The radiation and geographic expansion of primates through diverse climates

Significance

Textbooks often portray primates as originating, evolving, and dispersing exclusively within warm tropical forests. This tends to come from fossil evidence distributed across northern latitudes typically characterized as tropical. However, accumulating independent evidence suggests that nontropical climates were common across these regions during early primate evolution. By employing a geographic model capable of inferring ancestral locations within a phylogenetic framework while accounting for continental drift, we find that, contrary to widespread assumptions, early primates primarily inhabited cold and temperate climates. This research suggests that primates evolved and dispersed through diverse climates before becoming largely confined to modern warm tropical forests.

Abstract

One of the most influential hypotheses about primate evolution postulates that their origin, radiation, and major dispersals were associated with exceptionally warm conditions in tropical forests at northern latitudes (henceforth the warm tropical forest hypothesis). However, this notion has proven difficult to test given the overall uncertainty about both geographic locations and paleoclimates of ancestral species. By the resolution of both challenges, we reveal that early primates dispersed and radiated in higher latitudes, through diverse climates, including cold, arid, and temperate conditions. Contrary to expectations of the warm tropical forest hypothesis, warmer global temperatures had no effect on dispersal distances or the speciation rate. Rather, the amount of change in local temperature and precipitation substantially predicted geographic and species diversity. Our results suggest that nontropical, changeable environments exerted strong selective pressures on primates with higher dispersal ability – promoting the primate radiation and their subsequent colonization of tropical climates millions of years after their origin.

The idea that early primates—including both modern primates and those belonging to the broader Euprimates clade (1)—originated, radiated, dispersed, and thrived in the tropical forests, has been the dominant narrative for more than four decades (2–18). It is often reported that the primate radiation dramatically expanded, both geographically and taxonomically, in association with the global warming of the Paleocene–Eocene Thermal Maximum (PETM), when the range of the tropical forest presumably reached high latitudes in the Holarctic continents (3, 10, 17–22). We henceforth refer to these collective notions as the warm tropical forest hypothesis–i.e. that warm tropical forests have been instrumental in defining primate origins and evolution. The fact that the current distribution of extant primate species is largely restricted to a narrow range of tropical temperatures is also used to bolster support for the warm tropical forest hypothesis. However, current paleoclimatic evidence derived both from spore-pollen fossils and general circulation model simulations does not indicate tropical climates at the key locations where early fossil primates have been discovered or in the continents where they most likely originated (23–25). For example, climate reconstructions indicate that the Bighorn Basin and Chalk Butte in North America, as well as key fossil-bearing sites in Western Europe, were not tropical before and during the PETM (23).

The apparent discrepancy between the expectations of the warm tropical forest hypothesis and that early primate fossils are found in nontropical climates could stem from two factors. First, there are inherent temporal, spatial, and taphonomic biases in the fossil record. These biases have made the warm tropical forest hypothesis a longstanding challenge to evaluate. This is because the fossil record may tell us about the places, times, and, most importantly, the climates where fossilization was most likely—rather than where primates evolved and diversified. That is, a higher probability of fossilization in nontropical climates (26) could lead to the apparent lack of tropical climates in the early fossil record of primates. Second, despite the prevalent notion of warm tropical forest origins of primates, there is still considerable ambiguity in the terminology used to define and classify climate. Indeed, previous work on primate paleo-environments (27) has noted the extent of this issue: “it seems that every work on the paleoecology of ape’s environments adopts one or other of the numerous classifications of vegetation structure, never the same”. A nonexhaustive literature search retrieves at least 10 names associated with the type of climate where primates are proposed to have originated—all incorporating concepts like tropical, warm, and wet: tropical rainforest (17), tropical climate (3), tropical flora (16), continuous evergreen forest belt (18), tropical plants (10), lush forest (10), paratropical forest (17), tropical angiosperm biome (11), warm forest (12), and wet forest (12).

Here, we overcome the enduring difficulty of evaluating the warm tropical forest hypothesis through the resolution of those two main problems. First, we combine climatic inferences from the fossil record with inferences based on the location of common ancestors–i.e. internal nodes within the most recent and comprehensively sampled Euarchonta phylogeny (28). To do this, we use a modified version of the Geographic (Geo) model implemented in BayesTraits v5 (29, 30) that accounts for historical inhabitable environments (e.g. ancient seas)—as well as incorporating phylogenetic and topological uncertainty (Materials and Methods). Such an approach complements data derived from fossils and extant species locations as it can reveal hitherto unknown historical climates in unsampled regions where ancestral primates lived. Ancestral species could have dispersed over vast distances, potentially up to about 20,000 km in just 20,000 y (18), resulting in significant geographic and climatic differences between fossil sites and ancestral species locations. Second, after extracting geographic and climate data for fossils and ancestral species, it was essential to use a formal and standard climate classification criterion to make the hypothesis testable and reproducible, avoiding the ambiguity and complexity of climate definitions found in the literature. It is possible to achieve this by using climatic data derived from the Hadley center general circulation Coupled Model (Materials and Methods) (31–33) to classify the paleoclimates using the Köppen-Geiger (KG) classification system (34, 35).

