Cell context-dependent CFI-1/ARID3 functions control neuronal terminal differentiation

SUMMARY

AT-rich interaction domain 3 (ARID3) transcription factors are expressed in the nervous system, but their mechanisms of action are largely unknown. Here, we provide, in vivo, a genome-wide binding map for CFI-1, the sole C. elegans ARID3 ortholog. We identify 6,396 protein-coding genes as putative direct targets of CFI-1, most of which encode neuronal terminal differentiation markers. In head sensory neurons, CFI-1 directly activates multiple terminal differentiation genes, thereby acting as a terminal selector. In motor neurons, however, CFI-1 acts as a direct repressor, continuously antagonizing three transcriptional activators. By focusing on the glr-4/GRIK4 glutamate receptor locus, we identify proximal CFI-1 binding sites and histone methyltransferase activity as necessary for glr-4 repression. Rescue assays reveal functional redundancy between core and extended DNA-binding ARID domains and a strict requirement for REKLES, the ARID3 oligomerization domain. Altogether, this study uncovers cell-context-dependent mechanisms through which a single ARID3 protein controls the terminal differentiation of distinct neuron types.

Graphical Abstract

In brief

Li et al. provide a genome-wide binding map for CFI-1, the sole C. elegans ortholog of the ARID3 family of transcription factors. Through biochemistry and molecular genetics, the study offers mechanistic insights into how the same ARID3 protein can control the molecular identity of distinct neuron types.

INTRODUCTION

Members of the AT-rich interaction domain (ARID) family of proteins are found in plants, yeast, fungi, and invertebrate and vertebrate animals.^1-4 ARID family proteins are expressed either ubiquitously or in a tissue-specific fashion and control various biological processes, such as cell proliferation, differentiation, and embryonic patterning.^3,4 Mutations in ARID family proteins are associated with cancer and several neurodevelopmental disorders.^5-10

Humans possess 15 ARID family proteins, divided into seven subfamilies (ARID1–ARID5, JARID1, and JARID2) based on the degree of sequence similarity.⁴ Orthologs to each subfamily are found in insect genomes, indicating the minimum age of ARID proteins.⁴ In the nematode C. elegans, orthologs are found for five subfamilies (LET-526/ARID1, SWSN-7/ARID2, CFI-1/ARID3, ARID-1/ARID4, and RBR-2/JARID1).⁴ The ARID, after which the family is named, was first identified in ARID3 proteins. These bind DNA in a sequence-specific manner, prefer AT-rich sequences, and are known to function as transcription factors.⁴ The ARID5 subfamily also encodes transcription factors,^2,3 but the remaining five subfamilies (ARID1, ARID2, ARID4, JARID1, and JARID2) encode proteins that bind DNA in a non-sequence-specific manner.² For example, ARID1A, ARID1B, and ARID2 constitute subunits of the SWI/SNF (BAF/PBAF) chromatin-remodeling complex that can move and/or eject nucleosomes. Although the precise functions of all ARID proteins are not known, accumulating evidence suggests that they can act either as positive or negative regulators of gene transcription or as components of chromatin-remodeling complexes.^1-4

Mouse Bright (Arid3a) and Drosophila dead ringer (retained) are the founding members of the ARID family and belong to the ARID3 subfamily, which is specific to metazoans. Single orthologs exist in C. elegans (CFI-1) and Drosophila (dead ringer), whereas mammals contain three orthologs (ARID3A–ARID3C) (Figure 1A).^1,4,11,12 Defining features of ARID3 proteins are (1) the extended ARID (eARID), an ~40-residue-long region next to the core ARID,^3,13 and (2) the REKLES domain at the C terminus. The ARID and eARID contact DNA,^14,15 whereas REKLES is a multifunctional domain required for oligomerization and protein-protein interactions.^16,17

Figure 1. Mapping genome-wide CFI-1 binding with ChIP-seq.

(A) Schematic of CFI-1, Drosophila Dead ringer (Dri), and mouse Arid3a-c. The ARID (yellow) and REKLES (blue) domain are shown.

(B) Diagram of the endogenous 3xflag::cfi-1 reporter allele. Bottom: dashed boxes highlight regions shown in (C)–(E).

(C–E) Expression of the 3×FLAG::CFI-1 fusion protein is confirmed by immunostaining (DAPI, blue; anti-FLAG, red) in motor neurons (C), muscles (arrowheads) and head neurons (D), and tail neurons (E). Scale bars, 4 μm.

(F) Fingerprint plot indicating localized strong enrichment of CFI-1 binding events in the genome. 50% of the maximum number of reads is contained in 86% of all genomic bins, indicating that 14% of the genome contains 50% of reads.

(G) Heatmap of the CFI-1 ChIP-seq signal around 1.0 kb of the center of binding peaks.

(H) Summary plot of the CFI-1 ChIP-seq signal; 95% confidence interval (gray area) around 3.0 kb of the TSS. Average signal peak is detected at ~140 bp upstream of the TSS.

(I) Pie chart summarizing genomic distribution of the CFI-1 ChIP-seq signal.

(J) Graph summarizing protein class ontology analysis (WormCat 2.0) of global CFI-1 target genes. 6,351 genes were analyzed, 4,678 of which have known protein class terms (1,673 genes were unassigned by WormCat 2.0). Asterisks to the left of each bar represent enriched categories (significantly overrepresented relative to the entire genome, Fisher’s exact test). Terminal differentiation genes, 2,947 of the 4,678 (63%); Gene expression, 806 of the 4,678 (17.2%).

(K) Venn diagram showing that 77.1% of protein-coding genes (4,931 of 6,396) are bound by CFI-1 and expressed in cfi-1+ neurons based on RNA-seq (CenGEN data).

ARID3 proteins have several early developmental roles, identified by genetic studies. Mice lacking Bright/Arid3a display early embryonic lethality because of defects in hematopoiesis.¹⁸ Bright/Arid3a is best studied in B cell lineages, where it acts as an activator and increases immunoglobulin transcription.^18-21 However, Bright/Arid3a is also critical for embryonic stem cell differentiation.^22-24 In this context, it can either activate or repress gene expression.²⁴ Like mice lacking Bright/Arid3a, null mutants for dead ringer in Drosophila display early lethality.^25,26 Dead ringer is essential for anterior-posterior patterning and muscle development in the fly embryo; it can act either as an activator or repressor of gene transcription.^25,27,28 Last, Arid3a and Arid3b have been associated with tumorigenesis by acting as direct inducers of cell cycle regulators.^29,30

Genetic studies in Drosophila and C. elegans have also identified late developmental roles for ARID3 proteins in the nervous system. In Drosophila larvae, dead ringer is expressed in distinct neuron types and controls axonal pathfinding,^31-33 although its downstream targets in neurons remain unknown. In C. elegans, cfi-1 is selectively expressed in head muscle and several neuron types: IL2 sensory neurons; AVD, PVC, and LUA interneurons; and various classes of motor neurons.^12,34 Because C. elegans animals that lack cfi-1 are viable,¹² these mutant strains provided a glance into the potential functions of ARID3 proteins during post-embryonic life. Candidate approaches that examined a handful of effector genes encoding neurotransmitter (NT) biosynthesis proteins and receptors suggested that CFI-1 acts as an activator of gene expression in sensory neurons (IL2) and interneurons (AVD and PVC).^12,35-37 However, in ventral nerve cord motor neurons, CFI-1 is thought to act as a repressor of the glutamate receptor gene glr-4/GRIK4.³⁸ The molecular mechanisms underlying the differential activities of CFI-1 (activator versus repressor) in these distinct neuron types remain unknown. Elucidating such mechanisms in C. elegans may provide clues as to how CFI-1 orthologs in other species control cell differentiation. Last, unbiased approaches to identify the in vivo targets of CFI-1 (and any other ARID3 protein) in the nervous system are currently lacking, preventing a comprehensive understanding of the neuronal functions controlled by ARID3 transcription factors.

Here, we performed chromatin immunoprecipitation and sequencing (ChIP-seq) for CFI-1. By generating an in vivo binding map on the C. elegans genome, we identified 6,396 protein-coding genes as putative direct targets of CFI-1, the majority of which (77%) are expressed in post-mitotic neurons. Gene Ontology analysis suggests that CFI-1 is primarily involved in the process of neuronal terminal differentiation. To gain mechanistic insight into how CFI-1 controls the terminal differentiation of different neuron types, we focused on head sensory neurons (IL2 class) and nerve cord motor neurons (DA, DB, VA, and VB classes). In sensory IL2 neurons, CFI-1 exerts a dual role: it acts directly to activate and indirectly to repress distinct terminal differentiation genes (e.g., NT receptors and ion channels). In nerve cord motor neurons, however, CFI-1 acts directly to repress expression of the glutamate receptor gene glr-4/GRIK4. CRISPR-Cas9-mediated mutagenesis of endogenous CFI-1 binding sites suggests that proximal binding to the glr-4 locus is necessary for repression. Importantly, histone methyltransferase activity and REKLES, an ARID3-specific oligomerization domain, are required for glr-4 repression. Altogether, our study offers mechanistic insights into cell-context-dependent functions of CFI-1 (ARID3), a critical regulator of the terminal differentiation program of distinct neuron types.

