Supplementary material, Bridging macroecology and macroevolution in the radiation of sigmodontine rodents
Maestri, R., Luza, A.L., Hartz, S.M., Freitas, T.R.O., Patterson, B.D. April 2022
DEM_Elev = raster used in the maps presented in the main text Lon-lat-Disparity.txt = cell coordinates (latlong) PresAbs_228sp_Neotropical_MainDataBase_Ordenado = composition table (sites x species) shape.v.RData = RData ventral skull shape size.v.RData = RData ventral skull size Sigmodontinae.ventral.dig = output TpsDig Sigmodontinae.ventral.listed.pruned = list of species and individuals used in the analysis Sigmodontinae.ventral.listed = list of species and individuals used in the analysis Sigmodontinae_285speciesTree.Tre = consensus tree Sigmodontinae_413species100Trees.trees = 100 fully resolved phylogenies
This document presents the supplementary material of our article. The supplementary information embraces analyzes for species with skull and phylogenetic data (216 species), and also Oryzomyalia species. The figures and tables presented here follow the same order and settings as the ones shown in the main text:
1 - A table with the GLM coefficients, showing the differences in slope of SES disparity ~ SES MPD relationship between empirical and simulated datasets.(This table is similar to the Table 1 presented in the main text).
2 - A plot depicting the relationship between SES-MPD (Mean Pairwise (Phylogenetic) Distance between species) and the SES-disparity produced by empirical (skull size and shape) and three simulated data sets generated (Brownian motion[BM], Early-Burst[EB], and Ornstein-Uhlenbeck [OU]). (This panel is similar to the Fig. 2 presented in the main text).
3 - A panel of five maps comprising the A) empirical morphological disparity (as measured by Rao’s Entropy), B) disparity compared to a randomization-based null model, C) disparity compared to an evolutionarily oriented null model (OU for disparity in skull size, BM for disparity in skull shape), D) cells presenting significant deviations of disparity compared to a randomization-based null model (all values presented in B), E) cells presenting significant deviations of disparity compared to a evolutionarily oriented null model (all values presented in C). This figure is similar to the Fig. 3 presented in the main text.
4 - A panel comprising four maps, showing the disparity expected by i) a randomization-based null model, ii) BM model, iii) EB model, iv) OU model. The correlation between these values is presented in a separate table after the map. (This panel is similar to the Fig. 4 presented in the main text).
The values of parameters, estimated through macroevolutionary models (using ‘fitContinuous’ in univariate datasets (skull size), and ‘mvgls’ in multivariate datasets (skull shape)), are only presented here in the supplementary material.
The relationship between species richness and skull disparity (before and after a randomization-based null model)
Fig. S2.1. The relationship between assemblage-level species richness (SR) and observed/empirical values of the empirical phenotypic disparity of skull size (Rao’s entropy index) (left), the disparity compared to a randomization-based null model (middle), and the disparity compared to an evolution-oriented model (Ornstein-Uhlenbeck) (right).
Results considering the complete set of species (413 species), and using the complete sample of 100 fully resolved phylogenies from Upham et al. (2019) to consider phylogenetic uncertainty on parameter estimates.
Fig. S2.3. Density plot showing the estimates of each model parameter for univariate and multivariate macroevolutionary models. In the top we show the estimates of the parameter ‘sigma’ for the BM, EB, and OU models of evolution (estimated by evolutionary models (fitContinuous) applied to univariate trait data (skull size)). In the middle we show the estimates of the parameters ‘beta’ and ‘alpha’, which are specific from EB and OU, respectively. In the bottom, we show the estimates of the parameter ‘sigma’ for the BM model (evolutionary models (mvGLS) applied to multivariate trait data (skull shape)). In univariate models, the density represents variation in estimates across the 100 fully resolved phylogenies. In the multivariate model, the density represents variation of sigma across traits and phylogenies (112 simulated ‘landmarks’ and 100 fully resolved phylogenies). Results for the dataset of 413 species.
Below we present a table of GLM coefficients that considers phylogenetic uncertainty while testing which model produced a simulated disparity closer to the empirical disparity (results alternative to Table 1 in the main text). It shows show that the disparity simulated by the OU model gets closer to the empirical disparity than the disparity estimated by other models (Table S2.1). The GLM estimates were averaged (and its standard deviation calculated) across 100 runs per phylogeny (for 100 fully resolved phylogenies).
Table S2.1: Averaged GLM estimates, obtained by averaging the estimates across the 100 fully resolved phylogenies. Results for the dataset of 413 species.