Our approach brings the unique opportunity to explicitly and quantitatively test the various predictions made by the longstanding warm tropical forest hypothesis. If ancestral species evolved and relied on tropical forests for their dispersal, we would expect to find most early phylogenetic nodes to be reconstructed within the KG climate category Tropical (the type “A” climate, SI Appendix, Fig. S1 A and S1 B) (35). The Tropical climate category refers to an environment that is hot all year round with average annual temperature over 18 °C (34, 35). This climate includes multiple subcategories: Tropical Rainforest (wet), Tropical Monsoon (short dry season), and Tropical Savannah (distinct dry season), which each differ in the annual amount and pattern of rainfall (34, 35). Since the KG Tropical category encompasses a wide variety of subclimate categories, we are adopting a conservative approach in what constitutes a tropical climate. Furthermore, if early primates dispersed greater distances during the warmest global conditions of the PETM – when the tropical forests reached their widest latitudinal extension (17, 18, 20) – we would expect to observe a negative association between the total distance traveled across the globe for each species (pathwise, from root to tip—henceforth D_PATHWISE) and time (SI Appendix, Fig. S1C). Similarly, we expect to observe a positive effect of global mean temperature on D_PATHWISE (SI Appendix, Fig. S1 E). If early primates were also more speciose during the PETM, we would expect to observe a negative association between pathwise node count (the pathwise number of nodes, from root to tip, henceforth NC_PATHWISE) and time (SI Appendix, Fig. S1D). NC_PATHWISE can be regarded as a speciation rate metric (36). Finally, we should observe a positive effect of global mean temperature on NC_PATHWISE (SI Appendix, Fig. S1F).

To test these predictions, we carried out phylogenetic generalized least squares (PGLS) models using D_PATHWISE as the response variable, with time and global temperature as predictors. Then, we carried out phylogenetic generalized linear mixed models (37) (PGLMM) using the same predictors but using NC_PATHWISE as response variable with a Poisson distribution (Materials and Methods).

Results

Crown Primates’ Ancestral Location and Paleoclimate.

Using a single median phylogenetic tree (from a set of 100 median dated trees reconstructed in this study; SI Appendix, Fig. S2 and Materials and Methods), the geographic distribution of the most recent common ancestor of crown primates was inferred to have existed within North America in 8 of 10 Geo analyses. In the other two models, Western Europe was the most likely location. When we ran the analyses on the full set of 100 median phylogenetic trees, incorporating both topological and branch length uncertainty, we found North America as the location for the common ancestor in 70% of trees, and Western Europe in the other 30% of trees (SI Appendix, Supporting Text and Table S1). Those two likely locations are expected if we consider the location of some of the earliest fossil primates such as Teilhardina magnoliana and T. brandti in North America (20, 38), and T. belgica in Western Europe (39).

However, there is controversy as to the continental origins of primates owing to their almost synchronous first appearances in the fossil records of Asia, Europe, and North America (18, 20). Some previous research has additionally suggested that primates could have originated in Africa (40). Thus, we explicitly compared the model fit of each of these four potential continental origins (Fig. 1 A and B) by running four additional models, each assigning one of four paleo-continents to the node representing the primate common ancestor and global paleomaps for all the other nodes in the tree. We ran 10 Geo analyses for each of the four models. Comparison of these models based on marginal likelihoods estimated by stepping-stone sampling (41) (Materials and Methods), showed that the model where the common ancestor was restricted to fall within North America was statistically preferred (Bayes Factor > 5; strong evidence; Fig. 1B) – this agrees with previous findings (38, 42). Finally, when we ran the Geo model on 200 down-sampled input datasets (Dataset S1) in which we mitigate sampling biases by reducing the number of fossils included from higher latitudes (SI Appendix, Supporting Text), we still obtained North America as the most likely location for the common ancestor in 80% of the 200 datasets (Dataset S2).

Fig. 1.

The common ancestor of Primates was found in a cold climate within North America. (A), world paleomap with the Köppen-Geiger main climates of 66 Mya. White circles on the map are the posterior distribution of coordinates for the common ancestor (white circle on the tree). These coordinates were inferred using the Geo model with paleomap restrictions. We ran four Geo models, restricting the location to be either in North America, Africa, Asia, or Europe. (B), mean annual paleotemperature extracted from the posterior distribution of coordinates across each continent. The continental locations are ordered by their marginal likelihood inferred by steppingstones (numbers in squared brackets), with the best fit model on top given the Bayes Factor model comparison.

The monthly paleotemperature and paleoprecipitation extracted from the posterior distribution of coordinates for the common ancestor in North America, indicates a cold KG climate category, classified as “D” (Fig. 1A). The Cold climate category is defined by a wide range of temperatures through the year, with temperatures of the hottest month over 10 °C and temperatures of the coldest month equal to or lower than 0 °C. The cold climate can also include years with dry summers (Ds), dry winters (Dw), or no dry season (Df). Within cold climates without dry seasons, it is still possible to find hot summers (Dfa) in which the summer temperature is equal to or over 22 °C, and warm summers (Dfb) in which the temperature is over 10 °C and equal to or lower than 22 °C for more than 4 mo a year (34, 35). The climate subcategory for the primate common ancestor was the “Dfa” type (SI Appendix).

Historical Climatic Transition.

