Abstract
Climate change is well understood to be a major threat to biodiversity, but sublethal impacts of high temperatures, such as reduced fertility, have been poorly studied. We examined a panel of 43 Drosophila species, finding that 19 experience significant fertility loss at temperatures up to 4.3°C cooler than their lethal temperature limits. We found that upper thermal fertility limits explain global distributions of species better than limits based on lethal temperatures. This suggests that limits to reproduction, rather than limits to survival, can underpin species distributions in nature. Given that high temperatures impair male fertility across a broad range of animals and plants, many species may be at increased risk of extinction due to inability to reproduce at high temperatures.
One Sentence Summary Species’ distributions and response to climate change are strongly affected by the temperature at which they lose fertility.
Main Text
We urgently need to understand how rises in temperature will impact biodiversity(1). To do this we must understand the physiological, behavioral and evolutionary factors that underpin species’ current thermal distributions(2, 3). Laboratory-derived estimates of the highest temperatures at which an organism can function provide measures of species’ thermal tolerances(4). These measures of upper thermal limits have improved the accuracy of functional species distribution models(5) which can be extrapolated to climate change scenarios, allowing ecologists to forecast future species distributions(6). Accurate predictions of species’ distributions are invaluable for prioritizing conservation efforts(7) and predicting the invasion of disease vectors(8).
Upper thermal tolerance limits are usually based on the temperatures that cause loss of coordinated movement, coma, respiratory failure, or death: the species’ critical thermal limit. Despite these being measured in artificial laboratory conditions, critical limits correlate reasonably well with species’ distributions(4) and have been used to estimate species’ capacity to tolerate temperature increases across their current distribution range; their ‘thermal safety margins’(3). However, persistence of populations is not determined solely by survival, but also by reproduction. There is evidence that sub-lethal temperatures cause losses in fertility in plants(9), insects(10-12), fish(13), aquatic invertebrates(14), birds(15) and mammals, including humans(16). These effects include direct impacts on physiological processes (10, 15, 17) and indirect influences via changes in behavior and phenology(18). Previously, we proposed that temperatures at which fertility is lost, the thermal fertility limits (TFLs) (18), may be a critical but understudied part of species’ true upper thermal limits. If TFLs are lower than critical limits, then many organisms will be more vulnerable to climate change than currently thought. If TFLs and critical limits correlate poorly then we may misidentify which species are most at risk from rising temperatures. Critically, we need to know whether TFLs measured in laboratory conditions can predict the distribution of natural populations better than measures of critical limits.
Here, we recorded three measures of upper thermal limits in adult males from 43 species of Drosophila fruit flies. First, we exposed flies to a 4-hour static heat stress at a range of temperatures from benign through to lethal(19). From these data we estimated the temperature at which 80% of living males are sterilized (TFL80) and the temperature that is lethal to 80% of individuals (LT80). Fertility was assayed at two time points: (i) over 1-6 days post-heat, to capture any immediate sterilizing effect of heat, and (ii) 7-days after heat-stress to capture any recovery of fertility or delayed sterility. These data allow us to compare fertility and survival thresholds under identical conditions. In a separate experiment, we measured the temperature at which males lose coordinated motor function under ramping temperature conditions (CTmax). This measure is widely used across animal systems to predict species’ sensitivity to climate change (3, 4).
We found that 11 of 43 species experience an 80% loss in fertility at cooler-than-lethal temperatures immediately following heat-stress (Fig S1). Interestingly, rather than seeing a recovery of fertility over time, the impact of high temperatures on fertility was more pronounced 7-days post heat stress (Fig 1). Using this delayed measure of fertility, nearly half of species (19/43) showed significant fertility loss. The difference between lethal and fertility limits ranged from 0°C to 4.3°C (mean = 1.15 ± 0.22°C), and LT80 and TFL80 predict dramatically different relative ranking of species’ robustness to high temperature (Fig S2).
