Heat dissipation in subterranean rodents: the role of body region and social organisation

The relatively warm and very humid environment of burrows presents a challenge for thermoregulation of its mammalian inhabitants. It was found that African mole-rats dissipate body heat mainly through their venter, and social mole-rats dissipate more body heat compared to solitary species at lower temperatures. In addition, the pattern of the ventral surface temperature was suggested to be homogeneous in social mole-rats compared to a heterogeneous pattern in solitary mole-rats. To investigate this for subterranean rodents generally, we measured the surface temperatures of seven species with different degrees of sociality, phylogeny, and climate using infrared thermography. In all species, heat dissipation occurred mainly through the venter and the feet. Whereas the feet dissipated body heat at higher ambient temperatures and conserved it at lower ambient temperatures, the ventral surface temperature was relatively high in all temperatures indicating that heat dissipation to the environment through this body region is regulated mainly by behavioural means. Solitary species dissipated less heat through their dorsum than social species, and a tendency for this pattern was observed for the venter. The pattern of heterogeneity of surface temperature through the venter was not related to sociality of the various species. Our results demonstrate a general pattern of body heat exchange through the three studied body regions in subterranean rodents. Besides, isolated individuals of social species are less able to defend themselves against low ambient temperatures, which may handicap them if staying alone for a longer period, such as during and after dispersal events.

avoid overheating 22,29 . Although subterranean mammals live in a thermally relatively stable environment whose overall T a does not decrease much below the thermoneutral zone (TNZ), most of them have dense fur that well conserve body heat [30][31][32][33] . This may complicate heat offloading at higher T a s or after digging. Finally, these mammals have reduced or even lack most body extremities such as tails and ear pinnae used for heat dissipation in non-fossorial mammals.
From this point of view it has for a long time been an enigma as to how subterranean mammals avoid overheating, especially during highly heat-producing activities. It was speculated that the excess metabolic heat is dissipated mainly through the venter in the Talas tuco-tuco Ctenomys talarum 34 . This heat dissipation pathway was later supported by the analysis of surface temperatures (T s ), i.e. temperature of the fur surface facing to the environment, in five African mole-rats (Bathyergidae, Rodentia) 31,35 . Both studies demonstrated that the feet which are generally less haired compared to other body regions in all bathyergids also play a relevant role in heat dissipation. It was also shown that two mole-rat species, the giant mole-rat Fukomys mechowii and the silvery mole-rat Heliophobius argenteocinereus, cool themselves very effectively, while burrowing via contact of their body with the colder surrounding soil, especially when the soil is wet 36 . Although interspecific differences in heat dissipation in mole-rats could be attributed mainly to the climate of the habitats 35 , remarkable differences were found, surprisingly, also between species from the same climatic conditions and habitats. Thus, social F. mechowii has a shorter and sparser fur allowing more effective body heat dissipation and thus higher T s, compared to the solitary H. argenteocinereus which has longer and denser fur across most of its body 31 . Higher heat dissipation for social species is disadvantageous at lower T a s, but is reduced by huddling 37,38 . Both species also differed in the pattern of T s on the venter. Whereas F. mechowii had a uniformly high T s distributed across its venter, H. argenteocinereus had a smaller area of high T s on the chest, and the rest of the ventral surface was remarkably colder making the venter heterogeneous in terms of T s 31 . However, interpretations regarding the heterogeneity of thermal windows related to differences in social organisation are preliminary because only one solitary and one social species have been compared so far (cf. 31 ).
Apart from fur, heat dissipation through a particular body region could be influenced by the characteristics of skin and fat tissue as exhibited in aquatic mammals 39 , or small terrestrial mammals 40 . Nevertheless, to be able to move easily through burrows, mole-rats and probably also other subterranean mammals do not store fat as a subcutaneous layer 41,42 . A recent analysis did not show any remarkable differences in parameters of skin and fat tissue between the dorsum and venter in F. mechowii which supports an exclusive role of fur in heat dissipation through the integument 43 . Different fur characteristics can thus have relevant behavioural and ecological consequences in subterranean rodents. For example, it has been suggested that well insulating fur allows H. argenteocinereus to colonise the highest altitudes of the Nyika Plateau in Malawi, whereas the sympatric social Whyte's mole-rat Fukomys whytei occurs at lower altitudes 44 . We may predict that individuals of social species, when dispersing singly, have higher thermoregulatory costs due to a less insulating fur because they cannot decrease heat losses by huddling, one of the most efficient energy saving mechanisms in small social mammals 20 . Although it is very difficult to obtain data on the duration of staying alone in social subterranean species in nature, several records indicate solitary individuals of social mole-rats stay for weeks, months or even years on their own [45][46][47][48] .