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) 0.41951107 0.04291587 9.836620 1.611533e-13
## MPD -0.08265248 0.01773169 -4.692754 7.277070e-03
## datasetBM -0.36583792 0.06069220 -6.111344 4.037407e-02
## datasetEB -0.39576278 0.06069220 -6.644241 5.016113e-02
## datasetOU -0.42095639 0.06069220 -6.991466 3.445204e-02
## MPD:datasetBM 0.27680513 0.02507639 11.154513 2.145854e-02
## MPD:datasetEB 0.25900668 0.02507639 10.463898 2.773115e-02
## MPD:datasetOU 0.14709810 0.02507639 5.935941 7.554150e-02
Table S2.2: Standard deviation of GLM estimates across the 100 fully resolved phylogenies. Results for the dataset of 413 species.
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) 0.04221835 0.003809334 1.171888 1.564391e-12
## MPD 0.02360367 0.002000536 1.341819 3.814276e-02
## datasetBM 0.41024451 0.005387211 6.821072 1.301869e-01
## datasetEB 0.41023149 0.005387211 6.931416 1.699229e-01
## datasetOU 0.42716073 0.005387211 6.978750 1.092794e-01
## MPD:datasetBM 0.24336830 0.002829185 9.656811 1.057659e-01
## MPD:datasetEB 0.26186764 0.002829185 10.450733 1.104965e-01
## MPD:datasetOU 0.21457778 0.002829185 8.717895 1.948662e-01
Using this data subset, the estimated parameters of macroevolutionary models were very similar to those estimated for the complete set of species (compare Figs. S2.1 and S2.2). We then produced a bivariate plot describing the relationship between SES-disparity and SES-MPD for empirical and simulated data sets. We found that this relationship (Fig. S2.3) was similar to the one reported in the main results (Fig. 2). Overall, the disparity produced by the OU model had a closer relationship with the empirical disparity than had the other models of evolution (BM and EB) (Fig. S2.5).
As presented in the results (mapping section), we counted the number of cells with values of disparity higher, equal, or lower than the null model (randomization-based null model, NULL) and the evolutionary models. We found that 10, 11 and 11 assemblages had a disparity lower than expected by the OU, EB and BM models, respectively; three assemblages had a disparity lower than expected by the null model. None assemblage had a disparity higher than predicted by BM, OU and EB models, and 90 assemblages had a disparity lower than the null model.
## $lower
## lowerNULL lowerOU lowerBM lowerEB
## TRUE 3 10 11 11
##
## $higher
## highNULL highOU highBM highEB
## TRUE 90 NA NA NA
When mapping the values of empirical and simulated trait values, we found a high agreement between the datasets of 413 and 216 species (Fig. S2.6).
Finally, we show the maps of null and simulated disparity (Fig. S2.7). We found that using a randomization-based null model is similar to simulating a trait using the OU model. This can be seen in the following map (and the correlations presented after the Fig. S2.7), where the null disparity and the disparity produced by the OU model produced highly correlated maps (rho=0.96, Table below the map). The correlation was also high between the null and BM disparity.
Correlation between values of disparity
## RAO_OBS.med_nulo obsBM obsEB obsOU
## RAO_OBS.med_nulo 1.0000000 0.8890401 0.8475222 0.9625762
## obsBM 0.8890401 1.0000000 0.9534437 0.9385652
## obsEB 0.8475222 0.9534437 1.0000000 0.9268018
## obsOU 0.9625762 0.9385652 0.9268018 1.0000000
The relationship was similar to the plots for the datasets of 413 and 216 species. However, the main difference is that we miss values of SES MPD higher than zero as we are focusing on phylogenetic distances within the subclade Oryzomyalia.
Fig. S2.8. Density plot showing the estimates of each model parameter for univariate and multivariate macroevolutionary models. In the top we show the estimates of the parameter ‘sigma’ for the BM, EB, and OU models of evolution (estimated by evolutionary models (fitContinuous) applied to univariate trait data (skull size)). In the middle we show the estimates of the parameters ‘beta’ and ‘alpha’, which are specific from EB and OU, respectively. In the bottom, we show the estimates of the parameter ‘sigma’ for the BM model (evolutionary models (mvGLS) applied to multivariate trait data (skull shape)). In univariate models, the density represents variation in estimates across the 100 fully resolved phylogenies. In the multivariate model, the density represents variation of sigma across traits and phylogenies (112 simulated ‘landmarks’ and 100 fully resolved phylogenies).Results for the dataset of Oryzomyalia species.
Fig. S2.9. Bivariate plot showing the relationship between SES MPD and SES disparity, for empirical and simulated disparity. Results produced by using the dataset of Oryzomyalia species.
The number of cells (assemblages) with disparity deviating from randomization-based and evolutionarily oriented null models was similar to the number found for the dataset of 216 species.