These data show that fertility loss at sub-lethal temperatures is common across Drosophila. This suggests that ignoring TFLs has led to the overestimation of thermal tolerance for many species. This may be particularly problematic for efforts to preserve rare and endangered species or control emerging pests and disease vectors. For example, we found that the crop pest species Zaprionus indianus (African fig fly) has a relatively high upper lethal limit (LT80 = 37.9°C) but a much cooler upper fertility limit (TFL80 = 34.8°C). This switches the species’ thermal hardiness rank from rank 5th to 17th of species here (Fig S2). This pest is currently expanding throughout the New World, and this disparity could have marked consequences for forecasting its putative range and hence where to concentrate control efforts(20).
Our work demonstrates that TFLs can be substantially lower than lethal temperatures when measured under the same conditions in the laboratory. However, the key question is whether TFLs actually shape organisms’ distributions in nature. To test this, we used existing data on the distribution of each Drosophila species and integrated this with global climate data for each location to estimate the maximum temperatures species are likely to encounter in nature(19). Previous work using 95 Drosophila species found that the mean maximum summer temperature measured in this way correlates with CTmax for species from dry (<1000ml annual rainfall) but not from wet (>1000ml annual rainfall) environments(4). We first verified this finding in our smaller dataset of 43 species; CTmax significantly interacts with annual precipitation to predict mean maximum environmental temperature (PGLS: F = 6.021,39, P = 0.018, adjR2 = 0.37). We then tested whether either LT80 or TFL80 measured 7-days after heat stress are better predictors of species’ thermal habitats than CTmax (Table S1). Across all species, we found that a model with TFL80 as the sole physiological predictor had the strongest correlation with mean maximum environmental temperature with no significant interaction with precipitation (PGLS: interaction term: F = 0.5311,39 P = 0.471, TFL80: F = 39.31,39 P < 0.001, adjR2 = 0.55). LT80 also did not significantly interact with precipitation to predict environmental temperature but explained considerably less variation than TFL80 (PGLS: interaction term: F1,39 = 0.99, P = 0.324, LT80: F1,39 = 0.99, P = 0.324, adjR2 = 0.31). Hence, TFL results in a 48% and 77% improvement in the accuracy (R2) of species distribution predictions compared to CTmax and LT80 respectively. This dramatic improvement is largely due to TFL80 more accurately predicting the thermal range for species from high humidity environments (Table S2 & Fig S3). We find qualitatively similar results if both lethal and sterilizing limits are included as predictors in the same model (Table S3). The power of TFL80 to predict the distribution of Drosophila in nature strongly suggests that fertility losses due to high temperature are an important, but previously ignored, determinant of where species can persist.
Predicting species’ vulnerability to current and future temperature extremes is critical to protect biodiversity(1). “Thermal safety margins” measure species’ vulnerability to climate change by taking the difference between an organism’s physiological thermal limit and the maximum air temperature it is likely to experience(21). Although safety margins often use complex measures of environmental temperature to improve accuracy (e.g. microhabitat availability, thermoregulation behavior of adults(3)), they remain reliant on relatively simple measures of thermal vulnerability. We find that TFL80 measured seven days after heat-stress produces significantly smaller safety margins than either LT80 or CTmax (LMER: χ2 = 40.36, df = 2, P < 0.001, Fig S4). For populations that already live near the upper edge of their thermal range, TFL-based safety margins are reduced from 2.59 ± 3.08°C under assumptions of LT80, to 1.43 ± 2.80°C. This predicts that 20 of the 43 species studied here are found in environments in which air temperatures exceed safety limits during the hottest part of the year. We illustrate the implications of this with case-study distribution models of Drosophila flavomontana; safety margins based on TFL80 predicts a 17.9% reduction in habitable landscape compared to an identical LT80-based model under current climate conditions (Fig 2A). This disparity between predictions based on sterility and lethality grew to 48.0% by the year 2070 under moderately optimistic future climate forecasts (RCP4.5, Fig 2B), and to 58.9% under pessimistic climate change scenarios (RCP8.5, Fig 2C). TFL-based models also predict that by 2070 the available habitat for D. flavomontana will have reduced by 42.3% and 62.9% under RCP4.5 and RCP8.5 respectively.
