Variation in the response to exercise stimulation in Drosophila: marathon runner versus sprinter genotypes

How animals behave under diverse situations can have strong implications for survival and reproductive success (Gaynor et al., 2019). For example, when detecting a predator, an animal might keep still in the hope of remaining unnoticed, or it might try to flee and risk attracting the predator's attention. Similarly, animals make choices about when and where to forage for food, whether to approach a potential mate or competitor, and how to react to environmental perturbations. The specific timing of when to forage impacts the probability of encountering predators or competitors targeting the same food source (Mukherjee and Heithaus, 2013). Adverse environmental conditions can lead to migratory behavior or the entry into hibernation or diapause (Fielenbach and Antebi, 2008; Richards, 2010). In addition, unpredictable environmental conditions can have profound impacts on behavior in animals – including humans – leading to stress and changes in cognitive function and physical wellbeing (Cameron and Schoenfeld, 2018; Hollis et al., 2013; Hurtubise and Howland, 2017; Peay et al., 2020). These examples illustrate the important consequences of animal behavior patterns throughout life.

Given the importance of animal behavior, it is not surprising that many aspects of animal behavior are regulated tightly. Genetics plays a large role in animal behavior and how animals respond to specific situations (Opachaloemphan et al., 2018; Shimaji et al., 2019; York, 2018). For example, genes have been identified that control courtship behavior. In Drosophila melanogaster, the gene fru (fruitless) impacts courtship behavior, and loss of function mutations of fru lead to male flies perceiving other males as potential mates, rather than females (Yamamoto and Koganezawa, 2013). Genes also impact circadian activity patterns, as well as sleep–wake cycles, and allelic variation in several clock genes can result in shifts in circadian patterns (Dubowy and Sehgal, 2017). In D. melanogaster, strains differing in the midday quiet phase (‘siesta’) are well documented. Variability between strains in activity patterns has been linked to the clock gene per (Cao and Edery, 2017; Yang and Edery, 2018), and clock gene expression has also been linked to activity differences between D. melanogaster and Drosophila suzukii (Plantamp et al., 2019). Looking specifically at circadian rhythm in D. melanogaster, Harbison et al. (2019) identified 268 genes linked to circadian patterns and demonstrated that there was significant variation in circadian behavior patterns in their naturally derived study population. Similarly, in an earlier study of the same population, several-hundred genes were identified as candidate loci impacting various parameters associated with sleep behavior (Harbison et al., 2013). These studies illustrate that many aspects of animal behavior are at least to some degree under genetic control.

Another behavior where the influence of genetics has been examined in detail is physical activity. Data from several animal systems suggest that the timing and duration of animal activity are controlled by genes (Aaltonen et al., 2010; Lightfoot, 2013; Sarzynski et al., 2016). Both mice and worms show differences in activity patterns between strains (Greene et al., 2016; Houle-Leroy et al., 2003; Mowrey et al., 2014; Stegeman et al., 2019). Mice strains selected for high or low levels of wheel running activity have been produced, and they maintain these differences throughout life (Marck et al., 2017). Studies with mice selected for a high level of wheel running demonstrate that the tendency to run (or exercise) and the physiological response to the activity are influenced by genetics (Kelly et al., 2017; Kolb et al., 2013). In Drosophila, previous work from my laboratory has identified 314 genetic variants impinging on the basal activity levels of the animals and an additional 80 variants that control how much the animals move with rotational stimulation (Watanabe et al., 2020). Other laboratories have studied how flies respond to a ‘startle’ trigger, a brief, sudden environmental perturbation, and again, this response is impacted by many different genes. Mackay et al. (2012) link 94 genetic variants to the startle response in the Drosophila melanogaster Genetics Reference Panel (DGRP). Thus, the available data suggest that many aspects of animal activity and locomotor behavior are under genetic control as well.

