Wildtype CantonS flies (WT_CS) and the period mutants (per^s, per^l, and per⁰¹) from Konopka and Benzer (1971) were used for the experiments. In the following the per⁰¹ mutants will just be named per⁰. We obtained the flies from the Kyriacou group, who state in their paper that the flies are congenic and cantonized (Vanin et al., 2012). We could not obtain further information about the number of performed crosses with CantonS flies, but the flies have been co-isogenised for 12 generations in 1992 as is described in detail in Greenacre et al. (1993). Flies were reared on cornmeal/agar medium consisting of 0.8% agar, 2.2% sugar-beet syrup, 8.0% malt extract, 1.8% yeast, 1.0% soy flour, 8.0% corn flour, and 0.3% hydroxybenzoic acid at 25°C and 60% relative humidity under a light-dark (LD) cycle of 12:12 h (h). We started the first competition experiments under T = 24 h in 2013 (running until 2015) and the second ones, depending on the mutant either under long and short T-cycles or in LL in 2014 (running until 2016).

Ten food vials (diameter 48 mm, height 104 mm, filled with 28.5 ml food), each containing 15 virgin female and 15 male flies each of wild-type (CS) and period mutant flies were used to start the competition experiments (in sum, 60 flies per vial). The mutant flies were either per^s, per^l or per⁰ (Table 1). All vials were kept in a climate chamber at 25°C ± 0.2°C with 60% ± 2% of relative humidity. The light condition was either a T-cycle of 24 h (LD 12:12), a T-cycle of 29 h (LD 14.5:14.5) or of 19 h (LD 9.5:9.5) or constant light (LL) depending on the period mutant flies used for the subset of the experiment (Table 1). Light intensity was always 100 lux. LL conditions were only used for competing wildtype flies and per⁰ mutants, because wildtype flies become arrhythmic under LL. Thus, the wildtype activity pattern is no longer distinguishable from that of the arrhythmic per⁰ mutants and any selective advantages of possessing a circadian clock should disappear.

At 25°C, the generation time of fruit flies is ∼10 days and we did not observe any evident differences in developmental timing in the mutants. Therefore, we transferred all flies from the vials to new vials with fresh food every 14 days. At this time, ∼400 flies of the new generation had eclosed. The parental flies had died before the flies were flipped (when the larvae grow, the food becomes soft and fluid and the adults submerge in it being finally eaten by the larvae). Every third generation, the old vials were kept for another 2 to 3 days for more flies to hatch to carry out a census on mutant allele frequency (Figure 1). At these censuses, 32 males from each vial of these later hatching flies where used for locomotor activity recording to determine the genotype of each fly. Since the period gene is on the X-chromosome of which males carry only one copy, the genotype of each male becomes evident in its free-running period (short, long, ∼24 h or arrhythmic). Based on these tests, the genotype distribution among all flies in the experimental vials was determined (see below). In sum, 320 males (stemming from 10 vials) were used for determination of male genotype distribution at every census. This procedure was repeated over 60 generations for the 24 h T-cycle experiment and over 54 generations for the short- and long T-cycles. Figure 1 illustrates the procedure. Most importantly, the population size in the single vials stayed approximately constant throughout the entire experiments as far as we could judge by eye; this observation suggests the existence of a carry capacity and thus of competition among flies, respectively, larvae.

The outdoor experiments were performed in 2013 and 2014 outside the Biocenter and at the bee-station of the University of Würzburg – sheltered from rain and direct sunlight – where the flies could sense all changes in light, temperature and relative humidity (Figure 2). The outline of the competitions assay was in principle similar to that described above. However, due to low and variable temperatures, the flies took far more time to develop and to generate the next generation. Therefore, only three generations could be raised during each summer period (June to September/early October). The third generation was investigated for the genotype distribution of WT_CS and period mutant flies as described above. We compared these data with the third generation data (=census 1) from the long-term indoor experiments, using a general linear model with a “quasibinomial” error structure due to overdispersion of data. As suggested by Crawley (2013) we used F-tests for model comparison (procedure “anova” in R) – separately for each mutant – between the full model and a model version with the effect of “year” removed (the 2 years of outdoor experiments and 1 year of indoor experiment). Even though these constitute three independent experiments we apply post hoc Bonferroni adjustment for multiple testing.

