Bergmann's rule across the equator: a case study in Cerdocyon thous (Canidae)

Authors


Summary

  1. The variation in cranial size of the crab-eating fox Cerdocyon thous was analysed in relation to latitude and several environmental variables throughout its distribution in South America.

  2. We tested the existence of clines to determine whether this canid follows Bergmann's rule to the north and south of the Equator. Also, using niche modelling, we analysed whether the climatic changes during the last glaciation could have influenced Bergmann's rule in this species. We quantified the size of the cranium of C. thous (n = 300). The data were divided into two groups: (i) south of the Equator (n = 163) and (ii) north of the Equator (n = 137). We performed correlations, OLS regressions and simultaneous autoregressions to analyse the relationship between the variation in size and different geographic and environmental variables. Data of occurrence (n = 594) together with ambient variables from the present and the last glacial maximum (LGM) were used to predict the occurrence of C. thous with the implementation of the maximum entropy method. Present-day and historical distribution maps were obtained.

  3. The variation in the size of the cranium of C. thous showed two trends. In the south of Equator, we observed that the size of the skull shows an inverse relationship with temperature-related variables and a positive one with precipitation, while in north of the Equator, we observed the opposite relationship. Populations south of the Equator follow Bergmann's rule showing increasing size with increasing latitude. To the north of the Equator, a non-Bergmannian pattern occurs because size decreases with increasing latitude.

  4. Niche modelling showed two present-day groupings in South America, one north of Amazonia and the other south. However, for the period of the LGM, four groups emerged, possibly related to the four subspecies presently described for C. thous. Therefore, it is possible that the observed pattern – southern populations following Bergmann's rule while northern populations reflecting the opposite – has been influenced by the events that occurred during the LGM that could have led to the differentiation of populations.

Introduction

Macroecology has provided new perspectives through which we can examine geographic distribution patterns of species, notably analyses of latitudinal/altitudinal gradients and of ecogeographic rules (Gaston, Chown & Evans 2008). One of the first and more well-known generalizations in ecology was suggested by Bergmann (1847) and establishes that homeothermic animals inhabiting higher latitudes and colder climates tend to have larger body sizes than related species living in warmer environments. This hypothesis, known as ‘Bergmann's rule’, postulates that this pattern is due to the fact that larger animals have a smaller surface area in relation to body volume, thus allowing better heat retention (or less heat loss) than smaller animals. However, classic Bergmannian patterns (or converse ones) are not always easy to explain, because in many cases, ambient temperature alone is not sufficient to account for all observed trends. Different ecological and evolutionary mechanisms have been put forward to complement or refute Bergmann's original thermoregulatory explanation. Body size clines could follow geographic differences in primary productivity (Rosenzweig 1968; Geist 1987); in seasonality and ambient predictability, leading to improved survival of larger animals in higher latitudes (Calder 1984; Lindstedt & Boyce 1985; Millar & Hickling 1990; McNab 1999); in differential predation (Medina, Martí & Bidau 2007); and others.

Clearly, many selective forces can affect body size in different species or populations, and regional- or global-scale deviations of Bergmann's rule may occur (Dayan et al. 1991), but independently of the driving mechanism leading to Bergmannian patterns, the rule is an empirical generalization whose meaning depends on its prevalence in homeothermic organisms (Meiri & Dayan 2003). Such mechanisms have been amply debated in the last six decades (i.e. Mayr 1956; Scholander 1956; Irving 1957), and the ecogeographic pattern has received wide support from studies of mammals (Ashton, Tracy & de Queiroz 2000; Meiri & Dayan 2003; Diniz-Filho et al. 2007) and birds (Ashton 2002).

Besides ambient factors that may presently influence body size, several other effects must be considered such as life history (Brown 1995; Haskell, Ritchie & Olff 2002; Angilletta et al. 2004; Shelomi 2012), interactions between species (Dayan et al. 1989), evolutionary history (Renaud, Benammi & Jaeger 1999; Blois, Feranec & Hadly 2008) and anthropic effects (McCoy 2012; Yom-Tov, Yom-Tov & Zachos 2013; and references therein). Recent advances in geographic information systems (GIS) allow species niche modelling on the basis of environmental attributes (Guisan & Thuiller 2005). These models have been successfully applied to several animal groups (Luoto, Kuussaari & Toivonen 2002; Raxworthy et al. 2003), in different ecosystems and temporal intervals (Werneck et al. 2011), producing reliable results (Elith et al. 2006) and generating useful information on historical processes.