RESULTS

A map of CFI-1/ARID3 binding on the C. elegans genome

To identify CFI-1 binding events, we first generated an endogenous reporter strain through in-frame insertion of the FLAG epitope sequence (3xFLAG) immediately after the cfi-1 start codon (Figure 1B). Immunostaining against FLAG on adult 3xflag::cfi-1 animals showed nuclear expression in head muscle cells as well as in neurons of the head, ventral nerve cord, and tail regions (Figures 1C-1E), indicating that this reporter allele faithfully recapitulates the known expression pattern of cfi-1.^12,38 Unlike cfi-1 null animals,¹² homozygous 3xflag::cfi-1 animals do not display any defects in posterior touch response (Figure S1). This suggests that insertion of the 3xFLAG sequence does not alter cfi-1 gene function. We therefore conducted ChIP-seq using a FLAG antibody on homozygous 3xflag::cfi-1 animals at the third larval stage (L3) because all cfi-1-expressing cells are generated by this stage.

Our ChIP-seq experiment revealed strong enrichment of CFI-1 binding in the C. elegans genome, identifying 14,806 unique binding peaks (q value cutoff, 0.05) (Figures 1F and 1G). The CFI-1 peaks are predominantly located between 0 and 2 kb upstream of transcription start sites (TSSs); 86.42% of CFI-1 binding occurs between 0 and 2 kb from TSSs (Figures 1H and 1I). The remaining binding events occur between 2 and 3 kb from TSSs (6.49%), at distal intergenic regions (4.07%), and at introns (1.65%) (Figures 1H and 1I). Thus, CFI-1 appears to act primarily at proximal regions (0–2 kb from TSSs) to regulate gene expression. By comparison, a previous in vitro study found that overexpressed Arid3a in mouse embryonic stem cells binds to proximal (56% of binding) and distal (42% of binding) cis-regulatory elements.²⁴ Last, de novo motif analysis of the 14,806 CFI-1 binding peaks identified DNA motifs for dozens of transcription factors, which constitute potential CFI-1 collaborators in different cell types (Figure S2).

The majority of CFI-1/ARID3 target genes encode neuronal terminal differentiation markers

ChIP-seq for CFI-1 generated an in vivo binding map of an endogenously tagged ARID3 protein. To identify the biological processes controlled by CFI-1, we subsequently conducted bio-informatic analysis of the 14,806 CFI-1 binding peaks and identified 6,396 protein-coding genes as putative CFI-1 targets (STAR Methods). Because the majority of cfi-1-expressing cells are neurons (Figures 1C-1E),^12,38 we reasoned that a significant portion of the 6,396 protein-coding genes may be expressed in the nervous system. To test this, we used available single-cell expression profiles (CeNGEN project: www.cengen.org) for all known cfi-1-expressing neurons (IL2, URA, AVD, PVC, LUA, DA, DB, VA, VB, DD, VD) and indeed found that 77.1% of the global CFI-1 targets (4,931 of 6,396) are expressed in these neurons (Figures 1J and 1K; Data S1). To gain insights into the biological functions of CFI-1, we conducted Gene Ontology (GO) analysis with WormCat.³⁹ Strikingly, the majority of CFI-1 putative target genes (63%) encode proteins essential for neuronal terminal differentiation (e.g., NT receptors, transporters, ion channels, transmembrane receptors, and cell adhesion molecules) (Figures 1J and 1K; Data S2). The second largest category (17.2% of CFI-1 targets) contains transcription factors, chromatin factors, as well as proteins involved in DNA/RNA metabolism (Figures 1J and 1K), suggesting that CFI-1 can affect gene expression indirectly through these factors. Altogether, the downstream targets identified by this unbiased approach suggest that CFI-1 likely plays a prominent role in neuronal terminal differentiation.

CFI-1/ARID3 acts directly to activate terminal differentiation genes in IL2 sensory neurons

Although cfi-1 is expressed in several neuron types, a handful of CFI-1 target genes have only been identified in the IL2 class, which contains 6 head sensory neurons (Figure 2A).^12,36 In these cells, genetic experiments suggested that CFI-1 influences gene expression positively and negatively.^12,36 In IL2, CFI-1 is known to activate five terminal differentiation genes (cho-1/ChT, unc-17/VAChT, gcy-19 [receptor-type guanylate cyclase], klp-6 [kinesin-like protein], and unc-5 [netrin receptor]) and repress two ion-channel-encoding genes (pkd-2/Polycystin-2 like 1 [PKD2L1] and lov-1/Polycystin-1 like 3 [PKD1L3]), both associated with polycystic kidney disease.⁴⁰ It has remained unknown, however, whether CFI-1 acts directly or indirectly. Using ChIP-seq, we found that CFI-1 binds directly to all known terminal differentiation genes (e.g., cho-1/ChT and unc-17/VAChT) that require cfi-1 gene activity for their activation in IL2 neurons (Figure 2B). However, we did not detect any binding in the cis-regulatory regions of genes repressed by cfi-1 (pkd-2 and lov-1) (Figure 2C). Such a lack of binding may be due to low sensitivity. However, our ChIP-seq appears to be sensitive enough to detect CFI-1 binding in the cis-regulatory region of genes that are activated by CFI-1 and also exclusively expressed in the six IL2 neurons (Figure S3A). Altogether, biochemical evidence (ChIP-seq) combined with genetic studies^12,36 suggests that, in IL2 sensory neurons, CFI-1 exerts a dual role: it directly activates a set of terminal differentiation genes but indirectly (likely via intermediary factors) represses expression of ion-channel-encoding genes (pkd-2/Polycystin-2 and lov-1/Polycystin-1 like 3) (Figures 2B and 2C).

Figure 2. CFI-1 directly activates terminal differentiation genes in IL2 sensory neurons.

(A) Location of IL2 head sensory neurons.

(B and C) IGV snapshots showing CFI-1 binding at loci of genes known to be activated (B) and repressed (C) by CFI-1 in IL2 neurons.

(D) Expression analysis of five fluorescent markers of IL2 terminal differentiation (cil-7, cwp-4, ddn-3, degl-2, and tba-6) in WT and cfi-1 (ot786) mutant animals at L4. N = 10–15. Quantification of the number of IL2 neurons and fluorescence intensity is provided on the right. Scale bars, 4 μm. Two-tailed t test was performed. *p < 0.05, **p < 0.01, ***p < 0.001. CFI-1 binding at cil-7, cwp-4, ddn-3, degl-2, and tba-6 is shown below the microscopy images.

(E) Model summarizing targets of CFI-1 in IL2 neurons.

Although seven targets of CFI-1 (cho-1, unc-17, gcy-19, klp-6, unc-5, pkd-2, and lov-1) are known to date in IL2 neurons, it remains unclear whether CFI-1 broadly affects the IL2 terminal differentiation program. We therefore tested six additional markers of IL2 terminal differentiation (sodium channels: degl-1 and degl-2; cilium proteins: cil-7, cwp-4, and ddn-3; tubulin: tba-6) for cfi-1 dependency.^13,37 Expression of five markers (cil-7, cwp-4, ddn-3, degl-2, and tba-6) is significantly reduced in IL2 neurons of cfi-1 mutants (Figures 2D and S3). ChIP-seq showed CFI-1 binding in four of these five genes (Figure 2D). Altogether, our findings support the idea of CFI-1 acting directly to activate expression of multiple IL2 terminal differentiation genes (Figure 2E).

We next undertook an unbiased approach to investigate whether genes expressed in mature IL2 neurons are also bound by CFI-1. To this end, we used available single-cell expression profiles (CeNGEN project: www.cengen.org) and selected the most highly expressed genes (top 1,000) in IL2 neurons. Strikingly, 701 of these genes (70.1%) are bound by CFI-1 (Figure S4A). Among these 1,000, GO analysis uncovered 489 terminal differentiation genes, 74.4% of which are bound by CFI-1 (Figure S4A; Data S3). As a negative control, we conducted the same analysis for 4 neuron types that do not normally express cfi-1 and detected a lower percentage of CFI-1 binding (Figure S4). This analysis provides biochemical evidence to support the idea that CFI-1 acts directly to broadly control the terminal differentiation program of IL2 neurons.

Given its expression in multiple neuron types, how does CFI-1 specifically control the terminal differentiation of IL2 neurons? In these cells, CFI-1 cooperates with the POU homeodomain transcription factor UNC-86 to activate expression of a handful of terminal differentiation genes,³⁶ but whether such cooperation extends broadly to other IL2-expressed genes remains unclear. Therefore, we bioinformatically searched for the presence of the UNC-86 binding site in the 701 IL2-expressed genes bound by CFI-1. In these 701 genes, there are 1,312 CFI-1 binding peaks, of which 877 (66.8%) contain at least one UNC-86 binding site (p < 0.001) (Figure S4B). It is therefore likely that CFI-1 and UNC-86 cooperate and act directly to broadly affect gene expression in IL2 neurons.

CFI-1/ARID3 acts directly to repress endogenous glr-4/GRIK4 expression in cholinergic motor neurons

Our findings in IL2 sensory neurons suggest that CFI-1 is a direct activator and indirect repressor of gene expression (Figure 2E). Next, we interrogated the function of CFI-1 in cholinergic nerve cord motor neurons that control locomotion (Figure 3A). Using an endogenous reporter allele,³⁴ we found that cfi-1 is selectively expressed in 29 cholinergic motor neurons (DA, DB, VA, and VB classes) located in the mid-body region (Figure 3A). Next, we asked whether, in these neurons, CFI-1 functions as an activator of gene expression. To this end, we examined five available motor neuron-specific terminal differentiation markers (twk-40 and twk-43 [TWiK potassium channels], ncs-2 [neuronal calcium sensor], npr-29 [neuropeptide], and dbl-1 [Bmp-like]) for cfi-1 dependency.⁴¹ These showed no difference in expression in motor neurons of cfi-1(−) mutants (Figure S5). Consistent with this, a previous study also found that three other terminal differentiation markers (acr-5 [acetylcholine receptor], del-1 [SCNN1 sodium channel], and inx-12 [gap junction protein]) are not affected in motor neurons of cfi-1(−) mutants.³⁸ Interestingly, our ChIP-seq data indicate that all of these eight genes, as well as ~75% of DA, DB, VA, and VB terminal differentiation genes (Figure S4), are bound by CFI-1 (Figure S5), raising the possibility of CFI-1 operating redundantly with other transcription factors to activate gene expression in motor neurons.