In the current study, we recorded T s using infrared thermography (IRT), and core body temperature (T b ) to analyse heat dissipation ability, the presence of thermal windows, and the role of social organisation in seven species of subterranean rodents over a T a gradient from 10 to 35 °C. The studied species originated from different phylogenetic lineages (families Bathyergidae, Octodontidae, Spalacidae), climatic zones, and continents with representatives from both social and solitary species (for details see Table 1). As T b may differ among the tested species, we introduce T diff parameter defined as the difference between T b and T s of a particular body region (the dorsum, venter, and feet) at each T a to quantify heat dissipation. Higher values of this parameter indicate lower heat dissipation from the respective body region to its surrounding. Since T diff is determined mainly by the value of T s , while T b changes much less, we did not test T s in most analyses to avoid redundant results similar to those of T diff . We focused mainly on both extreme T a s (10 and 35 °C), and a T a within TNZ of all species (30 °C). We hypothesized that: (a) Smaller T diff will be found on the venter compared with the dorsum and feet at lower T a s confirming the existence of the ventral heat dissipation surface in all species. At these low T a s, larger T diff on the feet and dorsum is a consequence of vasoconstriction in the feet, and better fur insulation of the dorsum. Furthermore, T diff will be smaller in all three body regions at the highest T a s when body heat needs to be dissipated at the highest possible rate from all body surfaces to avoid overheating. (b) T diff will be generally smaller on the venter and dorsum in social compared to solitary species indicating easier heat dissipation through these regions in social species. No differences in T diff are expected for the feet because they are less haired in both solitary and social species. (c) The pattern of T diff along the ventral body region will be homogeneous in social species, and heterogeneous in solitary species. The heterogeneity in T diff in solitary species will be caused by smaller T diff on the less furred chest compared to higher T diff of other areas of the venter. Next, the venter will be homogeneous in T diff in all species at highest T a s, when body heat needs to be dissipated by the whole ventral surface to avoid overheating.

Material and methods
Tested animals. Altogether 73 individuals from seven species of subterranean rodents differing in body mass, phylogenetic relatedness, and sociality were studied (Table 1). All animals were adult non-breeders, or their breeding history was unknown in solitary species, but none of them showed signs of recent breeding, which may theoretically influence measured parameters. For the purpose of this study, we used the following taxa.  52 . Further information about the species including number of individuals used in the study, their physiology and ecology is shown in Table 1. All experiments were done on captive animals. Georychus capensis, C. hottentotus, and B. suillus, were captured about four months before the experiment, and kept in the animal facility at the University of Pretoria, South Africa (temperature: 23 °C; humidity: 40-60%, photoperiod: 12L:12D). The animals were housed in plastic boxes with wood shavings used as a bedding. Cryptomys hottentotus and G. capensis were fed with sweet potatoes; B. suillus with sweet potatoes, carrots, and fresh grass. Fukomys anselli, F. "Nsanje", N. galili, and S. cyanus were kept for at least three years in captivity (or born in captivity) before the experiment in the animal facility at the University of South Bohemia in České Budějovice, Czech Republic (temperature: African mole-rats 25 °C, N. galili and S. cyanus 23 °C; humidity: 40-50%, photoperiod: 12L:12D). The animals were kept in terraria with peat as a substrate and fed with carrots, potatoes, sweet potatoes, beetroot, apple, and rodent dry food mix ad libitum.
Experimental design. We measured T b and T s in all species at six T a s (10, 15, 20, 25, 30 and 35 °C). Each individual of all species was measured only once in each T a . Measurements were conducted in temperature controlled experimental rooms in České Budějovice and Pretoria. Each animal was tested on two experimental days.
The animals were placed in the experimental room individually in plastic buckets with wood shavings as bedding. On the first day, the experimental procedure started at T a 25 °C. They spent 60 min of initial habituation in the first T a after which T b and T s were measured as described in the following paragraphs. The T a was then increased to 30 °C and 35 °C, respectively. After the experimental room reached the focal T a , the animals were left minimally 30 min in each T a to acclimate, and the measurements were repeated. Considering their relatively small body size, tested animals were very likely in thermal equilibrium after this period because mammals of a Table 1. Characteristics of seven studied species of subterranean rodents and their habitats. Means ± SD of body masses are given. Numbers in parentheses after the species names represent sample sizes. TNZ, thermoneutral zone; AP, annual precipitation; AMT, annual mean temperature; MT, minimal temperature; TAR, temperature annual range (climatic parameters were downloaded from the Worldclim database 49 ; due to the unknown locality of S. cyanus, climatic data from the whole species range were used). a J. Okrouhlík www.nature.com/scientificreports/ comparable body mass are thermally equilibrated after similar period of acclimation [53][54][55][56] . On the second day, the procedure was repeated with the initial T a 20 °C and decreasing to 15 °C and 10 °C, respectively. The time span between the measurements of the same individual in different T a was at least 150 min. Between experimental days, the animals were kept at 25 °C in the experimental room (individuals of social species were housed together with their family members).
Body temperature measurements. We used two sets of equipment to measure animal T b and T s . In B.