Several lines of evidence have been used to assess the relevance of climatic fluctuations and the changes in vegetation in the biological diversification of the Neotropical region (Moritz et al. 2000; Carnaval et al. 2009; Werneck et al. 2011). The model known as ‘refugia theory’ was centred in the tropical forests and suggested that climatic changes during the Pleistocene, especially during the last glacial maximum (LGM) about 21 000 years ago, led to fragmentation of the tropical forests creating refugia separated by savannas and other open and dry formations. This could have promoted the expansion or retraction of the distributional ranges of species leading to diversification, fuelled by forest fluctuations (Haffer 1969; Pennington, Prado & Pendry 2000). Meanwhile, the open formations could have served as corridors for connecting populations to the north and south of Amazonia during this period (Pennington, Prado & Pendry 2000; Pennington et al. 2004).

The effects of climate change have had a great influence on flora and fauna and have been well documented in mammalian evolutionary history (Patterson & Pascual 1972; Janis 1993; MacFadden 2000). The canids reached South America after the formation of the Panama isthmus about 2·3 mya through at least two independent colonization events, producing the origin of the two present-day groups of canids of the continent. Thus, a rapid radiation seems to have occurred, which could be related to the retraction and expansion of the glaciers and to the climatic changes in the Andes at the culmination of the Pleistocene (Markgraf 1989; Perini, Russo & Schrago 2010).

Cerdocyon thous (Linnaeus, 1766) is a South American endemic, and the genus is usually considered as monotypic (Berta 1982; Tedford, Taylor & Wang 1995). It is a generalist species, and its diet varies according to the time of the year and the inhabited region being one of the more plastic species within the Neotropics, thus having a large geographic distribution (Trovati, de Brito & Duarte 2007). Cerdocyon thous has a disjunct distribution in South America with two large dispersion areas, north and south of the Equator, showing a clear preference for open habitats (Bisbal 1989; Sillero-Zubiri et al. 2004; Trovati, de Brito & Duarte 2007). Its geographic distribution in the north of the Equator extends on northern Colombia and almost all Venezuela with the exception of the southern Amazonia state and north of Guyana and Surinam. South of the Equator, the species is found in the north-eastern, central and southern Brazil, Paraguay, central-northern Argentina, Bolivia and Uruguay (Sillero-Zubiri et al. 2004). Five subspecies have been recognized within its distributional range: C. t. germanus, restricted to the high savannas of central Colombia; C. t. aquilus, in savannas and forests of Colombia and Venezuela; C. t. thous, found in the Guyanas, eastern Amazonia and northern Brazil; C. t. azarae, in north-eastern and central Brazil; and C. t. entrerrianus, in southern Brazil, northern Argentina, Paraguay, western Bolivia, and Uruguay (Cabrera 1931, 1958; Tate 1939; Berta 1982; Bisbal 1988). However, this subspecific classification is not without criticism, once many diagnostic characters were mainly based on differences in fur colour, which shows a wide intraspecific variation. Nevertheless, evident morphological differences have been observed between both subspecies isolated by the Orinoco river in Venezuela (Bisbal 1988), as well as between populations from northern, eastern and southern South America (Machado & Hingst-Zaher 2009).

Although a vast literature analysing Bergmann's rule in carnivores exists, there is a clear bias towards studies of taxa in the northern hemisphere, especially those with temperate climates, in relation to tropical ones (Meiri, Dayan & Simberloff 2004). Furthermore, very few investigations have focused on tropical taxa with distributions in the north and south of the Equator (e.g. Graves 1991; Brumfield & Remsen 1996); thus, C. thous represents an excellent mammalian model to assess the effects of ambient variables and historical processes influencing body size across the Equator.

In the present work, using geometrical morphometric procedures, we studied whether the body size of C. thous, as represented by cranial measurements, follows Bergmann's rule across its geographic distribution north and south of the Equator; further, through ecological niche modelling, we studied the possible historical effects of the LGM in structuring populations and influencing Bergmannian patterns.

Material and methods

Obtention of Data for Geometric Morphometrics

A total of 300 crania of C. thous, covering most of its geographic distribution, were analysed (Fig. 1). Individuals were identified at the species level on the basis of discrete morphological characters sensu Berta (1982, 1988) and Tedford, Taylor & Wang (1995). Adult individuals were identified using global skull size, the state of the cranial sutures and tooth eruption. The specimens were obtained from several museum collections: Argentina, Fundación Félix de Azara, Museo de Ciencias Naturales ‘Florentino Ameghino’, Museo de Ciencias Naturales de La Plata, and Universidad Nacional de Salta; Brasil, Museu de Zoologia da Universidade de São Paulo, Museu Nacional do Rio de Janeiro, Museu de Ciências Naturais da Pontifícia Universidade Católica de Minas Gerais, Bolivia, Museo de Historia Natural ‘Noel Kempff Mercado’, Museo Nacional de Historia Natural de La Paz; Venezuela, Museo de la Estación Biológica de Rancho Grande, Museo de Ciencias Naturales de Caracas.

