Phase 1: Radiation of basal Cavioidea s.s.

The most ancient records of Cavioidea s.s. come from late Oligocene beds of Patagonia (Deseadan SALMA [South American Land Mammal Age]; 24.5–29 Ma; [61], [62]). Two taxa are known from this age: Asteromys punctus [49], [52], [63] and Chubutomys simpsoni [52]. Both taxa are only known by fragmentary specimens composed by mandibular fragments. The relatively limited fossil record of Deseadan age does not reveal by itself the presence of a radiation during this time. However, when the phylogenetic hypotheses obtained in the parsimony analysis of combined dataset are calibrated against the geological age of fossil taxa, a basal cavioid radiation is revealed by the presence of four ghost lineages that must have originated in the Deseadan age (in addition to the two species recorded for this age; Fig. 1). These four ghost lineages are present in all the most parsimonious trees (MPTs) and extend the minimum age of origin of the lineages leading to Luantus initialis, Luantus minor, Chubutomys leucoreios, and the clade formed by Luantus propheticus and more derived cavioids back to the late Oligocene (Deseadan SALMA). All these forms appear later in the fossil record, in early Miocene beds referred to the Colhuehuapian SALMA (18–20.5 Ma; [61], [62]). Therefore, the calibrated phylogenies extend the evolutionary origins of these lineages at least 4 Myr before their first appearance in the fossil record. The extent of these ghost lineages is caused by the relatively derived phylogenetic position of the Oligocene taxon Chubutomys simpsoni (and the paraphyly of the genus Luantus; Fig. 1). The character support for such a derived position of Chubutomys simpsoni is, however, moderately low (as it takes two extra steps to place it basally within Cavioidea s.s. and bootstrap and jackknife frequencies of the two nodes basal to Chubutomys simpsoni are approximately 40% and 60%; Fig. 2; see Document S1).

The discovery of a basal cavioid radiation in the late Oligocene contrasts with previous interpretations about the early evolution of Cavioidea s.s. [52], [55], [64]. These traditional hypotheses postulated that the early phases of cavioid evolution proceeded through gradual changes from the late Oligocene to the early Miocene, reflected in the progressive appearance of Asteromys and the species of Luantus in the fossil record. Our phylogenetic analysis rejects this interpretation, placing Chubutomys simpsoni deeply nested within basal cavioids given the presence of apomorphic features of its dentition (e.g., protohypsodont condition of the teeth and the absence of mesofossetid in early ontogenetic stages). This implies a previously undetected radiation of all basal lineages of Cavioidea s.s. that must have occurred at least by the late Oligocene. The optimization of character transformations in these basal nodes of Cavioidea s.s. indicates that major evolutionary novelties must have appeared in this late Oligocene radiation, including changes from a mesodont to a protohypsodont dentition and other dental modifications such as the appearance of cement and enamel discontinuities and the loss of fossettes/ids during ontogeny.

The ghost lineages provide a minimum estimate for the age of these basal nodes of Cavioidea s.s., so it is possible that the actual diversification of this group predated the Deseadan SALMA (24.5–29 Ma). The pre-Deseadan record of fossil rodents in South America however provides relevant information to evaluate this possibility. The youngest pre-Deseadan rodent assemblage is known from La Cantera horizon of Central Patagonia (Upper Puesto Almendra Member, Sarmiento Formation; 29.5–31.1 Ma; [62]), which has yielded numerous rodent specimens but none belonging to Cavioidea s.s. (see also Node 2 in Calibrated Nodes). The La Cantera rodents include plesiomorphic taxa of Cavioidea (Dasyproctidae?) and representatives of Octodontoidea and Chinchilloidea [65]. Earlier faunal assemblages with fossil rodents from South America are limited to Tinguiririca in Chile (31.5–37 Ma; [29], [61]) and possibly the Santa Rosa fauna of Peru (estimated as late Eocene or Oligocene; [66]). The fossil rodents from these localities are also plesiomorphic forms of Cavioidea. Furthermore, the recently discovered rodent fauna from the Yahuarango Formation (ca. 41 Ma; middle Eocene; [28]) has yielded the oldest known South American rodents, which are basal forms of Caviomorpha that are not closely related to Cavioidea. The absence of Cavioidea s.s. in all these pre-Deseadan rodent faunas is compatible with the hypothesis of a basal radiation of this group close to or in the late Oligocene.

