Because C. elegans has several different mechanisms of pathogen avoidance that can be detected via different signals (olfactory cues, gustatory cues, illness-induced signals, bacterial metabolites, bacterial RNAs, etc.), pathogen avoidance behavior has been tested using several different assays (as reviewed in Liu and Zhang, 2020). Among these are variations on the bacterial choice assay, which is modeled on standard chemotaxis choice assays (two small bacterial spots with azide to trap worms at their first choice) (Bargmann et al., 1993; Hart, 2006; Zhang et al., 2005; Jin et al., 2016; Lee and Mylonakis, 2017; Ooi and Prahlad, 2017; Liu et al., 2018; Moore et al., 2019), variations on single-worm head-turning assays (Ha et al., 2010), and the leaving assay (PA14 spot in middle of plate, no other bacteria, no azide, and monitored over long periods for worms leaving the spot, e.g. 8–24 hr) (Meisel et al., 2014; Miller et al., 2015; Ma et al., 2017; Horspool and Chang, 2017; Hao et al., 2018; Singh and Aballay, 2019; Wolfe et al., 2019; Filipowicz et al., 2022). Each assay measures responses on a specific timescale; e.g., the two-bacteria choice assay measures initial preference of a population of animals over an hour, the head-turning assay measures immediate preference of individual animals, and the leaving assay (which is often performed overnight) measures final preference over many hours of PA14 exposure. Therefore, the conditions used in one assay might not be appropriate for all assays.

The standard bacterial choice assay is performed on plates, with two small spots of bacteria (here, PA14 and OP50) separated by several centimeters that the worms must traverse in response to olfactory cues (Figure 1A); monitoring the accumulation of worms as they become paralyzed with the anesthetic sodium azide at each bacterial spot allows the experimenter to accurately assess the worm’s first choice. Studies by many groups, including Zhang et al., 2005; Meisel et al., 2014; Ooi and Prahlad, 2017; Pereira et al., 2020; Filipowicz et al., 2022, and others over the past two decades have previously demonstrated that C. elegans are initially attracted to PA14 without prior exposure to the pathogen (‘naïve’) but switch their behavior to avoidance of PA14 following exposure to PA14. Our lab has also consistently replicated those naïve attraction and learned avoidance results: in fact, we see that the mild naïve attraction to PA14 is fairly consistent between individual plates and individual replicates over many experiments across several studies (Figure 1B–D, Supplementary file 1; Moore et al., 2019; Kaletsky et al., 2020; Moore et al., 2021; Sengupta et al., 2024; Seto et al., 2025). This initial naïve attraction to PA14 serves as a necessary baseline for all further experiments. That is, if no naïve attraction to PA14 is observed and instead the worms already avoid PA14 prior to PA14 exposure, it might be difficult to observe any additional learned avoidance of PA14 after training. Failure to observe naïve attraction to PA14 might also suggest that the experimenters’ assay conditions are not the same as those used by previous groups. A recent study failed to consistently observe this initial naïve avoidance of PA14 (Gainey et al., 2024); therefore, we investigated possible origins of this deviation in these results from other groups’ previously published results.

To determine possible origins of deviations in naïve choice assay results, we carried out the two-bacteria choice assay under standard conditions (1 hr of a binary choice of PA14 vs OP50, which includes the paralytic sodium azide added to the bacterial spots) as well as the conditions reported by Gainey et al., which primarily differs by omitting azide and letting the worms choose in the absence of azide for 1 hr at room temperature, and then moving the plates to cold temperature (4°C) prior to counting, as described by Gainey et al., 2024. We found that when azide is present at the bacterial spots to trap naïve worms at their first choice, the worms demonstrate attraction to PA14, as has been previously shown, but in the absence of azide, worms show significant avoidance of PA14 even without prior exposure to the pathogen (Figure 1E and F); the difference between the +azide and -azide chemotaxis indices is both significant and large in magnitude (a difference of 0.57 in CI and p<0.0001; Figure 1F). We note that these no-azide results are almost identical to the results that Gainey et al. reported for their naive P0s, while the +azide results are similar to our previously published naïve data (Figure 1B–D), as well as published data from other labs.

To better understand the source of the discrepancy with the Gainey et al. results, we first investigated the worms’ behavior at 4°C after the hour of the assay at room temperature; while we do see that the worms continue to move (4°C does not immediately paralyze worms), in fact, the final choice index of the population was already reached at the end of the room temperature period before they were placed in the cold (Figure 1G, t=0); therefore, the discrepancy in attraction to/avoidance of PA14 seems to have occurred during the initial part of the assay, when the worms were choosing for 1 hr at room temperature without azide. This may be particularly relevant for the Gainey et al. study, as they reportedly moved the worms at different times (‘These no-azide assay plates were moved to 4°C after 30–60 min’), which likely induced further variance. We also observed worms still traveling away from the PA14 spot at 1 hr, further underscoring the point that performing the assay without azide fails to trap the worms at their first choice, even if the worms are transferred to 4°C, because the worms have already made the switch from attraction to avoidance by that point.

