-type , phoQW104C-A128C , or ΔphoQ S . enterica Typhimurium . Following incubation , PhoQ-dependent gene expression was determined for intracellular S . enterica Typhimurium ( Figure 5D ) . Wild-type bacteria experienced 41- , 29- , 36- , and 51-fold increases in gene expression for pagD , pagO , phoN , and phoP , whereas phoQW104C-A128C experienced increases of 12- , 9- , 9- , and 33-fold , respectively , relative to ΔphoQ bacteria . These data may indicate that wild-type acidic pH or divalent cation sensing contribute an approximate threefold to fourfold increase in PhoQ-dependent gene expression relative to phoQW104C-A128C; however , a significant amount of gene expression ( ≥ninefold ) appeared to be independent of acidic pH or divalent cation sensing . These findings are consistent with in vitro results which show that acidic pH and CAMP are additive signals for PhoQ ( Prost et al . , 2007 ) . These data indicate that maximal PhoQ-dependent gene expression in macrophages requires acidic pH or divalent cation sensing . Furthermore , these findings reveal that the phoQW104C-A128C allele promotes significant induction of PhoQ-dependent gene expression , suggesting CAMP or alternative host factors , other than acidic pH and divalent cation limitation , are a major signal for S . enterica Typhimurium within BMMΦ vacuoles .
Salmonellae encounter changing environments within the macrophage phagosome and other mammalian host sites during infection . These environments include a variety of antimicrobial factors for which the bacteria must regulate inducible resistance mechanisms in order to survive . These bacterial resistance mechanisms are essential for successful infection , necessitating tight regulation by sensors such as PhoQ . Our study defines α4 and α5 in the PhoQ PD as a pH-responsive structural element that experiences a change in its dynamic behavior upon transition to acidic pH . Furthermore , mutations within the PhoQ PD , predicted to destabilize hydrophobic packing and hydrogen bonding between the α/β-core and α4 and α5 , resulted in loss of PhoQ repression . Limiting flexibility between these structural elements by introduction of a disulfide bond inhibited PhoQ activation by acidic pH and divalent cation limitation . We suggest that PhoQ has evolved α4 and α5 as a unique pH-responsive structural element within the PD , effectively replacing the ligand-binding site that is often found in a similar location in other structurally related PDC sensor domains ( Cho et al . , 2006; Cheung and Hendrickson , 2008 ) . This study provides insights for a refined model of PhoQ activation ( Figure 6 ) . At neutral pH and millimolar divalent cation concentration ( Figure 6 , left ) , the PhoQ PD is anchored to the inner membrane in a repressed state via cation-bridges , a rigid α/β-core and TDK network , and quiescent α4 and α5 . Acidic pH or divalent cation limitation promotes a change in α4 and α5 from a stable to a dynamic state ( Figure 6 , middle ) . Acidic pH-induced flexibility in α4 and α5 may destabilize divalent cation salt-bridges between the inner membrane and acidic patch , thereby promoting a loss of divalent cation-mediated repression . Lack of divalent cation salt-bridges between the PhoQ PD acidic patch and inner membrane due to divalent cation limitation may result in electrostatic repulsion between the acidic patch and inner membrane , releasing α4 and α5 , and favoring a more flexible state in this structural element . Changes in the relationship between α4 and α5 and the α/β-core surrounding H157 are transmitted to the dimerization interface and TDK network proximal to the membrane resulting in alterations of the transmembrane domain , cytoplasmic HAMP domain , and ultimately resulting in increased PhoQ kinase activity . 10 . 7554/eLife . 06792 . 015Figure 6 . Model of PhoQ activation and repression . ( Left ) At neutral pH and millimolar divalent cation concentration , the PhoQ PD is maintained in a repressed conformation due to rigidified interactions between the α/β-core ( yellow spheres ) , α4 and α5 , and salt-bridges ( bronze spheres ) formed between the acidic patch ( red spheres ) and inner membrane . ( Middle ) Transition to a mildly acidic ( left protomer ) or divalent cation limited ( right protomer ) environment promotes flexibility in α4 and α5 ( bent arrows ) and conformational dynamics in the α/β-core surrounding H157 ( teal spheres ) . Movement in α4 and α5 due to acidic pH or divalent cation limitation destabilizes salt-bridges between the acidic patch and inner membrane perturbing the TDK network ( blue spheres ) resulting in activation . ( Right ) CAMP ( magenta helices ) intercalates into the inner membrane and promotes PhoQ activation by directly interacting with the PhoQ transmembrane domains and/or by disrupting local phospholipid packing ( left protomer ) and/or by overcoming constraints in α4 and α5 ( right monomer , bent arrow ) . DOI: http://dx . doi . org/10 . 7554/eLife . 06792 . 015 Our model of PhoQ activation and repression has similarities to a recently proposed two-state computational model in which the PD is predicted to experience broad conformational changes within the periplasmic dimerization interface and acidic patch ( Molnar et al . , 2014 ) . This model is consistent with predictions previously made in relation to the discovery of the divalent cation bridges between the PhoQ acidic patch and negatively charged membrane phospholipids ( Bader et al . , 2005; Cho et al . , 2006 ) . Molnar et al suggest that the PhoQ PD assumes alternative conformations as the acidic patch moves away from the membrane in the absence of divalent cation . Our findings that restricting movement in α4 and α5 inhibits PhoQ activation by acidic pH and divalent cation limitation supports the model that the acidic patch and α4 and α5 must remain dynamic for proper signaling . Furthermore , our observations that PhoQ activation by acidic pH and divalent cation limitation are separable from CAMP-mediated activation may indicate that distinct conformational states exist for each of the unique PhoQ-activating and -repressing stimuli . Results presented here indicate that α4 and α5 within the PhoQ PD do not adopt a distinct conformation under activating pH conditions , but rather exchange between a rigid , repressed state and an ensemble of conformations that together constitute the acidic pH-activated state . Unlike other pH-sensors that utilize discrete histidine protonation as a mechanism of activation ( Perier et al . , 2007; Dawson et al . , 2009; Müller et al . , 2009; Choi et al . , 2013; Williamson et al . , 2013 ) , PhoQ activation by acidic pH does not appear to rely strictly on protonation of histidines . Mutagenesis of H157 , which is located in the PhoQ PD α/β-core and is observed to form hydrogen bonds with T129 on α4 and the backbone hydroxyl of T180 on the β7-α6 loop , results in a modest increase in activity; however , it does not account for the entire pH-mediated activation state ( Prost et al . , 2007 ) . Mutation of the other two histidines in the PD ( H137 and H120 ) did not significantly affect activation or repression of PhoQ ( data not shown ) . An alternate mechanism of PhoQ activation by acidic pH may involve pH-induced conformational changes within the periplasmic dimerization interface . Notably , an intermolecular disulfide at position T61C in the PhoQ PD dimerization interface results in increased PhoQ-dependent reporter activity at pH 5 . 5 , suggesting that conformational changes in the dimerization interface can directly affect signal transduction ( data not shown ) . Furthermore , analytical ultracentrifugation analysis revealed that the PhoQ PD dimer dissociates as the pH decreases ( data not shown ) . Additionally , work by Molnar et al support the notion that conformational changes with the dimerization interface are concomitant with activation and repression . Recently , it was shown that the Helicobacter pylori chemotaxis receptor , TlpB , utilizes a unique mechanism to sense acidic pH ( Goers Sweeney et al . , 2012 ) . Interestingly , the TlpB PD may sense pH by adopting a ‘relaxed’ conformation at low pH due to decreased hydrogen bonding to a coordinated urea molecule . It is plausible that similar relaxation may occur in the PhoQ PD between the α/β-core and helices α4 and α5 upon exposure to acidic pH . Therefore , pH-induced conformational changes resulting in structural relaxation or flexibility may be an important mechanism by which pH can be sensed . Our previous work suggests that PhoQ activation by acidic pH and CAMP proceed via different mechanisms ( Bader et al . , 2005; Prost et al . , 2007 ) . In this study , we have shown that activation by acidic pH and divalent cation limitation are separable from CAMP-mediated activation by rigidifying the interaction between α4 and α5 and the α/β-core . Perhaps , CAMP circumvents α4 and α5 and activates PhoQ via direct interactions within the transmembrane regions adjacent to the acidic patch or disrupts local phospholipid packing promoting conformational changes in the periplasmic and transmembrane domains ( Figure 6 , right ) . Alternatively , it is plausible that CAMP functions as a large steric ‘wedge’ . In this scenario , CAMP is recruited to the acidic patch of PhoQ resulting in conformational changes in the PD overcoming any repressive structural constraints between α4 and α5 and the α/β-core . Determining the host signals which activate PhoQ has been difficult; past investigations have relied on alterations to host processes via chemical inhibitors to neutralize acidic compartments or targeted mutagenesis to remove known PhoQ-activating antimicrobial peptides from host animals ( Alpuche Aranda et al . , 1992; Martin-Orozco et al . , 2006; Richards et al . , 2012 ) . Although informative , these studies do not account for unintended host-cell changes due to chemical neutralization of acidic vesicles and organelles or uncharacterized host peptides or molecules which may activate the system . For example , distinct pH-gradients are established and required in eukaryotic vesicular trafficking pathways and chemical neutralization of acidic compartments within these pathways results in a variety of cellular disturbances including inhibition of acidic hydrolases and proteases , perturbation of molecular sorting and recycling , endocytosis and exocytosis dysregulation , and disruption of vesicular fusion events ( Dean et al . , 1984; Mellman et al . , 1986; Casey et al . , 2010 ) . Furthermore , chemical neutralization of the SCV is detrimental to intracellular S . enterica Typhimurium as it results in decreased bacterial survival ( Rathman et al . , 1996 ) , presumably due to lack of virulence factor expression . Therefore , it is plausible that the use of chemical inhibitors to block or neutralize acidic host-compartments inhibits processes involved in CAMP maturation or trafficking , thereby preventing activation of PhoQ . It is plausible that undefined host molecules or conditions , that require defined pH-gradients , activate PhoQ in vivo . Our observations that activation by acidic pH and divalent cation limitation are not required for significant increases in PhoQ-dependent gene expression in BALB/c macrophage and that PhoQ-dependent gene expression is induced in CRAMP-deficient macrophage ( Richards et al . , 2012 ) indicate that uncharacterized host factors , which are likely to be a variety of different cationic antimicrobial molecules , activate PhoQ . Furthermore , multiple host peptides may activate PhoQ in vivo as various host and synthetically derived cationic peptides can activate the system; this is consistent with the large PhoQ PD acidic patch which likely evolved to bind diverse CAMP ( Bader et al . , 2003 , 2005; Shprung et al . , 2012 ) . From these observations , it is reasonable to speculate that the acidic patch , located on the α4/α5 structural unit within the PhoQ PD of bacteria that interact with animals , evolved to sense a variety of cationic peptides , as the PhoQ PD from environmental bacteria such as Pseudomonas aeruginosa lack these important structural features ( Prost et al . , 2008 ) . Though it is tempting to speculate that CAMP may be the dominant PhoQ-stimulant during systemic infection , it is important to remember that sensing may be redundant or host-compartment specific . Further experiments will need to be performed to examine the contribution of acidic pH and divalent cation sensing to PhoQ-mediated bacterial survival during transition from the intestinal tract to systemic environments and determine if ‘bacterial innate immunity’ or the recognition of multiple mammalian signals is redundant . Additionally , perhaps acidic pH and divalent cation sensing by PhoQ are functions required for survival in ex vivo environments , beyond animal hosts . The work in this study led to the construction of a specific S . enterica Typhimurium PhoQ mutant that is inhibited for activation by acidic pH or divalent cation limitation . This mutant has wild-type virulence in susceptible and resistant mouse models of systemic infection suggesting that , at least in these models , acidic pH and divalent cation sensing are dispensable for virulence . Although we have shown that the PhoQW104C-A128C disulfide forms in purified proteins and in bacteria grown in culture , we did not measure disulfide formation for W104C-A128C within host tissues . Host compartments replete with strong oxidizing agents , such as the macrophage phagosome , could potentially disrupt disulfide bond formation . However , multiple Salmonella virulence factors and homeostatic processes that occur in the macrophage phagosome require disulfide bond formation , indicating that S . enterica has robust mechanisms to regulate redox potential and resist hyperoxidation of thiols in the periplasm ( Ellermeier and Slauch , 2004; Miki et al . , 2004; Lippa and Goulian , 2012 ) . Additionally , we have observed altered PhoQ-dependent gene expression from phoQW104C-A128C S . enterica Typhimurium within macrophage phagosomes similar to in vitro grown bacteria . Therefore , it is highly likely that efficient formation of the W104C-A128C disulfide bond occurs inside host compartments . In summary , we provide novel detail to the mechanism by which S . enterica Typhimurium PhoQ is activated by acidic pH . We have identified residues and secondary structural elements within the PD which contribute to acidic pH sensing and are important for PhoQ signal transduction . Furthermore , structural studies have led to the engineered bifurcation of PhoQ signaling capabilities; separating acidic pH and divalent cation sensing from CAMP signaling . This discovery has allowed us to determine the contribution of acidic pH and divalent cation sensing to S . enterica Typhimurium virulence and will provide valuable insights to the spatial-temporal regulation of PhoQ during pathogenesis .
Bacterial strains , plasmids , and primers used in this study can be found in Tables 2 , 3 . S . enterica Typhimurium strain 14028s was the wild-type strain used in this study and all subsequent strains and mutants were derived from this strain . Unless otherwise stated , all alkaline phosphatase activity assays were performed in the CS1081 background with CS1084 as the wild-type control and various alleles of phoQ basally expressed from pBAD24 . Alkaline phosphatase activity assays were also performed on wild type ( KH127 ) and phoQW104C-A128C ( KH130 ) recombined on to the chromosome of CS1081 . Bacterial strains were grown in either LB broth or modified N-mm as indicated . Activation of the phoN::TnphoA reporter was utilized as previously described ( Bader et al . , 2005; Prost et al . , 2007 ) . Briefly , bacterial strains were grown overnight in modified N-minimal media pH 7 . 5 containing 1 mM MgCl2 . In the morning , cultures were washed once in the appropriate media and diluted 1:100 in to fresh modified N-minimal media containing either 10 μM , 1 mM , or 10 mM MgCl2 and buffered with either 0 . 1 M Tris or 0 . 1 M MES to pH 7 . 5 or pH 5 . 5 , respectively . Unless stated otherwise , the base growth media is N-minimal media pH 7 . 5 supplemented with 1 mM MgCl2 and 100 μg∙ml−1 ampicillin . Following dilution into fresh media , cultures were grown for 5 hr shaking at 37°C . To study phoN::TnphoA reporter activation in the presence of CAMP , overnight cultures were washed once in N-minimal media pH 7 . 5 containing 1 mM MgCl2 and diluted 1:100 into the same growth media . Cultures were then grown to OD600 0 . 2 , treated with 5 μg∙ml−1 of C18G peptide ( Anaspec , Fremont , CA ) , and grown shaking at 37°C for 90 min . Following incubation , alkaline phosphatase activity was measured . Alkaline phosphatase activity assays were performed according to standard protocol on cultures grown in duplicate and repeated on at least three independent occasions . 10 . 7554/eLife . 06792 . 016Table 2 . Strains and plasmids used in this studyDOI: http://dx . doi . org/10 . 7554/eLife . 06792 . 016StrainDescriptionSourceCS09314028s wild type S . enterica TyphimuriumATCCCS1081CS093 phoQ::TPOP phoN::TnphoABader et al . , 2005CS1083CS1081 pBAD24Bader et al . , 2005CS1084CS1081 pBAD24-phoQBader et al . , 2005CS1399CS1081 pBAD24-phoQI88NThis workCS1400CS1081 pBAD24-phoQY89NThis workKH45CS1081 pBAD24-phoQI102CThis workKH140CS1081 pBAD24-phoQL105DThis workCS1402CS1081 pBAD24-phoQT124NThis workCS1403CS1081 pBAD24-phoQV126EThis workCS1404CS1081 pBAD24-phoQT129IThis workCS1405CS1081 pBAD24-phoQT131PThis workCS1406CS1081 pBAD24-phoQL132PThis workKH28CS1081 pBAD24-phoQL133CThis workCS1407CS1081 pBAD24-phoQD150GThis workCS1408CS1081 pBAD24-phoQA153PThis workCS1409CS1081 pBAD24-phoQM155VThis workCS1410CS1081 pBAD24-phoQV178DThis workCS1374CS1081 pBAD24-phoQW104CThis workCS1386CS1081 pBAD24-phoQA128CThis workCS1382CS1081 pBAD24-phoQW104C-A128CThis workKH48CS1081 pBAD24-phoQW104SThis workKH49CS1081 pBAD24-phoQA128SThis workKH50CS1081 pBAD24-phoQW104S A128SThis workCS1101BL21 pET11a-phoQ 45-190- ( His ) 6Bader et al . , 2005KH85NEB SHuffle T7 express pET11a-phoQW104C-A128C 45-190- ( His ) 6This workKH23phoQ::tetRAThis workKH163phoQW104C-A128CThis workCS1350ΔphoQProst et al . , 2008KH127phoQ phoN105::TnphoAThis workKH130phoQW104C-A128C phoN105::TnphoAThis workKH111CS093 pWSK129KanThis workKH112CS093 pWSK29AmpThis workKH113phoQW104C-A129C pWSK29AmpThis workKH114ΔphoQ pWSK29AmpThis work10 . 7554/eLife . 06792 . 017Table 3 . Primer sequences used in this studyDOI: http://dx . doi . org/10 . 7554/eLife . 06792 . 017Primer # ( name ) Sequence ( 5′–3′ ) LP135 ( RM_Fwd ) CTGGTCGGCTATAGCGTAAGTTTTGLP136 ( RM_Rev ) CACGTATACGAACCAGCTCCACACLP178 ( I88N_Fwd ) CGACCATGACGCTGAATTACGATGAAACGGLP179 ( I88N_Rev ) CCGTTTCATCGTAATTCAGCGTCATGGTCGLP180 ( Y89N_Fwd ) CCATGACGCTGATTAACGATGAAACGGGCLP181 ( Y89N_Rev ) GCCCGTTTCATCGTTAATCAGCGTCATGGKH81 ( I102C_Fwd ) GACGCAGCGCAACTGTCCCTGGCTGATTAAAAGKH82 ( I102C_Rev ) CTTTTAATCAGCCAGGGACAGTTGCGCTGCGTCLP184 ( T124N_Fwd ) CTTCCATGAAATTGAAAACAACGTAGACGCCACCLP185 ( T124N_Rev ) GGTGGCGTCTACGTTGTTTTCAATTTCATGGAAGLP186 ( V126E_Fwd ) GAAATTGAAACCAACGAAGACGCCACCAGCACLP187 ( V126E_Rev ) GTGCTGGTGGCGTCTTCGTTGGTTTCAATTTCLP188 ( T129I_Fwd ) CAACGTAGACGCCATCAGCACGCTGTTGLP189 ( T129I_Rev ) CAACAGCGTGCTGATGGCGTCTACGTTGKH192 ( L105D_Fwd ) GCGCAACATTCCCTGGGATATTAAAAGCATTCAACKH193 ( L105D_Rev ) GTTGAATGCTTTTAATATCCCAGGGAATGTTGCGCLP190 ( L131P_Fwd ) CAACGTAGACGCCACCAGCCCACTGTTGAGCGAAGACCATTCLP191 ( L131P_Rev ) GAATGGTCTTCGCTCAACAGTGGGCTGGTGGCGTCTACGTTGLP192 ( L132P_Fwd ) GACGCCACCAGCACGCCATTGAGCGAAGACCATTCLP193 ( L132P_Rev ) GAATGGTCTTCGCTCAATGGCGTGCTGGTGGCGTCKH85 ( L133C_Fwd ) CACCAGCACGCTGTGTAGCGAAGACCATTCKH86 ( L133C_Rev ) GAATGGTCTTCGCTACACAGCGTGCTGGTGLP194 ( D150G_Fwd ) GTACGTGAAGATGGCGATGATGCCGAGLP195 ( D150G_Rev ) CTCGGCATCATCGCCATCTTCACGTACLP196 ( A153P_Fwd ) GAAGATGACGATGATCCCGAGATGACCCACLP197 ( A153_Rev ) GTGGGTCATCTCGGGATCATCGTCATCTTCLP198 ( M155V_Fwd ) GACGATGATGCCGAGGTAACCCACTCGGTAGCLP199 ( M155V_Rev ) GCTACCGAGTGGGTTACCTCGGCATCATCGTCLP200 ( V178D_Fwd ) CCATCGTGGTGGACGATACCATTCCGLP201 ( V178D_Rev ) CGGAATGGTATCGTCCACCACGATGGLP141 ( W104C_Fwd ) GCGCAACATTCCCTGCCTGATTAAAAGCATTCLP142 ( W104C_Rev ) GAATGCTTTTAATCAGGCAGGGAATGTTGCGCLP145 ( A128C_Fwd ) GAAACCAACGTAGACTGCACCAGCACGCTGTTGLP146 ( A128C_Rev ) CAACAGCGTGCTGGTGCAGTCTACGTTGGTTTCKH61 ( W104S_Fwd ) CAGCGCAACATTCCCAGCCTGATTAAAAGCATTCKH62 ( W104S_Rev ) GAATGCTTTTAATCAGGCTGGGAATGTTGCGCTGKH63 ( A128S_Fwd ) GAAACCAACGTAGACAGCACCAGCACGCTGTTGKH64 ( A128S_Rev ) CAACAGCGTGCTGGTGCTGTCTACGTTGGTTTCLP164 ( T48C_Fwd ) GTAAGTTTTGATAAAACCTGCTTTCGTTTGCTGCGCGLP165 ( T48C_Rev ) CGCGCAGCAAACGAAAGCAGGTTTTATCAAAACTTACLP168 ( K186C_Fwd ) CCATTCCGATAGAACTATGCCGCTCCTATATGGTGTGLP169 ( K186C_Rev ) CACACCATATAGGAGCGGCATAGTTCTATCGGAATGGKH35 ( T48S_Fwd ) GTTTTGATAAAACCAGCTTTCGGCTGCGKH36 ( T48S_Rev ) CGCAGCAAACGAAAGCTGGTTTTATCAAAAKH39 ( K186S_Fwd ) CATTCCGATAGAACTAAGTCGCTCCTATATGGTGKH40 ( K186S_Rev ) CACCATATAGGAGCGACTTAGTTCTATCGGAATGKH45 ( PhoQ_tetRA_knock-in_Fwd ) GAATAAATTTGCTCGCCATTTTCTGCCGCTGTCGCTGCGGTTAAGACCCACTTTCACAKH46 ( PhoQ_tetRA_knock-in_Rev ) CCTCTTTCTGTGTGGGATGCTGTCGGCCAAAAACGACCTCCTAAGCACTTGTCTCCTGKH93 ( ST-PhoQ_N-term_Fwd ) ATGAATAAATTTGCTCGCCATTTTCKH94 ( ST-PhoQ_N-term_Rev ) TTATTCCTCTTTCTGTGTGGGKH265 ( ST-rpoD_Fwd_qRT ) GGGATCAACCAGGTTCAATGKH266 ( ST-rpoD_Rev_qRT ) GGACAAACGAGCCTCTTCAGKH269 ( ST-pagD_Fwd_qRT ) GTTCAGGCCATTGTTCTGGTKH270 ( ST-pagD_Rev_qRT ) TAATCTGCCTGGCTTGCTTTKH273 ( ST-pagO_Fwd_qRT ) CGGGCTTAACTATCGCAATCKH274 ( ST-pagO_Rev_qRT ) CAGCAGAAATAAGCGCAGTGKH275 ( ST-phoP_Fwd_qRT ) TGCCAGGGAAGCTGATTACTKH276 ( ST-phoP_Rev_qRT ) CAGCGGCGTATTAAGGAAAGKH277 ( ST-phoN_Fwd_qRT ) CCGGCTTACCGCTATGATAAKH278 ( ST-phoN_Rev_qRT ) CGCTTACATCTGCATCCTCA All PhoQ alleles with point-mutations were generated on pBAD24-phoQ or pET11a-phoQ using the appropriate primers pairs ( Table 3 ) and a standard site-directed mutagenesis protocol or Gibson assembly ( Gibson et al . , 2009 ) . To generate phoQW104C-A128C on the S . enterica Typhimurium 14028s chromosome , lambda red allelic exchange methods were utilized ( Gerlach et al . , 2007 ) . Briefly , to engineer phoQW104C-A128C on the S . enterica Typhimurium chromosome , a tetracycline resistant cassette ( tetRA ) was amplified using primers KH45 and KH46 . The resulting phoQ::tetRA amplicon was recombined into CS093 generating phoQ::tetRA ( KH23 ) . Primers KH93 and KH94 were used to amplify phoQW104C-A128C from pBAD24-phoQW104C-A128C ( CS1382 ) . The phoQW104C-A128C amplicon was recombined into phoQ::tetra ( KH23 ) . Positive clones for phoQW104C-A128C recombination were identified via Bochner selection ( Bochner et al . , 1980 ) . Chromosomal phoQW104C-A128C was then transduced into a clean S . enterica Typhimurium 14028s background via P22 phage transduction . phoQW104C-A128C positive clones were confirmed via DNA sequencing . To identify residues in the PhoQ PD that when mutated result in increased phoN::TnphoA activity , we performed a random mutagenesis screen as previously described ( Cho et al . , 2006 ) . Briefly , pBAD24-phoQ was randomly mutagenized using primers LP135 and LP136 to introduce one mutation per 500 bp in the phoQ PD using the GeneMorph II EZClone Domain Mutagenesis Kit ( Agilent Technologies , Santa Clara , CA ) . The resulting mutagenized pBAD24-phoQ plasmids were transformed into CS1081 and grown overnight on LB plates containing XP substrate ( Sigma 104 , Sigma-Aldrich Corp . , St . Louis , MO ) , ampicillin 100 μg∙ml−1 , and 10 mM MgCl2 . In the morning , plates were screened for blue colonies indicative of phoN::TnphoA alkaline phosphatase activity and PhoQ activation by divalent cation limitation . Approximately , 50 , 000 colonies were screened . 103 blue colonies were chosen and sequenced , yielding 26 single amino acid substitutions . Mutations identified in the screen were independently engineered in a clean pBAD24-phoQ background using the appropriate primers pairs ( Table 3 ) and a standard site-directed mutagenesis protocol . Phenotypes were confirmed by alkaline phosphatase activity assays in duplicate on at least three separate occasions . The PhoQ PD ( strain CS1101 ) was purified as previously described ( Bader et al . , 2005 ) . PhoQW104C-A128C PD ( strain KH85 ) was purified from SHuffle T7 Express E . coli ( NEB , Ipswich , MA ) . Purification and storage of PhoQW104C-A128C PD was performed according to the same methods as wild-type PhoQ PD . Disulfide bond formation in PhoQW104C-A128C PD was confirmed by SDS-PAGE . Briefly , strain CS1101 or KH85 were grown in LB media supplemented with 100 mg∙l−1 ampicillin for all non-labeling experiments . For 15N-labeling and NMR experiments , strain CS1101 was grown in MOPS minimal medium supplemented with 100 mg∙l−1 ampicillin and 1 g∙l−1 15N-ammonium chloride . Expression strains were grown to mid-log phase and IPTG was added to 0 . 5 mM . Cultures were induced for 4–6 hr , harvested by centrifugation and lysed using a French Pressure Cell . Inclusion bodies were isolated by centrifugation , washed once in 50 mM sodium phosphate pH 8 . 0 300 mM NaCl , resuspended in 20 mM sodium phosphate pH 8 . 0 100 mM NaCl 7 M urea , and incubated on ice for 1 hr . Samples were then ultracentrifuged at 50 , 000 rpm for 30 min . The supernatant was rapidly diluted into ice cold 20 mM sodium phosphate pH 8 . 0 . Samples were filtered and purified using a 5 ml HisTrap HP nickel column ( GE Healthcare Bio-Sciences , Pittsburgh , PA ) according to standard protocol . Purified protein was then applied to a Superdex-200 gel filtration column ( GE Healthcare Bio-Sciences ) equilibrated with 20 mM sodium phosphate pH 6 . 5 150 mM NaCl 0 . 1 mM EDTA . PhoQ containing fractions were pooled , concentrated to approximately 0 . 25 mM and stored at −80°C in 10% glycerol . PhoQW104C-A128C PD was expressed by growing strain KH85 at 37°C to mid log phase in LB medium supplemented with 100 mg∙l−1 ampicillin . IPTG was added to 0 . 5 mM and protein expression was maintained overnight at 20°C . ( 1H , 15N ) -HSQC-NMR spectra of PhoQ PD were collected as a function of pH previously ( Prost et al . , 2007 ) . Briefly , uniformly 15N- and 13C-labeled PhoQ PD was prepared to 1 . 2 mM in 20 mM sodium phosphate buffer pH 6 . 5 , 150 mM NaCl , 20 mM MgCl2 , 0 . 1 mM EDTA , and 10% ( vol/vol ) D2O . The pH was lowered by approximately 0 . 5 units at a time by addition of microliter aliquots of 500 mM DCl . HSQC spectra from the pH-titration were collected at pH 6 . 5 , 6 . 0 , 5 . 5 , 4 . 9 , 4 . 1 , and 3 . 5 . Standard triple-resonance experiments were collected a pH 3 . 5 for assignments . Assignments from pH 3 . 5 were translated to higher pH conditions by tracking chemical shifts through the titration series . NMR experiments were performed at 25°C on a Bruker DMX 500 MHz spectrometer equipped with a triple-resonance , triple-axis gradient probe . Data were processed and analyzed using the programs NMRPipe/NMRDraw ( Delaglio et al . , 1995 ) and NMRView ( Johnson and Blevins , 1994 ) . To identify regions of the PhoQ PD affected by pH , ( 1H , 15N ) -HSQC-NMR spectra of the PhoQ PD were compared at pH 6 . 5 and 3 . 5 . Resonances that experienced a chemical shift perturbation ( CSP ) greater than 0 . 08 ppm and/or that broadened beyond detection were considered significantly affected by pH . CSPs were calculated using the formula ( ( Δ1H ) + ( Δ15N/5 ) ) 1/2 . Resonances that did not meet these criteria were considered unaffected . 36 residues within the PhoQ PD could not be unambiguously categorized into these groups because of missing/ambiguous assignments and/or crowding in the spectra . Wild-type ( CS093 ) , phoQW104C-A128C ( KH163 ) , and ΔphoQ ( CS1350 ) S . enterica Typhimurium were grown overnight in N-mm 1 mM MgCl2 pH 7 . 5 . In the morning , the cultures were normalized to OD600 2 . 0 and diluted 1:50 into fresh N-mm 1 mM MgCl2 pH 7 . 5 and grown at 37°C , 250 rpm . At approximately OD600 0 . 2 , the cultures were normalized to OD600 0 . 2∙ml−1 , washed once , and resuspended in 1 ml of either N-mm pH 7 . 5 1 mM MgCl2 , pH 5 . 5 1 mM MgCl2 , pH 7 . 5 10 μM MgCl2 , or pH 7 . 5 1 mM MgCl2 5 μg∙ml−1 . The cultures were grown shaking at 37°C , 250 rpm . After 1 hr , the cultures were immediately pelleted at 4°C , the media was aspirated , and placed on ice . RNA was collected using the Trizol Max Bacterial RNA Isolation Kit ( Ambion , Thermo Fisher Scientific , Grand Island , NY ) and RNeasy mini kit ( Qiagen , Netherlands ) . cDNA was generated using SuperScript III First-Strand ( Invitrogen , Thermo Fisher Scientific , Grand Island , NY ) . Synthesis Supermix for qRT-PCR ( Invitrogen ) . Quantitative RT-PCR was performed using SYBR GreenER qPCR SuperMix Universal ( Invitrogen ) and a BioRad CFX96 thermocycler for S . enterica Typhimurium rpoD , pagD , pagO , phoN , and phoP target transcripts using the appropriate qRT primers ( Table 3 ) . Relative gene expression was determined using the 2−Δ ΔCT method ( Livak and Schmittgen , 2001 ) . rpoD was used as the calibrator and gene expression was normalized to ΔphoQ . The S . enterica Typhimurium PhoQW104C-A128C PD structure ( PDB 4UEY ) was acquired by crystallizing the purified protein using a Mosquito crystallization robot ( TTP Labtech , United Kingdom ) and Nextal Classic Suite , Nextal Classic Suite II , Protein complex Suite ( Qiagen ) and JBScreen Classic HTS II ( Jena Bioscience , Germany ) . The progress of crystallization at 20°C was monitored using a temperature controlled robot ( Rock imager system , Formulatrix , Bedford , MA ) . Crystals appeared after 2 weeks . Optimized crystals of the PhoQW104C-A128C PD were formed in 0 . 1 M Bis-Tris pH 6 . 5 200 mM Magnesium chloride 25% ( wt/vol ) PEG3350 . Crystals of the PhoQW104C-A128C PD were mounted in nylon loops ( Hampton Research , Aliso Viejo , CA ) and directly frozen in liquid nitrogen . Diffraction data of the crystals were collected at ALBA synchrotron ( BL13 XALOC , Barcelona , Spain ) . Crystals were kept at 100 K and 200 diffraction images at 1° were recorded on a Pilatus 6M detector ( Dectris , Baden , Switzerland ) . Diffraction data were processed and scaled using the XDS software package ( Kabsch , 2010 ) . Data were truncated at lower resolution according to the recently defined CC* correlation factor ( Karplus and Diederichs , 2012 ) . Molecular replacement trials were performed using the program MOLREP and the model of the S . enterica Typhimurium PhoQ PD from the PDB databank ( PDB 1YAX ) ( Cho et al . , 2006; Vagin and Teplyakov , 2010 ) . The structure was refined using the PHENIX program package ( Afonine et al . , 2012 ) after rebuilding the structure in COOT ( Emsley et al . , 2010 ) . Structure details and PDB entries are given in Table 1 . Model quality was assessed using the Molprobity server ( http://molprobity . biochem . duke . edu/ ) . Prior to CD data collection , purified PhoQ PD and PhoQW104C-A128C PD were exchanged into 20 mM sodium phosphate buffer pH 5 . 5 150 mM NaCl 1 mM MgCl2 using a 5 ml HiTrap desalting column ( Amersham ) and treated with or without an approximate 1000 molar excesses of TCEP hydrochloride ( Sigma ) pH 5 . 5 for 4 hr to reduce the disulfide bond formed between W104C and A128C . Following TECP treatment , protein samples were exchanged in to 20 mM sodium phosphate buffer pH 5 . 5 150 mM NaCl 1 mM MgCl2 , with or without 1 mM TCEP and equilibrated overnight at 4°C . Following buffer exchange and equilibration , protein samples were concentrated and prepared to 17 μM for CD analysis . Disulfide bond reduction was monitored by SDS-PAGE prior to performing CD experiments . All CD data collection was performed on an Aviv model 420 spectrometer fitted with a total fluorescence accessory module and thermoelectric cuvette holder using a 1 mm pathlength quartz cuvette . Wavelength scans were performed for each sample prior to thermal denaturation from 260 to 195 nm at 25°C , sampling every 1 nm , with a 3 s averaging time per reading . CD-monitored thermal denaturation data was collected at 212 nm , from 25°C to 95°C , in 1°C increments , with a 3 s averaging time per reading , and 30 s temperature equilibration between readings . Raw thermal denaturation data were normalized to give the fraction unfolded protein assuming a two-state denaturation process ( Kamal et al . , 2002 ) . All CD experiments were reproduced on at least three separate occasions . BALB/c or A/J mice were ordered from Jackson Laboratories and virulence phenotypes for strains of S . enterica Typhimurium were determined by competition or single-strain inoculation . Competition experiments were performed similarly to previously described ( Freeman et al . , 2003 ) . Briefly , cultures of KH111 , KH112 , KH113 , and KH114 were grown overnight in LB media with the appropriate antibiotic and prepared by serial dilution in PBS . The inoculum for IP competition experiments was prepared by equally mixing 2 . 5 × 105 cfu of KH111 ( strain A ) with 2 . 5 × 105 cfu of KH112 , KH113 , or KH114 ( strain B ) in 2 ml PBS . The inoculum for PO competition experiments was prepared by equally mixing 5 × 107 cfu of KH111 ( strain A ) with 5 × 107 cfu of KH112 , KH113 , or KH114 ( strain B ) in 2 ml PBS . 6- to 8-week old female BALB/c mice were administered 0 . 2 ml of the mixture , for a total inoculation of 1 × 105 bacteria for IP infections or 5 × 108 bacteria for PO infections . For PO competition experiments , mice were deprived food for 5 hr prior to administering bacteria by oral gavage . The inoculum was confirmed for each experiment by plating dilutions on LB media supplemented with either 50 µg∙ml−1 kanamycin or 100 µg∙ml−1 ampicillin . Mice were euthanized by CO2 asphyxiation at 48-hpi ( IP ) or 96-hpi ( PO ) and spleens were harvested and homogenized in PBS . Homogenized spleens were serial diluted and plated on LB media supplemented with either 50 µg∙ml−1 kanamycin or 100 µg∙ml−1 ampicillin in order to determine the cfu∙ml−1 bacterial burden for each strain . The competitive index ( CI ) for each strain was calculated using the following formula: CI = ( strain B cfu∙ml−1 spleen/strain A cfu∙ml−1 spleen ) / ( strain B cfu∙ml−1 inoculum/strain A cfu∙ml−1 inoculum ) . For single-strain experiments , cultures of CS093 , KH163 , and CS1350 were grown overnight in LB media and prepared by serial dilution in PBS . The inoculum was confirmed for each experiment by plating dilutions on LB media . 6- to 8-week old female BALB/c or A/J mice were infected IP with approximately 1 × 103 cfu in 0 . 2 ml PBS . Mice were euthanized by CO2 asphyxiation at 48- and 96-hpi and spleens were harvested and homogenized in PBS . Homogenized spleens were serial diluted and plated on LB media in order to determine the cfu∙ml−1 bacterial burden for each strain . All mouse experiments were performed with IACUC approval . Wild type ( CS093 ) , phoQW104C-A128C ( KH163 ) , and ΔphoQ ( CS1350 ) were grown overnight in N-mm pH 7 . 5 1 mM MgCl2 . The following morning , the strains were washed in the appropriate N-mm , normalized , and diluted to 0 . 05 OD600 in either N-mm pH 7 . 5 or pH 5 . 5 supplemented with 1 mM MgCl2 . The strains were grown in a rolling drum at 37°C . At the indicated time-points , the bacterial strains were diluted 1:10 in PBS and their OD600 was monitored . Bone marrow was isolated from the femurs of BALB/c mice obtained from Jackson Laboratories and differentiated for 7 days in RPMI 1640 media ( Gibco #22400-089 , Thermo Fisher Scientific , Grand Island , NY ) supplemented with 10% FBS and L-929 cell supernatant following standard protocols . Following differentiation , bone-marrow derived macrophages were seeded into 24-well plates and incubated overnight . Bone-marrow derived macrophages were infected in triplicate with CS093 , KH163 , or CS1350 S . enterica Typhimurium and bacterial survival determined using a standard gentamicin-protection assay . Briefly , CS093 , KH163 , and CS1350 were grown overnight in LB media . The following morning , bacterial cultures are washed in PBS and suspended in RPMI 1640 at the appropriate concentration . BALB/c bone-marrow derived macrophages in 24-well plates ( 2 × 105 per well ) were washed with PBS and infected in triplicate with CS093 , KH163 , or CS1350 ( M . O . I . of 10 ) in RPMI 160 supplemented with 10% FBS , synchronized by centrifugation at 1000 rpm for 5 min at RT , and incubated for 30 min . Following incubation , infected macrophage monolayers were washed with PBS , incubated with media supplemented with 100 µg∙ml gentamicin−1 ( Sigma ) for 90 min and maintained at 15 µg∙ml−1 gentamicin for the duration of the experiment . Bacterial intracellular survival was determined by lysing infected macrophage with 1% Triton X-100 in PBS at the indicated time-points and plating serial dilutions on LB media for cfu counting . BALB/c bone marrow-derived macrophages were seeded into 6-well plates ( 1 × 107 per well ) and infected in triplicate with CS093 , KH163 , or CS1350 S . enterica Typhimurium using a standard gentamicin-protection protocol . 30 min post-infection , extracellular bacteria were harvested , lysed in Max Bacterial Enhancement Reagent ( Ambion ) and RNA was stabilized with Trizol ( Ambion ) . 4 hr post-infection , media was aspirated , infected macrophages were solubilized in Trizol to stabilize total RNA and triplicates where pooled . Trizol samples were stored at −80°C . RNA was prepared according to the Trizol Reagent protocol , treated with TURBO DNA-free DNase ( Ambion ) , and RNA quality was monitored using a 2200 TapeStation ( Agilent Technologies ) . cDNA was generated using SuperScript III First-Strand Synthesis Supermix for qRT-PCR ( Invitrogen ) . Quantitative RT-PCR was performed using SYBR GreenER qPCR SuperMix Universal ( Invitrogen ) and a BioRad CFX96 thermocycler for S . enterica Typhimurium rpoD , pagD , pagO , phoN , and phoP target transcripts using the appropriate qRT primers ( Table 3 ) . Relative gene expression was determined using the 2−Δ ΔCT method ( Livak and Schmittgen , 2001 ) . rpoD cDNA generated from extracellular bacteria harvested 30 min post-infection was used as the calibrator . Analysis and modeling of the three-dimensional protein structures was carried out using the PyMOL molecular viewer ( Schrodinger , 2010 ) . | Salmonella bacteria cause illnesses in humans , such as food poisoning and typhoid fever . In response to a Salmonella infection , immune cells known as macrophages detect and engulf the bacteria . The conditions inside the macrophage ( which include an acidic pH and high levels of antimicrobial molecules ) can destroy some bacteria . However , Salmonella bacteria ( which are also called salmonellae ) can sense and counteract these hostile conditions; this allows them to remodel their surface to survive and reproduce inside macrophages and continue to cause disease . A protein known as PhoQ , which is found on the surface of Salmonella bacteria , is a sensor that detects when the bacterium is inside a macrophage and so needs to boost its defenses . The PhoQ sensor is able to respond to acidity , the absence of divalent cations—such as magnesium and calcium ions—and certain antimicrobial peptide molecules . These conditions and components are used inside macrophages to try and kill the bacteria , but it was not known which of these signals PhoQ actually senses during an infection . Hicks et al . established how the sensor region of PhoQ changes when it is exposed to acid . This knowledge enabled variants of this protein to be constructed that do not respond when exposed to acidic conditions or low levels of divalent cations . Salmonellae that have these modified PhoQ sensors were still able to infect macrophages and cause disease in mice . These findings suggest that antimicrobial peptide sensing alone is sufficient to trigger the bacteria's defenses inside host organisms . Understanding how salmonellae detect antimicrobial factors could help with the development of new treatments for the diseases caused by these bacteria . Furthermore , the new tools developed by Hicks et al . could be applied to other systems to characterize how bacteria interact with their host environment during infection . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"biochemistry",
"and",
"chemical",
"biology",
"microbiology",
"and",
"infectious",
"disease"
] | 2015 | Acidic pH and divalent cation sensing by PhoQ are dispensable for systemic salmonellae virulence |
The membrane-bound transcription factor ATF6α is activated by proteolysis during endoplasmic reticulum ( ER ) stress . ATF6α target genes encode foldases , chaperones , and lipid biosynthesis enzymes that increase protein-folding capacity in response to demand . The off-state of ATF6α is maintained by its spatial separation in the ER from Golgi-resident proteases that activate it . ER stress induces trafficking of ATF6α . We discovered Ceapins , a class of pyrazole amides , as selective inhibitors of ATF6α signaling that do not inhibit the Golgi proteases or other UPR branches . We show that Ceapins block ATF6α signaling by trapping it in ER-resident foci that are excluded from ER exit sites . Removing the requirement for trafficking by pharmacological elimination of the spatial separation of the ER and Golgi apparatus restored cleavage of ATF6α in the presence of Ceapins . Washout of Ceapins resensitized ATF6α to ER stress . These results suggest that trafficking of ATF6α is regulated by its oligomeric state .
Activating transcription factor six alpha ( ATF6α ) is a type-II transmembrane protein localized in the endoplasmic reticulum ( ER ) where , with its close homolog ATF6β , it functions as an ER stress sensor in one of the three principal branches of the unfolded protein response ( UPR ) ( Haze et al . , 1999; Gardner et al . , 2013 ) . ATF6α target genes are exclusively cytoprotective , functioning to increase the folding capacity of the ER and restore ER homeostasis ( Adachi et al . , 2008; Wu et al . , 2007 ) . Cells or animals lacking ATF6α show impaired survival upon ER stress ( Wu et al . , 2007; Yamamoto et al . , 2007 ) . When demand exceeds the folding capacity of the ER , ATF6α is transported from the ER to the Golgi apparatus , where sequential cleavage by two Golgi-resident proteases – site-1 and site-2 proteases ( S1P and S2P ) , respectively - releases its N-terminal domain ( ATF6α-N ) from the membrane as a functional b-Zip transcription factor . ATF6α-N is then imported into the nucleus ( nuclear translocation ) where it activates transcription of its target genes ( Ye et al . , 2000 ) . The mechanism that retains ATF6α in the ER and then releases it to allow transport to the Golgi apparatus is unknown . Deciphering the mechanism of ATF6α’s regulated trafficking is essential to understanding how proteostasis is maintained in the ER . The mechanism whereby ATF6α senses ER stress has remained a mystery since its discovery in 1998 ( Yoshida et al . , 1998 ) . The lumenal domain of ATF6α ( ATF6α-LD ) dictates whether ATF6α localizes to the ER or to the Golgi apparatus ( Chen et al . , 2002; Schindler and Schekman , 2009 ) . In fact , the soluble ATF6α-LD alone is sufficient to sense ER stress ( Sato et al . , 2011 ) , and attaching the ATF6α-LD to the constitutively transported SNARE protein Sec22 is sufficient to retain Sec22 in the ER of unstressed cells , allowing its trafficking to the Golgi apparatus only upon ER stress ( Schindler and Schekman , 2009 ) . Thus ATF6α-LD is ATF6α‘s stress-sensor and regulates its trafficking . It is unclear what aspect of the folding environment ATF6α-LD is sensing , or if ATF6α senses misfolded proteins directly . The molecular machinery required to move ATF6α from the ER to the Golgi apparatus , the COPII coat , is positioned on the cytosolic side of the ER membrane . Activation of ATF6α therefore requires transmitting the signal from the ER lumen to the cytosol ( Schindler and Schekman , 2009; Nadanaka et al . , 2004 ) . It is unknown whether ATF6α interacts with the COPII coat directly or requires a transmembrane adaptor or traffics by bulk-flow after release from a transport-incompetent state . SREBP , the membrane bound transcription factor that responds to low cholesterol , has an elegant trafficking mechanism involving binding of a retention factor , INSIG , to a transport factor , SCAP ( Brown and Goldstein , 2009 ) . Indeed it is SCAP , not SREBP , that binds to and senses cholesterol and this binding regulates the interaction of SCAP both with its retention factor INSIG and with the COPII coat ( Sun et al . , 2007; Motamed et al . , 2011 ) . SREBP itself neither senses the signal nor interacts with the trafficking machinery . Similarly , a putative ACAP ( ATF6α cleavage activating protein ) and/or a putative INSIG-like ATF6α retention factor may remain to be discovered . To gain insight into the mechanism whereby ATF6α senses stress we performed high-throughput cell-based screens to identify small molecule modulators of ATF6α signaling . In the accompanying paper , we describe the identification of Ceapins , a class of pyrazole amides that inhibit selectively the processing of ATF6α by S1P and S2P in response to ER stress but not the other UPR sensors , including – surprisingly – ATF6β , or SREBP . Using Ceapins to interrogate each step of ATF6α activation , we show here that Ceapins prevent selection of ATF6α into COPII vesicles by retaining it in foci in the ER membrane . Removing the requirement for trafficking by bringing together substrate and proteases restored cleavage in the presence of Ceapins . Ceapins induce rapid clustering of ATF6α , indicating that its oligomeric state plays a key role to regulate its trafficking and thereby activation in response to ER stress . Based on its mode of action corralling ATF6α into ER-restricted foci , we named Ceapins after the Irish verb 'ceap , ' meaning 'to trap' .
To understand how Ceapins inhibit ATF6α activation and processing in response to ER stress , we monitored ATF6α trafficking . To this end , we used a U2-OS cell line , which stably expresses a fluorescent GFP-ATF6α fusion protein at low levels . We followed nuclear translocation of GFP-ATF6α-N , the proteolytic fragment resulting from GFP- ATF6α cleavage in response to ER stress . In unstressed cells , GFP-ATF6α localized to the ER ( Figure 1A and E , green ) . As expected , after induction of ER stress by treatment with thapsigargin ( Tg ) , which inhibits the ER calcium pump , or tunicamycin ( Tm ) , which inhibits N-linked glycosylation , we observed a fraction of the GFP fluorescence in the nucleus , apparent by co-localization with DAPI-stained nuclear DNA ( Figure 1B and F , DNA in purple , co-localization indicated by white color in overlay ) . Induction of ER stress in the presence of active Ceapin-A7 ( Figure 1C and G ) but not of the inactive Ceapin analog A5 ( Figure 1D and H ) prevented nuclear translocation of GFP-ATF6α-N and led to an accumulation of GFP fluorescence in discrete foci ( quantified in Figure 1—figure supplement 1 ) . We have previously shown [accompanying manuscript; Gallagher et al . , 2016] that under these conditions active Ceapin analogs block ATF6α proteolysis , indicating that the foci correspond to a pool of uncleaved GFP-ATF6α . 10 . 7554/eLife . 11880 . 003Figure 1 . Ceapins induce foci formation and prevent ER-stress induced nuclear translocation of GFP-ATF6 . ( A–H ) U2-OS cells stably expressing GFP-ATF6α were treated either with vehicle ( A , E ) or ER stress inducer ( B–D and F–H ) in the absence ( A , B , E and F ) or presence of active ( 6 μM Ceapin-A7 , C , G ) or inactive ( 6 μM Ceapin-A5 , D , H ) Ceapin analogs for five hours prior to fixation and fluorescent imaging of GFP-ATF6α ( green ) and DNA ( magenta ) . In unstressed cells ( A , E , DMSO ) GFP-ATF6α is in the ER . Addition of either 100 nM thapsigargin ( B ) or 2 . 5 μg/mL tunicamycin ( F ) induces nuclear translocation of cleaved GFP-ATF6 . The presence Ceapin-A7 ( C , G ) but not the inactive Ceapin analog A5 ( D , H ) prevents nuclear translocation . Scale bar is 10 μm . ( I–N ) Time-lapse images of U2-OS cells stably expressing GFP-ATF6α treated either with vehicle ( I , DMSO ) , ER stress ( J , 100 nM Tg ) , ER stress plus 5 μM Ceapin-A1 ( K , IC50 4 . 9 ± 1 . 2 μM ) , ER stress plus 5 μM Ceapin-A7 ( L , IC50 0 . 59 ± 0 . 17 μM ) or Ceapin analogs alone ( M , 5 μM Ceapin-A1; N , 5 μM Ceapin-A7 ) . The addition of Ceapin analogs induces formation of GFP-ATF6α foci and either partially ( K , Ceapin-A1 ) or completely ( L , Ceapin-A7 ) inhibits nuclear translocation of GFP-ATF6α in response to ER stress . Scale bar is 10 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 11880 . 00310 . 7554/eLife . 11880 . 004Figure 1—figure supplement 1 . Quantification of nuclear translocation assay with Ceapin Analogs . Cells were treated without ( white bar ) or with ( colored bars ) ER stressor ( 100 nM Tg , solid bars or 2 . 5 μg/mL Tm , patterned bars ) in the absence ( black bars ) or presence of active ( 6 μM Ceapin-A7 , purple bars ) or inactive ( 6 μM Ceapin-A5 ) Ceapin analogs . Means from three different wells are plotted; error bars are 95% confidence limits . Statistical analysis is one-way ANOVA of all groups . DOI: http://dx . doi . org/10 . 7554/eLife . 11880 . 00410 . 7554/eLife . 11880 . 005Figure 1—figure supplement 2 . Active but not inactive analogs of Ceapin induce foci formation and prevent nuclear translocation of GFP-ATF6 . ( A–H ) Time-lapse images of U2-OS cells stably expressing GFP-ATF6α treated either with vehicle ( A , C , E , G , DMSO ) , ER stressor ( B , D , F , H , 100 nM Tg ) , in the absence ( A , B ) or presence ( C–H ) of Ceapin analogs . Addition of active Ceapin analogs , 5 μM Ceapin-A1 ( C , D , IC50 = 4 . 9 ± 1 . 2 μM ) or 5 μM Ceapin-A3 ( E , F , IC50 = 6 . 9 ± 0 . 7 μM ) but not the inactive Ceapin analog , 5 μM Ceapin-A5 ( G , H , IC50 > 30 μM ) induce foci formation of GFP-ATF6 . Cells treated with ER stressor ( B , 121 min time point ) show nuclear translocation of GFP-ATF6 . Active Ceapin analogs ( D , F , 121 min time point ) but not the inactive analog ( H , 121 min time point ) prevent this ER-stress induced nuclear translocation and GFP-ATF6α remains in foci . Scale bar is 10 μm . Images are representative of at least three independent experiments where three positions per well were imaged in each experiment . DOI: http://dx . doi . org/10 . 7554/eLife . 11880 . 00510 . 7554/eLife . 11880 . 006Figure 1—figure supplement 3 . GFP-ATF6 foci persist for up to 24 hr after addition of Ceapin A7 . ( A–H ) U2-OS cells stably expressing GFP-ATF6α were either untreated ( A ) or treated either with vehicle ( C , E , G , DMSO ) or Ceapin-A7 ( 6 μM B , D , F , H ) for five minutes ( B ) , fifteen ( C , D ) , eighteen ( E , F ) or twenty-four ( G , H ) hours prior to fluorescent imaging of GFP-ATF6α . In untreated ( A ) or vehicle treated cells ( C , E , G , DMSO ) GFP-ATF6α is in the ER . Addition of 6 μM Ceapin-A7 induces foci formation of GFP-ATF6 that persists for up to twenty-four hours . DMSO concentration for all wells was 0 . 034% . DOI: http://dx . doi . org/10 . 7554/eLife . 11880 . 006 To characterize foci formation further , we next followed the cells in real time using live-cell imaging prior to and after induction of ER stress ( Figure 1I–N; Figure 1 , Videos 1–6 ) . Treatment with vehicle alone showed ER localization that did not change over time ( Figure 1I ) . In contrast , after induction of ER stress GFP fluorescence first accumulated in a perinuclear region , consistent with movement of GFP-ATF6α to the Golgi apparatus , and then accumulated in the nucleus , consistent with proteolytic processing and nuclear import of the resulting GFP-ATF6α-N ( Figure 1J ) . Addition of either active Ceapin-A7 or Ceapin-A1 induced rapid foci formation of GFP-ATF6α , while inhibiting nuclear accumulation ( Figure 1K and L ) . In contrast , the inactive Ceapin analog A5 failed to induce foci formation ( Figure 1—figure supplement 2 ) . Importantly , we observed that active but not inactive Ceapin analogs induce GFP-ATF6α foci even in the absence of ER stress ( Figure 1M and N , Figure 1—figure supplement 2 ) and these foci persist for up to twenty-four hours ( Figure 1—figure supplement 3 ) . These results suggest that Ceapins inhibit ATF6α signaling by capturing it in foci . Interestingly we also see foci in cells subjected to ER stress alone at later time points corresponding to the time point at which attenuation of ATF6α signaling would initiate ( Figure 1J , 90 min time point and Video 2 ) ( Haze et al . , 2001; Rutkowski et al . , 2006 ) . To assess if Ceapin-induced GFP-ATF6α foci depict a terminal state of ATF6α destined for degradation , we performed washout experiments and followed GFP-ATF6α foci using live cell imaging ( Figure 2 and Videos 7–9 ) . Cells treated with active Ceapin analogs ( Ceapin-A1 and Ceapin-A7; Figure 2B and C ) showed rapid formation of GFP-ATF6α foci . We allowed foci to form for 17 min , then washed the cells , and added media without inhibitors . Washout of both Ceapin analogs led to rapid dissolution of GFP-ATF6α foci , indicating the foci formation was reversible ( Figure 2B and C ) . Cells treated with vehicle alone showed no change in GFP-ATF6α localization throughout the washout experiment ( Figure 2A ) . We observed the same washout kinetics in cells pretreated for three hours with cycloheximide to inhibit protein synthesis , a time point at which it is reasonable to expect any newly translated GFP-ATF6α had folded and matured ( Heim et al . , 1994; 1995; Cormack et al . , 1996; Li et al . , 1998; Sacchetti , 2001; Sacchetti et al . , 2001; Zhang et al . , 2006; Pédelacq et al . , 2006; Ugrinov and Clark , 2010 ) ( Figure 2—figure supplement 1 and Videos 10–13 ) . This result indicates that the same molecules of GFP-ATF6α clustered into foci by Ceapins are redistributed in the ER upon washout . Time-lapse imaging of U2-OS cells stably expressing GFP-ATF6α treated either with vehicle ( Video 1 , DMSO ) or ER stressor ( 100 nM Tg ) in the absence ( Video 2 ) or presence active Ceapin analogs ( Video 3 , 10 μM Ceapin-A1 ) , ( Video 4 , 1 μM Ceapin-A7 ) or with active Ceapin analogs alone ( Video 5 , 10 μM Ceapin-A1 ) , ( Video 6 , 1 μM Ceapin-A7 ) . Images were acquired every minute and videos play at five frames per second . These videos are supplementary to Figure 1 . 10 . 7554/eLife . 11880 . 007Video 1 . GFP-ATF6α expressing U2-OS cells treated with vehicle . DOI: http://dx . doi . org/10 . 7554/eLife . 11880 . 00710 . 7554/eLife . 11880 . 008Video 2 . GFP-ATF6α expressing U2-OS cells treated with ER stressor . DOI: http://dx . doi . org/10 . 7554/eLife . 11880 . 00810 . 7554/eLife . 11880 . 009Video 3 . GFP-ATF6α expressing U2-OS cells treated with ER stressor and Ceapin-A1 . DOI: http://dx . doi . org/10 . 7554/eLife . 11880 . 00910 . 7554/eLife . 11880 . 010Video 4 . GFP-ATF6α expressing U2-OS cells treated with ER stressor and Ceapin-A7 . DOI: http://dx . doi . org/10 . 7554/eLife . 11880 . 01010 . 7554/eLife . 11880 . 011Video 5 . GFP-ATF6α expressing U2-OS cells treated with Ceapin-A1 . DOI: http://dx . doi . org/10 . 7554/eLife . 11880 . 01110 . 7554/eLife . 11880 . 012Video 6 . GFP-ATF6α expressing U2-OS cells treated with Ceapin-A7 . DOI: http://dx . doi . org/10 . 7554/eLife . 11880 . 01210 . 7554/eLife . 11880 . 013Figure 2 . Ceapin-induced foci are reversible and correlate with inhibition of ATF6 . ( A–C ) Time-lapse images of U2-OS cells stably expressing GFP-ATF6α treated either with vehicle ( A , DMSO ) or active Ceapin analogs ( B , 10 μM Ceapin-A1 ) , ( C , 1 μM Ceapin-A7 ) for seventeen minutes to allow foci formation . Cells were then washed once with PBS and then media without compound was added . Scale bar is 10 μm . Images are representative of three independent experiments where three positions per well were imaged for each experiment . ( D–G ) Time-lapse images of U2-OS cells stably expressing GFP-ATF6α treated either with vehicle ( A , B , DMSO ) or 10 μM Ceapin-A1 ( C , D ) . After fifteen minutes , the cells were washed with PBS and then media without ( A , C , DMSO ) or with ER stressor ( B , D , 100 nM Tg ) was added . After washout , ATF6α inhibitor foci resolve ( C , D ) and nuclear translocation of GFP-ATF6α occurs with similar kinetics in cells initially treated with either DMSO ( B ) or Ceapin-A1 ( D ) . Scale bar is 10 μm . Images are representative of three independent experiments where three positions per well were imaged for each experiment . DOI: http://dx . doi . org/10 . 7554/eLife . 11880 . 01310 . 7554/eLife . 11880 . 014Figure 2—figure supplement 1 . Redistribution of GFP-ATF6 from foci into the ER after washout of Ceapin-A1 does not require protein synthesis . ( A–D ) Time-lapse images of U2-OS cells stably expressing GFP-ATF6α pretreated with either cycloheximide ( A , B 0 . 1 μg/ml CHX ) or vehicle ( C , D ethanol ) for three hours prior to imaging . During imaging , cells were treated either with vehicle ( A , C DMSO ) or Ceapin-A1 ( B , D 10 μM Ceapin-A1 ) for sixteen minutes to allow foci formation . Cells were then washed once with PBS with ( A , B ) or without ( C , D ) cycloheximide and then media without compound with ( A , B ) or without ( C , D ) cycloheximide was added . Scale bar is 10 μm . Images are representative of three independent experiments where three positions per well were imaged for each experiment . DOI: http://dx . doi . org/10 . 7554/eLife . 11880 . 014 Time-lapse imaging of U2-OS cells stably expressing GFP-ATF6α treated either with vehicle ( Video 7 , DMSO ) or active Ceapin analogs ( Video 8 , 10 μM Ceapin-A1 ) , ( Video 9 , 1 μM Ceapin-A7 ) . Seventeen minutes after compound addition cells were washed once with PBS and then fresh media without compound was added . Images were acquired every minute and videos play at five frames per second . These videos are supplementary to Figure 2 . 10 . 7554/eLife . 11880 . 015Video 7 . Addition and washout of vehicle to GFP-ATF6α expressing cells . DOI: http://dx . doi . org/10 . 7554/eLife . 11880 . 01510 . 7554/eLife . 11880 . 016Video 8 . Addition and washout of Ceapin-A1 to GFP-ATF6α expressing cells . DOI: http://dx . doi . org/10 . 7554/eLife . 11880 . 01610 . 7554/eLife . 11880 . 017Video 9 . Addition and washout of Ceapin-A7 to GFP-ATF6α expressing cells . DOI: http://dx . doi . org/10 . 7554/eLife . 11880 . 017 Time-lapse imaging of U2-OS cells stably expressing GFP-ATF6α pretreated either with vehicle ( Video 10 , Video 11 , Ethanol ) or protein synthesis inhibitor ( Video 12 , Video 13 , 0 . 1 μg/ml cycloheximide ) for three hours prior to imaging . During imaging , cells were treated either with vehicle ( Video 10 , Video 12 , DMSO ) or Ceapin-A1 ( Video 11 , Video 13 , 10 μM Ceapin-A1 ) . Sixteen minutes after compound addition cells were washed once with PBS ± protein synthesis inhibitor and then fresh media without compound ± protein synthesis inhibitor was added . Images were acquired every minute and videos play at five frames per second . These videos are supplementary to Figure 2 . 10 . 7554/eLife . 11880 . 018Video 10 . Addition and washout of vehicle to GFP-ATF6α expressing cells pretreated with vehicle . DOI: http://dx . doi . org/10 . 7554/eLife . 11880 . 01810 . 7554/eLife . 11880 . 019Video 11 . Addition and washout of Ceapin-A1 to GFP-ATF6α expressing cells pretreated with vehicleDOI: http://dx . doi . org/10 . 7554/eLife . 11880 . 01910 . 7554/eLife . 11880 . 020Video 12 . Addition and washout of vehicle to GFP-ATF6α expressing cells pretreated with cycloheximide . DOI: http://dx . doi . org/10 . 7554/eLife . 11880 . 02010 . 7554/eLife . 11880 . 021Video 13 . Addition and washout of Ceapin-A1 to GFP-ATF6α expressing cells pretreated with cycloheximide . DOI: http://dx . doi . org/10 . 7554/eLife . 11880 . 021 To test if foci dissolution reflected the release of ATF6α from Ceapin-mediated inhibition , we repeated the washout experiment , except that after the washout the replacement media contained either vehicle or thapsigargin to induce ER stress ( Figure 2D–G ) . After the washout , both control and Ceapin-treated cells responded to ER stress with the same kinetics , showing nuclear accumulation of GFP-ATF6α-N ( Figure 2E and G , 85 min time point ) . Neither Ceapin-treated cells nor vehicle-treated cells that received media without thapsigargin showed nuclear accumulation of GFP-ATF6α-N ( Figure 2D and F , 85 min time point ) , confirming that the washout and imaging procedures alone did not induce ER stress . These results suggest that the dissolution of GFP-ATF6α foci restores an activatable pool of GFP-ATF6α . Thus , Ceapins are reversible inhibitors of ATF6α . To further examine the subcellular localization of the Ceapin-induced GFP-ATF6α foci , we fixed and stained cells with an antibody to giantin to mark the Golgi apparatus . In the absence of ER stress , we observed little overlap between GFP-ATF6α and giantin ( Figure 3A; giantin staining in purple , see arrowheads in zoomed view ) . In contrast , in the presence of ER stress , we observed clear co-localization of GFP-ATF6α and giantin ( Figure 3B ) . Golgi apparatus localization was markedly enhanced upon treatment of ER stressed cells with S1P inhibitor , which blocks the Golgi-resident protease that initiates ATF6α processing ( Figure 3C ) . This result was expected as under these conditions GFP-ATF6α traffics to the Golgi apparatus , where it accumulates because it cannot be cleaved ( Okada et al . , 2003 ) . In contrast , GFP-ATF6α foci formed in the presence of Ceapin-A1 and ER stress did not co-localize with giantin ( Figure 3D ) . This result suggests that ATF6α is not cleaved in the presence of active Ceapin analogs because it does not traffic to the Golgi apparatus upon induction of ER stress . 10 . 7554/eLife . 11880 . 022Figure 3 . Ceapins retain tagged-ATF6α as foci in the ER and prevent trafficking of tagged ATF6α to the Golgi apparatus . ( A-–D ) U2-OS cells stably expressing GFP-ATF6α were treated either with vehicle ( A , DMSO ) , ER stress ( B , 100 nM Tg ) , ER stress and site-1-protease inhibitor ( C , 1 μM Pf-429242 ) or ER stress and Ceapin-A1 ( D , 10 μM Ceapin-A1 ) for thirty minutes prior to fixation and fluorescent imaging of GFP-ATF6α ( green ) , anti-Giantin to mark the Golgi apparatus ( red in RGB , purple in GM insets ) and DNA ( blue ) . ( A ) Unstressed cells have minimal co-localization of GFP-ATF6α and Giantin . ( B ) ER stress induces trafficking of GFP-ATF6α to the Golgi apparatus where it colocalizes with Giantin . ( C ) ER stress combined with the site-1 protease inhibitor inhibits cleavage of GFP-ATF6α in the Golgi apparatus causing GFP-ATF6α to accumulate there . ( D ) ER stress combined with Ceapin-A1 shows minimal colocalization of GFP-ATF6α with Giantin , indicating GFP-ATF6α has not trafficked to the Golgi apparatus in the presence of the Ceapin-A1 . ( E–L ) 293 T-REx cells stably expressing doxycycline inducible 3xFLAG-ATF6α were treated either with vehicle ( E , I , DMSO ) , ER stress ( F , J , 100 nM Tg ) , ER stress and Ceapin-A1 ( G , K , 5 μM Ceapin-A1 ) or Ceapin-A1 alone ( H , L , 5 μM Ceapin-A1 ) for thirty minutes prior to fixation and fluorescent imaging of 3xFLAG-ATF6α ( green ) and DNA ( blue ) and either an ER marker Calnexin ( E–H , red in RGB , purple in GM inset ) or a Golgi apparatus marker Giantin ( I–L , red in RGB , purple in GM inset ) . ER stress induced Golgi trafficking of 3xFLAG-ATF6α ( J , arrowheads ) is prevented by the addition of the Ceapin-A1 ( K ) . Ceapin-A1 either in combination with ER stress ( G ) or alone ( H ) induces formation of 3xFLAG-ATF6α foci that co-localize with ER tubules ( arrowheads ) . ( M–P ) U2-OS cells stably expressing GFP-ATF6α were treated either with vehicle ( M , DMSO ) , ER stress ( N , 100 nM Tg ) , ER stress and active Ceapin analogs ( O , 5 μM Ceapin-A1 ) , ( P , 5 μM Ceapin-A7 ) . After time-lapse imaging for 2 . 4 hr , cells were fixed and stained for GFP-ATF6α ( green ) , anti-GRP94 to mark the ER ( red ) and anti-Giantin to mark the Golgi apparatus ( blue ) . ER stress induced trafficking to the Golgi apparatus ( N ) is blocked by the Ceapin analogs , and the induced GFP-ATF6α foci remain co-localized with ER tubules even after almost 2 . 5 hr of ER stress ( O , P ) . Note that fixation conditions to visualize the ER and Golgi apparatus are not suitable for imaging the nuclear translocated fraction of GFP-ATF6α ( see Materials and methods ) . Higher magnification panels underneath each image show each channel singly in greyscale ( middle row ) , pairwise merges bottom row ) of GFP-ATF6α ( green ) with either ER ( magenta , bottom left ) or Golgi markers in ( magenta , and triple merge ( bottom row , right ) of GFP-ATF6α ( green ) , ER ( red ) and Golgi ( blue ) . In each panel , scale bars are 10 μm and boxed inserts are 7 x 7 μm ( A–L ) or 11 . 8 x 11 . 8 μm ( M–P ) . DOI: http://dx . doi . org/10 . 7554/eLife . 11880 . 022 We next asked whether our GFP-ATF6α fusion protein faithfully reported on ATF6α biology without influence from the GFP tag . To this end , we used immunofluorescence in 293 T-REx cells that stably express a doxycycline-inducible 3xFLAG-6xHis-tagged ATF6α ( 3xFLAG-ATF6α ) . In unstressed cells , we found 3xFLAG-ATF6α in the ER , co-localizing with the ER marker calnexin ( Figure 3E ) . Thirty minutes after induction of ER stress , a portion of 3xFLAG-ATF6α had moved out of the ER ( Figure 3F , arrowheads in zoomed views ) . Treatment of these cells with Ceapin-A1 , either in the presence ( Figure 3G ) or absence ( Figure 3H ) of ER stress , led to formation of 3xFLAG-ATF6α foci that decorated ER tubules ( arrowheads in zoomed views ) . As expected , ER stress induced co-localization of 3xFLAG-ATF6α with giantin ( Figure 3J ) . No such co-localization was observed upon co-treatment with Ceapin-A1 ( Figure 3K ) . Thus the Ceapin-induced formation of ATF6α foci is independent of the nature of the tag and likely reflects an intrinsic property of ATF6α . To ask to what degree Ceapins completely block or just delay GFP-ATF6α transport to the Golgi apparatus , we induced ER stress in U2-OS cells in the absence or presence of active Ceapin analogs and imaged cells after a prolonged 2 . 4 hr incubation . We stained cells for an ER ( GRP94 ) and a Golgi apparatus ( giantin ) marker . In unstressed cells , GFP-ATF6α co-localizes with the ER marker with only marginal Golgi marker co-localization ( Figure 3M ) . After 2 . 4 hr of ER stress , GFP-ATF6α showed pronounced Golgi apparatus localization with some ER staining remaining ( Figure 3N ) . ( Note that fixation conditions that best preserve ER structure are not optimal for nuclear staining , making nuclear translocation difficult to see in these images . ) In the presence of active Ceapin analogs , GFP-ATF6α foci decorated ER tubules and show minimal co-localization with the Golgi marker ( Figure 3O and P ) . Taken together , these results show that in different cell types and using different ATF6α variants , Ceapins induce ATF6α foci and prevent ATF6α trafficking to the Golgi apparatus in response to ER stress . Using qPCR analysis , we showed in the accompanying manuscript that Ceapins inhibit ATF6α signaling . These analyses did not rely on the use of engineered reporters or over-expression of ATF6α [accompanying manuscript; Gallagher et al . , 2016] . In contrast , the discovery of Ceapin-dependent ATF6α foci formation described here relied on the use of tagged and over-expressed fusion proteins . To address possible concerns arising from this approach , we next utilized a polyclonal antibody against ATF6α developed by the Mori lab ( Haze et al . , 1999 ) to image endogenous ATF6α in U2-OS cells ( Figure 4 ) . We already used this antibody to show that endogenous ATF6α is no longer proteolyzed in response to ER stress [accompanying manuscript; Gallagher et al . , 2016] . 10 . 7554/eLife . 11880 . 023Figure 4 . Endogenous ATF6α is in foci in unstressed cells and these foci are not changed by Ceapin-A7 either in the presence or absence of ER stress . ( A–D ) Nuclear translocation of endogenous ATF6α in response to ER stress is inhibited by Ceapin-A7 . U2-OS cells treated with either vehicle ( A , A’ , A” ) or ER stress ( 100 nM Tg ) , in the absence ( B , B’ , B” ) or presence of Ceapin-A7 ( C , C’ , C” , 6 μM Ceapin-A7 ) or with Ceapin-A7 alone ( D , D’ , D” ) for two hours prior to fixation and fluorescent imaging of endogenous ATF6 , anti-GM130 to mark the Golgi apparatus , GRP94 to mark the ER and DAPI to mark DNA . ( A–D ) Greyscale images of ATF6α for each treatment . ( A’–D’ ) Merged images of ATF6α ( green ) and nuclear staining ( purple ) . ( A”–D” ) Triple color merges of ATF6α ( green ) with ER ( GRP94 , red ) and Golgi markers ( GM130 , blue ) . Boxed insets in ( A”–D” ) are shown below either as greyscale images of each channel ( top row ) or double ( green/magenta ) or triple ( green/red/blue ) merged images ( bottom row ) . White arrows in boxed inserts point to Golgi staining . Scale bar is 10 μm and boxed inserts are 7 x 7 μm . ( E ) Quantification of nuclear translocation of endogenous ATF6 . Plotted is the ratio of nuclear to ER intensity of ATF6α signal per cell as a box plot , whiskers are minimum and maximum values of the data . Statistics show the results of unpaired , two-tailed t-tests between indicated groups . Data plotted is from one of two independent experiments , each with at least twenty cells per treatment group . DOI: http://dx . doi . org/10 . 7554/eLife . 11880 . 023 We analyzed changes in ATF6α staining in response to ER stress in U2-OS cells . To our surprise , we found that , even in unstressed cells , endogenous ATF6α was not evenly distributed but found in small foci that were finely distributed over the ER network ( Figure 4A–D ) . These observations contrasted with those described above made in cells over-expressing tagged versions of ATF6α that were diffusely ER-localized . After two hours of ER stress , a portion of endogenous ATF6α co-localized with the Golgi marker GM130 ( Figure 4B” , arrowheads in inserts ) , and we observed a significant portion of ATF6α staining in the nucleus ( Figure 4B and B’ , quantified in Figure 4E , bar 2 , p<0 . 0001 ) . Ceapin-A7 reduced co-localization of ATF6α with GM130 ( GM130 , Figure 4C” , arrowheads in inserts ) and blocked nuclear accumulation of ATF6α , retaining it in ER foci ( Figure 4C’ , quantified in Figure 4E , bar 3 , p<0 . 0001 ) . After two hours of treatment with Ceapin-A7 in the absence or presence of ER stress , ATF6α foci appeared larger and brighter than in unstressed cells ( Figure 4C , D ) . The data show that endogenous ATF6α largely phenocopies the behavior of its tagged-variants characterized above and reveal that at physiological expression levels and in the absence of ER stress ATF6α is already clustered in small foci in the ER membrane . Ceapins may further stabilize these foci , trapping ATF6α in the ER and thus preventing its Golgi apparatus trafficking and nuclear translocation . Our results suggest a simple model: Ceapins inhibit ATF6α cleavage by antagonizing its transport to the Golgi apparatus , thereby preventing the encounter of substrate and proteases required to liberate ATF6α-N from the membrane . To test this notion , we treated 293 T-REx cells expressing 3xFLAG-ATF6α with brefeldin A ( Figure 5A ) . Brefeldin A treatment fuses the Golgi apparatus and ER compartments ( Fujiwara et al . , 1988 ) and thus relocalizes both S1P and S2P proteases to the ER where they process ATF6α ( Shen and Prywes , 2004 ) . 10 . 7554/eLife . 11880 . 024Figure 5 . Collapsing the Golgi apparatus on to the ER restores cleavage of 3xFLAG-ATF6α in the presence of Ceapins . ( A ) 293 T-REx cells stably expressing doxycycline inducible 3xFLAG-ATF6α were treated either with vehicle ( ethanol ) or Brefeldin A ( 5 μg/mL BFA ) in the absence or presence of increasing concentrations of either S1P inhibitor ( Pf-429242 ) or active ( Ceapin-A1 , Ceapin-A3 ) or inactive ( Ceapin-A5 ) Ceapin analogs for one hour prior to harvesting lysates for Western Blot analysis of 3xFLAG-ATF6α . ( B ) 293 T-REx cells stably expressing doxycycline inducible 3xFLAG-ATF6α were treated either with vehicle ( DMSO ) or ER stressor ( 100 nM Tg ) in the absence or presence of increasing concentrations of either S1P inhibitor ( Pf-429242 ) or active ( Ceapin-A1 , Ceapin-A3 ) or inactive ( Ceapin-A5 ) Ceapin analogs for one hour prior to harvesting lysates for Western Blot analysis of 3xFLAG-ATF6α . For both A and B inhibitor concentrations were 0 . 5 , 5 , 15 , 25 μM respectively . GAPDH is shown as a loading control . Black arrowheads – 3xFLAG-ATF6α , white arrowheads – 3xFLAG-ATF6α-N . White lines in A indicate where intervening lane has been removed . DOI: http://dx . doi . org/10 . 7554/eLife . 11880 . 024 As expected , ATF6α-N was produced upon treatment of cells with brefeldin A ( Figure 5A , lanes 2 , 12 , 14 , 24 ) or thapsigargin ( Figure 5B , lanes 26 , 36 , 38 , 48 ) . Treatment with both S1P inhibitor and either brefeldin A ( Figure 5A , lanes 7–10 ) , or thapsigargin ( Figure 5B , lanes 31–34 ) , prevented ATF6α cleavage indicating that changing the subcellular localization of the protease had no effect on the efficacy of its inhibitor . In contrast , treatment of cells with active Ceapin analogs did not prevent production of ATF6α-N in brefeldin A-treated cells ( Figure 5A , lanes 3–6 and lanes 15–18 ) , whereas it did prevent production of ATF6α-N in thapsigargin-treated cells ( Figure 5B , lanes 27–30 and lanes 39–42 ) . As expected , the inactive Ceapin analog A5 had no effect on either treatment ( Figure 5A , lanes 19–22 and Figure 5B , lanes 43–46 ) . These results show that Ceapins do not convert ATF6α to an inaccessible or uncleavable form . Instead , cleavage of ATF6α in response to ER stress is inhibited because ATF6α does not traffic to the compartment where the proteases required for cleavage reside . The foci of GFP-ATF6α observed with Ceapin treatment resemble the staining pattern observed for proteins that localize to ER exit sites . To test if in the presence of Ceapins GFP-ATF6α accumulated in ER exit sites , we co-stained for SEC16 and SEC31A , which are established markers of ER exit sites and the COPII coat assembling there ( D'Arcangelo et al . , 2013; Hughes et al . , 2009 ) ( Figure 6A–E ) , arrowheads in zoomed view mark ER exit sites that stained for both SEC16 and SEC31A ) . To this end , we induced ER stress in U2-OS cells expressing GFP-ATF6α in the absence or presence of Ceapin analogs . In unstressed cells , we observed minimal overlap of GFP-ATF6α foci and SEC16/SEC31A double-positive foci ( Figure 6A , quantified in Figure 6F ) . When cells were treated with ER stress , the amount of overlap ( indicated by triple-positive foci ) increased ( Figure 6B and F ) , consistent with GFP-ATF6α passing through ER exit sites on its way to the Golgi apparatus . 10 . 7554/eLife . 11880 . 025Figure 6 . In the presence of Ceapins , GFP-ATF6α is no longer selected for transport from the ER in COP II vesicles . ( A–E ) . U2-OS cells stably expressing GFP-ATF6α were treated either with vehicle ( A , DMSO ) , or ER stressor ( B , 100 nM Tg ) , in the absence or presence of either active ( C , 10 μM Ceapin-A1 , D , 1 μM Ceapin-A7 ) or inactive ( E , 10 μM Ceapin-A5 ) Ceapin analogs for thirty minutes prior to fixation and fluorescent imaging of GFP-ATF6α ( green ) , the COP II outer coat component Sec31A ( red ) , the transmembrane ER exit site marker Sec16 ( blue ) and GRP94 to mark the ER ( not shown ) . Punctae containing both Sec31A and Sec16 were denoted ER exit sites ( ERES ) and are marked with arrowheads in the inserts . Scale bar is 10 μm . Inserts are 3 . 3 μm x 3 . 3 μm or 39x zoom of lower magnification images . Note that the exposure times for GFP signal that is diffuse throughout the ER are not the same as those for GFP-ATF6α in foci ( see Materials and methods ) . ( F ) Quantification of the correlation of GFP-ATF6α with Sec31A within Sec31A / Sec16 positive ERES for a single experiment where for at least 6 cells per condition , each cell imaged as a stack of 6 slices . Plotted is the mean and standard deviation of the mean per cell correlation of GFP-ATF6α and Sec31A . Statistical analysis used unpaired two-tailed t-tests . *** indicates p< 0 . 0009 , **** indicates p= 0 . 0001 , ns stands for non-significant . Variances were not different between treatments ( F test ) . ( G ) The mean number of ERES analyzed per cells . Plotted is mean number of ERES measured per cell from either two or three independent experiments . None of the treatments were reproducibly statistically significantly different from each other . ( H-I ) Time lapse images of U2-OS cells stably expressing GFP-ATF6α ( green ) and transiently transfected with mRFP-p125A ( purple ) to mark ERES . Cells were treated with ER stressor ( 100 nM Tg ) in the absence ( H ) or presence ( I ) of Ceapin-A7 and images were acquired at one frame every five minutes . Scale bar is 10 μm . Images are representative of at least three independent experiments . DOI: http://dx . doi . org/10 . 7554/eLife . 11880 . 02510 . 7554/eLife . 11880 . 026Figure 6—figure supplement 1 . In the presence of Ceapins , GFP-ATF6 is no longer selected for transport from the ER in COP II vesicles . Quantification of the correlation of GFP-ATF6α with Sec31A within Sec31A / Sec16 positive ERES for a two or three independent experiments . Plotted is the mean per experiment of the mean per cell correlation of GFP-ATF6α and Sec31A . For each treatment , at least 41 cells per treatment ( three experiments ) or 21 cells per treatment ( two experiments ) in total were analyzed . DOI: http://dx . doi . org/10 . 7554/eLife . 11880 . 026 As expected , induction of ER stress in the presence of active Ceapin analogs produced GFP-ATF6α foci; however , these foci were non-overlapping with the SEC16/SEC31A positive ER exit sites ( Figure 6C and D ) , while the inactive Ceapin analog A5 did not prevent co-localization of GFP-ATF6α with SEC16 / SEC31A ( Figure 6E ) . In all cases , quantification confirmed the visual impression ( Figure 6F and Figure 6—figure supplement 1 ) . While we could observe a trend for fewer exit sites in ER stressed cells , the mean number of ERES counted per cell was not statistically different between conditions ( Figure 6G ) . Using live cell imaging , we followed GFP-ATF6α trafficking in response to ER stress in U2-OS cells that were transiently transfected with a fluorescent marker for ER exit sites , mRFP-p125A ( Klinkenberg et al . , 2014 ) ( Figure 6H ) . At a thirty-minute time point after ER stress induction , GFP-ATF6α co-localized with mRFP-p125A ( Figure 6H , top panels ) . In contrast , when we treated the cells with ER stressor and active Ceapin-A7 , GFP-ATF6α foci remained distinct from mRFP-p125A foci ( Figure 6I ) . Taken together with our previous data showing that cleavage of neither SREBP in response to low cholesterol nor ATF6β in response to ER stress , which also require COPII mediated transport from the ER to the Golgi apparatus and both S1P and S2P , are inhibited by Ceapins [accompanying manuscript; Gallagher et al . , 2016] , it is clear that Ceapins do not block COPII-mediated trafficking . Instead , our data suggest that Ceapins act to selectively prevent ATF6α being selected as cargo to leave the ER during ER stress ( Figure 7 ) . 10 . 7554/eLife . 11880 . 027Figure 7 . Model: Ceapins trigger an oligomeric state of ATF6α resembling the adapted / attenuated state . ( A ) Ceapin treatment inhibits COPII-mediated trafficking of ATF6a from the ER to the Golgi . In unstressed cells , ATF6α is in foci in the ER . Upon ER stress , ATF6a moves from the ER to the Golgi , where it is proteolyzed to release ATF6α-N that translocates to the nucleus and activates transcription of ATF6α target genes . After prolonged ER stress ATF6α signaling attenuates and ATF6α foci are seen in the ER . The appearance and stability of these foci are increased by Ceapin treatment in the absence or presence of ER stress . ( B ) In this model , Ceapin treatment stabilizes the inactive or attenuated form of ATF6a , which we hypothesize is an oligomeric state of ATF6α . In contrast , the monomeric form is stabilized by unfolded proteins and this form is capable of binding to the COPII coat and exiting the ER upon ER stress . DOI: http://dx . doi . org/10 . 7554/eLife . 11880 . 027
The mechanism of ATF6α activation has remained poorly understood . Here we show that one of the earliest steps in ATF6α activation – its selection as cargo to leave the ER – can be inhibited using Ceapins , a newly identified chemical scaffold . In cells treated with Ceapins , ATF6α clusters in foci that do not leave the ER , because it no longer engages with ER exit sites upon ER stress . Consequently , ATF6α is trapped in its uncleaved , inactive state . Importantly , Ceapins no longer inhibit proteolysis and activation of ATF6α upon lifting the requirement for ATF6α trafficking from the ER for proteolytic processing by relocalizing the Golgi-resident S1P and S2P proteases back to the ER . This is in contrast to the S1P inhibitor , which as expected , inhibits S1P independently of its subcellular localization . Thus Ceapins do not drive ATF6α into an uncleavable form that would occlude the interaction with S1P . This Ceapin-mediated inhibition of ER-to-Golgi trafficking is selective for ATF6α without inhibiting its closely related homolog ATF6β whose transport is similarly regulated by ER stress [accompanying manuscript; Gallagher et al . , 2016] . This is the first evidence that there is a difference in the activation mechanisms employed by these two highly similar proteins . To date , ATF6α and ATF6β were thought to be equivalent in their mechanism of stress-sensing and trafficking . These data become even more surprising , because the lumenal domains of ATF6α and ATF6β are highly conserved and regulate ER stress-sensing and ER exit . Similarly , Ceapins do not interfere with SREBP signaling in response to cholesterol depletion , further underscoring that neither stress-sensing , cargo selection , nor ER-Golgi transport are generally affected by the drug . Ceapins could act on either side of the membrane to inhibit ATF6α trafficking . Its high level of selectivity strongly suggests that Ceapins act by binding to ATF6α or to an ATF6α–dedicated accessory factor that remains to be identified . How ATF6α is retained in the ER in unstressed cells and how it is allowed to enter transport vesicles upon ER stress remains unknown . Our data showing that Ceapins trap ATF6α in the ER membrane in foci distinct from ER exit sites suggest a simple model of its action ( Figure 7 ) : Ceapins bind to ATF6α or some unknown accessory factor , stabilizing an oligomeric state that is not transport competent . In an oligomer , the interface required for interaction with the COPII coat could be buried , while in its monomeric form ATF6α would be recognized as cargo . This model is attractive because of the speed with which ATF6α foci form and dissolve after addition or removal of Ceapins ( Figure 2 ) . It is further supported by data showing that all active Ceapin analogs tested induced foci formation of ATF6α . In contrast , inactive Ceapin analogs either did not induce ATF6α foci or induced unstable foci that quickly disassembled during drug treatment . Moreover , dissolution of foci after washout of Ceapin analogs restored ER stress-induced transport to the Golgi apparatus . Foci formation of ATF6α is also observed in the absence of Ceapin . We found that unlike over-expressed ATF6α fusion proteins , endogenous ATF6α in unstressed cells is found in foci , suggesting that the resting , ER-retained state of ATF6α is an oligomer or higher order complex . In addition , ATF6α signaling attenuates in the face of unmitigated ER stress: although ATF6α is still plentiful in the ER , its proteolytic cleavage to produce ATF6α-N is only observed at early time points after induction of ER stress ( Haze et al . , 2001; Rutkowski et al . , 2006 ) . Consistent with this , we observed formation of ATF6α foci in cells treated with ER stress alone at time points at which attenuation of ATF6α signaling occurs . Our model poses that activation of ATF6α is achieved by shifting the equilibrium from a higher-order complex to a smaller entity ( most likely an ATF6α monomer [Nadanaka et al . , 2007] ) that can be productively recruited into transport vesicles and that attenuation is achieved by shifting the equilibrium back . Ceapins prevent the formation of the transport-competent form even in the presence of ER stress . In this view , Ceapins engage the mechanisms that control normal retention/attenuation of ATF6α . Our model is consistent with data showing that ATF6α exists in monomeric and oligomeric forms in the ER but that only the monomeric form is found in the Golgi apparatus ( Nadanaka et al . , 2007 ) . Both stress-induced ER-Golgi trafficking and oligomerization of ATF6α are regulated by its stress-sensing lumenal domain . Thus the ATF6α ER-lumenal sensor domain would respond to stress in the ER by conversion from an oligomer to a monomer that would allow the information to be transmitted across the membrane to the cytosolic side initiating interactions with the COPII trafficking machinery . Among signaling transmembrane proteins there is precedence for conformational changes on one side of the bilayer leading to subsequent changes on the other side – notably for SCAP , the cholesterol sensor / trafficking adaptor of SREBP ( Sun et al . , 2007; Motamed et al . , 2011 ) , but also for the epidermal growth factor receptor ( EGFR ) ( Endres et al . , 2013 ) . Conformational epitopes that regulate COPII mediated trafficking have also been shown for Sec22 ( Mancias and Goldberg , 2007 ) and for the potassium channel Kir2 . 1 ( Ma et al . , 2011 ) . How ER stress is sensed by ATF6α is unknown . While it has been shown that ATF6α activation correlates with the release of BiP ( Shen et al . , 2002; 2005 ) , there is also evidence for ATF6α activation induced by direct ligand binding . These two principles are not mutually exclusive , as previously reconciled for the activation principles of IRE1 ( Pincus et al . , 2010 ) . Evidence for activation by ligand binding comes from models of physiological ER stress in cardiac cells . In this system , ER stress induces expression of thrombospondin ( Thbs4 ) , which then binds to the lumenal domain of ATF6α and activates ATF6α signaling ( Lynch et al . , 2012 ) . As with IRE1 , ATF6α may bind newly accumulating misfolded proteins directly as ligands to activate signaling ( Gardner and Walter , 2011 ) . We envision that Ceapins act as antagonists to prevent such binding and stabilize ATF6α in its inactive state .
Human bone osteosarcoma ( U2-OS ) cells were obtained from the American Type Culture Collection ( ATCC HTB-96 , ATCC Manassas , VA ) . U2-OS cells stably expressing GFP-HsATF6α were purchased from Thermo Scientific ( 084_01 ) and cultured with 500 μg/mL G418 ( Roche 04 727 878 001 ) to maintain expression of GFP-ATF6α . 293 T-REx cells expressing doxycycline-inducible 6xHis-3xFLAG-HsATF6α are derived from ( Tet ) -ON 293 human embryonic kidney ( HEK ) cells ( Clontech , Mountain View , CA ) ( Cohen and Panning , 2007 ) and are described elsewhere ( Sidrauski et al . , 2013 ) [accompanying manuscript; Gallagher et al . , 2016] . Commercially available cell lines were authenticated by DNA fingerprint STR analysis by the suppliers . All cell lines were visually inspected using DAPI DNA staining and tested negative for mycoplasma contamination . Growth media was DMEM with high glucose ( Sigma D5796 ) supplemented with 10% FBS ( Life technologies , Carlsbad , CA # 10082147 ) , 2 mM L-glutamine ( Sigma G2150 ) , 100 U penicillin 100 μg/mL streptomycin ( Sigma P0781 ) . Two days prior to compound addition 300 μL of 1 . 375 x 104 U2-OS-GFP-ATF6α cells per ml were plated per well in 96 well imaging plate ( ibidi , Madison , WI 89626 ) and sealed with breathable seals ( E&K Scientific , Santa Clara , CA T896100 ) . Immediately prior to addition to cells , compounds were diluted to 6x in media from 500x DMSO stock and 60 μL 6x was added to cells for 1x final ( 0 . 2% DMSO ) . After 5 hr , media was removed and cells were fixed in 4% paraformaldehyde ( PFA ) ( Electron Microscopy Sciences , Hatfield , PA 15714 ) in PHEM buffer ( 60 mM PIPES , 25 mM HEPES , 10 mM EGTA , 2 mM MgCl2-hexahydrate , pH 6 . 9 ) for 15 min RT . Cells were permeabilized with PHEM-Tx ( PHEM containing 0 . 1% Triton X-100 , two washes , 5 min RT ) , washed twice in PHEM , blocked in PHEM containing 2% normal goat serum ( Jackson Immunoresearch Laboratories , West Grove , PA , 005-000-121 ) for 1 hr RT . Primary antibodies were incubated in blocking solution overnight at 4 degrees . Cells were washed three times in PHEM-Tx then incubated with secondary antibodies and nuclear stain ( DAPI , Molecular Probes , Eugene , OR , D-1306 , 5 μg/mL ) in blocking solution for 2 hr RT protected from light . Cells were washed three times PHEM-Tx , twice PHEM . Antibodies used were rat anti-GRP94 9G10 ( abcam , Cambridge , MA ab2791 ) , mouse anti-GFP 3E6 ( Invitrogen , Carlsbad , CA , A11120 ) , anti-rat-Alexa-555 ( Invitrogen A21434 ) , anti-mouse-Alexa-488 ( Invitrogen A11029 ) , each at 1:1000 dilution . Plate was imaged on a spinning disk confocal with Yokogawa CSUX A1 scan head , Andor iXon EMCCD camera ( Andor USA , South Windsor , CT ) and 20x Plan Apo Objective NA 0 . 79 ( Nikon , Melville , NY ) . Using the μManager high-content screening plugin 'HCS Site Generator' ( Edelstein et al . , 2014 ) 49 fields per well were acquired for mean cell number per well of 368 ± 12 . Images were analyzed using CellProfiler ( Carpenter et al . , 2006 ) , MATLAB R2014a ( Mathworks , Natick , MA ) and GraphPad Prism version 5 . 0f ( GraphPad Software , La Jolla , CA ) as previously described [accompanying manuscript; Gallagher et al . , 2016] . Masks for the ER and nucleus of each cell were created using the GRP94 and DAPI staining respectively . The ratio of the GFP intensity in the nucleus versus the ER was calculated for each cell and plotted as a histogram per well . A threshold for the minimum ratio of nuclear to ER signal corresponding to an activated ( i . e . nuclear localized ATF6 ) cell was calculated as the minimum nuclear: ER ratio greater than 1 where the number of ER stressed cells ( Tg ) was greater than the corresponding unstressed control . Percent activation per well was calculated as the percentage of cells per well with a nuclear: ER ratio greater than the calculated threshold for that plate . Mean percent activation per well for a minimum of three replicate wells per treatment was plotted; error bars are 95% confidence limits . Compounds are annotated as hits if they show percent activation more than three standard deviations lower than the mean of ER-stress treated control . U2-OS cells expressing GFP-ATF6α were plated at 7500 cells per well in eight-well ibiTreat μSlide ( ibidi 80826 ) in growth media containing 500 μg/mL G418 two days prior to compound addition . For imaging , growth media was replaced with 250 μL imaging media per well . Imaging media is Leibovitz’s L-15 medium , no phenol red ( Life Technologies 21083–027 ) supplemented with 10% fetal bovine serum ( heat-inactivated , Life Technologies 10082147 ) , 2 mM L-glutamine ( Sigma Aldrich G2150 ) , 100 U penicillin and 0 . 1 mg/mL streptomycin ( Sigma Aldrich P0781 ) and 0 . 45% glucose . Immediately prior to addition to cells , compounds were diluted to 6x in imaging media from 1000x DMSO stock and 50 μL 6x was added to cells for 1x final ( 0 . 2% DMSO ) . For longer time courses ( >5 hr ) final DMSO concentration was 0 . 034% . Cells were imaged at 37 degrees Celsius on a spinning disk confocal with Yokogawa CSUX A1 scan head , Andor iXon EMCCD camera and 40x Plan Apo air Objective NA 0 . 95 with a 1 . 5x tube lens for additional magnification giving 60x final . Three positions per well were marked and imaged per experiment . Images were acquired using 488 nm laser at a rate of one frame per minute with 250 ms exposure time for each . At least five images were acquired per well prior to addition of compounds . Compounds were added between frames , two wells per addition . The frame preceding compound addition for each well was annotated t = 0 min . Cells were plated , vehicle or Ceapins were added and imaging was initiated as described above for live imaging . For washout , fifteen minutes after Ceapin analog or vehicle addition media was removed , wells were washed with 300 μL of PBS; PBS was removed and replaced with 300 μL of imaging media containing either DMSO or 100 nM Tg to induce ER stress . DMSO concentration was equal between all wells for both foci formation and induction of ER stress . Imaging used the 40x Plan Apo air Objective NA 0 . 95 either with or without the tube lens set to 1 . 5x additional magnification . Cells were plated in eight well μslides as described above for live imaging . Cells in growth media were treated with either 0 . 1 μg/ml cycloheximide in ethanol ( Sigma C7698 ) or ethanol alone as vehicle for three hours prior to imaging . Prior to imaging , cells were washed once with 300 μL PBS containing either cycloheximide or ethanol and then placed in 250 μL imaging media containing either cycloheximide or ethanol . Immediately prior to addition to cells , compounds were diluted from 1000x DMSO stock to 6x in imaging media containing either cycloheximide or ethanol and 50 μL 6x was added to cells for 1x final . During imaging , cells were treated either with Ceapin-A1 ( 10 μM in DMSO ) or DMSO alone as vehicle in the presence of either cycloheximide or ethanol . For washout , sixteen minutes after Ceapin analog or vehicle addition media was removed , wells were washed with 300 μL of PBS containing either cycloheximide or ethanol; PBS was removed and replaced with 300 μL of imaging media containing either cycloheximide or ethanol . Final ethanol and DMSO concentration was equal between all wells ( 0 . 1% each ) . Imaging used the 20x Plan Apo air Objective NA 0 . 75 and images were acquired once per minute . In the presence of ethanol , Ceapin-A7 induced foci did not wash out and so this analog could not be tested in this experiment . 293 T-REx cells expressing doxycycline-inducible 6xHis-3xFLAG-HsATF6a were plated at 20000 cells per well in an eight-well collagen IV coated μSlide ( Ibidi 80822 ) in growth media one day prior to compound addition . After five hours , doxycycline ( Sigma D9891 ) was added to 50 nM final . Immediately prior to addition to cells , compounds were diluted to 10x in media from 1000x DMSO stock and 30 μL 10x was added to cells for 1x final ( 0 . 2% DMSO ) . After thirty minutes media containing compounds was removed and cells were fixed in 4% PFA in PHEM buffer as described above for U2-OS cells . Cells were permeabilized with PHEM containing 0 . 1% Triton X-100 ( Sigma T9284 ) ( PHEM-Tx ) , three washes each five minutes at room temperature . After two washes with PHEM cells were blocked in PHEM buffer containing 2% normal goat serum ( Jackson Immunoresearch Laboratories 005-000-121 ) ( blocking solution ) for 1 hr RT . Primary antibodies were incubated in blocking solution overnight at 4 degrees . Cells were washed three times in PHEM-Tx then incubated with secondary antibodies in blocking solution for 2 hr RT protected from light . Cells were washed three times PHEM-Tx with the second wash containing 5 μg/mL DAPI . Prior to imaging cells were washed twice with PHEM buffer . Antibodies used were mouse anti-FLAG M2 ( Sigma F1804 ) , rabbit anti-Calnexin ( Cell Signaling Technology , Danvers , MA , 2679S ) , rabbit anti-Giantin ( abcam 24586 ) , anti-mouse-Alexa-488 ( Invitrogen A11029 ) , anti-rabbit-Alexa-546 ( Invitrogen A11010 ) , each at 1:1000 dilution except anti-Calnexin , which was used at 1:100 . Slides were imaged on a spinning disk confocal with Yokogawa CSUX A1 scan head , Andor iXon EMCCD camera and 100x ApoTIRF objective NA 1 . 49 ( Nikon ) . U2-OS cells expressing GFP-ATF6α were plated at 7500 cells per well in eight-well ibiTreat μSlide ( Ibidi 80826 ) in growth media containing 500 μg/mL G418 two days prior to compound addition . Immediately prior to addition to cells , compounds were diluted to 6x in media from 1000x DMSO stock and 50 μL 6x was added to cells for 1x final ( 0 . 2% DMSO ) . After compound incubation ( various times ) media containing compounds was removed and cells were washed once quickly in PBS . Ice-cold methanol was added to fix and permeabilize cells and the slides were incubated for five minutes at minus thirty degrees Celsius . Cells were washed three times four minutes each with PHEM buffer and then blocked in PHEM buffer containing 2% normal goat serum ( Jackson Immunoresearch Laboratories 005-000-121 ) ( blocking solution ) for 1 hr RT . Primary antibodies were incubated in blocking solution overnight at 4 degrees . Cells were washed three times in PHEM buffer then incubated with secondary antibodies in blocking solution for 2 hr RT protected from light . Cells were washed four times PHEM buffer , if necessary the second wash containing 5 μg/mL DAPI . Antibodies used were mouse anti-Sec31A ( BD Biosciences , San Jose , CA , 612351 ) , rabbit anti-Sec16 KIAA0310 ( Bethyl Laboratories , Montgomery , TX , A300-648A ) , rabbit anti-Giantin ( abcam 24586 ) , rat anti-GRP94 9G10 ( abcam ab2791 ) , anti-mouse-Alexa-405 ( Invitrogen A31553 ) , anti-rabbit-Alexa-546 ( Invitrogen A11010 ) , anti-rabbit-Alexa-633 ( Invitrogen A21071 ) , anti-rat-Alexa-633 ( Invitrogen A21094 ) each at 1:1000 dilution . Slides were imaged on a spinning disk confocal with Yokogawa CSUX A1 scan head , Andor iXon EMCCD camera and 100x ApoTIRF objective NA 1 . 49 ( Nikon ) . For analysis of ERES the exposure time for GFP-ATF6α in cells treated with active Ceapin analogs was shortened to prevent overexposure inflating the size of GFP-ATF6α foci . In unstressed and stressed controls , the GFP-ATF6α signal is distributed throughout the ER and is dimmer . U2-OS cells ( no reporters ) were plated at 7500 cells per well in eight-well ibiTreat μSlide ( Ibidi 80826 ) in growth media two days prior to compound addition . Immediately prior to addition to cells , compounds were diluted to 6x in media from 1000x DMSO stock and 50 μL 6x was added to cells for 1x final ( 0 . 2% DMSO ) . After two hours of compound treatment , cells were fixed and stained as for nuclear translocation assay in GFP-ATF6α expressing U2-OS cells ( described above ) with the following changes . Primary antibodies used were rabbit polyclonal anti-ATF6α ( 1:250 , generous gift from Kazutoshi Mori ) , rat anti-GRP94 9G10 , ( 1:1000 , abcam ab2791 ) and purified mouse anti-GM130 clone 35 ( 1:250 , BD Biosciences 610823 . Secondary antibodies were all raised in goat and used at 1:1000 dilution - anti-rabbit-Alexa-488 ( Invitrogen A11034 ) , anti-rat-Alexa-555 ( Life Technologies A21429 ) and anti-mouse-633 ( Invitrogen A21050 ) . Nuclear stain ( DAPI , Molecular Probes D-1306 , 5 μg/mL ) was added in second of four PHEM-Tx washes after secondary antibody incubation and cells were washed twice in PHEM buffer before imaging in PHEM buffer . Slides were imaged on a spinning disk confocal with Yokogawa CSUX A1 scan head , Andor iXon EMCCD camera and 100x ApoTIRF objective NA 1 . 49 ( Nikon ) . Images were analyzed using CellProfiler ( Carpenter et al . , 2006 ) , MATLAB R2014a and GraphPad Prism 5 as previously described [accompanying manuscript; Gallagher et al . , 2016] . Masks for the ER and nucleus of each cell were created using the GRP94 and DAPI staining respectively . The ratio of the GFP intensity in the nucleus versus the ER was calculated for each cell and plotted as a boxplot . Statistics performed were unpaired two-tailed t-tests; similar results were obtained using one-way analysis of variance ( ANOVA ) . Two days prior to drug treatment 2 x 105 6xHis-3xFLAG-HsATF6α 293 T-REx cells per well were plated in 24 well plates ( Corning , Corning , NY , 3526 ) . The following day , expression of tagged ATF6α was induced using 50 nM doxycycline . Eighteen hours later either Brefeldin A ( 5 μg/mL final , in ethanol , Sigma Aldrich B6542 ) or ER stressor ( 100 nM Tg final , in DMSO , Sigma T9033 ) with or without inhibitors was added to cells and incubated for one hour . Vehicle was added to ensure the final concentration of either ethanol or DMSO was the same for all samples . Inhibitors used were S1P inhibitor ( Pf-429242 , Pfizer , New York , NY ) or Ceapin analogs Ceapin-A1 , Ceapin-A3 or the inactive Ceapin analog A5 . All compounds were added from 1000x stock solutions . After one hour , media was removed , and 200 μL of scraping buffer ( 10 μM MG132 ( Sigma C2211 ) , 1x complete protease inhibitor ( Roche Diagnostics , Pleasanton , CA , 05056489001 ) in phosphate buffered saline ( PBS , Sigma Aldrich D8537 ) ) was added to each well . Cells were scrapped into 1 . 5 mL eppendorf tubes , centrifuged at 3000 x g for five minutes at four degrees . Each cell pellet was resuspended in 50 μL 5x lysis buffer ( 200 mM Tris-HCl pH 8 . 0 , 1% SDS , 40 mM dithiothreitol , 30% glycerol , pinch of bromophenol blue ) supplemented with 10 μM MG132 and 1x complete protease inhibitor . Lysates were incubated on ice for twenty minutes , vortexed at full speed for five minutes at four degrees , incubated on ice for a further ten minutes , boiled for five minutes and centrifuged at 1000 x g for one minute at room temperature prior to loading . 7 . 5 μL of each sample was loaded on fifteen well ten percent mini-protean TGX gels ( Bio-Rad Laboratories , Hercules , CA , 4561036 ) . Gels were blotted onto 0 . 2 mM nitrocellulose membrane ( Perkin Elmer , Santa Clara , CA , NBA083C00 ) and western blotted according to standard techniques . Blocking solution was 5% milk in PBS-Tween . Antibodies used were mouse anti-FLAG ( M2 , Sigma A2220 ) and rabbit anti-GAPDH ( abcam ab9485 ) , anti-mouse-HRP conjugate ( Promega Corporation , Madison , WA , W4021 ) and anti-rabbit-HRP conjugate ( Promega Corporation W4011 ) . Horseradish peroxidase substrate ( SuperSignal West Dura Extended Duration Substrate , Pierce Biotechnology , Rockford , IL , 34075 ) and Kodak X-OMAT film ( Fisher Scientific , Waltham , MA , IB1651496 ) were used to detect protein bands . Images were analyzed using CellProfiler 2 . 1 . 1 . Briefly , the Sec31A and Sec16 images for each slice of each stack were multiplied so that only pixels containing both Sec31A and Sec16 fluorescence would be non-zero . In this resulting image , ERES were identified as objects with a diameter range of 0 . 167 – 1 . 67 μm . Thresholding was automatic and clumped objects were separated based on intensity . The resulting outlines of ERES were used as masks to count the intensity of Sec31A ( 405 nm ) , Sec16 ( 561 nm ) , GFP-ATF6α ( 488 nm ) and GRP94 ( 633 nm ) within ERES in the original images . The correlation between fluorophores was calculated on a pair-wise basis for all four and the results for correlation of Sec31A and GFP-ATF6α in double Sec31A / Sec16 positive punctae are shown in Figure 6 . Data from CellProfiler was imported into MatLab R2014a and organized by slice , cell and compound treatment . Results were imported into GraphPad Prism version 5 . 0 for statistical analysis and plotting . CellProfiler Pipelines and MatLab scripts are provided as source code files . Three independent experiments were analyzed . One day prior to transfection , U2-OS cells stably expressing GFP-ATF6α were plated at a density of 7500 cells per well of an eight well ibiTreat μslide ( Ibidi 80826 ) in growth media without 500 μg/mL G418 . The following day , cells were washed once in PBS ( Sigma Aldrich D8537 ) and growth media without antibiotics or selection agent was added . 2 ng of pmRFP-p125A ( kind gift from Meir Aridor ) and 78 ng of salmon sperm DNA ( carrier DNA , Life Technologies 15632011 ) were transfected per well using Fugene HD transfection reagent ( Promega E2311 ) in OptiMEM reduced serum media ( Gibco 31985 ) according to manufacturers instructions . After six hours , media was replaced with growth media supplemented with 500 μg/mL G418 to ensure expression of GFP-ATF6 . The transient transfection protocol was optimized to prevent transfection-induced ER stress ( as visualized by ER versus nuclear localization of GFP-ATF6α in transfected cells ) and to minimize over-expression of pRFP-125A to prevent distortion of ERES . Only unstressed cells with normal ER and ERES morphology were selected for imaging . Twenty-four hours after transfection cells media was exchanged for imaging media ( as described above ) . Cells were imaged on a spinning disk confocal with Yokogawa CSUX A1 scan head , a 100x ApoTIRF objective NA 1 . 49 ( Nikon ) and a 565 nm long pass filter to split the emission light between two Andor iXon EMCCD cameras . For each cell , a z-stack with 0 . 25 μm steps was acquired every five minutes . Compounds were mixed to 6x from 1000x DMSO solutions in imaging media and added to cells for 1x final after acquisition of two z-stacks per cell . Post-acquisition , images from the cameras were aligned using the GridAligner plugin for ImageJ written by Nico Stuurman ( http://valelab . ucsf . edu/~nstuurman/ijplugins/GridAligner . html ) with the affine matrix calculated from reference images taken of a NanoGrid ( an array of sub-wavelength sized holes , Miraloma Tech LLC , San Francisco , CA , A00011 ) using trans-illumination imaging . This imaging set-up is not suitable for long-term imaging of the cells . | Newly made proteins must be folded into specific three-dimensional shapes before they can perform their roles in cells . Many proteins are folded in a cell compartment called the endoplasmic reticulum . The cell closely monitors the quality of the work done by this compartment . If the endoplasmic reticulum has more proteins to fold than it can handle , unfolded or misfolded proteins accumulate and trigger a stress response called the unfolded protein response . This increases the capacity of the endoplasmic reticulum to fold proteins to match the demand . However , if the stress persists , then the unfolded protein response instructs the cell to die to protect the rest of the body . A protein called ATF6α is one of three branches of the unfolded protein response . This protein is found in the endoplasmic reticulum where it is inactive . Endoplasmic stress causes ATF6α to move from the endoplasmic reticulum to another compartment called the Golgi apparatus . There , two enzymes cut ATF6α to release a fragment of the protein that then moves to the nucleus to increase the production of the machinery needed to fold proteins in the endoplasmic reticulum . In a related study , Gallagher et al . identified a group of small molecules called Ceapins , which inhibit ATF6α activity . Here , Gallagher and Walter investigate how Ceapins act on ATF6α . The experiments show that Ceapin causes ATF6α molecules to form clusters that prevent the protein from moving to the Golgi apparatus by keeping it away from the machinery that moves proteins between these compartments . When the enzymes that cut ATF6α are sent to the endoplasmic reticulum , Ceapin treatment no longer prevents ATF6α activation , which shows that these small molecules specifically inhibit the stress-induced movement of ATF6α . When Ceapins are washed out of cells , the ATF6α clusters fall apart and ATF6α can now move to the Golgi . These experiments show that ATF6α is actively held in the endoplasmic reticulum by a mechanism that is stabilized by Ceapins . Gallagher and Walter propose that the small clusters of ATF6α in unstressed cells act to keep this protein in the endoplasmic reticulum . However , when cells experience stress , the ATF6α clusters fall apart to allow the protein to move to the Golgi . The next steps following on from this work are to find out what these clusters are , how they are influenced by endoplasmic reticulum stress and exactly how the Ceapins stabilize these clusters . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"cell",
"biology"
] | 2016 | Ceapins inhibit ATF6α signaling by selectively preventing transport of ATF6α to the Golgi apparatus during ER stress |
Fossils were thought to lack original organic molecules , but chemical analyses show that some can survive . Dinosaur bone has been proposed to preserve collagen , osteocytes , and blood vessels . However , proteins and labile lipids are diagenetically unstable , and bone is a porous open system , allowing microbial/molecular flux . These ‘soft tissues’ have been reinterpreted as biofilms . Organic preservation versus contamination of dinosaur bone was examined by freshly excavating , with aseptic protocols , fossils and sedimentary matrix , and chemically/biologically analyzing them . Fossil ‘soft tissues’ differed from collagen chemically and structurally; while degradation would be expected , the patterns observed did not support this . 16S rRNA amplicon sequencing revealed that dinosaur bone hosted an abundant microbial community different from lesser abundant communities of surrounding sediment . Subsurface dinosaur bone is a relatively fertile habitat , attracting microbes that likely utilize inorganic nutrients and complicate identification of original organic material . There exists potential post-burial taphonomic roles for subsurface microorganisms .
Fossils have traditionally been thought to retain little original organic material after undergoing decay and diagenesis . However , recent discoveries of relatively intact macromolecular organic material in fossils and sub-fossils challenge this view . These include ancient DNA ( Meyer et al . , 2012; Orlando et al . , 2013 ) and peptide ( Buckley , 2015; Demarchi et al . , 2016; Cappellini et al . , 2018 ) sequences in sub-fossils , as well as ancient biomolecules such as sterols ( Melendez et al . , 2013 ) , melanin ( Vinther et al . , 2008 ) , amino acids ( Curry et al . , 1991 ) , and porphyrins ( Wiemann et al . , 2018a ) . These findings show that organic remains can potentially persist for thousands to millions of years , depending on the biomolecules and environmental conditions . Such remains have already provided important insights into evolution , including the origins of our species ( Krause et al . , 2010 ) and the affinities of extinct Pleistocene megafauna ( Welker et al . , 2015 ) . In theory , millions to tens of millions of years old organic remains could offer palaeontologists new insights and a unique window into the biology of organisms distantly related to any living species . Such organic molecular fossils could potentially shed light on the biology and evolution of extinct organisms , including their coloration , structure , behavior , and phylogeny , providing unique insights into past life , and the origins of present life . However , it remains unclear how long different types of organic molecules and organic structures can survive and under which conditions . DNA , which is relatively unstable , is thought to persist no longer than a million years under optimal conditions ( Orlando et al . , 2013 ) . In comparison , structural proteins such as collagen are more stable , however , and are predicted to persist for longer ( Nielsen-Marsh , 2002 ) , although how much longer is unclear . Pigments such as melanin and porphyrins are highly stable and can persist for hundreds of millions of years ( Gallegos and Sundararaman , 1985; Vinther , 2015 ) . Dinosaur bone has previously been reported to contain endogenous organic remains such as DNA , collagen , osteocytes , erythrocytes , and blood vessels ( Pawlicki et al . , 1966; Pawlicki and Nowogrodzka-Zagórska , 1998; Schweitzer et al . , 2005a; Schweitzer et al . , 2005b; Schweitzer et al . , 2007a; Schweitzer et al . , 2007b; Schweitzer et al . , 2008; Schweitzer et al . , 2009; Schweitzer et al . , 2013; Schweitzer et al . , 2014; Schweitzer et al . , 2016; Asara et al . , 2007; Organ et al . , 2008; Schweitzer , 2011; Bertazzo et al . , 2015; Cleland et al . , 2015; Schroeter et al . , 2017 ) . These reports , if verified , could change the study of macroevolution and the physiology of extinct organisms , particularly considering the potential for protein sequence data to shed light on the biology and systematics of extinct organisms . Most of these reports rely on structural observations , mass spectrometry , and immunohistochemistry . Sub-fossil and fossil vertebrate remains are primarily composed of bone , dentine , and/or enamel . These represent calcified tissues with both a mineral component , primarily calcium phosphate , and a protein component that is dominated by collagen . As such collagen is a common target in the analysis of ancient organic remains . Collagen is also non-labile relative to many other vertebrate proteins because of its decay resistance , partly due to its triple helical quaternary structure and high concentration of thermally stable amino acids ( Engel and Bächinger , 2005; Persikov et al . , 2005; Sansom et al . , 2010; Wang et al . , 2012 ) , and it is therefore reasonable to predict that collagen would be more resistant to microbial decay and diagenesis than many other proteins . Others have criticized reports of ancient collagen based on mass spectrometric results , suggesting that they represent laboratory or environmental contamination ( Buckley et al . , 2008; Buckley et al . , 2017; Bern et al . , 2009 ) or statistical artefacts ( Pevzner et al . , 2008 ) . The use of antibodies to detect ancient collagen is also problematic since they are known to cause occasional false positives ( True , 2008 ) and have been suggested to do so in fossil samples ( Saitta et al . , 2018 ) . Furthermore , various organic and inorganic demineralization products of fossil bone that morphologically resemble blood vessels , osteocytes , and erythrocytes have alternatively been identified as biofilm or a network of microbiological materials ( Kaye et al . , 2008 ) , degraded and distorted organic contamination ( Saitta et al . , 2017a ) , or minerals such as pyrite/iron oxide framboids ( Martill and Unwin , 1997; Kaye et al . , 2008 ) . Reports of dinosaur protein and complex organic structure preservation are problematic for several reasons . Firstly , it remains unclear how such organics would be preserved for tens of millions of years . If endogenous , putative dinosaur soft tissues should contain diagenetically unstable proteins and phospholipids ( Bada , 1998; Briggs and Summons , 2014 ) , vulnerable to hydrolysis ( Eglinton and Logan , 1991; Zuidam and Crommelin , 1995 ) , although the released fatty acid moieties from phospholipids could be stabilized through in situ polymerization into kerogen-like aliphatic structures ( Stankiewicz et al . , 2000; Gupta et al . , 2006a; Gupta et al . , 2006b; Gupta et al . , 2007a; Gupta et al . , 2007b; Gupta et al . , 2008; Gupta et al . , 2009 ) . At 25°C and neutral pH , peptide bond half-lives from uncatalyzed hydrolysis are too short to allow for Mesozoic peptide preservation , although hydrolysis rates can be decreased through terminal modifications and steric effects on internal bonds ( Kahne and Still , 1988; Radzicka and Wolfenden , 1996; Testa and Mayer , 2003 ) . Estimates based on experimental gelatinization suggest that , even when frozen ( 0°C ) , relatively intact collagen has an upper age limit of only 2 , 700 , 000 years ( Nielsen-Marsh , 2002 ) . Secondly , the instances of dinosaur peptide preservation reported are older than the oldest uncontested protein preservation reported by at least an order of magnitude . The oldest non-controversial peptides include partially intact peptides from 3 . 4 Ma in exceptionally cold environments ( Rybczynski et al . , 2013 ) , as well as short peptides bound to eggshell calcite crystals from 3 . 8 Ma stabilized via unique molecular preservation mechanisms ( Demarchi et al . , 2016 ) . The youngest non-avian dinosaur bones are 66 million years old; on both theoretical and empirical grounds , it seems exceptional that original proteins could persist for so long . Furthermore , a long-term trend of protein loss and increasing contamination in ancient organismal remains , such as bone , has been shown ( Armstrong et al . , 1983; Dobberstein et al . , 2009; High et al . , 2015; High et al . , 2016 ) . Fossil bones are open systems capable of organic and microbial flux ( Bada et al . , 1999 ) . Such a system might lead not only to the loss of endogenous organics , but also to the influx of subsurface microorganisms that could complicate the detection of any surviving organics , as well as potentially metabolizing them . The possibility of a microbiome inhabiting fossil bone is very high , especially given that decades of research have revealed the existence of a substantial ‘deep biosphere’ of living microorganisms actively degrading everything organic from shallow soil organic matter to deeply buried petroleum ( Onstott , 2016; Magnabosco et al . , 2018 ) , even in million year old permafrost ( Amato et al . , 2010 ) . Since there are theoretical and empirical reasons to believe that dinosaur organics are unlikely to persist for tens of millions of years , and given the potential for contamination , we argue that the null hypothesis is that complex biomolecules ( e . g . nucleic acids or proteins ) recovered from dinosaur bones are not original material , more likely representing recent contamination . This hypothesis makes a series of testable predictions: ( 1 ) organic material recovered from fossil dinosaur bone will differ in composition ( both chemistry and structure ) from modern vertebrate proteins and tissues , beyond differences expected through normal diagenesis; ( 2 ) fossils will show evidence for microbial presence ( e . g . , through nucleic acids or protein ) ; ( 3 ) fossil bone organic material will show signatures of recent biological activity ( e . g . L-amino acid dominance or 14C abundance , which would suggest that the fossils are not isolated from surface processes ) . Here , chemical and molecular analyses of freshly collected , aseptically acquired , Late Cretaceous surface-eroded and excavated subterranean dinosaur bones , when compared to associated sediment and soil , younger fossil , and modern bone controls , show evidence for a contemporary microbiome . Analyses were conducted using variable pressure scanning electron microscopy ( VPSEM ) , energy dispersive X-ray spectroscopy ( EDS ) , light microscopy , attenuated total reflectance Fourier-transform infrared spectroscopy ( ATR FTIR ) , pyrolysis-gas chromatography-mass spectrometry ( Py-GC-MS ) , high-performance liquid chromatography ( HPLC ) , radiocarbon accelerator mass spectrometry ( AMS ) , Qubit fluorometer , epifluorescence microscopy ( propidium iodide ( PI ) and SYTO 9 staining ) , and 16S rRNA gene amplicon sequencing . In addition to finding little evidence for the preservation of original proteinaceous compounds , our findings suggest that bones not only act as open systems just after death and exhumation , but also act as favorable habitats as fossils in the subsurface . Microbial communities appear to be localized inside the dinosaur bones collected here .
Samples of Late Cretaceous fossil dinosaur bone , along with associated sediment and soil controls were obtained from the Dinosaur Park Formation ( Late Campanian ) in Dinosaur Provincial Park , Alberta , Canada ( Appendix 1—figures 8–20 , Appendix 1—table 2 ) . The Dinosaur Park Formation is a well-sampled , alluvial-paralic unit deposited during a transgressive phase of the Western Interior Seaway . A diverse vertebrate fauna has been documented from the formation by more than a century of collection ( Currie and Koppelhus , 2005 ) . The bone samples were collected from a monodominant bonebed ( BB180 ) of the centrosaurine Centrosaurus apertus ( Ornithischia; Ceratopsidae ) , located 3 m above the contact with the underlying Oldman Formation ( precise location data available at the Royal Tyrrell Museum of Palaeontology ) . The mudstone-hosted bone-bearing horizon is an aggregation of disarticulated but densely packed bones , with a vertical relief of 15–20 cm . Similar to other ceratopsid bonebeds from the same stratigraphic interval ( Ryan et al . , 2001; Eberth and Getty , 2005 ) , the recovered skeletal remains are nearly exclusively from Ceratopsidae , and with all diagnostic ceratopsid material assignable to Centrosaurus apertus , with the site interpreted as a mass-death assemblage . Fossil material was collected under a Park Research and Collection Permit ( No . 16–101 ) from Alberta Tourism , Parks and Recreation , as well as a Permit to Excavate Palaeontological Resources ( No . 16–026 ) from Alberta Culture and Tourism and the Royal Tyrrell Museum of Palaeontology , both issued to CM Brown . Sandstone and mudstone overburden was removed with pick axe and shovel ( ~1 m into the hill and ~1 m deep ) to expose a previously unexcavated region of the bonebed , stopping within ~10 cm of the known bone-bearing horizon . A few hours after commencement of overburden removal , excavation of the mudstone to the bone-bearing horizon was conducted using awl and scalpel . Subterranean Centrosaurus bones ( identified as a small rib and a tibia ) were first discovered and exposed to the air under typical paleontological excavation techniques to allow for rapid detection of bones . At this point , aseptic techniques were then implemented to expose more of the bone in order to determine its size and orientation . It is necessary to qualify the usage of the term ‘aseptic’ in this study . Paleontological field techniques have changed little over the last century , and it is practically impossible to excavate fossils in a truly sterile manner ( e . g . the process of matrix removal induces exposure , the wind can carry environmental contaminants onto exposed fossils , etc . ) . Considering this , the term ‘aseptic’ is used here to acknowledge the inability to provide completely sterile sampling conditions , while still indicating that efforts were taken to minimize contamination of the samples . Our success at reasonably reducing contamination is supported by the fact that our samples yielded consistent and interpretable results . During aseptic excavation and sampling , nitrile gloves washed in 70% ethanol and a facemask were worn . All tools ( i . e . awl , scalpel , Dremel saw ) were sterilized with 10% bleach , followed by 70% ethanol , and then heat-treated with a propane blowtorch at the site . Bone samples several centimeters long were obtained using a diamond-coated Dremel saw or utilizing natural fractures in the bone . Certain segments of the bones , designated herein as matrix-surrounded subterranean Centrosaurus bone , were sampled without first removing the surrounding matrix , although fractures in the mudstone did appear during sampling so that the samples cannot be said to have been unexposed to the air , especially prevalent in the small rib sample sent to Princeton University for analysis . Also sampled were the aseptically excavated but completely exposed portions of the subterranean bone immediately next to the matrix-surrounded region , designated herein as uncovered subterranean Centrosaurus bone . In other words , these were the regions of the bone fully exposed using aseptic techniques after initial discovery of the bone in order to determine size and orientation . All samples were collected in autoclaved foil without applying consolidants , placed in an ice cooler kept in the shade , and brought back to the field camp freezer that evening . Additionally , surface-eroded bone from BB180 and on the same ridge above BB180 , mudstone excavated from the overburden-removed area of BB180 and several cm below the weathered surface of the same ridge above BB180 , and topsoil on the same ridge above BB180 were similarly aseptically acquired and stored ( i . e . sterile tools , foil , and personal wear; kept cool ) . In total , eight bone samples , eight sediment samples , and two soil samples were collected . Samples were transported to the Royal Tyrrell Museum of Palaeontology in a cooler . Following accession at the museum , similar sets of samples were either mailed to Princeton on ice or transported via plane to Bristol without refrigeration with a maximum time unrefrigerated under 24 hr ( i . e . both Princeton and Bristol received a sample of matrix-surrounded bone , BB180 mudstone , topsoil , etc . ) . Note that warming after cold storage could lead to condensation , altering the behavior of any potential microbiome . Upon arrival , samples were stored at 4°C in Bristol or −80°C in Princeton as required for analysis . The aseptically collected Dinosaur Provincial Park fossil bone , mudstone , and soil samples were compared to younger fossils and modern bone . Chicken ( Gallus gallus domesticus ) bone was obtained frozen from a Sainsbury’s grocery store in Bristol , UK and was kept refrigerated ( 4°C ) . Other controls included amino acid composition data from a reference bone ( fresh , modern sheep long bone ) and radiocarbon data from an 82–71 ka radiocarbon-dead bovine right femur used as a standard from the literature ( Cook et al . , 2012 ) . Black , fossil sand tiger shark teeth ( Carcharias taurus ) eroded from Pleistocene-Holocene sediments were non-aseptically collected from the surface of the sand on a beach in Ponte Vedra Beach , Florida , USA without applied consolidants and stored at room temperature . It should be noted that Florida experiences a high-temperature climate relative to many samples typically studied for palaeoproteomics . Teeth samples represent a mix of dentine and enamel as opposed to normal bone tissue , with relative concentrations depending on how easily the different tissues fragmented during powdering . The decision to include subfossil shark teeth was made based on their ready availability ( i . e . they are incredibly common fossils and are easy to collect from the surface of the sand ) , the minimal loss to science when destructively analyzed due to their ubiquity , and that the protein composition of the tooth dentine would be dominated by collagen , as in bone . Technical grade humic acid was also purchased from Sigma Aldrich as an additional control .
VPSEM and EDS of HCl demineralized , freeze-dried dinosaur bones revealed that vessels ( and rare fibrous fragments ) ( Figure 1A , D–E , H–J ) were white , Si-dominated with O present , contained holes , and were sometimes infilled with a slightly more prominent C peak internally . Vessels occurred alongside white quartz crystals , which had strong Si peaks and overall were elementally similar to the vessels , and smaller reddish minerals , originally presumed to be iron oxide or pyrite , but which had high-Si content with Ba also present . Demineralization products differed from those of chicken bone ( Figure 1C , G , M ) and Pleistocene-Holocene shark tooth ( Figure 1B , F , L ) , which were much more homogenous and consisted of large fibrous masses . These more recent samples were enriched in C , O , N , and S , but the shark tooth also had a strong Fe signature and a relatively more prominent S peak than the chicken bone . The chicken demineralisation product was white , while that of the shark tooth was black . These results show that the dinosaur bone yielded different structures when the bone apatite was removed compared to the more recent bone ( i . e . primarily vessels as opposed to large fibrous masses ) . Furthermore , the dinosaur vessels are relatively inorganic in composition compared to the more recent bone , consistent with a mineralized biofilm ( Schultze-Lam et al . , 1996; Decho , 2010 ) . ATR FTIR of a HCl demineralized , freeze-dried vessel from subterranean Centrosaurus bone revealed somewhat poorly resolved , broad organic peaks ( Figure 2C ) that were close in position to peaks that might be expected from various CH , CO , and amide bonds , as well as water , phosphate , and potentially carbonate and silicate bonds ( Lindgren et al . , 2011; Surmik et al . , 2016; Lee et al . , 2017 ) ; also see publicly available NIST libraries ) . Pleistocene-Holocene shark tooth ( Figure 2B ) and modern chicken bone ( Figure 2A ) demineralization products similarly revealed peaks consistent with organic and phosphatic peaks , and the chicken bone had particularly strong organic peaks relative to phosphate . Maintaining close contact of the sample to the Ge crystal was difficult , resulting in the poorly resolved peaks , especially in the shark tooth sample . These results show how , although potentially poorly resolved , the ATR FTIR peaks in the dinosaur bone demineralization products could be consistent with various organic bonds present in more recent bone demineralization products . However , note that these bonds are relatively simple and could therefore be present in various organic molecules . Furthermore , they are not necessarily ancient , endogenous , or protein-derived . Data-rich Py-GC/MS results are primarily used here as a fingerprinting method via total ion chromatograms in order to complement the other analyses of this study . Centrosaurus bone had a low pyrolysate yield ( Figure 3B ) as evidenced by the significant column bleed at the end of the run and contained mostly early eluting compounds . Similarly , humic acid also contained many early eluting pyrolysis products ( Figure 3D ) . The pyrogram for Centrosaurus bone does not match that of modern collagen-containing bone ( Figure 3A ) , which contained many clear protein pyrolysis products such as nitriles and amides , and was most similar to mudstone matrix ( Figure 3C ) . Subterranean Centrosaurus bone pyrolysates included alkylated benzenes and some polycyclic aromatic hydrocarbons ( e . g . naphthalenes and fluorenes ) , and these can also be detected in the surrounding mudstone matrix ( Figure 3C ) and humic acid standard ( Figure 3D ) . Weak n-alkane/n-alkene doublets were possibly detected in the Late Cretaceous bones ( Figure 4A–D ) , and such doublets were also observed in the surrounding mudstone matrix ( Figure 3C ) and humic acid standard ( Figure 3D ) . Variation in the conspicuousness of these doublets between the matrix-surrounded and uncovered subterranean Centrosaurus bone samples was apparent ( Figure 5A–D ) . These results show how the dinosaur bone lacked any clear pyrolysis products indicative of high levels of protein preservation and instead had a chemical composition that more closely resembles potential environmental sources ( i . e . mudstone matrix or humic acids ) than bone proteins . Homologous series of n-alkane/n-alkene doublets may signify the presence of a kerogen-like substance which could potentially be an ancient lipid-derived geopolymer in the dinosaur bone . Interpretation of amino acid data is restricted here to only those samples that were prepared to counter peak suppression ( KOH-treated; Dickinson et al . , 2019 ) , although examination of the conventionally prepared ( High et al . , 2016 ) samples results in similar patterns , albeit with more noise ( Appendix 1—figures 21–28 , Appendix 1—tables 3–13 ) . Matrix-surrounded subterranean Centrosaurus bone had a total hydrolysable amino acid ( THAA ) compositional profile that did not match collagen ( Figure 6A , F ) . The matrix-surrounded subterranean Centrosaurus bone appeared to be dominated by Gly , with Tyr also prominent , while being highly depleted in all the other amino acids . Surface-eroded Late Cretaceous bone from the same outcrop showed a different THAA composition to the matrix-surrounded subterranean Centrosaurus bone , even when examining bone eroded out of the BB180 quarry itself ( Figure 6B , F ) . Furthermore , the uncovered subterranean Centrosaurus bone did not match the matrix-surrounded subterranean bone and was similar to the surface-eroded Late Cretaceous bone in THAA composition . Relative Gly concentration in surface-eroded Late Cretaceous bone was not as high as in the matrix-surrounded subterranean Centrosaurus bone , where Gly dominated the compositional profile . The surface-eroded Late Cretaceous bone showed somewhat more similarity to collagen in THAA compositional profile than did the matrix-surrounded subterranean Centrosaurus bone , but ultimately did not align ( Figure 6C , F ) . These results suggest that not only did the subterranean dinosaur bone not have an amino acid composition similar to collagen ( i . e . Gallus and reference bone ) , but also that exposure to the surface changes the amino acid profile within these Cretaceous fossils . Subterranean Centrosaurus bone had a far lower THAA concentration ( summed total of all amino acids measured ) than did the modern chicken bone ( Figure 7A ) , as would be expected , and the uncovered subterranean Centrosaurus bone showed a higher THAA concentration than the matrix-surrounded subterranean Centrosaurus bone ( Figure 7B ) . Surface-eroded Late Cretaceous bone showed relatively high variability in THAA concentrations , with some higher THAA concentrations than subterranean Centrosaurus bone . These results are consistent with the expectation that any potential proteins present in the subterranean dinosaur bone would be reduced in concentration compared to bone in vivo , an expectation that might hold regardless of whether proteins are endogenous or exogenous . Late Cretaceous bone tended to be L-amino acid dominated when amino acids were above detection limit ( Table 1 ) . Surface-eroded Late Cretaceous fossil bone seemed to show more variability in D/L values than the subterranean bone samples . Similar to the samples described here , other non-aseptically collected , room-temperature-stored Jurassic and Cretaceous surface-eroded bones have low amino acid concentrations and lack significant concentrations of D-amino acids ( Appendix 1—tables 3–4 ) . These low levels of racemization suggest that the amino acids in the dinosaur bone are not particularly ancient . The adjacent mudstone matrix did not match the subterranean Centrosaurus bone in THAA compositional profile ( Figure 6D , F ) . Surface-eroded Late Cretaceous bone showed some degree of similarity to topsoil in THAA composition ( Figure 6E , F ) , as did the various mudstone samples . Matrix-surrounded subterranean Centrosaurus bone showed the most different THAA compositional profile within the study ( i . e . greatest separation from other data points in PC space ) . All these groups plotted separately in the PCA from modern collagen ( Figure 6F ) . The greatest variation between the samples of this study was in relative Gly and Tyr concentrations , with the matrix-surrounded subterranean Centrosaurus bone tending to have notably higher Gly and Tyr than collagen . These results suggest that subterranean dinosaur bone had a different amino acid composition than the surrounding mudstone and that the amino acid composition changes upon surface exposure , approaching that of topsoil . Topsoil showed a greater THAA concentration than subterranean and surface-eroded Centrosaurus bones , but not as high as modern chicken bones ( Figure 7A ) . Mudstone tended to have a very low THAA concentration , even compared to some of the Late Cretaceous bone samples ( Figure 7B ) . The highest THAA concentration in mudstone appeared to be observed in the mudstone matrix adjacent to the subterranean Centrosaurus bone . When amino acids were above detection limit , mudstone was L-amino acid dominated , similar to the Late Cretaceous bone ( Table 1 ) . Topsoil , on the other hand , showed consistently moderate levels of racemization . These results show that topsoil contained a high amino acid concentration with relatively high rates of protein degradation , indicative of active biological accumulation and recycling , while mudstone contained low concentrations with very recent amino acids , indicative of low residence times of proteins within the mudstone . The fossil bones appeared to show instances of relatively greater accumulation of amino acids than the mudstone but with very recent amino acids , indicative of preferential localization of biologically active amino acids to the bone compared to the mudstone , but with less amino acid content than topsoil . Pleistocene-Holocene surface-eroded shark teeth had THAA compositional profiles that closely matched collagen ( Figure 6A , C , F ) and fairly high amino acid concentration with THAA concentrations between those of subterranean Centrosaurus bone and modern chicken bone ( Figure 7A ) . Pleistocene-Holocene surface-eroded shark teeth , unlike the Late Cretaceous bone and mudstone , had consistently high racemization ( Table 1 ) , even more so than the topsoil sample . Ethanol rinsing appeared to lower amino acid concentration in the shark teeth but did not strongly affect the THAA compositional profile ( Figures 6A , C , F and 7A ) . These results suggest that the Pleistocene-Holocene teeth contained detectable , ancient amino acids consistent with endogenous collagen . Total organic carbon ( TOC ) content was higher in the subterranean and surface-eroded Centrosaurus bone than the matrix , even the directly adjacent matrix , and was comparable to that found in the topsoil ( Table 2 ) . However , the organic carbon content in the Centrosaurus bones was significantly lower than the 82–71 ka Yarnton bovine bone sample known to contain well-preserved ( radiocarbon-dead ) collagen ( Cook et al . , 2012 ) . TOC in the Centrosaurus bone was not found to be radiocarbon dead , but did exhibit lower F14C values than both the mudstone and especially the topsoil . Assuming all endogenous bone C is radiocarbon ‘dead’ , based on these F14C values , a simple two-end-member mixing model would suggest that ~26% of the C in subterranean Centrosaurus bone originates in the adjacent mudstone matrix ( for formula , see Appendix 1 under the section entitled Carbon analysis ) . The fossil dinosaur bone therefore yielded a TOC content similar to relatively rich environmental carbon sources , such as topsoil , but not as high as more recent bone proteins . Although , some of the C in the fossil dinosaur bone is potentially ancient , there is still a sizable contribution of recent C from the immediate environment , consistent with the presence of a microbiome . DNA concentration was about 50 times higher in subterranean Centrosaurus bone than in adjacent mudstone matrix ( Table 3; Appendix 1—table 25 ) . PI staining for DNA on EDTA demineralized Centrosaurus bone revealed multi-cell aggregates forming organic vessel and fibrous conglomerate structures that fluoresce red ( Figure 8A–D ) . The DNA concentration in the bone indicates a cell concentration of ~4×108 cells/g ( calculation of cell abundance from DNA based on that of Magnabosco et al . , 2018; also see Appendix 1—table 26 ) . This is fairly similar to the observed THAA concentration indicating ~3×108 cells/g ( calculation of cell abundance from total amino acids based on that of Onstott et al . , 2014 and Lomstein et al . , 2012 ) , consistent with the idea that the amino acids within the bone are likely to be largely cellular ( i . e . lipid-bound within living organisms ) due to the discrepancy between DNA and amino acid stability over time . The DNA concentration in the adjacent mudstone matrix indicate a cell concentration of ~5×106 cells/g , but the observed THAA concentration is consistent with a cell concentration of ~2×109 cells/g . The greater amino acid abundance is a common feature of marine sediment and likely represents the amino acids of a microbial necromass ( e . g . Braun et al . , 2017 ) . The adjacent mudstone matrix contains amino acids that seem to largely represent dead prokaryote remains , unlike the amino acids in the dinosaur bone that seem to largely represent a more recent , likely living community in comparison ( i . e . the adjacent mudstone matrix has a greater amino acid concentration relative to the DNA concentration than does the dinosaur bone ) . These results suggest that the subterranean dinosaur bone was enriched in cell-bound DNA relative to the mudstone matrix . Furthermore , EDTA-extracted structures appeared to contain DNA from cells that aggregate within these structures , consistent with a modern biofilm; the DNA itself had possibly been exposed due to the EDTA treatment . The 16S rRNA gene amplicon sequencing revealed the predominance of Actinobacteria and Proteobacteria in subterranean Centrosaurus bone . Sequences affiliated with classes Nitriliruptoria and Deltaproteobacteria were more abundant relative to adjacent mudstone or even the surface scrapings from the bone itself ( Figure 9 ) . The majority of the sequences within Deltaproteobacteria were identified as belonging to the family Desulfurellaceae , which contains some sulfur-respiring species . However , the short reads prevented species level identification . In Centrosaurus bone , about 30% of sequences were phylogenetically close to the genus Euzebya , a deeply branching , aerobic , marine Actinobacterium ( Appendix 1—figure 58 ) . Furthermore , PCA of the species-level percentage data from these eight samples shows that the differences between the interior bone cores and the mudstone or bone surface scrapings is greater than the difference between the mudstone and bone surface scrapings ( Appendix 1—figure 60 ) . Likewise , one-way permutational multivariate analysis of variance ( PERMANOVA ) performed in PAST3 software of species-level sequence percentages of the two replicates of each of the four sample categories in Figure 9 yielded significant differences ( Euclidean similarity index; 9999 permutations; F = 53 . 16; p-value=0 . 0084 ) , with greater similarity between the mudstone and bone surface scrapings than between either of these and the interior bone core samples ( Appendix 1—table 27 ) . These results suggest that the subterranean dinosaur bone contained a different microbial community than the surrounding mudstone matrix with some species potentially impacting fossil bone taphonomy and chemical composition . Our initial sequence data , furthermore , suggests that some of these microbes might represent rare , poorly understood taxa .
Previous studies have often reported purported endogenous ‘soft tissues’ within fossil dinosaur bone ( Pawlicki et al . , 1966; Schweitzer et al . , 2005a; Schweitzer et al . , 2005b; Schweitzer et al . , 2007a; Schweitzer et al . , 2007b; Schweitzer et al . , 2008; Schweitzer et al . , 2009; Schweitzer et al . , 2013; Schweitzer et al . , 2014; Schweitzer et al . , 2016; Asara et al . , 2007; Organ et al . , 2008; Schweitzer , 2011; Bertazzo et al . , 2015; Cleland et al . , 2015; Schroeter et al . , 2017 ) . However , these studies often do not fully address fossil bones being open systems that are biologically active . This can be seen in field observations , in Dinosaur Provincial Park and elsewhere , where fossil bone is frequently colonized by lichen on the surface or overgrown and penetrated by plant roots in the subsurface . This forces researchers to consider that subsurface biota ( e . g . plant roots , fungi , animals , protists , and bacteria ) could contaminate bone . Given that fungi can produce collagen ( Celerin et al . , 1996 ) , the need to rule out exogenous sources of organics in fossil bone is made all the greater . Even deeply buried bone has the potential to be biologically active , given the high concentration of microorganisms in continental subsurface sedimentary rock ( Magnabosco et al . , 2018 ) . The analyses presented here are consistent with the idea that far from being biologically ‘dead’ , fossil bone supports a diverse , active , and specialized microbial community . Given this , it is necessary to rule out the hypothesis of subsurface contamination before concluding that fossils preserve geochemically unstable endogenous organics , like proteins . We detected no evidence of endogenous proteins in the bone studied here and were therefore unable to replicate claims of protein survival from deep time , such as the Mesozoic ( Pawlicki et al . , 1966; Schweitzer et al . , 2005a; Schweitzer et al . , 2007a; Schweitzer et al . , 2007b; Schweitzer et al . , 2008; Schweitzer et al . , 2009; Schweitzer et al . , 2013; Schweitzer et al . , 2014; Schweitzer et al . , 2016; Asara et al . , 2007; Organ et al . , 2008; Schweitzer , 2011; Bertazzo et al . , 2015; Cleland et al . , 2015; Schroeter et al . , 2017 ) . In contrast , recent Pleistocene-Holocene material often shows clear , and multiple lines of , evidence for endogenous , ancient collagen . These may be found even when the fossil ( dentine/enamel in this case ) is stained black , indicating taphonomic alteration , and the sample is found exhumed in a warm climate and not treated with aseptic techniques . Detection of specific organic signatures in fossils ( e . g . amide bands in FTIR or Raman spectroscopy ) requires corroborating evidence before claims of ancient proteins can be substantiated . In addition to reliable markers of general protein presence ( e . g . amide , succinimide , or piperazine pyrolysis products ) , evidence is required to identify the type of protein ( i . e . amino acid composition or sequence ) as well demonstrate its endogenous origin ( e . g . localization ) and age ( i . e . degree of degradation as revealed by amino acid racemization , post-translational modifications such as deamidation , or peptide length/degree of hydrolysis ) . Degradation of collagen polypeptides follows a pattern of gradual hydrolysis of amino acids at the terminal ends followed by catastrophic degradation and rapid hydrolysis due to rupture of the triple helix quaternary structure , making the resulting gelatinous fragments susceptible to rapid leaching from the bone or microbial degradation ( Collins et al . , 1995; Collins et al . , 2009; Dobberstein et al . , 2009 ) . It might therefore be suspected that if ancient collagen does indeed persist in a fossil bone , then such preservation would often provide clear , strong structural and chemical signatures like that in the Pleistocene-Holocene shark teeth . Recently it has been suggested that techniques that do not provide information on the precise sequence or post-translational modification of peptides , such as Py-GC-MS or HPLC amino acid analysis , are outdated for palaeoproteomic studies ( Cleland and Schroeter , 2018 ) . This might be the case when samples are very young and from cold environments , in which case , more precise mass spectrometric analyses such as liquid chromatography-tandem mass spectrometry might be employed early on in the course of research with elevated confidence that ancient proteins are capable of persisting in the sample . However , our results here suggest that techniques like Py-GC-MS or HPLC that give more general information on protein presence versus absence or general amino acid composition should be considered frontline approaches when dealing with samples of significant age and/or thermal maturity ( e . g . Demarchi et al . , 2016; Hendy et al . , 2018; Cappellini et al . , 2018 ) . Treating Mesozoic bone that has experienced diagenesis , low latitudes , and permineralisation identically to more recent , less altered bone is ill-advised , and any work on such samples should employ these fundamental methods before attempting to sequence peptides that might not be present , ancient , or endogenous . Fossil bone has fairly high concentrations of recent organics ( e . g . L-amino acids , DNA , and non-radiocarbon dead organic C ) , even when buried and often in comparison to the immediate environment . Fossil bone likely provides an ideal , nutrient-rich ( e . g . phosphate , iron ) open system microbial habitat inside vascular canals capable of moisture retention . The absence of evidence for endogenous proteins and the presence of a diverse , microbial community urge caution regarding claims of dinosaur bone ‘soft tissues’ . Microbes can colonize bones while buried , likely traveling via groundwater . Therefore , it is unsurprising that the prevalence of these ‘soft tissues’ is not correlated with overburden depth above the fossil or cortical versus cancellous bone tissue ( Ullmann et al . , 2019 ) . Rather , minimum distance from the surface is probably of importance and microbes likely readily colonize a variety of bone tissue types since both presumably behave as open systems . Our results support the hypothesis that at least some ‘soft tissue’ structures derived from demineralised fossil bones represent biofilms . We suggest that unless in an inaccessible form ( e . g . kerogen , depending on microbial metabolic ability ) or matrix ( e . g . well-cemented concretion ) , endogenous dinosaur organics that survive prior taphonomic processes ( e . g . diagenesis ) may be subject to subsequent microbial metabolic recycling . The study of fossil organics must consider potential microbial presence throughout a specimen’s taphonomic history , from early to late . Microbial communities interact with fossils immediately following death and after burial , but prior to diagenesis . Microbes are known to utilize bone and tooth proteins ( Child et al . , 1993 ) and fossil evidence of early fungal colonization has even been detected ( Owocki et al . , 2016 ) . More recent microbial colonization of fossil bone will occur as it nears the surface during uplift and erosion in the late stages of the taphonomic process . Furthermore , given that microbes can inhabit the crust kilometres below the surface ( Magnabosco et al . , 2018 ) , it might be predicted that bone remains a biologically active habitat even when buried hundreds of meters deep for millions of years . The extensive potential for microbial contamination and metabolic consumption makes verifying claims of Mesozoic bone protein extremely challenging . | The chances of establishing a real-world Jurassic Park are slim . During the fossilization process , biological tissues degrade over millions of years , with some types of molecules breaking down faster than others . However , traces of biological material have been found inside some fossils . While some researchers believe these could be the remains of ancient proteins , blood vessels , and cells , traditionally thought to be among the least stable components of bone , others think that they have more recent sources . One hypothesis is that they are in fact biofilms formed by bacteria . To investigate the source of the biological material in fossil bone , Saitta et al . performed a range of analyses on the fossilized bones of a horned dinosaur called Centrosaurus . The bones were carefully excavated in a manner to reduce contamination , and the sediment the bones had been embedded in was also tested for comparison . Saitta et al . found no evidence of ancient dinosaur proteins . However , the fossils contained more organic carbon , DNA , and certain amino acids than the sediment surrounding them . Most of these appeared to have a very recent source . Sequencing the genetic material revealed that the fossil had become a habitat for an unusual community of microbes that is not found in the surrounding sediment or above ground . These buried microbes may have evolved unique ways to thrive inside fossils . Future work could investigate how these unusual organisms live and whether the communities vary in different parts of the world . | [
"Abstract",
"Introduction",
"Materials",
"and",
"methods",
"Results",
"Discussion"
] | [
"biochemistry",
"and",
"chemical",
"biology",
"microbiology",
"and",
"infectious",
"disease"
] | 2019 | Cretaceous dinosaur bone contains recent organic material and provides an environment conducive to microbial communities |
Generative models , such as predictive coding , posit that perception results from a combination of sensory input and prior prediction , each weighted by its precision ( inverse variance ) , with incongruence between these termed prediction error ( deviation from prediction ) or surprise ( negative log probability of the sensory input ) . However , direct evidence for such a system , and the physiological basis of its computations , is lacking . Using an auditory stimulus whose pitch value changed according to specific rules , we controlled and separated the three key computational variables underlying perception , and discovered , using direct recordings from human auditory cortex , that surprise due to prediction violations is encoded by local field potential oscillations in the gamma band ( >30 Hz ) , changes to predictions in the beta band ( 12-30 Hz ) , and that the precision of predictions appears to quantitatively relate to alpha band oscillations ( 8-12 Hz ) . These results confirm oscillatory codes for critical aspects of generative models of perception .
It has long been apparent that brain responses do not simply represent input from sensory organs , but that they are modulated by context and expectation , giving rise to phenomena such as priming , mismatch negativity and repetition suppression . These can be explained if perceptual systems are based on internal generative models of the environment which are constantly updated based on experience . Predictive coding ( Rao and Ballard , 1999 ) is a popular account of perception , in which internal representations generate predictions about upcoming sensory input , characterised by their mean and precision ( inverse variance ) ( Friston , 2005; Friston and Kiebel , 2009 ) . Sensory information is processed hierarchically , with backward connections conveying predictions , and forward connections conveying violations of these predictions , namely prediction errors . Qualitatively , prediction errors are the mismatch between the prediction and incoming sensory information , but the term is often used without a quantitative definition . One quantitative formulation of prediction error is surprise ( Friston and Kiebel , 2009 ) , which is the negative log probability of a sensory event , given the prior prediction . This definition takes into account the precision of predictions , such that the same prediction violation causes greater surprise where predictions are more precise . Prediction errors act to produce changes in predictions , thereby updating and refining internal models of the environment , and reducing subsequent prediction errors . These variables are illustrated in Figure 1 . There is substantial overlap between predictive coding and other accounts of perception based on internal generative models ( Friston , 2008 ) . The crucial common feature of any generative model of perception is the brain’s use of hidden states to predict observed sensory inputs , thus the methods and findings of this study are applicable to all generative perceptual models . 10 . 7554/eLife . 11476 . 003Figure 1 . Computational variables involved in perceptual inference . The graph displays a schematic probability distribution ( solid curve ) representing the prior prediction about the fundamental frequency ( f ) of an upcoming auditory stimulus ( ft ) , where t simply refers to the number or position of the stimulus within a sequence . This prediction is characterised by its mean ( μt ) and precision ( Πt ) , which is the inverse of its variance ( σ2 ) . The incongruence between the actual ft and the prediction can be expressed either as a ( non-precision-weighted ) prediction error ( ξt ) , that is , the absolute difference from the prediction mean , or as surprise ( St ) , that is , the negative log probability of the actual ft value according to the prediction distribution . As a result of a mismatch with bottom up sensory information , the prediction changes ( dashed line ) . The change to the prediction ( Δμt ) is calculated simply as the absolute difference between the old ( μt ) and new ( μt+1 ) prediction means . Note that the curves on the graph display changing predictions on account of a stimulus ( i . e . Bayesian belief updating ) as opposed to the more commonly encountered graph in this field of research where the curves indicate the prior prediction , the sensory information and the posterior inference about the individual stimulus ( i . e . Bayesian inference ) . DOI: http://dx . doi . org/10 . 7554/eLife . 11476 . 003 The functional unit of neocortex , the canonical microcircuit ( Haeusler and Maass , 2007 ) , has recently been interpreted in light of predictive coding models ( Bastos et al . , 2012 ) , revealing appropriate neuronal properties and internal/external connectivity to carry out the necessary neuronal computations . It is thus hypothesised that superficial cell populations calculate prediction errors , manifest as gamma-band oscillations ( >30 Hz ) , and pass these to higher brain areas , while deep cell populations encode predictions , which manifest as beta band oscillations ( 12–30 Hz ) and pass these to lower brain areas ( Bastos et al . , 2012 ) . The layer-specific separation of higher and lower frequency oscillations ( Spaak et al . , 2012 ) , and the forward/backward asymmetry of high/low frequency oscillations ( Buschman and Miller , 2007; Fontolan et al . , 2014; van Kerkoerle et al . , 2014 , Bastos et al . , 2015 ) , are supported by direct evidence . A number of studies have found oscillatory gamma magnitude to correlate with the unexpectedness of incongruence of stimuli ( Arnal et al . , 2011; Brodski et al . , 2015; Todorovic et al . , 2011 ) , but it remains unclear exactly what computational variable they represent . While there is a strong case that beta oscillations are involved in top-down neural communication , evidence specifically linking beta oscillations to predictions is presently limited and indirect ( Arnal and Giraud , 2012 ) , but includes observations that there is interdependence of gamma and subsequent beta activity in both in vivo ( Haenschel et al . , 2000 ) and in silico ( Kopell et al . , 2011 ) studies and that omissions of expected stimuli induce a beta rebound response ( Fujioka et al . , 2009 ) . An oscillatory correlate of precision , to our knowledge , has not been proposed , though precision might affect the magnitude of gamma responses to prediction violations ( Feldman and Friston , 2010 ) . While an oscillatory correlate is possible , a case has been made that neuromodulatory connections alone , for instance from the basal forebrain cholinergic system , may be sufficient to dynamically mediate precision in sensory hierarchies ( Feldman and Friston , 2010; Kanai et al . , 2015 ) . Direct evidence for correlates of processes inherent in perceptual inference requires being able to quantitatively manipulate predictions during an experiment , which has not so far been achieved . In the present study , we sought to dissociate and expose the neural signatures of four key variables in predictive coding and other generative accounts of perception , namely surprise , prediction error , prediction change and prediction precision . Here , prediction error refers to absolute deviation of a sensory event from the mean of the prior prediction ( which does not take into account the precision of the prediction ) . We hypothesised that surprise ( over and above prediction error ) would correlate with gamma oscillations , and prediction change with beta oscillations . The possibility of an oscillatory code for precision was also explored .
Direct cortical recordings were made from the auditory cortices of three awake humans undergoing invasive monitoring for epilepsy localization , while they listened to a pitch stimulus with a fundamental frequency ( usually referred to as ‘f0’; hereafter just ‘f’ for clarity ) that varied according to simple rules ( Figure 2 ) . Local field potential ( LFP ) data were decomposed using Morlet wavelets , separated into evoked and induced components , and regressed against the four perceptual inference variables of interest which were calculated by Bayes-optimal inversion of the sequence of f values assuming full knowledge of the rules by which they were generated ( Figure 2—figure supplements 1 and 2 ) . 10 . 7554/eLife . 11476 . 004Figure 2 . Algorithm and example stimulus . ( A ) The stimulus is composed of a series of concatenated segments , differing only in fundamental frequency ( f ) . At any time , a given f population is in effect , characterised by its mean ( μ ) and standard deviation ( σ ) . For each successive segment , there is a 7/8 chance that that segment’s f value will be randomly drawn from the present population , and a 1/8 chance that the present population will be replaced , with new μ and σ values drawn from uniform distributions . ( B ) Example section of stimulus . ( Bi ) Dots indicate the f values of individual stimulus segments , of 300 ms duration each . Four population changes are apparent . ( Bii ) Spectrogram of the corresponding stimulus , up to 5 kHz , on a colour scale of -60 to 0 dB relative to the maximum power value . The stimulus power spectrum does not change between segments , and the only difference is the spacing of the harmonics . DOI: http://dx . doi . org/10 . 7554/eLife . 11476 . 00410 . 7554/eLife . 11476 . 005Figure 2—figure supplement 1 . Generative model and inversion scheme . Schematic of the full three-dimensional model space , with dimensions indicating population mean ( σ ) , population standard deviation ( σ ) and stimulus fundamental frequency ( f ) . To ease computational demands , the model space was discretised . Each point in model space corresponds to the probability of a particular f , given a specific μ and σ , that is , P ( f|μ , σ ) . Therefore each column ( along the f dimension ) gives the probability distribution P ( f|μ , σ ) , and corresponds to the forward model . The planes of the model space , conversely , indicate the probability of each given combination of μ and σ , given a particular f value , that is , P ( μ , σ|f ) , or in other words the inverse model . To generate priors based on a series of observed f values , the scalar product of the planes for each of these f values is taken , and the resulting plane scaled to a sum of 1 . This plane represents the estimates of the hidden states μ and σ and is then used to weight the columns of the model space . The weighted model space is averaged into a single column ( forward model ) , and scaled to a sum of 1 , thus providing optimal priors on the assumption that the f population does not change . The priors assuming a population change are derived from the same procedure , but with uniform weighting across the model space . The change and no change priors are then weighted according to the probability of a population change , and then summed . The only part of the process not illustrated here is the inference of population changes , determining how many preceding f values form part of the model inversion , which is explained in Equation 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 11476 . 00510 . 7554/eLife . 11476 . 006Figure 2—figure supplement 2 . Example Bayes-optimal prior predictions generated by model inversion . Observed f values from a section of the stimulus ( white dots ) , overlaid on prior predictions ( colour scale ) based on previous observations of f , using the model inversion scheme . DOI: http://dx . doi . org/10 . 7554/eLife . 11476 . 00610 . 7554/eLife . 11476 . 007Figure 2—figure supplement 3 . Regressor correlations . Correlation matrices between instantaneous and time-lagged values of the main regressors . Note the strong instantaneous mutual positive correlations between Δf , S and Δμ , and the negative correlations between Π and preceding values of Δf , S and Δμ . DOI: http://dx . doi . org/10 . 7554/eLife . 11476 . 007 In keeping with prior hypotheses , both surprise and prediction error ( the latter not taking into account the precision of predictions ) were associated with significant gamma band responses in the LFP . We first established which of these variables explained the LFP data better . Figure 3 shows the strong correlation between these variables ( A ) , the explanatory power of each with respect to the LFP data ( B ) , and the unique explanatory power of each after partialling out the other variable ( C ) . The extremely strong correlation between surprise and prediction error ( r = 0 . 92 over 8000 samples ) necessitated this partial analysis ( C ) in order to examine the independent contribution of each variable to the observed LFP data . Both variables correlated positively with gamma magnitude , but surprise showed a stronger correlation in all three subjects . In the partial analysis ( C ) , residual surprise ( after partialling out prediction error ) correlated positively with gamma magnitude , whereas residual prediction error ( after partialling out surprise ) showed only a weak negative correlation in two subjects , and no correlation in one subject . At group level , these correlations were significantly different to each other at p<0 . 01 corrected , thus we concluded that surprise is the better correlate of gamma magnitude , and used this measure for further analysis . 10 . 7554/eLife . 11476 . 008Figure 3 . Comparison between surprise and prediction error . ( A ) Correlation between surprise ( S ) and non-precision-weighted prediction error ( ξ ) , with each dot indicating an individual stimulus segment and the line indicating a linear regression fit . ( B/C ) Mean Pearson product moment correlation coefficients ( r ) between St or ξt , and gamma oscillation magnitude ( 30–100 Hz ) in the 90–500 ms period following the onset of stimulus segment t . Regression coefficients were calculated for each time-frequency point , after partialling out the influences of current and preceding/subsequent values of all other regressors , and then averaged across time and frequency; these processing steps diminished the absolute size of the correlation values . In C , the influence of S on ξ , and ξ on S , was also partialled out , thus exposing the unique contribution of each variable to explaining the observed neural response . Partial S showed a higher mean correlation , across subjects , with gamma magnitude than partial ξ ( p<0 . 01 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 11476 . 008 Figure 4 shows , at group level , the spectrotemporal pattern of induced and evoked oscillations uniquely attributable to each of the three perceptual variables of interest: surprise ( S ) , change in prediction mean ( Δμ ) , precision of predictions ( Π ) , as well as the change in f value from one stimulus to the next ( Δf ) . The latter measure was not a perceptual variable of interest but was included for comparative purposes as it approximately represents the ‘pitch onset response’ which is a robust and familiar response in auditory neurophysiology ( Griffiths et al . , 2010 ) . Data significant at p<0 . 05 corrected ( based on a non-parametric permutation approach [Maris and Oostenveld , 2007] ) are shown in the left column of each group , and all data in the right column . Individual subject data are shown in Figure 4—figure supplement 1 . As these perceptual variables were highly correlated ( Figure 2—figure supplement 3 ) , instantaneously and over time , regression for the main analysis was based on the residuals after partialling out these correlated influences , such that only the unique explanatory contribution of each variable , with respect to the LFP data , was measured . While the correlation values observed were small in absolute terms ( Pearson’s r < 0 . 1 ) , reassuringly these r values were of the same scale of magnitude as those for change in frequency ( Δf ) , which represents a robust auditory response . Furthermore , the LFP data variance explained by the entire model ( Figure 4—figure supplement 1 ) was around 1% . In accordance with our hypotheses , surprise ( S ) correlated positively , across subjects , with gamma oscillations , beginning at around 100 ms from segment onset , and this was significant by 200 ms . Also in accordance with our hypothesis , changes to predictions ( Δμ ) correlated positively with beta oscillations coinciding with the onset of the subsequent stimulus segment ( about 100 ms after ) , which again was significant . Prediction precision ( Π ) correlated positively with delta-alpha ( 2–12 Hz ) frequency oscillations ( for the whole 0–300 ms period from segment onset ) , although this was only significant in the alpha frequency range , and fell slightly below significance in the delta-theta range . Given the strong negative correlation between Π and the preceding values of S and Δf , it seemed likely that the low-frequency correlates of these were being mutually attenuated by the partialisation process . For this reason , and to search for correlates of the commonalities between key variables , we repeated the analyses with only the contemporaneous value of Δf being partialled out ( Figure 4—figure supplement 3 ) . This analysis found highly significant correlates of precision ( Π ) in the full delta-alpha range , spanning the previous , current and subsequent segments , but we cannot attribute the delta-theta component to Π with absolute confidence . To respect the exploratory nature of our search for oscillatory correlates of Π , a further variation of the analysis ( Figure 4—figure supplement 4 ) omitted Π altogether ( including the partialisation of other variables with respect to it ) . Results were quantitatively stronger ( due to reduced partialisation ) , but qualitatively similar except that S contained a strong negative delta-alpha band correlation coincident with the subsequent stimulus segment . As S for one segment is negatively correlated with Π of the subsequent segment , it is thus not presently clear how much of this low-frequency correlation is with S ( negatively ) , and how much with Π ( positively ) . However , given its sole significant association in the main analysis ( Figure 4 ) is with Π , and that a low-frequency correlate of S is not expected based on prior literature , we favour the interpretation that low-frequency oscillations are a correlate of the precision of prior predictions . Evoked results are shown in the right hand sections of Figure 4 plus its figure supplements 3 and 4 . Unlike the induced results , there was no qualitative distinction between the timing or frequency profiles of the different variables ( this is particularly evident in Figure 4—figure supplement 4 ) . The only significant evoked responses in the main analysis ( Figure 4 ) were to Δf and Δμ , while S only showed significant evoked correlates when Π was omitted from the analysis ( Figure 4—figure supplement 4 ) . 10 . 7554/eLife . 11476 . 009Figure 4 . Spectrotemporal profiles associated with key perceptual inference variables . Each plot illustrates the mean Pearson product moment correlation coefficient ( r ) , across stimulus-responsive electrodes and across subjects , between induced oscillatory amplitude , at each time-frequency point , and the regressor of interest , that is , a time-frequency ‘image’ of the oscillatory correlates of a particular perceptual variable . Time is represented on horizontal axes , and frequency on vertical . Dashed lines indicate the division between the previous ( t-1 ) , current ( t ) and subsequent ( t+1 ) stimulus segment ( vertical lines ) , and between frequency bands ( horizontal lines ) . Each row of plots represents one regressor . The left-hand group contains induced correlates , and the right-hand group evoked , with the left-hand column in each group showing data points significant at p<0 . 05 corrected . Regressors are partialised with respect to each other , such that only the unique contribution of each to explaining the overall oscillatory data is displayed . ( A ) Spectrotemporal correlates of the three fundamental variables for perceptual inference . Note that the grey area in the upper left plot reflects the spectrotemporal region of interest ( ROI ) analysis used for correlates of surprise ( i . e . >30 Hz ) . Outside of the ROI analysis , no significant correlates were observed below 30 Hz . ( B ) The variable ‘pitch change’ indicates the overall response to a changing stimulus and is included to illustrate the magnitude and time-frequency distribution of a typical and robust auditory response . δ = delta ( 0–4 Hz ) , θ = theta ( 4–8 Hz ) , α = alpha ( 8–12 Hz ) , β = beta ( 12–30 Hz ) , γ = gamma ( 30–100 Hz ) . DOI: http://dx . doi . org/10 . 7554/eLife . 11476 . 00910 . 7554/eLife . 11476 . 010Figure 4—figure supplement 1 . Individual subject induced correlates . Notation and conventions are as for Figure 4 , except that each column denotes a single subject . All data are displayed , as significance testing was not performed at the individual subject level . DOI: http://dx . doi . org/10 . 7554/eLife . 11476 . 01010 . 7554/eLife . 11476 . 011Figure 4—figure supplement 2 . Total variance explained by the model . This figure shows the total variance explained by the four regressor of interest ( applicable to the stimulus segment 0-300 ms ) , including both their overlaps and their unique explanatory contributions . The plot shows the mean across electrodes and subjects . DOI: http://dx . doi . org/10 . 7554/eLife . 11476 . 01110 . 7554/eLife . 11476 . 012Figure 4—figure supplement 3 . Results without mutual partialisation . This figure is equivalent in all respects to Figure 4 , except that the four regressors have not been partialised with respect to each other , hence results are not only limited to the unique explanatory power of each , but also include all overlaps between regressors . ( A ) The three perceptual inference regressors have been partialised only with respect to the contemporaneous pitch change value . ( B ) The pitch change regressor has been subjected to no partialisation at all . DOI: http://dx . doi . org/10 . 7554/eLife . 11476 . 01210 . 7554/eLife . 11476 . 013Figure 4—figure supplement 4 . Results with prediction precision omitted . This figure is equivalent in all respects to Figure 4 , except that the analysis it displays does not include prediction precision . Hence , this quantity has not been partialised out of the other regressors . Also , a region-of-interest analysis is no longer applied to correlates of surprise . Note stronger but qualitatively similar gamma correlates of surprise , and beta correlates of precision , but the major difference that surprise is now correlated negatively with low-frequency induced oscillations coincident with the subsequent stimulus segment . DOI: http://dx . doi . org/10 . 7554/eLife . 11476 . 01310 . 7554/eLife . 11476 . 014Figure 4—figure supplement 5 . Electrode positions . Electrodes used for analysis . White-filled circles each indicate one electrode . All electrodes shown were included in the presented analyses and were selected based on showing significant responses to the experimental stimulus , as described in the Methods section . Top row: lateral surface of the cerebral hemisphere recorded from . Solid black circles mark the insertion point for the Heschl’s gyrus depth electrode . Bottom row: depth electrode contacts used for analysis ( white-filled circles ) , shown in the context of the surface of the superior temporal plane . Depth electrode contacts were positioned along the axis of Heschl’s gyrus . S = superior , I = inferior , A = anterior , P = posterior , M = medial , L = lateral . DOI: http://dx . doi . org/10 . 7554/eLife . 11476 . 01410 . 7554/eLife . 11476 . 015Figure 4—figure supplement 6 . Distribution of correlations across electrodes . Distribution of regressor correlations . Each coloured plot represents one subject . Within each plot , the columns represent individual electrodes ( positions displayed in A ) , with the vertical line separating Heschl’s gyrus ( HG ) from superior temporal gyrus ( STG ) electrodes . HG electrodes are arranged medial ( left ) to lateral ( right ) . Rows indicate the four main partial regressors . Colour values indicate the relative similarity between the correlation pattern for that subject/electrode combination and the mean correlation pattern across electrodes and subjects ( range -1 to 1 ) . Δf = absolute change in f value ( octaves ) compared to previous , S = surprise , Δμ = absolute change in prediction mean ( octaves ) , Π = precision of prediction . DOI: http://dx . doi . org/10 . 7554/eLife . 11476 . 015 In light of the above results , this study provides the first direct demonstration that beta oscillations are involved in updating the content of sensory predictions . We have also found that the precision of predictions is correlated to the magnitude of alpha oscillations , and possibly delta and/or theta also , thereby raising the possibility of an oscillatory mechanism for the control of precision . Interestingly , this correlate of precision was not time-locked to the stimuli ( evoked ) , despite delta-theta oscillations showing strong phase entrainment by stimuli when predicting stimulus timing ( Arnal and Giraud , 2012 ) , and the period of the stimulus segments ( 300 ms ) falling within the delta-theta range . Possibilities include that there is an evoked correlate closely shared by both surprise and precision which the present methods are unable to disambiguate , or that this time-locking of low-frequency oscillations is entrained preferentially by low-level stimulus features ( such as changes in stimulus power spectrum ) as opposed to the higher level feature of temporal pitch employed in the present study . Existing generative accounts of perception have not proposed a specific oscillatory correlate for the precision of predictions in predicting what a sensory stimulus will be . However , the present findings are not without precedent , as theta ( 4–8 Hz ) and alpha ( 8–12 Hz ) oscillations are implicated in mechanisms to predict when a stimulus will occur , with theta phase aligning to the expected stimulus onset ( Arnal and Giraud , 2012 ) , and alpha magnitude has been found to correlate with the probability of a stimulus change occurring ( Bauer et al . , 2014 ) . Both theta ( Canolty et al . , 2006 ) and alpha ( Jensen and Mazaheri , 2010 ) oscillations modulate higher frequency oscillations , through phase-amplitude coupling , and thereby segregating sensory responses into specific temporal windows . There is also antagonism between theta/alpha and beta/gamma oscillation magnitudes ( Spaak et al . , 2012 ) which , in the context of a theta/alpha code for precision , might indicate that over coarse time scales neuronal populations alternate between states of precise predictions ( with theta/alpha predominating ) and states of prediction violation ( with beta/gamma predominating ) . However , the present results would suggest more than simple reciprocal antagonism and reveal the specific computational role of each oscillation type . While an oscillatory code for precision could have far-reaching implications , we must respect the fact that this is a novel finding and thus requires corroboration from additional studies with alternative methodology . Perhaps most strikingly , we have shown that the key variables theoretically necessary for sensory inference have distinct oscillatory profiles , with little to no overlap between these , which show remarkable consistency across subjects . Furthermore , each oscillatory frequency band correlates with a distinct computational variable . Thus , the present findings may be able to retrospectively aid in the interpretation of a large number of studies examining induced oscillations . As generative accounts of perception are generic across stimulus dimensions and sensory modalities , and perhaps even all of brain function ( for instance , if action is understood as a method of resolving prediction errors [Friston et al . , 2006] ) , the applicability of the results may be very broad indeed . The present paradigm is instantly portable to any sensory modality and , given the absence of any training or task requirement , to any species .
The basis of the experiment was an algorithm ( Figure 2A ) in which stimulus segments varied across only one perceptual dimension , and values were drawn randomly from populations , that is , Gaussian distributions , each characterised by its mean ( μ ) and standard deviation ( σ ) . These populations constituted hidden states that were not directly observable , but whose parameters ( i . e . μ and σ ) could be inferred . The populations were randomly changed according to simple rules , such that subjects could be expected to unconsciously learn these rules in order to minimise uncertainty about upcoming stimuli . The rules were that for each stimulus segment , there was a 7/8 chance that its value would be drawn from the existing population , and a 1/8 chance that a new population would come into effect . Once a new population came into effect , it became the ‘existing’ population . Each population had its μ and σ drawn randomly from uniform distributions . The 1/8 transition probability , and the other parameters described below , were chosen in order to maximise the dissociation between the perceptual inference variables under study . The algorithm was implemented in the auditory domain , with stimuli taking the form of harmonic complexes , containing only unresolved harmonics ( by high-pass filtering from 1 . 8 kHz ) . Each harmonic had a random phase offset , which was preserved across all segments , stimuli and subjects . The variable dimension was fundamental frequency ( f0; hereafter just ‘f’ for simplicity ) , which is the major determinant of perceived pitch . Population μ was limited to the range 120–140 Hz , and σ to the range 1/128–1/16 octaves . Stimulus segments were 300 ms in duration and were smoothly concatenated to avoid any transients at the transitions between segments . This was achieved by defining instantaneous frequency at every point in the stimulus , by calculating the cumulative sum of this , and then by creating harmonics individually in the time domain as follows in Equation 1: ( 1 ) aT=sin ( 2πr+2πh1s∑t=1Tft ) where a is the amplitude of the waveform , T is the current time point ( measured in samples ) , t is all previous time points , r is the random phase offset for the harmonic , h is the number of the harmonic , s is the sampling rate and f is the instantaneous frequency . This procedure was repeated for every harmonic , from below the high-pass to above the Nyquist frequency . To prevent aliasing , the stimulus was generated at 88 . 2 kHz sampling rate , then downsampled to 44 . 1 kHz . The segment duration of 300 ms was chosen as the minimum duration that would capture most of the transient response to the onset of pitch within a stimulus , based on previous work ( Griffiths et al . , 2010 ) . 8000 stimulus segments were presented to each subject , consisting of four blocks of 2000 segments . Blocks were generated by the same rules but were each independently randomly generated . See Figure 2B for an example section of the stimuli . Subjects were three patients undergoing invasive electrode monitoring for localisation of medically refractory epilepsy prior to resective surgery . Subjects were not known to have any major cognitive deficits or clinically significant hearing impairment , and none had lesions in the region of auditory cortex . Informed consent for experimentation was obtained from all subjects , and research procedures were approved by the University of Iowa Institutional Review Board . Stimuli were presented diotically , via insert earphones ( ER4B; Etymotic Research , Elk Grove Village , IL ) through molds fitted to the subject’s ear , at the loudest comfortable volume . During the experiments , subjects engaged in an irrelevant auditory task to maintain attention , but a specific performance on this task was not required . This task involved detecting a change to the timbre of individual stimulus segments ( 64 targets over 8000 segments ) , which was unrelated to their frequencies or underlying population parameters . Subject 1 performed well on the task , and subjects 2 and 3 performed poorly , with high false alarm rates . The first 100 stimulus segments , and 10 segments following each target and false alarm , were removed from analysis . Recordings were made from one hemisphere in each subject ( Subjects 1 and 3: right , Subject 2: left ) . All subjects had an 8-contact depth electrode placed along the axis of Heschl’s gyrus , including anatomically and physiologically defined primary auditory cortex , and a subdural grid overlying superior temporal gyrus . Local field potential data were downsampled to 1 kHz , and electrical noise was filtered out . Time-frequency decomposition was performed with Morlet wavelet convolution , oversampled at 2 Hz frequency resolution and 10 ms time resolution , in the time range -300 to 600 ms from segment onset ( i . e . spanning previous , current and subsequent segments ) and frequency range 2–100 Hz . The upper frequency bound was limited to 100 Hz in light of previous observations of a lack of qualitative response difference to pitch stimuli between the 80–100 Hz range and higher frequencies ( Griffiths et al . , 2010 ) . The number of cycles per wavelet increased linearly from 1 cycle at 2 Hz to 10 cycles at 100 Hz . The absolute value ( i . e . amplitude ) of the wavelet coefficients was calculated for artefact rejection purposes , and these were normalised for each frequency ( i . e . shifted/scaled to mean 0 and standard deviation 1 ) . For each trial , both the mean ( across time and frequency ) and maximum normalised amplitude value was recorded , and the frequency histograms of these were plotted . Thresholds for trial rejection were set , by visual inspection , at the upper limit of the normal distribution of responses , beyond which trials were assumed to contain artefacts . After removal of segments at the start of the experiment , following target segments , and with outlying amplitude values , 89 , 86 and 87% of segments remained , for the three subjects , respectively . Data were processed for all electrodes either in Heschl’s gyrus or over the superior temporal gyrus . The procedure for selecting electrodes for the final analysis is described later . Ideal prior predictions , for each stimulus segment were calculated by inverting the algorithm used to generate the frequency values for the stimulus . Human subjects tend to implicitly make near-optimal inferences based on available sensory evidence and past experience ( Ernst and Banks , 2002; Körding and Wolpert , 2004 ) , but in the present study , all that was assumed was an approximate concordance between the subjects’ actual inferences and the optimal inferences that we modelled . This seems highly probable given the simplicity of the algorithm and its conformation to Gaussian statistics . Prediction calculation was achieved as follows , using a model inversion scheme illustrated in Figure 2—figure supplement 1: An example section of the stimulus , along with Bayes optimal prior predictions generated by the above algorithm , are shown in Figure 2—figure supplement 2 . As the regressors ( the variables calculated above ) were highly correlated with each other , both instantaneously and over neighbouring segments ( see Figure 2—figure supplement 3 ) , these were partialised with respect to each of the other regressors , and the preceding and subsequent two values of both the other regressors and themselves . This conservative approach removed a lot of explanatory power from these regressors , but was necessary to be able to uniquely attribute observed neural correlates to a specific process . Importantly , the partialisation did not qualitatively alter the results , except in obscuring the distinction between correlates of S and Δµ . Non-partialised results are shown in Figure 4—figure supplement 3 . Electrodes were included for further analysis ( i . e . included in the averaging process ) if they showed a significant response to the stimulus as a whole , based on the single largest response value in time-frequency space . Analysis was based on a 300 ms time window , which was randomly displaced by up to +/- 300 ms for each segment in each permutation ( and undisplaced for the actual data ) before averaging across segments . The electrodes selected using this procedure are displayed in Figure 4—figure supplement 5 . In all subjects , electrodes were included from both primary and non-primary auditory cortex . For each partialised regressor , complex time-frequency data were subject to a two-stage regression approach . First , the complex ( i . e . with phase data retained ) for every electrode-time-frequency point were regressed against it to yield a pair ( real and imaginary ) of Pearson product moment correlation coefficients ( r ) . The modulus of these constituted the evoked ( time-locked ) response . To calculate the induced response , the residuals from this regression ( i . e . discarding the evoked component ) were converted to amplitude ( by taking their modulus ) , and the regression was performed again . All r values were subjected to a Fisher Z transformation prior to further analysis . Inspection of these responses found no qualitative differences between the responses observed in different divisions of auditory cortex , hence the correlations were averaged over electrodes for further analysis . To quantify the distribution of correlation strengths , the pattern , across time and frequency , of induced correlation coefficients for each regressor was averaged across electrodes and subjects . This pattern was used as a filter , in that it was scalar multiplied with the correlation coefficient pattern for each electrode for each subject , then averaged across time and frequency to yield a single correlation value for that electrode/subject combination . These values , for each regressor , for each subject , were divided by the largest absolute correlation value for that regressor , in order to represent relative correlation strengths on a scale of -1 to 1 . These correlation values are displayed in Figure 4—figure supplement 6 , and show no systematic dissociation between the anatomical distributions of the correlations for the different regressors . Statistical analyses were all performed using a permutation approach ( Maris and Oostenveld , 2007 ) , using 100 permutations ( in each of which the experimental parameter of interest was removed through randomization , e . g . by shuffling regressor values across stimulus segments ) and a significance threshold of p<0 . 05 corrected . In each permutation , the statistical measure of interest is calculated , and the largest value within each permutation is added to a null distribution , from which a statistical threshold for significance is set . The measure of interest in this approach varied according to the analysis being performed ( see below ) , and included the values at individual time-frequency points averaged across electrodes , mean values across time and frequency at individual electrodes , and mean values across both time-frequency points and electrodes . To compare surprise and prediction error , for each of these regressors , the mean of the correlation coefficients ( across time , frequency , electrode and subject ) was calculated within the time window 90–500 ms from segment onset and the frequency window 30- 100 Hz . This was performed once with the regressors partialised as previously described ( i . e . for current and adjacent values of Δf , Π and Δµ ) , and again with additional partialisation of each regressor with respect to the other . This latter analysis measures the unique contribution of each regressor in explaining the observed data over and above the other , and was the analysis subjected to statistical analysis . For the main correlation analysis , significance testing was performed on average correlation values across the three subjects . Points in time-frequency space exceeding the permutation-derived threshold were considered significant . Due to the strong prior hypothesis about gamma oscillations correlating with surprise or prediction error , the statistical analysis was repeated for these regressors but with only frequencies in the gamma range ( 30–100 Hz ) being included in the analysis . | Our perception of the world is not only based on input from our senses . Instead , what we perceive is also heavily altered by the context of what is being sensed and our expectations about it . Some researchers have suggested that perception results from combining information from our senses and our predictions . This school of thought , referred to as “predictive coding” , essentially proposed that the brain stores a model of the world and weighs it up against information from our senses in order to determine what we perceive . Nevertheless , direct evidence for the brain working in this way was still missing . While neuroscientists had seen the brain respond when there was a mismatch between an expectation and incoming sensory information , no one has observed the predictions themselves within the brain . Sedley et al . now provide such direct evidence for predictions about upcoming sensory information , by directly recording the electrical activity in the brains of human volunteers who were undergoing surgery for epilepsy . The experiment made use of a new method in which the volunteers listened to a sequence of sounds that was semi-predictable . That is to say that , at first , the volunteers heard a selection of similarly pitched sounds . After random intervals , the average pitch of these sounds changed and they became more or less variable for a while before randomly changing again . This approach meant that the volunteers had to continually update their predictions throughout the experiment In keeping with previous studies , the unexpected sounds , which caused a mismatch between the sensory information and the brain’s prediction , were linked to high-frequency brainwaves . However , Sedley et al . discovered that updating the predictions themselves was linked to middle-frequency brainwaves; this confirms what the predictive coding model had suggested . Finally , this approach also unexpectedly revealed that how confident the volunteer was about the prediction was linked to low-frequency brainwaves . In the future , this new method will provide an easy way of directly studying elements of perception in humans and , since the experiments do not require complex learning , in other animals too . | [
"Abstract",
"Introduction",
"Results",
"and",
"discussion",
"Materials",
"and",
"methods"
] | [
"short",
"report",
"neuroscience"
] | 2016 | Neural signatures of perceptual inference |
Cephalopods have evolved nervous systems that parallel the complexity of mammalian brains in terms of neuronal numbers and richness in behavioral output . How the cephalopod brain develops has only been described at the morphological level , and it remains unclear where the progenitor cells are located and what molecular factors drive neurogenesis . Using histological techniques , we located dividing cells , neural progenitors and postmitotic neurons in Octopus vulgaris embryos . Our results indicate that an important pool of progenitors , expressing the conserved bHLH transcription factors achaete-scute or neurogenin , is located outside the central brain cords in the lateral lips adjacent to the eyes , suggesting that newly formed neurons migrate into the cords . Lineage-tracing experiments then showed that progenitors , depending on their location in the lateral lips , generate neurons for the different lobes , similar to the squid Doryteuthis pealeii . The finding that octopus newborn neurons migrate over long distances is reminiscent of vertebrate neurogenesis and suggests it might be a fundamental strategy for large brain development .
Cephalopod mollusks represent an invertebrate lineage that exhibits morphological as well as behavioral complexity reminiscent of vertebrates . Studying species from this group thus brings an opportunity to understand the genetic drivers of the development of nervous systems that evolved convergently with vertebrates . The adult cephalopod mollusk Octopus vulgaris has a highly centralized brain containing about 200 million nerve cells in the supra- and subesophageal mass and two optic lobes ( Young , 1963; Young , 1971 ) , yet the cellular and molecular mechanisms driving brain development remain poorly understood . At hatching , the O . vulgaris brain counts about 200 , 000 cells and occupies roughly one fourth of the total body , indicating extensive embryonic neurogenesis ( Budelmann , 1995; Giuditta et al . , 1971; Packard and Albergoni , 1970 ) . In general , neural progenitor cells are generated from ectodermal cells and divide symmetrically and asymmetrically to generate all neurons of the nervous system ( Florio and Huttner , 2014; Kriegstein and Alvarez-Buylla , 2009; Wodarz and Huttner , 2003 ) . In clades harboring species with diffuse nerve nets such as cnidaria and hemichordates , the proliferating neural progenitor cells are distributed throughout the ectoderm generating local neurons , while in ( sub ) phyla with a centralized nervous system including vertebrates , arthropods and some annelids , the neural progenitor cells are grouped in the neurectoderm . The proliferating neural progenitor cells either remain on the apical surface of the neurectoderm in vertebrates and annelids , or internalize as is the case for insects ( Cunningham and Casey , 2014; Götz and Huttner , 2005; Hartenstein and Stollewerk , 2015; Lowe et al . , 2003; Meyer and Seaver , 2009; Rentzsch et al . , 2017; Simionato et al . , 2008; Taverna et al . , 2014; Urbach and Technau , 2004 ) . In addition , long-distance migration of neurons has been described for developing vertebrate brains , in which neurons born in different zones follow long trajectories to their final location , where they intermingle to form complex circuits ( García-Moreno and Molnár , 2020 ) . Although neuronal migration has been described in developing invertebrate nervous systems as well , this process is generally limited to restricted cell populations , for example the Q , CAN and HSN neuroblasts in Caenorhabditis elegans ( Blelloch et al . , 1999; Forrester et al . , 1998; Montell , 1999 ) , or to short-range migratory events , for example in the Drosophila visual system ( Apitz and Salecker , 2015; Bhat , 2007; Morante et al . , 2011 ) . Molecular studies in vertebrates , Drosophila , C . elegans , but also cnidarians , have revealed a set of regulatory transcription factors involved in neurogenesis ( Arendt et al . , 2008; Bertrand et al . , 2002; Galliot et al . , 2009; Hirth , 2010; Layden et al . , 2012 ) . First , group B SRY-related HMG box genes ( soxB genes ) regulate the generation of the neurectoderm and also maintain neurectodermal cells in an undifferentiated and proliferative state ( Guth and Wegner , 2008; Sarkar and Hochedlinger , 2013 ) . After neurectoderm formation , members of the superfamily of basic helix-loop-helix ( bHLH ) transcription factors control the specification of neural progenitor cells and also activate neuronal differentiation pathways ( Bertrand et al . , 2002; Vervoort and Ledent , 2001 ) . The bHLH protein superfamily is divided in groups A-F , based on structural and biochemical properties . bHLH genes such as atonal , neurogenic differentiation ( neuroD ) and neurogenin ( ngn ) , and achaete-scute ( asc ) complex members are classified in group A . The role of bHLH genes is best described in nervous system development where members of this group A ( and some of group B ) regulate the patterning , differentiation and specification of neurons , from sponges to primates ( Powell and Jarman , 2008; Simionato et al . , 2007; Vervoort and Ledent , 2001 ) . Although the level of differentiation of progenitor cells in which these factors are expressed and the sequential expression of bHLH genes differs slightly across species , they all steer progenitor cells towards a neural fate . After this neural progenitor commitment , regulatory genes such as musashi , prospero , and embryonic lethal abnormal vision ( elav ) ( vertebrate hu ) , will activate programs to initiate differentiation of progenitor cells into neurons ( Choksi et al . , 2006; Okano et al . , 2002; Pascale et al . , 2008 ) . In the end , postmitotic neurons will mature , form synapses , and produce neurotransmitters to form a fully functional central nervous system ( CNS ) . These neurogenic processes are poorly studied in lophotrochozoans , and above all , in lophotrochozoans with a complex nervous system such as O . vulgaris . How and where neurons are generated , whether they migrate and what factors drive their differentiation and integration remain largely unknown . One lineage tracing study in the squid Doryteuthis pealeii revealed that cells surrounding the eye placode contribute to the anterior chamber organ , the supraesophageal mass , the buccal mass and the buccal ganglion ( Koenig et al . , 2016 ) . Our study is the first combining gene expression data with functional lineage tracing to explain neurogenesis in cephalopods . The high fecundity of O . vulgaris , spawning small and transparent eggs that can be kept in a remote tank system , as well as the available genomic data recently made it a tractable species for embryonic studies ( Deryckere et al . , 2020; Zarrella et al . , 2019 ) . Here , we have visualized the anatomy of brain development in 3D using light sheet microscopy . Expression analysis of several transcription factors , proliferation and neural markers revealed the spatial and temporal pattern of O . vulgaris brain development from stage IX onwards . Lastly , lineage-tracing studies using CFDA-SE flash labeling proved that the neurogenic area is spatially and temporally patterned . Our data suggest that O . vulgaris deploys a neurogenic transcription factor sequence during neurogenesis , as well as extensive neural migration , that are reminiscent of vertebrate mechanisms of large brain generation .
O . vulgaris displays a direct embryonic development , giving rise to actively feeding paralarvae . Their embryonic development takes approximately 40 days at 19°C and has been classified in 20 major stages I-XX , with some stages subdivided in an early and late part ( Deryckere et al . , 2020; Naef , 1928 ) . Organogenesis starts at Stage VII . 2 and can be split in early , mid- , and late phases ( Figure 1A ) . We used paraffin sectioning and light sheet imaging combined with DAPI staining to document the different steps of brain development in 2 and 3 dimensions . In the early organogenesis phase , the brain anlagen ( cordal ) are elongated and interconnected ( in three dimensions ) ( Figure 1A , B ) as also described by Shigeno et al . for Octopus bimaculoides ( Shigeno et al . , 2015 ) . On the most anterior side , lateral and posterior to the mouth and foregut , the cerebral cord ( CC ) will give rise to the lobes of the supraesophageal mass ( SEM ) , while on the posterior side of the embryo , the palliovisceral ( PVC ) and pedal ( PC ) cords will form the subesophageal mass ( SUB ) . The two bilateral optic cords ( OC ) will differentiate and grow to form the optic lobes ( OL ) ( Figure 1A , C ) . At hatching , the central brain that now surrounds the esophagus is considerably big and contains densely packed nuclei ( Figure 1D ) . In contrast to the growing brain cords , the tissue adjacent to the eye primordia first grows in size from Stage VII . 2 to Stage XV . 2 , and then shrinks to disappear at hatching , suggesting this tissue might contribute to the inner head structures ( Figure 1A ) . This tissue was first described as ‘Kopflappen’ by Marquis and later as ‘anterior chamber organ’ by Koenig et al . , 2016; Marquis , 1989; Shigeno et al . , 2001; Yamamoto et al . , 2003 . However , the term ‘anterior chamber organ’ was introduced by Young , referring to neurovenous tissue adjacent to the adult eye , that presumably regulates the fluid content between the lens and cornea ( Young , 1970 ) . The term was most likely wrongly adopted embryonically as the tissue at that stage is not restricted to the anterior side of each eye , but is rather placed lateral to the eye fields . Based on 3D light sheet imaging , we could clearly show that the structure is also connected to the central brain , through a stream-like transition zone . Because of its position and shape , we renamed the tissue to ‘lateral lips’ ( LL , indicated in mint green in Figure 1C , E ) and introduced the terms ‘anterior transition zone’ ( ATZ ) and ‘posterior transition zone’ ( PTZ ) for the region that interconnects the lateral lips with the central brain on the anterior or posterior side of the embryo , respectively ( Figure 1C , F ) . In order to map the ( early ) patterns of neuronal development in O . vulgaris , we studied the expression of the pan-neuronal elav gene . ELAV/Hu RNA-binding proteins are a family of splicing factors , predominantly present in differentiating neurons , from the moment they exit the cell cycle ( Colombrita et al . , 2013 ) . Through blast searches against the full-length transcriptome of O . vulgaris embryos and paralarval brains generated in this study ( see Materials and methods ) , we identified three candidate elav transcripts . We performed a phylogenetic analysis on these ELAV proteins together with 27 ELAV sequences from 20 other species , five non-neural ELAV sequences from four other species and five PolyA-binding sequences in order to root the tree ( Figure 2—figure supplement 1 ) . One of the O . vulgaris ELAV candidate proteins nested with vertebrate HuC/D and invertebrate neural ELAV homologs . We refer to this sequence as Ov-ELAV . The two other candidate proteins could be identified as non-neural ( Figure 2—figure supplement 1 ) . We also performed a conserved domain ( CD ) -Search and detected an ELAV/HuD family splicing factor domain which is characteristic of ELAV proteins . We mapped the expression of Ov-elav using in situ hybridization at embryonic stages IX , XI , XIII , XV . 2 , XVII , XIX . 1 , and XX . 2 ( Figure 2A–G , Figure 2—figure supplement 2 ) . These stages serve as a starting point for the characterization of nervous system development during organogenesis ( Stage IX-XVII ) and maturation phases ( Stage XIX . 1 and XX . 2 ) of embryonic development . Ov-elav transcripts were detected throughout development with little expression at Stage IX in the cerebral , palliovisceral , pedal , and optic cords ( Figure 2A ) . Strong staining was observed from Stage XI onwards , with low-level staining in the lateral lips surrounding the eye placode and more intense staining in all cords and surrounding the mouth ( Figure 2B ) . A similar pattern was observed at subsequent stages with highest staining intensity in the central brain cords/masses and low intensity staining in the lateral lips . At Stage XIII , a region in between the lateral lips and the central brain on the posterior side of the embryo showed intermediate Ov-elav expression . This region corresponds to the posterior transition zone ( PTZ ) , introduced earlier ( Figure 2C ) . From Stage XV . 2 onwards , cells on the anterior side of the embryo and laterally from the supraesophageal mass , corresponding to the anterior transition zone ( ATZ ) , showed intermediate Ov-elav expression levels as well ( Figure 2D , E ) . In the optic lobes at stages XIX . 1 and XX . 2 , Ov-elav expression was highest in the medulla and inner granular layer , but lower in the outer granular layer . The ATZ and PTZ shrink toward the end of embryonic development and seem to become integrated in the developing brain ( Figure 2F , G ) . In summary , our data indicate that neuronal differentiation is present at early organogenesis phases already , and the transition zones as well as the growing cords and masses contain large numbers of differentiating neurons . To address the maturation of these neurons , we investigated the expression pattern of synaptotagmin . Synaptotagmin is a presynaptic calcium sensor necessary for neurotransmitter release , present in mature synapses and thus in differentiated , functional neurons ( Poskanzer et al . , 2003 ) . In addition , we also visualized neuronal somata and neuropil using immunoreactivity against acetylated alpha-tubulin , as previously described for multiple cephalopod species ( Figure 2H–U , Figure 2—figure supplement 3; Jung et al . , 2018; Kingston et al . , 2015; Scaros et al . , 2018; Shigeno et al . , 2015; Shigeno and Yamamoto , 2002; Wollesen et al . , 2009; Wollesen et al . , 2012 ) . In O . vulgaris , we observed expression of Ov-syt in the retina from Stage XI onwards ( Figure 2H , I ) . In the central brain , it was first expressed in the outer layers of the optic cord at Stage XIII , and in the optic lobe medulla , supra- and subesophageal masses from Stage XV . 2 onwards , pointing to a sequential maturation ( Figure 2J–N ) . First low-level immunoreactivity against acetylated alpha tubulin was found from Stage XV . 2 onwards within the optic lobes ( Figure 2O–U ) . We therefore used the adult terminology ( optic lobe , supra- and subesophageal mass ) for CNS annotation from this stage onwards , since the embryonic cords demonstrate clear signs of differentiation . The developing octopus brain thus consists mainly of Ov-elav expressing cells , and it remains unclear where these cells are initially generated . A previous report in the squid D . pealeii suggested that cells surrounding the eye placode contribute to the brain ( Koenig et al . , 2016 ) . This area might therefore harbor a proliferative zone contributing to embryonic neurogenesis . In order to locate proliferating cells that might contribute to CNS development , we used immunoreactivity against phospho-histone H3 ( PH3 ) and mapped the expression pattern of Ov-pcna . PCNA is a nuclear protein required for DNA replication and repair , and functions as a cofactor of DNA polymerase-delta in eukaryotes and archaea ( Barry and Bell , 2006; Baserga , 1991; Prelich et al . , 1987 ) . Expression of pcna is tightly regulated and peaks at late G1 and S phases of the cell cycle ( Santos et al . , 2015 ) . The expression of the O . vulgaris homolog was mapped using in situ hybridization at embryonic Stages VII . 2 , IX , XI , XIII , XV . 2 , XVII , XIX . 1 , and XX . 2 . Ov-pcna transcripts were detected throughout embryonic development ( Figure 3A–D , I–L ) . While at Stages VII . 2 and IX , transcripts were found in most embryonic tissues ( mouth apparatus , retina , lateral lips and cords , in the latter at lower intensity ) , transcripts gradually disappeared in the optic , cerebral , palliovisceral and pedal cords from Stage XI onwards ( Figure 3A–C ) . At Stage XI , most Ov-pcna expressing cells that were found adjacent to the eye , located to an area that is Ov-elav negative and that might be part of the epithelium lining the sinus ophthalmicus ( Figure 3C ) . Ov-pcna transcripts were dispersed throughout the lateral lips and were absent from the transition zones and central brain at Stages XIII-XX . 2 ( Figure 3D , I–L ) . Since transcripts of Ov-pcna might still be present at low level in cells that finished DNA replication , we also mapped the phosphorylation of histone H3 during embryonic development ( Figure 3E–H , M–P ) . Histone H3 is phosphorylated on serine 10 and serine 28 during early mitosis and dephosphorylated at the end of mitosis in eukaryotes ( Hans and Dimitrov , 2001; Prigent and Dimitrov , 2003; Wei et al . , 1998 ) . Immunohistochemistry on embryonic tissue from Stage VII . 2 to XX . 2 showed that mitotic activity in the embryonic head region was high at the beginning of organogenesis , with the highest level of dividing cells present in the mouth apparatus , at the apical side in the developing retina and spread throughout the lateral lips ( Figure 3E–H , M–P ) . Consistent with Ov-pcna expression , the number of PH3-positive cells in the cerebral , optic , palliovisceral and pedal cords was limited , with possibly very few , single , dim and seemingly randomly organized positive cells present until Stage XIII ( Figure 3E–H ) . At late organogenesis and maturation stages , PH3-positive cells located to the lateral lips and were absent from the central brain ( Figure 3M–P ) . Given that the elav-positive areas are almost devoid of dividing cells , and the transition zone represents an area connecting the mitotically active lateral lips to the growing CNS , our data strongly suggest that the lateral lips represent a major octopus neurogenic zone . In order to find molecular support for the neurogenic character of the lateral lip cells , we mapped the expression of conserved genes involved in neural stem cell specification and differentiation , namely soxB1 , ascl1 , ngn , and neuroD . First , to spatially map the extent of the neurectoderm , we studied the expression of soxB1 . Sox genes have been divided into seven groups ( A-G ) based on the sequence of their high mobility group ( HMG ) -box that binds to DNA in a sequence-specific manner ( Sasai , 2001; Wegner , 1999 ) . Members of the SoxB group play important roles in ( early ) neurogenesis in various bilaterians and can be further divided in two sub-groups SoxB1 and SoxB2 based on sequence differences outside the HMG-box and on their activity ( Bowles et al . , 2000; She and Yang , 2015 ) . Using BLASTn and tBLASTn searches , we identified O . vulgaris members for groups B-E , but not for group A which is mammalian specific or group G which is also restricted to particular vertebrate lineages ( Bowles et al . , 2000; Heenan et al . , 2016 ) . We could also not identify SOXF family members in the available cephalopod transcriptomes of O . vulgaris , O . bimaculoides , Sepia officinalis , and Euprymna scolopes ( Figure 4—figure supplement 1 ) . Based on phylogenetic analysis including 31 SOXB , 3 SOXC , 4 SOXD , 4 SOXE , and 3 SOXF amino acid sequences from other species together with 5 TCF/LEF sequences to root the tree , we identified single O . vulgaris homologs for SOXC , D and E and two SOXB homologs ( Figure 4—figure supplement 1 ) . One O . vulgaris SOXB protein nested within the SOXB2 clade and the other one within the SOXB1 clade , the latter named Ov-SOXB1 . CD-Search of this Ov-SOXB1 protein revealed the highly conserved SOX-TCF HMG-box and SOXp superfamily domains . Ov-soxB1 transcript levels were high in the head region at all embryonic stages examined ( Figure 4—figure supplement 2 ) . At Stage IX , Ov-soxB1 was expressed in the lateral lips , eyes and surface ectoderm surrounding the mouth apparatus ( esophagus , radula , salivary gland ) . In the developing central brain , Ov-soxB1 was expressed in the cerebral cord , but not in the palliovisceral , pedal and optic cords ( Figure 4—figure supplement 2A–D ) . At Stage XI , Ov-soxB1 was still highly expressed in the lateral lips and cerebral cord , and transcripts were now also present in the optic cord in a patched pattern ( Figure 4—figure supplement 2E–H ) . At Stage XIII , Ov-soxB1 transcripts were most highly expressed in the cerebral cord and showed a similar patched pattern in the optic cords . Transcripts were also present in the pedal cord and limited in the palliovisceral cord ( Figure 4—figure supplement 2I–L ) . At Stage XV . 2 , Ov-soxB1 expression was high in the lateral lips and supraesophageal mass . In the subesophageal mass , Ov-soxB1-positive cells were more spread and in the optic lobes , Ov-soxB1 transcripts showed the highest expression in the inner granular layer , while expression was more distributed in the outer granular layer and medulla ( Figure 4—figure supplement 2M–P ) . At Stage XVII , Ov-soxB1 transcripts were still numerous in the central brain , which has started to form neuropil ( Figure 4—figure supplement 2Q–T ) . At Stages XIX . 1 and XX . 2 , Ov-soxB1 was expressed in the lateral lips , the supra- and subesophageal mass and optic lobes where it was most highly expressed in the inner granular layer ( Figure 4—figure supplement 2U-b ) . Taken together , Ov-soxB1 was not only expressed in putative stem cells , but also in postmitotic ( Ov-elav+ ) areas , as has been described for other invertebrate species . To identify a transcription factor that would mark neural progenitors only , we mapped the expression of two proneuronal bHLH group A transcription factors ascl1 and neurogenin , and one neuronal differentiation bHLH transcription factor neuroD . By sequence similarity searches , we identified O . vulgaris homologs for achaete-scute , neurogenin and neuroD . Phylogenetic analysis of the atonal-related bHLH transcription factors including 15 NEUROD and 8 NEUROGENIN/TAP protein sequences from other species effectively assigned Ov-neuroD and Ov-ngn to their respective subfamilies ( Figure 4—figure supplement 3 ) . bHLH proteins possess a bHLH domain which is involved in DNA binding and dimerization ( Jones , 2004; Murre et al . , 1989 ) . CD-Search identified such conserved domains in both sequences . In addition , phylogenetic analysis for the achaete-scute-related bHLH gene Ov-ascl1 together with 27 other ASCa protein sequences and 10 ASCb protein sequences placed Ov-ASCL1 within the proneuronal ASCa subgroup of the bHLH superfamily ( Figure 4—figure supplement 4 ) . CD-Search revealed a highly conserved DNA-binding bHLH domain . Achaete-scute homologs are involved in vertebrate neural identity determination and are key regulators in Drosophila neuroblast generation ( Cabrera et al . , 1987; Skeath and Carroll , 1992; Vervoort and Ledent , 2001 ) . In O . vulgaris , Ov-ascl1 transcripts were detected throughout embryonic development at all stages tested ( Figure 4A–G , Figure 4—figure supplement 5 ) . At Stage IX , Ov-ascl1 was expressed in the lateral lips surrounding the eye primordia . Expression could also be observed in retinal cells of the developing eyes and in the tissue delineating the mouth apparatus , and was absent from the cerebral , pedal , palliovisceral and optic cords ( Figure 4A ) . At subsequent stages of organogenesis ( XI , XIII , XV . 2 , and XVII ) , Ov-ascl1 was expressed in the lateral lips and was absent from the developing cords/brain lobes ( Figure 4B–E , low level staining that did not consistently appear in all replicates was considered background ) . During maturation and right before hatching , the number of cells expressing Ov-ascl1 significantly decreased with the thinning lateral lips and transcripts were still absent from the optic lobes , supra- and subesophageal masses ( Figure 4F , G ) . Similar to achaete-scute , homologs of neurogenin are highly conserved proneural genes expressed in early nervous system development , before neuronal differentiation in vertebrates , but also annelids ( Simionato et al . , 2007; Simionato et al . , 2008; Sur et al . , 2017; Vervoort and Ledent , 2001 ) . In O . vulgaris embryos , we found Ov-ngn transcripts in the lateral lips throughout organogenesis and maturation phases , while transcripts were absent from the developing brain cords ( Stage IX-XX . 2 , Figure 4H–N , Figure 4—figure supplement 6 , low level staining that did not consistently appear in all replicates was considered background ) . In addition , very few Ov-ngn transcripts were present in both transition zones at Stages XIII to XX . 2 ( Figure 4J–N ) . Equivalent to Ov-ascl1 , the number of Ov-ngn expressing cells decreased considerably in the maturation phase with the thinning of the lateral lips and at Stage XX . 2 , only few cells were expressing Ov-ngn ( Figure 4M , N ) . In summary , expression of both Ov-ascl1 and Ov-ngn consistently marks the proliferative lateral lips , with Ov-ascl1 expressing cells being more numerous compared to Ov-ngn expressing cells . The second atonal-related bHLH transcription factor studied here was NeuroD ( Figure 4O–U , Figure 4—figure supplement 7 ) . Its vertebrate and annelid homologs are expressed at the time neurons differentiate ( Sur et al . , 2017; Vervoort and Ledent , 2001 ) . At the beginning of organogenesis ( Stage IX and XI ) , Ov-neuroD was expressed in the cerebral , palliovisceral , pedal and optic cords of the central nervous system . Transcripts were also detected in the lateral lips , although in fewer cells compared to the cords ( Figure 4O , P ) . At Stage XI in the optic cords , a medio-lateral expression gradient was observed with high Ov-neuroD at the medial side and low Ov-neuroD at the lateral side , closer to the eyes . Ov-neuroD also seemed more highly expressed in the cerebral and palliovisceral cords compared to the pedal cord ( Figure 4P ) . At Stage XIII , Ov-neuroD transcripts were absent from the region of Ov-ascl1 expressing cells in the lateral lips , but were present adjacent to those cells on the posterior side of the embryo next to the sinus ophthalmicus , and mark the posterior transition zone . In addition , Ov-neuroD expression in the cerebral , palliovisceral and optic cords was significantly reduced compared to earlier stages and expression in the pedal cord was elevated ( Figure 4Q ) . At Stages XV . 2 and XVII , high level Ov-neuroD expression marked both the anterior and posterior transition zones . In the central brain , transcripts were present at low level in the optic lobes on the medial side , and in the outer layers of the supra- and subesophageal masses ( Figure 4R , S , low level staining in the developing cords/lobes that did not consistently appear in all replicates was considered background ) . A similar expression pattern was visible at Stages XIX . 1 and XX . 2 with clear expression in the transition zones ( Figure 4T , U ) . Ov-neuroD transcripts thus label the transition zones that connect the proliferative lateral lips to the postmitotic central brain . While our data indicate that the lateral lips are a proliferating region with cells expressing the typical neurogenic transcription factors Ov-ascl1 and Ov-ngn , it is not clear whether these represent different progenitor types . In addition , it was not proven yet that Ov-neuroD effectively labeled postmitotic cells . In a hybridization chain reaction experiment combined with immunohistochemistry , we show that Ov-ngn and Ov-ascl1 are likely expressed in a different subset of progenitor cells , since their expression was not overlapping ( Figure 5A–C ) . Furthermore , co-staining with the PH3 antibody demonstrated that Ov-ascl1+ progenitor cells seem more proliferative compared to Ov-ngn+ progenitors , that rarely overlap with the PH3+ population at this late-organogenesis stage ( Figure 5A–G ) . The proximity of Ov-ngn+ cells to dividing Ov-ascl1+ progenitors could be a sign of asymmetric progenitor division during neurogenesis in octopus . In addition , Ov-neuroD expressing cells did not co-localize with PH3 immunoreactive cells , indicating that Ov-neuroD is absent from mitotically active cells in octopus embryos , at least at Stage XV . 2 ( Figure 5H–M ) . If the progenitor cells in the lateral lips identified in this study would contribute neurons to the CNS , postmitotic neurons would need to travel long distances from the lateral lips to the central brain . However , direct proof of such neuronal cell migration is still lacking . In order to map the progeny of cells generated in the lateral lips , we performed lineage tracing experiments using the fluorescent dye CFDA-SE . This technique has been applied in vivo to study temporal neurogenesis patterns in the mammalian cerebral cortex because of the short lifetime of the dye when not incorporated in cells and thus high temporal specificity ( Govindan et al . , 2018 ) . We labeled different populations of cells in the lateral lips in early , mid or late organogenesis phases ( Stages IX , XII . 1 , or XV . 2 ) and traced their progeny to the maturation phase ( Stage XIX . 2 ) . First , we will focus on the supra- and subesophageal masses ( Figure 6 ) . At hatching , the major lobes described for the adult supra- and subesophageal masses can be distinguished , with the supraesophageal mass consisting of buccal , inferior frontal , superior frontal , subfrontal , vertical , subvertical , basal , dorsal basal and medial basal lobes , and the subesophageal mass consisting of the palliovisceral and pedal lobes ( Figure 6A–C; Young , 1971 ) . For these regions , we identified progenitor cells in the lateral lips that generate output to the different lobes in the supra- and subesophageal masses ( Figure 6D–J ) . Specifically , a major contribution to the palliovisceral lobe found its origin in progenitor cells in the dorsal-posterior quadrant of the lateral lips at Stage IX , as was suggested but not empirically proven by Koenig et al . , 2016 ( example in Figure 6D–E ) . Other ventral-posterior progenitor populations generated a limited output to the palliovisceral lobe ( orange and red injection spots in Figure 6D , G , I ) . In addition , progenitor cells in the posterior lateral lip distinctly contributed cells to the inferior and superior frontal lobes of the supraesophageal mass at all stages , which has not been reported before by Koenig et al . , 2016 ( example in Figure 6G–H ) . Progenitor cells in the ventral lateral lips also produced cells destined to the basal lobes ( basal lobe , dorsal basal lobe , medial basal lobe ) of the supraesophageal mass ( orange and red injection spots in Figure 6D , G , I ) . The majority of cells located in the supraesophageal mass , however , were derived from ventral-anterior progenitor cells in the lateral lips ( example in Figure 6D , F , I , J ) . In contrast to the supraesophageal mass and the palliovisceral lobe , few labeled progenitor populations generated cells for the pedal lobe of the subesophageal mass . Apart from the ventral-anterior progenitor populations at Stage XV . 2 ( example in Figure 6I–J ) , we did not identify progenitor populations that gave rise to a significant number of cells that migrated to the pedal lobe . Occasionally , we identified few , single randomly dispersed cells in the pedal or palliovisceral lobes originating from more ventrally located progenitor populations . Considering the very low number of cells compared to the major output from those progenitors , these cells were not depicted in the overview . Focusing on the output to the optic lobe and peduncle complex ( Figure 7 ) , we identified progenitor cells in the dorsal-anterior quadrant of the lateral lips that gave rise to cells in the optic lobes , for which labeled cells generally located to the inner and outer granular layers of the cortex ( black populations with an asterisk in Figure 7A , C , E; example of progeny in Figure 7D ) . Progenitor cells in the posterior lateral lips at Stages IX and XII . 1 generated optic lobe cells that resided in the medulla , while at Stage XV . 2 , more cells located to the optic lobe cortex as well ( example in Figure 7A , B , E , F ) . Ventral-anterior lateral lip progenitors did not generate optic lobe cells . We also identified a clear spatial patterning of progenitors that generate cells for the peduncle complex ( olfactory and peduncle lobe ) . Progenitors in the posterior and ventral lateral lips generated cells destined to this complex ( example in Figure 7A–B ) , while populations on the dorsal-anterior side did not . Taken together , our lineage tracing study identified spatial and temporal patterning in the lateral lips , which generate neurons for specific brain regions . To determine the trajectory that the progeny of lateral lip cells is taking before entering the brain , we performed a short-term lineage tracing study . Hereto , populations of cells in the lateral lips were labeled with CFDA-SE at Stage XIV . Embryos were then allowed to grow for 48–72 hr ( reaching Stage XV . 2 ) at which point they were fixed , cleared and imaged with a light sheet microscope to map the location of CFSE positive cells ( Figure 8A , Figure 8—figure supplement 1 ) . We then manually tracked the labeled cells , and reconstructed their trajectory that revealed a continuous stream of cells starting in the lateral lips , passing the posterior ( and dorsal ) side of the lateral lips and the posterior transition zone , before entering the optic lobe ( Figure 8B , C ) . On a series of optical sections , the trajectory can be followed in 2D ( Figure 8D–R ) . Labeled cells in the lateral lips ( intense labeling , Figure 8G–L ) divided and migrated posteriorly and entered the posterior transition zone ( labeling intensity decreased , Figure 8D–F ) . Then , cells could be traced towards the ventral side of the embryo in the posterior transition zone ( Figure 8F–P ) after which they occupied all layers in the optic lobe ( Figure 8G–R ) . Forty-eight hr tracing combined with HCR on thin sections showed migrating cells in the posterior transition zone that expressed Ov-neuroD and low-level Ov-elav , indicative of newly formed neurons ( Figure 8—figure supplement 2A–D ) . The first cells that reached the optic lobe expressed Ov-elav , confirming their neuronal identity ( Figure 8—figure supplement 2B , E ) . Populations labeled at a different location in the lateral lips showed similar trajectories ( even the most anterior labeled population in Figure 8—figure supplement 1 ) , with cells passing the dorsal and posterior lateral lips before entering the posterior transition zone and then the optic lobe . Taken together , cells destined for the optic lobe seem to take a defined path via the posterior transition zone .
To determine when and where the first postmitotic immature neurons and differentiated neurons are formed in O . vulgaris embryos , we studied the expression of Ov-elav and Ov-syt , respectively . In the organogenesis phase , Ov-elav is expressed in embryonic cells lining the yolk envelope that form the cords , as in O . bimaculoides ( Shigeno et al . , 2015 ) . The expression pattern is also consistent with the description of brain precursor regions proposed by Marquis after cytological studies in O . vulgaris , reinforcing the use of elav as a reliable marker for young octopus neurons ( Marquis , 1989 ) . We observed low level expression in all cords at Stage IX . In O . bimaculoides , expression was already reported from Stage VII . 2 onwards , pointing toward rapid neuronal differentiation ( Shigeno et al . , 2015 ) . In contrast to both octopus species , Sof-elav1 expression in the cuttlefish Sepia officinalis is unequally distributed over the different cords , and disappears toward the end of embryonic development , pointing to a different timing of neuronal differentiation and a more advanced maturation of the Sepia brain at hatching ( Buresi et al . , 2013 ) . Consistent with this , S . officinalis embryos have been described to respond to tactile and chemical as well as visual cues from within the egg capsule from Stage XV . 1 and XVI onwards , respectively ( Romagny et al . , 2012 ) ( for interspecies stage comparison , see Deryckere et al . , 2020 ) . In our hands , O . vulgaris embryos only seem to react to visual stimuli ( chromatophore contraction in response to light change ) and mechanical stimuli ( mantle contraction after tapping the chorion ) from Stage XIX . 1 onwards ( preliminary observation ) . Furthermore , neural processes visualized with immunostaining against acetylated alpha-tubulin revealed the presence of neurites in the cerebral , palliovisceral , pedal , and optic cords of O . bimaculoides and S . officinalis embryos as early as Stage VIII , while we only observed expression of Ov-syt and presence of acetylated alpha-tubulin in the central brain from Stages XIII and XV . 2 onwards , respectively ( Baratte and Bonnaud , 2009; Shigeno et al . , 2015 ) . These findings further support the delayed maturation of the brain in O . vulgaris , that perhaps uses its paralarval phase to complete maturation of the nervous system . Ov-soxB1 expression is not restricted to neurectodermal progenitor regions . Consistent with Sof-soxB1 expression in S . officinalis , Ov-soxB1 is expressed at high level in the surface ectoderm , in the developing eyes , and is absent from the gills and stellate ganglia ( Focareta and Cole , 2016 ) . Next to ( neur ) ectodermal expression , soxB1 is also expressed in sensory epithelia in vertebrates , invertebrate mollusks and acoelomate worms ( Focareta and Cole , 2016; Guo et al . , 2010; Kiernan et al . , 2005; Le Gouar et al . , 2004; Neves et al . , 2007; Semmler et al . , 2010 ) . In addition , SOXB1 proteins are present in both neurectodermal stem cells and differentiated neurons in certain species . Drosophila soxN and Schmidtea polychroa soxB1 for example are expressed in early specification events in the CNS , but also in differentiated parts where they are involved in neuronal differentiation and axonal patterning , suggesting a dual role for protostome soxB1 ( Ferrero et al . , 2014; Girard et al . , 2006; Monjo and Romero , 2015; Phochanukul and Russell , 2010 ) . While such a general function in neuronal differentiation of vertebrate soxB1 factors has not been shown , some subtypes of ( inter ) neurons do require SOXB1 proteins for proper specification and migration ( Cavallaro et al . , 2008; Ekonomou et al . , 2005; Panayi et al . , 2010 ) . Similar to many other species , our expression data suggest a dual role for Ov-soxB1 , in early nervous system development to specify neural fate , and later on to steer neural cell differentiation . After neurectoderm establishment and neural stem cell formation regulated by SOXB1 transcription factors , neural progenitors must be specified . In most bilaterians , this function has been attributed to members of the bHLH family of transcription factors , including atonal , neurogenin , neuroD and achaete-scute subfamilies . Our study is the first one mapping the expression of ascl1 in cephalopods and together with the expression pattern of Ov-ngn , suggests the presence of a neurogenic domain in the lateral lips , outside the cords of the central brain ( Figure 9A , B ) . In the annelids Capitella teleta and P . dumerilii , neurogenin and ash1 are both expressed in actively proliferating neural progenitor cells in the neurectoderm ( Figure 9C ) . While Ct-ngn positive cells stay on the apical side , the Ct-ash1-positive progenitors ingress and undergo limited division before becoming Ct-elav positive ( Demilly et al . , 2013; Meyer and Seaver , 2009; Simionato et al . , 2008; Sur et al . , 2017; Sur et al . , 2020 ) . In O . vulgaris , the Ov-ascl1 progenitor population seems to be more proliferative compared to the Ov-ngn progenitor population , at least at Stage XV . 2 ( late organogenesis ) , which suggests their sequential expression could be turned around in cephalopods ( Figure 9B , C ) . In addition , the Ov-elav expressing neurons in the cords are not located immediately basal from the Ov-ascl1 or Ov-ngn positive pool of neural progenitor cells . In particular , a population of Ov-neuroD expressing cells is found in the transition zones , in between the Ov-ascl1+ and Ov-ngn+ lateral lips and the developing Ov-elav+ brain , suggesting an intermediate , Ov-neuroD+ population is present in O . vulgaris ( Figure 9A , B ) . While lost in Drosophila and Ciona intestinalis , neuroD genes can be widely found in bilaterians , where they steer the differentiation of neurons ( Ledent et al . , 2002; Stollewerk and Simpson , 2005 ) . In contrast , in the trunk of the annelid P . dumerilii ( but not C . teleta ) and in the developing nervous system of the planaria S . polychroa and S . mediterranea , neuroD seems broadly expressed in the neurectoderm , which suggests that the role of neuroD might not be conserved in Spiralia ( Figure 9C; Cowles et al . , 2013; Meyer and Seaver , 2009; Monjo and Romero , 2015; Simionato et al . , 2008; Sur et al . , 2020 ) . However , co-localization studies of neuroD with progenitor marker genes are still lacking in these species . In vertebrates , ascl1 and ngn are expressed in a complementary fashion , both exerting proneuronal functions in neurogenesis ( Bertrand et al . , 2002; Castro et al . , 2011; Farah et al . , 2000; Gradwohl et al . , 1996; Lee et al . , 1995; Ma et al . , 1996 ) . Similar to our data in octopus , neurogenin expression precedes , but also partially overlaps with that of neuroD ( Grimaldi et al . , 2008 ) . In contrast , Drosophila has only one neurogenin/neuroD homolog called tap , which relates better to neurogenin based on sequence similarities , but does not have a proneural role and is expressed in a few neurons , regulating their differentiation , axonal growth and guidance ( Gautier et al . , 1997; Vervoort and Ledent , 2001; Yuan et al . , 2016 ) . In arthropods , the members of the achaete-scute subfamily are expressed in proneural clusters and promote the generation of neural progenitor cells from quiescent ectodermal cells ( Bertrand et al . , 2002; Cubas et al . , 1991; Quan and Hassan , 2005; Skeath and Carroll , 1991; Stollewerk and Chipman , 2006 ) . In cnidaria as well , ashA promotes neurogenesis during its development ( Layden et al . , 2012; Figure 9C ) . In O . bimaculoides , neurogenin and neuroD transcripts were detected in the prospective cerebral , palliovisceral and pedal cords at Stage VIII , based on whole mount in situ hybridization , but did not distinguish the lateral lips from the cordal areas ( Shigeno et al . , 2015 ) . Based on cross-sections , neurogenin transcripts were found on the surface of the embryo , and neuroD more at the level of the cords , similar to our observations in O . vulgaris ( Shigeno et al . , 2015 ) . In the vertebrate neural tube , NEUROD was also found on the basal side , in postmitotic neuronal and glial cells and is required for the differentiation of neurons in the inner ear , cerebellum and hippocampus ( reviewed in Dennis et al . , 2019 ) . Its expression pattern in O . vulgaris suggests that upon differentiation , neurons express Ov-neuroD before expressing Ov-elav , which is similar to the role of Ov-neuroD as neuronal-differentiation bHLH transcription factor , as described for vertebrates and the annelid C . teleta ( Bertrand et al . , 2002; Farah et al . , 2000; Lee et al . , 1995; Sur et al . , 2020 ) . We also revealed possible overlapping expression of Ov-soxB1 with Ov-ascl1 and Ov-ngn in the lateral lips throughout development , suggesting a similar activation of Ov-ascl1 and Ov-ngn by Ov-soxB1 in the presumptive neurectoderm ( Figure 9B; Amador-Arjona et al . , 2015; Ferrero et al . , 2014 ) . Therefore , Ov-ngn and Ov-ascl1 are expressed at the right time and place to be the major proneuronal genes for the formation of the central nervous system in O . vulgaris . Our data further substantiate the conserved expression of proneural bHLH transcription factors , which suggests they might already have been present in the ur-bilateria . The lateral lips harbor proliferating cells with a signature of neural progenitors , while mature neurons are found in the cords . The overt absence of proliferative cells in the developing brain in an organism with such a big and centralized CNS is striking , but not uncommon for mollusks . Mitosis in the Aplysia CNS is infrequent from early embryogenesis to adulthood ( Jacob , 1984 ) . Compared to other invertebrates with large brains such as insects , however , this seems to be a rather unique strategy . The developing optic lobe in Drosophila for example has dividing neuroblasts in two main proliferation centers inside the lobe . These neuroblasts had previously invaginated from the neuroepithelium and start dividing only when in the brain ( Álvarez and Díaz-Benjumea , 2018; Apitz and Salecker , 2015; Green et al . , 1993; Hofbauer and Campos-Ortega , 1990; Walsh and Doe , 2017 ) . The physical disconnection between the neural stem cells and their progeny in octopus suggests that secondary progenitor cells or neurons migrate towards the central brain where they integrate . In lophotrochozoans such as the annelids C . teleta and P . dumerilii , dividing neural progenitors are located on the apical side of the neuroepithelium and it is their progeny that ingresses to form the CNS ( Meyer and Seaver , 2009; Monjo and Romero , 2015; Sur et al . , 2020 ) . While resembling the postmitotic migration in O . vulgaris , their migratory path is short and the number of neurons remains limited , making direct comparison with O . vulgaris difficult . As described here and by Shigeno et al . , the nervous system of cephalopods originates from a system of cords , similar to the nervous system of primitive mollusks like Aculifera including chitons , and in contrast to Concifera including gastropods , whose CNS knows a ganglionic origin ( Richter et al . , 2010; Shigeno et al . , 2015; Sumner-Rooney and Sigwart , 2018 ) . Gastropod ganglia seem to derive from ectodermal cells in the body wall that proliferate and delaminate/migrate inwards to join the developing ganglia . These studies suggest that similar to annelids , neural progenitor cells in the gastropod mollusk Aplysia californica , divide in proliferative zones in the body wall and their progeny migrates few cell lengths to the nearest ganglion ( Demian and Yousif , 1975; Jacob , 1984 ) . Depending on the location in the proliferative zone , cells migrate either in a columnar stream , or individually using pseudopodia . Evidence for cell migration in cephalopods came from Loligo vulgaris , where in vitro cultures of the ‘oculo-ganglionar complex’ showed extensive migration of cells and differentiation into bi-and multipolar neurons ( Marthy and Aroles , 1987 ) . Here , we showed that the progeny of octopus lateral lip progenitors migrates over long distances before integration in the CNS . Strikingly , we observed that the migratory path does not just represent the shortest route to destiny . Instead , we found that independent from the anterior-posterior axis , progenitor populations in the dorsal lateral lips generate cells for the optic lobe that pass through the posterior transition zone . Cells thus migrate from the most anterior part towards the posterior part of the lateral lips before entering the posterior transition zone and eventually migrating in all directions , spreading throughout the optic lobe . This suggests active , directed migration controlled by guidance cues . Which cell intrinsic and/or extrinsic cues govern the migratory process , remains to be studied . Active neural migration guided by extrinsic cues is common in vertebrates that build their brain from a large , spatially patterned and proliferating neural epithelium folded into a tube . Neurons are generated in a temporally controlled fashion and migrate away from the progenitor zone surrounding the ventricles to form the grey matter . The latter entirely consists of postmitotic neurons that start growing neurites and connect while additional postmitotic neurons are migrating in between them to their target region ( Marín et al . , 2010; Marín and Rubenstein , 2003; Paridaen and Huttner , 2014; Taverna et al . , 2014 ) . In that respect , the development of the O . vulgaris brain seems strikingly similar . Yet O . vulgaris seems to pattern the lateral lips in such a manner that entire brain lobes are derived from specific areas in the lateral lips , whereas specification in vertebrates links spatial ( i . e . progenitor area ) or temporal patterning to cell types that are intermingled in certain regions . An example is the subpallium , that generates ( among other cell types ) cortical interneurons , while the dorsal pallium generates pyramidal neurons , and both are mixed in the cerebral cortex . Another example is the postnatal V-SVZ , where different areas generate distinct interneuron types destined for the olfactory lobe ( Hatten , 1999; Marín and Rubenstein , 2001; Marín and Rubenstein , 2003; Medina and Abellán , 2009 ) . Neurons that are born in a specific region of the lateral lips thus seem instructed to migrate to a certain brain lobe , but how they get specified to different cell types within that lobe remains unclear . Indeed , cells from one injection site mostly spread out over a whole lobe , except for the optic lobe , where we identified lateral lip subregions that are biased to generate optic lobe cortex or optic lobe medulla neurons . Interestingly , the proportion of cells in those layers increased when progenitor labeling was performed at later stages , indicating that cell types might be specified in a temporal manner as well . To investigate this observation in more detail , additional molecular markers for different neuronal subpopulations need to be identified . Furthermore , while proneural bHLH transcription factors are likely involved in specifying and maintaining neural progenitor identity , the transcription factors that serve as terminal selectors need to be revealed . Lastly , as the lateral lips shrink to almost completely disappear towards the end of development , there must be a second , yet undefined source of progenitors , which generates additional neurons in post-hatching stages for the adult brain .
Live O . vulgaris embryos were obtained from the lab of E . Almansa ( IEO , Tenerife ) , transferred to the lab of Developmental neurobiology and kept in a closed standalone system ( Deryckere et al . , 2020 ) . Embryos were observed , staged and sampled daily , followed by overnight fixation in 4% paraformaldehyde ( PFA ) in phosphate buffered saline ( PBS ) . After a wash in PBS , embryos were manually dechorionated with tweezers and transferred to embedding cassettes ( Tissue-Tek Biopsy 6-Chamber Cassette , Sakura ) . For paraffin processing , the cassettes were immersed in 0 . 9% NaCl overnight before progressive dehydration and paraffin-embedding using an Excelsior AS Tissue Processor and HistoStar Embedding Workstation ( Thermo Scientific ) . 6 μm-thick transversal sections were made for subsequent immunohistochemistry or in situ hybridization . Embryo sections were processed using an automated platform ( Ventana Discovery , Roche ) for direct fluorescent staining . Primary antibodies mouse anti-Acetylated alpha Tubulin ( Sigma T6793 ) , rabbit anti phospho-histone H3 ( Ser10 ) ( Millipore 06–570 ) , and goat anti fluorescein ( Novus Biologicals NB600-493 ) and secondary antibodies donkey anti-mouse Alexa 488 , donkey anti-rabbit Alexa 555 , and donkey anti-goat Alexa 488 ( Life Technologies ) were each diluted in Pierce Immunostain or antibody diluent ( Roche ) and incubated at a final concentration of 1:300 . Sections were then incubated in DAPI and mounted in Mowiol . Images were acquired with a Leica DM6 upright microscope and minimum/maximum displayed pixel values were adjusted in Fiji ( Schindelin et al . , 2012 ) . Images used in the figures represent the staining pattern observed in multiple embryos ( number of replicates presented in Supplementary file 1 ) . In order to construct a full-length transcriptome of O . vulgaris embryos and paralarval brains , the Iso-Seq method was used . RNA was extracted from a pool of 25 Stage XI-XII embryos using Tri-reagent ( Invitrogen ) and the Qiagen Micro kit ( Qiagen ) . cDNA was synthesized with the Clontech SMARTer PCR cDNA Synthesis Kit ( Takara Bio Inc ) . RNA was also extracted from dissected brains of one-day old paralarvae in a similar manner and cDNA was synthesized using the NEBNext cDNA Synthesis and amplification kit . Both samples were sequenced on the PacBio Sequel at the Genomics Core at KU Leuven ( Belgium ) following the protocol recommended by PacBio . Only cDNAs containing polyA-tails were selected , with the aim to retrieve full-length transcripts . This resulted in a total of 12 , 017 , 703 subreads for the embryo and 15 , 426 , 835 subreads for the brain sample . The raw data files were processed with SMRT Link release 9 . 0 . 0 software . The IsoSeq 3 . 3 pipeline was followed to generate consensus reads ( inc polish , min . passes = 1 ) . Lima ( -isoseq ) was used to retain full-length fragments that possess both primers only , to remove unwanted primer combinations and to orient the sequences . Subsequently , Poly ( A ) tails were trimmed and concatemers were removed . This resulted in 22 , 757 and 28 , 490 high-quality polished isoforms for the embryo and hatchling brain samples , respectively . Data have been deposited in SRA under the following accession number PRJNA718058 . Putative homologs of O . vulgaris achaete-scute , neurogenin , neuroD , elav , soxB1 , synaptotagmin and pcna genes were identified using tBLASTn searches against the ISOseq transcriptomes . O . vulgaris hits hereafter named Ov-ascl1 , Ov-ngn , Ov-neuroD , Ov-elav , Ov-soxB1 , Ov-syt , and Ov-pcna were then blasted against the NCBI database to verify sequence homology . Primers were designed ( primer sequences in Supplementary file 2 ) to isolate a 500–1000 bp fragment from mixed-stage O . vulgaris embryo cDNA ( synthesized using Superscript III Reverse Transcriptase ( Invitrogen ) ) ( probe sequences in Supplementary file 3 ) . The resulting PCR products were TA cloned into the pCRII-TOPO vector ( Invitrogen ) and sequenced by LGC Genomics ( Berlin ) . After plasmid linearization , anti-sense digoxigenin- ( DIG ) labeled RNA probes were generated using an Sp6- or T7-RNA polymerase and DIG RNA labeling mix ( both Roche ) following the manufacturer’s protocol . The probes were cleaned using Micro Bio-Spin P-30 Gel Columns with RNase-free Tris Buffer ( BioRad ) . Paraffin sections were processed using an automated platform ( Ventana Discovery , Roche ) with RiboMap fixation and BlueMap detection kits ( Roche ) for in situ hybridization . In short , sections are deparaffinated , heated to 37°C , post-fixed and pretreated . Then , a 4 min digestion with proteinase K ( Roche , 1:1000 in PBS-DEPC ) is followed by probe titration ( 100–300 ng per slide dependent on the probe , dissolved in Ribohybe reagent ( Roche ) ) , denaturation at 90°C for 6 min and hybridization at 70°C for 6 hr . Three stringency washes in 0 . 1X SSC at 68°C for 12 min each are followed by post-fixation . The anti-DIG-Alkaline phosphatase antibody ( Roche ) is added and sections are incubated for 30 min after which a colorimetric signal ( BCIP/NBT ) is developed for 4–9 hr ( probe dependent ) . The tissue is counterstained with Red Counterstain II ( Roche ) , followed by dehydration and mounting using Eukitt quick-hardening mounting medium ( Sigma ) . Bright-field images were taken with a Leica DM6 upright microscope and background was subtracted in Photoshop . Images used in the figures represent the expression pattern observed in multiple embryos ( number of replicates presented in Supplementary file 1 ) . HCR-3 . 0-style probe pairs for fluorescent in situ mRNA visualization were generated for Ov-ascl1 , Ov-elav , Ov-neuroD and Ov-ngn . Hereto , we used the insitu_probe_generator ( Null and Özpolat , 2020 ) , followed by BLAST searches using Blast2GO ( Conesa et al . , 2005 ) to minimize potential off-target hybridization . DNA oPools were ordered from Integrated DNA Technologies , Inc ( probe sets in Supplementary file 4 ) and dissolved in DNase/RNase-Free distilled water ( Invitrogen ) . HCR amplifiers with fluorophores B1-Alexa Fluor-546 , B2-Alexa Fluor-647 , and B3-Alexa Fluor-488 were ordered from Molecular Instruments , Inc . The Molecular Instruments HCR v3 . 0 protocol for FFPE human tissue sections , based on Choi et al . , 2016 and Choi et al . , 2018 was followed ( Choi et al . , 2016; Choi et al . , 2018 ) . Described here are adaptations from this protocol . Paraffin sections were baked at 65°C for 30 min and subsequently deparaffinized with Xylene ( 2 x 4 min ) and 100% EtOH ( 3 x 4 min ) . To permeabilize the tissue , slides were treated with proteinase K ( Roche , 1:3000 in PBS-DEPC ) for 5 min at 37°C . Slides were then rinsed 2 x 2 min with autoclaved MQ and immediately processed for HCR . After a 30 min pre-hybridization step , probe solution ( 0 . 4 pmol per probe in probe hybridization buffer ) was incubated overnight . The next day , 4 . 5 pmol of hairpin h1 and 4 . 5 pmol of hairpin h2 were snap-cooled ( 95°C for 90 s , 5 min on ice followed by 30 min at room temperature ) and added to 75 µL of amplification buffer . After overnight amplification , excess hairpins were removed by washing 3 x 10 min with 5X SSCT . After HCR , we proceeded with immunohistochemistry as described above and lastly , sections were incubated in DAPI and mounted in Mowiol . Images were acquired using a confocal microscope ( Fluoview FV1000 , Olympus ) and minimum/maximum displayed pixel values were adjusted in Fiji ( Schindelin et al . , 2012 ) . Autofluorescence in the different channels and potential aspecific amplifier binding were assessed first . Then , all probes were tested individually before multiplexing . HCR was performed on at least three slides per probe , on at least two different embryos . To determine homology , phylogenetic analyses of ASH , NEUROD , NEUROG , ELAV , and SOX families were performed . Full-length protein sequences ( when available ) were obtained using BLASTp , tBLASTn or word search in NCBI or from published articles ( accession numbers in Supplementary file 5 ) . In the case of E . scolopes , the BLAST genome server together with peptide sequences on cephalopodresearch . org were used ( Belcaid et al . , 2019 ) . In the case of O . vulgaris , nucleotide sequences were obtained using tBLASTn searches against the Iso-Seq transcriptomes , followed by a search for open reading frames using the ORFfinder tool in NCBI , which also provided the translated protein sequences ( O . vulgaris protein sequences in Supplementary file 6 ) . All protein sequences were aligned using the ‘MUSCLE Alignment’ feature ( Edgar and Sjölander , 2004 ) within MEGA-X ( Kumar et al . , 2018 ) . The matrix was then trimmed using TrimAI ( Capella-Gutiérrez et al . , 2009 ) via the Automated one algorithm . The best-fit substitution model for each alignment was determined using the Bayesian information criterion in IQ-TREE ( Nguyen et al . , 2015 ) . Maximum likelihood analyses using the LG+G4 ( ELAV ) , JTT+I+G4 ( NEUROD/NGN ) , VT+I+G4 ( ASH ) , or JTT+G4+F ( SOX ) substitution models for protein evolution were also performed in IQ-TREE , with branch supports calculated by 10 . 000 Ultrafast bootstrap replicates ( UFBoot2 ) from maximum 1000 iterations ( stopping rule ) ( Hoang et al . , 2018; Nguyen et al . , 2015 ) . The produced consensus trees were rooted and visualized with FigTree v1 . 4 . 4 ( Rambaut , 2018 ) . Domains in the predicted protein sequences of candidate O . vulgaris homologs were identified with NCBIs’ conserved domain database ( CDD ) search tool ( Lu et al . , 2020 ) . For injection of live O . vulgaris embryos , a glass capillary ( 3-000-203-G/X , Drummond ) , that was pulled using a Laser-Based Micropipette Puller ( P-2000 , Sutter Instrument; Heat 450 , Fil 4 , Vel 150 ) , and opened at 30 μm was mounted on a micromanipulator ( M3301L , WPI ) and connected to a FemtoJet ( Eppendorf ) via an injection tube . Excess seawater was removed from the egg using a tissue after which the egg was transferred to a Sylgard-coated Petri dish ( Sylgard 170 , Dowsil ) . The egg was stabilized with tweezers and the capillary was inserted in the embryonic tissue , from the stalk side through the chorion , in an angle of about 10–30 ° relative to the dorso-ventral axis of the embryo . A single 50–100 nL dose of a carboxyfluorescein diacetate succinimidyl ester ( CFDA-SE ) working solution ( 1 mM for trajectory mapping , 0 . 1 mM for long term tracing in filtered sea water , 3% FastGreen ) was injected in the lateral lips anterior , posterior , dorsal or ventral from the eye placode at developmental Stage XIV for trajectory mapping ( n = 8 ) and Stages IX ( n = 20 ) , XII . 1 ( n = 21 ) and XV . 2 ( n = 16 ) for long-term tracing . CFDA-SE is a non-fluorescent and highly membrane-permanent molecule that is cleaved by intracellular esterases once taken up by cells , resulting in a trapped , fluorescent carboxyfluorescein succinimidyl ester ( CFSE ) molecule that is bound to amino groups of intracellular proteins ( Progatzky et al . , 2013; Quah et al . , 2007 ) . The dye thus acutely labels all cells present at the injection location , and their progeny due to its long-lasting stability . After injection , embryos were placed back in a Petri dish with filtered sea water . Successful uptake of the dye was verified after 30 min using a fluorescence binocular ( SteREO Discovery . V8 with AxioCam MRc5 ( Zeiss ) ) . Individual eggs were then incubated in a 96-well plate in filtered sea water , 2% Penicillin/Streptomycin , in a 19°C incubator in the dark ( Heratherm IMC18 , Thermo Scientific ) . Viable embryos were sampled after 48 or 72 hr for trajectory mapping or when reaching Stage XIX . 2 for long-term tracing , fixed overnight in 4% PFA in PBS and then stored in PBS until dechorionation and clearing or paraffin embedding . Dechorionated embryos were cleared before light sheet imaging as previously described ( Deryckere et al . , 2020 ) . Whole mount immunohistochemistry was performed for long term CFDA tracing , during the PBS washing steps in between incubation in ScaleCUBIC-1 and ScaleCUBIC-2 as follows: after clearing in ScaleCUBIC-1 , embryos were washed four times; one time for 5 min and three times for 2 hr in PBS supplemented with 0 . 3% Triton X-100 ( PBS-T ) and then incubated in primary antibody solution in PBS-T ( goat anti-fluorescein ( Novus Biologicals NB600-493 , 1:500 , pre-incubated overnight with non-injected embryos ) ) for 2 days at 4°C . Embryos were then washed three times for 2 hr in PBS-T , secondary antibody solution was added ( donkey anti-goat Alexa 488 ( Life Technologies , 1:300 ) ) and incubated overnight at 4°C after which the samples were washed three times for 2 hr in PBS-T and then incubated in 1/2-water diluted ScaleCUBIC-2 at 37°C . Cleared and stained embryos were glued with their yolk sack on a metal plunger and imaged using a Zeiss Z1 light sheet microscope ( Carl Zeiss AG , Germany ) in low-viscosity immersion oil mix ( Mineral oil , Sigma M8410 and Silicon oil , Sigma 378488 , 1:1 ) . Then , 3D reconstructions were generated in Arivis ( Vision4D , Zeiss Edition 3 . 1 . 4 ) . For lineage tracing after CFDA-SE injection , the distribution of the progeny was mapped using optical sections . In addition , for trajectory mapping , the region including most CFSE labeled cells was manually traced using the objects drawing tool in Arivis , in order to visualize the trajectory in 3D . This same tool was also used to reconstruct the eye , optic lobe , lateral lips and posterior transition zone at Stage XV . 2 for Figure 1 , and to reconstruct the different lobes in the central brain at hatching for Figure 6 . | Octopuses have evolved incredibly large and complex nervous systems that allow them to perform impressive behaviors , like plan ahead , navigate and solve puzzles . The nervous system of the common octopus ( also known as Octopus vulgaris ) contains over half a billion nerves cells called neurons , similar to the number found in small primates . Two thirds of these cells reside in the octopuses’ arms , while the rest make-up a central brain that sits between their eyes . Very little is known about how this central brain forms in the embryo , including where the cells originate and which molecular factors drive their maturation in to adult cells . To help answer these questions , Deryckere et al . studied the brain of Octopus vulgaris at different stages of early development using various cell staining and imaging techniques . The experiments identified an important pool of dividing cells which sit in an area outside the central brain called the ‘lateral lips’ . In these cells , genes known to play a role in neural development in other animals are active , indicating that the cells had not reached their final , mature state . In contrast , the central brain did not seem to contain any of these immature cells at the point when it was growing the most . To investigate this further , Deryckere et al . used fluorescent markers to track the progeny of the dividing cells during development . This revealed that cells in the lateral lips take on a specific neuronal fate before migrating to their target region in the central brain . Newly matured neurons have also been shown to travel large distances in the embryos of vertebrates , suggesting that this mechanism may be a common strategy for building large , complex brains . Although the nervous system of the common octopus is comparable to mammals , they evolved from a very distant branch of the tree of life; indeed , their last common ancestor was a worm-like animal that lived about 600 million years ago . Studying the brain of the common octopus , as done here , could therefore provide new insights into how complex nervous systems , including our own , evolved over time . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"developmental",
"biology",
"evolutionary",
"biology"
] | 2021 | Identification of neural progenitor cells and their progeny reveals long distance migration in the developing octopus brain |
Brain age is a widely used index for quantifying individuals’ brain health as deviation from a normative brain aging trajectory . Higher-than-expected brain age is thought partially to reflect above-average rate of brain aging . Here , we explicitly tested this assumption in two independent large test datasets ( UK Biobank [main] and Lifebrain [replication]; longitudinal observations ≈ 2750 and 4200 ) by assessing the relationship between cross-sectional and longitudinal estimates of brain age . Brain age models were estimated in two different training datasets ( n ≈ 38 , 000 [main] and 1800 individuals [replication] ) based on brain structural features . The results showed no association between cross-sectional brain age and the rate of brain change measured longitudinally . Rather , brain age in adulthood was associated with the congenital factors of birth weight and polygenic scores of brain age , assumed to reflect a constant , lifelong influence on brain structure from early life . The results call for nuanced interpretations of cross-sectional indices of the aging brain and question their validity as markers of ongoing within-person changes of the aging brain . Longitudinal imaging data should be preferred whenever the goal is to understand individual change trajectories of brain and cognition in aging .
The concept of brain age is increasingly used to capture interindividual differences in the structure , function , and neurochemistry of the aging brain ( Cole and Franke , 2017 ) . The biological age of the brain is estimated typically by applying machine learning to magnetic resonance imaging ( MRI ) data to predict chronological age . The difference between predicted brain age and actuackal chronological age ( brain age delta ) reflects the deviation from the expected norm and is often used to index brain health . Brain age delta has been related to brain , mental , and cognitive health , and proved valuable in predicting outcomes such as mortality ( Cole et al . , 2018; Cole and Franke , 2017; Elliott et al . , 2019 ) . To different degrees , it is assumed that brain age delta reflects past and ongoing neurobiological aging processes ( Cole and Franke , 2017; Elliott et al . , 2019; Franke and Gaser , 2019; Smith et al . , 2020 ) . Hence , it is common to interpret positive brain age deltas as reflecting a steeper rate of brain aging; often dubbed as accelerated aging ( here both terms are used interchangeably ) ( Cole and Franke , 2017; Franke and Gaser , 2019; Smith et al . , 2019 ) . The assumption that brain age delta reflects an ongoing process of faster or slower neurobiological aging implies that there should be a relationship between cross-sectional and longitudinal estimates of brain age . Alternatively , individual deviations from the expected brain age could capture constant interindividual differences in brain structure that remain stable throughout the lifespan , reflecting early genetic and environmental influences ( Deary , 2012; Elliott et al . , 2019; Walhovd et al . , 2016 ) . These perspectives offer fundamentally divergent interpretations of higher brain age ( delta ) in groups experiencing specific life events , brain disorders , and other medical problems . Here , we tested whether brain age – derived from structural T1-weighted ( T1w ) morphological features – is related to accelerated brain aging , early-life factors , or a combination of both . If interindividual variations of brain age reflect variations in rates of ongoing brain aging ( Figure 1a ) , cross-sectional brain age delta should be positively associated with brain decline measured longitudinally . Here , we quantified individual brain change as the annual rate of change of brain age delta ( brain age deltalong ) . In addition , we also assessed brain change with a composite score of structural brain change as obtained using principal component ( PC ) analysis of change and change in the different raw structural brain features . These analyses were performed in two independent cohorts , both divided into a cross-sectional model generation ( training ) and a longitudinal , hypothesis testing ( test ) dataset . If cross-sectional variations in brain age reflect differences in brain structure established early in life , one should observe a relationship between brain age and influences associated with stable , lifelong effects on brain structure . Here , we selected two congenital factors: self-reported birth weight and polygenic scores for brain age ( PGS-BA ) , for which lifelong effects on age-related phenotypes have been shown ( Walhovd et al . , 2012; Walhovd et al . , 2020; Figure 1b ) . Birth weight reflects normal variation in body ( and brain ) size as well as prenatal conditions , whereas PGS-BA quantifies genetic liability of having a higher brain age .