Locomotor activity was recorded in Drosophila Activity Monitors (DAM; Trikinetics system, Waltham, MA, United States) at 25°C and 100 lux as described previously (Schlichting and Helfrich-Förster, 2015). Light was provided by white LEDs (Lumitronix, LED Technik, Hechingen, Germany). We recorded the activity of male wildtype, per^s, per^l and per⁰ flies (1) before the competition experiments, (2) at the censuses during the competition experiments and (3) in parallel to the competition experiments, in flies of each genotype that were kept without competition under the same conditions. The flies were recorded for 4–7 days under LD 12:12 cycles (light intensity 100 lux) and consecutively for 2–3 weeks under constant darkness (DD). In addition, we recorded male wildtype flies and per⁰ mutants under constant light of 100 lux to confirm that the great majority of flies in both strains are arrhythmic under such conditions. Furthermore, we recorded the activity of male wildtype flies and per^s and per^l mutants under the different T-cycles (T = 19 h and T = 29 h) in order to see how the phasing of their activity bouts changed under these conditions.

The raw activity data was exported as text files by the DAM-System Software, displayed as actograms using a Fiji¹ plugin – ActogramJ (v0.9, (Schmid et al., 2011) and saved as pdf files. Average activity profiles during the LD-cycles were calculated for each genotype as described in Schlichting and Helfrich-Förster (2015). From the average activity profile the fly’s overall activity (=number of beam crosses) during the first 3 h of the day [Zeitgeber Time (ZT) 0 to 3] was determined for each individual fly. Data were tested for normal distribution and investigated for genotype influences by a One-Way ANOVA followed by a post hoc test with Bonferroni adaptation. To determine rhythmicity and the endogenous free-running period of every single fly in DD, the raw data were analyzed by χ²-periodogram analysis in ActogramJ. At p > 0.05 the flies were assumed to be arrhythmic. From the periodogram analyses, the proportion of wildtype and mutant flies was determined for each competition pair in the competition assay, and the frequency to which the mutant flies persisted in the population was plotted for every census in a diagram.

Overall fitness depends on several components, e.g., fertility, mating success, survival during development, number of offspring, etc. In order to reveal the fitness of the clock mutant in comparison to wildtype flies in more detail, we determined several fitness components. All tests were done under LD12:12 with 100 lux during the light period and a constant temperature of 25°C.

All statistics were performed with R, version 3.4.4 (R Core Team, 2018).

Our first aim was to compare the activity patterns of wildtype flies and the period mutants under LD 12:12 and constant darkness (DD) with the reported data and to check whether per⁰ and wildtype flies behave similarly arrhythmic under constant light (LL). Figure 4A shows typical actograms and activity profiles for wildtype flies, per^s, per^l, and per⁰ mutants under LD and DD, while Figure 5 depicts typical actograms of wildtype flies and per⁰ mutants under LD and LL. Under LL, the great majority of wildtype and per⁰ flies were arrhythmic (details see legend of Figure 5). Under DD conditions, only per⁰ mutants exhibited arrhythmic activity patterns, all other fly strains were clearly rhythmic, whereby wildtype flies free-ran with a period close to 24 h, per^s with short period and per^l with long period (Figure 4; detailed period calculation see in section “Changes in free-running period over the course of the indoor competition experiment”). As reported previously (Hamblen-Coyle et al., 1992), wildtype flies exhibited two activity bouts around lights-on and lights-off (also called morning and evening peaks). Morning and evening activity anticipated lights-on and lights-off, respectively, and were separated by a siesta. per^s mutants showed a much earlier and shorter siesta and their evening activity occurred already in the early afternoon. The morning activity might take place already before lights-on, but it seemed largely suppressed by darkness. In contrast, per^l mutants, exhibited a large morning activity that extended until midday. The following siesta was very late and stretched until lights-off. Evening activity started after lights-off and appeared largely suppressed by darkness, as was the case for the morning activity of per^s mutants. The calculation of the activity amounts during the first 3 h of the day, revealed significant differences between wildtype flies and per^s and per^l mutants (Figure 4C): per^l mutants were much more active than wildtype flies, while per^s mutants were less active. per⁰ mutants lacked the siesta and showed no evident morning and evening activity bouts, but instead merely responded to lights-on and lights-off.

