Rapid assessment of the temporal function and phenotypic reversibility of neurodevelopmental disorder risk genes in Caenorhabditis elegans

Neurodevelopmental disorders such as autism spectrum disorder (ASD) and intellectual disability (ID) are highly genetically heterogeneous and are accompanied by a range of cognitive and behavioral phenotypes including sensory processing and learning impairments (American Psychiatric Association, 2013; Boyle et al., 2011; de la Torre-Ubieta et al., 2016; Iakoucheva et al., 2019; Sanders et al., 2019). More severe cases of these disorders can cause significant challenges for affected individuals and their families (Boyle et al., 2011; de la Torre-Ubieta et al., 2016; Iakoucheva et al., 2019; Sanders et al., 2019). Recently, there has been remarkable progress in identifying genetic risk factors that contribute to diverse neurodevelopmental disorders, with hundreds of genes now implicated in ASD and ID alone (de la Torre-Ubieta et al., 2016; De Rubeis et al., 2014; Deciphering Developmental Disorders Study, 2015; Iakoucheva et al., 2019; Iossifov et al., 2014; Sanders et al., 2019; 2015; Satterstrom et al., 2020; Vissers et al., 2016). Variants in the majority of these genes (e.g. 89/102 ASD-associated genes; 87%) are thought to confer risk through haploinsufficiency as the individual carries one loss-of-function allele with insufficient residual function from the remaining copy (Satterstrom et al., 2020). The identification of how these variants contribute to disorder pathology suggests re-expression therapies, where a second functional allele is introduced to restore protein levels to compensate for the decreased function of the faulty allele, could be a viable treatment option.

Historically, it was assumed that any treatment targeting neurodevelopmental disorder risk genes would need to be administered very early in development to be effective. This long-held assumption was challenged by reports that re-expression of several risk gene orthologs could reverse multiple altered neurophysiological and/or behavioral phenotypes in adult mice (Creson et al., 2019; Ehninger et al., 2008; Gao et al., 2020; Guy et al., 2007; Mei et al., 2016; Vogel-Ciernia et al., 2013; Zeier et al., 2009). In addition, inactivating orthologs of some of these risk genes in adult mice could also induce the phenotypic impairments previously associated only with altered gene function during development (Ehninger et al., 2008; Guy et al., 2007). Together, these findings suggest there may be a degree of temporal flexibility in the neurodevelopmental processes these genes contribute to, and that some genes typically associated with neurodevelopment may continue to have important functions well into adulthood.

The handful of reports that show the possibility of phenotypic reversibility with gene reactivation in adult mice offer critical insights into which genes are promising candidates for future re-expression-based therapies. However, because of limitations including cost, developmental rate and technical difficulties (e.g. injection of viral vectors for large numbers of animals, etc.), very few risk genes have been tested for adult phenotypic reversibility. Further, the rapidly growing number of risk genes identified in recent years has exacerbated this problem and drastically increased the need for candidate prioritization to better direct research efforts. While most neurodevelopmental disorder risk genes are highly expressed early in pre-natal development (Jin et al., 2020; Parikshak et al., 2013; Satterstrom et al., 2020; Willsey et al., 2013), many continue to be expressed well into adulthood, and we currently do not know if the relationship between temporal expression patterns and inferred temporal functional windows is a significant predictor of whether a gene will be suitable for re-expression therapies. Assessing the phenotypic reversibility of neurodevelopmental disorder risk genes in more high-throughput model organisms offers the ability to rapidly screen a large number of genes to aid in prioritizing risk genes for further study in mammalian models.

The nematode Caenorhabditis elegans offers multiple advantages to systematically assess the temporal requirements and phenotypic reversibility of neurodevelopmental disorder risk genes. There are C. elegans orthologs for a high number of neurodevelopmental risk genes [e.g. >80% of high-confidence ASD risk genes (McDiarmid et al., 2020a,b)], and these genes have repeatedly been shown to be so well conserved that in many cases expression of the human risk gene can compensate for loss of the C. elegans ortholog (Kaletta and Hengartner, 2006; Levitan et al., 1996; McDiarmid et al., 2018a,b; Post et al., 2020). C. elegans has rapid development, growing from egg through well-characterized larval stages (L1, L2, L3 and L4) to egg-laying adults within 3 days. The hermaphroditic reproduction of C. elegans enables large colonies of genetically identical animals to be rapidly and cheaply cultivated. In addition, multiple genomic tools are available to precisely control the spatial and/or temporal activity of genes in vivo (Ashley et al., 2021; Au et al., 2019; Dickinson and Goldstein, 2016; Nance and Frøkjær-Jensen, 2019; Wang et al., 2017; Zhang et al., 2015). Lastly, the phenotypic profiles of hundreds of animals can be simultaneously assessed using automated tracking systems which capture and analyze the impact of genetic perturbations on morphological, sensory and learning behaviors in real time (Husson et al., 2012; McDiarmid et al., 2018a,b; Swierczek et al., 2011).