Chronological age ( Figure 1c ) was predicted based on regional and global features from structural T1w MRI , including cortical thickness , area , volume , and gray-white matter contrast , as well as subcortical volume and intensity imaging-derived phenotypes ( |N| = 365 ) . See a list of the different structural features used in the model in Supplementary files 1 and 2 , and Figure 1d for pairwise correlations with age . The model was trained on 38 , 682 participants ( age range = 44 . 8–82 . 6 years ) with a single MRI from the UK Biobank ( Miller et al . , 2016 ) using gradient boosting as implemented in XGBoost ( https://xgboost . readthedocs . io ) and optimized using 10-fold cross-validation and a randomized hyperparameter search . The trained model ( Figure 1e ) was then used to predict brain age for an independent test dataset of 1372 participants with two MRIs each ( age range = 47 . 2–80 . 6 years , mean [SD] follow-up = 2 . 3 [0 . 1] years ) . The predictions – applied to the longitudinal test set – revealed a high correlation between chronological and brain age ( r = 0 . 82 ) with mean absolute error ( MAE ) = 3 . 31 years and root mean squared error ( RMSE ) = 4 . 14 years ( Figure 1f ) , comparable to other brain age models using UK Biobank MRI data ( Cole , 2020a ) . We used generalized additive models ( GAM ) to correct for the brain age bias , that is , the underestimation of brain age in older individuals and vice versa; a regression-to-the-mean bias ( Smith et al . , 2019 ) . Brain age delta was calculated as the residual from the GAM fit . Brain age delta at baseline and follow-up were strongly correlated ( r = 0 . 81 ) . To establish generalizability , we replicated our results using a different machine learning algorithm – a LASSO-based approach ( Cole , 2020a ) – and an independent training and test ( longitudinal ) dataset from the Lifebrain consortium ( Walhovd et al . , 2018 ) with up to 11 . 2 years of follow-up ( 3292 unique participants , age range = 18 . 0–94 . 4 years; technical and biological replication ) . See Figure 1—figure supplement 1 and Supplementary file 3 for additional demographic information . All the codes used to generate the results are available alongside the article and at https://github . com/LCBC-UiO/VidalPineiro_BrainAge , ( Vidal-Piñeiro , 2021; copy archived at swh:1:rev:2044c6ca40e0b8f99c9190c6edfde8ca76b559ac ) . First , we tested whether cross-sectional brain age delta was associated with brain age deltalong – that is , annual rate of change in brain age delta – using linear models controlling for age , sex , scanning site , and estimated intracranial volume ( eICV ) . We selected the centercept ( brain age delta at mean chronological age ) , instead of baseline brain age delta , to avoid statistical dependency between indices . Cross-sectional and brain age deltalong were weakly , but negatively , associated in the UK Biobank ( β = –0 . 016 [±0 . 008] delta/year , t ( p ) = –2 . 0 ( . 04 ) , r2 = 0 . 002 , Figure 2a ) . Cross-sectional and brain age deltalong were unrelated using a LASSO regression approach ( β = –0 . 003 [±0 . 006] delta/year , t ( p ) = –0 . 5 ( . 65 ) , r2 = 0 . 001 , Figure 2b ) , and in the Lifebrain replication sample ( β = –0 . 007 [±0 . 01] delta/year , t ( p ) = –0 . 6 ( . 53 ) , r2 = 0 . 001 , Figure 2c ) . Post-hoc equivalence tests showed that positive relationships with β > 0 . 010 delta/year would be rejected in all three analyses , thus confirming a lack of a meaningful relationship between cross-sectional and longitudinal brain age ( Materials and methods and Figure 2—figure supplement 1 ) . UK Biobank ( gradient boosting ) results remained not significant when brain age delta was derived by time points 1 and 2 as two independent training sets ( 10-fold cross-validation; uncorrected delta values ) , thus avoiding potential confounds with age-bias correction ( t ( p ) = 0 . 3 ( . 76 ) ) . Lifebrain results remained unaffected after including follow-up interval as an additional covariate or restricting the analysis to participants with long follow-up intervals ( >4 years; n = 424 ) . The relationship between cross-sectional and brain age deltalong was not significant in both cases ( β = –0 . 008 [±0 . 01] delta/year , t ( p ) = –0 . 7 ( . 45 ) ; β = –0 . 008 [±0 . 007] delta/year , t ( p ) = –1 . 1 ( . 26 ) ) . We additionally tested whether cross-sectional and longitudinal brain age delta ( brain age deltalong ) were associated with a composite measure of longitudinal brain change or with change in any of the structural MRI features . See Materials and methods for details . Cross-sectional brain age delta was unrelated to a principal component of change ( β = –0 . 009 [±0 . 01] year , t ( p ) = –0 . 7 ( . 46 ) , r2 = 0 . 001 ) . We did not find a significant relationship when brain age delta was computed with neither a LASSO algorithm nor using the Lifebrain sample ( β = –0 . 02 [±0 . 01] year , t ( p ) = −1 . 7 ( 0 . 09 ) , r2 = 0 . 002; β = 0 . 007 [±0 . 006] year , t ( p ) = 1 . 3 ( 0 . 2 ) , r2 = 0 . 001 ) . In contrast , brain age deltalong was associated with a principal component of change in the UK Biobank dataset as well as in both replication analyses ( all tests p<0 . 001 ) . See Figure 2—figure supplement 2 for a visual representation . For specific features , cross-sectional brain age delta was significantly related to change – in the expected direction – of features capturing lateral ventricle expansion and white matter hypointensities ( p<0 . 05 Bonferroni-corrected ) . Brain age deltalong related to change in 45 of the features pertaining to four different modalities . The results were replicated both using the LASSO algorithm and the Lifebrain dataset ( Figure 2—figure supplement 3 and Supplementary file 4 ) . Finally , we estimated the rate of aging effects using a cross-sectional model by estimating the scaling of the size of delta with age as defined in Smith et al . , 2019 . The scaling ( δ ) of brain age delta ( δ ) throughout the datasets’ age range was = 0 . 14 and 0 . 09 for the UK Biobank and the Lifebrain datasets . This corresponds to an increase in the spread of brain age delta of |δ| = 0 . 38 and 0 . 37 years – when moving from youngest to oldest – in the UK Biobank and the Lifebrain datasets , suggesting that brain age delta only modestly reflects rate of aging effects . Next , we tested whether birth weight was associated with brain age delta or change in brain age delta . Linear mixed models were used to fit time ( from baseline; years ) , birth weight , and its interaction on brain age delta using age at baseline , sex , scanning site , and eICV as covariates . Birth weight was significantly related to brain age delta ( β = –0 . 70 [±0 . 30] year/kg , t ( p ) = −2 . 3 ( 0 . 02 ) , r2 = 0 . 009 , Figure 3a ) , but not to delta change ( β = 0 . 02 [±0 . 09] year/kg , t ( p ) = 0 . 3 ( 0 . 79 ) , Figure 3c ) . Birth weights were limited to normal variations at full term ( from 2 . 5 to 4 . 5 kg; n = 770 unique individuals ) but see Figure 3—figure supplement 1 for results with varying cutoffs . The results were not affected by excluding individuals being part of multiple births ( p=0 . 02 ) and were replicated using the LASSO approach ( β = –0 . 79 [±0 . 29] year/kg , t ( p ) = −2 . 8 ( 0 . 006 ) , r2 = 0 . 009 , Figure 3b and d ) . Finally , we tested whether PGS-BA related to brain age delta and change in brain age delta ( n = 1339 ) . PGS-BA was computed using a mixture-normal model based on a genome-wide association study ( GWAS ) of the brain age delta phenotype in the UK Biobank training dataset . To test the association , linear mixed models were used as above along with the top 10 genetic PCs to account for population structure . PGS-BA was positively associated with brain age delta ( β = 0 . 54 [±0 . 09] year , t ( p ) = 9 . 4 ( <0 . 001 ) , r2 = 0 . 02 , Figure 4a ) and negatively associated with brain age delta change ( β = –0 . 06 [±0 . 03] year , t ( p ) = −2 . 4 ( 0 . 02 ) , Figure 4c ) in the independent test dataset . Likewise , PGS-BA was associated with brain age delta derived from the LASSO algorithm ( β = 0 . 53 [±0 . 09] year , t ( p ) = 10 . 4 ( <0 . 001 ) , r2 = 0 . 02 , Figure 4b ) but not to brain age delta change ( β = –0 . 001 [± . 02] year , t ( p ) = 0 . 0 ( 1 . 0 ) , Figure 4d ) . See Figure 4—figure supplement 1 for GWAS results . The association between PGS-BA and brain age delta remained significant when using as covariates the top 10 genetic components derived from the full UK Biobank sample ( p<0 . 001 in both analyses ) .
We used large training datasets to estimate the brain age models and the PGS scores leading to robust PGS-BA and brain age estimates . Self-reported birth weight ( Nilsen et al . , 2017 ) and cross-sectional brain age ( Franke and Gaser , 2012 ) are highly reliable measures; thus , our analyses are well-powered to detect small effects ( Zuo et al . , 2019 ) . The reliability of brain age deltalong is , however , unknown . Strictly speaking , brain age delta is a prediction error from a model that maximizes the prediction of age in cross-sectional data and thus partially also reflects noise . Given that deltalong is estimated as the difference between two deltacross estimates , it will hence have higher noise than the cross-sectional estimates , reducing the power in identifying potential associations between longitudinal and cross-sectional delta . This may be compounded by the relatively short interscan interval in the UK Biobank ( ≈2 years ) . However , our sample size ( n > 1200 ) ensures that the tests performed in this study are well-powered to detect small effects , even if deltalong has mediocre reliability ( Zuo et al . , 2019 ) . Further , replication of our null results in the Lifebrain sample with more observations and longer follow-up times reduces the likelihood of noise as the main factor behind the lack of relationship . Furthermore , previous studies have found that changes in brain age are partly heritable ( Brouwer et al . , 2021 ) and relate to , for instance , cardiometabolic risk factors ( Beck , 2021 ) , suggesting that it captures biologically relevant signals ( i . e . , has predictive validity ) , although with substantially different origins from cross-sectional brain age . Although the reliability of deltalong needs to be formally tested , the null relationship between deltacross and deltalong does not seem to be a result of a low-powered test . We speculate that our results partially generalize to other normative and residual-based modeling approaches , as well as to developmental samples . There is considerable evidence in the literature that birth weight and genetic risk for neurodegenerative conditions affect brain structure from early life ( Raznahan et al . , 2012; Walhovd et al . , 2020; Walhovd et al . , 2016 ) . Brain age models are related to other models such as normative brain charts ( Bethlehem , 2021; Dong et al . , 2020 ) – akin to normative anthropometric charts – the main difference being that brain age models predict , rather than control for , age ( Marquand et al . , 2019 ) . Both types of models produce normative brain scores , which are uncorrelated with age ( Butler et al . , 2021 ) . Thus , caution is required when interpreting these scores as indices of brain aging without availability of longitudinal data . Developmental samples may , however , reflect slightly stronger relationships between cross-sectional brain age delta and ongoing brain change as brain changes during early-life development typically occur at a faster pace than in middle or later life . Similarly , for specific disease groups such as Alzheimer’s disease patients ( Franke and Gaser , 2012 ) , interindividual brain variation in brain age might reflect to a greater extent prevailing loss of brain structure . Moreover , the variance associated with factors other than ongoing development/aging might be more limited in early than later age since influences leading to interindividual variations in brain structure have a shorter span to accumulate . That is , as time from birth increases , chronological age as a marker of individual development is reduced . Finally , many genetic and environmental factors relate to lifelong stable differences in brain age beyond birth weight and PGS-BA . However , both variables are congenital and show stable associations through the lifespan ( Raznahan et al . , 2012; Walhovd et al . , 2020 ) without strong evidence that they relate to brain change after adolescence . Thus , birth weight and PGS-BA are paradigmatic for showing how interindividual differences in brain age emerge early in life . The present study does not provide a systematic understanding of these influences but presents a framework for interpreting the impact such measures may exert on age-related phenotypes . The results call for caution in interpreting brain-derived indices of aging based on cross-sectional MRI data and underscore the need to rely on longitudinal data whenever the goal is to understand the trajectories of brain and cognition in aging .
The main sample was drawn from the UK Biobank neuroimaging branch ( https://www . ukbiobank . ac . uk/ Miller et al . , 2016 ) . 38 , 682 individuals had MRI available at a single time point and were used as the training dataset . 1372 individuals had longitudinal data and were used as the test dataset . The present analyses were conducted under data application number 32048 . The Lifebrain dataset ( Walhovd et al . , 2018 ) included datasets from five different major European Lifespan cohorts: the Center for Lifespan Changes in Brain and Cognition cohort ( LCBC , Oslo; Walhovd et al . , 2016 ) , the Cambridge Center for Aging and Neuroscience study ( Cam-CAN; Shafto et al . , 2014; Taylor et al . , 2017 ) , the Berlin Study of Aging-II ( Base-II; Bertram et al . , 2014 ) , the University of Barcelona cohort ( UB; Rajaram et al . , 2016; Vidal-Piñeiro et al . , 2014 ) , and the BETULA project ( Umeå; Nilsson et al . , 2010 ) . Furthermore , we included data from the Australian Imaging Biomarkers and Lifestyle flagship study of ageing ( AIBL; Ellis et al . , 2009 ) . In addition to cohort-specific inclusion and exclusion criteria , individuals aged <18 years , or with evidence of mild cognitive impairment , or Alzheimer’s disease were excluded from the analyses . 1792 individuals with only one available scan were used for the Lifebrain training dataset . 1500 individuals with available follow-up of >0 . 4 years were included in the test dataset . Individuals had between 2 and 8 available scans each . Sample demographics for the UK Biobank and the Lifebrain samples are provided in Supplementary file 3 . See also Figure 1c and Figure 1—figure supplement 1 for a visual representation of the age distribution in the UK Biobank and the Lifebrain datasets . UK Biobank ( North West Multi-Center Research Ethics Committee [MREC]; see also https://www . ukbiobank . ac . uk/the-ethics-and-governance-council ) and the different cohorts of the Lifebrain replication dataset ( Supplementary file 5 ) have ethical approval from the respective regional ethics committees . All participants provided informed consent . See https://biobank . ctsu . ox . ac . uk/crystal/crystal/docs/brain_mri . pdf for details on the UK Biobank T1w MRI acquisition . UK Biobank and Lifebrain MRI data were acquired with 3 and 10 different scanners , respectively . T1w MRI acquisition parameters for both the Lifebrain and the UK Biobank are summarized in Supplementary file 6 . We used summary regional and global metrics derived from T1w data . For UK Biobank , we used the imaging-derived phenotypes developed centrally by UK Biobank researchers ( Miller et al . , 2016 ) and distributed via the data showcase ( http://biobank . ctsu . ox . ac . uk/crystal/index . cgi ) . See preprocessing details in https://biobank . ctsu . ox . ac . uk/crystal/crystal/docs/brain_mri . pdf . This procedure yielded 365 structural MRI features , partitioned in 68 features of cortical thickness , area , and gray-white matter contrast , 66 features of cortical volume , 41 features of subcortical intensity , and 54 features of subcortical volume . See the list of features in Supplementary files 1 and 2 . Lifebrain data were processed on the Colossus processing cluster , University of Oslo . Similar to the UK Biobank pipeline , we used the fully automated longitudinal FreeSurfer v . 6 . 0 . pipeline ( Reuter et al . , 2012 ) for cortical reconstruction and subcortical segmentation of the structural T1w data ( http://surfer . nmr . mgh . harvard . edu/fswiki Dale et al . , 1999; Fischl et al . , 1999; Fischl and Dale , 2000 ) and used similar atlases for structural segmentation and feature extraction . We used birth weight ( kg ) from the UK Biobank ( field #20022 ) . Participants were asked to enter their birth weight at the initial assessment visit , the first repeat assessment visit , or the first imaging visit . In the case of multiple birth weight instances , we used the latest available input . n = 894 participants from the test dataset had available data on birth weight . The main analysis was constrained to normal variations in birth weight between 2 . 5 and 4 . 5 kg ( n = 770; Walhovd et al . , 2012 ) due to lower reliability of extreme scores and to tentatively remove participants potentially with severe medical complications associated with prematurity . Detailed information on genotyping , imputation , and quality control was published by Bycroft et al . , 2018 . For genetic analyses , we only included participants with both genotypes and MRI scans . Following the recommendations from the UK Biobank website , we excluded individuals with failed genotyping , who had abnormal heterozygosity status , or who withdrew their consents . We also removed participants who were genetically related – up to the third degree – to at least another participant as estimated by the kinship coefficients as implemented in PLINK ( Chang et al . , 2015 ) . For the GWAS we used 38 , 163 individuals from the training dataset . Polygenic risk scores were computed using the test dataset consisting of 1339 individuals with longitudinal MRI . We performed GWAS analysis on the training dataset and the brain age delta-semi-corrected phenotype using the imputed UK Biobank genotypes . To control for subtle effects of population stratification in the dataset , we computed the top 10 PCs using the PLINK command –pca on a decorrelated set of autosome single-nucleotide polymorphisms ( SNPs ) . The set of SNPs ( n = 101 , 797 ) were generated by using the PLINK command , --maf 0 . 05 , --hwe 1e–6 , --indep-pairwise 100 50 0 . 1 . The –glm function from PLINK was used to perform GWAS on about 9 million autosomal SNPs , including age , sex , and the top 10 PCs as covariates . See Manhattan and quantile-quantile ( QQ ) plots in Figure 4—figure supplement 1 . Note that our results corroborated the same association region reported in Jonsson et al . , 2019 with a smaller sample . The GWAS results for the training dataset were used to compute PGS ( PGS-BA ) in the independent test dataset ( n = 1339 participants ) . We used the recently developed method PRS-CS ( Ge et al . , 2019 ) to estimate the posterior effect sizes of SNPs that were shown to have high quality in the HapMap data ( International HapMap 3 Consortium et al . , 2010 ) . Rather than estimating the polygenicity of brain age delta from our data , we assumed a highly polygenic architecture for brain age delta by setting the parameter --phi = 0 . 01 ( Boyle et al . , 2017 ) . The remaining parameters of PRS-CS were set to the default values . PGS was based on 654 , 725 SNPs and was computed on the independent test data using the --score function from PLINK . SNPs were aligned with HapMap 3 SNPs ( autosome only as provided by PRC-CS ) and posterior effects were estimated . We also computed the population structures PCs’ in the test dataset using the same procedure as in the training dataset . All statistical analyses were run with R version 3 . 6 . 3 https://www . r-project . org/ . We used the UK Biobank as the main sample and the Lifebrain cohort for independent replication . The main description refers to the UK Biobank pipeline , though Lifebrain replication followed identical steps unless otherwise stated . For replication across machine learning pipelines , we used a LASSO regression approach for age prediction , adapted from ( Cole , 2020b ) . See more details in Cole , 2020a . The correlation between LASSO-based and Gradient Boosting-based brain age deltas was 0 . 80 . We used machine learning to estimate each individuals’ brain age based on a set of regional and global features extracted from T1w sequences . We estimated brain age using gradient tree boosting ( https://xgboost . readthedocs . io ) . We used participants with only one MRI scan for the training dataset ( n = 36 , 682 ) and participants with longitudinal data as test dataset ( n = 1372 ) . All variables were scaled prior to any analyses using the training dataset metrics as reference . The model was optimized in the training set using a 10-fold cross-validation randomized hyperparameter search ( 50 iterations ) . The hyperparameters explored were number of estimators [seq ( 100:600 , by = 50 ) ] , learning rate ( 0 . 01 , 0 . 05 , 0 . 1 , 0 . 15 , 0 . 2 ) , maximum depth [seq ( 2:8 , by = 1 ) ] , gamma regularization parameter [seq ( 0 . 5:1 . 5 , by = 0 . 5 ) ] , and min child weight [seq ( 1:4 , by = 1 ) ] . The remaining parameters were left to default . The optimal parameters were number of estimators = 500 , learning rate = 0 . 1 , maximum depth = 5 , gamma = 1 , and min child weight = 4 predicting r2 = 0 . 68 variance in chronological age with MAE = 3 . 41 and RMSE = 4 . 29 . See visual representation in Figure 1f . Next , we recomputed the machine learning model using the entire training dataset and the optimal hyperparameters and used it to predict brain age for the test dataset ( Figure 1e ) . These metrics are similar or better than other brain age models using UK Biobank MRI data ( Cole , 2020a; de Lange et al . , 2019 ) and the cross-validation diagnostics . We used GAM to correct for the brain age bias estimation ( Smith et al . , 2019 ) ; r = –0 . 54 for the test dataset . Note that we used GAM fittings as estimated in the training dataset so delta values in the test dataset are not centered to 0 . Brain age delta was estimated as the GAM residual . The correlation between brain age delta corrected based on the training vs . the test fit was r > 0 . 99 . Also , GAM-based bias correction led to similar brain age delta estimations to linear and quadratic-based corrections ( r > 0 . 99 ) . The diagnostics for the LASSO-based model were as follows: variance explained ( r2 ) = 0 . 69/0 . 69; MAE = 3 . 36/3 . 28; RMSE = 4 . 21/4 . 04; age bias = –0 . 56/–0 . 52 for the training and predicted datasets . See representation of the brain age prediction in Figure 2—figure supplement 2 . The raw data were gathered from the UK Biobank , the Lifebrain cohort , and the AIBL . Raw data requests are specific to each cohort . UK Biobank and AIBL data are available upon application to UK Biobank and at https://aibl . csiro . au upon corresponding approvals . For the Lifebrain cohorts , requests for raw MRI data should be submitted to the corresponding principal investigator . See contact details in Supplementary file 5 . MRI data is not openly available as participants did not consent to share publicly their data . Access to data is available upon reasonable requests and transfer agreements . Different sample agreements are required for each dataset . Statistical analyses in this article are available alongside the article and will be available at https://github . com/LCBC-UiO/VidalPineiro_BrainAge . All analyses were performed in R 3 . 6 . 3 . The scripts were run on the Colossus processing cluster , University of Oslo . UK Biobanks’ data acquisition , MRI preprocessing , and feature generation pipelines are freely available ( https://www . fmrib . ox . ac . uk/ukbiobank ) . For the Lifebrain cohorts , the image acquisition details are summarized in Supplementary file 6 . MRI preprocessing and feature generation scripts were performed with the freely available FreeSurfer software ( https://surfer . nmr . mgh . harvard . edu/ ) . For bash-sourcing scripts , please contact the corresponding author . | Scientists who study the brain and aging are keen to find an effective way to measure brain health , which could help identify people at risk for dementia or memory problems . One popular marker is ‘brain age’ . This measurement uses a brain scan to estimate a person’s chronological age , then compares the estimated brain age to the person’s actual age to determine whether their brain is aging faster or slower than expected for their age . However , since brain age relies on one brain scan taken at one point in time , it is not clear whether it really measures brain aging or if it might capture brain differences that have been present throughout the individual’s life . Studies comparing individual brain scans over several years would be necessary to know for sure . Now , Vidal-Piñeiro et al . show that the brain-age measurement does not reflect faster brain aging . In the experiments , the researchers compared repeated brain scans of thousands of individuals over 40 years of age . The experiments showed that deviations from normative brain age detected in a single scan reflected early life differences more than changes in the brain over time . For example , people with older-looking brains were more likely to have had a low birth weight or to have a combination of genes associated with having an older looking brain . Vidal-Piñeiro et al . show that brain age mostly reflects a pre-existing brain condition rather than brain aging . The experiments also suggest that genetics and early brain development likely have a strong impact on brain health throughout life . Future studies trying to test or develop brain-aging measurements should use serial measurements to track brain changes over time . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"neuroscience"
] | 2021 | Individual variations in ‘brain age’ relate to early-life factors more than to longitudinal brain change |
Ribosome biogenesis is a complex and energy-demanding process requiring tight coordination of ribosomal RNA ( rRNA ) and ribosomal protein ( RP ) production . Given the extremely high level of RP synthesis in rapidly growing cells , alteration of any step in the ribosome assembly process may impact growth by leading to proteotoxic stress . Although the transcription factor Hsf1 has emerged as a central regulator of proteostasis , how its activity is coordinated with ribosome biogenesis is unknown . Here , we show that arrest of ribosome biogenesis in the budding yeast Saccharomyces cerevisiae triggers rapid activation of a highly specific stress pathway that coordinately upregulates Hsf1 target genes and downregulates RP genes . Activation of Hsf1 target genes requires neo-synthesis of RPs , which accumulate in an insoluble fraction and presumably titrate a negative regulator of Hsf1 , the Hsp70 chaperone . RP aggregation is also coincident with that of the RP gene activator Ifh1 , a transcription factor that is rapidly released from RP gene promoters . Our data support a model in which the levels of newly synthetized RPs , imported into the nucleus but not yet assembled into ribosomes , work to continuously balance Hsf1 and Ifh1 activity , thus guarding against proteotoxic stress during ribosome assembly .
Ribosome assembly is the most energy demanding process linked to cell growth and requires coordinated production of processed ribosomal RNAs ( rRNAs ) , ribosomal proteins ( RPs ) and ribosome biogenesis ( RiBi ) factors . This massive biosynthetic program permits rapidly growing yeast cells to produce about 2000 ribosomes per minute ( Warner , 1999 ) , which is critical for sustaining high rates of growth ( mass accumulation ) and proliferation . At the same time , though , ribosome assembly poses a constant threat to cellular protein homeostasis and continued growth , since it requires the coordinated and large-scale assembly of four rRNAs with 79 different RPs , the latter of which are known to be highly prone to aggregation ( David et al . , 2010; Pillet et al . , 2017; Rand and Grant , 2006; Weids et al . , 2016 ) . Indeed , unassembled RPs in metazoans have long been known to activate p53 , through titration of its negative regulator MDM2 , and conserved p53-independent pathways that respond to perturbations in ribosome assembly are now beginning to emerge ( James et al . , 2014 ) . Given the absence of p53 in yeasts , Saccharomyces cerevisiae is a promising model system in which to uncover ancestral processes that might monitor ribosome assembly to regulate growth and protein homeostasis in eukaryotes . Heat shock factor 1 ( Hsf1 ) is a central actor in Protein Quality Control ( PQC ) and protein homeostasis ( proteostasis ) in eukaryotes , in both stressed and unstressed cell , and in pathological situations ( Li et al . , 2017 ) . Notably , Hsf1 is a direct modulator of tumorigenesis and becomes essential , as it is in budding yeast ( Solís et al . , 2016 ) , to support growth of malignant cells ( Santagata et al . , 2013 ) . Hsf1 prevents protein aggregation and proteome imbalance by driving the expression of a small regulon including genes encoding essential chaperones ( Hsp70/Hsp90 ) , nuclear/cytoplasmic aggregases , and proteasome components ( Solís et al . , 2016; Mahat et al . , 2016; Pincus et al . , 2018 ) . Interestingly , studies in budding yeast reveal that the Ribosome Quality Control complex ( RQC ) , conserved from yeast to human ( Brandman et al . , 2012 ) , increases Hsf1 activity under conditions of translation stress . However , many essential aspects of Hsf1 regulation remain to be elucidated , in particular whether its transcriptional activity is linked to ribosome biogenesis itself . Recently , a conserved PQC mechanism referred to as Excess Ribosomal Protein Quality Control ( ERISQ ) was described that specifically recognizes unassembled RPs in the nucleus and targets them for proteasome degradation ( Sung et al . , 2016a; Sung et al . , 2016b ) , thus illuminating observations made 40 years ago showing that excess RPs are rapidly degraded ( Gorenstein and Warner , 1977; Warner , 1977 ) . Sung and colleagues showed that the ubiquitin ligase Tom1 plays an important role in ESRIQ by preventing the accumulation of detergent-insoluble RPs . The potential role of Hsf1 in ERISQ has not yet been explored . Given the tremendous investment of cellular resources involved in ribosome production ( Warner , 1999 ) and the fact that a decrease of ribosome abundance protects cells against proteotoxic stress ( Guerra-Moreno et al . , 2015; Mills and Green , 2017 ) , it might be expected that cells have evolved mechanisms to rapidly decrease RP gene transcription in the face of defects in ribosome assembly , in order to both save energy and reestablish cellular proteostasis . In S . cerevisiae , RP gene transcription is known to be tightly regulated according to growth conditions through the stress-sensitive transcription factor ( TF ) Ifh1 . Thus , Ifh1 is rapidly released from RP promoters only minutes following inhibition of the conserved eukaryotic growth regulator Target Of Rapamycin Complex 1 ( TORC1 ) kinase ( Schawalder et al . , 2004 ) . Although it has been shown that Ifh1 promoter binding is coordinated with RNA polymerase I ( RNAPI ) activity upon prolonged TORC1 inhibition to help balance RP and rRNA production ( Albert et al . , 2016; Rudra et al . , 2007 ) , how Ifh1 is removed from RP gene promoters to immediately downregulate their expression following stress remains a mystery . Furthermore , possible links between RP gene expression , ribosomal assembly and the protein homeostasis transcription program driven by Hsf1 remain important open questions . In this study , we uncover a novel regulatory pathway , hereafter referred to as the Ribosomal Assembly Stress Response ( RASTR ) , that allows rapid and simultaneous up-regulation of protein homeostasis genes and downregulation of RP genes following disruption of various steps in ribosome biogenesis ( rRNA production , processing or RP assembly ) . We show that RASTR is highly specific to the RP and Hsf1 regulons , with little or no effect on a much larger group of genes implicated in the Environmental Stress Response ( ESR ) . Importantly , RASTR requires neo-synthesis of RPs following stress and is linked to the accumulation of RP aggregates , which we propose lead to Hsf1 activation , through chaperone competition , and to the sequestration of Ifh1 in an insoluble nucleolar fraction . Notably , we show that protein synthesis inhabitation via cycloheximide treatment leads to a transcriptional response opposite to that of RASTR , supporting a model in which unstressed cells constantly monitor nuclear levels of unassembled RPs and use this information to balance expression of Hsf1 target genes with those encoding RPs . Finally , we demonstrate that RASTR is the initial transcriptional response to inactivation of TORC1 kinase , supporting a key role for this regulatory pathway in the activation of a protein homeostasis transcriptional program that allows cells to cope with the proteotoxic consequences of disruptions to ribosome biogenesis .
In an effort to better understand the role of the two major eukaryotic DNA topoisomerases in protein-coding gene transcription , we generated yeast strains in which Top1 , Top2 , or both of these enzymes are rapidly degraded by the auxin-induced degron ( AID ) method ( Nishimura et al . , 2009 ) and confirmed by western blotting that significant depletion of either protein was obtained between 10 and 20 min following auxin addition to the medium , and that Top2 depletion , as expected , prevents cell growth ( Figure 1—figure supplement 1A , B ) . We then performed ChIP-seq analysis of RNA polymerase II ( RNAPII ) in the Top1-AID , Top2-AID and Top1/2-AID strains at 20 and 60 min following auxin addition ( Figure 1A–C , Supplementary file 1 ) . As expected ( Brill et al . , 1987; Brill and Sternglanz , 1988 ) , the absence of Top2 had little or no effect on RNAPII distribution ( Figure 1B ) . However , Top1 depletion triggered a rapid response at two specific groups of genes: upregulation of Hsf1 target genes and downregulation of RP genes ( Figure 1A , D; Figure 1—figure supplement 1C , D ) . Remarkably , this response was transient , as both groups of genes returned to normal levels ( i . e . before auxin addition ) by 60 min . This re-equilibration was dependent upon Top2 since it failed to occur in the Top1/2-AID strain , where prolonged auxin treatment led to significant dysregulation of many other RNAPII-transcribed genes ( Figure 1C ) . Since upregulation of proteostasis-related genes and downregulation of RP genes are characteristic of many different stress responses , we decided to quantify the effect of Top1 depletion on transcription all gene groups that have been classified as part of the general ‘Environmental Stress Response’ ( ESR; Gasch et al . , 2000 ) , which include an additional group of stress-induced genes regulated by the Msn2/4 TFs and a large suite of genes involved in ribosome biogenesis ( RiBi genes ) . This analysis shows clearly that Top1 depletion , as well as depletion of both Top1 and Top2 ( at the early 20 min time point ) , triggers a highly specific stress response linked to RP genes and Hsf1 target genes ( Figure 1A–C ) . Such a targeted response is unlikely to result from a global topological effect on RNAPII recruitment but would instead appear to be the consequence of the activation of a specific signaling pathway that is more restricted in nature than the ESR . To explore the target ( s ) of this hypothetical signaling pathway at RP genes , we monitored by qPCR ChIP the promoter association of three TFs ( Rap1 , Fhl1 and Ifh1 ) that operate at the majority ( >90% ) of these 138 genes ( Knight et al . , 2014 ) . Interestingly , we found that the activator Ifh1 is rapidly released from RP gene promoters after topoisomerases depletion ( Figure 1E ) , whereas Rap1 and Fhl1 , which bind directly to RP promoter DNA , are not affected ( Figure 1—figure supplement 1E–F ) . To confirm that Hsf1 is indeed required for upregulation of genes following Top1 depletion , we used the anchor-away technique ( Haruki et al . , 2008 ) to rapidly remove Hsf1 from the nucleus ( 30 min treatment with rapamycin; Solís et al . , 2016 ) before initiating Top1-AID degradation by auxin addition ( Figure 1F ) . Efficient nuclear depletion of Hsf1 was confirmed by the inability of the Hsf1-FRB , Top1-AID strain to form colonies in the presence of rapamycin ( Figure 1—figure supplement 1G ) . Note that the strain used in this and all other anchor-away experiments contains the TOR1-1 mutation and is thus resistant to the normal physiological consequences of rapamycin treatment , which inactivates the growth-promoting TORC1 kinase ( Heitman et al . , 1991; Loewith and Hall , 2011 ) . This experiment revealed that Hsf1 nuclear depletion completely abolishes activation of stress genes following Top1 depletion without affecting down-regulation of RP genes ( Figure 1G ) . Therefore , activation of stress-induced genes following Top1 depletion is completely Hsf1-dependent , whereas repression of RP genes is independent of Hsf1 or the induction of its target genes . We would also note that the stress pathway induced by Top1 depletion is unusually restricted in comparison to many other stress responses that are often grouped together as the Environmental Stress Response ( ESR; Gasch et al . , 2000 ) , since Msn2/4 target genes are not induced and RiBi genes are not downregulated ( Figure 1A , G ) . Although it may seem surprising that depletion of topoisomerases can induce a Hsf1-dependent stress response , formation of distinct nuclear foci by the Btn2 aggregase and perinuclear accumulation of the proteasome subunit Pre6 following Top1/2 degradation ( Figure 2—figure supplement 1A , B ) both point to the induction of proteotoxic stress in the nucleus ( Miller et al . , 2015a ) . Top1 was initially identified through a mutation ( mak1 ) defective in large ribosomal subunit production ( Thrash et al . , 1985 ) and was later shown to be required for proper rRNA synthesis ( Brill et al . , 1987; El Hage et al . , 2010; French et al . , 2011 ) . Consistent with these findings , we observed a strong reduction of pre-rRNA synthesis as shown by decreased [3H]-adenine pulse-labeling of the RNAPI-transcribed 35S pre-rRNA , and two co-transcriptionally cleaved products , 27S and 20S pre-rRNAs , as early as 10 min after initiation of Top1 ( or Top1 and Top2 ) depletion by addition of auxin to the medium ( Figure 2—figure supplement 1C ) . This decreased rRNA synthesis is accompanied by an elongation defect , as shown by the accumulation of truncated pre-rRNAs that were initially described by the Tollervey laboratory ( El Hage et al . , 2010; Figure 2A ) . Further downstream , the rapid defect in rRNA production caused by inhibition of RNAPI elongation leads to unbalanced production of 40S and 60S ribosomal subunits , with a marked deficiency of the large ( 60S ) subunit relative to the small ( 40S ) subunit ( Figure 2B ) . This would be expected to create a disequilibrium between RP and rRNA production , and more specifically an excess of unassembled RPs . Consistent with this , we detect accumulation of both and large and small subunit proteins ( Rpl3 and Rps8 , respectively ) in trailing fractions of polysome gradients ( Figure 2C ) . These observations strongly suggest that RPs fail to be incorporated normally into ribosomes immediately following topoisomerase degradation . In addition , this sedimentation profile may also reflect the presence of disassembling or incompletely assembled pre-60S particles . RPs are known to be prone to aggregation ( David et al . , 2010; Pillet et al . , 2017; Rand and Grant , 2006; Weids et al . , 2016 ) and recent reports show that newly synthetized , unassembled RPs accumulate in aggregates in response to ribosome assembly stress ( Sung et al . , 2016a; Sung et al . , 2016b ) . Significantly , we also observed accumulation of RPs in an insoluble fraction following topoisomerase degradation ( Figure 2D ) . The observations described above led us to hypothesize that the transcriptional response to Top1 degradation is a consequence of defective ribosome assembly , perhaps driven by the proteotoxic stress caused by the accumulation of unassembled RPs . To challenge this idea , we measured the transcriptional response to three different perturbations to ribosome biogenesis: depletion of two essential ribosome assembly factors ( Utp8 and Utp13 ) and treatment of cells with diazaborine . Utp8p is a member of the t-UTP subcomplex of 90S pre-ribosomes and its depletion inhibits rDNA transcription , leading to a reduction of the primary 35S pre-rRNA transcript and subsequent processing intermediates ( Gallagher et al . , 2004 ) . In contrast , depletion of Utp13 ( a member of the UTP-B subcomplex ) interferes with downstream processing and synthesis of 40S subunits and causes decreased 18S rRNA levels without affecting the levels of the 25S or 5 . 8S rRNAs ( Gallagher et al . , 2004 ) . Diazaborine , an inhibitor of the essential Drg1 AAA-ATPase , rapidly blocks mid-late steps of 60S subunit maturation ( Loibl et al . , 2014 ) . Remarkably , all these treatments triggered a similar transcriptional response to that which occurs following Top1 depletion , namely a specific downregulation of RP genes and upregulation of Hsf1 target genes ( Figure 2E–G; Supplementary file 2 ) , which we refer to as the ‘Ribosome Assembly STress Response’ ( RASTR ) . Hsf1 activity is stimulated by many different types of cellular stress , including stalled ribosomes . A pioneering study reported that a set of proteins termed the RQC binds to 60S ribosomal subunits containing stalled polypeptides and leads to their degradation . In the process , the RQC triggers a specific stress signal that leads to Hsf1 target gene activation ( Brandman et al . , 2012 ) . Thus , cells lacking a component of the RQC , the Tae2 protein , fail to activate Hsf1 following translational stress . To ask if RASTR might be related to the RQC , we induced Top1/2 degradation in tae2-Δ cells . We found that activation of two Hsf1 target genes ( SSA1 and HSP42 ) and downregulation of two RP genes ( RPL30 and RPL39 ) was unaffected by deletion of TAE2 ( Figure 2—figure supplement 1D ) and conclude that RQC does not play a role in RASTR . These results highlight that cells have developed distinct mechanisms to adapt the Hsf1 transcriptional program to defects in both ribosome activity and ribosome assembly . Although many studies would support the notion that Hsf1 activation during RASTR occurs through sequestration of its inhibitory partner Hsp70 by RP aggregates ( Krakowiak et al . , 2018; Shi et al . , 1998; Zheng et al . , 2016 ) , it is less clear how ribosome assembly stress could trigger release of Ifh1 from RP gene promoters . We reported previously that the association of Ifh1 with RP gene promoters in growing cells is rapidly disrupted ( within 5 min ) following inhibition of the growth-promoting TORC1 kinase by addition of rapamycin to the medium ( Schawalder et al . , 2004 ) . More recently ( Albert et al . , 2016 ) , we found that stable release of Ifh1 from RP gene promoters ( measured 20 min after rapamycin addition ) requires its C-terminal domain together with a complex of proteins containing casein kinase 2 ( CK2 ) and two RiBi factors , Utp22 and Rrp7 , with which Ifh1 interacts to form the CURI complex ( Rudra et al . , 2007 ) . Thus , in ifh1-ΔC cells the truncated protein is rapidly released but later returns to RP gene promoters following TORC1 inhibition . This led us to propose two distinct mechanisms controlling the promoter release of Ifh1 following stress: one operating at a short timescale ( <5 min ) and the other on a long timescale ( ~20 min ) . Interestingly , ifh1-ΔC promoter release is stable following Top1 depletion , suggesting that an unknown mechanism regulates Ifh1 during RASTR ( Figure 3A ) . The fact that RP gene repression and Hsf1 target gene activation occur with identical kinetics following RASTR activation ( Figure 3B ) , and that Ifh1 concentrates in nuclear foci rapidly after topoisomerase depletion ( Figure 3C ) , suggests that Ifh1 could be sensitive to the accumulation of unassembled RPs in the nucleus , as is presumably the case for Hsf1 . Several lines of evidence are consistent with this hypothesis . To begin with , in cells lacking Tom1 , an E3 ligase required for degradation of unassembled RPs ( Sung et al . , 2016a; Sung et al . , 2016b ) , but not in TOM1 cells , Ifh1 accumulates in prominent nuclear foci even in the absence of stress ( Figure 3D ) . This suggests that Ifh1 aggregates in cells that are unable to efficiently degrade excess RPs , even under optimal growth conditions . Consistent with this , the published mass spectrometry data of insoluble fractions from cells either treated with the proteasome inhibitor bortezomib or lacking Tom1 clearly identified Ifh1 , together with RPs , RiBi proteins and two Hsp70 proteins , Ssa1 and Ssa2 , inhibitory partners of Hsf1 ( Figure 3—figure supplement 1A ) . These data indicate that Ifh1 could be trapped in an insoluble cellular fraction in the absence of Tom1 and thus decrease the pool of Ifh1 able to bind with RP gene promoters . To test this possibility , we combined deletion of TOM1 with a mutant allele of IFH1 ( ifh1-AA ) that weakens its interaction with RP gene promoters . Remarkably , tom1-Δ is synthetically lethal with ifh1-AA ( Figure 3E ) supporting the notion that RP aggregation could directly impact on Ifh1 promoter binding . Lastly , to exclude the possibility that the genetic interaction between TOM1 and the mutated allele of IFH1 could be linked to the growth defect of this mutation , we examined another mutated allele of IFH1 ( ifh1-6 ) that triggers a similar growth defect ( Figure 3E ) . Importantly , we showed in a previous study that this ifh1-6 mutant protein remains bound at high levels to RP genes promoter even under stress conditions ( Albert et al . , 2016 ) . Remarkably , tom1-Δ is not synthetically lethal with ifh1-6 ( Figure 3E ) , supporting the notion that genetic interaction with ifh1-AA is directly linked to the ability of Ifh1 to bind RP gene promoters . To assess directly whether Ifh1 is sequestered in aggregates during RASTR , we analyzed by mass spectrometry the insoluble fraction following topoisomerase depletion . As previously reported for tom1-Δ cells ( Sung et al . , 2016a ) , the insoluble fraction is enriched in chaperones and RPs ( Figure 3F , G ) . We also noted a strong increase in RiBi factors , primarily those implicated in biogenesis of the large ribosomal subunit ( Figure 3G ) . Importantly , Ifh1 was never detected in an insoluble fraction in the absence of stress but was invariably detected in these fractions following topoisomerase depletion ( Supplementary file 3 ) . This rapid sequestration of Ifh1 may be sufficient to explain the observed downregulation of RP genes during RASTR . Given their fast turnover rate , nuclear accumulation in the absence of ribosome assembly and propensity to aggregate , unassembled RPs could be ideally positioned to rapidly signal ribosome biogenesis defects ( Sung et al . , 2016a; Milkereit et al . , 2001 ) . To evaluate the importance of newly synthetized RPs in RASTR , we blocked their production by cytoplasmic anchoring of Ifh1 before topoisomerase depletion ( Figure 4A ) . It is important to note that Ifh1 binding is highly specific to RP genes ( Knight et al . , 2014 ) and that the transcriptional effect of its nuclear depletion is restricted to RP genes and a very small number of additional targets ( Supplementary file 4 ) . Although Ifh1 depletion by anchor-away may not be complete ( the strain still grows on plates containing rapamycin , albeit slowly , even though Ifh1 is essential for growth; Figure 4—figure supplement 1A ) , it nonetheless leads to a significant and highly specific decrease in RP gene transcription as measured by RNAPII ChIP-seq ( Figure 4B; Supplementary file 4 ) . Interestingly , Ifh1 depletion also leads to aberrant rRNA processing ( Figure 4C ) as would expected in conditions where RP levels become limiting ( Reiter et al . , 2011 ) . Remarkably , we noted that 60 min of Ifh1 anchoring alone , in the absence of topoisomerase depletion , also caused a significant down-regulation of Hsf1 target genes ( Figure 4D; Supplementary file 4 ) even though Ifh1 is absent from the promoters of these genes ( Figure 4—figure supplement 1B ) , suggesting that the Hsf1 transcriptional program is continuously influenced by RP production . Consistent with this idea , we found that upregulation of Hsf1 target genes was either abolished or strongly reduced ( Figure 4E , F , G; Supplementary file 4 ) when Top1 or Top1 and Top2 were degraded following nuclear depletion of Ifh1 , indicating that RP production is required for Hsf1 target gene activation during RASTR . As an alternative approach to test the requirement for de novo RP synthesis to initiate RASTR we used cycloheximide treatment , which blocks all translational elongations , thus leading to rapid depletion of the nuclear pools of RPs ( Figure 5A; Gorenstein and Warner , 1977; Warner , 1977; Reiter et al . , 2011; Lam et al . , 2007 ) . As reported by others ( Reiter et al . , 2011 ) , we confirmed that cycloheximide alone also triggers a rapid arrest of rRNA processing ( Figure 5B ) . Quite strikingly , we observed a transcriptional response to cycloheximide treatment exactly opposite to that induced by RASTR , namely Hsf1 target gene downregulation and RP gene upregulation ( Figure 5C; Supplementary file 5 ) . This finding suggests that even in unstressed cells RP production may contribute to a basal level of Hsf1 activation while at the same time limiting Ifh1 activity at RP gene promoters . Significantly , treatment of cells undergoing Top1 or Top1/2 depletion with cycloheximide ( auxin + CHX ) completely abolished both RP gene repression and activation of Hsf1 target genes ( Figure 5D , E; Supplementary file 5 ) . Indeed , Hsf1 target gene activation under these conditions is lower than in untreated cells and not significantly different than that seen in cells treated with cycloheximide alone ( Figure 5—figure supplement 1A ) . These findings clearly demonstrate that RASTR is dependent upon de novo protein synthesis . Importantly , it was recently reported that cycloheximide treatment efficiently prevents aggregation of newly synthesized RPs following proteasome inhibition ( Sung et al . , 2016a ) . Perhaps as a direct consequence of this , we found that CHX treatment also leads to strong reduction of Ifh1-eGFP nuclear foci that are observed in cells lacking the ubiquitin ligase Tom1 , which is specifically required for efficient degradation of unassembled RPs ( Sung et al . , 2016a; Sung et al . , 2016b; Figure 5—figure supplement 1B ) . Furthermore , Ifh1 dis-aggregation following cycloheximide exposure is associated with increased Ifh1 binding at a RP gene promoter , which becomes significant in tom1-Δ cells ( Figure 5—figure supplement 1C ) . Similarly , cycloheximide treatment also prevents the release of Ifh1 from RP gene promoters in response to the activation of RASTR by topoisomerase depletion ( Figure 5—figure supplement 1D ) . Considered as a whole , these data suggest that both RP and Ifh1 subnuclear structures ( aggregates ) are dynamic , promoted by de novo RP production upon RASTR initiation , and capable of influencing Ifh1 promoter binding . Consistent with this view , we observed a large increase in cells that accumulate RP ( Rpl25 ) or Ifh1 nuclear aggregates during RASTR that is abolished in the presence of cycloheximide ( Figure 5F , G; Figure 5—figure supplement 1E , F ) . Taken together with the strong reduction of Hsf1 target gene activation following Ifh1 cytoplasmic anchoring , both in the presence and absence of topoisomerase degradation , our observations on the effect of cycloheximide highlight the interwoven nature of RP and Hsf1 target gene regulons and support the notion that unassembled , aggregated RPs constitute the primary RASTR-induced signal capable of regulating both Ifh1 and Hsf1 activities , albeit in an opposite direction . More generally , these data indicate that newly synthetized RPs , in both stressed and unstressed cells , operate as a central hub in coordinating the expression of RP genes themselves with the Hsf1-dependent activation of chaperone and proteasome genes . We next turned our attention to the potential involvement of RASTR during more general stress responses that might also rapidly affect ribosome assembly . In an initial set of experiments , we inactivated the conserved growth-promoting TORC1 kinase by treatment of cells with rapamycin , which is known to mimic a major part of the environmental stress response , including osmotic and redox stress , as well as carbon , nitrogen , phosphate or amino acid starvation ( Loewith and Hall , 2011 ) . As reported previously , rapamycin triggers a rapid arrest of rRNA processing ( Figure 6A ) and a decrease of RP and RiBi gene expression ( Figure 6B , C; Supplementary file 6 ) . Interestingly , we noted that Hsf1 target genes are transiently up- and downregulated at 5 and 20 min , respectively , following rapamycin addition ( Figure 6B , C; Supplementary file 6 ) , suggesting that RASTR is activated at the early time point but shortly thereafter turned off . Consistent with this view , it has been reported that RP production ceases around 15 min after rapamycin treatment ( Reiter et al . , 2011 ) , which we suggest would turn off the signal for RASTR , thus explaining the downregulation of Hsf1 target genes observed at 20 min . To explore this hypothesis further , we took advantage of our observation that cycloheximide treatment prevents RASTR activation by either Top1 or Top1/2 degradation ( Figure 5 ) and treated cells with cycloheximide 5 min before rapamycin addition of either 5 or 20 min ( Figure 6D , see schematics of experimental protocols below the respective panels ) . Remarkably , this specifically prevented RP gene repression and Hsf1 target gene activation at 5 min following rapamycin addition , whereas at the longer time point ( 20 min ) RP genes and Hsf1 were regulated independently of cycloheximide ( Figure 6D , E; Supplementary file 6 ) . Consistent with the early block in RP gene downregulation being due to a failure to initiate RASTR immediately following rapamycin addition , we showed that cycloheximide pre-treatment prevents release of Ifh1 from RP gene promoters at 5 min , but not at 20 min following rapamycin treatment ( Figure 6F ) . The effects of rapamycin treatment described above are fully consistent with our previous report demonstrating that regulation of RP gene transcription following TORC1 inactivation by rapamycin operates through two distinct mechanisms at short and long timescales , with the latter dependent on RNAPI activity and the CURI complex ( Albert et al . , 2016 ) . The short timescale mechanism described here , which is dependent upon continued protein synthesis and presumably mediated by RASTR , allows cells to rapidly arrest RP production and avoid or minimize proteotoxic stress induced by arrested ribosome assembly . The second mechanism permits the resumption of RP production only when rRNA synthesis also resumes ( Albert et al . , 2016 ) . These two mechanisms could be particularly useful to rapidly adapt ribosome production to new growth conditions . To explore the possible generality of rapid RASTR-mediated shut-down of RP gene transcription in response to stress , we measured the transcriptional response to heat shock , which is known to transiently downregulate both RP gene and rRNA transcription ( Gasch et al . , 2000; Kos-Braun et al . , 2017 ) . As expected , we observed strong downregulation of RP genes and upregulation of Hsf1 target genes only 5 min following a shift in temperature to 40°C ( Figure 7A–C , left panels ) that was also accompanied by strong upregulation of Msn2 target genes and downregulation of RiBi genes ( Figure 7A , left panel ) . The heat-shock transcriptional response is thus much broader than that to ribosome assembly stress , despite their common effects on RP and Hsf1 target gene expression . Remarkably , though , we found that cycloheximide pre-treatment prevents the strong and immediate repression of RP genes following heat shock ( Figure 7A , B; right panels; Supplementary file 6 ) , consistent again with the idea that this facet of the heat-shock response is identical to that which occurs during RASTR . Importantly though , Hsf1 target genes are still activated following heat shock in the presence of cycloheximide ( Figure 7A , C; right panels ) , presumably because the unfolding of thermo-labile proteins induced by heat shock is alone sufficient to activate Hsf1 even in the absence of continuing RP synthesis . Nevertheless , the striking requirement for de novo protein synthesis for RP gene downregulation following heat shock strongly suggests that RASTR plays an integral role in this component of the heat-shock response and thus may constitute the earliest transcriptional response , at the level of RNAPII , to a wide variety of stress conditions .
In this study , we demonstrate the existence of a regulatory mechanism , which we refer to as the ribosome assembly stress response , or RASTR , that allows yeast cells to specifically coordinate the activity of two TFs , Hsf1 and Ifh1 , with the functional state of ribosome assembly . Our data and several previous reports suggest that rapid ribosome biogenesis is a potentially proteotoxic process , in large part due to accumulation of unassembled RPs , whose production needs to be carefully coordinated at the transcriptional level , at least in yeast , together with that of chaperones and proteasome components ( Figure 8 ) . In this perspective , RASTR may play an essential role in minimizing the proteostasis burden imposed by high ribosome production rates , particularly under fluctuating environmental conditions . Our data indicate that disequilibrium at any step-in ribosome biogenesis ( rRNA transcription , early or late rRNA processing or assembly ) will lead to RASTR activation until the pool of unassembled RPs decreases . Consequently , proteasome inhibition ( Kos-Braun et al . , 2017 ) , dNTP depletion ( Gómez-Herreros et al . , 2013 ) , DNA damage ( Conconi et al . , 2005 ) , nutrient and thermal stress ( Liu et al . , 1996; Tsang et al . , 2003 ) , all of which are known to alter rRNA transcription or processing , or ribosome assembly ( reviewed in James et al . , 2014 ) , are likely to activate RASTR , as do the genetic perturbations at different step of ribosome assembly described here . Taken together with previous reports , our data thus point to a critical role for RASTR in the transcriptional networks regulating both growth and protein homeostasis . It is important to note that RPs are among the most abundant ubiquitinylated proteins that accumulate in the nucleus of proteasome-deficient S . cerevisiae and human cells ( Sung et al . , 2016a; Sung et al . , 2016b; Lam et al . , 2007; Mayor et al . , 2007 ) , suggesting that the synthesis of RPs and their assembly into ribosomes must be tightly coordinated with the cell’s proteostasis capacity . Consistent with this view , we show here that induction of ribosome assembly stress is correlated with the rapid accumulation of RPs in a detergent-insoluble fraction and that blocking de novo RP production , either by anchoring away Ifh1 or treating cells with cycloheximide , diminishes or abolishes a key transcriptional consequence of RASTR , namely upregulation of Hsf1 target genes . Our observations thus strongly suggest that RP aggregates are an important activating signal for RASTR . Nevertheless , we and others ( Sung et al . , 2016a ) detect a large number of additional proteins that accumulate in an insoluble fraction upon ribosome assembly stress , including many RiBi proteins ( e . g . numerous rRNA helicases and processing factors ) , the RP gene activator Ifh1 , and chaperones . At present , we do not know the precise molecular nature of these aggregates , which presumably accumulate in the nucleolar space , or even whether they represent a common structure . In any event , our data suggest that RP- and Ifh1-containing aggregates are highly dynamic since the promoter release of Ifh1 and activation of Hsf1 following Top1 degradation are rapidly reversed by Top2 compensation ( Figure 1 ) and Ifh1 aggregates that appear in tom1-Δ cells quickly diminish following cycloheximide treatment ( Figure 5—figure supplement 1A ) . We imagine that this provides a strong selective advantage by allowing cells to rapidly recover from a transient disruption of ribosome biogenesis . Accumulation of proteins in insoluble fractions or aggregates underlies numerous diseases , as well as aging ( Saarikangas and Barral , 2015; Tuite and Melki , 2007 ) . However , certain protein aggregates appear to be dynamic structures that contribute to cellular fitness by protecting the cell during stress ( Cherkasov et al . , 2013; Douglas et al . , 2008; Grousl et al . , 2018; Kaganovich et al . , 2008; Miller et al . , 2015b ) . For example , stress granules and P-bodies , two of the most intensively studied insoluble macromolecular aggregates , have emerged as important cytoplasmic regulators of gene expression by controlling the processing , sequestering and/or degradation of specific RNA transcripts ( Decker and Parker , 2012; Mahboubi and Stochaj , 2017 ) . Interestingly , it has recently been reported that nucleolar proteins can form sub-compartmental structures by promoting liquid-liquid phase separation ( Berry et al . , 2015; Feric et al . , 2016 ) . Although further work will be required to characterize the composition , assembly and function of these nucleolar membrane-less structures , they are attractive candidates for regulatory hubs that could act by sensing ribosome biogenesis stress and controlling adaptive responses . With respect to the present study , we imagine that liquid phase-separated structures in the nucleolus could be directly involved in sequestration of Ifh1 during RASTR , as well as the titration of Hsp70 that we propose leads to Hsf1 activation . A challenge for future studies will be to characterize the physical properties of these postulated structures and their relevance to the transcriptional outputs that we measure here . We showed previously ( Albert et al . , 2016 ) that in strains where RNAPI is constitutively active ( Laferté et al . , 2006 ) Ifh1 is released from RP gene promoters shortly after TORC1 inhibition by rapamycin treatment ( ~5 min ) but returns only 15 min later . This promoter re-binding does not occur in wild-type cells due to the action of two RiBi proteins ( Utp22 and Rrp7 ) that can sequester Ifh1 in the CURI ( Casein Kinase 2/Utp22/Rrp7/Ifh1 ) complex ( Albert et al . , 2016; Rudra et al . , 2007 ) allowing to re-align RP gene expression with RNAPI activity . These findings revealed that S . cerevisiae has developed temporally distinct mechanisms to regulate RP gene expression . One of the keys finding in the present study is to confirm the existence of a two-step process in RP gene regulation and to link the first step to control of the Hsf1 regulon . We propose that this short timescale mechanism results from a rapid rise in unassembled RPs that occurs immediately following TORC1 inhibition , or other stresses that disrupt ribosome assembly , such as depletion of Top1 or RiBi factors . Following ribosome assembly stress , we imagine that the rapid induction of Hsf1 target genes , in combination with the arrest of RP gene transcription , contributes to the eventual clearing of proteotoxic unassembled RPs from the nucleus , thereby removing the signal that promotes RASTR and leads to Ifh1 promoter release and Hsf1 activation . Accordingly , RP production is abolished by 20 min following TORC1 inhibition ( Reiter et al . , 2011 ) , suggesting that RASTR becomes inoperant , thus explaining the downregulation of Hsf1 genes ( Figure 6B ) and the switch to a secondary regulatory mechanism , involving sequestration of Ifh1 in the CURI complex , to align RP expression with rRNA production ( Albert et al . , 2016 ) . Consistent with this view , the short time scale mechanism of RP gene downregulation is abolished by attenuating proteotoxicity through translation inhibition ( cycloheximide treatment ) whereas the long time scale process is insensitive to translation arrest but can be prevented by expression of a constitutively active RNAPI ( Albert et al . , 2016; Laferté et al . , 2006 ) . These two independent mechanisms adapt RP gene expression to both rRNA production and ribosome assembly , thus minimizing the accumulation of unassembled RPs . As alluded to above , ribosome assembly stress in higher eukaryotes has been studied extensively in the context of ribosomopathies , diseases often associated with RP gene haplo-insufficiencies , RP gene point mutations or mutations in RiBi factors . One hypothesis put forward to explain these observations is that unassembled RPs trigger a feedback mechanism that decreases transcription of ribosome biogenesis genes by inhibiting c-Myc function and arrests cell growth through p53 activation ( Dai et al . , 2007; Liu et al . , 2016 ) . It has also been recently reported that the rRNA helicase DDX21 binds to and activates RP gene promoters in a manner that may be sensitive to the status of ribosome biogenesis ( Calo et al . , 2015 ) . Taken together , these findings suggest that unassembled RPs could mediate an ancestral process to regulate ribosome biogenesis conserved from prokaryotes ( Nomura , 1999 ) to eukaryotes . Transcriptome analysis immediately following ribosome assembly stress in mammalian cells will be required to understand the interplay between these different mechanisms and may also uncover novel pathways . Our work also provides insights into the connection between ribosome assembly and Hsf1 that was first revealed in a report from the Churchman lab that appeared as our work was being prepared for publication ( Tye et al . , 2018 ) . Hsf1 is a key sensor of proteotoxic stress in all eukaryotes that controls a common set of chaperones conserved from yeast to human . One protective function reported for Hsf1 is its ability to reduce protein aggregate formation leading to neurogenerative diseases ( Neef et al . , 2014 ) . On the other hand , Hsf1 also exerts a pro-oncogenic function through its ability to promote proteostasis in rapidly growing tumor cells ( Mendillo et al . , 2012; Santagata et al . , 2011 ) . Despite its central function , a holistic understanding of the regulatory mechanisms that govern Hsf1 activity still missing . Our work and that of Tye et al . ( 2018 ) demonstrates that Hsf1 activity is tightly linked to ribosome biogenesis in yeast , in a manner independent of the previously described RQC mechanism that contributes to the dissociation of aberrant nascent polypeptides from the ribosome ( Brandman et al . , 2012 ) . These two mechanisms highlight the central importance of ribosome assembly and activity in regulation of cellular protein homeostasis through Hsf1 . Although it is currently unknown if RASTR is conserved in metazoans , we note that RPs are also subjected to a high turnover rate compared to other nuclear components in mammalian cells and that proteasome or ribosome assembly inhibition trigger a rapid accumulation of RPs in the nucleus , whereas arrest of translation has an opposite effect ( Sung et al . , 2016a; Lam et al . , 2007 ) . Importantly , it was reported that cycloheximide treatment also abolishes Hsf1 activity in mammalian cell by an unknown mechanism ( Santagata et al . , 2013 ) . We propose that a dynamic balance between unassembled and assembled RPs could be sensed by Hsf1 to constantly adjust protein homeostasis transcription programs in eukaryotes with translational flux , proteolysis and the rate of ribosome assembly ( Figure 8 ) , since disruption or hyperactivation of any of these processes will rapidly change nuclear levels of free RPs . Given the growing body of evidence linking Hsf1 activity to numerous diseases associated with proteotoxic stress , but also rapid cell growth in cancer , it will be of great interest to challenge this model in the future .
Cultures of 50 mL in YPAD were collected at OD6000 . 4–0 . 8 for each condition . The cells were crosslinked with 1% formaldehyde for 5 min at room temperature and quenched by adding 125 mM glycine for 5 min at room temperature . Cells were washed with ice-cold HBS and resuspended in 3 . 6 mL of ChIP lysis buffer ( 50 mM HEPES-Na pH 7 . 5 , 140 mM NaCl , 1 mM EDTA , 1% NP-40 , 0 . 1% sodium deoxycholate ) supplemented with 1 mM PMSF and 1x protease inhibitor cocktail ( Roche ) . Samples were aliquoted in Eppendorf tubes and frozen . After thawing , the cells were broken using Zirconia/Silica beads ( BioSpec ) . The lysate was spun at 13 , 000 rpm for 30 min at 4°C and the pellet was resuspended in 300 μl ChIP lysis buffer + 1 mM PMSF and sonicated for 15 min ( 30 s ON - 60 s OFF ) in a Bioruptor ( Diagenode ) . The lysate was spun at 7000 rpm for 15 min at 4°C . Antibody ( 1 μg / 300 μL of lysate , Abcam ab5131 ) was added to the supernatant and incubated for 1 hr at 4°C . Magnetic beads were washed three times with PBS plus 0 . 5% BSA and added to the lysates ( 30 μL of beads/300 μL of lysate ) . The samples were incubated for 2 hr at 4°C . The beads were washed twice with ( 50 mM HEPES-Na pH 7 . 5 , 140 mM NaCl , 1 mM EDTA , 0 . 03% SDS ) , once with AT2 buffer ( 50 mM HEPES-Na pH 7 . 5 , 1 M NaCl , 1 mM EDTA ) , once with AT3 buffer ( 20 mM Tris-Cl pH 7 . 5 , 250 mM LiCl , 1 mM EDTA , 0 . 5% NP-40 , 0 . 5% sodium deoxycholate ) and twice with TE . The chromatin was eluted from the beads by resuspension in TE + 1% SDS and incubation at 65°C for 10 min . The eluate was transferred to an Eppendorf tube and incubated overnight at 65°C to reverse the crosslinks . The DNA was purified using High Pure PCR Cleanup Micro Kit ( Roche ) . DNA libraries were prepared using TruSeq ChIP Sample Preparation Kit ( Illumina ) according to manufacturer’s instructions . The libraries were sequenced using an Illumina HiSeq 2500 and the reads were mapped to the sacCer3 genome assembly using HTSStation ( shift = 150 bp , extension = 50 bp; David et al . , 2014 ) . To compare depleted versus non-depleted cells , we divided the signal from the + auxin and/or rapamycin and/or cycloheximide samples by the signal from the – auxin and/or rapamycin and/or cycloheximide ( vehicle ) samples and log2 transformed this value . All data from publicly available databases were mapped using HTS Station ( http://htsstation . epfl . ch; David et al . , 2014 ) . Strains used in this study are listed in Supplementary file 7 . For ChIP-qPCR , the primer sequences used a listed in Supplementary file 8 . Experiments were typically performed with log phase cells harvested between OD600 0 . 4 and 0 . 8 . Anchor-away of FRB-tagged proteins was induced by the addition of rapamycin ( 1 mg/ml of 90% ethanol/10% Tween stock solution ) to a final concentration of 1 μg/ml ( Haruki et al . , 2008 ) . Depletion of AID-tagged protein was induced by the addition of auxin ( 3-indoloacetic acid ) at 500 μM final concentration . Arrest of translation was induced by the addition of cycloheximide to a final concentration of 25 μg/ml . Cells are treated with diazaborine to a final concentration of 50 ug/ml . Cells were grown overnight at 30°C in SC medium ( 0 . 67% nitrogen base without amino acids ( BD ) , 2% dextrose supplemented with amino acids mixture ( AA mixture; Bio101 ) , adenine , and uracil ) . Cells were diluted and were harvested when OD600 reached 0 . 4 . Cells were spread on slides coated with an SC medium patch containing 2% glucose . Stacked images were recorded ( Intelligent Imaging Innovations ) at a spinning disc confocal inverted microscope ( Leica DMIRE2 ) using the 100x oil objective and an Evolve EMCCD Camera ( Photometrics ) . Isolation of protein aggregates from yeast cells was performed as described previously ( Koplin et al . , 2010 ) with slight modifications . 50 OD600 units ( 50 ml ) of exponentially growing cells were harvested , and cell pellets were frozen in liquid N2 . The cell pellets were resuspended in 1 ml lysis buffer ( 20 mM Na-phosphate pH 6 . 8 , 10 mM DTT , 1 mM EDTA , 0 . 1% Tween , 1 mM PMSF , protease inhibitor cocktail and 100 units/ml zymolyase ) and incubated at 30° C for 30 min . Chilled samples were treated by tip sonication ( 20% , 10 s , 2x ) and centrifuged for 20 min at 600 g at 4°C . Aggregated proteins were pelleted at 16 , 000 g for 20 min at 4°C . After removing supernatants , insoluble proteins were washed once with Wash I buffer ( 20 mM Na-phosphate pH 6 . 8 , 500 mM NaCl , 5 mM EDTA , 2% NP-40 , 1 mM PMSF , and protease inhibitor cocktail ) , and centrifuged at 16 , 000 g for 20 min at 4°C . Insoluble proteins were washed with Wash II buffer ( 20 mM Na-phosphate pH 6 . 8 , ice-cold ) , pelleted and sonicated ( 2x for 10 s ) in 40 μl of Wash II buffer . For analysis by SDS-PAGE ( 4–12% acrylamide ) and subsequent western blotting , proteins were first boiled in Laemmli buffer . 1x of the total cell lysate and 25x of the insoluble pellet fraction were separated and analyzed by Coomassie Blue staining or immunoblotting . Proteins were identified by shotgun mass spectrometry analysis at the Functional Genomics Center Zurich ( ETH , Zurich ) following TCA precipitation ( 20% ) and acetone washing , according to posted procedures . Database searches were performed by using the Mascot ( SwissProt , all species; SwissProt , yeast ) search program , using very stringent settings in Scaffold ( 1% protein FDR , a minimum of 2 peptides per protein , 0 . 1% peptide FDR ) . Yeast cells growing exponentially were treated or not with auxin for 20 min . 50 µg/mL cycloheximide ( Sigma ) was added directly to the culture medium . Cells were collected by centrifugation , rinsed with buffer K [20 mM Tris-HCl pH 7 . 4 , 50 mM KCl , 10 mM MgCl2] supplemented with 50 µg/mL cycloheximide and collected again by centrifugation . Dry pellets were resuspended with approximately one volume of ice-cold buffer K supplemented with 1 mM DTT , 1 × Complete EDTA-free protease inhibitor cocktail ( Roche ) , 0 . 1 U/µL RNasin ( Promega ) and 50 µg/mL cycloheximide . About 250 µL of ice-cold glass beads ( Sigma ) were added to 500 µL aliquots of the resuspended cells and cells were broken by vigorous shaking , three times 2 min , separated by 2 min incubations on ice . Extracts were clarified through two successive centrifugations at 13 , 000 rpm and 4°C for 5 min and quantified by measuring absorbance at 260 nm . About 30 A260 units were loaded onto 10–50% sucrose gradients in buffer K , and then centrifuged for 150 min at 39 , 000 rpm and 4°C in an Optima L-100XP Ultracentrifuge ( Beckman-Coulter ) using a SW41Ti rotor without brake . Following centrifugation , 18 fractions of 500 µl each were collected from the top of the gradients with a Foxy Jr . apparatus ( Teledyne ISCO ) . The absorbance at 254 nm was measured during collection with a UA-6 device ( Teledyne ISCO ) . Metabolic labeling of pre-rRNAs was performed as previously described ( Tollervey et al . , 1993 ) with the following modifications . Strains were grown in synthetic glucose medium lacking adenine to an OD600 of 0 . 8 . Auxin ( 0 . 5 mM ) was next added to the cultures and cells were labeled for 2 min with [2 , 8-3H]-adenine ( NET06300 Perkin Elmer ) at 0 , 10 , 20 and 30 min following the addition of auxin . Cell pellets were frozen in liquid nitrogen . RNA extractions and Northern hybridizations were performed as previously described ( Beltrame and Tollervey , 1992 ) . For high-molecular-weight RNA analysis , 2 μg of total RNA were glyoxal denatured and resolved on a 1 . 2% agarose gel . Note that Northern hybridization was performed on [2 , 8-3H]-adenine labeled RNA . The membrane was first exposed to reveal neo-synthetized transcripts , and subsequent Northern hybridization revealed rRNA transcript abundance . Read counts for all RNAPII ChIP-seq experiments ( integrated counts over the complete open reading frame of all protein-coding genes ) are given in Supplementary file 1–6 . Primary processed sequence files will be made available at Gene Expression Omnibus ( GEO accession number GSE125226 ) . | When yeast cells are growing at top speed , they can make 2 , 000 new ribosomes every minute . These enormous molecular assemblies are the protein-making machines of the cell . Building new ribosomes is one of the most energy-demanding parts of cell growth and , if the process goes wrong , the results can be catastrophic . The proteins that make up the ribosomes themselves are sticky . Left unattended , they start to form toxic clumps inside the compartment that houses most of the cell’s DNA , the nucleus . A protein called Heat shock factor 1 , or Hsf1 for short , plays an important role in the cell's quality control systems . It helps to manage sticky proteins by switching on genes that break down protein clumps and prevent new clumps from forming . Hsf1 levels start to rise whenever cells are struggling to keep up with protein production . If it is half-finished ribosomes that are causing the problem , cells can stop making ribosome proteins . The protein in charge of this in yeast is Ifh1 . It is a transcription factor that sits at the front of the genes for ribosome proteins , switching them on . When yeast cells get stressed , Ifh1 drops away from the genes within minutes , switching them off again . Yet how this happens , and how it links to Hsf1 , is a mystery . To start to provide some answers , Albert et al . disrupted the production of ribosomes in yeast cells and examined the consequences . This revealed a new rescue response , that they named the “ribosome assembly stress response” . Both Hsf1 and Ifh1 are sensitive to the build-up of unfinished ribosomes in the nucleus . As expected , Hsf1 activated when ribosome proteins started to build up , and switched on the genes needed to manage the protein clumps . The effect on Isfh1 , however , was unexpected . When the unassembled ribosome proteins started to build up , it was the clumps themselves that pulled the Ifh1 proteins off the genes . The unassembled ribosomes proteins seemed to be stopping their own production . Low levels of clumped ribosome proteins in the nuclei of unstressed cells also helped to keep Hsf1 active and pull Ifh1 off the ribosome genes . It is possible that this provides continual protection against a toxic protein build-up . These findings are not only important for understanding yeast cells; cancer cells also need to produce ribosomes at a very high rate to sustain their rapid growth . They too might be prone to stresses that interrupt their ribosome assembly . As such , understanding more about this process could one day lead to new therapies to target cancer cells . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"chromosomes",
"and",
"gene",
"expression",
"cell",
"biology"
] | 2019 | A ribosome assembly stress response regulates transcription to maintain proteome homeostasis |
Current models successfully describe the auditory cortical response to natural sounds with a set of spectro-temporal features . However , these models have hardly been linked to the ill-understood neurobiological changes that occur in the aging auditory cortex . Modelling the hemodynamic response to a rich natural sound mixture in N = 64 listeners of varying age , we here show that in older listeners’ auditory cortex , the key feature of temporal rate is represented with a markedly broader tuning . This loss of temporal selectivity is most prominent in primary auditory cortex and planum temporale , with no such changes in adjacent auditory or other brain areas . Amongst older listeners , we observe a direct relationship between chronological age and temporal-rate tuning , unconfounded by auditory acuity or model goodness of fit . In line with senescent neural dedifferentiation more generally , our results highlight decreased selectivity to temporal information as a hallmark of the aging auditory cortex .
Age-related hearing loss is a frequent cause for a decline in speech comprehension , particularly in complex acoustic environments . As the temporal envelope of speech is crucial for speech comprehension ( Shannon et al . , 1995 ) , the ability to accurately encode temporal features in the auditory system is pivotal for successful perception of speech and conspecific vocalizations ( Peelle and Wingfield , 2016 ) . Age-related deficits have been observed in different temporal psychoacoustic tasks ( e . g . gap detection; Snell et al . , 2002 ) . Those psychoacoustic studies suggest that the precision of processing of temporal cues declines with age ( e . g . Gordon-Salant et al . , 2006 ) . While age-related hearing loss has been attributed to a variety of peripheral declines in auditory coding ( e . g . Gates and Mills , 2005 ) , fewer studies have looked at more central levels of auditory processing . Yet , declines in temporal processing are hypothesized to be not exclusively peripheral in origin , but more central than at least the mid-brain ( Recanzone , 2018 ) . In the aged macaque auditory cortex , neurons become more broadly tuned to temporal modulations and temporal fidelity of cortical responses decreases ( Ng and Recanzone , 2018 ) . Topographically , a neural dedifferentiation is suggested to take place in the older brain: In young macaques , the neuronal response to the inter-stimulus interval in primary field A1 and caudolateral belt area ( CL ) differs in that A1 neurons have shorter response latencies . In aged animals , however , no difference was observed between A1 and CL such that aged A1 neurons had an equivalent response latency to CL neurons ( Ng and Recanzone , 2018 ) . Further , in aged neurons , a shift in neural coding strategy from less temporal coding towards more rate coding of temporal modulations was evident ( Overton and Recanzone , 2016; for a review of both coding strategies see Joris et al . , 2004 ) . For those neurons that still adhered to a temporal coding strategy , the temporal fidelity decreased , although the absolute number of neurons responsive to temporal modulations was unaffected by age ( Overton and Recanzone , 2016 ) . At higher processing levels , only a number of relatively unspecific changes have been reported in the older human listener ( for review see Peelle and Wingfield , 2016 ) . For example , the frontal and cingulo-opercular cortical response is elevated during a challenging speech comprehension or gap detection task ( Erb and Obleser , 2013; Vaden et al . , 2015; Vaden et al . , 2020 ) . This has been interpreted as compensatory mechanism in response to loss of sensory acuity . Aging has been hypothesized to lead to a neural dedifferentiation in sensory cortices ( e . g . , Park et al . , 2004 ) . A more recent unspecific observation is that with age , decoding accuracy of stimulus conditions from auditory cortical fMRI responses declines ( Lalwani et al . , 2019; but see discussion for contrasting findings in MEG speech tracking , for example Presacco et al . , 2016 ) . This finding in turn has been linked to an imbalance of excitation and inhibition in older adults , more specifically a reduction of GABA levels in auditory cortex ( Lalwani et al . , 2019 ) . Accumulating evidence supports the theory of an age-related loss of inhibition in sensory cortices ( Caspary et al . , 2008 ) . A recent model of cortical processing ( Chi et al . , 2005 ) has led to substantial progress in our understanding of how natural sounds become represented in the auditory cortex . A series of studies have shown that cortical processing of sounds is optimized to represent the spectro-temporal modulations which are typically present in conspecific vocalizations such as speech or animal calls ( Santoro et al . , 2014; Hullett et al . , 2016; Santoro et al . , 2017; Erb et al . , 2019a ) . However , those studies have not looked at the pressing case of the aged human cortex . It remains unknown whether — and if so , which — general response properties are altered in aged auditory cortex . The present study applies an fMRI encoding and decoding approach to compare auditory cortical responses to natural sounds in young and older humans . We focus on the age-group comparison at the level of representation of fundamental acoustic features ( spectro-temporal modulations ) . Results show that while the large-scale topographic organization of acoustic features is preserved in the older auditory cortex , age groups differ in tuning to temporal modulations: Human aging thus appears to be accompanied by an anatomically and functionally specific broadening of temporal-rate tuning in auditory cortex .
We compared three models of hemodynamic responses to sound . The models respectively describe the fMRI responses to the ( 1 ) speech stream ( foreground ) , ( 2 ) textures ( background ) , or ( 3 ) the mixture of both streams ( for modulation spectra of the different streams see Figure 1—figure supplement 2 ) .
Using natural stimuli , we revealed topographic maps of acoustic feature preference for frequency , spectral and temporal modulations . Thus , we show that topographic maps of feature preference can be observed under approximately natural hearing situations . Tonotopic maps exhibited the well-established mirror-symmetric high-low-high frequency gradients across Heschl’s gyrus and adjacent areas ( Figure 3 ) consistent with abundant evidence in human ( Formisano et al . , 2003; Talavage et al . , 2004; Humphries et al . , 2010; Woods et al . , 2010; Da Costa et al . , 2011; Striem-Amit et al . , 2011; Langers and van Dijk , 2012; Moerel et al . , 2012; Moerel et al . , 2014 ) and non-human primates ( Merzenich and Brugge , 1973; Morel et al . , 1993; Kosaki et al . , 1997; Rauschecker et al . , 1997; Bendor and Wang , 2008; Baumann et al . , 2013; Joly et al . , 2014; Baumann et al . , 2015 ) . Notably , tonotopic maps in younger versus older adults were highly correlated , indicating topographical stability of tonotopic organization across age groups . The well-demonstrated existence of tonotopic maps notwithstanding , a topographic organization for temporal or spectral modulations in auditory cortex is less clear ( Schönwiesner and Zatorre , 2009 ) . Here , in both young and older adults we observed a medial-to-lateral low-to-high gradient for temporal modulations across the superior temporal plane ( Figure 4a ) , while for spectral modulations , the locus with the highest resolution ( ~1 . 5 cyc/oct ) was observed in lateral Heschl’s gyrus ( Figure 4b ) . A topographic representation for modulation rate has previously been observed in the primate auditory cortex ( Baumann et al . , 2015 ) where fast temporal acoustic information was preferably encoded in caudal auditory regions ( Camalier et al . , 2012; Kusmierek and Rauschecker , 2014 ) and slow temporal information in rostral areas R and RT ( Liang et al . , 2002; Bendor and Wang , 2008 ) . Whereas earlier human fMRI studies had failed to show a clear topography for modulation rate ( Giraud et al . , 2000; Schönwiesner and Zatorre , 2009; Overath et al . , 2012; Leaver and Rauschecker , 2016 ) , recent human fMRI ( Santoro et al . , 2014 ) and electrocorticography ( ECoG ) studies ( Hullett et al . , 2016 ) proposed the presence of a posterior-to-anterior high-to-low rate gradient in human auditory cortex . However , comparison of those studies is hampered by the differences in coverage and ranges of temporal modulations: While Hullett et al . , 2016 examined representation of slow temporal modulation rates present in speech ( 1–2 Hz ) only along the STG with ECoG , Santoro et al . , 2014 analyzed rates of 1–27 Hz , whereas Baumann et al . , 2015 presented rates of up to 512 Hz with more comprehensive coverage using fMRI . While previous studies had presented single natural sounds ( Erb et al . , 2019a ) , or even synthetic sounds such as AM stimuli ( Baumann et al . , 2015 ) or dynamic ripples ( Schönwiesner and Zatorre , 2009 ) , here , we presented a continuous stream of speech embedded in an acoustically rich background of sound textures . As an important foundation for the main conclusions of this manuscript , we here demonstrate the capacity to derive meaningful tonotopic maps from natural stimulus conditions , at conventional field strengths ( 3 T ) and in special populations ( older participants ) . Although an inherent caveat of the use of natural stimulus conditions is the difficulty of measuring psychoacoustic performance , this experimental design has the advantage that it closely approximates natural listening conditions ( Hamilton and Huth , 2020 ) . As synthetic sounds lack both the behavioural relevance and the statistical structure of natural sounds , they activate auditory cortex differently than natural stimuli ( Theunissen et al . , 2000; Bitterman et al . , 2008 ) . Methodologically , our encoding approach relies on the assumption that acoustic features that maximally activate single voxels are more accurately encoded . However , higher responses may not necessarily mean better encoding . The second , complementary decoding analysis incorporates the whole range of values rather than only the maximal values . The reconstruction accuracy explicitly quantifies the amount of information about a set of stimulus features that is available in a region of interest . Accurate reconstruction of acoustic features in test sounds indicates that these features are reproducibly mapped into distinct spatial patterns . As data from individual voxels are jointly modelled , this combined analysis of signals from multiple voxels increases the sensitivity for stimulus information that may be represented in patterns of responses , rather than in individual voxels ( Santoro et al . , 2017 ) . Under the assumption that the reconstruction accuracy reflects tuning properties of neuronal populations , the observed decrease in selectivity reveals a broadening of temporal tuning functions with age . This finding provides mechanistic insights into the changes of temporal processing in the aging auditory cortex: The broadened temporal tuning , reminiscent of an age-related decrease in temporal precision in the midbrain ( Anderson et al . , 2012 ) , may underlie the difficulties older adults experience in speech comprehension in noisy environments ( Anderson et al . , 2011 ) . While the present study lacks more direct measures of encoding along the auditory pathway , it is important to note that rate selectivity in the older listeners was statistically unrelated to their pure-tone audiograms ( this holds true for both presented measures of tuning , the selectivity index [Figure 5] and the variance across temporal rates [Figure 6] ) . Thus , the broadening of temporal tuning is not explicable merely by peripheral hearing loss . Future studies need to carefully assess the consequences of altered cortical temporal tuning on behavioural and psychoacoustic performance ( Erb et al . , 2019b; Flinker et al . , 2019; Holmes and Griffiths , 2019 ) , a measure that was precluded by the use of natural sounds in the current study . The observation of decreased temporal selectivity may seem at odds with previous accounts of elevated cortical responses to the temporal speech envelope in aging . Evidence for better reconstruction of temporal envelopes from older listeners’ MEG signals ( Presacco et al . , 2016 ) and larger N1 amplitudes in older adults with widened auditory filters ( Herrmann et al . , 2016; Henry et al . , 2017 ) suggest an enhanced cortical envelope representation . However , an inflated cortical representation of the speech envelope has been argued to rather be a maladaptive response as compensation of the lack of temporal precision in the midbrain ( Anderson et al . , 2012 ) . Neuronal coding of temporal modulations follows two distinct principles , that is rate and temporal coding ( Joris et al . , 2004 ) . While rate coding constitutes a variation of overall spike rate as a function of modulation frequency , temporal coding is achieved through phase-locking to the stimulus envelope . Fast temporal modulations ( >50 Hz ) are typically encoded through the rate code , while neurons tuned to slow temporal rates ( <50 Hz ) mostly synchronize to the sounds’ modulations ( temporal code , Sachs , 1984; Joris et al . , 2004 ) . It remains unclear how hemodynamic responses reflect these two types of neuronal coding and how those are in turn related to the observed age-related broadening of tuning functions . Evidence from electrophysiology suggests a change in coding strategy across neuronal populations in A1 from less temporal coding towards more rate coding of temporal modulations in aged animals ( Ng and Recanzone , 2018 ) . Interestingly , the observed age-related decrease in selectivity to temporal modulations was restricted to areas HG/HS and PT , the putative homologues of areas A1 and CL in the macaque ( Kaas and Hackett , 2000 ) . Both areas have been shown to exhibit altered temporal encoding in older macaques ( Ng and Recanzone , 2018 ) . In a direct comparison of the vector strengths ( a metric of periodicity of the neuronal response to a modulated signal ) as a function of the inter-stimulus interval , responses differed in young A1 and CL neurons such that A1 neurons had higher vector strength . This is consistent with the notion that CL neurons process predominantly spatial ( rather than temporal ) information . Therefore , so the theory , CL neurons would not necessarily need to pass on the temporal coding of A1 neurons . This difference between belt and core areas was abandoned in the older animals , where aged A1 were similar to both the aged and young CL neurons , indicating that the temporal fidelity of A1 responses decreases with age . However , a good temporal fidelity of A1 neurons is thought to be critical for temporal processing . We speculate that such an age-related decline in temporal processing may be linked to an imbalance of excitation and inhibition observed in aging ( Voytek et al . , 2015 ) . Inhibitory ( e . g . GABAergic or glycinergic ) neurotransmitters have been shown to increase response synchrony to modulated stimuli in both the cochlear nucleus and the inferior colliculus ( Koch and Grothe , 1998; Backoff et al . , 1999 , for review see Caspary et al . , 2008 ) , suggesting that inhibitory neurotransmission is of particular importance to precise neural timing and , thus , the adequate tracking of temporal sound features . An inherent challenge to age group comparisons is to distinguish any unspecific decrease of SNR in the data of older participants ( e . g . , due to vascular changes that impact the BOLD response Garrett et al . , 2017 or due to the typical increase in movement artefacts ) from specific age-related sensory processing changes . Most reassuringly , however , a multiple regression model taking into account overall goodness of the encoding model fit ( i . e . , sound identification accuracy ) did not alter the observed relationship between age and temporal rate tuning . The present study aimed to narrow the gap between recent progress in modelling the neural processing of acoustic features from natural stimuli on the one hand , and the understanding of potential senescent changes in these cortical stimulus representations on the other hand . Although the large-scale topographic organization of acoustic features appears preserved in the auditory cortex of older compared to younger listeners , age-related differences in the marginal profiles of multi-voxel MTFs were evident . Tuning to slow temporal rates which abounds in natural sounds and especially in speech ( Erb et al . , 2019a ) was markedly sharper in young compared to older participants . Consistent with previous findings in the macaque , these results suggest that temporal rate selectivity in auditory cortex declines in normal aging . The specificity of this decline , confined to primary auditory and adjacent areas and sparing tonotopic representations , opens a new lead in the ongoing search for tractable neurobiological signatures of older adults’ widely observed deficits in speech comprehension in noisy environments .
We invited a total of n = 75 participants for scanning from which we had to exclude one participant due to excessive movement , three participants due to incidental neurological findings , five participants due to their inability to understand speech in noise , one participant due to broken headphones , one participant due to age ( 43 years , i . e . , could not be assigned to young or older group ) . The remaining n = 64 participants were right-handed , young ( n = 33; aged 18–32 , mean 24 . 7 years18 female ) and older ( n = 31; aged 51–78 , mean 63 . 8 years , 15 female ) native German speakers . Simulations ( custom matlab code ) showed that a two-sample permutation test at a conventional type I error rate of 5% with a sample size of N = 27 per group can detect medium to large effects of Cohen’s d = 0 . 75 ( as can be expected for the neural fMRI measures under consideration , for example Alavash et al . , 2019 ) with a satisfactory power of 80% . The power of our procedure reduces accordingly if the true effect in the population is smaller but remains over 60% also for true effects closer to Cohen’s d = 0 . 5 . Younger participants had self-reported normal hearing . Older participants’ hearing ranged from normal hearing to mild hearing loss . The older participants were part of a cohort which is regularly tested in the lab ( e . g . , Alavash et al . , 2019 ) . They were recruited based on their audiograms that had been acquired within the two years prior to the experiment ( for audiograms see Figure 1—figure supplement 1a ) and were excluded from the study if the pure-tone-average ( PTA ) of the better ear exceeded 30 dB HL . On average , hearing thresholds are expected to increase by approximately 0 . 25–1 . 7 dB HL per year at the age range of 48–79 years ( Wiley et al . , 2008 ) . Thus , we considered 2-year-old audiograms to be valid . All participants gave informed consent and were financially compensated or received course credit . All procedures were approved by the local ethics committee of the University of Lübeck ( ethical approval AZ 16–107 ) . Participants listened to 64 min of a freely narrated audiobook ( Hertha Müller , ‘Die Nacht ist aus Tinte gemacht’ ) presented against a competing stream of resynthesized natural sounds ( ‘sound textures’; McDermott and Simoncelli , 2011 ) at 0 dB SNR . Textures were synthesized from the spectro-temporal modulation content of a large set of real-life sounds ( n = 192 ) , including speech and vocal samples , music pieces , animal cries , scenes from nature and tool sounds that had been used in previous studies ( Moerel et al . , 2013; Santoro et al . , 2017 ) . Texture synthesis parameters were as follows: Frequency range = 0 . 02–10 kHz , number of frequency bands = 30 , sampling rate = 20 kHz , temporal modulation range = 0 . 5–200 Hz , sampling rate = 400 Hz; maximum number of iterations = 60 . Texture exemplars of 5 s length were concatenated to form the background stream . The order of exemplars was pseudo-randomized across participants ( in four different sound orders ) . In total , each exemplar was repeated four times . Participants were asked to listen to the story and answer three four-choice questions on its semantic content after each run ( 24 questions in total ) ; chance level was thus proportion correct = 0 . 25 . Four-answer choice questions were displayed on a screen and participants responded via a button box in their right hand . Both the young and older group performed on average above chance ( see Figure 1—figure supplement 1b ) , but young participants performed significantly better than older participants ( mean proportion correct difference = 0 . 1 , p=0 . 009 [permutation test] , Cohen’s d = 0 . 65 ) . We acquired functional and structural MRI at 3T ( Siemens Magnetom Skyra ) with a 64-channel RF head array coil . Functional T2*-weighted data were acquired using an echo planar imaging sequence . We collected eight runs of eight minutes each ( each run contained 519 volumes ) using continuous scanning . The acquisition parameters were as follows: repetition time ( TR ) = acquisition time ( TA ) = 947 ms , echo time ( TE ) = 28 ms , acceleration factor = 4 , flip angle = 60° , field of view ( FOV ) = 200×200 mm , 52 slices; voxel size = 2 . 5 mm isotropic ( whole-brain coverage ) . Field maps for intensity inhomogeneity correction were acquired after every second run ( TR = 610 ms , TE1 = 4 . 92 , TE2 = 7 . 38 , flip angle = 60° , FOV = 200×200 mm , 62 slices , voxel size = 2 . 5 mm isotropic ) . Anatomical T1-weighted images were acquired using an MPRAGE sequence ( TR = 2400 ms , time to inversion [TI]=1000 ms , TE = 3 . 16 ms , flip angle = 8° , FOV = 256×256 mm , number of slices = 176 , voxel size = 1 mm isotropic , GRAPPA acceleration factor = 2 ) . T2-weighted images were collected at the end of the session ( TR = 3200 ms , TE = 449 ms , FOV = 256×256 mm , number of slices = 176 , voxel size = 1 mm isotropic , GRAPPA acceleration factor = 2 ) . Results included in this manuscript are based on a preprocessing pipeline of fMRIPprep 1 . 2 . 4 which is based on Nipype 1 . 1 . 6 ( Esteban et al . , 2019 ) . The following two paragraphs on preprocessing are based on an automatically generated output of fMRIPprep . The modulation content of the stimuli was obtained by filtering the sounds within a biologically plausible model of auditory processing ( Chi et al . , 2005 ) . This auditory model consists of an early stage that models the transformations that acoustic signals undergo from the cochlea to the midbrain; and a cortical stage that accounts for the processing of the sounds at the level of the auditory cortex . We derived the spectrogram and its modulation content using the ‘NSL Tools’ package ( available at http://www . isr . umd . edu/Labs/NSL/Software . htm ) and customized Matlab code ( The MathWorks Inc , Matlab 2014b/2018a ) . Spectrograms for all sounds were obtained using a bank of 128 overlapping bandpass filters with equal width ( Q10dB = 3 ) , spaced along a logarithmic frequency axis over a range of f = 180–7040 Hz . The output of the filter bank was band-pass filtered ( hair cell stage ) . A midbrain stage modelled the enhancement of frequency selectivity as a first-order derivative with respect to the frequency axis , followed by a half-wave rectification and a short-term temporal integration ( time constant τ = 8 ms ) . Next , the auditory spectrogram was further analyzed by the cortical stage , where the modulation content of the auditory spectrogram was computed through a bank of 2-dimensional filters selective for a combination of spectral and temporal modulations . The filter bank performs a complex wavelet decomposition of the auditory spectrogram . The magnitude of such decomposition yields a phase-invariant measure of modulation content . The modulation selective filters have joint selectivity for spectral and temporal modulations , and are directional , that is they respond either to upward or downward frequency sweeps . Filters were tuned to spectral modulation frequencies of Ω = [0 . 3 , 0 . 4 , 0 . 8 , 1 . 3 , 2 . 3 , 4] cyc/oct , temporal modulation frequencies of ω = [1 , 2 , 4 , 8 , 16 , 32] Hz , and centre frequencies of f = [232 , 367 , 580 , 918 , 1452 , 2297 , 3633 , 5746] Hz . Our rationale for this choice of values was to use a decomposition roughly covering the temporal and spectral modulations present in the acoustic energy of natural sounds we used ( for spectro-temporal modulation content of the sounds see Figure 1—figure supplement 2 ) . To avoid overfitting , for the decoding analyses in regions of interest ( ROIs ) , we reduced the number of features such that filters were tuned to spectral modulation frequencies of Ω = [0 . 25 , 0 . 5 , 1 , 2 , 4] cyc/oct , temporal modulation frequencies of ω = [1 , 2 . 4 , 5 . 7 , 13 . 5 , 32] Hz , and centre frequencies of f = [277 , 576 , 1201 , 2502 , 5213] Hz . The filter bank output was computed at each frequency along the tonotopic axis and then averaged over time . This resulted for the encoding ( ROI decoding ) analysis in a representation with 6 ( 5 ) spectral modulation frequencies × 6 ( 5 ) temporal modulation frequencies × 8 ( 5 ) tonotopic frequencies = 288 ( 125 ) parameters to learn . The time-averaged output of the filter bank was averaged across the upward and downward filter directions ( Santoro et al . , 2014 ) . Those processing steps were applied to all stimuli , resulting in an [N x F] feature matrix S of modulation energy , where N is the number of sounds , and F is the number of features in the modulation representation . Prior to acoustic feature extraction , the continuous sounds ( ~8 min per run ) were cut into snippets of the length of the TR ( 947 ms ) resulting in a total of n = 4152 sounds which were subdivided into training ( n = 3114 ) and test sounds ( n = 1038 ) for the cross-validation procedure ( see below ) . This resulted in an equivalent temporal resolution of the feature matrix and the fMRI data . Note that for the encoding analysis ( see below ) , the number of parameters to estimate is thus smaller than the number of observations in the training set ( n = 3114 training sounds ) . In the decoding analysis , the number of fitted features is limited by the number of voxels instead ( Equation 5 ) . Each feature was convolved with the standard double gamma model for the hemodynamic response function peaking at 4 s . We applied two modelling approaches to fMRI data as described in Erb et al . , 2019b; Santoro et al . , 2014; Santoro et al . , 2017 ( Figure 1 ) . In a first univariate encoding approach , we calculated a modulation transfer function ( MTF ) for each individual voxel . The MTF characterizes how faithfully the modulation content of the stimulus gets transferred to the voxel . By assigning the feature value with the maximal response to each voxel , we obtained maps of a voxel’s best features across the auditory cortex . In a second multivariate decoding approach , data from voxels were jointly modelled within a model-based decoding framework . The combined analysis of signals from multiple voxels increases the sensitivity for stimulus information that is represented in patterns of activity , rather than in individual voxels . Further , the accuracy with which those features can be reconstructed provides an explicit measure of the amount of information about sound features available in cortex . | It can often be difficult for an older person to understand what someone is saying , particularly in noisy environments . Exactly how and why this age-related change occurs is not clear , but it is thought that older individuals may become less able to tune in to certain features of sound . Newer tools are making it easier to study age-related changes in hearing in the brain . For example , functional magnetic resonance imaging ( fMRI ) can allow scientists to ‘see’ and measure how certain parts of the brain react to different features of sound . Using fMRI data , researchers can compare how younger and older people process speech . They can also track how speech processing in the brain changes with age . Now , Erb et al . show that older individuals have a harder time tuning into the rhythm of speech . In the experiments , 64 people between the ages of 18 to 78 were asked to listen to speech in a noisy setting while they underwent fMRI . The researchers then tested a computer model using the data . In the older individuals , the brain’s tuning to the timing or rhythm of speech was broader , while the younger participants were more able to finely tune into this feature of sound . The older a person was the less able their brain was to distinguish rhythms in speech , likely making it harder to understand what had been said . This hearing change likely occurs because brain cells become less specialised overtime , which can contribute to many kinds of age-related cognitive decline . This new information about why understanding speech becomes more difficult with age may help scientists develop better hearing aids that are individualised to a person’s specific needs . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"neuroscience"
] | 2020 | Temporal selectivity declines in the aging human auditory cortex |
During cortical synaptic development , thalamic axons must establish synaptic connections despite the presence of the more abundant intracortical projections . How thalamocortical synapses are formed and maintained in this competitive environment is unknown . Here , we show that astrocyte-secreted protein hevin is required for normal thalamocortical synaptic connectivity in the mouse cortex . Absence of hevin results in a profound , long-lasting reduction in thalamocortical synapses accompanied by a transient increase in intracortical excitatory connections . Three-dimensional reconstructions of cortical neurons from serial section electron microscopy ( ssEM ) revealed that , during early postnatal development , dendritic spines often receive multiple excitatory inputs . Immuno-EM and confocal analyses revealed that majority of the spines with multiple excitatory contacts ( SMECs ) receive simultaneous thalamic and cortical inputs . Proportion of SMECs diminishes as the brain develops , but SMECs remain abundant in Hevin-null mice . These findings reveal that , through secretion of hevin , astrocytes control an important developmental synaptic refinement process at dendritic spines .
The cerebral cortex receives synaptic inputs from various cortical and subcortical areas including the thalamus . In the mouse brain , innervation of the cortex by projecting neurites called axons begins during embryonic development and continues for the first several postnatal days ( Garel and Lopez-Bendito , 2014 ) . Only after the axons project to their approximate target areas , hosting their suitable postsynaptic partners , does an intense period of synapse formation occur , corresponding roughly to the second and third postnatal weeks in mice ( Li et al . , 2010 ) . Cortical excitatory synapses , which primarily use the neurotransmitter glutamate , are formed between dendritic protrusions called spines and axonal projections coming from two predominant inputs: intracortical and thalamic . Though the bulk of the cortical synapses from both of these inputs are made during the same early postnatal synaptogenic period ( P5–P21 ) ( Nakamura et al . , 2005 ) , whether they form through similar or differential mechanisms is unclear . Intracortical and thalamocortical connections can be distinguished as they primarily contain either vesicular glutamate transporter-1 ( VGlut1 ) or VGlut2 in their presynaptic terminals , respectively ( Kaneko and Fujiyama , 2002 ) . In most cortical areas , VGlut1-positive ( VGlut1+ ) intracortical projections greatly outnumber the VGlut2+ thalamic projections . The cellular and molecular mechanisms through which thalamocortical connections are established and maintained , despite steep competition from the vastly more abundant intracortical axons , have yet to be elucidated . Concurrently with the synaptogenic period , non-neuronal cells called astrocytes begin to populate the cortex , producing and secreting factors that promote synaptogenesis ( Eroglu and Barres , 2010 ) . For example , hevin ( a . k . a . synaptic cleft-1 or SPARC-like 1 ) is an astrocyte-secreted extracellular matrix protein that localizes to the clefts of excitatory synapses ( Johnston et al . , 1990; Lively et al . , 2007 ) and promotes excitatory synaptogenesis ( Kucukdereli et al . , 2011 ) . Using a combination of in vivo and in vitro approaches , here we show that in the cortex , hevin is specifically required for the formation of thalamocortical synapses . Moreover , using three-dimensional reconstructions of serial-section electron microscopy ( ssEM ) -imaged dendrites and axons , we show that P25 hevin KO dendritic spines often make multiple excitatory contacts with different axons . Multiply-innervated spines were identified nearly a half-century ago by Jones and Powell ( 1969 ) . The majority of these spines make two distinct types of synapses: one excitatory with an asymmetric postsynaptic density ( PSD ) opposed to an axon with round presynaptic vesicles , and one inhibitory synapse with a symmetrical PSD and flattened vesicles ( Jones and Powell , 1969; Knott et al . , 2002; Chen et al . , 2012 ) . A small percentage , however , make multiple excitatory contacts ( termed SMECs , i . e . , spines with multiple excitatory contacts ) . We found that SMEC structures are frequent earlier in development at P14 in WT but essentially disappear by P25 , indicating that they represent a transient stage in synaptic spine maturation . This developmental refinement is impaired in hevin KO mice . Moreover , using both confocal imaging as well as immunolabeling of electron micrographs with antibodies specific to VGlut1 and VGlut2 , we found that these SMECs often contact thalamic and cortical axons simultaneously . These results suggest that , during cortical synaptic development , dendritic spines serve as sites of competition between thalamic and cortical axons . Through secretion of hevin , astrocytes help maintain thalamic inputs onto cortical neurons and facilitate resolution of SMECs into singly-innervated spines .
The astrocyte-secreted synaptogenic protein hevin ( a . k . a . SPARC-like 1/SPARCL-1 or Synaptic Cleft-1/SC1 ) increases the number of synapses made between retinal ganglion cells ( RGCs ) in vitro and is required for the correct formation and maturation of RGC synapses onto their postsynaptic target , the superior colliculus ( Kucukdereli et al . , 2011 ) . Hevin expression is not restricted to the retinocollicular system . Hevin is expressed throughout the cortex in a developmentally regulated manner , peaking during P15–P25 , a time period that coincides with intense synapse formation , maturation and elimination events in the cortex ( Figure 1A , B ) . Staining for cell-specific markers confirms that hevin expression is largely restricted to astrocytes in the cortex ( Figure 1C–E ) . 10 . 7554/eLife . 04047 . 003Figure 1 . Hevin expression by astrocytes is developmentally regulated in the cortex . ( A ) Representative Western blots showing the developmental timeline for hevin expression in mouse cortex and hippocampus ( tubulin was used as a loading control ) . ( B ) Quantification of Western blot analysis of hevin expression shows high expression between P15–P25 . Data is presented as fold change compared to P1 levels ( n = 3 animals per age; p < 0 . 05; one-way ANOVA with Dunnett's post hoc test ) . ( C ) Schematic diagram of a coronal slice through mouse brain shows the synaptic zone of primary visual cortex ( V1 ) where EM , IHC and Golgi-cox staining analyses were performed . Layer II/III neurons of the visual cortex heavily project their dendrites to this region ( D ) IHC staining reveals that hevin expression ( green ) overlaps with all astrocytes ( left , arrow ) and a small subset of neurons ( middle , asterisk ) in V1 , with no overlap seen with microglia ( right , arrowhead ) . Cell-specific markers in red: Aldh1L1-EGFP for astrocytes , NeuN for neurons , Iba1 for microglia . Scale bar , 50 µm . ( E ) The rarely occurring GFAP+ astrocytes ( red ) in healthy visual cortex also express hevin ( green ) . Scale bar , 10 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 04047 . 003 In order to determine the role of hevin in cortical synaptic development , we investigated synaptic connectivity in the synaptic zone ( S/Z , a . k . a . Layer I ) of the mouse primary visual cortex ( V1 region ) in littermate age-matched hevin KO and wild-type ( WT ) mice using immunohistochemistry ( IHC ) . Excitatory pyramidal neurons from upper cortical layers project extensive dendritic trees to this region and form a large number of the cortical synaptic connections . Cortical neurons receive two main classes of excitatory inputs: ( 1 ) the intracortical connections that are VGlut1 positive ( VGlut1+ ) , and ( 2 ) the sensory pathway inputs from the thalamus that are VGlut2+ . The majority of the excitatory connections within the S/Z are of intracortical origin . On the other hand , the bulk of thalamocortical connections are made onto layer IV with a subset projecting to other layers including S/Z ( Figure 2A ) ( Kaneko and Fujiyama , 2002; Khan et al . , 2011 ) . To determine the role of hevin in the formation of these different classes of cortical synapses , we quantified the number of synaptic puncta as the co-localization of the presynaptic VGlut1 or VGlut2 with postsynaptic PSD95 at P25 in littermate hevin-KO and WT mice . This synapse quantification assay takes advantage of the fact that pre- and post-synaptic markers are within separate compartments ( axons and dendrites , respectively ) , but they appear to co-localize at synapses due to their close proximity ( Ippolito and Eroglu , 2010 ) . Surprisingly , we found that the number of VGlut1+ intracortical synapses in the S/Z of hevin KOs were significantly higher when compared to littermate WTs at P25 ( Figure 2B ) . In contrast , thalamocortical VGlut2+ synapses were profoundly reduced in hevin KOs compared to WTs at P25 ( Figure 2C ) . It is important to note that the appearance of co-localized VGlut/PSD95 puncta is not merely due to chance , since the randomization of puncta by rotating the channels out of alignment by 90° nearly abolished occurrence of co-localized puncta in both the WT and KO ( Figure 2—figure supplement 1 ) . 10 . 7554/eLife . 04047 . 004Figure 2 . Hevin is required for proper thalamocortical innervation of V1 . ( A ) Schematic of synaptic input to area V1 . Most inputs are intracortical ( green ) and contain primarily VGlut1 . The thalamus sends VGlut2+ projections ( red ) to various layers of V1 ( primarily Layer IV ) . Competition occurs between VGlut1+ and VGlut2+ terminals for the same postsynaptic targets . ( B ) Co-localization of VGlut1 ( green ) and PSD95 ( magenta ) revealed an increase in intracortical synapses ( co-localized puncta; arrows ) in the synaptic zone of P25 hevin KO V1 ( n = 3 z-stacks per animal , 5 animals per genotype; p < 0 . 05 , nested ANOVA ) . Scale bar , 5 µm . ( C ) VGlut2 ( red ) and PSD95 ( cyan ) co-localization ( arrows ) , representing thalamocortical synapses , was significantly decreased in P25 hevin KO compared to P25 WT ( n = 3 z-stacks per animal , 4 animals per genotype; p < 0 . 05 , nested ANOVA ) . Scale bar , 5 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 04047 . 00410 . 7554/eLife . 04047 . 005Figure 2—figure supplement 1 . Co-localized synaptic puncta are not due to random chance resulting from dense synaptic staining . Confocal Z-stacks were split into two channels ( pre- and post-synaptic ) , the presynaptic channel was rotated 90° out of alignment , then the two channels were re-merged and analyzed for co-localized synaptic puncta . For both VGlut1/PSD95 ( left ) and VGlut2/PSD95 ( right ) co-localized puncta , puncta density was significantly decreased in the rotated image ( red ) compared to the original unrotated image ( black ) in both the WT and hevin KO mice ( n = 3 z-stacks per animal , 4 animals per genotype; p < 0 . 01 , Student's t test ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04047 . 00510 . 7554/eLife . 04047 . 006Figure 2—figure supplement 2 . Deficient thalamocortical connectivity in hevin KO is not due to decreased cortical neuron density . ( A ) Left: V1 stained for neuronal marker NeuN ( red ) and DAPI ( green ) revealed no differences in the gross morphology of cortical layers between P25 WT and hevin KO . Scale bar , 50 µm . Right: Quantification of neuronal density showed no significant differences throughout the cortices of P25 WT vs Hevin KO ( n = 3 sections per animal , 3 animals per genotype; p > 0 . 05 , ANCOVA ) . ( B ) Top: Staining for VGlut2+ thalamic projections showed decreased intensity across multiple cortical layers , including S/Z , Layer II/III and Layer IV in hevin KO V1 . White *=pial surface . Scale bar , 50 µm . Bottom: Quantification of VGlut2 pixel intensity for P25 WT and hevin KO across cortical layers I–IV ( *p < 0 . 05 , ANCOVA ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04047 . 00610 . 7554/eLife . 04047 . 007Figure 2—figure supplement 3 . Reduced thalamocortical synapse density in hevin KO V1 is not due to deficient geniculocortical connectivity . ( A ) Left: Schematic diagram of a coronal slice through mouse brain highlights the dorsal lateral geniculate nucleus ( dLGN ) which sends VGlut2+ projections to V1 . Right: NeuN ( red ) staining reveals no difference in neuronal density between P25 WT and hevin KO dLGN ( WT: 14 . 52±1 . 10 cells per 10 , 000 μm2 , KO: 14 . 46±1 . 34 cells per 10 , 000 μm2; n = 3–4 slices per animal , 3 animals per genotype; p = 0 . 94 , Student's t test ) . Scale bar , 50 µm . ( B ) Left: Diagram of the AAV-FLEX-GFP viral vector . The portion of the vector encoding for GFP is inverted , preventing its expression . Upon Cre recombination , the flanking lox sites are excised and the GFP region is flipped , resulting in GFP expression . Right: The AAV-FLEX-GFP virus ( green ) was injected into the dLGN to label projection neurons . Rabies virus glycoprotein-coated Lenti-FuGB2-Cre ( brown ) was injected into V1 that retrogradely transports Cre recombinase expression to the dLGN . In this dual-injection system , only neuronal projections that originate in dLGN and terminate in V1 undergo Cre recombination and thereby express GFP . ( C ) Representative images showing GFP+ projection neurons in the dLGN of both the WT and hevin KO at P30 ( approximately 12 days after viral injection ) . ( D ) Images of GFP+ projections in the S/Z of both WT and hevin KO show that lack of hevin does not impair the ability of thalamic axons to reach their target regions . Scale bars: 1 mm ( main images ) ; 40 µm ( insets ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04047 . 00710 . 7554/eLife . 04047 . 008Figure 2—figure supplement 4 . Electrophysiological analysis of V1 cortical neurons in hevin KO . ( A ) Representative traces from mEPSC recordings in Layer II/III neurons of V1 from WT and hevin KO . Scale bars: 20 pA and 100 ms . ( B ) Frequency of mEPSC was not significantly different in P23-P26 hevin KO mice compared to age-matched WT ( n = 13 cells from 3 WT mice , 18 cells from 5 hevin KO mice; p > 0 . 05 , Student's t test ) . ( C ) Amplitude of mEPSC was not significantly different in P23–P26 hevin KO mice compared to age-matched WT ( n = 13 cells from 3 WT mice , 18 cells from 5 hevin KO mice; p > 0 . 05 , Mann–Whitney U test ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04047 . 008 The severe loss of thalamocortical synapses in hevin KOs was not due to a gross cellular defect in hevin KO cortices , because we found that neuronal density and layering was similar between P25 WT and hevin KO ( Figure 2—figure supplement 2A ) . However , we observed a significant reduction in VGlut2+ synaptic terminals across multiple cortical layers in the hevin KOs ( Figure 2—figure supplement 2B ) . This reduction in thalamic synaptic terminals was not due to the lack of thalamic neurons in hevin KOs at the dorsal lateral geniculate nucleus ( dLGN ) that project to V1 ( Figure 2—figure supplement 3A ) . Moreover , by using a viral approach to trace the thalamic axons that project to V1 cortex , we found that lack of hevin did not impair the ability of thalamic projections to reach the S/Z ( Figure 2—figure supplement 3B–D ) . Hevin expression in the cortex starts to increase by the end of first postnatal week ( corresponding to the start of synaptogenic period ) , peaking in abundance during the critical periods of plasticity in the cortex; hevin levels also remain high in the adult ( Figure 1B ) . Therefore , we next investigated whether loss of hevin also alters synapse numbers in early postnatal ( P7 ) or adult ( 12-week-old ) mice . The number of VGlut1+ synapses in P7 hevin KOs trended towards an increase when compared to littermate WTs , but this increase was not yet significant ( Figure 3B ) . However , similar to P25 , P7 hevin KO cortex showed a severe deficit in VGlut2+ synapses ( Figure 3C ) . This finding indicates that hevin is required for the formation of thalamocortical synapses in the early developing mouse brain . 10 . 7554/eLife . 04047 . 009Figure 3 . Hevin is important for both the formation and long-term maintenance of thalamocortical synapses . ( A ) Timeline of cortical synaptic development in mice . Axonal projections from the LGN reach their target areas in V1 shortly after birth . Around the time of eye opening , there is a period of intense synapse formation that gradually gives way to processes involved in synapse maturation and refinement , including synapse elimination . Multiple critical periods for different forms of plasticity in the visual system occur during this period of synapse formation and refinement . ( B ) At P7 , the beginning of the synaptogenic period , co-localization of VGlut1 ( green ) and PSD95 ( magenta ) revealed a trend towards an increase in intracortical synapses ( co-localized puncta; arrows ) in the synaptic zone of hevin KO V1 ( n = 3 z-stacks per animal , 3 animals per genotype; p = 0 . 082 , nested ANOVA ) . ( C ) VGlut2 ( red ) and PSD95 ( cyan ) co-localization ( arrows ) , representing thalamocortical synapses , was significantly decreased in P7 hevin KO compared to P7 WT ( n = 3 z-stacks per animal , 3 animals per genotype; p < 0 . 05 , nested ANOVA ) . Scale bars , 5 µm . ( D ) In the mature brain ( 12-weeks-old ) , hevin KO mice no longer have a discrepancy in VGlut1/PSD95 synaptic puncta when compared to WT ( n = 3 z-stacks per animal , 5 animals per genotype; p > 0 . 05 , Student's t test ) . ( E ) Deficient VGlut2/PSD95 synapse formation is still present in the mature hevin KO brains ( n = 3 z-stacks per animal , 5 animals per genotype; p < 0 . 01 , Student's t test ) . Scale bars , 5 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 04047 . 009 The specific loss of thalamocortical synaptic connectivity in hevin KOs was not merely a developmental delay in thalamocortical synaptogenesis , since VGlut2+ synapse density was still significantly lower in the 12-week old hevin KO adults compared to WT ( Figure 3E ) . However , the number of VGlut1+ synapses in the hevin KOs returned to normal levels in the adult ( Figure 3D ) , indicating that the changes we observed in the VGlut1+ synaptic connectivity in the developing ( P25 ) hevin KOs are due to a transient offsetting of the reduced VGlut2+ connections by VGlut1+ intracortical synapses . In agreement with this possibility , recordings from P23–P26 mice did not show any significant differences in miniature excitatory postsynaptic currents ( mEPSCs ) in layer 2/3 pyramidal neurons of hevin KOs when compared to their littermate WTs ( Figure 2—figure supplement 4 ) . Taken together , these results show that hevin is required for normal formation and maintenance of VGlut2+ thalamocortical connections in the cortex . Our findings also demonstrate that lack of hevin results in a transient increase in intracortical synapses . This increase in intracortical synapses could either be mediated through a homeostatic mechanism that compensates for lost thalamic input and/or be driven by a transient competitive advantage for cortical axons over the thalamic inputs to establish synapses . Our analyses of excitatory synaptic development in hevin KO mice revealed an important role for hevin in the development of thalamocortical circuitry ( Figures 2–3 ) . Therefore , we next investigated whether hevin is sufficient to induce thalamocortical synaptogenesis . To address this question we first utilized in vitro assays with purified cortical and thalamic neurons . To do so we immunopurified cortical neurons from hevin KO pups and plated them either alone or in the presence of equal number of purified thalamic neurons ( Figure 4A ) . Next we treated these cells with growth media with or without hevin and determined the effect of hevin treatment on the number of VGlut1+ or VGlut2+ excitatory synapses made onto cortical neurons ( Figure 4B–E ) . Hevin treatment did not increase the number of VGlut1+ synaptic puncta ( determined as the co-localization of VGlut1 and PSD95 ) in either the cortical neuron only cultures or in the cortical/thalamic neuron co-cultures ( Figure 4B , C ) . Hevin also did not significantly affect VGlut2/PSD95-positive synapses in the cortical neuron-only cultures ( Figure 4D , E ) , which were already at low quantities due to the fact that only a small portion of cortical axons expresses VGlut2 ( Wallen-Mackenzie et al . , 2009 ) . However , hevin treatment significantly increased VGlut2+ synapse formation onto cortical neurons in cortical-thalamic co-cultures ( Figure 4D , E ) . These in vitro evidence strongly suggest that hevin specifically induces formation of thalamocortical synapses . 10 . 7554/eLife . 04047 . 010Figure 4 . Hevin specifically induces thalamocortical synapse formation in vitro and in vivo . ( A ) Schematic of the cortical/thalamic neuron co-culture system . ( B ) Representative images of cortical neurons cultured for 14 days in vitro ( DIV ) , either alone or in equal densities with thalamic neurons , with or without hevin treatment . Insets show individual channels for VGlut1 ( green ) and PSD95 ( magenta ) staining , as well as the merged image . Co-localized puncta ( white , arrows ) represent synapses . Scale bars: 10 µm ( main image ) , 5 µm ( inset ) . ( C ) Hevin did not induce VGlut1 synapse formation onto cortical neurons , with or without thalamic neurons also present ( n = 30 cells per condition; p > 0 . 05 , Student's t test between GM and hevin treatments ) . ( D ) Same as B , except VGlut2 appears as red and PSD95 in cyan . ( E ) Hevin strongly induces VGlut2/PSD95 synapse formation when cortical neurons are cultured together with thalamic neurons ( n = 30 cells per condition; *p > 0 . 01 , Student's t test between GM and hevin treatments ) . Upon hevin treatment VGlut2/PSD95 synapses are recruited heavily to neuronal soma and proximal dendrites . ( F ) Hevin protein was stereotactically injected directly into Layer II/III of hevin KO V1 . When compared to vehicle-injected control , hevin-injected cortex had distinctly thickened VGlut2 staining throughout S/Z and upper II/III ( asterisk ) , as well as a dense appearance of VGlut2+ axon tracks throughout II/III ( arrow ) . Scale bar , 100 µm . ( G ) Hevin-injected V1 had more VGlut2/PSD95 co-localized puncta ( arrowheads ) than vehicle-injected controls . Scale bar , 5 µm . ( H ) Hevin injection significantly increased the number of VGlut2/PSD95 synapses in hevin KO cortex ( n = 2 z-stacks per animal , 3 animals per treatment; p < 0 . 01 , Student's t test ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04047 . 01010 . 7554/eLife . 04047 . 011Figure 4—figure supplement 1 . Hevin does not induce intracortical synapse formation in vivo . Top: Synaptic staining in V1 does not show a difference in VGlut1/PSD95 co-localization ( arrowheads ) between hevin-injected brains and vehicle-injected controls . Scale bar , 5 µm . Bottom: Quantification of VGlut1/PSD95 synapses in hevin-injected cortex confirms that hevin does not increase the number of VGlut1+ synapses ( n = 2 z-stacks per animal , 3 animals per treatment; p > 0 . 05 , Student's t test ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04047 . 011 Virally tracing the thalamocortical axons that innervate the S/Z in hevin KOs revealed that thalamocortical projections are intact in the hevin KO ( Figure 2—figure supplement 3B–D ) , but they do have defects in establishing thalamocortical synapses . Therefore , we next tested whether injection of hevin into the developing cortex was sufficient to increase thalamocortical synaptic connectivity in vivo . To do so , pure hevin protein was directly injected into Layer II/III of P13 hevin KO V1 ( Figure 4F ) . After 3 days , the brains were fixed and immuno-stained for pre- and postsynaptic markers and then imaged by confocal microscopy approximately 100 µm laterally to the injection site ( Figure 4F ) . Compared to vehicle-injected littermate controls , hevin-injected cortices showed a robust increase in VGlut2+ thalamocortical terminal staining throughout the S/Z and Layer II/III directly adjacent to the site of injection ( Figure 4F , asterisk and arrow , respectively ) . Quantitative analysis of co-localized VGlut2/PSD95 synaptic puncta in the S/Z also revealed a significant increase in thalamocortical synapses in hevin-injected cortices compared to the vehicle-injected littermates ( Figure 4G , H ) . By contrast , the number of co-localized VGlut1/PSD95 synaptic puncta was not affected by hevin injection ( Figure 4—figure supplement 1 ) . Combined with our in vitro data , these in vivo experiments demonstrate that hevin specifically induces thalamocortical synaptic connectivity without affecting intracortical connectivity . In the cortex , the majority of excitatory synaptic contacts are compartmentalized onto submicron structures called dendritic spines ( Harris and Kater , 1994 ) . Cortical dendritic spines , including those in V1 , follow a stereotypic maturation timeline ( Irwin et al . , 2001 ) . Long , highly motile filopodia-type protrusions abundant in early development give way to short , stable , wide-headed mushroom spines in the mature brain ( Figure 5—figure supplement 1A ) ( Kaneko et al . , 2012 ) . The spine maturation timeline coincides to a large extent with the expression of hevin protein , which reaches its highest levels at P25 ( Figure 1A , B ) . This observation prompted the following question: Does the deficient thalamocortical connectivity in hevin KO coincide with aberrant synaptic morphology ? In order to address this question , we investigated whether hevin is involved in the structural development of spine synapses . To do so , we analyzed dendritic morphology in the secondary and tertiary dendrites of layer II/III pyramidal neurons , which receive the majority of the excitatory connections within the S/Z . Analyses of spines in V1 of littermate hevin KO and WT mice at P25 by Golgi-Cox staining showed a significant increase in immature filopodia-like protrusions in P25 hevin KOs concomitant with a decrease in mature mushroom spines compared to littermate WT controls ( Figure 5—figure supplement 1B ) . There was no significant difference in total protrusion density between genotypes ( WT , 1 . 12 ± 0 . 03 spines/µm; KO , 1 . 07 ± 0 . 03 spines/µm; n = 45 dendrites per condition; p > 0 . 05 , Student's t test ) . The dendritic arborization of layer II/III neurons was also similar between hevin KO and WT mice , showing that lack of hevin does not lead to overt problems in dendritic morphology ( Figure 5—figure supplement 1C ) . These results indicated that the astrocyte-secreted synaptogenic protein hevin is important for spine maturation in the cortex . To understand the role of hevin in dendritic spine maturation at ultra-high resolution , we next employed ssEM in littermate P25 WT and hevin KO mice ( Kuwajima et al . , 2013 ) . Three-dimensional ( 3D ) -EM reconstructions , visualizing dendrites , spines and synapses , confirmed the structural immaturity of hevin KO dendrites ( Figure 5A ) . Analysis of postsynaptic density ( PSD ) area revealed smaller , asymmetric ( i . e . , excitatory ) synapse area in hevin KO V1 ( Figure 5B ) , indicative of synaptic immaturity . Despite the deficits in synapse morphology , overall asymmetric synapse density was not significantly affected in hevin KO V1 ( WT , 2 . 16 ± 0 . 16 synapses/µm; KO , 2 . 63 ± 0 . 27 synapses/µm; n = 12 dendrites per condition; p = 0 . 15 , Student's t test ) . Since hevin is primarily expressed and secreted by astrocytes , we postulated that hevin KOs may have altered astroglial contact at synapses , but no difference in astrocyte contact was found between WT and hevin KO synapses ( Figure 5C ) . Interestingly , a fraction of excitatory synapses in the hevin KOs was made directly onto the dendritic shafts rather than on spines ( Figure 5D ) , a configuration that was rare in the WTs at P25 . This observation reveals that hevin is required for the proper compartmentalization of excitatory synapses onto spines . Taken together , these findings show that hevin function is required for the proper maturation and localization of excitatory synapses in the cortex . 10 . 7554/eLife . 04047 . 012Figure 5 . Hevin is required for dendritic maturation and proper localization of excitatory synapses . ( A ) Example 3D reconstructions of dendrites in P25 WT ( left , blue ) and hevin KO ( right , pink ) V1 . Asymmetric PSD locations ( i . e . , excitatory synapses ) are shown in red . Scale cubes , 0 . 5 µm3 . ( B ) Asymmetric PSD size was decreased in P25 hevin KO vs WT ( n = 278 WT synapses , 293 KO synapses; p < 0 . 025 , Kolmogorov–Smirnov two-sample test ) . ( C ) Left: EM examples from P25 WT and P25 hevin KO show the ‘tripartite synapse’ arrangement of postsynaptic dendritic spines ( yellow ) , presynaptic axonal boutons ( green ) and glial processes ( blue ) . Scale bar , 250 nm . Right: Quantification revealed no difference in the percentage of excitatory synapses contacted by glial processes in P25 WT vs P25 hevin KO ( n = 4 dendrites per animal , 2 animals per genotype; p = 0 . 52 , Student's t test ) . ( D ) Left: Excitatory synapses , made by axons ( green ) onto dendritic spines ( yellow ) in P25 WT , were readily seen on dendritic shafts ( yellow ) in the hevin KO . Scale bar , 250 nm . Right , Top: Example hevin KO dendrite with multiple excitatory shaft synapses ( arrows ) . Scale cube , 0 . 5 µm3 . Right , Bottom: Quantification of excitatory shaft synapse density in P25 hevin KO compared to P25 WT ( n = 4 dendrites per animal , 3 animals per genotype; p < 0 . 01 , Student's t test ) . ( E ) SMEC density was increased in P25 hevin KO compared to WT ( arrows indicate excitatory PSDs on SMECs; n = 4 dendrites per animal , 3 animals per genotype; p < 0 . 01 , one-way ANOVA with Fisher's LSD posthoc test ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04047 . 01210 . 7554/eLife . 04047 . 013Figure 5—figure supplement 1 . Structural immaturity across multiple spine types in hevin KO V1 . ( A ) Multiple spine types can be found in developing visual cortex , progressing in maturity from the long , thin filopodia-type protrusions to the enlarged-head mushroom spines ( and occasional branched spines ) . ( B ) Top: Golgi-cox stained secondary dendrites from Layer II/III pyramidal neurons in V1 . Scale bar , 5 µm . Immature filopodia-like spines were frequent in hevin KOs ( red arrowheads ) whereas mature mushroom spines ( blue arrows ) were seen in P25 WT . Bottom: Quantification of spine type densities revealed that immature spine types , including filopodia and long thin spines , were more frequent in the P25 hevin KO compared to WT . More intermediate-to-mature spine types , including short thin , stubby and mushroom , were less frequent in the P25 hevin KO compared to WT ( n = 15 dendrites per animal , 3 animals per genotype; p < 0 . 01 , Student's t test ) . ( C ) Top: Camera lucida images of representative Layer II/III pyramidal neurons in V1 from P25 WT and hevin KO . Bottom: Scholl analysis ( p = 0 . 70 ) and neurite outgrowth ( p = 0 . 26 ) analyses of Golgi-cox stained pyramidal neurons showed no significant differences between P25 WT and hevin KO ( n = 4 neurons per animal , 3 animals per genotype; ANCOVA ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04047 . 013 Our 3D analyses revealed that , in addition to the above-mentioned structural deficits in the excitatory synapses , a considerable number of dendritic spines receive more than one excitatory synapse in hevin KOs ( Figure 5E ) . These SMECs ( Spines with Multiple Excitatory Contacts ) are distinctly different from branched spines , in which multiple spine heads are connected to the same spine neck ( Kirov et al . , 1999 ) . Furthermore , SMECs should not be confused with multisynaptic boutons ( MSBs ) , where a single presynaptic axonal bouton makes contact with multiple dendritic spines ( Kirov et al . , 1999 ) . SMEC density was significantly higher in P25 hevin KO mice compared to WT ( Figure 5E ) . Because hevin KO dendrites displayed other signs of immaturity , we postulated that SMECs may represent an earlier stage in excitatory synapse maturation . To investigate if SMECs occur in the context of normal synaptic development , we created ssEM-derived 3D reconstructions of dendrites in the synaptic zone of WT V1 at P14 , an age when dendritic spine structures are not yet fully mature . Electron micrographs revealed the existence of SMECs in P14 V1 in which a single postsynaptic spine contained more than one asymmetric PSD ( Figure 6A ) . 3D reconstructions from ssEM confirmed that each PSD on a SMEC was contacted by a different presynaptic axon ( Figure 6B ) . This ruled out SMECs as having either a single perforated PSD or multiple PSDs opposed to the same axon . Several configurations of SMECs were detected; some in which two axons synapsed on opposite sides of the same spine head ( Figure 6B , left ) , and others with one PSD on the head and a second PSD on either the neck or base of the spine ( Figure 6B , right ) . SMECs were primarily of the thin spine type , though we also found numerous examples of filopodia and mushroom SMECs . Remarkably , 25% of all excitatory connections are formed onto SMECs at P14 , a finding that may have gone unnoticed if not for the spatial resolution offered by 3D ssEM . The prevalence of SMECs is largely decreased by P25 ( Figure 6C , D ) , indicating that SMECs represent a transient stage in excitatory synaptic maturation . 10 . 7554/eLife . 04047 . 014Figure 6 . Spines with Multiple Excitatory Contacts ( SMECs ) represent a developmental stage in the maturation of dendritic spine structures . ( A ) Electron micrograph of a SMEC: a dendritic mushroom spine ( yellow ) making asymmetric contacts ( red arrows ) with two different axonal boutons ( green ) . Scale bar , 0 . 5 µm . ( B ) Different spatial arrangements observed in SMECs . Small green circles denote the location of glutamatergic vesicles within the axons . ( C ) SMECs ( arrows ) decrease from P14–P25 in WT . ( D ) Quantification of the percentage of excitatory synapses made onto SMECs in P14 WT and P25 WT mice ( n = 4 dendrites per animal , 3 animals per age; p < 0 . 05 , one-way ANOVA with Fisher's LSD posthoc test ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04047 . 014 SMECs are a transient structure observed during normal synaptic development , but what purpose do they serve ? When multiple axons are contacting a single spine , they will have to share the postsynaptic machinery available within that spine . Such a configuration would potentially provide means to drive competition between neighboring inputs for postsynaptic resources , and the activity levels of the presynaptic axons contacting that spine could directly influence this competition . To determine whether specific axonal populations were contacting SMECs , we completed full 3D reconstructions of axons contacting SMECs in P25 hevin KOs and P14 and P25 WTs . We found that if an axon made a connection with a SMEC , it had nearly a 50% chance of contacting another SMEC nearby ( 42 . 5% in P14 WT , 45 . 4% in P25 WT , 41 . 6% in P25 hevin KO; Figure 7A , Video 1 ) , which is vastly higher than what would be expected by chance ( 6 . 4% in P14 WT , 0 . 5% in P25 WT , 3 . 1% in P25 hevin KO ) . This specific preference of certain axons for SMECs suggested that SMECs are targeted by particular subpopulations of axons which may be in competition with their neighbors for common postsynaptic spines . 10 . 7554/eLife . 04047 . 015Figure 7 . SMECs represent potential sites for competition between thalamocortical and intracortical projections . ( A ) Reconstructed axon ( green ) from P25 hevin KO contacting multiple SMECs on two different dendrites . Quantification revealed that axons that synapsed with at least one SMEC also synapsed with another SMEC roughly 50% of the time . Scale cube , 0 . 5 µm3 . i–ii , Zoomed-in images reveal that each SMEC makes an excitatory synapse ( red; arrow ) with the reconstructed axon as well as a second excitatory synapse ( yellow; arrowhead ) with an additional axon ( not shown ) . ( B ) Immuno-EM image from V1 in P14 WT showing a SMEC simultaneously contacting a VGlut1+ ( green ) and VGlut2+ ( red ) axonal bouton . Higher magnification images ( below ) highlight the size difference between the small VGlut1 ( green arrowheads ) and large VGlut2 ( red arrowheads ) Nanogold particles . Scale bar , 250 nm . DOI: http://dx . doi . org/10 . 7554/eLife . 04047 . 01510 . 7554/eLife . 04047 . 016Video 1 . An axon ( green ) from Layer II/III V1 reconstructed in 3D from ssEM images of P25 hevin KO . SMEC 1 from Dendrite 1 ( blue ) makes Synapse 1 ( red ) with this axon and Synapse 2 ( yellow ) with a second unshown axon . On a different part of this same axon , it makes Synapse 3 ( red ) with SMEC 2 from Dendrite 2 ( pink ) that also contacts a third ( also unshown ) axon at Synapse 4 ( yellow ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04047 . 016 We found that SMECs are still abundant in P25 hevin KOs ( Figure 5E ) , and in hevin KOs the number of VGlut2 synapses are reduced whereas VGlut1 synapses are increased at this age ( Figure 2B , C ) . Therefore , we postulated that SMECs may be sites for simultaneous innervation by VGlut1+ cortical axons and the VGlut2+ thalamic projections ( Figure 2A ) . To provide evidence for this possibility , we next performed immuno-EM analysis using VGlut1 and VGlut2-specific antibodies on conventional 2D-EM sections from P14 WT mice and observed that a SMEC can indeed receive simultaneous cortical ( VGlut1+ ) and thalamic ( VGlut2+ ) inputs ( Figure 7B ) . It was previously shown that , during early cortical development , VGlut2+ thalamic inputs establish numerous contacts within the S/Z; however , as the cortex develops , the majority of these connections are pruned and VGlut1+ inputs dominate this region ( Miyazaki et al . , 2003 ) . During this period , confocal microscopy images of VGlut1+ and VGlut2+ terminals first appear to frequently overlap , but this overlap resolves over time ( Nakamura et al . , 2007 ) . Based on our findings , we postulated that the close positioning of VGlut1+ and VGlut2+ terminals at the same SMEC might yield an apparent co-localization between these two presynaptic markers in conventional light microscopy . If that is the case we expected hevin to affect the segregation of VGlut1 and VGlut2 terminals during cortical development . To test this possibility , we used IHC to compare the apparent co-localization of the two VGluts in the S/Z of V1 in WTs and hevin KOs at P15 and P25 . We found that , in agreement with earlier findings from other brain regions and layers ( Herzog et al . , 2006; Nakamura et al . , 2007 ) , 10 . 97% of VGlut terminals appeared to co-localize in P15 WT , which was significantly reduced in P25 WT ( Figure 8—figure supplement 1A , B ) . In hevin KOs , co-localization of VGlut1/VGlut2 was originally lower than WT at P15 , perhaps due to the overall loss of VGlut2+ synapses . Interestingly , the frequency of overlap in the P25 hevin KO was essentially unchanged compared to P15 and was significantly higher than the P25 WT ( Figure 8—figure supplement 1A , B ) . These results suggest that hevin is important for the resolution of the VGlut1/VGlut2 overlap ( and , potentially , SMECs ) . Analysis of rotated images confirmed that the observed VGlut1/VGlut2 overlap by confocal microscopy is not due to random chance ( Figure 8—figure supplement 1C ) . The VGlut1/VGlut2 overlap has previously been attributed to temporary co-expression of these proteins in the same synaptic terminals ( Nakamura et al . , 2007 ) . To test if this co-localization is mainly due to the resolution limit of light microscopy rather than the expression of these two VGluts at the same terminal , we used high-resolution structured illumination microscopy ( SIM ) imaging of presynaptic puncta in the synaptic zones of P15 WT mice ( since the VGlut1/2 overlap is highest at this age ) . SIM uses constructive and deconstructive interference of excitation light at the focal plane of the objective to illuminate a sample with a series of sinusoidal stripes; from the resulting moiré fringes it is possible to generate super-resolution data and produce a minimum two-fold improvement in resolution over confocal microscopy ( Schermelleh et al . , 2010 ) . Indeed , SIM imaging of VGlut1 and VGlut2 puncta in P15 WT showed virtually no co-localized presynaptic puncta ( Figure 8—figure supplement 1D ) . By increasing the maximum distance with which to detect co-localized puncta , SIM analysis eventually reached a point at which VGlut1/VGlut2 overlap approached the level observed in P15 WT by confocal imaging; this occurred near the practical resolution of confocal microscopy ( Figure 8—figure supplement 1E ) . This provided strong evidence that the overlap of VGlut1/VGlut2 puncta we observed with confocal microscopy ( Figure 8—figure supplement 1A , B ) was due to adjacent presynaptic puncta rather than co-expression in the same terminal . Despite the knowledge that SMECs are increased in the hevin KO ( Figure 5E ) and can receive simultaneous VGlut1/VGlut2 synaptic inputs ( Figure 7B ) , the vast majority of inputs to hevin KO cortex remain VGlut1-positive ( Figure 2B ) . Therefore , the possibility arises that most SMECs actually receive inputs from multiple intracortical axons , rather than existing as sites for simultaneous thalamic and cortical innervation . Taking advantage of the finding that co-localized VGlut1/VGlut2 puncta in light microscopy is predominantly due to close expression by different presynaptic terminals ( Figure 8—figure supplement 1D , E ) , we next imaged presynaptic puncta in close proximity to dendritic spines with confocal microscopy in order to quantify the types of inputs made onto SMECs . Using an in utero electroporation ( IUE ) approach at embryonic day 15 . 5 ( E15 . 5 ) we specifically labeled cortical layer II/III neurons with green fluorescent protein ( GFP ) ( Figure 8—figure supplement 2A , B ) . We then harvested brains at specific developmental ages and co-stained for VGlut1 and VGlut2 . S/Z dendrites and presynaptic puncta were imaged via confocal Z-stacks and reconstructed in 3D with the Imaris processing package ( Figure 8—figure supplement 2C , D ) . Presynaptic puncta in close proximity to the dendrite were identified using a Matlab algorithm embedded in Imaris ( see ‘Materials and methods’ ) . This analysis method was first confirmed in P15 WT , an age when SMECs are still prevalent . Unisynaptic spines ( i . e . , spines contacting only one presynaptic puncta; either VGlut1 or VGlut2 ) and SMECs in various configurations were successfully detected ( Figure 8—figure supplement 2D , E ) . In accordance with a role for SMECs in synaptic competition during development , the majority of SMECs were innervated by both VGlut1 and VGlut2 inputs at this age , with fewer examples of VGlut1/VGlut1 and VGlut2/VGlut2 SMECs ( Figure 8—figure supplement 2E ) . Using the same parameters , we then quantified the difference in presynaptic innervation of SMECs between WT and hevin KO V1 at P21 . In addition to confirming the overall increase in SMECs in the hevin KO , the 3D reconstructions revealed that the majority of these SMECs indeed received mixed VGlut1/VGlut2 inputs ( Figure 8A , B ) , rather than having multiple VGlut1 inputs as the overall presynaptic puncta density would suggest . Taken together , these findings reveal that SMECs are typically sites of contact by cortical and thalamic axons , potentially representing a novel paradigm for synaptic competition between these two main sources of presynaptic input onto cortical spines . Our results indicate that resolution of SMECs into single synapse spines is a developmental process that is regulated by astrocytes through secretion of hevin . 10 . 7554/eLife . 04047 . 017Figure 8 . Hevin is critical for the resolution of VGlut1/VGlut2-innervated SMECs . ( A ) Representative Imaris 3D reconstructions of GFP-labeled dendrites ( blue ) in the S/Z of P21 WT and hevin KO V1 . Presynaptic puncta are rendered as elliptical ‘spots’ ( green = VGlut1; red = VGlut2 ) . Numerous unisynaptic spines can be seen in the WT ( arrow = VGlut1 spine; arrowheads = VGlut2 spines ) . In hevin KO , a SMEC can be seen contacting both a VGlut1 and VGlut2 ‘spot’ ( asterisk ) . A unisynaptic VGlut1 spine ( arrow ) is also present . Scale bar , 0 . 5 µm . ( B ) Quantification of SMEC density at P21 shows that the increase in total SMECs in the hevin KO is driven by the VGlut1/VGlut2 SMEC subtype ( 3 animals/genotype , n = 15 dendrites per condition; p < 0 . 01 , Student's t test ) . ( C ) Model for astrocytic control of thalamocortical connectivity by hevin . Left: In early V1 synaptic development , intracortical ( primarily VGlut1 , or VG1 ) axons compete with thalamocortical ( primarily VGlut2 , or VG2 ) axons for synapses , occasionally forming synapses on the same dendritic spine ( resulting in a SMEC ) . In the WT , astrocytes secrete hevin which stabilizes VGlut2+ synapses , resulting in discrete populations of VGlut1+ and VGlut2+ unisynaptic spines . In the hevin KO , VGlut2+ synapses cannot be properly stabilized . These sites either remain in competition with VGlut1 , explaining the persistence of SMECs in hevin KO , or become lost , resulting in more VGlut1+ synapses overall . DOI: http://dx . doi . org/10 . 7554/eLife . 04047 . 01710 . 7554/eLife . 04047 . 018Figure 8—figure supplement 1 . Overlap of VGlut1 and VGlut2 in light microscopy is due to close proximity of different presynaptic terminals . ( A ) Apparent overlap between VGlut1 ( green ) and VGlut2 ( red ) by light microscopy appears as yellow puncta ( arrows ) . Representative images show VGlut puncta at P15 and P25 in both WT and hevin KO . Scale bar , 5 µm . ( B ) Quantification of the apparent overlap in VGlut1/VGlut2 puncta , which is decreased between P15 and P25 in the WT . In the hevin KO , the VGlut1/VGlut2 overlap is slightly decreased at P15 but is unchanged as development continues to P25 , at which point it is significantly higher than WT ( n = 3 z-stacks per animal , 3–4 animals per genotype; *p < 0 . 01 , **p < 0 . 05 , nested ANOVA ) . ( C ) To show that the apparent overlap in VGlut1 and VGlut2 puncta was not due to random chance in densely stained tissue , confocal Z-stacks were split into two channels ( VGlut1 and VGlut2 ) , the VGlut2 channel was rotated 90° out of alignment , then the two channels were re-merged and analyzed for apparent co-localization of presynaptic puncta . For all conditions analyzed , overlap frequency was significantly decreased in the rotated image ( red ) compared to the original unrotated image ( black ) ( n = 3 z-stacks per animal , 3–4 animals per genotype; p < 0 . 01 , Student's t test ) . ( D ) Representative SIM image of VGlut1 ( green ) /VGlut2 ( red ) -stained S/Z in P15 WT cortex . At the high resolution afforded by SIM ( ∼100 nm ) , VGlut1 and VGlut2 presynaptic puncta do not appear co-localized . Scale bar , 5 µm . ( E ) Scatterplot showing the apparent overlap in VGlut1/VGlut2 puncta at differing co-localization distance thresholds . When the limit of co-localization is 100 nm between puncta , approaching the resolution of SIM , virtually no co-localization is detected . By increasing the allowable distance between co-localized puncta ( closer to the resolution of confocal microscopy ) , VGlut1/VGlut2 overlap frequency eventually approaches the levels previously detected by confocal imaging ( n = 3–4 Z-stacks each from 3 animals ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04047 . 01810 . 7554/eLife . 04047 . 019Figure 8—figure supplement 2 . Imaging presynaptic terminals in proximity to dendritic spines . ( A ) Schematic for the IUE technique . Anesthetized dams have their uterine horns exposed at E15 . 5 . DNA plasmid containing GFP with loading dye is injected into the lateral ventricle of each pup , followed by electric pulses to facilitate uptake of the plasmid . The horns are then placed back into the dam , the incision sutured , and the dam is allowed to recover and give birth to the electroporated litter . ( B ) 10× magnification image taken in V1 at P15 showing that IUE resulted in specific GFP labeling of neurons in cortical Layer II/III . Secondary and tertiary dendrites from these neurons , which project to the S/Z , are then imaged by confocal microscopy . ( C ) Left: A single optical section taken from a confocal Z-stack shows a GFP-labeled dendrite ( blue ) in the S/Z along with surrounding presynaptic puncta ( VGlut1-green; VGlut2-red ) . Middle: Imaris surface rendering was used to image the dendrite's structure , including spines , in 3D . Right: The ‘Spots’ function of Imaris allowed for the resolution of individual presynaptic puncta in three-dimensional space . Scale bar , 1 µm . ( D ) After applying the Matlab ‘Spots close to surface’ algorithm in Imaris , in order to isolate spots within 0 . 2 µm of the dendrite , spines with closely-associated presynaptic puncta can be quantified . An asterisk indicates the location of a SMEC which is contacted by both a VGlut1 and VGlut2 terminal . ( E ) Quantification in P15 WT V1 showing the percentage of spines that contact various presynaptic puncta , including the various subtypes of SMECs ( 3 animals , n = 9 dendrites ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04047 . 019
Synaptic maturation and refinement , wherein appropriate synapses are strengthened while superfluous connections are eliminated , are critical for the establishment of functional neuronal circuitry . In the developing cortex , the process of refinement includes the resolution of synaptic competition between incoming thalamic projections and the resident intracortical axons . In the present study , we show that an astrocyte-secreted synaptogenic protein , hevin , is required for the connectivity of thalamocortical synapses . Furthermore , we found that thalamic and cortical axons simultaneously innervate dendritic spines , uncovering a potential role for spines as sites of synaptic competition . Based on our findings , we propose that , in the developing visual cortex , SMECs serve as sites at which cortical and thalamic inputs compete ( Figure 8C ) . This competition is normally resolved by P25 with the establishment of single synapse spines contacting either VGlut1+ or VGlut2+ terminals . This resolution is dependent on astrocyte-secreted hevin to stabilize VGlut2+ synapses . In the absence of hevin , the VGlut2+ connections cannot be stabilized thus cannot effectively compete with VGlut1+ terminals for postsynaptic spines , resulting in increased VGlut1+ synapses and the persistence of SMECs ( Figure 8C ) . Since hevin is a synaptogenic protein , our initial findings showing no overall reduction in excitatory synaptic density in area V1 of hevin KO mice were surprising . However , this observation was explained by a severe and specific reduction of in the number of thalamocortical connections in hevin KOs , whereas the abundance of intracortical synapses was increased . These results reveal that during development the formation and stabilization of VGlut1+ intracortical and VGlut2+ thalamocortical synapses are differentially regulated . Our findings also suggest that the reduction in VGlut2+ synapses in hevin KO V1 enable synaptic takeover by the more abundant VGlut1+ projections . In agreement with this possibility , previous observations of synaptic competition have shown that retraction of one axon is often succeeded by the expansion of another ( Walsh and Lichtman , 2003 ) . The specificity of hevin for VGlut2+ synaptic stabilization raises the question of why we observed global structural immaturity of dendrites in hevin KO V1 . During development , spontaneous retinal activity relayed by thalamus facilitates proper patterning and connectivity within visual cortex ( Ackman and Crair , 2014 ) . Furthermore , it was previously shown that perturbations in synaptic competition for eye-specific territories in V1 can compromise dendritic spine refinement ( Mataga et al . , 2004 ) . Defective thalamocortical connectivity in hevin KO mice may therefore be responsible for driving generalized spine synapse immaturity in the cortex . In agreement with this possibility , we found that a hevin KOs had increased numbers of excitatory synapses made onto dendritic shafts . It is postulated that excitatory shaft synapses are among the first synapses to form along dendrites ( Reilly et al . , 2011 ) . These shaft synapses may be induced to become spines by presynaptic activity during the early stages of postnatal development ( P8–P12 ) ( Kwon and Sabatini , 2011 ) ; our results suggest that thalamocortical connectivity in general and/or hevin in particular may be playing a role in the shaft to spine transition of some synapses . Hevin has been known to modulate cell adhesion ( Sullivan and Sage , 2004 ) , and its positioning in the synaptic cleft ( Lively et al . , 2007 ) makes it a prime candidate for organizing and stabilizing pre- and postsynaptic cell adhesion molecules . Because hevin is preferentially inducing thalamocortical synapses , there may be specific hevin interactors present on these synapses that distinguish them from the intracortical ones . Hevin may also act as a ‘protection signal’ ( Sanes and Lichtman , 1999 ) to prevent elimination of thalamocortical synapses . Alternatively , hevin is a generalized protective signal but VGlut1+ intracortical synapses may have redundant stabilization/protection mechanisms in place that can compensate for the lack of hevin , whereas with no such compensation available the VGlut2+ thalamocortical population is lost . Our investigation of hevin KO cortex presents an interesting possibility that cortical and thalamic axons compete for postsynaptic targets at single dendritic spines . The presence of SMECs in the central nervous system was described a long time ago ( Jones and Powell , 1969 ) . Perhaps due to the difficulty of observing these synaptic structures , because of their unique geometrical arrangement , this early finding of SMECs was largely ignored and their purpose remained unknown . Instead , a simple ‘one spine: one excitatory synapse’ view of connectivity prevailed . The one spine: one synapse configuration provides a context in which spines can compartmentalize calcium and filter membrane potentials in an input-specific manner ( Yuste , 2013 ) . Though important , these features of spines may not represent the full extent of their functions in the developing CNS . Our findings indicate that SMECs represent a stage in excitatory synaptic development . In agreement with this , multiply-innervated filopodia have been seen in the juvenile CA1 region of hippocampus ( Fiala et al . , 1998 ) ; these protrusions were proposed to ‘sample’ alternative axonal partners during a highly active period of synaptogenesis . In addition , live imaging of green fluorescent protein ( GFP ) -labeled dendrites in acute hippocampal slices from P10–12 mice showed that spines sample nearby synaptic resources even after making a stable contact ( Konur and Yuste , 2004 ) . This sampling was proposed to trigger synaptic competition at individual spines . However , these experiments inferred the presence of axonal boutons from labeling of the presynaptic marker synaptophysin , raising the possibilities that this staining could have been the result of multiple presynaptic release sites ( i . e . , active zones ) on the same synapse or even different boutons belonging to the same axon . By using ssEM-derived 3D reconstructions of synaptic structures , here we show that multiple independent axons compete for synaptic territory on single spines . Furthermore , our finding that SMECs are contacted by different axonal populations ( i . e . , VGlut1+ and VGlut2+ ) suggests that establishment of synaptic networks in the cortex depends on the outcome of synaptic competition at spines , demonstrating a dynamic new role for spines in synaptic development . In conclusion , here we show that the astrocyte-secreted hevin is required for the proper establishment of thalamocortical synapses . Moreover , this process occurs on spines serving as simultaneous contact sites for thalamic and cortical inputs during development , a finding that expands the current view of spines as input filters or calcium buffers . These results may also have important clinical implications . Hevin is strongly upregulated in reactive astrocytes in disease conditions ( McKinnon and Margolskee , 1996 ) and has also been linked to neurological disorders , including autism , schizophrenia , suicide and depression ( Purcell et al . , 2001; Jacquemont et al . , 2006; Kahler et al . , 2008; Vialou et al . , 2010; Zhurov et al . , 2012; De Rubeis et al . , 2014 ) . Abnormal spine maturation and connectivity have also been observed in these and other diseases ( Fiala et al . , 2002; De Rubeis et al . , 2013; Kim et al . , 2013 ) , including the presence of SMEC-like ‘giant spines’ in hippocampi from patients with severe temporal lobe epilepsy ( Witcher et al . , 2010 ) . Future studies may determine if impaired resolution of axonal competition by astrocytes drives the dendritic spine deficiencies observed in these conditions , providing a novel cellular target for therapeutic strategies .
For synaptic puncta analysis of mouse V1 , hevin KO mice on a 129/Sve background and littermate age-matched WT controls were perfused intracardially with Tris-Buffered Saline ( TBS , 25 mM Tris-base , 135 mM NaCl , 3 mM KCl , pH 7 . 6 ) supplemented with 7 . 5 µM heparin followed with 4% paraformaldehyde ( PFA; Electron Microscopy Sciences , PA ) in TBS . The brains were then removed and were fixed with 4% PFA in TBS at 4°C overnight . The brains were cryoprotected with 30% sucrose in TBS overnight and were then embedded in a 2:1 mixture of 30% sucrose in TBS:OCT ( Tissue-Tek , Sakura , Japan ) . Brains were cryosectioned ( coronal ) at 20 µm using Leica CM3050S ( Leica , Germany ) . Sections were washed and permeabilized in TBS with 0 . 2% Triton-X 100 ( TBST; Roche , Switzerland ) three times at room temperature . Sections were blocked in 5% Normal Goat Serum ( NGS ) in TBST for 1 hr at room temperature . Primary antibodies ( guinea pig anti-VGlut1 1:3500 [AB5905 , Millipore , MA] , guinea pig anti-VGlut2 1:7500 [135 404 , Synaptic Systems , Germany] , rabbit anti-VGLUT2 1:750 [135 403 , Synaptic Systems] , rabbit anti-PSD95 1:300 [51–6900 , Invitrogen , CA] ) were diluted in 5% NGS containing TBST . Sections were incubated overnight at 4°C with primary antibodies . Secondary Alexa-fluorophore conjugated antibodies ( Invitrogen ) were added ( 1:200 in TBST with 5% NGS ) for 2 hr at room temperature . Slides were mounted in Vectashield with DAPI ( Vector Laboratories , CA ) and images were acquired on a Leica SP5 confocal laser-scanning microscope . 3–5 animals/genotype/age were stained with pre- ( VGlut1 or VGlut2 ) and post-synaptic ( PSD95 ) marker pairs as described previously ( Kucukdereli et al . , 2011 ) . Three independent coronal sections per each mouse , which contain the V1 visual cortex ( Bregma −2 . 5–−3 . 2 mm , Interaural 1 . 3–0 . 6 mm [Franklin and Paxinos , 2001] ) were used for analyses . 5 µm thick confocal z-stacks ( optical section depth 0 . 33 µm , 15 sections/z-stack , imaged area/scan = 20 , 945 µm2 ) of the synaptic zone in area V1 were imaged at 63× magnification on a Leica SP5 confocal laser-scanning microscope . Maximum projections of three consecutive optical sections ( corresponding to 1 µm total depth ) were generated from the original z-stack . Analyses were performed blind as to genotype . The Puncta Analyzer plugin that was developed by Barry Wark ( available Source code 1 ) for ImageJ 1 . 29 ( NIH; http://imagej . nih . gov/ij/ , version ImageJ 1 . 29 is available at http://labs . cellbio . duke . edu/Eroglu/Eroglu_Lab/Publications . html ) was used to count the number of co-localized , pre- , and post-synaptic puncta . This quantification method is based on the fact that pre- and post-synaptic proteins ( such as VGluts and PSD95 ) are not within the same cellular compartments and would appear co-localized only at synapses due to their close proximity . Previous studies showed that this quantification method yields an accurate estimation of the number of synapses in vitro and in vivo which were previously confirmed by other methods such as EM and electrophysiology by us and others ( Christopherson et al . , 2005; Eroglu et al . , 2009; Kucukdereli et al . , 2011 ) . Details of the quantification method have been described previously ( Ippolito and Eroglu , 2010 ) . Briefly , 1 µm thick maximum projections are separated into red and green channels , background subtracted ( rolling ball radius = 50 ) , and thresholded in order to detect discrete puncta without introducing noise . The Puncta Analyzer plugin then uses an algorithm to detect the number of puncta that are in close alignment across the two channels , yielding quantified co-localized puncta . In order to calculate % of WT co-localization , co-localized puncta values for WT were averaged , then all image values ( WT and KO ) were converted to % of the calculated WT average . For co-localization of VGlut1 and VGlut2 , three P25 hevin KO mice on a 129/Sve background , three littermate P25 WT controls and three P15 WT controls were perfused , sectioned , stained and imaged as described previously for synaptic staining . Primary antibodies ( guinea pig anti-VGLUT1 1:3500 [AB5905 , Millipore] , rabbit anti-VGLUT2 1:750 [135 403 , Synaptic Systems] ) were diluted in 5% NGS containing TBST . For cell staining , three P25 WT mice and one Aldh1L1-eGFP mouse ( in which astrocytes are labeled with eGFP; MMRRC , UC Davis , CA ) were perfused and sectioned as described previously for synaptic staining . Sections containing Layer II/III of V1 visual cortex ( Bregma −2 . 5–−3 . 2 mm , Interaural 1 . 3 to 0 . 6 mm [Franklin and Paxinos , 2001] ) or dorsal LGN ( Bregma −1 . 7–−2 . 9 mm , Interaural 2 . 1–0 . 9 mm [Franklin and Paxinos , 2001] ) were washed and permeabilized in TBS with 0 . 2% Triton-X 100 ( TBST; Roche ) three times at room temperature . Sections were blocked in 5% Normal Donkey Serum ( NDS ) in TBST for 1 hr at room temperature . Primary antibodies ( mouse anti-GFAP 1:1000 [G3893 , Sigma , MO] , mouse anti-NeuN clone A60 1:1000 [MAB377 , Millipore] , rabbit anti-Iba1 1:500 [019–19 , 741 , Wako , Japan] , goat anti-hevin [a . k . a . SPARCL1] 1:500 [AF2836 , R&D Systems , MN] ) were diluted in 5% NDS containing TBST . Sections were incubated overnight at 4°C with primary antibodies . Secondary Alexa-fluorophore conjugated antibodies ( Invitrogen ) were added ( 1:200 in TBST with 5% NDS ) for 2 hr at room temperature . Slides were mounted in Vectashield with DAPI ( Vector Laboratories ) and images were acquired at 63× magnification on a Leica SP5 confocal laser-scanning microscope . WT mice were perfused with PBS intracardially to clear blood before the brains were removed . Cortex and hippocampus were dissected out and homogenized in ice-cold solubilization buffer ( 25 mM Tris pH 7 . 2 , 150 mM NaCl , 1 mM CaCl2 , 1 mM MgCl2 ) containing 0 . 5% NP-40 ( Thermo Scientific , MA ) and protease inhibitors ( Roche ) . The protein concentrations of the lysates were determined using micro BCA protein assay kit ( Pierce , IL ) . Samples for SDS-PAGE were prepared at 1 μg protein/μl concentration using 5× SDS-PAGE buffer ( Pierce ) . 10 μg of protein was loaded into each well . Samples were resolved by SDS-PAGE on 4–15% polyacrylamide gels ( BioRad , CA ) and were transferred onto an Immobilon-FL PVDF membrane ( Millipore ) . Blots were blocked in 50% fluorescent blocking buffer in PBS ( MB-070 , Rockland , PA ) containing 0 . 01% Tween-20 for 1 hr at room temperature . Blots were then incubated with primary antibody dilutions in blocking buffer ( goat anti-SPARCL1 1:2000 [AF2836 , R&D Systems] , rabbit anti-β-tubulin 1:1000 [926–42 , 211 , Li-Cor , NE] ) overnight at 4°C . Fluorescently labeled secondary antibodies ( Li-Cor ) were diluted ( 1:5000 ) in the same buffer as primary antibodies and western blots were incubated with secondary antibodies for 2 hr at room temperature in the dark . Detection was performed using the Li-Cor Odyssey System . Acute coronal slices ( 350 μm thick ) were prepared from P23–P26 Hevin-null animals and littermate wildtype controls . Mice were deeply anesthetized with tribromoethanol ( Alfa Aesar , MA ) and transcardially perfused with ice-cold high-sucrose artificial cerebrospinal fluid ( ACSF ) equilibrated with 95% O2 and 5% CO2 ( carboxygenated ) . Brains were removed and sectioned in ice-cold sucrose ASCF on a Leica VT1200S . Slices recovered in carboxygenated standard ACSF at room temperature for a minimum of 1 hr . Whole-cell patch-clamp recordings of miniature excitatory postsynaptic currents ( mEPSCs ) were recorded at 30°C in ACSF supplemented with tetrodotoxin ( Tocris , UK ) and picrotoxin ( Sigma ) , with a continuous perfusion rate of 2–3 ml/min . Membrane potential was held at −70 mV . Junction potential was uncorrected . Recording pipette internal solution ( pH 7 . 2 ) contained ( in mM ) : 103 CsOH , 103 D-gluconic acid , 2 . 8 NaCl , 5 TEA-Cl , 20 HEPES , 0 . 2 EGTA , 5 Lidocaine N-ethyl chloride , 4 ATP-Mg , 0 . 3 GTP-Na , 10 Na2 Phosphocreatine , 0 . 025 Alexa 488 Hydrazide ( Invitrogen ) and approximately 10 K2S04 ( to bring solution to 300 mOsm ) . Cells from Layer II/III of the visual cortex ( 30–80 microns ventral from the ventral border of Layer I ) were visualized with a 40× water-immersion objective ( LUMPlanFI , 40×/0 . 80 water immersion ) under an Olympus BX51WI microscope equipped with infrared differential interference contrast optics , reflected fluorescence system , and OLY-150 camera ( Olympus , Japan ) . Signals were recorded using a MultiClamp 700B amplifier and DigiData 1322A ( Molecular Devices , CA ) . Cells with capacitance of less than 100 picofarads were excluded to prevent inclusion of interneurons and pyramidal morphology was confirmed post-recording by visualizing the Alexa 488 dye . Pipette resistance ranged from 2 . 8–4 . 1 MΩ . Signals were collected at 10 kHz , unfiltered . Cells with poor technical quality of recordings were excluded from analysis based on predefined criteria of >25 MΩ access resistance or >30% change in series resistance or >6 pA peak-to-peak noise . Miniature EPSC events were analyzed off-line using MiniAnalysis software ( Synaptosoft , GA ) . P18 hevin KO and WT littermate mice were deeply anesthetized with intraperitoneal injection of ketamine ( 150 mg/kg ) /xylazine ( 15 mg/kg ) . Using Nanoject ( Drummond , PA ) , mice were stereotactically injected by 50 nl of EF-1a promoter-driven Flex-AAV-GFP within the dLGN [AP: −2 . 0; ML: 2 . 0; DV: 2 . 3 from brain surface] . To visualize specific neurons in dLGN that are directly connected with visual cortex , 100 nl of rabies virus glycoprotein-coated Lenti-FuGB2-Cre ( synapsin promoter ) ( Kato et al . , 2011 ) was infected into the V1 region of visual cortex [AP: −3 . 5; ML: 2 . 5; DV: 0 . 3 from brain surface] . 2 weeks after infection , brains were removed , postfixed overnight at 4°C , and then cryo-protected with 30% sucrose in TBS . Brains were cut into 50 μm coronal sections by cryostat ( Leica CM 3000 ) . Sections were counterstained with DAPI ( Sigma ) . After washing three times , the sections were coverslipped with FluorSave ( CalBioChem , Merck , Germany ) aqueous mounting medium . For the axonal fiber tracing , images were taken by tile scan imaging using LSM 710 confocal microscope ( Zeiss , Germany ) with a 10× objective under control of Zen software ( Zeiss ) . Neurons from either cortex or thalamus were purified from P1 hevin KO pups by sequential immunopanning as follows: Following dissection , cortex/thalamus was digested for 30 min in papain ( Worthington , NJ ) . Papain digestion was then inhibited in sequential low/high concentrations of ovomucoid inhibitor ( Worthington ) and the resultant digested tissue was passaged through a 20 µm Nitex mesh filter ( Sefar , NY ) . The cell solutions then underwent negative immunopanning ( to remove nontarget cells and debris ) on 2× Bandeiraea Simplicifolia Lectin I-coated petri dishes ( Vector Laboratories ) , AffiniPure goat-anti mouse IgG+IgM ( H+L ) ( Jackson Immunoresearch Laboratories , PA ) coated dish , and AffiniPure goat-anti rat IgG ( H+L ) ( Jackson ) coated dish . A round of positive panning , using rat anti-neural cell adhesion molecule L1 antibody , clone 324 ( MAB5272 , Millipore ) , was used to isolate neurons from other cell types ( predominantly astrocytes ) to greater than 95% purity . Cortical cells or mixed cortical/thalamic cells ( 1:1 ratio ) were then cultured in serum-free medium containing BDNF , CNTF , and forskolin on laminin-coated coverslips as previously described ( Christopherson et al . , 2005; Kucukdereli et al . , 2011 ) . Recombinant hevin protein was purified as described previously ( Kucukdereli et al . , 2011 ) . Neurons were cultured for 3 days , then were treated for 36 hr with AraC to kill any contaminating mitotic cells ( i . e . , astroglia ) , then were cultured with 90 nM hevin or hevin-free growth media for an additional 9 days . Synapse quantification of cortical/thalamic cultures follows the procedure outlined in Kucukdereli et al . ( 2011 ) with the exception of the antibodies used: primary antibodies against VGlut1 ( 1:1000; guinea pig; Millipore ) , VGlut2 ( 1:500; rabbit; Synaptic Systems ) , and PSD95 ( 1:500; mouse; Neuromab , CA ) ; secondary antibodies consisted of Alexa-conjugated antibodies diluted 1:1000 in antibody buffer . Imaging was performed on the AxioImager M1 ( Zeiss ) at 63× magnification . Only cortical neurons were imaged and thalamic neurons were avoided for imaging by the appearance of bright VGlut2 staining within the cell soma . P13 hevin KO and WT littermate mice were deeply anesthetized with intraperitoneal injection of ketamine ( 150 mg/kg ) /xylazine ( 15 mg/kg ) . With Nanoject , mice were stereotactically injected with either recombinant hevin protein ( 200 ng in Dulbecco's PBS , Gibco , CA ) or vehicle control ( DPBS; 100 nl ) into layer II/III of area V1 [AP: −2 . 1; ML: 2 . 3; DV: 0 . 25 from brain surface] . After 3 days , pups were anesthetized with Avertin then perfused transcardially with TBS containing heparin followed by 4% PFA . Brains were harvested , post-fixed overnight in 4% PFA , then cryoprotected in 30% sucrose-TBS . Sections ( 20 µm ) were cut on a cryostat ( Leica ) and stained for VGlut1/VGlut2/PSD95 as described above . Imaging was performed on a Leica SP5 confocal microscope in area V1 approximately 100 µm laterally to the site of injection . Golgi-cox staining was performed on hevin KO and littermate WT control mice ( n = 3 mice per genotype ) as described in the FD Rapid GolgiStain Kit ( FD NeuroTechnologies , MD ) . Dye-impregnated brains were embedded in Tissue Freezing Medium ( Triangle Biomedical , NC ) and were rapidly frozen on ethanol pretreated with dry ice . Brains were cryosectioned coronally at 100 µm thickness and mounted on gelatin-coated microscope slides ( LabScientific , NJ ) . Sections were stained according to the directions provided by the manufacturer . Three independent coronal sections per each mouse , which contain the V1 visual cortex ( Bregma −2 . 5–−3 . 2 mm , Interaural 1 . 3–0 . 6 mm [Franklin and Paxinos , 2001] ) were imaged . Layer II/III pyramidal neurons were identified by their distance from pia and their distinct morphologies . Secondary and tertiary dendrites of these neurons were selected for analysis . Z-stacks of Golgi-stained dendrites ( up to 80 microns total on z-axis; optical section thickness = 0 . 5 µm ) were taken at 63× magnification on a Zeiss AxioImager M1 . Series of TIFF files corresponding to each image stack were loaded into RECONSTRUCT software ( Fiala , 2005 ) ( freely available at http://synapses . clm . utexas . edu ) . For each series , 3 × 10 µm segments of dendrites were chosen for analysis . 15 dendrites were analyzed per animal making a total of 45 dendrites per condition . Analyses were performed blind as to genotype . Dendritic spines were identified on the selected dendritic segments; more than 500 spines per genotype were analyzed . Spines were analyzed by the rapid Golgi analysis method described in Risher et al . ( 2014 ) . Briefly , using the ‘Draw Z-trace’ tool in RECONSTRUCT , the three-dimensional length of each spine ( from the point where the spine neck contacted the dendritic shaft out to the tip of the spine head ) was measured . Spine head width was measured by drawing a straight line across the widest point of each spine in a single image of the z-series . These measurements were exported to Microsoft Excel ( Microsoft , WA ) , where a custom macro was used to classify spines based on the spine length , width , and length:width ratio measurements obtained in RECONSTRUCT . Spines were categorized based on the following hierarchical criteria: ( 1 ) Branched = more than one spine head attached to same spine neck; ( 2 ) Filopodium = length > 2 µm; ( 3 ) Mushroom = width > 0 . 6 µm; ( 4 ) Long thin = length > 1 µm; ( 5 ) Thin = length:width ratio > 1; ( 6 ) Stubby = length:width ratio ≤ 1 . For quantification of neurite outgrowth and branching , 100 µm thick coronal Golgi-cox stained sections were visualized using a Zeiss AxioImager D2 microscope . A total of 24 V1 Layer 2–3 neurons ( 4 neurons per animal , 3 animals per condition ) were selected for dendritic tracing from P25 hevin null and littermate WT controls . Tracing was performed with Neurolucida tracing tool ( MBF Bioscience , VT ) . Convex hull analysis was used to measure total dendritic length and area , while Sholl analysis was used to determine dendritic complexity/branching . All analyses were done with NeuroExplorer ( MBF Bioscience ) software . For ssEM analysis of mouse V1 , P14 WT controls , P25 hevin KO mice and their littermate P25 WT controls ( 3 mice per genotype/age , all on a 129/Sve background ) were first transcardially perfused with warm PBS solution to clear out blood cells , and then with warm ( 37°C ) 2% PFA , 2 . 5% glutaraldehyde ( EMS ) , 2 mM CaCl2 , and 4 mM MgCl2 in 0 . 1 M cacodylate buffer ( EMS ) ( pH 7 . 4 ) under Tribromoethanol ( Sigma ) anesthesia . 400 µm thick coronal sections per each mouse , which contain the V1 visual cortex ( Bregma −2 . 5–−3 . 2 mm , Interaural 1 . 3–0 . 6 mm [Franklin and Paxinos , 2001] ) were cut with a tissue chopper ( Stoelting , IL ) and area V1 was dissected out with a #11 scalpel blade . V1 slices were immersed in 2% glutaraldehyde , 2 mM CaCl2 , and 4 mM MgCl2 in 0 . 1 M cacodylate buffer ( pH 7 . 4 ) and fixed overnight at 4°C . At the Duke Electron Microscopy Service core facility , slices were rinsed 3 × 5 min in 0 . 1 M phosphate buffer ( PB ) and postfixed in 1% OsO4 ( Sigma ) while heating in a microwave ( 2 min on , 2 min off , then 2 min on at 70% power with vacuum ) . After rinsing 2 × 5 min with 0 . 1 M PB , they were dehydrated in ethanol/acetone series enhanced with 40 s of microwave processing . They were next incubated in 50:50 acetone:epoxy overnight at room temperature . After two changes of straight epon 3 × 3 min in the microwave , slices were left to stand for 30 min and then embedded in 100% epon resin at 60°C for 48 hr . Ultrathin serial sections ( 45–50 nm ) were cut from a small trapezoid positioned 50–100 µm below the pial surface corresponding to the synaptic zone ( a . k . a . layer I ) which contain the dendrites of layer II/III neurons . Serial sectioning , processing and photography were carried out by the Electron Microscopy Core at Georgia Regents University , following a protocol adapted from Harris et al . ( 2006 ) . Series consisting of 100–150 consecutive micrographs each were blinded as to condition prior to analysis . Serial sections were aligned and synaptic structures were traced using RECONSTRUCT software . Section thickness was calculated with the cylindrical diameters method ( Fiala and Harris , 2001 ) . Dendrites were chosen for analysis on the basis of ( 1 ) spanning at least 75 consecutive serial sections , ( 2 ) measuring between 0 . 4–0 . 8 µm in diameter in cross-section ( to exclude large , apical dendrites and only include secondary and tertiary dendrites ) and ( 3 ) having at least 1 spine ( to exclude aspinous dendrites from interneurons ) . PSD area was calculated by multiplying the two-dimensional length on each section by average section thickness and the total number of sections on which the PSD appears . For immuno-EM analysis of mouse V1 , 3 P14 WT mice on a 129/Sve background were transcardially perfused with warm ( 37°C ) 4% PFA , 0 . 2% glutaraldehyde , 2 mM CaCl2 , and 4 mM MgSO4 in 0 . 1 M cacodylate buffer ( pH 7 . 4 ) under Tribromoethanol anesthesia . 100 µm thick coronal sections per each mouse , which contain the V1 visual cortex ( Bregma −2 . 5–−3 . 2 mm , Interaural 1 . 3–0 . 6 mm [Franklin and Paxinos , 2001] ) were cut with a vibratome ( Leica ) and immersed in the perfusion fixative at 4°C . At the Electron Microscopy Core at Georgia Regents University , slices were rinsed 3 × 5 min in Hepes buffered saline ( HBS ) . Non-specific binding sites were permeablilized and blocked for 30 min with HBS , 10% BSA ( Sigma ) , and 0 . 025% Triton X-100 ( Sigma ) . Permeabilization solution was replaced with ice-cold guinea pig anti-VGlut1 1:750 ( Millipore ) in HBS plus 1% BSA and 0 . 0025% Triton X-100 , and slices were incubated at 4°C overnight on a shaker . After washing 3 × 5 min in HBS-0 . 05% BSA , slices were incubated in anti-guinea pig Nanogold 1:250 ( Nanoprobes , NY ) at 4°C overnight on a shaker . Slices were washed 3 × 5 min in HBS-0 . 05% BSA , then four changes of distilled H2O for 2 hr , then incubated for 2 hr on a shaker in 0 . 5 ml of GoldEnhance EM ( Nanoprobes ) mixed according to manufacturer's directions . Slices were washed thoroughly in ice-cold H2O to stop the gold enhancement . After washing 2 × 5 min in HBS , slices were incubated in rat anti-VGlut2 1:250 ( MabTechnologies , Inc . , GA ) in HBS plus 1% BSA and 0 . 00025% Triton X-100 at 4°C overnight on a shaker . After washing 3 × 5 min in HBS-0 . 05% BSA , slices were incubated in anti-rat Nanogold 1:250 ( Nanoprobes ) at 4°C overnight on a shaker . Slices were washed 3 × 5 min in HBS-0 . 05% BSA , then fixed at room temperature in perfusion fixative for 20 min . After four changes of distilled H2O for 2 hr , slices were incubated for 3 hr on a shaker in 0 . 5 ml of GoldEnhance EM solution . Slices were washed thoroughly in ice-cold H2O to stop the gold enhancement , washed 2 × 5 min in HBS , then washed 3 × 5 min in 0 . 1 M cacodylate buffer in preparation for processing and embedding . Slices were post-fixed in 1% OsO4 plus 1/5% potassium ferrocyanide in cacodylate buffer for 1 hr . Slices were washed 3 × 10 min in cacodylate buffer , post-fixed for 1 hr in 1% OsO4 in cacodylate buffer , then washed 3 × 10 min in distilled H2O . Slices were stained in 2% aqueous uranyl acetate on a shaker at room temperature for 1 hr , then washed 3 × 5 min in distilled H2O . They were then dehydrated in an ascending ethanol series ( 50% , 70% , 90% and 100% ) for 5–10 min each , with 100% repeated 3 × 10 min . Slices went through 2 × 10 min changes of propylene oxide , were placed in a 1:1 mixture of propylene oxide: Embed 812 resin mixture ( EMS ) for 1 hr , then put in 100% Embed 812 overnight on a rotator . Slices were flat embedded so that the plane of sectioning was perpendicular to the slice's surface , polymerized at 60°C for 24 hr . Thin sections were cut with a diamond knife on a Leica EM UC6 ultramicrotome , collected on copper grids and stained with lead citrate . Sections were observed in a JEM 1230 transmission electron microscope ( JEOL , Japan ) at 110 kV . Areas positioned 50–100 µm below the pial surface corresponding to the synaptic zone ( a . k . a . layer I ) , containing the dendrites of layer II/III neurons , were imaged with an UltraScan 4000 CCD camera and First Light Digital Camera Controller ( Gatan Inc . , PA ) . WT brains were harvested and cryoprotected at P15 following 4% PFA fixation . Sections ( 20 µm ) were cut on a cryostat ( Leica ) and stained for IHC using primary antibodies against VGlut1 ( 1:500; guinea pig; Millipore ) and VGlut2 ( 1:750; rabbit; Synaptic Systems ) followed by Alexa-conjugated secondary antibodies . Sections were imaged using a Zeiss ELYRA PS1 microscope . 3D structured illumination images of the S/Z of V1 were captured and images subsequently processed using Zeiss SIM algorithms . To quantify co-localized VGlut1 and VGlut2 puncta , SIM-processed image files were opened in Imaris ( Bitplane , Switzerland ) and spot channels generated for the synaptic markers using dimensions determined empirically from averaged measurements . Matlab ( Mathworks , MA ) was subsequently used to only show those puncta within 100 nm , 200 nm , or 300 nm of one another . Timed pregnant wild type WT ( CD1 , Charles River , MA ) and hevin KO ( 129/Sve ) dams were utilized for IUE . All electroporations were performed at embryonic day ( E ) 15 . 5 in order to target neocortical layer 2/3 pyramidal neurons . Dams were sedated with continuously vaporized isofluorane and cesarean sectioned to expose both uterine horns . 1 μg of DNA plasmid containing shControl-CAG-EGFP with loading dye was injected into one lateral ventricle of each embryo using a pulled glass pipette . Five 50 ms pulses of 50 V spaced 950 ms apart were applied with tweezertrodes ( positive paddle against the skull over the injection site , the negative paddle across the body away from the placenta ) using the BTX ECM 830 ( Harvard Apparatus , MA ) . Warm PBS was applied to embryos and dam to prevent drying . Following the electroporation , the uterine horns were returned to the abdominal cavity and the peritoneum , anterior muscle , and skin were sutured separately . The dam was then placed on a heating pad to recover and monitored daily following the surgery . All procedures for animal surgery and maintenance were performed in accordance with Duke Institutional Animal Care and Use Committee . Electroporated brains were harvested and cryoprotected at P21 following 4% PFA fixation . Sections ( 40 µm ) were cut on a cryostat ( Leica ) and stained for IHC using primary antibodies against GFP ( 1:750; chicken; Millipore ) , VGlut1 ( 1:500; guinea pig; Millipore ) , and VGlut2 ( 1:750; rabbit; Synaptic Systems ) followed by Alexa-conjugated secondary antibodies . GFP- expressing secondary/tertiary dendrites in the S/Z , along with surrounding VGlut1/2 presynaptic puncta , were imaged on a Zeiss 780 inverted confocal microscope at 63× with 8× optical zoom at 0 . 13 µm optical section thickness . Z-stacks were deconvolved with Huygens image processing software ( Scientific Volume Imaging , The Netherlands ) and then imported into Imaris for analysis . In Imaris , dendrites were reconstructed in 3D using either the Surfaces or FilamentTracer tool . Discrete VGlut puncta were resolved with the Spots tool . Presynaptic puncta within 0 . 2 µm of dendrites were then isolated using the Find Spots Close to Filament/Surface Matlab algorithm . Spines were then quantified by eye on the basis of their associated presynaptic partners , with the analyst blinded as to the genotype . Statistica ( StatSoft , OK ) was used for all statistical analyses . Variability between different IHC synaptic staining pairs was controlled for by the use of nested design hierarchical ANOVAs ( under Generalized Linear Models in Statistica ) with experimental pair nested within condition . Graphical data are presented as mean ± s . e . m . | The central nervous system—which is made up of the brain and spinal cord—processes information from all over the body . The information travels through cells called neurons , which connect to each other at junctions called synapses . A single neuron can receive information from many different places because it is covered with protrusions known as dendritic spines that enable it to form synapses with a variety of other neurons . In recent years , it has become apparent that brain cells other than neurons can influence synapse formation . The most abundant cells in the central nervous system are star-shaped cells known as astrocytes , which secrete molecules that control the timing and extent of synapse formation . Many previous studies on synapses have used a type of neuron found in the eye—called retinal ganglion cells—because these cells can be purified and grown in the laboratory in the absence of astrocytes . Under these conditions , they form very few synapses . However , in the presence of astrocytes the retinal ganglion cells form many more synapses , which is thought to be due to a protein called hevin and several other proteins that are secreted by the astrocytes . Risher et al . studied a region of the brain called the cerebral cortex in mice that were missing hevin . In the cortex of normal mice , the neurons generally form synapses with other neurons within the cortex , or with neurons from other parts of the brain that send long-distance projections into the cortex . The experiments revealed that fewer of these long-distance synapses formed in the cortex of the mice missing hevin compared to normal mice . When hevin was injected directly into the brains of the mice , more long-distance synapses were formed . Using a technique called three-dimensional electron microscopy , Risher et al . examined the structure of the synapses . In mice missing hevin , the synapses were much smaller and the dendritic spines were thin and long , indicating that they were not fully grown . The images also show that in normal mice , the dendritic spines often have multiple synapses when the animal is young , but many are lost as the brain matures so that only a single synapse remains in each dendritic spine . However , multiple synapses persist in the dendritic spines of mice lacking hevin , which could lead to competition between short and long distance synapses and may contribute to neurological diseases . These results indicate that astrocytes are crucial for controlling the formation of synapses in dendritic spines . In humans , defects in hevin have been implicated in autism , schizophrenia and other neurological conditions . Future studies will seek to determine the precise role of astrocytes in these conditions , which may help us to develop new therapies . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"cell",
"biology",
"neuroscience"
] | 2014 | Astrocytes refine cortical connectivity at dendritic spines |
Exposure to stress increases the risk of developing mood disorders . While a subset of individuals displays vulnerability to stress , others remain resilient , but the molecular basis for these behavioral differences is not well understood . Using a model of chronic social defeat stress , we identified region-specific differences in myelination between mice that displayed social avoidance behavior ( ‘susceptible’ ) and those who escaped the deleterious effect to stress ( ‘resilient’ ) . Myelin protein content in the nucleus accumbens was reduced in all mice exposed to stress , whereas decreased myelin thickness and internodal length were detected only in the medial prefrontal cortex ( mPFC ) of susceptible mice , with fewer mature oligodendrocytes and decreased heterochromatic histone marks . Focal demyelination in the mPFC was sufficient to decrease social preference , which was restored following new myelin formation . Together these data highlight the functional role of mPFC myelination as critical determinant of the avoidance response to traumatic social experiences . Editorial note: This article has been through an editorial process in which the authors decide how to respond to the issues raised during peer review . The Reviewing Editor's assessment is that all the issues have been addressed ( see decision letter ) .
Exposure to stress increases the risk of developing affective disorders such as depression and post-traumatic stress disorder . While stress leads to maladaptive behavioral responses in a subset of humans , others are capable of coping and remain resilient . Differences in the behavioral response to stress can also be detected in experimental mouse models , thereby highlighting the degree of conservation of this response . However , the cellular and molecular basis underlying resilience or susceptibility to negative experiences remains poorly defined . We and others have previously reported that animal models of psychosocial stressors , such as social isolation ( Liu et al . , 2012; Liu et al . , 2016; Makinodan et al . , 2016; Makinodan et al . , 2017; Makinodan et al . , 2012 ) , chronic social defeat stress ( CSDS ) ( Cathomas et al . , 2019; Lehmann et al . , 2017 ) , and chronic variable stress ( Liu et al . , 2018 ) , lead to transcriptional , translational , or ultrastructural changes in oligodendrocytes and myelination . Here we tested the hypothesis that myelinating glia serves a causal role in behavioral susceptibility or resilience following stress exposure . We examined social behaviors , ultrastructural changes in myelination as well as epigenetic modifications in oligodendrocytes in brain regions that have been implicated in depressive-like behavior after a well-established social defeat paradigm ( Berton et al . , 2006; Golden et al . , 2011; Hodes et al . , 2014; Krishnan et al . , 2007; Vialou et al . , 2010 ) . We also provide mechanistic insights into the region-specific differences between the phenotypes , which we attributed to defective oligodendrocyte progenitor differentiation . To provide direct causal evidence , we carried out focal demyelination in the medial prefrontal cortex and showed aversive social behavior in animals undergoing demyelination and a resolution of the behavioral effect consequent to new myelin formation . Together , we suggest the functional role of region-specific myelination in determining depressive-like social behavior .
We adopted a mouse model of chronic social defeat stress ( CSDS ) ( Golden et al . , 2011 ) , in which mice were exposed to an aggressor challenge for 10 days ( Figure 1A ) and tested for social behavior afterwards . While some mice showed signs of social withdrawal , characterized by reduced social interaction time when a conspecific mouse is present and reduced social interaction ratio ( i . e . susceptible mice ) , a subset escaped this deleterious consequence ( i . e . resilient mice ) , and were virtually indistinguishable from the control group , which were not exposed to any aggressors ( Figure 1B–C ) . Next , we sought to determine whether there was any myelination difference between susceptible and resilient mice . We focused our analysis on the nucleus accumbens ( NAc ) and the medial prefrontal cortex ( mPFC ) , two brain regions shown to play a critical role in determining stress responses ( Heshmati et al . , 2018; Han and Nestler , 2017 ) and displaying myelin transcriptional or structural impairment after a stressful experience ( Liu et al . , 2012; Liu et al . , 2016; Makinodan et al . , 2016; Lehmann et al . , 2017; Liu et al . , 2018; Zhang et al . , 2016 ) . In the NAc , a significant reduction of myelin basic protein ( MBP ) was detected in all defeated mice , regardless of their behavioral responses ( control , 3 . 6 ± 0 . 4%; susceptible , 2 . 4 ± 0 . 3%; resilient 2 . 0 ± 0 . 3%; Figure 1D–E ) . However , no significant differences were detected in the length of myelinated segments measured by MBP immunoreactivity ( Figure 1F–H ) or in myelin thickness ( Figure 1I–K ) among groups . Pearson coefficients correlation showed no significant correlation between the length of MBP-covered segments and social interaction ratio in either control or defeated group ( control , r = −0 . 2242 , p=0 . 5934 , defeated , r = −0 . 3483 , p-0 . 3240 , Figure 1G–H ) . Altogether , these results suggest that myelination in the NAc uniformly responds to stress and does not distinguish susceptibility and resilience following CSDS . In contrast , the mPFC displayed a unique myelination phenotype following CSDS . While the levels of MBP did not significantly differ between susceptible and resilient mice ( control , 6 . 2 ± 0 . 5%; susceptible , 6 . 1 ± 0 . 9%; resilient , 5 . 3 ± 0 . 8%; Figure 2A–B ) , the length of myelinated segments indicated by MBP immunoreactivity showed a significant positive correlation with social interaction in defeated mice ( Figure 2C–E ) . Importantly , such correlation was not detected in the control ( unstressed ) group , suggesting that changes in the length of myelinated segments represent an adaptive response to the social defeat stress . To more accurately quantify internodal length , we conducted immunohistochemical analysis using antibodies specific for the contactin-associated protein ( Caspr ) , which marks the paranodal regions ( Figure 2F ) . Also in this case , a significant positive correlation between internodal length and social interaction ratio was detected only in the defeated mice , with shorter internodal lengths identified in susceptible mice ( Figure 2G–H ) . Myelin was also thinner in the susceptible - but not in the resilient – mice , compared to controls ( Figure 2I–K ) . Therefore , region-specific myelination differences in the mPFC could -at least in part- explain the behavioral differences between susceptible and resilient mice in response to stress . To determine whether reduced myelin content in the mPFC of susceptible mice was limited to the internodal length , we further performed a detailed quantitative immunohistochemical analysis on oligodendrocyte lineage cells . No significant difference in the overall number of OLIG2+ cells was detected ( Figure 3A , C ) , thereby ruling out decreased survival of oligodendrocyte lineage cells in response to stress . However , compared to resilient and controls , the susceptible mice were characterized by a significantly higher number of NG2+ progenitor cells ( control , 39 . 2 ± 2 . 3 mm−2; susceptible , 55 . 3 ± 4 . 4 mm−2; resilient , 25 . 8 ± 4 . 8 mm−2; Figure 3A , C ) , and lower number of CC1+ mature oligodendrocytes ( control , 91 . 9 ± 5 . 7 mm−2; susceptible , 62 . 7 ± 5 . 2 mm−2; resilient , 86 . 1 ± 5 . 8 mm−2; Figure 3B–C ) . Consistent with defective differentiation of NG2+ cells in the mPFC of susceptible mice , a reduction of the histone modification marks associated with differentiation ( H3K9me3 ) was also detected ( pixel/area: control , 1373 . 6 ± 113 . 3; susceptible , 695 . 3 ± 127 . 9; resilient , 1186 . 0 ± 106 . 0; Figure 3D ) . Together these data suggest that social stress might have at least two main effects on oligodendrocyte lineage cells in the mPFC: it promotes myelin remodeling resulting in shorter internodal length and fewer wraps and possibly impairs in the epigenetic program of oligodendrocyte progenitor differentiation , resulting in fewer differentiated oligodendrocytes . The data above suggested an interesting correlation between myelination in the mPFC and social avoidance behavior in the susceptible mice . To test the causality of this finding , we induced myelin loss by focal injection of lysolecithin ( LPC ) into the mPFC and asked whether this manipulation would be sufficient to induce behavioral changes . LPC injection is a well characterized model of toxic demyelination , with early myelin loss ( detectable one week after injection ) followed by spontaneous repair , due to the formation of new myelin by newly differentiated oligodendrocytes ( occurring three weeks after injection ) ( Jeffery and Blakemore , 1995 ) . We reasoned that behavioral differences detected in mice at these two time points after LPC injection would support a causal link between social preference performance and myelin content in the mPFC ( Figure 4A ) . Indeed , the kinetics of demyelination and remyelination after LPC injection was validated by the detection of reduced MBP immunoreactivity at the 7dpi followed by spontaneous recovery of immunoreactivity by 21dpi ( Figure 4B ) . At the early time point ( 7dpi ) , LPC-injected mice displayed reduced social preference behavior compared to saline-injected controls ( Figure 4C ) . This difference in social interaction behavior was no longer detectable after 3 weeks ( Figure 4D ) , when myelination recovered to normal level ( Figure 4E ) . Therefore , we conclude that myelin content in the mPFC is a critical determinant of social behavior . Altogether , our study reveals region-specific epigenetic dysregulation of oligodendrocyte progenitor differentiation and subsequent defective adult myelination as maladaptive mechanisms occurring only in susceptible mice after exposure to repeated social stress . We have previously reported that myelination defects were detected in socially isolated adult mice , prior to the appearance of social avoidance behavior ( Liu et al . , 2012 ) . Here , we show that social avoidance behavior can be detected after chronic social defeat stress as well as after focal demyelination in the mPFC , and could therefore be caused by hypomyelination . Furthermore , promoting myelination has been shown to rescue depressive-like behavior in socially isolated mice ( Liu et al . , 2016 ) . On the same note , normal social behavior was restored following the spontaneously occurring remyelination in LPC-injected mice . However , it is important to note that , social stress did not induce a toxic effect on myelin , whereas LPC did . No cellular toxicity was detected in the mPFC of susceptible mice or in mice undergoing social isolation ( Liu et al . , 2012 ) . In contrast , we detected fewer mature oligodendrocytes and more progenitors lacking epigenetic marks of differentiation , suggesting an altered epigenetic program . For this reason , we interpret the lower myelin content in the mPFC of susceptible mice as resulting from impaired oligodendrocyte progenitor differentiation , possibly as maladaptive response to social stress . While our data support an inefficient production of new myelin , the detection of a positive correlation between intermodal length and social avoidance behavior , suggests that reorganization of paranodal loops could also be affected . Indeed , shorter internodal length , consequent to impaired myelin formation , has been previously shown to decrease nerve conduction in the optic nerve ( Etxeberria et al . , 2016 ) . It is , therefore , conceivable that the reduced length of myelinated segments detected in the mPFC of susceptible mice may result in slower conduction and functionally result in the characteristic social avoidance behavior in response to the social stress . Finally , we suggest that new myelin is formed in the mPFC of resilient mice as an adaptive mechanism to the repeated episodes of aggression . It is conceivable that the formation of new myelin in resilient mice could favor the establishment of neuronal circuits allowing the escape of negative impact following traumatic stress ( Krishnan et al . , 2007; Fagundes et al . , 2013; Fenster et al . , 2018; Ménard et al . , 2017; Russo et al . , 2012 ) , as oligodendrocytes are known to regulate conduction speed and play a crucial role in synchronizing neuronal networks ( Saab and Nave , 2017 ) . This explanation is in agreement with the increasing evidence from mice and squirrel monkeys , which suggests stress resilience may arise from active coping strategies , rather than a passive response , defined as lack of adaptive response ( Russo et al . , 2012; Lyons et al . , 2009 ) . The molecular basis for resilience has been studied extensively in the context of neuronal cells , the immune and neuroendocrine systems ( Ménard et al . , 2017; Russo et al . , 2012 ) . Here we proposed an alternative , although not mutually exclusive explanation involving myelinating glia . One possibility for new myelin formation as a coping strategy is associated with increased neuronal activity in the resilient mice , as reported by a greater degree of FosB , or ΔFosB expression in glutamatergic neurons of mPFC of resilient mice following social defeat ( Covington et al . , 2010; Lehmann and Herkenham , 2011 ) . Optogenetic stimulation of mPFC has been shown to help resilience phenotype in social defeated mice ( Covington et al . , 2010 ) . Although not characterized in the previous study ( Covington et al . , 2010 ) , optogenetic stimulation has been shown to promote oligodendrogliogenesis and new myelin formation ( Gibson et al . , 2014 ) . An alternative mechanism could involve inflammatory cytokines , such as interleukin-6 ( IL-6 ) . IL-6 has been identified as a major cytokine that contributes to the development of depression in human ( Dowlati et al . , 2010; Erta et al . , 2012 ) . In animal models of stress , systemic IL-6 , was the only differentially regulated cytokine that distinguished resilient mice from susceptible and control mice ( Hodes et al . , 2014 ) . Although systemic changes of IL-6 could not account for the region-specific differences in myelination in susceptible and resilient mice , it is known that IL-6 can be produced by neurons , astrocytes , microglia or endothelial cells in the central nervous system ( Erta et al . , 2012 ) . Several transcriptomic studies suggest that oligodendrocyte progenitors express IL-6 receptors ( Zeisel et al . , 2015; Zhang et al . , 2014 ) . Therefore , it is intriguing to think that IL-6 could be up-regulated in a region-specific pattern with the ability to impact oligodendrocyte progenitor differentiation and new myelin formation in specific regions of the adult brain . Overall this study extends our knowledge on the functional role of adult myelination by providing a mechanism for adaptation to social stress encounters , which ultimately result in the expression of resilience .
All experimental C57Bl/6J male mice ( 7 weeks ) were obtained from the Jackson Laboratory ( Bar Harbor , Maine ) and allowed one-week acclimation prior to the start of experiment . Retired male CD1 breeders used as the aggressors were obtained from Charles River ( Wilmington , Massachusetts ) . All mice were maintained in a temperature- and humidity-controlled facility on a 12 hr light-dark cycle with food and water ad libitum . All procedures were carried out in accordance with the Institutional Animal Care and Use Committee guidelines of the Icahn School of Medicine at Mount Sinai , Hunter College and Advanced Science Research Center at City University of New York . Chronic social defeat stress was performed as previously published ( Berton et al . , 2006; Golden et al . , 2011; Krishnan et al . , 2007; Vialou et al . , 2010; Wilkinson et al . , 2009 ) with slight modification . Briefly , male C57 mice were exposed to a novel aggressive CD1 male mouse for 5 min/day , after which the mice were separated by a Plexiglas barrier that allows for sensory contact without further physical interaction . Control mice were housed two animals/cage under the same conditions as their experimental counterparts but without the presence of an aggressive CD1 mouse . Twenty-four hours after the last of 10 daily defeat or control episodes , mice were evaluated in a social interaction test during the light cycle , as previously described ( Liu et al . , 2012 ) , then one-way ANOVA tests were performed to assess statistical differences and assess social avoidance . Social interaction ratio was calculated by dividing the time spent in the interaction zone when a conspecific mouse is present by no subject present in the enclosure area . Defeated mice with a social interaction ratio below one are defined as ‘susceptible’ , while those with a social interaction ratio above one are defined as ‘resilient’ . Mice were processed for standard electron microscopy ( EM ) analysis as previously described ( Liu et al . , 2012 ) . Briefly , the mounted section was trimmed to encompass a 1 μm2 region of layers 4–6 of the PFC , thin sectioned at 90 nm , stained with uranyl acetate and lead citrate , and mounted on 200 mesh copper grids . Ten images at 10 , 000X were collected per mouse using a transmission electron microscope JEOL JEM 1400Plus equipped with a Gatan CCD camera . g-ratios were determined by dividing the diameter of the axon by the diameter of the entire myelinated fiber . ImageJ was used to measure both axon caliber and myelin fiber diameter for a minimum of 100 myelinated axons per mouse . All analyses were performed blind to the experimental conditions . One-way ANOVA tests were performed to assess statistical differences . Mice were anesthetized and then perfused , cryopreserved , embedded , and sectioned as previously described ( Liu et al . , 2012 ) . Immunohistochemistry was performed as previously described ( Liu et al . , 2012 ) with primary antibodies against trimethylated histone 3 lysine 9 ( H3K9me3 , 1:100; ab8898 , Abcam ) , CC1 ( 1:100; OP80 , Calbiochem ) , myelin basic protein ( MBP , 1:500; SMI99 , Covance ) , OLIG2 ( 1:200 , ab81093 , Abcam ) , NG2 ( 1:200; AB5320 , EMD Millipore ) or Caspr ( 1:100 , ab34151 , Abcam ) . Stained sections were visualized using confocal microscopy ( LSM800 Meta confocal laser scanning microscope , Carl Zeiss Micro-Imaging ) . For NG2 , CC1 , OLIG2 cell counts , and H3K9me3 intensity quantifications , 4–6 20x fields were taken per mouse . For MBP-covered segments and internodal length marked by Caspr , 4–6 fields were taken per mouse followed by quantifications using ImageJ . One-way ANOVA tests were performed to assess statistical differences . For correlation of internodal length with social interaction ratio , data normality was determined using D’Agostino and Person test in GraphPad Prism 8 . Pearson correlation coefficients were calculated if data passed normality test . While under deep anesthesia induced by inhaled isoflurane , experimental C57BL/6J mice were surgically injected with 1 μl 1% lysolecithin ( l-α-lysophosphatidylcholine , Sigma-Aldrich ) dissolved in saline , or saline as sham control , bilaterally to the medial prefrontal cortex using a pulled capillary glass pipet at the following stereotaxic coordinates: anterioposterior , +1 . 5 mm; mediolateral from bregma , 0 . 5 mm; and dorsoventral-below the surface of the dura , 1 . 5 mm . The needle was left in place for an additional 2 min to avoid back flow of the lysolecithin or saline . Muscle and skin incisions were sutured with gut and nylon sutures , respectively . To reduce postoperative pain after recovery from anesthesia , animals received a subcutaneous injection of buprenorphine ( 1 . 0 mg/kg ) . Animals were monitored closely following surgery and were tested with social interaction tests at 7- and 21 days post injection . | High levels of stress do not have the same effect on everybody: some individuals can show resilience and recover quickly , while other struggle to cope . Scientists have started to investigate how these differences may find their origin in biological processes , mainly by focusing on the role of neurons . However , neurons represent only one type of brain cells , and there is increasing evidence that interactions between neuronal and non-neuronal cells play an important role in the response to stress . Oligodendrocytes are a common type of non-neuronal cells which shield and feed nerve cells . In particular , their membrane constitutes the myelin sheath , a protective coating that insulates neurons and allows them to better communicate with each other using electric signals . Bonnefil et al . explored whether differences in oligodendrocytes could affect how mice responded to social stress . The rodents were exposed to repeated attacks from an aggressive mouse five minutes a day for ten days . After this period , ‘susceptible’ mice then avoided future contact with any other mice , while resilient animals remained interested in socializing . Comparing the brain areas of resilient and susceptible mice revealed differences in the oligodendrocytes of the medial prefrontal cortex , the part of the brain that controls emotions and thinking . Susceptible animals had fewer mature oligodendrocytes and their neurons were covered in thinner and shorter segments of myelin sheaths . There was also evidence that , in these animals , the genes that regulate the maturation of oligodendrocytes were more likely to be switched off . Taken together , these results may suggest that , in certain animals , social stress disrupts the genetic program that controls how oligodendrocytes develop , potentially leading to neurons communicating less well . To explore whether reduced amounts of myelin could be linked to decreased social behavior , Bonnefil et al . then damaged the myelin in the medial prefrontal cortex in another group of rodents . The mice were then less willing to interact with other animals until new sheaths had formed . The results by Bonnefil et al . undercover how changes in non-neuronal cells can at least in part explain differences in the way individuals respond to stress . Ultimately , this knowledge may be useful to design new strategies to foster resilience . | [
"Abstract",
"Introduction",
"Results",
"and",
"discussion",
"Materials",
"and",
"methods"
] | [
"research",
"communication",
"neuroscience"
] | 2019 | Region-specific myelin differences define behavioral consequences of chronic social defeat stress in mice |
GPCRs regulate all aspects of human physiology , and biophysical studies have deepened our understanding of GPCR conformational regulation by different ligands . Yet there is no experimental evidence for how sidechain dynamics control allosteric transitions between GPCR conformations . To address this deficit , we generated samples of a wild-type GPCR ( A2AR ) that are deuterated apart from 1H/13C NMR probes at isoleucine δ1 methyl groups , which facilitated 1H/13C methyl TROSY NMR measurements with opposing ligands . Our data indicate that low [Na+] is required to allow large agonist-induced structural changes in A2AR , and that patterns of sidechain dynamics substantially differ between agonist ( NECA ) and inverse agonist ( ZM241385 ) bound receptors , with the inverse agonist suppressing fast ps-ns timescale motions at the G protein binding site . Our approach to GPCR NMR creates a framework for exploring how different regions of a receptor respond to different ligands or signaling proteins through modulation of fast ps-ns sidechain dynamics .
Our understanding of the molecular underpinnings of GPCR function has been greatly advanced over the past two decades through a combination of X-ray crystal structures , computational simulations , and spectroscopic studies of protein dynamics . Crystals of bovine rhodopsin provided the first high-resolution picture of a GPCR’s architecture ( Palczewski et al . , 2000; Li et al . , 2004 ) , and structures of photoactivation intermediates ( Nakamichi and Okada , 2006; Salom et al . , 2006 ) and retinal-free opsin ( Park et al . , 2008; Scheerer et al . , 2008 ) further documented the structural transitions involved in rhodopsin activation . For GPCRs activated by diffusible ligands , crystal structures of the human β2 adrenergic receptor ( β2AR ) with inverse agonists ( Rosenbaum et al . , 2007; Cherezov et al . , 2007 ) , agonists ( Rosenbaum et al . , 2011; Rasmussen et al . , 2011a; Warne et al . , 2011 ) , and bound G protein ( Rasmussen et al . , 2011b ) provided a molecular basis for understanding how diffusible agonist binding can promote structural changes in a receptor to enhance signaling . Subsequent crystal structures have revealed the ligand binding pockets for GPCRs of diverse function , responding to biogenic amines , purines , lipids , peptides , and proteins . While the sequences of these GPCRs and their orthosteric ligand binding pockets are highly diverse , the overall structures are remarkably similar , with Cα rmsd values between unrelated receptors in the 2–3 Å range . How these diverse ligands activate signaling by a small number of G proteins and arrestins through a common structural scaffold remains a central problem for the GPCR field . Contributing to this problem is the fact that most of the existing X-ray GPCR structures are of antagonist-bound receptors locked in inactive conformations . For the few GPCRs that have been captured in both inactive and active conformations , common structural changes include inward movements of transmembrane ( TM ) helices at the orthosteric pocket , rearrangements of common core residues through the helical bundle , and rigid-body outward movements of TM6 at the cytoplasmic surface to expose G protein binding epitopes ( Katritch et al . , 2013 ) . Beyond the static pictures of GPCR conformations seen in these crystal structures , experimental evidence for complex dynamic behavior has emerged from spectroscopic studies on purified receptors . Unlike rhodopsin , which exhibits an efficient and ordered photon-induced transition from the dark state conformation to metarhodopsin II ( Choe et al . , 2011a; Choe et al . , 2011b ) , ligand-activated GPCRs generally show considerable basal activity that is reduced by inverse agonists . 19F NMR and EPR experiments on the β2AR ( Liu et al . , 2012a; Manglik et al . , 2015 ) and A2AR ( Ye et al . , 2016 ) provided evidence for at least four different conformational states populated by these receptors , which are differentially stabilized by ligands of different efficacy . 1H/13C 2D NMR studies on 13Cε-methionine-labeled β2AR ( Nygaard et al . , 2013; Kofuku et al . , 2014 ) showed that a high free-energy barrier exists between inactive and active conformations ( with exchange on the millisecond timescale ) , and agonist binding alone is only capable of weakly populating the active conformation . These studies , in addition to similar measurements on μ-opioid receptor chemically modified to introduce NMR-active 13C-methyl labels onto lysine residues ( Sounier et al . , 2015 ) , suggest that ligand-activated GPCRs are weakly coupled allosteric systems in which multiple energetic inputs ( i . e . agonist and G protein binding ) are required to predominantly populate the active conformation . Within the slow millisecond-timescale global interconversion between functional GPCR conformations ( Vilardaga et al . , 2003 ) , local structural rearrangements of sidechains in the receptor ( referred to here as ‘microswitches’ ) have been observed in comparisons of inactive and active GPCR crystal structures ( Katritch et al . , 2013 ) , NMR data ( Liu et al . , 2012a; Manglik et al . , 2015; Ye et al . , 2016; Bokoch et al . , 2010 ) and MD simulations ( Vanni et al . , 2009; Dror et al . , 2011a ) . The importance of repacking of sidechains from TM3 , TM5 , and TM6 beneath the ligand binding pocket , known as the ‘conserved core triad , ’ was identified by comparison of structures of β2AR ( Rosenbaum et al . , 2011; Rasmussen et al . , 2011a ) and the μ-opioid receptor ( Huang et al . , 2015 ) . Further down the TM bundle , an activated microswitch near the site of G protein binding has been found to represent a common feature of GPCRs that have been crystallized in both active and inactive conformations ( Venkatakrishnan et al . , 2016 ) . A working hypothesis for GPCR activation is that a set of loosely coupled microswitches connecting the orthosteric pocket and G protein binding site are activated by agonists in a non-concerted fashion , and stabilizing subsets of these rearrangements can lead to alternate overall conformations that have different signaling properties . One of the most powerful ways to probe sidechain dynamics and their contributions to biological processes is provided by NMR-based studies of isotopically-labeled methyl groups ( Ruschak and Kay , 2010 ) . To apply these techniques to study the sidechains in these microswitches , one approach would entail isoleucine/leucine/valine ( ILV ) labeling ( Goto et al . , 1999; Gardner et al . , 1998 ) and perdeuteration of a wild-type GPCR , to generate proteins with 1H/13C-labeled methyl groups within an otherwise 2H/12C-labeled background . Such samples are ideally suited for acquiring 1H/13C methyl TROSY-based ( Tugarinov et al . , 2003a; Ollerenshaw et al . , 2003 ) relaxation NMR data , allowing the quantitative determination of methyl order parameters and relative motion of sidechains on the ps-ns timescale ( Tugarinov et al . , 2005; Ishima and Torchia , 2000 ) . Advances in labeling methodology and pulse sequences have enabled such measurements on large macromolecular systems , such as the 80 kDa enzyme malate synthase ( Tugarinov and Kay , 2003b; Sandala et al . , 2007 ) , the 670 kDa archaeal proteasome core particle ( Sprangers and Kay , 2007; Religa et al . , 2010 ) , and the 1 MDa GroEL/GroES complex ( Fiaux et al . , 2002; Horst et al . , 2005 ) . NMR spectroscopic interrogation of methyl groups has also been used to dissect the energetic contribution of sidechain entropy to important biological phenomena such as protein-protein interactions ( Marlow et al . , 2010 ) and transcription factor binding to double-stranded DNA ( Tzeng and Kalodimos , 2012 ) . These previous applications suggest that characterization of sidechain dynamics using methyl NMR approaches should be feasible for a GPCR purified in a detergent micelle , with an aggregate molecular weight in the 100–150 kDa range . A major hurdle in carrying out such experiments is that perdeuteration and ILV labeling conventionally rely on expression in E . coli ( Goto et al . , 1999 ) , while wild-type GPCRs typically require a eukaryotic host to ensure proper folding , glycosylation , trafficking , and stability in the plasma membrane . GPCR NMR studies have primarily used supplementation with labeled amino acids ( Nygaard et al . , 2013; Kofuku et al . , 2014; Okude et al . , 2015; Isogai et al . , 2016; Kofuku et al . , 2012 ) or covalent modification ( Liu et al . , 2012a; Sounier et al . , 2015; Bokoch et al . , 2010; Eddy et al . , 2016 ) to incorporate isotopic probes into protein from Sf9 cells , and perdeuteration with incorporation of labeled ILV methyl probes is largely intractable for this established eukaryotic system . Another limitation is that purified wild-type GPCRs are typically prone to aggregation and are not highly thermostable , limiting acquisition times and sample concentration . While high quality 2D NMR spectra of GPCRs have been obtained for receptors with thermostabilizing mutations ( Isogai et al . , 2016 ) , such mutations can affect the receptor’s dynamics and functional properties , including basal activity and maximal G protein stimulation . We have addressed many of these practical issues with our generation ( Clark et al . , 2015 ) of highly-deuterated 1H/13C Ile δ1-methyl labeled proteins in Pichia pastoris , a methylotrophic yeast ( Cereghino and Cregg , 2000 ) that is adaptable to D2O-based minimal media , will utilize the Ile precursor α-ketobutyrate , and enables expression of GPCRs at levels sufficient for crystallography ( Shimamura et al . , 2011; Hino et al . , 2012; Yurugi-Kobayashi et al . , 2009 ) . Here we apply this labeling method to the wild-type human A2AR , an important regulator of vascular function with a well-developed pharmacology including the natural hormone adenosine and high-affinity synthetic agonists and antagonists ( Preti et al . , 2015; de Lera Ruiz et al . , 2014 ) . A2AR has been crystallized in inactive ( Jaakola et al . , 2008; Liu et al . , 2012b ) , intermediate ( Lebon et al . , 2011; Xu et al . , 2011 ) and active ( Carpenter et al . , 2016 ) conformations , and has been studied by 19F NMR to characterize its conformational equilibria ( Ye et al . , 2016 ) . We expressed and purified milligram quantities of labeled wild-type human A2AR and were able to resolve 20 out of 29 expected peaks in the Ile δ1 region in 1H/13C TROSY HMQC experiments . Through structure-guided mutagenesis , we assigned four of these signals to specific functionally important residues , including I923 . 40 of the conserved core triad ( Huang et al . , 2015 ) and I292 at the cytoplasmic surface . We collected spectra in the presence of the high-affinity full agonist NECA and the inverse agonist ZM241385 , and also measured the effects of monovalent cations ( Na+ versus K+ ) on these spectra . We further carried out a modified triple-quantum ( 3Q ) relaxation experiment ( Sun et al . , 2011 ) to quantify the relative fast ps-ns motions of these sidechains while the receptor is bound to ligands of opposing efficacy . These data represent an important first step towards understanding how agonists can activate the fast motions of specific sidechains to facilitate conformational changes of a GPCR .
The first challenge in collecting methyl NMR spectra on a wild-type GPCR was achieving high-level expression and milligram scale purification from Pichia pastoris . We initially cloned several GPCRs ( including A2AR ) into a modified methanol-inducible expression vector with the wild-type MFα signal sequence at the N-terminus to direct incorporation into the membrane . Several receptors , such as the wild-type M2 muscarinic receptor and A2AR , showed reliable and reproducible expression , however much of the expressed proteins were present in the immature unprocessed form ( without MFα cleavage ) indicating that they had not been properly localized to the plasma membrane ( Figure 1—figure supplement 1 ) . To improve the efficiency of production of mature , folded receptors , we made several modifications to the MFα sequence , based on biochemical and genetic precedents ( Rakestraw et al . , 2009; Lin-Cereghino et al . , 2013 ) for improved processing efficiency ( Figure 1A , Materials and methods ) . To validate that Pichia-expressed A2AR is functional , we carried out radioligand binding assays measuring [3H]-ZM241385 saturation and NECA competition on membranes after cell wall disruption and zymolyase treatment . The Kd for [3H]-ZM241385 was 0 . 42 nM in NaCl-containing buffer , while the Ki for NECA was 210 nM in low ionic strength and 420 nM in KCl-containing buffer , in agreement with previous binding studies ( Xu et al . , 2011; Carpenter et al . , 2016; Bertheleme et al . , 2013 ) on heterologously expressed A2AR ( Figure 1—figure supplement 2 ) . We developed a purification protocol for A2AR from Pichia-derived membranes , in which we can solubilize and purify ~0 . 5 mg of monodisperse biochemically pure wild-type A2AR in dodecylmaltoside ( DDM ) detergent from 1 L of culture ( Figure 1—figure supplement 1 ) . Coupling this protocol with our previously-described labeling method ( Clark et al . , 2015 ) , we were able to generate samples of wild-type A2AR that are highly deuterated both within the protein ( 2H at non-exchangeable sites except Ile δ1 methyl groups ) and the surrounding buffer ( 99% D2O and deuterated DDM ) . We prioritized the use of a wild-type A2AR construct to report on the dynamics of a physiologically and functionally relevant GPCR sample . To confirm that the high level of deuteration did not alter the receptor’s functional properties , we reconstituted the perdeuterated purified A2AR together with purified Gs heterotrimer in phospholipid vesicles and carried out activation assays by [35S]GTPγS binding . Saturating the receptor with the full agonist NECA led to an 8-fold stimulation of [35S]GTPγS binding , and the inverse agonist ZM241385 produced a 50% reduction in basal activity , similar to values measured for protonated A2AR ( Figure 1B ) . These data show that the perdeuterated A2AR is capable of activating Gs at wild-type levels in vitro , and can serve as a valid model for GPCR dynamics . Our first NMR characterization of A2AR used 1H/13C TROSY HMQC spectra of the protein in the presence of different ligands or Gs . We initially expressed and solubilized the receptor with the low-affinity antagonist theophylline , and exchanged other ligands onto the receptor during washing in the Co2+-affinity and gel filtration steps ( Materials and methods ) . In this way we were able to purify A2AR samples with the inverse agonist ZM241385 , the full agonist NECA , or no ligand . The resulting 2D HMQC spectra , requiring ~100 μM GPCR and a minimum of 2 hr acquisition at 30°C on a cryoprobe-equipped 800 MHz spectrometer , are shown in Figure 2 . Overall , in a liganded sample , we could resolve 20 peaks in the Ile δ1 region of the spectrum , out of 29 Ile residues present in our construct ( 28 from the receptor and one from the C-terminal Protein C tag ) . The chemical shifts of the agonist-bound and inverse agonist-bound A2AR are similar ( Figure 2A , B ) , showing subtle changes that are comparable in magnitude to those observed previously for 13C-methyl-methionine probes in β2AR ( Kofuku et al . , 2014; Kofuku et al . , 2012 ) . The spectrum of the unliganded ( apo ) A2AR shows a loss of most of the well-dispersed peaks present in the ligand-bound spectra ( Figure 2C ) , which could result from protein instability or conformational exchange on the μs-ms timescale . Lastly , we formed and isolated a high-affinity complex between purified NECA-bound A2AR and purified Gs heterotrimer in the absence of GDP or GTP ( Figure 2—figure supplement 1 , Materials and methods ) . Like the apo spectrum , the G protein complex spectrum had fewer well-dispersed Ile peaks than receptor/ligand complexes ( Figure 2D ) , consistent with the increase in molecular weight of the complex . We also note the appearance of a few distinct peaks ( most prominently at 0 . 85 ppm 1H , 11 . 8 ppm 13C ) in the G protein complex , consistent with the conformation of the complexed receptor being distinguishable from the apo receptor ( Rasmussen et al . , 2011b; Carpenter et al . , 2016 ) . Assignment of 1H/13C peaks in the Ile δ1 region by NMR was hampered by the modest dispersion of our spectra and the limited stability of A2AR during data acquisition . Therefore , we took a site-specific mutagenesis approach , wherein we collected HMQC spectra of samples containing point mutations at a small subset of functionally important Ile sites ( Figure 2 and Figure 2—figure supplement 2 ) . This process was complicated by the sensitivity of the receptor to mutations at several of these Iles: changes to Val , Leu , or Met often drastically reduced purification yields and/or led to significantly altered spectra; for an example , see the I106V spectra in Figure 2—figure supplement 2 . Three of the assignments ( I923 . 40 , I2386 . 40 , and I292 ) could be made primarily through observation of ZM241385-bound mutant spectra . By comparing WT and I92V spectra in the presence of ZM241385 , we see a clear loss of signal at ~0 . 68ppm/13 . 6ppm ( 1H/13C ) that we can assign to the δ1 methyl of Ile92 . The NECA-bound I92V spectrum is missing a subset of peaks compared to the ZM241385-bound spectrum , possibly due to compromised affinity of NECA for the I92V mutant ( Figure 1—figure supplement 2D ) . Ile238 and Ile292 are well-dispersed peaks , allowing for unambiguous assignment through loss of peaks at 0 . 72 ppm/10 . 2 ppm and 1 . 12 ppm/12 . 7 ppm , respectively ( 1H/13C ) . The mutant spectra of I2386 . 40 and I292 have reciprocal effects on each other , where I238V causes a full peak-width shift in the I292 signal ( upfield in 1H ) , and I292L causes a ~ 0 . 5 peak-width shift in the I238 signal ( upfield in 13C ) . The reciprocity between these mutant spectra is consistent with the proximity of these residues in the A2AR structure . Mutation of I2747 . 39 in the ligand binding pocket ( within van der Waals contact of the ligands in the respective crystal structures ) to Val or Met led to a disappearance of some peaks , reminiscent of the apo spectrum . However , comparison of ZM241385-bound I274M and I274V spectra narrowed the choice of the assignment of I2747 . 39 to one of two peaks , with the greater solvent accessibility as revealed by solvent PRE effect ( see below and Figure 4 ) of one peak serving as the deciding factor of assignment at 0 . 55 ppm/14 . 1 ppm ( 1H/13C ) . As well as disappearance of many peaks , the spectrum of I274M A2AR in the presence of ZM241385 displays several apparent peak doublings , suggesting that the mutant protein is undergoing slow conformational exchange , potentially related to the slow interconversion between different inactive states of the receptor ( Ye et al . , 2016 ) . In addition , we assigned a peak to the Ile residue of the C-terminal Protein C tag in our construct by comparing spectra between differently tagged constructs ( not shown ) . The remaining unassigned peaks are referred to as a-o in the rest of this work ( Figure 3A ) . The chemical shift changes between ZM241385-bound and NECA-bound A2AR are small overall ( Figure 2A , B ) . In the A2AR-Gs complex spectrum ( Figure 2D ) , the peak for I923 . 40 appears to shift and simultaneously drop in intensity , strongly suggesting that I923 . 40 is involved in conformational changes on several timescales while the receptor is bound to Gs . A2AR contains a well-characterized Na+ binding site within the receptor core ( Liu et al . , 2012b ) , and Na+ has been shown to act as a negative allosteric modulator of A2AR activation ( Gao and Ijzerman , 2000 ) . Due to the presence of 150 mM NaCl in the experiments shown in Figure 2 , one possible explanation for the relatively small chemical shift differences between NECA- and ZM241385-bound spectra ( Figure 3A , B ) is that high [Na+] is suppressing structural changes in the agonist-bound receptor . To test this possibility , we acquired HMQC spectra for A2AR samples with 150 mM KCl substituting for NaCl . The spectrum of ZM241385-bound receptor in KCl is very similar to that in the presence of NaCl ( Figure 3C and Figure 3—figure supplement 1B ) , with the peak assigned to I923 . 40 shifting slightly upfield in the 13C dimension . The spectrum of NECA-bound A2AR , on the other hand , is dramatically different in KCl and NaCl ( Figure 3D , Figure 3—figure supplement 1D ) , with multiple peaks shifting and decreasing in intensity . However , the KCl samples were significantly less stable than those in NaCl , showing markedly decreased spectral quality after 1 ( NECA-bound ) or 2 ( ZM241385-bound ) hours of acquisition . Given the weak signal-to-noise and poor stability of the KCl samples , we were only able to perform NMR relaxation experiments ( below ) on A2AR in the presence of NaCl . In addition to ligand-induced chemical shift changes , we used our system to characterize the solvent accessibility of residues within the ZM241385-bound A2AR ( with NaCl ) using solvent paramagnetic relaxation enhancements ( PREs ) . To do so , we used the soluble paramagnetic Gd3+-DTPA probe ( Petros et al . , 1990 ) , which enhances the relaxation rates of protons in a distance dependent manner within roughly 15 Å of the protein/solvent interface ( Madl et al . , 2011 ) . By acquiring 1H/13C TROSY HMQC spectra in the absence and presence of Gd3+-DTPA , we could compare peak intensities between the two to establish solvent PRE levels . Examining our data , we see that the average peak intensity in the Gd3+-DTPA containing spectrum is about 0 . 4 of that in the control spectrum , suggesting that the Ile δ1 methyl groups of A2AR in the DDM micelle are quite close to solvent on average ( Figure 4 ) . Of the four most strongly affected peaks , three peaks correspond to assigned residues ( I2386 . 40 , I2747 . 39 , and I292 ) . The accessibility of I2747 . 39 is expected due this amino acid’s presence in the solvent-exposed ligand-binding pocket . Broadening of I292 at the junction between TM7 and Helix 8 is also somewhat expected due to its position at the cytoplasmic surface . The broadening observed for I2386 . 40 is more surprising , given that it is further embedded within the membrane and buried within the protein surface of A2AR . Crystal structures of multiple GPCRs , including A2AR , β2AR , μOR , and rhodopsin have all revealed ordered solvent networks within the protein core in contact with the position equivalent to I2386 . 40 . In ZM241385-bound A2AR , this residue is also in close proximity to the Na+ site ( Liu et al . , 2012b ) , one packing layer toward the cytoplasmic surface . The solvent PRE effect seen for this residue could reflect breathing of the structure to expose this region to the Gd3+-DTPA complex . To characterize the motions of the Ile sidechains within A2AR , we sought to carry out a triple quantum ( 3Q ) relaxation experiment developed by Kay , Tugarinov and colleagues ( Sun et al . , 2011 ) to quantify sidechain dynamics in large macromolecules . In this experiment , two related series of 2D 1H/13C HMQC-based spectra were acquired , both of which start by generating transverse 1H magnetization composed of single quantum ( SQ ) coherences . During a subsequent variable delay period , intra-methyl 1H−1H dipolar cross-correlated relaxation mechanisms lead to a portion of these SQ coherences evolving into 3Q coherences; the rate of this conversion ( η ) is proportional to the S2axis methyl order parameter . Practically , the degree of SQ->3Q evolution can be quantitated by examining paired ‘forbidden’ and ‘allowed’ spectra , which report on the fraction of SQ coherences which do and do not evolve into 3Q coherences , respectively . As previously reported ( Sun et al . , 2011 ) , this allows the simple calculation of the η rate for each specific methyl group from the ratio of peak intensities in the forbidden and allowed spectra . Our initial attempts to apply this method to purified A2AR were hampered by the fact that the wild-type receptor is insufficiently stable at 30°C to permit the measurement of the requisite paired spectra at many relaxation delays and with a high enough signal-to-noise ratio . We therefore explored the possibility of reducing the number of forbidden spectra required to accurately measure η values . Using similarly 2H , 12C ( 1H , 13C δ1 methyl ) labeled maltose binding protein ( MBP ) as a test case ( Materials and methods ) , we collected 3Q datasets with different numbers of forbidden spectra paired with a constant number of allowed spectra ( Figure 5 , Figure 5—figure supplement 1 ) . We found that pairing five allowed spectra with a single forbidden spectrum , we were able to faithfully recapitulate the η values measured using five pairs of forbidden and allowed experiments ( Figure 5B , C ) . This η value is proportional to the S2axis order parameter , which quantifies the amplitude of motion of the methyl group on the ps-ns timescale that is faster than global molecular tumbling . With this modified analysis of 3Q relaxation data and our Ile δ1 1H/13C-methyl-labeled and deuterated GPCR samples , we measured η values for Ile sidechains in A2AR bound to either ZM241385 or NECA in NaCl . The resulting values for the 20 Ile peaks and relative changes between the samples with different ligands are shown in Figure 6 . The error in these measurements is large due to the limited signal-to-noise of our A2AR spectra , and we did not attempt to convert these η measurements to S2axis values . However , in comparing the two datasets some features of the Ile dynamics can be discerned . The average η value for the ZM241385-bound sample is higher than for NECA , indicative of greater overall rigidity of sidechains with inverse agonist . At the individual residue level , we observe a diversity of ligand-dependent changes in η , with some Ile residues becoming more rigid with agonist while most become more flexible . As an internal control , the peak we assigned to the C-terminal Protein C tag shows the lowest η value measured ( i . e . greatest flexibility ) with little ligand-dependence . Among the other assigned peaks , I292 shows the largest difference between ligands , becoming more flexible in the NECA-bound sample . I923 . 40 and I2747 . 39 display more modest increases in flexibility with agonist , while the dynamics of I2386 . 40 are largely unchanged by ligand ( Figure 6B ) . Interestingly , these differences in fast timescale dynamics occur in the presence of NaCl , in which chemical shifts for the agonist- and inverse agonist-bound states were nearly identical . To our knowledge , these data represent the first reported effort to experimentally quantify site-specific sidechain dynamics in a GPCR or comparable human integral membrane protein . While we were able to measure sidechain dynamics in this challenging A2AR sample , the associated errors are quite large . As such , we turned to molecular dynamics ( MD ) simulations to provide independent validation of our dynamics measurements . We extracted S2axis order parameters for Ile δ1 methyl groups from ~80 ns trajectories ( post-equilibrium ) of ZM241385- and NECA-bound A2AR in DDM micelles , using overlapping 30-ns windows to estimate the standard deviation in the order parameters ( Figure 6—figure supplement 1 ) . The S2axis order parameters for some methyl groups ( for example , I60 and I80 in ZM241385; I106 and I302 in NECA ) show significant variations across different windows , suggesting that slower timescale motions involving these residues contribute to the sidechain dynamics measured on the fast timescale . Interestingly , the differences between ZM241385- and NECA-bound dynamics at three of the four assigned sites ( I2386 . 40 , I2747 . 39 , and I292 ) are qualitatively similar to what we see by NMR . The fourth site , I923 . 40 , shows the opposite trend , with higher S2axis , i . e . more rigidity , in the NECA-bound state . The simulations were set up with bulk Na+ ions , but none specifically occupying the binding site near I923 . 40 , which could account for this discrepancy .
Sidechain dynamics represent an important functional component of protein behavior . In cases of protein-ligand ( Marlow et al . , 2010; Frederick et al . , 2007 ) and protein-DNA ( Tzeng and Kalodimos , 2012 ) interactions that have been intensively studied , changes in entropy arising from modified sidechain dynamics in complexes were found to substantially contribute toward the overall free energy of binding . For GPCRs , changes in sidechain dynamics may play an energetic role in binding to ligands and G proteins , however we have focused on allosteric mechanisms connecting these two functionally important binding sites . To experimentally assess the roles of sidechain dynamics in a wild-type GPCR , we set out to create a labeled sample that would be amenable to relaxation NMR methods . The spectra we obtained for the labeled Ile δ1 methyl groups of perdeuterated A2AR are comparable or superior to previously published NMR spectra on other wild-type GPCRs , however the limited dispersion and signal-to-noise contributed to significant error in the relaxation values derived from the data ( Figure 6A ) . The Na+ site first seen in the high-resolution structure of A2AR bound to ZM241285 is conserved throughout most Class A GPCRs , and Na+ exerts a negative allosteric effect on A2AR activation by bridging residues on TM3 and TM7 and stabilizing the inactive conformation ( Liu et al . , 2012b ) . This effect can be observed pharmacologically as a NaCl-dependent decrease in agonist affinity ( Carpenter et al . , 2016; Gao and Ijzerman , 2000 ) . For A2AR expressed in P . pastoris , we observe a 9-fold decrease in NECA affinity when membranes are incubated in 150 mM NaCl buffer versus 150 mM KCl buffer ( Figure 1—figure supplement 2 ) , similar to previously reported values ( Carpenter et al . , 2016 ) . Several of the Ile residues that we assigned ( i . e . I923 . 40 and I2386 . 40 ) are in close proximity to the Na+ binding site , which should make for sensitive probes of local structure in this part of the receptor core . Our 2D NMR spectra indicate a strong dependency on combined lack of Na+ and binding of agonist to stabilize significant structural changes in the receptor core ( Figure 3 ) . This observation is also consistent with the active conformation seen in the crystal structure of the A2AR/mini-Gs complex ( Carpenter et al . , 2016 ) , which results in collapse of the Na+ site due to inward movements of TM3 and TM7 . In contrast to our data , the crystal structure of a NECA-bound thermostabilized A2AR mutant construct ( Lebon et al . , 2011 ) showed significant structural changes relative to the inactive conformation even with a high [Na+] beyond its EC50 ( ~50 mM ) for negative allosteric modulation ( Gao and Ijzerman , 2000 ) . Unfortunately the weak signal and stability of our samples in KCl precluded measurement of 3Q relaxation dynamics under these conditions . For the Na+-bound A2AR samples , we can correlate our observations of ligand-dependent changes in the relaxation rate of specific sites to previous structural and biophysical studies of GPCRs . I2747 . 39 at the ligand binding site makes direct contact with the adenine or adenine analogue rings of both NECA and ZM241385 in their respective crystal structures ( Jaakola et al . , 2008; Liu et al . , 2012b; Lebon et al . , 2011; Carpenter et al . , 2016 ) ( Figure 7A ) . In our relaxation datasets , I2747 . 39 was more rigid in terms of its fast sidechain motions with inverse agonist bound , and more flexible with NECA occupying the binding pocket . NMR experiments previously showed different ligand-dependent conformations of the extracellular surface region of β2AR bound to agonists compared to inverse agonists ( Bokoch et al . , 2010 ) . In addition , MD simulations on β2AR predicted greater mobility of agonists relative to inverse agonists in the orthosteric pocket of the inactive-state structure ( i . e . not bound to Gs ) ( Dror et al . , 2011b ) . Our data suggests that the fast dynamics of residues at the orthosteric pockets of Class A GPCRs may be correlated with the functional efficacy of the bound ligand . The residue I923 . 40 is one of the three residues of the conserved core triad in A2AR , along with P1895 . 50 and F2426 . 44 , which interact at a layer beneath the orthosteric pocket further toward the cytoplasmic surface ( Figure 7B ) . Consistent with the rearrangements of this microswitch region between the inactive and active conformations of multiple GPCRs , we observe a lower η value for I923 . 40 in the NECA dataset versus ZM241385 , indicating greater sidechain flexibility when A2AR is agonist-bound . In several GPCRs , including β2AR and μOR , the conserved core triad is unchanged at a static structural level with an agonist bound but without a G protein or nanobody to further stabilize the active conformation . In contrast , crystal structures of A2AR bound to agonists alone ( Figure 7B ) showed an intermediate active-like conformation in the region surrounding I923 . 40 ( Lebon et al . , 2011; Xu et al . , 2011 ) . Our data indicate that low [Na+] is required for NECA alone to stabilize structural rearrangements surrounding I92 ( Figure 3 , Figure 3—figure supplement 1 ) . However agonist alone ( even with high [Na+] ) is enough to at least promote the fast motions of the sidechains in this microswitch , which may reduce the activation energy for the observed packing rearrangement to the active conformation . The residue I2386 . 40 is situated further toward the G protein binding site at a critical region for GPCR activation , where it packs against TM7 and undergoes significant outward movement in the transition to the active conformation ( Figure 7C ) . I2386 . 40 is also one helical turn on TM6 above L2356 . 37 , which participates in a conserved microswitch between inactive and active conformations for multiple GPCRs ( Venkatakrishnan et al . , 2016 ) . I292 is present at the linker between TM7 and Helix 8 , where it packs against Y2887 . 53 of the highly conserved NPXXY motif ( Katritch et al . , 2013 ) ( Figure 7D ) . NMR studies of 13C-dimethyllysine-labeled μOR showed that peak broadening of a lysine probe in Helix 8 ( near the position equivalent to I292 ) was more sensitive to agonist than probes at the cytoplasmic ends of TM5 and TM6 ( 21 ) . Further , in the structure of A2AR/mini-Gs ( Carpenter et al . , 2016 ) , the engineered Gs protein makes direct contact with I292 at the TM7-Helix 8 junction . In our relaxation dataset , the η value of I2386 . 40 is low ( i . e . more flexible ) and essentially independent of the ligand , while I292 undergoes a large change from more rigid to more flexible from ZM241385 to NECA . As mentioned above , the environment surrounding I2386 . 40 is loosely packed in A2AR structures , with an ordered solvent network that may allow for relative freedom of motion for this sidechain . In contrast , despite the solvent exposure of I292 ( Figure 4 ) , inverse agonist binding suppresses the fast dynamics of this residue relative to agonist binding ( Figure 6 ) . Outward movement of TM6 is one of the hallmarks of GPCR activation seen in crystal structures , and the cytoplasmic ends of TM6 and TM7 are separated from each other in the structures of β2AR and A2AR bound to Gs ( Rasmussen et al . , 2011b; Carpenter et al . , 2016 ) . Our data suggest that the responsiveness of sidechain dynamics to ligands in these two regions may be largely uncoupled , and that the inverse agonist activity of ZM241385 could partly arise from its allosteric suppression of dynamics at the cytoplasmic end of TM7 . Changes in η for other peaks in our dataset may provide further insights into the regulation of different regions of the receptor by ligands , depending on assignment of the other Ile residues in our spectra . Beyond our studies of the A2AR , we can now apply the methods for labeling and NMR spectroscopy described here to other GPCRs and eukaryotic membrane proteins . Since the microswitches discussed above were identified by comparison of different GPCR structures ( Huang et al . , 2015; Venkatakrishnan et al . , 2016 ) , it will be instructive to apply our methods to other receptors to see if the same patterns of ligand-dependent changes apply across the GPCR superfamily or change depending on the cognate ligand or preferred G protein signaling partner . Further , sidechain dynamics may be among the biophysical properties of GPCRs that are altered when receptors are bound to allosteric modulators or biased agonists , classes of ligands that are of increasing value and importance in GPCR pharmacology and drug development . In addition to GPCRs , many other disease-relevant human integral membrane proteins ( such as ABC transporters and ion channels ) are currently impossible to study by NMR methods , largely due to the challenges of expression , labeling , and perdeuteration in E . coli . Our approach has the potential to make these systems tractable for similar NMR measurements of sidechain dynamics .
The cDNA for wild-type human ADORA2A adenosine A2A receptor was cloned into the pPICZ vector for expression in Pichia pastoris with a modified MFα secretion signal combining previous precedents ( Rakestraw et al . , 2009; Lin-Cereghino et al . , 2013 ) . Modifications to the MFα greatly increased the amount of fully-processed receptor present at the plasma membrane . Briefly , the mutations V22A , G40D , L42S , V50A , V52A , and F55L were introduced , and residues 57–70 were removed from the signal sequence . The receptor expression construct was terminated at residue 316 and the N-glycosylation site at Asn154 was mutated to glutamine ( Jaakola et al . , 2008 ) . The gene for the receptor was followed by a 8x His tag and a Protein C tag at the C-terminus . All point mutations were created using standard QuikChange protocols . Plasmids were linearized by incubation with PmeI ( NEB ) and inserted via electroporation into freshly prepared competent KM71H cells ( Invitrogen , https://www . thermofisher . com/order/catalog/product/C18200 ) . Clones were screened for integration efficiency with increasing amount of Zeocin ( Invitrogen ) in selection media and further selected through expression screens and western blots . The best expressing clones were stored as glycerol stocks at −80°C . As these yeast clones were only used for protein expression/purification and membrane preparation for binding assays , no authentication or mycoplasma contamination testing was performed . For large scale growth in natural abundance media , a small amount of freshly streaked cells was inoculated into a 10 mL culture of BMGY media ( 1% glycerol , 100 mM potassium phosphate pH 6 . 0 , 1 . 34% YNB ( yeast nitrogen base ) , 0 . 004% histidine , 4 × 10−5% biotin ) and shaken at 28°C overnight at 250 rpm . The pilot culture was used to inoculate multiple liters of BMGY and shaken until saturation is reached ( OD600 ~20–30 ) . The total culture was spun down in sterile bottles at 4000 rpm for 30 min and resuspended in equal volume of BMMY media ( 100 mM potassium phosphate pH 6 . 0 , 1 . 34% YNB , 0 . 004% histidine , 4 × 10−5% biotin ) without methanol . Cultures continued shaking for ~8 hr at 28°C to allow for metabolism of residual glycerol . Protein expression was induced with the addition of 0 . 5% v/v methanol and the temperature was reduced to 20°C . An additional 0 . 5% methanol was added every 12 hr to maintain expression . Cells were harvested after 36–48 hr and pelleted by centrifugation . Pellets were stored at −80°C . For expression of cells in deuterated cultures , the cells were first adapted to deuterated media as follows . A small amount of freshly streaked cells was inoculated into a 50 mL culture of BMGY containing 90% D2O/10% H2O ( Cambridge Isotope Laboratories , Inc . , Tewksbury MA ) . The culture was shaken at 28°C until an OD600 of 8–10 was reached ( typically ~24 hr ) . Once that OD was reached , 200 μL of the 90%/10% culture was inoculated into 50 mL of BMGY media made with 100% D2O and protonated glycerol . The culture was shaken at 28°C until an OD600 of 8–10 was reached ( typically ~48 hr ) . 200 μL of the 100% culture was inoculated into 50 mL of BMGY media made with 100% D2O including d8-glycerol as the carbon source ( Cambridge Isotope Laboratories , Inc . ) . This culture was incubated until reaching an OD600 of ~10 and the entire culture was inoculated into large scale cultures of BMGY media again made with 100% D2O including d8-glycerol . The large scale cultures were shaken until saturation ( OD600 of ~20–30 ) and then spun down in sterile bottles at 4000 rpm for 30 min . The cells were resuspended in BMMY made in 100% D2O without methanol and continued shaking for 12–16 hr to metabolize residual d8-glycerol . One hour prior to induction , 200 mg/L of labeled α-ketobutyric acid ( methyl-13C , 99%; 3 , 3-D2 , 98%; Cambridge Isotope Laboratories , Inc . ) was added to the culture . Ten minutes prior to induction , dry theophylline was added to the culture to a final concentration of 4 mM . Protein expression was induced with the addition of 0 . 5% d4-methanol ( Cambridge Isotope Laboratories , Inc . ) and the temperature was reduced to 20°C . Expression was maintained by further additions of 0 . 5% d4-methanol every 12 hr , and cells were harvested by centrifugation after 36–48 hr and stored at −80°C . Cell pellets were thawed and resuspended in lysis buffer ( PBS containing 10% glycerol , 4 mM theophylline , 2 mM EDTA , and protease inhibitors ( 160 μg/mL benzamidine , 2 . 5 μg/mL leupeptin , 1 mM PMSF , 1 μM E-64 ) ) . Cells were passed through a high-pressure microfluidizer ( Microfluidics M-110P ) three times at 24 , 000 psi with a cooling period between passes . LongLife Zymolyase ( G-Bio Sciences ) was added to the total lysate at a concentration of 15 U/mL and stirred at 37°C for 1 hr . Total membranes were isolated by centrifugation at 140 , 000 rcf for 30 min and then washed by douncing in an equal volume of lysis buffer followed by an additional centrifugation step . Membranes were then resuspended in hypotonic buffer ( 10 mM HEPES pH 7 . 5 , 2 mM EDTA , 4 mM theophylline , protease inhibitors ) by douncing and stirred at 4°C for 30 min , followed by centrifugation again to isolate membranes . Membranes were dounced in buffer containing 500 mM NaCl , 50 mM HEPES pH 7 . 5 , 20% glycerol , 1% DDM ( Anatrace ) , 4 mM theophylline , and protease inhibitors and stirred for 2 hr at 4°C . Insoluble material was spun out by centrifugation at 140 , 000 rcf for 30 min . The resultant supernatant was incubated with TALON resin ( Clontech , Mountain View CA ) pre-equilibrated in 250 mM NaCl , 50 mM HEPES 7 . 5 , 0 . 05% protonated DDM , 5% glycerol , 4 mM theophylline , and 30 mM imidazole . Additional imidazole was added to the supernatant to the final concentration of 30 mM to minimize background binding . Batch binding continued overnight at 4°C . Following batch binding , the resin was washed with a series of buffers to exchange the protonated DDM into deuterated DDM and exchange on the high affinity ligands ZM241385 or NECA . Buffers were made in D2O and all contain 250 mM NaCl , 50 mM HEPES pH 7 . 5 , 5% glycerol , 0 . 05% DDM ( protonated or deuterated; Anatrace ) , 20 mM imidazole , and 10 μM ZM241385 or 20 μM NECA ( Tocris ) . Buffers contain different ratios of protonated:deuterated detergents , and were added sequentially: ( A ) 4:0; ( B ) 3:1; ( C ) 2:2; ( D ) 1:3; ( E ) 0:4 . Protein was eluted from TALON with buffer E + 250 mM imidazole . Eluted A2AR was concentrated in 100 kDa MWCO Amicon concentrators ( Millipore ) and injected on a Superdex200 column ( GE Healthcare , Chicago IL ) equilibrated in 150 mM NaCl , 20 mM HEPES pH 7 . 5 , 0 . 05% deuterated DDM , and 10 μM ZM241385 or 20 μM NECA made in D2O . Samples for KCl experiments were purified in the same manner , only substituting 150 mM KCl instead of NaCl in the SEC buffer . Previous studies have shown the importance of using cholesterol hemisuccinate ( CHS ) as a component of the micelle to preserve A2AR function ( Weiss and Grisshammer , 2002 ) and increase the thermostability of the receptor ( Liu et al . , 2012b ) . However , when CHS was included in our purification , we observed significant artifacts in the Ileδ1 region of our NMR spectra that hampered data collection and analysis . Ye et al . utilized an XAC ligand affinity column as an additional chromatography step during purification ( Ye et al . , 2016 ) that we did not include in this study . Our purification is similar to what has been published in structural studies of A2AR , following a general purification scheme of IMAC followed by size exclusion chromatography prior to crystallization in detergent ( Hino et al . , 2012; Carpenter et al . , 2016; Sun et al . , 2017 ) or lipidic cubic phase ( Jaakola et al . , 2008; Liu et al . , 2012b; Xu et al . , 2011 ) . These structures contain A2AR bound to agonists , antagonists , or in complex with an engineered ‘mini-Gs’ protein . Given our biochemical evidence ( Figure 1—figure supplement 1 , Figure 2—figure supplement 1 ) , we are confident that our purification protocol outlined above produces a high level of folded and functional A2AR , which is able to efficiently bind to GS in solution . Gs heterotrimer was purified from Trichoplusia ni cells grown in ESF921 media ( Expression Systems ) . Cells were lysed in hypotonic buffer containing 10 mM Tris pH 7 . 4 , 100 μM MgCl2 , 5 mM β-mercaptoethanol , 10 μM GDP , and protease inhibitors . After centrifugation , membranes were dounced and solubilized in buffer containing 100 mM NaCl , 20 mM HEPES pH 7 . 5 , 1% sodium cholate , 0 . 05% DDM , 5 mM MgCl2 , 5 mM β-mercaptoethanol , 10 μM GDP , 5 mM imidazole , protease inhibitors , and a 1:150 , 000 vol dilution of CIP ( NEB ) . The soluble material was incubated with pre-equilibrated Ni2+-NTA resin for 2 hr at 4°C . Detergent exchange into deuterated DDM was performed on-column by mixing volumes of buffer E1 ( 100 mM NaCl , 20 mM HEPES pH 7 . 5 , 1% sodium cholate , 0 . 05% deuterated DDM , 5 mM MgCl2 , 5 mM β-mercaptoethanol , 10 μM GDP , 20 mM imidazole , protease inhibitors ) and buffer E2 ( 50 mM NaCl , 20 mM HEPES pH 7 . 5 , 0 . 1% deuterated DDM , 1 mM MgCl2 , 5 mM β-mercaptoethanol , 10 μM GDP , 20 mM imidazole , protease inhibitors ) in the following E1:E2 ratio: 1:1; 1:3; 1:9; 1:19 at a flow rate of 1 mL/min . Gs was eluted with buffer E2 supplemented with 200 mM imidazole . Following elution , MnCl2 was added to a final concentration of 1 mM , and the pooled eluate ( ~10 mL volume ) was incubated with 10 μL lambda phosphatase , 1 μL CIP , and 1 μL Antarctic phosphatase for 30’ at 4°C . Sample was diluted with buffer E2 to reduce imidazole concentration and applied to a pre-equilibrated 2 mL Q-sepharose column . The resin was washed with 6 CV of 100 mM NaCl , 20 mM HEPES pH 7 . 5 , 0 . 04% deuterated DDM , 1 mM MgCl2 , 100 μM TCEP , 10 μM GDP and eluted with wash buffer supplemented with an additional 250 mM NaCl . Eluted sample was concentrated in a 10 kDa MWCO concentrator to 1 mL and diluted 1:1 with buffer containing 20 mM HEPES pH 7 . 5 , 1 . 5 mM EDTA , 1 . 65 mM MgCl2 , 100 μM TCEP , 1 μM GDP to reduce final DDM and NaCl concentrations . Glycerol was added to 20% , and aliquots were flash frozen in LN2 and stored at −80°C until needed . Functional Gs was conservatively estimated to constitute at least 40% of the total purified sample . A2AR-Gs complex formation was carried out as follows . SEC-purified A2AR was mixed with Gs at a 1:8 w/w ratio in the presence of 1 mM EDTA , 3 mM MgCl2 , 10 μM NECA , 1 mM MnCl2 , and a 1:100 dilution of lambda phosphatase ( NEB ) . The complex was incubated at room temperature for 1 hr , then apyrase ( NEB , Ipswich MA ) was added at a 1:2000 vol dilution followed by an additional hour incubation at room temperature . CaCl2 was added to a final concentration of 2 mM and loaded at a flow rate of 10 mL/hr on Protein C antibody resin ( Sigma-Aldrich , St . Louis MO ) pre-equilibrated in 150 mM NaCl , 20 mM HEPES pH 7 . 5 , 0 . 05% DDM , 2 mM CaCl2 , 10 μM NECA . ProC resin was washed with 1 CV buffer at 10 mL/hr , followed by 5 CV buffer at 30 mL/hr . Complex was eluted in 2 CV with 150 mM NaCl , 20 mM HEPES pH 7 . 5 , 0 . 05% DDM , 10 μM NECA , 5 mM EDTA , 0 . 2 mg/mL ProC peptide ( EDQVDPRLIDGK ) . Freshly made TCEP was added to a final concentration of 100 μM and was spin-concentrated prior to injection on a Superdex200 column equilibrated in 150 mM NaCl , 20 mM HEPES pH 7 . 5 , 0 . 05% DDM , and 10 μM NECA . A2AR-Gs complex formation for NMR proceeded as above with a few modifications . Gs heterotrimer purification was carried out as above with the final exchange steps containing deuterated DDM instead of protonated DDM . Deuterated , labeled , SEC-purified A2AR was mixed with Gs at a 1:8 w/w ratio in the presence of 3 mM MgCl2 in a final volume of 1 mL . The complex was incubated at room temperature for five minutes , and then a 1/2000 vol of apyrase ( NEB ) was added . The complex was incubated for an additional hour at room temperature , and then was injected on a Superdex200 column incubated in 150 mM NaCl , 20 mM HEPES pH 7 . 5 , 0 . 05% deuterated DDM , 20 μM NECA . Peak fractions were concentrated to ~100 μL for NMR experiments . NMR spectra were collected at 30°C on 50–100 μM A2AR ( complexed with ZM241285 or NECA ) and 100 μM MBP ( complexed with β-cyclodextrin ) using a Bruker AVANCE III HD 800 MHz spectrometer with a cryogenically-cooled TCI probe . All NMR samples were approximately 100 μl in a 3 mm Shigemi tube . NMR data were processed using NMRpipe ( Delaglio et al . , 1995 ) and analyzed using NMRViewJ ( Johnson , 2004 ) . Parameters derived from relaxation data were obtained with a non-linear least-squares fitting programmed in the Python SciPy library ( Jones et al . , 2001; More , 1977 ) . Errors in these parameters were generated using a nonparametric Monte Carlo bootstrap approach ( Efron , 1981 ) with 1000 simulated datasets , each of which consisted of synthetic datapoints with generated values centered on the measured peak intensities ( and errors determined from the noise of the experimental spectra ) . Each simulated dataset was individually fit; parameters from these fits were averaged to generate the reported values ( as average ± standard deviation ) . Unassigned isoleucine methyl peaks in A2AR are referred to by a peak ID ( Figure 3A ) , sorted by increasing difference between inverse agonist- and agonist-bound relaxation rates ( Figure 6B ) . 1H/13C-HMQC spectra of the methyl region were collected with a 13C spectral width of 22 ppm ( 8 ppm ) centered at 14 ppm ( 9 . 5 ppm ) for A2AR ( MBP ) , with 32 complex pairs collected . Due to short sample lifetimes , some A2AR HMQC spectra were collected as several sequential experiments with 64 scans per t1 point; if peak intensities and locations in these spectra remained consistent ( for example , see Figure 2—figure supplement 3 ) , they were summed after processing . Solvent PRE experiments were carried out by collecting an HMQC spectrum on an A2AR sample , then adding Gd3+ complexed with diethylenetriaminepentaacetic acid ( DTPA; Sigma-Aldrich ) to a final concentration of 1 mM and collecting a matched HMQC spectrum . The PRE effect was measured as the ratio of peak intensities in the paramagnetic sample to those in the diamagnetic sample . Triple quantum ( 3Q ) relaxation experiments were conducted using a pulse sequence kindly provided by Prof . Lewis Kay ( University of Toronto ) ( Sun et al . , 2011 ) . The carrier was centered on the methyl region ( 0 . 8 ppm ) in the 1H dimension , while 13C dimension center and spectral width were as above for HMQC spectra . The 3Q relaxation experiments were run as pseudo-4D experiments with the relaxation delays and alternating forbidden/allowed experiments as the third and fourth dimensions . An NMRpipe ( Delaglio et al . , 1995 ) script was used to divide the data and process each spectrum . For MBP , spectra were collected with relaxation delays of 0 . 8 , 2 , 4 , 8 , and 16 ms for both forbidden and allowed experiments . Peaks ( corresponding to I317 and I333 ) with low intensities ( defined as less than ten-fold greater than noise in the 16 ms forbidden spectrum ) were not used for relaxation analysis . Peak intensities were measured using the NMRViewJ ( Johnson , 2004 ) Rate Analysis module by fitting each peak to an ellipse and calculating the volume . Ratios of peak intensities were fit to the following equation as a function of relaxation delay ( Sun et al . , 2011 ) : ( 1 ) IforbIall=3Nall4Nforbηtanhη2+δ2Tη2+δ2-δtanhη2+δ2T where N is the number of scans for each experiment , T is the relaxation delay , δ ( <0 ) is a term for the coupling between rapidly and slowly decaying single-quantum coherences , and η is a relaxation rate , defined as the difference between slow and fast relaxation rates of single-quantum transitions for methyl protons: ( 2 ) η=R2 , HF-R2 , HS2∝τcSaxis2 proportional to the methyl axis order parameter S2axis and correlation time τc . All relaxation rates were constrained to be >0 while δ was constrained to be <0 . Due to short lifetimes of A2AR samples ( ~12–14 hr at 30°C ) , the relaxation experiment was modified and run with five allowed experiments ( 0 . 8 , 2 , 4 , 8 , and 14 ms ) and fewer forbidden experiments . To extract values of η , the forbidden:allowed peak intensity ratios at those few relaxation delays were fit to Equation 1 as above . Simultaneously , the allowed experiment peak intensities were fit to: ( 3 ) Iall=A32exp-R2 , HST+exp-R2 , HFT where A is a scaling constant , T is the relaxation delay , and RS and RF are as above ( Figure 5—figure supplement 1 ) . To determine which relaxation delays for the limited forbidden experiments gave values of η that best agreed with those determined from a full dataset , the MBP relaxation dataset was analyzed using all five allowed experiments and different combinations of 1 or two forbidden experiments . The differences in η values for each peak , as well as the sum of squared differences across all peaks , were compared for each analysis and a single relaxation delay of 8 ms was selected for A2AR relaxation experiments ( Figure 5 , Figure 5—figure supplement 1 ) . For processing , two separate scripts were used to extract the five allowed experiments and sum up the five forbidden experiments . Allowed peak intensity and forbidden:allowed peak intensity ratios were calculated and fit to Equations 1 and 2 , respectively , as above , to derive values for η at each peak . Ligand binding experiments on membranes containing A2A receptor were carried out based on previously published protocols ( Xu et al . , 2011 ) . Pichia cells expressing each construct were used to generate membranes as follows . Cells were resuspended in lysis buffer ( PBS containing 10% glycerol , 2 mM EDTA , and protease inhibitors ) and incubated for 2 hr at 37°C with Zymolyase 20T ( AMS Bio , Abingdon UK ) at a final concentration of 50 U/mL . Crude membranes were isolated through centrifugation at 40 , 000 rcf for 30 min and dounced in storage buffer ( 150 mM NaCl , 50 mM HEPES pH 7 . 5 , 10% glycerol , protease inhibitors ) . Large cell debris was removed by a low speed spin at 1000 rcf for 10 min , and remaining membranes were subjected to a high speed spin at 140 , 000 rcf for 30 min . Pellets were dounced in a minimal volume of storage buffer , flash frozen in LN2 , and stored at −80°C until needed . Saturation binding was carried out by incubating 1 . 5–2 μg of membranes with different concentrations of [3H]-ZM241385 ( 50 Ci/mmol , American Radiolabeled Chemicals , Inc . , St . Louis MO ) between 0 . 019 and 10 nM in assay buffer ( 50 mM Tris pH 7 . 4 , 10 mM MgCl2 , 1 mM EDTA , with 150 mM NaCl or 150 mM KCl as needed ) containing 0 . 1% protease-free BSA in a final volume of 250 μL per tube . Reactions were incubated at room temperature for 1 hr . Non-specific binding was determined using reactions containing 10 mM theophylline . Reactions were separated on a vacuum manifold using GF/C filters ( pre-soaked in assay +0 . 5% PEI ) to retain membranes and discard unbound ligand . After washing four times with cold assay buffer , bound radioactivity was quantified using a scintillation counter . For competition binding experiments , aliquots of membranes were incubated with 0 . 5 nM [3H]-ZM241385 , and varying concentrations of cold NECA , from 0 . 5 nM to 300 μM , were included in the binding reactions . All binding experiments were carried out as three independent experiments , each performed in duplicate . Data analysis and fitting was performed with GraphPad Prism ( GraphPad Software Inc . ) . 9 μg of Gs trimer was added to a tube and preincubated on ice for 15 min with 75 μL of 1650 μM SAPE ( 1-stearoyl-2-arachidonoyl-sn-glycero-3-phosphoethanolamine; Avanti Lipids ) , 980 μM porcine brain phosphatidylserine ( Avanti Lipids ) , 180 μM cholesteryl hemisuccinate ( Steraloids ) in 20 mM Hepes pH 8 . 0 , 100 mM NaCl , 0 . 4% deoxycholate , 0 . 04% sodium cholate . 35 pmol of A2A receptor and HMEN buffer ( 20 mM HEPES pH 8 . 0 , 100 mM NaCl , 3 mM MgCl2 , 1 mM EDTA ) was added to the tube for a total volume of 150 μL . The sample was incubated an additional 5 min and applied to an Ultrogel AcA34 column ( Sigma-Aldrich ) equilibrated in HMEN buffer . BSA was added to fractions at a concentration of 0 . 1 mg/mL and vesicles were flash-frozen for storage at −80°C . Receptor recovery was monitored by total ZM241385 binding and ~1 mL after void volume was used in GTPγS binding assays . Recovery of receptor in this pool was ~13% . 5 μL of vesicles was assayed in a total of 50 μL with 20 mM HEPES pH 8 . 0 , 100 mM NaCl , 2 mM MgCl2 , 1 mM EDTA , 1 mM DTT , 0 . 1 mg/mL BSA , 100 nM GTPγS with 35S GTPγS as a tracer in either no ligand , 10 μM NECA agonist , or 100 nM ZM241385 antagonist . Samples were incubated at 30°C for 10 min and stopped with 50 μL of quench buffer ( 20 mM Tris pH 8 , 100 mM NaCl , 10 mM MgCl2 , 0 . 1 mM DTT , 1 mM GTP , and 0 . 1% Lubrol ) and incubated on ice for 10 min . Samples were filtered over BA85 nitrocellulose and filters counted on a liquid scintillation counter after washing 4x with 20 mM Tris pH 8 . 0 , 100 mM NaCl , 10 mM MgCl2 . | Almost every aspect of the human body – from our senses to our moods – depends , in one way or another , on a large family of proteins called G-protein-coupled receptors . These receptor proteins , known as GPCRs for short , detect signals from outside the cell and trigger activity within the cell . This allows cells to gather information from their surroundings and to communicate with each other . Importantly , since GPCRs regulate many processes in the body that are involved in disease , it is perhaps unsurprising that over a third of all approved drugs target these receptors . Like all proteins , GPCRs are long chain-like molecules with a repetitive backbone and short branches called sidechains . Each sidechain has its own chemical properties and electrical charge , which can affect how different parts of the chain interact with each other and what shape the protein can adopt . This in turn can influence how strongly a drug or other molecule can bind to a receptor protein . Protein crystallography is one technique that has been used to better understand how the different GPCRs are built and how they work . The technique involves growing crystals from pure samples of the protein; this locks millions of copies of the protein in place and provides a snapshot of its shape . However , GPCRs – and especially their sidechains – are flexible and can adopt different shapes , which cannot be seen fully by only looking at protein crystals . Now , Clark , Dikiy et al . used another technique called nuclear magnetic resonance spectroscopy , or NMR for short , to understand how drugs affect the fast moving sidechains within a GPCR . First , genetically modified yeast was used to create samples of a GPCR called the adenosine receptor A2A that were labelled with specific markers which made it easier to measure the structure and flexibility of the protein by NMR . This approach revealed that too much sodium in the sample’s solution supresses the large structural changes that occur in the A2A receptor when it binds to a drug . Moreover , it showed that the sidechains of several regions on the receptor move in different ways depending on whether the receptor binds to an activating drug or an inhibiting drug . These findings lay the groundwork for understanding how the movements of sidechains help to activate or inhibit GPCRs , and will complement on-going studies using protein crystals . Moreover , the new approach to producing labelled proteins could be applied to other types of proteins that until now could not be studied with NMR due to practical limitations . In future , this may help scientists to better understand how drugs affect these proteins and to develop new treatments for a whole range of diseases . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"biochemistry",
"and",
"chemical",
"biology",
"structural",
"biology",
"and",
"molecular",
"biophysics"
] | 2017 | Ligand modulation of sidechain dynamics in a wild-type human GPCR |
Influenza is notable for its evolutionary capacity to escape immunity targeting the viral hemagglutinin . We used deep mutational scanning to examine the extent to which a high inherent mutational tolerance contributes to this antigenic evolvability . We created mutant viruses that incorporate most of the ≈104 amino-acid mutations to hemagglutinin from A/WSN/1933 ( H1N1 ) influenza . After passaging these viruses in tissue culture to select for functional variants , we used deep sequencing to quantify mutation frequencies before and after selection . These data enable us to infer the preference for each amino acid at each site in hemagglutinin . These inferences are consistent with existing knowledge about the protein's structure and function , and can be used to create a model that describes hemagglutinin's evolution far better than existing phylogenetic models . We show that hemagglutinin has a high inherent tolerance for mutations at antigenic sites , suggesting that this is one factor contributing to influenza's antigenic evolution .
