1. Neuroscience
Download icon

Homeostatic plasticity fails at the intersection of autism-gene mutations and a novel class of common genetic modifiers

  1. Özgür Genç
  2. Joon-Yong An
  3. Richard D Fetter
  4. Yelena Kulik
  5. Giulia Zunino
  6. Stephan J Sanders
  7. Graeme W Davis  Is a corresponding author
  1. Department of Biochemistry and Biophysics Kavli Institute for Fundamental Neuroscience University of California, San Francisco, United States
  2. Department of Psychiatry UCSF Weill Institute for Neurosciences University of California, San Francisco, United States
  3. School of Biosystem and Biomedical Science, College of Health Science, Korea University, Republic of Korea
Research Article
  • Cited 0
  • Views 1,221
  • Annotations
Cite this article as: eLife 2020;9:e55775 doi: 10.7554/eLife.55775

Abstract

We identify a set of common phenotypic modifiers that interact with five independent autism gene orthologs (RIMS1, CHD8, CHD2, WDFY3, ASH1L) causing a common failure of presynaptic homeostatic plasticity (PHP) in Drosophila. Heterozygous null mutations in each autism gene are demonstrated to have normal baseline neurotransmission and PHP. However, PHP is sensitized and rendered prone to failure. A subsequent electrophysiology-based genetic screen identifies the first known heterozygous mutations that commonly genetically interact with multiple ASD gene orthologs, causing PHP to fail. Two phenotypic modifiers identified in the screen, PDPK1 and PPP2R5D, are characterized. Finally, transcriptomic, ultrastructural and electrophysiological analyses define one mechanism by which PHP fails; an unexpected, maladaptive up-regulation of CREG, a conserved, neuronally expressed, stress response gene and a novel repressor of PHP. Thus, we define a novel genetic landscape by which diverse, unrelated autism risk genes may converge to commonly affect the robustness of synaptic transmission.

Introduction

Autism Spectrum Disorder (ASD) is a polygenic disorder with a complex underlying genetic etiology (Bourgeron, 2015). Advances in whole genome sequencing and genome-wide association studies have dramatically expanded our understanding of the genetic architecture of ASD. In particular, the identification of rare de novo mutations that confer high risk for ASD has generated new molecular insight (De Rubeis et al., 2014; Iossifov et al., 2014; Sanders et al., 2015). Yet, even in cases where a rare de novo mutation confers risk for ASD, additional processes are likely to contribute to the ASD phenotype (Leppa et al., 2016; Peter et al., 2019) including the engagement of adaptive physiological mechanisms (Gaugler et al., 2014; Gibson, 2009; Hartman et al., 2001; Hou et al., 2019; Kitano, 2007; Plomp et al., 1992; Sackton and Hartl, 2016; Sardi and Gasch, 2018; Bourgeron, 2015).

Homeostatic plasticity, in particular, has garnered considerable attention as an adaptive physiological process that might be relevant to the phenotypic penetrance of ASD mutations (Antoine et al., 2019; Bourgeron, 2015; Mullins et al., 2016; Nelson and Valakh, 2015; Ramocki and Zoghbi, 2008). Yet, very little is known at a mechanistic level regarding the interface of homeostatic plasticity and ASD genetics. There remains ongoing debate regarding whether homeostatic plasticity is normally induced or whether it is impaired in the context of rare de novo mutations that confer risk for ASD (Antoine et al., 2019; Bourgeron, 2015; Ramocki and Zoghbi, 2008). And, there is no mechanistic information regarding how rare de novo mutations that confer risk for ASD might be connected to the signaling mechanisms that are essential for the induction and expression of homeostatic plasticity.

It is well established that homeostatic signaling systems function throughout the central and peripheral nervous systems to stabilize neural function following a perturbation that can be of genetic, immunological, pharmacological or environmental origin (Davis, 2006; Marder, 2011; Turrigiano, 2011). Evidence for this has accumulated by measuring how nerve and muscle respond to the persistent disruption of synaptic transmission, ion channel function or neuronal firing. In systems ranging from Drosophila to human, cells have been shown to restore baseline function in the continued presence of these perturbations by rebalancing ion channel expression, modifying neurotransmitter receptor trafficking and modulating neurotransmitter release (Davis, 2013; Hengen et al., 2013; Maffei and Fontanini, 2009; Watt and Desai, 2010). There is evidence that homeostatic signaling systems function at the level of individual cells and synapses (Davis, 2013). There is also evidence that homeostatic signaling systems influence the function of neural circuitry (Deeg and Aizenman, 2011; Hengen et al., 2013; Maffei and Fontanini, 2009; Nelson and Valakh, 2015).

We set out to determine whether there exists a molecular interface between mutations in ASD gene orthologs in Drosophila and the induction or expression of presynaptic homeostatic plasticity. Presynaptic homeostatic plasticity (PHP) is an evolutionarily conserved form of homeostatic plasticity, observed in Drosophila, mice and humans (Davis, 2013). PHP has been documented at both central and peripheral synapses in response to differences in target innervation (Liu and Tsien, 1995) altered postsynaptic excitability (Davis, 2006; Marder and Goaillard, 2006; Mullins et al., 2016), following chronic inhibition of neural activity (Kim and Ryan, 2010; Zhao et al., 2011) and following disruption of postsynaptic neurotransmitter receptor function (Henry et al., 2012; Jakawich et al., 2010). The mechanisms of PHP have a remarkable ability to modulate and stabilize synaptic transmission, with an effect size that can exceed 200% (Müller and Davis, 2012; Ortega et al., 2018).

Many of the rare de novo mutations that confer high risk for ASD are considered to be heterozygous loss of function (LOF) mutations (Bourgeron, 2015; De Rubeis et al., 2014; Iossifov et al., 2014; Sanders et al., 2015). Therefore, we examine the phenotype of heterozygous LOF mutations in five different ASD gene orthologs. We make several fundamental advances. First, we demonstrate that these individual heterozygous LOF mutations have no overt effect on baseline transmission or PHP. However, we demonstrate that PHP is sensitized to failure. Next, we sought to define the molecular mechanisms that connect ASD gene orthologs to the mechanisms of PHP. A genome-scale screen and subsequent systems-genetic analyses yielded unexpected insight. We do not simply identify genes that, when mutated, enhance the phenotype of individual ASD gene mutations. We discovered genes that, when their function is diminished by heterozygous LOF mutations, commonly modify multiple ASD gene orthologs, causing a selective failure of homeostatic plasticity. Thus, we define the first class of common phenotypic modifiers of ASD genes in any system. Finally, we do not stop with the identification of a novel class of ASD gene modifiers. We proceed to characterize how homeostatic plasticity fails in one such condition. The mechanism we discovered is also unexpected and illuminates the complexity by which double heterozygous gene-gene interactions can generate a cellular or organismal phenotype. We demonstrate maladaptive, enhanced expression of a gene known as Cellular Repressor of E1A Stimulated Genes (CREG), a gene that is conserved from Drosophila to human and expressed in the brain (Yang et al., 2011).

Taken together, we define a novel, unexpected genetic architecture that connects heterozygous LOF mutations in ASD-associated gene orthologs with the mechanisms of homeostatic plasticity. In particular, the observation that PHP is commonly sensitized by multiple, different ASD genes, and the fact that we identify and characterize common phenotypic modifiers of five different ASD genes, defines a novel means by which a diversity of ASD-associated risk genes may converge to affect synaptic transmission. We propose that this information may be relevant to new therapeutic approaches that might someday modify ASD phenotypic severity, regardless of the underlying genetic mutation(s) that confer risk for ASD.

Results

We began an investigation of ASD gene orthologs in Drosophila by acquiring heterozygous null mutations in five genes; RIMS1, CHD8, CHD2, WDFY3 and ASH1L (Figure 1A; Note: throughout we use the human nomenclature). Heterozygous null mutations were analyzed, as opposed to use of RNAi-mediated gene knockdown, in order to more precisely reflect the proposed genetic perturbations in human.

Figure 1 with 2 supplements see all
Heterozygous ASD gene mutations do not affect baseline transmission or PHP.

(A) Schematic of the Drosophila locus for CHD8, ASH1L, CHD2, WDFY3 and RIMS1 with gene disruptions indicated. (B) Representative EPSP and mEPSP traces for indicated genotypes (+ / - PhTx for each genotype, left traces and right traces respectively) (C–D) Quantification of mEPSP amplitude (C) and EPSP amplitude (D) in the absence and presence of PhTx (open and filled bars respectively). (E) The percent change of mEPSP and quantal content as indicated, comparing the presence and absence of PhTx for each genotype with Student’s t-test (two tail), *p<0.05, **p<0.01. Sample sizes for data reported (C–E) are as follows (n reported for each genotype -/+ PhTx): wild type: n = 36/47; CHD8/+: n = 7/8; ASH1L/+: n = 15/25; WDFY3/+: n = 8/7; CHD2/+: n = 8/19; RIMS1/+: n = 20/30. (F–H) Scatter plots of quantal content (y axis) versus mEPSP amplitude (x axis) for wild type (left), RIMS1/+ mutant (middle) and the CHD8/+; RIMS1/+ double heterozygous mutant. Each symbol represents an individual muscle recording. Inset: representative traces (+ / - PhTx). Exponential data fit (black line, R2-value inset, calculated based on a linear fit). Dashed lines encompass 95% of all data (absent in (H) for clarity). Below each graph (F–H), boxes display percent PHP (+ / - PhTx for each genotype), statistical values compared to baseline (H).

All five of these genes are considered high confidence ‘category 1’ ASD-associated genes based on SFARI Gene (Simons Simplex Collection, 2020). All five of these genes have clear Drosophila orthologs. Further, we demonstrate that all five genes are expressed in Drosophila third instar motoneurons based on a Patch-Seq analysis of gene expression (Figure 1—figure supplement 1). The five ASD gene orthologs were also chosen to reflect a broad range of biological activities that are associated with the numerous ASD-associated genes identified to date. The RIMS1 gene is a synaptic scaffolding protein that localizes to and organizes sites of neurotransmitter release, termed active zones. The CHD8 and CHD2 genes encode chromatin remodeling factors that localize to the cell nucleus. WDFY3 encodes a phosphatidylinositol 3-phosphate-binding protein and regulator of autophagy and intracellular signaling. ASH1L encodes a member of the trithorax group of transcriptional activators and is found in the cell nucleus. A supplemental table (Supplementary file 1, Supplemental Table S1) includes known disease associations for each of these five human genes, and links to web-based genetic and genomic resources. A survey of biochemical and genetic interaction networks in Drosophila demonstrates no known interactions among these five genes (Flybase). In humans, there appear to be no known direct biochemical interactions among these genes. Yet, heterozygous LOF mutations in each of these genes are associated with risk for ASD in humans.

Heterozygous ASD gene mutations have normal synaptic transmission and PHP

We analyzed baseline neurotransmission and presynaptic homeostatic plasticity (PHP) at the Drosophila neuromuscular junction (NMJ) as a model glutamatergic synapse. At the Drosophila NMJ, PHP is induced by application of sub-blocking concentrations of the postsynaptic glutamate receptor antagonist philanthotoxin-433 (PhTx; 5–10 µM), diminishing the average postsynaptic depolarization caused by the release of single synaptic vesicles (miniature excitatory postsynaptic potential; mEPSP). Decreased mEPSP amplitude initiates a potentiation of presynaptic neurotransmitter release that precisely offsets the magnitude of the PhTx perturbation and, thereby, maintains evoked excitatory postsynaptic potential amplitude (EPSP) at baseline levels prior to the application of PhTx (Figure 1B–E; Davis, 2013; Frank et al., 2009).

First, we characterized baseline synaptic transmission and the rapid induction of PHP in heterozygous null mutations of all five ASD-associated genes, defined above. We find no significant change in baseline neurotransmission, including average mEPSP amplitude, average EPSP amplitude and quantal content (Figure 1B–E). Following application of PhTx, we find that heterozygous null mutations in all five ASD gene orthologs do not alter the expression of PHP (Figure 1B–E). Specifically, PhTx significantly diminished the average mEPSP amplitude in each heterozygous mutant and induced a statistically significant increase in quantal content that restored EPSP amplitudes toward wild type values. We conclude that all five heterozygous mutations express normal PHP.

Genetic interaction of RIMS1 with either CHD8, ASH1L or CHD2 impairs PHP

Tests of genetic interaction are commonly used to determine if two genes have a function that converges on a specific biological process. While genetic interactions cannot be interpreted to reflect participation in a linear signaling pathway, such an analysis can define signaling relationships among genes that are independent of whether the encoded proteins interact biochemically. Thus, genetic interactions have been a powerful means to explore new signaling systems in model organisms, an approach that is being increasingly utilized in cancer biology (Ashworth et al., 2011Mair et al., 2019Chan and Giaccia, 2011O'Neil et al., 2017; Baetz et al., 2004; Bharucha et al., 2011) One approach, formally termed ‘second site non-complementation (SSNC)’ or ‘non-allelic non-complementation’, is particularly powerful when a gene of interest is essential for cell or organismal viability, such as CHD8 and CHD2. In brief, if two heterozygous null mutations, each having no observable phenotype when tested alone, create a phenotype when combined in a single organism, then the genes are said to genetically interact according to SSNC. We apply this approach here.

The Drosophila RIMS1 ortholog was previously demonstrated to be a central component of the presynaptic machinery necessary for PHP (Müller et al., 2012). Genetic interactions with heterozygous null mutations in Drosophila RIMS1 have been used to link genes to the mechanism of PHP (Harris et al., 2018; Hauswirth et al., 2018; Ortega et al., 2018). First, we confirm that PHP is robustly expressed in the RIMS1/+ heterozygous null mutant (Figure 1E–G). The average magnitude of homeostatic compensation is indistinguishable from wild type (p>0.1). When we plot the relationship between mEPSP amplitude and quantal content for every individual recording, there is a strong negative correlation observed in both wild type (R2 = 0.66) and RIMS1/+ (R2 = 0.64) (Figure 1F and G, respectively).

Next, we analyzed the heterozygous CHD8/+ mutant, which also shows robust PHP (Figure 1C–E) and a strong negative correlation between mEPSP amplitude and quantal content (R2 = 0.75; Figure 1H, gray points and black line). However, animals harboring heterozygous mutations in both RIMS1 and CHD8 (CHD8/+; RIMS1/+) show a complete failure of PHP (Figure 1H). The correlation of mEPSP amplitude and quantal content is abolished (Figure 1H, red points and red line; R2 = 0.01). The percent homeostasis in the double heterozygote is decreased to less than 10%, not statistically different from baseline (Figure 1H, box; p=0.6), and highly statistically different from both CHD8/+ and RIMS1/+ alone (p<0.01). We conclude that CHD8 can be linked, directly or indirectly, to the mechanisms of PHP. We propose that the heterozygous LOF mutation in CHD8 weakens the robustness of PHP, thereby associating an ASD-associated chromatin remodeling factor with homeostatic mechanisms that ensure robust synaptic transmission.

Next, to test the generality of this effect, we created double heterozygous mutant combinations of RIMS1/+ with the remaining ASD orthologs that we examine in this study (ASH1L, CHD2 and WDFY3) (Figure 1—figure supplement 2). The ASH1L/+, RIMS1/+ double heterozygous animal shows a complete failure of PHP (Figure 1—figure supplement 2). The percent PHP expression is decreased from 152% in the ASH1L/+ mutant, to 114% in the double heterozygote, which is not different from baseline (p=0.2) and represents a highly significant suppression (p<0.01) compared to the ASH1L/+ mutant alone (p<0.01) (Figure 1—figure supplement 2B). A similar analysis of the CHD2/+; RIMS1/+ double heterozygous animal shows a significant suppression of PHP (p=0.01), although significant PHP remains expressed in the double heterozygous animals (Figure 1—figure supplement 2C). Finally, the WDFY3/+; RIMS1/+ double heterozygous animal shows robust PHP (Figure 1—figure supplement 2D) that is indistinguishable from either the WDFY3/+ or the RIMS1/+ single heterozygotes. Taken together, these results suggest that there may be an unexpected connection between three unrelated ASD gene orthologs (CHD8 and CHD2 and ASH1L) and the mechanisms of PHP, given that all three genes interact with RIMS1. Based on these data, we pursued a genome-scale forward genetic screen to interrogate and better define the molecular interface of these ASD gene orthologs and the rapid induction of PHP.

Forward genetic screen for altered baseline transmission and PHP

The screen that we performed is diagrammed in Figure 2A. We took advantage of a collection of small chromosomal deficiencies (5–50 genes per deficiency, each with known chromosomal breakpoints) that tile the 3rd chromosome, uncovering approximately 6000 genes in total. For every double heterozygous combination of RIMS1/+ with a heterozygous deficiency, we performed multiple (n = 3–15) intracellular recordings, quantifying mEPSP amplitude, EPSP amplitude, quantal content (EPSP/mEPSP), resting membrane potential and input resistance. Recordings were made in the presence of PhTx to induce PHP. If the baseline EPSP is normal and quantal content is increased compared to wild type, then we can conclude that PHP is normally expressed. In these instances, we expect that baseline transmission was also normal in the absence of PhTx. However, if EPSPs are diminished in a given genetic combination (RIMS1/+, Df/+) and quantal content is not increased compared to wild type, then there are two possible origins: 1) the double mutant impairs baseline transmission or 2) baseline transmission is normal and PHP is selectively impaired. In these instances, the double heterozygous mutant combinations were re-assessed in the absence of PhTx to test for altered baseline transmission.

Screen for common genomic modifiers of ASD-associated gene mutations.

(A) Diagram of genetic screen. (B) Screen results are shown with yellow circles representing average data per genotype. Fit (solid blue line) and confidence interval (dotted lines encompassing 95% of all data) from RIMS1/+ are overlaid. Black horizontal line defines one standard deviation from population mean (yellow circles). Gray area encompasses potential hits residing outside the RIMS1/+ confidence interval and below the solid line. Two modifiers are shown in the absence (dark red circles) and presence of PhTx (light red circles, dark outline) (C) Approximate location of hits (red lines) on chromosome 3. (D) Representative traces for indicated genotypes in the presence and absence of PhTx as indicated. (E) Average percent change in mEPSP amplitude (gray bars) and quantal content (red bars) in presence of PhTx compared to baseline. One-way ANOVA and posthoc Tukey’s multiple comparisons; ****p<0.0001 for quantal content (QC).

Double heterozygous combinations that strongly affected muscle resting potential or input resistance were not observed. We uncovered two instances where a mutant combination (RIMS1/+, Df/+) caused a specific deficit in baseline transmission, without altering PHP. We did not isolate any double heterozygous conditions with dramatic defects in mEPSP amplitude. The majority of double mutant combinations specifically affected the expression of PHP. This was unexpected.

Double mutant combinations were determined to disrupt PHP by satisfying one of two criteria. First, the average quantal content (+PhTx) had to be more than one standard deviation below the population mean of all genotypes (Figure 2B, solid horizontal black line). Second, average quantal content had to reside outside a boundary that encompasses 95% of all individual recordings made in the RIMS1/+ mutant alone (Figure 2B, black dashed lines). Two example ‘hits’ are shown in red (Figure 2B; dark red point shows data in the absence of PhTx and light red point shows data recorded in the presence of PhTx, and the red lines simply connect the points for a given genotype for the purposes of data display). We also present a complete data set for a single hit from the screen as a standard format bar graph with representative traces (Figure 2D,E). Note that the heterozygous deficiency has normal baseline transmission and PHP (p>0.1 One-way ANOVA with post-hoc Tukey multiple comparison), but when combined with a heterozygous mutation in RIMS1, PHP completely and selectively fails.

In total, our screen identified, and we subsequently confirmed, 20 small deficiencies that cause PHP to fail when combined with RIMS1/+. To achieve a final list of 20 hits, each potential hit was re-validated in a second set of experiments, increasing sample sizes (generally 7–16 NMJ). During the process of re-validation, we rule out approximately one third of the potential hits selected from the screen. The identified deficiencies are randomly distributed across the 3rd chromosome (Figure 2C). The screen was empirically validated by the identification, blind to genotype, of deficiencies that uncovered the RIMS1 locus, as well as the Pi3K68D locus (not included in hit list), previously shown to interact as a double heterozygous mutant with RIMS1/+ (Hauswirth et al., 2018). Furthermore, the rim binding protein (RBP) locus was not identified as disrupting PHP, consistent with the previously published observation that a rbp/+ mutant does not interact with RIMS1/+ for PHP (Müller et al., 2015). However, rbp/+ did interact with RIMS1/+ for baseline neurotransmitter release as expected based on previously published data (not shown) (Müller et al., 2015). No other genes previously implicated in the mechanisms of PHP were present in the deficiencies isolated in our screen. It is important to note that, according to a formal genetic analysis, no strong conclusion can be made regarding the negative result of a double heterozygous genetic interaction (see Supplemental Tables S2-S5 for further detailed information on the screen results).

Finally, we assessed whether there was any relationship between the number of genes that were deleted within a given deficiency and the robustness of PHP. One hypothesis is that the additive effects of multiple, heterozygous gene mutations would increase for larger deficiencies and PHP would be increasingly compromised. That was not the case (Figure 3). There was no correlation between the number of genes uncovered by a given deficiency and EPSP amplitude recorded in the presence of PhTx (R2 = 0.003; Figure 3A). Thus, impaired PHP cannot be accounted for by a simple additive accumulation of genetic mutations within a given deficiency.

Absence of an additive effect of gene heterozygosity on synaptic transmission or PHP.

(A) Scatter plot showing the number of genes deleted (y axis) versus quantal content (x axis) in the presence of PhTx for all deficiencies tested. Each circle represents average data from an individual muscle recording for an individual deficiency. Red line shows the fit with a Pearson coefficient of 0.003. (B) Schematic of two deficiency alleles showing the extent of the deletion (yellow bars) and the genes deleted (red boxes) (C) Representative EPSP and mEPSP traces for indicated genotypes D) Quantification of EPSP, mEPSP amplitude and quantal content for the indicated genotypes. All deficiencies recorded as heterozygous mutations in the presence of RIMS1/+).

Identification of common phenotypic enhancers of multiple unrelated ASD genes

The results of our forward genetic screen, encompassing approximately one third of the Drosophila genome, might identify genetic interactions specific to RIMS1. However, we reasoned that because RIMS1 also showed a strong genetic interaction with CHD8, as well as CHD2 and ASH1L (Figure 1; Figure 1—figure supplement 2), a portion of the hits from our screen might also interact with these genes. When initial experiments confirmed that this was the case, we expanded our analysis to encompass all five of the ASD-associated gene orthologs from Figure 1. Thus, we performed a systems-genetic test of all possible double heterozygous genetic interactions, using wild type and the five ASD gene orthologs introduced in Figure 1 combined with wild type and five hits (deficiencies) randomly selected from our forward genetic screen. In total, we tested 36 genetic combinations for baseline transmission and PHP, recording every genotype in the presence and absence of PhTx (Figure 4).

