Cross-tolerance evolution is driven by selection on heat tolerance in Drosophila subobscura

Sampling and maintenance

D. subobscura females were collected in the spring 2014 at the Botanical Garden of the Universidad Austral de Chile (Valdivia, Chile; 39° 48′S, 73° 14′W) using plastic traps containing banana/yeast baits. Two hundred females were collected and placed individually in plastic vials containing David’s killed-yeast Drosophila medium to establish isofemale lines. In the next generation, 100 isofemale lines were randomly selected, and 10 females and 10 males per line were placed in an acrylic cage to establish a large, outbred population. In the next generation, the flies from this cage were randomly divided into three population cages (R1, R2, and R3), attempting to assign the same number of flies to each cage. After three generations, the flies in each replicate cage were randomly divided into four population cages, trying to assign the same number of flies to each cage. This procedure established 12 population cages, which were randomly assigned to each artificial selection protocol in triplicate: fast-ramping selection, fast-ramping control, slow-ramping selection, and slow-ramping control lines (Fig. S1). During the selection experiments, population cages were maintained at 18 °C (12:12 light-dark cycle, 50–60% humidity) in a discrete generation, controlled larval density regime (Castañeda, Rezende & Santos, 2015). Each population cage had a population size of 1,000–1,500 breeding adults.

Heat tolerance selection

For each replicate cage, 120 four-day-old virgin females were randomly collected and placed individually in vials containing food medium. Two four-day-old males were randomly collected from each replicate cage and placed in the same vial as each female from the same replicate line. They were allowed to mate for 2 days, during which females laid eggs in the vial. Then, males were removed from the vials and females were placed individually in capped 5-mL glass vials. The vials containing eggs were placed in an incubator at 18 °C with a 12:12 light-dark cycle and 50–60% humidity. The offspring from these vials (hereafter referred to as "offspring vials") will be used to found the next generation of each replicate according to the selection protocol explained below.

The glass vials containing each mated female were attached to a plastic rack and immersed in a water tank (tank size: 35 cm × 42 cm × 45 cm; filling volume: 60 L) with an initial temperature of 28 °C (± 0.03 °C), controlled by a heating immersion unit (model ED; Julabo Labortechnik, Seelbach, Germany). After an equilibration period of 10 min, the temperature was increased to 0.08 °C min⁻¹ for the slow-ramping selection protocol or 0.4 °C min⁻¹ for the fast-ramping selection protocol. Assays were stopped when all female flies collapsed. Each assay was recorded using a high-resolution camera (model D5100; Nikon, Tokyo, Japan) and then visualized to score the knockdown temperature for each fly, defined as the temperature at which each fly ceased to show coordinated movements. After each assay females from each replicated line assigned to the selection protocols were ranked by knockdown temperature. Four virgin females from the offspring vials of the 40 flies with the highest knockdown temperature (top 30% of each assay) were selected at random to establish the next generation. For the fast and slow control lines, the knockdown temperature was measured as described above, but the progeny was randomly selected from the offspring vials to establish the next generation (i.e., regardless of the knockdown temperature of their mother). This artificial selection experiment was performed for 16 generations, after which flies from each selection treatment were placed in separate acrylic cages and maintained without selection (e.g., relaxed selection) at 18 °C and a 12:12 light-dark cycle.

Knockdown time in static assays

Eggs were collected from each population cage and transferred to vials at a density of 40 eggs/vial. At 4 days of age, ten females and ten males from each population cage were tested to measure their heat knockdown time at four different static temperatures: 35, 36, 37, and 38 °C. This experimental design allowed the measurement of 960 flies (10 flies × 2 sexes × 4 static temperatures × 4 selection treatments × 3 replicated lines). Static assays were performed similarly to knockdown temperature assays, but constant temperatures were used instead of ramping temperatures. A total of 240 flies were measured for each static temperature, except for the assay at 35 °C (178 flies) because two flies died before the start of the assay, and a video file of one assay was corrupted (data for 60 flies were lost). For the 37 °C assay, four flies died before the assay began, and the collapse time could not be measured for six flies. Finally, for the 38 °C assay, three flies died before the start of the assay and the collapse time could not be measured for five flies. Heat knockdown assays were performed in generation 23 (Fig. S1).

