Main

Memory loss is among the most debilitating symptoms of aging. As human lifespan continues to increase1, a challenge must be addressed: how can we slow cognitive aging? Age-related cognitive decline starts in middle age2, with learning and memory decreasing more rapidly than other behavioral declines, even in the absence of disease3,4. While mechanisms underlying memory regulation and cognitive aging have been studied for decades, therapies for age-related cognitive decline are still lacking. Neuron-based mediators of memory have been the primary focus, as the brain is the functional ___location of cognition and memory. However, the role of peripheral tissues on brain function is becoming appreciated. For example, mouse parabiosis experiments revealed blood factors that influence cognitive function5,6. Whether any tissues outside neurons play a similar role in the regulation of Caenorhabditis elegans memory, and therefore might act as possible targets of midlife or late-life interventions to slow cognitive decline, is underexplored.

The molecular machinery required to form and maintain short-term and long-term memories is conserved from C. elegans to mammals7,8. For example, pathways controlling presynaptic vesicle transport and release are vital components regulating memory maintenance in both worms and humans9,10,11,12. Like mammals, C. elegans can learn and form both short-term and long-term associative memories3,8; memory is measured using a Pavlovian-style appetitive associative learning and memory assay3. Young adult wild-type worms retain short-term memory for 1–2 h3; short-term memory decreases with age and is lost in midlife3. Long-term memory induced by multiple training sessions persists for 16–24 h and requires gene transcription by the cAMP response element-binding protein (CREB)3, which is activated in a single pair of interneurons (AIM) upon spaced training7. Long-term associative memory (LTAM) ability is lost even earlier in adulthood than short-term memory, correlating with declining CREB levels3. CREB is a conserved regulator of long-term memory across organisms, including Aplysia, C. elegans, Drosophila, and mice7. Overexpression of CREB specifically in neurons3 or gain-of-function in Gαq/GNAQ signaling upstream of CREB13 each enhances worms’ long-term memory, including in aged worms, and activation of Gαq/GNAQ in aged mice restores memory12. CREB expression decreases with age both in C. elegans3 and in mammals14, a trend paralleling age-related memory decline3,14. This implies that sustaining or restoring CREB levels/activity might present a strategy to prevent or slow cognitive aging.

The IIS longevity pathway is critical for regulation of aging in worms through mammals9. This pathway both autonomously and non-autonomously regulates longevity15, metabolism16,17 and reproduction18,19,20,21,22. Additionally, DAF-2 plays tissue-specific roles in longevity and dauer formation regulation23. We previously found that reducing IIS via mutation of the insulin/IGF-like receptor, daf-2, significantly extends associative memory and slows cognitive aging3. We also identified genes responsible for IIS regulation of this memory improvement through transcriptomic analysis of neurons isolated from daf-2 mutants10. However, a non-autonomous role for IIS in regulation of cognitive function has not previously been explored.

While investigating tissue-specific functions of IIS activity using the auxin-inducible degron (AID) protein degradation system to selectively degrade DAF-2 (ref. 23), we made the surprising discovery that loss of DAF-2 in the hypodermis, the metabolic epidermal tissue of the worm20, significantly extends memory. That is, in addition to the expected memory improvement by DAF-2 degradation in neurons, hypodermal DAF-2 degradation significantly improves memory. We then found that the hypodermis signals to neurons via the OSM-11 Notch ligand and Notch signaling components in neurons. Using single-nucleus RNA sequencing (RNA-seq) of neurons, we identified transcriptional changes in individual neurons downstream of hypodermal IIS signaling, revealing that hypodermal DAF-2 degradation broadly induces expression of CREB, and hypodermal DAF-2’s extended memory is dependent on CREB. Unlike hypodermal DAF-2 depletion memory extension, the memory enhancement caused by DAF-2 degradation in neurons is independent of CREB. We further established that memory regulation by the hypodermis is mediated by Notch ligand signaling from the hypodermis to neurons and is dependent on CREB. Finally, reduction of hypodermal IIS or increased Notch signaling via OSM-11 expression in aged worms can each rescue memory decline. Our results suggest that the memory extension induced by the reduction of IIS in daf-2 mutants is the result of the combination of CREB-dependent and CREB-independent gene expression changes regulated by hypodermal and neuronal IIS, respectively. These data establish a role for non-autonomous regulation of CREB-dependent memory and cognitive maintenance by hypodermal IIS signaling and downstream Notch signaling. Thus, memory can be extended by insulin signaling in both the brain and the body, and DAF-2 activation rescues memory function in aged animals.

Results

Reduced hypodermal or neuron IIS extends associative memory

We previously found that reduction of IIS improves C. elegans associative memory in both young and aged adults and extends memory beyond 5 h after a single training session3. To test memory, we use a Pavlovian positive olfactory association assay3: short starvation followed by conditioning (training) of the animals with food + butanone (a neutral odorant) increases the worms’ preference for butanone, indicating learning of this positive food–butanone association; holding the trained animals on food without butanone, then testing their chemotaxis to butanone at later times, measures associative memory duration (Extended Data Fig. 1a,b and Extended Data Table 1). We measure chemotaxis before training (‘naive’) to ensure that the chemotaxis index (CI) is low (~0.2 for wild type), indicating low stress. During the naive chemotaxis assay, we also assess whether there are any motility issues that would affect interpretation of the remainder of the assay; if the worms cannot move away from the origin, we do not perform learning and memory assays on those animals, to avoid confusing high naive preference with learning or memory. Immediately after conditioning (odor + food training), we assess the trained CI; learning (t = 0 after training) is the difference between naive and trained CI (Extended Data Fig. 1a,b). Worms’ preference for butanone is tested at selected time points after a single training session (short-term associative memory (STAM); Extended Data Fig. 1a,b). Thus, we can distinctly assess each function (motility, naive butanone attraction, associative learning and associative memory). Consistent with our previous results, the insulin-like receptor mutant daf-2 improves memory maintenance, retaining memory for more than 5 h after conditioning, while wild-type worms have forgotten the association between food and butanone by 2 h3 (Fig. 1a and Extended Data Fig. 1d,e).

Fig. 1: Reduced insulin signaling in the hypodermis and neurons extends associative memory.
figure 1

a, Young adult (day 1) daf-2(e1370) mutants have extended memory up to 6 h after one training session, while N2 wild-type worms’ STAM lasts for <2 h. b, Auxin-induced DAF-2 degradation in neurons increases associative memory in day 2 adults. c–e, Auxin-induced DAF-2 degradation in other tissues (intestine (c); gonadal sheath (d); germ line (e)) has no effect on memory. f, Auxin-induced DAF-2 degradation in the hypodermis improves memory in young adults. b–f, Auxin treatment started 20 h before memory training. In the estimation plots, mean difference (bootstrap sampling distribution) is depicted as a dot. 95% confidence interval indicated by the vertical bar. g–i, Naive CI (g), CI (h) and learning index (i) of neuronal DAF-2-AID worms. h,i, Neuronal DAF-2 degradation extends memory up to 3 h. j, Memory was evaluated by calculating the rate of change in CI between the learning and 1-h time points across replicates. k, Naive chemotaxis and chemotaxis and learning index plots of hypodermal DAF-2-AID worms. l,m, Hypodermal DAF-2 degradation extends memory up to 6 h after a single training session, chemotaxis index (l) and learning index (m) of hypodermal DAF-2 AID worms. n, Slope analysis calculated as in j. o, Comparison of memory improvement induced by hypodermal or neuronal DAF-2 degradation. p, Lifespan and memory increases upon tissue-specific DAF-2 degradation. Lifespan changes and P values reported from ref. 23. Memory increase is reported as the average increase in median memory relative to controls across all biological replicates. Memory P values from 2-h learning index. b,f, Red indicates comparison of note for genetic interpretation. b–f, Representative experiments from two to three biological replicates. Each dot indicates individual chemotaxis assay plate containing ~80 worms. Two-way repeated-measures analysis of variance (ANOVA) and Tukey post hoc tests were performed. In the box plots, the center line denotes the median, box range indicates the 25th–75th percentile, and whiskers denote minimum–maximum values. a,h,i,l,m, Representative experiment from three biological replicates. Mean ± s.e.m. Two-way repeated-measures (ANOVA) and Bonferroni post hoc tests were performed. n ≥ 4 per time point. g,k,j,n, Two-sided unpaired t-test. NS, not significant, *P ≤ 0.05, **P ≤ 0.01, ***P ≤ 0.001, ****P ≤ 0.0001.