A behavior that has been increasingly used in high-throughput model organisms to investigate the biological function of neurodevelopmental disorder risk genes and functional impact of disorder-associated variants is habituation (Blok et al., 2022; Kepler et al., 2020). Habituation is a highly conserved form of non-associative learning observed as a decrement in responding to a repetitive stimulus (Rankin et al., 2009; Thompson and Spencer, 1966). Habituation is thought to be an important building block for higher cognitive functions and enable ongoing shifts in behavioral strategy (McDiarmid et al., 2019; Schmid et al., 2015). Alterations in the ability to habituate have been reported in ASD, ID and schizophrenia (McDiarmid et al., 2017) and are hypothesized to contribute to more complex behavioral symptoms (Green et al., 2015; Kavšek, 2004; Kleinhans et al., 2009; Massa and O'Desky, 2012; Williams et al., 2013).

Here, we developed a strategy to assess the phenotypic reversibility and temporal functional windows of three neurodevelopmental disorder risk genes in vivo using CRISPR-Cas9 auxin-inducible degradation (AID) and high-throughput machine vision phenotyping (Fig. 1). We took advantage of the genetic tractability and rapid, stereotyped development of C. elegans to precisely investigate the effects of degrading and re-expressing the protein products of three genes – EBF3•unc-3, BRN3A•unc-86 and DYNC1H1•dhc-1 (‘•’ denotes orthologous relationship between human and C. elegans genes) – at multiple developmental time points in hundreds of age-synchronized genetically identical animals. We found that some level of phenotypic reversibility was possible for each risk gene if re-expression occurred early in post-embryonic development, but only re-expressing EBF3•unc-3 and DYNC1H1•dhc-1 could reverse multiple phenotypic alterations if re-expression began later in life. More broadly, we provide an adaptable strategy and important examples/criteria that illustrate a path towards prioritizing neurodevelopmental disorder risk genes for further study and therapeutic development.

We selected three neurodevelopmental disorder risk genes [as identified by Simons Foundation Autism Research Initiative (Abrahams et al., 2013); Satterstorm et al. (2020)] for reversibility analysis based on the availability of C. elegans strains that contain a neurodevelopmental disorder risk gene ortholog tagged with an auxin-inducible degron at the endogenous locus using CRISPR-Cas9 (see Materials and Methods). The AID system relies on tagging the gene of interest with a short degron peptide tag as well as transgenic expression of TIR1, which is an E3 ubiquitin ligase typically found only in plants (Nishimura et al., 2009; Zhang et al., 2015). In the presence of the auxin class plant hormone indole-3-acetic acid (IAA), TIR1 can associate with the AID degron, adding a poly-ubiquitin chain to the protein of interest, causing it to be degraded by the proteasome (Nishimura et al., 2009; Zhang et al., 2015) (Fig. 1A). We chose to use the AID system as it enables temporal control of protein degradation, which can be reversed by transferring populations of worms to culture plates without IAA (McDiarmid et al., 2020a,b; Zhang et al., 2015). Since the AID degron is tagged to the endogenous locus, protein expression is restored using the native regulatory machinery, therefore bypassing the biological confounds associated with conventional approaches that rely on overexpression. Importantly, our laboratory and others have shown that IAA exposure does not cause any overt effects on C. elegans morphology, locomotion, short-term learning or mechanosensory processing phenotypes (McDiarmid et al., 2020a,b; Zhang et al., 2015).

We systematically assessed the functional consequence of multiple developmental degradation time windows in vivo by transferring animals on or off plates containing IAA at precise time points in C. elegans development (Fig. 1B). We used our high-throughput machine vision tracking system, the Multi-Worm Tracker (MWT) (Swierczek et al., 2011), to quantify 30 phenotypes spanning morphology, baseline locomotion, sensory responding and learning while animals were subjected to a short-term mechanosensory habituation behavioral paradigm (Fig. 1C,D). Our phenotypic features included multiple measures of mechanosensory responding and habituation learning, as both are disrupted across neurodevelopmental disorders (Green et al., 2019, 2015; McDiarmid et al., 2017; Williams et al., 2013), and because we have previously shown that different components of a single behavioral response can be mediated by genetically dissociable underlying mechanisms (Ardiel et al., 2018; Kindt et al., 2007; McDiarmid et al., 2019; Randlett et al., 2019). Inclusion of a range of phenotypes not only aids in characterizing gene function across development, but also enables any unexpected phenotypes caused by protein re-expression to be captured.