Figure 4 with 1 supplement see all
Identification of common modifiers of diverse ASD-associated mutations.

(A) Genetic interaction matrix showing average mEPSP (top two matrix) and EPSP (bottom two matrix) amplitudes in the absence (left) and presence (right) of PhTx, as indicated. Values are according to lookup codes at left. Each individual box represents average data for a double-heterozygous mutant at intersection of x and y axes. Sample size (number of NMJ recordings) is displayed for each box (top) and are identical below (bottom). (B) Top matrix (orange and gray) is organized as in (A). Average percent change in quantal content (+ PhTx) compared to baseline (- PhTx), values according to lookup code. Bottom panel, shows data from top panel re-plotted diagramming p-values for the observed percent change in quantal content (+ / - PhTx), values according to lookup code. Student’s t-test (two tail) comparing each genotype + / - PhTx. (C) Scatter plots of quantal content (y axis) versus mEPSP amplitude (x axis) for CHD8/+ (left), and CHD8/+;Df(3)24410/+. (D) Scatter plot as in (C) for ASH1L/+ and ASH1L/+,/Df(3)7963/+. Each dot represents average data from an individual muscle recording. Fits as indicated. R2 values as indicated (calculated based on linear fit). (E) Representative traces for indicated genotypes (+ / - PhTx) (F) Percent change in mEPSP (gray bars) and quantal content (red bars) in presence of PhTx compared to baseline. One-way ANOVA and posthoc Tukey’s multiple comparisons; *p<0.05.

To facilitate visual interpretation, genetic interaction data are presented as a heat map superimposed on a matrix representing all genetic combinations, in the presence or absence of PhTx (Figure 4A). All but one genotype responded to the application of PhTx with decreased mEPSP amplitudes (Figure 4A, mEPSP; compare top left matrix with top right matrix, the transition from blue to red indicates diminished average mEPSP for each genetic combination). Thus, we induced homeostatic pressure in 35 out of 36 genetic combinations (CHD8/+ with Df(3)7562/+ being the exception). Next, we demonstrate that all heterozygous deficiencies (x-axis) or heterozygous ASD-associated gene mutations (y-axis), when crossed to the wild type strain (w1118) showed normal EPSP amplitudes in the absence and presence of PhTx, demonstrating robust induction of PHP (Figure 4A, EPSP, bottom right matrix). Next, nearly all (23 out of 25) of the double heterozygous combinations show normal EPSP amplitudes in the absence of PhTx, demonstrating normal baseline neurotransmission (Figure 4A, bottom left matrix). However, a majority (16 of 25) of the double heterozygous genetic combinations showed a failure of PHP in the presence of PhTx (Figure 4A, red and light-red boxes, bottom right matrix). In Figure 4B, we also plot the induction of PHP for each double heterozygous combination by calculating the percent change in quantal content following PhTx application (Figure 4B, top matrix). Here, if quantal content does not change (<15% change; gray), then PHP is impaired or blocked. Moderate increases in quantal content (15–30% change; light orange) suggest suppression of PHP, in some instances being statistically significant suppression (see below).

We performed statistical analyses for each double mutant combination, asking whether there was a statistically significant increase in quantal content for a given double mutant in the presence of PhTx compared to that same double mutant combination in the absence of PhTx (Figure 4B, bottom matrix.) Note that we are testing whether PHP is induced in a given double heterozygous mutant combination (an individual square in the matrix), comparing quantal content in the absence and presence of PhTx. We do not compare PHP expression among different double heterozygous mutant combinations. The colors gray and ‘light pink’ each reflect a complete block of PHP, an effect that is observed in the majority of double mutant combinations. As a complementary statistical analysis, we tested the differences between individual genotypic conditions (quantal content in each box) versus the wild type quantal content (One-way ANOVA with Dunnett’s multiple comparisons) (Figure 4—figure supplement 1). In this case, if PHP is blocked, then there will be a statistically significant difference compared to wild type. Again, 18 of 25 comparisons are significantly different. The genotypic comparison against wild type verified the analysis based on individual genotypic comparisons (Figure 4B). It should be noted that, in a few instances, minor differences were observed caused by a change in quantal content that was significant (-/+PhTx), but which remained smaller compared to wild type and therefore became significant. Thus, comparisons within genotypes (-/+PhTx) seem to assess the presence or absence of PHP most accurately (Figure 4B).

Our data demonstrate that four out of five deficiencies, isolated in our forward genetic screen as interacting with RIMS1, also cause PHP to fail when combined with any one of four different heterozygous ASD-associated gene mutations (Figure 4B, bottom). The pattern of PHP blockade is not uniform. WDFY3, CHD8 and CHD2 show a common pattern of interactions with the same three deficiencies. However, ASH1L interacts with only two out of the five tested deficiencies. To our knowledge, this is the first demonstration, in any system, of common phenotypic enhancement for multiple, independent and unrelated ASD gene orthologs.

In Figure 4C–F, we elaborate on three of the genetic interactions with data presentations that are more detailed. We show evidence of normal PHP in CHD8/+ (Figure 4C, replicated from Figure 1). In the adjacent graph (Figure 4C, right), we show evidence of a strong disruption of PHP in the double heterozygous combination of CHD8/+ with a heterozygous deficiency (Df(3)24410/+) isolated in genetic screen. A similar analysis is presented for the ASH1L/+ heterozygous gene mutation and the interaction with a different heterozygous deficiency (Figure 4D). Finally, a third genetic interaction is presented in a format that is standard for the field of homeostatic plasticity (Figure 4E,F), inclusive of representative traces (Figure 4E) and bar graphs with associated statistical analyses (Figure 4F). Note that values for all recordings are presented (Supplementary file 1 Supplemental Table S6). Several additional controls were performed to validate and extend the findings reported for our genetic interaction data set. First, we note that all double heterozygous mutant combinations are adult viable. Thus, it was possible to inspect adult animals for phenotypes that might indicate altered signaling. Inspection of the compound eye and wings (bristles, wing veins and size) demonstrate wild type tissue morphogenesis (data not shown).

PDPK1 and PPP2R5D are common phenotypic enhancers of multiple ASD gene orthologs

We isolated the causal single gene mutations within two of the deficiencies isolated from our screen. To do so, we tested smaller sub-deficiencies that mapped within the originally isolated deficiencies. Sub-deficiency mapping either identified the causal gene, or a limited number of candidates. We subsequently tested individual gene candidates with established single gene mutations or RNAi. The process of mapping to single genes, therefore, included several rounds of independent phenotype verification. The first two instances in which we have isolated single causal genes are presented. Each candidate gene was tested individually against all five ASD gene orthologs, using previously published mutations (Figure 5). For both genes, we confirmed the same set of genetic interactions that occurred when analyzing the deficiency that included the identified gene (Figure 5).

Figure 5 with 3 supplements see all
Single genes are common modifiers of diverse ASD-associated mutations.

(A) Schematic of the PPP2R5D gene locus and the PPP2R5D104 deletion mutation (red horizontal bar). (B) Representative traces for indicated genotypes. Bar graph (right) shows percent change in mEPSP (gray) and quantal content (red) (+ / - PhTx). (C) Data as in (B) for rescue of the double heterozygous CHD2/+ and PPP2R5D104/+ mutant by incorporation of a CHD2 translocation (CHD2WT/+). (D) Schematic of the PDPK1 gene locus with the PDPK133 deletion mutation (red horizontal bar). (E) Representative traces for indicated genotypes. Bar graph (right) as in B. (F) Data as in (C) for the genomic rescue of double heterozygous CHD2/+ and PDPK133/+ mutants. One-way ANOVA, Dunnett’s multiple comparisons *p<0.05, **p<0.01, ***p<0.001, ****p<0.0001 for (B) and (E), Student’s t-test, two-tailed for (C) and (F), n.s. p>0.05).

The first gene that we identified encodes a serine threonine kinase encoded by the PDPK1 gene (PDK1 in Drosophila). PDPK1 is a master controller of cellular metabolism, as well as cellular and synaptic growth at the Drosophila NMJ (Cheng et al., 2011). The second gene encodes a regulatory subunit of the PP2A phosphatase encoded by the PPP2R5D gene (wrd in Drosophila) (Viquez, 2006). PPP2R5D is also a master controller of cellular metabolism (Bernal et al., 2014), as well as cellular and synaptic growth at the Drosophila NMJ (Viquez, 2006). Both proteins are present at the Drosophila NMJ (Cheng et al., 2011; Viquez et al., 2009; Viquez, 2006) Single-cell Patch-Seq experiments confirmed the expression of these genes in Drosophila motoneurons at third-instar (Figure 1—figure supplement 1). Intriguingly, the PPP2R5D gene has recently been associated with intellectual disability and autism in human (Loveday et al., 2015; Shang et al., 2016).

Next, we demonstrate that both genes, PDPK1 and PPP2R5D, are common modifiers of multiple heterozygous ASD-associated gene mutations (Figure 5). To underscore the specificity of the double heterozygous genetic interactions (Figure 5B,E), we restored the expression of the CHD2 gene to wild type levels in the CHD2/+; PPP2R5D/+ double heterozygous combination. This was achieved using a previously published translocation of the CHD2 gene locus that allowed us to put back one copy of the CHD2 gene (CHD2WT/+) in the background of the CHD2/+; PPP2R5D/+ double mutant. We demonstrate that PHP is fully restored (Figure 5C). An identical series of experiments was performed with a previously characterized PDPK1 mutation (Figure 5E,F). Thus, PDPK1 and PPP2R5D represent the first known common phenotypic modifiers of ASD gene orthologs, causing PHP to fail (see also Supplementary file 1 Supplemental Table S7 for values and statistics). The data also underscore that deficiencies isolated in our forward genetic screen can be resolved to the activity of single genes. As such, the screen may have identified a novel class of common phenotypic modifier.

Another series of control experiments were performed. We asked whether neuronal morphology was substantially altered in the heterozygous ASD-associated gene mutations and in select double heterozygous genetic interactions (Figure 5—figure supplement 1). We do find evidence that the heterozygous CHD8/+ mutation predisposes the NMJ to modest overgrowth, consistent with CHD8 influencing brain development in other systems (Gompers et al., 2017). But, this effect does not become more severe when combined with either the PDPK1 or PPP2R5D mutation as double heterozygotes. Thus, we conclude that altered synaptic growth is not highly correlated with the block of PHP in these double heterozygous combinations.

Deletion of PPP2R5D impairs the robustness of PHP

Given that PPP2R5D and PDPK1 both genetically interact with multiple ASD-associated gene mutations, and given that several of the ASD-associated genes sensitize PHP toward failure, we considered whether PPP2R5D and PDPK1 are also directly involved with the induction of PHP. Unlike many of the ASD genes, both PPP2R5D and PDPK1 are viable as homozygous deletion mutations, allowing a direct test of their involvement in PHP. To our surprise, neither PPP2R5D nor PDPK1 can be classified as strictly essential for the mechanisms of PHP based on analysis of homozygous LOF mutants (Figure 5—figure supplement 2). PHP is fully expressed in the homozygous PDPK1 mutant (Figure 5—figure supplement 2), demonstrating that this gene is not required. There is a statistically significant suppression of PHP in the homozygous LOF mutation in PPP2R5D suggesting a role for this gene in the rapid induction of PHP, but without being strictly necessary (Figure 5—figure supplement 2).

We note that both PDPK1 and PPP2R5D control signaling that directly intersects with the AKT/mTOR pathway, a signaling system that is associated with ASD in human (Alessi et al., 1997; Manning and Toker, 2017; Yeung et al., 2017). The mTOR signaling proteins S6K and Tor have both been implicated in the long-term maintenance of PHP. However, both are dispensable for the rapid, PhTx-dependent induction of PHP (Cheng et al., 2011; Penney et al., 2012). Never-the-less, the possible connection to Tor signaling prompted us to revisit our screen data and ask whether mutations affecting the broader AKT/mTOR signaling system might also be common ASD-gene modifiers. The genes Akt, S6K, TSC1, TSC2, and PTP61F are all encoded on the Drosophila third chromosome. All of these genes were present within the deficiencies that were tested in our screen. But, none were identified as a hit in our unbiased forward genetic screen. Although the lack of a genetic interaction cannot be used to conclude the absence of a role for these genes in the PHP effects that we observe, it seems likely that PPP2R5D and PDPK1 have other targets relevant to the intersection of ASD-gene mutations and the rapid induction of PHP. Consistent with this possibility, PDPK1 and PPP2R5D are predicted to have opposing actions on AKT, yet both genes participate in the blockade of PHP when combined with a mutation in one of the five ASD-associated gene mutations (see discussion). Furthermore, as demonstrated below, one mechanism by which PHP is blocked is novel and unexpected.

Dissecting the mechanism of impaired PHP in a single double heterozygous mutant combination

It is rare for a genetic study to define, precisely, how a double heterozygous interaction creates a synthetic phenotype if the two genes do not encode proteins that biochemically interact. Simply put, there are a vast number of possible mechanisms by which SSNC could occur (Yook et al., 2001). None-the-less, we attempted to do so for at least one double heterozygous combination. Although this represents only a single mechanism of SSNC, it could provide proof of principle for how PHP is affected in other ASD gene interactions. We chose the genetic interaction of PPP2R5D/+ with CHD8/+. This combination was chosen because CHD8 is among the most common ASD de novo gene mutations. Furthermore, the genetic interaction is highly penetrant.

We began by pursuing additional phenotypic analyses, looking for clues in a wider variety of cellular and electrophysiological measures. It is possible that the genetic interaction of PPP2R5D/+ with CHD8/+ could indirectly affect PHP expression by altering motoneuron firing properties. Therefore, we analyzed intrinsic excitability and neuronal firing by patch clamp electrophysiology of larval motoneurons. There is no change in motoneuron firing frequency in response to a series of step current pulse injection. Likewise, there are no changes in action potential amplitude, cell input resistance or rheobase comparing wild type with each single heterozygous mutation and the double heterozygote (Figure 5—figure supplement 3). Thus, aberrant excitability is not linked to impaired PHP.

Ultrastructural correlate of impaired PHP: altered presynaptic membrane trafficking

Next, we turned to electron microscopy to determine whether the genetic interaction of PPP2R5D/+ with CHD8/+ affects the presynaptic release site. Ultrastructural changes have previously been linked to impaired PHP (Harris et al., 2018). Thin section transmission electron microscopy (EM) was used to examine the synapse, defined as a characteristic increase in pre- and postsynaptic membrane electron density, opposing clustered presynaptic vesicles and a characteristic presynaptic density, termed a T-bar. We find that the ultrastructure of CHD8/+ alone was wild type (Figure 6B,D,E). The ultrastructure of PPP2R5D/+ alone was wild type (Figure 6A,D,E). However, the double heterozygous mutant showed evidence of large membrane structures surrounding the presynaptic release site and apparent stalled endocytic events, appearing adjacent to sites of neurotransmitter release where compensatory synaptic vesicle endocytosis occurs (Figure 6C, insets). Quantification of vesicle size reveals a large increase in average intracellular vesicle diameter for all vesicles within 150 nm of the base of the presynaptic release site, defined by the T-bar structure (Figure 6D,E), again selective to the double heterozygous mutant. These data provide a striking visual confirmation of the genetic interaction between PPP2R5D/+ and CHD8/+. And, this is further evidence linking the action of a chromatin-remodeling factor (CHD8) to the stability of synaptic transmission.

Figure 6 with 1 supplement see all
ASD gene-modifier interaction causes impaired synaptic membrane organization.

(A–C) Representative electron microscopy images of individual boutons inclusive of (A) PPP2R5D/+, (B) CHD8/+ and (C) CHD8/+; PPP2R5D/+ double heterozygous mutant. Insets (C) show individual active zones taken from the image on the left (red rectangles) (D) Cumulative probability distribution of the vesicle size for wild type (w1118) and CHD8/+ and PPP2R5D/+ single mutants, as well as the CHD8/+; PPP2R5D/+ double heterozygous mutants. Each point reflects the average at a single active zone. (E) Plot of individual data points for each genotype as shown in (D). Sample sizes (D, E): Animal number: w1118 N = 2, CHD8/+ N = 2; PPP2R5D/+ N = 2; CHD8/+; PPP2R5D/+ N = 3. Active zone number: in same genotypic order n = 12, n = 23, n = 10, n = 14; Vesicle number n = 97, n = 171, n = 85; n = 89. (F) Representative electron microscopy images for individual boutons for indicated genotypes after stimulation with 50 Hz for 10 s and rapid fixation. Insets show active zones for wild type (top) and the CHD8/+; PPP2R5D/+ double heterozygous mutant (bottom) taken from the images on the left (red rectangles) (G) An example image from the CHD8/+; PPP2R5D+ double heterozygous mutant with larger vesicles having a crenulated appearance after stimulation. (H) Cumulative probability distribution of the vesicle size for wild type (w1118) and the double heterozygous mutant CHD8/+; PPP2R5D/+ after stimulation and rapid fixation. Each point reflects the average at a single active zone. (I) Plot of individual data points for data in (H). Sample sizes (H, I): Animal number: w1118 with stimulation N = 2, CHD8/+; PPP2R5D/+ with stimulation N = 2; Active zone number n = 10, 21. Vesicle number n = 101, n = 175. One-way ANOVA Tukey’s multiple comparisons, ****p<0.001, n.s. p>0.05.

Given the appearance of enlarged vesicles at or near the presynaptic release site, we repeated our ultrastructural analysis of the double heterozygous mutant, fixing the synapse immediately (~1–5 s) after strong stimulation of presynaptic release (50 Hz stimulation, 10 s). In wild type, there was no change in the number or appearance of presynaptic vesicles when fixed immediately following the stimulus. However, in the double heterozygous mutant condition (CHD8/+; PPP2R5D/+) we found that intracellular vesicles were further increased in size and took on a crenulated appearance (Figure 6F–I). These data are consistent with the enlarged vesicles being endosomal intermediates, arguing that the process of vesicle recycling is altered in the double heterozygous mutant. In further support of this idea, we demonstrate enhanced synaptic depression in response to high frequency (50 Hz) stimulation (Figure 6—figure supplement 1). Recently, homozygous CHD8 loss of function mutations have been linked to defects in synaptic vesicle endocytosis at the Drosophila NMJ (Latcheva et al., 2019). Our data underscore that that an endocytosis phenotype can be uncovered in the heterozygous CHD8/+ mutant in the context of the PPP2R5D/+ mutant. Regardless of the underlying molecular mechanism leading to this EM phenotype and associated physiological deficits (a topic for future study), these data present a striking, visual confirmation of a strong synthetic genetic interaction between PPP2R5D/+ and the CHD8/+ heterozygous mutations. Furthermore, these data link the activity of a chromatin remodeling factor, present in the nucleus (CHD8), to a profound synaptic defect. Experiments detailed below, including genetic rescue, confirm the specificity of this EM phenotype.

Differential gene expression analyses

One possible reason that genes isolated from our screen are common modifiers of diverse ASD genes is that each modifier is a direct transcriptional target of the ASD mutants. It is possible to assess this by RNAseq. To our knowledge, side-by-side differential gene expression analysis has yet to be performed for multiple heterozygous ASD-associated gene mutant backgrounds. We performed whole genome RNAseq analysis for wild types and the four heterozygous ASD mutants (four biological replicates) (Figure 7A). We asked whether any of the genes contained within the 20 deficiencies identified in the screen (37 genes; a number arrived at following sub-deficiency mapping and sub-selection based on gene expression in nerve or muscle) are commonly altered in all four of ASD-associated mutants (Figure 7A, orange data points). None were commonly differentially regulated (p-value=0.096 for ASH1L/+; p-value=0.636 for WDFY3/+; p-value=0.392 for CHD2/+; p-value=0.112 for CHD8/+; Wilcoxon sign rank test two-sided). We conclude that common down-regulation of identified genetic modifiers cannot account for the common impairment of PHP that we observe electrophysiologically.

Differential gene expression analysis identifies CREG.

(A) Volcano plot display of differentially expressed genes (DEX) for each heterozygous mutant versus wild type. Candidate ASD-gene modifiers are indicated (orange dots). Horizontal dashed line indicates cutoff of adjusted p-values (0.05). (B) Matrix shows all intersections of DEXs from the four indicated genotypes (see Database S1). Filled circles in the matrix indicate sets that are part of the intersection between genotypes. Bar graphs on the top show the total number of DEXs for each set, ordered by the size of intersection. (blue, up-regulated; red, down-regulated). (C) Individual genes are listed at the intersection of each genotypes. (D) Schematic showing the selection of 14 genes uniquely dysregulated in CHD8/+; PPP2R5D/+ double heterozygous mutants. (E) Volcano plot display of DEX calculated as CHD8/+; PPP2R5D/+ versus wild-type, CHD8/+; PPP2R5D/+ versus CHD8/+ and CHD8/+; PPP2R5D/+ double heterozygotes versus PPP2R5D/+ alone.

Next, we asked whether the ASD-associated gene mutations might cause common changes in gene expression, with potential relevance to a common disruption of PHP. We define all differentially expressed genes common to at least two ASD mutations (Figure 7B,C). While there are individual genes that are commonly differentially regulated, a GO database analysis of differentially expressed genes did not reveal any consistent change in a gene category across all four genotypes. The patterns of gene dysregulation do not predict any pattern of genetic interactions documented in our systems-genetic analysis. Finally, while there are genes that are commonly dysregulated in multiple ASD gene orthologs, there are only two genes that are commonly down-regulated in all four ASD mutants (FBgn0027578 [Nepl21] and FBgn0037166 [CG11426]) (Figure 7C). FBgn0027578 encodes a metalloprotease of the Neprilysin family, with homology to endothelin converting enzyme one in human, of unknown function in the nervous system. FBgn0037166 encodes phosphatidic acid phosphatase type 2, which is expressed in the Drosophila nervous system, but of unknown function. There is no obvious means to connect the down regulation of these two genes to impaired homeostatic signaling, although future experiments will explore these genes in greater depth. Furthermore, there is no clear connection between FBgn0027578 or FBgn0037166 and the roles of either PDPK1 or PPP2R5D in the nervous system. Thus, a transcriptional analysis of heterozygous ASD gene mutations alone did not allow us to make clear progress toward understanding the mechanisms of impaired PHP.