If our data for Drosophila can be extrapolated to other organisms, it is likely that male fertility losses at high temperatures are common and can occur at substantially lower temperatures than the species upper lethal limit. The limited data on fertility at extreme temperatures supports this, with losses in male fertility at high temperatures seen across a broad diversity of organisms(18). It is possible that behavioral thermoregulation will reduce the impact of high temperatures on fertility in nature. Drosophila are able to behaviorally thermoregulate, moving to leaf litter, shade, or higher altitudes, with many species able to survive high temperature periods by aestivating as adults, eggs or pupae(22). Despite this, we still find that the distribution of species is predicted by thermal fertility limits. Even species that might be predicted to have high thermal tolerance can show evidence of thermal fertility losses. For instance, the zebra finch, a high temperature adapted desert-dwelling endothermic species with naturally high body temperature and good thermoregulation, shows substantial damage to sperm at temperatures it regularly experiences in nature(15).
Our work suggests that temperature-driven fertility losses may be a major threat to biodiversity during climate change. We urgently need to understand the range of organisms likely to suffer similar fertility losses in nature, and the traits that predict vulnerability. However, we currently do not understand the physiology underlying variation in TFLs between species, nor the selective forces that created this variation. Ultimately, we need to know whether evolution for higher TFLs will allow species to adapt to a warming environment.
Funding
NERC grant NE/P002692/1 to TP, AB, AH & RS. SNF P300PA_177830 to AM. NIHR HPRU EZI to SM.
Author Contributions
Conception: TP, AB, RS, AH, SP. Methodology & data collection: SP, BW, NW. Data curation: SP & SM. Analysis: SP, AM & SM. Original Draft: SP, TP, AB, RS & AH. Review & Editing: All authors.
Data and materials availability
All data and analysis R code will be deposited on Dryad upon acceptance of this manuscript.
List of Supplementary Materials
Materials and Methods
Tables S1-S4
Supplementary Text
References (23 - 36)
Materials and Methods
Animal maintenance
All species were kept in temperature-controlled rooms at their noted “rearing temperature” (see table S4) selected based on observations of when the laboratory populations were most stable. All stocks were kept at 12:12 L:D and ambient humidity. Stocks were maintained at moderate density (20 – 50 flies per 23ml vial or 50 – 100 flies per 300ml bottle culture). Cultures were tipped to fresh food every 7 days and a new generation was made every 4 – 6 weeks depending on the speed of a species development. Species were reared on one of four food types; A = ASG (10g agar, 85g Sucrose, 20g yeast extract, 60g maize, 1000ml H2O, 25ml 10% Nipagin), B = Banana (10g agar, 30g Yeast extract, 150g pulped banana, 50g Molasses, 30g Malt Extract, 25ml 10% Nipagin, 1000ml H2O), P = Propionic (10g Agar, 20g Yeast extract, 70g cornmeal, 10g soya flour, 80g Malt Extract, 22g Molasses, 14ml 10% nipagin, 6ml Propionic acid, 1000ml H2O), M = Malt (10g agar, 20g Yeast, 60g Semolina, 80g Malt, 14ml 10% Nipagin, 5ml Propionic acid, 1000ml H2O). To standardize across species, we only included males in our TFL assay that were sexually mature. This inevitably led to variation in the age of males (days since eclosion) between species based on published age-to-maturity (e.g. 34) and our own observations (Table S4 for details).
Species verification
Most Drosophila species were gained from stock centers, other research groups, or through field collections. Where species were not distinguishable through evident morphological traits, we verified position within the phylogeny by nucleotide sequence identity.
