Adult single-nucleus neuronal transcriptomes of insulin signaling mutants reveal regulators of behavior and learning

Summary

Gene expression in individual neurons can change during development to adulthood and can have large effects on behavior. Additionally, the insulin/insulin-like signaling (IIS) pathway regulates many of the adult functions of Caenorhabditis elegans, including learning and memory, via transcriptional changes. We used the deep resolution of single-nucleus RNA sequencing to define the adult transcriptome of each neuron in wild-type and daf-2 mutants, revealing expression differences between L4 larval and adult neurons in chemoreceptors, synaptic genes, and learning/memory genes. We used these data to identify adult new AWC-specific regulators of chemosensory function that emerge upon adulthood. daf-2 gene expression changes correlate with improved cognitive functions, particularly in the AWC sensory neuron that controls learning and associative memory; behavioral assays of AWC-specific daf-2 genes revealed their roles in cognitive function. Combining technology and functional validation, we identified conserved genes that function in specific adult neurons to control behavior, including learning and memory.

Graphical abstract

Highlights

• •Day 1 adult worms exhibit behavioral and transcriptional differences from L4 larva
• •csGPCR expression patterns change greatly from L4 to day 1 adulthood
• •Daf-2 animals’ chemosensory neurons exhibit the most changes in gene expression
• •Transcriptional changes in the AWC are functionally linked to learning and memory

St. Ange et al. provide a single-nucleus RNA sequencing atlas of adult wild-type and IIS mutant Caenorhabditis elegans neurons, identifying genes that turn on from L4 to adulthood, including GPCRs. Functional behavior experiments revealed the roles of genes expressed in single neurons, including previously unidentified IIS targets.

Introduction

Caenorhabditis elegans is a powerful system to discover conserved mechanisms of shared traits, including lifespan regulation, neuronal functions, and cognitive behaviors that are conserved from worms to mammals. With only 302 adult neurons of 118 neuron types, individual neurons can have a great impact on behavior. However, there is also surprising complexity in each worm neuron; for example, unlike the “one olfactory neuron/one receptor” model of the mouse olfactory bulb,¹ C. elegans encodes many more than 302 receptors in its genome. Individual chemoreceptors expressed in specific neurons determine specific functions of C. elegans neurons.² Where each receptor is expressed in adult neurons and whether their expression changes in specific mutants is currently unknown.^3,4,5 Other genes such as synaptic and signaling proteins play specific roles in individual neurons as well. Therefore, we must understand the transcriptional profile of individual neurons to better understand regulation of behavior. Because we are interested in understanding how insulin/insulin growth factor-1 signaling (IIS), which extends lifespan and improves memory,^6,7 affects expression in specific neurons to regulate neuron-specific behaviors, being able to rapidly identify transcriptome changes in individual neurons in mutants is critical.

Previously, we carried out pan-neuronal transcriptional analysis,^8,9 which is useful for identifying and characterizing previously unknown neuronal genes, but it lacks the cellular resolution to determine gene expression in individual neuron types. Bulk sequencing of the entire nervous system cannot detect the location of differential expression and might omit differential expression targets that are only present in a subset of neurons. We also used bulk sequencing of specific isolated neurons⁸ (six mechanosensory neurons), but this approach requires labeling the specific neuron type prior to sorting, limiting exploratory studies. Recent whole-worm single-cell transcriptomic data^10,11,12,13 distinguish cell types broadly, but may lack sufficient resolution and depth in neuronal expression (see Figure 2) to answer questions about individual neurons. The CeNGEN database,¹¹ which is based on single-cell sequencing of late larval (L4) stage wild-type (WT) C. elegans hermaphrodite neurons, is an excellent resource for neuron-specific gene expression for all 118 neuron types. However, identifying genes required in adult neurons, or comparing animals of different genetic backgrounds or beyond the L4 larval stage, requires an alternative approach. Other studies have used whole animals rather than specifically isolating neurons, but because germline nuclei can vastly outnumber other nuclei types, some whole-worm, germline-intact approaches may not provide sufficient single-neuron-type-level information. Additionally, approaches using germlineless mutants might affect the function and transcription of neurons.¹⁴ Finally, to address questions about daf-2 (insulin/IGF-1-like receptor) mutants, we must carry out direct experimental comparisons between mutant and WT adults. Therefore, higher-resolution transcriptional profiling of adult neurons from germline-intact animals is necessary.

Figure 2.

Methods and metrics of different RNA sequencing datasets

Five RNA sequencing dataset strains, isolation approach, metrics, and validation presented alongside our dataset.

Single-nucleus RNA sequencing^15,16 (snSeq) has become the gold standard for mammalian neuronal transcriptional analyses because the identity of the neuron is retained when mRNA is isolated from the nucleus, in contrast to methods where cytoplasmic mRNA from interconnected neurons can be confounded. We previously used snSeq to compare hippocampal neurons from aged WT and GNAQ-overexpression mice¹⁷; here, we adapted this method for adult C. elegans neuronal nuclei to characterize the neuronal transcriptomes of day 1 adult WT and daf-2 mutants. We used these data to identify neuronal transcriptome shifts during the development from L4 to adulthood, genes expressed in adult WT neurons, and daf-2-dependent transcriptional changes in individual neurons. Like behavior, neuronal gene expression changes significantly between L4 and young adult (day 1), including differential expression of G protein-coupled receptors (GPCRs) in sensory neurons. Neuron-specific IIS analysis revealed significant differences in the regulation of metabolic and longevity genes. We identified new gene functions in WT behavior and neuron-type-specific daf-2-regulated genes that play important roles in regulating sensory and cognitive behaviors. In addition to providing an atlas of day 1 adult WT and daf-2 neuronal gene expression, these data demonstrate that single-nucleus analysis of adult neurons is a powerful tool for understanding neuron-specific changes with the onset of adulthood and in behavioral mutants, and for the identification of conserved mechanisms of enhanced neuron-specific functions in daf-2 mutants.

Results

snSeq of adult C. elegans neurons

C. elegans develops through a series of four larval stages separated by molts, culminating in reproductively competent adults (day 1 adults). Odor preferences change during development.^2,14 We observed significant differences in the behavior of L4 larvae and day 1 adults; specifically, the associative learning ability of L4 larvae is significantly lower than that of adults (Figure 1A). Therefore, data from larval-stage neurons, which is the primary existing source of single-cell data available,¹¹ may not accurately reflect gene expression in adult neurons.

Figure 1.

Single-nucleus RNA sequencing of adult Caenorhabditis elegans neurons

(A) L4 larvae have impaired learning ability (two biological replicates). Each dot is individual chemotaxis plate (average 150 worms per plate). Boxplots: center line, median; box range, 25th–75th percentiles; whiskers denote minimum–maximum values. Unpaired, two-tailed Student’s t test. ∗∗∗∗p < 0.0001.

(B) Single-nucleus neuron isolation (Prgef:his-58::GFP) and sequencing method (see STAR Methods).

(C) Heatmaps comparing L4 transcriptional data (CeNGEN)¹¹ to our day 1 transcriptional data.

(D) Uniform manifold approximation and projection (UMAP) of all 88,497 quality-controlled nuclei (both wild type [WT] and daf-2).

(E) A hierarchical dendrogram based on gene expression in each day 1 adult neuron, color-coded by functional subtype.

Many C. elegans neurons have long processes and intricate synaptic connections, and conventional single-cell RNA sequencing (RNA-seq) preparation can cause axon breakage, thus losing synapses and allowing RNA to leak out of the cell body. In fact, we previously took advantage of these neuron disruptions to isolate synaptic fragments and identify synaptically enriched mRNA transcripts.¹⁸ By contrast, transcripts inside the nucleus are closely linked with neuron identity. Therefore, we isolated neuronal nuclei from animals expressing a pan-neuronal histone-GFP tag (three biological replicates per genotype). Mechanical and chemical lysis were employed to obtain a suspension of nuclei, which were then stained with Hoechst dye (Figure S1A) and fluorescence-activated cell sorting (FACS) sorted for both GFP and Hoechst signal (Figure 1B); RNA was isolated and sequenced (STAR Methods). After Cell Ranger processing and ambient RNA removal (SoupX), samples were assessed for quality. Number of features (genes), number of counts (unique molecular identifiers [UMIs]), and percentage of mitochondrial transcripts were assessed, and cutoffs were set to filter out empty droplets and doublets. Each dataset required its own cutoff ranges (STAR Methods) (Figure S1B). In total, 88,497 neuronal nuclei were obtained after quality control, with an average of 756 UMIs per nucleus and 526 genes per nucleus (Figure 2). Thus, our sequencing approach using just three replicates of WT neurons has quality metrics similar to those of the CeNGEN dataset¹¹ (Figure 2). We detected a high percentage of previously identified synaptic and neuronal transcripts (91.5%), suggesting that our method successfully identified neuronal transcripts (Figures 1C and 2; Tables S1 and S2), despite analyzing slightly fewer neurons (63,565 vs. 70,296).

Neuronal gene expression differs significantly between L4 larvae and day 1 adults

Because adults differ from L4 larvae in their behavior (Figure 1A), we wondered which genes change expression from L4 to day 1 of adulthood. Our snSeq detected 94% of previously identified 542 synaptically localized transcripts¹⁸ (Table S1), while a comparable single-cell dataset of L4 neurons¹¹ detected only 66% of the 542 genes (Figure 1C). In general, neuronally enriched genes⁹ are largely present in both datasets (87.8%), suggesting that the highest-expressed genes in neurons are consistent from L4 to adults (Figure 1C; Table S2), although some genes appeared only in L4 larval or day 1 adult neurons. While 65% of synaptically enriched genes are shared, many more are detected in adult neurons (29.1%) than in L4 (2.4%), likely due to a shift in expression from L4 larvae to adulthood rather than technical differences, since single-cell sequencing is more likely than snSeq to identify more transcripts. Similarly, while 63% of memory genes are shared between L4 and day 1 neurons, more of these long-term associative memory (LTAM) genes¹⁹ appear in adult neurons (20.4%) that are not present in L4 (3.7%). It is unlikely that the higher representation of these genes in adult neurons is a detection artifact, as the L4 larval dataset sampled more than twice the cells (46,627 WT nuclei vs. 100,955 cells; Figure 2). Our data reveal shifts in gene expression from the late larval stage to adulthood that may correlate with changes in behavior.

snSeq identifies 107 of 118 neuron classes

To assign cell identities to the clusters, we used a combined systematic and manual curation approach (Figure 1D; STAR Methods; Table S3). On average, we found 355 nuclei per annotated neuron cluster (710 across two genotypes, WT and daf-2) on the same order of magnitude as the CeNGEN clusters (549.2 cells per cluster; Figure 2).

The 302 neurons of C. elegans are sorted into specific classes based on left/right pairs and higher-fold symmetry, reducing our expectation of distinct neuron clusters to 118 specific neuron classes. Annotations from our dataset match 107 of the 118 neuron classes (Figures 1D and 2). We identified clusters that correspond to a significant proportion of the neuron population of hermaphrodite C. elegans, including 83 of the 87 sensory neurons (95.4%), 67 of the 77 interneurons (87.0%), 101 of the 118 motor neurons (85.6%), and 17 of the 20 pharyngeal neurons (85.0%). Non-neuronal cells, such as intestine, hypodermis, pharynx, sperm, and excretory cells were identifiable and distinguishable from neurons.