In this study, I re-examined the activity data collected by my laboratory from the DGRP strains in response to rotational stimulation to look closer at the activity patterns exhibited by the animals (Watanabe et al., 2020). While my laboratory's previous analysis focused on the total amount of activity shown by animals in response to exercise stimulation, here, I focused on the timing of this activity throughout the 2 h exercise period, as a previous study has shown that some animals stop moving after a short period of time of rotational stimulation, while others continue to move or ‘exercise’ (Mendez et al., 2016; Watanabe and Riddle, 2017). Specifically, I examined how the animals respond to rotational stimulation to determine whether there are ‘marathon runner’ strains that remain highly active throughout the 2 h exercise period and ‘sprinter’ strains that only respond to the rotation with a short burst of activity before stopping. Focusing on the first 30 min of the 2 h exercise period, there was extensive variation in the DGRP strains in how much activity the animals performed in these 30 min, ranging from 0.1% to 85.3%. This variation was impacted strongly by sex and genotype, and a genome-wide association study (GWAS) identified approximately 60 genes linked to the amount of activity performed in the first 30 min of the exercise period. Among these genes were 10 high-confidence candidate genes confirmed in a validation study. Thus, the tendency of fly strains to respond to rotational stimulation with a ‘marathoner’ or ‘sprinter’ pattern of activity is impacted by genetic variation similar to many other aspects of activity including locomotor activity.

The data used for the analyses presented in this paper were described initially in Watanabe et al. (2020). Briefly, 3–7 day old virgin animals from the DGRP strain collection were exercised in groups of 10 animals for 2 h using the rotational exercise quantification system (REQS). The rotation speed of the REQS was set to 4 rpm, and the experiment was carried out in an environmental chamber (25°C, 70% humidity, 12 h:12 h light:dark cycle) at 12:00 h (5 h after lights on). The REQS recorded beam crossings in 5 min intervals. For each DGRP strain, 10 replicate sets of 10 flies per sex were assayed if possible. To determine how the strains in the DGRP collection differed in their exercise patterns, the data from Watanabe et al. (2020) were recoded. For each sample, the percent of exercise performed during the first 30 min of the 2 h exercise period was calculated. The raw data can be found in Table S1.

Basic statistical analyses were carried out in R (http://www.R-project.org/), using the VCA package for quantitative genetics analyses. The variance components were estimated using the restricted maximum likelihood (REML) approach for mixed models. The model included line (L, random), sex (fixed) and their interaction (fixed). Genetic (σ_G²) and environmental (σ_E²) variance compose the phenotypic variance (σ_P²=σ_G²+σ_E²) with the genetic variance including both the variance due to line and the sex by line interaction (σ_G²=σ_L²+σ_L×S²) and the environmental variance defined as the within-line variance. Broad sense heritability (H²) was H²=σ_G²/σ_P². Coefficients of genetic and environmental variance are CV_G=100σ_G/mean and CV_E=100σ_E/mean. The cross-sex genetic correlation was calculated as r_MF=σ_L²/(σ_L²+σ_L×S²), and genetic correlation was r_g=σ_L²/√[σ_L²(female data)×σ_L²(male data)].

The GWAS was carried out with the DGRP webtool (http://dgrp2.gnets.ncsu.edu/) (Huang et al., 2014; Mackay et al., 2012), using line means for the fraction of activity performed in the first quarter of the exercise period as input (Table S1). The webtool provides the results from two different analyses: a mixed model analysis and a regression analysis. These two models are applied to the combined sex data, the data from each sex, and the difference between the sexes to identify genetic variants linked to the trait. Genetic variants with a nominal P-value of 10⁻⁵ or less in any of the analyses were considered candidate loci (Table S2). This nominal P-value cutoff was chosen based on a review of other studies using the DGRP strain collection, many of which use a P-value of 10⁻⁵ cutoff to select genes for further consideration (Arya et al., 2015; Campbell et al., 2019; Dembeck et al., 2015; Jordan et al., 2012; Mackay et al., 2012; Swarup et al., 2013; Weber et al., 2012). Manhattan plots were produced using the R package ‘qqman’. Genetic variants were classified based on the annotation available from the DGRP website, using custom Perl scripts and R (http://www.R-project.org/).

