The expression of the unc-42 transcription factor was previously analyzed using antibody staining and reporter constructs including 2.6 kb of sequences upstream of the unc-42 locus (Baran et al., 1999). Tentative cellular identifications of sites of expression were provided for many, but not all, expressing cells (Supplementary file 1). We revisited this expression data with a set of three distinct reagents: a chromosomally integrated, multi-copy fosmid-based reporter construct in which the C-terminus of the locus was tagged with gfp, as well as two different engineered strains in which we inserted either gfp or TagRFP at the 3′ end of the unc-42 locus using CRISPR/Cas9 genome engineering (Figure 1A). All three reagents showed the same expression pattern, with the only difference being that the rfp strain showed a delayed onset of expression in the embryo, likely due to delayed fluorophore maturation.

The availability of landmark strains for individual neuron types, particularly the novel NeuroPAL landmark strain that allows for disambiguation of all 118 neuron classes (Yemini et al., 2021), allowed us to determine the complete pattern of unc-42 expression during larval development and adulthood. This analysis substantially revised and extended the previously reported expression pattern by Baran and colleagues (Baran et al., 1999) (comparison is shown in Supplementary file 1). Specifically, we found that in the adult nervous system unc-42 is strongly and consistently expressed in 40 neurons located in the head of the worm that fall into 15 anatomically distinct neuron classes (Figure 1B, C, Supplementary file 1, Figure 1—figure supplement 1). At the first larval stage, expression is detected in the same set of neuron classes as observed in the adult, with the addition of very weak and inconsistent expression in AVJ and SIA that disappears by the adult stage (Figure 1—figure supplement 1). In the embryo, expression is first observed in a few neuroblasts before the bean stage. At the bean stage, when most neurons have terminally divided, unc-42 expression commences and reaches the full complement of unc-42(+) cells at the 1.5-fold stage (Figure 1B; Figure 1—figure supplement 1 for quantification of levels). The cellular sites of expression in the embryonic nervous system appear to be the same as observed post-embryonically, a notion further supported by recent scRNA data (Packer et al., 2019). No expression is observed outside the nervous system.

At first sight, the complete set of neurons that express UNC-42 in the mature nervous system do not share obvious commonalities. UNC-42(+) neurons include sensory, inter-, and motor neurons and display distinct neurotransmitter identities (glutamatergic, cholinergic, or peptidergic) (Figure 1—figure supplement 1). Moreover, there is no obvious lineage relationship among the UNC-42(+) neurons (Figure 1, Figure 1—figure supplement 2). The main worm neuropil can be abstracted into a laminar structure with specific clusters (‘strata’) or neighborhoods of neurons defined by ultrastructural adjacencies of neuron processes (Brittin et al., 2021; Moyle et al., 2021). Despite all UNC-42(+) neurons projecting into this neuropil (Figure 1E), UNC-42(+) axons are not exclusively part of one stratum (Brittin et al., 2021; Moyle et al., 2021; Supplementary file 2). However, we noted a striking theme of the 40 neurons (15 neuron classes) that express UNC-42(+): they form a network of densely connected neurons (Figure 1D, E).

We tested whether this observation can be explained by chance alone, using a previously described approach that analyzed gene expression in relation to synaptic connectivity (Arnatkeviciūtė et al., 2018). We first calculated the chance that any two neurons from a random set of 40 neurons in the connectome are connected via chemical or electrical synapses. We then compared this to the chance that any two neurons from a set of 40 unc-42-expressing neurons are synaptically connected. We found that unc-42-expressing neurons indeed are more likely to be synaptically connected to each other than neuron pairs sampled at random from the whole connectome (Supplementary file 3).

One potential problem of this methodology is that among a set of randomly chosen 40 neurons several neurons may not be in close enough physical proximity to potentially form synapses. To address this limitation, we utilized a recently introduced unbiased approach, termed network differential gene expression (NDGE) analysis, with an appropriate null model that takes into account membrane contact (Taylor et al., 2021) to estimate the significance of unc-42 expression with synaptic linkage. This approach is a generalization of differential gene expression analysis (Wang et al., 2019) that is in widespread use in single-cell RNA-sequencing literature where the gene expression differences between groups of cells are statistically tested. In NDGE, the groups that are compared are synaptically linked versus non-synaptically linked neuron pairs. Paired multiplicative gene expression that assesses ‘homomeric’ expression of a gene in synaptically linked neurons yields a p-value that denotes the significance of a gene-pair towards promoting or inhibiting synaptic linkage. The statistical significance of this analysis is driven by a carefully generated null distribution where the connectome is randomly rewired while maintaining its topological properties such as its degree distribution and membrane adjacency (Rao et al., 1996). Since many possible combinations of genes could by chance show spurious associations to synaptic linking, we perform false discovery rate (FDR) procedure that limits such false discoveries to 5% (Benjamini and Hochberg, 1995). This analysis indeed confirms the significant association of UNC-42 expression with synaptic linkage (p=5.78×10⁻⁵) (Figure 1F).

We further extended these two synaptic association tests to all members of the homeobox gene family that are selectively expressed in a subset of neurons in the mature nervous system (Reilly et al., 2020). This analysis shows that the expression of a total of eight homeobox genes each shows a significant enrichment in synaptically connected neurons, both by the NDGE analysis and the methodology by Arnatkeviciūtė et al., 2018 (Supplementary file 3).