Epidemic influenza poses an annual threat to human health largely because the virus rapidly evolves to escape the immunity elicited by previous infections or vaccinations . The most potent form of anti-influenza immunity is antibodies targeting the virus’s hemagglutinin ( HA ) protein ( Yewdell et al . , 1979; Wiley et al . , 1981; Caton et al . , 1982 ) . The virus evades these antibodies primarily by accumulating amino-acid substitutions in HA's antigenic sites ( Smith et al . , 2004; Das et al . , 2013; Koel et al . , 2013; Bedford et al . , 2014 ) . Remarkably , HA undergoes this rapid evolution while retaining the ability to fold to a highly conserved structure that performs two functions essential for viral replication: receptor binding and membrane fusion ( Wiley and Skehel , 1987; Russell et al . , 2004 ) . HA is therefore highly ‘antigenically evolvable’ in the sense that it can accommodate rapid antigenic change without compromising its structural and functional properties . Two factors that undoubtedly contribute to HA's rapid antigenic evolution are influenza's high mutation rate and the strong selection that immunity exerts on the virus . However , it is unclear whether these factors are sufficient to fully explain HA's antigenic evolution . For instance , while some other error-prone viruses ( such as HIV and hepatitis C ) also exhibit rapid antigenic evolution of their surface proteins ( Burton et al . , 2012 ) , other viruses with comparable mutation rates ( such as measles ) show little propensity for antigenic change ( Sheshberadaran et al . , 1983; Duffy et al . , 2008 ) , despite the fact that evasion of immunity would presumably confer a selective benefit . A variety of explanations ranging in scale from ecological to molecular can be posited to account for these differences in rates of antigenic evolution ( Lipsitch and O’Hagan , 2007; Koelle et al . , 2006; Heaton et al . , 2013 ) . One hypothesis is that HA has a high inherent tolerance for mutations in its antigenic sites , thereby conferring on influenza the evolutionary capacity to escape from anti-HA antibodies with relative ease . Testing this hypothesis requires quantifying the inherent mutational tolerance of each site in HA . This cannot be done simply by examining variability among naturally occurring viruses , since the evolution of influenza in nature is shaped by a combination of inherent mutational tolerance and external immune selection . For example , the rapid evolution of HA's antigenic sites in nature could reflect the fact that these sites are especially tolerant of mutations , or it could be purely a consequence of strong immune selection . Traditional experimental approaches using site-directed mutagenesis or serial viral passage are also inadequate to quantify inherent mutational tolerance—while such experimental techniques have been used to determine the effect of specific mutations on HA , they cannot feasibly be applied to all possible individual amino-acid mutations . Recently Heaton et al . ( 2013 ) used transposon mutagenesis to show that HA is tolerant to the random insertion of five to six amino-acid sequences at several locations in the protein . However , the relevance of this tolerance to insertional mutations is unclear , since HA's actual antigenic evolution involves almost entirely point substitutions , with only a very low rate of insertions and deletions . Here we use the new high-throughput experimental technique of deep mutational scanning ( Fowler et al . , 2010; Araya and Fowler , 2011 ) to comprehensively quantify the tolerance of HA to amino-acid mutations . Specifically , we create mutant libraries of the HA gene from the H1N1 strain A/WSN/1933 ( WSN ) that contain virtually all of the ≈4 × 104 possible individual codon mutations , and therefore virtually all of the ≈104 possible amino-acid mutations . We use these mutant libraries to generate pools of mutant influenza viruses , which we estimate incorporate at least 85% of the possible HA codon mutations and 97% of the possible amino-acid mutations . We then passage these viruses to select for functional variants , and use Illumina deep sequencing to determine the frequency of each HA mutation before and after this selection for viral growth . Since these experiments measure the impact of mutations in the absence of immune selection , they enable us to quantify HA's inherent preference for each amino acid at each site in the protein . We show that these quantitative measurements are consistent with existing knowledge about HA structure and function , and can be used to create an evolutionary model that describes HA's natural evolution far better than existing models of sequence evolution . Finally , we use our results to show that HA's antigenic sites are disproportionately tolerant of mutations , suggesting that a high inherent tolerance for mutations at key positions targeted by the immune system is one factor that contributes to influenza's antigenic evolvability .
Our strategy for deep mutational scanning ( Fowler et al . , 2010; Araya and Fowler , 2011 ) of HA is outlined in Figure 1 . The wildtype WSN HA gene was mutagenized to create a diverse library of mutant HA genes . This library of mutant genes was then used to generate a pool of mutant viruses by reverse genetics ( Hoffmann et al . , 2000 ) . The mutant viruses were passaged at a low multiplicity of infection to ensure a linkage between genotype and phenotype . The frequencies of mutations before and after selection for viral growth were quantified by Illumina deep sequencing of the mutant genes ( the mutDNA sample in Figure 1 ) and the mutant viruses ( the mutvirus sample in Figure 1 ) . An identical process was performed in parallel using the unmutated wildtype HA gene to generate unmutated viruses in order to quantify the error rates associated with sequencing , reverse transcription , and virus growth ( these are the DNA and virus samples in Figure 1 ) . The entire process in Figure 1 was performed in full biological triplicate ( the replicates are referred to as #1 , #2 , and #3 ) . In addition , a repeat of the Illumina sample preparation and deep sequencing was performed for replicate #1 to quantify the technical variation associated with these processes . 10 . 7554/eLife . 03300 . 003Figure 1 . Schematic of the deep mutational scanning experiment . The Illumina deep-sequencing samples are shown in yellow boxes ( DNA , mutDNA , virus , mutvirus ) . Experimental steps and associated sources of mutations are shown in blue text , while sources of error during Illumina sample preparation and sequencing are shown in red text . This entire process was performed in biological triplicate . DOI: http://dx . doi . org/10 . 7554/eLife . 03300 . 003 The deep mutational scanning strategy in Figure 1 requires creating mutant libraries of the HA gene . We wanted to assess the impact of all possible amino-acid mutations . Most mutagenesis techniques operate at the nucleotide level , and so frequently introduce single-nucleotide codon changes ( e . g . , GGA → cGA ) but only very rarely introduce multi-nucleotide codon changes ( e . g . , GGA → cat ) . However , several PCR-based techniques have recently been developed to introduce random codon mutations into full-length genes ( Firnberg and Ostermeier , 2012; Bloom , 2014; Jain and Varadarajan , 2014 ) . We used one of these techniques ( Bloom , 2014 ) to create three replicate codon-mutant libraries of the WSN HA gene ( Supplementary file 1 ) . Sanger sequencing of 34 individual clones indicated that the libraries contained an average of slightly over two codon mutations per gene , with a very low rate of insertions and deletions ( less than 0 . 1 per gene ) . The number of mutations per clone was distributed around this average in an approximately Poisson fashion ( Figure 2 ) . The mutations consisted of a mix of one- , two- , and three-nucleotide codon changes , and were roughly uniform in their nucleotide composition and location in the gene ( Figure 2 ) . 10 . 7554/eLife . 03300 . 004Figure 2 . Properties of the HA codon-mutant library as assessed by Sanger sequencing of 34 individual clones drawn roughly evenly from the three experimental replicates . ( A ) There are an average of 2 . 1 codon mutations per clone , with the number per clone following a roughly Poisson distribution . ( B ) The codon mutations involve a mix of one- , two- , and three-nucleotide mutations . ( C ) The nucleotide composition of the mutant codons is roughly uniform . ( D ) The mutations are distributed uniformly along HA's primary sequence . ( E ) There is no tendency for mutations to cluster in primary sequence . Shown is distribution of observed pairwise distances between mutations in multiply mutated clones vs the expected distribution when the mutations are placed independently in the clones . All plots show results only for substitution mutations; insertion/deletion mutations are not shown . However , only two insertion/deletion mutations ( 0 . 06 per clone ) were identified . The data and computer code used to generate this figure are at https://github . com/jbloom/SangerMutantLibraryAnalysis/tree/v0 . 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 03300 . 004 The genes in each mutant library were cloned at high efficiency into a bidirectional influenza reverse-genetics plasmid ( Hoffmann et al . , 2000 ) . Each of the library replicates contained at least six-million unique clones—a diversity that far exceeds the 104 unique single amino-acid mutations and the ≈4 × 104 unique single codon mutations to the HA gene . The vast majority of possible codon and amino-acid mutations are therefore represented many times in each plasmid mutant library , both individually and in combination with other mutations . The HA plasmid mutant libraries were used to generate pools of mutant influenza viruses by reverse genetics ( Hoffmann et al . , 2000 ) . Briefly , this process involves transfecting cells with the HA plasmid mutant library along with plasmids encoding the other seven genes from the WSN strain of influenza . Although the cells were transfected with a very large diversity of HA plasmids , we were uncertain what fraction of the genes encoded on these plasmids would actually be productively packaged into a virus . In an attempt to maximize the diversity in the viral pools , the mutant viruses were generated by transfecting several dozen wells of cells . The logic behind this scheme was to maintain substantial diversity even if only a subset of viral mutants stochastically predominated in each individual well of cells . A different replicate virus pool was generated for each of the three HA plasmid mutant libraries . The mutant viruses generated for each replicate were passaged at a relatively low multiplicity of infection ( MOI ) of 0 . 1 to reduce the probability of co-infection , thereby creating a link between viral genotype and phenotype . This genotype-phenotype link is essential to ensure that the sequenced HA gene matches the protein on the surface of the virus . Our previous work with NP ( Bloom , 2014 ) indicates that one low MOI passage is sufficient to create a strong genotype-phenotype link , since that previous work found that the results obtained after one viral passage were extremely similar to the results obtained after two viral passages . In order to maintain a diversity of over two-million infectious viral particles , we performed the passaging in a total of 2 . 4 × 107 cells . We used Illumina sequencing to quantify the frequencies of mutations before and after selection for viral growth . For each replicate , we sequenced HA from the unmutated plasmid , the plasmid mutant library , virus produced from the unmutated plasmid , and mutant virus produced from the plasmid mutant library—these are the DNA , mutDNA , virus , and mutvirus samples in Figure 1 . For the DNA and mutDNA samples , the HA gene was amplified directly from the plasmids by PCR . For the virus and mutvirus samples , the HA gene was first reverse-transcribed from viral RNA and was then amplified by PCR . In all cases , template quantification was performed prior to PCR to ensure that >106 initial HA molecules were used as templates for subsequent amplification . In order to reduce the sequencing error rate , the HA molecules were fragmented to roughly 50 nucleotide fragments using Illumina's transposon-based Nextera kit , and then sequenced with overlapping paired-end reads ( Figure 3—figure supplement 1 ) . We only called codon identities for which both paired reads concur—this strategy substantially increases the sequencing fidelity , since it is rare for the same sequencing error to occur in both reads . For each sample , we obtained in excess of 107 overlapping paired-end reads that could be aligned to HA ( Figure 3—figure supplement 2 ) . As shown in Figure 3—figure supplement 3 , the read depth varied somewhat along the primary sequence , presumably due to known weak biases in the insertion sites for the Nextera transposon ( Adey et al . , 2010 ) . However , these biases were fairly mild , and so we obtained well over 2 × 105 unique paired reads for nearly all HA codons . Figure 3 shows the frequency of mutations in each sample as quantified by deep sequencing . The DNA samples derived from unmutated HA plasmid show a low frequency of apparent mutations which are almost exclusively composed of single-nucleotide codon changes—the frequency of these apparent mutations reflects the rate of errors from the PCR amplification and subsequent deep sequencing . The virus samples created from the unmutated plasmid show only a slightly higher frequency of mutations , indicating that reverse-transcription and viral replication introduce only a small number of additional mutations . As expected , the mutDNA samples derived from the plasmid mutant libraries show a high rate of one- , two- , and three-nucleotide mutations , as all three types of mutations were introduced during the codon mutagenesis . The mutvirus samples derived from the mutant virus pools exhibit a mutation rate that is substantially lower than that of the mutDNA samples . Most of the reduction in mutation frequency in the mutvirus samples is due to decreased frequencies of nonsynonymous and stop-codon mutations; synonymous mutations are only slightly depressed in frequency . As stop-codon and nonsynonymous mutations are much more likely than synonymous mutations to substantially impair viral fitness , these results are consistent with purifying selection purging deleterious mutations during viral growth . 10 . 7554/eLife . 03300 . 005Figure 3 . The per-codon frequencies of mutations in the samples . The samples are named as in Figure 1 , with the experimental replicate indicated with the numeric label . The DNA samples have a low frequency of mutations , and these mutations are composed almost entirely of single-nucleotide codon changes—these samples quantify the baseline error rate from PCR and deep sequencing . The mutation frequency is only slightly elevated in virus samples , indicating that viral replication and reverse transcription introduce only a small number of additional mutations . The mutDNA samples have a high frequency of single- and multi-nucleotide codon mutations , as expected from the codon mutagenesis procedure . The mutvirus samples have a lower mutation frequency , with most of the reduction due to fewer stop-codon and nonsynonymous mutations—consistent with purifying selection purging deleterious mutations . The data and code used to create this plot is available via http://jbloom . github . io/mapmuts/example_WSN_HA_2014Analysis . html; this plot is the file parsesummary_codon_types_and_nmuts . pdf described therein . The sequencing accuracy was increased by using overlapping paired-end reads as illustrated in Figure 3—figure supplement 1 . The overall number of overlapping paired-end reads for each sample is shown in Figure 3—figure supplement 2 . A representative plot of the read depth across the primary sequence is shown in Figure 3—figure supplement 3 . DOI: http://dx . doi . org/10 . 7554/eLife . 03300 . 00510 . 7554/eLife . 03300 . 006Figure 3—figure supplement 1 . The overlapping paired-end Illumina sequencing strategy . ( A ) Sequencing accuracy was increased by fragmenting the HA gene to pieces roughly 50 nucleotides in length , and then using overlapping paired-end 50 nucleotide Illumina sequencing reads . Codon identities were only called if the reads overlapped and concurred on the codon identity . ( B ) The distribution of actual HA fragment lengths for a representative sample . The plot in ( B ) is the file replicate_3/DNA/replicate_3_DNA_insertlengths . pdf described at <monospace>http://jbloom . github . io/mapmuts/example_WSN_HA_2014Analysis . html . DOI: http://dx . doi . org/10 . 7554/eLife . 03300 . 00610 . 7554/eLife . 03300 . 007Figure 3—figure supplement 2 . The total number of reads for each sample . For all samples , the majority of reads could be paired and aligned to the HA sequence . However , the exact fraction of reads that could be paired varied somewhat among samples due to variation in the efficiency with which the HA gene was fragmented to the target length of 50 nucleotides . This plot is the file alignmentsummaryplot . pdf described at http://jbloom . github . io/mapmuts/example_WSN_HA_2014Analysis . html . DOI: http://dx . doi . org/10 . 7554/eLife . 03300 . 00710 . 7554/eLife . 03300 . 008Figure 3—figure supplement 3 . The per-codon read depth as a function of primary sequence . This plot is typical of the samples . The read depth varied fairly consistently as a function of primary sequence , presumably due to biases in the positions at which the HA gene tended to fragment . This plot is the file replicate_3/DNA/replicate_3_DNA_codondepth . pdf described at http://jbloom . github . io/mapmuts/example_WSN_HA_2014Analysis . html . DOI: http://dx . doi . org/10 . 7554/eLife . 03300 . 008 Inspection of Figure 3 also demonstrates an important advantage of introducing the mutations at the codon rather than the nucleotide level . While there is a low but non-zero rate of errors ( from sequencing , PCR , or reverse-transcription ) that lead to single-nucleotide codon changes ( as judged by the DNA and virus samples ) , errors that lead to multi-nucleotide codon changes are negligible because it is extremely rare for a single codon to experience two errors . We similarly expect that any do novo mutations or reversions that arise during viral growth should be limited to single-nucleotide changes given the short duration of viral passage in our experiments . The fact that our mutant libraries were constructed at the codon rather than the nucleotide level means that the vast majority ( 54 of 63 ) possible mutations to each codon involve multiple nucleotide changes , and so the sequencing results for these mutations can be analyzed essentially at face value , without having to worry about confounding errors . For the remaining ( 9 of 63 ) possible mutations that only involve a single-nucleotide codon change , we have attempted to statistically correct for the error rates estimated from our controls as described in the ‘Materials and methods’ . It is important to assess the completeness with which the experiments sampled all possible HA mutations . Several problems could limit mutational sampling: mutations might be absent from the plasmid mutant libraries due to biases in the codon mutagenesis , mutations that are present in the plasmid mutants might fail to be incorporated into viruses due to stochastic bottlenecks during virus generation by reverse genetics , or the sequencing read depth might be inadequate to sample the mutations that are present . The most straightforward way to assess these issues is to quantify the number of times that each possible multi-nucleotide codon mutation is observed in the mutDNA and mutvirus samples . Restricting the analysis to multi-nucleotide codon mutations avoids the confounding effects of sequencing and reverse-transcription errors , which cause almost exclusively single-nucleotide changes . Figure 4 shows the number of times that each mutation was observed in the combined sequencing data for the three biological replicates; Figure 4—figure supplement 1 shows the same data for the replicates individually . More than 99 . 5% of multi-nucleotide codon mutations are observed at least five times in the combined sequencing data from the plasmid mutant libraries ( mutDNA samples ) , and ≈ 97 . 5% of all such mutations are observed at least five times in sequencing of the mutDNA for each individual replicate . These results indicate that the vast majority of codon mutations are represented in the plasmid mutant libraries . 10 . 7554/eLife . 03300 . 009Figure 4 . The number of times that each possible multi-nucleotide codon mutation was observed in each sample after combining the data for the three biological replicates . Nearly all mutations were observed many times in the mutDNA samples , indicating that the codon mutagenesis was comprehensive . Only about half of the mutations were observed at least five times in the mutvirus samples , indicating either a bottleneck during virus generation or purifying selection against many of the mutations . If the analysis is restricted to synonymous multi-nucleotide codon mutations , then about 85% of mutations are observed at least five times in the mutvirus samples . Since synonymous mutations are less likely to be eliminated by purifying selection , this latter number provides a lower bound on the fraction of codon mutations that were sampled by the mutant viruses . The redundancy of the genetic code means that the fraction of amino-acid mutations sampled is higher . The data and code used to create this figure are available via http://jbloom . github . io/mapmuts/example_WSN_HA_2014Analysis . html; this plot is the file countparsedmuts_multi-nt-codonmutcounts . pdf described therein . Similar plots for the individual replicates are shown in Figure 4—figure supplement 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 03300 . 00910 . 7554/eLife . 03300 . 010Figure 4—figure supplement 1 . Plots like those in Figure 4 for the individual biological replicates . ( A ) replicate 1 , ( B ) replicate 2 , and ( C ) replicate 3 . These plots are the files replicate_1/countparsedmuts_multi-nt-codonmutcounts . pdf , replicate_2/countparsedmuts_multi-nt-codonmutcounts . pdf , and replicate_3/countparsedmuts_multi-nt-codonmutcounts . pdf described at http://jbloom . github . io/mapmuts/example_WSN_HA_2014Analysis . html . DOI: http://dx . doi . org/10 . 7554/eLife . 03300 . 010 In contrast , only 53% of multi-nucleotide codon mutations are observed at least five times in the combined sequencing data for the mutant viruses ( mutvirus samples ) , and only ≈ 26% of such mutations are observed at least five times in sequencing of the mutvirus for each individual replicate ( Figure 4 , Figure 4—figure supplement 1 ) . However , these numbers are confounded by the fact that many mutations are deleterious , and so may be absent because purifying selection has purged them from the mutant viruses . A less confounded measure is the frequency of synonymous multi-nucleotide mutations , since synonymous mutations are less likely to be strongly deleterious . About 85% of such mutations are observed at least five times in the combined mutvirus samples , and ≈ 51% of such mutations are observed at least five times in the mutvirus samples for the individual replicates ( Figure 4 , Figure 4—figure supplement 1 ) . Note that these numbers are only a lower bound on the fraction of codon mutations sampled by the mutant viruses—even synonymous mutations to influenza are sometimes strongly deleterious ( Marsh et al . , 2008 ) , and so some of the missing synonymous codon mutations may have been introduced into mutant viruses but then purged by purifying selection . Furthermore , the redundancy of the genetic code means that the fraction of possible amino-acid mutations sampled is substantially higher than the fraction of codon mutations sampled . Specifically , if 85% of possible codon mutations are sampled at least five times in the combined libraries ( as Figure 4 indicates ) , then our simulations suggest that ≈ 97% of possible amino-acid mutations will have also been sampled at least five times ( ‘Materials and methods’ ) . Overall , these results indicate that nearly all mutations are represented in the plasmid mutant libraries . Virus generation by reverse genetics does introduce a bottleneck—but fortunately , this bottleneck is sufficiently mild that at least half of all possible codon mutations are still sampled at least five times by the mutant viruses in each individual replicate . Combining the data for the three replicates brings the coverage of possible codon mutations to around 85% , and the coverage of possible amino-acid mutations to 97% . Therefore , the sampling of mutations is sufficiently complete to provide information on the effects of most amino-acid mutations when the data from the three experimental replicates are combined . We quantified the effects of mutations in terms of site-specific amino-acid ‘preferences’ . These preferences are the expected frequency of each amino acid at each site in the mutant viruses in a hypothetical situation in which all amino acids are introduced at that site at equal frequency in the initial plasmid mutant library ( Bloom , 2014 ) . Because many of the HAs in our libraries contain several mutations , these preferences do not simply correspond to the fitness effect of each individual mutation to the WSN HA—rather , they represent the average effect of each mutation in a collection of closely related HA mutants . Mutations to amino acids with high preferences are favored by selection , while mutations to amino acids with low preferences are disfavored . The amino-acid preferences are inferred from the deep sequencing data using a Bayesian statistical framework in which the observed counts are treated as draws from multinomial distributions with unknown parameters representing the initial mutagenesis rate , the various error rates , and selection as represented by the preferences ( see ‘Materials and methods’ for details ) . Figure 5 shows the amino-acid preferences for the entire HA gene inferred from the combined data from all three biological replicates . As can be seen from this figure , some sites have strong preferences for one specific amino acid , while other sites are tolerant of a variety of different amino acids . As described in Table 1 , the inferred amino-acid preferences are consistent with existing knowledge about mutations and residues affecting HA stability , membrane fusion , proteolytic activation , and receptor binding ( Nakajima et al . , 1986; Martin et al . , 1998; Qiao et al . , 1999; Stech et al . , 2005 ) . This concordance suggests that the deep mutational scanning effectively captures many of the structural and functional constraints on HA . 10 . 7554/eLife . 03300 . 011Figure 5 . The amino-acid preferences inferred using the combined data from the three biological replicates . The letters have heights proportional to the preference for that amino acid , and are colored by hydrophobicity . The first overlay bar shows the relative solvent accessibility ( RSA ) for residues in the HA crystal structure . The second overlay bar indicates Caton et al . antigenic sites or conserved receptor-binding residues . The sequence is numbered sequentially beginning with 1 at the N-terminal methionine—however , this first methionine is not shown as it was not mutagenized . Figure 5—figure supplement 1 shows the same data with H3 numbering of the sequence . The data and code used to create this figure are available via http://jbloom . github . io/mapmuts/example_WSN_HA_2014Analysis . html; this plot is the file sequentialnumbering_site_preferences_logoplot . pdf described therein . DOI: http://dx . doi . org/10 . 7554/eLife . 03300 . 01110 . 7554/eLife . 03300 . 012Figure 5—figure supplement 1 . A plot matching that shown in Figure 5 except that the HA sequence is numbered using the H3 numbering scheme . This plot is the file H3numbering_site_preferences_logoplot . pdf described at http://jbloom . github . io/mapmuts/example_WSN_HA_2014Analysis . html . DOI: http://dx . doi . org/10 . 7554/eLife . 03300 . 01210 . 7554/eLife . 03300 . 013Table 1 . The amino-acid preferences inferred from the combined experimental replicates are consistent with existing knowledge about HA structure and functionDOI: http://dx . doi . org/10 . 7554/eLife . 03300 . 013Site in sequential numberingSite in H3 numberingExisting knowledgeInferred amino-acid preferences127117 ( HA1 ) Mutation from S to P creates a temperature-sensitive defect ( Nakajima et al . , 1986 ) The preference for S is 30 times higher than the preference for P174161 ( HA1 ) Mutation from Y to H creates a temperature-sensitive defect ( Nakajima et al . , 1986 ) The preference for Y is 25 times higher than the preference for H3441 ( HA2 ) Mutation from G to E abolishes HA fusion activity ( Qiao et al . , 1999 ) The preference for G is 11 times higher than for E343327 ( HA1 ) A basic residue ( R or K ) is required for HA proteolytic activation ( Stech et al . , 2005 ) The combined preferences for R and K ( 0 . 87 ) far exceed those of all other amino acids combined10898 ( HA1 ) Receptor-binding residue , is Y in >99% of natural H1 HAsThe preference for Y ( 0 . 61 ) exceeds those of all other amino acids combined166153 ( HA1 ) Receptor-binding residue , is W in >99% of natural H1 HAsThe preference for W ( 0 . 65 ) exceeds those of all other amino acids combined196183 ( HA1 ) Receptor-binding residue , is H in >99% of natural H1 HAsThe preference for H ( 0 . 69 ) exceeds those of all other amino acids combined203190 ( HA1 ) Receptor-binding residue , is D in 90% of natural H1 HAsThe highest preference is for the chemically similar E207194 ( HA1 ) Receptor-binding residue , is L in 97% of natural H1 HAsThe preference for L ( 0 . 55 ) exceeds those of all other amino acids combined208195 ( HA1 ) Receptor-binding residue , is Y in >99% of natural H1 HAsThe preference for Y ( 0 . 72 ) exceeds those of all other amino acids combined239226 ( HA1 ) Receptor-binding residue , is Q in ≈99% of natural H1 HAsQ is one of three amino acids with a high preference241228 ( HA1 ) Receptor-binding residue , is G in >99% of natural H1 HAsThe preference for G ( 0 . 57 ) exceeds those of all other amino acids combinedThe conserved receptor-binding residues listed in this table are those delineated in the first table of Martin et al . ( 1998 ) that also have at least 90% conservation among all naturally occurring H1 HAs in the Influenza Virus Resource ( Bao et al . , 2008 ) . Despite the general concordance between the inferred amino-acid preferences and existing knowledge , it is important to quantify the experimental error associated with the deep mutational scanning . We sought to quantify two factors: technical variation due to inaccuracies and statistical limitations during Illumina sample preparation and deep sequencing , and biological variation due to stochasticity in the viral mutants that were generated and enriched during each replicate of the experiment . Figure 6A shows the correlation between biological replicate #1 and a technical repeat of the Illumina sample preparation and deep sequencing for this biological replicate . There is a very high correlation between the preferences inferred from these two repeats , indicating that technical variation has only a very minor influence on the final inferred amino-acid preferences . Figure 6B–D show the correlation among the three different biological replicates . Although the biological replicates are substantially correlated , there is also clear variation . Most of this variation is attributable to amino acids which in one replicate are inferred to have preferences near the a priori expectation of 0 . 05 ( there are 20 amino acids , which in the absence of data are all initially assumed to have an equal preference of 120 ) , but in another replicate are inferred to have a much higher or lower preference . Such variation arises because the mutant viruses for each biological replicate only sample about 50% of the possible codon mutations ( see previous section ) , meaning that there is little data for some mutations in any given replicate . Fortunately , combining the three biological replicates greatly increases the coverage of possible mutations ( see previous section ) . Therefore , inferences made from the combined data ( as in Figure 5 ) should be substantially more accurate than inferences from any of the individual replicates . This idea is supported by the results below , which quantify the extent to which the inferred preferences accurately describe natural HA evolution . 10 . 7554/eLife . 03300 . 014Figure 6 . Correlations among the amino-acid preferences inferred using data from the individual biological replicates . ( A ) The preferences from two technical repeats of the sample preparation and deep sequencing of biological replicate #1 are highly correlated . ( B ) – ( D ) The preferences from the three biological replicates are substantially but imperfectly correlated . Overall , these results indicate that technical variation in sample preparation and sequencing is minimal , but that there is substantial variation between biological replicates due to stochastic differences in which mutant viruses predominate during the initial reverse-genetics step . The Pearson correlation coefficient ( R ) and associated p-value are shown in the upper-left corner of each plot . The data and code used to create this figure are available via http://jbloom . github . io/mapmuts/example_WSN_HA_2014Analysis . html; these plots are the files correlations/replicate_1_vs_replicate_1_repeat . pdf , correlations/replicate_1_vs_replicate_2 . pdf , correlations/replicate_1_vs_replicate_3 . pdf , and correlations/replicate_2_vs_replicate_3 . pdf described therein . DOI: http://dx . doi . org/10 . 7554/eLife . 03300 . 014 As our paper was under review , Wu et al . ( 2014 ) published the results of using a similar strategy to examine the effects of mutations to the WSN HA . In their study , the HA gene was mutated at the nucleotide level , so their experiments surveyed only amino-acid mutations accessible by single-nucleotide codon changes . As a result , they provide data on the effects of only about 20% of the 19 × 564 = 10716 amino-acid mutations examined in our study . Despite this limitation , their study provides a large dataset of mutational effects to which we can compare our results . Figure 7 compares the mutational effects determined in our study to those from Wu et al . ( 2014 ) . There is a highly significant correlation between the results of the two studies—but the inferred mutational effects are certainly not identical . Because Wu et al . ( 2014 ) do not provide the data for replicates of their experiment , we are unable to assess whether the variability between the two different studies exceeds the variability between experimental replicates within each study . So one can imagine both biologically interesting and uninteresting explanations for the imperfect correlation between the results of the two studies . The interesting explanation is that differences in experimental methodology could lead to different selection pressures on specific mutations: for instance , Wu et al . ( 2014 ) use A549 cells while we use MDCK-SIAT1 cells , and perhaps the impact of certain mutations is dependent on the cell line . The uninteresting explanation is that the imperfect correlation is simply due to noise in the experimental measurements . Unfortunately , it is not straightforward to distinguish between these two explanations . This difficulty in pinpointing reasons for inter-study variation highlights a limitation of the high-throughput experimental methodology employed by ourselves and Wu et al . ( 2014 ) : while such experiments provide a wealth of data , numerous factors can create noise in these data ( sequencing errors , population bottlenecks , epistasis among mutations , etc ) . Realizing the full potential of such studies will therefore require extensive experimental controls and biological replicates to quantify errors and noise to enable comparisons across data sets . 10 . 7554/eLife . 03300 . 015Figure 7 . Correlation of the site-specific amino-acid preferences determined in our study with the “relative fitness” ( RF ) values reported by Wu et al . ( 2014 ) . Wu et al . ( 2014 ) report RF values for 2350 of the 564×19 = 10716 possible amino-acid mutations to the WSN HA examined in our study ( they only examine single-nucleotide changes and disregard certain types of mutations due to oxidative damage of their DNA ) . To compare across the data sets , we have normalized their RF values by the RF value for the wildtype amino-acid ( which they provide for only 2264 of the 2350 mutations ) . We then correlate on a logarithmic scale these normalized RF values with the ratio of our measurement of the preference for the mutant amino acid divided by the preference for the wildtype amino acid , using the preferences from our combined replicates . For mutations for which Wu et al . ( 2014 ) report an RF of zero , we assign a normalized RF equal to the smallest value for their entire data set . There is a significant Pearson correlation of 0 . 48 between the data sets , indicating that both our experiments and those of Wu et al . ( 2014 ) are capturing many of the same constraints on HA . The data and code used to create this figure are available via http://jbloom . github . io/mapmuts/example_WSN_HA_2014Analysis . html; this plot is the file correlation_with_Wu_et_al . pdf described therein . DOI: http://dx . doi . org/10 . 7554/eLife . 03300 . 015 Nonetheless , Figure 7 shows that there is a highly significant correlation between the results of these two high-throughput studies , despite differences in experimental methodology and unquantified sources of experimental noise . This fact suggests that both studies capture fundamental constraints on HA’s mutational tolerance . In the remaining sections , we apply the more comprehensive data generated by our study to address questions about HA’s natural evolution and antigenic evolvability . Do the results of our deep mutational scanning experiment accurately reflect the real constraints on HA ? Table 1 uses an anecdotal comparison to a small number of existing experimental studies to suggest that they do . However , a more systematic way to address this question is to compare the inferred amino-acid preferences to the actual patterns of HA evolution in nature . To make such a comparison , we created an alignment of HA sequences from human and swine influenza viruses descended from a common ancestor closely related to the virus that caused the 1918 influenza pandemic . Figure 8 shows a phylogenetic tree of these sequences . The WSN HA used in our deep mutational scanning falls relatively close to the root of this tree . 10 . 7554/eLife . 03300 . 016Figure 8 . A phylogenetic tree of human and swine H1 HA sequences descended from a common ancestor closely related to the 1918 virus . The WSN virus used in the experiments here is a lab-adapted version of the A/Wilson Smith/1933 strain . Human H1N1 that circulated from 1918 until 1957 is shown in blue . Human seasonal H1N1 that reappeared in 1977 is shown in purple . Swine H1N1 is shown in red . The 2009 pandemic H1N1 is shown in green . This tree was constructed using codonPhyML ( Gil et al . , 2013 ) with the substitution model of Goldman and Yang ( 1994 ) . This plot is the file CodonPhyML_Tree_H1_HumanSwine_GY94/annotated_tree . pdf described at http://jbloom . github . io/phyloExpCM/example_2014Analysis_Influenza_H1_HA . html . Figure 8—figure supplement 1 shows a tree estimated for the same sequences using the substitution model of Kosiol et al . ( 2007 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 03300 . 01610 . 7554/eLife . 03300 . 017Figure 8—figure supplement 1 . A phylogenetic tree of the same sequences shown in Figure 8 , this time inferred using the substitution model of Kosiol et al . ( 2007 ) . This tree is extremely similar to that in Figure 8 , indicating the inferred topology is robust to the exact choice of codon-substitution model . This plot is the file CodonPhyML_Tree_H1_HumanSwine_KOSI07/annotated_tree . pdf described at http://jbloom . github . io/phyloExpCM/example_2014Analysis_Influenza_H1_HA . html . DOI: http://dx . doi . org/10 . 7554/eLife . 03300 . 017 The crudest comparison is simply to correlate amino-acid frequencies in the natural sequences to the experimentally inferred amino-acid preferences . Figure 9 shows that the inferred preferences are substantially although imperfectly correlated with the natural amino-acid frequencies . However , this comparison is problematic because it fails to account for the contingent and limited sampling of mutations by natural evolution . While the deep mutational scanning is designed to sample all possible mutations , only a fraction of theoretically tolerable mutations have fixed in natural H1 HAs due to the finite timespan during which evolution has been exploring possible sequences ( in other words , evolution is not at equilibrium; see Povolotskaya and Kondrashov , 2010 ) . Therefore , an amino-acid frequency of close to one among the natural HA sequences in Figure 8 might imply an absolute functional requirement for that amino acid—or it might simply mean that natural evolution has not yet happened to fix a mutation to another tolerable amino acid at that site . 10 . 7554/eLife . 03300 . 018Figure 9 . The frequencies of amino acids among the naturally occurring HA sequences in Figure 8 vs the amino-acid preferences inferred from the combined replicates ( Figure 5 ) . Note that a natural frequency close to one or zero could indicate absolute selection for or against a specific amino acid , but could also simply result from the fact that natural evolution has not completely sampled all possible mutations compatible with HA structure and function . The Pearson correlation coefficient ( R ) and associated p-value are shown on the plot . This plot is the file natural_frequency_vs_preference . pdf described at http://jbloom . github . io/phyloExpCM/example_2014Analysis_Influenza_H1_HA . html . DOI: http://dx . doi . org/10 . 7554/eLife . 03300 . 018 A better approach is therefore to treat natural evolution as a non-equilibrium dynamic process , and ask whether the inferred amino-acid preferences accurately describe this process . This type of analysis can be done using the likelihood-based statistical framework for phylogenetics developed by Felsenstein ( 1973 , 1981 ) . Specifically , we fix the phylogenetic tree topology to that shown in Figure 8 and then assess the likelihood of the natural sequences given a specific evolutionary model after optimizing the branch lengths of the tree . Evolutionary models that more accurately describe HA sequence evolution will have higher likelihoods , and the relative accuracy of models can be quantified by comparing their likelihoods after correcting for the number of free parameters using AIC ( Posada and Buckley , 2004 ) . Previous work has described how experimental measurements of amino-acid preferences can be combined with known mutation rates to create a parameter-free phylogenetic evolutionary model from deep mutational scanning data ( Bloom , 2014 ) . Table 2 and Table 3 compare the fit of evolutionary models based on the experimentally inferred amino-acid preferences with several existing state-of-the-art models that do not utilize this experimental information ( Goldman and Yang , 1994; Kosiol et al . , 2007 ) . The model based on amino-acid preferences inferred from the combined experimental data from the three replicates describes the evolution of the naturally occurring HA sequences far better than the alternative models , despite the fact that the latter have a variety of free parameters that are optimized to improve the fit . Models based on amino-acid preferences inferred from the individual experimental replicates also fit the data better than existing models—however , the fit is poorer than for the model that utilizes the data from all three replicates . This result is consistent with the fact that the individual replicates are incomplete in their sampling of the mutational effects , meaning that aggregating the data from several replicates improves the accuracy of inferred preferences . Overall , these comparisons show that the deep mutational scanning reflects the actual constraints on HA evolution substantially better than existing quantitative evolutionary models . 10 . 7554/eLife . 03300 . 019Table 2 . An evolutionary model derived from the experimentally inferred amino-acid preferences describes the HA sequence phylogeny in Figure 8 far better than a variety of existing state-of-the-art modelsDOI: http://dx . doi . org/10 . 7554/eLife . 03300 . 019ModelΔ AICLog likelihoodParameters ( optimized + empirical ) Combined0 . 0−24088 . 70 ( 0 + 0 ) replicate 3303 . 2−24240 . 30 ( 0 + 0 ) Combined , Halpern and Bruno500 . 6−24339 . 00 ( 0 + 0 ) replicate 1535 . 4−24356 . 40 ( 0 + 0 ) replicate 3 , Halpern and Bruno657 . 8−24417 . 60 ( 0 + 0 ) replicate 2876 . 2−24526 . 80 ( 0 + 0 ) GY94 , gamma ω , gamma rates882 . 6−24517 . 013 ( 4 + 9 ) replicate 1 , Halpern and Bruno983 . 2−24580 . 30 ( 0 + 0 ) GY94 , gamma ω , one rate1109 . 7−24631 . 512 ( 3 + 9 ) replicate 2 , Halpern and Bruno1190 . 0−24683 . 70 ( 0 + 0 ) KOSI07 , gamma ω , gamma rates1620 . 5−24834 . 964 ( 4 + 60 ) GY94 , one ω , gamma rates1859 . 4−25006 . 412 ( 3 + 9 ) KOSI07 , gamma ω , one rate1883 . 0−24967 . 263 ( 3 + 60 ) KOSI07 , one ω , gamma rates2378 . 8−25215 . 163 ( 3 + 60 ) GY94 , one ω , one rate2544 . 5−25350 . 011 ( 2 + 9 ) KOSI07 , one ω , one rate3040 . 0−25546 . 762 ( 2 + 60 ) combined , randomized5632 . 8−26905 . 10 ( 0 + 0 ) replicate 1 , randomized6002 . 4−27089 . 90 ( 0 + 0 ) replicate 3 , randomized6138 . 8−27158 . 10 ( 0 + 0 ) replicate 2 , randomized6477 . 8−27327 . 60 ( 0 + 0 ) combined , randomized , Halpern and Bruno7072 . 8−27625 . 10 ( 0 + 0 ) replicate 1 , randomized , Halpern and Bruno7795 . 0−27986 . 20 ( 0 + 0 ) replicate 3 , randomized , Halpern and Bruno7891 . 8−28034 . 60 ( 0 + 0 ) replicate 2 , randomized , Halpern and Bruno8494 . 4−28335 . 90 ( 0 + 0 ) The model is most accurate if it utilizes data from the combined experimental replicates , but it also outperforms existing models even if the data are only derived from individual replicates . Models are ranked by AIC ( Posada and Buckley , 2004 ) . GY94 indicates the model of Goldman and Yang ( 1994 ) , and KOSI07 indicates the model of Kosiol et al . ( 2007 ) . The nonsynonymous/synonymous ratio ( ω ) and the substitution rate are either estimated as a single value or drawn from a four-category gamma distribution . Randomizing the experimentally inferred preferences among sites makes the models far worse . The models work best fixation probabilities are computed from the preferences using the first equation proposed in Bloom ( 2014 ) . The table also shows the results if the fixation probabilities are instead computed using the equation of Halpern and Bruno ( 1998 ) as described in Bloom ( 2014 ) . This table is the file H1_HumanSwine_GY94_summary . tex described at http://jbloom . github . io/phyloExpCM/example_2014Analysis_Influenza_H1_HA . html . Table 3 shows the results when the tree topology is instead estimated using the substitution model of Kosiol et al . ( 2007 ) . 10 . 7554/eLife . 03300 . 020Table 3 . An evolutionary model derived from the experimentally inferred amino-acid preferences also outperforms existing models for the tree topology in Figure 8—figure supplement 1DOI: http://dx . doi . org/10 . 7554/eLife . 03300 . 020ModelΔAICLog likelihoodParameters ( optimized + empirical ) Combined0 . 0−24082 . 50 ( 0 + 0 ) replicate 3304 . 8−24234 . 90 ( 0 + 0 ) Combined , Halpern and Bruno494 . 4−24329 . 70 ( 0 + 0 ) replicate 1534 . 2−24349 . 60 ( 0 + 0 ) replicate 3 , Halpern and Bruno653 . 2−24409 . 10 ( 0 + 0 ) replicate 2869 . 4−24517 . 20 ( 0 + 0 ) GY94 , gamma ω , gamma rates876 . 7−24507 . 813 ( 4 + 9 ) replicate 1 , Halpern and Bruno976 . 8−24570 . 90 ( 0 + 0 ) GY94 , gamma ω , one rate1101 . 0−24621 . 012 ( 3 + 9 ) replicate 2 , Halpern and Bruno1180 . 4−24672 . 70 ( 0 + 0 ) KOSI07 , gamma ω , gamma rates1609 . 0−24823 . 064 ( 4 + 60 ) GY94 , one ω , gamma rates1856 . 2−24998 . 612 ( 3 + 9 ) KOSI07 , gamma ω , one rate1867 . 3−24953 . 163 ( 3 + 60 ) KOSI07 , one ω , gamma rates2367 . 9−25203 . 463 ( 3 + 60 ) GY94 , one ω , one rate2548 . 3−25345 . 611 ( 2 + 9 ) KOSI07 , one ω , one rate3028 . 0−25534 . 562 ( 2 + 60 ) Combined , randomized5628 . 0−26896 . 50 ( 0 + 0 ) replicate 1 , randomized5993 . 6−27079 . 30 ( 0 + 0 ) replicate 3 , randomized6138 . 0−27151 . 50 ( 0 + 0 ) replicate 2 , randomized6475 . 2−27320 . 10 ( 0 + 0 ) combined , randomized , Halpern and Bruno7069 . 4−27617 . 20 ( 0 + 0 ) replicate 1 , randomized , Halpern and Bruno7786 . 8−27975 . 90 ( 0 + 0 ) replicate 3 , randomized , Halpern and Bruno7889 . 2−28027 . 10 ( 0 + 0 ) replicate 2 , randomized , Halpern and Bruno8496 . 0−28330 . 50 ( 0 + 0 ) This table differs from Table 2 in that it uses the tree topology inferred with the model of Kosiol et al . ( 2007 ) rather than Goldman and Yang ( 1994 ) . This table is the file H1_HumanSwine_KOSI07_summary . tex described at http://jbloom . github . io/phyloExpCM/example_2014Analysis_Influenza_H1_HA . html . The amino-acid preferences inferred from the deep mutational scanning reflect the inherent mutational tolerance of sites in HA . In contrast , the evolution of HA in nature is shaped by a combination of HA's inherent mutational tolerance and external selection pressures . Specifically , the evolution of HA in humans is strongly driven by selection for mutations that alter antigenicity ( Yewdell et al . , 1979; Wiley et al . , 1981; Caton et al . , 1982; Smith et al . , 2004; Das et al . , 2013; Koel et al . , 2013; Bedford et al . , 2014 ) . The fact that such antigenic mutations fix at high frequency implies some degree of mutational tolerance at antigenic sites , since no mutations would fix if these sites were under absolute structural or functional constraint . However , it is not possible to tell from natural sequences alone whether antigenic sites are unusually mutationally tolerant compared to the rest of HA , or whether their rapid evolution is solely because they are under strong external immune selection . To address this issue , we used the results of the deep mutational scanning to compare the inherent mutational tolerance of antigenic sites to the rest of the HA protein . Caton et al . ( 1982 ) mapped the antigenic sites of the H1 HA from A/Puerto Rico/8/1934 ( PR8 ) , which is closely related to the WSN HA used in our experiments . We therefore defined the ‘Caton et al . antigenic sites’ as the WSN residues homologous to those mapped by Caton et al . ( 1982 ) with the exclusion of a single site that has gained glycosylation in the WSN HA relative to the PR8 HA ( see ‘Materials and methods’ for details ) . One possible concern is that Caton et al . ( 1982 ) mapped antigenic sites largely by selecting monoclonal-antibody escape mutants , and so these sites might be biased towards being more mutationally tolerant . We therefore also made a broader classification of ‘antigenic sites and contacting residues’ consisting of the Caton et al . antigenic sites plus all surface-exposed residues in contact with these sites ( see ‘Materials and methods’ for details ) . This broader classification includes all residues in regions of the HA surface targeted by antibodies , and so should not be biased by whether sites are amenable to the selection of monoclonal-antibody escape mutants . We hypothesized that both sets of antigenic sites would have unusually high mutational tolerance . For comparison , we used two classifications of receptor-binding residues ( ‘Materials and methods’ ) . The first classification consists of residues that have important roles in receptor binding ( Martin et al . , 1998 ) and are conserved in H1 HAs; these residues are mostly deep in the binding pocket . The second classification consists of all residues that contact the sialic-acid receptor in the crystal structure , regardless of their level of conservation . We hypothesized that the core set of conserved receptor-binding residues would have unusually low mutational tolerance , but that the set of all receptor-binding residues would have typical levels of mutational tolerance since influenza routinely escapes from antibodies that target the periphery of the receptor-binding pocket ( Koel et al . , 2013 ) . The positions of the Caton et al . antigenic sites and the conserved receptor-binding residues in the primary sequence are indicated by the top overlay bar in Figure 5 . Visual inspection suggests that the conserved receptor-binding residues are indeed relatively intolerant of mutations ( have a strong preference for one specific amino acid ) , whereas the Caton et al . antigenic sites are relatively tolerant of mutations ( have roughly equivalent preferences for many amino acids ) . For a more quantitative analysis , we computed a site entropy from the inferred amino-acid preferences—larger site entropies indicate a higher inherent tolerance for mutations . The site entropies of all residues are displayed on the HA protein structure in Figure 10 . Visual inspection suggests that both classifications of antigenic sites have unusually high mutational tolerance , whereas the conserved receptor-binding residues have unusually low mutational tolerance . 10 . 7554/eLife . 03300 . 021Figure 10 . Inherent mutational tolerance of HA’s receptor-binding residues and antigenic sites . ( A ) Surface of HA with one monomer colored by site entropy as determined by the deep mutational scanning; blue indicates low mutational tolerance and red indicates high mutational tolerance . ( B ) The structure shows residues classified as antigenic sites by Caton et al . ( 1982 ) in colored spheres; the plot shows site entropy vs relative solvent accessibility ( RSA ) of these residues ( red triangles ) and all other HA1 residues in the crystal structure ( blue circles ) . ( C ) Antigenic sites of Caton et al . ( 1982 ) plus all other surface-exposed residues that contact these sites . ( D ) Conserved receptor-binding residues . ( E ) All receptor-binding residues . Table 4 shows that residues in ( B ) and ( C ) have unusually high mutational tolerance , residues in ( D ) have unusually low mutational tolerance , and residues in ( E ) do not have unusual mutational tolerance . The data and code to create all panels of this figure is provided via http://jbloom . github . io/mapmuts/example_WSN_HA_2014Analysis . html . The structure is PDB 1RVX ( Gamblin et al . , 2004 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 03300 . 021 We next tested whether these visual observations were supported by a rigorous statistical analysis . A confounding factor in comparing mutational tolerance across different sets of residues is that sites with higher solvent accessibility are typically more tolerant of mutations ( Bustamante et al . , 2000; Ramsey et al . , 2011 ) . To correct for this fact , we computed the relative solvent accessibility ( RSA ) for all residues in the HA crystal structure . Residues with RSAs close to zero are buried and are expected to be fairly intolerant of mutations , whereas residues with RSAs substantially greater than zero are surface exposed and are expected to be fairly tolerant of mutations . Figure 10 plots site entropy as a function of RSA for HA1 residues . This figure shows that sites with higher RSA are more mutationally tolerant as expected . However , the figure also suggests that both classifications of antigenic sites are more mutationally tolerant than other residues with equivalent RSA . The figure also suggest that the conserved receptor-binding residues are less mutationally tolerant than other residues with equivalent RSA , whereas the set of all receptor-binding residues have fairly typical mutational tolerance . These observations are supported by the statistical analyses in Table 4: even after correcting for RSA , there is a significant trend for antigenic sites to have high mutational tolerance , and for conserved receptor-binding residues to have low mutational tolerance . 10 . 7554/eLife . 03300 . 022Table 4 . The antigenic sites are more significantly mutationally tolerant than other HA1 residues with similar relative solvent accessibility ( RSA ) , the conserved receptor-binding residues are significantly less mutationally tolerant than other similar residues , and sites in the more expansive set of all receptor-binding residues have typical levels of mutational toleranceDOI: http://dx . doi . org/10 . 7554/eLife . 03300 . 022Model: site entropy ∼ RSA + ( Caton et al . antigenic site ) + intercept PropertyEstimateStandard errorp-value RSA1 . 290 . 12<10−10 Caton et al . antigenic site0 . 300 . 091 . 6 × 10−3Model: site entropy ∼ RSA + ( antigenic site or contacting residue ) + intercept PropertyEstimateStandard errorp-value RSA1 . 220 . 13<10−10 antigenic site or contacting residue0 . 230 . 072 . 2 × 10−3Model: site entropy ∼ RSA + ( conserved receptor binding ) + intercept PropertyEstimateStandard errorp-value RSA1 . 380 . 11<10−10 conserved receptor binding−0 . 520 . 161 . 7 × 10−3Model: site entropy ∼ RSA + ( all receptor binding ) + intercept PropertyEstimateStandard errorp-value RSA1 . 400 . 11<10−10 all receptor binding−0 . 180 . 110 . 12The sets of residues analyzed here are those shown in Figure 10 . Shown here are the results of multiple linear regression of the continuous dependent variable of site entropy ( as computed from the amino-acid preferences ) vs the continuous independent variable of RSA and the binary variable of being a receptor-binding residue or being an antigenic site . The data and code used to perform these analyses are available via http://jbloom . github . io/mapmuts/example_WSN_HA_2014Analysis . html . Overall , these results show that antigenic sites in HA have unusually high inherent mutational tolerance , suggesting that this property combines with external immune selection to contribute to HA’s rapid antigenic evolution . These results also show that while a core group of conserved residues deep in the receptor-binding pocket have unusually low mutational tolerance , the bulk of residues that contact the receptor are not under exceptional constraint . This fact probably explains why HA is able to escape from antibodies targeting the periphery of the receptor-binding pocket ( Koel et al . , 2013 ) , and why only rare antibodies that penetrate deep into this pocket are broadly neutralizing ( Whittle et al . , 2011 ) . The foregoing results show that the antigenic sites in HA have an unusually high inherent tolerance for mutations . Is this antigenic evolvability an exceptional feature of HA , or is it commonly shared by other viral proteins ? Ideally one would compare HA to the major surface antigens of other viruses with high ( e . g . , HIV ) and low ( e . g . , measles ) rates of antigenic evolution—but unfortunately comparable data sets for these other viruses are not yet available . Therefore , we instead compared the antigenic evolvability of HA to that of influenza nucleoprotein ( NP ) , a protein for which we have recently performed a similar deep mutational scanning experiment ( Bloom , 2014 ) . The adaptive immune system targets NP via cytotoxic T-lymphocytes ( CTLs ) ( Valkenburg et al . , 2011 ) . Although the selection exerted by these CTLs is believed to be weaker than the antibody-mediated selection on HA’s antigenic sites ( Bhatt et al . , 2011 ) , influenza does benefit from mutations in NP that promote escape from CTLs ( Berkhoff et al . , 2007; Valkenburg et al . , 2013 ) . However , whereas HA rapidly evolves to escape from antibodies , NP does not appear to have any special propensity for rapid evolution of the epitopes targeted by CTLs . Instead , mutations in NP's CTL epitopes are often deleterious and require secondary permissive or compensatory mutations to fix without a fitness cost ( Rimmelzwaan et al . , 2004; Berkhoff et al . , 2005 , 2006; Gong et al . , 2013 ) . Therefore , we hypothesized that unlike HA's highly evolvable antigenic sites , NP's CTL-antigenic sites would not possess unusually high inherent mutational tolerance . To test this hypothesis , we used a previously described delineation of epitopes in NP from the human H3N2 strain A/Aichi/2/1968 with experimentally validated human CTL responses ( Gong and Bloom , 2014 ) . In this delineation , less than a quarter of NP’s sites participate in multiple CTL epitopes . We used the results of our previous deep mutational scanning of NP to compare the inherent mutational tolerance of sites that participate in multiple CTL epitopes to all other sites in NP . As shown in Figure 11 and Table 5 , the NP sites involved in multiple CTL epitopes have an inherent mutational tolerance that is indistinguishable from other sites in the protein . Therefore , NP does not possess any special inherent mutational tolerance in its CTL epitopes . This finding implies that a high level of antigenic evolvability is not a general feature of all viral proteins , but is instead at least somewhat unique to HA . 10 . 7554/eLife . 03300 . 023Figure 11 . The inherent mutational tolerance of NP's CTL epitopes is indistinguishable from that of non-epitope sites in NP . The plot shows the site entropy vs relative solvent accessibility ( RSA ) of NP residues that participate in multiple CTL epitopes ( red triangles ) and all other NP residues in the crystal structure ( blue circles ) . Visual inspection suggests that the epitope sites have mutational tolerance comparable to other sites , and this result is supported by the statistical analysis in Table 5 . Note that unlike for HA , there is no trend for RSA to correlate with site entropy—this could be because many of NP’s surface-exposed sites are constrained by interactions with viral RNA . The CTL epitopes are those delineated in the first supplementary table of Gong and Bloom ( 2014 ) . The site entropies are computed from a previously described deep mutational scan of NP , and are the values in the first supplementary file of Bloom ( 2014 ) ; the RSA values are also taken from that reference . The data and code used to generate this plot is available via http://jbloom . github . io/mapmuts/example_WSN_HA_2014Analysis . html; the plot itself is the file NP_CTL_entropy_rsa_correlation . pdf described therein . DOI: http://dx . doi . org/10 . 7554/eLife . 03300 . 02310 . 7554/eLife . 03300 . 024Table 5 . There is no statistically significant difference between the inherent mutational tolerance of NP sites involved in multiple CTL epitopes and all other NP residuesDOI: http://dx . doi . org/10 . 7554/eLife . 03300 . 024Model: NP site entropy ∼ RSA + ( multiple CTL epitopes ) + interceptPropertyEstimateStandard errorp-valueRSA−0 . 050 . 070 . 52multiple CTL epitopes−0 . 040 . 040 . 31The table shows the result of multiple linear regression of the continuous dependent variable of site entropy ( as computed from the amino-acid preferences ) vs the continuous independent variable of RSA and the binary variable of participating in multiple CTL epitopes . The data set analyzed here is plotted in Figure 11 . The data and code used to perform this analysis are available via http://jbloom . github . io/mapmuts/example_WSN_HA_2014Analysis . html .
A fundamental challenge in studying the natural evolution of influenza is separating the effects of external selection pressures from inherent structural and functional constraints . The evolutionary patterns observed in natural sequences are shaped by a combination of inherent mutational tolerance and external pressures such as immune selection , and the analysis of such sequences is further confounded by the fact that influenza is not at evolutionary equilibrium . Here we have quantified the inherent mutational tolerance of influenza HA by using deep mutational scanning ( Fowler et al . , 2010; Araya and Fowler , 2011 ) to simultaneously assess the impact on viral growth of the vast majority of the ≈104 possible amino-acid mutations to influenza HA . The information obtained from the deep mutational scanning is consistent with existing knowledge about the effects of mutations on HA function and structure . For instance , the deep mutational scanning shows strong selection for specific amino acids known to play important roles in HA's receptor-binding activity , fusion activity , and proteolytic activation ( Martin et al . , 1998; Qiao et al . , 1999; Stech et al . , 2005 ) . Similarly , at the sites of known temperature-sensitive mutations to HA ( Nakajima et al . , 1986 ) , the deep mutational scanning identifies the more stabilizing amino-acid as more favorable . Broader trends from the deep mutational scanning are also in agreement with current thinking about mutational effects . For example , the deep mutational scanning finds that there is strong purifying selection against stop-codon mutations and many nonsynonymous mutations , but that there is only weak selection against synonymous mutations . All of these results suggest that the deep mutational scanning faithfully captures both the specific and general effects of mutations on HA . The comprehensive information generated by the deep mutational scanning can be used to create quantitative evolutionary models for analyzing HA sequence phylogenies . Here we have shown that an evolutionary model constructed from our deep mutational scanning data describes the evolution of human and swine H1 HAs far better than existing state-of-the-art models for sequence evolution . We anticipate that separating HA's inherent mutational tolerance from external selection should also eventually allow the external selection pressures to be studied in greater detail . For example , one might imagine that sites in HA that exhibit evolutionary patterns that deviate from the quantitative model created from our deep mutational scanning are likely to be under external selection . Future work that augments deep mutational scanning with specific experimentally defined selection pressures ( such as antibodies against HA ) could aid in further elucidation of the forces that shape influenza evolution . It also may be possible to utilize high-throughput experimental data on mutational effects to better estimate the fitness of naturally occurring strains in a way that aids in prediction of the year-to-year strain dynamics of influenza ( Łuksza and Lässig , 2014 ) . The deep mutational scanning also enabled us to assess the extent to which HA's inherent mutational tolerance contributes to influenza’s antigenic evolvability . It remains a mystery why error-prone RNA viruses differ so widely in their capacity for evolutionary escape from immunity , with some ( e . g . , influenza and HIV ) undergoing rapid antigenic evolution while others ( e . g . , measles ) show little antigenic change on relevant timescales ( Lipsitch and O’Hagan , 2007; Koelle et al . , 2006; Heaton et al . , 2013 ) . Our data demonstrate that the antigenic sites in HA are unusually tolerant to mutations , implying that inherent evolutionary plasticity at sites targeted by the immune system is one factor that contributes to influenza's rapid antigenic evolution . This high mutational tolerance at antigenic sites could itself be a property that influenza has evolved to aid in its antigenic escape—or it might simply be an unfortunate coincidence that the immune system focuses on especially plastic portions of HA . In either case , it is intriguing to speculate whether a high inherent mutational tolerance in antigenic sites is also a feature of other antigenically variable RNA viruses . Application of the deep mutational scanning approach used here to additional viruses should provide a means to address this question .
Illumina sequencing data are available at the SRA , accession SRP040983 ( http://www . ncbi . nlm . nih . gov/sra/ ? term=SRP040983 ) . Source code and a description of the computational process used to analyze the sequencing data and infer the amino-acid preferences is at http://jbloom . github . io/mapmuts/example_WSN_HA_2014Analysis . html . Source code and a description of the computational process used for the phylogenetic analyses is available at http://jbloom . github . io/phyloExpCM/example_2014Analysis_Influenza_H1_HA . html . A variety of different numbering schemes for HA are used in the literature . Unless noted otherwise , residues are numbered here using sequential numbering of the WSN HA protein sequence ( Supplementary file 1 ) starting with one at the N-terminal methionine . In some cases , the number of the corresponding residues in the widely used H3 numbering scheme is also indicated . These numbering systems can be interconverted using the Python script available at https://github . com/jbloom/HA_numbering . The HA codon-mutant library was generated using the oligo-based PCR mutagenesis protocol described previously by Bloom ( 2014 ) . The only differences from that protocol were that HA was used as the template rather than NP , and that only two overall rounds of mutagenesis were performed , rather than the three rounds used by Bloom ( 2014 ) . This reduction in the number of rounds of mutagenesis reduced the average number of codon mutations from the ≈ three per clone in Bloom ( 2014 ) to the ≈ two per clone shown in Figure 2 . The libraries were created in full biological triplicate , meaning that each experimental replicate was derived from an independent plasmid mutant library . The end primers for the mutagenesis were 5′-cgatcacgtctctgggagcaaaagcaggggaaaataaaaacaac-3′ and 5′-gatacacgtctcatattagtagaaacaagggtgtttttccttatatttctg-3′ ( these primers include BsmBI restriction sites ) . The mutagenic primers were ordered from Integrated DNA Technologies , and are listed in Supplementary file 2 . The final products from the codon mutagenesis PCR were gel purified and digested with BsmBI ( R0580L; New England Biolabs , Ipswich , Massachusetts ) . The BsmBI-digested HA was ligated into a dephosphorylated ( Antarctic Phosphatase , M0289L; New England Biolabs ) and BsmBI-digested preparation of the bidirectional reverse-genetics plasmid pHW2000 ( Hoffmann et al . , 2000 ) using T4 DNA ligase ( M0202S; New England Biolabs ) . Column-purified ligations were electroporated into ElectroMAX DH10B T1 phage-resistant competent cells ( 12033-015; Invitrogen , Carlsbad , California ) and plated on LB plates containing 100 μg/ml of ampicilin . A 1:4000 dilution of each transformation was plated in parallel to enable estimation of the number of unique transformants—we obtained at least two-million unique colonies per transformation . For each replicate of the codon-mutant library , we performed three transformations to generate approximately six-million independent clones per replicate library . Control ligations lacking an insert yielded at least 100 times fewer colonies , indicating a very low rate of background self-ligation of the pHW2000 plasmid . The transformants from each HA mutant library replicate were pooled , cultured in LB supplemented with ampicillin , and mini-prepped to generate the HA codon mutant plasmid libraries . For the Sanger sequencing analysis shown in Figure 2 , we picked and prepped 34 independent colonies for sequencing . The full analysis of this Sanger sequencing is available at https://github . com/jbloom/SangerMutantLibraryAnalysis/tree/v0 . 2 . The HA mutant plasmid libraries were used to generate pools of mutant influenza viruses by reverse genetics ( Hoffmann et al . , 2000 ) . Cocultures of 293T and MDCK-SIAT1 cells were transfected with equal amounts of HA ( either unmutated or one of the plasmid mutant libraries ) cloned into pHW2000 as described above , plus the seven other WSN genes in bidirectional reverse-genetics plasmids ( pHW181-PB2 , pHW182-PB1 , pHW183-PA , pHW185-NP , pHW186-NA , pHW187-NA , pHW188-NS ) , which were kind gifts from Robert Webster of St . Jude Children's Research Hospital . Overall , six viral rescues and passages were performed , each using a different HA plasmid preparation: the three HA mutant library replicates ( eventually yielding the mutvirus samples in Figure 1 ) and three independent unmutated HAs ( eventually yielding the virus samples in Figure 1 ) . Each of the viral rescues was performed by transfecting multiple wells of cells in an effort to increase the diversity of the rescued viruses . Specifically , two 12-well dishes were transfected per rescue . Cells were plated at 2 × 105 293T cells and 5 × 104 MDCK-SIAT1 cells per well in D10 ( DMEM supplemented with 10% heat-inactivated FBS , 2 mM L-glutamine , 100 U of penicillin/ml , and 100 μg of streptomycin/ml ) , and then each well was transfected with 1 μg of total plasmid DNA ( 125 ng of each of the eight plasmids ) using the BioT transfection reagent ( Bioland B01-02 , Paramount , California ) . At 12 to 18 hr post-transfection , the medium was changed to our WSN viral growth media: Opti-MEM supplemented with 0 . 5% heat-inactivated FBS , 0 . 3% BSA , 100 U of penicillin/ml , 100 μg of streptomycin/ml , and 100 μg of calcium chloride/ml . This media does not contain trypsin since viruses with the WSN HA and NA are trypsin independent ( Goto and Kawaoka , 1998 ) . Viral supernatants were collected 72 hr post-transfection , and the supernatants from the different wells were pooled for each viral rescue . These pooled supernatants were then clarified by centrifugation at 2000×g for 5 min , aliquoted , and frozen at −80 oC . Aliquots were then thawed and titered by TCID50 ( see below ) . For viral passage , each viral rescue replicate was passaged in four 10-cm dishes . Briefly , 6 × 106 MDCK-SIAT1 cells per 10-cm dish in WSN viral growth media were infected with 6×105 infectious particles ( multiplicity of infection of 0 . 1 ) . Since there are four dishes for each replicate , this maintains a diversity of 2 . 4 × 106 TCID50 units per replicate . The passaged viral supernatants were collected at 50 hr post-infection , and the supernatants for the four plates were pooled for each replicate . These pooled supernatants were clarified at 2000 × g for 5 min , aliquoted , and frozen at −80 oC . Aliquots were then thawed and titered by TCID50 . The viruses were titered by TCID50 ( 50% tissue culture infectious dose ) . In this assay , 10 μl of a 1:10 dilution of the viral supernatant to be titered was added to the first row of a 96-well tissue culture plate containing 90 μl of WSN viral growth media . At least one no-virus control supernatant was included on each plate as a negative control . The virus was then serially diluted 1:10 down the rows of the plates , and then 5 × 103 MDCK-SIAT1 cells were added to each well . The plates were then incubated at 37°C , and scored for cytopathic effects caused by viral growth after for 65–72 hr . Virus titers were calculated by the method of Reed and Muench ( 1938 ) implemented via the Python script at https://github . com/jbloom/reedmuenchcalculator . The deep sequencing samples were prepared from PCR amplicons that were generated exactly as described for the DNA , mutDNA , virus , and mutvirus samples in Bloom ( 2014 ) . The viral RNA template for the virus and mutvirus were isolated using freshly purchased Trizol reagent ( 15596-026; Life Technologies ) in order to avoid any oxidative damage associated with old reagents . After performing reverse transcription as described in Bloom ( 2014 ) , quantitative PCR ( qPCR ) was used to quantify the number of HA cDNA molecules to ensure that there were at least 106 unique template molecules before beginning the subsequent PCR amplification . The qPCR primers were designed based on those described by Marsh et al . ( 2007 ) , and were 5′-taacctgctcgaagacagcc-3′ and 5′-agagccatccggtgatgtta-3′ . The PCR amplicons were fragmented and barcoded using the custom modification of Illumina's Nextera kit using the protocol described in Bloom ( 2014 ) . Samples were barcoded as follows: DNA–N701 , mutDNA–N702 , virus–N704 , and mutvirus–N705 . For each of the three biological replicates , these four samples were pooled and sequenced on their own Illumina lane with 50-nucleotide paired-end reads as described in Bloom ( 2014 ) . For the technical sequencing repeat of biological replicate #1 , the library preparation and sequencing were repeated from the same viral RNA templates . This technical repeat therefore only quantifies variation associated with sample preparation and sequencing , whereas the biological replicates also quantify variation associated with the processes of codon-mutant library creation , virus generation , and virus passage . The deep sequencing data was analyzed using the mapmuts computer program ( Bloom , 2014 ) . A description of the analysis approach and the resulting data files and figures produced are available at http://jbloom . github . io/mapmuts/example_WSN_HA_2014Analysis . html . Briefly , paired reads were overlapped as illustrated in Figure 3—figure supplement 1 and then aligned to HA . Reads were retained only if both reads in the pair passed the default Illumina filter , had average Q-scores of at least 25 , overlapped for at least 30 nucleotides with no more than one mismatch , and the overlap aligned to the HA gene with no more than six mismatches . Figure 3—figure supplement 2 shows the number of reads for each sample that met these criteria . Most reads that did not meet these criteria failed to do so because they could not be paired with at least 30 nucleotides of overlap—a situation that arises when the HA fragment produced by the Nextera fragmentation produces a fragment smaller than 30 nucleotides or larger than 70 nucleotides . Codon identities were called only if both overlapped paired reads agreed on the identity of the codon . This requirement reduces the error rate , because it is rare for both paired reads to independently experience the same sequencing error . As shown in Figure 4 , we estimated that 85% of possible codon mutations were sampled at least five times by the mutant viruses . To estimate the fraction of amino-acid mutations that would have been sampled , we simulated randomly selecting 85% of the mutant codons from the HA sequence , and determined that these codons encoded ≈97% of the amino-acid mutations . The counts of each codon identity in the deep sequencing data was used to infer the ‘preference’ of each site for each amino acid as described in Bloom ( 2014 ) . This inference was also done using the mapmuts computer program as detailed at http://jbloom . github . io/mapmuts/example_WSN_HA_2014Analysis . html . Briefly , the preference πr , a of site r for amino-acid a represents the expected frequency of that amino acid in a hypothetical library where each amino-acid is introduced at equal frequency . Specifically , the expected frequency fr , xmutvirus of mutant codon x at site r in the mutvirus sample is related to the preference for its encoded amino-acid A ( x ) byfr , xmutvirus=ϵr , x+ρr , x+μr , x×πr , A ( x ) ∑yμr , y×πr , A ( y ) , where ϵr , x is the rate at which site r is erroneously read to be codon x , ρr , x is the rate at which site r is erroneously reverse-transcribed to codon x , and μr , x is the rate at which site r is mutagenized to codon x in the mutant DNA sample . These unknown error and mutation rate parameters are inferred from the DNA , virus , and mutvirus samples using the Bayesian approach described in Bloom ( 2014 ) . Inferences of the posterior mean preferences πr , a were made separately for each replicate of the experiment , and the correlations among these inferences from different replicates are in Figure 6 . The final ‘best’ inferred preferences from the combined data of the three biological replicates were obtained by averaging the preferences obtained from the three biological replicates . These final inferred preferences are provided in Supplementary file 3 and displayed graphically in Figure 5 . The site entropies in Figure 10 and Table 4 were calculated from the amino-acid preferences as hr=∑a πr , a×log2πr , a . These site entropies are therefore in bits . Higher site entropies indicate a higher inherent mutational tolerance . The inferred amino-acid preferences were compared to amino-acid frequencies in an alignment of naturally occurring H1N1 HAs from swine and human lineages descended from a close relative of the 1918 virus . Briefly , all full-length H1 HAs from these hosts were downloaded from the Influenza Virus Resource ( Bao et al . , 2008 ) . Up to three sequences per host and year were randomly subsampled and used to build a phylogenetic tree . Clear outliers from the molecular clock ( typically lab artifacts or mis-annotated sequences ) were iteratively excluded and the trees were rebuilt . The final sequence alignment is in Supplementary file 4 . This alignment was used to build the phylogenetic trees in Figure 8 and Figure 8—figure supplement 1 with codonPhyML ( Gil et al . , 2013 ) using the codon-substitution model of ( Goldman and Yang , 1994 ) or ( Kosiol et al . , 2007 ) with empirical codon frequencies determined using the CF3x4 method ( Pond et al . , 2010 ) or the F method , respectively . In both cases , the nonsynonymous-synonymous ratio ( ω ) was drawn from four gamma-distributed categories ( Yang et al . , 2000 ) . A description of this process is at http://jbloom . github . io/phyloExpCM/example_2014Analysis_Influenza_H1_HA . html . We compared the accuracy with which the naturally occurring HA phylogeny was described by an evolutionary model based on the experimentally measured amino-acid preferences vs several standard codon-substitution models . These comparisons were used made using HYPHY ( Pond et al . , 2005 ) and phyloExpCM ( Bloom , 2014 ) . A description of this analysis is at http://jbloom . github . io/phyloExpCM/example_2014Analysis_Influenza_H1_HA . html . Briefly , the phylogenetic tree topology was fixed to that shown in Figure 8 or Figure 8—figure supplement 1 . The branch lengths and any free parameters of the evolutionary model were then optimized by maximum likelihood . The experimentally determined evolutionary models were constructed from the inferred amino-acid preferences reported here and the experimentally measured mutation rates reported in Bloom ( 2014 ) . The ‘fixation probabilities’ were computed using either the Metropolis-like relationship described in Bloom ( 2014 ) or the relationship proposed by Halpern and Bruno ( 1998 ) . The results of these comparisons are in Tables 2 and 3 . All of these comparisons show that the experimentally determined evolutionary models are far superior to the various standard models . The WSN HA studied here has a high degree of sequence identity to the HA crystallized in PDB 1RVX ( Gamblin et al . , 2004 ) . It is this HA structure that is shown Figure 10 . The relative solvent accessibilities ( RSA ) values in Figure 5 and Figure 10 were calculated by first determining the absolute solvent accessibilities of the residues in the full trimeric HA in PDB 1RVX with the DSSP ( Joosten et al . , 2011 ) webserver at http://www . cmbi . ru . nl/hsspsoap/ , and then normalizing by the maximum solvent accessibilities given by Tien et al . ( 2013 ) . Several sub-classifications of HA residues were performed . Conserved receptor-binding sites were any residues listed in the first table of Martin et al . ( 1998 ) that are also conserved in at least 90% of H1 HAs . These residues are listed in Table 1 . All receptor-binding residues were any residues with any atom within 5 Å of the substrate in PDB 1RVX ( Gamblin et al . , 2004 ) . No constraint is placed on whether or not these residues are conserved in natural sequences . The residues that fall into this classification are ( in sequential numbering of the WSN HA ) : 108 , 147 , 148 , 149 , 150 , 151 , 158 , 166 , 168 , 196 , 198 , 199 , 203 , 207 , 238 , 239 , and 241 . The Caton et al . ( 1982 ) antigenic-site residues are classified based on antigenic mapping of the A/Puerto Rico/8/1934 ( H1N1 ) HA . Specifically , these are any residues listed in the third table of Caton et al . ( 1982 ) with the following exceptions: residue 182 ( H3 numbering ) is not considered for the reason explained on page 421 of Caton et al . ( 1982 ) , residue 273 ( H3 numbering ) is not considered for the reason explained on page 422 of Caton et al . ( 1982 ) , and residue 129 ( H3 numbering ) is not considered because it has gained a glycosylation site in the WSN HA that is not present in the A/Puerto Rico/8/1934 ( H1N1 ) HA and mutation of this WSN glycosylation site can strongly affect viral growth ( Deom et al . , 1986 ) . Overall , this gives the following set of antigenic residues , listed by sequential numbering of the WSN HA with the H3 number in parentheses: 171 ( 158 ) , 173 ( 160 ) , 175 ( 162 ) , 176 ( 163 ) , 178 ( 165 ) , 179 ( 166 ) , 180 ( 167 ) , 169 ( 156 ) , 172 ( 159 ) , 205 ( 192 ) , 206 ( 193 ) , 209 ( 196 ) , 211 ( 198 ) , 182 ( 169 ) , 186 ( 173 ) , 220 ( 207 ) , 253 ( 240 ) , 153 ( 140 ) , 156 ( 143 ) , 158 ( 145 ) , 237 ( 224 ) , 238 ( 225 ) , 87 ( 78 ) , 88 ( 79 ) , 90 ( 81 ) , 91 ( 82 ) , 92 ( 83 ) , and 135 ( 122 ) . A second classification is done that includes the Caton et al . ( 1982 ) plus any surface-exposed residues that are in contact with these residues , using an α-carbon to α-carbon distance of ≤6 . 0Å as the threshold for being in contact and classifying residues are solvent-exposed if they have an RSA of at least 20% . The rationale for this second classification is that the mapping by Caton et al . ( 1982 ) may have been biased towards inherently variable sites , and so other surface-exposed residues that contact these sites could also be antigenic . This classification adds the following 28 residues ( listed by sequential numbering of the WSN HA ) to the 29 Caton et al . ( 1982 ) residues: 85 , 86 , 89 , 126 , 132 , 136 , 137 , 138 , 142 , 148 , 150 , 154 , 155 , 157 , 170 , 184 , 185 , 187 , 202 , 203 , 207 , 210 , 212 , 221 , 235 , 236 , 239 , and 252 . | Influenza is a major threat to human health largely because the flu virus evolves rapidly to escape recognition by the immune system . These ongoing changes also explain why flu vaccines become less effective over time and need to be reformulated every year . Hemagglutinin is a protein on the surface of the flu virus that helps the virus bind to and infect host cells . The surface proteins of most viruses are recognized by the immune system , and influenza hemagglutinin is no exception . However , hemagglutinin is unusual in that it evolves exceptionally rapidly to avoid being recognized by the immune system . This raises an important question: what is it about the influenza hemagglutinin protein that allows it to change so readily ? Thyagarajan and Bloom address this question by making mutant copies of the gene that encodes the hemagglutinin protein . There are over 10 , 000 ways in which the protein can be mutated , and Thyagarajan and Bloom managed to make the vast majority of the possible changes . The mutated genes were then re-introduced into the virus , and the mutant viruses were allowed to replicate in cells for several generations . Thyagarajan and Bloom sequenced the viruses that had replicated—which meant that the mutant copies of the hemagglutinin protein in these viruses still worked—and looked to see where in the protein the changes had occurred . Those regions that rarely changed included the part of the protein that binds to host cells , whereas other regions—especially those that are recognized by the immune system—were much more likely to contain mutations . Thyagarajan and Bloom then went on to show that not all influenza proteins share hemaglutinin's capacity to change the regions targeted by the immune system , suggesting that this capacity is possibly a unique feature of this protein . Thyagarajan and Bloom also suggest that this capacity to tolerate mutations in parts of proteins that are recognized by the immune system might be important for shaping a virus's ability to evolve to escape this recognition . Future work is now needed to see how tolerant to mutations other viral proteins are , and to reveal which properties of a protein determine its tolerance to mutations . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"evolutionary",
"biology",
"microbiology",
"and",
"infectious",
"disease"
] | 2014 | The inherent mutational tolerance and antigenic evolvability of influenza hemagglutinin |
We previously discovered histones bound to cytosolic lipid droplets ( LDs ) ; here we show that this forms a cellular antibacterial defense system . Sequestered on droplets under normal conditions , in the presence of bacterial lipopolysaccharide ( LPS ) or lipoteichoic acid ( LTA ) , histones are released from the droplets and kill bacteria efficiently in vitro . Droplet-bound histones also function in vivo: when injected into Drosophila embryos lacking droplet-bound histones , bacteria grow rapidly . In contrast , bacteria injected into embryos with droplet-bound histones die . Embryos with droplet-bound histones displayed more than a fourfold survival advantage when challenged with four different bacterial species . Our data suggests that this intracellular antibacterial defense system may function in adult flies , and also potentially in mice .
Histones are fundamental components of eukaryotic chromatin , and are therefore abundant in essentially all animal cells , for example , in humans , there are tens of millions of histone molecules per cell . While not generally appreciated , histones and histone fragments are surprisingly bactericidal in in vitro assays ( Hirsch , 1958 ) . Thus , in principle histones could provide animals with a potent supply of microbicides for protection against bacterial invasion . In some cases , histones are released extracellularly , and then contribute to innate immunity against bacteria . For example , in the skin mucosa of catfish , secreted histone H2A contributes critically to the organismal defense against bacteria ( Cho et al . , 2002 ) , and histones H3 and H4 are prominent antimicrobial agents in sebaceous gland secretions ( Lee et al . , 2009 ) . Many bacterial pathogens invade cells and replicate intracellularly . This infectious strategy allows evasion of many host innate and adaptive immune mechanisms , including antibody-based defenses in vertebrates . In principle , then , histones normally present in cells could confer protection against microbes that exploit intracellular niches . However , until now this has seemed unlikely for two reasons . First , histones typically are located in the nucleus , excluding contact with cytosolic bacteria . Second , histones are generally believed to be predominantly bound to DNA , with only minuscule amounts of excess histones . In fact , because unconfined histones cause genomic instability , hypersensitivity to DNA-damaging agents , and lethality ( Saffarzadeh et al . , 2012 ) , cells minimize excess free histones by active mechanisms ( Singh et al . , 2009a , 2009b ) . Given these constraints , it is not surprising that histone-mediated intracellular antibacterial responses have not been investigated extensively . Lipid droplets ( LDs ) are ubiquitous fat storage organelles that store triglycerides and sterols for energy production and as biosynthetic precursors . In addition to their fundamental role in lipid homeostasis , LDs have recently been proposed to act as protein sequestration sites ( Cermelli et al . , 2006; Welte , 2007 ) . Histones , in particular , have been detected on LDs in a number of animal cells and tissues , including C . elegans , fly embryos , moth fat bodies , and mammalian leukocytes , insulin producing β-cells , and muscles ( Cermelli et al . , 2006; Wan et al . , 2007; Yang et al . , 2010; Zhang et al . , 2011; Larsson et al . , 2012; Zhang et al . , 2012 ) . Might these cytosolic extra-nuclear histone deposits allow cells to exploit the antimicrobial properties of histones without incurring the risks typically caused by excess histone accumulation ? We address this question using early Drosophila embryos . In these embryos , LDs sequester large amounts of histones ( Cermelli et al . , 2006 ) , and new work has provided mutants lacking droplet-bound histones ( Li et al . , 2012 ) . Here , we report that while histones are usually sequestered on droplets , they are released in response to the presence of bacterial cell wall components . These histones kill Gram-positive as well as Gram-negative bacteria , both in vitro and in vivo . Embryos injected with bacteria have considerably higher survival success when they have droplet-bound histones , establishing that histones on LDs contribute to innate immunity in embryos . A similar survival advantage was observed in adult flies . Simulation of bacterial infection in mice induces histone accumulation on LDs in the liver , a key organ for fighting infection; this observation suggests that lipid droplet histones may represent an ancient host defense strategy .
Our earlier study ( Cermelli et al . , 2006 ) established the presence of histones ( usually believed to be nuclear ) in cytosolic lipid droplets purified from Drosophila embryos . Based on our discovery of immunity-related mRNAs on the LDs ( unpublished ) , we considered whether histones might play a role in the flies' immune response . When we found that there was a precedent for histone antibacterial activity ( Hirsch , 1958 ) , we hypothesized that the histones bound to the cytosolic LDs might act as an antibacterial system . To test whether LDs could indeed inhibit bacterial growth , we performed a traditional plate assay ( Figure 1A ) . Dilute suspensions of bacteria were grown in the presence or absence of potential antimicrobial agents ( LDs equivalent to ∼500 μg total proteins and controls ) , and colony forming units ( CFU ) on an agar plate were counted after 24 hr of incubation with the agents . For both Gram-negative ( Escherichia coli ) and Gram-positive ( Staphylococcus epidermidis ) bacteria , LDs decreased the CFU dramatically ( Figure 1A , compare buffer vs LD ) , indicating that the droplets have an antimicrobial property . These effects were highly reproducible in multiple trials ( Figure 1B ) . Complementary disc-diffusion assays confirmed the microbicidal effects of the droplets ( Figure 1C , D ) . 10 . 7554/eLife . 00003 . 003Figure 1 . LDs kill bacteria via droplet bound histones . ( A ) . Representative plates in a colony forming assay , showing growth of Gram-negative ( Escherichia coli DH5α , top ) and Gram-positive ( Staphylococcus epidermidis , bottom ) bacteria , where a known amount of bacteria were incubated at 37°C either in buffer alone , or with LDs pre-treated with or without anti-histone antibodies . In buffer ( left , ‘Buffer’ ) , many colonies ( white spots ) were observed , but in the presence of LDs ( LD ) , the observed number of colonies was greatly decreased , demonstrating an antibacterial effect of the LDs . Pre-treatment of the droplets with anti-histone antibodies abolished this effect ( LD + Anti-histones ) . ( B ) . Quantification of colony forming assay in A . Each bar represents the mean number of observed colonies , in three independent trials , presented with the standard error . ( C ) . Disc diffusion assay over a lawn of E . coli DH5α . A potential antibacterial agent is placed on a small sterile piece of filter paper ( white circle ) ; a cleared area ( darker region ) indicates antibacterial activity . Positive control: the antibiotic kanamycin ( Antibiotic ) ; growth inhibition region indicated by the red arrow . Negative control: buffer ( Buffer ) . LDs isolated in the presence ( LD-CaCO3 ) or absence ( LD ) of alkaline carbonate were spotted on sterile discs; bacterial inhibition was observed in the untreated droplets ( LD ) but not in the carbonate-treated droplets ( LD-CaCO3 ) . The filter papers are 7 mm in diameter . ( D ) . Quantification of the size of the clear zone in the disc diffusion assay from C . Fifteen independent disc diffusion assays were performed with purified LDs from Drosophila embryos , the antibiotic kanamycin , or buffer . Antimicrobial activity of compounds was quantified as the diameter of the clear zones surrounding the filter papers after subtraction of filter papers diameter ( 1 A . U . = 0 . 1 mm ) . ( E ) . Use of a gel-overlay assay to determine the identity of the anti-bacterial protein ( s ) on the LDs . Proteins extracted from LDs ( LD , left lane ) were run in duplicate on an AU-gel; murine cryptdin 4 ( Crp 4 , right lane ) served as positive control . After electrophoresis , the gel was split . One half ( left ) was stained by Coomassie Blue; the histone bands are indicated by a blue arrow and the crp 4 control is indicated by the green arrow . The other half ( right ) was used in a gel overlay assay ( see ‘Materials and methods’ ) to reveal regions of the gel able to inhibit bacterial growth ( inhibition by LD is indicated by the red arrow and that by crp 4 control is indicated by the violet arrow ) . Inhibition of bacterial growth due to proteins on the LDs was only observed in a single region , corresponding to the histones ( red arrow ) . Consistent with this , mass spectrometry of proteins cut from the Coomassie gel corresponding to the killing region identified predominantly histones H2A and H2B ( see ‘LDs have antimicrobial activity’ ) . ( F ) . E . coli ML35 cultures were transiently ( ∼1 hr ) exposed to commercial calf thymus pan-histone proteins ( Sigma ) or gel-extracted LD-histones ( from the gel-overlay assay ) . Both preparations show similar potency for bacterial killing . DOI: http://dx . doi . org/10 . 7554/eLife . 00003 . 003 What is the molecular nature of this killing activity ? LDs are complex organelles with both lipids and proteins . Thus , one potential source of antibacterial activity might be fatty acids released due to breakdown of the abundant triglycerides; because of their detergent-like properties , they might destroy bacterial membranes and thus impair bacterial viability . We therefore treated purified LDs with alkaline washes . This treatment removes electrostatically bound droplet proteins , but not lipids and proteins attached via hydrophobic interactions ( Brasaemle et al . , 2004 ) . Indeed , as previously reported ( Cermelli et al . , 2006 ) , these washes greatly reduced the levels of the electrostatically bound histone H2B , but not of the hydrophobically bound LSD-2 . These treated droplets no longer had antibacterial activity ( Figure 1C , LD-CaCO3 ) , suggesting that the antimicrobial activity requires electrostatically bound proteins . Our previous study identified hundreds of lipid droplet proteins ( Cermelli et al . , 2006 ) , many present in low enough copy number to be detectable only by silver stain . In principle , any of them might be responsible for the antibacterial activity , but it seems likely that the active agents would be present in high copy number . We thus performed a traditional gel overlay assay ( Figure 1E ) , visualizing these high-copy number proteins by Coomassie stain . Droplet proteins were separated by Acid Urea ( AU ) gel electrophoresis , and the gel was overlain with nutrient agar seeded with bacteria . As in the disc assay , antibacterial proteins diffuse into the agar and locally inhibit bacterial growth , as shown by the positive control mouse α-defensin cryptdin-4 ( Figure 1E , gel overlay violet arrow ) . The lipid-droplet lane contained significant antibacterial activity in only one location ( Figure 1E , gel overlay red arrow ) . MS analyses of proteins eluted from this location identified histones H2A and H2B as the predominant proteins present ( spectra not shown ) . Therefore , the bactericidal activity of LDs is most likely due to histones , consistent with their known antimicrobial activities in vitro ( Hirsch , 1958 ) . Importantly , pretreatment of purified LDs with anti-histone antibodies abolished or markedly reduced the droplet bactericidal activity ( Figure 1A , 1B-LD + Anti-histones ) . Thus , we conclude that the majority of the in vitro antibacterial activity of LDs from early Drosophila embryos is due to histones . We performed two tests to determine if these cytosolic droplet-bound histones are different from nuclear histones . First , using mass spectrometry , we compared post-translational modifications on droplet-bound and nuclear histones and found no major distinctions , suggesting that unique post translation modifications are not responsible for association of histones to LDs . Since our analysis was mostly qualitative , we cannot rule out the possibility that differences might arise from quantitative changes in post-translation modifications . We did identify several acetylation sites in histones H2A ( serine 1 and lysines 5 and 8 ) and histone H2B ( lysines 7 , 11 , 14 and 17 ) . Some of these acetylation sites were previously found in shrimp histones ( Ouvry-Patat and Schey , 2007 ) . The histones previously examined for antimicrobial activity are reported to be un-acetylated ( Kim et al . , 2000 ) , though there was no indication of whether acetylation affected antimicrobial efficacy . Second , we compared the antibacterial potencies of droplet-derived histones with commercial calf thymus histones isolated from a calf thymus nuclear fraction in both bactericidal ( killing bacteria outright ) and bacteriostatic ( inhibiting the growth or reproduction of bacteria ) activity assays . Histones extracted from droplets using AU gel electrophoresis were combined with a suspension of E . coli ML35 for 1 hr , and bacterial cell survival was determined by measuring CFU . This assay showed that histones are bactericidal and that droplet-derived histones and commercial pan-histones purified from calf thymus do not differ significantly in antibacterial efficacy ( Figure 1F ) . Although these studies indicate that histones present on embryonic LDs kill bacteria in vitro , it remained unclear whether histones make a meaningful contribution to the overall antibacterial defense in the embryos . In vivo , droplet-bound histones may have different properties , due to the presence of binding partners or the physiological state of the bacteria , their effective concentration might not be high enough to kill , or , relative to other antibacterial mechanisms , the contribution of histones might be negligible . To test the significance of the histones on LDs in vivo , we took advantage of the recent identification of the putative histone receptor on droplets , jabba ( CG42351 ) , a novel 42 kDa protein ( Li et al . , 2012 ) . Jabba is present on LDs in wild-type flies and is required for histone localization on droplets; in Jabba mutants LDs are present , but histones are absent from the LDs ( Figure 2 ) . We used two Jabba alleles , derived independently . Jabbazl01 is due to imprecise excision of a P element inserted near the Jabba promoter , and Jabbaf07560 is due to the insertion of a PBac element in the middle of the Jabba coding region . The two alleles were generated from entirely different genetic backgrounds , ruling out genetic background effects . In the wild type , fusions between GFP and the histone H2Av are present both in nuclei and in cytoplasmic rings ( Figure 2A ) , a characteristic appearance of droplet-targeted proteins ( Cermelli et al . , 2006 ) , but cytoplasmic rings are undetectable in Jabba embryos . Second , when living embryos are centrifuged , LDs separate from the rest of the embryonic content and form a distinct layer ( Cermelli et al . , 2006 ) . Immunostaining of such centrifuged embryos reveals abundant histone signal in the droplet layer in the wild type , but not in Jabba embryos ( Figure 2B , C ) . Third , Western blotting reveals high levels of histones on droplets purified from wild-type embryos , but not on droplets purified from Jabba mutants ( Figure 2D , E ) . Such purified droplets also differed in their antimicrobial activity in vitro: droplets from wild-type embryos resulted in a 10-fold decrease in bacterial growth relative to buffer alone , but droplets from two different independently isolated Jabba mutant strains displayed essentially no killing activity ( Figure 2G ) , as expected if the killing activity of wild-type droplets is indeed due to histones . 10 . 7554/eLife . 00003 . 004Figure 2 . Presence of extranuclear histones depends on the Jabba protein . ( A ) Histone H2Av GFP is not detectable in cytoplasmic puncta of Jabbazl01 embryos . Both genotypes show strong signal in nuclei . ( B ) . By immunostaining , endogenous H2A and H2B are absent from the lipid droplet layer ( LD ) of centrifuged Jabbazl01 embryos . ( C ) . Histone H2A is absent from the lipid droplet layer in centrifuged Jabbaf07560 embryos . BF is the bright field image and LD is the lipid droplet layer . ( D ) . Equal amounts of proteins from purified LDs were compared by Western analysis . Droplets from Jabbazl01 embryos lack histones H2A and H2B . The droplet-bound Khc protein serves as loading control . ( E ) . When compared side by side , similar reductions in droplet-bound histones were found for both the independently isolated Jabba alleles Jabbaf07560 and Jabbazlo1 . ( F ) . Western blot of equal numbers of unfertilized wild-type and Jabba mutant embryos . Overall levels of histone H2A and H2B are significantly reduced in the Jabba mutants . ( G ) . LDs purified from embryos of two independently isolated Jabba mutants revealed no bacterial killing activity in antibacterial plate assays , with bacterial growth comparable to buffer alone , in contrast to droplets purified from wild-type embryos which dramatically decreased bacterial growth . DOI: http://dx . doi . org/10 . 7554/eLife . 00003 . 004 In Jabba mutant embryos , the overall levels of histones are much reduced relative to the wild type ( Figure 2F ) , presumably because the histones not sequestered on the droplets are degraded . At least in other systems , unconfined histones are rapidly eliminated via proteolysis ( Singh et al . , 2009a , 2009b ) , a protective mechanism against the detrimental effects of free histones . Jabba mutants are viable and fertile and develop into apparently healthy adults ( though likely with compromised immune systems , see Figure 4 ) . In particular , embryos hatch at wild-type rates . These mutants therefore make it possible to ask if the extra-nuclear pool of histones affects the outcome of bacterial infections in vivo . We microinjected ( Figure 3A ) very early wild-type or Jabba embryos ( less than 1-hr old ) with a GFP-expressing E . coli strain; living bacteria are easily identified by GFP fluorescence; fluorescence fades after bacterial cell death ( Lowder et al . , 2000 ) . Monitoring overall GFP fluorescence provided a measure of live bacteria in the infected embryos . The injection protocol per se does not apparently harm either genotype , since buffer-only injection resulted in high and similar hatching success ( Figure 3B ) . 10 . 7554/eLife . 00003 . 005Figure 3 . LD bound histones can kill bacteria in vivo . ( A ) Schematic representation of embryo microinjection . Early embryos collected within half an hour of laying were injected with a bacterial suspension , as detailed in ‘Materials and methods’ . ( B ) . Wild-type and Jabba mutant embryos show similar survival when injected with buffer alone . Wild-type and Jabba mutants ( Jabbaf07560 , Jabbazl01 ) embryos were injected with microinjection buffer ( no bacteria ) and the percentage survival was scored 72 hr post injection . ( C ) Bacteria grow only in embryos lacking droplet-bound histones . Approximately equal numbers of GFP labeled bacteria ( E . coli strain YD133 ) were injected into wild-type and Jabba mutant embryos ( Jabbaf07560 ) and the growth of bacteria inside embryos was monitored at various times post injection . ( D ) . Drosophila embryos lacking droplet-bound histones have reduced survival due to bacterial infection . Approximately equal numbers of bacteria were injected into wild-type and Jabba mutant ( Jabbazl0 and Jabbaf07560 ) embryos and embryo survival after 72 hr was normalized to the buffer-only injected embryos ( in B ) . The bacterial strains used were Staphylococcus epidermidis ( Gram-positive ) ; E . coli DH 5α ( Gram- negative ) ; Listeria monocytogenes ( Gram-positive and intracellular ) ; and Bacillus subtilis ( hlyA ) , modified Bacillus subtilis expressing listeria hemolysin-A protein ( Gram-positive and intracelluar ) . DOI: http://dx . doi . org/10 . 7554/eLife . 00003 . 00510 . 7554/eLife . 00003 . 006Figure 3—figure supplement 1 . 2 hr: Additional images of wild-type and Jabba mutant embryos with fluorescent bacteria , 2 hr after bacterial injection . DOI: http://dx . doi . org/10 . 7554/eLife . 00003 . 00610 . 7554/eLife . 00003 . 007Figure 3—figure supplement 2 . 24 hr: Additional images of wild-type and Jabba mutant embryos with fluorescent bacteria , 24 hr after bacterial injection . DOI: http://dx . doi . org/10 . 7554/eLife . 00003 . 00710 . 7554/eLife . 00003 . 008Figure 3—figure supplement 3 . 48 hr: Additional images of wild-type and Jabba mutant embryos with fluorescent bacteria , 48 hr after bacterial injection . DOI: http://dx . doi . org/10 . 7554/eLife . 00003 . 008 Under the conditions employed , bacterial numbers in wild-type embryos decreased with time ( Figure 3C ) . 2 hr after injection , numerous individual bacteria were obvious in the embryos , and we estimate on the order of 84 bacteria per embryo ( see ‘Materials and methods’ ) . By 24 hr , the number of visible bacteria had decreased substantially , and we estimate on the order of 33 bacteria per embryo . By 48 hr , there typically were either no surviving bacteria or only a few detectable bacterial cells , and on average we estimate 2–6 bacteria were present . Thus , some innate immune mediator ( s ) limits bacterial cell viability in wild-type embryos in this experimental system . This limit on bacterial viability was lost in the Jabba embryos ( Figure 3C ) . The appearance of multiple individual bacteria 2 hr post-infection was similar to the wild type , and we estimate approximately 79 bacteria per embryo . However , bacterial cell numbers increased dramatically by 24 hr , with 840 bacteria on average . By 48 hr , we estimate that thousands of bacteria were present in the Jabba embryos ( Figure 3C , compare WT to Jabbaf07560 at 24 and 48 hr; also see Figure 3—figure supplements 1–3 ) . These results indicate that loss of histones on droplets due to the absence of functional Jabba correlates with susceptibility to massive bacterial overgrowth . Because bacterial numbers are controlled , and bacteria are ultimately eliminated in wild-type embryos , it suggests that droplet-associated histones contribute to immunity against the introduced bacteria , consistent with their in vitro bactericidal capability . It seemed likely that the immunity observed in this experimental bacterial infection would have consequences for the embryo . To test this , we challenged wild-type and Jabba mutant embryos with different bacterial species ( Figure 3A ) and assessed the effects of genotype on survival of the embryos ( Figure 3D ) . We first injected embryos with two laboratory strains of bacteria: Gram-negative E . coli DH5α and Gram-positive S . epidermidis . In each case , experiments were done in parallel , with 50–100 bacteria being injected into multiple embryos of each genotype . Embryos that hatched into larvae were scored as surviving . Injection of bacteria delayed wild-type embryonic development , but caused only a minimal increase in lethality , relative to buffer-only injected embryos ( normalized survival more than 80%; see Figure 3D , S . epidermidis and E . coli DH5α , orange bars ) . In contrast , for the two independently derived Jabba mutants ( both lacking droplet-bound histones; Figure 2D ) , the same treatment resulted in high lethality , with normalized survival of less than 20% ( Figure 3D ) . Thus , Jabba embryos exhibited at least a fourfold decrease in survival when injected with either species of bacteria . Such a difference in survival would provide a huge survival advantage in nature . E . coli and S . epidermidis do not typically grow intracellularly , but bacterial pathogens that grow intracellularly are not well characterized in flies . We therefore took advantage of two species of bacteria with well-characterized intracellular mechanisms of infection , Bacillus subtilis engineered to express Listeria's hemolysin-A protein ( Bielecki et al . , 1990 ) , and thus able to enter cells and reproduce in the cytosol , and also Listeria monocytogenes whose infectious life cycle typically involves growth in the cytosol of mammalian cells ( Tilney and Portnoy , 1989 ) . At moderate injection dosages of 50–100 bacteria per embryo , there was again good survival for wild-type embryos , but not for the Jabba embryos ( Figure 3D , B . subtilis and L . monocytogenes ) . In conclusion , a marked survival difference between the wild-type ( more survival ) and Jabba mutant embryos was observed when infected with bacteria , for all of the bacterial species tested . Could Jabba function in adults ? Consistent with a possible role in facilitating stable localization of histones to the cytoplasm , Jabba is expressed in a variety of adult tissues and in both sexes , with especially high expression levels in fat body and ovaries , according to microarray and RNA-seq data available on FlyBase ( Chintapalli et al . , 2007; McQuilton et al . , 2012 ) . To test whether this Jabba protein present in the adults might contribute to a similar LD-histone system , we used a traditional bacterial challenge assay , where bacteria were introduced into adult flies by pricking the flies under the wing with a metallic needle dipped in either sterile buffer or a concentrated bacterial suspension . Pricking either the wild type ( black curve , Figure 4A , B ) or Jabba-mutant adults ( red curves , Figure 4A , B ) with the buffer-dipped needle resulted in low long-term mortality , with roughly a 20–30% mortality at 4 days . At the dose of Listeria used , mortality of the pricked wild-type adults was approximately the same as the buffer-pricked adults ( the purple mortality curve in Figure 4A is within experimental error of the black curve ) . However , for the Jabba-mutant adults , pricking with a bacterial-dipped needle was quite lethal ( Figure 4A , B , brown curve ) , with less than 5% survival at 4 days—a 14-fold difference from wild-type survival . 10 . 7554/eLife . 00003 . 009Figure 4 . The Jabba protein contributes to improved survival for adult flies . ( A and B ) Adult Drosophila lacking Jabba ( A: Jabbaf07560; B: Jabbazl01 ) have reduced survival when challenged by bacteria . Wild-type and Jabba mutant ( Jabbazl0 and Jabbaf07560 ) adult flies were infected with Listeria monocytogenes as detailed in ‘Materials and methods’ , and fly survival was monitored over the course of 4 days . ( C ) . Representative plates in a colony forming assay , showing bacterial colonies on agar plates streaked with cytosolic extract from bacteria infected adult flies . ( D ) Western blots of histone H2B from equal amounts of cytosolic extracts from wild type and Jabba mutant adult flies , showing that overall levels of H2B were significantly reduced in the Jabba mutants . DOI: http://dx . doi . org/10 . 7554/eLife . 00003 . 009 How likely is it that the underpinnings of the survival difference reflect the same mechanism ? We first examined relative bacterial load via a plate assay using cytoplasmic extract from the buffer or bacterial-pricked adults to seed the plate . From wild-type or Jabba-mutant adult buffer-pricked cytoplasm , typically less than three colonies were observed . In contrast , for the bacterial-pricked adults , initial counts were typically on the order of 400 CFUs ( Figure 4C ) , and by day 3 went down significantly for the wild type ( 50 ) but less so for the surviving Jabba-mutant flies ( 320 ) . Presumably , the Jabba-mutant flies that died ( not assayed ) had even higher bacterial counts . While a complete investigation of the kinetics of bacterial load is beyond the scope of this paper , as in the embryos , these results suggest that the observed lethality correlates with differences in bacterial load . Finally , we looked for the presence of histones in the adult cytoplasm . First , cytoplasmic lysates were made from 1- to 2-day-old adult wild-type or Jabba-mutant flies as detailed in ‘Materials and methods’ , and then equal amounts of the lysates from both classes of adults were blotted to detect histone H2B ( Figure 4D ) . Compared to the wild-type , the amount of H2B detected was lower in the Jabba-mutant background ( threefold ) , consistent with the embryo data . In conclusion , while more work remains to understand the role of Jabba and histones in adult immunity , our initial data is consistent with the hypothesis that the embryonic system described above may function in adult flies as well . Since excess free histones are deleterious for the cell overall ( Gunjan and Verreault , 2003 ) , droplet-bound histones are likely relatively immobilized: we expected them to be sequestered on droplets and not free to diffuse . Indeed , when purified droplets are incubated in excess buffer , there is no detectable loss of histones from the droplets , or appearance of histones in the buffer ( Figure 5A , B , UB control ) . Thus , the histones indeed appear to be stably bound to LDs . This might limit their ability to reach the bacteria since the diffusion constant of a 0 . 5-µm droplet is expected to be much lower than the diffusion constant of a free histone . 10 . 7554/eLife . 00003 . 010Figure 5 . Bacterial cell wall components release droplet bounds histones in a dose dependent manner . ( A ) . Increasing concentrations of lipopolysaccharide ( LPS ) in the buffer releases droplet bound histones from purified LDs . ( B ) . Lipoteichoic acid ( LTA ) causes the dose dependent release of histones from purified droplets . LDs were purified from wild-type Drosophila embryos , re-suspended in buffer , and incubated for 2 hr at room temperature with different concentrations of LPS or LTA . LDs ( LD ) were then separated from the under-layer buffer ( UB ) and both were processed for SDS-PAGE . Western Blot analysis was carried out with H2B histone antibodies . DOI: http://dx . doi . org/10 . 7554/eLife . 00003 . 010 These observations are seemingly contradictory: histones are stably bound to droplets , yet they can kill bacteria well . We thus hypothesized that the bacteria may induce release of the histones from the droplets . Pathogen-Associated Molecular Patterns , components of the bacterial envelope , would be particularly well positioned to induce such a release , as they are present on the surface of bacteria and thus accessible . Indeed , organisms often detect bacterial infections due to the presence of LPS ( Heumann and Roger , 2002 ) or LTA ( Wergeland et al . , 1989 ) , major pro-inflammatory constituents of Gram-negative and Gram-positive bacterial cell envelopes , respectively . We therefore incubated purified LDs in the presence or absence of LPS or LTA . Histones were detected in the buffer ( UB , Figure 5A , B ) only when LPS or LTA were included , and histone amounts increased with increasing levels of the cell envelope components ( Figure 5A , B ) ; concomitantly , histones attached to the LDs decreased ( LD , Figure 5A , B ) . Thus , LPS and LTA induce release of histones from the droplets in a dose-dependent manner . Histones on LDs are not restricted to Drosophila . In particular , specific histones have been identified on LDs purified from a number of mammalian cell lines and tissues ( Smolenski et al . , 2007; Wan et al . , 2007; Zhang et al . , 2011; Larsson et al . , 2012 ) . Thus , this defense system may be widely conserved . As a preliminary test , we looked at droplets in the liver , as this organ removes pathogens and microbial products from the blood , and plays a key role in the body's immune response ( Mackay , 2002 ) . LDs were purified from murine liver using a previously established protocol ( Turro et al . , 2006 ) ; the hepatocyte lipid-droplet resident protein ( Turro et al . , 2006 ) ALDI was enriched 103-fold ( Figure 6A ) , confirming the success of the fractionation . By Western blotting , we detected histone H1 in the droplet fraction , using three different specific antibodies generated in different species ( Figure 6B ) . This does not represent contamination with nuclei or chromatin since histones H2A , H2B and H3 were not detectable ( Figure 6A ) . Normalized for total protein , the H1 levels on LDs were comparable to the levels of H1 histone in a purified nuclear fraction ( Figure 6B—n ) , and were clearly much higher than in whole-liver homogenates ( Figure 6B—h ) . Like other histones , histone H1 also has antibacterial activity in vitro ( Parseghian and Luhrs , 2006 ) . 10 . 7554/eLife . 00003 . 011Figure 6 . Histones are on mammalian LDs and respond to LPS . ( A ) . Western blot analysis of LDs ( LD ) purified from hepatocytes of mice injected with ( + ) or without ( − ) LPS . Antibodies against ALDI , histones H2A , H2B and H3 were used . Whole liver homogenate ( h ) was used for comparison and as a control . ( B ) The presence of histone H1 ( H1 ) on LDs ( LD ) purified from hepatocytes of mice injected with ( + ) or without ( − ) LPS was detected by immunoblot , and more H1 was present on droplets purified from LPS-treated animals . Equal total proteins from the nuclear fraction ( n ) and from whole liver homogenate ( h ) were used as comparison . ( C ) . Mice were injected intraperitoneally with ( + ) or without ( − ) LPS , and transaminase levels ( AST and ALT ) and cytokine levels ( IL-6 and TNFα ) were quantified in units/l or units/ml in the serum; asterisk indicates statistical significance ( p=0 . 05 ) , confirming that LPS injection provoked the expected biological response . ( D ) . Western blot analysis of histone H1 released into the buffer ( UB ) when purified LDs , from the liver of infection induced mice , were treated with LPS ( + ) . Histone H1 is either minimally detected or not at all detected in the buffer in the absence of LPS ( − ) . The band at 97 kDa in C and D represents histone H1 oligomers . DOI: http://dx . doi . org/10 . 7554/eLife . 00003 . 011 Intriguingly , the levels of histone H1 on LDs increased under conditions that mimicked a systemic infection . Mice were intraperitoneally injected with LPS ( 20 ng ) , and assayed 16 hr later . LPS injection resulted in the expected hepatic injury ( Senga et al . , 2008 ) , as assessed by the increased presence of serum transaminases ( AST and ALT ) and cytokines ( IL-6 and TNFα; Figure 6C ) . Relative to untreated controls , H1 levels in the lipid-droplet fraction were clearly increased in the LPS challenged animals ( average ratio of H1 in LPS injected to H1 in uninjected was 1 . 6 ± 0 . 3 ( mean ± SEM ) , significantly different with a p=0 . 025 by Student t-test; note that the amount of the lipid droplet marker protein ALDI went down slightly in the LPS-injected samples , and if we use this as a standard , and normalize the detected ratio , it becomes 1 . 86 instead of 1 . 6 ) . Like the Drosophila histones , H1 was efficiently released from the LDs by LPS ( Figure 6D ) , consistent with the hypothesis that the accumulated H1 on the hepatic LDs can be released by the presence of cytosolic bacteria , and thus might contribute to an antibacterial response . These two observations—that systemic LPS increases histones on murine droplets , but that direct treatment of purified droplets with LPS releases the histones—might appear contradictory , but we believe they are not . In the first case , the systemic LPS is known to activate numerous defense pathways which could then result in increased loading of droplets with histones . Further , lacking cytosolic bacteria , the systemic circulating LPS is unlikely to be present inside the cells at high enough dosage to release the histones from the droplets . In the second case , purified droplets with histones are exposed directly to LPS , mimicking the presence of bacteria inside of cells . In this latter case , there are no signaling effects , because in the purified system no signaling apparatus is present . Thus , this second assay looks only at the direct effect of high levels of LPS interacting with histones , causing the histones to release from the droplets .
Here , we suggest that LDs contribute to host defense by sequestration and regulated release of histones . Analogous to mammalian mothers depositing antibodies across the placenta into the embryo as it develops , we suspect that maternal deposition of the histone laden lipid droplets into the embryo may prevent transmission of any bacterial infection from mother to egg , and may also protect the embryo from subsequent bacterial infections , resulting , for example , from physical damage to the chorion . The histones are stably bound to Drosophila and murine LDs , but bacterial cell envelope components promote their release . As tested in the embryos , and in a limited way in adult flies , the system appears efficient , resulting in a more than fourfold improvement in survival for challenges by multiple types of Gram-positive and Gram-negative bacteria , including the known intracellular pathogen Listeria . Histones and histone fragments had previously been reported to play a role in extracellular antibacterial defense ( Fernandes et al . , 2004; von Kockritz-Blickwede and Nizet , 2009 ) . We propose here that by using histone-droplet sequestration , cells can position histones in the cytosol , so that they are both available to interact with cytosolic bacteria , and also can evade constitutive pathways responsible for degrading free histones . Determining whether the murine droplet-bound histones actively contribute to host defense remains for future studies . In principle , any kind of cell has the ability to form LDs , and thus employing droplet-bound histones as antibacterial defence might be quite general . Indeed , proteomic studies have revealed histones on LDs in tissues from larval and adult insects as well as in a range of mammalian cells ( Smolenski et al . , 2007; Wan et al . , 2007; Yang et al . , 2010; Zhang et al . , 2011; Larsson et al . , 2012 ) , including mouse liver ( Figure 6 ) . Such a role of droplets in the antibacterial response may in part explain why the number and size of LDs increase in disease , from osteoarthritis to liver degeneration and cartilage overproliferation ( Tilney and Portnoy , 1989; Bielecki et al . , 1990; Lowder et al . , 2000; Gunjan and Verreault , 2003; Figueredo et al . , 2009; Yang et al . , 2010 ) . In mammalian leukocytes and macrophages , LDs have already been suggested to participate in the regulation of the host response to infection , by modulating the production of inflammation mediators like eicosanoids ( Bozza and Viola , 2010 ) . Our result discloses an additional role of droplets in innate immunity , a role potentially conserved from flies to mammals . Since many medically relevant bacterial pathogens enter the cytosol ( e . g . Listeria monocytogenes , Shigella flexneri , Burkholderia pseudomallei , Francisella tularensis and Rickettsia spp; Ray et al . , 2009 ) and histones are also reported to have anti-fungal properties ( De Lucca et al . , 2011 ) , and many such pathogenic fungi also enter the cytosol ( e . g . , Cryptococcus neoformans; Bliska and Casadevall , 2009 ) , it may be that the system described here has surprisingly wide utility . At least in the case of bacteria , histone release from the droplets is triggered by bacterial cell envelope components . This release likely allows an appropriate cellular response: freeing enough histones to kill bacteria , but minimizing problems ( Gunjan and Verreault , 2003 ) of excess free histones interfering with endogenous cellular processes . We speculate that this release is achieved via direct binding between LPS and histones , as histones can indeed directly bind LPS ( Bolton and Perry , 1997; Augusto et al . , 2003 ) and the presence of LPS did not significantly affect the antibacterial activity of histones in the plate assay . We propose that the negatively charged LPS or LTA neutralize the positively charged histones and therefore weaken their electrostatic interactions with Jabba on the LDs; histone binding to droplets is charge-sensitive , as indicated by the ability of CaCO3 to detach the histones from the droplets . If histones are so effective , why load them not at high levels onto all LDs ? The observation that droplet-bound H1 increases in mouse liver in response to a simulated bacterial infection might suggest that there is a cost to storing histones on droplets . Cells likely have to balance immediate histone availability with possible undesirable uncontrolled histone release from droplets due to metabolic consumption of the underlying droplets , or alternatively , potential secondary effects due to the excess histones causing saturation of histone-modifying enzymes ( Singh et al . , 2010 ) .
Drosophila embryos ( Oregon-R strain ) were collected , aged and dechorionated with 50% bleach . LDs were purified as previously described ( Cermelli et al . , 2006 ) . Briefly , LDs were isolated from total embryo lysates by sucrose gradient ultracentrifugation and solubilized in NP40 lysis buffer ( 10 mM Tris–HCl , pH 7 . 4 , 0 . 5 mM EDTA , 1% NP40 ) . Protein concentration was determined by Bradford dye-binding assay ( Sigma-Aldrich , MO , USA ) . In some cases , the isolation of LDs was performed in the presence of 100 mM sodium/calcium carbonate , according to Brasaemle et al . , ( 2004 ) , to remove proteins bound via electrostatic interactions . The presence of proteins histones H2A and H2B , and kinesin heavy chain in droplets was monitored using immunoblot analysis with anti-H2A ( Leach et al . , 2000 ) , anti-H2B ( Upstate Biotechnologies , Lake Placid , NY , USA ) , and anti-Khc ( Cytoskeleton , CO , USA ) , respectively . Proteins were separated by SDS-PAGE and electro transferred to PVDF/nitrocellulose membranes . The membranes were blocked with 5% non-fat milk or 5% bovine serum albumin for an hour and then incubated with appropriate primary antibodies at desired concentrations for 1 hr at room temperature or over night at 4°C . Peroxidase conjugated donkey anti-rabbit/goat anti-mouse ( 1:10 , 000; Jackson ImmunoReseach ) were used as secondary antibodies and the signals were monitored by Novex ECL kit ( Invitrogen , CA , USA ) . Embryos were centrifuged , heat-fixed , and stained with anti H2A and H2B antibodies ( 1:2000 ) as described ( Cermelli et al . , 2006 ) . H2Av-GFP expressing embryos were imaged live without any fixation . Micrographs were acquired on a Leica SP5 confocal microscope or a Nikon Eclipse E600 fluorescence microscope with a 4 MP Spot Insight camera . Images were processed in Adobe Photoshop and assembled with Adobe Illustrator . In order to visualize the GFP bacteria inside microinjected embryos , a LSM710 confocal microscope was used . To evaluate the antibacterial property of purified LDs/LD components , we performed colony forming assays , gel overlay assays as well as disc diffusion assays . Duplicate samples of proteins from purified LDs were separated by AU-gel electrophoresis . One lane was stained by Coomassie Blue and used to identify the approximate position of histones in the unstained lane . This region of the unstained lane was cut out , macerated , and incubated overnight in 5% acetic acid . The solution with extracted proteins was lyophilized , and the pellet resuspended in 1% TSB . LDs proteins were separated by 1-D SDS-PAGE and stained by Coomassie Blue . Proteins bands were excised , in-gel digested by trypsin , extracted , and concentrated for LC-MS/MS analysis as described ( Huang et al . , 2001 ) . The tryptic digests were analyzed by LC-MS/MS using a nanoLC system ( Eksigent , Inc . , MA , USA ) coupled with Linear Ion Trap ( LTQ ) -Orbitrap XL mass spectrometer ( Thermo-Electron Corp , OH , USA ) . The LC analysis was performed using a capillary column ( 100 μm i . d . × 150 mm length ) packed with C18 resins ( GL Sciences , CA , USA ) and the peptides were eluted using a linear gradient of 2–35% B in 105 min; ( solvent A , 100% H2O/0 . 1% formic acid; solvent B , 100% acetonitrile/0 . 1% formic acid ) . A cycle of one full FT scan mass spectrum ( 350–1800 m/z , resolution of 60 , 000 at m/z 400 ) was followed by 10 data-dependent MS/MS acquired in the linear ion trap with normalized collision energy ( setting of 35% ) . Target ions selected for MS/MS were dynamically excluded for 30 s . We used Oregon-R as our wild-type stock . We employed two mutant alleles of Jabba ( also known as CG42351 ) : Jabbazl01 is a promoter deletion and expresses no Jabba protein in early embryos; Jabbaf07560 is a transposable element insertion between two coding exons of Jabba , resulting in a severely truncated Jabba protein . A comprehensive molecular and phenotypic description of Jabba mutant alleles will be published elsewhere ( Li et al . , 2012 ) . Alleles Jabbazl01 and Jabbaf07560 were derived independently , and thus likely share few , if any , unknown secondary mutations . Both alleles eliminate droplet-bound histones ( Figure 3D , Figure 3—figure supplements 1–3 ) . Cultures of the bacterial strain of interest ( E . coli DH5α , S . epidermidis , L . monocytogenes , B . subtilis [hly] and YD133-GFP; an E . coli K-12 bacterial strain expressing GFP ) were grown to the log phase . An appropriate volume of the bacterial suspension was pelletted , washed with PBS and re-suspended in injection buffer ( 5 mM KCl , 0 . 1 mM sodium phosphate pH 6 . 8 , 5% [vol/vol] of McCormick green food color ) . Precellular blastoderm stage embryos were injected manually using a Narishige IM300 microinjector . Injected embryos were maintained at 25°C in a fly incubator and the survival was monitored daily until they hatched into larvae . Percentage survival was normalized with respect to the survival of embryos injected with buffer only . An estimate of the bacterial load in the GFP-bacteria injected cases was done by image analysis . We searched in each embryo for the region ( s ) with the most bacteria visible , and quantified that number , using images with same field of view . At least three different embryos were used in each group . At the same time , we noted how many such high-bacteria fields were typically present in an embryo , as well as typical bacterial counts in the other fields . We estimate the embryo can be covered by approximately 10 independent fields of 40 µ each . For the wild-type , at t = 2 hr , the average number of bacteria in the ( maximal ) field of view was 24 . 3 ± 6 . 4 , and there were between three and four such fields per embryo , with few bacteria elsewhere , leading to an estimate of ∼84 bacteria per embryo . For the Jabba mutant embryo , the average number of bacteria in the ( maximal ) field of view was 23 . 3 ± 3 . 2 , and there were between three and four such fields per embryo , with few bacteria elsewhere , leading to an estimate of ∼79 bacteria per embryo . For the t = 24 wt embryos , the average maximal field had 21 . 7 ± 9 . 2 bacteria , but there was only one such field per embryo , typically with 3–4 other areas with 1–5 bacteria per field , resulting in an estimate of 33 bacteria per embryo . For the t = 24 Jabba mutant embryos , approximately 1/3 of the fields had the high bacterial count ( ∼180 ± 16 . 1 bacteria ) , and the rest typically had 10–30 bacteria per field , leading to an estimate of 840 bacteria per embryo . Finally , for the t = 48 hr wild-type , at most only one or two fields had any detectable bacteria ( some embryos had none ) so we estimate between 2–6 bacteria per embryo . In contrast , for the t = 48 Jabba embryo , every field was full of bacteria ( 200–250 bacteria per field ) , so assuming 10 such fields we estimate 2000–2500 bacteria per embryo . Adult flies ( both wild type and Jabba mutant , 2–3 days old ) were subjected to an infection assay . The flies were anesthetized with CO2 , and pricked under the wing with a fine metallic needle ( 33G ) which was first dipped into either buffer or a suspension of the bacteria Listeria monocytogene . To make the bacterial suspension , a dilute overnight culture was grown to an absorbance of 0 . 5–0 . 6 at wavelength 595 nm . From this log-phase bacterial suspension , 5 ml was pelletted and the pellet washed with PBS . This bacterial pellet was then suspended in 200 μl of PBS and the metallic needle was dipped in for the bacterial infection assays . After the infection , the flies were maintained in 25°C fly incubator . Every 24 hr , the number of dead and live flies was counted . The percentage of survival calculated from these counts was then plotted against days post injection as shown in Figure 4 . Approximately 10 live flies from the both wild type and Jabba mutant at days 0 and 3 were sacrificed and cytosolic extracts were prepared using ready prep protein extraction kit ( Bio-Rad Laboratories , Inc , CA , USA ) according to the manufacturers' protocol . Equal amounts of cytosolic extracts were then plated on agar plates to estimate the relative viable cytosolic bacterial load as shown in Figure 4C . The quantification of band intensity in the Figure 4C was done using Image J software . C57BL/6 mice were kept under a controlled humidity and lighting schedule with a 12-hr dark period . All animals received human care in compliance with institutional guidelines regulated by the European Community . Food and water were available ad libitum . Male mice of approximately 8–12 weeks old were intraperitoneally injected with 20 ng LPS diluted in NaCl 0 . 9% or with an equivalent volume of NaCl and fasted for 16 hr . Hepatic LDs were isolated exactly as described in Turro et al . , ( 2006 ) ( Wan et al . , 2007 ) . The protein fraction of LDs was separated by precipitation with 33% of cold TCA and the resulting pellets resuspended in 50 μl of 250 mM Tris , 2% SDS . The protein content was determined by the method of Lowry , and equal amount of protein of LDs was separated by SDS-PAGE . Western blotting was performed as described previously . In some experiments , the LDs obtained from LPS-treated mice were diluted 1:3 in a 250-mM Tris buffer and incubated with 1 mg/ml of LPS or an equivalent volume of NaCl for 2 hr at 4°C . A soluble fraction of histones and other proteins was separated from the intact LDs by centrifugation at 16 , 000×g for 15 min in an Eppendorf microfuge . Finally , equal volumes of soluble fractions were separated by SDS-PAGE and the presence of H1 determined by Western blotting . The primary antibodies used in immunoblot analysis of mouse samples were anti-histone H1 , clone AE-4 from UPSTATE biotechnology ( Millipore , MA , USA ) , Anti histone H1: sc-10806 , anti histone H3 ( FL-136 ) : sc-10809 and anti histone H2B ( FL-126 ) : sc-10808 ( from Santa Cruz Biotechnology , CA , USA ) , anti histone H1 . 2 , anti histone H2A and anti histone H2B ( from Abcam , Cambridge , UK ) . The polyclonal anti-ALDI used in this study is described in Turro et al . , 2006 . | Histones are proteins found in large numbers in most animal cells , where their primary job is to help DNA strands fold into compact and robust structures inside the nucleus . In vitro , histones are very effective at killing bacteria , and there is some evidence that histones secreted from cells provide protection against bacteria living outside cells . However , many types of bacteria are able to enter cells , where they can avoid the immune system and go on to replicate . In principle histones could protect cells against such bacteria from the inside , but for many years this was thought to be unlikely because most histones are bound to DNA strands in the cell nucleus , whereas the bacteria replicate in the cytosol . Moreover , free histones can be extremely damaging to cells , so most species have developed mechanisms to detect and degrade free histones in the cytosol . Recently , however , it was discovered that histones can bind to lipid droplets—organelles in the cytosol that are primarily used to store energy—in various animal cells and tissues . Now , Anand et al . have demonstrated that histones bound to lipid droplets can protect cells against bacteria without causing any of the harm normally associated with the presence of free histones . In in vitro experiments with lipid droplets purified from Drosophila embryos , they showed that histones bound to lipid droplets could be released to kill bacteria . The histones were released by lipopolysaccharide or lipoteichoic acid produced by the bacteria . The effect was also observed in vivo: using four different bacterial species , Anand et al . injected similar numbers of bacteria into Drosophila embryos that contained histones bound to lipid droplets , and also into embryos that had been genetically modified so that they did not contain such droplet-bound histones . While most of the normal embryos survived , the vast majority of the embryos without droplet-bound histones died . Similar results were also found in experiments on adult flies , along with evidence which suggests that histones might also provide defenses against bacteria in mice . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"microbiology",
"and",
"infectious",
"disease",
"immunology",
"and",
"inflammation"
] | 2012 | A novel role for lipid droplets in the organismal antibacterial response |
When unperturbed , somatic stem cells are poised to affect immediate tissue restoration upon trauma . Yet , little is known regarding the mechanistic basis controlling initial and homeostatic ‘scaling’ of stem cell pool sizes relative to their target tissues for effective regeneration . Here , we show that TEAD1-expressing skeletal muscle of transgenic mice features a dramatic hyperplasia of muscle stem cells ( i . e . satellite cells , SCs ) but surprisingly without affecting muscle tissue size . Super-numeral SCs attain a ‘normal’ quiescent state , accelerate regeneration , and maintain regenerative capacity over several injury-induced regeneration bouts . In dystrophic muscle , the TEAD1 transgene also ameliorated the pathology . We further demonstrate that hyperplastic SCs accumulate non-cell-autonomously via signal ( s ) from the TEAD1-expressing myofiber , suggesting that myofiber-specific TEAD1 overexpression activates a physiological signaling pathway ( s ) that determines initial and homeostatic SC pool size . We propose that TEAD1 and its downstream effectors are medically relevant targets for enhancing muscle regeneration and ameliorating muscle pathology .
Maintenance and repair of many tissues depend on reserve populations of tissue-specific stem cells . In the unperturbed state , these reserve cells are maintained at a defined pool size . Upon tissue damage , they expand to give rise to progenitors that can differentiate for efficient and timely tissue restoration as well as self-renew to re-establish the stem cell pool . With age , tissue function and regenerative capacity decline due to alterations in stem cell intrinsic function , microenvironment ( niche ) , and systemic cues . Collectively , these changes lead to a decline in the number of functional stem cells ( Oh et al . , 2014 ) . The skeletal muscle system provides an experimental paradigm for understanding stem cell-based tissue regeneration . Because of the essentiality of skeletal muscles for mobility , metabolic homeostasis , and thermogenesis , the clinical relevance of resident muscle stem cells cannot be over-stated . Skeletal muscle tissue can mount effective tissue restoration even after up to 50 weekly myotoxin-induced injuries in the mouse ( Luz et al . , 2002 ) . The muscle stem cells driving this remarkable lifelong regenerative response are satellite cells ( SCs ) , which were initially discovered by Alexander Mauro in Xenopus ( Mauro , 1961 ) . SCs comprise a small population ( 5 to 7% of all muscle nuclei ) of undifferentiated and quiescent mono-nucleated cells that reside between the basal lamina and the surface of skeletal muscle cells; a microenvironment referred to as their niche ( Cardasis and Cooper , 1975; Mauro , 1961 ) . SCs express the transcription factor Pax7 ( Seale et al . , 2000 ) . Lineage-tracing experiments unequivocally demonstrate that Pax7-expressing ( Pax7+ ) SCs are of somitic origin ( Lepper and Fan , 2010 ) , and they are the major contributing source of myonuclei in both postnatal development and injury-induced regeneration ( Lepper et al . , 2009 ) . Ablation of the Pax7+ cell population in adult mice establishes that they are essential for muscle regeneration ( Lepper et al . , 2011; McCarthy et al . , 2011; Murphy et al . , 2011; Sambasivan et al . , 2011 ) . Pax7+ SCs first appear under the basal lamina at late stages of fetal myogenesis ( day 15 . 5 of gestation in the mouse ) ( Gros et al . , 2005; Relaix et al . , 2005; Schienda et al . , 2006 ) . After birth , they remain highly active to supply myonuclei for muscle growth but gradually cease to differentiate by 3–4 weeks ( Lepper et al . , 2009; White et al . , 2010 ) . Therefore , SCs appear to be set-aside as a quiescent pool of stem cells during this time frame for muscle homeostasis and regeneration throughout lifetime . Tremendous insights into the molecular regulation of muscle stem cells during adult regeneration and aging have been gained in recent years ( Brack and Munoz-Canoves , 2015; Dumont et al . , 2015a ) . The primary foci of those studies are regulators for SC proliferation , differentiation , and self-renewal during adult and aged muscle homeostasis and following injury . Information derived from these studies helps formulate strategies towards restoring a functionally competent SC pool to combat muscle aging and muscular dystrophies . By contrast , a significant gap exists in our current knowledge pertaining to the regulation of SCs shortly after birth when they provide myonuclei to drive tissue growth and transition into quiescence , i . e . their initial establishment . In particular , how the initial SC population is proportionally scaled with respect to the muscle size during the postnatal period is entirely unknown . In adult homeostasis , SC numbers per muscle fiber in the mouse ( Collins et al . , 2005 ) or per defined myofiber length in human ( Boldrin and Morgan , 2011 ) reveal very minimal inter-fiber variation . Teleologically , SC number to muscle size scaling would be most relevant to the replicative potential of SCs per given muscle tissue volume in regeneration . It seems reasonable to presume that the physiological SC/muscle ratio is developmentally set at a maximum in a given muscle group for effective regeneration later on in life . To what extent this ratio can be altered without changing muscle size or impacting regenerative potential remains unclear . To date , loss of function studies have helped identify many necessary components for the initial fate specification and later maintenance of SCs . As such , changes in SC number in these animal models are not directly relevant to their initial scaling during the critical early postnatal period . By contrast , a ‘gain-of-function’ mouse model with super-numeral SCs is much desired , as its molecular dissection would much more likely enable the discovery of the molecular mechanism ( s ) sufficiently altering the initial SC pool size . Moreover , a robust SC hyperplasia mouse model could allow addressing the relevant clinical question of whether bestowing a tissue with extra stem cells would be of benefit or of detriment to the regenerative process of skeletal muscle and in other organismal systems . Yet , currently it is unknown whether the SC pool can be increased in the absence of overall muscle tissue hypertrophy . Here , we report a murine SC hyperplasia model and investigate the effects on regenerative myogenesis . We had generated transgenic mice that express hemagglutin-tagged ( HA ) -TEAD1 driven by a 6 . 5 Kb muscle creatine kinase ( MCK ) control region ( Tsika et al . , 2008 ) . TEAD1 belongs to the TEA domain transcription factor family , which serves important functional roles in multiple embryonic contexts by activating gene expression when bound to the Hippo signaling pathway effectors Yap or Taz ( Zhao et al . , 2008 ) . Muscle-intrinsic changes effected by TEAD1 overexpression include a prominent transition toward a slow muscle contractile phenotype ( Tsika et al . , 2008 ) . Here , we provide comprehensive analyses of the SC compartment and make the discovery of a robust , up to 6-fold increase of quiescent SCs in TEAD1-Tg mice . As a functional consequence , TEAD1-Tg muscle regenerates at a faster rate post-injury likely owing to their increased SC number as well as the increased proliferation rates observed when the myofiber-associated SCs are activated . Remarkably , in the mdx mouse model for Duchenne muscular dystrophy , skeletal muscle pathology is significantly ameliorated by TEAD1-overexpression , which is likely contributed in part by an increase in utrophin expression . Lastly , we provide evidence implicating TEAD1-Tg skeletal muscle fibers in the regulation of mouse SC number . This work implicates a role for TEAD1-induced , myofiber-derived signaling that can scale the initial SC pool size during the perinatal period . Additionally , TEAD1 overexpression appears to alter myofiber stability in the context of dystrophic disease . Further study of this mouse model will be invaluable in elucidation of physiological pathways controlling stem cell numbers , and may prove to be an entry point to modulating SC number in clinical settings .
A remarkable feature of skeletal muscle is its ability to adapt to changing physiologic demands via reversible modulation of tissue size and of fiber-type composition , to accommodate differential force exertion or contractile usage patterns , respectively . To this end , we previously identified a regulatory role for Tead1 in the induction of slow muscle gene expression: skeletal muscle of TEAD1-Tg mice features a transition to the slow muscle contractile phenotype ( Tsika et al . , 2008; Vyas et al . , 1999 ) . To further confirm this transition , we employed immunofluorescence ( IF ) analysis to tibialis anterior ( TA ) muscle cross-sections from adult TEAD1-Tg and wild type ( Wt ) sibling mice using antibodies against fiber-type specific myosins ( Figure 1—figure supplement 1 ) . Fast glycolytic myofibers express IIX myosin and make up approximately half of all TA fibers of Wt mice ( Figure 1—figure supplement 1A ) . By contrast , this fiber type is absent in TEAD1-Tg TA muscle ( Figure 1—figure supplement 1B; quantification in Figure 1I ) . Reciprocal analysis using an antibody detecting any myosin but the IIX type confirmed the complete loss of this fiber type as all TA muscle fibers stained positive in TEAD1-Tg samples ( Figure 1—figure supplement 1C–D; quantification in Figure 1I ) . While the low percentage of slow twitch ( I myosin+ ) fibers is not affected , a more than 2-fold increase in IIa myosin+ fibers suggests that the fast glycolytic fibers are replaced in large part by fast oxidative fibers in TEAD1-Tg TA muscles ( Figure 1—figure supplement 1E–H; quantification in Figure 1I ) . Whether additional tissue alterations accompany this TEAD1-induced change in fiber-type composition is unknown . For further characterization , we decided to evaluate muscle fiber number ( hyperplasia ) and size ( hypertrophy ) in TEAD1-Tg mice . 10 . 7554/eLife . 15461 . 003Figure 1 . Skeletal muscle is histologically indistinguishable between Wt and TEAD1-Tg mice . ( A–P ) H and E ( A–B , E–F , I–J , M–N ) and Sirius Red ( C–D , G–H , K–L , O–P ) stains of TA ( A–D ) , EDL ( E–H ) , Plantaris ( I–L ) , and Soleus ( M–P ) from Wt ( A , C , E , G , I , K , M , O ) and TEAD1-Tg mice ( B , D , F , H , J , L , N , P ) . Insets are magnified images of the representative areas within the tissue . ( Q–S ) EDL fiber number ( Q ) , fiber area ( R ) , and fiber diameter ( S ) are quantified for both genotypes ( n > 3 adult mice for all measurements ) . DOI: http://dx . doi . org/10 . 7554/eLife . 15461 . 00310 . 7554/eLife . 15461 . 004Figure 1—figure supplement 1 . Ratios of type II fibers are changed in TEAD-Tg muscle . Wt ( A , C , E , G ) and TEAD1-Tg ( B , D , F , H ) TA muscles were queried for percentages of type IIX fibers ( A , B ) , fibers that were not IIX ( C , D ) , type IIa fibers ( E , F ) , and type I fibers ( G , H ) . The TA muscle was chosen for analysis because it is comprised of all muscle fiber types . The results were then quantified ( I ) . n > 3 adult mice for all measurements , p<0 . 05 represented by ( * ) , p<0 . 005 represented by ( ** ) , p<0 . 0005 represented by ( *** ) . DOI: http://dx . doi . org/10 . 7554/eLife . 15461 . 004 We investigated the size of skeletal muscle of multiple hind limb muscle groups from TEAD1-Tg mice by weight and by histological staining to determine muscle fiber size and number . Four muscle groups were chosen based on their fiber-type compositions as follows: primarily fast-twitch glycolytic fibers ( EDL and plantaris ) , slow-twitch oxidative fibers ( soleus ) , and mixed fast- and slow-fiber types ( TA ) ( Figure 1 ) . Histological staining by haematoxylin and eosin ( H and E ) , as well as , by Sirius Red of the chosen muscle groups from 3-months old TEAD1-Tg and Wt littermates did not reveal histopathology as myofiber nuclei were peripherally located , and interstitial spacing between individual fibers was normal with no excess collagen or mononucleated cells ( Figure 1A–P ) . TEAD1-Tg mice also did not display muscle hypertrophy , as weights of all analyzed hind limb muscles were unchanged ( Table 4 ) . Because its fibers extend the entire length of the muscle , the EDL muscle facilitates accurate determination of myofib er number and size in cross sections . No differences were detected in myofiber number ( Figure 1Q ) , or in myofiber size by cross-sectional area ( Figure 1R ) and fiber diameter ( Figure 1S ) between TEAD1-Tg and Wt samples . All together , we conclude that forced expression of TEAD1 does not induce any overt pathological ( i . e . fibrosis ) or size ( i . e . hypertrophy or hyperplasia ) alterations of skeletal muscle . Reports of a higher density of SCs in slow-twitch compared to fast-twitch muscles ( Gibson and Schultz , 1983; Schmalbruch and Hellhammer , 1977 ) prompted us to investigate the SC compartment in TEAD1-Tg muscle , which has an increase in slow muscle gene expression ( Tsika et al . , 2008 ) . We determined SC number by assessing Pax7+ cells on TA muscle sections by IF staining ( Figure 2A , Bi; quantification in Figure 2I ) . A striking 5-fold increase in Pax7+ cells in TEAD1-Tg was found as compared to Wt TA muscle ( Figure 2I ) . This increase is of much greater magnitude than expected from the ~2-fold difference in SC number between slow- and fast-twitch muscles ( Gibson and Schultz , 1983; Schmalbruch and Hellhammer , 1977 ) . The increase in SC number was consistent between 3 independent TEAD1-Tg mouse lines ( L12 , L4 , L14 ) excluding the possibility of transgene integration sites accounting for the observed increase in SCs ( Figure 2I ) . For the rest of this study , we utilized line L12 exclusively for consistency . 10 . 7554/eLife . 15461 . 005Figure 2 . TEAD1-Tg hind limb features SC hyperplasia . ( A–Hi ) Pax7 IF in hind limb muscle groups: TA ( A–Ai , B–Bi ) , Plantaris ( C–Ci , D–Di ) , Soleus ( E–Ei , F–Fi ) , and EDL ( G–Gi , H–Hi ) . Wt ( A–Ai , C–Ci , E–Ei , G–Gi ) and TEAD1-Tg muscles ( B–Bi , D–Di , F–Fi , H–Hi ) are represented . Ai–Hi are zoomed images of areas represented by the white boxes in A–H , arrows indicate Pax7 ( green ) positive nuclei ( blue ) . ( I–K ) Quantification of SC hyperplasia ( fold increase above Wt ) in TA sections from three independent TEAD1 transgenic lines with different transgene insertion sites ( I ) . Quantification of TA sections from Wt and TEAD1-Tg line L12 for fold increase of myonuclei per fiber , SC per fiber , and SC per myonuclei ( J ) . Quantification of plantaris , soleus , and EDL muscle group sections for fold increase in SC number in TEAD1-Tg line L12 compared to WT ( K ) . n > 3 adult mice for all measurements , p<0 . 05 represented by ( * ) , p<0 . 005 represented by ( ** ) , p<0 . 0005 represented by ( *** ) . DOI: http://dx . doi . org/10 . 7554/eLife . 15461 . 005 The absence of myofiber hyperplasia and/or hypertrophy ( Figure 1 ) argues for a specific SC hyperplasia in TEAD1-Tg muscle . To confirm this , we quantified myofiber nuclei in TA muscles , which revealed no differences in the myonuclei to myofiber ratio , yet the SC to myonuclei ratio is increased 5-fold when comparing TEAD1-Tg and Wt samples , thus establishing a selective effect in SC hyperplasia ( Figure 2J ) . We also detected significant increases in SCs among the groups of muscle of TEAD1-Tg mice , including ~4 . 5-fold increase in the plantaris , ~6-fold increase in the soleus , and ~2 . 5-fold increase in the EDL ( Figure 2C–Hi; quantification in Figure 2K ) . The SC hyperplasia of the predominantly slow-twitch soleus is particularly noteworthy as it argues against the increase in SCs being a simple consequence of myofiber conversion to the slow phenotype ( Tsika et al . , 2008 ) . Collectively , these findings rule out skeletal muscle hypertrophy and hyperplasia , and demonstrate specific SC hyperplasia in TEAD1-Tg fast- and slow-twitch skeletal muscles of the lower hind limb . During muscle development , SCs gradually become quiescent by 3–4 weeks after birth ( Lepper et al . , 2009; White et al . , 2010 ) . Intriguingly , peak HA-TEAD1 expression is detected prior to this time , at around postnatal day 14 ( PN 14d ) ( Tsika et al . , 2008 ) . Therefore , we next examined whether TEAD1-Tg muscles show SC hyperplasia during this early postnatal time period . SCs were quantified in hind limb muscles of TEAD1-Tg and Wt sibling mice at designated postnatal time points ( PN 10d , PN 12d , PN 14d and PN 20d ) , as was done for adult muscles at three months ( Figure 2 , Materials and methods ) . To monitor SC proliferation during the postnatal period , we applied the thymidine analogue EdU . At PN 10d and PN 12d , we did not detect differences in the numbers of sublaminal Pax7+ cells or in the percentages of EdU+/Pax7+ cells between TEAD1-Tg and Wt sibling TA muscles ( Figure 3A , Biii; quantification in Figure 3E , G ) . At PN 14d , TEAD1-Tg TA muscles had begun accumulating an ~1 . 5-fold greater number of Pax7+ cells when compared to Wt TA muscles ( Figure 3E , F ) . This difference became greater at PN 20d when TA muscles boasted a >3-fold increase in SCs in TEAD1-Tg versus Wt mice ( Figure 3C , Diii; quantification in Figure 3E-F ) . Quantification of the percentage of sublaminal Pax7+/EdU+ cells at PN 20d revealed a significant 2-fold increase in the percentage of proliferative SCs in TEAD1-Tg compared to Wt TA muscles ( Figure 3G ) . Conceivably , SCs could be protected from cell death in TEAD1-Tg muscle , which could contribute to the greater number of SCs . To assay for SC apoptosis , we performed TUNEL labeling coupled with anti-Pax7 IF on TA muscle sections from TEAD1-Tg and Wt siblings at PN 14d when initial SC increases are detected ( Figure 3H , I ) . A single instance of co-labeling was detected in the TEAD1-Tg sections ( 0 . 15% ) while none were observed in the Wt sections ( Figure 3J ) suggesting TEAD1-Tg SC protection from apoptosis does not drive the SC hyperplasia . Together , these data indicate that TEAD1-Tg muscle accumulates super-numeral SCs beginning at ~2 weeks after birth via extending the proliferation period . Such super-numeral SCs appear to be excluded from fusion with the myofiber ( no increase in the myonuclei/fiber ratio ) , providing the basis for the specific increase of SCs over myonuclei number . 10 . 7554/eLife . 15461 . 006Figure 3 . SC hyperplasia arises during perinatal stages in TEAD1-Tg mice . ( A–Diii ) Wt ( A–Aiii , C–Ciii ) and TEAD1-Tg ( B–Biii , D–Diii ) SC numbers and proliferation were assessed . Postnatal day 12 ( PN 12d , A–Biii ) and day 20 ( PN 20d , C–Diii ) are represented in IF images . Pax7 ( green ) alone is shown in A , B , C , and D , and combined with DAPI ( blue ) and laminin ( red ) in the merged images Ai , Bi , Ci , Di . EdU ( red ) alone is represented in Aii , Bii , Cii , Dii and combined with Pax7 ( green ) in merged images Aiii , Biii , Ciii , Diii , arrows indicate nuclei labeled with both EdU and Pax7 while arrowheads are Pax7 positive nuclei that show no EdU label . ( E–G ) Numbers of SCs normalized to total myofibers per image is quantified for PN 10d , 12d , 14d , and 20d Wt or TEAD1-Tg TA sections ( E ) . Fold increase in SC numbers in TEAD1-Tg TA sections compared to Wt was quantified for PN 10d , 12d , 14d , and 20d ( F ) . Fold increase in percent EdU positive SCs in TEAD1-Tg TA sections compared to Wt was quantified for PN 12d and 20d ( G ) . ( H–J ) Apoptosis rates of SCs were assessed for PN 14d . Representative images for Wt ( H ) and TEAD1-Tg muscle ( I ) show Pax7 ( green ) and TUNEL ( red ) stained nuclei ( blue ) . Percent of SCs that were TUNEL positive or negative was quantified ( J ) . n > 3 mice for all measurements , p<0 . 05 represented by ( * ) , p<0 . 005 represented by ( ** ) , p<0 . 0005 represented by ( *** ) . DOI: http://dx . doi . org/10 . 7554/eLife . 15461 . 006 During adult homeostasis , SCs localize to the myofiber sublaminal space , are highly polarized , and maintain a quiescent state , all of which is characterized by typical molecular marker expression and a non-proliferative state . To determine if super-numeral SCs of TEAD1-Tg mice have acquired the quiescent muscle stem cell state , we performed extensive marker and proliferation analyses in adult ( 3-months old ) mice . First , we determined proper niche localization of SCs using anti-Pax7 and anti-laminin dual-IF staining ( Figure 4A , B ) . All SCs of TEAD1-Tg muscle localized to the sublaminal space ( Figure 4G ) . Next , we analyzed SC polarity using β1-Integrin as a basal marker and M-Cadherin as an apical marker ( Figure 4C–F ) . Similar to controls , ≥80% of SCs displayed proper polarity ( Figure 4H ) . Calcitonin receptor ( CTR ) is specifically expressed by quiescent SCs and important for maintenance of quiescence ( Fukada et al . , 2007; Yamaguchi et al . , 2015 ) . Like Wt SCs , SCs from TEAD1-Tg mice express CTR ( Figure 4I–Ji; quantification in Figure 4KJ ) and also the general stem cell marker , CD34 ( Figure 4K ) . These data suggest that super-numeral adult SCs of TEAD1-Tg mice maintain proper cell polarity , niche interaction , as well as the quiescent state . To probe the quiescent state more rigorously , we conducted long-term proliferation assays via administering BrdU in the drinking water for a month and assaying for BrdU+ nuclei in the sublaminal space ( Figure 4L–M ) . No differences were found in proliferation between SCs of TEAD1-Tg and those of Wt muscles ( Figure 4N–O ) . All together , we conclude that super-numeral SCs of TEAD1-Tg muscle acquire and maintain proper niche occupation and cell polarity , as well as , the quiescent state typical of adult muscle stem cells . 10 . 7554/eLife . 15461 . 007Figure 4 . SC localization , marker expression , and long-term proliferation are no different in TEAD1-Tg compared to Wt mice . ( A–H ) Localization of SC ( Pax7 , green ) within the muscle fiber basal lamina ( laminin , red ) shown by IF of adult Wt ( A ) and TEAD1-Tg ( B ) TA sections and quantified for TEAD1-Tg TA sections ( G ) . Polarized Integrin-β1 expression ( Integrin , green ) in SCs ( Pax7 , red ) was assessed on TEAD1-Tg and Wt isolated EDL fibers ( myofiber , mf ) and found to be basally localized in all instances where the position of the SC on the fiber allowed for assessment ( C , D ) . M-Cadherin ( Mcad , green ) polarization in SCs ( Pax7 , nuclei shown in red ) was assessed in TA sections for apical localization ( E , F ) , which is quantified in H . Non-polarized Mcad in Wt and TEAD1-Tg samples is likely due to imperfect SC orientation within the section . I–K ) Assessment of quiescent marker ( calcitonin receptor , CTR ) expression in SCs ( Pax7 , Ii and Ji ) on Wt ( I ) and TEAD1-Tg ( J ) isolated EDL fibers are represented by IF and quantified in K along with the general stem cell marker , CD34 . L–N ) Following a month-long BrdU treatment , Wt ( L ) and TEAD1-Tg ( M ) adult TA sections were assessed for long-term proliferation . Since myonuclei are non-proliferative and SC nuclei present the only other sublaminal nuclear species , this assay cumulatively captures any SC proliferation over the one-month long period . Very low numbers of BrdU nuclei were detected in both sublaminal and interstitial compartments of TA muscles from TEAD1-Tg and Wt mice . BrdU ( green ) labeled nuclei ( blue ) within the basal lamina ( red ) of the myofiber were quantified ( N ) and normalized to SC number ( O ) . For G , H , and K n > 50 cells , while for N , 3 mice were quantified for each genotype . DOI: http://dx . doi . org/10 . 7554/eLife . 15461 . 007 Since we found SCs of TEAD1-Tg skeletal muscle to have a normal quiescent phenotype ( Figure 4 ) , we next asked whether they retain the functional capacity to regenerate tissue after injury . To determine this , the TA muscles of adult ( 2–3 months old ) TEAD1-Tg and Wt littermates were injured by intra-muscular injection of cardiotoxin ( CTX; see Materials and methods ) . Histological analyses of muscle regenerates were performed 3 , 7 , and 14 days post injury ( dpi , Figure 5A–Fi ) . Remarkably , while not detectable in Wt samples , very small regenerative myotubes could be detected histologically at 3 dpi in TEAD1-Tg regenerates ( Figure 5A–Bi ) , indicating that the regeneration process is accelerated in muscle with super-numeral SCs . To confirm this , we applied IF analyses using embryonic myosin heavy chain ( eMyHC ) as a marker for early regenerated myotubes and found widespread and robust eMyHC expression in TEAD1-Tg muscle regenerates , which is in contrast to Wt regenerates , for which eMyHC expression is more sparse and less robust ( Figure 5I–Ji ) . At 7 and 14 dpi , regenerative myofibers with centrally located myonuclei were present in both Wt and TEAD1-Tg samples , with no evidence of fibrosis or immune cell infiltration ( Figure 5C–Fi ) , demonstrating that TEAD1-Tg muscle retains full regenerative capacity . Consistent with the accelerated formation of new myofibers at 3 dpi , when we quantified the sizes of regenerated myofibers at 7 dpi , we found both fiber area and diameter to be significantly increased in TEAD1-Tg muscle regenerates compared to Wt ( Figure 5G–H ) . By 14 dpi , no difference in fiber size was detected . Maturation of regenerative myofibers is normal in TEAD1-Tg TA muscles as fiber sizes were no different compared to Wt at 35 dpi ( Figure 5G–H ) . These data demonstrate that TEAD1-Tg muscle features accelerated kinetics of initial myotube formation upon injury . Despite this accelerated regeneration process and super-numeral SCs , we are surprised that no myofiber hypertrophy resulted from the injury-induced regenerative myogenesis . We therefore suggest that these two processes are separable . 10 . 7554/eLife . 15461 . 008Figure 5 . TEAD1-Tg muscle exhibits faster regeneration than Wt muscle upon injury with cardiotoxin . ( A–Fi ) H and E stains of Wt ( A–Ai , C–Ci , E–Ei ) and TEAD1-Tg ( B–Bi , D–Di , F–Fi ) TA muscle 3 days ( 3 dpi , A–Bi ) , 7 days ( 7 dpi , C–Di ) , or 14 days ( 14 dpi , E–Fi ) after injury with cardiotoxin ( CTX ) . Panels Ai–Fi show a higher magnification of the regenerating muscle ( indicated by arrows and central nuclei ) . A lack of arrows in Ai is due to a lack of identifiable young fibers at that stage in Wt muscle . G–H ) Quantification of fiber area ( G ) and fiber diameter ( H ) for 7 dpi , 14 dpi , and 35 dpi . I–Ji ) IF of embryonic myosin heavy chain ( eMyHC , I–Ji ) localized to regenerating fibers in Wt ( I–Ii ) and TEAD1-Tg ( J–Ji ) TA muscles 3 days after injury by CTX . Images Ii and Ji show a zoomed-in region indicated by the white boxes in images I and J . n > 3 mice for all samples quantified . p<0 . 05 represented by ( * ) , p<0 . 005 represented by ( ** ) , p<0 . 0005 represented by ( *** ) . DOI: http://dx . doi . org/10 . 7554/eLife . 15461 . 008 We applied an additional injury paradigm via the myotoxin BaCl2 , which is thought to display less toxicity towards SCs compared to cardiotoxin ( Boldrin et al . , 2012; Gayraud-Morel et al . , 2009 ) , to more accurately elucidate the rapid kinetics of initial muscle repair ( 3 , 5 and 7 dpi; Figure 6A–Hi ) . Confirming our observations made with the cardiotoxin injury paradigm ( Figure 5 ) , we found significantly larger myofibers in TEAD1-Tg compared to Wt control regenerates demonstrated by increased fiber area ( Figure 6I ) and diameter ( Figure 6J ) at both 5 and 7 dpi . We conducted IF analyses of myogenic marker expression to examine the timing of myogenic differentiation in more detail ( Figure 6K–R ) . Myogenin is a terminal marker of myogenic differentiation and its expression gradually increases from 3 to 5 dpi; while eMyHC expression typically initiates at 3 dpi , peaks at 4 and 5 dpi , and then diminishes by 6 to 7 dpi , at which point it becomes replaced by adult myosin heavy chain isoforms . Consistent with the earlier detection of new myotubes by 3 dpi by histology in TEAD1-Tg compared to Wt regenerates ( Figure 5A , B ) , we found both Myogenin and eMyHC to be increased at this early regeneration time point in TEAD1-Tg compared to Wt samples ( Figure 6K–N ) . By 5 dpi , while the number of Myogenin+ cells is still increased in TEAD1-Tg compared to Wt regenerates ( Figure 6O–P ) , eMyHC expression is already reduced in TEAD1-Tg regenerative myofibers ( Figure 6R ) compared to the 3 dpi time point ( Figure 6N ) and the peak expression observed in Wt 5 dpi skeletal muscle regenerates ( Figure 6P ) , suggesting it is already being replaced by more mature myosins . Taken together , these data demonstrate acceleration of skeletal muscle regeneration by super-numeral SCs . 10 . 7554/eLife . 15461 . 009Figure 6 . TEAD1-Tg muscle exhibits faster regeneration than Wt muscle upon injury with BaCl2 . ( A–Hi ) H and E stains of Wt ( A–Di ) and TEAD1-Tg ( E–Hi ) TA muscle 3 days ( B–Bi , F–Fi ) , 5 days ( C–Ci , G–Gi ) , or 7 days ( D–Di , H–Hi ) after injury with BaCl2 or uninjured ( A–Ai , E–Ei ) . Images Ai–Hi are higher magnification views of regenerating areas of muscle ( regenerating fibers indicated by black arrows ) . I–J ) Quantification of fiber area ( I ) and fiber diameter ( J ) for 5 days and 7 days after BaCl2 injury . K–R ) IF images show myogenin ( green ) and DAPI ( blue; K , M , O , Q ) or embryonic myosin heavy chain ( green ) with laminin ( red; L , N , P , R ) localized to regenerating fibers in Wt ( K–L , O–P ) and TEAD1-Tg ( M–N , Q–R ) TAs 3 days and 5 days after BaCl2 injury . For quantification of fiber number and diameter n=3 mice were used . p<0 . 05 represented by ( * ) , p<0 . 005 represented by ( ** ) , p<0 . 0005 represented by ( *** ) . DOI: http://dx . doi . org/10 . 7554/eLife . 15461 . 009 SCs represent a self-renewable reservoir of stem cells able to repeatedly provide myonuclei for muscle repair . To test if super-numeral SCs of TEAD1-Tg muscle can self-renew , we assayed for their capacity to support repeated regeneration bouts . For this , we injured the TA muscles of TEAD1-Tg and Wt littermates three times , allowing 5 weeks between injuries for complete regeneration ( Figure 7A ) . Indeed , robust regeneration in both singly ( Figure 7B , D ) and triply ( Figure 7C , E ) injured TA muscles of both Wt and TEAD1-Tg mice were observed . We did not detect any excess collagen deposition in TEAD1-Tg TA muscles after one or three injury-induced regeneration bouts compared to Wt TA muscles ( Figure 7F–I ) . These data demonstrate that super-numeral SCs of TEAD1-Tg skeletal muscle retain regenerative capacity after repeated injury and suggest that these SCs self-renew ( to maintain their numbers as cells are recruited for fusion into myofibers ) . To test if these SCs participated in and contributed to the regeneration process prior to replenishing the Pax7+ muscle reserve cell pool , we interrogated whether Pax7+ SCs of skeletal muscle regenerates had proliferated by administering EdU during the proliferative period of the regeneration process to mark cells in S-Phase . Two weeks after injury , we could readily detect EdU+/Pax7+ cells in both Wt and TEAD1-Tg samples ( Figure 7J–K ) . We did not detect any statistically significant differences in the percentages of EdU+/Pax7+ SCs between Wt and TEAD1-Tg samples ( Figure 7L ) . These data further support that SCs of TEAD1-Tg mice are self-renewing . Furthermore , we not only detected sublaminal Pax7+ SCs in singly- , but also in triply-injured TA muscles of both TEAD1-Tg and Wt littermates ( Figure 7M–P ) . Quantification of the number of Pax7+ SCs revealed that super-numeral SCs are maintained in skeletal muscle regenerates of TEAD1-Tg mice ( Figure 7Q–R ) , though the magnitude of the typical 5-fold increase in SCs in uninjured muscle is reduced to ~3 . 5-fold after one and ~3-fold after three injuries ( Figure 7R ) . This reduction in the magnitude of the SC hyperplasia after three injuries could reflect a diminished self-renewal capacity . All together , these data demonstrate that super-numeral SCs of TEAD1-Tg muscle can self-renew and maintain the tissue’s high capability for regeneration after repeated traumatic insults . 10 . 7554/eLife . 15461 . 010Figure 7 . SC hyperplasia persists through multiple injuries . Adult mouse TA muscles were injured with BaCl2 multiple times with 35-day regeneration periods between injuries as indicated ( A ) . This injury paradigm applies to panels B–I , M–P ) H and E stains show muscle after 35 days of regeneration following one injury ( B , D ) or 3 injuries ( C , E ) for Wt ( B , C ) or TEAD1-Tg muscle ( D , E ) . ( F–I ) Sirius Red stains show the connective tissue after 35 days of regeneration following one injury ( F , H ) or 3 injuries ( G , I ) for Wt ( F , G ) or TEAD1-Tg muscle ( H , I ) . ( J–L ) Wt ( J ) and TEAD1-Tg ( K ) TA muscles were injured with CTX and regenerated for 2 weeks . EdU was given on days 2–5 of regeneration . IF of Pax7 ( green ) , laminin ( red ) , EdU ( white ) , and counterstained with DAPI ( blue ) allowed for quantification of sublaminal Pax7 and EdU positive cells ( L ) . Myonuclei in images are EdU positive but reduced in intensity due to larger nuclear volume . M–P ) Numbers of Pax7 expressing cells were assessed by IF of Pax7 ( green ) and laminin ( red ) , counterstained with DAPI ( blue ) after 35 days of regeneration following one injury ( M , O ) or 3 injuries ( N , P ) for Wt ( M , N ) or TEAD1-Tg muscle ( O , P ) and is quantified as SC averages per area ( Q ) and fold increase relative to Wt ( R ) . Uninjured data from Figure 2 displayed for comparison . n > 3 adult mice for all measurements , p<0 . 05 represented by ( * ) , p<0 . 005 represented by ( ** ) , p<0 . 0005 represented by ( *** ) . DOI: http://dx . doi . org/10 . 7554/eLife . 15461 . 010 To challenge the regenerative capacity of TEAD1-Tg muscle further , we subjected the tissue to a state of chronic degeneration . To accomplish this , we bred TEAD1-Tg mice to the mdx mouse , a mouse model of Duchenne muscular dystrophy , which is the most severe form of the muscle wasting diseases . These mice lack the sarcolemmal protein dystrophin , which results in greatly destabilized muscle fibers and thus , chronically injured and regenerating muscle tissue , evidenced by the presence of many regenerative myofibers with centrally located nuclei , a hallmark pathological feature of this disease . Histological analyses revealed greatly reduced numbers of regenerative muscle fibers in mdx; TEAD1-Tg compared to mdx TA muscles of 2–3 months old littermates ( Figure 8A–B ) . The percentage of fibers with central nuclei was reduced from ~55% in Wt to less than 20% in TEAD1-Tg muscles ( Figure 8C ) , implying less degeneration in mdx; TEAD1-Tg muscle . We also noted a concomitant significant reduction in myofiber hypertrophy , a secondary pathological feature of the mdx dystrophy model ( Figure 8D–E ) . To further probe the extent of improvement of the dystrophic phenotype , we assayed for fibrosis , another histological hallmark feature of skeletal muscle dystrophy . Trichrome stainings revealed a significant reduction in fibrotic area in mdx; TEAD1-Tg compared to mdx muscle ( Figure 8F–G; quantification in Figure 8H ) . Dystrophic muscle fibers have an altered sarcolemmal permeability , which can be assessed via Evans Blue Dye ( EBD ) uptake . While large numbers of EBD positive fibers where detected in mdx muscle , EBD incorporation by fibers of mdx; TEAD1-Tg mice was largely stunted ( Figure 8I–J; quantification in Figure 8K ) . The absence of evidence for worsened muscle pathology and by contrast , an amelioration of the dystrophy argues for SCs being functional in dystrophic TEAD1-Tg muscle . Lastly , we assayed for SCs in dystrophic muscle of mdx and mdx; TEAD1-Tg littermates ( Figure 8L–M ) . Quantification of Pax7+ cells revealed an ~2-fold increase in SCs in mdx; TEAD1-Tg mice ( Figure 8N ) . While less dramatic compared to the fold increase observed in healthy muscle , the SC hyperplasia in dystrophic muscle is quite substantial considering the baseline SC numbers are already increased due to higher SC proliferation in the chronically de- and regenerating environment . The magnitude of the SC hyperplasia is potentially also affected by impaired dystrophin-deficient SCs ( Dumont et al . , 2015b ) . 10 . 7554/eLife . 15461 . 011Figure 8 . Pathologies of muscular dystrophy are ameliorated by TEAD1-Tg expression . Histology , IF analyses , and qRTPCR on dystrophic muscle from mdx mice modeling chronic injury . A–C ) Asterisks in H and E stained muscle sections indicate regenerated fibers in mdx ( A ) and mdx; TEAD1-Tg mice ( B ) . The percent of fibers that are regenerative is quantified for each genotype in C . D–E ) The fiber area ( D ) and fiber diameter ( E ) for mdx or mdx; TEAD1-Tg TA muscle is also quantified . F–H ) Trichrome staining was employed to label fibrotic areas in mdx ( F ) or mdx; TEAD1-Tg ( G ) TA muscle . Percent fibrotic area was quantified ( H ) . I–K ) Membrane leakage was assessed by Evan’s Blue Dye incorporation into the myofibers of mdx ( I ) or mdx; TEAD1-Tg ( J ) TA muscle . Positive fibers per area were quantified ( K ) . ( L–N ) Numbers of Pax7 expressing cells were assessed by IF of Pax7 ( green ) , and laminin ( red ) , counterstained with DAPI ( blue ) in mdx ( L ) or mdx; TEAD1-Tg ( M ) . TA muscles are quantified as fold increase relative to mdx in N . Utrophin ( Utrn ) expression was quantified in mdx or mdx; TEAD1-Tg TA muscle via qRTPCR ( O ) . n > 3 adult mice for all measurements , p<0 . 05 represented by ( * ) , p<0 . 005 represented by ( ** ) , p<0 . 0005 represented by ( *** ) . DOI: http://dx . doi . org/10 . 7554/eLife . 15461 . 011 Whether increased SCs , altered myofiber properties , or both contribute to the amelioration of the dystrophic pathology in mdx; TEAD1-Tg muscle is unclear . Besides differing in their contractile properties , slow-twitch muscle features higher utrophin levels compared to fast-twitch muscle ( Gramolini et al . , 2001 ) . Utrophin can functionally substitute for dystrophin ( Rafael et al . , 1998 ) . Since TEAD1-Tg muscle features a transition to the slow contractile muscle protein phenotype , we decided to investigate utrophin expression in mdx; TEAD1-Tg skeletal muscle . Quantitative PCR revealed an ~3-fold increase in utrophin expression in mdx; TEAD1-Tg compared to mdx muscle samples ( Figure 8O ) . Additionally , we assayed for revertant myofibers via anti-dystrophin IF . No increased reversion rates were found in mdx; TEAD1-Tg compared to mdx muscle samples ( data not shown ) . These data suggest that stabilization of the sarcolemma via utrophin up-regulation contributes to amelioration of the dystrophic muscle pathology of mdx; TEAD1-Tg mice . The HA-tagged TEAD1 transgene is under the control of the muscle creatine kinase promoter and presumed to be exclusively expressed by the myofiber and not its associated SCs . To confirm this , we determined the expression of the TEAD1 transgene with respect to SCs or the myofiber plasmalemma by co-IF for HA and Pax7 in TEAD1-Tg skeletal muscle ( Figure 9A–Biii ) . We found no overlap of HA+ and Pax7+ nuclei ( Figure 9A–Aiii ) . All HA expression was localized below the dystrophin+ domain revealing that the only TEAD1 expressing nuclear species are those within the myofiber ( Figure 9B–Biii ) . To more rigorously probe MCK promoter-controlled TEAD1 transgene expression , we grew myoblasts in heterogeneous cultures containing both proliferative myoblasts as well as differentiated myocytes . We then assayed for transgene expression ( HA ) in post-mitotic differentiated myocytes ( Myogenin+ ) . All HA staining was contained within the domain of Myogenin+ cells ( Figure 9C ) . These data demonstrate that the TEAD1 transgene is not expressed by proliferative myoblasts but only by differentiated myogenic cells . Reciprocally , we probed for TEAD1 transgene expression ( via HA ) in Pax7+ reserve cells . As expected , transgene expression was excluded from the domain of Pax7+ reserve cells ( Figure 9C ) . 10 . 7554/eLife . 15461 . 012Figure 9 . SC hyperplasia derives from cell-cell signaling between the myofiber and SCs . ( A–Aiii ) IF images of TEAD1-Tg TA adult muscle show Pax7 expressing SCs ( green; A ) and HA tagged TEAD1 ( red; Ai ) , and are merged in image ( Aii ) and a zoomed-in merged image ( Aiii ) to show that these genes are expressed in distinct compartments . B–Biii ) IF of HA-TEAD1 ( red , Bi ) and dystrophin ( green , B ) show that the TEAD1 transgene is expressed in nuclei of the myofiber in TA adult muscle . Transgene expression was also queried by HA colocalization with Myogenin ( Myog ) and Pax7-expressing reserve cells in differentiated cell culture experiments after 3 and 5 days of differentiation , respectively ( C ) . ( D–J ) In vitro differentiation of myoblasts derived from Wt ( D–F ) or TEAD1-Tg ( G–I ) hind limb muscle shows equivalent expression of MyoD ( D , G; green ) , Myogenin ( E , H; green ) , and myosin heavy chain ( F , I; green ) and is quantified in J ( n> 480 cells for all ) . ( K–M ) Cultures of Wt ( K ) or TEAD1-Tg ( L ) myoblasts show equivalent EdU label ( green ) , which is quantified in M ( n>375 cell for each ) . N–P ) Muscle fibers and associated SC were isolated and grown in culture for 72 hr with EdU in the growth media after 24 hr in culture . A minimum number of 50 fibers were quantified . Cell clones labeled with Pax7 ( green ) , DAPI ( blue ) , and/or EdU ( red ) can be observed on isolated muscle fibers from Wt ( N ) or TEAD1-Tg ( O ) mice and are quantified in P . The average number of Pax7-positive cells ( T0 ) and the average number of clones on muscle fibers after 72 hr ( T72 ) in growth media are quantified in Q . Clone sizes for TEAD1-Tg and Wt fibers at 72 hr post myofiber isolation in R ( n=3 mice each ) . After 72 hr in culture , the ratio of self-renewing SCs to differentiating cells within these clones is quantified in U ( n>150 clone ) for both Wt ( S ) and TEAD1-Tg ( T ) isolated fibers cultured in 10% FBS/ DMEM . p<0 . 05 represented by ( * ) , p<0 . 005 represented by ( ** ) , p<0 . 0005 represented by ( *** ) . DOI: http://dx . doi . org/10 . 7554/eLife . 15461 . 012 The above results predict that the hyper-proliferation of SCs in TEAD1-Tg skeletal muscle ( Figure 3 ) depends on their associated differentiated progeny , in which the TEAD1 transgene is expressed via the MCK promoter , i . e . the myofiber . To test this prediction , we performed in vitro myoblast culture experiments either in the absence or presence of the associated TEAD1-Tg myofiber ( Figure 9D–U ) . First , we cultured primary myoblasts derived from either TEAD1-Tg or Wt hind limb muscles ( Figure 9D–M ) . We detected no differences in the differentiation or proliferation capacities between them . Upon serum starvation , myoblasts from either genotype readily differentiated as indicated by early ( MyoD ) , late ( Myogenin ) , and terminal ( myosin heavy chain ) differentiation marker expression ( Figure 9D–I; quantified inFigure 9J ) . Proliferation assays using EdU incorporation , also revealed no differences between primary myoblasts from TEAD1-Tg and from Wt muscles ( Figure 9K–L; quantified in Figure 9M ) . To determine if the TEAD1-Tg myofiber affects proliferation of its attached cohort of SCs , we cultured single myofibers from EDL muscles of TEAD1-Tg and Wt littermates and determined SC proliferation via cumulative marking of Pax7+ cells in S-Phase via EdU ( Figure 9N–O; quantification in Figure 9P ) . The majority ( ~92% ) of SCs on Wt fibers proliferated . The proportion of proliferating SCs was significantly increased on TEAD1-Tg myofibers , as all Pax7+ cells are positive for EdU ( Figure 9P ) . No statistically significant differences in apoptotic rates were found via anti-Caspase-3/Pax7 co-IF staining ( ~0 . 8±1 . 4% versus ~3 . 2±1% Caspase3+/Pax7+ cells on Wt and TEAD1-Tg fibers , respectively; p>0 . 05 ) . To determine the effect of the increased proliferation rate on SC clone formation , we quantified SCs on single myofibers at t=0 hr , and number of clones as well as clone size at t=72 hr on both Wt and TEAD1-Tg myofibers ( Figure 9Q–R ) . As expected , we found significantly increased numbers of SCs associated with freshly isolated TEAD1-Tg compared to Wt myofibers ( Figure 9Q ) . At t=72 hr , the number of SC clones vastly exceeded that of the starting number of SCs for both Wt and TEAD1-Tg samples suggesting that some clones split into 2 or more clones ( Figure 9Q ) . Yet , significantly greater numbers of cells were found per clone on TEAD1-Tg compared to Wt fibers ( Figure 9R ) . This evidence suggests that increased proliferation yields larger clones on TEAD1-Tg cultured myofibers rather than fusion of clones derived from multiple SCs , as clone fracturing occurs more frequently than clone fusion and at similar rates between Wt and TEAD1-Tg samples ( Figure 9N–R ) . We next assayed cells of these clones for differentiating ( MyoD+ ) , expanding ( MyoD+/Pax7+ ) , and self-renewing ( Pax7+ ) cell fates ( Figure 9S–T ) . Quantification revealed that hyper-proliferation of SCs on TEAD1-Tg myofibers does not alter the relative distributions of these cell fates , but instead involves a proportional increase in the sizes of each fraction ( Figure 9U ) . These data further confirm the SC self-renewal capacity in vivo ( Figure 7 ) . A model of increased proliferation yielding a larger number of progeny , and consequently more SCs , is consistent with our in vivo postnatal proliferation rates ( Figure 3 ) and explains the hyperplasia phenotype . We conclude that SC hyperplasia of TEAD1-Tg skeletal muscle is a non-cell autonomous phenomenon , which involves the coordinated scaling of the expanding , differentiating , and self-renewing myogenic cell populations .
SC hyperplasia in TEAD1-Tg mice is universal among the muscle groups analyzed , including pre-dominantly fast- or pre-dominantly slow-twitch muscle groups . Since the TEAD1-Tg features a transition to the slow-twitch muscle contractile phenotype , i . e . loss of type IIX fibers ( Figure 1—figure supplement 1; Tsika et al . , 2008 ) , the pervasiveness of the SC hyperplasia among the different muscles analyzed is particularly noteworthy . Higher SC densities have been found in slow-twitch compared to fast-twitch muscles ( Gibson and Schultz , 1983; Schmalbruch and Hellhammer , 1977; Collins et al . , 2005 ) . As such , the fast- to slow-twitch fiber transition may account for the SC hyperplasia in TEAD1-Tg mice . However , the pre-dominantly slow-twitch soleus muscle , which is devoid of type IIX fibers , features the largest increase ( 6-fold ) in SCs in TEAD1-Tg mice . Hence , the increase in SCs cannot solely be a ‘passenger effect’ of the fast- to slow-twitch fiber transition . The SC hyperplasia originates at the time of peak TEAD1 transgene expression around PN 14d ( Figure 3 ) and eventually features normal quiescent adult SCs indistinguishable from their Wt counterparts ( Figure 4 ) . In particular , we utilized this new mouse model of SC hyperplasia to investigate whether an increase in the physical number of stem cells had any impact on skeletal muscle regeneration and discovered faster regeneration kinetics in TEAD1-Tg mice ( Figures 5 and 6 ) . The SC hyperplasia persists through multiple and chronic injuries ( Figures 7 and 8 ) and notably , derives from cell-cell signaling between the myofiber and the SCs ( Figure 9 ) . All together , these features uniquely prime the TEAD1-Tg mouse model for the future discovery of the elusive signaling pathway ( s ) regulating muscle stem cell number . Though SC hyperplasia is not well studied , other genetic or pharmacologically induced mouse models have reported an expansion of the SC pool in adult mice . For example , skeletal muscle mutant for Collagen VI has an ~2 . 4-fold increase in SCs . As mutations in this gene cause myopathy or muscular dystrophy in humans ( Lampe and Bushby , 2005 ) , Collagen VI mutant mice also feature dystrophic muscle , which leads to SC activation as evidenced by increased proliferation and apoptosis ( Urciuolo et al . , 2013 ) . Hence an increase in SCs is also observed in mdx mice ( Boldrin et al . , 2009 ) . As such , this model features impaired regeneration and a failure to sustain the SC hyperplasia during injury-induced regenerative myogenesis ( Urciuolo et al . , 2013 ) . Likewise , ectopic application of Wnt7a in adult regenerative muscle drives an ~2-fold expansion of proliferative SCs . No accelerated kinetics of muscle repair were noted but muscles featured both hypertrophy and hyperplasia ( Le Grand et al . , 2009 ) . Subsequently , Wnt7a was shown to directly act on the myofiber for induction of hypertrophy ( von Maltzahn et al . , 2012a ) . The expansion of SCs by Wnt7a can be further increased by the co-application with fibronectin . Whether the additional SC increase impacted myofiber hypertrophy and/or hyperplasia was not reported ( Bentzinger et al . , 2013 ) . The source of Wnt7a in skeletal muscle is unresolved . Also , the Wnt7a-induced increase in SCs may be indirectly coupled to the myofiber hypertrophy , precluding a concrete conclusion about the signaling for the SC expansion . While informative , the above mouse models are not useful in elucidating the molecular regulation underlying initial scaling of SC pool size during the perinatal period . They feature many distinct aspects including a less robust increase of SCs , a non-developmental origin of SC increase , overall muscle tissue hypertrophy and/or failure to generate a stably quiescent pool of adult SCs . The TEAD1 transgene induced SC hyperplasia is of the greatest known magnitude to-date , stably established during early postnatal development and features a normal adult quiescent phenotype . The timing of the SC increase in TEAD1-Tg muscle is particularly noteworthy and should incite new investigations of the aforementioned mouse models during the early postnatal period to further unravel this fundamental 'stem cell scaling' process in skeletal muscle . Furthermore , while myofiber-specific TEAD1 overexpression reveals a surprising plasticity of the muscle stem cell niche to harbor greatly increased SCs , we do not know if there are any functional requirements for the Tead transcription factor family in the regulation of SC number . Further studies of Tead loss-of-function mouse models are needed . The activity of early postnatal SCs is markedly different from their adult counterparts . Adult SCs are kept in a quiescent state , while early postnatal SCs are highly proliferative providing differentiated myonuclei for muscle growth as they progressively exit the cell cycle to be set aside as quiescent stem cells during the first three weeks after birth ( Lepper et al . , 2009; White et al . , 2010 ) . The signals that orchestrate the SC transition from the proliferative to the quiescent state are unknown but likely entail both ‘proliferation’ and ‘quiescence’ cues from the local niches leading to the formation of an adult muscle stem cell pool of defined size . Super-numeral SCs and increased SC proliferation were first detected in TEAD1-Tg muscle during this early postnatal period at 2 weeks after birth ( Figure 3 ) . Since TEAD1-Tg muscle is not hypertrophic and does not accumulate more nuclei than Wt ( Figures 1 and 2 ) , a 'Stop' signal must exist , and be still intact in TEAD1-Tg mice to prevent further fusion by excessive SCs . Thus , we propose that the extra rounds of SC proliferation are a selectively symmetric expansion of the stem cell pool . The likely mechanism driving the SC hyperplasia is over-expression or prolonged expression of physiologically normal ‘proliferation’ factor ( s ) , or reduced expression of ‘quiescence’ factor ( s ) in early postnatal TEAD1-Tg muscle . Therefore , this animal model may also be useful to uncover the physiological signaling underlying the complex transition from a proliferative juvenile progenitor to a quiescent adult stem cell . Applying EdU to monitor SC proliferation both in vivo ( Figure 7 ) and in vitro ( Figure 9 ) , we found super-numeral SCs of TEAD1-Tg muscle to be self-renewing stem cells . This conclusion is further supported by the observation that the TEAD1 transgene induced SC hyperplasia is maintained even over several injury-induced regeneration cycles ( Figure 7 ) . Of note , the magnitude of the hyperplasia is slightly lessened with multiple bouts of regeneration . It is possible that the SC self-renewal capacity is slightly reduced in TEAD1-Tg mice . It is also possible that the regenerating tissue lacks the ability to promote SC hyperplasia to the same degree as the early postnatal myofiber . The adult niche is destroyed by injury leading to a transient loss of the TEAD1-expressing cell , i . e . the muscle fiber . This is in contrast to the ‘intact’ early postnatal niche , which drives the establishment of super-numeral SCs in TEAD1-Tg mice and features intimate contact between the TEAD1-expressing muscle fiber and its SCs . Moreover , there appear to be differences in the regulation of proliferating early postnatal versus quiescent adult SCs . Temporally-controlled inactivation of Pax7 in SCs shortly after birth leads to an immediate dramatic loss in regenerative capacity while inactivation in the adult leads to a slow progressive loss of SCs , which is reflected by normal regeneration over the short-term and impaired regeneration over the long-term ( Günther et al . , 2013; Lepper et al . , 2009 ) . It is possible that early postnatal SCs are uniquely primed to receiving proliferation and quiescence cues from their local niches , which leads to the formation of a properly proportioned adult muscle stem cell pool . Such cues may be expressed at a higher level or for a longer period of time in the developing muscles than in adult regenerating muscles of TEAD1-Tg mice . Or , the adult SC niche may be more restrictive to proliferation , and thus limiting to the extent of re-establishment of the SC hyperplasia during regeneration . For example , the stiffness of the extracellular matrix from skeletal muscle has been demonstrated to increase with age ( Wood et al . , 2014 ) , and substrate elasticity has been shown to affect the regenerative competence of SCs ( Gilbert et al . , 2010 ) . Along this line of thought , it is interesting to note that neither developmental nor regenerative muscles of TEAD1-Tg mice produce hypertrophy despite the SC hyperplasia ( Figures 1 , 5 and 6 ) . It seems reasonable that upon acquiring the ‘correct’ number of myonuclei , to efficiently support the transcriptional and metabolic needs , the myofiber sends 'dominant' signal ( s ) to prevent additional fusion by SCs , and possibly to stop SC proliferation . A longer period of SC proliferation during early postnatal development versus adult regeneration could be the basis for the plasticity of the magnitude of the SC hyperplasia . The muscle histopathology found in the mdx mouse model was significantly mitigated in the TA muscle of mdx; TEAD1-Tg mice ( Figure 8 ) . This finding can likely be accounted for based on a TEAD1-induced slow-oxidative phenotype in both fast and slow-twitch muscles , which we determined previously by high-resolution electrophoretic separation and quantification of native myosin heavy chain isoforms and confirmed here by IF ( Tsika et al . , 2008 ) . This notion is consistent with previous studies demonstrating that slow-twitch oxidative fibers in human and murine dystrophic muscles are less susceptible to the degenerative outcome of muscular dystrophies than are fast-twitch fibers ( Moens et al . , 1993; Webster et al . , 1988; Consolino and Brooks , 2004 ) . More recent therapeutic efforts shown to mitigate the dystrophic pathology in the mdx model have utilized transgenic ( calcineurin , peroxisome proliferator-activated receptor-γ coactivator-1α , AMP-activated protein kinase , Kruppel-like factor 15 ) or pharmacological ( Resveratol , AICAR , GW501516 , Wnt7a ) mediators , all previously shown to remodel dystrophic muscle to reflect a greater proportion of slow-oxidative fiber-types ( Stupka et al . , 2006; Chakkalakal et al . , 2004; Handschin et al . , 2007; Hori et al . , 2011; Tabebordbar et al . , 2013; Ljubicic et al . , 2014; von Maltzahn et al . , 2012b; Morrison-Nozik et al . , 2015 ) . It is therefore conceivable that the TEAD1-Tg-induced slow-oxidative myofiber phenotype ameliorates dystrophic pathology in mdx muscle in similar fashion to the aforementioned studies . In particular , utrophin , which can functionally substitute for dystrophin , is elevated in slow-twitch when compared to fast-twitch muscle ( Gramolini et al . , 2001; Rafael et al . , 1998 ) . Elevated utrophin transcript levels in mdx; TEAD1-Tg muscle are a highly likely contributor to the betterment of the dystrophic pathology ( Figure 8 ) . Additional mechanisms are likely at play . Calcineurin is a direct transcriptional activator of utrophin and skeletal-muscle specific overexpression results in increased utrophin protein levels and results in amelioration of the dystrophic pathology ( Stupka et al . , 2006 ) . Yet , the extent of the observed amelioration is significantly smaller when compared to the mdx; TEAD1-Tg model . Additional properties of the myofiber affected by the TEAD1 transgene could further protect the fiber from degeneration . It is also possible that the increased number of SCs in mdx; TEAD1-Tg muscle contributes to a more speedy repair process , thereby stabilizing the dystrophic muscle . Or , both mechanisms may contribute . Additionally , it would be of great interest to quantify SCs in the murine models above to determine if their numbers are increased as well , which could also potentially contribute to an ameliorated disease state . It is conceivable that the size of the resident stem cell pool could be modulated with , and thus , be directly linked to the size of tissue , for example via physical expansion or reduction of available stem cell niches . To extend this idea , it is of interest to consider the exquisite molecular mechanisms that have evolved to initially establish and then maintain proper skeletal muscle tissue size . Signaling pathways including Hippo , IGF-1 , Wnt , calcium ( via NFAT ) , and TGF-β contribute to the regulation of myofiber size and number . In particular , the TGF-β family member myostatin and its receptor activin receptor type-IIB ( ActRIIB ) play prominent roles , as both genetic inactivation of myostatin and ActRIIB antagonism result in a double muscling phenotype in mice with both hyperplasia and hypertrophy ( McPherron et al . , 1997; Zhou et al . , 2010; Lee and McPherron , 2001 ) . However , it remains controversial whether a corresponding SC increase accompanies and is required for the increase in muscle mass by inhibition of myostatin signaling ( McCroskery et al . , 2003; Wang and McPherron , 2012; Zhou et al . , 2010; Lee et al . , 2012 ) . Inactivation of ActRIIB specifically in the myofiber causes myofiber hypertrophy arguing against myofiber size regulation being dependent on the SC pool ( Lee et al . , 2012 ) . Likewise , myostatin inhibition has been shown to result in skeletal muscle hypertrophy in the absence of satellite cell activity ( Amthor et al . , 2009 ) . Canonical Wnt signaling has been implicated in regulating myogenic progenitor and myofiber number during fetal myogenesis and muscle repair after injury ( Hutcheson et al . , 2009; Murphy et al . , 2014; Rudolf et al . , 2016 ) . Similarly , calcium signaling via NFAT proteins has been implicated in regulating skeletal muscle size ( Schulz and Yutzey , 2004 ) . In our previous work , we found both Wnt and NFAT signaling to be significantly reduced in TEAD1-Tg muscle ( Tsika et al . , 2008 ) . Yet , we did not find muscle tissue size to be altered ( Figure 1 ) . This apparent discrepancy can likely be attributed to the late onset of MCK promoter driven TEAD1 transgene expression , which happens after muscle differentiation . Thus , neither differentiation nor fusion is affected in these mice . And the non-autonomous induction of super-numeral SCs results in the uncoupling of the molecular regulation of SC number from muscle size in TEAD1-Tg mice . Similarly , muscle groups of different ontology appear to maintain different SC to muscle nuclei ratios , i . e . branchiomeric muscles have significantly fewer SCs compared to somite-derived muscles ( Ono et al . , 2010 ) . Still , muscle fibers and their associated cohort of SCs have co-evolved to maintain a relatively fixed ratio . While the evolutionary selection pressures for both processes are unknown they are of great biological curiosity .
hTEAD1 transgenic mice were previously described ( Tsika et al . , 2008 ) . Age-matched C57BL/6 siblings served as Wt controls . mdx mice ( ID: 001801 ) were obtained from the Jackson Laboratory ( ME ) . Injury with cardiotoxin ( 10 µM , Sigma C9759 ) to the TA ( after anesthesia ) used 50 µl for adult ( >2 months ) mice . For injury by BaCl2 ( after anesthesia ) , TA muscles were injected with 25 µl of 1 . 2% ( w:v ) Barium Chloride ( Sigma 217565 ) in PBS . For multiple rounds of injury , TA muscles were injected with 25 μl of BaCl2 for each injury and allowed to regenerate for 35 days between injuries . EdU ( Invitrogen; CA ) was administered as previously described ( Lepper et al . , 2009 ) . BrdU ( 5-bromo-2’-deoxyuridine , Sigma B5002 ) was provided in drinking water ( 0 . 8 mg/mL ) for 30 consecutive days . All procedures were approved by IACUC . For adult animals , tail DNA was used for genotyping by PCR . For perinatal mice , toe DNA was used . DNA was extracted using the ExtractN’Amp kit ( Sigma XNAT2 ) following the manufacturer’s instructions . PCR reactions were carried out using GoTaq polymerase ( Promega M8291 ) with buffers supplied by the manufacturer with 0 . 1 mM dNTPs and 2 . 5 mM MgCl2 . PCR products were resolved in a 2% agarose gel , stained with 0 . 5 µg/mL ethidium bromide ( Gibco 15585011 ) , and digitally imaged with a Bio-Rad Gel Doc system for record keeping . Primer sequences , product sizes , and PCR conditions are in Table 1 . 10 . 7554/eLife . 15461 . 013Table 1 . PCR genotyping and RT primers , and conditions . DOI: http://dx . doi . org/10 . 7554/eLife . 15461 . 013GeneForward primerReverse primerSizeTEAD-HA5’-ATCCATGCTTGTTACCTTCAG-3’5’-ACTACAAGGACGATGACAAG-3’460 bpInternal Control5’-CAGCTCTACATCACCTGCCA-3’5’-CACTGGGAAGAGACACTCAG-3’520 bpPCR conditionsDenature: 94°C for 30 s Anneal: 56°C for 30 s Extend: 72°C for 60 s ( 34 cycles ) Dystrophin ( WT ) 5’-GCGCGAAACTCATCAAATATGCGTGTTAGTGT-3’5’-GATACGCTGCTTTAATGCCTTTAGTCACTCAGATAGTTGAAGCCATTTTG-3’134 bpDystrophin ( mdx ) 5’-GCGCGAAACTCATCAAATATGCGTGTTAGTGT-3’5’-CGGCCTGTCACTCAGATAGTTGAAGCCATTTTA-3’117 bpPCR conditionsDenature: 94°C for 20 s Anneal: 60°C for 20 s Extend: 72°C for 20 s ( 5 cycles ) Denature: 94°C for 20 s Anneal: 64°C for 20 s Extend: 72°C for 20 s ( 23 cycles ) utrophin5’-AGTATGGGGACCTTGAAGCC-3’5’-CGAGCGTTTATCCATTTGGT-3’cycloA5’-ATTTCTTTTGACTTGCGGGC-3’5’-AGACTTGAAGGGGAATG-3’actin5’-CCCTAAGGCCAACCGTGAA-3’5’-CAGCCTGGATGGCTACGTACA-3’ Total RNA was extracted and purified from TA muscles using the Direct-zol RNA MiniPrep Kit ( ZymoResearch , R2051 ) . Reverse transcription was performed using M-MLV Reverse Transcriptase ( Thermo Scientific , 28025013 ) . Real Time PCR was performed using the BioRad CFX96 Real-Time System and primers listed in Table 1 . Data analysis was performed using the CFX manager . Utrophin expression was normalized to both actin and cyclophilin A expression . For IF staining , muscles were isolated , partially imbedded in tragacanth ( Sigma 9000-65-1 ) on a slice of cork , and flash frozen in isopentane ( Sigma 78–78-4 ) cooled by liquid nitrogen . Samples were cryo-sectioned at 10 µm and mounted on Superfrost Plus slides ( VWR 48311–703 ) . Following fixation for 10 min ( 4% PFA/PBS ) on ice and PBS wash , antigen retrieval was performed ( Dako S1699 ) . Slides were placed in a solution of Sudan black dye ( 0 . 1% , Sigma 199664 ) in ethanol ( 70% ) for 20 min to lessen autofluorescence . Samples were then blocked with 10% normal goat serum diluted in 0 . 02% triton-100/PBS ( PBT ) . Subsequent primary and secondary antibodes were used at concentrations described ( Tables 2 and 3 ) diluted in 2% FBS in 0 . 02% PBT for one hour to overnight in a humidified chamber . TUNEL labeling was performed immediately following the IF staining protocol using the ApopTag Red In Situ Apoptosis Detection Kit ( Millipore , S7165 ) . Coverslips were mounted with a drop of mounting media ( Vector Laboratories , H1200 ) and sealed . All IF images were obtained using Nikon E800 and Zeiss Axioscop microscopes ( described below ) . 10 . 7554/eLife . 15461 . 014Table 2 . Primary antibodies used for immunofluorescence . The monoclonal antibodies for Pax7 , Myog , Myosin , IIx type myosin , IIa type myosin , non-IIx type myosin , I type myosin , and embryonic myosin , developed by Stefano Schiaffino , C Lucas , A Kawakami , WE Wright , DA Fischman , and HM Blau were obtained from the Developmental Studies Hybridoma Bank , created by the NICHD of the NIH and maintained at The University of Iowa , Department of Biology , Iowa City , IA 52242 . DOI: http://dx . doi . org/10 . 7554/eLife . 15461 . 014AntibodiesHostDilutionSourceAnti-Pax7mouse ( IgG1 ) 1:4-5DSHB Pax7 ( RRID:AB_528428 ) Anti-Lamininrabbit1:3000Sigma L9393 ( RRID:AB_477163 ) Anti-HArabbit1:1000Invitrogen PA1-985 ( RRID:AB_559366 ) Anti-HArabbit1:100Sigma H6908 ( RRID:AB_260070 ) Anti-HAmouse ( IgG1 ) 1:100Covance mms-101P ( RRID:AB_2314672 ) Anti-B1 Integrinrat1:200Abcam AB95623 ( RRID:AB_10676803 ) Anti-Mcadherinmouse1:200Santa Cruz SC81471 ( RRID:AB_2077111 ) Anti-Calcitonin Receptorrabbit1:250AbD Serotec AHP635 ( RRID:AB_2068967 ) Anti-CD34rat1:25BD Biosciences 553731 ( RRID:AB_395015 ) Anti-Dystrophinmouse1:1000Genetex GTX27164 ( RRID:AB_386029 ) Anti-Dystrophinrabbit1:100Abcam AB15277 ( RRID:AB_301813 ) Anti-MyoDrabbit1:50Santa Cruz SC760 ( RRID:AB_2148870 ) Anti-MyoDmouse ( IgG1 ) 1:200Santa Cruz SC32758 ( RRID:AB_627978 ) Anti-Myogeninmouse1:20DSHB F5D ( RRID:AB_528355 ) Anti-Myogeninmouse ( IgG1 ) 1:200Santa Cruz SC12732 ( RRID:AB_627980 ) Anti-MyHCmouse1:20DSHB MF20 ( RRID:AB_2147781 ) Anti-eMyHCmouse ( IgG1 ) 1:400DSHB F1 . 652 ( RRID:AB_528358 ) Anti-myosin ( IIX ) mouse ( IgM ) 1:2DSHB 6H1 ( RRID:AB_2314830 ) Anti-myosin ( non-IIX ) mouse ( IgG1 ) 1:2DSHB BF-35 ( RRID:AB_2274680 ) Anti-myosin ( IIA ) mouse ( IgG1 ) 1:2DSHB SC-71 ( RRID:AB_2147165 ) Anti-myosin ( I ) mouse1:2DSHB BA-D5 ( RRID:AB_2235587 ) Anti-cleaved caspase 3rabbit1:100Cell Signaling 9664S ( RRID:AB_2070042 ) 10 . 7554/eLife . 15461 . 015Table 3 . Secondary antibodies used for immunofluorescence . DOI: http://dx . doi . org/10 . 7554/eLife . 15461 . 015HostAntigenFluorophoreDilutionSourceGoatRabbit IgGAlexa 4881:1000Invitrogen A11034 ( RRID:AB_2576217 ) GoatMouse IgGAlexa 4881:1000Invitrogen A11001 ( RRID:AB_2534069 ) GoatMouse IgG1Alexa 4881:100 - 1:1000Invitrogen A21121 ( RRID:AB_2535764 ) GoatRabbit IgGAlexa 5681:1000Invitrogen A11011 ( RRID:AB_2534078 ) GoatRabbit IgGAlexa 5461:100Invitrogen A11035 ( RRID:AB_2534093 ) GoatMouse IgGAlexa 5681:1000Invitrogen A11031 ( RRID:AB_2534090 ) GoatRat IgGAlexa 5461:200Invitrogen A11081 ( RRID:AB_2534125 ) GoatMouse IgMAlexa 5941:1000Invitrogen A21044 ( RRID:AB_2535713 ) EdU was injected at 0 . 1 mg/20g bodyweight . EdU was detected in muscle sections via Click-iT Kit ( Invitrogen C10640 ) . BrdU was detected following IF staining by antigen retrieval ( BD biosciences 550803 ) and ABC amplification ( Vector Laboratories PK4000 ) . In cultured cells grown on chamber slides ( Sigma C7182 ) coated in matrigel ( Fisher CB-40234 ) , EdU was added to a concentration of 10 µM for 45 min and detected by Click-iT Kit following IF procedures . In experiments of EdU labeled fiber-associated cell clones , EdU ( 10 µM ) was provided throughout the culturing period beginning at t=24 hr . H and E staining has been previously described ( Lepper et al . , 2009 ) . Sections were stained with Sirius Red using protocol and reagents of the Picro-Sirius Red staining kit ( American MasterTech KTPSRPT ) following fixation in Bouin’s solution ( Sigma HT01032 ) for 30 min at 56°C . Trichrome stains were performed according to manufacturer’s instructions using the Gomori’s One-Step Trichrome Kit ( Polysciences Inc . 24205 ) . All histology was imaged on the Zeiss Stemi SV11 ( for low magnification ) and Nikon E800 DIC microscopes ( for high magnification ) described below . Evans Blue dye ( EBD , Sigma E2129 ) was administered to mice at a 1% concentration in PBS via intra-peritoneal injection ( 10 μL/g bodyweight ) . Mice were sacrificed 20 hr later and the TA and EDL were harvested , flash frozen , then sectioned at 8 μm . Sections were fixed in cold acetone ( −20°C ) for 10 min , rinsed with PBS , and mounted . The dye’s fluorescence was visualized via red light excitation . Muscle size was determined by weight ( Table 4 ) and by measurements and quantifications of muscle fibers . Cross-sections ( by cryostat sections at 10 µm ) of muscles were stained for dystrophin by IF or subjected to H and E stains ( for mdx muscle samples ) . Using ImageJ software , fibers were manually outlined and then measured via ImageJ for area and minimum ferret diameter . 10 . 7554/eLife . 15461 . 016Table 4 . Weights ( mgSD ) of respective muscle groups in Wt vs . TEAD1-Tg mice . TEAD1-Tg hind limb muscles are equivalent in weight to Wt . Weight measurements ( mg ) and standard deviation for TA , EDL , Soleus , and Plantaris muscles from TEAD1-Tg or Wt mice . By t-test no significant difference was found between genotypes ( n > 5 muscles for all measurements ) . DOI: http://dx . doi . org/10 . 7554/eLife . 15461 . 016TAEDLSoleusPlantarisWt41 . 0 ± 5 . 59 . 3 ± 1 . 98 . 3 ± 1 . 416 . 5 ± 2 . 5TEAD1-Tg38 . 6 ± 3 . 49 . 5 ± 1 . 59 . 7 ± 0 . 914 . 3 ± 2 . 7 Genotyped animals were killed by cervical dislocation and hind limb muscles were removed and minced with scissors . Samples were digested with collagenase and dispase as described previously ( Springer et al . , 2002 ) . Dissociated muscles were then transferred to 15 mL conical tubes with 10 mL myoblast media ( 20% fetal bovine serum/5% horse serum in DMEM ) , spun down in a clinical centrifuge , resuspended in myoblast media , and passed through a 70 µm filter . The remaining cells were twice pre-plated onto uncoated tissue culture dishes for 30 min to remove fibroblasts . The remaining cells were quantified by hemocytometer and placed at equal numbers into wells of chamber slides ( Sigma C7182 ) coated with matrigel ( Fisher CB-40234 ) . These cells were differentiated for 2–5 days in media containing 2% horse serum in DMEM . For IF , these cells were fixed with prewarmed 4% PFA/PBS for 10 min , then permeabilized for 15 min in 0 . 5% PBT and washed with PBS . Primary and secondary antibodies were diluted in goat blocking buffer ( blocking powder [Perkin Elmer P1012] in normal goat serum [Genetex GTX73245] ) and applied for one hour to overnight with PBS washes in between primary and secondary antibody applications . After washing , the chamber portion was removed from the slide , and coverslips were mounted using a drop of fluoromount mounting media ( Fisher OB100-01 ) containing DAPI . Genotyped animals were killed by cervical dislocation and EDL muscles were placed in a solution of 0 . 2% collagenase/DMEM for 90 min in a 37°C shaking ( 80 rpm ) water bath . Digested muscles were placed in a pre-warmed solution of DMEM on horse serum coated tissue culture dishes for one hour . The muscles were then titurated with a large bore glass pipette to loosen outer myofibers , which were collected and moved to a new horse serum-coated dish of 10% FBS/DMEM or growth media ( 20% FBS/ 5% horse serum/ DMEM ) via a small bore glass pipette . This continued until sufficient myofibers were obtained . Myofibers were incubated 48 or 72 hr in media ( replaced daily ) . Myofibers were fixed in pre-warmed 4% PFA/PBS for 10 min . The myofibers were then washed three times in PBS and moved to a 1 . 5 mL microfuge tube . IF of isolated myofibers proceeded as follows: permeabilization ( 0 . 5% PBT ) occurred for 15 min , myofibers were washed with PBS , then primary and secondary antibodies in goat blocking buffer were applied for one hour to overnight with PBS washes in between primary and secondary antibody applications . Fibers were mounted on Superfrost Plus slides in a drop of mounting media containing DAPI , covered with a coverslip , and sealed with nail polish . Error bars in histograms represent standard deviation ( SD ) over either the mean or fold change . Using an excel spreadsheet , all data was subjected to a two-tailed t test or chi square test ( indicated in figure legend ) when appropriate to determine statistical differences . Statistical significance was defined as a p value <0 . 05 . Sample size was predetermined based on published SC counts , which reveal very low variability , and on preliminary data revealing a highly robust and large change in the number of SCs in TEAD1-Tg mice . The Nikon E800 upright epifluorescence microscope uses a 100 watt mercury arc lamp fluorescent source . Images were taken with Hamamatsu Orca-Flash 4 . 0 LT sCMOS camera under objectives: Plan Fluor 10x ( NA 0 . 30 ) , 20x ( NA 0 . 50 ) , 40x ( NA 0 . 75 ) . The Zeiss Axioskop upright epifluorescence microscope uses an 89North PhotoFluor metal halide fluorescent source . Images were taken with a Zeiss AxioCam monochrome CCD camera under objectives: Plan-Neofluar 10x ( NA 0 . 30 ) , 20x ( NA 0 . 50 ) , Plan-Neofluar 40x ( NA 0 . 75 ) . The Zeiss Stemi SV11 dissecting microscope was used with a Canon EOS Rebel T1i DSLR camera under a 1x objective . The Nikon E800 upright microscope w/ DIC optics was used with a Canon EOS Rebel T3i DSLR camera under objectives: Plan Fluor 4x ( NA 0 . 13 ) , Plan Apo 10x NA 0 . 45 and Plan Fluor 40x ( NA 1 . 30 Oil ) . | Skeletal muscles are primarily composed of cells called muscle fibers , which attach to bones via tendons . These muscle fibers contract to help move the body . Muscle also contains a population of muscle stem cells that repair injured tissue . Normally , in adult skeletal muscle , these stems cells are in a resting state . However , upon injury , the stem cells become activated , divide to increase in number and then develop into new muscle fibers to replace those that were damaged . The balance between the number of stem cells and the size of the muscle must be tightly regulated to ensure that there are enough stem cells to fully regenerate the tissue after injury . However , little is known about how tissues keep their number of stem cells in proportion with their overall size . Previous attempts to make mice with more muscle stem cells invariably also created mice with larger muscles overall . This raised the question: is it possible to increase the numbers of stem cells without changing the size of the muscle ? Now , Southard , Kim et al . show it is possible and report that mice engineered to overproduce a protein called Tead1 in their muscle fibers have up to 6-times more stem cells yet normally sized muscles . Tead1 is a transcription factor that controls the activity of a number of genes as part of a major signaling pathway . The stem cells in mice that overproduce Tead1 began to increase in number two weeks after the mice were born because they went through additional rounds of cell division before they entered the resting state . Further experiments then showed that having more stem cells meant that the muscles were repaired more quickly after an injury . Additionally , when mice with extra Tead1 had a mutation that normally leads to muscle wasting , experiments showed that the progression of the disease was stunted . Southard , Kim et al . also show that the muscle fibers that are directly attached to the muscle stem cells are needed for the stem cells to increase in number in the Tead1-overexpressing mice . Together these findings suggest that a signal from the muscle fiber to its stem cells regulates the size of the stem cell population in the tissue . The next challenge is to uncover the molecule ( or molecules ) that signals from the muscle fiber to the stem cells and to gain deeper insight into how the Tead1 protein can counteract the effects of a muscle wasting disease . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"stem",
"cells",
"and",
"regenerative",
"medicine"
] | 2016 | Myofiber-specific TEAD1 overexpression drives satellite cell hyperplasia and counters pathological effects of dystrophin deficiency |
Eukaryotic genomes are organized dynamically through the repositioning of nucleosomes . Isw2 is an enzyme that has been previously defined as a genome-wide , nonspecific nucleosome spacing factor . Here , we show that Isw2 instead acts as an obligately targeted nucleosome remodeler in vivo through physical interactions with sequence-specific factors . We demonstrate that Isw2-recruiting factors use small and previously uncharacterized epitopes , which direct Isw2 activity through highly conserved acidic residues in the Isw2 accessory protein Itc1 . This interaction orients Isw2 on target nucleosomes , allowing for precise nucleosome positioning at targeted loci . Finally , we show that these critical acidic residues have been lost in the Drosophila lineage , potentially explaining the inconsistently characterized function of Isw2-like proteins . Altogether , these data suggest an ‘interacting barrier model , ’ where Isw2 interacts with a sequence-specific factor to accurately and reproducibly position a single , targeted nucleosome to define the precise border of phased chromatin arrays .
Chromatin consists of the nucleic acids and proteins that make up the functional genome of all eukaryotic organisms . The most basic regulatory and structural unit of chromatin is the nucleosome . Each nucleosome is defined as an octamer of histone proteins , which is wrapped by approximately 147 base pairs of genomic DNA ( Luger et al . , 1997; Kornberg , 1974 ) . The specific positioning of nucleosomes on the underlying DNA can have significant effects on downstream processes , such as promoter accessibility and molecular recruitment , which ultimately serve to alter gene expression ( Lai and Pugh , 2017 ) . Despite decades of research , the mechanisms leading to precise nucleosome locations in cells are still being defined . Nucleosome positioning is dynamically established by a group of enzymes known as ATP-dependent chromatin remodeling proteins ( ChRPs ) ( Zhou et al . , 2016 ) . Extensive biochemical and structural characterization has been performed on this group of proteins from various families ( Clapier et al . , 2017 ) . The chromodomain-helicase-DNA binding ( CHD ) and imitation switch ( ISWI ) families of ChRPs have been characterized as nonspecific nucleosome sliding and spacing factors in vitro ( Stockdale et al . , 2006; Hauk et al . , 2010; McKnight et al . , 2011; Kagalwala et al . , 2004; Lusser et al . , 2005; Tsukiyama et al . , 1999; Pointner et al . , 2012 ) . In yeast , flies , and mammals , ChRPs generate evenly spaced nucleosome arrays at transcription start sites and organize genomic chromatin at other defined boundaries ( Pointner et al . , 2012; Lee et al . , 2007; Mavrich et al . , 2008a; Valouev et al . , 2011; Krietenstein et al . , 2016; Wiechens et al . , 2016; Baldi et al . , 2018; Gkikopoulos et al . , 2011; Zhang et al . , 2011 ) . However , relatively little is known about the in vivo biological regulation of these spacing factors , and it is not understood how they can accurately and reproducibly position nucleosomes throughout the genome in different cellular contexts . A widely accepted model is that ChRPs pack nucleosome arrays against a noninteracting barrier , such as an unrelated DNA binding protein or another nucleosome ( Krietenstein et al . , 2016; Zhang et al . , 2011; Mavrich et al . , 2008b ) . In this way , general regulatory factors ( GRFs ) could establish chromatin landscapes with differing nucleosome arrays in response to changes in the cellular environment . In support of this model , nucleosome arrays near GRFs and other DNA binding elements appear to be phased relative to the binding motifs of the sequence-specific DNA binding factors in cells and in biochemically reconstituted cell-free systems ( Krietenstein et al . , 2016; Baldi et al . , 2018; Yan et al . , 2018 ) . This model suggests that boundaries of nucleosome arrays are determined by the binding of barrier factors . Implicit in this barrier model are the assumptions that ChRPs act as nonspecific nucleosome spacing machines throughout the genome and that specific ChRP and GRF interactions are not required to establish nucleosome positions . While this model provides a good explanation for how phased nucleosome arrays can be established throughout the genome by a combination of DNA binding factors and nonspecific chromatin remodeling factors , the fundamental assumptions of the barrier model have not been thoroughly tested . It has been shown through genetic and recent biochemical experiments that members of the ISWI family of ChRPs functionally interact with transcription factors in vivo ( Krietenstein et al . , 2016; Gelbart et al . , 2005; Goldmark et al . , 2000; Fazzio et al . , 2001; Yadon et al . , 2013 ) . One of the most well-defined interacting partners of ISWI proteins is the meiotic repressor unscheduled meiotic gene expression ( Ume6 ) , which is found in yeasts . It has been previously demonstrated that Ume6 and Isw2 , an ISWI-containing ChRP complex in Saccharomyces cerevisiae ( homologous to the ATP-dependent chromatin assembly factor [ACF] complex in humans and flies ) , share genetic targets of repression and likely interact physically ( Goldmark et al . , 2000 ) . While interactions with sequence-specific DNA binding proteins can potentially determine precise nucleosome targeting and final nucleosome positions ( Donovan et al . , 2019; Bowman and McKnight , 2017; McKnight et al . , 2016 ) , the mechanisms through which physical interactions between Isw2 and any genomic recruitment factor like Ume6 influence nucleosome positioning activity in cells have not been defined . For example , it is not known how these physical interactions occur or what role they play in the biochemical outcomes of chromatin remodeling reactions and the resulting downstream biological outputs . In this work , we have successfully identified the mechanism of interaction between Isw2 and Ume6 in S . cerevisiae . By taking a protein dissection approach combined with genome-wide nucleosome profiling , we have identified a previously uncharacterized helical domain in Ume6 that allows for Isw2 binding , specific genomic recruitment , and precise nucleosome positioning outcomes . We further demonstrate that conserved attributes of this helical domain are observed in the cell cycle regulator Swi6 , which we have identified as a new Isw2-recruitment adapter protein that allows for specific nucleosome positioning at Mbp1/Swi6 ( MBF ) and Swi4/Swi6 ( SBF ) targets . We have also determined that the transcription factor-interacting interface of Isw2/ACF-like remodeling complexes contains a few key and highly conserved residues within the WAC ( WTSF/Acf1/cbp146 ) domain . Finally , we show that these residues , which are essential for directional , sequence-specific remodeling , were lost in the evolution of the Drosophila lineage , where extensive biochemical , genetic , and genomic characterization has been performed on the ITC1 ortholog ACF .
We wished to understand how the conserved Isw2 protein complex in yeast behaves genome-wide and at specific promoter nucleosomes at target sites . Yeast Isw2 has been characterized extensively in biochemical assays , which all suggest that it has nonspecific DNA binding , ATP hydrolysis , nucleosome sliding , mononucleosome centering , and nucleosome spacing activities ( Stockdale et al . , 2006; Kagalwala et al . , 2004; Lusser et al . , 2005; Tsukiyama et al . , 1999; Dang and Bartholomew , 2007; Dang et al . , 2006; Hota et al . , 2013; Kassabov et al . , 2002; Zofall et al . , 2004; Zofall et al . , 2006 ) . These nonspecific nucleosome mobilizing activities suggest that the Isw2 protein should be able to organize nucleosome arrays against a barrier across the genome in yeast cells since ( 1 ) it is estimated that there are enough Isw2 molecules for every 10–20 nucleosomes in the genome ( Gelbart et al . , 2005 ) , ( 2 ) Drosophila melanogaster ACF , an Isw2 ortholog , can organize nucleosomes into evenly spaced arrays ( Baldi et al . , 2018 ) , and ( 3 ) other nonspecific and related nucleosome spacing factors can globally space nucleosomes across the genome in yeast and other organisms ( Pointner et al . , 2012; Wiechens et al . , 2016; Gkikopoulos et al . , 2011; Zhang et al . , 2011 ) . To first determine how Isw2 positions nucleosomes in S . cerevisiae , we examined nucleosome positioning activity in an isw1/chd1 deletion background to remove known and potentially overlapping global spacing factors and highlight ‘isolated positioning activity’ by Isw2 . When examining the positioning of nucleosomes with and without Isw2 at all yeast pre-initiation complex sites ( PICs ) , it is evident that Isw2 activity is specialized at only a subset of target sites ( Figure 1A ) . As seen previously ( Gkikopoulos et al . , 2011; Ocampo et al . , 2016 ) , no global nucleosome spacing or organizing activity is detected by Isw2 alone ( Figure 1—figure supplement 1 ) . Close inspection of Isw2-targeted PICs suggests that Isw2 can only organize a single PIC-proximal nucleosome , while subsequent nucleosomes become more poorly phased as the distance from the initially positioned nucleosome increases ( Figure 1A , Figure 1—figure supplement 2 ) . Importantly , the PICs that display specific Isw2-directed activity are bound by Isw2 , while those lacking any detectable nucleosome organization by Isw2 are unbound ( Figure 1A , middle panel ) . It has been shown that Isw2 associates with sequence-specific DNA binding factors , such as the transcriptional repressor Ume6 ( Goldmark et al . , 2000; Fazzio et al . , 2001 ) . Isw2 activity at Ume6-bound loci has been previously characterized as precise , with Isw2 reproducibly moving nucleosomes until the predicted edge of the nucleosome core particle is 30 base pairs from the center of the Ume6 binding motif ( McKnight et al . , 2016 ) . Because of the connection to Ume6 , we examined nucleosome positions in an isw1/chd1 background in the presence and absence of Isw2 to determine whether Isw2 is similarly restricted at known target sites . Again , we determined that Isw2 is efficient at positioning the Ume6-proximal nucleosome but positioning of nucleosomes decays rapidly as the distance from the proximal nucleosome increases , suggesting that Isw2 may only position single nucleosomes at target sites ( Figure 1B , clusters 1 and 3 ) . Nucleosomes also appear to always be positioned toward Ume6 motifs as nucleosome positions in the absence of Isw2 are always more distal to the Ume6 motif than when Isw2 is present . Finally , these nucleosomes are positioned with the dyad only separated from the Ume6 motif by 100 nucleotides rather than the ~200 nucleotides that would be expected between dyads in a nucleosome array based on Isw2 preferentially leaving 60 base pairs of linker DNA between nucleosomes in vitro ( Kagalwala et al . , 2004; Tsukiyama et al . , 1999 ) . Of note , a subset of Ume6-bound sites do not display Isw2-dependent nucleosome remodeling ( Figure 1B , cluster 2 ) . We have observed slightly reduced chromatin immunoprecipitation ( ChIP ) signal for Ume6 at these sites ( Figure 1—figure supplement 1C ) . We speculate that for cluster 2 sites where Ume6 is bound , the Isw2 complex might in fact be recruited but that the nearest nucleosome is too distant from the recruitment site , making it out of reach of the remodeler and thus resulting in no change in nucleosome position at these sites . The observations that ( 1 ) Isw2 is solely required to move single nucleosomes at target sites , ( 2 ) Isw2 does not have global nucleosome spacing/organizing activity , and ( 3 ) Isw2 moves nucleosomes within 100 nucleotides of bound Ume6 suggest that Isw2 behavior in cells is distinct from our understanding of Isw2 activity from decades of biochemical characterization . Similarly , these specific movements toward Ume6 ( a barrier ) are inconsistent with previous biophysical studies , where ISWI proteins were shown to move nucleosomes away from inert DNA-bound factors ( Li et al . , 2015 ) . Because of these inconsistencies , we wished to know if Isw2 followed the ‘barrier model’ for positioning nucleosomes at Ume6-bound targets . To initially test this , we created a variant Ume6 construct where all residues were deleted except for the DNA binding domain . This Ume6 ( ∆2–763 ) construct binds to the same targets as full-length Ume6 ( Figure 1—figure supplement 3 ) . However , Isw2 does not appear to have any activity on global Ume6-proximal nucleosomes in the presence of the Ume6 DNA binding domain alone as nucleosomes in this strain occupy identical positions when Ume6 or Isw2 are completely absent ( Figure 1B ) . In the presence of full-length Ume6 , the Isw2 complex appears to be necessary and sufficient for moving motif-proximal nucleosomes as nucleosome positions in the ISW2/isw1/chd1 strain could achieve identical motif-proximal nucleosome positions as the wild-type strain . Additionally , the CHD1/isw1/isw2 and ISW1/chd1/isw2 strains were unable to move any Ume6-proximal nucleosomes ( Figure 1—figure supplements 4 and 5 ) , which strongly argues that Ume6 is not acting as a passive barrier against which nucleosome spacing factors can pack nucleosomes . Instead , these data are more consistent with the recent characterization of Isw2 as a ‘puller' ( Kubik et al . , 2019 ) , with Ume6 being a DNA-bound factor that may immobilize Isw2 to create leverage for ‘pulling’ . Consistent with this immobilized pulling model and consistent with the directional movement of single nucleosomes toward Ume6-bound sites , artificially tethered chromatin remodeling proteins were previously shown to always move nucleosomes toward target sites ( Donovan et al . , 2019 ) . We suspected that Ume6 and Isw2 likely interact in a specific fashion to faithfully select and precisely move single-target nucleosomes toward a recruitment motif ( Figure 1C ) . To determine which region ( s ) on Ume6 are required for specific nucleosome positioning by Isw2 , we initially created a panel of N-terminal Ume6 truncations to determine when nucleosome positioning by Isw2 is lost ( Figure 2—figure supplement 1 ) . This initial truncation panel was necessary due to the poor overall conservation of the Ume6 protein even within related yeasts , as well as the disordered structure predicted by Phyre2 ( Kelley et al . , 2015 ) . Our truncation panel indicated that Isw2 activity was retained if the N-terminus was deleted to residue 322 but lost when deleted to residue 508 . Closer inspection of the residues between 322 and 508 revealed a conserved region with a proline-rich segment followed by a predicted alpha helix , altogether spanning Ume6 residues 479–508 ( Figure 2A ) . Deletion of residues 2–479 preserved Isw2-positioned nucleosomes at Ume6 sites , while an internal deletion of 480–507 in the context of an otherwise full-length Ume6 abrogated nucleosome positioning by Isw2 ( Figure 2A , Figure 2—figure supplement 2 ) . Importantly , Ume6 ∆2–479 and Ume6 ∆2–508 showed identical binding as measured by ChIP ( Figure 2—figure supplement 3 ) , indicating that the loss of nucleosome positioning is not due to the loss of Ume6 binding . Since this region is proximal to the characterized Sin3-binding domain in Ume6 ( Washburn and Esposito , 2001 ) , we wished to validate that the newly determined Isw2-recruitment helix is independent from the Sin3-binding domain . Ume6 recruits both Isw2 and Sin3-Rpd3 for full repression of target genes ( Goldmark et al . , 2000; Fazzio et al . , 2001 ) . If either Isw2 or Sin3-Rpd3 is present , there is partial repression at Ume6-regulated genes . However , if Sin3-Rpd3 and Isw2 are both lost , Ume6 targets are fully de-repressed . We examined transcriptional output at Ume6 genes in Ume6 ( ∆2–479 ) +/– Rpd3 and Ume6 ( ∆2–508 ) +/– Rpd3 . Transcription was modestly increased at Ume6 targets in Ume6 ( ∆2–508 ) /RPD3+ compared to Ume6 ( ∆2–479 ) /RPD3+ ( Figure 2—figure supplement 4 ) , which would be expected if only Isw2 is lost when residues 479–508 are deleted . More convincingly , only a modest increase in transcription was seen at Ume6 targets in the Ume6 ( ∆2–479 ) /∆rpd3 strain , suggesting that Isw2 is still present , while the Ume6 ( ∆2–508 ) /∆rpd3 strain displayed extreme induction of Ume6-regulated genes , suggesting that both Isw2 and Rpd3 activity are absent ( Figure 2B , Figure 2—figure supplement 5 ) . Finally , we wanted to know if the predicted helix consisting of Ume6 residues 479–508 was sufficient to bring Isw2 nucleosome positioning activity to Ume6 target sites . To test this , we employed the SpyCatcher/SpyTag system ( Zakeri et al . , 2012 ) , which creates a spontaneous covalent bond between a short SpyTag peptide and a SpyCatcher domain . We fused the SpyTag peptide to the C-terminus of Ume6 ( ∆2–596 ) , a construct that is incapable of positioning motif-proximal nucleosomes ( Figure 2—figure supplement 1 ) . We then appended Ume6 residues 479–508 to the C-terminus of the SpyCatcher domain and introduced this fusion on a yeast expression plasmid driven by the ADH1 promoter . In yeast cells , this would create a fusion protein where the helical element is ectopically displayed on the C-terminus of a DNA binding competent but nucleosome-positioning-deficient construct , connected via a SpyTag-SpyCatcher linker . This fusion protein was capable of fully recapitulating Isw2-positioned nucleosomes at a subset of Ume6 sites ( Figure 2C , Figure 2—figure supplements 6 and 7 ) . Perhaps not surprisingly , considering the non-native positioning of the recruitment helix in this fusion construct , not all Ume6 sites were able to gain proper nucleosome positioning with this chimeric system ( Figure 2—figure supplement 6 ) . We conclude that the region spanning residues 479–508 in Ume6 is a yeast-conserved Isw2-recruitment domain and is required and sufficient for recruiting Isw2 nucleosome positioning activity to Ume6 targets . While dissecting the Isw2-recruitment domain in Ume6 , we discovered that deleting the MBP1 gene resulted in ectopic nucleosome positioning at a subset of Mbp1 target loci , which was identical to mispositioned nucleosomes in a ∆isw2 strain . Mbp1 is a conserved cell cycle regulator that complexes with Swi6 to form the MBF complex ( Koch et al . , 1993 ) . This complex activates the transition from G1 to S and includes the conserved function of regulating Start-specific transcription ( Koch et al . , 1993; Breeden , 1996 ) . To determine how Mbp1 recruits Isw2 , we similarly made truncations of Mbp1 to determine at which point nucleosome positioning no longer resembles wild-type positioning and reflects ∆isw2 positioning instead . The DNA binding element in Mbp1 resides in the extreme N-terminus ( Figure 3A ) spanning residues 2–124 ( Nair et al . , 2003 ) , so a panel of C-terminal truncations was created . However , before examining the full panel of truncations , we observed that nucleosome positioning was already identical to ∆isw2 positioning in Mbp1 ∆562–833 , the first C-terminal truncation examined ( Figure 3—figure supplement 1 ) . This extreme C-terminal region interacts with Swi6 ( Figure 3A ) , so we speculated that Swi6 may be responsible for recruiting Isw2 . As predicted , deletion of the SWI6 gene led to ectopic nucleosome positions identical to ∆mbp1 and ∆isw2 strains at the small subset of Mbp1 targets . We conducted sequence alignment and conservation analyses between the helical element in Ume6 and full-length Swi6 from multiple yeast species ( Figure 3A ) . We noticed a similarly conserved surface-exposed helix ( Foord et al . , 1999 ) in the cell cycle regulating protein Swi6 ( Figure 3A ) . Intriguingly , the function of this helical element has not been determined despite its sequence conservation . Because Swi6 also interacts with Swi4 to form the highly conserved SBF complex ( Koch et al . , 1993 ) , we speculated that deletion of either Swi6 , Swi4 , or Isw2 could potentially lead to ectopic nucleosome positions at a subset of SBF targets . Indeed , we observed ectopic nucleosome positioning at the HSP12 locus ( an SBF target ) when either Isw2 , Swi6 , or Swi4 was absent ( Figure 3A ) . Wild-type nucleosome positions were observed in the absence of Mbp1 , indicating that this is specific to SBF . Similarly , wild-type nucleosome positions were observed at Mbp1 targets when Swi4 was missing ( Figure 3A ) , again suggesting that MBF and SBF have individual Isw2-targeting capacity at their respective binding sites . Swi6 appears to be an adapter protein responsible for recruiting Isw2 to Mbp1 and Swi4 sites since Swi6 has no intrinsic DNA binding domain . To determine if Isw2 recruitment to Mbp1 sites was sufficient to recapitulate proper nucleosome positioning , we again used a SpyTag-SpyCatcher approach ( Figure 3B ) . Mbp1 was truncated to the DNA binding domain alone ( Mbp1 1–136 ) , which abolishes its interaction with Swi6 but still allows for proper genomic localization . This truncation construct was appended with SpyTag , and nucleosome positions were examined in the absence of any SpyCatcher partner present . As expected , we observed aberrant chromatin structure identical to the ∆isw2 strain near the Isw2-dependent Mbp1 targets , adjacent to Mbp1 consensus motifs ( Figure 3B ) . We then introduced SpyCatcher fused to the helical element from Ume6 , which was characterized above for bringing Isw2 to Ume6-bound loci . Introduction of the SpyCatcher-Ume6 fusion to the Mbp1 ( 1–136 ) -SpyTag background resulted in the rescue of proper Isw2-directed nucleosome positioning at Mbp1 sites ( Figure 3B ) . Altogether , these data strongly support our model that these conserved , putatively helical sequences are important for recruiting Isw2 to establish proper chromatin structure at multiple sequence-specific motifs throughout the genome . We also implicate Swi6 as an adapter protein for bringing Isw2 to a small subset of both Swi4 and Mbp1 targets to create Isw2-specific nucleosome positioning at these genes . Finally , the ectopic display of an Isw2-recruitment helix can recapitulate proper Isw2-directed nucleosome positioning , further supporting the notion that a small epitope is necessary and sufficient for communicating specific nucleosome positioning outputs to the Isw2 chromatin remodeling protein . The Isw2 complex contains two major subunits ( Figure 4A ) . The catalytic subunit Isw2 harbors the energy-producing ATPase domain flanked by biochemically well-defined autoregulatory domains ( Clapier and Cairns , 2012; Yan et al . , 2016; Ludwigsen et al . , 2017 ) with a C-terminal HAND-SANT-SLIDE domain , thought to bind linker DNA ( Zofall et al . , 2004 ) and interact with the accessory subunit Itc1 . Itc1 contains an N-terminal WAC domain , thought to bind to and sense extranucleosomal DNA and help with nucleosome assembly in the Drosophila ortholog ACF1 ( Fyodorov and Kadonaga , 2002 ) . Itc1 links to Isw2 through a DDT domain ( DDT is named for "DNA-binding homeobox-containing proteins and different transcription factors" ) ( Fyodorov and Kadonaga , 2002 ) . The ~350 amino acid N-terminal region of human Acf1 was shown to bind both extranucleosomal linker DNA and the histone H4 tail , suggesting an allosteric mechanism through which ISWI complexes can set proper spacing between nucleosomes ( Hwang et al . , 2014 ) . Though this work was performed with human ACF complex , Hwang et al . demonstrated that removal of residues 2–374 in S . cerevisiae was lethal , suggesting a critical and conserved role of these residues in establishing proper chromatin structure in vivo ( Hwang et al . , 2014 ) . Because of the geometry of the Isw2 complex , with the N-terminus of Itc1 sensing DNA information distal to the nucleosome onto which the catalytic subunit is engaged , we speculated that the N-terminus of Itc1 would be the most likely component of the Isw2 complex for interacting with epitopes in DNA-bound recruitment factors . We first attempted to recapitulate the result from Hwang et al . and made the identical Itc1 ( ∆2–374 ) deletion . Isw2 containing Itc1 ( ∆2–374 ) did not display any defects in nucleosome sliding using a gel mobility shift assay that detects nucleosome centering by Isw2 ( Figure 4B ) . Surprisingly , this construct was not lethal in our W303 background , but phenocopied a ∆isw2 strain by displaying identical ectopic nucleosome positioning at all Isw2 target sites throughout the genome ( Figure 4C ) . Since proper targeted nucleosome positioning was lost when this large N-terminal region was removed , but complex formation and catalytic activity were maintained , we strongly suspected that the Isw2 targeting domain resided in the Itc1 N-terminus . We created a panel of truncations in this region , guided by sequence conservation through humans , and determined whether wild-type or ∆isw2 positions were observed throughout the genome . All truncations tested resulted in loss of positioning at Isw2 targets , and we were able to narrow the targeting region entirely to the highly conserved WAC domain . Deletion of the WAC domain ( Itc1 residues 24–130 ) produced identically ectopic nucleosome positions compared to ∆isw2 at target loci ( Figure 4C ) and genome-wide ( Figure 4D ) . We conclude that the WAC domain of Itc1 is the component of the Isw2 complex responsible for coupling with epitopes on DNA-bound factors such as Ume6 , Swi6 , and all other Isw2 targeting proteins with yet-to-be-defined recruitment epitopes . To confirm that the WAC domain can interact with Isw2 targets throughout the genome , we created Itc1 ( 1-73 ) -FLAG and Itc1 ( 1-132 ) -FLAG constructs based on two differentially conserved regions within the full WAC domain ( Figure 5A ) . Neither of these constructs contains the DDT domain , so they are incapable of forming a complex with endogenous Isw2 . We performed ChIP-Seq to determine if these WAC domain constructs could associate with Isw2 targets without complexing with the Isw2 catalytic domain ( Figure 5B , Figure 5—figure supplement 1 ) . Genome-wide binding demonstrates large , but not complete overlap of Isw2 ( K215R ) -FLAG ChIP peaks with both Itc1 ( 1–73 ) -FLAG and Itc1 ( 1–132 ) -FLAG , strongly suggesting that the Itc1 region from 1 to 73 alone can interact with Isw2 targets . We noticed that the Itc1 signal and Isw2 signal were offset at target genes such that Itc1 ( 1–73 ) or Itc1 ( 1–132 ) was upstream and Isw2 was closer to the nucleosome that was selected for repositioning ( Figure 5B , Figure 5—figure supplement 1 ) . Genome-wide analysis showed that Itc1 ( 1–73 ) was associated with approximately half of Isw2-bound loci and was offset from the catalytic subunit at all co-bound sites ( Figure 5C ) . In all cases , Itc1 ( 1–73 ) was found upstream of the nucleosome that was repositioned , and Isw2 was located on top of the selected nucleosome . Nucleosomes were always shifted toward the Itc1 subunit ( Figure 5C ) . This geometry matches what was seen by ChIP-Exo mapping with Isw2 subunits at Reb1 target sites ( Yen et al . , 2012 ) . We propose a mechanism where the Itc1 WAC domain interacts with a DNA-bound factor , which constrains the Isw2 catalytic subunit to select the proper proximal nucleosome and reposition it toward the immobilized Itc1 ( Figure 5D ) . This is again consistent with the recently proposed ‘pulling’ model ( Kubik et al . , 2019 ) , but we postulate that Itc1 is anchored to a DNA-bound factor such as Ume6 to allow Isw2 to pull nucleosomes toward the proper location . There is an abundance of literature suggesting that Drosophila ACF complex , the Isw2 ortholog , is a nonspecific nucleosome spacing and assembly factor that evenly spaces phased nucleosome arrays against defined genomic barriers ( Lusser et al . , 2005; Baldi et al . , 2018; Fyodorov and Kadonaga , 2002 ) . We wondered if the WAC domain of Drosophila Acf1 was different from that of Itc1 , so we performed sequence alignment of WAC domains and compared to Acf1 from the Drosophila genus . While sequence alignment demonstrated widespread conservation of the WAC domain , one striking feature was exposed: the Drosophila genus underwent reversal or loss of negative charge at multiple residues that are strictly or mostly acidic in other representative organisms ( Figure 6A ) . Two of these residues are strictly acidic in all organisms except members of the Drosophila genus ( E33 and E40 in Itc1 ) . The other two ( E43 and D70 in Itc1 ) are more loosely conserved , though they are strictly positive charge in Drosophila . We made charge-reversal mutations in S . cerevisiae Itc1 to recapitulate the D . melanogaster residues at each of these positions either pairwise ( a , b and c , d to separate the strictly conserved acidic versus loosely conserved acidic nature ) or simultaneously ( a , b , c , d ) to reverse all charges to the D . melanogaster sequence . We assessed whether charge reversal was sufficient to abrogate targeted nucleosome positioning at Isw2 targets across the yeast genome ( Figure 6B ) . Strikingly , the E33R/E40H double mutation ( a , b ) was enough to completely abolish Isw2 activity at specific and known Isw2 targets ( Figure 6B ) and at all genomic loci where Isw2 activity is observed ( Figure 6C ) . Mutation of the less-conserved acidic residues E43R/D70K ( c , d ) retained Isw2-directed nucleosome positioning . As expected , mutation of all four acidic residues ( a , b , c , d ) E33R/E40H/E43R/D70K resulted in complete loss of Isw2-targeted activity across the genome ( Figure 6B , C ) . We conclude that the Drosophila genus lost critical acidic residues that are essential for targeted nucleosome positioning by S . cerevisiae Isw2 , potentially explaining the disconnect between the Drosophila ACF literature and what we have characterized herein . It is possible that the increase in positive charge simultaneously increases nonspecific binding of Drosophila Acf1 to extranucleosomal DNA , and these charge reversals may help explain the nonspecific spacing behavior of Acf1 observed in Drosophila . We also believe that there is strong potential that humans and most other organisms have retained targeting potential as they retain mechanistically important acidic residues present in yeast Itc1 . In support of conservation , targeted nucleosome array formation has previously been observed in humans at specific transcription factor sites including CTCF , JUN , and RFX5 ( Wiechens et al . , 2016 ) .
Collectively , our results give rise to an ‘interacting barrier model’ as an alternative means of genomic nucleosome positioning by introducing a targeted interaction between an epitope contained within condition-specific transcription factors and ISWI-type ChRPs ( Figure 7 ) . We show that a recruitment factor , the sequence-specific repressor Ume6 , harbors a helical domain that interacts with the N-terminus of the Isw2 accessory protein Itc1 . Further , we reveal this geometrically restricts the binding of the Isw2 catalytic subunit to a motif-proximal nucleosome . The complex then remodels the nucleosome , repositioning it to a specific distance from the Ume6 recognition motif . At this point and for reasons to be elucidated , this complex is strained or inactivated , and it fails to remodel any further , leaving the nucleosome in a precise location with respect to the bound recruitment factor . The activity of Isw2 and the interacting barrier sets the absolute phase of a nucleosome array that is propagated by true nonspecific spacing activities of Chd1 and Isw1 in yeast , as previously described ( Gkikopoulos et al . , 2011; Zhang et al . , 2011; Ocampo et al . , 2016 ) . This ‘interacting barrier model’ of chromatin organization is more comparable to the factor-targeted activities of SWI/SNF than the nonspecific array spacing of CHD family remodelers and is potentially conserved through humans based on conservation of key interacting residues in Itc1 ( Figure 6A ) and the observation that Isw2 orthologs can precisely position nucleosomes adjacent to specific factors in the human genome ( Wiechens et al . , 2016 ) . Together , we show that coupling between an epitope on an interacting barrier and a conserved chromatin remodeling protein leads to robust , directional , and specific nucleosome organization at genomic regulatory elements . Our data suggest that some small peptide domains embedded within transcription factors can nucleate nucleosome arrays of over 1 kb in length in vivo through an interaction with evolutionarily conserved ChRPs . Unlike the arrays established by nonspecific ChRPs , these nucleosome arrays are organized in a sequence-specific and directional manner . Establishing large swaths of chromatin structure by appending a small epitope on a genome-associated protein creates opportunity for diversity with few evolutionary constraints . Only changes in relatively small DNA binding motifs and the small peptide sequences with which they interact can have a large impact on chromatin structure . Supporting this notion , we were able to identify a strikingly similar motif to that found in Ume6 in the unrelated cell cycle regulator Swi6 , which we identified as a new Isw2-recruitment adapter for Swi4 and Mbp1 . For these reasons , we find it likely that more ChRP-interacting motifs will be discovered in multiple transcription factors from a variety of organisms , and these motifs may play a significant role in sequence-specific nucleosome positioning for precisely phased and tunable nucleosome arrays in eukaryotic genomes . Importantly , the identification of such epitopes in human cells could lead to the development of targeted drugs to specifically disrupt defined remodeler–transcription regulator interactions . We found that Isw2 acts on specific targets through these specific transcription factor interactions rather than acting on all nucleosomes genome-wide . We therefore speculate that Isw2 is in a globally repressed state in cells and activated solely on target nucleosomes . This inactivity is not consistent with work in vitro and may be caused by a regulatory interaction that has not been previously observed in biochemical systems . For example , an unknown inhibitory factor that interacts with Isw2 or the nucleosome in cells may be lost during protein purification , allowing for the ubiquitous Isw2 chromatin remodeling activity observed in vitro . Additionally , it is conceivable that Isw2 is unable to bind to linker DNA the same way in a genomic context as it can bind in vitro , potentially due to the presence of unknown chromatin interacting components , molecular crowding , chromatin folding , or other physiological differences not recapitulated in vitro . Maintaining Isw2 in an inactive state may allow organisms to conserve energy by controlling errant ATP hydrolysis while simultaneously enabling for rapid changes in chromatin structure and cellular output in differing contexts . It will be of great interest to determine how interactions with recruitment factor epitopes may alter the activity of Isw2 to elicit such precise nucleosome positioning outcomes in a cellular context . The WAC domain , a broad N-terminal region of the Itc1 accessory protein , has been previously characterized as a DNA binding element that shares sequence conservation from flies to humans ( Fyodorov and Kadonaga , 2002; Ito et al . , 1999 ) . In this work , we have identified a previously undefined function of the WAC domain in mediating protein–protein interactions between ChRPs and transcription factors in vivo . We have further demonstrated that this mediation requires two conserved acidic residues within the WAC domain , which may allow future work to distinguish the DNA and protein binding capacity of the broader WAC domain region . Intriguingly , these critical acidic residues that are conserved between yeast , humans , mice , and fish have undergone an evolutionary charge reversal in Drosophila . It is conceivable that this charge reversal establishes a more general role in generating chromatin structure for the single ISWI-type protein found in flies , as opposed to the more specialized and context-dependent roles of the many ISWI-type ChRPs found in other organisms . A recent model suggested strong interplay between the human Acf1 ( yeast Itc1 ) N-terminus , extranucleosomal DNA , and the histone H4 tail ( Hwang et al . , 2014 ) . In this model , the human Acf1 N-terminus binds to extranucleosomal DNA in nucleosomes with long linker length , allowing the Snf2h ( Isw2 ) catalytic subunit to engage the H4 tail . Snf2h engagement of the H4 tail relieves known autoinhibitory interactions ( Clapier and Cairns , 2012; Ludwigsen et al . , 2017 ) , thereby activating the remodeling complex . When linker DNA length shortens , the N-terminus of Acf1 switches to binding the H4 tail , thus displacing the Snf2h catalytic subunit and inactivating the complex through autoinhibition . This model was used to describe how ISWI complexes can be allosterically inactivated when linker DNA length shortens on nucleosomes and is a mechanistic model for how nucleosome length sensing can be achieved . Our results indicate that the N-terminus of Itc1 does not have a primary cellular function of length sensing and nucleosome spacing . If the Itc1 N-terminus can bind H4 tail and transcription factor epitopes similarly to extranucleosomal DNA and H4 tail , the Hwang et al . model ( Hwang et al . , 2014 ) can mechanistically explain the precise distance measurements made at targeted sites in cells . In this speculative model , Itc1 binds a targeting epitope at a genomic locus when the upstream nucleosome is far away . This orients the catalytic subunit on the appropriate nucleosome , which is remodeled toward the recruitment site . When the length between the nucleosome and the recruiting epitope is short enough , the Itc1 N-terminus may bind the H4 tail to inactivate Isw2 through autoinhibition . This function of binding the positively charged H4 tail may be facilitated by the clustered acidic residues in the WAC domain or may be mediated by another domain within the broadly defined 374 base pair Itc1 N- terminus implicated in putative H4 tail binding . Determining whether this interplay between a transcription factor epitope and the H4 tail can tune distance measurements in cells will be important in future biophysical characterizations . What advantage may an interacting barrier provide that a general barrier cannot , particularly since a noninteracting barrier can still phase nucleosome arrays ? We envision at least two major advantages of the interacting barrier model . First , an interacting barrier can behave directionally while a non-interacting barrier cannot . Indeed , Isw2 seems to be positioned on a specific barrier-proximal nucleosome through interactions between the WAC domain of Itc1 and the epitope on Ume6 or Swi6 . Directionality allows for more refined establishment of transcriptionally relevant chromatin arrays . Second , an interacting barrier can be modulated in a condition-specific manner through post-translational modification of the small epitope on the Isw2-recruitment factor . For example , one of the proteins that we identified as containing an Isw2- recruitment helix in yeast is Swi6 , a critical regulator of the cell cycle in the G1/S transition . Interestingly , only three Swi6-regulated genes were identified as Swi6-mediated Isw2-recruitment sites . It is thus likely that the Swi6–Isw2 interaction can be tuned by cellular context , which is not possible for noninteracting barriers . Importantly , a tunable interacting barrier allows for continuous expression of the barrier and the ability to alter its barrier activity . This is a versatile mechanism through which chromatin structure may be spatiotemporally regulated in a dynamic fashion through these ChRP-recruitment factor interactions . We predict that other targeting factors likely exist , which might recruit the Isw2 complex to specific sites in the genome through a variety of different domains , and that this creates a mechanism by which remodeling occurs at specific subsets of genomic loci in response to the presence of these recruitment factors . The expression level of transcription factors is controlled by cellular context , such as cell cycle progression or stress response , and in this way the recruitment of Isw2 to specific sites might also be linked to changes in cell conditions .