Gene size of candidate genes was evaluated using data from FlyBase (dmel-all-r6.25.gtf; www.flybase.org) (Thurmond et al., 2019) and a permutation test implemented in R (R Development Core Team, 2018). The gene size of the candidate genes was compared to the genes identified by Schnorrer et al. (2010) as involved in muscle development, again using a permutation test implemented in R (http://www.R-project.org/).

To validate the GWAS results, two additional GWAS were carried out. The initial GWAS described above is based on 10 measurements for each sex/genotype combination derived from 10 vials of 10 flies each. To determine how repeatable the GWAS results were, we split the data into two independent datasets, one with phenotypic means based on vials 1–5 and one with phenotypic means based on vials 6–10. As not all sex/genotype combinations had 10 vials, the dataset based on vials 6–10 was missing some DGRP lines present in the dataset based on vials 1–5, leading to a lower power to detect significant variants. The two new datasets were processed using the DGRP2 webtool (http://dgrp2.gnets.ncsu.edu/) (Huang et al., 2014; Mackay et al., 2012) as described above, and the results were compared with each other as well as with the results from the GWAS using the complete dataset.

The genetic variants with a nominal P-value of 10⁻⁵ or less were used for this analysis, with variants not associated with genes and duplicates removed. Several tools were used for this analysis including the Database for Annotation, Visualization and Integrated Discovery (DAVID, v6.8; Huang da et al., 2009a,b), Webgestalt (WEB-based Gene SeT AnaLysis Toolkit 2019; Liao et al., 2019), FlyEnrichr (Kuleshov et al., 2016) and PANTHER (Protein ANalysis THrough Evolutionary Relationships, v14.1; Mi et al., 2019). Multiple testing corrections (false discovery rate, FDR) were used for all analyses.

I was interested in determining whether there is variation in the response to exercise stimulation among different Drosophila genotypes. During rotational exercise stimulation, flies walk or hop along the walls of their enclosure trying to reach the top of the enclosure, but many of them stop walking at some point during the 2 h exercise period. Thus, I wanted to know whether there was a Drosophila equivalent of ‘sprinters’ that would preferably exercise for a short time period after the initiation of exercise stimulation, and ‘marathoners’ that would remain active throughout most of the exercise period. To achieve this goal, previously published data by my laboratory (Watanabe et al., 2020) reporting on the activity observed in the DGRP strain collection in response to a 2 h rotational exercise stimulation using the REQS (Watanabe and Riddle, 2017) was re-analyzed. Specifically, I calculated the percent of activity performed in the first quarter of the exercise period as a measure of the tendency to behave like a sprinter versus a marathoner. In the DGRP population, the percent of activity in the first quarter of the exercise period varied widely, ranging from 0.1% to 85.3% when the data from the two sexes was combined (Fig. 1). Similarly, this range was 0.1–88.6% for males and 0.0–93.1% for females. On average, animals typically performed 42.4% of the activity during the first quarter of the exercise period (40.7% for males, 43.9% for females). If animals exercised continuously during the 2 h exercise period, one would expect them to perform 25% of the total activity in the first 30 min (30 min/120 min). For 137 out of the 156 DRGP strains examined, the activity performed during the first quarter of the exercise period differed significantly from 25% (t-test or Wilcoxon test), suggesting that continuous activity is rare. Together, these data show that Drosophila strains differ significantly in how they respond to exercise stimulation, with some animals active mostly during the initial phase of the exercise period and others remaining active throughout.