The first neurodevelopmental disorder risk gene we assessed was EBF3•unc-3, which is a highly conserved transcription factor. In C. elegans, unc-3 acts to specify the identity of different neuronal classes by initiating and maintaining the expression of class-specific effector genes (Fig. 2A) (Kratsios et al., 2017; Prasad et al., 2008, 1998). Variants in EBF3 have been associated with multiple neurodevelopmental disorders including ASD and ID, and are thought to confer risk through haploinsufficiency or by interfering with DNA binding (Chao et al., 2017; Lopes et al., 2017; Sleven et al., 2017; Tanaka et al., 2017). In C. elegans, unc-3 loss-of-function results in severe locomotion and coordination defects caused by undifferentiated/abnormal identity of cholinergic motor neurons in the ventral nerve cord (Brenner, 1974; Feng et al., 2020; Kratsios et al., 2017; Prasad et al., 1998).

In our paradigm, continuous degradation of UNC-3 from egg through to adulthood produced uncoordinated locomotion, an inability to respond to mechanosensory stimuli, and severe alterations in several other morphological and behavioral phenotypes (Fig. 2B,D and Fig. 3). Early adult inactivation of unc-3 (achieved by beginning IAA exposure at the final larval stage, L4) resulted in effective protein degradation (Fig. S1) and induced impairments for the majority of affected phenotypes, supporting previous findings that unc-3 function is continuously required throughout development (Feng et al., 2020; Kratsios et al., 2017; Li et al., 2020) (Fig. 2B). Importantly, impairments in mechanosensory response probability and several other altered phenotypes were partially rescued in animals when UNC-3 was degraded during development but was re-expressed from the endogenous locus starting in early adulthood (animals taken off IAA immediately after L4) (Fig. 2C).

Our initial findings of partial rescue of some phenotypes following protein re-expression starting at early adulthood motivated us to explore whether earlier restoration of unc-3 function would produce more effective rescue. Re-expression of UNC-3 during late post-embryonic development (ending IAA exposure at L2) resulted in an almost complete rescue of impairments in response probability (Fig. 2E). Further, starting UNC-3 degradation during late post-embryonic development produced more severe impairments, which were similar to those present with continuous degradation, suggesting unc-3 plays a critical role between L2 and L4 for reversal probability (Fig. 2E). Lastly, we explored the phenotypic consequences of exposing or removing 3-day-old animals (young adult, 72 h post-hatch) from IAA. Re-expression of UNC-3 in adulthood did not rescue impairments in response probability, reaffirming that the crucial functional period of unc-3 occurs during development. In line with this, starting UNC-3 degradation at 72 h post-hatch did not induce impairments, suggesting that the role of unc-3 in maintaining the expression of terminal identity genes throughout the lifespan may not be required for normal behavior once the nervous system has fully developed.

Assessment across the three temporal conditions revealed a striking reciprocal pattern of unc-3 temporal function for response probability (Fig. 2E-G). The degree of phenotypic impairment for response probability induced by inactivation of unc-3 at a given developmental time point almost perfectly mirrored the degree of reversibility possible with re-expression (Fig. 2E-G). While some other phenotypes followed this reciprocal pattern, others did not (Fig. 3A-F). For example, reversal duration seems to be mediated by a mechanism that acts earlier in development, with reversibility only possible with early re-expression (Fig. 3G-I). Taken together, these results clearly demonstrate that the temporal windows of gene function can drastically vary by phenotype, and that both the degree of reversibility and number of reversible phenotypes can be influenced by how early re-expression occurs in development. Importantly, these findings reveal that unc-3 expression can rescue multiple phenotypic alterations relatively late in life, prioritizing this gene as a candidate for further study.

BRN3A•unc-86 (also known as POU4F1) is a POU-type transcription factor that plays conserved roles in the initiation and maintenance neuronal identity across species (Badea et al., 2009; Serrano-Saiz et al., 2018; Xiang et al., 1995; Zou et al., 2012). Variants in BRN3A have been associated with abnormal development of sensory structures, including auditory (Huang et al., 2001) and visual cells (Badea et al., 2009). BRN3A may also be affected by a copy number variant identified in a family with ASD that harbor a deletion 153.5 kp upstream of the BRN3A locus (Salyakina et al., 2011). While the gene sequence is not affected, upstream regulatory components of BRN3A may be altered, leading to abnormal development (Salyakina et al., 2011). In C. elegans, unc-86 is thought to be required across the lifespan to maintain the expression of terminal identity genes in multiple neuronal subtypes (Serrano-Saiz et al., 2018; Sze and Ruvkun, 2003) (Fig. 4A).