The simplest interpretation of the discrepancy between the ±azide assay conditions is that the worms are initially attracted to and choose PA14; in the presence of azide, the worms remain at this first-choice spot, but over the course of an hour-long choice assay in the absence of azide, the worms learn to avoid PA14 and migrate to the OP50 spot. This phenomenon was already carefully demonstrated by Ooi and Prahlad, 2017, with a time-course assay: in the absence of azide to trap the worms, the worms are initially attracted to PA14, but by 45 min the worms show no preference for PA14, and by 1 hr the worms instead avoid PA14 – i.e., within the time of the hour-long assay, in the absence of azide to trap the worms at their first choice, they learn to avoid PA14. Under the no-azide assay conditions, it would be difficult to observe any additional learned avoidance that is the result of training, since the learned preference for OP50 (PA14) is already very high when the worms have been allowed to change from their first choice to their second choice. Therefore, the simplest interpretation is that one cannot accurately measure naïve preference in the absence of azide, and one cannot subsequently properly assess subsequent learned avoidance after training without azide. It is notable that the overwhelming majority of Gainey et al.’s experiments were done without azide; because a third party (Ooi and Prahlad, 2017) already showed that there is a substantial difference when no azide is used, as we have shown here, Gainey et al.’s no-azide data cannot be compared with assays done with azide, including all of our work.

We repeated this ±azide preference assay with another Pseudomonas pathogen, P. fluorescens 15 (PF15), that we have also found to induce transgenerational avoidance for five generations via small RNAs (Seto et al., 2025) and observed similar results (Figure 1H and I). That is, the preference for PF15 is similar to worms’ preference for PA14 over OP50, but in the absence of azide, the worms learn to avoid PF15, as they did with PA14. Therefore, we conclude that the inclusion of azide in the choice assay is necessary to properly assess the worms’ first choice; in the absence of azide, the worms’ naïve choice is not being measured, so it cannot be used to then compare with post-training behavior to assess learned or inherited avoidance.

C. elegans uses many different environmental cues to inform its behavioral decisions; in fact, avoidance of PA14 can be induced by bacterial metabolites, by olfactory cues, by gustatory cues, and by innate immunity pathways. These cues induce avoidance in the P0 generation, and they and other mechanisms can induce avoidance and changes in survival or development in the F1 generation via distinct mechanisms (Pereira et al., 2020; Hong et al., 2021; Burton et al., 2020; Pender et al., 2025), but do not extend beyond the F1 generation. Our studies used a single condition for training, PA14 grown on plates at 25°C and training at 20°C for 24 hr (see Figure 1A). Under these conditions, we consistently observe learned avoidance in mothers (P0; Figure 2A, B, and C) that is then transgenerationally inherited (F2; Figure 2D–F). Therefore, our PA14 growth and/or training conditions must be distinct from those growth and training paradigms that only produce P0 or F1 effects.

In searching for the relevant TEI-inducing factor, we tested and eliminated DNA, proteins, and large RNAs as the TEI-inducing component and found that small RNAs isolated from PA14 grown at 25°C on plates could induce avoidance and TEI (Kaletsky et al., 2020). That is, small RNAs isolated from 25°C plate-grown PA14 pipetted onto Escherichia coli prior to 24 hr C. elegans training were able to induce avoidance in those animals and the transgenerational inheritance of that avoidance, just as we had observed from 24 hr PA14 lawn-training experiments. We had also noticed that PA14 grown at lower temperatures or in liquid could not induce transgenerational inheritance of learned avoidance. Capitalizing on these observations, we used differential small RNA sequencing to identify the set of small RNAs that are expressed in PA14 grown at 25°C on plates that are not expressed in liquid or at low temperatures; we then cloned each of those six small RNAs into E. coli and tested these individual sRNAs for their ability to induce avoidance and TEI (Kaletsky et al., 2020). This set of experiments revealed that the small RNA P11 – and only P11 – could induce avoidance in mothers and inheritance of avoidance for four generations (Kaletsky et al., 2020). C. elegans trained on E. coli expressing P11 sRNA induced the P0-F4 avoidance of PA14 without making the worms sick and without inducing innate immune pathways, but while also inducing daf-7::gfp expression in the ASI (Kaletsky et al., 2020). We observed that E. coli-P11 (indicated by the yellow points) consistently induces P0 avoidance (Figure 3A and B), and this learned avoidance is inherited transgenerationally (Figure 3C and D). Therefore, P11 sRNA is both necessary and sufficient to induce the transgenerational inheritance of learned PA14 avoidance.