Under long and short T-cycles, the activity profiles of wildtype flies and per^s and per^l mutants changed drastically (Figures 6A,B). Under short T-cycles (19 h), wildtype flies did not at all anticipate lights-on but showed instead a long-lasting pronounced morning activity resembling very much the morning activity of per^l mutants under 24-h cycles (Figure 4A). Evening activity of wildtype flies shifted completely into the night, even more than that of per^l mutants under 24-h cycles. Under long T-cycles (29 h), wildtype flies showed early morning and evening activity, again resembling the activity profiles of per^s mutants under 24 h cycles with the exception that the nocturnal morning activity of wildtype flies was clearly visible, while it appeared suppressed in per^s mutants. The mutants’ activity profile under 19 and 29 h cycles, respectively, came close to the activity profile of wildtype flies under 24-h cycles. Overall, this experiment revealed that, under the T-cycles that matched the endogenous period of the mutants, the mutants had an almost “wildtype-like” phase-relationship of morning and evening activity to lights-on and lights-off, while wildtype flies behaved “mutant-like”. Consequently, the mutants should have a selective advantage over the wildtype flies under T-cycles, at least if fitness depends solely on the phasing of their activity bouts.

Figure 7 depicts the sampled frequency at which the period mutant flies persisted in the fly population over time when grown in competition to WT_CS flies. We directly compared the outcome of the experiment under the 24-h cycle with the outcome under the 29-h T-cycle for the per^l mutants, with the outcome under the 19-h T-cycle for the per^s mutants and with the outcome under LL for the per⁰ mutants, respectively. Under the 24-h day, the frequency of the mutants quickly declined in most of the food vials for all three period mutants (Figure 7 left). While per⁰ mutants declined to low proportions in all food vials (both, 24-h cycle and LL) and presumably completely disappeared from several vials (what we cannot say with certainty, as we only genotyped a fraction of the male population on each census), the per^l mutants persisted in at least 4 vials (in one of them to 80% and in another one to 40%) under 24-h cycle and in all ten vials under the 29-h cycle. per^s mutants persisted in at least 3 vials under both T-cycles. On average, per^l and per^s mutants remained to a certain low percentage in the population (per^l ∼20%, per^s ∼5%; Thick line in Figure 7 left).

To test whether these results could also be explained by genetic drift, we created 95% confidence intervals for the “true” frequency of mutant males in the population and (hatched gray lines in Figure 7) and just for samples of 32 males (hatched black lines) under the assumption of pure genetic drift. Because only males were genotyped, the period mutants are linked to the sex chromosome, data sets represent time-series, and a census was only taken every 3rd generation, a direct (analytical) calculation of confidence intervals is not possible in this case. Instead, we simulated the competition experiments exactly replicating the protocol of population initialization, generation transfer, and random sampling of 32 males every 3rd generation as applied in the experiments but under the (neutral) assumption of identical fitness of wildtype and mutant genotypes and the assumption of random mating and random sex determination. To account for sampling effects at any of these steps, discrete values for the number of females and males as well as the different genotypes by drawing random values from a binomial (two sexes, two possible male genotypes) or multinomial (three female genotypes) distribution, were assumed. These simulations were replicated 100,000 times and the lower and upper 2.5% quantiles were then used to define the 95% confidence intervals for both, the frequency of the mutant allele in the whole population and for the frequency in just a sample of 32 males. The calculated confidence intervals are shown in Figure 7. The results for the per⁰ and the per^s mutants can clearly not be explained by genetic drift alone but indicate strong selection against the mutants in both, the 24-h cycles and the LL, respectively, the 19-h cycle. For per^l, however, the results under the 29-h cycle are consistent with genetic drift and also under the 24-h cycle some of the populations show fluctuations in the mutant’s frequency consistent with genetic drift (but see below for deviations from the genetic drift hypothesis).