As this dataset included animals with very low activity levels as measured by beam-crossing in the REQS (Watanabe and Riddle, 2017), I investigated the relationship between total activity and how much activity was performed during the first quarter of the exercise period to ensure that the findings were not driven mainly by these low-activity samples. Overall, there was a negative correlation between the total amount of activity performed and the fraction of activity performed in the first quarter of the exercise period (−0.42, P=5.005e−08; significant in both males and females; Fig. S1). Thus, animals that have a high total amount of activity tend to perform a lower fraction of activity early, while animals with lower total activity levels tend to perform a large fraction of activity early. Not surprisingly, animals with the highest activity levels tend to perform 25–30% of their activity in the first quarter of the exercise period. This finding is robust, even if samples with a total activity level of less than 100 beam-crossings were removed (correlation: −0.43, P=2.287e−08). A negative correlation was also observed between the fraction of activity performed in the first quarter of the exercise period and the baseline activity levels of animals in the DGRP, as reported by my laboratory previously (Watanabe et al., 2020). This correlation is lower (−0.126 in males and −0.246 in females) than the correlation with total induced activity, and was only significant for females (P=0.007588; Fig. S1). These analyses suggest that animals with high overall activity levels typically are active for longer periods of time, and animals with low overall activity levels tend towards shorter bouts of activity immediately following stimulation but then cease to move.

Next, the relationship between the activity pattern exhibited by the animals and lifespan was analyzed. Typically, it is thought that higher levels of activity result in better health and longer lifespan (Erickson et al., 2019; Kujala, 2018; Partridge et al., 2018). However, there are limits to this relationship, and sometimes high activity levels are linked to oxidative damage and possibly costs to the individual (Margotta et al., 2018; Williams, 2018). When my laboratory (Watanabe et al., 2020) previously investigated the relationship between basal and induced animal activity and lifespan, we found only a small, negative correlation (r=−0.199) between exercise-induced activity and the lifespan of DGRP strains reported by Ivanov et al. (2015). Given that my analysis is based on a re-analysis of the exercise-induced activity levels from Watanabe et al. (2020), it was surprising to see a positive correlation between lifespan data from Ivanov et al. (2015) and the amount of activity performed early in the exercise period examined here (Fig. S2; r=0.191, P=0.03111). Using a second lifespan dataset, the correlation between lifespan and the amount of activity performed early in the exercise period was similar, but not statistically significant (r=0.158, P=0.07954) (Durham et al., 2014). These analyses imply that animals that respond to the rotational stimulation with a short spurt of activity and then settle down might have a small lifespan advantage compared with animals that continue to respond to the rotational stimulation by increased activity levels.

The above result suggested that it might also be of interest to look at the startle response of the DGRP strains. In Drosophila, startle response refers to the behavioral response to a brief perturbation such as tapping the vial the flies are housed in (Jordan et al., 2006). This phenotype has been studied previously by the Mackay lab in the DGRP strains (Mackay et al., 2012). Given the fact that some of the animals in the activity data set analyzed here responded only briefly to the rotational stimulus, it might be hypothesized that these animals might also have a lower startle response. Looking at the correlation between startle response and the fraction of activity performed early in the exercise period, there was a small negative correlation (r=−0.153, Pearson's product-moment correlation), but it was not statistically significant (P=0.08). In addition, the startle response did not show a correlation with lifespan in either the Ivanov et al. (2015) or Durham et al. (2014) datasets (r=0.101, P=0.16; r=0.038, P=0.62). These findings indicate that a brief response to an external stimulus (tapping in the startle response, rotation in the exercise study) is not generally associated with longer lifespan.

Next, I examined the relationship between exercise activity patterns in males and females of the same DGRP strain. In the dataset presented here, the amount of activity performed in the first 30 min of the 2 h exercise period in males and females was correlated strongly, with a correlation of 0.73 (P<2.2e−16; Fig. S2C) and a cross-sex genetic correlation of 0.785 (for additional quantitative genetics analyses, see Table 1). This level of correlation between the sexes is similar to the correlation reported previously for total exercise activity levels (r=0.83) (Watanabe et al., 2020), suggesting that activity patterns and the amount of exercise performed are well matched in males and females. However, sex still significantly influences how much activity the animals perform during the first 30 min of the exercise period (P=1.027e−08, Kruskal–Wallis rank sum test), as does the overall genotype of the animals (‘strain effect’, P<2.2e−16, Kruskal–Wallis rank sum test). Generally speaking, females tend to perform a higher fraction of the total activity early in the exercise period than males, but this is dependent on the genotype or strain. Thus, while the amount of exercise performed early in the exercise period is strongly influenced by sex and genotype, there is also a strong correlation between the sexes of the same DGRP strain.