All yeast strains were derived from the parent strain S . cerevisiae W303 RAD5+ . Gene deletions were made by replacing the gene of interest with antibiotic resistance markers amplified from pAG vectors . C-terminal deletions of genes were also made by replacing the region to be deleted with antibiotic resistance markers . N-terminal gene deletions were made by first replacing the region to be deleted with a URA3 marker , and then counterselecting with 5-fluoroorotic acid ( FOA ) to delete the URA3 . Ume6-helix was introduced to yeast through plasmid transformation of a p416 vector containing the Ume6 helix fused to the SpyCatcher protein ( Zakeri et al . , 2012 ) . To make SpyTagged yeast strains , a C-terminal 3x FLAG tag followed by the SpyTag sequence ( AHIVMVDAYKPTK ) ( Zakeri et al . , 2012 ) was cloned into a pFA6a vector . Tags were then inserted at the endogenous locus of interest by homologous recombination of PCR products from the respective tagging vectors using selectable drug markers . Cells were grown at 30°C and 160 rpm in yeast extract–peptone–2% glucose ( YPD ) medium unless otherwise indicated . Strains were streaked from glycerol stocks onto 2% agar YPD plates and grown at 30°C for 2–3 days . An isolated colony was then grown overnight in 25 ml of YPD . This pre-culture was used to inoculate 25 ml of YPD at an OD600 of 0 . 2 , which was grown to an OD600 of 0 . 6–0 . 8 for chromatin analysis . Yeast containing nonintegrating plasmids ( p416 ) were grown in SD ( - ) Ura overnight , diluted to OD600 = 0 . 2 in YPD and grown to OD600 = 0 . 6–0 . 8 for chromatin analysis . Cells were then fixed with 1% formaldehyde and harvested for chromatin analysis . Yeast strains containing the Isw2 variants of interest appended with a FLAG tag were grown at 30°C to an OD600 of ~1 . Yeast were pelleted , washed with binding buffer ( 25 mM HEPES , pH 7 . 8 , 300 mM NaCl , 0 . 5 mM EGTA [ethylene glycol-bis ( β-aminoethyl ether ) -N , N , N' , N'-tetraacetic acid] , 0 . 1 mM EDTA [ethylenediamine tetraacetic acid] , 2 mM MgCl2 , 20% glycerol , 0 . 02% NP-40 , 2 mM beta-mercaptoethanol , 1 mM PMSF [phenylmethylsulfonyl fluoride] , 1× protease inhibitor cocktail [Expedeon , Cambridge UK] ) , and then lysed via cryogrinding . Yeast powder was incubated with binding buffer for 90 min before the addition of 200 μl bed volume anti-FLAG magnetic beads ( Sigma M2; Sigma , St . Louis MO USA ) . After 3 hr incubation at 4°C , beads were collected with magnets and washed three times with binding buffer and three times with elution buffer ( 25 mM HEPES pH 7 . 8 , 500 mM NaCl , 0 . 5 mM EGTA , 0 . 1 mM EDTA , 2 mM MgCl2 , 20% glycerol , 0 . 02% NP-40 , 2 mM beta-mercaptoethanol , 1 mM PMSF , 1× protease inhibitor cocktail [Expedeon , Cambridge UK] ) . A 0 . 5 mg/ml solution of FLAG peptide in 100 ml elution buffer was then added to the beads and allowed to incubate for 30 min . This process was repeated three more times for a total of four elutions . Elutions were analyzed by silver staining and combined by estimated purity for aliquoting and storage at −80° . Sliding assays were performed at least three independent times with reproducible results . Recombinant yeast histones were purified as previously described ( Luger et al . , 1999 ) and dialyzed by gradient salt dialysis onto the Widom 601 positioning sequence to create end-positioned nucleosomes with 60 base pairs of linker DNA ( Lowary and Widom , 1998 ) . Nucleosome sliding was performed at 25°C in sliding buffer ( 50 mM KCl , 15 mM HEPES , pH 7 . 8 , 10 mM MgCl2 , 0 . 1 mM EDTA , 5% sucrose , 0 . 2 mg/ml bovine serum albumin , with or without 5 mM ATP ) by incubating 1 ml or 1 . 5 ml of purified Isw2 with 12 . 5 nM reconstituted mononucleosomes for 40 min in 6 ml reaction volume . Reactions were quenched by diluting 1:2 with solution containing 3 mM competitor DNA and 5% sucrose . Native PAGE ( 6% ) was used to separate the positioning of the mononucleosomes , with Cy5 . 5-labeled nucleosomal DNA detected by a LiCor Odyssey FC imager ( LI-COR Biosciences , Lincoln NE USA ) . Micrococcal nuclease digestions were performed with a minimum of two biological replicates as previously described ( Rodriguez et al . , 2014 ) . Briefly , cells were grown to mid-log phase and fixed with 1% formaldehyde . Chromatin was digested with 10 , 20 , and 40 units of MNase for 10 min . Proper nuclease digestion of DNA was analyzed by agarose gel , and samples with approximately 80% mononucleosomes were selected for library construction . After crosslink reversal , RNase treatment , calf intestine phosphatase ( NEB , Ipswich MA , USA ) treatment , and proteinase K digestion , mononucleosome-sized fragments were gel-purified and resulting DNA was used to construct libraries with the NuGEN Ovation Ultralow kit per the manufacturer’s instructions . Libraries were sequenced at the University of Oregon’s Genomics and Cell Characterization Core Facility on an Illumina NextSeq500 on the 37 cycle , paired-end , High Output setting , yielding approximately 10–20 million paired reads per sample . Chromatin immunoprecipitation was performed with biological replicates as previously described ( Rodriguez et al . , 2014 ) . Briefly , cells were grown to mid-log phase , fixed with 1% formaldehyde , and lysed by bead-beating in the presence of protease inhibitors . Chromatin was fragmented by shearing in a Bioruptor sonicator ( Diagenode , Denville NJ , USA ) for a total of 30 min ( high output , 3 × 10′ cycles of 30 s on , 30 s off ) . Sonication conditions were optimized to produce an average fragment size of ∼300 base pairs . FLAG-tagged protein was immunoprecipitated using FLAG antibody ( Sigma , St . Louis MO , USA ) and Protein G magnetic beads ( Invitrogen , Waltham MA , USA ) . After crosslink reversal and proteinase K digestion , DNA was purified using Qiagen MinElute columns and quantified by Qubit High-Sensitivity fluorometric assay . Libraries were prepared using the NuGEN Ovation Ultralow kit by the manufacturer’s instructions and sequenced at the University of Oregon’s Genomics and Cell Characterization Core Facility on an Illumina NextSeq500 with 37 cycles of paired-end setting , yielding approximately 10 million single-end reads per sample . Only the first read ( R1 ) of each paired read was taken for downstream alignments and processing . For RNA-Seq ( minimum two biological replicates ) , RNA was purified by hot acid phenol extraction followed by polyA selection and strand-specific library construction using the NuGEN Universal Plus mRNA Kit according to the manufacturer’s instructions . Libraries were sequenced on an Illumina NextSeq500 on the 37 cycle , paired-end , high-output setting . Paired-end reads were quality filtered for adapter contamination and low-quality ends using trimmomatic ( Bolger et al . , 2014 ) . After quality filtering , an average of 10 . 5 million reads per paired-end sample remained . Surviving reads were mapped to the S . cerevisiae reference genome ( Cunningham et al . , 2015 ) using STAR ( V . 2 . 5 . 3 ) ( Dobin et al . , 2013 ) . Gene counts were quantified from uniquely aligning reads using HTSeq ( V . 0 . 9 . 1 ) ( Anders et al . , 2015 ) . Differential gene expression was performed using DESeq2 ( V . 1 . 22 . 2 ) ( Love et al . , 2014 ) , and expression graphs were generated using ggplot2 ( Wickham , 2016 ) . MNase sequencing data were analyzed as described previously ( McKnight and Tsukiyama , 2015 ) . Briefly , paired-end reads were aligned to the S . cerevisiae reference genome ( Cunningham et al . , 2015 ) using Bowtie 2 ( Langmead and Salzberg , 2012 ) and filtered computationally for unique fragments between 100 and 200 bp . Dyad positions were calculated as the midpoint of paired reads , then dyad coverage was normalized across the S . cerevisiae genome for an average read/bp of 1 . 0 . Dyad coverage is displayed in all figures . Nucleosome alignments to transcription Ume6 binding sites were performed by taking average dyad signal at each position relative to all 202 intergenic instances of a Ume6 motif center ( WNGGCGGCWW ) . PIC locations were obtained from Rhee and Pugh , 2012 . For ChIP-Seq data , single-end reads were aligned to the S . cerevisiae reference genome with Bowtie 2 and total read coverage was normalized such that the average read at a genomic location was 1 . 0 . ChIP peaks were called using a 400 bp sliding window with a threshold average enrichment within the window of 3 . 0 . Data were visualized using Integrated Genome Browser ( Freese et al . , 2016 ) . The datasets generated during this study are available in the GEO Database with accession code GSE149804 . Sequencing data sets can be accessed in the Gene Expression Omnibus with accession number GSE149804 . | DNA encodes the genetic instructions for life in a long , flexible molecular chain that is packaged up neatly to fit inside cells . Short sections of DNA are wound around proteins to form bundles called nucleosomes , and then spun into chromatin fibres , a more compact form of DNA . While nucleosomes are a fundamental part of this space-saving packaging process , they also play a key regulatory role in gene expression , which is where genes are decoded into working proteins . Placing nucleosomes at regular intervals along DNA invariably controls which parts of the DNA – and which genes – the cell’s machinery can access and ‘read’ to make proteins . But the nucleosomes’ positions are not fixed , and gene expression is a dynamic process . The cell often uncoils and repackages its DNA while molecular motors called chromatin remodelling proteins move nucleosomes up and down the DNA , exposing some genes and obstructing others . One group of chromatin remodelling proteins are called Imitation Switch ( ISWI ) complexes . It has long been thought that these complexes position nucleosomes with little regard to the underlying DNA sequence or the genes encoded , that is to say in a non-specific way . However , this theory has not been thoroughly tested . It is possible that ISWI complexes actually place nucleosomes on certain parts of DNA at particular times in an organism’s development , or in response to other environmental factors . Except how such precision is achieved remains unknown . To test this alternative theory of nucleosome positioning , Donovan et al . studied ISWI proteins and nucleosomes in common baker’s yeast . This involved systematically removing sections of ISWI proteins to see whether the complexes could still position nucleosomes , and which parts of the proteins where essential for the job . By doing so , Donovan et al . identified multiple ‘targeting’ proteins that bind to ISWI proteins and deliver the complexes to specific target sequences of DNA . From there , the complex remodels the nucleosome , positioning it at a specific distance from its landing site on DNA , as further experiments showed . This research provides a new model for explaining how nucleosomes are positioned to package DNA and control gene expression . Donovan et al . have identified a new mechanism of interaction between nucleosomes and chromatin remodelling proteins of the ISWI variety . It is possible that more interactions of this kind will be discovered with further research . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"chromosomes",
"and",
"gene",
"expression",
"biochemistry",
"and",
"chemical",
"biology"
] | 2021 | Basis of specificity for a conserved and promiscuous chromatin remodeling protein |
Our understanding of the antigen presentation pathway has recently been enhanced with the identification that the tapasin-related protein TAPBPR is a second major histocompatibility complex ( MHC ) class I-specific chaperone . We sought to determine whether , like tapasin , TAPBPR can also influence MHC class I peptide selection by functioning as a peptide exchange catalyst . We show that TAPBPR can catalyse the dissociation of peptides from peptide-MHC I complexes , enhance the loading of peptide-receptive MHC I molecules , and discriminate between peptides based on affinity in vitro . In cells , the depletion of TAPBPR increased the diversity of peptides presented on MHC I molecules , suggesting that TAPBPR is involved in restricting peptide presentation . Our results suggest TAPBPR binds to MHC I in a peptide-receptive state and , like tapasin , works to enhance peptide optimisation . It is now clear there are two MHC class I specific peptide editors , tapasin and TAPBPR , intimately involved in controlling peptide presentation to the immune system .
Major histocompatibility complex ( MHC ) class I molecules convey a selection of the peptidome of a cell to the immune system . As well as being of crucial importance for the detection of intracellular pathogens , particularly viruses , it is now apparent that peptide presentation by MHC class I is extremely relevant in the recognition of tumours ( Rötzschke et al . , 1990; Duan et al . , 2014; Gubin et al . , 2014; Snyder et al . , 2014; Yadav et al . , 2014 ) . Despite this , the molecular mechanisms controlling peptide selection for immune recognition are still poorly understood . Currently , the loading of peptide onto MHC class I is known to be orchestrated by the co-factor tapasin and is thought to occur predominantly within the peptide loading complex ( PLC ) in the endoplasmic reticulum ( Sadasivan et al . , 1996; Li et al . , 1997; Ortmann et al . , 1997; Lehner et al . , 1998; Tan et al . , 2002 ) . Tapasin enhances the rate and the extent of MHC class I peptide loading and improves the discrimination that occurs between peptides to ensure MHC class I molecules are loaded with high-affinity peptides , thus prolonging cell surface expression of MHC class I molecules ( Williams et al . , 2002; Howarth et al . , 2004; Chen and Bouvier , 2007; Wearsch and Cresswell , 2007; van Hateren et al . , 2013 ) . MHC class I allomorphs differ in their dependence on tapasin for efficient peptide loading ( Greenwood et al . , 1994; Peh et al . , 1998; Williams et al . , 2002 ) . Furthermore , the ability to select and assemble with an optimal peptide cargo in the absence of tapasin is inversely correlated to the enhancement observed in the presence of tapasin ( Williams et al . , 2002; van Hateren et al . , 2013; Rizvi et al . , 2014 ) . It has recently become apparent that , in addition to tapasin , there is a second MHC class I-specific chaperone , the tapasin-related protein TAPBPR ( Boyle et al . , 2013 ) . In contrast to tapasin , TAPBPR is not an integral component of the peptide loading complex and cannot compensate for the loss of tapasin ( Boyle et al . , 2013; Hermann et al . , 2013 ) . Currently , the precise function of TAPBPR in the MHC class I pathway is unknown ( Hermann et al . , 2015 ) . A functional role in MHC class I antigen presentation is supported by the finding that the presence of TAPBPR slows the anterograde trafficking of MHC class I and prolongs the association of MHC class I with the PLC . This raises the possibility that TAPBPR serves as an additional quality control checkpoint in the MHC class I antigen presentation pathway ( Boyle et al . , 2013 ) . Although tapasin and TAPBPR only share 22% identity ( Teng et al . , 2002 ) , both bind MHC class I , and their orientation on MHC class I is similar ( Dong et al . , 2009; Hermann et al . , 2013 ) . This shared orientation raises the possibility that there is some common functionality between the two molecules , particularly in regard to the ability to influence the peptide repertoire presented on MHC class I . Here , we sought to determine whether , like tapasin , TAPBPR has a peptide editing function .
In order to test whether TAPBPR has peptide editing functionality we chose an approach analogous to that previously established for tapasin , namely a cell free in vitro assay that allows monitoring of peptide binding to MHC class I in real time via fluorescence anisotropy ( Chen and Bouvier , 2007 ) . Due to the intrinsically low affinity between tapasin and MHC class I molecules in vitro , soluble tapasin function on MHC class I is only measurable when the two molecules are brought into close proximity via a Jun/Fos leucine zipper ( Chen and Bouvier , 2007 ) or when tapasin is conjugated to ERp57 ( Wearsch and Cresswell , 2007 ) . Since TAPBPR is not integrated into the PLC and can be co-immunoprecipitated with MHC I ( Boyle et al . , 2013 , Hermann et al . , 2013 ) , it is likely that TAPBPR has a higher affinity for MHC class I than tapasin . We therefore tested whether TAPBPR functioned in vitro without the need for an artificial intermolecular tether . To this end , the luminal domains of TAPBPR were cloned into the pHLsec expression vector and transiently transfected into HEK293F cells . This resulted in the efficient production of a secreted form of TAPBPR with a C-terminal His-tag , allowing for purification from the culture supernatant using Ni-affinity and size exclusion chromatography . TAPBPR protein eluted as a single major peak of high purity as verified by Coomassie staining following sodium dodecyl sulfate polyacrylamide gel electrophoresis ( SDS-PAGE ) ( Figure 1A ) . TAPBPRTN5 in which the isoleucine at position 261 was mutated to lysine , which inhibits the interaction of TAPBPR with MHC class I ( Hermann et al . , 2013 ) , was produced in the same manner and was also efficiently expressed and produced a major single protein peak following size-exclusion chromatography ( Figure 1B ) . Differential scanning fluorimetry revealed melting temperatures for TAPBPR and TAPBPRTN5 to be between 51 . 5°C and 52 . 5°C indicating that the single point mutation did not adversely affect TAPBPR folding ( Figure 1C ) . These mammalian-expressed , purified TAPBPR proteins were used in subsequent fluorescence anisotropy experiments to characterise the equilibrium and kinetic parameters between MHC class I and different peptide ligands . 10 . 7554/eLife . 09617 . 003Figure 1 . Expression of TAPBPR . Size exclusion chromatograms of ( A ) TAPBPR and ( B ) TAPBPRTN5 purified from cell culture supernatants . The protein peaks were analysed by SDS-PAGE followed by Coomassie staining . ( C ) Differential scanning fluorimetry of TAPBPR and TAPBPRTN5 demonstrate equivalent thermal denaturation profiles . The data is representative of two independent experiments . DOI: http://dx . doi . org/10 . 7554/eLife . 09617 . 003 To investigate if TAPBPR is able to directly edit peptides on MHC class I molecules in a manner similar to that attributed to tapasin , we adapted the in vitro peptide exchange assay previously developed for measuring tapasin function ( Chen and Bouvier , 2007 ) . Using fluorescent polarisation we tested whether TAPBPR enhanced peptide dissociation from HLA-A2 , an allomorph that co-immunoprecipitates readily with TAPBPR ( Hermann et al . , 2013 ) . HLA-A*02:01 molecules were loaded with the fluorescent peptide FLPSDC*FPSV ( C* indicates the position of the fluorophore ) before excess unlabelled FLPSDCFPSV peptide was added in the presence or absence of TAPBPR . Dissociation of FLPSDC*FPSV was only apparent in the first six hours in the presence of competitor peptide and TAPBPR ( Figure 2A ) . Dissociation of FLPSDC*FPSV was also observed when a vast excess of unlabelled NLVPMVATV was used in the presence of TAPBPR ( Figure 2B ) . To ensure the enhancement of peptide dissociation was a consequence of direct interaction between TAPBPR and HLA-A2 , we repeated the peptide dissociation assays using TAPBPRTN5 ( I261K ) which inhibits the ability of TAPBPR to bind to HLA-A2 ( Hermann et al . , 2013 ) . No enhancement in the dissociation of FLPSDC*FPSV was observed in the presence of TAPBPRTN5 ( red line Figure 2B ) . Thus the ability of TAPBPR to enhance peptide dissociation is a direct consequence of TAPBPR binding to HLA-A2 . 10 . 7554/eLife . 09617 . 004Figure 2 . TAPBPR functions as a peptide loading catalyst and peptide editor for HLA-A2 . ( A , B ) Dissociation , ( C ) association and ( D ) peptide competition of fluorescent peptide FLPSDC*FPSV on HLA-A*02:01 in the absence or presence of TAPBPR or TAPBPRTN5 as measured by fluorescence polarisation . ( A ) 0 . 15 µM HLA-A*02:01fos molecules were mixed with 1 . 2 µM human β2m and loaded with 0 . 1 μM FLPSDC*FPSV and ( B ) 0 . 5 µM HLA-A*02:01 molecules were loaded with 0 . 125 μM FLPSDC*FPSV . The complexes were then split and incubated with 1000 fold molar excess of ( A ) FLPSDCFPSV or ( B ) NLVPMVATV with either A ) buffer ( No protein ) or supplemented with 0 . 75 μM TAPBPR , or ( B ) buffer ( No protein ) or supplemented with 0 . 25 μM TAPBPR or TAPBPRTN5 . Note , the slight difference in dissociation rate observed in A & B is likely to be related to the concentration of TAPBPR used in each experiment and not to the sequence of the competing peptide used . The data shown in Figure 2A , B is representative of 13 independent experiments , which all produced similar results . ( C ) 0 . 5 µM HLA-A*02:01 molecules were made peptide receptive and then the binding of 0 . 125 µM FLPSDC*FPSV was followed in the presence or absence of 0 . 05 µM TAPBPR or TAPBPRTN5 . One representative association experiment of 13 experiments is shown . ( D ) 0 . 5 µM HLA-A*02:01 molecules were made peptide-receptive and incubated with 0 . 125 μM high affinity peptide FLPSDC*FPSV and various concentrations of the lower affinity competing peptide NLVPMVATV ( 0–20 μM ) in presence or absence of 0 . 25 μM TAPBPR or TAPBPRTN5 . One of two experiments is shown . ( E–G ) Comparison of the dissociation of seven peptides from HLA-A*02:01fos in the presence or absence of TAPBPR or tapasin-Jun as performed in Figure 2—figure supplement 1 . The results of three ( E ) and four ( F , G ) independent experiments were combined and are shown . Data from each experiment was processed in GraphPad Prism using one-phase exponential decay non-linear regression . The half-lives that were calculated for the dissociation of the indicated peptide in the presence of either ( E ) Competitor only , ( F ) Competitor and Tapasin-Jun or ( G ) Competitor and TAPBPR were plotted as a percentage of the half-life calculated for the dissociation of KLWEAESK*L in the equivalent condition . Error bars show the standard deviation of the relative half-lives measured in the different experiments . While there were slight modifications of experimental conditions between replicate experiments , results consistent with the presented results were observed . DOI: http://dx . doi . org/10 . 7554/eLife . 09617 . 00410 . 7554/eLife . 09617 . 005Figure 2—figure supplement 1 . Dissociation of seven different peptides from HLA-A*02:01fos in the presence or absence of TAPBPR or Tapasin-Jun . 0 . 15 µM HLA-A*02:01fos molecules were mixed with 1 . 5 µM human β2m and made peptide receptive by UV exposure and loaded with 0 . 1 µM of the indicated labelled peptide overnight at 4°C . Next day the sample was split and either: 1000 excess unlabelled FLPSDCFPSV in buffer ( No protein ) ; 1000 excess unlabelled FLPSDCFPSV supplemented with 0 . 35 µM TAPBPR; or 1000 excess unlabelled FLPSDCFPSV supplemented with 0 . 1 μM Tapasin-Jun was added . One experiment from a total of four experiments is shown . DOI: http://dx . doi . org/10 . 7554/eLife . 09617 . 005 Next we examined the effect that TAPBPR has on the kinetics of peptide association with HLA-A2 . To test this , HLA-A*02:01 molecules refolded with a UV conditional ligand were made peptide-receptive by photolysis and the binding of the fluorescent peptide FLPSDC*FPSV was monitored by fluorescence polarisation in the absence or presence of TAPBPR . We found that TAPBPR catalysed binding of FLPSDC*FPSV to peptide-receptive HLA-A*02:01 ( Figure 2C ) . In contrast , no enhancement in the association of FLPSDC*FPSV with HLA-A2 was observed in the presence of TAPBPRTN5 ( Figure 2C ) , further demonstrating that a direct association between TAPBPR and HLA-A2 is required for TAPBPR to influence HLA-A2 peptide binding characteristics . To ascertain whether TAPBPR is able to discriminate between peptides , competition assays were performed in which peptide-receptive HLA-A*02:01 molecules were incubated with a mix of two peptides; 0 . 125 µM of the high affinity labelled peptide FLPSDC*FPSV , and a variable concentration of the peptide NLVPMVATV , which binds HLA-A*02:01 with lower affinity than FLPSDCFPSV . In the presence of TAPBPR , NLVPMVATV became a poorer competitor ( Figure 2D ) . TAPBPRTN5 had no effect on peptide competition ( Figure 2D ) . These results suggest that , like tapasin ( Chen and Bouvier , 2007 ) , TAPBPR can enhance the selection of high affinity peptides for binding to MHC I in vitro . Next we compared the ability of TAPBPR or tapasin-jun to enhance dissociation of FLPSDC*FPSV and a panel of six additional peptides bound to HLA-A2 . The dissociation rates of all peptides were enhanced in the presence of TAPBPR and in the presence of tapasin-jun as compared to competing peptide alone ( Figure 2—figure supplement 1 ) . However , peptide specific differences were apparent in the ability of tapasin-jun and TAPBPR to enhance dissociation . As different concentrations of active tapasin-jun and TAPBPR proteins were present in these assays , it was not possible to directly compare the dissociation half-lives of specific peptides in the presence of tapasin-jun with TAPBPR . Therefore we compared the hierarchies of dissociation relative to one peptide ( KLWEAESK*L ) over four independent experiments . This revealed reproducible differences in the hierarchies of peptide dissociation between tapasin-jun and TAPBPR on HLA-A2 ( Figure 2E–G ) . Interestingly the half-life hierarchy for tapasin was similar to the hierarchy observed in the absence of either cofactor ( competitor only ) ( Figure 2E , F ) , with the exception of the dissociation of GLDDIKDLK*V whose dissociation was relatively much slower in the presence of tapasin-jun compared to the other peptides than was observed in the presence of competitor only . This suggests that tapasin enhancement of peptide dissociation generally parallels peptide-MHC complex stability , as previously reported ( Howarth et al . , 2004 ) . However , the half-life of peptides in the presence of TAPBPR does not follow this trend ( Figure 2G ) . Specifically we observed that in comparison to the other peptides , TAPBPR enhanced dissociation of KLVK*EVIAV to a much smaller extent than was apparent in the presence of tapasin-jun or absence of either cofactor ( competitor only ) ( Figure 2E–G ) . The same was also true of FLPSDC*FPSV ( Figure 2E–G ) . Taken together our results suggest firstly that both tapasin and TAPBPR enhance peptide dissociation , and secondly that there are subtle differences in regard to their peptide specificity . We previously reported that TAPBPR has preference for HLA-A68 over HLA-B15 expressed in HeLa cells ( Boyle et al . , 2013 ) . We wondered if TAPBPR generally exhibited preference for HLA-A allomorphs or whether TAPBPR might interact with HLA-B allomorphs . To explore this , we expressed a small panel of individual HLA allomorphs in the MHC class I negative cell line 721 . 221 , and determined their interaction with endogenous TAPBPR , in a system in which competition between MHC class I allomorphs is absent . Immunoprecipitation of TAPBPR , followed by western blotting for MHC class I , confirmed the strong association between TAPBPR and the HLA-A allomorphs HLA-A2 and HLA-A68 ( Figure 3 ) . However , an association between TAPBPR and HLA-B allomorphs was also detected , although this was weaker ( HLA-B8 and -B15 ) , or below the limits of detection ( HLA-B40 ) , compared to that for HLA-A2 ( Figure 3 ) . Therefore , although there is clearly a strong association with those HLA-A allomorphs tested , TAPBPR interacts with a broader range of HLA molecules than originally appreciated ( Boyle et al . , 2013 ) . 10 . 7554/eLife . 09617 . 006Figure 3 . TAPBPR associates with HLA-A and HLA-B molecules . TAPBPR was isolated by immunoprecipitation ( using R014 ) from the MHC class I negative cell line 721 . 221 and 721 . 221 stably transduced with HLA-A2 , A68 , B8 , B15 or B40 . Western blot analysis was performed for TAPBPR and the MHC class I heavy chain ( using HCA2 and HC10 ) on lysates ( labelled lys ) and TAPBPR immunoprecipitates as indicated . The data is representative of three independent experiments . DOI: http://dx . doi . org/10 . 7554/eLife . 09617 . 006 Given the interactions detected between TAPBPR and the studied HLA-B allomorphs , we asked whether TAPBPR was involved in peptide selection on HLA-B*08:01 , an allomorph which interacts weakly with TAPBPR ( Figure 3 ) and which is well-characterised in regard to tapasin function ( Greenwood et al . , 1994; Lehner et al . , 1998; Peh et al . , 1998; Chen and Bouvier , 2007; Wearsch and Cresswell , 2007 ) . We first asked whether TAPBPR enhanced peptide dissociation from HLA-B*08:01fos molecules . We found that dissociation of EIYK*RWIIL was enhanced in the presence of TAPBPR ( Figure 4A and Figure 4—figure supplement 1A ) . No enhancement of the dissociation of EIYK*RWIIL was observed in the presence of TAPBPR using HLA-B8:01-T134K in which the T134 residue had been mutated to lysine , which abrogates binding of TAPBPR to MHC I ( Hermann et al . , 2013 ) ( Figure 4B and Figure 4—figure supplement 1B ) , demonstrating that the enhanced peptide dissociation was dependent on a direct interaction between TAPBPR and HLA-B8 . We also tested the ability of TAPBPR to enhance dissociation of two additional peptides , ELRSRK*WAI and FLRGRK*YGL from HLA-B*08:01 . The dissociation of ELRSRK*WAI was not altered in the presence of TAPBPR ( Figure 4C and Figure 4—figure supplement 1C ) and only a very slight change in dissociation of FLRGRK*YGL was observed in the presence of TAPBPR ( Figure 4D and Figure 4—figure supplement 1D ) . Therefore , our results suggest that TAPBPR can enhance peptide dissociation from HLA-B8 but that this effect is peptide specific . Tapasin has also been reported to exhibit peptide specificity in regard to peptide dissociation ( Chen and Bouvier , 2007 ) . Like TAPBPR , tapasin-jun accelerated dissociation of fluorescein isothiocyanate ( FITC ) labelled versions of both EIYK*RWIIL and FLRGRK*YGL from HLA-B*08:01fos , while the dissociation of ELRSRK*WAI was insensitive to tapasin-jun ( Chen and Bouvier , 2007 ) . 10 . 7554/eLife . 09617 . 007Figure 4 . TAPBPR can function as a peptide loading catalyst and peptide editor for HLA-B8 . ( A–D ) Dissociation , ( E–G ) association and ( H ) peptide competition of fluorescent peptides on HLA-B*08:01 in the absence or presence of TAPBPR . 2 μM HLA-B*08:01fos or HLA-B*08:01fos T134K molecules were mixed with 20 μM human β2m and made peptide receptive then loaded with 1 μM ( A , B ) EIYK*RWIIL ( C ) ELRSRK*WAI or ( D ) FLRGRK*YGL . Dissociation was subsequently followed after the addition of a 250 molar excess of ( A , B ) EIYKRWIIL ( C ) ELRSRKWAI , or ( D ) FLRGRKYGL in the absence or presence of 0 . 75 µM TAPBPR . A total of eight dissociation experiments have been conducted for wild-type HLA B*08:01fos , and the T134K mutant was included in six of these experiments . ( E–G ) 0 . 6 μM HLA B*08:01fos molecules were mixed with 6 µM human β2m and made peptide-receptive , then the binding of 1 µM of ( E ) EIYK*RWIIL , ( F ) ELRSRK*WAI or ( G ) FLRGRK*YGL was followed in the absence or presence of 0 . 175 µM TAPBPR . One of three experiments is shown . ( H ) 0 . 6 μM HLA-B*08:01fos molecules were mixed with 0 . 6 µM human β2m and were made peptide-receptive , then incubated with 0 . 1 μM high affinity peptide ELRSRK*WAI and various concentrations of the lower affinity competing peptide EIYKRWIIL ( 0–25 μM ) in presence or absence of 0 . 3 μM TAPBPR . One of six experiments is shown . Fluorescence polarisations measurements were taken after being left at room temperature overnight . While there were slight modifications of experimental conditions between replicate experiments , results consistent with the presented results were observed . DOI: http://dx . doi . org/10 . 7554/eLife . 09617 . 00710 . 7554/eLife . 09617 . 008Figure 4—figure supplement 1 . Dissociation of three different peptides from HLA B*08:01fos in the presence or absence of TAPBPR . 2 µM HLA-B*08:01fos molecules or T134K mutant were mixed with 20 µM human β2m and made peptide receptive by UV exposure and then loaded with 1 μM of the indicated peptide overnight at 4°C . Dissociation was subsequently followed at room temperature after the addition of a 250 molar excess of ( A B ) EIYKRWIIL ( C ) ELRSRKWAI , or ( D ) FLRGRKYGL in the absence ( black dots ) or presence of 0 . 75 µM TAPBPR ( blue dots ) for up to 240 hr . The results of three independent experiments are combined , with each polarisation measurement plotted as a percentage of the initial polarisation measurement taken in the absence of competing peptide from the same experiment . DOI: http://dx . doi . org/10 . 7554/eLife . 09617 . 008 As with HLA-A2 , we observed that TAPBPR could enhance peptide binding to peptide-receptive HLA-B*08:01fos molecules . However , this was only observed with EIYK*RWIIL ( Figure 4E ) and the binding of ELRSRK*WAI or FLRGRK*YGL to peptide receptive HLA-B*08:01fos was not altered in the presence of TAPBPR ( Figure 4F , G ) . This contrasts with the results previously published by Chen and Bouvier , who observed that tapasin-jun enhanced binding of FITC labelled versions of ELRSRK*WAI and FLRGRK*YGL to peptide-receptive HLA-B*08:01fos molecules ( Chen and Bouvier , 2007 ) . Taken together the data indicate that TAPBPR can enhance the loading of peptide onto HLA-B*08:01 , and that this is peptide specific . Furthermore , there are differences in peptide specificity between tapasin and TAPBPR with respect to HLA-B*08:01 peptide-loading . Finally we showed that , like tapasin ( Chen and Bouvier , 2007 ) , TAPBPR increased peptide exchange on HLA-B*08:01 since the low affinity peptide EIYKRWIIL became a poorer competitor against the high affinity labelled peptide ELRSRK*WAI in the presence of TAPBPR ( Figure 4H ) . Given that TAPBPR functions as a peptide exchange catalyst for MHC class I in vitro , we next determined if TAPBPR expression had any effect on the peptide repertoire expressed by MHC class I molecules in cells . To do this , we created a HeLa cell line in which TAPBPR was knocked out using the clustered regularly interspaced short palindromic repeats ( CRISPR ) system . We found that upon treatment with IFN-γ , TAPBPR expression was induced in HeLa , but not in the HeLa-TAPBPR KO line ( Figure 5A ) . Tapasin expression was not affected by the sgRNA targeted to TAPBPR ( Figure 5A ) . In the absence of TAPBPR , cell surface expression of MHC class I in IFN-γ treated HeLa did not significantly change as determined by W6/32 staining and flow cytometry ( Figure 5B ) . When the amino acid sequences of the peptides eluted from MHC class I molecules from IFN-γ treated HeLa and HeLa-TAPBPR KO cells were compared , the most noticeable difference was a change in the diversity of peptides isolated ( Figure 5C ) . In IFN-γ treated HeLa cells , we isolated 1607 different peptides on MHC class I , compared with 2074 in IFN-γ treated HeLa-TAPBPR KO cells ( Figure 5C ) . 1398 peptides were shared between the two cell lines , but 209 peptides were unique to HeLa and 676 were unique to HeLa-TAPBPR KO ( Figure 5C ) . When the peptides were subdivided into those predicted to bind to HLA-A*68:02 or HLA-B*15:03 based on anchor sequences , an increased repertoire was observed for both allomorphs in the absence of TAPBPR ( Figure 5C ) . Consistent with these findings , a similar pattern of increased peptide diversity was also observed in IFN-γ treated HeLa-S cells upon TAPBPR depletion using shRNA to TAPBPR ( Figure 5—figure supplement 1 ) . 10 . 7554/eLife . 09617 . 009Figure 5 . TAPBPR expression alters the peptide repertoire presented by MHC class I on cells . ( A ) The TAPBPR:MHC class I complex was immunoprecipitated from IFN-γ treated HeLa and HeLa-TAPBPR KO cells . Western blot analysis was performed for TAPBPR ( mouse anti-TAPBPR ) , tapasin ( Rgp48N ) , MHC class I ( HC10 ) and calnexin on lysates and TAPBPR immunoprecipitates as indicated . The data is representative of three independent experiments . ( B ) Cytofluorometric analysis of MHC class I detected with W6/32 on IFN-γ treated HeLa ( Blue line ) and HeLa-TAPBPR KO cells ( Red line ) . Staining with an isotype control on both cell lines ( dashed lines ) is included as a control . ( C , D ) Peptide-MHC class I complexes were isolated by affinity chromatography using W6/32 from ( C ) IFN-γ treated HeLa and HeLa-TAPBPR KO cells or ( D ) HeLa and HeLa overexpressing WT-TAPBPR . Eluted peptides were analysed using LC-MS/MS . Graphs show the total number of peptides and also the number of peptides assigned as HLA-A*68:02 and HLA-B*15:03 binders based on their peptide motifs using the online programme NetMHC . The number of peptides shared between the cell lines ( white bar ) and the number of peptides unique to each cell line ( coloured bars ) is shown . The data was generated from tandem MS analysis performed five times on one immunoprecipitate . Two independent biological repeats have been performed in two different cell lines ( HeLa-S cells shown in Figure 5—figure supplement 1 , and KBM-7 cells shown in Figure 6 ) in which a similar pattern of increased peptide diversity in TAPBPR depleted cells was observed . ( E–H ) Conservation of P2 and C-terminal anchor residues ( PΩ ) . Plots show prevalence of classic peptide anchors on ( E , F ) HLA-A*68:02 or ( G , H ) HLA-B*15:03 . Bars show classic anchor conservation at P2 , PΩ and P2/PΩ combined for ( E , G ) shared peptides found in both IFN-γ treated HeLa and HeLa TAPBPR KO cells ( white bar ) ( presumably permitted expression in the presence of TAPBPR ) and peptides unique to HeLa TAPBPR KO + IFNγ cells ( red bar ) ( presumably restricted in the presence of TAPBPR ) , ( F , H ) shared peptides found in both HeLa and HeLa over-expressing TAPBPR ( white bar ) ( presumably permitted expression in the presence of TAPBPR ) and peptides unique to HeLa ( green bar ) ( presumably restricted in the presence of TAPBPR ) . DOI: http://dx . doi . org/10 . 7554/eLife . 09617 . 00910 . 7554/eLife . 09617 . 010Figure 5—figure supplement 1 . Increased peptide diversity on MHC class I in the absence of TAPBPR in IFN-γ treated HeLa-S cells . Approximately 2 × 108 HeLa-S and HeLa-S depleted of TAPBPR expression using shRNA specific for TAPBPR ( Boyle et al . , 2013 ) were induced with IFN-γ treated for 72 hr . This was repeated three time ( Batch no . 1–3 ) in order to harvest >5 × 108 cells of each line . ( A ) To confirm the efficient induction of TAPBPR by IFN-γ and the depletion of TAPBPR in shTAPBPR treated cells , a small fraction of the cells were lysed in 1% digitonin-TBS , TAPBPR was immunoprecipitated using the rabbit anti-TAPBPR polyclonal R014 and western blot analysis was performed for TAPBPR ( using mouse anti-TAPBPR ) . ( B ) Peptide-MHC class I complexes were isolated using affinity chromatography using W6/32 from IFN-γ treated HeLa-S and shTAPBPR depleted HeLa-S cells . Eluted peptides were analysed using LC-MS/MS . The Graph show the total number of shared and unique peptides , and also the number of peptides assigned as HLA-A*68:02 and HLA-B*15:03 binders based on their peptide motifs . Note: surface expression of MHC class I detected by W6/32 is similar in both cells as demonstrated in Boyle et al . , 2013 . The number of peptides shared between the cell lines ( white bar ) and the number of peptides unique to each cell line ( coloured bars ) is shown . ( C&D ) Conservation of P2 and C-terminal anchor residues ( PΩ ) . Plots show prevalence of classic peptide anchors on ( C ) HLA-A*68:02 or ( D ) HLA-B*15:03 . Bars show classic anchor conservation at P2 , PΩ and P2/PΩ combined for shared peptides found in both IFN-γ treated HeLa and HeLa shTAPBPR cells ( white bar ) ( presumably permitted expression in the presence of TAPBPR ) and peptides unique to HeLa shTAPBPR IFNγ cells ( red bar ) ( presumably restricted in the presence of TAPBPR ) . DOI: http://dx . doi . org/10 . 7554/eLife . 09617 . 01010 . 7554/eLife . 09617 . 011Figure 5—figure supplement 2 . TAPBPR expression does not influence peptide length . The average length of peptide eluted from HLA-A*68:02 and HLA-B*15:03 was calculated using GraphPad Prism and compared between ( A ) IFN-γ treated HeLa and HeLa-TAPBPR KO cells and ( B ) HeLa cells and HeLa overexpressing TAPBPR . Error bars show SEM of peptide length . The p value was calculated by two-tailed t-test . DOI: http://dx . doi . org/10 . 7554/eLife . 09617 . 01110 . 7554/eLife . 09617 . 012Figure 6 . Distinct alterations to the peptide repertoire expressed on MHC class I molecules in IFN-γ treated KBM-7 cells upon TAPBPR depletion or tapasin deficiency . Cytofluorometric analysis of ( A ) HLA-A2 expression ( detected with BB7 . 2 ) and ( D ) HLA-B expression ( detected with 4E ) on the cell surface of WT , TAPBPR stably depleted ( shTAPBPR ) and tapasin deficient ( TPN- ) KBM-7 cells treated with and without IFN-γ for 48 hr . The data is representative of three independent experiments . ( B , E ) Peptide-MHC class I complexes were isolated using affinity chromatography using ( B ) BB7 . 2 ( HLA-A2 ) and ( E ) B1 . 23 . 2 ( HLA-B , -C ) from IFN-γ treated WT , TAPBPR stably depleted ( shTAPBPR ) and tapasin deficient ( TPN- ) KBM-7 cells . Eluted peptide were analysed using LC-MS/MS . Graphs show the total number of peptides . The data was generated from tandem MS analysis performed five times on one immunoprecipitate . Two independent biological repeats comparing the peptide repertoire expressed on MHC class I under conditions of TAPBPR competency and TAPBPR deficiency in two other cell lines have been performed ( HeLa-M cells shown in Figure 5 and HeLa-S cell shown in Figure 5—figure supplement 1 ) in which a similar pattern of increased peptide diversity upon TAPBPR depletion was observed . ( C , F ) Venn diagrams show the overlap of peptides on ( C ) HLA-A2 and ( F ) HLA-B , -C eluted from IFN-γ treated WT , TAPBPR stably depleted ( TAPBPR KD ) and tapasin deficient ( Tapasin KO ) KBM-7 cells . DOI: http://dx . doi . org/10 . 7554/eLife . 09617 . 012 To complement these findings we compared the amino acid sequences of peptides eluted from MHC class I molecules isolated from HeLa and HeLa cells over-expressing TAPBPR by mass spectrometry . Again the most obvious difference was a change in the diversity of peptides isolated ( Figure 5D ) . We found the total number of different peptides isolated from MHC class I was decreased in cells over-expressing TAPBPR , from 819 MHC class I restricted peptides from HeLa cells down to 296 MHC class I restricted peptides isolated from HeLa cells over-expressing TAPBPR . 210 peptides were shared between the two cell lines ( Figure 5D ) . Again , when these peptides were further sub-divided into those predicted to bind to HLA-A*68:02 and -B*15:03 based on their motif , an increased repertoire was observed for both allomorphs in the absence of TAPBPR ( Figure 5D ) . Taken together , these results suggest that TAPBPR restricts the peptide repertoire presented on MHC class I molecules . We compared the anchor residues found at the P2 and C-terminal ( PΩ ) position of the shared peptides , eluted from either TAPBPR negative or TAPBPR expressing cells , which will presumably be permitted release in TAPBPR expressing cells , with unique peptides exclusively eluted from TAPBPR negative cells ( HeLa or HeLa-TAPBPR KO+IFN-γ ) , which are therefore usually removed or restricted by TAPBPR . We found there was an enhancement of canonical anchor residues in the peptides eluted from TAPBPR expressing cells ( Figure 5E–H ) . For example , in IFN-γ treated HeLa cells 53% of the shared peptides predicted to be bound to HLA-A*A68:02 ( which had presumably been subjected to TAPBPR function ) had classic anchor residues ( P2 = threonine or valine , PΩ = valine or leucine ) compared to 43% of the peptides found exclusively in TAPBPR deficient cells which are the peptides that TAPBPR usually restricts ( Figure 5E ) . More striking was the restriction of peptides predicted to be bound by HLA-B*15:03 in cells over-expressing TAPBPR ( Figure 5H ) with only 5% of the peptides eluted exclusively from TAPBPR deficient cells ( i . e . the TAPBPR restricted peptides ) containing classical anchor motifs ( P2 = glutamine or lysine , PΩ = tyrosine or phenylalanine ) compared to 24% of the peptides eluted from TAPBPR over expressing cells ( i . e . the TAPBPR permitted peptides ) . This suggests TAPBPR restriction in vivo helps to remove some peptides of lower affinity and therefore assists in improving peptide selection on MHC class I . We found TAPBPR expression did not have a dramatic effect on peptide length ( Figure 5—figure supplement 2 ) . We also compared the effect of TAPBPR depletion with tapasin deficiency on the MHC class I peptide repertoire from IFN-γ treated KBM-7 cells ( see [Hermann et al . , 2013] for knockdown efficiency ) . Surface expression of HLA-A2 was similar in IFN-γ treated wild-type ( WT ) , TAPBPR depleted and tapasin deficient KBM-7 cells ( Figure 6A ) . While TAPBPR depletion resulted in increased peptide diversity on HLA-A2 ( Figure 6B ) , tapasin deficiency did not significantly change the total number of peptides presented on HLA-A2 ( Figure 6B ) , although the peptide repertoire was significantly different between IFN-γ treated WT and tapasin deficient KBM-7 cells ( Figure 6C ) . Surface expression of HLA-B40 was similar in IFN-γ treated WT and TAPBPR depleted cells , but was reduced in tapasin deficient KBM-7 cells ( Figure 6D ) . TAPBPR depletion and tapasin deficiency produced different effects on peptides presented by HLA-B40 and -C03 in IFN-γ treated KBM-7 cells , with TAPBPR depletion increasing peptide diversity and tapasin deficiency decreasing the number of peptides presented ( Figure 6E , F ) . A similar effect of tapasin deficiency decreasing peptide number has recently been reported for the mouse MHC I molecule H-2Db and Kb ( Kanaseki et al . , 2013 ) . Therefore , although in vitro TAPBPR and tapasin both function as peptide exchange catalysts , within the cellular environment the two molecules shape the peptide repertoire in distinct ways . Given our finding that TAPBPR functions as a peptide exchange catalyst in vitro and restricts the peptide repertoire presented on MHC class I on cells , we next asked if the MHC class I molecules bound to TAPBPR in cells were peptide-receptive and whether they could be released from TAPBPR by high-affinity peptide . TAPBPR was isolated in complex with MHC I from IFN-γ induced KBM-7 cells ( which are HLA-A2+ ) , then incubated in the presence or absence of 100 μM of the HLA-A2 binding peptide NLVPMVATV . Western blot analysis of the proteins co-immunoprecipitated with TAPBPR revealed a >90% reduction in the amount of HLA-A2 bound to TAPBPR following the addition of the NLVPMVATV peptide ( Figure 7 ) . To ensure this was a consequence of direct peptide binding to HLA-A2 , we tested NLVPMVATV peptide variants in which the anchor residues at position 2 and 9 were substituted , thus lowering the avidity of the peptide . Less dissociation of HLA-A2 from TAPBPR was observed using NAVPMVATV ( Δ2 ) and NLVPMVATM ( Δ9 ) peptides , in which one anchor residue had been substituted , while the NAVPMVATM ( Δ2/9 ) peptide , in which both anchor residues were substituted was unable to induce any HLA-A2 dissociation ( Figure 7 ) . Next we asked whether peptide loaded HLA-A2 molecules were released from TAPBPR following peptide addition by performing a second immunoprecipitation using BB7 . 2 from the peptide-treated eluates . Peptide loaded HLA-A2 was released from TAPBPR upon incubation with the NLVPMVATV peptide but not in the absence of the peptide ( lanes 2 and 1 respectively Figure 7 ) . Only trace amounts of peptide loaded HLA-A2:β2m molecules were released from TAPBPR using the Δ2 and Δ9 peptides ( lane 3 and 4 , Figure 7 ) , while no peptide loaded HLA-A2 was released from TAPBPR using the Δ2/9 peptide ( lane 5 Figure 7 ) . Together these results demonstrate that MHC I molecules bound to TAPBPR are in a peptide-receptive state and that the association between TAPBPR and MHC I is reduced by high affinity peptide resulting in peptide loaded MHC I being released from TAPBPR . 10 . 7554/eLife . 09617 . 013Figure 7 . The MHC I bound to TAPBPR is peptide receptive . The TAPBPR:MHC I complex was immunoprecipitated from IFN-γ induced KBM-7 cells . Equal aliquots were divided , then incubated -/+ 100 μM ( 20 μg ) of the indicated peptide ( WT:NLVPMVATV , Δ2: NAVPMVATV , Δ9: NLVPMVATM , Δ2/9:NAVPMVATM ) for 30 min at 4°C . Subsequently all eluates ( - or + peptide ) were re-immunoprecipitated with BB7 . 2 . Extensive washing was performed to remove any released MHC I before denaturation . Western blot analysis was performed for TAPBPR , and HLA-A2 ( using HCA2 ) under reducing conditions . Data shown are representative of three independent experiments . DOI: http://dx . doi . org/10 . 7554/eLife . 09617 . 013 HLA allomorphs differ in their dependency on tapasin for efficient surface expression . Previous studies have reported that , although the quantity of naturally processed peptides stably bound by HLA-A*02:01 is significantly reduced in the absence of tapasin ( Barber et al . , 2001 ) , surface expression of HLA-A2 is relatively unaffected in the absence of tapasin ( Greenwood et al . , 1994; Barber et al . , 2001 ) . In contrast , HLA-B7 expression is reduced in the absence of tapasin ( Rizvi et al . , 2014 ) . As TAPBPR is also involved in peptide selection on MHC class I molecules , we sought to examine the effect of TAPBPR depletion on surface expression of both HLA-A2 and HLA-B7 in the presence and the absence of tapasin . To investigate this , 721 . 221 ( tapasin positive , MHC class I negative ) ( Shimizu et al . , 1988 ) and 721 . 220 ( tapasin negative , HLA-A , -B negative , but HLA-Cw1 positive ) ( Greenwood et al . , 1994 ) cells were transduced with HLA-A2 and -B7 and then depleted of TAPBPR using shRNA . This resulted in a significant , although not complete , reduction of TAPBPR expression ( Figure 8A ) . Tapasin expression in 721 . 221 cells was not altered in cells expressing shTAPBPR ( Figure 8A ) . Consistent with our previous findings in tapasin deficient KBM-7 cells ( Hermann et al . , 2013 ) , we found an increased interaction between HLA-A2 and TAPBPR in the absence of tapasin in the 721 cell line series ( compare TAPBPR IP lanes 1 and 3 , Figure 8A ) which was apparent even though the expression of TAPBPR appeared to be slightly higher in . 220 cells compared to . 221 cells . This phenomenon does not appear to be restricted to HLA-A2 , as we similarly observed an increased interaction between HLA-B7 and TAPBPR in the absence of tapasin ( compare TAPBPR IP lanes 1 and 3 , Figure 8A ) . The increased association between TAPBPR and MHC class I occurs despite the fact that the steady state level of both HLA-B7 and HLA-A2 is severely reduced in the absence of tapasin ( compare lysate lanes 1 and 3 , Figure 8A ) . We found that the steady state level of HLA-A2 at the cell surface was not affected by TAPBPR depletion in tapasin sufficient cells ( left panel Figure 8B ) . However , the depletion of TAPBPR in tapasin negative cells resulted in a ∼75% decrease in cell surface expression of HLA-A2 ( right panel Figure 8B ) . In contrast , using an antibody to the Bw6 epitope we observed a significant decrease in the surface expression of HLA-B7 upon depletion of TAPBPR , even in the presence of tapasin ( left panel Figure 8C ) . This is despite the poor ability of HLA-B7 to co-immunoprecipitate with TAPBPR , as compared with HLA-A2 with TAPBPR ( Figure 8A ) . Only residual expression of HLA-B7 was observed upon the depletion of TAPBPR in the tapasin negative cell 721 . 220 ( right panel Figure 8C ) . These results suggest that HLA allomorphs differ in their dependency on TAPBPR , and that tapasin and TAPBPR work together to optimise surface expression of MHC class I molecules . 10 . 7554/eLife . 09617 . 014Figure 8 . TAPBPR depletion reduces MHC class I stability . ( A ) The MHC class I negative cell line 721 . 221 and the tapasin and HLA-A and -B negative cell line 721 . 220 were transduced with HLA-A2 and –B7 . shRNA specific for TAPBPR ( shTAPBPR ) was used to produce TAPBPR depleted versions of these cell lines . TAPBPR was isolated by immunoprecipitation from these four cell lines . Western blot analysis was performed for TAPBPR , tapasin , HLA-B ( 3B10 . 7 ) , HLA-A2 ( HCA2 ) , and calnexin on lysates and TAPBPR immunoprecipitates as indicated . This data is representative of three independent repeats . Cytofluorometric analysis of ( B ) HLA-A2 ( detected with BB7 . 2 ) or ( C ) HLA-B7 ( detected with anti-Bw6 antibody ) on . 221 ( A2+ , B7+ ) ( grey filled histogram ) , . 221 ( A2+ , B7+ ) shTAPBPR ( red line histogram ) , . 220 ( A2+ , B7+ ) ( blue line histogram ) , . 220 ( A2+ , B7+ ) shTAPBPR ( green line histogram ) . Staining on the non-transduced 721 . 221 and 721 . 220 cells with BB7 . 2 and Bw6 ( grey solid line ) or with an isotype control ( grey dashed line ) are included as controls . The data is representative of three independent experiments . ( D ) Thermal stability of HLA-A2 expressed in . 221 ( A2+ , B7+ ) and . 220 ( A2+ , B7+ ) -/+ stable depletion of TAPBPR ( shTAPBPR ) . Cells were radiolabelled for 60 min with [35S] cysteine/methionine , lysed , then equal aliquots of cleared lysates were either kept at 4°C or heated at 22°C , 37°C or 50°C for 12 min . Peptide loaded HLA-A2 was immunoprecipitated using BB7 . 2 post-preclear . After separation by SDS-PAGE , the signal intensity of the radiolabelled HLA-A2 band was determined by phosphorimaging . The results are representative of four independent experiments . The graph shows the percentage of peptide loaded HLA-A2 recoverable at each temperature as a percentage of the signal intensity at 4°C . Error bars show SEM from four independent experiments . Surface expression of ( E ) HLA-A2 ( detected with BB7 . 2 ) and ( F ) HLA-B7 ( detected with BB7 . 1 ) on cells treated with 5 µg/ml brefeldin A ( BFA ) , which inhibits egress of newly assembled molecules from the ER , for 0 , 0 . 5 , 3 , 6 and 16 hr . The level of remaining HLA-A2 and HLA-B7 at each time point is expressed as percentage of the mean fluorescence at time 0 . Error bars represent SEM of duplicate samples and the data is representative of three independent experiments . DOI: http://dx . doi . org/10 . 7554/eLife . 09617 . 014 Comparison of the anchor residues in TAPBPR-permitted versus TAPBPR-restricted peptides suggests TAPBPR enhances the selection of peptides with canonical anchor residues and consequently may help reduce the cell surface presentation of peptides with lower affinity ( Figure 5E–H ) . To test directly whether HLA molecules are indeed loaded with low affinity peptides in the absence of TAPBPR , we compared the stability of MHC class I molecules in the presence and absence of TAPBPR . First , the thermal stability of HLA-A2 molecules expressed in TAPBPR-competent and TAPBPR-depleted cells in the presence or absence of tapasin was investigated by heating lysates of radiolabelled cells for a short time , followed by immunoprecipitation with BB7 . 2 ( specific for peptide loaded HLA-A2 ) . We found that the thermal stability of HLA-A2 was reduced in TAPBPR depleted cells ( compare grey bar with the red bar in Figure 8D ) . As the thermal stability of MHC class I correlates with the affinity of its peptide cargo ( Williams et al . , 2002 ) , these results support the hypothesis that TAPBPR assists in the selection of stable peptides on MHC class I molecules . However , the effect of TAPBPR depletion on HLA-A2 thermal stability was not as severe as that observed in the absence of tapasin ( compared red and blue bars in Figure 8B ) . The thermal stability of peptide loaded HLA-A2 was further decreased when TAPBPR was depleted in tapasin negative cells ( green bar in Figure 8B ) . These results are consistent with the decreased steady state levels of HLA-A2 observed in TAPBPR depleted and tapasin deficient cells as shown in Figure 8A . Next , to examine whether MHC class I complexes with lower stability are released onto the cell surface in the absence of TAPBPR , we examined the decay rates of MHC class I molecules expressed at the cell surface in TAPBPR-competent and TAPBPR-depleted cells . Both HLA-A2 and –B7 exhibited faster decay rates from the cell surface in TAPBPR-depleted cells ( Figure 8E‚F ) , suggesting that in the absence of TAPBPR MHC class I molecules containing lower affinity peptides can escape to the cell surface . For HLA-A2 , tapasin deficiency had a more severe effect than TAPBPR depletion on the rate of decay ( compare the blue line with the red line in Figure 8E ) while for HLA-B7 , tapasin deficiency and TAPBPR depletion had a similar effect on the decay rate ( see the overlap of the blue and red line in Figure 8F ) . For both HLA-A2 and –B7 the loss of both tapasin and TAPBPR had an accumulative effect on MHC class I stability ( see green line in Figure 8E , F ) . Together , these results clearly demonstrate a role for TAPBPR in the selection of stable peptides for MHC class I molecules in vivo and support the hypothesis that TAPBPR can enhance peptide optimisation in the absence of tapasin .
The PLC is considered to be the major site where MHC class I molecules are loaded with high affinity peptides , with the tapasin-ERp57 conjugate described as the functional unit ( Momburg and Tan , 2002; Williams et al . , 2002; Howarth et al . , 2004; Wearsch and Cresswell , 2007 ) . Our experiments indicate that TAPBPR is also involved in peptide selection for MHC class I molecules . Like tapasin , TAPBPR can catalyse the dissociation of peptides from peptide-MHC I complexes , enhance the loading of peptide-receptive MHC I molecules , and discriminate between peptides based on affinity in vitro . Therefore , it is now apparent that there are two MHC class I specific peptide exchange catalysts in the antigen presentation pathway intimately involved in selecting peptide for presentation to the immune system . We found that the luminal domains of TAPBPR were able to efficiently function in MHC class I peptide selection in vitro . This suggests that the luminal region of TAPBPR alone has sufficient affinity for MHC class I to promote peptide loading and to catalyse peptide exchange and does not require other co-factors or to be artificially tethered to MHC class I to perform this function . This is in contrast to soluble tapasin , which others have found does not associate productively with MHC class I in vitro alone , and instead requires either conjugation to ERp57 or to be artificially tethered to MHC class I ( Chen and Bouvier , 2007; Wearsch and Cresswell , 2007 ) . We speculate the difference in affinity of tapasin and TAPBPR for MHC class I may have evolved as a consequence of the distinct cellular environment the two peptide exchange catalysts function: tapasin functioning in the context of the PLC , where other proteins collectively form a mutually supportive interaction scaffold , and TAPBPR functioning outside the confines of the PLC . Although both tapasin and TAPBPR clearly function as peptide exchange catalysts in vitro , the activities of the two proteins are not identical . Firstly , we observed differences in the ability of tapasin-jun and TAPBPR to promote dissociation of peptides from HLA-A*02:01 . Secondly , we found TAPBPR was unable to promote the binding of ELRSRK*WAI or FLRGRK*YGL to peptide receptive HLA-B*08:01fos molecules , whereas Chen & Bouvier found that tapasin-jun enhanced the binding of FITC labelled variants of both these peptides ( Chen and Bouvier , 2007 ) . This indicates there are some qualitative differences between the activity of TAPBPR and tapasin . Clearly , a new picture concerning how peptides are selected for presentation by MHC class I molecules is emerging . Our findings are consistent with tapasin and TAPBPR working together in a sequential manner ( Figure 9 ) . The initial peptide loading of MHC I molecules is likely to occur within the PLC and to be assisted by tapasin . Within this relatively peptide-rich environment , peptide loading and exchange mediated by tapasin increases the affinity of peptide bound to MHC class I . We propose that outside of the PLC , the stability of peptide:MHC I complexes is monitored by TAPBPR . Those molecules loaded with high affinity peptides will either circumvent TAPBPR-mediated peptide editing completely or pass through this quality control checkpoint . For MHC class I molecules loaded with peptides of intermediate or low affinity , TAPBPR-mediated peptide editing proceeds . The unique environments in which tapasin and TAPBPR have evolved to operate in are likely to shape the peptide repertoire in different ways . The relatively peptide-rich environment of the PLC may allow tapasin to promote peptide binding and produce a relatively broad peptide repertoire . However , presumably as an MHC I molecule moves away from the PLC , the gradient of optimal peptides is likely to decrease and therefore TAPBPR-mediated peptide editing events within such an environment may favour peptide dissociation rather than association . Thus we speculate that , in a peptide rich environment , acceleration of both association and dissociation ( by tapasin ) contributes to editing; while in a peptide dilute environment , acceleration of dissociation only ( by TAPBPR ) may dominate editing , consistent with the increased peptide repertoire observed in TAPBPR deficient cells . Thus , TAPBPR may restrict peptide presentation by enhancing the selection of high affinity peptide on MHC class I . 10 . 7554/eLife . 09617 . 015Figure 9 . Model of the relationship between tapasin and TAPBPR in shaping the peptide repertoire expressed on MHC class I . ( A ) Upstream of the PLC , in an environment in which peptides suitable for MHC class I binding are likely to be at a low concentration , TAPBPR might help stabilise a peptide-receptive conformation of MHC class I , but would not directly assist in peptide loading , peptide exchange or directly increase the abundance of peptide:MHC complexes due to a shortage of peptide . ( B ) The optimal peptide loading environment for MHC class I is presumably within the PLC where the concentration of peptides suitable for MHC class I is the highest and where MHC I are held in a peptide-receptive conformation by tapasin . ( C ) Once peptide:MHC I complexes are released from the PLC , TAPBPR-mediated dissociation would attempt to remove peptide from MHC class I . If a high affinity peptide was bound to the MHC I molecules , TAPBPR would not be able to remove it . If TAPBPR can cause the dissociation of peptide , peptide exchange could ensue if suitable replacement peptides were available . However , the further away from the PLC and/or the further down the MHC class I peptide gradient TAPBPR works there may be a shortage of suitable replacement peptides . In this situation TAPBPR is unlikely to load the peptide-receptive MHC class I and may assist in the recycling of MHC I molecules . By functioning as a peptide exchange catalyst in a relatively peptide deficient environment , TAPBPR could increase the affinity of the peptide:MHC I complexes or restrict the diversity of peptide:MHC I complexes which are presented to the cell surface . DOI: http://dx . doi . org/10 . 7554/eLife . 09617 . 015 Further in-depth analysis is required to dissect the intricate relationship between tapasin and TAPBPR in determining the final peptide repertoire presented on the cell surface . In particular it will be important to understand how different MHC class I allomorphs depend on these two specific chaperones . Intriguingly , we have observed differences in the functional effects of TAPBPR on peptide selection by HLA-A2 and –B7 . For HLA-A2 , TAPBPR catalyses both the peptide dissociation and association in vitro , in a similar manner described for the function of tapasin , albeit with different peptide preference . Therefore , there may be shared , but not identical , functionality between tapasin and TAPBPR regarding peptide selection onto HLA-A2 , i . e . in a peptide abundant environment both chaperones could potentially load and edit peptides bound to HLA-A2 . This may explain , at least in part , why surface expression of HLA-A2 is relatively tapasin-independent ( Lewis et al . , 1998 ) . Surface expression of HLA-A2 is also TAPBPR-independent in the presence of tapasin , and it is not until tapasin and TAPBPR are absent or depleted that a severe effect is observed on HLA-A2 surface expression . In the absence of both tapasin and TAPBPR , surface expression of HLA-A2 resembles that of the HLA-A2 T134K mutant ( Lewis et al . , 1996; Peace-Brewer et al . , 1996 ) . Therefore the discovery of a peptide editing function for TAPBPR helps explain the previous discrepancies regarding the phenotype of the T134K mutant ( in which the HLA-A2 mutant is unable to interact with either tapasin or TAPBPR and exhibits low cell surface expression ) and tapasin-deficient cells ( in which the HLA-A2 binds TAPBPR and exhibits high cell surface expression ) . Our results clearly demonstrate that TAPBPR also plays a significant role in peptide selection by HLA-B molecules . This importance of TAPBPR in HLA-B biology was underappreciated previously when investigating the interaction between TAPBPR and HLA-B in HeLa cells ( Boyle et al . , 2013 ) . Although all the HLA-B molecules investigated here ( HLA-B7 , -B8 , -B15 and -B40 ) exhibit weak or transient association with TAPBPR as compared to HLA-A2 in wild-type cells , the interaction between HLA-B7 and TAPBPR was significantly increased in the absence of tapasin as shown in Figure 8 . TAPBPR clearly influenced peptide selection on all of these HLA-B molecules: fluorescent polarisation experiments demonstrate that TAPBPR functions as a peptide loading and exchange catalyst for EIYK*RWIIL on HLA-B8; TAPBPR deficiency in IFN-γ induced HeLa and KBM-7 cells alters the peptide repertoire expressed by HLA-B15 and HLA-B40 respectively; and depletion of TAPBPR in 721 . 221 results in a significant decrease in the surface expression of HLA-B7 , even in the presence of tapasin . Taken together , these results suggest that TAPBPR is responsible for restricting peptides on HLA-B allomorphs . Clearly , peptide selection by MHC class I molecules is far more complex than previously anticipated . Given that a single T-cell receptor ( TCR ) can in principle see many different peptide-MHC class I complexes ( Sewell , 2012; Wooldridge et al . , 2012 ) , the two MHC specific chaperones of the MHC class I antigen presentation pathway , may be the crucial influence in selecting immune responses .
The luminal domains of TAPBPR and TAPBPRTN5 ( I261K ) ( Hermann et al . , 2013 ) were cloned into pHLsec containing a C-terminal His-tag and an N-terminal signal sequence ( Aricescu et al . , 2006 ) . To produce secreted TAPBPR and TAPBPRTN5 , 250 μg of the TAPBPR pHLsec plasmids were mixed with 1 . 5 ml of 1 mg/ml PEI 25K ( Sigma , St Louis , MO ) in 25 ml phosphate-buffered saline ( PBS ) , followed by incubation at room temperature for 20 min . The solution was added dropwise to HEK293F cell in FreeStyleTM 293 Expression Medium ( Gibco , Thermo Fisher Scientific , UK ) seeded at a density of 1 × 106 cells/ml in a volume of 250 ml in a 1 L Erlenmeyer conical tissue flask and incubated for three days at 37°C under constant shaking at 125 rpm . The cell culture supernatant was harvested , filtered and purified using Ni-NTA affinity chromatography ( Invitrogen , Thermo Fisher Scientific ) . Beads were isolated , washed in 20 mM Tris pH 7 . 4 , 200 mM NaCl , 20 mM Imidazole and TAPBPR was eluted with 20 mM Tris pH 7 . 4 , 200 mM NaCl , 300 mM Imidazole . Protein-containing fractions were analysed by SDS-PAGE followed by Coomassie staining , pooled and concentrated . The concentrate was exchanged into 50 mM Tris , 150 mM NaCl pH 8 . 0 buffer and separated on a Superdex S200 10/300 column ( GE Heathcare , UK ) by size exclusion chromatography . Protein-containing fractions were analysed by SDS-PAGE followed by Coomassie staining , pooled and further concentrated . The concentrate was snap frozen in liquid nitrogen and stored at -80°C . DSF experiments were performed in 48-well plates using 50 µL reactions consisting of 2 µg of purified TAPBPR protein and 5x SyPRO orange dye ( Invitrogen molecular probes , Thermo Fisher Scientific ) in PBS pH 7 . 4 . The melt curve was performed using Bio-Rad MiniOpticon reverse transcription-polymerase chain reaction ( RT-PCR ) thermal cycler between 20°C and 95°C in 1°C steps with 20 s equilibration time per step . The protein melting temperature ( Tm ) was taken as the inflexion point of the sigmoidal melting curve , obtained by curve fitting using DSF scripts ( Niesen et al . , 2007 ) and GraphPad Prism software . pHN1+ plasmids encoding the mature human beta 2-microglobulin protein ( β2m hereafter ) or the ER luminal domains of HLA A*02:01 protein with C-terminal BirA motif were obtained from Prof P Moss ( University of Birmingham , UK ) . pGM-T7 plasmids encoding the ER luminal domains of wild-type HLA B*08:01 or the T134K point mutant , each with C-terminal fos leucine zipper sequences , were provided by Prof M Bouvier ( Chen and Bouvier , 2007 ) . The sequence encoding HLA-B*08:01fos or the T134K point mutant were excised by restriction enzyme digestion and sub-cloned into pET22b plasmid ( Novagen , Merck Millipore , Germany ) . DNA encoding the ER luminal domains of HLA A*02:01 with C-terminal Fos leucine zipper sequence was created by PCR: nucleotides encoding the ER luminal domains of A*02:01 were amplified using primers 5'-GATATACCATGGGCTCTCACTCC-3' and 5'-CGGAACCTCCCTCCCATC-3’ and pHN1 A*02:01 DNA; while nucleotides encoding the Fos leucine zipper were amplified from pET22b HLA B*08:01fos using primers 5'- GGAGGTTCCGGCGGTC-3' and 5'-CGCAAGCTTTTAATGGGCGG-3' . The purified products from both PCR reactions were used in a third PCR reaction to create A*02:01fos using primers 5'-GATATACCATGGGCTCTCACTCC-3' and 5'-CGCAAGCTTTTAATGGGCGG-3' . Following agarose gel electrophoresis and digestion of the purified product with restriction enzymes the sequence encoding HLA-A*02:01fos was cloned into pET22b . DNA encoding His tagged human tapasin-jun ( Chen and Bouvier , 2007 ) was amplified by PCR using 5'-GTCAGATCTGGACCCGCGGTGATCG-3' and 5'-CTCGGTACCCTAATGGTGATGGTGATG-3' primers , and ligated to pMT/BiP plasmid ( Invitrogen , Thermo Fisher Scientific ) following agarose gel electrophoresis and digestion of the purified product with BglII and KpnI restriction enzymes . The His6 tag present in the Jun leucine zipper portion was then replaced by PCR mutagenesis with an HA epitope using 5'-GTCAGATCTGGACCCGCGGTGATCG-3' and 5'-GTAGGTACCCTAAGCGTAGTCTGGGACGTCGTATGGGTAGTTCATGACTTTCTG-3' primers . The resulting DNA ( encoding the luminal domains of human tapasin , GGSGG linker , thrombin cleavage site , Jun leucine peptide , HA tag , and stop codon ) was transferred by restriction enzyme digestion to pMT/BiP plasmid modified to encode puromycin resistance . Stable polyclonal transfectants of S2 cells were obtained by transfecting 1 µg of tapasin-jun DNA using Fugene 6 ( Roche Applied Science , UK ) , and puromycin selection . Transfectants were adapted to EX-CELL 420 Serum-Free Medium ( Sigma ) , and tapasin-jun expression was induced with 500 µM CuSO4 . Supernatants were harvested 6 days later . Tapasin-jun was captured using anti-HA-agarose ( Sigma ) , washed with 20 mM Tris pH7 . 4 , 150 mM NaCl and eluted using 1 mg/ml HA peptide in 20 mM Tris pH 7 . 4 , 150 mM NaCl . SDS PAGE electrophoresis and Coomassie staining was used to select fractions containing high concentrations of tapasin-jun , which were dialysed against 20 mM Tris-HCl pH 7 . 5 , 150 mM NaCl at 4°C . The protein was snap frozen in liquid nitrogen and stored at -80°C . The following HLA-A*02:01 binding peptides were used: the UV-labile peptide KILGFVFjV ( j represents 3-amino-3- ( 2-nitro ) phenyl-propionic acid ) , the fluorescent peptides FLPSDC*FPSV , KLWEAESK*L , FLLAEDTK*V , KLVK*EVIAV , YLVAEK*VTV , GLDDIKDLK*V , YLENGK*ETL ( C* and K* denotes TAMRA labelled cysteine or lysine ) , non-labelled peptides FLPSDCFPSV , NLVPMVATV and three variants of this peptide NAVPMVATV ( Δ2 ) , NLVPMVATM ( Δ9 ) , NAVPMVATM ( Δ2/9 ) . The following HLA-B*08:01 binding peptides were used: the UV-labile peptide FLRGRAjGL , the fluorescent peptides ELRSRK*WAI , EIYK*RWIIL or FLRGRK*YGL ( K* denotes TAMRA labelled lysine ) , and the non-labelled peptides ELRSRKWAI , EIYKRWIIL or FLRGRKYGL . Tamra labelled and unlabelled peptides used in fluorescence polarisation were synthesised by GL Biochem Ltd ( Minhang , Shanghai ) . All other peptides were synthesised by Peptide Protein Research Ltd ( Funtley , Fareham , UK ) . Peptide-loaded MHC I or MHC I-fos complexes were obtained as in ( Garboczi et al . , 1992 ) by refolding solubilized inclusion bodies of MHC I or MHC I-fos heavy chains with solubilized inclusion bodies of human β2m and UV-labile MHC class I specific peptides . Fluorescence polarization measurements were taken using an Analyst AD ( Molecular Devices ) with 530 nm excitation and 580 nm emission filters and 561 nm dichroic mirror . All experiments were conducted at room temperature in duplicate and used PBS supplemented with 0 . 5 mg/ml bovine gamma-globulin ( Sigma ) and 0 . 5 mM dithiothreitol , in a volume of 60 μl . Binding of TAMRA-labelled peptide is reported in millipolarisation units ( mP ) and is obtained from the equation , mP = 1000 × ( S - G × P ) / ( S + G × P ) , where S and P are background-subtracted fluorescence count rates ( S = polarised emission filter is parallel to the excitation filter; P = polarised emission filter is perpendicular to the excitation filter ) , and G ( grating ) is an instrument- and assay-dependent factor . The following TAPBPR specific antibodies were used: PeTe4 ( a mouse mAb raised against the luminal domains of human TAPBPR which recognise a native conformation of the protein ) ( Boyle et al . , 2013 ) , R014 ( a rabbit polyclonal raised against the luminal domains of human TAPBPR ) , R021 ( a rabbit polyclonal raised against the cytoplasmic tail of human TAPBPR ) , a mouse anti-TAPBPR mAb raised against the membrane distal domain of TAPBPR ( ab57411 , Abcam , UK ) . The following MHC class I specific antibodies were used: HC10 ( a mouse mAb that recognises a PxxWDR motif at aa 57–62 in the α1 domain of the MHC class I heavy chain , found commonly in HLA-B and –C , and also in a few HLA-A allomorphs ) ( Stam et al . , 1986; Perosa et al . , 2003 ) , HCA2 ( a mouse mAb that recognises HLA allomorphs containing a xLxTLRGx motif at aa 77–84 on the α1 domain ) ( Stam et al . , 1990 Sernee et al . , 1998 ) , 3B10 . 7 ( a rat monoclonal antibody with broad specificity for HLA I independent of conformation ) ( Lutz and Cresswell , 1987 ) , BB7 . 2 ( a mouse mAb specific for β2m bound and peptide loaded HLA-A2 ) , Bw6 ( a mouse mAb specific for MHC class I allomorphs containing the Bw6 epitope SLRNLRG at aa 77–83 ) ( One Lambda , Thermo Fisher Scientific , Canoga Park , CA ) , BB7 . 1 ( a mouse mAb specific for HLA-B7 ) ( AbDSerotec ) , W6/32 ( a pan MHC I mAb which recognises a conformation epitope on the MHC I α2 domain , dependent on β2m and peptide ) ( Barnstable et al . , 1978 ) , 4E ( a mouse monoclonal reactive against HLA-B as well as a limited number of HLA-A molecules ( A29 , Aw30 , Aw31 , Aw32 ) ( Yang et al . , 1984 ) , B1 . 23 . 2 ( a mouse mAb that recognises both β2m association and -free HLA-B , -C as well as some HLA-A heavy chains ) ( Rebaï and Malissen , 1983; Apps et al . , 2009 ) . Other antibodies used were Pasta1 ( a mAb which recognises native tapasin ) ( Dick et al . , 2002 ) and Rpg48N ( tapasin specific rabbit polyclonal ) ( both kind gifts from Peter Cresswell , Yale University School of Medicine , New Haven , CT ) , and rabbit anti-calnexin ( Enzo life Sciences , UK ) . Primary antibodies were detected with horseradish peroxidase ( HRP ) conjugated secondary antibodies ( Dako , UK ) and goat-anti-mouse Alexa 647 ( molecular probes ) . In addition isotype control antibodies were also used ( Dako ) . HEK293T were maintained in Dulbecco's modified eagle's medium ( DMEM;Gibco ) , HeLa-M , HeLa-S , HeLa-S depleted of TAPBPR using shRNA ( Boyle et al . , 2013 ) , 721 . 221 and 721 . 220 cells were maintained in Roswell Park Memorial Institute ( RMPI ) 1640 media ( Gibco ) , and KBM-7 cells were maintained in Iscove's Modified Dulbecco's Medium ( Gibco ) all supplemented with 10% fetal calf serum , 100 U/ml penicillin and 100 µg/ml streptomycin ( Gibco ) at 37°C and 5% CO2 . To induce the expression of endogenous TAPBPR , cells were treated with 50 U/ml of IFN-γ ( Peprotech , UK ) at 37°C for 48–72 hr . HLA-A*02:01 , -A*68:01 , -B*08:01 , -B*15:10 and –B*40:02 were cloned into the lentiviral vector pHRSINcPPT-SGW . HLA-B*07:02 was cloned into the lentiviral vector pHRSIN-C56W-UbEM producing HLA-B7 under the spleen focus-forming virus promotor and the green fluorescent protein ( GFP ) -derivative protein emerald under an ubiquitin promotor . For TAPBPR depletion , the lentiviral shRNA plasmid V2LHS_135531 on the pGIPZ backbone was used ( Open Biosystems , GE healthcare ) . Lentiviral plasmids were transfected into HEK-293T cells along with the packaging plasmid pCMVR8 . 91 and envelope plasmid pMD-G using TransIT-293 ( Mirus , Madison , WI ) . Forty eight and seventy two hours post-transfection , the filtered supernatants were used to transduce HeLa , 721 . 221 or 721 . 220 cells . For HLA molecules transduced into 721 . 221 , the transduction efficiency was determined by the surface expression of MHC class I using W6/32 antibody with flow cytometry . To make the 721 . 221 and 721 . 220 expressing both HLA-A2 and –B7 , HLA-A2 was first stably transduced into the cell lines with transduction efficiency determined by surface expression of HLA-A2 using BB7 . 2 antibody with flow cytometry . These HLA-A2 positive cells were subsequently transduced with HLA-B7 pHRSIN-C56W-UbEM with transduction efficiency determined by GFP expression in the cells . To select cells depleted of TAPBPR puromycin selection was used post transduction with V2LHS_135531 . HeLa TAPBPR KO lines were generated using CRISPR as previously described ( Ran , et al . , 2013 ) . To target TAPBPR expression , sgRNAs were chosen using the CRISPR design tool ( http://crispr . mit . edu ) , which bind in exon 2 of TAPBPR ( Crispr7: GCGAAGGACGGTGCGCACCG ) and cloned into pSpCas9 ( BB ) -2A-Puro . To generate HeLa TAPBPR KO cell lines , cells were seeded at a confluence of 80% in 6 well format and transfected with TAPBPR-CRISPR plasmids in the absence of serum using Lipofectamine2000 ( Invitrogen , Thermo Fisher Scientific ) . Twenty four hours after transfection , the medium was replaced with complete DMEM containing 4 µg/ml puromycin ( Invivogen , San Diego , CA ) . After 48 hr , the medium was replaced with complete DMEM without puromycin . Subsequently , the clones were selected and the absence of TAPBPR protein was verified by immunoprecipitation/western blot analysis . Harvested cells were washed in PBS , pelleted , then lysed in either 1% digitonin ( Merck Millipore ) Tris-buffered saline ( TBS ) ( 20 mM Tris-HCl pH7 . 4 , 150 mM NaCl , 5 mM MgCl2 , 1 mM ethylenediaminetetraacetic acid ) supplemented with 10 mM N-ethylmaleimide ( NEM ) ( Sigma ) , 1mM phenylmethylsulfonyl fluoride ( PMSF; Sigma ) and protease inhibitor cocktail ( Roche ) for 30 min at 4°C . Nuclei and cell debris were pelleted by centrifugation at 13 , 000 × g for 10 min and supernatants were precleared on IgG-sepharose ( GE Healthcare , United Kingdom ) and Protein A Sepharose ( Generon , UK ) for 1 hr at 4°C with rotation . Immunoprecipitation was performed with the indicated antibody and Protein A Sepharose for 1 hr at 4°C with rotation . Following immunoprecipitation , the beads were washed thoroughly in 0 . 1% detergent-TBS to remove the unbound protein . All samples were heated at 80°C for 10 min in reducing sample buffer ( 125 mM Tris-HCL pH 6 . 8 , 4% SDS , 20% glycerol , 0 . 04% bromophenol blue with 100 mM β-mercaptoethanol ) . Proteins were separated by SDS-PAGE and transferred onto an Immobilon transfer membrane ( Millipore , Billerica , MA ) . Membranes were blocked using 5% ( w/v ) dried milk , 0 . 1% ( v/v ) Tween 20 in PBS for 30 min , followed by incubation with the indicated primary antibody for 1 hr . After washing , membranes were incubated with species-specific HRP conjugated secondary antibodies , before detection by enhanced chemiluminescence reagent ( GE Healthcare ) . Cells were starved in methionine and cysteine free RPMI for 30 min at 37°C , then labelled using EasyTag Express [35S]-protein labeling mix ( Perkin Elmer ) for 60 min at 37°C . Cells were lysed in 1% Triton X-100 ( Sigma ) TBS containing 10 mM NEM , 1 mM PMSF and protease inhibitors for 30 min at 4°C . Equal aliquots of clarified cell lysates were either kept at 4°C or heated at 22 , 37 or 50°C for 12 min , before returning to 4°C . Immunoprecipitation and SDS-PAGE were performed as above . Gels were subsequently fixed in 12% acetic acid , 40% methanol and dried . Images were obtained using a phosphor screen ( Perkin-Elmer ) or on film . PhosphorImager analysis was performed using Typhoon Trio variable mode imager ( GE Healthcare ) together with ImageQuantTL software . Densitometry of the MHC class I HC band was performed . The amount of recoverable HLA-A2 remaining at each temperature was determined as a percentage of the signal intensity at 4°C . Graphs were generated using GraphPad Prism software . High-resolution images were obtained using film . Cells were washed in PBS , then incubated for 20 min at 4°C with W6/32 , BB7 . 2 , BB7 . 1 , 4E or a HLA-Bw6 specific antibody . Isotype control antibodies were used as negative controls . Cells were washed in ice-cold PBS , then the bound primary antibody was subsequently detected with species-specific Alexa Fluor 647 secondary antibodies ( Molecular Probes ) . Cells were analysed on a BD FACS Calibur 4- ( BD Biosciences , East Rutherford , NJ ) colour analyser and data were analysed using FlowJo software . For MHC class I cell surface decay experiments , 5 µg/ml brefeldin A was incubated with cells for 0 , 0 . 5 , 3 , 6 , and 16 hr followed by staining with BB7 . 2 or BB7 . 1 . HLA ligands from 5 × 108 cells per cell line were isolated by immunoaffinity chromatography . Cells were lysed in buffer containing PBS , 0 . 6% 3-[ ( 3-cholamidopropyl ) dimethylammonio]-1-propanesulfonate ( CHAPS ) , and Complete protease inhibitor , shaken for 1 hr at 4°C and subsequently sonicated for 1 min . Following centrifugation for 1 . 5 hr to remove debris , the supernatant was applied on affinity columns overnight at 4°C . Columns were previously prepared by coupling antibodies to CNBr-activated Sepharose ( GE Healthcare , Buckinghamshire , England ) ( 1 mg antibody/40 mg Sepharose ) . The antibody W6/32 was used to isolate all HLA class I alleles in HeLa cells . For KBM-7 cells , BB7 . 2 was used to isolate HLA-A2 and B1 . 23 . 2 was used to isolate HLA-B and -C molecules . On the second day the columns were eluted in eight steps using 0 . 2% trifluoroacetic acid . The eluate was passed through a 10 kDa filter ( Merck Millipore , Darmstadt , Germany ) to yield the HLA-ligands in solution . The filtrate was desalted with C18 ZipTips ( Merck Millipore , Darmstadt , Germany ) and subsequently concentrated using a vacuum centrifuge ( Bachofer , München , Germany ) . Sample volume was adjusted for measurement by adding 1% ACN/0 . 05% TFA ( v/v ) . With an injection volume of 5 µl HLA ligands were loaded ( 100 µm × 2 cm , C18 , 5 µm , 100 Å ) and separated ( 75 µm × 50 cm , C18 , 3 µm , 100 Å ) on Acclaim Pepmap100 columns ( Dionex , Sunnyvale , CA ) using an Ultimate 3000 RLSCnano uHPLC system ( Dionex ) . A gradient ranging from 2 . 4 to 32% of ACN/H2O with 0 . 1% formic acid was used to elute the peptides from the columns over 140 min at a flow rate of 300 nl/min . Online electrospray ionisation ( ESI ) was followed by tandem mass spectrometry ( MS ) analysis in a LTQ Orbitrap XL instrument ( ThermoFisher Scientific , Bremen , Germany ) . Survey scans were acquired in the Orbitrap mass analyzer with a resolution of 60 , 000 and a mass range of 400–650 m/z . Peptides with a charge state other than 2+ or 3+ were rejected from fragmentation . Fragment mass spectra of the five most intense ions of each scan cycle were recorded in the linear ion trap ( top5 CID ) . Normalised collision energy of 35% , activation time of 30 ms and isolation width of 2 m/z was utilised for fragment mass analysis . Dynamic exclusion was set to 1 s . The rawwere processed against the human proteome as comprised in the Swiss-Prot database ( www . uniprot . org , status: Dec12th , 2012; 20 . 225 reviewed sequences contained ) using MASCOT server version 2 . 3 . 04 ( Matrix Science , Boston , MA ) and Proteome Discoverer 1 . 4 ( Thermo Fisher Scientific ) . Oxidation of methionine was allowed as dynamic peptide modification . A mass tolerance of 5 ppm or 0 . 5 Da was allowed for parent- and fragment masses respectively . Filtering parameters were set to a Mascot Score<20 , search engine rank = 1 , peptide length of 8-12 AA , achieving a false discovery rate ( FDR ) of 5% as determined by an inverse decoy database search . The netMHC programme ( http://www . cbs . dtu . dk/services/NetMHC/ ) was used to predict the binding of peptides to specific HLA allomorphs ( Lundegaard et al . , 2008a , 2008b ) . To determine classic peptide anchor conservation , TAPBPR permitted peptides ( shared peptides ) and TAPBPR restricted peptides ( peptides unique to HeLa-TAPBPR KO cells or peptides unique to HeLa cells ) were compared and the frequency of classic anchors at P2 or PΩ calculated for both HLA-A*68:02 or HLA-B*15:03 . | Our immune system protects us from infections and destroys cells that are turning cancerous . A group of proteins called MHC class I molecules are essential for this protection . These molecules let the immune system know what is going on inside our cells by displaying chopped up fragments of proteins ( or peptides ) on the surface of the cell . If these peptides are from infectious or disease-causing agents the immune system is triggered into action and can recognise and kill the cell . There is still much to discover regarding how MHC molecules choose which peptides to display . A very complex pathway within our cells controls this displaying of peptides to the immune system . Recently , a protein called TAPBPR was identified as a new player in MHC class I biology , but its role was unclear . Hermann , van Hateren et al . now reveal that TAPBPR plays a central role in restricting which peptides are loaded onto and presented by MHC class I molecules . The results suggest that TAPBPR acts as a quality control checkpoint , closely monitoring and ensuring that the loaded peptide is stable within the MHC class I molecule . The discovery of TAPBPR's role in peptide selection increases our understanding of how peptides are chosen and stabilized , and sets the stage for learning more about how cells decide which peptides to reveal to the immune system . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"cell",
"biology",
"immunology",
"and",
"inflammation"
] | 2015 | TAPBPR alters MHC class I peptide presentation by functioning as a peptide exchange catalyst |
Ammonium serves as key nitrogen source and metabolic intermediate , yet excess causes toxicity . Ammonium uptake is mediated by ammonium transporters , whose regulation is poorly understood . While transport can easily be characterized in heterologous systems , measuring transporter activity in vivo remains challenging . Here we developed a simple assay for monitoring activity in vivo by inserting circularly-permutated GFP into conformation-sensitive positions of two plant and one yeast ammonium transceptors ( ‘AmTrac’ and ‘MepTrac’ ) . Addition of ammonium to yeast cells expressing the sensors triggered concentration-dependent fluorescence intensity ( FI ) changes that strictly correlated with the activity of the transporter . Fluorescence-based activity sensors present a novel technology for monitoring the interaction of the transporters with their substrates , the activity of transporters and their regulation in vivo , which is particularly valuable in the context of analytes for which no radiotracers exist , as well as for cell-specific and subcellular transport processes that are otherwise difficult to track .
Transport proteins play critical roles in cellular uptake and release as well as subcellular distribution of ions and metabolites . Due to their hydrophobic nature the characterization of transporters is challenging , and activity measurements typically depend on the use of radiotracers and heterologous expression systems . One of the major breakthroughs for characterizing channels has been the development of electrophysiology tools , enabling characterization of electrogenic transport in live cells ( Neher and Sakmann , 1992 ) . However , patch clamping has largely been limited to the analysis of processes at the cell membrane , or cells at the surface of sliced tissues . A new level of insight was provided by fluorescent proteins enabling us to observe cellular processes in living cells of intact tissues ( Tsien , 2006 ) . Fluorescence Resonance Energy Transfer ( FRET ) sensors have allowed us to monitor analyte levels and dynamics in live cells with minimal invasive methods ( Okumoto et al . , 2012 ) . Particular FRET sensors can even be used to monitor biophysical processes such as membrane potential or tension between molecules in live cells , as well as the conversion of substrates of enzyme reactions by proteases and kinases . Yet to date , we have not been able to directly follow the activity of transporters and to monitor their regulation in vivo . We rationalized that it may be possible to create transport activity sensors by monitoring structural rearrangements in a transporter with the help of environmentally sensitive genetically encoded fluorophores . Based on the pioneering work of Miyawaki et al . ( 1997 ) , FRET can be used to probe conformational rearrangements and has widely been exploited to develop genetically encoded sensors for small molecules . An alternative , again developed by Tsien's lab ( Baird et al . , 1999 ) , is the use of circularly permutated GFP variants as reporters for conformational rearrangements , which have been exploited to create sensors for small molecules such as calcium , zinc and maltose ( Baird et al . , 1999; Marvin et al . , 2011 ) . Circularly-permutated versions of GFP ( cpGFP ) have successfully been used to monitor structural rearrangements of calmodulin ( Baird et al . , 1999; Nagai et al . , 2001; Nakai et al . , 2001; Wang et al . , 2008; Akerboom et al . , 2009 ) . cpGFP is a protein in which the β-barrel structure has been rearranged in a way that ligand-induced conformational changes can influence the chromophore environment and consequentially the fluorescent properties of a chimeric sensor . Depending on pH , GFP and its variants have two excitation maxima ( λex ∼ 395 and 475 nm ) ( Tsien , 1998 ) . Protein-cpGFP fusions , in which structural rearrangements affect the state of the chromophore as well as the hydrogen bonds between the chromophore and the surrounding β-barrel , can in principle be exploited to develop sensors that monitor transporter activity . Here we tested the hypothesis that ammonium transporters can effectively report transporter activity in live cells when fused to cpGFP . Ammonium transport is key to nitrogen nutrition of bacteria , fungi and plants . Although mammals are unable to assimilate ammonium into amino acids , ammonium transporters play key roles in renal ammonium secretion and male fertility ( Biver et al . , 2008 ) . Transporters for ammonium are conserved in bacterial , fungal , plant and animal genomes . Ammonium transport is electrogenic and is mediated by high affinity ammonium transporters of the AMT/MEP/Rhesus protein superfamily ( Marini et al . , 1994; Ninnemann et al . , 1994; Tremblay and Hallenbeck , 2009 ) . The genome of the plant Arabidopsis contains six AMT paralogs , four of which ( AMT1;1 , 1;2 , 1;3 , and 1;5 ) are partially redundant and together essential for ammonium acquisition ( Yuan et al . , 2007 ) . In plants , ammonium acquisition is highly regulated and is subject to feedback inhibition , potentially as a means of preventing accumulation of ammonium to toxic levels ( Wang et al . , 1993; Kronzucker et al . , 2001 ) . Recent results show that ammonium-triggered phosphorylation of a critical threonine in the cytosolic C-terminus of AMT1;1 leads to inhibition of ammonium transport by allosteric regulation in the trimeric transporter complex ( Loqué et al . , 2007; Lanquar et al . , 2009 ) . The close homolog AMT1;3 , also functions as an electrogenic high-affinity ammonium transporter involved in nitrogen uptake in Arabidopsis roots ( Gazzarrini et al . , 1999 ) . AMT1;3 functions in homo- and hetero-trimeric complexes with the coexpressed AMT1;1 and is allosterically regulated ( Yuan et al . , 2013 ) . AMT1;3 not only functions as a transporter , but in addition controls root architecture by a process similar to the one found for the yeast MEP2 transceptor ( dual function transporter and receptor ) , a protein that mediates ammonium transport and regulates pseudohyphal growth ( Boeckstaens et al . , 2007; Lima et al . , 2010 ) . The term transceptor had been introduced by Thevelein to describe the dual function of the general amino acid permease Gap1p or the phosphate transporter Pho84p in transport and signaling ( Kriel et al . , 2011 ) . Most of the progress in the ammonium transport field has been made with the help of yeast and Xenopus laevis oocytes as heterologous expression systems . By contrast , in planta studies mostly rely on the use of either stable isotopes ( 15N ) , or radiotracer analogs ( 14C-methylammonium ) . Monitoring ammonium-induced depolarization of membrane potential has successfully been deployed in plants ( Wang et al . , 1994 ) , yet it can not differentiate between transport mediated by individual AMTs and non-selective cation channels . Therefore , it has not been possible to identify the networks that regulate the substrate-triggered feedback inhibition of ammonium uptake in plant roots . Thus there is a need for new tools that enable us to monitor transport activity in intact root tissues . Here we created an activity sensor by inserting cpGFP into a cytosolic loop of the Arabidopsis ammonium transporter AMT1;3 . Thorough characterization demonstrates that the chimera reports processes that occur in the transport cycle , although at present it is not possible to differentiate substrate binding or steps in the transport cycle . We project that this approach is transferable to other proteins , a notion supported by the successful creation of additional sensors for the Arabidopsis paralog AMT1;2 as well as the distantly related yeast ammonium transceptor MEP2 .
To engineer a transporter that reports substrate-dependent changes in conformation , we inserted fluorescent proteins ( FP ) into Arabidopsis thaliana AMT1;3 ( Figure 1 ) . Monomeric teal fluorescent protein ( mTFP ) , Venus or a modified circularly permutated GFP ( mcpGFP ) , which had successfully been employed to create highly sensitive , single-fluorophore sensors for calcium and maltose ( Tian et al . , 2009; Marvin et al . , 2011 ) , were inserted into intracellular loops of AMT1;3 . The functionality of the transporter fusions was tested by complementation of ammonium uptake deficiency in a yeast mutant lacking endogenous ammonium transporters ( Marini et al . , 1997 ) . AMT1;3 activity was extremely sensitive to manipulation of loops 7–8 and 9–10 or the cytosolic C-terminus ( Figure 1B ) . However , modification of loop 5–6 ( L5–6 , position K233 ) by insertion of either two amino acid residues ( encoded as part of the restriction site XbaI ) or mcpGFP was tolerated ( Figure 1B ) . In contrast , mTFP or Venus insertions inhibited transport activity . Interestingly , L5–6 is located between the two pseudo-symmetric halves of the protein and connects two transmembrane helices ( TMH-V and -VI ) that contain residues postulated to be directly involved in recruitment , gating and substrate translocation ( Figure 1C; Andrade et al . , 2005 ) . Moreover , TMH-V has been proposed to oscillate during substrate transport ( Andrade et al . , 2005; Inwood et al . , 2009 ) . L5–6 thus may be a prime location for detecting structural rearrangements through readout of a conformation-sensitive FP . 10 . 7554/eLife . 00800 . 003Figure 1 . Design of fusion constructs . ( A ) Topological representation of AMT1;3 by HMMTOP ( Tusnady and Simon , 2001 ) . Eleven TMH are organized in a pseudo-symmetric structure ( TMH I-V and TMH VI-X ) with an extra terminal TMH-XI that directs the C-terminus to the cytosol . The position of the residues preceding the insertion points of FPs in L5–6 , 7–8 , 9–10 and in the C-tail are indicated . Residues D202 , G460 and T464 , important for the activity of the transporter , are also shown . ( B ) The functionality of the transporters was measured as growth of the yeast Δmep1 , 2 , 3 mutant transformed with AMT1;3-FP fusions and grown on solid media containing 2 mM NH4Cl or 1 mM arginine ( growth control ) as the sole nitrogen source for three days . Numbers indicate the position in AMT1;3 preceding the insertion site . Vector: empty vector served as the negative control . ( C ) Three-dimensional model of one AMT-mcpGFP chimeric protein based on the crystal structures of Af-AMT1 ( 2B2H ) and cpGFP ( 3evp ) . One monomer is shown in cartoon and the rest of the trimeric complex is represented as a shaded surface in the background . mcpGFP ( green ) was inserted in position 233 of L5–6 of AMT1;3 , connecting TMH-V and -VI ( red and blue , respectively ) . DOI: http://dx . doi . org/10 . 7554/eLife . 00800 . 003 In initial tests , addition of ammonium to yeast cells expressing the AMT1;3-mcpGFP fusion did not lead to detectable changes in fluorescence intensity ( FI; data not shown ) . However , it is known that the cpGFP-based sensors for calcium and maltose are exquisitely sensitive to modifications in the linker sequences connecting to the respective binding proteins ( Wang et al . , 2008; Akerboom et al . , 2009; Marvin et al . , 2011 ) . To explore whether linker modification would affect the ability of the chimera to detect structural rearrangements of AMT1;3 during transport , we created 24 variants by altering the linkers connecting AMT1;3 and mcpGFP , exploiting information generated during optimization of calcium and maltose sensors ( Wang et al . , 2008; Akerboom et al . , 2009; Marvin et al . , 2011 ) . All 24 variants maintained transport activity ( Figure 2A ) ; importantly some variants showed a FI response to addition of 1 mM NH4Cl ( Figure 2B ) . The variant showing the strongest response ( 40% FI change ) was named AmTrac ( Ammonium Transporter coupled to mcpGFP ) . In AmTrac , mcpGFP had been inserted into position 233 of AMT1;3 , centrally located in the cytosolic loop L5–6 and flanked via linker sequences containing Leu-Glu and Phe-Asn at the N- and C-terminus , respectively ( Figure 2C , D ) . To explore the tolerance of L5–6 of AMT1;3 to mcpGFP insertion , we varied the insertion position in the loop ( Figure 3A ) . Both transport activity and fluorescence response were highest in variants with mcpGFP inserted in central positions of L5–6 , with a clear correlation between activity and FI change ( Figure 3B ) . Loop length also proved to be critical . In variants with incremental deletions in the middle of L5–6 around the insertion point of mcpGFP , deletion of four or more residues abolished transport activity and the fluorescence response ( Figure 3C ) . Importantly , all variants that responded to ammonium with a FI change were functional transporters , indicating that AmTrac reports transport activity . Fluorescent variants that were able to mediate transport but showed no FI change ( such as AMT1;3-mcpGFP-233 ) were used as control sensors to exclude potential effects of other parameters such as intracellular ammonium , proton accumulation or depolarization on the FI change . 10 . 7554/eLife . 00800 . 004Figure 2 . Development of AmTrac . ( A ) Growth of the yeast Δmep1 , 2 , 3 mutant transformed with fusion variants on solid media containing 2 mM NH4Cl or 1 mM arginine ( growth control ) as the sole nitrogen source for 3 days . Composition of linkers connecting AMT1;3 and mcpGFP are indicated . Linkers at the N- and C-termini of mcpGFP are indicated in letter code and separated by a slash . In the cases of variants 1–6 , no linkers were inserted between the C-terminal sequence of mcpGFP and the second part of AMT1;3 . ( B ) Screen of 24 linker variants for fluorescence intensity before addition of ammonium and fluorescence intensity change after addition of 1 mM NH4Cl ( mean ± SE; n = 3 ) . Variant 16 ( in red ) , carrying LE/FN as linkers , named AmTrac , showed the highest change in fluorescence intensity . ( C ) Schematic representation of AmTrac . Linkers between AMT1;3 and mcpGFP are indicated in yellow . ( D ) Protein sequence of AmTrac . Residues in ‘yellow’ constitute synthetic linker segments . Residues in ‘green’ correspond to the mcpGFP moiety . Numbers indicate amino acid position in AMT1;3 . DOI: http://dx . doi . org/10 . 7554/eLife . 00800 . 00410 . 7554/eLife . 00800 . 005Figure 3 . Effect of L5 on transport and fluorescence response . ( A ) Amino acid sequence of the L5–6 region of AMT1;3 . Residues in ‘blue’ correspond to TMH 5 ( left ) and TMH 6 ( right ) . ( B ) Influence of the insertion position of mcpGFP into L5 on transport activity and fluorescence response . Left panels show the growth assay of the yeast Δmep1 , 2 , 3 strain transformed with insertion variants on solid media containing 2 mM NH4Cl or 1 mM arginine ( growth control ) as the sole nitrogen source for 3 days . Numbers indicate the insertion site within AMT1;3 ( residue preceding the point of insertion of mcpGFP ) . Right graph shows the fluorescence response of the variants to addition of the indicated concentrations of NH4Cl . Data were normalized to the water-treated control ( 0; mean ± SE; n = 2 ) . ( C ) Growth and response of AmTrac variants with deletions in L5–6 . Growth was analyzed as described in Figure 2A . Numbers in the left column indicate the position of the insertion in AMT1;3; two numbers indicate residues preceding and following the mcpGFP insertion . Right column named “Response” indicates whether the corresponding variant responded to addition of 1 mM NH4Cl with a fluorescence change . The original AmTrac corresponds to insertion after aa 233 . DOI: http://dx . doi . org/10 . 7554/eLife . 00800 . 005 Initial experiments showed that the FI response of AmTrac was proportional to the ammonium concentration added to the yeast cells ( Figure 4A ) . The mcpGFP insertion into AMT1;3 did not significantly affect the affinity constant for ammonium transport ( Km ∼ 50 µM ) when tested in Xenopus oocytes using two-electrode voltage clamp ( TEVC ) analysis ( Figure 4B ) . Notably , the affinity constant for the FI response of AmTrac to NH4Cl ( EC50 55 ± 7 µM ) is strikingly similar to that reported here in oocytes and to the reported transport kinetics of AMT1;3 as measured in roots of Arabidopsis plants expressing only AMT1;3 ( Km 57 µM; Figure 4C; Yuan et al . , 2007 ) . 10 . 7554/eLife . 00800 . 006Figure 4 . Characterization of AmTrac transport and responses . ( A ) Currents recorded in single oocytes injected with water , AMT1;3 , AmTrac , or AmTrac-LS , and perfused with NH4Cl at the indicated concentrations . ( B ) Kinetics of NH4+-induced currents of AMT1;3 , AmTrac , and AmTrac-LS . The Kms were 55 ± 15 μM , 51 ± 24 μM , and 57 ± 19 μM , respectively . The data were fitted to Michaelis–Menten kinetics . Oocytes were clamped at −120 mV ( independent data from three different oocytes recorded from three different frogs ) . ( C ) Titration of the fluorescence response of AmTrac in yeast ( blue , left y-axis ) and of ammonium uptake of AMT1;3 in plants ( red , right y-axis; Yuan et al . , 2007 ) . Data are normalized to water-treated controls ( 0 ) ( mean ± SE; n = 3 ) . ( D ) Response of a single yeast cell expressing AmTrac to pulses of NH4Cl at the indicated concentrations ( blue frames ) . ( E ) Substrate specificity . Yeast cells expressing the sensor were treated with the indicated salts at 1 mM concentration . Data are normalized to water-treated control ( 0 ) ( mean ± SE; n = 3 ) . Only data for the ammonium treatments were significantly different from control ( SNK test: *p<0 . 01 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 00800 . 006 The saturation kinetics observed for the FI response can be interpreted as the dose-dependent conversion of an increasing number of transporters into a new state , for example from inactive to active or unbound to bound . We would assume that this process is reversible , so that AmTrac returns to the inactive or unbound state when ammonium is removed . To test the reversibility of the sensor response directly , we recorded the FI of single cells trapped in a microfluidic device and exposed to ammonium pulses . AmTrac responses were detectable in single cells , were concentration-dependent , and were readily reversible , demonstrating that AmTrac can be used in vivo to quantify ammonium transport activity and to observe kinetics of the chimeric transporter in response to substrate availability ( Figure 4D ) . The unique selectivity of AMT1 for ammonium over potassium was retained for FI responses ( Figure 4E ) . Also other cations did not induce a change in FI . NH4NO3 and ( NH4 ) 2SO4 elicited the same FI change as NH4Cl , demonstrating that chloride does not affect the FI response . Furthermore , addition of equimolar concentrations of sodium , potassium , calcium , magnesium , manganese or zinc did not trigger a change in FI ( Figure 4E ) . Together these data strongly support the hypothesis that AmTrac measures ammonium concentrations and/or reports conformational state changes of the transporter . The FI of AmTrac was relatively low , limiting the dynamic range and the signal-to-noise ratio ( SNR ) . To create sensors with enhanced SNR , we randomly substituted residues in the two-amino acid linker directly preceding mcpGFP . We identified AmTrac versions that maintained transport activity and showed high FI and a large ammonium-induced FI response . Interestingly , the majority of the brightest variants carried a Ser residue instead of a Glu in the position immediately preceding the mcpGFP insertion ( Figure 5A ) . AmTrac-LS , a sensor with a Leu-Ser linker , was as responsive to ammonium as AmTrac ( ∼40% FI change ) , but was approximately fivefold brighter ( Figure 5B ) . Fluorescence excitation and emission spectra of intact yeast cells expressing the improved variant AmTrac-LS were recorded . The excitation spectra revealed two maxima , a minor peak of the protonated chromophore at λex ∼ 380 nm and a major peak of the deprotonated chromophore at λex ∼ 495 nm . For the emission spectra one maximum was detected at λem ∼ 513 nm ( Figure 5C ) . The spectra are similar to those reported for unmodified EGFP . No shifts in the fluorescence maxima were observed , while FI decreased by 30–40% in a concentration-dependent manner when ammonium was added . The linker modifications had no significant effect on the transport kinetics ( tested for AmTrac-LS; Figure 4B ) , nor on the EC50 of the fluorescence response ( shown for AmTrac-IS , Ile-Ser linker; Figure 5D ) . AmTrac-LS responses were monitored in single yeast cells as described above and also showed reversibility of the FI change when ammonium was removed ( Figure 5E ) . 10 . 7554/eLife . 00800 . 007Figure 5 . Development of improved variants of AmTrac . ( A ) Screen for improved sensor variants . Fluorescence intensity and fluorescence response of ∼350 random variants of the C-terminal linker of mcpGFP to addition of 1 mM NH4Cl . The brightest and most responsive variants ( in red ) were sequenced and composition of the linker ( LE in AmTrac ) is reported as letter code . Note that most variants carry a Ser residue in the last position of the linker . ( B ) Fluorescence intensity and fluorescence response after addition of 1 mM NH4Cl , normalized to values of AmTrac ( 100% ) of yeast expressing the best variants identified in ( A ) ( mean ± SE; n = 3 ) . ( C ) Steady-state fluorescence spectra of AmTrac-LS recorded at λexc 485 nm and λem 514 nm , respectively , with addition of NH4Cl at the indicated concentrations , or water ( as control ) . Fluorescence intensities were normalized to the major peak of the water control ( =1 ) . ( D ) Titration of the fluorescent response of AmTrac ( blue line ) and AmTrac-IS ( red line ) . Data are normalized to water-treated controls ( 0; mean ± SE: n = 3 ) . AmTrac kinetics shown here for comparison are the same as Figure 4C . ( E ) Single cell response of AmTrac-LS . Individual cells trapped in a microfluidic plate were perfused with 50 mM MES buffer pH 6 . 0 , or a pulse of 50 µM NH4Cl in buffer ( blue box ) . Data were normalized to the initial value ( mean ± SE; n =3 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 00800 . 007 To further corroborate the correlation between transport activity and ammonium-induced FI change , mutations ( D202N , G460D , T464D ) known to inactivate AMT1 transporters were introduced into AmTrac ( Loqué et al . , 2007; Neuhäuser et al . , 2007 ) . While these mutations did not affect plasma membrane localization of AmTrac ( Figure 6A ) , both transport activity and fluorescence response were abolished ( Figure 6B , C ) , indicating that transport activity is tightly correlated with the FI response of AmTrac . Alternatively , accumulation of intracellular ammonium , and not its transport , could affect mcpGFP fluorescence . Since the Δmep1 , 2 , 3 yeast strain lacks endogenous ammonium transporters , ammonium uptake in these cells depends on episomally expressed transporters . AmTrac mutants unable to transport ammonium also prevent intracellular ammonium accumulation . To discriminate between the hypotheses that AmTrac reports conformational states of the transporter or responds to accumulation of ammonium in the cytosol , AmTrac and AmTrac-D202N , -G460D , -T464D mutants were expressed in wild type yeast carrying endogenous ammonium transporters ( MEPs; Figure 6B ) . In the presence of an independent set of functional ammonium transporters , none of the transport-deficient AmTrac versions produced an FI response ( Figure 6D ) , providing strong evidence that cytosolic ammonium accumulation did not elicit the FI response . As a further test , generation of intracellular ammonium from arginine catabolism ( Marini et al . , 1997 ) failed to trigger an AmTrac response ( Figure 6E ) . Taken together , these results indicate that AmTrac responds to extracellular and not intracellular ammonium levels . However our data do not exclude the possibility that other mutations can uncouple of the two processes , that is , to generate a protein that responds with a FI change to addition of ammonium but is unable to mediate transport across the membrane . 10 . 7554/eLife . 00800 . 008Figure 6 . AmTrac mutant analysis . ( A ) Confocal z-section of yeast expressing AmTrac or its inactive variants carrying mutations D202N , G460D or T464D . Bar 10 µm . ( B ) Growth complementation of the Δmep1 , 2 , 3 or wt yeast expressing AmTrac or its inactive variants on solid media containing 2 mM NH4Cl or 1 mM arginine ( growth control ) as sole nitrogen source for 3 days . Endogenous MEPs in wt strain are not affected by expression of mutant variants . ( C ) Fluorescence response of Δmep1 , 2 , 3 or ( D ) wt yeast expressing AmTrac or the transport-inactive variants D202N , G460D or T464D . Data were normalized to water-treated controls ( 0 ) ( mean ± SE; n = 3 ) . Only yeast cells expressing AmTrac showed significantly different responses ( SNK test: p<0 . 01 ) . ( E ) Response of AmTrac to addition of arginine . Fluorescence response of Δmep1 , 2 , 3 yeast cells expressing AmTrac treated with the indicated concentrations of arginine or 1 mM NH4Cl . Data were normalized to water-treated controls ( 0 ) ( mean ± SE; n = 3 ) . Only the ammonium treatment was significantly different from control ( SNK test: *p<0 . 01 ) . Note that arginine addition to yeast cells has been shown to lead to increased cytosolic levels of ammonium ( Marini et al . , 1994 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 00800 . 008 The allosteric trans-activation mechanism used by AMTs for feedback inhibition at elevated ammonium levels ( Loqué et al . , 2007; Lanquar et al . , 2009 ) provided us with an opportunity to test whether restoration of transport activity in non-functional AmTrac mutants reconstitutes the FI response . Previous work had shown that the C-terminus of AMT1 acts as a trans-activation domain in the trimeric AMT1 complex ( Loqué et al . , 2007 ) . Mutations affecting the cytosolic C-terminus block AMT1 activity , and activity can be restored by suppressor mutations in the cytosolic loops or mutations in the pore region ( Loqué et al . , 2007; Neuhäuser et al . , 2007 ) . A saturating multicopy suppressor screen with the inactive mutant AmTrac-T464D ( analogous to T460D in AMT1;1 ) ( Loqué et al . , 2007 ) identified eight gain-of-function mutations ( Figure 7A and Table 1 ) , seven in the pore region and one pseudo-reversion , D464V ( Figure 7B–D ) . The extent to which the suppressors were able to restore transport activity , as measured by growth , was highly correlated with the FI response ( R2=0 . 72; Figure 7E , F ) , further supporting the tight link between transport activity and FI response . 10 . 7554/eLife . 00800 . 009Figure 7 . Suppressor mutants restore transport and fluorescence response . ( A ) Growth of the yeast Δmep1 , 2 , 3 strain transformed with suppressor mutants and grown on solid media containing the indicated concentrations of NH4Cl , ( NH4 ) 2SO4 ( as anion control ) or 1 mM arginine ( growth control ) as the sole nitrogen source for 3 days . Note that yeast expressing AmTrac-T464D-A141E grew poorly at high concentrations of ammonium , suggesting high capacity transport activity leading to ammonium toxicity . ( B ) Lateral view and ( C ) cytoplasmic side view of AfAmt1 according to the crystal structure ( Andrade et al . , 2005 ) . The corresponding residues in AfAmt1 that suppress the T464D mutation in AmTrac are indicated by spheres . TMH-V is shown as red helix , TMH-VI in blue . The connecting L5–6 is labeled . Note that residues corresponding to cis-suppressing mutations reside in the internal pore region . ( D ) Sequence alignment of AMT1;3 from Arabidopsis ( At-Amt1;3 ) and Af-Amt1 from A . fulgidus . The residues belonging to TMH domains of the two pseudo-symmetric halves of AfAMT1 are shown in red and blue . C-terminal TMH-XI of AfAMT1 is shown in grey , with white font . Predicted TMHs of AMT1;3 are highlighted in grey , with black font . The corresponding residues identified in the suppressor screen of AmTrac-T464D are indicated in both sequences ( yellow ) . ( E ) Correlation between transport efficiency ( growth in 2 mM NH4Cl ) and fluorescence change after addition of 1 mM NH4Cl of the suppressor mutants . Data are normalized to values obtained for AmTrac ( =100 ) ( mean ± SD; n = 3 ) . ( F ) Fluorescence response of suppressors to addition of the indicated concentrations of NH4Cl . Data were normalized to water-treated controls ( 0 ) ( mean ± SE; n = 3 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 00800 . 00910 . 7554/eLife . 00800 . 010Table 1 . List of suppressor mutations , their location and frequencyDOI: http://dx . doi . org/10 . 7554/eLife . 00800 . 010MutationLocation#%Q61ETM 135Q61HTM 159F64STM 135I140STM 312A141ETM 32646A142PTM 31323I150TTM 324D464VC-tail35Total56100 Interestingly , one of the suppressor mutants , AmTrac-T464D-A141E , mediated hypersensitivity to elevated ammonium concentrations ( Figure 7A ) . Hypersensitivity , as shown in the case of the AtAMT1;1-T460D suppressor mutant Q57H ( Loqué et al . , 2009 ) can be caused by increased ammonium uptake capacity . TEVC analysis demonstrates that AmTrac-T464D-A141E functions as a low affinity/ high capacity ammonium transporter with a Km ∼ 3 . 5 mM ( Figure 8A ) . The fluorescence response of AmTrac-T464D-A141E ( AmTrac-100µ ) also increased ( EC50 = 101 ± 10 µM; Figure 8B ) . Once again , affinity and activity of the transporter correlated with the fluorescence change , suggesting a tight link between transport and fluorescence response . 10 . 7554/eLife . 00800 . 011Figure 8 . Characterization of a high capacity sensor variant . ( A ) Kinetics of NH4+-induced currents of AmTrac-100 . The Km was 3 . 4 ± 0 . 56 mM . The data were fitted to Michaelis–Menten kinetics . Oocytes were clamped at −120 mV ( independent data from three different oocytes recorded from three different frogs ) . ( B ) Titration of the fluorescent response of AmTrac ( blue line ) and AmTrac-100µ ( red line ) . Data are normalized to water-treated controls ( 0; mean ± SE: n = 3 ) . AmTrac kinetics shown here for comparison are the same as Figure 4C and Figure 5D . DOI: http://dx . doi . org/10 . 7554/eLife . 00800 . 011 To test whether the strategy of creating activity state sensors by mcpFP insertion is generalizable , we inserted mcpGFP into the central loop of other ammonium transporters; that is , the paralog AMT1;2 from Arabidopsis and the transceptor MEP2 of S . cerevisiae ( Marini et al . , 1997 ) . Phylogenetic analyses place AMT1;3 and AMT1;2 into the same subfamily . Yeast MEPs belong to a distantly related clade; and MEP2 and AMT1;3 share only 22% identity at the protein level . Notwithstanding , insertion of mcpGFP in the center of L5–6 of both AMT1;2 and MEP2 ( after amino acids R242 and R217 , respectively ) with the same linkers as in AmTrac-LS , generated two functional sensors ( named AmTrac1;2 and MepTrac ) that maintained transport activity ( Figure 9A ) and responded to ammonium addition with a fluorescence change ( Figure 9B ) . The original MepTrac showed only a low response when tested at pH 6 . In contrast to AMTs , MEP2 has a pH optimum at low pH . Histidine-194 in the pore of MEP2 has been shown to be responsible for the lower pH optimum ( Boeckstaens et al . , 2008 ) . A variant of MepTrac carrying the H194E mutation showed a significantly increased FI response , in agreement with the broader pH optimum conferred by this mutation ( Boeckstaens et al . , 2008; Figure 9B ) . Similar to AmTrac , the response of AmTrac1;2 , MepTrac and MepTrac-H194E to ammonium addition were dose-dependent . It is noteworthy that the AmTrac1;2 and MepTrac sensors were generated in a single step by using the linker and position information acquired during AmTrac development . We hypothesize that based on these results it will be possible to convert other transporters with similar twofold pseudo-symmetry into activity state sensors as well ( Forrest , 2013 ) . 10 . 7554/eLife . 00800 . 012Figure 9 . AmTrac1;2 and MepTrac response . ( A ) Growth complementation of Δmep1 , 2 , 3 yeast expressing MEP2 , MepTrac , their H194E mutants or AmTrac1;2 on solid media containing 2 mM NH4Cl or 1 mM arginine ( growth control ) as sole nitrogen source for 3 days . ( B ) Fluorescence response of Δmep1 , 2 , 3 yeast expressing AmTrac1;2 , MepTrac or MepTrac-H194E to NH4Cl at the indicated concentrations at pH 4 or 6 . AmTrac1;2 was tested at pH 6 . Data are normalized to water-treated control ( 0 ) ( mean ± SE; n = 3; SNK test: *p<0 . 01 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 00800 . 012
To create sensors that report substrate binding , a circularly-permutated form of GFP ( Tian et al . , 2009 ) was inserted into the center of the cytosolic loop in AMT1;3 ( and AMT1;2 and MEP2 ) at the pseudo-symmetry axis between the two antiparallel repeats of the transporter . The characteristics of the AmTrac fluorescence response , namely ( i ) the indistinguishable transport and fluorescence response kinetics , ( ii ) the reversibility , ( iii ) the tight link between transport activity for the FI response , ( iv ) the correlated shifts in affinity for ammonium transport and FI response , and ( v ) the restoration of the FI response by reconstitution of transport activity in suppressor mutants , strongly indicate that AmTrac sensors measure processes that occur during the transport cycle , or are intimately and quantitatively connected to transport activity . Since binding of the substrate to the transporter is part of the transport cycle , our data do not exclude the possibility that the sensors report extracellular binding of ammonium and therefore measure extracellular ammonium levels rather than activity . AMTs have been suggested to function as both transporters and receptors for ammonium ( Lima et al . , 2010 ) . Other transporters with a dual function as transporters and signaling proteins include UhpT from E . coli ( Schwöppe et al . , 2003 ) , Pho84p and GAP1p from Saccharomyces ( Thevelein and Voordeckers , 2009 ) and Chl1 from Arabidopsis ( Ho et al . , 2009 ) . The mechanism by which signaling is transmitted is unclear , specifically how substrate binding or conformational rearrangements during the transport cycle are necessary for triggering the receptor activity . Recent work by the Thevelein group has shed some light on the relative role of binding vs transport in this process . Some transporter-like receptors , such as the glucose transporter-like SNF3p and RGT1p from S . cerevisiae , appear to be unable to transport , suggesting that binding is sufficient for triggering a conformational rearrangement that is detected by intracellular signaling proteins ( Thevelein and Voordeckers , 2009 ) . The yeast Pho84 phosphate transceptor functions both in phosphate transport and mediates activation of the protein kinase A pathway . The finding that phosphate analogs that do not appear to be transported retain signaling activity is consistent with the hypothesis that a complete transport cycle is not required for signaling . However , further analyses suggested that ligand binding is insufficient for signaling , therefore the authors concluded the receptor activity is triggered by specific conformational rearrangements that occur during a specific phase of the transport cycle ( Popova et al . , 2010 ) . Mutations in the presumed proton-binding site severely reduced transport but not signaling , demonstrating that transport and signaling activities are separable ( Samyn et al . , 2012 ) . Combined analysis of the action of amino acid analogs and mutagenesis of the putative binding site of the amino acid transceptor GAP1 on transport and signaling have indicated that , similar as in PHO84 , signaling requires a ligand-induced conformational change that corresponds to a step in the transport cycle but does not appear to require a the full cycle ( Van Zeebroeck et al . , 2009 ) . It is therefore conceivable that the fluorescent change observed in the fluorescent ammonium sensors presented here is due to either ( i ) ammonium binding , ( ii ) a specific set of steps in the transport cycle , or ( iii ) the full transport cycle of the transporter . Crystal structures of AMTs from Escherichia coli ( Khademi et al . , 2004; Zheng et al . , 2004 ) and Archaeoglobus fulgidus ( Andrade et al . , 2005 ) did not reveal obvious conformational differences when generated in the presence or absence of ammonium; observations that led to the proposition that AMTs are rigid gas channels ( Khademi et al . , 2004 ) . Electrophysiological studies challenged the hypothesis that AMTs transport the uncharged form ammonia by demonstrating that plant AMTs are electrogenic and show minimal or no ammonia conductivity ( Mayer et al . , 2006 ) . The transport of the charged form of ammonium is further supported by the finding that mutations in the pore release the transporter from a gating control , leading to massively increased transport capacity and an up to 100-fold increase in the Km ( Figure 8A; Loqué et al . , 2009 ) . Moreover , we provided evidence for the existence of open and closed states by demonstrating an allosteric feedback inhibition of transport activity by phosphorylation of residues in the trans-activating cytosolic C-terminus ( Loqué et al . , 2007; Lanquar et al . , 2009 ) . The fluorescence response of AmTrac variants and MepTrac supports the hypothesis that AMT1;3 , AMT1;2 and MEP2 undergo conformational changes during transport . Our results are consistent with the hypothesis that TMH-V and -VI of AMT move relative to each other as a result of ammonium binding or during ammonium translocation , thus affecting the conformation of L5–6 and thereby gating the inserted mcpGFP . cpGFP has successfully been deployed to monitor conformational changes in a variety of proteins , for example , calmodulin ( Baird et al . , 1999; Nagai et al . , 2001 , 2004; Nakai et al . , 2001 ) , metal ion binding domains ( Baird et al . , 1999; Mizuno et al . , 2007 ) , a periplasmic sugar binding protein ( Marvin et al . , 2011 ) , PKG isoforms for measuring cGMP ( Nausch et al . , 2008 ) , and a voltage sensing domain of a protein phosphatase ( Ci-VSP ) ( Gautam et al . , 2009 ) . While none of the sensors measures activity , they report ensemble changes in conformational states of the respective proteins . Here we demonstrate that insertion of cpGFP can be more widely used to monitor activity states of proteins , specifically ammonium transporters . Interestingly , the pseudo-symmetry of AMTs with an inverted repeat of a five TMH containing domain is comparable to that found in the LeuT transporter ( Krishnamurthy and Gouaux , 2012 ) . LeuT carries a substrate-binding site at the interface of the two repeats and undergoes a transport cycle for which multiple states have been identified ( Forrest and Rudnick , 2009; Rudnick , 2011 ) . Similarly , the transport pore of the AMTs is located between the two pseudo-symmetric halves , with TMH-V and -VI ( connected by L5–6 ) carrying key residues for ammonium translocation . Our data are consistent with the occurrence of structural rearrangements during transport; crystal structures for different states will be required to determine the conformational state during transport . In summary , we developed an in vivo reporter of activity states for a set of ammonium transporters . The combination of two ammonium transport activity sensors from different organisms will be useful for dissecting endogenous control mechanisms , because heterologous proteins are not expected to be subject to the same regulatory control and can thus serve as references . As a next step , we will deploy AmTrac and MepTrac in both Arabidopsis roots and in yeast to study their regulation in vivo and to identify regulators . The measurement of flux through a transporter by reporting state changes , as shown here , might be applicable to other transporters or enzymes for monitoring in vivo fluxes , for example , in the context of neurotransmission . Such in vivo flux analyses will aid in understanding the mode of action of proteins and may serve as tools for drug discovery .
All transporter and sensor constructs were inserted in the yeast expression vector pDRf1-GW , containing the f1 replication origin , GATEWAY cassette , PMA1 promoter fragment , ADH terminator , and the URA3 cassette for selection in yeast ( Loqué et al . , 2007 ) . The 1494 bp ORF of AMT1;3 from Arabidopsis ( At3g24300 ) was used as base for AmTrac construction . The XbaI restriction site ( tctaga ) was inserted in different positions of AMT1;3 ( after amino acids 233 , 312 , 364 and 448 ) using Kunkel mutagenesis ( Kunkel et al . , 1991 ) . mcpGFP was generated by amplifying the domains of EGFP corresponding to amino acids 150–239 and 1–144 with the primer pairs EGFP-150-for/EGFP-239-rev and EGFP-1-for/EGFP-144-rev , respectively ( see Supplementary file 1 ) . The two amplified fragments were gel-purified by a commercial kit ( Machery-Nagel , Düren , Germany ) , digested by AgeI ( New England Biolabs , Ipswich , MA ) and ligated by T4 DNA ligase ( New England Biolabs ) . The resulting cpGFP , where the domains 150–239 and 1–144 were connected by the linker coding GGTGGS , was cloned into a pGEM-Teasy ( Promega , Madison , WI ) . The additional internal mutations M154K , V164A , S176G , D181Y , T204V , A207K , V190I ( positions referring to eGFP sequence ) , which had been shown to improve stability of cpGFP ( Akerboom et al . , 2009; Tian et al . , 2009 ) , were introduced by Kunkel mutagenesis ( Kunkel et al . , 1991 ) to generate mcpGFP . Finally , mcpGFP was amplified with the primers cpGFP-for and cpGFP-rev ( Supplementary file 1 ) . Similarly , mTFP and Venus were amplified with primers mTFP-for and mTFP-rev , and Venus-for Venus-rev , respectively ( Supplementary file 1 ) . The purified mcpGFP- , mTFP- and Venus-encoding fragments were digested by XbaI ( New England Biolabs ) and ligated into XbaI digested pDRf1-GW containing AMT1;3 versions that harbor the sequence corresponding to the XbaI restriction site in different cytosolic loops , to generate the fusion constructs AMT1;3-mcpGFP , AMT1;3-TFP and AMT1;3-Venus in positions 233 , 312 , 364 and 448 ( Figure 1B ) . To vary the linker regions between mcpGFP and AMT1;3 in position 233 ( Figure 2A ) , we took advantage of homologous recombination in yeast between two DNA fragments sharing sequence homology . We co-transformed yeast with the pDR-AMT1;3 opened at position 233 by XbaI digestion , and the mcpGFP fragments amplified by PCR with the primer pair cpGFP-hom-for/rev containing the variable sequence encoding the linkers ( Supplementary file 1 ) . The amplification products contained mcpGFP flanked by the variable linker sequences and about 30 bp homologous to the region around the insertion point 233 of AMT1;3 . The transformed yeast contained the pDR-AMT-mcpGFP vectors resulting from insertion of the ∼800 bp mcpGFPs with linkers into the vector backbone , as confirmed by DNA sequencing . Similarly , we used homologous recombination to generate random variants of the linker preceding mcpGFP ( Figure 5A ) . For this case , mcpGFP was amplified with the primer pairs cGFP-hom-for-deg and cpGFP-hom-rev-FN ( Supplementary file 1 ) . Homologous recombination was also used to insert mcpGFP in different positions along L5–6 ( Figure 3B ) and to generate deletions in the loop ( Figure 3C ) . In this case , the 30-bp-long regions of the primers that overlapped the AMT1;3 sequence flanked the different insertion points ( 228–236 ) and contained the appropriate deletions . Point mutations for inactivation of AMT1;3 and AmTrac were generated by Kunkel mutagenesis ( D202N , G460D , T464D; Figure 6 ) . MepTrac was generated by overlapping PCR . The 1500 bp S . cerevisiae MEP2 cDNA was amplified in two separate halves ( corresponding to amino acids 1–217 and 218–499 ) with the primers attB1-MEP2-for and MEP2-217-rev , and MEP2-218-for and attB2-MEP2-rev , respectively ( see Supplementary file 1 ) . Primers contained the MEP2-specific sequences plus the Gateway attB sites for subsequent BP cloning . A third fragment containing the mcpGFP with the linkers coding for the amino acids LS and FN ( preceding and following the mcpGFP , respectively ) was generated by amplifying the mcpGFP from the pDR-Amtrac-LS template with the primers cpGFP-LS-MEP2-for and cpGFP-LS-MEP2-rev ( Supplementary file 1 ) . The primers contained a region specific to mcpGFP , a sequence coding for the linkers and about 30 bp homologous to the region around the insertion point 217 of MEP2 . The three fragments were gel-purified by a commercial kit ( Machery-Nagel ) and 1 µl of each fragment was added to a PCR reaction tube ( Phusion Taq; Finnzymes Thermo Fisher , Waltham , MA ) containing 200 µM dNTPs and amplified for 10 cycles without addition of any primer . 1 µl of this reaction was further amplified for additional 30 cycles in a PCR reaction tube ( Phusion Taq; Finnzymes ) containing 200 µM dNTPs and the primers attB1-MEP2-for and attB2-MEP2-rev . The amplified fragment was gel-purified by a commercial kit ( Machery-Nagel ) and cloned into a pDON221 vector by Gateway BP reaction ( Invitrogen Life Technology , Paisley , United Kingdom ) , following manufacturer’s instructions . The resulting pENTRY-MepTrac construct was introduced into TOP10 competent cells ( Invitrogen ) . The yeast vector harboring MepTrac was then created by Gateway LR reaction between pENTRY-MepTrac and pDRf1-GW , following manufacturer’s instructions . Sequence of Meptrac in the resulting pDR-MepTrac was verified by sequencing . MepTrac-H194E was generated by single-point mutagenesis of MepTrac by a commercial kit ( QuickChange; Stratagene Agilent , Santa Clara , CA ) . AmTrac1;2 was generated by overlapping PCR by the same strategy used for MepTrac . The cassette of mcpGFP with LS/FN linkers was inserted into the A . thaliana AMT1;2 cDNA ( 1545 bp , At1g64780 ) after the codon for the amino acid R242 , in the middle of predicted loop5–6 . Primers are listed in Supplementary file 1 . For functional assays in Xenopus oocytes , the cDNAs of AMT1;3 , AmTrac , AmTrac-LS , and AmTrac-100µ were cloned into the oocyte expression vector pOO2-GW ( Loqué et al . , 2009 ) . Yeast strains used in this study were 31019b [mep1Δ mep2Δ::LEU2 mep3Δ::KanMX2 ura3] , a strain in which all three endogenous MEP ammonium transporter genes had been deleted , and its parental strain 23344c [ura3] , considered here as wt ( Marini et al . , 1997 ) . Yeast was transformed using the lithium acetate method ( Gietz et al . , 1992 ) and selected on solid YNB ( minimal yeast medium without amino acids and without ammonium; Difco BD , Franklin Lakes , NJ ) supplemented with 3% glucose and 1 mM arginine . Single colonies were inoculated in 5 ml liquid YNB supplemented with 3% glucose and 0 . 1% proline under agitation ( 230 rpm ) at 30°C until OD600nm=0 . 8 . The liquid cultures were diluted 10−1 , 10−2 , 10−3 , 10−4 , 10−5 and 10−6 in water and 5 µl of each dilution was spotted on solid YNB medium buffered with 50 mM MES/Tris , pH 5 . 2 and supplemented with 3% glucose and either NH4Cl , ( NH4 ) 2SO4 or 1 mM arginine as the sole nitrogen source . After 3 days of incubation at 30°C , cell growth was documented by flatbed scanning the plate at 300 dpi in grayscale mode . For quantification of yeast growth ( Figure 7E ) , the pixel intensity of each spot ( here 10−2 dilution ) were quantified using ImageJ software , and normalized to AmTrac spot intensity ( 100% ) . Quantification of spot intensity of 10−1 and 10−3 dilutions gave similar results ( not shown ) . Fluorimetric analyses are described in detail at Bio-protocol ( Ast et al . , 2015 ) . Yeast cultures were washed twice in 50 mM MES buffer , pH 6 . 0 , and resuspended to OD600nm=0 . 5 in MES buffer supplemented with 5% glycerol to delay cell sedimentation . For MepTrac and MepTrac-H194E measurements , cells were washed twice in purified water and resuspended in 25 mM sodium citrate buffer , pH 4 . 0 or 6 . 0 , supplemented with 5% glycerol . Fluorescence was measured by a fluorescence plate reader ( Safire; Tecan , Männedorf , Germany ) , in bottom reading mode using a 7 . 5 nm bandwidth for both excitation and emission ( Bermejo et al . , 2010 , 2011 ) . To measure fluorescence response to substrate addition , 50 µl of substrate ( dissolved in water as 500% stock solution ) were added to 200 µl of cells in a 96-well plate ( Greiner , Monroe , NC ) . Fluorescence was measured as emission at λem 513 nm using excitation at λexc 488 nm . Response data are presented as ( Fwater − Ftreatment ) /Ftreatment . Steady-state emission measurements were performed using a Fluoromax-P fluorescence spectrometer ( Horiba Jobin Yvon , Kyoto , Japan ) in 3 . 5-ml silica cuvettes ( Hellma Analytics , Müllheim , Germany ) . 2 ml liquid yeast cultures were supplemented with 0 . 5 ml NH4Cl ( 10 mM , 1 mM , 100 µM or 10 µM ) or water as control . The samples were allowed to incubate for 20 min . The excitation and emission spectra were recorded at λem 514 nm and λexc 485 nm , respectively , with a step size of 1 nm and five repeats were taken for averaging . Untransformed yeast cells served as blank . Confocal sections of yeast cells expressing the sensors ( Figure 6A ) were acquired on an inverted confocal laser scanning microscope ( SP5; Leica , Wetzlar , Germany ) . To record fluorescence intensities in single cells over time , yeast cells were trapped as a single cell layer in a microfluidic perfusion system ( Y04C plate; Onyx , Cellasic , Hayward , CA ) and perfused with either 50 mM MES buffer , pH 6 . 0 , or buffer supplemented with NH4Cl ( Bermejo et al . , 2010 , 2011 ) . Cells were imaged on a spinning disk confocal microscope ( Yokogawa CSU-X1 , Mitaka , Japan; Leica DMI6000 ) equipped with a motorized stage ( ASI ) . Fluorescence was excited by a solid state laser at 488 nm; emission was detected using a 525/50 nm filter set ( Semrock , Rochester , NY ) and an electron multiplying charge coupled device ( EMCCD ) camera ( Evolve; Photometrics , Tucson , AZ ) . Measurements were taken every 2 min , with 100 ms exposure time using Slidebook 5 . 0 image acquisition software ( Intelligent Imaging Innovations , Denver , CO ) . To account for lateral shift during imaging , the image stacks were post-registered using the StackReg plugin for ImageJ ( Thevenaz et al . , 1998 ) . Fluorescence pixel intensity was quantified using Fiji software; single cells were selected and analyzed with the help of the ROI manager tool . For in vitro transcription , pOO2-AMT1;3 and AmTracs were linearized with MluI . Capped cRNA was transcribed in vitro by SP6 RNA polymerase using mMESSAGE mMACHINE kits ( Ambion , Austin , TX ) . Xenopus laevis oocytes were obtained by surgery , manually dissected , and defolliculated with collagenase ( Sigma , St . Louis , MO ) . The oocytes were injected by Roboinjector ( Multi Channel Systems , Reutlingen , Germany; ( Pehl et al . , 2004; Leisgen et al . , 2007 ) ) with distilled water ( 50 nl as control ) , AMT1;3 cRNA ( 50 ng in 50 nl ) , AmTrac cRNA ( 50 ng in 50 nl ) , AmTrac-LS cRNA ( 50 ng in 50 nl ) , or AmTrac-100µ cRNA ( 50 ng in 50 nl ) . Cells were kept at 16°C in ND96 medium containing 96 mM NaCl , 2 mM KCl , 1 . 8 mM CaCl2 , 1 mM MgCl2 , and 5 mM HEPES , pH 7 . 4 , containing gentamycin ( 50 μg/μl ) . Measurements were made in a solution as previously described ( Loqué et al . , 2009 ) . Oocytes were voltage-clamped at −120 mV and measured by two-electrode voltage-clamp ( TEVC ) Roboocyte system ( Multi Channel Systems ) ( Pehl et al . , 2004; Leisgen et al . , 2007 ) . All experiments were performed in triplicate , unless specified otherwise . Reported values represent mean and standard deviation . The effects of treatments on the fluorescence response were compared using analysis of variance ( ANOVA ) : the factors ( sample , treatment ) were treated as fixed factors . Prior to analyses , homogeneity of the variances was tested by Cochran’s test . Whenever the ANOVA revealed significant differences among treatments , post hoc comparisons were performed using the SNK test . ANOVAs were performed using the GMAV5 software package ( University of Sydney , Australia ) . | Ammonium provides a vital source of nitrogen for bacteria , fungi and plants , and is produced by animals as a waste product of metabolism . High levels of ammonium can be toxic , so all organisms need to control their uptake or excretion of this substance . Ammonium transporters , which are highly conserved from bacteria to plants to humans , are essential for this process but , along with transporters in general , they are hard to study . Their activity can be examined in vitro by expressing them in heterologous systems—that is , in cells other than those in which they are naturally found . But in vivo studies must rely on indirect techniques such as monitoring radioactive isotopes or membrane potentials , and these cannot distinguish between the activity of ammonium transporters and uptake of ammonium through other routes . One approach that has been successful in other fields is the use of fluorescent proteins that can signal conformational changes—such as those that occur when a transporter is activated—by a shift in fluorescence . Green fluorescent protein ( GFP ) is a commonly used fluorescent indicator , and a particularly useful variant is ‘circularly permutated GFP’ . This is GFP in which parts of the amino acid sequence have been rearranged without fundamentally changing the overall structure or function of the protein . Circularly permutated GFP can be fused to another protein in such a way that a conformational change in the second protein triggers a change in fluorescence that can be detected by fluorescence spectroscopy or microscopy . Now , De Michele et al . have applied this approach to the study of both plant and yeast ammonium transporters . They constructed a library of fusion proteins made up of circularly permutated GFP and an ammonium transporter from the plant Arabidopsis thaliana—and found one version that functioned normally as a transporter but also produced a detectable change in fluorescence that correlated precisely with transporter activity . De Michele et al . then used the same method to produce fluorescent indicator fusion proteins of two more ammonium transporters—a second isoform from Arabidopsis and one from yeast . These fluorescent sensors should be a great boon to researchers studying the ammonium transport system . Moreover , this approach could in theory be applied to other transporter proteins that are currently difficult to study , and so could help to open up research into a variety of transport processes . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"plant",
"biology",
"cell",
"biology"
] | 2013 | Fluorescent sensors reporting the activity of ammonium transceptors in live cells |
Proper kinetochore-microtubule attachments , mediated by the NDC80 complex , are required for error-free chromosome segregation . Erroneous attachments are corrected by the tension dependence of kinetochore-microtubule interactions . Here , we present a method , based on fluorescence lifetime imaging microscopy and Förster resonance energy transfer , to quantitatively measure the fraction of NDC80 complexes bound to microtubules at individual kinetochores in living human cells . We found that NDC80 binding is modulated in a chromosome autonomous fashion over prometaphase and metaphase , and is predominantly regulated by centromere tension . We show that this tension dependency requires phosphorylation of the N-terminal tail of Hec1 , a component of the NDC80 complex , and the proper localization of Aurora B kinase , which modulates NDC80 binding . Our results lead to a mathematical model of the molecular basis of tension-dependent NDC80 binding to kinetochore microtubules in vivo .
Chromosome segregation errors lead to aneuploidy and micronuclei formation , which are closely associated with cancer , infertility , and birth defects ( Santaguida and Amon , 2015 ) . Accurate chromosome segregation is believed to result from a process that actively suppresses potential errors . The mechanism of error correction remains unclear , but extensive evidence suggests that it is based on the regulation of the attachment of microtubules to chromosome via the kinetochore , a protein complex assembled at centromeres ( Godek et al . , 2015 ) . Previous works suggested that error correction is largely due to the detachment of kinetochore microtubules ( kMTs ) being regulated by the tension across centromeres , which selectively destabilizes erroneous kMT attachments bearing low tension and stabilizes proper attachments under high tension ( Nicklas and Ward , 1994; Liu et al . , 2009; Akiyoshi et al . , 2010; Lampson and Cheeseman , 2011; Godek et al . , 2015 ) . However , the molecular mechanism of the tension-dependent regulation of kMT attachments is still poorly understood . The highly conserved NDC80 complex is the major coupler of the kinetochore to microtubules ( Cheeseman et al . , 2006; DeLuca et al . , 2006 ) . In human mitotic cells , ~240 NDC80 complexes are recruited at the outer layer of each kinetochore ( Suzuki et al . , 2015 ) and interact with ~20 kMTs by directly binding to them ( Cheeseman and Desai , 2008; Maiato et al . , 2004; Rieder , 1982 ) . In vitro experiments showed that the binding affinity of NDC80 for microtubules decreases upon the phosphorylation of the N-terminal tail of Ndc80/Hec1 protein by Aurora B kinase ( Cheeseman et al . , 2006; Zaytsev et al . , 2014 , 2015 ) , which may explain the contribution of Aurora B to error correction ( Tanaka et al . , 2002 ) . It is unclear how the biochemical activities of NDC80 and Aurora B result in tension-dependent kMT detachment . The lack of techniques to measure the binding of the NDC80 to kMTs in vivo has been a major obstacle to investigate this .
Inspired by previous work ( Posch et al . , 2010 ) , we sought to develop a Förster Resonance Energy Transfer ( FRET ) based approach to directly measure the association between the NDC80 complex and kinetochore microtubules ( kMTs ) in living cells . We engineered U2OS cells stably expressing Nuf2 , a subunit of the NDC80 complex , N-terminally labeled with a cyan fluorescent protein , mTurquoise2 ( Figure 1A ) . In this same cell line , we also inserted a tetracysteine ( TC ) motif at the C-terminus of β-tubulin ( TUBB ) using CRISPR-induced homologous recombination , which becomes fluorescent after binding to the membrane-permeable dye FlAsH ( Hoffmann et al . , 2005 ) ( Figure 1A and B ) . The small size ( six amino acids ) of the TC motif minimizes the negative effects of labeling the C-terminus of tubulin , allowing the engineered cells to successfully pass through mitosis ( Andresen et al . , 2004 ) . CRISPR-mediated endogenous tubulin tagging ensures low cell-to-cell variation and a high fraction of labeled β-tubulin , which was estimated to be 26 . 1 ± 5 . 4% ( SD ) ( Figure 1—figure supplement 1 and see Supplemental experiments in Materials and methods ) . We used time-correlated single photon counting ( TCSPC ) fluorescence lifetime imaging microscopy ( FLIM ) to quantitatively measure FRET between mTurquoise2 and TC-FlAsH in tissue culture cells ( Figure 1—figure supplement 2 ) . TCSPC FLIM-FRET provides fluorescence decay curves of the donor fluorophore at each pixel location . If a donor fluorophore has a single-exponential fluorescence decay curve when not engaged in FRET , then when it is engaged in FRET the fluorescence decay curve will also be single-exponential , but with a shorter lifetime . A pixel containing a mixture of such donor fluorophores engaged in FRET and not engaged in FRET displays a fluorescence decay curve that is a sum of two exponentials . Bayesian analysis of the fluorescence decay curves provides a bias-free measurement of the relative fraction of the two exponentials , and hence the fraction of donor fluorophores engaged in FRET ( Yoo and Needleman , 2016; Kaye et al . , 2017 ) . In contrast to intensity-based FRET methods , FLIM-FRET is capable of quantifying the fraction of donor fluorophores engaged in FRET when donors and acceptors are differentially distributed in cells , and it is less prone to errors arising from instrumental artefacts and photobleaching ( Berezin and Achilefu , 2010 ) . We first characterized the fluorescence decay of mTurquoise2-NDC80 in the absence of FRET by performing FLIM measurement on the engineered U2OS cells ( mTurquoise2-NDC80/β-tubulin-TC ) that were not exposed to FlAsH ( Figure 1B , top ) . We found that their fluorescence decays are well described as a single exponential with a lifetime of 3 . 75 ± 0 . 09 ns ( SD ) ( Figure 1C , left , and Figure 1—figure supplement 3A ) . As discussed above , this single exponential decay profile is expected when the donor fluorophores do not engage in FRET . We next measured the fluorescence decay of the mTurquoise2-NDC80 in the presence of FlAsH labeling of microtubules . In this case , a single exponential provided a poor fit to the data , exhibiting significant systematic deviations ( Figure 1C , right ) . The fluorescence decay in the presence of FlAsH labeling was well fit by a sum of two exponentials with lifetimes 3 . 71 ± 0 . 04 ns ( SE ) and 0 . 75 ± 0 . 12 ns ( SE ) ( Figure 1D ) . The long lifetime of the two-exponential fit was indistinguishable from the lifetime in the absence of FRET ( p=0 . 68 , two-sided Z-test ) , and thus corresponds to the non-FRET donor population . Therefore , the short-lifetime species is the FRET donor population . The relative amplitude of the short- and long-lifetime exponentials are 0 . 13 ± 0 . 01 ( SE ) and 0 . 87 ± 0 . 01 ( SE ) , respectively , thus 13 ± 1% ( SE ) of donor fluorophores are engaged in FRET . Having demonstrated our ability to measure FRET between mTurquoise2-NDC80 and FlAsH in tissue culture cells , we explored if the FRET is due to the binding of NDC80 to kMTs . We first engineered an alternative construct with mTurquoise2 conjugated to the distally located C-terminus of Nuf2 , far removed from kMTs . This alternative construct displayed only a single long-lifetime state in either the presence or absence of TC-FlAsH , arguing that FRET does not result from non-specific interactions ( Figure 1—figure supplement 3B , C and E ) . Incubating cells with nocodazole to depolymerize microtubules caused a reduction ( p<10−10 , two-sided Z-test ) of NDC80 FRET fraction from 13 ± 1% ( SE ) to 3 ± 1% ( SE ) ( Figure 1—figure supplement 3D ) . Thus , FRET strongly depends on the presence of microtubules . We next investigated if NDC80 that is close to kMTs , but not bound to them , can lead to appreciable FRET . Answering this requires knowing the Förster radius between mTurquoise2 and TC-FlAsH , which we determined to be 5 . 90 ± 0 . 10 nm ( SE ) through a combination of FLIM measurements and Monte Carlo simulations ( see Figure 1—figure supplement 4 and Supplemental experiments in Materials and methods ) . We next performed large-scale Monte Carlo simulations of mTurquoise2-NDC80 at various distances between the calponin homology ( CH ) domain of Hec1/Ndc80 protein ( the NDC80 complex’s microtubule binding domain adjacent to mTurquoise2 [Alushin et al . , 2010] ) and FlAsH-labeled microtubules and simulated the fluorescence decay curves , which revealed that NDC80 more than 8 nm away from the kMT do not contribute to the short-lifetime FRET state ( Figure 1—figure supplement 5A and B ) . Thus , FRET only results when the CH domain of Hec1 is very close to the surface of kMTs , consistent with the short-lifetime species being NDC80 complexes whose Hec1 CH domains are bound to kMTs , an interpretation further supported by biological perturbation experiments described below . Even though FRET results only from NDC80 bound to kMTs , the measured FRET fraction is not identical to the fraction of NDC80 bound to kMTs because not all tubulin heterodimers are labeled with TC-FlAsH . Using large scale Monte Carlo simulations of mTurquoise2-NDC80 bound to FlAsH-labeled microtubules , we generated fluorescence decay curves for various NDC80 binding fractions , and estimated the resulting NDC80 FRET fractions from a fit to a two-exponential decay ( see Figure 1—figure supplement 5C and Supplemental experiments in Materials and methods ) . We found that the NDC80 FRET fraction increases linearly with the NDC80 binding fraction with a slope of 0 . 42 ± 0 . 08 , indicating that 42% of attached mTurquoise2-NDC80 contribute to the short-lifetime FRET state ( Figure 1—figure supplement 5C ) . Thus , the measured FRET fraction of 13% in Figure 1D corresponds to 31% of NDC80 complexes being bound to kMTs . Using the FLIM-FRET measurements of NDC80-kMT binding , we first investigated how NDC80-kMT binding evolves over the course of mitosis . We found that the average NDC80-kMT binding gradually increases as mitosis progresses , with NDC80 FRET fraction rising from 7% in early prometaphase to 14% in late metaphase , and reaching about 18% in anaphase ( corresponding to NDC80 binding fractions of 17% in prometaphase; 33% in late metaphase; and 43% in anaphase ) ( Figure 2A ) . This temporal change in NDC80-kMT binding may underlie the previously observed decrease in kMT turnover throughout mitosis ( Kabeche and Compton , 2013; Zhai et al . , 1995 ) . The change in the average NDC80-kMT binding over the course of error correction in prometaphase could be due to a cell cycle-dependent coordinated regulation of NDC80 across kinetochores ( coordinated regulation ) , an independent modulation of NDC80 on different chromosomes ( chromosome-autonomous regulation ) , or a combination of both . To investigate the contribution of chromosome-autonomous regulation , we sought to determine if different populations of kinetochores in prometaphase exhibit different NDC80-kMT binding . We compared the extent of the NDC80-kMT binding of kinetochores centered at the metaphase plate to those located off-centered ( Figure 2B ) , and found that the centered kinetochores exhibit 2 . 0 ± 0 . 4 times higher NDC80 binding than the off-centered kinetochores . We next investigated how the NDC80-kMT binding of centered and off-centered kinetochores change with time in prometaphase . As mitosis progresses and the chromosomes align to the metaphase plate , the number of kinetochores in the center region increases while the number of kinetochores in the off-center region decreases ( Figure 2C ) . NDC80-kMT binding continuously increases over time for the kinetochores located in the center region , but remains constant with the FRET fraction of ~7% for the kinetochores in the off-center region ( Figure 2D ) . The observation of differences in NDC80 binding between different subpopulations of kinetochores strongly argues for the existence of chromosome-autonomous regulation , which might be modulated by tension , Aurora kinases A and B , pathways that control the conversion of lateral to end-on kMT attachments , or other factors ( Godek et al . , 2015; DeLuca et al . , 2018 ) . We speculate that the temporal increase in NDC80-kMT binding of centered kinetochores is due to the gradual decrease in the number of kinetochores with erroneous attachment that transiently lie on the metaphase plate ( Magidson et al . , 2011 ) . The constant NDC80-kMT binding of off-centered kinetochores argues for a lack of temporal regulation of this subpopulation . After demonstrating that different population of kinetochores exhibits different NDC80-kMT binding throughout prometaphase , we next investigated the factors contributing to chromosome-autonomous regulation of the interaction between NDC80 and kMTs . Aligned chromosomes in U2OS cells oscillate around the metaphase plate , with microtubules attached to the leading and trailing kinetochores primarily depolymerizing and polymerizing , respectively ( Tirnauer et al . , 2002; Armond et al . , 2015 ) ( Figure 3A ) . The distance between sister kinetochores ( referred to as K-K distance ) fluctuates during the oscillation ( Magidson et al . , 2011 ) , as the centromere deforms in response to the dynamic change in tension ( Figure 3A ) . Therefore , chromosome oscillation provides a window to study how NDC80 binding is related to kMT dynamics and centromere tension in a physiologically relevant condition . We first asked whether NDC80 binding is different on leading and trailing kinetochores . We acquired time-lapse movies of 17 metaphase cells , tracked their kinetochores , identified sister kinetochores by their relative motions ( Figure 3B ) , and quantified the NDC80 binding fraction in groups of kinetochores with similar velocities using FLIM-FRET analysis . We found that the NDC80 FRET fraction is higher at trailing kinetochores ( 12 . 8 ± 0 . 5% , SEM ) than leading kinetochores ( 11 . 4 ± 0 . 5% , SEM ) , regardless of their speeds ( Figure 3C ) , suggesting that NDC80 preferentially binds to polymerizing kMTs in vivo . The preferential binding is statistically significant ( p<0 . 02 , two-sided Z-test ) , yet small , presumably because leading and trailing kinetochores have a mixture of both polymerizing and depolymerizing MTs ( Armond et al . , 2015 ) . This differential binding of NDC80 provides an explanation for the higher detachment rate of depolymerizing microtubules from kinetochores in vitro ( Akiyoshi et al . , 2010 ) , and may give insight into the nature of kMT attachments ( Dumont et al . , 2012 ) . The detachment rate of kMTs from kinetochores was shown to be reduced when tension was increased using glass needles in classic micromanipulation experiments by Bruce Nicklas ( Nicklas and Koch , 1969 ) . Since the NDC80 complex is the predominant coupler of the kinetochore to microtubules ( Cheeseman et al . , 2006; DeLuca et al . , 2006 ) , we hypothesized that the tension-dependent detachment of kMTs results from tension-dependent NDC80-kMT binding . To test this possibility , we next investigated the correlation between NDC80 FRET fraction and centromere tension , inferred by K-K distance , during chromosome oscillations . We used FLIM-FRET analysis to measure the NDC80 binding in groups of sister kinetochores with similar K-K distances , and observed a highly significant positive correlation ( p<0 . 005 ) between NDC80 FRET fraction and K-K distance ( Figure 3D ) . We observed no significant correlation between K-K distance and kinetochore velocity ( p=0 . 75 ) , arguing that NDC80 binding is independently regulated by these two factors ( Figure 3—figure supplement 1 ) . In the absence of microtubules , the rest length of K-K distance in human cell is 0 . 73 ± 0 . 04 μm ( Tauchman et al . , 2015 ) , significantly shorter than the K-K distances during metaphase oscillations . Thus , in order to investigate a wider range of K-K distance , we treated cells with taxol , a microtubule-stabilizing drug , which greatly reduced K-K distances ( 0 . 90 ± 0 . 10 μm , taxol vs . 1 . 19 ± 0 . 19 μm , untreated , SD , p<10−30 ) ( Figure 3E ) as well as NDC80 FRET fraction ( Figure 3D ) . As an alternative way to reduce the tension , we inhibited Eg5 with 5 μM S-trityl-L-cysteine ( STLC ) ( Skoufias et al . , 2006 ) . STLC-treated cells form monopolar spindles with reduced K-K distances ( 0 . 87 ± 0 . 10 μm , SD , p<10−30 ) ( Figure 3E ) . In these monopolar spindles , NDC80 FRET fractions from poleward-facing kinetochores was positively correlated with the K-K distance ( p<0 . 05 ) , while anti-poleward-facing kinetochores displayed reduced NDC80 FRET fraction with no significant correlation with K-K distance ( p=0 . 46 ) , presumably because many of these kinetochores are monotelically attached from the poleward side ( Figure 3—figure supplement 2 ) . The correlation between NDC80 FRET fraction and K-K distance was similar between taxol- and STLC-treated cells ( p=0 . 15 , see Figure 3D ) , arguing that the relation between NDC80 FRET fraction and K-K distance is insensitive to the precise perturbation used to reduce tension . Combining the data of untreated , taxol-treated and STLC-treated ( only poleward-facing kinetochores ) cells , we found that the NDC80 FRET fraction continually increases with K-K distance over the full range of K-K distance ( positive correlation , p<0 . 00005 ) ( Figure 3D ) . The extent of variation of NDC80-kMT binding with K-K distance is comparable to the extent of variation over the course of mitosis , from prometaphase to anaphase onset , as well as the extent of difference between centered and off-centered kinetochores in late prometaphase ( compare Figure 3D with 2A and D ) . Aurora B kinase is one of the best characterized components of the error correction process , and the N-terminal tail of the Hec1/Ndc80 protein in the NDC80 complex is a known substrate of Aurora B kinase that contains nine phosphorylation sites ( Tanaka et al . , 2002; Biggins et al . , 1999; Cheeseman et al . , 2006; DeLuca et al . , 2006; Ciferri et al . , 2008; Hauf et al . , 2003 ) . Previous biochemistry experiments demonstrated that the phosphorylation state of Hec1 modulates its binding to microtubules in vitro ( Cheeseman et al . , 2006; Zaytsev et al . , 2014 , 2015 ) . We used our FLIM-FRET technique to investigate the relationship between Aurora B kinase activity and NDC80-kMT binding in cells . We first added the ATP-competitive Aurora B inhibitor , ZM447439 , to late prometaphase cells , and observed a gradual increase in NDC80 FRET fraction over ~10 min , from 9% to nearly 18% ( corresponding to 21% NDC80 binding fraction before Aurora B inhibition and 43% after the inhibition ) ( Figure 4A and Figure 4—figure supplement 1A ) . Thus , Aurora B is a major modulator of NDC80-kMT binding in cells . This modulation of NDC80-kMT binding could occur directly , through phosphorylation of the N-terminal tail of the Hec1/Ndc80 protein , or indirectly , through other Aurora B substrates which are at kinetochores or which influence spindle assembly ( Carmena et al . , 2012; Krenn and Musacchio , 2015 ) . To further investigate the Aurora B modulation of NDC80 binding , we used non-phosphorylatable mutants of Hec1/Ndc80 protein , in which all nine identified Aurora B target sites are mutated to either aspartic acid ( a phospho-mimicking mutation ) or alanine ( a phospho-null mutation ) ( DeLuca et al . , 2011 ) . We replaced the endogenous Hec1 with wild-type Hec1 or the phosphomimetic mutant of Hec1 by RNAi knockdown and rescue ( Figure 4B , see Materials and methods ) . The NDC80 FRET fraction decreased with the number of phospho-mimicking mutations , from 16 . 0 ± 0 . 5% with 9A-Hec1 ( all nine phosphorylation sites substituted with Ala ) to 5 . 8 ± 0 . 5% with 2D-Hec1 ( two sites , S44 and S55 , substituted with Asp while the others with Ala ) and to 4 . 8 ± 0 . 3% with 9D-Hec1 ( all nine sites substituted with Asp ) ( mean ± SEM , Figure 4C ) . The average NDC80 FRET fraction of WT-Hec1 ( 9 . 3 ± 0 . 9% ) was similar , but slightly higher than 2D-Hec1 , consistent with previous results arguing that on average there are zero to two sites phosphorylated per Hec1/Ndc80 protein in prometaphase and metaphase ( Zaytsev et al . , 2014 ) . We next sought to determine if the increased binding of NDC80 to kMTs upon Aurora B inhibition is caused by a change of phosphorylation of the N-terminal tail of Hec1/Ndc80 protein . To this end , we added the Aurora B inhibitor , ZM447439 , to cells with endogenous Hec1 replaced with 2D-Hec1 , whose nine identified Aurora B target sites in the N-terminal tail cannot be phosphorylated . After the addition of ZM447439 , the NDC80 FRET fraction increases by only 0 . 02 ± 0 . 01 ( SE ) , which is statistically significant ( p<0 . 001 ) , but substantially smaller than the increase observed in cells with endogenous Hec1 ( 0 . 09 ± 0 . 01 ( SE ) ) ( compare Figure 4D–4A ) . Thus , the modulation of NDC80 binding to kMTs by Aurora B predominantly occurs through the phosphorylation of the N-terminal tail of Hec1/Ndc80 . We next quantified how Aurora B inhibition influences Aurora B activity in cells . We performed FLIM measurement on a cytoplasmic Aurora B FRET biosensor ( Fuller et al . , 2008 ) , which contains a kinesin-13 family Aurora B substrate whose phosphorylation obstructs intramolecular FRET between mTurquoise2 and YPet ( Figure 4—figure supplement 1B ) . During ZM447439 treatment , we found a continual reduction in the fraction of the Aurora B sensors in the non-FRET state , a proxy for Aurora B phosphorylation , from 0 . 540 ± 0 . 007 ( SEM ) to 0 . 368 ± 0 . 012 ( SEM ) ( Figure 4E and Figure 4—figure supplement 1C ) . Nuf2-targeted Aurora B sensor responded to the ZM447439 treatment with similar kinetics , arguing that the time scale of response to Aurora B inhibition is insensitive to the spatial location of the substrate ( Figure 4—figure supplement 1D ) . As the typical time scale of drug uptake is far slower than typical phosphorylation/dephosphorylation kinetics ( Thurber et al . , 2014; Huang et al . , 1997 ) , it is reasonable to assume that the phosphorylation level of Aurora B substrate is at steady state at each time point , so plotting the measured NDC80 binding fraction ( converted from FRET fraction ) vs . phosphorylated level ( converted from Aurora B sensor non-FRET fraction ) at each time point reveals their relationship . This analysis showed a graded dependence of NDC80-kMT binding on phosphorylation ( Figure 4F ) , which is consistent with the impact of phosphomimetic Hec1 mutants on NDC80 binding ( compare Figure 4C and F ) and results from previous in vitro assays ( Zaytsev et al . , 2014 , 2015 ) . The increased NDC80-kMT binding after Aurora B inhibition may underlie the reduction in detachment of kMTs from kinetochores after Aurora B inhibition , observed in photoactivation experiments ( Cimini et al . , 2006 ) . Aurora B regulates kinetochore-microtubule interactions , but its contribution to the tension-dependent stabilization of kinetochore-microtubule attachments is controversial ( Campbell and Desai , 2013; Salimian et al . , 2011; Akiyoshi et al . , 2010; Liu et al . , 2009; Tanaka et al . , 2002; Zaytsev et al . , 2016; Godek et al . , 2015; Lampson and Cheeseman , 2011; Haase et al . , 2017 ) . We next sought to determine if the correlation between NDC80-kMT binding and centromere tension that we observed ( Figure 3D ) is caused by the phospho-regulation of NDC80 by Aurora B . After replacing the endogenous Hec1 protein with 9A-Hec1 , which cannot be phosphorylated on the nine mutated target sites , we no longer observed a significant correlation between NDC80 FRET fraction and K-K distance ( p=0 . 20 ) ( Figure 5A ) . 9A-Hec1-expressing cells displayed significantly larger K-K distance than unperturbed cells ( 1 . 36 ± 0 . 21 µm , SD , p<10−30 ) ( Figure 5B ) , consistent with previous studies ( Tauchman et al . , 2015; Zaytsev et al . , 2014 ) . To investigate a wider range of K-K distance , we treated cells with taxol or STLC ( Figure 5-supplement figure 1A ) , as described above , and found no correlation over the full range of K-K distance ( p=0 . 29 , Figure 5A and B ) . Since the strong correlation between NDC80 binding and K-K distance ( Figure 3D ) is eliminated in the non-phosphorylatable Hec1 mutant 9A-Hec1 ( Figure 5A ) , this argues that the phosphorylation of the N-terminal tail of Hec1 is responsible for the correlation between NDC80-kMT binding and K-K distance . As Aurora B is believed to be the primary kinase that phosphorylates the N-terminal tail , this further suggests that the activity of Aurora B is responsible for the correlation between NDC80-kMT binding and K-K distance . Aurora B is localized to centromeres in prometaphase and metaphase ( Carmena et al . , 2012 ) . We next investigated if this localization is important for the correlation between NDC80-kMT binding and K-K distance . We used the haspin kinase inhibitor , 5-iodotubercidin ( 5-ITu ) , which has previously been shown to compromise the recruitment of Aurora B to inner-centromeres ( Wang et al . , 2012 ) ( Figure 5C ) . After 10 min of exposure of cells to 5-ITu , INCENP , a member of the chromosome passenger complex ( CPC ) , which also includes Aurora B ( Carmena et al . , 2012 ) , was drastically reduced at centromeres ( Figure 5D ) . Treating cells with 5-ITu for over 15 min did not significantly alter the average K-K distance ( 1 . 16 μm ± 0 . 18 μm , 5-ITu vs . 1 . 19 μm ± 0 . 19 μm , untreated , SD ) or the overall average fraction of NDC80 bound to kMTs ( FRET fraction , 11 . 79 ± 0 . 02% 5-ITu vs . 11 . 87 ± 0 . 02% , untreated , SEM ) , but eliminated the correlation between K-K distance and NDC80-kMT binding ( p=0 . 41 , Figure 5E and F ) . In order to investigate a wider range of K-K distance , we treated cells with both taxol and 5-ITu or with both STLC and 5-ITu ( Figure 5—supplement figure 1B ) , and found no correlation over the full range of K-K distance ( p=0 . 96 , Figure 5E and F ) . Our observation that the mislocalization of Aurora B from centromeres does not affect the K-K distance but compromises the correlation between NDC80 binding and K-K distance argues that the correlation shown in Figure 3D is due to tension causing an increase in NDC80-kMT binding , rather than NDC80-kMT binding causing an increase in tension . Moreover , the tension dependency of NDC80-kMT binding in human tissue culture cells depends on Aurora B recruitment to centromeres by haspin kinase , further arguing in favor of models in which phosphorylation by Aurora B plays a central role in chromosome autonomous error correction . We were surprised that the average NDC80-kMT binding does not significantly change after mislocalizing Aurora B with 5-ITu ( compare Figures 3D and 5E , and see Figure 5G ) . This suggests that Aurora B can still act on NDC80 even after the concentration of Aurora B at centromeres is greatly reduced . Consistent with this hypothesis , the Aurora B activity at kinetochores assessed by Nuf2-targeted Aurora B FRET sensor was not changed by 5-ITu treatment ( Figure 5H ) . Furthermore , when cells treated with 5-ITu were exposed to the Aurora B inhibitor ZM447439 , NDC80-kMT binding increased ( to NDC80 FRET fraction of 0 . 17 ± 0 . 01 , SEM ) and Aurora B activity at kinetochores decreased ( to Aurora B FRET sensor non-FRET fraction of 0 . 55 ± 0 . 01 , SEM ) , indistinguishable from the levels in cells not subject to 5-ITu exposed to ZM447439 ( Figure 5G and H ) . Thus , tension dependency of NDC80-kMT binding is conferred by Aurora B recruited to centromeres through a haspin-dependent pathway , while the average level of NDC80-kMT binding is also set by Aurora B , but in a manner that is not dependent on haspin . The extent to which Aurora B phosphorylates NDC80 depends on the activity of Aurora B and the concentration of Aurora B at NDC80 . To further investigate how the haspin-dependent pool of Aurora B confers tension dependency to NDC80-kMT binding , we next examined how Aurora B localization depends on K-K distance . We used spinning-disk confocal microscopy to image mNeonGreen-Nuf2 , to locate NDC80 , and INCENP-mCherry , to measure the distribution of Aurora B . We localized NDC80 to sub-pixel accuracy and identified sister kinetochore pairs ( see Figure 6A and Materials and methods ) . For each pair of kinetochores , we measured the intensity of INCENP-mCherry at the location of NDC80 , normalized on a cell-by-cell basis . Plotting the intensity of INCENP-mCherry at NDC80 as a function of K-K distance revealed a highly significant anti-correlation ( p<10−4 , Figure 6B ) . To explore a larger range of K-K distances , we treated cells with 10 μM taxol . Combining the data of untreated and taxol-treated cells , we found that the intensity of INCENP-mCherry at NDC80 linearly decreases with K-K distance over the full range of K-K distance ( p<10−6 ) . This observation suggests that the tension dependency of NDC80 binding may result from the decrease of Aurora B at NDC80 with increasing K-K distance . We next investigated how the concentration of Aurora B at the location of NDC80 is influenced by haspin inhibition . In the presence of 10 μM 5-ITu , the concentration of Aurora B at NDC80 was greatly reduced and independent of K-K distance ( Figure 6B ) . The lack of correlation between Aurora B concentration at NDC80 and K-K distance may explain the lack of tension dependency between NDC80-kMT binding and K-K distance upon haspin inhibition . We speculate that the finite concentration of Aurora B at NDC80 after haspin inhibition is the pool of Aurora B that maintains the average level of NDC80-kMT binding as described above . Taking together , our data suggest that the concentration of Aurora B at NDC80 determines the extent of NDC80 phosphorylation , which in turn determines the level of NDC80-kMT binding . To further explore this possibility , we sought to determine the relationship between Aurora B concentration at NDC80 and NDC80-kMT binding . We plotted the NDC80-kMT binding ( converted from the NDC80 FRET fraction in Figures 3D and 5E ) vs . the Aurora B concentration ( converted from the normalized INCENP-mCherry intensity in Figure 6D ) for each K-K distance , both with and without haspin inhibition ( Figure 7A , see Materials and methods ) . This revealed a highly nonlinear relationship: when the Aurora B concentration is lower than ~5 µM , the NDC80 binding fraction is independent of the Aurora B concentration , while for higher concentrations , the NDC80 binding fraction decreases with the Aurora B concentration ( Figure 7A ) . We constructed a mathematical model to determine if this nonlinear relationship can be explained by the known biochemistry of Aurora B and NDC80 ( Figure 7A ) and observed change in the concentration of Aurora B at NDC80 ( Figure 6B ) . In this model , we assume that there are two independent pools of Aurora B , haspin-dependent and haspin-independent , both of which engage in intermolecular autoactivation by phosphorylation in trans ( Zaytsev et al . , 2016; Xu et al . , 2010; Sessa et al . , 2005; Bishop and Schumacher , 2002 ) , and are inactivated by phosphatases ( Zaytsev et al . , 2016; Sessa et al . , 2005; Kelly et al . , 2007; Rosasco-Nitcher et al . , 2008 ) . The activated Aurora B phosphorylates NDC80 , which changes the binding affinity of NDC80 for kMTs ( Cheeseman et al . , 2006; Zaytsev et al . , 2014; Zaytsev et al . , 2015 ) . This model can be solved analytically , and is sufficient to account for the relationship between NDC80 phosphorylation and NDC80-kMT binding ( Figure 4F ) , and the relationship between Aurora B concentration at NDC80 and NDC80-kMT binding ( Figure 7A ) . In this model , the nonlinear relationship between Aurora B concentration at NDC80 and NDC80-kMT binding ultimately results from the activation dynamics of Aurora B: at low concentrations , dephosphorylation by phosphatases overwhelm the in trans autoactivation , but above a threshold Aurora B concentration , A* , these two processes balance , leading to steady state level of activated Aurora B that further increases with increasing Aurora B concentration . We next investigated if the same model can recapitulate the tension dependency of NDC80-kMT binding . Inputting the measured linear relationship between Aurora B concentration at NDC80 and K-K distance ( Figure 6B ) into the model reproduced the observed tension-dependent behavior of NDC80-kMT binding ( Figure 7B , left ) . Performing a similar procedure with the data from haspin inhibited cells ( Figures 5E and 6B ) revealed that the model successfully predicts both the level of NDC80-kMT binding upon haspin inhibition and its independence on K-K distance ( Figure 7B , right ) . Thus , this model provides a self-consistent , quantitative explanation of how the tension dependency of NDC80-kMT binding results from the biochemistry of Aurora B and NDC80 , and the measured change in Aurora B concentration at NDC80 with tension , without the need to invoke diffusible gradients or additional mechanochemistry .
In this study , we developed a method to quantitatively measure the binding of the NDC80 complex to microtubules at individual kinetochores in human tissue culture cells . Our method uses TCSPC FLIM-FRET which , in contrast to intensity-based FRET , allows quantitative measurements of the fraction of molecules engaging in FRET , even with spatially varying concentrations of donors and acceptors . We calibrated our measurements using control experiments and Monte Carlo simulations , allowing us to convert the fraction of donor-labeled Nuf2 engaged in FRET to the fraction of NDC80 complexes bound to kMTs . This technique can be extended to the quantitative assessment of other protein-protein interactions in living cells . Using this technique , we demonstrated that NDC80-kMT binding is regulated during prometaphase in a chromosome-autonomous manner . We observed that NDC80-kMT binding is strongly correlated to centromere tension , to an extent which is sufficient to account for the changes in NDC80-kMT binding over the course of prometaphase and metaphase . We characterized how Aurora B modulates NDC80-kMT binding in cells , which we found predominantly occurs through the phosphorylation of the N-terminal tail of Hec1 . We showed that the correlation between NDC80-kMT binding and centromere tension is dependent on the phosphorylation of the N-terminal tail of Hec1 . We determined the concentration of Aurora B at the locations of NDC80 , which decreases with increasing centromere tension . Mislocalizing Aurora B by inhibiting haspin kinase eliminated the tension dependency of NDC80-kMT binding , but did not change its average level . The observation that inhibiting haspin removes the correlations between NDC80-kMT binding and tension , and between Aurora B localization and tension , but does not affect the distribution of K-K distances , argues that these correlations are caused by the influence of tension on NDC80-kMT binding and Aurora B localization . A simple mathematical model of Aurora B autoactivation and NDC80 phosphorylation and binding can quantitatively explain these results . Taking together , this leads to a biophysical model of the tension dependency of NDC80-kMT interactions , which arises from the nonlinearity of Aurora B autoactivation and the change in Aurora B concentration at NDC80 with centromere tension . The FLIM-FRET technique developed in this study measures the fraction of NDC80 complexes whose Hec1 CH domains are bound to kMTs . Hypothetically , changes in NDC80 binding fraction might result from either alterations in the binding affinity of NDC80 or variations in the number of kMTs . Our results argue that the increase in NDC80 binding with increasing tension is caused by changes in affinity , because: ( 1 ) the addition of taxol causes a reduction in tension and a reduction in NDC80 binding ( Figure 3D and E ) , but an increase in the number of kMTs ( McEwen et al . , 1997 ) ; ( 2 ) taxol-treated and STLC-treated cells exhibit the same correlation between tension and NDC80 binding ( Figure 3D ) , suggesting that the decrease in binding with decreasing tension occurs independently of the mechanism of perturbation . Furthermore , previous estimates have found that each kMT can be contacted by approximately 40 NDC80 complexes ( Zaytsev et al . , 2014 ) , and since there are approximately 240 NDC80 complexes per kinetochore ( Suzuki et al . , 2015 ) , roughly 6 kMTs are sufficient for every NDC80 complex to be within reach of potential binding sites . Thus , NDC80 binding is expected to be insensitive to the number of kMTs if there are at least 6 kMTs . As on average there are ~20 kMTs per kinetochore in human cells ( Maiato et al . , 2004; Rieder , 1982 ) , small changes in the number of kMTs are unlikely to modify the fraction of NDC80 complexes bound to kMTs . However , in circumstances where there are very few kMTs , for example in early prometaphase , changes in the number of kMTs will lead to changes in NDC80 binding . We hypothesize that the observed increase in NDC80 binding over the course of prometaphase to metaphase results from the combination of changes in kMT number and the changes in NDC80 binding affinity , due to the continual increase in tension at those times ( Magidson et al . , 2011 ) . The difference in NDC80 binding between poleward-facing and anti-poleward-facing kinetochores in STLC-treated cells is likely caused by the difference in kMT numbers ( Figure 3—figure supplement 2 and Figure 5—figure supplement 1 ) . This effect is particularly likely to explain the difference in NDC80 binding between sister kinetochores with 9A-Hec1 , since these sister kinetochores are expected to exhibit the same NDC80 binding affinity ( Figure 5—figure supplement 1A ) . Error correction is believed to result from the regulation of the detachment of kMTs from kinetochores ( Godek et al . , 2015 ) . As NDC80 is the primary coupler of kinetochores to microtubules ( Cheeseman et al . , 2006; DeLuca et al . , 2006 ) , it is reasonable to hypothesize that the rate of kMT detachment from kinetochores might largely be governed by NDC80-kMT binding . Consistent with this , previous work showed that mutating NDC80 changes the number of kMTs in a manner that argues that increasing NDC80-kMT binding increases the stability of kMTs ( Guimaraes et al . , 2008; Zaytsev et al . , 2014 ) . Our work further supports the connection between NDC80-kMT binding and kMT stability by comparing our results with previous measurements of the rate of kMT detachment from kinetochores: NDC80-kMT binding increases during mitotic progression ( Figure 2 ) , while kMT stability increases ( Zhai et al . , 1995; Kabeche and Compton , 2013 ) ; NDC80-kMT binding increases in response to Aurora B inhibition ( Figure 4 ) , which causes an increase in kMT stability ( Cimini et al . , 2006 ) ; NDC80 preferentially binds to polymerizing kMTs over depolymerizing kMTs ( Figure 3C ) , while reconstituted kinetochores bind more strongly to polymerizing microtubules than depolymerizing microtubules ( Akiyoshi et al . , 2010 ) ; NDC80-kMT binding increases with increasing tension ( Figure 3D ) , and the stability of kMTs increase with increasing tension ( Nicklas and Koch , 1969; Akiyoshi et al . , 2010 ) . These comparisons argue that the NDC80-kMT binding is a major determinant of the kMT detachment rate . Hence , we propose that tension dependency of kMT detachment from kinetochores , which is believed to underlie error correction , results from the tension dependency of NDC80-kMT binding . If correct , this implies that error correction ultimately results from the nonlinear autoactivation of Aurora B and the consequent phosphoregulation of NDC80-kMT binding . Further testing this proposal will require additional quantitative measurements of kMT detachment , errors , and error correction , in combination with measurement of NDC80-kMT binding using the FLIM-FRET method presented here .
U2OS cell lines ( ATCC , HTB-96 ) were maintained in Dulbecco’s modified Eagle’s medium ( DMEM , Thermo Fisher ) supplemented with 10% Fetal Bovine Serum ( FBS , Thermo Fisher ) , and 50 IU ml−1 penicillin and 50 μg ml−1 streptomycin ( Thermo Fisher ) at 37°C in a humidified atmosphere with 5% CO2 . Cells were validated as mycoplasma free by PCR-based mycoplasma detection kit ( Sigma Aldrich ) . All live-cell FLIM and spinning-disk confocal microscopy imaging were performed as follows . Cells were grown on a 25 mm diameter , #1 . 5-thickness , round coverglass coated with poly-D-lysine ( GG-25–1 . 5-pdl , neuVitro ) to 80~90% confluency . The cells were incubated in imaging media , which is FluoroBrite DMEM ( Thermo Fisher ) supplemented with 4 mM L-glutamine ( Thermo Fisher ) and 10 mM HEPES , for 15 ~ 30 min before imaging . The coverglass was mounted on a custom-built temperature controlled microscope chamber at 37°C , while covered with 1 ml of imaging media and 2 ml of white mineral oil ( VWR ) . An objective heater ( Bioptech ) was used to maintain the objective at 37°C . We confirmed that the cells can normally divide longer than 6 hr in this condition . Only cells displaying proper chromosome alignment , normal spindle morphology , and high signal-to-noise ratio were selected for imaging and analysis . A tetracysteine ( TC ) tag , CCPGCC , was genetically attached to the C-terminal end of tubulin beta class I ( TUBB ) , an isotype of β-tubulin that is predominantly expressed in U2OS ( assessed by qPCR , data not shown ) and most other cancer cells ( Leandro-García et al . , 2010 ) . The attachment of the TC tag was achieved by CRISPR-induced homologous recombination to ensure the consistent expression of labeled β-tubulin . ssDNA ( IDT ) with TC tag ( 5’-TGCTGTCCCGGCTGTTGC-3’ ) and ~80 bp-long homology arms was used as a donor DNA . pSpCas9 ( BB ) −2A-GFP ( Addgene plasmid # 48138 ) ( Ran et al . , 2013 ) was utilized as a backbone for the plasmid carrying a sgRNA ( 5’-GAGGCCGAAGAGGAGGCCUA-3’ ) and Cas9 . The plasmid and the donor ssDNA were simultaneously delivered into U2OS cells by electroporation ( Nucleofector 2b and Amaxa Cell Line Nucleofector Kit V , Lonza ) . The insertion of the TC tag was verified through a PCR-based genotyping with primers 5'-GCATGGACGAGATGGAGTTCAC-3' and 5'-CCAGCCGTGTTTCCCTAAATAAG-3' , qPCR , and a fluorescence imaging after FlAsH-EDT2 staining . The U2OS cells expressing TC-tagged β-tubulin were further engineered to stably express Nuf2 N-terminally labeled with mTurquoise2 ( Goedhart et al . , 2012 ) by retroviral transfection , three times with different antibiotic selections , 1 μg ml−1 puromycin , 2 μg ml−1 blasticidin , and 200 μg ml−1 hygromycin ( all from Thermo Fisher ) . The retroviral vectors and their information are available on Addgene ( plasmid #: 80760 , 80761 , 80762 ) . Monoclonal cell line was obtained by single cell sorting . The protocol for the association of FlAsH-EDT2 with β-tubulin-TC in cell was adapted from the previous study ( Hoffmann et al . , 2010 ) so as to maximize the labeling fraction while maintaining cell viability . The engineered U2OS cells expressing β-tubulin-TC were grown to 80~90% confluency in a 30 mm cell culture dish , and then were gently washed with Opti-MEM ( Thermo Fisher ) twice , and then stained in 2 ml Opti-MEM with 1 μM FlAsH-EDT2 ( Thermo Fisher ) for 2 hr . To reduce the non-specific binding of FlAsH , the stained cells were subsequently incubated in Opti-MEM containing 250 μM 1 , 2-Ethanedithiol ( EDT , Alfa Aesar ) for 10 min , followed by a gentle wash with Opti-MEM . The cells were incubated in DMEM with 10% FBS for 6~10 hr before imaging , because they were found to be interphase-arrested for the first ~5 hr after the incubation with 250 μM EDT . Every buffers and media above were pre-warmed at 37°C before use . All incubation steps were performed at 37°C in a humidified atmosphere with 5% CO2 . Schematic instrumental setup of FLIM is shown in Figure 1—figure supplement 2A , and more details can be found in previous work ( Yoo and Needleman , 2016 ) . FLIM measurements were performed on a Nikon Eclipse Ti microscope using two-photon excitation from a Ti:Sapphire pulsed laser ( Mai-Tai , Spectral-Physics ) with an 80-MHz repetition rate and ~70 fs pulse width , a galvanometer scanner ( DCS-120 , Becker and Hickl ) , TCSPC module ( SPC-150 , Becker and Hickl ) and two hybrid detectors ( HPM-100–40 , Becker and Hickl ) . Objective piezo stage ( P-725 , Physik Instrumente ) and motorized stage ( ProScan II , Prior Scientific ) were used to perform multi-dimensional acquisition , and a motor-driven shutter ( Sutter Instrument ) was used to block the excitation laser between acquisitions . The wavelength of the excitation laser was set to 865 nm . 470/24 and 525/30 bandpass emission filters ( Chroma ) were mounted on each detector , and a dichroic beam splitter ( FF506-Di03 , Semrock ) was used for the simultaneous detection of mTurquoise2 and FlAsH fluorescence . The excitation laser was expanded to overfill the back-aperture of a water-immersion objective ( CFI Apo 40 × WI , NA 1 . 25 , Nikon ) . The power of the excitation laser was adjusted to 1 . 1~1 . 5 mW at the objective . All the electronics were controlled by SPCM software ( Becker and Hickl ) and μManager ( Edelstein et al . , 2014 ) . Scanning area was set to either 13 . 75 × 13 . 75 μm or 27 . 5 × 27 . 5 μm , and the pixel size was set to 107 nm . Each image was acquired for 3~5 s of integration time . Acquisition interval was set to 13 s for the FLIM-FRET data in Figures 3 and 5 , and 60~90 s for the FLIM-FRET data in Figures 2 and 4 . Three or four z-sections , separated by 1 μm , were acquired for each time point . No photo-bleaching or photo-damage was observed in this imaging condition . Mitotic phases were judged by the arrangement of kinetochores . For the kinetochore FLIM-FRET measurements shown in Figure 2–5 , custom-built MATLAB graphical user interphase ( GUI ) ( available at http://doi . org/10 . 5281/zenodo . 1198705 [Yoo , 2018b]; copy archived at https://github . com/elifesciences-publications/FLIM-Interactive-Data-Analysis ) was used to import Becker and Hickl FLIM data , track kinetochores , identify kinetochore pairs , extract the FLIM curve from each kinetochore , and estimate the FLIM parameters using a nonlinear least-squared fitting or Bayesian FLIM analysis , as described below and in previous work ( Yoo and Needleman , 2016 ) . The GUI also allows the users to scrutinize and manually correct the kinetochore trajectories and pairing . The kinetochore tracking algorithm was adapted from a particle tracking algorithm ( Pelletier et al . , 2009 ) , and the pair identification was performed by selecting pairs of kinetochores with distances and velocity correlations in predefined ranges . Drift correction was done by measuring correlation between two consecutive spindle images . The velocity vt of kinetochore ( in Figure 3 ) was estimated from the position x ( t ) using the five-point method:v ( t ) ≈ −x ( t+2Δt ) +8x ( t+Δt ) −8 ( t−Δt ) +x ( t−2Δt ) 12Δt Leading and trailing kinetochores ( in Figure 3 ) were determined based on the velocities and the relative positions of paired sister kinetochores . The metaphase plate ( in Figure 2 ) was determined by finding an equidistant plane between the two spindle poles ( that were manually located based on spindle images ) . Poleward-facing and anti-poleward-facing kinetochores ( in Figures 3 and 5 ) in STLC-treated cells were determined based on the relative positions of paired sister kinetochores and the position of the spindle pole , which was approximated by the average of the positions of kinetochores in each time point . Fluorescence decay curves from individual kinetochores at each time point contain only a few hundreds of photons . In this low photon count regime , FLIM analysis with conventional least-squared nonlinear regressions results in significantly biased estimate for the parameters ( Kaye et al . , 2017; Rowley et al . , 2016 ) . Therefore , we used a Bayesian approach , which has been described and tested previously ( Yoo and Needleman , 2016; Kaye et al . , 2017 ) , and is briefly explained below . Let θ be the set of parameters of the FLIM-FRET model , and y=yi be the observed FLIM data , where yi is the number of photons detected in the i-th time bin of the FLIM curve . Then the posterior distribution of θ ( assuming a uniform prior distribution ) isp ( θ|y ) ∝∏i=1NP ( tar∈[ ( i−1 ) Δt , iΔt]|θ ) yiwhere tar is the photon arrival time , and N is the number of time bins . Since the size of the time bin ( Δt , ~50 ps ) is much smaller than the time scale of fluorescence decay ( ~ns ) , the probability that the arrival time tar falls in the i-th time bin can be approximated by a Riemann sum:P ( tar∈[ ( i−1 ) Δt , iΔt]|θ ) ≅∑k= ( i−1 ) K+1k=iKhθ ( kΔt∼ ) Δt∼where hθ is the discretized FLIM model , Δt~ is the size of time bin with which instrument response function ( IRF ) is measured , and the ratio K=ΔtΔt~ is the ADC ratio , which is set to 16 for our data . hθkΔt~ can be written as the convolution between the IRF and an exponential decay model , gθ:hθ ( kΔt∼ ) = ( IRF∗ ( Agθ+ ( 1−A ) ) ( kΔt∼ ) ≅∑lmIRF[l−bshift] ( Agθ ( ( k−l ) Δt∼ ) + ( 1−A ) ) where mIRF is the IRF measured with the finest time bins of size Δt~ , and bshift is an integer parameter that determines the approximate shift of measured IRF relative to the theoretical IRF . 1-A indicates the relative contribution of noise that is uniformly distributed over time . The exponential decay model gθ ( td ) is set to exp-tdτ for the single-exponential decay model or 1-fFRETe-tdτD+fFRETe-tdτFRET for the two-exponential decay model , where 0≤fFRET≤1 is the FRET fraction . The posterior distribution was computed by Gibbs sampling if the number of free parameters is greater than 3 , or by grid sampling otherwise ( for example , when both long and short lifetimes are fixed ) . The instrument response function ( IRF ) was acquired by measuring second-harmonic generation from a urea crystal . Negative control FLIM measurements on the engineered cells ( mTurquoise2-NDC80/β-tubulin-TC ) not incubated with FlAsH were performed for every experiment and the fluorescence decay curves extracted from kinetochores were analyzed with a single-exponential FLIM-FRET model to determine the long non-FRET lifetime , which is usually 3 . 7 to 3 . 8 ns . The short FRET lifetime was estimated by performing a two-exponential Bayesian FLIM-FRET analysis on the aggregated FLIM data of kinetochores in each cell stained with FlAsH while fixing the non-FRET lifetime to the value pre-determined from the negative control . Then we performed a two-exponential Bayesian FLIM-FRET analysis , with both FRET and non-FRET lifetimes fixed to the predetermined values , on FLIM data from each kinetochore . Kinetochores were grouped by time ( Figures 2 and 4 ) , positions ( Figure 2B–D ) , velocities ( Figure 3C ) , and K-K distances ( Figures 3D , 5A and E ) . The posterior distributions in a group of kinetochores were multiplied and then marginalized to obtain the mean and SEM of the FRET fraction . We previously confirmed that this way of combining posterior distribution gives an unbiased estimate of the mean FRET fraction ( Kaye et al . , 2017 ) . NDC80 binding fraction was calculated by dividing NDC80 FRET fraction by the conversion factor 0 . 42 , which had been determined by the calibration shown in Figure 1—figure supplement 5C . An Aurora B FRET sensor was constructed by replacing CyPet in a previous construct ( Addgene plasmid # 45215 ) ( Fuller et al . , 2008 ) with mTurquoise2 . The FRET sensor contains a kinesin-13 family Aurora B substrate whose phosphorylation results in its binding to the forkhead-associated domain in the sensor , which constrains the sensor to be in an open conformation and obstructs intramolecular FRET between mTurquoise2 and YPet ( Figure 4—figure supplement 1B ) . Hence , the non-FRET fraction of the Aurora B FRET sensor is proportional to the Aurora B activity . The cytoplasmic Aurora B FRET sensor was stably expressed in U2OS cells by retroviral transfection ( plasmid available on Addgene , plasmid # 83286 ) . The Nuf2-targeted Aurora B FRET sensor was transiently transfected by electroporation ( Nucleofector 2b , Lonza; Ingenio Electroporation Kit , Mirus ) a day before imaging . The non-FRET fraction of the Aurora B FRET sensor was measured by FLIM-FRET in the same way as NDC80 FRET measurements described above . The exponential decay models ybinding ( t ) =A1-exp-It≥0tτ+c and yAurorat=Aexp-It≥0tτ+c were fitted to the time courses of NDC80 FRET fraction and FRET sensor non-FRET fraction after ZM447439 , respectively ( Figure 4A , D and E and Figure 4—figure supplement 1D ) , where It≥0 is equal to 0 if t is less than zero , and 1 otherwise . The estimated parameter values are given in Table 1: The fraction of Aurora B phosphorylation sites in NDC80 , fphos ( x-axis of Figure 4F ) , was converted from the non-FRET fraction of Aurora B FRET sensor , fsensor ( y-axis of Figure 4E ) , as follows . First , we assumed that fsensor increases linearly with fphos . Our result ( Figure 4C ) and previous work ( Zaytsev et al . , 2014 ) suggest that Ndc80 has about one phospho-residue out of nine phosphorylation sites in late prometaphase , based on which we assumed that fphosWT=1/9 before Aurora B inhibition , and fphosZM=0 after the full Aurora B inhibition . Since fsensor were measured to be fsensorWT=0 . 540±0 . 007 ( SEM ) before Aurora B inhibition and fsensorZM=0 . 368±0 . 012 ( SEM ) after the full Aurora B inhibition ( Figure 4E ) , we converted fsensor to fphos by:fphos=fsensor-fsensorZMfsensorWT-fsensorZMfphosWT-fphosZM+fphosZM=0 . 646fsensor-0 . 368 The fbound vs fphos data in Figure 4F was fit using a NDC80 binding model:fbound=1+K0+K0'fphos-1which is derived in Mathematical modeling section below . We used three different non-phosphorylable mutant Hec1 ( gift from Jennifer DeLuca ) in which all nine identified Aurora B target sites in the N-terminal tail are mutated to either Asp ( phospho-mimicking mutation ) or Ala ( phospho-null mutation ) ( DeLuca et al . , 2011; Zaytsev et al . , 2015 , 2014 ) : 9A-Hec1 ( all nine sites substituted with Ala ) , 2D-Hec1 ( two sites , S44 and S55 , substituted with Asp , while the other seven sites with Ala ) , and 9D-Hec1 ( all nine sites substituted with Asp ) . WT-Hec1 and the mutant Hec1 are C-terminally labeled with LSSmOrange . LSSmOrange signal at kinetochores were assessed to ensure the expression of the substituting Hec1 in cells . Cells were grown to 50% confluence on a 10 cm petri dish in DMEM supplemented with 10% FBS andpenicillin-streptomycin ( P/S ) as described above . To knock down endogenous Hec1/Ndc80 protein , we used a FlexiTube siRNA duplex targeted to the 5’ UTR of the Hec1 gene ( 5’-TCCCTGGGTCGTGTCAGGAAA-3’ , QIAGEN Hs_KNTC2_7 SI02653567 ) . We incubated 240 pmol of the siRNA in 1 . 2 mL Opti-MEM ( ThermoFisher 51985091 ) for 5 min with periodic flicking . We simultaneously incubated 8 μL of Lipofectamine RNAiMax ( ThermoFisher 13778030 ) in 1 . 2 mL Opti-MEM for 5 min . We then combined the siRNA and Lipofectamine solutions and incubated at room temperature for 30 min with periodic flicking . Prior to adding the siRNA-lipid complex , we washed the cells once with PBS and then replaced the media with 8 mL Opti-MEM supplemented with 10% FBS . We then added the entire 2 . 4 mL siRNA mixture to the cells dropwise and incubated the cells at 37°C for 30 hr . Following the incubation , we nucleofected 2 μg of plasmid encoding WT- , 9A- , 2D- , or 9D-Hec1 along with an additional 30 pmol of Hec1 siRNA into 1 million cells using a Lonza Nucleofector 2b . We spread these cells evenly over three 35 mm dishes containing 25 mm poly-D-lysine coated coverslips and 2 mL of Opti-MEM supplemented with 10% FBS and P/S . We incubated overnight at 37°C for 18 hr before staining with TC-FlAsH and FLIM-FRET imaging as described above . Cells were incubated with 5 μM Nocodazole ( Sigma Aldrich ) for >10 min for microtubule depolymerization . Aurora B inhibition was performed by adding 3 μM of ZM447439 ( Enzo Life Sciences ) during imaging . Taxol ( Enzo Life Sciences ) treatment was performed at 10 μM final concentration for >10 min . S-Trityl-L-cysteine ( STLC , Sigma Aldrich ) treatment was performed at 5 μM final concentration for >60 min to induce monopolar spindles . For the haspin kinase inhibition , cells were treated with 10 μM 5-iodotubercidin ( 5-ITu , Enzo Life Sciences ) for >10 min . The double treatment of 5-ITu and taxol or STLC was performed sequentially by treating cells with 10 μM taxol or 5 μM STLC and then adding 10 μM 5-ITu . Here we describe the mathematical model presented in Figure 7 in detail . The model predicts NDC80 binding fraction from Aurora B concentration at NDC80 in three steps: ( 1 ) Aurora B activation dynamics , consisting of autoactivation in trans and deactivation , which determines the concentration of active Aurora B from the concentration of Aurora B; ( 2 ) NDC80 phosphorylation , which is dependent on the active Aurora B concentration; and ( 3 ) NDC80-kMT binding , which is governed by the phosphorylation level of NDC80 . Protein structure illustrations were generated by The PyMOL Molecular Graphics System , Version 2 . 0 Schrödinger , LLC . The statistical test used , sample size ( number of cells and kinetochores ) , dispersion and precision measures can be found in figure legends , Results , or below . All curve fittings , except FLIM data analysis ( which is separately explained above ) , were performed by Levenberg-Marquardt algorithm with residuals weighted by the inverse of y-errors , and the corresponding 95% confidence intervals were calculated by predint function in MATLAB . To assess the significance of correlation , we determined p-value from 1-α , where α is the smallest confidence level that makes zero contained in the confidence interval of the slope of the linear fit . | When a cell divides , each new cell that forms needs to contain a complete set of DNA , which is stored in structures called chromosomes . So first , the chromosomes duplicate , and the two copies are held together . A protein structure known as a kinetochore then forms on each copy of the chromosome . The kinetochores act as a pair of hands that pull the chromosome copies apart and toward opposite sides of the dividing cell . They do this by grabbing protein ‘ropes’ called microtubules that extend toward the chromosomes from each side of the cell . Kinetochores grip the microtubule ropes more tightly when the connection is under greater tension . This helps the kinetochores to remain attached to the microtubules that will separate the chromosome copies while releasing the microtubules that would pull both copies to the same side . Previous research has shown that hundreds of finger-like structures made out of a protein group called NDC80 extend from each kinetochore ‘hand’ and attach to the microtubules . What remains a mystery is whether and how the NDC80 fingers grip the microtubules more tightly when tension is greater in cells . Yoo et al . developed a technique for counting how many of the available NDC80 fingers of a single kinetochore are attached to microtubules within a living human cell . The new technique combines genetic engineering , fluorescence imaging and statistical methods to quantify the attachment of NDC80 to microtubules over time and space . Yoo et al . found that more NDC80 bound to microtubules when there was greater tension . This relationship between binding and tension depends on an enzyme called Aurora B , which modifies the tip of each NDC80 finger and consequently changes the binding of NDC80 to microtubules . Yoo et al . further showed that Aurora B needs to be properly placed between two kinetochore hands to make NDC80-microtubule binding dependent on tension . Without this tension dependency , chromosomes could segregate unevenly into the newly formed cells – a problem that can lead to cancer , infertility and birth defects . The results presented by Yoo et al . therefore expand our understanding of how these diseases originate and may eventually help researchers to develop new treatments for them . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"cell",
"biology"
] | 2018 | Measuring NDC80 binding reveals the molecular basis of tension-dependent kinetochore-microtubule attachments |
Cell fate determination during development often requires morphogen transport from producing to distant responding cells . Hedgehog ( Hh ) morphogens present a challenge to this concept , as all Hhs are synthesized as terminally lipidated molecules that form insoluble clusters at the surface of producing cells . While several proposed Hh transport modes tie directly into these unusual properties , the crucial step of Hh relay from producing cells to receptors on remote responding cells remains unresolved . Using wing development in Drosophila melanogaster as a model , we show that Hh relay and direct patterning of the 3–4 intervein region strictly depend on proteolytic removal of lipidated N-terminal membrane anchors . Site-directed modification of the N-terminal Hh processing site selectively eliminated the entire 3–4 intervein region , and additional targeted removal of N-palmitate restored its formation . Hence , palmitoylated membrane anchors restrict morphogen spread until site-specific processing switches membrane-bound Hh into bioactive forms with specific patterning functions .
Hedgehog ( Hh ) morphogens are dually lipidated 19 kDa proteins that are firmly anchored to the cell membrane of producing cells . Production of all active Hhs begins with autocatalytic cleavage of a precursor molecule by its C-terminal cholesterol transferase domain ( Porter et al . , 1996b ) . This results in cholesteroylated vertebrate Sonic hedgehog ( Shh ) and Drosophila Hh . Next , Hh acyltransferase ( Hhat , also designated Skinny hedgehog or Raspberry ) attaches a palmitoyl group to a conserved N-terminal cysteine that becomes exposed after signal peptide cleavage ( Chamoun et al . , 2001; Lee and Treisman , 2001; Micchelli et al . , 2002 ) . Hh palmitoylation is critical for later signaling , demonstrated by mutation of the N-terminal cysteine to serine or alanine ( C25 > A/S in ShhC25A/S , C85 >A/S in Drosophila HhC85A/S ) which abolishes palmitoylation and results in morphogen inactivity ( Chamoun et al . , 2001; Chen et al . , 2004; Dawber et al . , 2005; Goetz et al . , 2006; Kohtz et al . , 2001; Lee et al . , 2001; Pepinsky et al . , 1998 ) . However , why N-palmitoylation is required for Hh signaling in vivo is still unclear . Another unusual feature of all Hhs is their multimerization at the surface of producing cells which requires binding to the long , unbranched heparan sulfate ( HS ) chains of cell surface HS proteoglycans ( HSPGs ) called glypicans ( Chang et al . , 2011; Ortmann et al . , 2015; Vyas et al . , 2008 ) . The Hh cholesterol modification is sufficient to drive this process ( Feng et al . , 2004; Gallet et al . , 2006; Koleva et al . , 2015; Ohlig et al . , 2011 ) . Despite membrane anchorage and cell-surface HS association , the multimeric Hhs initiate the Hh response in distant cells that express the Hh receptor Patched ( Ptc ) . The question of how dual-lipidated Hh clusters manage to travel and signal to remote target cells is intensely investigated . The most current models propose lipidated Hh transport on filopodia called cytonemes ( Bischoff et al . , 2013; Sanders et al . , 2013 ) or on secreted vesicles called exosomes ( Gradilla et al . , 2014 ) to bridge the distance between Hh-producing and receiving cells . Hh release through cell-surface-associated proteases , called sheddases , has also been suggested . In vitro , membrane-proximal shedding not only releases Hh ectodomains from their lipidated N-terminal peptides ( Dierker et al . , 2009; Ohlig et al . , 2011 ) but also activates Hh clusters . This is because N-terminal lipidated peptides block adjacent Hh-binding sites for the receptor Ptc and , thereby , render Hh at the cell membrane inactive . By cleaving these inhibitory peptides during release , sheddases unmask Ptc binding sites of solubilized clusters and thereby couple Hh solubilization with its bioactivation . In this model , the N-palmitate plays two indirect roles for Hh biofunction: first , it ensures reliable membrane-proximal positioning of inhibitory N-terminal peptides as a prerequisite for their efficient proteolytic processing , and second , by its continued association with the cell membrane , it ensures that only fully processed ( =activated ) Hh clusters are released . This model therefore predicts that inhibition of N-palmitoylation will result in release of inactive soluble proteins with masked Ptc-binding sites ( Jakobs et al . , 2014; Jakobs et al . , 2016; Ohlig et al . , 2011; Ohlig et al . , 2012 ) . It also predicts that impaired or delayed processing of dual-lipidated Hh will strongly reduce its release and bioactivity in vivo . By uncovering a dominant negative , cell-autonomous function of non-palmitoylated HhC85S in endogenous Hh , we here support the first prediction . By using a series of transgenic Drosophila melanogaster lines that express untagged Hh , biologically inactive HhC85S , or N-truncated variants thereof in posterior and anterior wing disc compartments , we provide strong evidence that Hh clusters form by direct protein-protein contact and that unprocessed N-terminal peptides block Ptc binding of adjacent endogenous Hhs . As a consequence , we suggest that , due to their reduced activity , soluble clusters with masked Ptc-binding sites impair direct patterning of the 3–4 intervein region of the wing . Supporting this mechanism , targeted deletion on non-palmitoylated inhibitory peptides restores 3–4 intervein formation . We also show that impaired or delayed processing of lipidated Hh strongly reduces its solubilization , and hence its bioactivity , in vivo . We demonstrate that the HS-binding Cardin-Weintraub ( CW ) motif serves as the preferred N-terminal Hh processing site in vivo , and that impaired processing of this site completely abolishes direct 3–4 intervein wing patterning . Additional targeted deletion of N-palmitate restores wing patterning , demonstrating that one role of palmitoylated membrane anchors is to prevent the release of un- or incompletely processed Hh clusters in vivo . These genetic data are supported by the nano-structure of Hh clusters as revealed by immunoelectron microscopy ( IEM ) and provide new insights into how Hh relay from the producing cell membrane or between membranes could be achieved .
A first step in decoding possible Hh solubilization modes is to characterize the composition and organization of Hh substrates . It has been previously shown that Hh forms light microscopically visible clusters at the surface of producing cells ( Chen et al . , 2004; Gallet et al . , 2006; Ortmann et al . , 2015; Sanders et al . , 2013; Vyas et al . , 2008 ) . However , the nanoscale structure of these heteroprotein complexes has not been determined . We therefore expressed Shh together with Hh acyltransferase in HEK293-derived Bosc23 cells to produce authentic cell surface Hh clusters for IEM analysis . To this end , we used several different α-Shh antibodies and secondary antibodies conjugated to 5 nm or 10 nm gold particles . Three α-Shh antibodies detected Shh in variably sized cell surface clusters , with the largest complexes exceeding sizes of 100 nm . Notably , as shown in Figure 1a–h and Figure 1—figure supplement 1 , many clusters consisted of linear arrangements ( Figure 1a–d ) or contained linear arrays of closely packed gold particles ( Figure 1f , g , arrowheads ) . Nearest-neighbour analysis of the angular distribution between the three most proximal gold particles ( Figure 1i ) confirmed that most arrangements were rectangular ( 90° ) or linear ( 180° ) , the latter being consistent with Hh multimerization using linear HS chains of glypican HSPGs as templates ( Chang et al . , 2011; Vyas et al . , 2008; Schuermann et al . , 2018 ) . Hh linearization during cell-surface multimerization is further consistent with previous structural and biochemical data which suggest a zig-zag arrangement of Hh monomers ( Figure 1j ) and variably sized Drosophila Hh and vertebrate Shh , ranging from 80 kDa to 600 kDa ( Chang et al . , 2011; Chen et al . , 2004; Jakobs et al . , 2014; Ohlig et al . , 2011; Ohlig et al . , 2012 ) . We therefore next aimed to genetically confirm direct Hh clustering in vivo by using Drosophila melanogaster wing development as a model . The fly wing develops from the imaginal wing disc ( Figure 2a , bottom ) . The wing primordium at the center of the wing disc differentiates into the wing blade proper , which shows a characteristic pattern of five longitudinal veins ( L1-5 ) , an anterior cross vein ( connecting L3 and L4 ) and a posterior cross vein ( connecting L4 and L5 ) ( Figure 2a , top ) ( Hartl and Scott , 2014 ) . The anterior/posterior ( a/p ) boundary is located slightly anterior to the position of L4 in the adult wing ( Figure 2a , red dashed line ) . Hh is produced in the posterior wing disc compartment under the control of the transcription factor Engrailed ( en ) ( Tabata et al . , 1992; Zecca et al . , 1995 ) , which acts indirectly on Hh expression through the repression of the transcriptional Hh repressor Cubitus interruptus ( Ci ) ( Bejarano and Milán , 2009 ) . Hh then moves across the a/p boundary into the anterior compartment , where it binds to Ptc ( Ingham et al . , 1991 ) . During its movement , Hh forms a gradient of decreasing concentration with increasing distance from the a/p border which corresponds to differential activation of different Hh target genes . Up to ten cell diameters from the a/p boundary , high Hh levels directly pattern the central L3-L4 region of the wing ( Mullor et al . , 1997; Strigini and Cohen , 1997 ) by stabilizing Ci155 . More distal regions , up to 12–15 cell diameters from the a/p border , depend on Dpp , which is secreted in a stripe just anterior to the a/p boundary in response to low Hh levels . Hh thus plays a role in Drosophila wing patterning by controlling the spatially defined expression of target genes at the a/p border . We exploited the Hh-regulated wing patterning response as a simple and reliable in vivo assay to test the functional consequences of proteolytic Hh processing . Specifically , we addressed the formation and positioning of longitudinal L3-L4 veins , and investigated whether Hh proteolytic processing in cells of the posterior compartment is a prerequisite for its signaling activity in cells of the anterior compartment ( Crozatier et al . , 2004 ) . To this end , comparable amounts of Hh and Hh variants ( Figure 2b , Supplementary file 1 ) were expressed from one specific attP 51C landing site on the second chromosome ( Bateman et al . , 2006 ) using the Gal4/UAS system ( Brand and Perrimon , 1993 ) . In the posterior compartment , Hh was expressed under en-Gal4 control , which is referred to as en >Hh , while in the anterior compartment , Hh was expressed in a stripe of cells under ptc-Gal4 control , referred to as ptc >Hh . As previously shown ( Crozatier et al . , 2004; Lee et al . , 2001; Mullor et al . , 1997; Strigini and Cohen , 1997 ) , en >Hh expanded the L3-L4 intervein area and , as a concomitant effect , reduced the L2-L3 intervein space ( Figure 2c ) . By contrast , en-regulated overexpression of non-palmitoylated , biologically inactive HhC85S ( en >HhC85S ) resulted in L3-L4 veins being proximally apposed and the formation of ectopic anterior cross veins ( Figure 2d ) ( Crozatier et al . , 2004; Lee et al . , 2001 ) , suggesting that HhC85S competes with bioactive wild-type Hh ( Lee et al . , 2001 ) . This phenotype is consistent with the higher Hh concentrations required for activation of the target genes , ptc and collier , and L3-L4 development , than those required for the activation of dpp , which patterns the remainder of the wing ( Hooper , 2003; Méthot and Basler , 1999; Mohler et al . , 2000; Strigini and Cohen , 1997; Vervoort et al . , 1999 ) . Wing phenotypes were quantified by dividing the proximal L3-L4 areas by the L2-L3 areas ( Figure 2e , f ) . This revealed significant Hh gain of function upon Hh overexpression in the posterior compartment or loss of function upon HhC85S overexpression ( en >GFP served as a normal control: L3-L4/L2-L3 = 0 . 074 ± 0 . 002; en >Hh = 0 . 108 ± 0 . 005 ( +46% ) , p<0 . 0001; en >HhCC85S=0 . 032 ± 0 . 002 ( -57% ) , p<0 . 0001 ) . To investigate the molecular basis of the dominant-negative HhC85S activity in wing disc tissues , we spatially disconnected HhC85S expression from endogenous Hh expression by using ptc >HhC85S . In the event that biologically inactive HhC85S would impair the response to Hh in a non-cell autonomous manner , for example , by binding to and blocking the receptor Ptc , ptc >HhC85S wing phenotypes should be comparable , or even more severe than those observed in en >HhC85S wings . Alternatively , if unprocessed N-terminal HhC85S peptides directly inhibit Ptc binding of associated endogenous Hh produced in the same compartment , we expected ptc >HhC85S wing phenotypes to be less severe than those observed in en >HhC85S wings . Indeed , HhC85S expression under ptc control had little effect on wing development ( Figure 2g , h: ptc >GFP: 0 . 084 ± 0 . 001 , ptc >HhC85S: 0 . 078 ± 0 . 002 ( -7% ) , p=0 . 0026 ) , suggesting that HhC85S cell-autonomously interferes with endogenous Hh , possibly by the random mixing of inactive HhC85S and wild-type Hh at the cell surface ( Figure 2i , left ) . In this mixed association , unprocessed HhC85S N-terminal peptides block wild-type Hh-receptor-binding sites in trans . By contrast , ptc>HhC85S expression in the anterior compartment prevents mixed cluster formation and therefore does not affect the controlled secretion and signaling of endogenous Hh ( Figure 2i , right ) . To independently confirm that HhC85S dominant-negative function requires direct Hh/HhC85S association with the same clusters , we expressed unlipidated monomeric HhNC85S ( Porter et al . , 1996a ) in vitro and in vivo ( Figure 3a ) . We observed that the expression of soluble HhNC85S under en-control did not affect endogenous Hh function in vivo ( Figure 3b ) , as expected from its exclusion from lipidated Hh clusters at the cell surface . As shown in Figure 3c , relative L3-L4/L2-L3 ratios obtained from three independent HhNC85S lines ( 1-3 ) revealed that wing development was not significantly affected ( en >GFP: 0 . 074 ± 0 . 002 , en >HhNC85S ( 1 ) : 0 . 07 ± 0 . 001 ( -5% ) ( p=0 . 0707 ) , en >HhNC85S ( 2 ) : 0 . 074 ± 0 . 001 ( ±0% ) ( p=0 . 9419 ) , en >HhNC85S ( 3 ) : 0 . 075 ± 0 . 002 ( +1% ) ( p=0 . 6050 ) , 20 wings were quantified in each line ) . Notably , HhNC85S expression under ptc control resulted in a small , yet significant gain-of-function phenotype ( Figure 3d , e , ptc >GFP: 0 . 084 ± 0 . 001 , ptc >HhNC85S ( 1 ) : 0 . 091 ± 0 . 002 ( +8% ) ( p=0 . 0013 ) , ptc >HhNC85S ( 2 ) : 0 . 089 ± 0 . 002 ( +6% ) ( p=0 . 0348 ) , ptc >HhNC85S ( 3 ) : 0 . 093 ± 0 . 001 ( +11% ) ( p<0 . 0001 ) , 20 wings were quantified in each line ) . This is consistent with the concept that Hh inactivation by adjacent unprocessed N-terminal peptides in trans is restricted to clustered , but not unclustered proteins ( Ohlig et al . , 2011; Ohlig et al . , 2012 ) . We conclude that the lack of Hh inhibition by monomeric HhNC85S ( Figure 3f ) , even if expressed in the same cells , is consistent with required direct association of palmitoylated and non-palmitoylated morphogens for dominant-negative HhC85S function . We next determined the molecular basis of the cell-autonomous inhibitory activities of HhC85S . In vitro , N-terminal peptides block Ptc-receptor-binding sites of adjacent Hh molecules in the cluster in trans ( Figure 4—figure supplement 1a–c ) ( Ohlig et al . , 2011 ) . Thus , we predicted that N-terminal truncation of HhC85S should restore Hh biofunction in mixed clusters . To test this idea and to mimick Hh processing observed in L3 Drosophila larvae ( Figure 4—figure supplement 1d , e ) , we consecutively deleted N-terminal amino acids 86–91 ( HhC85S;Δ86-91 ) to 86–100 ( HhC85S;Δ86-100 ) ( Figure 4a ) and confirmed unimpaired protein expression ( Figure 4b ) and multimerization ( Figure 4—figure supplement 2 ) . All ten constructs were then inserted into the attP-51C landing site on the second chromosome to ensure comparable expression . At least three independent transgenic fly lines were derived from each construct and crossed with the en-Gal4 driver line . We observed unchanged or moderately changed L3-L4/L2-L3 intervein ratios between en >HhC85S and en >HhC85S;Δ86-91 to en >HhC85S;Δ86-97 adult wings ( Figure 4c–e , i and Figure 4—figure supplement 3a ) . However , protein truncation beyond residue R97 gradually restored the biological activity of mixed clusters: en >HhC85S;Δ86-98 and en >HhC85S;Δ86-99 fly wings showed partially restored wing patterning ( Figure 4f , g , i ) and , strikingly , the posterior expression of HhC85S;Δ86-100 fully restored normal wing patterning ( Figure 4h , i and Figure 4—figure supplement 3a ) . Cell-autonomous inhibitory activities of HhC85S and restored wing patterning upon targeted coexpression of en >HhC85S;Δ86-100 were confirmed with the independent en-driver lines en ( 2 ) -Gal4 and hh-Gal4 , both controlling transgene expression in the wing disc ( Figure 4—figure supplements 4 and 5 ) . These results are consistent with the assembly of Hh clusters by direct protein-protein contacts as a prerequisite for the inhibitory activity of unprocessed N-terminal peptides . We also observed that wing phenotypes varied between and within fly lines ( labeled 1–4 in Figure 4—figure supplement 3b ) . This variability can be explained by slightly different expression levels or by small stochastic changes in Hh/HhC85S cluster composition with increasing relative amounts of HhC85S , resulting in stronger dominant negative phenotypes . Indeed , temperature-dependent Gal4-regulated transgene amounts ( Duffy , 2002 ) affected dominant-negative wing phenotypes: At 29°C , increased amounts of HhC85S relative to ( fixed ) endogenous Hh inhibited Hh function more strongly , whereas reduced transgene expression at 18°C inhibited Hh function less strongly ( Figure 4—figure supplement 6 ) . Taken together , we conclude that N-palmitate serves to ensure reliable membrane-association of inhibitory N-termini , making quantitative peptide processing a prerequisite for the solubilization of fully activated clusters ( Figure 5a ) . As a consequence , Hh concentrations at any position in the gradient will strictly correlate with their biological activities ( i . e . their Ptc-binding capacities ) . Impaired N-palmitoylation in this scenario reduces Hh bioactivity to variable degrees , depending on the relative number of unprocessed N-terminal peptides in soluble clusters ( Figure 5b ) . This essential ‘cleavage/activation control’ function is confirmed by fully restored Hh biofunction upon targeted coexpression of en >HhC85S;Δ86-100 ( Figure 5c ) . In our model , N-palmitate tethers incompletely processed Hh clusters to the cell membrane to prevent their release . To test this hypothesis , we utilized a cell culture model employing Bosc23 cells . To achieve quantitative Hh N-palmitoylation in vitro , we used bicistronic mRNA constructs to couple Shh ( the vertebrate Hh ortholog ) and Hh acyltransferase expression in the same cells . We then compared the release of fully lipidated Shh , non-palmitoylated ShhC25S , and variants carrying the extended C-terminal membrane anchor N190SVAAKSG-YPYDVPDYA-G198 ( G198 represents the cholesterol-modified glycine; italicized underlined letters represent the tag , Figure 6 ) ( Jakobs et al . , 2014 ) . Proteins were detected by polyclonal α-Shh antibodies and monoclonal α-HA antibodies on the same ( stripped ) blots . Grayscale blots were inverted , colored ( green: α-Shh signal , blue: α-HA signal ) and merged to identify proteins bound by both antibodies ( yielding bright blue/cyan signals ) and proteins bound by only α-Shh antibodies ( green signals ) . As shown in Figure 6a , dual-lipidated Shh and ShhHA yielded strong cellular signals but were absent from media , indicating impaired release . By contrast , non-palmitoylated ShhC25S;HA was effectively converted into a C-terminally truncated soluble morphogen , as indicated by an electrophoretic size shift and lack of α-HA antibody reactivity ( compare the cellular ( c ) material in each lane 1 with corresponding media in each lane 3 ) . Three independent quantifications of dual-lipidated Shh , cholesterylated ShhC25S , and non-lipidated ShhNC25S in cells and media ( Figure 6b ) confirmed that N-palmitoylation controls protein solubilization in vitro ( 1 hr release: ShhC25A 94 ± 3 arbitrary units ( a . u . ) , ShhNC25A 265 ± 10 a . u . , p<0 . 0001 , n = 3; 4 hr release: Shh 18 ± 2 a . u . , ShhC25A 238 ± 6 a . u . , p>0 . 0001 , n = 3 , ShhNC25A 500 ± 58 a . u . , p>0 . 001 , n = 2; values express ratios between solubilized/cell-associated proteins ) . Accordingly , coexpression of dual-lipidated Shh and ShhC25A in the same Bosc23 cells resulted in mixed clusters and thereby a four-fold reduction in ShhC25A release ( Shh+ShhC25A: 25 . 7 ± 5% , ShhC25A alone was set to 100% , p<0 . 0001 , n = 7 ) ( Figure 6c ) . Importantly , we further observed that dual-lipidated , N-terminally HA-tagged HAShh was not released ( Figure 6d ) . In this construct , the HA tag was inserted at the position of the membrane-proximal CW motif , shifting this previously identified sheddase target site ( Ohlig et al . , 2012 ) distally while not affecting its HS-binding capacity . To test whether the same modification would also impair release of fly Hh , we inserted an HA tag between corresponding Hh amino acids L91 and G92 , resulting in the N-terminal HAHh sequence C85GPGRGL91-YPYDVPDYAG92-RHRARN ( bold letters represent the CW motif that is shifted nine amino acids away from the preferred membrane proximal site of sheddase activity ) . We also used Hh , non-palmitoylated HhC85S and HhC85S;Δ86-100 as controls . HAHh was expressed in S2 cells , its unimpaired multimerization confirmed ( Figure 6—figure supplement 1 ) , and cellular and soluble proteins compared by SDS-PAGE and immunoblotting ( Figure 6e ) . We observed that all proteins were produced in S2 cells , as indicated by strong α-Hh antibody binding to all cellular forms . In contrast , only low levels of HAHh that retained the tag were solubilized , suggesting that N-terminal processing was impaired in S2 cells . From these experiments , we conclude that N-palmitoylation controls Hh release from the cell surface and restricts possible modes of Hh solubilization to shedding ( Figure 6f , g ) . We next generated transgenic flies expressing HAHh ( Figure 7a ) in the posterior compartment . The HA-tagged protein , due to its invariable association with the membrane ( Figure 6e ) and direct association with endogenous Hh in mixed clusters , was expected to impair endogenous Hh release and to lead to severe dominant-negative mis-patterning phenotypes . Indeed , HAHh expression in the posterior compartment at 25°C largely arrested fly development at the pupal and pharate stages , leading to defective head development characteristic of Hh loss of function ( Torroja et al . , 2004 ) . Of 230 pharates counted , only three imagos hatched with smaller wings lacking anterior structures ( Figure 7b ) , again characteristic of Hh loss of function ( Bejarano et al . , 2012 ) . Reduced transgene expression at 18°C largely reversed pharate lethality: 77% of en> HAHh pharates hatched ( 293 flies from 393 pupae ) but wing development was still impaired with all analyzed wings lacking all or most of the L3-L4 intervein area ( Figure 7c ) . This phenotype resembles wing phenotypes of flies expressing non-diffusible HhCD2 ( Strigini and Cohen , 1997 ) or with impaired activity of Hh signaling components such as Fused or Collier ( Col ) ( Ascano and Robbins , 2004; Vervoort et al . , 1999 ) , while the distal ‘widening’ of L3 ( Figure 7c ) is consistent with impaired Hh repression of Iroquois-regulated L3 formation ( Crozatier et al . , 2004 ) . Notably , the observation that HAHh expression under ptc-control showed only minor effects ( Figure 7h , i ) confirms cell-autonomous Hh repression by direct HAHh contacts in mixed clusters , and suggests that palmitoylated HAHh peptides restrain these mixed clusters at the cell membrane . We therefore expected that additional C > S mutagenesis , by removing the membrane anchor ( Figure 7a ) , would revert the observed severe mis-patterning phenotypes due to impaired cluster release into milder forms caused by partially impaired Hh binding to Ptc , as described earlier . Indeed , additional mutagenesis of the palmitate acceptor cysteine in en >HAHhC85S flies fully reversed pharate lethality at 25°C and led to wing phenotypes comparable to those of en >HhC85S flies ( compare Figure 7d with 7 f and Figure 7e with 7 g ) . The Hh gradient emanating from the posterior compartment activates the Hh target genes engrailed ( en ) , collier ( col ) , patched ( ptc ) and decapentaplegic ( dpp ) in stripes of anterior cells adjacent to the a/p border . Via Hh-responsive accumulation and nuclear access of Ci155 , en and col are induced in a 5- to 7 cell wide anterior stripe , ptc in a 10 cell wide stripe and dpp in a 12–15 cell wide stripe , and the presence and width of these stripes of target gene expression is differentially sensitive to Hh dose ( Chen and Struhl , 1996; Strigini and Cohen , 1997 ) . Far from the Hh source , Ci155 is depleted to form the repressor CiR , and Hh target genes are repressed . Cells receiving minimal amounts of Hh activate dpp transcription , cells receiving an intermediate amount of Hh activate the expression of col and ptc in addition to that of dpp , and Hh-dependent anterior en transcription ( but not posterior , Hh-independent en transcription ) is located closest to the a/p border ( Figure 8a , b ) . Col in the high and intermediate zones down-regulates Dpp responses: This results in the future L3-L4 intervein ( Mohler et al . , 2000; Vervoort et al . , 1999 ) . We used this system to investigate the impact of our mutant forms of Hh on the expression of en , ptc and dpp . We confirmed that posterior Hh overexpression expanded dpp-LacZ expression anteriorly ( Figure 8c ) and , consistent with established en >Hh expansion of the L3-L4 intervein area ( Figure 2c ) , we confirmed that posterior Hh overexpression expanded ptc-LacZ expression in the presumptive L3-L4 region ( Figure 8d ) . En-controlled expression of the HA-tagged protein at 18°C , in contrast , did not much affect dpp-LacZ expression in the anterior compartment ( Figure 8e ) , but abolished all ptc-LacZ reporter expression and restricted en-expression posteriorly ( Figure 8f ) . This confirms that the complete loss of L3-L4 intervein tissue in adult en >HAHh wings is caused by insufficient Hh levels at the a/p border , and supports the idea that coexpressed HAHh impaired Hh release from the posterior compartment of the wing disc . We note that the observed expansion of dpp expression can be best explained by abrogated Ptc-mediated Hh internalization that normally restricts the Hh gradient ( Chen and Struhl , 1996 ) . Consistent with our concept of N-palmitate serving as a membrane anchor to prevent unregulated Hh release , HAHhC85S coexpression restored ptc-LacZ and dpp-LacZ expression ( Figure 8g , h ) to levels comparable to those detected in HhC85S expressing wing discs ( Figure 8i , j ) , and additional deletion of the unpalmitoylated N-terminal peptide reverted the expanded area of dpp-LacZ expression to wild-type range ( Figure 8k , l ) . Together , these experiments confirm that N-terminal Hh processing converts the insoluble Hh cluster into truncated , bioactive morphogen , and that the palmitate anchor controls completion of this process . To confirm that impaired processing of palmitoylated Hh variants prevents their solubilization , while impaired processing of unpalmitoylated Hh N-termini affects Ptc receptor binding of soluble clusters ( merely reducing their bioactivity ) , we macroscopically analyzed wings of single and compound transgenic fly lines expressing Hh from the attP 51C landing site on chromosome 2 and HAHh or HAHhC85S from one specific attP2 landing site on chromosome 3 . As shown earlier , if expressed under en-Gal4 control , Hh and HAHhC85S strongly affected wing development: En-controlled HAHhC85S reduced the formation of L3-L4 intervein tissue ( Figure 9a ) , and en >Hh expanded the L3-L4 intervein area ( Figure 9b ) . Targeted coexpression of both proteins fully reverted dominant-negative HAHhC85S function in 80% of wings and expanded this area in the remaining 20% of wings ( 56 wings were analyzed , Figure 9c , d ) . This indicates that increased Hh amounts ‘titer out’ dominant-negative HAHhC85S function . In contrast , en >Hh did not significantly correct dominant-negative HAHh mis-patterning phenotypes . As previously shown , HAHh expression in the posterior compartment arrested fly development at pupal and pharate stages . All wings of about 4% imagos that hatched ( n = 12/320 ) lacked most anterior structures ( Figure 9d , e ) . Hh coexpression partially reversed pharate lethality ( 22% imagos hatched , n = 41/182 ) , but all analyzed wings still lacked the complete L3-L4 intervein area ( Figure 9d , f ) . The most likely explanation for this is that increased relative Hh amounts ‘dilute’ the average number of permanent membrane anchors of any given mixed cluster , with only a limited compensatory effect on Hh release and activity due to the remaining tethers . We confirmed cell-autonomous Hh repression by using another Gal4-line ( 34B-Gal4 ) that drives transgene expression in cells that form the most proximal parts of the wing where hh is normally not expressed ( Figure 9g ) ( Brand and Perrimon , 1993 ) . 34B-Gal4-controlled Hh expression in these cells results in phenotypes resembling a natural hh gain-of-function allele , hhMoonrat ( Figure 9g´ , arrow ) ( Tabata and Kornberg , 1994 ) . Phenotypes resulting from ectopic hhMoonrat expression are usually mild , varying between overgrowth of anterior wing tissue to slight disorganization of the wing margin and the addition of extra vein material . We observed that 34B > HAHhC85S and 34B > HAHh expression did not affect wing development , confirming spatial disconnection of 34B-directed transgene expression from posterior endogenous Hh production and biological inactivity of both proteins ( Figure 9h , i ) . In compound 34B > Hh;HAHhC85S wings , the activity of mixed clusters was reduced ( Figure 9j , arrowhead ) , while it was completely abolished in 34B > Hh;HAHh wings ( Figure 9k , l ) . This is expected from impaired Ptc-binding of Hh;HAHhC85S clusters in the former situation versus blocked release of Hh;HAHh clusters in the latter situation . Finally , we investigated the expression of the Hh target gene ptc in flies expressing Hh and HAHhC85S alone and in combination . As shown earlier , posterior Hh overexpression expanded ptc-LacZ expression ( compare Figure 10a with Figure 10b ) , and en-controlled expression of the HA-tagged non-palmitoylated protein strongly reduced ptc-LacZ reporter expression ( Figure 10c ) . Consistent with the restored formation of L3-L4 intervein tissue in adult en >Hh;HAHhC85S wings ( Figure 9c ) , and occasionally gain-of-function in these wings , ptc-LacZ expression in the anterior compartment of the wing disc was expanded ( Figure 10d ) . This shows that coexpressed Hh fully restored dominant-negative HAHhC85S function by expanding ptc-LacZ target gene expression in the presumptive L3-L4 region in the anterior compartment and demonstrates that receiving cells respond to the morphogen . Together , these experiments confirm a functional link between Hh lipidation , formation of linear cell surface clusters and proteolytic processing of lipidated N-terminal peptides in vivo . Processing serves to convert the lipidated morphogen cluster at the cell surface into the active form ( Figure 11 ) . Therefore , the N-palmitate membrane anchor and membrane-proximal CW-residues are functionally linked since the palmitoylation ensures quantitative CW-cleavage as a prerequisite for full Hh activation in vivo .
It is well established that cell-surface HS chains assist in Hh multimerization as a prerequisite for the generation of light microscopically visible aggregates at the cell surface ( Ortmann et al . , 2015; Vyas et al . , 2008 ) . Here , we provide ultrastructural data showing that a significant fraction of Hh assembles into extended linear arrays , consistent with the long unbranched HS-chain structure that scaffolds Hh clusters ( Vyas et al . , 2008 ) , observed crystal lattice interactions of the vertebrate Shh ortholog ( Pepinsky et al . , 1998 ) and functional in vitro data ( Ohlig et al . , 2011; Ohlig et al . , 2012 ) . By exploiting the Drosophila wing development model which is dependent on differential Hh signaling for the formation of distinct wing structures , we further show that N-terminal peptides can block Ptc-receptor-binding of Hh clusters in vivo . Consistent with this , expression of N-truncated Hh mutants in Drosophila revealed that inhibitory peptide removal unmasks Ptc-binding sites and mediates direct , high threshold tissue patterning ( Strigini and Cohen , 1997 ) . Yet , contrary to previous observations on N-truncated Shh ( Ohlig et al . , 2011 ) , we note that all artificially truncated HhC85S;Δ variants were functionally inert . We explain this inactivity by HhC85S;Δ misfolding due to possible intramolecular chaperone function of the 84 amino acid N-terminal Hh pre-peptide ( Eder and Fersht , 1995 ) or unproductive Ptc binding of artificially truncated proteins as described for monomeric ShhN ( Williams et al . , 1999 ) . We currently investigate these possibilities by insertion of a tobacco etch virus ( TEV ) protease recognition site into the putative Hh target site to allow for sequence-specific HhC85S cleavage and activation after controlled TEV protease expression in the fly ( Harder et al . , 2008 ) . We previously showed that proteolytic conversion targets the N-terminal CW-site in vitro ( Dierker et al . , 2009; Ohlig et al . , 2012 ) . We show here that insertion of HA peptides , which displaces this cleavage site distally without affecting Hh N-palmitoylation ( Hardy and Resh , 2012 ) and HS-dependent multimerization , is sufficient to impair endogenous and transgenic Hh high threshold biofunction in vivo , apparently without affecting low threshold Dpp-mediated Hh activity . Rescue of Hh biofunction by the additional mutation of the cysteine acceptor shows that N-palmitate anchors the unprocessed peptide to the cell membrane to safeguard Hh release . These findings are consistent with enhanced Drosophila Hh release upon RNAi-mediated knockdown of Hh acyltransferase activity ( Chamoun et al . , 2001 ) and increased Shh tethering to cell membranes by palmitate ( Konitsiotis et al . , 2014; Levental et al . , 2010 ) . Importantly , while S- and O-linked palmitate moieties are susceptible to enzymatic deacylation by palmitoyl-protein thioesterases ( Kakugawa et al . , 2015 ) , amide-linked Hh palmitate is thioesterase resistant . This suggests that Hh relay from posterior subcellular structures – at least at some point – requires proteolytic processing of sheddase-accessible , membrane-proximal terminal target peptides . Support for this idea comes from the published replacement of the C-terminal Hh target peptide with transmembrane-CD2 ( Strigini and Cohen , 1997 ) . Resulting Hh-CD2 fusion proteins remain permanently membrane associated and generate wings with one single central vein in the region normally occupied by veins L3 and L4 , while leaving Dpp-mediated anterior and posterior wing patterning intact . We note that this phenotype is strikingly similar to the en >HAHh phenotype described here . Moreover , required Hh transfer between protruding cell-cell contact structures emanating from the Hh-sending and Hh-receiving compartments , called cytonemes , was recently indicated by impaired Ptc signaling and internalization at contact sites with Hh-CD2 ( González-Méndez et al . , 2017 ) . While the exact mechanism by which Hh is liberated from the posterior cytoneme membrane was not addressed , proteolytic Hh relay and reception at cytoneme contact sites was suggested by the authors , and is supported by the results shown in our work ( Figure 1e ) . In addition to cytoneme contact sites , other subcellular structures of P-compartment cells release Hh from the membrane ( Figure 12 ) . It has been suggested that the Hh gradient in Drosophila wing imaginal discs consists of apical and basolateral secreted pools formed as a consequence of initial apical Hh secretion , subsequent reinternalization , and apical ( D'Angelo et al . , 2015 ) or basolateral ( Callejo et al . , 2011 ) resecretion , both depending on the endosomal sorting complex required for transport ( ESCRT ) . Pools of Hh and ESCRT proteins are then secreted together into the extracellular space ( Gradilla et al . , 2014; Matusek et al . , 2014; Vyas et al . , 2014 ) , Hh being transported on ( Bischoff et al . , 2013 ) or inside of ( Chen et al . , 2017 ) basolateral cytonemes , or apically released to promote Hh long-range activity ( Ayers et al . , 2010 ) ( Figure 12 ) . While Hh shedding may target several of these pools , timely and reliable paracrine Hh function through proteolytic release and extracellular apical diffusion alone ( Figure 12b’’ ) is difficult to envision for two reasons . First , patterning of folded epithelia , such as the Drosophila imaginal disc , poses a problem if spreading were to occur out of the plane of the epithelial cell layer through diffusion or flow , as this would result in morphogen loss into the peripodial space and loss of long-range Hh function . The second limitation is that it normally takes much time for diffusing molecules to travel long distances away from the source because the timescale of diffusion increases with the square of the distance ( Berg , 1993; Müller and Schier , 2011 ) . Cytoneme- or exosome-mediated basolateral transport , followed by proteolytic Hh relay over short distances at membrane contact sites ( González-Méndez et al . , 2017 ) , would effectively solve both problems , as would the idea of heparan sulfate proteoglycan ‘restricted’ Hh transport at the apical cell surface ( Han et al . , 2004 ) . Our future aim is to distinguish between these possibilities . We also aim to characterize the Hh release factor Shifted ( Glise et al . , 2005 ) , a soluble protein with structural similarities to vertebrate Scube2 sheddase enhancers ( Jakobs et al . , 2014; Jakobs et al . , 2016; Jakobs et al . , 2017 ) , to identify the elusive ‘Hh sheddase’ . Finally , we are currently investigating the important question of whether C-terminal Hh processing contributes to its in vivo biofunction in the wing disc and in other developing tissues requiring Hh signaling over shorter ranges , such as in the developing eye ( Ma et al . , 1993 ) . In conclusion , we propose that palmitate-controlled quantitative Hh shedding from the cell surface constitutes an essential step in Hh transmission and high-threshold tissue patterning in vivo . While we have used Drosophila wing development to elucidate this molecular process , we expect our results to also be relevant to other Hh-dependent developmental programs and to Hh ligand-dependent cancer induction and progression ( Amakye et al . , 2013 ) .
The following fly lines were used: Ptc-Gal4 ( ptc> ) : w[*]; P ( w[+mW . hs]=GawB ) ptc[559 . 1] , Bloomington stock #2017; En-Gal4e16E ( En> ) : P ( en2 . 4-GAL4 ) e16E , FlyBaseID FBrf0098595; Hh-Gal4 ( hh> ) : w[*];; P ( w[+mC]=Gal4 ) hh[Gal4] , Bloomington stock #67046; en ( 2 ) -Gal4 ( en ( 2 ) > ) : w1118;; P ( GMR94D09-Gal4 ) , Bloomington stock #48011; 34B-Gal4 ( 34B> ) : y1w[*];; P ( w[+mW . hs]=GawB ) 34B , Bloomington stock #1967 . These lines were crossed with flies homozygous for UAS-hh or variants thereof . All Hh cDNAs cloned into pUAST-attP were first expressed in Drosophila S2 cells to confirm correct protein processing and secretion . Transgenic flies were generated by using the landing site 51 C1 by BestGene or in-house by using strain PhiC31 ( X ) ; attPVK37; attP2 that possesses the landing sites VK37 and attP2 . Cassette exchange was mediated by germ-line-specific phiC31 integrase ( Bateman et al . , 2006 ) . Ptc-LacZ reporter flies were kindly provided by Jianhang Jia , Markey Cancer Center , and Department of Molecular and Cellular Biochemistry , University of Kentucky College of Medicine , Lexington , USA . Wing discs were fixed , permeabilized and stained with anti-β-galactosidase antibodies ( Cappel , MP Biomedicals ) and Cy3-conjugated goat-α-rabbit antibodies ( Jackson Immuno Research ) . Posterior , Hh-producing cells were detected with monoclonal antibodies directed against engrailed ( en 4D9 , DSHB ) and Alexa488-conjugated donkey-α-mouse antibodies ( Thermo Fisher ) . Images were taken on a LSM 700 Zeiss confocal microscope using ZEN software . Maximum intensity projections are shown . Hh cDNA ( nucleotides 1–1416 , corresponding to amino acids 1–471 of D . melanogaster Hh ) and HhN cDNA ( nucleotides 1–771 , corresponding to amino acids 1–257 ) were inserted into pENTR , sequenced , and cloned into pUAST for protein expression in S2 cells or the generation of transgenic flies . Mutations were introduced by QuickChange Lightning site-directed mutagenesis ( Stratagene ) . Primer sequences and sequence information is shown in Supplementary file 1 . S2 cells ( RRID: CVCL_Z232 ) were cultured in Schneider’s medium ( Invitrogen ) supplemented with 10% fetal calf serum ( FCS ) and 100 μg/ml penicillin/streptomycin . The cells were obtained from C . Klämbt , University of Münster , Germany , and tested negative for mycoplasma . S2 cells were transfected with constructs encoding Hh and HhN variants together with a vector encoding an actin-Gal4 driver by using Effectene ( Qiagen ) and cultured for 48 hr in Schneider’s medium before protein was harvested from the supernatant . Shh constructs were generated from murine cDNA ( NM_009170 ) by PCR ( primers are listed in Supplementary file 1 ) . Hh acyltransferase cDNA ( NM_018194 ) was obtained from ImaGenes and cloned into pIRES ( ClonTech ) for bicistronic Shh/Hh acyltransferase coexpression in the same transfected cells . This resulted in N-palmitoylated , C-cholesterylated proteins . Bosc23 cells ( RRID: CVCL_4401 ) were cultured in Dulbecco's modified Eagle’s medium ( Lonza ) supplemented with 10% FCS and 100 µg/ml penicillin-streptomycin . The cells were obtained from D . Robbins , University of Miami , USA , authenticated via by PCR-single-locus-technology ( Eurofins Forensics ) , and tested negative for mycoplasma . Bosc23 cells were transfected with PolyFect ( Quiagen ) and cultured for 48 hr , the medium was changed , and Shh was secreted into serum-free medium for the indicated times . All media were ultracentrifuged for 30 min at 125 , 000 g , and the proteins were TCA precipitated and analyzed by 15% SDS-PAGE and western blotting with polyvinylidene difluoride membranes . Blotted proteins were detected by α-HA antibodies ( mouse IgG; Sigma ) , α-Shh antibodies ( goat IgG; R and D Systems ) , or α-Hh ( rabbit IgG , Santa Cruz Biotechnology ) . Incubation with peroxidase-conjugated donkey-α-goat/rabbit/mouse IgG ( Dianova ) was followed by chemiluminescent detection ( Pierce ) . Photoshop was used to convert grayscale blots into merged RGB pictures for improved visualization of terminal peptide processing . Drosophila third-instar larvae were collected and transferred into a microcentrifuge to which 1 ml lysis buffer was added ( PBS containing 1% ( v/v ) Triton X-100 ) . Larvae were homogenized with a micropestle and the solution was cleared at 15 , 000 rpm for 15 min at 4°C . The supernatant was sterile-filtered ( 45 μm ) and transferred into a fresh microcentrifuge tube for gel filtration analysis . All processings were conducted at 4°C . Gel filtration analysis was performed on an Äkta protein purifier ( GE Healthcare ) on a Superdex200 10/300 GL column ( Pharmacia ) equilibrated with PBS at 4°C . Eluted fractions were TCA precipitated and analyzed by SDS-PAGE as described earlier . Signals were quantified by using ImageJ . Sequence analysis was conducted on the CFSSP secondary structure prediction server ( http://www . biogem . org/tool/chou-fasman/ ) . All statistical analyses were performed in GraphPad Prism by using the Student’s t test ( two-tailed , unpaired , confidence interval 95% ) . For wing quantifications , 10 male and 10 female wings were analyzed for each data set and ratios between L3-L4 intervein areas and L2-L3 intervein areas determined . All error estimates are standard errors of the mean ( SEM ) . Shh-expressing Bosc23 cells were fixed overnight at 4°C in 4% paraformaldehyde/glutaraldehyde , washed in PIPES , and dehydrated in a graded ethanol series ( 30% EtOH , 4°C , 45 min; 50% EtOH , −20°C , 1 hr; 70% EtOH , −20°C , 1 hr; 90% EtOH , −20°C , 1 . 5 hr; 100% EtOH , −20°C , 1 . 5 hr; 100% EtOH , −20°C , 1 . 5 hr ) . Dehydrated cells were embedded in Lowicryl K4M embedding medium by using the Lowicryl K4M Polar Kit ( Polysciences ) . Cells were then embedded in gelatin capsules , centrifuged twice for 15 min at 1500 rpm , and incubated overnight at −35°C . For polymerization , the resin was UV irradiated for 2 days at −35°C . The embedded samples were cut into 60 nm sections , washed in PBS containing 5% BSA ( pH 7 . 4 ) , and incubated for 2 hr in PBS-BSA containing primary antibodies ( α-Shh antibodies from R and D , GeneTex , and Cell Signaling at 1:20 dilution ) . Samples were washed five times in PBS-BSA and once in Tris-BSA . Secondary antibodies conjugated to 5 nm and 10 nm gold nanoparticles were diluted in Tris-BSA buffer and incubated with the cell sections for 1 hr . Afterwards , samples were washed five times in Tris-BSA and once in dH2O . Contrasting was done with 2% uranyl acetate ( 15 min ) and Reynold’s lead citrate ( 3 min ) . Finally , immunogold-labeled cell sections were analyzed by using a transmission electron microscope ( CM10 , Philips Electron Optics ) . The transgenic fly lines Hh-CW ( lacking the putative N-terminal Hh processing site ) , Hh-CW/HA ( a variant having this site replaced with a hemagglutinin tag ) and HhHS ( carrying a C-terminally inserted HA-tag ) generated in the course of this study that support the phenotypes described in the manuscript are available upon request from the corresponding author ( KG ) . We plan to publish these new lines separately in the future . | Each cell in a developing embryo receives information that determines what type of body structure it will form . In fruit flies , this information is partly given by a protein called Hedgehog . In the embryo cells that receive it , Hedgehog can trigger a series of events which activate certain genes and thereby regulate structure formation . The Hedgehog proteins are produced by a different organizing group of cells: from there they transport within the embryo , creating a gradient . Depending on where a responding cell is in the embryo , it receives a different amount of Hedgehog , which gives the cell its identity . For example , Hedgehog proteins form a gradient across a fruit fly’s developing wing , which creates a visible vein pattern . How Hedgehog proteins form gradients is enigmatic , however , because once produced , they cling to the cells that created them . The reason for this unusual behavior is that the two ends of the Hedgehog protein are attached to a different fat molecule . In particular , one extremity is linked to a fat molecule called palmitate . These ends’ fatty additions anchor Hedgehog to the cells that produced them . Then , the tethered proteins gather together to form chain-like clusters where they inactivate each other: the extremity with the palmitate ‘hides’ the portion of the neighboring protein that binds to the receiving cells . It is still unclear how Hedgehog can be activated and released to reach these faraway cells . One hypothesis is that an enzyme comes to the clusters and frees the proteins by cutting both of Hedgehog’s fatty anchors . Thanks to how the palmitate tethers Hedgehog to the cell , the protein is positioned in such a way that when the enzyme makes its snip , the binding site on the neighboring Hedgehog gets exposed: this protein is activated and , when also cut by the enzyme , released . Here , Schürmann et al . create an array of mutant Hedgehog proteins – for example some without palmitate , some with palmitate that cannot be removed by the enzyme – and study how they affect the development of the wing’s pattern in the fruit fly . Coupled with the imaging of the clusters , these experiments support the hypothesis that the palmitate anchor is necessary so that Hedgehog proteins can be turned on before diffusing away . The Hedgehog family of proteins is also present in humans , where it presides over the development of the embryo but is also involved in cancer . Understanding how Hedgehog works in the fruit fly could lead to new discoveries in humans too . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"developmental",
"biology",
"cell",
"biology"
] | 2018 | Proteolytic processing of palmitoylated Hedgehog peptides specifies the 3-4 intervein region of the Drosophila wing |
PP2A-B56 is a serine/threonine phosphatase complex that regulates several major mitotic processes , including sister chromatid cohesion , kinetochore-microtubule attachment and the spindle assembly checkpoint . We show here that these key functions are divided between different B56 isoforms that localise to either the centromere or kinetochore . The centromeric isoforms rely on a specific interaction with Sgo2 , whereas the kinetochore isoforms bind preferentially to BubR1 and other proteins containing an LxxIxE motif . In addition to these selective binding partners , Sgo1 helps to anchor PP2A-B56 at both locations: it collaborates with BubR1 to maintain B56 at the kinetochore and it helps to preserve the Sgo2/B56 complex at the centromere . A series of chimaeras were generated to map the critical region in B56 down to a small C-terminal loop that regulates the key interactions and defines B56 localisation . Together , this study describes how different PP2A-B56 complexes utilise isoform-specific interactions to control distinct processes during mitosis .