Desiccation and starvation resistance

Eggs from each replicate cage were collected and placed in vials at a density of 40 eggs/vial. Only fast control lines were measured as control lines. This decision was based on logistical reasons (i.e., the high number of vials) and statistical support because fast and slow control lines did not differ in their knockdown times and CT_max values (see the Results section).

For desiccation resistance assays, flies were separated by sex. Five flies of the same sex were placed in a 15 mL plastic vial containing five CaSO₄-based desiccant droplets (Drierite®; W.A. Hammond Drierite Co., Xenia, OH, USA). The vials were sealed with parafilm to prevent water enters to vials (flies had no access to food or water during the assay). For starvation resistance assays, flies were separated by sex and the five flies of the same sex were placed in a 15 mL plastic vial with a mixture of agar and water (flies had access to water but no food). The vials containing flies were placed in an incubator connected to a PELT-5 thermal controller (Sable Systems International, Las Vegas, NV, USA), which allowed to maintain a constant temperature of 18 °C. A cold light connected to a timer was placed inside the incubator to maintain a 12L:12D photoperiod cycle throughout the experiments.

For both desiccation and starvation resistance assays, the number of live flies was counted every 3 h until all the flies were dead. Desiccation and starvation resistance were measured in 126 vials containing five flies each, respectively (7 vials × 2 sexes × 3 selection treatments × 3 replicate lines). These experiments were conducted using flies from generation 24 (Fig. S1).

Statistical analysis

Normality and homoscedasticity were tested for all variables, and the knockdown times were squared root transformed to meet the parametric assumptions (Quinn & Keough, 2002). All analyses were performed with R software (R Core Team, 2024).

Heat tolerance

The selection effect was tested separately for the fast- and slow-ramping datasets because the knockdown temperature is higher in fast-ramping assays than in slow-ramping assays (Chown et al., 2009; see Mesas, Jaramillo & Castañeda, 2021). For the knockdown time analysis, a mixed linear model with ramping selection (fixed effect with fast-control, slow-control, fast-selection, and slow-selection lines as levels), sex (fixed effect with females and males as levels), and replicate lines nested within the thermal selection (random effect with replicates 1, 2 and 3 as levels) was performed using the library lme4 package for R (Bates et al., 2015). Fixed effects were tested using a type-III ANOVA, and the random effect was tested using a likelihood ratio test (LRT) by comparing the models that included or excluded replicate lines. Both tests were carried out using the library lmerTest package for R (Kuznetsova, Brockhoff & Christensen, 2017). If the selection effect was significant, the emmeans package for R (Lenth, 2025) was used to perform posteriori comparisons with a false-discovery rate (FDR) approach.

Knockdown times were also used to plot the survival curves based on the Kaplan-Meier formula using the survfit function implemented in the survival package for R (Therneau, 2024).

Thermal death time curves (TDT)

Average knockdown times were calculated for each sex, replicate line, and selection treatment combination (Table S1). These values were regressed against the assay temperatures according to Eq. (1) (Rezende, Castañeda & Santos, 2014):

(1) $$lo{g}_{10}t={\displaystyle \frac{C{T}_{max}-T}{Z},}$$where T is the assay static temperature (°C), CT_max is the upper thermal limit (°C), t is the knockdown time (min), and z is the thermal sensitivity. These curves allowed the estimation of CT_max as the extrapolated temperature that would result in a knockdown time of log₁₀ t = 0 (i.e., knockdown time at 1 min) and the estimation of the thermal sensitivity (z = –1/slope), where lower z value, the higher the thermal sensitivity.