Source data

Full size image

Zhang et al. used the AID degradation system to selectively degrade DAF-2 in individual tissues to determine daf-2’s tissue-specific roles in the regulation of lifespan, dauer, and RNA and protein metabolism23. This system allows DAF-2 degradation with temporal control and only in specific tissues in adult animals23, avoiding any possible developmental effects due to systemically reduced IIS. The authors noted that DAF-2 is difficult to visualize due to its localization on the plasma membrane; therefore, they used GFP fusions to verify that DAF-2 was degraded in the specific tissues without affecting DAF-2 in other tissues in these TIR-1 lines23.

We treated these DAF-2 AID worm strains with auxin on day 1 of adulthood, then carried out our single-training associative memory assay on day 2 (young adults; Extended Data Fig. 1c) after DAF-2 was degraded in neurons, intestine, gonadal sheath, germ line, muscle, or hypodermis (Fig. 1b–f and Extended Data Fig. 1f–k). As expected, neuronal DAF-2 degradation significantly extended memory duration (Fig. 1b and Extended Data Figs. 1f and 2a), confirming that the auxin treatment successfully degrades DAF-2, consistent with the role of neuron-autonomous IIS in memory and behavior10. Degradation of DAF-2 in the intestine, gonadal sheath, or germ line had no effect on memory (Fig. 1c–e and Extended Data Figs. 1g–i,k and 2b–d). The muscle DAF-2 AID strain shows significant defects in learning before AID treatment (Extended Data Fig. 1k), complicating interpretation. Surprisingly, however, we found that DAF-2 degradation specifically in the hypodermis dramatically increases memory duration (Fig. 1f and Extended Data Figs. 1j and 2e). The magnitude of memory improvement induced by hypodermal DAF-2 degradation is similar to that of neuronal DAF-2 degradation (Fig. 1b and Supplementary Table 1).

Next, we examined the duration of memory induced by neuronal and hypodermal degradation of DAF-2. Surprisingly, while auxin-induced neuronal DAF-2 degradation extended memory beyond 3 h after training (Fig. 1g–j), hypodermal DAF-2 degradation extended memory beyond 6 h after training (Fig. 1k–o), similar to the memory duration of daf-2 mutants (Fig. 1a and Extended Data Fig. 1d).

To determine whether memory improvement is solely the consequence of increased lifespan, we compared the extension of lifespan by DAF-2 degradation in a specific tissue23 and memory improvement (Fig. 1p). We found no correlation between extension of lifespan by DAF-2 degradation in a specific tissue and the effect on memory; for example, hypodermal DAF-2 degradation has the highest mean memory increase, but causes only a modest lifespan increase23, while DAF-2 degradation in the intestine nearly doubles lifespan and, like daf-2 mutants, increases dauer formation23, as we verified (100% dauer; Extended Data Fig. 2h) but has no effect on memory (Fig. 1c and Extended Data Fig. 1g and 2a). These results eliminate the hypothesis that simply extending lifespan increases memory, suggesting that lifespan and memory are regulated by DAF-2 through distinct mechanisms.

Identification of hypodermal IIS transcriptional targets

To uncover the molecular mechanisms of non-autonomous memory regulation from hypodermal IIS, we identified downstream transcriptional targets regulated by hypodermal IIS using whole-worm RNA-seq. Comparison of the transcriptional profiles of day 2 non-auxin control and hypodermal DAF-2-degraded ( + auxin) worms revealed distinct transcriptional profiles (principal component analysis (PCA); Fig. 2a). Similarly, neuronal DAF-2-regulated genes (Fig. 2b) are distinct from hypodermal DAF-2-regulated genes (Fig. 2c,d). Although we manipulated DAF-2 in the hypodermis, most genes upregulated by hypodermal IIS reduction are enriched in the nervous system and neurons, followed by the intestine and hypodermis (Extended Data Fig. 3a), consistent with our behavioral results (Figs 1 and 2c and Supplementary Table 2).

Fig. 2: Hypodermal DAF-2(−) transcriptional targets are distinct from neuronal DAF-2(−) targets.
figure 2

a, PCA of mRNA-seq from whole worms with hypodermal DAF-2 degradation and neuronal DAF-2 degradation. Day 2 adult worms; six biological replicates per condition. b, Volcano plot of neuronal DAF-2(−)-regulated genes. Green dots indicate upregulated genes (log2 fold change ≥ 0.5). Gray dots indicate downregulated genes (log2 fold change ≤ −0.5). False discovery rate (FDR) ≤ 1%. c, Volcano plot of hypodermal DAF-2(−) upregulated and downregulated genes (gray dots, log2 fold change ≥ 0.5 or ≤ −0.5, FDR ≤ 1%. Gold dots indicate genes expressed in neurons. Blue dots indicate genes expressed in the hypodermis. Purple dots indicate genes expressed in neurons and enriched in the hypodermis20,24. d, Comparison of upregulated genes from neuronal and hypodermal DAF-2(−) animals. All upregulated genes, FDR < 1%. Hypergeometric test. e, RNAi knockdown (from young adult (L4)) of the hypodermal DAF-2-regulated transcriptional targets pqn-73, fmo-2, or far-3 does not block memory enhancement induced by hypodermal DAF-2 degradation in young adults (tested on day 3). One biological replicate. f, ins-19 expression is downregulated upon hypodermal DAF-2 degradation, but not upon neuronal DAF-2 degradation. Statistics from DESeq2 (adjusted P values). g,h, CI (g) and learning index (h) plots of ins-19 mutant. Mean ± s.e.m. Two-way repeated-measures ANOVA, Bonferroni post hoc tests. n ≥ 4 per time point. n represents the number of chemotaxis plates at each time point. Each plate contains ~80 worms for all memory experiments. i, Memory was evaluated by calculating the rate of change in CI between the learning and 1-h time points across replicates. Two-sided unpaired t-test. e,f, Two-way repeated-measures ANOVA. In the box plots, the center line denotes the median, box range indicates the 25th–75th percentile, and whiskers denote minimum–maximum values. Representative experiment for two to three biological replicates unless specified otherwise. *P ≤ 0.05, **P ≤ 0.01, ***P ≤ 0.001, ****P ≤ 0.0001.

Source data

Full size image

To assess the roles of genes downstream of hypodermal IIS, we knocked down the top upregulated genes and tested memory. far-3, the gene with the highest fold change (Fig. 2c), encodes a fatty acid binding protein that is expressed in the hypodermis, intestine and neurons, and is regulated by Notch signaling key components20,24. pqn-73 encodes a prion-like protein; we previously found that pqn-73 is expressed in arcade cells and neurons, and is required for LTAM7. fmo-2 encodes a flavin-containing monooxygenase expressed in the hypodermis and neurons, and activation of fmo-2 in C. elegans extends lifespan25. While far-3 has DAF-16 binding elements (DBEs) and one DAF-16 associative element (DAE) in its upstream promoter, pqn-73 and fmo-2 have no DBEs or DAEs in their promoters and are thus less likely to be direct targets of the insulin signaling pathway in the hypodermis (Supplementary Table 3). However, knockdown of far-3, pqn-73, or fmo-2 did not block memory enhancement induced by hypodermal DAF-2 degradation; in fact, pqn-73 and far-3 knockdown each slightly increased memory (Fig. 2e and Extended Data Fig. 3b–d), suggesting these genes were not responsible for the improved short-term memory induced by hypodermal IIS reduction. We next focused on the hypodermal DAF-2(−) downregulated insulin-like peptide, ins-19, which is downregulated only upon hypodermal DAF-2(−) conditions and not in neuronal DAF-2(−) animals (Fig. 2f). Because ins-19 mutants have significantly higher learning and memory, consistent with previous results26 (Fig. 2g–i and Extended Data Fig. 3e), hypodermis IIS-induced memory extension may be mediated in part by ins-19 reduction.

