The coupling between healthspan and lifespan in Caenorhabditis depends on complex interactions between compound intervention and genetic background

Oxidative stress response varies in a compound-, strain-, and age-specific manner

Reactive oxygen species can serve as critical signaling molecules that promote healthy biology, but when ROS production and/or defenses become imbalanced with age, ROS can promote aging [44, 45]. Cells experience oxidative stress due to metabolic activity and environmental stressors. The ability to resist oxidative stress is fundamental to the maintenance of cellular function [46], and therefore oxidative stress resistance is commonly used measure of animal health. Of note, C. elegans N2 lifespan has been linked to oxidative stress resistance, with long-lived C. elegans insulin pathway mutants demonstrating increased survival under oxidative stress [27]. We sought to assess whether NP1, resveratrol, and propyl gallate, compounds that showed species- and strain-specific effects on lifespan, would exhibit the same pattern of effects on oxidative stress resistance, and if NP1, propyl gallate or resveratrol could confer health effects in strains that did not respond to those compounds with lifespan increases. We tested oxidative stress resistance in the presence of superoxide generator paraquat at two ages: early mid-life (adult day 6 for C. elegans, adult day 8 for C. briggsae; time points selected based on our previously detailed survival analyses in these genetically diverse backgrounds). We also tested older ages for the C. briggsae strains (adult day 16 for C. briggsae vs. adult day 12 for C. elegans, as C. briggsae display a later age of senescent decline [34]) to investigate whether compound effects on oxidative stress resistance were age dependent. Specifically, the different time assays enable us to assess if longevity-promoting compounds ameliorate age-related declines in oxidative stress resistance, rather than merely increasing oxidative stress resistance overall.

We found the effects of DR mimetic NP1 on oxidative stress resistance to vary in tested strains across both elegans and briggsae species (Figure 2A; Supplementary Figure 1). In particular, the strong species-specific effect that we previously reported for lifespan with NP1 treatment [34] was not evident for oxidative stress resistance. More specifically, one strain, C. briggsae AF16, showed a small age-related increase in oxidative stress resistance (no significant effect at early mid-life; 15% increase in median survival at late mid-life, p = 0.0288). A second strain, C. elegans MY16, demonstrated an age-by-compound interaction, with a moderately negative oxidative stress response at early mid-life (25% decrease in median survival, p = 0.00135), but a robust positive response at late mid-life age (61% increase in median survival, p = 0.00477). C. briggsae HK104 had an overall robust decrease in oxidative stress resistance at both ages (33–35% decrease in median survival, p < 0.001; we note that this mirrors the decreased lifespan previously seen with HK104 NP1 treatment [34]). The three remaining strains C. elegans N2 and JU775, and C. briggsae JU1348, showed no significant effect of NP1 treatment on oxidative stress resistance. We conclude that NP1 treatment in genetically diverse backgrounds induces a range of oxidative stress resistance responses. Importantly, for NP1, oxidative stress response does not correlate well with previously reported longevity benefits across strains [34].

Our previously published lifespan results showed that the antioxidant propyl gallate had a weak but positive effect on lifespan in C. elegans, and no effect on lifespan in C. briggsae [34]. Much like NP1, we found that in most cases propyl gallate did not lead to increased oxidative stress resistance (five of the six strains tested, Figure 2B). Only one strain, C. elegans N2, showed an overall increase in oxidative stress resistance, which was more robust in early mid-life (36% increase in median survival at early mid-life, p = 0.00128; 24% increase at late mid-life, p = 0.00323).