Our climate reconstructions across phylogenetic nodes reveal that the dominant climate inhabited by ancestral species changed dramatically through time (Fig. 2 A and B). Such a changeable climate is contrary to the general notion that primates mostly relied on tropical climates (17). Rather, we show that most early primates occupied cold, temperate, and dry climates (Fig. 2 A and B). This result does not support a major expectation of the warm tropical forest hypothesis, i.e., a climate reconstruction of all or most of the early phylogenetic internal nodes falling within a Tropical KG climate (Fig. 2A).

Fig. 2.

Primates evolved through diverse climates. (A), primate median phylogenetic tree with the Köppen-Geiger main climates categories for all internal nodes. The climate reconstruction was obtained by extracting the monthly paleotemperature and paleoprecipitation (simulated under the HadCM3BL-M2.1aD climate model) from the Geo model posterior node coordinates. (B), lineages through time intervals of 1 My. Fossils and extant species (phylogenetic tips) are also included in this plot. Colors represent the proportion of lineages inhabiting each climate category through time.

Our patterns of climatic reconstruction stand over the full set of 100 phylogenetic trees (Dataset S3), which means that our results are robust to topological and divergence-time uncertainty in the phylogenetic tree, and to the uncertainty of ancestral locations inferred with the Geo model (SI Appendix, Table S1). Our ancestral climatic reconstruction remains qualitatively similar even when we ran all the analyses on the 200 down-sampled datasets that are heavily biased against nontropical fossils in northern latitudes (SI Appendix, Supporting Text and Datasets S1 and S2). If the taphonomic bias favoring fossilization in nontropical climates was influencing our node climates to be reconstructed as nontropical, then we should observe an increase of tropical climates across phylogenetic nodes when running the analyses on input data that randomly down sample the nontropical fossils; but this expectation was not supported by the evidence (Dataset S2). Contrary to this expectation, the results show that there was an enduring major representation of nontropical climates across all nodes corresponding to early primate history, across all the 200 datasets (Dataset S2). This means that, even when the input datasets for the Geo model analyses are heavily biased toward tropical fossils, our inferred pattern of ancestral climates is maintained.

Notably, our down-sampled datasets included the random removal of several species with uncertain phylogenetic positions such as Ekgmowechashalidae, and we also obtained similar results when including or excluding all the fossils with problematic phylogenetic position like Parvimico materdei, Dolichocebus annectens, and Ucayalipithecus perdita (Datasets S4 and S5). We can therefore conclude that these species of questionable phylogenetic positionare not influencing our results.

Then, looking at the change in climate along branches (i.e., between main KG climate categories; arid, cold, temperate, and tropical, Fig. 2A) we see that climate transitions are relatively rare. Specifically, only ~22% of phylogenetic branches showed such a transition from ancestral to descendant node. However, subclimatic (Dataset S6) transitions are relatively more common (39% of branches). Within the ~22% of branches that showed transitions between main KG climate categories, the most common ones were from arid to tropical (~5% of branches), followed by tropical to temperate (4% of branches) (Fig. 3A). The least frequent transitions were from tropical to cold, and from cold to tropical (< 1% of branches; Fig. 3A). Transitions from cold to any other climate were also rarely observed (< 2% of branches; Fig. 3A). Transitional branches were associated with longer geographic dispersal distances (median distance for transitional branches = 561 km; median distance for nontransitional branches = 137 km), which means that major historical colonization of novel climates (as defined by the main KG climate categories) was linked to long distance dispersals. Taken together, our results demonstrate that primates have had the ability to disperse and colonize diverse climates, where longer dispersal distances have tended to be associated with major climatic transitions.

Fig. 3.

Primates historically transitioned across diverse climates. (A), transition between the Temperate (Top), Arid (Left), Tropical (Bottom), and Cold (Right), main climates, for all primates. Arrow size represents the proportion of phylogenetic branches with the respective transitions. (B), climatic transitions for early primates, who were living between 65 and 47.8 Mya. (C), climatic transitions for species that were living between 47.8 and 23.03 Mya. (D), climatic transitions for species that were living from 23.03 Mya to the present.

To delve further into the historical transition between main KG climate categories and to provide a comprehensive understanding of its temporal pattern, we elaborate on the transitions across three temporal windows, namely, from 66 to 47.8 Mya (early), 47.8 to 23.03 (middle), and 23.03 to the present (late; Fig. 3 B–D). Most of the transitions in the early primate radiation were from the cold to the temperate climate (Fig. 3B). In the middle of the radiation, most of the transitions occurred from the temperate to the arid climate (Fig. 3C). During the late radiation, major transitions occurred from the arid to the tropical climate (Fig. 3D). This late radiation pattern occurred seemingly coeval to a global trend toward drier and cooler climates of the Neogene (since 23.03 Mya), and the expansion of the major mid-latitude deserts (e.g., Saharo-Arabia). During the Neogene, the Earth cooled and experienced the onset of ice sheet expansion and expansion of the Mid-Latitude deserts. Perhaps such climatic changes may have caused the dispersals and transitions into tropical climates (43–45).

Global Temperature Is Decoupled from Biogeographic Movement and Speciation.

Time had a significant positive effect on both D_PATHWISE and NC_PATHWISE (SI Appendix, Table S2), meaning that dispersal distances and speciation rates were higher toward the present. This result is expected by virtue of the fact that longer-lived primate lineages have had more time to move and speciate - this is why it is important to account for the effect of time when evaluating the effect of additional covariates like global temperature (GT) in the regressions.