Protein Phosphatase 2A ( PP2A ) is a major class of serine/threonine phosphatase that is composed of a catalytic ( C ) , scaffold ( A ) and regulatory ( B ) subunit . Substrate specificity is mediated by the regulatory B subunits , which can be subdivided into four structurally distinct families: B ( B55 ) , B’ ( B56 ) , B’ ( PR72 ) and B’’ ( Striatin ) ( Seshacharyulu et al . , 2013 ) . In humans , the B subunits are encoded by a total of 15 separate genes which give rise to at least 26 different transcripts and splice variants; therefore , each of the four B subfamilies are composed of multiple different isoforms ( Seshacharyulu et al . , 2013 ) . Although these isoforms are thought to have evolved to enhance PP2A specificity , there is still no direct evidence that isoforms of the same subfamily can regulate specific pathways or processes . Perhaps the best indirect evidence that they can comes from the observation that B56 isoforms localise differently during mitosis ( Bastos et al . , 2014; Nijenhuis et al . , 2014 ) . However , even in these cases , it is still unclear how this differential localisation is achieved or why it is needed . We addressed this problem by focussing on prometaphase , a stage in mitosis when PP2A activity is essential to regulate sister chromatid cohesion ( Kitajima et al . , 2006; Riedel et al . , 2006; Tang et al . , 2006 ) , kinetochore-microtubule attachments ( Foley et al . , 2011; Kruse et al . , 2013; Suijkerbuijk et al . , 2012; Xu et al . , 2013 ) and the spindle assembly checkpoint ( Espert et al . , 2014; Nijenhuis et al . , 2014 ) . Crucially , all of these mitotic functions are controlled by PP2A-B56 complexes that localise to either the centromere or the kinetochore . The kinetochore is a multiprotein complex that assembles on centromeres to allow their physical attachment to microtubules . This attachment process is stochastic and error-prone , and therefore it is safeguarded by two key regulatory processes: the spindle assembly checkpoint ( SAC ) and kinetochore-microtubule error-correction . The SAC preserves the mitotic state until all kinetochores have been correctly attached to microtubules , whereas the error-correction machinery removes any faulty microtubule attachments that may form . The kinase Aurora B is critical for both processes because it phosphorylates the kinetochore-microtubule interface to destabilise incorrectly attached microtubules and it reinforces the SAC , in part by antagonising Knl1-PP1 , a kinetochore phosphatase complex needed for SAC silencing ( Saurin , 2018 ) . These two principal functions of Aurora B are antagonised by PP2A-B56 , which localises to the Knl1 complex at the outer kinetochore by binding directly to BubR1 ( Foley et al . , 2011; Kruse et al . , 2013; Suijkerbuijk et al . , 2012; Xu et al . , 2013 ) . This interaction is mediated by the B56 subunit , which interacts with a phosphorylated LxxIxE motif within the kinetochore attachment regulatory domain ( KARD ) of BubR1 ( Wang et al . , 2016a; Wang et al . , 2016b ) . As well as localising to the outer kinetochore , PP2A-B56 also localises to the centromere by binding to shugoshin 1 and 2 ( Sgo1/Sgo2 ) ( Kitajima et al . , 2006; Riedel et al . , 2006; Rivera et al . , 2012; Tang et al . , 2006; Tanno et al . , 2010; Xu et al . , 2009 ) . The crystal structure of Sgo1 bound to PP2A-B56 has been solved to reveal a bipartite interaction between Sgo1 and the regulatory and catalytic subunits of the PP2A-B56 complex ( Xu et al . , 2009 ) . This interaction is thought to allow centromere-localised PP2A-B56 to counteract various kinases , such as Aurora B , which remove cohesin rings from chromosome arms during early mitosis in higher eukaryotes ( Marston , 2015 ) . The result is that cohesin is specifically preserved at the centromere where it is needed to resist the pulling forces exerted by microtubules . As well as preserving cohesion at the centromere , PP2A-B56 is also thought to balance the net level of Aurora B activation in this region ( Meppelink et al . , 2015 ) . In human cells , B56 isoforms are encoded by five separate genes ( B56α , β , γ , δ and ε ) . The interaction interfaces involved in BubR1 and Sgo1 binding are extremely well conserved between all of these B56 isoforms ( Figure 1—figure supplement 1 ) . This explains why BubR1 and Sgo1 appear to display no specificity for individual B56 isoforms ( Kitajima et al . , 2006; Kruse et al . , 2013; Riedel et al . , 2006; Xu et al . , 2013; Xu et al . , 2009 ) , and why these isoforms have been proposed to function redundantly at kinetochores during mitosis ( Foley et al . , 2011 ) . However , one crucial observation throws doubt over this issue of redundancy: individual B56 isoforms localise differentially to either the kinetochore or centromere in human cells ( Meppelink et al . , 2015; Nijenhuis et al . , 2014 ) . It is therefore not easy to reconcile this differential localisation with the evidence presented above , which implies that the centromere and kinetochore receptors for B56 do not display any selectivity for individual isoforms . This caused us to readdress the question of redundancy and isoform specificity in human cells .
PP2A-B56 isoform localisation to the centromere and kinetochore was visualised in nocodazole-arrested HeLa Flp-in cells expressing YFP-tagged B56 subunits . This revealed that while some B56 isoforms localise predominantly to the sister kinetochore pairs marked by Cenp-C ( B56γ and δ ) , others localise mainly to the centromere defined by Sgo2 ( B56α and ε ) , and one isoform displayed a mixed localisation pattern ( B56β ) ( Figure 1a , b ) . B56 isoforms have been proposed to act redundantly at the kinetochore in human cells ( Foley et al . , 2011 ) , therefore we readdressed this question in light of their differential localisation . B56α and B56γ were chosen as representative members of the centromere and kinetochore-localised pools , respectively , since these isoforms could both be readily detected by western blot analysis of HeLa cell lysates ( Figure 1—figure supplement 2 ) . Furthermore , both genes were endogenously tagged using CRISPR/Cas9-mediated homologous recombination to demonstrate consistent expression and differential localisation to either the centromere or kinetochore ( Figure 1—figure supplement 3 ) . All B56 isoforms were then depleted , except for either B56α or B56γ ( Figure 1—figure supplement 2 ) , to determine whether these endogenous isoforms could support centromere and kinetochore functions . Centromeric PP2A-B56 is important for maintaining sister chromatid cohesion during mitosis in human cells ( Marston , 2015 ) . In agreement with our differential localisation data , only the centromere-localised B56α was able to support proper centromeric cohesion ( Figure 1c ) . In fact , we observed no difference in the extent of sister chromatid splitting when comparing loss of all B56 isoforms to a situation when only B56γ is retained ( Figure 1c ) . Therefore , sister chromatid cohesion can be supported by a B56 isoform that localises primarily to the centromere ( B56α ) , but not by one that localises to the kinetochore ( B56γ ) . To examine which B56 isoforms can support kinetochore functions , we first focussed on SAC signalling . The SAC is activated at kinetochores by the phosphorylation of ‘MELT’ repeats on Knl1 by the kinase Mps1 ( London et al . , 2012; Shepperd et al . , 2012; Yamagishi et al . , 2012 ) . These phosphorylated repeats recruit a variety of SAC proteins to the kinetochore , which are then assembled into an inhibitory complex that is released into the cytosol to prevent mitotic exit ( Saurin , 2018 ) . PP2A-B56 antagonises this process , as evidenced by the fact that removal of B56 from kinetochores prevents Knl1-MELT dephosphorylation and delays mitotic exit following Mps1 inhibition in nocodazole ( Espert et al . , 2014; Nijenhuis et al . , 2014 ) . Therefore , we sought to address whether these effects were dependent on specific B56 isoforms . As expected , simultaneous depletion of all B56 isoforms enhanced basal Knl1-MELT phosphorylation in nocodazole , delayed MELT dephosphorylation upon Mps1 inhibition with AZ-3146 ( Hewitt et al . , 2010 ) , and prevented mitotic exit under identical conditions ( Figure 1d–f ) . Importantly , these effects were all rescued when endogenous B56γ was preserved , but not if only B56α remained ( Figure 1d–f ) . Kinetochore PP2A-B56 also has well-established roles in chromosome alignment where it is needed to antagonise Aurora B and allow initial kinetochore-microtubule attachment to form ( Foley et al . , 2011; Kruse et al . , 2013; Suijkerbuijk et al . , 2012; Xu et al . , 2013 ) . Knockdown of all B56 isoforms produced severe defects in chromosome alignment , as expected , and these effects could be rescued by preserving B56γ , but not B56α ( Figure 1g ) . In summary , only the kinetochore-localised B56γ , and not the centromeric B56α , can support SAC silencing and chromosome alignment in human cells . Overexpression of GFP-B56α has previously been shown to rescue kinetochore-microtubule attachment defects following the depletion of all PP2A-B56 isoforms in human cells ( Foley et al . , 2011 ) . To understand the discrepancy with our data , we performed the same assays as previously , but this time expressing siRNA-resistant YFP-B56α or YFP-B56γ to rescue the knockdown of all endogenous B56 isoforms . Under these conditions , both exogenous B56 isoforms were able to rescue MELT dephosphorylation , SAC silencing and chromosome alignment ( Figure 1—figure supplement 4 ) . The ability of exogenous YFP-B56α to support kinetochore functions can be explained by the fact that it is highly overexpressed , which leads to elevated centromere and kinetochore levels in comparison to the endogenous YFP-B56α situation ( Figure 1—figure supplement 5 ) . We therefore conclude B56α acts primarily at the centromere , but it can still function at the kinetochore when overexpressed . In summary , under endogenous conditions , PP2A-B56 isoforms localise differentially to the centromere or kinetochore where they carry out specific roles in sister chromatid cohesion , SAC silencing and chromosomal alignment . We next sought to determine the molecular explanation for differential B56 isoform localisation . This was difficult to reconcile with existing structural data mapping the interaction between PP2A-B56 and the reported kinetochore and centromere receptors - BubR1 and Sgo1 – since these demonstrate that the key interacting residues are well conserved between all B56 isoforms ( Figure 1—figure supplement 1 ) ( Wang et al . , 2016a; Wang et al . , 2016b; Xu et al . , 2009 ) . Furthermore , biochemical studies could not detect a difference in association between different B56 isoforms and either BubR1 or Sgo1 ( Kitajima et al . , 2006; Kruse et al . , 2013; Xu et al . , 2013 ) . Therefore , we decided to first test whether BubR1 or Sgo1 were the only receptors for B56 at the kinetochore and centromere . At the centromere , Sgo1 and Sgo2 can both bind to PP2A-B56 ( Rivera et al . , 2012; Tanno et al . , 2010; Xu et al . , 2009 ) , but Sgo1 is considered the primary receptor because it is more important than Sgo2 for protecting cohesion in mitosis ( Huang et al . , 2007; Kitajima et al . , 2005; Kitajima et al . , 2006; Llano et al . , 2008; McGuinness et al . , 2005; Rivera et al . , 2012; Tang et al . , 2006; Tanno et al . , 2010 ) . However , this critical role in cohesin maintenance could also be explained by PP2A- independent effects ( Hara et al . , 2014 ) . Furthermore , although Sgo1 has been implicated in PP2A-B56 recruitment to centromeres ( Liu et al . , 2013a; Liu et al . , 2013b; Nishiyama et al . , 2013; Tang et al . , 2006 ) , the only study that has directly compared the contribution of Sgo1 and Sgo2 to centromeric PP2A-B56 recruitment , has concluded that Sgo2 is more important ( Kitajima et al . , 2006 ) . We therefore set out to clarify the role of Sgo1 and Sgo2 in controlling the recruitment of B56 isoforms to the centromere in human cells . Depletion of Sgo2 , but not Sgo1 , caused a significant reduction in B56α levels at the centromere ( Figure 2a–d ) . It is important to note that the quantification in Figure 2b and d cannot distinguish between kinetochore and centromere localisation , and whilst Sgo1 depletion did not reduce B56 , it did appear to shift its localisation towards the kinetochore ( see zoom panel in Figure 2c ) , an effect that has previously been seen by others ( Meppelink et al . , 2015 ) . Line plots analysis , which can quantify localisation across the centromere-kinetochore axis , demonstrates that Sgo1 depletion caused Sgo2 and B56α to spread out from the centromere towards the kinetochore ( Figure 2e ) . This is due to inefficient anchoring of Sgo2 at centromeres because combined Sgo1 and Sgo2 depletion completely removed B56α from kinetochores and centromeres ( Figure 2f , g ) . We therefore conclude that , as suggested previously by others ( Kitajima et al . , 2006 ) , Sgo2 is the primary centromeric receptor for PP2A-B56 during mitosis . However , Sgo1 also contributes to centromeric B56 localisation primarily by helping to anchor the Sgo2-B56 complex at the centromere , perhaps by bridging an interaction with cohesin rings or by helping to preserve centromeric cohesion ( Hara et al . , 2014; Liu et al . , 2013b ) . We next examined whether specific binding to Sgo1 and/or Sgo2 could explain differential B56 isoform localisation . To address this , we artificially relocalized Sgo1 or Sgo2 to the inner kinetochore , by fusing it to the kinetochore-targeting domain of CENP-B ( CB ) . This location was chosen , even though it partially overlaps with the endogenous centromeric B56 pool , because it is still accessible to Aurora B . This may be important because phosphorylation of Sgo2 by Aurora B has been proposed to be needed for B56 interaction ( Tanno et al . , 2010 ) . Whereas CB-Sgo1 was able to localise additional B56α and B56γ to the inner kinetochore ( Figure 2—figure supplement 1 ) , CB-Sgo2 was only able to recruit additional B56α ( Figure 2h–k ) . To confirm that endogenous Sgo2 displayed selectivity for specific B56 isoforms , we used a Designed Ankyrin Repeat Protein ( DARPin ) that can bind to GFP with high affinity ( Brauchle et al . , 2014 ) . The DARPin was fused to dCas9 to enable the selective targeting of YFP-tagged B56α or B56γ to a repetitive region on chromosome 7 ( Chr7 ) . This assay confirmed that only B56α , and not B56γ , was able to co-recruit endogenous Sgo2 to this region ( Figure 2l , m ) . Considering Sgo2 is the primary centromeric receptor for B56 ( Figure 2a , b ) ( Kitajima et al . , 2006 ) , this provides an explanation for why only a subset of B56 isoforms localise to the centromere . At the kinetochore , PP2A-B56 binds to a phosphorylated LxxIxE motif in BubR1 ( Kruse et al . , 2013; Suijkerbuijk et al . , 2012; Xu et al . , 2013 ) and this interaction is mediated by a binding pocket on B56 that is completely conserved in all isoforms ( Figure 1—figure supplement 1 ) ( Hertz et al . , 2016; Wang et al . , 2016a; Wang et al . , 2016b ) . Therefore , we hypothesised that additional interactions may help to stabilise specific B56 isoforms at the kinetochore . In agreement with this hypothesis , BubR1 depletion or mutation of the LxxIxE binding pocket in B56γ ( B56γH187A ) reduced but did not completely remove B56γ from kinetochores/centromere ( Figure 3a–d ) . This is not due to knockdown efficiency or penetrance of the mutation , because residual B56 could still be detected after BubR1 depletion in B56γH187A cells ( Figure 3—figure supplement 1a , b ) . Interestingly , the remaining B56γ in these situations spreads out between the kinetochore and centromere ( Figure 3e , f ) , which implies that B56γ uses additional interactions to be maintained at this region . A targeted siRNA screen identified critical roles for Knl1 and Bub1 , which , when depleted , completely abolished B56γ recruitment to kinetochores ( Figure 3—figure supplement 1c–f ) . Knl1 recruits Bub1 to kinetochores , and Bub1 scaffolds the recruitment of BubR1 ( Johnson et al . , 2004; Overlack et al . , 2015; Primorac et al . , 2013 ) . However , in addition to this , Bub1 also phosphorylates histone-H2A to localise Sgo1 to histone tails that are adjacent to the kinetochore ( Baron et al . , 2016; Kawashima et al . , 2010; Kitajima et al . , 2005; Liu et al . , 2013a; Tang et al . , 2004; Yamagishi et al . , 2010 ) . Since Sgo1 can bind to B56γ ( Figure 2—figure supplement 1 ) we examined its role in the kinetochore recruitment of this isoform . Sgo1 depletion reduced B56γWT at kinetochores and completely removed B56γH187A ( Figure 3g , h ) . Moreover , this was specific for Sgo1 , because Sgo2 depletion had no effect ( Figure 3—figure supplement 1g , h ) . To test whether this was due to direct binding to Sgo1 , we generated a B56γ Sgo1-binding mutant ( B56γΔSgo1 ) , which we confirmed was defective in binding CB-Sgo1 in vivo ( Figure 3—figure supplement 2 ) . This mutation reduced the recruitment of B56γWT to kinetochores and completely abolished the recruitment of B56γH187A ( Figure 3i , j ) , in a manner that was similar to the effect of Sgo1 depletion ( Figure 3g , h ) . This demonstrates that Bub1 establishes two separate arms that cooperate to recruit B56γ to kinetochores: it binds directly to BubR1 , which interacts via its LxxIxE motif with B56γ , and it phosphorylates Histone-H2A to recruit Sgo1 , which additionally helps to anchor B56γ at kinetochores . The B56-Sgo1 interaction is unlikely to explain B56 isoform specificity at kinetochores , since Sgo1 interacts with both B56α and B56γ when recruited to centromeres ( Figure 2—figure supplement 1 ) . We therefore focussed on the LxxIxE interaction with BubR1 to quantitatively assess the binding to B56α and B56γ . Immunoprecipitations of equal amounts of B56α and B56γ from nocodazole-arrested cells demonstrated that BubR1 bound preferentially to B56γ ( Figure 4a , b ) . Moreover , a panel of antibodies against other LxxIxE containing proteins ( Hertz et al . , 2016 ) , demonstrated that LxxIxE binding was generally reduced in B56α immunoprecipitates ( Figure 4a , b ) . B56γ has been shown to display slightly higher affinities for some LxxIxE containing peptides in vitro ( Wu et al . , 2017 ) , which , in principle , could allow this isoform to outcompete B56α for binding . However , a simple competition model is unlikely to explain differential kinetochore localisation , since we observe no change in B56α localisation when all other B56 isoforms are present or knocked down ( Figure 4c , d ) . Instead , we favour the hypothesis that binding to LxxIxE motifs is specifically perturbed in PP2A-B56α complexes during prometaphase . We next searched for the molecular explanation for differential B56 isoform localisation . To do this , we generated four chimaeras between B56α and B56γ by joining the isoforms in the loops that connect the α-helixes ( Figure 5a ) . Immunofluorescence analysis demonstrated that B56γ localisation switched from kinetochores to centromeres in chimaera 4 ( Figure 5b , c ) . Furthermore , this region alone ( i . e . the region that is different between chimaeras 3 and 4 ) is sufficient to switch localisation to the centromere when transferred into B56γ , and the corresponding region in B56γ can induce localisation to the kinetochore if transplanted into B56α ( Figure 5—figure supplement 1 ) . We generated four additional chimaeras to narrow down this region even further to amino acids 405–425 in B56α , which contains an α-helix and a small loop that juxtaposes the catalytic domain in the PP2A-B56γ complex ( Figure 5d–f ) ( Xu et al . , 2006 ) . Importantly , switching just four amino acids within this loop in B56α to the corresponding residues in B56γ ( B56αTKHG ) was sufficient to relocalise B56α from centromeres to kinetochores ( Figure 5g–i ) . Furthermore , the B56αTKHG remained functional and holoenzyme assembly was unperturbed ( Figure 5—figure supplement 2 ) . In summary , a small C-terminal loop in B56 defines whether B56 localises to centromeres , via Sgo2 , or to kinetochores , via an LxxIxE interaction with BubR1 . We next addressed whether the B56αTKHG mutant switched the Sgo2 and LxxIxE binding properties of B56α . In-cell interaction assays demonstrated that B56αTKHG , in contrast to B56αWT , was not efficiently recruited to the centromere by CB-Sgo2 ( Figure 6a , b ) , and was unable to co-recruit endogenous Sgo2 to the repeat region on chromosome 7 , when re-localised there using dCas9-DARPin ( Figure 6c , d ) . Furthermore , in addition to these effects on Sgo2 interaction , the YFP-B56αTKHG mutant showed an enhanced ability to bind LxxIxE containing proteins and , in particular , BubR1 , following immunoprecipitation from nocodazole-arrested cells ( Figure 6e , f ) . Therefore , we conclude that the small EPVA loop in B56α is necessary for the interaction with Sgo2 and the centromere and , in addition , it is also required to fully repress binding to LxxIxE motifs and the kinetochore . Importantly , this loop is not sufficient to induce either of these effects when transplanted alone into B56γ , because B56γEPVA is not lost from the kinetochore or gained at the centromere ( Figure 6—figure supplement 1a ) . Instead , a region immediately C-terminal to the EPVA ( amino acids 414–453 in B56α ) is also required to induce centromere binding , and a small helix N-terminal to the EPVA ( amino acids 374–386 in B56α ) is needed to repress kinetochore binding ( Figure 6—figure supplement 1b ) . Therefore , although the regions that define centromere and kinetochore localisation overlap at the EPVA loop , they have different distal requirements that demonstrates that they are not identical ( Figure 6g ) .
This work demonstrates how different B56 isoforms localise to discrete subcellular compartments to control separate processes during mitosis . Differential B56 isoform localisation has previously been observed in interphase ( McCright et al . , 1996 ) and during the later stages of mitosis ( Bastos et al . , 2014 ) , which implies that B56 isoforms may have evolved to carry out specific functions , at least in part , by targeting PP2A to distinct subcellular compartments . The differential localisation we observe during prometaphase arises because B56 isoforms display selectivity for specific receptors at the centromere and kinetochore . The centromeric isoform B56α binds preferentially to Sgo2 via a C-terminal stretch that lies between amino acids 405 and 453 ( Figure 6—figure supplement 1 ) . A key loop within this region juxtaposes the catalytic domain and contains an important EPVA signature that is critical for Sgo2 binding and is unique to B56α and B56ε . This sequence is also conserved in Xenopus B56ε , which has previously been shown to selectively bind to Sgo2 , when compared to B56γ ( Rivera et al . , 2012 ) . We therefore propose that a subset of B56 isoforms ( B56α and ε ) utilize unique motifs to interact with Sgo2 and the centromere during mitosis . How then can these results be reconciled with the fact that Sgo1 appears to be more important than Sgo2 for the maintenance of cohesion during mitosis ( Huang et al . , 2007; Kitajima et al . , 2005; Kitajima et al . , 2006; Llano et al . , 2008; McGuinness et al . , 2005; Rivera et al . , 2012; Tang et al . , 2004; Tang et al . , 2006; Tanno et al . , 2010 ) ? Firstly , it is important to note that Sgo1 can compete with the cohesin release factor , WAPL , for cohesin binding ( Hara et al . , 2014 ) , thereby protecting cohesion independently of PP2A . In addition , Sgo1 could help cells to tolerate the loss of Sgo2 , because Sgo2 depletion does not fully remove PP2A-B56 from the centromere , and the pool that remains under these conditions is dependent on Sgo1 ( Figure 2a–g ) . Therefore , the residual Sgo1-PP2A-B56α/ε that remains at centromeres following Sgo2 depletion could be sufficient to preserve cohesion . Finally , Sgo1 is needed to preserve Sgo2-PP2A-B56 at the centromere ( Figure 2e ) and it can also bind directly to the SA2–Scc1 complex ( Hara et al . , 2014; Liu et al . , 2013b; Tanno et al . , 2010 ) . Therefore , perhaps Sgo1 also helps to position Sgo2-PP2A-B56 so that it can dephosphorylate nearby residues within the cohesin complex . It will be important in future to examine the interplay between Sgo1 , Sgo2 and PP2A-B56 at centromeres . The kinetochore B56 isoforms bind to BubR1 via a canonical LxxIxE motif within the KARD ( Hertz et al . , 2016; Kruse et al . , 2013; Suijkerbuijk et al . , 2012; Xu et al . , 2013 ) . Although the LxxIxE binding pocket is completely conserved in all B56 isoforms ( Figure 1—figure supplement 1 ) , we observe a striking preference in the binding of B56γ over B56α to many LxxIxE containing proteins during prometaphase ( Figure 4 ) . We hypothesise that this is due to repressed binding between LxxIxE motifs and B56α during prometaphase , because LxxIxE binding ( Figure 6e , f ) and kinetochore accumulation ( Figure 5h , j ) can both be enhanced by mutation of the EPVA loop in B56α ( B56αTKHG ) . We cannot , however , exclude the possibility that the corresponding TKHG sequence in B56γ positively regulates LxxIxE interaction and kinetochore localisation . Considering that this region also controls Sgo2 and centromere binding , a simple explanation could be that Sgo2 interaction obscures the LxxIxE binding pocket . However , this appears unlikely for four reasons: 1 ) Sgo2 depletion does not relocalise B56α to kinetochores ( Figure 2a , b ) , 2 ) Sgo2 depletion does not enhance the ability of B56α to bind to BubR1 or other LxxIxE motifs during mitosis ( Figure 6—figure supplement 2 ) , 3 ) centromere and kinetochore binding can occur together in certain B56α-γ chimaeras ( Figure 6—figure supplement 1b ) , and 4 ) the regions that define each of these localisations do not fully overlap ( Figure 6g ) . Although we believe these results imply that Sgo2 is unlikely to block LxxIxE interaction , in vitro experiments with purified components would ultimately be needed to formally rule this out . If not Sgo2 , then what could limit the kinetochore accumulation of B56α ? We speculate that another interacting partner , or alternatively a tail region within a PP2A-B56 subunit , might obscure or modify the conformation of the LxxIxE binding pocket in PP2A-B56α complexes . An important additional finding of this work is that Sgo1 contributes to the B56γ signal observed at the kinetochore ( Figure 3g–j ) . This likely requires Sgo1 to be bound to histone tails , because it also depends on Bub1 , the kinase that phosphorylates histone H2A to recruit Sgo1 ( Baron et al . , 2016; Kawashima et al . , 2010; Kitajima et al . , 2005; Liu et al . , 2013a; Tang et al . , 2004; Yamagishi et al . , 2010 ) ( Figure 3—figure supplement 1 ) . It is not currently clear whether a Sgo1:PP2A-B56γ complex simply contributes to the signal observed at kinetochores or whether it may help to physically retain BubR1:PP2A-B56 at the kinetochore , for example , by directly interacting with the BubR1:PP2A-B56 complex . The interfaces between BubR1-B56 and Sgo1-PP2A-B56 do not appear to be overlapping , at least based on current structural data ( Wang et al . , 2016a; Wang et al . , 2016b; Xu et al . , 2009 ) , which implies that Knl1-bound BubR1-B56 could potentially be anchored towards histone tails by Sgo1 . We were unable to detect Sgo1 in YFP-BubR1 immunoprecipitates ( results not shown ) , however , this could simply reflect an interaction that is either transient or unstable away from kinetochores . It will be important in future to clarify exactly how Sgo1 collaborates with BubR1 to control B56 localisation and , in particular , to determine whether Sgo1 can interact with BubR1:PP2A-B56 complexes directly . If such a complex can exist , then this could have important implications for SAC signalling and tension-sensing . In summary , the work presented here explains how different members of the PP2A-B56 family function during the same stage of mitosis to control different biological processes . This is the first time that such sub-functionalisation has been demonstrated between isoforms of the same B family . It is currently unclear why such specialisation is necessary or at least preferable to a situation whereby all B56 isoforms operate redundantly , as initially suggested ( Foley et al . , 2011 ) . One possibility is that the use of different B56 isoforms allows PP2A catalytic activity to be regulated differently in specific subcellular compartments: for example , by enabling interactions or post-translational modifications that are specific for the B56 subunits . In this respect , protein inhibitors of PP2A-B56 have been shown to function specifically at the centromere ( SET ( Chambon et al . , 2013 ) ) and at the kinetochore ( BOD1 ( Porter et al . , 2013 ) ) ; therefore , it would be interesting to test whether these inhibitors display selectivity for certain PP2A-B56 isoforms . Future studies such as this , which build upon the work presented here , may ultimately help to reveal novel ways to modulate the activity of specific PP2A-B56 complexes . The recent development of selective inhibitors of related PP1 regulatory isoforms to combat neurodegenerative diseases ( Das et al . , 2015; Krzyzosiak et al . , 2018 ) , provides a proof-of-concept that successful targeting of specific serine/threonine phosphatase isoforms is both achievable and therapeutically valuable .
HeLa Flp-in cells ( Tighe et al . , 2008 ) , stably expressing a TetR , were authenticated by STR profiling ( Eurofins ) and cultured in DMEM supplemented with 9% tetracycline-free FBS , 50 μg/mL penicillin/streptomycin and 2 mM L-glutamine . All cell lines were routinely screened ( every 4–8 weeks ) to ensure they were free from mycoplasma contamination . All HeLa Flp-in cells stably expressing a doxycycline-inducible construct were derived from the HeLa Flp-in cell line by transfection with the pCDNA5/FRT/TO vector ( Invitrogen ) and the FLP recombinase , pOG44 ( Invitrogen ) , and cultured in the same medium but containing 200 μg/mL hygromycin-B . Plasmids were transfected using Fugene HD ( Promega ) according to manufacturer’s protocol . 1 µg/mL doxycycline was added for ≥16 hr to induce protein expression in the inducible cell lines . Thymidine ( 2 mM ) and nocodazole ( 3 . 3 µM ) were purchased from Millipore , MG132 ( 10 µM ) and AZ-3146 from Selleck Chemicals , doxycycline ( 1 µg/mL ) from Sigma , 4 , 6- diamidino-2-phenylindole ( DAPI , 1:50000 ) from Invitrogen , calyculin A ( 10 µM in 10% EtOH ) from LC labs , RO-3306 ( 10 µM ) from Tocris and hygromycin-B from Santa Cruz Biotechnology . pCDNA5-YFP -B56α , β , γ1 , γ3 , δ and ε were amplified from pCEP-4xHA-B56 ( Addgene plasmids 14532–14537; deposited by D . Virshup , Duke-NUS Graduate Medical School , Singapore ) and subcloned into pCDNA5-LAP-BubR1WT ( Nijenhuis et al . , 2014 ) through Not1 and Apa1 restriction sites . B56γ1 and B56γ3 were corrected to start on M1 and not 11 , and the R494L mutation in B56γ3 was corrected . pCDNA5-YFP-B56α and pCDNA5-YFP-B56γ1 were made siRNA-resistant by site-directed mutagenesis ( silent mutations in the coding sequence for E102 and L103 in B56α , and T126 and L127 in B56γ ) . All B56α and B56γ1 mutants were created by site-directed mutagenesis from pCDNA5-YFP-B56α and pCDNA5-YFP-B56γ1 , respectively . The B56α–γ chimeras were generated by Gibson assembly with pCDNA5-YFP-B56α and pCDNA5-YFP-B56γ used as templates for the PCR reaction . vsv-CENP-B-Sgo1-mCherry ( Meppelink et al . , 2015 ) was used to make vsv-CENP-B-Sgo2-mCherry , by removing Sgo1 and adding Sgo2 via Gibson assembly from pDONR-Sgo2 ( a gift from T . J . Yen ) . The Sgo1 binding mutant in B56γ ( B56γ ΔSgo1 ) was created by site directed mutagenesis to create three mutations: Y391F , L394S and M398Q . The dCas9-DARPIN-flag was created by digesting pHAGE-TO-dCas9-3xmCherry ( Addgene #64108 ) with BamHI and XhoI to remove 3xmCherry and replace with a synthesised DARPIN-flag that binds to GFP with high affinity ( Brauchle et al . , 2014 ) . The gRNA targeting a repetitive region on chromosome seven was generated by PCR mutagenesis to introduce the gRNA sequence ( GCTCTTATGGTGAGAGTGT ( Chen et al . , 2016 ) ) into the pU6 vector . Cells were transfected with 20 nM siRNA using Lipofectamine RNAiMAX Transfection Reagent ( Life Technologies ) according to the manufacturer’s instructions . For simultaneous knockdown of all B56 isoforms ( B56pool ) the single B56 isoform siRNA were mixed at equimolar ratio of 20 nM each . The siRNA sequences used in this study are as follows: B56α ( PPP2R5A ) , 5’-UGAAUGAACUGGUUGAGUA-3’; B56β ( PPP2R5B ) , 5’-GAACAAUGAGUAUAUCCUA-3’; B56γ ( PPP2R5C ) , 5’-GGAAGAUGAACCAACGUUA-3’; B56δ ( PPP2R5D ) , 5’-UGACUGAGCCGGUAAUUGU-3’; B56ε ( PPP2R5E ) , 5’-GCACAGCUGGCAUAUUGUA-3’; Sgo1 , 5’-GAUGACAGCUCCAGAAAUU-3’; Sgo2 , 5’-GCACUACCACUUUGAAUAA-3’; BubR1 , 5’-AGAUCCUGGCUAACUGUUC-3’; Knl1 , 5’-GCAUGUAUCUCUUAAGGAA-3’; Bub1 5’-GAAUGUAAGCGUUCACGAA-3’; Control ( GAPDH ) , 5’-GUCAACGGAUUUGGUCGUA-3’; . All siRNA oligos were custom made and purchased from Sigma , except for Sgo1 , which was ordered from Dharmacon ( J-015475–12 ) . For reconstitution of B56 isoforms or mutants , HeLa Flp-in cells were transfected with 100 nM B56pool or mock siRNA and , in some experiments , 20 nM additional control , Sgo1 , Sgo2 , BubR1 , Bub1 or Knl1 siRNA . Cells were transfected with the appropriate siRNA for 16 hr , after which they were arrested in S phase for 24 hr by addition of thymidine . Subsequently , cells were released from thymidine for 8–10 hr and arrested in prometaphase by the addition of nocodazole . YFP-B56 expression was induced by the addition of doxycycline during and following the thymidine block . For BubR1 knockdowns and for all chromosome alignment assays , cells were released from thymidine for 6 . 5 hr and arrested at the G2/M boundary with RO3306 for 2 hr . Cells were then released into nocodazole ( BubR1 experiments ) or normal growth media ( alignment assays ) for 15 mins before MG132 was then added for 30 mins to prevent mitotic exit . For alignment assays , this is critical to analyse the synchronous alignment of mitotic cells over a 45 min period . Hela-FRT cells were transfected with B56pool , B56βγδε , B56αγδε or control siRNA for 16 hr , treated with thymidine for 24 hr and released into normal growth media for 6 . 5 hr . Cells were then arrested at the G2/M boundary with RO3306 for 2 hr before release into nocodazole for 1 hr . Mitotic cells were isolated and incubated with hypotonic buffer ( 20 mM Hepes ( pH7 . 0 ) , 1 mM MgCl2 , 20 mM KCl , 2 mM CaCl2 ) for 10 min at room temperature before being spun onto slides using a Cellspin cytocentrifuge ( Tharmac ) . Slides were airdried for 1 min and then fixed in 4% formaldehyde in PBS for 10 min at room temperature . Blocking and immunofluorescence staining ( for Cenp-C to visualise split kinetochore pairs ) was carried out as described below . The percentage of cells with at least one split sister kinetochore pair was quantified . Cells were transfected with dCas9-DARPIN-flag and a guide RNA that targets a repetitive region on chromosome 7 ( at 1:3 ratio of dCas9:gRNA ) . Doxycycline was added to induce YFP-B56 isoform expression and 48 hr later cells arrested in mitosis with nocodazole were fixed , stained and imaged for co-localisation of YFP-B56 isoforms and Sgo2 . Only cells containing defined Flag-dCas9 spots that also co-recruited YFP-B56 were imaged . The majority of these spots recruited YFP-B56 , but the dCas9 spots themselves were only readily detectable in mitotic cells . For the CB-Sgo1/2 expression experiments , the endogenous Sgo1/2 was still present during these assays . 800 base pair homology arms that span left and right of the start codon of B56α and B56γ were custom synthetized by Biomatik . A NaeI ( B56γ ) /SwaI ( B56α ) restriction site was place between the homology arms and used to insert a YFP tag by Gibson assembly . Guides were designed to span the start codon ( using http://crispr . mit . edu/ ) so that their complementary sequences are interrupted following successful homologous recombination . Flp-in HeLa Cas9 cells were generated and transfected with the YFP-homology arm vector and guide RNAs ( B56α: gatgtcgtcgtcgtcgccgccgg B56γ: gtcaacatctagacttcagcggg ) in a 1:1 ratio . Cas9 expression was then induced by addition of doxycycline and FACS was performed 2 weeks later to sort cells and enrich for the YFP-expressing population . For time-lapse analysis , cells were plated in 24-well plates , transfected and imaged in a heated chamber ( 37°C and 5% CO2 ) using a 10x/0 . 5 NA on a Zeiss Axiovert 200M Imaging system , controlled by Micro-manager software ( open source: https://www . micro-manager . org/ ) . Images were acquired with a Hamamatsu ORCA-ER camera every 4 min using 2 × 2 binning . For immunofluorescence , cells were plated on High Precision 1 . 5H 12 mm coverslips ( Marienfeld ) . Following the appropriate treatment , cells were pre-extracted with 0 . 1% Triton X-100 in PEM ( 100 mM Pipes , pH 6 . 8 , 1 mM MgCl2 and 5 mM EGTA ) for 1 min followed by addition of 4% PFA/PBS for 2 min; cells were subsequently fixed with 4% paraformaldehyde in PBS for 10 min . Coverslips were washed with PBS and blocked with 3% BSA in PBS + 0 . 5% Triton X-100 for 30 min , incubated with primary antibodies for 16 hr at 4°C , washed three times with PBS and incubated with secondary antibodies plus DAPI for an additional 2–4 hr at room temperature in the dark . Washed coverslips were then mounted on a glass slide using ProLong antifade reagent ( Molecular Probes ) . All images were acquired on a DeltaVision Core or Elite system equipped with a heated 37°C chamber , with a 100x/1 . 40 NA U Plan S Apochromat objective using softWoRx software ( Applied precision ) . Images were acquired at 1 × 1 binning using a CoolSNAP HQ2 camera ( Photometrics ) and processed using softWorx software and ImageJ ( National Institutes of Health ) . All images displayed are maximum intensity projections of deconvolved stacks . All displayed immunofluorescence images were chosen to most closely represent the mean quantified data . For kinetochore quantification of immunostainings , all images within an experiment were acquired with identical illumination settings and analysed using ImageJ ( for experiments in which ectopic proteins were expressed , cells with comparable levels of exogenous protein were selected for analysis ) . Kinetochore quantification was performed as previously ( Saurin et al . , 2011 ) . For quantification of B56 localization , The Cenp-C channel was used to choose 5 random kinetochore pairs per cell that lie on the same 0 . 2 μm Z-plane . A line was then drawn through the kinetochore pairs ( using ImageJ ) , with the first Cenp-C kinetochore peak at 0 . 2 µm from the start of the line . An ImageJ macro ( created by Kees Straatman , University of Leicester and modified by Balaji Ramalingam , University of Dundee ) was used to simultaneously measure the intensities in each channel across the line . The signal from the five kinetochore pairs was averaged and normalized to the maximum signal in each channel . For chromosome alignment assays , misalignments were score as mild ( 1 to 2 misaligned chromosomes ) , intermediate ( 3 to 5 misaligned chromosomes ) , and severe ( >5 misaligned chromosomes ) . For mitotic exit assays , time from entry into mitosis ( defined by the rounding up of the cell ) to mitotic exit ( defined by the separation of the sister chromatids or flattening down of the cell in nocodazole +AZ-3146 ) were recorded for 50 cells . Data is presented as cumulative percentage of mitotic exit over time . Flp-in HeLa cells were treated with thymidine and doxycycline for 24 hr and subsequently released into fresh media supplemented with doxycycline and nocodazole for 16 hr . Mitotic cells were isolated by mitotic shake off and lysed in lysis buffer ( 50 mM Tris , pH 7 . 5 , 150 mM NaCl , 0 . 5% TX-100 , 1 mM Na3VO4 , 5 mM ß-glycerophosphate , 25 mM NaF , 10 nM Calyculin A and complete protease inhibitor containing EDTA ( Roche ) ) on ice . The lysate was incubated with GFP-Trap magnetic beads ( from ChromoTek ) for 2 hr at 4°C on a rotating wheel in wash buffer ( same as lysis Buffer , but without TX-100 ) at a 3:2 ratio of wash buffer:lysate . The beads were washed 3x with wash buffer and the sample was eluted according to the protocol from ChromoTek . Samples were them processed for SDS-Page and immunoblotting using standard protocols . For quantification of relative immunoprecipitation levels , scanned immunoblots were analyzed using Image Studio Lite ( LI-COR Bioscences ) . A rectangle of the same size was drawn around each band and the intensity within the band ( minus the background ) was calculated . The immunoprecipitated protein was used as a control , and each band was normalized to it . All antibodies were diluted in 3% BSA in PBS . The following primary antibodies were used for immunofluorescence imaging ( at the final concentration indicated ) : mouse α-GFP ( clone 4E12/8 , a gift from P . Parker; 1:1000 ) , chicken α-GFP ( ab13970 , Abcam; 1:5000 ) , mouse α- Sgo1 ( clone 3C11 , H00151648-M01 , Abnova; 1:1000 ) , rabbit α-Sgo2 ( A301-262A , Bethyl; 1:1000 ) , mouse α-BubR1 ( clone 8G1 , 05–898 , Upstate/Millipore; 1:1000 ) , mouse α-VSV ( clone P5D4 , V5507 , Sigma; 1:1000 ) , rabbit α-Knl1 ( ab70537 , Abcam; 1:1000 ) , rabbit α-Bub1 ( A300-373A , Bethyl; 1:1000 ) , mouse α-FLAG ( clone M2 , F3165 , Sigma , 1:10000 ) guinea pig α-Cenp-C ( PD030 , MBL; 1:5000 ) and rabbit α-pMELT-Knl1 directed against T943 and T1155 of human Knl1 ( Nijenhuis et al . , 2014 ) , 1:1000 ) . Secondary antibodies used were highly-cross absorbed goat α-rabbit , α-mouse , α-guinea pig or α-chicken coupled to Alexa Fluor 488 , Alexa Fluor 568 , or Alexa Fluor 647 ( Life Technologies ) ; all were used at 1:1000 . The following antibodies were used for western blotting ( at the final concentration indicated ) : rabbit α-GFP ( custom polyclonal , a gift from G . Kops; 1:5000 ) , mouse α-B56γ ( clone A-11 , sc-374379 , Santa Cruz Biotechnology; 1:1000 ) , mouse α-B56α ( clone 23 , 610615 , BD; 1:1000 ) , mouse α-B56δ ( clone H-11 , sc-271363 , Santa Cruz , 1:1000 ) , rabbit α-B56ε ( ARP56694-P050 , Aviva , 1:1000 ) , mouse α-PPP2CA ( clone 1D6 , 05–421 , Millipore; 1:5000 ) and rabbit α-PPP2R1A ( clone 81G5 , #2041 , CST; 1:1000 ) , rabbit α-BubR1 ( A300-386A , Bethyl; 1:1000 ) , rabbit α-Axin ( C76H11 , CST; 1:1000 ) , rabbit α-GEF-H1 ( 155785 , Abcam; 1:1000 ) , rabbit α-Kif4a ( A301-074A , Bethyl; 1:1000 ) , rabbit α-RepoMan ( HPA030049 , Sigma; 1:1000 ) and rabbit α-Actin ( A2066 , Sigma; 1:5000 ) and mouse α-alpha-Tubulin ( clone B-5-1-2 , T5168 , Sigma , 1:5000 ) . Secondary antibodies used were goat α-mouse IgG HRP conjugate ( Bio-Rad; 1:2000 ) and goat α-rabbit IgG HRP conjugate ( Bio-Rad; 1:5000 ) . Mann-Whitney U test was performed to compare experimental groups in all kinetochore/centromere quantification graphs , whereas two-tailed , unpaired t-test with Welch’s correction was performed to compare experimental groups in all other graphs ( using Prism seven software ) . The n numbers for kinetochore/centromere quantification statistics were derived from the individual cells ( i . e . biological replicates ) , which were always from at least three separate experiments ( i . e . technical replicates ) with similar results . The n numbers for the statistics in all other graphs were defined by the number of experimental repeats . The SD bars displayed in each graph shows the variation between the means of the experimental repeats . The statistical comparisons most pertinent for the conclusions are shown in the figures and legends . The original data for all experiments displayed in graphs can be found in the raw data source file , which also contains the actual statistical values . | The cells in our body are a hive of activity , but that activity must be kept under control . This is never more critical than when a cell divides , because unchecked cell division can lead to cancer . Fortunately , enzymes called kinases and phosphatases exist to control the countless proteins in a cell; these enzymes help ensure that each step of cell division is complete before moving on to the next . Kinases control other proteins by adding bulky phosphate groups to them , while phosphatases remove those groups . For a long time , phosphatases were assumed to be less specific than their kinase counterparts . Yet it has now become clear that phosphatases achieve specificity by interacting with a range of regulatory subunits . A phosphatase called PP2A oversees a number of key steps in cell division by working together with its regulatory B56 subunit . In human cells , there are five separate genes that encode B56 subunits , and all of these B56 ‘isoforms’ were thought to exert the same influence on the PP2A phosphatase . The fact , however , that different isoforms are found at different locations within the cell suggested otherwise . To investigate this , Vallardi et al . focused on a particular stage of cell division when the activity of the PP2A-B56 complex is essential . Before a cell divides it duplicates its genetic material and the two copies of each chromosome are held together until the cell is ready to pull them apart . The experiments compared two representative B56 isoforms: one that concentrates at the centromere , the region where the copied chromosomes are held together; and another found at the kinetochore , a nearby structure that is involved in pulling the two chromosomes apart . By eliminating all but one isoform and measuring the ensuing activity of the PP2A-B56 complex , Vallardi et al . could differentiate between the main regulatory roles of each isoform . These experiments showed that B56 isoforms control separate processes during cell division , which mirrors their different locations within the cell . Next , Vallardi et al . looked at the receptor proteins that recruit each isoform to its position . Removing or relocating different receptors showed how they anchor select B56 isoforms in different positions while the associated PP2A enzymes get to work on different processes . Further experiments using ‘hybrid’ subunits made from parts of two different B56 isoforms then helped to reveal the site on the B56 subunits that determines which receptors they bind to . Together these findings show that slight differences between each B56 isoform ultimately dictate where they localise and what processes they control when cells divide . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"cell",
"biology"
] | 2019 | Division of labour between PP2A-B56 isoforms at the centromere and kinetochore |
Visual perception across a broad range of light levels is shaped by interactions between rod- and cone-mediated signals . Because responses of retinal ganglion cells , the output cells of the retina , depend on signals from both rod and cone photoreceptors , interactions occurring in retinal circuits provide an opportunity to link the mechanistic operation of parallel pathways and perception . Here we show that rod- and cone-mediated responses interact nonlinearly to control the responses of primate retinal ganglion cells; these nonlinear interactions , surprisingly , were asymmetric , with rod responses strongly suppressing subsequent cone responses but not vice-versa . Human psychophysical experiments revealed a similar perceptual asymmetry . Nonlinear interactions in the retinal output cells were well-predicted by linear summation of kinetically-distinct rod- and cone-mediated signals followed by a synaptic nonlinearity . These experiments thus reveal how a simple mechanism controlling interactions between parallel pathways shapes circuit output and perception .
Perceptual interactions between rod- and cone-mediated signals have been studied for nearly 200 years; these interactions influence the spatial , temporal and chromatic sensitivities of human vision at light levels ranging from moonlight to dawn or dusk ( for review see Buck , 2004 , 2014 ) . Despite the impact of rod-cone interactions on vision , little is known about the mechanistic basis of such interactions . Our aims here were to relate perceptual rod-cone interactions to retinal mechanisms and by doing so to provide an example of how the mechanisms controlling parallel processing in neural circuits impact computation and human perception .
We first compared rod-cone interactions in the output signals of the primate retina with those observed perceptually ( Figure 1 ) . We focused on nonlinear interactions between brief increment flashes that preferentially elicited responses from ON retinal circuits; ganglion cell spike outputs in response to these flashes were dominated by excitatory synaptic input ( Figure 1—figure supplement 1 ) , further simplifying the circuitry involved . ON parasol ganglion cells ( which project to magnocellular layers of the lateral geniculate nucleus ) exhibited particularly robust responses to such flashes . We used dim short-wavelength flashes to preferentially activate rod photoreceptors , and brighter long-wavelength flashes to preferentially activate L-cone photoreceptors ( Figure 1A , B; Figure 1—figure supplement 2 ) . 10 . 7554/eLife . 08033 . 003Figure 1 . Asymmetric nonlinear rod-cone interactions . ( A ) Diagram of the retinal circuits that convey rod- and cone-generated signals to the brain . Dim short-wavelength light preferentially activates rod photoreceptors whereas long-wavelength light preferentially activates long wavelength ( L ) cone photoreceptors ( see Figure 1—figure supplement 2 ) . ( B ) Protocol for testing for nonlinear rod-cone interactions electrophysiologically . Neural responses in ON parasol ganglion cells to rod ( top ) and cone ( middle ) flashes were recorded separately , and the sum of these responses ( bottom , gray ) was compared to trials in which the rod and cone flashes were delivered together ( bottom , black ) . ( C ) Average spike responses from an example cell comparing rod → cone ( top ) and cone → rod ( bottom ) interactions . ( D ) Summary data across cells comparing rod → cone and cone → rod interaction indices for spikes and excitatory postsynaptic currents ( EPSCs ) . ( E ) Average spike responses from an example cell comparing rod → cone interactions for short ( 0 . 2 s , top ) and long ( 0 . 8 s , bottom ) time offsets . ( F ) Summary data across cells: short intervals were 0 . 2–0 . 3 s and long intervals were ≥0 . 8 s . ( G ) Protocol used to test for nonlinear rod-cone interactions in human perception ( also see ‘Materials and methods’ and Figure 1—figure supplement 3 ) . Observers fixated on a cross while rod and cone flash sequences ( similar to electrophysiology stimuli ) were delivered to their peripheral retina ( ∼10° eccentricity ) . Observers compared the perceived brightness of two test flashes , one in the same location as the initial flash and the other spatially offset . ( H ) Comparison of perceptual rod → cone and cone → rod interactions for 8 human observers . ( I ) Comparison of perceptual rod → cone interactions for short and long time offsets . Each marker in panels D and F represents the average interaction for a single ganglion cell . Each marker in H and I represents the average monocular pathway-specific interaction for a single observer ( see ‘Materials and methods’ and Figure 1—figure supplement 3 ) . All retinal recordings ( B–F ) from whole mount retina . DOI: http://dx . doi . org/10 . 7554/eLife . 08033 . 00310 . 7554/eLife . 08033 . 004Figure 1—figure supplement 1 . Rod → cone interactions in RGC spike output are largely dependent on interactions present in the RGC's excitatory synaptic inputs . ( A ) Single cell example of rod → cone interactions in spikes ( left ) , excitatory synaptic input ( middle ) and inhibitory synaptic input ( right ) . ( B ) Interaction indices ( II ) for excitatory inputs resembled those observed in spikes . ( C ) Interactions in inhibitory inputs were highly variable and did not match spike interactions observed in the same cells . ( D ) Example cell from dynamic-clamp experiments to test contributions of inhibitory synaptic input to rod → cone interactions in spike output . Spike responses elicited by excitatory and inhibitory inputs ( left ) were compared to those elicited by excitatory inputs alone ( right ) . ( E ) Population data plotting interaction indices from dynamic-clamp experiments in D . All recordings from whole mount retina . DOI: http://dx . doi . org/10 . 7554/eLife . 08033 . 00410 . 7554/eLife . 08033 . 005Figure 1—figure supplement 2 . Rod → cone interactions and photoreceptor selectivity depend on mean luminance . ( A ) Photoisomerization calibration values for flashes used in this study . ( left ) Emission spectra of the blue and red LEDs and absorption spectra for the rod and L-cone photoreceptors in primate ( Baylor et al . , 1984; Baylor et al . , 1987 ) ( gray ) . ( right ) Emission spectra of the filtered blue and red phosphors used in perceptual experiments . ( B ) ( left ) Spike responses of an ON parasol ganglion cell to short- and long-wavelength flashes in the dark . Under these conditions , the response to the long-wavelength flash has two kinetically-distinct components that correspond to activation of cones ( fast initial component ) and rods ( prolonged component ) ( Dunn et al . , 2007; Schwartz et al . , 2013 ) . Both components were suppressed by a prior short-wavelength flash . ( middle ) Increasing the mean luminance reduces the gain of the rods and the rod bipolar pathway . Under these conditions the long-wavelength flash elicited a clear cone response , however the rod component was suppressed at 3 R*/rod/s and absent at 30 R*/rod/s . Nonlinear rod-cone interactions were clear at these backgrounds . ( right ) At a mean luminance of 300 R*/rod/s ganglion cell responses were fast and transient regardless of which LED was used; retinal interactions between blue and red flashes were largely absent under these conditions . All recordings from whole mount retina . DOI: http://dx . doi . org/10 . 7554/eLife . 08033 . 00510 . 7554/eLife . 08033 . 006Figure 1—figure supplement 3 . Design of psychophysical experiments presented in Figure 1 . ( A ) Assuming that retinal and cortical gains are multiplicative , retinal interactions were isolated by comparing the brightness matches from trials in which the adapt and test flashes were delivered to the same eye ( e . g . , blue and red stimuli delivered to eye #2 ) to those obtained from trials in which flashes were delivered to separate eyes ( e . g . , blue stimulus delivered to eye #1 and red stimuli to eye #2 ) . ( B ) Two example trials from a naive observer's session ( Rod-Cone ( short ) Task v2: same eye trial in black , separate eye trial in red ) . Observers adjusted the intensity of the bottom test flash ( ‘y-axis’; flash position spatially offset from that of adapt flash ) to match their perception of the top test flash ( same location as adapt flash ) . After completing a total of 6 crossings , users were automatically advanced to the next trial . When test and adapt flashes were delivered to the same eye , interactions were stronger than when flashes were delivered to separate eyes . ( C ) Individual sessions for each observer . ( top ) Comparison of rod → cone ( short ) and cone → rod ( short ) sessions . ( bottom ) Comparison of rod → cone ( short ) and rod → cone ( long ) sessions . ( D ) Mean interactions ( Figure 1F , I ) split by task version: Task v1 ( squares ) and Task v2 ( circles ) . ( top ) Comparison of mean rod → cone ( short ) and cone → rod ( short ) interactions . ( bottom ) Comparison of mean rod → cone ( short ) and rod → cone ( long ) interactions . Each marker represents average data from a single observer . DOI: http://dx . doi . org/10 . 7554/eLife . 08033 . 006 The linear sum of ON parasol responses to rod- and cone-preferring flashes delivered individually differed substantially from the response to the flashes delivered together ( Gouras and Link , 1966 ) ( Figure 1B ) —that is , rod- and cone-mediated signals interact nonlinearly within the retina . We quantified the strength of the nonlinear interactions using an interaction index ( see ‘Materials and methods’ ) ; an interaction index of 0 corresponds to linear summation of the responses , whereas an interaction index of 1 indicates that the ‘adapt’ flash ( i . e . , the first flash in the sequence ) completely suppressed the response to the ‘test’ flash ( i . e . , the second flash in the sequence ) . Switching the identity of the adapt and test flashes revealed a surprising asymmetry in these interactions: cone responses produced a modest suppression of subsequent rod responses ( cone → rod interaction ) , while rod responses produced a strong suppression of subsequent cone responses ( rod → cone interaction; Figure 1C , D ) . This asymmetry held for rod- and cone-preferring adapt flashes that produced similar amplitude responses in the RGC ( as in Figure 1C ) . Rod → cone interactions were restricted to a <1 s time window following the adapt flash ( Figure 1E , F ) . ON midget ganglion cells , another prominent retinal ganglion cell type in primate retina , exhibited qualitatively similar rod-cone interactions ( data not shown ) . Do asymmetric rod-cone interactions occur perceptually ? To answer this question we asked human observers to match the perceived brightness of two test flashes—one at the same location as the adapt flash and the other displaced spatially ( Figure 1G ) . Comparing matches from trials in which the adapt flash was located in the same or the opposite eye as the test flashes allowed us to separate interactions that likely originated in the retina ( monocular-specific interactions ) from those that likely originated in the cortex ( binocular interactions; see ‘Materials and methods’ and Figure 1—figure supplement 3 ) . Monocular-specific interactions shared several features with those observed in the responses of ON parasol ganglion cells , although several issues , including uncertainty about how retinal and cortical interactions combine , precluded quantitative comparison: ( 1 ) rod → cone interactions were stronger than cone → rod interactions ( Figure 1H ) ; and , ( 2 ) rod → cone interactions were stronger for 0 . 2–0 . 3 s intervals between adapt and test flashes than for 1 s intervals ( Figure 1I ) . These similarities suggest that retinal rod-cone interactions contribute substantially to perceptual interactions; this similarity motivated investigation of where and how the retinal interactions occur . Previous studies provide clear , testable predictions for where rod → cone interactions might originate ( Kolb , 1977 , 1979; Tsukamoto et al . , 2001; Field et al . , 2009 ) . Rod-mediated signals can traverse the retina through the dedicated rod bipolar circuitry and/or through cones via rod-cone gap junctions and then through the associated cone bipolar circuitry ( DeVries and Baylor , 1995; Schneeweis and Schnapf , 1995 ) ( Figure 1A ) . To determine which route dominates under our experimental conditions , we compared RGC sensitivity to rod- and cone-preferring flashes with the sensitivity of L-cone synaptic terminals and horizontal cells . We adjusted the strengths of rod and cone flashes until they elicited similar amplitude responses in ON RGCs ( with resulting flash strengths within a factor of two of those used for probing rod-cone interactions in Figure 1 ) . Then , in the same retinal slice , we measured voltage responses from either L-cone synaptic terminals ( Figure 2B ) or horizontal cells ( Figure 2C ) to the same flashes; horizontal cells receive direct synaptic input from cones and hence provide a direct measure of cone synaptic output ( Trumpler et al . , 2008 ) ( Figure 2A ) . The ratios of the amplitudes of the rod-mediated to cone-mediated responses were ∼8 times smaller in L-cone synaptic terminals and horizontal cells than in ON RGCs ( Figure 2D ) . Thus , at these light levels rod-mediated signals appear to be routed through the dedicated rod bipolar circuitry until they reach the inner retina . Independent pharmacological manipulation of the rod and cone bipolar circuits provided further support for this conclusion ( Figure 2—figure supplement 1 ) . This situation differs considerably from mouse retina , where rod signals are transmitted in parallel through rod and cone ( via rod-cone gap junctions ) bipolar circuits at these light levels ( Grimes et al . , 2014; Ke et al . , 2014 ) . 10 . 7554/eLife . 08033 . 007Figure 2 . Rod and cone-mediated signals are largely independent upstream of the cone bipolar → RGC synapse . ( A ) Retinal circuit diagram . ( B–C ) Rod and cone flashes that produced near-equal amplitude responses in RGCs elicited strong cone responses and weak rod responses in the axon terminals of L-cones ( B ) and in horizontal cells ( C ) . ( D ) Summary across cells . Ratios of the responses to rod- and cone-preferring stimuli in horizontal ( H ) and L-cone terminals plotted vs the response ratios in RGCs to the same flashes from the same slices . ( E ) Rod → cone interactions were absent in circuit elements upstream of the RGC: ( left ) horizontal cells , ( middle ) AII amacrine cells and ( right ) cone bipolar cells . ( F ) Mean rod → cone interaction indices for all horizontal , AII amacrine , cone bipolar and retinal ganglion cells ( mean ± SEM , number of cells in parenthesis ) . Interaction indices for horizontal cells ( p = 0 . 49 ) , AIIs ( p = 0 . 45 ) and cone bipolars ( p = 0 . 49 ) did not differ significantly from 0 , while those of RGCs did ( p < 10−7 ) . All recordings except the RGCs reported in F ( whole mount ) from retinal slices . DOI: http://dx . doi . org/10 . 7554/eLife . 08033 . 00710 . 7554/eLife . 08033 . 008Figure 2—figure supplement 1 . Pharmacological evidence that rod- and cone-mediated signals traverse the retina through largely distinct circuitry . ( A ) Circuit diagram illustrating the logic for using NBQX , an AMPAR antagonist . ( B ) Rod- and cone-preferring stimuli were adjusted to produce clear responses in AII amacrine cells . Bath application of NBQX selectively eliminated input from the rod bipolar pathway . ( C ) Population data for experiments in B; response ratios in NBQX are plotted vs response ratios in control conditions . If the majority of the rod-mediated signal traversed the retina through cone bipolar cells , the rod-mediated response should be largely insensitive to NBQX and the data would lie near the dashed line . Instead , the greater sensitivity of the rod-mediated response to NBQX indicates that most of this response is conveyed through the rod bipolar cell . Each round markers represents an AII amacrine cell recorded in the current-clamp configuration ( with corresponding K-based internal solution—see ‘Materials and methods’ ) whereas each square marker represents an AII amacrine cell recorded in the voltage-clamp configuration ( with corresponding Cs-based internal solution—see ‘Materials and methods’ ) . Similar results were obtained regardless of recording configuration . All voltage-clamp recordings from retinal slices , and current-clamp recordings ( including B ) from whole mount retina . DOI: http://dx . doi . org/10 . 7554/eLife . 08033 . 008 Although rod-cone coupling appears to make a relatively modest contribution to the strength of the rod responses observed in the retinal output , such coupling could still account for rod → cone interactions—for example , if rod-mediated signals alter the gain of the cone output synapse . This did not appear to be the case: under conditions in which the RGC excitatory synaptic inputs exhibited strongly nonlinear responses , horizontal cells , AII amacrine cells , and cone bipolar cells all combined rod- and cone-mediated responses linearly—that is , the linear sum of the responses delivered individually matched the response to the flashes delivered sequentially ( Figure 2E ) . The absence of nonlinear rod → cone interactions in circuit components upstream of the synapse between ON cone bipolar cells and RGCs , and presence of such interactions in the RGC excitatory synaptic inputs ( Figure 2F ) , indicates that such interactions must occur at the cone bipolar output synapse itself . Because this synapse is shared by both the rod and cone bipolar circuits , it could function as a gate keeper , controlling which types of photoreceptor signals are transmitted to an ON parasol RGC . What mechanisms mediate rod-cone interactions at the cone bipolar → RGC synapse ? One possibility is that transmission of the response to the adapt flash depletes the pool of presynaptic vesicles , resulting in a depressed response to the test flash . This mechanism , however , cannot easily account for the greater strength of rod → cone interactions compared to cone → rod interactions ( Figure 1D , E ) since similar synaptic activation , regardless of origin , would be expected to produce similar levels of vesicle depletion . An alternative , but not exclusive , hypothesis is that kinetically-distinct rod- and cone-mediated signals sum linearly before passing through a common synaptic nonlinearity ( i . e . , the cone bipolar → ganglion cell synapse ) . Consistent with this hypothesis , the kinetics of the rod- and cone-mediated voltage responses of ON cone bipolar cells differed substantially ( Figure 3A; similar differences were observed in AII amacrine responses ) : rod-mediated responses were biphasic , consisting of an initial depolarization ( rightward blue arrow in Figure 3B ) followed by a slow hyperpolarization ( i . e . , overshoot; leftward blue arrow in Figure 3B ) ; the latter component depended on inhibition within the rod bipolar circuit ( Figure 3—figure supplement 1 ) . Cone-mediated responses , in comparison , were faster and substantially less biphasic ( red arrows in Figure 3B ) . In this case , even if rod- and cone-mediated responses sum prior to the nonlinearity , the hyperpolarization associated with the rod-mediated response could suppress the ability of a subsequent cone-mediated response to traverse the nonlinear synapse and produce a response in the ganglion cell . The smaller hyperpolarization of the cone-mediated response , however , would suppress a subsequent rod-mediated response less and hence result in a smaller nonlinear interaction . 10 . 7554/eLife . 08033 . 009Figure 3 . Rod → cone interactions exhibit a similar time course as the observed overshoot in cone bipolar cell voltage following a rod flash . ( A ) Cone bipolar voltage responses to rod and cone flashes exhibited distinct kinetics . ( B ) Illustration of impact of kinetic differences on cone bipolar synaptic output . Rightward blue and red arrows indicate depolarizing components of rod- and cone-mediated responses , while leftward arrows indicate hyperpolarizing components . ( C ) Time course of rod → cone interactions probed by delivering paired flashes across a range of temporal offsets . ( D ) Summary of time dependence of rod → cone interactions ( individual cells in gray , population averages binned by Δt in black ) . Similar to the overshoot of the rod-mediated response in cone bipolar cells ( A ) , rod → cone interactions were maximal when flashes were offset by 0 . 1–0 . 3 s . On cone bipolar cell recordings from retinal slices , and On parasol recordings from whole mount retina . DOI: http://dx . doi . org/10 . 7554/eLife . 08033 . 00910 . 7554/eLife . 08033 . 010Figure 3—figure supplement 1 . Inhibition creates a rapid overshoot in rod-mediated signals at mean light levels where both rods and cones are active . ( A ) top Rod-mediated responses in AII amacrine cells were monophonic in darkness and became biphasic as mean luminance was increased to 20 R*/rod/s . bottom Inhibitory blockers significantly reduced the fast ( ∼200 ms after the flash ) component of the overshoot , likely reflecting suppression of presynaptic inhibition to the rod bipolar cell terminal . After returning to control solution for >8 min the fast overshoot recovered . ( B ) Population data . The inhibitory cocktail ( INH , green ) contained 2 μM Strychnine ( Str ) , 25 μM SR95531 ( SR ) and 50 μM TPMPA ( T ) to block glycine , GABAa- and GABAc- receptors , respectively . Number of recordings are indicated in parenthesis . Bars represent the mean biphasic index across recordings and error bars indicate the SEM . ( C ) Population data for pharmacological effects on the biphasic index . Bars represent the mean percentage of the biphasic index remaining in drug across cells; error bars indicate the SEM . Paired , two-tailed Student's t-tests were used to test significance; * and ** correspond to p-values less that 0 . 05 and 0 . 01 , respectively . Current-clamp recordings from AII amacrine cells in whole mount retina . DOI: http://dx . doi . org/10 . 7554/eLife . 08033 . 010 The mechanistic hypothesis illustrated in Figure 3B predicts that rod → cone suppression should be maximal when a presynaptic test signal arrives during the peak of the presynaptic hyperpolarizing overshoot associated with the adapt flash response . To test this prediction , we examined rod → cone interactions present in the excitatory synaptic inputs to ON parasols across a range of time offsets ( Figure 3C ) . Indeed , rod → cone suppression exhibited a time course consistent with that of the hyperpolarization in the rod-mediated ON cone bipolar voltage response ( Figure 3A , C , D ) . Across cells , rod → cone suppression was maximal for time offsets between 0 . 1 and 0 . 3 s , with little or no suppression for longer or shorter offsets ( Figure 3D ) . The behavior at time offsets <0 . 1 s is consistent with previous work showing that responses to simultaneously delivered rod and cone stimuli can be described by linear summation followed by saturation for strong stimuli ( Enroth-Cugell et al . , 1977; Cao et al . , 2010 ) . Can linear summation followed by a synaptic nonlinearity quantitatively account for the measured rod → cone and cone → rod interactions ? To answer this question , we characterized the relation between a time-varying stimulus ( input ) and the excitatory ganglion cell response ( output ) using a linear-nonlinear cascade model ( LN model; see ‘Materials and methods’ ) , and then used this description to generate a parameter-free model for rod → cone and cone → rod flash interactions ( Figure 4B–E ) . We derived the LN model components ( i . e . , the linear filter and static nonlinearity ) for both rod and cone inputs using gaussian noise stimuli . Linear filters for rod inputs were slower and more biphasic than those for cone inputs ( Figure 4B ) , consistent with the differences observed in ON cone bipolar voltage responses ( Figure 3A ) . The nonlinearities derived from rod and cone inputs were similar ( Figure 4C ) , consistent with a location in a shared element in the rod and cone circuits . 10 . 7554/eLife . 08033 . 011Figure 4 . Linear summation followed by a rectifying nonlinearity can account for rod-cone interactions . ( A ) Linear-nonlinear ( LN ) model construction . A time-varying rod- or cone-preferring stimulus and the resulting RGC excitatory synaptic inputs were used to derive the linear filter and static nonlinearity that relate the stimulus to the response . ( B ) Normalized linear filters for rod and cone stimuli . ( C ) Nonlinearities for rod and cone stimuli . ( D ) Rod-cone interactions were predicted by summing scaled and temporally-offset ( i . e . , 0 . 2 s ) rod- and cone-preferring filters and passing the result through a common nonlinearity ( see ‘Materials and methods’ ) . ( E ) Predicted rod → cone ( left ) and cone → rod ( right ) interactions . ( F ) Measured rod → cone and cone → rod interactions . ( G ) Population data comparing predictions from the LN models to experimental observations . Each point represents a single cell in which LN model components and rod-cone interactions were measured . All recordings from whole mount retina . DOI: http://dx . doi . org/10 . 7554/eLife . 08033 . 011 LN model components—the separate linear filters for rod and cone responses and a single common nonlinearity—were used to predict rod → cone and cone → rod flash interactions . First , we predicted responses to rod and cone flashes delivered independently . For example , for the rod flash response we scaled the amplitude of the rod linear filter so that its output , passed through the nonlinearity , matched the amplitude of the measured rod flash response . We followed the same procedure to scale the cone linear filter to match the measured cone flash response . We then predicted interactions between these flashes by offsetting the scaled linear filters in time to represent the time offset between flashes , linearly summing the two scaled filters , and passing the result through the common nonlinearity ( Figure 4D ) . Rod-cone interactions and LN model components were measured in the same cell so that the predicted ( Figure 4E ) and measured ( Figure 4F ) interactions could be directly compared . Across cells , the predicted and measured interaction indices were strongly correlated ( Figure 4G ) . Both the predicted and measured interaction indices scaled with the degree of rectification in the nonlinearity as measured by the LN model ( data not shown ) . The greater strength of rod → cone compared to cone → rod interactions in the LN model predictions depends directly on the rod linear filter being more biphasic than the cone filter; specifically , ∼200 ms after the flash , the overshoot of the rod filter is maximally effective at suppressing the ability of the cone-mediated response to traverse the nonlinearity . Similarly , predicted interactions peaked for time offsets near 200 ms , as observed experimentally ( Figure 3D ) . Thus a simple model based on linear summation of kinetically-distinct rod and cone responses followed by a rectifying nonlinearity can quantitatively predict the strength of interactions between rod- and cone-mediated flash responses , including the asymmetry in relative interaction strength .
The results here bear on the mechanistic basis of rod-cone interactions and more generally on parallel processing and computation in neural circuits . Our work leads to two broad conclusions . First , perceptual measures of rod-cone interactions and their dependence on light level are often interpreted in the context of the different routes that rod-mediated signals could take through the retina ( reviewed in Buck , 2004 , 2014 ) . Surprisingly , we found that the rod bipolar circuit remains the dominant route through which rod-mediated signals traverse the retina at light levels well above cone threshold . This lack of mixing of rod- and cone-mediated signals both constrains potential sites of interaction and provides ample opportunities for selective processing , for example to shape the kinetics of rod-mediated signals distinctly from cone-mediated signals . Second , perceptual rod-cone interactions can be complex—for example , the asymmetric nonlinear interactions we investigate here . We show that these apparently complex interactions can be explained by a simple neural mechanism: linear summation of kinetically-distinct outputs from parallel circuits followed by a shared synaptic nonlinearity . This simple picture provides a key step towards understanding the mechanistic basis of the diverse rod-cone interactions that influence so many aspects of visual perception .
Experiments were conducted on whole mount or slice ( 200 μm thick ) preparations of primate retina as previously described ( Dunn et al . , 2007; Trong and Rieke , 2008 ) . In brief , pieces of retina attached to the pigment epithelium were stored in ∼32–34°C oxygenated ( 95% O2/5% CO2 ) Ames medium ( Sigma , St Louis , MO ) and dark-adapted for >1 hr . Pieces of retina were then isolated from the pigment epithelium under infrared illumination and either flattened onto polyL-lysine slides ( whole mount ) or embedded in agarose and sliced . Once under the microscope , tissue was perfused with Ames medium at a rate of ∼8 ml/min . Non-human primate retina was obtained through the Tissue Distribution Program of the Regional Primate Research Center at the University of Washington . Extracellular recordings from ON parasol retinal ganglion cells used ∼3 MΩ electrodes containing Ames medium . Voltage-clamp whole-cell recordings were conducted with electrodes ( RGC: 2–3 MΩ , AII: 5–6 MΩ ) containing ( in mM ) : 105 Cs methanesulfonate , 10 TEA-Cl , 20 HEPES , 10 EGTA , 2 QX-314 , 5 Mg-ATP , 0 . 5 Tris-GTP and 0 . 1 Alexa ( 488 , 555 or 750 ) hydrazide ( ∼280 mOsm; pH ∼7 . 3 with CsOH ) . Current-clamp whole-cell recordings were conducted with electrodes ( AII: 5–6 MΩ , CB: 8–12 MΩ , HC: 5–6 MΩ , L-cone terminals: 10–12 MΩ ) containing ( in mM ) : 123 K-aspartate , 10 KCl , 10 HEPES , 1 MgCl2 , 1 CaCl2 , 2 EGTA , 4 Mg-ATP , 0 . 5 Tris-GTP and 0 . 1 Alexa ( 488 , 555 or 750 ) hydrazide ( ∼280 mOsm; pH ∼7 . 2 with KOH ) . Cone terminals were identified by following axons from cone cell bodies to the margin of the outer plexiform layer . In initial experiments , cell types were confirmed by fluorescence imaging following recording . NBQX ( 10 μM; Tocris ) was added to the perfusion solution as indicated in Figure 2—figure supplement 1 . To isolate excitatory or inhibitory synaptic input , cells were held at the estimated reversal potential for inhibitory or excitatory input of ∼−60 mV and ∼+10 mV . These voltages were adjusted for each cell to maximize isolation . Dynamic clamp procedures followed those described previously ( Murphy and Rieke , 2006 ) . Absolute voltage values have not been corrected for liquid junction potentials ( K+-based = −10 . 8 mV; Cs+-based = −8 . 5 mV ) . The dichoptic apparatus consisted of two 60 Hz LCD computer monitors ( 1920 × 1200 Dell , model U2412M ) controlled by a Mac mini computer running Psychtoolbox for Matlab ( Brainard , 1997; Pelli , 1997 ) . The observer's left and right eyes viewed separate monitors . NDF0 . 6 , ‘Bright pink’ , and ‘Scarlet’ gel filters ( Rosco E-colour , Stamford , CT ) were mounted to the front of each monitor to control luminance and suppress the transmission of wavelengths between 500–600 nm , thus improving photoreceptor selectivity for the red and blue phosphors . The apparatus was uniquely aligned for each observer's session . Monitor translation ( coarse adjustments ) and image translation ( fine adjustments ) were made sequentially under the direction of the observer to maximize overlap of the images in the two eyes and invoke binocular fusion . After acceptable fusion was achieved , observers were dark-adapted for 10–15 min before beginning the tasks . Observers occasionally lost fusion within a session; in these cases the session was exited and the alignment procedure was repeated before returning to the task . Two versions of the matching task produced qualitatively similar results . In the first version ( v1 ) , observers repeatedly adjusted the brightness of the bottom test flash and then manually indicated a perceptual match . In the second version ( v2 ) , observers adjusted the intensity of the bottom test flash until they had reversed the direction of their adjustments 6 times , at which point they were automatically advanced to the next trial ( Figure 1—figure supplement 3 ) . In both versions of the task the adjustment increments were reduced by 1/3 after each crossing until reaching the minimum permissible intensity adjustment set by the discrete monitor intensity values . For each session , observers ran 10–16 trials under each condition ( e . g . , rod → cone [short] , rod → cone [long] , cone → rod [short] ) ; monocular and binocular trials were randomly interleaved . Matches were averaged across trials within each session before calculating the interaction index . In Task v1 , we excluded data from trials in which observers achieved fewer than 2 crossings and from any conditions with 3 or fewer measurements/trials . In Task v2 , trial matches were calculated by averaging the midpoints between contiguous pairs of crossing values , with each midpoint weighted by the inverse of the step size between that pair of crossings . In both tasks , retinal interactions were inferred by comparing the perceptual matches obtained when adapt and test flashes were delivered to the same vs separate eyes . Interactions arising from separate eye delivery must occur in higher brain regions with access to information from both eyes ( e . g . , cortical interactions ) whereas interactions arising from same eye delivery would also include retinal interactions . We assumed that retinal and cortical gain factors were multiplicative and occurred in series . Thus suppressive interactions attributable to the retina were taken as the ratio of the perceptual sensitivities for 1 vs 2 eye delivery . Interaction indices were calculated for each session before averaging across sessions . The strength of perceptual interactions attributable to the retina were defined using an interaction index analogous to that used for the physiological experiments:II=1−S1eyeS2eye , where S1eye is the sensitivity ratio for single eye delivery ( i . e . , intensity ratio of the perceptual match ) and S2eye is the sensitivity ratio for separate eye delivery . This index is 0 if interactions for one and two eye stimuli have identical strengths and 1 if the adapt flash completely suppresses perception of the test flash for one but not two eye stimuli . Each observer conducted at least 1 training session before data was included for analysis . These training sessions helped observers ( 1 ) become comfortable with binocular fusion of images from distinct monitors , ( 2 ) improve their fixation while flashes are delivered to the peripheral retina and ( 3 ) become comfortable with the user interface ( e . g . , entering responses , transitions between conditions ) . Visual perception experiments were conducted according the human subject guidelines laid forth by the University of Washington . For electrophysiology experiments , full field illumination ( diameter: 500–560 μm ) was delivered to the preparation through a customized condenser from blue ( peak power at 460 nm ) or red ( peak power at 640 nm ) LEDs ( Figure 1—figure supplement 2 ) . Light intensities ( photons/μm2/s ) were converted to photoisomerization rates ( R*/photoreceptor/s ) using the estimated collecting area of rods and cones ( 1 and 0 . 37 μm2 , respectively ) , the stimulus ( i . e . , LED or monitor ) emission spectra and the photoreceptor absorption spectra ( Baylor et al . , 1984 , 1987 ) . In Figures 1 , 3 the blue LED provided a constant illumination of ∼20 R*/rod/s . In Figure 4 the blue and red LEDs produced a mean of ∼20 R*/rod/s and ∼200 R*/L-cone/s . Rod- and cone- preferring flashes were 10 ms in duration . For visual perception experiments , red ( peak power at 640 nm ) and blue ( peak power at 444 nm ) spots of ∼2° were presented for 16 ms at ∼10° eccentricity in human observers . The filtered backlight of the monitor produced a mean luminance of ∼1 R*/rod/s . Flash isomerization estimates for all experiments are presented in Figure 1—figure supplement 2 . We used a linear-nonlinear ( LN ) model to test whether summation of rod- and cone-mediated responses followed by a shared synaptic nonlinearity could account for rod-cone interactions . The linear filter ( L ) and nonlinearity ( N ) were estimated using measured excitatory synaptic inputs to On parasol ganglion cells in response to 50% contrast Gaussian noise ( 0–60 Hz bandwidth ) . Stimulus wavelength and intensity were chosen to emphasize rod- or cone-mediated responses ( see Figure 1—figure supplement 2 ) . Model components were estimated using standard approaches ( Rieke , 1997; Chichilnisky , 2001; see Figure 4A ) . In brief , correlating the response with the stimulus estimated the linear filter ( Figure 4B ) . Comparing the filter output ( i . e . , the stimulus convolved with the filter ) with the measured response provided an estimate of the nonlinearity ( Figure 4C ) . Ganglion cell responses to flashes were then predicted by scaling the linear filter according to the flash strength and passing the scaled filter through the nonlinearity ( Figure 4D ) . Nonlinear interactions cause the neural response to the test flash to differ in the presence or absence of the adapt flash . We quantified these interactions in our electrophysiology data by using an interaction index , defined as:II=1−RpairRsingle , where Rsingle is the integral of the response to the test flash delivered alone and Rpair is the integral of the response to the test flash when preceded by the adapt flash ( i . e . , the response to the paired flashes minus the response to the adapt flash alone ) . Rod signals recorded in AII amacrine cells were biphasic under the luminance conditions tested here ( i . e . , ∼20 R*/rod/s; Figure 3—figure supplement 1 ) . We quantified this property using a biphasic index , defined as:BI=AHAH+AD , where AD is the maximum amplitude of the depolarization and AH is the maximum amplitude of the hyperpolarization in the first 400 ms of the baseline corrected response . Electrophysiology example traces presented throughout the figures represent the average of 5–20 raw responses to the same stimuli . All data are presented as mean ± SEM and two-tailed paired Student's t-tests were used to test significance . | The inner surface at the back of the eye is called the retina and contains two types of light-sensitive cells: rod cells and cone cells . Rods outnumber cones by roughly twenty to one and are responsible for vision under low light levels . Cone cells , by contrast , provide detailed vision in bright light , as well as the ability to see in color . Rods and cones provide input to two distinct networks of cells that convey information in parallel to cells called ganglion cells , which then relay this information out of the retina . However , the signals from activated rods can feed into the cone pathway at several points , meaning that the responses of rods and cones are not independent . At dawn and dusk—and indeed under street lighting at night—rods and cones are both active and interactions between rod and cone responses influence many aspects of vision , including sensitivity to color and contrast . Grimes et al . have now identified a neural mechanism behind these interactions by combining measurements of human vision with recordings of electrical activity in retinas from non-human primates . The experiments confirmed that activating either type of photoreceptor briefly suppresses the responses of the other , although unexpectedly rods inhibit cones more than cones inhibit rods . The site of this interaction is the connection—or synapse—between the very last cell in the cone pathway and the retinal output cells . Prior to this ‘gateway’ synapse , rod and cone-mediated responses are largely independent . Vision at dawn and dusk is shaped by a complex set of interactions between rod and cone signals—such as the ability of activated rods to change color perception at dusk . These findings show that these seemingly complex behaviors can arise from simple interactions at the level of neural circuits . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"short",
"report",
"neuroscience"
] | 2015 | A simple retinal mechanism contributes to perceptual interactions between rod- and cone-mediated responses in primates |
Ecological variation influences the appearance and maintenance of tool use in animals , either due to necessity or opportunity , but little is known about the relative importance of these two factors . Here , we combined long-term behavioural data on feeding and travelling with six years of field experiments in a wild chimpanzee community . In the experiments , subjects engaged with natural logs , which contained energetically valuable honey that was only accessible through tool use . Engagement with the experiment was highest after periods of low fruit availability involving more travel between food patches , while instances of actual tool-using were significantly influenced by prior travel effort only . Additionally , combining data from the main chimpanzee study communities across Africa supported this result , insofar as groups with larger travel efforts had larger tool repertoires . Travel thus appears to foster tool use in wild chimpanzees and may also have been a driving force in early hominin technological evolution .
What evolutionary pressures have favoured tool use in some species , including chimpanzees and humans , but not others ? Recent work in non-human primate species has focussed on the role of ecological variables for the emergence of tool use ( Fox et al . , 1999; Humle and Matsuzawa , 2002; Möbius et al . , 2008; Spagnoletti et al . , 2012; Sanz and Morgan , 2013 ) . These studies have enlightened our understanding of how ecology influences animal culture ( Whiten et al . , 1999; Laland and Janik , 2006 ) and are also informative for modelling early hominin lifestyle ( Susman and Hart , 2015 ) . For non-human primates , Fox and colleagues proposed three hypotheses to test the relation between ecological factors and the innovation of feeding-related tool use in primates , i . e . , the ‘invention’ , ‘necessity’ , and ‘opportunity’ hypotheses ( Fox et al . , 1999 ) . While the invention hypothesis states that new forms of tool use are rare chance events , which spread through social learning ( Fox et al . , 2004 ) , the necessity and opportunity hypotheses state that ecological factors can have an additional influence ( Sanz and Morgan , 2013 ) . While the necessity hypothesis predicts that tool use emerges as a response to food scarcity , the opportunity hypothesis predicts this emergence as a response to encounters with either the material needed to make a tool or the resources to be extracted by the tools ( Koops et al . , 2013 ) . The current literature has generated conflicting and inconclusive results concerning the different ecological hypotheses , even within the same species ( chimpanzees ( Pan troglodytes verus; necessity: Yamakoshi , 1998; opportunity: Koops et al . , 2013; inconclusive: Furuichi et al . , 2015 ) ; capuchins ( Sapajus spp . , necessity: Moura and Lee , 2004; opportunity: Spagnoletti et al . , 2012 ) ; bonobos ( Pan paniscus; inconclusive: Furuichi et al . , 2015 ) ; see ( Sanz and Morgan , 2013 ) for a review ) . Research on non-primates has generated an additional ecological hypothesis , the ‘relative profitability hypothesis’ to explain the emergence of tool use , which is based on optimal foraging theory and work on New Caledonian crows ( Corvus moneduloides ) . This hypothesis states that tool use can develop as a strategy to obtain dietary components difficult to obtain without tools , but only if this is more profitable than non-tool-based strategies and as long as the ecological conditions , such as low predation pressure , allow it ( Rutz and St Clair , 2012 ) . Koops et al . ( 2014 ) proposed an enlarged opportunity hypothesis , which includes not only ecological but also social and cognitive opportunities as drivers of tool use innovation and maintenance . In this view , necessity cannot explain tool use in animals because of the lack of correlations between selected environmental indicators and tool use . In particular , in a study with unhabituated chimpanzees of Nimba forest , Guinea , there was no correlation between fruit availability and ant remains in faeces , a proxy for stick use ( Koops et al . , 2013 ) . Additionally , there was no relation between feeding-related tool use variants and the number of dry months across chimpanzee sites through Africa , further suggesting that tool use did not emerge out of necessity ( Koops et al . , 2014 ) . In contrast , support for the necessity hypothesis comes from another study with the nearby habituated chimpanzees of Bossou , Guinea , where nut-cracking increased when fruit availability was low , suggesting that tool use is a fall back strategy during periods of food scarcity ( Yamakoshi , 1998 ) . One explanation for these conflicting results is that the necessity hypothesis is difficult to test . For instance , Sanz and Morgan ( 2013 ) argue that the abundance of preferred food is a poor proxy for necessity and that even low levels of these foods may not be sufficient to trigger significant behavioural changes . Second , necessity may be driven by the lack of particular micronutrients essential for survival but that do not account for a major part of the diet ( Sanz and Morgan , 2013 ) . Necessity-based tool inventions , in other words , may not always function to compensate for low caloric intake . A third problem with the necessity hypothesis may also be due to the narrow focus of the analyses conducted to test it , e . g . feeding opportunities determined through phenological surveys , with no data on ( a ) whether animals actually seize these opportunities , ( b ) their variation across large timescales ( Gruber et al . , 2012a ) , ( c ) the energetic costs incurred to benefit from them ( Pontzer and Wrangham , 2004; Lehmann et al . , 2007; Amsler , 2010 ) , and ( d ) the differential needs of individuals across time . In this respect , while analysing entire communities or populations can be useful , for instance by correlating phenological variables with tool use frequencies or tool repertoire sizes of entire communities ( see below ) , individual needs may differ substantially within groups , suggesting that additional levels of analysis may be necessary to test the necessity hypothesis . In this study , we were interested in the role of ecological factors in the emergence of chimpanzee tool use at the individual level . We studied how individuals of a chimpanzee community known for its limited tool use behaviour , the Sonso community of Budongo Forest ( Pan troglodytes schweinfurthii ) , behaved in an experimental foraging task that required tool use . Although Sonso chimpanzees use tools in non-feeding contexts , such as for personal hygiene or communication , they have only been observed to use one type of tool to access resources , which consists of folding and chewing a handful of leaves to make a sponge , usually to collect water ( Reynolds , 2005 ) . Recently , some members of the Sonso community have learned a new technique , moss-sponging , to access mineral-rich suspensions from a clay pit ( Hobaiter et al . , 2014 ) . The Sonso community is part of a larger population of about 700 chimpanzees living in Budongo Forest , which most likely show the same limitations in tool use behaviour ( Gruber et al . , 2012a ) . For this reason , they constitute ideal subjects to study the emergence of new tool use behaviours , unlike other populations that already have complex food-related tool repertoires ( Whiten et al . , 1999 ) . We analysed data from a long-term field experiment , the honey-trap experiment , in which subjects were exposed to a novel foraging task that could only be solved with a tool . In doing so , we controlled for opportunity-based ecological aspects by presenting subjects with a standardised apparatus , which consisted of a small cavity drilled into a portable log , filled with liquid honey ( Gruber et al . , 2009 , 2011 ) . Our goal was to test individuals under conditions of high ecological validity , over an extended period of time ( 2009–2015 ) , with an unprecedented subject pool of over 50 individuals of a fully habituated community . In contrast to previous studies , our experimental approach allowed us to carry out analyses at the individual level , by comparing individuals in their interactions with the apparatus ( Gruber et al . , 2009 , 2011 ) . In our previous work , we found that 10 of 52 individuals ( 19 . 2% ) who engaged with the apparatus proceeded to manufacture a leaf-sponge to extract artificially provided honey ( Gruber et al . , 2009 , 2011; Gruber , 2016 ) . This behaviour is customarily used by wild chimpanzees to drink water , but there are no reports of chimpanzees using this behaviour to collect naturally available honey from bee nests . During our experiments , we also recorded two individuals using a stick to access the honey , but only after much exposure and experimental facilitation ( see Material & Methods ) , and in contrast to another Ugandan chimpanzee community , where stick use was customary to access experimentally provided honey ( Gruber et al . , 2009 , 2011 ) . In the current study , we combined our long-term experimental data and behavioural observations to determine the natural parameters that influenced individual variation in engagement with the apparatus and the use of tools . As our experimental design controlled for opportunity , we were able to assess the influence of two key necessity-related variables , feeding time spent on ripe fruits ( a proxy for food availability ) and travel effort ( a proxy for energetic demands ) , measured as the proportion of travel in the activity budget , on individuals’ ( a ) engagement time with the apparatus and ( b ) probability of tool use . As we had no specific predictions concerning the relevant time intervals , we carried out these analyses incorporating data from different time periods prior to interaction with the apparatus . Second , to determine whether any eventual patterns characterised chimpanzees as a species , we ran a cross-population comparison of travel behaviour and fruit feeding in relation to differences in food-related tool repertoires comparing data from all long-term chimpanzee communities . Finally , we discuss how our findings can shed light on the different hypotheses outlined above , and how they can contribute to a unifying model of the emergence of tool use .
We analysed a total of 292 experimental trials ( N = 52 subjects , mean/median number of trials per individual: 5 . 6/3 . 0 , range: 1–39 ) . Mean engagement time with the apparatus was 111 s ( N = 292 trials , range: 1–1275 s ) . In 21 of these trials ( 7 . 2% ) , subjects also used a tool . These cases were distributed over 16 different experimental days ( 11 with a single tool-user , five with two successive tool-users ) . For each trial , we determined the preceding travel and ripe fruit feeding behaviour of the subject by systematically varying the time periods before each experiment ( ranging from 1 to 13 weeks ) . To this end , we determined the proportion of all scans that contained travel and ripe fruit feeding for the focal individual of the test subject’s party . This is a reasonable approach since members of a chimpanzee party typically engage in the same behaviour at a given time ( see Material and methods ) . The first model assessed how a subject’s engagement time with the apparatus was related to ripe fruit feeding , travel time and time period . This model was significant overall ( linear mixed-effects model , likelihood ratio test ( LRT ) : X2 = 188 . 1 , df = 10 , p<0 . 0001 , R2m = 0 . 33 , Table 1 ) , with a significant three-way interaction between ripe fruit feeding , travel time and time period ( LRT: X2 = 5 . 77 , df = 1 , p = 0 . 0163 , Figure 1A ) . Specifically , when subjects fed little on ripe fruits , they engaged more with the apparatus , provided they also travelled much . This effect was modulated by the duration the subject was recorded in the same condition . For example , chimpanzees engaged more with the apparatus if they had travelled more and had consumed less ripe fruits for longer than shorter periods of time ( Figure 1A , lower panel ) . However , when subjects spent much time feeding on ripe fruits , there was less variation in time spent engaging with the apparatus , regardless of prior travel time . In addition , older individuals and males engaged less with the apparatus than young individuals and females . 10 . 7554/eLife . 16371 . 003Table 1 . Results of LMM for the engagement of the Sonso chimpanzees with the honey-trap experiment . p-values for intercept and terms comprised in the three-way interaction are omitted . Reference levels for categorical predictors are female ( sex ) , and no ( tool use ) . p-values resulted from likelihood ratio tests . DOI: http://dx . doi . org/10 . 7554/eLife . 16371 . 003β± setp95% CIIntercept0 . 040 . 270 . 14Ripe fruit feeding0 . 040 . 050 . 80Time period−0 . 000 . 01−0 . 11Travel time0 . 080 . 051 . 73Sex ( male ) −0 . 310 . 40−0 . 780 . 4517−1 . 100 , 0 . 473Age1 . 210 . 0913 . 180 . 00001 . 028 , 1 . 387Tool use ( yes ) 1 . 250 . 1210 . 660 . 00001 . 021 , 1 . 481Auto correlation−0 . 300 . 01−36 . 130 . 0000−0 . 313 , −0 . 281Ripe fruit : Time period−0 . 000 . 01−0 . 54Ripe fruit : Travel time−0 . 020 . 02−1 . 32Time period : Travel time0 . 010 . 011 . 44Ripe fruit : Time period : Travel time−0 . 020 . 01−2 . 400 . 0163−0 . 030 , −0 . 00310 . 7554/eLife . 16371 . 004Figure 1 . The relationship between ripe fruit feeding , travel time , time period and engagement in the honey experiment ( A , Figure 1—source data 1 ) and ripe fruit feeding , travel time and tool use during the experiment ( B , Figure 1—source data 2 , 3 and 4 ) . Each panel shows the relationship between ripe fruit feeding , travel time and engagement , respectively use of tools , for time periods of 1 , 7 and 13 weeks . All variables were standardized to a mean = 0 and SD = 1 . For better readability , colour gradients along the model planes reflect predicted values along the vertical axis ( engaged in experiment ) : larger values appear in red and smaller values in blue . DOI: http://dx . doi . org/10 . 7554/eLife . 16371 . 00410 . 7554/eLife . 16371 . 005Figure 1—source data 1 . Engagement data . DOI: http://dx . doi . org/10 . 7554/eLife . 16371 . 00510 . 7554/eLife . 16371 . 006Figure 1—source data 2 . Tool data 1 week . DOI: http://dx . doi . org/10 . 7554/eLife . 16371 . 00610 . 7554/eLife . 16371 . 007Figure 1—source data 3 . Tool data 7 weeks . DOI: http://dx . doi . org/10 . 7554/eLife . 16371 . 00710 . 7554/eLife . 16371 . 008Figure 1—source data 4 . Tool data 13 weeks . DOI: http://dx . doi . org/10 . 7554/eLife . 16371 . 008 Concerning the occurrence of tool use ( observed in 21 of 292 trials; 7 . 2% ) , we built three generalized linear mixed models at three different time periods ( 1 week , 7 weeks and 13 weeks ) because a single model analogous to the one presented above did not converge . Each of these models included the interaction between ripe fruit feeding and travel . We found that only the 1-week model was significantly different from its corresponding null model ( LRT: 1 week: X2 = 12 . 0 , df = 5 , p = 0 . 0346 , R2m = 0 . 30; 7 weeks: X2 = 7 . 6 , df = 5 , p = 0 . 1810 , R2m = 0 . 19; 13 weeks: X2 = 8 . 5 , df = 5 , p = 0 . 1299 , R2m = 0 . 18 ) . Contrary to the previous engagement time model , we did not find any effect of the interaction between ripe fruit feeding and travel time on the likelihood of tool use ( all p>0 . 1 , Table 2 ) . However , we found a significant main effect of travel time on the probability of tool use , which increased with travel time ( Table 3 , Figure 1B top panel ) . No such result was found for ripe fruit feeding , although the effect went into the expected direction ( i . e . , more tool use with less ripe fruit feeding ) . For the other two time periods , the estimated effects of travel time and ripe-fruit feeding also went into the expected directions ( Table 3 , Figure 1B ) . 10 . 7554/eLife . 16371 . 009Table 2 . Likelihood ratio tests for full model and the interaction between ripe fruit feeding and proportion of travel time for the tool use models . Null models contained the random effects structure and the auto-correlation term . DOI: http://dx . doi . org/10 . 7554/eLife . 16371 . 009Full vs . null model ( df = 5 ) Interaction Ripe fruit : Travel time ( df = 1 ) Time periodX2pX2p1 week11 . 990 . 03490 . 250 . 61697 weeks7 . 580 . 18100 . 020 . 893113 weeks8 . 520 . 12990 . 430 . 511610 . 7554/eLife . 16371 . 010Table 3 . Model results for GLMMs testing the occurrence of tool use . p-values are presented only for the first model as the two other models were not significant ( see Table 2 ) . All numeric predictor variables were standardized to mean = 0 and SD = 1 . For sex , ‘female’ is the reference level . DOI: http://dx . doi . org/10 . 7554/eLife . 16371 . 0101 week7 weeks13 weeksβ± sezpβ± sezpβ± sezpIntercept−3 . 550 . 42−8 . 470 . 0000−3 . 390 . 37––−3 . 400 . 37––Ripe fruit feeding−0 . 240 . 26−0 . 930 . 3525−0 . 330 . 24––−0 . 350 . 24––Travel time0 . 670 . 302 . 250 . 02420 . 300 . 27––0 . 320 . 26––Sex ( male ) −0 . 070 . 58−0 . 120 . 90620 . 040 . 54––0 . 040 . 55––Age−0 . 560 . 33−1 . 720 . 0855−0 . 560 . 29––−0 . 550 . 29––Auto-correlation0 . 800 . 184 . 440 . 00000 . 520 . 16––0 . 500 . 16–– Finally , we analysed our data set on published estimates of diet and travel related behaviour of nine habituated wild chimpanzee communities ( Table 4 ) . In accordance with the results found in our analysis , we found that larger tool repertoires were associated with lower percentages of fruit consumption ( Spearman’s rho = −0 . 43 , N = 9 , Figure 2A ) and higher percentages of travel ( rho = 0 . 61 , N = 9 , Figure 2B ) . When using average distance travelled per day , we found again a positive relationship with size of tool repertoire ( rho = 0 . 77 , N = 7 , Figure 2C ) . Similar to our experimental data , the effect of the travel-related variables was larger than the effect of ripe fruit feeding . 10 . 7554/eLife . 16371 . 011Table 4 . Data set for the cross-community comparison of nine wild chimpanzee study sites . Number of tools used were taken from Sanz and Morgan ( 2007 ) , except for Fongoli . DOI: http://dx . doi . org/10 . 7554/eLife . 16371 . 011SubspeciesSite/ groupNumber of tools% fruit in diet% travelDaily travel distance ( km ) ReferenceverusBossou1360 . 319 . 5–Hockings et al . ( 2009 , 2012 ) Fongoli10*60 . 811 . 03 . 3*Bogart and Pruetz ( 2011 ) ; Pruetz and Bertolani ( 2009 ) Tai/North1185 . 022 . 03 . 7§Boesch and Boesch-Achermann ( 2000 ) ; Boesch et al . ( 2006 ) ; Herbinger et al . ( 2001 ) troglodytesGoualougo1156 . 012 . 8†–Morgan and Sanz ( 2006 ) ; Sanz† ( 2004 ) schweinfurthiiGombe1243 . 013 . 63 . 9§Wrangham ( 1977 ) Kanyawara266 . 611 . 02 . 1§Pontzer and Wrangham ( 2004 ) ; Potts et al . ( 2011 ) Mahale/M531 . 018 . 6‡4 . 8¶Huffman‡ ( 1990 ) ; Matsumoto-Oda¶ ( 2002 ) ; Nishida and Uehara , ( 1983 ) Ngogo491 . 514 . 03 . 0#Amsler# ( 2010 ) ; Potts et al . ( 2011 ) Sonso165 . 57 . 52 . 1**Bates and Byrne** ( 2009 ) ; Fawcett ( 2000 ) ; Newton-Fisher ( 1999 ) * Jill Pruetz , personal communication; travel estimate based on data from rainy season;percentage of travel in daily budget: † from her table 6 . 2 , taking the highest value ( range: 7 . 6−12 . 8 ) as travel activity was likely underestimated because of low habituation ( Sanz , 2004 , p . 169 ) ;‡ from his table 12 . 2 , mean over individuals of both sex in the year 1985;daily travel values: § average calculated across sex following Pontzer and Wrangham ( 2004 ) ;# from her table I , calculated as sum of hourly averages over a 10-hr activity day , based on males only;¶ from her figure 4 , calculated across seasons and sex;** calculated from the average provided for each sex . 10 . 7554/eLife . 16371 . 012Figure 2 . Relationship between percentage of fruit in the diet ( A ) , percentage of travel in the activity budget ( B ) , daily travel path ( km , C ) , and the number of feeding-related tools described in currently documented long-term habituated chimpanzee communities . See Table 4 for details . DOI: http://dx . doi . org/10 . 7554/eLife . 16371 . 012
Our results indicate that travel is directly related to the probability of tool use behaviour in wild chimpanzees . Our data first showed that the combination of low ripe-fruit availability and high travel effort increased their motivation to engage with a foraging problem that required tool use . Specifically , in situations of low fruit availability , the subjects spent more time engaging with the apparatus than at times of high ripe fruit availability , suggesting that they were possibly more inclined to explore alternative hard-to-get food possibilities . Our second finding was that tool use was mainly driven by short-term changes in daily travel , and less so by fruit availability . Specifically , tool use increased with increasing amounts of travel before the experiment , but this was mostly a short-term effect , up to one week prior to an experiment ( Figure 1B , Table 3 ) . Taken together , these results suggest that travel generates extra energetic costs in situations of low fruit abundance and that tool use is more likely to appear if ecological situations force chimpanzees to explore alternative feeding options in situations of high energy expenditures . In this respect , tool use itself does not appear to be fostered by resource limitation , but rather by increased energetic costs . While tool use was interpreted as a fall back strategy in response to food scarcity in Bossou ( Yamakoshi , 1998 ) , in line with the original definition of necessity ( Fox et al . , 1999 ) , this effect may not be observed in communities that do not display habitual feeding-related tool use behaviour , such as Sonso . The Budongo Forest has been described as a rich habitat where periods of extreme food scarcity are absent ( Newton-Fisher , 1999 ) , which may prevent chimpanzees from experiencing extreme necessity . Food availability nevertheless undergoes seasonal fluctuations ( Reynolds , 2005 ) and , over the last decade , the food supply has noticeably gone down , in part due to anthropogenic activities ( Babweteera et al . , 2012 ) . The Sonso chimpanzees have responded with behavioural adaptations to the disappearance of their original food resources ( Reynolds et al . , 2015 ) , which suggests that detailed analyses are needed to better understand how food variation affects chimpanzee behaviour . Overall , our results suggest that chimpanzees are more eager to exploit difficult resources when the ecological conditions are more demanding relative to average conditions , both in terms of low food availability and high amounts of energy required to obtain the food . While high travel effort in itself is not necessarily linked with low diet quality in chimpanzees ( e . g . Riedel et al . , 2011 ) , our analyses show that a combination of the two favours the exploration of alternative food resources , which creates opportunities for acquiring new tool behaviours . We interpret these findings as support for the more general idea that necessity can also drive invention in wild chimpanzees , when energetic demands are high . Necessity , in other words , is likely to be a major factor in driving the emergence of tool use behaviour in chimpanzees , if it is redefined to take into account both energetic costs and opportunities to compensate these costs . These results underline the importance of individual-based analyses that take into account data on both energetic expenditure and intake , with potentially important implications for theories about the origin of tool use behaviour . Our results are in line with the ‘relative profitability hypothesis’ , which states that extractive tool use will occur if it is relatively more profitable than other alternative foraging strategies that do not rely on tool use ( Rutz and St Clair , 2012 ) . If increased travel effort represents an extra energetic cost , then tool use is a relatively more profitable strategy , especially if this occurs in ecologically challenging situations , which may trigger the switch from non-tool to tool-based foraging . Interestingly , the chimpanzees of Budongo Forest have increased their crop raiding habits over the last decade ( Tweheyo et al . , 2005 ) , a probable response to a general decrease in food availability in the forest ( Babweteera et al . , 2012 ) . As such , the innovation of novel tool use may only be one possible reaction to a changing environment , highlighting the flexibility of chimpanzees in dealing with changes in food availability ( Hockings et al . , 2015 ) . Another facet of the relative profitability hypothesis is that tool use may provide individuals with a selective advantage over non-tool-using individuals ( Patterson and Mann , 2011 ) , as it provides them access to an energetically valuable resource , although in only 7% of trials did subjects succeed to do so . Perhaps this is not so surprising as tool use innovation is itself rarely observed in the animal kingdom ( Shumaker et al . , 2011 ) and only some species will develop tool use under identical ecological conditions ( Rutz and St Clair , 2012 ) , a reasoning that may apply at the population or individual level , as suggested by the current study . While alternate strategies , such as crop-raiding , may contribute in part to the general lack of tool use inventions in this community , it is equally possible that psychological mechanisms can explain some of the observed patterns , offering insights into the ‘invention hypothesis’ . Here , one important result of our study is that the large majority of the tool-using individuals ( 19 of 21 cases , 90 . 5% ) applied a familiar technique , leaf-sponging , in the experiment , behaving much different from when extracting honey from natural bee nests with their hands . Nevertheless , while adapting an existing behaviour to a novel context may be considered an innovation ( Reader and Laland , 2003; Reader et al . , 2016 ) , only two individuals chose a different technique by attempting to use sticks . However , these two individuals did not incorporate this behaviour into their repertoire , raising questions about how wild chimpanzees represent artefacts as potentially useful tools ( Gruber et al . , 2015; Gruber , 2016 ) . Additional studies are needed to explore the cognitive processes underlying chimpanzee tool use and , particularly , to decipher how ecological pressures and cognitive factors interact to lead to tool use innovation . A neglected aspect in this study were the social opportunities for individuals to engage with the device or observe others to do so ( see Koops et al . ( 2014 ) ) . In our study , engagement with the apparatus overlapped between social contexts ( Figure 3 ) , suggesting that the presence of others did not prevent subjects from engaging with it . However , it is less clear how the presence of others influenced the use of tools . Six individuals used a tool while being alone , seven others while in the company of family members , and eight in the company of other group members , to the effect that the current study cannot disentangle the relative role of social competition . Although tool-users spent more time with the apparatus and consumed more honey ( Gruber ( 2016 ) and see Table 1 ) , it is unlikely that this was because they monopolized the log . Rather , these individuals had developed a successful technique to recover the honey , compared to others who abandoned the apparatus earlier ( Gruber , 2016 ) . However , social influences are also in terms of social learning opportunities . As described elsewhere ( Gruber , 2016 ) , chimpanzees were generally tolerant to each other , but it is unclear whether they learnt from each other that leaf-sponge use was a suitable solution to extract honey . Social learning is a reasonable explanation for three individuals , but individual learning cannot be ruled out , largely because leaf-sponging was already part of their behavioural repertoire . Nevertheless , wild chimpanzees can learn socially from each other , even in a competitive context , and it is equally possible that this may even enhance social learning as it facilitates close observations of the novel behaviour ( Hobaiter et al . , 2014 ) . 10 . 7554/eLife . 16371 . 013Figure 3 . Range of engagement time of Sonso chimpanzees with the honey-trap experiment depending on the social context ( alone , family-unit , or social ) . DOI: http://dx . doi . org/10 . 7554/eLife . 16371 . 013 From our data , we conclude that the emergence of tool use in our group was due to a combination of necessity ( energetic demands ) , opportunity ( inaccessible honey ) and relative profitability ( lack of alternatives ) , suggesting that ecological and temporal aspects of resources availability as well as individual efforts all played a role ( Gruber , 2013 ) . While it is important to quantify the food available over the entire home range , it is also important to take into account the temporal variation of food availability and its consequences on the relative attractiveness of alternative foods simultaneously available to individuals , even for foods as attractive as honey . We concur with Koops et al . ( 2014 ) that individuals must be exposed to the right ecological opportunities , in our case the honey-trap apparatus , and that the probability of tool use may directly depend on the frequency with which they will encounter this challenge , a parameter we controlled for in our experiment . For tool acquisition and spread to appear , the right social settings may also have to be present ( Sanz and Morgan , 2013 ) , under the form of opportunities for close observation ( Hobaiter et al . , 2014 ) . In our case , encountering tools left by others ( i . e . discarded sticks ) did not appear to constitute a sufficient condition for social learning ( Gruber et al . , 2011 ) . Our data also suggest that individual differences need to be taken into account . Finally , while cognitive abilities are likely to play a part in innovation and learning ( Gruber , 2016 ) , the emergence of tool use may also depend on whether it is relatively more profitable to do so , at any given time ( Rutz and St Clair , 2012 ) . Here , our data suggest that energetic demands resulting from individual variation in diet and travel effort directly influence the probability of tool use . Is the relationship between travel and tool use generally found in the Hominidae ? Our analysis of nine chimpanzee communities , although limited by the availability of published data on travel effort and tool use , suggests that our findings represent a general pattern . This analysis corroborated our empirical findings that travel effort and fruit consumption have opposing effects on tool repertoire size , and that travel effort , which arguably is best represented by the average daily distance walked by the chimpanzees , is likely to be more important than fruit consumption in explaining variation in tool repertoires between chimpanzee populations . In the long run , with more chimpanzee communities being currently habituated spanning across their entire ecological range ( e . g . savanna in Fongoli , Pruetz , 2006 ) , future studies will have to disentangle how environmental changes influence the relationship between tool use , energy intake and expenditure across a larger sample of chimpanzee populations . Regarding other great apes , both gorillas ( genus Gorilla ) and most orangutans ( genus Pongo ) show limited to no feeding-related tool use and interestingly they spend significantly less time travelling per day compared to chimpanzees , which suggests that their energy requirements are lower ( Pontzer and Wrangham , 2004; Pontzer et al . , 2016 ) . Nevertheless , because travel is mostly arboreal in orangutans , more work is needed to estimate how this compares to chimpanzees , particularly with respect to Sumatran orangutans ( Pongo abelii ) for whom interesting variation in feeding-related tool use behaviour has been described ( van Schaik et al . , 2003; Gruber et al . , 2012b ) . The most promising comparison may come from the chimpanzees’ closest relatives , the bonobos ( Pan paniscus ) , where the lack of tool use has been connected to smaller travel distance between food patches and reduced feeding competition ( Wrangham and White , 1988 ) . The estimated daily travel effort for bonobos ( 2 . 6 km , Furuichi et al . ( 2008 ) ) is comparable with some chimpanzee communities , incidentally the ones with the smallest tool repertoires for the species ( Kanyawara , Ngogo and Sonso , all in Uganda ) , underlining a possible convergence in ecological pressures faced by these populations ( Gruber et al . , 2010 ) . Interestingly , some convergence can also be found with modern humans . For instance , modern human hunter-gatherers walk on average 11 . 4–14 . 1 km/day ( Marlowe , 2005; Pontzer et al . , 2012 , 2016 ) and have the most diverse tool repertoire of all Hominidae , much beyond anything reported from the great apes ( Marlowe , 2010 ) . Combined , the results of the present study and the data from the three living hominines ( Homo , Pan ) reviewed here suggest an important role of travel in the emergence of tool use , but this needs to be tested across more study groups in different habitats and species . Whether this pattern holds for larger taxonomic groups beyond hominines remains to be investigated , taking into account the various ecological conditions faced by each species . In conclusion , our findings suggest that tool use in hominids evolved in reaction to environmental changes that made preferred food harder to obtain . By extension , our results have direct implications for understanding hominid technological evolution , particularly in relation to the evolution of locomotor behaviour in the early stages of human evolution , as hominids faced similar ecological pressures . In effect , a number of major biogeographic events in the human lineage occurred at times of climate instability and it has been suggested that the development of tool use and sociality in hominins could constitute adaptive responses to heightened habitat instability ( Potts , 2013 ) . Australopithecines , for instance , evolved in a changing environment at the beginning of the Pliocene , where they faced more patchy resources of potentially lower quality ( Foley and Gamble , 2009; Potts , 2013 ) . Our findings support the view that tool use is connected to energy gain in a changing environment and that using tools is a response to increased costs of travel and lower quality of available food . In parallel , the adoption of bipedalism , which is less energetically costly than the quadrupedal and bipedal locomotion of chimpanzees , also allowed minimizing energy expenditure ( Pontzer et al . , 2009 ) . Efficient , human-like bipedalism and tool use may have had complementary effects on travel costs , allowing both energy gain through the exploitation of novel ecological niches and energy economy during locomotion . Whether their development to unrivalled levels is what led to the dispersal of early humans throughout Africa and the advent of complex technology around 3 . 0 million years ago ( Foley and Gamble , 2009; Harmand et al . , 2015 ) remains to be investigated .
The data were collected in the Sonso chimpanzee community of the Budongo Forest Reserve , Uganda ( 1°350–1°550 N , 31°180–31°420 E ) , at a mean altitude of 1050 m within 482 km2 of continuous medium-altitude semi-deciduous forest ( Reynolds , 2005 ) . Rainfall in the Budongo Forest follows a bimodal pattern with two main rainy seasons between March and May and between September and November ( Figure 4A , Reynolds , 2005 ) . Habituation of the community started in 1990 with all residents identified , around 70 over the last eight years . The Sonso chimpanzees are notable for their complete lack of feeding-related tool-using behaviour with the exception of leaf- and moss-sponging ( Hobaiter et al . , 2014 ) . Data included in the analysis consisted of six years of experimental data , collected between 2009 and 2015 . We combined our experimental data with observational data collected between 2008–2015 , up to three months before each experimental trial . 10 . 7554/eLife . 16371 . 014Figure 4 . Temporal variation in climate in the Budongo Forest ( A ) and in feeding behaviour of the Sonso community ( B , C ) during the period covering the experimental trials . Months during which experiments were conducted are highlighted in red . ( A ) To define the climate factor , we calculated monthly cumulative rainfall and mean temperatures , extracted from daily values for rainfall , minimum temperature and maximum temperature in Budongo Forest available from 2001 through 2015 ( Budongo Conservation Field Station long-term data 2001–2015 ) . These monthly values were subjected to a principal component analysis ( function ‘princomp’ in the stats package R v . 3 . 1 . 1 , R CoreTeam ( 2014 ) ) . The climate factor corresponds to the scores of the first component of this analysis , which explained 64% of variance . Larger values along this axis correspond to larger values of rainfall , higher minimum temperature and lower maximum temperature as compared to smaller values along the climate factor . For reference , monthly cumulative rainfall is also plotted in this panel ( dashed line ) . Both variables were standardized to mean = 0 and SD = 1 . As such , values of 0 indicate average climate/rainfall ( horizontal grey line ) . Out of 19 months with experimental days , 10 were characterised by above-average climate/rainfall and 9 by below-average climate/rainfall . ( B ) Variation in ripe fruit feeding behaviour . Shown are monthly median values of the proportion of ripe fruit in the diet for individuals that were observed at least five times feeding during a given month . Grey bars indicate quartiles and the horizontal dashed line represents the mean value across all individual-months . ( C ) Variation in fig feeding . Shown are monthly median values of the proportion of figs in the diet for individuals that were observed at least five times feeding on ripe fruit during a given month . Grey bars indicate quartiles and the horizontal dashed line represents the mean value across all individual-months . DOI: http://dx . doi . org/10 . 7554/eLife . 16371 . 014 The Sonso chimpanzees are opportunists in relation to honey consumption , acquiring honey from natural bee nests ( Xylocopa and Apis genus ) . Honey acquisition does not involve any tool use and is carried out with limited success only ( T . Gruber , personal observation ) . In our honey-trap experiment , we provided subjects with the opportunity to systematically engage with a foraging problem , a 16 cm deep hole drilled into a 50 cm long log of about 25 cm diameter . The honey-trap experiment , by closely mimicking a natural setting , has proven its ecological validity , with over 80 individuals in two unhabituated and two habituated communities engaging with the experiment ( Gruber et al . , 2009 , 2011 , 2012a ) . The hole contained natural honey up to about 10 cm below the surface , which could only be extracted with the help of a suitable tool ( Gruber et al . , 2009 , 2011 ) . Honeycombs were positioned so that they covered the hole to prevent insects , such as bees and ants , from entering it . Finally , a stick was potentially placed next to the log or directly plugged into the honey , depending on the experimental condition ( Gruber et al . , 2011 ) . The apparatus was only set up when no chimpanzees were around and the experimenter ( TG ) always left the experimental area before the arrival of a subject . Several such apparatuses were in operation throughout the study period , all of them positioned at different locations throughout the Sonso territory . We never limited access to the apparatus , so that several individuals could participate during a given experimental day , possibly simultaneously . Our final sample consisted of 292 trials , involving 52 individuals ( over 70% of the total Sonso community ) , on 96 experimental days . In 124 cases of 292 ( 42 . 5% ) , the tested subjects were strictly alone while in 86 cases of 292 ( 29 . 5% ) we tested individuals within a family unit . Finally , in 82 cases out of 292 ( 28% ) , other individuals joined the tested subject and also engaged with the honey-trap . These trials were also counted as engagement with the apparatus if the individual attempted to recover the honey ( Figure 3 ) . Experimental days were spread over six years ( between 2009-2010 and 2012-2015 , about 19 experimental days per year ) , with several weeks without experiment between each set of trials . Engagement time was defined as the time spent by a subject actively seeking to recover honey from the apparatus . Any attempt at playing with the log , or simply resting on the log was not included . In total , we observed 21 distinct tool use occurrences by 11 individuals: six in the alone context , seven in the family unit context and eight in the social context . For three of the latter trials , this occurred during social trials when other individuals had been using a tool before them . Because it has been shown that chimpanzee sponging is influenced socially ( Hobaiter et al . , 2014 ) , we cannot exclude that these individuals may have been influenced by the previous individual engaging with the log . However , it is also possible that chimpanzees opted for a tool solution independently in each of these cases ( see discussion in Gruber ( 2016 ) ) . For this reason , we considered each of the 21 instances independent from each other . Experiments were carried out both in dry and wet seasons , to control for a potential effect of seasonality and to encompass the entire range of ecological variation in terms of possible food offer available to the chimpanzees ( Figures 4 and 5 ) . Over the last decades , Budongo Forest has been described as a rich environment where chimpanzees do not face periods of food scarcity comparable to the ones experienced by other chimpanzee populations ( Newton-Fisher , 1999 ) . For instance , a study conducted during this period found that there was no positive relationship between food availability and party size , a marker of food scarcity ( Newton-Fisher et al . , 2000 ) . Nevertheless , recent research has shown that the food supply has steadily decreased in Budongo Forest , suggesting the possible appearance of periods of food scarcity for the resident primate populations ( Babweteera et al . , 2012 ) . Interestingly , when correlating party size with proportion of ripe fruit feeding over the whole duration of our study , we found an overall relationship close to 0 , reflecting the results of the earlier study by Newton-Fisher et al . , ( 2000 ) . However , we saw a large variation between years , with years ( 2009 , 2013 , 2014 ) where the relationship follows the more conventional chimpanzee pattern ( i . e . more ripe fruit feeding coincides with larger parties ) , years ( 2010 , 2011 ) where this pattern follows an opposite direction , and years ( 2008 , 2012 ) where there is no clear pattern ( Figure 5B ) . Additionally , even within a particular year , we observed substantial variation in monthly ripe fruit consumption ( range: 0 . 00 to 0 . 93 , Figure 4B ) . Similarly , there was also large variation in time spent feeding on fig species across the duration of our study ( Figure 4C ) , with fig species often considered a fallback food for chimpanzees , and their consumption a potential marker of food scarcity ( Marshall and Wrangham , 2007; Harrison and Marshall , 2011 ) . Our experiments , spread across this spectrum , thus allowed us to test the potential effect of food scarcity and travel effort across a large range of ecological situations . 10 . 7554/eLife . 16371 . 015Figure 5 . Within- and between-year variation in the relationships between ripe fruit feeding and ( A ) travel time and ( B ) mean monthly party size . In ( A ) , each black line represents the regression line of travel time on ripe fruit feeding within a month , based on data from focal individuals . Thus , per panel 12 lines are depicted , except for 2015 for which data were available only for the first three months . The red line depicts the average regression over the respective year . In ( B ) , each line represents a regression line of monthly average party size on average monthly ripe fruit feeding proportion . Each line is based on data from a random selection of parties ( limited to one party per day ) to calculate the monthly average party size . The randomization was repeated 20 times , resulting in 20 regression lines per panel . The panel for 2015 is based on regressions with only three data points as data were only available for the first three months of 2015 . DOI: http://dx . doi . org/10 . 7554/eLife . 16371 . 015 Long-term data on party composition as well as foraging and ranging behaviour have been collected by trained field assistants since the beginnings of the project . During focal animal follows , the field assistants note every 30 minutes a focal individual’s activity ( feeding , travel , resting , grooming ) and , if feeding , the plant species and the plant part ( ripe fruit , unripe fruit , leaves , flowers , bark ) consumed . In addition , party composition is recorded by noting all adolescent and adult individuals in the focal animal’s party . Data for dependent juvenile individuals are extracted from their mother's behaviour . To increase our sample on feeding and travel behaviour of individuals , we assumed that all party members expressed the same behaviour as a party’s focal individual . This approach is justified given an analysis of a subset of our data for which the activity for all party members ( in addition to the focal individual ) was recorded . Across 31 , 278 party scans , the mean proportion of individuals that expressed the same behaviour as the party’s focal individual was 0 . 8 ( median = 1 . 0 , range: 0 . 0–1 . 0 ) . For each subject who participated in an experimental trial , we calculated separately its time spent feeding on ripe fruit and its time travelling in the following way . We identified all data points in our behavioural database in which the subject was present in an observed party ( regardless of whether the subject was the focal animal or not , see above ) . We then noted the respective focal animal’s activity and plant part eaten . In other words , we considered the focal animal as representative for the experimental subject as long as they were part of the same party . Because juveniles who engaged with the experiments were still dependent to their mother at the time of the experiment ( and therefore are not considered as individual points in the database ) , we extracted these data from their mother's data . For this study , we analysed N = 40 , 908 data points collected by nine experienced field assistants between 2008 and 2015 . From this database , we calculated ripe fruit feeding and travel time as proportions , i . e . as the number of data points feeding on ripe fruit relative to all data points spent feeding , and travelling relative to all observations of that subject ( or its respective focal animal , if the subject was in the party but not itself the focal animal , see above ) . Because we had no a priori expectation as to what time period was meaningful to the chimpanzees , we considered different time periods , ranging from one week prior to the experiment up to 13 weeks before the experiment ( i . e . approximately three months ) , using one-week increments . Note that we did this in a cumulative fashion , i . e . a given 2-week data point included the data of the first week before an experiment , a given 3-week data point included data from weeks 1 and 2 , and so forth . We controlled for this inter-dependence statistically ( see below ) . In this way , we assembled a total of 292×13 = 3 , 796 data points . Out of these , we had to exclude 52 data points because no observational data were available for a given subject ( mostly during the shorter time periods ) . Our final data set comprised 3 , 744 data points , including data from 50 subjects that participated in the honey experiment . Our objective was to investigate how ecological parameters ( feeding on ripe fruits and travelling ) influenced the time the subjects engaged with the apparatus and whether a tool was used during the experiment . Our main predictor variables were the proportions of time spent feeding on ripe fruit and time spent travelling , plus their interaction . Further , we included a 3-way interaction between ripe fruit feeding , travel time and time period , reasoning that any effect of feeding and/or travelling may be short or long term . The time period variable indicated the number of weeks ( range: 1–13 , i . e . about three months ) over which an individual’s travel and feeding data were accumulated prior to an experimental trial . Finally , we searched the published literature for estimates of the travel and feeding behaviour of wild chimpanzees . In particular , we collected data on activity budget , tool use and diet from long-term habituated chimpanzee communities for which tool-use behaviour was known ( N = 9 , Table 4 ) . When possible , we used fruit consumption and travel data from the same study , as this would directly connect the travel effort with the food consumed at the time of the study . When this was not possible , we extracted or calculated the values from the literature . If there were more than one value for any of the variables , we selected the values that had been estimated the closest to each other . For tool use , we only took into account feeding-related tool use behaviour , as reviewed by Sanz and Morgan ( 2007 ) . We used estimates of travel in activity budget and proportion of fruits in the diet to compare with our experimental data . We calculated non-parametric ( Spearman ) correlations between these values and the number of different tools used in the respective communities . We also ran an additional correlation between number of tools and daily travelled distance when these data were available . | There is currently much debate about the origins of animal culture , including why some animals have acquired the ability to use tools . Ecological problems often lead to the innovation of new tools . For example , a particular desirable food item may not be reachable without using a tool , or environmental conditions may make it difficult for an animal to find food without help . Gruber et al . investigated how particular ecological factors influenced the use of tools in wild chimpanzees by combining controlled field experiments and observational data . When the ecological conditions were the most demanding , wild chimpanzees engaged most with the honey-trap experiment , an experiment where they had to use a tool to extract honey from a cavity dug in a log . Chimpanzees spent a longer time engaging with the apparatus when not much food was available and they had to travel more to obtain it . However , actual tool use during the experiments was only influenced by the travel effort made by the chimpanzees before they engaged with the log , not by how much fruit they had eaten beforehand . In a larger analysis that included data from all of the long-term field sites with habituated chimpanzees , Gruber et al . found that chimpanzee communities that travel further on a daily basis use a wider range of tools to acquire food . These results suggest that travel is an important factor to consider when studying how tool use evolved . Furthermore , these results can be extrapolated to humans , who both travel further and use a greater variety of tools than chimpanzees . Although innovation and culture are closely linked , innovation is mostly performed by individuals whereas culture is a social process . However , both are shaped by the environment . The next step will therefore be to disentangle and quantify the different contributions of environmental , individual and group factors in explaining how culture evolves . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Material",
"and",
"methods"
] | [
"ecology"
] | 2016 | Travel fosters tool use in wild chimpanzees |
Gene copy number alterations , tumor cell stemness , and the development of platinum chemotherapy resistance contribute to high-grade serous ovarian cancer ( HGSOC ) recurrence . Stem phenotypes involving Wnt-β-catenin , aldehyde dehydrogenase activities , intrinsic platinum resistance , and tumorsphere formation are here associated with spontaneous gains in Kras , Myc and FAK ( KMF ) genes in a new aggressive murine model of ovarian cancer . Adhesion-independent FAK signaling sustained KMF and human tumorsphere proliferation as well as resistance to cisplatin cytotoxicity . Platinum-resistant tumorspheres can acquire a dependence on FAK for growth . Accordingly , increased FAK tyrosine phosphorylation was observed within HGSOC patient tumors surviving neo-adjuvant chemotherapy . Combining a FAK inhibitor with platinum overcame chemoresistance and triggered cell apoptosis . FAK transcriptomic analyses across knockout and reconstituted cells identified 135 targets , elevated in HGSOC , that were regulated by FAK activity and β-catenin including Myc , pluripotency and DNA repair genes . These studies reveal an oncogenic FAK signaling role supporting chemoresistance .
Ovarian carcinoma is the most lethal gynecologic malignancy in the United States ( Siegel et al . , 2018 ) . High-grade serous ovarian carcinoma ( HGSOC ) , the most prevalent histologic tumor subtype ( Matulonis et al . , 2016 ) , is treated with a combination of cytoreductive surgery and carboplatin ( DNA damage generation ) and paclitaxel ( microtubule-stabilizing drug ) chemotherapy . Cure is highly dependent on elimination of microscopic disease ( Narod , 2016 ) . Approximately 80% of patients with HGSOC exhibit serial disease recurrence , develop resistance to platinum chemotherapy , and die ( Bowtell et al . , 2015 ) . Although platinum chemotherapy is effective at creating DNA damage and triggering cell apoptosis , subpopulations of tumor cells can survive this stress ( Pogge von Strandmann et al . , 2017 ) . Tumor sequencing has revealed complexity and heterogeneity among HGSOC ( Cancer Genome Atlas Research Network , 2011 ) . DNA breakage and regions of chromosomal gain or loss are common ( Patch et al . , 2015 ) . Gains at 8q24 occur in most HGSOC tumors and encompass the MYC oncogene at 8q24 . 21 ( Gorringe et al . , 2010 ) . Although MYC expression is frequently high in HGSOC , the clinical significance remains unclear . MYC supports pluripotent stem cell generation and contributes to chemoresistance ( Fagnocchi and Zippo , 2017; Kumari et al . , 2017; Li et al . , 2019 ) . Myc protein expression is regulated by Wnt/β-catenin signaling , which is both essential for embryonic development and activated in many tumors ( Shang et al . , 2017 ) . Wnt and Myc fall within the 10 most prevalent signaling pathways in cancer ( Sanchez-Vega et al . , 2018 ) . Wnt signaling is tightly regulated by the stability , subcellular localization , and transcriptional activity of β-catenin , which supports cancer stem cell ( CSC ) survival and chemoresistance ( Condello et al . , 2015; Nagaraj et al . , 2015 ) . Platinum can , paradoxically , also select for ovarian cancer ‘stemness’ through undefined mechanisms ( Wiechert et al . , 2016 ) . Increased aldehyde dehydrogenase ( ALDH ) activity , arising from elevated expression of a family of cellular detoxifying enzymes , is one hallmark of ovarian CSCs ( Raha et al . , 2014; Silva et al . , 2011 ) . Culturing cells as tumorspheres in vitro increases chemotherapy resistance , ALDH expression , cell de-differentiation and stemness ( Shah and Landen , 2014; Malta et al . , 2018 ) . Notably , HGSOC dissemination involves tumorsphere growth and survival within ascites ( Pogge von Strandmann et al . , 2017 ) . The PTK2 gene at 8q24 . 3 , encoding focal adhesion kinase ( FAK ) , is frequently amplified in breast , uterine , cervical , and ovarian tumors ( Kaveh et al . , 2016 ) . FAK is a cytosolic tyrosine kinase canonically activated by matrix and integrin receptors controlling cell motility ( Mitra et al . , 2005 ) . Autophosphorylation at tyrosine 397 ( pY397 ) is a hallmark of FAK activity ( Kleinschmidt and Schlaepfer , 2017 ) . HGSOC tumors with PTK2 gains exhibit elevated FAK expression and FAK Y397 phosphorylation ( Cancer Genome Atlas Research Network , 2011; Zhang et al . , 2016 ) . Metastatic HGSOC tumor micro-environments are enriched with matrix proteins that are FAK activators ( Pearce et al . , 2018 ) . FAK knockdown and FAK inhibitor studies support an important role for FAK in promoting invasive tumor growth ( Ward et al . , 2013; Tancioni et al . , 2014 ) , yet the targets downstream of FAK are varied and may be tumor or stroma context-dependent ( Sulzmaier et al . , 2014; Haemmerle et al . , 2016 ) . Interestingly , phenotypes associated with FAK knockout can be distinct from FAK inhibition , since kinase-inactive FAK retains important scaffolding roles ( Lim et al . , 2008 ) . Several ATP-competitive FAK inhibitors have been developed . Acceptable Phase I safety profiles in patients with advanced solid tumors ( Jones et al . , 2015; Soria et al . , 2016; Hirt et al . , 2018 ) have enabled current Phase II combinatorial clinical trials with FAK inhibitors in pancreatic , mesothelioma , and non-small cell lung carcinoma ( NCT02758587 and NCT02546531 ) . In ovarian and prostate carcinoma preclinical models , FAK inhibition ( VS-6063 , defactinib ) enhanced taxane-mediated tumor apoptosis ( Kang et al . , 2013; Lin et al . , 2018 ) . While inhibitors of FAK and Myc exhibit combinatorial activity in promoting HGSOC cell apoptosis in vitro ( Xu et al . , 2017 ) , it remains uncertain whether gains in 8q24 encompassing PTK2 are associated with specific HGSOC cell phenotypes or responses to therapy , as determinants of FAK pathway dependence in tumors remain unknown . Herein , we molecularly characterize a new murine model of ovarian cancer that displays spontaneous gains in the Kras , Myc , and FAK genes among other striking similarities to HGSOC phenotypes . By using a combination of genetic FAK knockout and rescue , pharmacological inhibition , sequencing and bioinformatics , we identify a non-canonical FAK activity-dependent linkage to β-catenin leading to differential mRNA target expression of Myc and other targets supporting pluripotency and DNA repair . Our studies linking intrinsic FAK activity to platinum resistance support the combinatorial testing of FAK inhibitors for recurrent ovarian cancer .
HGSOC is characterized by p53 inactivation and genomic copy number alterations ( CNAs ) , though no preclinical models exist to study cell phenotypes associated with ovarian tumor CNAs . Murine ID8 cells , are spontaneously-immortalized clonal ovarian epithelial cells that form slow-growing tumors in C57Bl/6 mice ( Roby et al . , 2000 ) . ID8 cells do not contain common oncogenic mutations and express wild type p53 . Targeted p53 inactivation promotes ID8 tumor growth and sensitivity to platinum chemotherapy ( Walton et al . , 2016; Walton et al . , 2017 ) . Passage of ID8 cells through C57Bl/6 mice can enhance ID8 tumorigenic potential via undefined mechanisms ( Clark et al . , 2016; Mo et al . , 2015; Ward et al . , 2013 ) . We previously isolated aggressive ID8-IP cells , lethal in mice within 40 days ( Ward et al . , 2013 ) , via early recovery of ascites-associated cells and anchorage-independent expansion ex vivo ( Figure 1A ) . Total exome sequencing ( 90% of exons sequenced at 100X ) of ID8 and ID8-IP cells revealed 19619 shared , 29373 ID8 unique , and 11800 ID8-IP unique gene variants . However , less than 1% of exon variants identified were detected by RNA sequencing ( ~60 million clean reads/replicate ) . No equivalent mutations were found in COSMIC , the Catalogue of Somatic Mutations in Cancer . In addition to non-synonymous mutations previously identified in ID8 cells ( Walton et al . , 2016 ) , we detected two additional changes in Hjurp . In ID8-IP cells , new mutations were identified in Xxylt1 and Atxn10 . Overall , the mutational burden within both ID8 and ID8-IP cells is low . To determine if genetic copy number alterations underlie ID8-IP phenotypes , exome sequencing read values and bioinformatic analyses were used to map sites of DNA gains or loss across chromosomes using ID8 as a reference ( Figure 1—source data 1 ) . Gains in murine chromosome cytoband regions 6qD1-G3 , 15qD3-F3 , and 15qA1-D3 were present in ID8-IP cells ( Figure 1B , green circles ) . These correspond to human cytobands 12p12 . 1 , 8q24 . 2 , and 8q24 . 3 . The latter two represent one of the most amplified regions in HGSOC ( Cancer Genome Atlas Research Network , 2011; Li et al . , 2019 ) . The gain in cytoband 15qA1-D3 is in addition to chromosome 15 polyploidy detected by ID8 karyotyping ( Roby et al . , 2000 ) . Notably , common gene gains in ID8-IP and HGSOC include Kras , Myc , and Ptk2 ( encoding FAK ) that support proliferation , stem cells , and adhesion signaling , respectively ( Table 1 ) . Herein , these ID8-IP cells will be termed KMF to denote gains in Kras , Myc , and FAK genes . Murine KMF cells contain several gains or losses in genes common to the top 20 set of genes altered in HGSOC ( Table 1 ) . In HGSOC patients , simultaneous gains in KRAS , MYC , and PTK2 gains co-occur in 42% of tumors; PTK2 and MYC co-occur in an additional 32% of HGSOC patients ( Figure 1C ) . Thus , more than 70% of HGSOC tumors contain combined gains at PTK2 and MYC loci . PTK2 copy number gains are linearly proportional to PTK2 mRNA ( R2 = 0 . 66 ) and FAK protein ( R2 = 0 . 61 ) levels in HGSOC tumors ( Figure 1C—figure supplement 1 ) . Elevated PTK2 mRNA levels in HGSOC are associated with decreased patient relapse-free survival ( n = 1435 , p=0 . 0009 , and hazard ratio = 1 . 25 ) ( Figure 1—figure supplement 2A ) . Bioinformatic analyses identified a set of 36 genes on different chromosomes in HGSOC that exhibit a significant and at least a two-fold change in tumors with elevated PTK2 . This 36 gene set was associated with a significant shorter time to relapse ( n = 575 , p=0 . 0024 , hazard ratio = 1 . 37 ) ( Figure 1—figure supplement 2B , C ) . Together , these results support the importance of PTK2 gains as a marker for poor prognosis . ID8 cells exhibit an epithelial-like morphology and poor growth as colonies in semi-solid methylcellulose media . KMF cells exhibit a mesenchymal morphology , form foci in two-dimensional ( 2D ) cell culture , and readily form 3D colonies in methylcellulose ( Figure 2A , B ) . When grown in serum-free supplement-enhanced tumorsphere media ( PromoCell ) under anchorage-independent conditions for 5 days , ID8 and KMF cells remain ~95% viable and can be analyzed for signaling differences . By immunoblotting , KMF tumorspheres expressed elevated FAK , increased FAK Y397 phosphorylation , and decreased E-cadherin and β-catenin protein levels relative to ID8 cells and normalized to actin ( Figure 2C ) . Treatment of KMF tumorspheres with a glycogen synthase kinase-3 inhibitor ( GSK3i ) increased β-catenin protein and nuclear transcriptional activity whereas GSK3i addition had no effects on β-catenin levels or activity in ID8 cells ( Figure 2D ) . Additionally , greater than 10% of KMF tumorspheres possessed high ALDH activity with less than 1% of ID8 cells being ALDH-positive ( Figure 2E ) . These results support the notion that KMF cells have gained enhanced CSC characteristics . To determine if ID8 and KMF cells possess distinct transcriptional signatures , RNA sequencing was performed ( Figure 2—source data 1 ) . Using FPKM ( Fragments Per Kilobase of transcript per Million mapped reads ) threshold values greater than one , 10800 shared , 744 ID8 enriched , and 402 KMF elevated transcripts were identified ( Figure 2F ) . Top 20 Reactome signaling pathways upregulated in KMF cells include Cell Cycle Control , Mitotic Checkpoint , DNA Repair , and Rho GTPase signaling ( Figure 2G ) . Elevated cell cycle mRNA levels are consistent with enhanced KMF tumorsphere growth . However , elevated levels of mitotic checkpoint inhibitors such as p21CIP1 were also constitutively and highly expressed in KMF cells ( Figure 2H ) . Deregulated p21CIP1 levels can occur in p53-deficient cells ( Georgakilas et al . , 2017 ) , but no mutations in p53 were detected by KMF exome sequencing and steady-state levels of KMF p53 protein are low . As DNA repair pathway targets are also increased in KMF cells ( Figure 2G ) , ID8 and KMF tumorsphere viability was measured after exposure to different concentrations of cisplatin ( CP ) over 5 days ( Figure 2I ) . KMF cells possessed increased intrinsic resistance to CP cytotoxicity with greater than a 10-fold difference in EC50 values compared to ID8 . Taken together , this new KMF model exhibits noteworthy phenotypic similarities to drug-resistant HGSOC . A small subset of HGSOC patients are treated with neoadjuvant carboplatin and paclitaxel chemotherapy to reduce tumor burden prior to undergoing surgery ( Matulonis et al . , 2016 ) . However , some tumor cells , such as CSCs , escape CP-mediated apoptosis and survive chemotherapy ( Wiechert et al . , 2016 ) . FAK protein and FAK tyrosine phosphorylation ( pY397 FAK ) levels are elevated in primary HGSOC tumors compared to normal tissue ( Zhang et al . , 2016 ) , but it is not known whether chemotherapy alters FAK phosphorylation . To evaluate this , serial sections of paired primary biopsies and tumors obtained at the time of cytoreductive surgery following neoadjuvant carboplatin and paclitaxel chemotherapy were analyzed by immunohistochemical staining and quantitative image analyses ( Figure 3—figure supplement 1 ) . A high degree of Pax8 ( tumor marker ) and pY397 FAK co-localized staining was detected in primary biopsy samples with FAK pY397 exhibiting both cytoplasmic and nuclear localization ( Figure 3A , B ) . Several of these tumor cells stained positive for the Ki-67 proliferation marker . FAK pY397 staining was higher in ovarian tumor compared to surrounding stromal cells ( Figure 3—figure supplement 1 ) . Surprisingly , pY397 FAK staining remained elevated in non-necrotic tumor samples obtained after multiple cycles of neoadjuvant chemotherapy ( Figure 3—figure supplement 1 ) . By comparing samples from the same patients pre- and post-chemotherapy , we found pY397 FAK levels trended significantly upward in tumors surviving neoadjuvant chemotherapy ( Figure 3D ) , further supporting an association between FAK signaling and tumor chemoresistance . FAK pY397 is canonically considered a marker associated with cell adhesion or increased tissue stiffness ( Sulzmaier et al . , 2014 ) . Unexpectedly , pY397 FAK staining was also observed within Pax-8-positive ascites tumorspheres that also displayed active β-catenin staining ( Figure 4A ) . This was unanticipated , since FAK Y397 phosphorylation is rapidly lost when human platinum-resistant OVCAR3 cells are removed from adherent 2D culture and placed in suspension ( Figure 4B ) . However , extended time course analyses of OVCAR3 cells cultured in anchorage-independent PromoCell media revealed that FAK Y397 phosphorylation was restored as OVCAR3 cells clustered to form tumorspheres within 2–3 days ( Figure 4B , C ) . Surprisingly , CP ( 1 µM ) treatment of OVCAR3 tumorspheres ( EC50 >10 µM ) triggered increased FAK Y397 and β-catenin Y142 phosphorylation ( Figure 4D ) . As β-catenin Y142 is a direct FAK substrate promoting β-catenin activation in endothelial cells ( Chen et al . , 2012 ) , our findings support the notion that adhesion-independent non-canonical FAK activation occurs during tumorsphere formation and in response to CP stimulation . As increased FAK Y397 phosphorylation can occur upon CP treatment , we investigated the effect of low dose CP treatment ( 1 µM ) in the presence or absence of a FAK inhibitor ( VS-4718 , 1 µM ) over 5 days on tumorsphere formation , ALDH activity , and cell viability ( Figure 5 ) . Cisplatin EC50 values for growth inhibition were 13 µM and 31 µM for OVCAR3 and KMF tumorspheres , respectively . CP treatment resulted in increased tumorsphere formation and ALDEFLUOR activity in OVCAR3 and KMF cells , consistent with this low CP dose serving as an activation-type stress ( Figure 5A , B ) . In contrast , FAK inhibitor ( FAKi ) reduced tumorsphere formation and ALDEFLUOR activity compared to control-treated OVCAR3 and KMF cells ( Figure 5A , B ) . FAKi was not directly cytotoxic , since only CP combined with FAKi decreased tumorsphere viability ( Figure 5C ) . Single agent CP or FAKi treatment did not alter KMF ( Figure 5—figure supplement 1 ) or OVCAR3 ( Figure 5—figure supplement 2 ) growth or viability in 2D culture . Under 3D conditions , FAKi reduced FAK Y397 phosphorylation and resulted in an elevated percentage of KMF and OVCAR3 cells in G1 phase of the cell cycle ( Figure 5—figure supplements 1 and 2 ) . The finding that FAKi decreased 3D cell proliferation , and that FAKi exhibits combinatorial activity with low-dose CP to promote apoptosis , highlights a potential therapeutic combination . Phosphoinositide 3-kinase ( PI3K ) -elicited Akt activation is one of several survival signaling pathways downstream of FAK ( Sulzmaier et al . , 2014 ) . More than half of HGSOC tumors harbor genetic lesions that can elevate PI3K activity ( Hanrahan et al . , 2012 ) . A2780 human ovarian carcinoma tumor cells contain activating mutations in PI3KCA and inactivation of PTEN - alterations that can promote Akt activation ( Domcke et al . , 2013 ) . OVCAR10 cells similarly exhibit elevated Akt phosphorylation and both A2780 and OVCAR10 cells are resistant to FAKi ( 1 µM ) effects on 3D cell proliferation ( Tancioni et al . , 2014 ) . To determine if in vitro acquisition of increased CP resistance alters FAK signaling , intermittent CP exposure ( 10 µM ) and cell recovery was used to generate OVCAR10-CP ( EC50 = 9 µM ) and maintain A2780-CP70 cell ( EC50 = 60 µM ) selection . Immunoblotting revealed constitutively elevated FAK pY397 within tumorspheres of CP-resistant compared to parental cells ( Figure 6A ) . In addition , we find that CP-resistant A2780-CP70 and OVCAR10-CP cells exhibited a newly acquired dose-dependent sensitivity to FAKi growth inhibition as tumorspheres ( Figure 6B ) , but not when the same cells were grown in 2D conditions ( Figure 6B—figure supplement 1 ) . FAKi treatment of A2780-CP70 and OVCAR10-CP tumorspheres was accompanied by an increased percentage of G1 phase cells , decreased cyclin D1 expression , but not increased apoptosis ( Figure 6—figure supplement 2 ) . Both A2780-CP70 and OVCAR10-CP tumorspheres possessed increased ALDH activity compared to parental cells ( Figure 6—figure supplement 3 ) and this was dependent on FAK activity ( Figure 6E ) . FAKi selectively prevented A2780-CP70 and OVCAR10-CP methylcellulose colony formation ( Figure 6F ) but did not inhibit parental A2780 or OVCAR10 colony formation ( Figure 6—figure supplement 2 ) . To determine the effects of combinatorial FAKi and low dose CP treatments , OVCAR10-CP colony formation was evaluated in the presence of DMSO ( control ) , FAKi ( 1 µM ) , CP ( 1 µM ) , or FAKi and CP combination ( Figure 6G , I ) . Single agent FAKi reduced colony size ( Figure 6G ) , consistent with an inhibitory effect on tumorsphere proliferation . FAKi with CP prevented colony formation ( Figure 6H ) and this was associated with increased OVCAR10-CP apoptosis ( Figure 6I ) . Only the combination of FAKi with CP triggered increased A2780-CP70 apoptosis ( Figure 6I ) . These results support the notion that selection for platinum resistance can result in the acquired dependence on FAK activity for tumorsphere growth . Moreover , FAK inhibition in combination with CP can trigger CP-resistant tumorsphere apoptosis . DTomato plus luciferase-labeled A2780 or A2780-CP70 cells were orthotopically injected into mice to assess the combinatorial potential of FAKi ( VS-4718 ) and cisplatin plus paclitaxel ( CPT ) chemotherapy on paired CP-sensitive and CP-resistant tumors established in immune-deficient mice ( Figure 7 ) . A2780 tumor growth was insensitive to FAKi ( Figure 7—figure supplement 1 ) , consistent with limited FAKi effects on A2780 growth in vitro . As expected , CPT chemotherapy inhibited A2780 tumor growth , but did not affect the resistant A2780-CP70 tumors . FAKi with CPT did not yield additional anti-tumor effects on A2780 tumor growth . In dramatic contrast , in A2780-CP70 tumors , single-agent FAKi treatment reduced tumor mass approximately 40% compared to controls ( Figure 7A , D ) , and the combination of FAKi with CPT chemotherapy resulted in an even greater reduction in tumor growth ( Figure 7B–D ) . Interestingly , CPT treatment increased FAK Y397 phosphorylation ( Figure 7E , F ) and ALDH-1A1 staining ( Figure 7G ) in non-necrotic regions of A2780-CP70 tumors ( Figure 7—figure supplement 2 ) . Adding FAKi to CPT chemotherapy suppressed FAK Y397 phosphorylation , reduced ALDH-1A1 tumor staining , and increased tumor apoptosis in vivo ( Figure 7E , H ) . These results show that the combination of a FAK inhibitor with CP exhibits selective anti-tumor effects on CP-resistant A2780-CP70 tumors in vivo . To provide genetic support for the role of FAK in intrinsic CP resistance , CRISPR/Cas9 targeting was used to inactivate murine Ptk2 exon four in KMF cells ( Figure 8 ) and human PTK2 exon 3 of OVCAR3 cells ( Figure 8—figure supplement 1 ) to create FAK knockout ( KO ) cells . No difference in 2D adherent cell growth was detected and FAK KO clones were isolated . Sanger sequencing confirmed unique deletions/insertions predicted to terminate FAK protein translation in each of four Ptk2 alleles identified in KT3 and KT13 FAK KO clones ( Figure 8—figure supplement 2 ) . Exome sequencing of FAK KO clone KT13 ( 90% of exons sequenced at 100X ) detected only 165 unique variants , including the four Ptk2 alterations , indicating that CRISPR targeting was specific and that the FAK KO KT13 genome is similar to parental KMF cells ( Figure 8—source data 1 ) . CRISPR inactivated FAK but not expression of the FAK-related Pyk2 kinase ( Figure 8A ) . In KT3 and KT13 clones , total β-catenin protein levels were constitutively lower than KMF cells ( Figure 8A ) and this corresponded to decreased β-catenin transcriptional activity ( Figure 8B ) . When cultured in PromoCell tumorsphere media under anchorage-independent conditions , FAK KO cell viability remained high after 5 days ( Figure 8C ) . However , FAK KO cells exhibited sensitivity to CP-mediated cytotoxicity compared to parental KMF cells . Similar results were obtained comparing parental and FAK KO OVCAR3 cells ( Figure 8—figure supplement 1 ) . Importantly , phenotypes of decreased tumorsphere formation , ALDEFLUOR activity , and CP resistance in FAK KO OVCAR3 cells were rescued by GFP-FAK re-expression ( Figure 8—figure supplement 1 ) . These results connect FAK expression to CP resistance and CSC-associated phenotypes . To establish a genetic linkage with FAK activity , FAK KO KT13 cells were stably reconstituted with GFP-FAK wildtype ( WT ) or a catalytically-inactive ( K454R ) GFP-FAK point mutation ( Figure 8D ) ( Lim et al . , 2010 ) . In 3D anchorage-independent conditions , GFP-FAK-WT and GFP-FAK-R454 were equally expressed , but only GFP-FAK-WT was phosphorylated at Y397 ( Figure 8D ) . This result confirms that intrinsic FAK kinase activity facilitates FAK Y397 phosphorylation . To identify proteomic differences in an unbiased manner , lysates of KT13 FAK KO , FAK-WT , and FAK-R454 cells were analyzed by mass spectrometry ( Figure 8—source data 2 ) . Elevated levels of extracellular matrix ( collagen type six and laminin ) , surface receptors ( N-cadherin and Nectin-2 ) , β-catenin , and Wnt signaling targets ( GPC4 ) ( Sakane et al . , 2012 ) were present in FAK-WT compared to FAK KO cells . These differences were maintained in FAK-WT versus FAK-R454 cells ( Figure 8E ) . As FAK can regulate β-catenin levels in colon carcinoma cells ( Gao et al . , 2015 ) , the mass spectrometry results implicate intrinsic FAK activity in supporting Wnt-β-catenin signaling in KMF cells . To test whether stabilized β-catenin was sufficient to complement KMF FAK KO cell phenotypes , an activated β-catenin point-mutant ( ΔGSK ) lacking the regulatory GSK3β phosphorylation sites ( Barth et al . , 1999 ) was expressed in KT13 FAK KO cells ( Figure 8F ) . A series of assays were performed comparing KMF , FAK KO , FAK-WT , FAK-R454 , and FAK KO β-catenin ΔGSK expressing cells . In 3D conditions , FAK KO proliferation was less than KMF cells and this was rescued by FAK-WT , FAK-R454 , and β-catenin ΔGSK ( Figure 8G ) . Notably , FAK-R454 cells grew in 3D culture , whereas parental KMF cells treated with FAKi exhibit growth defects ( Figure 5 ) . In contrast , FAK activity was required for clustering of KMF cells into tumorspheres and this phenotype was also supported by β-catenin ΔGSK expression ( Figure 8H ) . FAK-WT restored total ALDEFLUOR activity , ALDH-1A2 , ALDH-1B1 , and Myc protein levels in FAK KO cells equivalent to parental KMF cells ( Figure 9A , C ) . Expression of FAK-WT and β-catenin ΔGSK but not FAK-R454 significantly enhanced FAK KO resistance to CP cytotoxicity in vitro ( Figure 9C ) . Together , these results link intrinsic FAK activity and β-catenin in supporting KMF CSC and intrinsic CP resistance phenotypes . Although oral FAKi administration can inhibit tumor growth in mice ( Sulzmaier et al . , 2014 ) , it remains unclear whether this is mediated by FAK inhibition within tumor , stroma , or multiple cell types . Parental KMF , FAK KO , and FAK-WT cells were labeled with a dual reporter ( luciferase and dTomato ) and injected within the intraperitoneal cavity of C57Bl/6 mice to test whether FAK is essential for tumor establishment . At Day 24 , luciferase imaging revealed significant KMF tumor burden whereas FAK KO tumor cells were only weakly detected ( Figure 9D—figure supplement 1 ) . At Day 28 , flow cytometry enumeration of dTomato-positive peritoneal cells revealed significantly fewer FAK KO compared KMF and FAK-WT tumor cells ( Figure 9E ) . In an independent experiment over 21 days , FAK KO and FAK-R454 cells did not grow in vivo as did FAK-WT tumors ( Figure 9F ) . Surprisingly , β-catenin ΔGSK also did not promote FAK KO tumor growth ( Figure 9F ) . This result contrasted with the rescue of FAK KO tumorsphere formation , ALDEFLUOR activity , and CP resistance in vitro by β-catenin ΔGSK expression ( Figure 8 ) . Importantly , these results reveal that intrinsic FAK activity is essential for KMF tumor establishment in mice . Moreover , β-catenin signaling was not sufficient to support KMF tumor growth in the absence of FAK . FAK controls various gene transcriptional networks ( Sulzmaier et al . , 2014; Serrels et al . , 2017 ) . As FAK KO cells are deficient in a number of different phenotypes , we performed RNA sequencing from KT13 FAK KO , FAK-WT , FAK-R454 , and β-catenin ΔGSK cells grown in PromoCell to determine FAK activity-dependent , -independent , and β-catenin-specific patterns of differential gene expression . Using an FPKM cutoff greater than one , 1040 mRNA transcripts were increased two-fold or more by FAK compared to FAK KO cells and significant after multiple testing correction ( Figure 10—source data 1 ) . By filtering out transcripts that were elevated in FAK-R454 versus FAK KO cells ( FAK activity-independent targets ) , 591 genes were identified as FAK activity-dependent and showed KEGG ( Kyoto Encyclopedia of Genes and Genomes ) pathway enrichment for MAPK Signaling , Cell Cycle , Axon Guidance , and DNA Replication . After unbiased filtering of the 591 FAK activity-dependent transcripts with those elevated by β-catenin ΔGSK , 241 shared transcripts were identified as co-regulated by FAK activity and β-catenin ( Figure 10A ) . KEGG enrichments were MAPK Signaling , Cell Cycle , Hippo Signaling , and Pluripotency . Last , the 241 shared FAK and β-catenin murine targets were filtered against genes elevated in HGSOC by querying TCGA . Notably , 135 targets matched , 77 exhibit frequent gains in 20% of tumors , and 19 genes were elevated in greater than 50% of HGSOC patient tumors ( Figure 10A ) . Myc was the most common gene target and immunoblotting of KMF , FAK KO , and FAK-WT lysates confirmed the regulation of Myc protein expression by FAK ( Figure 9B ) . Although Myc is a common target of β-catenin ( Sanchez-Vega et al . , 2018 ) , this result is surprising as both Myc and Ptk2 ( FAK ) DNA loci exhibit gains in KMF cells ( Table 1 ) . However , FAK knockout and FAK re-expression in human OVCAR3 cells also showed FAK-mediated regulation of Myc protein levels ( Figure 8—figure supplement 2 ) . Together , these results place FAK , β-catenin , and Myc within a common signaling pathway activated in ovarian cancer . KEGG pathway analyses of FAK activity and β-catenin supported targets in HGSOC reveal conservation in Pluripotency , Hippo Signaling , and Cell Cycle ( Figure 10A ) which include targets supporting platinum resistance and stemness phenotypes . Quantitative PCR using independent experimental samples verified at least twofold changes in genes linked to DNA repair ( Recq4l , Spc24 , Pkmyt1 , Dscc1 , and Rnf144b ) or pluripotency ( Wnt4 , Klf5 , Amigo2 , Fos , and Ndufa4l2 ) that were elevated by FAK-WT re-expression ( Figure 10B ) . Although our studies did not establish causality of mRNA changes with FAK phenotypes supporting CP resistance and pluripotency of KMF cells , these targets are part of the FAK activity- and β-catenin-regulated 135 genes elevated in HGSOC . KMF cells are a unique murine model with profound similarities to HGSOC and will be made available to the research community .
Platinum-resistant ovarian carcinomas have complex tumor genomes , few targetable mutations , and no effective treatments ( Patch et al . , 2015 ) . Gene breakage , gains , or losses are common drivers of tumor cell phenotypes . Using a new in vivo-evolved murine ovarian cancer model termed KMF - denoting gains in genes for Kras , Myc , and FAK – we demonstrate the functional significance of PTK2 ( FAK ) gains observed in HGSOC tumors . KMF cells exhibit more aggressive tumor growth , greater tumorsphere formation in vitro , elevated FAK Y397 phosphorylation , increased β-catenin and ALDH activities , and increased resistance to cisplatin-mediated cytotoxicity compared to parental ID8 cells . In both KMF and human OVCAR3 ovarian carcinoma cells , we identify tumorsphere-associated non-canonical FAK signaling as supporting CSC phenotypes and intrinsic cisplatin resistance . This context-dependent signaling is consistent with an oncogenic role for FAK activation in ovarian cancer . Although MYC amplification at 8q24 . 21 in HGSOC is associated with a poor prognosis ( Goode et al . , 2010 ) , less is known about PTK2 amplification at 8q24 . 3 . We show that over 70% of HGSOC patient tumors contain gains at both PTK2 and MYC loci , that PTK2 copy number parallels PTK2 mRNA and FAK protein increases , and that elevated PTK2 mRNA levels are associated with decreased patient disease-free survival . We identify a set of 36 genes associated with PTK2 gain predictive of HGSOC relapse . Additionally , we identify Myc as part of a set of 135 genes increased in murine KMF cells in a FAK kinase-dependent manner that also are highly expressed in HGSOC tumors . As FAK has also been proposed to function downstream of Wnt-Myc signaling in intestinal regeneration and tumorigenesis ( Ashton et al . , 2010 ) , the contribution of FAK in support of Wnt signaling may be mediated by multiple mechanisms ( Chen et al . , 2012; Chen et al . , 2018; Gao et al . , 2015; Kolev et al . , 2017 ) . Moreover , recent studies show that FAK and β-catenin over-expression cooperate to induce hepatocellular carcinoma ( HCC ) in mice ( Shang et al . , 2019 ) and that FAK promotes CSC-like phenotypes in HCC cells ( Fan et al . , 2019 ) . While platinum and taxane chemotherapy kills most ovarian tumor cells , we unexpectedly find that FAK activation is elevated in the residual tumor cells of patients undergoing chemotherapy , in mouse tumors , and in isolated ovarian carcinoma tumorspheres after cisplatin chemotherapy ( Figure 10C ) . Previous studies showed increased FAK Y397 phosphorylation during the processes of acquired CP resistance of cultured ovarian carcinoma cells ( Villedieu et al . , 2006 ) . This is consistent with studies linking chemotherapy to selective CSC survival ( Wiechert et al . , 2016 ) and we show that FAK inhibition compromises ALDH levels and CSC generation . Notably , platinum-resistant cells can acquire FAK dependence for growth . This dependence was manifest when culturing ovarian carcinoma cells as tumorspheres and this 3D selective phenotype has been observed in breast ( Tanjoni et al . , 2010 ) and squamous cell carcinoma models ( Serrels et al . , 2012 ) . In the A2780 and A2780-CP70 models , we found selective FAK dependence for growth in the CP resistant but not parental CP-sensitive cells . Via complementary approaches including pharmacological inhibition , FAK knockout , and FAK re-expression , we show that FAK signaling sustains both intrinsic and acquired resistance to cisplatin chemotherapy in part via promoting β-catenin activation ( Figure 10C ) . Notably , a FAK to β-catenin signaling linkage functions as an adaptive chemotherapy resistance pathway in BRAF mutated colon cancer ( Chen et al . , 2018 ) . Stabilized β-catenin ΔGSK expression in KMF FAK KO cells supported canonical Wnt target genes , yet β-catenin ΔGSK was unexpectedly insufficient to rescue FAK KO growth as tumors . This may be due to weak oncogenic activity of the β-catenin ΔGSK construct ( Barth et al . , 1999 ) or due to a supporting requirement for FAK . Additionally , the FAK-associated protein Rgnef is also essential for KMF tumorsphere growth and protection from oxidative stress ( Kleinschmidt et al . , 2019 ) . We show that FAK expression and intrinsic activity are essential for KMF tumor growth and that elevated FAK activity and Y397 phosphorylation is an acquired and targetable cellular adaptation of cisplatin resistance in HGSOC . In cell culture , cisplatin resistant cells acquired dependence on FAK activity to maintain proliferation as 3D tumorspheres without alterations in 2D growth . Single agent pharmacological FAK inhibition did not promote apoptosis of platinum-resistant ovarian cells . Rather , the combination of FAK inhibition ( genomic and pharmacological ) with cisplatin-triggered apoptosis of platinum-resistant cells as tumorspheres in vitro and prevented the growth of platinum-resistant tumors in mice . To this end , a phase I/II clinical trial for recurrent platinum-resistant ovarian cancer termed ROCKIF ( Re-sensitization of platinum-resistant Ovarian Cancer by Kinase Inhibition of FAK , NCT03287271 ) has been initiated . ROCKIF will investigate whether the small molecule FAK inhibitor defactinib , in combination with carboplatin and paclitaxel chemotherapy , can provide benefit for this difficult to treat patient population .
The dTomato with luciferase lentiviral vector , pUltra-Chili-Luc , was a gift from Malcolm Moore ( Addgene #48688 ) . The lentiviral vector MSCV-β-catenin ( ΔGSK-KT3 ) -IRES-GFP was a gift from Tannishtha Reya ( Addgene #14717 ) . The CRISPR/Cas9 plasmid pSpCas9n ( BB ) −2A-Puro was a gift from Feng Zhang ( Addgene #48141 ) . M50 Super 8x TOPFlash was a gift from Randall Moon ( Addgene #12456 ) . Lentiviral vectors for murine GFP-FAK and GFP-FAK R454 in pCDH-CMV-MCS-Puro ( System Biosciences ) were used as described ( Chen et al . , 2012 ) . FAKi ( VS-4718 ) was from Verastem Inc FAKi , cisplatin ( Enzo Life Sciences ) or staurosporine ( Calbiochem ) were dissolved in DMSO for in vitro studies . VS-4718 was suspended in 0 . 5% carboxymethyl cellulose ( Sigma ) and 0 . 1% Tween 80 ( Sigma ) in sterile water and administered twice daily by oral gavage for tumor experiments . For mouse experiments , cisplatin and paclitaxel ( APP Pharmaceuticals ) were obtained from the Moores Cancer Center Pharmacy . Human ovarian carcinoma A2780 , A2780-CP70 , and OVCAR10 cell lines were from Denise Connolly ( Fox Chase Cancer Center , PA ) . NIH OVCAR3 cells were from the Division of Cancer Treatment and Diagnosis Tumor Repository , National Cancer Institute ( Frederick , MD ) , murine ovarian ID8 cells were from Katherine Roby ( University of Kansas Medical Center ) , and KMF cells were isolated from peritoneal ascites of ID8-injected C57Bl6 tumor-bearing mice as described ( Ward et al . , 2013 ) . Intermittent CP exposure ( 10 µM for 24 hr ) , cell recovery ( 7 days ) , and repeated exposure-recovery ( five times ) was used to generate OVCAR10-CP cells and maintain A2780-CP70 cells . All cells were from validated sources and were evaluated for mycoplasma contamination . OVCAR3 FAK knockout cells were created using CRISPR/Cas9 targeting . pSpCas9n ( BB ) −2A-Puro was used to deliver guide RNAs ( ACTGGTATGGAACGTTCTCC and TGAGTCTTAGTACTCGAATT ) targeting exon 3 of human PTK2 . Transfected cells were enriched by puromycin ( 1 µg/ml , 3 days ) and clones selected by dilution . Loss of FAK expression was verified by immunoblotting . DNA sequencing was used to verify insertions/deletions introducing stop codons in PTK2 exon 3 . FAK re-expressing cells were generated by lentiviral transduction of OVCAR3 FAK KO clone AB21 , puromycin selection , enrichment by flow cytometry , and GFP-FAK protein expression verified by immunoblotting . KMF FAK KO cells were generated by CRISPR/Cas9 targeting . pSpCas9n ( BB ) −2A-Puro was used to deliver two independent guide RNAs ( ACTTACATGGTAGCTCGCGG and CACTCCCACAGCCATCCTAT ) targeting exon 4 of murine Ptk2 . Transfected cells were enriched by puromycin selection ( 3 . 5 µg/mL for 24 hr ) and clones selected by dilution . Loss of FAK expression was verified by immunoblotting . DNA sequencing was used to verify insertions/deletions introducing stop codons in murine Ptk2 exon 4 . GFP-FAK-WT , GFP-FAK-R454 , and ΔGSK β-catenin ( GFP expressed independently ) were generated by lentiviral ( for FAK ) or retroviral ( for β-catenin ΔGSK ) transduction , puromycin or hygromycin selection , enrichment by flow cytometry , and protein expression verified by immunoblotting . For adherent 2D growth , cells were maintained in DMEM ( OVCAR10 , OVCAR10-CP , ID8 , and KMF ) or RPMI 1640 ( A2780 , A2780-CP70 , and OVCAR3 ) supplemented with 10% fetal bovine serum ( FBS , Omega Scientific ) , 1 mM non-essential amino acids , 100 U/ml penicillin , and 100 µg/ml streptomycin on cell culture-treated plastic plates ( Costar ) . For growth as tumorspheres , cells were seeded in poly-hydroxyethyl methacrylic acid ( poly-HEMA ) coated Costar plates ( non-adherent ) in serum-free CSC media ( 3D Tumorsphere Medium XF , PromoCell GmbH ) at cell dilutions recommended by the manufacturer . Prior to tumor initiation experiments , KMF , A2780 , and A2780-CP70 cells were transduced with a lentiviral vector expressing dTomato and luciferase ( pUltra-Chili-Luc ) or mCherry ( pCDH-CMV-MSCI ) and enriched by fluorescence sorting . For 2D growth , cells were seeded ( 3 × 105 cells per well ) in tissue culture-treated 6-well plates ( Costar ) . At the indicated time , cells were enumerated and stained with Trypan blue ( ViCell XR , Beckman ) . Alternatively , cell metabolic activity was measured by a colorimetric XTT assay ( Sigma ) . For 3D tumorspheres , cells were seeded at 10 , 000 cells/ml equivalent in poly-HEMA-coated 6- , 24- , or 96-well plates ( Costar ) for 5 days . At the indicated times , 3D tumorspheres were phase-contrast imaged ( Olympus CKX41 ) , enumerated ( ViCell XR ) , or collected by centrifugation . Spheroid size was determined using Image J ( NIH ) . Alternatively , cell metabolic activity was measured by a colorimetric XTT assay ( Sigma ) . For methylcellulose colony formation , cells were suspended in 1% methylcellulose diluted in 2D growth media , 104 plated in six-well poly-HEMA-coated plates , and colony formation analyzed after 21 days . Cells from methylcellulose colonies were collected by dilution-dispersion in PBS , centrifugation at 400 xg , and washed in PBS prior to enumeration or cell lysis . Cells were used at passage 10 to 35 and mycoplasma testing was performed every 3 months . For all experiments , triplicate experimental points were evaluated ( technical replicates ) and experiments were repeated at least two times ( biological replicates ) . De-identified human ovarian cancer tissue specimens from consented patients were obtained from the Fox Chase Cancer Center ( FCCC ) Biosample Repository Facility ( BRF ) under Institutional Review Board ( IRB ) approved protocols ( IRB 11–866 and IRB 08–851 ) . FCCC staff queried the BRF sample database to identify participants that received carboplatin and paclitaxel neoadjuvant chemotherapy . Biopsy specimens were obtained from FCCC Surgical Pathology , sectioned , H and E stained , and reviewed by a board-certified pathologist . FFPE blocks from the biopsy and the corresponding surgical resection blocks banked by the BRF were cut to obtain one H and E stained slide and six additional unstained sections . One section each from pre-treatment biopsy and post-neoadjuvant treatment surgical resection specimen was stained for Pax8 by the FCCC Histopathology Facility . The remainder of unstained slides were sent to UCSD for additional staining performed under UCSD IRB-approved protocol ( IRB 110805 ) . Cells were collected as a single cell suspension by limited trypsin treatment , fixed in 70% ethanol and stored at −20°C overnight . Cells were incubated in 100 μl of PBS containing DNAse-free RNAse ( 100 μg/mL , Qiagen ) . After 45 min , propidium iodide ( 10 μg/mL ) was added prior to flow cytometry and analyzed using FlowJo ( v9 . 5 . 1 ) and ModFit LT ( Verity Software House ) software . Integrated reporter: 293 T cell transfection with a β-catenin DNA binding reporter ( 7X-TCF repeat sequence AGATCAAAGGgg ) driving eGFP ( 7TGP , Addgene #24305 , gift from Roel Nusse ) was packaged with psPAX2 ( gift from Didier Trono , Addgene #12260 ) and pMD2 . G ( gift from Didier Trono , Addgene #12259 ) using Fugene HD ( Thermo ) according to the manufacturer’s instructions . Media was replaced after 6 hr , conditioned media with virus collected ( 36 hr ) , centrifuged ( 500 g ) , and target cells infected for 24 hr with polybrene ( 8 µg/ml Sigma ) . Transduced cells were selected by puromycin ( 1 µg/ml , 5 days ) . Proliferating cells were seeded in 3D tumorsphere growth conditions ( PromoCell ) , DMSO or GSK3β inhibitor ( CHIR99021 , Sigma , 3 µM ) added after 24 hr , and cells evaluated by flow cytometry ( BD FACSCelesta ) after 5 days . For transient transfection of a β-catenin TOPFlash reporter , 3 . 5e4 cells were seeded in triplicate in 24-well plates and co-transfected with M50 Super 8x TOPFlash with firefly and Renilla luciferase expression vectors using Fugene HD ( Promega ) . After 5 hr , media was replaced and cells treated with DMSO or 3 µM GSKi ( CHIR99021 ) . After 24 hr , cells were evaluated using the Dual‐Luciferase Reporter Assay system ( Promega , E1910 ) and a plate luminometer ( Dynex Tech . , VA ) . For annexin V and 7-AAD staining , cells were cultured in adherent or non-adherent conditions as described above for 24 hr or 5 days , respectively , with increasing cisplatin treatment ( 0 to 100 μM ) . Cells were dissociated using limited trypsin treatment , washed by centrifugation , suspended in annexin-V binding buffer ( 10 mM Hepes , 140 mM NaCl and 2 . 5 mM CaCl2 ) and incubated with allophycocyanin-conjugated Annexin V ( eBioscience ) and 7-aminoactinomycin D ( 7-AAD ) for 10 min at RT prior to analysis on a FACSCalibur flow cytometer ( BD Biosciences ) . Post-acquisition analyses were performed using CellQuest Pro ( BD Biosciences ) or FlowJo software . For AlamarBlue ( Life Technologies ) assays , cells were cultured in 96-well poly-HEMA-coated plates as above for 5 days . AlamarBlue reagent ( Life Technologies ) was added to each sample and incubated at 37°C at 5% CO2 for 24 hr . Viability was analyzed by resorufin production via absorbance at 570/600 nm using a Synergy HTX spectrophotometer ( BioTek Instruments ) . Cells ( 10 , 000 in 90 µl ) were plated in tissue culture-treated 96-well plates ( Costar ) . At 24 hr , increasing concentrations of cisplatin were added in growth media ( 10 µl ) , and the number of viable cells determined at 72 hr using the CellTiter 96 AQueous One Solution Cell Proliferation Assay ( Promega ) . Measurement of cell resistance to CP-induced cytotoxicity were performed by colorimetric XTT cell staining ( Sigma ) . In 2D culture , OVCAR10-CP and OVCAR10 cells exhibit 10 . 9 ± 2 . 2 µM and 1 . 4 ± 0 . 6 µM EC50 values to CP treatment , respectively . A2780-CP70 and A2780 cells exhibit 62 . 2 ± 8 . 7 µM and 5 . 6 ± 3 . 2 µM EC50 values to CP treatment , respectively . EC50 values were calculated using Prism ( v7 , GraphPad ) . The ALDEFLUOR fluorescent assay ( Stemcell Technologies ) was used to measure cell-associated ALDH activity . Cells were cultured as tumorspheres , treated with the indication concentrations of cisplatin or VS-4718 for 5 days , collected by centrifugation , dissociated by trypsinization , resuspended in Aldefluor assay buffer containing ALDH substrate ( BODIPY-aminoacetaldehyde ) , and incubated for 45 min at 37°C with or without the ALDH inhibitor diethylamino-benzaldehyde ( DEAB ) . AldeRed substrate ( EMD Millipore ) was used with cells expressing GFP . Individual gates were used to determine the percentage of ALDEFLUOR-positive cells per experimental point relative to DEAB-inhibitor treated controls . For analysis of ALDH activity in ascites-associated cells , pooled isolates from peritoneal washings of each experimental group were dissociated by trypsinization , treated with red blood cell lysis buffer ( Biolegend ) , and processed as described above . Total RNAs were extracted using PureLink RNA Mini Kit ( Thermo ) and cDNA prepared using the High-Capacity cDNA Reverse Transcription Kit ( Thermo ) from 1 µg total RNA . Target transcripts were amplified using a LightCycler 480 ( Roche Applied Science ) , Premix Ex Taq probe qPCR Kit , iTaq Universal SYBR Green Supermix ( Bio-Rad ) with cDNA template and primers Table 2 . according to manufacturer instructions . Target gene expression was normalized to 60S ribosomal protein L19 ( RPL19 ) as a housekeeping gene control . Transcript levels were calculated using the ΔΔCT ( cycle threshold ) method . Protein extracts of cells were prepared using a lysis buffer containing ( 25 mM HEPES , pH 7 . 5 , 150 mM NaCl , 10% glycerol , 10 mM MgCl2 , 1 mM EDTA , 10 mM NaF , 1 mM Na3VO4 ) with 1% NP-40 , 0 . 25% sodium deoxycholate , 0 . 1% SDS and protease inhibitors ( Roche Diagnostics ) . Tumors were homogenized in lysis buffer without sodium deoxycholate using Precellys24 ( Bertin Instruments ) bead disruption . Total protein levels in lysates were determined through a bicinchoninic acid assay ( Pierce ) , proteins were resolved by SDS-PAGE ( NuPAGE 4–12% Tris-Bis gels , Thermo ) , and transferred to polyvinylidene difluoride membranes ( Immobilon-FL , Millipore ) for immunoblotting . Levels of protein expression and/or phosphorylation were detected with specific primary antibodies and IRDye 680 goat anti-mouse and IRDye 800 goat anti-rabbit secondary antibodies . Protein bands were visualized and quantified using the Odyssey Infrared Imaging System ( Li-Cor Biosciences ) . Alternatively , HRP-conjugated secondary antibodies were visualized by chemiluminescence detection ( ChemiDoc , BioRad ) . All animal experiments were performed in accordance with The Association for Assessment and Accreditation for Laboratory Animal Care guidelines and approved by the UCSD Institutional Animal Care and Use Committee ( S07331 ) . A2780 or A2780-CP70 tumor growth was evaluated by IP injection of 4 million pChili-Luciferase-labeled cells mixed with Matrigel into 9-week-old female NOD SCID gamma mice ( Jackson Laboratory ) . IVIS imaging ( Day 4 , 11 , 18 , and 23 ) was used to monitor tumor growth . On Day 5 , mice were randomized to a control ( saline injection ) ; chemotherapy group ( CPT ) receiving IP injection of cisplatin ( 3 mg/kg ) plus paclitaxel ( 2 mg/kg ) at Day 5 , 12 , and 19; VS-4718 FAK inhibitor ( 100 mg/kg ) via oral gavage twice daily ( BID ) ; or CPT plus FAK inhibitor treatment . At Day 24 , mice were euthanized , omental tumors excised , and remaining peritoneal metastatic sites quantified by dTomato fluorescence using an OV100 Small Animal Imaging Station ( Olympus ) and ImageJ software . For KMF intraperitoneal tumor growth , cells were transduced with a lentiviral vector expressing dTomato and luciferase ( pUltra-Chili-Luc ) and were enriched by FACS . Cells were mixed with PBS + 50% Matrigel ( Growth factor reduced , Corning ) for a final concentration of 4 × 106 cells per 200 μL for injection in 10-week-old C57Bl/6N mice ( Charles River ) . Tumor growth was monitored via luciferase bioluminescent imaging ( IVIS , Perkin Elmer ) . At the indicated times , ascites-associated cells were recovered by peritoneal washings by injection and immediate removal of PBS ( 5 ml ) , followed by erythrocyte lysis ( RBC lysis buffer , eBioscience ) , Accutase ( Corning ) treatment for cell dissociation , total cell enumeration ( ViCell XR , Beckman ) and trypan blue staining ( viability >95% ) . Flow cytometry ( BD LSRFortessa ) was used to identify dTomato+ and CD45-negative ( rat anti-mouse CD45 , clone 30-F11 , BD Biosciences ) tumor cells . Exome sequencing was performed by Novogene ( Beijing , China ) , using genomic DNA ( 1 µg ) isolated from ID8 or KMF cells . Genomic DNA was sheared into 180–280 bp fragments using a Covaris Sonicator ( Covaris ) . Exome enrichment and sequencing libraries were generated using Agilent SureSelect Mouse All Exon kit ( Agilent Technologies ) following manufacturer’s recommendations . Each exome was sequenced using a 150 bp paired-end protocol on the Illumina HiSeq platform , generating 47M reads for the ID8 sample and 61M reads for the KMF sample . ( https://software . broadinstitute . org/gatk/best-practices ) . Reads were aligned with BWA MEM 0 . 7 . 12 ( Li and Durbin , 2009 ) to mouse genome GRCm38_68 . Variants were called with GATK 3 . 4 according to the Broad Institute’s best practices ( https://software . broadinstitute . org/gatk/best-practices ) ( McKenna et al . , 2010 ) . Processing after alignment was carried out with SAMtools v . 1 . 1 ( Li and Durbin , 2009 ) . Variants were annotated with ANNOVAR ( Wang et al . , 2010 ) . Copy number variants were called from the same alignments with CNVkit ( Talevich et al . , 2016 ) , visualized in the Integrative Genomics Viewer ( Robinson et al . , 2011 ) using standard parameters , with ID8 as normal and KMF as tumor samples . Ninety percent of exons were sequenced at 100X . Total RNA was isolated from cells growing in suspension using PureLink RNA Mini Kit ( Thermo Fisher ) . Three independent samples of RNA were isolated from ID8 or KMF cells grown in 3D PromoCell XF media as tumorsperes for 5 days at various cell passages . RNA sequencing was performed by Novogene ( Beijing , China ) . Three replicate RNA samples were obtained from KT13 FAK KO , GFP-FAK WT , GFP-FAK R454 ( kinase-inactive ) , and FAK KO expressing a ΔGSK β-catenin . RNA library preparation was performed using NEB Next Ultra RNA Library Prep Kit ( New England Biolabs ) . Each transcriptome was sequenced using a 150 bp paired-end protocol on the Illumina HiSeq platform . At least 60 million clean reads were generated per sample . Reads were mapped ( >90% ) to the reference genome using TopHat2 ( Kim et al . , 2013 ) . Novogene analyses used ClusterProfiler software for enrichment analysis , including GO Enrichment , DO Enrichment , KEGG and Reactome database enrichment to analyze and visualize functional profiles of genomic coordinates , genes and gene clusters . Novogene performed differential expression analysis of two conditions/groups by using the DESeq2 R package . Clustering and grouping analyses used transcripts with FPKM values > 1 and an adjusted p value < 0 . 05 . Each dataset was subject to Gene Set Enrichment Analysis ( GSEA ) and The Molecular Signatures Database ( MSigDB ) analysis . The murine FAK activity and β-catenin targets were compared with the total list of all genes gained in HGSOC by Genomic Identification of Significant Targets in Cancer ( GISTIC , ov_tcga ) . Survival analysis was performed using a database of ovarian cancer samples ( Pénzváltó et al . , 2014 ) . The TCGA dataset was used to link copy number gains to gene expression ( Cancer Genome Atlas Research Network , 2011 ) . Samples with copy number gains were designated into one cohort and all remaining samples were designated into a second cohort . Gene expression was compared between cohorts using a non-parametric Mann-Whitney test . Genes with a fold change over two and a p-value below 1E-04 were accepted as statistically significant . The mean expression of all significant genes was computed for each sample and was used in subsequent analyses for the selected gene . Cox proportional hazards regression was performed for relapse-free survival and for overall survival . Correlation between mRNA expression and survival was assessed using the Kaplan-Meier plotter ( Gyorffy et al . , 2012 ) for PTK2 mRNA levels in 1435 annotated ovarian cancer patient samples . Selections were: relapse-free survival , split patients by median , stage ( all ) , histology ( serous ) , grade ( all ) , debulk ( all ) , and chemotherapy treatments ( all ) . ECM-enriched protein extracts from tumorsphere cultures in PromoCell were prepared by trypsin digestion as described ( Ojalill et al . , 2018 ) . Peptides were separated by a nanoflow HPLC system ( Easy-nLC1000 , Thermo ) coupled to a Q Exactive Hybrid Quadrupole-Orbitrap Mass Spectrometer ( Thermo ) . A full MS ( mass spectrum ) scan over the mass-to-charge ( m/z ) range of 300–2000 with a resolution of 140 , 000 followed by data-dependent acquisition of with an isolation window of 2 . 0 m/z and a dynamic exclusion time of 20 s was performed . The top 10 ions were fragmented by higher energy collisional dissociation ( HCD ) with a normalized collision energy of 27 and scanned over the m/z range of 200–2000 with a resolution of 17 , 500 . After the MS2 scan for each of the top 10 ions had been obtained , a new full MS scan was acquired and the process repeated until the end of the 70 min run . Three repeated runs per sample were performed . Tandem mass spectra were searched using MaxQuant software ( v1 . 5 . 2 . 8 ) against reviewed ( SwissProt ) mouse sequences of UniProtKB release 2018_08 . Peptide-spectrum-match- and protein-level false discovery rates were set at 0 . 01 . Carbamidomethyl ( C ) , as a fixed modification , and oxidation ( M , P , K ) as dynamic modifications were included . A maximum of two missed cleavages was allowed . The LC-MS profiles were aligned , and the identifications were transferred to non-sequenced or non-identified MS features in other LC-MS runs ( matching between runs ) . The extracted ion intensities of all peptides matching to the same protein from the three technical replicates were summed . Proteins were determined as detected in sample identification was derived from at least two unique peptide identifications . Contaminant proteins ( according to the contaminants listed in MaxQuant ) , reverse identifications , and identifications only by site were removed . Only the proteins containing ‘cell membrane’ , ‘plasma membrane’ , ‘cell surface’ , ‘extracellular matrix’ or ‘secreted’ in the cellular component gene ontology or in the subcellular location definition in the UniProt database were included in the final list . Samples were normalized by sum of protein intensities . Mouse tumors were divided into thirds and either processed for protein lysates , fixed in formalin , or frozen in optimal cutting temperature compound . For immunohistochemical staining , paraffin-embedded tumors were sectioned , mounted onto glass slides , deparaffinized , rehydrated , processed for antigen retrieval , and peroxidase quenched as described ( Tancioni et al . , 2014 ) . Tissues were blocked ( PBS with 1% BSA , and 0 . 1% Triton X-100 ) for 45 min at room temperature and incubated with anti-PAX8 ( 1:200 ) , anti-FAK ( 1:200 ) , anti-Ki67 ( 1:500 ) , anti-active β-catenin ( 1:800 ) or anti-pY397 FAK ( 1:100 ) in blocking buffer overnight . FAK pY397 antibodies were pre-incubated with 200-fold molar excess of FAK pY397 peptide ( Abcam ) for 12 hr at RT prior to use in IHC staining . Processing with biotinylated goat-anti-rabbit or goat-anti-mouse IgG , Vectastain ABC Elite , and diaminobenzidine were used to visualize antibody binding . Slides were counterstained with hematoxylin . Colon or breast carcinoma tumor samples were used as controls for active β-catenin staining . High-resolution digital scans were acquired ( Aperio CS2 scanner ) using Image Scope software ( Leica Biosystems ) . Images were also acquired using an upright microscope ( Olympus BX43 ) with a color camera ( Olympus SC100 ) . A board-certified pathologist evaluated H and E , Pax8 , pY397 FAK , or Ki67 stained images of patient tumor samples in a blinded manner . Quantification was performed using Aperio Image Analysis software ( v12 . 3 . 0 . 5056 ) using the positive pixel count ( v9 ) algorithm . Pax8-positive regions were identified and then these regions were manually-transposed onto images from FAK pY397-stained serial section slides . Average intensity ( I-Avg ) values were obtained and percent FAK pY397 was calculated . Frozen tumors were thin sectioned ( 7 µm ) using a cryostat ( Leica ) , mounted onto glass slides , fixed with acetone ( or with 4% paraformaldehyde ) for 10 min , permeabilized ( PBS with 0 . 1% Triton ) for 1 min , and blocked ( PBS with 8% goat serum ) for 2 hr at room temperature . Sections were incubated in anti-ALDH1A1 ( 1:100 ) or anti-pY397 FAK ( 1:100 ) in PBS with 2% goat serum overnight . Antibody binding was detected with goat anti-rabbit conjugated with Alexa Fluor-488 . Cell nuclei were visualized using Hoechst 33342 stain ( Thermo ) . Images were sequentially captured at 20X magnification ( UPLFL objective , 1 . 3 NA; Olympus ) using a monochrome charge-coupled camera ( ORCA ER; Hamamatsu ) , an inverted microscope ( IX81; Olympus ) , and Slidebook software ( Intelligent Imaging ) . Images were pseudo-colored , overlaid , merged using Photoshop ( Adobe ) , and quantified using Image J . Statistical difference between groups was determined using one-way or two-way ANOVA with Tukey , Bonferroni’s or Fisher’s LSD post-hoc analysis . Differences between pairs of data were determined using an unpaired two-tailed Student’s t test . For the IHC analysis the differences between pairs of data were calculated using a paired two-tailed Student’s t test . All statistical analyses were performed using Prism ( GraphPad Software , v7 ) . p-Values of <0 . 05 were considered significant . | Ovarian cancer is one of the deadliest types of cancer in women . There are two main reasons for the aggressiveness of this cancer . First , ovarian cancer cells can spread to other parts of a woman’s body before she has been diagnosed , where the cells grow as tiny clumps or spheres of tumor cells , also called tumorspheres . Second , in the majority of patients , some ovarian cancer cells will develop resistance to the chemotherapy used . It is not clear exactly how these tumor cells become resistant to therapy . One way in which cells could do this is by gaining extra copies of genes that remove toxic substances or repair DNA , which help them withstand the therapy . Here , Osterman , Ozmadenci , Keinschmidt , Taylor , Barrie , Jiang , Bean , Sulzmaier et al . set up a new experimental method to study how some ovarian cancer cells resist chemotherapy . Comparing ovarian cancer cells from mice at early and late stages of the disease showed that the later-stage , more aggressive cells had more genetic changes . One of these changes affected the gene for a protein called FAK , which was found to have more copies than normal . The FAK protein is an enzyme that helps cancer cells move around . In cells from mice with late-stage cancer , FAK was over-active and present at high levels . When these cells grew as tumorspheres , the tumors were more resistant to chemotherapy than their early-stage counterparts . In patients who have received chemotherapy , surviving tumor cells also exhibit high levels of FAK activity . Human ovarian cancer cells that are resistant to chemotherapy can be grown into tumors in mice , where they retain their resistance to chemotherapy . However , if chemotherapy is combined with a drug that targets the FAK enzyme , the tumors shrink . This experiment highlights a possible weak spot of these tumor cells . To understand how FAK makes ovarian cancer cells resistant to chemotherapy , Osterman et al . deleted the gene for FAK from the cells and then looked at how this changed the levels of activation of different genes . They found that , in addition to its effects on cell movement , FAK also activated a group of genes that increase resistance to chemotherapy and repair damaged DNA . This better understanding of how ovarian cancer cells resist chemotherapy could lead to new therapies . In particular , there is now a clinical trial for women with chemo-resistant ovarian cancer in which standard chemotherapy is combined with an inhibitor of the FAK protein . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"cancer",
"biology"
] | 2019 | FAK activity sustains intrinsic and acquired ovarian cancer resistance to platinum chemotherapy |
Learning about temporal structure is adaptive because it enables the generation of expectations . We examined how the brain uses experience in structured environments to anticipate upcoming events . During fMRI ( functional magnetic resonance imaging ) , individuals watched a 90 s movie clip six times . Using a hidden Markov model applied to searchlights across the whole brain , we identified temporal shifts between activity patterns evoked by the first vs . repeated viewings of the movie clip . In many regions throughout the cortex , neural activity patterns for repeated viewings shifted to precede those of initial viewing by up to 15 s . This anticipation varied hierarchically in a posterior ( less anticipation ) to anterior ( more anticipation ) fashion . We also identified specific regions in which the timing of the brain’s event boundaries was related to those of human-labeled event boundaries , with the timing of this relationship shifting on repeated viewings . With repeated viewing , the brain’s event boundaries came to precede human-annotated boundaries by 1–4 s on average . Together , these results demonstrate a hierarchy of anticipatory signals in the human brain and link them to subjective experiences of events .
A primary function of the brain is to adaptively use past experience to generate expectations about events that are likely to occur in the future ( Clark , 2013; Friston , 2005 ) . Indeed , anticipation and prediction are ubiquitous in the brain , spanning systems that support sensation , action , memory , motivation , and language ( den Ouden et al . , 2010 ) . For example , the visual system takes advantage of the world’s relative stability over space and time to anticipate upcoming input ( de Lange et al . , 2018; Summerfield and Egner , 2009 ) . The majority of studies examining anticipatory signals , however , have tested anticipation based on memory for relatively simple associations between pairs of discrete stimuli , such as auditory tones , lines , dots , oriented gratings , or abstract objects ( e . g . , Alink et al . , 2010; Gavornik and Bear , 2014; Hindy et al . , 2016; Kok et al . , 2012; Kok et al . , 2014; Kok and Turk-Browne , 2018 ) . These studies have found anticipatory signals about a single upcoming stimulus in a variety of brain regions , from perceptual regions ( Kok et al . , 2012 ) to the medial temporal lobe ( Hindy et al . , 2016; Kok and Turk-Browne , 2018 ) . How does the brain use repeated experience in naturalistic environments to anticipate upcoming sequences of events that extend farther into the future ? Prior work has shown that the brain integrates information about the recent past over a hierarchy of timescales ( Aly et al . , 2018; Hasson et al . , 2015; Kurby and Zacks , 2008 ) . Lower-order areas primarily represent the current moment , whereas higher-order areas are sensitive to information from many seconds or even minutes into the past . Higher-order regions with longer timescales play a critical role in organizing perceptual input into semantically meaningful schematic representations ( Baldassano et al . , 2017; Baldassano et al . , 2018 ) . What is less clear is whether this hierarchy also exists in a prospective direction: as we move from lower-order perceptual systems into higher-order areas , do these regions exhibit different timescales of anticipation into the future ? We previously found that higher-order regions did exhibit anticipatory signals when individuals had prior knowledge of the general structure of a narrative ( Baldassano et al . , 2017 ) . But these individuals only had knowledge of information at relatively long timescales ( e . g . , the general sequence of events , and not moment-by-moment perceptual features ) , so we were unable to assess whether they could generate expectations across the timescale hierarchy . Here , we examine how the brain anticipates event boundaries in familiar sequences of actions . We used a naturalistic narrative stimulus ( a movie ) , in which regularities are present at multiple timescales . For example , upon second viewing of a movie , one can anticipate the next action to be taken in a given scene , the next character to appear , the next location that is visited , and the last scene of the movie . The presence of predictability at multiple timescales in the same stimulus enables us to identify varying timescales of anticipation in the brain that co-exist simultaneously . We hypothesized that the timescale of anticipation in the brain would vary continuously , with progressively higher-order regions ( e . g . , prefrontal cortex ) anticipating events that are further in the future compared to lower-order regions ( e . g . , visual cortex ) . To test this , we examined brain activity with functional magnetic resonance imaging ( fMRI ) while individuals watched a 90 s clip from the movie The Grand Budapest Hotel six times . To uncover anticipation in the brain , we used a searchlight approach in which , for each region of interest , we fit a hidden Markov model ( HMM ) to identify temporal shifts between multivariate activity patterns ( functionally hyperaligned across individuals using the shared response model [SRM] ) evoked by the first viewing of the movie clip compared to repeated viewings ( Figure 1 ) . This model assumes that the brain’s response to a structured narrative stimulus consists of a sequence of distinct , stable activity patterns that correspond to event structure in the narrative ( Baldassano et al . , 2017 ) . We could then identify , on a timepoint-by-timepoint basis , the extent to which viewers were activating event-specific brain activity patterns earlier in subsequent viewings of the movie , by drawing on their prior experience . Because the HMM infers a probability distribution over states , it is able to detect subtle shifts between viewings; activity patterns may reflect a combination of current and upcoming events , and the degree of anticipation can vary throughout the clip . We also compared the brain’s event boundaries ( identified by the HMM ) to subjective event boundary annotations made by a separate group of participants . This allowed us to test how the relationship between the brain’s events and subjective event boundaries changes with repeated viewings . Together , this approach allowed us to characterize the nature of hierarchical anticipatory signals in the brain and link them to behavioral measures of event perception .
To identify anticipatory signals in the brain , we examined TR-by-TR brain activity patterns during each of the six viewings of the movie clip . For each spherical searchlight within the brain volume , we fit an HMM jointly to all repetitions , to identify a sequence of event patterns common to all viewings and the timing of spatial pattern changes for each viewing . At each timepoint for each viewing , the HMM produced a probability distribution that describes the mixture of event patterns active at that timepoint . Computing the expected value of this distribution provides an index of how the brain transitions through event patterns on each viewing , allowing us to identify how this timing shifts within each region of the brain . Our analysis revealed temporal shifts in event patterns in many brain regions , including lateral occipital cortex , angular and supramarginal gyri , lateral and anterior temporal lobe , lateral and medial prefrontal cortex ( mPFC ) , and insular cortex ( Figure 2 ) . The magnitude of this shift varied along a posterior-to-anterior temporal hierarchy ( Spearman’s rho = 0 . 58 , p=0 . 0030 ) , with the most anterior regions in the temporal pole and prefrontal cortex showing shifts of up to 15 s on subsequent viewings compared to the first viewing . This hierarchy persisted even when computed on the unthresholded anticipation map including voxels that did not meet the threshold for statistical significance ( Spearman’s rho = 0 . 42 , p=0 . 0028; see Figure 2—figure supplement 1 ) . There were no significant correlations with the left-to-right axis ( rho = 0 . 06 , p=0 . 41 for thresholded map; rho = 0 . 12 , p=0 . 29 for unthresholded map ) or the inferior-to-superior axis ( rho = 0 . 07 , p=0 . 28 for thresholded map; rho = −0 . 11 , p=0 . 73 for unthresholded map ) . We obtained a similar map when comparing the first viewing to just the sixth viewing alone ( see Figure 2—figure supplement 2 ) . We compared how this hierarchy of anticipation timescales related to the intrinsic processing timescales in each region during the initial viewing of the movie clip . Identifying the optimal number of HMM events for each searchlight , we observed a timescale hierarchy similar to that described in previous work , with faster timescales in sensory regions and slower timescales in more anterior regions ( Figure 2—figure supplement 3a ) . Regions with longer intrinsic timescales also showed a greater degree of anticipation with repeated viewing ( Figure 2—figure supplement 3b ) . We also compared these results to those obtained by using a simple cross-correlation approach , testing for a fixed temporal offset between the responses to initial and repeated viewing . This approach did detect significant anticipation in some anterior regions , but was much less sensitive than the more flexible HMM fits , especially in posterior regions ( Figure 2—figure supplement 4 ) . Our data-driven method for identifying event structure in fMRI data does not make use of information about the content of the stimulus , leaving open the question of how the HMM-identified event boundaries correspond to subjective event transitions in the movie . One possibility is that the brain’s event boundaries could start well-aligned with event boundaries in the movie and then shift earlier ( indicating anticipation of upcoming stimulus content ) . Alternatively , they may initially lag behind stimulus boundaries ( reflecting a delayed response time on initial viewing ) and then shift to become better aligned with movie scene transitions on repeated viewings . Finally , both patterns may exist simultaneously in the brain , but in different brain regions . We asked human raters to identify event transitions in the stimulus , labeling each ‘meaningful segment’ of activity ( Figure 3 ) . To generate a hypothesis about the strength and timing of event shifts in the fMRI data , we convolved the distribution of boundary annotations with a hemodynamic response function ( HRF ) as shown in Figure 4 . We then explored alignment between these human-annotated event boundaries and the event boundaries extracted from the brain response to each viewing , as shown in Figure 1d . In each searchlight , we cross-correlated the brain-derived boundary timecourse with the event annotation timecourse to find the temporal offset that maximized this correlation . We found three clusters in the middle temporal gyrus ( MTG ) , fusiform gyrus ( FG ) , and superior temporal sulcus ( STS ) in which the optimal lag for the repeated viewings was significantly earlier than for the initial viewing , indicating that the relationship between the brain-derived HMM event boundaries and the human-annotated boundaries was changing with repeated viewings ( Figure 5 ) . The HMM boundaries on the first viewing were significantly later than the annotated boundaries in FG and STS , while the optimal lag did not significantly differ from 0 in MTG ( 95% confidence intervals for the optimal lag , in seconds: MTG = [−0 . 27 , 2 . 86]; FG = [0 . 14 , 1 . 99]; STS = [1 . 48 , 8 . 53] ) . The HMM boundaries on repeated viewings were significantly earlier than the annotated boundaries in all three regions ( 95% confidence intervals for the average optimal lag , in seconds: MTG = [−4 . 06 , –1 . 83]; FG = [−1 . 56 , –0 . 26]; STS = [−3 . 06 , –1 . 69] ) .
One region showing long-timescale anticipatory signals was the bilateral anterior insula . This region has been linked to anticipation of diverse categories of positive and negative outcomes ( Liu et al . , 2011 ) , including outcomes that will be experienced by other people ( Singer et al . , 2009 ) . The movie stimulus used in our experiment depicts an interview in which the protagonist is initially judged to have ‘zero’ experience but then ends up impressing the interviewer , allowing for anticipation of this unexpected social outcome only on repeat viewings . Other regions showing long timescales of anticipation include the medial prefrontal cortex ( mPFC ) , which tracks high-level narrative schemas ( Baldassano et al . , 2018 ) and has been proposed to play a general role in event prediction ( Alexander and Brown , 2014 ) , and lateral prefrontal cortex , including the inferior frontal gyrus , which processes structured sequences across multiple domains ( Uddén and Bahlmann , 2012 ) . We also observed shorter-timescale anticipation throughout lateral occipital and ventral temporal cortex , which , though primarily thought to process bottom-up visual information , also exhibits event-specific patterns during recall ( Chen et al . , 2017 ) . A top-down memory-driven signal could be responsible for driving anticipatory activation in these regions during repeated movie viewing ( Finnie et al . , 2021 ) . Future work incorporating eye-tracking measurements could determine whether anticipatory eye movements can account for the temporal shifts in these regions , or if this anticipation is separate from the representation of the current retinal input . We did not observe widespread anticipatory signals in primary sensory areas , although some prior fMRI studies have been able to observe such signals in early regions such as V1 ( Alink et al . , 2010; Ekman et al . , 2017; Hindy et al . , 2016; Kok et al . , 2012 ) . One possibility is that the rich , ongoing sensory input dominated relatively small anticipatory signals in these regions . Paradigms involving periods without any sensory input ( e . g . , occasionally removing the audiovisual movie from the screen during repeated viewings ) may be necessary to detect these subtle signals . Alternatively , ultra-fast fMRI sequences ( Ekman et al . , 2017 ) or alternative imaging modalities ( discussed below ) may be required to track anticipation at a subsecond scale . Previous work has identified cumulatively longer timescales up the cortical hierarchy but has primarily focused on representations of the past . Lerner et al . , 2011 demonstrated hierarchical cortical dynamics in participants who listened to variants of a 7 min narrative that was scrambled at different timescales ( e . g . , paragraphs , sentences , or words ) . Response reliability , measured as the correlation in BOLD activity timecourses across individuals , varied based on the timescale of scrambling , with higher-level brain regions responding consistently to only the more-intact narrative conditions . This led to the idea that higher-order brain regions contain larger ‘temporal receptive windows’ than lower-order areas , in that their activity at a given moment is influenced by relatively more of the past . Likewise , using intracranial EEG ( iEEG ) , Honey et al . , 2012 observed progressively longer temporal receptive windows in successive stages of the cortical hierarchy in participants who watched intact and scrambled versions of the movie Dog Day Afternoon . These findings can be described by the process memory framework ( Hasson et al . , 2015 ) , where hierarchical memory timescales process , represent , and support longer and longer units of information . We found that this hierarchy also exists in the prospective direction , with the degree of anticipatory temporal shifts increasing from posterior-to-anterior regions of the brain . Furthermore , regions with longer intrinsic processing timescales showed further-reaching anticipation . These results extend the process memory framework , suggesting that the timescales in these regions are relevant not only for online processing and memory , but also for future anticipation or simulation . Although prior work has uncovered anticipatory and predictive coding in the brain , most studies have examined fixed , shorter timescales of anticipation . Moreover , these shorter timescales have often been studied using simple , non-narrative stimuli such as objects moving across the screen , short visual sequences , and visual pattern completion tasks ( Alink et al . , 2010; Ekman et al . , 2017; Gavornik and Bear , 2014; Hindy et al . , 2016; Kok et al . , 2012 ) . Some studies have used dynamic movie stimuli , but anticipation was measured via correlations between initial and repeated viewing of a movie at a constant fixed lag of 2 s ( Richardson and Saxe , 2020 ) . Such an approach is not well suited to capturing dynamic levels of anticipation within and across brain regions . Research investigating longer timescales of anticipation , such as learning future state representations in a maze task , examined single timescales up to 30 s ahead in OFC-VMPFC regions ( Elliott Wimmer and Büchel , 2019 ) . Some studies that use narrative stimuli have examined specially constructed texts in order to manipulate predictions about upcoming sentences; for example , work by Kandylaki et al . , 2016 demonstrated that predictive processing of referents in narratives can be modulated by voice ( passive vs . active ) and causality ( high vs . low ) . Our results show that in a naturalistic setting , in which structure exists at many timescales , anticipation at multiple levels can occur in parallel across different brain regions . We found anticipation up to approximately 15 s into the future with our 90 s stimulus , but future work with stimuli of longer duration could uncover even longer timescales of anticipation , on the scale of minutes . Simultaneously maintaining expectations at varying timescales could allow for flexible behaviors , because different timescales of anticipation may be helpful for a variety of tasks and actions . Taking action to avoid immediate harm or danger would require shorter timescales of prediction , whereas cultivating social relationships demands predictions on longer timescales . These results are consistent with those of Baldassano et al . , 2017 , in which some participants listening to an audio narrative had advance knowledge of the high-level events of the story ( because they had previously watched a movie version of the narrative ) . Using a similar HMM approach as in this paper , the authors observed shifts in event boundaries in higher-level regions including angular gyrus , posterior medial cortex , and mPFC . In the current study , however , participants were repeatedly exposed to an identical movie stimulus , allowing them to generate expectations at a broad range of timescales , including the timescales of fast-changing low-level visual features . This novel approach allowed us to observe for the first time that anticipation occurs in both low- and high-level regions , with shorter-timescale anticipation in visual occipital regions and the furthest-reaching anticipatory signals in prefrontal cortex . Our model detects anticipation as temporal shifts in events , and though timepoints can reflect ‘mixed’ event assignments , it assumes that the underlying event patterns themselves ( Figure 1b ) are constant . This view of anticipation is complementary to other theories of predictive representations , in which event patterns themselves should change over time to incorporate future information . One example is the ‘successor representation’ model from the field of reinforcement learning , which describes a representation in which each state ( here , event representation ) comes to include features of future events , weighted by their likelihood of occurring and their distance into the future ( Dayan , 1993 ) . Successor representations can also be constructed at multiple scales ( by changing the relative weighting of events near vs . far in the future ) . Such multi-scale representations are useful for goal-directed prediction that require multiple stages of planning ( Momennejad and Howard , 2018; Brunec and Momennejad , 2019 ) . Future work could explore how these two different theories could be integrated to model both mixing of event patterns and temporal shifts in the activation of these event patterns . The current fMRI study is complementary to investigations of memory replay and anticipation that use MEG and iEEG . In an MEG study , Michelmann et al . , 2019 found fast , compressed replay of encoded events during recall , with the speed of replay varying across the event . Furthermore , an iEEG investigation found anticipatory signals in auditory cortex when individuals listened to the same story twice ( Michelmann et al . , 2020 ) . In another MEG study , Wimmer et al . , 2020 found compressed replay of previously encoded information . Replay was forward when participants were remembering what came after an event , and backward when participants were remembering what came before an event . The forward replay observed in the Wimmer et al . study may be similar to the anticipatory signals observed in the current study , although there was no explicit demand on memory retrieval in our paradigm . Thus , one possibility is that the anticipatory signals observed in MEG or iEEG are the same as those we observe in fMRI , except that they are necessarily sluggish and smoothed in time when measured via a hemodynamic response . This possibility is supported by fMRI work showing evidence for compressed anticipatory signals , albeit at a slower timescale relative to MEG ( Ekman et al . , 2017 ) . An alternative possibility is that the anticipatory signals measured in our study are fundamentally different from those captured via MEG or iEEG . That could explain why we failed to find widespread anticipatory signals in primary visual or primary auditory cortex: the anticipatory signals in those regions might have been too fast to be captured with fMRI , particularly when competing with incoming , dynamic perceptual input . Future studies that obtain fMRI and MEG or iEEG in participants watching the same movie would be informative in that regard . It is possible that fMRI may be particularly well suited for capturing relatively slow anticipation of stable events , as opposed to faster anticipatory signals relating to fast sub-events . Nevertheless , advances in fMRI analyses may allow the detection of very fast replay or anticipation , closing the gap between these methods and allowing more direct comparisons ( Wittkuhn and Schuck , 2021 ) . One limitation of the current work is the reliance on one movie clip . Movie clips of different durations might yield different results . For example , it is an open question whether the duration of anticipation scales with the length of the movie and playback speed or if the amount of anticipation is fixed ( Lerner et al . , 2014; Baumgarten et al . , 2021 ) . Furthermore , the content of the movie and how frequently event boundaries occur may change anticipation amounts . That said , anticipatory signals in naturalistic stimuli have been observed across multiple studies that use different movies and auditorily presented stories ( e . g . , Baldassano et al . , 2017; Michelmann et al . , 2020; also see Michelmann et al . , 2019; Elliott Wimmer and Büchel , 2019; Wimmer et al . , 2020 ) . Thus , it is likely that anticipatory hierarchies will also replicate across different stimuli . There may nevertheless be important differences across stimuli . For example , the specific regions that are involved in anticipation may vary depending on what the most salient features of a movie or narrative are ( e . g . , particular emotional states , actions , conversations , or perceptual information ) . The detection of varying timescales of anticipation in the brain can be applied to multiple domains and modalities of memory research . Future work could explore even shorter timescales using other neuroimaging modalities , or longer timescales using longer movies or narratives from TV series that span multiple episodes . Furthermore , the impact of top-down goals on the hierarchy of anticipation timescales could be explored by using different tasks that require different levels of anticipation , such as anticipating camera angle changes vs . location changes . Brain stimulation studies or studies of patients with brain lesions could also explore the extent to which anticipation in lower-level regions relies on feedback from higher-level regions ( Auksztulewicz and Friston , 2016; Kiebel et al . , 2008 ) . The increased use of naturalistic , dynamic stimuli in neuroscience , and the development of methods to analyze the resulting data , has opened many avenues for research exploring flexible , future-oriented behavior . Our results and analysis approach provide a new framework for studying how anticipatory signals are distributed throughout the cortex , modulated by prior memory , and adaptive for improving comprehension and behavior .
We used data collected by Aly et al . , 2018 . Thirty individuals ( 12 men , age: M = 23 . 0 years , SD = 4 . 2; education: M = 15 . 3 years , SD = 3 . 2; all right-handed ) watched movie clips from The Grand Budapest Hotel while undergoing fMRI . None of the participants reported previously seeing this movie . We analyzed data from the Intact condition , during which participants watched a continuous 90 s clip from the movie in its original temporal order . This clip was watched six times , interspersed with other video clips that are not considered here . This Intact clip depicts an interview scene between the protagonist and his future employer inside of the Grand Budapest Hotel . Stimuli and data are available on OpenNeuro: https://openneuro . org/datasets/ds001545/versions/1 . 1 . 1 . Data were acquired on a 3T Siemens Prisma scanner with a 64-channel head/neck coil using a multiband echo planar imaging ( EPI ) sequence ( repetition time = 1 . 5 s; echo time = 39 ms; flip angle = 50°; acceleration factor = 4; shift = 3; voxel size = 2 . 0 mm iso ) . T1-weighted structural images ( whole-brain high-resolution; 1 . 0 mm iso ) were acquired with an MPRAGE sequence . Field maps ( 40 oblique axial slices; 3 mm iso ) were collected to aid registration . The fMRI scan took place over three experimental runs , each of which contained two presentations of the Intact movie clip ( as well as other movie clips not considered here ) . The first three EPI volumes of each run were discarded to allow for T1 equilibration . Data preprocessing was carried out in FSL , and included brain extraction , motion correction , high-pass filtering ( max period = 140 s ) , spatial smoothing ( 3 mm FWHM Gaussian kernel ) , and registration to standard Montreal Neurological Institute ( MNI ) space . After preprocessing , the functional images for each run were divided into volumes that corresponded to each of the video clips presented within that run , and only the two Intact clips within each run are considered further . Finally , each voxel’s timecourse was z-scored to have zero mean and unit variance . Fourteen individuals ( nine men ) were asked to mark event boundaries corresponding to the same 90 s Intact clip from The Grand Budapest Hotel as shown to the fMRI participants . Each participant was asked to pause the clip at the end of a meaningful segment and to record the time and a brief title corresponding to the segment ( Figure 3 ) . Specifically , they were given the following instructions: The movie clip can be divided into meaningful segments . Record the times denoting when you feel like a meaningful segment has ended . Pause the clip at the end of the segment , write down the time in the spreadsheet , and provide a short , descriptive title . Try to record segments with as few viewings of the movie clip as possible; afterward , record the number of times you viewed the clip . Although participants were allowed to watch the clip multiple times , they were instructed to minimize and report the number of viewings needed to complete the task . No participant reported watching the clip more than three times . Group-averaged fMRI data were fit with the event segmentation model described by Baldassano et al . , 2017 . This HMM assumes that ( 1 ) events are a sequence of discrete states , ( 2 ) each event is represented in the brain by a unique spatial activity pattern , and ( 3 ) all viewings of the movie evoke the same sequence of activity patterns in the same order ( though possibly with different timings ) . We fit the HMM jointly to all six viewings . This fitting procedure involved simultaneously estimating a sequence of event activity patterns that were shared across viewings , and estimating the probability of belonging to each of these events for every timepoint in all six datasets . The model was fit with seven events; this number was chosen to match the approximate timescale of the semantic events in the narrative , matching the mean number of events annotated by human observers ( mean = 6 . 5 ) . After fitting the HMM , we obtain an event by timepoint matrix for each viewing , giving the probability that each timepoint belongs to each event . Note that because this assignment of timepoints to events is probabilistic , it is possible for the HMM to detect that the pattern of voxel activity at a timepoint reflects a mixture of multiple event patterns . This allows us to track subtle changes in the timecourse of how the brain is transitioning between events . We took the expectation over events at each timepoint , yielding curves showing the average event label at each timepoint for each viewing . To compute shifts in time between the first viewing and the average of repeated viewings , the area under the curve ( AUC ) was computed for each viewing . We then computed the amount of anticipation as the average AUC for repeated viewing ( viewings 2–6 ) minus the AUC for the first viewing . In a supplementary analysis , we compared the first viewing to the last viewing only . To convert to seconds , we divide by the vertical extent of the graph ( number of events minus 1 ) and multiplied by the repetition time ( 1 . 5 s ) . We then performed a one-tailed statistical test ( described below ) to determine whether this difference was significantly positive , indicating earlier event transitions with repeated viewing . Not only does this approach provide a way of quantifying anticipation , it gives us a trajectory of the most likely event at any given timepoint , as well as the onset and duration of each event . We obtained whole-brain results using a searchlight analysis . We generated spherical searchlights spaced evenly throughout the MNI volume ( radius = 5 voxels; stride = 5 voxels ) . We retained only the searchlights with at least 20 voxels which were inside a standard MNI brain mask and for which at least 15 participants had valid data for all viewings . We then used the SRM ( Chen et al . , 2015 ) to functionally hyperalign all participants into shared 10-dimensional space ( jointly fitting the alignment across all six viewings ) and averaged their responses together . This produced a 10 feature by 60 timepoint data matrix for each of the six viewings , which was input to the HMM analysis described above . After running the analysis in all searchlights , the anticipation in each voxel was computed as the average anticipation of all searchlights that included that voxel . To assess statistical significance , we utilized a permutation-based null hypothesis testing approach . We constructed null datasets by randomly shuffling each participant’s six responses to the six presentations of the movie clip . The full analysis pipeline ( including hyperalignment ) was run 100 times , once on the real ( unpermuted ) dataset and 99 times on null ( permuted ) datasets , with each analysis producing a map of anticipation across all voxels . A one-tailed p-value was obtained in each voxel by fitting a normal distribution to the null anticipation values , and then finding the fraction of this distribution that exceeded the real result in this voxel ( i . e . , showed more anticipation than in our unpermuted dataset ) . Voxels were determined significant ( q<0 . 05 ) after applying the Benjamini-Hochberg FDR correction , as implemented in AFNI ( Cox , 1996 ) . To determine if anticipation systematically varied across the cortex in the hypothesized posterior-to-anterior direction , we calculated the Spearman’s correlation between the Y-coordinate of each significant ( q<0 . 05 ) voxel ( indexing the position of that voxel along the anterior/posterior axis ) and the mean amount of anticipation in that voxel . To obtain a p-value , the observed correlation was compared to a null distribution in which the Spearman’s correlation was computed with the null anticipation values from the permutation analysis described above , in which the order of the viewings was randomly scrambled for each participant . For comparison , the correlation was also computed for the X ( left-right ) and Z ( inferior-superior ) axes . This analysis was repeated on unthresholded anticipation maps , to examine if this hierarchy remained even when including regions whose anticipation amounts did not reach statistical significance . To relate the timescales of anticipation to the intrinsic timescales of brain regions during the first viewing , we fit the HMM on the first viewing alone , varying the number of events from 2 to 10 . The HMM was trained on the average response from half of the participants ( fitting the sequence of activity patterns for the events and the event variance ) and the log-likelihood of the model was then measured on the average response in the other half of the participants . The training and testing sets were then swapped , and the log-likelihoods from both directions were averaged together . Hyperalignment was not used during this fitting process , to ensure that the training and testing sets remained independent . The number of events that yielded the largest log-likelihood was identified as the optimal number of events for that searchlight . The optimal number of events was then compared to the anticipation timescale in that region ( from the main analysis ) , using Spearman’s correlation . For comparison , we also ran a searchlight looking for anticipatory effects using a non-HMM cross-correlation approach . Within each searchlight , we obtained an average timecourse across all voxels and correlated the response to the first viewing with the average response to repeated viewings at differing lags . Using the same quadratic-fit approach for identifying the optimal lag described below , we tested whether the repeated-viewing timecourse was significantly ahead of the initial-viewing timecourse ( relative to a null distribution in which the viewing order was shuffled within each subject ) . The p-values obtained were then corrected for FDR . We compared the event boundaries identified by the HMM within each searchlight to the event boundaries annotated by human observers . To obtain an event boundary timecourse from the annotations , we convolved the number of annotations ( across all raters ) at each second with the HRF ( Figure 4 ) . Separately , we generated a continuous measure of HMM ‘boundary-ness’ at each timepoint by taking the derivative of the expected value of the event assignment for each timepoint , as illustrated in Figure 1d . Moments with high boundary strength indicate moments in which the brain pattern was rapidly switching between event patterns . We cross-correlated the HMM boundary strength timecourse for each viewing with the annotated event boundary timecourse , shifting the annotated timecourse forward and backward to determine the optimal temporal offset ( with the highest correlation ) . We measured the timing of the peak correlation by identifying the local maximum in correlation closest to 0 lag , then fitting a quadratic function to the maximum correlation lag and its two neighboring lags and recording the location of the peak of this quadratic fit . This produced a continuous estimate of the optimal lag for each viewing . We measured the amount of shift between the optimal lag for the first viewing and the average of the optimal lags for repeated viewings , and obtained a p-value by comparing to the null distribution over maps with permuted viewing orders ( as in the main analysis ) , then performed an FDR correction . We identified three gray matter clusters significant at q<0 . 05 . To statistically assess whether the optimal lags differed from 0 in the three searchlights maximally overlapping these three clusters , we repeated the cross-correlation analysis in 100 bootstrap samples , in which we resampled from the raters who generated the annotated event boundaries . We obtained 95% bootstrap confidence intervals for maximally correlated lag on the first viewing and for the average of the maximally correlated lags on repeated viewings . Data preprocessing scripts and python code to reproduce all the results in this paper are available at https://github . com/dpmlab/Anticipation-of-temporally-structured-events ( copy archived at swh:1:rev:8fbd488c04d47148f9a53048de5d05a90e1c1663 ) . Results in MNI space can be viewed at https://identifiers . org/neurovault . collection:9584 . | Anticipating future events is essential . It allows individuals to plan and prepare what they will do seconds , minutes , or hours in the future . But how the brain can predict future events in both the short-term and long-term is not yet clear . Researchers know that the brain processes images or other sensory information in stages . For example , visual features are processed from lines to shapes to objects , and eventually scenes . This staged approach allows the brain to create representations of many parts of the world simultaneously . A similar hierarchy may be at play in anticipation . Different parts of the brain may track what is happening now , and what could happen in the next few seconds and minutes . This would provide a way for the brain to forecast upcoming events in the immediate , near , and more distant future at the same time . Now , Lee et al . show that the regions in the back of the brain anticipate the immediate future , while longer-term predictions are made in brain regions near the front . In the experiments , study participants watched a 90-second clip of the movie ‘The Grand Budapest Hotel’ six times while undergoing functional magnetic resonance imaging ( fMRI ) . Then , Lee et al . used computer modeling to compare the brain activity captured by fMRI during successive viewings . This allowed the researchers to watch participants’ brain activity moment-by-moment . As the participants repeatedly watched the movie clip , their brains began to anticipate what was coming next . Regions near the back of the brain like the visual cortex anticipated events in the next 1 to 4 seconds . Areas in the middle of the brain anticipated 5 to 8 seconds in the future . The front of brain anticipated 8 to 15 seconds into the future . Lee et al . show that many parts of the brain work together to predict the near and more distant future . More research is needed to understand how this information translates into actions . Learning more may help scientists understand how diseases or injuries affect people’s ability to plan and respond to future events . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"neuroscience"
] | 2021 | Anticipation of temporally structured events in the brain |
As we navigate through the world , eye and head movements add rotational velocity patterns to the retinal image . When such rotations accompany observer translation , the rotational velocity patterns must be discounted to accurately perceive heading . The conventional view holds that this computation requires efference copies of self-generated eye/head movements . Here we demonstrate that the brain implements an alternative solution in which retinal velocity patterns are themselves used to dissociate translations from rotations . These results reveal a novel role for visual cues in achieving a rotation-invariant representation of heading in the macaque ventral intraparietal area . Specifically , we show that the visual system utilizes both local motion parallax cues and global perspective distortions to estimate heading in the presence of rotations . These findings further suggest that the brain is capable of performing complex computations to infer eye movements and discount their sensory consequences based solely on visual cues .
Retinal images of the environment are altered by self-generated rotations such as eye or head movements . In order to perceive the world accurately , the component of retinal patterns resulting from such rotations needs to be discounted by the visual system . How the brain achieves such a rotation-invariant visual representation of the world remains unclear . Visually guided navigation is an important context in which achieving rotation-invariance is critical for accurate behavior ( Gibson , 1950; Warren and Saunders , 1995; Grigo and Lappe , 1999 ) . For example , while walking down a sidewalk and simultaneously looking at a passing car using eye or head rotations , the brain must discount the visual consequences of the self-generated rotations to estimate and maintain one's direction of translation ( i . e . , heading ) . Self-motion results in retinal velocity patterns known as ‘optic flow’ ( Gibson , 1950 ) . During translations , the resulting retinal pattern is generally an expansionary or contractionary radial flow field from which the point of zero velocity ( Focus of Expansion , FOE ) can be used to estimate heading ( Tanaka et al . , 1986; Warren et al . , 1988; Duffy and Wurtz , 1995; Britten , 2008 ) . However , eye or head rotations alter this flow pattern such that deciphering heading requires decomposing the resultant optic flow into translational and rotational components ( Figure 1A ) . Psychophysical ( Royden et al . , 1992; Royden , 1994; Crowell et al . , 1998 ) and electrophysiological ( Bradley et al . , 1996; Page and Duffy , 1999; Zhang et al . , 2004 ) studies have often emphasized the role of non-visual signals , such as efference copies of self-generated eye/head movements , in discounting rotations to estimate heading . Such non-visual signals can represent several different sources of rotation , including eye-in-head ( REH ) , head-on-body ( RHB ) , and body-in-world ( RBW ) movements ( Figure 1B ) . Critically , retinal image motion is determined by the translation and rotation of the eye relative to the world ( TEW and REW , Figure 1B ) , such that extracting heading from optic flow requires compensating for the total rotation of the eye-in-world ( where , REW = REH + RHB + RBW ) . Therefore , in general , multiple non-visual signals would need to be added to achieve a rotation-invariant estimate of heading , potentially compounding the noise that is associated with each signal ( Gellman and Fletcher , 1992; Li and Matin , 1992; Crowell et al . , 1998 ) . 10 . 7554/eLife . 04693 . 003Figure 1 . The problem of dissociating translations and rotations , and experimental approaches . ( A ) Optic flow patterns during self-motion ( shown as planar projections onto a flat image ) . Forward translations result in symmetric flow patterns ( black vector fields ) with a focus of expansion ( FOE ) indicating heading . When rotations are added to forward translations , the resultant optic flow pattern has an FOE shift in the direction of the added rotation ( rightward rotation: red , leftward rotation: blue ) . ( B ) VIP receives both visual and non-visual signals that may be used to achieve rotation-invariant heading estimates . Visual optic flow signals contain information about translation and rotation of the eye in the world ( TEW , REW ) whereas non-visual signals ( efference copies ) may contain information about rotation of eye-in-head ( REH ) , rotation of head-on-body ( RHB ) , or rotation of body-in-world ( RBW ) . ( C ) Visual stimuli simulating translations in eight directions spanning the entire horizontal plane were presented to the monkey . ( D ) Schematic showing the translation and rotation parameters in the simulated 3D cloud . Inset shows the trapezoidal velocity profile of translation and rotation during the course of a trial ( 1500 ms ) . ( E ) During the ‘Real pursuit ( RP ) ’ condition , the optic flow stimulus on the screen simulated translation , while rotation was added by having the monkey smoothly pursue a visual target that moved leftward or rightward across the screen . During the ‘Simulated pursuit ( SP ) ’ condition , the monkey fixated at the center of the display while optic flow simulated combinations of translation and eye rotation . During real and simulated pursuit , the optic flow patterns projected onto the monkey's retina were nearly identical . DOI: http://dx . doi . org/10 . 7554/eLife . 04693 . 00310 . 7554/eLife . 04693 . 004Figure 1—figure supplement 1 . Dependence of translational and rotational optic flow properties on viewing distance . Translational optic flow vectors ( left column ) decrease in magnitude as the distance of the plane being viewed increases . Rotational optic flow ( middle column ) , however , remains constant irrespective of the viewing distance . When these translation and rotation flow fields are added , the resultant FOE shift consequently varies with distance to the plane ( right column ) . Therefore , in a 3D environment where objects are present at varying distance from the observer , no single FOE exists . For the stimulus parameters used in this study , the nearest depth plane of the simulated 3D cloud ( 25 cm ) results in a 20° shift in FOE; at the screen depth of 35 cm , the shift is 33° and for any plane beyond 45 cm ( 45–125 cm ) , the FOE is undefined as the optic flow is dominated by rotations . DOI: http://dx . doi . org/10 . 7554/eLife . 04693 . 004 Alternatively , rotation-invariance can theoretically be achieved exclusively through visual processing ( Longuet-Higgins and Prazdny , 1980; Rieger and Lawton , 1985 ) . If the brain can use optic flow to directly estimate and discount rotations of the eye-in-world ( REW ) , such mechanisms may provide a complementary and potentially more efficient way to decompose rotations and translations to achieve invariant heading perception . Psychophysical studies have provided evidence that visual cues may play a role in estimating heading in the presence of rotations ( Grigo and Lappe , 1999; Li and Warren , 2000; Crowell and Andersen , 2001; Li and Warren , 2002 , 2004; Royden et al . , 2006 ) . However , electrophysiological evidence for the role of visual cues is ambiguous , in part because previous neurophysiological studies either did not include visual controls for eye rotation ( Zhang et al . , 2004 ) , simulated rotations incorrectly ( Bradley et al . , 1996; Shenoy et al . , 1999 , 2002 ) or employed insufficient analysis methods ( Bradley et al . , 1996; Shenoy et al . , 1999 , 2002; Bremmer et al . , 2010; Kaminiarz et al . , 2014 ) ( see ‘Discussion’ ) . We recorded neural activity from the macaque ventral intraparietal area ( VIP ) to evaluate the relative roles of visual and non-visual cues in computing heading in the presence of rotations . To elucidate the role of visual cues , we accurately simulated combinations of translations and rotations using visual stimuli containing a variety of cues present during natural self-motion . Our results provide novel evidence that ( 1 ) a subpopulation of VIP neurons utilizes visual cues to signal heading in a rotation-invariant fashion and ( 2 ) both local motion parallax and global perspective cues present in optic flow contribute to these computations . In addition , we find that visual and non-visual sources of rotation elicit similar responses in VIP , suggesting multi-sensory combination of both visual and non-visual cues in representing rotations . We further show that rotation-invariance is distinct from the reference frame used to represent heading , and provide additional support that heading representation in VIP is close to eye-centered ( Chen et al . , 2013 ) .
When rotation and translation occur simultaneously , the resulting pattern of retinal velocity vectors can differ substantially from the radial optic flow patterns observed during pure translation . This change is often conceptualized as a shift in the focus of expansion ( FOE ) ( Warren and Hannon , 1990; Bradley et al . , 1996; Shenoy et al . , 1999 , 2002 ) . However , in a visual scene with depth structure , adding rotation results in different FOE shifts at different depths ( Zhang et al . , 2004 ) . This is due to a key difference in the properties of optic flow resulting from translations and rotations—the magnitudes of translational optic flow vectors decrease with distance ( depth ) , whereas rotational optic flow vectors are independent of depth ( Longuet-Higgins and Prazdny , 1980 ) . Hence , for more distal points in a scene , rotations produce a larger FOE shift ( Figure 1—figure supplement 1 ) . For the translation and rotation parameters used in this study , the nearest plane in the 3D cloud ( 25 cm ) results in an FOE shift of approximately 20° . However , for any plane farther than 45 cm , the resultant optic flow has an undefined FOE ( Figure 1—figure supplement 1 , top row ) . The simulated 3D cloud ranged from 25 cm to 125 cm , resulting in a large volume of the stimulus space having undefined FOE shifts . Since FOE shift is an ill-defined measure of the visual consequence of rotations , we simply refer to the net visual stimulation associated with simultaneous translation and rotation as the ‘resultant optic flow’ . Forward translations result in an expansionary flow field , for which adding a rightward rotation causes a rightward shift of the focus of expansion ( for any given plane ) . On the other hand , backward translations produce a contractionary flow field and adding a rightward rotation results in a leftward shift in the focus of contraction ( Figure 2A ) . If a neuron signals heading regardless of the presence of rotations , then its tuning curves during real and simulated pursuit should be identical to the heading tuning curve measured during pure translation ( Figure 2B ) . For a neuron that instead represents the resultant optic flow rather than the translation component ( heading ) , a transformation of the tuning curve is expected due to the added rotations . As a result of the opposite shifts expected for forward ( expansionary flow field ) and backward translations ( contractionary flow field ) , the heading tuning curve of a neuron preferring forward headings would have a peak that shifts to the right and a trough that shifts to the left during rightward eye rotation; together , these effects cause a skewing of the tuning curve ( Figure 2C , red curve ) . For the same neuron , leftward eye rotation would cause the peak to shift to the left and the trough to shift to the right , thus having an opposite effect on the shape of the tuning curve ( Figure 2C , blue curve ) . Neurons that prefer lateral headings , which are common in VIP ( Chen et al . , 2011 ) , may in fact , show no shift in the peak . But , since opposite shifts are expected for forward and backward headings , the resulting tuning curve may exhibit substantial bandwidth changes ( Figure 2D ) . 10 . 7554/eLife . 04693 . 005Figure 2 . Predicted transformations of heading tuning curves due to rotations . ( A ) Forward and backward translations result in expansion and contraction flow fields , respectively ( row 1 ) . Adding rotation causes the FOE to shift in opposite directions for forward and backward translations ( rows 2 , 3 ) . ( B , C , D ) Hypothetical heading tuning curves showing the predicted transformations due to rotations ( rightward , red; leftward , blue ) . ( B ) Schematic illustration of rotation-invariant heading tuning curves . ( C ) Schematic representing a cell that responds to resultant optic flow ( no rotation tolerance ) with a heading preference of straight ahead ( 90° ) . Rightward rotation causes a rightward shift of the tuning curve for forward headings ( around 90° ) , and a leftward shift for backward headings ( around 270° ) . The opposite pattern holds for leftward rotations . Here , the net result of rotation is a skewing of the tuning curve . ( D ) Schematic tuning of a cell with a leftward heading preference ( 180° ) and no rotation tolerance . In this case , the tuning bandwidth increases for leftward rotations and decreases for rightward rotations . The opposite bandwidth changes would be observed for a cell with a 0° heading preference ( see Figure 2—figure supplement 1 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04693 . 00510 . 7554/eLife . 04693 . 006Figure 2—figure supplement 1 . Schematic showing tuning curve transformations for hypothetical neurons with different heading preferences . For cells that prefer lateral headings ( 0° , 180° ) , rotations ( rightward: red , leftward: blue ) cause changes in tuning bandwidth . The expected change in bandwidth is opposite for cells preferring 0° and 180° . For cells preferring forward or backward motion , rotations cause opposite directions of shifts in the peak and trough of the tuning curve , thus changing the shape of the tuning curve . DOI: http://dx . doi . org/10 . 7554/eLife . 04693 . 006 Therefore , under the null hypothesis that neural responses are simply determined by the resultant optic flow , the expected effect of rotation on heading tuning is not simply a global shift of the tuning curve , as was assumed previously ( Bradley et al . , 1996; Page and Duffy , 1999; Shenoy et al . , 1999 , 2002; Bremmer et al . , 2010; Kaminiarz et al . , 2014 ) . Further illustrations of the expected effects of rotation for hypothetical neurons with different heading preferences are shown in Figure 2—figure supplement 1 . We designed our quantitative analysis of heading tuning curves specifically to account for these previously unrecognized complexities ( see ‘Materials and methods’ ) . Heading tuning curves ( translation only ) can be compared to real pursuit ( RP ) and simulated pursuit ( SP ) tuning curves ( translation + rotation ) to evaluate whether a VIP neuron signals heading invariant to rotations ( Figure 2B ) , or whether it responds to the resultant optic flow ( Figure 2C , D ) . Figure 3A shows heading tuning curves for an example neuron during pure translation ( black curve ) , as well as during rightward ( red ) and leftward ( blue ) rotations added in RP and SP conditions . The tuning curves in this example show only minor changes during RP indicating that the cell signals heading in a manner that is largely invariant to eye rotation , consistent with previous findings for real eye rotation ( Zhang et al . , 2004 ) . Interestingly , the tuning curves of the same neuron during SP also change very little , showcasing the role of visual signals in compensating for rotation . Thus , rotation invariance in VIP that was previously attributed to non-visual signals ( Zhang et al . , 2004 ) might also be driven by visual cues . 10 . 7554/eLife . 04693 . 007Figure 3 . Heading tuning curves from two example VIP neurons . Five tuning curves were obtained per cell: one pure translation curve ( black ) , two real pursuit ( RP , left column ) curves , and two simulated pursuit ( SP , right column ) curves ( rightward rotation: red , leftward rotation: blue ) . Black horizontal line indicates baseline activity . Red and blue stars in the left column ( RP ) indicate responses during pursuit in darkness , and in the right column ( SP ) indicate responses to simulated eye rotation . ( A ) This neuron has largely rotation-invariant tuning curves in both RP and SP conditions ( shifts not significantly different from 0 , CI from bootstrap ) , and has significant rotation responses during both pursuit in darkness and simulated rotation ( compared to baseline; Wilcoxon signed rank test p < 0 . 05 ) . ( B ) This example neuron shows significant bandwidth changes during SP ( shifts >0° , CI from bootstrap ) , similar to the prediction of Figure 2D . Of the rotation-only conditions , the cell only responds significantly during rightward pursuit in darkness ( Wilcoxon signed-rank test p = 0 . 01 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04693 . 00710 . 7554/eLife . 04693 . 008Figure 3—figure supplement 1 . Bandwidth changes observed in data . Bandwidths of linearly interpolated tuning curves were calculated as the full width at half height ( FWHH ) . The difference in FWHH between the pure translation and rotation-added tuning curves ( reds: rightward , blues: leftward rotation ) are plotted for cells with lateral heading preferences ( since the largest bandwidth changes would occur for cells preferring lateral motion ) . The bandwidth changes observed are in the directions predicted by Figure 2 and Figure 2—figure supplement 1 . The change in bandwidths at 0° and 180° headings are significantly different for both RP ( Wilcoxon rank sum test; leftward , rightward: p < 0 . 001 ) and SP ( Wilcoxon rank sum test; leftward , rightward: p < 0 . 001 ) conditions . DOI: http://dx . doi . org/10 . 7554/eLife . 04693 . 00810 . 7554/eLife . 04693 . 009Figure 3—figure supplement 2 . Heading tuning curves from two example VIP neurons that preferred forward headings . Tuning curves are measured and represented as in Figure 3 . ( A ) This neuron has tuning that is largely invariant to rotations during both real and simulated pursuit . ( B ) This example neuron shows significant shifts in the tuning peak during both RP and SP . Since this neuron prefers forward translations , the shifts observed are similar to the predictions made in Figure 2C . DOI: http://dx . doi . org/10 . 7554/eLife . 04693 . 009 Data for another example VIP neuron ( Figure 3B ) reveal RP tuning curves that are also largely consistent in shape with the pure translation curve , but which have larger response amplitudes during leftward pursuit . During simulated pursuit , however , the tuning curves of this neuron show clear bandwidth changes . Thus , this second example neuron appears to rely more on non-visual cues to discount rotations . Note that this example neuron preferred lateral headings ( leftward ) and showed large bandwidth changes during SP , as predicted in the schematic illustration of Figure 2D . Such bandwidth changes were observed consistently among VIP neurons that preferred lateral translations; specifically , rightward rotations increased bandwidth for cells preferring rightward headings ( ∼0° ) and decreased bandwidth for cells preferring leftward headings ( ∼180° ) , with the opposite pattern holding for leftward rotations ( Figure 3—figure supplement 1 ) . We find analogous results for cells that preferred forward/backward translations . Specifically , we find neurons with tuning curve peaks around forward/backward heading that are invariant to added rotations ( Figure 3—figure supplement 2A ) , as well as neurons for which the tuning curve peaks shift with real and simulated pursuit ( Figure 3—figure supplement 2B ) , as shown in the simulations in Figure 2C . Because of these changes in tuning curve bandwidth or shape , analysis of the effects of rotation on heading tuning requires more complex and rigorous approaches ( Figure 4—figure supplement 1 ) than the cross-correlation or rank-order methods used in previous studies ( Bradley et al . , 1996; Shenoy et al . , 1999 , 2002; Bremmer et al . , 2010; Kaminiarz et al . , 2014 ) . It is also critical to distinguish between changes in response gain and changes in the shape ( Figure 4—figure supplement 2; see Discussion ) of tuning curves , which our analysis allows because we sample the entire heading tuning curve ( Mullette-Gillman et al . , 2009; Chang and Snyder , 2010; Rosenberg and Angelaki , 2014 ) . As shown in Figure 4—figure supplement 1 , the first step in the analysis involves normalizing each RP and SP tuning curve to match the dynamic range of the pure translation tuning curve . Following this transformation , the change in the shape of the RP and SP tuning curves can be measured without ambiguity . To account for the expected changes in bandwidth and skew , partial shifts of the tuning curve were measured separately for forward ( 0°:180° ) and backward ( 180°:360° ) headings . Thus , four shift values were obtained from each neuron for both real and simulated pursuit , corresponding to forward/backward headings and left/right rotation directions . These four values were averaged for each neuron to quantify the transformation in shape and obtain one shift metric for RP tuning curves and one for SP tuning curves ( see ‘Materials and methods’ , Figure 4—figure supplement 1 ) . Results are summarized for the population of recorded neurons ( n = 72; from two monkeys ) in Figure 4 . A shift of 0° implies that the neuronal representation of translation is invariant to rotation ( i . e . , the shape of heading tuning curves are highly similar , as in Figure 3A ) . A positive shift indicates under-compensation for rotation , such that responses change in a manner consistent with the resultant optic flow . Negative shifts indicate that the tuning curve transformation was in the direction opposite to that expected based on the resultant optic flow . This can be interpreted as an over-compensation for rotation . As noted earlier , though the FOE shift for the nearest depth plane ( 25 cm ) in our stimuli is 20° , a majority of the cloud volume ( 45–125 cm deep ) is dominated by rotations , such that the resultant optic flow has undefined FOEs . This implies that neurons should show shifts that are generally much larger than 20° if they do not discount the rotations and merely represent the resultant optic flow . 10 . 7554/eLife . 04693 . 010Figure 4 . Scatterplot and marginal distributions of shifts measured during real pursuit ( RP ) and simulated pursuit ( SP ) using 3D cloud stimuli ( n = 72 cells ) . A shift of 0° indicates rotation-invariance . Positive and negative shifts indicate under-compensation and over-compensation for rotation , respectively . Grey shaded area corresponds to shifts >20° ( conservative estimate of shift for cells with no tolerance to rotations ) . Red data points correspond to the shifts associated with the example cells shown in Figure 3 . Error bars depict bootstrapped 95% confidence intervals ( CI ) . Colored regions of marginal distributions indicate shifts ≤20° . Darker colors indicate shifts not significantly different from 0° . Uncolored histograms indicate shifts significantly >20° . Diagonal histogram shows difference in RP and SP shifts for each neuron with a median of −6 . 0° indicating that for most cells SP shifts tended to be larger than RP shifts ( significantly <0°; Wilcoxon signed-rank test p = 0 . 02 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04693 . 01010 . 7554/eLife . 04693 . 011Figure 4—figure supplement 1 . Method for analyzing tuning curve shifts . ( A ) Example tuning curves of a single neuron for RP and SP conditions . Black–pure translation; red–rightward rotation; blue–leftward rotation . ( B ) First , the offset and gain of the RP/SP tuning curves are corrected to match the offset and gain values of the pure translation tuning curve . ( C , D ) To account for bandwidth changes in the shift calculations , the RP/SP tuning curves are split into halves corresponding to forward ( 0:180° ) and backward ( 180:360° ) headings . The pure translation tuning curve is then circularly shifted to minimize the sum squared error with each half of the leftward rotation ( C ) and rightward rotation tuning curves ( D ) . This yields four shift values ( shown in C , D ) each , for SP and RP , which are averaged . ( E ) We simulated noisy neuronal tuning curves that have different response amplitudes , offsets and bandwidth/shape changes similar to real data ( see ‘Materials and methods’ ) . The expected shifts for the simulated data should correspond to 20° . The mean shifts ( based on 10 repetitions ) were not significantly different from 20° ( t-test; p = 1 ) , indicating that our analysis method correctly extracts tuning curve changes despite variations in shape , response amplitude , or baseline response . DOI: http://dx . doi . org/10 . 7554/eLife . 04693 . 01110 . 7554/eLife . 04693 . 012Figure 4—figure supplement 2 . Problems with previous approaches to measuring shifts in the absence of full tuning curve measurements . Previous studies ( Bradley et al . , 1996; Shenoy et al . , 1999; Shenoy et al . , 2002; Bremmer et al . , 2010; Kaminiarz et al . , 2014 ) evaluated heading tuning curves in a narrow range around straight ahead ( white region around 90° in panel A ) . ( A ) Two hypothetical tuning curves with different bandwidths and amplitudes . ( B ) Correcting for the difference in response amplitudes reveals a clear difference in bandwidths , which may reflect a lack of compensation for rotation . The rank order of the responses ( 1–1 , 2–2 , 3–3 ) would be identical for the two curves , which previous methods would erroneously interpret as evidence for rotation-invariance ( Bremmer et al . , 2010; Kaminiarz et al . , 2014 ) . ( C ) Simulated tuning curves with peaks roughly near straight ahead . The underlying von Mises functions ( to which Poisson noise was added ) have peaks at headings of 80° , 100° and 120° , resulting in a simulated shift of 20° . ( D ) The cross-correlation function between the translation only ( black ) and rotation-added ( red—leftward rotation , blue—rightward rotation ) tuning curves . Reflecting the true shift of 20° that was introduced into the tuning curves ( before noise was added ) , the cross-correlation functions peak near a lag of +20° . ( E ) Simulated tuning curves with peaks at 180° and different bandwidths . The difference in the width of the full tuning curves corresponded to a 20° shift ( at half-height ) . ( F ) Cross-correlation functions for the simulated neurons with lateral heading preferences are quite flat and show no evidence of a peak near a lag of +20° . ( G ) Shifts from 10 sets of simulated tuning curves ( with different noise samples ) were measured using cross-correlation for heading preferences ranging from 0° to 180° . The gain and offsets of the tuning curves were randomized for each set . The mean shift ( black markers ) approaches the true shift ( 20° , dashed line ) for tuning curves with heading preferences near 90° , but the mean shifts are grossly inaccurate for simulated neurons with lateral heading preferences . In contrast , Figure 4–figure supplement 1E shows that our analysis method correctly estimates tuning curve shifts regardless of heading preference . DOI: http://dx . doi . org/10 . 7554/eLife . 04693 . 012 In the RP condition , 22/72 ( 30 . 6% ) neurons showed shifts that were not significantly different from zero ( bootstrap 95% CI ) ; these cells can be considered to represent heading in a rotation-invariant fashion . For SP , 17/72 ( 23 . 6% ) neurons had shifts that were not significantly different from zero , indicating that purely visual cues were sufficient to achieve rotation-invariance in these neurons . Only 13/72 ( 18 . 1% ) neurons during RP and 19/72 ( 23 . 4% ) neurons during SP showed shifts that were significantly greater than 20° , suggesting that only a minority of VIP neurons simply represent the resultant optic flow . The median shift of the population during RP is 8 . 5° , which is significantly less than the 13 . 8° median shift observed during SP ( Wilcoxon signed-rank test; p = 0 . 02 ) , indicating greater tolerance to rotations in the presence of both non-visual and visual cues . However , both median shifts are significantly greater than 0° ( Wilcoxon signed-rank test; p < 0 . 001 ) , and less than 20° ( Wilcoxon signed-rank test; RP: p < 0 . 001 , SP: p = 0 . 005 ) suggesting that , on average , VIP neurons do not simply represent the resultant optic flow , but rather signal heading in a manner that is at least partially tolerant to rotations . Together , these findings indicate that VIP can signal heading in the presence of rotations using both visual and non-visual cues . Importantly , this tolerance to rotations is observed even when only visual cues are present ( SP ) . The previous section shows that VIP neurons can use visual cues to signal heading in the presence of rotations , but it is unclear if the rotational component is also represented . During real pursuit , the rotation arises from a movement of the eye relative to the head . In this case , both non-visual and visual sources of information about the rotation are available . These two sources of information differ in that the non-visual source signals the rotation of the eye relative to the head ( REH ) and the visual source signals the rotation of the eye relative to the world ( REW ) . Previous studies have shown that VIP receives efference copies of pursuit eye movements ( Colby et al . , 1993; Duhamel et al . , 1997 ) , reflecting an REH signal . However , no previous studies have tested if VIP also carries an REW signal based on visual rotation information present in optic flow . To test whether neurons in VIP signal rotations based on both non-visual and visual cues , we analyzed data from interleaved rotation-only trials ( leftward and rightward rotations ) in which the monkey either pursued a target in darkness ( non-visual REH signal ) or fixated centrally while the visual stimulus simulated a rotation ( visual REW signal ) with the same velocity profile as pursuit in darkness . We found that about half of the rotation responses were significantly different from baseline activity during both real and simulated rotations ( 144 responses from 72 cells; 73/144 , 50 . 7% during pursuit in darkness and 78/144 , 54 . 2% during simulated rotation ) . Since we only tested horizontal ( yaw axis ) rotations at a single constant velocity , it is likely that more VIP neurons are responsive to rotation , but prefer different rotation velocities or axes of rotation . In our experiments , the REW signal is equivalent to the REH signal since only eye rotations are considered . Therefore , similarity between the efference copy signal ( REH ) and the neural responses to purely visual rotation stimuli ( REW ) would suggest the presence of an integrated ( visual and non-visual ) REW signal in VIP . We find that the baseline-subtracted responses to these two types of rotation stimuli are significantly correlated ( rightward rotation: Spearman r = 0 . 50 , p < 0 . 001; leftward rotation: Spearman r = 0 . 39; p = 0 . 001 ) , supporting the presence of a rotation signal derived from purely visual cues ( REW ) in area VIP ( Figure 5A ) . Furthermore , the difference in response between rightward and leftward rotations ( Figure 5B ) shows that many VIP neurons exhibit direction-selective responses to rotation . We also find significant correlation between the differential responses ( left—right rotation ) during real and simulated rotation ( Spearman r = 0 . 59; p < 0 . 001 ) . These results support the hypothesis of multi-sensory convergence of visual and non-visual cues to provide consistent rotation information , which may be critical for encoding rotations , in addition to achieving a rotation-invariant representation of heading . 10 . 7554/eLife . 04693 . 013Figure 5 . Neural responses to pure rotation stimuli . ( A ) Scatterplot and marginal distributions of baseline-subtracted rotation responses . The monkey either pursued a target across a dark screen ( pursuit in darkness ) or fixated centrally as rotation was simulated in the 3D dot cloud ( simulated rotation ) . Filled marginal distributions indicate significant rotation responses compared to baseline ( t-test , p ≤ 0 . 05 ) . Red and blue symbols denote rightward and leftward rotations , respectively . ( B ) Scatterplot of differences between leftward and rightward rotation responses . Filled marginal distributions indicate significant differences between leftward and rightward rotation responses ( t-test , p ≤ 0 . 05 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04693 . 013 It is important to note that , in general , retinal motion corresponding to REW is a combination of REH , rotation of the head-on-body ( RHB ) , and body-in-world ( RBW ) . And each of these different rotations will be accompanied by different efference copy ( non-visual ) signals . If VIP neurons represent REW based on non-visual signals , then they would have to represent a combination of all efference copy signals: REW = REH + RHB + RBW . Although we cannot test this directly with our data , the correlations observed in Figure 5 allow for the possibility that VIP neurons represent REW based on both visual and non-visual cues . Results from the 3D cloud experiment ( Figure 4 ) demonstrate , for the first time at the neural level , a clear contribution of visual cues in achieving a rotation-tolerant representation of heading . To gain a deeper understanding of the visual mechanisms involved in dissociating translations and rotations , we investigated which optic flow properties are used by the visual system to infer self-motion from visual cues . Helmholtz and Southall ( 1924 ) and Gibson ( 1950 ) suggested that local motion parallax information plays an important role in deciphering self-motion based on the depth structure of a scene . In a 3D environment , two points can have similar retinal locations , but different depths . As illustrated in Figure 1—figure supplement 1 , optic flow vectors resulting from observer translation are dependent on depth , producing different retinal velocities for points at different depths . This difference in velocity between nearby points at different depths gives rise to local motion parallax . Rotations , on the other hand , produce image motion that is not depth-dependent , and therefore lacking local motion parallax . As a result , for a rich 3D environment , computing the local difference between optic flow vectors corresponding to points at different depths allows the rotational component of optic flow to be subtracted away ( Longuet-Higgins and Prazdny , 1980; Rieger and Lawton , 1985; Warren and Hannon , 1990 ) , and the singularity point of the resulting motion parallax field ( Figure 6A ) corresponds to the observer's heading . This solution requires rich depth structure in the scene , which is not always present . For instance , walking through a dense forest provides robust local motion parallax cues , but walking towards a wall or through an open field , does not . 10 . 7554/eLife . 04693 . 014Figure 6 . Role of dynamic perspective cues in signaling rotation-invariant heading . ( A ) Optic flow fields during combined translation and rotation at two different depth planes have different FOE shifts . The dotted circle indicates true heading . Subtracting these flow fields yields a motion parallax field that eliminates the rotational component and the point of zero local motion parallax corresponds to the true heading . ( B ) Rotational optic flow can be decomposed into laminar flow and dynamic perspective cues . Dynamic perspective cues may signal eye rotations even in the absence of depth structure . ( C ) Scatterplot and marginal distributions of shifts measured using the fronto-parallel plane stimulus during real and simulated pursuit ( n = 34 cells ) . Format as in Figure 4 . Open and filled symbols denote data collected during binocular and monocular viewing , respectively . Errorbars denote bootstrapped 95% CIs . All filled histograms indicate shifts significantly <37° . Dark colored histogram bins indicate cells with shifts not significantly different from 0° . Uncolored bars indicate shifts ≥37° . DOI: http://dx . doi . org/10 . 7554/eLife . 04693 . 014 In addition to local motion parallax cues resulting from observer translation , optic flow also contains global components of motion that convey information about observer rotation . When a pure eye rotation is simulated using optic flow stimuli , the image contains distortions resulting from the changing orientation of the eye relative to the scene , that we term ‘dynamic perspective cues’ ( see Kim et al . , 2014 for more details ) . A correct simulation of rotational optic flow can thus be characterized as a combination of laminar flow and dynamic perspective cues ( Figure 6B ) . Importantly , these cues are independent of the depth structure of the scene and are present in scenes having rich 3D structure as well as scenes consisting of a single plane . Theoretical studies have proposed that such cues may play an important role in estimating and discounting the rotational component of optic flow to estimate heading ( Koenderink and van Doorn , 1976 , 1981; Grigo and Lappe , 1999 ) . A recent electrophysiological study in MT provides evidence that the visual system may be capable of using these dynamic perspective cues to disambiguate the sign of depth specified by motion parallax ( Kim et al . , 2014 ) . To examine the role of dynamic perspective cues , we conducted a second set of experiments using a fronto-parallel ( FP ) plane of dots with zero disparity . These visual stimuli contain global perspective cues to rotation , as in the 3D cloud stimulus , but lack local motion parallax cues . For 11/34 neurons recorded , the stimulus was viewed binocularly; the remaining cells were recorded while the monkey viewed the stimulus monocularly with the eye contralateral to the recording hemisphere . In contrast to previous studies , which kept the simulated distance to a FP wall constant over the duration of a trial ( Bradley et al . , 1996; Shenoy et al . , 1999 , 2002 ) , the simulated distance of the FP plane changed , in our stimuli , from 45 cm at the beginning to 18 cm at the end of the trial . This more accurately simulates the real world situation in which approaching a wall reduces its distance from the observer over time . As a result , the speed of the translation component of optic flow increased over time for forward translations as the distance to the wall decreased ( Figure 1—figure supplement 1 ) . Since rotational optic flow is invariant to the distance from a wall ( Figure 1—figure supplement 1 ) , the resulting shift in FOE due to added rotations changed over time in our stimulus . During the middle 750 ms of a forward translation stimulus , real or simulated pursuit results in an average FOE shift of 37° . Hence , heading tuning shifts significantly smaller than 37° would provide evidence for the hypothesis that the visual system can use dynamic perspective cues to discount rotations . Figure 6C summarizes the shifts in heading tuning measured during presentation of the FP plane stimulus . The median shifts across the population for real pursuit ( 14 . 3° ) and simulated pursuit ( 21 . 5° ) were both significantly less than the 37° expected if there were no tolerance for rotations ( Wilcoxon signed-rank test; p < 0 . 005 ) . The median values were also significantly different from each other ( Wilcoxon signed-rank test; p = 0 . 03 ) and greater than 0° ( Wilcoxon signed-rank test; p < 0 . 001 ) . Furthermore , 8/34 ( 23 . 5% ) neurons during RP and 5/34 ( 14 . 7% ) neurons during SP had shifts that were not significantly different from 0° ( darker colors in Figure 6C ) , implying rotation-invariant heading responses . Only 6/34 ( 17 . 6% ) neurons during RP and 12/34 ( 35 . 3% ) neurons during SP showed shifts that were statistically greater than or not different from 37° ( 95% CI; see ‘Materials and methods’ ) . These results indicate that , even in the absence of non-visual signals and 3D visual cues such as local motion parallax , a large sub-population of VIP neurons can use global perspective cues to at least partially mitigate the effect of rotations on heading tuning . Shifts measured during simulated pursuit in the 3D cloud experiments were significantly less than shifts measured using the FP plane ( Wilcoxon rank sum test; p = 0 . 02 ) . This implies that both local motion parallax cues arising from translations , and global features such as dynamic perspective cues arising from rotations play important roles in visually dissociating translations and rotations . Since the eyes physically rotate during real pursuit , but the head does not , previous studies interpreted rotation-invariant heading tuning as evidence that VIP neurons represent self-motion in a head-centered reference frame ( Zhang et al . , 2004 ) . In contrast , studies that measured heading tuning with the eye and head at different static positions have revealed an eye-centered reference frame for visual heading tuning in VIP ( Chen et al . , 2013 , 2014 ) . On the surface , these results appear to be incompatible with each other . However , we posit that the issues of rotation-invariant heading tuning and reference frames are not necessarily linked . Indeed , we show below that VIP neurons can discount rotations without signaling translation direction in a head-centered reference frame . The key to reconciling these issues is appreciating that , during eye pursuit , the eye-centered reference frame rotates relative to a subject's heading ( Figure 7A ) . As the eye rotates , the direction of translation remains constant in head-centered coordinates ( Figure 7A , dashed green lines ) . However , in the rotating eye-centered reference frame , the translation direction relative to the eye changes over time , such that the focus of expansion moves across the retina ( Figure 7B ) . This is true for both the RP and SP conditions . During RP , the eye physically moves and the FOE remains constant on the screen , whereas during SP , the eye remains stationary as the FOE drifts across the screen . Hence , the temporal change in the translation direction with respect to the retina is the same during both real and simulated pursuit . 10 . 7554/eLife . 04693 . 015Figure 7 . Distinguishing reference frames from rotation invariance . ( A ) Schematic of a rightward eye rotation while translating forward . As the eye position changes during smooth pursuit , the eye reference frame ( ERF , black axes ) rotates relative to the head ( REH ) and the direction of translation in the world , TEW . Since the head is not rotating relative to the world , the head reference frame ( HRF , green axes ) remains constant with respect to the heading . ( B ) In retinal co-ordinates , the translation component of optic flow changes with eye position and results in a drifting FOE ( x ) across the retina . The translation direction represented by the FOE changes from right of straight ahead to left of straight ahead for rightward rotations . ( C , D ) Heading corresponding to the largest firing rate gradient was identified for each neuronal tuning curve and the temporal responses at that heading were evaluated . Dashed straight lines show the predicted population response slopes based on the assumption of an eye-centered reference frame . The population average of the normalized time course of firing rate is plotted for each condition type—translation only ( grey ) , rightward rotation ( red ) and leftward rotation ( blue ) for real pursuit ( C ) and simulated pursuit ( D ) . Shaded regions indicate standard errors . The significant positive and negative trends observed are consistent with a reference frame that is intermediate between eye- and head-centered , but closer to an eye-centered reference frame . DOI: http://dx . doi . org/10 . 7554/eLife . 04693 . 015 In our experimental protocol , as well as that of previous studies ( Bradley et al . , 1996; Shenoy et al . , 1999 , 2002; Zhang et al . , 2004 ) , the average eye position during the translation-only , real pursuit and simulated pursuit conditions is the same ( centered on the screen ) over the duration of a trial . Therefore , the average eye position is the same as the average head position . As a result , time-averaged neural responses may provide insight into what signal is represented ( heading or resultant optic flow ) , but not about whether these signals are represented in an eye- or head-centered reference frame . To evaluate reference frames , responses must be examined with the eye at different positions relative to the simulated heading . In our experiments , we can examine the temporal responses of neurons to study reference frames since the translation direction in eye coordinates changes over time . Accordingly , an eye-centered representation of heading would result in systematic temporal response variations due to the rotating reference frame , and these variations would be different for leftward and rightward rotations of the eye . In contrast , a head-centered representation would result in responses that are constant over time , and similar for rightward and leftward rotations during both real and simulated pursuit . For a neuron representing heading in an eye-centered reference frame , a rightward eye rotation would result in an upward trend in firing rate over time for headings along the positive slope of the tuning curve . In contrast , a leftward eye rotation would result in a downward trend ( Figure 7C , D , dashed lines ) . It is important to note that these trends are determined by the changing eye position and are independent of how tolerant the heading representation is to rotations ( i . e . , the extent of compensation ) . The degree of rotation compensation would result in a shift in the mean firing rate away from the pure translation responses ( as discussed in previous sections ) , irrespective of the reference frame in which translations are represented . Therefore , neurons can represent translations invariant to rotations in either a head-centered or an eye-centered reference frame . In order to evaluate the underlying reference frame for representing translations in area VIP , we examined the temporal changes in firing rate for each neuron over the same 750 ms epoch used in the rest of the analyses . If neurons signal heading in an eye-centered reference frame , the largest temporal variations in firing rate will occur at headings along the steepest portion of the tuning curve . Therefore , we identified the heading corresponding to the largest positive gradient for each tuning curve , and examined the temporal dynamics of responses for that direction . In order to determine the expected temporal changes in firing rate under the assumption of an eye-centered reference frame , the slope of the tuning curve at the heading corresponding to the largest gradient was calculated for each normalized tuning curve . The average predicted slopes for the population based on our data were ±0 . 41/s during real pursuit and ±0 . 4/s for simulated pursuit ( dashed lines in Figure 7C , D ) . These predictions , based on an eye-centered reference frame hypothesis , were compared to the average time course of normalized responses of the population of VIP neurons ( see ‘Materials and methods’ for details ) . VIP population responses show trends in the directions predicted by an eye-centered reference frame , but are inconsistent with the expectation for a head-centered reference frame ( red , blue curves in Figure 7C , D ) . The slopes observed in VIP responses correspond to an intermediate reference frame that lies closer to an eye-centered frame than a head-centered reference frame . Specifically , for real pursuit , average responses increased for rightward eye rotation ( slope = 0 . 29/s , 95% CI = [0 . 21 0 . 37] , linear regression ) and decreased for leftward rotation ( slope = −0 . 24/s , 95% CI = [−0 . 16–0 . 32] ) . These slopes are significantly different from 0 and ∼65% as steep as the predictions of the eye-centered reference frame , thus indicating an intermediate reference frame . Since the temporal response profile was essentially flat during the translation only condition ( slope = 0 . 01/s , 95% CI = [−0 . 04 0 . 06] ) and opposite trends are observed for rightward vs leftward rotations , these response patterns cannot be explained by other basic aspects of neural response dynamics , such as adaptation . Interestingly , similar trends are also observed during simulated pursuit ( rightward: slope = 0 . 28/s , 95% CI = [0 . 21 0 . 35]; leftward: slope = −0 . 26/s , 95% CI = [−0 . 17–0 . 35] , linear regression ) , for which the eye does not physically rotate . These slopes are again about two-thirds as steep as expected based on the eye-centered reference frame hypothesis . Whereas previous studies have demonstrated a role of non-visual signals in estimating the position of the eye or head relative to the body ( Squatrito and Maioli , 1997; Lewis et al . , 1998; Klier et al . , 2005 ) , these results suggest that visual signals in VIP carry information about changes in eye position even in the absence of efference copy signals . In other words , the temporal dynamics of an eye rotation may be inferred from the rotational components of optic flow and used to modulate neural responses during simulated pursuit . This further strengthens the functional role of visual signals in VIP for estimating rotational information and contributing to a rotation-invariant heading representation .