Analysis of 30 quantitative phenotypes revealed that continuous degradation of UNC-86 specifically impaired mechanosensory response probability (Fig. 4B). Animals were hyporesponsive to mechanosensory stimuli, with a lower likelihood of responding throughout the tracking session, while other parameters of the reversal response (e.g. duration and speed) were not affected. Re-expression of UNC-86 in early adulthood did not rescue these impairments (Fig. 4C) and exposing animals to IAA from L4 onwards did not induce the hyporesponsive phenotype seen with continuous degradation (Fig. 4B) even though effective degradation occurred (Fig. S2). Together, these findings suggest a primarily early developmental role for unc-86 in regulating mechanosensory responses. While we found that impairing unc-86 from L4 onwards did not induce any significant impairments (Fig. 4D), a previous study found that inactivating unc-86 at L4 using a temperature-sensitive allele resulted in impaired chemotaxis to multiple odorants (Sze and Ruvkun, 2003). These differences may suggest that unc-86 has distinct temporal functional windows in different neuronal classes (i.e. unc-86 is continuously required for the function of chemotaxis neurons but is only required in early development for function of mechanosensory neurons).

We next investigated the phenotypic consequences of degrading and re-expressing UNC-86 starting at an earlier developmental time point (specifically 24 h after age-synchronization during late post-embryonic development; beginning after L2). Restoration of unc-86 expression beginning at this earlier time point was sufficient to completely rescue impairments in mechanosensory hyporesponsivity (Fig. 4E). In contrast, we found that degrading UNC-86 from this same time point onwards also impaired response probability (Fig. 4F). These results suggest the window of phenotypic reversibility for unc-86 extends into early post-embryonic development. Moreover, these results indicate that as long as unc-86 has played its role in neurodevelopment in this early critical window, it is no longer required for normal mechanosensory responding in adulthood.

DYNC1H1•DHC-1 is an essential motor protein implicated in diverse processes including cell division and cargo transport along microtubules (e.g. retrograde axonal transport in neurons) (Cianfrocco et al., 2015) (Fig. 5A). Variants in DYNC1H1 have been implicated in several neurodevelopmental disorders including ID and ASD (Satterstrom et al., 2020; Willemsen et al., 2013). Determining the biological functions of DYNC1H1•dhc-1 throughout development has been challenging as dynein loss-of-function results in early embryonic lethality in multiple model organisms (Hamill et al., 2002; Harada et al., 1998; Howell and Rose, 1990; Mains et al., 1990; Robinson et al., 1999). As a result, the role of DYNC1H1•dhc-1 in behavior remains relatively uncharacterized.

Here, we used the AID system to ubiquitously degrade DHC-1. As expected from loss-of-function alleles, continuous degradation of dynein was lethal. Degrading dynein in early adulthood (starting IAA exposure immediately after L4) also resulted in lethality, indicating that dynein function remains essential throughout the late stages of C. elegans development. To determine whether dynein function is essential in adulthood, we ubiquitously degraded dynein in 3-day-old adults (beginning 72 h post-hatch). We found that degrading dynein in adulthood was not lethal, allowing us an opportunity to investigate the biological functions of dynein in adult animals.

Despite its broad expression and essential role in early development, degradation of dynein later in life revealed surprisingly specific roles in adult behavior (Fig. 5A-F). Adult-specific degradation of dynein did not cause severe alterations in morphology or baseline locomotion. Instead, only select components of the mechanosensory reversal response were affected while others were left intact. Response probability was unaffected, as the proportion of worms that responded to each mechanosensory stimulus was similar to the no IAA control (Fig. 5B). However, degrading dynein at 72 h post-hatch caused animals to display deeper habituation of response duration (Fig. 5C) and a slower response speed compared to animals not exposed to IAA (Fig. 5D). Interestingly, the absolute speed trace across the entire experiment shows animals have the ability to respond as fast as control animals but decrement their response speed faster with repeated stimulation (Fig. 5E). These findings suggest that, in adulthood, dynein may function to promote normal habituation of response duration and speed, but not response probability. Taken together, these results support the hypothesis that different components of habituation can be mediated by distinct mechanisms, and reveal novel, adult-specific roles for DYNC1H1•dhc-1 in mechanosensory responding and habituation (Fig. 5F).

