Extant species fail to estimate ancestral geographical ranges at older nodes in primate phylogeny

Abstract

A clade’s evolutionary history is shaped, in part, by geographical range expansion, sweepstakes dispersal and local extinction. A rigorous understanding of historical biogeography may therefore yield insights into macroevolutionary dynamics such as adaptive radiation. Modern historical biogeographic analyses typically fit statistical models to molecular phylogenies, but it remains unclear whether extant species provide sufficient signal or if well-sampled phylogenies of extinct and extant taxa are necessary to produce meaningful estimates of past ranges. We investigated the historical biogeography of Primates and their euarchontan relatives using a novel meta-analytical phylogeny of over 900 extant (n= 419) and extinct (n = 483) species spanning their entire evolutionary history. Ancestral range estimates for young nodes were largely congruent with those derived from molecular phylogeny. However, node age exerts a significant effect on ancestral range estimate congruence, and the probability of congruent inference dropped below 0.5 for nodes older than the late Eocene, corresponding to the origins of higher-level clades. Discordance was not observed in analyses of extinct taxa alone. Fossils are essential for robust ancestral range inference and biogeographic analyses of extant clades originating in the deep past should be viewed with scepticism without them.

1. Introduction

Geography plays a critical role in shaping a clade’s macroevolutionary history [1–5]. However, in order to understand the distinct contributions of vicariance, dispersal and in situ diversification to biogeographic patterns within a clade, comparative biologists must first infer where the ancestors of the clade occurred [6]. As a result, considerable effort has gone into the development of statistical methods for estimating historical biogeographic ranges using comparative data and time-scaled molecular phylogenies [7–12].

Intuitively, fossil data may provide evidence of a clade’s past geographical range that cannot be obtained from extant taxa alone [13–15]. The inclusion of extinct taxa in phylogenetic biogeographic analyses, either as terminal taxa in their own right [16] or as priors on the areas occupied at nodes [17], has uncovered evidence for ancient vicariance [16] and dispersal [18,19] events and yielded ancestral range estimates that fall outside the distribution of Recent taxa [17,20–22]. However, the difficulty in obtaining comprehensive sampling of fossils in phylogenetic analyses means that most past biogeographic studies that have investigated the impact of combining extant and extinct taxa have either focused on inferring the ancestral ranges of lower-level clades, such as families or genera, or used sparse sampling of fossils to investigate the historical biogeography of larger clades. As a result, it remains unclear whether complete sampling of extant higher-level clades is sufficient to infer their ancestral biogeographic ranges, or whether dense fossil sampling is needed to obtain reliable ancestral range estimates.

The mammalian order Primates present an opportunity to evaluate the importance of fossil data for ancestral biogeographic range inference at basal nodes of molecular phylogeny. Living primates are geographically widespread and can be found across Africa, Madagascar, Asia, and Central and South America. Biogeographic analyses based on molecular phylogeny [23] have suggested an Asian origin for crown group Primates, as well as crown clades such as Strepsirrhini (lemurs and lorises), Catarrhini (Old World Monkeys and apes) and Anthropoidea (South and Central American monkeys and the catarrhines). However, the first primates appeared near-simultaneously in Europe, Asia and North America [24–29] and palaeontological evidence indicates that their geographical distribution has varied substantially over time (electronic supplementary material, figure S1). Correspondingly, estimates of where major clades of primates originated and how they came to occupy their current geographical distributions differ substantially between studies based on extant taxa alone and those that consider information from the fossil record. For example, palaeontological evidence suggests that crown group strepsirrhines [30,31] (but see [32]) and catarrhines [33,34] originated in Africa, while the origins of crown and stem group Primates may have occurred in Eurasia [27,31] or North America [27], depending on which stem-taxa are sampled or whether non-primate outgroups are considered [35,36]. The placement of a number of African taxa along the anthropoid stem lineage further suggests that the ancestor of the South American platyrrhines may have colonized that continent from Africa [34,37], rather than from Asia as implied by phylogenies of extant taxa alone [23].

Here, we generate the most comprehensive estimate of primate phylogeny produced to date, to our knowledge, using a meta-analytical approach [38,39], and use the time-scaled topology to re-evaluate hypotheses for the origins of major primate sub-clades using probabilistic biogeographic modelling. We compared historical biogeographic inference at key nodes in primate phylogeny based on extant-only, extinct-only, and complete taxon sampling, both with and without the inclusion of non-primate outgroups. Our results highlight that the importance of fossil data for ancestral character state inference applies also to historical biogeographic analyses, but also suggest some caveats to uncritical integration of fossil data into phylogenetic tests.