We extracted DNA from 2-3 adult male flies with DNeasy kits (Qiagen) following the manufacturer’s invertebrate protocol. We PCR amplified a portion of the mitochondrial universal barcode gene cytochrome oxidase subunit 1 using the primers C1-J-1718 (5’ – GGAGGATTTGGAAATTGATTAGT – 3’) and C1-N-2191 (5’ – CCCGGTAAAATTAAAATATAAACTTC – 3’) using HotStart Taq (Promega) with (5-minute initial heating, 30 cycles at 95°C for 30s, 56 for 30s, and 72°C for 30, with an final elongation step of 72°C for 120s). PCR products were visualized by SYBRSafe-stained gel electrophoresis and cleaned up using Exonuclease I and Shrimp Alkaline Phosphatase incubation per supplier protocol (BioLine). We then used BigDye based sequence reactions with both forward and reverse primers, followed by NaOH & ethanol precipitation clean-up and precipitation before sequences were analyzed with an ABi 3500XL Genetic Analyzer. Forward and reverse sequences for each species were aligned to derive a consensus sequence. For species with a putative ID based on stock center or collaborator expertise we assessed the clustering of these sequences with publicly available CO1 sequences from the same species available on the BOLD database (boldsystems.org). Clustering was performed with the MAFFT server tool, and clustering via a neighbor-joining tree was visualized, all based on default parameters. For species that were unknown (two repleta-like species collected from the field in Madeira and Morocco), we performed the same process as above, but included multiple sequences from multiple species in the repleta species group – initial BLAST of these sequences on the NCBI database had given an approximate idea of the species ID.
Measuring upper thermal limits
We assayed three metrics of upper thermal limits in males from 43 species of Drosophila: Lethal Temperature 80% (LT80), Thermal Fertility Limit 80% (TFL80) and Maximum Critical Temperature (CTmax). We chose to test for thermal fertility limits in only males because of the substantial evidence that male fertility is particularly vulnerable to high temperatures(18).
LT80 & TFL80
Newly eclosed virgin adult males were collected over a 48-hour period into Drosophila vials and allowed to sexually mature for 7 – 21 days depending on species (see Table S4). Males were kept in groups of maximum 10 individuals to minimize intrasexual aggression. Once mature, males were transferred to fresh 23ml fly vials with standard maize-sugar-yeast agar “ASG” medium’ food and allocated to 3D-printed floating plastic racks in pre-heated waterbaths (Grant TXF200) set to a range of temperatures (see Table S4 for temperature ranges). ASG food was used during heating for all species because preliminary work had shown that survival under heat stress was influenced by food type. Floating racks were weighted with ball-bearings at each corner to keep vials containing flies submerged so that the waterline was above the cotton-wool stopper. This ensured that flies were exposed to homogenous temperature conditions inside the vials. In each run of the experiment we monitored the temperature inside a fresh ASG vial without flies with a k-type thermocouple attached to a data logger (Pico Technology TC-08). Males were heated for 4 hours between ∼10am - ∼2pm and then returned to temperature-controlled rooms set to the species’ benign temperature. We scored the survival of males the next morning to account for any immediate recovery or delayed lethality of the temperature treatment. We scored flies as dead if they were completely immobile or had become stuck on their backs in the fly food – the latter would indicate that the flies were completely incapacitated by the treatment temperature which would likely be lethal in the wild.
Surviving males were then individually aspirated into separate, fresh vials containing their species’ designated food type and 3-4 sexually mature virgin females. Males remained in these vials at their benign temperature and allowed to mate freely for 6 days. This allows us to score any immediate lasting effect of temperature on fertility. After 6 days males were transferred to a second fresh vial with 1-2 more virgin females and allowed to mate for 24 hours. This allowed us to test for recovery of fertility or any delayed sterilizing effect of temperature. Females in both sets of vials were left to lay eggs for 3-5 days (depending on fecundity of the species) after the male had been removed. Fertility was scored qualitatively as the presence or absence of any offspring: vials were routinely scored for the presence of larval flies by checking for the larvae themselves or their distinctive track marks in the food. Juveniles flies were allowed to develop through to adult eclosion before being frozen and the number of adult offspring counted.