To further investigate the role of dynein across development, we took advantage of the ability to activate the AID system cell specifically and obtained a line of C. elegans that allowed for specific and reversible degradation of dynein only in neurons by driving TIR1 expression under the rab-3 promoter (Fig. 6A). Continuous pan-neuronal degradation of dynein did not cause lethality even though effective degradation was observed (Fig. S3), offering us an unprecedented opportunity to determine the phenotypic consequences of decreased dynein function in the nervous system throughout development and whether the resulting impairments were reversible.

Continuous pan-neuronal degradation of dynein caused a broad range of sensory responding impairments, including a low probability of responding to mechanosensory stimuli, rapid habituation of response duration, and lower response speed (Fig. 6B-E and Fig. 7A-D). In this case, multiple phenotypes showed distinct temporal functional windows and reversibility patterns. Re-expression of dynein in early adulthood (ending IAA exposure immediately after L4) did not restore normal mechanosensory responding, but instead resulted in severely impaired habituation of response probability (animals were hyperresponsive and did not learn to decrease their likelihood of responding to repeated stimuli) (Fig. 6B). Adult pan-neuronal degradation of dynein (beginning IAA exposure at L4) did not alter initial response probability but did decrease habituation of response probability (Fig. 6B). For response duration, adult re-expression of dynein was sufficient to fully rescue the impairment seen with continuous degradation, but adult degradation did not induce the response duration impairment (Fig. 6C). Consistent with our findings for ubiquitous dynein degradation in 72-h-old adult animals, continuous degradation of pan neuronal dynein also caused animals to exhibit slower reversal speed. The speed impairment could be rescued with dynein re-expression in early adulthood and was also induced by adult degradation (Fig. 6D), suggesting that dynein is continuously required in neurons to mediate response speed. In addition, while continuously degrading dynein did not alter animal curvature, both adult-specific degradation and re-expression of DHC-1 led to alterations in curvature (Fig. 7E).

We next investigated the phenotypic consequence of degrading and re-expressing dynein in neurons at an earlier time point in development. Re-expressing dynein during early post-embryonic development (starting IAA exposure at L2) fully rescued impairments in response probability while starting degradation at L2 produced a similar level of impairment in response probability as the continuous degradation condition (Fig. 6F). Importantly, the lack of the novel hyperresponsive reversal probability phenotype that occurred when dynein was re-expressed at L4 suggests that for certain genes there will be crucial windows in development when re-expression must occur to avoid inducing alternative impairments. For reversal duration and speed, impairments in both phenotypes were fully rescued with re-expression starting at L2 (Fig. 6G,H). Earlier degradation did not induce impairments in reversal duration, suggesting the crucial functional window of DHC-1 for reversal duration occurs prior to L2 but is not required throughout the lifespan (Fig. 6G). Earlier degradation did lower reversal speed (Fig. 6H), providing more evidence that DHC-1 is continuously required for this phenotype. Together, these results reveal many new roles for dynein in the developing and adult nervous system and illustrate the diversity of temporal functional windows that can be observed for a single gene (Fig. 6E).

All neurodevelopmental disorder risk genes assessed here showed post-embryonic reversibility for at least one impaired phenotype, with several showing phenotypic reversibility relatively late in development (e.g. after stopping IAA exposure at L4; Fig. 8A). Looking across all genes assessed, at least one phenotype within each phenotypic class (morphology, locomotion, mechanosensory responding, and learning) could be reversed later in life, suggesting a degree of flexibility in when multiple cellular processes can occur during development (Fig. 8A). However, not all phenotypes caused by the inactivation of a single gene could be reversed, even if reversibility was possible for other affected phenotypes (Fig. 8A). These results suggest that multiple phenotypic disruptions stemming from a single affected gene can show distinct windows of reversibility (Fig. 8A). In addition, there were also cases where the same organism-level phenotypic disruption (e.g. impaired mechanosensory responding) could be rescued later in development by re-expression of one of the genes, but not others (Fig. 8A-C). Based on our results, EBF3•unc-3 is a promising candidate for further research as many impaired phenotypes could be reversed even when gene function was restored at later stages of development (Fig. 8A), and re-expression did not induce novel phenotypic disruptions (Figs 2 and 3). Further, identifying which phenotypes are most commonly affected across risk genes may provide insight into points of convergence in the mechanisms that are altered in neurodevelopmental disorders (Fig. 8B). We found that reversal probability metrics appeared to be both the most affected and reversible phenotypes across all genes assessed (Fig. 8B,C). This finding supports previous work from multiple model organisms which found that inactivating neurodevelopmental disorder risk genes commonly impaired habituation of response probability (Fenckova et al., 2019; McDiarmid et al., 2020a,b) and that impairments in habituation caused by inactivation of the ASD risk gene ortholog neuroligin (NLGN1/2/3/X•nlg-1) could be partially rescued with adult re-expression (McDiarmid et al., 2020a,b).