Figure 1.

Geographic distribution of the individuals of Cerdocyon thous in South America. The dotted line separates the different subspecies proposed by Cabrera (1931). (1) C. t. entrerrianus, (2) C. t. azarae, (3) C. t. thous, (4) C. t. aquilus and (5) C. t. germanicus.

The skulls were photographed in dorsal view with a Digital Sony DSC-H10 8·1 mp camera (Sony, Japan). The support on where they were photographed as well as the camera, were always levelled, and all the photographs were taken by the same person, as to minimize possible effects of distortion. A total of 27 landmarks in two dimensions were digitalized on the obtained images with the tpsdig v2·16 software (Rohlf 2010; Fig. 2). Procrustes superposition (Dryden & Mardia 1998) was performed with the software morphoj (Klingemberg 2008). Centroid size (Cs), a measurement of landmark dispersion around their centre of gravity, was used as size estimator. It is calculated as the square root of the sum of the squared distances of all landmarks from their centroid (for details of mathematical procedures and applications, see Bookstein 1991 and Zelditch et al. 2004). To evaluate possible photographic errors, the value of the Cs was preliminarily analysed separately on the right and left sides of the dorsal view using 16 landmarks and also the length of skull (cm). Both sides produced similar results between different size estimators (Fig. 3); thus, the final analysis of Cs involved 27 landmarks as shown in Fig. 2.

Figure 2.

Twenty-seven landmarks selected for analysis shown on the image of the cranium of Cerdocyon thous in dorsal view. Landmark (1) end of premaxilla; (2–3) premaxilla–maxilla suture; (4–5) lateral point of the nasal; (6) point of the nasal; (7–8) curve corresponding to the end of P2; (9–10) premaxilla–nasal–maxilla suture; (11) nasal–frontal suture; (12–13) jugal–maxilla suture; (14–15) interorbital constriction; (16–17) end of the postorbital process; (18–19) point of the frontal process; (20–21) postorbital constriction; (22) frontal–parietal suture; (23–24) end of the squamosal; (25–26) end of the occipital crest; (27) inion. Illustrated by my friend Marcelo Ribaya.

Figure 3.

Regressions of the greatest length of skull (a), of right-side landmarks (b) and of the left-side landmarks (c) on latitude. Significance of the regression slopes is indicated. The dotted line represents the Equator and separates both linear regressions.

Data Analysis

To test the possible existence of sexual size dimorphism that could affect our results, we performed anova analyses of Cs between females, males and sexually undetermined individuals in each subspecies. No significant differences within subspecies were observed (P > 0·05); thus, all individuals were analysed jointly regardless of sex. Results are shown in Fig 4.

Figure 4.

anova tests of sexual size dimorphism between females (F), males (M) and sexually undetermined individuals (U) of the subspecies C. t. entrerrianus, C. t. azarae and C. t. aquilus y C. t. thous, respectively. Results of the statistical analyses are indicated.

For ecogeographic variation, a total of 23 variables, 22 climatic and one topographic, obtained from Cramer & Leemans (2001) and the WORLDCLIM data base (available: www.worldclim.com). Because the proximity to large human settlements can influence the quantity and quality of available food for the species and thus affect body size (McCoy 2012), we obtained population density values for the year 2000 of the relevant geographic areas to test possible anthropic effects using the data base at International Earth Science Information Network (CIESIN), Columbia University, and Centro Internacional de Agricultura Tropical (CIAT; available at http://sedac.ciesin.columbia.edu/gpw; Table 1). To process the data base the diva-gis 7·1 software was used (http://www.diva-gis.org/download).

Table 1. List of geographic, topographic, anthropic (A-D) and environmental (1-22) variables used for correlation-regression analyses [ordinary least squares (OLS) and spatial autoregression (SAR)] and ecological niche modelling (ENM)
VariableCodeOLS and SARENM
  1. Crosses indicate the variables used for each type of analysis.