Figure 3. CFI-1/Arid3a represses glr-4/GRIK4 in ventral cord motor neurons.

(A) Summary of endogenous expression of cfi-1 and glr-4 in cholinergic motor neuron subtypes of the nerve cord (DA, DB, VA, VB, and AS) and retrovesicular ganglion (SAB). Expression patterns were determined by colocalization with neuron-subtype-specific reporters. See also Figure S6. Colored boxes represent expression; white boxes indicate no detectable expression.

(B) CFI-1 binding signal on the glr-4 locus and design of the endogenous glr-4 reporter allele (2xNLS::mScarlet::SL2::glr-4).

(C) Representative micrographs showing expression of 2×NLS:mScarlet:SL2:GLR-4 in SAB neurons of WT and cfi-1(−) animals. The 3 arrowheads point to the cell body of each SAB neuron. In cfi-1 (−) mutants, one SAB neuron is not located on the same focal plane with the other 2 SAB neurons. Hence, only 2 SAB neurons are shown. A bright white signal to the left of the SAB neurons indicates glr-4 expression in head neurons.

(D) 2xNLS::mScarlet::SL2::glr-4 expression in ventral cord motor neurons (white arrowheads) in WT and cfi-1 (−) animals (L4). All MNs that express this reporter were counted. Ectopic expression of glr-4 was detected in cfi-1(−) mutants. Bottom: each dot in the graph of total number of MNs (left) represents an individual animal. Each dot in the fluorescence intensity quantification graph (right) represents an individual MN with glr-4 expression. Only MNs in the anterior nerve cord are included in the quantification. Two-tailed t test was performed. *p < 0.05, **p < 0.01, ***p < 0.001. N ≥ 15.

(E) RNA-seq data (CeNGEN) showing expression of glr-4 in SAB, DB, VB, AS, VA, and DA motor neurons.

(F) Single-molecule (sm) mRNA fluorescence in situ hybridization (FISH) for glr-4 in MNs of WT and cfi-1(−) animals (L1 stage). Left: images showing glr-4 mRNA (Cy5, red) in a motor neuron in cfi-1(−) mutants (nucleus: DAPI, blue). Right: quantification of the number of glr-4 transcripts in MNs of WT and cfi-1(−) animals. Two-tailed t test was performed. *p < 0.05, **p < 0.01, ***p < 0.001. Error bars, SEM. N ≥ 18.

(G) RT-PCR for glr-4 in WT and cfi-1(−) mutants (whole-worm lysates). Error bars, SEM.

(H) Model summarizing glr-4 regulation by CFI-1.

Scale bars, 2 μm.

The terminal differentiation gene glr-4 (ortholog of human GRIK4 [glutamate inotropic receptor kainate type subunit 4]) is the only known CFI-1 target in motor neurons, where it is negatively regulated by CFI-1,³⁸ providing an opportunity to obtain mechanistic insights into how ARID3 proteins mediate gene repression in the nervous system. We therefore carried out an in-depth investigation focused on glr-4.

Because previous studies employed transgenic reporters,^38,42 the endogenous expression pattern of glr-4 in C. elegans neurons remained unclear. We therefore generated an endogenous glr-4 reporter allele by inserting the 2xNLS::mScarlet::SL2 cassette immediately after the start codon (STARMethods) and established the glr-4 expression pattern in motor neurons with single-cell resolution (Figures 3A and S6). Consistent with previous studies,^38,42 we observed high levels of glr-4 (mScarlet) expression in head neurons as well as in SAB motor neurons that innervate head muscle (Figures 3A and 3C). This endogenous reporter also revealed new sites of expression. In the nerve cord of wild-type (WT) animals, we detected low levels of glr-4 (mScarlet) expression in 17 of the 29 cfi-1-expressing motor neurons as well as in AS motor neurons, which do not express cfi-1 (Figures 3A and S6). Further, the observed levels of glr-4 expression in head (SAB) and ventral cord motor neurons were independently confirmed by available single-cell RNA sequencing (scRNA-seq) data (CeNGEN project: www.cengen.org) (Figure 3E).

Next, we tested whether endogenous glr-4 expression in motor neurons depends on cfi-1 gene activity. Indeed, glr-4 (mScarlet) expression is increased in motor neurons of cfi-1 loss-of-function mutants because we observed a higher number of cells expressing mScarlet, and at higher levels, compared with WT motor neurons (Figure 3D). No effects on glr-4 (mScarlet) were observed in SAB neurons because they do not express cfi-1 (arrowheads in Figure 3C). Moreover, we performed single-molecule (sm) mRNA fluorescence in situ hybridization (FISH) in WT and cfi-1 mutant animals. Loss of cfi-1 led to increased levels of glr-4 mRNA in nerve cord motor neurons (Figure 3F). Last, these results were corroborated by RT-PCR in WT and cfi-1 mutant animals (Figure 3G). Altogether, we conclude that CFI-1 limits the endogenous expression of glr-4/GRIK4 in nerve cord motor neurons. In WT animals, we can either detect low or no glr-4 expression in motor neurons, whereas loss of cfi-1 results in robust glr-4 expression in these cells (Figure 3H). Because ChIP-seq revealed extensive CFI-1 binding immediately upstream of the glr-4 locus (Figure 3B), we propose that CFI-1 acts as a direct repressor of glr-4/GRIK4.

Cfi-1/ARID3 is required to maintain glr-4 repression in nerve cord motor neurons

The cfi-1-expressing motor neurons (DA, DB, VA, and VB) are generated in two waves (DA/DB are born during embryogenesis; VA and VB at larval stage 1 [L1]) (Figures 4A and 4B). Although all cfi-1-expressing motor neurons have been generated by larval stage 2 [L2], we found no glr-4 (mScarlet) expression in WT animals at this stage. However, we observed a progressive increase in the number of WT motor neurons expressing low levels of glr-4 (mScarlet) at subsequent stages (L3, L4, and day 2 adult), indicating a correlation between glr-4 expression and motor neuron maturation (Figure 4A).

Figure 4. CFI-1 is sufficient and continuously required to repress glr-4.

(A) Quantification of the number of MNs expressing glr-4 in WT and cfi-1 (−) at four developmental stages: L2, L3, L4, and day 2 adults. Two-tailed t test was performed. *p < 0.05, **p < 0.01, ***p < 0.001. N ≥ 12.

(B) Diagram showing the timeline of the auxin experiment.

(C) Quantification of the endogenous glr-4 reporter in the ethanol group (control) and auxin group on kas16[mNG::AID::cfi-1]; otTi28[unc-11prom8+ehs-1prom7+rgef-1prom2::TIR1::mTurquoise2::unc-54 3′UTR] animals. otTi28 drives expression of TIR1 specifically in neurons. Two-tailed t test was performed. *p < 0.05, **p < 0.01, ***p < 0.001. N = 15.

(D) Representative images showing loss of glr-4::tagRFP expression in SAB neurons (arrowheads) upon overexpression (OE) of cfi-1 with an SAB-specific promoter (unc-4). Right: quantification of the number of SAB neurons expressing glr-4::tagRFP. Two-tailed t test was performed. *p < 0.05, **p < 0.01, ***p < 0.001. N ≥ 14. Errorbars, SEM. Scale bar, 4 μm.

(E) Model summarizing CFI-1 sufficiency in SAB neurons.

To test when cfi-1 gene activity is required for repression, we monitored endogenous glr-4 (mScarlet) expression in motor neurons of WT and cfi-1 mutant animals at larval (L2, L3, and L4) and adult (day 2) stages. Compared with controls, we identified a statistically significant increase in the number of glr-4 (mScarlet)-expressing motor neurons at L3, L4, and adult (day 2) stage (Figure 4A). Next, we used a conditional cfi-1 allele (mNG::3xFLAG::AID::cfi-1) that enables temporally controlled CFI-1 protein depletion upon administration of the plant hormone auxin.^34,43 Depletion of CFI-1 during the first 2 days of adulthood led to a significant increase in the number of motor neurons expressing glr-4 (mScarlet) (Figures 4B and 4C), suggesting that cfi-1 is continuously required to maintain glr-4/GRIK4 repression in the adult (Discussion).

cfi-1/ARID3 is sufficient to repress glr-4 expression

To test whether cfi-1 is sufficient to repress glr-4, we ectopically expressed cfi-1 in the SAB neurons. Using an SAB-specific-promoter (unc-4) to drive cfi-1, we observed a significant decrease in the number of SAB neurons expressing a glr-4 reporter gene (Figures 4D and 4E), indicating that cfi-1 is not only necessary (Figure 3) but also sufficient to repress glr-4 expression.