We systematically investigated the effect of degrading and re-expressing multiple neurodevelopmental disorder risk gene orthologs across a suite of morphological, locomotor, sensory and learning phenotypes in thousands of freely behaving animals using our high-throughput machine vision tracking system. Taking advantage of the CRISPR-Cas9 AID system allowed us to test whether restoring protein levels through re-expression from the endogenous locus was sufficient for phenotypic rescue at multiple time points throughout development. We found that each gene displayed unique temporal functional windows and phenotypic profiles (Fig. 8). DYNCH1•dhc-1 function is continuously essential for development and plays a specific role in adult mechanosensory behavior. The transcription factor BRN3A•unc-86 can only reverse phenotypic disruptions early in post-embryonic development and is not required for adult function, suggesting a primarily developmental role in mechanosensory responding. The transcription factor EBF3•unc-3 displayed a range of temporal requirements throughout the lifespan and can reverse multiple phenotypic disruptions later in life. In addition to the three genes tested in this study, previous findings from our laboratory provide the temporal requirements for another neurodevelopmental disorder risk gene, the synaptic cell adhesion molecule NLGN1/2/3/4/X•nlg-1. Adulthood re-expression of NLGN1/2/3/4/X•nlg-1 can partially reverse impairments in habituation of response probability; however, once neuroligin has functioned to build a circuit capable of normal sensory processing, it is no longer required in adulthood for normal short-term habituation (McDiarmid et al., 2020a,b). Together, these results reveal a remarkable diversity in temporal phenotypic profiles across neurodevelopmental disorder risk genes that would be missed by approaches that focus on a single phenotype or developmental time point. The approach established in this study can be used to systematically assess the temporal requirements and phenotypic reversibility of neurodevelopmental disorder risk genes at an unprecedented throughput to prioritize risk genes for further assessment.

Using neuron-specific reversible protein degradation, we provide the first description of the role of dynein in behavior across development. We found that continuous degradation of dynein specifically in neurons affected the majority (23/30) of the phenotypes assessed, including multiple morphology phenotypes. Interestingly, we found that both degrading and re-expressing dynein in neurons during early adulthood impaired habituation of mechanosensory response probability. Previous work from our laboratory has found that habituation of response probability is affected by developmental stage such that habituation becomes deeper with age due to circuit rewiring and reduced sensitivity (Beck and Rankin, 1993; Rai and Rankin, 2007; Timbers et al., 2013). The impaired habituation phenotypes seen with both development and adult-specific dynein degradation could both stem from an immature nervous system, such that impairing protein function during early development temporarily impedes developmental processes from occurring, whereas early adult inactivation freezes the nervous system in an immature state. Determining whether this neurodevelopmental freeze mechanism, or alternative, more complex mechanisms (e.g. changes in synaptic physiology) mediate these habituation impairments requires further study. Taken together, these results reveal that dynein has several ongoing functions in the nervous system to modulate sensory and learning behaviors after its essential period in development.

Using high-throughput model systems to rapidly assess the temporal function and phenotypic reversibility of neurodevelopmental risk genes provides critical insight into the emerging principles that should be considered in future re-expression studies. Importantly, we found restoring protein expression in adulthood may induce novel phenotypes that were not observed when proteins were continuously degraded (e.g. altered habituation with neuron-specific re-expression of dynein). This finding reveals the importance of assessing a large number of morphological and behavioral phenotypes to ensure novel adverse phenotypes do not arise when gene function is restored later in development. These studies also offer a reminder to the diversity of functions of a single gene across development. Although a gene may play an important role in a phenotype under study, it also may have other functions that are not immediately obvious. Only by studying multiple phenotypes over the span of development can we begin to understand the breadth of what a given gene contributes to the organism. Future re-expression studies should not only assess whether restoring gene function rescues the well-documented cell functions of the gene of interest, but also capture multiple organism-level phenotypes such as forms of learning or other behaviors commonly altered in neurodevelopmental disorders. Overall, our results suggest that earlier protein re-expression will almost always enable more phenotypic reversibility, yet identification of genes with longer reversibility windows and multiple reversible phenotypes in high-throughput model systems should be a key principle in prioritizing candidates for further study.