suillus, G. capensis, and C. hottentotus, T b was measured by intraperitoneally injected PIT tags (< 1 g, LifeChip with Bio-Thermo Technology; Destron Fearing Corp., Dallas, Texas, USA, accuracy 0.5 °C, resolution 0.1 °C). A vet injected the tags under anaesthesia (Isoflurane) three months before the experiment. The tags were calibrated by the manufacturer and were read using a Global Pocker Reader EX (Destron Fearing Corp., Dallas, Texas, USA). T s was measured by FLIR SC325 thermal camera (FLIR Systems, Inc., Wilsonwille, Oregon, USA; sensitivity < 50 mK, accuracy ± 2%, frame rate 31 fps, calibrated by the manufacturer). Core body temperatures of F. "Nsanje", F. anselli, N. galili, and S. cyanus were measured using a RET-2 temperature probe (Physitemp Instruments LLC., Clifton, New Jersey, USA; tip diameter 3.2 mm, inserted at least 2 cm in the rectum, accuracy 0.1 °C) connected to Thermalert TH-8 (Physitemp Instruments LLC., Clifton, New Jersey, USA; resolution 0.1 °C). This procedure took less than 30 s. The apparatus was verified against a thermometer (EL-USB-2-LCD+; Lascar Electronics Ltd., Salisbury, UK; overall accuracy 0.45 °C) calibrated by an accredited laboratory. Both means of measuring T b have been shown to provide almost identical results in small mammals including one species belonging to African mole-rats [57][58][59] . Surface temperature was measured using a Workswell WIRIS thermal imaging system (ver. 1.5.0, Workswell s.r.o., Praha, Czech Republic, sensitivity 30 mK, accuracy ± 2%, frame rate 9 fps, calibrated by the manufacturer). Both thermal cameras were calibrated prior to measurement of each animal, and we used the same software to process the raw thermograms (see below).
To measure T s , a focused radiometric video of the animal was taken with its different body parts exposed perpendicularly to the camera. More specifically, the animals were held hanging by the loose skin around the tail, and their dorsal and ventral body regions were sequentially exposed to the camera. This procedure took less than 30 s. To ensure unbiased T s measurements, all fans in the experimental room were switched off during measurements, and a non-uniformity correction (type of calibration of the camera) was performed just before measuring each animal. Since the seven tested species were of different sizes, and the lenses of the two thermal cameras had different focal lengths, and thus field of views, the animals were filmed at different distances from the camera to fully utilise the resolution of the two cameras. Distances of animals to the camera lens were 38-42 cm for N. galili, F. anselli, F. "Nsanje", and S. cyanus, 53-57 cm for C. hottentotus, 64-68 cm for G. capensis, and 86-92 cm for B. suillus. Room air temperature and humidity were monitored throughout the trials (EL-USB-2-LCD+; Lascar Electronics Ltd., Salisbury, UK; calibrated by a certified laboratory). Humidity in both labs ranged from 47 to 72% during all experiments.
To assess the heat dissipation (which is related mainly to insulative properties of the integument) of each species regardless of their actual value of T b , we introduced T diff parameter defined as the difference between T b and T s of a particular body region of each individual (the dorsum, the venter, and the feet) at each T a .
Handling stress may influence T b and may also evoke vasoconstriction on the periphery and thus potentially affect T s 58 . To test the potential influence of handling procedures employed in our study on T s and T b , we carried out several simple experiments. Firstly, we measured T b of three individuals of F. mechowii (species not included in the present study, but in 31 ) by intraperitoneal probes (G2-HR E-Mitter, Starr Life Sciences Corp., Oakmont, Pennsylvania, USA; the only species with these probes in our breeds), and found no significant differences in T b prior to or after the 30 s handling period (T b was 34.0 ± 0.6 °C and 34.1 ± 0.7 °C before and after handling, respectively; paired t-test: t = −0.42, df = 3, p = 0.70). Secondly, to exclude the possibility of a long-term effect of repeated handling on T b , we simulated the measurement procedure with another three individuals of the same species with intraperitoneally injected PIT tags obtaining T b s without direct contact. We placed mole-rats singly in a bucket and after 150 min we measured their T b . Subsequently, we lifted them for a period of one min and returned them into the bucket. After 150 min, we again measured their T b s. Lifting of the mole-rats and temperature measurement was repeated once again (Note that 150 min is the minimal period between two manipulations of each individual in our experiment). The T b s did not change during these three subsequent measurements (Generalized Least Squares model [GLS]: F = 0.7, p = 0.544; T b values were 33.14 ± 0.6, 33.13 ± 0.77 and 33.23 ± 0.92 °C). Thirdly, to rule out a possible effect of handling on T s , we measured dorsal T s in ten individuals of F. "Nsanje" (species included in the present study) before and after the handling period which took usually less than 30 s. Similarly, we did not find significant differences (T s was 30.5 ± 1.1 °C and 30.3 ± 1.1 °C before and after the 30 s handling, respectively; paired t-test: t = 1.78, df = 9, p = 0.11). It should be noted that all animals are accustomed to this handling, as they undertake it on a weekly basis during routine activities, such as weighing, cleaning of terraria, and miscellaneous behavioural and physiological experiments. We therefore suggest that the different approaches in obtaining T b , and different time in captivity for experimental animals did not affect our results substantially.
Thermogram processing and analysis. The thermographic camera produces (sequences of) images in the infrared range, so-called thermograms. Sharp and high contrast radiometric thermograms of the dorsum, the venter and one hind foot as a representative of the feet, were captured from the raw radiometric video in the software CorePlayer (ver. 1.7.70.320; Workswell s.r.o., Praha, Czech Republic; https ://works well-therm al-camer a.com/works wellc orepl ayer), and processed into non-radiometric thermograms by specifying temperature calculation parameters in the software ThermoFormat (Workswell s.r.o., Praha, Czech Republic; https ://works welltherm al-camer a.com/works well-therm oform at). These parameters were entered as follows: fur emittance was www.nature.com/scientificreports/ set to 0.97 (e.g. 60 ); air humidity, air temperature, and reflected temperature were entered as means of experimental room humidity and air temperature during the course of T s measurement of each species at given T a ; distance was set as the distance of camera lens from the animal. Although Šumbera et al. 31 identified a few other body surface areas through which body heat is mainly passively dissipated in mole-rats (peripalpebral, nose, and ear area), we did not include them in our study due to their relatively small area, and thus very low contribution to overall heat dissipation. Processed thermograms were used to infer mean T s of the whole ventral and dorsal body region, and the feet necessary for the calculation of T diff . The analysed body regions were manually marked out with a polygonal region of interest in the CorePlayer (see Fig. 1 as an example for polygon of the whole venter). The region of interest was adjusted to fit onto the part of the body region of the particular individual perpendicularly oriented to the camera lens. Additionally, the venter was divided in the anterior-posterior axis into five areas ( Fig. 1: 1-between the front legs, 2-to the widest part of the chest, 3-to the end of the rib cage, 4-to the hips, and 5-between hind legs without the anogenital area).