CFI-1 binding sites proximal to glr-4 are necessary for repression in motor neurons

CFI-1 binds to proximal and distal cis-regulatory elements upstream of glr-4 (Figure 5A). To precisely identify the elements through which CFI-1 mediates repression, we conducted a cis-regulatory analysis in the context of transgenic reporter animals. When tagRFP was driven by distal regulatory elements (2.23 kb or 938 bp), we did not observe differences in the number of tagRFP-expressing motor neurons between WT and cfi-1 mutants (Figure 5B). However, we found an increase in the number of tagRFP-expressing motor neurons in cfi-1 mutants when tagRFP was driven by a proximal 538-bp element (Figure 5B), suggesting that this element contains sequences necessary for CFI-1 repression. A longer 3.7-kb element and a translational reporter (GLR-4::GFP) driven by a 4.9-kb element yielded similar results (Figure 5B). Interestingly, a shorter tagRFP reporter (3.14 kb) that specifically lacks the region with the most proximal CFI-1 binding peak did not show an increase in the number of tagRFP-expressing motor neurons in cfi-1 mutants (Figure 5B), suggesting that proximal CFI-1 binding to the glr-4 locus is needed for repression.

Figure 5. CFI-1 directly represses glr-4 by binding to its promoter via a conserved binding motif.

(A) CFI-1 ChIP-seq tracks on the glr-4 locus. A series of glr-4 transgenic reporters is depicted. Right: the CFI-1 DNA binding motif. Using bioinformatics analysis, 13 CFI-1 binding motifs were identified upstream of glr-4 that overlap with CFI-1 binding peaks.

(B) Quantification of MNs expressing glr-4 reporters in WT and cfi-1(−) animals. Two-tailed t test was performed. *p < 0.05, **p < 0.01, ***p < 0.001. All reporters were analyzed in the adult (day 2), except for the 538 bp and 4.9 kb reporters (analyzed at L4). N ≥ 13. Error bars, SEM.

(C) Schematic of the 2xNLS::mScarlet::SL2::glr-4^{11CFI–1 sites MUT} allele. Point mutations were introduced to all 11 CFI-1 binding motifs at the glr-4 promoter.

(D) Quantification of MNs expressing 2xNLS::mScarlet::SL2::glr-4 and 2xNLS::mScarlet::SL2::glr-4^{11CFI–1 sites MUT} alleles in day 2 adults. *p < 0.05, **p < 0.01, ***p < 0.001.

We next sought to determine whether proximal CFI-1 binding sites are required for glr-4 repression. The CFI-1 site (NNATHDNN) has been determined previously in vitro through protein binding microarrays (Figure 5A).⁴⁴ Within the most proximal region of glr-4, we identified 11 predicted CFI-1 binding sites (Figure 5A). To test their functionality, we introduced nucleotide substitutions to all 11 sites in the context of the endogenous glr-4 (mScarlet) reporter through CRISPR-Cas9 genome editing (Figure 5C). This manipulation nearly phenocopied the cfi-1 null mutant phenotype; it led to a dramatic increase in the number of mScarlet-expressing motor neurons (Figures 5C and 5D). We conclude that CFI-1 binding sites located in the most proximal region of glr-4 (likely within 538 bp from ATG) are necessary for its repression in motor neurons.

The transcription factor UNC-3 (Collier/Ebf) and two Hox proteins (LIN-39 and MAB-5) activate basal levels of glr-4/GRIK4 expression in motor neurons

Because glr-4 is expressed at basal levels in cholinergic motor neurons of WT animals (Figure 3), we reasoned that this occurs because of CFI-1 antagonizing the function of glr-4 activators in these neurons. The transcription factor UNC-3 (Collier/Ebf) and the Hox proteins LIN-39 (Scr/Dfd/Hox4-5) and MAB-5 (Antp/Hox6-8) are known to act as transcriptional activators in cholinergic motor neurons^38,45 (Figure 6A), leading us to hypothesize that they can also activate glr-4 expression. Indeed, we found that expression of the endogenous glr-4 (mScarlet) reporter is reduced in unc-3 loss-of-function mutant animals (Figure 6B). Importantly, the decrease of glr-4 expression was significantly exacerbated in unc-3; lin-39; mab-5 triple mutants compared with unc-3 single mutants, indicating that these three factors cooperate to activate glr-4 in cholinergic motor neurons (Figure 6B). Interestingly, the Hox requirement is only revealed in the absence of unc-3 gene activity because glr-4 expression appears normal in lin-39; mab-5 double mutants.

Figure 6. CFI-1 represses glr-4 by counteracting three terminal selectors.

(A) Summary of the endogenous expression of cfi-1/Arid3, glr-4/GRIK4, unc-3/COE, lin-39 /HOX, and mab-5/HOX in five cholinergic motor neuron subtypes and SAB neurons. Colored boxes represent expression, while white boxes indicate no detectable expression. Only MNs located in the nerve cord are shown for DA, DB, VA, VB, and AS subtypes.

(B) Quantification of 2xNLS::mScarlet::SL2::glr-4 expression in WT, cfi-1(−), lin-39(−); mab-5(−) double-mutant, and unc-3(−); lin-39(−); mab-5(−) triple-mutant animals (L4). Two-tailed t test. *p < 0.05, **p < 0.01, ***p < 0.001. N = 15. Error bars, SEM.

(C) ChIP-seq binding peaks for UNC-3 (in WT and cfi-1(−)), CFI-1, LIN-39, and MAB-5 at the glr-4 locus. Point mutations were introduced to the proximal COE motif (UNC-3 binding site), resulting in the 2xNLS::mScarlet::SL2::glr-4^{COE1 MUT} allele.

(D) Quantification of the 2xNLS::mScarlet::SL2::glr-4 reporter in WT and unc-3(−) animals and of the 2xNLS::mScarlet::SL2::glr-4^{COE1 MUT} reporter in day 2 adults. Two-tailed t test was performed. *p < 0.05, **p < 0.01, ***p < 0.001. N ≥ 13.

(E) Quantification of the 2xNLS::mScarlet::SL2::glr-4 reporter in WT, cfi-1(−), unc-3(−), and cfi-1(−); unc-3(−) double mutants and of the 2xNLS::mScarlet::SL2::glr-4^{COE1 MUT} allele in WT and cfi-1(−) animals (day 2 adults). Two-tailed t test was performed. *p < 0.05, **p < 0.01, ***p < 0.001. N ≥ 13.

(F and G) Representative images (F) and fluorescence intensity quantification (G) of the 2xNLS::mScarlet::SL2::glr-4 reporter in WT, cfi-1(−), met-2(−), met-2(−); set-25(−), and cfi-1(−); met-2(−);set-25(−) animals at L4. Scale bars, 10 mm. In (F), a white dashed line indicates the C. elegans gut (autofluorescent). White arrowheads, MN nuclei expressing mScarlet. Two-tailed t test. *p < 0.05, **p < 0.01, ***p < 0.001. N ≥ 15.

The most proximal UNC-3 biding site is necessary for glr-4 expression in motor neurons

Interrogation of available ChIP-seq data for UNC-3, LIN-39, and MAB-5 showed overlapping binding upstream of glr-4 (Figure 6C), suggesting a direct mode of activation by these factors. To functionally test this notion, we focused on UNC-3 because its binding site (termed the COE motif) is well defined in the C. elegans genome.^34,46 Through a bioinformatics search (STAR Methods), we found two COE motifs, one proximal and one distal to the glr-4 locus (Figure 6C). Using CRISPR-Cas9 genome editing, we introduced nucleotide substitutions to the proximal COE motif (COE1) in the context of the endogenous glr-4 (mScarlet) reporter (Figure 6C). Animals carrying this glr-4 (mScarlet)^{COE1 MUT} reporter allele showed a significant reduction in the number of mScarlet-expressing motor neurons, reminiscent of the effect seen in unc-3 (−) null mutants (Figure 6D). These data indicate that, in WT animals, the most proximal COE motif is necessary for basal glr-4 expression in motor neuros.

CFI-1 antagonizes the ability of UNC-3 to activate glr-4 expression in motor neurons

Because UNC-3 activates basal glr-4 expression in motor neurons of WT animals, we wondered whether it also controls the increased levels of glr-4 expression observed in cfi-1 mutants. We found this to be the case through double mutant analysis. The number of mScarlet-expressing cells in cfi-1; unc-3 mutants is dramatically decreased compared with cfi-1 single mutants (Figure 6E). These data indicate that CFI-1 antagonizes the ability of UNC-3 to activate glr-4 (mScarlet) expression in motor neurons.

Because the proximal UNC-3 binding site (COE1 motif) is required to activate glr-4 expression in WT motor neurons (Figure 6D), this site may also be necessary for the increased expression of glr-4 in motor neurons of cfi-1 mutants. Indeed, we observed a significant reduction in the number of mScarlet-expressing cells in cfi-1 mutants carrying the glr-4 (mScarlet)^{COE1 MUT} reporter compared with the intact version of the glr-4 (mScarlet) reporter (Figure 6E). However, this reduction was not as strong as seen in cfi-1; unc-3 double mutants, suggesting that the second COE motif (COE2) is likely used by UNC-3 to drive glr-4 expression in cfi-1 mutants (Figures 6C and 6D).

Last, we hypothesized that the extensive binding of CFI-1 (four binding peaks identified by ChIP-seq) immediately upstream of the glr-4 locus may limit the ability of UNC-3 to access the locus (Figure 6C), resulting in basal levels of glr-4 expression in WT motor neurons. If this were the case, then UNC-3 binding would increase on the glr-4 locus of cfi-1 mutants to drive higher levels of glr-4 expression. To test this, we performed ChIP-seq for UNC-3 in WT and cfi-1 null mutant animals (at the L3 stage). We found that UNC-3 binding on the glr-4 locus remains largely unaltered upon cfi-1 loss (Figure 6C). Although our ChIP-seq lacks motor neuron specificity (whole-animal lysates were used), these data do raise the possibility that UNC-3 can access the glr-4 locus in the presence and absence of CFI-1.