Phase 2: Radiation of euhypsodont cavioids.

The euhypsodont condition represents the presence of teeth with continuous growth (lacking a root) and was one of the major evolutionary novelties in the history of Cavioidea s.s. The most ancient records of euhipsodont cavioids come from the early Miocene Santa Cruz Formation of Central Patagonia (Santacrucian SALMA; 15.7–16.5 Ma; [61], [67]–[69]). This unit has yielded thousands of specimens of cavioids, including the three youngest records of protohypsodont species (L. toldensis, Phanomys mixtus, P. vetulus) and the five oldest euhypsodont species of Cavioidea s.s. (Eocardia montana, E. fissa, E. excavata, Schistomys erro, S. rollinsi). The sudden appearance of multiple euhypsodont lineages at this time suggests the occurrence of an evolutionary radiation that not only includes the five above mentioned taxa but also three ghost lineages leading to slightly younger taxa of the Colloncuran SALMA (13,8–15.5 Ma; [70]): 1) Matiamys elegans, 2) E. robertoi, and 3) the clade formed by E. robusta and more derived cavioids (see Fig. 1). Thus, unlike the case of basal cavioids, the diversification of euhypsodont lineages is more evident in the number of species known from the Santa Cruz Formation than in the diversity inferred from ghost lineages. The morphological data provide strong support for the Santacrucian radiation of euhypsodont cavioids. The node formed by Eocardia and more derived cavioids and the node formed by Phanomys and more derived cavioids (see Fig. 1) have bootstrap and jackknife frequencies between 82% and 85% (Fig. 2; see Document S1). The Bremer support for these nodes is only moderate (Bremer  = 2) but forcing any of the pre-santacrucian cavioids (e.g., L. propheticus, L. minor) within this clade results in markedly suboptimal topologies (i.e., at least ten extra steps).

The exceptional Santacrucian fossil record and the sudden appearance of a high diversity of euhypsodont cavioids can be interpreted in two alternative ways. On the one hand, the fossil record could be actually capturing the early offshoots of a major radiation, characterized by the acquisition of a key evolutionary innovation (euhypsodonty), which has been interpreted as an adaptation to grazing. The acquisition of the euhypsodont condition may have been a critical innovation at this time given the major environmental changes recorded in Patagonia, which includes high volcanic activity related to the uplift of the Andes (Quechua phase; [71]), a change from grasslands and woodlands biomes, and a general drop in humidity and temperature [72], [73].

On the other hand, the exceptional Santacrucian fossil record could cause a Lagerstätten effect: the sudden first appearance of multiple lineages merely due to the presence of favorable conditions for fossilization. In this way, some lineages may have existed before the Santacrucian but were not captured by the fossil record just because the preservation potential of rodents in older sediments was unfavorable. A critical analysis of the cavioid fossil record in sediments older than the Santa Cruz Formation, however, lends support to the existence of a major radiation during the Santacrucian SALMA. The Pinturas Formation contains a sedimentary sequence that is slightly older than the Santa Cruz Formation, dated at 17.7 Ma [67]. The rodent fossil record in this unit is extremely rich, being the second-best record from the Oligocene-mid Miocene of Patagonia (after the Santacrucian). Out of the thousands of rodent specimens known from the Pinturas Formation, all the cavioid taxa are protohypsodont and there is no record of euhypsodont species in this large sample [55], [74]. Based on the high quality of the rodent fossil record of the Pinturas Formation, we believe that the sudden appearance of multiple euhypsodont species at the Santa Cruz Formation represents a true evolutionary radiation. Furthermore, the numerous cladogenetic events inferred to occur during the Santacrucian in the calibrated phylogeny (Fig. 1) are restricted to a very specific part of the cavioid tree, rather than distributed in different parts of the phylogeny. As noted by Calvin and Forey [75], such scenarios are more likely to result from a genuine radiation event. If the sudden diversity increase were due to a Lagerstätten effect, there would be no expectation to find all the new lineages clustered in one region of the phylogenetic tree.

Phase 3: Diversification of Caviidae.