If the decrease of the period mutants in the fly population depends only on the presence and speed of circadian clock, per^l mutants are expected to improve performance against wildtype flies under the 29-h T-cycle and per^s mutants under the 19-h T-cycle, while per⁰ mutants should be equally fit as wildtype flies under LL conditions. However, only per^l mutants performed better under the 29-h T-cycle than under the 24-h cycle, while per^s and per⁰ mutants performed equally poor under the 19-h T-cycle and LL as they did under the 24-h cycle, respectively (Figure 7 right). Under the 29-h T-cycle, per^l mutants persisted in all 10 vials to varying percentages, on average to ∼50%. In two vials, the mutants outcompeted the wildtype flies, meaning that in the end they represented presumably 100% of the population. This cannot be due to genetic drift. The final frequency of per^l mutants in the other vials are within the confidence intervals of genetic drift and might theoretically be caused by it (see above). However, the strong decline of the mutant frequency at the beginning of the experiment (lasting until census 6) that is followed by a clear increase in frequency are hard to reconcile with genetic drift. In particular, the comparison between the shape of the “frequency curves” at T = 24 h and T = 29 h shows that per^l mutants perform clearly better under the 29-h day, which supports the resonance hypothesis. Nevertheless, the poor performance of the per^s and per⁰ mutants under both T-cycles clearly shows that additional factors contribute to the fitness of the mutants. These factors can be related or completely independent of the circadian clock.

In the outdoor experiments, the generation time of the flies was ∼1 month and we could breed through only three generations in each summer. In both summers, 2013 and 2014, the wildtype flies clearly outnumbered the mutants in the third generation (Figure 8). Roughly speaking, the mean proportion of the mutant fell from 0.5 to about 0.25 in all the experiments significantly deviating from the initial 1:1 ratio (p < 0.001) indicating selection against the mutant alleles. There was no significant difference in the percentage of remaining mutants between the 2 years under outdoor conditions and also not between the outdoor and indoor conditions (d.f. 2, 23, deviance reduction = 10.25, F = 3.065, p > 0.05 following post hoc Bonferroni correction). The fact that the decline in the mutant abundance was about similar in the out- and indoor experiments suggests that the different conditions, in particular also the differences in the light regime, between indoor and outdoor conditions do not have a relevant impact on the mutants’ success. However, different selective response may also not have been possible due to limited genetic variance in the inbred initial populations.

In the original paper (Konopka and Benzer, 1971), per^s mutants were reported to free-run with a period of ∼19 h and per^l mutants with a period of ∼ 28 h under constant darkness. We determined the free-running period and the activity patterns under LD12:12 for 60 wildtype and mutant flies, respectively, before they underwent the competition experiment. We found a mean period of 24.3 ± 0.04 h for wildtype flies, a period of 19.0 ± 0.03 h for per^s mutants and a period of 27.6 ± 0.06 h for per^l mutants. Thus, while the free-running period of per^s mutants matched the one previously reported, the free-running period of our per^l mutants was about 0.4 h shorter than originally reported and only 3.3 h longer than that of wildtype flies. To see whether the periods remained the same during the competition experiments, we determined them in the flies that were recorded every census throughout the competition experiment. The results are plotted in Figure 9. To our surprise, we found that per^l mutants lengthened their period continuously, at the beginning very fast (at census 4 they reached already 29 h) and later slower. At the end they reached a period of 31.1 ± 0.87 h (± SD) in the T = 24 h experiment and a period of 30.5 ± 0.71 h (± SD) in the T = 29 h experiment. This period lengthening was similar in all vials (Figure 9B), indicating that it was not due to founder effects or genetic drift. The other fly strains (wildtype flies and per^s mutants) kept their period rather constant throughout the different experiments (Figure 9). The same was true for the wildtype flies and per^s mutants that were kept alone, but maintained and recorded in parallel to the competing flies (data not shown). Again, the situation was a bit different for the per^l mutants (pale blue line in Figure 9). per^l mutants kept alone, without competition with the wildtype flies did also lengthen their period over time but this lengthening was far less than for the flies kept in competition with wildtype flies: per^l mutants reached a maximal period of 28.7 ± 0.57 h (±SD) in the first experiment (T = 24 h) and of 28.6 ± 0.62 h (±SD) in the second experiment (T = 29 h). This shows that a period lengthening of ∼2 h was caused by growing per^l flies together with wildtype flies.