Finally, we tested the effect of resveratrol on oxidative stress resistance, another compound that we showed confers a species-specific lifespan response [34]. Unlike NP1 and propyl gallate, resveratrol had a widespread effect, and increased oxidative stress resistance to statistical significance in three of the six strains tested (Figure 2C). C. elegans N2 exhibited an increase in oxidative stress resistance at both ages (27% increase in median survival at early mid-life, p = 0.01061; 17% increase at late mid-life, p = 0.00688), while C. elegans JU775 exhibited increased survival in an age-dependent manner (no significant effect at early mid-life; 13% increase in median survival at late mid-life, p = 0.001988). C. elegans MY16, along with C. briggsae AF16 and HK104 trended towards a general increase in oxidative stress resistance, although not all effects were significant. The remaining strain, C. briggsae JU1348, showed no significant change in oxidative stress resistance with resveratrol treatment.

Overall, NP1 and propyl gallate, compounds that CITP previously observed to increase lifespan [34] have little to no effect on the oxidative stress resistance healthspan metric; resveratrol, however, exerted a positive effect on oxidative stress resistance in a species- specific manner.

We conclude that resistance to paraquat toxicity-associated oxidative stress does not correlate strongly with CITP longevity outcomes for select compounds NP1, propyl gallate or resveratrol.

Thermotolerance varies in a compound-, species-, strain-, and age-specific manner

Thermotolerance has been implicated as a reasonably close correlate of increased longevity [20, 47], with long-lived C. elegans mutants being relatively thermotolerant (physiological temperatures for wild-type N2 C. elegans range from 15–25°C; thermotolerance is typically described as increased survival after transient exposure to noxious temperatures (above 26°C to at least 36°C)) [48]. To determine the thermotolerance effects of NP1, propyl gallate and resveratrol, we aged animals at 20°C in the presence of the compound and then subsequently measured survival after shifting culture to 32°C on compound-free plates (see Supplementary Figure 2 for Kaplan-Meier survival curves).

We found that NP1 ameliorated age-related thermotolerance decline in two of the six Caenorhabditis strains tested (Figure 3A), C. elegans N2 and MY16 (no significant change at early mid-life; 42–53% increase in median survival at late mid-life; p = 0.00611- N2; p < 0.0001 – MY16). Conversely, we found that C. briggsae HK104 + NP1 exhibited accelerated loss of thermotolerance late in life (no effect at early mid-life; 28% decrease in median survival at late mid-life, p < 0.001).

Propyl gallate treatment conferred a moderate positive late life increase in thermotolerance in all three C. elegans strains tested (Figure 3B; no significant effect at early mid-life; 9–34% increase in median survival at late mid-life, p < 0.0001 – N2, p = 0.00138 – JU775, p = 0.00238 – MY16). Within this group, the effect size varied in a strain-specific manner. In contrast, thermotolerance in the C. briggsae strains tested was not affected by propyl gallate treatment.

Finally, resveratrol caused a significant increase in thermotolerance in both species and five of the six strains (Figure 3C). More precisely, we saw a compound-by-age interaction in four of the six strains tested. These strains exhibited an increase in ability to withstand thermal stress in an age-dependent manner (no effect at early mid-life; 12–51% increase in median survival at late mid-life, p < 0.0001 for all C. elegans strains, p < 0.05 for C. briggsae HK104). C. briggsae AF16 was positively affected overall by resveratrol treatment, with an increase in survival at both ages tested (34% increase in median survival at early mid-life, p = 0.00186; 17% increase at late mid-life, p = 0.03046), although the size of the effect decreased with age. The remaining strain, C. briggsae JU1348, showed no change in thermotolerance with resveratrol treatment.

Overall, we find the heat resistance associated with lifespan-extending compounds varies with compounds for tests of NP1, propyl gallate, and resveratrol in a genetically diverse test set. NP1 treatment conferred strain- and age-dependent thermotolerance, while propyl gallate elicited generally species- and age-specific responses. We found that resveratrol conferred the most robust and widespread thermotolerance, increasing heat stress survival in nearly every strain, with most strains exhibiting age-dependent effects, and the effect at the younger ages being non-significant but trending towards an increase. Notably, however, thermotolerance outcomes do not correlate with oxidative stress outcomes. We conclude that distinct measures of stress response can be differentially regulated by the compound interventions we tested.