Hierarchical clustering by transcriptome reveals functional relationships

Next, we asked how similar gene expression in specific neurons is by hierarchical clustering (Figure 1E). Our dendrogram closely grouped neurons by type, except for a few motor and pharyngeal neurons, indicating that our sequencing approach captured the cellular identities of different types of neurons. Although neurons arise from specific embryonic lineages, these transcriptomic-based dendrograms differ from anatomy-based lineage dendrograms. For example, URX and AQR/PQR clusters neighbor one another, which is not surprising due to their shared O₂- and CO₂-sensing abilities. However, these neurons arise from different lineages: URX develops from the AB cell linage, while AQR and PQR are from the Q cell lineage. Similarly, the AVH and AVF interneurons cluster together transcriptomically, reflecting their shared functions as interneurons; however, AVH neurons develop from the AB cell lineage, while AVF neurons are from P1 or W cell lineages. The close clustering of the sensory ALN/PLN neurons and SDQ/SAA interneurons might reflect their shared O₂-sensory functions. Other neurons cluster together unexpectedly, such as the ALA, DVA, and DVC interneurons, possibly due to their dual functions as interneurons and mechanosensory neurons, and the AIM and BDU interneurons, suggesting functional similarity. This hierarchical clustering of adult neuron transcriptomes provides insights into their previously understudied adult functions that may be lacking from existing anatomical and developmental lineage information.

Fluorescence imaging validates gene expression differences between L4 and day 1

We generated promoter::GFP reporters of several of the most highly expressed genes in our adult dataset that were not detected in L4 neurons. The serpentine receptor gene srz-64 is highly expressed in the adult ADL neuron cluster (Figure 3A) but is not present in L4 data¹¹; indeed, fluorescence of Psrz-64::GFP significantly increases from late larval L4 stage to day 1 of adulthood (Figures 3B and 3C). Similarly, Y65B4A.4, which is highly detected in the ADL cluster (Figure 3D), and M04C7.4, which is highly detected in the BAG neuron cluster (Figure 3G), both increase their GFP fluorescence from L4 to adulthood (Figures 3E, 3F, 3H, and 3I), demonstrating that our method detected real biological changes in expression upon development to adulthood.

Figure 3.

Promoter-GFP analysis of genes that increase in expression from L4 to day 1

(A) srz-64 expression (inset: ADL).

(B) Psrz-64::GFP animals on L4 and day 1.

(C) Quantification of L4 and day 1 adult animals; n = 61.

(D) Y65B4A.4 expression (inset: ADL).

(E) Y65B4A.4p::GFP animals on L4 and day 1.

(F) Quantification of L4 and day 1 adult animals; n = 60 neurons, two biological replicates.

(G) M04C7.4 expression (inset: BAG neuron cluster).

(H) Representative images of M04C7.7p::GFP animals on L4 and day 1.

(I) Quantification of L4 and day 1 adult animals; n = 46 neurons scored across 2 biological replicates. Points on the box and whisker plots represent intensities of individual neurons. Statistical analysis: two-tailed t tests on each replicate; pooled results shown. ∗∗∗p < 0.001; ∗∗∗∗p < 0.0001. In total, four reporter strains were tested, and the three shown here were significant.

Neuron-specific transcriptional data provide functional information

GPCRs are membrane proteins that serve as receptors for many environmental stimuli. GPCRs bind a wide variety of molecules, from hormones and odors to neurotransmitters²⁰ and neuropeptides.⁵ While other organisms encode 200–400 GPCRs,^21,22,23 the C. elegans genome encodes over 1,300 chemosensory GPCRs (csGPCRs),^24,25 as well as 153 peptidergic GCPRs (pepGCPRs), 16 aminergic, 3 muscarinic, and 19 GPCRs of other families²⁵; the 1,341 C. elegans csGPCRs are broken into three super-families and six solo families.²⁵ The expression of 375 of the csGPCRs (28%) were previously examined using promoter-GFP reporters²⁵; the location and functions of the remaining 72% remain unknown. We detected 742 csGPCRs in specific neurons at a threshold of 1% detection across a cluster and a normalized average expression value of 0.001, more than doubling the known sites of expression. We generated a matrix of the percentage of cells expressing each GPCR across all clusters (Table S4); the ADL neuron expresses the most csGPCRs (286) in adult neurons (Figures 4A and S1A; Tables S4 and S5). Sensory neurons express the highest number of csGPCRs, with the ASJ, ASH, ASK, AWA, ASI, ADF, and AWC expressing dozens to hundreds of receptors (Figure 4A). Chemosensory neurons expressed significantly more csGPCRs than any other neuron types (Figure 4A); by contrast, pepGCPRs are expressed at a lower and more consistent level across all neuron types (Figure 4B; Table S6). In fact, no neuron expresses only one GPCR, while the vast majority (∼77%) express more than 10 GPCRs. There was no significant over- or under-representation of any one family, and our dataset identified 50%–100% of most families (Figure S2B). Our data suggest that many of these sensory neurons and some motor neurons might have the ability to sense information through many receptors simultaneously.

Figure 4.

Functional analysis of GPCRs

(A) Chemosensory GPCR (csGPCR) expression in specific neuron classes, colored by functional type; top 35 csGPCR-expressing neuron classes. Inset: csGPCR expression by neuron functional class.

(B) Peptidergic GPCR (pepGPCR) expression in specific neuron subtypes, colored by functional type; top 39 pepGPCR-expressing neuron classes. Inset: pepGPCR expression by neuron functional class. One-way ANOVA with Tukey’s post hoc analysis; ∗p < 0.05; ∗∗p < 0.01; ∗∗∗∗p < 0.0001.

(C) Heatmap of changes in csGPCRs in chemosensory neurons from L4 to day 1, broken down by GPCR family; overall percentage of the changes in each neuron shown by stacked bar plots. GPCR expression was detected in both L4 and day 1 neurons (gray), neither dataset (white), only in L4 neurons (CeNGEN)¹¹ (blue), or only in day 1 adult neurons (red).

(D) Chemotaxis of adult RNAi-treated unc-119p::sid-1 (RNAi-sensitized) worms to benzaldehyde upon adult-specific RNAi knockdown of the top day 1 AWC-only expressed chemoreceptor genes. Pooled data from at least two biological replicates. ∗∗∗∗p < 0.0001. One-way ANOVA with Bonferroni post hoc analysis.

(E and F) Chemotaxis of adult RNAi-treated unc-119p::sid-1 (RNAi-sensitized) worms to butanone (E) or pentanedione (F) is not affected by srd-5 knockdown.

(G) Chemotaxis of adult RNAi-treated unc-119p::sid-1 (RNAi-sensitized) worms to isoamyl alcohol is significantly reduced by srd-5 knockdown. ∗∗∗p < 0.001.

(H) srd-5 expression across all neuron clusters in our combined dataset; srd-5 is primarily expressed in the AWC (inset).

(I) Representative composite images (GFP+TD) and GFP images of Psrd-5::GFP animals on L4 and day 1; arrow shows srd-5p:GFP expression turning on in the AWC in adults.

(J) Percentage of animals showing expression in the AWC^off and AWC^off+AWC^on in L4 and day 1 adults.

(F–H) Two-tailed t test, representative figure from three biological replicates.

(A, B, and F–H) Boxplots: center line, median; box range, 25th–75th percentiles; whiskers denote minimum–maximum values.

We find that the shifts in gene expression from larvae to adults are distributed across all families of csGPCRs (Figure 4C; Table S7). The ciliated amphid ADL neuron, which is involved in several chemosensory roles, including mediating social feeding behaviors and modulating other chemosensory responses, expresses almost 300 csGPCRs, many of which appear to switch “on” in the shift from L4 to adults, while others switch “off.” In fact, the ADL L4 to day 1 adult transcriptional shift includes at least 88 chemoreceptor genes—11% of the total number of csGPCRs in the worm genome. Our data suggest that shifts in GPCR expression accompany the transition from L4 larvae to day 1 of adulthood, indicating corresponding shifts in behaviors.

Functional analysis of GPCRs in the AWC neuron

We are particularly interested in GPCRs in the ciliated amphid odor-sensing neuron AWC²⁶ because it is required for olfactory associative learning and memory.^7,27 We previously found that a gain-of-function mutation in a highly conserved Gαq protein, EGL-30/GNAQ, expressed specifically in the AWC is sufficient to rejuvenate memory in aged worms,²⁸ and activation of its mammalian homolog, Gnaq, in the hippocampus similarly rescued memory in aged mice.^1,17 Therefore, characterizing the expression of GPCRs in individual neurons may shed light on conserved neuronal functions.

We knocked down the most highly expressed candidate csGPCR genes in adults and tested chemotaxis toward odorants detected by AWC.^7,26,27,28 Benzaldehyde is sensed by both AWC_on and AWC_off; although most RNAi treatments had no effect, knockdown of srd-5 reduced chemotaxis to benzaldehyde (Figure 4D). srd-5 knockdown did not affect chemotaxis to butanone (Figure 4E), a chemical sensed only by the AWC_on, or to pentanedione (sensed by AWC_off) (Figure 4F). However, chemotaxis to isoamyl alcohol (AWC_on and AWC_off) was reduced by srd-5 knockdown (Figure 4G). Our data suggested that the SRD-5 csGPCR is required for detection of odorants sensed by both AWC_on and AWC_off, but not for sensing of odorants solely by the AWC_on (butanone) or AWC_off (pentanedione).

Previously, srd-5 was found to be expressed only in the AWC_off in L4 animals.²⁵ We confirmed that srd-5 expression is limited to one of the AWC neurons in L4 animals (Figures 4H–4J); however, in adults, srd-5 is expressed in both AWC neurons (Figures 4I, 4J, and S3D). This broadened expression upon adulthood correlates with the role of srd-5 in sensing odors that require both AWCs. The identification of the role of SRD-5 in AWC-regulated behaviors demonstrates the utility of profiling and functionally testing the csGPCR landscape in individual neuron types.

Sequencing data provides locations of potential neuropeptide-receptor interactions

The interactions between neuropeptides and their cognate receptors were recently functionally probed in vitro to identify potential extrasynaptic signaling ligand/receptor pairs, describing a neuropeptidergic network that is independent of the connectome of the worm.^3,5 However, these predictions rely on L4 larval neuron expression; whether and where each receptor is expressed in adults is unknown. We identified neuronal sites of potential neuropeptide-receptor interactions; by comparing the neuronal sites of neuropeptide (Table S8) and neuropeptide receptor transcript expression (Table S9), potential pairwise interactions can be proposed. For example, NLP-58 interacts with TKR-1 and TKR-2⁵; nlp-58 is expressed in the OLQ and URA neurons, while tkr-2 is expressed in the ADL neuron, revealing a candidate interaction between the OLQ/URA and ADL neurons (Figure S4C). Although the ADL and OQR are physically connected,²⁹ the URA-ADL peptidergic interaction may bypass the synaptic connectome. Similarly, nlp-12 is expressed in the DVA interneuron, while its cognate receptor, ckr-1,⁵ is expressed in RIS (Figure S4D); these neurons are not thought to be directly connected through chemical synapses or gap junctions, but instead may signal via the NLP-12 neuropeptide. Similar maps of adult neuropeptide-receptor interactions can be built by combining these datasets (Tables S8 and S9), revealing adult neuropeptidome connections that are not obvious from the connectome of the worm.