In total, 14742 males were exposed to heat treatment, of which 10925 survived and 9064 went on to be tested for heat-induced sterility. The average sample size at each temperature per species was 36.5 ± 0.5sem, ranging from 10 – 88 individuals.
Ctmax
The measures of upper tolerance described above are performed under static heat-stress conditions – this is necessary for estimating TFLs because the phenotype “fertility” can only be observed after the heat-stress treatment, so matching a male’s fertility to a stress temperature requires that he only experiences a single stressful environment. LT80 is therefore a directly comparable measure to TFL80. However, several studies in Drosophila and other animals have used estimates of temperature tolerance measured under gradually ramping heat conditions to estimate species response to climate change (4, 7, 23, 24). Because of this, and because previous work has shown that estimate of upper critical thermal limits can depend on methodological context(25), we also measured upper critical limits of our 43 Drosophila species under ramping heat conditions – critical thermal maxima (CTmax). This allows us to compare how static and dynamic measures of critical heat tolerance predict species’ distributions. It also allows us to compare upper TFLs to two widely used measures of upper critical limits.
We measure CTmax as the mean temperature at which male flies lose coordinated movement and are unable to right themselves. Individual males were collected as virgin adults from fly stocks and allowed to sexually mature (see Table S4). Flies were then anesthetized with CO2 and isolated into individual glass sample tubes sealed with a rubber stopper. Sample tubes were left at ambient room temperature for at least 30 minutes to allow males to fully recover from CO2 anesthetization – preliminary trials found no difference in CTmax between anesthetized and non-anesthetized male D. melanogaster. Sample tubes then were attached to a custom-built plastic rack and held at a 45° angle with elastic banding to facilitate the experimenter identifying when flies collapsed. The rack and vials were submerged in a glass aquarium held at the rearing temperature for the given species by a water circulator (Grant TXF200) (18, 23 or 25°C – Table S4). The temperature was then increased at a rate of 0.1°C per minute until all flies were incapacitated. The temperature at which flies collapsed for 30 seconds and did not right themselves after being physically disturbed by the experimenter (tapping the vial) was scored. The temperature of the waterbath was continuously recorded throughout the assay with a data logger (Elitech RC-61). The mean sample size for each species for CTmax assays was 19.53 ± 1.77.
Repeatability
Due to the logistical constraints in the number of flies that could be processed at a time and the differences in temperature range required to stress each species it was impossible to assay upper limits across all 43 species simultaneously. Further, inter-species variation in development time prohibited us from synchronizing all 43 species to run the assay in a completely randomized block design. Instead, we assayed each species separately to independently estimate CTmax, TFL80 and LT80. These assays were run between June 2018 – November 2019.
To test the repeatability of our measure of Lethal Temperature (LT80) we ran a single control block in February 2020 with 22 species that we could obtain enough virgin males that would reach sexual maturity in synchrony. We simultaneously exposed these species to a range of 6 temperatures each that covered their estimated LT80 identified when we had previously run them individually. We then tested the strength of the correlation between our original LT80 estimates and those given by this replicate block and determined if any given species showed significantly different estimates based on whether the confidence intervals of the two runs overlapped. Sample sizes for this control run ranged from 3 – 10 males per temperature per species, and involved 1152 male flies in total.
As a test of repeatability for our TFL80 estimates we ran two independent TFL assays on Drosophila virilis. This species was selected because it had demonstrated a clear difference between TFL80 and LT80 limits. The two repeat blocks were run 6 months apart and data were collected by two separate researchers. We compare the estimated TFL80 from these independent runs and consider differences to be significant if the 95% confidence intervals of our estimates do not overlap.
Repeatability of CTmax was estimated by randomly allocating species from the same rearing temperatures into blocks of 3 - 5 species and measuring their mean CTmax as described above. We compared these mean Ctmax values with those obtained when species were run individually when we collected the CTmax data used in analyses.