2. Material and methods

(a) . Primate meta-analytical phylogeny

We used the metatree pipeline ([38,39]; electronic supplementary material, figure S2) to generate a comprehensive meta-analytic phylogeny for living and extinct primates, as well as their dermopoteran and scandentian sister lineages. This approach is similar to traditional matrix representation with parsimony (MRP) supertree methods but uses a formal set of rules to accommodate phylogenetic uncertainty in source studies, remove redundant datasets or down-weight datasets based on similar underlying matrices, and reconcile taxonomic information to a common source [39]. Our final dataset comprised 140 character matrices (138 morphological, one palaeoproteomic, one genetic) from 116 sources. Full details of data collection, curation and analysis are provided in the electronic supplementary material. A long-standing concern regarding MRP supertrees is a tendency to recover clades that are not present in any of the source studies [40,41]. To evaluate clade support, we computed a modified version of the reduced qualitative support (rQS) metric [42,43] that is appropriate for metatrees (see the electronic supplementary material). This metric ranges from −1, indicating that a bipartition in the metatree is contradicted by all characters in all source trees, and +1 indicating that the bipartition is consistent with all characters in all source trees.

(b) . Tip-dating

We used $\mathtt{B}\mathtt{E}\mathtt{A}\mathtt{S}\mathtt{T}$ v. 2.6.6 [44] to timescale, our majority rule consensus metatree under a relaxed uncorrelated lognormal molecular clock and fossilized birth–death (FBD) tree prior. Full details of model choices, prior distributions, Markov chain Monte Carlo settings and diagnoses of chain convergence are provided in the electronic supplementary material. We used protein-coding mitochondrial genes (7227 bp) for 420 extant and recently extinct species, partitioned into codon positions 1 + 2 and position 3, with each assigned $\mathrm{GTR}+\Gamma$ models of sequence evolution following [45] to provide relative branch length information. For downstream analyses, we obtained the median tree, based on the Kendall–Colijn distance metric [46], from the posterior sample using the $\mathtt{t}\mathtt{r}\mathtt{e}\mathtt{e}\mathtt{s}\mathtt{p}\mathtt{a}\mathtt{c}\mathtt{e}$ library [47].

(c) . Historical biogeography analyses

We modelled the evolution of biogeographic ranges in Primates using a dispersal, extinction, cladogenesis (DEC) model [8,9]. In this framework, range shifts must occur via dispersal to a new area (i.e. an increase in the number of areas occupied) followed by extinction in the original area, with an instantaneous rate matrix determining the rates of dispersal and extinction. We therefore also considered a version of the DEC model with jumps (DEC + J) that allows range shifts to occur via sweepstakes dispersal [12]. Model parameters were estimated in a maximum-likelihood framework using the $\mathtt{B}\mathtt{i}\mathtt{o}\mathtt{G}\mathtt{e}\mathtt{o}\mathtt{B}\mathtt{e}\mathtt{a}\mathtt{r}\mathtt{s}$ [48] package for $\mathtt{R}$, with model selection based on comparison of Akaike weights. Following previous work on the historical biogeography of Primates [23–28,31,49], we used a modified, quasi-continental level coding, recognizing five geographical areas: North America (including Central America and the Caribbean), South America, Eurasia (including India), Afro-Arabia (including mainland Africa and the Arabian Peninsula) and Madagascar (electronic supplementary material, table S1). Although the Indian sub-continent was home to an endemic primate fauna during its early Cenozoic history as an island continent [50], we do not model it as a distinct region here for consistency with previous work, though such analyses would be fruitful in the future. To aid in comparison of our results to those of Springer and colleagues [23], a maximum of two areas were permitted for each node in all analyses, we assumed that transitions between areas were equiprobable, regardless of distance and transition probabilities were modelled as time-constant. To investigate whether the inclusion of non-primate euarchontan outgroups affects historical biogeographic inference [27,35], we repeated all analyses with and without scandentian and dermopteran taxa included in the tree.