Hypodermal-dependent memory requires the OSM-11 Notch ligand

We next sought to determine how the hypodermis—a structural and metabolic tissue—regulates memory, which is largely thought to be a neuron-autonomous process. We hypothesized that hypodermal IIS targets might include a secreted, diffusible protein that signals to neurons. We cross-referenced hypodermis-specific gene expression in adult C. elegans20 with genes that are upregulated in daf-2 mutants compared to daf-16;daf-2 worms (class I genes)16,17; 15 class I candidate genes are expressed in the hypodermis and are predicted to encode secreted proteins (Fig. 3a). Among those 15 genes, osm-11 is expressed in the hypodermis27,28 and has the highest fold-induction following adult hypodermal DAF-2 degradation (Supplementary Table 2 and Fig. 3b; P < 0.0034; log2fold change = 0.46). osm-11 is not changed upon neuronal DAF-2 reduction (Fig. 3b). osm-11’s promoter is a direct DAF-16 target29 with both a DBE and a DAE (Supplementary Table 3), indicating that osm-11 may be a direct target of the hypodermal IIS pathway. osm-11 encodes a secreted Notch ligand expressed in the hypodermis and spermatheca of adult worms27 (Fig. 3c,d). Finally, OSM-11 is secreted from the hypodermis and non-autonomously regulates octanol avoidance and quiescence28. OSM-11 is expressed in the seam cells of the hypodermis but not in neurons (osm-11p::gfp; Fig. 3c,d). Therefore, the Notch ligand OSM-11 emerged as our top candidate for mediating IIS signaling from the hypodermis to neurons directing memory improvement.

Fig. 3: Notch signaling components including osm-11 are required for extended memory.
figure 3

a, A total of 15 of 1,436 class I (daf-2 upregulated) genes are exclusively expressed in the hypodermis and are predicted to encode secreted proteins. b, osm-11 expression is upregulated upon hypodermal but not neuronal DAF-2 degradation. DESeq2 (adjusted P values). c, osm-11 is expressed exclusively in the hypodermis. Day 1, n = 21 worms. d, osm-11 is not expressed in neurons in young (day 2) and older adults (day 5). osm-11p::gfp (green, hypodermis), neurons = DiI (red). n = 15–20 worms/age. c,d, Representative images. Scale bars, 100 μm. e, osm-11 adult-only RNAi knockdown (starting late L4) abolishes memory enhancement induced by hypodermal DAF-2 degradation (STAM on day 3 adults). f, Memory evaluated by the rate of change in CI between 0-h and 2-h time points across replicates. g, Hypodermis-specific osm-11 overexpression enhances memory in young (day 2) adults. h, Memory evaluated by the rate of change in CI between 0-h and 1-h time points across replicates. i–k, lin-12 knockdown in flp-18-expressing neurons (i) (RIG, AIY, RIM, and AVA) or the RIG alone (k) (twk-3 promoter) abolishes memory enhancement induced by hypodermal DAF-2 degradation, while lin-12 knockdown in the AIY interneurons does not (j). CI plot (left), learning index at 2 h (middle). Memory evaluated as in (f). l,m, RNAi knockdown of sel-8 (l) or lag-1 (m) (starting late L4) blocks memory enhancement induced by hypodermal DAF-2 degradation (neuronal RNAi-sensitized background). l, Memory evaluated as in h, or for m as in f. n, sel-8 adult-only RNAi knockdown is not required for memory enhancement induced by neuronal DAF-2 degradation (neuronal RNAi-sensitized background). n = 1 replicate. e,i–n, Red indicates comparison of note for genetic interpretation. b,e,i–n, Mean ± s.e.m. Two-way repeated-measures ANOVA and Tukey or Bonferroni post hoc test were used (g,i–n). In the box plots, the center line denotes the median, box range indicates the 25th–75th percentile, and whiskers denote minimum–maximum values. n ≥ 4 chemotaxis plates/time point, each dot represents an individual assay plate with ~100 worms. Representative experiment from two to three biological replicates unless specified. f,h,i,j–n, Slope analysis, two-sided unpaired t-test. *P ≤ 0.05, **P ≤ 0.01, ***P ≤ 0.001, ****P ≤ 0.0001.

Source data

Full size image

We next tested whether OSM-11 is required for hypodermal-degraded DAF-2 memory extension: osm-11 knockdown abrogated the memory increase induced by hypodermal DAF-2 degradation in young adults (Fig. 3e,f and Extended Data Fig. 4a,b), suggesting that reduced hypodermal IIS acts via the secreted protein OSM-11 to regulate associative memory.

To determine if we can extend memory in wild-type worms by increasing Notch signaling, we overexpressed OSM-11 specifically in the hypodermis using the osm-11 promoter in young adults, and tested memory; OSM-11 overexpression increased associative memory (Fig. 3g,h and Extended Data Fig. 4b). Overexpression of osm-11 slightly increases learning, but also significantly improves memory (Fig. 3g,h). Although OSM-11 overexpression slightly increases lifespan (Extended Data Fig. 4c), its effect on memory is proportionally greater. These results suggest that OSM-11 is required for hypodermal IIS memory enhancement and is sufficient to enhance memory in the absence of additional inputs.

Hypodermal IIS-regulated memory requires the Notch pathway

Notch signaling is activated when Notch ligands bind to the receptors LIN-12 or GLP-1 (refs. 30,31), which are then cleaved, releasing the intracellular signaling transducing fragment, the Notch intracellular ___domain32,33. The Notch intracellular ___domain then translocates to the nucleus and binds to the transcription factor LAG-1 and coactivator LAG-3/SEL-8, forming a transcriptionally active ternary complex to regulate gene transcription34. Therefore, we hypothesized that OSM-11 might signal from the hypodermis to non-autonomously activate neuronal Notch signaling. If so, then other key components of Notch signaling might be required in neurons for the memory induced by hypodermal IIS reduction.

To identify candidate-receiving cells for the OSM-11 ligand, we reduced the LIN-12 Notch receptor in a subset of neurons (RIG, AIY, RIM, and AVA interneurons) using a hairpin RNA-mediated interference (RNAi) strain (flp-18p::lin-12(RNAi))28,35. Indeed, knockdown of lin-12 in those interneurons blocked the memory enhancement induced by hypodermal IIS reduction (Fig. 3i and Extended Data Fig. 4d,e). (nmr-1p::lin-12(RNAi) worms were not healthy, preventing analysis of the effect of lin-12 knockdown in RIM and AVA in memory.) The knockdown of lin-12 only in AIY neurons (ttx-3p::lin-12(RNAi)), an interneuron that is critical for LTAM formation in wild-type worms13, did not abrogate memory improvement (Fig. 3j and Extended Data Fig. 4f,g). However, lin-12 knockdown driven by the RIG-associated twk-3 promoter36 (twk-3p::lin-12(RNAi)) significantly blocked the memory increase induced by hypodermal IIS reduction (Fig. 3k and Extended Data Fig. 4h,i), suggesting that the RIG might be a receiving cell for the OSM-11 ligand through the LIN-12 Notch receptor.

The GLP-1 Notch receptor is also expressed in many neurons, including multiple ciliated sensory neurons28; glp-1 knockdown in ciliated neurons, even in the absence of auxin treatment, caused severe general learning and chemotaxis defects (Extended Data Fig. 5a–c), preventing the subsequent analysis of memory. To test whether GLP-1 might act in the RIG neuron, we constructed the hairpin RNAi of glp-1 under the twk-3 promoter; however, loss of glp-1 in the RIG had no effect on hypodermal DAF-2 reduction-mediated memory (Extended Data Fig. 5d,e), suggesting that LIN-12 is the sole Notch receptor required in the RIG neuron for this signaling pathway.

While we were unable to obtain healthy strains reducing sel-8 and lag-1 specifically in the RIG neuron, knockdown of sel-8 (Fig. 3l and Extended Data Fig. 5f,g) or lag-1 (Fig. 3m and Extended Data Fig. 5h,i) in neuronal RNAi-sensitive, hypodermal DAF-2-degraded adults reduced hypodermal DAF-2 degradation-mediated memory improvement. Knocking down sel-8 had no effect on memory improvement induced by neuronal DAF-2 degradation (Fig. 3n), suggesting that neuronal DAF-2 degradation increases associative memory through mechanisms distinct from Notch signaling. Together, our results suggest that components of the Notch signaling pathway, including LIN-12 function in the RIG, LAG-1, and SEL-8, are required in neurons for the extended memory induced by hypodermal DAF-2 degradation.