The effect of compounds on locomotory health is largely species- specific

Previous work has shown that motility decreases in aging C. elegans [3, 16, 21, 49]. As locomotory ability is used in human clinical health assessment assays, we investigated whether lifespan-extending compounds would improve C. elegans motility, particularly later in life. To measure locomotion features, we used the CeleST platform [49, 50] to acquire eight measures of swimming ability at days 5, 9, and 12 of adulthood, respectively (the selected timepoints showed age-dependent differences in baseline measurements, i.e., are associated with detectable aging between the timepoints); see Supplementary Figure 3 for results with the eight measures of movement; swimming features measured by CeleST are improved in long-lived C. elegans mutants [49]. While the eight measures provide a broad range of information on the swimming of each of the strains, it was not clear which measures best capture the decline of locomotion with age across our genetic diversity panel. We therefore combined the information from all eight measures for each strain into a single multivariate composite measure using a linear discriminate analysis to weight and combine the individual measurements (Materials and Methods, Supplementary File, and Online Materials [51] provide an in-depth description of methodology). This approach accounts for interdependency among measures and the unique movement properties of the different strains. We used a strain-specific composite score in our analyses.

We found that NP1 had the most robust and widespread effect on swimming ability, improving locomotion in all six strains tested (Figure 4). In C. elegans, the effect of NP1 was age-dependent, with NP1 treatment slowing the rate of decline in swimming ability in all strains. We observed the largest effect in C. elegans N2 (17% increase in mean swimming score at day 12, p < 0.0001), while C. elegans JU775 and MY16 exhibited small but significant improvements (2–5% increase with age, p < 0.05 – JU775; 3% increase at adult day 9, p = 0.0085 – MY16). Interestingly, the effect of NP1 on swimming ability in C. briggsae was strikingly robust. JU1348 showed a reduction in age-related locomotory decline (12–16% increase in mean swimming score at adult days 9 and 12, p < 0.0001), while both AF16 and HK104 had an overall increase in swimming ability at all ages (9–14% increase in mean swimming score, p < 0.0001 for all ages with the exception of HK104 day 9, where p = 0.0028). Although NP1 conferred an overall increase in the locomotory ability of HK104, the effect was greatest in young animals, meaning that NP1 increased the relative rate of decline in strain HK104.

We found that propyl gallate treatment improved swimming ability in a species-specific manner, similar to the propyl gallate effect on lifespan (Figure 4). All three C. elegans strains showed an age-related improvement in swimming ability, with the effect size increasing at the older ages tested. Propyl gallate improved locomotory ability in N2 by 5% (p = 0.0105) at adult day 5, 7% (p = 0.0005) at day 9, and 21% (p < 0.0001) at day 12. This trend was also observed in both JU775 and MY16 to a lesser extent (4–6% maximum increase in mean composite swimming score during adult days 9 and 12; p ≤ 0.0003 – JU775, p < 0.05 – MY16), indicating that propyl gallate reduces the rate of age-related locomotory decline in C. elegans. In contrast, propyl gallate treatment had no effect on swimming ability in any C. briggsae strain tested.

We found the effect of resveratrol on locomotory ability was also species-specific (Figure 4). Specifically, resveratrol treatment improved swimming ability with age in C. elegans JU775 and N2 (5–11% increase in mean composite swimming score at day 9, 4–10% increase at day 12; p ≤ 0.0001 – N2, p ≤ 0.0015 – JU775), slowing the age-related decline in locomotion in both strains. This effect, however, is dependent on genetic background, as swimming ability in C. elegans MY16 was not affected by resveratrol. Likewise, resveratrol treatment had no effect on swimming in any C. briggsae strain.