A. LatitudeLATX 
B. LongitudeLONX 
C. ElevationELEXX
D. Human Population DensityHPDX 
1. Annual Mean TemperatureAMTX 
2. Mean Diurnal RangeMDRX 
3. IsothermalityISOXX
4. Temperature SeasonalityTSEXX
5. Maximum Temperature of Warmest MonthMTWMX 
6. Minimum Temperature of Coldest MonthMTCMX 
7. Temperature Annual RangeTARXX
8. Mean Temperature of Wettest QuarterMTWQX 
9. Mean Temperature of Driest QuarterMTDQX 
10. Mean Temperature of Warmest QuarterMTQXX
11. Mean Temperature of Coldest QuarterMTCQXX
12. Annual PrecipitationANPX 
13. Precipitation of Wettest MonthPWMX 
14. Precipitation of Driest MonthPDMXX
15. Precipitation SeasonalityPSEXX
16. Precipitation of Wettest QuarterPWWXX
17. Precipitation of Driest QuarterPDQXX
18. Precipitation of Warmest QuarterPWQX 
19. Precipitation of Coldest QuarterPCQX 
20. Actual EvapontranspirationAETX 
21. Potential EvapotranspirationPETX 
22. Water BalanceWBX 

Because C. thous is present in two great geographically isolated areas, the samples were separated into two groups: (i) comprising all individuals south of the Equator (n = 163) and (ii) those north of the Equator (n = 137) with the aim of assessing whether different trends in size variation occur in both groups. To analyse geographic variation in size, linear regressions between log-transformed Cs and latitude were performed. A series of parametric (Pearson) and nonparametric (Spearman) correlations between log Cs and the geographic, ambient and anthropic variables were carried out to discard those that were not correlated with Cs, which left seven useful variables (Table 2). Conventional statistical analysis assumes that all observations are independent. However, the possible spatial autocorrelation among data may produce an overestimation of the number of independent observations in spatial studies (Peres-Neto 2006) that may lead to the rejection of H0 (type I error), assuming a false correlation between morphological and climatic variables. Different models tend to decrease the effect of spatial autocorrelation, but caution is always recommended when interpreting the results (Bini et al. 2009). In this study, we used univariate and multivariate simultaneous autoregression (SAR) between log Cs and the seven chosen climatic variables using Akaike's Information Criterion (AIC) to select the best model fitting the data, with the sam 4·0 software (Rangel, Diniz-Filho & Bini 2006; http://www.ecoevol.ufg.br/sam). Comparisons were made between the north and south of the Equator.

Table 2. Parametric (Pearson) and nonparametric (Spearman) correlations between log10 skull centroid size of Cerdocyon thous and geographic and climatic variables
AreaVariabled.f.Pearson's r F P Spearman's r t P
  1. Only those variables that produced statistically significant results are shown. For abbreviations of variables, see Table 1. Total: whole C. thous sample; south: subequatorial sample; north: supra-equatorial sample; d.f., degrees of freedom; F, Fisher–Snedecor statistic; t, Student's statistic; P, probability; NS, non-significant. Note that correlation coefficients for LAT are marked with an asterisk; this is because in our analysis, and negative values were assigned to southern latitudes. Thus, they represent truly Bergmannian trends. The negative correlations north of the Equator indicate a converse Bergmannian pattern.

TotalLAT299−0·487*93·06<0·001−0·533*−10·94<0·001
LON2990·22816·53<0·0010·3095·63<0·001
AMT299−0·20312·85<0·001−0·234−4·18<0·001
MDR2990·39354·76<0·0010·4607·93<0·001
ISO299−0·42766·87<0·001−0·484−9·60<0·001
TSE2990·40257·72<0·0010·4789·43<0·001
MTDQ299−0·39330·24<0·001−0·316−5·79<0·001
MTCQ299−0·29328·18<0·001−9·297−5·39<0·001
PWQ2990·35142·09<0·0010·3546·57<0·001
SouthLAT161−0·377*26·85<0·001−0·295*−3·94<0·001
LON161−0·2672·94<0·001−0·292−3·90<0·001
AMT161NSNS
MDR1610·31117·38<0·001−0·292−3·89<0·001
ISO161−0·242−10·110·002−0·207−2·710·002
TSE1610·2258·610·0040·2373·110·004
MTDQ161−0·2379·600·002−0·2383·140·002
MTCQ161−0·1835·610·019−0·182−2·37<0·019
PWQ1610·36925·48<0·0010·3054·09 N<0·001
NorthLAT137−0·42630·20<0·001−0·401−5·13<0·001
LON137NSNS
AMT1370·34117·88<0·0010·4085·24<0·001
MDR137NSNS
ISO137−0·2176·740·01−0·227−2·73<0·001
TSE137NSNS
MTDQ1370·36921·40<0·0010·4055·18<0·001
MTCQ1370·33417·09<0·0010·3965·04<0·001
PWQ137−0·2609·900·002−0·324−4·01<0·001