Candidate mechanisms for impaired PHP based on differential gene expression analysis

Next, we continued with our focus on characterizing the homeostasis defect in the CHD8/+; PPPR25D/+ double heterozygous mutant combination. We repeated the RNAseq differential gene expression analysis comparing the double heterozygous condition to three control conditions, inclusive of wild type and each single heterozygous mutant alone. In this manner, we sought to identify synergistic effects on gene expression that could not be accounted for in either single heterozygous mutant alone (Figure 7D,E). As expected, many of the differentially expressed genes documented in the double heterozygous mutant, when compared to wild type, could be accounted for by subsequent comparisons to each single heterozygous mutant. However, a small number of genes (14 genes; 5 upregulated and nine downregulated) appear to be synergistically differentially expressed in the double heterozygous mutant compared to all three control conditions (Figure 7D,E). We successfully replicated altered expression of four genes in the double heterozygous mutant combination by quantitative RT-PCR (Figure 8A,B). Of these genes, CREG stood out as being robustly and dramatically up-regulated. Upon closer inspection, CREG showed a slight, but significant, up-regulation in the CHD8/+ mutant (Figure 7A), and this was enhanced by the presence of the heterozygous PPP2R5D/+ mutation (Figure 7E, left). Next, we confirmed the up-regulation of CREG in the third instar larval central nervous system by quantitative RT-PCR (Figure 8—figure supplement 1B). Finally, we took advantage of a previously published gene expression data set (Parrish et al., 2014) and document CREG expression in motoneurons throughout embryonic and larval development. CREG is strongly expressed in embryonic motoneurons (20–24 hr after egg laying – AEL), after which expression levels plummet (Figure 8—figure supplement 1A).

Figure 8 with 3 supplements see all
CREG is a homeostatic repressor that blocks PHP and regulates synapse ultrastructure.

(A) Quantification of transcriptional changes calculated by RNAseq for four genes (CREG, Ect3, PEPCK2 and Cyp6a23) in CHD8/+; PPP2R5D/+ double heterozygous mutant versus wild-type. (B) Quantification of the transcriptional changes for the same genes in (B) by qPCR. (C) Schematic of the Drosophila CREG locus. The positions of two transposon insertion mutations are shown (red triangles). (D) Average CREG transcript levels calculated by qPCR are shown for the indicated genotypes E) Representative EPSP and mEPSP traces for indicated genotypes. (F), Bar graph (right) shows percent change in mEPSP (black filled) and quantal content (no fill) (+ / - PhTx). Sample size indicated as (–PhTx/+PhTx): wild type n = 17/15; CHD8/+; PPP2R5D/+ n = 11/14; CregM1 n = 8/7; CregM1/+ n = 8/8; CHD8/+; PPP2R5D/+; CregM1/+ n = 8/10; CHD8/+; PPP2R5D/+; CregM2/+ n = 8/8. (G) Representative electron microscopy images of individual active zones from indicated genotypes (double heterozygous mutant at left, triple heterozygous mutant at right). Scale bar:100 nm. (H) Individual data points (vesicle size) shown for indicated genotypes. (I) Cumulative probability distribution of vesicle size for genotypes shown in (H). Sample sizes for (H, I): Animal number: wild type N = 2, CHD8/+; PPP2R5D/+ N = 3; CHD8/+; PPP2R5D/+; CregM1/+ N = 3. Active zone number: in same genotypic order n = 12, n = 14, n = 12; Vesicle number n = 97, n = 89, n = 112. (J) Scatter plots of quantal content (y axis) versus mEPSP amplitude (x axis) for wild type (left), OK371-Gal4 > UAS Creg (middle, red) and Tub-Gal4 >UAS Creg (right, blue). Fits as indicated. R2 values as indicated (calculated based on linear fit). (K) Percent change in mEPSP (gray bars) and quantal content (red bars) in presence of PhTx compared to baseline. Sample sizes as in (F), wild type n = 8/6; OK371-Gal4 > UAS) Creg n = 14/12; Tub-Gal4 >UAS) Creg n = 11/11. n.s. p>0.05, **p<0.01, ***p<0.001, ****p<0.0001.

CREG is a homeostatic repressor

CREG (Cellular Repressor of E1A-stimulated Genes) encodes an endosomal/lysosomal localized glycoprotein that is linked to stress responses in other systems as well as to the homeostatic maintenance of the vascular epithelium (Ghobrial et al., 2018; Kowalewski-Nimmerfall et al., 2014). Mammalian orthologs are expressed in the brain (Yang et al., 2011). However, CREG function has never been addressed in the nervous system of any organism. Given that we observe a strong synaptic internal membrane phenotype in the CHD8+; PPP2R5D/+ double heterozygous mutant, and given that CREG localizes to the endo-lysosomal system, we chose to study CREG in greater detail, asking if it is causally involved in PHP.

Two independent transposon insertion mutations were identified, residing in the Drosophila CREG gene locus (Figure 8C). The CREGM1 transposon completely abolishes CREG expression and a heterozygous CREGM1/+ mutant reduces CREG expression (CREGM1 = zero expression compared to wild type, 3 biological and three technical replicates; CREGM1/+ = 51.5 ± 3.1% wild type expression, 3 biological and three technical replicates). Next, we generated a triple heterozygous mutant combination (CHD8+; PPP2R5D/+, CREGM1/+) and find that the CREGM1/+ allele attenuates the up-regulation of CREG gene transcript in the triple heterozygous mutant background, a suppression effect of approximately 50%, as predicted (Figure 8D). Then, we repeated this analysis with the CREGM2 allele. This allele has a minor effect on baseline CREG expression (73.2 ± 2.7% wild type expression, 3 biological and three technical replicates). However, we discovered that this transposon insertion caused a complete block of CREG up-regulation in the triple heterozygous mutant combination, suggesting that this transposon insertion, residing in 3’ UTR, may disrupt a transcription regulatory motif (Figure 8D).

Next, we asked whether the triple heterozygous mutant combinations, in which CREG up-regulation is either attenuated or abolished, would rescue the expression of homeostatic plasticity and synaptic ultrastructure. In both triple mutant combinations (CHD8+; PPP2R5D/+, CREGM1/+) and (CHD8+; PPP2R5D/+, CREGM2/+), the expression of PHP is fully rescued (Figure 8E,F). These data are consistent with the conclusion that the abnormally enhanced levels of CREG transcription are responsible for the block of homeostatic plasticity seen in the double heterozygous mutant combination. If true, then we might also see rescue of the ultrastructural phenotype in the CHD8+; PPP2R5D/+, CREGM1/+ triple mutant. Indeed, this is the case (Figure 8G). We observe full rescue of synaptic ultrastructure. Thus, preventing the dramatic up-regulation of CREG, without abolishing CREG expression, is sufficient to restore membrane trafficking and PHP to the presynaptic nerve terminal of the CHD8+; PPP2R5D/+ double mutant combination.

It is possible that CREG is a novel suppressor of PHP. However, it is also possible that CREG mediates this effect only in the context of the other two heterozygous mutations. To address this possibility, we generated a UAS-CREG transgenic line, allowing cell-type specific overexpression of the CREG gene. Over-expression of CREG in a wild type background using either a ubiquitously expressed source of GAL4 (tubulin-GAL4), or a GAL4-line that is selective to motoneurons (OK371-GAL4), causes a complete block of PHP (Figure 8J–K). As a control for adverse developmental effects of CREG overexpression, we analyzed NMJ anatomy and find no substantive effects on NMJ growth or morphology that could account for the absence of PHP (Figure 8—figure supplement 2).

Finally, we assessed the consequences of the heterozygous and homozygous loss of function mutations on baseline neurotransmission and PHP. The CREGM1 allele abolished expression (see above) and is the focus of these analyses. Neither the heterozygous nor homozygous animals affected expression of PHP (Figure 8F). The CREGM1/+ heterozygous animals had no effect on baseline transmission compared to wild type (wild-type QC = 40.8 ± 2.2 n = 10; CREGM1/+ QC = 38.6 (±3.0) n = 8; Student’s t-test; p>0.5). The CREGM1 homozygous allele decreased baseline transmission by ~18% (CREGM1 QC = 33.5 ± 3.0 n = 8; p=0.02). Clearly, neither baseline release nor PHP are potentiated, demonstrating that the rescue of PHP in the triple heterozygous mutant condition cannot be considered an additive effect of the heterozygous CREGM1 mutation. Taken together, our data are consistent with the conclusion that CREG is a homeostatic repressor, one of very few identified to date (Spring et al., 2016). This finding underscores the complexity of interpreting the double heterozygous mutant combinations that cause blockade of PHP.

Assessing the generality of CREG as a mechanism for impaired PHP

In mammals, there are two CREG genes and CREG2 is expressed in the brain (Yang et al., 2011). A recent study provides evidence that CREG2 expression is enhanced in layer four excitatory neurons, isolated from human postmortem ASD patient brain tissue (Velmeshev et al., 2019), suggesting possible relevance. This fact prompted us to ask whether over-expression of CREG is the primary mechanism responsible for the disruption of PHP, or whether it is just one of many. Our existing gene expression analysis demonstrates that CREG is not up-regulated in the other heterozygous mutations (ASH1L or CHD2 or WDFY3; Figure 7). This was extended to the PDPK1/+ mutant and, again, CREG levels are not increased. Finally, we analyzed two additional double heterozygous mutant combinations (PDPK1/+ with CHD2/+ as well as PPP2R5D/+ with CHD2/+). CREG was not up-regulated compared to single heterozygous controls. Finally, we repeated the ultrastructural analysis for a second genetic combination (PPP2R5D/+ with CHD2/+). No phenotype of enlarged vesicles or endomembranes was observed (Figure 8—figure supplement 3). From these data, we conclude that the aberrant over-expression of CREG is not a universal cause of impaired PHP in the double heterozygous interactions. In the future, a systematic test of all genetic combinations identified in our screen may define whether CREG over-expression is unique to a single genetic interaction or whether it is reflected in a subset of gene interactions.

Discussion

In this study, we make several fundamental advances. First, we provide evidence that mutations in multiple different ASD-associated genes sensitize homeostatic plasticity to fail (Figure 9A,B). Second, using genome-scale forward genetics and subsequent systems-genetic analyses, we identify the first phenotypic modifiers that commonly enhance five different ASD-associated gene mutations, causing a specific failure of PHP (Figure 9A). Third, we identify PDPK1 and PPP2R5D as common phenotypic modifiers of multiple ASD-associated genes and, thereby, define a mechanistic link between synaptic transmission, PHP and chromatin remodeling complexes in the neuronal nucleus (Figure 9A). Finally, we define how PHP fails at the intersection of an ASD-associated gene mutation and phenotypic modifier. The mechanism is unexpected, involving the maladaptive up-regulation of a novel repressor of homeostatic plasticity (CREG) (Figure 9C). We demonstrate that up-regulation of CREG cannot explain other gene-gene interactions, underscoring the potential complexity of gene-gene interactions and the common failure of PHP. Regardless of potential mechanistic complexity, our data argue that impaired PHP may be a common pathophysiological effect downstream of LOF mutations in five different ASD-associated genes. If our data can be extended to additional ASD genes, and to other experimental systems including human neurons, then it may be possible to use this information to advance therapeutic approaches that modify ASD phenotypic severity regardless of the underlying genetic mutation(s) that confer risk for ASD.

Summary and Model.

(A) Summary of genetic interactions. RIMS1 interacts with three of four ASD gene orthologues, impairing PHP. RIMS1 interactions identified in a genetic screen as modifiers are shown below in green. Each modifier interacts with multiple ASD heterozygous mutations in a RIMS1 independent manner, disrupting PHP. The data present a complex network of gene-gene interactions (yellow and green) that diminish the robustness of PHP. Df(3)1 refers to Df(3)7562 and Df(3)2 refers to Df(2)24953. (B) Homeostatic signaling systems robustly ensure stable neural function. However, the homeostatic signaling system itself is sensitive to genetic perturbation. We demonstrate that PHP is sensitive to mutations in multiple genes that were identified as ASD risk factors. In at least one instance, this is due to the up-regulation of a PHP interfering factor (CREG) and the red star indicates this a possible mechanisms more generally. (C) Complexity of interpreting double heterozygous gene-gene interactions. Signaling systems are not blocked by heterozygous gene mutations, but are likely to be attenuated to some degree. The combined effect of two higher-order heterozygous gene mutations creates a downstream, intersectional effect that is very difficult to predict. In the case of this paper, we succeeded in identifying a novel intersection causing up-regulation of CREG, which disrupts the homeostatic signaling system.

The loss or impairment of PHP could contribute to the phenotypic penetrance of an ASD gene mutation in multiple ways. Impaired PHP is expected to render the nervous system less robust to perturbation including the effects of environmental stress, immunological stress, or genetic mutation (Davis, 2013; Davis, 2006). If an ASD-associated gene mutation leads to neural developmental defects, then loss of PHP would be expected to exacerbate the functional consequences. According to the same logic, loss of PHP might enhance the adverse effects of environmental or immunological stress, both of which are thought to contribute to ASD pathophysiology (Beversdorf et al., 2018; Modabbernia et al., 2017). Finally, loss of PHP could be relevant to the appearance or severity of ASD comorbidities, including epilepsy.

It should be emphasized that failed homeostatic plasticity cannot be determined by simply assessing the phenotype of a heterozygous ASD-associated gene mutation. The observation of a phenotype, such as altered E/I balance or impaired neurotransmission, could reflect failure of homeostatic plasticity, or it could reflect the outcome of successful homeostatic mechanisms that constrained a phenotype that might otherwise have been more severe (Davis, 2013; Kulik et al., 2019). Ultimately, the loss or impairment of homeostatic plasticity can only be determined by a direct test of homeostatic robustness; specifically referring the ability of a neuron, synapse or neural circuit to respond to a perturbation and sustain normal function in the continued presence of the perturbation (Davis, 2013; Davis, 2006). Thus, our data set the stage for similar analyses in other model organisms, potentially extending the connection between ASD-associated gene mutations and the robustness of PHP or other forms of homeostatic plasticity.

The specificity of gene-gene interactions that cause PHP to fail

The genetic interactions that we document in our study appear to be highly specific. First, our genetic screen was based on the use of deficiency chromosomes that uncover 5–50 genes each, rendering those genes heterozygous. Thus, each deficiency can be considered to test pairwise gene-gene interactions among all the genes contained in the deficiency. According to this logic, we tested in excess of 50,000 double heterozygous gene combinations and discovered only 20 interactions that cause PHP to fail. Although it is unlikely that genes are completely randomly distributed throughout the genome, this calculation still has merit and emphasizes the rarity of gene-gene interactions that cause PHP to fail. In addition, we found no correlation between the number of genes deleted in a heterozygous deficiency and the impairment of PHP. Thus, the likelihood of a genetic interaction does not increase with the number of genes that are rendered heterozygous. Finally, it should be emphasized that PHP is a robust physiological process that is not unusually susceptible to the effects of genetic mutations. Previously, forward genetic have observed low rates of gene discovery. Two such screens tested transgenic RNAi against nearly every kinase and phosphatase encoded in the Drosophila genome, a gene set that includes prominent signaling proteins, the majority of which had no effect on PHP induction or expression (Brusich et al., 2015; Hauswirth et al., 2018). With this information as a background, the identification of genes that commonly enhance multiple ASD genes, causing PHP to fail, seems extraordinarily.

The rapid induction versus long-term expression of PHP

There are two well-established methods to induce expression of PHP. Application of PhTx induces PHP within minutes, a process that can be maintained for hours (Frank et al., 2006). In addition, a mutation in the non-essential GluRIIA subunit of postsynaptic glutamate receptors drives persistent expression of PHP. Since the GluRIIA mutation is present throughout the life of the organism, it is inferred that this reflects the long-term maintenance of PHP. Although this distinction reflects only the duration of the perturbation (acute versus genetic), recent work does argue that the acute induction of PHP may transition to another long-term expression mechanism (Harris et al., 2018; Harris et al., 2015). Indeed, screens based on the acute versus long-term PHP have identified different candidate genes, even when screening a common transgenic RNAi collection (Brusich et al., 2015; Hauswirth et al., 2018).

It remains unknown whether one form of PHP is more relevant regarding the intersection of homeostatic plasticity with diseases or disorders of the nervous system. In the present study, the acute induction of PHP can be considered a type of ‘stress test’. If the rapid induction of PHP fails, we can infer that the neurons are less robust to perturbation. In the future, it will be interesting to systematically determine whether the gene-gene interactions identified here also uniformly perturb PHP induced by the GluRIIA mutation. However, such an analysis is beyond the scope of the present study.

Common phenotypic enhancers of multiple ASD gene orthologs

How can the existence of common phenotypic modifiers be explained? We began our study with the demonstration that heterozygous LOF mutations in four unrelated ASD-associated genes including RIMS1 (presynaptic scaffolding protein), CHD8 (chromatin helicase), CHD2 (chromatin helicase) and ASH1L (transcriptional activator and histone methyltransferase), all sensitize the expression of PHP to fail (Figure 1, Figure 1—figure supplement 2, Figure 9B). One possibility, therefore, is that PHP is commonly sensitized to fail by heterozygous LOF mutations in each of the five ASD gene orthologs that we chose to study. If so, then a phenotypic modifier that interacts with one of these genes might also be expected to commonly interact with the other ASD genes. In other words, commonality arises because of the unexpected finding that each ASD gene ortholog has an activity that, when diminished, impairs the robustness of PHP. Our data generally support this model, given that three of four ASD genes interact with RIMS1 to block PHP. According to this model, we provide the first evidence that sensitization of PHP is a common pathophysiological effect downstream of multiple ASD genes with, as yet, unrelated biological activities.

The finding that ASD gene mutations sensitize PHP to fail does not require that each ASD gene participate in the actual mechanisms of PHP. RIMS1 is a core component that is required for PHP (Müller et al., 2012). However, a gene such as CHD8 might compromise PHP indirectly by causing some form of cellular stress that interacts with the mechanisms of PHP (Figure 8J,K). Indeed, it was previously demonstrated that simultaneous induction of two different forms of homeostatic plasticity creates interference and homeostatic failure (Bergquist et al., 2010). The same argument can apply to the novel class of common phenotypic modifiers. Some modifiers may represent core components of PHP, including PPP2R5D, which seems to suppress PHP when knocked out (Figure 5—figure supplement 2). However, the PDPK1 knockout has no effect on PHP and, therefore, may interact with the mechanisms of PHP indirectly. Thus, we cannot rule out the possibility that compounded cellular stressors occasionally intersect and cause PHP to fail.

Novel mechanisms impair PHP; CREG-dependent suppression of PHP

We explored, in detail, how PHP fails at the intersection of CHD8/+ and PPP2R5D/+. First, we discovered a profound effect on synaptic ultrastructure that was not observed in either single heterozygous mutation. This provided dramatic visual proof of a strong, genetic interaction between these two heterozygous gene mutations. Next, we demonstrate that this strong, genetic interaction is not a consequence of extensive transcriptional dysregulation. Indeed, when the effects of each heterozygous gene mutation are taken into account, only 14 genes show evidence of altered transcription. A single gene, CREG, was subsequently demonstrated to be the cause of impaired PHP and disrupted presynaptic membrane trafficking. Although CREG2 is not upregulated in the heterozygous CHD8/+ mouse, a recent study provides evidence that CREG2 expression is enhanced in layer four excitatory neurons, isolated from human postmortem ASD patient brain tissue (Gompers et al., 2017; Velmeshev et al., 2019).

It remains to be determined how loss of PPP2R5D causes further dysregulation of CREG in the background of CHD8/+. One possibility is that CREG is a stress-response gene, and up-regulation occurs at the intersection of two cellular stresses. Other alternatives remain plausible, including a direct connection between CHD8 and CREG that is modulated by PPP2R5D-mediated signaling. The biochemical and transcriptional relationships will be defined in subsequent work and are beyond the scope of our current study. The generality of this genetic interaction will also be explored. We note, for example, that CREG shows a mild increase only in the CHD8/+ mutant, not in the other three ASD-associated genes (Figure 7A). This does not rule out CREG participating in genetic interactions involving other ASD-associated genes, but it might suggest additional mechanisms will be engaged.

CREG encodes a glycoprotein that localizes within the endo-lysosomal system and may also be secreted. In mammals, there are two CREG genes and CREG2 is expressed in the brain (Yang et al., 2011). There is generally more information regarding the function of CREG1, which is an effector of tissue homeostasis in the vascular epithelium (Ghobrial et al., 2018). In this capacity, CREG seems to function as a stress response factor, influencing the activity of several potent signaling systems (Ghobrial et al., 2018). Our current phenotypic analyses suggest that increased levels of CREG may directly impact the integrity of synaptic vesicle membrane recycling and, either directly or indirectly, interface with the homeostatic potentiation of vesicle release. Thus, while a full dissection of CREG activity remains for future studies, our data argue that CREG has an activity that could be directly coupled to vesicle release and recycling, an ideal situation to normally limit the homeostatic potentiation of vesicle fusion.

Relevance and conclusions

It is well established that genetic context can profoundly influence the phenotypic severity of disease-causing gene mutations. For example, in mice, it has been shown that genetic context (strain background) influences phenotypic penetrance in an Alzheimer’s disease model (Neuner et al., 2019). In humans, systematic screening of the phenotypically normal population has identified individuals that are resistant to the effects of well-established, debilitating disease causing mutations, an effect termed ‘resilience’ that is attributed to the effects of genetic context (Chen et al., 2016; Friend and Schadt, 2014). It seems plausible that the common phenotypic enhancers, identified in our genetic screen, could represent a mechanism by which genetic context influences the phenotypic penetrance of ASD-associated gene mutations. We recognize that PDPK1 and PPP2R5D have fewer than expected LOF and missense mutations in humans (http://exac.broadinstitute.org). It remains to be determined if this will be the case with additional modifier genes. Furthermore, we note that PHP is completely blocked at the intersection of ASD gene mutations and the common modifiers we identify. Therefore, subtle changes in the expression or function of common phenotypic modifiers, perhaps caused by mutations in enhancer/promoter regions, could impact expression or robustness of PHP with cascading negative phenotypic consequences. If our findings can be extended to other systems, including humans, it is conceivable that our emerging mechanistic understanding could be used to restore the beneficial effects of homeostatic plasticity and alleviate aspects of ASD phenotype, irrespective of individual genetic makeup.