Single-nucleus RNA-seq of hypodermal DAF-2-degron worm neurons

Next, we sought to better understand how hypodermal IIS and Notch signaling might affect transcription in individual neurons. Previously, we carried out adult pan-neuronal RNA-seq of wild-type and daf-2 worms to identify neuronal genes that are required for daf-2’s extended memory10. We also recently identified transcriptional changes in individual neurons altered in daf-2 adults37 using a method we recently adapted from mammalian neuronal single-nucleus RNA-seq protocols37; single-nucleus RNA-seq allows transcriptional analysis of individual neurons. To identify transcriptional changes in individual neurons altered by hypodermal DAF-2 degradation, we first constructed a hypodermal DAF-2-AID strain expressing a pan-neuronal GFP-histone tag; these worms retain the ability to enhance memory after auxin treatment (Fig. 4a). We isolated GFP neuronal nuclei from day 2 no-auxin control and hypodermal DAF-2-degraded ( + auxin) worms and performed single-nucleus RNA-seq (Fig. 4b). We obtained 122,834 neuronal nuclei with an average of 456 unique molecular identifiers per cell and detected 370 genes per cell (Extended Data Fig. 6). We successfully identified 106 clusters that match 118 specific neuron classes (Fig. 4c), with no missing or extra clusters after auxin treatment (Supplementary Table 4).

Fig. 4: Neuronal single-nucleus RNA-seq upon hypodermal DAF-2 knockdown reveals broad neuronal changes.
figure 4

a, Neuronal single-nucleus RNA-seq was performed on hypodermal DAF-2 AID worms with GFP-tagged HIS-58; neuronal histone tagging does not affect memory enhancement upon hypodermal DAF-2 degradation. Mean ± s.e.m. n ≥ 4 chemotaxis plates at each time point, where each dot represents an individual assay plate with ~100 worms. Two-way repeated-measures ANOVA and Bonferroni post hoc tests were performed. Representative experiment from three biological replicates. ****P ≤ 0.0001. b, Schematic workflow of neuronal single-nuclei RNA-seq of hypodermal DAF-2 AID worms as performed in ref. 37. c, A total of 105 unique neuronal clusters were identified across two biological replicates each of hypodermal DAF-2 AID worms ± auxin. d, Significant Gene Ontology terms are shown from analysis of genes upregulated (log2 fold change ≥ 0.1 or ≤ −0.1, FDR ≤ 5%) across all neurons upon DAF-2 AID relative to the non-auxin control. e, KEGG pathway analysis of upregulated genes in the single-nucleus sequencing dataset. d,e, Results from gProfiler. f,g, AWC-upregulated differentially expressed genes from hypodermal DAF-2(−) were compared to those from whole-worm daf-2 mutants37. g, AWC chemosensory neuron differentially expressed genes. Two-sided unpaired t-test. d–g, Differentially expressed genes determined by Wilcoxon rank-sum test.

Source data

Full size image

Gene Ontology analysis of neuronal genes differentially expressed upon hypodermal DAF-2 degradation revealed upregulation of pathways related to voltage-gated channel activity, calcium signaling, G-protein-coupled receptor activity, and synaptic vesicles (Fig. 4d,e). Additionally, MAPK signaling, which is known to be required for synaptic plasticity and memory38, was enriched among the neuronal upregulated genes. These data suggest that the reduction of hypodermal DAF-2 may enhance memory by increasing neuronal activity and synaptic plasticity39,40,41. The FOXO and mTOR longevity-related pathways were enriched among neuronal genes (Fig. 4e), consistent with the role of IIS in C. elegans15,16,23,42,43. Ribosomal genes were downregulated in neurons (Extended Data Table 2), implying that there might be reduced ribosomal biogenesis in neurons following hypodermal IIS reduction, which may conserve energy and promote neuronal health44,45.

We were next interested in neuron-specific changes that might mediate enhanced memory upon hypodermal IIS reduction. The AWC pair of chemosensory neurons detect a variety of chemoattractants, including butanone, and are required for short-term associative learning and memory in both wild-type and daf-2 animals3,13,37,46,47. We previously performed single-nucleus RNA-seq on wild-type daf-2 mutant animals and identified transcriptional changes in the AWC neurons that result from whole-worm IIS reduction37. We compared the hypodermal IIS and whole-worm IIS changes in the AWC neurons and found that DAF-2 reduction exclusively in the hypodermis accounts for ~20% (7/36) of the daf-2 AWC-upregulated genes (Fig. 4f). Shared upregulated genes include neprilysin (nep-26), ist-1, C36C9.5, Rho GEF Trio/unc-73, the VH1 phosphatase vhp-1, and the transcription factors STAT/sta-1 and nhr-21 (Fig. 4g). Notably, nep-26, vhp-1, and unc-73 have previously been shown to be required for the extended short-term memory of daf-2 mutants37.

Broad CREB upregulation in hypodermal DAF-2 AID worm neurons

The transcription factor CREB and its downstream transcriptional targets are essential for long-term memory3,5,7. Our single-nucleus RNA-seq results showed that crh-1/CREB was upregulated in many neurons upon DAF-2 degradation in the hypodermis (Fig. 5a–c); notably, RIG is not the major site of CREB expression change, suggesting that CREB may be an indirect rather than direct target of Notch signaling. We identified 35 neuronal genes upregulated by hypodermal DAF-2 depletion that are also CREB targets induced by long-term memory training (CREB/LTAM7; Fig. 5d), and 36 neuronal genes induced upon hypodermal DAF-2 degradation that are orthologs of genes upregulated by gain-of function GNAQ expression in the mammalian hippocampus (Extended Data Fig. 6i). Most of these overlapping genes, including those that we previously determined to be crucial for learning and memory (for example, casy-1, egl-30, irk-3, kin-2, pkc-1, unc-43, and unc-73), exhibited increased expression in memory-related and sensory-related neurons—AIY, AIM, RIA, AIA, AWA, AWB, and AWC (Fig. 5d and Extended Data Fig. 6i). For instance, casy-1, which encodes the transmembrane protein calsyntenin, is upregulated across all three conditions (hypodermal DAF-2 degradation, CREB-LTAM and GNAQ(gf)) and is required for learning3,7,12. Similarly, unc-43, the C. elegans ortholog of the CaMKII type II calcium/calmodulin-dependent protein kinase, is required for memory maintenance7,8 and is induced by all three treatments.

Fig. 5: CREB is required for hypodermal IIS reduction-induced memory, but not for neuronal IIS reduction-induced memory.
figure 5

a, crh-1 is differentially expressed in many neurons upon hypodermal DAF-2-degradation. Red indicates upregulated in +auxin condition compared to no-auxin control. b, crh-1 levels in neurons with significantly upregulated expression upon hypodermal DAF-2 reduction. Wilcoxon rank-sum test. FDR ≤ 5%. c, crh-1 expression is increased broadly across the nervous system. Fold change across all neurons is shown. d, Hypodermal DAF-2(−)-upregulated CREB/LTAM genes in select neurons7. Wilcoxon rank-sum test. e, Adult-only RNAi knockdown (starting from late L4) of the known CREB/LTAM genes unc-43, unc-73, or casy-1 impairs 3-h associative memory extension hypodermal DAF-2(−) animals. n ≥ 4 chemotaxis plates at each time point, where each dot represents an individual assay plate with ~100 worms. Hypodermal DAF-2 AID performed in neuronal RNAi-sensitized background. f, Young (day 1) daf-2(e1370) mutants have extended memory beyond 5 h after a single training session, while daf-2(e1370);crh-1(n3315) worms’ memory lasts 2 h less. crh-1(n3315) worms’ memory lasts <2 h. g, Memory was evaluated by calculating the rate of change in CI between the learning-indicated time points across replicates. Two-sided unpaired t-test. h, Loss of CREB does not affect the memory extension of neuronal DAF-2 degradation until hour 4 after a single training session in young (day 2) adult animals. i, crh-1 mutation abolishes the memory induced by hypodermal IIS reduction. j, crh-1 is required for learning and memory in osm-11-overexpressing worms. e,f,h–j, Representative experiment from three biological replicates. Two-way repeated-measures ANOVA and Bonferroni post hoc tests were performed. Mean ± s.e.m. n ≥ 4. *P ≤ 0.05, **P ≤ 0.01, ***P ≤ 0.001, ****P ≤ 0.0001.