Point estimates of TFL80 & LT80
We chose 80% thresholds as our physiological limits for both LT and TFL because i) we wanted to compare viability and fertility measures on comparable scales, and ii) a loss of 80% of males is likely to represent a substantial threat to population stability for most species.
We found that general linear models fit our qualitative (1/0 data) fertility and survival data poorly, and overestimated 80% fertility loss thresholds considerably. Instead, we generated point estimates of 80% thresholds with dose response models (’drm()’), implemented in the ‘drc’ package in R. We used 3-parameter versions of the log-logistic model which fixes the lower limit of the predicted curve to 0 to reflect that we expect fertility and survival to eventually reach 0 over an infinite range of temperatures. The models then estimate the point at which we see an 80% reduction in survival or fertility relative to the upper limit present in the data. In this way, these models allow for a given species to inherently have some degree of sterility even at benign temperatures, and to estimate the temperature required to reduce this by 80%. Percentage fertility was analyzed for surviving flies only, i.e males killed by the heat treatment were not considered to be sterile.
In analyses in which we predict geographic range by physiological limits we use TFL80 estimates that are ‘significantly’ lower than the species’ LT80 estimate – for the remaining species TFL80 point estimates are defaulted to the LT80 to be conservative. We considered the difference between TFL80 and LT80 to be ‘significant’ if the 95% confidence intervals of the two point-estimates do not overlap.
Correlations between upper thermal limits
We used phylogenetically controlled linear models to test how much variation in TFL80 is explained by variation in LT80. These measures are taken from the same assays and are often very similar, so this correlation gives a quantitative assessment of the mismatch between these two thermal tolerance traits.
CTmax and LT80 inherently use different heating regimes to capture upper critical limits. Most notably, CTmax includes an element of temperature ramping, and so may allow species to plastically harden to high temperature stress. To test how well these two measures capture the same component of thermal interspecific variation, and are therefore proxies for each other, we used phylogenetically controlled linear models to explore the correlation between TFL80 and CTmax. Whilst we did not expect these methods to produce identical point estimates of lethal temperatures for a given species, we did hypothesize that these correlations will have a high R2 and a slope estimate close to 1.
If there is a universal upper limit to fertility across Drosophila then we would expect the difference between TFL and critical temperatures to be greater in species with higher temperature LT80 and CTmax. To test this, we used phylogenetically controlled linear models to test for significant correlations between ΔLimits (the difference between TFL80 and LT80) and both LT80 and CTmax.
Accounting for phylogeny
All of the species we used in this study are present in a phylogeny published by Patrik Röhner and colleagues(26): We use this phylogenetic tree as the basis of all our phylogenetically controlled analyses.
We tested for phylogenetic signal in all three upper thermal limits (LT80, TFL80 and CTmax). We used ‘pgls()’ in ‘caper’ to estimate if Pagel’s lambda was significantly different from 0 (no phylogenetic signal) or 1 (complete Brownian motion). The maximum likelihood method was used to estimate lambda.
When analyzing correlations amoung traits, and between traits and geographic distributions we corrected for phylogenetic signal in model residuals with ‘pgls()’ in the R package ‘caper’. Even where phylogenetic signal was estimated to be not significantly different to 0, we retained the correction in the model because these models give qualitatively identical results to uncorrected linear models using ‘lm()’.
Environmental variables
Geographic distribution data for each species were obtained manually from TaxoDros.uzh.ch as coordinates and location names. These were systematically cleaned by deleting any entry set to zero decimal places in both coordinates. Then, they were globally cleaned by removing locations with vague names such as entire countries or US states. Plots of each species location on a map were then visually inspected, and outliers were investigated and removed or corrected when appropriate.
The cleaned dataset of geolocations was then integrated with rasterized bioclimate data from the WordlClim V2.0 database (27). Our intention was not to fully reassess multiple climate variables as predictors for Drosophila distributions, because this has received extensive attention in the existing literature(4, 24). Instead, we tested if TFLs correlate with variables previously identified as important for predicting physiological limits and distribution in Drosophila; the maximum temperature reached in the hottest summer month (WorldClim Bio5) and annual precipitation (WorldClim Bio12). Means of each climate variable for each species were calculated across all global locations. We also calculated the mean of each variable independently for locations in the North and South hemispheres for species whose range straddles the equator.