Data analyses.
In all analyses performed in this study, T diff was used as a dependent variable assuming Gaussian error distribution. In all models described below, body region (the dorsum, the venter, and feet), social organisation (social and solitary), and ventral area (1-5) were treated as explanatory categorical variables, whereas T a was treated as a covariate. Means ± SD are given throughout the text. The sex of the tested animals was not included as an explanatory variable, mainly because of a relatively low number of males and females tested. Nevertheless, we did not expect any differences because there are no remarkable sex differences in thermal biology of subterranean rodents apart from body mass if females are not pregnant or lactating pups 25 . To test our hypotheses, we fitted the models on subsets of data divided according to body region and selected T a values. All analyses were carried out in R 61 and all figures were edited using the program Inkscape 0.92 (https ://inksc ape. org/). www.nature.com/scientificreports/ T b , T s , and T diff over the T a gradient. Prior to testing of our hypotheses, we performed the following basic statistical exploratory analyses of T b and T s . For each species, we firstly calculated a piecewise linear regression implemented in the R package Segmented 62 to assess whether there is a breakpoint T a , at which the regression curve characterising the change in T b (expressed as mean values for each T a for each species) for a given species changes its slope. Secondly, we calculated a linear regression to characterise a change in T s (expressed as mean values for each T a for each species) in response to increasing T a . We used Bonferroni procedure to correct the α of the tests since we calculated one piecewise linear regression for T b for each of the seven species (adjusted α = 0.0071) and 21 linear regressions for T s (adjusted α = 0.0024).
For each species separately, we tested whether T diff differs amongst the three body regions, i.e. the dorsum, the venter, and the feet, using GLS marginal models implemented in the nlme package 63 . One GLS model was calculated for each of three T a s − 10 and 35 °C as extremes in order to obtain information about T diff in cold and hot conditions, respectively, and 30 °C which represents TNZ of all the studied species (Table 1). In all GLS models, the body region was the explanatory variable, and the identity of tested individuals was included to avoid pseudo-replication. We used the Bonferroni procedure to correct the significance level of the tests because we tested each T a separately for seven species and three T a s (adjusted α = 0.0024), and then performed a similar post-hoc comparison 13-times (adjusted α = 0.0038).
For each species separately, we first calculated three linear regressions to test whether the slope characterising the change of T diff in response to increasing T a for each of the three body regions is different from zero. For each body region, T diff value for a given T a was calculated as the mean obtained from all individuals within a given species, for which we obtained a complete set of T s values from all three body regions at all six T a s. Afterwards, a homogeneity-of-slopes model was calculated to test whether these three regression lines are parallel to each other, i.e. there is no interaction between body region and T a . The function "emtrends" implemented in the R package emmeans 64 was used to calculate post-hoc comparisons to find differences between the slopes for the dorsum, the venter, and the feet in these models. We used the Bonferroni procedure to correct the significance level of the tests we performed for each of the seven species: (a) three similar regression models (adjusted α = 0.0024) and (b) altogether seven homogeneity-of-slope models with a post-hoc test (adjusted α = 0.0071).

T diff of the three body regions in solitary and social species.
To assess the effect of the social organization on T diff (expressed as a mean value for each T a for each species) in each of the three body regions, we fitted Generalized Linear Mixed Models in a Bayesian framework using Markov Chain Monte Carlo sampling algorithm (MCMC) implemented in the R package MCMCglmm 65 . This approach was used to account for nonindependence in T diff measurements arising due to shared phylogenetic ancestry, and therefore a random effect of phylogeny was included in all models. We controlled for phylogenetic relatedness using the subset tree of the mammal phylogeny by Upham et al. 66 . The set of subtrees was retrieved via an online tool: vertlife.org/phylosubsets/. The maximum credibility tree was then inferred using the R package phangorn 67 , and taxa labels were changed using the R package phytools 68 in order to merge the taxa with the comparative dataset.
Parameter estimates for fixed and random effects of each model were obtained from sampling posterior distributions by running 2,500,000 MCMC iterations with a burn-in period of 20% and a thinning interval sampling each 1000th iteration. Estimates of each parameter are thus based on 2000 samples from a posterior distribution. In all models, we used weakly informative priors for variance components (V = 1, ν = 0.02), and a default prior specification for fixed effects, i.e. mean centred on zero with a very large variance (µ = 0, V = 10 8 ), to let the posterior be determined mostly by the information in the data. Mixing and convergence of MCMC chains were assessed by visual inspection of both time-series and density plots, as well as by calculating autocorrelation among successive MCMC samples. There was no apparent trend among MCMC samples in time-series plots, and autocorrelation was low in each fitted model. Effective sample size was ~ 2000 for all estimated parameters in all models, suggesting generally good mixing and convergence properties of MCMC chains.