Materials and methods

Key resources table
Reagent type
(species) or resource
DesignationSource or referenceIdentifiersAdditional
information
Genetic reagent (D. melanogaster)w1118Bloomington Drosophila Stock CenterBDSC:3605
Genetic reagent (D. melanogaster)rim103PMID:23175813
Genetic reagent (D. melanogaster)kis1Bloomington Drosophila Stock CenterBDSC:431
Genetic reagent (D. melanogaster)chd1[1] and chd1
[1], chd1[wt]
PMID:21177652
Genetic reagent (D. melanogaster)ash1-mimicBloomington Drosophila Stock CenterBDSC:23524
Genetic reagent (D. melanogaster)bchs58Bloomington Drosophila Stock CenterBDSC:9887
Genetic reagent (D. melanogaster)UAS-CREGthis paperFly carrying UAS-Creg transgene
Genetic reagent (D. melanogaster)wrd104PMID:16957085
Genetic reagent (D. melanogaster)pdk1PMID:21930778
Genetic reagent (D. melanogaster)ok371-gal4PMID:16378756
Genetic reagent (D. melanogaster)tubulin-gal4PMID:21930778
Genetic reagent (D. melanogaster)3rd chromosome deficiency collectionBloomington Drosophila Stock Center
Genetic reagent (D. melanogaster)Creg-m1Bloomington Drosophila Stock CenterBDSC:42140
Genetic reagent (D. melanogaster)Creg-m2Bloomington Drosophila Stock CenterBDSC:22800
Genetic reagent (D. melanogaster)MN1b-gal4Bloomington Drosophila Stock CenterBDSC:40701
Genetic reagent (D. melanogaster)MN1s-gal4Bloomington Drosophila Stock CenterBDSC:49227
Genetic reagent (D. melanogaster)uas-cd8:gfpPMID:10197526
Chemical compound, drugPhilanthotoxin-433Santa Cruz BiotechnologyGH28782
AntibodyAnti-brp
(Mouse monoclonal)
Developmental Studies Hybridoma BankRRID:AB_2314866IF (1:100)
AntibodyAnti-dlg
(Rabbit monoclonal)
PMID:29303480IF (1:1000)
AntibodyCy3 anti-rabbitJackson Immuno-research LaboratoriesRRID:AB_2338000IF (1:500)
AntibodyAlexa488 anti-mouseJackson Immuno-research LaboratoriesRRID:AB_2338840IF (1:500)
Software, algorithmIgor Pro 8.03WavemetricsRRID:SCR_000325
Software, algorithmGraphpad PRISM 7.04GraphpadRRID:SCR_002798
Software, algorithmAdobe Illustrator CC 2018AdobeRRID:SCR_010279
Software, algorithmMiniAnalysis 6.0.7SynaptoSoftRRID:SCR_002184
Software, algorithmSlideBook 6Intelligent ImagingRRID:SCR_014300
Sequence-based reagentCREG primerApplied BiosystemsDm02135967_g1
Sequence-based reagentEct3 primerApplied BiosystemsDm02139373_g1
Sequence-based reagentPepck2Applied BiosystemsDm02366462_s1
Sequence-based reagentCyp6a23Applied BiosystemsDm01824231_g1
Sequence-based reagentrpl32Applied BiosystemsDm02151827_g1
Commercial assay or kitRNeasy Plus Mini kitQIAGENID:74134
Commercial assay or kitSuperScript III First-Strand synthesis systemInvitrogenCat# 18080051
Commercial assay or kitTURBO DNA-free kitThermoFisherCat# AM1907
Commercial assay or kitTaqMan Fast Universal PCR Master MixApplied BiosystemCat# 4352042
Commercial assay or kitLexogen’s Split RNA Extraction KitLexogenCat# 008
Commercial assay or kit3’mRNA-Seq Library Prep KitLexogenCat# 015
Commercial assay or kitSingle Cell/Low Input RNA Library Prep KitNew England BiolabsCat# E6420S
Recombinant DNA reagentpTW (Gateway vector)DGRCCat# 1129
Recombinant DNA reagentpENTR-dTOPO Cloning KitInvitrogenCat# K240020
Recombinant DNA reagentCreg cDNADrosophila Genomics Resource CenterGH28782

Fly stocks

Request a detailed protocol

All Drosophila stocks were kept and raised on standard food at 25°C. RIMS1 was previously described (Müller et al., 2012). PPP2R5D was a gift from Dr. Aaron Diantonio. Chd2[1] and Chd2[1],Chd2wt flies were gifts from Dr. Alexandra Lusser and Dr. Dmitry Fyodorov. All other Drosophila stocks were obtained from the Bloomington Drosophila Stock Center unless otherwise noted. W1118 is used as wild-type controls. 3rd chromosome deficiency fly stocks are balanced over TM6b and all stock are on the W1118 background.

Molecular biology

Request a detailed protocol

Drosophila CREG cDNA was obtained by amplifying the single open reading frame from genomic DNA by PCR and cloning directly in to the pENTR vector (Gateway Technology; Invitrogen). We engineered a CACC site in the forward primer for the subsequent Gateway reaction: forward primer for pUASt-creg: 5′ CACCATGGATTCGGACAGCACC 3′; reverse primer for pUASt-creg with a stop codon, 5′ TCA ATT CGA AAC AGC GTA ATA 3′. The final construct were sequenced to ensure there were no mutations. The creg cDNA was then cloned into proper destination vector obtained from the Drosophila Gateway Vector Collection (Carnegie Institution, DGRC barcode #1129). Transgenic lines were generated and mapped using standard methods.

Electrophysiology

Request a detailed protocol

All current clamp recordings were performed from muscle six, at the second and third segment of the third-instar Drosophila with an Axoclamp 900 amplifier (Molecular Devices). The composition of the extracellular solution (HL3) is (in mM) 70 NaCl, 5 KCl, 10 MgCl2, 10 NaHCO3, 115 sucrose, 4.2 trehalose, 5 HEPES. Ca2+ concentration in the extracellular solution is 0.35 mM unless otherwise noted. Homeostatic plasticity was induced by incubating the larvae with Philantotoxin-433 (PhTx, 15–20 µM, Sigma) for 10 min as previously described (Frank et al., 2006; Genç et al., 2017). Quantal content calculation is made by dividing average EPSP to mEPSP. mEPSPs were analyzed with MiniAnalysis program (Synaptosoft). All other physiology data were analyzed with custom written functions in Igor 6 (Wavemetrics Inc). Data collected from a minimum of two animals from two independent crosses.

Immunohistochemistry

Request a detailed protocol

Third-instar larvae were dissected, fixed in Bouin’s fixative or 4% PFA in PBS, and immunostained with previously described methods (Eaton et al., 2002; Harris et al., 2015). Third instar larvae were dissected with cold HL3 and immediately fixed with PFA (4%) and incubated overnight at 4 C with primary antibodies (rabbit anti-Dlg, 1:1000; anti-Brp 1:100, Life Technologies). Alexa-conjugated secondary antibodies were used for secondary staining (Jackson Laboratories 1:500). An inverted epifluorescence deconvolution confocal microscope (Axiovert 200, Zeiss) equipped with a 100X objective (N.A. 1.4), cooled CCD camera (CoolSnap HQ, Roper Scientific) was used to acquire images. All acquisition, deconvolution and analysis were done by Slidebook 5.0 software (3I, Intelligent Imaging). Structured illumination microscopy (Nikon LSM 710 equipped with 63X objective and Andor Ixon EMCCD camera) was used to perform Brp puncta and Dlg labeling experiments. Bouton numbers were quantified as described previously (Harris et al., 2015).

RNA extraction and library preparation for RNA sequencing

Request a detailed protocol

RNA was extracted from the adult heads (5–7 days post-pupation) of heterozygous mutants of four genotypes (ASH1L/+ and WDFY3/+ and CHD2/+ and CHD8/+) and wild types with four biological replicates per group by using Lexogen’s RNA Extraction Kit (Lexogen). RNA quality was checked with Bioanalyzer (Agilent Technologies Inc) prior to library amplification. 3’mRNA-Seq Library Prep Kit for Illumina (FWD) from Lexogen was used for first strand cDNA, second strand synthesis, dsDNA purification, i7 single indexing, library amplification and final library purification. To estimate the PCR cycle numbers for library amplification, qPCR was done by using PCR Add-On Kit (Lexogen). Purified final libraries were quality tested by using Agilent Bioanalyzer 2100 with High Sensitivity DNA chips (Agilent Technologies Inc). Qubit fluorometer (ThermoFisher Scientific) was used to quantify the concentration of the final library. Barcoded libraries are then sequenced using an Illumina HiSeq 4000 at 50 bp single-end reads in the CAT genomic facility at UCSF. There was a median of 14.3 million (M) mapped reads per sample (interquartile range, IQR: 8.0 M-20.6M).

Read mapping and quantification of gene expression

Request a detailed protocol

Read count and transcript per million reads mapped (TPM) were determined using Salmon software version 0.12.0. A reference genome index for Salmon was created according to developer’s instructions for the Drosophila melanogaster genome BDGP6 (Ensembl v92). Reads mapping and quantitation was simultaneously performed to individual transcripts.

Differential expression across heterozygous mutant flies

Request a detailed protocol

Differential expression of heterozygous mutant flies was determined by pooling samples from the same genotype. Gene expression profiles between mutant and wild type were collated using the R package tximport (version 1.6.0). The R package DESeq2 (version 1.18.1) used raw gene counts to determine differentially expressed genes (DEGs) by genotype with the linear model [Gene counts ~Batch + Genotypes]. Protein coding and lincRNA genes defined by the BDGP6 were included in differential expression. Expression was adjusted for batch to account for difference between fly lines, tissue source, and library preparation. The p-values were adjusted for Benjamini-Hochberg Procedure through DESeq2 with a target alpha = 0.1, and genes were considered DEGs at FDR < 0.05 and±50% expression changes.

Single cell Patch-seq

Request a detailed protocol

To obtain the cytoplasmic content of the cell, we performed whole-cell tight-seal patch clamp in motoneurons expressing GFP (Kulik et al., 2019). We established whole-cell configuration with leak currents less than 100 pA. We gently sucked the cytoplasmic content of the cell by applying a negative pressure to the patch-pipette. Then, we pulled the individual motoneuron from the tissue while visually confirming the GFP fluorescent signal at the tip of the pipette. Immediately after, we immersed the pipette tip in a test tube containing the Cell Lysis Buffer and RNAse inhibitor medium and broke the pipette tip by gently touching to the tube wall. The content of the pipette tip was ejected by applying positive pressure. We pooled 4–8 motoneurons for one reaction.

We used the Low Input RNA: cDNA Synthesis, Amplification and Library Generation kit from NEB (New England Biolabs Inc) to isolate, reverse-transcribe the RNA and prepare the libraries for sequencing. Following the reverse transcription and template switching, we amplified the cDNA by PCR. Amplified cDNA was cleaned up by using SPRI beads. The quality and quantity of the amplified cDNA was assessed by Bioanalyzer (Agilent Technologies Inc). After fragmentation and adaptor ligation, adaptor-ligated DNA were enriched with i7 primer and universal primer by PCR-amplification. Amplified libraries were quality checked by Biolanalyzer with High Sensitivity DNA chips (Agilent Technologies Inc) and the quantity was measured by Qubit fluorometer (ThermoFisher Scientific). Barcoded libraries were sequenced using an Illumina HiSeq 4000 at 100 bp paired-end reads in the CAT genomic facility at UCSF.

Patch-seq transcriptional analysis

Request a detailed protocol

(Raw reads were first processed with flexbar version 3.5.0 (https://github.com/seqan/flexbarRoehr, 2019) to remove adapters specific to the NEBNext library prep, using parameters as described in https://github.com/nebiolabs/nebnext-single-cell-rna-seq (Shtatland and Langhorst, 2018). The reads were then processed with HTStream v.1.1.0 (https://github.com/s4hts/HTStream; Hunter, 2019) to perform data QA/QC, remove Illumina adapter contamination, PCR duplicates, and low-quality bases/sequences.

The trimmed reads were aligned to the Drosophila melonogaster genome v.BDGP6.22 (http://ensembl.org/Drosophila_melanogaster/Info/Annotation) with annotation release version 98 using the aligner STAR v. 2.7.0e (Dobin et al., 2013) to generate raw counts per gene. On average, 93.7% of the trimmed reads aligned to the Drosophila genome, and 80% of the trimmed reads uniquely aligned to an annotated Drosophila gene.

Differential expression analyses were conducted using limma-voom in R (limma version 3.40.6, edgeR version 3.26.7, R 3.6.1). Prior to analysis, genes with fewer than five counts per million reads in all samples were filtered, leaving 8598 genes. The differential expression analysis was conducted independently for the two experiments represented in the samples.

qPCR

Request a detailed protocol

RNA was extracted from third-instar larval CNS or adult heads (5-7 days post-pupation) with RNeasy Plus Micro kit (Qiagen). RNA isolation was followed with DNase digestion with Turbo DNA-free (Ambion). For the first strand synthesis Super Script II RT was used (Invitrogen). Taqman Fast Universal PCR solution was mixed with TaqMan probe with an Applied Biosystems FAM dye. RPL32 was amplified as an internal control. Expression fold-changes are quantified by ddCT method. Data represent three biological and three technical replicates.

Electron microscopy

Request a detailed protocol

Electron microscopy experiments were performed as previously described (Harris et al., 2015). For high-frequency stimulation experiments, larval fillet preparations were fixed immediately (1–5 s) following stimulation. Data are acquired from at least two animals.

Statistical analysis of physiology and morphology data

Request a detailed protocol

Average values are presented as mean ± standard error of mean. All statistical tests are indicated in the figure legends, referring to individual panels within the figure. For multiple comparisons, we used one-way ANOVA, followed by Dunnett’s or Tukey multiple comparisons. To test the difference between two groups, we used unpaired two-tailed Student’s t-test. Pearson correlation coefficients were calculated following a linear-fit of the X-Y (quantal size vs. quantal content) data, although supra-linear best-fits are sometimes displayed, purely for the purpose of display.

References

  1. 1
  2. 2
  3. 3
  4. 4
  5. 5
  6. 6
  7. 7
  8. 8
  9. 9
  10. 10
  11. 11
  12. 12
  13. 13
  14. 14
  15. 15
  16. 16
  17. 17
  18. 18
  19. 19
  20. 20
  21. 21
  22. 22
  23. 23
  24. 24
  25. 25
  26. 26
  27. 27
  28. 28
  29. 29
  30. 30
  31. 31
  32. 32
  33. 33
  34. 34
  35. 35
  36. 36
  37. 37
  38. 38
  39. 39
  40. 40
  41. 41
  42. 42
  43. 43
  44. 44
  45. 45
  46. 46
  47. 47
  48. 48
  49. 49
  50. 50
  51. 51
  52. 52
  53. 53
  54. 54
  55. 55
  56. 56
  57. 57
  58. 58
  59. 59
  60. 60
  61. 61
  62. 62
  63. 63
  64. 64
  65. 65
  66. 66
  67. 67
  68. 68
  69. 69
  70. 70
  71. 71
    SFARI gene
    1. Simons Simplex Collection
    (2020)
    Accessed October 1, 2019.
  72. 72
  73. 73
  74. 74
  75. 75
  76. 76
  77. 77
  78. 78
  79. 79
  80. 80
    Rules of Nonallelic noncomplementation at the synapse in Caenorhabditis elegans
    1. KJ Yook
    2. SR Proulx
    3. EM Jorgensen
    (2001)
    Genetics 158:209–220.
  81. 81

Decision letter

  1. Ronald L Calabrese
    Senior Editor; Emory University, United States
  2. Hugo J Bellen
    Reviewing Editor; Baylor College of Medicine, United States
  3. C Andrew Frank
    Reviewer; University of Iowa, United States

In the interests of transparency, eLife publishes the most substantive revision requests and the accompanying author responses.

Acceptance summary:

The concept that Drosophila can contribute to the molecular underpinnings of some aspects of autism is exciting. The data that you present in the manuscript will therefore be of wide interest to researchers who study neurodevelopmental disorders like ASD as well as neuroscientists who use animal models to study synapse function.

Decision letter after peer review:

[Editors’ note: the authors submitted for reconsideration following the decision after peer review. What follows is the decision letter after the first round of review.]

Thank you for choosing to send your work, "Homeostatic plasticity commonly fails at the intersection of autism gene mutations and novel phenotypic modifiers", for consideration at eLife. Your submission has been assessed by a Senior Editor in consultation with a member of the Board of Reviewing Editors. Although the work is of significant interest and the manuscript is very ambitious, the reviewers felt that there were missing pieces of information and experiments that preclude acceptance at this stage. We know that you appreciate eLife's policy to not ask for revisions that the reviewers feel would require more than two months of work. Consequently, we are declining this version at this time. Nonetheless, we would be willing to consider a resubmission at some later date if you feel that you can deal with these critiques.

Three very knowledgeable reviewers read your work and there is a clear consensus. They all think the manuscript has the backbone of a story that could be developed into an interesting eLife paper. but they all require a substantial number of experiments. The reviewers discussed your work and one reviewer nicely summarized it all as follows.

"Regarding additional experiments, by my reading, the requested additions would be:

1) Evaluations of the chronic maintenance of homeostatic plasticity (for several key genotypes). This was requested by two separate reviews. If the consensus ends up being to move forward with a revision request, we would need to have a discussion about which genotypes are important for the authors to examine.

2) The authors need to provide a much better evaluation of CREG gene function and its role in PHP as a potential homeostatic repressor. Two separate reviews requested overexpression of CREG to demonstrate its activity as a repressor. Two reviews requested more information of its loss-of-function phenotypes. Two reviews requested better information about its expression (especially at the larval stage where the homeostatic phenotypes are measured).

3) There also seems to be a consensus that the authors would need to either: 1) redo underpowered experiments that are important for main conclusions; 2) and/or conduct pairwise combinations that were excluded without explanation (Figure 1); 3) and/or providing an affirmative rationale for choosing to follow some pairwise combinations but not others.

The same reviewer also asked "tissue-specific expression information should be shared for new genes whose roles in PHP induction are delineated (CHD8, ASH1L, CHD2, WDFY3, PPP2R5D, PDPK1, CREG). The request in this case is informational (e.g. existing tissue-specific RNAseq data or in situ data), not a request to generate de novo transgenic lines for all of these genes. I did a brief search and found the expression pattern of nearly all genes : ASHIL homologue has a Pacman clone that is tagged with GFP (Venken et al., 2006); CHD2 homologue has T2A-GAL4 (Lee et al., 2018; my web site); WDEY3 hom. has a T2A-GAL4 (same); PPP2R5D is tagged with GFP (Nagarkar-Jaiwal et al., 2015; my web site); PDPK1 is tagged with T2A_GAL4 and GFP (same); CREG has a MiMIC (Venken et al., 2011). This means that there is a plethora of reagents that are easily accessible to determine expression and in some case protein localization. Many of these are also mutants and allow many elegant experiments that would answer the reviewer’s concerns.

Reviewer #1:

The manuscript by Genç et al. details a series of genetic interactions that affect homeostatic synaptic plasticity at the Drosophila NMJ. The first main finding is that a rim/+ genetic background (RIMS1/+) sensitizes the NMJ to heterozygous loss of autism-associated gene homologs in Drosophila. The authors detail how several double heterozygous genetic conditions abrogate an acute expression of presynaptic homeostatic plasticity (PHP).

The authors searched for additional interactors, screening Drosophila Chromosome III using Deficiency (Df) lines – all in a rim/+ genetic background. The second main finding is that several Df/+ conditions synergistically impair PHP with rim/+. The Df/+ conditions also synergistically impair PHP with heterozygous loss of autism-associated gene homologs. For two cases, the authors identified single genes (PPP2R5D and PDPK1) uncovered by Df chromosomes that phenocopied the Df/+ enhancement phenotype when mutated. For a CHD8/+ PPP2R5D/+ double heterozygous condition, synaptic membranes appeared highly disorganized, and there are large synaptic vesicles.

Finally, the authors tested a hypothesis that some of these synergistic conditions could cause adaptive (or maladaptive) changes in gene transcription. Using RNAseq they found that CREG gene expression increased for several conditions, namely the CHD8/+PPP2R5D/+ condition. When tested as a potential regulator of PHP, the authors found that a CREG/+ heterozygous condition restored PHP to the impaired CHD8/+PPP2R5D/+ genetic background. They concluded that CREG was a homeostatic repressor.

Main points:

1) PhTx (tested) vs. chronic glutamate receptor loss (untested): Recent papers in the field suggest that the acute expression of PHP and the chronic maintenance of PHP have important similarities and differences. This paper only tests acute expression of PHP, using PhTx as a homeostatic challenge. Yet there is no information about how NMJs of these genotypic backgrounds respond to chronic glutamate receptor subunit loss or knockdown.

There are many unknowns about how these genes might influence synaptic function. The authors correctly state that the effects on PHP could be due to indirect cellular stress or other mechanisms. In terms of organizing this new information and drawing more informed conclusions, tests of PHP maintenance for a subset of the genotypes would be helpful - especially since maintenance might reflect some long-term transcriptional changes. Some possible genotypes to test (given their importance in the paper):

CHD8/+; RIMS1/+

PPP2R5D/+; RIMS1/+

PPP2R5D/CHD8

CHD8/+; PPP2R5D/CREG

2) Expression. Several new genes are described that are relevant to presynaptic biology and PHP. It is not at all clear where or when these genes might be expressed – and how many of the PHP-blocking conditions might affect synaptic ultrastructure like the CHD8/+ PPP2R5D/+.

To the extent that it exists, tissue-specific expression information should be shared for new genes whose roles in PHP induction are delineated (CHD8, ASH1L, CHD2, WDFY3, PPP2R5D, PDPK1, CREG). The request in this case is informational (e.g. existing tissue-specific RNAseq data or in situ data), not a request to generate de novo transgenic lines for all of these genes.

Related: The authors seem to be arguing that CREG upregulation is a special maladaptive case that impairs PHP, at least in the case of the CHD8/+, PPP2R5D/+ genetic background. There is a TRiP-overexpression (TOE) construct for CREG at Bloomington. One clear prediction from the paper is that if this TOE line were used to upregulate CREG (in the correct tissue) that it would block PHP – and maybe it would affect ultrastructure too?

3) Ultrastructure (Figure 6). These images are beautiful, but it is not clear what these phenotypes mean regarding PHP – or if these phenotypes might be generalizable to the other autism gene interactions. The large membrane structures are reminiscent of the large structures seen in endocytosis mutants. Endocytic mutants at the NMJ often have an electrophysiological signature (e.g. large minis). Were any recordings conducted post-stimulation? The baseline physiology for the PPP2R5D/+; CHD8/+ condition (Supplementary file 1 table 7) does not look like an endocytic mutant, but maybe post-tetanus it would.

4) This reviewer found some claims that pieces of data provided the first evidence of a phenomenon, but the claims were not correct, even within the realm of the Drosophila NMJ itself. Specific Examples:

The idea that CHD8 had never before been connected to presynaptic biology is not right. At the Drosophila NMJ, there are well-documented defects for kismet mutants in both development and presynaptic neurotransmitter release.