Comparing upper thermal limits as predictors of species’ thermal environment
We used three distinct modelling approaches to test which measure of upper thermal limits best predicts species’ thermal range:
Firstly, we fitted separate phylogenetically controlled linear models with either TFL80, LT80 or CTmax as independent predictors, and species’ Tmax (WorldClim Bio5) as a response variable. In all three models we included mean annual precipitation (WorldClim Bio12 or “Pann”) as an interacting covariate, because previous work has found that the power of CTmax to predict species’ thermal environments degrades in more humid habitats(4). In these models all predictors are centered and scaled. We then used AIC-based model selection to reduce models to minimum-adequate models, and compared the slope estimate, model structure and R2 of the competing models.
Secondly, we directly replicated the method of (4) in which the authors split species into those that experience greater-than and less-than 1000mm of rainfall (“wet” and “dry” hereafter). The authors of (4) found that CTmax correlated well for “dry” species but not for “wet” species. We ran phylogenetically corrected linear models with Tmax as a response and either CTmax, LT80 or TFL80 and predictors for both ‘dry’ and ‘wet’ species. We qualitatively compared the R2 of these models as indicators of model fit.
Finally, we included both CTmax and TFL80 and their interaction with ‘Pann’ into a single phylogenetically controlled model of Tmax. We did not include LT80 in this analysis because a priori Variance Inflation Factor tests (VIF) on non-phylogenetically controlled versions of this model indicated that LT80 correlated very strongly with both other measures. We reduced the maximal model through AIC-based stepwise simplification starting with higher order interaction terms to reach a minimum adequate model.
Calculating and modelling thermal safety margins
Thermal safety margins were calculated as the difference between a species’ physiological upper thermal limit and the mean maximum summertime temperature they are likely to experience across all of their known global distribution locations. As per(4), we calculated this in two ways: firstly by taking the mean Tmax for each species, to estimate safety margins for the ‘central’ populations of a species (termed “Central Safety Margin”). Secondly by taking the mean Tmax + 1 SD to account for populations of species that live at the upper thermal range of species’ distribution (the “distribution safety margin). Safety margins were calculated separately using either CTmax, TFL80 or LT80 as the physiological limit.
We also treated populations in the North and Southern hemispheres as equivalent to separate species, because previous work has found minor differences in safety margins in the North and South hemisphere, likely because of biases in sampling efforts(3).
Safety margins were modelled as phylogenetically controlled linear mixed models with ‘hemisphere’ (N or S) and ‘physiological limit’ (CTmax, TFL80 or LT80) as linear predictors, absolute latitude (degrees2) as a quadratic predictor, and species identity as a random intercept.
Predicting current and putative future distributions of D. flavomontana
As a case-study demonstration of the importance of TFLs, we use MaxEnt v3.4.1 (28) to model potential distributions of D. flavomontana. Occurrence data were obtained from TaxoDros.uzh.ch and points within 25km of each other were clustered together to avoid oversampling (N = 24). First, we chose climatic and other variables to predict a potential distribution of D. flavomontana that is independent of the upper thermal limit. After excluding variables that strongly correlated or had little to no impact on the model results (<3% contribution), we used four of the tested 15 variables: mean winter temperature (WC Bio 11, from the WorldClim2.0 database, precipitation seasonality (WC Bio 15), an agricultural index (29), and elevation (30). We then ran 10 cross-validations with maximum iterations of 5000, using 10000 random pseudo-absences. To distinguish suitable from unsuitable regions, we used the threshold that maximized Youden’s index (maximum training sensitivity plus specificity) (31). The MaxEnt mean output grid is shown in Figure S5.