Overall, we found that compounds that extend lifespan in particular strains also improve mobility when administered to those same strains. It is noteworthy, however, that we also found that NP1, which did not have a lifespan effect in the C. briggsae genetic background, could nonetheless enhance adult swimming ability. Thus, NP1 extends locomotory health more broadly than it enhances longevity in a genetically diverse test set.

Thermotolerance, but not oxidative stress or swimming ability, correlates with lifespan

Do health measurements correlate with lifespan, especially in response to a pharmacological intervention? Because our health assays were terminal (for throughput considerations and need to minimize manipulation stress, we did not recover animals used in the swimming assays for use in parallel lifespan studies), we do not have health and lifespan measurements for the same individuals. However, we were able to correlate each health measure score on a combined strain, compound, and age basis (Supplementary Figure 4) and found that thermotolerance was the most predictive of median lifespan, particularly at days 12 (C. elegans) and 16 (C. briggsae) of adulthood (R² = 0.58 at early mid-life, R² = 0.83 at late mid-life). In contrast, neither oxidative stress resistance nor swimming ability correlated well with median lifespan, with swimming ability on the days assayed being a particularly poor predictor of the effect of a compound on longevity (R² = 0.08, 0.097, 0.048, for days 5, 9, and 12 of adulthood, respectively). Our data suggest that despite experimental variability in thermotolerance outcomes, thermal resistance might be the best proxy for screening for longevity promoting drugs, an idea that remains to be tested more extensively.

Swimming ability and oxidative stress resistance, but not thermotolerance, are highly reproducible across labs

We observed some variability within our healthspan datasets and statistically evaluated the sources of this variation (Table 1). For oxidative stress resistance, ~34% of the total variance is attributed to genetic background, ~7.8% of total variance is attributed to lab-specific effects, and ~9.8% of the total variance is a result of variation within labs. This distribution of variation is similar to the variation we observed in our previously published lifespan datasets [34, 38]. The sources of variation within the swimming dataset are weighted towards individual variation, with ~79.2% of total variance attributable to differences in individual swimming performance. Variability at the genetic level only accounts for 5.7% of the variance, with nearly half of that (2.4%) attributable to species differences. Furthermore, among-lab differences only contributed to ~5.8% of the total variation, and within lab variance contributed to ~9.3% of the total variation. Both oxidative stress resistance and swimming are thus reproducible across labs.

We found thermotolerance outcomes to be variable in comparison to longevity, locomotion and oxidative stress measures. For thermotolerance, ~24.4% of the total variation is attributed to genetic background, while we find ~28.5% of variation is attributable to among lab differences, and ~15.8% to variation within labs. Variability may reflect difficult-to-control environmental factors such as localized/fluctuating differences in incubator temperature (or temperature distribution), or humidity variation. Regardless of root cause, however, data reveal that even against the backdrop of variability in thermotolerance, the thermotolerance measure may serve as a plausible tool for assessing the chances that a given intervention might exert longevity benefit.

Genetic background can profoundly affect the efficacy of a tested intervention to promote survival or health

Although many screens for longevity extension in C. elegans have been published [54–57], genetic background effects remain an unexplored variable in these screens as nematode studies have been largely confined to the canonical C. elegans N2 lab strain. The CITP seeks to address the issue of genetic background variability in compound screening by testing a panel of Caenorhabditis strains that represent evolutionary divergence similar to that found between mice and humans [35–37]. As we have previously published, genetic background is important when assessing the effects of pharmacological interventions on lifespan [34]. Here we selected three anti-aging compounds (NP1, propyl gallate, and resveratrol) for which we had previously conducted comprehensive studies of longevity [34] and we characterized their impact on two measures of physiological resilience, oxidative stress resistance (Figure 2) and thermotolerance (Figure 3), and on swimming ability (Figure 4), a primarily neuromuscular phenotype thought to reflect general health and vigor.