The idea that CREG is the first-ever documented repressor of PHP in any system is not correct. Loss-of-function conditions at the NMJ have been documented that relieve blocks of PHP maintenance (two examples are loss of FasII and loss of 4E-BP). By the criteria set out by the current data set, these earlier examples would also constitute homeostatic repressors.

Reviewer #2:

The manuscript by Genc and colleagues examines the loss of presynaptic homeostatic plasticity (PHP) in heterozygous combinations as a model for polygenic disorders. Using the loss of PHP in trans-heterozygous combination as readout, the authors uncovered genetic interactions among unrelated autism genes (RIMS1, CHD8, CHD2, WDFY3 and ASH1L) with fly orthologs. They expanded this strategy to screen for novel modifiers of PHP using the RIMS1/+ background and a series of overlapping deficiencies on 3rd chromosome. The screen identified modifiers in 20 lines (about 10% of the Df screened). Interestingly, some of these RIMS1- interacting Dfs also caused PHP defects when screened against the other autism genes (CHD8, CHD2, WDFY3 and ASH1L). The authors mapped two of the interacting loci to a kinase and phosphatase, PDPK1 and PPP2R5D, and confirmed that they interact genetically with most of the autism genes in this study. However, neither enzyme is required for PHP; also, the molecular basis for their observed interaction with RIMS1 and other loci remains unknown. Finally, the authors searched for differentially expressed transcripts (using RNAseq from adult heads) and reported that CREG was elevated in CHD8/+; PPPR25D/+ trans-heterozygotes. Importantly, lowering the dose of CREG rescued the PHP defects of CHD8/+; PPPR25D/+. How CREG restores the PHP in this trans-heterozygote and whether CREG is elevated in any of the other synthetic phenotypes remain open questions.

The study aims to model the genetic heterogenicity of autistic patients. The approach is sound, and the amount of data is substantial and of high quality, but the manuscript only tackles genetic interactions that the authors cannot explain nor connect to any previous PHP pathways. These numerous synthetic phenotypes seem merely "synthetic" without some anchoring into known phenomena; the same is true for the accompanying EM micrographs.

1) The description of genetic interaction concepts is unnecessarily long and convoluted and must be consolidated.

2) The screen itself is poorly described. Did the authors use the Bloomington Df kit for their screen? If not, specify. A table with the reagents used here should be provided.

3) The authors chose to compare a small and a large Df in lieu of examining additive effects for gene heterozygosity (Figure 3). These arguments are weak. Instead, a more detailed description of the screen results would be more meaningful and should be included here.

For example, how many of the Dfs tested contain known PHP modulators? What was the outcome of the screen for those Dfs? How does that outcome compare with prior single gene analyses?

4) When a Df covering a known PHP player did not show the expected interaction, what does this mean for the results of the screen?

5) It is somewhat anticlimactic that the PDPK1 and PPP2R5D are not required for acute PHP. However, the authors hint to a possible role in chronic homeostasis, more similar to the mTOR signaling components. This probably has already been examined and should be included here.

6) The RNAseq analysis was performed with adult heads due to understandable limitations. But the qPCR validation of CREG up-regulation must be done in third instar larvae, at the stage when PHP is recorded.

7) The authors propose that CREG is a homeostatic repressor. This predicts that overexpression of CREG will block PHP. The authors should directly test this prediction which will strengthen their conclusions.

8) Reducing the dose of CREG would be expected to restore PHP in other trans-heterozygote combinations. This should be examined here since a positive result could expand the relevance of this finding beyond an isolated trans-heterozygous case presented here.

9) Another elegant solution to link the disconnected parts of this study would be to overexpress CREG in some of the single heterozygotes (RIMS1, CHD8, CHD2, WDFY3, ASH1L – or PDPK1, PPP2R5D) and block PHP.

10) Using qPCR to search for increased levels of CREG in (third instar) mutants lacking PHP would be an easy, fast way to implicate CREG in other known PHP pathways and will tremendously benefit the author's arguments.

Reviewer #3:

Genc et al. present in this study findings arguing that five seemingly unrelated autism associated genes are linked to presynaptic homeostatic plasticity (PHP) and synaptic transmission. They propose through genetic interaction studies in Drosophila that phenotypic enhancers could influence the robustness and phenotypic penetrance of these autism associated gene variants. Although several aspects of this study appear to support their conclusion, the reliability and reproducibility of their findings is hampered by conceptual and technical concerns. Addressing these concerns would help to strengthen the study.

1) The authors propose that five selected autism genes converge on PHP, but they only examine the genetic interaction between two pairs (RIMS1 and CHD8) and (RIMS1 and ASH1L). It is unclear why the authors do not complete the pairwise interaction studies for the remaining two genes. No rationale is provided to explain this incomplete analysis.

2) Due to the possibility of second site mutations contributing to the genetic interaction findings, it is important to assess other loss of function alleles of RIMS1, CHD8, and ASH1L in the genetic interaction studies. As well as for the remaining two selected autism genes. Furthermore, can these findings be ameliorated with genomic rescue of RIMS1, CHD8, ASH1L, etc in the pairwise genetic interaction studies?

3) What is the rationale for choosing only five of the 20 deficiencies that showed interactions with RIMS1 to test their interactions with other gene? The authors need to provide a rationale for this selection. Otherwise it raises the question as to whether the remaining deficiencies were tested but did not show as strong of an interaction effect.

4) Critical resource and technical information is lacking, which make it difficult to interpret or reproduce the study in the future. In particular, fly stocks (origin, type of mutation, specific breakpoints, stock center, etc) was not reported. What were the specific deficiency lines – including the "20 small deficiencies that cause PHP to fail"?

5) It is often unclear in the manuscript and figures if the sample size "n" is referring to animals, neurons, synapses, vesicles, etc. The sample size is also very low (n of only three) for some of the studies, raising the concern as to whether the studies were adequately powered to detect a statistical difference. For example, "we performed multiple (n=3-15) intracellular recordings".

6) The authors argue that CREG is a homeostatic repressor that blocks PHP and regulates synaptic ultrastructure (Figure 8). It would be important to show the phenotype of the CREG single mutant. What is the ultrastructural phenotype of the CREG single mutant?

7) EM studies are performed in tissue fixed 1-5 seconds after high frequency stimulation with a sample size as low as only 2 animals. How does the difference between 1 to 5 seconds post-high frequency stimulation affect the findings?

8) The rationale for selecting RIMS1, CHD8, CHD2, WDFY3, and ASH1L is still unclear. The authors state they were selected as they were "category 1 or 2 ASD-associated genes based on SFARI Gene". However these two categories encompass many more than these five genes. Were these the only genes with fly orthologs and available resources? Were there other reasons for choosing these genes?

9) The authors need to be careful not to overstate or misstate their findings. Particularly as the premise of the paper hinges upon 5 selected genes associated with autism, and far more than 5 genes are implicated in autism. Providing a better rationale for selecting these 5 genes would help frame their findings better.

[Editors’ note: further revisions were suggested prior to acceptance, as described below.]

Thank you for submitting your article "Homeostatic plasticity fails at the intersection of autism-gene mutations and a novel class of common genetic modifiers" for consideration by eLife. Your article has been reviewed by three peer reviewers, and the evaluation has been overseen by Hugo Bellen as the Reviewing Editor and Ronald Calabrese as the Senior Editor. The following individual involved in review of your submission has agreed to reveal their identity: C. Andrew Frank (Reviewer #1).

The reviewers have discussed the reviews with one another and the Reviewing Editor has drafted this decision to help you prepare a revised submission.

We would like to draw your attention to changes in our revision policy that we have made in response to COVID-19 (https://elifesciences.org/articles/57162). Specifically, when editors judge that a submitted work as a whole belongs in eLife but that some conclusions require a modest amount of additional new data, as they do with your paper, we are asking that the manuscript be revised to either limit claims to those supported by data in hand, or to explicitly state that the relevant conclusions require additional supporting data.

Our expectation is that the authors will eventually carry out the additional experiments and report on how they affect the relevant conclusions either in a preprint on bioRxiv or medRxiv, or if appropriate, as a Research Advance in eLife, either of which would be linked to the original paper.

Another issue that was raised and all reviewers agreed upon is that one cannot tell which 18-20 interacting Dfs fall into the gray shaded area of Figure 2B. This would be a matter of a text revision or a table that indicates something about the physiological data of those Dfs identified in the screen. Specifically, save a few labeled examples, the data points in the figures do not actually match up with specific Dfs, and therefore, it is difficult to interpret the findings in the context of what was expected/unexpected a priori. To address this, the authors need to do a couple of things. First, they would need to publish a long supplemental table or Excel file for all 168 screened genotypes. The type of data expected are: genotype, average mini amplitude, average evoked amplitude, calculated quantal content, and n. This should not be too onerous to do (depending on how the data are currently organized).

The reviewers agreed that the main message, combinatorial loss of autism-associated genes can result in profound synaptic phenotypes, including blocked PHP, is valuable. This is an interesting and novel idea in itself, and worthy of publication in eLife.

Below are issues that all can be addressed in a timely fashion.

Summary and General Evaluation

Complex neurodevelopmental disorders like Autism Spectrum Disorder (ASD) are not well understood, either on genetic and neurophysiological levels. The study by Genç et al. uses a genetic model to tackle this knowledge gap. The study uses the fruit fly Drosophila melanogaster and its model synapse, the neuromuscular junction (NMJ). The core finding is clear: in Drosophila, double heterozygous combinations of ASD-related genes can manifest in striking synaptic phenotypes. Moving forward, it stands to reason that genetic combinations in Drosophila system (and other models) could be exploited, potentially to understand biology underlying ASD and other neurodevelopmental disorders.

The work reports novel scientific findings. First, double heterozygous losses of Drosophila homologs of high confidence ASD genes impair a form of homeostatic synaptic plasticity, called presynaptic homeostatic potentiation (PHP). Second, heterozygous Deficiency (Df) lines synergistically cause PHP to fail when combined with a rim/+ (RIMS1/+) genetic background; a subset of those Df lines synergistically interact with other heterozygous ASD gene combinations to impair PHP expression. Third, the authors identify two genes – PPP2R5D and PDPK1 – uncovered by two of the Df chromosomes that interacted with RIMS1/+. Heterozygous loss of PPP2R5D and PDPK1 phenocopy the corresponding Df/+ condition. Moreover, heterozygous loss-of-function mutations in PPP2R5D and PDPK1 interact with other heterozygous ASD gene homologs, again resulting in instances of PHP failure.

The authors follow one genetic combination as an example. By doing this, they looked deeper, on a level of cell biology (by EM) and on a level of gene transcription (by RNAseq), and they complemented that work with electrophysiological analyses. For the CHD8/+ PPP2R5D/+ double heterozygous combination, EM shows that synaptic membranes at the NMJ are highly disorganized, there are large synaptic vesicles, and there is clear evidence of endocytic defects both visually and by electrophysiology. Using RNAseq, the authors find that CREG is upregulated in the CHD8/+ PPP2R5D/+ condition. They test the idea that CREG upregulation is maladaptive for the synapse. Partial loss of CREG reverses the electrophysiology and EM phenotypes of the CHD8/+ PPP2R5D/+ genetic combination. Consistently, transgenic overexpression of CREG blocks PHP, and the authors conclude that CREG is a novel suppressor of homeostatic plasticity.

A prior review requested testing the chronic maintenance of PHP vs. the acute induction of PHP.

The paper examines the acute induction of PHP. For the prior submission, this reviewer offered that examining the chronic maintenance of PHP would be helpful – especially since ASD-related mutations could represent modeling a chronic condition. The authors made a counterargument that the PHP induction paradigm is a stress test of sorts – and their core findings remain robust with this test, even in the absence of testing long-term maintenance. This is well reasoned. The authors correctly state that testing the maintenance of PHP could be a logical extension of the work, but the subject of a different study. This is a point added to the text of the article.

The prior review requested more evaluation of CREG gene function as a novel homeostatic repressor.

The set of experiments pursued by the authors is very nicely done and firmly establish CREG as a homeostatic repressor at the NMJ. They also implicate CREG upregulation in the EM endocytic phenotypes.

The prior review requested more complete analyses of the possible ASD-related gene combinations.

The authors have responded to this request experimentally. The main new work is summarized in Figure 1—figure supplement 2, which summarizes all of the pairwise one-gene combinations of ASD-related heterozygous conditions.

The prior review requested expression data for new genes characterized in this study.

The authors responded to this request in a creative way, by performing a PatchSeq analysis and examining gene expression in motor neurons that innervate larval body wall muscles. The results are consistent with the electrophysiology: both types of motor neurons tested express the genes tested in the study, and at similar levels. Those data are summarized in Figure 1—figure supplement 1. Future studies can use this as a starting point to interrogate the roles that those genes and their gene products might play in motor neurons. Roles in other tissue types are also possible.

Revisions required for this paper:

1) Clarify or modify broad statements about lack of information or claims of "novelty", that falters in the context of existing information in the literature. The study will be strengthened by building connections with existing knowledge, which will minimize the impression that these findings are "synthetic" without anchoring in known phenomena.

a) In discussing the random selection of RIMS1, CHD8, CHD2, WDFY3, and ASH1L the authors state that there are no known biochemical and genetic interactions among these five genes in Drosophila, and no known direct biochemical interactions in humans. However, it is curious that the authors fail to also state the potential for interactions between these genes by incorporating many findings in mouse models indicating a potential convergence between these genes on neurodevelopment and function. Presenting this at the outset will further strengthen the study by utilizing prior literature to support the probability of identifying disease relevant genetic interactions.

b) By extension, the known disease associations for these genes should be presented and referenced, as well as if they are all impacting similar neurologic functions in humans. All five genes have known disease associations: OMIM: RIMS1 (#603649, Cone-rod dystrophy type 7), CHD8 (#615032, Susceptibility to autism type 18), CHD2 (#615369, Epileptic encephalopathy childhood onset), WDFY3 (#617520, Primary microcephaly type 18), and ASH1L (#617796, autosomal dominant mental retardation type 52).

c) Heterozygous loss of function mutations in CHD8, CHD2, WDFY3, and ASH1L all cause neurodevelopmental disorders, including autism, epilepsy, and microcephaly. In the context of the human data, the authors' finding that PHP in the single heterozygous flies is normal suggests that PHP may not be the most important mechanism driving autism phenotypes in neurodevelopmental disorders. In contrast, although pathogenic variants in RIMS1 have only been reported with a photoreceptor phenotype in humans, there is a recent publication by Peter et al., 2019 (PMID 30949922) presenting evidence in humans for pleiotropic and complex effects involving RIMS1 in sporadic autism spectrum disorder. Therefore, the authors can consider strengthening the conceptual impact of their study from the perspective that variants in RIMS1 may increase the susceptibility to autism spectrum disorder but require other polygenic factors to manifest the disorder.

d) The authors state that the "five ASD gene orthologs do not share any known common biological activity. Therefore, the identification of common genetic modifiers is completely unexpected." This is a confusing statement to make given that CHD8 and CHD2 are in the same subfamily of chromodomain helicase DNA-binding proteins, indicating a common biological activity, and ASH1L localizes to the chromatin. Based on these common functions and subcellular localization it is not entirely unexpected to find common genetic modifiers. These statements should be revised. Noting the potential for common biological activity will not detract from the potential impact of the authors' findings of common genetic modifiers through their screening approach.

e) The authors state that "there is no clear connection, biochemically or genetically, to the role of PDPK1 or PPP2R5D in the nervous system." The authors need to clarify this statement as there are rodent models demonstrating a genetic role for PDPK1 and PPP2RD5 in the nervous system. There is altered brain development in PDPK1 KO mice (Lawlor et al., EMBO J, PMID 12110585) and abnormal tau pathology in the brains of PPP2RD5 KO mice (Louis et al., PNAS, PMID 21482799). Based on the literature there is clear connection for a role of PDPK1 and PPP2R5D in the nervous system.

f) The authors propose that upregulation of CREG in PPP2R5D/+;CHD8/+ underlies the failure of PHP, which is further supported by the complete block of PHP when CREG is overexpressed ubiquitously or in motor neurons. They identified two CREG mutant alleles with differential effects on normalizing CREG levels in Figure 8, panel D (m2 restored the double mutant to wildtype, but not m1) suggesting that m2 is possibly a stronger loss of function allele. But surprisingly, m2 seems to have a reduced rescue effect compared to m1 (Figure 8, panel F). Is there a difference in CREG expression level between these two alleles? The authors only report the m1 transposon completely abolishes CREG expression and do not show the data. They do not report the effect of the m2 transposon. The effect on CREG expression from both alleles should be examined and shown as this may indicate a dosage sensitivity for CREG.

g) A rationale for selecting only 5 of the 20 deficiencies showing interactions with RIMS1 is still lacking. Were they selected based on the type of genes in the deficiencies (brain specific in mammals), known or unknown function of the genes, prior implications in homeostatic plasticity, availability of reagents, strength of the interaction effect, etc?

2) Technical concerns

a) There appears to be incomplete presentation of some datasets. The authors report that "altered NMJ growth was not commonly observed in the majority of genetic interactions tested (Figure 5—figure supplement 1)", implying that multiple pairwise interactions were examined. In fact, this is not the case. Figure 5—figure supplement 1 shows NMJ morphology data for wildtype, single heterozygotes for PDPK, PPP2R5D, CHD8, CHD2, ASH1, WDFY3. The only genetic interactions shown were for CHD8 with PDPK1 and CHD8 with PPP2R5D. The data actual shows an increase in bouton number for the CHD8/+;PPP2R5D/+ genotype compared to wildtype and PPP2R5D/+ alone, it did not reach statistical significance compared to CHD8/+ alone but perhaps this is due to the sample size? The data for the other pairwise genetic interactions should also be included in order for the authors to make the statement that altered NMJ growth was not commonly observed.

b) Sample size still remains unclear for Figures 5—figure supplement 2, Figure 5—figure supplement 3, Figure 8F, Figure 8K. Please clarify the sample size so that the interpretation and reproducibility of the data is improved.

c) In Figure 8, what does the sample size "n" in panels H and I represent? The authors note in the figure legend that "individual data points shown for indicated genotypes". Clarify if the individual data points represent individual vesicles, average vesicle size per synapse, or average vesicle size per animal. How many synapses were quantified per animal? How many animals per genotype?

d) Similarly, in Figure 6, what is the sample size representing? Individual vesicles, average size of vesicles per boutons, per animal? How many animals and synapses per animal were assessed?

e) It is important to provide all the necessary information about reagents and resources used for rigor and reproducibility. The authors now include a table S1 showing all of the deficiency stocks that were used in the study. But it remains unclear which of the listed 168 deficiency stocks were the "20 small deficiencies that cause PHP to fail when combined with RIMS1/+". The authors provide a schematic showing the distribution of these deficiencies in Figure 2C, but it is important to also denote in Table S1 the stock ID number corresponding to these 20 small deficiencies. This should be very straightforward to indicate in table S1.

f) The authors assessed the consequences of CREG heterozygous and homozygous loss of function mutations on baseline neurotransmission and PHP and describe the findings in the text. This is an important control data. For comparison to the other genotypes it would be helpful to also include the data graphically in Figure 8, panel F.

g) You stated: "First, we present every single deficiency in the graph (Figure 3). We provide representative examples of one large and one small deficiency. The reviewer seems to have misread the figure and our text in this regard." The reviewers did not find the data, legend or text presenting the interacting deficiencies. Figure 3 has no such graph. They are wondering whether you refer to the drawing in Figure 2C. Do the authors imply that this is the summary result that presents every single deficiency? Are we missing something? Moreover, the stock numbers for the Bloomington deficiencies provided in the table do not match the ones described in the text.

Information is needed for why data sets are not shown (ie CREG expression level in m1 allele), clarity on what the sample size represents in multiple datasets, details as to how many synapses, boutons, animals are included in the NMJ and EM datasets.

https://doi.org/10.7554/eLife.55775.sa1

Author response

[Editors’ note: the authors resubmitted a revised version of the paper for consideration. What follows is the authors’ response to the first round of review.]

Three very knowledgeable reviewers read your work and there is a clear consensus. They all think the manuscript has the backbone of a story that could be developed into an interesting eLife paper. but they all require a substantial number of experiments. The reviewers discussed your work and one reviewer nicely summarized it all as follow.

"Regarding additional experiments, by my reading, the requested additions would be:

1) Evaluations of the chronic maintenance of homeostatic plasticity (for several key genotypes). This was requested by two separate reviews. If the consensus ends up being to move forward with a revision request, we would need to have a discussion about which genotypes are important for the authors to examine.

We respectfully disagree. We pioneered the field of presynaptic homeostatic plasticity and are well aware of the emerging distinctions between the acute induction of PHP and the effects induced by a genetic mutation in the glutamate receptor (Frank et al., 2006; Frank et al., 2009). Indeed, in a recent publication, we delineated a bifurcating signaling cascade that controls both the rapid induction and long-term expression of homeostatic plasticity (Harris et al., 2017). However, we would like to point out that the injection of PhTx into Drosophila larvae was originally performed and demonstrated to induce a lasting homeostatic effect, persisting for many hours (Frank et al., 2006). It remains unclear how PhTx-induced homeostatic plasticity relates to GluRIIA induced homeostatic plasticity. One might be an extension of the other, or they might be independent effects with over-lapping mechanisms.

In justifying a request for substantial addition of new data (analysis of new triple and quadruple mutants and analyses thereof) the reviewers argue that GluRIIA-induced PHP is most relevant to autism. We disagree for the following reasons:

a) Currently, there has yet to be a clear experimental link between PHP and the mechanisms of autism. Thus, we have already made an advance and it was achieved systematically across many genotypes.

b) We use the acute induction of PHP to probe the capacity of the nervous system to respond to a homeostatic challenge. This is the sole purpose of our study. Consider the following analogy: It is common for physicians to diagnose pre-diabetes by challenging a patient with a glucose tolerance test. This is an acute perturbation, and the physician determines whether the individual is capable of robust, glucose homeostasis. This is precisely the logic that we use. Can the nervous system respond to an acute challenge with a robust homeostatic response? This could reasonably be directly relevant to ASD.

c) Is the GluRIIA mutation more directly relevant to ASD, as argued by the reviewers? To our knowledge, there is no evidence for a systematic change in AMPA receptor abundance in the ASD brain (changes do occur in fragile-X animals, but not generally across many different models). So, the argument that there is some unique importance of using the GluRIIA mutation is not well substantiated. Indeed, a counter argument is possible. Because the GluRIIA mutant is present throughout neural development, there is the very real possibility of creating effects in the GluRIIA-ASD-modifier triple mutants that are unique and not relevant to ASD. By using an acute perturbation, any developmental effects are restricted to the ASD orthologs and newly identified modifiers that we are studying.

d) We generally agree that the GluRIIA-dependent mechanisms of PHP could be a logical, future extension of our work. But, any such analysis should be performed in the same systematic manner as performed here – a thorough interrogation of many dozens of different genotypes, clearly beyond the scope of our current paper.

e) Finally, we have added two paragraphs to our Discussion regarding the difference between the acute induction and long-term expression of homeostatic plasticity. The paragraphs are copied here:

“The rapid induction versus long-term expression of PHP

There are two well-established methods to induce expression of PHP. Application of PhTx induces PHP within minutes, a process that can be maintained for hours (Frank et al., 2006). […] In the future, it will be interesting to systematically determine whether the gene-gene interactions identified here also uniformly perturb PHP induced by the GluRIIA mutation. However, such an analysis is beyond the scope of the present study.”