Using Eq. (1), 24 TDT curves (2 sexes × 3 replicate lines × 4 selection protocols) were fitted, from which CT_max and z values were estimated as described above. A linear model with ramping selection treatment (levels: fast-control, slow-control, fast-selection, and slow-selection lines), sex (levels: females and males), and their interaction was performed to evaluate their effects on CT_max and z values. TDT analysis did not include replicate lines as a random effect because only one CT_max and z value was estimated by each replicate line. If the selection treatments had a significant effect on CT_max or z, an FDR approach was used to perform posteriori comparisons. Additionally, a linear mixed model with ramping selection (fixed effect with fast-control, slow-control, fast-selection, and slow-selection lines as levels), sex (fixed effect with females and males as levels), and replicate lines nested within the thermal selection (random effect with replicates 1, 2 and 3 as levels), and assay temperatures (as covariate) was fitted on the knockdown time using the lmer package for R.

Desiccation and starvation resistance

To determine the lethal time at which 50% of flies of each vial were dead (LT₅₀), a generalized linear model following a binomial distribution was fitted with the proportion of flies alive as the dependent variable and time as the predictor variable. The generalized linear model was run using the glm function of the lme4 package for R (Bates et al., 2015). The LT₅₀ of each vial was then estimated using the function dose.p from the MASS package for R (Venables & Ripley, 2002).

To estimate the median LT₅₀ and the 95% confidence intervals for each selection treatment and sex, LT₅₀ was transformed into a survival object using the surv and survfit functions of the survival package for R (Therneau, 2024). This procedure also allowed to estimate the survival curves in each vial. Finally, to test the effect of selection treatment (fixed effect with levels control, fast-selection, and slow-selection lines), replicate lines (random effect with levels R1, R2 and R3), and sex (fixed effect with levels females and males) on desiccation and starvation resistance, a mixed effect Cox proportional regression model was fitted with LT₅₀ as the dependent variable. The mixed effect Cox model was run using the coxme function of the coxme package (Therneau, 2024). To evaluate the variability among replicate lines, the same model was ran, excluding the replicate line effect, using the cox function of the survival package (Therneau, 2024). An LRT was used to compare mixed vs. fixed models.

Knockdown temperature evolution

Knockdown temperature evolved in response to artificial selection for increased heat tolerance, regardless of the ramping assay protocol. For the fast-ramping comparison, selected lines showed higher knockdown temperature than control lines (mean fast-ramping selection lines ± SD = 37.71 ± 0.68 °C and mean fast-ramping control lines ± SD = 37.23 ± 0.79 °C; F_1,4 = 32.0, P = 0.005). For the slow-ramping comparison, selected lines showed higher knockdown temperature than control lines (mean slow-ramping selection lines ± SD = 35.48 ± 0.66 °C and mean fast-ramping control lines ± SD = 34.97 ± 0.72 °C; F_1,4 = 41.7, P = 0.003). These results were previously reported by Mesas, Jaramillo & Castañeda (2021) and are reported here to show that selected lines used in this study evolved higher thermal tolerance compared to control lines.

Knockdown time evolution

As expected, the knockdown time decreased significantly as the static temperatures increased (F_1,877 = 649.1, P < 2 × 10⁻¹⁶). The knockdown time (mean ± SE) for each static assay are as follows: 35 °C = 33.77 ± 0.87 min; 36 °C = 16.98 ± 0.44 min; 37 °C = 8.82 ± 0.23 min; and 38 °C = 6.68 ± 0.18 min.

Knockdown times only differed significantly between selection treatments when flies were assayed at 36 and 37 °C (Table 1; Fig. 1). At these temperatures (Figs. 1C, 1E), slow-ramping selection lines showed higher knockdown time than slow-ramping control lines (36 °C: P = 0.005 and 37 °C: P = 0.004; Tables S1, S2). Also, fast-ramping selection lines showed higher knockdown time than fast-ramping control lines (36 °C: P = 0.005 and 37 °C: P < 0.001; Table S1, S2). In addition, fast-ramping selection lines showed higher knockdown time at 37 °C than slow-ramping selection lines (P < 0.001) but not at 36 °C (P = 0.053) (Table S1, S2; Figs. 1C, 1E), whereas fast-ramping control and slow-ramping control lines owed the similar knockdown time at all static temperatures (Table S2; Fig. 1). On the other hand, replicate lines showed similar knockdown time at all static temperatures (LRT: χ²₁ = 0, P = 1). In addition, females showed higher knockdown time than males but only when flies were assayed at 35 and 38 °C (Table 1; Figs. 1B, 1H). Finally, non-significant interactions between selection and sex were found for all static temperatures (Table 1).