We then constrained the suitable area by the upper physiological thermal limits measured above: i) the lethal temperature measured as LT80 (35.8°C), and ii) the TFL80 (31.4°C). We use the maximum weekly temperature of the warmest month as reference (WC Bio5). Using WC Bio 05 directly as a predictor variable for D. flavomontana had close to no effect on the distribution (0.1% contribution).
For future predictions, we chose two carbon emission scenarios: RCP4.5 (moderately optimistic) and RCP8.5 (pessimistic). We use precipitation and temperature predictions of the NorESM1-M model from the NEX-NASA GDDP data set (32), as this model produces a median trend in terms of temperature increase(33). The agricultural index and elevation had to be used as present day data. Predicting and constraining the possible future distribution followed the same procedure as for the current distribution.
Supplementary text
Repeatability of knockdown CTmax assay
Correlations between TFL, LT and CTmax
The correlation between TFL80 and LT80 is significant (PGLS: F1,41 = 47.86, P <0.001, coefficient = 1.631) but LT80 only explains just over half of the variation in TFL80 (PGLS: adjR2 = 0.53,).
The correlation between absolute TFL80 and CTmax assays is significant (PGLS: F1,41 = 19.40, P <0.001) but considerably weaker than a perfect proxy (PGLS: R2 = 0.30, coefficient = 0.631). These analyses were phylogenetically controlled, with no significant phylogenetic signal found among model residuals (λ = 0.543ns).
The correlation between the two measures of upper lethal limits (LT80 and knockdown CTmax) is significant and positive (estimate = 0.668, F1,41 = 89.176, P <0.001) but also shallower than would be expected of exact proxies. However, this relationship explains 67.74% of the variation in LT80 across species. LT80 and knockdown CTmax are different measures of upper thermal limits and so deviation from parity is not unexpected. Notably, knockdown CTmax inherently contains an element of thermal ramping, whilst LT80 employs a single static temperature. These analyses were phylogenetically controlled, with a significant phylogenetic signal found among model residuals (λ = 0.570**).
The difference between TFL80 and LT80 (ΔLimits) shows no significant relationship with either CTmax or LT80 (PGLS; knockdown CTmax: coefficient = 0.042, F1,41 = 0.127, P = 0.72, R2 = -0.02; LT80: coefficient = -0.015, F1,41 = 0.011, P = 0.92, R2 = 0.02). This demonstrates that critical limits cannot be used as a direct proxy for TFLs, and so TFLs are capturing biological variation that measurements of critical limits do not. This also suggests that there is not simply an unavoidable upper limit to fertility shared among species with differing upper critical limits.
Phylogenetic signal in upper thermal limits
Previous empirical work has found that absolute upper critical limits show strong phylogenetic signal(37). This suggests that these limits may be difficult to rapidly evolve in response to changes in the environment and suggests that species are sorted into their environment by their evolutionary history, rather than locally adapted to it(4). We find similar degrees of phylogenetic signal in our measures of critical upper limits (LT80 λ = 0.94 P0 <0.001, P1 = 0.108; κ = 0.866, P = 0.001), (CTmax λ = 0.853, P0 <0.001, P1 < 0.001; κ = 0.51, P < 0.001). We also find phylogenetic signal in our estimate of TFL80, however it is less strong than the signal for either upper lethal limit (λ = 0.722, P0 = 0.01, P1 = 0.004; κ = 0.32, P = 0.005). The difference between LT80 and TFL80 (ΔLimits hereafter) are explained by evolutionary relatedness between species (λ = 0.823, P0 = < 0.001, P1 = 0.024, κ = 0.857, P = 0.001), but this is primarily because the majority of species in the Sophophora subgenus have a ΔLimit of 0°C. Whether this means that species are able to rapidly evolve TFLs within the range of their ΔLimits is an open question.
Acknowledgements
Natasha Mannion and Dr Angela Sims for assistance with experiments, Dr Ben Longdon and Dr Katherine Roberts for flies, Rowan Connell for designing 3D-printed equipment.