2) The authors need to provide a much better evaluation of CREG gene function and its role in PHP as a potential homeostatic repressor. Two separate reviews requested overexpression of CREG to demonstrate its activity as a repressor. Two reviews requested more information of its loss-of-function phenotypes. Two reviews requested better information about its expression (especially at the larval stage where the homeostatic phenotypes are measured).

We thank the reviewers for prompting us to do these experiments. We have addressed all three requests for additional information.

a) We have generated the necessary transgenic animals and performed the requested analyses. Overexpression of CREG, either specifically within motoneurons or ubiquitously, blocks homeostatic plasticity without affecting baseline transmission. The new data are added to Figure 8.

b) We have added an analysis of the loss of function mutation, focusing on the mutation that abolishes CREG expression. There is no effect on baseline transmission or PHP. The new data are added to the text.

c) We have added an extensive amount of information regarding CREG expression including assessing CREG expression in motoneurons at four stages of neural development and assessing the up-regulation of CREG in larval motoneurons. These data have been added as a new supplemental figure (new Figure 8—figure supplement 1).

3) There also seems to be a consensus that the authors would need to either: 1) redo underpowered experiments that are important for main conclusions; 2) and/or conduct pairwise combinations that were excluded without explanation (Figure 1); 3) and/or providing an affirmative rationale for choosing to follow some pairwise combinations but not others.

We have performed additional experiments of pairwise gene-gene combinations as requested regarding the data originally presented in Figure 1. We now demonstrate that RIMS1/+ interacts with CHD2/+ and ASH1L/+ but not with WDFY3/+. These data support our existing models and greatly strengthen the paper. We thank the reviewers for prompting us to do this work. The data are now presented as a new supplemental figure (new Figure 1—figure supplement 2).

Comment 3, continued: “redo underpowered experiments”

The comment regarding sample size is unfair. The reviewer quotes a statement that specifically refers to our screen (for which no statistical comparisons are made between genotypes) – yet the reviewer implies that this statement reflects upon other data in our paper (which is factually incorrect). We stated in our original paper:

“The screen that we performed is diagrammed in Figure 2A. We took advantage of a collection of small chromosomal deficiencies (5-50 genes per deficiency, each with known chromosomal breakpoints; listed in Supplemental Table 1) that tile the 3rd chromosome, uncovering approximately 6000 genes in total. For every double heterozygous combination of RIMS1/+ with a heterozygous deficiency, we performed multiple (n=3-15) intracellular recordings, quantifying mEPSP amplitude, EPSP amplitude, quantal content (EPSP/mEPSP), resting membrane potential and input resistance.”

A genetic screen, such as ours, is designed to identify “hits” as efficiently as possible. In our genetic screen, as is common practice, we do not make any claims regarding the statistical validity of any individual observation. The only purpose of a screen is to identify “hits”. We establish well-justified parameters to select “hits”. This is done in a conservative fashion in order to protect the experimenter from performing un-necessary follow-up work. Importantly, every “hit” is re-validated. We are now more explicit, stating, “To achieve a final list of 20 hits, each potential hit was re-validated in a

second set of experiments, increasing sample sizes (generally 7-16 NMJ). During the process of revalidation, we rule out approximately one third of the potential hits selected from the screen.”

Furthermore, the reviewer is referred to the fact that multiple “hits” from our screen were re-validated, multiple times throughout our manuscript. First, the hits were identified. Second, they were revalidated as a hit. Third, five of the deficiencies were incorporated into a matrix of multiple gene-gene interactions, a further re-validation. Finally, two deficiencies were resolved to individual genes, requiring multiple rounds of re-validation. And, the single genes precisely recapitulate the results of the deficiencies in which they originally resided. Surely, at this point, the extensive nature of revalidation, with statistically relevant sample sizes at every stage, should inspire statistical confidence.

How did we determine adequate sample sizes? We performed a power analysis to estimate necessary sample size. We have an effect size of 150-200% with a low standard deviation. Estimation of sample size suggests a minimal sample size of three to four, which is obviously very small. In our primary screen, when efficiency is emphasized, we use a sample size of 3-15, as reported. None of the data from the screen are compared statistically – the use of sample sizes here is simply for transparent reporting. In our study, when statistical comparisons are made, sample sizes were never less than n=7. In instances where effect size was on the smaller end of the range, sample sizes are generally larger to ensure confidence.

In our study, we analyze more than 40 genotypes electrophysiologically, under two conditions (with and without PhTX), representing more than 80 experimental conditions – not including our screen. In only two experiments out of more than 80 experimental conditions are sample sizes equal to 4 (in the original submission, there were four instances, and new data have been incorporated). These instances were controls that adhered to high effect size and low variance. In every other experiment (>80 conditions) sample size was never less than 7, generally in excess of 10 and often greatly exceeding these sizes. All the data are transparently presented in figures with individual data points and in tables with sample sizes. In every instance, sample sizes were sufficiently powered and all statistical tests and values reported.

Comment 3, “provide affirmative rationale for choosing to follow some pairwise combinations but not others”

If we understand this criticism correctly, the reviewer seems to suggest that we follow every single double heterozygous combination (36 were systematically tested), a request that seems out of line with the existing literature and reviewing standards.

It is possible that the reviewer simply wants additional justification for why we chose to follow the CHD8/+; PPP2R5D/+ double heterozygous mutant. It was a highly penetrant double mutant and CHD8 is one of the most well-established risk factors for ASD. As we now state in our updated text:

“It is rare for a genetic study to define, precisely, how a double heterozygous interaction creates a synthetic phenotype if the two genes do not encode proteins that biochemically interact. Simply put, there are a vast number of possible mechanisms by which SSNC could occur (Yook et al., 2001). None-the-less, we attempted to do so for at least one double heterozygous combination. Although this represents only a single mechanism of SSNC, it could provide proof of principle for how PHP is affected in other ASD gene interactions. We chose the genetic interaction of PPP2R5D/+ with CHD8/+. This combination was chosen because CHD8 is among the most common ASD de novo gene mutations. Furthermore, the genetic interaction is highly penetrant.”

Reviewer #1:

The manuscript by Genç et al. details a series of genetic interactions that affect homeostatic synaptic plasticity at the Drosophila NMJ. The first main finding is that a rim/+ genetic background (RIMS1/+) sensitizes the NMJ to heterozygous loss of autism-associated gene homologs in Drosophila. The authors detail how several double heterozygous genetic conditions abrogate an acute expression of presynaptic homeostatic plasticity (PHP).

The authors searched for additional interactors, screening Drosophila Chromosome III using Deficiency (Df) lines – all in a rim/+ genetic background. The second main finding is that several Df/+ conditions synergistically impair PHP with rim/+. The Df/+ conditions also synergistically impair PHP with heterozygous loss of autism-associated gene homologs. For two cases, the authors identified single genes (PPP2R5D and PDPK1) uncovered by Df chromosomes that phenocopied the Df/+ enhancement phenotype when mutated. For a CHD8/+ PPP2R5D/+ double heterozygous condition, synaptic membranes appeared highly disorganized, and there are large synaptic vesicles.

Finally, the authors tested a hypothesis that some of these synergistic conditions could cause adaptive (or maladaptive) changes in gene transcription. Using RNAseq they found that CREG gene expression increased for several conditions, namely the CHD8/+ PPP2R5D/+ condition. When tested as a potential regulator of PHP, the authors found that a CREG/+ heterozygous condition restored PHP to the impaired CHD8/+ PPP2R5D/+ genetic background. They concluded that CREG was a homeostatic repressor.

Main points:

1) PhTx (tested) vs. chronic glutamate receptor loss (untested): Recent papers in the field suggest that the acute expression of PHP and the chronic maintenance of PHP have important similarities and differences. This paper only tests acute expression of PHP, using PhTx as a homeostatic challenge. Yet there is no information about how NMJs of these genotypic backgrounds respond to chronic glutamate receptor subunit loss or knockdown.

There are many unknowns about how these genes might influence synaptic function. The authors correctly state that the effects on PHP could be due to indirect cellular stress or other mechanisms. In terms of organizing this new information and drawing more informed conclusions, tests of PHP maintenance for a subset of the genotypes would be helpful -especially since maintenance might reflect some long-term transcriptional changes. Some possible genotypes to test (given their importance in the paper):

CHD8/+; RIMS1/+

PPP2R5D/+; RIMS1/+

PPP2R5D/CHD8

CHD8/+; PPP2R5D/CREG

We respectfully disagree with this reviewer see above for our response to major comment 1.

2) Expression. Several new genes are described that are relevant to presynaptic biology and PHP. It is not at all clear where or when these genes might be expressed – and how many of the PHP-blocking conditions might affect synaptic ultrastructure like the CHD8/+ PPP2R5D/+.

To the extent that it exists, tissue-specific expression information should be shared for new genes whose roles in PHP induction are delineated (CHD8, ASH1L, CHD2, WDFY3, PPP2R5D, PDPK1, CREG). The request in this case is informational (e.g. existing tissue-specific RNAseq data or in situ data), not a request to generate de novo transgenic lines for all of these genes.

The reviewers asked that we determine the expression of each of the genes that we analyze in our manuscript, inclusive of seven genes. The reviewers suggested a number of resources including transposable element-mediated gene tags, some of which remain un-validated. We reasoned, however, that the only way that we could quickly and definitively address this comment was to pursue a Patch-Seq gene-expression profiling experiment, assaying gene expression selectively in the two cells that we analyze electrophysiologically.

We performed Patch-Seq experiments analyzing gene expression in type 1b and type 1s motoneurons. We demonstrate that all of the genes studied in our manuscript are expressed in both of these motoneuron types (at very similar levels). These data are now included as a new supplemental figure (Figure 1—figure supplement 1).

There is added significance to the addition of these new data. We recently demonstrated that homeostatic plasticity is globally induced at all connections to muscle 6, but the expression of homeostatic plasticity can vary depending upon the experimental conditions. Some conditions favor expression by type 1b motoneurons and some by type 1s motoneurons (Genc et al., 2019). By demonstrating that all ASD genes queried in our study are expressed by both motoneurons, it seems likely that genetic mutations will perturb both cells. Indeed, the expression levels are remarkably similar comparing MN1s and MN1b.

(continued)…how many of the PHP-blocking conditions might affect synaptic ultrastructure like the CHD8/+ PPP2R5D/+.

This reviewer seems unaware of the low-throughput nature of electron microscopy. We originally provided data for two controls, two heterozygous mutants, a double mutant, and the triple heterozygous mutant combination (an experiment that included its own independent control).

Furthermore, we now have added three more genotypes to our EM analysis based on this request. Finally, the reviewer should also note that we now include an electrophysiological analysis to support our conclusions made based on electron microscopy (see below).

Electron microscopy is not a screening tool. The HHMI funded laboratory of Dr. Bellen has been able to fully support such an effort in the past, but that is both remarkable and almost completely unique. Please note that the majority of papers examining homeostatic plasticity at the NMJ do not include quantitative EM, inclusive of papers recently published in eLife by other laboratories.

Related: The authors seem to be arguing that CREG upregulation is a special maladaptive case that impairs PHP, at least in the case of the CHD8/+, PPP2R5D/+ genetic background. There is a TRiP-overexpression (TOE) construct for CREG at Bloomington. One clear prediction from the paper is that if this TOE line were used to upregulate CREG (in the correct tissue) that it would block PHP – and maybe it would affect ultrastructure too?

This is a reasonable prediction and something that we have been able to test directly. We have added two types of data to our manuscript that address this issue:

a) We now present the developmental expression profile for CREG in motoneurons at four time points including the embryo, 1st instar, 2nd instar, early third instar and late third instar. The data demonstrate that CREG is expressed highly in the embryo and then is turned off. However, in the double heterozygous mutant combination, CREG is induced at the late third instar stage (4-fold induction). These new data clearly validate our assertion that CREG is up-regulated in a specific manner in the double mutant.

b) We generated a UAS-CREG transgenic line to drive CREG expression in a tissue-specific manner. Over-expression of CREG either ubiquitously (Tubulin-GAL4) or specifically in motoneurons (OK371-GAL4) is sufficient to completely block homeostatic plasticity. We did not perform additional ultrastructure.

3) Ultrastructure (Figure 6). These images are beautiful, but it is not clear what these phenotypes mean regarding PHP – or if these phenotypes might be generalizable to the other autism gene interactions. The large membrane structures are reminiscent of the large structures seen in endocytosis mutants. Endocytic mutants at the NMJ often have an electrophysiological signature (e.g. large minis). Were any recordings conducted post-stimulation? The baseline physiology for the PPP2R5D/+; CHD8/+ condition (Supplementary file 1 table 7) does not look like an endocytic mutant, but maybe post-tetanus it would.

There are two points that we would like to make.

a) We now include an entire figure in which we analyze synaptic rundown, a parallel test of a defect in vesicle endocytosis. These data are presented in new Figure 6—figure supplement 1. The data argue in favor of an endocytic defect, as reported by EM.

b) We agree that the EM effect may or may not be directly related to homeostatic plasticity. Nonetheless, the EM provides a striking visual demonstration of the effects of the double mutant combination. And, it is clearly a consequence of increased CREG in the double mutant context based on the rescue of the EM in the triple mutant. We believe that these data will be important for the field, as it might encourage those laboratories studying synapses in the rodent central nervous system to examine synaptic ultrastructure. We state in our text:

“These data provide a striking visual confirmation of the genetic interaction between PPP2R5D/+ and CHD8/+. And, this is further evidence linking the action of a chromatinremodeling factor (CHD8) to the stability of synaptic transmission.”

4) This reviewer found some claims that pieces of data provided the first evidence of a phenomenon, but the claims were not correct, even within the realm of the Drosophila NMJ itself. Specific Examples:

The idea that CHD8 had never before been connected to presynaptic biology is not right. At the Drosophila NMJ, there are well-documented defects for kismet mutants in both development and presynaptic neurotransmitter release.

Kismet (Drosophila ortholog of CHD8) has been studied previously by gene knockdown, reducing expression levels below 50%. No phenotype has been reported in the heterozygous mutation to our knowledge. Regardless, we have removed the offending text as it was un-necessary.

The idea that CREG is the first-ever documented repressor of PHP in any system is not correct. Loss-of-function conditions at the NMJ have been documented that relieve blocks of PHP maintenance (two examples are loss of FasII and loss of 4E-BP). By the criteria set out by the current data set, these earlier examples would also constitute homeostatic repressors.

The reviewer is correct that over-expression of a specific isoform of Fas2 has been demonstrated to disrupt PHP. We now cite this paper.

Reviewer #2:

The manuscript by Genc and colleagues examines the loss of presynaptic homeostatic plasticity (PHP) in heterozygous combinations as a model for polygenic disorders. Using the loss of PHP in trans-heterozygous combination as readout, the authors uncovered genetic interactions among unrelated autism genes (RIMS1, CHD8, CHD2, WDFY3 and ASH1L) with fly orthologs. They expanded this strategy to screen for novel modifiers of PHP using the RIMS1/+ background and a series of overlapping deficiencies on 3rd chromosome. The screen identified modifiers in 20 lines (about 10% of the Df screened). Interestingly, some of these RIMS1- interacting Dfs also caused PHP defects when screened against the other autism genes (CHD8, CHD2, WDFY3 and ASH1L). The authors mapped two of the interacting loci to a kinase and phosphatase, PDPK1 and PPP2R5D, and confirmed that they interact genetically with most of the autism genes in this study. However, neither enzyme is required for PHP; also, the molecular basis for their observed interaction with RIMS1 and other loci remains unknown. Finally, the authors searched for differentially expressed transcripts (using RNAseq from adult heads) and reported that CREG was elevated in CHD8/+; PPPR25D/+ trans-heterozygotes. Importantly, lowering the dose of CREG rescued the PHP defects of CHD8/+; PPPR25D/+. How CREG restores the PHP in this trans-heterozygote and whether CREG is elevated in any of the other synthetic phenotypes remain open questions.

The study aims to model the genetic heterogenicity of autistic patients. The approach is sound, and the amount of data is substantial and of high quality, but the manuscript only tackles genetic interactions that the authors cannot explain nor connect to any previous PHP pathways. These numerous synthetic phenotypes seem merely "synthetic" without some anchoring into known phenomena; the same is true for the accompanying EM micrographs.

First, we thank the reviewer for commenting that our approach is sound. However, we find the subsequent criticism to be quite strange. We have pioneered something new in several respects. 1. We have determined precisely how combinations of heterozygous mutations can cause a synthetic disruption of PHP. This demonstration is proof of principle that double heterozygous gene-gene interactions could have implications for the etiology of ASD (or other genetic disorders of the nervous system). 2. We present evidence that homeostatic plasticity can fail at the intersection of two heterozygous genetic mutations without either gene being a core component of the homeostatic machinery. The data support the conclusion that the ASD orthologue mutation sensitizes the homeostatic signaling system to fail, and the modifier mutation exacerbates the effect, causing PHP to fail. A normally robust process, PHP, is rendered fragile and then fails. This is a fundamentally different concept compared to the standard search for core molecular mechanisms.

This is new territory and we suggest the following parallel. Anyone working in a rodent model system knows that a phenotype can be dramatically enhanced or suppressed by moving a mutation onto a new genetic background. It is very rare, but not unprecedented, to determine the mechanism by which a genetic background influences phenotypic penetrance. When successful, the discoveries are quite unexpected. This is effectively what we are doing by systematically testing double heterozygous mutant combinations. It stands to reason that the mechanisms we identify will be new. This possibility is further underscored by two studies in yeast, where systematic double heterozygous mutant combinations were tested (see our Introduction). Gene-gene interactions were identified that were entirely unpredicted. Given that ASD gene mutations are often heterozygous loss of function mutations, they will likely interact with standing (heterozygous) genetic variation in the human genome, unique to each individual. The reviewer is referred to the following papers:

Regarding the concept of resilience to disease causing mutations:

S. M. Neuner, S. E. Heuer, M. J. Huentelman, K. M. S. O’Connell, C. C. Kaczorowski, Harnessing Genetic Complexity to Enhance Translatability of Alzheimer’s Disease Mouse Models: A Path toward Precision Medicine. Neuron. 101, 399-411.e5 (2019).

R. Chen et al., Analysis of 589,306 genomes identifies individuals resilient to severe Mendelian childhood diseases. Nat. Biotechnol.34, 531–538 (2016).

S. H. Friend, E. E. Schadt, Clues from the resilient. Science344, 970–972 (2014).

Regarding double heterozygous gene interactions:

Ashworth A, Lord CJ, Reis-Filho JS. 2011. Genetic interactions in cancer progression and treatment. Cell145:30–8. doi:10.1016/j.cell.2011.03.020.

Bharucha, N. et al. A Large-Scale Complex Haploinsufficiency-Based Genetic Interaction Screen in Candida albicans: Analysis of the RAM Network during Morphogenesis. PLoS Genet.7, e1002058 (2011).

Glazier, V. E. et al. Genetic analysis of the Candida albicans biofilm transcription factor network using simple and complex haploinsufficiency. PLOS Genet.13, e1006948 (2017).

Chan DA, Giaccia AJ. 2011. Harnessing synthetic lethal interactions in anticancer drug discovery. Nat Rev Drug Discov10:351–64. doi:10.1038/nrd3374

Sardi M, Gasch AP. 2018. Genetic background effects in quantitative genetics: gene-by-system interactions. Curr Genet64:1173–1176. doi:10.1007/s00294-018-0835-7

Finally, regarding our electron microscopy. We agree that causality is never established. Nonetheless, the EM provides clear, visual demonstration of the synthetic nature of the gene-gene interaction that is undeniable, even for those readers who are not well versed in electrophysiology. The fact that this phenotype is rescued by the heterozygous mutation in CREG (the triple heterozygous mutant combination) is clear evidence of a link to CREG. We clearly state:

“Regardless of the underlying molecular mechanism leading to this EM phenotype and associated physiological deficits (a topic for future study), these data present a striking, visual confirmation of a strong synthetic genetic interaction between PPP2R5D/+ and the CHD8/+ heterozygous mutations. Furthermore, these data link the activity of a chromatin remodeling factor, present in the nucleus (CHD8), to a profound synaptic defect.”

1) The description of genetic interaction concepts is unnecessarily long and convoluted and must be consolidated.

We have shortened this section of the Introduction.

2) The screen itself is poorly described. Did the authors use the Bloomington Df kit for their screen? If not, specify. A table with the reagents used here should be provided.

A table is now provided. We respectfully disagree with the criticism that the screen is poorly described.

3) The authors chose to compare a small and a large Df in lieu of examining additive effects for gene heterozygosity (Figure 3). These arguments are weak. Instead, a more detailed description of the screen results would be more meaningful and should be included here.

For example, how many of the Dfs tested contain known PHP modulators? What was the outcome of the screen for those Dfs? How does that outcome compare with prior single gene analyses?

The reviewer is factually incorrect at several levels. This is perhaps understandable, since this type of analysis is rarely performed on a systematic basis.

a) First, we present every single deficiency in the graph (Figure 3). We provide representative examples of one large and one small deficiency. The reviewer seems to have misread the figure and our text in this regard.

b) Our arguments are not weak. This is the first time that a collection of small deficiencies has been analyzed electrophysiologically. We demonstrate a clear lack of correlation between the number of genes deleted by a deficiency versus effects on synaptic transmission. We state in our text,

“…we assessed whether there was any relationship between the number of genes that were deleted within a given deficiency and the robustness of PHP. One hypothesis is that the additive effects of multiple, heterozygous gene mutations would increase for larger deficiencies and PHP would be increasingly compromised. That was not the case (Figure 3). There was no correlation between the number of genes uncovered by a given deficiency and EPSP amplitude recorded in the presence of PhTx (R2 = 0.003; Figure 3A). Thus, impaired PHP cannot be accounted for by a simple additive accumulation of genetic mutations within a given deficiency.”