(i) . Comparing ancestral range inferences based on extant and extinct taxa

Biogeographic analyses are generally conducted using molecular phylogenies of extant taxa, without consideration of fossil data. Therefore, we first pruned extinct taxa from our metatree, and generated maximum-likelihood estimates of the ancestral range for each node in the phylogeny of extant taxa alone. To investigate whether robust sampling of extinct taxa changed estimates of ancestral range evolution in primates, relative to the extant-only analyses, we then performed a second set of analyses using the complete time-scaled metatree. We compared the results of biogeographic inference with and without fossils in two ways. First, we counted the proportion of mismatched inferences for shared nodes between complete and extant-only trees using the $\mathtt{c}\mathtt{o}\mathtt{m}\mathtt{p}\mathtt{a}\mathtt{r}\mathtt{e}\mathtt{P}\mathtt{h}\mathtt{y}\mathtt{l}\mathtt{o}$ function in $\mathtt{a}\mathtt{p}\mathtt{e}$ [51]. Previous analyses have shown a strong influence of fossil data on ancestral range estimates at deep nodes in molecular phylogenies, particularly where extant representatives no longer occupy areas that are well represented in the fossil record (e.g. [21, 22]). To quantitatively evaluate whether mismatches were randomly distributed across shared nodes or whether there is a significant association between node depth and estimation mismatch, we also performed a logistic regression of congruence between node estimates on node age, using the $\mathtt{g}\mathtt{l}\mathtt{m}$ function in the $\mathtt{s}\mathtt{t}\mathtt{a}\mathtt{t}\mathtt{s}$ library [52].

Most discussions surrounding the importance of fossil data in macroevolutionary and biogeographic inference have focused on how parameter estimates change when fossil data are included in molecular phylogenies. However, it is unclear whether using data from fossil taxa alone results in the same kinds of systematic errors and biases as the use of extant taxa alone. If not, then it might be concluded that historical biogeographic analyses would be best served by a focus, where possible, on erecting phylogenetic hypotheses for extinct rather than extant representatives of clades. To evaluate this possibility, we conducted a third set of analyses using a metatree pruned to only extinct taxa. We compared ancestral range estimates to those from the complete metatree in the same way as above. If extinct taxa provide more signal than extant taxa regarding ancestral ranges at deep nodes in a phylogeny, we would expect to recover no significant association between node depth and mismatched inference at common nodes between the complete and extinct-only trees.

(ii) . Time-stratified rates and area adjacency

The assumption of time-constant probabilities of dispersal between continents is contradicted by abundant geological evidence (reviewed in [53]). It is possible that discrepancies between historical biogeographic analyses based on extant taxa alone and those based on fossil evidence could be ameliorated by modelling time-varying dispersal probabilities or by limiting dispersal to between adjacent continents. We therefore repeated our analysis of extant euarchontan historical biogeography using (i) a time-stratified dispersal matrix that allowed for variable dispersal probabilities between regions based on geological evidence, (ii) time-stratified adjacency matrices that limited the potential colonizers of new areas to those coming from adjacent continents that share a land connection [9], and (iii) time-stratified dispersal matrices and adjacency matrices simultaneously. Full details of the time-stratified model are provided in the electronic supplementary material.

3. Results

(a) . Meta-analytical phylogeny

The final MRP matrix contained 934 species and specimen-level operational taxonomic units. The strict consensus metatree is generally well-resolved, but lacks resolution in some key clades, such as Catarrhini (electronic supplementary material, figure S3). Inspection of the Adams (electronic supplementary material, figure S4) and majority rule consensus (MRC; electronic supplementary material, figure S5) shows that this is owing to the presence of a few rogue taxa, and that all major primate clades are otherwise recovered. On this basis, we used the MRC for time-scaling purposes, with polytomies resolved by molecular data or under the FBD model. A full description of the topology is provided in the electronic supplementary material, but we note here that the relationships recovered in our metatree are broadly consistent with previously published estimates of primate phylogeny and nodes in the MRC are generally well-supported; 92% of nodes receive positive rQS scores, indicating that they are supported by at least one source study, while of the 6% (n = 43) of nodes that received negative rQS scores, a majority (n = 26) involved pairs of taxa, often within the same genus. Eleven nodes received ambiguous (rQS = 0) support (electronic supplementary material, figure S6).