The set of unc-42(+) neurons includes a single sensory neuron involved in nociceptive behavior (Kaplan and Horvitz, 1993) and downstream inter- and motor neurons that have been shown, mostly through microsurgical removal or genetic manipulations, to shape the locomotory response to aversive cues, including neurons involved in reversal behavior, omega turns, backward locomotion, restriction of head movement, pausing, and locomotory speed (Bhattacharya et al., 2019; Chalfie et al., 1985; Chatzigeorgiou et al., 2013; Ezcurra et al., 2011; Gray et al., 2005; Hamakawa et al., 2015; Hart et al., 1999; Hart et al., 1995; Hilliard et al., 2002; Hilliard et al., 2004; Kindt et al., 2007; Komuniecki et al., 2012; Sambongi et al., 1999; Shen et al., 2016; Walker et al., 2009; Yeon et al., 2018). Using an automated WormTracker system (Yemini et al., 2013), we found that animals carrying the canonical allele of unc-42, e419 (a premature stop codon in the homeobox; Baran et al., 1999), display behavioral defects that match the defects observed after functional disruption of normally unc-42(+) neurons (Figure 2; Supplementary file 4). This match is not simply the result of unc-42 mutant animals being completely immobile since there are many locomotory components that are unaffected in unc-42 mutants (Supplementary file 4). Together with the previously reported inability of unc-42 mutants to respond to aversive sensory cues (Baran et al., 1999), this analysis suggests that unc-42 is essential for unc-42(+) neurons to exert their proper function.

The restriction of unc-42 expression to a set of synaptically highly interconnected neurons made us first ask whether unc-42 affects neuronal communication among the unc-42(+) neurons, using again the canonical e419 nonsense allele. To this end, we systematically examined the expression of enzymes and transporters that mark the distinct neurotransmitter identities of the unc-42(+) neurons. Some of this analysis had already previously been done for 6 of the 15 unc-42(+) neurons: unc-42 was shown to be required for the glutamatergic identity of the ASH and AIB neurons, as assessed by loss of eat-4/VGLUT expression (Bhattacharya et al., 2019; Serrano-Saiz et al., 2013) and the cholinergic identity of the RMD, SMD, SIB, and RIV motor neurons, as assessed by unc-17/VAChT and cho-1/ChT expression (Pereira et al., 2015). We extended this analysis to the remaining nine cholinergic neuron classes in the unc-42(+) circuit. We found that loss of unc-42 affects unc-17/VAChT expression and hence cholinergic identity in all but one of the normally unc-42(+) neurons (Figure 3A–C). A summary of the effect of unc-42 on neurotransmitter identity is provided in the context of a circuit diagram that displays all the unc-42(+) neurons (Figure 3D).

We examined the receiving end of chemical synaptic neurotransmission by analyzing the expression of ionotropic glutamate (Glu) receptors, nmr-1, glr-1, glr-2, glr-4, and glr-5, and two ionotropic acetylcholine (Ach) receptors, acr-2 and acr-15, in the unc-42(+) circuit. The expression and unc-42 dependence of the ionotropic Glu receptors had been examined before (Baran et al., 1999; Brockie et al., 2001; Wightman et al., 2005). However, we found that some of these receptors are expressed in different set of unc-42(+) cells than previously reported likely explained by our usage of updated reagents that more precisely identify neuron classes. We found the expression of each of these genes to be affected in a cell type-specific manner in unc-42 mutants (Figure 4).

Moving from glutamatergic receptors to cholinergic receptors, we found that loss of unc-42 also affects the expression of cholinergic receptor systems, specifically the AChR subunits acr-2 and acr-15 in the RIV, RMD, SAA, and SMD neurons and expression of the AChR-like des-2 gene in the AVD neurons (Figure 5A–C). Moreover, we found that expression of the tyramine-gated chloride channel, LGC-55, which makes the unc-42(+) circuit responsive to tyramine signaling from the RIM neurons (Donnelly et al., 2013), is affected in the AVB, RMD, and SMD neurons of unc-42 mutants (Figure 5D). A summary of the effect of unc-42 on neurotransmitter receptor expression is provided in the context of a circuit diagram that displays all the unc-42(+) neurons (Figure 5E). We note that the effect of unc-42 on neurotransmitter expression (presynaptic neurotransmitter synthesis/transport and/or postsynaptic neurotransmitter) is often not fully expressive or fully penetrant, indicating that unc-42 is not the sole determinant of these neuronal identity features. This is a recurrent theme in the analysis of unc-42 function throughout this paper. We will return to this point at the end of this paper when we discuss transcriptional cofactors of unc-42.

Extending our analysis beyond chemical synaptic transmission, we asked whether neuromodulatory signaling by neuropeptides is affected within or to/from unc-42(+) neurons. Previous work has already shown three neuropeptides (flp-1, flp-32, nlp-15) and one neuropeptide receptor (npr-9) to be dependent on unc-42 in specific neuron types (Bhattacharya et al., 2019; Serrano-Saiz et al., 2013; Wightman et al., 2005; Wood and Ferkey, 2019). We extended this analysis by examining the expression of seven additional neuropeptide-encoding genes (producing at least 23 distinct neuropeptides) and of eight additional neuropeptide receptors. Several of these peptides and receptors are known cognate ligand/receptor pairs. For example, the flp-7 and flp-12 neuropeptides activate the frpr-8 receptor (I. Beets, pers. comm.) and the flp-18 neuropeptide binds to the npr-4 and npr-11 receptors (Cohen et al., 2009). We found that the highly patterned expression of these peptides and receptors is severely affected in unc-42(e419) mutant animals (Figure 6A–O). We again summarize these findings in a circuit diagram (Figure 6P), and we again note the often partial penetrance of the effects, as mentioned in the previous section.