The oldest caviid known is Prodolichotis pridiana [76] from the La Venta section of Colombia (La Victoria and Villa Vieja formations; Laventan SALMA, middle Miocene, 11.8–13.5 Ma; [77]). This taxon has been traditionally referred to either Dolichotinae [76] or Caviinae [78] given the presence of derived similarities shared with these two lineages (e.g., presence of the nasolacrimal foramen in the maxilla). However, the morphological data used in the combined phylogenetic analysis presented here unequivocally places P. pridiana as the sister group of the third major lineage of caviids: Hydrochoerinae (including Kerodon; Fig. 1). This position is supported by the presence of two unambiguous synapomorphies (i.e., p4 with three lobes [character 45] and frontals not convex [character 59]). This position is however weakly supported given that P. pridiana can be positioned as the sister group of dolichotines with a single extra step or basally within Caviidae with two extra steps. Similarly, bootstrap and jackknife frequencies of the node joining P. pridiana and Hydrochoerinae range between 5% and 7% (Fig. 2; see Document S1).

Slightly older fossils recorded in deposits of Colloncuran age represent successive sister taxa of Caviidae (e.g., Guiomys, Microcardiodon; see Fig. 1) so that based on the available information P. pridiana is the most ancient Caviidae. The phylogenetic position of P. pridiana implies that the diversification of Caviidae in its three major lineages must have occurred, at least, by Laventan times (11.8–13.5 Ma; [69], [79]). After this time, the caviid fossil record has an extensive gap (of at least 2.7 Myr) until the late Miocene Arroyo Chasicó Formation of Central Argentina (Chasicoan SALMA; 6.1–9.07 Ma), in which derived specimens of the three modern lineages of caviids are recorded (i.e., Caviinae, Dolichotinae, Hydrochoerinae; see Calibrated Nodes). The only caviid remain that fill the extensive gap between the Laventan and Chasicoan records is a single isolated tooth from the Río Frias Formation (Mayoan SALMA) that is possibly related to basal hydrochoerines [80] (see Calibrated Nodes).

Therefore, the known fossil record lacks adequate information for thoroughly understanding the timing and radiation of modern lineages of crown caviids. The basic evolutionary pattern that can be inferred from the available fossil record is that: a) the initial split of Caviidae in three major lineages must have occurred at least 11.8–13.5 Ma (based on the Laventan record of P. pridiana and its phylogenetic position) and b) that the three modern and morphologically distinct lineages of Caviidae (Hydrochoerinae, Dolichotinae, Caviinae) were already present, abundant, and diverse about 6.1–9.07 Ma (based on the derived caviid fauna of late Miocene Arroyo Chasicó Formation and the ghost lineages that can be inferred from their phylogenetic positions; see Fig. 1).

Four calibration constraints.

Owing to the incorporation of fossils in the phylogeny, these are the most tightly constrained analyses of the diversification timing of Caviidae (Fig. 4). The estimate of the time of diversification of Caviidae using gamma prior distribution yields a mean estimate of 18.8 Ma (leftmost blue circle in row 4 of Fig. 4; Table 2) and a 95% HPD ranging between 12.66 and 24.77 Ma (Table 3) in the four independent MCMC runs (ESS = 480). As in most analyses, when a normal prior distribution is used the estimates are slightly younger, yielding a mean age of 17 Ma (rightmost blue circle in row 4 of Fig. 4; see Table 2) and 95% HPD ranging from 10.75 to 24.43 Ma (Table 3). In both cases, the age of the oldest fossil caviid (P. pridiana; Laventan age; purple column in Fig. 4) is younger than the mean age estimate but falls within the 95% HPD (blue bar on tree of Fig. 4; Table 3).

The initial diversification of Caviinae and the split between Hydrochoerinae and Dolichotinae are inferred to occur two to five million years after the initial diversification of Caviidae, with mean gamma estimates of 16.37 and 14.42 Ma for these two clades and normal estimates of 14.08 and 12.30 Ma (Fig. 4). An interesting pattern in these estimates is the coincident time of diversification of three major modern and morphologically distinct linages of Caviidae (Fig. 4): Dolichotinae, Hydrochoerinae (node 3), and the caviine lineage leading to the genus Galea. The age of one of these nodes was constrained by a prior distributions (node 3; see Calibrated Nodes) but the ages retrieved for the diversification of Dolichotinae (yellow circles in row 4 of Fig. 4) and Galea (red circles in row 4 of Fig. 4) are highly congruent, with mean ages of 6.91 and 6.79 (normal distribution of calibration constraints; see Table 2) and 8.25 and 7.96 (gamma distribution of calibration constraints; see Table 2). The estimated ages of these nodes lie within the range of ages of the Chasicoan sediments, in which the oldest members of caviines, dolichotines, and hydrochoerines are recorded.