In addition, we found that time windows for when a gene is required and when re-expression can reverse impairments did not always align. While EBF3•unc-3 showed reciprocal functional and reversibility windows for many affected phenotypes, we found other genes (e.g. nlg-1 and dhc-1) where certain phenotypic impairments were not induced if the protein was degraded in early adulthood, but phenotypic rescue was possible if protein levels were restored at that same time point in development. In addition, while EBF3•unc-3 and BRN3A•unc-86 have relatively similar functions and temporal mRNA expression levels (Fig. S4), we identified stark differences in their reversibility profiles, indicating that temporal mRNA levels are not necessarily predictive of when a gene in needed for a given phenotype. Together, these findings highlight the need for systematic assessment of the reversibility windows of neurodevelopmental disorder risk genes across different developmental time points, even if there is prior indication of when the gene normally functions.

As the number of genes assessed increases, we may uncover patterns in the molecular attributes that enable certain genes to rescue impairments more broadly than others. For example, EBF3•unc-3 is a putative pioneer transcription factor that may be able to more efficiently remodel chromatin and rewire transcriptional networks outside of its typical developmental window compared to other transcription factors. The approach developed in this study can be adapted to determine the temporal requirements and effects of gene reactivation for cellular phenotypes or more complex behaviors in C. elegans and other model systems. It is important to note that the temporal requirements of gene function and phenotypic reversibility observed in more evolutionarily distant model organisms, such as C. elegans, may not always be conserved in mammals. Nonetheless, information gained from high-throughput model organisms is increasingly valuable as it enables rapid assessment and prioritization of the growing list of risk genes to gain insight into principles governing neurodevelopment and how a nervous system adapts to the re-introduction of a previously inactive protein.

Prior to IAA experiments, all strains were maintained on Petri plates containing nematode growth medium (NGM) that were seeded with Escherichia coli strain OP50 following standard experimental procedures (Brenner, 1974). 96 h post-hatch hermaphrodite animals were used for all experiments.

All human orthologs of all AID strains available at the Caenorhabditis Genetics Center (CGC) were identified using the Alliance of Genome Resources ortholog prediction tool and OrthoList 2 (Agapite et al., 2020; Kim et al., 2018; McDiarmid et al., 2020a,b). AID strains for which the human ortholog corresponded to a known neurodevelopmental disorder risk gene [based on lists generated by recent large-scale sequencing studies and manual literature search (Belmadani et al., 2019; De Rubeis et al., 2014; McDiarmid et al., 2020a,b; Satterstrom et al., 2020)] were selected for analysis. Note that throughout the paper the ‘•’ symbol is used to denote the relationship between the human gene and C. elegans ortholog under study (e.g. DYNCH1•dhc-1).

Auxin administration was performed by transferring animals to bacteria-seeded NGM plates containing IAA (McDiarmid et al., 2020a,b; Zhang et al., 2015). To prepare IAA plates, a 400 mM stock solution of IAA (Thermo Fisher, Alfa Aesar™ #A1055614) was created by dissolving IAA in ethanol. Molten NGM was prepared and allowed to cool to approximately 50°C. The IAA stock was then diluted into separate flasks of molten NGM agar to final concentrations of 0.025 mM, 1 mM and 4 mM. The NGM agar+IAA mixture was then poured into Petri plates and allowed to dry in the dark for 72 h. IAA plates were then seeded with 50 µl of E. coli OP50 liquid culture 48 h before use. All plates were stored in the dark at room temperature (20°C) in a temperature and humidity-controlled room (McDiarmid et al., 2020a,b; Zhang et al., 2015).

Age synchronization by egg lay was used to create the experimental groups for phenotypic analysis as previously described (McDiarmid et al., 2020a,b). For age synchronization, five gravid adults were placed onto either NGM or IAA plates and allowed to lay eggs for 4 h before removal (resulting in 50-100 animals per plate). For the development specific-degradation conditions, approximately 240 progeny were manually transferred from IAA plates onto six regular NGM plates (∼40 animals per plate) either 24 h (at L2), 48 h (at L4) or 72 h (at early adulthood) after synchronization (egg lay). For adult-specific degradation conditions, approximately 240 progeny were manually transferred from regular NGM plates onto six IAA plates (∼40 animals per plate) 24 h (at L2), 48 h (at L4) or 72 h (at early adulthood) after synchronization. All plates remained in the dark other than when animals were being manually transferred to preserve the integrity of IAA. Four to six plates were run for each experimental condition.