Many of these unc-42-dependent neuropeptides and neuropeptide receptors have been previously shown to be involved in locomotory behavior, and we found that these behavioral defects match subsets of the defects observed in unc-42 mutants. For example, animals lacking flp-18, nlp-15, npr-9, and npr-11 display decreased backwards motion (i.e., decreased reversals) (Bhardwaj et al., 2018; Campbell et al., 2016; Chalasani et al., 2010; Yemini et al., 2013), thereby phenocopying the unc-42 defects that we have described here (Figure 2). Among these, flp-18 and npr-11 are expressed in command interneurons that are responsible for backward locomotion, and we found their expression to be unc-42 dependent in these neurons (Figure 6C, K). npr-11 is also expressed in the AIB interneuron, and, like AIB-ablated animals (Gray et al., 2005), animals lacking npr-11 also phenocopy unc-42 in that they conduct fewer omega turns. We found that npr-11 expression in AIB is downregulated in unc-42 mutants (Figure 6K).

In order to further evaluate the potential role of additional neuropeptides and neuropeptide receptors in the locomotory behaviors associated with unc-42 function, we analyzed the behavior of animals that lack the neuropeptide flp-21 and the neuropeptide receptors ntr-1 and ntr-2. We found that, like unc-42 mutants, flp-21 and ntr-2 mutant animals displayed decreased backward motion (Figure 7A, Figure 7—figure supplement 1). flp-21 is regulated by unc-42 in the ASH neuron (Serrano-Saiz et al., 2013), the sole sensory neuron in the circuit responsible for backwards motion in response to noxious stimuli. Other behaviors in these mutant animals that phenocopied unc-42 were head bend mean (ntr-1 and ntr-2) and absolute midbody speed (ntr-1) (Figure 7A). The neurons responsible for these behaviors are yet to be elucidated. Taken together, locomotory defects observed in unc-42 mutant animals match those observed upon loss of neuropeptidergic signaling systems that are transcriptionally regulated by unc-42.

Beyond neuropeptidergic communication, unc-42 may also affect signaling via other internal signaling systems. We infer this from the observation that unc-42 also affects the expression of the orphan GPCR sra-11 in the AVB neurons (Figure 7—figure supplement 2). sra-11 is involved in an associative learning paradigm and responds to an as yet unknown, likely internal signal (Remy and Hobert, 2005).

Direct sensory input into the unc-42(+) circuit is provided by the polymodal ASH neurons. Among other modalities, ASH senses high osmolarity (Kaplan and Horvitz, 1993) and this sensory paradigm is disrupted in unc-42 mutants (Baran et al., 1999). A putative osmosensor, osm-10, is expressed in ASH (Hart et al., 1999), and osm-10 expression is disrupted in unc-42 mutants (Figure 7—figure supplement 3; Wood and Ferkey, 2019). ASH also expresses many putative sensory receptors of the G-protein-coupled receptor family, likely involved in the chemorepulsive function of ASH (Vidal et al., 2018). The expression of several of these GPCRs, as well as downstream G-alpha proteins, were previously found to require unc-42 (Baran et al., 1999; Serrano-Saiz et al., 2013; Wood and Ferkey, 2019). We added another GPCR, srh-15, to the list of unc-42-regulated GPCRs (Figure 7—figure supplement 3). Like other transcriptional regulators that control the individuality of distinct sensory neuron types (Alqadah et al., 2015; Masoudi et al., 2018), we found that unc-42 does not control expression of the pansensory cilia gene osm-6 (Figure 7—figure supplement 3). In conclusion, unc-42 controls the proper specification of the sensory neuron that provides sensory input into the unc-42(+) nociceptive reflex circuit.

The data described above demonstrates that unc-42 affects sensory input into the set of 15 interconnected neurons, as well as communication within these interconnected neurons. We next asked whether apart from controlling neuronal circuit activity unc-42 may also impact on the assembly of unc-42(+) into functional circuitry. We first examined systematically and quantitatively whether neurons that normally express UNC-42 are generated and whether their soma adopt their correct positions in unc-42 mutant animals. To this end, we again used the NeuroPAL neuronal landmark strain, which labels the position of all neurons with a number of markers, including a panneuronally expressed marker (Yemini et al., 2021). We found that all normally unc-42-expressing neurons are generated in unc-42 null mutants, that they express the panneuronal reporter normally, and that they are positioned in a manner that is not quantitatively different from their relative position in wild type animals (Figure 8A, Figure 8—figure supplement 1).

Previous analysis has described that some command interneurons display axon pathfinding defects along the ventral nerve cord in unc-42 mutants (Baran et al., 1999). Since command interneurons are only a small subset of neurons that express unc-42, we sought to expand this analysis to other unc-42(+) neurons. This proved to be a substantial challenge because, as we described above (and further below), the vast majority of molecular markers that label individual unc-42(+) neurons fail to be expressed in unc-42 mutant animals. Two exceptions include the srd-10::gfp transgene that labels ASH axodendritic morphology and the hlh-34::gfp transgene that labels the morphology of the AVH interneuron. Since the expression of both reporters is either not affected or only mildly affected in unc-42 mutants, they allowed us to visualize ASH and AVH process outgrowth, revealing that unc-42 is indeed required for proper axon extension of both neuron classes; unc-42 is required for proper AVH axon extension along the ventral nerve cord, and unc-42 is required for ASH axons to reach the dorsal midline (Figure 8B, C).