Single-nucleus daf-2 sequencing reveals heterogeneity of neuronal regulation

The IIS pathway regulates longevity,⁶ dauer formation,³⁰ reproduction,³¹ and behavior,⁷ including cognitive functions.⁷ We previously found that the insulin/IGF receptor mutant daf-2 extends short-term (STAM) and LTAM.^7,8 daf-2 mutants also better maintain motor functions, axon regeneration ability, and memory ability with age.^32,33,34,35 Previously, we performed both whole-worm^36,37 and neuron-specific RNA-seq^8,9 to identify IIS targets, and discovered that the neuron-specific targets of daf-2 are distinct from its whole-body targets.^8,36,37 These neuronal targets regulate phenotypes, including learning, memory, and axon regeneration.⁸ Whether IIS changes gene expression in individual neurons that in turn have effects on specific behaviors is unknown. For example, differential expression specifically in the AWC may correlate with the extended abilities of daf-2 in butanone associative learning and memory.^7,8

We performed snSeq on daf-2 mutants (three biological replicates) to identify differentially expressed neuronal IIS targets that were masked in our previous pan-neuronal RNA-seq analyses.^8,37 We found that 980 and 695 genes were up- or down-regulated, respectively, in at least one cell type (log₂(fold change [FC]) >0.25 or <−0.25, adjusted p < 0.05; Figures 5A and S5A; Table S10). Although our single-nucleus dataset is enriched in canonical and previously identified daf-2 targets^8,37 (Figures 5A and S5B), we were surprised to find that about two-thirds are previously unidentified daf-2 targets (Figure 5A).

Figure 5.

Genes differentially expressed between daf-2 and N2 vary across cell types

(A) Previously and newly identified daf-2 upregulated (678) and downregulated (419) genes; three biological replicates of WT and daf-2 samples.

(B) Euclidean distance of each neuron’s mean daf-2 vector and mean N2 vector; top 40 neuron subtypes with the largest distance shown (full graph shown in Figure S5B). Inset: average Euclidean distance of each neuron class. One-way ANOVA with Tukey’s post hoc analysis.

(C and D) Gene interaction network based on daf-2 vs. N2 upregulated and downregulated genes. The shorter length of the edges correlates with a stronger connection.

(E) KEGG pathway gene set enrichment analysis (GSEA)-normalized enrichment scores across different cell types. Empty boxes indicate insufficient overlap between the differentially expressed genes in that cluster and the KEGG pathway gene set to calculate the GSEA enrichment score. Red: pathway enriched in daf-2-upregulated genes; blue: pathway enriched in daf-2-downregulated genes.

(F and G) Differential expression of selected genes across neuron subtypes. Colors indicate the direction of differential expression fold change, and dot size indicates statistical significance from the Wilcoxon rank-sum test. (F) Differentially expressed genes of daf-2-regulated targets present in previous studies.^8,32,37 (G) Genes newly identified to be differentially expressed in daf-2 neurons.

(H) Naive preference was determined using choice assays to OP50 versus PA14 bacteria. Each dot represents an individual choice assay plate (average 60 worms per plate). Unpaired, two-tailed Student’s t test. Data from four biological replicates are shown.

(I) Neuron-type expression of insulin-like peptides.

(J) Neuron-type expression of receptor-type guanylyl cyclases.

(K) Average number of daf-2-upregulated GPCRs in each neuronal class. One-way ANOVA with Tukey post hoc analysis. The full comparison is shown in Figure S10A.

(L) Neuron-subtype expression of chemoreceptors that are upregulated in daf-2 neurons. The full heatmap of GPCR expression is shown in Figure S9B.

(B, H, K) Boxplots: center line, median; box range, 25th–75th percentiles; whiskers denote minimum–maximum values. ∗∗p < 0.01; ∗∗∗p < 0.001; ∗∗∗∗p < 0.0001.

To determine whether some neurons are more affected by daf-2 mutation than others, we calculated the Euclidean distance of each neuron cluster between WT (N2) and daf-2 (Figure 5B; Table S11). Sensory neurons, especially chemosensory neurons (e.g., ASK, ADF, ADE, ASH, ASJ) are generally more distant between daf-2 and N2 (Figure 5B, inset) compared to other neuron types, consistent with the notable changes in the sensory behaviors of daf-2.^7,38,39 In particular, gene expression in the ASK neuron, which has roles in multiple types of sensory behaviors, is the most different between daf-2 and N2 (Figures 5B and S5B); this difference is also reflected in the principal-component analysis (PCA) plot of ASK (Figure S5C). Some motor neurons also exhibit large differences between N2 and daf-2, consistent with extended motor behaviors in aged daf-2 worms.^33,40 We also compared the number of differentially expressed genes in each neuronal cluster. The number of differentially expressed genes roughly correlates with the number of cells in a cluster, because more cells lead to more genetic material and more sequencing depth. However, the “peaks” in the graph, which correspond to neuron types with larger changes compared with clusters with a similar size, correspond with the clusters with a large N2 vs. daf-2 distance (e.g., ASK, PDB, ASER, AIM, ASG, ASEL) (Figure S5D). We also examined the correlation between the differential expression of each neuron cluster; similar neuron types share highly correlated daf-2-regulated differential transcriptomic changes (Figure S6A).

Global analyses reveal new pathways

We performed gene network analysis by calculating the regulatory strength between all daf-2 significantly differentially expressed genes using tree-based ensemble methods (GENIE3⁴¹) to investigate the co-regulation and interaction of these differentially expressed genes (Figure S7). Several insulin-signaling-related genes form an interaction network with daf-2, including insulin-like peptide genes (ins-6, ins-9, ins-24), a forkhead transcription factor (fkh-7), and the skn-1/Nrf transcription factor (Figure 5C). Interestingly, there are also other genes in this hub that have not been previously associated with insulin signaling, such as the NAD⁺ ADP-ribosyltransferase tank-1/pme-5, the fatty acid (polyketide) synthase pks-1, the serpentine receptor srm-1, a protein tyrosine phosphatase receptor (T13H5.1), and an elongation factor (K10C3.5). PME-5 is involved in the C. elegans apoptosis DNA damage response pathway.⁴² K10C3.5 expression is induced by UV exposure in daf-2 mutants, and its induction is DAF-16 dependent.⁴³ K02D3.1 is significantly downregulated in daf-16 mutants⁴⁴ and upregulated in skn-1⁴⁵ mutant background. Our network analysis also suggested that mechanosensory genes as well as an epsilon class serpentine receptor (sre-45) are differentially expressed in daf-2 mutants (Figure 5D). Together, these data suggest that there are daf-2-regulated neuronal functions that have not yet been identified, and these data may serve as a resource for building hypotheses.

Kyoto Encyclopedia of Genes and Genomes (KEGG) analysis of daf-2-regulated genes revealed that FOXO, MAPK, mTOR signaling, autophagy, and other longevity-related pathways are upregulated in most neurons, consistent with canonical IIS/FOXO pathway functions (Figures 5E and S8A; Table S12). By contrast, purine metabolism, calcium signaling, and inositol-phosphate metabolism pathways—all associated with neuronal activity—were downregulated in most neurons. Although decreased activity seems surprising, neuronal hyperactivity has been linked with cognitive decline and neurodegeneration^46,47; it is possible that the downregulation of calcium and inositol signaling genes might reduce hyperactivation with age. The AGE-RAGE signaling pathway, which is involved in binding advanced glycation end damage products produced by the Maillard reaction and is associated with diabetes,⁴⁸ is also downregulated in select neurons. Some pathways are more heterogeneous, such as ribosomal genes, which were upregulated in daf-2 in some neurons (e.g., LUA, AIY, AVL) and downregulated in others (e.g., RIC, AVB, RMF/SMB). Ribosomal biogenesis is universally downregulated in longevity mutants to reduce energy expenditure,^49,50 but our data suggest that this global downregulation could have neuron-specific exceptions.

snSeq analysis reveals daf-2 differential regulation in individual neurons

Newly identified IIS targets may be genes that are consistently differentially expressed across a large set of neurons that were not previously detected by other methods or may have large FCs in a small set of neurons. We observe that canonical daf-2-upregulated (class 1) genes (e.g., mtl-1, fkh-7, dod-6, lys-7) and downregulated (class 2) genes (e.g., ilys-5, ctc-1, nduo-2) were differentially expressed across many neurons (Figure 5F), consistent with our previous pan-neuronal results,⁸ but also exhibit neuron-specific differential expression (e.g., fkh-7 in ASER, dod-6 in ASEL, igeg-2 in AVH; Figure S8B). By contrast, several newly identified daf-2-regulated genes (e.g., flp-16, zip-5, maco-1, odr-2, slo-2, nlp-46) were differentially expressed in only a few neurons (Figure 5G). These newly identified daf-2-regulated genes control many aspects of the nervous system. For example, ODR-2 regulates odor sensation and pheromone imprinting,^26,51 mec-7 encodes a β-tubulin gene that regulates mechanosensory neuron development and touch sensitivity,⁵² and ZIP-5 regulates axon regeneration³⁵ and pathogen-induced pheromone responses.⁵³ The extended health and behaviors of daf-2 animals may result from such single neuron-specific gene changes that were previously masked. In fact, we found that over half were differentially expressed in a single neuron, and over 98% of genes were differentially expressed in fewer than 30 neuron types (Figure S8C). That is, the expression of a particular gene may differ in only a handful of neurons, making it difficult to discover from previous methods, even if that gene expression difference in a particular neuron is critical for functional differences. Other genes are significantly upregulated in one neuron but downregulated in another; thus, its net differential expression in previous bulk analyses might be negligible. For example, daf-7 is upregulated in daf-2 ASJ, ASG, and ADE neurons, but downregulated in ASI and ASER. Because increased daf-7 expression in either ASJ⁵⁴ or ASI⁵⁵ is sufficient to induce a switch from attraction to PA14 to avoidance, we wondered whether daf-2 mutants might have altered chemotaxis to PA14. Indeed, we find that daf-2 mutants show a higher avoidance of PA14 (Figure 5H).

Insulin-like peptides, particularly ins-17 (AIZ), ins-24 (ASI), and ins-6 (ASI), were differentially expressed in specific chemosensory and interneurons of daf-2 mutants (Figure 5I). Similarly, receptor-type guanylyl cyclases were differentially expressed specifically in the ASEL (gcy-7), ASER (gcy-1, -3, -4, -13), ASG (gcy-15), and AFD (gcy-23, gcy-29) sensory neurons in daf-2 mutants (Figure 5J).