(b) . Tip-dating

Tip-dating yielded mean divergence time estimates for crown euarchontan nodes that are generally younger than those based on node-calibrated molecular phylogenies, and with narrower credible intervals (figure 1; electronic supplementary material, S7; table 1). The basal splits within Euarchonta (scandentians from primatomorphs) and Primatomorpha (dermopterans from primates) occurred during the latest Cretaceous (95% highest posterior density intervals exclude the Cenozoic), and the origin of crown group Primates is restricted to the early Palaeogene, approximately 63 Ma, prohibiting the possibility that this clade originated in the Late Cretaceous [23,45] (table 1). Divergence date estimates for younger clades are more comparable with previous estimates (table 1).

Figure 1.

Overview of the Euarchontan meta-analytical phylogeny (median tree from tip-dating analyses, $n=902\hspace{0.17em}\mathrm{species}$). Sub- and Infraorders are labelled, as well as paraphyletic stem clades. Stem anthro., stem anthropoids; O − formes, Omomyiformes (including Tarsiidae); Loris., Lorisiformes; stem streps., stem strepsirrhines; derm, Dermoptera; scan, Scandentia. For more detailed trees, including tip-labels, see the electronic supplementary material and Dryad repository [54]. (Online version in colour.)

Table 1.

Mean (95% credible interval) divergence dates for major Primate crown clades as estimated in this study and two previously published, large scale phylogenies. (Results from analysis ‘A’ reported from dos Reis [45]. All ages in millions of years ago (Ma).)

clade	this study	Springer et al. [23]	dos Reis et al. [45]
Euarchonta	73.91 (70.22–77.72)	—	91.75 (81.91–104.14)
Primatomorpha	72.43 (69.39–75.48)	—	82.29 (76.58–88.68)
Primates	63.71 (61.20-66.17)	67.84 (60.99-76.72)	74.44 (70.03–79.17)
Strepsirrhini	51.96 (47.53–56.64)	54.23 (48.75–57.22)	62.71 (58.77–66.83)
Haplorhini	60.24 (58.29–62.46)	61.16 (57.62–69.59)	70.63 (66.52–75.02)
Anthropoidea	41.02 (38.45–43.75)	40.60 (33.55–49.48)	45.04 (41.81–48.33)
Platyrrhini	22.83 (21.40–24.42)	23.32 (19.26–27.49)	25.32 (23.60–27.49)
Catarrhini	30.03 (27.89–32.13)	25.07 (19.67–32.83)	32.58 (30.39–35.11)
Cercopithecidae	20.27 (18.16–22.24)	13.17 (8.93–18.27)	16.17 (19.88–25.43)
Hominoidea	21.55 (18.05–24.83)	17.36 (12.44–23.90)	24.01 (22.21–26.06)
Hominidae	17.32 (14.69–19.64)	15.09 (11.04–20.08)	20.29 (18.76–21.97)

(c) . Historical biogeographic analyses

(i) . Comparing ancestral range inferences based on extant and extinct taxa

Parameter estimates for the DEC and DEC + J models fitted to the metatrees of extant taxa are consistent with a dispersal and cladogenetic dominated biogeographic history, with a negligible contribution of within-area extinction to biogeographic range evolution (table 2). DEC + J is moderately preferred over DEC for the extant-only tree, receiving approximately two-thirds of model weight. Ancestral range inference is broadly concordant with previous estimates derived from molecular phylogeny (figure 2a; electronic supplementary material, table S2). Regardless of the model (DEC or DEC + J), or whether outgroups were included or excluded, we infer Eurasian origins for most major euarchontan and primate crown clades. Notable divergences from this pattern include the ancestor of crown group Stepsirrhini, which is inferred to have occurred in Eurasia and Madagascar with dispersal to Africa in the ancestor of crown Lorisiformes, and the ancestor of crown group anthropoids, which is estimated to have spanned Eurasia and South America (electronic supplementary material, figures S8-11). When non-primate outgroups are excluded (electronic supplementary material, figures S8 and S9) we recover Eurasia + Madagascar as the most likely ancestral range for crown Primates based on extant species alone (p[Eurasia + Madagascar] = 0.78, p[Eurasia] = 0.069).

Table 2.

Log-likelihoods (LnLk) and dispersal (d), extinction (e) and jump-dispersal (j) parameter estimates for DEC and DEC + J analyses with and without outgroups across the three sampling regimes; extant taxa only, fossil taxa only and all taxa.