When we evaluated the effect of GT on D_PATHWISE and NC_PATHWISE while accounting for time, we did not find a significant effect on either of the two response variables (SI Appendix, Table S2). These results do not support the warm tropical forest hypothesis expectation, i.e., that the past warmer global temperatures, including those of the PETM, were associated with the highest species dispersal distances and speciation rates (SI Appendix, Fig. S1 E and F).

GT may not explain primate biogeographic movement and speciation because of the natural mismatch between global and local environmental conditions (SI Appendix, Fig. S3). Thus, perhaps local environmental conditions like local temperature (LT) and local precipitation (LP) specific to the region where each species lives or lived might relate positively to D_PATHWISE and NC_PATHWISE. To assess this expectation, we tested the effects of LT and LP in our phylogenetic regression models. Such local variables were obtained from paleocoordinates for fossils, and from current coordinates for extant species (Materials and Methods). Our results show that LT had a significant effect on both response variables (positive on D_PATHWISE, negative on NC_PATHWISE; SI Appendix, Table S2), which is in line with our expectation. LP, on the other hand, had a significant negative effect on NC_PATHWISE only (SI Appendix, Table S2). However, the effect size of LT and LP was less than 3% of the variance in D_PATHWISE and NC_PATHWISE (SI Appendix, Table S3). The relatively low amount of variance explained means that there is still uncertainty about what the main factors are that have driven the dispersal and speciation of primates across multiple continents.

The Rate of Change in LT and LP Substantially Explains Dispersal and Speciation.

It has been proposed that the rate of climate change could be a factor of paramount importance in determining the geographic and evolutionary dynamics of primates (21, 46–50). In fact, during the PETM, GT not only increased to one of the highest on record but also those increases were exceptionally fast (45, 51). Also, our results show that ancestral species dispersed and transitioned across diverse KG climatic subcategories (Dataset S6), which differ substantially in the annual pattern of both LT and LP (34, 35). Examining the local climatic variables allows us to explicitly test whether changeable local environments in a species’ history influenced D_PATHWISE and NC_PATHWISE. Therefore, we tested the effect of the rate of local climate change on species dispersal and speciation by including the pathwise rate of local temperature (LT_RATE) and the pathwise rate of local precipitation (LP_RATE) as additional covariates in our phylogenetic regressions (Materials and Methods). LT_RATE and LP_RATE are the cumulative change of LT and LP across the phylogenetic branches linking the common ancestor of all primates with every fossil and extant species, divided by time (Materials and Methods; though the effect of these variables is identical when we do not divide by time). Crucially, these changes are not directional, i.e., changes can be to either cooler, warmer, drier, or wetter local conditions.

We found that the effect size of LT_RATE and LP_RATE was 19% and 2% for D_PATHWISE, respectively (SI Appendix, Table S3). Both rates of local changes related positively to D_PATHWISE (Fig. 4 B and C), which means that primates dispersed longer distances when the LT and LP changed at higher rates, irrespective of changes to warmer or colder temperatures or to drier or wetter conditions. On the other hand, the LP_RATE had a significant positive effect on NC_PATHWISE (Fig. 4F and SI Appendix, Table S2), explaining 14% of the variance (SI Appendix, Table S3). This positive association means that primates diverged into new species more frequently when the total amount of annual LP changed at higher rates. When the local environment became drier or wetter, rapidly over time, speciation rate was higher.

Fig. 4.

Primates dispersed and radiated under variable rates of change in local climate. (A), local temperature (LT) had a significant positive effect on pathwise distance (D_PATHWISE). (B), the pathwise rate of local temperature (LT_RATE) had a significant positive effect on D_PATHWISE. (C), the pathwise rate of local precipitation (LP_RATE) had a significant positive effect on both D_PATHWISE. (D), the pathwise node count (NC_PATHWISE) was positively associated with time. (E), local precipitation (LP) had a significant negative effect on NC_PATHWISE. (F), LP_RATE had a significant positive effect on NC_PATHWISE. We show the predictor variables that were significant in both the median and the sample of phylogenetic trees (Materials and Methods). Light-colored lines represent the posterior distribution of phylogenetic regression slopes. Darker lines represent the mean slopes of the posterior distribution.

The positive effect of LT_RATE and LP_RATE on dispersal and speciation was robust to several sources of uncertainty (Materials and Methods). These sources of uncertainty include: the inferred ancestral locations reconstructed at phylogenetic nodes (SI Appendix, Tables S1, S4 and S5), the multiple continents proposed for primate origins (SI Appendix, Table S1), the spatial variation in LT and LP (SI Appendix, Tables S4 and S5), and the uncertainty associated with the topology and branch lengths of the phylogenetic tree (SI Appendix, Tables S4 and S5).

Discussion

Our results suggest that early primates moved through, evolved, speciated, and went extinct in, as well as mostly lived in the nontropical climates of the northern continents. Primates dispersed away from cold and variable climates, toward warmer and more stable ones. It was only late in their evolutionary history that primates colonized and diversified in tropical climates (SI Appendix, Fig. S4). Our results are in stark contradiction to the warm tropical forest hypothesis that has prevailed for decades as the most likely explanation for primate origin and evolution (2, 17). This discrepancy is mainly attributable to bringing a new generation of phylogenetic and biogeographical methodologies to bear on the question of primate biogeography and its links to climate. These methods have the potential to realistically infer past geographical movement across the continents. In turn, this means that we can reveal climates outside the range of those seen in extant species—something which is not possible with other methods. In addition, our ability to extract local rather than global environmental conditions can explain why our results do not show strong support for the warm tropical forest hypothesis. SI Appendix, Fig. S3, shows how divergent the local temperature has been from the global temperature during the primate radiation. Taken together, our approach allows us to test the warm tropical forest hypothesis in a far more nuanced fashion than has ever before been possible.