To identify daf-2-differentially expressed GPCRs, we plotted the expression (Figure S9A) and differential expression (Figure S9B) of csGPCRs, pepGCPRs, and other GPCRs in daf-2 neurons. While GCPRs are widely expressed (Figure S9A), they are selectively differentially expressed only in a subset of neurons (Figure S9B). csGPCRs are generally upregulated in daf-2 mutants (Figures S10A and S10B), which is also true for pepGCPRs (Figures S10A and S10B). Chemosensory neurons on average have the most upregulated GPCRs (∼5 per neuron), significantly higher than other neuron types (Figure 5K). Chemosensory neurons are the site of 76% of all upregulated csGPCRs (Figure S10B); many csGPCRs were upregulated in daf-2 in a single sensory neuron type—for example, srj-18 in the ASK and str-144, str-130, and str-50 in the AWC neuron (Figure 5L). The fact that most of these GPCRs are upregulated in daf-2 mutants (Figure 5K) suggests that daf-2 neurons may have a unique advantage over WT in chemosensory-related behaviors. Other GPCRs are also differentially expressed in daf-2 mutants, but in various neuron types (Figure S9B). Some of these GPCRs have important neurotransmitter signaling functions, including dopamine receptor dop-3 (ASK, AVK) and glutamate receptor mgl-3 (ASK), serotonin receptor ser-4 (ASI and AWA), octopamine receptors ser-3 (AVK) and ser-6 (ADE), and tyramine receptors ser-2 (DVA) and tyra-2 (AWC) (Figures S9A and S9B).

Genes that normally decline with age are upregulated in daf-2 neurons

daf-2 mutants extend cognitive and neuronal behaviors,^7,28,35,38 some of which may depend on gene expression changes in individual neurons. We previously used pan-neuronal transcriptional profiling to identify genes whose expression declines with age.³² These genes are enriched in neuronal and synaptic function,^32,56 and their downregulation correlates with age-dependent behavioral deficits. With the exception of the ubiquitously downregulated lysozyme ilys-5 (Figure S11A) and the dietary restriction overexpressed droe-4 (AWA), most of these genes are upregulated in daf-2 mutants (Figure S11A). While the ubiquitously upregulated genes include collagens,⁹ the DAF-16 target dct-16,⁵⁷ a nematode allergen (npa-1), and the cpr-9 and asp-3 proteases, the majority of these age-downregulated genes are upregulated only in a small subset of daf-2 neurons. For example, several mec genes are specifically upregulated in the mechanosensory touch neurons, while insulins, neuropeptides (nlp-, flp-), neuropeptide receptors (pdf-1, pdf-2, tkr-3), chemosensory genes (odr-2, hot-1), ion channels, transmembrane transporters, and synaptic transmission genes are upregulated in specific neurons (Figure S11A). Genes that decline with age in WT neurons³² are enriched in all of the daf-2 neurons (Figure S11B). Unbiased promoter analysis revealed the FOXO/DAF-16 binding element, suggesting that many of the daf-2-regulated genes are direct DAF-16 targets⁵⁸ (Figure S11C; Table S13).

Promoter::GFP analysis validates daf-2-upregulated gene expression results

To test our snSeq neuron-specific predictions, we imaged promoter::GFP strains for daf-2-regulated genes that fall into different categories. For example, the insulin-like peptide ins-6 is a class I gene^36,37 that is significantly upregulated in daf-2 in bulk neuronal sequencing⁸ and has been previously implicated in olfactory learning⁵⁹ and dauer entry.^60,61 Our snSeq data suggested that ins-6 is significantly upregulated in ASI and ASJ neurons (Figure 6A); imaging of Pins-6::GFP in WT and daf-2 mutant backgrounds confirmed this expression pattern, validating our snSeq data (Figures 6B and 6C).

Figure 6.

Promoter-GFP analysis of genes that are differentially expressed in daf-2 neurons

(A) ins-6 is upregulated in daf-2 neurons in the ASI and ASJ neurons.

(B) Representative images of ins-6p::GFP expression in day 1 WT (PHX2685) and daf-2 (CQ820) adult worms. Arrows: ASJ and ASI neurons.

(C) Quantification of ASI and ASJ cell bodies in ins-6p::GFP worms; n = 51 and 54, respectively.

(D) flp-5 expression is upregulated in daf-2 RMG neurons.

(E) Representative images of flp-5p::GFP expression in day 1 WT (NY2049) and daf-2 (CQ821) adult worms. Arrow: RMG neuron cell bodies.

(F) Quantification of RMG cell bodies in flp-5p::GFP worms; n = 32 and 41, respectively.

(G) nlp-21 is upregulated in daf-2 PDA/AS/DA neurons.

(H) nlp-21p::GFP expression in day 1 WT (KG2430) and daf-2 (CQ822) adult worms. Arrows: ventral nerve cord neurons AS or DA.

(I) Quantification of AS/DA cell bodies in nlp-21p::GFP; n = 46 and 39, respectively.

(J) Y39G8B.7 expression is upregulated in daf-2 AWA neurons.

(K) Y39G8B.7p::GFP expression in day 1 WT (CQ807) and daf-2 (CQ819) adult worms. Arrow: AWA neuron cell bodies.

(L) Quantification of AWA cell bodies in Y39G8B.7p::GFP worms; n = 69 and 74, respectively.

(M) flp-21 is upregulated in daf-2 RMG neurons.

(N) Representative images of flp-21p::GFP expression in day 1 WT (PHX3212) worms treated with control (pAD12) or daf-2 RNAi. Arrows: ASJ and ASI neurons.

(O) Quantification of RMG cell body GFP presence in flp-21p::GFP worms. p value obtained from Fisher’s exact test; n = 50 and 63, respectively. All five strains tested for promoter analysis yielded significant results and are shown here.

(A, D, G, J, and M) SCT-normalized expression level shown. Adjusted p value: Wilcoxon’s rank-sum test methods. (B, E, H, K, and N) Error bar: 20 μm. (C, F, I, and L) ∗∗∗∗p < 0.0001. p value obtained from Student’s t test. All fluorescence intensity quantified from 12-bit or 16-bit maximum projection images (FIJI).

Other genes are expressed ubiquitously, but only differentially expressed or oppositely expressed in daf-2 mutants in a small subset of neurons or specific neurons. For example, the FMRF-like peptide gene flp-5 is expressed in several neurons; its expression in mechanosensory neurons promotes food searching,⁶² and was downregulated in whole-worm³⁷ and bulk neurons.⁸ Our transcriptome data suggested that flp-5 expression is significantly upregulated by daf-2 specifically in the RMG neuron (Figure 6D), which we confirmed in daf-2;flp-5p::GFP worms (Figures 6E and 6F).

Still other genes were not detectable in previous bulk pan-neuronal sequencing approaches, but were revealed by snSeq. For example, bulk sequencing did not show nlp-21 as significantly upregulated in daf-2 neurons,⁸ but we detected nlp-12 upregulation in the PDA/AS/DA cluster (Figure 6G). NLP-21 is located in motor neuron-dense core vesicles, and is affected by Rab2 interactors rund-1 and cccp-1.⁶³ We found that daf-2 nlp-12p::GFP ventral cord neurons are significantly more fluorescent than WT (Figures 6H, 6I, and S12D). Other genes were not known previously to be expressed in neurons. A class I gene³⁶ of unknown function, Y39G8B.7, was not previously reported to be expressed in neurons¹¹ or differentially expressed in daf-2 neurons.⁸ Y39G8B.7 is required for the lifespan extension of the daf-2;set-21 double mutant compared to daf-2, indicating its expression may be required for histone methylation functions.⁶⁴ Our snSeq data suggested that Y39G8B.7 is not expressed in WT neurons, but its expression increases in AWA neurons of daf-2 mutants (Figure 6J); Y39G8B.7p::gfp fluorescence confirms this expression change in the AWA of daf-2 mutants (Figures 6K and 6L).

The FMRF-like peptide FLP-21, which plays roles in nociceptive heat avoidance,⁶⁵ dauer entry,⁶⁶ and behavioral quiescence,⁶⁷ was not significantly upregulated in daf-2 bulk neuronal sequencing,⁸ but snSeq indicates upregulation in RMG neurons of daf-2 (Figure 6M). We detected flp-21p::NLS::GFP fluorescence in the RMG neuron in fewer than 20% of WT worms, but daf-2 RNAi increased this expression detection level to over 50% (Figures 6N and 6O). Together, these promoter:gfp results validate our snSeq data for different classes of daf-2-regulated genes.

AWC-specific daf-2 targets function in butanone sensation and associative memory

The AWC chemosensory neurons sense butanone, benzaldehyde, and isoamyl alcohol and are required for butanone appetitive associative learning and memory enhancement in daf-2 mutants.^7,26,27,28 The AWC transcriptome is significantly enriched in the expression of ion transport, protein kinase, and ATP- and guanosine triphosphate (GTP)-binding activities, indicating its role in signal transduction (Figure 7A).

Figure 7.

AWC-specific daf-2-regulated genes regulate chemosensation and cognitive functions

(A) Significant Gene Ontology (GO) terms for AWC-enriched genes in the WT and daf-2 AWC combined dataset (log₂[FC (AWC/total)] > 0.25, adjusted p < 0.05, Wilcoxon rank-sum test).

(B) GO terms of AWC-specific daf-2-upregulated genes.

(C) GO terms of AWC-specific daf-2-downregulated genes.

(B and C) |daf-2 vs. N2 log₂(FC)| >0.25, adjusted p < 0.05, Wilcoxon rank-sum test. GO terms from WormCat 2.0.⁶⁸p values of the GO terms were Bonferroni corrected.

(D) Expression distribution of AWC-upregulated daf-2 genes. Expression level density of (gray) N2 or (red) daf-2 AWC cells. Adjusted p value: Wilcoxon rank-sum test. Bold: with identified mammalian homologs.

(E) daf-2 and N2 short-term associative memory (STAM) results (day 6 shown). Representative figure from three biological replicates. Error bar: SEM. Two-way ANOVA with Sidak’s post hoc analysis.

(F) Feature Plot expression level of sre-19. Scale bar: SCT normalized expression level. Inset 1: sre-19 expression in the AWC. Inset 2: sre-19 expression distribution in N2 and daf-2 AWC neurons.

(G) STAM of day 5 neuronal-RNAi-sensitive daf-2 worms treated with adult-specific sre-19 RNAi. Representative figure from three biological replicates. ∗∗p = 0.0098; two-way ANOVA with Sidak’s post hoc analysis. Error bars: SEM.

(H and I) Two-hour STAM of day 5 neuronal RNAi-sensitive daf-2 worms treated with adult-specific control RNAi (L4440 vector) or candidate gene RNAi. daf-16 RNAi as a positive control. Each dot represents a chemotaxis plate (average 150 worms per plate). Representative image of two biological replicates. One-way ANOVA with Dunnett’s post hoc analysis. Boxplots: center line, median; box range, 25th–75th percentiles; whiskers denote minimum–maximum values.

(J) Mammalian homologs of daf-2-upregulated genes in the AWC. Mammalian homologs obtained from Panther. Gene description obtained from Alliance of Genomes.

∗p < 0.05; ∗∗p < 0.01; ∗∗∗p < 0.001; ∗∗∗∗p < 0.0001; ns, not significant.