While the analysis presented in the previous section suggests that overall genotype strongly impacts activity timing during exercise stimulation, to ensure that a GWAS is appropriate, heritability for the fraction of activity performed during the first quarter of the exercise period was calculated. The broad sense heritability is 0.596, suggesting that there is a sufficiently high fraction of the phenotype under genetic control to warrant a GWAS. The GWAS was carried out using the DGRP GWAS webtool (http://dgrp2.gnets.ncsu.edu/), which implements two different statistical models, a mixed model and a regression model (Huang et al., 2014; Mackay et al., 2012). Four different GWAS were carried out with both of these models: one on the combined data from the two sexes, one for male data, one for female data, and one for the phenotypic difference between the sexes. Using a nominal P<10⁻⁵ cutoff, there were 99 distinct genetic variants contributing to how the animals respond to rotational exercise stimulation, in the male (43), female (19) or combined datasets (19), or contributing to the difference between the two sexes (33; Table S2). The Manhattan plots in Fig. 2 illustrate the genome-wide distribution of the single nucleotide polymorphisms (SNPs) linked to the amount of exercise performed early for females (Fig. 2A) and males (Fig. 2B). Overall, the results from the two analysis models, regression versus mixed model, were very consistent, with the mixed model generally detecting ∼90% of the variants detected by the regression model as well as 10–20% additional variants. There was only one genetic variant shared between the results from the male and female data (a SNP in bab1), and the combined analysis shared more variants with the results from males (12) than with those from females (2). The variants identified as responsible for the sex differences in this activity phenotype were very distinct and only shared one variant with the male data (an intergenic variant) and none with the combined or female data. These findings suggest that genetic variants impact activity patterns and that different loci are important in males and females.

Overall, these 99 variants are associated with a total of 58 genes in the Drosophila genome. While there are several regions of high linkage disequilibrium (LD) among the 99 genetic variants identified, most of the SNPs in LD with each other correspond to the same gene (there were four gene pairs with weaker LD scores of <0.8; Fig. S3A), suggesting that the 58 genes identified likely represent independent effects on the activity behavior phenotype. The candidate variants are found in genic as well as intergenic regions (Fig. 3); 59% of the candidate variants are in introns and 14% are in intergenic regions (‘unknown’ relationship to genes). These values are similar to the genome-wide distribution of SNPs, where 50% are in introns and 21% of variants are found in intergenic regions (Fig. 3, right), and the two distributions were not statistically different from each other (chi-square test, P=0.1543). Thus, in terms of their functional classification, the candidate variants detected in the GWAS reflect the overall distribution of variants included in the analysis.

To further characterize the candidate gene variants identified in the GWAS, I carried out functional enrichment and GO term analyses with a variety of tools. The basic GO term analysis tools DAVID (Huang da et al., 2009a,b) and PANTHER (Mi et al., 2019) did not identify any significantly enriched categories after correction for multiple testing. The WebGestalt functional enrichment tool (Liao et al., 2019) identified the terms ‘tissue migration’ and ‘epithelial cell differentiation’ as significant enriched in the candidate gene set (P=0.033 and P=0.045 after FDR multiple testing correction). In addition to traditional GO term analysis, FlyEnrichr (Kuleshov et al., 2016) also carries out a variety of functional enrichment analyses by mining data from FlyBase (Thurmond et al., 2019) and PubMed. Analyzing the loss-of-function phenotypes reported in FlyBase for the candidate gene set, FlyEnrichr identifies ‘midgut constriction’ as well as a number of muscle-related terms as significantly enriched (Fig. 4A). Muscle-related terms also appear in other analyses, such as ‘embryonic visceral muscle’ in the allele phenotype analysis, and terms such as ‘flight-defective’, ‘paralytic’, ‘hypoactive’ in the analyses mining phenotype data from PubMed (Fig. 4B) and incorporating expression data in addition to the phenotype data (Fig. 4C). Together, these analyses suggest that the GWAS analysis identified genes associated with muscle function and with reported mutant phenotypes linked to mobility and animal activity.