Histone methyltransferase activity is required for glr-4 repression in motor neurons

To gain mechanistic insights, we sought to identify histone modifications necessary for glr-4 repression in nerve cord motor neurons. We initially focused on MET-2 (SETDB1 ortholog), a histone methyltransferase responsible for mono- and dimethylation of lysine 9 of histone 3 (H3K9),^47,48 because a previous study implicated met-2 in gene repression in egg laying motor neurons,⁴⁹ which do not express cfi-1 (Figure S6). Similar to cfi-1 mutants, we observed increased expression of glr-4 in cholinergic motor neurons of met-2 mutant animals using transgenic (Figure S7) and endogenous reporters (Figures 6F and 6G), providing genetic evidence that mono- and dimethylation of H3K9 is needed for glr-4 repression. Through cell-type-specific rescue, we found that met-2 acts cell autonomously in nerve cord motor neurons to repress glr-4 (Figure S7). In addition to mono- and dimethylation, H3K9 trimethylation is also implicated in glr-4 repression. By testing animals lacking gene activity of set-25 (SUV39H2 ortholog), a histone methyltransferase that specifically mediates H3K9 trimethylation,⁴⁸ we found that the effect on glr-4 expression was significantly exacerbated in met-2(−); set-25(−) double mutants compared with met-2(−) animals (Figures 6F, 6G, and S7). Altogether, our genetic analysis indicates that H3K9 methylation, the epigenetic hallmark of heterochromatin,⁵⁰ is necessary for glr-4 repression in nerve cord motor neurons.

Next, we tested whether cfi-1, met-2, and set-25 operate in the same genetic pathway or act in parallel pathways to repress glr-4 by generating cfi-1(−); met-2(−); set-25(−) triple mutants. We found no evidence for an additive effect; the increase in glr-4 expression in these animals is comparable with the increase observed in met-2(−) single and met-2(−); set-25(−) double mutants (Figures 6F and 6G), supporting the idea that cfi-1, met-2, and set-25 potentially operate in the same genetic pathway.

The core ARID and REKLES domains of CFI-1 are required for glr-4 repression

To gain molecular insights into ARID3-mediated gene repression, we deleted portions of the CFI-1 protein and then conducted rescue assays in cfi-1 mutant animals to assess glr-4 expression. ARID3 proteins are defined by the eARID, an ~40-residue-long highly conserved domain immediately following the core ARID (Figure 7A). Because structural studies on Dead ringer showed that eARID contacts DNA,^2,14,15 we tested whether the CFI-1 eARID is required for glr-4 repression. Transgenic expression of either WT CFI-1 or CFI-1 lacking the eARID (ΔeARID) in motor neurons of cfi-1-null mutant animals led to complete rescue. That is, glr-4 expression was no longer observed in motor neurons when either WT or ΔeARID CFI-1 was provided (p = 0.37) (Figure 7B), indicating that the eARID is dispensable for CFI-1-mediated gene repression. Next, we mutated the helix-turn-helix (HTH) domain within the core ARID region because the HTH domain of Dead ringer contacts the major groove of DNA.^2,14,15 Transgenic expression of CFI-1 lacking the HTH domain (ΔHTH) in motor neurons of cfi-1 mutants led to significant repression of glr-4, but this effect is not identical to WT CFI-1 (p = 0.05) (Figure 7B), arguing for a minor role of the HTH domain in glr-4 repression. Similarly, transgenic expression of CFI-1 lacking the entire core ARID (ΔARID) (including the HTH domain) in motor neurons of cfi-1 mutants led to partial rescue (i.e., ΔARID CFI-1 did not completely repress glr-4 expression compared with WT CFI-1 [p = 0.0011]; Figure 7B), suggesting that the core ARID is necessary for repression. Intriguingly, we found evidence of redundancy between the core ARID and eARID; simultaneous deletion of both (ΔARID+ΔeARID) led to a greater effect on glr-4 compared with individual deletions of ΔARID and ΔeARID (Figures 7A and 7B). Last, the REKLES oligomerization domain, which is specific to all ARID3 proteins across phylogeny,^16,17 is required for glr-4 repression (Figures 7A and 7B). Although we cannot discard the possibility that mutated versions of CFI-1 protein lack stability, our overall analysis suggests functional redundancy between the core ARID and eARID and further demonstrate that the core ARID and REKLES are required for glr-4 repression in motor neurons.

Figure 7. Protein motif analysis of CFI-1.

(A) Six CFI-1 constructs tested for rescue effects: WT cDNA, ΔARID, ΔeARID, ΔHTH, ΔARID + ΔeARID, and DREKLES. Two intrinsically disordered regions IDR1 (amino acids [aa] 4–56) and IDR2 (aa 121–157) were identified by NetSurfP-3.0 and PONDR (shown in gray).

(B) Quantification of MNs showing expression of glr-4::tagrfp in cfi-1(ot786) mutant animals expressing the rescue constructs at L1. Two-tailed t test: *p < 0.05, **p < 0.01, ***p < 0.001. Error bars, SEM.

(C) Schematic model summarizing our findings in SAB and nerve cord motor neurons.

DISCUSSION

Our current knowledge of ARID3 functions in the nervous system remains rudimentary. In vertebrate nervous systems, the role of ARID3 proteins is completely unknown. In Drosophila, dead ringer has been implicated in control of axonal pathfinding.^31-33 In C. elegans, a handful of effector genes have been identified as CFI-1 targets in distinct neuron types.^12,35,36,38 Hence, a comprehensive understanding of ARID3-mediated biological processes in the nervous system is lacking. Here, we globally identified 6,396 protein-coding genes as putative direct targets of CFI-1, most of which (77%) are expressed in post-mitotic C. elegans neurons. GO analysis suggested a prominent role of CFI-1 in neuronal terminal differentiation because 63% of its target genes encode NT receptors, transporters, ion channels, transmembrane receptors, cell adhesion molecules, etc. Further, our study offers mechanistic insights into how a single ARID3 protein controls the terminal differentiation program of distinct neuron types by uncovering cell-context-dependent CFI-1 functions.

CFI-1 acts as a terminal selector in IL2 sensory neurons

Transcription factors that bind directly to the cis-regulatory region of terminal differentiation genes (e.g., NT biosynthesis components and receptors, ion channels, neuropeptides, and membrane proteins) and activate their expression have been termed “terminal selectors.”^51-53 Terminal selectors are continuously required in individual neuron types to initiate and maintain expression of terminal differentiation genes, safeguarding neuronal functionality throughout life. To date, terminal selectors have been described in C. elegans, Drosophila, simple chordates, and mice,⁵³ suggesting an evolutionarily conserved role of these critical regulators of neuronal differentiation. For most terminal selectors, however, biochemical evidence of direct binding to their target genes is currently lacking. For example, CFI-1 is a candidate terminal selector in IL2 sensory neurons because five terminal differentiation genes (cho-1/ChT, unc-171VAChT, gcy-19 [receptor-type guanylate cyclase], klp-6 [kinesin-like protein], and unc-5 [netrin receptor]) fail to be properly expressed in IL2 neurons of cfi-1 mutants.³⁶ Here, we identified five additional CFI-1 target genes encoding IL2 terminal differentiation markers (sodium channel degl-2; cilia proteins: cil-7, cwp-4, and ddn-3; tubulin: tba-6). Through ChIP-seq, we provide biochemical evidence that CFI-1 binds directly to most of these genes (9 out of 10), acting as a direct activator (Figure 2E). Further, our analysis of the top 1,000 highly expressed genes in IL2 neurons revealed that the majority of CFI-1 binding (~70%) occurs at terminal differentiation genes, consistent with an in silico prediction study for CFI-1 binding.³⁵ Altogether, a synthesis of our findings with the aforementioned studies indicates that CFI-1 functions as a bona fide terminal selector in IL2 neurons and directly activates scores of terminal differentiation genes. This mode of action is reminiscent of mouse Arid3a, known to also act as a direct activator of gene expression in cells outside of the nervous system.^19,20,24

Accumulating evidence suggests that terminal selectors act in combination with other transcription factors to determine the differentiation of individual neuron types.^35,54 Our findings (Figure S4B), together with previous genetic and in silico prediction studies, support the idea that CFI-1 collaborates with the POU homeodomain transcription factor UNC-86 to directly activate IL2-specific terminal differentiation genes.^35,36 Similarly, it has been proposed that CFI-1 collaborates with two different homeodomain proteins, UNC-42/Prop-1 like and CEH-14/LIM, to control the terminal differentiation of AVD and PVC interneurons,^35,55 respectively, although the underlying mechanisms remain unclear.

Insights into ARID3-mediated gene repression

In IL2 sensory neurons, CFI-1 exerts a dual role (Figure 2). It functions as direct activator of IL2-specific terminal differentiation genes and repressor of ion channel-encoding genes (pkd-2 and lov-1) normally expressed in neurons responsible for pheromone detection (CEM neurons).⁵⁶ Hence, it promotes IL2 terminal differentiation and inhibits an alternative neuronal identity. However, in nerve cord motor neurons, CFI-1 functions as a direct repressor of the glutamate receptor-encoding gene glr-4/GRIK4. Thus, a single ARID3 protein can control the terminal differentiation program of distinct neuron types through mechanisms that depend on cell context, possibly because of CFI-1 participating in distinct neuron-type-specific regulatory complexes that function either as dedicated activators or repressors.