We also noted that loss of unc-42 affects the proper axon extension to the dorsal midline of a subset of other sensory neurons that normally do not express unc-42 (OLL, AWA, AWB) (Figure 8E, F). This apparent cell non-autonomous effect is selective; for example, the ASI neurons are not affected in unc-42 mutants (Figure 8—figure supplement 2A). The apparent non-autonomy of unc-42 function is further exemplified in the ventral nerve cord, where previous work has revealed axon outgrowth defects in unc-42 mutants in the HSN, PVQ, and PVP neurons (Wightman et al., 1997), none of which normally express unc-42 (Figure 1). As we will show below, unc-42 affects the expression of the secreted unc-6/Netrin axon guidance cue in command interneurons, which may provide a straightforward explanation for the non-autonomous function of unc-42.

The effect of unc-42 on axon outgrowth already suggests a role for unc-42 in the proper assembly of the nociceptive circuit marked by unc-42 expression. To take this analysis one step further, we asked whether unc-42 may also affect the generation of chemical synapses in this circuit. To this end, we made use of the availability of unc-42 mutant animals that were fixed and embedded by Nichol Thomson in the context of early efforts by the Brenner lab to systematically analyze behavioral mutants by electron microscopy (Brenner, 1973). For this analysis, the unc-42(e270) allele was used, which contains a missense mutation in a highly conserved residue of the homeodomain (Baran et al., 1999). We sectioned these blocks and traced all of the axonal process in the nerve ring area (~15 µm) of a single animal. We chose the nerve ring because this is where the majority of synaptic connections between unc-42(+) neurons are made. Tracing processes through 309 sections, we were able to assign a total of 19 processes to specific neuron types, 8 of which normally express unc-42 (see Materials and methods).

We found that loss of unc-42 does not affect the overall relative arrangement of the 19 identified neuronal processes with the part of the nerve ring that we were able to analyze (Figure 9A, B). We quantified this arrangement through counting sections in which two identified neuron pairs are adjacent to one another and found that the processes that are adjacent to one another in wild type animals retain their adjacency in unc-42 mutants (Figure 9C, D). In terms of synaptic connectivity, we did not observe synaptic defects among neurons that normally do not express unc-42 (Figure 9C, D). However, we found that chemical synaptic connectivity is reduced in neurons that normally express unc-42 in the pre- and postsynaptic neurons (Figure 9C, D). The most striking example of an affected connection in which both neurons are unc-42(+) is between the interneurons SAAVL and AVAL. For this connection, the wild type animal makes 22 synapses across 50 serial sections and the unc-42 mutant animal makes only 2 synapses across two serial sections in spite of the normal adjacency of the two processes. In addition to connections where both the pre- and postsynaptic neurons are normally unc-42(+), some of the connections where only the postsynaptic neuron is normally unc-42(+) are also disrupted. For example, in the connection between the presynaptic unc-42(-) sensory neuron OLLR and the postsynaptic unc-42(+) motor neuron SMDVL, the wild type animal makes five synapses in a total of 11 sections, whereas the unc-42 mutant animal makes only one synapse across two sections. The above-mentioned process adjacency data indicates that the synaptic defects that we observed cannot be explained by loss of adjacency alone. For example, the wild type SAAVL and AVAL processes are adjacent for 158 sections, whereas in the unc-42 mutant animal these two neurons are adjacent for 131 sections, a reduction much milder than the loss of chemical synaptic connectivity (22 synapses over 50 serial sections vs. 2 synapses across to 2 serial sections in the unc-42 mutant). In conclusion, our fine-grained anatomical analysis reveals that while axons are placed correctly in the analyzed unc-42 mutant animal, there are defects in the generation of synaptic contact. This is consistent with the hypothesis that unc-42 is required for proper circuit assembly.

With the limitations of sample size of the electron micrographical analysis in mind, we pursued alternative means to assess synaptic connectivity in unc-42 mutants with greater quantitative rigor. Due to the widespread effect of unc-42 on neuron-class-specific molecular markers, we could not easily utilize GRASP technology to label chemical synapses. However, we were able to assess the integrity of electrical synapses in unc-42 mutant animals. Electrical synaptic contacts are abundant among the unc-42(+) neurons (Figure 1D; Cook et al., 2019; White et al., 1986) and are formed by members of the innexin gap junction proteins (Hall, 2017). We had previously mapped the expression pattern of all neuronally expressed innexins (Bhattacharya et al., 2019). Several of them are indeed expressed in the unc-42(+) nociceptive circuit. inx-18a and inx-19 show a particularly good match, and we examined the expression pattern in unc-42 mutant animals. We found that in unc-42 mutant animals inx-18a and inx-19 expression is selectively downregulated in those neurons that normally express unc-42. Specifically, inx-19 expression is affected in ASH sensory neuron, AVA, AVB, AVD, and AVE command interneurons, in the peptidergic AVK interneurons, and the RMD head motor neurons (Figure 10B), while inx-18a expression is affected in the AVK interneurons, the AVA, AVB, and AVD command interneurons, and the RIV inter/motor neurons (Figure 10C). Expression of the very broadly expressed innexin unc-7 is also affected in a number of neurons in unc-42 mutant animals (Figure 10D). Lastly, in previous studies on the function and regulation of the innexin inx-6, we had already shown that unc-42 affects inx-6 expression in the AIB neurons (Bhattacharya et al., 2019).