The idea that most early primates inhabited cold climates across high latitudes may seem to be at odds with fossil evidence concerning the paleoclimates of northern latitudes around 55 Mya. For instance, fossil evidence from Canada’s high Arctic suggests a mild temperate climate during the early to middle Eocene, with winter temperatures at or just above freezing, summer temperatures reaching ≥ 20 °C, and high precipitation (52). However, using the KG classification system, the fossil evidence agrees with our results. Our results show that the common ancestor of primates inhabited cold climates with no dry seasons and hot summers (Dfa, Dataset S6); which is defined by summer temperatures equal or over 22 °C (34, 35). The take-home message here is that warm temperatures alone are not enough evidence to support a tropical or temperate climate, and that fossil and paleoclimate data can define similar climates when the data are classified under the same criterion.

In the context of the effect of temperature on primate dispersal and evolution, we observe a clear, direct effect of local temperature, whereas global temperature shows no meaningful effect owing to their divergent patterns during the primate radiation. Specifically, local mean temperature fell below the global mean early in the radiation and rose to above-average levels during the middle and later stages (SI Appendix, Fig. S3). The differential effect of local versus global temperature on primate evolution may help understand why important primate adaptations are decoupled from global temperature. For example, the fossil record shows that primate body size evolution was unaffected by periods of global cooling, periods of global warming, and relatively stable periods (53). To test hypotheses linking body size evolution with temperature—like the Cope-Bergmann hypothesis (54)—we should place more emphasis on the direction and rate of local temperature changes (55).

Local geographical location and climate are critically important for making inferences about the physiological nature of the primate common ancestor. This information can provide previously unknown clues to help us understand how the common ancestor might have lived and thrived in its ecological context. It is possible that hibernation or torpor might have been a survival strategy to thrive in such cold and seasonal conditions (56). This idea finds support in several small primates that live in unpredictable environments with freezing winter temperatures. For example, dwarf lemurs (genus Cheirogaleus) exist in cold climates with scarce resources by entering continuous hibernation for several months (56). Specifically, the highland-dwelling dwarf lemurs C. crossleyi and C. sibreei, dig themselves into the ground for hibernation beneath a soft layer of plant roots, humus, and leaves, where they are protected from freezing temperatures (56, 57). Whether the earliest primates were able to enter either torpor or hibernation is still debated (56), but some studies suggest that they could, possibly owing to having exceptionally low metabolic rates given their inferred small size (8). Small primates have high energy needs per unit of body mass and typically depend on scarce resources that are rich in energy, such as insects, small vertebrates, saps, and gums (8, 10). Given this, the primates’ common ancestor could have lived in environments with low overall productivity or marked seasonality by lowering their metabolic rate or by temporally abandoning their normal homeothermic state (8). Future studies could investigate the genomic signature of heterothermy evolution in ancestral primates, given evidence of accelerated convergent evolution in conserved noncoding elements among hibernating mammals (58) and selection for energy metabolism genes in colobine monkeys under cold climates (49).

Our results have implications for some of the main hypotheses proposed to explain primate origins. For example, the visual predation (2) and terminal branch feeding (16) hypotheses postulate that the adaptations that set modern primates apart from other mammals evolved as a response to dietary specialization. Both assume that selective processes took place in a warm, wet, and tropical forest environment. However, our results, supporting a cold climate, suggest that this dietary specialization—whether for visual predation or for plant products—might have occurred in forests adapted to cold rather than tropical climates. Some studies suggest that the origin of mammalian clades might be linked to the radiation of rosids (59), an angiosperm clade that includes many orders well-adapted to temperate climates and cold temperatures outside the tropics (60). Indeed, fossils of rosid plants, which are characteristic of coniferous and mixed forest ecosystems, have been found in North America around 66 Mya (61), in the region we found most likely to be the primate common ancestor origin. Future work might seek to find a direct link between the geographical movement of these angiosperm taxa and primates. Such a connection would lend more support to the notion that primate origins were influenced by mixed forest environments.

This study highlight the idea that the prevalence of extant primates highly adapted to warm and stable tropical climates are the result of a long evolutionary process that started with selection on ancestral species thriving in a variety of colder and more seasonal paleoclimates of the northern latitudes (Figs. 1 and 2). From this colder and more seasonal local paleoenvironmental setting, the surviving species that started to diversify were those species able to disperse longer geographic distances, toward different—but more stable climates (Fig. 2).

The potential main selective force on dispersal ability was the rate of local environmental changes in temperature and precipitation, which we find to have both substantially predicted longer dispersal distance (SI Appendix, Table S3). This result agrees with inferences made from individual-based models that explore the effect of rates of climate change on dispersal evolution (62). By assuming that species track geographically shifting conditions, these models show that under rapid rates of local climate change, species evolve to have higher dispersal capacity (62). Given that populations can cross large landscape gaps, this evolutionary process in turn could increase the rate of speciation by range fragmentation. To explain the remainder of the variance in primate dispersal capacity and speciation rates, it may be of interest to look at the extent to which other ecological and geographic factors, like body mass and geographic range size, relate to dispersal and speciation over long evolutionary timescales (63–65).