Three calibration constraints.

These analyses sequentially exclude one of the four calibration constraints (see Calibrated Nodes and analyses 3 to 10 in Table 1), to test their influence in inferring the time of diversification of Caviidae. The estimate of the initial diversification of Caviidae excluding node 1, node 2, or node 3 (central blue circles in row 3 of Fig. 4) yield similar results as the estimate with four calibration points (mean age ranging between 15.48 and 20.01 Ma and 10.05–26.64 Ma 95% HPD; see Tables 2, 3). However, when the caviine calibration constraint (node 4) is excluded, the retrieved age are markedly older (leftmost blue circles in row 3 of Fig. 4), and the age of the oldest fossil caviid (Prodolichotis pridiana) lies outside the 95% HPD (15.9–27.28 Ma; see Table 3).

As with the estimates of the diversification of Caviidae, the estimates on the time of diversification of the modern linages of Caviidae are highly sensitive to the exclusion of node 4 from the calibration set. Excluding any of the other three calibration constraints (nodes 1, 2, or 3) the mean ages of these nodes fall within the Chasicoan SALMA (see Table 2). However, when node 4 is excluded from the calibration set, the estimated age of these nodes is older. This is especially marked when a gamma distribution is used, yielding mean age estimates that predate the Chasicoan age (leftmost red and yellow circles in row 3 of Fig. 4; Table 2). Interestingly, the mean ages inferred for node 4 (Microcavia+Cavia) when this node is not calibrated is remarkably old, with mean ages of 14.78 and 17.45 Ma and 95% HPD (Table 3) lying outside the Chasicoan (significantly predating the appearance of derived caviines in the fossil record) This suggests that either this caviine lineage has a higher evolutionary rate than the other caviids or that its fossil record failed to capture more than 60% of their evolutionary history. These two hypotheses are plausible explanations but more data and research is needed for testing them.

Two calibration constraints.

These analyses retrieved disparate results on the time of diversification of Caviidae (see analyses 11 through 22 in Table 1). Out of the 12 conducted analyses, six of them yielded 95% HPD age estimates for the initial diversification of Caviidae that are older than the first fossil caviid and one of them yielded 95% HPD age estimates that are younger than the first fossil caviid (see Table 3). The six analyses that excluded the caviine calibration constraint (node 4 in Fig. 4; see Calibrated Nodes) yielded the oldest age estimates for the initial diversification of caviids (Table 2). The inclusion of the caviine calibration point (node 4) in combination with the calibration of the basal nodes of Cavioidea (node 1 or node 2) yielded mean age estimates (central blue circles in row 2 of Fig. 4) that are similar to those of the analysis with four constrained nodes (Table 2). The 95% HPD of these analyses included the age of appearance of fossil caviids (Table 3). However, when the two calibration constraints were node 4 and node 3, the estimated mean age and 95% HPD resulted exceedingly young (Table 3) in comparison with the caviid fossil record (mean age estimates ranging between 7.53 and 9.99 Ma, using a normal and gamma distribution; rightmost blue circles in row 2 of Fig. 4; Table 2).

The conducted analyses also retrieved highly variable results on the time of diversification of the major modern lineages of Caviidae mentioned above (Fig. 4). As for the estimates on the age of Caviidae, when the caviine calibration (node 4) was excluded, the inferred ages of several nodes are markedly older, with most mean estimates of most nodes dating over 9.5 Ma (leftmost red and yellow circles in row 2 of Fig. 4; Table 2). Again, this is particularly noticeable when a gamma distribution is used in the prior probability of the calibration constraints. When node 4 is included in the calibration set in combination with the basal most calibration constraints (nodes 1 or 2) the estimates on the diversification ages of the modern lineages of Caviidae retrieved mean ages that fall within the Chasicoan SALMA (central yellow and red circles in row 2 of Fig. 4; Table 2). The use of only the two most recent constraints (nodes 3 and 4) yields mean ages that are much younger than the Chasicoan age (rightmost yellow and red circles in row 2 of Fig. 4; see Table 2). In most of these estimates, however, the 95% HPD is notably large and extends for more than 10 Myr (for both normal and gamma prior distributions; see Table 3) covering from the Pleistocene to the Miocene (thus completely including the Chasicoan age).