For each of the three body regions separately, we fitted a model of the formula: T diff ~ T a + social organisation + T a : social organisation (the same T diff datasets as for linear regressions were used for these models). Beside the random effect of the phylogeny accounting for evolutionary history of the studied taxa, we included also a random effect of species to account for species-specific effects on the variability in T diff in response to T a . This model allowed us to test whether the slopes for the change in T diff as a response to increasing T a differ between solitary and social species. Interaction term was considered only when its effect in a model was significant (see highest posterior density (HPD) intervals and/or pMCMC in Table 3), and the deviance information criterion (DIC) indicated a better fit (lower DIC) of the model with than without interaction. If the effect of interaction was not significant, we ran the model without the interaction term, and interpreted only the effects of T a and social organization, respectively.
T diff along the venter in solitary and social species. For each species individually, we tested whether T diff differs among the five ventral areas using GLS models. One GLS model was calculated for each T a of 10, 30 and 35 °C. In all GLS models, the ventral area was the explanatory variable (a factor with five levels: ventral area 1-ventral area 5, see Fig. 1). We used the Bonferroni procedure to correct the significance level of the tests since we tested each of three T a s separately for seven species (adjusted α = 0.0024). In all GLS models, an identity of tested individuals was included to avoid pseudo-replication.
In addition, for each of the three T a , we fitted one MCMCglmm of the form: T diff ~ ventral area + social organisation + ventral area: social organisation, with random effects of the phylogenetic relatedness and species. This model allowed us to test whether the pattern of mean change in T diff (calculated as the mean from all individuals of each species entering GLS analysis mentioned above) across five ventral areas differed between solitary and social species. The effect of interaction term was considered significant if the HPD intervals did not

Results
T b , T s , and T diff over the T a gradient. In all species apart from C. hottentotus and S. cyanus, T b was stable from 10 up to 25 °C (the slopes of segmented regressions were close to 0), showing breaking points between 26.6 and 28.8 °C, after which T b started to increase (see Table 2 for details). For C. hottentotus, T b increased across the whole T a range (the slope was significantly different from 0 for T b values before the breaking point), but there was a relatively steeper increase between 30 and 35 °C (breaking point at 29.3 ± 0.2 °C, Table 2). For S. cyanus, T b was stable only from 10 to 20 °C, with a breaking point at T a of 23.1 ± 1.1 °C. Surface temperature for the dorsum, the venter, and the feet increased in all species over the range of T a (Linear regressions, all p < 0.002 were significant after the Bonferroni procedure, Fig. 2 and Table 2). In contrast to T b , T s of all three regions showed markedly higher differences among the species, with the highest value for the venter at lower T a s, but these differences became smaller at 30 and 35 °C (Fig. 2).
For each of the seven species, T diff differed among the three body regions at 10 °C (p < 0.0001, Supplementary  Table S2). In N. galili, F. anselli, F. "Nsanje", and S. cyanus, T diff was the lowest for the venter, and the highest for the feet, with the dorsum having intermediate T diff significantly different from both (p < 0.001, for all pairwise comparisons see Supplementary Table S2). In B. suillus, G. capensis, and C. hottentotus, T diff was lower only for the venter than the dorsum (p < 0.001), and the feet did not differ from either of the two.
At 30 °C, T diff differed among the three body regions in N. galili, B. suillus, and F. anselli only (p < 0.0001, Supplementary Table S2). While both the venter and the feet had significantly lower T diff than the dorsum in N. galili (p < 0.001), T diff was lower for the feet than for both the venter and the dorsum in B. suillus (p < 0.001). In F. anselli, the ventral T diff was lower than the dorsal one, with T diff of the feet being the highest (p < 0.001).
At 35 °C, T diff differed among the three body regions only in N. galili, B. suillus, and G. capensis (p < 0.0001, Supplementary Table S2). The feet of N. galili had lower T diff than the venter and the dorsum, with the latter having the highest T diff (p < 0.001). In both B. suillus and G. capensis, the feet had lower T diff than both the dorsum and the venter (p < 0.001) which had similar T diff .
For each of the seven species, T diff decreased with increasing T a in all three body regions (see Table 3 for statistical details and Fig. 3). Among the three body regions, the ventral T diff showed the smallest change in response to increasing T a (i.e. a less steep slope), and it differed significantly from those of both the dorsum and the feet in five species. In S. cyanus, it differed from the feet only, and in C. hottentotus, no differences among all three respective slopes were found. Besides, the slopes for the dorsum were significantly steeper than slopes for the feet in F. anselli, F. "Nsanje", S. cyanus, and N. galili.