Our snSeq data suggest that daf-2 AWC neurons differ significantly from WT in the expression of GTPases, guanine nucleotide exchange factors (GEFs), transcription factors, and transmembrane transport and oxidoreductase activities (Figures 7B and 7C; Table S14). Upregulated genes include seven-transmembrane, serpentine, and tyramine receptors; neprilysin; synaptic proteins; axon guidance proteins; potassium voltage-sensitive channels; the ODR-2 GPI-linked signaling protein; signaling proteins; the lin-42 microRNA regulator; VH1-related genes; transcription factors; and the Dietary Restriction OverExpressed gene droe-4 (Figure 7D; Table S10).

The AWC is the center of butanone olfactory learning and memory.^28,69 daf-2 worms have significantly higher STAM than WT in both young and older animals⁷ (Figures 7E and S13B). The expression of the serpentine receptor gene sre-19 is confined to the AWC neurons (Figure 7F); therefore, we tested whether it is required for improved memory of daf-2. To avoid developmental defects, we reduced gene function in RNAi-sensitive worms solely in adulthood. We performed the STAM assay on day 5, when WT worms have no memory, but daf-2 memory is still intact. Although adult-only reduction of sre-19 did not affect motility or the ability to sense butanone (Figures S13C and S13D), sre-19 was indeed required for the extended STAM of aged daf-2 (Figure 7G).

We also selected top daf-2-upregulated AWC-expressed genes (Figure 7D) to test their requirement in the extended memory of daf-2 (Figures 7H and 7I). Most of these genes, as well as others in the daf-2 AWC-upregulated set, have conserved mammalian homologs (Figure 7J) and may play important roles in cellular signaling and regulation in memory formation and storage. Seven of the 12 tested candidates (nep-26, vhp-1, odr-2, zip-5, dmd-10, unc-73, and tyra-2) are required for extended STAM of aged daf-2 worms (Figures 7H and 7I). Our data suggest that not only are we able to reliably recapitulate gene expression changes in specific neurons but also that the changes in expression that we have identified in specific neurons contribute to the changes in daf-2 behavior.

Discussion

Here, we used single-nucleus sequencing to transcriptionally characterize the neurons of C. elegans in WT and daf-2 adults. Our results indicate that snSeq identifies targets in specific neurons that were masked in whole-worm, bulk neuron-specific, or low-depth sequencing. Because L4 larvae differ from adults in neuron-regulated behaviors, the identification of neuron-specific transcriptional differences helps us uncover the causes of these differences. For example, we find that L4 larvae cannot carry out butanone associative learning behavior, while day 1 adults can, and most genes that we previously identified in studies of learning and memory are only expressed after the transition from larvae to adulthood. Moreover, the transcriptomes of adult neurons reveal potential functional similarities that are not obvious from lineage or morphological data.

Transcriptional changes of a single gene within an individual neuron can be detected using snSeq, allowing us to better describe the expression profiles of the C. elegans large family of GPCRs. While the expression of 28% of these GPCRs was already known, we determined the site of expression of an additional 367 GPCRs. Remarkably, most neurons express not just one, but multiple GPCRs, and some sensory neurons express dozens to hundreds of these receptors, suggesting that these neurons have the potential to sense many inputs simultaneously. For example, the ciliated amphid sensory neuron ADL expresses more than 250 GPCRs; how these receptors respond to and integrate this vast number of inputs, and change in different mutant backgrounds, will be interesting to decipher. We also used this information to identify and test candidate GPCRs for AWC-neuron-related chemosensory functions and to identify potential neuronal sites of candidate neuropeptide-receptor interactions. The SRD-5 csGPCR, for example, is necessary specifically for benzaldehyde and isoamyl alcohol chemotaxis, but not for other AWC sensory functions. Our data also provide expression data for candidate neuropeptide-receptor interactions, helping us to better understand the adult peptidergic signaling connectome.

snSeq is also a powerful tool for identifying single-neuron differential gene expression changes in mutants, which cannot be inferred from existing datasets of WT neurons. Changes in single neurons that were masked in whole-worm, pan-neuronal, and even single-cell analyses of daf-2 mutants are identifiable in our snSeq data. For example, we found that over half of daf-2-regulated genes were only differentially expressed in one neuron type and were identified as neuronal daf-2 targets for the first time here. Other genes have surprisingly restricted expression changes; for example, the serpentine receptor sre-19 was expressed only in the AWC neuron, yet it is required for the ability of daf-2 to extend STAM. Likewise, insulin-like peptides have restricted differential expression; ins-4 was downregulated only in ASI and VC neurons in daf-2 mutants, which corresponds to its function in pheromone-mediated learned avoidance change.⁷⁰ ins-6 was upregulated in daf-2 ASI and ASJ neurons, which correlates with its function in aversive memory formation in ASI.⁵⁹ We also found that guanylyl cyclases, which function in salt chemotaxis, olfaction, CO₂ sensation, and thermotaxis, were differentially expressed in daf-2 only in the sensory neurons mediating these responses, correlating with the altered behaviors of daf-2.^71,72 Most of these expression changes were not obvious in bulk RNA-seq, likely because the changes are restricted to a single or small number of neurons, or in some cases might have been masked by decreases in other cells, indicating that deep, neuron-specific snSeq provides the resolution to identify IIS targets that might have been masked in pan-neuronal, whole-worm, or single-cell analyses. Thus, snSeq is a powerful tool for identifying genes that only change in a subset of neurons and would be masked in tissue-specific or whole-worm analyses.

We also detected new AWC-specific daf-2-upregulated genes that are required for AWC-mediated learning and memory. Of these genes, several were known to have neuronal functions, but they had not been previously associated with AWC or daf-2 phenotypes. For example, ZIP-5, the worm homolog of the mammalian transcription factor CEBP, functions in axon regeneration of daf-2 worms,³⁵ but here we found that is also required for STAM. ODR-2 is important for AWC chemosensory functions⁷³ and was implicated in odor imprinting,⁵¹ but it was not previously known to be required for the extended memory of daf-2 (Figures 5F and 7H). The transcription factor DMD-10/DMRT is required for ASH-mediated osmolarity avoidance,⁷⁴ but its requirement in the AWC neuron had not been previously shown. Likewise, the TYRA-2 tyramine receptor/mammalian adrenergic receptor ADRA2, which is required for imprinted memory of pathogen avoidance,⁷⁵ is also required for the extended memory of daf-2 (Figures 7D and 7I). The Rho GEF UNC-73/ARHGEF25 regulates axon guidance, motility, and learning in C. elegans, and its homolog Trio is required for normal spatial learning in mice, but its connection with the insulin signaling pathway was not previously described.^28,76 The role of UNC-73 in the memory extension of daf-2 echoes our recent finding that axon guidance factors are required for the extended memory of activated G_αq proteins in aged worms and mice.¹⁷ These genes may promote structural and functional integrity of the neuron and facilitate metabolic processes and signal transport, improving these behaviors relative to WT, and maintaining these behaviors with age.

Our data further describe the expression landscape of IIS differentially expressed genes and link them to new roles downstream of daf-2 to regulate learning and memory. Our sequencing results provide new deep transcriptional data for individual adult C. elegans WT neurons, and our functional testing of more than two dozen genes revealed functions for these genes in chemotaxis, learning, and memory. These results suggest that we can use snSeq to identify cell-type-specific changes in adult mutants to uncover new conserved candidates that regulate sensory and cognitive behaviors.

Limitations of the study

While we found that sequencing three biological samples of each genotype was sufficient to resolve most of the C. elegans adult neuron types (107 of 119), a few remained unresolvable. Additionally, we can only assess genes that we were able to measure in each cell type; we cannot comment on genes that may have been below the threshold of detection.

Resource availability

Lead contact

Further information and requests will be fulfilled by the lead contact, Coleen T. Murphy (ctmurphy@princeton.edu).

Materials availability

Strains generated in this study are available upon request. This study did not generate new unique reagents.

Data and code availability

snRNA-seq data have been deposited at GEO and are publicly available as of the date of publication (PRJNA1027859). All other data are available in the main text or supplemental information. Any additional information required to reanalyze the data reported in this paper is available from the lead contact upon request. No new code was generated.

Acknowledgments

We thank Christina DeCoste and the Princeton FACS Core, Jennifer Miller, Jean Arly Vomar, Wei Wang, and the Princeton Genomics Core for their assistance, the C. elegans Genetics Center for strains, and members of the Murphy lab for suggestions on the manuscript. Funding was obtained from the Simons Foundation (811235SPI to C.T.M.), the China Scholarship Council (to Y.W.), NSF GRFP (DGE-2039656 to J.S.A.), the National Institutes of Health grant F32 Fellowship (AG079490 to M.E.S.), and the NIH Office of the Director Pioneer Award (NIGMS DP1GM119167 to C.T.M.).

Author contributions

Conceptualization: R.K., J.S.A., Y.W., and C.T.M. Methodology: R.K., J.S.A., and M.E.S. Investigation: Y.W., J.S.A., R.K., M.E.S., R.S.M., and S.Z. Visualization: J.S.A. and Y.W. Funding acquisition: C.T.M. Project administration: C.T.M. Supervision: R.K. and C.T.M. Writing – original draft: Y.W., J.S.A., and C.T.M. Writing – review & editing: Y.W., J.S.A., and C.T.M.

Declaration of interests

The authors declare no competing interests.