Through an in-depth analysis of glr-4/GRIK4, our findings advance our understanding of how ARID3 proteins mediate gene repression in several aspects. First, cis-regulatory analysis combined with mutagenesis of endogenous CFI-1 binding sites strongly suggest that proximal CFI-1 binding to the glr-4 locus is required for repression (Figure 5). Second, CFI-1 antagonizes three conserved transcription factors (UNC-3/Ebf, LIN-39/Hox4-5, and MAB-5/Hox6-8) that directly activate basal levels of glr-4/GRIK4 expression in nerve cord motor neurons (Figure 6). Third, we found that deletion of highly conserved protein domains (eARID and HTH) predicted to bind DNA (based on Dead ringer structural studies) did not affect the ability of CFI-1 to repress glr-4 (Figure 7). However, combined deletion of eARID with core ARID led to failure to repress glr-4, suggesting functional redundancy between these protein domains. Last, REKLES, an oligomerization domain specific to all ARID3 proteins across phylogeny, is required for glr-4 repression.

One unresolved question is how exactly CFI-1 represses gene expression in C. elegans motor neurons. We found that H3K9 methylation, the epigenetic hallmark of heterochromatin, is necessary for glr-4 repression in nerve cord motor neurons (Figures 6 and S7). Moreover, our analysis supports the idea of cfi-1 operating in the same genetic pathway with the histone methyltransferases met-2 (Setdb1) and set-25 (Suv39h1) to ensure glr-4 transcriptional silencing. Hence, we favor a model where a sequence-specific transcription factor (CFI-1) recruits the histone methyltransferases MET-2 and SET-25 to the glr-4 locus, generating facultative heterochromatin.⁵⁰ Additional possibilities include the following. (1) Because proximal binding is required for glr-4 repression, CFI-1 may interfere with the function of the basal transcription complex.⁵⁷ (2) Based on patterning studies in Drosophila, where Dead ringer physically interacts with the transcriptional corepressor Groucho,^27,28 CFI-1 may cooperate with UNC-37 (Groucho ortholog in C. elegans) to repress glr-4. (3) In mouse stem cells, Arid3a directly represses pluripotency genes by recruiting histone deacetylases (HDACs).²⁴ Future studies are needed to determine whether CFI-1 acts via any of these repressive mechanisms in motor neurons.

CFI-1 continuously antagonizes the activator function of terminal selectors

The glr-4 gene receives positive regulatory input from three conserved transcription factors (UNC-3/Ebf, LIN-39/Hox4-5, and MAB-5/Hox6-8) and negative input from CFI-1. UNC-3 and the Hox proteins LIN-39 and MAB-5 are known terminal selectors in C. elegans nerve cord motor neurons.^45,46,58,59 As such, they are continuously required from embryonic to adult stages to activate expression of multiple terminal differentiation genes.^34,45 On the other hand, our constitutive (null alleles) and conditional (temporally controlled protein depletion) approaches also revealed a continuous requirement for CFI-1 in motor neurons (Figure 4). Hence, we propose that CFI-1 is required continuously to prevent high levels of glr-4 expression driven by terminal selectors (UNC-3 and Hox) (Figure 7C). Such a model is also supported by the continuous requirement of two other repressor proteins (BNC-1/Bnc1 and MAB-9/Tbx20) in C. elegans motor neurons.³⁸ Antagonism between repressor proteins and terminal selectors has also been reported in C. elegans touch receptor neurons, where EGL-44/Tead3 and EGL-46/Insm2 repress terminal selector target genes.⁶⁰ Further, studies in mice indicated that the activity of two terminal selectors, Nurr1 and Crx, which control dopamine and photoreceptor neuron identities, respectively, is counteracted by repressor proteins (Otx2 and Nr2e3).^61,62 Additional work is needed, however, to determine whether all of these repressor proteins (e.g., EGL-44 and Otx2) act directly or are required continuously.

Evolutionary implications

The terminal selector UNC-3 is required to maintain cfi-1 expression in nerve cord motor neurons.³⁴ Hence, the repressor protein (CFI-1) and glr-4 are targets of UNC-3, generating an incoherent feedforward loop (FFL) (Figure 7C). In SAB motor neurons, however, CFI-1 is not expressed, and UNC-3 is able to drive high levels of glr-4 expression (Figure 7C).⁶³ From an evolutionary perspective, incoherent FFLs have been proposed to diversify a ground state into various substates.⁶⁴ One can envision an ancestral state where an UNC-3 ortholog is present in a relatively homogeneous population of motor neurons but the recruitment of a repressor (CFI-1) enabled their diversification. Hence, the UNC-3 → CFI-1 −I glr-4 incoherent FFL may help distinguish, at the molecular level, nerve cord motor neurons that control locomotion from SAB motor neurons that control head movement. In agreement with this idea, incoherent FFLs are known to diversify gustatory neurons in C. elegans and photoreceptor cells in Drosophila.^65,66

Limitations of the study

Our findings stress the importance of cell context in ARID3 transcription factor activity. However, the precise molecular mechanisms through which CFI-1 acts as an activator in IL2 sensory neurons and as a repressor in nerve cord motor neurons warrant further investigation. Two unresolved questions remain. How do CFI-1 and UNC-86 (POU) cooperate to activate the IL2 terminal differentiation program? Does CFI-1 ensure transcriptional silencing of glr-4 in motor neurons by recruiting the histone methyltransferases MET-2 (SETDB1) and SET-25 (SUV39H1) to induce H3K9 methylation? Last, our study strongly suggests that CFI-1 acts directly to control gene expression in sensory and motor neurons. However, ChIP-seq was performed on whole-animal lysates because protocols for neuron type-specific ChIP-seq are not yet established in C. elegans. Hence, it remains possible that a fraction of the CFI-1 binding events described here for IL2 sensory neurons and nerve cord motor neurons occur in other cfi-1-expressing neurons.

STAR★METHODS

RESOURCE AVAILABILITY

Lead contact

Further information and requests for resources and reagents should be directed to and will be fulfilled by the lead contact, Paschalis Kratsios (pkratsios@uchicago.edu).

Materials availability

C. elegans strains generated in this study are available upon request.

Data and code availability

Sequencing data have been deposited at GEO under accession code GSE205628. Moreover, all data generated or analyzed for this study are included in the manuscript and supporting files. Microscopy data reported in this paper will be shared by the lead contact upon request.

This paper does not report original code.

Any additional information required to reanalyze the data reported in this paper is available from the lead contact upon request.

EXPERIMENTAL MODEL AND SUBJECT DETAILS

C. elegans hermaphrodite animals were grown at 20°C or 25°C on nematode growth media (NGM) plates supplied with E. coli OP50 as food source.⁶⁷ Larval stage 2, 3, and 4, as well as young adult (day 2) animals were analyzed as described in main text and figures. All strains used or generated for this study are listed in key resources table.