Niche Modelling

Data on the presence of C. thous were obtained from the different surveyed museum collections, from the ‘species link’ data base (available online at: http://splink.cria.org.br) and from personal observations and communications. A total of 237 localities covering most of the species distribution were recorded. Recent climatic variables (1960–1990) were downloaded from worldclim. Climatic data for the LGM (21 mya Kb) were obtained from the Community Climate System Model (CCSM). Original data were downloaded from the Paleoclimate Modeling Intercomparison Project (PMIP; available online at: http://www.ncdc.noaa.gov/paleo/model.html) with a resolution of 2·5 arcmin (c. 5 km). Highly correlated variables (r > 0·9) were maintained in the model on the basis of their biological relevance following the procedure described by Rissler & Apodaca (2007). Ten variables of 20 available in worldclim were used, including elevation (Table 1). The bioclimatic slices were cut to include latitudes from 12N°54′ to 40S°21′, and longitudes from 82W°22′ to 33W°28′, which represent all zones with bioclimatic conditions compatible with the occurrence of C. thous. The modelling of C. thous distribution during the LGM and the present was performed with the method of maximum entropy with the maxent software (Phillips & Dudík 2008), which has proved to be superior to other modelling algorithms used traditionally to produce predictions about distribution maps (Elith et al. 2006). In the present work, the precision of the model was estimated from the area under the curve (AUC) method, which is considered an effective indicator of the performance of the model and frequently used to assess the consistency of the model's projections on the distribution of species (Manel, Williams & Ormerod 2001; Pearson et al. 2006). More details are given in the study by Phillips (2008).

Results

Clinal Body Size Variation

The analysis of body size variation regarding latitude both sides of the Equator, revealed two distinct trends: Individuals south of the Equator showed a significant increase in size with increasing latitude (F = 32·44; r2 = 0·16; P < 0·01) according to the expectations of Bergmann's rule. However, those populations north of the Equator showed the opposite trend, a significant decrease in body size with increasing latitude (= 22·17; r2 = 0·14; P < 0·01), thus following the converse to Bergmann's rule (Fig. 5).

Figure 5.

Variation in cranium size of Cerdocyon thous in relation to latitude. Centroid size (Cs) was log-transformed. The dotted line represents the Equator and separates both linear regressions.

When partial SARs were performed, significant correlations between body size and the climatic variables MDR, ISO, TSE, MTDQ, MTCQ and PWQ were verified south of the Equator, while populations in the north showed significant correlations between body size and the climatic variables AMT, ISO, TSE, MTDQ, MTCQ and PWQ, indicating that some climatic variables differ in their effects on populations north and south of the Equator. Two different trends were observed: south of the Equator, temperature variables MTDQ and MTCQ showed a negative correlation in relation to body size according to Bergmann's rule. However, both parameters showed a positive correlation with body size north of the Equator, opposing Bergmann's pattern. The analysis of the precipitation variable PWQ also showed opposite trends in the populations north and south of the Equator: body size was positively correlated with PWQ in the southern populations, which could be related to ambient productivity and resource availability, while a negative correlation occurred for the north of the Equator (Table 3). From multivariate SARs, it can be seen that the model that best predicts body size south of the Equator includes MDR, ISO and PWQ, which significantly explains c. 20% of the variation. In the north of the Equator, the best model includes AMT, ISO, MTDQ, MTCQ and PWQ. MTDQ was significantly explaining 20% of the total variation in body size (Table 4).

Table 3. Univariate simultaneous autoregressions (SARs) between centroid size (Cs) and different environmental variables (predictors) in the groups of Cerdocyon thous, north and south of the Equator
RegionVariableCoefficient OLSCoefficient SARPredictor (r2)Predictor + space (r2) F P
SouthAMT−0·227−0·2930·0170·0322·7450·099
MDR0·9601·0370·0960·11617·159<0·001
ISO−1·670−1·7750·0590·07710·0070·002
TSE0·0100·0120·0500·0718·4630·004
MTDQ−0·330−0·3520·0560·0749·4670·002
MTCQ−0·270−0·2900·0330·0515·5230·020
PWQ0·1400·1290·1360·13425·253<0·001
NorthAMT0·7800·7630·1160·13517·742<0·001
MDR−0·070−0·282<0·0010·0220·0140·908
ISO−2·040−1·9220·0470·0606·6700·011
TSE0·0500·0410·0210·0382·9160·090
MTDQ0·7500·7240·1360·15021·200<0·001
MTCQ0·7700·7420·1110·12916·939<0·001
PWQ−0·130−0·1150·0670·0779·7400·002
Table 4. Multivariate Simultaneous Autoregressions (SARs) between Centroid Size (Cs) and different environmental variables (predictors) in the groups of Cerdocyon thous, north and south of the Equator. The best model for each region was chosen from the lower value of Akaike's Information Criterion (AIC)
RegionVariablesOLS coefficientSAR coefficientPredictors (r2)Predictors + space (r2) F P
SouthModel (3 predictors)  0·2100·22013·730<0·001
MDR0·6200·720<0·01
ISO−1·030−1·120<0·01
PWQ0·1100·100<0·01
NorthModel (5 predictors)  0·1900·2006·260<0·001
AMT2·9903·2700·180
ISO2·8802·6400·070
MTDQ3·2002·7600·030
MTCQ−5·570−5·3900·050
PWQ−0·080−0·0800·130

When comparing body size variances as estimated by Cs north and south of the Equator, a difference is observed even considering that the northern region is smaller in area. However, when comparing the variances of the climatic variables, a clear difference is apparent: a higher variance occurs in the populations south of the Equator (Table 5).