To determine if these three compounds induced lifespan extension by affecting a root cause in the aging process, we determined whether they conferred broad health benefits. We observed that if a compound extended lifespan in a particular strain, that compound typically slowed the rate of decline across the healthspan measures tested, with some exceptions (Figure 5). This raises an interesting question: if a compound modulates a core aging process, then why are the observed compound effects dependent on the genetic background? This may result from background differences in permeability, compound turnover, and/or pathway tuning that can vary with genetic background, or may suggest that the underlying cause(s) of aging varies across genetic backgrounds, or may indicate that the compounds are affecting symptoms of aging instead of causes of aging. This separation in phenotypes suggests that interventions that do not exhibit a benefit in one age-dependent measure may show benefits when assayed using an alternative measure. We therefore sought to determine if compound interventions promote health benefits in genetic backgrounds that do not exhibit lifespan extension. Interestingly, pharmacological interventions in the strains that saw no lifespan change after compound treatment had varying and sometimes robust effects on the rate of health decline, both positively and negatively. In this way, although health effects can vary, we observe that healthspan can generally be uncoupled from lifespan, and that anti-aging effects of compounds could be seen on health regardless of lifespan effects.

Thermotolerance as a proxy for longevity outcomes?

Given the relationship among different health measures and lifespan, we sought to determine which health measures are most reproducible and informative for targeted screens for anti-aging treatments. In the case of the three health measures that we evaluated, thermotolerance was the only one for which compound intervention effects correlated well with lifespan effects across genetic backgrounds (Supplementary Figure 4). Increased thermotolerance corresponds to increased longevity in many studies [20, 58–60]. The fact that lifespan correlates significantly with thermotolerance across compounds and genetic backgrounds suggests that the ability to maintain cellular homeostasis is a key factor in not just maintaining health, but also crucial in preventing death. Thermal stress causes unfolding of cellular proteins, and the induction of stress response pathways that induce organelle-specific molecular chaperone production [61, 62]. We therefore suggest the correlation between lifespan and thermotolerance as being likely attributed to compound administration promoting protein homeostasis, which when addressed could result in lifespan extension and increased thermotolerance.

Despite the relationship of thermotolerance to lifespan, thermotolerance may only be used sparingly in future CITP compound screening for three reasons: (1) given the correlation between thermotolerance and lifespan, there may be little additional benefit for using thermotolerance assays given that positive hits will likely be identified by lifespan studies alone, (2) although in principle thermotolerance provides a faster result, the approach used here actually ended up being more labor intensive than standard longevity assays (see Methods), and (3) most importantly, among our three assays, thermotolerance was the least reproducible across labs (Table 1).

Oxidative stress resistance and swimming locomotion reflect aging processes that are uncorrelated with lifespan

In contrast to thermotolerance, oxidative stress resistance, the other physiological resilience measure we used in these studies, does not correlate well with lifespan (Supplementary Figure 4). Although oxidative damage has been proposed as a cause for aging [63–65], this may not be surprising for two reasons. First, ROS signaling plays a crucial role in multiple cellular processes [45] and certainly the positive and negative aspects of ROS signaling play into the outcome equation [44]. Indeed, it is known that antioxidant treatment does not guarantee increased lifespan, indicating that oxidative stress may not be a primary driver of mortality [31–33]. Secondly, “oxidative stress” is an umbrella term that covers a wide range of phenomena, with sub-cellular localization of the oxidative stress and differing reactive oxide species resulting in differing levels of cellular stress or benefit. In this study we only tested a severe oxidative stressor, paraquat, and our results may have been different had we used a lower concentration of paraquat, a different oxidative stressor, or a more physiological deleterious ROS source.

Similar to oxidative stress, we found that swimming ability, a measure of neuromuscular function [66], does not correlate well with median lifespan across the different genetic backgrounds tested (Supplementary Figure 4). While oxidative stress resistance and swimming ability do not predict ultimate lifespan, these measures remain of particular interest when exploring the effect of pharmacological interventions on aging. Of note, the measures of oxidative stress and swimming prowess showed high reproducibility among labs (Table 1), and their separability from lifespan can facilitate the identification of interventions that treat important symptoms of aging that may not be found by screening lifespan alone.