One calibration constraint.

These analyses also retrieved disparate results on the timing of caviid evolution (analyses 23 through 30 in Table 1). All analyses that used the calibration of the two basal nodes of Cavioidea (node 1 and node 2; see Calibrated Nodes) retrieved remarkably old mean ages (leftmost blue circles in row 1 of Fig. 4; Table 2) with 95% HPD that are older than the first appearance of caviids in the fossil record (Table 3). Conversely, when node 4 is used as the single calibration point, the 95% HPD for the time of the initial diversification of caviids is younger than the first appearance of this group in the fossil record (rightmost blue circles in row 1 of Fig. 4; see Table 3). Finally, node 3 is the only calibration point that retrieves mean ages close to the age of the most ancient caviid P. pridiana (central blue circles in row 1 of Fig. 4; Table 2) and 95% HPD that includes this age (Table 3).

The estimates on the time of diversification of the major modern lineages of Caviidae are also highly sensitive to the choice of a single calibration constraint. When any of the two basal calibrations were used (node 1 and node 2), the mean ages of the Dolichotinae and Galea nodes are notably old (ranging from 9.79 to 12.84 Ma with both normal and gamma prior distributions; four leftmost red and yellow circles in row 1 of Fig. 4; Table 2). Furthermore, using these calibrations, the 95% HPD of the age of two other modern lineages of caviids (nodes 3 and 4) are in most cases exceedingly old (i.e., older than the Chasicoan SALMA; Table 3). Conversely, when the caviine age (node 4) is used as the only calibration constraint, the mean ages and the 95% HPD for the time of diversification of modern caviid lineages are younger than their first appearance in the fossil record at the Chasicoan SALMA (rightmost yellow and red circles in row 1 of Fig. 4; Table 3). Finally, as in the case of the age of Caviidae, when node 3 is the only calibration constraint the mean ages lie close to the upper bound of the Chasicoan age (5.3 to 5.7 Ma; see Fig. 4 and Table 2) and the 95% HPD of all the modern lineages includes the Chasicoan age (Table 3).

Summary of molecular clock estimates.

These results indicate a considerable degree of rate heterogeneity among groups of Cavioidea. The two oldest and most basal calibration constraints (node 1 and node 2; see Calibrated Nodes) lead to inferences of a much slower rate of evolution than when the caviine calibration constraint (node 4) is used, and rate estimates derived from calibrations of node 3 are intermediate between these two extremes. Consequently, estimates on the age of Caviidae and on the radiation of the major modern lineages of caviids based only on the basal calibrations (or basal nodes and node 3) are much older than those retrieved when node 4 (or node 4 and node 3) is used for calibrating the relaxed molecular clock (Fig. 4). This sensitivity to the choice of calibration constraint is also reflected in the 95% HPD that show variable results depending on the calibrate node/s. These distributions, however, converge to a similar result as the number of calibrated nodes increases, as exemplified for the 95% HPD of the age of Caviidae (Fig. 5). Based on these results, the use of four calibration points better accommodates the rate heterogeneity of the group as a whole, because these bracket the nodes of interest (Caviidae and the principal modern lineages of Caviidae). Therefore, we will discuss below the molecular clock inferences based on four calibration points (row 4 of Fig. 4) in terms of their agreement with the diversification pattern inferred from the fossil record of early caviids.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 5. 95%HPD of age estimates for the time of diversification of Caviidae as estimated for each of the 30 different Bayesian relaxed molecular clock analyses conducted in this study.

https://doi.org/10.1371/journal.pone.0048380.g005

Morphological data.