T diff of the three body regions in solitary and social species. The comparison among the slopes of regression lines characterising the response of T diff to increasing T a for the dorsum, the venter, and the feet differed between solitary and social species (Fig. 4). For the dorsum, T diff was significantly higher in solitary than in social species across the whole T a range (pMCMC < 0.0001), and it decreased relatively more in solitary (slope − 0.52) than in social species (slope − 0.39) for a given increase in T a (Fig. 4A and Table 4, DIC = 134.4 and 119.8 for models without and with the interaction between T a and social organisation, respectively). For the venter, the slopes characterising the change in T diff in response to increasing T a did not differ between solitary and social species (slope − 0.23, DIC = 93.4 and 95.6 for models without and with the interaction between T a and Table 2. Results of piecewise linear regressions characterising a change in core body temperature (T b ) and linear regressions characterising a change in the surface temperature (T s ) for three body regions in response to increasing ambient temperature (T a ) in seven subterranean rodent species. Breaking points and slopes are provided as mean ± SEM; p values for the slopes before the breaking point are in parentheses; statistically significant tests after the Bonferroni procedure α = 0.0071 for T b and α = 0.0024 for T s are marked with asterisks.

Species T a breaking point for T b (°C)
Slope before the breaking point of T b (p value) www.nature.com/scientificreports/ social organisation, respectively; pMCMC = 0.884), but there was a tendency (pMCMC = 0.079) for higher T diff in solitary compared with social species (Fig. 4B and Table 4). For the feet, T diff did not differ between solitary and social species across the T a range (slope − 0.59, Fig. 4C, and Table 4, DIC = 145.8 and 148.1 for models without and with the interaction between T a and social organisation, respectively; pMCMC = 0.902).  Table 6

Discussion
To the best of our knowledge, this study represents the highest number of endotherm species, where heat dissipation has been analysed within a single study using the IRT approach. In several aspects, this allows us to generalise our findings for subterranean rodents. We found that the main heat dissipation areas in all species are present at the venter and the feet. With respect to the effect of sociality, solitary species dissipate less heat through the dorsum than social species and the same tendency was found for the venter. For the feet, there were no differences between both groups. Finally, the pattern of heat dissipation along the venter was not consistent between species and could not be attributed to sociality. For the purpose of our study, we defined T diff as the difference between T b and T s in each individual and each T a measured. The value of this parameter is predominantly driven by T s which increases remarkably over the gradient of T a for all three body regions with T b being much more stable (Fig. 2). Applying this approach can be important especially in comparative studies using multiple species differing in T b and species-specific change of T b over the gradient of T a . This can be illustrated by the results of the segmented analysis when the increase of T b started at 23.1 °C as found in S. cyanus, but at 29.3 °C as observed for C. hottentotus (Table 2).
Our results demonstrate that the tested body regions play different roles in heat dissipation in subterranean rodents. As predicted (based on 31,34 ), at the coldest T a measured (10 °C) we found lower T diff on the venter compared to the dorsum in all studied species (and in F. anselli, F. "Nsanje", S. cyanus, N. galili also compared to the feet) showing relatively higher heat dissipation from this body region ( Fig. 3 and Supplementary Table S2). The warmer venter in lower T a s compared to other body surfaces was recently also found in three social mole-rat species 35 . This finding corresponds with the pattern found in some other mammals. A venter suitable for heat dissipation can be advantageous as its heat flow may be controlled behaviourally 12,21,31,69 . Although we did not quantify the behaviour of all individuals during the experiments, closing of a ventral thermal window by curling up into a ball was observed in most individuals at lower T a s (see Supplementary Fig. S2). Adopting this posture, animals not only minimise heat losses through their venter, but also reduce their effective surface-to-volume ratio bringing additional energetic savings 70 (note that the same posture is adopted also by torpid small nonvolant mammals 71 ). On the contrary, at high T a s (≥ 30 °C), tested animals usually exposed their venter to the environment mainly by lying on their back with the legs outstretched trying to maximize heat transfer to the surroundings (Supplementary Fig. S2). A large ventral thermal window should thus be advantageous, especially in subterranean mammals because they are predicted to lose the excessive metabolic heat during digging by means of conduction when the venter is in direct contact with the excavated soil and burrow floor 36 .
Our findings also confirm an important role of the feet in thermoregulation of subterranean rodents as has been also shown in two studies on African mole-rats 31,35 . In laboratory rats, the feet together with the distal parts of legs comprise around 10% of the total body surface and serve as an important heat dissipating region 72 . The feet of rodents adapted for digging and removing the soil are larger than in non-fossorial rodents 73,74 , and thus represent a significant proportion of the body surface. The importance of the feet in the thermoregulation of subterranean rodents is likely even higher because of lack of other body appendages, and due to their large surface-to-volume ratio. The feet are usually not well haired which, together with their bare pedal surface, makes them suitable for fast and effective heat transfer. Due to their relatively large size and heat convection via direct Table 3. Results of the linear regressions testing whether the slopes characterising a decrease in the change of the difference between core body and surface temperatures (T diff ) in response to increasing ambient temperatures (T a ) differ from zero for the dorsum (D), venter (V) and feet (F) and of the homogeneity-ofslope models (Body region × T a ) testing whether these slopes differ among the three body regions in seven subterranean rodent species. The degrees of freedom were 1 and 4 for the linear regressions and 2 and 12 for the homogeneity-of-slope models; the goodness of fit of the regression is presented by the adjusted R 2 ; the p value of post-hoc pairwise comparisons for the homogeneity-of-slope models are shown in the columns V vs. D, V vs. F, and D vs. F; statistically significant results after Bonferroni correction α = 0.0024 a or α = 0.0071 b applied are marked with an asterisk.

Results of linear regressions testing non-zero slopes
Results of homogeneity-of-slope models www.nature.com/scientificreports/ contact with the substrate, the feet are far more effective than small areas around the eyes or ears which were also found to lose heat in mole-rats 31 .