It is important to recognize that the significance of visual cues in discounting rotation extends beyond eye pursuit to head-on-body ( RHB ) and body-in-world ( RBW ) rotations as well . The efference copy signals for each of these sources of rotation depend on the specific motor commands generating the movement . If we consider that eye , head , and body rotations are often generated simultaneously , multiple efference copy signals must be added together and subsequently discounted from the resultant optic flow to signal heading accurately . Each of these non-visual signals is associated with signal-dependent noise ( Gellman and Fletcher , 1992; Li and Matin , 1992; Crowell et al . , 1998 ) ; thus , combining multiple , potentially independent , efference copy signals to estimate rotations may not always be an efficient solution for the brain . On the other hand , the information contained in visual cues is independent of the source of rotation and represents rotation of the eye relative to the world ( REW ) . The REW information present in optic flow inherently reflects the sum of all the different sources of rotation ( REW = REH + RHB + RBW ) and thus provides direct information regarding the total rotation of the eyes during self-motion . Therefore , visual signals may have important advantages when the goal is to accurately estimate heading in the presence of self-generated rotations . However , we also face situations in which visual information may be sparse , such as driving at night on an open road ( limited visual range and depth structure ) , where non-visual signals may be crucial . As expected , given the brain's propensity towards multi-sensory integration , we find that both visual and non-visual signals contribute to discounting rotations to represent heading . Real pursuit shifts are smaller than simulated pursuit shifts , and both types of shifts are smaller for a dense 3D cloud than a fronto-parallel plane . Given the variety of efference copy signals present in parietal cortex ( Andersen , 1997 ) and the correlation observed between the REH ( pursuit in darkness ) and REW ( pure simulated rotation ) responses in our data ( Figure 5 ) , we postulate that VIP contains an integrated representation of rotation that relies on both visual signals and efference copy inputs . However , to conclusively test these theories , experiments with multiple rotation velocities and directions as well as different sources of rotation ( e . g . , eye vs head pursuit ) need to be conducted . How these visual rotation cues are combined with efference copy signals and other non-visual sensory cues to rotation ( e . g . , vestibular inputs ) warrants further investigation . Several human psychophysical studies have assessed pursuit compensation during heading estimation based on visual and non-visual cues . However , owing to variations in experimental protocols , visual stimuli , and instructions given to the subjects , the results of these studies vary substantially . If we consider studies that used 3D cloud stimuli , we find that some studies report large errors in heading perception ( the difference between reported heading and true heading ) in the absence of efference copy signals ( Royden et al . , 1992; Royden , 1994; Banks et al . , 1996 ) , whereas other studies report that subjects are able to accurately perceive their heading based on purely visual stimuli ( Warren and Hannon , 1988 , 1990; van den Berg and Brenner , 1994 ) . In order to compare results across these studies , we calculated the degree of compensation as the difference between the error in heading perception and the shift in FOE based on the experimental parameters , normalized by the expected shift in FOE ( [FOE shift-heading error]/FOE shift ) . The rotation compensation observed in these studies during simulated pursuit ( only visual cues ) ranged from 100% to 20% for a 3D cloud stimulus ( based on the depth plane corresponding to the screen distance ) . Studies with smaller compensatory effects ( Royden et al . , 1992; Royden , 1994; Banks et al . , 1996 ) concluded that optic flow was insufficient for estimating translations in the presence of rotations . However , these studies used visual stimuli with a small field of view and a low density of dots in the 3D cloud , thus limiting the visual information available for estimating heading in the presence of rotations ( Grigo and Lappe , 1999 ) . Despite these limitations in the visual stimuli , the compensatory effects were greater than 0 . Moreover , other studies have shown that richer visual stimuli , including landmarks ( Li and Warren , 2000 , 2004; Royden et al . , 2006 ) and larger fields of view ( van den Berg and Brenner , 1994; Grigo and Lappe , 1999 ) , resulted in larger compensatory effects based on purely visual cues . In this study , using a 3D cloud stimulus , we observed a large and continuous range of compensatory effects , including a substantial subset of VIP neurons that compensated completely for rotations , as well as neurons that do not compensate at all or even over-compensate for rotations . Since the experimental parameters used in our study and the various behavioral papers are different , it is difficult to compare our results quantitatively with the published behavioral findings . However , the fact that we find moderate , but significant , compensation during simulated pursuit is broadly consistent with the psychophysical literature . Furthermore , depending on how the population of VIP neurons is decoded , a substantial range of behavioral effects might be expected . For instance , if the rotation-invariant neurons are selectively decoded to estimate heading , it should be possible for VIP to support behavioral responses with compensation close to 100% . On the other hand , if all VIP neurons are decoded with equal weights , we would expect the behavioral errors to be comparable to the mean compensation observed in the neural population . It is also important to note that in many behavioral studies , subjects made small but significant errors even during real pursuit ( Freeman , 1999; Freeman et al . , 2000; Crowell and Andersen , 2001 ) , consistent with our finding that the average compensation among VIP neurons is not complete even when both visual and non-visual cues to rotation are available . Some psychophysical studies attribute the errors observed during simulated pursuit to the misinterpretation of path-independent rotations ( such as eye pursuit during straight translations ) as motion along a curved path ( Royden , 1994; Royden et al . , 2006 ) . In behavioral studies that eliminate this ambiguity through specific instruction to subjects , heading errors during simulated pursuit are reported to be largely reduced ( Li and Warren , 2004; Royden et al . , 2006 ) . This provides further evidence that the visual system is indeed capable of estimating rotation-invariant heading based on purely visual stimuli . It is also possible that the range of compensation observed in our data could be a result of this perceptual ambiguity . To evaluate how the brain resolves this ambiguity , neurophysiological studies using both path-independent rotations and curved path stimuli are needed . Previous physiological studies emphasized the contribution of efference copy signals to achieving rotation invariance ( Bradley et al . , 1996; Page and Duffy , 1999; Shenoy et al . , 1999; Zhang et al . , 2004 ) . However , these studies could not conclusively establish a contribution of visual rotation cues to heading tuning for various reasons . Some studies did not use a simulated pursuit condition and therefore could not disambiguate visual and non-visual contributions to the rotation-invariance of heading tuning they observed ( Page and Duffy , 1999; Zhang et al . , 2004 ) . On the other hand , Bradley et al . ( 1996 ) and Shenoy et al . ( 1999 ) ; ( 2002 ) included a simulated pursuit condition in their experiments , but the visual stimulus used to simulate pursuit was incorrect . To mimic pursuit , they simply added laminar flow to their fronto-parallel plane ( i . e . no local motion parallax cues ) optic flow stimuli , and thus their stimuli lacked the dynamic perspective cues necessary to accurately simulate eye rotations on a flat display . When rendering visual stimuli , dynamic perspective cues should be incorporated any time the eye changes orientation relative to the scene ( Kim et al . , 2014 ) . If eye rotation is simulated ( incorrectly ) as laminar flow on a flat screen , then it should not be possible for neurons to exhibit rotation-tolerant heading tuning because the addition of laminar motion simply shifts the focus of expansion in the flow field , and does not provide any rotation cues . Indeed , Bradley et al . ( 1996 ) found that MSTd neurons did not compensate for rotations when pursuit was simulated in this manner . In contrast , Shenoy et al . ( 1999 ) ; ( 2002 ) reported that MSTd neurons show considerable tolerance to rotation when pursuit was simulated as laminar flow , despite the fact that little or no rotation tolerance was reported psychophysically by the same laboratory for simulated pursuit ( Crowell et al . , 1998 ) . Compared to Bradley et al . ( 1996 ) , Shenoy et al . ( 2002 ) used a smaller display size and yet observed larger compensatory effects . This finding contradicts theoretical and psychophysical studies that have established that a larger display size should improve pursuit compensation based on visual cues ( Koenderink and van Doorn , 1987; Grigo and Lappe , 1999 ) . We believe that the counter-intuitive results obtained by Shenoy et al . ( 1999 ) ; ( 2002 ) stem from the fact that the boundary of their visual stimuli moved across the retina during real and simulated pursuit ( but not during the fixation condition ) , and thus stimulated different regions of the visual field in and around the receptive field of a neuron over time . Such a moving image boundary defined only by the rotation velocity would not occur under natural conditions as a result of eye rotations . By changing the region of visual space that was stimulated over the course of a trial , Shenoy et al . ( 1999 ) ; ( 2002 ) likely induced changes in the amplitude ( response gain ) or shape of heading tuning curves . Shenoy et al . ( 1999 ) ; ( 2002 ) measured heading tuning over a narrow range ( ±32° ) around straight ahead , and estimated shifts in tuning by cross-correlation analysis . While cross-correlation is invariant to gain changes , it only provides an accurate measure of tuning shifts if the tuning curve has a clear peak within the range of headings tested ( Figure 4—figure supplement 2C , D; see ‘Materials and methods’ ) . In contrast , cells that prefer lateral headings generally have monotonic tuning curves around straight ahead ( e . g . , Figure 4—figure supplement 2E ) , and this generally yields rather flat cross-correlation functions with no clear peak ( e . g . , Figure 4—figure supplement 2F ) . As a result , cross-correlation analysis produces fairly accurate estimates of shifts for cells with heading preferences within the range of headings tested , but does not provide reliable shifts for neurons with monotonic tuning functions in that range ( Figure 4—figure supplement 2G ) . These simulations suggest that the degree of rotation compensation reported previously ( Shenoy et al . , 1999 , 2002 ) may have been inaccurate for neurons with monotonic tuning around straight-forward , which are common in areas MSTd ( Gu et al . , 2006 , 2010 ) and VIP ( Chen et al . , 2011 ) . This may also help explain the partial rotation compensation observed by Shenoy et al . ( 1999 ) ; ( 2002 ) in their ( incorrect ) simulated rotation condition , which contained no relevant visual cues that could be used to compensate for rotation . In contrast to the cross-correlation method , our method for measuring shifts works well for cells with all heading preferences ( Figure 4—figure supplement 1E ) , and is robust to variations in the gain , offset and shape of tuning curves . More recently , Bremmer et al . ( 2010 ) and Kaminiarz et al . ( 2014 ) reported that neurons in areas MSTd and VIP , respectively , show rotation-invariant heading tuning based solely on visual cues . However , these studies only measured neural responses to three headings ( forward , 30° leftward , and 30° rightward ) , and defined rotation-tolerance based on a rank-ordering of heading responses across the different eye movement conditions . Since absolute firing rates were not considered , it is likely that shifts in tuning curves could go undetected by this method in the presence of gain fields or bandwidth changes . For instance , this analysis would report identical rank-order for the tuning curves shown in Figure 4—figure supplement 2A , and would erroneously classify them as rotation-invariant . In addition , the authors did not attempt to compare their results to the tuning shifts that would be expected if neurons do not compensate for rotation . Consider that , in their ground-plane stimuli ( e . g . , Figure 1 of Kaminiarz et al . , 2014 ) , rotation has a large effect on slow-speed optic flow vectors near the horizon , and high-speed foreground vectors are much less altered . For neurons with receptive fields below the horizontal meridian or those with responses dominated by high speeds , one might not expect the rank ordering of heading responses to change even if neurons do not compensate for rotation . Thus , the results of these studies are difficult to interpret . By comparison with the above studies , we accurately simulated eye rotations such that correct 2D and 3D visual cues are present in the stimuli . We also measured full heading tuning curves and our analysis methods allowed us to disambiguate changes in response gain from shifts or shape changes in the tuning curve . By using a large display and maintaining the same area of retinal stimulation for all viewing conditions ( see ‘Materials and methods’ ) , we eliminated artifacts that likely confounded the results of some previous studies ( Shenoy et al . , 1999 , 2002 ) . Therefore , we are confident that our findings in the simulated rotation condition reflect a true contribution of visual cues to the problem of dissociating translations and rotations . In order to navigate through the environment and interact successfully with objects , it is imperative that we distinguish visual motion caused by self-generated movements from that caused by external events in the world ( Probst et al . , 1984; Wallach , 1987; Warren and Saunders , 1995 ) . For instance , the visual consequences of eye or head rotations need to be discounted in order to accurately perceive whether an object is stationary or moving in the world . The neuroscience literature has extensively studied and emphasized the contribution of efference copy signals to discounting self-generated movements in several sensory systems ( Andersen , 1997; Cullen , 2004; Klier et al . , 2005 ) . We have presented novel evidence for an alternative solution that is available to the visual system—using large-field visual motion cues to discount self-generated rotations . The ability of VIP neurons to represent heading during rotations , even in the absence of efference copy signals , suggests that visual mechanisms may make substantial contributions to a variety of neural computations that involve estimating and accounting for self-generated rotations . The contribution of visual cues may be especially important in situations where efference copy signals are either unreliable or absent . For instance , driving along a winding path and looking in the direction of instantaneous heading does not result in any eye or head movements relative to the body ( i . e . , no efference copy signals ) . However , such curvilinear motion still introduces rotational components in the optic flow field and disrupts the FOE . In order to estimate such motion trajectories , the visual system would need to decompose self-motion into both translational and rotational components . This study suggests that such trajectory computations based purely on optic flow may be feasible . How the visual system may implement such computations warrants further research and may provide useful insights to neuroscientists as well as those in the fields of computer vision and robotic navigation .
Two adult rhesus monkeys ( Macaca mulatta ) , weighing 8–10 kg , were chronically implanted with a circular molded , lightweight plastic ring for head restraint and a scleral coil for monitoring eye movements ( see Gu et al . , 2006; Fetsch et al . , 2007; Takahashi et al . , 2007 for more detail ) . Following recovery from surgery , the monkeys were trained to sit head restrained in a primate chair . They were subsequently trained using standard operant conditioning to fixate and pursue a small visual target for liquid rewards , as described below . All surgical materials and methods were approved by the Institutional Animal Care and Use Committees at Washington University and Baylor College of Medicine , and were in accordance with NIH guidelines . The primate chair was affixed inside a field coil frame ( CNC Engineering , Seattle , WA , USA ) with a flat display screen in front . The sides and top of the coil frame were covered with a black enclosure that restricted the animals' view to the display screen . A three-chip DLP projector ( Christie Digital Mirage 2000 , Kitchener , Ontario , Canada ) was used to rear-project images onto the 60 × 60 cm display screen located ∼30 cm in front of the monkey ( thus subtending 90° × 90° of visual angle ) . Visual stimuli were generated by an OpenGL accelerator board ( nVidia Quadro FX 3000G ) . The display had a pixel resolution of 1280 × 1024 , 32-bit color depth , and was updated at the same rate as the movement trajectory ( 60 Hz ) . Behavioral control and data acquisition were accomplished by custom scripts ( see Source code 1 ) written for use with the TEMPO system ( Reflective Computing , St . Louis , MO , USA ) . Visual stimuli were presented for a duration of 1500 ms during each trial and consisted of various combinations of eight heading directions in the horizontal plane ( Figure 1C ) and two rotational directions ( leftward and rightward ) . Translation and rotation velocities followed a trapezoidal profile in which the velocity was constant ( translation: 24 cm/s , rotation: 17°/s ) during the middle 750 ms ( Figure 1D ) of the stimulus period . The optic flow stimuli were generated using a 3D rendering engine ( OpenGL ) to accurately simulate combinations of observer translation and rotation . In the 3D cloud protocol , the virtual environment consisted of a cloud of dots that was 150 cm wide , 100 cm tall , 160 cm deep and had a density of 0 . 002 dots/cm3 . The part of the cloud visible to the monkey was clipped in depth to range from 25 cm to 125 cm ( relative to the observer ) at all times . This clipping ensured that the same volume of dots was visible to the monkey over the duration of a trial as we simulated a translation of 27 cm through the cloud . The stimulus was rendered as a red-green anaglyph that the monkey viewed stereoscopically through red/green filters . In the second experimental protocol , a fronto-parallel plane ( FP ) of dots was rendered with a density of 0 . 2 dots/cm2 . The plane was rendered with zero binocular disparity and was viewed by the monkey either binocularly or monocularly , without any red/green filters . During the course of a trial ( 1500 ms ) , the 27 cm translation resulted in the simulated distance of the wall changing from 45 cm at the beginning , to 18 cm at the end . We simulated this change in wall distance to better replicate the real world situation of approaching a fronto-parallel wall . Apart from replacing the 3D cloud with a FP plane and the removal of binocular disparity in the stimuli , all other experimental parameters ( such as velocity profiles , trial types , stimulus duration , etc ) were the same as in the 3D cloud experiment . During each session , the monkey's eye position was monitored online using the implanted scleral search coil . Only trials in which the monkey's eye remained within a pre-determined eye window ( see below ) were rewarded with a drop of juice . Trials were aborted if the eye position constraints set by the eye window were violated . The experiment consisted of three main trial types: pure translation , translation + real eye pursuit ( RP ) , and translation + simulated pursuit ( SP ) . ( i ) For the pure translation condition , the monkey fixated a visual target at the center of the screen and maintained fixation within a 2° eye window while the optic flow stimuli were presented . Optic flow stimuli simulated eight headings within the horizontal plane , corresponding to all azimuth angles in 45° steps . The pure translation stimuli were rendered by translating the OpenGL camera along one of the eight headings with the velocity profile shown in Figure 1D . ( ii ) For the real pursuit ( RP ) condition , the animal actively pursued a moving target while the same translational optic flow stimuli as above were presented on the display screen . A rightward rotation trial started when the fixation target appeared 9 . 5° to the left of center . Once the monkey fixated this target ( within 1000 ms ) , it moved to the right following a trapezoidal velocity profile ( Figure 1D ) . Analogously , leftward pursuit trials began with the target appearing on the right and moving leftward . The monkey was required to pursue the moving visual target and maintain gaze within a 4° eye window during the acceleration and deceleration periods ( 0:375 ms and 1125:1500 ms ) . During the middle 750 ms of the trial ( constant velocity phase ) , the monkey was required to maintain gaze within a 2° window around the visual target . Importantly , the optic flow stimulus was windowed with a software rendered aperture that moved simultaneously with the pursuit target . Thus , the area of the retina being stimulated during the RP trials remained constant over time , eliminating potential confounds from moving the stimulus across the receptive field over time ( see ‘Discussion’ ) . ( iii ) For the simulated pursuit ( SP ) condition , optic flow stimuli accurately simulated combinations of the same eight headings with leftward or rightward rotations , while the monkey fixated at the center of the screen ( 2° window ) . These stimuli were rendered by translating and rotating the OpenGL camera with the same trapezoidal velocity profile of the moving target in the RP condition . This ensured that the retinal optic flow patterns in the RP and SP conditions were identical ( assuming accurate pursuit in the RP condition ) . The area of retinal stimulation was also identical in the SP and RP conditions . In addition to these main stimulus conditions , the experimental protocol also included three types of pure rotation conditions for both leftward and rightward directions: ( i ) eye pursuit over a black background ( with the projector on ) , ( ii ) eye pursuit over a static field of dots , and ( iii ) pure rotational optic flow in a 3D cloud ( simulated rotation-only ) . We also included a blank screen during visual fixation and a static field of dots during fixation to measure the spontaneous activity and baseline visual response of the neurons , respectively . Therefore , each block of trials ( for both 3D cloud and FP protocols ) consisted of 48 unique stimulus conditions: eight directions * ( 1 translation only +2 RP + 2 SP ) + 8 controls . To record from single neurons extracellularly , tungsten microelectrodes ( FHC; tip diameter , 3 µm; impedance , 1–3 MΩ at 1 kHz ) were inserted into the cortex through a transdural guide tube , using a hydraulic microdrive . Neural voltage signals were amplified , filtered ( 400–5000 Hz ) , discriminated ( Plexon Systems ) , and displayed on SpikeSort software ( Plexon systems ) . The times of occurrence of action potentials and all behavioral events were digitized and recorded with 1 ms resolution . Eye position was monitored online and recorded using the implanted scleral search coil . Raw neural signals were also digitized at a rate of 25 kHz using the Plexon system for off-line spike sorting . VIP was first identified using MRI scans as described in detail in Chen et al . ( 2011 ) . Electrode penetrations were then directed to the general area of gray matter around the medial tip of the intraparietal sulcus with the goal of characterizing the entire anterior-posterior extent of area VIP—typically defined as the intraparietal area with directionally selective visual responses ( Colby et al . , 1993; Duhamel et al . , 1998 ) . To determine direction selectivity , we presented a patch of drifting dots for which the size , position , and velocity could be manipulated manually with a computer mouse . We used this mapping procedure to characterize the presence or absence of strong visual drive as well as the direction and speed selectivity of multi-unit and single-unit activity . At each location along the anterior–posterior axis , we first identified the medial tip of the intraparietal sulcus and then moved laterally until there was no longer a directionally selective visual response in the multi-unit activity . During each experimental session , we inserted a single microelectrode into the region of cortex identified as VIP . Single unit action potentials were then isolated online using a dual voltage-time window discriminator . Within the region of gray matter identified as VIP , we recorded from any neuron that showed robust visual responses during our search procedure . Once a single unit was isolated , we ran the 3D cloud protocol with all conditions randomly interleaved ( 72 neurons ) . Each stimulus was repeated at least four , and usually five , times . At the end of the 3D cloud protocol , if isolation of the neuron remained stable , we ran the fronto-parallel plane ( FP ) protocol for 4–5 repetitions ( 34 neurons ) . For the FP protocol , the red/green stereo glasses were either removed during the binocular viewing sessions ( 11/34 ) , or replaced with an eye patch during the monocular viewing sessions ( 23/34 ) , such that the eye ipsilateral to the recording hemisphere was occluded . Analysis of spike data and statistical tests were performed using MATLAB ( MathWorks ) . Tuning curves for the different stimulus conditions ( translation only , RP , SP ) were generated using the average firing rate of the cell ( spikes/s ) during the middle 750 ms of each successfully completed trial . This analysis window was chosen such that rotation/translation velocities were constant and the monkey was pursuing or fixating the visual target in the small 2° window . To determine the effect of rotations on neural responses , the translation only tuning curve was compared to the RP/SP tuning curves . Previous studies ( Bradley et al . , 1996; Page and Duffy , 1999; Shenoy et al . , 1999 , 2002; Zhang et al . , 2004; Kaminiarz et al . , 2014 ) only measured tuning curves over a narrow range of headings around straight ahead . Without measuring the full tuning curve , it is very difficult to distinguish between gain fields and shifts in the tuning curves ( Mullette-Gillman et al . , 2009; Chang and Snyder , 2010; Rosenberg and Angelaki , 2014 ) . Furthermore , these previous studies assumed that rotations would cause a global shift of the tuning curve in the absence of pursuit compensation . However , as shown in Figure 2 and Figure 2—figure supplement 1 , rotations can change the shape of the tuning curve , including both skew and bandwidth changes . Therefore , we suspect that the cross-correlation methods or rank-ordering of responses used in previous studies are insufficient to characterize all of the changes in heading tuning due to rotations ( see also Figure 4—figure supplement 2 ) . To account for these more complex changes in heading tuning curves , we developed a novel 3-step analysis procedure , as illustrated for an example cell in Figure 4—figure supplement 1 . Step 1: we measured the minimum and maximum responses of the pure translation tuning curve . The lowest response ( trough ) and amplitude ( maximum—minimum ) of the RP/SP tuning curves were then matched to those of the pure translation curve by vertically shifting and scaling the responses , respectively . Step 2: because the predicted effects of rotation are opposite for forward and backward headings ( Figure 2A ) , RP and SP tuning curves were split into heading ranges of 0–180° and 180–360° . We tested whether each half of the tuning curve was significantly tuned using an ANOVA ( p ≤ 0 . 05 ) . All the tuning curves were then linearly interpolated to a resolution of 1° . Step 3: for half-curves that showed significant tuning , we performed a shift analysis as follows . The pure translation tuning curve was circularly shifted ( in steps of 1° ) to minimize the sum-squared error with each half of the RP/SP tuning curves . For neurons that were significantly tuned in all conditions and in both direction ranges , this analysis yielded four shift values for real pursuit and four shifts for simulated pursuit . In order to quantify the transformation of heading tuning due to rotations , the four shift values were averaged to arrive at one shift value for real pursuit and one shift for simulated pursuit for each cell . The 95% confidence intervals ( CIs ) for the shifts plotted in Figures 4 and 6C , were calculated using a bootstrap analysis . Bootstrapped tuning curves for translation only , real pursuit , and simulated pursuit were generated by resampling responses with replacement . The same offset , gain and shift calculations were performed on each one of 300 bootstrapped tuning curves to produce a distribution of shifts for each neuron from which the 95% CI was calculated by the percentile method . In order to test the efficacy of our analysis method , we simulated heading tuning curves using von Mises functions ( Equation 1 ) , with gain ( A ) , preferred direction ( φ ) , and width ( k ) as free parameters ( Swindale , 1998 ) . ( 1 ) VM ( θ ) =Ae{k[cos ( θ−φ ) −1]} . To simulate the tuning curve transformations caused by adding rotational optic flow , a second shape parameter ( Equation 2 ) and skew ( Equation 3 ) terms were added to the von Mises functions as follows: ( 2 ) VMwidth ( θ ) =Ae{k[cos ( θ−φ+σsin ( θ−φ ) ) −1]} , ( 3 ) VMskew ( θ ) =Ae{k[cos ( θ−φ+γ ( cos ( θ−φ ) −1 ) −1]} , where , σ is the second shape parameter ( such that slope of the function at half-height can vary independently of the width at half-height ) and γ is the skew parameter ( see Swindale , ( 1998 ) for more details ) . The second shape parameter ( σ ) was manipulated to yield rotation-added tuning curves with bandwidth changes of 40° ( 20° on each half of the tuning curve ) for cells preferring close to lateral translations ( [340°:20°] , [160°:200°] ) . For cells preferring all other headings ( close to forward or backward translations ) , the skew parameter ( γ ) was manipulated to yield a 20° shift in the peaks of the rotation-added tuning curves . Random gain values ranging from 0 . 66 to 1 . 33 were used to scale the rotation-added tuning curves and random offset values ( 0–40 spikes/s ) were also added to the tuning curves corresponding to leftward and rightward rotations . Poisson random noise was added to all tuning curves ( averaged over five simulated stimulus repetitions ) and the curves were sampled at heading intervals of 45° , similar to the recorded data . Shifts were measured between the translation only and rotation-added curves using the partial shift analysis method described above . The mean shifts resulting from 10 sets of simulated tuning curves with heading preferences ranging from 0:360° are shown in Figure 4—figure supplement 1E . These simulations demonstrate that our method is capable of accurately measuring shifts in the presence of gain , offset and shape changes for neurons with a variety of heading preferences . To compare our method with the cross-correlation method used in previous studies ( Bradley et al . , 1996; Shenoy et al . , 1999 , 2002 ) , von Mises functions with Poisson noise were generated as described above ( Equations 1–3 ) , but were sampled and analyzed as described in those papers . Specifically , simulated tuning curves were generated by sampling the von Mises functions at headings in the range of ±32° around straight ahead , with 8° sampling intervals . To match the previous studies , the resulting data were then smoothed with a three-point moving average and interpolated using a spline function at 1° intervals ( Figure 4—figure supplement 2C , E ) . The rotation-added tuning curves were horizontally shifted in 1° increments relative to the translation-only curve and the maximum correlation coefficient between the curves was measured using the equation described in Shenoy et al . ( 1999 ) ( Figure 4—figure supplement 2D , F ) . This analysis was repeated for 10 sets of simulated tuning curves ( different random noise samples ) for 10 different heading preferences in the range from 0:180° ( Figure 4—figure supplement 2G ) . Since this analysis was based only on the narrow heading range of ±32° around straight forward , we did not simulate neurons with backwards heading preferences in the range of 180–360° because such neurons would have little response in this heading range . In contrast with our analysis , this cross-correlation method resulted in unreliable tuning shifts for simulated neurons with heading preferences outside the narrow range of measured headings ( Figure 4—figure supplement 2G ) . To test the rotating reference frame hypothesis ( Figure 7 ) , the gradient of firing rate was calculated at each heading on each measured tuning curve and the heading associated with the largest positive gradient was selected . The predicted slopes for an eye-centered reference frame were calculated as the average gradient for all the neurons for a given condition ( dashed lines in Figure 7C , D ) . To test whether the temporal responses match this prediction , the time course of firing rate was measured at the heading associated with the largest positive gradient , for neurons recorded during the 3D cloud protocol . For sharply tuned neurons , it is possible that the true largest gradient lay between sampled headings . Hence , the measured largest gradient could be part of the peak or trough of the tuning curve . To account for such instances in the data , we excluded tuning curves for which the mean response at the largest gradient heading was not significantly different ( t-test; p ≤ 0 . 05 ) from the responses of its immediate neighboring headings ( 29/360 total tuning curves from 72 cells ) . The time course of firing rate during each trial for the selected heading was calculated by convolving the spike events with a Gaussian kernel ( σ = 25 ms ) . The temporal responses from all selected tuning curves were averaged by condition and used to calculate the mean and standard errors shown in Figure 7C , D . | When strolling along a path beside a busy street , we can look around without losing our stride . The things we see change as we walk forward , and our view also changes if we turn our head—for example , to look at a passing car . Nevertheless , we can still tell that we are walking in a straight-line because our brain is able to compute the direction in which we are heading by discounting the visual changes caused by rotating our head or eyes . It remains unclear how the brain gets the information about head and eye movements that it would need to be able to do this . Many researchers had proposed that the brain estimates these rotations by using a copy of the neural signals that are sent to the muscles to move the eyes or head . However , it is possible that the brain can estimate head and eye rotations by directly analyzing the visual information from the eyes . One region of the brain that may contribute to this process is the ventral intraparietal area or ‘area VIP’ for short . Sunkara et al . devised an experiment that can help distinguish the effects of visual cues from copies of neural signals sent to the muscles during eye rotations . This involved training monkeys to look at a 3D display of moving dots , which gives the impression of moving through space . Sunkara et al . then measured the electrical signals in area VIP either when the monkey moved its eyes ( to follow a moving target ) , or when the display changed to give the monkey the same visual cues as if it had rotated its eyes , when in fact it had not . Sunkara et al . found that the electrical signals recorded in area VIP when the monkey was given the illusion of rotating its eyes were similar to the signals recorded when the monkey actually rotated its eyes . This suggests that visual cues play an important role in correcting for the effects of eye rotations and correctly estimating the direction in which we are heading . Further research into the mechanisms behind this neural process could lead to new vision-based treatments for medical disorders that cause people to have balance problems . Similar research could also help to identify ways to improve navigation in automated vehicles , such as driverless cars . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"neuroscience"
] | 2015 | Role of visual and non-visual cues in constructing a rotation-invariant representation of heading in parietal cortex |
Intermediate filaments ( IF ) are a major component of the metazoan cytoskeleton and are essential for normal cell morphology , motility , and signal transduction . Dysregulation of IFs causes a wide range of human diseases , including skin disorders , cardiomyopathies , lipodystrophy , and neuropathy . Despite this pathophysiological significance , how cells regulate IF structure , dynamics , and function remains poorly understood . Here , we show that site-specific modification of the prototypical IF protein vimentin with O-linked β-N-acetylglucosamine ( O-GlcNAc ) mediates its homotypic protein-protein interactions and is required in human cells for IF morphology and cell migration . In addition , we show that the intracellular pathogen Chlamydia trachomatis , which remodels the host IF cytoskeleton during infection , requires specific vimentin glycosylation sites and O-GlcNAc transferase activity to maintain its replicative niche . Our results provide new insight into the biochemical and cell biological functions of vimentin O-GlcNAcylation , and may have broad implications for our understanding of the regulation of IF proteins in general .
Intermediate filaments ( IF ) are a major component of the metazoan cytoskeleton , distinct from the actin and microtubule systems ( Lowery et al . , 2015; Herrmann and Aebi , 2016; Chernyatina et al . , 2015; Köster et al . , 2015; Leduc and Etienne-Manneville , 2015 ) . Humans express over 70 IF proteins , including both cytoplasmic ( e . g . , vimentin , keratins , neurofilaments ) and nuclear ( lamins ) members , many with tissue-specific functions ( Szeverenyi et al . , 2008 ) . All IF proteins comprise a central , conserved α-helical rod domain , as well as amino-terminal head and carboxy-terminal tail domains of varying lengths ( Lowery et al . , 2015; Herrmann and Aebi , 2016; Chernyatina et al . , 2015; Köster et al . , 2015; Leduc and Etienne-Manneville , 2015 ) . IF proteins homo- or heterodimerize through the parallel association of their rod domains into coiled coils , forming an elongated dimer of ~45–48 nm for cytoplasmic IFs and ~50–52 nm for nuclear lamins ( Quinlan et al . , 1986; Aebi et al . , 1986 ) . These dimers laterally associate in antiparallel fashion to form tetramers , which in turn assemble into ~65 nm unit-length filaments ( ULFs ) composed of eight tetramers ( Herrmann and Aebi , 2016; Chernyatina et al . , 2015; Herrmann et al . , 1996 ) . Finally , ULFs associate end-to-end to assemble mature IFs , measuring ~10 nm across ( Lowery et al . , 2015; Herrmann and Aebi , 2016; Chernyatina et al . , 2015 ) . Unlike actin- or microtubule-based structures , IFs are nonpolar and do not serve as tracks for molecular motors . Instead , IFs contribute to the mechanical integrity of the cell through their unique viscoelastic properties ( Lowery et al . , 2015; Herrmann and Aebi , 2016; Chernyatina et al . , 2015; Köster et al . , 2015; Leduc and Etienne-Manneville , 2015 ) . In general , the IF network is flexible under low strain but stiffens and resists breakage under an applied force ( Janmey et al . , 1991; Fudge et al . , 2003; Guzmán et al . , 2006; Kreplak et al . , 2005 ) . Remarkably , individual IFs can be stretched up to 3 . 6-fold before rupture , demonstrating their elastic nature , as compared to actin cables or microtubules ( Kreplak et al . , 2005 ) . The IF network is also highly dynamic in vivo , with IF subunits ( likely tetramers ) exchanging rapidly at many points along mature filaments ( Mendez et al . , 2010; Goldman et al . , 2012; Miller et al . , 1991; Vikstrom et al . , 1989; Ho et al . , 1998; Martys et al . , 1999; Vikstrom et al . , 1992; Nöding et al . , 2014 ) . Similarly , the IF cytoskeleton quickly reorganizes in response to numerous physiological cues , including cell cycle progression , migration , spreading , and growth factor stimulation ( Lowery et al . , 2015; Herrmann and Aebi , 2016; Chernyatina et al . , 2015; Köster et al . , 2015; Leduc and Etienne-Manneville , 2015; Yoon et al . , 1998; Helfand et al . , 2003 ) . IFs participate in many cellular processes , including maintenance of cell shape , organelle anchoring , cell motility , and signal transduction ( Helfand et al . , 2011; Ben-Ze'ev , 1984 ) . For example , vimentin , among the most widely studied IF proteins , is required for mesenchymal cell adhesion , migration , chemotaxis , and wound healing in both cell culture and animal models ( Ivaska et al . , 2007; Yamaguchi et al . , 2005; Eckes et al . , 2000; Rogel et al . , 2011; Menko et al . , 2014 ) . Vimentin IFs also contribute to the mechanical properties of the cytoplasm , and stabilize and localize mitochondria ( Nekrasova et al . , 2011; Buehler , 2013; Guo et al . , 2013; Eckes et al . , 1998 ) . Importantly , genetic lesions that dysregulate the IF cytoskeleton cause a wide range of human diseases , including skin and nail disorders ( keratins ) , cardiomyopathies ( desmin ) , lipodystrophy , muscular dystrophy and progeria ( lamins ) , giant axonal neuropathy and Charcot-Marie-Tooth disease ( neurofilaments ) , and cataracts ( vimentin ) ( Omary , 2009; Omary et al . , 2004 ) . In addition , the ectopic expression of wild type ( WT ) vimentin is a hallmark of the epithelial-to-mesenchymal transition , and is widely observed in human metastatic cancers ( Satelli and Li , 2011; Nieto , 2011; De Craene and Berx , 2013 ) . IFs are likely functionally important in this context , because vimentin expression levels correlate with the invasive phenotypes of breast , prostate , and other epithelial cancers ( Satelli and Li , 2011; Wei et al . , 2008; Zhu et al . , 2011; Vuoriluoto et al . , 2011 ) . Finally , the IF cytoskeleton is also intimately involved in host-microbe interactions ( Geisler and Leube , 2016; Mak and Brüggemann , 2016 ) . For example , changes in vimentin IFs are implicated in the adhesion , invasion , and replication of a wide range of bacteria , including such important pathogens as Chlamydia trachomatis ( Kumar and Valdivia , 2008; Jorgensen et al . , 2011; Snavely et al . , 2014; Bednar et al . , 2011 ) , Mycobacterium tuberculosis ( Garg et al . , 2006; Mahesh et al . , 2016 ) , Streptococcus pyogenes ( Bryant et al . , 2006; Icenogle et al . , 2012; Lin et al . , 2015 ) and Salmonella enterica ( Murli et al . , 2001; Guignot and Servin , 2008 ) . Despite this broad pathophysiological significance , the regulation of IF cytoskeleton morphology , dynamics , and signaling functions remains incompletely understood . Several recent lines of evidence indicate that post-translational modifications ( PTMs ) are an important mode of IF regulation , and indeed all IFs are subject to extensive PTMs , including phosphorylation , ubiquitination , sumoylation , acetylation , farnesylation , and glycosylation ( Snider and Omary , 2014 ) . However , in most cases , the functional impact of these PTMs on IF structure and function is poorly characterized . To better understand the dynamic regulation of the IF cytoskeleton , we focused on O-linked β-N-acetylglucosamine ( O-GlcNAc ) , an intracellular form of protein glycosylation that reversibly decorates serine and threonine residues on thousands of nuclear , cytoplasmic , and mitochondrial proteins . In mammals , O-GlcNAc is added by O-GlcNAc transferase ( OGT ) and removed by O-GlcNAcase ( OGA ) ( Figure 1A ) ( Hanover et al . , 2010; Hart et al . , 2011; Hart , 2014 ) . O-GlcNAc cycling controls many processes , including nutrient sensing , cell cycle progression , and apoptosis ( Hanover et al . , 2010; Hart et al . , 2011 ) , and is essential , as genetic ablation of OGT or OGA is lethal in mice ( Shafi et al . , 2000; Keembiyehetty et al . , 2015; Yang et al . , 2012 ) . In addition , aberrant O-GlcNAc cycling is implicated in numerous human diseases , including cancer ( Hart et al . , 2011; Ma and Vosseller , 2013; Yi et al . , 2012; Singh et al . , 2015 ) , diabetes ( Vaidyanathan and Wells , 2014; Hardivillé and Hart , 2014; Ma and Hart , 2013 ) , cardiac dysfunction ( Erickson et al . , 2013; Erickson , 2014; Dassanayaka and Jones , 2014; Darley-Usmar et al . , 2012 ) , and neurodegeneration ( Yuzwa et al . , 2012; Vaidyanathan et al . , 2014; Yuzwa and Vocadlo , 2014; Zhu et al . , 2014 ) . Interestingly , numerous IF proteins are O-GlcNAcylated ( King and Hounsell , 1989; Chou et al . , 1992 ) . For example , keratin-18 glycosylation is required for the recruitment and activation of the pro-survival kinase Akt , and mice expressing an unglycosylatable keratin-18 mutant are sensitized to chemical injury of the liver and pancreas ( Ku et al . , 2010; Ku et al . , 1996 ) . Several neurofilament proteins are also glycosylated , especially in axons , and neurofilament-M O-GlcNAcylation is reduced in both human Alzheimer’s disease patients and a rat model of amyotrophic lateral sclerosis , suggesting a potential role for dysregulated IF glycosylation in neurodegeneration ( Dong et al . , 1996; Dong et al . , 1993; Lüdemann et al . , 2005; Deng et al . , 2008; Cheung and Hart , 2008 ) . Finally , vimentin is O-GlcNAcylated on several sites , primarily in its head domain ( Slawson et al . , 2008; Wang et al . , 2007 ) . Changes in vimentin glycosylation have been observed in models of differentiating adipocytes ( Ishihara et al . , 2010 ) and neurons ( Farach and Galileo , 2008 ) , and a recent study observed a correlation between vimentin O-GlcNAcylation and the invasive potential of cholangiocarcinoma ( Phoomak et al . , 2017 ) , implicating glycosylation changes in both the physiological and pathological functions of vimentin . These studies suggest that O-GlcNAcylation is a prominent mode of IF regulation in homeostasis and disease alike . Nevertheless , the mechanistic and functional impacts of O-GlcNAcylation on IF proteins remain largely unexplored . Here , we report that vimentin O-GlcNAcylation is required for the structure and function of IFs in human cells . We demonstrate that site-specific glycosylation of vimentin mediates its self-association , and is required in human cells for both IF morphology and its facilitating role in cell migration . In addition , we provide evidence that vimentin glycosylation is co-opted by an intracellular bacterial pathogen to produce a replicative niche , revealing a new connection between IF O-GlcNAcylation and microbial pathogenesis . Our results provide new insight into the physiological role of vimentin glycosylation in particular , and may have broad implications for our understanding of the dynamic regulation of the IF cytoskeleton in general .
Vimentin is O-GlcNAcylated at multiple sites in its head domain ( Slawson et al . , 2008; Wang et al . , 2007 ) , but the functional consequences of this modification are poorly understood . Like other intracellular PTMs , O-GlcNAc can exert a wide range of biochemical effects on its substrates , including conformational change , relocalization or destruction ( Hanover et al . , 2010; Hart et al . , 2011; Shafi et al . , 2000; Keembiyehetty et al . , 2015; Mondoux et al . , 2011; Bond and Hanover , 2013 ) . Because IF proteins self-associate into homo-oligomeric complexes , we hypothesized that vimentin O-GlcNAcylation might influence its protein-protein interactions . Indeed , O-GlcNAc regulates protein-protein interactions in other contexts , including chromatin remodeling , nutrient sensing , and the interaction between keratins and Akt ( Ku et al . , 2010; Fujiki et al . , 2011; Tarbet et al . , 2018; Dentin et al . , 2008 ) . However , physiological O-GlcNAc-mediated interactions are often low-affinity , sub-stoichiometric , and transient , presenting a technical barrier to studying them ( Hanover et al . , 2010; Hart et al . , 2011; Shafi et al . , 2000; Keembiyehetty et al . , 2015; Mondoux et al . , 2011; Bond and Hanover , 2013; Tarbet et al . , 2018 ) . To address these challenges , we used a chemical biology method to capture and characterize O-GlcNAc-mediated protein-protein interactions in living cells ( Yu et al . , 2012 ) . In this strategy , cells are metabolically labeled with a GlcNAc analog bearing a diazirine photocrosslinking moiety , termed ‘GlcNDAz' ( Yu et al . , 2012 ) . GlcNDAz is accepted by the GlcNAc salvage pathway , converted to the nucleotide-sugar UDP-GlcNDAz , and used by OGT to decorate its native substrates ( Yu et al . , 2012 ) . Brief treatment of GlcNDAz-labeled live cells with UV light triggers the covalent crosslinking of O-GlcNDAz moieties to any binding partner proteins within ~2–4 Å of the sugar ( Yu et al . , 2012 ) . Because of this short radius , GlcNDAz crosslinking occurs exclusively at sites where the glycan contributes to the interaction interface , without crosslinking to distant or nonspecific proteins ( Yu et al . , 2012 ) . Therefore , GlcNDAz is a powerful tool for identifying direct , glycosylation-mediated interactions between endogenous proteins in live cells ( Yu et al . , 2012 ) . To determine whether vimentin engages in O-GlcNAc-mediated protein-protein interactions , we treated human cells with GlcNDAz and UV , and analyzed lysates by immunoblot ( IB ) . Endogenous vimentin crosslinked into distinct , high molecular weight bands that were diazirine- and UV-dependent ( Figure 1B ) , indicating that O-GlcNAc mediates protein-protein interactions between vimentin and one or more binding partners . Overexpression of OGT caused an increase in vimentin crosslinks ( Figure 1C ) , whereas overexpression of OGA ( Figure 1C ) or treatment with a small molecule inhibitor of OGT , abbreviated 5SGlcNAc ( Gloster et al . , 2011 ) ( Figure 1—figure supplement 1A ) , reduced crosslinking , confirming that this assay reports on authentic O-GlcNAc-mediated interactions , and not a nonspecific action of the GlcNDAz probe . We next sought to identify the O-GlcNAc-mediated binding partner ( s ) of vimentin detected in our crosslinking assay . We created a myc-6xHis-tagged human vimentin construct and confirmed that it crosslinked via GlcNDAz similarly to endogenous vimentin ( Figure 1—figure supplement 1B ) . Then , we purified preparative amounts of vimentin crosslinks from transfected cells via tandem immunoprecipitation ( IP ) and immobilized metal affinity chromatography , and analyzed the crosslinks by mass spectrometry ( MS ) -based proteomics ( Figure 1—figure supplement 1C , Figure 1—figure supplement 2D , E and Figure 1—source data 1 ) . We obtained 80% tryptic peptide coverage of vimentin in these samples , without significant enrichment of other proteins , beyond common contaminants ( Figure 1—figure supplement 2 and Figure 1—source data 1 ) . This result suggested that the vimentin GlcNDAz crosslinks represent homotypic , O-GlcNAc-mediated vimentin-vimentin interactions . In vivo , vimentin exists in a range of assembly states , from soluble tetramers and ULFs , to relatively insoluble mature IFs ( Lowery et al . , 2015; Herrmann and Aebi , 2016; Chernyatina et al . , 2015; Köster et al . , 2015; Leduc and Etienne-Manneville , 2015 ) . To determine the assembly state of the crosslinked vimentin species , we performed a well-established differential extraction procedure ( Ridge et al . , 2016 ) on GlcNDAz-crosslinked cells . The GlcNDAz crosslinks of endogenous vimentin extracted into a denaturing buffer but not low or high ionic strength non-denaturing buffers , indicating that the crosslinks occur within the highly assembled filament population ( Figure 1D ) ( Ridge et al . , 2016 ) . Importantly , extracted crosslinks remained soluble when exchanged from denaturing into non-denaturing buffers , demonstrating that GlcNDAz crosslinking per se does not reduce vimentin solubility ( Figure 1—figure supplement 2F ) . To further confirm that crosslinks occur within assembled IFs , we treated cells with β , β’-iminodipropionitrile ( IDPN ) , which blocks vimentin assembly beyond the ULF state ( Durham , 1986 ) . Consistent with previous reports , IDPN treatment abrogated vimentin IFs and caused the accumulation of vimentin aggregates ( Figure 1—figure supplement 2G ) ( Durham , 1986 ) . In the crosslinking assay , IDPN also suppressed the formation of GlcNDAz-dependent adducts , indicating that the crosslinks occur in assembly states beyond ULFs ( Figure 1E ) . Using quantitative IBs , we detected ~10% of endogenous vimentin crosslinked into GlcNDAz-dependent adducts ( Figure 1—figure supplement 2H ) , though this measurement likely significantly underestimates the fraction of vimentin that engages in O-GlcNAc-mediated interactions , since several steps in the GlcNDAz crosslinking protocol are less than 100% efficient ( Yu et al . , 2012; Rodriguez et al . , 2015 ) . Based on these results , we concluded that O-GlcNAc-mediated vimentin-vimentin interactions are prevalent in cells , and occur primarily within assembled IFs , but not in soluble pools of lower-order assembly states . The Hart lab previously mapped several glycosylation sites on vimentin’s flexible head domain ( Slawson et al . , 2008; Wang et al . , 2007 ) . We used these results to identify vimentin O-GlcNAc sites that contribute to its homotypic , glycosylation-mediated interactions . We mutated each reported O-GlcNAcylation site to alanine and screened these constructs in the GlcNDAz crosslinking assay ( Figure 2 ) . Mutation of several individual residues , including T33 , S34 , and S39 , reduced vimentin crosslinking , while mutation of S49 abolished all detectable crosslinks . Because of the dramatic reduction of crosslinking in the S49A mutant , we measured the fraction of total vimentin glycosylation occurring at S49 . We transfected cells with vector , WT or S49A vimentin and then labeled them with GalNAz , an azide-bearing unnatural monosaccharide that we have described previously as a metabolic reporter of O-GlcNAcylation ( Boyce et al . , 2011; Palaniappan et al . , 2013; Chen et al . , 2017 ) . S49A mutant vimentin exhibited ~80% the GalNAz signal of WT ( Figure 2—figure supplement 1 ) , suggesting that approximately one-fifth of vimentin glycosylation occurs on S49 under these conditions . Together , these results suggest that robust and site-specific O-GlcNAcylation in the vimentin head domain mediates homotypic protein-protein interactions within IFs . We next tested whether O-GlcNAc-mediated interactions influence vimentin IF structure or function in live human cells . We used CRISPR/Cas9 methods to delete endogenous vimentin from two different human cell lines , selected single cell-derived clones , and confirmed the absence of vimentin mRNA and protein ( Figure 3—figure supplement 1A , B ) . Then , we stably reconstituted individual vimentin−/− clones with WT or unglycosylatable point-mutant versions of a well-characterized vimentin-mEmerald fusion protein ( Mendez et al . , 2010; Yoon et al . , 1998; Helfand et al . , 2011; Hookway et al . , 2015 ) to permit live-cell visualization of IFs . We verified that vimentin-mEmerald was expressed at uniform levels comparable to endogenous vimentin ( Figure 3—figure supplement 1C , D ) , and that the mEmerald signal precisely coincided with anti-vimentin immunofluorescence in cells expressing both endogenous vimentin and vimentin-mEmerald ( Figure 3—figure supplement 2E ) . These results confirmed that the vimentin-mEmerald construct is a faithful proxy for the untagged protein in this system . In our panels of vimentin-mEmerald-reconstituted cells , we included Y117L , a point-mutation in the rod domain that blocks vimentin assembly beyond the ULF stage ( Meier et al . , 2009 ) , as a positive control for IF disruption . We used fluorescence-activated cell sorting , IB and fluorescence microscopy to ensure equal expression of the various WT and mutant vimentin-mEmerald transgenes across reconstituted vimentin−/− cell lines ( Figure 3A , Figure 3—figure supplements 1D , 3A ) . Cells reconstituted with WT vimentin-mEmerald exhibited canonical IF morphology , whereas the Y117L-expressing cells lacked assembled filaments and instead displayed punctate structures consistent with ULFs ( Figure 3A , B , Figure 3—figure supplement 3A , B ) . These results indicate that our reconstituted cell systems recapitulate the previously reported characteristics of WT and Y117L vimentin ( Meier et al . , 2009; Helfand et al . , 2011; Robert et al . , 2014 ) . We also observed dramatic alterations in vimentin IF organization in several unglycosylatable point-mutants . For example , the S34A mutant , which had an intermediate phenotype in our crosslinking assay ( Figure 2 ) , displayed a partial defect in IF formation , with both punctate and filamentous fluorescence detected ( Figure 3A , Figure 3—figure supplement 3A ) . Interestingly , the S49A mutant , which lacked detectable O-GlcNAc-mediated interactions in the GlcNDAz assay ( Figure 2 ) , exhibited a significant reduction of assembled IFs in live cells , displaying a higher proportion of punctate vimentin fluorescence as compared to WT ( Figure 3A , B , Figure 3—figure supplement 3A , B ) . Importantly , the S49A mutation had no impact on vimentin stability ( Figure 3—figure supplement 2 ) , further ruling out an effect of vimentin expression level in this system . These results suggested that O-GlcNAc on specific residues of vimentin , particularly S49 , may be required for its homotypic assembly into mature IFs in live cells . However , S/T→A mutations eliminate all O-linked PTMs at the mutated site , including both O-GlcNAcylation and phosphorylation . Therefore , phenotypes arising from S/T→A mutations may be due to loss of phosphorylation , glycosylation , or both . Vimentin S49 is a known glycosylation site ( Slawson et al . , 2008; Wang et al . , 2007 ) but not a reported phosphorylation site , suggesting that the phenotypes we observed in the S49A mutant ( Figures 2 and 3 , Figure 3—figure supplement 3 ) are due to the loss of O-GlcNAc , and not the loss of phosphorylation . To further test this hypothesis , we reconstituted vimentin−/− cells with a phosphomimetic S49E mutant . Vimentin-S49E-expressing cells displayed a loss of filament morphology and a proportion of puncta indistinguishable from the S49A mutant ( Figure 3 and Figure 3—figure supplement 3 ) . These results suggest that the abnormal vimentin IF structures observed with the S49A mutant in live cells are due to the loss of glycosylation , not loss of phosphorylation , at S49 . We next investigated whether vimentin O-GlcNAcylation is required for known functions of IFs . The IF cytoskeleton facilitates cell migration , and vimentin−/− cells and tissues exhibit migration defects ( Ivaska et al . , 2007; Yamaguchi et al . , 2005; Eckes et al . , 2000; Rogel et al . , 2011; Menko et al . , 2014 ) . We used a well-characterized Transwell assay ( Justus et al . , 2014 ) to measure cell migration by vimentin-reconstituted cell lines across a collagen matrix . Consistent with prior reports in other systems ( Ivaska et al . , 2007; Yamaguchi et al . , 2005; Eckes et al . , 2000; Rogel et al . , 2011; Menko et al . , 2014 ) , vimentin−/− cells stably transduced with empty vector or the Y117L mutant construct were impaired in serum-stimulated migration , relative to cells expressing WT vimentin ( Figure 4 , Figure 4—source data 1 ) . Interestingly , vimentin-S49A-expressing cells also exhibited migration defects compared to WT ( Figure 4 , Figure 4—source data 1 ) . This result suggested that O-GlcNAc on vimentin may be required for optimal cell migration . To further confirm this hypothesis , we assayed the serum-induced migration of vimentin-reconstituted cells upon treatment with 5SGlcNAc or Thiamet-G , a specific small molecule inhibitor of OGA ( Yuzwa et al . , 2008 ) ( Figure 4 , Figure 4—source data 1 ) . 5SGlcNAc or Thiamet-G treatment each significantly inhibited the serum-induced migration of cells expressing WT vimentin , but not cells lacking vimentin or expressing the S49A mutant ( Figure 4 , Figure 4—source data 1 ) . Taken together , these results indicate that O-GlcNAcylation of vimentin , especially on S49 , is required for optimal serum-stimulated cell migration . Several intracellular pathogens co-opt the IF cytoskeleton to stabilize the large membrane-bound vacuoles in which they replicate , presumably by providing structural scaffolds ( Geisler and Leube , 2016; Mak and Brüggemann , 2016 ) . For example , we previously showed that the obligate intracellular pathogen Chlamydia trachomatis recruits a meshwork of vimentin IFs to its vacuolar ‘inclusion’ compartment , stabilizing its replicative niche and shielding bacterial components from the host’s cytoplasmic innate immune surveillance machinery ( Kumar and Valdivia , 2008; Jorgensen et al . , 2011; Snavely et al . , 2014; Bednar et al . , 2011 ) . The molecular mechanisms of Chlamydia-induced vimentin recruitment and reorganization are incompletely understood . To examine the potential role of vimentin O-GlcNAcylation in these processes , we treated HeLa cells reconstituted with WT vimentin-mEmerald with vehicle control , 5SGlcNAc or Thiamet-G , infected them with Chlamydia , and visualized pathogen-containing inclusions and vimentin IFs by fluorescence microscopy ( Figure 5A , Figure 5—figure supplement 1 ) . We found that 5SGlcNAc treatment inhibited the formation of a vimentin meshwork around the inclusion , consistent with a requirement for vimentin glycosylation in IF remodeling during Chlamydia infection ( Figure 5A ) . 5SGlcNAc treatment reduced the average size of the inclusions in infected cells and increased the number of extra-inclusion bacteria ( Figure 5B ) , demonstrating that OGT activity is required for inclusion expansion and integrity . Thiamet-G treatment had little effect in these assays , likely because basal levels of O-GlcNAcylation are already sufficient for optimal Chlamydia replication ( Figure 5 , Figure 5—figure supplement 1 ) . We next tested whether O-GlcNAcylation of vimentin itself is required for inclusion integrity . We infected vimentin-reconstituted HeLa cell lines with Chlamydia and visualized vimentin IFs and bacteria by fluorescence microscopy . Compared to WT vimentin-expressing cells , cells expressing either the S49A or Y117L mutant vimentin displayed reduced IF recruitment to the inclusions , smaller average inclusion size , and larger numbers of bacteria escaping into the cytoplasm ( Figure 5C , D ) . These data suggest that the site-specific glycosylation of vimentin itself is required for IF remodeling during Chlamydia infection . To further test this hypothesis , we infected empty vector-reconstituted vimentin−/− cells with Chlamydia in the presence or absence of 5SGlcNAc . In contrast to our observations with cells expressing WT vimentin ( Figure 5A , B ) , 5SGlcNAc had no impact on Chlamydia inclusion size or extra-inclusion bacteria in cells lacking vimentin , indicating that vimentin , but not other host- or pathogen-encoded targets , is required for the effects of 5SGlcNAc in this context ( Figure 5E ) . We concluded that both OGT activity and vimentin glycosylation sites are required for IF reorganization during Chlamydia infection , and are co-opted by this pathogen to promote inclusion integrity and growth .
The IF cytoskeleton plays critical roles in both physiological and pathological processes , but how cells regulate IF structure and function remains poorly understood . We provide evidence that site-specific glycosylation of the vimentin head domain regulates its homotypic association in human cells , and is required for both IF morphology and cell migration ( Figure 6 ) . In addition , site-specific O-GlcNAcylation of vimentin is exploited by an intracellular pathogen to promote its own replication , underlining the importance of IF dynamics in disease states . Because many IF proteins are O-GlcNAc-modified ( King and Hounsell , 1989; Chou et al . , 1992; Ku et al . , 2010; Dong et al . , 1996; Dong et al . , 1993; Lüdemann et al . , 2005; Deng et al . , 2008; Cheung and Hart , 2008; Slawson et al . , 2008; Wang et al . , 2007; Srikanth et al . , 2010; Kakade et al . , 2016; Tao et al . , 2006 ) , our results may provide new insight into the regulation of both vimentin in particular and the IF cytoskeleton in general . Early evidence for vimentin glycosylation was reported nearly twenty-five years ago ( Shikhman et al . , 1993 ) and confirmed with sophisticated MS methods thirteen years later ( Slawson et al . , 2008; Wang et al . , 2007 ) . However , the biochemical and cellular effects of this modification have remained largely uncharacterized . Through GlcNDAz crosslinking ( Figure 2 ) and live-cell imaging ( Figure 3 , Figure 3—figure supplement 3 ) , we determined that the vimentin glycosylation sites S34 , S39 and especially S49 are required for normal homotypic vimentin-vimentin interactions and for IF morphology in live human cells . These O-GlcNAc-mediated interactions likely occur within assembled IFs and not smaller oligomeric states ( Figure 1D , E ) . It is well established that portions of the vimentin head domain are required for filament assembly ( Eriksson et al . , 2004 ) , and that head domain PTMs ( especially phosphorylation ) govern IF dynamics in vivo ( Helfand et al . , 2011; Eriksson et al . , 2004; Chou et al . , 1990; Eriksson et al . , 1992; Sihag et al . , 2007; Goto et al . , 2002; Chan et al . , 2002 ) . Our results are consistent with this general model of regulated filament assembly/disassembly through head domain PTMs ( Figure 6 ) . Interestingly , however , internal deletion mutations within the head domain demonstrate that S49 itself is not required for vimentin filament formation in vitro or in vivo ( Shoeman et al . , 2002 ) . Therefore , we propose that the phenotypes observed in the S49A vimentin mutant are caused by a loss of PTM regulation , rather than simply a requirement for serine at that site . Given the very short crosslinking radius of the GlcNDAz reagent ( ~2–4 Å ) ( Yu et al . , 2012 ) and our MS proteomics analysis of the crosslinked complexes ( Figure 1—figure supplement 1C and Figure 1—figure supplement 2D , E ) , the O-GlcNAc-mediated interactions we observe are likely homotypic dimeric or trimeric vimentin adducts . The precise molecular nature of these crosslinks , and why we observe multiple discrete crosslinked complexes , remain uncertain . The high MS proteomic coverage we obtained of vimentin and the lack of other proteins in the crosslinks ( Figure 1—figure supplement 2 and Figure 1—source data 1 ) suggest that the complexes do not contain other binding partner proteins or known protein-based PTMs of vimentin , such as SUMO ( Kaminsky et al . , 2009; Wang et al . , 2010 ) . Because vimentin is multiply phosphorylated ( Helfand et al . , 2011; Eriksson et al . , 2004; Chou et al . , 1990; Eriksson et al . , 1992; Sihag et al . , 2007; Goto et al . , 2002; Chan et al . , 2002 ) , the crosslinks of distinct molecular weight ( Figures 1B and 2 ) could in principle represent different phospho-forms of vimentin . However , several commercially available anti-phospho-vimentin antibodies failed to recognize the crosslinked complexes by IB ( not shown ) , and phosphatase treatment of crosslinked adducts had no effect on their SDS-PAGE migration ( Figure 1—figure supplement 3 ) , arguing against differential phosphorylation as an explanation . Instead , we postulate that the discrete bands in the GlcNDAz-induced vimentin complexes represent different crosslinking geometries of vimentin multimers . Because the O-GlcNAc-mediated interactions we detect occur primarily in assembled IFs ( Figure 1D , E ) , an individual glycan on the vimentin head domain ( e . g . , at residue S49 ) may contact the head domain of its dimeric partner , and/or the rod domain of a vimentin molecule in an adjacent dimer , and/or another vimentin molecule within the same ULF . These different crosslinking geometries might produce migration differences in our GlcNDAz/SDS-PAGE assay ( Figure 1B ) . Irrespective of the molecular identities of the individual adducts , the fact that we observe a highly reproducible crosslinking pattern suggests that O-GlcNAc mediates specific contacts among vimentin molecules , and not a large variety of nonspecific interactions , which would produce variable or smeared crosslinking ( Yu et al . , 2012 ) . This evidence of specific contacts is perhaps surprising , given the widely accepted model that the head domain of vimentin is intrinsically disordered ( Guharoy et al . , 2013 ) . It is possible that , in vivo , O-GlcNAcylation mediates site-specific contacts between glycans on one vimentin molecule and discrete glycan recognition sites or domains on adjacent vimentin molecules within assembled filaments . Interestingly , several prior reports identified a GlcNAc-binding property of vimentin on the cell surface ( Ise et al . , 2010; Ise et al . , 2011; Kim et al . , 2012; Komura et al . , 2012 ) . Although the physiological relevance of extracellular vimentin is controversial , these biochemical observations are reminiscent of our own , in that vimentin is reported to bind to GlcNAc residues in both cases . However , two mutations in vimentin that abolish extracellular GlcNAc binding , E382A and E396A ( Komura et al . , 2012 ) , do not reduce intracellular GlcNDAz crosslinking ( Figure 2—figure supplement 2 ) . Therefore , our observations are distinct from the reported glycan-binding property of cell surface vimentin ( Ise et al . , 2010; Ise et al . , 2011; Kim et al . , 2012; Komura et al . , 2012 ) . The vimentin site ( s ) that bind O-GlcNAc moieties are not readily discernible from our MS studies , due to the unpredictable masses and fragmentation spectra of crosslinked adducts . However , the identification of these sites will be an important priority for future studies , because this information may elucidate the biophysical regulation of IF assembly and dynamics by O-GlcNAc . At the cellular level , specific glycosylation sites of vimentin , including S34 , S39 and especially S49 , are essential for IF morphology and for facilitating serum-stimulated migration ( Figures 3 and 4 , Figure 3—figure supplement 3 ) . We propose that O-GlcNAcylation at one or more of these sites influences the assembly and/or disassembly of vimentin filaments , as has been described for other PTMs . Indeed , phosphorylation of several residues in the vimentin head domain promotes IF disassembly in response to multiple stimuli ( Helfand et al . , 2011; Eriksson et al . , 2004; Chou et al . , 1990; Eriksson et al . , 1992; Sihag et al . , 2007; Goto et al . , 2002; Chan et al . , 2002 ) . O-GlcNAcylation and phosphorylation often compete for nearby or identical residues on specific protein substrates , giving rise to a complex functional interplay between these PTMs ( Hart et al . , 2011; Butkinaree et al . , 2010 ) . Moreover , specific O-GlcNAc sites required for in vivo IF assembly in our work , such as S34 and S39 ( Figure 3 , Figure 3—figure supplement 3 ) , can also be phosphorylated ( Helfand et al . , 2011; Eriksson et al . , 2004; Chou et al . , 1990; Eriksson et al . , 1992; Sihag et al . , 2007; Goto et al . , 2002; Chan et al . , 2002 ) . Therefore , reciprocal PTMs at certain sites may have antagonistic effects , with glycosylation of S34 or S39 promoting IF assembly , while phosphorylation at those sites induces disassembly . In contrast , vimentin S49 is required for normal IF assembly in human cells ( Figures 2 and 3 , Figure 3—figure supplement 3 ) and is a known glycosylation site ( Slawson et al . , 2008; Wang et al . , 2007 ) but is not a reported phosphorylation site . These observations suggest a functional role for O-GlcNAc , but not O-phosphate , at this residue . In support of this hypothesis , a phosphomimetic S49E mutant phenocopies an S49A mutant in IF morphology ( Figure 3 ) , and an OGT inhibitor impacts on cell migration and Chlamydia inclusion size only in the presence of WT vimentin , but not in cells expressing S49A mutant vimentin or lacking vimentin ( Figures 4 and 5E ) . Taken together , these results strongly indicate that ( de ) glycosylation of S49 is required for WT levels of cell migration and pathogen-induced IF remodeling . Although we cannot strictly rule out the possibility that S49 is phosphorylated under as-yet untested conditions , we suggest that S49 O-GlcNAcylation is required for homotypic vimentin-vimentin interactions ( Figure 2 ) , filament assembly ( Figure 3 ) , and downstream behaviors of the vimentin IF cytoskeleton in normal and pathological contexts ( Figures 4 and 5 ) . Finally , our results indicate that O-GlcNAc-mediated vimentin IF assembly is required for Chlamydia inclusion expansion and integrity , because we observed smaller inclusions and more cytoplasmic bacteria in Chlamydia-infected cells expressing Y117L or S49A mutants of vimentin ( Figure 5C , D ) . Chlamydia remodels the IF cytoskeleton to stabilize the inclusion during its replication phase , and deploys the CPAF protease to dismantle the IF cytoskeleton and promote bacterial release at late stages of infection ( Kumar and Valdivia , 2008; Jorgensen et al . , 2011; Snavely et al . , 2014; Bednar et al . , 2011 ) . Our current data extend these results by demonstrating that vimentin filament remodeling by Chlamydia depends on both OGT activity ( Figure 5A , B ) and on the site-specific glycosylation of vimentin itself ( Figure 5C–E ) . We detected WT levels of CPAF-mediated vimentin cleavage with the S49A and Y117L mutants ( Figure 5—figure supplement 2 ) , indicating that the phenotypes we observe are not due to differential vimentin cleavage during infection . The molecular mechanisms by which Chlamydia co-opts OGT activity and vimentin glycosylation sites ( e . g . , S49 ) to reshape the IF cytoskeleton remain to be determined , and will be an interesting focus of future studies . In conclusion , we have shown that site-specific glycosylation of vimentin mediates its homotypic protein-protein interactions , and is required in human cells for IF morphology and cell migration . In addition , Chlamydia relies on OGT activity and particular vimentin glycosylation sites to remodel the IF cytoskeleton and promote inclusion expansion and integrity . Our results provide new insight into the biochemical and cell biological functions of vimentin O-GlcNAcylation . In addition , our work may have broad implications for other IF proteins . Numerous IF proteins are dynamically glycosylated in vivo ( King and Hounsell , 1989; Chou et al . , 1992; Ku et al . , 2010; Dong et al . , 1996; Dong et al . , 1993; Lüdemann et al . , 2005; Deng et al . , 2008; Cheung and Hart , 2008; Slawson et al . , 2008; Wang et al . , 2007; Srikanth et al . , 2010; Kakade et al . , 2016; Tao et al . , 2006 ) , and the S49 residue of vimentin is conserved both among vertebrate vimentin orthologs and in human desmin , another type III IF protein ( Figure 6—figure supplement 1 ) . Therefore , site-specific O-GlcNAcylation may be a general mode of regulating IF dynamics and function . In the future , pharmacological modulation of O-GlcNAcylation may provide a way to manipulate filament form and function for therapeutic benefit in diseases of dysregulated IFs , ranging from neurodegeneration and cardiomyopathy to cancer .
Thiamet-G was synthesized as described ( Yuzwa et al . , 2008 ) by the Duke Small Molecule Synthesis Facility ( DSMSF ) . Ac45SGlcNAc was synthesized as described ( Gloster et al . , 2011 ) and was a gift of Benjamin Swarts , Central Michigan University . Ac3GlcNDAz-1P ( Ac-SATE ) 2 was synthesized in-house or by the DSMSF as described ( Yu et al . , 2012 ) . All other chemicals were purchased from Sigma unless otherwise noted . Lambda phosphatase was purchased from New England Biolabs . Cells were maintained in Dulbecco’s Modified Eagle’s Medium ( DMEM ) supplemented with 10% fetal bovine serum ( FBS ) , 100 units/ml penicillin and 100 µg/ml streptomycin and kept at 37°C with 5% CO2 . Cell lines were obtained from ATCC or the Duke Cell Culture Facility . In all cases , authenticity was verified using morphology , karyotyping , and PCR-based approaches ( e . g . , short tandem repeat profiling ) and tested negative for Mycoplasma ( PCR test ) by the vendor at the time of purchase . HeLa and 293T cells were selected because they are well-established model systems for the human IF cytoskeleton and Chlamydia infection . O-GlcNAc-mediated protein crosslinking was performed essentially as described ( Yu et al . , 2012 ) . Briefly , Ac3GlcNDAz-1P ( Ac-SATE ) 2 ( GlcNDAz precursor ) was added to a final concentration of 100 µM to the culture medium of 293T cells stably expressing AGX1 ( F383G ) ( Yu et al . , 2012 ) . DMSO served as the vehicle control treatment . Dishes were incubated in the dark for 24 hr , dosed again with GlcNDAz precursor , and incubated for an additional 24 hr . To crosslink and harvest , dishes were placed on ice and washed carefully twice with 10 ml of cold phosphate-buffered saline ( PBS ) twice . With dishes still on ice , 4 ml of cold PBS were added , lids were removed , and plates were exposed to 365 nm UV light for 20 min . Then , cells were resuspended in PBS by scraping and/or pipetting , pelleted by centrifugation , and lysed in either 8 M urea for IB , or IP buffer ( 150 mM NaCl , 20 mM Tris pH 7 . 4 , 1% Triton X-100 , 0 . 1% SDS ) for IPs . The concentrations of protein samples were obtained via bicinchoninic acid ( BCA ) assay according to the manufacturer’s instructions ( Thermo ) and protein concentration was normalized across all samples within each experiment . One-third the volume of 4X SDS-PAGE loading buffer was added and the sample was heated to 95°C for 3 min ( except in the case of samples prepared in 8 M urea ) . Samples were then loaded onto a polyacrylamide gel and run at 165 V . Unless otherwise noted , IBs detection was performed by enhanced chemiluminescence ( ECL ) . For ECL IBs , SDS-PAGE gels were transferred onto PVDF membrane using a BioRad semi-dry transfer system with transfer buffer ( 25 mM Tris , 192 mM glycine , 0 . 1% SDS , pH 8 , 20% methanol ) using standard methods . After transfer , membranes were blocked in 5% bovine serum albumin ( BSA ) in Tris-buffered saline with Tween ( TBST; Tris-buffered saline , 0 . 1% Tween-20 ) overnight at 4°C with gentle rocking . Primary antibody dilutions were prepared in TBST+BSA and incubated overnight with gentle shaking at 4°C . The next day , membranes were washed three times with TBST and incubated with the appropriate secondary antibody for 1 . 5 hr at room temperature . Membranes were again washed three times with TBST , developed with Advansta ECL reagent , and exposed to film . The following primary antibodies were used: Anti-O-GlcNAc RL2 ( Santa Cruz sc-59624 ) , anti-myc 9E10 ( Santa Cruz sc-40 ) , anti-tubulin ( Sigma T6074 ) , anti-nucleoporin-62 ( BD Biosciences 610498 ) , anti-vimentin ( clone D21H3 , Cell Signaling 5741 ) , anti-vimentin ( clone V9 , Sigma-Aldrich V6389 ) , anti-phospho- ( Ser/Thr ) Akt substrate ( Cell Signaling 59624 ) , anti-GFP ( ThermoFisher; A11122 ) , MOMP ( a gift from Ken Fields , University of Kentucky ) . The following horseradish peroxidase-conjugated secondary antibodies were used: Goat anti-rabbit IgG ( Southern Biotech 4030–05 ) , goat anti-mouse IgG ( Southern Biotech 1030–05 ) , goat anti-mouse κ light chain ( Southern Biotech 1050–05 ) . For quantitative fluorescent IBs , samples were separated by SDS-PAGE , electroblotted onto nitrocellulose membranes , blocked and probed as described above . The following secondary antibodies were used: Goat anti-rabbit IgG ( H + L ) Alexa Fluor 594 conjugate ( Thermo Fisher Scientific A-11012 ) , goat anti-mouse IgG ( H + L ) Alexa Fluor 594 conjugate ( Thermo Fisher Scientific A-11005 ) , goat anti-rabbit IgG ( H + L ) IRDye 800CW conjugate ( Li-Cor , 925–32211 ) , goat anti-mouse IgG ( H + L ) IRDye 800CW conjugate ( Li-Cor , 925–32210 ) . Blots were washed in TBST and scanned and analyzed on a Li-Cor Odyssey imaging system . The sequence of human vimentin isoform 1 was amplified using PCR and the following primers: Forward: 5’ cgcggatccgccaccatgtccaccaggtccgtg 3’; Reverse: 5’ cgtctagattcaaggtcatcgtgatgctga 3’ . The resulting PCR product and pcDNA3 . 1 ( + ) /myc-His A vector ( Invitrogen ) were digested with BamHI and XbaI ( New England Biolabs ) and the vector treated with calf intestinal alkaline phosphatase ( New England Biolabs ) according to the manufacturer’s instructions . PCR product and vector were mixed in a 3:1 molar ratio and incubated with T4 DNA ligase ( New England Biolabs ) in the provided buffer for 1 hr . This reaction was transformed into E . coli strain DH5α and the resulting colonies were cultured , miniprepped and sequenced using standard methods . The OGT-myc/6xHis and OGA-myc constructs have been described previously ( Boyce et al . , 2011 ) . The sequence of mEmerald-vimentin-N-18 ( Addgene plasmid #54301 ) was amplified using primers complementary to vimentin and GFP ( Forward: 5’ caccgccaccatgtccaccaggtccgtg 3’; Reverse: 5’ cgcaacgaattctcaatgtccaccaggtccgt ) . The resulting PCR product was cloned into a pENTR vector using the pENTR Directional TOPO cloning kit ( Fisher Scientific #K240020 ) according to the manufacturer’s instructions . This vector was used in a Gateway LR Clonase II Enzyme reaction , according to manufacturer’s instructions ( Invitrogen ) , with pLenti CMV Neo DEST ( 705-1 ) ( Addgene plasmid #17392 ) , or pLenti CMV/TO Hygro DEST ( Addgene plasmid # 17291 ) . Vimentin mutants were created from WT expression constructs via standard site-directed mutagenesis protocols using Agilent online primer design tools , oligonucleotides synthesized by IDT , and Phusion polymerase according to the manufacturer’s instructions ( Thermo Fisher ) . Mutagenesis was performed on vimentin-mEmerald in the pENTR vector prior to transfer to the appropriate destination vector . The integrity of all constructs was confirmed by Sanger sequencing through the entire open reading frame of each construct ( Eton Bioscience ) . 30 µl of Trans-IT-293 transfection reagent ( Mirus ) was added to 750 µl of Opti-MEM medium ( Life Technologies ) , vortexed and centrifuged briefly , and incubated at room temperature for 15 min . Then , 10 µg DNA was added to the mixture and vortexing and centrifuging were repeated , followed by an additional 15 min at room temperature . The entire reaction was added drop-wise to adherent cells . Samples were extracted in IP buffer ( 150 mM NaCl , 20 mM Tris pH 7 . 4 , 1% Triton X-100 , 0 . 1% SDS ) , diluted to 1 mg/ml total protein concentration , and 3 µg of antibody was added for each 1 mg of protein used . IPs were incubated at 4°C overnight with gentle agitation . The next day , washed protein-A/G agarose beads ( Thermo ) were added to the tubes and rotated at room temperature for 1 . 5 hr . The beads were then isolated by centrifuging at 600 g for 1 min and the supernatants were discarded . Beads were washed four times with 1 ml of IP buffer and eluted with three times the bead volume of elution buffer ( 8M urea , 150 mM NaCl; if the samples were to be further purified by nickel , this buffer also contained 10 mM imidazole ) . Differential extraction was performed essentially as described ( Ridge et al . , 2016 ) . Briefly , 293 T cells were grown to confluency , treated with GlcNDAz precursor and UV-crosslinked as above , and washed three times on the plate with PBS containing 2 mM MgCl2 at room temperature . After removing PBS , attached cells were incubated in 1 ml of low-detergent buffer ( 10 mM MOPS pH 7 , 10 mM MgCl2 , 1 mM EGTA , 0 . 15% Triton X-100 , plus 7 . 5 µl saturated phenylmethane sulfonyl fluoride ( PMSF ) solution in ethanol , added fresh , in 1x PBS ) for five minutes at room temperature with gentle agitation . The buffer was removed and cleared by benchtop centrifugation , and the resulting supernatant was reserved as the soluble cell fraction . Plates were then incubated on ice in 1 ml of ice-cold high-detergent buffer ( 10 mM MOPS pH 7 , 10 mM MgCl2 , 1% Triton X-100 , plus 7 . 5 µl saturated PMSF solution and 50 mg Benzonase , added fresh , in 1x PBS ) for 3 min . Then , 250 µl of ice-cold 5 M NaCl was added , cells were resuspended by pipetting , and samples were cleared by centrifugation . The resulting supernatant was saved as the cytoskeletal fraction , including intermediate vimentin assembly states ( Ridge et al . , 2016 ) . Finally , the pellet was resuspended by pipetting in 250 µl of 8 M urea in PBS to solubilize the remaining material ( e . g . , assembled IFs ) ( Ridge et al . , 2016 ) . Zeba spin desalting columns ( Thermo Fisher , 89890 ) were washed with 1 ml of IP buffer and centrifuged at 1000 g for 2 min three times , according to the manufacturer’s instructions . Samples were applied to the center of the column and 40 µl of IP buffer was applied after sample absorption into the resin . The column was centrifuged for 2 min at 1000 g and sample was collected in 1 . 5 ml centrifuge tubes . 293T cells stably expressing AGX1 ( F383G ) ( Yu et al . , 2012 ) were transfected with a vimentin-myc-6xHis construct , treated with GlcNDAz precursor , and UV-crosslinked as above . Crosslinked and uncrosslinked vimentin-myc-6xHis was isolated first through anti-myc IP , as above . Samples were eluted from the protein A/G beads in 8 M urea , 150 mM NaCl and 10 mM imidazole in PBS and incubated with nickel-NTA resin ( Qiagen ) rotating at room temperature for 2 hr . The resin was washed three times with the same buffer and eluted with 8 M urea , 150 mM NaCl and 250 mM imidazole in PBS for 15 min . Eluents were separated by SDS-PAGE and stained with InstantBlue gel stain ( Thermo Fisher ) . Bands corresponding to crosslinks were excised by hand , and analyzed by in-gel digest and MS/MS proteomics by the Duke Proteomics and Metabolomics Shared Resource . For more details , please see https://genome . duke . edu/cores-and-services/proteomics-and-metabolomics/protein-characterization GalNAz labeling was performed essentially as described ( Boyce et al . , 2011; Palaniappan et al . , 2013; Chen et al . , 2017 ) . Briefly , 293T cells were incubated with DMSO vehicle or 100 µM Ac4GalNAz for 24 hr . To harvest , cells were washed twice with PBS and resuspended in click lysis buffer ( 1% Triton X-100 , 1% SDS , 150 mM NaCl , 20 mM Tris pH 7 . 4 , 5 µM PUGNAc , protease inhibitor cocktail ) . After sonication and centrifugation , protein concentration was measured via BCA assay . 2 . 5 mg of protein were used in 875 µl total reaction volume for each click reaction . To these reactions were added 5 mM sodium ascorbate , 25 µM alkyne-Cy5 probe ( Palaniappan et al . , 2013 ) , 100 µM TBTA , and 1 mM CuSO4 . Samples were incubated in the dark at room temperature for 1 hr with gentle agitation and quenched with 10 mM EDTA . Samples were then diluted with 1% Triton X-100 , 150 mM NaCl , 20 mM Tris pH 7 . 4 , 1 mM EDTA , 5 µM PUGNAc and protease inhibitors to bring SDS concentration to 0 . 1% . 6 . 75 µg anti-myc ( 9E10 ) antibody were added to each sample and IPs were incubated at 4°C overnight with gentle agitation . The next day , washed protein-A/G agarose beads ( Thermo ) were added to the tubes and rotated at room temperature for 1 hr . The beads were then isolated by centrifuging at 600 g for 1 min and the supernatants were discarded . Beads were washed four times with 1 ml of IP buffer , eluted with three times the bead volume of SDS-PAGE loading buffer , and boiled for 5 min at 95°C . Samples were analyzed by SDS-PAGE and fluorescence scanning on a Li-Cor Odyssey imaging system . Three single guide RNA ( sgRNA ) sequences targeting the human vimentin locus were designed and validated via the Surveyor assay ( Ran et al . , 2013 ) by the Duke Functional Genomics facility: Vim-1: 5’ GGACGAGGACACGGACCTGG 3’; Vim-2: 5’ CATCCTGCGGTAGGAGGACG 3’; Vim-3: 5’ GGACACGGACCTGGTGGACA 3’ . An sgRNA targeting the AAVS1 ‘safe harbor’ locus ( Sadelain et al . , 2011 ) was used as a control . HeLa or 293T cells were added to a 6-well plate at ~40–60% confluency and allowed to attach for 24 hr . Then , cells were infected by adding 50 µl of a Cas9 lentivirus ( Addgene plasmid #52961 ) , produced by the Duke Functional Genomics Facility , in DMEM plus 8 µg/ml polybrene drop-wise to each well . Plates were centrifuged at 700 g for 30 min and incubated under standard conditions for 24 hr . Then , the medium was changed to DMEM without polybrene and the cells were allowed to recover for 3 days . Cells were selected and maintained with blasticidin ( 5 µg/ml for 293T , 3 µg/ml for HeLa ) for several passages before infection with sgRNA viruses . HeLa or 293T cells stably transduced with lentiviral Cas9 were added to a 6-well plate at ~40–60% confluency and allowed to attach for 24 hr . Then , cells were infected with sgRNA-expressing lentivirus produced by the Duke Functional Genomics Facility in DMEM plus 8 µg/ml polybrene , and 50 µl of virus was added drop-wise to each well . Plates were centrifuged at 700 g for 30 min and incubated under standard conditions for 24 hr . Then , the medium was changed to DMEM without polybrene and the cells were allowed to recover for 3 days . Cells were selected and maintained with puromycin ( 0 . 5 µg/ml for 293T , 1 . 5 µg/ml for HeLa ) for several passages . Single cells were sorted into 96-well plates by a DiVa fluorescence-activated cell sorter ( BD Biosciences ) at the Duke Cancer Institute Flow Cytometry Shared Resource ( DCI FCSR ) to obtain individual clones . Clones were screened for successful vimentin deletion both by IP/IB with a monoclonal vimentin antibody ( V9 , Sigma ) and quantitative RT-PCR ( qPCR ) . For qPCR , cellular mRNA was extracted with an RNeasy kit according to the manufacturer’s instructions ( Qiagen ) . RNA was used to generate cDNA using SuperScript II reverse transcriptase according to the manufacturer’s instructions ( Thermo Fisher #18064014 ) . cDNA was diluted five-fold , and triplicate reactions were performed using Sir Master Mix ( Life Tech-Power Sybr no . 4367659 ) according to manufacturer’s instructions . Reactions were performed on a StepOnePlus Real-Time PCR system ( Applied Biosystems ) . The following qPCR primers and cycling conditions were used: Forward: 5’-AGTGTGGCTGCCAAGAACCT 3’ Reverse: 5’-GAGGGACTGCACCTGTCTCC 3’ Forward: 5’-CACTCTTCCAGCCTTCCTTC 3’ Reverse: 5’-GGATGTCCACGTCACACTTC 3’ StepTemperature ( °C ) Time ( minutes ) CyclesInitial denaturation95101Denaturation Annealing Extension95 60 720 . 15 0 . 3 0 . 340Final extension72101Storage4∞- 1 . 2 × 106 293 T cells were plated at ~60–70% confluency in a 6 cm dish 24 hr prior to transfection . Cells were transfected with: 12–18 hr post-transfection , the medium was changed to 4 ml of DMEM with 30% FBS . At 48 hr post-transfection , the medium containing the virus was harvested and replaced with fresh DMEM . At 72 hr post-transfection , the virus-containing medium was again harvested , combined with the previous medium , and filtered through a 0 . 45 µm PVDF filter . Filtered supernatants were then used to infect target cells as described above . HeLa cells were selected and passaged in 500 µg/ml G418 , and 293T cells with 100 µg/ml hygromycin . After trypsinizing , cells were resuspended in serum-free DMEM ( +penicillin/streptomycin ) and passed through a sterile 30 µm filter ( Sysmex CellTrics , 04-004-2326 ) . Cells were sorted on a DiVa fluorescence-activated cell sorter ( BD Biosciences ) at the DCI FCSR to obtain the top third of highest-expressing cells , as judged by GFP fluorescence ( vimentin-mEmerald signal ) . Imaging of vimentin-mEmerald was performed using a confocal laser scanning microscope ( LSM 880; Zeiss ) equipped with an automatic stage , Airyscan detector ( Hamamatsu ) and diode ( 405 nm ) , argon ion ( 488 nm ) , double solid-state ( 561 nm ) , and helium-neon ( 633 nm ) lasers . Images were acquired using a 60x/1 . 4 NA oil objective ( Zeiss ) and deconvolved using automatic Airyscan Processing in the Zen Software ( Zeiss ) . Cells were rinsed twice with 37°C PBS and fixed with 1% formaldehyde ( Sigma ) in PBS for 10 min . Cells were permeabilized with PBS containing 0 . 1% Triton X-100 ( Sigma ) for 10 min and blocked with TBS containing 5% BSA ( Equitech-Bio ) and 0 . 1% Triton X-100 . A mouse antibody against vimentin ( clone V9 , Santa Cruz Biotech sc-57678 ) was diluted in TBS containing 5% BSA and 0 . 1% Triton X-100 and incubated with the samples overnight at 4˚C . Cells were washed three times with PBS and then incubated with a goat anti-rabbit ( H + L ) Alexa Fluor 647-conjugated secondary antibody ( Thermo Fisher ) diluted in TBST containing 5% BSA for 1 hr at room temperature . Coverslips were washed five times with PBS and mounted on slides using ProLong Diamond anti-fade mounting medium with DAPI ( Thermo Fisher ) . Cells were imaged using a confocal laser scanning microscope ( LSM 880; Zeiss ) equipped with an automatic stage , Airyscan detector ( Hamamatsu ) and diode ( 405 nm ) , argon ion ( 488 nm ) , double solid-state ( 561 nm ) , and helium-neon ( 633 nm ) lasers . Images were acquired using a 60x/1 . 4 NA oil objective ( Zeiss ) and deconvolved using automatic Airyscan Processing in the Zen Software ( Zeiss ) . Transwell migration assays were performed essentially as described ( Justus et al . , 2014 ) . Briefly , cells were plated at approximately 70% confluency and allowed to attach for 24 hr . Then , the medium was removed and replaced with DMEM containing penicillin/streptomycin but lacking FBS for 72 hr . 6 . 5 mm Transwell plates with 8 . 0 µm pore polycarbonate membrane inserts were collagen-coated by incubating individual inserts in 50 µg/ml collagen solution from bovine skin ( Sigma-Aldrich , C4243-20ML ) for 1 hr at 37°C , UV-sterilized in a biosafety cabinet , and re-hydrated with FBS-free DMEM for 1 hr . FBS-starved cells were trypsinized and counted , and 30 , 000 cells per replicate were added to each insert with either FBS-containing or FBS-free DMEM on the opposite side . Cells were permitted to migrate for 24 hr under standard culture conditions . The assay was stopped by fixing cells in ice-cold methanol for 10 min at −20°C . Then , inserts were stained with crystal violet solution ( 30% methanol , 0 . 1% crystal violet in PBS ) overnight . After staining , inserts were washed three times in PBS and the non-migrated cells were gently removed with a cotton swab . Four non-overlapping fields of view per insert were imaged with a 10x objective of a Nikon TE200 inverted microscope . Cells were counted manually using Fiji ( NIH ) . For filament morphology quantification , the number of puncta- and filament-containing cells was normalized as a percent of total cells counted and analyzed by ANOVA followed by pairwise t-tests , with p<0 . 05 considered significant . Transwell migration assays were normalized to percent of control ( i . e . , WT vimentin , vehicle-treated , serum-stimulated ) migration and analyzed by ANOVA followed by pairwise t-tests , with p<0 . 05 considered significant . Chlamydia trachomatis serotype LGV-L2 , strain 434/Bu ( CTL2 ) was propagated in Vero cells . Infectious elementary bodies ( EBs ) were derived from Vero-infected cells at 44 hr post-infection ( hpi ) . Infected cells were rinsed twice with PBS , lysed in water for 10 min , and diluted in buffer ( 7 . 2 mM K2HPO4 , 3 . 8 mM KH2PO4 , 218 mM sucrose , 4 . 9 mM L-glutamic acid , pH 7 . 4 ) . Cell lysates were subsequently sonicated and stored at −80°C . HeLa cells were seeded at a density of 5 × 104 cells/well on glass coverslips ( Bellco Glass , Inc . ) . Coverslips were pre-coated with 30 μg/ml type I collagen ( Thermo Fisher ) in 20 mM acetic acid ( Spectrum ) for 5 min and rinsed twice with medium . Cells were maintained in medium containing 0 . 5 mg/ml G418 and rinsed three times with medium lacking G418 just prior to infections . CTL2 EBs were added at a multiplicity of infection of three and infections were synchronized by centrifugation ( 1000 g , 20 min ) at 10°C . The medium was replaced and infected cells were cultured under standard conditions for 30 hr . To inhibit OGT or OGA activity , the medium of infected cells was replaced with medium containing DMSO ( vehicle ) , 50 µM 5SGlcNAc , or 50 µM Thiamet-G at 10 hpi . HeLa cells were rinsed twice with warm PBS and fixed with 3 . 7% formaldehyde in PBS for 20 min . Cells were quenched with 0 . 25% NH4Cl , permeabilized with PBS containing 0 . 1% Triton X-100 for 10 min , and blocked with PBS containing 2% BSA and 0 . 1% Triton X-100 . A mouse antibody against MOMP ( Santa Cruz , sc-57678 ) and a rabbit antibody against cap1 ( A . Subtil , Institut Pasteur ) were diluted in PBS containing 2% BSA and 0 . 1% Triton X-100 . The secondary antibodies goat-anti-mouse ( H + L ) Alexa Fluor 647 ( Thermo Fisher ) and goat-anti-rabbit ( H + L ) Alexa Fluor 555 ( Thermo Fisher ) were diluted in PBS containing 2% BSA and 0 . 1% Triton X-100 . Coverslips were washed five times with PBS and mounted on slides using Fluorsave mounting media ( CalBiochem ) . Cells were imaged using a confocal laser scanning microscope ( LSM 880; Zeiss equipped with an automatic stage , Airyscan detector ( Hamamatsu ) and diode ( 405 nm ) , argon ion ( 488 nm ) , double solid-state ( 561 nm ) , and helium-neon ( 633 nm ) lasers . Images were acquired using a 60x/1 . 4 NA oil objective ( Zeiss ) and deconvolved using automatic Airyscan Processing in the Zen Software ( Zeiss ) . To quantify Chlamydia inclusion size , images were imported into ImageJ ( NIH ) and converted to 8-bit TIFF and binary image files to demarcate individual inclusions . The area of each cap1-positive inclusion and the number of extra-inclusion bacteria were exported and plotted in the R software . Datasets were analyzed in R using Levene’s Test to assess equal variance , followed by either a Student’s t-test or Welch’s t-test , with p<0 . 05 considered significant . Vimentin−/− HeLa cells stably transduced with empty vector or WT , S49A or Y117L vimentin-mEmerald were mock-infected or infected with CTL2 Chlamydia ( MOI = 0 . 5 ) for 30 hr , washed twice with cold PBS , incubated in ice-cold buffer ( 50 mM Tris pH 7 . 4 , 150 mM NaCl , 1 mM EDTA , 1 mM PMSF , 1% Triton X-100 , and protease inhibitors ( Roche ) ) for 30 min and then sonicated on ice for 10 s . Cell lysates were cleared by centrifugation at 8000 rpm for five minutes at 4°C . Supernatants were diluted in SDS-PAGE sample buffer and heated to 95°C for five minutes . Equal volumes of sample were analyzed by IB . Rabbit antibodies against GFP ( ThermoFisher A11122 ) and MOMP ( a gift from Ken Fields ) , a mouse antibody against α-tubulin ( Sigma-Aldrich , Clone B-5-1-2 ) , and secondary anti-rabbit and anti-mouse antibodies ( Li-Cor Biosciences ) conjugated to infrared dye were diluted in PBS containing 5% nonfat milk ( weight/volume ) and 0 . 1% Tween-20 , and sequentially incubated on the membrane prior to scanning with the Odyssey imaging system ( Li-Cor Biosciences ) . We have adhered to the definition of ‘biological replicates’ outlined by Blainey et al . ( 2014 ) and paraphrased as independent , parallel measurements of biologically distinct samples to capture biological variation . By contrast , technical replicates are repeated measures of the same biological sample to determine the noise associated with experimental procedures or instruments ( Blainey et al . , 2014 ) . Individual experiments shown in the figures are representative of at least three biological replicates . Statistical standards and tests for particular experiments are detailed in the appropriate subsections of the Materials and Methods and figure legends . For more information , please see the eLife Transparent Reporting form associated with this work . | Like the body's skeleton , the cytoskeleton gives shape and structure to the inside of a cell . Yet , unlike a skeleton , the cytoskeleton is ever changing . The cytoskeleton consists of many fibers each made from chains of protein molecules . One of these proteins is called vimentin and it forms intermediate filaments in the cytoskeleton . Many different types of cells contain vimentin and a lot of it is found in cancer cells that have spread beyond their original location to other sites in the body . Cells use chemical modifications to regulate cytoskeleton proteins . For example , through a process called glycosylation , cells can reversibly attach a sugar modification called O-GlcNAc to vimentin . O-GlcNAc can be attached to several different parts of vimentin and each location may have a different effect . It is not currently clear how cells control their vimentin filaments or what role O-GlcNAc plays in this process . Using genetic engineering , Tarbet et al . produced human cells in the laboratory with modified vimentin proteins . These altered proteins lacked some of the sites for O-GlcNAc attachment . The goal was to see whether the loss of O-GlcNAc at a specific location would affect fiber formation and cell behavior . The results showed one site where vimentin needs O-GlcNAc to form fibers . Without O-GlcNAc at this site , cells could not migrate towards chemical signals . In addition , in normal human cells , Chlamydia bacteria hijack vimentin and rearrange the filaments to form a cage around themselves for protection . However , the cells lacking O-GlcNAc on vimentin were resistant to infection by Chlamydia bacteria . These findings highlight the importance of O-GlcNAc on vimentin in healthy cells and during infection . Vimentin’s contribution to cell migration may also help to explain its role in the spread of cancer . The importance of O-GlcNAc suggests it could be a new target for therapies . Yet , it also highlights the need for caution due to the delicate balance between the activity of vimentin in healthy and diseased cells . In addition , human cells produce about 70 other vimentin-like proteins and further work will examine if they are also affected by O-GlcNAc . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"cell",
"biology"
] | 2018 | Site-specific glycosylation regulates the form and function of the intermediate filament cytoskeleton |
Drosophila has helped us understand the genetic mechanisms of pattern formation . Particularly useful have been those organs in which different cell identities and polarities are displayed cell by cell in the cuticle and epidermis ( Lawrence , 1992; Bejsovec and Wieschaus , 1993; Freeman , 1997 ) . Here we use the pattern of larval denticles and muscle attachments and ask how this pattern is maintained and renewed over the larval moult cycles . During larval growth each epidermal cell increases manyfold in size but neither divides nor dies . We follow individuals from moult to moult , tracking marked cells and find that , as cells are repositioned and alter their neighbours , their identities change to compensate and the pattern is conserved . Single cells adopting a new fate may even acquire a new polarity: an identified cell that makes a forward-pointing denticle in the first larval stage may make a backward-pointing denticle in the second and third larval stages .
There are three larval stages in the development of Drosophila , L1–L3 ( Szabad et al . , 1979; Dambly-Chaudière and Ghysen , 1986; Campos-Ortega and Hartenstein , 1997 ) . Ventrally , each segment of the abdominal epidermis has a belt of thorny denticles ( Lohs-Schardin et al . , 1979; Dougan and DiNardo , 1992; Campos-Ortega and Hartenstein , 1997; Payre et al . , 1999 ) . Each belt is built by seven lines of cells ( Figure 1 ) that together produce seven imperfectly defined rows of denticles ( Price et al . , 2006; Walters et al . , 2006 ) ; denticle rows 0 and 1 are made by cells of the posterior compartment ( P ) ( Dougan and DiNardo , 1992 ) and rows 2–7 by the anterior compartment ( A ) . The denticles in rows 2 , 3 , 5 , 6 and 7 point backwards while those in rows 0 , 1 and 4 point forwards ( Lohs-Schardin et al . , 1979 ) . The cells that make denticle rows 2 and 5 in the embryo have also an additional function: they are the tendon cells that link epidermal cells to muscles ( Figure 1C–E; Hatini and DiNardo , 2001; Volohonsky et al . , 2007; reviewed in Volk ( 1999 ) ) . 10 . 7554/eLife . 01569 . 003Figure 1 . The arrangement of epithelial cells differs between embryo and larva . Anterior is to the left in all figures . ( A and B ) An embryo is shown in ( A ) and the pre-L3 ( i . e . , the new third instar epidermis developing under the second instar cuticle ) in ( B ) . The seven rows of pre-denticles are numbered and arrows indicate the ventral midline . The pre-denticles are labelled with utrp::GFP and the cell outlines with DE-cad::GFP ( both in green ) . Posterior cells and pre-denticles ( rows 1 in ( A ) and rows 0 and 1 in ( B ) ) show red because en . GAL4 is driving expression of UAS . cherry::moesin in the entire P compartment . ( C and D ) Embryo ( C ) and pre-L3 ( D ) . Pre-denticles labelled with utrp::GFP as above . The muscle-attaching tendon cells are marked by sr . GAL4 driving expression of UAS . cherry::moesin ( red ) . In the embryo ( C ) sr . GAL4 marks the pre-denticles of rows 2 and 5 , made by the two lines of tendon cells in the embryo . In the pre-L3 larva ( D ) note the actin palisades in the tendon cells that are labelled in both green and red . In the larva , no pre-denticles are made by these two lines of cells . ( E and F ) show the cuticular denticles of the L1 ( E ) and L3 ( F ) larvae . Scale bars are 10 µm . ( G ) Diagrams of the embryonic and larval ventral epithelium . The green numbers indicate rows of denticles in L1 , the red numbers in L2 and L3 . Their polarities are indicated . Note the many changes between embryo and larvae ( see also Figure 2 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01569 . 003 During each moult cycle , the larval epidermis secretes a new cuticle under the old one and when this process is completed , the old cuticle is sloughed off . Over the two larval moult cycles the epidermal cells do not change in number , but they undergo endoreplication of their DNA and grow considerably ( Edgar and Orr-Weaver , 2001 ) . Here we describe how the epidermal cells behave during the three larval stages and ask how the patterns of muscle attachments and cuticular denticles are maintained .
During embryogenesis , the actin-based pre-denticles , the precursors of the cuticular denticles , are formed temporarily at the apico-posterior limits of the cells and all point backwards ( Figure 1A; Dickinson and Thatcher , 1997 ) . However , by the L1 stage the completed denticles of rows 1 and 4 now point forwards ( Figure 1E; Lohs-Schardin et al . , 1979 ) and it is not clear when or how this change of polarity occurs . However , our observations suggest that rows 1 and 4 have broader cells and start to behave differently from the other rows shortly before stage 16 , at the beginning of cuticle formation ( Figure 2A ) . 10 . 7554/eLife . 01569 . 004Figure 2 . Convergent extension in the anteroposterior axis between pre-L1 and L2 . ( A ) Mid stage embryo with pre-denticles . The seven rows of pre-denticles ( 1–7 ) are indicated . At first all pre-denticles are found on the posterior boundaries of the appropriate rows . However , in later embryos pre-denticle rows 1 and 4 , the two rows that will make denticles pointing forwards , are now situated near the middle of the cells . This suggests that some movement of the pre-denticles may be part of polarity reversal . Also it may be relevant that cell lines I and IV of the embryo are the only lines that make extra lines of cells in the larva , and therefore contribute to convergent extension . Labelling for ( A–E ) : The pre-denticles are labelled with utrp::GFP and the cell outlines with DE-cad::GFP ( A and B ) or DE-cad::tomato ( C–E ) . ( B ) Late stage embryo , before moulting to L1 but after actin pre-denticles have gone . The pattern is similar to the earlier embryo , with two lines of cells between the tendon cells ( II and V ) . ( C ) Mid stage embryo showing the marked portion of the epidermis . The rectangle demarcates a segment in the anteroposterior axis and the region between the pair of ventral sensorial papillae ( p1 ) ( Dambly-Chaudière and Ghysen , 1986 ) in the mediolateral axis . The total number of cells and the numbers along the axes were counted , see ( F ) . ( D ) The pre-L2 stage . By this stage the cells have rearranged and extended in the anteroposterior axis: the number of cells in that axis has increased from ca 14 to 18 cells per segment . The fixed rectangle has changed shape dramatically but contains the same number of cells as in the embryo ( Table 1 ) . This change of dimensions is due to convergent extension that involves cell rearrangement as well as alterations in the shapes of cells . ( E ) The pre-L3 stage . The pattern of cells and the shape of the fixed rectangle resembles that in the pre-L2 . ( F ) Quantitation of the evidence for convergent extension: Boxplots ( Frigge et al . , 1989 ) of the number of cells enclosed by fixed pattern landmarks remains the same while these cells extend in the anteroposterior axis and narrow in the mediolateral axis . Pairwise comparisons using t tests with Bonferroni adjustment ( ns = not significant , *** = p<0 . 001 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01569 . 004 Actin pre-denticles are also found in the epidermis of the later parts of L1 and L2 as they each begin to construct the subsequent stage ( we call such larvae pre-L2 and pre-L3 ) . The pre-denticles do not all point backwards: in both pre-L2 and pre-L3 stages the pre-denticles of rows 1 and 4 point forwards ( Figure 1B ) . Row 0 denticles , which are present in only the L2 and L3 stages , also come from pre-denticles that point anteriorly . Therefore , unlike the embryonic pre-denticles , all the larval pre-denticles are an accurate harbinger of the orientation of the denticles that they will make . The larva grows substantially and moults twice with only subtle changes to the pattern ( Dambly-Chaudière and Ghysen , 1986; Hartenstein and Campos-Ortega , 1986 ) . It has been assumed that the cells become polytene as they do in Calliphora ( Pearson , 1974 ) , and that they neither divide nor die . These assumptions led to the reasonable expectations that the arrangement of the cells as well as their identities are conserved throughout the three larval stages ( Szabad et al . , 1979; Dambly-Chaudière and Ghysen , 1986; Hartenstein and Campos-Ortega , 1986; Bate and Martínez Arias , 1993; Campos-Ortega and Hartenstein , 1997 ) . However , as we now demonstrate , both these expectations are mistaken . In the embryo , the lines of epidermal cells that will produce the denticles of L1 are more tightly compacted than those cells that do not produce denticles ( Price et al . , 2006; Walters et al . , 2006 ) and include tendon cells that themselves make the pre-denticles of rows 2 and 5 ( Figures 1C and 2; Hatini and DiNardo , 2001 ) . Note that in what follows there are generalisations as accurate as we can make them but , in truth , each segment differs slightly from the next . Sometimes the lines of cells and the denticle rows are incomplete or partially duplicated , occasional cells are difficult to allocate or sit in an ambiguous position . The tendon cells are separated by the two lines of cells that will make denticle rows 3 and 4 ( Figure 1C , E ) . The embryonic P compartment is two-cells wide in the anteroposterior axis , the posterior of these two lines of cells making row 1 denticles ( Figure 1A , G; Dougan and DiNardo , 1992 ) . In the larva , the arrangement of the cells differs from the embryo in three major respects: first , unlike the tendon cells of the embryo , the tendon cells of the larva do not themselves make denticles . One row of tendon cells is located between denticle rows 1 and 2 and the other between denticle rows 4 and 5 ( Figure 1B , D , F ) . Second , in the embryo there are two lines of cells between the tendon cells , while in the larva the tendon cells are separated by three lines of cells . Third , in the embryo , the P compartment is two cells wide , but it becomes about four cells wide in the larva ( Figure 1A , B ) . These changes occur prior to the L2 stage and clearly involve a reorganisation of the cells that gives a substantial increase in length , along the anteroposterior axis ( Figure 2 ) . Nevertheless , in spite of this cell rearrangement , the cuticular pattern is very similar in all the three larval stages ( Figure 1E , F ) ; suggesting that some cells must be reallocated to different fates during the transition from the embryo to the L2 larva . We have quantified the arrangement and number of cells , studying individuals during embryogenesis and revisiting the same individuals as pre-L3 larvae . Other individuals were studied as pre-L2 larvae . The number of cells in a defined rectangular portion of the segment remained constant in all three stages at a mean of about 73 cells ( Figure 2C–F ) , confirming that the epidermal cells do not divide or die . In the embryo , the average number of lines of cells found in the anteroposterior axis of this portion was about 14 but it increased to 18 in the pre-L2 larva and remained unchanged thereafter and up to the pre-L3 stage ( Figure 2C–F ) . Also , the proportions of a fixed rectangular region of the segment changed between embryo and the pre-L3 larva . The ratios of the lengths of the anteroposterior to mediolateral axes were compared; there was a large change in the shape of this rectangle from the embryo to the L2 and L3 larval stages ( Figure 2 ) . We measured the shape changes separately in the denticulate and naked cuticle; the cells in these two regions rearranged in a similar way , each one increasing by roughly 2 cells in the anteroposterior axis ( data not shown ) . Thus , both ways of describing the rearrangement of the cells argue that , between the L1 and L2 larval stages , the ventral epithelial cells converge into the midline and extend in the anteroposterior axis . This mode of cell rearrangement is found in many systems and is known as convergent extension ( reviewed in Keller , 2002; Wallingford et al . , 2002 ) . To investigate in more detail how the cells rearrange , we induced clones soon after the time of cellularisation in the embryo and found these usually divided one to two times during late embryogenesis , giving rise to small clones of marked cells ( Table 1 ) . We then followed individual clones during the next two larval stages ( Figure 3 ) . To track the cells in the clones we needed to number the lines of cells ( in roman numerals ) to distinguish them from the rows of denticles that they contribute to ( see Figure 3 , ‘Materials and methods’ ) . For example , a marked cell might divide and label three cells that belong to line III in the embryo; all such cells would form pre-denticles of row 3 . This clone could then be revisited in the pre-L3 to see if the same three cells were still forming row 3 pre-denticles in the larva and ask whether the disposition of these cells in the larva differed from the earlier arrangement in the embryo . 10 . 7554/eLife . 01569 . 005Table 1 . Summary of all clones analysed in both embryo and L3 of the same individualDOI: http://dx . doi . org/10 . 7554/eLife . 01569 . 005Clone #12345678910ELELELELELELELELELEL−I22 −I22 −I44 −I22 −I44 −I22 −I11 −I44 −I44 −I44 −II41 I’41 I’31 I’31 I’42 I’ *41 I’31 I’41 I’41 I’42 I’ *3 I3 I2 I2 I2 I3 I2 I3 I3 I2 III44 T133 T133 T133 T122 T122 T144 T122 T144 T122 T1III11 III22 III22 III33 III11 III44 III11 III22 III11 III22 IIIIV43 IV’42 IV’31 IV’22 IV’43 IV’11 IV’42 IV’31 IV’31 IV’32 IV’1 IV2 IV2 IV1 IV2 IV2 IV2 IV1 IVV22 T222 T233 T222 T222 T233 T233 T222 T222 T222 T2VI22 VI11 VI11 VI22 VI11 VI11 VI22 VI11 VI22 VI22 VIVII55 VII22 VII22 VII11 VII11 VII22 VII44 VII11 VII11 VII22 VII+VII22 +VII11 +VII22 +VII22 +VII22 +VII44 +VII44 +VII44 +VII22 +VII55 +VIIThe row headers at the left show the line of cells in the embryo from which each clone originates . The numbers of cells in each clone are shown in arabic numerals . The locations of these same cells ( within the lines of cells ) in the larva are highlighted in bold roman numerals . Cells in lines I and IV of the embryo contribute to two separate lines of cells in the larva; within the denticulate region , apart from lines I and IV , each embryonic cell contributes to only one line of cells in the larva . Cells in line –I in the embryo never make denticles in embryo or larva , but we do not know if they contribute to two lines of cells in the larva . However , since the P compartment is made of two lines of cells in the embryo and four lines of cells in the larva , one extra row has to come from somewhere and line –I is the obvious suspect . ( * ) Only one these two cells showed denticles in the larva . 10 . 7554/eLife . 01569 . 006Figure 3 . Cells rearrange and change both identity and polarity during larval development . ( A–D ) Four individuals are shown; on the left as embryos ( A–D ) while the right column shows the same four individuals as pre-L3 larvae ( A’–D’ ) . The drawings indicate the disposition of the cells of each marked clone ( apical profiles filled in green ) in the individual embryo and larva respectively . Clones were induced with sry . FLP in the blastoderm stage , studied in the later embryo in the pre-L1 stage and then revisited in the pre-L3 . The cells of the clones are marked with cherry::moesin ( red cell membranes and pre-denticles ) , stinger::GFP and Cd8::GFP . All the cells , clone and non-clone are marked with utrp::GFP ( pre-denticles are labelled in green ) and DE-cad::tomato ( cell outlines labelled in red ) . Numbers I–VII indicate the lines of cells while numbers 1–7 mark the rows of pre-denticles in the embryo ( green digits ) and the larva ( red digits ) . The cells labelled with single black dots are the T1 tendon cells . Scale bars are 10 µm . ( A–A’ ) Clone of four cells . In the embryo , cells of the clone mark pre-denticles of row 1 . In the larva ( A’ ) , cells mark pre-denticles of both row 0 ( 1 cell ) and row 1 ( 3 cells ) . ( B–B’ ) Clone of two cells . In the pre-L1 embryo , cells in the clone mark pre-denticles of row 2 . In the pre-L3 larva ( B’ ) , the same cells are the tendon cells . No pre-denticles are marked in the larva . ( C–C’ ) Clone of two cells . In the embryo , cells in the clone mark pre-denticles of row 3 while in the larva the same two cells mark pre-denticles of row 2 . ( D–D’ ) Four marked cells in the denticle belt . In the embryo , two cells mark pre-denticles of row 4 and in the larva these same cells mark pre-denticles of row 3 . In the embryo two cells make pre-denticles of row 5 and are , presumably , tendon cells . In the larva these same cells make frank T2 tendon cells . Note the tendon cells are small with smaller nuclei , presumably of lower polytenic values than the epidermal cells . Note in Figure D’ that there is a muscle from the adjacent more anterior segment labelled with utrp::GFP and that this muscle attaches to a T1 cell , the most anterior cell of the segment—exactly as in the adult ( see Krzemień et al . , 2012 ) . See further cases of clones in Figure 4 . DOI: http://dx . doi . org/10 . 7554/eLife . 01569 . 006 The number of cells per clone varied between one and four , or rarely five , ( Table 1 and other clones studied but not included therein ) and , for each clone ( n = 121 ) , this number did not change significantly between the embryo ( n = 404 cells ) and the pre-L3 larva ( n = 410 ) ; confirming that cell divisions and cell death , if any , are rare in the larval epidermis . In our opinion this small inconsistency is due to recording error , caused by weaker staining in the embryo . The lack of any epidermal cell death was confirmed by means of a caspase marker . The Apoliner marker stains apoptotic cells in embryos of stage 15 as previously reported ( Bardet et al . , 2008 ) and we found no cell death in the epidermis of L1 or L3 larvae ( data not shown ) . Single clones never extended across the ventral midline that divides left from right , presumably because any clone begins with a single cell that can only be on one side of the presumptive mesoderm . Remarkably , in the embryonic denticle belts , the constituent cells of every one of those clones were arranged in a line , parallel to the mediolateral axis . Note also that the cells in the denticulate area are strongly elongated in the mediolateral axis and closely packed . However , clones that were located in the naked cuticular region between the denticle belts tended to be much less aligned . Also , the cells in the naked area , while still somewhat elongated in the mediolateral axis , were more isodiametric than those in the denticulate regions ( Figures 2E and 4C ) . Observations on the embryo do not appear to show any preferred orientation in the mitoses ( Campos-Ortega and Hartenstein , 1997; Pfeiffer et al . , 2000 ) . Perhaps the denticulate cells become aligned after they have been formed and , if so , it is intriguing that this movement is localised to only part of the segment . 10 . 7554/eLife . 01569 . 007Figure 4 . Some informative clones shown in embryo and pre-L3 . ( A ) A two-celled clone in lines–I and I of the embryo . Only the larval cell I exhibits pre-denticles . There is a line of cells I’ ( unlabelled ) in between lines–I and I in the larva . ( B ) A five-celled clone in line VII , it makes row 7 denticles in the embryo and row 6 denticles in the larva . ( C ) A four-celled clone that is in the naked cuticle posterior to line VII; this clone is not stretched out in the mediolateral axis , as the denticulate clones are . T1 tendon cells are labelled with black dots and T2 with red dots . DOI: http://dx . doi . org/10 . 7554/eLife . 01569 . 007 Clone tracking demonstrated that the epithelial cells shift during postembryonic development and change their neighbours . This was particularly clear in clones within lines I and IV . Look , for example at Figure 3A: in the late embryo the clone consists of 4 cells belonging to line I , they are arranged side by side along the mediolateral axis and all form row 1 denticles , pointing forwards . Later , in L3 , we see the same clone of four cells but now one cell has stepped out of line relative to the remaining three; this cell now becomes part of line I’ and makes denticles of row 0 . In every case any shift in the cells was anteriorwards: for example the cells from line IV in the embryo contributed about equally to denticle rows 3 and 4 in the L3 larva ( Figures 3D and 5 ) —observations arguing that line IV cells in the embryo are the source of the extra line of cells ( IV’ ) that is found between the two rows of muscle attachments in the larva . Line III , VI and VII do not rearrange but they do move forwards relative to the pattern ( see ‘Cells change both identity and polarity’ ) . The other lines of cells are stable: cell lines II and V are the tendon cells , sites of muscle attachments in the embryo; in the larva they continue to be muscle attachments ( Figure 3B , D ) . Taken together , as summarised in Figure 5 , the clones show that , of the cells producing denticles , many change their positions relative to the pattern as they develop from embryo to larva . Also it is clear that all cell lines do not contribute equally to convergent extension , some cell lines turn into two cell lines in the larva , others make only one . 10 . 7554/eLife . 01569 . 008Figure 5 . Summary of the embryo-to-larva transition . ( A ) Diagram showing the studied 9 lines of cells ( green roman numerals ) and rows of denticles ( green arabic numerals ) in the embryo and the rows of larval cells and denticles they produce in the larva ( numerals in red ) . Both denticle rows 0 and 1 in the larva arise from line I cells of the embryo . Also , a small percentage of these line I cells contribute to line I’ but make no denticles . Both rows 3 and 4 of the larva come from line IV cells in the embryo . II and V lines of cells are the tendon cells responsible for muscle attachments in the embryo that make tendon cell lines T1 and T2 in the larva . Cells of line VI and VII in the embryo produce denticles of row 5 and 6 respectively in the larva . The diameter of the circles indicates the proportion of the cells involved , for example 30% of all line I cells contribute to row 0 in the larva . ( B ) The changes between embryo and larva . Lines of cells both in front of and behind the denticles are shown . Cells shown in blue belong to the posterior compartment . Some cells of lines I and IV cells move anteriorly to form lines I’ and IV’ , respectively . Line VII cells produce row 6 denticles in the larva . DOI: http://dx . doi . org/10 . 7554/eLife . 01569 . 008 Clones within the naked cuticle behave differently; first , they are usually roundish or elongated in the anteroposterior axis ( not in the mediolateral axis as are the denticulate clones ) ; second , the clones do not change much in overall shape between the embryo and L3 . However , any changes of fate would be difficult to detect as we do not have useful markers for cell fate in that region of the cuticle . The measurements mentioned above argue that there is also cell rearrangement in the naked cuticle . The only useful signpost is the row of tendon cells present near the posterior end of the A compartment . We have seen that individual epithelial cells can contribute to one row of denticles in the embryo and to another in the larva . Nearly 30% of line I cells , which all form row 1 denticles in the late embryo and L1 , moved forward to contribute to line I’ and most of these produced row 0 denticles in L2 and L3 ( Figures 3A and 5; Table 1 ) . Are all the row 0 denticles made by cells that originated from line I in the embryo ? To test we looked at those line I’ larval cells that do not make denticles . It is possible that any denticle-free gaps in the larval row 0 could be due to line I’ cells in the larva , if they originated from line –I in the embryo . We therefore checked the fate of 29 line –I cells and asked if any went on to make denticles of row 0 ( or row 1 ) in L2 and L3 . None of these cells made denticles in the larva and none joined line I’ in the larva ( Figures 5 and 4A; Table 1 ) . It appears therefore that all the row 0 denticles in line I’ of the larva originate from cells that made row 1 denticles; they appear to ‘remember’ their embryonic propensity to form denticles , even though the larval denticles they make are distinct from those of row 1 . Even so , of 12 cells originating in line I ( that made row 1 denticles in the embryo and became line I’ in the larva ) , 2 failed to make denticles in the larva . So not every cell ‘remembers’ . The most striking evidence for change of identity came from line IV cells . Some individual line IV cells made row 4 denticles in the L1 stage but row 3 denticles in L2 and L3 ( Figures 3D and 5 ) . Row 4 denticles point forwards and row 3 denticles point backwards . These cases prove that an identified epidermal cell , even as it grows in size and increases its DNA content without division , can undergo a change in overt polarity during normal larval development . In an experimental situation involving ectopic expression of Dachsous , it was possible to change the polarity of epidermal cells as they developed through the larval stages ( Repiso et al . , 2010 ) , but a change of polarity as a normal part of development has neither been suspected nor found before . Regarding the other lines of cells: cells from lines III , VI and VII were found to produce distinct types of denticles at different stages of development: line III cells made only row 3 denticles in L1 but all made row 2 denticles in L3 ( Figures 3C and 5 ) . Similarly cells from lines VI and VII that made rows 6 and 7 denticles in the embryo made rows 5 and 6 denticles in the larva ( Figures 3C , 4 and 5 ) . Thus their identities change and with a particular bias; for they all make more anterior pattern elements in the larva than they did in the embryo . Cells from lines II and V make row 2 and 5 denticles in the embryo but they do not make denticles at all in the larval stages ( Figures 3B , D and 5 ) . In so far as denticle position , size and polarity are tokens of cell identity , then these observations prove that single cells change their identity during normal development . Furthermore , the results argue for the existence of supracellular systems that act continuously to maintain a stable cuticular pattern in spite of considerable rearrangement of the cells ( DiNardo et al . , 1994; Heemskerk and Dinardo , 1994; Alexandre et al . , 1999 ) . In order to maintain the same denticle pattern , the cells must compensate for changes in the behaviour of the tendon cells: we have shown that , although the embryonic tendon cells produce denticles ( Hatini and DiNardo , 2001 ) , the larval tendon cells do not . The instruments of this change are unknown . It may be relevant however that the expression of the stripe ( sr ) gene , that drives the differentiation of tendon cells , has two isoforms expressed at different times ( Volohonsky et al . , 2007 ) . The embryonic tendon cells express the B isoform of sr . Later , in the pre-L1 stage , when the muscles have already attached to the tendon cells and the pre-denticles have been formed , the tendon cells begin to express also the srA isoform ( Frommer et al . , 1996; Becker et al . , 1997; Vorbruggen and Jackle , 1997 ) . Perhaps the expression of srA could change the nature of the tendon cells ? To test we expressed srA and srB prematurely and ectopically in the early embryo in clones and found that the early expression of either of these forms failed to block the formation of pre-denticles and denticles in L1 . However , in the larva , cells in those clones over-expressing either SrA or SrB acquired the actin palisades characteristic of muscle-attached tendon cells in L3 ( Figure 1B , D ) and did not form any denticles ( Figure 6 ) . One simple interpretation is that , although sr expression allows denticle formation in the embryo , it blocks denticle formation in the larva . 10 . 7554/eLife . 01569 . 009Figure 6 . Clones expressing the stripe gene block L2 and L3 denticle formation . ( A–A’’ ) Two clones ( labelled with cherry::moesin and Cd8::GFP ) in adjacent segments in one individual make denticles in pre-L1 but no denticles in L3 . The B isoform of sr is expressed in these clones which are found in the row 3 of the fourth and the row 4 of the fifth abdominal segments . ( B–B’ ) One clone expressing the A isoform of sr in the embryo makes denticles in rows 2 to 4 of the sixth abdominal segment , but makes no denticles in the larva . The denticle rows are disturbed in this case . DOI: http://dx . doi . org/10 . 7554/eLife . 01569 . 009 Our results do not argue for complete lability in cell identities . For example , it would be unexpected , if , in normal development , a cell were to change from A to P identity . Indeed there is no case of cells producing row 2 ( A compartment ) in the embryo that make row 1 denticles ( P compartment ) in the larva , even though this would only require a small anteriorwards shift in cell fate , as happens to other lines of cells . Our results raise queries about the stability of PCP . What do we conclude if the polarity of a cell changes during development ? In Drosophila many experiments suggest that an individual cell’s polarity is not imposed by pervasive forces but depends , locally , on its interactions with its neighbours . For example , wild type cells that are adjacent to a nascent clone of cells lacking the frizzled gene form hairs of reversed polarity ( Gubb and Garcia-Bellido , 1982 ) . Now we report changes , during normal development , in the polarity of larval cells that do not divide . In both cases it is likely that polarity of a cell is driven by interactions with contacting cells . During development from embryo to larva , we have shown that cells , as a consequence of rearrangement , acquire new neighbours . Moreover , if these new neighbours were to present different amounts of Dachsous or Fat on their abutting membranes , this would lead to different arrangements of heterodimers and consequent polarity changes ( Ma et al . , 2003; Casal et al . , 2006; Repiso et al . , 2010 ) . Our kind reviewers have asked us to review how we name the denticle rows and to consider alternative nomenclatures . The naming of the rows is done objectively in both L1 and L2/3 but a comparison between the stages is open to various interpretations . Comparison is problematic because we have shown that two lines of cells that make denticles in L1 do not make denticles in L2/3 . Since the number of denticle rows is about the same in all three larval stages , it follows that the two replacement denticular rows in L2/3 must be made by cells that either did not have denticular identity in L1 , or by cells that contributed to different denticle rows in L1 . First , we name the denticle rows according to their position in the series of rows , with some input from interpretation . For example , in L2/3 , we name the extra row at the front , row 0 because of its position vis-à-vis the compartment border and because it is one row anterior to row 1—row 1 is itself the most posterior row in the P compartment in all larval stages and was named long ago and extensively studied in the developing L1 ( Dougan and DiNardo , 1992 ) . Row 2 is so named because it is the next row after row 1 . Row 2 in L1 , as we and others ( Dilks and DiNardo , 2010 ) have described , is made by the tendon cells , T1 . But we show here that , in L2/3 , the second row of denticles we now call row 2 is made by the cells lying posterior to T1 , and the clones show indubitably that the cells that make this row in L2/3 are different from those that made row 2 in L1 . Row 3 is named as the next row back from row 2 in both L1 and L2/3 and and the other rows in sequence , up to and including row 7 . Second , the rows are named because of their characteristics . In L1 , the rows differ in polarity and also in structure . In L1 , most of the denticles are similar , although variable , sizes , but row 5 has clearly larger denticles . However in L2/L3 the rows are more individually sized and can be distinguished on those grounds ( e . g . , row 3 denticles are small and row 5 denticles particularly large , see Figure 1 ) . In addition we name the rows because of their polarity; the denticles of rows 0 , 1 and 4 point forward in all larval stages . The problems arise when one compares row numbers in L1 with those in L2/3: for example one reviewer questioned our nomenclature and proposed that maybe row 2 of L1 is not made in L2/3 , so what we are calling row 2 in L2/3 is more homologous to row 3 ( of L1 ) than row 2 ( of L1 ) . But if we were to rename our row 2 ‘row 3’ in L2/3 ( as he/she suggests ) then we would have to rename the more posterior rows and if we were to do that then ‘row 5’ in L2/3 would point forwards ( as row 4 does in all larval stages in our nomenclature ) and ‘row 6’ in L2/3 would have the largest denticles ( like row 5 in all larval stages in our nomenclature ) . These inconsistencies argue against using the reviewer’s alternative nomenclature; in our view all the characteristics of the different denticle rows should be used as objectively as possible to help name them . The functional outcome of the developmental processes we describe is the segmented pattern of denticle bands . This pattern appears to be important for crawling ( Dixit et al . , 2008 ) and therefore , presumably , for function . If so , it needs to be conserved not only during the evolution of species ( McGregor et al . , 2007 ) but also in all the three larval stages of each species . However , conflicting with this need , is the fundamental and atavistic process of convergent extension that , necessarily , causes cell rearrangement . This conflict is resolved by the use of supracellular mechanisms , mechanisms that can build a fixed pattern in a shifting and labile cell population ( DiNardo et al . , 1994; Heemskerk and Dinardo , 1994; Alexandre et al . , 1999 ) . To achieve the same final pattern in spite of cell rearrangement it becomes necessary to drive , as we have found , changes in individual cell identity . We believe that the Drosophila embryo is exceptional because , throughout the growth of these Diptera , the epidermal cells are neither dividing nor dying . This makes our finding that individual cells become reallocated to different fates and polarities particularly eloquent . The ability of developmental processes to conserve pattern in a changing cellular constitution is emphasised when related species are compared . In classic work with the development of the vulva in nematode species , ( Eizinger and Sommer , 1997; reviewed in Sommer ( 1997 ) ) showed how two similar species build a common structure by contrasting molecular and cellular mechanisms . Illustrating again that the pressure of selection acts directly on the outcome but only indirectly on the mechanisms of development .
Flies were reared at 25°C in standard food . The Flybase ( Marygold et al . , 2013 ) entries of the relevant constructs used in this work are the following: en . GAL4: Scer/GAL4en-e16E; sr . GAL4: srmd710; DE-cad::GFP: shgUbi-p63E . T:Avic\GFP-rs; DE-cad::tomato: shgKI . T:Disc\RFP-tdTomato; sqh . utrp::GFP: Hsap\UTRNsqh . T:Avic\GFP-EGFP; UAS . cherry::moesin: MoeScer\UAS . P\T . T:Disc\RFP-mCherry; UAS . stinger::GFP: Avic\GFPStinger . Scer\UAS . T:nls-tra; UAS . srA: srA . Scer\UAS; UAS . srB: srb . Scer\UAS . tub>stop>GAL4: P{GAL4-αTub84B ( FRT . CD2 ) . P}; UAS . Cd8::GFP: Mmus\Cd8aScer\UAS . T:Avic\GFP . UAS . Apoliner: P{UAS-Apoliner}5 . Act . GAL4: P{Act5C-GAL4}17bFO1 . ( Figure 1 ) w; DE-cad::tomato/en . GAL4; UAS . cherry::moesin/utrp::GFP . ( Figure 1 ) w; DE-cad::tomato , utrp::GFP; sr . GAL4/UAS . cherry::moesin . ( Figure 2 ) w; DE-cad::tomato utrp::GFP/CyO . ( Figure 2 ) w; DE-cad::GFP/utrp::GFP . ( Figure 3 , Figure 4 ) w; tub>stop>GAL4 UAS . GFP utrp::GFP/DE-cadherin::tomato UAS . stinger::GFP;UAS . cherry::moesin/sry . FLP . ( Figure 6 ) w; tub>stop>GAL4 UAS . cd8::GFP utrp::GFP/UAS . SrA ( or UAS . SrB ) ; UAS . cherry::moesin/sry . flp . A transgene expressing FLP specifically at the blastoderm stage was created by placing the sry-alpha promoter ( −331 to +130 ) ( Schweisguth et al . , 1989 ) upstream of a cDNA encoding FLP ( Golic and Lindquist , 1989 ) . Transgenic flies were obtained by P-element transformation . Several such transgenes were recombined with tub>stop>GAL4 and P{UAS-GFPS65T . nls} ( Neufeld et al . , 1998 ) and a combination that gave a useful frequency of clones was selected for further work . The sry-alpha promoter is only expressed for about 15 min during cellularisation ( Schweisguth et al . , 1990 ) . Clones are therefore expected to be induced before or after the first blastoderm division , which follows cellularisation . Since most clones comprised 4 cells and since epidermal cells divide three times on average ( Vincent and O’Farrell , 1992; Foe and Alberts , 1983 ) , it appears that clones were mostly induced soon after the first blastoderm division . Stage 15 embryos with clones of 1–4 cells in the abdominal segments A2–A6 were mounted in a drop of Voltalef 10S oil on a microscope slide and imaged using a Leica SP5 confocal microscope . The embryos were subsequently removed , kept at 25°C on an agar plate with fresh yeast paste and reexamined 48–60 hr later ( i . e . , in pre-L3 or early L3 stages ) . To identify the lines of cells labelled , we used a combination of features: the number of rows of pre-denticles , the localisation of the tendon cells and the distinct types of pre-denticles specific for each row . For example , some III cells that did not completely align with each other were still scored as III cells based on their position relative to the tendon cells and also on the size and number of their pre-denticles—usually in the larva , III cells make about four big pre-denticles , whereas cells of line IV tend to form more denticles per cell , but of smaller size . A quadrilateral abcd was drawn using as reference the ventral papillae a’b’c’d’ whose sides ac and bd line up with the P/A boundary between the third and fourth and the fourth and fifth abdominal segments respectively ( Figure 4B ) . Around 10 straight lines were then drawn between and perpendicular to the lines ac and bd . The number of cells intersected by these lines was counted and the average of the ten lines used to estimate the number of cells in the A/P axis ( Figure 4F ) . We also measured separately the number of cells in the denticular region ( rows 1 to 7 in the embryo and 0 to 6 in the larva ) as well as the cells forming the naked cuticle between two denticle belts ( comprising the posterior half of the A compartment plus the anterior half of the P compartment in the same segment ) . For the total number of cells we counted the cells that were , entirely or partially , inside the quadrilateral . For statistical analysis we used the R programming language and software environment ( R Core Team , 2013 ) . | Fly larvae grow in an unusual way . Most embryos grow by increasing the number of cells in the embryo , but fly larvae grow by increasing the size of their cells . The epidermal cells in the growing larvae secrete a hard skin or cuticle that is renewed three times as they grow . This cuticle is decorated with teeth called denticles that the larvae use to grip surfaces as they crawl on them . The denticles are arranged in six rows during all three larval stages . It has long been assumed that if a cell in the first larval stage makes the denticles belonging to a given row , then the same cell will make denticles in the same row in the second and third larval stages . Now Saavedra et al . report that this assumption is mistaken and that the epidermal cells rearrange extensively between the first and second larval stages , and that cells acquire different identities to keep the pattern constant . Saavedra et al . marked small groups of cells in the embryo and plotted the positions of these cells as the larvae progressed through the three stages of development . These measurements showed that as the larvae grow , the cells changed their positions relative to each other . This meant that , in order to keep essentially the same pattern of denticle rows , the cells had to change their identity . Some of these changes were quite dramatic . Consider , for example , the embryonic cells that make the denticles in the second of the six rows during the first larval stage . In the embryo , these cells are tendons and attach to the muscles needed for crawling . Saavedra et al . found that these cells remain attached to the same muscles throughout growth , but that they do not make denticles during the second and third larval stages . Instead , the denticles in the second row of later larval stages are made by other cells , and these new second row cells are not attached to any muscles . In another example of these changes , some cells make denticles that point away from the head during the first larval stage , and then make denticles that point towards the head during later stages . Thus , cells can change both their identity ( e . g . , whether they are attached to muscles or not ) and their orientation ( also known as the cell polarity ) during the development of a larva . The work of Saavedra et al . illustrates how organisms adapt developmental mechanisms that have been stabilised for millions of years and for this reason limit the kinds of morphological changes that are possible . | [
"Abstract",
"Introduction",
"Results",
"and",
"discussion",
"Materials",
"and",
"methods"
] | [
"developmental",
"biology",
"cell",
"biology"
] | 2014 | Plasticity of both planar cell polarity and cell identity during the development of Drosophila |
RET can be activated in cis or trans by its co-receptors and ligands in vitro , but the physiological roles of trans signaling are unclear . Rapidly adapting ( RA ) mechanoreceptors in dorsal root ganglia ( DRGs ) express Ret and the co-receptor Gfrα2 and depend on Ret for survival and central projection growth . Here , we show that Ret and Gfrα2 null mice display comparable early central projection deficits , but Gfrα2 null RA mechanoreceptors recover later . Loss of Gfrα1 , the co-receptor implicated in activating RET in trans , causes no significant central projection or cell survival deficit , but Gfrα1;Gfrα2 double nulls phenocopy Ret nulls . Finally , we demonstrate that GFRα1 produced by neighboring DRG neurons activates RET in RA mechanoreceptors . Taken together , our results suggest that trans and cis RET signaling could function in the same developmental process and that the availability of both forms of activation likely enhances but not diversifies outcomes of RET signaling .
The neurotrophic receptor tyrosine kinase RET plays critical roles in many biological processes , including kidney genesis , spermatogenesis , and development of enteric , sensory , autonomic , and motor neurons ( Runeberg-Roos and Saarma , 2007; Ibáñez , 2013 ) . Loss of RET signaling leads to Hirschprung's disease , while RET gain of function has been implicated in various human carcinomas ( Runeberg-Roos and Saarma , 2007; Santoro and Carlomagno , 2013 ) . In addition , activation of the RET signaling pathway has potential applications in the treatment of Parkinson's disease and promotion of spinal cord ( SC ) regeneration following injury ( Bespalov and Saarma , 2007; Deng et al . , 2013 ) . Therefore , it is critical to thoroughly understand RET signaling mechanisms . RET is the common signaling receptor for the glial cell line-derived neurotrophic factor ( GDNF ) family of ligands ( GFLs ) , which includes GDNF , neurturin ( NRTN ) , artemin , and persephin . For RET activation and signaling , GFLs first bind to a GPI-linked GDNF family receptor alpha ( GFRa ) , which then associates with RET to form an active signaling complex ( Airaksinen and Saarma , 2002 ) . In vertebrates , the GFRas and their high-affinity ligand pairs are GFRa1 and GDNF ( Jing et al . , 1996; Treanor et al . , 1996 ) , GFRa2 and NRTN ( Baloh et al . , 1997; Buj-Bello et al . , 1997; Klein et al . , 1997 ) , GFRa3 and artemin ( Baloh et al . , 1998 ) , and GFRa4 and persephin ( Yang et al . , 2007 ) . RET can be activated by GFRas expressed in the same cell ( cis signaling ) or by GFRas ( mainly GFRa1 ) produced from other sources ( trans signaling ) in vitro ( Paratcha et al . , 2001; Ledda et al . , 2002 ) . The existence of both cis and trans activation has been proposed to diversify RET signaling by either recruiting different downstream effectors or changing the kinetics or efficacy of kinase activation ( Tansey et al . , 2000; Paratcha et al . , 2001 ) . Consistent with the trans signaling model , Gfra1 is expressed in the target fields of many RET+ neurons during development and can promote axon growth upon GDNF treatment in culture ( Trupp et al . , 1997; Yu et al . , 1998; Paratcha et al . , 2001 ) . However , the ‘cis-only’ mouse model , in which Gfra1 is expressed under the control of the Ret promoter in a Gfra1 null background , produced no overt phenotypes in many Ret-dependent developmental processes , suggesting that trans signaling may not play a major physiological role ( Enomoto et al . , 2004 ) . Recently , trans RET signaling has been implicated in the development of inhibitory cortical interneurons , nigral dopaminergic neurons , and enteric lymphoids , and in perineural invasion by cancer cells ( Canty et al . , 2009; Kholodilov et al . , 2011; Patel et al . , 2012; He et al . , 2014 ) . Nevertheless , the physiological functions of trans RET signaling and whether cis and trans signaling lead to the same or different biological outcomes in vivo remain largely unresolved . Aβ mechanoreceptors are large diameter somatosensory neurons mediating discriminative touch , which innervate layers III–V of the SC . They can be broadly divided into rapidly adapting ( RA ) and slowly adapting ( SA ) mechanoreceptors based on their adaptation properties to sustained mechanical stimuli ( Fleming and Luo , 2013 ) . Previously , we and other labs identified that a small population of mouse DRG ( dorsal root ganglion ) neurons , the early RET+ DRG neurons , develop into RA mechanoreceptors , and that Ret is required cell autonomously for the growth of their third order central projections innervating the dorsal SC ( dSC ) ( Bourane et al . , 2009; Luo et al . , 2009; Honma et al . , 2010 ) . RET in RA mechanoreceptors encounters environments in which both cis and trans activation are possible , providing a good model system to study the physiological functions of trans RET signaling . RA mechanoreceptors express Ret and Gfra2 ( Bourane et al . , 2009; Luo et al . , 2009; Honma et al . , 2010 ) , whereas Gfra1 is highly expressed in their target field ( Treanor et al . , 1996; Yu et al . , 1998 ) and by neighboring DRG neurons during development ( Luo et al . , 2009; Honma et al . , 2010 ) . Here , we found that the central projection deficit of RA mechanoreceptors is negligible in postnatal Gfra2 and Nrtn mutant mice , which is in great contrast to the severely affected Ret mutant mice . By genetically tracing RA mechanoreceptors in different mutant mouse lines during development , we showed that the initial growth of the third order central projections of RA mechanoreceptors depends on the cis activation of RET via GFRa2 and NRTN . However , central projections of Gfra2 null RA mechanoreceptors gradually recover during development . Gfra1 null mice show no obvious central projection deficit by itself , but Gfra1;Gfra2 double null mice have similar cell death and central projection deficits to those of Ret null mice . Moreover , we showed that Gfra1 is non-detectable in most RA mechanoreceptors , thus RET in RA mechanoreceptors is most likely activated by GFRa1 in trans . Finally , we determined that RET in Gfra2 null RA mechanoreceptors responds to GDNF in DRG explant culture , and this responsiveness is mediated by GFRa1 from neighboring DRG neurons ( trans activation ) . Taken together , our results indicate that combinatorial cis and trans RET signaling promote survival and central projection growth of RA mechanoreceptors in vivo .
Since RET can be activated by GFLs/GFRas either in cis or in trans ( mainly by GDNF/GFRa1 ) in vitro , we asked if the expression patterns of Gfra1 , Gfra2 , Gdnf , and Nrtn in the developing SC and DRGs would provide insight into RET signaling in RA mechanoreceptors in vivo . We performed in situ hybridization for Ret , Gfra1 , Gfra2 , Gdnf , and Nrtn on embryonic day 13 . 5 ( E13 . 5 ) and E15 . 5 wild-type DRG and SC sections . Double in situ hybridizations that characterize the expression of Gfra1 and Gfra2 in different populations of DRG neurons have been previously conducted ( summarized in Figure 1—figure supplement 1K [Luo et al . , 2009] ) . Similar to previous characterization ( Molliver et al . , 1997; Luo et al . , 2007 , 2009 ) , Ret is expressed in motor neurons and a mix of small and large diameter DRG neurons at E13 . 5 and E15 . 5 ( Figure 1—figure supplement 1A–B ) . Most large diameter RET+ DRG neurons at these stages are the early RET+ DRG neurons , which develop into RA mechanoreceptors ( Bourane et al . , 2009; Luo et al . , 2009 ) . Gfra1 is highly expressed in some DRG neurons and motor neurons as well , but these GFRa1+ DRG neurons come from NTRK1+ precursors and are not early RET+ RA mechanoreceptors ( Yu et al . , 1998; Luo et al . , 2009; Honma et al . , 2010 ) . In addition , Gfra1 is highly expressed in the dorsal root entry zone and the dSC , which are the target fields of the central projections of RA mechanoreceptors ( Figure 1—figure supplement 1C–D ) . Gfra2 is expressed in a small number of large diameter DRG neurons , which were previously shown to be RA mechanoreceptors ( Bourane et al . , 2009; Luo et al . , 2009 ) , and some SC cells and motor neurons at these stages ( Figure 1—figure supplement 1E–F; Oppenheim et al . , 2000 ) . Nrtn is diffusely expressed at a low level in the SC and DRGs at both E13 . 5 and E15 . 5; Gdnf transcript is barely detected at E13 . 5 but is clearly expressed in DRG and motor neurons at E15 . 5 ( Figure 1—figure supplement 1G–J ) . Thus , based on the expression patterns of RET signaling components in the developing SC and DRGs , RET in the central projections and cell bodies of developing RA mechanoreceptors could potentially be activated in cis by NRTN/ GFRa2 or in trans by GDNF/ GFRa1 , which may come from neighboring DRG neurons , dorsal root entry zone cells , or dSC cells . RA mechanoreceptors depend on RET for the growth of their third order central projections innervating layers III–V of SC . In postnatal Ret mutant mice , VGLUT1+ puncta , which label pre-synaptic terminals of mechanoreceptors and proprioceptors ( Hughes et al . , 2004; Paixão et al . , 2013 ) , are greatly reduced in layers III–V , indicating deficits in the third order central projections of RA mechanoreceptors ( Luo et al . , 2009 ) . Since RA mechanoreceptors express a high level of Gfra2 but not any other Gfras ( Luo et al . , 2009 ) , it is likely that RET in RA mechanoreceptors is activated by NRTN/GFRa2 in cis . Indeed , we previously found that Pacinian corpuscles , a subtype of RA mechanosensory end organs in the periphery , are not formed in Ret , Gfra2 , or Nrtn mutant mice , supporting that NRTN/GFRa2-RET cis signaling occurs in RA mechanoreceptors ( Luo et al . , 2009 ) . Here , we asked if NRTN-GFRa2/RET cis signaling is required for the growth of RA mechanosensory central projections as well . We performed immunostaining of VGLUT1 with postnatal day 7 ( P7 ) Gfra2GFP/GFP null and Nrtn−/− null SC sections . No significant decrease of VGLUT1+ puncta in layers III–V of SC is observed in Gfra2 and Nrtn null mice ( Figure 1A–C , Figure 1—source data 1 [p = 0 . 96] , and data not shown ) . This result suggests that unlike RET signaling in the peripheral branches of RA mechanoreceptors , cis activation of RET by GFRa2 and NRTN may be dispensable for the normal development of central projections of RA mechanoreceptors . 10 . 7554/eLife . 06828 . 003Figure 1 . P7 Gfra2 mutant mice show normal dorsal spinal cord ( dSC ) VGLUT1 staining and Gfra1 null mice display normal rapidly adapting ( RA ) mechanoreceptor central projections at E13 . 5 . ( A–B ) Anti-VGLUT1 immunostaining of P7 SC sections from Gfra2GFP/+ control ( A ) and Gfra2GFP/GFP null ( B ) mice . VGLUT1 staining labels presynaptic terminals of mechanosensory neurons , which are found in layers III–V of the dSC ( outlined in white ) . Note that green fluorescent protein ( GFP ) driven from the Gfra2 locus cannot be visualized directly . Therefore , positive signal indicates presynaptic VGLUT1+ puncta and not GFRa2+ primary afferent axons . ( C ) Quantification of VGLUT1+ puncta in dSC , which is displayed as a percentage of VGLUT1+ pixels compared to the control pixel count . The similar density of VGLUT1+ puncta between mutant and control tissue suggests that cis RET signaling via GFRa2 is dispensable for the growth of RA mechanosensory central projections at P7 . ( D ) Quantification of GFP+;NF200+ neurons , which indicate RA mechanoreceptors , per DRG section . The non-significant decrease in RA mechanoreceptor number per section in Gfra2 nulls suggests that most RA mechanoreceptors are not dependent on cis RET signaling for survival . ( E–F ) Anti-GFP immunostaining of RA mechanoreceptor central projections in E13 . 5 Gfra1+/−;RetCFP/+ control ( E ) and Gfra1−/−;RetCFP/+ mutant ( F ) SC sections . The increased CFP signal in Gfra1 null dSC is likely due to the precocious expression of Ret in some dSC neurons of Gfra1 mutants . ( G ) Quantification of CFP+ pixel number in dSC . The lack of a reduction in CFP+ axons in Gfra1 mutant dSC indicates that trans signaling via GFRa1 is not required for the initial growth of RA mechanosensory third order central projections . ( H ) Quantification of number of CFP+ neurons per DRG section indicates no loss of RA mechanoreceptors in Gfra1 mutants at E13 . 5 . C: cervical level , T: thoracic level , L: lumbar level . Scale bars = 50 μm . Error bars represent SEM . n . s . = p > 0 . 05 , * = p < 0 . 05 . Source data are provided in Figure 1—source data 1 , 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 06828 . 00310 . 7554/eLife . 06828 . 004Figure 1—source data 1 . VGLUT1 dSC staining and RA mechanoreceptor number in P7 Gfra2 mutants . DOI: http://dx . doi . org/10 . 7554/eLife . 06828 . 00410 . 7554/eLife . 06828 . 005Figure 1—source data 2 . RA mechanoreceptor central projections and cell number in E13 . 5 Ret , Gfra1 , Gfra2 , and Nrtn mutants . DOI: http://dx . doi . org/10 . 7554/eLife . 06828 . 00510 . 7554/eLife . 06828 . 006Figure 1—figure supplement 1 . Expression of Ret , Gfras , and Gfls in developing spinal cord ( SC ) and DRG . ( A–J ) In situ hybridization of mouse SC and DRG at E13 . 5 and E15 . 5 for Ret ( A–B ) , Gfra1 ( C–D ) , Gfra2 ( E–F ) , Gdnf ( G–H ) , and Nrtn ( I–J ) . Ret is expressed in DRG neurons and motor neurons at E13 . 5 and E15 . 5 . Ret is also expressed in dSC from E15 . 5 . Gfra1 is expressed in DRG neurons , motor neurons , dorsal root entry zone , and dSC at both stages . Note that expression of Gfra1 in the dorsal root entry zone and dSC is largely Ret independent . Gfra2 is expressed in large diameter DRG neurons and in motor neurons . Nrtn and Gdnf are barely detected in DRG and SC at E13 . 5 and display increased expression in DRGs at E15 . 5 . ( K ) Schematic of temporal expression of Ret and Gfra co-receptors in DRG neurons , which is adapted from previous studies ( Luo et al . , 2007 , 2009; Molliver et al . , 1997 ) . RA mechanoreceptors ( red cells ) are early RET+ DRG neurons , which begin to express Ret and Gfra2 from E10 . 5 or earlier . All other RET+ DRG neurons develop from NTRK1+ precursors and depend on NTRK1 signaling for their expression of Ret and Gfras . Intermediate RET+ neurons ( blue cells ) express Ret and Gfra1 from E13 . 5 . The late RET+ non-peptidergic nociceptors express Ret from E15 . 5 , and begin to express a low level of Gfra2 around P0 . Scale bar = 100 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 06828 . 00610 . 7554/eLife . 06828 . 007Figure 1—figure supplement 2 . Ret is required for the growth of RA mechanosensory third order central projections at E13 . 5 . ( A ) Schematic of development of RA mechanosensory central projections . RA mechanoreceptors grow central and peripheral axons soon after neurogenesis , generating first order branches ( red color ) . Upon reaching the dSC , the central axons bifurcate and send second order longitudinal branches rostrally and caudally ( blue color ) . Around E13 . 5 , third order interstitial projections ( green color ) from the longitudinal branches innervate layers III–V of the dSC and develop complex branching patterns . Synaptic connections between mechanoreceptors and dSC neurons ( light blue dots ) develop from E18 . 5 . ( B–G ) Anti-NF200 and anti-GFP immunostaining of E13 . 5 RetCFP/+ ( B–D ) and RetCFP/CFP ( E–G ) SC . The dSC innervations of RA mechanosensory fibers are outlined by white dotted line . ( H ) Quantification of CFP+ pixel number in the dSC , which is displayed as a percentage of pixel number relative to control . There is a significant decrease in CFP+ axons innervating the SC in Ret mutants , suggesting that the initial growth of RA mechanosensory third order projections depends on RET signaling . ( I ) Quantification of the number of CFP+ neurons per DRG section . There is no significant change in the number of CFP+ neurons per DRG section , suggesting that there is no cell death or downregualtion of CFP expression in E13 . 5 Ret null RA mechanoreceptors . Scale bar = 50 μm . C = Cervical , T = Thoracic , L = Lumbar . Error bars represent SEM . n . s . = p > 0 . 05 , *** = p < 0 . 001 . Source data are provided in Figure 1—source data 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 06828 . 00710 . 7554/eLife . 06828 . 008Figure 1—figure supplement 3 . Generation of Gfra1 conditional and null alleles . ( A ) Schematic of generation of Gfra1 conditional and null alleles . See supplemental ‘Materials and methods’ for additional details . ( B ) Predicted peptide sequence of the truncated GFRa1 protein after the excision of exon 6 . Black letters represent amino acids which share identity with wild-type protein sequence . The loss of exon 6 causes a frame shift , leading to the inclusion of amino acids which do not match the wild-type sequence ( blue letters ) . The frame shift also introduces a premature stop codon following amino acid 179 . ( C–D ) In situ hybridization of Gfra1 in P0 Gfra1+/+ control ( C ) and Gfra1−/− null ( D ) DRG sections shows a complete loss of Gfra1 transcript in Gfra1 null tissue . DOI: http://dx . doi . org/10 . 7554/eLife . 06828 . 008 To determine whether RA mechanoreceptors survive without Gfra2 , we quantified the number of GFP+;NF200+ neurons per DRG section in P7 Gfra2GFP/+ controls and Gfra2GFP/GFP nulls . Green fluorescent protein ( GFP ) is expressed from the Gfra2 locus and most of GFP+;NF200+ neurons indicate RA mechanoreceptors in Gfra2GFP mice . In agreement with our previous findings at P0 ( Luo et al . , 2009 ) , we found a slight but non-significant decrease in RA mechanoreceptor number between controls and mutants ( Figure 1D , Figure 1—source data 1 [p = 0 . 34] ) . Therefore , cis RET signaling via GFRa2 does not seem to be critical for the early postnatal survival of RA mechanoreceptors . To understand the mechanism of RET signaling that controls growth of RA mechanosensory central projections , we genetically traced RA mechanoreceptors in Ret , Gfra1 , Gfra2 , and Nrtn mutant mice at different developmental stages . We first used Ret mutant mice , which serve as a positive control for the central projection deficit , to determine a robust method for visualizing RA mechanosensory interstitial branches at E13 . 5 . We compared two methods that have been previously used . One is immunostaining of neurofilament-200 ( NF200 ) , which is expressed by large diameter DRG neurons , including RA mechanoreceptors , SA mechanoreceptors , and proprioceptors ( Bourane et al . , 2009 ) . The other is to use a knockin/null allele of Ret ( Honma et al . , 2010 ) , RetCFP ( Uesaka et al . , 2008 ) , in which cyan fluorescent protein ( CFP , a variant of GFP ) is expressed from the Ret locus . Although Ret is expressed in both RA mechanoreceptors and some other DRG neurons at E13 . 5 ( Luo et al . , 2009 ) , central projections of non-RA mechanoreceptor RET+ neurons , most of which develop into nociceptors , do not innervate the dSC until E15 . 5 or later ( Ozaki and Snider , 1997 ) . In addition , the expression of Ret in dSC neurons is not obvious until E15 . 5 ( Figure 1—figure supplement 1B ) . Thus , the RetCFP allele may allow us to specifically visualize central projections of RA mechanoreceptors at E13 . 5 . To compare these two methods , we performed anti-NF200 and anti-GFP immunostaining on SC sections of E13 . 5 RetCFP/+ control and RetCFP/CFP null embryos . We observed a decrease in the density of NF200+ fibers in the dorsal horn ( Figure 1—figure supplement 2B , E ) . This decrease of NF200+ central projections , however , is not dramatic . This is because the NF200 antibody also recognizes central projections of SA mechanoreceptors and proprioceptors , which develop in a manner temporally comparable to RA mechanoreceptors . In contrast , CFP+ fibers innervating the dSC display a dramatic reduction in Ret null mice ( Figure 1—figure supplement 2C , F ) . Ret null CFP+ fibers reach the dorsal surface of the SC but rarely grow interstitial branches innervating layers III–V . We quantified the number of CFP+ pixels in the dorsal horn ( displayed as percentage of CFP+ pixels normalized to the control ) as a proxy for the extent of axon growth and found a significant decrease in CFP+ fibers in Ret mutant dorsal horn ( Figure 1—figure supplement 2H and Figure 1—source data 2 , [p < 0 . 001] ) . This result suggests that the RetCFP allele is a valid tool for visualizing central projection deficits of RA mechanoreceptors at E13 . 5 . Since Ret signaling can positively regulate the expression of its own signaling components or control neuronal survival ( Luo et al . , 2007; Baudet et al . , 2008; Golden et al . , 2010 ) , it is conceivable that the lack of dSC CFP+ fibers could be due to a downregulation of CFP expressed in Ret null RA mechanoreceptors or death of RA mechanoreceptors . To exclude these possibilities , we quantified the number of CFP+ neurons in DRGs . We found that the number of CFP+ neurons per DRG section was not statistically different between Ret heterozygotes and null mice ( Figure 1—figure supplement 2I and Figure 1—source data 2 ) . In addition , the intensity of GFP+ fibers at the dorsal surface of the SC is comparable between Ret mutant and control mice . Therefore , the loss of CFP+ fibers in the dorsal horn of E13 . 5 Ret mutants must mainly be due to a deficit in growth of interstitial central axons , but not due to the down-regulation of CFP expression or the death of RA mechanoreceptors . The finding that dSC VGLUT1 staining is largely normal in postnatal Gfra2 and Nrtn null mice suggests that cis RET signaling may be dispensable for RA mechanosensory central projections . To determine if the development of RA mechanosensory central projections depends on the trans activation of RET via GFRa1 and GDNF , we generated Gfra1 null ( Gfra1− ) mice ( Figure 1—figure supplement 3A–B and ‘Materials and methods’ ) . In situ hybridization of Gfra1 control and null DRG sections showed that Gfra1 transcripts are not produced in mice homozygous for this mutant allele ( Figure 1—figure supplement 3C–D ) . In addition , kidneys are not formed in these Gfra1 null mice ( data not shown ) , a phenotype consistent with previously reported Gfra1 null mice ( Cacalano et al . , 1998; Enomoto et al . , 1998 ) . Thus , the Gfra1− allele we generated is a null allele . If trans activation of RET via GFRa1 is required for the growth of interstitial central projections of RA mechanoreceptors , we expect to see a decrease of central projections of RA mechanoreceptors in the dSC of Gfra1 null mice . To test this idea , we generated E13 . 5 Gfra1+/−;RetCFP/+ control and Gfra1−/−;RetCFP/+ mutant embryos to examine RA mechanosensory central projections at this stage ( Figure 1E–F ) . We found that innervation of dSC by CFP+ fibers was not reduced upon Gfra1 ablation ( Figure 1G , Figure 1—source data 2 ) . Additionally , the lack of Gfra1 function did not lead to a decrease of CFP+ DRG neurons ( Figure 1H , Figure 1—source data 2 ) . Together , our results suggest that trans activation of RET via GFRa1 is not required for the survival or central projection growth of RA mechanosensory neurons at E13 . 5 . Since no deficit was observed in the central projections of RA mechanoreceptors in E13 . 5 Gfra1 mutants , we next asked whether cis RET signaling is required for the initial growth of RA mechanosensory third order central projections . We crossed the RetCFP allele into Gfra2 and Nrtn null mice and examined central projections of RA mechanoreceptors at E13 . 5 ( Figure 2A–D ) . In contrast to what we observed at P7 , at this early development stage CFP+ central projections of RA mechanoreceptors are greatly reduced in both Gfra2 and Nrtn null SC sections ( Figure 2E–F , Figure 1—source data 2 , Gfra2 mutant has 9 . 50 ± 1 . 44% of control staining at thoracic levels [p < 0 . 001] ) . In addition , similar to the E13 . 5 Ret mutant mice , the number of CFP+ DRG neurons in Gfra2 and Nrtn null mice is comparable to that of control mice ( Figure 2G–H , Figure 1—source data 2 ) , suggesting that the loss of CFP+ fibers in the dSC of these mutant mice is due to a deficit in the interstitial central projection growth of RA mechanoreceptors . Thus , at E13 . 5 , Gfra2 and Nrtn null mice phenocopy the central projection deficit of Ret mutant mice , which suggests that RET is activated by NRTN/GFRa2 in cis for the initial growth of RA mechanosensory central projections . 10 . 7554/eLife . 06828 . 009Figure 2 . Gfra2 and Nrtn null mice show reduced RA mechanoreceptor central projections at E13 . 5 . ( A–D ) Anti-GFP immunostaining to visualize RA mechanosensory central projections in E13 . 5 dSC sections of Gfra2GFP/+;RetCFP/+ control ( A ) , Gfra2GFP/GFP;RetCFP/+ mutant ( B ) , Nrtn+/−;RetCFP/+ control ( C ) , and Nrtn−/−;RetCFP/+ mutant ( D ) mice . ( E–F ) Quantification of CFP+ pixel number in dSC of Gfra2 ( E ) and Nrtn ( F ) mice . The dramatic reduction in CFP+ axons in Gfra2 and Nrtn nulls at E13 . 5 suggests that cis activation of RET is required for the initial growth of RA mechanosensory third order central projections . ( G–H ) Quantification of number of CFP+ neurons per DRG section in Gfra2 ( G ) and Nrtn ( H ) mice . Similar number of CFP+ DRG neurons between control and mutant mice indicates that cell death of RA mechanoreceptors or downregulation of RetCFP allele do not occur at E13 . 5 when cis RET signaling is ablated . Scale bar = 50 μm . Error bars represent SEM . n . s . = p > 0 . 05 , ** = p < 0 . 01 *** = p < 0 . 001 . Source data are provided in Figure 1—source data 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 06828 . 009 If Ret , Gfra2 , and Nrtn null mice phenocopy each other at E13 . 5 , why do their postnatal VGLUT1 staining patterns look so different ( Figure 1 and [Luo et al . , 2009] ) ? One possibility is that since Ret has a much broader expression pattern than Gfra2 in the dSC and DRGs , the dramatic loss of VGLUT1 staining in layers III–V of SC may be caused by the loss of RET signaling both in RA mechanoreceptors and other RET+ cells . For Gfra2 and Nrtn mutant mice , though central projection deficits of RA mechanoreceptors may persist postnatally , VGLUT1+ puncta from SA mechanoreceptors could mask the phenotype . Alternatively , RA mechanosensory central projections in Gfra2 and Nrtn mutant mice could recover at later developmental stages due to the function of other RET signaling mechanisms . To differentiate these possibilities , we examined central projections of RA mechanoreceptors in Gfra2 null mice through development . We focused on Gfra2 instead of Nrtn mutant mice because: ( 1 ) the cell autonomous requirement of a co-receptor is the key to differentiate cis vs trans RET signaling; and ( 2 ) Gfra2 and Nrtn null mice display very similar phenotypes of RA mechanoreceptors . Since Ret begins to be expressed in additional populations of DRG neurons ( Molliver et al . , 1997; Luo et al . , 2007 ) and dSC cells ( Figure 1—figure supplement 1B ) from E15 . 5 , we can't use the RetCFP allele to visualize the central projections of RA mechanoreceptors at late developmental stages . To overcome this problem , we used a tandem allele ( see ‘Materials and methods’ and Figure 3—figure supplement 1 ) of an inducible Cre allele of Ret ( RetCreERT ) and Rosa26 conditional red fluorescent protein ( RosaTdt ) . We combined these alleles with early ( E11 . 5 and E12 . 5 ) 4-hydroxy tamoxifen ( 4-HT ) treatment to specifically trace RA mechanoreceptors , as previously established ( Luo et al . , 2009 ) . We generated Gfra2GFP/+; RetCreERT; RosaTdt control and Gfra2GFP/GFP; RetCreERT; RosaTdt mutant mice and examined their SC and DRG sections at E15 . 5 . Tdt+ fibers innervate layers III–V of the SC , which is consistent with specific genetic tracing of RA mechanoreceptors ( Luo et al . , 2009 ) . In addition , the majority of Tdt+ DRG neurons are RET+ , GFRa2+ ( reported by the expression of GFP ) , but NTRK1− at E15 . 5 ( Figure 3A–J ) , further supporting the specific labeling of RA mechanoreceptors . We found that central projections of Gfra2 null RA mechanoreceptors are also decreased at E15 . 5 ( Figure 3K–N , Figure 3—source data 1 , Gfra2 mutant has 55 . 13 ± 2 . 82% of control staining at the thoracic level [p < 0 . 001] ) . Since the number of labeled DRG neurons is not significantly reduced in the mutant mice ( Gfra2 mutants have 79 . 52 ± 8 . 39% of control cell number [p = 0 . 06] ) , the central projection phenotype mostly reflects a growth deficit at this developmental stage . Noticeably , the relative reduction of innervation in Gfra2 null mice at E15 . 5 is less severe compared to that of E13 . 5 mutants ( Figure 2 ) , suggesting that central projections of Gfra2 null RA mechanoreceptors may start to recover at this stage . 10 . 7554/eLife . 06828 . 010Figure 3 . Central projection growth deficit of Gfra2 null RA mechanoreceptors at E15 . 5 . ( A–I ) E15 . 5 Gfra2GFP/+;RetCreERT/+;RosaTdt DRG sections stained with anti-RET ( A–C ) , anti-NTRK1 ( D–F ) , and anti-GFP ( G–I ) . ( J ) Quantification of percentage of Tdt+ DRG neurons which co-express RET ( 96 . 16 ± 0 . 28% ) , NTRK1 ( 6 . 56 ± 0 . 18% ) , and GFP driven from the Gfra2 locus ( 86 . 48 ± 0 . 55% ) . The expression profile of Tdt+ neurons confirms that this genetic labeling strategy specifically labels RA mechanoreceptors . ( K–L ) Visualization of Tdt+ RA mechanosensory central projections in dSC of E15 . 5 Gfra2GFP/+; RetCreERT/+; RosaTdt control ( K ) and Gfra2GFP/GFP; RetCreERT/+; RosaTdt mutant ( L ) SC sections . ( M ) Quantification of Tdt+ pixels in dSC , which is displayed as a percentage normalized to dSC Tdt+ pixels of the within litter controls . Gfra2 mutant mice have 55 . 13 ± 2 . 82% of control staining ( p < 0 . 001 ) . Note that although Gfra2 null RA mechanoreceptors still have a central projection deficit at E15 . 5 , the reduction at this stage is less severe than the deficit observed at E13 . 5 . ( N ) Quantification of number of Tdt+ neurons per DRG section , which is displayed as a percentage normalized to Tdt+ neurons of the within litter controls . Gfra2 mutant mice have 79 . 52 ± 8 . 39% of control cell number ( p = 0 . 06 ) , which suggests that the survival of RA mechanoreceptors is not dependent on cis signaling at this stage . Scale bars = 100 μm ( A–I ) , 50 μm ( K–L ) . Error bars represent SEM . n . s . = p > 0 . 05 , *** = p < 0 . 001 . Source data are provided in Figure 3—source data 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 06828 . 01010 . 7554/eLife . 06828 . 011Figure 3—source data 1 . RA mechanoreceptor central projections and cell number in E15 . 5 Gfra2mutants . DOI: http://dx . doi . org/10 . 7554/eLife . 06828 . 01110 . 7554/eLife . 06828 . 012Figure 3—figure supplement 1 . Generation of tandem RetCreERT;RosaTdt allele . The Ret and Rosa loci are located ∼5 megabases apart on mouse chromosome 6 . RetCreERT/+ mice were crossed to RosaTdt/Tdt mice to generate RetCreERT/+;RosaTdt/+ mice , which were crossed back to RosaTdt/Tdt mice . Occasionally , an interchromosomal recombination event occurred between the Ret and Rosa loci , which caused RetCreERT and RosaTdt to be located on the same chromosome . Recombinants were identified by genotyping for the RetCreERT allele and the homozygous presence of the RosaTdt allele . The chromosome containing both RetCreERT and RosaTdt alleles is called the tandem RetCreERT;RosaTdt allele and maintained by mating with RosaTdt/Tdt mice . DOI: http://dx . doi . org/10 . 7554/eLife . 06828 . 012 To determine if RA mechanoreceptors require Ret but not Gfra2 for their central projection growth at later developmental stages , we generated E18 . 5 RetCreERT/+;RosaTdt control and RetCreERT/CreERT;RosaTdt mutant embryos ( RetCreERT is a null allele of Ret ) . Consistent with previous results ( Bourane et al . , 2009; Luo et al . , 2009; Honma et al . , 2010 ) , we found that RA mechanosensory central projections are greatly reduced in the Ret mutant mice ( Figure 4A , C , I , Figure 4—source data 1 , Ret mutant has 35 . 86 ± 4 . 97% of control staining at thoracic levels [p < 0 . 001] ) . In addition , we counted the number of Tdt+ neurons in L4/L5 DRGs and found that the number of Tdt+ RA mechanoreceptors is dramatically reduced as well ( Figure 4B , D , J , Ret mutant has 52 . 52 ± 7 . 76% of control cell number [p < 0 . 001] ) . Taken together , these results suggest that Ret is absolutely required for both survival and central projection growth of RA mechanoreceptors at E18 . 5 . 10 . 7554/eLife . 06828 . 013Figure 4 . Ret and Gfra2 null mice display different central projection and cell survival deficits at E18 . 5 . ( A–H ) SC sections and whole mount L4/L5 DRGs of Tdt labeled RA mechanoreceptor from E18 . 5 RetCreERT/+;RosaTdt control ( A–B ) , RetCreERT/CreERT;RosaTdt mutant ( C–D ) , Gfra2GFP/+;RetCreERT/+;RosaTdt control ( E–F ) , and Gfra2GFP/GFP;RetCreERT/+;RosaTdt mutant ( G–H ) embryos . ( I ) Quantification of Tdt+ pixels in dSC , which is displayed as a percentage normalized to dSC Tdt+ pixels of the within litter controls . ( J ) Quantification of the number of Tdt+ DRG neurons per whole-mount L4/L5 DRG , which is displayed as a percentage normalized to Tdt+ neurons of the within litter controls . Ret mutants have significant decreases in RA mechanosensory axons innervating the dSC and in the number of Tdt+ RA mechanoreceptors , suggesting that Ret mutants have deficits in both the growth of third order central projections and the survival of RA mechanoreceptors at E18 . 5 . In contrast , Gfra2 nulls have only minor deficits in RA mechanosensory central projection growth and the survival or RA mechanoreceptors , suggesting that an additional GFRa2 independent but RET-dependent mechanism functions in these processes . Scale bar = 50 μm . Error bars represent SEM . * = p < 0 . 05 , *** = p < 0 . 001 . Source data are provided in Figure 4—source data 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 06828 . 01310 . 7554/eLife . 06828 . 014Figure 4—source data 1 . RA mechanoreceptor central projections and cell number in E18 . 5 Ret , Gfra2 , Gfra1 , and Gfra1;Gfra2 mutants . DOI: http://dx . doi . org/10 . 7554/eLife . 06828 . 01410 . 7554/eLife . 06828 . 015Figure 4—figure supplement 1 . Gfra2 null RA mechanoreceptors retain phospho-S6 expression . ( A–B ) Anti-GFP ( green ) and anti-phospho-S6 ( red ) staining of P0 Gfra2GFP/+ control ( A ) and Gfra2GFP/GFP null ( B ) DRG sections . ( C ) Quantification of percentage of GFP+ neurons which express phospho-S6 shows no significant change in proportion of phospho-S6+ RA mechanoreceptors ( 92 . 53 ± 1 . 55% of Gfra2GFP/+ GFP+ DRG neurons express phospho-S6 , 93 . 66 ± 0 . 14% of Gfra2GFPGFP+ GFP+ DRG neurons express phospho-S6 , p = 0 . 51 ) . Therefore , active RTK signaling seems to still occur in Gfra2 null RA mechanoreceptors . DOI: http://dx . doi . org/10 . 7554/eLife . 06828 . 015 In contrast , central projections of Tdt+ Gfra2 null RA mechanoreceptors are only slightly reduced at E18 . 5 ( Figure 4E , G , I , Figure 4—source data 1 , Gfra2 mutant has 86 . 34 ± 4 . 48% of control staining at thoracic levels [p = 0 . 01] ) . At P7 , almost no difference is observed ( data not shown ) . Similarly , the number of Tdt+ RA mechanoreceptors is only slightly reduced in Gfra2 null mice ( Figure 4F , H , J Gfra2 mutant has 84 . 01 ± 5 . 16% of control cell number [p = 0 . 04] ) , indicating that extensive cell death of RA mechanoreceptors resulting from an absence of RET signaling does not occur in Gfra2 null mice . The discrepancy between E18 . 5 Ret and Gfra2 mutant phenotypes suggests that RET signaling still occurs in neonatal Gfra2 null RA mechanoreceptors . To demonstrate this , we quantified the expression of phospho-S6 ribosomal protein , which is downstream of RET/PI3K/mTOR signaling ( Plaza-Menacho et al . , 2010 ) , in RA mechanoreceptors . We found that the proportion of GFP+ RA mechanoreceptors which express phospho-S6 in P0 Gfra2GFP/+ control and Gfra2GFP/GFP mutant DRGs was similar ( Figure 4—figure supplement 1 [p = 0 . 51] ) . This result is consistent with the idea that RET activation occurs in neonatal RA mechanoreceptors without Gfra2 . Collectively , our results suggest that Gfra2 null RA mechanoreceptors display a central projection deficit at E13 . 5 but recover during later development , which explains the almost normal VGLUT1 staining in layers III–V of SC at P7 . In addition , our data indicate that from E15 . 5 , an additional GFRa2 independent but RET-dependent mechanism begins to play a role in promoting the survival and central projection growth of RA mechanoreceptors . To determine if this GFRa2-independent but RET-dependent mechanism requires GFRa1 , we examined genetically labeled Gfra1+/−;RetCreERT/+;RosaTdt control and Gfra1−/−;RetCreERT/+;RosaTdt mutant SC and DRGs at E18 . 5 . Similar to E13 . 5 , neither RA mechanosensory central projections nor their number is significantly decreased in Gfra1 null mice ( Figure 5A–D , I–J , Figure 4—source data 1 ) , suggesting that simply disrupting trans activation of RET via GFRa1 is not sufficient to block Ret signaling in RA mechanoreceptors . 10 . 7554/eLife . 06828 . 016Figure 5 . Gfra1;Gfra2 double null mice phenocopy Ret mutants at E18 . 5 . ( A–H ) SC sections and whole mount L4/L5 DRGs of Tdt labeled RA mechanoreceptors from E18 . 5 Gfra1+/−;RetCreERT/+;RosaTdt control ( A–B ) , Gfra1−/−;RetCreERT/+;RosaTdt mutant ( C–D ) , Gfra1+/−; Gfra2GFP/+;RetCreERT/+;RosaTdt control ( E–F ) and Gfra1−/−; Gfra2GFP/GFP;RetCreERT/+;RosaTdt double null ( G–H ) embryos . ( I ) Quantification of Tdt+ pixels in dSC , which is displayed as a percentage normalized to dSC Tdt+ pixels of the within litter controls . ( J ) Quantification of number of Tdt+ DRG neurons per DRG , which is displayed as a percentage normalized to Tdt+ neurons of the within litter controls . Gfra1 mutants have no significant deficits in RA mechanosensory third order projections or cell survival at E18 . 5 , indicating that ablating trans signaling alone is not sufficient to disrupt the development of RA mechanoreceptors . However , loss of both cis and trans signaling in Gfra1;Gfra2 double nulls leads to a significant loss of RA mechanosensory third order projection growth and cell number , suggesting that both cis and trans RET signaling contribute to the development of RA mechanoreceptors . Scale bars = 50 μm . Error bars represent SEM . n . s . = p > 0 . 05 , *** = p < 0 . 001 . Source data are provided in Figure 4—source data 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 06828 . 016 The lack of Ret-mutant-like survival and central projection phenotypes of RA mechanoreceptors in both Gfra1 and Gfra2 single null mice made us wonder if cis and trans RET signaling function in the same developmental process and thus loss of one co-receptor can be compensated for by the other . To test this idea , we generated Gfra1;Gfra2 double knockout mice , in which RA mechanoreceptors were specifically labeled with Tdt using the RetCreERT;RosaTdt tandem allele . We examined control and double null SC sections and DRGs at E18 . 5 . We found that Tdt+ RA mechanosensory central projections are greatly reduced in the dSC ( Figure 5E , G , I , Figure 4—source data 1 , Gfra1;Gfra2 double mutant has 27 . 25 ± 2 . 09% of control staining at thoracic levels [p < 0 . 001] ) . In addition , fewer Tdt+ RA mechanoreceptors remain in the double knockout DRGs ( Figure 5F , H , J , Gfra1;Gfra2 double mutant has 38 . 17 ± 2 . 65% of control cell number [p < 0 . 001] ) , indicating that a significant number of RA mechanoreceptors die in the absence of Gfra1 and Gfra2 . Strikingly , the extent of reduction in both cell number and central projections of RA mechanoreceptors is comparable between the Ret null and Gfra1:Gfra2 double null mice . Thus , our in vivo analyses strongly suggest that RET in RA mechanoreceptors is activated via both GFRa1 and GFRa2 . Is RET in RA mechanoreceptors activated by GFRa1 in cis or trans ? Although Gfra1 is not widely expressed in RA mechanoreceptors in wild-type mice , could it be upregulated to compensate for the loss of Gfra2 in the Gfra2 null mice ? To address these questions , we conducted double fluorescent in situ hybridization of Gfra1 and GFP with E14 . 5 Gfra2GFP/+ control and Gfra2GFP/GFP null DRG sections . We found that a comparable low number of GFP+ neurons expressed Gfra1 transcript in both mutants and controls ( Figure 6A–C [p = 0 . 52] ) , suggesting that Gfra1 is not upregulated in Gfra2 null RA mechanoreceptors . In addition , we performed in situ hybridization of Gfra1 with P0 Gfra2GFP/+;Ntrk1+/− control , Gfra2GFP/+;Ntrk1−/− null , and Gfra2GFP/GFP;Ntkr1−/− double null DRG sections . We previously showed ( Luo et al . , 2009 ) that Gfra1 is expressed in NTRK1+ DRG neurons and that the expression of Gfra1 is completely lost in Ntrk1 null mice . Here , we found that while Gfra1 expression was observed in Gfra2GFP/+;Ntrk1+/− control DRGs , no Gfra1 expression was observed in either Gfra2GFP/+;Ntkr1−/− null or Gfra2GFP/GFP;Ntkr1−/− double null DRGs ( Figure 6D–F ) . This result indicates that the expression of Gfra1 in Gfra2 null DRG neurons still fully depends on NTRK1 signaling and thus it must be expressed in the non-RA mechanoreceptors . Moreover , we performed quantitative RT-PCR ( QPCR ) for Gfra1 transcripts in DRGs from E13 . 5 , E15 . 5 , and E18 . 5 Gfra2GFP/+ control and Gfra2GFP/GFP mutant embryos . We found no significant difference in the expression of Gfra1 between control and mutant DRGs at any stage ( Figure 6—figure supplement 1 , Figure 6—source data 1 ) , suggesting that Gfra1 is not transcriptionally upregulated in DRG neurons upon Gfra2 ablation . 10 . 7554/eLife . 06828 . 017Figure 6 . Gfra1 is not upregulated in Gfra2 null RA mechanoreceptors . ( A–B ) Double fluorescent in situ hybridization against GFP and Gfra1 on E14 . 5 Gfra2GFP/+ control ( A ) and Gfra2GFP/GFP null ( B ) DRG sections . ( C ) Quantification of percentage of GFP+ neurons which co-express Gfra1 . 12 . 81 ± 3 . 92% of control GFP+ neurons express Gfra1 , and 16 . 17 ± 3 . 31% of Gfra2 null GFP+ neurons express Gfra1 ( p = 0 . 52 ) . The comparable low number of DRG neurons co-expressing GFP and Gfra1 in control and Gfra2 nulls suggests that Gfra1 normally is not expressed in most RA mechanoreceptors and that no upregulation of Gfra1 occurs in Gfra2 null RA mechanoreceptors . ( D–F ) In situ hybridization against Gfra1 in P0 Gfra2GFP/+;Ntrk1+/− control ( D ) , Gfra2GFP/+;Ntrk1−/− null ( E ) , and Gfra2GFP/GFP;Ntrk1−/− double null ( F ) DRG and SC sections . Black border outlines DRG . In control DRG sections , Gfra1 is expressed in some DRG neurons . In Gfra2GFP/+;Ntrk1−/− null DRG sections , Gfra1 transcript is not detected because the DRG neurons which normally express detectable levels of Gfra1 don't survive in the absence of Ntrk1 . In Gfra2;Ntrk1 double null mice , no Gfra1 expression is detected in DRG neurons as well , which further supports that upregulation of Gfra1 doesn't occur in Gfra2 null RA mechanoreceptors . Scale bars = 50 μm . Error bars represent SEM . n . s . = p > 0 . 05 . DOI: http://dx . doi . org/10 . 7554/eLife . 06828 . 01710 . 7554/eLife . 06828 . 018Figure 6—source data 1 . QPCR of Gfra1 in embryonic Gfra2 null DRGs . DOI: http://dx . doi . org/10 . 7554/eLife . 06828 . 01810 . 7554/eLife . 06828 . 019Figure 6—figure supplement 1 . Quantitative RT-PCR ( QPCR ) of Gfra1 in Gfra2 null DRGs . QPCR for Gfra1 from cDNAs generated from E13 . 5 , E15 . 5 , and E18 . 5 Gfra2GFP/+ control and Gfra2GFP/GFP null DRGs . ( A ) ΔCT values ( cycles to reach threshold for Gfra1 minus cycles to reach threshold for Gapdh , a housekeeping gene ) are not significantly different between control and mutant DRGs at E13 . 5 , E15 . 5 , or E18 . 5 , suggesting that transcription of Gfra1 is not changed in Gfra2 mutants . Error bars represent standard deviation , n . s . = p > 0 . 05 . ( B–D ) Relative quantification of Gfra1 expression levels at E13 . 5 ( B ) , E15 . 5 ( C ) , and E18 . 5 ( D ) calculated by 2−ΔΔCT . Error bars represent range of expression based on 2−ΔΔCT calculated ± the standard deviation of CT . Source data are provided in Figure 6—source data 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 06828 . 019 Although Gfra1 transcript in most RA mechanoreceptors is below the detection level of in situ hybridization , it remains possible that an undetectable amount of GFRa1 could function in cis to promote RET signaling in RA mechanoreceptors . To exclude this possibility and to demonstrate that RET in RA mechanoreceptors is indeed activated by GFRa1 in trans , we used DRG explants from E14 . 5 embryos of different mutant backgrounds and treated these explants with NRTN , GDNF , GFRa1 plus GDNF , or GFRa1 alone . In E14 . 5 explants harboring the RetCFP allele , the cell bodies and axons of RET+ neurons , some of which are RA mechanoreceptors , can be identified by anti-GFP immunostaining . We found that CFP+ neurons in RetCFP/+ control DRG explants grow long axons upon NRTN , GDNF , or GFRa1 plus GDNF , but not GFRa1 alone treatment ( Figure 7—figure supplement 1A–D , I , and Figure 7—source data 1 ) . In addition , the number of CFP+ DRG neurons is reduced in GFRa1 alone culture ( Figure 7—figure supplement 1J , Figure 7—source data 2 ) , suggesting that either cell death or down-regulation of Ret , and thus CFP expression , occur in the absence of RET signaling . Similarly , CFP+ neurons in RetCFP/CFP null DRG explants lost their responsiveness to GFLs completely ( Figure 7—figure supplement 1E–H , I–J , Figure 7—source data 2 ) , suggesting that this assay reflects RET-dependent signaling . Next , we examined DRG explants harboring the Gfra2GFP allele , which drives a much lower level of GFP expression than RetCFP . Although some small diameter DRG neurons also express Gfra2 around P0 or later ( Luo et al . , 2007 ) , in this Gfra2GFP mouse line GFP is mainly detected in the large diameter RA mechanoreceptors ( Luo et al . , 2009 ) , which express a much higher level of Gfra2 . Therefore , anti-GFP staining of E14 . 5 Gfra2GFP DRG explants should specifically show RA mechanoreceptors . Since GFP+ axons of these explants could not be reliably imaged and quantified due to the low level of GFP expression , we approximated the extent of RET signaling in Gfra2GFP DRG explants by quantifying the number of discernable GFP+ cell bodies . We found that Gfra2GFP/+ control DRG neurons show robust responses upon GFL application ( Figure 7A–D , Q , Figure 7—source data 2 ) . Interestingly , Gfra2GFP/GFP null DRG neurons lost their responsiveness to NRTN , but retain GFP expression in the presence of either GDNF or GFRa1 plus GDNF ( Figure 7E–H , Q , Figure 7—source data 2 ) . These results suggest that a GFRa2-independent but RET-dependent mechanism can mediate GDNF responsiveness of RA mechanoreceptors . 10 . 7554/eLife . 06828 . 020Figure 7 . RA mechanoreceptors utilize GFRa1 produced by neighboring neurons to respond to GDNF . ( A–P ) DRG explants from Gfra2GFP/+ control ( A–D ) , Gfra2GFP/GFP null ( E–H ) , Gfra2GFP/+;Ntrk1−/− null ( I–L ) , and Gfra2GFP/GFP;Ntrk1−/− double null ( M–P ) embryos grown for 1 day in vitro and stained with anti-GFP antibody . Explants were treated with NRTN ( 50 ng/ml ) , GDNF ( 100 ng/ml ) , GDNF ( 100 ng/ml ) plus GFRa1 ( 300 ng/ml ) , or GFRa1 ( 300 ng/ml ) , respectively . Schematic next to each genotype depicts the presence of RET and GFRas in each condition , and green color indicates cells detected by anti-GFP staining . ( Q ) Quantification of number of GFP+ neurons per 10 , 000 μm2 of explant in Gfra2GFP/+ control and Gfra2GFP/GFP null explants . GFP driven from the Gfra2 locus indicates RET signaling activity . Gfra2 control explants display many GFP+ neurons upon NRTN , GDNF , and GDNF plus GFRa1 treatment , but do not respond to GFRa1 alone . Gfra2 null explants lose their responsiveness to NRTN , but remain responsive to GDNF and GDNF plus GFRa1 . ( R ) Quantification of number of GFP+ neurons per 10 , 000 μm2 of explant in Gfra2GFP/+;Ntrk1−/− null and Gfra2GFP/GFP;Ntrk1−/− double null explants . In a Ntrk1 null background , expression of Gfra1 is lost in non-RA-mechanoreceptor DRG neurons . Gfra2GFP/+;Ntrk1−/− null explants respond to NRTN , GDNF , and GDNF plus GFRa1 . In this case , it is likely that GDNF activates RET signaling by interacting with GFRa2 ( Jing et al . , 1997; Sanicola et al . , 1997; Rossi et al . , 1999 ) . In contrast , Gfra2;Ntrk1 double null DRG explants show GFP expression upon treatment with a combination of GDNF and GFRa1 , but completely lose their responsiveness to GDNF . These results indicate that Gfra2 null RA mechanoreceptors do not express GFRa1 at a functional level and they depend on GFRa1 produced by neighboring NTRK1+ neurons to respond to GDNF . See Figure 7—source data 2 for quantification . ( S–V ) Western blot analysis of cell lysates and concentrated supernatants from cultured dissociated DRG neurons of E18 . 5-P1 wild-type , Gfra2 null , and Gfra1 null mice . ( S ) The specificity of the anti-GFRa1 antibody was confirmed by the loss of a doublet at the predicted size of GFRa1 in Gfra1 null cell lysates . GFRa1 was also detected in the supernatants of wild-type and Gfra2 null cultures , but not Gfra1 null cultures , indicating that GFRa1 is shed from the membrane of DRGs of both wild-type and Gfra2 mutants . Note that the size of cleaved GFRa1 is slightly smaller than that tethered to cells , which is consistent with previous publication ( Paratcha et al . , 2001 ) . Following detection of GFRa1 , membranes were stripped and probed for β-actin , which served as a loading control and confirmation that the supernatant fraction was not contaminated with cells or cellular debris ( lower panel ) . ( T ) The specificity of the anti-GFRa2 antibody was confirmed by the loss of a band ∼75 kDa in Gfra2 null cell lysates . The larger than predicted size of GFRa2 may be due to post-translational modifications . Two GFRa2 specific bands were also detected in the supernatants of wild-type and Gfra1 null cultures , but not Gfra2 null cultures , indicating that GFRa2 is also shed from DRG cell membranes . The size of cleaved GFRa2 is also smaller than that tethered to cells . ( U–V ) Densimetric quantification of anti-GFRa1 blots shows no significant change in the level of GFRa1 produced by cells ( U ) or released into the media ( V ) , which suggests that there is no compensation for the loss of GFRa2 through changes in the expression or release of GFRa1 . See Figure 7—source data 3 for quantification . Error bars represent SEM . Scale bars = 50 μm . n . s . = p > 0 . 05 , *** = p < 0 . 001 . Source data are provided in Figure 7—source data 2 , 3 . DOI: http://dx . doi . org/10 . 7554/eLife . 06828 . 02010 . 7554/eLife . 06828 . 021Figure 7—source data 1 . Quantification of axonal growth in Ret mutant DRG explants . DOI: http://dx . doi . org/10 . 7554/eLife . 06828 . 02110 . 7554/eLife . 06828 . 022Figure 7—source data 2 . GFP+ neuron number in Gfra2 null and Gfra2;Ntrk1 double null explants . DOI: http://dx . doi . org/10 . 7554/eLife . 06828 . 02210 . 7554/eLife . 06828 . 023Figure 7—source data 3 . Densimetric measurements of GFRa1 in DRG cell extracts and supernatants . DOI: http://dx . doi . org/10 . 7554/eLife . 06828 . 02310 . 7554/eLife . 06828 . 024Figure 7—figure supplement 1 . RetCFP null DRG explants lose responsiveness to GFLs . ( A–H ) DRG explants from E14 . 5 RetCFP/+ control ( A–D ) and RetCFP/CFP null ( E–H ) embryos grown for 1 day in vitro and stained with anti-GFP antibody . Explants were treated with NRTN ( 50 ng/ml ) , GDNF ( 100 ng/ml ) , GDNF ( 100 ng/ml ) plus GFRa1 ( 300 ng/ml ) , or GFRa1 ( 300 ng/ml ) , respectively . Schematic next to each genotype depicts the presence of RET and GFRas in each condition , and green color indicates cells detected by anti-GFP staining . ( I ) Quantification of number of axonal intersections at a 200 μm distance from the edge each explant for Ret control and null explants . CFP+ neurons in control explants grow numerous axons upon treatment with NRTN , GDNF , or GDNF plus GFRa1 , but not in response to GFRa1 alone . Ret null explants do not grow axons upon treatment with GFLs . ( J ) Quantification of number of CFP+ neurons per 10 , 000 μm2 of explant in Ret control and null explants . Since RET signaling positively regulates Ret expression , CFP driven from the Ret locus serves as a readout of RET signaling activity as well . Ret control explants have many CFP+ neurons upon NRTN , GDNF , and GDNF plus GFRa1 treatment , but Ret null explants do not respond to treatment with GFLs . Scale = 50 µm . Source data are provided in Figure 7—source data 1 , 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 06828 . 024 How can Gfra2 null RA mechanoreceptors retain their responsiveness to GDNF ? It could be due to: ( 1 ) a very low level of GFRa1 is expressed in RA mechanoreceptors , which activates RET in cis in the presence of GDNF; or ( 2 ) GFRa1 expressed by neighboring DRG neurons binds GDNF and activates RET in RA mechanoreceptors in trans . To differentiate between these possibilities , we cultured E14 . 5 Gfra2GFP/GFP;Ntrk1−/− double mutant DRGs . Since the expression of Gfra1 in non-RA mechanoreceptor DRG neurons fully depends on NTRK1 signaling , as shown previously ( Luo et al . , 2009 ) and above ( Figure 6D–F ) , GFRa1 should be depleted from non-RA mechanoreceptors in Gfra2GFP/GFP;Ntrk1−/− double mutant DRGs . Therefore , if GFRa1 is expressed at a low level in RA mechanoreceptors and activates RET in cis , the double null explants should retain their responsiveness to GDNF . On the other hand , if GFRa1 expressed by neighboring neurons activates RET in RA mechanoreceptors in trans , the GDNF responsiveness would be lost in the Gfra2;Ntrk1 double nulls . Here , we found that Gfra2GFP/+;Ntrk1−/− control explants were responsive to NRTN , GDNF , and GDNF plus GFRa1 , but not GFRa1 alone ( Figure 7I–L , R , Figure 7—source data 2 ) . In contrast , Gfra2GFP/GFP;Ntrk1−/− double null explants only respond to GDNF plus GFRa1 , but not to NRTN , GDNF , or GFRa1 ( Figure 7M–P , R , Figure 7—source data 2 ) . The loss of responsiveness of RA mechanoreceptors to GDNF in Gfra2;Ntrk1 double null DRG explants strongly suggests that Gfra1 is not expressed at a functional level in RA mechanoreceptors and that RET in Gfra2 null RA mechanoreceptors is activated by exogenous GFRa1 from the neighboring DRG neurons in trans . Trans activation of RET could occur by direct contact between membranes of cells which express either RET or GFRa1 , or by soluble GFRa1 which is shed from the cell surface . To determine whether GFRa1 is released by DRG neurons , we cultured dissociated DRGs from E18 . 5-P1 wild-type , Gfra2−/− , and Gfra1−/− mice . We collected cell lysates and concentrated media from days 3–6 in vitro and then performed Western blot analysis . Immunoblotting with anti-GFRa1 revealed a doublet at ∼55–65 kDa in wild-type and Gfra2−/− cell lysates , which was absent in the Gfra1−/− samples ( Figure 7S , lanes 1–3 ) , confirming the specificity of the anti-GFRa1 antibody . A positive band of ∼55 kDa was present in concentrated supernatants of wild-type and Gfra2−/− but not Gfra1−/− cultures ( Figure 7S , lanes 4–6 ) , suggesting that soluble GFRa1 is shed from neonatal DRG cells . Together with reports of GFRa1 being shed by Sciatic nerve Schwann cells , immortalized neuronal progenitors ( Paratcha et al . , 2001 ) , and adult DRG explants ( He et al . , 2014 ) , our findings indicate that GFRa1 can be released by many cell types during both developmental and adult stages . Therefore , it is possible that RET in RA mechanoreceptors is activated in trans by both soluble GFRa1 and GFRa1 tethered to the membranes of neighboring cells . In addition , although there is no significant increase of Gfra1 transcripts in Gfra2 null DRGs by in situ or QPCR ( Figure 6 and Figure 6—figure supplement 1 ) , it remains possible that post-transcriptional regulation may occur to alter the translation , perdurance , or release of GFRa1 . To test this possibility , we quantified the amount of GFRa1 in cell lysates and supernatants of wild-type and Gfra2−/− cultures by densitometry . We found that the amount of GFRa1 expressed in the cell or shed into the media was not significantly different between wild-type and Gfra2 null cultures ( Figure 7U–V , Figure 7—source data 3 ) . Therefore , Gfra2 null DRGs do not produce or release more GFRa1 protein to compensate for the loss of Gfra2 . We also investigated whether GFRa2 is normally shed by DRGs . The specificity of the anti-GFRa2 antibody was confirmed by the absence of a ∼75 kDa band from Gfra2 null cell lysates , which was present in both wild type and Gfra1 null cultures ( Figure 7T , lanes 1–3 ) . Furthermore , secreted GFRa2 band was also present in the supernatants of wild-type and Gfra1 DRG cultures , but not in Gfra2 null cultures . Therefore , both GFRa1 and GFRa2 are normally released by DRGs during early postnatal development . As described above , the central projections of Gfra2 null RA mechanoreceptors display a severe , Ret-like deficit at E13 . 5 , but begin to recover from E15 . 5 , which is due to compensation by trans signaling via GDNF/GFRa1 . Why is trans RET signaling able to compensate for the loss of cis signaling during late embryonic development , but not at E13 . 5 ? One possible reason for the delay is the availability of trans signaling components . Our in situ hybridization data suggest that Gfra1 is expressed at high levels at both E13 . 5 and E15 . 5 , but the expression of Gdnf is greatly increased in DRGs from E13 . 5 to E15 . 5 ( Figure 1—figure supplement 1 ) . To provide additional evidence for the dynamic expression of Gdnf during development , we examined DRG and SC sections of E13 . 5 and E15 . 5 GdnfLacZ/+ ( Moore et al . , 1996 ) embryos using X-Gal staining . We found that the expression of LacZ increased significantly in DRGs from E13 . 5 to E15 . 5 ( Figure 8A–E [p < 0 . 001] , Figure 8—figure supplement 1 ) . In addition , X-Gal staining was found in the E15 . 5 dorsal root , the pathway through which DRG central projections travel to reach the dSC ( Figure 8A–D , black arrows ) . Thus , the expression of Gdnf seems to significantly increase in both the DRG and dorsal root from E13 . 5 to E15 . 5 , providing a possible explanation for why the trans compensation occurs from E15 . 5 . 10 . 7554/eLife . 06828 . 025Figure 8 . Dynamic expression of GDNF during development . ( A–D ) X-Gal staining of E13 . 5 ( A , C ) and E15 . 5 ( B , D ) GdnfLacZ/+ DRG and SC sections ( also see Figure 8—figure supplement 1 ) . Arrows indicate dorsal roots , which express Gdnf at E15 . 5 , but not E13 . 5 . ( E ) Quantification of LacZ+ cells per DRG section , normalized to DRG area , reveals a significant increase in the number of cells expressing Gdnf from E13 . 5 to E15 . 5 . E13 . 5 embryos have 4 . 41 ± 0 . 82 LacZ+ cells/unit area of DRG , E15 . 5 embryos have 17 . 73 ± 0 . 70 LacZ+ cells/unit area of DRG ( p < 0 . 001 ) . Error bars represent SEM . Scale bars = 200 μm ( A–B ) , 100 μm ( C–D ) . *** = p < 0 . 001 ( F ) Model of cis and trans signaling at cell bodies and central branches of RA mechanoreceptors . GFRa2 is co-expressed with RET in RA mechanoreceptors and can activate RET in cis . GFRa2 can also be shed from the membrane and may activate RET in its soluble form . GFRa1 is expressed in neighboring DRG neurons , dorsal root entry zone cells , and dSC cells . GFRa1 present at the membrane of these cells may directly contact the cell bodies or processes of RA mechanoreceptors to activate RET in trans . In addition , soluble GFRa1 released from these cells may also activate RET in RA mechanoreceptor in trans . DOI: http://dx . doi . org/10 . 7554/eLife . 06828 . 02510 . 7554/eLife . 06828 . 026Figure 8—figure supplement 1 . Gdnf LacZ expression in DRGs at E13 . 5 and E15 . 5 . DRG sections with large , strongly LacZ+ neurons were observed at both E13 . 5 ( A ) and E15 . 5 ( B ) . Note that there are many more LacZ+ DRG neurons at E15 . 5 . Sections with such cells were observed in all embryos and were usually found in distal anterior and distal posterior segments . Sections with smaller reactive cells , as shown in Figure 6 , were observed more frequently . Scale = 100 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 06828 . 026
Previous expression analyses revealed that Gfra1 is expressed more broadly than Ret , and cells which express Gfra1 usually lie adjacent to Ret-expressing cells ( Trupp et al . , 1997; Yu et al . , 1998 ) . This expression pattern suggests that GFRa1 may have RET-independent functions or that GFRa1 may interact with RET expressed on the surfaces of other cells in trans . Indeed , evidence for both ideas has been demonstrated . GFRa1 and GDNF interact with NCAM in neurons and Schwann cells to promote neurite outgrowth and Schwann cell migration ( Paratcha et al . , 2003; Nielsen et al . , 2009 ) . Additionally , GFRa1 and GDNF are required for the proper migration of cortical GABAergic interneurons and can act as ligand-dependent adhesion molecules for synapse formation , independent of RET and NCAM ( Pozas and Ibáñez , 2005; Ledda et al . , 2007 ) . Recently , it was also shown that GFLs have additional roles in cortical development via interactions with Syndecan-3 , likely independent of RET , GFRas , and NCAM ( Bespalov et al . , 2011 ) . On the other hand , RET can be activated by GFRa1 and GDNF in trans using both heterologous cells and tissue explants ( Paratcha et al . , 2001; Ledda et al . , 2002; Patel et al . , 2012; He et al . , 2014 ) . Trans RET signaling may affect many cellular processes , including directional axonal outgrowth and promotion of axon regeneration ( Paratcha et al . , 2001; Airaksinen and Saarma , 2002; Ledda et al . , 2002 ) . Evidence for physiologically relevant in vivo function of trans RET signaling , however , has remained less conclusive . Enomoto et al . ( 2004 ) generated a ‘cis-only’ mouse model in which Gfra1 is expressed in all RET-expressing cells , but not in cells that do not express RET . Using this model , they found that the major RET-dependent developmental processes were completely normal , suggesting that trans signaling is likely to be irrelevant for most RET-dependent processes . Results from this model , however , may not necessarily preclude a physiological role for trans signaling . Not only does this model present a loss of trans signaling , but it also presents a gain of function: Gfra1 is expressed at a high level in RET+ cells which may not normally express this co-receptor . If cis and trans RET activation lead to similar physiological outcomes , any deficits due to the loss of trans signaling may be masked by a gain of cis signaling . Indeed , the gain of function of GFRa1 was recently demonstrated in enteric hematopoietic cells derived from cis-only mice ( Patel et al . , 2012 ) . Thus , whether trans RET signaling has any physiological function in development has remained an open question . We aimed to address this question by analyzing the survival and growth of RA mechanosensory central projections using loss-of-function mouse lines . Here , we found that loss of cis signaling , via ablation of either Gfra2 or Nrtn , produces a central projection deficit during early embryonic development ( Figure 2 ) . Our findings are consistent with previous observations using a different Ret knock-in line and NRTN ectopic expression ( Honma et al . , 2010 ) , but differ from the findings using anti-GFRa2 staining to visualize RA mechanosensory central projections at E13 . 5 ( Bourane et al . , 2009 ) . This is likely due to different subcellular localization of CFP and GFRa2 , as well as the expression of GFRa2 by some dSC cells ( Figure 1—figure supplement 1E , F ) , which may mask the RA mechanoreceptor phenotype . Interestingly , this phenotype recovers during late embryonic development and the central projections of Gfra2 null mice seem nearly normal postnatally ( Figures 1 , 4 ) . Thus , our results suggest that cis RET signaling is required for the initial growth of RA mechanosensory central projections , but an additional cis signaling independent process takes place during later development . Indeed , the loss of both cis and trans signaling in Gfra1;Gfra2 double mutants recapitulates the Ret phenotype ( Figures 5 , 6 ) . Furthermore , using DRG explant and dissociated culture , we demonstrated that soluble GFRa1 is normally released by DRGs and that GFRa1 produced by NTRK1+ DRG neurons present a potential source to activate RET in RA mechanoreceptors in trans to promote their survival ( Figure 7 ) . Taken together , our results suggest that trans RET signaling contributes to the development of RA mechanoreceptors in vivo . Nevertheless , the exact subcellular locus of trans RET activation in RA mechanoreceptors remains speculative . The expression pattern of Gfra1 suggests that trans RET activation is possible in the axons of RA mechanoreceptors along their path to dSC , and/or at the cell body within the DRG . Although individual DRG cell bodies are surrounded by satellite glial cells , large macromolecules and proteins are able to invade the space between the neuron and satellite cell ( Hanani , 2005 ) , suggesting trans RET activation by soluble GFRa1 could occur within DRGs . Signaling of neurotrophic RTKs , such as NTRK1 , NTRK2 , and NTRK3 , is critical for the specification and survival of numerous classes of neurons ( Ernsberger , 2009 ) . RET signaling also plays important roles in survival , differentiation , and specification of distinct neuronal classes ( Enomoto , 2005 ) . For example , RET signaling components are absolutely required for the survival of enteric neurons ( Taraviras et al . , 1999 ) , but their roles in DRG neuron survival are more complicated to dissect . Previously , it was reported that the number of total DRG neurons is not significantly reduced in neonatal and early postnatal Ret mutants ( Luo et al . , 2007 ) . At that time , specific molecular markers or genetic approaches for labeling RA mechanoreceptors had not been identified , so it was impossible to specifically assay the role of RET signaling in the survival of this neuronal population . Given that RA mechanoreceptors represent a very small proportion of the total DRG neurons ( Molliver et al . , 1997; Luo et al . , 2007 ) , a partial loss of this population may not lead to a significant change in cell counts of total DRG neurons . In another paper ( Luo et al . , 2009 ) , it was proposed that RA mechanoreceptors depend on NRTN-GFRa2/Ret signaling for their development . This was based on the findings that GFRa2 is the only co-receptor expressed in RA mechanoreceptors and that Ret , Gfra2 , and Nrtn null mice display the same no-Pacinian-corpuscle phenotype . Since RET can't be used as a molecular marker to quantify the number of RA mechanoreceptors in Ret null mice , the number of P0 RET+/NTRK1− and Gfra2GFP DRG neurons , most of which indicate the RA mechanoreceptors , was quantified in Nrtn and Gfra2 nulls . No significant change in cell number between mutants and controls was found , suggesting that Gfra2 and Nrtn are not required for survival of neonatal RA mechanoreceptors . These results are interesting in light of the current findings . Here , we show that when cis signaling via NRTN/GFRa2 is perturbed ( as was tested in [Luo et al . , 2009] ) , trans signaling via GDNF/GFRa1 can activate RET in RA mechanoreceptors to support their survival and central projection growth . When the number of genetically labeled RA mechanoreceptors was quantified in different mutant backgrounds , we found only marginal changes in Gfra2 nulls but drastic decreases in Ret mutants at E18 . 5 ( Figure 4 ) . The slight difference between the current and previous findings regarding the loss of RA mechanoreceptors in Gfra2 nulls at P0 ( Luo et al . , 2009 ) is likely due to different methods in identifying and quantifying RA mechanoreceptors . In short , the current study clarifies that RET signaling is required for RA mechanoreceptor survival but simple disruption of cis RET signaling components may not reveal this deficit . What is the purpose for RET to be activated both in cis and in trans ? Do cis and trans signaling activate different cellular responses and influence distinct developmental processes , or do cis and trans signaling exert similar physiological effect ? Here , we found that , in RA mechanoreceptors , cis and trans signaling seem to produce similar biological outputs in vivo . Our results demonstrate that cis and trans signaling can compensate for the loss of each other to promote both the central projection growth and survival of RA mechanoreceptors ( Figure 8F ) . This compensatory ability suggests that the existence of both cis and trans activation is likely to enhance but not diversify outcomes of RET signaling . Consistent with this notion , a recent study found that peripheral nerves secrete both GDNF and GFRa1 , which attracts perineural invasion of heterogeneous cancer cells , some of which expresses Ret and Gfras , while some express only Ret ( He et al . , 2014 ) . In an attempt to show that GFRa1 is normally released from wild-type DRG cells for trans RET signaling , we found the same for GFRa2 ( Figure 7 ) . Given that soluble GFRa2 could also activate RET in trans with NRTN ( Worley et al . , 2000 ) , our finding raises many interesting questions , such as whether all GFRas are secreted and whether ‘cis’ and ‘trans’ RET signaling normally co-exist even when RET and GFRas are expressed in the same cells . It seems plausible that even for GFRas co-expressed with RET ( usually defined as ‘cis’ signaling ) , such as GFRa2 in RA mechanoreceptors , it could be secreted and then upon NRTN binding activates RET in the cell from which it was released ( ‘trans’ activation ) . Although cis and trans activation of RET lead to a similar biological outcome in the growth and survival of RA mechanoreceptors , it is worth noting that substantial differences likely exist between the signaling processes of cis and trans RET activation . Cis RET activation might be more efficient , given that GFRas and RET are located in the same membrane . In addition , cis and trans signaling could differ in the kinetics of recruitment of RET to lipid rafts upon GFLs stimulation , interactions with downstream-associated proteins , and the longevity of activated RET and downstream effectors ( Tansey et al . , 2000; Paratcha et al . , 2001 ) . Additionally , it remains possible that although the gross structure of the RA mechanosensory central projections seems mostly normal in mice lacking either cis or trans signaling , more precise aspects of circuit formation , such as specific synapse formation , which are beyond the resolution of current analysis , may differentially depend on cis or trans signaling . Finally , it will be interesting to see whether cis and trans signaling can produce similar biological outcomes in other systems , as we have shown here for RA mechanoreceptors . Future experiments to carefully dissect cis and trans RET signaling in other types of cells and tissues will address these issues .
Mice except GDNFlacZ line were raised in a barrier facility in Hill Pavilion , the University of Pennsylvania . All procedures were conducted according to animal protocols approved by Institutional Animal Care and Use Committee ( IACUC ) of the University of Pennsylvania and National Institutes of Health guidelines . GDNFlacZ mice were raised in accordance with the European Community Council Directive of November 24 , 1986 ( 86/609/EEC ) , and approved by the ethics . Most mice used in this paper were described previously: RetCreERT , RetCFP , Nrtn+/− ( purchased from the Jackson lab ) , Gfra2GFP ( re-derived using sperm provided by Dr Jeffery Milbrandt at the Washington University ) , Rosa26Tdt , and GdnfLacZ mice ( Moore et al . , 1996; Heuckeroth et al . , 1999; McDonagh et al . , 2007; Uesaka et al . , 2008; Luo et al . , 2009; Madisen et al . , 2010 ) . The Ntrk1− allele was generated by crossing the floxed TrkAF592A allele ( Chen et al . , 2005 ) to germline Cre mice . The generation of Gfra1 conditional and null mice and the RetCreERT;RosaTdt tandem allele are described below . All mice except GdnfLacZ were maintained on a mixed C57BL/6J and CD1 background . GdnfLacZ mice were maintained on a C57BL/6N background . Except for Gfra1;Gfra2 double null animals ( n = 2 ) , at least three animals per genotype were examined . N-values for explants are listed in Figure 7—source data 1 , 2 . We generated Gfra1 conditional knockout mice , in which loxP sites flank exon 6 of Gfra1 , by homologous recombination . Mice harboring the floxed allele were crossed to germ line Cre mice , resulting in a Gfra1 allele lacking exon 6 . The loss of Gfra1 transcript in Gfra1−/− mice was confirmed by in situ hybridization of mutant and control DRGs ( see Figure 1—figure supplement 3 ) . Since Ret and Rosa26 loci are located only ∼5 megabases apart on mouse chromosome 6 , we generated a tandem configuration of RetCreERT and RosaTdt alleles , which are linked during meiosis and became a great genetic advantage for our experiments ( Figure 3—figure supplement 1 ) . We used this tandem RetCreERT;RosaTdt allele to specifically label RA mechanoreceptors in different mutant mouse lines described in the text . We set up timed pregnancy mating for mice in the evening and checked mice for vaginal plugs the following morning . The time when a female mouse was found to have a plug was counted as embryonic day 0 . 5 ( E0 . 5 ) . We treated embryos harboring the RetCreERT;RosaTdt reporter allele with 4-hydroxy-tamoxifen ( 4-HT , 2 mg and 1 mg at E11 . 5 , and E12 . 5 , respectively ) by oral gavage to pregnant female mice to specifically label RA mechanoreceptor population . Spinal columns of embryos and neonatal mice at the desired developmental stages were dissected out and directly immersed in PBS/4% paraformaldehyde ( PFA ) for 2 to 4 hr at 4°C . Postnatal mice were sacrificed with CO2 , transcardially perfused with 4% PFA , and spinal columns were dissected out and post-fixed with 4% PFA for 2 hr at 4°C . They were then cryo-protected in 1× PBS , 30% sucrose overnight . 20-µm frozen sections of SC and DRGs were cut using a CM1950 cryostat ( Leica , Buffalo Grove , IL ) . Immunostaining of SCs and DRG sections were performed as described previously ( Niu et al . , 2013 ) . Antibodies used are as follows: rabbit anti-GFP ( A-11122 , 1:2000 , Invitrogen , Carlsbad , CA ) , chicken anti-GFP ( GFP-1020 , 1:1000 , Aves , Tigard , OR ) , chicken anti-NF200 ( NF-H , 1:500 , Aves ) , rabbit anti-NF200 ( N4142 , 1:1000 , Sigma , St . Louis , MO ) , rabbit anti-cRet ( 18121 , 1:50 , IBL , Minneapolis , MN ) , rabbit anti-NTRK1 ( 06-574 , 1:1000 , Millipore , Temecula , CA ) , guinea pig anti-VGLUT1 ( AB5905 , 1:1000 , Millipore ) , rabbit anti-phospho-S6 ( 2215s , 1:200 , Cell Signaling , Beverly , MA ) , and Alexa Fluorescent conjugated Goat or Donkey secondary antibodies ( 1:500 , Invitrogen or Jackson ImmunoResearch , West Grove , PA ) . Embryos of the desired age were eviscerated and fixed in 1% PFA , 2 mM MgCl2 , 5 mM EGTA , 0 . 02% NP40 , and 0 . 2% glutaraldehyde in phosphate buffer ( pH 7 . 4 ) for 1 . 5 to 2 hr at 4°C . Vibratome sections were incubated for 30 min in washing solution ( 2 mM MgCl2 , 0 . 02% NP-40 in phosphate buffer pH 7 . 4 ) . LacZ reaction was developed with X-gal ( 1 mg/ml ) at 37°C . DIG- or FITC-labeled riboprobes were synthesized using a DIG or FITC RNA labeling kit ( 11175025910 , Roche , Indianapolis , IN ) . Template for GFP was amplified by PCR and subcloned into vector pGEM-T Easy ( A1360 , Promega , Madison , WI ) . Antisense RNA probes for Ret , Gfra1 , Gfra2 , Gdnf , and Nrtn were generated as previously described ( Luo et al . , 2009 ) . The detailed procedures of in situ hybridization and double fluorescent in situ hybridization were performed as described previously ( Fleming et al . , 2012 ) . DRGs from E13 . 5 , E15 . 5 , and E18 . 5 Gfra2GFP/+ and Gfra2GFP/GFP embryos were dissected and rapidly frozen on dry ice . RNA was extracted with the GeneJet RNA Purification Kit ( K0731 , Fermentas , Vilnius , Lithuania ) and cDNAs were generated using Super-Script III First-Strand Synthesis System ( 18080-51 , Invitrogen ) . 500 ng of total RNA was used for each RT reaction in a total volume of 25 μl . QPCR reactions were performed in triplicate for three samples of each age and genotype . QPCR reactions contained SYBR Green PCR master mix ( 4309155 , Life Technologies , Carlsbad , CA ) , 0 . 5 μM of each primer , and 3 μl ( for Gfra1 ) or 1 μl ( for Gapdh ) of cDNA template per 15 μl reaction . Reactions were run and analyzed on a StepOnePlus Real-Time PCR System ( Applied Biosystems , Carlsbad , CA ) . Primers used were Gapdh ( 5′-CCACCAACTGCTTAGCCCCC-3′ and 5′-GCAGTGATGGCATGGACTGTGG-3′ ) and Gfra1 ( 5′-TGTCTTTCTGATAATGATTACGGA-3′ and 5′-CTACGATGTTTCTGCCAATGATA-3′ ) . p-values between samples were calculated from ΔCT values with the Student's t-test , and relative concentrations were calculated by the 2−ΔΔCT method ( Livak and Schmittgen , 2001 ) . E14 . 5 embryos were removed from the dam and placed in F-12 media ( 11765-047 , Invitrogen ) on ice . SCs with attached DRGs were dissected from the spinal column , and individual DRGs were removed and placed in fresh F-12 on ice . Using a dissecting needle , DRGs were cleaned and bisected , and then placed in fresh F-12 . Explants were grown on Superfrost Plus slides ( 22-034-979 , Fisher , Waltham , MA ) coated with poly-L-lysine ( P1274 , Sigma , 0 . 1 mg/ml in ddH2O overnight at 4°C ) and laminin ( 354232 , BD , Franklin Lakes , NJ ) , 20 μg/ml in HBSS [14170122 , Invitrogen] at 37°C for one to 3 hr ) . Immediately before placing explants on the slide , slides were washed with HBSS and DRG culture medium ( Neurobasal medium [21103-049 , Invitrogen] , 1× B27 [17504-044 , Invitrogen] , 100 U/ml penicillin/streptomycin [15140-122 , Invitrogen] , 2 mM L-glutamine [25030-081 , Invitrogen] , and 35 mM glucose ) . DRG culture media supplemented with the appropriate recombinant proteins ( 50 ng/ml Nrtn [477-MN-025 , R&D , Minneapolis , MN] , 100 ng/ml GDNF [512-GF-010 , R&D] , 300 ng/ml GFRa1 [560-GF-100 , R&D] , or 100 ng/ml GDNF and 300 ng/ml GFRa1 ) were added to the culture dish . Four to six DRG explants were placed on each slide and the culture dishes were carefully moved to a 37°C incubator and left undisturbed overnight . Following 16–24 hr of incubation , cultures were rinsed with PBS and fixed in 4% PFA in PBS for 30 min at room temperature . Immunofluorescence was then performed directly in the culture dish using antibody dilutions described above . Following secondary antibody , explant slides were mounted on microscope slides using Superglue , and coverslipped with Fluoromount-G ( 0100-01 , Southern Biotech , Birmingham , AL ) and 22 × 22 mm coverglass . DRGs from E18 . 5-P1 mice were collected into HBSS on ice . DRGs were first digested in 0 . 5 mg/ml collagenase ( LS4186 , Worthington , Lakewood , NJ ) plus 100 U/ml penicillin/streptomycin , 10 mM HEPES , and 1× MEM vitamins ( M6895 , Sigma ) in MEM ( 11095072 , Invitrogen ) at 37°C for 30 min , followed by a second digestion with 0 . 05% trypsin ( 25200056 , Invitrogen ) plus 100 U/ml penicillin/streptomycin , 10 mM HEPES , and 1× MEM vitamins in MEM at 37°C for 30 min . Digestion was stopped by adding 5% FBS and 10 mM HEPES in HBSS . Cells were then triturated with a fire polished Pasteur pipette to a homogenous solution . The cells were then pelleted at 500×g for 5 min and resuspended in DRG culture media , as described above , supplemented with 50 ng/ml NRTN , 100 ng/ml GDNF , and 50 ng/ml NGF ( 556-NG-100 , R&D ) . Cells were plated in six-well collagen-coated plates ( Millipore , PICL06P05 ) and cultured at 37°C and 5% CO2 . After 2 days , media were removed and cells were rinsed with warmed Neurobasal media . 2 ml of fresh DRG culture media supplemented with NRTN , GDNF , and NGF ( but without B27 ) was added to each well . After 2 days , media were removed and saved at 4°C with added protease inhibitors ( P8340 , Sigma ) . Fresh media supplemented with growth factors but without B27 were then added to each well . After an additional 2 days , media were removed and pooled with previously collected media , and additional protease inhibitor was added . The cells were rinsed twice with PBS , and then lysed directly in the well by the addition of 70 μl 2× sample buffer ( 0 . 125 M Tris pH 6 . 8 , 20% glycerol , 4% SDS , 0 . 16% bromophenol blue , 10% 2-mercaptoethanol ) and scraping , followed by heating at 95°C for 5 min . All cell lysates were then brought to a total volume of 140 μl with 1× PBS . Supernatants were centrifuged at 14 , 000×g for 15 min to clear cellular debris , and then were concentrated to ∼30 μl with Amicon 30 kDa filters ( UFC503024 , Millipore ) , then mixed with an equal volume of 2× sample buffer and heated at 95°C for 5 min . Duplicate 4–15% gradient mini-Protean TGX gels ( 456-1084 , Biorad , Hercules , CA ) were used to run samples . 40 μl of cell lysate of each genotype or one third of the total volume of concentrated supernatant of each genotype was used . Both gels were then transferred to nitrocellulose membrane and blocked in 3% BSA in TBS plus 0 . 1% Tween-20 ( TBST ) for 1 hr at room temperature . Membranes were then incubated overnight with either goat anti-GFRa1 ( 0 . 2 μg/ml , AF560 , R&D ) or goat anti-GFRa2 ( 0 . 2 μg/ml , AF613 , R&D ) in blocking solution overnight at 4°C . Following washes with TBST , membranes were incubated with donkey anti-goat-AP ( SC-2022 , 1:5000 , Santa Cruz Biotechnology , Santa Cruz , CA ) in blocking solution for 1 hr at room temperature . After washes , AP was detected with CDP-Star ( T2218 , Applied Biosystems ) and membranes were imaged with a Chemi-Doc system ( BioRad ) . Following imaging , membranes were stripped with 2× 10 min stripping buffer ( 0 . 2 M glycine , 0 . 1% SDS , 1% Tween-20 , pH 2 . 2 ) , followed by 2× 10 min wash with PBS and 2× 5 min wash with TBST . Membrane was then probed with rabbit anti-β-actin ( sc-130656 , 1:400 , Santa Cruz Biotechnology ) and goat anti-rabbit-AP ( T2191 , 1:5000 , Applied Biosystems ) following the above procedure , except that all blocking and antibody incubations were performed in 5% milk in TBST . Western blot densitometry was performed with ImageJ . Three cultures per genotype were analyzed . Densitometry measurements for each antibody were performed on three blots running independent culture samples . Relative quantifications were performed using β-actin in the cell lysates as a measure of total protein per lane , and optical density values for GFRa1 were scaled accordingly . Because an equal proportion of total lysates was run in each lane , total β-actin per cell lysate lane was used as a proxy for cell number , and was therefore used to normalize protein levels in the supernatant lanes ( equal proportion of total supernatant volume were run in each lane ) . Cell lysate and supernatant samples were scaled to wild-type quantifications of respective sample type and reported in arbitrary units . Student's t-test was used to measure significance of differences between genotypes . Fluorescent images of SC/DRG sections were acquired on a Leica SP5II confocal microscope . DRG explant cultures were imaged on Leica DM5000B microscope . Bright field images were taken using Leica DM5000B microscope . For histological analysis , at least six sections per specified spinal/DRG level per animal were analyzed . For quantification of genetically labeled neuron number in E18 . 5 embryos , whole-mount L4/L5 DRGs were imaged and total Tdt+ cell number was counted in each DRG . Except for Gfra1;Gfra2 double null animals ( n = 2 ) , at least three animals per genotype were examined . N-values for explants are listed in Figure 7—source data 1 and 2 . Pixel counts for central projections were generated by counting the number of pixels at each intensity level ( 0–256 ) in an outlined immunoreactive area in ImageJ . Background staining was subtracted by counting pixel number of each intensity level in a non-immunoreactive region of the tissue . The minimal intensity level at which two consecutive levels displayed a pixel count of zero was taken as the threshold cut of background fluorescence . Pixel counts of real staining were then calculated by summing the pixel counts for all intensity levels above the defined background level . Column graphs were generated in GraphPad Prism 5 . All error bars are ± standard error of the mean ( SEM ) , unless otherwise specified . All statistical analyses were performed using SAS version 9 . 3 ( SAS Inc . , Cary , NC ) . Due to differences in labeling efficiency across litters in 4-HT treated mice , quantification for SC section staining and whole mount DRGs were performed by normalizing to controls within the same litter . For all explant quantifications , GFP+ neuron number per 10 , 000 μm2 was calculated for each explant . For RetCFP explants , a circle with a radius 200 μm larger than the explant was drawn around the explant in ImageJ , and the number of CFP+ axons which crossed the circle was counted . | During development , cells send and receive numerous signaling molecules . In order to trigger a biological response , such signaling molecules must first bind to a specific receptor protein , often located on the cell surface . These receptor proteins can either work alone or with partner proteins called co-receptors . When the co-receptor is produced by the same cell as the receptor , it is called cis signaling . When the co-receptor is produced by other cells , it is called trans signaling . RET is one such receptor that is important for the development of the nervous system and many other biological processes . It interacts with a particular family of signaling molecules , the glial cell line-derived neurotrophic factor ( GDNF ) family ligands , which first bind to a co-receptor , GFRα , before binding to RET . These co-receptors can come from the same cell as RET , or from a different cell . Previous studies have indicated that RET can receive both cis and trans signals using cultured cells , but it was not clear whether both types of signal occur during normal development and contribute to the same biological processes . Fleming , Vysochan et al . investigated this question by analyzing the roles of RET signaling in a type of mouse neuron that is involved in sensing touch . RET is important for the survival and development of these neurons , which express both RET and its co-receptor GFRa2 . Another RET co-receptor , GFRa1 , is produced by other cells that are next to the cell bodies and projections of these touch-sensing neurons . To investigate the roles of different GFRa co-receptors further , Fleming , Vysochan et al . generated a variety of mouse mutants , including mice with mutations in one or both types of co-receptor . The neurons in mice lacking both co-receptors shared the same defects as the neurons in the mice lacking RET . Loss of either co-receptor alone did not produce these abnormalities . This indicates that both co-receptors can mediate the normal development of these neurons , with GFRa2 signaling in cis and GFRa1 signaling in trans . Fleming , Vysochan et al . propose that cis and trans RET signaling can lead to the same biological outcomes in these neurons . Future experiments should reveal if cis and trans RET signaling contribute towards common biological processes in other cell types inside the body as well . Such findings might also be important for understanding the role of RET signaling in cancer and other human diseases . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"developmental",
"biology",
"neuroscience"
] | 2015 | Cis and trans RET signaling control the survival and central projection growth of rapidly adapting mechanoreceptors |
Electron microscopy-based connectomics aims to comprehensively map synaptic connections in neural tissue . However , current approaches are limited in their capacity to directly assign molecular identities to neurons . Here , we use serial multiplex immunogold labeling ( siGOLD ) and serial-section transmission electron microscopy ( ssTEM ) to identify multiple peptidergic neurons in a connectome . The high immunogenicity of neuropeptides and their broad distribution along axons , allowed us to identify distinct neurons by immunolabeling small subsets of sections within larger series . We demonstrate the scalability of siGOLD by using 11 neuropeptide antibodies on a full-body larval ssTEM dataset of the annelid Platynereis . We also reconstruct a peptidergic circuitry comprising the sensory nuchal organs , found by siGOLD to express pigment-dispersing factor , a circadian neuropeptide . Our approach enables the direct overlaying of chemical neuromodulatory maps onto synaptic connectomic maps in the study of nervous systems .
A comprehensive understanding of nervous system function requires knowledge of not only precise neuronal connectivity but also the unique combinations of molecules expressed by each neuron . Connectomics using serial-sectioning electron microscopy ( EM ) aims to map the synapse-level connectivity of entire neural circuits ( Morgan and Lichtman , 2013 ) . However , wiring diagrams provide an incomplete picture of the nervous system , since they lack information about the molecular nature of neurons . One important class of molecules are neuromodulators , including monoamines and neuropeptides , that actively shape the output of circuits by modifying synaptic function and neuron excitability ( Bargmann , 2012; Bargmann and Marder , 2013; Bucher and Marder , 2013; Marder , 2012 ) . The direct mapping of specific neuromodulators to their source cells in synapse-level anatomical maps would help to integrate connectomic and neuromodulatory perspectives , thereby enriching our understanding of circuit function . Several approaches have been developed to assign molecular identities to neurons on EM sections . Genetically encoded tags with enzymatic activity , such as miniSOG and APEX probes , allow the resolution of cells or subcellular structures by EM following incubation with a substrate that is converted to electron-dense deposits ( Martell et al . , 2012; Shu et al . , 2011 ) . Fixation-resistant protein tags compatible with EM procedures and observable by correlative light and electron microscopy ( CLEM ) or directly by immunogold labeling ( immunoEM ) , such as the smGFPs ( Viswanathan et al . , 2015 ) , have also been developed . In CLEM , fluorescently labeled neurons or organelles can be imaged live prior to fixation , sectioning , and EM imaging of the specimen . Fiducial markers allow the correlation of cellular structures in the immunofluorescence ( IF ) and EM images ( Colombelli et al . , 2008; Maco et al . , 2013; Maco et al . , 2014; Urwyler et al . , 2015 ) . However , the neuron-specific probes employed by these technologies must be expressed using transgenesis , limiting their use to one or a few markers , such as in the ‘two-tag’ labeling approach ( Lin et al . , 2015 ) . An alternative approach to assigning molecular identities to neurons is the use of endogenous targets in fixed specimen , as in array tomography , a method employing sequential IF with several different antibodies on large series of sections ( Collman et al . , 2015; Micheva and Smith , 2007 ) . IF imaging can be combined with scanning EM imaging of selected sections to resolve ultrastructural detail . The sequentially acquired IF and EM images can be registered , allowing for correlative analysis of images of different modalities ( Collman et al . , 2015 ) . However , the sequential IF procedure compromises the membrane contrast in EM , preventing the reliable tracing of fine processes across series of sections ( Collman et al . , 2015 ) . Array tomography also relies on embedding the specimen in porous , hydrophilic acrylic resins to allow optimal immunolabeling . For large-scale serial sectioning projects , epoxy resins , such as Epon , are favored due to their higher stability during sectioning and imaging and the optimal ultrastructural preservation they provide ( Bock et al . , 2011; Briggman et al . , 2011; Bumbarger et al . , 2013; Ohyama et al . , 2015; Randel et al . , 2015; White et al . , 1986 ) . Unfortunately , the Epon-embedding procedure , including osmium-fixation and resin polymerization at 60°C , compromises the immunogenicity of many endogenous targets ( Brorson , 1998; Brorson and Reinholt , 2008; De Paul et al . , 2012 ) . Current immunolabeling approaches therefore sacrifice the mechanical stability and ultrastructural contrast of the sample for optimal immunolabeling . One exception is short amidated neuropeptide antigens that show good immunopreservation in epoxy-embedded samples ( Hamanaka et al . , 2010; Koizumi et al . , 1989; Merighi et al . , 1992; Yasuyama and Meinertzhagen , 2010 ) . Such antigens represent promising targets in an attempt to combine connectomics on Epon-embedded samples and immunogold labeling for specific neuromodulators . Neuropeptides are often chemically modified , including C-terminal amidation ( Eipper et al . , 1992 ) , which confers stability and high immunogenicity to even very short peptides ( Conzelmann and Jékely , 2012 ) . Furthermore , neuropeptides show neuron-type specific expression and are distributed throughout the axon of peptidergic neurons ( Wong et al . , 2012; Zupanc , 1996 ) , representing useful markers for neuron-type identification . Here , we introduce serial-multiplex immunogold ( siGOLD ) , a method for immunolabeling connectomes . siGOLD involves immunogold labeling of small subsets of sections from large series with different neuron-type-specific neuropeptide antibodies . The molecularly identified neurons and their synaptic partners are then reconstructed from the entire aligned series using standard EM-based connectomics . We established siGOLD using larval stages of Platynereis dumerilii , a marine annelid that has recently emerged as a powerful model for circuit neuroscience , genetics , and whole-body connectomics ( Backfisch et al . , 2013; Bannister et al . , 2014; Gühmann et al . , 2015; Randel et al . , 2014; Randel et al . , 2015; Tosches et al . , 2014; Veedin-Rajan et al . , 2013; Zantke et al . , 2014 ) . We identified several amidated neuropeptide epitopes that showed long-term immunopreservation in Epon-embedded samples , allowing us to simultaneously obtain high ultrastructural detail and specific immunogold signal . Using siGOLD with 11 distinct antibodies on the same specimen , we identified several neuropeptide-containing neuron profiles in a whole-body Platynereis larval serial EM dataset ( Randel et al . , 2015 ) . Furthermore , taking advantage of the whole-body series , we fully reconstructed several peptidergic neurons identified by siGOLD in the Platynereis larva . We also identified and reconstructed the postsynaptic partners of selected peptidergic neurons , focusing on the nuchal organs , paired , putatively chemosensory organs with high structural complexity and variability among the annelids ( Purschke , 1997; Purschke , 2005; Purschke et al . , 1997; Schlötzer-Schrehardt , 1987 ) . Our work demonstrates that siGOLD can be used in large serial EM datasets to assign molecular identities to multiple neurons using different markers and to fully reconstruct and analyze the synaptic connectivity of these neurons at EM resolution .