To test whether loss of innexins phenocopies the loss of unc-42, we tracked inx-18, inx-19, and unc-7 mutant animals and found that several components of locomotory behavior were affected in inx-19 and unc-7 mutant animals in a manner that is similar to unc-42 mutant animals (Figure 7A, Figure 7—figure supplement 1). Taken together, the prominent loss of expression of innexins is a strong indication that electrical synaptic connectivity is not properly established in unc-42 mutants.

Considering the wiring defects of unc-42 mutants, we asked whether unc-42 may affect the expression of genes with known or potential roles in cell/cell recognition. The most obvious candidate is the UNC-6/Netrin protein, which is known to affect both axon pathfinding and synapse formation in C. elegans and other systems (Colón-Ramos et al., 2007; Hedgecock et al., 1990). UNC-6 is expressed in a highly restricted manner during C. elegans development, and the few sites of neuronal expression include the command interneurons (Wadsworth et al., 1996; Weinberg et al., 2018). We found that unc-6/Netrin null mutants show defects in amphid sensory neuron axon outgrowth that mimic the effects observed in unc-42 mutants. For example, the ASH axons display a 40% penetrant axon extension defect in unc-6(ev400) null mutants (n = 91), comparable to what is observed in unc-42 mutants (Figure 8). We found that in unc-42 mutants unc-6/Netrin expression (as assessed with an unc-6 fosmid-based reporter reagent) (Weinberg et al., 2018) is eliminated from all command interneurons (Figure 11A). We attempted to assess whether unc-42 axon pathfinding defects can be rescued by force-expressing unc-6 in an unc-42 mutant background using the unc-42-independent nmr-1 promoter as driver for unc-6. Rescue was not observed, possibly due to the late onset of the nmr-1 driver expression (Figure 8—figure supplement 2B). No earlier, unc-42-independent command interneuron-restricted drivers are currently available.

The synaptically connected unc-42(+) neurons also express a host of cell surface molecules with potential roles in axon pathfinding and synapse formation. We examined 10 cell surface proteins (nine of them of the IgSF family members) that show expression in subsets of unc-42(+) neurons for their dependence on unc-42. These include genes previously shown to be involved in axon pathfinding and/or axon fasciculation (lad-2/L1CAM, rig-6/Contactin, ncam-1/NCAM) and/or organizing synaptic structure and/or function (nlg-1/Neuroligin, syg-1/KirreL, rig-5/IgLON, oig-1, rig-3) (Hashimoto et al., 2009; Howell et al., 2015; Kim and Emmons, 2017; Maro et al., 2015; Schwarz et al., 2009; Shen and Bargmann, 2003; Wang et al., 2008), as well as two IgCAMs with presently unknown functions (rig-1 and rig-5). These proteins display a unique combinatorial expression in each unc-42(+) neurons (with the exception of AVK and AVH). We found that expression of each one of these 10 genes is profoundly affected in unc-42 mutant animals (Figure 11B–J), summarized in Figure 11K.

To assess possible function of these cell surface molecules in the unc-42-dependent nociceptive reflex circuit, we tested animals that carry mutations in either of seven of the unc-42-dependent cell recognition molecules for locomotory defects. We found that nlg-1/Neuroligin and three IgSF members, rig-6/Contactin, ncam-1/NCAM and rig-3 phenocopy subsets of the unc-42 locomotory defects (Figure 7B, Figure 7—figure supplement 1). Similar to unc-42 mutants, rig-6 and ncam-1 mutants display fewer omega turns and are expressed, in an unc-42-dependent manner, in omega turn controlling neurons (ASH, RIV, or AIB). ncam-1(-) animals also phenocopy unc-42(-) animals in regard to increased pausing, a behavior that is controlled by the AIB neuron, where the expression of ncam-1 is unc-42 dependent. nlg-1 mutants phenocopy unc-42 mutants in decreased backwards motion, a behavior that is controlled by the AVA command interneuron, where the expression of nlg-1 is unc-42-dependent (Figure 7B). Cases in which we observed no or limited locomotory defects are not conclusive because several of the available mutants are not clear molecular nulls. As with other cases described above where a target gene mutation phenocopies aspects of the unc-42 mutant phenotype, it is important to realize that the phenocopy does not prove that the respective target gene indeed can be made responsible for the unc-42 mutant phenotype (e.g., the target gene could affect the phenotype from a complete different cell where unc-42 displays no function). Nevertheless, the phenocopy is an encouraging prerequisite for being a functionally relevant target of unc-42.

We observed no obvious defects in axon pathfinding or synaptic vesicle clustering (rab-3::gfp marker) of a subset of unc-42(+) neurons (ASH, AIB, SAA) in rig-6, rig-3, ncam-1, and nlg-1 mutants (Figure 8—figure supplement 3), but since RAB-3::GFP labels presynaptic sites indiscriminately, this approach lacks the anatomical resolution to draw any definitive conclusions. More analysis will be required to assess whether these genes may affect synaptic connectivity of unc-42(+) neurons.

The analysis described above raises a number of questions. First, does UNC-42 control the many target genes described above directly or indirectly? Second, how does UNC-42 activate distinct target genes in distinct neuron types? Using the single cell transcriptome atlas of all C. elegans neurons (Taylor et al., 2021) and a phylogenetic footprinting pipeline, described in the accompanying paper by Glenwinkel et al., we found that functionally validated UNC-42 binding sites are enriched in the cellular transcriptomes of each one of the 15 neuron classes that express UNC-42 including those genes whose expression we have shown here to be unc-42-dependent (representative examples are shown in Figure 12A) (Glenwinkel et al., accompanying paper). This analysis strongly suggests that UNC-42 directly activates the expression of terminal gene batteries in all unc-42(+) neurons.