Dorsum (R 2 adj) a Venter (R 2 adj) a Feet (R 2 adj)
In this regard, our results indicate very effective vasoconstriction taking place in the feet to decrease heat losses at the lowest T a s. This is supported by the feet being the body region with the lowest T s , and thus highest T diff in four of seven tested species at T a of 10 °C (Fig. 2, 3, and Supplementary Table S2). The absence of significant differences between the feet and the other two body regions at this T a in B. suillus, G. capensis, C. hottentotus indicates a very effective insulative ability of their haired body surfaces rather than the absence of effective vasoconstriction in their feet. Importantly, at 35 °C, the feet T diff was significantly smaller than the T diff of other two body regions in three tested solitary species, N. galili, B. suillus, and G. capensis (Fig. 3, Supplementary  Table S2). This demonstrates intensive vasodilatation and dissipation of the excessive body heat which cannot be effectively dissipated by other body parts because of the presence of well insulating fur (based on data on fur characteristics for all three species, Vejmělka and Šumbera unpubl. data). Heat dissipation via the feet is likely very relevant, similarly to the venter, especially during digging because of their contact with soil. McGowan et al. 35 even suggested that the feet with reduced hair coverage are important at high T a s when other conductive heat losses are working at their maximal capacity. Since the feet contribute to increased heat dissipation at high T a s, but effectively decrease heat loss at low T a s, they serve as a typical body-extremity-located thermal window similarly to other mammals (e.g. 3-6 ). www.nature.com/scientificreports/ The importance of the feet in heat exchange was supported also by behavioural observations during the experiment. At low T a s, we observed that many tested animals were sitting on the hind feet with the front feet not in contact with the cold floor ( Supplementary Fig. S2). It was also documented that several individuals of H. argenteocinereus had some of their feet surprisingly warm while shivering at 10 °C 31 . This indicates that the feet may play an active role in dissipation of excessive heat produced by shivering as suggested by authors, or alternatively, occasional warming of the feet could be related to protection against cold damage by repeated short-term redirection of blood flow into peripheral colder body parts, as showed across diverse mammalian taxa including foxes, rats, or humans (e.g. 4,75,76 ).
We demonstrated that tested species and probably other subterranean rodents effectively combine heat dissipation through the venter and the feet, yet each of these body regions acts as a different type of thermal exchanger. The first type emits heat more or less continuously (as indicated by constantly relatively low T diff , and by its less steep slope over the gradient of T a compared to the other two body regions; see Fig. 3), and is represented by being less furred, and thus less thermally insulated venter 31 . When necessary in cold conditions, heat dissipation through this surface can be reduced behaviourally by changing the body posture (see above). The second type is represented by the feet with the steeper slope of the T diff change. In the cold temperatures, heat loss may be prevented effectively as for instance indicated in gerbils 77 . In this regard, the way in which these mechanisms, when acting simultaneously, contribute to heat dissipation is intriguing, and needs to be further explored in subterranean rodents and mammals in general.
Considering the effect of sociality on heat dissipation, Šumbera et al. 31 found that social F. mechowii had higher T s on both the venter and the dorsum, but not the feet compared with solitary H. argenteocinereus. Our results partially support the effect of sociality on heat dissipation in larger number of species. As expected, the www.nature.com/scientificreports/ role of the feet in heat dissipation was independent of social organization, as demonstrated by almost identical regression lines for solitary and social species (Fig. 4C). On the contrary, solitary species had significantly higher T diff on the dorsum across tested T a s, especially at lower temperatures (Fig. 4A). Solitary species thus conserve heat more effectively through this body region especially at lower T a s (cf. 31 ). Although the mean ventral T diff of solitary species was about 2 °C higher across experimental T a s (Fig. 4B), this difference only approached significance. This might be caused mainly by a large intraspecific variability of the ventral T diff among the four studied social species (compare Fig. 4A,B). Indeed, S. cyanus and C. hottentotus had markedly higher T diff than two Fukomys Table 4. Relationships between the difference between core body and surface temperature (T diff ) of the three body regions, ambient temperature (T a ), and social organisation (solitary vs. social) in seven species of subterranean rodents. 2.5% HPD and 97.5% HPD represent lower and upper highest posterior density intervals; pMCMC denotes to the p value obtained from MCMCglmm; T a × Social organisation = interaction between T a and social organisation.  Table 5. The results of GLS models comparing the difference between core body and surface temperature (T diff ) among the five ventral areas at ambient temperature (T a ) of 10, 30 and 35 °C in each of the seven subterranean rodent species. For each T a , the venter was tested separately; n denotes to the number of individuals tested for each T a ; statistically significant results after Bonferroni correction α = 0.0024 applied are marked with an asterisk; post-hoc comparisons among the areas are depicted by letter combinations in Supplementary Fig. S1. www.nature.com/scientificreports/ species at the lowest T a of 10 °C (see Fig. 3). We may speculate that such difference can indicate better ventral fur insulation in these two species, as they occupy harsher environmental conditions compared to tropical habitats of both Fukomys species (see Table 1). Further studies that will focus on the fur quality including more species from various environmental conditions could help to better understand this concept. Higher heat dissipation from the surface of social subterranean species is linked with lower insulative properties of their fur 31,33 . Dense heat conserving fur is probably not necessary in social species because they can take advantage of social thermoregulation at low T a s 37,38 . While individuals of social species are temporarily single such as during digging, burrow maintaining, and patrolling, metabolic heat produced by their activities could help to keep a stable T b (c.f. 36,78 ). On the contrary, solitary species cannot use huddling. Instead, solitary species can minimise body surfaces exposed to the environment solely by curling up during the rest, as observed in all solitary species at low T a s in the present study, and also during several radio-tracking studies under natural conditions [79][80][81] . Denser and thus well insulating fur across the dorsum and flanks of solitary species 31,33 is important because this body region is always exposed to the environment when the animal is rolled into a ball. On the other hand, less insulating fur in social species is advantageous at high T a s, or whilst digging when it is necessary to dissipate surplus metabolic heat to avoid overheating. In this situation, better heat conserving fur of solitary species complicates heat dissipation as we observed higher T diff on the dorsum, and in some species also on the venter, compared to the feet at the highest studied T a in solitary species only (see Supplementary Table S2). Thus, their feet likely play a crucial role in heat dissipation at high T a and/or during digging in solitary species.