		DEC				DEC + J
clade	sample	LnLk	AICw	d	e	LnLk	AICw	d	e	j
Euarchonta	extant	−91.04	0.33	1.17 × 10−3	0	−89.33	0.67	9.96 × 10−4	0	5.07 × 10−4
	fossil	−386.04	0.00	4.52 × 10−3	2.58 × 10−3	−314.39	1.00	2.70 × 10−4	0	1.73 × 10−2
	complete	−459.78	0.00	2.93 × 10−3	7.03 × 10 −4	−403.02	1.00	6.33 × 10−4	0	7.97 × 10−3
Primates	extant	−87.38	0.36	1.26 × 10−3	0	−85.79	0.64	1.08 × 10−3	0	4.83 × 10−4
	fossil	−376.79	0.00	4.61 × 10−3	2.16 × 10−3	−307.27	1.00	2.89 × 10−4	0	1.73 × 10−2
	complete	−448.20	0.00	3.08 × 10−3	5.42 × 10−3	−394.21	1	6.92 × 10−4	0	7.98 × 10−3

Figure 2.

Ancestral ranges estimates for higher-level euarchontan phylogeny under the best-fitting DEC + J model of range evolution. The top row shows ancestral range estimates for (a) extant taxa alone, (b) extant plus extinct taxa, and (c) fossil taxa alone assuming time-constant dispersal rates and equal probabilities between all areas. On the bottom row, ancestral range estimates for extant primate phylogeny are shown under (d) a time-stratified rate matrix and (e) a time-stratified adjacency matrix. Pie charts indicate ranges that span two areas; ? indicate nodes that yield different ancestral range estimates when alternate topologies are considered for key fossil lineages (see Results and Discussion). (Online version in colour.)

Analysis of the complete dataset yielded larger estimated dispersal rates than did the extant-only analyses (table 2). For the DEC model, we also recovered non-zero estimates for extinction rates, though DEC + J accounted for this by elevating jump dispersal rates instead. Model comparisons indicate that DEC + J is overwhelmingly supported for this dataset. We infer that Euarchonta had a Eurasian + North American origin (figure 2b; electronic supplementary material, figures S12 and S13), but crown Primatomorpha and total group Primates are inferred to have originated exclusively in North America. Although we infer a Eurasian origin for crown Primates, ancestral range estimates for other crown clades diverged from those in the extant-only analysis (figure 2b; electronic supplementary material, table S2). Crown Strepsirrhini, Lorisiformes, Catarrhini, Cercopithecoidea, Homininae and Hominini are all inferred to have had at least partially Eurasian origins based on molecular phylogeny alone, but all are inferred to be exclusively Afro-Arabian based on the complete tree, while crown group Haplorhini is inferred to have originated in North America. A wide-ranging ancestor that occurred in Afro-Arabia and Madagascar is inferred for crown Lemuriformes + Chiromyiformes, rather than a single Malagasy origin of the clade.

Bizarrely, we infer that the crown anthropoid ancestor (Catarrhini + Platyrrhini) occurred in South America rather than Eurasia or Afro-Arabia, while total group catarrhines are inferred to have been distributed across Afro-Arabia and South America. We investigated whether the recovery of some South American fossil primates outside of total group Platyrrhini could be the cause of these results by individually pruning them and rerunning ancestral range inference. Recovery of Parvimico materdei, Dolichocebus annectens (probably in error, see the electronic supplementary material) and Ucayalipithecus perdita as stem anthropoids had no influence on ancestral range inference at the crown anthropoid or total group catarrhine node (electronic supplementary material, figures S14 and S15). However, recovery of CPI-6487, an as-yet un-named ?m3 from the late Eocene of Peru [37] as a stem catarrhine did have an impact, and its removal resulted in an inferred Afro-Arabian origin for both crown Anthropoidea and total group Catarrhini (figure 2b; electronic supplementary material, figure S16).

The proportion of node estimates that differ between pairs of analyses is small overall (13–14 nodes, 3.2–3.3%). However, mismatches qualitatively appear to correspond to the origins of named higher-level clades that are of particular interest to many evolutionary biologists and palaeontologists (figure 2). Indeed, across all models and outgroup treatments, mismatches are significantly associated with deeper divergence events (Z = −5.65 to −5.69; p < 0.001; figure 3a; electronic supplementary material, figure S17A and table S3), and the probability of congruent inference between analyses dips below 0.5 during the late Eocene.

Figure 3.