In order to selectively label individual neurons in large-scale serial EM datasets , we established an immunoEM procedure to label ultrathin sections with neuronal cell-type specific antibodies . We reasoned that immunoEM performed on only a few layers from a large series of sections could identify neuron profiles that contain the antigen ( Figure 1A ) . We first performed immunoEM on 40-nm serial sections from the ventral nerve cord ( VNC ) of a 72 hr post-fertilization ( hpf ) Platynereis larva ( specimen HT9-5 , Figure 1B , C ) . For specimen preparation , we used a conventional serial TEM protocol including high-pressure freezing , fixation with a freeze substitution medium containing 2% osmium tetroxide and 0 . 5% uranyl acetate , and embedding in Epon . We also developed a procedure for the safe handling of several grids in parallel during the immunostaining and contrasting procedure . We optimized the immunolabeling protocol to achieve high specificity for immunoEM and high ultrastructural detail . In our protocol , we use secondary antibodies coupled to ultra small gold particles and a silver-enhancement procedure . We also fine-tuned the contrast-staining protocol to optimize contrast for both gold labeling and ultrastructural detail . 10 . 7554/eLife . 11147 . 003Figure 1 . Development of the siGOLD method . ( A ) Schematic flowchart of the siGOLD labeling approach from high-pressure freezing and freeze substitution ( HPF-FS ) to tracing and 3D reconstruction . Ni , nickel grid , Cu , copper grid . ( B ) SEM micrograph of a 72 hpf Platynereis larva . ( C ) Schematic of the HT9-5 sample showing the position of the ventral nerve cord ( VNC ) , ventral view . Colored lines indicate where cross-sections through the VNC were taken , near the base of the circumesophageal connectives at the level of the first commissure . Layer number ( s ) followed by neuropeptide ID are indicated for each colored line . Dashed line indicates the gap ( approximately 10 missing sections ) between the first and second series of sections . Scale bar: ( B ) 50 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 11147 . 003 In preliminary tests , we found strong and localized labeling in neurites using 11 different polyclonal antibodies generated against short amidated neuropeptides of Platynereis ( Table 1 ) . 10 . 7554/eLife . 11147 . 004Table 1 . List of antibodies usedDOI: http://dx . doi . org/10 . 7554/eLife . 11147 . 004NP precursor nameAbbreviationAntigenFMRFamideFMRFa ( C ) FMRFaRYamideRYa ( C ) VFRYaMyoinhibitory peptide/Allatostatin BMIP ( C ) AWNKNSMRVWa or ( C ) VWaRGWamideRGWa ( C ) RGWa or ( C ) GWaProenkephalinPENK ( C ) YGDLSFSNSNYaLuqinLUQ ( C ) WRPQGRFaAllatotropinATO ( C ) GFRTGAYDRFSHGFaPigment dispersing factorPDF ( C ) NPGTLDAVLDMPDLMSLaLeucokininLEUC ( C ) KFTPWAaFVamideFVa ( C ) AHRFVa or ( C ) FVaFVRIamideFVRIa ( C ) FVRIaThe full name and abbreviation of neuropeptide precursors that contain the neuropeptides used for immunoEM . The FMRFa , RYa , MIP short and long , RGWa , FVa short and long , and FVRIa antibodies have been described previously ( Conzelmann and Jékely , 2012; Conzelmann et al . , 2011; Conzelmann et al . , 2013a; Jékely et al . , 2008 ) . All 11 neuropeptides are amidated ( a ) . A Cys ( C ) was added to the N-terminus of each peptide to allow coupling during immunization and affinity purification . All antibodies were generated in rabbits . For FVa long , a rat antibody was also generated . To test the specificity and reproducibility of immunoEM with the 11 neuropeptide-antibodies , we collected two sets of transverse serial sections from the first trunk segment of the HT9-5 specimen . Four to 18 consecutive serial sections were collected on each EM grid for immunoEM . For each antibody , we stained two grids separated by a serial distance of approximately 50 sections ( Figure 1C ) . We imaged the VNC region in each section at a resolution of 2 . 22 nm/pixel followed by stitching and alignment of the images . We found strong and localized labeling in only a small subset of neurites for each antibody ( Figure 2A , Shahidi , et al . , 2015 ) . In consecutive sections , the same neurite was often strongly labeled with the same antibody ( Figure 2A , FVa , layers 1–5 ) . In sections collected on different grids that were labeled with different antibodies , we found distinct patterns of neurite-specific labeling ( Figure 2B , C ) . In many sections , we could observe dense core vesicles ( DCVs ) in the cytoplasm of the gold-labeled neurites , indicative of the peptidergic nature of these cells ( Figure 2A–F , Shahidi , et al . , 2015 ) . In high-resolution ( 0 . 22 nm/pixel ) images , we could observe gold labeling associated with DCVs ( Figure 2D–F ) , suggesting that our immunoEM procedure labeled mature neuropeptides residing inside these vesicles . Several ultra-small gold particles were not enlarged during the silver-enhancement procedure but were also associated with DCVs ( Figure 2D–F ) . 10 . 7554/eLife . 11147 . 005Figure 2 . Immunolabeling with neuropeptide antibodes on Epon sections . ( A ) Representative micrographs with immunogold labeled axons for the neuropeptide antibodies indicated . For the FVa neuropeptide antibody , five adjacent sections are shown ( FVa Layer 1–5 ) . ( B , C ) Neurite-specific labeling in adjacent sections ( seven sections apart ) labeled with different antibodies . ( D–F ) High-resolution micrographs of immunogold labeled , silver-enhanced gold particles ( arrows ) , and unenhanced ultra small gold particles ( arrowheads ) Scale bar: ( A-C ) 1 μm; ( D–F ) 200 nm . High-resolution images are available in ( Shahidi , et al . , 2015 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 11147 . 005 Next , we comprehensively scored the distribution of gold particles in the VNC series ( Figure 3 ) . We selected all neurites that were labeled with two or more gold particles in any of the immunoEM sections and traced these neurites across all sections . We also selected 50 control neurites along a coronal transect spanning the VNC and traced these across all sections ( Figure 3A ) . We then counted all gold particles in each traced neurite and tabulated the results ( Figure 3—source data 1 ) . The gold particles were also summed for each antibody to show a concise summary of all gold labels ( Figure 3B and D ) . Since not all ultra-small gold particles were enhanced during silver-enhancement ( Figure 2D–F ) and non-enhanced particles are not visible on the lower resolution images we used for scoring , the gold counts likely underestimate the intensity of gold labeling . 10 . 7554/eLife . 11147 . 006Figure 3 . Identification of peptidergic neurites by siGOLD in a 72 hpf Platynereis specimen ( HT9-5 ) . ( A , C ) Anterior view of EM cross-section through the VNC near the first commissure . Dorsal side of larva is to the top . Strongly labeled neurites were analyzed across the whole VNC region . Control axon profiles were analyzed along a transect ( dotted line ) , two axon profiles were sampled every 1 μm . Total of 72 and 63 axons were examined for the first and second series of sections respectively ( an approximately 10-section gap occurs between the two series ) . Colored cell profiles indicate gold labeled neuropeptidergic axons . Different shades of a color represent an approximation of labeling intensity . Positive axons are tagged with neuron number , neuropeptide name , and total number of gold particles per total number of layers for that neuropeptide . ( B , D ) Tables show number of gold particles per axon for each neuropeptide . All strongly labeled axons across the VNC and control axons along the sampled transect are shown . Data were arranged according to the spatial distribution of the corresponding neurites in the VNC . Each sampled axon was traced across all layers and counted for its total number of gold particles . Columns indicate neurons and an ID with ‘n’ is given to each neuron in the first series of sections and ‘N’ for neurons in the second series of sections . Rows indicate neuropeptide immunogold labels . Different shades of the same color indicate intensity of gold labeling . Totals are shown for each row ( neuropeptide ) and each column ( axon profile ) . In the final column of each table , the total number of layers stained for each neuropeptide is shown . Scale bar: 5 μm . Gold scores are available in Figure 3—source data 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 11147 . 00610 . 7554/eLife . 11147 . 007Figure 3—source data 1 . Tabulated counts of all gold particles in each traced neurite in HT9-5 . ( Sheet 1 ) Detailed gold counts throughout all layers analyzed from the HT9-5 specimen . ( Sheet 2 ) Tabulated data shown in Figure 3B , D . ( Sheet 3 ) DCV counts and gold counts in selected neurites . DOI: http://dx . doi . org/10 . 7554/eLife . 11147 . 00710 . 7554/eLife . 11147 . 008Figure 3—figure supplement 1 . Number of DCVs in selected neurite profiles along 100 layers . Number of DCVs scored along 100 layers in selected neurites in the HT9-5 dataset . The gold counts for the corresponding profiles are shown in Figure 3—source data 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 11147 . 008 Our quantifications revealed a high neurite-specificity of immunogold labeling with all 11 neuropeptide antibodies ( Figure 3 ) . In sections where we omitted the primary antibody , we did not see any gold labeling in any of the traced neurites ( Figure 3B ) . The VNC in our sample contained approximately 1600 neurite cross-sections and the majority of these showed no gold labeling . However , for each antibody we could identify a small number of neurites that consistently showed strong labeling across different sections ( Figure 3 , Figure 3—source data 1 ) . The pattern of the labeled neurites in the VNC showed bilateral symmetry , supporting the labeling of specific peptidergic neuron populations on the left and right sides of the body . Most neurites that were labeled on one grid were also strongly labeled on their grid pair with the same antibody separated by approximately 50 sections . There were some exceptions , for example neurite n5 that was labeled strongly with the FVamide antibody in layers 18–23 but not in layers 69–74 ( Figure 3—source data 1 ) . To determine whether the lack of labeling was due to the lack of DCVs in these sections , we scored the number of DCVs across all sections for selected neurites ( Figure 3—figure supplement 1 ) . We found that DCVs were non-uniformly distributed in the neurites and some sections completely lacked DCVs . Importantly , we only detected gold labeling on sections that contained DCVs in the respective neurite ( Figure 3—source data 1 ) . siGOLD also allowed us to detect the coexpression of some neuropeptides in the same neurons . For example , we found coexpression of FVa and PDF neuropeptides in a subset of the neurites labeled by these two antibodies ( Figure 3D ) . Some antibodies showed extensive overlap in labeling , including the FMRFa , luqin and RYa antibodies . To test whether the antibodies specifically recognize the neuropeptides used for immunization , we performed morpholino-mediated knockdown experiments of proneuropeptide expression . The specificity of the MIP antibody was demonstrated previously ( Williams et al . , 2015 ) . We performed microinjections with 10 different translation-blocking morpholinos , one targeting each remaining proneuropeptide . We then performed triple IF in whole-mount 72 hpf larval samples using an acetylated tubulin antibody , a rat FVa antibody , and the respective rabbit neuropeptide antibodies . The acetylated tubulin antibody allowed us to exclude developmental abnormalities caused by morpholino injection . The rat FVa antibody provided a further control to exclude that morpholino injection affected proneuropeptide processing in general . For 8 out of 10 antibodies ( FVa , FMRFa , ATO , FVRIa , PDF , RGWa , PENK , LEUC ) , morpholino injection strongly reduced IF signal with the respective antibody ( Figure 4—figure supplement 1 ) . For the FMRFa antibody , we did not detect staining in the VNC following morpholino injection but we could still detect staining in the head . The FMRFa antibody therefore likely recognizes other antigens , such as other RFa peptides in the head ( Conzelmann et al . , 2013b ) . For the RYa and LUQ antibodies , we did not see a reduction in staining intensity in the VNC following morpholino injection . These two antibodies may cross-react with other R[Y F]amide peptides as suggested by the labeling of an overlapping set of neurites with the FMRFa , LUQ , and RYa antibodies ( Figure 3 ) . Without knowing the exact antigen specificity for these antibodies , we can nevertheless use them as specific markers of R[Y|F]amidergic neurons . siGOLD labeling of consecutive sections with 11 antibodies revealed the arrangement of several peptidergic neurites in the VNC ( Figure 3 ) . To test if this spatial arrangement is consistent between siGOLD and IF labeling of whole specimens , we analyzed the labeling with all 11 antibodies in whole-mount Platynereis 72 hpf larval samples . All antibodies labeled subsets of longitudinal axons spanning the entire length of the VNC and occurring in different mediolateral positions ( Figure 4 ) . 10 . 7554/eLife . 11147 . 009Figure 4 . Whole-mount IF of Platynereis larvae with antibodies raised against neuropeptides labels distinct subsets of neuronal tracks in the VNC . ( A ) Ventral view of Platynereis VNC , stained with an anti-acetylated tubulin antibody . This is the registered average VNC generated from scans of 36 larvae . ( B–L ) Whole-mount IF of Platynereis larvae with an antibody raised against ( B ) FVa , ( C ) RGWa , ( D ) FVRIa , ( E ) RYa , ( F ) PDF , ( G ) LEUC , ( H ) LUQ , ( I ) MIP , ( J ) ATO , ( K ) PENK , ( L ) FMRFa . Whole-mount scans were cropped to show only the VNC region . Scale bar: 20 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 11147 . 00910 . 7554/eLife . 11147 . 010Figure 4—figure supplement 1 . Morpholino-mediated knockdown of proneuropeptides followed by whole-mount IF indicates antibody specificities . Whole-mount IF of 72 hpf uninjected control Platynereis larvae and larvae micro-injected with translation-blocking morpholinos ( MOs ) targeting different neuropeptide precursor genes . FVa-knockdown and control larvae were stained with an antibody raised against FVa and counterstained with acetylated tubulin . All other control and knockdown larvae were co-stained with an antibody raised against FVa and an antibody raised against the MO target peptide indicated and counterstained with acetylated tubulin . All larvae are shown in ventral view . For each neuropeptide , three control larvae ( top ) and three MO-knockdown larvae ( bottom ) are shown . ( A ) FVa knockdown and control larvae with antibody raised against FVa . ( B ) FMRFa knockdown and control larvae with antibody raised against FMRFa . ( C ) LUQ knockdown and control larvae with antibody raised against LUQ . ( D ) ATO knockdown and control larvae with antibody raised against ATO . ( E ) FVRIa knockdown and control larvae with antibody raised against FVRIa . ( F ) PDF knockdown and control larvae with antibody raised against PDF . ( G ) RGWa knockdown and control larvae with antibody raised against RGWa . Note that the staining in the head is eliminated following knockdown . The staining in the VNC was too weak in these samples . ( H ) RYa knockdown and control larvae with antibody raised against RYa . ( I ) PENK knockdown and control larvae with antibody raised against PENK . ( J ) LEUC knockdown and control larvae with antibody raised against LEUC . Scale bar: 30 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 11147 . 010 To display the spatial relationships of 11 distinct IF labels , we employed image registration to a reference template of whole-body confocal scans . High-accuracy image registration of different specimens is possible in Platynereis , due to the stereotypic anatomy and neuronal connectivity of the larvae ( Asadulina et al . , 2012; Randel et al . , 2015; Tomer et al . , 2010 ) . For optimal VNC registration , we generated an unbiased average whole-body reference template based on the non-rigid registration of the acetylated tubulin IF signal from 36 specimens . We aligned whole-body confocal scans of IF specimens to this reference using non-rigid image registration ( Figure 5A , B; Video 1 ) . We then took virtual cross sections from this registered dataset to analyze the spatial relationships of distinct peptidergic axons ( Figure 5C–E ) . The arrangement of peptidergic axons in the VNC reconstructed by IF and image registration was similar to that of the VNC reconstructed by siGOLD ( Figure 5F–H ) . To allow a more direct comparison of the spatial relationship of peptidergic axons , we performed double IF with each of the RGWa , PDF , and LEUC antibodies generated in rabbits in combination with a FVa antibody generated in rats . The spatial relationships of the neurites labeled with the RGWa , PDF , and LEUC antibodies relative to six prominent FVa neurites was very similar between IF and siGOLD ( Figure 5I , J ) . We could also detect the coexpression of FVa and PDF in some axons , both by double-IF and siGOLD . The high specificity of immunoEM , the reproducibility of labeling across sections , and the spatial correspondence of the labels between siGOLD and whole-mount IF indicate that we could accurately identify different peptidergic neurites in a serial EM dataset . 10 . 7554/eLife . 11147 . 011Figure 5 . Axonal arrangements in the VNC detected by whole-body IF spatially match those detected by siGOLD . ( A , B ) Ventral overview of individual registered full-body IF with antibodies raised against 11 different neuropeptides ( colors ) . Image is cropped to show only the VNC in the first segment . White dashed box indicates the region where a 5 μm virtual transverse section shown in ( C–E ) was taken . ( C–E ) 5 μm virtual transverse section of individual registered full-body IFs from ( A , B ) in anterior view to indicate their spatial position in the VNC relative to each other . ( C ) RGWa , FVa , PENK , MIP , ( D ) FVRIa , PDF , LUQ , FMRFa ( E ) RYa , LEUC , ATO . ( F ) Schematic overview of ( C–E ) indicating relative positioning of individual registered antibody stainings in the VNC , anterior view . ( G , H ) Reconstruction of neurites labeled by siGOLD in the VNC of specimen HT9-5 with antibodies raised against 11 different neuropeptides , ( G ) anterior view , ( H ) ventral view . For comparison with registered IF labeling in ( A–F ) . ( I ) Position of neurites in specimen HT9-5 siGOLD-labeled with FVa and RGWa ( top ) , FVa and PDF ( middle ) , and FVa and LEUC antibodies ( bottom ) . For comparison with double-IF in ( J ) . ( J ) 2 μm virtual transverse sections of the VNC of 72 hpf Platynereis larvae double-stained with the FVa antibody ( orange ) and the RGWa ( yellow , top ) , the PDF ( blue , middle ) or the LEUC ( brown , bottom ) antibodies . Scale bars: ( A–E ) 15 μm , ( F–J ) 5 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 11147 . 01110 . 7554/eLife . 11147 . 012Figure 5—source data 1 . Acetylated tubulin reference signal used for image registration . The reference is based on the registration and averaging of the acetylated tubulin signal from 36 whole-body confocal stacks . The file is a mult-image TIFF stack and can be opened e . g . with Fiji or ImageJ . DOI: http://dx . doi . org/10 . 7554/eLife . 11147 . 01210 . 7554/eLife . 11147 . 013Video 1 . Neuronal arrangements in the ventral nerve cord detected by whole-body IF and image registration . Ventral view of 72 hpf Platynereis . VNC ( grey ) of average whole-body reference template stained with acetylated tubulin , generated from scans of 36 individuals . Onto this reference scaffold , we project individual registered immunostaining patterns generated by the different neuropeptide antibodies . In order of appearance: FVa ( orange ) , RGWa ( yellow ) , LEUC ( brown ) , FVRIa ( lavender ) , RYa ( forest green ) , LUQ ( magenta ) , MIP ( aqua ) , ATO ( sky blue ) , FMRFa ( lime green ) , PENK ( red ) , PDF ( royal blue ) . DOI: http://dx . doi . org/10 . 7554/eLife . 11147 . 013 Next , we tested siGOLD on a whole-body EM series of 5056 sections encompassing an entire 72 hpf Platynereis larva ( specimen HT9-4 ) ( Randel et al . , 2015 ) . During sectioning of this larva , we set aside series of 1–6 sections for later immunoEM ( Figure 6—source data 1 ) . In the full-body series , we used the 11 rabbit antibodies ( Table 1 ) to label a total of 154 sections ( 3% of all sections ) distributed along the length of the larva ( Figure 6A ) . We used at least two grids for each antibody , separated by up to 1000 sections . Gold labeling in a whole-body context allowed us to identify several strongly labeled neurite profiles throughout the body . It is worth noting that sections were successfully gold labeled up to 3 years after sectioning the specimen , demonstrating the long-term stability of neuropeptide antigens in Epon sections . 10 . 7554/eLife . 11147 . 014Figure 6 . siGOLD labeling in a whole-body serial EM dataset ( HT9-4 ) . ( A ) Schematic of the HT9-4 specimen showing the position of the VNC ( grey ) , ventral view . The entire larva was fully sectioned and imaged . Colored lines indicate the position of sections that were used for immunolabeling . For each line , layer number ( s ) followed by the name of the neuropeptide that was immunogold-labeled in those layers are indicated . ( B ) Ventral and anterior views of all fully traced immunogold-labeled peptidergic neurons . Detailed layer information is available in Figure 6—source data 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 11147 . 01410 . 7554/eLife . 11147 . 015Figure 6—source data 1 . Complete layer statistics of the sections and images of HT9-4 . The table is updated from ( Randel et al . , 2015 ) with the information on all immunoEM layers . DOI: http://dx . doi . org/10 . 7554/eLife . 11147 . 015 We identified and traced 83 neurons ( 67 of them with a soma; Figure 6B; Video 2 ) with different peptidergic identities . We mapped the distribution of peptidergic axons crossing a section in the first trunk segment in a position comparable to that of HT9-5 and found similar patterns across both siGOLD-labeled larvae and the IF samples ( Figure 7A , B; compare to Figure 5 ) . For selected neurons , we quantified the number of gold particles in every immunolabeled section in HT9-4 ( Figure 8 , Figure 9 ) . These counts again demonstrated that we could repeatedly label the same neurons in different sections , often spaced several hundred sections apart . 10 . 7554/eLife . 11147 . 016Figure 7 . Position of siGOLD-labeled neurites in a cross-section of the VNC in HT9-4 . ( A ) TEM image of a VNC cross-section with the segmented profiles of peptidergic neurites identified by siGOLD . Different colors represent different neuropeptide-antibody labeling . ( B ) Segmented profiles of peptidergic axons in a cross-section through the VNC near the base of the circumesophageal connectives at the first commissure . All traced neuropeptidergic axons crossing the VNC at that level in the HT9-4 specimen are shown . Scale bar: ( B ) 5 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 11147 . 01610 . 7554/eLife . 11147 . 017Figure 8 . siGOLD labeling and whole-body neuron reconstructions in HT9-4 . ( A , G ) Full body IF labeling of FVa and PDF-positive cells , ventral view and anterior view , respectively . Note that ( A ) is 72 hpf and ( G ) is 48 hpf . Arrowheads point to neuron cell-bodies that were traced in EM reconstructions . ( B–F , H–L ) Fully reconstructed neurons , identified using the siGOLD method in the full body HT9-4 dataset . Dashed arrows indicate immunogold labeled layers along the neurite , layer number ( s ) and number of gold particles per layer are shown . Reconstructed FVa ( B–F ) and PDF ( H–L ) positive cells . Scale bars: ( A ) 50 μm; ( G ) 30 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 11147 . 01710 . 7554/eLife . 11147 . 018Figure 9 . siGOLD labeling and whole-body neuron reconstructions in HT9-4 . ( A , C ) Full body IF labeling of RGWa and MIP-positive cells in 72 hpf larvae , anterior views . Arrowheads point to neurons that were traced in EM reconstructions . ( B , D ) Traced neurons , identified using the siGOLD method in the full body HT9-4 dataset . Dashed arrows indicate immunogold-labeled layers along the neurite , layer number ( s ) and the number of gold particles per layer are shown . Reconstructed RGWa ( B ) and MIP ( D ) positive cells . Scale bars: ( A , C ) 30 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 11147 . 018 The reconstruction of several neurons in the whole-body serial EM dataset allowed us to further test the specificity of the immunogold labels by comparing the morphology and position of reconstructed neurons to neurons that were identified by whole-body IF . We found comparable cellular morphologies and positions for six FVa , several PDF ( Figure 8 ) , two RGWa , and four MIP neurons ( Figure 9 ) between siGOLD and IF . This close anatomical correspondence further supports the specificity of siGOLD and demonstrates that the method can be used to unambiguously tag fully reconstructed neurons in serial EM . 10 . 7554/eLife . 11147 . 019Video 2 . 3D view of the 72 hpf HT9-4 larva showing fully-traced peptidergic neurons identified by siGOLD . Ventral view of the 72 hpf Platynereis siGOLD-tagged peptidergic neurons are shown in the following colors: FVa ( orange ) , RGWa ( yellow ) , LEUC ( brown ) , FVRIa ( lavender ) , RYa ( forest green ) , LUQ ( magenta ) , MIP ( aqua ) , ATO ( sky blue ) , FMRFa ( lime green ) , PENK ( red ) , PDF ( royal blue ) . Other non-labeled traced neurons ( grey ) provide the outline of the larva . DOI: http://dx . doi . org/10 . 7554/eLife . 11147 . 019 The whole-body serial EM dataset combined with siGOLD labeling allows the reconstruction of circuits of neurons with specific peptidergic identities . To demonstrate this , we focused on the nuchal organs in the Platynereis larval head as an example . The nuchal organ is a paired putative chemosensory organ in the annelid head with sensory neurons projecting a sensory dendrite into an olfactory pit . The olfactory pit is covered by cuticle and is associated with a patch of motile cilia ( Figure 10A–C; Video 3 ) ( Purschke , 2005 ) . We found that several sensory neurons of the nuchal organs ( SNnuch ) were strongly labeled by the PDF antibody in three different PDF-labeled sections in the head ( Figure 6A; Table 2 ) . Of the 35 SNnuch cells we identified , 15 were labeled with the PDF antibody , with some of the neurons showing strong gold labeling in multiple sections ( Table 2 ) . This is consistent with the labeling of the nuchal organ with the PDF antibody in IF ( Figure 8G ) . We traced all SNnuch neurons and identified their direct postsynaptic targets . We found that the most strongly connected neurons , receiving up to 22 synapses from SNnuch cells , were two pairs of interneurons ( INarc ) with a unique biramous morphology and contralaterally projecting axons ( Figure 10D–F; Figure 10—figure supplement 1; Video 4 ) . Two of the INarc neurons were previously shown to be postsynaptic to the photoreceptor cells of the larval eyespots and have connections to the ventral motorneurons ( Randel et al . , 2015 ) . This represents a potential functional path linking the nuchal organs to the locomotor apparatus . We also identified several other neurons that received a few synapses from the SNnuch cells , including two interneurons ( INRGWa-dcr1 , INRGWa-dcl1 ) that were identified by siGOLD labeling to express the RGWa neuropeptide ( Figure 9A , B ) . Additionally , 87 neurons received only one synapse from one of the SNnuch cells ( Figure 10—figure supplement 2 ) . The strong connectivity of the INarc cells to the nuchal organ relative to these other neurons suggests that the INarc cells represent the functionally relevant targets in the circuit . 10 . 7554/eLife . 11147 . 020Figure 10 . Reconstruction of a siGOLD-labeled peptidergic circuit . ( A , B ) SEM image of the nuchal organs ( arrowhead ) in the dorsal-posterior head of a 72 hpf Platynereis larva , dorsal view . Boxed area is shown enlarged in ( B ) . Arrowheads point at motile cilia above the olfactory pit . ( C ) TEM image of a cross section of the nuchal organ showing the olfactory pit with sensory neurons and microvilli . Arrowheads point at sensory cilia . Sensory endings in the olfactory pit are highlighted in magenta , an epithelial cell underneath the cuticle is highlighted in blue . ( D ) 3D reconstruction of the nuchal organ circuit . SNnuch neurons ( red ) , and INarc ( orange ) interneurons are shown . The photoreceptor cells of the adult eyes are shown in blue as a reference . ( E , F ) 3D reconstruction of INarc ( orange ) interneurons . ( G ) Graph representation of SNnuch connectivity . Nodes represent neurons or groups of neurons , edges represent synaptic connections . The number of synapses is indicated on each arrow . Edge thickness is proportional to the square root of the number of synapses . ( H , I ) Presynaptic sites ( red dots ) in SNnuch neurons and adult-eye photoreceptor cells ( PRC ) with their soma on the right ( H ) or left ( I ) side of the body . ( J , K ) Peptidergic synapses in SNnuch neurons . ( L ) A glutamatergic synapse in an adult eye photoreceptor cell . Scale bar: ( A ) 30 μm; ( C ) 1 μm; ( L ) 150 nm . DOI: http://dx . doi . org/10 . 7554/eLife . 11147 . 02010 . 7554/eLife . 11147 . 021Figure 10—figure supplement 1 . Morphology of SNnuch and INarc neurons . ( A , D ) Four examples of SNnuch sensory neurons with the presynaptic ( red ) and postsynaptic sites shown . ( E , H ) Morphology of the four INarc interneurons with the presynaptic ( red ) and postsynaptic sites shown . DOI: http://dx . doi . org/10 . 7554/eLife . 11147 . 02110 . 7554/eLife . 11147 . 022Figure 10—figure supplement 2 . Number of synaptic inputs from SNnuch cells to all postsynaptic targets . All postsynaptic targets of the SNnuch neurons are listed . DOI: http://dx . doi . org/10 . 7554/eLife . 11147 . 02210 . 7554/eLife . 11147 . 023Table 2 . Number of gold particles in SNnuch neurons in three layers labeled with the PDF antibody . DOI: http://dx . doi . org/10 . 7554/eLife . 11147 . 023LayerLayerSNnuch703723787SNnuch703723787r1020l1000r2000l2104r3000l3000r4007l4000r5000l5000r6200l60310r7000l7002r8012l8103r9000l9000r10000l10000r11000l11000r12030l12610r13132l13996r14000l141800r15-r20000l15000Number of gold particles in SNnuch neurons identified in three different immunogold layers labeled with the PDF antibody in all SNnuch sensory neurons . 10 . 7554/eLife . 11147 . 024Video 3 . Reconstruction of the nuchal organ . Reconstruction of the nuchal organ from the dorsal head of the 72 hpf Platynereis larva dataset HT9-4 . The nuchal organ consists of three multiciliated cells and 16 sensory cells which project a sensory cilium and branching microvilli into an olfactory pit just below the cuticular layer . DOI: http://dx . doi . org/10 . 7554/eLife . 11147 . 02410 . 7554/eLife . 11147 . 025Video 4 . Reconstruction of the nuchal organ circuit . The nuchal organ sensory cells ( SNnuch , red ) connect to two pairs of INarc interneurons ( orange ) . The adult eye photoreceptors ( dark blue ) are shown for reference . Several other neurons are shown in pale grey to highlight the shape of the larval nervous system . DOI: http://dx . doi . org/10 . 7554/eLife . 11147 . 025 Mapping the position of presynaptic sites of SNnuch cells revealed that these cells form synapses bilaterally ( Figure 10 H , I ) . This is in contrast to the adult eye photoreceptor cells that only form ipsilateral synapses ( Figure 10 H , I ) ( Randel et al . , 2014 ) . To characterize the synapses of the PDF-positive SNnuch cells in more detail , we examined several synapses in high-resolution ( 2 . 22 nm/pixel ) images ( Randel et al . , 2015 ) . We found that SNnuch presynaptic sites contained only DCVs ( Figure 10J , K ) . These synapses could be clearly distinguished from classical neurotransmitter synapses ( Figure 10L ) ( Randel et al . , 2014 ) based on the larger size of the synaptic vesicles ( mean diameter=63 nm , S . D . = 8 . 4 , n=100 vesicles ) and the electrondense core of the vesicles ( Figure 10J , K ) . Furthermore , mapping classical neurotransmitter markers by in situ hybridization did not reveal the presence of any classical neurotransmitters in the nuchal organ region ( Randel et al . , 2014 ) . Although we did not directly immunogold label any of the SNnuch synapses , these observations suggest that the SNnuch neurons signal to their target interneurons by PDF-containing vesicles concentrated at peptidergic synapses .
In this paper , we introduced siGOLD , a method to molecularly identify specific neurons in large serial EM datasets based on the immunogold labeling of subsets of sections followed by serial reconstruction of the tagged neurons . Due to the sparse labeling of a few sections across many , siGOLD allows the use of several different antibody labels , each at different sections , within a single dataset . siGOLD relies on continuous series of sections of sufficient quality that allow the tracing of neurites in order to reconstruct the morphologies of labeled neurons . Since siGOLD requires the sparse labeling of small subsets of sections , it is ideally used with markers that broadly label a neuron’s morphology . We established siGOLD using neuropeptide antibodies for two reasons . First , neuropeptides are distributed along the entire length of axons , due to their active circulation throughout the axon ( Wong et al . , 2012 ) . Second , neuropeptide antigens have unique properties , including their small size , high abundance , concentration in DCVs , and frequent C-terminal amidation , a modification that confers increased stability and immunogenicity to mature neuropeptides ( Conzelmann and Jékely , 2012; Eipper et al . , 1992 ) . In agreement with the extreme stability of neuropeptide immunoreactivity , we could perform immunoEM of ‘set-aside’ layers using neuropeptide antibodies years after sectioning and the acquisition of the complete image series had been completed . We did not test other antibodies , but several generally used antibody markers , including antibodies against neurotransmitters , transporters , or enzymes ( Collman et al . , 2015 ) , could in principle be suitable for neuron identification using siGOLD . We established siGOLD on Epon-embedded samples that allowed robust sectioning of thousands of sections and provided high ultrastructural detail . Since Epon-embedding is known to compromise the immunogenicity of some antigens ( Brorson , 1998; Brorson and Reinholt , 2008; De Paul et al . , 2012 ) , labeling for such antigens may require the use of alternative embedding resins such as Lowicryl HM-20 that provides excellent ultrastructural contrast and is compatible with many antibodies ( Collman et al . , 2015 ) . siGOLD is compatible with conventional TEM using sections collected on slotted , plastic-coated nickel grids ( Skepper and Powell , 2008 ) . It is also compatible with sectioning methods that collect long ribbons of sections on a glass slide ( Blumer et al . , 2002 ) or the high-throughput automatic tape-collecting ultramicrotome ( ATUM ) method ( Hayworth et al . , 2014 ) . However , siGOLD is not compatible with sectioning procedures that destroy the sections , such as focused ion beam ( FIB ) SEM ( Reyntjens and Puers , 2001 ) or the serial block-face method ( Denk and Horstmann , 2004 ) . The siGOLD method differs from conjugate light-electron array tomography ( Collman et al . , 2015 ) and CLEM in several aspects . Array tomography was developed to provide detailed molecular profiling of individual synapses , and uses repeated immunolabeling of every section . siGOLD does not rely on the staining of every section with multiple markers , but rather the staining of sparsely distributed sections with one marker each . The aim of siGOLD is to assign molecular tags to several different neurons and to trace them through many layers . Array tomography and CLEM involves the registration of separately acquired IF and EM images , and the relocation of the sample between the different imaging setups can be challenging and time consuming ( Timmermans and Otto , 2015 ) . In contrast , siGOLD relies on the direct immunoEM labeling of sections and no registration step is needed . With siGOLD , we enjoy the full resolving power of the electron microscope , and in this study we could also identify individual DCVs that carry strong immunogold signal in the neurite profile of specific peptidergic neurons . One shortcoming of the immunoEM approach is that sections cannot be relabeled after they have been contrasted and exposed to the electron beam . However , given that in siGOLD only a subset of the sections is processed for immunoEM , it is possible to use an arbitrarily large number of antibodies at different sections from a series that encompasses a large volume of tissue , allowing the multiplex identification of neuron types . We demonstrated that siGOLD can be used to molecularly identify neurons in large serial EM datasets , in conjunction with the reconstruction of the synaptic connectivity of these neurons . We reconstructed the sensory neurons of the nuchal organs and their postsynaptic partners , representing a candidate chemotactic circuit in the Platynereis larval head . siGOLD labeling revealed that the sensory neurons express the PDF neuropeptide . The presence of DCVs and lack of clusters of clear vesicles at presynaptic sites in these neurons suggest the presence of peptidergic transmission to the target interneurons . The bilateral projection of the nuchal organ sensory neurons is similar to that seen for olfactory receptor neurons in Drosophila . These neurons form bilateral projections , and the lateralization of odor processing is achieved by an asymmetry in neurotransmitter release ( Gaudry et al . , 2013 ) . The nuchal organ circuit may use a similar computational strategy , and this is likely different from the processing of visual information ( Randel et al . , 2014; Randel et al . , 2015 ) . The molecular information obtained by siGOLD provides a useful entry point to the genetic investigation of this putative chemotactic circuit in Platynereis . The mapping of peptidergic neurons by siGOLD , together with the recent identification of several neuropeptide receptors in Platynereis ( Bauknecht and Jékely , 2015 ) , open up the possibility of a cellular-level analysis of peptidergic neurotransmission in diverse circuits in Platynereis . In total , we have identified 83 different peptidergic neurons out of approximately 2000 total in the Platynereis whole-body dataset . Further connectome tracing in these data will provide a rich source of information about specific peptidergic circuit motifs in this animal . The siGOLD approach and the use of neuropeptide antibodies could also be adapted to enrich connectome data with molecular information in other organisms . In Platynereis , as well as vertebrates , C . elegans , and Drosophila , the majority of mature neuropeptides are amidated ( Jékely , 2013; Mirabeau and Joly , 2013 ) , and several antibodies are available ( e . g . ( Johard et al . , 2008; Nässel , 1993 ) ) or could readily be generated . Some neuropeptide antibodies have already been shown to work with immunoEM . For example , PDF neuropeptide-containing neurons were identified in the Drosophila brain by immunoEM and their local synaptic inputs were reconstructed from a series of sections ( Yasuyama and Meinertzhagen , 2010 ) . This and other antibodies could be used for siGOLD labeling in larger-scale connectome projects in Drosophila . Similar protocols and reagents could also be established for other organisms . siGOLD could also be adapted to organisms where transgenic tools to deliver EM-compatible markers are not available . In such organisms , the use of cross-species antibodies ( Conzelmann and Jékely , 2012; Nässel , 1993 ) is a promising approach that would also allow the comparison of specific peptidergic neurons and their circuits across species . The direct overlaying of chemical neuromodulatory maps onto synaptic connectomic maps by siGOLD opens up new possibilities for the study of nervous systems .
Fixation and embedding was carried out on 72 hpf Platynereis larvae ( HT9-4 and HT9-5 ) as described previously ( Conzelmann et al . , 2013a ) . Forty nanometer serial sections were cut on a Reichert Jung Ultracut E microtome using a 45º DiATOME Diamond knife . The sections were collected on single-slotted copper grids ( NOTCH-NUM 2_1 mm , Science Service , Munich ) with Formvar support film . For the HT9-5 serial sections , the samples were collected on single-slotted nickel grids coated with Formvar support film . The ribbons of sections were picked up from underneath and on the under-side ( the notch side ) of the grids . This method allowed the section ribbons to stay afloat and stretch while being dried in the oven , eliminating most wrinkles from the sections . The section statistics for the HT9-4 specimen ( or NAOMI ) were previously described ( Randel et al . , 2015 ) and an updated table with the immunoEM information is shown in Figure 6—source data 1 . For HT9-5 , 200 sections were cut from the VNC ( first segment ) and imaged for tracing , segmentation and analysis . To meet the demand for consistent section processing in large-scale serial reconstruction projects , we have developed an effective method of grid handling based on a method first described in Rowley and Moran ( 1975 ) . The following method has proven safe for contrast staining , immunolabeling , and carbon coating . It stabilizes the grids during the staining procedures and saves time by eliminating the need for one-by-one grid manipulation . A 6 × 12 hole microwell mini tray plate ( NUNC™ Brand MicroWell® Mini Trays ) was cut into two 3 × 12 hole plate strips . Holes were drilled through each of the plate’s micro-cups using a fine drill to later allow the plastic support film to dry evenly on both sides . Plates were rinsed with 75% ethanol , followed by distilled water , then sonicated to eliminate dust and burr particles . This procedure produced two lightweight plates similar to that described in Rowley and Moran ( 1975 ) with the additional benefit of 'raised' , separated walls . This approach prevents cross-contamination by diffusion of solutions during the staining and labeling procedures . A strip of Formvar support film was used to coat the microwell mini plate on the topside of the plate . A small water droplet ( 10 µl ) was placed on each hole of the plate on the support film using a filtered syringe . A grid containing previously cut sections was placed ( sections facing up ) on each droplet and dried in the 50ºC oven until the grids and the plastic support film had fused . Grids were then ready for immunolabeling and contrast staining by inverting and matching to the micro-cups of a full mini tray with desired solutions ( Figure 1A – Mini tray ) . For immunoEM , we have eliminated the etching step prior to primary antibody incubation . We have observed that the treatment with the etching solution destroys ultrastructural integrity . In our technique , the use of sheep serum and bovine serum albumin mixed with Tween-20 at pH of 7 . 9 greatly enhanced the immunogold signal . The following protocol was performed over a 1-day period and is an optimized version of a protocol by Stierhof ( Stierhof et al . , 1991 ) for Epon-embedded sections . All incubations up to the silver enhancement point were done in the TBST-BGN ( Tris buffer saline with Tween-20 , BSA , Fish Gelatin and Normal Sheep Serum , pH 7 . 9; see section below on preparing the buffer ) . First , sections were blocked with TBST-BGN for 10 min; blocked sections were then incubated with primary antibody ( 1:25 dilution ) in TBST-BGN for 2 hr . Grids were washed twice with TBST-BGN for 5 min , and then incubated with secondary goat anti-rabbit IgG antibody with gold conjugates ( AURION Ultra Small Immunogold Reagents; size 0 . 8 nm ) at 1:50 dilution for 1 hr . Grids were then washed with TBST-BGN twice for 5 min , then washed twice with filtered distilled water for 5 min , followed by fixation with 1% glutaraldehyde in filtered distilled water for 5 min . Sections were washed in filtered distilled water twice for 5 min . Silver enhancement was applied with Aurion silver-enhancement kit for up to 47 min . The time interval of silver enhancement was dependent on temperature and concentration of the silver-enhancement solution . Gold size growth was approximately 15–20 nm within this time interval , a size appropriate for observation of gold particles in EM . Silver enhancement was stopped with three 3-min washes in filtered distilled water . Excess water was gently wicked from the back of the plate and individual holes with wedges of Whatman filter paper . Grids were dried in a 50°C oven and then contrasted ( see above ) . Buffer concentration and consistency is very important in immunological reactions . We compared many protocols to achieve the best buffer for our experiments . We found that a saline buffer with Tris , Tween , bovine serum albumin ( BSA ) , fish Gelatin , and normal serum ( TBST-BGN ) , worked well . Our solutions were always freshly prepared on the day of the experiment . To prepare the buffer for immunoEM , three separate solutions were made: ( 1 ) 0 . 61 g Trizma , 0 . 90 g NaCl , and 70 ml of distilled water , ( 2 ) 1 . 0 g of BSA , 1 . 5 ml normal sheep serum , 500 µl Tween-20 , and 20 ml distilled water , and ( 3 ) 0 . 01 g of fish Gelatin and 10 ml of distilled water . Solution 3 was heated to dissolve and then cooled before being added to solutions 1 and 2 . Finally , solutions 1–3 were combined and gently mixed . The buffer solution was then adjusted to pH of 7 . 9 using 1 M HCl and syringe filtered with a 0 . 45 µm filter . Image acquisition of TEM serial sections was performed on a FEI TECNAI Spirit transmission electron microscope equipped with an UltraScan 4000 4X4k digital camera using the image acquisition software Digital Micrograph ( Gatan Software Team Inc . , Pleasanton ) and SerialEM ( Mastronarde , 2005 ) . The images for the HT9-4 and HT9-5 samples were scanned at a pixel resolution of 5 . 71 nm/pixel and 2 . 22 nm/pixel , respectively . Image stitching and alignment were accomplished using TrakEM2 ( Cardona et al . , 2010; Cardona et al . , 2012 ) . All structures were segmented manually as area-lists , and exported into 3Dviewer and Blender as previously described ( Asadulina et al . , 2015 ) . Tracing and annotation of the connectome were performed with CATMAID , a collaborative annotation toolkit for large-scale image data ( Saalfeld et al . , 2009; Schneider-Mizell et al . , 2015 ) . Antibodies were generated by immunizing rats or rabbits with synthetic , amidated peptides . Some of the antibodies were used in previous studies ( Conzelmann and Jékely , 2012; Conzelmann et al . , 2011; Conzelmann et al . , 2013a; Jékely et al . , 2008 ) . All peptides contained an N-term Cys that was used for coupling . Antibodies were affinity purified from sera as previously described ( Conzelmann and Jékely , 2012 ) . Immunostainings were carried out as previously described ( Conzelmann and Jékely , 2012 ) . For triple IF , we used secondary antibodies coupled to three different fluorophores ( FVa rat , AF488 secondary antibody rat; other neuropeptide rabbit , AF647 secondary antibody rabbit; acetylated-tubulin mouse , high-fidelity AF555 secondary antibody mouse ) . For image registration , an average full-body acetylated tubulin reference template was generated for 72 hpf Platynereis using a modification of a previously described method ( Figure 5—source data 1 ) ( Asadulina et al . , 2012 ) . We used full-body scans of 36 larvae generated on a Zeiss LSM 780 confocal microscope with ZenBlue software . All stacks were oriented and their centers of mass were aligned . Stacks were averaged , and all stacks were aligned to this first average template using affine transformation . The 24 registered stacks most similar to the first average ( as determined by an iteration metric ) were then used to create an affine-transformed average . All original oriented stacks were then aligned to the affine-transformed average using affine and deformable transformation . The 24 registered stacks most similar to the affine-transformed average were then used to create an affine/deformable-transformed average . All original oriented stacks were then aligned to the affine/deformable-transformed average using affine and deformable transformation . The 24 registered stacks most similar to the affine/deformable-transformed average were then used to generate the final average acetylated tubulin reference template . The final average stack was unbiased and was used for image registration . Imaris was used for image processing ( adjusting brightness and contrast uniformly ) and to take virtual cross-sections of the VNC for comparison with siGOLD samples . Morpholino injections were performed as previously described ( Conzelmann et al . , 2013a ) . We used the following morpholinos to target the various neuropeptide precursor genes ( GeneTools , LLC ) : Pdu-FMRFa-start MO CCACTGGTCCCTCATGGCAGGGTTT , Pdu-PDF-start MO CTGAACTGTCTTGCTTGATCCCATC , Pdu-ATO-start MO CACAGGACTACCTTCATTTTTCTGA , Pdu-LUQ-start MO GTATTTGCACAACATAGTGATAGTC , Pdu-PENK-start MO GAGGAGGACCACCAATATCTTCATC , Pdu-RYa-start MO TATAGACATGACACCTTGTTGGAGT , Pdu-LEUC-start2 MO TCTTGGCTGAACTCATTGCGGCC , Pdu-FVa-start MO CCATCCGCCCACGCTCATATGCATC , Pdu-FVRIa-start MO CCCCCTTCATACTGTCACAACGGAC , Pdu-RGWa-start MO CGACGACCCCCTGTAGCTTCATGTC . | In the nervous system , cells called neurons connect to each other to form large “neural” networks . The most powerful method that is currently available for tracing neurons and mapping the connections between them is called electron microscopy . This requires slicing brain tissue into ultrathin sections , which are then imaged one by one . However , while electron microscopy provides highly detailed information about the structure of the connections between neurons , it does not reveal which molecules the neurons use to communicate with each other . To address this question , Shahidi et al . have developed a new approach called ‘siGOLD’ . Unlike previous approaches , siGOLD allows signal molecules inside cells to be labeled with protein tags called antibodies without compromising the ability to examine the tissue with electron microscopy . The technique was developed using the larvae of a marine worm called Platynereis . A single larva was sliced into 5000 sections thin enough to view under an electron microscope , and 150 of these were selected to represent the entire body . Because neurons are typically long and thin , individual neurons usually spanned multiple slices . To identify the neurons , Shahidi et al . then applied an antibody that recognizes a specific signal molecule to a subset of the slices . The antibodies were labeled with gold particles , which show up as black dots under the electron microscope . Because the molecules recognized by the antibodies are present all along the neuron , and because individual neurons extend over multiple slices , it was possible to trace single neurons by labeling only a small number of slices . Repeating this process in different subsets of slices with antibodies that bind to different signal molecules allowed entire neural circuits to be mapped . In the future , Shahidi et al . ’s approach could be adapted to study neural networks in other organisms such as flies , fish and mice . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"M"
] | [
"neuroscience"
] | 2015 | A serial multiplex immunogold labeling method for identifying peptidergic neurons in connectomes |
Malate and lactate dehydrogenases ( MDH and LDH ) are homologous , core metabolic enzymes that share a fold and catalytic mechanism yet possess strict specificity for their substrates . In the Apicomplexa , convergent evolution of an unusual LDH from MDH produced a difference in specificity exceeding 12 orders of magnitude . The mechanisms responsible for this extraordinary functional shift are currently unknown . Using ancestral protein resurrection , we find that specificity evolved in apicomplexan LDHs by classic neofunctionalization characterized by long-range epistasis , a promiscuous intermediate , and few gain-of-function mutations of large effect . In canonical MDHs and LDHs , a single residue in the active-site loop governs substrate specificity: Arg102 in MDHs and Gln102 in LDHs . During the evolution of the apicomplexan LDH , however , specificity switched via an insertion that shifted the position and identity of this ‘specificity residue’ to Trp107f . Residues far from the active site also determine specificity , as shown by the crystal structures of three ancestral proteins bracketing the key duplication event . This work provides an unprecedented atomic-resolution view of evolutionary trajectories creating a nascent enzymatic function .
The common ancestor of the eukaryotic Apicomplexa evolved nearly 1 billion years ago ( Douzery et al . , 2004 ) , and its modern descendants comprise a large phylum of intracellular parasites that are currently responsible for numerous devastating metazoan diseases , including malaria ( Plasmodium ) , toxoplasmosis ( Toxoplasma ) , cryptosporidiosis ( Cryptosporidium ) , cyclosporiasis ( Cyclospora ) , and babesiosis ( Babesia ) . A key event in the early evolution of the Apicomplexa was the acquisition of a malate dehydrogenase ( MDH ) via lateral gene transfer from α-proteobacteria ( Golding and Dean , 1998; Madern , 2002; Zhu and Keithly , 2002 ) . Following a gene duplication event roughly 700–900 Mya , one copy of this MDH evolved a novel substrate specificity to become a highly specific lactate dehydrogenase ( LDH ) that is now essential to the life cycle of many modern apicomplexans ( Royer et al . , 1986 ) . As a core metabolic enzyme that evolved independently of metazoan LDH , the unique apicomplexan LDH has attracted significant attention as a potential drug target ( Gomez et al . , 1997; Read et al . , 1999; Cameron et al . , 2004; Conners et al . , 2005 ) . However , the molecular and evolutionary mechanisms that drove this switch in substrate specificity are currently unknown . LDH and MDH are homologous , 2-ketoacid oxidoreductases that share both a protein fold ( Rossmann et al . , 1975; Figure 1—figure supplement 1 ) and a common catalytic mechanism ( Birktoft and Banaszak , 1983; Clarke et al . , 1986; Hart et al . , 1987a , 1987b; Clarke et al . , 1988; Waldman et al . , 1988; Figure 1 ) . Both enzymes are found in central metabolism: MDH catalyzes the interconversion of oxaloacetate and malate in the citric acid cycle , and LDH converts pyruvate to lactate in the final step of anaerobic glycolysis . Despite their structural and catalytic similarities , modern apicomplexan LDHs and MDHs have extraordinarily strict substrate specificity . For example , Plasmodium falciparum ( Pf ) MDH and LDH each prefer their respective substrates by over six orders of magnitude . The biophysical basis for this extraordinary substrate preference is presently an unresolved question . 10 . 7554/eLife . 02304 . 003Figure 1 . Schematic of M/LDH superfamily active site and catalytic mechanism . MDH reduces oxaloacetate to malate , in which the R-group is a methylene carboxylate group . LDH reduces pyruvate to lactate , in which the R-group is a methyl group . Key conserved active site residues are shown in black; substrate is shown in blue . The oxidized 2-ketoacid form of the substrate is at left; the reduced 2-hydroxy acid form is shown at right . The R-group of the substrate interacts with Arg102 in MDHs and Gln102 in canonical LDHs . Both Arg109 and position 102 are found in the mobile ‘specificity loop’ that closes over the active site . DOI: http://dx . doi . org/10 . 7554/eLife . 02304 . 00310 . 7554/eLife . 02304 . 004Figure 1—figure supplement 1 . Fold architecture in the LDH and MDH superfamily . At left is CpMDH ( blue , 2hjr ) , at right is PfLDH ( vermilion and olive , 1t2d ) . The Rossmann fold domain , which binds the NADH cofactor , is show as light blue in CpMDH and vermilion in PfLDH . The active site is found at the interface of the two domains . In CpMDH , the ‘opposing loop’ is highlighted in yellow ( see text ) . In PfLDH , the six-residue insertion is highlighted in yellow . DOI: http://dx . doi . org/10 . 7554/eLife . 02304 . 004 A conspicuous structural difference between apicomplexan MDHs and LDHs is an insertion within the active site loop of the LDHs ( Bzik et al . , 1993; Dunn et al . , 1996; Figure 2 ) . In the LDH/MDH superfamily , closure of this loop over the active site is rate-limiting during catalysis ( Waldman et al . , 1988 ) , and mutations within this loop have large effects on activity and substrate specificity ( Wilks et al . , 1988 ) . For example , simply mutating Gln102 to Arg in the specificity loop of Bacillus stearothermophilus ( Bs ) LDH converts the enzyme into an MDH , shifting specificity from a 103-fold preference for pyruvate to a 104-fold preference for oxaloacetate ( Wilks et al . , 1988 ) ( Figure 1 , residue numbering is based on the dogfish LDH convention [Eventoff et al . , 1977] ) . In fact , all known MDHs have an Arg at position 102 , while canonical LDHs have a Gln , and consequently residue 102 has been called the 'specificity residue' ( Chapman et al . , 1999 ) . Residue 102 is thought to contribute to substrate discrimination by balancing the substrate charge within the active site: the positively charged Arg in MDHs forms a salt bridge with the C4 carboxylate of oxaloacetate , whereas the neutral Gln in canonical LDHs packs with the C3 methyl of pyruvate ( Figure 1 ) . Yet , attempts to convert an MDH into an LDH by mutating Arg102 to Gln have met with limited success ( Nicholls et al . , 1992; Cendrin et al . , 1993 ) . In the apicomplexan LDHs , residue 102 is not a Gln but a Lys , a relatively conservative substitution compared to the MDH Arg . It is currently not understood why Plasmodium LDHs lack activity towards oxaloacetate , despite having a positively charged sidechain at residue 102 similar to MDHs ( Gomez et al . , 1997; Dando et al . , 2001; Winter et al . , 2003; Brown et al . , 2004; Kavanagh et al . , 2004; Shoemark et al . , 2007 ) . 10 . 7554/eLife . 02304 . 005Figure 2 . Apicomplexan M/LDH active sites . Structures of CpMDH ( blue , 2hjr ) and PfLDH ( vermilion , 1t2d ) superposed using THESEUS . The ligands ( oxalate and NAD+ ) are from 1t2d and colored white . Side chains of important residues are shown as sticks and the six-residue insert of PfLDH is highlighted in yellow . Note how the PfLDH Trp107f overlays Arg102 from CpMDH . Residues in the insertion are labeled using numbers and letters to maintain consistency with homologous positions in the dogfish LDH . DOI: http://dx . doi . org/10 . 7554/eLife . 02304 . 00510 . 7554/eLife . 02304 . 006Figure 2—figure supplement 1 . Sequence alignment of the specificity loop from apicomplexan M/LDHs with ancestral sequences . DOI: http://dx . doi . org/10 . 7554/eLife . 02304 . 00610 . 7554/eLife . 02304 . 007Figure 2—figure supplement 2 . Alanine scanning of PfLDH specificity loop . Logarithm of pyruvate kcat/KM of PfLDH and each mutant . Labels on x-axis describe the mutation tested in the WT PfLDH background . DOI: http://dx . doi . org/10 . 7554/eLife . 02304 . 00710 . 7554/eLife . 02304 . 008Figure 2—figure supplement 2—source data 1 . Kinetic parameters for PfLDH alanine-scan . DOI: http://dx . doi . org/10 . 7554/eLife . 02304 . 00810 . 7554/eLife . 02304 . 009Figure 2—figure supplement 3 . Crystal structure of PfLDH-W107fA mutant . ( A ) Crystal lattice of the W107fA mutant ( left ) compared to the WT PfLDH ( right ) . ( B ) Superposition of the WT PfLDH ( olive ) and the W107fA mutant ( vermilion ) . The structures are highly similar throughout , expect for the active site loop ( at top ) , which is closed in the WT and partially disordered and open in the mutant . DOI: http://dx . doi . org/10 . 7554/eLife . 02304 . 009 Apicomplexan LDH evolved from the duplication of an ancestral MDH gene ( Golding and Dean , 1998; Zhu and Keithly , 2002 ) . Gene duplication is widely considered the major force that has driven the evolutionary diversity of protein functions ( Innan and Kondrashov , 2010 ) . There are three general ways duplicated genes can be fixed in a population by selection: ( 1 ) ‘dosage selection’ , beneficial increase in dosage due to multiple copies , ( 2 ) ‘subfunctionalization’ , specialization of previously existing functions , or ( 3 ) ‘neofunctionalization’ , creation of a novel function through the accumulation of beneficial , gain-of-function mutations ( Ohno , 1970 ) . Most mutations , however , are either neutral or detrimental . A new duplicated gene typically degrades to a crippled pseudogene before it can acquire the rare beneficial mutations needed to confer a selectable function ( Walsh , 1995; Lynch and Conery , 2000 ) . Hence , classical neofunctionalization has fallen out of favor in preference for models that begin with the duplication of a multifunctional protein , such as ‘specialization’ and ‘subfunctionalization’ models . Currently the molecular and evolutionary mechanisms that create novel functions in gene duplicates are fiercely debated ( Force et al . , 1999; Conant and Wolfe , 2008; Innan and Kondrashov , 2010; Soskine and Tawfik , 2010 ) , and there are few clear examples of classic neofunctionalization or gain-of-function mutations ( Zhang and Rosenberg , 2002; Bridgham et al . , 2008; Voordeckers et al . , 2012 ) . The apicomplexan LDH and MDH enzyme family provides an exceptional model system for investigating several long-standing questions in molecular evolution , including the mechanisms available to convergent evolution , the number of mutations required to produce a nascent function , the role of promiscuous intermediates during evolution of function , and the effects of epistasis on evolutionary irreversibility . In order to identify the biophysical and evolutionary mechanisms responsible for pyruvate specificity in apicomplexan LDHs , we have reconstructed ancestral proteins along the evolutionary trajectories leading to modern apicomplexan MDHs and LDHs ( Figure 3B ) . We kinetically and structurally characterized the ancestral proteins together with multiple evolutionary intermediates . This work provides a clear example of neofunctionalization in protein evolution and the first crystal structures documenting the evolution of a new enzyme . We show that apicomplexan LDHs evolved as the result of few mutations of large effect via the classic neofunctionalization of a duplicated MDH gene . 10 . 7554/eLife . 02304 . 010Figure 3 . Phylogeny of apicomplexan M/LDH superfamily . ( A ) 1844 taxa . The tree is colored according to function ( LDH—vermilion; MDH—blue; HicDH—moss ) . The N-terminal Rossmann-fold of glucosidases and aspartate dehydrogenases ( AspDHs ) was used to root the phylogeny . Numbers highlight convergent events of LDH evolution from MDHs: 1–Canonical LDHs , 2–Trichomonad LDHs , and 3 , 4–apicomplexan LDHs . The shaded clades have highly significant supports ( Anisimova and Gascuel , 2006 ) . ( B ) Apicomplexan M/LDH Clade . A close-up of the apicomplexan portion of the phylogeny in A , similarly colored by function . aLRT supports for each group: α-proteobacteria MDHs , 15; apicomplexan LDHs , 11; Plasmodium LDHs , 333; Cryptosporidium MDHs , 54; Cryptosporidium LDHs , 202 . Ancestral reconstructed proteins are labeled at internal nodes ( AncMDH1 , AncMDH2 , AncMDH3 , AncLDH ) . The focus of the present work is the gene duplication at node 3 . DOI: http://dx . doi . org/10 . 7554/eLife . 02304 . 01010 . 7554/eLife . 02304 . 011Figure 3—figure supplement 1 . Phylogeny of M/LDH superfamily . Same phylogeny as Figure 3A with select branch supports shown ( aLRT supports ) . DOI: http://dx . doi . org/10 . 7554/eLife . 02304 . 01110 . 7554/eLife . 02304 . 012Figure 3—figure supplement 2 . Phylogeny of M/LDH superfamily . Same phylogeny as Figure 3A . The tree is colored by domain of life ( eubacterial—vermilion; eukaryotic—blue; archeal—magenta ) . DOI: http://dx . doi . org/10 . 7554/eLife . 02304 . 01210 . 7554/eLife . 02304 . 013Figure 3—figure supplement 3 . Apicomplexan M/LDH Clade . Same phylogeny as Figure 3B with aLRT branch supports and clades shown in full detail . DOI: http://dx . doi . org/10 . 7554/eLife . 02304 . 013
A maximum likelihood phylogeny of representatives of all known LDH and MDH proteins provides strong support for five distinct protein clades ( Figure 3A , Figure 3—figure supplement 1 ) : canonical LDHs , ‘LDH-like’ MDHs , mitochondrial-like MDHs , cytosolic-like MDHs , and the poorly characterized HicDHs ( hydroxyisocaproate-related dehydrogenases ) , confirming previous phylogenetic analyses ( Golding and Dean , 1998; Madern , 2002; Zhu and Keithly , 2002; Madern et al . , 2004 ) . The HicDH clade are close sequence homologs of a known hydroxyisocaproate dehydrogenase . They all possess a residue other than a Gln or an Arg at the ‘specificity’ position 102 , as well as insertions of varying lengths within the catalytic loop between residues 102 and 109 . Despite these alterations within the catalytic loop , all other catalytic residues ( Arg109 , Asp168 , Arg171 , and His195 ) are conserved . Only one taxon within the HicDH clade has been functionally characterized , DHL2_LACCO , which is a specific hydroxyisocaproate dehydrogenase ( Feil et al . , 1994 ) . These observations suggest that the clade features dehydrogenases with altered substrate specificity . Except for the HicDHs , which are exclusively eubacterial , both eukaryotic and eubacterial enzymes are found in all major clades ( Figure 3—figure supplement 2 ) . The ‘LDH-like’ MDH clade additionally contains archaeal dehydrogenases , which are basal and group to the exclusion of the bacterial MDHs . Intriguingly , three different groups of LDH proteins cluster with high confidence outside of the canonical LDH clade . A set of trichomonad LDHs found in the cytosolic-like MDH clade are thought to have evolved from a recent gene duplication of an MDH ( Wu and Fiser , 1999 ) . The Trichomonads appear to lack a canonical LDH . A prominent eukaryotic group of LDH and MDH proteins from the Apicomplexa nests deep within the bacterial ‘LDH-like’ MDHs , sister to many Rickettsiales sequences , signifying a horizontal gene transfer event from α-proteobacteria to the eukaryotic Apicomplexa . We find no evidence that the Apicomplexa have canonical LDH or conventional eukaryotic-type MDH ( either cytosolic- or mitochondrial-like MDHs ) , despite searching in many available complete apicomplexan genomes ( multiple Eimeria , Neospora , Toxoplasma , Plasmodium , and Cryptosporidium species ) ( Heiges et al . , 2006; Gajria et al . , 2008; Aurrecoechea et al . , 2009 ) . In the Apicomplexa , LDH activity has apparently evolved independently twice ( Figure 3B , Figure 3—figure supplement 3 ) , once in a lineage leading to Plasmodium-related species and once in Cryptosporidium . The apicomplexan portion of the LDH/MDH gene phylogeny is consistent with recent apicomplexan species phylogenies constructed from concatenated protein sequences ( Templeton et al . , 2009 ) . We rooted the MDH/LDH phylogeny using the Rossmann fold domain of the distantly related α/β-glucosidases and aspartate dehydrogenases as outgroups . The ML root position apparently splits the tree into two large groups: one which contains the cytosolic- and mitochondrial-like MDHs , which are largely dimeric , and another which contains the canonical LDHs , ‘LDH-like’ MDHs , and HicDHs , which are primarily tetrameric ( Figure 3—figure supplement 1 ) . While the ML root position is robust to variation in taxon coverage , the exact location is poorly supported . Nevertheless , there is strong support for a root position within the central MDH section of the tree and outside of the five identified clades , including the canonical LDH clade ( confidence level >0 . 99985 according to the aLRT ) , indicating that the canonical LDHs evolved from an ancestral MDH . The global rooting and the location of the three separate LDH groups , deep within MDH clades , indicate that LDH enzymes have evolved convergently from MDHs at least four times in the superfamily . In the present work , our focus is on the convergent evolution of the unusual apicomplexan LDHs . With the α-proteobacterial ‘LDH-like’ MDHs as the closest outgroup , the apicomplexan enzymes are split into two main groups: LDHs belonging to Toxoplasma , Plasmodium , and related protists , and MDHs belonging to Plasmodium and Cryptosporidium . Apart from their atypical phylogenetic position , the apicomplexan MDHs appear as typical α-proteobacterial ‘LDH-like’ MDHs , containing all the key catalytic residues including Arg102 . The Cryptosporidium LDHs are an exception , being nested within the apicomplexan MDH clade partitioned from the rest of the apicomplexan LDHs . Cryptosporidium LDHs have a Gly at position 102 and are thought to be a product of an independent , convergent duplication event ( Madern et al . , 2004 ) . In contrast , the large apicomplexan LDH clade is demarcated by a unique , conserved five-residue insertion in the active site loop . While the apicomplexan LDH and MDH proteins are moderately divergent , with about 45% sequence identity , the differences are largely confined to exterior residues removed from the active sites . One important difference is that the apicomplexan LDHs have Lys102 for the ‘specificity residue’ , rather than a Gln as found in the canonical LDHs ( Figure 2—figure supplement 1 ) . Apicomplexan proteins frequently contain numerous insertions relative to proteins from other species ( Feng et al . , 2006; Kissinger and DeBarry , 2011 ) , a characteristic thought to result from various factors , including high AT genome content , DNA strand slippage , double strand break repair , high recombination rates , and selection pressure for parasite antigenic variation . Except for Met106 , the amino acid and coding sequence immediately flanking the apicomplexan LDH loop insertion is largely conserved with α-proteobacterial MDHs ( Figure 2—figure supplement 1 ) . It is therefore likely that a mutation ‘expanded’ the Met106 codon to code for six residues , resulting in the observed five-residue insertion and the Met106Lys mutation . Henceforth we will refer to this expansion mutation as the ‘six-residue loop insertion’ . In the modern apicomplexan enzymes , the six-residue insertion in the LDH specificity loop ( positions 99–112 ) induces two significant structural changes relative to MDH ( Figure 2 ) . First , LDH residue Lys102 is excluded from of the active site , unlike the corresponding Arg102 in MDH , which is enclosed within the active site and participates in functionally important interactions with the substrate . Second , LDH Trp107f , which is part of the novel insertion , occupies the same space as Arg102 in MDH ( by convention , residues in the insertion are labeled using numbers and letters to maintain consistency with homologous positions in the dogfish LDH , Figure 2 ) . The only prominent structural difference between the active sites of the LDH and MDH proteins is the replacement of MDH Arg102 with LDH Trp107f . Trp107f is positioned where it could presumably interact with the distinguishing C3 methyl of the pyruvate substrate , while MDH Arg102 interacts with the C4 carboxylate of oxaloacetate ( Chapman et al . , 1999 ) . As a bulky , hydrophobic residue , Trp107f could recognize pyruvate in preference to oxaloacetate by two mechanisms: a hydrophobic interaction with the pyruvate C3 methyl vs the negatively charged oxaloacetate methylene carboxylate and steric occlusion of the methylene carboxylate of oxaloacetate . Furthermore , Trp107f is conserved in all apicomplexan LDHs ( Figure 2—figure supplement 1 ) , suggesting negative selection and functional importance . We therefore hypothesized that Trp107f plays an important role in pyruvate recognition . We tested the functional importance of residues in the specificity loop in PfLDH with an ‘alanine scan’ by individually mutating each residue in positions 101–108 to an alanine ( Figure 2—figure supplement 2 , note Ala103 was mutated to a serine ) . We assessed the activity of the mutants using kcat/Km , a measure of enzymatic specificity and catalytic efficiency , as determined from steady state kinetic assays . Mutating Trp107f to Ala reduced pyruvate activity by five orders of magnitude , whereas mutations at all other positions had effects less than a single order of magnitude , including the canonical specificity residue at position 102 . The Trp107fAla mutation affects both kcat ( 1500-fold decrease ) and Km ( 50-fold increase ) . To assess the effects of Trp107fAla mutation on the specificity loop conformation , we solved the crystal structure of PfLDH-W107fA ( 1 . 1 Å ) in the presence of oxamate and NADH . The protein crystallizes in the same space group as the wild-type PfLDH , with nearly identical cell dimensions ( Figure 2—figure supplement 3A ) . In the W107fA mutant , the specificity loop is disordered between residues Thr101 and Arg109 , as is often seen in structures in which the loop is in the open conformation . In the mutant , residues 112–115 are in a linear α-helical conformation , in contrast to the wild-type PfLDH closed state which has a very prominent 60° kink in the α-helix at Pro114 . Thus , the only significant difference between the wild-type and mutant structures is that the PfLDH-W107fA specificity loop is found in the open conformation , consistent with weaker binding of substrate ( Figure 2—figure supplement 3B ) . These results indicate that Trp107f is necessary for pyruvate activity in apicomplexan LDHs , and that it has become the new ‘specificity residue’ despite the fact that Trp107f does not align in sequence with the canonical specificity residue at position 102 ( Figure 2—figure supplement 1 ) . During evolution , the six-residue insertion displaced the canonical specificity residue at position 102 and apparently switched substrate preference in apicomplexan LDHs . If this insertion is sufficient for pyruvate recognition , then adding the insertion to a modern apicomplexan MDH should convert the enzyme to an LDH . To test this hypothesis , we incorporated the six-residue insertion from PfLDH into the catalytic loop of PfMDH ( PfMDH-INS ) and the Cryptosporidium parvum ( Cp ) MDH ( CpMDH-INS ) . The chimeric proteins showed a >100-fold reduction in oxaloacetate activity with no significant gain in pyruvate activity ( Figure 4 ) . Like other MDHs , the apicomplexan MDHs have an Arg at position 102 that is important for oxaloacetate recognition; in the modern apicomplexan LDHs position 102 is a Lys . The Arg102Lys mutation may be necessary to eliminate oxaloacetate activity and increase pyruvate activity . Therefore , we also mutated Arg102 to Lys in the PfMDH chimera ( PfMDH-R102K-INS ) . However , this mutation reduced activity towards oxaloacetate by another 100-fold , with no increase in pyruvate activity ( Figure 4 ) . 10 . 7554/eLife . 02304 . 014Figure 4 . Specificity switching in apicomplexan M/LDHs . Blue horizontal bars ( left ) quantify activity towards oxaloacetate; vermilion horizontal bars ( right ) quantify activity towards pyruvate . Activity is measured as log10 ( kcat/KM ) , where kcat/KM is in units of s-1M−1 . Error bars are shown as small black brackets and represent 1 SD of the mean from triplicate measurements . INS refers to the presence of the six-residue insertion from PfLDH , DEL refers to the removal of the six-residue insertion . Relative specificity ( RS ) is the ratio of kcat/KM for pyruvate vs oxaloacetate , with positive log10 ( RS ) representing a preference for pyruvate and negative log10 ( RS ) representing a preference for oxaloacetate . All logarithms are base 10 . DOI: http://dx . doi . org/10 . 7554/eLife . 02304 . 01410 . 7554/eLife . 02304 . 015Figure 4—source data 1 . Kinetic parameters for modern constructs . DOI: http://dx . doi . org/10 . 7554/eLife . 02304 . 015 Alternatively , it may be possible to revert a modern apicomplexan LDH to MDH-like specificity by deleting its six-residue loop insertion . To test this hypothesis we removed the insertion from PfLDH and from the Toxoplasma gondii ( Tg ) LDH2 ( constructs PfLDH-DEL and TgLDH2-DEL ) . However , deleting the insertion from the modern LDHs abolishes pyruvate activity with no significant gain of oxaloacetate activity ( Figure 4 ) . Both of these deletion mutants retain a Lys at position 102 , but a specific MDH likely requires an Arg at position 102 . Mutating Lys102 to Arg in PfLDH-DEL results in a two order-of-magnitude gain in oxaloacetate activity ( Figure 4 ) . However , this mutant fails to recapitulate the level of oxaloacetate activity seen in modern apicomplexan MDHs . In the modern enzymes , substrate specificity cannot be switched with mutations involving the loop insert and position 102 , indicating that additional residues govern substrate preference . The apicomplexan LDH and MDH phylogeny strongly suggests that after ( or coincident with ) the crucial gene duplication event , the nascent LDH branch gained pyruvate activity due to the six-residue insertion in the specificity loop . This presents a conundrum , as our mutation trials in the modern enzymes failed to recapitulate the historical swap in specificity . However , the modern apicomplexan LDH and MDH enzymes differ by over 200 residues in addition to the loop insert and Arg102Lys , differences that have accumulated in the descendants of the ancestral MDH and LDH . Any of these differences may detrimentally affect the ability to switch substrate specificity with the insertion in the modern enzymes . We therefore reasoned that the ancestral background may be necessary for swapping specificity with the loop insertion . To test this , we reconstructed and characterized four key ancestral enzymes: AncMDH1 , the ancestral protein that was transferred from α-proteobacteria to the archaic Apicomplexa , AncMDH2 , the last common ancestor of all apicomplexan MDHs and LDHs , found at the critical duplication event , AncMDH3 , the last common ancestor of all modern apicomplexan MDHs , and AncLDH , the last common ancestor of modern apicomplexan LDHs ( Figure 3B ) . All four ancestral proteins are highly active in steady state kinetic assays , with substrate preferences and catalytic efficiencies that are similar to their modern apicomplexan descendants ( Figure 5 ) , despite sharing only 49–71% sequence identity with the modern apicomplexan proteins ( Figure 5—figure supplement 1 ) . AncMDH1 , AncMDH2 , and AncMDH3 are highly specific MDHs with negligible pyruvate activity , having even greater activity towards oxaloacetate than modern Plasmodium and Cryptosporidium MDHs ( Figure 5 ) . AncLDH is a highly active and specific LDH , with very low activity towards oxaloacetate ( Figure 5 ) . 10 . 7554/eLife . 02304 . 016Figure 5 . Evolution of novel LDHs in Apicomplexa . The activities of ancestral and modern apicomplexan M/LDHs are plotted on the corresponding nodes of the protein phylogeny . Nodes are numbered as in Figure 3B . The y-axis of the bar graphs is log ( kcat/KM ) , with oxaloacetate in blue and pyruvate in vermilion . RbMDH is a representative α-proteobacterial MDH from Rickettsia bellii . T . gondii has two LDH proteins ( TgLDH1 and TgLDH2 ) , each expressed at different stages of the life cycle ( Dando et al . , 2001 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 02304 . 01610 . 7554/eLife . 02304 . 017Figure 5—source data 1 . Kinetic parameters for ancestral/modern phylogeny . DOI: http://dx . doi . org/10 . 7554/eLife . 02304 . 01710 . 7554/eLife . 02304 . 018Figure 5—figure supplement 1 . Sequence identity of ancestral and modern proteins . DOI: http://dx . doi . org/10 . 7554/eLife . 02304 . 018 AncLDH differs from AncMDH2 by 66 residues , including the six-residue insertion and Arg102Lys . We investigated the evolutionary trajectory from AncMDH2 to AncLDH by characterizing three different mutations in the AncMDH2 background: the addition of AncLDH's six-residue insertion to the AncMDH2 specificity loop , Arg102Lys , which assesses the effect of changing the canonical specificity residue , and the remaining 59 residues that separate AncLDH from AncMDH2 , simultaneously changed to their AncLDH identities . Incorporating the loop insertion into AncMDH2 confers significant pyruvate activity with minimal effect on oxaloacetate activity , resulting in a highly active , bifunctional enzyme ( AncMDH2-INS , Figure 6 ) . In contrast , the Arg102Lys mutation in the AncMDH2 background ( AncMDH2-R102K , Figure 6 ) reduces oxaloacetate activity by more than a 100-fold , with no increase in pyruvate activity . The 59 mutations in the AncMDH2 background have a minimal effect on the activity towards both substrates ( AncMDH2-59Mut , Figure 6 ) . Note that the AncMDH2-59Mut construct is equivalent to a modified AncLDH construct with the Lys102Arg mutation and the insertion deleted from the loop . Therefore , only two changes—Lys102Arg and the loop deletion—are sufficient to convert the AncLDH construct to a highly active and specific MDH . 10 . 7554/eLife . 02304 . 019Figure 6 . Specificity switching in ancestral MDH2 . INS refers to the reconstructed six-residue insertion from AncLDH . 59Mut is described in the text . Relative specificity ( RS ) is described in legend of Figure 4 . DOI: http://dx . doi . org/10 . 7554/eLife . 02304 . 01910 . 7554/eLife . 02304 . 020Figure 6—source data 1 . Kinetic parameters for ancestral specificity switch mutants . DOI: http://dx . doi . org/10 . 7554/eLife . 02304 . 02010 . 7554/eLife . 02304 . 021Figure 6—source data 2 . Source data for figure supplement 7 . Kinetic parameters for alternative ancestral proteins . DOI: http://dx . doi . org/10 . 7554/eLife . 02304 . 02110 . 7554/eLife . 02304 . 022Figure 6—figure supplement 1 . Histogram of ancestral reconstructions . Reconstructed residues binned according to posterior probability ( PP ) of the predicted residue . DOI: http://dx . doi . org/10 . 7554/eLife . 02304 . 02210 . 7554/eLife . 02304 . 023Figure 6—figure supplement 2 . Histogram of ancestral reconstructions . Reconstructed residues binned according to posterior probability ( PP ) of the predicted residue . DOI: http://dx . doi . org/10 . 7554/eLife . 02304 . 02310 . 7554/eLife . 02304 . 024Figure 6—figure supplement 3 . Histogram of ancestral reconstructions . Reconstructed residues binned according to posterior probability ( PP ) of the predicted residue . DOI: http://dx . doi . org/10 . 7554/eLife . 02304 . 02410 . 7554/eLife . 02304 . 025Figure 6—figure supplement 4 . Histogram of ancestral reconstructions . Reconstructed residues binned according to posterior probability ( PP ) of the predicted residue . DOI: http://dx . doi . org/10 . 7554/eLife . 02304 . 02510 . 7554/eLife . 02304 . 026Figure 6—figure supplement 5 . Histogram of ancestral reconstructions . Reconstructed residues binned according to posterior probability ( PP ) of the predicted residue . DOI: http://dx . doi . org/10 . 7554/eLife . 02304 . 02610 . 7554/eLife . 02304 . 027Figure 6—figure supplement 6 . Histogram of ancestral reconstructions . Reconstructed residues binned according to posterior probability ( PP ) of the predicted residue . DOI: http://dx . doi . org/10 . 7554/eLife . 02304 . 02710 . 7554/eLife . 02304 . 028Figure 6—figure supplement 7 . Alternative ancestral enzymes . INS refers to the reconstructed six amino acid insertion from AncLDH* . 58Mut refers to remaining residue differences between AncMDH* and AncLDH* that are not R102K or INS . Relative specificity ( RS ) is described in legend of Figure 4 . DOI: http://dx . doi . org/10 . 7554/eLife . 02304 . 028 Combinations of these mutations confirm that the insertion is primarily responsible for the evolution of pyruvate activity . Adding the 59 mutations to AncMDH2-INS ( resulting in a construct that differs from AncLDH by only one residue ) has little additional effect ( AncMDH2-INS-59Mut , Figure 6 ) . Surprisingly , the combination of Arg102Lys and the 59 mutations , a construct that differs from AncLDH by just the six-residue insertion , yields a crippled MDH enzyme with 1000-fold less oxaloacetate activity than AncMDH2 ( AncMDH2-R102K-59Mut , Figure 6 ) . However , the combination of Arg102Lys and the loop insertion in the AncMDH2 background is sufficient to confer pyruvate activity and specificity comparable to AncLDH ( AncMDH2-INS-R102K , Figure 6 ) . Ancestral sequence reconstruction is a difficult statistical problem that strongly relies on evolutionary assumptions , which may be unrealistic , and on available sequence data , which is inherently incomplete . The likelihood and Bayesian ancestral reconstruction methodology that we use produces the most probable ancestral sequence given certain evolutionary model assumptions , along with a posterior probability for alternative amino acids at each position ( Figure 6—figure supplements 1–6 ) . Ambiguous residues are generally associated with positions of low conservation and presumably less functional importance . The reconstructed AncMDH2 and AncLDH sequences have 31 and 48 ambiguous positions , respectively , all of which are located outside of the ‘first active site shell’ ( defined as within 6 Å of the substrate ) . In order to verify that these sequence ambiguities do not affect our kinetic results , alternative ancestral sequences were reconstructed and assayed . We tested the robustness of our ancestral proteins by constructing alternative ancestors based on perturbed sequence data , evolutionary assumptions , and phylogenetic methodology . Both phylogenies give very similar relationships , and Figure 2B summarizes both equally well . The alternative AncMDH2 ( AncMDH2* ) differs from AncMDH2 by 27 residues; the alternative AncLDH ( AncLDH* ) differs from AncLDH by 19 residues . The alternative ancestral reconstructions behave very similar to the prior reconstructions . AncMDH2* is a strict MDH , and AncLDH* is a strict LDH ( Figure 6 , Figure 6—figure supplement 7 ) . Addition of the six-residue insertion from AncLDH* to AncMDH2* confers pyruvate specificity without adversely affecting oxaloacetate activity ( AncMDH2*-INS , Figure 6—figure supplement 7 ) . In the AncMDH2* background , mutating Arg102 to Lys together with the 58 mutations from AncLDH* yields a poor enzyme with little pyruvate activity ( AncMDH2*-R102K-58Mut ) . The kinetic behavior of these AncMDH2* constructs closely matches those seen with the corresponding AncMDH2 constructs ( AncMDH2 , AncMDH2-INS , and AncMDH2-R102K-59Mut , Figure 6 ) . In order to understand the structural changes during evolution that shifted the enzymatic substrate specificity of the apicomplexan dehydrogenases , we determined the high-resolution crystal structures of three ancestral proteins bracketing the key duplication event: AncMDH2 , AncLDH* , and AncMDH2-INS . Each protein was crystallized with multiple substrates or ligands ( lactate , malate , and oxamate inhibitor ) and with NADH ( resolution ranging from 1 . 35 Å to 2 . 05 Å ) . Unfortunately , in all crystals with malate , the malate spontaneously converted to pyruvate and/or lactate via redox reactions and decarboxylation . In the crystals constructed with AncMDH2 and lactate , a phosphate was seen in the active site rather than lactate . In the following , therefore , the descriptions of the models are primarily based on enzymes crystallized with oxamate inhibitor or lactate , which are highly similar . All three ancestral proteins adopt the same overall fold and conformation as the modern , descendant enzymes . In particular , the ancestral active sites and specificity loops are highly similar to their modern counterparts . The AncMDH2 structure superposes closely with the modern CpMDH structure ( Vedadi et al . , 2007 ) ( ∼0 . 6 Å RMSD for the loop-closed states ) , although differing at ∼119 residue positions ( 62% sequence identity , Figure 7A ) . In the modern and ancestral MDHs , all residues within the first shell of the active sites ( within 6 Å of the substrate ) are identical , and the active site conformations are correspondingly highly similar ( Figure 7B ) . The first shell active site residues comprise Arg102 , Arg109 , Leu112 , Asn140 , Leu167 , Asp168 , Arg171 , His195 , Met199 , Gly236 , Gly237 , Ile239 , Val240 , Ser245 , Ala246 , and Pro250 . 10 . 7554/eLife . 02304 . 029Figure 7 . Ancestral and modern dehydrogenase structures . ( A ) Superposition of CpMDH and AncMDH2 . Superposition of AncMDH2 structure ( blue , 4plw , chain C ) and CpMDH ( aquamarine , 2hjr , chain A ) . Ligands from AncMDH2 are shown in gray; ligands from CpMDH are in white . ( B ) Active site detail of Cp MDH and AncMDH2 . Side chains of catalytic residues highlighted as sticks . ( C ) Superposition of apicomplexan LDHs and AncLDH* . Superposition of AncLDH* structure ( vermilion , 4plg , chain A ) and four apicomplexan LDHs ( deep olive , PfLDH , 1t2d , chain A , Plasmodium berghei ( Pb ) LDH , 1oc4 , chain B , TgLDH1 , 1pzh , chain A , TgLDH2 , 1sow , chain B ) . Ligands from AncLDH* are shown in gray , ligands from apicomplexan LDHs are in white . The ‘opposing loop’ and residues 236 and 246 ( discussed in text ) are highlighted in cyan . ( D ) Active site detail of apicomplexan LDHs and AncLDH* . Side chains of catalytic residues highlighted as sticks . ( E ) Superposition of ancestral dehydrogenases . Superposition of AncMDH2 ( blue , 4plw , chain C ) , AncLDH* ( vermilion , 4plg , chain A ) , and AncMDH2-INS ( magenta , 4ply , chain F and 4plv , chain B ) . Ligands are shown in gray . ( F ) Active site detail of ancestral dehydrogenases . Side chains of catalytic residues highlighted as sticks . DOI: http://dx . doi . org/10 . 7554/eLife . 02304 . 02910 . 7554/eLife . 02304 . 030Figure 7—figure supplement 1 . Crystallographic statistics table for AncLDH* , AncMDH2 , AncMDH2-INS , and PfLDH-W107fA structures . DOI: http://dx . doi . org/10 . 7554/eLife . 02304 . 030 Compared to the modern MDH , only slight differences are seen in the substrate loop backbone and the positioning of the Arg102 and Arg109 sidechains , which are the only residues from the specificity loop that directly interact with the substrate . However , these modest conformational differences are largely within coordinate error , as the loop residues have some of the highest B-factors in the structures . Furthermore , AncMDH2 was crystallized with lactate/oxamate and NADH , while CpMDH was crystallized with citrate and ADPR ( an NADH analog lacking the nicotinamide ring ) . Citrate is roughly three times larger than lactate and has likely affected the position of substrate loop in the CpMDH structure . We also crystallized AncMDH2-INS , a bifunctional AncMDH2 construct with the six-residue specificity loop insertion . This AncMDH2-INS construct represents a possible intermediate along the evolutionary trajectory between the MDH duplication event and the ancestral apicomplexan LDH . AncMDH2-INS was successfully co-crystallized with both oxamate/NADH and lactate/NADH . In the loop-closed state , the specificity loop adopts an LDH-like confirmation with Trp107f occupying the specificity position and Arg102 oriented into solution , similar to how Lys102 is positioned in the modern and ancestral LDH structures ( Figure 7F ) . The lactate and oxamate structures are highly similar ( RMSD ∼0 . 2 Å ) , and the active site architectures are nearly indistinguishable . The three ancestral proteins , AncMDH2 , AncLDH* , and AncMDH2-INS , are all highly similar ( RMSD 1 . 20 Å ) with the main structural differences found in the conformation of the specificity loop ( Figure 7E , RMSD ∼0 . 9 Å excluding residues in the specificity loop ) . Otherwise the first shell active site residues are identical between AncMDH2-INS and AncLDH* , and the conformations of the active sites are correspondingly similar ( Figure 7F ) . Given the known importance of position 102 , the ‘specificity residue’ , in substrate recognition , we wondered whether different residues at position 102 could confer pyruvate activity . Position 102 in fact differs in the four convergent LDH families: Gln in canonical LDHs ( Wilks et al . , 1988 ) , Lys in the apicomplexan LDHs , Gly in Cryptosporidium LDHs ( Madern et al . , 2004 ) , and Leu in trichomonad LDHs ( Wu and Fiser , 1999 ) . Could the ancestral apicomplexan MDH have evolved pyruvate specificity by any of these alternative routes ? To answer this question , we evaluated the potential of these different amino acids at the 102 position to confer pyruvate specificity in the AncMDH2 background . Each mutation increases pyruvate activity , but none result in a highly specific LDH . The canonical mutation ( Arg102Gln ) results in the largest gain in pyruvate activity ( 2800-fold ) and the smallest loss of oxaloacetate activity ( 2500-fold ) ( Figure 8 ) . Additionally , we tested whether the full six amino acid insertion was required to confer pyruvate specificity in AncMDH2 or if simply mutating Arg102 to Trp was sufficient . The Arg102Trp mutation all but abolishes activity towards both substrates , indicating that the loop insertion was necessary to switch the specificity residue ( Figure 8 ) . 10 . 7554/eLife . 02304 . 031Figure 8 . Alternative LDH mutations in AncMDH2 . Relative specificity ( RS ) is described in legend of Figure 4 . DOI: http://dx . doi . org/10 . 7554/eLife . 02304 . 03110 . 7554/eLife . 02304 . 032Figure 8—source data 1 . Kinetic parameters for specificity residue mutants . DOI: http://dx . doi . org/10 . 7554/eLife . 02304 . 032
Substrate recognition in the canonical MDHs and LDHs is thought to be determined by a ‘specificity residue’ in the active site loop at position 102 . All known MDHs have Arg at position 102 , while canonical LDHs have Gln ( Chapman et al . , 1999 ) . In the classic explanation of the molecular mechanism of substrate specificity , residue 102 discriminates between pyruvate and oxaloacetate primarily via charge conservation ( Wilks et al . , 1988 ) . In MDHs , the positively charged Arg interacts with and balances the negatively charged carboxylate of oxaloacetate . If pyruvate were to bind in the active site , loop closure would result in a buried and unbalanced positive charge , which is unfavorable . In canonical LDHs , the neutral Gln interacts with the neutral pyruvate methyl group . Oxaloacetate binding would similarly result in the unfavorable burial of an unbalanced negative charge . In the apicomplexan LDHs , evolution has converged on pyruvate specificity using an alternative molecular mechanism . Residue 102 is not a Gln but a positively charged Lys , similar to Arg102 of MDHs , leading many researchers to wonder why apicomplexan LDHs lack activity towards oxaloacetate ( Gomez et al . , 1997; Dando et al . , 2001; Winter et al . , 2003; Brown et al . , 2004; Kavanagh et al . , 2004; Shoemark et al . , 2007 ) . However , during the evolution of the apicomplexan LDH from the ancestral MDH , the six-residue insertion in the active site loop shifted both the position and identity of the ‘specificity residue’ from Arg102 to Trp107f . Due to the insertion , residue 102 no longer interacts with the substrate and is extruded from the active site . In contrast , the hydrophobic Trp107f packs against the C3 methyl of the pyruvate substrate . Similar to the canonical LDH , oxaloacetate binding would result in an unbalanced and buried negative charge . As a large bulky residue , Trp107f can also occlude binding of the larger oxaloacetate , in which a methylene carboxylate replaces the pyruvate methyl . However , as discussed in detail below , this simplistic explanation is complicated by long-range epistatic interactions . When the six-residue insertion is introduced into the modern apicomplexan MDH , specificity is not switched; both specificity and activity are lost . Similarly , removal of the insertion from the modern apicomplexan LDHs fails to swap specificity and kills the enzymes . Therefore , while Trp107f is necessary for substrate specificity in the apicomplexan enzymes ( as indicated by the alanine scan mutations ) , it is insufficient to confer specificity . The bifunctionality of AncMDH2-INS and AncMDH2-INS-59Mut also presents a conundrum . Why do these constructs have high activity towards both pyruvate and oxaloacetate substrates ? The crystal structure of AncMDH2-INS offers few clues , since the loop insertion , including Trp107f , adopts the same conformation as seen in AncLDH and the modern apicomplexan enzymes . Both the AncMDH2-INS and AncMDH2-INS-59Mut constructs have an Arg at position 102 , like the MDHs . In fact , the bifunctional AncMDH2-INS-59Mut enzyme differs from the strict AncLDH by only a R102K mutation , suggesting that Arg102 is responsible for the oxaloacetate activity of AncMDH2-INS and AncMDH2-INS-59Mut . We speculate that perhaps the enzymes change conformation depending upon the substrate . When using pyruvate , these bifunctional enzymes may adopt an LDH-like conformation in which Trp107f interacts with the substrate ( as seen in the crystal structure ) . On the other hand , when presented with oxaloacetate , perhaps Trp107f flips out of the active site , and Arg102 flips in to interact with substrate in a manner similar to the canonical MDHs . We are currently testing this hypothesis . Our data show that apicomplexan LDHs evolved from a horizontally transferred proteobacterial MDH by a classic neofunctionalization mechanism of gene duplication . Because debasement to a pseudogene is much more likely to occur prior to the evolution of a novel function , neofunctionalization has fallen out of favor as a mechanism for the evolution of novel functions . A variety of alternative specialization models have been proposed that feature a reduced risk of non-functionalization . Though differing in details , all specialization models feature a promiscuous common ancestor of the duplicated proteins . The reconstructed AncMDH2 , which represents the last common ancestor of the apicomplexan MDH and LDHs , is a highly active and specific MDH , preferring oxaloacetate over pyruvate by seven orders of magnitude ( Figure 6 ) . The activity of AncMDH2 towards pyruvate is barely detectable , requiring a high enzyme concentration to quantify . AncMDH2's kcat for pyruvate is 0 . 07 s−1 , with a Km of 20 mM , while the physiological concentration of pyruvate is estimated to be about three orders-of-magnitude lower ( e . g . , ∼50 μM in human erythrocytes [Garrett and Grisham , 2005] , the Plasmodium host during its blood stage ) . Based on these kinetic parameters , each AncMDH2 reduces one pyruvate molecule per hour . While the enzyme can be forced to reduce pyruvate in vitro , this negligible activity is unlikely to have been subjected to selection in vivo . Therefore , the various specialization hypotheses , which require a promiscuous ancestor , are poor models for apicomplexan LDH evolution . Activity towards pyruvate increased by over seven orders of magnitude on the evolutionary lineage between AncMDH2 and AncLDH , indicating neofunctionalization . One of the most favored specialization models is ‘escape from adaptive conflict’ ( EAC ) ( Des Marais and Rausher , 2008 ) . EAC holds that functional specialization is driven by an inability to simultaneously optimize multiple functions on a single protein scaffold . Gene duplication relieves this constraint and allows for the independent optimization of conflicting functions . Although the apicomplexan AncMDH2 is highly specific , promiscuous intermediates did play a role in the functional transition between AncMDH2 and AncLDH . AncMDH2-INS and AncMDH2-INS-59Mut have high levels of MDH and LDH activity in a single protein scaffold ( Figure 6 ) . Both the presence of bifunctional intermediates and the high specificity of AncMDH2 conflict with fundamental predictions of the EAC specialization model . The evolution of apicomplexan LDHs involved strong epistasis that has profoundly influenced the convergent evolution of pyruvate activity . Epistasis refers to interactions between residues that potentiate the effects of a mutation depending on the presence or absence of other residues ( Harms and Thornton , 2010 ) . Epistasis can constrain the order of mutations and the pathways accessible to evolution , and hence it is of great importance in understanding the evolution of novel functions . In the apicomplexan dehydrogenases , the evolutionary mutations that switched specificity from oxaloacetate to pyruvate ( the six-residue insertion and Arg102Lys ) are insufficient to confer pyruvate activity in modern apicomplexan MDHs ( PfMDH-R102K , PfMDH-INS , CpMDH-INS , PfMDH-R102K-INS , Figure 4 ) . However , these mutations are sufficient to confer pyruvate function and specificity in the AncMDH2 background ( AncMDH2-INS , AncMDH2-INS-R102K , Figure 6 ) . Similarly , removal of the insert from the modern LDHs ( PfLDH-DEL and TgLDH2-DEL , Figure 4 ) kills the enzymes , while removal of the insert from the ancestral LDH ( AncMDH2-R102K-59Mut , Figure 6 ) results in a weak MDH . The different effects of these mutations , depending upon the sequence of the rest of the protein , provide direct evidence of epistatic interactions . Why do these historical mutations ‘work’ in the ancestral enzymes , but not in the modern ones ? Epistatic interactions are often mediated by direct physical contact . For example , the active site of the ancestral MDH could have certain residues that the modern MDH lacks , residues that interact with the insertion and allow it to preferentially bind pyruvate . However , the active sites of the ancestral and modern MDHs are identical in sequence and virtually indistinguishable in structure ( Figure 7B ) , as are the active sites of the ancestral and modern LDHs ( Figure 7D ) and the AncMDH2-INS intermediate ( Figure 7F ) . In fact , the active sites of the MDHs and the LDHs are also identical in sequence except for the 102 position , and they are otherwise highly structurally similar . Therefore , residues remote from the active sites necessarily affect the substrate specificity of the enzymes . In principle , these long-range epistatic residue interactions could differentially modify the structure of the active site . Certain residues found in the ancestral MDH , but not in the modern MDH , could position the active site residues so that they allowed the insertion to confer pyruvate activity . In this scenario the active site residues of the ancestral and modern MDHs would be identical , but their conformations would differ due to interactions with residues in other parts of the protein . However , the crystal structures reveal ancestral , intermediate , and modern active sites that are nearly indistinguishable , suggesting that epistasis has modified the protein dynamics or shifted the energy landscape , effects that are largely invisible to static crystal structures . Interestingly , Bacillus subtilis ( Bs ) LDH reverts to an MDH with only a single mutation , Gln102Arg , indicating a lack of complicating epistatic effects ( Wilks et al . , 1988 ) . The kinetics of wild-type BsLDH with pyruvate are comparable to those for the Gln102Arg mutant with oxaloacetate ( e . g . , BsLDH has a kcat/KM for pyruvate of 4 . 2 × 106 M−1 s−1 , and the BsLDH-Q102R mutant has the same kcat/KM for oxaloacetate ) . However , BsLDH likely is an exception in the LDH/MDH superfamily , since the reverse mutation ( Arg102Gln ) fails to switch specificity in MDHs from two other species ( Nicholls et al . , 1992; Cendrin et al . , 1993 ) . In Haloarcula marismortui ( Hm ) MDH , the Arg102Gln mutation switches specificity , but the mutant's kcat/KM for pyruvate is 200-fold less than the wild-type's kcat/KM for oxaloacetate . The Arg102Gln mutation in Escherichia coli ( Ec ) MDH is even less effective , as it converts a highly active MDH to an enzyme with low activity on both substrates ( 10 , 000-fold lower kcat/KM ) . Hence , the strong epistasis observed in apicomplexan LDH and MDHs is likely a general phenomenon within the superfamily . LDH evolved convergently from MDH four separate times in the superfamily , but did the activity evolve by the same mechanism each time ? Each event has resulted in a different change at the specificity residue ( position 102 ) within the catalytic loop . However , the epistatic effects seen in the apicomplexan , H . marismortui , and E . coli dehydrogenases indicate that in general position 102 is not solely responsible for the transition from MDH to LDH . In order for the historical LDH mutations to confer pyruvate specificity , additional residues must be present to provide a permissive background ( Figure 8 ) . Due to the presence of different sets of permissive mutations , LDH activity has evolved from an MDH under epistatic constraints by a different mechanism four separate times . The evolution of AncLDH from AncMDH2 involves a shift in substrate specificity by 12 orders-of-magnitude . Through the characterization of possible evolutionary intermediates , we have found that just two mutations are responsible for the great majority of this switch: the six-residue insertion and the Arg102Lys point mutation . Mutagenesis within the insertion indicates that only a single position , Trp107f , contributes strongly to pyruvate activity and specificity . Both the insertion and Arg102Lys have a large effect on preference for pyruvate vs oxaloacetate , although by differentially affecting activity towards each substrate . Incorporating the six-residue insertion into AncMDH2's substrate loop results in a 12 , 000-fold gain in pyruvate activity with little effect on oxaloacetate activity ( Figure 6 ) . Conversely , mutating Arg102 to Lys reduces oxaloacetate activity by more than 2500-fold , with minimal effect on pyruvate activity ( Figure 6 ) . The apicomplexan LDH six-residue insertion is an exceptionally large gain-of-function mutation: it enhances pyruvate activity by more than four orders of magnitude while barely affecting oxaloacetate activity . In contrast , other well-studied mutations of large effect are often predominantly deleterious towards one function while modestly enhancing another . The textbook example of a gain-of-function mutation is Gln102Arg in BsLDH , which causes a 107-fold change in the enzyme's specificity ( Wilks et al . , 1988 ) . The Gln102Arg mutation reduces pyruvate activity by more than 8000-fold , while enhancing activity towards oxaloacetate by only 1000-fold . Another example is given by E . coli isocitrate dehydrogenase ( IDH ) , where seven mutations are necessary to switch the cofactor specificity from a 7000-fold preference for NADP to a 200-fold preference for NAD ( Chen et al . , 1995 ) . Within this set of mutations , two reduce specificity for NADP by 6000-fold , whereas the rest enhance NAD usage 200-fold . Thus , while mutations can have both deleterious and beneficial effects on different functions , the deleterious effects typically appear greater than enhancement . In previous ancestral sequence reconstruction studies , mutations of large effect are in fact usually loss-of-function rather than gain-of-function ( e . g . , RNaseA [Jermann et al . , 1995] , chymase [Wouters et al . , 2003] , and glucocorticoid receptors [Bridgham et al . , 2006; Ortlund et al . , 2007; Carroll et al . , 2008 , 2011] ) . In these studies , the modern proteins are generally specific for one substrate , whereas the ancestral proteins are promiscuous . Furthermore , the activity of the ancestral protein is comparable to the modern descendants . Therefore , these proteins specialized by accumulating deleterious mutations , with the modern , specialized activity being the ‘last function standing’ . For example , the ancestral glucocorticoid receptor binds three steroid hormones tightly ( EC50 <10 nM for aldosterone , deoxycorticosterone , and cortisol ) , while the modern receptors bind only cortisol ( EC50 ∼ 100 nM ) ( Ortlund et al . , 2007 ) . Seven historical mutations produced the modern cortisol preference by completely eliminating aldosterone and deoxycorticosterone sensitivity yet reducing cortisol sensitivity only 50-fold . In other ancestral reconstruction studies , function-enhancing mutations have relatively minor effects , all less than a 50-fold gain in kcat/KM ( Zhang and Rosenberg , 2002; Voordeckers et al . , 2012; Risso et al . , 2013 ) . During the evolution of the malate and lactate dehydrogenase superfamily , pyruvate activity has converged multiple times despite strong constraints due to epistasis . While epistasis may constrain evolutionary options locally , there are nevertheless multiple ways to ‘skin the cat’ in more distant regions of protein sequence space . The apicomplexan enzymes provide a clear example of neofunctionalization in protein evolution and thereby validate the plausibility of this particular mechanism of gene duplication . Specialization mechanisms may be more common , but the evolution of novel function does not require a promiscuous genesis . The PDB accession codes for the coordinates and structure factor files reported in this paper are 4PLC , 4PLF , 4PLG , 4PLH , 4PLT , 4PLV , 4PLW , 4PLY , and 4PLZ .
Protein sequences used in the phylogenetic analyses were identified through searches of the non-redundant database ( Pruitt et al . , 2009 ) with the BLASTP algorithm ( Altschul et al . , 1990 ) using selected query sequences . All sequences from these searches that returned BLASTP E-values <10−7 were downloaded from NCBI ( www . ncbi . nlm . nih . gov ) . Multiple complete apicomplexan genomes ( Heiges et al . , 2006; Gajria et al . , 2008; Aurrecoechea et al . , 2009 ) were also searched for LDH and MDH homologs in order to fill out the apicomplexan portion of the tree ( using a more lenient significance cutoff of E-values <10−4 ) . Redundant sequences , synthetic constructs , and sequences from PDB files were removed . To reduce phylogenetic complexity , sequences were curated based on character length and pairwise sequence identity within each dataset ( as described below ) . The dataset used for the construction of the non-redundant phylogeny ( Figure 3A ) was generated using four query sequences , UniProt IDs ( UniProt Consortium , 2013 ) : MDHC_HUMAN , LDH_THEP1 , MDHP_YEAST , and LDH6A_HUMAN . Multiple sequences were necessary to generate full coverage , due to the low sequence identity across the superfamily , which can be less than 20% between members . Sequences were removed if their character length was less than 280 or greater than 340 . Limits were chosen to remove truncated/partial sequences and those featuring large insertions or terminal extensions . Sequences greater than 97% identical , determined by pairwise alignment within the dataset , were also removed . This level of identity provides a high level of detail within the tree while accelerating computational time by removing redundant taxa . The final dataset contains 1844 taxa . Residue numbering in the text is based on the dogfish LDH convention ( Eventoff et al . , 1977 ) for consistency with previous work . A multiple sequence alignment of this dataset was generated using the program MUSCLE ( Edgar , 2004 ) . A maximum likelihood ( ML ) phylogenetic tree was inferred with PhyML 3 . 0 ( Guindon et al . , 2010 ) using the LG substitution matrix ( Le and Gascuel , 2008 ) and estimating the gamma parameter ( 12 categories ) and empirical amino acid frequencies . The starting tree was generated by Neighbor-Joining ( BIONJ ) and searched by Nearest Neighbor Interchange ( NNI ) ; tree topology , branch lengths , and rate parameters were optimized . Branch supports were estimated with the approximate likelihood ratio test ( aLRT ) , as implemented in PhyML , represented as either the raw aLRT statistic ( roughly >8 is considered highly significant ) or the confidence level that the clade is correct ( Anisimova and Gascuel , 2006 ) . The outgroup for rooting the L/MDH phylogeny was identified through a profile analysis of the Rossmann fold ( Rao and Rossmann , 1973 ) , based on a method used for OB folds and SH3 domains ( Theobald and Wuttke , 2005 ) . All structurally characterized Rossmann folds with 40% or less sequence identity were identified from ASTRAL SCOP 1 . 73 protein domain sequence database ( Chandonia et al . , 2004 ) . Each of the 193 domains identified was searched against the SwissProt database ( Boeckmann et al . , 2003 ) using BLASTP . A multiple sequence alignment for each query and SwissProt sequences with BLASTP E-values <10−10 was created using MUSCLE . Each alignment was cropped to the limits of the original query . COMPASS ( Sadreyev et al . , 2003 ) was then used to generate an all-against-all scoring matrix for the 193 multiple sequence alignments . The E-values generated by COMPASS were converted to evolutionary distances as described in Theobald and Wuttke ( 2005 ) . A weighted least-squares phylogenetic analysis of the distance matrix was performed using PAUP ( Swofford , 2003 ) . First order taxon jackknifing ( Lanyon , 1985; Siddall , 1995 ) was used to determine the robustness of tree topology , with a consensus tree calculated from all analyses . Rossmann fold domains from α- and β-glucosidases and aspartate dehydrogenases ( AspDH ) were identified from the profile–profile analysis as grouping with the Rossmann fold domain from L/MDHs . An L/MDH dataset was constructed for use with the outgroup to create a rooted phylogeny . This dataset was generated by querying four sequences , UniProt IDs: MDHC_HUMAN , LDH_THEP1 , MDHP_YEAST , and LDH6A_HUMAN , against the SwissProt database using BLASTP . All sequences from these searches that returned BLASTP E-values <10−7 were downloaded from NCBI ( www . ncbi . nlm . nih . gov ) . Redundant sequences , synthetic constructs , and sequences from PDB files were removed . Also , four taxa identified as ubiquitin-conjugating enzymes were removed due to sequence length . This SwissProt L/MDH dataset contained 595 taxa . An outgroup dataset was constructed by querying three sequences , UniProt IDs: LICH_BACSU , AGAL_THEMA , and ASPD_THEMA , against the SwissProt database using BLASTP . All sequences from these searches that returned BLASTP E-values <10−7 were downloaded from NCBI ( www . ncbi . nlm . nih . gov ) . Redundant sequences , synthetic constructs , and sequences from PDB files were removed . The outgroup dataset contained 62 taxa . The SwissProt LDH , MDH , AspDH , and glucosidase datasets were combined and a multiple sequence alignment was generated using the program MUSCLE . The C-terminal domain of the glucosidases and AspDHs were removed from the MUSCLE alignment . A ML phylogenetic tree was inferred from the alignment with PhyML using the LG substitution matrix ( Whelan and Goldman , 2001 ) with the gamma parameter estimated over 10 categories , no invariant sites , and estimating empirical amino acid frequencies . The initial tree was obtained by BIONJ and searched by NNI; tree topology , branch lengths , and rate parameters were optimized . Robustness of root positioning was evaluated with two truncated alignments , one with the LDH and ‘LDH-like’ MDH sequences removed and the other with the cytosolic and mitochondrial MDH sequences removed . Truncated alignments were input to PhyML for phylogenetic analysis using the parameters described above . The dataset for the alternative phylogeny ( used in reconstructing alternative ancestors ) is smaller and focused on apicomplexan taxa . It was generated by BLASTP searches with four query sequences , UniProt IDs: MDHC_PIG , Q76NM3_PLAF7 , C6KT25_PLAF , and MDH_WOLPM for full coverage of the superfamily . Sequences were removed if their length was less than 290 or greater than 340 . The dataset was culled to 60% identity , but the apicomplexan clade was filled back to 97% identity to gain resolution within the clade of interest . The final dataset contained 277 taxa . A multiple sequence alignment of this dataset was produced using the program MUSCLE . The ML tree was inferred with PhyML 3 . 0 using the LG substitution matrix and estimating the gamma parameter ( 12 categories ) and empirical amino acid frequencies . The starting tree was generated by Neighbor-Joining ( BIONJ ) and searched by Nearest Neighbor Interchange ( NNI ) ; tree topology , branch lengths , and rate parameters were optimized . Sequences at internal nodes in phylogenies were inferred using the codeml program from the PAML software package ( Yang , 2007 ) . Posterior amino acid probabilities at each site were calculated using the LG substitution matrix , given the ML tree generated by PhyML . The initial ancestral reconstruction assumed the background amino acid frequencies implicit in the LG matrix , while the alternative reconstruction estimated background frequencies from the sequence alignment of the alternative dataset . N-/C-termini of ancestral sequences were modified manually to match those of the closest modern sequence ( determined by branch length ) . Escherichia coli codon-optimized coding sequences were constructed for the Plasmodium falciparum MDH ( gi#: 86171227 ) , Cryptosporidium parvum MDH ( gi#: 32765705 ) , Toxoplasma gondii LDH1 ( gi#: 237837615 ) , Toxoplasma gondii LDH2 ( gi#: 2497625 ) , Rickettsia bellii MDH ( gi#: 91205459 ) , and ancestrally inferred protein sequences . These coding sequences were synthesized and subcloned into pET-24a , bypassing the N-terminal T7-tag but using the C-terminal 6xHis-tag . PfLDH ( gi#: 124513266 ) with six His residues added to the C-terminus was synthesized and subcloned into pET-11b . All gene synthesis and subcloning was performed by Genscript ( Piscataway , NJ ) . All point mutations were made using the QuikChange Lightning kit from Agilent ( Santa Clara , CA ) and synthesized primers from IDT ( Coralville , IA ) . Plasmids were transformed in BL21 DE3 ( pLysS ) E . coli cells ( Invitrogen , Grand Island , NY ) for expression . Cells were grown at 37°C with 225 rpm agitation in 2xYT media supplemented with 30 mM potassium phosphate , pH 7 . 8 and 0 . 1% ( wt/vol ) glucose . Once cultures reached an OD600 between 0 . 5–0 . 8 , cells were induced with 0 . 5 mM IPTG for 4 hr . Cells were collected by centrifugation at 10 , 000×g for 15 min and stored at −80°C . Cell pellets were thawed on ice , releasing lysozyme produced by the pLysS plasmid from within the cells , and resuspended in 15 ml lysis buffer ( 50 mM NaH2PO4 , pH 8 . 0 , 300 mM NaCl , 10 mM Imidazole ) with 375 units of Pierce Universal Nuclease ( Thermo Scientific , Rockford , IL ) per 1 . 5 l of culture . Once homogeneously resuspended , lysate was sonicated on ice at 35% amplitude ( 30 s ON , 20 s OFF , 2 min total ) . Insoluble cell debris was separated by centrifugation at 18 , 000×g for 20 min . Proteins were purified by nickel affinity chromatography . Clarified lysate was applied to a 5 ml HisTrap FF column ( GE Healthcare , Piscataway , NJ ) and eluted via an imidazole gradient from 10 mM to 500 mM on an AKTA Prime ( GE Healthcare , Piscataway , NJ ) . Fractions were analyzed by SDS-PAGE , pooled , and concentrated using Amicon Ultracel-10 K centrifugal filters ( Millipore , Billerica , MA ) . Finally , proteins were desalted into 50 mM Tris , pH 7 . 4 , 100 mM NaCl , 0 . 1 mM EDTA and 0 . 01% azide by either PD10 column ( GE Healthcare , Piscataway , NJ ) or gel filtration over a HiPrep 16/60 Sephacryl S-200 HR column ( GE Healthcare , Piscataway , NJ ) on an AKTA Purifier ( GE Healthcare , Piscataway , NJ ) . Enzyme concentrations were determined by absorbance at 280 nm , using extinction coefficients and molecular weights calculated by ExPASy's ProtParam tool ( http://web . expasy . org/protparam/ ) . Enzymatic reduction of pyruvate and oxaloacetate was monitored at 25°C by following the decrease in absorbance at 340 nm due to NADH oxidation on a Cary 100 Bio ( Agilent , Santa Clara , CA ) in 50 mM Tris , pH 7 . 5 , 50 mM KCl . All substrates were purchased from Sigma-Aldrich ( St . Louis , MO ) . NADH concentration was held constant at 200 μM while pyruvate/oxaloacetate concentrations were titrated . Enzyme concentrations ranged from 0 . 28 nM to 2 . 8 μM , depending on enzyme activity for a particular substrate . All experiments used 1-cm path-length quartz cuvettes with 500 μl final volume of reaction mixture . Kinetic parameters were estimated by chi-squared fitting to either the Michaelis-Menton equation ( v/[E]t = kcat [S]/ ( KM + [S] ) ) or a substrate inhibition equation ( v/[E]t = kcat [S]/ ( KM + [S] + [S]2/Ki ) ) using the KaleidaGraph software . Three datasets were fit using a modified substrate inhibition equation with KM = Ki for identifiability and to prevent the Ki being less than KM . These datasets were: AncMDH2-INS-59Mut oxaloacetate and AncMDH2-R102Q for both oxaloacetate and pyruvate . Kinetic constants kcat , KM , and kcat/KM are consistently reported in units of s−1 , M , and s-1M−1 , respectively . Aqueous oxaloacetate spontaneously decarboxylates to pyruvate at 25°C and neutral pH at a rate of ∼3 × 10−5 s−1 ( approximately 10% per hr ) ( Wolfenden et al . , 2011 ) . As a result , oxaloacetate preparations contain appreciable pyruvate contamination ( approximately 1–3% from Sigma-Aldrich , depending on batch ) and must be handled with care . All oxaloacetate stock solutions were made fresh before each assay and kept on ice to keep decarboxylation to a minimum . For enzymes with low pyruvate activity , the oxaloacetate decarboxylation has a negligible affect on measured rates . However , enzymes with appreciable pyruvate activity can display an apparent , artifactual oxaloacetate activity that is due to pyruvate contamination ( Parker and Holbrook , 1981; Wilks et al . , 1988; Shoemark et al . , 2007 ) . In this work , seven such proteins are PfLDH , PfLDH-K102R , TgLDH1 , TgLDH2 , AncLDH , AncLDH* , and AncMDH2-INS-R102K . For these proteins , oxaloacetate activity was assayed at high enzyme concentration ( 600 nM–1 μM ) , resulting in a biphasic ΔA340 trace with an initial burst in which pyruvate is rapidly consumed followed by a slower linear phase representing oxaloacetate reduction . The post-burst ( slow ) phase of the ΔA340 trace was used to quantify the oxaloacetate catalytic rate ( Parker and Holbrook , 1981; Wilks et al . , 1988 ) . This procedure controls for the standing pyruvate contamination but does not account for the relatively slow spontaneous decarboxylation during the assay . Hence , the oxaloacetate kcat/Km values for the seven enzymes with high pyruvate activity should be considered upper limits on the true oxaloacetate activity . The low or negligible oxaloacetate activities of these seven enzymes were further confirmed by undetectable malate/NAD+ reactions in spectroscopic steady state enzyme assays , and the absence of malate product as determined from 1D proton NMR ( 3 μM enzyme , 5 mM oxaloacetate , 5 mM NADH in NaCl/Pi/D2O pH 7 . 5 over four hour reaction ) ( Shoemark et al . , 2007 ) . Crystallization trials were conducted by hanging-drop vapor-diffusion at room temperature using Crystal Screen and Crystal Screen 2 from Hampton Research ( Aliso Viejo , CA ) . Drops consisting of 2 μl reservoir solution and 2 μl protein stock were equilibrated against 1 ml of reservoir solution . Crystals of the ancestral proteins were identified from condition #43 of Crystal Screen ( 30% ( wt/vol ) polyethylene glycol 1500 ) and further refined by adding 0 . 1 M sodium HEPES . Crystals of the ternary complexes were grown at room temperature by hanging-drop vapor-diffusion with 4 μl drops of 1:1 precipitating buffer:protein . AncMDH2 ( 25 mg/ml ) was co-crystallized with 2 mM oxamate/NADH in 35% ( wt/vol ) PEG-1500 , 0 . 1 M sodium HEPES , pH 7 . 5 and with 2 mM L-lactate/NADH in 30% ( wt/vol ) PEG-1500 , 0 . 1 M sodium HEPES , pH 7 . 3 . AncMDH2-INS ( 18 mg/ml ) was co-crystallized with 1 mM oxamate/NADH and 1 mM L-lactate/NADH in 25% ( wt/vol ) PEG-1500 , 0 . 1 M sodium HEPES , pH 8 . 1 . AncLDH* ( 20 mg/ml ) was co-crystallized with 2 mM oxamate/NADH and 2 mM L-lactate/NADH in 20% ( wt/vol ) PEG-1500 , 0 . 1 M sodium HEPES , pH 7 . 5 . Pf LDH_W107fA ( 20 mg/ml ) was co-crystallized with 1 . 2 mM oxamate/2 mM NADH in 22% ( wt/vol ) PEG-1000 . All crystals were cryoprotected with a 30% ( wt/vol ) dextrose solution ( 15 mg dextrose dissolved in 50 μl reservoir solution ) . Crystals were harvested from the drop , soaked in 15% ( wt/vol ) dextrose solution for 3 min , transferred to the 30% solution , and flash-frozen immediately in liquid N2 . Diffraction datasets were collected at the SIBYLS beamline ( 12 . 3 . 1 , Lawrence Berkeley National Laboratory , Berkeley , CA ) . All datasets were indexed , integrated , and scaled with XDS/XSCALE ( Xds , 2010 ) . Datasets included all reflections that were significant according to the CC ( 1/2 ) criterion ( flagged as ‘*’ in the XDS output ) , which typically extended to much weaker data than the conventional 2 sigma cutoff ( Karplus and Diederichs , 2012; Diederichs and Karplus , 2013 ) . The resolution of the data is defined as the resolution bin where CC ( 1/2 ) = 0 . 5 . Structures were solved by molecular replacement using AutoMR in PHENIX ( Adams et al . , 2010 ) . Homology models for the AncMDH2 and AncLDH* datasets were generated by the Phyre2 server ( Kelley and Sternberg , 2009 ) . The AncMDH2 homology model was based on the structure for Cryptosporidium parvum MDH ( 2hjr , 62% sequence identity , Vedadi et al . , 2007 ) , while the model for AncLDH* was based on the Toxoplasma gondii LDH1 structure ( 1pzf , 65% sequence identity , Kavanagh et al . , 2004 ) . AncMDH2-INS datasets were solved using the AncMDH2 structure as the model . Pf LDH_W107fA dataset was solved using the P . falciparum LDH structure ( 1t2d ) structure as a model . All models were improved by rounds of manual building in Coot ( Emsley et al . , 2010 ) and refinement by phenix . refine in PHENIX . Model quality of all structures was validated with MolProbity ( Davis et al . , 2007; Chen et al . , 2010 ) in PHENIX . All structural alignments were generated using THESEUS ( Theobald and Wuttke , 2008 ) . Structure images were rendered with PyMOL . In all structure models , the enzyme is generally found in either of two states , the loop-open state or the loop-closed state , and often both states are observed in different monomers ( chains ) in the asymmetric unit . Loop-open states are generally more disordered ( with higher B-factors ) and sometimes the chain could not be traced for the loop . In loop-open states the substrate typically also has weak electron density and cannot be fit reliably . In loop-closed states the substrate generally has strong electron density , but complications arose for proteins co-crystallized with lactate and malate . NADH will eventually oxidize to NAD+ spontaneously in the crystals and during data collection due to oxidative radiation damage . Malate can also oxidize to oxaloacetate in the crystals , and oxaloacetate is expected to spontaneously decarboxylate to pyruvate relatively quickly ( over the multi-week time frame of crystallization ) . Additionally , pyruvate spontaneously reacts with NAD+ to form a covalent pyruvate-NAD adduct ( White et al . , 1976 ) . Although we added NADH to the crystallization mixture , we cannot reliably determine whether NADH , NAD+ , or a mixture of the two is found in the crystals , and we have variable evidence for the pyruvate-NAD adduct . In both of the structures crystallized with malate ( 4plc and 4ply ) , the substrate electron density and loop conformation indicated that the malate had converted to pyruvate and/or lactate . Therefore , for the enzymes crystallized with lactate and malate , the fine details of substrate conformation and identity in the models should be viewed with skepticism , as we are fairly confident we have an unresolvable mixture of pyruvate , lactate , and pyruvate-NAD adduct in the active site , all of which will produce similar electron density . These concerns with substrate do not apply to the oxamate models ( although the uncertainty in cofactor state still remains ) . Finally , in the 4plw model ( AncMDH2 crystallized with lactate ) , clear phosphates were seen in the active site instead of lactate , presumably carried through from the initial purification step . | How are new genes created ? Most of the mutations in the genome of an organism place the organism at some sort of disadvantage , but a small number confer an advantage . The beneficial changes are usually retained by subsequent generations and can ultimately lead to the creation of new genes . An example is the gene that encodes an enzyme called lactate dehydrogenase ( LDH ) . This enzyme is involved in anaerobic respiration , the process that allows organisms to produce energy without using oxygen . The LDH enzyme is found in many species of animals and parasites , including those that spread malaria and other diseases . However , there are important differences in the structures of the LDH enzyme in animals and some parasites , like the malarial Plasmodium , because the genes for the enzymes in these two groups evolved separately . The parasite version of the LDH enzyme evolved hundreds of millions of years ago from an enzyme with a similar structure called malate dehydrogenase , which was inherited from bacteria . To work out how the LDH enzyme developed , Boucher et al . predicted and built the ancestral proteins that would have formed as the bacterial enzyme evolved into LDH . Studying these structures revealed that two mutations were mainly responsible for this evolution: six amino acids were added to the active site of the enzyme , and one amino acid ( at position 102 ) was replaced by a different amino acid . However , introducing the same mutations into a modern version of the bacterial enzyme did not produce a working form of the LDH enzyme . This suggests that other amino acids , further away from the active site , also influenced how LDH evolved . The structures found by Boucher et al . reveal that the enzymes evolved as a result of a gene duplicating , followed by one of the copies evolving a new function . However , some of the mutations responsible for the novel function occurred far from the active site , and it is still unknown how they exert their functional effects . Untangling the important mutations from the mundane will be necessary to fully understand how protein functions are created and how to control them—both of which will aid in developing effective drugs that target essential parasite proteins . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"evolutionary",
"biology",
"structural",
"biology",
"and",
"molecular",
"biophysics"
] | 2014 | An atomic-resolution view of neofunctionalization in the evolution of apicomplexan lactate dehydrogenases |
The development of outgrowths from plant shoots depends on formation of epidermal sites of cell polarity convergence with high intracellular auxin at their centre . A parsimonious model for generation of convergence sites is that cell polarity for the auxin transporter PIN1 orients up auxin gradients , as this spontaneously generates convergent alignments . Here we test predictions of this and other models for the patterns of auxin biosynthesis and import . Live imaging of outgrowths from kanadi1 kanadi2 Arabidopsis mutant leaves shows that they arise by formation of PIN1 convergence sites within a proximodistal polarity field . PIN1 polarities are oriented away from regions of high auxin biosynthesis enzyme expression , and towards regions of high auxin importer expression . Both expression patterns are required for normal outgrowth emergence , and may form part of a common module underlying shoot outgrowths . These findings are more consistent with models that spontaneously generate tandem rather than convergent alignments .
The development of plant shoots involves iterative formation of outgrowths . Shoot apical meristems produce leaf primordia , which in turn provide the setting for the initiation of new outgrowths such as serrations and leaflets . A common developmental module has been proposed to underlie the generation of both leaves and leaf-derived outgrowths ( Barkoulas et al . , 2008; Hay et al . , 2006 ) . A key feature of the module is an epidermal site of high intracellular auxin , located at the centre of convergence of the polarised auxin efflux carrier , PIN1 ( Barkoulas et al . , 2008; Benková et al . , 2003; Hay et al . , 2006; Reinhardt et al . , 2000 , 2003; Scarpella et al . , 2006 ) . The generation of polarity convergence sites has been most commonly explained by the up-the-gradient model , whereby cells localise PIN1 towards the neighbouring cell with the highest concentration of intracellular auxin ( Bilsborough et al . , 2011; Jönsson et al . , 2006; Smith et al . , 2006 ) . This mechanism is parsimonious because it can spontaneously generate spaced centres of polarity convergence without additional assumptions about the presence of auxin sources or sinks . Molecular mechanisms accounting for up-the-gradient behaviour have been proposed to involve detection of auxin-induced stress gradients ( Heisler et al . , 2010 ) or auxin transport sensing ( Cieslak et al . , 2015 ) . An alternative model for formation of sites of polarity convergences assumes that PIN1 becomes localised to cell membranes in proportion to the rate of auxin efflux across the membrane ( Mitchison , 1980; Rolland-Lagan and Prusinkiewicz , 2005 ) . A possible molecular mechanism for how cells might sense auxin flux has recently been proposed ( Cieslak et al . , 2015 ) . Although originally proposed to account for venation patterns , this flux-based model has also been shown to be compatible with the patterns of epidermal PIN1 polarity in the vegetative shoot apical meristem ( Stoma et al . , 2008 ) . Polarity convergence formation can be accounted for by assuming that sub-epidermal pro-vascular PIN1 strands are induced in regions of elevated auxin and cause a local depletion of auxin from the epidermis . This depletion causes PIN1 polarities to reorient and generate a site of convergence , which then raises auxin levels at its centre through transport . A problem with this model is that it predicts a transient drop in intracellular auxin concentration during early stages of convergence formation , which is not supported experimentally ( Brunoud et al . , 2012; Heisler et al . , 2005 ) . A further type of model for coordinating PIN1 orientations requires neither responding to auxin gradients between cells nor sensing auxin flux . Instead , the indirect coupling model involves intracellular polarity components that can establish cell polarity without external asymmetries in auxin distribution ( Abley et al . , 2013 ) . The two polarity components each exist in two forms: a diffusible cytoplasmic form ( A and B ) and a more slowly diffusing membrane bound form ( A* and B* ) . Interactions between the components leads to A* and B* being localised at opposite ends of the cell . PIN1 is recruited to the membrane by one of the components ( A* ) , causing a polarised PIN1 distribution . Polarities of neighbouring cells are coupled indirectly through a feedback mechanism in which extracellular auxin inhibits A* and thus PIN1 accumulation . This model results in polarities being oriented away from regions of high extracellular auxin and towards regions with low extracellular auxin . It can generate coordinated polarities for a field of cells , but it is unclear whether it can generate centres of PIN1 convergence . A convenient system for testing the models is the formation of ectopic 3D outgrowths from the abaxial leaf surface of kanadi1kanadi2 ( kan1kan2 ) mutants ( Eshed et al . , 2004 ) . These outgrowths can be considered as intermediates between leaf primordia and serrations . Similar to leaf primordia , the outgrowths emerge perpendicular to the main plane of the tissue , and like serrations , they are derived from the leaf . Because of their emergence from the abaxial lamina which can be readily imaged , these outgrowths are more amenable to time-lapse imaging than serrations which are often obscured by neighbouring cotyledon tissue and curving of the leaf edge . kan1kan2 outgrowths have elevated intracellular auxin at their tips , and their formation depends on specific patterns of auxin biosynthetic enzyme expression ( Wang et al . , 2011 ) . They provide a test bed for studying convergence site formation in a starting context which , unlike the apex , is not complicated by the prior patterns of continual primordium initiation . However , the dynamics of auxin accumulation , PIN1 polarity and expression of auxin biosynthesis genes at early stages of kan1kan2 outgrowth emergence have not been described . Moreover , the role of CUC genes , needed for formation of primordia and serrations ( Nikovics et al . , 2006 , Aida et al . , 1997 ) has not been determined . We show through time-lapse imaging that , similar to leaf primordia and serrations , kan1kan2 outgrowths are preceded by centres of PIN1 polarity convergence with elevated intracellular auxin . These convergent polarities arise within the context of an initial proximo-distal PIN1 polarity field and are promoted by CUC2 activity . An exploration of model behaviours reveals that models may be classified into two types: those that spontaneously generate convergent aligments ( e . g . the up-the-gradient ) and those that spontaneously generate tandem alignments ( e . g . flux-based and indirect coupling ) . Both types of model can account for the generation of an initial proximo-distal polarity field , followed by the formation of convergences with elevated intracellular auxin . However , unlike the convergent alignment model , tandem alignment models require the appearance of local regions with elevated auxin import . In support of tandem alignment models , we show that expression of the LAX1 auxin importer is elevated in regions of polarity convergence at the tips of kan1kan2 outgrowths , and that AUX/LAX importer genes are required for normal kan1kan2 outgrowth development . Additionally , we show that expression of YUCCA1 and YUCCA4 auxin biosynthetic enzymes tends to be elevated in regions of polarity divergence in kan1kan2 and WT leaves , an observation which is also most readily consistent with tandem alignment models . Thus , tandem alignment models provide parsimonious explanations for the developmental module underlying outgrowth emergence .
To characterise patterns of epidermal PIN polarity associated with the development of kan1kan2 outgrowths , a PIN1::PIN1:GFP reporter ( Benková et al . , 2003 ) was imaged on the abaxial surface of wild-type ( WT ) and kan1kan2 leaves . Here we focus on PIN1 since this is the predominant epidermally expressed PIN that shows polar intracellular distributions in wild-type leaves ( Guenot et al . , 2012 ) . In young kan1kan2 leaf primordia , the epidermal PIN1 polarity pattern was similar to that observed in WT: PIN1 polarities were oriented towards the leaf tip in both cases ( Figure 1A , B ) . At later stages of development , WT leaves retained a proximo-distal polarity pattern , and then no-longer showed detectable expression of the PIN1::PIN1:GFP reporter ( Figure 2A ) . The loss of PIN1::PIN1:GFP expression did not occur uniformly throughout the WT leaf because some patches of cells retained expression for longer than other regions of the leaf ( Figure 2A iii ) . By contrast to WT , kan1kan2 leaves maintained PIN1 expression until later stages of development and formed centres of PIN1 polarity convergence , located at the tips of emerging ectopic outgrowths ( Figure 2B iv and C iv ) . Cells at the centres of convergence had highly polarised PIN1 , oriented towards the interface between three or four neighbouring cells ( Figure 2C iv ) . To determine whether the centres of convergence preceded outgrowth emergence , leaves were imaged over several days and cell lineages that gave rise to convergences traced back through the time-course of the experiment ( Figure 2C , yellow dots , arrows and lines ) . This revealed that cells closest to the centre of convergence at the outgrowth tip descended from one or two cells , which were already at a centre of polarity convergence prior to outgrowth emergence ( Figure 2B , C ) . Taken together , these findings show that , prior to kan1kan2 outgrowth emergence , centres of polarity convergence develop within a proximodistally oriented polarity field . 10 . 7554/eLife . 18165 . 003Figure 1 . PIN1::PIN1:GFP expression in young WT and kan1kan2 leaf primordia . ( A ) WT primordium of leaf 1 , showing abaxial epidermis ( a total of 10 leaves were imaged over 2 separate experiments ) . ( B ) As for A , but for a kan1kan2 primordium ( a total of 15 leaves were imaged over three separate experiments ) . White arrows indicate inferred polarities . Dashed white lines indicate leaf outlines . Scale bars = 20 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 18165 . 00310 . 7554/eLife . 18165 . 004Figure 2 . PIN1::PIN1:GFP polarity patterns in WT and kan1kan2 leaf development . Confocal images of the PIN1::PIN1:GFP marker in the abaxial epidermis of the same WT leaf primordium imaged over a period of 2 days . Approximate leaf widths ( measured from projections of the z stacks ) are given above each image . Times from the beginning of the experiment are: i ) 0 hrs , ii ) 12 hrs , iii ) 36 hrs , iv ) 48 hrs . Cells indicated by white dots in iii ) retain detectable expression of the marker for longer than other cells . This time-lapse data is supported by snapshot images taken for at least five leaves at each developmental stage in a separate experiment . ( B ) Confocal images of the PIN1::PIN1:GFP marker in the abaxial epidermis of a kan1kan2 leaf primordium , prior to and during the emergence of an ectopic outgrowth . For each time point , a surface view of the abaxial epidermis ( left ) and a side view of a 3D rendering of the confocal z-stack ( right ) are shown . The side views allow the emergence of the outgrowth to be monitored . The time relative to when an outgrowth could clearly be observed and the estimated leaf width are given above each image . Yellow arrows in iv ) indicate an ectopic centre of PIN1 polarity convergence at the tip of an emerging outgrowth . ( C ) Magnified images of the regions outlined by the dashed grey rectangles in B , showing the development of the centre of PIN1 polarity convergence at the tip of the emerging outgrowth . Yellow dots , arrows and lines indicate the cells that form the centre of convergence in iv ) . Arrows indicate inferred PIN1 polarities and lines indicate inferred axes of PIN1 distributions . Scale bars = 20 µm . The scale bar in C i ) applies to all panels in C . This data is representative of tracking three out of three kan1kan2 leaves that developed ectopic outgrowths , each in a separate experiment , and with snapshot images from at least three leaves at each developmental stage ( in another separate experiment ) . DOI: http://dx . doi . org/10 . 7554/eLife . 18165 . 004 To investigate how the pattern of PIN1 polarity is related to the dynamics of intracellular auxin accumulation , a DR5::GFP reporter ( Benková et al . , 2003 ) was imaged in kan1kan2 leaves over the course of outgrowth development . At early stages , DR5::GFP signal was detected exclusively at the leaf tip ( Figure 3A ) , similar to the pattern described for the WT leaf ( Mattsson et al . , 2003; Scarpella et al . , 2006 ) . Approximately one day prior to the first observation of an outgrowth , locally elevated DR5::GFP signal was detected in groups of epidermal cells in the proximal region of the lamina , where outgrowths typically form ( Figure 3B i , yellow arrow ) . At later stages , these regions of DR5::GFP expression persisted , and could be found throughout emerging outgrowths ( Figure 3C , D , yellow arrows ) . Thus , both intracellular auxin activity maxima and centres of PIN1 polarity convergence precede and predict sites of outgrowth emergence in kan1kan2 leaves . 10 . 7554/eLife . 18165 . 005Figure 3 . DR5::GFP expression in a kan1kan2 leaf during outgrowth development . Confocal images of DR5::GFP in the same kan1kan2 leaf imaged over a period of 3 days . Times relative to the first observation of an outgrowth , and leaf widths , are given above images . B ii , C ii and D ii show optical sections through 3D renderings of confocal z-stacks . DR5::GFP ( green ) and auto-fluorescence plus CUC2::RFP ( red ) channels are shown ( CUC2::RFP is used to help show the leaf outline ) . Yellow arrows in D indicate the region of DR5::GFP activity at the tip of an outgrowth and yellow arrows in B and C indicate the same DR5::GFP expressing cells tracked back in time prior to outgrowth emergence . White arrow heads indicate high DR5::GFP signal in stipules . Scale bars = 50 µm . This data is representative of tracking 4 out of 4 kan1kan2 leaves that developed outgrowths ( in two experiments ) , and of snapshot data taken for at least 3 leaves at each of the developmental stages shown ( in another separate experiment ) . DOI: http://dx . doi . org/10 . 7554/eLife . 18165 . 005 cuc2 mutants fail to form centres of PIN1 polarity convergence in the leaf margin and do not develop serrations ( Bilsborough et al . , 2011; Nikovics et al . , 2006 ) . To investigate whether , like leaf serrations , kan1kan2 outgrowths depend on the CUC2 transcription factor , we generated kan1kan2cuc2 mutants . Leaves of kan1kan2 mutants consistently produce ectopic , finger-like outgrowths from the abaxial leaf surface ( 97% of 160 leaves observed had at least one outgrowth , and the mean number of outgrowths per leaf was 12 . 4 ) ( Figure 4A , B ) . By contrast , mature leaves of kan1kan2cuc2 mutants only occasionally developed such outgrowths ( only 8% of leaves observed developed outgrowths , and the mean number of outgrowths was 1 . 3 ) ( Figure 4C , D ) . Some kan1kan2cuc2 leaves also developed ridge-like thickenings of the abaxial surface or serrations in the leaf margin ( Figure 4—figure supplement 1A , B ) . 10 . 7554/eLife . 18165 . 006Figure 4 . Leaf phenotype of the kan1kan2cuc2 mutant . ( A ) Whole kan1kan2 plant . ( B ) OPT images of a kan1kan2 leaf showing abaxial leaf surface . ( C ) Whole kan1kan2cuc2 plant . ( D ) OPT images of a kan1kan2cuc2 leaf . Scale bars in A and C = 1 cm , scale bars in B and D = 1 mm . See Figure 4—figure supplement 1 for more details of the kan1kan2cuc2 phenotype . DOI: http://dx . doi . org/10 . 7554/eLife . 18165 . 00610 . 7554/eLife . 18165 . 007Figure 4—figure supplement 1 . Abaxial ridges and serrations produced in kan1kan2cuc2 mutants . ( A ) Example of a kan1kan2cuc2 leaf with an abaxial ridge-like thickening ( indicated by blue arrow ) . ( B ) Example of a kan1kan2cuc2 leaf with serrations . Scale bars = 1 mm . DOI: http://dx . doi . org/10 . 7554/eLife . 18165 . 007 To test if CUC2 is required for the formation of epidermal sites of PIN1 convergence , we generated kan1kan2cuc2 plants with the PIN1::PIN1:GFP reporter . Time-lapse imaging of PIN1::PIN1:GFP in the abaxial epidermis of the first two leaves of this mutant revealed that centres of convergence did not form ( Figure 5 ) . At early stages of leaf development , a proximodistal PIN1 polarity field was observed ( Figure 5 i ) , but similar to WT leaves , at later stages the expression of PIN1::PIN1:GFP was lost throughout most of the abaxial epidermis ( Figure 5B i–iii ) . Expression of the reporter was maintained in groups of epidermal cells which did not show centres of PIN1 convergence ( Figure 5B ii ) . 10 . 7554/eLife . 18165 . 008Figure 5 . PIN1::PIN1:GFP in leaf one of the kan1kan2cuc2 mutant . ( A ) Confocal image of PIN1::PIN1:GFP in the abaxial epidermis of the first leaf primordium of a kan1kan2cuc2 mutant . ( B ) Time-lapse confocal images of the abaxial side of the first leaf of a kan1kan2cuc2 mutant , taken at successive time points . Bi , ii and iii are maximum intensity projections of the abaxial side of the leaf , and therefore signal from both epidermal and sub-epidermal cell layers is combined . The data set shown here is representative of that obtained by tracking three kan1kan2cuc2PIN1::PIN1:GFP leaves ( from three different seedlings ) , and of snapshot images of 10 kan1kan2cuc2 first leaf primordia taken at each of the developmental stages shown . ( C ) Maximum intensity projection of the abaxial side of leaf 5 of a kan1kan2cuc2 mutant , showing an example of epidermal centres of PIN1 convergence ( on left and right sides of the leaf ) . Scale bars = 50 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 18165 . 008 In upper leaves of kan1kan2cuc2 mutants ( leaves 3 to 6 ) , ectopic primordium-like bumps were occasionally observed on the abaxial leaf surface ( in approximately 10% of leaves observed , n = 50 ) ( Figure 5C ) . These bumps were associated with centres of PIN1 polarity convergence linked with sub-epidermal PIN1 strands . The frequency of centres of PIN1 convergence in kan1kan2cuc2 mutants was much lower than in the kan1kan2 background where approximately 95% of leaves 3–6 were observed to have at least one centre of PIN1 polarity convergence ( n = 50 ) . Thus , formation of kan1kan2 outgrowths and polarity convergence sites is largely dependent on CUC2 , although this requirement can occasionally be circumvented . To gain insight into how CUC2 might promote the formation of epidermal centres of PIN1 convergence , we characterised the expression pattern of CUC2 at the time when centres of PIN1 polarity convergence form in kan1kan2 leaves . The CUC2::RFPer reporter ( which causes expression of endoplasmic reticulum-localised RFP ) was time-lapse imaged together with PIN1::PIN1:GFP in the abaxial epidermis of the first two kan1kan2 leaves . At early stages of leaf development , when PIN1 polarities are oriented distally , elevated expression of CUC2::RFPer was detected throughout the proximal half of the lamina ( Figure 6A , B ) . Before outgrowth emergence , expression of the reporter was lost from cells towards the base of the leaf , and from early centres of PIN1 polarity convergence ( Figure 6C ii , D ii ) . This resulted in groups of cells within the proximal half of the lamina with CUC2::RFPer expression , surrounded by non-CUC2::RFPer expressing cells . Later on these groups of CUC2::RFP expressing cells were seen to be at the distal side of each epidermal PIN1 polarity convergence ( Figure 6D iii ) and at the distal base of emerging outgrowths ( Figure 6D iii , side view , blue arrow head ) . This region is the boundary between the outgrowth and the abaxial lamina and is analogous to the sinus region of serrations where CUC2 is also expressed ( Nikovics et al . , 2006; Bilsborough et al . , 2011 ) . 10 . 7554/eLife . 18165 . 009Figure 6 . Expression of CUC2::RFPer and PIN1::PIN1:GFP during kan1kan2 outgrowth development . ( A ) Confocal images of a CUC2::RFPer reporter in the abaxial epidermis of a young kan1kan2 first leaf primordium . ( B ) Combined CUC2::RFPer ( red ) and PIN1::PIN1:GFP confocal channels for the same leaf as in A . White arrows indicate inferred polarity orientations . ( C ) and ( D ) Time-lapse imaging of CUC2::RFP and PIN1::PIN1:GFP in the abaxial epidermis of the first leaf of another seedling from that in A and B , over 58 hr prior to outgrowth development . C: CUC2::RFPer , D: combined channels , as in B . Times relative to outgrowth emergence and approximate leaf widths are given above each image . The right-hand images in D ii and iii show optical sections through 3D renderings of the confocal z-stacks at the position of the centre of convergence . White and yellow arrows indicate inferred PIN1 polarity orientations . Yellow dots and arrows indicate the cells that gave rise to the outgrowth tip in C iii and D iii . Blue arrow heads in D iii indicate elevated CUC2::RFPer expression distal to a centre of PIN1 convergence and at the base of the emerging outgrowth on its distal side . Data is representative of that obtained by tracking outgrowth development in five kan1kan2 CUC2::RFPEer PIN1::PIN1:GFP leaves , in three separate experiments . ( E ) Zoomed-in image of the region outlined by the dashed rectangle in D iii , showing PIN1 polarities within and surrounding the domain of elevated CUC2::RFPer expression . ( F ) Confocal images of CUC2::RFPer ( red ) and PIN1::PIN1:GFP ( green ) associated with an ectopic kan1kan2 outgrowth at a later stage of development to that in C–E . Dotted blue line outlines the domain with elevated CUC2::RFPer expression at the distal boundary between an outgrowth and the main lamina . ( G ) CUC2::RFPer expression in two WT first leaf primordia ( ii also shows PIN::PIN1:GFP in green ) . Blue arrow in ii indicates a site of elevated CUC2::RFPer expression in the leaf margin . Red signal in the centre of the lamina in ii ) is non-ER localised and therefore due to auto-fluorescence . Data is representative of of that obtained by imaging nine leaves at each of the developmental stages shown , across two separate experiments . Scale bars = 50 μm . Dashed white lines indicate leaf outlines . DOI: http://dx . doi . org/10 . 7554/eLife . 18165 . 009 We next investigated how the expression pattern of CUC2::RFPer is related to the polarity of PIN1 in emerging outgrowths . Inspection of PIN1 polarities using the PIN1::PIN1:GFP reporter revealed that in developing outgrowths , cells close to the regions of elevated CUC2::RFPer expression frequently had PIN1 polarities oriented away the regions of high CUC2 expression ( Figure 6E , F ) . On the proximal side of the CUC2::RFPer expression domain ( i . e . in cells of the emerging outgrowth ) , PIN1 polarities were oriented proximally , away from cells with high CUC2::RFP expression and towards a centre of convergence at the outgrowth tip; while on the distal side of the CUC2::RFPer expression domain , polarity appeared to point distally . In contrast to kan1kan2 leaves , in WT , CUC2::RFPer expression was mainly restricted to the leaf base and to regions of the margin associated with centres of PIN1 convergence involved in serration development ( Figure 6G; Bilsborough et al . , 2011 ) . Thus , ectopic outgrowth development in kan1kan2 leaves is associated with the ectopic expression of CUC2 in the abaxial lamina . Similar to leaf serrations , the normal development of kan1kan2 outgrowths and the centres of PIN1 polarity convergence that precede them is dependent on CUC2 . We next compare different models of PIN1 polarity for their ability to capture the observed PIN1 polarity patterns and for their consistency with the role of CUC2 in PIN1 polarity convergence formation . To test possible mechanisms that could underlie the observed epidermal polarity patterns , we compare the abilities of up-the-gradient , flux-based and indirect coupling models to account for them . We first characterise basic model behaviours , and then investigate the assumptions needed for each model to capture the observed polarity patterns and how they may be tested experimentally . The flux and up-the-gradient models are implemented as in previous publications , using the simplifying assumption that auxin moves directly from cell to cell and omitting an explicit representation of the cell wall ( Bilsborough et al . , 2011; Feugier et al . , 2005; Jönsson et al . , 2006; Rolland-Lagan and Prusinkiewicz , 2005; Smith et al . , 2006; Stoma et al . , 2008 ) . By contrast , we explicitly represent the cell wall in our implementations of the indirect coupling model . In this model , we also represent cell membranes and walls with several compartments per cell edge ( allowing lateral diffusion of components to be simulated ) . However , as in previous work , flux and up-the-gradient models are implemented with a single compartment per edge and no lateral diffusion of components along the cell edge ( Bilsborough et al . , 2011; Feugier et al . , 2005; Jönsson et al . , 2006; Rolland-Lagan , 2008; Smith et al . , 2006; Stoma et al . , 2008 ) . With the flux-based model , we assume a linear feedback between flux and PIN1 allocation , which tends to generate broad regions of coordinated polarisation rather than narrow canalised strands that arise with super-linear feedback ( Feugier et al . , 2005; Stoma et al . , 2008 ) . Details of all the model assumptions are given in the Materials and methods . Previously simulations of up-the-gradient and flux-based models have considered polarity patterns in fields of cells all of which have the potential to polarise ( Jönsson et al . , 2006; Rolland-Lagan , 2008; Smith et al . , 2006; Stoma et al . , 2008 ) . However , to get a clearer view of the properties of each model , we begin by considering a group of cells having initially uniform auxin concentration and PIN1 distribution , and with only the central cell having the ability to relocate its PIN1 according to the rules specified by each model ( Figure 7A–C ) . We find that the central cell becomes polarised with both the up-the-gradient and flux-based models , assuming that there are small random fluctuations ( noise ) in the initial concentrations of PIN1 in the cell membranes and that auxin flux from the central cell can modify the initially homogeneous auxin concentration in the neighbouring tissue ( Figure 7A and B ) . In the up-the-gradient model , a cell edge with elevated PIN1 causes an increase in auxin concentration in the neighbouring cell . This elevated auxin concentration in the neighbour feeds back to cause an increase in PIN1 allocation to the cell edge ( Figure 7A ii , purple arrow ) . Assuming a limited pool of PIN1 in the cell , there will be a corresponding reduction in PIN1 allocation to other cell edges . In the flux-based model polarity arises because if one cell edge has slightly elevated PIN1 ( due to small random fluctuations ) , this edge will have a higher rate of auxin efflux , which feeds back to favour further recruitment of PIN1 to the given cell edge ( Figure 7B ii , purple arrow ) . Recruitment of PIN1 to other cell edges is inhibited because the different edges of a cell compete to export auxin into surrounding cells . 10 . 7554/eLife . 18165 . 010Figure 7 . Comparison of basic behaviours of indirect coupling , flux-based and up-the-gradient models . ( A–C ) Investigating the ability of cells to polarise in the absence of pre-established external asymmetries or polarisable neighbours for up-the-gradient ( A ) , flux-based ( B ) and indirect coupling ( C ) models . Panels i ) show the initial state of the simulations . In each simulation , all cells initially have the same auxin concentration ( indicated by the intensity of green , where darker green indicates a higher concentration ) . In the case of up-the-gradient or flux-based models ( A , B ) , noise is present in the initial levels of PIN1 ( shown by red lines at cell edges ) . In the case of indirect coupling ( C ) , a single central cell has noise in the levels of A* and B* . Panels ii ) show intermediate states of the simulations , and panels iii ) show the final states . In this and subsequent simulations , PIN1 polarisation is indicated by black arrows , where the arrow points towards the region of the cell with highest PIN1 . ( A ) For the up-the-gradient model , if a given cell edge has slightly elevated PIN , this causes the juxtaposed neighbour to have an increased auxin concentration . This increased auxin concentration in the neighbour feeds-back to cause increased PIN1 recruitment to the given cell edge ( purple arrow in ii ) . ( B ) For the flux-based model , a cell edge with a slightly elevated PIN1 concentration causes an increased auxin efflux rate . This increased auxin efflux feeds-back to cause an increased recruitment of PIN1 to the given cell edge ( purple arrow in ii ) . ( C ) For the indirect coupling model , polarity is generated independently from PIN1 and auxin through A* and B* polarity components ( purple arrow in ii ) which in turn causes polarisation of PIN1 . ( D–F ) Behaviour of up-the-gradient ( D ) , flux-based ( E ) and indirect coupling ( F ) models for 2D arrays of polarisable cells in the absence of pre-established asymmetries . D and E were initialised with noise in the initial concentrations of auxin and F was initialised with noise in the levels A* and B* in the membrane . Details of all simulations can be found in the Materials and methods . See Figure 7—figure supplement 1 for a further comparison of the requirements for polarisation . DOI: http://dx . doi . org/10 . 7554/eLife . 18165 . 01010 . 7554/eLife . 18165 . 011Figure 7—figure supplement 1 . Comparison of models’ ability to generate polarity for a cell surrounded by a fixed environment . In each case the central cell is initialized with noise in the concentrations of PIN at each cell edge and all seven cells have the same auxin concentration . The auxin concentration in each outside neighbor of the central cell is fixed throughout the simulation so that the central cell cannot induce asymmetries in its environment . DOI: http://dx . doi . org/10 . 7554/eLife . 18165 . 011 As previously described , the indirect coupling model also generates a polarised distribution of PIN1 in the central cell ( Figure 7C ) . This is due to the auto-activating and mutually inhibitory activities of membrane bound A* and B* polarity components , combined with relatively fast diffusion of their cytoplasmic forms , which generates cell polarity . PIN1 is then recruited to the membrane with high A* ( Marée et al . , 2006; Jilkine et al . , 2007 ) . Thus , all three models have the potential to generate polarity in a single polarisable cell surrounded by non-polarisable neighbours in an initially uniform field of auxin concentration . For the indirect coupling model , this polarity arises through partitioning of the polarity components , while for the flux-based and up-the-gradient models it arises because small fluctuations in PIN1 distribution create variations in auxin flux or concentrations which feed back to reinforce the polarity . There is , however , a key difference between the up-the-gradient and other models . If auxin concentration in the surrounding medium remains fixed throughout the simulation ( equivalent to voltage-clamping in neurophysiology ) , then the up-the-gradient model does not generate polarity ( Figure 7—figure supplement 1 ) . In contrast , both the flux-based and indirect coupling model present polarity even under such settings ( Figure 7—figure supplement 1 ) . In a previous paper we overlooked the potential effect of feedback in the flux-based and up-the-gradient models for single polarisable cells , and implicitly assumed that such cells would not polarise ( Abley et al . , 2013 ) . Moreover , we did not distinguish between lack of pre-established asymmetric cues versus lack of being able to establish asymmetric cues ( e . g . through auxin transport ) . The above simulations show that this distinction needs to be taken into account to appreciate that , whereas the flux-based model can present intracellular partitioning ( i . e . establish polarity in the absence of asymmetric cues or polarisable neighbours ) , the up-the-gradient model relies on its ability to change its surrounding to establish and maintain polarization . Thus , both the flux-based and up-the-gradient model share commonalities with the indirect coupling model regarding intracellular partitioning , albeit through a more indirect mechanism involving auxin transport between cells . We next consider an array of polarisable cells in the absence of pre-established asymmetries . Noise is present in initial auxin concentrations ( for the up-the-gradient and flux-based models ) , or in the A*-B* polarity components ( for the indirect coupling model ) , and all cells have the ability to relocate PIN1 ( Figure 7D–F ) . In this situation the up-the-gradient model generates convergent polarities , in which spaced groups of cells orient their PIN1 polarity towards a central region ( Figure 7D ) ( Jönsson et al . , 2006; Smith et al . , 2006 ) . This is because fluctuations lead to competing centres of high auxin concentration which orient polarities towards them . By contrast , the flux-based model tends to generate swirled patterns of polarity , in which polarities are coordinated in tandem between neighbouring cells ( Figure 7E ) . Convergent polarities are disfavoured because if two cells have PIN1 oriented towards each other , both cells would experience a low net auxin efflux across their PIN1-rich ends due to transport towards them from the opposing neighbour . The low net auxin efflux would cause removal of PIN1 from the membranes and relocalisation to edges juxtaposed with a PIN1-free edge of a neighbouring cell , promoting tandem alignments . Similar to the flux-based model , indirect coupling also generates swirled patterns of polarity ( Figure 7F , Abley et al . , 2013 ) . Convergent alignments are disfavoured because if two cells have PIN1 oriented towards each other , both cells transport auxin to the intervening extracellular space . Accumulation of auxin in the extracellular space then inhibits A* , and therefore PIN1 , in adjacent membranes . This destabilises convergent polarities and favours tandem alignments . Thus , in the absence of pre-established asymmetries , the models spontaneously generate two types of polarity pattern . The flux-based and indirect coupling models both generate tandem alignments , and are subsequently referred to as tandem alignment models; while the up-the-gradient model generates convergent alignments , and is subsequently referred to as a convergent alignment model . We next explore the assumptions that need to be added to these models for them to account for the epidermal PIN1 polarity patterns observed in WT and kan1kan2 leaves . With each model , we attempt to capture the initial proximodistal polarity field observed in both WT and kan1kan2 , where polarities are aligned in tandem along the proximodistal leaf axis with high intracellular auxin at the tip ( Figure 1 ) . We then explore the assumptions required for the formation of centres of PIN1 convergence with elevated intracellular auxin , such as those observed on the abaxial side of the main lamina at later stages of kan1kan2 leaf development ( Figure 2B , C , Figure 3 ) . The convergent alignment model has been proposed to account for the initial centre of convergence at the distal end of the leaf primordium ( Jönsson et al . , 2006; Smith et al . , 2006 ) . For a proximodistal polarity field to be maintained within the growing primordium , the convergent alignment model requires maintenance of an increasing auxin concentration gradient from the leaf base to the leaf tip . One way this can be achieved is through net auxin removal at the leaf base . Net auxin removal from a region of tissue could be achieved through a decreased rate of auxin biosynthesis , an increased rate of transport away from the region , or an increased rate of auxin degradation or conjugation . Here net removal is achieved using elevated auxin degradation rates ( Figure 8A ) . Combining this assumption with an initially elevated intracellular auxin concentration at the leaf tip , a proximodistal polarity field can be maintained within an array . The intracellular auxin concentration is kept high at the distal end of the array through transport towards this region ( Figure 8A ) , similar to experimental observations using the DR5::GFP reporter ( Figure 3 , Mattsson et al . , 2003; Scarpella et al . , 2006 ) . As the primordium increases in size ( Figure 8B ) , the auxin concentration tends to become shallow at a distance from the leaf tip . This can result in the spontaneous formation of a centre of convergence with elevated intracellular auxin within the proximal region of the lamina ( Figure 8B ) , similar to that observed in kan1kan2 mutants ( Figure 2B , C ) . The length of the tissue at which the polarity reversal occurs depends on the parameters values used in the model . The failure to form such a convergence centre in WT leaves could be explained by the PIN1 polarity pattern becoming fixed at early stages of development , preventing polarity reorientation . However , how this could be achieved is unclear , although it has been proposed that absence of CUC2 activity may play a role in fixing polarity ( Bilsborough et al . , 2011 ) . 10 . 7554/eLife . 18165 . 012Figure 8 . Formation of a proximo-distal polarity field and centres of convergence in the convergent alignment model . ( A ) Formation of a proximo-distal polarity field due to the presence of an elevated initial auxin concentration at the leaf tip ( orange cells ) and an elevated rate of auxin removal from the leaf base ( blue cells ) . The graph shows the concentration of intracellular auxin for each cell in the column marked with the grey arrow . ( B ) As for A , but for a larger tissue . A centre of polarity convergence forms within the proximal half of the lamina . DOI: http://dx . doi . org/10 . 7554/eLife . 18165 . 012 In the flux-based model , polarities become oriented away from auxin sources and towards auxin sinks . A proximodistal polarity pattern may therefore be generated with net auxin production ( a source ) proximally and elevated net auxin removal ( a sink ) distally ( Figure 9A ) . Here we simulate net auxin production in proximal regions with an increased auxin biosynthesis rate , but it could also be achieved with a reduced auxin degradation rate or increased auxin influx from ( or decreased auxin efflux to ) tissues beyond those represented in the simulation . Likewise , net removal from the leaf tip is simulated using elevated degradation , but could also occur through reduction in biosynthesis , increase in conjugation , or via transport into underlying tissue layers ( Bayer et al . , 2009; Scarpella et al . , 2006 ) . 10 . 7554/eLife . 18165 . 013Figure 9 . Formation of a proximo-distal polarity field in the flux-based and indirect coupling models . ( A ) Formation of a proximo-distal polarity field in the flux-based model due to the presence of elevated auxin biosynthesis at the leaf base ( orange cells ) and an elevated rate of auxin removal from the leaf tip ( blue cells ) . The graph on the right shows the concentration of intracellular auxin for the column marked with the grey arrow . ( B ) As for A , but with elevated auxin import ( green cell outlines ) at the distal end of the tissue . ( C ) As for A , but for the indirect coupling model . ( D ) As for B , but for the indirect coupling model . DOI: http://dx . doi . org/10 . 7554/eLife . 18165 . 013 With these assumptions the flux-based model generates a proximodistal pattern of polarities . However , intracellular auxin concentrations tend to be lowest at the distal end of the tissue ( Figure 9A ) , inconsistent with experimental observations . This inconsistency can be avoided if the leaf tip acts as a weak instead of a strong auxin sink . The initial drop in intracellular auxin at the tip is then followed by a rise to a high level because of the transport towards this region ( Stoma et al . , 2008; Walker et al . , 2013 ) . Another possibility is that the leaf tip has an elevated auxin import rate as well as an elevated rate of auxin removal ( Figure 9B ) . Elevated auxin import in the distal-most cells causes increased auxin flux towards them , encouraging polarities to point distally . The elevated import also causes the distal-most cells to accumulate intracellular auxin . Although auxin is synthesised at the highest rate in the proximal-most files of cells , in the final state of the simulation , maximal auxin concentrations are found at the distal end ( Figure 9B ) . This is due to polar transport of auxin away from the base , towards the leaf tip . Similar to the flux-based model , with indirect coupling an initial proximodistal alignment of polarities can be generated if the proximal region ( leaf base ) has net auxin production ( here due to elevated auxin biosynthesis generating high extracellular auxin ) , while the distal region ( leaf tip ) has elevated net auxin removal ( here due to elevated auxin degradation generating low extracellular auxin ) . As in the flux-based model , elevated auxin degradation at the distal end of the tissue can lead to low intracellular auxin in this region ( Figure 9C ) . Combining an elevated rate of auxin import at the leaf tip with the elevated degradation ensures low extracellular auxin concentrations in this region ( encouraging polarities to point distally ) and allows distal cells to accumulate intracellular auxin ( Figure 9D ) . Thus , the two tandem alignment models both have similar requirements for the generation of a proximodistal polarity field with an intracellular auxin maximum in the distal region . The predictions of the tandem alignment models for the generation of a proximodistal polarity field differ from those of the convergent alignment model . Tandem alignment models predict that the leaf base has net production of auxin , whilst the convergent alignment model predicts that this region has net auxin removal . Also , tandem alignment models require that the leaf tip has an elevated rate of auxin import and removal to account for high auxin in this region , whereas the convergent alignment model has no such requirement . A further difference is that unlike the convergent alignment model , the two tandem alignment models do not spontaneously generate centres of polarity convergence upon an increase in tissue size ( the results of simulations shown in Figure 9A–D are similar for a range of tissue sizes ) . Thus , with tandem alignment models , further assumptions are needed to generate centres of PIN1 polarity convergence . With tandem alignment models , a site of convergence could arise by a cell having higher levels of auxin removal , caused by export to underlying cells . This cell could be positioned in regions of elevated intracellular auxin , as proposed by Stoma et al , 2008 . However , induction of auxin removal alone can result in a drop in intracellular auxin concentration before convergence formation ( Stoma et al . , 2008 ) , and it is unclear how such regions of elevated intracellular auxin could emerge in the proximal half of the kan1kan2 abaxial lamina . One possible explanation for the generation of convergence sites in tandem alignment models is that the proximal region of the abaxial lamina has an extended region of auxin biosynthesis . This region corresponds to the region of ectopic CUC2::RFP expression in the proximal region of young kan1kan2 primordia ( Figure 6A , C i ) . In the tandem alignment models , auxin biosynthesis in this region leads to a shallow peak in intracellular auxin proximal to the distal limit of auxin biosynthesis ( Figure 10A , C ) . This peak arises as a consequence of elevated auxin biosynthesis in this domain , together with polarised transport which shifts the position of the peak distally . If intracellular auxin above a threshold concentration activates auxin import , a region with elevated import arises , leading to a centre of polarity convergence ( Figure 10B , D ) . Elevated auxin removal needs to arise in addition to elevated auxin import since import alone causes very high levels of intracellular auxin to accumulate at the centre of convergence . This can disrupt convergence formation by preventing the maintenance of low extracellular auxin ( in the indirect coupling model ) , and by increasing total auxin efflux from the central cell ( in the flux-based model ) . Unlike the model of Stoma et al . , 2008 , where only auxin removal is induced by high intracellular auxin , the combination of elevated auxin import and removal helps to ensure that a dip in intracellular auxin concentration does not occur prior to convergence formation . 10 . 7554/eLife . 18165 . 014Figure 10 . Formation of centres of polarity convergence in the flux-based and indirect coupling models . ( A ) A proximo-distal polarity field is initially established with the flux based model due to the presence of elevated auxin biosynthesis in the proximal half of the leaf ( orange cells ) and elevated rates of auxin import and removal from the leaf tip ( blue cells with green outlines ) . The graph on the right shows the concentration of intracellular auxin for the column marked with the grey arrow . Note that there is a broad peak of intracellular auxin concentration within the proximal domain of elevated auxin synthesis ( black arrow ) . ( B ) Subsequent stage of the simulation in A . If it is assumed that intracellular auxin above a threshold concentration induces elevated auxin import and removal , one or more cells within the proximal domain of elevated auxin biosynthesis are induced to have elevated levels of import and removal . Neighbouring cells reorient their polarity to point towards the high import cells , which accumulate elevated levels of intracellular auxin . ( C ) As for A ) , but for the indirect coupling model . ( D ) As for B ) , but for the indirect coupling model . See Figure 10—figure supplement 1 for an alternative version of the indirect coupling model which incorporates D6-kinase like activity . DOI: http://dx . doi . org/10 . 7554/eLife . 18165 . 01410 . 7554/eLife . 18165 . 015Figure 10—figure supplement 1 . Incorporation of a D6 protein kinase-like activity into the indirect coupling model . D6 kinase activity was added into simulations of proximodistal polarity establishment ( A ) , ( compare with Figure 9D ) and convergence formation ( B ) ( compare with Figure 10D ) . We assume that D6 protein kinase localizes to the A* end of the cell and locally enhances the export of auxin by PIN1 . Since A* determines the distribution of PIN1 in the simple form of the indirect coupling model , this addition of D6-kinase-regulated PIN1 activity where A* is present makes little difference to the model’s behaviour . DOI: http://dx . doi . org/10 . 7554/eLife . 18165 . 015 D6 protein kinases are polarly localized in cells independently of PIN proteins and their phosphorylation of PIN proteins is required for PIN-mediated auxin efflux ( Barbosa et al . , 2014 , Zourelidou et al . , 2009 ) . We added a representation of D6 kinase activity to the indirect coupling model , as cell polarity in this system does not depend on PIN activity , and found that this does not qualitatively affect model behavior ( Figure 10—figure supplement 1 ) . This comparison of models gives rise to specific predictions that may be used to distinguish between them experimentally . Tandem alignment models require that , at the leaf tip and at the tips of outgrowths ( where centres of convergence are located ) , the epidermis has elevated rates of auxin import and auxin removal ( which could be removal into underlying tissue layers ) . Locally elevated auxin import is not required for the convergent alignment model , as sites with slightly elevated intracellular auxin reinforce themselves . The tandem alignment models predict that the epidermis towards the leaf base has elevated net auxin production , whilst the convergent alignment model predicts that it has net auxin removal . Moreover , a broader band of auxin biosynthesis at the base of the abaxial side of the kan1kan2 leaf provides one possible mechanism for inducing convergence site formation with tandem alignment models . To test whether auxin import is elevated at centres of polarity convergence , we determined the expression pattern of auxin importers in kan1kan2 mutant leaves . The AUX/LAX family of auxin importers includes four genes in A . thaliana ( AUXIN RESISTANT 1 ( AUX1 ) , LIKE AUXIN RESISTANT 1 ( LAX1 ) , LIKE AUXIN RESISTANT 2 ( LAX2 ) , and LIKE AUXIN RESISTANT 3 ( LAX3 ) ) which encode proteins that actively transport auxin from the extracellular space into the cytoplasm ( Parry et al . , 2001; Péret et al . , 2012; Yang et al . , 2006 ) . We focus on AUX1 and LAX1 expression since both genes are expressed in the leaf epidermis and therefore may contribute towards the auxin import activity predicted by the models ( Kasprzewska et al . , 2015 ) . Consistent with previous reports , expression of LAX2::GUS was detected in leaf vascular tissue ( Figure 11—figure supplement 1A ) and expression of LAX3::GUS was absent from leaves but present in vascular tissue of hypocotyls and roots ( Bainbridge et al . , 2008; Kasprzewska et al . , 2015 ) ( Figure 11—figure supplement 1B , C ) . As expected from the tandem alignment models and consistent with a previous report , LAX1::GUS was expressed at the tips of young wild-type leaf primordia , at the tips of serrations , and in the midvein , but was absent from the rest of the leaf lamina ( Kasprzewska et al . , 2015 ) ( Figure 11A ) . In kan1kan2 leaves , LAX1::GUS was expressed in the same regions as in WT , but was also ectopically expressed in groups of a few cells in proximal regions of the abaxial epidermis ( Figure 11B ) . These groups of cells were at the tips of developing outgrowths ( Figure 11B ii and iii ) . To investigate whether LAX1::GUS expression preceded outgrowth emergence , sections of kan1kan2LAX1::GUS seedlings stained to reveal GUS activity were imaged . This revealed that local epidermal sites of elevated LAX1::GUS expression were present in the abaxial lamina prior to outgrowth emergence ( Figure 11C i ) , consistent with a role of LAX1 in the generation of centres of polarity convergence . Transverse sections also confirmed that LAX1::GUS expression at the tips of outgrowths is located in the epidermis ( Figure 11C ii ) . 10 . 7554/eLife . 18165 . 016Figure 11 . Expression of LAX1::GUS and AUX1::AUX1:YFP in WT and kan1kan2 leaves . ( A ) Expression pattern of LAX1::GUS in WT leaves one and two . LAX1::GUS was expressed at the tips of developing primordia ( arrows in ( i ) , black dashed lines indicate leaf outlines , arrow heads indicate stipules ) ( a total of 6 plants ( with 12 young leaves ) were imaged across two separate experiments ) and at the tips of serrations ( ii ) ( arrows indicate serrations ) ( a total of 9 plants ( 18 leaves ) were imaged across three separate experiments ) . Leaf widths are given above images . ( B ) Expression pattern of LAX1::GUS in kan1kan2 leaves . LAX1::GUS was expressed at the tips of primordia ( i ) ( a total of 7 plants ( 14 leaves ) were imaged across two separate experiments ) and at the tips of outgrowths ( ii and iii ) ( arrows ) ( 20 out of 21 leaves imaged across three separate experiments ) . ii ) shows leaf 1 , iii ) shows leaf 3 . ( C ) Transverse sections through GUS stained kan1kan2 LAX1::GUS seedlings , showing points of LAX1::GUS expression before outgrowths have emerged ( black arrows in i ) ( data supported by serial sections of 3 other leaves at similar developmental stages in a separate experiment ) and at the tips of developing outgrowths ( ii ) ( a total of 8 leaves were imaged across two separate experiments ) . Dashed orange line in i ) indicates leaf outline . ( D ) AUX1::AUX1:YFP expression in leaf 1 of WT , showing abaxial surface ( i ) ( a total of 15 leaves were imaged across two separate experiments ) and adaxial surface ( ii ) ( a total of 4 leaves were imaged in two separate experiments ) of two different leaves . ( E ) AUX1::AUX1:YFP expression in leaf 1 of a kan1kan2 mutant , showing abaxial surface . Arrow points to the tip of an emerging outgrowth with locally elevated AUX1::AUX1:YFP signal ( a total of 9 leaves were imaged , across three different experiments ) . ( F ) Time-lapse confocal imaging of AUX1::AUX1:YFP in the abaxial epidermis of the first leaf of a kan1kan2 seedling . White arrows mark the positions of cells that eventually gave rise to the AUX1::AUX1:YFP expressing cells in the developing serration on the right side of the leaf . Yellow arrows mark the positions of cells which eventually gave rise to the tip of the ectopic outgrowth on the left side of the leaf . Red and green images in iv ) , v ) and vi ) show side views , from the left hand side of the leaf ( showing that the outgrowth emerged at the time point shown in v ) . Red shows auto-fluorescence and a CUC2::RFP marker ( used to show the leaf contours ) , and green shows AUX1::AUX1:YFP signal . Times from the beginning of the experiment at which images were taken are: i ) 0 hrs , ii ) 22 hr 40 min , iii ) 31 hr 10 min , iv ) 46 hr 40 min , v ) 55 hr 40 min , vi ) 74 hr 40 min . Data is consistent with tracking experiments performed for two other kan1kan2 leaves that developed ectopic outgrowths , and with snapshot images of at least 6 leaves before and after outgrowth emergence . ( G ) Mean number of outgrowths ( +/- standard error of the mean ) in rosette leaves of kan1kan2 plants carrying mutant alleles of AUX/LAX genes . n numbers are: kan1kan2: 84 leaves from 10 plants , kan1kan2aux1lax1: 110 leaves from 10 plants , kan1kan2aux1lax1lax2: 136 leaves from 10 plants , kan1kan2aux1lax1lax2lax3:200 leaves from 10 plants . ( H ) Convergent alignment model where elevated intracellular auxin causes elevated rates of auxin import and removal . The model is initialised with elevated auxin degradation at the leaf base ( blue cells ) and an elevated rate of auxin import and degradation at the leaf tip ( blue cells with green outlines ) . Centres of convergence form within the proximal half of the lamina and when the intracellular auxin concentration exceeds a threshold level , an elevated rate of auxin import and removal is induced ( blue cells with green outlines ) . Scale bars = 50 µm . See Figure 11—figure supplement 1 for LAX2::GUS and LAX3::GUS expression . DOI: http://dx . doi . org/10 . 7554/eLife . 18165 . 01610 . 7554/eLife . 18165 . 017Figure 11—source data 1 . Counts of outgrowths used to generate Figure 11G . DOI: http://dx . doi . org/10 . 7554/eLife . 18165 . 01710 . 7554/eLife . 18165 . 018Figure 11—figure supplement 1 . Expression patterns of LAX2::GUS and LAX3::GUS . ( A ) Expression of LAX2::GUS in a kan1kan2+/- leaf , showing expression in leaf vascular tissue . Scale bar = 100 μm . ( B ) Expression of LAX3::GUS in a WT seedling , showing expression is absent from leaves , but present in the hypocotyl and root . Scale bar = 1 mm ( C ) Expression of LAX3::GUS in a kan1kan2 seedling . Scale bars = 100 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 18165 . 018 In wild-type primordia , AUX1::AUX1:YFP was expressed in all cells of the abaxial epidermis , with strongest expression at the leaf margin ( Figure 11Di ) . On the adaxial leaf surface , expression was excluded from most of the epidermis , but detected in cells close to and within the leaf margin ( Figure 11Dii ) . kan1kan2 leaves showed a similar pattern of AUX1::AUX1:YFP expression on their adaxial surface , but the expression pattern in the abaxial epidermis differed from that in wild type . Expression was found throughout distal regions of the leaf , but was concentrated in cells of emerging outgrowths in proximal regions ( Figure 11E ) . Time-lapse confocal imaging revealed that , prior to outgrowth formation , AUX1::AUX1:YFP expression was absent from proximal regions ( Figure 11Fiii , iv , yellow arrow ) , but became detectable as outgrowths emerged ( Figure 11Fv–iv , yellow arrows ) . Since centres of PIN1 polarity convergence form at least one day before outgrowth emergence , these observations suggest that strong AUX1 expression is not involved in their initial development . However , local elevation of AUX1 expression at outgrowth tips could play a role in the stabilisation and maintenance of convergent polarities . To investigate whether AUX/LAX auxin importers are required for the generation of ectopic outgrowths , we generated kan1kan2aux1lax1lax2lax3 hextuple mutants . Loss of all four AUX/LAX genes caused a reduction in the mean number of outgrowths per rosette leaf from 12 . 4 ( in kan1kan2 mutants ) to 1 . 6 ( in kan1kan2aux1lax1lax2lax3 mutants ) ( Figure 11G ) . To determine the contribution of lax2 and lax3 to the loss of outgrowths , kan1kan2aux1lax1 quadruple and kan1kan2aux1lax1lax2 quintuple mutants were also generated . Loss of function of aux1 and lax1 alone caused a reduction in the number of outgrowths compared with kan1kan2 plants ( Figure 11G ) . This reduction was further increased by a loss of LAX2 activity ( kan1kan2aux1lax1lax2 mutants have a further loss of outgrowths [Figure 11G] ) . Since LAX2::GUS expression was only detected in sub-epidermal tissue , this finding supports a role of sub-epidermal auxin importer expression in the generation of outgrowths . LAX2-mediated sub-epidermal auxin import below centres of convergence could contribute to removal of auxin from the centre of convergence . Despite the absence of LAX3::GUS reporter expression in leaves , loss of LAX3 caused a further reduction in the number of outgrowths ( kan1kan2aux1lax1lax2 mutants generated more outgrowths than kan1kan2aux1lax1lax2lax3 mutants , [Figure 11G] ) . This suggests that LAX3 may function redundantly with other members of the AUX/LAX family in the generation of ectopic outgrowths , or that differences in genetic background between the multiple mutants has an effect on outgrowth frequency ( see Materials and methods ) . The findings that auxin importers are expressed in the epidermis at the tips of leaves and emerging kan1kan2 outgrowths , and play a role in outgrowth development , meet the requirement of tandem alignment models ( Figures 9 and 10 ) . In its most parsimonious form , there is no need for elevated auxin import at the leaf tip or at subsequent centres of convergence with the convergent alignment model ( Figure 8B ) . However , this does not make the model incompatible with elevated auxin import at these sites . If it is assumed that intracellular auxin above a threshold concentration induces elevated auxin import and removal , elevated import and removal may be induced after convergence formation ( Figure 11H ) . This does not cause disruption of the centres of convergence . Distally oriented polarities near the leaf tip may also be achieved if these cells have an elevated rate of auxin import and removal ( rather than an elevated initial auxin concentration ) ( Figure 11H ) . Thus , induction of auxin import by elevated intracellular auxin concentrations is compatible with the convergent alignment model , though it renders the model less parsimonious . To test whether CUC2 could promote the formation of kan1kan2 outgrowths by influencing expression of auxin importers , we examined the expression of LAX1::GUS and AUX::AUX1:YFP reporters in the kan1kan2cuc2 background . Sites of detectable LAX1::GUS expression were lost from the kan1kan2cuc2 lamina ( Figure 12A ) . Additionally , in kan1kan2cuc2 leaves , AUX1::AUX1:YFP was expressed uniformly throughout the abaxial lamina ( Figure 12B ) , suggesting that CUC2 represses the expression of AUX1 in the region surrounding outgrowths . In support of this conclusion , AUX1::AUX1:YFP and CUC2::RFP were found to be expressed in mutually exclusive domains surrounding outgrowths , with AUX1::AUX1:YFP expression being excluded from the region of high CUC2::RFP expression in the outgrowth axil ( Figure 12C ) . 10 . 7554/eLife . 18165 . 019Figure 12 . LAX1 and AUX1 expression in kan1kan2cuc2 mutants . ( A ) Expression of LAX1::GUS in the abaxial surface of leaf 1 of a kan1kan2cuc2 mutant . Data representative of images from ten seedlings in two separate experiments . ( B ) AUX1::AUX1:YFP expression in leaf 3 and 4 of a kan1kan2cuc2 mutant . Images representative of those taken from fifteen seedlings in two separate experiments . ( C ) CUC2::RFPer and AUX1::AUX1:YFP expression in leaf 3 of a kan1kan2 mutant leaf i ) AUX1::AUX1:YFP , ii ) CUC2::RFPer ( red ) and AUX1::AUX1:YFP ( green ) , iii ) CUC2::RFPer . White arrows indicate region of low AUX1::AUX1:YFP and elevated CUC2::RFPer expression at the distal base of developing outgrowths . Images representative of those taken from six seedlings in two experiments . Approximate leaf widths are given above each image . Scale bars = 50 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 18165 . 019 These findings suggest that CUC2 is required for both the regions of elevated LAX1 expression at outgrowth tips and the regions of low AUX1::AUX1:YFP expression surrounding emerging outgrowths . Thus , CUC2 contributes to locally elevated auxin import at the outgrowth tip . These findings are compatible with a role for CUC2 in promoting convergence formation in the context of a tandem alignment polarity mechanism . We next test whether centres of convergence have elevated auxin removal rates as well as elevated rates of import . One way that auxin could be removed from epidermal centres of convergence is through transport into underlying tissue layers . In the shoot apical meristem , and in serrations , epidermal centres of PIN1 convergence are linked with sub-epidermal strands of PIN1 expression with PIN1 oriented auxin away from the epidermis ( Hay et al . , 2006; Reinhardt et al . , 2003; Scarpella et al . , 2006 ) . To test whether this is also the case for kan1kan2 outgrowths , we immuno-localised PIN1 in transverse cross-sections of kan1kan2 leaves before ( Figure 13A ) and after ( Figure 13B ) outgrowth emergence . At both time points , epidermal centres of convergence were coupled with sub-epidermal strands of cells with elevated PIN1 expression ( Figure 13 ) . PIN1 in these strands was polarised away from the epidermis , towards underlying tissue ( Figure 13A ) , suggesting that the strands provide a route for the removal of auxin from the epidermis . This observation provides a mechanism for auxin removal required for the tandem alignment models ( Figure 10B , D ) . It is compatible with the convergent alignment model ( Figure 11H ) but not required ( Figure 8B ) . 10 . 7554/eLife . 18165 . 020Figure 13 . PIN1 immuno-localisation in transverse cross-sections of kan1kan2 leaves . ( A ) PIN1 immuno-localisation in a kan1kan2 leaf before outgrowth emergence . i ) and ii ) are consecutive sections through the tissue ( each section is 8 µm thick ) . A centre of PIN1 polarity convergence can be seen in i ) and the site of this convergence is shown with yellow arrows in ii ) . Arrows show the inferred directions of PIN1 polarity . The adaxial side of the leaf is at the top of the image . 4 leaves with convergences were imaged across two separate experiments . ( B ) PIN1 immuno-localisation in a kan1kan2 leaf in which outgrowths have begun to emerge . Dotted white lines indicate the strands of cells with elevated levels of PIN1 and the leaf outline . 3 leaves with emerging outgrowths were imaged across two separate experiments . All scale bars = 50 μm , except those in the zoomed in panels in A , which are 10 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 18165 . 020 The epidermis at the leaf base is predicted to have net auxin production by the tandem alignment models , and one way this could be achieved is through elevated auxin biosynthesis . To identify likely locations of elevated auxin production in the leaf , we analysed the expression patterns of genes encoding YUCCA auxin-biosynthesis enzymes in wild-type and kan1kan2 leaves . A . thaliana has six YUCCA genes , three of which ( YUCCA1 ( YUC1 ) , YUCCA2 ( YUC2 ) and YUCCA4 ( YUC4 ) ) are expressed in the leaf and are redundantly required for kan1kan2 outgrowth development ( Cheng et al . , 2007 , 2006; Wang et al . , 2011 ) . YUC1::GUS was previously reported to be expressed at the base of wild-type leaf primordia , in all cell layers ( Wang et al . , 2011 ) . Using altered staining conditions to restrict the diffusion of the coloured GUS product , we found that the expression of this reporter was restricted mainly to outer cell layers at the base of the wild-type leaf , with strongest expression on the adaxial side ( Figure 14A i , ii ) . This adaxial expression was detected in very young wild-type leaf primordia emerging from the shoot apical meristem ( Figure 14B ) , and expression at the leaf base persisted during leaf development ( Figure 14C ) . 10 . 7554/eLife . 18165 . 021Figure 14 . YUCCA1::GUS expression in WT leaves . ( A ) Expression of YUC1::GUS in a WT leaf 1 primordium . i ) abaxial epidermis ( 10 plants were imaged across two separate experiments ) ii ) transverse cross-sectional view of leaf 1 at a similar developmental stage ( adaxial ( ad ) is towards the top ) at the position marked by the dashed black line in i ) . Dashed orange line indicates leaf outline . Data consistent with serial sections taken through 3 other leaves . ( B ) Transverse cross-sectional view of a YUC1::GUS expressing wild-type vegetative meristem . Yellow arrow with black outline indicates region of YUC1::GUS expression at the boundary of an emerging primordium . Dashed orange lines indicate leaf primordia outlines . Data consistent with sections from 3 meristems , across two separate experiments . ( C ) Expression of YUC1::GUS at a later stage of WT leaf development . Scale bars = 50 µm . 9 leaves imaged across two separate experiments . ( D ) In the presence of elevated auxin production in the proximal row of cells ( orange ) , and elevated auxin import and removal in the distal row of cells ( blue cells with green outline ) , the convergent alignment model generates a divergent polarity field , with proximally oriented polarity in the proximal half of the tissue , and distally oriented polarity in the distal half . DOI: http://dx . doi . org/10 . 7554/eLife . 18165 . 021 The expression of YUC1::GUS in the epidermis at the base of primordia matches the predictions of the tandem alignment models . When this region of auxin production is added to the convergent alignment model along with auxin import and removal at the leaf tip , the formation of a proximo-distal polarity field may be disrupted . A divergent polarity field may be established , with proximally oriented polarities towards the base of the tissue , and distally oriented polarities towards the tip ( Figure 14D ) . This pattern emerges because cells with sufficiently elevated rates of auxin synthesis at the leaf base acquire increased intracellular auxin concentrations , causing polarities to orient towards them . However , it is possible that auxin biosynthesis in regions of YUC1 expression does not contribute sufficiently to influence polarity . Compared to wild-type leaves , kan1kan2 mutants tended to have a broader distribution of YUC1::GUS in the proximal half of young leaf primordia ( Figure 15A i ) . Transverse sections through these primordia revealed that the expression domain extended further along the proximodistal axis on the abaxial side of the leaf ( the region where centres of convergence form ) than on the adaxial side ( Figure 15A ii and iii ) . This is consistent with the tandem alignment models for convergence formation , where a broad domain of auxin biosynthesis in the proximal region of the leaf induces cells with elevated auxin import and removal , causing convergence formation ( Figure 10 ) . As in WT , YUC1::GUS expression in the basal-most cells of the leaf persisted and could still be detected in leaves of around 500 µm in width ( Figure 15B iv ) . Consistent with a role of auxin biosynthesis in inducing cells with elevated auxin import ( Figure 10 ) , the majority ( 98% ) of kan1kan2yuc1yuc4 mutant rosette leaves failed to form outgrowths and lacked sites of elevated LAX1::GUS expression within the lamina ( Figure 15C , compare with Figure 11B ) . 10 . 7554/eLife . 18165 . 022Figure 15 . Expression of YUCCA1::GUS in kan1kan2 leaves . ( A ) Expression of YUC1::GUS in kan1kan2 leaf one primordia . i ) abaxial epidermis ( a total of 15 leaves imaged , in three separate experiments ) , ii ) transverse cross-sectional view of leaf 1 at a similar developmental stage , at an approximate position along the proximo-distal axis marked with the pink dashed line in i , iii ) transverse cross section through the same leaf as shown in ii ) , at the approximate position marked with the green dashed line in i . Data consistent with serial sections for 3 other kan1kan2 leaves . ( B ) Expression of YUC1::GUS in kan1kan2 mutant leaf 1 primordia at progressive stages of development . Black arrows indicate the band of YUC1::GUS expression , which is at the distal base of outgrowths in iii ) and iv ) . Arrow heads indicate an outgrowth tip which does not express YUC1::GUS . At least 8 leaves were imaged for each developmental stage , across two experiments . ( C ) Expression of LAX1::GUS in first two leaves of a kan1kan2yuc1yuc4 seedling ( a total of 20 leaves were imaged in three separate experiments ) . ( D ) Expression of YUC1::GUS in leaf one of kan1kan2cuc2 mutant seedlings ( 15 seedlings were imaged at each developmental stage across three separate experiments ) . Scale bars = 50 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 18165 . 022 Before kan1kan2 outgrowths were detected , YUC1::GUS was expressed in an additional band , a few cells wide , running across the abaxial side of the leaf , approximately one-third of the way from the leaf base ( Figure 15B ii , black arrow ) . As outgrowths emerged , the band of YUC1::GUS expression was present at their base on their distal side ( the outgrowth axil ) ( Figure 15B iii and iv , black arrow ) and absent from the outgrowth tips ( Figure 15B iii , black arrow head ) . The expression pattern of YUC1::GUS is similar to that of CUC2::RFP ( Figure 6 ) . To investigate whether CUC2 functions upstream of YUC1 expression in kan1kan2 leaves , we generated kan1kan2cuc2 mutants with the YUC1::GUS reporter . In young kan1kan2cuc2 leaves , the proximal domain of YUC1::GUS expression was reduced in size ( Figure 15D i ) and at later stages of leaf development the ectopic band of YUC1::GUS expression was lost ( Figure 15D ii ) . Thus , CUC2 is required for the ectopic expression of YUC1::GUS in kan1kan2 leaves . To test the potential effects of the pattern of YUC1::GUS expression close to outgrowths on PIN1 polarity patterns , we introduced a pre-pattern of auxin biosynthesis similar to that observed for YUC1 into simulations of the different models . In tandem alignment models , a band of cells with elevated auxin production introduced after the establishment of a proximodistal polarity field generates a region of divergent polarities centred on the band with elevated production ( Figure 16A , C ) . If the band of elevated auxin biosynthesis is introduced together with a cell with elevated auxin import and removal located on the proximal side of the band , a polarity convergence tends to form centred on the cell with elevated import and removal ( Figure 16B , D ) . This behaviour is consistent with the finding that centres of PIN1 polarity convergence are located at the tips of outgrowths , where auxin importer expression is elevated . 10 . 7554/eLife . 18165 . 023Figure 16 . Effect of a band of locally elevated auxin biosynthesis on flux-based , indirect coupling and up-the-gradient models . ( A ) Flux-based model . A proximodistal polarity field is initially established due to elevated auxin production at the leaf base ( orange cells ) and elevated auxin import and removal at the leaf tip ( blue cells with green outline ) . After distally oriented polarities are established , 3 cells are given an elevated rate of auxin biosynthesis ( orange cells ) , causing polarities to diverge away from this region . Graph shows intracellular auxin concentrations for the column of cells marked with the grey arrow . ( B ) As for A , but with a cell with elevated auxin import and removal ( blue cell with green outline ) on the proximal side of the band with elevated auxin synthesis . ( C ) As for A , but for the indirect coupling model . ( D ) As for B , but for the indirect coupling model . ( E ) Up the gradient model . A proximo-distal polarity field is initially established through elevated auxin removal at the leaf base ( blue cells ) and an initially elevated auxin concentration at the leaf tip ( same initial conditions as Figure 8A ) . Three cells ( isolated orange cells ) are then given an elevated auxin biosynthesis rate , causing polarities to orient towards them . ( F ) As for E , but a cell with elevated auxin import and removal is also added on the proximal side of the band with elevated auxin synthesis . DOI: http://dx . doi . org/10 . 7554/eLife . 18165 . 023 In contrast to the tandem alignment models , when a band of elevated auxin synthesis is introduced in the convergent alignment model , polarities tend to orient towards the band ( Figure 16E ) . This may occur even in the presence of a cell with locally elevated auxin import and removal on the proximal side of the band if the rate of auxin import in this cell is not sufficiently high ( Figure 16F ) . This behaviour is inconsistent with the observation that PIN1 polarities orient towards regions with elevated LAX1 and AUX1 expression , even in the presence of a band with elevated YUC1 expression in the outgrowth axil . According to these simulations , the patterns of YUC1-mediated auxin biosynthesis close to kan1kan2 outgrowths are more compatible with the tandem rather than convergent alignment models . However , the presence of YUC expression in regions of polarity divergence at the leaf base and outgrowth axil may be reconciled with the convergent alignment model if it is assumed that YUC expression appears as a consequence of low auxin in these regions following polarity reorientation . Consistent with this possibility , the expression of YUC1 , YUC2 , YUC4 and YUC6 has been shown to be down-regulated by auxin , and upregulated by expression of an auxin biosynthesis inhibitor ( Suzuki et al . , 2015 ) . Given such negative feedback acting upon YUC expression , it is possible that local auxin-biosynthesis by YUCs does not raise auxin concentrations sufficiently to affect PIN1 polarities . YUC2::GUS and YUC4::GUS were previously reported to be expressed at the tips of wild-type leaves , in contrast to expression of YUC1::GUS which is expressed at the base ( Wang et al . , 2011 ) . This expression pattern of YUC2::GUS and YUC4::GUS is more consistent with the convergent alignment model . However , at early stages of leaf development , when YUC1::GUS is expressed at the leaf base , YUC4::GUS and YUC2::GUS were found to be absent from the leaf tip ( Wang et al . , 2011 ) . The proximo-distal PIN1 polarity pattern was only observed in young WT primordia ( Figure 1A , Figure 2A ) . This suggests that , at the time that the proximodistal PIN1 polarity field is observed , the leaf base is the main site of YUC-mediated auxin biosynthesis . YUC2::GUS and YUC4::GUS were also previously reported to be expressed at the tips of kan1kan2 outgrowths ( whether this expression domain is epidermal or sub-epidermal was not shown ) . However , whether expression of these genes in regions of polarity convergence precedes outgrowths or appears later was not determined . We therefore performed time-lapse imaging of a YUC4::GFP reporter to capture the dynamics of YUC4 expression ( Figure 17 ) . This revealed that , approximately one day prior to outgrowth emergence , when centres of PIN1 convergence were previously determined to form , this reporter was expressed in the epidermis , distal to where outgrowths subsequently emerged ( Figure 17 i , ii ) . As outgrowths emerged , they maintained epidermal expression of YUC4 in their axils , the region where YUC1 is also expressed ( Figure 17 iii , iv , compare with Figure 15B ii–iv ) . Emerging outgrowths also showed a second region of expression of YUC4::GFP in epidermal and sub-epidermal tissue at the outgrowth tip ( Figure 17 iii , iv ) . These findings suggest that , at the time that centres of PIN1 polarity convergence form , YUC4 is expressed in the epidermis distal to the centre of convergence . YUC4 expression at the outgrowth tip then appears after convergence formation . The epidermal expression domain distal to polarity convergences in the outgrowth axil is similar to that inferred for YUC1 and is more consistent with tandem than convergent alignment models . Similar to YUC1 , expression of YUC4::GFP within the abaxial lamina was lost in the kan1kan2cuc2 background ( Figure 17B ) . 10 . 7554/eLife . 18165 . 024Figure 17 . Time-lapse imaging of YUC4::GFP in kan1kan2 leaves . ( A ) Confocal images of YUC4::GFP in the same kan1kan2 leaf imaged over a period of 3 days . Times relative to outgrowth emergence , and leaf widths , are given above images . Yellow circles indicate cells located distal to the outgrowth after its emergence , and their ancestors prior to outgrowth emergence . Note that these cells express YUC4::GFP before and after outgrowth emergence , but the region of expression at the outgrowth tip only appears following outgrowth emergence . Data is representative of tracking 5 out of 5 kan1kan2 leaves that developed ectopic outgrowths across two experiments . ( B ) YUC4::GFP expression in leaf one of a kan1kan2cuc2 mutant . Data representative of 13 leaves imaged in two separate experiments . Arrow head indicates stipule . Scale bars = 50 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 18165 . 024
kan1kan2 outgrowths arise ectopically as a consequence of reduced abaxial identity , and thus differ from leaf primordia and serrations . Despite these differences , we show that general elements of outgrowth formation apply to shoot outgrowths ( kan1kan2 outgrowths , primordia and serrations ) regardless of their specific context . 1 ) Similar to leaf primordia and serrations ( Bilsborough et al . , 2011; Hay et al . , 2006; Reinhardt et al . , 2003; Scarpella et al . , 2006 ) , kan1kan2 outgrowths are preceded by epidermal centres of PIN1 polarity convergence . By tracking cells , we show that these centres go on to form outgrowth tips , similar to results from tracking emerging floral primordia ( Heisler et al . , 2005 ) . Thus , as the outgrowth develops , the centre of convergence corresponds to the distal end of a proximodistal field . 2 ) Like primordia and serrations ( Bilsborough et al . , 2011; Heisler et al . , 2005 ) , kan1kan2 outgrowths are preceded by locally elevated intracellular auxin . 3 ) Expression of YUCCA auxin biosynthesis enzymes is elevated in the boundaries that separates leaves , serrations and kan1kan2 outgrowths from the tissues that they emerged from . The axillary boundary ( located in the axils of primordia and distal sides of serrations and kan1kan2 outgrowths ) has divergent PIN1 polarity orientations ( Hay et al . , 2006; Heisler et al . , 2005; Wang et al . , 2011 ) . In the case of YUC4 , we show that axillary expression in kan1kan2 outgrowths precedes outgrowth emergence . 4 ) Centres of polarity convergence at the tips of leaves ( Bainbridge et al . , 2008 ) , serrations ( Kasprzewska et al . , 2015 ) and kan1kan2 outgrowths are associated with regions of elevated epidermal expression of the auxin importer , LAX1 . 5 ) Epidermal centres of PIN1 polarity convergence are coupled to sub-epidermal strands of cells with elevated PIN1 expression and inwardly oriented polarity ( Bayer et al . , 2009; Hay et al . , 2006; Reinhardt et al . , 2003 ) . 6 ) Like primordia and serrations ( Nikovics et al . , 2006; Aida et al 1997 ) , formation of outgrowths is promoted by CUC genes which become strongly expressed distal to the outgrowth . We show that in kan1kan2 leaves , CUC2 promotes formation of PIN1 convergence sites and both elevated YUCCA and auxin importer expression . Taken together , these results support the hypothesis that a common module is associated with the development of outgrowths . This outgrowth module operates iteratively , first in the creation of the leaf with a proximodistal polarity field , and then in the creation of leaf outgrowths such as serration and leaflets ( Hay et al . , 2006; Bilsborough et al . , 2011 ) , or ectopic kan1kan2 outgrowths . Although centres of convergence associated with elevated intracellular auxin and sub-epidermal PIN1 strands were previously shown to be associated with primordia and serrations , here we extend the notion of the outgrowth module to include elevated auxin biosynthesis and auxin import ( features 3 , 4 and 6 above ) and show that it also applies to outgrowths emerging from the plane of the leaf . The observation that a loss of PIN1 convergences and outgrowths in kan1kan2cuc2 mutants is correlated with a loss of ectopic auxin biosynthesis and import supports an important role of these features in the outgrowth module . These observations raise the question of how the various features of the outgrowth module are generated . We found that models for PIN1 convergence site formation ( feature 1 ) fall into two groups . Convergent alignment models ( e . g . up-the-gradient ) spontaneously generate spaced centres of polarity convergence in a in a field of cells with initially uniform auxin distribution ( Figure 7D , Jönsson et al . , 2006; Smith et al . , 2006; Cieslak et al . , 2015; Heisler et al . , 2010 ) . By contrast , tandem alignment models ( e . g . flux-based and indirect coupling ) generate swirled patterns of tandemly aligned polarity in this context ( Figure 7E , F ) . Convergent alignment models provide the most parsimonious explanation for the observed PIN1 polarity ( feature 1 ) , intracellular auxin ( feature 2 ) , and spacing of outgrowths , as all of these aspects arise naturally in a uniform field ( Bilsborough et al . , 2011; Jönsson et al . , 2006; Smith et al . , 2006 ) . However , there is no expectation of elevated expression of auxin biosynthesis enzymes in the regions of polarity divergence in the axils of leaves and their outgrowths ( feature 3 ) . Indeed , auxin biosynthesis in these regions can cause polarities to orient towards regions of elevated auxin biosynthesis , rather than away from them as is observed experimentally ( Hay et al . , 2006; Heisler et al . , 2005; Wang et al . , 2011 ) . The convergent alignment model may be reconciled with the observed expression of YUC1 and YUC4 auxin biosynthesis genes in regions of polarity divergence if it is assumed that this expression does not raise auxin levels sufficiently to influence polarity . Rather than playing a role in polarity reorientation , YUC expression may then be a consequence of low auxin in these regions following polarity reorientation , as expression of YUC genes has been shown to be down-regulated by auxin ( Suzuki et al . , 2015 ) . There is also no expectation of elevated auxin import in regions of polarity convergence ( feature 4 ) for the convergent alignment model . The parsimony of the convergent alignment model arises from the reinforcement of auxin levels at centres of convergence through auxin transport without any need for additional mechanisms , such as elevated auxin import . Elevated auxin importer expression can be incorporated within a convergent alignment model , by postulating that high levels of intracellular auxin induce auxin import . This approach was adopted to account for the effect of auxin import on leaf serration development , and with suitable parameters auxin import can play a role in reinforcing convergent alignment ( Kasprzewska et al . , 2015 ) . However , with some convergent alignment models , incorporation of auxin import may have a disruptive effect . It has been proposed that cells may orient polarity up auxin gradients by responding to auxin levels in their neighbours indirectly by sensing auxin flux ( Cieslak et al . , 2015 ) . Assuming all cells in a tissue have an equal capacity to import and export auxin , cells with elevated intracellular auxin concentration will have higher net rates of auxin flux away from them , into neighbouring cells . This flux could be sensed to orient PIN1 allocation up the auxin gradient . However , this behaviour would be disrupted by elevated levels of auxin importer expression at centres of convergence , because elevated auxin import in the high auxin cells would reduce the net flux from them into their neighbours . Based on PIN1 data and intracellular auxin levels alone ( features 1 and 2 ) , tandem alignment models provide a less parsimonious explanation than convergent alignment models . Special mechanisms need to be postulated for tandem alignment models for PIN1 convergence sites to form , such as induction of regions with elevated auxin import and removal . The observation of sub-epidermal strands of cells with inwardly oriented PIN proteins ( feature 5 ) fits the requirement of auxin removal for convergence formation in the tandem alignment models . However , auxin removal alone , through formation of sub-epidermal PIN1 strands ( Stoma et al . , 2008 ) would lead to a transient dip in epidermal auxin concentrations , which is not observed experimentally ( Brunoud et al . , 2012; Heisler et al . , 2005 ) . Here , in the simulations we present , the transient dip does not occur if centres of convergence have elevated auxin import in addition to removal . The observed elevated levels of auxin import at convergence centres ( feature 4 ) is therefore an expected feature for tandem alignment models . A functional role for auxin import is also supported by the observed disruption of outgrowth initiation in aux1lax1lax2lax3 meristems ( Bainbridge et al . , 2008 ) and kan1kan2aux1lax1lax2lax3 leaves . A further property of tandem alignment models is that polarities tend to become oriented away from auxin sources . It is therefore expected that sites of divergent polarity , such as the axillary regions of outgrowths would be auxin sources , compatible with elevated expression of auxin biosynthesis enzymes in these regions ( feature 3 ) . Also supporting these predictions of tandem alignment models , loss of ectopic auxin biosynthesis and import in the kan1kan2cuc2 mutant correlates with a loss of PIN1 convergence points and ectopic outgrowths . Thus , when the auxin import and biosynthesis data are considered along with the PIN1 polarity and intracellular auxin levels , tandem alignment models may give a more a parsimonious explanation for the data as a whole than convergent alignment models . The contribution of CUC gene activity ( feature 6 ) can be incorporated into both convergent and tandem alignment models . Our data suggest that in wild-type leaves CUC2 expression is excluded from the abaxial lamina and confined to serrations by the action of abaxial identity determinants including KAN1 and KAN2 . According to the convergent alignment model , CUC2 may be required to enable PIN1 polarity to become reoriented rather than remaining fixed in the abaxial lamina as occurs in wild type ( Bilsborough et al . , 2011 ) . According to the tandem alignment models , ectopic expression of CUC2 could drive auxin synthesis ( YUCCA expression ) and induction of ectopic sites of elevated auxin import and removal , and thus centres of convergence . CUC2 expression distal to the outgrowth could also maintain these regions as sites of high auxin synthesis and polarity divergence . The results presented here highlight distinctive elements in models for tissue cell polarity in plants and animals . The ability of all three plant models analysed to generate cell polarity in a non-polarising context ( Figure 7A–C ) contrasts with models of animal planar cell polarity ( Strutt and Strutt , 2009 ) , where polarity is proposed to either be dependent on pre-established asymmetric cues ( Amonlirdviman , 2005; Fischer et al . , 2013; Le Garrec et al . , 2006 ) or polarisable neighbours ( Burak and Shraiman , 2009 ) . Models may therefore be classified into two broad categories: those in which polarity of a cell depends on a polarising environment and those that do not ( Figure 18 ) . Although most animal models belong to the first category , a version that does not require a polarising environment , termed direct coupling , has been proposed ( Abley et al . , 2013 ) . 10 . 7554/eLife . 18165 . 025Figure 18 . Classification of plant and animal polarity models . Models of plant and animal epidermal polarity may be classified according to whether they require the presence of pre-established asymmetries or polarisable neighbours ( referred to as a polarising environment ) for cell polarity to arise . All plant models tested can generate cell polarity without a polarising environment , whilst animal models except the direct coupling model ( [4] [Abley et al . , 2013] ) either require an asymmetric signal ( [1] [Le Garrec et al . , 2006] [2] [Amonlirdviman , 2005] ) or polarisable neighbours ( [3] [Burak and Shraiman , 2009] ) for cell polarity to arise . A further classification is made according to the behaviours of groups of cells in the absence of pre-established asymmetries . In this context , the indirect coupling ( [5] [Abley et al . , 2013] ) and flux-based ( [6] [Mitchison , 1980; Rolland-Lagan and Prusinkiewicz , 2005; Stoma et al . , 2008] ) models of plant polarity generate swirled patterns of polarity with tandem alignments between cells . In contrast , the up-the-gradient model ( [7] [Bilsborough et al . , 2011; Jönsson et al . , 2006; Smith et al . , 2006] ) generates convergent alignments of cell polarities . The stress-based model of PIN1 polarity ( [8] [Heisler et al . , 2010] ) also fits in the convergent alignment category as it generates similar behaviours with respect to auxin concentrations to the up-the-gradient model . The animal models that are capable of polarising in the absence of pre-established asymmetries generate tandem alignments of polarities ( [4] direct coupling model [Abley et al . , 2013] , and ( 3 ) Burak and Shraiman , 2009 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 18165 . 025 For plant models , cells may polarise in a non-polarising environment in two ways . One is through small fluctuations in polarity leading to altered auxin distribution or flux which then feeds back to enhance polarity . This occurs in the up-the-gradient and flux-based models ( Figure 7A , B [van Berkel et al . , 2013] ) . It contrasts with current animal models in which polarity does not feed back to influence the distribution of the extracellular ligand . Another class of plant models proposes that cells become polarised without requirement for altered distribution or flux of auxin , through processes internal to the cell , as assumed for indirect coupling ( Figure 7C ) . It remains to be seen whether the indirect coupling model can also capture the generation of sub-epidermal strands of cells with elevated levels of membrane-localised PIN1 ( feature 5 ) . Such strands are observed experimentally during pro-vascular development and can be accounted for with a flux-based model assuming superlinear feedback between flux and PIN1 allocation ( Scarpella et al . , 2006; Bayer et al . , 2009; Rolland-Lagan and Prusinkiewicz , 2005; Feugier et al . , 2005; Stoma et al . , 2008 ) . Our analysis illustrates the value of comparing multiple models when analysing experimental data rather than focusing on one in isolation . Each model alone can be modified to account for experimental observations , showing that fitting to data is not conclusive . However , models may differ in the predictions they make prior to the data being obtained , and the parsimony with which experimental data can be accounted for . Moreover some models may make very similar predictions even though underlying molecular mechanisms are different . For example , two types of tandem alignment model - flux-based and indirect coupling - give similar predictions for auxin import and synthesis . Similarly , a concentration-based or stress-based version of the convergent alignment model gives similar predictions for auxin turnover . Other types of experiment will be needed to discriminate between these further levels of explanation .
All kan1kan2 mutants carry kan1-2 and kan2-1 alleles in the Landsberg erecta ( Ler-0 ) background ( Eshed et al . , 2001 ) except in the kan1kan2yuc1yuc4 background , where plants carry kan1-11 and kan2-5 alleles ( Wang et al . , 2011 ) . The cuc2 mutant allele used is cuc2-3 and was initially in the Columbia ( Col-0 ) background ( Hibara et al . , 2006 ) . All marker lines were originally in the Col-0 background . PIN1::PIN1:GFP , DR5::GFP ( Benková et al . , 2003 ) ; LAX1::GUS , LAX2::GUS , LAX3::GUS ( Bainbridge et al . , 2008 ) ; YUC1::GUS , YUC2::GUS , YUC4::GUS ( Cheng et al . , 2006 ) were described previously . The AUX1::AUX1:YFP line corresponds to the AUX1-YFP116 construct described by ( Swarup et al . , 2004 ) . The CUC2::RFP line was kindly provided by Patrick Laufs prior to publication . The YUC4::GFP reporter was provided by Yunde Zhao and Youfa Cheng . The YUC4 promoter was amplified with the following primers: Y4pro3p_XmaI: cccggggtcgactaataaaagcgaaagagag and Y4pro5p_HindIII: aagcttatgtccaacatgcatgcg . The 2916 bp PCR product was digested with HindIII and XmaI , and cloned into pBJ36-ERGFP vector . The promoter was subcloned from pBJ36-ERGFP into the pART27 vector and transformed into Col-0 background by floral dipping and subsequently selecting for kanamycin resistant transgenic plants . All images of marker construct expression in WT were obtained using the original marker lines in the Col-0 background . The kan1kan2cuc2 mutant was generated by crossing cuc2 mutants with kan1kan2+/- plants ( kan1kan2 mutants are sterile ) . In the F2 of this cross , kan1kan2+/- cuc2 individuals were identified by screening for the kan1kan2+/- abberent fruit phenotype , and by PCR-based screening for cuc2-3 homozygous individuals . The presence of a T-DNA insertion in the CUC2 locus was detected by using a forward primer , which binds to the T-DNA: 5’-TCCATAACCAATCTCGATACAC-3’ and a reverse primer which binds within the CUC2 locus downstream of the T-DNA insertion ( referred to as CUC2-R ) : 5’- GGAGGCTAAAGAAGTACCATTC-3’ . The presence of a WT copy of CUC2 was detected using a forward primer CUC2-F: 5’-AATATCCATCCACATTATTACCAC-3’ , which binds to CUC2 upstream of the site of the T-DNA insertion in cuc2-3 , along with CUC2-R . One quarter of the offspring of kan1+/- kan2 cuc2 individuals were kan1kan2cuc2 mutants , and could be identified at the seedling stage ( and therefore selected for experiments ) based on the upwardly curled cotyledon phenotype of kan1kan2 mutants . Marker constructs were introduced into the kan1kan2 and kan1kan2cuc2 backgrounds by carrying out the corresponding crosses using kan1kan2+/- or kan1kan2+/-cuc2 mutants . The kan1kan2aux1lax1lax2lax3 , kan1kan2aux1lax1lax2 and kan1kan2aux1lax1 mutants were generated by crossing kan1kan2+/- plants with the aux1lax1lax2lax3 quadruple mutant ( Bainbridge et al . , 2008 ) . In segregating F2s from this cross , kan1kan2+/- plants with desired mutant alleles of aux/lax genes were selected as described in Bainbridge et al . , 2008 . The kan1kan2+/- line used to generate the kan1kan2aux1lax1lax2lax3 mutants was in the Ler-0 background , and the aux/lax quadruple mutant was in the Col-0 background . To check that the loss-of outgrowth phenotype in hextuple mutants was not due to the introduction of an unknown genetic element from the Col-0 background , we determined whether the reduction in the number of outgrowths segregated with the mutant LAX1 allele . An F3 family of plants was used that segregated for mutations in KAN2 and LAX1 , but was homozygous for kan1 , aux1 , lax2 and lax3 mutant alleles . Of 156 kan1kan2aux1lax2lax3 plants segregating for lax1 , 120 ( 77% ) developed many outgrowths and 36 ( 23% ) had a reduced number and size of outgrowths . All of the plants with few outgrowths were homozygous for the lax1 mutation , whilst all siblings with many outgrowths that were genotyped ( 40 in total ) all had a wild-type copy of LAX1 . Such segregation analyses were not carried out for kan1kan2aux1lax1 and kan1kan2aux1lax1lax2 mutants thus it is possible that the reduction in outgrowths in these lines is due to introduction of an unknown genetic element during the cross between Col-0 and Ler backgrounds . However , two independent families of kan1kan2aux1lax1lax2 plants ( derived from different F2 plants ) had similar losses of outgrowths ( data provided in source data file for Figure 11G ) . The loss of outgrowths in kan1kan2aux1lax1lax2lax3 plants was also observed in two independent families quantified in two different experiments ( see source data file for Figure 11G ) . Plants were sown on plates for confocal imaging experiments and to select those to be used for crosses . Seeds were surface-sterilized before plating on Petri dishes with solid Murashige and Skoog medium as described ( Sauret-Güeto et al . , 2013 ) and grown in controlled environment room at 20°C in long day conditions ( 8 hr dark and 16 hr light under fluorescent white light at a photon fluence rate of 100 µmol·m−2·s−1 ) . When appropriate , medium was supplemented with Kanamycin 50 µg/ml ( to select for PIN1::PIN1:GFP , AUX1::AUX1:YFP , LAX1::GUS , LAX2::GUS , LAX3::GUS , YUC1::GUS and YUC4::GFP constructs ) . kan1kan2 plants carrying aux/lax or cuc2 mutant alleles were sown on soil prior to characterisation of their phenotype . Seeds were stratified directly ( without sterilisation ) and then sown in John Innes Centre A . thaliana Soil Mix ( Levington F2 compost with Intercept and grit at a 6:1 ratio ) and were grown under long-day conditions ( 16 hr light and 8 hr dark ) in a glasshouse , supplemented with artificial light , at approximately 22°C . The kan1kan2yuc1yuc4 mutants used are as described in ( Wang et al . , 2011 ) . kan1kan2+/-yuc1+/-yuc4 LAX1::GUS plants were selected in an F2 of a cross between kan1kan2+/- yuc1+/- yuc4 plants and LAX1::GUS plants . F2 seedlings with the LAX1::GUS construct were by selected for sowing seeds on plates containing 50 µg/ ml Kanamycin . Those that were kan1kan2+/- mutants were identified based on their fruit phenotype and the presence of the yuc1+/-yuc4 genotype was selected for by PCR-based genotyping . The yuc1 and yuc4 alleles have T-DNA insertions , and mutant alleles were screened for using the following primers for yuc1: LBb1: 5’-GCGTGGACCGCTTGCTGCAAC-3’ and YUC1-R: 5’-CCTGAAGCCAAGTAGGCACGTT-3’ and the following primers for yuc4: LBb1: 5’-GCGTGGACCGCTTGCTGCAAC-3’ and YUC4-R: 5’-GCCCAACGTAGAATTAGCAAG-3’ . WT copies of YUC1 were screened for using YUC1-F: 5’-GGTTCATGTGTTGCCAAGGGA-3’ and YUC1-R and WT copies of YUC4 were screened for using YUC4-F 5'-CCCTTCTTAGACCTACTCTAC-3’ and YUC4-R . To analyse the expression of LAX1::GUS in kan1kan2yuc1yuc4 mutants , the offspring of kan1kan2+/-yuc1+/-yuc4 LAX1::GUS or kan1kan2+/yuc1+/-yuc4 LAX1::GUS+/- plants were sown on plates containing 50 µg/ ml Kanamycin and seedlings with a kan1kan2 leaf phenotype ( narrow , upwardly pointing leaves ) were selected , genotyped for YUC1 and YUC4 , and stained to reveal GUS activity . Of these seedlings , one quarter did not have outgrowths , and genotyping showed they were kan1kan2yuc1yuc4 mutants . For tracking experiments , seedlings were first grown on plates ( until 5 days after stratification for kan1kan2 and kan1kan2cuc2 mutants , and until 4 days after stratification for WT seedlings ) . Seven seedlings were then transferred into a tracking chamber ( Sauret-Güeto et al . , 2013; Calder et al . , 2015 ) where they were kept for the duration of the imaging experiment . During this period , there was a constant flow of liquid media ( 1/4 strength Murashige and Skoog , 0 . 75% sucrose , 1 . 1 μg / ml MES , pH 5 . 8 ) at 1 μl /s through the growth chamber . Seedlings in the chamber were imaged using a Zeiss EXCITER Laser Confocal Microscope every 6 to 24 hr ( times are provided for specific experiments in figure legends ) . To image GFP and YFP markers , a 488-nm line of an argon ion laser was used . Emitted light was filtered through a 500-550-nm band-pass filter . To image RFP markers , a 543 nm helium-neon laser was used and emitted light was filtered using a 560-615-nm band-pass filter . A 40x oil objective was used for all experiments . Between imaging , the chamber containing seedlings was kept at 20°C , with 16 hr light , 8 hr dark cycles . In confocal imaging experiments without tracking , seedlings were mounted in water on microscope slides . Confocal z-stacks were converted into individual PNG images using Bioformats converter ( http://cmpdartsvr3 . cmp . uea . ac . uk/wiki/BanghamLab/index . php/BioformatsConverter ) . Z-stacks of PNG images were then rendered in 3D using Volviewer ( http://cmpdartsvr3 . cmp . uea . ac . uk/wiki/BanghamLab/index . php/VolViewer#Download ) . All snapshots shown in figures were taken from Volviewer . Leaf widths were measured from maximum intensity projections in image J and thus do not take into account the 3D curvature of the leaf . OPT imaging was performed as described by ( Lee et al . , 2006 , Sharpe et al . , 2002 ) . Following PI staining ( described below ) , plants were embedded in 1% low melting point agarose and embedded samples were kept in methanol overnight to dehydrate . To clear the agarose before scanning , the methanol was replaced with 2 parts ( v/v ) Benzyl Benzoate , 1 part Benzyl Alcohol and samples were left for 12 hr until almost transparent . A prototype OPT device was used to image embedded leaves ( Lee et al . , 2006; Sharpe et al . , 2002 ) . A 20-W halogen lamp connected to the OPT device was used to collect visible light transmission images . Images were reconstructed into png slices and visualised in 3D using VolViewer . For propidium iodide staining , dissected leaves were placed into 10% ( v/v ) acetic acid , 50% ( v/v ) methanol solution and kept at 4°C overnight . Samples were then washed with water and dehydrated with two washes each in 40% ( v/v ) ethanol , 60% ( v/v ) ethanol and then 80% ( v/v ) ethanol . The samples were boiled in 80% ( v/v ) ethanol at 80°C for 2–10 min in a water bath and then rehydrated with two washes each in 60% ( v/v ) ethanol , 40% ( v/v ) ethanol , 20% ( v/v ) ethanol , and water . Samples were then treated with alfa-amylase [3 mg of alpha-amylase ( SIGMA , UK ) in 10 ml of 20 mM phosphate buffer ( pH 7 ) , 2 mM NaCl , 0 . 25 mM Ca2Cl] overnight at 37°C , followed by three water washes , and then treated with 1% periodic acid [1% in solution from SIGMA , UK] ( an oxidising agent ) for 40 min at room temperature . Following two washes with water , PI staining solution [333 mM sodium metabisulphite , 0 . 5 M HCl , 148 μM PI] was added and incubated for 1–2 hr at room temperature until the material appeared pink in colour . PI staining solution was then removed and the samples were washed with water . GUS staining with and without wax embedding was performed as described by Sessions et al . ( 1999 ) , but here using 6 mM Potassium Ferrocyanide and 6 mM Potassium Ferricyanide and with an 18 hr incubation with the GUS staining solution . Immuno-localisation of PIN1 was performed as described ( Conti and Bradley , 2007 ) , using a goat anti-PIN1 ( aP-20 ) polyclonal primary antibody ( Santa Cruz Biotechnology , TX , USA ) ( RRID:AB_670528 ) diluted 1:300 and a Cy3 conjugated rabbit anti-goat secondary antibody ( Jackson Immunoresearch , PA , USA ) , diluted 1 in 300 . All models were created using the VVe modelling environment , an extension of the VV system ( Smith et al . , 2003 ) , which in turn is an extension of L-systems . All models are provided as supplementary source code files . Each model can be run using L-studio with VVe , which can be downloaded from http://algorithmicbotany . org/virtual_laboratory/ . Instructions on how to run models are available as supplementary information ( Supplementary file 1 ) . The indirect coupling model was implemented based on the description in Abley et al . , 2013 except in some simulations we use rectangular rather than hexagonal cell geometries and we introduce an explicit representation of PIN1 to all models . The details of these differences are described below . All parameter values are as described in Abley et al . , 2013 , unless otherwise detailed in Table 1 . 10 . 7554/eLife . 18165 . 026Table 1 . Parameter values used in simulations of the indirect coupling model . DOI: http://dx . doi . org/10 . 7554/eLife . 18165 . 026SymboldescriptionunitvalueΔtnumerical time steps ( seconds ) 0 . 05Rcarea of cytoplasmic compartmentµm2260Rwarea of cell wall compartmentµm22 . 88 ( for long cell edges ) ; 3 . 0 ( for short cell edges ) lnlength of membrane compartmentsµm2 . 88; 3 . 0lwlength of wall compartmentsµm2 . 88; 3 . 0cPINdefault initial concentrations of PIN1 in cytoplasmic compartmentsAu/ µm20 . 003ρPINPIN1 default membrane binding rateµm/s0 . 03τA*-dependent promotion of PIN1 bindingµm2/Au . s2µPINPIN1 default membrane unbinding rate/s0 . 004DPINPIN1 diffusion in cell membraneµm2/s0 . 1ΨPINPIN-dependent active auxin efflux rateµm2/Au . s40ɛlimit for noise addition during initialisation of A* and B* concentrationsdimensionless0 . 0166γAuxAuxin-dependent promotion of A* to A conversionµm2/Au . s0 . 75*ρAuxproduction rate of AuxinAu/µm2 . s1 . 0 × 10−4*µAuxdegradation rate of Auxin/s0 . 01*vininflux auxin permeabilityµm/s0 . 75*All parameters relate to either equations specified here or in Abley et al . , 2013 . * In the simulations used to generate Figure 7C and Figure 7—figure supplement 1C , γAux = 0 . 9 µm2/Au . s , ρAux = 1 . 3 × 10−4 Au/µm2 . s and µAux = 0 . 02/s . Also , in some simulations , at tissue boundaries or centres of polarity convergence , auxin production , degradation and influx auxin permeability rates may vary from the background rates . Modulation of the influx auxin permeability rate is used to simulate elevated rates of auxin import . In Figure 9C , the auxin production rate in the bottom-most row of cells = 2 × 10−4 Au/µm2 . s , and the auxin degradation rate in the top most row of cells = 0 . 07/s . In Figure 9D , and at the beginning of the simulation used to generate Figure 10C and D the auxin production rate in the bottom-most row of cells = 2 × 10−4Au/µm2 . s . Throughout both simulations , in the top-most row , the auxin degradation rate = 0 . 015/s and the auxin influx permeability = 4 . 5 µm/s . In the simulation used to generate Figure 10C and D , changes in auxin production occur during the simulation and are described below . In this simulation , in the cell that forms the centre of convergence , the inwards permeation of auxin = 37 . 5 µm/s and the auxin degradation rate = 0 . 15/s . In Figure 16C and D , the proximo-distal polarity field is established as described for Figure 9D . In the three cells with elevated auxin production , the auxin production rate is 1 . 4 × 10−3Au/µm2 . s . In the cell with elevated auxin import and removal ( Figure 16D ) , the auxin degradation rate = 0 . 1 /s and the auxin influx permeability = 37 . 5 µm/s . In the simulations used to generate Figure 10—figure supplement 1A and B , parameter values are as for Figures 9D and 10D , respectively except ΨPIN = 80 µm2/Au . s and γAux = 0 . 85 µm2/Au . s . Here , c , the concentration of A* at which half of the PINs are activated = 0 . 22 Au/µm and the hill coefficient , h = 2 . Each rectangular cell is represented by a single central cytoplasmic compartment , 22 membrane compartments ( 6 membrane compartments for each long edge and 5 membrane compartments for each short edge ) , and surrounding cell wall compartments . Each cell wall compartment links two membrane compartments located in adjacent cells . The system is initialised with a default concentration of PIN1 in the cytoplasmic compartment of all cells . The rate of change of PIN1 concentration in a given membrane compartment depends on a background binding rate of PIN1 from the cytoplasm to the membrane , plus default unbinding from the membrane into the cytoplasm . It is also assumed that A* in a given membrane compartment promotes the binding of PIN1 to that membrane compartment and that PIN1 can diffuse between adjacent membrane compartments of the same cell . The equation describing the rate of change of PIN1 concentration in a given membrane compartment is: ( 1 ) ∂PIN∂t= ( ρPIN+τA∗ ) PINc−μPINPIN+DPIN∇2PIN where PIN is the concentration of PIN1 in the focal membrane compartment , with units of Arbitrary units ( Au ) / µm and PINC is the concentration of PIN1 in the cytoplasmic compartment of the same cell as the focal membrane compartment , with units of Au/µm2 . ρPIN is the default binding rate of PIN1 to the membrane , with units of µm/s , and τ is a constant describing the rate at which membrane-bound A* promotes the binding of PIN1 to the membrane , with units of µm2/Au . s . A* is the concentration of the A* polarity component in the focal membrane compartment , with units of Au/ µm . µPIN is the default unbinding rate of PIN1 from the membrane into the adjacent cytoplasmic compartment , with units of /s . DPIN is the diffusion constant of PIN1 in the membrane , with units of µm2/s . The corresponding equation describing the rate of change of PIN1 concentration for a given cytoplasmic compartment is: ( 2 ) ∂PINc∂t=−1Rc∑n∈N ( c ) ln ( ( ρPIN+τΑn* ) PINc−μPINPINn ) Where PINC is the concentration of PIN1 in the focal cytoplasmic compartment and PINn is the concentration of PIN1 in the membrane compartment n , in the neighbourhood of the cell c ( N ( c ) ) . Rc is the area of the cytoplasmic compartment and ln is the length of the nth membrane compartment . A*n is the concentration of A* in the membrane compartment n . ρPIN , τ and µPIN are as described above for Equation 1 . The export of auxin from a cytoplasmic compartment to a wall compartment depends on the concentration of PIN1 in the intervening membrane compartment . The equation presented by Abley et al . , 2013 , describing the rate of change of auxin ( referred to as mediator ( M ) in Abley et al . , 2013 ) , in cytoplasmic compartment i becomes: ( 3 ) dAuxidt=ρAux−μAuxAuxi+1Rclw∑n∈N ( c ) ( vinAuxw−voutAuxi−ψPINPINnAuxi ) Where Auxi is the concentration of auxin in the cytoplasmic compartment i , Auxw is the concentration of auxin in the wall compartment neighbouring the membrane compartment n in the neighbourhood of the cell c ( N ( c ) ) , ρAux is the production rate of auxin with units of Au/µm2 . s , Rc is the area of the cytoplasmic compartment , and µAux is the degradation rate of auxin with units of /s . PINn is the concentration of PIN1 in the nth membrane compartment of the cell . νin is the background permeation rate of auxin into the cytoplasm from the wall with units of µm/s and νout is the background permeation rate of auxin into the wall from the cytoplasm with units of µm/s . lw is the length of the cell wall compartment into/out of which auxin flux is occurring and Rc is the area of the cytoplasmic compartment . ψ PIN is the rate of PIN1-dependent active efflux of auxin from the cytoplasm into the wall with units of µm2/ Au . s . The equation describing the rate of change of auxin in wall compartments becomes: ( 4 ) ∂Auxwdt=−1Rwlw∑n∈N ( w ) ( vinAuxw−voutAuxc−ψPINPINnAuxc ) +DAux∇2Auxw where Rw is the area of the wall compartment , lw is the length of the wall compartment , Auxw is the concentration of auxin in the wall compartment , An* is the concentration of A* in the membrane compartment n in the neighbourhood of the given wall compartment w ( N ( w ) ) and Auxc is the concentration of auxin in the cytoplasm of the same cell as the membrane compartment n . ψ PIN is as described above and PINn is the concentration of PIN1 in the membrane neighbor , n , in the neighbourhood of the wall compartment w ( ( N ( w ) ) . DAux is the diffusion constant for auxin within the cell wall with units of µm2/s ( this constant relates to lateral diffusion between wall compartments as it is assumed that the concentration of auxin is uniform across the thickness of the wall ) . In the simulation used to generate Figure 10—figure supplement 1 , we introduce D6 kinase-like activity into the model by assuming that only a fraction of PINs are active and able to export auxin , and that this fraction depends on the local concentration of A* in the membrane . A* therefore represents D6 kinase-like activity . In this scenario , the term describing PIN1-mediated auxin efflux from the cell to the wall in Equations 3 and 4 is modified to become: ( 5 ) An∗hAn∗h+chΨPINPINnAuxc Where A*n is the concentration of A* in the same membrane compartment as PIN , h is the Hill coefficient ( here a value of 2 was used , giving a sigmoidal function ) , and c is the concentration of A* at which half of the PINs are activated . ψ PIN , PINn and Auxc are as described above . In all simulations of the indirect coupling model , there is no flux of auxin across the boundaries of the tissue . In the simulation used to generate Figure 10C and D ( positioning of cells with elevated auxin import and removal ) , in the first 6500 s of the simulation , the production rate of auxin , ρAux , in all cells in the proximal half of the array ( except those in the proximal-most row ) is 1 . 5 × 10−4 Au/ µm2 . s ( in the proximal most row ρAux = 2 × 10−4Au/µm2 . s ) . Then , between 6500 and 10 000 s of the simulation , the auxin production rate of all cells in the proximal half of the array ( including those in the proximal-most file ) is increased at every time step of the simulation: ( 6 ) dρAuxdt=β Where ρAux is the auxin production rate of a given cell and β is a constant describing the rate of increase of the auxin production rate , with units of Au/ µm2 . s2 ( β = 5 × 10−6 Au/ µm2 . s2 ) . As a consequence of this increase in the auxin production rate , after 10 000 s of the simulation , in the proximal-most file of cells , ρAux = 3 . 75 × 10−4 Au/ µm2 . s , and in the other cells in the proximal-most half of the leaf , ρAux = 3 . 25 × 10−4 Au/ µm2 . s . After 6 500 s of the simulation , if the auxin concentration of a cell exceeds a threshold , Timport , the inwards permeability of the cell to auxin is increased 50-fold ( vin = 37 . 5 µm /s ) and the auxin degradation of the cell is increased 10-fold so that µAux = 0 . 1 /s . Timport = 0 . 0285 Au/ µm2 . In the simulation used to generate Figure 10C and D , a noise term is added to the concentrations of auxin every 0 . 05 s of the simulation , according to the following equation: ( 7 ) Auxn= Aux+ ( θAux∗ Aux ) where Auxn is the concentration of auxin in a given cytoplasmic compartment after the addition of noise with units of Au/µm2 , Aux is the concentration of auxin in that compartment before the addition of noise ( with the same units ) , θAux is a random number drawn from a normal distribution , with mean 0 , and standard deviation 5 × 10−4 . Noise is added to the concentration of auxin in proportion to the square root of the auxin concentration . The flux-based model was implemented using a similar discretisation of cells as in previous implementations ( Rolland-Lagan and Prusinkiewicz , 2005; Stoma et al . , 2008 ) . Each cell is represented by a single central cytoplasmic compartment surrounded by six peripheral compartments representing cell edges ( one for each cell-cell interface ) . There is no representation of the cell wall . Each edge compartment of a cell is connected to the cytoplasmic compartment of the same cell and to the juxtaposed edge compartment of the neighbouring cell , unless the vertex is at a border of the tissue . The model is implemented based on previously described implementations ( Rolland-Lagan and Prusinkiewicz , 2005; Stoma et al . , 2008 ) . At the beginning of all simulations , the system is initialised with auxin in cytoplasmic compartments and PIN1 at edge compartments . The system may be initialised with noisy auxin concentrations in each cytoplasmic compartment , i: ( 8a ) Auxi ( t=0 ) =cA ( 1+θAux ) ( 8b ) θAux∈[−εAux , εAux] where Auxi ( t = 0 ) is the initial concentration of auxin in a given cytoplasmic compartment , i , with units of Au/ µm2 and cA is the default initial concentration of auxin at an edge compartment . θAux is a random number uniformly distributed between an upper and lower limit , ɛAux . The system may also be initialised with noisy PIN1 concentrations at each cell edge: ( 9a ) PINedge ( t=0 ) =cPINedge ( 1+θPIN ) ( 9b ) θPIN∈[−εPIN , εPIN] where PINedge ( t = 0 ) is the initial concentration of PIN1 at a given edge compartment , with units of Au/ µm and cPINedge is the default initial concentration of PIN1 at an edge compartment . θPIN is a random number uniformly distributed between an upper and lower limit , ɛPIN . PIN1 is recruited to a cell edge depending on the rate of auxin efflux across that edge , and PIN1 is removed from the edge at a background rate . The equation describing the rate of change of PIN1 concentration for a given cell edge , between cells i and j , is ( 10 ) dPINijdt={αϕi→j−γPINijifϕi→j≥0−γPINijifϕi→j<0 Where PINij is the concentration of PIN1 in cell i , at the cell edge between cells i and j , α is a dimensionless constant determining the extent to which flux promotes PIN1 allocation to the membrane and Φi→j is the auxin flux across the interface between cells i and j , with units of Au / µm . s . Outgoing fluxes , from cell i , to cell j are considered to be positive , and incoming fluxes are considered to be negative . γ is the unbinding rate of PIN1 from the cell edge , with units of /s . As previously described ( Rolland-Lagan and Prusinkiewicz , 2005 ) , we assume that once the concentration of PIN1 at the cell edge reaches a threshold value , Pmax , PIN1 can no-longer be allocated to the edge . This is equivalent to assuming that there is a maximum density of PIN1 proteins which may be present in the plasma membrane . In all versions of the model , the flux across a given cell edge , from cell i to cell j , at a given time point depends on passive permeability of auxin between cells and PIN-mediated auxin export . In the simulations used to generate Figures 9B , 10A , B and 16A , B , we assume that auxin may also be actively imported into cells , and that the rate of import depends on the auxin import rate of a given cell , and the auxin concentration in its neighbour . The flux across a given cell edge , from cell i to cell j , at a given time point is therefore given by ( 11 ) Φi→j=D ( Ai−Aj ) +T ( PINijAi−PINjiAj ) +IjAi−IiAj Where Φi→j is the flux across a given cell-cell interface , from cell i to cell j , D is a constant describing the passive permeation rate of auxin , with units of µm / s , Ai is the concentration of auxin in cell i , Aj is the concentration of auxin in cell j , PINij is the concentration of PIN1 in cell i , at the cell edge between cells i and j , and PINji is the concentration of PIN1 in cell j , at the cell edge between cell j and i . T is a constant describing the rate of PIN-mediated auxin transport , with units of µm2/ Au . s . Ij and Ii are parameters describing the auxin import rates of cells j and i respectively , with units of µm / s . The rate of change of auxin concentration in the cytoplasm of a given cell , i , depends on the production and degradation rates of auxin , and its flux into adjacent cells and is given by: ( 12 ) dAidt=ρ−μAi−1Ri∑ij∈N ( i ) ( Φi→j ) lij Where ρ is the production rate of auxin , with units of Au / µm2 . s , µ is the degradation rate of auxin , with units of /s , Ai is the concentration of auxin in the cytoplasm of cell i and Ri is the area of the cell i . Φi→j is the flux of auxin across the interface between cells i and j , in the neighbourhood of the cell i ( ( N ( i ) ) and lij is the length of the cell edge at the interface between cells i and j . In the simulations used to generate Figure 7B , E and Figure 7—figure supplement 1B , there is no flux of auxin across the boundaries of the tissue . To minimise boundary effects , in the simulations used to generate Figures 9A , B , 10A , B and 16A , B , the left and right boundaries of the tissue are connected so auxin flux can occur between them . All changes in concentrations were solved numerically using an explicit Euler integration method . In all simulations where the row of cells at the base of the tissue has an elevated rate of auxin production ( orange cells at the base in Figures 9A , B , 10A , B and 16 A , B ) , the auxin production rate in these cells = 0 . 0008 Au /µm2 . s . In Figure 16A , B , in the band of three cells with elevated auxin production , the auxin production rate = 0 . 032 Au /µm2 . s . In the simulation used to generate Figure 9A , in the distal row of cells , the auxin degradation rate = 0 . 5/s . In the simulations used to generate Figures 9B , 10A , B and 16 A , B , in the distal most row of cells with elevated auxin import and removal , I = 300 µm / s and μ = 0 . 04/s . In the simulation used to generate Figure 10B , in the induced cells with elevated auxin import and removal , I = 30 µm / s and μ = 0 . 5 /s . In the simulation used to generate Figure 16B , I = 100 µm / s and μ = 0 . 8 /s in the single cell with elevated auxin import and removal . In the simulation used to generate Figure 10A and B ( positioning of cells with elevated auxin import and removal ) , in the first 165 s of the simulation , the production rate of auxin , ρAux , in the proximal-most file of cells , = 0 . 0008 Au/ µm2 . s . In this first phase of the simulation , the production rate of auxin in all other cells in the proximal half of the array is 0 . 0005Au/ µm2 . s . Then , between 165 and 200 s of the simulation , the auxin production rate of all cells in the proximal half of the array ( including those in the proximal-most file ) is increased at every time step of the simulation according to Equation 6 . Here , β = 0 . 0001 Au/ µm2 . s2 . As a consequence of this increase in the auxin production rate , after 200 s of the simulation , in the proximal-most file of cells , ρ = 0 . 0043 Au/ µm2 . s , and in the other cells in the proximal-most half of the leaf , ρ = 0 . 004 Au/ µm2 . s . After 165 s of the simulation , if the auxin concentration of a cell exceeds a threshold , Timport , the auxin import of the cell is increased so that I = 30 µm / s and the auxin degradation rate is increased so that μ = 0 . 5 /s . Timport = 0 . 09 Au/ µm2 . After 20 s of the simulation , noise is added to the concentrations of auxin every 0 . 01 s of the simulation , according to Equation 7 . In this case , θAux is a random number drawn from a normal distribution , with mean 0 , and standard deviation 5 × 10−3 . The up-the-gradient model is implemented based on the description by Bilsborough et al . , 2011 . The tissue is represented as described for the flux-based model . The system is initialised with auxin in cytoplasmic compartments and PIN1 at edge compartments . In some simulations , the system is initialised with noisy PIN1 concentrations at each cell edge , as described by Equations 9a , b , or noisy auxin concentrations in each cell , as described by Equations 8a , b . PIN1 only exists in cell edge compartments and each cell has a fixed total concentration of PIN1 available to its cell edges . This total concentration of PIN1 is distributed between edge compartments at each time step according to an exponential function of the auxin concentrations in neighbouring cells . The equation describing the concentration of PIN1 in a given edge compartment in cell i , at the interface between cell i and j , at a given time step is: ( 13 ) PINij=PINibAj∑kbAk Where PINij is the concentration of PIN1 at the cell edge between cells i and j , in cell i . PINi is the total PIN1 concentration available to be divided between all the edges of cell i ( with units of Au/ µm ) , Aj is the concentration of auxin in the neighbour j and Ak is the auxin concentration in the neighbour k of cell i . The exponentiation base , b , controls the extent to which the auxin distribution in neighbouring cells influences PIN1 protein distribution at cell edges . Previous simulations have differed in the implementation of auxin production . Some assume that intracellular auxin feeds back to inhibit its own production ( Smith et al . , 2006; Bayer et al . , 2009; Bilsborough et al . , 2011; Feugier et al . , 2005 ) , whilst other assume that auxin production occurs independently of intracellular auxin concentration ( Rolland-Lagan and Prusinkiewicz , 2005; Jonsson et al . , 2006; Stoma et al . , 2008 ) . We use the latter assumption in all models . The rate of change of auxin concentration for a given cytoplasmic compartment depends on the rates of auxin production and degradation , and the rates of auxin flux between the given cell and its neighbours . This is described by the following equation: ( 14 ) dAidt=ρ−μAi−1Ri∑ij∈N ( i ) ( Φi→j ) lij10 . 7554/eLife . 18165 . 027Table 2 . Parameter values used in simulations of the flux-based model . DOI: http://dx . doi . org/10 . 7554/eLife . 18165 . 027SymboldescriptionunitvalueΔtnumerical time steps ( seconds ) 0 . 01Rarea of cytoplasmic compartmentµm2260llength of cell edge compartmentµm15 for short cells edges; 8 . 6 for the each of the two compartments of a long edge *CAdefault initial concentration of auxin in cytoplasmic compartmentsAu/ µm20 . 01CPINdefault initial concentration of PIN1 at cell edge compartmentsAu /µm0 . 01†ɛPINlimit for noise addition during initialisation of PIN1 concentrations at cell edgesdimensionless0 . 025**ɛAuxlimit for noise addition during initialisation of Auxin concentrationsdimensionless0 . 025‡TPIN-dependent active auxin efflux rateµm2/Au . s1γUnbinding rate of PIN1 from the cell edge/s0 . 1ρproduction rate of AuxinAu/µm2 . s0 . 0001§µdegradation rate of Auxin/s0 . 02αFlux-dependent promotion of PIN1 allocation to a cell edgedimensionless1#DPassive permeation rate of auxinµm/s5PmaxMaximum concentration of PIN1 at a cell edgeAu /µm0 . 01¶IAuxin import rateµm / s0* For cells with hexagonal geometries , each cell edge has a length of 10 µm . † Value given is that used in simulations used to generate Figure 7 . B and Figure 7—figure supplement 1B . In all other simulations , CPIN = 0 . ‡ Values apply to simulations used to generate Figure 7B ( ɛPIN ) and Figure 7—figure supplement 1B , and Figure 7E ( ɛAux ) . All other simulations are initialised without noise addition . § The value given for the auxin production rate applies to simulations used to generate Figures 9A , B , 10A , B and 16A , B . In the simulation used to generate Figure 7B and Figure 7—figure supplement 1B , ρ = 0 . 0025 Au/µm2 . s and in the simulation used to generate Figure 7E , ρ = 0 . 002 Au/µm2 . s . # The value given for , α , the flux-dependent promotion of PIN allocation to cell edges , applies to simulations used to generate Figures 9A , B , 10A , B and 16A , B . In the simulation used to generate Figure 7B , α = 4 × 10−3 and in the simulation used to generate Figure 7E , α = 3 . 2 × 10−3 . ¶ In the simulation used to generate Figures 7B , Figure 7—figure supplement 1B and 7E , Pmax = 0 . 04Au /µm . Where Ai is the auxin concentration of cell i , ρ is the auxin production rate , with units of Au /µm2 . s , µ is the auxin degradation rate , with units of / s and Ri is the area of the cell i , with units of µm2 . Φi→j is the auxin flux ( with units of Au / µm . s ) out of cell i , across the interface between cells i and j , in the neighbourhood of cell i ( ( N ( i ) ) . lij is the length of the cell edge at the interface between cells i and j , with units of µm . The flux of auxin between neighbouring cells is calculated as described by Equation 11 . Changes in auxin concentrations were solved numerically using an explicit Euler integration method . As for the flux-based model , in the simulations used to generate Figure 7A and D and Figure 7—figure supplement 1A , there is no flux of auxin across the boundaries of the tissue . To minimise boundary effects , in the simulations used to generate Figures 8A , B , 11H , 14D and 16 E , F , the left and right boundaries of the tissue are connected so auxin flux can occur between them . In simulations used to generate Figure 8A , B and Figure 16E , F , where the distal-most row of cells has an elevated initial auxin concentration ( row of orange cells in Figures 8A , B ) , the initial auxin concentration in these cells is 0 . 5 Au /µm2 . In all simulations where the proximal-most row of cells has a low initial auxin concentration and an elevated rate of auxin degradation ( blue cells in bottom row in Figures 8A , B , 11H , 16E , F ) , the initial auxin concentration in these cells is 0 Au /µm2 and their auxin degradation rate = 0 . 05/s . In the simulation used to generate Figure 14D , the initial auxin concentration in the distal-most row of cells = 0 . 5 Au /µm2 , whilst in all other cells , the initial auxin concentration is as stated in Table 3 . In the distal most row of cells , the rates of auxin import and degradation are elevated ( I = 20 µm / s and µ = 0 . 01 /s ) . In the proximal most file of cells , the auxin production rate is elevated so that ρ = 0 . 001 Au /µm2 . s10 . 7554/eLife . 18165 . 028Table 3 . Parameter values used in up-the-gradient simulations . DOI: http://dx . doi . org/10 . 7554/eLife . 18165 . 028SymboldescriptionunitvalueΔtnumerical time steps ( seconds ) 0 . 01Rarea of cytoplasmic compartmentµm2260llength of cell edge compartmentµm15 for short cells edges; 8 . 6 for the each of the two compartments of a long edge *CAdefault initial concentration of auxin in cytoplasmic compartmentsAu/ µm20 . 01 ( Figure 7A , D , Figure 7—figure supplment 1A . ) 0 . 005 ( other Figs . ) CPINdefault initial concentration of PIN1 at cell edge compartmentsAu /µm0 . 1†ɛPINlimit for noise addition during initialisation of PIN1 concentrations at cell edgesdimensionless0 . 025†ɛAuxlimit for noise addition during initialisation of Auxin concentrationsdimensionless0 . 025†TPIN-dependent active auxin efflux rateµm2/Au . s80ρproduction rate of AuxinAu /µm2 . s0 . 0003‡µdegradation rate of Auxin/s0 . 005PINiTotal amount of PIN1 in a cell available for binding to edge compartmentsAu /µm0 . 1DPassive permeation rate of auxinµm/s10bExponentiation base for PIN1 allocation to the membranedimensionless6IAuxin import rateµm / s0* For cells with hexagonal geometries , each cell edge has a length of 10 µm . † Values given for the initial concentration of PIN1 at cell edge compartments , and the limit for noise addition to this concentration , are those used in simulations used to generate Figure 7A and Figure 7—figure supplement 1A . In all other simulations , CPIN and ɛPIN = 0 . The value given for ɛAux applies only to Figure 7D , in all other simulations , ɛAux = 0 . ‡ Value given for the production rate of auxin applies to the simulations used to generate Figures 7D , 8A , B , 11H , 14D and 16E , F . In the simulation used to generate Figure 7A and Figure 7—figure supplement 1A , ρ = 0 . 0025 Au /µm2 . s . In the simulation used to generate Figure 11H , in the distal most row of cells , auxin import and degradation are elevated , as described for Figure 14D . When the auxin concentration exceeds a threshold , Timport , in cells other than those at the proximal and distal boundaries , the rates of auxin import and degradation are increased to be equal to those in the distal most row of cells ( I = 20 µm / s and µ = 0 . 01 /s ) . Timport = 0 . 08 Au /µm2 . In Figure 16E , F , in the band of three cells with elevated auxin production the auxin production rate = 0 . 04 Au /µm2 . s . In Figure 16F , in the cell with elevated auxin import and removal , I = 5 µm / s and µ = 0 . 03 /s . | Plants , unlike animals , are able to grow and develop throughout their lives . New leaves and flowers are made from outgrowths that constantly form at the tip of growing shoots . Groups of cells in the outer layer of the shoot tip arrange a protein called PIN1 so that it is more abundant on the cell surfaces that face towards the centre of the group . PIN1 transports a hormone called auxin out of plant cells and this “convergent” arrangement of PIN1 increases the levels of auxin in cells at the centre of the group , leading to the formation of a new outgrowth . However , it is not clear what causes these cells to position their PIN1 proteins in this way . Several hypotheses have been proposed to explain how convergent patterns of PIN1 form . For example , according to the “up-the-gradient” hypothesis , PIN1 is allocated to the end of a cell that is next to a cell with a higher level of auxin . Abley et al . have now compared predictions from computer models with new experimental data from a plant called Arabidopsis to evaluate three hypotheses for how convergent PIN1 patterns form . A computer model based on the up-the-gradient hypothesis naturally creates convergent PIN1 patterns , even if each cell starts off with the same level of auxin . On the other hand , models based on two other hypotheses generate tandem alignments of PIN1 so that auxin is transported in the same direction along lines of cells . Next , Abley et al . tested these models using mutant Arabidopsis plants that develop outgrowths from the lower surface of their leaves . These outgrowths form in a similar way to outgrowths at the growing shoot tip , but in a simpler context . The experiments show that the patterns of where auxin is produced in growing leaves were more compatible with the tandem alignment models than the up-the-gradient model . This suggests that plants use a tandem alignment mechanism to form convergences of PIN1 proteins that generate the local increases in auxin needed to make new outgrowths . This study only examined a single layer of cells on the plant surface . Other cell layers also show highly organised patterns of PIN1 proteins , so a future challenge is to extend the approach to study the entire 3D structure of new shoot outgrowths . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"plant",
"biology",
"developmental",
"biology"
] | 2016 | Formation of polarity convergences underlying shoot outgrowths |
Calcium ions ( Ca2+ ) are essential for many cellular signaling mechanisms and enter the cytosol mostly through voltage-gated calcium channels . Here , using high-speed Ca2+ imaging up to 20 kHz in the rat layer five pyramidal neuron axon we found that activity-dependent intracellular calcium concentration ( [Ca2+]i ) in the axonal initial segment was only partially dependent on voltage-gated calcium channels . Instead , [Ca2+]i changes were sensitive to the specific voltage-gated sodium ( NaV ) channel blocker tetrodotoxin . Consistent with the conjecture that Ca2+ enters through the NaV channel pore , the optically resolved ICa in the axon initial segment overlapped with the activation kinetics of NaV channels and heterologous expression of NaV1 . 2 in HEK-293 cells revealed a tetrodotoxin-sensitive [Ca2+]i rise . Finally , computational simulations predicted that axonal [Ca2+]i transients reflect a 0 . 4% Ca2+ conductivity of NaV channels . The findings indicate that Ca2+ permeation through NaV channels provides a submillisecond rapid entry route in NaV-enriched domains of mammalian axons .
Ca2+ ions crossing the neuronal plasma membrane are critically involved in depolarization and distribute in the cytosol in spatial microdomains and organelles to regulate a wide range of processes ranging from gene expression to fast transmitter release ( Berridge , 2006; Neher and Sakaba , 2008 ) . In axons , voltage-gated Ca2+ ( CaV ) channels at presynaptic terminals open in response to a single action potential ( AP ) , raising intracellular Ca2+ concentrations ( [Ca2+]i ) in nanodomains from ~50 nM up to ~10 µM to increase transmitter vesicle release rates by the power of ~4 ( Helmchen et al . , 1997; Schneggenburger and Neher , 2000 ) . In response to APs , large and local [Ca2+]i transients are typically also observed in the axon initial segment ( AIS ) and nodes of Ranvier ( Callewaert et al . , 1996; Bender and Trussell , 2009; Yu et al . , 2010; Gründemann and Clark , 2015; Zhang and David , 2016; Clarkson et al . , 2017 ) . At these sites , Ca2+ currents have been implicated in AP initiation and propagation by a local depolarizing action of the inward current or by activating the large conductance , Ca2+- and voltage-dependent K+ ( BKCa ) channels modulating burst firing probability and limiting frequency-dependent AP failure rates ( Bender and Trussell , 2009; Yu et al . , 2010; Hirono et al . , 2015 ) . The CaV channel subtypes identified in axons are both cell type- and species-dependent and include the T- , P/Q- or N-type CaV channels ( Callewaert et al . , 1996; Bender and Trussell , 2009; Yu et al . , 2010; Gründemann and Clark , 2015; Zhang and David , 2016 ) . At the AIS in particular the T-type Ca2+ channel mediates AP-dependent Ca2+ influx ( Bender and Trussell , 2009; Martinello et al . , 2015; Fukaya et al . , 2018; Jin et al . , 2019 ) . However , in the prefrontal cortical pyramidal neuron AIS about 70% of the AP-evoked [Ca2+]i remains following pharmacological block of T-type CaV channels ( Clarkson et al . , 2017 ) . Furthermore , evidence for a clustering of T-type CaV channels at the AIS is ambiguous and immunofluorescence or immuno-gold labeling studies show a density which is comparable to somatodendritic or spine compartments ( McKay et al . , 2006; Martinello et al . , 2015 ) . Several other mechanisms may contribute to axoplasmic [Ca2+]i elevations in the AIS . Firstly , Ca2+ levels could rise due to Ca2+-induced Ca2+ release from intracellular sources such as the endoplasmic reticulum ( ER ) . Most AISs contain ER cisternae organelles consisting of stacks of membranes expressing the store-operated ryanodine receptor ( RyR ) , inositol 1 , 4 , 5-triphosphate receptor 1 ( IP3R1 ) and sarcoplasmic ER Ca2+ ATPase ( SERCA ) pumps ( Benedeczky et al . , 1994; King et al . , 2014; Antón-Fernández et al . , 2015 ) . The coupling of transmembrane Ca2+ entry with intracellular store release may generate a local activity-dependent rise of [Ca2+]i . However , a contribution of ER stores to AIS [Ca2+]i remains to be directly demonstrated . Secondly , near the peak of the AP the electrogenic Na+-Ca2+ exchanger ( NCX ) reverses direction and imports Ca2+ . A reverse mode of operation has not only been implicated in pathological [Ca2+]i elevations in axons during hypoxia and injury ( Stys et al . , 1991; Iwata et al . , 2004 ) , but also occurs during trains of APs in nodes and neighboring internodes ( Zhang and David , 2016 ) . Finally , one alternative pathway that has yet to be directly examined in mammalian cortical axons involves the voltage-gated Na+ ( NaV ) channels . Studies in the squid giant axon combining electrophysiological recordings with Ca2+ imaging showed that an early component of depolarization-induced Ca2+ entry is tetrodotoxin ( TTX ) -sensitive ( Baker et al . , 1971; Meves and Vogel , 1973; Brown et al . , 1975 ) . Voltage-clamp recordings from axons and perfusing distinct ionic solutions provided a quantitative estimate that NaV channels may pass divalent Ca2+ ions with permeability ratios ( PCa/PNa ) up to 0 . 10 ( Hille , 1972; Meves and Vogel , 1973 ) . Ca2+ permeability of NaV channels has also been shown in cardiac cells and hippocampal neurons ( Akaike and Takahashi , 1992; Aggarwal et al . , 1997; Santana et al . , 1998 ) but whether this extends to the cortical axons remains to be examined . Here , using wide-field Ca2+ imaging with a high-speed CCD camera enabling detection of [Ca2+]i changes at high sensitivity and high temporal resolution ( Jaafari et al . , 2014; Ait Ouares et al . , 2016 ) , we explored the various pathways of Ca2+ entry in axons of rat thick-tufted neocortical layer 5 ( L5 ) pyramidal neurons . We found that during subthreshold depolarizations [Ca2+]i transients were highly compartmentalized to the AIS and nodes of Ranvier . While these transients were amplified by ER store release , the trigger was only modestly accounted for by CaV channels . The largest fraction of activity-dependent [Ca2+]i was TTX-sensitive and overlapped with the rapid gating of NaV channels . Experiments in HEK-293 cells transfected with the human NaV1 . 2 channel confirmed that TTX-sensitive Na+ currents were sufficient to generate [Ca2+]i elevations . Together , the data suggest that [Ca2+]i dynamics in the mammalian AIS are predominantly mediated by a rapid Ca2+ entry through NaV channels .