While genes linked to the central nervous system (CNS) and neuronal function do not constitute a significantly enriched group in the basic GO term analysis, there are genes with documented functions in the CNS among the candidate genes. For example, almost half of the genes identified (26 of 58) have documented links to the CNS reported in FlyBase (Thurmond et al., 2019), including highest expression levels in the CNS or reported defects linked to the CNS. For example, mib2 (mind bomb 2) encodes a protein required for regulating synaptic plasticity and memory formation, and Dmtn (Dementin) is required for normal brain development (Hopkins, 2013; Olesnicky et al., 2014). Also, some of the additional functional enrichment analyses involving mining of data from PubMed and FlyBase by FlyEnrichr (Kuleshov et al., 2016) suggest links to neuronal functions. These terms include, for example, ‘channel activity’ and ‘neuron differentiation’. Together, the reported functions of the candidate gene set suggests that the CNS and the muscles cooperate and coordinate to impact exercise patterns.

When investigating the candidate genes, a preponderance of large genes was noted. The candidate genes ranged in size from 1013 bp to 179,438 bp with a median of 16,152 bp and a mean of 29,116 bp. The results from a permutation test indicated that the candidate gene set includes genes significantly larger than the total gene set (all genes in the D. melanogaster genome, r6.25), which has a median gene size of 1691 bp and a mean of 5746 bp (P<0.003; Fig. S3B). Thus, the genes in the candidate gene set are significantly larger than genes typically found in the D. melanogaster genome.

To evaluate the importance of this finding, two approaches were employed. First, I carried out GO term and functional enrichment analyses for the top 1% of largest genes in the D. melanogaster genome (r6.25), focusing on the WebGestalt and FlyEnrichr analysis tools, as these tools identified significantly enriched terms for the candidate gene set (see above). In this analysis, WebGestalt identified 10 terms as significantly enriched, but they did not include ‘tissue migration’ or ‘epithelial cell differentiation’, which were detected in the candidate gene set. FlyEnrichr did not detect muscle associated terms as enriched in this set of large Drosophila genes (allele loss-of-function phenotype from FlyBase). The term ‘behavior defective’ appeared in both the candidate gene analysis above and this analysis of large genes, but ‘flight defective’ only appeared in the candidate gene analysis [‘phenotype – information from PubMed (AutoRIF)’]. However, the in the ‘phenotype – AutoRIF predicted z-score’ category, three of the four terms reported as significant in the candidate gene analysis (Fig. 4C) also appeared in this analysis of large genes in the Drosophila genome. Thus, at least some of the terms identified in the GO term analysis of the candidate genes, specifically those in Fig. 4C, might be due to the large size of genes of the candidate genes identified here. However, the majority of muscle-associated terms appear to be specific to the candidate gene set.

As a second approach to validating the functional enrichment analysis of the candidate genes and to evaluate the importance of the large size of these genes, I examined genes involved in muscle development, to determine whether these genes might be abnormally large. Neuronal genes tend to be larger than average (Gabel et al., 2015). Genes involved in muscle development were identified by Schnorrer et al. (2010) in a genome-wide RNAi screen. The average size of these genes was 7674 bp, which is larger than the largest 0.1% of gene sets of identical size pulled at random from the genome (P<0.001, permutation test), suggesting that genes important for muscle development, like genes important for neurons (Gabel et al., 2015), tend to be larger than the genome average. Why muscle genes might be larger than the genome average is unclear. However, the large size of the candidate gene set identified here might reflect this feature of genes functioning in muscle.