Ancestral range estimates derived from extant taxa alone (a) exhibit a significant, age-based discordance with those estimated from the metatree of extant and extinct species. Ancestral range estimates derived from fossil taxa alone (b) do not exhibit this bias and are more consistent with estimates derived from extant and extinct species. Results for analyses DEC + J with non-primate outgroups are shown here, and are representative of all node comparison tests (electronic supplementary material). Points are jittered to emphasize density. (Online version in colour.)

As with the complete tree, we recovered higher dispersal or jump dispersal rates for the extinct-only tree than for the extant-only tree, coupled with non-zero extinction rates for the DEC, but not DEC + J model (table 2). Unlike the analyses of extant species, ancestral range estimates from analyses of fossil species alone (electronic supplementary material, figures S18 and S19) were largely concordant with results obtained from analyses using the complete metatree (figure 2b; electronic supplementary material, table S2) and there is a weak but insignificant positive relationship between node age and the probability of a match between the fossil-only datasets and the complete metatree, regardless of outgroup treatment and model choice (Z = 1.46 to 1.5, p > 0.1; figure 3b; electronic supplementary material, figure S17B and table S3).

(ii) . Time-stratified rates and area adjacency

Analyses of extant euarchontan biogeography with time-stratified dispersal rate and/or adjacency matrices did not bring ancestral range estimates in line with those based on the complete or extinct-only datasets. DEC + J is more strongly supported than DEC under time-stratified rates (AIC_w = 0.75) and ancestral range estimates at most nodes are consistent with those based on extant taxa alone using a single rate matrix. However, the implausible range of Eurasia + Madagascar is smeared back to the crown euarchontan node, regardless of whether a DEC or DEC + J model was used (figure 2d; electronic supplementary material, figures S20 and S21). Analyses employing a time-stratified adjacency matrix failed to run under DEC as the lack of adjacency between Madagascar and any other landmass prevents dispersal to or from this area. Allowing for jump dispersal overcomes this issue (figure 2e; electronic supplementary material, figure S22), but we infer a Eurasian ancestry along the backbone for crown Euarchonta, Primatomorpha, Primates, Haplorhini, Anthropoidea and Catarrhini, with jump dispersal from Eurasia explaining most shifts. Madagascar is the maximum-likelihood estimate of the area of origin for crown Strepsirrhini (p = 0.3422), though this is only marginally preferred over an Afro-Arabian (p = 0.3416) or Eurasian (p = 0.316) origin. Combining time-stratified rate matrices and adjacency matrices yielded results that blend those recovered using the two in isolation, but with some additional novelties. For example, we recovered Eurasia + Afro-Arabia as the most likely range along the backbone of Euarchontan phylogeny, with an Afro-Arabian + Malagasy ancestor inferred for crown strepsirrhines (electronic supplementary material, figure S23).

4. Discussion

Concerns about the susceptibility of ancestral state estimation to model violations have long been voiced in studies of phenotypic trait evolution [55–57], where even a few well-placed fossils can have a dramatic impact on parameter estimates [58–61]. Recent historical biogeographic studies have yielded similar findings (e.g. [17,22]), strongly implying that ancestral range estimates based on extant taxa alone may fail to accurately capture patterns of biogeographic range evolution in the early phase of clade history owing to extinction in previously occupied areas [21]. Our results are consistent with this observation (figure 2). For example, we infer an Afro-Arabian, rather than Eurasian–Malagasy, origin for crown group Strepsirrhini (figure 2a–c), a result that is probably driven by the placement of several African taxa (Azibiidae, Djeblemuridae) as stem-strepsirrhines and by the placement of Karanisia as a stem lorisiform [31,49]. Similarly, our analyses, which incorporate abundant stem apes, cercopithecoids and catarrhines confirm that crown Catarrhini originated in the mid-Oligocene of Africa, as has previously been considered well-established based on fossil evidence [28,62,63], rather than in Eurasia (figure 2a,b). Other range estimates simply could not be recovered without fossil data. For example, we recover a well-supported (p = 0.94) North American origin for Haplorhini, largely owing to the inclusion of numerous omomyids that are recovered along the tarsiiform and anthropoid stem lineages. However, although the impact of fossil data on ancestral state estimates is well understood in general, ours is the first study that we are aware of to have statistically demonstrated that the congruence between biogeographic range estimations based on extant versus extinct taxa declines significantly with time. This finding has substantial implications for comparative biologists interested in understanding how geography has shaped the evolution and ecology of clades, but is not without caveats.