We noted that P0 avoidance induced by E. coli-P11 training (Figure 3A and B) is slightly lower than that induced by training on PA14 bacterial lawns (Figure 2B and C), suggesting that the P0 innate immunity-induced and other avoidance mechanisms are distinct from and additive with the sRNA pathway in mothers. In fact, the average level of avoidance induced by E. coli-P11 training is consistent in P0 and F2 (Figure 3C and D), and with the F2 avoidance levels induced by PA14 training (Figure 2E and F), consistent with our previous reports (Kaletsky et al., 2020; Moore et al., 2019). P11-induced avoidance remains consistent from P0 through F4 (Kaletsky et al., 2020), as we later also found for the small RNAs (Pv1, Pfs1) required for Pseudomonas vranovensis (Sengupta et al., 2024) and PF15 (Seto et al., 2025) avoidance. Similarly, while E. coli-P11 training induced daf-7::gfp expression in the ASI from P0 through F4 (Kaletsky et al., 2020), we did not observe any daf-7::gfp expression in the ASJ, consistent with the ASJ response being specifically induced by metabolites produced by PA14 (Meisel et al., 2014). Therefore, it appears that there are PA14 growth conditions that can induce some mechanisms of P0 and F1 avoidance, but will not induce transgenerational avoidance, because no P11 is expressed in PA14 under those conditions.

To more carefully address this hypothesis, we examined the levels of P11 sRNA from PA14 grown at 25°C on plates, 15°C on plates, and 37°C in liquid (Figure 3E, adapted from Kaletsky et al., 2020). When comparing 25°C plates with 15°C plates, we observed a 6.8-fold decrease in P11 levels at the lower temperature, and a 13.7-fold decrease in P11 when PA14 is grown in liquid rather than on plates, even at high temperatures. Therefore, it is clear that the growth of PA14 in non-25°C/plate-grown conditions may fail to elicit a P11-mediated avoidance response, because no P11 is expressed in PA14 under those conditions. Because Gainey et al., 2024, did not verify P11 expression in their bacterial cultures, it is not surprising that they also failed to observe P11-induced avoidance in the F2 generation in those experiments.

Previously, we showed that many components of small RNA uptake, RNA interference, small RNA processing pathways, and the germline are necessary for transgenerational inheritance of learned avoidance (Moore et al., 2019; Kaletsky et al., 2020). We find that the meg-1;meg-3 double mutants, which perturb P granules and the germline (Wang et al., 2014), can avoid PA14 after lawn training (Figure 3F) but are defective for small RNA-induced learning (worms trained on E. coli with small RNA purified from PA14 added to the lawn) (Figure 3G). Similarly, NRDE-3, an Argonaute protein that functions in nuclear RNAi (Guang et al., 2008), is also required for small RNA-induced learned avoidance (Figure 3H-J). Like other mutants required for the small RNA TEI pathway, the P0 generation of meg-3;meg-4 and nrde-3 mutants still avoid PA14 after training on intact PA14 lawns (Figure 3F and H) through other avoidance pathways (innate immunity, metabolites, etc.), once again highlighting the independence of these different avoidance-inducing pathways. Only the small RNA-induced avoidance is transgenerational, i.e., F2s of wild-type worms continue to avoid PA14 after lawn training, but F2s of nrde-3 mutants are defective for this inherited small RNA-induced avoidance (Figure 3J).

We also find that rde-3/mut-2, which encodes a ribonucleotidyltransferase that is involved in siRNA accumulation, piRNA-mediated silencing, and p(UG)ylation of small RNAs (Chen et al., 2005; Shukla et al., 2020; Priyadarshini et al., 2022) in epigenetic inheritance is also required for E. coli P11-induced learned avoidance (Figure 3K). Thus, components of small RNA transport, small RNA processing, RNAi function, germline components, chromatin factors, neuronal function, and the Cer1 retrotransposon (Moore et al., 2021) are required for the transgenerational inheritance of learned pathogen avoidance (Figure 4A).