This is a simple analysis and a straightforward conclusion based on the lack of a statistical correlation.

c) In a formal genetic analysis, if two heterozygous genes do not interact, then nothing can be concluded. This is because it remains unknown how each heterozygous mutation affects the function of the protein within the cell. The fact that we did not identify homeostatic genes residing on the 3rd Chromosome in our screen is, therefore, an un-interpretable event and not reported. With this fact stated, we did report (in the original text) precisely which genes were identified by the screen that were previously demonstrated to interact with RIMS1/+. We stated in our original text:

“The screen was empirically validated by the identification, blind to genotype, of deficiencies that uncovered the RIMS1 locus, as well as the Pi3K68D locus (not included in hit list), previously shown to interact as a double heterozygous mutant with RIMS1/+ (44). Furthermore, the rim binding protein (RBP) locus was not identified as disrupting PHP, consistent with the previously published observation that a rbp/+ mutant does not interact with RIMS1/+ for PHP (45). However, rbp/+ did interact with RIMS1/+ for baseline neurotransmitter release as expected based on previously published data28 (not shown)”.

d) Our screen is quite unique so a direct comparison to single gene mutant screens or screens based on RNAi-mediated knockdown is not obvious.

e) A table of all deficiencies used in our screen is now provided.

4) When a Df covering a known PHP player did not show the expected interaction, what does this mean for the results of the screen?

In a formal genetic analysis, if two heterozygous genes do not interact, then nothing can be concluded. This is because it remains unknown how each heterozygous mutation affects the function of the protein within the cell. The fact that we did not identify occasional homeostatic genes residing on the 3rd Chromosome is, therefore, an un-interpretable event and not reported (see above).

5) It is somewhat anticlimactic that the PDPK1 and PPP2R5D are not required for acute PHP. However, the authors hint to a possible role in chronic homeostasis, more similar to the mTOR signaling components. This probably has already been examined and should be included here.

We respectfully disagree. This is a major take home message and it is important. We are NOT attempting to identify new core components of homeostatic plasticity. Our screen is designed to identify double heterozygous gene-gene interactions that weaken the robustness of homeostatic plasticity in previously unexplored ways. Why is this important? As highlighted above, ASD gene mutations are often heterozygous loss of function mutations. These heterozygous mutations will likely interact with standing (heterozygous) genetic variation in the human genome, variation that is unique to each individual. Thus, it is important to investigate and try to understand how heterozygous gene-gene interactions affect the phenotype of an organism. Drosophila (and C. elegans) are two of the model organisms where such an analysis is theoretically possible. This is what we have achieved.

Finally, it is clear from our new summary (Figure 9A), and the data presented in Figure 2 (with the addition of new data in Figure 1—figure supplement 2) that multiple ASD heterozygous mutations sensitize the homeostatic system to failure, but each mutation alone has no effect. The interacting genes need not be homeostatic genes. Rather, heterozygous mutations in newly identified modifiers potentiate the deleterious effects of the ASD gene mutations, causing homeostatic plasticity to fail. This is a fundamentally different way of thinking about how homeostatic plasticity could be affected by gene mutations. We argue that this may have relevance to the variable phenotypic penetrance of neurological and/or psychiatric diseases caused by heterozygous loss of function mutations.

The reviewer is referred to three excellent publications in this regard, concerning an emerging field of disease resilience (below). This interpretation is the subject of a section of our Discussion termed “Genetic context”.

S. M. Neuner, S. E. Heuer, M. J. Huentelman, K. M. S. O’Connell, C. C. Kaczorowski, Harnessing Genetic Complexity to Enhance Translatability of Alzheimer’s Disease Mouse Models: A Path toward Precision Medicine. Neuron. 101, 399-411.e5 (2019).

R. Chen et al., Analysis of 589,306 genomes identifies individuals resilient to severe Mendelian childhood diseases. Nat. Biotechnol.34, 531–538 (2016).

S. H. Friend, E. E. Schadt, Clues from the resilient. Science (80-. ).344, 970–972 (2014).

6) The RNAseq analysis was performed with adult heads due to understandable limitations. But the qPCR validation of CREG up-regulation must be done in third instar larvae, at the stage when PHP is recorded.

This is a perfectly reasonable concern and we not only performed the requested experiment, but added considerably more information to our study. Now, we document CREG expression in motoneurons at four developmental time points including the embryo, first instar, second instar, early third instar and late third instar. We then validate the four-fold increase in CREG expression in third instar neurons in the double heterozygous condition (as requested). These data substantially improve our analysis and we thank this reviewer for prompting us to include these data.

7) The authors propose that CREG is a homeostatic repressor. This predicts that overexpression of CREG will block PHP. The authors should directly test this prediction which will strengthen their conclusions.

We generated a new UAS-CREG transgene and drove expression either in motoneurons or ubiquitously. In both experiments, PHP is blocked. These data are added to Figure 8.

8) Reducing the dose of CREG would be expected to restore PHP in other trans-heterozygote combinations. This should be examined here since a positive result could expand the relevance of this finding beyond an isolated trans-heterozygous case presented here.

We respectfully disagree. This prediction assumes that the same mechanism is induced in all double heterozygous combinations. We now report that three additional RNAseq experiments fail to show up-regulation of CREG in other double heterozygous mutant combinations. When these data are considered alongside of the new EM data that we present (also showing that the EM phenotype is not conserved across all double heterozygous genotypes), it can be concluded that the CREG result is not universal, but it is none-the-less remarkable. Regardless of the generality, our data emphasize a new way to consider how homeostatic plasticity can fail in the context of human diseases that have a genetic origin.

9) Another elegant solution to link the disconnected parts of this study would be to overexpress CREG in some of the single heterozygotes (RIMS1, CHD8, CHD2, WDFY3, ASH1L – or PDPK1, PPP2R5D) and block PHP.

We demonstrate that CREG overexpression blocks PHP. This effect occurs irrespective of any genetic background. So, the suggested experiment does not make logical sense.

10) Using qPCR to search for increased levels of CREG in (third instar) mutants lacking PHP would be an easy, fast way to implicate CREG in other known PHP pathways and will tremendously benefit the author's arguments.

We disagree with this assertion. There is no reason to assume that CREG is involved in any general mechanism given that it is invoked only at the intersection of two heterozygous gene mutations, neither of which causes CREG over-expression alone. The reviewer is intent upon linking our data to the existing knowledge of homeostatic plasticity mechanism. One reason our work is novel is that we have may have uncovered a fundamentally different way to affect the robustness of homeostatic plasticity. This is emphasized by the fact that the modifiers are not, strictly, essential for PHP when tested as homozygous mutations.

Reviewer #3:

Genc et al. present in this study findings arguing that five seemingly unrelated autism associated genes are linked to presynaptic homeostatic plasticity (PHP) and synaptic transmission. They propose through genetic interaction studies in Drosophila that phenotypic enhancers could influence the robustness and phenotypic penetrance of these autism associated gene variants. Although several aspects of this study appear to support their conclusion, the reliability and reproducibility of their findings is hampered by conceptual and technical concerns. Addressing these concerns would help to strengthen the study.

1) The authors propose that five selected autism genes converge on PHP, but they only examine the genetic interaction between two pairs (RIMS1 and CHD8) and (RIMS1 and ASH1L). It is unclear why the authors do not complete the pairwise interaction studies for the remaining two genes. No rationale is provided to explain this incomplete analysis.

This is a fair point. We now analyze all potential interactions. Three of the four show a clear block (to our surprise) and one (WDFY3/+) does not. These data have been added to the manuscript. This greatly strengthens our paper and we thank the reviewer for prompting us to add these data to the study.

2) Due to the possibility of second site mutations contributing to the genetic interaction findings, it is important to assess other loss of function alleles of RIMS1, CHD8, and ASH1L in the genetic interaction studies. As well as for the remaining two selected autism genes. Furthermore, can these findings be ameliorated with genomic rescue of RIMS1, CHD8, ASH1L, etc in the pairwise genetic interaction studies?

The reviewer has seemingly requested many hundreds of experiments. In the context of our work, we argue that this is not justified. First, the RIMS1 and CHD8 and WDFY3 have been previously published and validated at the NMJ. The mutation in CHD2 was validated in our study by showing rescue experiments in which a CHD2 gene translocation was placed in the double heterozygous mutant background, demonstrating rescue. Second, our screen essentially is a systematic test for second site (loss of function) interactions on the third chromosome and an interaction is, indeed, rare. Finally, we validate the major finding of our paper by performing several experiments including A) UAS-CREG over-expression (new data) replicates the double heterozygous block of PHP and B) by analysis of a triple heterozygous animal (both electrophysiologically and at the EM level) demonstrating that all phenotypes are rescued by simply reducing the dosage of the CREG gene.

3) What is the rationale for choosing only five of the 20 deficiencies that showed interactions with RIMS1 to test their interactions with other gene? The authors need to provide a rationale for this selection. Otherwise it raises the question as to whether the remaining deficiencies were tested but did not show as strong of an interaction effect.

We can only test a finite number of interactions because they must be performed by hand. We chose five deficiencies – simple as that. A similar question could be asked about virtually every scientific study – why choose to study a given gene? Here is another example: The laboratory of Dr. Dan Feldman recently published a nice piece of work examining three ASD gene mutations in mice (published in Neuron). Why did he choose those three? The point is that one must make choices. Our choices were random.

4) Critical resource and technical information is lacking, which make it difficult to interpret or reproduce the study in the future. In particular, fly stocks (origin, type of mutation, specific breakpoints, stock center, etc) was not reported. What were the specific deficiency lines – including the "20 small deficiencies that cause PHP to fail"?

A new table is provided listing all deficiencies used. Each stock number can be input to the Bloomington website and all relevant information ascertained. We would also like to point out something that we consider obvious, but might be necessary given the antagonistic tone of this reviewer. When mapping individual gene contributions within a deficiency, numerous sub-deficiencies are used followed by testing of individual genes. This represents many, repetitive, examples of verification (positive and negative) before gene identification is achieved. This is standard in the field.

5) It is often unclear in the manuscript and figures if the sample size "n" is referring to animals, neurons, synapses, vesicles, etc. The sample size is also very low (n of only three) for some of the studies, raising the concern as to whether the studies were adequately powered to detect a statistical difference. For example, "we performed multiple (n=3-15) intracellular recordings".

a) In the majority of figures, every data point is plotted. For the EM, we have clarified sample sizes. In tables, as in figures, the sample size is the recording number, given that the averages are for recordings, it cannot be otherwise.

b) The comment regarding sample size is unfair. The reviewer quotes a statement that specifically refers to our screen – yet the reviewer implies that this reflects upon other data in our paper (which is factually incorrect). We stated in our original paper:

“The screen that we performed is diagrammed in Figure 2A. We took advantage of a collection of small chromosomal deficiencies (5-50 genes per deficiency, each with known chromosomal breakpoints; listed in Supplemental Table 1) that tile the 3rd chromosome, uncovering approximately 6000 genes in total. For every double heterozygous combination of RIMS1/+ with a heterozygous deficiency, we performed multiple (n=3-15) intracellular recordings, quantifying mEPSP amplitude, EPSP amplitude, quantal content (EPSP/mEPSP), resting membrane potential and input resistance.”

A genetic screen, such as ours, is designed to identify “hits” as efficiently as possible. In our genetic screen, as is common practice, we do not make any claims regarding the statistical validity of any individual observation. The only purpose of a screen is to identify “hits”. We establish well-justified parameters to select “hits”. This is done in a conservative fashion in order to protect the experimenter from performing un-necessary follow-up work. Importantly, every “hit” is re-validated. We are now more explicit, stating, “To achieve a final list of 20 hits, each potential hit was re-validated in a second set of experiments, increasing sample sizes (generally 7-16 NMJ). During the process of revalidation, we rule out approximately one third of the potential hits selected from the screen.”

Furthermore, the reviewer is referred to the fact that multiple “hits” from our screen were re-validated, multiple times throughout our manuscript. First, the hits were identified. Second, they were revalidated as a hit. Third, five of the deficiencies were incorporated into a matrix of multiple gene-gene interactions, a further re-validation. Finally, two deficiencies were resolved to individual genes, requiring multiple rounds of re-validation. And, the single genes precisely recapitulate the results of the deficiencies in which they originally resided. Surely, at this point, the extensive nature of revalidation, with statistically relevant sample sizes at every stage, should inspire statistical confidence.

How did we determine adequate sample sizes? We performed a power analysis to estimate necessary sample size. We have an effect size of 150-200% with a low standard deviation. Estimation of sample size suggests a minimal sample size of three to four, which is obviously very small. In our primary screen, when efficiency is emphasized, we use a sample size of 3-15, as reported. None of the data from the screen are compared statistically – the use of sample sizes here is simply for transparent reporting. In our study, when statistical comparisons are made, sample sizes were never less than n=7. In instances where effect size was on the smaller end of the range, sample sizes are generally larger to ensure confidence.

In our study, we analyze more than 40 genotypes electrophysiologically, under two conditions (with and without PhTX), representing more than 80 experimental conditions – not including our screen. In only two experiments out of more than 80 experimental conditions are sample sizes equal to 4 (in the original submission, there were four instances, and new data have been incorporated). These instances were controls that adhered to high effect size and low variance. In every other experiment (>80 conditions) sample size was never less than 7, generally in excess of 10 and often greatly exceeding these sizes. All the data are transparently presented in figures with individual data points and in tables with sample sizes. In every instance, sample sizes were sufficiently powered and all statistical tests and values reported.

6) The authors argue that CREG is a homeostatic repressor that blocks PHP and regulates synaptic ultrastructure (Figure 8). It would be important to show the phenotype of the CREG single mutant. What is the ultrastructural phenotype of the CREG single mutant?

We have performed the requested experiments. The CREG mutant that abolishes gene expression was analyzed. We state in the text:

“Finally, we assessed the consequences of a the heterozygous and homozygous loss of function mutations on baseline neurotransmission and PHP. The CREGM1 allele abolished expression (see above). Neither CREGM1/+ heterozygous animals (n=8) nor the CREGM1 homozygous animals (n=8) and any effect on baseline transmission (quantal content; p>0.5) or the expression of PHP (CREGM1/+ = 143% PHP, p<0.05 compared to wild type; CREGM1 = 155% PHP, p<0.01 compared to wild type). Thus, loss of CREG is without effect on neuromuscular transmission or PHP. Taken together, our data are consistent with the conclusion that CREG is a homeostatic repressor, one of very few identified to date (Spring et al., 2016).”

We have not performed the ultrastructure of the CREG mutant since it has no effect on synaptic transmission, nor does it effect homeostatic plasticity.

7) EM studies are performed in tissue fixed 1-5 seconds after high frequency stimulation with a sample size as low as only 2 animals. How does the difference between 1 to 5 seconds post-high frequency stimulation affect the findings?

The only way to answer this question is to invoke the collaboration of either Eric Jorgensen or Shigeki Watanabe, who perform modern freeze-slam biology with highly specialized equipment from Leica. Indeed, we have attempted this experiment in collaboration with the Watanabe laboratory but, to date, the technology cannot be applied to third instar larvae due to tissue thickness, which prevents uniform freezing and tissue preservation. We have even attempted this in first instars, but tissue freezing remains a confound. We would like to point out that our manuscript represents an unusually thorough EM analysis by the standards of the field (noting, however, the exceptional and unique work of our C. elegans colleagues).

8) The rationale for selecting RIMS1, CHD8, CHD2, WDFY3, and ASH1L is still unclear. The authors state they were selected as they were "category 1 or 2 ASD-associated genes based on SFARI Gene". However these two categories encompass many more than these five genes. Were these the only genes with fly orthologs and available resources? Were there other reasons for choosing these genes?

We thought that we were clear, in this regard, in the text.

Here is the full history. This project began more than five years ago when Prof. Matt State moved to UCSF from Yale University. He supplied us with the identities of many of the most agreed upon de novo mutations that confer risk for ASD at that time. At that time, the list was much shorter than it is now. Nonetheless, the five genes that we decided to study remain category 1 genes, and remain well-acknowledged, high confidence ASD risk factors. It appears that we received good advice and made good choices in retrospect. It would be impossible for us to analyze more than five. No other study has engaged in such a phenotypic analysis. The only study to analyze multiple mutations in ASD, to our knowledge, is the recent work of Dr. Dan Feldman, which was an extraordinary assessment of three mutations in mice (published recently in Neuron). We do not known why he chose those three mutations to work on.

We stated in the original text a very clear set of rational for choosing these genes, to which we have added expression analysis. We stated,

“All five of these genes are considered high confidence “category 1” ASD-associated genes based on SFARI Gene (Simons Simplex Collection, 2018)34. All five of these genes have clear Drosophila orthologs. Further, we demonstrate that all five genes are expressed in Drosophila third instar motoneurons based on a Patch-Seq analysis of gene expression (Figure 1—figure supplement 1). The five ASD gene orthologs were also chosen to reflect a broad range of biological activities that are associated with the numerous ASD-associated genes identified to date. The RIMS1 gene is a synaptic scaffolding protein that localizes to and organizes sites of neurotransmitter release, termed active zones. The CHD8 and CHD2 genes encode chromatin remodeling factors that localize to the cell nucleus. WDFY3 encodes a phosphatidylinositol 3-phosphate-binding protein and regulator of autophagy and intracellular signaling. ASH1L encodes a member of the trithorax group of transcriptional activators and is found in the cell nucleus. A survey of biochemical and genetic interaction networks in Drosophila demonstrates no known interactions among these five genes (Flybase). In humans, there appear to be no known direct biochemical interactions among these genes. Yet, heterozygous LOF mutations in each of these genes are associated with risk for ASD in humans.”

9) The authors need to be careful not to overstate or misstate their findings. Particularly as the premise of the paper hinges upon 5 selected genes associated with autism, and far more than 5 genes are implicated in autism. Providing a better rationale for selecting these 5 genes would help frame their findings better.

Thank you for the criticism. However, our study is one of very few to begin to assess multiple ASD genes in parallel and to explore genetic interactions. We have revisited the text in response to this reviewer, but we argue that highlighting the novelty of our approach is both reasonable and appropriate at this time.

[Editors’ note: what follows is the authors’ response to the second round of review.]

Revisions required for this paper:

1) Clarify or modify broad statements about lack of information or claims of "novelty", that falters in the context of existing information in the literature. The study will be strengthened by building connections with existing knowledge, which will minimize the impression that these findings are "synthetic" without anchoring in known phenomena.

Please see individual responses to each specific criticism. In all cases, we quote the changes we make to the text directly in in this document in response to each reviewer query.

a) In discussing the random selection of RIMS1, CHD8, CHD2, WDFY3, and ASH1L the authors state that there are no known biochemical and genetic interactions among these five genes in Drosophila, and no known direct biochemical interactions in humans. However, it is curious that the authors fail to also state the potential for interactions between these genes by incorporating many findings in mouse models indicating a potential convergence between these genes on neurodevelopment and function. Presenting this at the outset will further strengthen the study by utilizing prior literature to support the probability of identifying disease relevant genetic interactions.

In many respects, this comment reflects the question that drove us to initiate our study back in 2013. At that time, as now, a major question in the field of ASD research is whether there are genetic or molecular commonalities that give rise to common phenotypes in human. The reviewer would like us to highlight the “potential for interactions between these genes” (the ASD genes). In other words, the reviewer would like us to continue the type of speculation that has been ongoing in the literature for a number of years. We entered this line of research to avoid such speculation and perform forward genetic gene discovery. So, we are hesitant to initiate our study with such speculation. There are many theories, some more robustly defended than others. For example, there is a theory that causal gene mutations in ASD may commonly disrupt the balance of excitation and inhibition, an idea first proposed by Michael Merznich and John Rubenstein (Rubenstein and Merznich, 2003) with recent experimental support (Antoine et al., 2019). There has been speculation that chromatin remodeling factors control the expression of “synaptic” genes (Sullivan et al., 2019). Homeostatic hypotheses have been put forward, including a nice review by T. Bourgeron (Bourgeron, 2015).

Rather than reviewing these many possible theories, or emphasizing phenotypic similarities among genes that may or may not have and biochemical or genetic interaction, we frame our paper from a well-established perspective in the field of genetics and evolutionary biology. At the start of the Introduction, we frame our paper by stating, “Advances in whole genome sequencing and genome-wide association studies have dramatically expanded our understanding of the genetic architecture of ASD. In particular, the identification of rare de novo mutations that confer high risk for ASD has generated new molecular insight (De Rubeis et al., 2014; Iossifov et al., 2014; Sanders et al., 2015). Yet, even in cases where a rare de novo mutation confers risk for ASD, additional processes are likely to contribute to the ASD phenotype including the engagement of adaptive physiological mechanisms that respond to the presence of an ASD risk associated gene mutation (Gaugler et al., 2014; Gibson, 2009; Hartman et al., 2001; Hou et al., 2019; Kitano, 2007; Plomp et al., 1992; Sackton and Hartl, 2016; Sardi and Gasch, 2018).” This is an idea that is also a major thesis of an excellent book on the evolution of complex systems titled, “The Plausibility of Life: Resolving Darwin’s Dilemma” by M.W. Kirschner and J.C. Gerhart (2005).

Out of respect for this reviewer’s comment and the reviewer’s apparent wish to acknowledge that there has been considerable phenotypic work on these genes, we now add a sentence to the first paragraph of our Results that refers the reviewer to a newly added table. We state, “A supplemental table (Supplementary file 1 table S1) includes known disease associations for each of these five human genes, and links to web-based genetic and genomic resources.” This table lists each of the five ASD genes and includes a web-link to the SFARI GENE website for each gene. This site curates the known human mutations in each gene, disease associations for each gene, and evaluates the robustness of the existing literature. It is an excellent resource. Further, for each of the five ASD-associated genes, we list the known disease associations. We hope that the inclusion of this table will satisfy the desire for additional information.

b) By extension, the known disease associations for these genes should be presented and referenced, as well as if they are all impacting similar neurologic functions in humans. All five genes have known disease associations: OMIM: RIMS1 (#603649, Cone-rod dystrophy type 7), CHD8 (#615032, Susceptibility to autism type 18), CHD2 (#615369, Epileptic encephalopathy childhood onset), WDFY3 (#617520, Primary microcephaly type 18), and ASH1L (#617796, autosomal dominant mental retardation type 52).