## $lower
## lowerNULL lowerOU lowerBM lowerEB
## TRUE 2 11 12 18
##
## $higher
## highNULL highOU highBM highEB
## TRUE 89 NA NA NA
The maps of observed and simulated disparity still resemble the ones shown in Fig. 3 (main text), and the ones just reported considering 216 species (Fig. S2.6). However, less assemblages had significantly lower and higher disparity than the disparity expected by the OU model (Fig. S2.10).
Finally, there was a high correlation between null and simulated disparity, especially for the OU model.
The correlation between the null and simulated disparity was similar to the correlation found for other datasets (the highest correlation was found between NULL and OU-simulated disparity, with a rho=0.98).
## RAO_OBS.med_nulo obsBM obsEB obsOU
## RAO_OBS.med_nulo 1.0000000 0.9494270 0.9593531 0.9822153
## obsBM 0.9494270 1.0000000 0.9778866 0.9691552
## obsEB 0.9593531 0.9778866 1.0000000 0.9723185
## obsOU 0.9822153 0.9691552 0.9723185 1.0000000
Results considering the consensus phylogeny. As it has 285 tips, the number of species used in trait simulations was 285 (for traits simulated with all tips), being 169 with occurrence, trait and phylogenetic data.
The figures we will present now have the same settings and order as the ones we presented above.
Parameter estimates for the univariate model fitted to skull size (table below), and for the multivariate model fitted to skull shape (Fig. S3.1).
## Estimates Parameter model
## 1 0.124 sigma BM
## 2 0.142 sigma EB
## 3 -0.012 beta EB
## 4 0.124 sigma OU
## 5 0.000 alpha OU
Bivariate plot similar to Fig. 2. The OU-resulting disparity is still closer to the empirical disparity than other models (see the table of coefficients and the difference in slope (i.e., the effect of the interaction between SR and the classes of null models)).
Fig. S3.2. Bivariate plot showing the relationship between SES MPD and SES disparity, for empirical and simulated disparity. Results for the data set of 285 species present in the consensus tree.
##
## Call:
## lm(formula = value ~ MPD * Data, data = av_disp)
##
## Residuals:
## Min 1Q Median 3Q Max
## -3.2975 -0.1929 -0.0287 0.1157 2.8196
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) 0.14841 0.02583 5.745 9.60e-09 ***
## MPD 0.04471 0.01258 3.553 0.000384 ***
## DataBM -0.12768 0.03653 -3.495 0.000477 ***
## DataEB -0.10271 0.03653 -2.812 0.004945 **
## DataOU -0.23521 0.03653 -6.439 1.29e-10 ***
## MPD:DataBM 0.19405 0.01780 10.903 < 2e-16 ***
## MPD:DataEB 0.17331 0.01780 9.738 < 2e-16 ***
## MPD:DataOU 0.09605 0.01780 5.397 7.04e-08 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 0.5911 on 6280 degrees of freedom
## Multiple R-squared: 0.1809, Adjusted R-squared: 0.1799
## F-statistic: 198.1 on 7 and 6280 DF, p-value: < 2.2e-16
We found that 28 assemblages had a disparity lower than the null model, whereas 7, 28 and 24 assemblages had disparity lower than the OU, BM and EB models, respectively. We found that 467 assemblages had a disparity higher than expected by the OU model, and 53 assemblages had a disparity higher than the null model. None assemblage had disparity higher than expected by the BM and EB models.
## $lower
## lowerNULL lowerOU lowerBM lowerEB
## TRUE 38 7 28 24
##
## $higher
## highNULL highOU highBM highEB
## TRUE 53 467 NA NA
Maps similar to Fig. 3, and to maps just reported considering other datasets (Fig. S2.6 and S2.10). The maps still resemble the ones found in the main text, but now we did not find cells (assemblages) with a disparity higher or lower than expected by an OU model of evolution.
Using the consensus phylogeny, we found variation relative to results presented above and in the main text. The maps of BM- and OU-simulated disparity showed higher spatial variation than reported previously.
The correlation between the null disparity and the simulated disparity was moderate to high. The highest correlation we found was between BM- and EB-simulated disparity (rho = 0.95). The correlation between the null disparity and the OU-simulated disparity was rho = 0.37.
## RAO_OBS.med_nulo obsBM obsEB obsOU
## RAO_OBS.med_nulo 1.0000000 0.8019505 0.7970357 0.3668096
## obsBM 0.8019505 1.0000000 0.9557484 0.4459697
## obsEB 0.7970357 0.9557484 1.0000000 0.4662957
## obsOU 0.3668096 0.4459697 0.4662957 1.0000000
Results considering skull shape, and the application of macroevolutionary models for multivariate trait simulation fitted to skull shape (using a BM model of evolution). The estimates of model parameters for datasets of 216 species and Oryzomyalia species were presented in Figs. S2.3 and S2.8.