To further evaluate the candidate genes controlling exercise behavior, I took advantage of the fact that phenotypic values in this study were based on 10 biological replicates per genotype/sex, i.e. 10 vials with 10 flies each per sex/genotype combination. Thus, it was possible to split the sample and calculate phenotypic means based on vials 1–5 and on vials 6–10, producing two non-overlapping datasets. The dataset based on vials 6–10 was somewhat smaller, as for some of the genotypes fewer than 10 replicate data points were available. When running additional GWAS using the DGRP2 webtool, 55 genes were identified from dataset A (vials 1–5) and 57 genes from dataset B (vials 6–10) (Table S2). Ten of these genes were shared between the two datasets, indicating that these genes represent the highest confidence candidate genes (see Fig. 5). These genes include bric a brac 1 (bab1), a transcriptional regulator important for pattern formation during development (Godt et al., 1993). belphegor (bor) encodes an ATPase with functions in neurons (Gilquin et al., 2010; Harel et al., 2016). Supervillin (Svil; CG13800) encodes an actin-binding protein, mutations in which cause problems with muscles and flightlessness (Schnorrer et al., 2010). Zasp67 (CG14168) functions in adult muscle development and mutations cause problems with muscles and flightlessness (González-Morales et al., 2019). Gαo (G-oalpha47A) is important for heart formation, and muscle and motor neuron phenotypes have been reported in mutants (Frémion et al., 1999; Patel et al., 2016). The fact that this group of genes appears in all three analyses suggests that they are the most promising for follow-up studies.

In this study of activity in response to rotational exercise stimulation, I have discovered that there are significant differences in the behavior patterns of different Drosophila strains. In the DGRP population, many strains perform more activity early in the exercise period. There are clearly genotypes that behave as ‘sprinters’, with most of their activity early in the exercise period, as well as ‘marathoner’ genotypes that maintain high activity throughout the exercise period. These behavioral differences are impacted by sex and genotype, and a set of 10 high-confidence candidate genes was identified that impact these behavioral patterns. The candidate genes include several genes linked to muscle and neuronal development, as well as the transcriptional regulator bab1. Thus, activity patterns in response to rotational stimulation seem to be controlled at least in part genetically.

These findings are in agreement with studies in other model organisms that suggest that activity patterns are influenced by genetics (Aaltonen et al., 2010; Lightfoot, 2013; Sarzynski et al., 2016). For example, genetically distinct rodent lines have been bred that are naturally high or low in voluntary wheel running (Heese et al., 2019; Hiramatsu and Garland, 2018; Kelly et al., 2014). In studies using forced swim tests, where rodents are placed in an inescapable water tank to determine how long they will try to escape, animals of different genotypes differ in their response behavior, with some genotypes attempting to escape for much longer than others (Can et al., 2012; Molendijk and de Kloet, 2019). In addition, studies with mice selected for high levels of voluntary wheel running demonstrate that baseline physiological measures as well as the response to physical activity show strain-specific differences, with high-activity strains showing an improved training response (Kelly et al., 2017). There are also studies of human athletic performance that suggest that certain genotypes are overrepresented among elite athletes (e.g. specific alleles of ACTN3 and ACE in elite sprinters; Papadimitriou et al., 2016) and that the overrepresented genotypes might be different between sprinters versus endurance athletes (Eynon et al., 2013). Thus, this study as well as others suggest that further genetic studies in both humans and model organisms will be essential to understand what determines animals' activity patterns and their physiological response to this activity.