Please see the addition of a new supplemental table (Supplementary file 1 table S1), as described in the answer to point (a) above. Within the SFARI GENE listing are all of the OMIM information requested by this reviewer.

c) Heterozygous loss of function mutations in CHD8, CHD2, WDFY3, and ASH1L all cause neurodevelopmental disorders, including autism, epilepsy, and microcephaly. In the context of the human data, the authors' finding that PHP in the single heterozygous flies is normal suggests that PHP may not be the most important mechanism driving autism phenotypes in neurodevelopmental disorders. In contrast, although pathogenic variants in RIMS1 have only been reported with a photoreceptor phenotype in humans, there is a recent publication by Peter et al., 2019 (PMID 30949922) presenting evidence in humans for pleiotropic and complex effects involving RIMS1 in sporadic autism spectrum disorder. Therefore, the authors can consider strengthening the conceptual impact of their study from the perspective that variants in RIMS1 may increase the susceptibility to autism spectrum disorder but require other polygenic factors to manifest the disorder.

We thank the reviewer for pointing us to this newer paper on RIMS1, one that we had not seen. We now reference this paper in the first paragraph of the Introduction. This is exciting work.

We would also like to make a point of clarification. The reviewer states, “Heterozygous loss of function mutations in CHD8, CHD2, WDFY3, and ASH1L all cause neurodevelopmental disorders, including autism, epilepsy, and microcephaly. In the context of the human data, the authors' finding that PHP in the single heterozygous flies is normal suggests that PHP may not be the most important mechanism driving autism phenotypes in neurodevelopmental disorders.”

Although the individual heterozygous mutations do not block PHP, we demonstrate that many of these gene mutations sensitize PHP to fail. Based on this finding, and based on the rest of our analyses, we argue that loss of PHP may contribute to the phenotypic severity of ASD, not the cause of ASD. We consider this distinction between causality and phenotypic severity to be a very important concept, and it is the framework for the entire study (see above). We also state this in our Discussion:

”It is well established that genetic context can profoundly influence the phenotypic severity of disease-causing gene mutations. For example, in mice, it has been shown that genetic context (strain background) influences phenotypic penetrance in an Alzheimer’s disease model (Neuner et al., 2019). In humans, systematic screening of the phenotypically normal population has identified individuals that are resistant to the effects of well-established, debilitating disease causing mutations, an effect termed “resilience” that is attributed to the effects of genetic context (Chen et al., 2016; Friend and Schadt, 2014). It seems plausible that the common phenotypic enhancers, identified in our genetic screen, could represent a mechanism by which genetic context influences the phenotypic penetrance of ASD-associated gene mutations.”

This is a framework that is not dissimilar to the argument of genetic complexity made in a number of recent studies, including that of Peter et al., 2019. But, we take the idea of genetic complexity, which is quite general, and provide evidence for how genetic complexity could manifest as neurophysiological defects through the failure of homeostatic mechanisms.

d) The authors state that the "five ASD gene orthologs do not share any known common biological activity. Therefore, the identification of common genetic modifiers is completely unexpected." This is a confusing statement to make given that CHD8 and CHD2 are in the same subfamily of chromodomain helicase DNA-binding proteins, indicating a common biological activity, and ASH1L localizes to the chromatin. Based on these common functions and subcellular localization it is not entirely unexpected to find common genetic modifiers. These statements should be revised. Noting the potential for common biological activity will not detract from the potential impact of the authors' findings of common genetic modifiers through their screening approach.

We would like to point out that the function of each gene was clearly stated in our original text. We state, “The RIMS1 gene is a synaptic scaffolding protein that localizes to and organizes sites of neurotransmitter release, termed active zones. The CHD8 and CHD2 genes encode chromatin remodeling factors that localize to the cell nucleus. WDFY3 encodes a phosphatidylinositol 3-phosphate-binding protein and regulator of autophagy and intracellular signaling. ASH1L encodes a member of the trithorax group of transcriptional activators and is found in the cell nucleus.

We also are quite specific about this statement and topic in our Introduction. We state, “A survey of biochemical and genetic interaction networks in Drosophila demonstrates no known interactions among these five genes (Flybase). In humans, there appear to be no known direct biochemical interactions among these genes. Yet, heterozygous LOF mutations in each of these genes are associated with risk for ASD in humans.”

Out of respect for the reviewer’s concern, we have removed that sentence. We agree that the sentence is somewhat redundant and it is less precise that the text that preceded it in our Introduction.

e) The authors state that "there is no clear connection, biochemically or genetically, to the role of PDPK1 or PPP2R5D in the nervous system." The authors need to clarify this statement as there are rodent models demonstrating a genetic role for PDPK1 and PPP2RD5 in the nervous system. There is altered brain development in PDPK1 KO mice (Lawlor et al., EMBO J, PMID 12110585) and abnormal tau pathology in the brains of PPP2RD5 KO mice (Louis et al., PNAS, PMID 21482799). Based on the literature there is clear connection for a role of PDPK1 and PPP2R5D in the nervous system.

With all due respect, the reviewer is confused on this point. This comment is taken out of context and the reviewer infers something that we do not state. We do not make any claim regarding a connection between PDPK1 and PPP2R5D. The sentence refers back to the two other genes that were identified in our RNAseq analyses. To prove this point, we simply copy more of the surrounding paragraph: We state, “… there are only two genes that are commonly down-regulated in all four ASD mutants (FBgn0027578 [Nepl21] and FBgn0037166 [CG11426]) (Figure 7C). FBgn0027578 encodes a metalloprotease of the Neprilysin family, with homology to endothelin converting enzyme 1 in human, of unknown function in the nervous system. FBgn0037166 encodes phosphatidic acid phosphatase type 2, which is expressed in the Drosophila nervous system, but of unknown function. There is no obvious means to connect the down regulation of these two genes to impaired homeostatic signaling, although future experiments will explore these genes in greater depth. Furthermore, there is no clear connection, biochemically or genetically, to the role of PDPK1 or PPP2R5D in the nervous system.”

To address any possible confusion that some other reviewer might have, we have revised this sentence. It now reads, “Furthermore, there is no clear connection between FBgn0027578 or FBgn0037166 and the roles of either PDK1 or PPP2R5D in the nervous system.”

f) The authors propose that upregulation of CREG in PPP2R5D/+;CHD8/+ underlies the failure of PHP, which is further supported by the complete block of PHP when CREG is overexpressed ubiquitously or in motor neurons. They identified two CREG mutant alleles with differential effects on normalizing CREG levels in Figure 8, panel D (m2 restored the double mutant to wildtype, but not m1) suggesting that m2 is possibly a stronger loss of function allele. But surprisingly, m2 seems to have a reduced rescue effect compared to m1 (Figure 8, panel F). Is there a difference in CREG expression level between these two alleles? The authors only report the m1 transposon completely abolishes CREG expression and do not show the data. They do not report the effect of the m2 transposon. The effect on CREG expression from both alleles should be examined and shown as this may indicate a dosage sensitivity for CREG.

We acknowledge that additional information would be important. In our prior submission, we stated that RNA expression was abolished in the M1 allele. We have now expanded upon this statement to include precise values, variance and the number of biological and technical replicates. We now state, “The CREGM1 transposon completely abolishes CREG expression and a heterozygous CREGM1/+ mutant reduces CREG expression (CREGM1 = zero expression compared to wild type, 3 biological and 3 technical replicates; CREGM1/+ = 51.5 ± 3.05% wild type expression, 3 biological and 3 technical replicates). Next, we generated a triple heterozygous mutant combination (CHD8+; PPP2R5D/+, CREGM1/+) and find that the CREGM1/+ allele attenuates the up-regulation of CREG gene transcript in the triple heterozygous mutant background, a suppression effect of approximately 50%, as predicted (Figure 8D). Then, we repeated this analysis with the CREGM2 allele. This allele has a minor effect on baseline CREG expression (73.2±2.7% wild type expression, 3 biological and 3 technical replicates). However, we discovered that this transposon insertion caused a complete block of CREG up-regulation in the triple heterozygous mutant combination, suggesting that this transposon insertion, residing in 3’ UTR, may disrupt a transcription regulatory motif (Figure 8D).”

g) A rationale for selecting only 5 of the 20 deficiencies showing interactions with RIMS1 is still lacking. Were they selected based on the type of genes in the deficiencies (brain specific in mammals), known or unknown function of the genes, prior implications in homeostatic plasticity, availability of reagents, strength of the interaction effect, etc?

This question was asked in the first round of review. In response, we stated: “We are human. We can only test a finite number of interactions because they must be performed by hand. We chose five deficiencies – simple as that.” Perhaps we can add more context with proof regarding the randomness of our selection. We did not select the 5 deficiencies because they were the “best” hits. Note that Df(3)24410 is among the weaker hits identified in our screen (see new Supplementary file 1 table S3-S5). We did not select these five deficiencies because they contain interesting genes. Given the number of genes within each deficiency, it is impossible to guess. Finally, there is some bias against the very largest Df’s, which seems appropriate for this experiment. But, at the same time, we did not actively select for the smallest of the Df’s. So, once again, we randomly selected 5 deficiencies and performed the analysis. As further context, when we began this experiment, we did not expect to observe a high degree of common enhancement. This finding was serendipitous and became a focus of the work from that point forward.

2) Technical concerns

a) There appears to be incomplete presentation of some datasets. The authors report that "altered NMJ growth was not commonly observed in the majority of genetic interactions tested (Figure 5—figure supplement 1)", implying that multiple pairwise interactions were examined. In fact, this is not the case. Figure 5—figure supplement 1 shows NMJ morphology data for wildtype, single heterozygotes for PDPK, PPP2R5D, CHD8, CHD2, ASH1, WDFY3. The only genetic interactions shown were for CHD8 with PDPK1 and CHD8 with PPP2R5D. The data actual shows an increase in bouton number for the CHD8/+;PPP2R5D/+ genotype compared to wildtype and PPP2R5D/+ alone, it did not reach statistical significance compared to CHD8/+ alone but perhaps this is due to the sample size? The data for the other pairwise genetic interactions should also be included in order for the authors to make the statement that altered NMJ growth was not commonly observed.

We thank the reviewer for highlighting this point. We agree that the statement highlighted by the reviewer is not a correct representation of the information in the supplemental figure. This information has not changed since the original submission to a different journal more than 24 months ago. We are pleased to change the text here, soften the conclusion and be more specific about the exact data presented. We now state, “We asked whether neuronal morphology was substantially altered in the heterozygous ASD-associated gene mutations and in select double heterozygous genetic interactions (Figure 5—figure supplement 1). We do find evidence that the heterozygous CHD8/+ mutation predisposes the NMJ to modest overgrowth, consistent with CHD8 influencing brain development in other systems (Gompers et al., 2017). But, but this effect does not become more severe when combined with either PDPK1 or PPP2R5D mutation as double heterozygotes. Thus, we conclude that altered synaptic growth is not highly correlated with the block of PHP in these double heterozygous combinations.”

b) Sample size still remains unclear for Figure 5—figure supplement 2, Figure 5—figure supplement 3, Figure 8F, Figure 8K. Please clarify the sample size so that the interpretation and reproducibility of the data is improved.

We thank the reviewer for pointing out this omission. The sample sizes have been listed in the figure legend for all genotypes in Figure 5—figure supplement 2, Figure 5—figure supplement 3, 8F and 8K.

c) In Figure 8, what does the sample size "n" in panels H and I represent? The authors note in the figure legend that "individual data points shown for indicated genotypes". Clarify if the individual data points represent individual vesicles, average vesicle size per synapse, or average vesicle size per animal. How many synapses were quantified per animal? How many animals per genotype?

We thank the reviewer for pointing out this omission. The sample sizes have been clarified the figure legend of Figure 8.

d) Similarly, in Figure 6, what is the sample size representing? Individual vesicles, average size of vesicles per boutons, per animal? How many animals and synapses per animal were assessed?

We thank the reviewer for pointing out this omission. The sample sizes have been clarified in the figure legend.

e) It is important to provide all the necessary information about reagents and resources used for rigor and reproducibility. The authors now include a table S1 showing all of the deficiency stocks that were used in the study. But it remains unclear which of the listed 168 deficiency stocks were the "20 small deficiencies that cause PHP to fail when combined with RIMS1/+". The authors provide a schematic showing the distribution of these deficiencies in Figure 2C, but it is important to also denote in Table S1 the stock ID number corresponding to these 20 small deficiencies. This should be very straightforward to indicate in table S1.

We are pleased to provide a set of new tables at this stage of the review process. We now present a total of seven supplemental tables. Included are the follow new data: Supplementary file 1 tables S3-S5 list all of the deficiencies that were tested, inclusive of the electrophysiological data acquired during the screen and sample sizes. We also list each of the deficiencies that were selected as “hits” in our screen. We list the stock number and full genotype. Please note: The values presented in graphical format in Figure 2 from our screen are the values that were initially recorded during our high-throughput screening mode and these values are listed in Supplementary file 1 table S2. In Supplementary file 1 table S3, we report values only for the “hits” and these values are different because they include larger sample sizes that were achieved when we verified each hit. Thus, Supplementary file 1 table S3 includes additional information, inclusive of our verification efforts. Finally, (Supplementary file 1 table S5), we list all of the genes that are uncovered by each of the deficiencies selected in our screen, according to information provided in Flybase. This should facilitate reader use of the information. These tables are cited in the Results section. We state, “(see Supplementary file 1 tables S3-S5 for further detailed information on the screen results).” We believe that we have exceeded the request for new information in this regard. We thank the reviewers for prompting us to provide these additional data.

f) The authors assessed the consequences of CREG heterozygous and homozygous loss of function mutations on baseline neurotransmission and PHP and describe the findings in the text. This is an important control data. For comparison to the other genotypes it would be helpful to also include the data graphically in Figure 8, panel F.

We have added data for homeostatic plasticity in the CregM1/+ and CregM1 to Figure 8F. We think that this now represents a full data set, as is commonly presented in homeostatic plasticity papers in the field. We also report the precise values for baseline neurotransmitter release in the text. We now state, “Finally, we assessed the consequences of the heterozygous and homozygous loss of function mutations on baseline neurotransmission and PHP. The CREGM1 allele abolished expression (see above) and is the focus of these analyses. Neither the heterozygous nor homozygous animals affected expression of PHP (Figure 8F). The CREGM1/+ heterozygous animals had no effect on baseline transmission compared to wild type (wild-type QC = 40.8±2.2 n=10; CREGM1/+ QC = 38.6 (±3.0) n=8; Student’s t-test; p>0.5). The CREGM1 homozygous allele decreased baseline transmission by ~18% (CREGM1 QC = 33.5±3.0 n=8; p=0.02). Clearly, neither baseline release nor PHP are potentiated, demonstrating that the rescue of PHP in the triple heterozygous mutant condition cannot be considered an additive effect of the heterozygous CREGM1 mutation. Taken together, our data are consistent with the conclusion that CREG is a homeostatic repressor, one of very few identified to date (Spring et al., 2016). “

g) You stated: "First, we present every single deficiency in the graph (Figure 3). We provide representative examples of one large and one small deficiency. The reviewer seems to have misread the figure and our text in this regard." The reviewers did not find the data, legend or text presenting the interacting deficiencies. Figure 3 has no such graph. They are wondering whether you refer to the drawing in Figure 2C. Do the authors imply that this is the summary result that presents every single deficiency? Are we missing something? Moreover, the stock numbers for the Bloomington deficiencies provided in the table do not match the ones described in the text.

The original reviewer comments stated, “The authors chose to compare a small and a large Df in lieu of examining additive effects for gene heterozygosity (Figure 3).” We interpreted this statement to mean that the reviewers had missed the fact that each data point in Figure 3A represents averaged data for every individual deficiency tested in our screen. In Figure 3A, we plot quantal content versus the number of genes deleted within each deficiency and note that there is no correlation. The legend of Figure 3A stated, “Scatter plot showing the number of genes deleted (y axis) versus quantal content (x axis) in the presence of PhTx for all deficiencies tested.”

In our response to reviewer criticism in the first round of review, we stated, “First, we present every single deficiency in the graph (Figure 3A – the entire figure is copied here for clarity). We provide representative examples of one large and one small deficiency. The reviewer seems to have misread the figure and our text in this regard.” We are sorry that there seems to remain some confusion. Currently, the reviewers state, “Figure 3 has no such graph”. Perhaps the graph was somehow deleted in the figure the reviewer received, or there remains some confusion about the precise graph that we are each referring to?

Finally, we re-checked the deficiency call-outs. The representative deficiencies (Figure 3B) are not “hits” from our screen (we never suggested that they were) and, therefore, were not represented in the last submission. The representative examples were chosen, simply, as representative deficiencies that had either many or few genes deleted, with remarkably similar quantal contents. We now supply, upon reviewer request, information for all of the deficiencies in our screen.

Information is needed for why data sets are not shown (ie CREG expression level in m1 allele), clarity on what the sample size represents in multiple datasets, details as to how many synapses, boutons, animals are included in the NMJ and EM datasets.

Please see above.

https://doi.org/10.7554/eLife.55775.sa2

Article and author information

Author details

  1. Özgür Genç

    Department of Biochemistry and Biophysics Kavli Institute for Fundamental Neuroscience University of California, San Francisco, San Francisco, United States
    Contribution
    Data curation, Formal analysis, Validation, Investigation, Visualization, Writing - review and editing
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0003-0635-3192
  2. Joon-Yong An

    1. Department of Psychiatry UCSF Weill Institute for Neurosciences University of California, San Francisco, San Francisco, United States
    2. School of Biosystem and Biomedical Science, College of Health Science, Korea University, Seoul, Republic of Korea
    Contribution
    Data curation, Formal analysis, Writing - review and editing
    Competing interests
    No competing interests declared
  3. Richard D Fetter

    Department of Biochemistry and Biophysics Kavli Institute for Fundamental Neuroscience University of California, San Francisco, San Francisco, United States
    Contribution
    Data curation, Formal analysis, Methodology, Writing - review and editing
    Competing interests
    No competing interests declared
  4. Yelena Kulik

    Department of Biochemistry and Biophysics Kavli Institute for Fundamental Neuroscience University of California, San Francisco, San Francisco, United States
    Contribution
    Data curation, Formal analysis
    Competing interests
    No competing interests declared
  5. Giulia Zunino

    Department of Biochemistry and Biophysics Kavli Institute for Fundamental Neuroscience University of California, San Francisco, San Francisco, United States
    Contribution
    Data curation, Formal analysis
    Competing interests
    No competing interests declared
  6. Stephan J Sanders

    Department of Psychiatry UCSF Weill Institute for Neurosciences University of California, San Francisco, San Francisco, United States
    Contribution
    Methodology, Writing - review and editing
    Competing interests
    No competing interests declared
  7. Graeme W Davis

    Department of Biochemistry and Biophysics Kavli Institute for Fundamental Neuroscience University of California, San Francisco, San Francisco, United States
    Contribution
    Conceptualization, Resources, Supervision, Funding acquisition, Visualization, Methodology, Writing - original draft, Project administration, Writing - review and editing
    For correspondence
    graeme.davis@ucsf.edu
    Competing interests
    Reviewing editor, eLife
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0003-1355-8401

Funding

National Institute of Neurological Disorders and Stroke (R35-NS097212)

  • Graeme W Davis

Simons Foundation (SFARI #401636)

  • Graeme W Davis

Simons Foundation (SFARI #402281)

  • Stephan J Sanders

National Institute of Mental Health (R01 MH110928)

  • Stephan J Sanders

Neurosciences Research Foundation (2017M3C7A1026959)

  • Joon-Yong An

The funders had no role in study design, data collection and interpretation, or the decision to submit the work for publication.

Acknowledgements

Supported by NINDS Grant (R35-NS097212) and Simons Foundation (SFARI #401636) to GWD, Simons Foundation (SFARI #402281) and NIMH (R01 MH110928) to SJS, and NRF-2017M3C7A1026959 to J-YA. We thank Matt State for comments and support and members of the Davis, State and Sanders labs for critical evaluation of the manuscript.

Senior Editor

  1. Ronald L Calabrese, Emory University, United States

Reviewing Editor

  1. Hugo J Bellen, Baylor College of Medicine, United States

Reviewer

  1. C Andrew Frank, University of Iowa, United States

Publication history

  1. Received: February 5, 2020
  2. Accepted: June 7, 2020
  3. Accepted Manuscript published: July 1, 2020 (version 1)
  4. Version of Record published: July 31, 2020 (version 2)

Copyright

© 2020, Genç et al.

This article is distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use and redistribution provided that the original author and source are credited.

Metrics

  • 1,221
    Page views
  • 224
    Downloads
  • 0
    Citations

Article citation count generated by polling the highest count across the following sources: Crossref, PubMed Central, Scopus.

Download links

A two-part list of links to download the article, or parts of the article, in various formats.

Downloads (link to download the article as PDF)

Download citations (links to download the citations from this article in formats compatible with various reference manager tools)

Open citations (links to open the citations from this article in various online reference manager services)

Further reading

    1. Neuroscience
    David Q Stoye et al.
    Research Article

    The mechanisms linking maternal stress in pregnancy with infant neurodevelopment in a sexually dimorphic manner are poorly understood. We tested the hypothesis that maternal hypothalamic-pituitary-adrenal axis activity, measured by hair cortisol concentration (HCC), is associated with microstructure, structural connectivity, and volume of the infant amygdala. In 78 mother-infant dyads, maternal hair was sampled postnatally, and infants underwent magnetic resonance imaging at term-equivalent age. We found a relationship between maternal HCC and amygdala development that differed according to infant sex. Higher HCC was associated with higher left amygdala fractional anisotropy (β = 0.677, p=0.010), lower left amygdala orientation dispersion index (β = −0.597, p=0.034), and higher fractional anisotropy in connections between the right amygdala and putamen (β = 0.475, p=0.007) in girls compared to boys. Furthermore, altered amygdala microstructure was only observed in boys, with connectivity changes restricted to girls. Maternal cortisol during pregnancy is related to newborn amygdala architecture and connectivity in a sexually dimorphic manner. Given the fundamental role of the amygdala in the emergence of emotion regulation, these findings offer new insights into mechanisms linking maternal health with neuropsychiatric outcomes of children.

    1. Neuroscience
    Tobias Lieblein et al.
    Research Article Updated

    The formation of oligomers of the amyloid-β peptide plays a key role in the onset of Alzheimer's disease. We describe herein the investigation of disease-relevant small amyloid-β oligomers by mass spectrometry and ion mobility spectrometry, revealing functionally relevant structural attributes. In particular, we can show that amyloid-β oligomers develop in two distinct arrangements leading to either neurotoxic oligomers and fibrils or non-toxic amorphous aggregates. Comprehending the key-attributes responsible for those pathways on a molecular level is a pre-requisite to specifically target the peptide's tertiary structure with the aim to promote the emergence of non-toxic aggregates. Here, we show for two fibril inhibiting ligands, an ionic molecular tweezer and a hydrophobic peptide that despite their different interaction mechanisms, the suppression of the fibril pathway can be deduced from the disappearance of the corresponding structure of the first amyloid-β oligomers.