Source data

Full size image

We next investigated whether these induced genes are required for the memory extended by hypodermal IIS signaling. We used hypodermal DAF-2 AID worms with a neuronal RNAi-sensitive background to knock down specific genes and then tested memory duration. We focused on genes that are also regulated by CREB-LTAM and/or GNAQ(gf), or are known to be involved in learning and LTAM (unc-43, unc-73, and casy-1)3,8,12,13. RNAi treatment was initiated in adult animals to avoid causing developmental defects, and no motility defects were observed (Extended Data Fig. 7a). We found that all three genes are required for extended memory after hypodermal DAF-2 degradation (Fig. 5e and Extended Data Fig. 7b–f). Together, our data indicate that reduced hypodermal IIS bolsters memory maintenance via upregulation of CREB and multiple target genes in neurons.

CREB is required only for hypodermal IIS-dependent memory

We previously found that CREB is required for long-term memory, and increasing CREB activity extends long-term memory, as CREB is activated during spaced training3,7; however, a role for CREB in extending short-term memory after a single training session is less well studied. The upregulation of crh-1 and CREB/LTAM target expression in many neurons (Fig. 5d) suggested that CREB may contribute to the enhanced memory resulting from hypodermal DAF-2 degradation, and may at least partially contribute to the prolonged short-term memory exhibited by daf-2 mutants (Fig. 1a and Extended Data Fig. 1d,e).

To investigate whether CREB is required for daf-2’s extended short-term memory, we tested the duration of memory after 1 h of training in daf-2(e1370);crh-1(n3315) double mutants. While daf-2(e1370) mutants’ memory lasts up to 6 h, loss of CREB (daf-2(e1370);crh-1(n3315) double mutants) reduces memory to 4 h (Fig. 5f–g and Extended Data Fig. 7g–i), which suggests that CREB is partially required for the full extended memory of daf-2 mutants.

Since DAF-2 degradation in the hypodermis and neurons, not other tissues, induces memory enhancement, we then tested where the partial CREB contribution in daf-2 mutants’ memory originates from. Loss of CREB increases naive CI, obscuring learning (Extended Data Fig. 7h) but has no effect on neuronal DAF-2 degradation-induced memory (Fig. 5h and Extended Data Fig. 8a–d), while loss of CREB from hypodermal DAF-degraded worms completely abrogates memory from 2 h onward (Fig. 5i and Extended Data Fig. 8e–h), suggesting that hypodermal IIS acts through CREB to extend memory, but neuronal IIS does not. Loss of crh-1 decreased the improved learning and memory induced by osm-11 overexpression, as well (Fig. 5j and Extended Data Fig. 8i,j), further suggesting that the memory increase exhibited by OSM-11 overexpression is similarly dependent on CREB. Together, these data suggest that the CREB-independent portion of daf-2’s extended memory (~2 h) is due to its direct DAF-2 reduction in neurons, while hypodermal IIS regulation of CREB via OSM-11 accounts for the remaining fraction of daf-2’s extended memory. We found that CREB is not required in the RIG neuron for the extended memory of hypodermal DAF-2 degradation animals (Extended Data Fig. 8k–m), suggesting that Notch signaling in the RIG must communicate to additional neurons, rather than acting in RIG itself, to regulate CREB.

Reduced hypodermal IIS in old worms slows cognitive aging

We previously found that aged worms lose the ability to learn and to remember with age, but reduction of insulin signaling results in prolonged maintenance of learning and associative memory3. However, these experiments were performed using mutants in which IIS was reduced in all tissues3,10. We wondered whether we could ameliorate aging-induced memory loss by intervening later in life, particularly when animals are already experiencing pronounced memory impairment. (By day 4–5 of adulthood, wild-type worms have no LTAM but can still move (Extended Data Fig. 9a–d); by day 7–8, learning and short-term memory are also abrogated3.) To address this question, we degraded DAF-2 in the hypodermis starting in mid-adulthood, when cognitive function is diminishing in wild-type animals3, and tested the worms’ learning and memory. Because memory declines earlier in adulthood than learning3, we started auxin treatment on day 5 and performed memory assays on day 6, when daf-2 mutants are still capable of learning and memory, but wild-type worms are not (Extended Data Fig. 9)26. Remarkably, DAF-2 degradation specifically in the hypodermis on day 5 significantly increased associative memory in day 6 worms, an age when wild-type worms can learn but have no memory (Fig. 6a and Extended Data Fig. 10a,b), indicating beneficial effects from the reduction of hypodermal IIS in cognitive aging. By contrast, DAF-2 degradation specifically in the neurons on day 5 did not increase associative memory in day 6 worms (Fig. 6b and Extended Data Fig. 10c,d). While both hypodermal and neuronal DAF-2 degradation in young (day 2) worms significantly improves 1-h learning index, only hypodermal DAF-2 degradation improves 1-h learning index in aged (day 6) animals (Fig. 6c and Extended Data Fig. 10e). Notably, the magnitude of memory increase from hypodermal DAF-2 degradation in midlife is comparable to the memory improvement observed in young worms (Fig. 6c), indicating the potential of this midlife intervention to mitigate cognitive aging. Together, these results suggest that reducing hypodermal insulin signaling via DAF-2 degradation specifically in the hypodermis, even in midlife, can enhance associative memory. This difference between hypodermal and neuronal DAF-2 degradation in aged worms highlights the difference between molecular mechanisms involved in the two pathways of IIS-mediated memory extension.

Fig. 6: Hypodermal DAF-2 degradation in aging worms slows cognitive decline through CREB-dependent pathway via Notch ligand OSM-11.
figure 6

a, Hypodermis-specific DAF-2 degradation in old adult animals improves STAM. b, Neuron-specific DAF-2 degradation in old adult animals does not increase STAM. a,b, Auxin started on day 5, while STAM was performed on day 6. c, Comparison of 1-h memory in young (day 2) versus old (day 6) adult animals with hypodermis-specific or neuron-specific DAF-2 degradation. Only hypodermis-specific DAF-2 degradation improves old (day 6) adult memory. d, osm-11 expression decreases with age26. c,d, In the box plots, the center line denotes the median, box range indicates the 25th–75th percentile, and whiskers denote minimum–maximum values. e,f, OSM-11 overexpression increases memory in aged (day 5) adults. CI (e) and learning index (f) are shown. g, Memory was evaluated by calculating the rate of change in CI between the learning and 0.5-h time points across replicates. h,i, Aged (day 6) daf-2(e1370);crh-1(n3315) learning and STAM is impaired relative to daf-2 worms. CI (h) and learning index (i) are shown. j, Slope analysis as in g using learning and 1-h time points. a,b,e,f,h,i, Representative experiment from three biological replicates. Mean ± s.e.m. Two-way repeated-measures ANOVA using data. Bonferroni post hoc tests. n ≥ 4 chemotaxis plates at each time point, where each dot represents an individual assay plate with ~80 worms. Representative experiment for three biological replicates. *P ≤ 0.05, **P ≤ 0.01, ***P ≤ 0.001, ****P ≤ 0.0001. c,d,g,j, Two-sided unpaired t-test.

Source data

Full size image

Our previous whole-worm RNA-seq study of aging animals26 revealed that osm-11 expression decreases with age (Fig. 6d). Worms overexpressing OSM-11 tested at day 5 of adulthood—an age when wild-type worms have poor learning ability and no functional memory3 (see control)—showed both significantly enhanced learning and memory (Fig. 6e–g and Extended Data Fig. 10f). Our data suggest that OSM-11 overexpression even in aged animals is sufficient to confer learning and memory improvements, consistent with our data on hypodermal DAF-2 degradation on older animals.