Table 5. Variance of the centroid (Cs) and the correlated environmental variables of Cerdocyon thous, north and south of the Equator
 Variance
CsAMTMDRISOTSEMTDQMTCQPWQ
South1·58701·31313·963·6612130771599142822592
North1·27466·5675·9228·052373559146110466
South/North1·241·504·132·2651·102·713·092·15

Ecological Niche Modelling

The distribution of C. thous predicted by the niche model for the present climatic conditions showed a good performance with AUC = 0·886. Two geographic regions can be distinguished: (i) comprising the Dry Diagonal (Cerrado, Caatinga and Chaco), and (ii) comprising the llanos of northern Venezuela, the Caribbean coast of Venezuela and Colombia, and part of the Guyanas (Fig. 6a). Besides these zones, small spots were observed in the savanna region of northern Brazil and Bolivia, suggesting a reduction in the distribution potential through the Amazonian basin. Precipitation during the wet season was among the variables that most influenced the distribution of the species, indicating that high levels of rainfall may limit the range of C. thous. High thermal amplitude also decreases the probability of occurrence of the species.

Figure 6.

Modelling of the distributional ranges of Cerdocyon thous for the climatic scenarios of (a) present-day, (b) Last glacial maximum (LGM) and (c) the bioregions of the South America, Chaco (CH), Cerrado (CE), Pantanal (PA), Caatinga (CA), Amazonas (AM) and savanna (SA). The dotted arrows show the possible corridors that linked the north and south populations of Equator during the LGM. The localities of presence are indicated by white circles, and the maxent probability values are listed as a scale of greys. Black corresponds to the logistic result of higher probability of occurrence of Cerdocyon thous. AUC refers to the area under the curve, an indicator of the performance of the model.

Furthermore, the projection of the present niche in the climatic scenario of the LGM showed four main groupings: (i) northern Venezuela and western Colombia, (ii) northern Brazil, (iii) north-eastern Brazil, corresponding to the Caatinga region and (iv) Chaco. It is important to note the existence of a corridor connecting the Chaco and Caatinga blocks (Fig. 6b).

Discussion

Clinal Variation in Body Size

Despite the vast literature on Bergmann's rule since its formulation in 1847, there is a notorious paucity of studies addressing the variation in body size of species or higher taxa both sides of the Equator. Gay & Best (1996) analysed the size variation in Puma concolor (Felidae) in North and South America, observing an increase in the size at higher latitudes, both north and south of the Equator. Likewise, two intraspecific studies in Andean Passeriformes, the grey-bellied Flowerpiercer Diglossa carbonaria (Icteridae) and the yellow-billed Cacique, Amblycercus holosericeus (Thraupidae), showed that Bergmann's rule is verified in both sides of the Equator, with birds becoming progressively larger as distances north and south of 0° latitude increase (Graves 1991; Kratter 1993). However, Brumfield & Remsen (1996) studying four Andean species of Cinnycerthia wrens (Troglodytidae) found a converse pattern both north and south of the Equator: birds showed a negative correlation between body size and latitude.

However, the case of Cerdocyon thous is very different from what other studies have shown. The species shows a significant amount of body size variation throughout its geographic distribution. In this investigation, we aimed to identify the main responsible factors for such wide variation. The most intriguing observation was that the individuals distributed south of the Equator followed the canonical Bergmann's rule (Tables 3 and 4), that is, animals from higher latitudes are larger in relation to those nearer the Equator, but this was not true for populations north of the Equator (Ashton, Tracy & de Queiroz 2000; Meiri & Dayan 2003). Since the formulation of Bergmann's rule, the main mechanism proposed to explain the observed patterns was thermoregulation; larger bodies allowed for better heat conservation (Rensch 1938; Mayr 1956).

However, this mechanism has not received complete support. Although many studies have stressed the importance of temperature on body size clinal variation in birds and mammals (e.g. Smith & Betancourt 1998; Ashton, Tracy & de Queiroz 2000), other studies suggest that factors such as precipitation, primary productivity, seasonality or resource availability may be equally or more important in generating Bergmann-like patterns (James 1970; Murphy 1985; Yom-Tov & Yom-Tov 2005; Yom-Tov & Geffen 2006). As indicated, populations north of the equator showed a significant decrease in size with increasing latitude (Fig. 5), showing a converse Bergmannian trend. Thus, both groups of populations have completely opposite geographic patterns of body size distribution.