Here we documented that representative pharmacological interventions in a genetically diverse Caenorhabditis test set result in a complex pattern of health effects, with genetic background and age both being important for overall outcome. Assay selection and design is clearly a factor in evaluation of intervention impact. Compounds can improve health measures in the absence of strong longevity outcome. Our results underscore that genetic background can be a crucial determinant in the evaluation of potential anti-aging, pro-resilience compounds. Although the findings presented here are limited to three compounds relevant to human interventions, our findings suggest that a personally tailored intervention that accounts for age, current health state, and genetic background may ultimately be necessary for optimal efficacy in the clinic.

Strains

Following standard CITP protocol, the following natural isolates were obtained from the Caenorhabditis Genetics Center (CGC) at University of Minnesota: C. elegans N2, MY16, JU775; C. briggsae AF16, HK104, JU1348. The N2 strain used was N2-PD1073, which is a clonal line derived from the N2 strain VC2010 used to generate a new N2 reference (VC2010-1.0) genome [68] that has been adopted by the CITP as a lab adapted control [38]. Animals were maintained at 20°C on 60 mm plates with NGM and E. coli OP50-1 and synchronized by timed egg-lays (for full SOP see ref. [67]). Animals were transferred to 35 mm NGM plates containing 51 μM FUdR and compound intervention (or the solvent DMSO in control plates) on the first, second and fifth day of adulthood, then once weekly when applicable, until healthspan measurements were initiated.

Interventions

The following compounds were selected to study impacts on health based on our previous findings for nematode longevity [34, 38]: NP1 (ChemBridge), propyl gallate (Sigma-Aldrich), and resveratrol (Cambridge Chemical). The same in-plate concentrations previously tested for lifespan effects from each compound were used here: 50, 200, and 100 μM, respectively. Animals were exposed to compound interventions only for the duration of adulthood up until health measurements were performed. Lifespan data for C. elegans and C. briggsae strains are from Lucanic et al. (2017) [34]; healthspan studies were conducted on identical strain stocks between 2017 and 2022.

Selection of ages for health assays

To measure health in aging adults we selected time points that (1) showed age-dependent differences in the baseline measurements (e.g., detectable aging between the timepoints), and (2) were justifiable based on the known physiological and demographic changes in normally aging adults. For example, C. elegans and C. briggsae hermaphrodites are self-fertile, with a normal reproductive period in the absence of males lasting for the first five to eight days of adulthood [69]. We therefore used the end of the period of self-reproduction to establish an “early-mid-life” timepoint for health assays. We also selected a “late-mid-life” timepoint to correspond to approximately the 95th survivorship centile [70] to minimize selection biases (see Supplementary Figure 5).

Heat stress

To measure the impact of each compound on organismal tolerance of heat stress, we implemented the following augmented Lifespan Machine (LM) protocol [38, 71]: 70 animals each placed on 50 mm tight-lidded petri plates with modified NGM and E. coli OP50-1 at 32°C humidity for a duration of four days. For each of two biological replicates (independent trials), we created two technical replicates (single plate) per strain and condition (age and compound or control) per lab. C. elegans were tested at adult days 6 and 12 while C. briggsae were tested at adult days 8 and 16. A full protocol is available online [67].

Oxidative stress

To measure the impact of each compound on organismal resilience to exogenous oxidative stress, we implemented the following ALM protocol [38, 71]: 70 animals were placed onto each 50 mm tight-lidded petri plate with modified NGM containing 40 mM paraquat (or methyl viologen dichloride, from Sigma-Aldrich), 51 μM FUdR, and E. coli OP50-1. The duration of the paraquat-exposure assays was dependent on the starting age of the animals due to the increased rate of mortality with age. Specifically, C. elegans on day 6 of adulthood and C. briggsae on day 8 of adulthood were assayed for 16 days, while C. elegans at 12 days of adulthood and C. briggsae at 16 days of adulthood were assayed for only 7 days. For each of two biological replicates, we created two technical replicates per strain and condition (age and compound or control) per lab. A full protocol is available online [67].