We previously found that aged daf-2 mutants retain learning and memory longer than do wild-type worms26. To determine if the memory improvement of older daf-2 mutants is the result of an integrated effect of CREB-dependent memory signaling from hypodermal IIS and CREB-independent extended STAM signaling from neuronal IIS, we tested memory duration in daf-2;crh-1 mutants on day 6, an age at which daf-2 mutants still have memory but wild-type worms do not26. While day 6 daf-2 mutants are still capable of memory beyond 3 h after conditioning, daf-2;crh-1 mutants have no learning or memory ability, suggesting that crh-1 is largely responsible for daf-2’s memory extension at mid-adulthood (Fig. 6h–j and Extended Data Fig. 10g). Together, our results indicate that, unlike the combined effects from hypodermal and neuronal IIS signaling in young daf-2 mutants, the enhanced memory of middle-aged daf-2 mutants is fully dependent on hypodermal IIS signaling and CREB activity.

Discussion

In this study, we present evidence of non-autonomous regulation of CREB-dependent associative memory mediated by the insulin pathway specifically in the hypodermis of C. elegans via Notch signaling from the hypodermis to neurons. We had previously identified transcriptional changes in memory7 and expression changes specifically in neurons in daf-2 animals10, but possible contributions to memory function from C. elegans’ non-neuronal tissues were unknown. Therefore, we were surprised to discover that degradation of the DAF-2 receptor solely in the worm’s hypodermis—a large syncytial tissue that we previously found to have a liver-like metabolic transcriptional signal20—increases memory to a greater magnitude and duration than degradation of DAF-2 in neurons.

C. elegans with elevated insulin signaling—that is, the opposite direction of daf-2 mutants or DAF-2 degradation—have learning impairments associated with high neuronal levels of kynurenic acid, whose precursor is generated in epidermal cells48,49, indicating negative involvement of non-neuronal tissues in learning and memory; however, the fact that reduced insulin signaling in non-neuronal tissues could have beneficial effects on memory had not been previously reported. In addition, we previously found that memory was extended beyond 5 h after one training session in daf-2 mutants, but whether any part of that memory is dependent on CREB was previously unknown. Here, we found that DAF-2 degradation exclusively in the hypodermis extends memory even longer than the memory extension induced by neuronal DAF-2 degradation, primarily through CREB activity.

Our single-nucleus sequencing data and subsequent memory assays implicated CREB in the regulation of hypodermal IIS-regulated memory. While memory extension by neuronal DAF-2 depletion is independent of CREB, hypodermal DAF-2 degradation requires CREB activity, indicating that the extended memory exhibited by daf-2 mutants is the result of the combined effect of extended STAM originating from neuronal IIS reduction and the expression of CREB and its targets in hypodermal IIS reduction neurons before training. By contrast with CREB-dependent long-term memory, which forms upon increased CREB activity during spaced training3,7, the improvements in learning and memory we observe in animals with hypodermal DAF-2 degradation are due to increased basal CREB activity and neuronal transcriptional changes prior to training. Furthermore, our transcriptomic and memory assay results suggest that the upregulation of a diffusible Notch ligand, OSM-11, mediates this memory improvement via downstream components of the Notch signaling pathway in neurons, including the Notch receptor LIN-12, the transcription factor LAG-1, and the coactivator SEL-8. Thus, the memory extension exhibited by daf-2 mutants is the result of the integration of enhanced short-term memory regulated by neuronal IIS and the formation of CREB-dependent memory induced by hypodermal IIS regulation of Notch signaling (Extended Data Fig. 10h). Surprisingly, daf-2’s improvement in learning and memory with age relative to wild type is entirely due to the hypodermal IIS–Notch–CREB activity, rather than IIS signaling in neurons. Together, these results define a mechanism for the regulation of memory in C. elegans by a non-neuronal tissue, the hypodermis.

Notch signaling is an evolutionarily conserved pathway, and its roles in development, cancer, and nervous system development in mammals50,51, as well as its role in C. elegans development, have been studied extensively52,53,54. However, Notch signaling’s post-developmental adult roles in C. elegans are limited27,28. Previous studies uncovered the neuron-autonomous role of neuronal Notch signaling in memory regulation in flies and mice: knocking down the Notch receptor specifically in neurons of the mushroom body, the brain center that plays an essential role in Drosophila learning and memory, causes deficits in long-term memory55,56, and postnatal disruption of Notch signaling in the mouse brain causes defects in learning and memory57. Additionally, Notch signaling in the brain regulates mammalian hippocampal plasticity through CREB signaling58 also through cell-autonomous signaling. In C. elegans, Komatsu et al. showed that OSM-11 is a secreted Notch ligand expressed in the hypodermis27, and Singh et al. found that OSM-11 is secreted from the hypodermis and acts non-autonomously to regulate C. elegans octanol avoidance and quiescence28, suggesting that Notch signaling regulates some neuronal functions by receiving signals from peripheral tissues. However, Notch signaling’s regulation of neuronal CREB activity and its role in memory, as well as its regulation by IIS, was previously unknown; while IIS is known to regulate direct effectors of longevity and other functions through the activity of the DAF-16 transcription factor15, its impact on the Notch signaling pathway in C. elegans neurons has not been previously reported. Together with the previous work on Notch signaling in worms, flies, and mice27,28,55,57,58, our data indicate that the IIS–Notch–CREB axis of memory regulation may be conserved.

The C. elegans hypodermis is one of the worm’s largest metabolic tissues and shares transcriptional similarity with human liver and blood plasma20. The hypodermis also aids in adaptation to external environmental changes, such as food scarcity and changes in osmolarity59. Our findings suggest that it is possible that this metabolic tissue may aid in remembering food odors: the hypodermis’ non-autonomous memory regulation might allow a long-lasting mechanism to respond more readily to future environmental changes, with different temporal dynamics than regulation in the neurons. Our observation of non-autonomous regulation of C. elegans neuronal function, and specifically learning and memory, suggests that the animal might have redundant systems in place to modulate memory function, a system that might be helpful under conditions of sustained food limitation. By engaging signaling in the neurons downstream of Notch signaling, worms may be able to have a more sustained memory response than insulin signaling in neurons alone might achieve, particularly in aging animals.

Interestingly, there may be a parallel in the effect of the mammalian liver on memory: the administration of systemic blood plasma from exercised aged mice—which secrete factors from the liver into the blood—improves cognitive function and neurogenesis in sedentary aged recipient mice60. Likewise, Alzheimer’s disease model rats receiving administration of a potent metabolic regulator, fibroblast growth factor 21, which is mainly produced by the liver, also exhibit ameliorated neurodegeneration and enhanced learning and memory61. Together, these data reveal a critical connection between non-neuronal metabolism and memory regulation, suggesting a similar mechanism across different animal models, and providing therapeutic insight to ameliorate cognitive aging or to treat neurodegenerative diseases by regulating metabolism.

Limitations of the study

Our genetic data support a role for LIN-12 function in the RIG neuron in Notch signaling regulation of memory, but we were unable to test the role of sel-8 or lag-1 specifically in the RIG; moreover, our single-nucleus RNA-seq data suggest that upregulation of crh-1/CREB expression in many neurons as a result of OSM-11/Notch signaling is widespread, rather than solely acting through the RIG. The most logical conclusion is that this regulation may occur indirectly via neuropeptide signaling62, of which there are hundreds of pair possibilities.

Methods

General worm maintenance

All strains were cultured using standard methods63. For all experiments, worms were maintained at 20 °C on plates made from nematode growth medium (NGM: 3 g l−1 NaCl, 2.5 g l−1 Bacto-peptone, 17 g l−1 Bacto-agar in distilled water, with 1 ml l−1 cholesterol (5 mg ml−1 in ethanol), 1 ml l−1 1 M CaCl2, 1 ml l−1 1 M MgSO4 and 25 ml l−1 1 M potassium phosphate buffer (pH 6.0) added to molten agar after autoclaving) or high growth medium (HGM: NGM recipe modified as follows: 20 g l−1 Bacto-peptone, 30 g l l−1 Bacto-agar and 4 ml l−1 cholesterol (5 mg ml−1 in ethanol); all other components same as NGM), with OP50 Escherichia coli for ad libitum feeding. To synchronize experimental animals, eggs were collected from gravid hermaphrodites by exposing the animals to an alkaline-bleach solution (for example, 1.5 ml sodium hypochlorite, 0.5 ml 5 N KOH, 8.0 ml water), followed by repeated washing of collected eggs in M9 buffer (6 g l−1 Na2HPO4, 3 g l−1 KH2PO4, 5 g l−1 NaCl and 1 ml−1 1 M MgSO4 in distilled water).