Although the inclusion of extinct taxa can help resolve ancestral states at deep nodes in phylogeny, fossil data should not be naïvely viewed as a panacea. Our analyses yield some confusing results that can be explained, but that suggest the need for better phylogenies of currently recognized taxa. Most strikingly, we infer an implausible South American ancestry for crown group anthropoids based on our complete and extinct-only metatrees, with subsequent dispersal to Africa implied for the ancestor of total group Catarrhini (figure 2b,c). Molecular evidence suggests a Eurasian origin for crown anthropoids [23], while fossil evidence has tended to indicate an Afro-Arabian origin, most likely from a parapithecid ancestor [34]. The fact that our unusual result arises entirely owing to the placement of an as-yet unnamed fossil that is represented by a single tooth of uncertain locus [37] is concerning and serves to highlight that the influential nature of fossil taxa may, on occasion, act as an impediment to robust macroevolutionary inference. A number of new Palaeogene and earliest Neogene South American fossil primates have been recovered outside of Platyrrhini and along the anthropoid stem in recent years [34,37,64], but the phylogenetic status of these taxa is highly uncertain owing to the very limited fossil material, typically a single tooth, and the preliminary nature of some analyses. Indeed, a recent tip-dating analysis [29] recovered the Early Miocene primate Parvimico as the sister to all other platyrrhines, rather than as the sister to crown anthropoids. We recommend that macroevolutionary researchers think carefully about whether to include fossils of such uncertain status in comparative phylogenetic analyses. However, it is also worth noting that, while inference at the crown anthropoid node is strongly influenced by a South American fossil of uncertain status, we also found that recovery of the Asian eosimiids and amphipithecids as paraphyletic sister lineages to crown-group anthropoids exerts a strong effect on biogeographic range estimates. Previous studies have tended to view ampithecids as either stem anthropoids, outside the African parapithecoids [34,65], or even as stem haplorhines [66]. Deletion of these clades yields an Afro-Arabian origin for anthropoids with subsequent colonization of South America in platyrrhines (electronic supplementary material, figure S24), even without removal of ‘non-platyrrhine’ South American fossils like CPI-6487. Resolution of the placement of eosimiids and amphipithecids within haplorhine phylogeny is therefore also of critical importance for resolving the biogeographic history of the anthropoid radiation.

Ree & San Martín [67] have voiced statistical concerns regarding DEC + J and the appropriateness of model comparison, though these claims have recently been disputed [68]. We found that although DEC + J did indeed tend to explain most dispersal and within-area extinction as jump dispersal [67], support for this model over DEC increased with the addition of data from extinct taxa (table 2). Much like island colonization [12], explicit modelling of sweepstakes dispersal in addition to range expansion may be more appropriate for continental-level biogeographic analyses, especially in deep time. Furthermore, the fact that analyses employing a geologically reasonable time-stratified adjacency matrix failed to run under DEC owing to the inability of Primates to colonize or disperse from Madagascar suggests that this model is statistically inadequate for data at continental scales where at least some sweepstakes dispersal must have occurred. Comparison of ancestral range estimates under the two models yields further support for the idea that DEC + J is a more appropriate model for intercontinental biogeography. Recent work has suggested that Chiromyiformes (aye-ayes) and Lemuriformes colonized Madagascar independently from the African mainland [49]. We find no support for this hypothesis owing to differences in the topological positions of Propotto and Plesiopithecus in our metatree, but even after altering chiromyiform relationships to be consistent with [49] we only recover support for dual colonizations of Madagascar under a DEC + J model (electronic supplementary material, figures S25 and S26). We suspect that, when dispersal events occur at low rates, the DEC model tends to smear ancestral ranges back in time, which would also explain the bizarre inference of a Malagasy + Eurasian origin for Primates in the extant-only analysis, and why this problem may be exacerbated by the use of time-stratified rate matrices (electronic supplementary material, figure S20). DEC + J avoids this effect at the cost of a tendency to explain all events as jump dispersal [67]; a reasonable compromise for analyses at the geographical scale used here. Future work that more finely delimits the biogeographic regions that primates occupy (e.g. [50,69–71]) may increase dispersal rates and it would be interesting to evaluate whether this increases support for DEC, relative to DEC + J, and yields more precise estimates of ancestral ranges.