We used SARs to discriminate the effects of space and the environmental predictors that could explain the geographic patterns shown for C. thous on both south and north of the Equator. Again, two different trends were observed, suggesting not only that different variables are probably involved in body size variation in both groups, but also that variables could be acting in contrasting ways (Tables 3 and 4).

For the populations south of the Equator, it was observed that the mean temperatures of the driest and coldest periods have a negative correlation with body size, that is, a higher mean temperature during these periods predicts smaller size, which agrees with Bergmann's rule. A strong correlation with seasonality was also observed: there exists a positive correlation with mean diurnal range and a negative one with isothermality [(MDR/TAR)*100)], indicating that the high variability of daily temperatures may act as a selective force on the evolution of body size, because in less predictable environments, larger individuals would be positively selected (Meiri, Dayan & Simberloff 2005). Seasonality is a plausible alternative or complementary mechanism to explain Bergmann's rule, because colder regions tend to be more seasonal (Ashton, Tracy & de Queiroz 2000; Ashton 2001a,b).

An increase in precipitation during the warmer season had also a positive correlation with body size. This fact was observed in previous studies (James 1970; Wigginton & Dobson 1999; Yom-Tov & Geffen 2006) and could reflect the relevance of primary productivity and resource availability as having a positive effect on body size, more than a direct impact of precipitation on organisms (Blois, Feranec & Hadly 2008). Primary productivity generally impacts carnivores only indirectly, but may have a direct effect on the omnivorous C. thous; fruits are an important element of its diet (Bisbal & Ojasti 1980; Facure & Monteiro-Filho 1996; Macdonald & Courtenay 1996).

In the populations north of the Equator, the situation is completely different. Mean annual, autumn and winter temperatures show a positive relationship with body size, while summer rainfall shows a negative one. Converse Bergmannian patterns are rare in endothermic animals (Katti & Price 2003; Meiri & Dayan 2003; Ochocinska & Taylor 2003; Medina, Martí & Bidau 2007) and are even more difficult to explain. It would be expected that populations north of the Equator had less body size variation than those south of the Equator, because of the difference in the geographic range. A smaller range reduces the probability of size clines (Meiri & Dayan 2003). Nevertheless, when comparing Cs variances, it was observed that body size variation is high and similar in both sides of the Equator, while the environmental variance is much lower north of the Equator. This suggests that environmental variables may not be the main factor influencing Cs among these northern populations.

North of the Equator, there exist two subspecies, C. t. aquilus and C. t. thous, which are probably allopatric and which show differences in body size (Bisbal 1988). C. t. thous is larger than C. t. aquilus, which leads to a non-Bergmannian pattern. Two main forces may have driven the subspecies to diverge in body size: genetic drift and natural selection. When the genetic flux becomes limited between populations, as it seems to be the case for these two subspecies isolated by the formidable barrier of the Orinoco river, rapid morphological differentiation and local adaptation may occur (Thompson 1998), and natural selection would thus be a powerful force driving phenotypic divergence (Ogden & Thorpe 2002).Another factor that could be affecting body size is resource availability. Bisbal & Ojasti (1980) showed that different populations of C. t. aquilus feed preferentially on small vertebrates, insects and fruits and that the diet composition varies in the dry and wet seasons, while C. t. thous eats mainly fruits and, to a less extent, small vertebrates, without showing great variations in diet along the year. However, these authors were not able to correlate size with diet. The present-day climatic factors and feeding habits do not seem to explain body size variation north of the Equator. The observed great phenotypic variation between geographically close populations that are isolated (at least partially by the Orinoco river) suggests that natural selection has a central role in size variation. However, a role of genetic drift in morphological differentiation cannot be disregarded (see 'Historical Factors'). We also did not find evidence that food availability was related to anthropic interference as suggested for other carnivores (Yom-Tov, Yom-Tov & Zachos 2013, and references therein).

Historical Factors

The prediction of the geographic distribution of C. thous for present-day climatic conditions indicates a high probability of occurrence in open habitats of South America (savannas and seasonal dry forests), agreeing with the proposals of several authors (Bisbal 1989; Trovati, de Brito & Duarte 2007), having two clear distribution areas, isolated by the Amazonian Forest. The consensual opinion is that these two regions were connected during the cold and dry periods of the LGM in the recent past, when savannas expanded and the Amazonian forest retracted (Mayle, Burbridge & Killeen 2000). Three corridors connecting the northern and southern savannas during the LGM have been proposed: (i) the Andean corridor, connecting the southern savannas with Colombia and Venezuela; (ii) the central Amazonian corridor, connecting the southern savannas with those of northern Amazonia; and (iii) the coastal corridor that would connect south with north through spots of savanna along the Atlantic coast (Haffer 1997; Cardoso da Silva & Bates 2002).