In the 24-h day, per^l mutants remained at ∼20% in the fly population and in one food vial they even dominated over the wildtype flies (Figure 7). The main reason for their relative higher competitive fitness, may lie in their longer endogenous period that enabled them to have a long and pronounced morning activity (Figure 4). Since courting and mating occurs mainly in the morning (Sakai and Ishida, 2001a, b), per^l males might be more successful in mating with females simply because they are rather active in the morning hours. It might be irrelevant for their competitive fitness when they afterward extend the siesta until the evening and start eating only in the night. Indeed, we found that per^l mutants had a relatively high mating success in the female mating preference assay (Figure 10D). Another positive factor for their fitness may be the mating preference of virgin WT_CS/per^l heterozygotes for per^l males (Figure 10C). These positive effects may have dominated over the lower survival rate of per^l larvae (Figure 10B) and the reduced number of adult offspring of per^l females (Figure 10E); note also that we only have egg-to-adult survival data only for homozygous individuals so that we do not know how strong the survival effect of period alleles is in heterozygous individuals. Earlier studies even reported a slower development of per^l mutants (Kyriacou et al., 1990a). In our study, we did not observe evident differences in developmental timing, but we cannot exclude that minor differences exist, especially not under competition. In any case, such differences had obviously no negative effects on the competitive fitness of per^l mutants.

per^s mutants persist in the population at ∼10%, independently of the external Zeitgeber period. Putative reasons for their persistence may be the strong preference of WT/per^s heterozygote females and even of homozygote wildtype females for per^s males. In particular, when the mutant allele becomes rare and thus hardly occurs in homozygous females, the mutant male benefit may compensate for the female fitness disadvantages thus resulting in a stable coexistence of wild-type and mutant alleles.

Perhaps more difficult to explain is the very low overall fitness of per^s mutants in comparison to wildtype flies. per^s males transfer slightly less number of sperm (Figure 10A), but as explained earlier this is unlikely to affect the reproductive fitness of the females. The low number of surviving per^s larvae is for sure reducing the fitness of per^s mutants and in contrast to per^l mutants this is not compensated by a higher mating success of per^s males (Figure 10D). Nevertheless, the per^s genotype had no negative effect on the number of adult offspring (Figure 10E), suggesting that per^s female may compensate the high larval mortality by laying more eggs. Once again, the most likely reason for the low fitness of per^s mutants may lie in the activity pattern of per^s males (Figure 4A). In contrast to per^l mutants and wildtype flies, per^s mutants are rather inactive during the morning. Before lights-on, their activity appears suppressed by darkness and after lights-on they show only a very brief morning activity before they enter the siesta. Consequently, the mating activity of male flies may be rather low and this fits with their rather low mating success (Figure 10D). Most interestingly, this may also explain why per^s mutants do not perform better under the 19-h day, because even then the morning activity of male mutants was lower than that of wildtype males (Figure 6). Nevertheless, still other factors that are independent of timing may contribute to the low fitness of per^s mutants.