Thick-tufted L5 pyramidal neurons , also called L5B or pyramidal tract neurons , are the largest pyramidal neurons in the cortex and integrate synaptic inputs from all cortical layers , playing a central role in cognitive tasks including perception ( Groh et al . , 2010; Ramaswamy and Markram , 2015; Takahashi et al . , 2016 ) . Their large axons ( ~1 . 5 µm in diameter ) send long-range output projections to the thalamus , striatum and spinal cord , but within the cortex branch sparsely and have a trajectory perpendicular to the pia providing an excellent anatomical arrangement to image and record from . To optically record the spatial profile of axonal [Ca2+]i we made somatic whole-cell patch-clamp recordings from neurons filled with the high-affinity Ca2+ indicator Oregon Green BAPTA 1 ( OGB-1 , 100 µM ) and imaged epifluorescence signals along the proximal region of the main axon ( Figure 1 ) . We first used subthreshold depolarizations evoked by artificial excitatory postsynaptic potentials ( a-EPSPs , 100 Hz , peak depolarization 17 . 0 ± 0 . 6 mV , n = 15; Figure 1a ) . Examination of the spatial profile revealed that Ca2+ signals were observed in the AIS and hot spots separated with regular distances along the axon ( locations 2 , 4 and 6; Figure 1a ) . In order to examine whether the [Ca2+]i hot spots corresponded to nodes of Ranvier , we post-hoc stained for βIV-spectrin and biocytin , and found indeed overlap between subthreshold [Ca2+]i rise and spectrin-enriched sites ( Figure 1b ) . In the same cells we examined the spatial profile of Δ[Ca2+]i in response to a single AP evoked with a brief square current injection ( Figure 1c ) . As expected from back- and forward-propagating APs with much higher depolarizations ( ~100 mV ) , large [Ca2+]i transients were observed widespread throughout all axonal and somatodendritic domains . Population analysis showed that AP-induced [Ca2+]i transients were similar between AIS and nodes ( one-way ANOVA followed by Tukey’s multiple comparison test , p<0 . 0001 , differences between all groups were significant ( p<0 . 05 ) except between AIS and node ( p=0 . 13 ) and between internode ( IN ) and dendrites ( p=0 . 85 ) ; Figure 1c , d ) . Interestingly , also during a-EPSPs the [Ca2+]i transients in the AIS and the first nodes were highly comparable , while [Ca2+]i signals in the internodal and dendritic domains were an order of magnitude smaller ( one-way ANOVA with Tukey’s multiple comparison test , p<0 . 0001 , differences between all groups were significant ( p<0 . 0001 ) , except between AIS and node ( p=0 . 38 ) and IN and Dend ( p=0 . 97 ) ; Figure 1a , d ) . Similar experiments in L5 neocortical pyramidal neurons in slices from human temporal cortex also revealed a-EPSP evoked Δ[Ca2+]i in the AIS , but not in the dendrite , suggesting that subthreshold sensitive [Ca2+]i transients are conserved across mammalian species ( Figure 1—figure supplement 1 ) . Together , these results show that activity-dependent [Ca2+]i transients are spatiotemporally compartmentalized and Ca2+ entry dynamics are similar in the axoplasm of the AIS and nodes . The thick-tufted L5 pyramidal neuron AIS contains a unique variant of cisternal organelle characterized by a continuous tubular organization of smooth ER , called the giant saccular organelle ( Antón-Fernández et al . , 2015 ) . Cisternal organelles with smooth ER express synaptopodin ( synpo ) , RyR , the IP3 receptor 1 , and SERCA that are thought to contribute to Ca2+ release , buffering and storage ( Bas Orth et al . , 2007; King et al . , 2014 ) . We hypothesized that these organelles could generate Ca2+-induced Ca2+-release , thereby contributing to domain-selective activity-dependent [Ca2+]i transients ( Figure 1 ) . Triple immunostaining for synpo , Ankyrin G and biocytin confirmed that the cisternal organelle was present along the entire axis of the AIS and spatially overlapped with the subthreshold-evoked [Ca2+]i transients ( n = 19; Figure 2a , b ) . However , while subthreshold depolarization-induced Ca2+ transients were present in the nodes , synaptopodin expression was not detected ( n = 10 nodes; Figures 1 and 2a ) . To experimentally test whether AIS Ca2+-store release contributes to activity-dependent [Ca2+]i transients we performed experiments with standard intracellular solution and subsequently re-patched the same cell with a solution containing ryanodine ( 200 µM , blocking RyR-mediated Ca2+ release ) and heparin ( 5 mg/ml , competitively inhibiting IP3-evoked Ca2+ release; Figure 2c ) . Blocking Ca2+ release significantly lowered ΔF/F Ca2+ peak transients in the AIS , both for the subthreshold- and AP-evoked [Ca2+]i changes ( a-EPSP , 53 . 2% reduction , p=0 . 006; AP , 34 . 3% reduction , p=0 . 02 , one-tailed ratio paired t-tests , n = 5; Figure 2d ) . Consistent with the AIS-specific location of the giant saccular organelle , store blockers had no effect on AP-evoked Δ[Ca2+]i in the basal dendrite ( p=0 . 48 , n = 3; Figure 2d ) . Furthermore , since the stores contribute significantly to AIS Ca2+ levels , blocking store release could act as a low-pass filter for Ca2+ level kinetics , reducing rise and decay times . However , blocking Ca2+-store release did not alter the rise- or decay time kinetics in the AIS ( τact , p=0 . 52; τde-act , p=0 . 18 , two-tailed paired t-tests , n = 5; Figure 2e ) . These data suggest that the giant saccular organelle amplifies activity-dependent [Ca2+]i changes selectively in the AIS . Ca2+ release from internal stores is likely triggered by Ca2+ entry via neuronal voltage-dependent plasmalemmal routes . To test whether AIS [Ca2+]i changes require Ca2+ from the extracellular space , we bath applied 2 . 5 mM of the Ca2+ chelator EGTA which effectively lowered the extracellular Ca2+ concentration ( [Ca2+]o ) from 2 mM to ~437 nM , thereby reducing the driving force for Ca2+ ( see Materials and methods ) . Ca2+ imaging at the AIS ( OGB-1 , 100 µM ) showed that EGTA almost fully abolished the subthreshold-evoked Δ[Ca2+]i ( 90 . 7% reduction , one-tailed ratio paired t-test , p=0 . 0031 , n = 4; Figure 3a , b ) . Similarly , the AP-generated Δ[Ca2+]i was almost extinguished after bath application of EGTA ( 92 . 8% reduction , p=0 . 0011 , n = 4; Figure 3a , b ) . Next , we hypothesized that the transmembrane pathway for Ca2+ entry in the AIS during subthreshold stimuli could be explained by the low-voltage gated CaV channels ( T- and R-type ) . However , bath application of the highly selective T-type ( CaV3 . 1–3 ) blocker TTA-P2 ( 1 µM , Choe et al . , 2011 ) or nickel ( Ni2+ , 100 µM ) did not significantly reduce Ca2+ signals ( one-tailed ratio paired t-tests; TTA-P2 , p=0 . 17 , n = 4; Ni2+ , p=0 . 063 , n = 5; Figure 3b , c ) . We next blocked R-type CaV channels , by puffing SNX-482 ( 500 nM ) locally to the AIS , but this did not lead to a reduction in subthreshold [Ca2+]i rise either ( SNX-482 , p=0 . 11 , n = 3 ) . Furthermore , consistent with their more depolarized voltage range of activation , the L-type channels did not affect subthreshold Δ[Ca2+]i ( 20 µM isradipine , p=0 . 14; 10 µM nimodipine , p=0 . 41 , both n = 4; Figure 3b ) and the block of N-type and P/Q-type channels , by local application of ω-conotoxin MVIIC ( 2 µM ) to the AIS , also failed to reduce Ca2+ signals ( p=0 . 42 , n = 5; Figure 3b ) . Furthermore , a combined block of T- and L-type channels did not affect the peak ΔF/F in the AIS ( TTA-P2 and isradipine , p=0 . 12 , n = 3; Figure 3b ) . Although application of the T-type blockers TTA-P2 and Ni2+ was ineffective to block subthreshold [Ca2+]i rise , in the same neurons it did reduce the peak ΔF/F evoked by a single AP by almost 20% ( TTA-P2 , 18 . 7% , p=0 . 021 , n = 4; Ni2+ , 19 . 7% block , p=0 . 013 , n = 5; Figure 3b , c ) . In addition , isradipine reduced the AP-evoked Δ[Ca2+]i in the AIS by 14 . 9% ( isradipine , p=0 . 0070 , n = 4 ) and the alternative L-type blocker nimodipine showed a non-significant blocking trend ( nimodipine , p=0 . 060 , n = 4; Figure 3b ) . A combined application of T- and L-type channel blockers ( 1 µM TTA-P2 and 20 µM isradipine ) caused a 27 . 2% reduction of peak ΔF/F , showing a sublinear summation of two blocking agents ( TTA-P2 and isradipine , p=0 . 0071 , n = 5; Figure 3b ) . In contrast , the R-type CaV channel blocker SNX-482 ( 500 nM ) did not reduce the AP-evoked Δ[Ca2+]i ( SNX-482 , p=0 . 29 , n = 3; Figure 3b ) . Local application of the ω-conotoxin MVIIC ( 2 µM ) showed a non-significant trend to block the peak ΔF/F ( 6 . 8% , p=0 . 064 , n = 5; Figure 3b ) . As a positive control experiment , we imaged a collateral bouton of the same neuron and used local application of ω-conotoxin MVIIC which almost completely abolished the peak ΔF/F by 91 . 5% , consistent with the presence of N- and P/Q-type CaV channel subtypes in presynaptic terminals ( p=0 . 021 , n = 3; Figure 3b ) . Finally , given the unexpected remaining [Ca2+]i transients in the AIS in the presence of CaV channel blockers , we hypothesized that NCX may contribute to activity-dependent [Ca2+]i increase in the AIS . At the resting membrane potential NCX exports Ca2+ from the cytoplasm to maintain [Ca2+]i near ~50 nM , however Na+ entry may promote instantaneous Ca2+ influx by a reverse mode of operation ( Yu and Choi , 1997; Stys and LoPachin , 1997; Figure 3b ) . To examine its contribution , we pharmacologically blocked NCX by combined bath application of KB-R7943 ( 20 µM ) and SN-6 ( 10 µM ) . The results showed , however , no change in the subthreshold nor AP-evoked Δ[Ca2+]i ( two-tailed ratio paired t-tests , a-EPSP , p=0 . 16; AP , p=0 . 13 , n = 5 , respectively; Figure 3b ) . In summary , these data show that while transmembrane Ca2+ influx from the extracellular space is required for activity-evoked Δ[Ca2+]i , none of the CaV channels played a role in the subthreshold depolarization , whereas T- and L-type CaV channels partially contributed to the AP-evoked influx . What could be the source of the remaining component of Ca2+ influx at the AIS ? Both in hippocampal neurons and heart muscle cells , Ca2+ currents have been described which are not blocked by Ni2+ nor by other known CaV channel antagonists , but instead are sensitive to the highly selective NaV channel blocker ( TTX ) , and therefore called ICa ( TTX ) ( Akaike and Takahashi , 1992; Aggarwal et al . , 1997 ) . In Na+-free extracellular solution ICa ( TTX ) resembles the Na+ current and activates at potentials as negative as –70 mV while peaking at –30 mV ( Akaike and Takahashi , 1992 ) . To examine the presence of ICa ( TTX ) in L5 axons we took advantage of the low-affinity indicator Oregon Green BAPTA 5N ( OGB-5N , 1 mM; Figure 4a ) , which gives smaller fluorescent signals but is linear over a wider range of [Ca2+]i compared to OGB-1 ( Kd 20 µM vs . 170 nM , respectively ) . We used the voltage-clamp configuration and injected depolarizing ramps of 50 ms with increasing slopes ( from 0 . 0 to 0 . 55 mV ms–1 ) with a maximum peak at ~95% of the AP threshold ( Figure 4b ) , thereby studying the same depolarization range and duration as the a-EPSPs used in Figures 1–3 . The results showed that Ca2+ influx was strongly compartmentalized to the AIS and nodal axolemma ( Figure 4a ) consistent with the a-EPSP evoked OGB-1 transients ( Figure 1 ) . Remarkably , bath application of TTX almost completely abolished [Ca2+]i elevations , even at depolarizations above the AP threshold ( at –54 . 5 mV , 92 . 8% block , Cohen’s d: 1 . 88 , two-way ANOVA , p<0 . 0001 , n = 4; Figure 4b , c ) . As an alternative to TTX we used the quaternary amine NaV channel inhibitor QX-314 , which plugs the open state of the NaV channel from the internal side . Similar to TTX , with 6 mM QX-314 added to the pipette solution voltage ramps did not evoke Ca2+ transients ( at –54 . 5 mV , 94 . 8% block , Cohen’s d: 1 . 92 , control vs . QX-314 , p<0 . 0001 , n = 4 , TTX vs . QX-314 , p=0 . 97 , n = 4; Figure 4c ) . Although QX-314 at this concentration has been reported to also block voltage-gated Ca2+ currents ( Talbot and Sayer , 1996 ) , subthreshold-evoked [Ca2+]i was not mediated by CaV channels ( Figure 3b ) . The near complete block by two distinct NaV channel blockers therefore indicates an important role of NaV channels in mediating subthreshold axonal Ca2+ influx . We next investigated whether NaV channels also contribute to AP-evoked Δ[Ca2+]i ( Figure 5a ) . To dissociate a putative role of NaV channels to pass Ca2+ ions from generating the AP depolarization of ~100 mV we first imaged Ca2+ at the AIS in current-clamp , subsequently applied 1 µM TTX and imaged Ca2+ transients evoked by the recorded AP waveform injected as a voltage command ( ‘AP-clamp’ ) . The results showed a near complete abolishment of Δ[Ca2+]i in the presence of TTX ( one-way ANOVA with Tukey’s multiple comparisons test , CC vs . VC , 89 . 5% reduction , p<0 . 0001 , n = 7; Figure 5a , b ) . However , this [Ca2+]i peak amplitude reduction could also be due to an incomplete voltage- and space-clamp of the AIS for fast voltage transients . The small diameter of the axon has a high axial resistance , acting as a low-pass filter for the antidromic AIS action potentials ( Hamada et al . , 2016 ) which also may attenuate the orthodromic voltage spread into the axon . To examine the possibility that axonal APs are attenuated in the somatic AP clamp configuration , we optically recorded JPW3028 , a fast fluorescent voltage indicator that remains stable over long recording periods and is highly linear over a large voltage range ( ~250 mV , Figure 5—figure supplement 1 ) . Consistent with the voltage loss , when we injected the AP-clamp in the soma in the presence of TTX and optically recorded Vm in the AIS , we observed a significant ~2 fold reduction in the AP peak amplitude ( one-way ANOVA with Tukey’s multiple comparisons test , VC vs . CC , p=0 . 014; Figure 5a , c ) . To restore peak depolarization in the presence of TTX and reliably compare the Ca2+ transients evoked by equal depolarization , we doubled the amplitude of the somatic AP-clamp ( VC ×2 ) . With this protocol both the peak depolarization and AP half-width in the AIS were indistinguishable from the control APs ( peak JPW , VC ×2 vs . CC , one-way ANOVA with Tukey’s multiple comparisons test , p=0 . 75 , n = 4 , Figure 5c , half-width in JPW , VC ×2 vs . CC , one-way ANOVA with Tukey’s multiple comparisons test , p=0 . 36 , n = 4 , not shown ) . TTX blocked 65 . 5% of the AP-evoked Δ[Ca2+]i ( peak OGB-5N , Cohen’s d: 5 . 49 , VC ×2 vs . CC , p<0 . 0001 , n = 8; Figure 5b ) . Taken together , these data suggest that a large fraction of both subthreshold-depolarization and AP-evoked Ca2+ ions enter the axoplasm through TTX-sensitive channels at the AIS . Whether ICa ( TTX ) is carried by a specific TTX-sensitive CaV channel or reflects Ca2+ permeating directly through the NaV channel remains debated ( Santana et al . , 1998; Cruz et al . , 1999; Heubach et al . , 2000; Chen-Izu et al . , 2001; Guatimosim et al . , 2001 ) . We hypothesized that if Ca2+ ions enter the AIS cytoplasm by flowing through NaV channels , ICa ( TTX ) should reflect the time course of INa . To measure submillisecond rapid events with fluorescence in the small axon ( diameter ~1 . 5 µm ) , we optimized multiple imaging parameters enabling the acquisition of fluorescence at 20 kHz ( see Materials and methods ) . Using a Na+-sensitive indicator ( sodium-binding benzofuran isophthalate , SBFI , 1 mM ) in combination with OGB-5N ( 1 mM ) showed that the two indicators were indistinguishable in their rising phase during an AP , suggesting that Ca2+ entry at the AIS may be as rapid as Na+ entry ( Figure 6—figure supplement 1 ) . In order to quantify the kinetics of INa and ICa more directly , we next used a voltage-clamp approach . Near ~20°C and –40 mV , NaV channels open at least one order of magnitude faster compared to the T-type CaV channels ( ~200 µs [Schmidt-Hieber and Bischofberger , 2010] vs . ~5 ms [Perez-Reyes et al . , 1998] , respectively ) which may be sufficiently different to compare against the kinetics of optically recorded [Ca2+]i at the AIS . To determine the specific activation kinetics of INa and T-type ICa ( ICaT ) in L5 pyramidal neurons we measured total inward current ( INa + ICa , ) by depolarizing the soma with a step to –35 mV and pharmacologically isolated Na+ and Ca2+ current components by bath application of 1 µM TTX or 100 µM Ni2+ , respectively ( Figure 6a ) . The activation time constant of the total inward current was identical to INa ( single exponential fit τtotal = 438 . 2 µs vs . τNa = 440 . 3 µs , one-way ANOVA with Tukey’s multiple comparison test , p>0 . 999 , n = 6 ) , whereas the total current was substantially faster in comparison to ICaT ( τtotal = 438 . 2 µs vs . τCaT = 4 . 8 ms , p<0 . 0001 , n = 5; Figure 6a ) . The initial fraction of the inward current was thus primarily generated by INa . The large difference in gating could provide a temporal window to distinguish Ca2+ entry via NaV channels or T-type CaV channels . Theoretical and experimental work show that low-affinity Ca2+ indicators , like OGB-5N , are capable of tracking rapidly activating Ca2+ currents when imaged at high speed: the first time derivative of ΔF/F ( dΔF/F dt–1 ) overlaps with electrically recorded ICa , providing a mean to optically resolve the time course of ICa ( Sabatini and Regehr , 1998; Jaafari et al . , 2014; Ait Ouares et al . , 2016 ) . Imaging OGB-5N ( 1 mM ) at 20 kHz in the AIS we observed that ΔF/F comprised of two separate time courses , a fast initial rise followed by a slower rising phase ( Figure 6b ) . Both components were almost completely abolished by TTX , leaving only a small transient reflecting putatively ICaT ( n = 6 ) . We quantitatively compared the activation time constants of INa , ICa_opt ( dΔF/F dt–1 ) and ICaT by resampling the electrically recorded INa and ICaT to 20 kHz and filtering both electrical and optical traces identically ( see Materials and methods ) . Multiple hallmarks of INa matched with ICa_opt: both traces showed a rapid inward component , followed by inactivation and a persistent component ( Figure 6b , c ) . In comparison , ICaT lacked both the rapid activation and inactivation time constants ( Figure 6a , c ) . Given the lower signal-to-noise ratio in the optical traces we fitted Boltzmann sigmoid functions to the rising phase to compare the slopes of the optically and electrically recorded currents ( Figure 6c ) . The average slope of ICa_opt was significantly faster compared to the activation of ICaT ( ~500 µs vs . ~2 . 5 ms , respectively , one-way ANOVA with Tukey’s multiple comparison test , p<0 . 0001 ) and slower compared to INa ( ~500 µs vs . ~200 µs , ICa_opt vs . INap<0 . 0001; Figure 6c , d ) . The small difference between INa and ICa_opt may be explained by the equilibration time of OGB-5N ( ~200 µs ) ( Ait Ouares et al . , 2016 ) , local differences between NaV channels in the soma and AIS or the presence of Ca2+-store release in the AIS . Together , the findings indicate that the current mediating [Ca2+]i at the AIS resembles NaV channel kinetics . The results suggest that Ca2+ ions could enter the cytoplasm by permeation through the NaV channel pore . Previous studies showed Ca2+ influx through the cardiac NaV1 . 5 channel ( Cruz et al . , 1999; Guatimosim et al . , 2001 ) . To examine whether NaV channel isoforms of the axon initial segment also enable Ca2+ influx we performed experiments in HEK-293 cells which were transfected with the human gene SCN2A encoding NaV1 . 2 channel with auxiliary β1 and β2 subunits and EGFP tag ( Materials and methods; Figure 7a ) . Whole-cell recording revealed Na+ currents in EGFP+ cells but not in non-transfected cells ( average peak current density –115 . 7 ± 28 . 4 pA/pF , n = 10 vs . –3 . 7 ± 1 . 9 pA/pF at –20 mV , n = 5 , respectively; Figure 7b ) . The inward currents were completely abolished by 1 µM TTX ( 96 . 1 ± 1 . 3% , n = 7 , one-tailed Wilcoxon matched-pairs signed rank test , p=0 . 0078 , Figure 7c ) and the voltage-dependence of activation and inactivation revealed midpoints at –25 . 4 ± 2 . 1 mV and –74 . 1 ± 3 . 9 mV , respectively ( n = 10 , Figure 7d , e ) , consistent with previous work ( Ben-Shalom et al . , 2017 ) , indicating a highly selective expression of NaV1 . 2 channels . Next , we filled the transfected cells with 100 µM OGB-1 and imaged the fluorescence changes in response to a train of depolarizing pulses ( 200 Hz for 1 s , –120 to –30 mV steps; Figure 7f ) . We observed an increase in ΔF/F in every EGFP+ cell , indicating an influx of Ca2+ ( average peak 0 . 46 ± 0 . 18% ΔF/F , range: 0 . 06–1 . 4% ΔF/F , n = 7; Figure 7f–h ) . To test whether the [Ca2+]i increase required NaV channel opening we bath applied TTX ( 1 µM ) , revealing a significant decrease in the peak ΔF/F ( 92 . 1 ± 3 . 8% reduction , one-tailed Mann Whitney test , p=0 . 012 , n = 4; Figure 7h ) . The results indicate that molecular expression and opening of NaV1 . 2 channels suffices to mediate transmembrane Ca2+ influx . Our findings are in agreement with the depolarization-induced Ca2+ entry in the squid axon which is tetrodotoxin ( TTX ) -sensitive and reflected a 1% conductivity of NaV channels for Ca2+ ions ( Baker et al . , 1971 ) . To estimate the conductivity ratios ( gCa/gNa ) in L5 axons we performed computational simulations . Ca2+ entry through NaV channels was implemented by adding an ohmic Ca2+ ion mechanism into a mathematical 8-state NaV channel model that calculated the current carried by Ca2+ ( ICa ( Na ) ) and Na+ ( INa ) ( see Materials and methods ) . A single compartment containing INa and ICa ( Na ) together with high voltage-gated and T type-gated CaV channel models ( ICaH and ICaT , respectively ) showed that with an axonal AP waveform ICa ( Na ) is activated during the first microseconds of AP onset , rapidly inactivates and is temporally separated from ICaH and ICaT ( Figure 8—figure supplement 1 ) . Next , to estimate the gCa/gNa we made a multicompartmental model of a L5 pyramidal neuron ( Figure 8a , including detailed reconstructions of the AIS and nodal domains ( see Figure 2a within Hamada et al . , 2016 ) . Based on multiple experimentally recorded parameters we constrained the model AP and found that a peak conductance density of NaV channels of 16 , 000 and 850 pS µm–2 in the AIS and soma , respectively , reproduced the recorded AP and matched with AP-evoked Δ[Na+]i imaged in the AIS ( see Materials and methods and Figure 8a ) . Subsequently , [Ca2+]i was simulated based on mathematical equations representing Ca2+ diffusion and extrusion , endogenous stationary Ca2+ buffers ( taken together as κs ) and was supplemented with the buffering capacities of the specific Ca2+ indicators ( Fink et al . , 2000 ) ( see Materials and methods ) . The Ca2+ extrusion threshold and rates were adjusted to approximate the experimentally imaged peak and decay time course of measured OGB-5N ΔF/F in the AIS ( Figure 8—figure supplement 2 ) . To determine the absolute rise in [Ca2+]i produced exclusively by NaV channels , we performed additional experiments in which we imaged [Ca2+]i while blocking CaV channels that contributed to AP-evoked Δ[Ca2+]i: T- and L-type calcium channels ( TTA-P2 and isradipine , respectively , see Figure 3b ) . Using calibrated ratiometric bis-Fura-2 ( 200 μM ) imaging , we found that during 1 AP , Ca2+ entry though NaV channels induces a peak Δ[Ca2+]i of 55 . 6 nM ( n = 4; Figure 8b , Figure 8—figure supplement 2 ) . Since ~35% of AP-evoked Δ[Ca2+]i is caused by internal store amplification ( Figure 2d ) ~36 nM is mediated by transmembrane Ca2+ entry via NaV channels ( Figure 8b ) . We subsequently simulated these experiments in the multicompartmental model by removing CaV channels and including the buffering properties of 200 µM bis-Fura-2 . Varying endogenous buffering ( κs ) between 1 and 100 we updated gCa/gNa to obtain a 36 nM rise of free [Ca2+]i at the AIS . A κs of ~100 corresponds to dendritic buffering capacities ( Cornelisse et al . , 2007 ) , while axonal buffering capacities are reported to be lower ( 10–40 , Klingauf and Neher , 1997; Jackson and Redman , 2003; Delvendahl et al . , 2015 ) . When changing κs between 10–40 the gCa/gNa ratio was 0 . 38% ( κs = 10: 0 . 37% , κs = 40: 0 . 39%; Figure 8c ) . Using the 0 . 38% conductivity ratio we next evaluated how Ca2+ currents through NaV and CaV channels spatiotemporally varied across the neuronal compartments without the buffering capacities of externally applied Ca2+ dyes ( Figure 9a , b and Figure 8—figure supplement 2 ) . The simulations showed that the Δ[Ca2+]i from one AP reached a peak concentration of ~800 nM in the AIS ( Figure 9a , b ) . Due to the high density of NaV channels in the AIS they contribute to the majority of Δ[Ca2+]i and cause a rise of [Ca2+]i within submillisecond from the start of the AP ( 450 nM within <150 µs from AP threshold , red arrow in Figure 9b ) . These results are likely to provide an underestimate of the total Δ[Ca2+]i since in the model Ca2+ release from giant saccular organelle was not simulated , which would result in a total AP-evoked Δ[Ca2+]i of ~1 . 2 µM . In the basal dendritic branches the AP has a slower rise time and broader half-width , causing dendritic [Ca2+]i to accumulate slower and to higher concentrations , consistent with our experimental findings ( Figure 1 ) . Because the dendritic NaV channel density is substantially lower , their contribution to the total [Ca2+]i is negligible . The distinct contribution of NaV and CaV channels to [Ca2+]i is clearly visible when comparing the different Ca2+ currents in the AIS , showing that the majority of the total ICa during an AP is carried by ICa ( Na ) ( Figure 9c ) . Simulations predict that the ICa ( Na ) rapidly inactivates during the AP while ICa activates more slowly and has an incomplete inactivation during the AP repolarization , likely becoming the dominant contribution to [Ca2+]i during the afterdepolarization and high-frequency spike generation ( Figure 9c and Figure 8—figure supplement 1 ) . Our experiments and computational simulations show that [Ca2+]i changes mediated by CaV or NaV channels act at distinct spatiotemporal scales . To experimentally test the differential impact of Ca2+ influx on the AP waveform , we analyzed the somatically recorded APs when using distinct blockers most effective in modulating [Ca2+]i in the AIS ( Figure 3 ) . Blocking both T- and L-type CaV channels , contributing to ~27% to AP-evoked Ca2+ at the AIS ( Figure 3 ) , significantly reduced the afterdepolarization and showed a trend to reduce the afterhyperpolarization ( ADP , p=0 . 039 and AHP , p=0 . 055 , respectively , two-tailed paired t-tests , n = 5 ) , without affecting other AP properties ( p>0 . 30; Figure 9d and Table 1 ) . In contrast , when lowering [Ca2+]o with EGTA , which abolished all AIS Ca2+ influx ( Figure 3 ) , the AP half-width significantly increased and the AHP was reduced ( AP half-width , p=0 . 035 and AHP , p=0 . 043 , two-tailed paired t-tests , n = 4; Figure 9d , Table 1 ) . These results are consistent with the temporal differences in AP-evoked AIS [Ca2+]i and suggest that NaV-mediated Ca2+ entry may act to open BKCa channels ( Figure 9e ) , thereby driving K+ efflux and facilitating axonal AP repolarization .
In the present study we identified Ca2+ permeation through NaV channels as a source for activity-dependent Ca2+ entry in mammalian axons . The findings were supported by independent and converging lines of evidence , ranging from anatomical compartmentalized [Ca2+]i transients at sites with high NaV channel densities , a pharmacological block by TTX , an overlap of optically recorded ICa with NaV channel gating and molecular evidence for Ca2+ influx mediated by the NaV1 . 2 channel . In axonal domains NaV channels are thus not only involved in electrically generating the upstroke of the action potential , but also contribute to cytoplasmic Ca2+ signaling . Ca2+ entry in the L5 pyramidal neuron AIS was in part mediated by T- and L-type CaV channels ( Figure 3 and Figure 5 ) in keeping with previous studies showing that CaV channel subtypes mediate activity-dependent Ca2+ changes in both central- and peripheral nervous system axons ( Callewaert et al . , 1996; Bender and Trussell , 2009; Yu et al . , 2010; Gründemann and Clark , 2015; Zhang and David , 2016; Clarkson et al . , 2017 ) . However , when quantifying the specific fraction of block CaV channels explained only ~35% of the total Ca2+ entry during a single AP ( Figure 3 ) . These results are consistent with recent 2-photon Ca2+ imaging from the prefrontal cortical pyramidal neuron AIS , showing that ~70% of the [Ca2+]i transients evoked by a train of three APs remains in the presence of T-type CaV channel block ( Clarkson et al . , 2017 ) . Here , we found that for APs the remaining ~65% of [Ca2+]i increase is actually TTX-sensitive and this accounted even for >90% of the subthreshold-induced [Ca2+]i changes ( Figure 4 and Figure 5 ) . That Ca2+ enters through NaV channels builds on landmark studies showing that the initial component of depolarization-induced Ca2+ entry in the squid giant axon is tetrodotoxin ( TTX ) -sensitive ( Baker et al . , 1971; Meves and Vogel , 1973; Brown et al . , 1975 ) . Squid axons even generate rapid spikes in the sole presence of Ca2+ ions ( Watanabe et al . , 1967 ) . A TTX-sensitive Ca2+ current has also been identified in hippocampal neurons ( Akaike and Takahashi , 1992 ) and more extensively investigated in cardiac myocytes ( Aggarwal et al . , 1997; Santana et al . , 1998; Cruz et al . , 1999; Heubach et al . , 2000; Chen-Izu et al . , 2001 ) . The incomplete selectivity of toxins and blockers for Ca2+ channels continued , however , to cast doubt about the precise identity of the TTX-sensitive Ca2+ current ( Cruz et al . , 1999; Chen-Izu et al . , 2001; Sun et al . , 2008 ) . Indeed , an alternative explanation for some of the present results is that TTX blocks a CaV channel subtype . This has been reported at very high concentrations ( 30 µM ) for CaV3 . 3 ( Sun et al . , 2008 ) , which is higher than what we used ( 1 µM ) . Although we cannot exclude the presence of a TTX-sensitive CaV channel at the AIS that is not blocked by one of the compounds used in the pharmacological screening ( Figure 3 ) , our optically recorded ICa provides biophysical evidence that the TTX-sensitive Ca2+ current follows the same rapid activation time course as the Na+ channel pore , incompatible with T-type channel kinetics ( Figure 6 ) . Importantly , in further support that NaV channels give rise to cytoplasmic Ca2+ changes , heterologous expression of α- and β-subunits of NaV1 . 2 , known to be expressed in the rodent and human AIS ( Garrido et al . , 2003; Hu et al . , 2009; Tian et al . , 2014 ) , showed that the channel proteins expressed in isolation were sufficient to mediate Ca2+ influx ( Figure 7 ) . These results are in support of the findings by Lederer and colleagues showing that heterologous expression of hNaV1 . 5 channel produces depolarization-evoked [Ca2+]i , if expressed with its β subunits ( Cruz et al . , 1999; Guatimosim et al . , 2001 ) . From an evolutionary point of view , some Ca2+ permeation of NaV channels is not surprising . NaV channels evolved from the CaV channel superfamily and share molecular structure both in their pore sequence and intracellular regulatory domains ( Zakon , 2012; Ben-Johny et al . , 2014 ) . Furthermore , the Born radii for Na+ and Ca2+ are comparable ( 1 . 68 and 1 . 73 Å , respectively ) and Ca2+ ions are known to enter the NaV channel pore to block Na+ permeation in a concentration-dependent manner ( Lewis , 1979; Armstrong and Cota , 1999 ) . Interestingly , single residue mutations in the selectivity filter of NaV channels suffices to increase Ca2+ ion permeation ( Heinemann et al . , 1992; Naylor et al . , 2016 ) . Our calculations indicate that the conductivity of the NaV channel for Ca2+ is ~0 . 4% ( Figure 7 ) . If we assume there exists proportionality between permeability and conduction we can apply the equation of permeability ~ conductance/concentration × valency2 ( Baker et al . , 1971; Meves and Vogel , 1973 ) . With our extracellular solutions [Ca2+]o/[Na+]o being 0 . 0148 and the valence ratio ( Ca2+/Na+ ) 2 being four we can calculate that NaV channels in mammalian axons have a PCa/PNa ratio of 0 . 06 . Notably , the value is in range of direct recordings for PCa/PNa in the sciatic nerve and squid axons ( 0 . 10 and 0 . 14 , Hille , 1972; Meves and Vogel , 1973 ) as well as recordings from NaV1 . 5 channels revealing PCa/PNa ratios of ~0 . 04 ( Cruz et al . , 1999 ) . Such permeability is orders of magnitude lower than PCa/PNa ratios of acetylcholine receptors or NMDA receptors ( 1 . 0 and 17 , respectively ) ( Lewis , 1979; Iino et al . , 1997 ) . An independence of Na+ and Ca2+ ions , modeled as ohmic conductances , will be a major simplification of the NaV channel under multi-ion conditions . Molecular dynamic studies of NaV channels showed that ionic interactions between Na+ and Ca2+ at the channel pore are complex ( Corry , 2013; Boiteux et al . , 2014; Naylor et al . , 2016 ) . The energy barrier of the selectivity filter strongly favors Na+ ions but can be flexible , changing in conformational states and consistent with modest Ca2+ permeation ( Corry , 2013; Boiteux et al . , 2014; Naylor et al . , 2016 ) . In future experiments it will be interesting to obtain more detailed permeability ratios ( PCa/PNa ) by recording changes in Erev with varying intra- and extracellular concentrations , fitting the data to mathematical solutions such as the electro-diffusion theory of Goldman-Hodgkin-Katz ( GHK ) extended with surface charge potentials ( Lewis , 1979; Campbell et al . , 1988 ) , or using Eyring–Läuger theory based on individual ionic rate constants ( Läuger , 1973 ) . Such experiments would in particular be interesting for NaV1 . 6 , the main isoform expressed in axonal domains ( Lorincz and Nusser , 2010; Kole and Stuart , 2012 ) . Although the Ca2+ conductivity of the channels is small it achieves near–micromolar Ca2+ changes in axons as NaV channels are clustered to very high densities ( ~1000 channels µm–2 ) at the AIS and nodes of Ranvier ( Neumcke and Stämpfli , 1982; Lorincz and Nusser , 2010; Kole and Stuart , 2012 ) . Consistent with this idea , our imaging experiments showed that also nodes of Ranvier produced subthreshold-activated Ca2+ entry ( Figure 1 and Figure 4 ) , suggesting that NaV channel mediated Ca2+ entry could play similar roles in these domains . At the AIS , the rapid opening of high densities of NaV channels may further act as a trigger to amplify [Ca2+]i via the activation of ryanodine receptors , mediating ER store release of Ca2+ from the giant saccular organelle , which extends continuously along the AIS of thick-tufted L5 pyramidal neurons ( Figure 2 ) . The rapid inactivation of NaV channels compared to the slow inactivation of CaV channels will lead to voltage- and time-dependent changes in the relative contribution of NaV and CaV channels to [Ca2+]i . In axons , a single AP will mostly lead to NaV-mediated Ca2+ entry while an increasing number of APs , or longer sustained depolarization , will lead to an accumulation of Ca2+ mediated byCaV channel activation . Indeed , Ca2+ entry via CaV channels has been identified as a major contributor to Ca2+ entry in sciatic nerve or Purkinje axons during trains of APs or prolonged depolarization for hundreds of milliseconds ( Callewaert et al . , 1996; Zhang and David , 2016 ) . However , in vivo recordings from L5 pyramidal neurons show that they typically fire sparsely and on average ~1–4 Hz ( de Kock et al . , 2007 ) and the half-width of axonal APs is about ~300 µs ( Kole et al . , 2007 ) . In this view , NaV-mediated Ca2+ entry may be the main source for activity-dependent [Ca2+]i in the excitable domains of axons under physiological conditions . What could be the functional role of NaV-mediated Ca2+ entry in axon initial segments and nodes of Ranvier ? One downstream target of submembranous axoplasmic [Ca2+]i may be regulation of NaV inactivation kinetics via their Ca2+/calmodulin domain at their C-terminus as has been demonstrated for multiple NaV subtypes , including NaV1 . 2 and NaV1 . 6 ( Sarhan et al . , 2012; Reddy Chichili et al . , 2013; Ben-Johny et al . , 2014; Wang et al . , 2014 ) . Another target of NaV-mediated Ca2+ could be to open axonal large-conductance BKCa channels . The BKCa channel opens with the cooperative action of membrane voltage and [Ca2+]i ≥10 µM to repolarize APs and shorten their duration ( Berkefeld et al . , 2010 ) . Across cell types there are considerable variations in the magnitude and time course of BKCa currents due to differences in nanodomain coupling with CaV channel isoforms . BKCa channels are exclusively activated by P/Q-type CaV channels in cerebellar Purkinje neurons with short AP durations ( Womack et al . , 2004 ) , but in rat chromaffin cells with wider APs , BKCa channels are coupled with the Q- and slower activating L-type CaV channels ( Prakriya and Lingle , 1999 ) . Also in L5 pyramidal neurons BKCa activation shortens the duration of somatic APs from ~1 ms to ~600 µs ( Yu et al . , 2010; Bock and Stuart , 2016; Roshchin et al . , 2018 ) . Considering the brief duration of axonal APs in the L5 pyramidal neurons ( ~300 µs , Kole et al . , 2007 ) NaV channels may provide both a precisely-timed voltage-dependent activation , via Na+ current , as well as a [Ca2+]i rise within 150 µs ( Figure 9 ) , to rapidly open BKCa channels and shape axonal AP repolarization . In agreement with this conjecture our data show that T- and L-type CaV channels are too slow to mediate somatic AP repolarization , leaving open the possibility that a NaV-BKCa channel nanodomain coupling provides the required Ca2+ signal . Firm evidence for such interaction would require mutating the selectivity filter of NaV channels to abolish Ca2+ but not Na+ permeation . Recent two-photon Ca2+ uncaging experiments already showed that the [Ca2+]i rise at the first node of Ranvier in L5 axons opens nodal BKCa channels to shorten the AP duration and facilitate the generation of high firing rates in the proximal axon ( Roshchin et al . , 2018 ) . Further downstream from the initiation site nodal BKCa channels in Purkinje axons play a role in augmenting the hyperpolarization following APs and facilitate recovery from NaV inactivation to prevent propagation failures ( Hirono et al . , 2015 ) . The dual permeation of axonal NaV channels for Na+ and Ca2+ ions may thus serve a common function; mediating the rapid electrical upstroke of the AP and via Ca2+ signaling activating K+ efflux to recover from inactivation and accelerating NaV channel availability for the next AP , representing a fine-tuning specifically to the needs of axonal AP generation and conduction fidelity .
All animal experiments were performed in compliance with the European Communities Council Directive 2010/63/EU effective from 1 January 2013 . They were evaluated and approved by the national CCD authority ( license AVD8010020172426 ) and by the KNAW animal welfare and ethical guidelines and protocols ( DEC NIN 14 . 49 , DEC NIN 12 . 13 , IvD NIN 17 . 21 . 01 and 17 . 21 . 03 ) . Written informed consent was obtained from patients and all procedures on human tissue were performed with the approval of the Medical Ethical Committee of the Amsterdam UMC , location VuMC and in accordance with Dutch license procedures and the Declaration of Helsinki . All data were anonymized . Young-adult male Wistar rats ( RjHan:WI ) were used at an age between P21 and P35 ( Charles River Laboratories and Janvier labs ) . Animals were deeply anaesthetized by 3% isoflurane inhalation , decapitated and 300 µm parasagittal slices containing the primary somatosensory cortex were cut with a Vibratome ( 1200S , Leica Microsystems B . V . ) within ice-cold artificial cerebrospinal fluid ( ACSF ) of the following composition ( in mM ) : 125 NaCl , 3 KCl , 25 glucose , 25 NaHCO3 , 1 . 25 Na2H2PO4 , 1 CaCl2 , 6 MgCl2 , saturated with 95% O2 and 5% CO2 ( pH 7 . 4 ) . Following a recovery period at 35°C for 35–45 min slices were stored at room temperature in the ACSF . Human slices were obtained from non-pathological cortex removed for the surgical treatment of deeper brain structures for mesial temporal lobe epilepsy . After resection , a block of the temporal lobe was placed within 30 s in ice-cold artificial cerebrospinal fluid ( ACSF ) slicing solution which contained 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 laboratory , as described in detail previously ( Testa-Silva et al . , 2014 ) . Transition time between resection of the tissue and preparation of the slices was <15 min . Neocortical slices ( ~350 µm thickness ) were cut in an ice-cold slicing solution , stored for 30 min at 34°C , and afterwards switched to room temperature in standard ACSF . Slices were subsequently transported ( <15 min ) towards the NIN ( KNAW ) in continuously carbogenated ACSF . Human embryonic kidney 293 cells ( HEK 293T/17 cell line , CRL-11268 obtained from ATCC ) were cultured in growth medium consisting of equal parts of Dulbecco’s modified Eagle’s medium ( DMEM ) ( DMEM Glutamax , Gibco , Thermo Fisher Scientific ) and Ham’s F10 nutrient mix ( Gibco , Thermo Fisher Scientific ) , supplemented with 10% fetal calf serum ( FCS ) and 1% penicillin–streptomycin . Cells were split twice a week by trypsinization and grown at 37°C with a humidified atmosphere containing 5% CO2 . STR profiling confirmed a 100% match with the HEK 293T cellline ( ATCC ) . Human SCN2A ( D-splice variant ) , encoding for the alpha subunit of the NaV1 . 2 channel was cloned in pcDNA3 . 1-IRES-GFP , and SCN1B/SCN2B , encoding for beta subunits 1 and 2 , was cloned into pcDNA3 . 1 . These vectors were described previously ( Ben-Shalom et al . , 2017 ) and obtained from Genscript ( Genscript , USA ) . The constructs were amplified in Stbl3 bacteria ( Genscript , USA ) and were purified using the GeneJET Plasmid Maxiprep kit ( ThermoFisher , USA ) according to the manufacturer’s protocols . The plasmids were transiently transfected into 70% confluent HEK-293 cells plated in 12-well plates . Per well , the transfection cocktail contained 500 ng pcDNA3 . 1-SCN2A-IRES-GFP , 290 ng pcDNA3 . 1-SCN1B- IRES-SCN2B and 5 μL of polyethylenimine ( PEI ) diluted in 100 μL 1% saline , incubated for 20 min at room temperature before addition to the culture medium . Cells were incubated with 100 μL of transfection cocktail in 1 mL of culture medium for 24 hr at 37°C in a humidified atmosphere containing 5% CO2 . Cells were trypsinised and used for electrophysiological recording typically 48 hr after transfection . For patch-clamp recording , slices were transferred to a customized upright microscope ( BX51WI , Olympus Nederland BV , or LNscope , Luigs and Neumann , Ratingen , Germany ) . The transmitted light path consisted of a custom made 850 nm Light Emitting Diode ( LED ) light source ( LZ1-10R602 , LED Engin , CA ) , collimated using an aspheric condenser lens ( ACL50832U-B , Thorlabs , Germany ) and adapted to the microscope lamp port using a custom 3D printed adapter and passed through an oblique or Dodt illumination condenser ( WI-OBCD , Olympus ) . The top 50 µm of the slice surface was visualized using an optical pathway consisting of a 60× water immersion objective ( N . A . 1 . 0 , LUMPLFLN60XW , Olympus or N . A . 1 . 1 , LUMPLFLN60XW , Olympus ) , 2× intermediate zoom attachment ( U-ECA , Olympus ) , camera splitter ( U-TRU , Olympus ) with inbuilt 180 mm tube lens on the back port and a 0 . 63× demagnifier ( U-TV0 . 63XC , Olympus ) projected the final image onto a high resolution CCD camera ( CoolSNAP-EZ , Photometrics ) , which was operated using μManager ( Edelstein et al . , 2014 ) . Based on the bright-field image large L5 neurons with an intact axon parallel and close to the surface were targeted for recording . Current-clamp recordings were made with Dagan BVC-700A amplifiers ( Dagan Corporation , MN , USA ) or AxoClamp 900A ( Molecular Devices Limited , UK ) . An Axopatch 200B ( Molecular Devices ) was used for voltage-clamp and AP-clamp experiments . The microscope bath was perfused with oxygenated ( 95% O2 , 5% CO2 ) ACSF consisting of ( in mM ) : 125 NaCl , 3 KCl , 25 glucose , 25 NaHCO3 , 1 . 25 Na2H2PO4 , 2 CaCl2 , and 1 MgCl2 . Patch pipettes were pulled from borosilicate glass ( Harvard Apparatus , Edenbridge , Kent , UK ) pulled to an open tip of 3–6 MΩ resistance . For all current-clamp , subthreshold voltage-clamp ramp and AP-clamp recordings the intracellular solution contained ( in mM ) : 130 K-Gluconate , 10 KCl , 4 Mg-ATP , 0 . 3 Na2-GTP , 10 HEPES and 10 Na2-phosphocreatine ( pH 7 . 25 adjusted with KOH , 280 mOsmol kg−1 ) . The liquid junction potential difference of –13 . 5 mV was corrected in all recordings . For morphological reconstruction , 5 mg ml−1 biocytin was routinely added . Voltage recordings were analogue low-pass filtered at 10 kHz ( Bessel ) and digitally sampled at 100 kHz using A-D converter ( ITC-18 , HEKA Elektronik Dr . Schulze GmbH , Germany ) and data acquisition software Axograph X ( v . 1 . 5 . 4 , Axograph Scientific , NSW , Australia ) . Bridge-balance and capacitances were fully compensated based on small current injections leading to minimal voltage errors . The recording temperature was 33 ± 1°C . Only cells with a stable bridge-balance ( <25 MΩ ) , resting membrane potential and AP shape throughout the recording session were included in the analysis . For voltage-clamp recordings of INa and ICa ( Figure 5 ) the bath was perfused with oxygenated ( 95% O2 , 5% CO2 ) extracellular recording solution consisting of ( in mM ) : 100 NaCl , 3 KCl , 25 glucose , 25 NaHCO3 , 1 . 25 Na2H2PO4 , 2 CaCl2 , 1 MgCl2 , 5 4-AP , 20 TEA-Cl , 0 . 02 CNQX , 0 . 05 D-AP5 , 0 . 02 ZD-7288 , 0 . 01 XE991 and 0 . 003 Gabazine ( SR-95531 ) . The intracellular solution contained ( in mM ) : 130 CsCl , 10 TEA-Cl , 10 HEPES , 4 Mg-ATP , 5 Na2-phosphocreatine and 0 . 3 Na2-GTP ( pH 7 . 25 adjusted with CsOH , 280 mOsmol kg−1 ) . A liquid junction potential of –5 . 6 mV was applied to the recordings . Series resistance was routinely compensated to >75% and the linear leak and capacitance off-line subtracted using a P/9 protocol with 10-fold scaled pulses . Current recordings were analogue low-pass filtered at 10 kHz ( Bessel ) and digitally sampled at 100 kHz using A-D converter ( ITC-18 , HEKA Elektronik Dr . Schulze GmbH , Germany ) and data acquisition software Axograph X ( v . 1 . 5 . 4 , Axograph Scientific , NSW , Australia ) . To improve voltage-clamp of the large and rapid Na+ currents the recordings were made at room temperature ( ~20°C ) . For recordings from HEK-293 cells they were transferred to a recording chamber which was continuously perfused with extracellular solution , containing ( in mM ) : 135 NaCl , 4 . 5 KCl , 2 CaCl2 , 1 MgCl , 10 HEPES and 11 Glucose . The intracellular solution contained ( in mM ) : 110 CsF , 10 NaCl , 20 EGTA and 10 HEPES . In the OGB-1 experiments , we added 100 µM OGB-1 , EGTA was omitted and CsF raised to 120 mM instead of 110 mM . The liquid junction potential difference of –10 mV was corrected for . Whole-cell patch-clamp recordings were made ~48 hr after transfection . Cells were recorded at room temperature ( ~20°C ) and continuously perfused with extracellular solution at a flow rate of 1 . 5 mL∙min–1 . Patch pipettes were pulled to a resistance of 2–3 MΩ . Round , isolated cells with a diameter >10 μm , a smooth cell surface and a moderate EGFP fluorescent signal were selected for recordings ( Figure 7a ) . HEK-293 cells had an average capacitance of 9 . 19 ± 0 . 70 pF ( n = 16 ) . The holding potential was –70 mV and voltage dependence of activation of NaV1 . 2 was determined by an activating protocol consisting of a hyperpolarizing pulse to –130 mV ( 20 ms ) followed by step pulses from –80 mV to +50 mV with increments of 10 mV for 20 ms . Voltage dependence of inactivation was assessed with voltage pulses from –130 mV to –30 mV with increments of 10 mV for 100 ms duration , followed by a depolarizing pulse to –20 mV for 20 ms . A P/5 leak subtraction protocol ( 10-fold scaling ) was used to subtract remaining capacitive and leak currents . Series resistance was not compensated . EGTA and blockers were added to the appropriate concentration to the ACSF and perfused . The extracellular Ca2+ ( [Ca2+]o ) was lowered by bath application of 2 . 5 mM EGTA and using the online maxchelator tool ( https://somapp . ucdmc . ucdavis . edu/pharmacology/bers/maxchelator/CaMgATPEGTA-TS . htm; Bers et al . , 2010 ) we calculated the remaining [Ca2+]o to be 437 nM , based on a recording temperature of 35°C , a pH of 7 . 4 and an ionic strength of 0 . 15 M of the free ions in our solution . To limit hyperexcitability in the presence of EGTA , we added synaptic blockers to the ACSF ( 20 µM CNXQ and 50 µM D-AP5 ) and kept the a-EPSP voltage peak amplitude constant by reducing the amplitude of the current injections in both control and EGTA measurements ( Figure 2 ) . To prevent precipitation of Ni2+ we used phosphate-free extracellular solutions containing ( in mM ) : 126 . 25 NaCl , 3 KCl , 25 glucose , 25 NaHCO3 , 2 CaCl2 , 1 MgCl2 and 0 . 1 Ni2+ . A > 10% increase in fluorescence baseline was observed in some experiments ( 3 out of 8 recordings ) , which were subsequently excluded . Two blockers ( SNX-478 and ω-conotoxin MVIIC ) were not perfused but were locally puffed using a Picospritzer III ( Intracel ) for 3 s ending 0 . 5 s before imaging to avoid vibration artifacts . Bovine serum albumin ( 0 . 1 mg/ml ) was added to the rACSF before ω-conotoxin MVIIC was introduced to minimize non-specific binding of the drug . To optically record [Ca2+]i and [Na+]i in axons , membrane impermeable Ca2+ and/or Na+ indicators were added to intracellular solutions . For Ca2+ we used OGB-1 ( 100 µM ) , OGB-5N ( 1 mM ) or bis-Fura-2 ( 200 µM ) and for Na+ imaging we used sodium-binding benzofuran isophthalate ( SBFI , 1–1 . 5 mM ) . Patch pipettes were first filled with dye-free solution for half of the tapered part of the pipette tip , then backfilled with the dye-containing solution . Fluorescence intensity at the AIS was monitored during dye loading and imaging started only when the indicators were fully equilibrated ( typically after 0 . 5–1 hr ) . Optical recordings of Ca2+ or Na+ dye fluorescence changes were obtained with wide-field epifluorescence microscopy . Fluorescence was collected by the same 60× water immersion objective , passed through the microscope tube lens ( U-TR30IR , Olympus ) and projected onto a rapid data-acquisition camera with relatively low spatial resolution ( 80 × 80 pixels ) but high dynamic range ( 14 bits ) and low read noise ( NeuroCCD-SM , RedShirtImaging LLC , Decatur , GA ) via a 0 . 1× or 0 . 35× demagnifier . The CCD frame corresponded to an area of approximately 320 or 91 µm2 in the object plane with each individual pixel receiving light from an area of ∼4 × 4 or 1 . 1 × 1 . 1 µm2 , respectively . High-speed recordings ( 20 kHz ) were always performed with the 0 . 35× demagnifier and with 3 × 3 binning of pixels on the chip , the ultimate pixel sizes in these recordings were ∼3 . 4 × 3 . 4 µm2 . The two recordings from human cells were performed under a 100× NA 1 . 1 Nikon objective ( MRL07920 ) in combination with a 0 . 1× demagnifier , resulting in a pixel size of ∼2 . 4 × 2 . 4 µm2 ( see Figure 1—figure supplement 1 ) . The epifluorescence light path consisted of an excitation LED light source , which was collimated using an aspheric lens ( ACL5040U-A , Thorlabs , Germany ) and the appropriate excitation filter , dichroic mirror and emission filter . For OGB-1 and OGB-5N excitation a 470 nm LED was used ( SP-01-B4 , Luxeon Star LEDs , Canada ) , the excitation light was filtered with 475/30 nm ( 475 nm center wavelength , 30 nm wide ) excitation filter , reflected to the preparation by a dichroic mirror with a central wavelength of 500 nm and the fluorescence light was passed through a 520 nm barrier filter ( U-MWB2 cube , Olympus ) . For SBFI excitation , LED light ( 365 nm LED , LZ1-10UV00 , Ledengin , USA ) was filtered by 357/44 nm filter ( FF01-357/44-25 , Semrock ) , a dichroic mirror with edge at 415 nm ( Di03-R405-t1−25 × 36 , Semrock ) reflected excitation light to the sample and the emission light was then passed through a long-pass colored glass filter with the edge at 400 nm ( FGL400 , Thorlabs ) . For combined Na+ and Ca2+ imaging ( Figure 6—figure supplement 1 ) , the light from the 365 and 470 LEDs was combined by a dichroic mirror with edge at 458 nm ( FF458-Di02 , Semrock ) and the filter set switched between trials . Light was directed through a fluorescence illuminator ( BX-RFA , Olympus ) equipped with a rectangular field stop providing an open area of 150 × 250 µm to reduce phototoxicity ( U-RFSS , Olympus ) . The cell body was positioned just outside the field stop and the axon in the middle parallel to the long side ( see e . g . Figure 1 ) . The critical benefit of epifluorescence measurements over two-photon imaging is increased light collection ( ~90% quantum efficiency , low-read noise of the CCD camera ) enabling a high sensitivity and temporal fidelity . We optimized all imaging parameters to obtain maximal signal to noise ratio , which allowed us to image at the maximum acquisition rate of 20 kHz . In addition to the light collection optimization and selective targeting of superficial neurons , multiple trials were averaged to improve signal-to-noise ratio ( typically 20–40 ) . Fluorescence signals were temporally aligned to the electrophysiological voltage or current signals . For optical recording of ICa , which requires the transformation of ΔF/F into the first time derivative , Ca2+ binding to the indicator must be proportional to [Ca2+]i and the endogenous buffering capacity to be low . Based on the submillisecond equilibration time of OGB-5N and imaging at the maximally possible frame rate of 20 kHz it was recently shown that in CA1 hippocampal neurons these conditions are met and optically recorded ICa tracks electrically recorded ICa , enabling the identification of CaV channel subtypes in dendrites ( Jaafari et al . , 2014 ) . Considering the low buffering capacities of endogenous buffers in the axon ( κs ≈ 20 ) ( Jackson and Redman , 2003; Delvendahl et al . , 2015 ) , we employed this technique in the AIS . Ca2+ imaging in HEK-293 cells was performed with 0 . 1 mM OGB1 added to the HEK-293 cell intracellular solution ( from which EGTA was omitted ) . The fluorescence was recorded during a 200 Hz 1 s train of depolarizing pulses from –120 to –30 or –20 mV ( corrected for liquid junction potential ) . Voltage imaging in neurons was performed as reported previously ( Hamada et al . , 2017 ) . Neurons were filled with intracellular solution containing JPW3028 ( 0 . 8 mM ) for typically 1 hr at room temperature , after which the patch pipette was retracted and the dye was left to diffuse into the lipid membranes for 1–4 hr . Subsequently the bath temperature was increased to 35°C and the cell was re-patched with normal intracellular solution . A 530 nm LED ( SP-05-G4 , Luxeon Star LEDs , Canada ) was used for excitation of the dye . The excitation light was filtered with a 530/20 nm filter ( BP510-550 , Olympus ) , reflected to the sample by a dicroic mirror with a center wavelength of 570 nm ( DM570 , Olympus ) and the emission light filtered by a 590 longpass filter ( BA590 , Olumpys ) . Data were collected at 20 kHz and low-pass filtered by a binomial filter ( one pass ) and averaged over 20–30 trials . Voltage imaging in HEK-293 cells was performed identically , with the only exception being that the experiments were performed at 20°C and the dye diffused equally in the small round cells , so imaging experiments were initiated 20 min after obtaining whole-cell configuration . The cells were maintained at –75 mV holding potential and the fluorescence of JPW3028 recorded at 1 kHz . The voltage command consisted of 100 ms steps of 50 mV increasing steps with a maximal step of +250 mV relative to holding potential . The average ΔF/F per voltage step was defined as the first 20 frames of each bleach corrected and normalized voltage step . Imaging data analysis was performed using Neuroplex ( Redshirt imaging ) , Axograph and Excel . Fluorescence signals were always background-subtracted . To correct for bleach effects , every 5th trial was recorded without current injection . A first order exponential was fitted to the average of the bleach trials and normalized to the peak . The average of the signal trials was divided by this trace to correct for bleach decay . Values for each ROI were defined as a fractional fluorescence change ( ΔF/Fbaseline ) , where Fbaseline is the raw intensity average of 10 frames before the signal ( subthreshold or AP ) was initiated . Pixels were color coded with ‘physics’ color scheme from FIJI image processing software ( NIH , USA ) ( Schindelin et al . , 2012 ) . For both OGB-1 and OGB-5N , we recorded the fluorescence in response to subthreshold stimuli , single APs and multiple APs . The ΔF/F response to subthreshold stimuli and a single AP was always below dye saturation . For OGB-1 , we recorded trains of APs and observed that a single AP was 25 ± 2 . 7% of dye saturation ( n = 8 ) and for OGN-5N , we observed a linear increase from 1 to 3 APs ( n = 3 ) , indicating that the fluorescence of a single AP was far from dye saturation . To compare kinetics between electrically and optically recorded currents , electrical currents were first downsampled to 20 kHz ( optical acquisition rate ) . The optical trace was differentiated and then inverted ( to mimic INa , which is conventionally depicted as a negative , inward current ) . All current traces were then filtered with a 3-window binomial filter of 50–150 iterations ( generally 100 ) . The traces were baselined to the current before the onset of the fast current and normalized to the peak of the current . Because the ICa_opt traces were too noisy to be fitted with a single exponential fit , conventionally used to obtain activation rise time , we used a Boltzmann sigmoid function to obtain the slopes of the traces . y= bottom+ top-bottom1+ex-x0k We fitted all traces using Axograph and the slope values ( k ) were used to compare activation kinetics . Although the INa peak amplitude recorded at the soma highly varied between neurons ( –36 . 61 ± 6 . 53 nA ) , the slope was nearly constant ( 207 ± 0 . 007 µs , n = 17 ) . To estimate the absolute [Na+]i and [Ca2+]i in response to a single AP , we used a ratiometric imaging approach . Patch pipettes were front filled with clear intracellular solution and back-filled with intracellular solution containing 1 . 5 mM SBFI ( Invitrogen ) or 200 µM bis-Fura-2 hexapotassium salt ( bF2 , Biotium ) . Fluorescent emission of ratiometric indicators depends on the ionic concentration and the excitation wavelength where an increase in [Ca2+]i produces an increase in bF2 fluorescence with the wavelength of 340 nm but a decrease with 385 nm . On the other hand , with SBFI an increase in [Na+]i decreases SBFI fluorescence at 340 nm but does not alter at 385 nm excitation wavelength . Using the ratio ( R ) corrects for differences in cytosolic volume or dye concentration differences along imaged compartments ( Langer and Rose , 2009 ) . The sources of excitation light were two LEDs ( Thorlabs ) with peaks at 340 nm and 385 nm , fitted with band pass excitation filters at 340/22 and 387/11 nm ( FF01-340/22-25 and FF01-387/11-25 , Semrock ) and combined by a dichroic mirror with a central wavelength of 376 nm ( FF376-Di01−25 × 36 , Semrock ) . The excitation light was reflected to the sample by a dichroic mirror with a central wavelength at 405 nm ( Di01-R405−25 × 36 , Semrock ) and passed through the objective to the sample . The fluorescent emission signals were passed through a 420 long pass filter ( Thorlabs ) . Ratiometric imaging was performed by alternatingly triggering each LED at the frame rate of the camera , as described previously ( Miyazaki and Ross , 2015 ) . This was achieved by combining custom designed Arduino/Parallax machines with Cyclops LED drivers . These hardware solutions allowed us to digitally control the voltage driving the LED , thus having maximum control over excitation light intensity . Fluorescence emission signals originating from each LED were separated with custom written software ( FrameSplitter . txt , Battefeld et al . , 2018 ) . The camera operated at 0 . 5–1 . 0 kHz , resulting in a ratiometric frame rate of 0 . 25–0 . 5 kHz . Per experiment the fluorescent signals were averaged for 40 to 120 trials . The ratio R was defined as F1/F2 , where F1 and F2 are the background-subtracted fluorescence intensities at excitation with 340 nm and 385 nm , respectively . In order to scale ratiometric bF2 signals to absolute changes in Ca2+ concentration , we used the standard equation for ratiometric measurements ( equation 1 in Figure 8—figure supplement 2; Grynkiewicz et al . , 1985 ) , which depends on KD , the dissociation constant , Rmin and Rmax , the ratio in zero and dye-saturating Ca2+ , respectively and the scaling factor ( Sf2/Sb2 ) , defined as the fluorescence intensity at excitation with 385 nm of zero Ca2+ divided by saturating Ca2+ . These values were determined in an ex situ calibration , by measuring the ratiometric signal of solutions containing 0 Ca2+ and a high [Ca2+] ( Figure 8—figure supplement 2 ) . The solutions closely mimicked intracellular solutions and contained ( in mM ) : 110 K-gluconate , 4 . 4 or 21 KCl , 0 or 10 CaCl2 , 3 . 8 or 5 . 36 MgCl2 , 10 HEPES , 4 Mg-ATP , 0 . 3 Na2-GTP , 10 Na2-phosphocreatine , 10 EGTA , and 0 . 2 bF2 . The final free [Ca2+] depends on interaction between Ca2+ , Mg2+ and EGTA and was calculated using the maxchelator tool ( Bers et al . , 2010 ) . [Ca2+]min was 0 and [Ca2+]max was 4 . 39 µM . We repeated the calibration experiment three times . In our experimental setting , the KD of bF2 was 507 . 3 ± 5 . 7 nM , which matches with the reported value of 525 nM in the presence of Mg2+ ( Molecular Probes Handbook , Thermofischer ) . We then performed ratiometric imaging in the AIS in response to a single AP . ∆R/R was calculated by dividing every ratio by the average of the baseline ratio before the onset of the AP ( similar to the conventional ∆F/F ) . We used the KD as determined from our calibration experiments . The Rmin was scaled to be ~95% of Rbaseline to result in resting [Ca2+]i of 50 nM and Rmax as established in the calibration experiments . If the LED intensity of the cellular recording was different from the calibration intensity used during the calibration experiments , the Rmax was corrected linearly , assuming that Rmin/Rmax was constant . These experiments showed that after a single AP , [Ca2+]free in the AIS rises with 55 . 6 ± 12 . 6 nM ( see Figure 8—figure supplement 2 ) . Because Rmin and Rmax were not measured in situ we verified the Δ[Ca2+]free with an alternative analysis that is independent of the exact values for KD , Rmin and Rmax ( equation 2 in Figure 8—figure supplement 2; Langer and Rose , 2009 ) . In this approach , changes in fluorescence ratio ΔR/R0 ( % ) are plotted versus [Ca2+]free , showing a nearly linear increase in ΔR/R0 ( % ) for low [Ca2+]free ( see Figure 8—figure supplement 2 ) . A linear fit to the region of [Ca2+]i between 0 and 193 nM indicated that a 1% increase in ΔR/RbF2 corresponded to a Δ[Ca2+]i of ~10 . 4 nM ( R2 = 0 . 99 , six concentrations , n = 3 repetitions ) . We measured an AP-evoked Δ[Ca2+]free of 52 . 5 ± 12 . 2 nM ( Figure 8—figure supplement 2 ) , in good support of the standard Ca2+ measurement approach . We analyzed the ratiometric SBFI data only using this second approach , which is standard for SBFI measurements ( Langer and Rose , 2009 ) . The two base calibration solutions contained ( in mM ) : 130 K-Gluconate or Na-Gluconate , 10 KCl of NaCl , 0 . 3 Tris2-GTP or Na2-GTP , 10 HEPES , 4 Mg2+ATP , 10 Tris2-phospocreatine or Na2-phospocreatine , and 1 . 5 SBFI , pH 7 . 25 adjusted with Tris base . These two base solutions provided a range of 0–160 . 6 mM [Na+] . When normalized to the ratio obtained in Na+-free solution ( R0 ) , a 1% increase in ΔR/RSBFI corresponded to a Δ[Na+]i of 0 . 35 mM for Δ[Na+]i between 0 and 48 mM ( R2 = 0 . 98; eight concentrations , n = 3 repetitions , Figure 8—figure supplement 2 ) . An AP evoked a Δ[Na+]i of 1 . 49 ± 0 . 2 mM ( Figure 8—figure supplement 2 ) . Following imaging experiments , the slices were fixed using 4% PFA in 0 . 1 M phosphate-buffered saline ( PBS ) , pH 7 . 4 for 20 min and stored in 0 . 1 M PBS , pH 7 . 4 at 4°C . For triple immunohistological labeling the slices were washed three times in PBS and then incubated in a blocking solution ( 10% normal goat serum , 0 . 5% Triton X-100 in PBS ) at room temperature for two hours , followed by 24 hr incubation at room temperature in the blocking solution containing primary antibodies: streptavidin Alexa-488 conjugate ( 1:500; Invitrogen ) , primary antibody for giant saccular organelle Synaptopodin ( rabbit; 1:500; Sigma-Aldrich Chemie ) and antibody for AIS marker: Ankyrin G ( mouse; 1:100; Neuromab ) or ßIV-spectrin ( mouse; 1:250; Neuromab , see also Key Resources Table ) . The slices were 3x washed in 0 . 1 M PBS and then incubated with secondary antibodies: Alexa-555 goat anti rabbit IgG ( 1:500; Invitrogen ) and Alexa 633 goat anti mouse IgG ( 1:500; Invitrogen ) . Subsequently , the slices were 3x washed in 0 . 1 M PBS and mounted with Vectashield mounting medium with 4 , 6-diamidino-2-phenylindole ( DAPI; Vector Laboratories ) . Images ( bit depth , 8 ) were collected as described previously ( Hamada et al . , 2016 ) . To align confocal images and the Ca2+ fluorescence images of the RedShirt CCD camera ( Figures 1 , 2 and 4 ) we used the original calibrated images of the two systems . The maximum Ca2+ fluorescence image was calibrated within the original optical path . We overlaid the maximum Ca2+ fluorescence image of the RedShirt camera and the maximum projection of the streptavidin image of the neuron morphology from confocal microscopy within ImageJ and applied only a rotation translation to visually match the two images based on the AP-evoked Ca2+ signals spreading into dendrites and axons . All model simulations were performed with NEURON ( v . 7 . 5 ) ( Hines and Carnevale , 2001 ) . A single compartment was created with length and diameter dimensions of 10 µm and nseg = 10 , with specific membrane capacitance of 1 . 0 µF cm–2 , specific membrane resistance of 25 kΩ cm2 and specific axial resistivity of 150 Ω cm . The resting membrane potential set to –77 mV using e_pass . Conductance models for Ca2+ were based on the high-voltage activated ( CaH ) and a T-type CaV channel model ( CaT ) obtained from ModelDB ( https://senselab . med . yale . edu/ModelDB/ ) ( Mainen and Sejnowski , 1996 ) . Ca2+ conductivity of NaV channels was modeled by including a standard ohmic Ca2+ ion mechanism with a reversal potential ( eca ) of +140 mV into a mathematical 8-state Na+ conductance model , computing simultaneously voltage- and time-dependence of the Ca2+ current ICa ( Na ) and INa based on experimentally constrained rate constants of somatodendritic and axonal INa ( Schmidt-Hieber and Bischofberger , 2010 ) . The kinetics of the voltage-gated conductance models was examined by fitting the current rise times with an exponential function for a –35 mV command potential , resampling the simulated traces to 20 kHz . The results showed that the INa in the model activated with 240 µs and ICa ( Na ) ( 0 . 5% conductivity ratio ) with 280 µs . In comparison , ICaT activated with 4 . 88 ms and ICaH with 6 . 51 ms . These time constants are well in range of the experimentally determined values for the TTX- and Ni2+-sensitive components recorded at the soma ( Figure 5 ) . For Figure 8—figure supplement 1 we used an AP recorded from the L5 pyramidal neuron AIS at 100 kHz ( threshold-to-peak , 94 mV; half-width duration of 285 µs [Hallermann et al . , 2012] as the command potential in VectorPlay linked to the SEClamp function in NEURON ( with Rs being infinitely small ) . Single compartmental models were run at dt of 10 µs at a nominal temperature of 33°C . Conductance-based multi-compartmental simulations were performed with an anatomically realistic reconstructed rat L5 pyramidal neuron ( NeuroMorpho . Org ID: NMO_75667 , Neuron Name 2014-04-01_1 ) . The morphology was acquired with a confocal microscope at 2048 × 2048 pixels ( 1 . 0 µm z-steps , Leica SP8 ) using a 40× oil immersion objective ( NA 1 . 3 ) scanning both the biocytin-streptavidin fluorescence and the ßIV-spectrin fluorescence . Uncompressed image stacks ( ~20 GB ) were imported and reconstructed into Neurolucida ( v . 10 , MBF Bioscience Inc , Germany ) , compartmentalized for the AIS and nodes as described previously ( Hamada et al . , 2016 ) and imported with the 3D import tool in NEURON ( Carnevale and Hines , 2006 ) . Multicompartmental simulations were performed to estimate the detailed ionic accumulation , concentration and diffusion in the proximal sites of the axon and match our experimental recordings as close as possible . Ca2+ diffusion , buffering and pump ( cdp ) mechanisms were implemented based on the algorithms described in the NEURON book ( Chapter 9 , example 9 . 8 in Carnevale and Hines , 2006 ) and on a previously published Ca2+ model ( Fink et al . , 2000 ) ( available at ModelDB , accession number 125745 , https://senselab . med . yale . edu/ModelDB/ ) . We implemented cdp . mod ( Fink et al . , 2000 ) with the following alterations: we removed all SERCA related parameters , updated some starting values to our experimental conditions and extended the models to report not only [Ca2+]i , but also to simulate the Ca2+ indicator response ΔF/F , using the equation:F = [dye]free+ c*[Ca2++ buffer][dye]total With c being a constant to scale simulated ΔF/F . Because the equation was used to match the simulation to experimental data with regard to the temporal dynamics of Ca2+ extrusion , the absolute amplitude of ΔF/F was not used and c was set to a nominal value of 6 . The different Ca2+ indicators used experimentally were implemented by adjusting the concentration of the exogenous buffer , and its known or measured KD . Static Ca2+ buffering properties of endogenous organelles ( κs ) were simulated with a TBufs of 100–400 µM and KDs of 10 µM , to mimic a κs of 10–40 ( Jackson and Redman , 2003; Delvendahl et al . , 2015 ) . To constrain the peak Na+ conductance densities ( g-Na ) we injected a 3 ms square current pulse in the somatic compartment and iteratively adjusted g-Na and g-K . We varied both their peak conductance densities as well as the voltage-dependence of activation of NaV and KV channels by constraining the model AP to the experimentally recorded AP of the same neuron , with the aim to match the AP both in the V-t as well as the phase-plane dimensions recorded and simulated at 100 kHz ( Figure 7c ) . To further constrain g-Na we compared the AP-evoked [Na+] with the experimental recordings using the ratiometric indicator SBFI , yielding a Δ[Na+]i of on average ~1 . 5 ± 0 . 2 mM ( n = 5 , imaged at 0 . 5 kHz; Figure 8—figure supplement 2 ) . CaV channels were incorporated based on previously published models ( Mainen and Sejnowski , 1996 ) and CaV channel conductance was separated in high- and low-voltage ( T-type ) activated channels and was varied between 2 and 4 pS µm–2 in the AIS , 8 and 4 pS µm–2 in the soma and ranged between 0 . 5 and 4 pS µm–2 in the dendrites . All statistical tests were done in GraphPad Prism 8 ( version 8 . 1 . 2 , GraphPad Software , Inc ) . Sample sizes for the pharmacological experiments were estimated based on the following assumptions: to observe a 50% block ( based on Bender and Trussell , 2009 ) with 25% standard deviation ( relative to mean ) with a power of 0 . 8 and a type I error probability of 0 . 05 , we would need a minimum of 4 paired recordings per treatment ( PS Software version 3 . 1 . 6 ) . The cutoff significance level ( P ) was 0 . 05 . Control peak ΔF/F values at the AIS in response to both subthreshold and AP signals were tested for normality . Since both data sets passed the D’Agostino and Pearson normality test , parametric tests were used to test all differences between peak OGB-1 ΔF/F . To compare the spatial differences in signal amplitude we used one-way ANOVAs with multiple comparisons with Tukey correction for false positives . A linear regression was used to assess the synaptopodin and AIS marker length ( Ankyrin G or ß4-spectrin ) relationship . We used one-tailed ratio ( compared log differences in the data set ) paired t-test when analyzing all our toxin data . Differences between toxin and control give a measure of absolute reduction; differences between logarithms give a measure of relative reduction log toxin – log control = log ( toxin/control ) . One tailed test was used on the premise that toxins reduce Ca2+ signals . The exception was the effect of NCX for which we used a two-tailed ratio paired t-test . The OGB-5N peaks in response to subthreshold depolarizations passed the D’Agostino and Pearson normality test and to compare the effects of TTX and QX-314 on subthreshold Ca2+ responses we used two-way-ANOVA with Sidak’s correction for false positives . The following data sets passed the Shapiro-Wilk test for normality , so we compared the means using parametric tests: rise and decay times of OGB1 before and after Ca2+ store release block , peak ΔF/F JPW3028 , peak ΔF/F OGB-5N and slopes of INa , ICa_opt and ICa , the ratio of subthreshold peak to AP peak between sodium ( SBFI ) and calcium ( OGB-5N ) fluorescence . The following data did not pass the Shapiro-Wilk test for normality , so we compared the means using nonparametric tests: the slope AP peak between sodium ( SBFI ) and calcium ( OGB-5N ) fluorescence , the peak INa and peak OGB-1 ΔF/F in HEK-293 cells . All data generated or analyzed are in the manuscript or supporting files . The source data files are provided for Figures 1–7 and Table 1 . The NEURON model morphology is available at NeuroMorpho . Org ID: NMO_75667 , Neuron Name 2014-04-01_1 and the mod file used to model Ca2+ diffusion and buffering is available at ModelDB , accession 125745 ( https://senselab . med . yale . edu/ModelDB/ ) , with adjustments described in ‘Model simulations with a multicompartmental model’ . | Nerve cells communicate using tiny electrical impulses called action potentials . Special proteins termed ion channels produce these electric signals by allowing specific charged particles , or ions , to pass in or out of cells across its membrane . When a nerve cell ‘fires’ an action potential , specific ion channels briefly open to let in a surge of positively charged ions which electrify the cell . Action potentials begin in the same place in each nerve cell , at an area called the axon initial segment . The large number of sodium channels at this site kick-start the influx of positively charged sodium ions ensuring that every action potential starts from the same place . Previous research has shown that , when action potentials begin , the concentration of calcium ions at the axon initial segment also increases , but it was not clear which ion channels were responsible for this entry of calcium . Channels that are selective for calcium ions are the prime candidates for this process . However , research in squid nerve cells gave rise to an unexpected idea by suggesting that sodium channels may not exclusively let in sodium but also allow some calcium ions to pass through . Hanemaaijer , Popovic et al . therefore wanted to test the routes that calcium ions take and see whether the sodium channels in mammalian nerve cells are also permeable to calcium . Experiments using fluorescent dyes to track the concentration of calcium in rat and human nerve cells showed that calcium ions accumulated at the axon initial segment when action potentials fired . Most of this increase in calcium could be stopped by treating the neurons with a toxin that prevents sodium channels from opening . Electrical manipulations of the cells revealed that , in this context , the calcium ions were effectively behaving like sodium ions . Human kidney cells were then engineered to produce the sodium channel protein . This confirmed that calcium and sodium ions were indeed both passing through the same channel . These results shed new light on the relationship between calcium ions and sodium channels within the mammalian nervous system and that this interplay occurs at the axon initial segment of the cell . Genetic mutations that ‘nudge’ sodium channels towards favoring calcium entry are also found in patients with autism spectrum disorders , and so this new finding may contribute to our understanding of these conditions . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"neuroscience"
] | 2020 | Ca2+ entry through NaV channels generates submillisecond axonal Ca2+ signaling |
Parasitic helminths infect over a billion humans . To survive in the low oxygen environment of their hosts , these parasites use unusual anaerobic metabolism — this requires rhodoquinone ( RQ ) , an electron carrier that is made by very few animal species . Crucially RQ is not made or used by any parasitic hosts and RQ synthesis is thus an ideal target for anthelmintics . However , little is known about how RQ is made and no drugs are known to block RQ synthesis . C . elegans makes RQ and can use RQ-dependent metabolic pathways — here , we use C . elegans genetics to show that tryptophan degradation via the kynurenine pathway is required to generate the key amine-containing precursors for RQ synthesis . We show that C . elegans requires RQ for survival in hypoxic conditions and , finally , we establish a high throughput assay for drugs that block RQ-dependent metabolism . This may drive the development of a new class of anthelmintic drugs . This study is a key first step in understanding how RQ is made in parasitic helminths .
Soil-transmitted Helminths ( STHs ) are major human pathogens ( WHO Expert Committee on the Control of Schistosomiasis , 2002 ) . Over a billion humans are infected with an STH — the roundworm Ascaris lumbricoides , the whipworm Trichuris trichuria , and the hookworm Necator americanus account for most of these infections ( WHO Expert Committee on the Control of Schistosomiasis , 2002 ) . STHs are transmitted from human to human via the soil where eggs from human faeces develop into infective stages which then enter new hosts ( reviewed in Brooker et al . , 2006 ) . On infection , STHs encounter a very different environment and require multiple strategies to be able to survive . One of the major changes is the availability of oxygen . While there is abundant oxygen outside their hosts , in many host tissues there is little available oxygen — this is especially true in the intestine where oxygen levels drop steeply to near anoxia in the lumen ( reviewed in Espey , 2013 ) . To survive in these hypoxic conditions , parasites must switch from aerobic respiration to anaerobic respiration; crucially , the anaerobic metabolic pathways that STHs depend on are unusual and are not used in any host ( Van Hellemond et al . , 1995 ) . Inhibiting these anaerobic pathways is therefore a possible way to kill the parasites while leaving the host unaffected . During aerobic respiration in helminths , the great majority of ATP is made in the mitochondrion ( Tielens , 1994; Tielens et al . , 1984 ) . Electrons enter the Electron Transport Chain ( ETC ) either at Complex I or via several quinone-coupled dehydrogenases ( QDHs from here on ) . These QDHs include Succinate Dehydrogenase ( Complex II ) and Electron-Transferring Flavoprotein Dehydrogenase ( ETFDH ) ( Komuniecki et al . , 1989; Ma et al . , 1993; Rioux and Komuniecki , 1984 ) . The electrons entering the ETC are first transferred to the lipid soluble electron carrier ubiquinone ( UQ ) ( Crane et al . , 1957; Mitchell , 1975 ) . From UQ , they are ultimately carried to Complex III then IV where they are finally transferred onto oxygen as the terminal electron acceptor ( see Figure 1a ) . Electron transport is coupled to proton pumping into the inner membrane space of the mitochondrion — this establishes a proton gradient which is used to power the F0F1-ATP synthase ( Mitchell , 1961 ) . When there is insufficient oxygen to accept electrons at Complex IV , or when inhibitors of Complex IV such as cyanide are present ( Antonini et al . , 1971; Nicholls et al . , 1972 ) , almost all animals stop using the ETC and rely on anaerobic glycolysis to make ATP , generating lactate ( Isom et al . , 1975; Meyerhof , 1927 ) . STHs , however , have evolved a different solution that allow them to survive months in the hypoxic host environment . Electrons still enter the ETC at Complex I , Complex I still pumps protons to generate the proton motive force ( PMF ) , and ATP is still made by the F0F1ATPase , powered by the PMF . However , rather than the electrons passing through the ETC to oxygen as the terminal electron acceptor , they exit the ETC immediately downstream of Complex I onto a number of alternative terminal electron acceptors ( Figure 1b ) ( reviewed in Hochachka and Mustafa , 1972; Müller et al . , 2012 ) . This transfer of the electrons out of the ETC and onto alternative electron acceptors requires the quinone-coupled dehydrogenases ( Kita , 1992; Ma et al . , 1993 ) . Under aerobic conditions these QDHs act as entry points to the ETC , transferring electrons from their substrates to UQ . Crucially , the reactions catalysed by these QDHs are reversed in anaerobic conditions — they now act as reductases transferring electrons out of the ETC and onto their products . For example , Complex II acts as a succinate dehydrogenase in aerobic conditions , generating fumarate; in anaerobic conditions , it reduces fumarate generating succinate as a terminal electron sink ( Figure 1c ) ( Kmetec and Bueding , 1961; Sato et al . , 1972; Saz and Vidrine , 1959; Van Hellemond et al . , 1995; Takamiya et al . , 1999 ) . In this way , an entry of electrons into the ETC from a variety of electron donors in aerobic conditions is reversed to provide an exit from the ETC onto a variety of electron acceptors in anaerobic conditions . The unusual ETC wiring used by STHs to survive anaerobic conditions requires an unusual electron carrier , rhodoquinone ( RQ ) ( Moore and Folkers , 1965 ) . RQ and UQ are highly related molecules — the sole difference is the presence of an amine group on the quinone ring of RQ ( Figure 1d ) . This changes the biophysical properties of the quinone ring: while UQ can accept electrons from the QDHs as they flow into the ETC under aerobic conditions , UQ cannot carry electrons of the correct electropotential to drive the reverse reactions in anaerobic conditions ( Fioravanti and Kim , 1988; Sato et al . , 1972 ) . RQ can carry such electrons ( Fioravanti and Kim , 1988; Sato et al . , 1972 ) , however , and the ability of STHs to drive their unusual anaerobic metabolism in their hosts is absolutely dependent on RQ . The single amine group that differs between UQ and RQ thus affects the health of over a billion humans . RQ is found in very few animal species — only helminths , molluscs and annelids are known to make RQ ( Allen , 1973; Fioravanti and Kim , 1988; Sato and Ozawa , 1969; Takamiya et al . , 1999; Van Hellemond et al . , 1995 ) . Since no host animals make RQ , inhibiting RQ synthesis or RQ use is a potentially powerful way to target parasites inside their host . Currently however little is known about RQ synthesis . The most mature studies have focused on the purple Proteobacterium R . rubrum , where RQ appears to derive from UQ ( Brajcich et al . , 2010 ) . RQ synthesis in R . rubrum requires the gene rquA ( Lonjers et al . , 2012 ) which is the first and thus far only gene known to be required for RQ synthesis in any organism . The role of rquA was initially unclear ( Lonjers et al . , 2012 ) , but very recently it has emerged that it is capable of converting UQ to RQ ( Bernert et al . , 2019 ) . In animals , the situation is even more blank: nothing is known about which genes are required for RQ synthesis , there are no clear rquA orthologues ( Lonjers et al . , 2012 ) , and there are no drugs that are known to prevent RQ synthesis . This is partly because no tractable animal model has been established in which to study RQ synthesis and use . Previous studies have shown that C . elegans , a free-living helminth , can make RQ ( Takamiya et al . , 1999 ) and that when C . elegans is exposed to hypoxic conditions it undergoes major metabolic changes that resemble those that occur in STHs when they are in the hypoxic environment of their hosts ( Butler et al . , 2012; Föll et al . , 1999 ) . This suggested to us that we could establish C . elegans as a model for dissecting the pathway of RQ synthesis and for screening for drugs that block RQ synthesis or use . We confirm that C . elegans makes RQ and also that it uses RQ-dependent metabolism when unable to use oxygen as a terminal electron acceptor . We show that in C . elegans RQ synthesis requires the activity of the kynurenine pathway . This pathway is key for the metabolism of tryptophan and generating metabolites that include 3-hydroxyanthranilate ( 3 HA ) . Our data suggest that 3 HA ( or a highly related molecule generated by the kynurenine pathway ) is the amine-containing precursor for RQ and that it is the source of the critical amine group on the quinone ring of RQ . Thus in helminths the amine group is present from the start of the RQ biosynthetic pathway and is not added as a late step to UQ as is the case in bacteria . We also show that C . elegans requires RQ to survive under conditions where oxygen cannot be used as an electron acceptor for the ETC . This allowed us to establish a high throughput screening assay to identify compounds that block RQ synthesis or RQ use . This is the first study to show how RQ , an electron carrier that affects the life of over a billion humans , is made in helminths . This will help towards the development of a new class of drugs to treat these major human pathogens .
C . elegans is a non-parasitic helminth that is easily genetically tractable ( Brenner , 1974; reviewed in Jones et al . , 2005 ) and can be used for efficient drug screens ( Burns et al . , 2015 ) . We wanted to establish C . elegans as a model to dissect the pathway for RQ synthesis in helminths and as a system in which we could efficiently screen for drugs that block the synthesis of RQ or use of RQ . We note that there are no other standard model organisms where this is possible: yeasts , insects , fish , and vertebrates do not make or use RQ so C . elegans is the sole genetically tractable animal model for this work . Like other helminths , C . elegans has previously been shown to make RQ ( Takamiya et al . , 1999 ) . We wanted to confirm this and determine whether we could define a simple experimental method to drive C . elegans to carry out similar RQ-dependent anaerobic metabolism as that used by parasitic helminths to survive in their hosts . We extracted quinones as described in Materials and methods and as shown in Figure 2a and Supp Figure 1 , C . elegans makes both UQ and RQ when maintained in normoxic conditions . We can therefore use C . elegans to genetically dissect the pathway for RQ synthesis . Our next step was to establish a simple method to drive C . elegans to use RQ-dependent metabolism that would allow high throughput drug screens . Parasitic helminths use RQ-dependent metabolism under low oxygen conditions ( Rioux and Komuniecki , 1984; Saz and Lescure , 1969; Tielens et al . , 1992 ) and previous studies showed that C . elegans shows similar metabolic shifts when exposed to hypoxic conditions ( Butler et al . , 2012; Föll et al . , 1999; ) . If possible , however , we wanted to avoid the use of hypoxic chambers . While hypoxia chambers are highly accurate ways of controlling oxygen levels , they are also very expensive and cumbersome for large-scale drug screens . We therefore turned to chemical methods of inducing a hypoxic state . Potassium Cyanide ( KCN ) is a potent inhibitor of Complex IV ( Antonini et al . , 1971; Nicholls et al . , 1972 ) — KCN inhibits oxygen binding to Complex IV and KCN treatment thus mimics the effect of hypoxia on the ETC . We tested whether treatment with KCN could drive C . elegans to use anaerobic metabolism that is similar to the RQ-dependent metabolism used by STHs in their hosts . The classic hallmark of RQ-dependent anaerobic metabolism in helminths is the generation of high levels of succinate through the reversal of Complex II ( Figure 1c ) ( Butler et al . , 2012; Saz and Lescure , 1969; Tielens et al . , 1992 ) . If C . elegans can indeed use the same anaerobic metabolism as parasitic helminths , there should be a build-up of succinate following KCN treatment . Furthermore , this should be dependent on Complex I activity , since Complex I is the sole source of electrons that are carried by RQ to drive the fumarate reductase activity of Complex II ( Figure 1b ) . We found that when C . elegans are exposed to KCN , they build up high levels of succinate as expected and that inhibiting Complex I with rotenone prevents succinate build-up ( Figure 2b ) . We thus find that C . elegans makes RQ and that treatment of C . elegans with KCN causes them to switch to a metabolic state that resembles that of STHs in their host . C . elegans is thus an excellent model in which to dissect RQ synthesis and to screen for compounds that alter RQ-dependent metabolism . RQ and UQ are highly related molecules — the sole difference is the presence of an amine group on the quinone ring of RQ ( Figure 1d ) . A critical question for RQ synthesis is where this amine group comes from and how it is generated . The best-defined current model for RQ synthesis comes from experiments in the proteobacterium R . rubrum . At least in this prokaryote , RQ is thought to be made by a late addition of the critical amine group to an existing molecule of UQ ( Brajcich et al . , 2010 ) . UQ is thus an obligate precursor of RQ and RQ synthesis requires initial synthesis of UQ ( Figure 3a ) . While this may be the case for R . rubrum , this is not the case in F . hepatica ( Van Hellemond et al . , 1996 ) or C . elegans . The clk-1 ( qm30 ) strain has a loss-of-function mutation in the C . elegans COQ7 orthologue that is required for hydroxylation of 5-demethoxyubiquinone to 5-hydroxyubiquinone , a late step in UQ synthesis — there is no detectable UQ in clk-1 ( qm30 ) homozygous animals . However , a previous study showed that there does appear to be RQ in this strain ( Jonassen et al . , 2001 ) . If there is no UQ , but there is RQ , then RQ is not derived from UQ , at least in helminths . The two models for RQ synthesis thus differ fundamentally — in one UQ is an obligate precursor ( Brajcich et al . , 2010 ) , in the other it is not ( Jonassen et al . , 2001 ) . Since this is a fundamental result , we wanted to confirm this before trying to dissect the pathway of RQ synthesis . We thus extracted and analysed quinones from either wild-type worms or clk-1 ( qm30 ) homozygous animals . We find that while there is no detectable UQ in clk-1 ( qm30 ) mutants , there is abundant RQ and indeed we find that RQ levels are essentially unchanged ( Figure 3b ) . We thus confirm that UQ is not an obligate precursor for RQ in C . elegans . If RQ is not generated by addition of the key amine group to an existing UQ molecule where does the amine group on RQ come from ? One possibility is it is added not to UQ but to a UQ precursor such as demethoxyquinone ( DMQ ) — such UQ precursors would still be present in the clk-1 ( qm30 ) mutant strain ( Figure 3a for schematic; DMQ is abundant in clk-1 ( qm30 ) ( data not shown ) ) . While this could in principle be the case , it is unlikely because amination of an aromatic ring is highly thermodynamically unfavourable ( reviewed in Downing et al . , 1997 ) . We therefore investigated an alternative possibility — that the critical amine group of RQ is not added in a late step of RQ synthesis but instead is present from the outset . A key initial step in UQ synthesis is the addition by COQ-2 of a polyprenyl tail to a p-hydroxybenzoate ring ( PHB — also often called 4-hydroxybenzoate ( 4-HB ) ) ( Momose and Rudney , 1972; Trumpower et al . , 1974 ) . PHB has no amine group — however , S . cerevisiae COQ2 is known to be able to use a variety of similar compounds as substrates for prenylation such as para-aminobenzoic acid and vanillic acid ( reviewed in Pierrel , 2017 ) . Given the potential substrate flexibility of COQ-2 , we hypothesised that RQ synthesis might start not with PHB but with a related molecule that contains an amine group already on the ring ( Figure 3a for schematic ) . In particular , we noted that yeast COQ2 is tolerant of substituents at positions 5 and 6 of the PHB structure ( reviewed in Pierrel , 2017 ) suggesting that this might be feasible enzymatically . We considered different candidate molecules as amine-containing ring structures that might act as precursors for RQ and focussed on anthranilate and 3-hydroxyanthranilate ( 3 HA ) as likely sources of the amine-containing ring in RQ . Anthranilate and 3 HA are made from the amino acid tryptophan via the kynurenine pathway ( Heidelberger and Gullberg , 1948; Kotake , 1936 ) and kynu-1 encodes the C . elegans kynureninase that is required for the generation of anthranilate and 3 HA ( Babu , 1974; Bhat and Babu , 1980; van der Goot et al . , 2012 ) . We examined the quinones present in kynu-1 ( e1003 ) mutants that lack kynureninase and while UQ levels are normal in the kynu-1 ( e1003 ) mutant animals , there is no detectable RQ ( Figure 3c ) . This suggests that the amine group on RQ derives from anthranilate or 3 HA , or some closely related product of kynureninase . To determine whether the key precursor is anthranilate , 3 HA , or both , we examined quinones in the afmd-1 ( tm4547 ) and kmo-1 ( tm4529 ) mutant strains . These contain loss-of-function mutations in either afmd-1 , which encodes a kynurenine formamidase , or in kmo-1 which encodes a kynurenine 3-monoxygenase . Based on current models for the kynurenine pathway , afmd-1 ( tm4547 ) strain cannot make either anthranilate or 3 HA whereas the kmo-1 ( tm4529 ) can generate anthranilate but not 3 HA . We find that both strains have reduced RQ levels compared with N2 ( Figure 3d ) suggesting that the key precursor is 3 HA . Although the reduction in RQ levels in these two mutants is less severe than in the kynu-1 ( e1003 ) strain , we note that while kynu-1 encodes the sole kynureninase in the C . elegans genome , there are paralogues of both afmd-1 and kmo-1 ( afmd-2 and kmo-2 respectively ) which may be functionally redundant . Furthermore , several of these enzymes act on multiple related substrates , which may permit additional routes of 3 HA synthesis that are not shown . If 3 HA is indeed the key precursor for RQ , it should be a substrate for the polyprenyltransferase COQ-2 — COQ-2 prenylates PHB in the first step of UQ synthesis and our model suggests it should prenylate 3 HA as the first step in RQ synthesis . Consistent with this , we identify a m/z peak corresponding to the predicted mass of prenylated 3 HA ( Supp Figure 2 ) , but find no equivalent peak for prenylated anthranilate ( data not shown ) . Both the genetic data and mass spec data thus show that RQ synthesis requires the kynurenine pathway to generate 3 HA and that 3 HA is a substrate for COQ-2 . As an additional confirmation that the amine group on RQ ultimately derives from tryptophan we tested whether a tryptophan-derived aromatic amino group is being incorporated into RQ . We fed C . elegans 15N-labelled bacteria for three generations either in the presence or absence of 14N tryptophan . As shown in Figure 4a , the sole source of any 14N incorporated into RQ is the 14N tryptophan . As expected , the RQ detected in animals fed with 15N bacteria alone is approximately all 15N labelled ( Figure 4b ) . However , if 14N tryptophan was added while worms are feeding on the 15N-labelled bacteria , ~50% of the RQ observed was 14N RQ — the sole source for this 14N was the added tryptophan ( Figure 4b ) . We note that if we add 14N 3 HA instead of 14N Trp we also detect significant 14N RQ , but see no 14N RQ if 14N anthranilate is added , further confirming that 3 HA is the key RQ precursor ( Figure 4c ) . Taken together , our data suggest that RQ does not derive from UQ , and that 3 HA is the source of the amine group on the quinone ring of RQ . We propose that the pathway of RQ and UQ synthesis are largely the same — the key difference is the presence or absence of the amine group on the initial aromatic ring substrate for COQ-2 . As shown in Figure 2 , C . elegans shows similar changes in metabolism when it is treated with KCN as STHs undergo when they adapt to the hypoxic environment of their host . For example , the classic hallmark of this RQ-dependent anaerobic metabolism is the generation of succinate by the action of Complex II as a fumarate reductase and C . elegans shows high levels of succinate when treated with KCN ( Figure 3b ) . To confirm that this generation of succinate is indeed RQ-dependent in C . elegans , we examined whether RQ-deficient kynu-1 ( e1003 ) mutant animals could generate succinate when exposed to KCN . While wild-type worms generate high levels of succinate when Complex IV is inhibited with KCN , RQ-deficient kynu-1 ( e1003 ) mutant animals do not ( Figure 5 ) , confirming that the metabolic shift we see when we expose C . elegans to KCN is not simply similar to that of parasitic helminths in their hosts , it also requires RQ . We have thus established that treating C . elegans with KCN drives them into an alternative metabolic state where they use RQ to drive the same anaerobic metabolism used by STHs in their hosts . To be able to screen efficiently for drugs that affect RQ synthesis or RQ-dependent metabolism , however , we need a direct phenotypic readout for RQ-utilisation rather than a molecular readout ( such as succinate generation ) . When do C . elegans require RQ-dependent metabolism and what are the consequences if they have no RQ ? Since RQ-dependent metabolism is being used in the presence of KCN we compared the sensitivity of wild-type worms and kynu-1 ( e1003 ) mutants to KCN and the ability of wild-type worms and kynu-1 ( e1003 ) mutants to survive in KCN for long periods . We found no significant differences in acute KCN sensitivity of wild-type worms and kynu-1 ( e1003 ) mutants — both slow their movement in the presence of KCN , stop moving completely by ~90 min ( Figure 6a ) , and remain immobile from there on when maintained in KCN . However , there was a dramatic difference in their ability to survive extended periods in KCN . We exposed worms to KCN for different lengths of time and then removed animals from KCN and assayed their movement over the next 3 hr as they recover from KCN treatment . When wild-type worms are removed from KCN , they rapidly recover movement ( Figure 6b; recovery is also shown as a movie in Video 1 ) — they can do this even after 24 hr of KCN treatment ( Figure 6c ) . However , kynu-1 ( e1003 ) mutants show greatly reduced ability to survive extended KCN treatment — they do not survive exposure to KCN for 12 hr or more ( Figure 6c ) . We confirmed this using a second mutant strain that also carries a loss-of-function mutation in kynu-1 , the kynu-1 ( tm4924 ) strain and find essentially identical results ( data not shown ) . RQ-dependent anaerobic metabolism thus allows C . elegans to survive extended periods where it cannot use oxygen as the terminal electron acceptor of the ETC . This provides a simple high throughput assay for drugs that specifically affect RQ-dependent metabolism: drugs that block RQ synthesis or the activity of RQ-dependent pathways should abolish the ability of worms to survive >12 hr in KCN . To test this , we used the compound wact-11 — this is a Complex II inhibitor that binds to the quinone-binding pocket of Complex II ( Burns et al . , 2015 ) . wact-11 is highly related to the anthelmintic flutolanil ( Burns et al . , 2015 ) and is highly selective for helminth Complex II ( Burns et al . , 2015 ) . Complex II is critical for RQ-dependent anaerobic metabolism where it acts as a RQ-dependent fumarate reductase — inhibitors of Complex II would thus be expected to prevent survival in KCN . When worms are treated with wact-11 alone , we see no impact on worm movement over the time course of our assay . However , when worms are treated with both wact-11 and KCN , they can no longer survive long-term KCN exposure ( Figure 6d ) indicating that this assay can allow discovery of drugs that block RQ-dependent anaerobic metabolism . Our assay should thus allow efficient screens for drugs that inhibit RQ synthesis or RQ utilization in vivo in a helminth under conditions where they require RQ , the first time this has been possible . The assay is image-based and quantitative and can be extremely high throughput when using rapid imaging platforms like the Phenalysys Parallux 2 ( used to make the movies of recovery in Video 1 ) which can measure the effect of 96 drugs on worm movement in under 5s . We note that not all drugs that kill C . elegans when in the presence of KCN ( i . e . that would be hits in this assay ) will affect RQ synthesis or RQ use and secondary screens will be required to further stratify the hits . Nonetheless , inhibitors of RQ synthesis or use can be discovered using this assay and we anticipate that this will yield novel compounds that may be effective anthelmintics . Finally , we took advantage of a set of mutant worm strains that are resistant to wact-11 treatment . Mutations that result in resistance to wact-11 treatment cluster in the quinone binding pocket of Complex II ( Burns et al . , 2015 ) and we reasoned that some of these might not only prevent the binding of wact-11 but might fortuitously also disrupt the binding of RQ and thus affect the ability of worms to survive extended exposure to KCN . We tested a number of point mutants that affect wact-11 sensitivity and found that most mutants appear similar to wild-type worms in their ability to survive long term KCN exposure ( Figure 7a and data not shown ) . However , we found that the G71E mutation results in worms that are unable to survive extended KCN exposure — the G71E animals thus resemble kynu-1 ( e1003 ) mutants . We note that this mutation sits right above the modelled binding site for the rhodoquinone ring , whereas a neighbouring mutation that sits two turns of an alpha-helix further away ( C78Y ) has no effect . We thus suggest that the G71E mutation affects the ability of C . elegans to bind RQ into the quinone binding pocket of Complex II and thus to drive RQ-dependent fumarate reduction as part of its RQ-dependent anaerobic metabolism .
RQ was first identified over 50 years ago ( Moore and Folkers , 1965 ) . It is absolutely required for the anaerobic metabolism used by parasitic helminths to survive in the hypoxic environment of the host gut where they can thrive for many months . The single amine group that differs between RQ and UQ is crucial for this — it allows RQ to carry electrons of the right electropotential to drive quinone-coupled dehydrogenases ( QDHs ) in reverse , acting as reductases ( Fioravanti and Kim , 1988; Sato et al . , 1972 ) . In aerobic conditions , QDHs carry electrons from a diverse set of electron donors and transfer them onto UQ and hence into the ETC; under anaerobic conditions , RQ carries electrons to the QDHs which then reduce a diverse set of electron sinks , providing an exit point for electrons from the ETC . The single amine group on the quinone ring of RQ allows parasites to carry out this unusual anaerobic metabolism and thus it affects the lives of over a billion humans . Despite the importance of RQ for human health , its synthesis has been elusive and no anthelmintics have been identified that affect RQ synthesis . Here , we used C . elegans genetics to identify the source of the key amine group on RQ , and to establish a pipeline for screening for new compounds that alter the ability of worms to make and use RQ . The critical question in RQ synthesis is where the amine group on the quinone ring comes from and how it is added . Previous studies from bacteria suggested that RQ synthesis uses UQ as a precursor ( Brajcich et al . , 2010 ) and that the amine group is added at a late stage in RQ synthesis . At least in bacteria , this seems to be the case — in R . rubrum , the gene rquA that is required for RQ synthesis encodes a methyltransferase that is related to the quinone methyltransferases UbiG/COQ3 and UbiE/COQ5 that act in UQ synthesis ( Lonjers et al . , 2012; Stairs et al . , 2018 ) . Very recently , RquA has been shown to be able to catalyse the conversion of UQ to RQ ( Buceta et al . , 2019 ) — in bacteria , RQ synthesis thus requires UQ synthesis and the key amine group is a late stage addition . Here , however , we show that the mechanism of RQ synthesis in C . elegans is different . In C . elegans ( and likely in helminths in general ) RQ synthesis does not require UQ as a precursor and , crucially , the critical amine group on the quinone ring of RQ is not added at a late stage in the synthesis of RQ but is present from the initial steps of RQ synthesis . We show that the amine group likely comes from 3-hydroxyanthranilate ( 3 HA ) , which is generated by degradation of tryptophan via the kynurenine pathway . We also demonstrate that RQ synthesis requires the kynurenine pathway . We note that during revision of the manuscript , a second group also found that RQ synthesis requires the kynurenine pathway ( Buceta et al . , 2019 ) . Thus , for the first time since its discovery in the early 1960s , we now have a key insight into how RQ is made in helminths and this has several implications for the search for novel anthelmintics that might affect RQ synthesis . First , we do not believe that there are completely separate dedicated pathways for UQ and for RQ synthesis in helminths . Instead , we suggest that UQ and RQ have a largely shared synthesis pathway . A key difference in RQ and UQ synthesis is the use of different initial substrates for the polyprenyltransferase COQ-2 . If PHB is used , the product will be UQ; if 3 HA ( or possibly a related product of tryptophan metabolism ) is used , the product will be RQ . This ‘different precursor , shared pathway’ model for RQ and UQ synthesis stands in contrast to the pathways used by bacteria to synthesise two other quinones , UQ and menaquinone ( MK ) . Facultative anaerobic bacteria including non-pathogenic E . coli as well as major human pathogens like M . tuberculosis make two quinones: UQ which is used as an electron carrier under aerobic conditions , and menaquinone ( MK ) ( reviewed in Meganathan and Kwon , 2009 ) , which acts as a carrier under anaerobic conditions . The synthesis pathways of UQ and MK are completely distinct and the genes involved are distinct ( reviewed in Meganathan and Kwon , 2009 ) . This separation of MK and UQ synthesis pathways has allowed the development of a number of promising compounds that act as specific inhibitors of MK synthesis ( Kurosu et al . , 2007 and Debnath et al . , 2012; reviewed in Boersch et al . , 2018 ) . We suggest that there may be no analogous inhibitors for ‘the RQ synthesis pathway’ in helminths since there does not appear to be a dedicated RQ pathway analogous to the dedicated MK synthesis pathway . This also raises intriguing questions of regulation of quinone content in helminths . While in bacteria the switch from UQ ( aerobic ) to MK ( anaerobic ) occurs through the use of two entirely separate biosynthetic pathways , in helminths the switch from UQ ( aerobic ) to RQ ( anaerobic ) must occur via the entry of different substrates into a largely shared pathway . We currently have little insight into how that switch occurs or how it is regulated . Second , the pathway we identify for RQ synthesis suggests novel targets for anthelmintics . The finding that the kynurenine pathway is the source of the key precursors for RQ synthesis suggests naively that helminth-specific inhibitors of the kynurenine pathway might act as potent anthelmintics . However , the human gut is likely to be a source of anthranilate and 3 HA from host metabolism or from the microbiome and inhibiting production of these molecules in the helminth itself might thus prove ineffective . A more likely target is COQ-2 , the enzyme that prenylates the amine-containing 3 HA ring as the first step in RQ synthesis . While host and parasite COQ-2 are orthologous enzymes , they have different substrate specificity: the host only makes UQ and not RQ , whereas the parasite COQ-2 must be able to use not only PHB for UQ synthesis but also amine-containing precursors like 3 HA for RQ synthesis . The ability of helminth COQ-2 to use amine-containing substrates efficiently thus opens up the possibility of helminth-specific COQ-2 inhibitors , a potential avenue for new anthelmintics . We believe that the in vivo assay we describe here may help identify such drugs . Our study also raises key new questions . One of the most intriguing to us is why RQ synthesis is so rare amongst animals . To date , only three groups of animals are known to make RQ: molluscs , annelids , and helminths ( Allen , 1973; Fioravanti and Kim , 1988; Sato and Ozawa , 1969; Takamiya et al . , 1999; Van Hellemond et al . , 1995 ) . If RQ synthesis and UQ synthesis largely share a common pathway , why doesn’t every animal that makes UQ also make RQ ? One possibility is that while some of the pathway for UQ and RQ synthesis is shared , RQ synthesis requires additional components that might only be present in RQ-synthesising species . For example , we note that while the UQ precursor PHB has a hydroxyl group at the four position , the RQ precursor 3 HA does not and this must be added by some as yet unknown enzyme . The other possibility is that key enzymes in the UQ and RQ synthesis pathway may have altered specificity in species that make RQ to allow them to use substrates containing an amine group for RQ synthesis as well as non-aminated substrates for UQ synthesis . Careful phylogenetic analysis may identify subtle sequence changes that could allow helminth enzymes to use amine-containing RQ precursors . Such molecular signatures that mark out RQ-utilising species have been found in proteins that bind both UQ and RQ — for example there are helminth-specific residues that lie around the quinone binding site in Complex II ( Burns et al . , 2015 ) . It is clear that there is no simple answer yet for the finding that RQ is made by so few animals but that this will likely emerge as more of the RQ synthesis pathway is uncovered . Finally , we note that several steps in the kynurenine pathway and in the ubiquinone synthesis pathway are catalysed by either monooxygenases or dioxygenases that require oxygen . These include TDO-2 ( Hayaishi et al . , 1957 ) and KMO-1 ( Detmer and Massey , 1985; Entsch et al . , 1976 ) in the kynurenine pathway and COQ-6 ( Ozeir et al . , 2015 ) and CLK-1 ( Marbois and Clarke , 1996 ) in the UQ synthesis pathway . RQ synthesis thus appears to require the availability of oxygen for these enzymes , an unexpected result since RQ is preferentially required in anaerobic conditions and is the predominant quinone in helminths living under anaerobic conditions . How might these oxygen-requiring steps be carried out for RQ synthesis ? It is possible that the helminth enzymes have evolved so that they can still operate under low oxygen conditions . Other oxygen-using proteins have evolved extremely high oxygen affinity in helminths — for example Ascaris haem is octameric and binds oxygen with ~25 , 000 times greater affinity than human haem ( Minning et al . , 1999 ) . Alternatively , these same enzymatic steps might be carried out by other enzymes in lower oxygen conditions . In E . coli , for example , in aerobic conditions UbiB and UbiF carry out the same hydroxylation modifications to the quinone ring as C . elegans COQ-6 and CLK-1 and mutation of either gene results in a lack of mature UQ in these bacteria ( reviewed in Meganathan and Kwon , 2009 ) . However , under anaerobic conditions , ubiB and ubiF mutants make normal levels of UQ suggesting that other enzymes carry out these reactions in low oxygen conditions ( Alexander and Young , 1978 ) . It is possible that there is an analogous set of enzymes that are required for RQ synthesis in low oxygen conditions — these would carry out similar reactions to COQ-6 and CLK-1 but without the requirement for oxygen . If they exist , they remain to be discovered . There is thus still much to be discovered about the regulation and the precise pathway of RQ synthesis in helminths . The results presented here provide a firm starting point and the assay we describe for drugs that affect RQ-dependent metabolism may lead to the discovery and development of a new class of anthelmintic drugs . Since resistance to known classes of anthelmintics is widespread among livestock parasites like H . contortus , C . oncophora and A . suum and is rising in human populations ( reviewed in Sangster et al . , 2018 ) , this will prove critical in the control and treatment of these major pathogens .
In addition to the traditional laboratory strain N2 , here we include work using strains clk-1 ( qm30 ) , kynu-1 ( e1003 ) , sdhc-1 ( tr357 ) , and sdhc-1 ( tr423 ) . The two sdhc-1 strains were provided by Dr . Peter Roy and all other strains were provided by the Caenorhabditis Genetics Centre or by the Mitani group of the National Bioresources Project . All worms were maintained on NGM agar plates seeded with E . coli OP50 as described elsewhere ( Stiernagle , 2006 ) and maintained at 20°C . Escherichia coli ( MG1655 ) was grown overnight at 37°C in M9 media prepared using 1 g/L 15N ammonium chloride ( Cambridge Isotopes ) as the nitrogen source . Bacteria were heat killed at 65°C for 15 min . The heated culture was used to seed NGM agar plates . 500 μL of 50 mg/ml tryptophan in water was spread on each 10 cm plate . Ten L4 nematodes were placed on each plate to lay eggs overnight at 20°C . Adult worms were removed following the egg laying period . After 5 days at 20°C the nematodes were collected and frozen at −80°C . Nematode samples were thawed and lysed via sonication . Quinone extraction solvent containing a 2:1 ratio of chloroform and methanol ( Thermo Fisher Optima LC-MS grade ) respectively was added to the samples . The organic phase of the sample was collected and then dried using nitrogen gas . Samples were resuspended in a 60:40 acetonitrile and isopropanol solution prior to analysis using APCI LC-MS . Quinones were analyzed by reverse phase chromatography on an Eclipse Plus C-18 RRHD column , 2 . 1 mM x 50 mm with 1 . 8 um packing operated in a thermostatted column compartment held at 70°C . Buffer A was 50% MeCN in water , Buffer B was 100% acetone with 0 . 01% formic acid . Starting conditions were 0 . 25 mL/min at 50% B . Gradient was 1 min hold , followed by increase 100% B at 5 min , hold 100% B until 7 min , then return to 50% B at 7 . 1 min and hold until 10 min . Samples were introduced from a HTC pal by injection of 5 μL sample into a 2 μL loop . Wash one was acetonitrile and wash two was isopropanol . Samples were ionised using a Multimode ionisation source ( Agilent ) operated in APCI mode , gas temp 350°C , vaporizer temp 350°C , drying gas 5 L/min , nebulizer 60 PSI , capillary voltage 4000 V , corona current 4 μA , skimmer voltage 70 V , octupole 1 RF 400 V . Samples were analyzed on a 6230 TOF , a 6545 Q-TOF , or a 6490 QQQ as indicated . Fragmentor voltage for TOF/QTOF analysis was 200 V . For QQQ analysis , ubiquinone nine was monitored by MRM of 795 . 6/197 . 3 at CID of 52 V; rhodoquinone-9 was monitored at 780 . 6/192 . 1 at CID of 52 V . All image-based experiments were conducted on L1 animals which were collected from mixed-stage plates and isolated using a 96 well 11 µM Multiscreen Nylon Mesh filter plate ( Millipore: S5EJ008M04 ) as described previously ( Spensley et al . , 2018 ) to a final concentration of ~100 animals per well . They were then incubated in a final concentration of 200 μM potassium cyanide for varying amounts of time . There are two key assays: the acute assay and the recovery assay . The acute assay monitors worm movement immediately following exposure to KCN every 5 min for a total of 3 hr . The recovery assays involve a KCN incubation of 3 , 6 , 9 , 12 , or 18 hr after which the KCN is diluted 6-fold with M9 buffer . Immediately after dilution , worm movement is monitored every 10 min for 3 hr . In both assays , worm movement is quantified using an image-based system as previously described ( Spensley et al . , 2018 ) . All data were normalised by the fractional mobility score of the M9-only control wells per strain per time point . For movies shown in Video 1 , N2 and kynu-1 ( tm4924 ) L1 worms were incubated in 200 μM KCN for 18 hr as per the recovery assay . Images were taken and processed using a Phenalysys Parallux 2 . Raw images were taken every second and segmented into a set of images per individual well . These are then processed by FFmpeg to generate a Quicktime movie . The accompanying graph plots the moving average of total-well FMS scores using a box filter with n = 60 , with every datapoint representing a minute of the assay . Solutions of potassium cyanide ( Sigma 60178–25G ) were made fresh prior to each experiment in phosphate buffered saline ( PBS ) and then diluted to a 5 mM stock solution in M9 buffer . 2X working concentrations were then prepared with M9 and the KCN stock solution . Wact-11 ( Chembridge ID 6222549 ) was kept frozen as a 100 mM stock in DMSO . Diluted wact-11 stocks were made in DMSO ( BioShop DMS666 ) to a concentration of 3 . 75 mM and kept frozen until day of use . 10X working concentrations were made with wact-11 stock , M9 buffer , and DMSO . All experiments were prepared to contain 0 . 8% v/v DMSO to control for any confounding effects of drug solvent . Assays were assembled in flat-bottomed polystyrene 96-well plates ( Corning 3997 ) to a total volume of 100 μL and 40 μL for the acute and recovery assays , respectively . Apart from assays including wact-11 which constituted half KCN solution , 10% wact-11 solution , and 40% worms in buffer , all other assays were comprised of equal parts worms in buffer and KCN solution . Worms were collected and isolated as described above . A final concentration of 7 . 5 L1s/10 µL was treated with final concentrations of 12 . 5 µL rotenone ( Sigma R8875 ) in 0 . 8% DMSO and with 0 . 8% DMSO alone and with 100 µM KCN in M9 or with M9 alone , 20 mL altogether in 40 mL plastic containers ( Blender Bottle 600271 ) and were on a shaker for 1 hr at room temperature . After 1 hr , samples were poured over 0 . 2 µM Nylaflo nylon filter membranes ( PALL 66604 ) over vacuum and once the supernatant had run through , the filter paper was placed in prepared 1 . 2 mL of 8:1:1 extraction solvent ( MeOH , HPLC Grade ( SA 34860 ) ; H2O , HPLC Grade ( Caledon 8801-7-40 ) ; CHCl3 , HPLC Grade ( SA 650498 ) ) in a 1 . 5 mL microfuge tube on dry ice . Tubes were inverted five times then vortexed . Samples were switched between −80°C and −20°C three times . Filters were then removed , and the tubes spun at 13 , 200 rpm at 4°C for 30 min . 1 mL of supernatant was transferred to a new tube and dried under dry N2 with <0 . 02% O2 at 5 PSI for 8 hr . Each sample was reconstituted with 30 µL HPLC grade water as was prepared labelled yeast reference . Samples and reference were spun at 13 , 200 rpm at 4°C for 5 min . 10 µL sample and reference were placed in an LC-MS sample vial ( Agilent 5190–2243 , cap is Agilent 5185–5820 ) and were fast spun at 1 , 000 rpm at 4°C . N2 and kynu-1 ( e1003 ) worms were washed and filtered as previously described . They were then placed in 1 . 5 mL microfuge tubes at 1 . 5 mL for a final concentration of 45 L1s , 300 µM KCN in M9 or M9 alone and were placed on a rotator for 6 hr at room temperature . After 6 hr , the samples were extracted and prepared as previously described . Succinate was extracted from an ion-paired reverse phase method IPRP method LC-MS run at an mzCenter of 117 . 0193 and a retention time of 690 s in the case of the kynu-1 experiment and from an Acid method LC-MS run at an mzCenter of 117 . 0193 and a retention time of 141 s in the case of the rotenone experiment . Both sets of samples were normalised to a labelled yeast reference . In the case of the kynu-1 experiment , they were further normalised to the median of all the extracted peaks for each sample . In both cases they were ultimately normalised to the mean unlabelled N2 sample treated with buffer or 0 . 8% DMSO . Plots were generated using {plotPeak} . Mitochondrial rhodoquinol-fumarate reductase from A . suum bound with rhodoquinone-2 ( PDB: 3VR8 ) was displayed on Chimera ( Pettersen et al . , 2004 ) and the C . elegans sequence was threaded by homology using Modeller ( Sali and Blundell , 1993; Webb and Sali , 2016 ) and the MSA with 15 iterations . | Parasitic worms infect more than a billion people worldwide , using a range of tricks to survive inside the human body . Some species can live for weeks inside the gut , a place with almost no oxygen . Yet exactly how they manage this is remains unclear . Scientists know that parasitic worms have an unusual way of making chemical energy when oxygen levels drop . Like human cells , worm cells use a series of molecular complexes called the electron transport chain . As electrons pass along the chain , they drive the production of chemical energy . Normally , oxygen sits at the end of the chain to receive the electrons . But , when there is no oxygen , almost all animals stop using the electron transport chain . A few animals can continue to use it by using other molecules to receive the final electrons instead of oxygen . To do that , they need a special electron carrier and , in worms , this electron carrier is rhodoquinone . Human cells do not use rhodoquinone , making it a prime target for drug design . If a drug could block rhodoquinone production , it might be able to stop worms surviving in the human intestines without harming the patient’s own cells . Yet , even though the scientific community has known about rhodoquinone for more than 50 years , it remains unclear how worms make this molecule . To find out , Del Borrello et al . examined the laboratory worm Caenorhabditis elegans . This worm is not a parasite , but it does make rhodoquinone . Del Borrello et al . developed a new way to study rhodoquinone production by blocking the normal route of the electron transport chain with cyanide . This causes the worms to switch to using rhodoquinone and is cheaper than raising the worms in low oxygen , making it easier to conduct high-throughput screening . A combination of chemistry and information from other species made it possible to identify candidate genes responsible for the production of rhodoquinone . Worms with faults in these genes revealed the key building blocks of rhodoquinone , and the early steps in its production . Removing any one of the genes made it harder for the worms to survive without oxygen . Although there are already effective drugs that kill parasitic worms , resistance is growing . A better understanding of rhodoquinone could lead to a new class of drugs to help control this major problem in global health . A drug that blocks any one of the production steps of rhodoquinone might be a future candidate for a new anti-parasitic worm therapy . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"biochemistry",
"and",
"chemical",
"biology",
"microbiology",
"and",
"infectious",
"disease"
] | 2019 | Rhodoquinone biosynthesis in C. elegans requires precursors generated by the kynurenine pathway |
Hematopoietic stem cells ( HSCs ) must ensure adequate blood cell production following distinct external stressors . A comprehensive understanding of in vivo heterogeneity and specificity of HSC responses to external stimuli is currently lacking . We performed single-cell RNA sequencing ( scRNA-Seq ) on functionally validated mouse HSCs and LSK ( Lin- , c-Kit+ , Sca1+ ) progenitors after in vivo pharmacological perturbation of niche signals interferon , granulocyte colony-stimulating factor ( G-CSF ) , and prostaglandin . We identified six HSC states that are characterized by enrichment but not exclusive expression of marker genes . External signals induced rapid transitions between HSC states but transcriptional response varied both between external stimulants and within the HSC population for a given perturbation . In contrast to LSK progenitors , HSCs were characterized by a greater link between molecular signatures at baseline and in response to external stressors . Chromatin analysis of unperturbed HSCs and LSKs by scATAC-Seq suggested some HSC-specific , cell intrinsic predispositions to niche signals . We compiled a comprehensive resource of HSC- and LSK progenitor-specific chromatin and transcriptional features that represent determinants of signal receptiveness and regenerative potential during stress hematopoiesis .
Stem cell therapy holds promises for numerous indications , including blood diseases , autoimmune diseases , neurodegeneration , and cancer ( Blau and Daley , 2019 ) . Despite being used in the clinic for over 30 years , hematopoietic stem cell ( HSC ) transplants remain a highly risky procedure . To better understand HSC regeneration , recent efforts have used single-cell RNA sequencing ( scRNA-Seq ) to discover novel markers to further enrich for functional HSCs ( Chen et al . , 2016; Cabezas-Wallscheid et al . , 2017; Wilson et al . , 2015; Rodriguez-Fraticelli et al . , 2020 ) . Yet , no consensus exists on the optimal marker combination to obtain the most purified HSCs in part because extensive functional heterogeneity within HSCs makes experimental evaluation challenging ( Haas et al . , 2018 ) . Both intrinsic and extrinsic factors have been implicated in regulating HSC function ( Zon , 2008; Morrison et al . , 1996 ) . The stem cell niche forms an important extrinsic regulator of HSCs as it anchors stem cells and maintains the balance between self-renewal and differentiation ( Morrison and Spradling , 2008; Morrison and Scadden , 2014 ) . Release of soluble signals from the niche such as interferons , prostaglandins , and growth factors , including stem cell factor ( SCF ) and G-CSF , has been shown to influence HSC function during homeostasis and upon injury ( Pinho and Frenette , 2019; Pietras et al . , 2016; Zhao et al . , 2014; Morales-Mantilla and King , 2018 ) . While known to be affected by a wide variety of extracellular signals , little is known about the heterogeneity and specificity of HSC responses to these external stimuli , nor is it understood how differential responses relate to functional diversity of HSCs . HSCs are also regulated cell intrinsically ( Zon , 2008; Morrison et al . , 1996 ) . Chromatin state is a crucial determinant of cell identity and behavior ( Klemm et al . , 2019 ) . Hematopoietic differentiation is a prime example of how cell fate changes associate with massive remodeling of the epigenetic landscape ( Avgustinova and Benitah , 2016 ) . Despite the current knowledge on regulators of HSC fate , few studies have assessed chromatin states in purified , in vivo-derived HSC populations ( Yu et al . , 2017; Lara-Astiaso et al . , 2014 ) due to technical limitations such as cell numbers . Recent advancements in single-cell chromatin accessibility sequencing ( scATAC-Seq ) provides a methodological framework for studying the diversity and uniqueness of HSC chromatin features at homeostasis and upon external stimulation ( Buenrostro et al . , 2018; Lareau et al . , 2019 ) . Here , we performed comprehensive scRNA-Seq and scATAC-Seq profiling on functionally validated mouse HSCs and examined in vivo transcriptional responses to pharmacological stimulation , mimicking signals from the stem cell niche . To encompass a wide variety of different transcriptional responses , we evaluated three different signaling pathways: an inflammatory pathway through stimulation or inhibition of prostaglandins by 16 , 16-dimethyl prostaglandin E2 ( dmPGE2 ) and indomethacin , a host-defense immune signaling pathway mediated by activating of TLR and interferon signaling with poly ( I:C ) , and a cellular mobilization pathway stimulated by the growth factor G-CSF . We found that unperturbed HSCs exist in fluent transcriptional states with different levels of marker gene enrichment . External stimulants can alter the cell distribution between HSC states to varying degrees depending on the stimulant as well as induce specific changes within cell states . Comparison of HSCs to multipotent LSK ( Lin- , c-Kit+ , Sca1+ ) progenitors allowed us to determine the specificity of transcriptional responses in HSCs . Finally , analysis of native HSC chromatin states revealed cell intrinsic heterogeneity that may prime HSC subpopulations for particular transcriptional responses following exposure to certain signals . The data is provided as a resource to the broader research community via an easily accessible web interactive application ( https://mouse-hsc . cells . ucsc . edu ) . This work provides a comprehensive description of the in vivo single-cell transcriptomic and epigenetic landscape of HSCs and multipotent LSK progenitors in response to common external stressors .
To investigate transcriptional responses to external signals , we profiled HSCs and multipotent progenitors ( MPPs ) after four distinct in vivo pharmacological perturbations with doses matching previous studies ( Figure 1A , see Materials and methods ) . Male and female mice were treated with one of three activators dmPGE2 , poly ( I:C ) , or G-CSF for 2 hr or administered the Cox1/2 inhibitor indomethacin ( ‘Indo’ ) for 1 week to deplete endogenous prostaglandins ( see Materials and methods ) . We chose a 2 hr treatment window for the extrinsic activators as we aimed to assess the immediate , direct effects of the external stimulants on HSCs and MPPs . After the respective drug treatments , HSC and MPP populations comprising the entire LSK compartment were isolated via fluorescence-activated cell sorting ( FACS ) ( Figure 1—figure supplement 1A ) . Through a limiting dilution transplantation assay ( LDTA ) and extreme limiting dilution assay ( ELDA ) analysis ( Hu and Smyth , 2009 ) , we determined HSC purity to be 1 in 8 ( Figure 1—figure supplement 1B-D ) . The LDTA confirmed that our isolation and purification procedure allowed for the profiling of functional , highly purified HSCs . Phenotypic marker composition within LSK cells remained largely consistent between different stimulations ( Figure 1—figure supplement 1E ) . An exception was the reduction of cells within the HSC compartment following dmPGE2 treatment , decreasing from 1 . 9% in control to 0 . 85% of LSK cells ( p-value = 6 . 4*10–4 , by differential proportion analysis [DPA]; Farbehi et al . , 2019 ) . To account for a potential phenotypic shift in HSC surface marker expression due to CD34 externalization , which would move functional HSCs to the MPP1 population , we compared the contribution of the later by scRNA-Seq-defined ‘stem cell state’ in HSCs and MPP1s . We found no increase in the ‘stem cell’ population in dmPGE2-treated MPP1s , compared to the control ( Figure 1—figure supplement 2H ) . After cell sorting , we subjected a total of 46 , 344 cells to scRNA-Seq using the 10× Genomics platform ( see Materials and methods ) . We obtained an average of 37 , 121 ( SD = 14 , 308 ) reads per cell and 2994 ( SD = 480 ) genes per cell ( Supplementary file 1 ) , indicative of a rich dataset that contained functionally validated HSCs . To determine how external stimulants affect specifically HSCs in vivo , we first analyzed a combination of highly purified control and treated HSCs but not MPPs cells ( Figure 1A ) . We applied a standard scRNA-Seq pipeline to filter and normalize UMI reads ( see Materials and methods ) . Separate analysis of male and female HSCs revealed minimal sexual dimorphism during both steady state and following perturbation with external stimulants ( Figure 1—figure supplement 3 , Supplementary files 2 and 3 ) . We therefore regressed out any sex-specific effects and controlled for other batch-specific confounders in further downstream analyses ( see Materials and methods ) . In the aggregated dataset , we detected a total of six HSC clusters ( Figure 1B ) . To ensure optimal choice of clustering hyperparameters , we used a data-driven approach ( Silhouette coefficient and Davies–Bouldin index ) that was validated by comparison of two independent biological scRNA-Seq replicates of control HSCs sorted from different mouse strains ( see Materials and methods , Figure 1—figure supplement 2A-D , Supplementary file 4 ) . The absence of clear separation into highly distinct clusters in uniform manifold approximation and projection ( UMAP ) space ( Figure 1B ) , together with fact that most marker genes were not exclusively expressed but rather enriched in a given cluster ( Figure 1—figure supplement 2E ) , suggests that the HSC clusters represent transcriptional states with continuous transitions as opposed to discrete subtypes of HSCs . We calculated a transcriptional score by combining the top enriched genes for each cluster ( Figure 1C , see Materials and methods ) to further illustrate the observation of gradual changes in transcriptional state within the HSC population . While transcriptional scores were most enriched in their respective clusters , expression dropped before and extended beyond cluster borders ( Figure 1B and C ) . Reactome and gene ontology ( GO ) term pathway enrichment analysis , comparison to previous studies of functionally characterized HSCs ( Materials and methods , Supplementary files 5 and 6 ) and manual curation of enriched genes ( Figure 1D , Figure 1—figure supplement 2E , Supplementary file 4 ) allowed to assign labels to each HSC cluster or state . Three HSC clusters made up 98% of control HSCs ( Figure 1F ) while the remaining 2% split into a ‘cell cycle’ cluster marked by genes such as Ki67 and an ‘Interferon’ cluster characterized by the expression of interferon-response genes Iigp1 , Isg15 , Ifit1 , and Oasl2 ( each 1% , Figure 1D and F ) . A prominent HSC subpopulation was defined by various immediate early genes ( IEGs ) including Nr4a1 , Ier2 , and Fos ( Figure 1D and Figure 1—figure supplement 2E ) and we therefore named this cluster ‘Activated’ . We eliminated the possibility that the ‘Activated’ cluster arose due to an unspecific artifact of the cell isolation procedure since LSKs did not have an ‘Activated’ cluster and the proportion of Nr4a1 expressing cells was much smaller ( Figure 3B and Figure 3—figure supplement 1B ) . HSCs have been tightly associated with decreased cell cycle activity ( Foudi et al . , 2009; Wilson et al . , 2008; Qiu et al . , 2014 ) . The cluster adjacent to the ‘Activated’ state was termed ‘quiescent’ because cells showed enrichment in expression of marker genes that have previously been linked to the most potent and quiescent HSCs ( Figure 1D , Figure 1—figure supplement 2F , Supplementary file 6; Cabezas-Wallscheid et al . , 2017; Chen et al . , 2016; Wilson et al . , 2015; Acar et al . , 2015; Gazit et al . , 2014; Balazs et al . , 2006; Komorowska et al . , 2017; Schneider et al . , 2016; Jeong et al . , 2009 ) . Furthermore , ‘quiescent’ HSCs did not express IEGs and expressed low levels of the ‘cell cycle’ score ( Figure 1B and C ) . The ‘metabolism’ cluster comprised the most metabolically active HSCs as evidenced by enrichment of transcripts involved in translation initiation ( Eif5a , Eif4a1 ) , nucleotide metabolism ( Nme1 , Dctpp1 ) , ribosome assembly ( Ncl , Nop56 , Nop10 , Npm1 ) and protein chaperones ( Hsp90 , Hsp60 ) ( Figure 1B and D , Supplementary file 4 ) . In conclusion , baseline HSCs were defined by three main transcriptional states , ‘Quiescent’ , ‘Activated’ , and ‘Metabolism’ ( Figure 1F ) with few HSCs residing in the ‘Interferon’ or ‘Cell cycle’ state . Transcriptional scores visualized that these HSC states were not exclusive and that HSC transcriptional state could be rather described by a combination of continuous gradients of marker genes . Therefore , subsequent analyses via discrete clusters provided an analytical tool to compare changes in transcriptional state as opposed to an exclusive assignment of cell identities . To determine how external stimulants affect transcriptional identity of HSCs , we evaluated changes in cell distribution between clusters ( Figure 1E and F ) as well as differentially expressed genes ( DEGs ) within each cluster using ‘model-based analysis of single-cell transcriptomics’ or MAST ( see Materials and methods; Finak et al . , 2015 , Supplementary file 7 ) . We further examined the relationship of genes that define each HSC cluster and genes perturbed by each external stimulant ( Figure 1D and H ) . A unified heatmap shows all HSC clusters for every perturbation ( rows ) and the averaged gene expression within these clusters for four cluster- or treatment-representative genes ( columns , up-only , Figure 1D , Supplementary files 4 and 8 , full heatmap in Figure 1—figure supplement 2I ) . To further identify distinct patterns of gene regulation in HSC clusters and visualize both up- and downregulated genes , we generated separate heatmaps for each individual perturbation ( Figure 1—figure supplement 4 , Supplementary file 8 ) . DmPGE2 and poly ( I:C ) stimulated genes showed enrichment for previously described signatures with the same stimulants ( Supplementary files 5 and 6 ) . G-CSF induced selected genes such as Myb and Spi1 ( Figure 1D ) and downregulated niche adhesion receptors ckit and Cd9 ( Figure 1—figure supplement 4B , purple arrows ) consistent with the growth factor’s role in myeloid differentiation ( Metcalf and Nicola , 1983 ) and mobilization ( Leung et al . , 2011; Bendall and Bradstock , 2014 ) , respectively . However , our G-CSF-induced gene set did not show any significant enrichment ( Supplementary files 5 and 6 ) with various previously reported G-CSF signatures ( Schuettpelz et al . , 2014; Pedersen et al . , 2016; Giladi et al . , 2018; Mervosh et al . , 2018 ) likely due to different timing of G-CSF treatment . Indomethacin only led to subtle changes in gene expression ( Figure 1—figure supplement 4D , Supplementary file 8 ) and cell distribution between HSC clusters remained unaffected ( Figure 1F ) . Both dmPGE2 and poly ( I:C ) caused a significant change in HSC cluster distribution which indicated a loss of the original transcriptional identity of some HSCs ( Figure 1D–F ) . In vivo treatment with dmPGE2 gave rise to a novel cluster that contained 55% of dmPGE2-treated HSCs ( Figure 1F ) and which was itself only composed of dmPGE2-treated cells ( Figure 1G ) . We called this cluster ‘Acute activation’ ( Figure 1B ) since marker genes included known cAMP-response genes such as Fosl2 ( Figure 1D and Figure 1—figure supplement 2E ) and the phosphodiesterases Pde10a , Pde4b , and Pde4d ( Figure 1D , Supplementary file 4 ) . The ‘Acute activation’ cluster displayed the highest transcriptional score of marker genes from the ‘Activated’ cluster ( besides the ‘Activated’ cluster itself ) including genes such as Klf2 which confirmed the close relationship between these two clusters ( Figure 1D and H and Figure 1—figure supplement 2G , p-value [Tukey’s honest significant differences , HSD] = 0 . 001 ) . dmPGE2-treated cells in other clusters also showed strong expression of target genes such as Tsc22d3 , but in contrast to the ‘Acute activation’ cluster the expression of cluster identity genes ( e . g . Txnip , Mllt3 ) was maintained in the dmPGE2-treated ‘quiescent’ cluster ( Figure 1D ) . Poly ( I:C ) treatment increased the proportion of HSCs in the ‘interferon’ cluster from 1% to 42% ( Figure 1F and p-value [DPA] <10–5 ) . The top 100 poly ( I:C ) -stimulated genes exhibited a 72% overlap with the top 100 marker genes of the ‘interferon’ cluster ( Figure 1H and p-value ( hypergeometric test , false discovery rate [FDR]-corrected ) = 10–144 , Supplementary file 9 ) suggesting that poly ( I:C ) treatment reinforces a transcriptional program that already exists endogenously in a small proportion of HSCs ( Figure 1 , Figure 1—figure supplement 2A-D ) . In contrast to dmPGE2 , the transcriptional response to poly ( I:C ) was strongest in the ‘interferon’ cluster since target genes , for example , Oasl2 or Peli1 , were less induced in the other poly ( I:C ) -treated clusters ( Figure 1D ) . Treatment with G-CSF led only to minimal shifts in HSC distribution ( Figure 1E ) and proportions between HSC clusters , respectively ( Figure 1F and p-value [DPA] >0 . 05 for all clusters ) . The transcriptional response for most G-CSF target genes such as Myb , Eif4ebp1 , or Ncl was strongest within the ‘metabolism’ cluster ( Figure 1D ) with a 34% overlap ( p-value [hypergeometric test , FDR-corrected] = 8 . 2*10–49 ) between ‘metabolism’ marker genes and G-CSF-induced genes ( Figure 1H , Supplementary file 9 ) . In summary a 2 hr in vivo pulse with poly ( I:C ) or dmPGE2 significantly altered distributions of HSCs between pre-existing transcriptional states and , in the case of dmPGE2 , allowed for a novel transcriptional state to surface . The fact that certain clusters ( e . g . ‘metabolism’ and ‘interferon’ ) responded more strongly to external stimuli combined with the observation that HSCs kept their baseline cluster identity to varying degrees strongly suggests that transcriptional heterogeneity does not only exist at baseline but also during HSCs’ response to extrinsic signals . To better understand how poly ( I:C ) induced interferon signaling , we evaluated different components of the TLR and interferon pathways in our single-cell clusters . Binding of poly ( I:C ) to Toll-like receptor 3 ( TLR3 ) ( Alexopoulou et al . , 2001 ) induces expression of Type I interferons ( IFNα and IFNβ ) , which in turn signal via IFNα/β receptor 1 ( Ifnar1 ) and 2 ( Ifnar2 ) heterodimers , all of which were expressed in HSCs ( Figure 4E ) . We identified two expression patterns in poly ( I:C ) -treated HSCs that were consistent with TLR and interferon receptor signaling . The first expression pattern ‘up interferon’ was driven by induction of poly ( I:C ) responsive genes across all cell states . In addition , these genes were already specifically enriched in the ‘interferon’ cluster in the absence of poly ( I:C ) stimulation ( Figure 2A ) . Genes within this group are either directly downstream of Type I interferon receptors , such as Stat2 and Irf9 , or act as effector proteins involved in viral interferon response such as Apobec3 and Eif2ak2 ( Figure 2A , Figure 1—figure supplement 4A ) . The high expression of several interferon-induced viral-response genes ( e . g . Bst2 , Ifitm3 , Ube2l6 , and Rnf213 ) in the control ‘interferon’ cluster might point to a state of general surveillance for viral infection at baseline ( Figure 2A , Figure 1—figure supplement 4A ) . The second expression pattern ‘up Toll-like receptor’ constituted poly ( I:C ) -induced genes that were predominantly found in the ‘interferon’ cluster with low expression at baseline in the control ‘interferon’ cluster ( Figure 2A , Figure 1—figure supplement 4A ) . Genes within this signature included Nfkbia , Peli1 , Map3k8 , and Rps6ka3 all of which are part of TNFα and Toll-like signaling pathways . This expression profile might therefore represent a more direct response to poly ( I:C ) interaction with Tlr3 . Comparison of differential expression patterns across cell states allowed us to distinguish between poly ( I:C ) -mediated TLR- and interferon-based signaling . Even though G-CSF did not change cell distribution between clusters ( Figure 1F ) , it induced DEGs , most within the HSC ‘metabolism’ cluster ( Figure 1D , Figure 1—figure supplement 4B ) . Hierarchical clustering suggested that G-CSF treatment drove the expression profile of the HSC ‘metabolism’ cluster closer toward the ‘cell cycle’ state ( Figure 1—figure supplement 4B ) . This shift was facilitated by induction of genes related to transcription , such as RNA binding proteins ( Hnrnpd , Hnrnpf , Hnrnpa2b1 ) , as well as splicing factors ( Srsf7 , Sf3b1 , Srsf2 ) ( ‘transcription’ , Figure 2B ) . G-CSF also increased expression of transcripts involved in translation ( ribosome biogenesis: Nop14 , Nip7 , Wdr43 , Wdr12 and translation initiation: Eif4a1 , Eif4ebp1 ) that were not expressed in the ‘cell cycle’ state at baseline ( ‘translation’ , Figure 2B ) . This may indicate a G-CSF-induced fate commitment toward differentiation . Overall , a 2 hr pulse of G-CSF pushed HSCs toward a more metabolically active state . Our scRNA-Seq data are consistent with the original description of G-CSF as a growth factor that regulates myeloid differentiation and indicates an early transcriptional response leading to HSC mobilization . To investigate external signaling in a more physiological setting , we orally treated mice for 1 week with indomethacin to deplete endogenous prostaglandins . Differential expression analysis identified only 21 genes ( 1 . 2-fold change cutoff , Figure 4C ) affected by indomethacin . Ten out of twelve upregulated genes can be classified as IEGs ( e . g . Fos , Fosb , Jun , Klf4 , or Klf6 ) ( Figure 1—figure supplement 4D , Supplementary file 8 ) . While cell proportions did not change between the HSC clusters ( Figure 1F ) , distribution of cells shifted slightly toward the periphery of the UMAP plot ( Figure 1E ) which was mirrored by increased expression of individual ‘Activated’ cluster marker genes such as Fos and other IEGs ( Figure 2C–D and Figure 2—figure supplement 1A-B ) . To further investigate the influence of endogenous prostaglandin depletion on cell state while taking the entire transcriptional landscape into account , we computed diffusion pseudotime ( DPT ) ( Haghverdi et al . , 2016 ) between the ‘Activated’ and ‘Quiescent’ cluster in HSCs . The cell with the combined highest expression of the three top cluster markers for the ‘Activated’ state ( Figure 2E , see Materials and methods ) was set as the ‘root cell’ and DPT was calculated originating from that root cell ( Figure 2F ) . Indomethacin-treated cells displayed a significant shift in overall pseudotime kernel density distribution , which is indicative of overall lower pseudotime ( Figure 2G , shift indicated by asterisk , p-value = 5 . 8*10–12 by Mann–Whitney U-test ) . No shift was observed when comparing the control to G-CSF-treated HSCs ( Figure 2H and p-value = 0 . 18 ) . Ranking cells for each treatment condition according to pseudotime and averaging gene expression in 10 equally sized bins ( quantile ranks 1–10 ) further illustrated the change in expression of Fos and other IEG genes following indomethacin , especially at lower pseudotimes ( Figure 2I and Figure 2—figure supplement 1C; indicated by asterisks ) . Genes that were not part of the ‘Activated’ gene signature , such as Ly6a , did not follow the same pattern ( Figure 2J ) , nor was a similar trend observed in response to G-CSF treatment ( Figure 2—figure supplement 1D ) . The pseudotime analysis of the scRNA-Seq data indicated a specific shift in IEG transcriptional state upon depletion of endogenous prostaglandins . To further confirm the effect of endogenous prostaglandins on IEGs in an orthogonal assay , we measured single-cell protein levels of FOS by intracellular flow cytometry . Across two independent experiments , a 7-day in vivo indomethacin treatment led on average to a 34% ( SD = 8 . 2% ) reduction in FOS mean fluorescent intensity ( MFI ) in HSCs ( p-value = 6 . 2 * 10–3 , t-test with Welch’s correction ) and a mean 35% ( SD = 8 . 6% ) decrease in LSKs ( p = 6 . 6 * 10–3 , Figure 2K–L ) . Overall , endogenous prostaglandin levels impacted both the transcriptional state and protein levels of FOS and potentially other IEGs . To evaluate specificity of transcriptional heterogeneity observed within HSCs and their response to external signals , we analyzed the transcriptome of the entire LSK compartment , which encompasses mostly MPPs and a small proportion ( ~2% ) of HSCs ( Figure 1—figure supplement 1A and E ) . Transcriptional responses and LSK cell states in phenotypically defined MPPs ( Cabezas-Wallscheid et al . , 2014; Pietras et al . , 2015 ) ( MPP0 , MPP1 , MPP2 , MPP3/4 , Figure 1—figure supplement 1A ) were profiled using a hashtag oligonucleotide ( HTO ) labeling strategy that is part of the cellular indexing of transcriptomes and epitopes by sequencing ( CITE-Seq ) methodology ( Figure 3—figure supplement 1A , C , D and Materials and methods Stoeckius et al . , 2018 ) . Cell hashing enables tracking of cell surface phenotypes in scRNA-Seq data through barcoding of cells with antibody conjugated DNA-oligos ( HTO barcoding ) . ScRNA-Seq gene expression of marker genes such as Cd34 , Cd48 , and Cd150 ( Slamf1 ) matched the surface phenotypes used for sorting of HTO-barcoded MPPs , confirming that our workflow was successful ( Figure 3—figure supplement 1B , E ) . We analyzed transcriptomic data from LSK cells as an aggregated set consisting of all four perturbations and control , analogous to the approach used for HSCs above . We discovered a total of eight LSK clusters , which similar to HSCs displayed gene expression enrichment as opposed to exclusive expression of marker genes ( Figure 3B , Figure 3—figure supplement 1B ) . These LSK clusters were labeled by analysis of enriched genes and pathways ( Figure 3E , Supplementary files 4-6 ) , their composition of phenotypically defined cell populations tracked by HTO barcoding ( Figure 3—figure supplement 1C and G ) and by comparing the top 100 enriched genes of LSK clusters to the earlier defined HSC clusters ( Figure 3A , Supplementary file 9 ) . Because the latter analysis only indicated similarity rather than full equivalence of HSC and LSK clusters , and to avoid ambiguity when evaluating HSCs and LSKs , all LSK clusters were denoted with the prefix ‘LSK-’ . LSK clusters most similar to the ‘quiescent’ HSC state by top enriched genes were named ‘LSK-primitive’ and ‘LSK-primed’ , respectively ( Figure 3A ) . These two clusters further expressed the highest level of the HSC ‘quiescence’ score ( Figure 3—figure supplement 1H , p-value ( Tukey’s HSD ) = 0 . 001 ) . The ‘LSK-primitive’ cluster encompassed the majority of phenotypic HSCs and was significantly depleted of MPP3/4s compared to all other clusters ( Figure 3—figure supplement 1F-G , DPA p-values < 0 . 02 ) . LSK cells in the ‘LSK-primed’ cluster represented a more committed state given their expression of Cd34 and Flt3 . Enrichment of Cd37 and Sox4 suggested priming toward a lymphoid fate ( Figure 3E; Sun et al . , 2013; Zou et al . , 2018 ) . In contrast to HSCs , a higher proportion of LSKs were in a metabolically active or cycling state ( 43% LSKs [Figure 3C] vs . 35% HSCs [Figure 1F] , p-value ( chi-squared test ) = 1 . 7*10–5 ) . In addition , the ‘LSK-metabolism’ cluster itself exhibited a stronger cell cycle signature compared to the HSC ‘metabolism’ cluster ( Figure 3A and increased expression of Ki67 and Top2a Figure 3E vs . Figure 1D ) . A small proportion of LSKs ( <1% , Figure 3B–C ) , comprising the ‘LSK-myeloid’ cluster , were defined by expression of genes such as Mpo , Ctsg , Fcer1g , and Cebpα ( Figure 3E ) . Consistent with previous reports ( Pietras et al . , 2015 ) , our data indicated that the ‘LSK-myeloid’ cluster was composed of MPP2s and MPP3/4 cells but no HSCs , MPP0s , or MPP1s ( Figure 3—figure supplement 1G ) . In summary , control-treated LSKs were distributed among four main clusters , those being ‘LSK-primed’ , ‘LSK-primitive’ , ‘LSK-metabolism’ , and ‘LSK-cell cycle’ , that together encompassed >99% of control LSK cells ( Figure 3C ) . Comparison to HSC clusters and HTO-barcoded MPPs allowed to define identities of LSK clusters . Consistent with previous functional studies , we found enrichment of phenotypically defined MPPs in corresponding transcriptional clusters ( e . g . MPP2 and -3 in ‘LSK-myeloid’ cluster ) . Compared to HSCs , baseline transcriptional heterogeneity in the LSK population was equally fluid but predominantly defined by an increased proportion of lineage-committed and mitotically active cells . Analogous to HSCs we evaluated the effects of external stimulants on LSKs by both assessing changes in LSK distributions between clusters and differential gene expression within LSK clusters ( Figure 3C , E and G and Figure 3—figure supplement 2 , Supplementary files 5-7 and 10 ) . Treatment with dmPGE2 or poly ( I:C ) gave rise to novel clusters that were absent in control LSKs ( Figure 3B-D , G ) . These treatment-induced LSK cell states displayed transcriptional profiles that were similar to the HSC equivalent cell states ( Figure 3A and E ) . Poly ( I:C ) treatment induced two interferon responsive clusters in LSKs , of which one showed higher mitotic activity ( ‘LSK-interferon cell cycle’ , Figure 3A , C and E ) . Like in HSCs , G-CSF and indomethacin treatment did not alter cell proportions within LSK clusters ( Figure 3C and G ) . In contrast to HSCs , in LSKs considerably less overlap existed between cluster-defining and stimulant-induced gene programs ( Figure 3F ) . The poly ( I:C ) -induced gene program had no match to a baseline cluster identity because no interferon responsive cluster was present in unperturbed LSK cells ( Figure 3C and F ) . For G-CSF a statistically significant but smaller ( 12% , p-value [hypergeometric test , FDR-corrected] = 10–10 ) overlap existed between G-CSF-induced genes that were also ‘LSK-metabolism’ marker genes compared to HSCs ( Figure 3F , Supplementary file 9 ) . Overall , poly ( I:C ) and dmPGE2 initiated a transcriptional program that altered the original LSK cell identity shifting cells between clusters . In contrast to HSCs , poly ( I:C ) induced the emergence of two new LSK cell clusters that did not exist in control . While responses to external stimuli were equally heterogeneous in the more differentiated LSK population , compared to HSCs , there was less crosstalk between LSK cell state heterogeneity at baseline and following perturbation of external signaling . To evaluate and compare the magnitude of transcriptional changes in HSCs and LSKs in greater detail , DEGs for all four treatments at three levels of expression changes across all clusters , that is , using a 1 . 5-fold change , 1 . 2-fold change , and no fold-change cutoff ( FDR < 0 . 01 see Materials and methods and Supplementary file 7 ) were compiled . We then aggregated genes based on common ( ‘up/down overlap’ ) or unique expression ( ‘up/down HSC/LSK only’ ) within HSCs or LSKs ( Figure 4A–D ) . G-CSF perturbed gene expression more strongly within LSKs ( green bars , Figure 4A ) whereas stimulation by poly ( I:C ) predominantly affected HSCs ( purple bars , Figure 4B ) . Receptor expression could not explain this difference since both the G-CSF receptor Csf3r and the type I interferon receptors Ifnar1 and Ifnar2 were expressed in a higher proportion of LSK cells compared to HSCs ( Figure 4E and F ) . For perturbation of prostaglandin signaling indomethacin was found to selectively affect HSCs ( Figure 4C ) whereas dmPGE2 led to a balanced effect on HSCs and LSKs , with neither compartment dominating the DEGs ( Figure 4D ) . In conclusion , different stimuli exhibited varying degrees of gene expression for either LSKs or HSCs . Receptor expression at baseline could not explain the variability of transcriptional responsiveness between HSCs and LSKs . To better understand HSC intrinsic factors regulating the transcriptional ‘receptiveness’ to signals and resulting heterogeneous responses , we assessed chromatin states using scATAC-Seq ( see Materials and methods ) of sorted HSCs and MPPs . We clustered cells based on chromatin accessibility in HSCs resulting in two clusters ( ‘HSC cluster 0’ and ‘HSC cluster 1’ , Figure 5B ) and LSK cells consisting of MPPs and HSCs resulting in eight clusters ( Figure 5C and Figure 5—figure supplement 1A-B , Materials and methods ) . To gain insight into the nature of the differentially accessible chromatin regions , we computed a per-cell transcription factor ( TF ) motif activity score using ChromVar ( Schep et al . , 2017 ) and evaluated enrichment of these scores across clusters . The motif activities of TFs CREB1 , NF-κB , and STAT3 that are immediately downstream of prostaglandins , poly ( I:C ) , and G-CSF ( Figure 5A ) , respectively , were homogeneously distributed in HSCs ( Figure 5D , Figure 5—figure supplement 1C ) and the majority of LSK clusters ( Figure 5E , Figure 5—figure supplement 1D and Supplementary file 11 ) . This result suggested that HSCs have an equally responsive potential to these external signals based on their accessible chromatin states . We did detect differential enrichment of motifs for TFs that are further downstream in the response to external signals . Interferon regulatory factors ( IRFs ) that bind interferon signaling response elements ( ISREs ) are induced by NF-κB signaling as well as direct targets of poly ( I:C ) intracellular binding ( Negishi et al . , 2018 , Figure 5A ) . The AP-1 motif can be bound by FOS and JUN , both are downstream effectors of the prostaglandin/CREB1 signaling pathway ( Luan et al . , 2015 , Figure 5A ) . We found differential ISRE enrichment in HSC cluster 1 ( log2FC = 0 . 57 , p-value ( logistic regression ) = 2 . 4*10–5 ) and AP-1 enrichment in HSC cluster 0 ( log2FC = 2 . 6 , p-value ( logistic regression ) = 3 . 0*10–63 , both indicated by asterisks , Figure 5D and Figure 5—figure supplement 1C ) . In addition , HSC cluster 0 displayed increased motif activity enrichment for several key HSC lineage-specific master TFs including RUNX ( log2FC = 1 . 3 , p-value ( logistic regression ) = 8 . 0*10–23 ) GATA ( log2FC = 0 . 68 , p-value ( logistic regression ) = 7 . 2*10–8 ) , and Pu . 1/SPI1 ( log2FC = 0 . 60 , p-value ( logistic regression ) = 1 . 9*10–9 , indicated by asterisks , Figure 5D and Figure 5—figure supplement 1C ) as well as SMAD , another signal-responsive TF ( log2FC = 0 . 87 , p-value ( logistic regression ) = 2 . 7*10–16 , Figure 5—figure supplement 1E , F ) . In LSK cells the same motifs were also enriched in some clusters ( top log2FC indicated by asterisks , Figure 5E and Figure 5—figure supplement 1D , log2FC and p-values in Supplementary file 11 ) . However , no corresponding cluster like HSC cluster 0 existed where all lineage-specific ( RUNX , GATA , and Pu . 1 ) and signaling TF motifs ( AP-1 , SMAD ) co-occured ( Figure 5E and Figure 5—figure supplement 1D ) . In summary , the chromatin state directly downstream of external stimulants could not explain variability in gene expression upon treatment in HSCs . Rather , our analysis implicated cell intrinsic heterogeneity of downstream effectors , such as AP-1 and IRFs that may govern differential transcriptional responses . While cluster enrichment of AP-1 and ISREs was not unique to HSCs , we observed a specific co-occurrence of AP-1 and HSC lineage-specific master factors suggestive of HSC unique chromatin architecture .
Here , we provide a comprehensive transcriptional and epigenetic single cell analysis of a highly purified , functionally validated HSC population . Our work reveals that HSCs exist in fluent transcriptional and epigenetic states rather than distinctly separated cell types . While we cannot entirely rule out that the continuous cell states arose from the noisy nature of scRNA-Seq sampling , this is unlikely given our observation that genes that vary along the same transcriptional gradients are also functionally correlated ( e . g . IEGs ) . External perturbations rapidly shifted HSC distribution between HSC states within hours of signaling , providing evidence that the transcriptional states are highly dynamic allowing HSCs to quickly transition between states . Interestingly , we observed heterogeneity of HSC responses to external stimuli which may be determined by the baseline transcriptional and epigenetic state supported by our single-cell chromatin studies . Preliminary findings suggested an HSC specific co-occurrence of signaling and lineage-specific TF motif activities that is consistent with previous observations in human hematopoietic progenitors ( Trompouki et al . , 2011; Choudhuri et al . , 2020 ) . Overall , our data indicates that the single-cell landscape of in vivo-derived , functional HSCs is likely made up of a unique chromatin architecture with fluent transcriptional states , some of which can be rapidly influenced by external signals . Our combined scRNA-Seq and cell hashing ( HTO barcoding ) approach allowed us to gain insights into the transcriptional landscape of HSCs and phenotypically defined MPP populations within the LSK compartment at steady state and following perturbations with extrinsic signals . Our results enabled us to connect the transcriptional profile on a single-cell level to the previously described phenotypic behaviors of these MPP populations ( ; Pietras et al . , 2015; Cabezas-Wallscheid et al . , 2014 ) . For example even though both MPP2 and MPP3 cells have been previously described as myeloid biased ( Pietras et al . , 2015 ) , our analysis allowed to determine the proportion of putative myeloid cells within MPP2 and MPP3/4 cells as well as the relative MPP2 and MPP3/4 composition of myeloid cells . The HTO barcoding method provided a flexible tool to evaluate and compare transcriptional profiles within phenotypically defined populations because the technology used here is not dependent on the availability of specifically conjugated antibodies against particular surface receptors . In addition , Xist expression was used to deconvolute pooled male and female cells . While our analysis revealed only minimal sexual dimorphism that is consistent with previous reports ( Nakada et al . , 2014; Gal-Oz et al . , 2019 ) , the negligible additional investment to obtain data from both sexes may become the default experimental design in mammalian scRNA-Seq experiments . Our work presents evidence for two value-adding pooling strategies that allow for further insights into cell populations analyzed by scRNA-Seq . We used a two-pronged strategy to assess the specificity of external perturbations in HSCs and LSKs . First , we determined changes of cell proportions between cell states . Second , we evaluated differential expression within particular cell states following stimulation . Comparison of cluster-enriched and treatment-induced genes allowed us to identify unique and common genes for a given perturbation or a specific cluster . In contrast to LSKs , HSCs exhibited a high degree of overlap between stimulant-induced and cluster marker-defined gene programs . These results suggest that even at baseline , HSC transcriptional heterogeneity is defined by differences in signaling activity . Changes in cell proportions between different clusters indicated further specificity for treatment and differentiation state . Poly ( I:C ) and dmPGE2 led to cellular shifts between distinct transcriptional states with poly ( I:C ) driving the formation of two novel interferon-related clusters in LSKs but not HSCs . The strength of transcriptional perturbation could not solely be estimated based on the distribution of cells within clusters alone . G-CSF did not change the cell proportions between clusters but rather elicited strong transcriptional responses within a given cell state . Comparison of DEGs within clusters in HSCs and LSKs indicated that HSCs display a smaller response across all clusters to G-CSF compared to LSK progenitors . In summary , scRNA-Seq enabled a number of analyses that uncovered novel , HSC-specific responses to external perturbations . We evaluated the effect of three complementary signaling pathways ( G-CSF , prostaglandin , and interferon ) on the transcriptional state of HSCs . Pharmacological perturbation of these signaling pathways allowed to tightly control critical experimental parameters ( e . g . genetic background of mice , timing of sample processing ) that mitigated potential confounders of the downstream analysis . With the exception of indomethacin , we chose a short treatment window of 2 hr to increase the likelihood of studying direct downstream effects of stimulants on HSCs . Analysis of DEGs within clusters indicated interferon- vs . Toll-like receptor response genes induced by poly ( I:C ) treatment . While we could not detect transcripts for Type I interferons in our scRNA-Seq data of HSCs or MPPs , it is possible that some of the interferon-response genes were induced indirectly by release of interferons from the niche . An interferon inducer similar to poly ( I:C ) has been previously shown to increase IFNα protein levels in the serum as early as 2 hr post in vivo injection ( Linehan et al . , 2018 ) . Future work using genetic models is needed to further dissect indirect vs . direct effects of external stimulants on HSCs . There is a tradeoff between the strength of a perturbation required for experimental robustness vs . studying signals that are more physiologically relevant but lead to more subtle changes within and between cells . Here , we evaluated response of HSCs to three different external activators mimicking niche signals that were dosed two to four orders of magnitude higher than what an animal would typically encounter during actual injury or infection ( Eyles et al . , 2008; Porter et al . , 2013; Hoggatt et al . , 2013; Sheehan et al . , 2015 ) . To assess niche-derived signals in a more physiological setting , we administered the Cox1/2 inhibitor indomethacin orally for 1 week to deplete endogenous prostaglandins . As expected , the changes in gene expression with indomethacin were much weaker than those observed after acute injection with dmPGE2 , G-CSF , and poly ( I:C ) . ScRNA-Seq analysis offers unique tools to evaluate gene expression changes in response to weak perturbations . Pseudotime analysis showed that depletion of endogenous prostaglandins using indomethacin led to a small but significant shift in the transcriptional state of HSCs . The effect of indomethacin on IEGs such as Fos was further validated in independent FACS experiments which showed that the transcriptional programs implicated through pseudotime were also found to be perturbed using this orthogonal assay . How exactly the increase in RNA levels of Fos observed in scRNA-Seq can be reconciled with decreased FOS protein levels determined by FACS analysis will need to be addressed in future experiments . Another important implication and potential caveat highlighted by our findings is that RNA and protein levels may not always positively correlate , even on a single-cell level . Regardless , scRNA-Seq technologies provide sensitive tools to interrogate subtle changes in cellular states . In summary , we showed that single-cell approaches provide a rich and sensitive tool to analyze transcriptional and epigenetic states of HSCs during homeostasis and upon external perturbation . We found that HSCs exist in dynamic cell states and external signals can induce rapid transitions between , as well as changes within , these HSC states . While our work did not reveal whether these transcriptional states are associated with specific niches in vivo , novel spatial transcriptomic approaches provide exciting new opportunities to address such questions ( Rodriques et al . , 2019 ) . Additionally , recently developed barcoding strategies enable assessment of treatment-induced transcriptional changes and functional potential of single cells within the same experiment ( Rodriguez-Fraticelli et al . , 2020 ) . Understanding endogenous levels of niche-derived factors and the associated transcriptional and epigenetic responses will advance our basic understanding of stem cells and their potential applications in the clinic .
Whole bone marrow was isolated from femur , tibia , hip , and vertebrae via gentle crushing using a mortar and pestle . Stem and progenitor cells were enriched via lineage depletions ( Miltenyi Biotech , 130-090-858 ) . Antibodies , dilutions , and vendors are listed in the Key resources table . Cells were stained for 1 . 5 hr based on published best practice protocols for assessing CD34 labeling ( Ema et al . , 2006 ) . HSCs ( LSK , CD48- , CD150+ , CD34- ) , MPP1s ( LSK , CD48- , CD150+ , CD34+ ) , MPP0s ( LSK , CD48- , CD150- ) , MPP2s ( LSK , CD48+ , CD150+ ) , and MPP3/4s ( LSK , CD48+ , CD150- ) were sorted on a FACSAria ( Becton Dickinson ) and representative sorting scheme is shown in Figure 1—figure supplement 1A . Purity of >80% was ensured by reanalyzing each sorted population . To determine appropriate sample sizes of mice and HSCs , we performed an initial experiment on fresh HSCs ( HSC Replicate 1 ) which yielded estimated number of 2382 cells ( after filtering ) , and which resolved biologically meaningful clusters ( Figure 1—figure supplement 2A ) . In subsequent experiments we therefore targeted obtaining a similar or higher cell number . For external stimulant treatment , we based our sample size of five male and five female mice on this initial experiment . Because of sample processing times , a maximum of two conditions could be performed on the same day , resulting in three separate days of experiments . To mitigate batch effects resulting from different experimental days , the following precautions were taken . ( 1 ) All mice included in the external stimulant treatment were ordered from the same batch from JAX . ( 2 ) Control mice were administered acidified water and injected with DMSO to control for both unspecific perturbations that might result from the external stimulant treatments . ( 3 ) All experiments were performed within less than 1 week and single-cell libraries were prepared together for all samples after the initial droplet reaction was frozen . ( 4 ) FACS gates were set up initially but left constant for each experiment . Single color controls as well as fluorescence minus one controls ensured that there was minimal day-to-day technical drift on the FACS instrument . BM extraction , lineage depletion , and surface marker staining were performed as described above . Cells were fixed and permeabilized for intracellular staining according to manufacturer’s instructions ( BD Biosciences , 554714 ) . Intracellular staining was performed for 30 min on ice . Samples were analyzed on an LSRII FACS analyzer . Recipient CD45 . 2 ( Jax #00664 ) mice were gamma-irradiated ( Cs-137 source ) with a split dose of 5 . 5 Gy each 1 day before transplantation . HSCs were isolated from CD45 . 1 ( Jax #002014 ) donors and transplanted with 200 , 000 whole bone marrow cells ( CD45 . 2 ) via retro-orbital injection . Donor cell engraftment was monitored monthly for 16 weeks using an LSRII FACS analyzer ( Becton Dickinson ) . Flow cytometry data were analyzed with FlowJo ( Tree Star ) . HSC frequency was calculated using the following website: http://bioinf . wehi . edu . au/software/elda/ . Male and female cells were sorted separately but pooled in equal ratios before further downstream processing . For CITE-Seq HTO labeling of MPP populations , 0 . 25 µg of TruStain FcX Blocking reagent ( Biolegend ) was added for 10 min on ice . Each MPP populations was labeled with 1 µg of TotalSeq antibody cocktail ( Biolegend , see Key resources table ) and incubated for 30 min on ice . After washing , cells were resuspended in small amounts , counted and pooled in equal ratios . Each drug treatment condition resulted in one pooled MPP and one HSC sample that were processed separately for scRNA-Seq according to manufacturer’s recommendations ( 10× Genomics , 3’ V2 for HSC Replicate 1 experiment and V3 for external stimulant treatments ) . Briefly , for pooled MPPs , no more than 10 , 000 cells were loaded . For HSCs , all sorted cells ( between 2222 sorted events for dmPGE2 and 12 , 017 sorted events for control ) were loaded on the 3’ library chip . For preparation of HTO – surface libraries manufacturer’s recommendations ( Biolegend ) were followed . For ATAC-Seq , HSCs and MPPs ( pooled MPP0 , MPP1 , MPP2 , and MPP3/4 ) were sorted as described above from five male and five female mice ( strain CD 45 . 2 [Ly5 . 2] , JAX strain #00664 ) . Nuclei were isolated and libraries were prepared using manufacturer’s recommendations ( 10× Chromium Single Cell ATAC ) . Libraries were sequenced on a Next-seq 500 , 75 cycle kit ( ‘Replicate 1’ , scRNA-Seq ) and NOVAseq 6000 , 100 cycle kit ( ‘Replicate 2’ and external stimulant treatments , scRNA-Seq , scATAC-Seq ) . All code and a detailed description of the analysis is available in a dedicated GitHub repository ( see link in key resources table ) . To ensure reproducibility the entire analysis ( except cellranger and CITE-Seq count ) was entirely performed in Docker containers . Containers used for the analysis are indicated in the Jupyter notebooks and corresponding images are available on dockerhub ( see link in key resources table ) . Interactive cell browser web app is available here: ( https://mouse-hsc . cells . ucsc . edu ) . Raw data are available with GEO accession code GSE165844 . Cellranger ( v3 . 0 . 1 ) command ‘mkfastq’ was used to demutliplex raw base call ( BCL ) files into individual samples and separate mRNA FASTQ files and HTO surface fastq files . The cellranger ‘count’ command was used with default options to generate gene by cell matrices from mRNA FASTQ files . CITE-Seq count ( version 1 . 4 . 3 ) was used to generate surface count by cell matrices from the HTO surface FASTQ libraries . For the fresh HSC Replicate 1 experiment cellranger ( version 2 . 1 . 0 ) was used for demultiplexing and count matrix generation . The mm10 reference genome was used for all alignments . For scATAC-Seq cellranger-atac mkfastq and count ( 1 . 2 . 0 ) was used for demultiplexing and alignment and generation of the fragment file . To generate the count matrix MACS2 was run with default parameters ( keeping duplicates ) on the aligned reads . Resulting peak summits were extended to 300 bp and counts were extracted from fragment file using a custom script ( see GitHub repository ) to generate a count matrix . The main parts of the bioinformatic analysis of scRNA-Seq data was performed using the python package scanpy ( Wolf et al . , 2018 ) . For filtering and quality control , best practice examples were followed ( Luecken and Theis , 2019 ) . Count matrices were filtered on a gene and cell level . Cells were excluded with either less than 3000 UMIs , less than 1500 ( LT ) , or 2000 ( MPPs ) genes or more than 20 , 000 ( LT ) or 30 , 000 ( MPPs ) counts . A cutoff of no more than 10% UMIs aligned to mitochondrial genes per cell was applied . Genes expressed in less than 20 cells were excluded from the analysis . Counts were normalized to 10 , 000 per cell and log transformed . Features ( genes ) were scaled to unit variance and zero mean before dimensionality reduction . To reveal the structure in the data , we built a neighborhood graph and used the leiden community detection algorithm ( Traag et al . , 2019 ) to identify communities or clusters of related cells ( see also below ) . The UMAP algorithm was used to embed the high-dimensional dataset in a low-dimensional space ( Becht et al . , 2018 ) . DPA was used for comparing cell proportions between clusters as previously described ( Farbehi et al . , 2019 ) . Interactive visualization app of scRNA-Seq data was prepared using UCSC Cell Browser package ( Speir et al . , 2021 ) . We used the DemuxEM ( Gaublomme et al . , 2019 ) implementation in pegasuspy to assign MPP surface identities and demultiplex the pooled MPP sample . First background probabilities ( ‘pg . estimate_background_probs’ ) were estimated using default settings and ‘pg . demultiplex’ was run adjusting the alpha and the alpha_noise parameter to maximize cell retrieval by singlet classification . Assignments were validated by plotting count matrix in UMAP space and observing four distinct clusters indicative for the four HTO labels that were pooled . The proportion of demultiplexed cells matched the original pooling ratio . Analysis of coexpression of sex-specific genes allowed for further validation of the doublet rate . Proportion of cells classified by DemuxEM as doublets exceeded doublet rate estimated by coexpression of sex-specific genes . Because of timing required for FACS and sample prep , it was impossible to obtain HSCs and MPPs from all conditions on 1 day ( see also ‘Sample size estimation and sample batching’ above ) . To evaluate if batch correction was needed , we determined scRNA-Seq clusters and enriched genes by processing each sample separately or by combined analysis of all samples . Even though similar scRNA-Seq clusters were found in individual samples , these populations were non-overlapping in the integrative analysis ( especially for G-CSF ) . To correct for the batch effects we used ComBat ( Johnson et al . , 2007 ) with default settings on the log2 expression matrix , allowing cells to be clustered by cell type or cell state . Batch correction results were similar when we used Scanorama ( Hie et al . , 2019 ) and Harmony ( Korsunsky et al . , 2019 ) but both of these methods appeared to be overcorrecting with respect to the dmPGE2-treated population . To correct for potential sex-specific differences Xist counts were regressed out . Raw data was used for all differential expression analyses and plotting of single-cell gene expression values . Batch-corrected counts were used for clustering and DPT analysis . Since HSCs and MPPs are highly purified cell populations , we did not observe any clearly separated clusters in UMAP space . To aid the optimal choice of hyperparameters for leiden clustering , we used a combination of Silhouette coefficient and Davies–Bouldin index . We first validated this approach using the PBMC3K ( from 10× genomics , scanpy . datasets . pbmc3k ( ) ) silver standard dataset . We iterated through a range of KNN nearest neighbors and Leiden resolution combinations measuring average Silhouette coefficient and Davies–Bouldin index in PCA space for each combination . Plotting the optimal value for Silhouette score and Davies–Bouldin index vs . increasing numbers of clusters allowed for the determination of appropriate cluster number for the dataset . For the PBMC dataset , there was a clear drop-off in optimal value after eight clusters , which is corroborated by most single-cell tutorials that also report eight clusters for this dataset . After validation of this approach on PBMCs , we assessed Silhouette coefficient and Davies–Bouldin index for different clustering results of our own HSC and MPP datasets . This allowed us to select the optimal hyperparameters for each cluster number . The approach was validated by comparing two independent biological replicates of control HSCs ( ‘Replicate 1’ and ‘Replicate 2’ ) . Differential expression analysis was performed using MAST ( Finak et al . , 2015 ) . This method is based on a Hurdle model that takes into account both the proportion of cells expressing a given transcript and transcript levels themselves while being able to control for covariates . Based on previous reports , differential expression cutoff was set at 1 . 2-fold ( Smillie et al . , 2019 ) and a more stringent cutoff of 1 . 5-fold was also included . Only genes that were expressed in at least 5% of the cells were considered for differential expression analysis . FDR ( Benjamini and Hochberg ) cutoff was set at 1% . For drug treatments , differential expression between treatment and control was assessed within the entire LSK or HSC dataset and within each cluster controlling for number of genes per cell and sex . For differential expression analysis between male and female cells at baseline , control datasets were analyzed with clusters and number of genes as a covariates . For sex-specific effects of drug treatments , samples were split by sex and analyzed separately . Resulting differential expression coefficients were compared between male and female cells . To identify gene signatures with common patterns , for each treatment average expression of DEGs was extracted per cluster , scaled ( z-score ) and grouped together by similarity using hierarchical clustering ( seaborn . clustermap , Euclidean distance , single linkage ) . For DPT analysis ( Haghverdi et al . , 2016 ) , cells from the ‘Quiescent’ and ‘Activated’ cluster were selected for the following treatments: control , indomethacin , and G-CSF . We recalculated PCA and UMAP embeddings in this reduced dataset . Re-clustering using the Leiden algorithm was used to exclude outlier cells and assess top enriched genes within the new ‘Activated’ cluster . Raw expression of the three top enriched genes ( Nr4a1 , Nr4a2 , Hes1 ) was summed to robustly select the most highly ‘Activated’ cell as a root cell . DPT was calculated with the following function in scanpy ( ‘sc . tl . dpt’ ) using default settings . Cells were ranked according to pseudotime and kernel density distribution was plotted using a bandwidth of 0 . 02 . The Mann–Whitney U-test was used to assess if cells from different samples are drawn from the same pseudotime distribution . To analyze gene expression across pseudotime , for each sample cells were split into 10 equally sized bins according to ascending pseudotime . Bin 1 contained the first 10% of cells with the lowest pseudotime and bin 10 contained the 10% of cells with the highest pseudotime . Average gene expression for representative genes were plotted for each bin and sample . We performed over-representation analysis comparing various gene sets of interest ( upregulated by stimulants , enriched in clusters ) to a reference gene set . Depending on the analysis , the reference gene set was composed of an entire database of pathways ( REACTOME , GO:BP ) , manually curated pathways of interest ( searching for keywords on MSigDB database and from relevant publications; Goessling et al . , 2011; Schuettpelz et al . , 2014; Pedersen et al . , 2016; Giladi et al . , 2018; Mervosh et al . , 2018; Patterson et al . , 2020; Cilenti et al . , 2021; Rodriguez-Fraticelli et al . , 2020; Cabezas-Wallscheid et al . , 2017 ) or gene sets generated from the analysis itself ( marker genes from other clusters ) . Enrichment was assessed using a hypergeometric test ( one-sided Fisher’s exact test ) and p-values were corrected for FDR using Benjamini–Hochberg . We deliberately choose to evaluate the top 100 genes for every pairwise cluster/treatment comparisons to be more intuitive to interpret and compare . Transcriptional scores for each cluster were calculated using the scanpy function ‘scanpy . tl . score_genes’ . Briefly the score represents the average expression of a set of genes subtracted with the average expression of a reference set of genes . The reference set is randomly sampled for each binned expression value . Mean scores per cluster were compared via ANOVA followed by Tukey’s HSD test for individual post hoc mean comparisons . The R package Signac ( version 0 . 2 . 5 ) , an extension of Seurat ( Stuart et al . , 2019 ) , was used for quality control , filtering of ATAC-Seq peaks counts and plotting . Quality of scATAC-Seq dataset was ensured by presence of nucleosomal banding pattern and enrichment of reads around transcription start sites . Cells were removed with a less than 1000 or more than 20 , 000 fragments in peaks . Male and female cells were classified according to absence or presence of Y-chromosome reads . Since distribution of male and female cells appeared uniform across all analyses , no downstream correction was taken for sex . Term frequency-inverse document frequency was used for normalization and dimensionality reduction was performed by singular value decomposition . Cells were clustered using the Louvain community finding algorithm after a neighborhood graph was built with k = 20 ( HSCs ) or k = 30 ( LSK ) nearest neighbors . To calculate TF motif scores , ChromVAR ( Schep et al . , 2017 ) was run with default parameters using the JASPAR 2018 motif database . Differential TF motif activity scores between clusters were calculated with the ‘FindMarkers’ function in Signac using a logistic regression and p-values were adjusted using a Bonferroni correction . | Most organs in the human body are maintained by a type of immature cells known as adult stem cells , which ensure a constant supply of new , mature cells . Adult stem cells monitor their environment through external signalling molecules and replace damaged cells as needed . Stem cell therapy takes advantage of the regenerative ability of immature stem cells and can be helpful for conditions such as blood diseases , autoimmune diseases , neurodegeneration and cancer . For example , hematopoietic stem-cell transplantation is a treatment for some types of cancer and blood disorders , in which stem cells are harvested from the blood or bone marrow and reintroduced into the body , where they can develop into all types of blood cells , including white blood cells , red blood cells and platelets . Hematopoietic stem-cell transplants have been in use for over 30 years , but they remain a highly risky procedure . One of the challenges is that outcomes can vary between patients and many of the factors that can influence the ‘regenerative’ potential of hematopoietic stem cells , such as external signalling molecules , are not well understood . To fill this gap , Fast et al . analysed which genes are turned on and off in hematopoietic stem cells in response to several external signalling molecules . To do so , three signalling pathways in mice were altered by injecting them with different chemicals . After two hours , the hematopoietic stem cells were purified and the gene expression for each cell was analysed . This revealed that the types of genes and the strength at which they were affected by each chemical was unique . Moreover , hematopoietic stem cells responded rapidly to external signals , with substantial differences in gene expression between individual groups of cells . Contrary to more specialised cells , the external signalling genes in some hematopoietic stem cells were already activated without being injected with external signalling molecules . This suggest that low levels of external signalling molecules released from their microenvironment may prepare stem cells to better respond to future stress or injuries . These results help to better understand stem cells and to evaluate how the signalling state of hematopoietic stem cells affects regeneration , and ultimately improve hematopoietic stem cell transplantation for patients . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"stem",
"cells",
"and",
"regenerative",
"medicine"
] | 2021 | External signals regulate continuous transcriptional states in hematopoietic stem cells |
In vitro selection of antibodies allows to obtain highly functional binders , rapidly and at lower cost . Here , we describe the first fully synthetic phage display library of humanized llama single domain antibody ( NaLi-H1: Nanobody Library Humanized 1 ) . Based on a humanized synthetic single domain antibody ( hs2dAb ) scaffold optimized for intracellular stability , the highly diverse library provides high affinity binders without animal immunization . NaLi-H1 was screened following several selection schemes against various targets ( Fluorescent proteins , actin , tubulin , p53 , HP1 ) . Conformation antibodies against active RHO GTPase were also obtained . Selected hs2dAb were used in various immunoassays and were often found to be functional intrabodies , enabling tracking or inhibition of endogenous targets . Functionalization of intrabodies allowed specific protein knockdown in living cells . Finally , direct selection against the surface of tumor cells produced hs2dAb directed against tumor-specific antigens further highlighting the potential use of this library for therapeutic applications .
Antibodies expended as the biochemical tools of choice to label antigens in cells or tissues . Over the past 20 years , recombinant methods have been developed to quickly select and improve monoclonal antibodies from highly diverse libraries . Recombinant antibodies can be selected from immune or naïve libraries . Immune libraries provide in general high affinity binders but , depending on the antigen , diversity is sometimes limited . Because natural antibody selection requires animal immunization , very conserved or toxic antigens should be avoided and , in general , only limited control of the immune response is possible . On the contrary , non-immune ( naïve ) libraries provide a higher diversity of binders even for antigens highly conserved in mammals , but high specificity and affinity can be reached only when selecting from very large functional libraries . Immune and naïve libraries are based on the manipulation of antibody fragment that retain binding capacity and specificity of the entire immunoglobulin G ( IgG ) . The smallest IgG portion capable of binding with high specificity an antigen is the Fv fragment consisting of the variable heavy ( VH ) and the variable light ( VL ) domains . In the case of single domain antibodies ( sdAb ) , a VH or a VL alone , is able to bind its target antigen . Each variable domain contains four conserved framework regions ( FRW or framework ) and three regions called CDR ( Complementarity Determining Regions ) corresponding to hypervariable sequences which determine the specificity for the antigen . VH and VL can be fused together using a synthetic linker and produced as a single protein in the form of a single chain Fv ( scFv ) . Easier to manipulate , they can be produced in several bacteria or eukaryote cell types , fused to various tags or functional domains . Interestingly , antibodies called HCAb in Camelidae ( Hamers-Casterman et al . , 1993 ) or IgNAR in sharks ( Greenberg et al . , 1995 ) have an antigen recognition part composed of only a VH domain . Camelid natural single domain VH , referred to as VHH or nanobodies , can be expressed as recombinant fragments and represent attractive alternatives over classical antibody fragments like scFvs because they are easy to manipulate and they are not limited by potential misfolding of the two domains ( Wörn and Plückthun , 1999 ) . It is noteworthy that VHH FRWs show a high sequence and structural homology with human VH domains of family III ( Muyldermans , 2013 ) and VHH have comparable immunogenicity as human VH ( Bartunek et al . , 2013; Holz et al . , 2013 ) . Thus , they further constitute very interesting agents for therapeutic applications , some of them are currently in phase II and Phase III clinical trials ( Ablynx Nanobodies; http://clinicaltrials . gov/ct2/results ? term=ablynx ) . Recombinant antibody fragments allowed not only to accelerate the identification of unique binders , but also the development of a novel type of tool: in this case , the antibodies are directly expressed in living cells as intracellular antibodies ( intrabodies ) , to trace or perturb endogenous target at the protein level . Some scFv or sdAb have indeed been directly expressed in eukaryotic cell as intrabodies to target with high specificity intracellular antigens . Several intrabodies have been used as fluorescent protein fusion to highlight endogenous antigen in cells in a spatio-temporal manner ( Nizak et al . , 2003a; Rothbauer et al . , 2006 ) . Intrabodies with intrinsic blocking activity have been reported ( Haque et al . , 2011; Shin et al . , 2005 ) , and several other approaches have been developed to allow a larger fraction of intrabodies to be used as inhibitory factors: forced co-localization ( Tanaka et al . , 2007 ) , suicide through proteasome targeting ( Joshi et al . , 2012; Melchionna and Cattaneo , 2007 ) , rerouting or sequestration to cell compartment ( Böldicke et al . , 2005 ) , degradation ( Caussinus et al . , 2011 ) . Depending on the target , such inhibitors may have potential in human therapy . Production of functional intrabodies depends on the stability of the antibody fragments in the reducing environment of the cytosol that does not allow disulfide bond formation between conserved cysteine . In this context , many advantages of the nanobody scaffold have been reported and , in particular , higher solubility , improved stability in a reducing environment ( Wesolowski et al . , 2009 ) , as well as higher expression yield and thermostability ( Jobling et al . , 2003 ) . For all these reasons , the nanobody scaffold represents an attractive option to generate functional intrabodies . Thus , we decided to create a non-immune recombinant antibody library of high diversity , based on a nanobody scaffold that would enable efficient in vitro antibody selection against virtually any antigen . Such a library should provide antibodies usable in conventional immune assays and be enriched in antibodies active in the intracellular environment . First , using a fusion assay in E . coli , a family of highly functional VHH scaffolds was isolated , optimized for intracellular expression and high stability . One particularly stable VHH scaffold consensus sequence was chosen from these selected antibodies . Additional changes were then introduced to reduce the distance between the Camelidae and human VH3 sequences . We confirmed by CDR grafting that this humanized synthetic scaffold ( hs2dAb ) was robust and functional . Statistics of amino-acid diversity in the CDRs were computed and these information were used to construct a high diversity phage display library of 3 . 109 independent hs2dAb . The library was then screened against diverse targets of various structures and origin . Highly specific antibodies were selected against EGFP , mCherry , β-tubulin , β-actin , heterochromatin protein HP1α , GTP-bound RHO , p53 and HER2 . Affinity measurement indicated that affinities in the nM range can be obtained using this library . As expected from our design , we further showed that hs2dAb are frequently usable as fluorescent intrabodies to track antigens in cells . We also showed that they can be functionalized to induce antigen knockdown . This thus represents the first report of a large and diverse synthetic single domain antibody library enabling fully in vitro selection of highly functional antibodies and intrabodies .
We reasoned that the usual lower quality of these non-immune libraries may come from ( 1 ) an antibody scaffold that may not allow robust folding and presentation of CDR region , ( 2 ) a lack of control of diversity in the CDR regions and ( 3 ) a frequent occurrence of incorrect clones due to the presence of unexpected mutation or empty clones . We designed a pipeline for the development of functional synthetic libraries that aims at overcoming these limitations ( Figure 1A ) . As a first step to construct a single domain antibody library enriched in highly stable and functional antibody fragments , we screened for a robust sdAb scaffold ( Figure 1—figure supplement 1 ) . Previously , we selected from immune or naïve llama VHH libraries several hundreds of clones ( Monegal et al . , 2012; Olichon and Surrey , 2007 ) . From this population , we identified a set of robust scaffolds using an assay that discriminates highly stable clones from clones prone to aggregation , or unfolding , in the bacterial cytoplasm ( Olichon and Surrey , 2007 ) . This assay is based on the fusion of HA-tagged chloramphenicol acetyl transferase ( CAT ) to the carboxy-terminus of VHH sequences ( Figure 1—figure supplement 1A ) . In these conditions , only bacteria expressing a functional VHH fusion in the reducing cytosol ( non aggregating , non degraded ) can grow in the presence of high antibiotic concentration , thus filtrating a sub-library of potential intrabodies . Expression yield in E . coli and apparent solubility as EGFP fusion in the mammalian cell cytoplasm were further assessed to select a set of robust antibody scaffolds . Strikingly , the consensus scaffold was matching the sequence of the most robust VHH framework , represented by a single domain antibody D10 ( hereafter named sdAbD10 ) . When compared to previously reported thermostable nanobodies ( Olichon et al . , 2007a ) or intrabodies ( Rothbauer et al . , 2006 ) obtained from immune libraries , sdAbD10 was found to provide higher antibiotic resistance in the chloramphenicol filter assay ( Figure 1—figure supplement 1A ) . Its expression yield in E . coli periplasm was in the higher range of soluble llama VHH fragments , allowing efficient and quantitative purification ( Figure 1—figure supplement 1B ) . Purified sdAbD10 showed excellent solubility , stability after treatment at 70°C and we did not observe aggregation when expressed as an intrabody fused to the EGFP in mammalian cells ( Figure 1—figure supplement 1C ) . 10 . 7554/eLife . 16228 . 003Figure 1 . Overview of scaffold selection , diversity design , and synthetic production of the NaLi-H1 library . The development of the NaLi-H1 library followed three lines of optimization . ( i ) A novel scaffold was defined by selecting a set of robust nanobodies using a CAT fusion assay ( 1 ) . A consensus was derived and mutations were introduced to humanize the scaffold ( 2 ) . Usability and efficacy of the novel scaffolds ( VHH and humanized ) were then confirmed evaluating their display on phage , expression in CHO cells and use as intrabodies ( 3 ) . In silico design was completed analyzing natural CDR diversity ( 4 ) and using this information to design synthetic CDRs . A fixed size of 7 aa was chosen for the CDR1 and CDR2 . 4 sizes ( 9 , 12 , 15 and 18 amino acids ) were chosen for CDR3 . Finally , the pHEN2 vector was improved by implementing a triple myc tag and inserting a toxic gene ( ccdb ) to ensure low frequency of empty clones during library construction ( 6 ) . Gene synthesis ( using a tri-nucleotide approach ) was ordered , synthetic sequences cloned into the novel pHEN2+ vector , transformed into bacteria and plated on 430 15 cm plates . 3 × 109 clones were obtained . Quality control was carried out using Sanger sequencing of 315 randomly picked clones and large scale sequencing of 56 000 clones . No redundant clone was found . The NaLi-H1 was then screened in various conditions and diversity , efficacy , versatility and affinity evaluated . DOI: http://dx . doi . org/10 . 7554/eLife . 16228 . 00310 . 7554/eLife . 16228 . 004Figure 1—figure supplement 1 . Robust scaffold identification . ( A ) Chloramphenicol acetyl transferase carboxy terminal fusion is a folding reporter allowing for the selection of soluble amino terminal VHH . Top: scheme of the construct expressed from pAO-VHH-CAT-HA vector . Bottom: Relative colony growth of selected VHH ( GFP4 and Lam1 chromobodies , or thermostable Re3 ) on chloramphenicol selection medium ( Cam ) . Serial dilution of E . coli culture expressing VHH . ( B ) Analysis of heat purified sdAbD10 by SDS-PAGE . Clone D10 was expressed in E . coli and protein secreted in the periplasm were extracted ( lane 1 ) . Periplasmic extract was subjected to heat treatment at 70°C and insoluble proteins were pelleted by centrifugation ( lane 2 ) . The soluble supernatant containing the VHH was then concentrated using Amicon filters ( lane 3 ) . ( C ) HeLa cells expressing a GFP fusion of sdAbD10 showing homo-dispersed fluorescence ( right ) compared to typical randomly chosen aggregating llama VHH considered as non intrabody ( left ) . DOI: http://dx . doi . org/10 . 7554/eLife . 16228 . 00410 . 7554/eLife . 16228 . 005Figure 1—figure supplement 2 . CDR3 loop grafting and synthetic scaffold validation . ( A ) Phages presenting each scaffolds ( sdAbD10 and hs2dAb ) bearing anti-lamin CDRs were produced in E . coli and supernatant were detected in Western Blot with an anti-pIII antibody ( NEB ) . Two bands are visible , one for wild-type pIII and one for the pIII fusion with single domain antibodies . ( B ) Dot blot analysis of the production of both scaffolds either as single domain antibodies in E . coli supernatant or as fusions with a human Fc domain and secreted by CHO cells ( Moutel et al . , 2009 ) . Serial dilutions of supernatant were revealed with an anti-His tag antibody or an anti-human Fc antibody . ( C ) Immunofluorescence of HeLa cells with recombinant antibody in both scaffolds labeling the nuclear rim structure characteristic of the nuclear lamina . ( D ) The anti-lamin based on the two scaffolds were transiently expressed in HeLa cells as GFP fusion . Living cells were imaged after 24 hr and showed that the hs2dAb recognized its intracellular target lamin . Bar = 10 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 16228 . 00510 . 7554/eLife . 16228 . 006Figure 1—figure supplement 3 . Nali-H1 library diversity . ( A ) Comparison between relative amino acid frequencies in CRD1 and CDR2 observed from 250 llama VHH isolated from a naïve library ( llama ) , the rationally designed ( Designed ) and the effective diversity in the synthetic library ( Nali-H1 ) as computed after sequencing of 2500 clones using NGS . The position of each amino acid is indicated ( CDR1 1; CDR1 2; etc ) . Amino Acid are indicated in single letter code . ( B ) Comparison between the designed diversity set for every amino acid position in the CDR3 region with the effective diversity observed after sequencing 2500 clones using NGS . Note that various CDR3 amino acid lengths are present in the library . They were almost evenly distributed with a little bias for shorter CDR3s ( 9 aa: 26%; 12 aa: 27 . 2%; 15 aa: 23 . 7%; 18 aa: 23 . 1% ) . For simplicity , we report in the figure the diversity observed for up to 18 aa . Note also that although cysteine occurs in natural llama CDRs , they were avoided by design and , accordingly , not found in Nali-H1 CDRs . DOI: http://dx . doi . org/10 . 7554/eLife . 16228 . 00610 . 7554/eLife . 16228 . 007Figure 1—figure supplement 4 . Solubility of secreted hs2dAb antibodies . ( A ) Sequential centrifugation carried out using clones selected from the NaLi-H1 library ( anti-GFP , anti-mCherry ) or obtained after CDR grafting into the hs2dAb scaffold ( anti-lamin ) . As a reference , the behaviour of a widely used natural anti-GFP antibody ( GFP4 or Chromobody ) was analysed . I: Input; P1: pellet after the 20 000 g centrifugation , P2: pellet obtained after the 250 000 g centrifugation , SN: supernatant of the 250 000 g centrifugation . Samples were analysed after western blotting using an anti-HIS antibody . Note that the apparent size of the lamin antibody and the Chromobody is lower because these antibodies are fused to only one myc tag ( expressed from the parental pHEN2-His-Myc plasmid ) . ( B ) , ( C ) : a set of clones randomly chosen in the NaLi-H1 library was expressed in a multiwell format and analysed as in A . Only-the pellet ( P ) and the supernatant ( SN ) of the 250 000 g centrifugation are shown . Two samples ( an anti-GFP hs2dAb and a randomly picked clone #1 ) were heated at 90°C for 10 min before being analysed by sequential centrifugation ( anti-GFP* and #1* ) . Note that in these conditions , a fraction of the hs2dAb is unfolded and found in the pellet . A significant resistance to temperature is however observed . Note that high molecular weight bands were systematically seen which may suggest that a fraction of the hs2dAb is in a multimeric form . The same behaviour was observed for the natural Chromobody . DOI: http://dx . doi . org/10 . 7554/eLife . 16228 . 007 To test whether the sdAbD10 scaffold composed of the FRW 1 to 4 was robust independently on the CDR sequences , we grafted the CDR loops of the lam1 VHH ( Rothbauer et al . , 2006 ) directed against laminB into the framework of the sdAbD10 ( sdAbD10-anti lamin ) . In parallel , partial humanization of the scaffold was also tested to figure out whether it affected dramatically its intrinsic properties . Seven residues of the synthetic scaffold sdAbD10 were altered towards the most represented in human VH3 while five other divergent residues were kept unchanged . The four llama VHH-specific amino acids hallmarks in the framework-2 region ( positions 42 , 49 , 50 , and 52 ) , which are essential to increase intrinsic solubility properties , as well as the Glutamine in position 103 , were maintained . We named the resulting hybrid single domain VH hs2dAb ( humanized synthetic single domain antibody ) ( Figure 1B ) . After grafting the CDR of the anti-lamin into these two scaffolds , we analyzed their display on the M13 surface ( Figure 1—figure supplement 2A ) and their production by E . coli and by CHO cells ( Figure 1—figure supplement 2B ) . Both scaffold showed similarly good efficacy . Importantly , both the sdAbD10 and hs2dAb scaffolds enabled functional display of the grafted CDRs and robust detection of endogenous lamin was observed by immunofluorescence staining of Hela cells ( Figure 1—figure supplement 2C ) . Last , Rothbauer et al . ( Rothbauer et al . , 2006 ) showed that the original anti-lamin VHH recognized its target antigen upon intracellular expression . Similarly , we observed that both the hs2dAb and the sdAbD10 scaffolds allowed the efficient intrabody use of the synthetic anti-lamin antibodies ( Figure 1—figure supplement 2D ) . This further indicated that the synthetic scaffolds were robust , non-aggregating and resisted to the reducing conditions found in the cytosol , while allowing the proper display of CDR loops . As partial humanization did not affect the properties of the sdAbD10 scaffold , we chose the hs2dAb scaffold as a unique framework to construct a diverse library of synthetic nanobodies endowed with the characteristic stability of these single domain antibodies while displaying an amino acid sequence closer to human VH3 . A synthetic diversity was introduced in the three CDRs by rationally controlling each position of the CDR1 and CDR2 using a set of amino acids that recapitulates partially natural diversity ( see the Appendix for more details ) while reducing the presence of the most hydrophobic residues in order to avoid the aggregation propensity ( see Material and methods ) . A constant length of 7 amino acids was selected for CDR1 and CDR2 . A large spectrum of Camelidae VHH CDR3 length is naturally observed and this loop is known to contribute strongly to antigen binding selectivity . Thus , we chose to use four different lengths of CDR3 to cover this spectrum ( 9 , 12 , 15 or and 18 amino acids ) and introduce random amino acid ( except cysteine ) at each position . Synthetic DNA was produced by the trinucleotide DNA assembly and amplification was carried out starting from 2 . 1011 different molecules , using only a few cycles of PCR ( PCR linearity validated by Q-PCR ) to prevent drift during amplification . The synthetic library was inserted into a modified pHEN2 phagemid vector containing a triple myc-tag and suicide gene ( ccdB ) that allowed positive selection of insert-bearing clones ( Bernard et al . , 1994 ) . Massive electroporation was carried out using E . coli TG1 cells and 430 large agar dishes ( 140 mm ) were used to ensure proper plating of the library . About 3 . 109 individual recombinant hs2dAb clones were obtained . We named this library NaLi-H1 ( for Nanobody Library-Humanized 1 ) . We first evaluated the quality of the NaLi-H1 library by sequencing 315 random clones . Only 13 sequences were found to be incorrect ( bearing an in-frame stop codon , missing one base , missing a large region [the CDR1 or CDR1-FWR1- CDR2] , or being empty ) . Thus , a total of about 4% of potential default was observed , which is rather low and only marginal in comparison to the 3 . 109 clones obtained . The diversity was then evaluated by sequencing 5 . 6 105 inserts using ion Torrent chips ( Life Technologies ) . This confirmed the quality of the library and showed that the four CDR3 lengths ( 9 , 12 , 15 or and 18 amino acids ) were present in similar proportions . The diversity and statistical distribution of amino acids in the CDRs were found to be as expected ( Figure 1—figure supplement 3 ) . To estimate the overall folding of the hs2dAb present in the library , we picked randomly 24 clones and tested their solubility in bacteria medium after secretion . Medium were centrifuged at 250 000 g and the supernatant and pellet were analyzed . All tested clones showed essentially complete solubility , at least as good as a natural Lama antibody ( GFP4 ) . Even after warming at 90°C for 10 min , the hs2dAb showed good solubility ( over 70% ) ( Figure 1—figure supplement 4 ) . The NaLi-H1 library was screened against a set of various antigens . Several standard phage display methods ( Hoogenboom , 2005 ) were used ( see Materials and methods for details ) : antigen adsorption on immunotube , native antigen captured on beads , direct selection at the cell surface . All conditions allowed the efficient recovery of diverse and functional antibodies . As a first screen to evaluate the quality of the library , we chose to select specific binders for the EGFP and mCherry fluorescent proteins . NaLi-H1 phages were panned against biotinylated EGFP or mCherry and 3 rounds of selection were carried out . Eighty clones were analyzed for each screening campaign . From the panning against EGFP , 37 non redundant nanobodies were shown to detect EGFP by phage ELISA . These antibodies were then used for immunofluorescence and 10 of them were found to efficiently stain EGFP in fixed HeLa cells ( Figure 2A ) . Similarly , selection against mCherry led to 6 positive binders ( Figure 2B ) . As shown in Figure 2A and B , no staining was obtained in untransfected cells . 10 . 7554/eLife . 16228 . 008Figure 2 . Selection of functional hs2dAb against various antigens . ( A ) HeLa cells were transfected with GFP-Rab6 , fixed using paraformaldehyde , permeabilized using saponin and stained with non-purified myc-tagged hs2dAb ( R3TF3 ) directed against EGFP and revealed with anti-Myc-Tag ( 9E10 ) and Cy3-labeled secondary antibodies . ( B ) HeLa cells transfected with mCherry-Rab6 , fixed and permeabilized as in A and stained using a myc-tagged non purified hs2dAb against mCherry ( C11 ) . The hs2dAb was then revealed using 9E10 and A488-labeled secondary antibodies . ( C ) Cells were fixed in methanol and co-stained with a non-purified anti-tubulin hs2dAb ( D5 ) fused to a human Fc domain and a mouse monoclonal anti-tubulin antibody ( DM1A ) , and revealed using an anti-Human Fc-A488 and an anti-Mouse-Cy3 secondary antibody , respectively . ( D ) hs2dAb F4 anti-beta-actin was used to stain MRC5 cells fixed with paraformaldehyde and post fixed with methanol . The hs2dAb was then revealed using 9E10 and A488-labeled secondary antibodies . Cells were co-stained by red fluorescent phalloidin to detect actin stress fibers . ( E ) A431 cells were fixed with 3% paraformaldehyde , permeabilized with 0 . 1% Triton and stained with the anti-p53 hs2dAb ( B7 ) fused to a human Fc domain together with a rabbit polyclonal antibody directed against p53 . Immuno-labeling was revealed using anti-Human Fc-Cy3 and anti-Rabbit-A488 secondary antibodies . ( F ) The hs2dAb antibody directed against HP1α ( A5 ) fused to a human Fc domain was used to stain HeLa cells fixed with paraformaldehyde and permeabilized with 0 . 1% TritonX100 . Cells were co-stained using a polyclonal rabbit antibody directed against HP1α and immuno-labeling was revealed using anti-Human Fc-Cy3 and anti-Rabbit-A488 secondary antibodies . ( scale bar = 10 µm ) . DOI: http://dx . doi . org/10 . 7554/eLife . 16228 . 00810 . 7554/eLife . 16228 . 009Figure 2—figure supplement 1 . Specificity of hs2dAb directed against tubulin and actin . ( A ) HeLa cells were left untreated ( top ) or incubated with 10 µM nocodazole for 90 min at 37°C ( bottom ) . Cells were then permeabilized using 0 . 2% tritonX100 , fixed using methanol ( −20°C , 4 min ) and immunolabelled using an anti-tubulin hs2dAb ( green ) and a polyclonal anti-tubulin antibody ( Red ) . Staining was essentially lost upon nocodazole treatment with only few nocodazole-stable microtubules being labelled . Bar = 10 µm ( B ) HeLa cells were left untreated ( top ) or incubated with 5 µM cytochalasin D for 60 min at 37°C ( bottom ) . Cells were then fixed using paraformaldehyde , permeabilized with saponin and immunolabelled using the anti-actin hs2dAb H2 ( green ) and a polyclonal anti-actin antibody ( Red ) . Staining was strongly reorganized upon cytochalasin D treatment . Bar = 10 µm ( C ) SDS-PAGE of 40 µg sample per well of WI38 whole cell extract was blotted and separated in stripes for each lane . Stripes were then incubated for immuno-detection with various hs2dAb ( lane 1–3: anti-actin hs2dAb; lane 4: non relevant [NR] control; lane 6: anti-tubulin hs2dAb ) directly used from E . coli culture supernatant and further revealed using myc-HRP antibodies . Control monoclonal anti-actin or anti-tubulin were used on stripes 5 and 7 , respectively . DOI: http://dx . doi . org/10 . 7554/eLife . 16228 . 00910 . 7554/eLife . 16228 . 010Figure 2—figure supplement 2 . Specificity of the anti-p53 hs2dAb . ( A ) RPE-1 were left untreated or irradiated with UV light ( 100 J/m2 ) . Cells were then fixed and stained using an anti-p53 hs2dAb together with an anti-p53 rabbit polyclonal antibody . Both the polyclonal and the hs2dAb antibodies detected the strong increase of p53 localization in the nuclei . ( B ) RPE-1 cells stably expressing an shRNA directed against p53 together with GFP were irradiated as in A and stained using the hs2dAb . No labeling was obtained in p53 KD cells . Bar = 10 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 16228 . 01010 . 7554/eLife . 16228 . 011Figure 2—figure supplement 3 . Specificity of the anti-HP1 hs2dAb . ( B ) . ( A ) GFP tagged HP1α , HP1b and HP1γ were expressed in Hela cells . Cells were then lysed and extract separated by SDS-PAGE , transferred on nitrocellulose filters . Immunodetection was carried out using an anti-GFP antibody ( left ) or an anti-HP1 hs2dAb . Although the selection was carried out against was HP1α , the hs2dAb efficiently detected GFP tagged HP1α , HP1β and HP1γ . The band indicated by a star likely represent the endogenous HP1 proteins . ( B ) Cells were transfected as in ( A ) , fixed and analyzed by immunofluorescence using the anti-HP1 hs2dAb . Nuclei were stained using DAPI . GFP tagged HP1α , HP1b and HP1γ were all efficiently detected by the recombinant antibody . Bar = 10 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 16228 . 011 In the next screen , two highly and constitutively expressed components of cell cytoskeleton , tubulin and β-actin , were targeted . Antibodies against tubulin were selected in native conditions ( Nizak et al . , 2003a ) using commercial biotinylated tubulin ( Cytoskeleton ) . After two rounds of selection , 3 out of 40 clones analyzed were shown to detect endogenous tubulin by immunofluorescence ( Figure 2C ) . As expected , staining was lost in cells treated with the microtubule destabilizing drug nocodazole , with only a few stable microtubules being labelled in these conditions ( Figure 2—figure supplement 1A ) . This antibody was recently used to stain microtubules by super-resolution imaging ( Mikhaylova et al . , 2015 ) . Among phage display methods , selection on antigens directly adsorbed on the surface of immunotube is often used as a cheap and straightforward method , despite the low capacity and the strong denaturation imposed by non-specific adsorption . A screen against coated β-actin led to the identification of 16 unique binders positive in phage ELISA . Seven of these antibodies decorated endogenous actin stress fibers by immunofluorescence in MRC5 cells ( Figure 2D ) as well as in other cell lines . Treatment of cells with cytochalasin D disorganizes actin fibers . Accordingly , staining with the hs2dAb was strongly altered ( Figure 2—figure supplement 1B ) . Three of the hs2dAb detected a single band at the molecular weight of β-actin in western blot from HeLa cell extract while one of the anti-tubulin detected a band at the correct molecular weight of tubulin ( Figure 2—figure supplement 1C ) . Actin and tubulin are strongly expressed cellular proteins . Another screen was thus performed to select binders directed against proteins expressed at a lower level . A first screen was carried out against the tumor suppressor p53 protein . The 83 first amino acids of the NP_000537 . 3 isoform were produced in bacteria fused to a SNAP tag , biotinylated in vitro and used as a target in the phage display selection . Among 12 clones positive in phage ELISA , 6 selected hs2dAb were shown to label endogenous p53 in immunofluorescence on A431 cells ( Figure 2E ) . The specificity of the staining was confirmed using RPE-1 cells ( Figure 2—figure supplement 2 ) . A low nuclear staining was observed in normal conditions , with some variability between cells . As expected , the intensity was enhanced upon UV-induced DNA damage . Such an increase was not observed in a RPE-1 cell line stably expressing an shRNA against p53 . A second screen was carried out against the heterochromatin protein HP1α . HP1α was produced in bacteria fused to an avitag to obtain a biotinylated recombinant protein . Biot-HP1α was then immobilized on streptavidin beads and used as a target for 3 rounds of selection . 5 individual hs2dAb were directly identified by immunofluorescence staining of HeLa cells . Figure 2F shows that selected antibodies were efficiently staining endogenous HP1 in the nucleus . Overexpression of the different HP1 variant followed by western blot and immunofluorescence analysis actually suggested that HP1β/γ were also detected by the antibody ( Figure 2—figure supplement 3 ) . Together these results showed that the NaLi-H1 synthetic library can be screened in various conditions against very different purified targets while leading to the rapid identification of diverse and specific binders that can be used in classical antibody-based staining methods . One of the main advantages of full in vitro immunization using display technologies is the control of antigen conformation and concentration . It allows to drive selection towards the desired outcome . For example , selection schemes can be devised to improve the recovery of high affinity binders endowed with low off-rate kinetics ( Lee et al . , 2007 ) , to target specific epitopes ( Even-Desrumeaux et al . , 2014; Vielemeyer et al . , 2009 ) , or to identify conformation sensitive-binders ( Haque et al . , 2011 ) . Recombinant antibody fragment library screening have , for example , provided several binders targeting selectively the active conformation of GTP binding proteins ( Dimitrov et al . , 2008; Nizak et al . , 2003a; Tanaka et al . , 2007 ) . We hypothesized that the NaLi-H1 synthetic library had enough diversity and functionality to enable the identification of selective conformational binders . A subtractive panning was performed to select conformation-specific antibodies directed against small GTPases from the RHO subfamily ( Chinestra et al . , 2012 ) . Small GTPases are molecular switch that cycle between an inactive and an active state when bound to GDP or GTP nucleotides , respectively . Mutant of small GTPases can be designed that adopt stably an active or inactive conformation . A constitutively active ( CA ) mutant RHOA L63 was expressed in HEK293 as a bait then freshly pulled down for panning to preserve its native conformation . To enrich in phage specific for GTP-bound RHOA , a depletion step was introduced from the second round of panning using GDP-bound RHO proteins . After four rounds of selection , clones were analyzed using phage ELISA against either wild type RHOA loaded with GTPγS ( a non-hydrolysable analogue of GTP ) or GDP-loaded RHOA . Forty clones presenting a differential ELISA signal in favor of the GTP loaded RHOA were sequenced . One antibody , represented by clone H12 , represented more than 50% of the population . We analyzed H12 binding specificity by ELISA on several purified RHO proteins expressed as GST fusion in E . coli . H12 recognized the constitutively active mutant RHOA L63 ( RHOA-CA ) which is bound to GTP due to impaired hydrolysis activity . A similar signal was obtained with wild type RHOA loaded with the non-hydrolysable GTP analogue GTPγS . In contrast , no binding was observed to the dominant negative RHOA N19 mutant RHOA-DN nor to GDP-loaded wild type RHOA ( Figure 3A ) . The capacity of H12 to specifically immunoprecipitate GTP-loaded RHOA from mammalian cell extracts was then evaluated in comparison to the standard method to assay RHO activity ( Ren et al . , 1999 ) . This pull down method is based on the RHO binding domain of Rhotekin fused to GST ( GST-RBD ) which is known to bind to the active conformation of RHO GTPase . The hs2dAb H12 bearing a carboxy-terminal CBD ( Chitin-Binding Domain ) was expressed in E . coli and immobilized on chitin beads . These beads were then incubated with HeLa cell extracts pre-treated with either GTPγS or GDP to load small GTPases with the respective nucleotide . The H12 hs2dAb was found to be highly selective of RHO loaded with GTPγS as it was unable to precipitate RHO from GDP loaded extract ( Figure 3B ) . A similarly strong conformation-specificity was found when using H12 for immunofluorescence staining ( Figure 3C ) . HeLa cells expressing the GFP-RHOA constitutively active mutant carrying the mutation Q63L ( GFP-RHOA CA ) or the dominant negative mutated T19N ( GFP-RHOA DN ) , were fixed and stained with H12 hs2dAb . Expression of the dominant negative mutant GFP-RHOA DN , did not lead to an increased signal over the background of un-transfected cells . In contrast , a staining with H12 was correlated with the level of GFP-RHOA CA mutant expression . Note that the signal does not fully overlap GFP fluorescence and appeared stronger at the cell border and in large zone where cell shape is strongly retracted by large bundled actin stress fibers induced by a sustained activation of the RHOA/ROCK pathway ( Mayer et al . , 1999 ) . This CA mutant , like active mutant of many small GTPases related to RAS still need to be activated by guanine nucleotide exchange factors to be loaded with GTP and display the active conformation . Thus , we believe that the H12 staining revealed the active form of this mutant in cells ( Figure 3C ) . All together , these results demonstrated that the H12 hs2dAb is selective for RHO GTPases in their active conformation , highlighting the performance and diversity of the NaLi-H1 library . 10 . 7554/eLife . 16228 . 012Figure 3 . Subtractive selection led to conformational or cell type specific hs2dAbs . ( A–C ) H12 is a conformational hs2dAb binding only to the GTP bound , activated state , of the RHOA GTPase: ( A ) ELISA using the H12 or anti-GST antibodies to reveal recombinant GST-RHOA wild type proteins loaded with either 100 µM GTP gamma S ( Black ) or 1 mM GDP ( White ) , or constitutively active mutant proteins GST-RHOA Q63L ( Grey ) . Means ± SEM . ( B ) A CBD tagged H12 pull down from HeLa cell extract loaded with100 µM GTP gamma S ( GTP ) or with 1 mM GDP as inputs . Western blot reveals RHOA at a similar level in 5% of both input but only on the GTP loaded extract in the CBD-H12 pull down . D5 anti tubulin was used as a negative control and the standard GST-RBD ( RHO binding domain of Rhotekin ) as a positive control of active RHO pull down . ( C ) Immunofluorescence on HeLa cells overexpressing GFP-RHOA CA ( constitutively active ) mutant or GFP-RHOA DN dominant negative mutant . H12 staining detected using a myc tag antibody revealed only cells overexpressing the constitutively active mutant with a pattern stronger at the cell periphery were RHOA activation is high . ( D ) Tumor cell surface subtractive selection scheme . ( E ) ELISA of hs2dAb F7 anti-HER2 on HER2 fused with a rabbit Fc versus binding on rabbit Fc at equimolar concentration . ( F ) hs2dAb F7 anti-HER2 decorated the SKBR3 membrane in immunofluorescence . SKBR3 cells were fixed with 3% paraformaldehyde and stained with F7 revealed by an anti-HisTag ( Sigma ) and an anti-MouseCy3 secondary antibody ( Jackson ) . ( G ) FACS analysis of F7 anti-HER2 on SKBR3 HER2 positive cells versus MCF10A HER2 negative cells . ( Scale bar = 10 µm ) . DOI: http://dx . doi . org/10 . 7554/eLife . 16228 . 01210 . 7554/eLife . 16228 . 013Figure 3—figure supplement 1 . non cropped western blot corresponding to Figure 3B detection RHOA . DOI: http://dx . doi . org/10 . 7554/eLife . 16228 . 013 The use of antibodies to target cells involved in pathologies like cancers or viral infection is one of the most promising therapeutic approach . Antibodies also represent a unique tool to identify novel targets at the cell surface . Using synthetic libraries like the NaLi-H1 library to carry out direct selection against the cell surface of tumor cells would strongly accelerate the identification of specific antigens while allowing extended control over the conditions of selection . A subtractive selection scheme was set up to identify antibodies selectively detecting the surface of breast tumor cells: phages displaying hs2dAbs were first depleted against a reference cell line before being selected against the target one ( Figure 3D ) . As a target cell line , we used the SKBR3 line , which is known to overexpress the HER2 cell surface protein , while the MCF7 cell line , negative for HER2 , was used to pre-adsorb the library . After the third round of bio-panning , 88 clones were analyzed by FACS and 58 were found to be positive when tested on SKBR3 cells and negative on MCF7 cells . Sequencing the 58 positive clones revealed that 15 independent binders had been selected . Although the subtractive selection was not performed on identical cell lines expressing or not HER2 , we observed strikingly that 12 clones out of 15 recognized HER2 exoplasmic domain by ELISA using HER2-Fc as a target antigen ( Figure 3E ) , suggesting that it behaves as a dominant differential epitope . As shown in Figure 4 , these antibodies efficiently detected HER2 at the cell surface by immunofluorescence ( Figure 3F ) or by FACS ( Figure 3G ) . These experiments demonstrated that the NaLi-H1 library will represent a unique tool to discover , in a rapid and cost effective manner , specific antibodies detecting antigens present at the surface of pathological cells . These antibodies may then be used to identify the corresponding target . 10 . 7554/eLife . 16228 . 014Figure 4 . Fluorescent intrabodies tracking endogenous proteins . Intracellular expression of hs2dAb . ( A ) ( top panel ) HeLa cells were co-transfected with GFP-Rab6 and a hs2dAb-mCherry anti-EGFP plasmids . The hs2dAb mCherry anti-EGFP colocalized perfectly with the Rab6 Golgi staining . ( bottom panel ) HeLa cells were co-transfected with Myr-palm-mCherry and a VHH-EGFP anti-mCherry plasmids . The VHH-EGFP anti-mCherry interacted with its target in vivo and colocalized perfectly with the mCherry staining at the plasma membrane . ( B ) SKBR3 cells were transfected with an anti-p53 hs2dAb-mCherry alone ( top panel ) , or together with full length p53-EGFP which concentrated the hs2dAb into the nucleus ( bottom panel ) . ( C ) GFP , used as a control ( top panel ) or a GFP-tagged anti-HP1 hs2dAb ( bottom panel ) were transiently expressed in HeLa cells ( green ) . In contrast to the GFP control , the GFP-tagged anti-HP1 strongly accumulated in the nucleus where it labeled nuclear condensations . ( Scale bar = 10 µm ) DOI: http://dx . doi . org/10 . 7554/eLife . 16228 . 014 Various antibody fragments have long been proposed to represent powerful tools when expressed in cells as intrabodies . Although several studies indeed report efficient use of intrabodies ( reviewed in [Kaiser et al . , 2014; Lobato and Rabbitts , 2003; Stocks , 2005] ) , this is limited to antibody scaffolds that resist to the reducing environment of the cytosol . We evaluated the use of the hs2dAb scaffold to develop intrabodies . Randomly chosen hs2dAbs were fused to a fluorescent protein and observed in living cells . In comparison to our previous experience with scFv or nanobody libraries in which a majority of tested antibodies showed aggregation when expressed in the cytosol as an EGFP-fusion , most of the hs2dAbs tested here gave a monodispersed fluorescence , and very few were showing aggregates . Monodispersed fluorescence of the EGFP has been proposed to be a convenient indicator of stability and potential use as intrabody ( Guglielmi et al . , 2011 ) . Several hs2dAb directed against actin , tubulin , EGFP or p53 were tested for their ability to trace intracellular antigens in living cells . None of the anti-tubulin or anti-actin antibodies tested were found localized on microtubule or microfilament , respectively , in living cells . We reasoned that this poor efficiency may be linked to the antigen denaturating conditions used during the selection of antibodies directed against actin . The existence of a large pool of unpolymerized actin and tubulin may also prevent efficient recruitment on the polymers . In contrast , several anti-GFP and anti-mCherry hs2dAbs were found to efficiently label their targets , like for example GFP-Rab6 or Myr-Palm-mCherry fusion proteins , in living cells ( Figure 4A ) . Similarly , the anti-p53 hs2dAb fused to mCherry were clearly accumulated in the nucleus of SKBR3 cells where endogenous p53 is also localized . This signal was enhanced in cells overexpressing p53-EGFP ( Figure 4B ) . This further confirmed the binding specificity of the hs2dAb anti-p53 while expressed in the reducing cytosol . Effective intrabodies were also obtained against HP1 . Figure 4C ( lower panel ) shows that anti-HP1 expressed as an EGFP tagged protein in HeLa cells localized in the nucleus where it labels condensed structures similar to HP1 usual staining . In contrast , a diffuse cytoplasmic and nuclear staining was obtained using non relevant hs2dAbs fused to EGFP ( Figure 4C , upper panel ) . These results indicate that functional intrabodies can be obtained at high frequency using the NaLi-H1 library . Intrabodies can be used , upon fusion with a fluorescent protein , to track the dynamics of their target in living cells ( see for example Nizak et al . , 2003a ) . Such an application is illustrated in the Video 1 where an hs2dAb directed against mCherry is used to track mCherry-fused Rab6 in living cells . Intrabodies may allow not only to track the dynamics of their cellular target , but also to perturb , or block , their activity . Our results indeed indicate that the H12 antibody was able to perturb endogenous RHO activity when expressed in the cytosol . We could not directly image enrichment of the H12 antibodies in cellular sub-domains in living cells but we observed that H12 may behave as an efficient intrabody carrying out co-immunoprecipitation experiments . H12 carrying a carboxy-terminal myc tag was expressed in HeLa cells together with either the CBD-fused RHOA DN or with the CBD-RHOA CA mutants . H12 was pulled-down by RHOA CA but not by inactive RHOA DN ( Figure 5—figure supplement 1 ) . H12 thus worked as an intrabody and kept its conformation sensitivity in the cytosol . Because RHO GTPases are involved in signaling pathways that promotes the actin cytoskeleton polymerization , we looked at functional effects induced by H12 overexpression . In contrast to un-transfected cells or cells transfected with various non-relevant EGFP fused hs2dAb ( Figure 5A , upper panel ) , we observed that cells expressing H12-EGFP were totally devoid of actin stress fibers ( Figure 5A , lower panel ) . This alteration in actin filament organization was associated with marked changed in cell shape characteristic of loss of intracellular mechanical forces and tension . As RHOA plays a major role in activating myosin II and actin cytoskeleton reorganization , our results suggested that H12 efficiently perturbed RHO-dependent signaling , mimicking the phenotype induced by the C3 exoenzyme RHO inhibitor ( Ridley and Hall , 1992 ) . 10 . 7554/eLife . 16228 . 015Video 1 . mcherry-Rab6 was transiently expressed in HeLa cells together with an anti-mCherry hs2dAb fused to GFP . 24 hr after transfection , cells were imaged using a spinning disk confocal microscope . DOI: http://dx . doi . org/10 . 7554/eLife . 16228 . 01510 . 7554/eLife . 16228 . 016Figure 5 . Targeting cellular proteins using inhibitory antibodies or by functionalizing antibodies to induce protein knockdown . ( A ) HeLa cells expressing transiently a EGFP-tagged non relevant hs2dAB ( top panel ) or EGFP-H12 anti RHO-GTP ( bottom panel ) were fixed 20 hr post transfection and stained using DAPI and Alexa 594 phalloidin to detect actin stress fibers . The H12 hs2dAb induced actin stress fibers disappearance and major cell shape change ( see cells outlined with a dotted line ) ( B ) Protein knockdown of H2B-EGFP mediated by functionalized inhibitory antibodies . HeLa S3 cell stably expressing histone H2B-EGFP were transfected with vectors expressing antibodies fused to an F-box ( F-Ib ) to induce degradation of the targeted cellular antigen . The F-GFP4 VHH ( DegradFP ) was used as a positive control ( top panel ) and a non-relevant hs2dAb as a negative control ( bottom panel ) . F-Ib were expressed using a bi-cistronic vector driving the co-expression of mitochondrial targeted mCherry . Protein interference is analyzed in cells displaying mCherry positive mitochondria ( mitoCherry channel ) . Efficient protein knockdown is obtained using the R3TF3 anti-EGFP intrabody . Note that not all nanobodies can be used as F-Ib because R3TG4 does not induce protein degradation . ( Scale bars = 20 µm ) ( C ) Fluorescence decay measurement of the protein interference assay was quantified by flow cytometry ( 10000 cells analyzed , from 3 independent replicates ) . GFP fluorescence intensity was quantified in the transfected and the untransfected subpopulations for each F-Ib . The ratio of each median of fluorescence ( transfected versus untransfected population ) was calculated as a percentage of GFP fluorescence intensity for one F-Ib . A strong decrease in fluorescence corresponding to protein knockdown was observed with F-GFP4 VHH and F-R3TF3 hs2dAb intrabodies while the non-relevant negative control and R3TG4 did not induce a decrease of fluorescence . ( D ) Cells were analyzed as in C but the cells were incubated in 1 µM MG132 or DMSO for 44 hr after transfection by the different F-Ib and fluorescence intensity was normalized using the non-relevant control . Protein knockdown was inhibited by MG132 . DOI: http://dx . doi . org/10 . 7554/eLife . 16228 . 01610 . 7554/eLife . 16228 . 017Figure 5—figure supplement 1 . Conformational selectivity of the H12 intrabody towards RHOA . GTP . HeLa cells were co-transfected for 24 hr with the anti RHO hs2dAb H12 fused to carboxy-terminal myc tag ( H12-myc ) or a non relevant control ( NR-myc ) that was a negative clone in a panning against FITC together with chitin binding domain ( CBD ) fusion of either the dominant negative mutant RHOA-N19 ( DN ) or the constitutively active mutant RHOA-L63 locked in the GTP bound state ( CA ) . Chitin beads pull down of CBD-RHOA-DN or CBD-RHOA-CA revealed the selective co-precipitation of H12-myc together with the RHO active mutant . The total level of CBD-RHOA or hs2dAb-myc proteins was revealed by loading 5% of the respective input . CBD-RHOA proteins were detected with anti CBD tag and the hs2dAb antibodies with an anti-myc tag antibody . DOI: http://dx . doi . org/10 . 7554/eLife . 16228 . 01710 . 7554/eLife . 16228 . 018Figure 5—figure supplement 2 . Protein knockdown set up using F-Ib degradation with anti-GFP intrabodies . ( A ) Model for F-Ib degradation by fusing Fbox domain to the VHH GFP4 inducing GFP target protein ubiquitiation followed by proteasome-dependent degradation of the target protein . ( B ) Schematic illustration of the bicistronic vectors: an Fbox or NoFbox domain , respectively , is fused to the N terminal part of hs2dAb and a transfection marker MTS-mCherry , labeling mitochondria in red , is co-expressed using an Internal Ribosome Entry Site ( IRES ) . ( C ) Fluorescence visualization HeLa S3 cells stably expressing H2B-GFP and transfected with NoF-VHH GFP4 or F-VHH GFP4 . Degradation by the degradFP was observed in cells expressing F-VHH GFP4 . Scale bar = 10 µm ( D ) Western blot quantification of protein knockdown mediated by F-GFP4 . ( E ) Quantification of GFP fluorescence by flow cytometry in MTS Cherry positive cells . DOI: http://dx . doi . org/10 . 7554/eLife . 16228 . 01810 . 7554/eLife . 16228 . 019Figure 5—figure supplement 3 . non cropped western blot corresponding to Figure 5—figure supplement 1 detection of RHOA , Myc tagged hs2dAb intrabodies , and GAPDH which is not in the main figure . DOI: http://dx . doi . org/10 . 7554/eLife . 16228 . 01910 . 7554/eLife . 16228 . 020Figure 5—figure supplement 4 . non cropped western blot corresponding to Figure 5—figure supplement 1D detection of tubulin , GFP and myc tag . DOI: http://dx . doi . org/10 . 7554/eLife . 16228 . 020 Identification of blocking antibodies is a challenging task and not all functional intrabodies are inhibitory . However , it is possible to functionalize non-blocking intrabodies to inhibit their target function . One approach relies on the ubiquitinylation and degradation of the recognized target as described by Caussinus et al . ( 2011 ) . This approach is based on the fusion of intrabodies to an F-box domain which allows interaction with Skip1 , a member of the SCF complex , an E3 ubiquitin ligase of the complex E1/E2/E3 ubiquitinylation machinery , that targets proteins to proteasome-dependent cellular degradation ( Caussinus et al . , 2011 ) ( Figure 5—figure supplement 2 ) . This approach was efficiently developed to target several EGFP fusion proteins in Drosophila using a single anti EGFP intrabody , named GFP4 , which is a robust and high affinity EGFP llama intrabody originally isolated from an immune library ( Rothbauer et al . , 2006 ) . To get insight into the relative functionality of hs2dAb for such a protein interference approach , several of the anti EGFP hs2dAb selected from the NaLi-H1 library were fused at their amino terminus to the Fbox domain and their efficacy was compared to the efficacy of the Fbox-GFP4 nanobody . To detect cells expressing Fbox-intrabody fusion proteins ( F-Ib ) , we constructed a bicistronic vector driving the co-expression of F-Ib together with a mitochondria-targeted mCherry ( Mito-mCherry ) ( Figure 5B ) . We expressed the F-Ib antibodies in a HeLa clone stably expressing EGFP fused to histone H2B ( Silljé et al . , 2006 ) and looked for EGFP-H2B depletion . As expected , F-GFP4 , also known as degradFP , induced a strong reduction of H2B-EGFP expression as analyzed by western blot ( Figure 5—figure supplement 2 ) . Accordingly , a strong reduction in nuclear fluorescence intensity was observed in cells expressing F-GFP4 ( see mito-mCherry positive cells , Figure 5B; Figure 5—figure supplement 2 ) . No effect was observed when expressing either GFP4 alone or a GFP4 fused to a truncated , nonfunctional , Fbox domain ( Figure 5—figure supplement 2 ) . When anti-EGFP clones selected from the NaLi-H1 library were tested , we observed that some of the hs2dAb that were found to active as fluorescent intrabodies failed to degrade H2B-EGFP when expressed as F-Ib . This highlights the fact that not all intrabodies can efficiently be functionalized with the F-box and that in vivo binding to a target is not the only parameter to consider . However , several hs2dAb anti-EGFP induced a complete disappearance of nuclear H2B-EGFP signal when expressed as F-Ib ( F-R3TF3 , Figure 5B ) while no reduction was observed when using an anti-EGFP that cannot be used as an intrabody ( F-R3TG4 , Figure 5B ) . FACS analysis showed a decreased of fluorescence intensity by as much as 70% ( Figure 5C ) . As expected , this effect was reversed in the presence of a proteasome inhibitor ( Figure 5D ) . Altogether , these experiments show that the hs2dAb scaffold enables the frequent selection of antibodies that can be expressed in the mammalian cell cytoplasm to be used as functional fluorescent or inhibitory intrabodies .
Here , we report the construction of the first large fully synthetic single domain antibody library based on a humanized scaffold derived from llama VHH . A set of robust nanobody scaffolds was first identified using a positive expression screening in E . coli cytosol . One very robust scaffold ( sdAbD10 ) was identified and was used as a base . After introduction of several modifications that aimed at humanizing its primary sequence , we designed the hs2dAb scaffold which is as stable as sdAbD10 while being closer to human VH3 . Our data indicate that the hs2dAb displays partial resistance and/or refolding after treatment for 10 min at 90°C . Using CDR grafting experiments we confirmed the efficacy and the stability of the synthetic scaffold to display CDR regions . Based on our prior experience on phage display libraries , immune or naïve llama VHH libraries ( Monegal et al . , 2012; Olichon and Surrey , 2007 ) or from scFv libraries ( Dimitrov et al . , 2008; Goffinet et al . , 2008; Nizak et al . , 2003a ) we then rationally designed CDR diversity with fixed CDR1 and CDR2 size and four CDR3 sizes ( 9 , 12 , 15 or and 18 amino acids ) . The power of modern gene synthesis approach permits to reach very high genetic diversity while controlling codon bias and cloning features . Fully random codon combination using NNN or NNK trinucleotide cannot prevent stop codon , undesired cysteine or hydrophobic residues to be incorporated , and it does not lead to the controlled probability of amino acid occurrence at a given position . Therefore a more rational design was implemented with defined set of codons for each CDR amino acid position so that it does not mimic natural diversity , in contrast to recently developed Fab synthetic libraries ( Prassler et al . , 2011; Zhai et al . , 2011 ) , but is rather optimized for intrinsic hydrophilicity or solubility . After a large-scale cloning of synthetic fragments , 3 billion independent clones were transformed in the bacteria . Library quality was confirmed by Sanger and next generation sequencing . The library was validated by screening against various targets and in each case specific and highly functional antibodies were obtained ( Table 1 ) . Various selection schemes yielded a large diversity of high affinity and high selectivity binders . Selections were carried out using purified antigen coated on polystyrene , on magnetic beads or directly on the cell surface . In many cases , two rounds of selection were sufficient to obtain selective binders . We usually analyzed only 80 randomly picked clones because the diversity of specific binders was systemically high . Only a few selections led to antibodies usable in western blotting ( anti-actin , ant-tubulin ) probably because most screenings were done using natively folded targets . Accordingly , selected hs2dAb performed very well in other conventional immunoassays like ELISA , FACS , immunoprecipitation or immunofluorescence . Affinity measurements done by surface plasmon resonance revealed KD values in the order of 10 nanomolar and up to 50 picomolar . Such high affinities are rather good and usually rarely observed for monovalent binders obtained without in vivo immunization or in vitro affinity maturation steps ( Figure 6; Table 2 ) . 10 . 7554/eLife . 16228 . 021Table 1 . Summary of screenings showing the number of unique clones giving positive signal . ( ND means non determined ) DOI: http://dx . doi . org/10 . 7554/eLife . 16228 . 021Positive clonesAntigenPhage ELISAIF/FACSIntrabodyRounds of panningGFP3710/ND4/102mCherryND6/ND2/63TubulinND3/ND0/32Actin167/ND1/73p53126/ND2/62RHOA-GTP248/ND3/84Her265/10ND310 . 7554/eLife . 16228 . 022Figure 6 . Affinity determination . Single cycle kinetics analysis was simultaneously performed on immobilized His fusion VHH antibodies ( 250–300 RU ) , with five injections of analytes ( EGFP , HER2 , RHOAQ63L and mCherry ) at 3 . 125 nM , 6 . 25 nM , 12 . 5 nM , 25 nM , and 50 nM . Analytes injections lasted for 120 s each and were separated by 10 s dissociation phases . At this time of buffer exchange , a slight refraction index discrepancy between the sample and the flow buffer can induce a drop in resonance unit . This common bulk effect , which is clearly visible on sensorgrams with a smaller scale range on the RU axis ( ie: R3SE4 , R2TB5 ) , does not affect the measurement of off-rate constant . Off-rate constant was calculated from an extended dissociation period of 10 min following the last injection according to the single cycle kinetics method . Each sensorgram ( expressed in RUs as a function of time in seconds ) represents a differential response where the response on an empty reference channel ( Fc1 ) was subtracted . The red curves correspond to the data and the black curves represent the fit done by the BIAevaluation software . Note that the fitted curve is almost identical to the data curve in some cases like for example the RHOA Q63L or the HER2 binding measurement . DOI: http://dx . doi . org/10 . 7554/eLife . 16228 . 02210 . 7554/eLife . 16228 . 023Table 2 . Binding affinities of 9 selected hs2dAb fused to a 6HIS tag measured by surface plasmon resonance single cycle kinetics method . Dissociation equilibrium constant KD corresponds to the ratio between off-rate and on-rate kinetic constant Koff/Kon . Non relevant hs2dAb were used as negative controls and gave no detectable binding signal . A positive control endowed with subnanomolar affinity , the GFP binder VHH-GFP4 , was analyzed in parallel to the GFP hs2dAbs . A KD of 1 . 55–10 M was measured for VHH-GFP4 which is similar to published values . The binding properties of the conformational H12 hs2dAb to the GTP loaded RHOA subfamily were measured using the L63 or L61 constitutively active mutants of RHO , RHOB , RHOC , RAC1 and CDC42 related small GTPases , as well as the negative mutant T19N of RHOA . ( 'no' means no detectable binding ) . DOI: http://dx . doi . org/10 . 7554/eLife . 16228 . 023hs2dAb-6xHisAntigenkon ( M−1 s−1 ) koff ( s−1 ) KD ( M ) R2TB5 anti-GFPGFP1 . 24 10+63 . 05 10−42 . 45 10−10R3SD1 anti-GFPGFP7 . 07 10+56 . 68 10−49 . 44 10−10R3SE4 anti-GFPGFP1 . 45 10+55 . 83 10−44 . 01 10−9Llama VHH GFP4GFP2 . 99 10+54 . 65 10−51 . 55 10−10D4 anti-Her2Her21 . 79 10+51 . 63 10−49 . 11 10−10A10 anti-Her2Her21 . 66 10+44 . 88 10−52 . 94 10−9B9 anti-CherrymCherry6 . 14 10+41 . 68 10−42 . 74 10−9E4 anti-CherrymCherry6 . 57 10+42 . 03 10−43 . 10 10−9B3 anti-CherrymCherry6 . 19 10+42 . 71 10−44 . 38 10−9H12 anti-RHO . GTPRHOA Q63L4 . 81 10+51 . 28−42 . 65 10−10H12 anti-RHO . GTPRHOB Q63L2 . 24 10+53 . 59−41 . 57 10−9H12 anti-RHO . GTPRHOC Q63L1 . 12 10+65 . 41−54 . 79 10−11H12 anti-RHO . GTPRHOA T19NnononoH12 anti-RHO . GTPRAC1 Q61L7 . 53 10+52 . 55−43 . 3 10−10H12 anti-RHO . GTPCDC42 Q61Lnonono The NaLi-H1 library thus enables the rapid selection of diverse and highly functional binders . Because it is a fully synthetic , non immune , library , it does not depend on animal experimentation , it is not limited by natural immunogenicity or toxicity of antigens and allows to develop and adjust the selection without ethic consideration . In addition , because all steps are carried out in vitro , conditions can be tightly controlled . This allowed to develop powerful differential selection and to identify conformation-specific antibodies . This also allowed to directly screen for antibodies directed against antigens specifically present at the surface of a particular cell type . Such a differential selection will be a powerful approach to identify novel antigen at the surface of tumor or infected cells . Such antibodies may also represent powerful tools for diagnostic and therapeutic applications to target cells in human pathologies . For example , after dimerization using Fc domains , hs2dAb antibodies may be used to target tumor cells and benefit from antibody-dependent cell-mediated cytotoxicity ( ADCC ) for example . They may also be used directly as the smallest antibody-derived domain naked as an agonist or antagonist or armed for enhanced toxicity . Similarly , it may be labeled using radioactive compounds ( e . g . 99mTc , 111In , 64Cu ) and used to image tumors in patient using positron emission tomography . Altogether , he NaLi-H1 library may accelerate the identification of novel potent tools to be used in human clinical applications . The synthetic scaffold we defined was based on the selection of a set of VHH able to fold properly in the bacteria cytosol . The goal was not only to define a robust scaffold that would be efficiently produced without aggregation but also to allow frequent selection of functional intrabodies . Intrabodies have been isolated from various antibody libraries ( Nizak et al . , 2003a; Rothbauer et al . , 2006; Tanaka and Rabbitts , 2010; Vercruysse et al . , 2010 ) as well as other protein scaffold like Darpin ( Tamaskovic et al . , 2012 ) or FN3 ( Koide et al . , 2012 ) which are devoid of cysteine . A peculiar feature of the NaLi-H1 library is that it is based on a humanized nanobody-like robust scaffold , stable in a reducing environment , while it still contains the two canonical cysteine residues . Stabilized nanobodies , human single domain scaffolds ( Christ et al . , 2007; Saerens et al . , 2005 ) and libraries ( Goldman et al . , 2006; Mandrup et al . , 2013 ) were reported before . However , to our knowledge , no synthetic library producing at high frequency functional intrabodies was developed before based on such stabilized scaffolds . Almost every hs2dAb antibodies we expressed in mammalian cytosol showed no sign of aggregation , which further supported the idea that the synthetic scaffold we designed is robust and highly resistant to reduction . Previous studies showed that functional intrabody identification often relied on additional steps of selection in a protein–protein interaction reporter system such as PCA ( Koch et al . , 2006 ) or bacterial 2 hybrid ( Pellis et al . , 2012 ) , yeast IACT ( Tanaka and Rabbitts , 2010 ) or F2H assays ( Zolghadr et al . , 2008 ) . Using the NaLi-H1 library , we observed that without using particular selection schemes , functional intrabodies were frequently obtained . Although we did not formally compare the NaLi-H1 library to previous llama naïve or semi synthetic libraries , the functionality of selected hs2dAb was compared to a sub-nanomolar affinity intrabody , the GFP4 nanobody , which has been extensively used ( Caussinus et al . , 2011; Kirchhofer et al . , 2010 ) . We observed by monitoring its signal-to-noise ratio and by using it in a protein knockdown assay that NaLi-H1 can provide highly functional hs2dAb which appeared as good as intrabodies from immune libraries . Intrabodies can be used in several applications like tracking of intracellular dynamics of endogenous proteins ( Nizak et al . , 2003a , 2003b; Rothbauer et al . , 2006 ) but the most appealing application is to use them for rapid protein inactivation in living cells . Intrabodies may be used to directly block their target proteins in cells . One of the conformation-sensitive antibody that was selected can be used to inhibit active RHO GTPase signaling in living cells and is as potent as the C3 exoenzyme toxin . Only few intrabodies have been described to be intrinsically inhibitors of protein activity ( Haque et al . , 2011; Vercruysse et al . , 2010 ) , and our results suggest that the NaLi-H1 library may enable rapid selection of inhibitory antibodies . The next challenge will be to select conformational sensors specifically directed against a particular member of closely related RHO subtypes ( RHOA/B/C ) which share more than 90% similarity in primary sequence . But in any case , our results show that the NaLi-H1 library allows the selection of efficient , conformation-specific , inhibitory intrabodies . Another way intrabodies may be used to inactivate their targets in living cells is to fuse the intrabody to a dominant inhibitory domain . Following an idea pioneered by Affolter and colleagues ( Caussinus et al . , 2011 ) , we showed here that intrabodies selected from the NaLi-H1 library can be fused to a proteasome-targeting domain to impose the specific degradation of their respective targets . This protein interference approach was validated using anti-EGFP hs2dAb and we believe that this approach will bring disruptive tools to generate rapid protein knockdown both in cell culture and in the animal . In summary , we have designed a novel nanobody scaffold endowed with improved stability and created a highly diverse library , the NaLi-H1 library , that was successfully screened to identify highly functional binders directed against very diverse targets . We believe that this library will allow the fast , and fully in vitro , identification of immunological tools usable both for fundamental and medical applications .
Artificial gene synthesis ( Mr Gene , GmbH , Germany ) composed of a 6His-Tag and a triple c-myc Tag was inserted into the pHEN2 phagemid vector ( Griffin 1 . library ) between NotI and BamHI sites . CcdB gene from pENTR4 vector ( Invitrogen - ThermoFisher Scientific , France ) was inserted into the pHEN2 vector between NcoI and NotI sites . This vector allows to express antibody fragments in fusion , upstream , with the pelB leader to drive secretion in the periplasm and downstream with the PIII protein of M13 phages . An amber stop codon is present between the antibody and the pIII . This stop codon is partially suppressed in SupE E . coli . For expression and purification of dimeric antibodies , hs2dAb were inserted in vectors derived from pFuse ( Invivogen , France ) as described in Moutel et al . ( 2009 ) . For intrabody expression in mammalian cells , hs2dAb were digested by NcoI and NotI and ligated into the pIb-mEGFP , pEGFP or the pmCherry vectors ( Clontech - Takara , USA ) . ( See the Appendix for more details ) . Previously selected VHHs from naïve or immune libraries were subcloned into pAOCAT ( Monegal et al . , 2012 ) using the NcoI and NotI restriction sites . Chloramphenicol resistance assay was performed using BL21 ( DE3 ) cells transformed with the pAOCAT-VHH fusion constructs . ( See the Appendix for more details ) . Details about the construction of the library can be found in the Appendix . In short , a synthetic design was ordered based on a statistic analysis of the diversity found in natural VHH and aiming at reducing hydrophobicity at some position . The size of the CDR1 and CR2 was fixed at 7 amino acids while 4 sizes of CDR3 were chosen ( 9 , 12 , 15 and 18 amino acids ) . Large -scale PCR was then carried out ensuring that at least 1010 DNA molecules were used as a matrix . Fragments were then cut and inserted into the pHEN2-3myc plasmid . The ligated DNA material was used to transform electrocompetent E . coli TG1 cells ( Lucigen Corp . , Middleton , United States ) . Serial dilution was used to count the total number of bacteria transformed . A potential diversity of 3 × 109 was calculated . Transformed bacteria plated on 430 2xYT-ampicillin agar dishes ( 140 mm ) , grown overnight at 37°C , scrapped and stored in 30% of glycerol at −80°C . IonTorrent sequencing library was prepared with the Ion Plus Fragment Library kit for AB Library Builder System ( Life Technologies - TermoFisher Scientific , France ) following manufacturer's instructions and was controlled on the Agilent 2100 Bioanalyzer ( Agilent Technologies , France ) with the High Sensitivity DNA Kit ( Agilent Technologies ) . The sequencing template was prepared by emulsion PCR with the Ion OneTouch 2 system and the Ion PGM Template OT2 400 Kit ( Life Technologies ) . Sequencing was performed on a IonTorrent Personal Genome Machine using the Ion PGM Sequencing 400 Kit and a 314v2 Ion chip ( Life Technologies ) . Human βActin was purchased from Sigma-Aldrich ( France ) . RHOA GTPase fused to either an amino terminal Chitin Binding Domain or a streptactin binding peptide were produced in HEK293 cells . EGFP ( as mCherry ) in fusion with a streptavidine binding peptide ( SBP ) were produced through in vitro translation system ( Roche Life Science , France ) and used directly for screening without the need for purification . Biotinylated Tubulin was purchased from Cytoskeleton , Inc . ( Denver , United States ) . For p53 , the 83 first amino acids of the NP_000537 . 3 isoform were produced in bacteria with a SNAP and His Tag , purified using Talon resin ( Takara - Clontech ) and biotinylated in vitro . HP1α was produced in bacteria with an avitag and a His Tag , and purified using Talon resin . For HER2 , the natural receptor was used as membrane protein target on SKBR3 cells . For more details see the Appendix for more details . Screening for ßactin was performed by panning in immunotubes as described ( Marks et al . , 1991 ) . Screening for EGFP , Tubulin and p53 were performed in native condition as described ( Nizak et al . , 2005 ) . Screening for HER2 was performed on surface cells as described ( Even-Desrumeaux and Chames , 2012 ) . Screening on RHO was performed in native condition . ( See the Appendix for more details ) . Individual clones were screened by monoclonal phage ELISA as described . ( See the Appendix for more details ) . After boiling in SDS-PAGE loading buffer , the samples were separated on a 12% SDS-PAGE and transferred to nitrocellulose membranes ( Whatman GmbH , Germany ) . Membranes were blocked in 3% non-fat milk-PBS with 0 . 2% Tween 20 for 1 hr at room temperature or overnight at 4°C . unpurified hs2dAb were used at 1/100 from culture supernatant and added to the membranes with an anti-hisTag antibody at 1/3000 ( Sigma-Aldrich ) for 90 min . Blots were then washed and incubated 1 hr with secondary anti-Mouse HRP labeled antibodies ( diluted at 1/10000 in PBS 0 . 1% Tween 20 ) ( Jakson ImmunoResearch Laboratories ) . After 5 washes with PBS 0 . 1% Tween 20 , secondary antibodies were then revealed using the SuperSignal chemoluminescent reagent ( Pierce ) and Hyperfilm ECL ( GE HealthCare ) . For RHO-GTP pull down , the primary anti RHOA mAb was used ( Cell Signaling Technology; 1/1000 ) . For protein knockdown experiments , 500 000 of transfected cells ( mCherry positive cells ) were sorted with a MoFlo Astrios flow cytometer ( Beckman Coulter ) . Cells were lysed with SDS-Tris lysis buffer ( Tris pH7 . 4 10 mM , SDS 1% supplemented with phosphatase and protease inhibitors ) . 20 µg of cell extracts were separated on 12 . 5% SDS-PAGE and electro transferred onto PVDF membranes . Blots were probed with a rabbit polyclonal anti-EGFP full length ( Santa Cruz , sc-8334 , 1:500 ) , a mouse monoclonal anti-α-tubulin ( Sigma , T5168 , 1:25000 ) and an anti-myc HRP antibody ( Novus Biologicals , NB600-341 , 1:40000 ) . Detection was performed using peroxydase conjugated secondary antibodies and Pierce ECL Western Blotting Substrate ( Thermo Scientific Pierce ) . Immunofluorescence screenings were performed on HeLa cells as described before ( Nizak et al . , 2005 ) . ( See the Appendix for more details ) . Hela or HeLa S3 H2B-EGFP Cells cultured on coverslips were transfected according to the CaPO4 or jet prime procedure with 1 µg DNA per well ( 24 wells plate ) or 10 µg DNA ( 10 cm2 diameter dish ) . Cells can be observed from 12 hr post-transfection on . For HER2 immunoassay , cell surface staining were performed in phosphate-buffered saline ( PBS ) supplemented with 1% SFV . 100 µL of supernatant ( 80 µL phages + 20 µL PBS/milk1% ) were incubated on 1 . 105 cells for 1 hr on ice . Phage binding was detected by a 1:300 dilution of anti-M13 antibody ( GE healthcare , France ) for 1 hr on ice followed by a 1:1000 dilution of PE-conjugated anti-Mouse antibody ( BD Bioscience , France ) for 45 min . Samples were analyzed by flow cytometry on a FACSCalibur using CellQuest Pro software ( BD Biosciences , France ) . In the protein knockdown experiments , 48 hr after transfection , at least 10000 HeLa S3 H2B-GFP cells were analyzed on a MoFlo Astrios flow cytometer ( Beckman Coulter France S . A . S ) for their GFP fluorescence intensity . This fluorescence was analyzed in mCherry transfected cells and non transfected cells . Flow cytometry data were analyzed with Kaluza software ( Beckman Coulter ) . 1 µM of proteasome inhibitor MG132 ( Sigma-Aldrich ) was used in the cell growth medium for 48 hr . Values reported represent median ± standard deviation ( SD ) of at least three independent experiments . p values were calculated with GraphPad Prism 6 ( RRID:SCR_002798 ) using a Student’s t test . **p<0 . 01; ***p<0 . 001; ****p<0 . 0001 . All binding studies based on SPR technology were performed on BIAcore T200 optical biosensor instrument ( RRID:SCR_008424 , GE Healthcare ) . Capture of single domain Hs2dAb-6xHis was performed on a nitrilotriacetic acid ( NTA ) sensorchip in HBS-P+ buffer ( 10 mM Hepes pH 7 . 4 , 150 mM NaCl , and 0 . 05% surfactant P20 ) ( GE Healthcare ) . The four flow cells ( FC ) of the sensorchip were used: one ( FC 1 ) to monitor nonspecific binding and to provide background corrections for analyses and the other three flow cells ( FC 2 , 3 , and 4 ) containing immobilized Hs2dAb-6xHis for measurement . For immobilization strategies , the four flow cells were loaded with nickel solution ( 10 μL/min for 60 s ) in order to saturate the NTA surface with Ni2+ and an extra wash using running buffer containing 3 mM EDTA after the nickel injection . Each His-tagged hs2dAb in running buffer was injected in flow cells at a flow-rate of 10 μL/min . The total amount of immobilized hs2dAb-6xHis was 250–300 resonance units . ( RUs; 1 RU corresponds approximately to 1 pg/mm2 of protein on the sensor chip ) . A Single-Cycle Kinetics ( SCK ) analysis to determine association ( on-rates ) , dissociation ( off-rates ) and affinity constants ( kon , koff and KD respectively ) was carried out . SCK method prevents potential inaccuracy due to sensorchip regeneration between cycles which are necessary in the conventional multiple cycle kinetics ( MCK ) ( Trutnau , 2006 ) . SCK binding parameters are evaluated for each injection according to the tools and fit models of the BIAevaluation software , giving similar values than MCK . As hs2dAb were smaller proteins than their respective antigens , hs2dAbs were captured on the sensorchip while the recombinant antigens were used as analytes . Analytes were injected sequentially with increased concentrations ranging between 3 . 125 nM to 50 nM in a single cycle without regeneration of the sensorship between injections . Binding parameters were obtained by fitting the overlaid sensorgrams with the 1:1 . Langmuir binding model of the BIAevaluation software version 1 . 0 . | Antibodies are proteins that form part of an animal’s immune system and can identify and help eradicate infections . These proteins are also needed at many stages in biological research and represent one of the most promising tools in medical applications , from diagnostics to treatments . Traditionally , antibodies have been collected from animals that had been previously injected with a target molecule that the antibodies must recognize . An alternative strategy that uses bacteria and bacteria-infecting viruses instead of animals was developed several decades ago and allows researchers to obtain antibodies more quickly . However , the majority of the scientific community view these “in vitro selected antibodies” as inferior to those produced via the more traditional approach . Moutel , Bery et al . set out to challenge this widespread opinion , using a smaller kind of antibody known as nanobodies . The proteins were originally found in animals like llamas and camels and are now widely used in biological research . One particularly stable nanobody was chosen to form the backbone of the in vitro antibodies , and the DNA that encodes this nanobody was altered to make the protein more similar to human antibodies . Moutel , Bery et al . then changed the DNA sequence further to make billions of different versions of the nanobody , each one slightly different from the next in the region that binds to the target molecules . Transferring this DNA into bacteria resulted in a library ( called the NaLi-H1 library ) of bacterial clones that produce the nanobodies displayed at the surface of bacteria-infecting viruses . Moutel , Bery et al . then screened this library against various target molecules , including some from tumor cells , and showed that the fully in vitro selected antibodies worked just as well as natural antibodies in a number of assays . The in vitro antibodies could even be used to track , or inactivate , proteins within living cells . The NaLi-H1 library will help other researchers obtain new antibodies that bind strongly to their targets . The approaches developed to create the library could also see more people decide to create their own synthetic libraries , which would accelerate the identification of new antibodies in a way that is cheaper and requires fewer experiments to be done using animals . These in vitro selected antibodies could help to advance both fundamental and medical research . | [
"Abstract",
"Introduction",
"Results",
"Discussion",
"Materials",
"and",
"methods"
] | [
"cell",
"biology",
"tools",
"and",
"resources",
"immunology",
"and",
"inflammation"
] | 2016 | NaLi-H1: A universal synthetic library of humanized nanobodies providing highly functional antibodies and intrabodies |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.