Finally, we can logically hypothesize that the evolutionary failure of early primates (and Plesiadapiformes; SI Appendix, Fig. S5 and Table S6), living in the colder and more fluctuating climates of the northern continents, was caused by their inability to keep moving toward warmer and more stable climates when their local environment changed too fast. As the main changes in driving species dispersal and speciation were not directional, then early species did not become extinct from global cooling (i.e., directional change to lower global temperatures) or forest contraction and disintegration in northern latitudes (i.e., directional changes to habitat reduction) (12). Consequently, species with a higher capacity to disperse from places with challenging and volatile environments are the only ones that have left their evolutionary trace in contemporary primate diversity.

Materials and Methods

Phylogenetic Trees.

Our comparative analyses were based on the most complete phylogeny of Euarchonta to date (28). This phylogeny includes 902 tips of which 419 are extant and 483 are fossils (Dataset S5). The Primates clade (excluding the incertae sedis Altanius orlovi (10)) contains 404 extant and 361 fossil tips. To address issues associated with zero-branch lengths and both temporal and topological uncertainty, we dated a random sample of 100 of the total set of most parsimonious topologies (MPTs) obtained by Wisniewski et al. (28), using a slightly modified version of their tip-dating procedure (28) and implemented in BEAST2 (66).

For each of the 100 MPTs, we obtained a posterior sample of dated trees from which we retrieved a single representative by calculating a median tree using the Kendall–Colijn metric (67) (Dataset S7). After removing potentially problematic taxa (SI Appendix, Supporting Text), this set of 100 median trees contains 894 tips spanning Euarchonta (Dataset S8). We conducted the comparative analyses based on each on these 100 median trees, and on an additional median tree calculated from the 100 median trees. We also conducted all comparative analyses after additionally excluding Parvimico materdei, Dolichocebus annectens, and Ucayalipithecus perdita (Dataset S4) as they (probably erroneously) were recovered as stem anthropoids (28). Finally, we conducted all comparative analyses on the original median tree available in ref. 28 (Dataset S5) for comparison. Results across all phylogenies were qualitatively similar (Dataset S5).

Geographic Distribution Data.

We obtained geographic coordinates (longitude and latitude) for every phylogenetic tip in order to infer the posterior distribution of ancestral locations at phylogenetic nodes.

For extant species, we downloaded distribution polygons from the IUCN Red List of Threatened Species (68) and generated a random sample of coordinates within each polygon. This approach allowed us to get an exhaustive representation of the extent of the geographic distribution for each species. There were three species in the tree which were absent from the IUCN database: Lepilemur mittermeieri, Microcebus mittermeieri, and Otolemur crassicaudatus. For each of these, we obtained their polygons using the Map of Life (MOL) database (69). We defined the number of random coordinates to generate based on the polygon's geographic area (SI Appendix, Table S8). This dataset for all coordinates is available as Dataset S9.

For most fossil species, we downloaded paleogeographic coordinates from the Paleobiology Database, PBDB (61). For fossil species with no information in PBDB, we obtained present-day coordinates from the localities where the fossils were found. Then, we reconstructed their paleogeographic coordinates using the “reconstruct” function of the chromosphere R package, version 0.4.1 (70), the PALEOMAP model (71), and both their first appearance date (FAD) and last appearance date (LAD). We randomly adjusted coordinates for sister taxa which fell within <50 m of one another (SI Appendix, Supporting Text). The complete coordinate dataset for fossils is available as Dataset S10.

Ancestral Locations Inference.

We inferred the posterior distribution of coordinates at internal phylogenetic nodes using a modified version of the Geo model (29) in BayesTraits v5 (30). The previous version of the Geo model reconstructed the posterior distribution of nodal longitude and latitude using a three-dimensional Cartesian coordinates system (x, y, and z) that considers the spherical nature of Earth. It estimates the posterior distribution of ancestral coordinates while sampling across all the coordinates within the geographic range of species. This approach avoids the potential biases of using one coordinate per phylogenetic tip such as is often done using either geographic centroids, mid latitudes, or longitudes. Changes in coordinates across the branches of the phylogenetic tree are modeled using Brownian motion which assumes that species disperse across the globe at a constant speed (distance over time). However, the Geo model can also estimate ancestral locations while considering continuous variation in dispersal speed across phylogenetic branches, i.e., the model considers the continuous variation of dispersal capacity. The speed of movement ranges from species quiescence (no movement per unit time), through constant movement in direct proportion of the passage of time (Brownian motion), to heterogeneous long-distance dispersals per unit time. Estimations of the speed of species’ dispersal across phylogenetic branches are based on the variable rate model (72). The variable rate model detects shifts away from a constant (background) speed expected under Brownian motion. We compared the constant and variable speed models by means of Bayes Factors (BF). The BF is calculated as double of the difference of the log marginal likelihoods of two models—estimated by stepping stone sampling in BayesTraits while considering the number of parameters of the model (i.e., model complexity) (41). Higher log marginal likelihoods represent better fitted models and by convention, BF > 2 indicates positive support, BF = 5 to 10 indicates strong support, and BF > 10 is considered very strong support for a model over the other (73).