per⁰ mutants appear to have no fitness advantage against wild-type flies. The initial decline in the first generations was in fact somewhat slower than that for the other two mutants (cf. Figure 7) but did not stabilize at low proportion as seems to be the case for other two mutants. They even lose quickly against wildtype flies under LL conditions, under which the wildtype flies are similarly arrhythmic and can therefore take no advantage of their circadian clock. These findings are hard to explain in light of the experiments performed in this study. In contrast to previous observations (Greenacre et al., 1993; Beaver et al., 2002), we found no reduction of transferred sperm in per⁰ mutants (Figure 10A). Furthermore, the survival of per⁰ from egg to pupa was much better than that of per^l and per^s mutants (Figure 10B) and homozygote wildtype flies showed a mating preference for per⁰ males (Figure 10C). The mating success of per⁰ males was also very good (Figure 10D) and heterozygous WT/per⁰ mutants showed even a tendency to have more adult offspring than wildtype flies (Figure 10E). Consequently, there must exist other factors that reduce the fitness of per⁰ mutants and that possible only contribute when the flies have to compete with wildtype flies. per⁰ mutants are reported to sleep less (Shaw et al., 2002), to have deficits in certain aspects of learning and memory (Chouhan et al., 2015), to show a lower neuronal activity in neurons that control fast escape responses (Megighian et al., 2001), and to be more sensitive to oxidative stress (Krishnan et al., 2008). Constant light (especially short-wavelength light) also generates stress, and this might be one reason why per⁰ flies have disadvantages against wildtype flies under such conditions. Altogether, this might explain the observed reduced reproductive fitness of per⁰ mutants when raised in competition to wildtype flies. To sort this out in more detail, additional experiments are needed, which also decipher the effects of population density on the mutant’s fitness under competition conditions.

One of the most surprising result of our study was the dramatic period lengthening of the per^l mutants over the course of the competition experiment. At the beginning, the mutants had slightly shorter periods than reported in previous studies (Konopka and Benzer, 1971), but already at census 2 (generation 6) they reached periods of ∼29 h, and then asymptotically approached periods of 31 h in both experiments (under the 24 h and the 29 h day; Figure 9). A selection for long periods by the 29 h cycle cannot be the cause of this period lengthening, because it occurred also under the 24 h cycle. Most interestingly, the 31 h period precisely coincides with the period that Ewer et al. (1990) measured in per^l mutants at 25°C. Possibly, during the long inbreeding, our per^l mutants have accumulated genetic modifiers that shortened their period so that it stayed closer to 24 h. This would imply that wildtype and mutant flies had not the identical genetic background and were probably not outcrossed with each other at the beginning of the experiment as we had presumed. During the gradual exchange of the genetic background between wildtype and per^l mutants in the competition experiment, the genetic modifiers may have got lost and the original period of per^l mutants has reappeared. This interpretation is supported by the fact that per^l kept in isolation do not show a comparable lengthening of their period. In wildtype flies and per^s mutants, we could not see any evident changes in period during the competition experiment. However, this does not mean that these flies had an identical genetic background. Future studies with the now perfectly “cantonized” period mutants are necessary to clarify this.

We have shown that possessing an endogenous clock that runs with a period close to 24 h has a clear fitness advantage when flies have to compete with others for food and mating partners. The right timing of activity, especially in the morning when courting and mating occurs, may be crucial for the survival in the population. In addition, we have shown that multiple other factors contribute to the performance of the mutants. Especially the per^s and per⁰ mutants are clearly less fit as compared to wildtype flies. This confirms previous observations that the period gene has pleiotrophic effects on physiology, metabolism and behavior, independent of its effect on timing. Furthermore, we have clear evidences that the genetic background contributes, too. At least per^l mutants appeared to have accumulated background mutations that modulated the effects of the per^l mutation on the circadian period in such a way that it remained closer to 24 h than in the original stocks maintained at 24 h cycles. After now having outcrossed the flies for >50 generations, we have certainly eliminated such background differences. Thus, we are in the perfect situation to test the impact of the period gene on fitness in more detail in future experiments.