If UNC-42 binding sites are enriched in all the UNC-42-dependent target genes described above, why are they not activated in all UNC-42(+) neurons? The most parsimonious explanation is that UNC-42 requires neuron-type-specific cofactors to activate distinct sets of neuron class-specific gene batteries. We found this scenario to indeed apply in many UNC-42(+) neuron classes. For example, we had previously shown that the EBF/Collier ortholog unc-3 affects the cholinergic identity of command interneurons in a manner similar to the unc-42 effect (Pereira et al., 2015). Consistent with UNC-3 and UNC-42 working together, we found that UNC-42 binding sites are co-enriched with UNC-3 sites in genes expressed in all command interneurons (Figure 12B). To functionally validate this potential interaction, we made use of the fact that unc-42 and unc-3 single mutants each only display a partially penetrant loss of unc-17/VAChT expression. Building unc-42; unc-3 double mutant animals, we found that in some of the affected neurons the defects are synergistic relative to the single null mutants, indicating that both genes cooperate to control cholinergic neurotransmitter identity in the command interneurons (Figure 3).

Since command interneurons come in different types, one may expect that UNC-42 and UNC-3 interact with distinct additional factors in distinct command interneuron types. Loss of the ARID-type cfi-1 transcription factor causes differentiation defects in one of the command neurons, the AVD neurons (Shaham and Bargmann, 2002) resembling those observed in unc-42 mutants. In contrast, the nuclear receptor fax-1 controls the expression of genes in AVA and AVE, but not AVD (Wightman et al., 2005). Since binding sites for CFI-1 and FAX-1 are also defined, we again used the phylogenetic footprinting pipeline described in the accompanying paper by Glenwinkel et al. We found that the AVD gene battery contains an enrichment for CFI-1 binding sites (in addition to the UNC-42 and UNC-3 binding site enrichment), while the other command interneurons display enrichments of FAX-1 binding sites (in addition to the UNC-42 and UNC-3 binding site enrichment) (Figure 12B). Notably, the AVK interneurons that require both fax-1 and unc-42 for their proper differentiation (Wightman et al., 2005) also display a co-enrichment for FAX-1 and UNC-42 binding sites. Hence, we surmise that UNC-42 cooperates with UNC-3 and CFI-1 to specify AVD identity and with UNC-3 and FAX-1 to specify AVA and AVE identity, and with FAX-1, but not UNC-3, to specify AVK identity.

Apart from synergistic activities of unc-42 and unc-3 in command interneurons, we identified another genetic interaction in the RMF and RMH head motor neurons. As described above, unc-42 single mutants have mild defects in execution of cholinergic fate of these neurons (unc-17/VAChT expression) (Figure 3). The LIM homeobox gene lim-4 was previously shown to be expressed in a number of mostly cholinergic head sensory, inter-, and motor neurons and acts as a terminal selector to control AWB and SMB neuron identity (Sagasti et al., 1999; Kim et al., 2015; Alqadah et al., 2015; Pereira et al., 2015). However, its function in other neurons was not previously investigated. We find that like unc-42 single mutants lim-4 single mutants have mild defects in the execution of cholinergic fate in the RMF and RMH head motor neurons (unc-17/VAChT expression) (Figure 3). These defects are strongly enhanced in unc-42; lim-4 double mutant (Figure 3).

Moving beyond cholinergic neurons, we considered the previously uncharacterized AVH interneuron, whose identity specification is affected in unc-42 mutants, as described above. The hlh-34 gene, a bHLH-PAS transcription factor, is exclusively expressed in the AVH neuron throughout the life of the AVH neuron (see Materials and methods). We used CRISPR/Cas9 to engineer a loss-of-function allele of hlh-34 (Figure 13A) and found AVH differentiation defects in these animals (Figure 13B, C) that match those observed in unc-42 animals. Moreover, hlh-34; unc-42 double null mutants show strongly enhanced mutant phenotypes (Figure 13C). Hence, UNC-42 may cooperate with HLH-34 to specify AVH identity.

In the glutamatergic AIB interneuron, we identified two potential collaborators for unc-42, the Meis-type homeobox gene unc-62 and the Pbx-type homeobox gene ceh-20, both of which co-expressed with unc-42 exclusively in the AIB neurons (Reilly et al., 2020). unc-62/Meis and ceh-20/Pbx mutant animals phenocopy the AIB differentiation defects observed in unc-42 mutant animals (Figure 13D–I). Specifically, loss-of-function alleles of both genes results in defects in expression of several key identity features of AIB, including loss of glutamatergic identity (eat-4/VGLUT expression) and loss of expression of several innexin genes (Figure 13D, E, G).

Lastly, in the SIB lateral motor neurons, the differentiation defects observed in unc-42 mutants are phenocopied by loss of the ceh-24 homeobox gene (Schwarz and Bringmann, 2017), suggesting that these two transcription factors may cooperate in the SIB neurons. Because the effect of both mutants is fully penetrant, we refrained from building unc-42; ceh-24 double mutants. The unc-42(+) SMD neurons had also been reported to express ceh-24 and require ceh-24 for their correct specification (Schwarz and Bringmann, 2017). However, using the cell identification and cell fate tool NeuroPAL, we found that ceh-24 is expressed in SMB, not SMD (Reilly et al., 2020), and, correspondingly, that in ceh-24 null mutants, SMB and not SMD differentiation is defective (Figure 13—figure supplement 1).