STAR★Methods

Key resources table

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Bacterial and virus strains
OP50 E. Coli (BactoBeads)	Sigma-Aldrich	Cat. #S29021
HT115 E coli	Caenorhabditis Genetics Center	HT115
P. aeruginosa: PA14	Tan et al.77	PA14
Chemicals, peptides, and recombinant proteins
2-Butanone, 99+%, extra pure	Acros Organics	Cat. #149670250
Isoamyl Alcohol	Millipore Sigma	Cat. #W205702
Benzaldehyde	Millipore Sigma	Cat. #B1334-100G
2,3-Pentanedione	Millipore Sigma	Cat. #241962
Molecular Probes Hoechst 33342	Thermo Fisher	Cat. #H3570
Sigma Protector RNase inhibitor	Sigma-Aldrich	Cat. #3335402001
1,1′-Dioctadecyl-3,3,3′,3′-tetramethylindocarbocyanine perchlorate (DiI)	Sigma-Aldrich	Cat. #41085-99-8
Critical commercial assays
10X Genomics Chromium X system using the Single Cell 3′ v3.1 Reagent Kits	10X Genomics	Cat. #1000
Illumina Tagment DNA Enzyme and Buffer kit	Illumina	–
Deposited data
Single nucleus RNA-seq data	This Paper	NCBI BioProject: PRJNA1027859
Experimental models: Organisms/strains
C. elegans strain N2 var. Bristol: wild type	Caenorhabditis Genetics Center	RRID:WB-STRAIN:WBStrain00000003
C. elegans strain CB1370: daf-2(e1370) III	Shen Lab	RRID:WB-STRAIN:daf-2(e1370)
C. elegans strain CQ757: wqIs7 [Prgef-1::his-58::GFP]	This paper	CQ757
C. elegans strain CQ758: wqIs7 [Prgef-1::his-58::GFP];daf-2(e1370) III	This paper	CQ758
C. elegans strain LC108: vIs69 [pCFJ90(Pmyo-2::mCherry +Punc-119::sid-1)] V	Calixto et al.78	LC108
C. elegans strain CQ745: daf-2(e1370) III;vIs69 [pCFJ90(Pmyo-2::mCherry +Punc-119::sid-1)] V	This paper	CQ745
C. elegans strain OH13859: otEx6423 [srd-5p::GFP +pha-1(+)]	Vidal et al.25	OH13859
C. elegans strain CQ806: oyls44 [odr-1::RFP]; otEx6423 [srd-5p::GFP]	This Paper	CQ806
C. elegans strain CQ823 wqEx99[Psrz-64::GFP + Pmyo-3::mCherry]	This Paper	CQ823
C. elegans strain CQ824 wqEx100 [PY65B4A.4::GFP + Pmyo-2::mCherry]	This Paper	CQ824
C. elegans strain CQ825 wqEx101 [PM04C7.4::GFP + Pmyo-2::mCherry]	This Paper	CQ825
C. elegans strain PHX2685: ins-6(syb2685[ins-6::T2A::3xNLS::GFP]) II	Sun and Hobert79	PHX2685
C. elegans strain NY2049: ynIs49 [flp-5p::GFP] V	Li Lab	NY2049
C. elegans strain KG2430: ceIs56 [unc-129p::ctns-1::mCherry + nlp-21p::Venus + ttx-3p::RFP]X	Edwards et al.80	KG2430
C. elegans strain CQ807: wqEx98 [Y39G8B.7p::GFP +myo-3p::mCherry]	This Paper	CQ807
C. elegans strain CQ819: daf-2(e1370) III; wqEx98[Y39G8B.7p:GFP + myo-3p:mCherry]	This Paper	CQ819
C. elegans strain CQ820: ins-6(syb2685[ins-6::T2A::3xNLS::GFP]) II, daf-2(e1370) III	This Paper	CQ820
C. elegans strain CQ821: daf-2(e1370) III, ynIs49 [flp-5p::GFP] V	This Paper	CQ821
C. elegans strain CQ822: daf-2(e1370) III, ceIs56 [unc-129p::ctns-1::mCherry + nlp-21p::Venus + ttx-3p::RFP]X	This Paper	CQ822
Recombinant DNA
Plasmid pL4440 RNAi	Addgene	RRID:Addgene_1654
Plasmid: pL4440-srd-5 RNAi	Ahringer RNAi Library	srd-5
Plasmid: pL4440-sre-19 RNAi	Ahringer RNAi Library	sre-19
Plasmid: pL4440-srx-50 RNAi	Ahringer RNAi Library	srx-50
Plasmid: pL4440-srt-42 RNAi	Ahringer RNAi Library	srt-42
Plasmid: pL4440-str-28 RNAi	Ahringer RNAi Library	str-28
Plasmid: pL4440-str-130 RNAi	Ahringer RNAi Library	str-130
Plasmid: pL4440-str-144 RNAi	Ahringer RNAi Library	str-144
Plasmid: pL4440-str-244 RNAi	Ahringer RNAi Library	str-244
Plasmid: pL4440-srab-16 RNAi	Ahringer RNAi Library	srab-16
Plasmid: pL4440-srsx-5 RNAi	Ahringer RNAi Library	srsx-5
Plasmid: pL4440-srx-2 RNAi	Ahringer RNAi Library	srx-2
Plasmid: pL4440-srd-12 RNAi	Ahringer RNAi Library	srd-12
Plasmid: pL4440-srt-59 RNAi	Ahringer RNAi Library	srt-59
Plasmid: pL4440-srt-24 RNAi	Ahringer RNAi Library	srt-24
Plasmid: pL4440-srt-59 RNAi	Ahringer RNAi Library	srt-59
Plasmid: pL4440-zip-5 RNAi	Ahringer RNAi Library	zip-5
Plasmid: pL4440-odr-2 RNAi	Ahringer RNAi Library	odr-2
Plasmid: pL4440-vhp-1 RNAi	Ahringer RNAi Library	vhp-1
Plasmid: pL4440-nep-26 RNAi	Ahringer RNAi Library	nep-26
Plasmid: pL4440-dod-6 RNAi	Ahringer RNAi Library	dod-6
Plasmid: pL4440-snt-2 RNAi	Ahringer RNAi Library	snt-2
Plasmid: pL4440-sre-19 RNAi	Ahringer RNAi Library	sre-19
Plasmid: pL4440-daf-16 RNAi	Ahringer RNAi Library	daf-16
Plasmid: pL4440-dmd-10 RNAi	Ahringer RNAi Library	dmd-10
Plasmid: pL4440-tyra-2 RNAi	Ahringer RNAi Library	tyra-2
Plasmid: pL4440-unc-73 RNAi	Ahringer RNAi Library	unc-73
Plasmid: pL4440-ags-3 RNAi	Ahringer RNAi Library	ags-3
Plasmid: pAD12 RNAi	Kenyon Lab	RRID:Addgene_34832
Plasmid: pAD12-daf-2 RNAi	Kenyon Lab	daf-2
Software and algorithms
GraphPad Prism version 8.0 or 9.0	GraphPad Software	https://www.graphpad.com
Cell Ranger version 7.1.0.	10X Genomics	https://support.10xgenomics.com/single-cellgene-expression/software/downloads/latest
R software for statistical computing v4.0.2	R Core Team, 2022	https://www.r-project.org/
Seurat: R toolkit for single cell genomics	Satija Lab	https://satijalab.org/seurat/
Meme Suite: Motif based sequence analysis	Bailey and Grant81	https://meme-suite.org/meme/tools/sea
fGSEA: Fast Gene Set Enrichment Analysis	Korotkevich et al.82	https://bioconductor.org/packages/release/bioc/html/fgsea.html
AUCell: Analysis of 'gene set' activity in single-cell RNA-seq data	Aibar et al.83	https://bioconductor.org/packages/release/bioc/html/AUCell.html
Cytoscape: a software environment for integrated models of biomolecular interaction networks	Shannon et al.84	https://cytoscape.org/
GENIE3: GEne Network Inference with Ensemble of trees	Aibar et al.83	https://bioconductor.org/packages/release/bioc/html/GENIE3.html

Experimental model and study details

Animal growth and maintenance

All strains were maintained at 20C for the duration of our experiments. Animals were maintained on plates made with high growth medium (HGM: 3 g/L NaCl, 20 g/L Bacto-peptone, 30 g/L Bacto-agar in distilled water, with 4 mL/L cholesterol (5 mg/mL in ethanol), 1 mL/L 1M CaCl2, 1 mL/L 1M MgSO4, and 25 mL/L 1M potassium phosphate buffer (pH 6.0) added to molten agar after autoclaving). All assays were performed on plates made with standard nematode growth medium (NGM: 3 g/L NaCl, 2.5 g/L Bacto-peptone, 17 g/L Bactoagar in distilled water, with 1 mL/L cholesterol (5 mg/mL in ethanol), 1 mL/L 1M CaCl2, 1 mL/L 1M MgSO4, and 25 mL/L 1M potassium phosphate buffer (pH 6.0) added to molten agar after autoclaving.⁸⁵ All experiments that did not involve RNAi treatment were seeded with OP50 E. Coli (From the CGC) for ad libitium feeding. Hypochlorite synchronization was used to developmentally synchronize experimental worms, where gravid hermaphrodites were exposed to an alkaline-bleach solution (e.g., 6 mL sodium hypochlorite, 2.5 mL KOH, 41.5 mL distilled water) to collect eggs, followed by repeated washes with M9 buffer (6 g/L Na2HPO4, 3 g/L KH2PO4, 5 g/L NaCl and 1 mL/L 1M MgSO4 in distilled water).⁸⁵

C. elegans strains

Strains used in this study:

CQ757: wqIs7 [Prgef-1::his-58::GFP],

CQ758: wqIs7 [Prgef-1::his-58::GFP];daf-2(e1370) III,

LC108: vIs69 [pCFJ90(Pmyo-2::mCherry + Punc-119::sid-1)] V,

CQ745: daf-2(e1370) III;vIs69 [pCFJ90(Pmyo-2::mCherry + Punc-119::sid-1)] V,

OH13859: otEx6423 [srd-5p::GFP + pha-1(+)],

CQ806: oyls44 [odr-1::RFP]; otEx6423 [srd-5p::GFP]

CQ823 wqEx99[Psrz-64::GFP + Pmyo-3::mCherry]

CQ824 wqEx100[PY65B4A.4::GFP + Pmyo-2::mCherry]

CQ825 wqEx101[PM04C7.4::GFP + Pmyo-2::mCherry]

PHX2685: ins-6(syb2685[ins-6::T2A::3xNLS::GFP]) II.

NY2049: ynIs49 [flp-5p::GFP] V.

KG2430: ceIs56 [unc-129p::ctns-1::mCherry + nlp-21p::Venus + ttx-3p::RFP]X.

CQ807: wqEx98[Y39G8B.7p::GFP + myo-3p::mCherry]

CQ819: daf-2(e1370) III; wqEx98[Y39G8B.7p:GFP + myo-3p:mCherry]

CQ820: ins-6(syb2685[ins-6::T2A::3xNLS::GFP]) II, daf-2(e1370) III

CQ821: daf-2(e1370) III, ynIs49 [flp-5p::GFP] V

CQ822: daf-2(e1370) III, ceIs56 [unc-129p::ctns-1::mCherry + nlp-21p::Venus + ttx-3p::RFP]X

Method details

Neuronal nuclei isolation

C. elegans neuronal nuclei were isolated using the following methods, modified from the mammalian single nucleus RNA-seq protocol from 10X Genomics. First, ∼400 μL of Day 1 worms were washed from HG plates. Three biological replicates of worms from each genotype were washed 3x in M9 buffer and transferred to a Dounce homogenizer (Kimble Glass Tissue Homogenizer, 88 mm overall length, Dounce 1 mL working capacity; Cat # 885300-0001) filled with 300 μL of ice-cold NP40 lysis buffer (10 mM Tris-HCl (Sigma-Aldrich, Cat. #T2194; pH 7.4), 10 mM NaCl (Sigma-Aldrich, Cat. # 59222C), 3 mM MgCl₂ (Sigma-Aldrich, Cat. #M1028), 0.05% Nonidet P40 Substitute (Sigma-Aldrich, Cat. # 74385), 1 mM DTT (Sigma-Aldrich, Cat. # 646563), and 1 U/μL RNase inhibitor (Sigma Protector RNase inhibitor; Sigma-Aldrich, Cat. # 3335402001), Nuclease-free water). Each sample was dounce homogenized 25-35x, using only the tight pestleuntil complete worm lysis was achieved. Samples were moved to low bind microcentrifuge tubes, incubated for 5 min on ice, and pipetted 10 times at two intervals during the incubation (at 2 min and 4 min) to disrupt tissues and cells. Nuclei suspensions were passed through a 40 μm filter into a conical tube, and then transferred to a low bind microcentrifuge tube. Samples were centrifuged at 1000 x g for 5 min at 4°C. The supernatant was removed and 1 mL of wash buffer (PBS +0.5% BSA (Miltenyi Biotec, Cat. # 130-091-376) + 1 U/uL of RNase inhibitor) was added with no mixing and incubated for 5 min on ice. After 5 min, the pellet was resuspended. Samples were centrifuged again at 1000 x g for 5 min at 4°C. The supernatant was removed, and the pellet was then resuspended in 500 μL of wash buffer. Hoechst stain (1:10,000 dilution; Molecular Probes Hoechst 33342, Thermo Fisher, Cat. #H3570) was added to each sample, and samples were passed through 5 μm syringe filters, directly into FACS tubes. Samples were incubated for at least 5 min on ice prior to FACS. Hoechst and GFP+ nuclei were sorted using a 70 μm nozzle and a flow rate of 3 on a BD Biosciences FACSAria Fusion sorter into a 1.5 mL low bind microcentrifuge tube containing collection buffer (500 μL of 0.5% BSA +1.5. U/μL RNase Inhibitor). Most nuclei maintained their round shape post-FACS and did not leak fluorescence, indicating that the nuclei were not damaged by sorting (Figure S1A). The instrument was washed with bleach between samples.