TDT curves evolution

Linear regressions between log₁₀(LT₅₀) and assay temperatures enabled the estimation of 24 TDT curves (4 selection treatments × 3 replicate lines × 2 sexes) with high coefficients of determination (mean R² = 0.946, range: 0.820–0.989; Table S4), confirming that heat knockdown time is linearly related to stressful sublethal temperatures. From these TDT curves, the average CT_max (mean ± SE) was 41.21 ± 0.02 °C, and the average z (mean ± SE) was 4.18 ± 0.01 °C. A two-way ANOVA showed that a significant effect of selection treatments on CT_max (F_3,16 = 5.16, P = 0.011; Fig. 2A), but no effect of sex (F_1,16 = 0.008, P = 0.92) or the interaction between selection and sex (F_3,16 = 2.21, P = 0.13). Posteriori comparisons indicated that slow-ramping selection lines showed higher CT_max than the slow-ramping control lines (t₁₆ = 2.86, P = 0.034). In addition, fast-ramping selection lines also showed higher CT_max than the fast-ramping control lines (t₁₆ = –2.49, P = 0.049). While control lines of the fast-ramping and slow-ramping treatments showed similar CT_max values (t₁₆ = 0.93, P = 0.44). Regarding z (i.e., thermal sensitivity; Fig. 2), it shows no significant effects of selection treatments (F_2,18 = 1.04, P = 0.37), sex (F_1,18 = 1.27, P = 0.28), nor the interaction between selection treatments and sex (F_2,18 = 1.82, P = 0.19). In summary, the evolution of a higher CT_max is not associated with an evolutionary change in thermal sensitivity (Fig. 2B). In addition, the relationship between TDT parameters did not change with the evolutionary increase of CT_max (r_{control-lines} = 0.981 and r_{selection-lines} = 0.929; Z-test = 1.06, P = 0.29).

Desiccation resistance evolution

Survival analysis showed a significant effect of sex and selection treatment on desiccation resistance, but not an interaction between both effects (Table S5). Variation among replicate cages was not significant (LTR: χ²₁ = 1.23, P = 0.27). Males showed a higher risk of desiccation than female flies (hazard ratio = 7.11, P < 2 × 10⁻⁷; Fig. 3). Females of selected lines showed higher desiccation resistance than control lines (LTR: χ²₂ = 6.72, P = 0.03; Fig. 3A). Specifically, females of the slow-ramping selection lines showed a higher desiccation resistance than females of the control lines (hazard ratio = 0.42, P = 0.009), whereas females of the fast-ramping selection and control lines showed similar desiccation risk (hazard ratio = 0.56, P = 0.072). On the other hand, males showed no differences in desiccation resistance between selected and control lines (LTR: χ²₂ = 1.88, P = 0.4; Fig. 3B).

Starvation resistance evolution

A significant effect of sex, selection treatment, and the interaction between the two effects on starvation resistance was found in the survival analysis (Table S6). Variation among replicate cages was not significant (LTR: χ²₁ = 0.0034, P = 0.95). Males had lower starvation resistance than female flies (hazard ratio = 22.75, P < 1 × 10⁻¹⁶; Fig. 4). In female flies (Fig. 4A), fast-ramping selection and slow-ramping selection lines showed a higher starvation risk than control lines (hazard ratio = 2.37, P = 0.009; and hazard ratio = 2.20, P = 0.014, respectively). In contrast, male flies had an opposite pattern (Fig. 4B): slow-ramping selection lines had a lower starvation risk than control lines (hazard ratio = 0.50, P = 0.03), but nonsignificant differences were found between fast-ramping selection and control lines (hazard ratio = 0.64, P = 0.16).