None assemblage had disparity higher than the null model, whereas only four assemblages had a disparity higher than the BM model. Nonetheless, 480 assemblages had a disparity lower than the null model, and 860 assemblages had a disparity lower than the BM model.
## $lower
## lowerNULL lowerBM
## TRUE 480 806
##
## $higher
## highNULL highBM
## <NA> NA 4
The BM model had a good fit to skull shape (Fig. S4.1, Table S4.1).
Table S4.1: Averaged GLM estimates, obtained by averaging estimates produced by each one of the 100 different phylogenies used to simulate traits and calculate disparity. Results for the dataset of 216 species.
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) -0.11989648 0.03947344 -3.079593 7.904859e-02
## MPD 0.54050076 0.01631589 33.482403 1.903306e-108
## datasetBM 0.16171537 0.05582387 2.935900 9.517206e-02
## MPD:datasetBM -0.09359235 0.02307415 -3.950603 3.464664e-02
Table S4.2: Standard deviation of GLM estimates, obtained by calculating the standard deviation of estimates produced by each one of the 100 different phylogenies used to simulate traits and calculate disparity. Results for the dataset of 216 species.
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) 0.08220315 0.003829300 2.186437 1.458757e-01
## MPD 0.03881256 0.002010355 3.550319 1.893766e-107
## datasetBM 0.41893270 0.005415448 7.607369 2.250030e-01
## MPD:datasetBM 0.22694128 0.002843071 9.681416 1.192826e-01
Assemblages presenting lower disparity than predicted by a BM model of evolution were located throughout South America except the eastern and southern region (Fig. S4.2).
## RAO_OBS.med_nulo obsBM
## RAO_OBS.med_nulo 1.0000000 0.9293195
## obsBM 0.9293195 1.0000000
## $lower
## lowerNULL lowerBM
## TRUE 458 1008
##
## $higher
## highNULL highBM
## TRUE 1 3
Table S4.3: Averaged GLM estimates, obtained by averaging estimates produced by each one of the 100 different phylogenies used to simulate traits and calculate disparity. Results for the dataset of Oryzomyalia species.
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) -0.1494278 0.03592063 -4.361487 5.493398e-02
## MPD 0.4177589 0.01146675 36.668705 4.549490e-164
## datasetBM 0.1755463 0.05079944 3.554921 1.002113e-01
## MPD:datasetBM -0.1572541 0.01621643 -9.384956 3.508392e-02
Table S4.4: Standard deviation of GLM estimates, obtained by calculating the standard deviation of estimates produced by each one of the 100 different phylogenies used to simulate traits and calculate disparity. Results for the dataset of Oryzomyalia species.
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) 0.09947100 0.003517084 3.258340 0.1681786
## MPD 0.04021881 0.001439523 2.968131 0.0000000
## datasetBM 0.41706099 0.004973908 8.257509 0.2235813
## MPD:datasetBM 0.19903811 0.002035793 11.852784 0.1393277
## RAO_OBS.med_nulo obsBM
## RAO_OBS.med_nulo 1.0000000 0.9634121
## obsBM 0.9634121 1.0000000
## $lower
## lowerNULL lowerBM
## TRUE 473 957
##
## $higher
## highNULL highBM
## <NA> NA 41
Table S4.5: Averaged GLM estimates, obtained by averaging estimates produced by each one of the 100 different phylogenies used to simulate traits and calculate disparity. Results for the dataset of 285 species present in the consensus tree.
##
## Call:
## lm(formula = value ~ MPD * Data, data = av_disp)
##
## Residuals:
## Min 1Q Median 3Q Max
## -2.46518 -0.19241 0.01479 0.18853 2.62946
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) -0.41323 0.02814 -14.687 < 2e-16 ***
## MPD 0.49420 0.01371 36.052 < 2e-16 ***
## DataBM 0.47329 0.03979 11.895 < 2e-16 ***
## MPD:DataBM 0.10713 0.01939 5.526 3.54e-08 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 0.6438 on 3140 degrees of freedom
## Multiple R-squared: 0.519, Adjusted R-squared: 0.5185
## F-statistic: 1129 on 3 and 3140 DF, p-value: < 2.2e-16
## RAO_OBS.med_nulo obsBM
## RAO_OBS.med_nulo 1.0000000 0.8845862
## obsBM 0.8845862 1.0000000
## RAO_OBS.med_nulo obsBM
## RAO_OBS.med_nulo 1.0000000 0.9351824
## obsBM 0.9351824 1.0000000