In this study, we introduce a modified extension to the original Geo model (29) that restricts reconstructed locations to points found only on land. Initially, all reconstructed locations are placed on land, and when proposing a new location, the closest point to the proposed location is identified on the map. If the closest point is found to be in the sea, the new location is assigned as zero probability (rejected), otherwise it is accepted or rejected based on its likelihood. Geography is not static through time; therefore, maps were created for different time periods (see below). As the phylogeny is time calibrated each internal node is assigned a map based on its age.

To restrict the space for ancestral location inferences, we used maps from the PALEOMAP project (71), which contains global maps for every My, during the last 76 My. This input map file for the Geo model analysis is available at Figshare (74). We matched every phylogenetic node with the closet paleomap given their ages. With this approach we ensured that the reconstructed longitudes and latitudes for the phylogenetic internal nodes fell within the ancestral configuration of continents. The ability to restrict reconstructions to valid paleocoordinates means that we could consider continental drift in the reconstruction of the primates’ ancestral geographic locations. Simulations demonstrate that the new model version accurately recovers ancestral locations (SI Appendix, Supporting Text and Fig. S6).

We ran 10 MCMC chains of 800 million iterations each, discarding the first 600 million iterations as burn in. We also ran four analyses where we restricted the paleomap for the node representing the crown Primates. We selected the four continents proposed as the ancestral location for the crown Primates: Asia, Africa, Europe, and North America. We ran 10 MCMC chains of 800 million iterations for each of the four model restrictions, discarding the first 600 million iterations as burn in. Then we compared all models by using Bayes Factors. Usually, when Europe is suggested as the place of origin for the common ancestor of primates, it is implicitly included within the coarser continental area of Eurasia (28, 75). This coarse discretization of Earth is a common practice given the limitations of biogeographical models that use discrete tip data in the phylogenetic tree. However, we leverage the higher geographical resolution of the Geo model to evaluate whether either Europe or Asia is a possible location for the origin of crown primates.

12-mo Paleoprecipitation and Paleotemperature.

We obtained global values of monthly precipitation and mean monthly surface temperature for a period spanning the entire evolutionary history of Euarchonta given the dates of both the median and set of phylogenetic trees, i.e., up to around 76 Mya. We used three approaches to get the monthly precipitation and temperature for extant, fossil, and phylogenetic nodes, respectively.

For extant species, we extracted monthly precipitation and temperature values from their geographic centroids from the WordClim Version 2 (76). We estimated the species geographic centroid within their IUCN and MOL polygons, using the terra R package, version 1.7.71 (Dataset S11).

For fossil species, we obtained the monthly paleoprecipitation and paleotemperature values from their paleocoordinates (Dataset S10) using world paleoclimatic simulations based on the Hadley center general circulation Coupled Model (HadCM3). Here, we use the HadCM3BL-M2.1aD model: See SI Appendix, Supporting Text, for a full description of this scheme. We used the FAD and LAD of each fossil to match paleoclimatic layers to the fossils (SI Appendix, Supporting Text).

Finally, for the internal phylogenetic nodes, we extracted the monthly paleoprecipitation and paleotemperature (matched by age, as above) from the HadCM3BL-M2.1aD model. We used the posterior distribution of coordinates that were reconstructed with the Geo model. As there are many phylogenetic nodes in the median tree that do not match exactly the age of each climatic layer (given their difference in age resolution), we also extracted the paleoclimate data using the node ages from the sample of 100 phylogenetic trees. With this approach we considered the uncertainty in node ages when matching them with their closet simulated paleoclimate layers. All our results were qualitatively similar when running the analyses on the median tree and across the sample of 100 phylogenetic trees (SI Appendix, Supporting Text).

Köppen-Geiger Climate Classification.

Using the monthly values of precipitation and mean monthly temperature for nodes, fossils, and extant species, we formally classified climates based on the Köppen-Geiger climate classification system (KG). The KG system classifies climates based on threshold values and seasonality of monthly air temperature and precipitation (35). It reflects climatic factors limiting vegetation growth and it aims to empirically map biome distributions around the world (35). The KG system classifies climates into five main categories. These main categories are Tropical A), Arid B), Temperate C), Cold D), and Polar E) climates—which are themselves divided among 30 subcategories (34, 35). For example, within the main category Tropical, there exists Tropical Rainforest, Tropical Savannah, and the Tropical Monsoon subcategories. The system can break down general terms that are equated in the hypothesis, such as “tropical” and “warm”. Warm temperatures can define KG climates that are fundamentally different. For example, the climate classified as “Af” (Tropical rainforest) having temperatures over 18 °C all year round, is starkly different from the climate “Dfa” (Cold, without dry season, and hot summer) which has temperatures in the hottest month over 22 °C (35).

It is important to note that the climate concept used in this study differs from the environment concept. Climates are first-order processes for environments (i.e., processes that explain environment assuming a completely flat, nonvariable landscape), but environments are also influenced by second-order processes such as focused recharge, orography, topology, lithology, etc. KG climates, for example, do not identify hydric wetlands and other microhabitat variability. Our approach allows us to separate the macroscale first-order climatic processes influencing radiation and dispersal, but do not show fine details of environmental variability.