We acknowledge that variation in the quality of the primate fossil record in time and space might affect our inferences. To partially address this concern, we compared the number of primate-bearing collections within in each biogeographic region over the Cenozoic to the total number of mammal-bearing collections (a taphonomic control [72]) using data from the Paleobiology Database. We found no evidence of substantial geographical bias in the primate fossil record; for all time periods lacking primate-bearing collections on a particular continent there are still a number of mammal-bearing collections, suggesting that the lack of primates represents a true absence rather than poor preservation potential. This interpretation is strengthened by the finding that for most, though not all time periods in which at least one mammal bearing collection is available, we still recover primates (such as in the Palaeocene of Africa), suggesting that when primates were present in a region, they had a relatively high probability of entering the fossil record. Nonetheless, some regions possess better mammalian fossil records for important time periods than do others and a single new fossil discovery could change our understanding of primate relationships and historical biogeography. An advantage of our meta-analytical approach is that new fossil discoveries can easily be incorporated into an updated estimate of primate phylogeny for biogeographic analysis [38,39].

The relatively small proportion of ancestral range estimates that differed between analyses based on extant taxa alone and extant and fossil data in combination (approx. $3\text{\%}$) may give some readers hope that historical biogeographic analyses are, for the most part, robust to the exclusion of fossil data. To some extent, this result is probably owing to the geographical scale at which our analyses were conducted, and it remains to be seen if this result would stand up to scrutiny in the face of more refined geographical codings. Regardless, the significant negative relationship between node age and consistency of ancestral range inference (figure 3a) suggests that comparative biologists should exercise extreme caution when drawing conclusions about clade origins from historical biogeographic analyses. Of particular concern is the observation that many of these nodes correspond to the origins of named higher taxa (figure 2), which are often the focus of historical biogeographic interest. The general consistency between results for our extinct-only and complete metatree analyses, along with the lack of a relationship between node age and consistency (figure 3b), reinforces the notion that fossils contribute more information to ancestral state estimation at more basal nodes than do extant taxa [59–61,73], and that inferences based on extinct taxa should be preferred, when available.

(a) . Conclusion

The best estimate of historical biogeography will ultimately come from the analysis that has the most comprehensive sampling of taxa from a clade’s entire evolutionary history. Large phylogenies are more readily available for living taxa than for extinct taxa, largely owing to the inherent difficulty in constructing large morphological character taxon matrices. As such, analyses of historical biogeography are often limited to small samples of extinct taxa or, more commonly, extant taxa alone. Our analyses of a comprehensive metatree for living and fossil primates reinforce the importance of fossil data in phylogenetic macroevolutionary analyses, particularly when estimating ancestral character states [58–61,73,74] and should serve as a stark warning against over-interpretation of macroevolutionary parameters estimated from phylogenies of extant species alone. In fact, our results emphasize that, on a per-taxon basis, a well-chosen fossil provides more signal to estimate ancestral character states than an extant species [73]. Given a robust estimate of a clade’s phylogeny, we should have more confidence in biogeographic estimates for clades with no modern representatives such as trilobites [75] or non-avian dinosaurs [76,77], than we should for old extant clades with no fossil representatives. While acknowledging the uncertainty that always comes with fossil data, we suggest that, when confronted with a choice between data types, extinct species should always be viewed as more valuable than living species in phylogenetic analyses of historical biogeography.

Data accessibility

Additional methods and results supporting this article have been uploaded as part of the online electronic supplementary material. All datasets and scripts used in this study are available from the Project’s GitHub Repository (https://github.com/graemetlloyd/ProjectPlanetOfTheApes) and the Dryad Digital Repository https://doi.org/10.5061/dryad.4j0zpc8czhttps://datadryad.org/stash/share/H0NcPKfSQlwo5Y4dg3crGZly1Fr_c4W3Tc7YBWD2Yjw [54].

The data are provided in electronic supplementary material [78].

Authors' contributions

A.L.W.: conceptualization, data curation, formal analysis, investigation, methodology, visualization, writing—original draft, writing—review and editing; G.T.L.: formal analysis, investigation, methodology, writing—original draft, writing—review and editing; G.S.: conceptualization, data curation, formal analysis, investigation, methodology, visualization, writing—original draft, writing—review and editing.

All authors gave final approval for publication and agreed to be held accountable for the work performed therein.

Conflict of interest declaration

The authors declare no competing interests.

Funding

No funding has been received for this article.