C. elegans strains

Strains are listed in Supplementary Table 5.

Primers used in this study

Three promoters were used to drive the sense and antisense expression of target genes: Pttx-3 (AIY interneurons), Ptwk-3 (RIG interneurons) and Pnmr-1 (RIM and AVA interneurons).

ttx-3 promoter64:

ttx-3 Pf = tcatgcatattcgattttttcagtaaacg

ttx-3 Pr = tttgacaccgaagacaattattatg

twk-3 promoter36:

twk-3 Pf = tacgccatccaattagtattttatgtctg

twk-3 Pr = ttttaagaagaacaaggaaaaagttgaag

nmr-1 promoter65:

nmr-1 Pf = gatgattatggaaccaaactcagaatttaatg

nmr-1 Pr= atctgtaacaaaactaaagtttgtcgtg

lin-12 for sense and antisense constructs:

Pf = atgcggatccctacgatttg

Pr = acactagccccttgctgaatc

lag-1 for sense and antisense constructs:

Pf = ggatgtggaatggaaattggagg

Pr = ctcttccatgcaactcgacacgc

glp-1 for sense and antisense constructs:

Pf = gtcctgactgcaaaactcctctattttc

Pr = acaaatgttgctcgcagttcaatccatag

crh-1 for sense and antisense constructs:

Pf = atggccacaatggcgagcacctc

Pr = tcacattccgtccttttcctttcggc

PCR products of twk-3p- or ttx-3-driven sense and antisense constructs were injected at 100 ng μl−1 each with 2 ng μl−1 myo-2p::GFP marker.

CQ801 was generated by injection at 10 ng µl−1 of a plasmid containing the genomic coding sequence of osm-11 gene with the unc-54 3′ untranslated region expressed under the control of 2.5 kb of upstream genomic DNA. As a co-injection marker and reporter of osm-11 expression, the same osm-11 promoter was used to express GFP with an unc-54 3′ untranslated region.

Auxin treatment

For auxin experiments, the standard HGM molten agar was supplemented with 2.5 ml 400 mM indole-3-acetic acid per 1 l of HGM: Alfa Aesar (A10556) freshly prepared in ethanol and plates were seeded with E. coli for ad libitum feeding. Synchronized AID worms were transferred to HGM with auxin plates for 16–20 h before behavior assay. For midlife experiments, worms were transferred at the L4 larval stage onto HGM plates supplemented with 500 ml l−1 0.1 M 5-Fluoro-2′-deoxyuridine (FUDR) for a final concentration of 0.05 M FUDR and were transferred back to standard HGM with or without auxin 20 h before memory assay.

RNAi treatment

For RNAi experiments, the standard HGM molten agar was supplemented with 1 ml l−1 1 M isopropyl β-d-1-thiogalactopyranoside and 1 ml l−1 100 mg ml−1 carbenicillin, and plates were seeded with HT115 E. coli for ad libitum feeding. All RNAi treatment starts from L4 larval stage and lasts for 3 days. RNAi experiments were performed using the standard feeding RNAi method. Bacterial clones expressing the control construct (empty vector, pL4440). All RNAi clones were sequenced before use.

Pavlovian appetitive associative assay

Animals were trained and tested for short-term memory as previously described3. Briefly, synchronized young or aged adult hermaphrodites were washed from HGM, RNAi or HGM supplemented with auxin plates with M9 buffer, allowed to settle by gravity, and repeatedly washed three more times with M9 buffer. Then animals were starved for 1 h in M9 buffer. For conditioning (food and 10% 2-butanone pairing), worms were then transferred to 10-cm NGM conditioning plates (seeded with OP50 E. coli bacteria and with 18 ml 10% 2-butanone (Acros Organics) dissolved in ethanol on the lid) for 1 h at 20 °C. After conditioning, the trained worms were tested for chemotaxis toward 10% butanone versus an ethanol control either immediately (0 h) or after being transferred to 10-cm NGM holding plates with fresh OP50 E. coli bacteria for specified time intervals (30 min–2 h). Before spotting 10% butanone or ethanol control on assay plates, 10% sodium azide (diluted using two parts sodium azide and one part nuclease-free water) was added to paralyze worms for analysis of the populations arriving at each odorant spot.

Chemotaxis indices were calculated as follows:

$${\rm{{Chemotaxis}}\; {\rm{index}}=\frac{\rm{no. of worms Butanone}-\rm {no. of worms Ethanol}}{\rm{no. of worms Total}-\rm{no. of worms Origin}}}$$

The calculation for the learning index is:

$${\rm{Learning}}\; {\rm{index}}={\rm{Chemotaxis}}\; {\rm{IndexTrained}}-{\rm{Chemotaxis}}\; {\rm{IndexNaive}}$$

Learning indices for extrachromosomal transgenic strains were analyzed by hand counting GFP-positive and GFP-negative worms at different locations at individual time points on the chemotaxis plates. Control animals for these experiments were the transgenic worms’ GFP-negative siblings.

Chemotaxis assay

Synchronized adult worms were tested for chemotaxis to 1% benzaldehyde in ethanol or 10% pyrazine in ethanol, using standard, previously described chemotaxis assay conditions46.

Microscopy

For DiI, worms were grown on regular seeded HGMs until day 2 adults (for young worms DiI staining), or until L4 and transferred to HGMs supplemented with FUDR to day 4 and then transferred to regular HGMs until day 5 (for older worms DiI staining). On the day of imaging, well-fed worms were washed three times by M9, and then resuspended in 1 ml M9 with 5 μl DiI stock solution (2 mg ml−1 DiI (Molecular Probes, D-282) in dimethyl formamide) for 3 h on a slow shaker at room temperature. Worms were washed with M9 twice before transferring them onto agar pads with sodium azide to visualize by Nikon Eclipse Ti microscope. z-stack multi-channel (DIC, mCherry/RFP and GFP) images of day 2 or 5 adult worms were acquired at ×60 magnification. Images were analyzed using Nikon NIS elements software (v5.42.03). For Auxin imaging experiments, worms were treated with auxin per the auxin treatment method.

Worm FACS (population enrichment)

Array-carrying adult D1 worms were sorted using COPAS Large Particle Sorter to select for fluorescent pharynx (array-carrying worms). This enriched population was bleached to synchronize eggs, and once grown, D1 animals were subsequently moved to auxin for STAM experiment, where fluorescent worms were counted manually.

RNA isolation

Synchronized day 2 adult worms were collected in M9 and washed repeatedly 3 times to remove excess bacteria. Worm pellets were crushed in liquid nitrogen and transferred to 850 μl TRIzol LS. Total RNA was extracted using the standard TRIzol/chloroform/isopropanol method followed by DNase digestion and Qiagen RNeasy Mini kit. Agilent Bioanalyzer RNA Pico chips were used to assess the quality and quantity of isolated RNA. mRNA libraries were prepared using the RNA-seq directional library prep on the Apollo 324 robot and were sequenced (65-nucleotide paired-end) on the Illumina NovaSeq S1 100-nucleotide flowcell v1.5 platform (yields ~1.3–1.6 billion reads).

RNA-seq data analysis

RNA-seq analysis was performed as previously described. FASTQC was used to assess read quality scores. The universal Illumina adaptor sequences were trimmed using Cutadapt v1.666. The trimmed reads were mapped to the C. elegans genome (UCSC Feb 2013, ce11/ws245) using STAR67. The reads aligning to individual genes were counted using htseq-counts (mode: union), and DESeq2 was used for differential expression analysis. Genes with an adjusted P value \(\le\) 0.05 were considered significantly differentially expressed.

Tissue query

worm.princeton.edu was used for tissue query from upregulated or downregulated gene lists (DESeq2 genes FDR < 1%).

Secreted protein prediction

SignalP-5.0 (DTU Health Tech) was used for secreted protein prediction from 75 genes exclusively expressed in the hypodermis.