KEY RESOURCES TABLE

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Antibodies
Monoclonal ANTI-FLAG M2 antibody	Sigma-Aldrich	Cat. No. F1804;RRID: AB_262044
Alexa Fluor 594 donkey anti-mouse antibody	Molecular Probes	Cat. No. A-21203; RRID: AB_141633
Bacterial and virus strains
E.coli: OP50-1	Caenorhabditis Genetics Center (CGC) (https://cgc.umn.edu/)	Wormbase ID: WBStrain00041971
Critical commercial assays
TOPO TA cloning kit	Invitrogen	Cat. No. K805010
Plasmid miniprep kit	QIAGEN	CAT:12125
Deposited data
CFI-1 ChIP-seq data	GEO	GSE205628
Experimental models: Organisms/strains
C. elegans: Wild-type N2	Caenorhabditis Genetics Center (CGC) N2; (https://cgc.umn.edu/)	WormBase ID: WBStrain00000001
C. elegans: glr-4 (kas29[2xNLS::mScarlet::SL2::glr-4 with COE1 motif mutated from CCCAAGG to TTTCCAA]	This study	KRA762
C. elegans: syb3680[2xNLS::mScarlet::SL2::glr-4]	This study	PHX3680
C. elegans: syb1778[3xFLAG::cfi-1]	This study	PHX1778
C. elegans: cfi-1(ot786)	This study	KRA464
C. elegans: kas16[mNG::AID::cfi-1]	This study	KRA345
C. elegans: otIs476 [glr-4tagRFP]; kasEx146[Punc-4::cfi-1_cDNA+pha-1 cDNA]	This study	KRA436
C. elegans: akEx32[glr-4_4.9 kb::gfp]	CGC	VM141
C. elegans: otEx6583[glr-4_3.7 kb::TagRFP + pha-1 (+)]	CGC	OH14196
C. elegans: otEx6579[glr-4_3.14 kb::TagRFP + pha-1 (+)]	CGC	OH14192
C. elegans: otEx6576[glr-4_2.23 kb::TagRFP + pha-1 (+)]	CGC	OH14189
C. elegans: syb5348[2xNLS::mScarlet::SL2::glr-4 with 11 CFI-1 binding sites mutated]	This study	PHX5348
C. elegans: unc-3(n3435)	CGC	MT10785
C. elegans: lin-39 (n1760);mab-5 (e1239)/ht2; lgIs58 [[gcy-32::gfp]]	CGC	LE4023
C. elegans: kasEx271[glr-4_938 bp::TagRFP + pha-1 (+)]	This study	KRA763
C. elegans: cfi-1(ot786); otIs476[glr-4::tagRFP]; kasEx272[cfi-1_cDNA + myo-2::gfp]	This study	KRA764
C. elegans: cfi-1(ot786); otIs476[glr-4::tagRFP]; kasEx273[cfi-1_cDNA_ARID_del + myo-2::gfp]	This study	KRA765
C. elegans: cfi-1(ot786); otIs476[glr-4::tagRFP]; kasEx274[cfi-1_cDNA_eARID_del + myo-2::gfp]	This study	KRA766
C. elegans: cfi-1(ot786); otIs476[glr-4::tagRFP]; kasEx276[cfi-1_cDNA_core ARID_del + eARID_del + myo-2::gfp]	This study	KRA796
C. elegans: cfi-1(ot786; otIs476[glr-4::tagRFP]; kasEx275[cfi-1_cDNA_HTH_del + myo-2::gfp]	This study	KRA767
C. elegans: lin-15B&lin-15A(n765); akEx31[glr-5::GFP + lin-15(+)]	CGC	VM133
C. elegans: otIs544 [cho-1(fosmid)::SL2::mCherry::H2B + pha-1(+)]	CGC	OH13646
C. elegans: otIs534 [cho-1(fosmid)::SL2::NLS::YFP::H2B]	CGC	OH12543
C. elegans: unc-30 (syb2344[unc-30::mNG::3xFLAG::AID]	This study	PHX2344
C. elegans: otEx4805 [twk-40_prom::NLS::DsRed + pha-1 (+)]	CGC	OH10723
C. elegans: otEx4806 [twk-43_prom::NLS::DsRed + pha-1 (+)]	CGC	OH10724
C. elegans: unc-42(e270); ctIs43 [dbl-1p::GFP + dbl-1p::GFP::NLS + unc-119(+)]	CGC	BW1946
C. elegans: pha-1(e2123) III; kasEx286[npr-29::TagRFP]	This study	KRA597
C. elegans: pha-1(e2123) III; kasEx284[ncs-2::TagRFP]	This study	KRA595
C. elegans: met-2(n4256) III; otIs476 [glr-4::RFP] V; kasEx111 [Plin-39::met-2 + myo-2::gfp]	This study	KRA400
C. elegans: met-2(n4256) III; otIs476 [glr-4::RFP] V; kasEx112 [Plin-39::met-2 + myo-2::gfp]	This study	KRA401
C. elegans: met-2(n4256) III;otIs476 [glr-4::RFP] V; kasEx113[Plin-39::met-2 + myo-2::gfp]	This study	KRA402
C. elegans: N2;Ex[cil-7p::gfp + klp-6p::mcherry + unc-122p::ds-red]	Lab of Dr. Junho Lee (Seoul National University, Seoul, Republic of Korea)	LJ871
C. elegans: N2;Ex[cwp-4p::gfp + unc-122p::ds-red]	Lab of Dr. Junho Lee (Seoul National University, Seoul, Republic of Korea)	LJ870
C. elegans: N2;Ex[ddn-3p::gfp + rol-6(su1006)]	Lab of Dr. Junho Lee (Seoul National University, Seoul, Republic of Korea)	LJ868
C. elegans: otIs825 [degl-1p::GFP + unc-122p::GFP]	CGC	OH17019
C. elegans: degl-2(syb5229[degl-2::SL2::gfp::H2B]) IV	CGC	PHX5229
C. elegans: N2;Ex[tba-6p::gfp + klp-6p::mcherry +unc-122p::ds-red]	Lab of Dr. Junho Lee (Seoul National University, Seoul, Republic of Korea)	LJ872
C. elegans: met-2(n4256) III; glr-4([syb3680[2xNLS:: mScarlet::SL2::glr-4]	This study	KRA797
C. elegans: met-2(n4256) III; set-25(n5021) III; glr-4([syb3680[2xNLS::mScarlet::SL2::glr-4]	This study	KRA798
C. elegans: cfi-1(ot786) I; met-2(n4256) III; set-25(n5021) III; glr-4([syb3680[2xNLS::mScarlet::SL2::glr-4]	This study	KRA799
Oligonucleotides
Forward primer to delete core ARID domain: GAATTGTCAGACAAACTTTCAAATCAATCTGATC	This study	Not applicable
Reverse primer to delete core ARID domain: GATTTGAAAGTTTGTCTGACAATTCGTACAAC	This study	Not applicable
Forward primer to delete HTH domain: GGACTAAATCTTCCATCATCTATCCTT CGAACACAATATCAAAAATATTTATATG	This study	Not applicable
Reverse primer to delete HTH domain: CATATAAATATTTTTGATATTGTGTTCG AAGGATAGATGATGGAAGATTTAGTCC	This study	Not applicable
Forward primer to delete eARID domain: GTGTGAAAAAGAGGCTCCATCATTTCCATTACC	This study	Not applicable
Reverse primer to delete eARID domain: GAAATGATGGAGCCTCTTTTTCACACTCATAATC	This study	Not applicable
Forward primer to delete REKLES domain: GCCGAGCAAATGTCAGAGTCTTAAGC GCCGG	This study	Not applicable
Reverse primer to delete REKLES domain: CTTAAGACTCTGACATTTGCTCGGCTCCATAG	This study	Not applicable
Forward primer to delete core ARID+eARID domains: CGAATTGTCAGACGCTCCATCATTTCCATTACC	This study	Not applicable
Reverse primer to delete core ARID+eARID domains: GAAATGATGGAGCCGTCTGACAATTCGTACAAC	This study	Not applicable
Software and algorithms
Version 2.3.69.1000, Blue edition	Carl Zeiss Microscopy	https://www.zeiss.com/microscopy/us/products/microscope-software/zen.html
GraphPad Prism 5	GraphPad Software, La Jolla, CA, USA	https://www.graphpad.com/scientific-software/prism/
Other
Zeiss, Axio Imager Z2	Carl Zeiss Microscopy	https://www.zeiss.com/microscopy/us/products/microscope-software/zen.html

METHOD DETAILS

Generation of transgenic animals carrying transcriptional fusion reporters and overexpression or rescue constructs

Reporter gene fusions for cis-regulatory analyses of glr-4 were made with PCR fusion.⁶⁸ Genomic regions were amplified and fused to the coding sequence of tagrfp followed by the unc-54 3′ UTR. PCR fusion DNA fragments were injected into young adult pha-1(e2123) hermaphrodites at 50 ng/μL together with pha-1 (pBX plasmid) as co-injection marker (50 ng/μL). To generate animals with cfi-1 overexpression in the SAB neurons, the unc-4 promoter was fused to the cDNA sequence of cfi-1 followed by the unc-54 3′ UTR. The fluorescent co-injection marker myo-2::gfp was used (2 ng/ul) and the PCR fusion DNA fragments were injected into young adult animals carrying the glr-4::tagrfp reporter at 50 ng/ul. To generate transgenic animals carrying different versions of the cfi-1 cDNA rescue constructs (WT, ΔARID, ΔeARID, ΔHTH, ΔREKLES), the cfi-1 enhancer driving expression in motor neurons³⁴ was fused to the corresponding version of cfi-1 cDNA followed by the unc-54 3′ UTR. The fluorescent co-injection marker myo-2::gfp was used (2 ng/ul) and the PCR fusion DNA fragments were injected into young adults of cfi-1(−) mutants carrying the glr-4::tagrfp reporter at 50 ng/ul. A similar approach was used to drive expression of met-2 cDNA in nerve cord motor neurons under the control of a lin-39 regulatory element (intron 1).⁶⁹

Targeted genome editing

The endogenous glr-4 reporter allele syb3680 [2xNLS::mScarlet::glr-4] (PHX3680 strain) was generated by SunyBiotech via CRISPR/Cas9 genome editing by inserting the 2xNLS::mScarlet cassette immediately after the ATG of glr-4. Moreover, the endogenous glr-4 reporter allele syb5348 [2xNLS::mScarlet::SL2::glr-4^{11 CFI-1 sites MUT}] that carries nucleotide substitutions in eleven CFI-1 binding sites was also generated by SunyBiotech. The endogenous glr-4 reporter allele kas29 [2xNLS::mScarlet::SL2::glr-4^{COE1 MUT}] that carries nucleotide substitutions in a single UNC-3 binding site (COE1 motif) was generated in the Kratsios lab by using homology dependent repair and inserting a synthesized DNA fragment that carries the desired mutations.

Microscopy

Imaging slides were prepared by anesthetizing worms with sodium azide (NaN₃, 100 mM) and mounting them on a 4% agarose pad on glass slides. Images were taken with an automated fluorescence microscope (Zeiss, Axio Imager Z2). Images containing several z stacks (0.50 μm intervals between stacks) were taken with Zeiss Axiocam 503 mono using the ZEN software (Version 2.3.69.1000, Blue edition). Representative images are shown following max-projection of 2–5 μm Z-stacks using the maximum intensity projection type. Image reconstruction was performed with Image J.⁷⁰

Motor neuron subtype identification

Motor neuron subtypes were identified based on combinations of the following factors: [1] co-localization with or exclusion from additional reporter transgene with known expression patterns; [2] Invariant position of neuronal cell bodies along the ventral nerve cord, [3] Birth order of specific motor neuron subtypes (e.g., during embryonic or post-embryonic stages); [4] Total cell numbers in each motor neuron subtype.

Bioinformatic prediction of binding motifs

Information of the CFI-1 and UNC-86 binding motifs is provided in the Catalog of Inferred Sequence Binding Preferences database (http://cisbp.ccbr.utoronto.ca). To predict and identify CFI-1 and UNC-86 binding motifs, we utilized tools provided by MEME (Multiple Expectation maximization for Motif Elicitation) bioinformatics suite (http://meme-suite.org/), and performed FIMO (Find Individual Motif Occurrences) motif scanning analysis. De novo motif analysis was conducted with HOMER (http://homer.ucsd.edu/homer/motif/).