Here, we have investigated two experimental factors that might affect investigators’ interpretation of PA14 training-induced avoidance choice assays: the expression of P11 sRNA in PA14 and proper execution of the choice assay itself. It would be incorrect to assume that because F2 avoidance is not demonstrated under all PA14 growth conditions that ‘the published model is unlikely to be relevant in a natural environment’ (Gainey et al., 2024). Instead, our results suggest the opposite: that the worms’ ability to interpret P11 expression is due to P11’s importance in regulating nitrogen fixation under biofilm-inducing conditions that also induce a pathogenic state that is important for C. elegans to remember over several generations. C. elegans appears to have evolved a mechanism to use P11 and other specific small RNAs as biomarkers of future infection, as the presence of those specific small RNAs correlates well with pathogenesis – and the specific small RNA detected by C. elegans in each case might be tightly correlated with the pathogenic potential of that particular Pseudomonas species. Furthermore, the fact that we found that other Pseudomonas species, including Pseudomonas from the worm’s microbiota, use the same bacterial sRNA/C. elegans target gene mechanism indicates that this natural example of trans-kingdom signaling and transgenerational avoidance is highly conserved. In those bacteria, the key small RNAs appear to be constitutively expressed, and thus even more likely to be encountered by C. elegans upon ingestion. Together, these data suggest that the transgenerational inheritance of small RNA-induced learned bacterial avoidance is likely to be widespread in the wild, indicating its likely importance as a mechanism to promote long-term survival of C. elegans in its natural habitat.

Worm strains were maintained at 20°C on high growth medium (HG) plates (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 1 M CaCl₂, 1 mL/L 1 M MgSO₄, and 25 mL/L 1 M potassium phosphate buffer [pH 6.0] added to molten agar after autoclaving) on OP50 using standard methods. The following worm strains were provided by the C. elegans Genetics Center (CGC): JH3229, meg-1(vr10); meg-3(tm4259) X; YY158, nrde-3(gg66) X; WM30, mut-2/rde-3(ne298) I.

OP50 was provided by the CGC. PA14 was a gift from Z. Gitai. OP50 and PA14 were grown in liquid cultures in LB at 37°C with shaking. E. coli strains expressing PA14 sRNA were cultured overnight in Luria broth supplemented with 0.02% arabinose wt/vol and 100 μg/mL carbenicillin. Overnight cultures of bacteria were diluted in LB to an optical density at 600 nm (OD 600)=1 and used to fully cover 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 1 M CaCl₂, 1 mL/L 1 M MgSO₄ and 25 mL/L 1 M potassium phosphate buffer [pH 6.0] added to molten agar after autoclaving) plates. For the preparation of control E. coli and E. coli expressing the PA14 P11 small RNA, bacteria were seeded on NGM plates supplemented with 0.02% arabinose and 100 μg/mL carbenicillin. All plates were incubated for 2 days at 25°C, unless specified otherwise (in separate incubators for control and pathogen-seeded plates). OP50 and PA14 small RNAs were collected as described in Kaletsky et al., 2020.

Worm preparation for training: synchronized L4 worms were washed off plates using M9 and left to pellet on the bench top for approximately 5 min. Then, 5 μL of worms were placed onto sRNA-spotted training plates, and 10 μL or 40 μL of worms were plated onto OP50 or E. coli expressing control/P11, or PA14-seeded training plates, respectively. Worms were incubated on training plates at 20°C in separate containers for 24 hr. After 24 hr, worms were washed off plates using M9 and washed an additional three times to remove excess bacteria.

Choice plates and naïve day 1 worms were prepared as in Moore et al., 2019, and Seto et al., 2025. On the day of the assay, choice assay plates were left at room temperature for 1 hr before use. For standard (sodium azide added) experiments, 1 μL of 1 M sodium azide was spotted onto each respective bacteria spot to be used as a paralyzing agent during choice assay and preserve first bacterial choice. For conditions without azide, no azide was added.

Worms were washed off training plates in 2 mL M9 into 1.5 mL tubes and allowed to pellet by gravity. Worms were washed two to three additional times in M9. Using a wide orifice pipet tip, 5 μL of worms (approximately 50–200 worms) were spotted at the bottom of the assay plate, midway between the bacterial spots. Plates were incubated at room temperature for 1 hr before moving plates to 4°C. Plates were left at 4°C for ~1 hr before manually counting the number of worms on each bacterial spot.

We collected all examples of N2 naïve chemotaxis (PA14 vs OP50 choice assays) prior to training, N2 P0 and F2 choice assays after P0 training on PA14 vs OP50, and P0 and F2 choice assays after P0 training on E. coli-P11 vs control E. coli from Moore et al., 2019, Kaletsky et al., 2020, Moore et al., 2021. These data are collected in Supplementary file 2. Multiple two-group estimation plots were generated for side-by-side comparison of multiple mean differences across independent experiments. Each mean difference is plotted as a bootstrap sampling distribution.