Species T a (°C) (n) Fixed term: ventral area
Different social organisation and fur characteristics resulting in different body heat dissipation may have relevant ecological consequences. For the naked mole-rat Heterocephalus glaber, the combination of small body mass and hairless surface resulting together in very high heat losses limits its occurrence in lowland areas of East Africa because it is not able to disperse through the cooler highlands in the western and northern borders of its distribution 82 . We may assume that higher heat losses in social species may handicap them during dispersal and forming of new families especially in colder climate if they disperse singly. Although it is very difficult to collect such information in nature, it seems that dispersing individuals of social species can stay alone for long period of time, likely for weeks, months, or even years [45][46][47][48] facing high thermal challenges during that time. In this regard, although anecdotal, the radio-tracking of F. mechowii revealed interesting finding related to this phenomenon 47 . Lövy et al. 47 found that a solitarily living female had a different circadian activity pattern and resting position than conspecific mole-rats living in a group. This female was most active during the hottest part of day, and usually rested in the curled-up body position while mole-rats from the family were most active during the night, and rested communally outstretched in their nest as indicated by a signal of their collars. Both changing of resting position and activity pattern likely minimised her heat losses. Interestingly, the peak occurred during the hottest part of the day and the pattern of its activity was more similar to that of the activity of the solitary H. argenteocinereus and N. galili 79,80 than to F. mechowii living in a family group 47 . Besides, less insulating fur of social species could explain why the social F. whytei does not colonise generally colder Afromontane grasslands from the lower altitudes of the Nyika Plateau where it is common in contrast to the solitary H. argenteocinereus, even though there is a higher food supply and more easily workable soil in grasslands 44 .
Finally, we analysed whether T diff differs along the venter, and if sociality influences the heterogeneity of ventral T diff distribution as suggested for two mole-rat species with contrasting sociality 31 . For this purpose, www.nature.com/scientificreports/ we divided the venter into five areas (Fig. 1), and for each species, we analysed the pattern of T diff at T a below, within, and above TNZ (Table 5). Although there were some statistically significant differences in the pattern of T diff among these ventral areas for particular species at T a s below and within TNZ (except for B. suillus and G. capensis; see Table 5 and Supplementary Fig. S1), the differences were usually small. Due to this and the absence of any consistent pattern within the tested species, we think that the differences have probably low biological importance. In C. hottentotus and S. cyanus, the differences were found only within TNZ, but not in the lowest T a . As expected, we found homogeneous pattern of T diff at the highest T a . Together with low values of T diff at this T a , it demonstrates that heat dissipation through the ventral surface is maximal. Similarly, no consistent differences in the pattern of the T diff change across the five ventral areas between solitary and social species were detected in either of the three T a s (Table 5). Nevertheless, limitations of the IRT approach should be considered when measuring T s of restricted body areas in small mammals, such as those five ventral areas explored in our study. For instance, the fact that the animals move and bend their bodies during the short recording might provide relatively variable data for such small areas. In this case, the analysis of the insulative quality of the fur might be a better approach to reveal potential differences if such heterogeneity exists. It should be noted that recent findings based on the histological analysis of small skin patches along the body in F. mechowii did not reveal any differences in the amount and structure of the adipose tissue and vascularisation along its venter 43 . This suggests fur having the main role in heat dissipation through the integument in mole-rats and subterranean rodents generally.
For future research on heat dissipation in subterranean rodents, histological analysis of the feet, analysis of T s of the only hairless rodent H. glaber, experiments based on fur shaving, direct measurement of thermal insulation of fur, and/or measuring of skin temperature with thermocouples could bring valuable information on how subterranean rodents handle heat dissipation which is a crucial aspect for mammals who frequent burrows for their whole lives. Table 6. Relationships between the difference between core body and surface temperature (T diff ), ventral area (five ventral areas, for details see Fig. 1) and social organisation (solitary vs. social) in seven species of subterranean rodents. 2.5% HPD and 97.5% HPD represent lower and upper highest posterior density intervals; pMCMC denotes to the p value obtained from MCMCglmm; ventral area × social organisation = interaction between ventral area and social organisation. www.nature.com/scientificreports/

Data availability
The datasets generated and analysed during the current study are available from the corresponding author on reasonable request. www.nature.com/scientificreports/