Neuronal nuclei isolation

C. elegans neuronal nuclei were isolated using the following methods37, modified from the mammalian single-nucleus RNA-seq protocol from 10x Genomics. Worms (CQ756) grew on regular HG until day 1 and were transferred to HG or HG + auxin plates on day 1. Then, ~400 μl of day 2 worms CQ756 were washed from HG or HG + auxin plates. Samples were washed 3× in M9 buffer and worm pellets were kept on ice until douncing. The 400 μl worm pellet was transferred to a 1-ml dounce homogenizer (Kimble Glass Tissue Homogenizer, 88-mm overall length, Dounce 1 ml working capacity; 885300-0001) filled with 300 μl of NP40 lysis buffer (10 mM Tris-HCl (Sigma-Aldrich, T2194; pH 7.4), 10 mM NaCl (Sigma-Aldrich, 59222 C; 5 M), 3 mM MgCl2 (Sigma-Aldrich, M1028; 1 M), 0.05% Nonidet P40 Substitute (Sigma-Aldrich, 74385), 1 mM dithiothreitol (Sigma-Aldrich, 646563), 1 U μl−1 RNase inhibitor (Sigma Protector RNase inhibitor; Sigma-Aldrich, 3335402001), Nuclease-free water). Each sample was Dounce homogenized 25–35×, using only the tight pestle. After 15 strokes, the degree of lysis was examined every five strokes, until appropriate lysis was achieved (no intact worms, with small worm fragments and pieces of empty cuticle present). Samples were moved to 2 ml low-binding microcentrifuge tubes and Dounce homogenizers were rinsed with 1 ml of NP40 lysis buffer; this rinse was added to the tubes. Samples were then incubated for 5 min on ice and pipetted twice after 2 min and 4 min, 10× each, with a P200 pipette. Suspensions were passed through a 40-μm filter into a 15-ml conical tube (each filter was rinsed with an additional 250 μl of lysis buffer), and then transferred to a 2-ml low-binding microcentrifuge tube. Samples were centrifuged at 1,000g for 5 min at 4 °C. The supernatant was removed, leaving ~100 μl behind, and 1 ml of wash buffer (PBS + 0.5% BSA (Miltenyi Biotec, 130-091-376) + 1 U ul−1 of RNAase inhibitor) was added with no mixing for 5 min on ice. After 5 min, the pellet was resuspended by mixing 3× with a P1000. Samples were centrifuged again at 1,000g for 5 min at 4 °C. The supernatant was removed, leaving ~100 μl behind, and the pellet was then resuspended in 500 μl of wash buffer. Hoechst stain (1:10,000 dilution; Molecular Probes Hoechst 33342, Thermo Fisher, H3570) was added to each sample, and samples were then passed through 5-μm syringe filters (pre-wetted with 500 μl of wash buffer) using a 1-ml syringe, directly into FACS tubes. Samples were incubated for at least 5 min on ice before FACS.

FACS

Hoechst and GFP-positive 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-binding microcentrifuge tube containing collection buffer (500 μl of 0.5% BSA + 1 U per μl RNase inhibitor). The instrument was washed with bleach between samples. Around 150,000–200,000 nuclei per sample were sorted for further analysis.

Single-nucleus RNA-seq library preparation and sequencing

After FACS, samples were centrifuged at 1,000g for 5 min at 4 °C. The supernatant was removed and nuclei were resuspended in 20 μl of collection buffer. The total number of nuclei for each sample was estimated (~3/4 loss) and single-nuclei suspension samples were loaded to the 10x Genomics Chromium X system using the Single Cell 3′ v3.1 Reagent Kits (10x Genomics) to generate and amplify cDNA. The amplified cDNA samples were purified with Ampure XP magnetic beads (Beckman Coulter), quantified by a Qubit fluorometer (Invitrogen), and examined on a Bioanalyzer with High Sensitivity DNA chips (Agilent) for size distribution. Illumina sequencing libraries were generated from the amplified cDNA samples using the Illumina Tagment DNA Enzyme and Buffer kit (Illumina). These libraries were examined by Qubit and Bioanalyzer, then pooled at equal molar amounts and sequenced on Illumina NovaSeq 6000 S Prime flowcells as 28 + 94-nucleotide paired-end reads following the standard protocol. Raw sequencing reads were filtered by Illumina NovaSeq Control Software and only the Pass-Filter reads were used for further analysis.

Alignment and quality control of data

Alignment of reads was performed using Cell Ranger version 7.1.0. SoupX was used to remove ambient RNA contamination on the Cell Ranger output files. To subset the data that represent broken or damaged nuclei, doublets or empty droplets, we set cutoffs by checking violin plots of genes per cell: the lower bound was 150–210 features per cell for each sample and the higher bound was between 750 and 1,000 features per cell depending on the sample.

Normalization, integration and clustering

In Seurat, we used the single-cell pipeline, where single-cell transform was used for normalization. Different numbers of principal components (80, 90, 100, 110, 120, 125, 130, 150, 200) and different resolutions (0.8, 0.9, 1.0 and 1.2) were used. In total, 125 PCs at a clustering resolution of 1 was used, which resulted in 106 clusters in our dataset.

Cluster labeling/cell-type analysis

We performed cluster annotation using a combination of a hypergeometric test and AUCell algorithm as described previously68. Detailed methods for cluster labeled are available in ref. 37.

Differentially expressed gene identification

The FindMarkers function was used in the Seurat package for differential expression identification. The Wilcoxon rank-sum test method was performed on each subsetted cluster, comparing auxin (−) cells and auxin (+) cells within the same cluster. Genes with a minimum percentage of expression in 25% of the cells were analyzed, and the significantly differentially expressed genes were identified if their log2(fold change) > 0.1 or < −0.1, and adjusted P value < 0.05. The genes significantly differentially expressed in at least one cluster are considered differentially expressed. gProfiler was used to identify Gene Ontology and KEGG terms enriched among all of the hypodermal DAF-2(−) upregulated genes (FDR < 0.05) from the single-nucleus RNA-seq dataset.

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.

Lifespan analysis

Lifespan analyses were performed for osm-11-overexpressing worms. Transgenic worms or their wild-type sibling controls were segregated at the L4 larval stage to start the assay. Every other day, worms were transferred to freshly seeded NGM plates. The first day of adulthood was defined at T = 0. Kaplin–Meier analysis with the log-rank (Mantel–Cox) method was used to compare lifespans between transgenic and non-transgenic siblings. Worms that ‘escaped’ or ‘bagged’ were censored on the day of the event (n = 96 per strain; two biological replicates were performed).

Dauer assay

Day 2 adult N2 and daf-2(e1370) worms were allowed to lay eggs for 1 h onto OP50-seeded NGM plates at 20 °C. For intestinal DAF-2 AID worms, eggs were laid on OP50-seeded NGM plates containing 0.25% ethanol or 1 mM auxin. Adults were removed and egg-containing plates were incubated at 27 °C for 72 h. The number of dauer and non-dauer worms were counted to calculate the percentage of dauers.

Statistics and reproducibility

Two-way ANOVA with Bonferroni post hoc tests were used to compare learning indices among multiple groups. One-way ANOVAs followed by Bonferroni post hoc tests for multiple comparisons were performed. Two-way ANOVAs were used between genotype (daf-2(e1370) and wild-type, control RNAi and daf-2 RNAi, control and osm-11 overexpression, hypodermal DAF-2 AID and hypodermal DAF-2 AID;crh-1(3315)) and different time point (0 h, 0.5 h, 1 h, 2 h, 3 h, 4 h, 5 h and 6 h) on learning indices with a significant interaction between factors leading to the performance of Bonferroni post hoc comparisons to determine differences between individual groups. Prism 10 software was used for all statistical analyses of behavior data using standard settings for ANOVA and post hoc tests. 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.

Experiments were repeated on separate days, using separate independent populations, to confirm that results were reproducible, and all replicates of memory assays are provided. Software and statistical details used for RNA-seq analyses are described in the Methods. No statistical methods were used to predetermine sample sizes, but our sample sizes are similar to those reported in previous publications3,8,23. Data distribution was assumed to be normal but this was not formally tested. Data collection and analysis were not performed blind to the conditions of the experiments. For feeding RNAi experiments and +/− auxin experiments, worms were randomly selected for each experiment for placement into the different experimental groups. This was achieved by randomly pipetting populations of worms into different groups from pooled starting/growth plates. No animals or data points were excluded from the analyses for any reason.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.