However, the relationship between genetics and activity, or athletic performance in humans, is not straight forward. When the performance of two genetically distinct strains of Drosophila on different diets was compared, the optimal diet differed between strains, and total caloric intake was a poor predictor of performance (Bazzell et al., 2013). In humans, studies suggest that while there are links between certain genetic variants and athletic performance, different genetic variants impact natural ability, the ability to perform a certain task without practicing and trainability, the ability to show increased performance in a task after training (Bouchard, 2019; Bouchard et al., 2011). In addition, natural ability and/or trainability are irrelevant in athletic performance if the individual is not motivated to train or does not have access to various interventions and training methods (Joyner, 2019; Sarzynski et al., 2016). This study suggests that animals such as Drosophila can be used to model diverse responses to exercise interventions, as all animals in this study received the same exercise stimulation, but the amount of exercise performed and the exercise behavior, i.e. sprinter versus marathoner, varied between the genotypes. This variability is reminiscent to what is seen in humans, where there are individuals that prefer short, high-intensity exercise and others that prefer longer, more moderate exercise periods. Thus, there are opportunities to exploit model systems like Drosophila with innovative study design to address questions related to exercise behavior and response.

Given the complex effects of physical activity on physiology, it might not be surprising that the link between physical activity and lifespan is not straight forward. Often, it is reported that increased physical activity promotes longevity (Erickson et al., 2019; Kujala, 2018; Partridge et al., 2018). However, we did not find such a correlation in our previous study on fly activity in general (Watanabe et al., 2020), nor did I see a strong positive effect of activity levels on lifespan in this study. Rather, it appears that animals responding to the rotational exercise stimulation with a short burst of activity might have a lifespan advantage. It is possible that the animals that respond to exercise stimulation with a short burst of activity and then settle down are less stressed by the novel experience than the animals that continue to be active, and thus the lifespan advantage might be due to a general stress response difference. Therefore, getting at the underlying cause for correlations between activity and longevity is challenging. Other studies also have suggested that the link between activity levels and longevity is complex, proposing the existence of a ‘sweet spot’ that maximizes benefits while minimizing harm (Arem et al., 2015; O'Keefe et al., 2018). In addition, some studies suggest that endurance, or even the maximum rate of oxygen consumption V̇_O2,max, might be a better predictor of lifespan (Clarke et al., 2015; Koch et al., 2011; Leeper et al., 2013). This complex set of findings indicates that additional, possibly long-term studies are needed to determine which of the many physiological effects of exercise mediate the positive impact of activity on lifespan, and how to promote those specific effects.

Finally, the results from the study presented here also suggest that Drosophila might be a good model to investigate exercise behavior and the link to diet, the gut microbiome and possibly disease. The impact of diet and gut microbiome is considered important for exercise in humans (Hughes, 2019). For example, the gut bacterium Akkermansia muciniphila has been reported to impact obesity (Everard et al., 2013). This effect appears to be mediated by a membrane protein, administration of which improved metabolism (Plovier et al., 2017) and increased physical activity and energy use (Depommier et al., 2020). Another example of the impact of the microbiome and pathogens on behavior is illustrated by Toxoplasma gondii, for which infections can lead to more impulsive behavior and increased activity (Martinez et al., 2018). Drosophila populations have a range of activity levels and different types of responses to the exercise stimulations such as the marathoner/sprinter behaviors described here. In addition, similar to the impact described for humans, the composition of the gut microbiome alters activity-related behaviors in Drosophila. For example, a 2018 study with germ-free Drosophila found an increase in walking speed and overall activity in the animals lacking a microbiome compared with the control animals (Schretter et al., 2018). Thus, using Drosophila resources such as the DGRP strains in conjunction with exercise systems such as the REQS provides an opportunity to examine the impact of the microbiome and diet on exercise activity and responses.

I would like to thank Dr M. Bamman (UAB) for the idea that inspired this analysis. I also would like to thank the various undergraduate students from the Riddle lab, who helped with the data recoding for this project: F. Archange, S. Avington, J. Favors, J. Harrison, I. Scott, M. Sherpa, S. Sims and H. Syed. In addition, I would like to thank the members of the Riddle lab and Dr H. Payami as well as two anonymous reviewers for helpful discussions and comments on the manuscript.