As summarized in Figure 14, we conclude that the specificity of UNC-42 activity is determined by cell-type-specific collaboration of UNC-42 with distinct cofactors. Such collaboration could, for example, be in the form of cooperative DNA binding, as previously observed for other Prd-type transcription factors (Wenick and Hobert, 2004), or could be in the context of forming a ‘transcription factor collective’ that operates additively to control target gene expression (Junion et al., 2012). Cohorts of collaborating terminal selectors have been termed ‘core regulatory complexes’ (Arendt et al., 2016).

There is much precedent in the literature for transcription factors operating in the context of feedforward loops in which a given transcription factor activates a downstream transcription factor to then cooperate with the downstream transcription factor to control entire batteries of effector genes. The first described example in the C. elegans literature is the UNC-86 POU homeodomain transcription factor, which activates the MEC-3 LIM homeodomain transcription factor to then co-regulate together with MEC-3 the expression of scores of touch neuron receptor genes (Duggan et al., 1998; Zhang et al., 2002). We therefore asked whether unc-42 controls the expression of the transcription factors that appear to cooperate with UNC-42 to control neuronal identity. We indeed find that the expression of ceh-24, cfi-1, and unc-3 reporter genes is strongly affected in unc-42 mutants (Figure 13—figure supplement 2). Moreover, fax-1 has previously been shown to be regulated by unc-42 (Wightman et al., 2005). In contrast, hlh-34, unc-62/Meis, and ceh-20/Pbx expression is unaffected in AVH and AIB, respectively. Together, this data lends support to the notion of the existence of positive feedforward loops in cell identity specification in a number of different contexts. More needs to be learned about these feedforward loops to understand why unc-42 activates these factors only in some, but not other neurons; possibly, the ability of unc-42 to control downstream transcription factors depends on the presence of cell-specific lineage cues that are present transiently when the respective neuron is born. A precedent for this is the activation of the terminal selector ceh-10 by its own cofactor, the terminal selector ttx-3, exclusively in the AIY interneuron, which requires transient Wnt signaling cues (Bertrand and Hobert, 2009).

Having defined the roles of UNC-42 in the neurons in which the protein is expressed, we circled back to our original observation that all the 15 distinct UNC-42(+) neuron classes are more highly interconnected than expected from any random set of 15 neuron classes. Intriguingly, several of the factors that collaborate with unc-42 in a neuron-type-specific manner are also expressed in synaptically connected neurons. This includes the ceh-24 NK2-type homeobox gene, which cooperates with unc-42 in the SAA and SIB neurons (summarized in Figure 14). In addition, ceh-24 is also expressed in the RME, SIA, and SMB neurons (Reilly et al., 2020). All five ceh-24 expression neurons are interconnected more heavily than expected by chance. This correlation is observed using the analysis as done by Arnatkeviciute et al., as well as the NDGE analysis that we described above (Supplementary file 1, Figure 1F). Similarly, the LIM homeobox gene lim-4 collaborates with unc-42 in some neurons, but is also expressed in additional sets of synaptically connected neurons (Supplementary file 1, Figure 1F). Lastly, the sites of expression of the COE-type transcription factor unc-3, an apparent cofactor for unc-42 in command interneurons (see above), are also significantly enriched for synaptically connected neurons based on the approach by Arnatkeviciute et al. and NDGE analysis. These UNC-3(+) neurons include the UNC-42(+) command interneurons, but also UNC-42(-) ventral cord motor neurons that are directly innervated by command interneurons. Taken together, one can imagine that the C. elegans connectome can be deconstructed into a series of overlapping groups of interconnected neurons whose interconnectivity is defined by what we propose to call ‘circuit organizer transcription factors’ (Figure 14). These factors operate as both terminal selectors to control molecular identify features of a neuron, such as its neurotransmitter identity, and may also organize neurons into synaptic circuits.

C. elegans strains used in this study are listed in Supplementary file 5. The wild type strain was Bristol N2. Worms were grown on nematode growth media (NGM) agar plates seeded with bacteria (OP50) as a food source.

All transgenes are referenced in Supplementary file 5. Contrary to a previous report (Cunningham et al., 2012), hlh-34 transgene reporters are expressed exclusively in AVH, not in AVJ. This cell identification was done with specific landmark strains and will be reported elsewhere. It is also consistent with recent scRNA data, which revealed hlh-34 expression exclusively in AVH (Taylor et al., 2021).

C. elegans were anesthetized using 100 mM sodium azide and placed on 5% agar pads on glass slides. All images were acquired using a Zeiss 880 laser-scanning confocal. Z-stack images (each ~0.5 μm thick) were acquired using the Zen software and analyzed using the Zen software or ImageJ. Representative images are shown following orthogonal maximum intensity projection of 2–25 z-stacks.

Reporter expression analysis was determined by confocal microscopy. Cell identification was done by assessing position and size using Nomarski optics and by crossing with neuronal landmark reporter strains eat-4 (otIs518, otIs388) (Serrano-Saiz et al., 2013), cho-1 (otIs544, otIs354) (Pereira et al., 2015), and NeuroPAL (otIs696) (Yemini et al., 2021). Expression in a subset of sensory neurons was confirmed by dye filling with DiD.

For quantification of axon length shown in Figure 8, confocal Z-stacks were opened using FIJI software and were loaded into the Simple Neurite Tracer plugin. Using this plugin, the axon emerging from the soma of AVH was traced across multiple Z-stacks and summed to calculate total axon length.