snRNA-seq library preparation and sequencing

After FACS, samples were centrifuged at 1000 x g for 5 min at 4°C. The supernatant was removed and nuclei were resuspended in 20 μL of collection buffer, and nuclei integrity was monitored by microscopy, with ∼90% of nuclei intact (no membrane blebbing) for each sample. The total number of nuclei for each sample was estimated and provided to the Princeton Genomics Core. Single nuclei suspension samples were loaded to the 10X Genomics Chromium X system using the Single Cell 3′ v3.1 Reagent Kits (10X Genomics Inc., CA) to generate and amplify cDNA. Up to 30,000 nuclei were loaded per 10X sample. The amplified cDNA samples were purified with Ampure XP magnetic beads (Beckman Coulter, CA), quantified by Qubit fluorometer (Invitrogen, CA), and examined on Bioanalyzer with High Sensitivity DNA chips (Agilent, CA) for size distribution. Illumina sequencing libraries were generated from the amplified cDNA samples using the Illumina Tagment DNA Enzyme and Buffer kit (Illumina, CA). These libraries were examined by Qubit and Bioanalyzer, then pooled at equal molar amount and sequenced on Illumina NovaSeq 6000 S Prime flowcells as 28 + 94 nt pair-end reads following the standard protocol. Raw sequencing reads were filtered by Illumina NovaSeq Control Software and only the Pass-Filter (PF) reads were used for further analysis.

Alignment and quality control of data

After Illumina sequencing, alignment of reads was performed using CellRanger version 7.1.0. Quality control of the data began with removal of ambient RNA contamination using SoupX on the Cell Ranger output files. SoupX calculated contamination fractions for all samples were between 0.03 and 0.12. Next, metrics of genes/cell, average UMIs/cell, and number of cells per cluster were determined. Violin Plots of genes per cell were generated to determine cutoffs: the lower bound was 200–300 features/cell for each sample (to remove damaged nuclei and empty droplets) and the higher bound was between 1000 and 2000 features/nucleus (to remove doublets) depending on the sample. Data outside of these cutoffs were excluded from further analysis.

Normalization, integration, and clustering

In Seurat, we followed the single cell pipeline, where single cell transform (SCT) was used for normalization. Prior to normalization, all data were pooled. SCT normalization is different from the normalization used for bulk sequencing datasets: it fits genes to a negative binomial distribution to normalize them across data instead of dividing everything by a ‘size factor’ as in bulk seq. Elbow plots were generated to determine the number of principal components (PCs) to include in the dimensional reduction, and the Louvain algorithm in the Seurat package⁸⁶ was used for unsupervised network clustering. We tested different numbers of PCs: 80, 100, 130, 150, 200, and different resolutions: 0.5, 0.7, 1.0, and 1.2. 150 PCs at a clustering resolution of 1 was used, which resulted in 100 clusters in our dataset without overclustering of neuron subtypes.

Cluster labeling/cell type analysis

We performed cluster annotation using a combination of systematic and manual approaches. Initially, we utilized the 'FindAllMarkers' function in the Seurat package to identify cluster-specific markers. These markers corresponded to genes showing significant differential expression (log2FC > 0.25) within each cluster compared to the average expression across all clusters. These markers also needed to be expressed in greater than 25 percent of cells for them to be used.

To further refine the annotation process, we compared the identified markers for each cluster with a curated list of known markers specific to each neuron type classification (Table S3). This comparison was conducted using a hypergeometric test and assessing the Bonferroni-corrected p-value for the overlapping gene regions. Clusters demonstrating a significant overlap (p < 0.01) with known anatomy markers were considered for the corresponding neuron type.

Additionally, we employed the AUCell algorithm as an independent mechanism for cluster annotation. Initially, we constructed a selected gene set collection utilizing known markers associated with each neuron type. Subsequently, we ranked the expression levels of each gene within every cell in our sequencing dataset. Using these rankings, we calculated the area-under-the-curve (AUC) value for each gene set in every cell.

To assign neuron types to cells, we generated a histogram of AUCell values and applied a threshold. In some cases, we manually adjusted the threshold to ensure approximately 5% of the cells were assigned to each neuron term. Subsequently, we assembled the cells into clusters and determined the percentage of cells within each cluster assigned to a specific neuron type. The neuron types with the highest five percentages were considered further for that cluster.

To make the final neuron type assignment for a cluster, we integrated the results from the hypergeometric test and the AUCell algorithm. If the outcomes from both methods aligned, we regarded that neuron type as the final annotation. In cases of disagreement, we manually examined gold-standard markers associated with that neuron type and made a manual decision regarding the appropriate annotation. The results of both mechanisms and the final annotation are summarized in Table S3. These orthogonal approaches increase our confidence in neuron classifications.

Threshold setting for expressed genes

When comparing our dataset to L4 data, we not only subsetted our data to only compare our wild-type data with the L4 data, we also applied expression level thresholds to the genes. We applied 4 thresholds (as done by CeNGEN¹¹) that require a varying percent of expression of a gene within a cluster (0.5, 1, 1.5, or 3) and a minimum average normalized expression value (0.001) (Table S5). When comparing our data to CeNGEN, we always compare our 2^nd threshold against their 2^nd threshold. These thresholds show enrichment of genes in specific places and help to cut down any noise in the data.

Hierarchical clustering

To generate a gene-expression-based hierarchical clustering dendrogram, we first aggregated the normalized gene expression level of each gene in a cluster to obtain the average gene expression vector for each cluster, then calculated the Euclidean distance matrix between clusters, and then hierarchically clustered using the “hclust” function with the “complete” linkage method.

Differentially-expressed gene identification

We used the FindMarkers function in the Seurat package for differential expression identification. We performed the Wilcoxon Rank-Sum Test method on each subsetted cluster separately, comparing N2 cells and daf-2 cells within the same cluster. Genes with a minimum percentage of expression in 10% of the cells were analyzed, and the significantly differentially expressed genes were identified if their log2(fold-change) > 0.25 or < −0.25, and adjusted p-value <0.05. The genes significantly differentially expressed in at least one cluster are considered differentially expressed.

For comparison with previous datasets, genes that are differentially expressed with a p-value <0.05 were considered differentially expressed in Kaletsky et al., 2016 dataset.⁸ Genes with a ranking <1500 or > −1500 and FDR <0.05 were considered differentially expressed in the Tepper et al., 2013 dataset.³⁷ Genes that have a daf-2 vs. N2 log2(fold-change) > 0.5 and p-adjusted <0.05 were considered differentially expressed in the Weng et al., 2024 dataset.³² Genes present in those 3 datasets were considered previously identified and the differentially expressed genes not present in these 2 datasets were considered newly identified.

When calculating the correlation of differential expression of clusters, each cluster was represented as a vector where each dimension is the differential expression log2(fold-change) of a gene, and Pearson’s correlation of each vector was calculated to generate a correlation graph for Figure S6A.

Gene differential expression heatmaps and dotplots were generated using the ggplot2 package in R and log2(fold-change) and p-adjusted values were extracted and plotted.⁸⁷

Gene co-regulation network analysis

We performed gene co-regulation network analysis using the daf-2 vs. N2 log2(fold-change) matrix. Using the GENIE3 package, we calculated the regulatory strength between all differentially expressed genes using tree-based ensemble methods.⁴¹ Then we selected the gene interaction of gene pairs with an interaction weight >0.04, resulting in a total of 793 interaction pairs, and visualized their interactions using the Cytoscape software 3.10.0.⁸⁴ We used the perfuse force OpenCL layout where the length of an edge reflects the weight of the interaction. To plot a subset of interaction, we selected the nodes and edges to plot, then calculated the perfuse force OpenCL layout independently.

Euclidean distance measure

We first subsetted the data into each cluster and further subsetted out N2 and daf-2 cells using the Seurat package. For N2 and daf-2 in each cluster, we generated a representative PC vector by averaging the column means of PCA embeddings from all cells in that cluster and genotype. Then, we calculated the Euclidean distance between the PC vector of N2 and daf-2 in each cluster. We ranked the distance from the largest to the smallest.

GSEA

For calculating GSEA for KEGG pathways, we downloaded the KEGG pathways from Worm Enrichr.⁸⁸ Next, we generated a ranked list of the differentially expressed genes in each cluster according to their log2(fold-change). We applied GSEA on the ranked list for each cluster using the fgsea package in R.⁸² The minimum size of the gene set to test is 3 and the maximum size is 100. Then, we plotted the normalized enrichment score for each pathway in the heatmap.

For calculating the GSEA enrichment score for the genes that downregulate with age, we input the ranked list of the differentially expressed genes in each cluster according to their log2(fold-change). Then, we generated a gene set of all genes that downregulate with age in wild-type neurons (log2FC > 2.0, p-adj <0.001) and calculated the gene set enrichment score of each neuron for this gene set.

For calculating GSEA enrichment score for the bulk sequencing datasets, we generated the ranked list from the Kaletsky et al., dataset based on fold-change, and another ranked list from Tepper et al., dataset based on ranking.^8,37 Then we generated a gene set for each cluster using only the significantly upregulated genes. We performed GSEA of these 2 ranked lists on each gene set where the minimum size of the gene set to test is 10 and the maximum size is 1000, and calculated the normalized enrichment score.

Gene ontology analysis

We identified AWC-enriched genes using the FindMarkers function in the Seurat package to identify genes significantly differentially expressed in the AWC compared with the background (log2(fold-change) > 0.1, p-adjusted <0.05). We identified the AWC daf-2 upregulated and downregulated genes using the significantly up- and down-regulated genes from the gene differential expression analysis. We measured GO terms using the WormCat 2.0⁶⁸ Website and chose the significantly enriched pathways (Bonferroni adjusted p-value <0.05).

Motif enrichment analysis

To identify the enriched transcription factor binding motifs of the daf-2 upregulated genes, the −1000 to −1 upstream region of the daf-2 upregulated genes were downloaded using the RSAT. The −1000 to −1 upstream region of all C. elegans protein-coding genes were retrieved as the background control sequences. We then input the sequence into the Simple Enrichment Analysis (SEA) function in the Meme-suite website (https://meme-suite.org/meme/tools/sea). We used Cis-BP 2.00 C elegans motif database as the input motif database. To identify enriched 7-mer oligos, we input the upstream region into the oligo-diff function in the RSAT website. The output enriched motifs and overrepresented oligomers are presented in Table S10.