The MWT was used for all behavioral tracking experiments (Swierczek et al., 2011). Each plate was subjected to the same short-term habituation behavioral paradigm (see Results and Fig. 1D) that began with a 5 min period to allow worms to acclimate to being placed on the MWT. After acclimation, we collected data for an additional 5 min period to assess baseline locomotion and morphology features (Fig. 1D). Following this baseline period, 30 mechanosensory stimuli were administered to the side of the Petri plate using an automated push-solenoid at a 10 s inter-stimulus interval (Fig. 1D). These non-localized mechanosensory stimuli cause animals to perform a reversal response, where animals briefly crawl backwards before resuming forward locomotion (Rankin et al., 1990). We quantified multiple mechanosensory sensitivity and habituation learning phenotypes from these reversals, which we have previously shown are mediated by genetically dissociable underlying mechanisms. After the 30th stimulus, a 5 min rest period occurred, which was followed by the administration of a final stimulus to assess short-term memory retention of habituation (spontaneous recovery, Fig. 1D). See Table S1 for full description of all phenotypes. All testing occurred in a temperature and humidity-controlled room at approximately 20°C (McDiarmid et al., 2020a,b; Swierczek et al., 2011).

We used the MWT software (version 1.2.0.2) to deliver mechanosensory stimuli and acquire images (Swierczek et al., 2011), and Choreography software (version 1.3.0_r103552) to quantify phenotypes. Choreography filters ‘–shadowless’, ‘–minimum-move-body 2’ and ‘–minimum-time 20’ were used to restrict analysis to animals that moved more than two body lengths and were tracked for 20 s or longer. The ‘MeasureReversal’ plug-in was used to identify animals that reversed within 1 s of the mechanosensory stimulus being administered (Swierczek et al., 2011). Choreography output files were organized using custom R scripts which are freely available at https://github.com/troymcdiarmid/reversibility. All phenotypic features were pooled across the four to six plate replicates (each plate replicate capturing 40-100 animals) per strain. The mean of each condition was then compared using a two-tailed unpaired t-test and Benjamini–Hochberg control of false discovery rate at 0.01. Figures were generated using the ggplot2 and complex heatmap packages in R (Gu et al., 2016; Wickham, 2016).

Degradation groups were created as described above. All images were taken using 96-h-old animals. Adult animals were anesthetized by being placed in 10 µl of 7.5 mM Levamisole (Sigma-Aldrich) and 0.225 M BDM (2,3-butanedione monoxime) (Sigma-Aldrich) on glass microscope slide containing a 2% agar pad. After animals were paralyzed, a 1.5 coverslip was placed on top the agar pad. A Leica SP8 white light laser confocal microscope and 63× oil immersion lens was used for imaging. Step size was 0.3 µm. GFP was excited using a 488 nm wavelength laser with emitted light collected through a 490-778 nm bandpass filter. mNeonGreen was excited using a 506 nm wavelength laser with emitted light collected through a 512-742 nm bandpass filter. For each condition, four to five animals were scored for the presence of GFP or mNeonGreen. Final figures were generated using ImageJ (National Institutes of Health, Bethesda, MD, USA).

The following strains used are available through the CGC: CA1200 ieSi57[eft-3p::TIR1::mRuby::unc-54 3'UTR+cbr-unc-119(+)] II; OH13988 ieSi57[eft-3p::TIR1::mRuby::unc-54 3'UTR+cbr-unc-119(+)] II; unc-3(ot837[unc-3::mNeonGreen::AID]) X; OH15227 unc-86(ot893[unc-86::3xFlag::mNeonGreen::AID]) III; CA1207 dhc-1(ie28[dhc-1::degron::GFP]) I; CA1210 ieSi57[eft-3p::TIR1::mRuby::unc-54 3'UTR+cbr-unc-119(+)] II; dhc-1(ie28[dhc-1::degron::GFP]) I.

The following strains were generated using standard genetic crosses: VG937 ieSi57[eft-3p::TIR1::mRuby::unc-54 3'UTR+cbr-unc-119(+)] II; unc-86(ot893[unc-86::3xFlag::mNeonGreen::AID]) III; VG946 mizSi6[rab-3p::TIR1::unc-54 3′UTR+LoxP pmyo-2::GFP::unc-54 3′UTR prps27::NeoR::unc-54 3′UTR; LoxP] V; dhc-1(ie28[dhc-1::degron::GFP]) I.

Successful crosses were determined through visual confirmation of fluorescent reporters as well as PCR-based genotyping using the following primers: TIR1 sequence: forward, 5′-GACCGTAACTCCGTCTCC-3′; reverse, 5′-CGTTGGTGGTGATGATTTGAC-3′; AID degron sequence: forward, 5′-CCTAAAGATCCAGCCAAACC-3′; reverse, 5′-CTTCACGAACGCCGC-3′; or forward, 5′-GATCCAGCCAAACCTCCGGC-3′; reverse, 5′-CTTCACGAACGCCGCCGC-3′.

The authors thank Dr Kota Mizumoto for the rab-3p::TIR1 strain that was used for neuron-specific degradation and the CGC (funded by National Institutes of Health Office of Research Infrastructure Programs, P40 OD010440) for other strains used within this research.