For quantification of RAB-3::GFP puncta shown in Figure 8—figure supplement 3, manual counting was performed using the ZEN software.

For quantification of cell nucleus positions, confocal Z-stacks were opened in the NeuroPAL ID software as described in Yemini et al., 2021. Cell nuclei were manually identified and aligned X, Y, and Z nucleus positions were assessed by measuring the distance from the point of origin, as determined by the NeuroPAL ID software. The X, Y, and Z nucleus positions of unc-42 mutant animals were evaluated individually and were not different as compared to wild type animals (data not shown). Additionally, a Euclidean distance was assessed using the distance formula, as shown in Figure 8—figure supplement 1.

unc-42(e270) was fixed as previously described (White et al., 1986). These fixed worms were then cut into 50 nm sections using RMC Powertome XL and collected onto grids. The nerve ring region of unc-42(e270) was then imaged either manually with a Phillips CM10 TEM or automatically with a JEOL 1400Plus TEM and the SerialEM software. Sections were then aligned and montaged, all of the axons in the nerve ring were serially traced, and synapses were annotated using the TrakEM2 software (Cardona et al., 2012). The region imaged, reconstructed, and annotated was ~15 μm in length and included 309 serial sections. Neurons were identified by characteristic synaptic and/or morphological features together with relative cell body position (Supplementary file 6). Synapse counts and axon adjacency counts were then extracted using scripts kindly provided by Christopher Brittin (Brittin et al., 2018). To compare to the unc-42(e270) synapse and axon adjacency counts to the previously described wild type (N2U) (Cook et al., 2019), the sections were aligned from the beginning of the RMEV neuron nucleus, a neuron that is easily identifiable based on morphology and position, to the anterior end of the nerve ring.

WormTracker assays were conducted and analyzed as previously described (Yemini et al., 2013). To avoid any potential variability due to room conditions, mutant and wild type strains were recorded simultaneously for each experiment. Briefly, individual L4 worms were placed on unseeded NGM plates. These worms were then tracked for 3 min with the WormTracker 2.0 (WT2) software, which tracks and records each worm with a camera.

For the categorical data shown in Figures 3–6, 8, 10, 11, and 13, statistical analysis was performed using Fisher’s exact test. For the numerical data shown in Figures 2, 7, and 8,, statistical analysis was performed using a one-way ANOVA followed by a post-hoc Tukey HSD test. Where appropriate, p-values were adjusted using a FDR correction for multiple testing.

The probability that a transcription factor is expressed in a set of neurons that are more interconnected than the whole connectome was calculated with a probability mass function using a binomial distribution. Analysis was performed as described (Arnatkeviciūtė et al., 2018) with modifications, using an updated transcription factor expression database. Connectivity data was taken from https://www.wormwiring.org and described in Cook et al., 2019.

The probability of having k success in n trials is given by the probability mass function:Pr(k;n,p)=Pr(X=k)=(nk)pk(1−p)n−k

We define the ‘probability of success’ (p) as the probability that any two given neurons are connected in the somatic (i.e., non-pharyngeal) C. elegans hermaphrodite connectome, excluding the pharyngeal neurons. This was calculated by examining all possible neuron pairs, excluding interclass pairs, where order matters (e.g., A–B is not the same as B–A), and totaling how many of these pairs were connected by either a chemical and/or electrical connection. Both electrical and chemical synapses were doubly counted (e.g., A > B and A < B were counted as two connections).

The ‘number of trials’ (n) was determined by totaling all possible pairs of neurons in which the transcription factor was expressed, excluding interclass pairs, where order matters.

The ‘number of successes’ (k) was determined for each transcription factor by totaling how many of these pairs were chemically and/or electrically connected.

A probability mass function calculation using a binomial distribution was then performed for each transcription factor, and a p-value was calculated. These p-values were then corrected for multiple testing using a FDR correction. Custom computation scripts are available at https://github.com/hobertlab/Berghoff_2021 (Berghoff and Hobert, 2021a, copy archived at swh:1:rev:2e64fea4812ce726f3e679dca3f691d3e866af43 Berghoff and Hobert, 2021b).

Transcription factor DNA binding motifs from the CISBP version 2.0 database (Weirauch et al., 2014) (unc-42: M03874_2.00,cfi-1:M01667_2.00,fax-1:M06432_2.00) and unc-3 (Kratsios et al., 2011) were used with TargetOrtho2.0 (Glenwinkel et al., 2014; Glenwinkel et al., unpublished) (FIMO p value threshold: 1e-4) to identify binding sites among orthologous coding gene loci in eight nematode species. Binding site enrichment tests were conducted using Python’s hypergeom function among C. elegans neuron class reporter genes. Reporter genes per neuron class are from the Hobert lab’s curated and recently updated collection of reporter genes (see Brain Atlas in Hobert et al., 2016). C. elegans coding gene annotations are from Wormbase version WS264. UNC-42 binding site enrichment tests: the expected proportion of binding sites is computed as the number of coding genes in the genome with at least one binding site divided by the total number of coding genes annotated. The observed proportion of binding sites is the number of neuron class-specific reporter genes that have at least one binding site in upstream intergenic or intronic regions. UNC-42 cofactor binding site enrichment tests: the expected proportion of coding genes with cofactor binding sites was computed by multiplying together the proportion of genes in the whole genome with at least one binding site match for both cofactors examined. The observed proportion is the number of neuron class reporter genes with at least one binding site from each cofactor.