Fluorescence microscopy

Fluorescence microscopy was performed on the Nikon AXR Confocal Microscope using the 60X oil immersion objective. Worms are synchronized and raised on HGM plates until adulthood. Day 1 adult worms were immobilized with sodium azide on agar pads. In experiments with RNA interference, RNAi was performed through feeding worms with HT115 bacteria with RNAi plasmids from hatching. We imaged the following fluorescently tagged proteins or fluorophores: GFP, Venus, mCherry, and DiI at 488 or 561 nm excitation wavelength. DiI staining were performed by adding 5uL of 2 mg/mL DiI into 1mL of M9 for 2 h before imaging. Images were collected using z stack with 0.5um increments. Images were processed and quantified through the FIJI software. Image processing included maximum intensity projection, thresholding, rotation, cropping, brightness/contrast, splitting color channels, and merge color channels. The neuronal identities of these reporter strains were confirmed using DiI staining and morphology (Figures S3A–S3C and S12). All quantifications were conducted on maximal projections of the raw data without brightness or contrast adjustments. Fluorescence intensity quantification was performed through thresholding or manual area selection and mean intensity of the area was quantified through FIJI. Detection quantification was performed by blind assessment of whether a neuronal GFP is detected in a worm or not. For quantifications from confocal images, N = number of worms quantified, unless otherwise specified.

Behavioral assays

All Short-Term Associative Memory assays (STAM) were performed as previously described.^7,89 Briefly, synchronized adult worms were washed and starved for an hour before training on plates with food and butanone (present on the plate lid) for 1 h to form an associative memory. Then worms were tested on chemotaxis plates with butanone and ethanol on opposite sites immediately after training for their learning ability or transferred to holding plates with food for 1 h or 2 h to subsequently test their memory ability. Naive chemotaxis was measured for each treatment group in each experiment by testing the chemotaxis index of untrained worms. The chemotaxis index was measured by the number of worms at the butanone location minus the number of worms at the ethanol location, divided by the total number of worms outside of the origin spot. The learning index is measured by subtracting the chemotaxis index of trained worms from the chemotaxis index of naive worms. For chemotaxis assays, chemotaxis toward 1% butanone (Acros Organics; Cat. #332828-25ML) or 1% benzaldehyde (Millipore Sigma Cat. #B1334-100G) in ethanol was accessed using standard chemotaxis conditions (Bargmann et al., 1993). All chemotaxis assays used Day 2 adult, LC108 worms that completed two days of adult-only RNAi treatment.

For behavioral experiments with RNA interference, we used LC108 (Punc-119::sid-1) or CQ740 (daf-2(e1370) III; vIs69 [pCFJ90(Pmyo-2::mCherry + Punc-119::sid-1)] V) worms to increase the penetrance of RNAi to the nervous system. We performed adult-only RNAi by transferring synchronized L4 worms onto fresh RNAi plates supplemented with IPTG, carbenicillin, FUdR and additionally freshly spotted IPTG prior to worm transfer.Worms were moved to fresh RNAi plates every 2 days before performing behavior experiments. We transfer the worms to RNAi plates without FUdR 1 day before performing behavior. For daf-2 STAM experiments, we performed the experiment on Day 5 to test effects on aging worms.

PA14 choice assay

Choice assays were performed as described in Moore et al., 2021.⁹⁰ Overnight liquid-grown cultures of PA14 were diluted in LB to an Optical Density (OD₆₀₀) = 1, and 25 μL of each bacterial suspension (PA14 or OP50 control) was spotted onto one side of a 60 mm NGM plate and incubated for 2 days at 25°C. Plates were moved to room temperature for 1 h before use. Immediately before use, 1 μL of 1M sodium azide was spotted onto each bacteria spot as a worm paralytic to capture the first choice made by each animal. Day 1 adult wild-type or daf-2 worms raised on OP50-seeded plates were collected and washed in M9. 5 μL of washed worm pellet was spotted at the bottom of the choice assay plate and incubated at room temperature for 1 h, when all worms were fixed at either the OP50 or PA14 spot. Worms were counted and the choice index was calculated (Choice Index = (Number of worms on OP50 – Number of worms on PA14)/(Total number of worms on OP50 + PA14)).

Quantification and statistical analysis

Unpaired, two-tailed Student’s t test was performed to compare chemotaxis or learning indices with only 2 conditions. One-way ANOVA with Tukey’s post-hoc analysis was used to compare results between multiple groups. two-way ANOVA with Sidak’s post-hoc analysis was used to compare the learning and memory curve between different genotypes and different time points with significant interaction between factors. One-way ANOVA with Bonferroni’s post-hoc analysis was used to check the chemotaxis index change between multiple groups. One-way ANOVA with Dunnett’s post-hoc analysis was used to identify memory changes after RNAi treatment to daf-2 worms. Wilcoxon Ran Sum test’s adjusted p-values are used for the identification of differentially expressed genes. Experiments were repeated on separate days, using separate independent populations, to confirm that results were reproducible. Prism 9 software was used for all statistical analyses. Software and statistical details used for RNA sequencing analyses are described in the method details section of the STAR Methods. Additional statistical details of experiments, including sample size (with n representing the number of chemotaxis assays performed for behavior, RNA collections for RNA-seq, and the number of worms for microscopy), can be found in the figure legends.

Published: December 4, 2024

Supplemental information

Table S1. Analysis of synaptically enriched genes, related to Figure 1

Comparison of 542 synaptically enriched genes¹⁸ across our wild-type day 1 data and CeNGEN’s 1^st threshold of genes detected supplementary data.¹¹ Genes detected were compared in both datasets, divided by the total 542, and multiplied by 100 to yield percent of detection. The first page of this table contains a summary of these results, and CeNGEN’s data is available on their website.

Table S2. Gene set lists neuron CREB GNAQ, related to Figure 1

In this table we compared our wild-type day 1 genes detected and CeNGEN’s 1^st threshold of detection against 3 datasets. The Neuronally Detected tab denotes a dataset of neuronally enriched genes.⁹ The CREB_LTAM tab denotes genes upregulated in a memory dataset collected by our lab.¹⁹ The final GNAQ tab contains orthologs of genes upregulated upon GNAQ gain of function.¹⁷ The first tab of this table contains a summary with information on each dataset.

Table S3. Cluster identification results, related to Figure 1

A curated list of markers for each neuron type is listed. Enriched genes for each identified cluster are shown (pct1 = cluster of interest, pct2 = all clusters). We determined the neuron-type anatomy of each cluster based on hypergeometric test and AUCell test results. The test results and the final annotation of each cluster are shown.

Table S4. WT csGPCRs, related to Figure 4

Analysis of chemoreceptor GPCR expression across cell types in our wild-type dataset. The “Summary and Info” tab contains information on what each tab in the table means as well as our thresholding method. The number of receptors depends on the threshold of detection used, and raw data should be considered along with this sheet. Throughout the manuscript we use our semi-liberal threshold (2): >1% detection in cells in the cluster and >0.001 averaged normalized expression across the cluster.

Table S5. All genes and thresholds in WT data, related to Figure 4

A spreadsheet containing averaged normalized expression data for all genes detected in our wild-type dataset separated into the 4 thresholds outlined in the Methods. Before each threshold tab, there is a summary tab detailing the spread of averaged normalized expression values throughout the set.

Table S6. WT peptidergic GPCRs, related to Figure 4

Analysis of neuropeptidergic GPCRs in our wild-type dataset. The “Summary and Info” tab contains information on what each tab in the table means as well as our thresholding method. The number of receptors depends on the threshold of detection used, and raw data should be considered along with this sheet. Throughout the manuscript we use our semi-liberal threshold (2): >1% detection in cells in the cluster and >0.001 averaged normalized expression across the cluster.

Table S7. CeNGEN comparisons to WT, related to Figure 4

This table highlights differences in csGPCR expression between L4 and Day 1 animals by comparing our 2^nd threshold of expression in our WT data to CeNGEN’s 2^nd threshold of expression. The “Summary and Info” tab gives information on what the differences are and which data are being compared.

Table S8. flps_nlps, related to Figures 4 and S4

Analysis of flps and nlps in our full WT and daf-2 dataset for increased resolution. This table contains a Summary and Info tab explaining all tabs in the table. The “Raw Data” tab contains detection of a gene across cells in a cluster. The “Percent Expression” tab contains the percent of cells in a cluster where a gene is detected. Next, we looked at flps and nlps that have known interactions discovered to see if we could place any in specific neurons to provide neuronal context to where the peptide is coming from. The “Known Interactors” tab contains a selection of interactors and their percent expression information and these interactors were characterized in Beets et al. 2023.⁵

Table S9. Receptor presence, related to Figures 4 and S4

Analysis of neuropeptide receptor presence in our full WT and daf-2 dataset for increased resolution. The “Summary and Info” Tab contains information on all tabs in the spreadsheet. The “Unfiltered Cell Expression” tab contains detection of a gene across cells in a cluster. The “Percent Receptor Expression” tab contains the percent of cells in a cluster where a gene is detected. The “Avg Expression of Receptors” tab contains RNA count information of the genes, and cutoffs were made to see which neurons receptors were enriched in. Finally the “Known Interactors” tab gives localization to some interactions characterized in Beets et al. 2023.

Table S10. daf-2 vs. N2 differentially expressed genes, related to Figure 5

daf-2 vs. N2 differential expression results from Wilcoxon Rank-Sum test of every cluster. Significantly differentially expressed genes of each cluster are shown as a separate tab with their gene name, p-value, average log2Fold-change (daf-2/N2), expression percentage pc1 (daf-2) and pc2 (N2), and adjusted p-value.

Table S11. daf-2 vs. N2 neuron-specific Euclidean distance, etc. analysis, related to Figure 5

Results from the daf-2 vs. N2 analyses. Raw results for the distance of mean PC vector between the 2 genotypes in each cluster are listed. Gene regulation network results are listed with gene names of the 2 interacting genes and their interaction strength calculated from GENIE3 for all genes interaction weight >0.04. The neuronal correlation matrix was shown with Pearson’s correlation coefficients between every 2-neuron cluster. Comparison of each cluster’s gene differential expression results with Kaletsky et al., 2016⁸ and Tepper et al., 2013³⁷ papers using GSEA were shown, with their enrichment score, normalized enrichment score, p-value, adjusted p-value, and log2error.

Table S12. daf-2 vs. N2 Kyoto Encyclopedia of Genes and Genomes pathways gene set enrichment analysis (GSEA), related to Figure 5

Results from the daf-2 upregulated and downregulated KEGG pathway analysis. The upregulated and downregulated pathways for each individual neuron cluster was calculated using GSEA and shown as different tabs. Results include pathway name, p-value, adjusted p-value, log2error, enrichment score, normalized enrichment score, overlap size, and leading-edge genes.

Table S13. Motif analysis results, related to Figure 7

Analysis of the AWC neurons. Lists of wild type and daf-2 combined AWC-enriched genes and AWC-enriched Gene Ontology Terms are shown. Lists of AWC daf-2-up- and down-regulated Gene Ontology Terms are shown.

Table S14. AWC-specific daf-2 analysis, related to Figure 7

Motif enrichment analysis of daf-2-upregulated genes. List of enriched TF-binding motifs from the upstream −1000 to −1 region of daf-2-upregulated genes. List of enriched 7-mers from the upstream −1000 to −1 region of daf-2-upregulated genes.