Finally, the KG classification has been widely applied to paleoclimate simulations (23, 24, 77). The KG system has several advantages over other climate classification systems in terms of applicability, comparability, and quantifiability (78). SI Appendix, Fig. S7 shows the world KG reconstructions for the present and for the past, using the WorldClim Version data, and the HadCM3BL-M2.1aD simulated paleoclimate data, respectively.

Here, we assigned KG climates following the revised KG climate classification of Peel et al (34) and using Wong Hearing et al.’s (77) R script.

Geographic Distances Moved from Root to Tip (D_PATHWISE).

We obtained a measure for the geographic distance that each primate species moved across the globe from the location of the most recent common ancestor (MRCA) of crown Primates (Dataset S12). For this, we added up the root-to-tip branchwise dispersal distance along phylogenetic paths, i.e., across all phylogenetic branches that link the MRCA with each tip in the median tree. First, we calculated the geographic distance per phylogenetic branch, using the Great Circle distance. For this measure, we used the median of the coordinates inferred with the Geo model (phylogenetic nodes), the median of inferred paleocoordinates (fossils) or the within-polygon centroids (extant species). Second, we added up the branch-wise distances along the paths linking the MRCA to the tip for each fossil and extant species.

Node Count from Root to Tip (NC_PATHWISE).

To study speciation, we used node count (NC) along phylogenetic paths (Dataset S12). There are alternative nonmodel-based tip-rate metrics used to study speciation rate, such as the inverse of equal splits (ES) or the inverse of terminal branch length (TB) (36). However, we preferred to use NC as it has been shown to be less influenced by potential biases and sources of uncertainty associated with branch-length estimation from empirical data (79). NC captures the average speciation rate over the entire phylogenetic path and weights all branch lengths equally. We did not use speciation metrics estimated from time-varying birth–death diversification models owing to the erroneous inference of the general diversification patterns when the variation in rate of sequence evolution is not properly considered during the time-tree inference (80). Additionally, it has also been shown in the context of phylogenetic regressions that NC is the response variable that exhibits the highest statistical power when compared to regressions using ES or TB as speciation metrics (36).

Local Precipitation and Temperature Rates of Change (LP_RATE and LT_RATE).

To get the total amount of change in temperature and precipitation along phylogenetic paths we used a three-step approach. First, we extracted the total local precipitation per year (LP) and the year-mean local temperature (LT). These data were extracted for the posterior coordinates of each phylogenetic node, and from the paleocoordinates for fossils and the geographic centroid for extant species (SI Appendix, Fig. S8). Second, we calculated the absolute difference of the median LP and LT, between the ancestral and descendant node for each phylogenetic branch. Third, we summed the per branch absolute differences between LP and LT along the paths that link the common ancestor with each fossil and extant species. We divided this variable by the total time along each path (i.e. path length) given the median time-calibrated phylogenetic tree.

Phylogenetic Regressions.

We performed two sets of phylogenetic regression analyses to study the correlates of D_PATHWISE and NC_PATHWISE. To study the correlates of D_PATHWISE, we used Bayesian PGLS regression models, estimating Pagel’s lambda in BayesTraits v5.

We first evaluated the effect of time and global average temperature (GT) on D_PATHWISE. Time was obtained from the path length of the time-calibrated median phylogenetic tree (Dataset S12). GT is available across the Phanerozoic at a resolution of 1 My and was obtained from Scotese et al. (51). We extracted the GT from all the paleocoordinates (given their FAD and LAD) for fossil tips, and from the centroid within the polygon for extant species (Dataset S12). We matched the fossils and extant coordinates with the GT values, according to their age.

Then, we evaluated the effect of local precipitation (LP) and local temperature (LT) on D_PATHWISE. The LP and LT variables correspond to the annual precipitation and annual mean temperature values extracted from all the paleocoordinates for fossil tips (given their FAD and LAD), and from the centroid within the polygon for extant species (Dataset S12). The annual precipitation and annual mean temperature for fossil tips were extracted from the HadCM3BL-M2.1aD simulation layers. The annual precipitation and annual mean temperature for extant species was extracted from the WorldClim version 2 data. Finally, we evaluated the effect of LP_RATE and LT_RATE on D_PATHWISE.

To study the correlates of NC_PATHWISE we performed PGLMMs using the MCMCglmm R package 2.35 (37). We used a Poisson distribution to model NC_PATHWISE. As in the PGLS analyses D_PATHWISE, we tested the effect of time, GT, LP, LT, LP_RATE, and LT_RATE.

For the PGLS regression we ran 6,000,000 iterations, sampling every 50,00 iterations and discarding the first 1,000,000 iterations as burn in. Statistical significance of predictor variables was estimated based on a P_MCMC metric. This metric is based on counting the percentage of regression parameters that are higher (or lower) than zero in the posterior distribution. When the regression parameter is higher (or lower) than zero over 95% of the posterior distribution, then the predictor variable is statistically significant. For PGLMM regressions we also ran 6,000,000 iterations, sampling every 50,000 iterations and discarding the first 1,000,000 iterations as burn in. Regression coefficients were judged to be significant according to the P_MCMC metric estimated by the “MCMCglmm” R function.

Supplementary Material

Appendix 01 (PDF)

Dataset S01 (GZ)

Data, Materials, and Software Availability

All study data are included in the supporting information. The input map file for BayesTraits is available at Figshare (https://doi.org/10.6084/m9.figshare.29608814.v2) (74).