Chromatin immunoprecipitation (ChIP)

ChIP assay was performed as previously described, with the following modifications.^71,72 Synchronized L1 cfi-1(syb1778 [3xFLAG::cfi-1]) worms and N2 worms were cultured on 10 cm plates seeded with OP50 at 20°C overnight. Late L2 worms were cross-linked and resuspended in FA buffer supplemented with protease inhibitors (150 mM NaCl, 10 μL 0.1 M PMSF, 100 μL 10% SDS, 500 μL 20% N-Lauroyl sarsosine sodium, 2 tablets of cOmplete ULTRA Protease Inhibitor Cocktail [Roche Cat.# 05892970001] in 10 mL FA buffer). For each IP experiment, 200 μL worm pellet was collected. The sample was then sonicated using a Covaris S220 with the following settings: 200 W Peak Incident Power, 20% Duty Factor, 200 Cycles per Burst for 1 min. Samples were transferred to centrifuge tubes and spun at the highest speed for 15 min. The supernatant was transferred to a new tube, and 5% of the material was saved as input and stored at −20°C. The remainder was incubated with FLAG antibody at 4°C overnight. Wild-type (N2) worms do not carry the 3xFLAG tag and serve as negative control. The cfi-1(syb1778[3xFLAG::cfi-1]) CRIPSR generated allele was used in order to immunoprecipitate the endogenous CFI-1 protein. On the next day, 20 μL Dynabeads Protein G (1004D) was added to the immunocomplex which was then incubated for 2 h at 4°C. The beads were then washed at 4°C twice with 150 mM NaCl FA buffer (5 min each), once with 1 M NaCl FA buffer (5 min). The beads were transferred to a new centrifuge tube and washed twice with 500 mM NaCl FA buffer (10 min each), once with TEL buffer (0.25 M LiCl, 1% NP-40, 1% sodium deoxycholate, 1 mM EDTA, 10 mM Tris-HCl, pH 8.0) for 10 min, twice with TE buffer (5 min each). The immunocomplex was then eluted in 200 μL elution buffer (1% SDS in TE with 250 mM NaCl) by incubating at 65°C for 20 min. The saved input samples were thawed and treated with the ChIP samples as follows. One (1) μL of 20 mg/mL proteinase K was added to each sample and the samples were incubated at 55°C for 2 hours then 65°C overnight (12–20 hours) to reverse cross-link. The immonuprecipitated DNA was purified with Ampure XP beads (A63881) according to manufacturer’s instructions.

ChIP-seq data analysis

Unique reads were mapped to the C. elegans genome (ce10) with bowtie2.⁷³ Peak calling was then performed with MACS2 (minimum q-value cutoff for peak detection: 0.005).⁷⁴ For visualization purposes, the sequencing depth was normalized to 1x genome coverage using bamCoverage provided by deepTools⁷⁵ and peak signals were shown in Integrated Genome Viewer.⁷⁶ Heatmap of peak coverage in regard to CFI-1 enrichment center was generated with NGSplot.⁷⁷ The average profile of peaks binding to TSS region was generated with ChIPseeker.⁷⁸

Immunocytochemistry

C. elegans animals carrying the cfi-1 (syb1778[3xFLAG::cfi-1]) allele (PHX1778 strain, SunyBiotech) were grown at 25°C on nematode growth media (NGM). We followed an immunocytochemical staining procedure described previously.⁷⁹ In brief, worms were prepared for staining following the freeze-crack procedure and they were subsequently fixed in ice-cold acetone (5 min) and ice-cold methanol (5 min). Worms were transferred using a Pasteur pipette from slides to a 50 mL conical tube that contained 40 mL 1XPBS. Following a brief centrifugation (2 min, 3,000 rpm), worms were pelleted and 1xPBS was removed. Next, the worms were incubated with 300 μL of blocking solution (1XPBS, 0.2% Gelatin, 0.25% Triton) for 30 min at room temperature (rolling agitation). Following removal of the blocking solution, worms were incubated over-night with a mouse monoclonal antibody against FLAG (Sigma-Aldrich, F1804) [1:1,000 dilution in PGT solution (1xPBS, 0.1% Gelatin, 0.25% Triton. Next, the primary antibody solution was removed and worms were washed five times with washing solution (1XPBS, 0.25% Triton). Worms were incubated with an Alexa Fluor 594 donkey anti-mouse IgG secondary antibody (1:1000 in PGT solution, A-21203, Molecular Probes) for 3 hours at room temperature. Following 5 washes, worms were mounted on a glass slide and examined at an automated fluorescence microscope (Zeiss, AXIO Imager Z2).

Putative CFI-1 target genes in various neuron types

The top 1,000 highest expressed genes in IL2, DA, DB, VA, VB, ASE, RMD, DVB, and PHC (transcripts per million, tpm) were mined from available single-cell RNA-sequencing data (CenGEN). This dataset was computationally compared to a dataset of CFI-1 ChIP-Seq targets using the ‘semi_join’ function in R (package Dplyr 1.0.7). This generated a new data frame containing genes in the scRNA-seq dataset that are also putatively bound by CFI-1. Similarly, the ‘set_diff’ function (Dplyr 1.0.7) was used to generate a new data frame containing genes that are expressed in IL2 based on scRNA-seq but which are not found in the CFI-1 ChIP-seq dataset. Gene list analysis (WormCat 2.0)³⁹ was performed on both data frames to functionally classify all genes based on protein class ontology.

Temporally controlled protein degradation

Temporally controlled protein degradation was achieved with the auxin-inducible degradation system.⁴³ TIR1 expression was driven by the pan-neuronal promoter in the transgene otTi28[unc-11prom8+ehs-1prom7+rgef-1prom2::TIR1::mTurquoise2::unc-54 3′UTR]. To induce degradation of CFI-1 proteins, we used the allele kas16[cfi-1::mNG::AID]. Worms at the L4 stage were grown at 20°C on NGM plates coated with 4 nM auxin (indole-3-acetic acid [IAA] dissolved in ethanol) or ethanol (negative control) for 2 days before testing (see figure legends for exact time in specific experiments). All plates were shielded from light.

Single molecule RNA fluorescent in situ hybridization (sm RNA FISH)

Synchronized L1 worms were collected from the plates and washed with M9 buffer 3 times. Worms were incubated in the fixation buffer (3.7% formaldehyde in 1x PBS) for 45 minutes at room temperature. Worms were then washed twice with 1x PBS, resuspended in 70% ethanol and left at 4°C for two nights. After removing the ethanol, worms were incubated in the wash buffer (10% formamide in 2x SSC buffer) for 5 minutes and the wash buffer was removed afterwards. A glr-4 probe was designed using the Stellaris Probe Designer website (Biosearch Technologies). The probe was mixed in hybridization buffer (0.1 g/mL dextran sulfate [Sigma D8906-50G], 1 mg/mL Escherichia coli tRNA [ROCHE 10109541001], 2 mM vanadyl ribonucleotide complex [New England Biolabs S1402s], 0.2 mg/mL RNase-free BSA [Ambion AM2618], 10% formamide) and added to the worms. The hybridization buffer was removed and worms were washed twice in wash buffer (DAPI was added during the second wash and incubated for 30 minutes in the dark for nuclear counterstaining). Worms were washed once in 2x SSC, incubated in GLOX buffer (0.4% glucose, 0.1 M Tris-HCl, 2x SSC) for 2 minutes for equilibration, and the resuspended in GLOX buffer with glucose oxidase and catalase added. The samples were then examined under the fluorescent microscope.

Real-time PCR assay for glr-4 expression level analysis

Synchronized L4 stage wildtype and cfi-1(−) worms were collected, and mRNA was extracted. cDNA library was prepared using the Superscript first strand cDNA synthesis kit (Invitrogen #11904-018). RT-PCR TaqMan assays for the genes glr-4 (assay ID: Ce02435302_g1) and pmp-3 (Ce02485188_m1) were performed, and the expression level of glr-4 was determined in each genotype after normalizing to the expression of the housekeeping gene pmp-3.

Harsh touch behavioral assay

Harsh touch was delivered with a platinum wire pick as previously described.⁸⁰ The stimulus was applied from above the animals by pressing down with the edge of the pick on the tail of non-moving adult animals. Each animal was tested only once. Worms that moved forward in response to harsh touch were scored as normal response. Results are presented as fractions of animals that responded normally.

QUANTIFICATION AND STATISTICAL ANALYSIS

For data quantification, graphs show values expressed as mean ± standard error mean. Statistical analyses were performed using unpaired t-test (two-tailed). Calculations were performed using the GraphPad QuickCalcs online software (http://www.graphpad.com/quickcalcs/). Differences with p < 0.05 were considered significant. Asterisks in figures indicate statistical significance as follows: *p < 0.05, **p < 0.01, ***p < 0.001.

Highlights.

• ChIP sequencing identifies direct target genes of CFI-1/ARID3 in C. elegans neurons

• CFI-1 directly activates effector genes necessary for sensory neuron (IL2) identity

• CFI-1, MET-2/SETDB1, and SET-25/SUV39H2 control gene repression in motor neurons

• The ARID3-specific oligomerization domain (REKLES) is required for gene repression

INCLUSION AND DIVERSITY

One or more of the authors of this paper received support from a program designed to increase minority representation in their field of research.