The data of the present study shed light on the possible historical distribution of C. thous. The modelling of the paleodistribution showed four main groupings with two groups in the north and two in the south, potentially connected south of the Equator, that could be related to the subspecies C. t. aquilus and C. t. thous, and C. t. azarae and C. t. entrerrianus, respectively, and agreeing with the proposal of Cabrera (1931). Our data do not support the hypothesis of a central corridor through Amazonia and is consistent with connections of the populations through the coastal corridor and the Andes (Fig. 6c).

The climatic changes associated with events of retraction and expansion of the Amazonian forest during the quaternary have been postulated by several biogeographers as an important factor in the speciation of South American tropical organisms (Whitmore & Prance 1987; Haffer & Prance 2001). Phylogeographic studies in C. thous have shown that populations from north-eastern Brazil exhibit a certain degree of genetic differentiation when compared with those of southern Brazil and Paraguay, while those from the Cerrado and Pantanal share haplotypes with both regions (Tchaicka et al. 2007). These data are consistent with the niche modelling performed by us: two groupings of high occurrence probability, one in the Caatinga and the other in Chaco, were observed with an intermediary contact zone during the LGM. The existence of two groupings north of the Equator during the LGM shows that a possible isolation that occurred in the past could have produced the differentiation of the subspecies C. t. aquilus localized in the north of Venezuela and Colombia, and C. t. thous localized in the north of Brazil and southern Guyanas.

The idea of isolation is reinforced by the fact that both subspecies show substantial differences in shape and size, C. t. thous being larger than C. t. aquilus. Both subspecies occur in allopatry, separated by the Orinoco river basin (Bisbal 1988).

Future research of population genetics will allow us to meet the levels of genetic diversity within subspecies, which will be important to detect possible population bottlenecks during the LGM. Likewise, phylogeographic studies comprising the entire distribution of the species will help to confirm or refute that these two subspecies have been isolated since the LGM. They will also indicate the level of divergence between the different subspecies, to determine whether C. thous is a single species or a species complex.

Factors affecting body size such as climatic conditions, resource availability and the composition of communities may vary frequently in the continent along time (Raia & Meiri 2011). The potential populations of C. t. thous during the LGM were possibly exposed to selective pressures different to those existing nowadays, which could have led to an increase in body size of this subspecies. Separate from the niche modelling, it can be observed that the potentially habitable area for C. thous was larger during the LGM than in the present (Fig. 6) and that the climatic change and the concomitant expansion of the forest have produced a reduction in open areas in South America (Werneck et al. 2011), which are suitable for the species. This situation would explain the presence of populations of C. t. thous in northern Brazil isolated from populations of the same subspecies in Venezuela (Sillero-Zubiri et al. 2004). Under this perspective, it may be expected that population bottlenecks have occurred for C. t. thous, and given that genetic drift has an important effect in fixating phenotypic characteristics in small populations (Stern & Orgogozo 2009), a larger size in C. t. thous could be the result of drift.

The mechanisms that produced the inversion of Bergmann's rule north of the Equator are still obscure, but the situation of the present geographic isolation of C. t. aquilus and C. t. thous, driving these subspecies to fixate unique adaptive characteristics, plus possible genetic drift events in C. t. thous could help to explain the unusual observed pattern.

In conclusion, two opposite trends with regarding geographic variation in body size were observed for Cerdocyon thous. First, populations south of the Equator show a strong Bergmannian pattern of spatial (clinal) body size variation negatively correlated with temperature. The positive correlations of body size with seasonality and precipitation also suggest a dependence on resource availability. Conversely, populations north of the Equator, comprising two subspecies, showed a completely opposite pattern. We consider that this deviation from Bergmann's rule in northern populations must have been influenced by historical factors, local adaptation and genetic drift.

Acknowledgements

We would like to thank Shai Meiri and Miguel Angel Olalla-Tarraga for their helpful criticism of a previous version of this manuscript. We are also extremely grateful to Dakota McCoy, Graham Slater, and an anonymous reviewer for their expert comments and suggestions. The Coordenação de Aperfeiçoamento de Pessoal de Nível Superior (CAPES-REUNI) and the Postgraduate Ecology Course of the Universidade Federal do Rio Grande do Norte are gratefully acknowledged for their support.

Ancillary