Heat and oxidative stress statistical analysis

Statistical analyses for survival were conducted as previously published [34, 38]. In brief, a mixed-model approach within each strain was used in which compound and age were treated as fixed effects, and laboratory site, experiment date, lifespan machine, and technical replicate were considered random effects. Effects of compounds on healthspan were tested via the compound-by-age interaction, with the compound or control specific change with age being defined as the “rate of decline” of that healthspan measure. We analyzed survival using both a general linear model with the lme4 (v1.1.23) [72] package and a mixed model Cox proportional hazard (CPH) model with the coxme (v2.2.16) package [73]. Each compound was tested as a planned comparison against its appropriate control using the multcomp (v1.4.13) package [74]. Analyses were conducted in the R statistical language. All relevant data, R-scripts, and output files are available online [75].

C. elegans swim test

To measure the impact of lifespan-enhancing compounds on the age-associated decline in agility or neuromuscular function, we implemented the C. elegans Swim Test (CeleST [49]). Briefly, animals were exposed to compound intervention during adulthood as described above until CeleST measurements were collected at adult ages day 5, 9, and 12. For two biological replicates at each of the three CITP sites, 40 animals were tested per condition (age and compound or control) per strain. For full experimental protocols see our online protocol [67]. The CeleST software was used to measure eight different parameters (Wave initiation rate, Body wave number, Asymmetry, Stretch, Curling, Travel speed, Brush stroke, and Activity index) [49, 50]. To facilitate comparisons between strains and compound treatments we generated a single composite swimming score.

CeleST composite score

CeleST provides video-based analysis of eight separate features of locomotion, from bending rate to travel speed⁵⁰. However, we do not necessarily have an a priori expectation as to how each variable might change with age, or how strains may differ in age-dependent changes among the eight measures. To maximize the differences between ages across all the measurements, we used linear discriminant analysis (LDA), with the eight original measurements in each record serving as the predictor variables and age-dependent decline being the primary discriminator (see Supplementary File) to reduce the number of dimensions from eight to one. Projecting the eight predictor variables onto a single axis creates the first linear discriminant function. This first linear discriminant function minimizes within-age variance, maximizes between-age variance, and maximizes the separability of the means of the ages [76]. While it cannot capture all the information provided by each original measurement, it captures as much as possible. Because it maximizes the differences between group levels (ages in our case), linear discriminant analysis is often used to predict group membership after a training dataset is used. To avoid confounding strain specific differences in swimming, our LDA was performed on the untreated control data independently for each strain to generate first linear discriminant functions for each strain. This function provides the composite movement score that best captures the decline in movement with age. We then used the coefficients for each of the original eight predictor variables in the strain specific first linear discriminant functions as strain specific weightings of the eight measures to generate composite scores that maximized the ability to separate the animals in the control set by age. Those strain specific weightings were then used to generate a composite score of treated animals to analyze for compound and age effects (see Supplementary File for in depth methodology). While the main text presents an analysis of swimming using the composite score, analyses of the individual CeleST measurements are available in the Supplementary Materials (Supplementary Figure 3). All relevant R-scripts and raw output are available online [51, 67]. To analyze changes in swimming behavior with age, the composite score was used as the dependent in mixed effects general linear models built for each strain in R using the lme4 package (v1.1.27.1) [72], with compound and age as fixed effects and laboratory site, research technician, experiment date, and video as nested random effects. Significant age by compound interactions, as determined using the R car package (v3.0.11) [77], were used to determine the effect of the compound on the rate of decline of swimming with age.