The morphological dataset was expanded from a recent study [48], [49] through the incorporation of six crown caviid taxa (four fossils, two extant) and ten new morphological characters. The morphological dataset included fossil and living representatives of Cavioidea s.s. that were selected as the ingroup. The dataset include all the known stem-group fossil taxa (“eocardiids”), at least one representative of each extant genus of Caviidae, and nine extinct species of caviids (see Document S2). Outgroup taxa included representative species of Dasyproctidae, Cuniculidae, and Echimyidae, the latter of which was used to root the topologies (see Document S2). The character sampling is based on 69 craneo-mandibular and 27 dental characters (see Document S3).

Molecular data.

The DNA sequences of extant caviid species were gathered from GenBank for two nuclear and two mitochondrial genes: exon 10 of growth hormone receptor (Ghr), intron 1 of transthyretin (Tth), 12 subunit ribosomal RNA (12s), and cytochrome b (cyb). Sequences of these genes were available for nine extant representatives of Caviidae and the three outgroup taxa (see Document S2 for GenBank accession numbers). Sequences of two of the outgroup taxa (Proechimys and Dasyprocta) have been assembled from two different species of each genus (see Document S2). These sequences have been successfully used by previous authors to resolve relationships of caviomorph and/or cavioid rodents [13], [18], [26], [40] and were therefore thought to provide adequate levels of divergence for conducting molecular clock estimates.

Combined dataset.

The phylogenetic dataset consisting of the 96 morphological characters and the 4014 characters of the four genes (Ghr, Tth, 12s, cyb) is available as Dataset S1 and also at DataDryad http://datadryad.org/ (doi:10.5061/dryad.v5p71).

Data analysis.

The DNA sequences of each of the four genes (12s, cyb, Ghr, and Tth) were compiled from several previous analyses (see Document S2), and were aligned using CLUSTAL X [99] using the default values of gap opening (10/100) and gap extension (0.1). The leading and trailing gaps were replaced with missing entries. After alignment, the leading and trailing ends of the sequences with no homologous sequences in other species were deleted. The alignment resulted in 961 base pairs (bp) for 12s gene, 1140 bp for cyb gene, 1099 bp for Tth gene, and 814 bp for Ghr gene.

The combined dataset of the 96 morphological characters was concatenated with the DNA sequences of the four genes (12s, cyb, Ghr, and Tth), scoring fossil taxa with missing entries for the DNA partitions. This dataset contained a total of 40 taxa and a total of 4110 characters. An equally weighted parsimony analysis was conducted treating gaps as missing data in TNT 1.1 [100], [101]. The heuristic search consisted in 1000 replicates of a Wagner tree with random addition sequence of taxa and followed by TBR branch swapping, collapsing zero-length branches under the strictest criterion. After this procedure, a final round of TBR branch swapping to find all most parsimonious trees (MPTs).

The results of the cladistic analysis (Fig. 1) are congruent with previous phylogenetic hypotheses [41], [47]–[49] and with the topology obtained in the Bayesian analysis of the molecular partition (see below). Further details on the parsimony analysis are given in the Document S1.

Calibration constraints.

Bayesian approaches to molecular clock estimates, as implemented in BEAST v. 1.6, allows using different prior probability distributions to calibrate selected node ages (calibration constraints). As noted by several authors in recent years [98], [110]–[115], the choice of calibration points is a critical step in molecular clock studies but is rarely discussed at length. We have explored the calibration of up to four nodes located both above and below the nodes of interest (diversification of caviid lineages). We performed 30 different analyses varying the number of calibrated nodes (from 1 to 4 calibrated nodes) and using two different prior distributions used for calibrating the age of these nodes (normal and gamma distributions; see below and Table 1). The analyses are numbered from 1 to 30, starting with the run with four constrained nodes that used a normal (analysis 1) or gamma (analysis 2) distribution. Analyses 3 to 10 used only three calibration constraints. Analyses 11 to 22 used two calibration constraints. Analyses 23 to 30 used a single calibration constraint.

Prior distributions of the ages of these four nodes were defined based on the available chronostratigraphic information of the fossil record, considering the phylogenetic placement of fossils in the phylogenetic analysis (see Fig. 1) and the uncertainties in the age of the fossiliferous sediments. Two different prior distributions were alternatively used (Normal and Gamma), which represent different levels of confidence on the true absence of a given lineage in sediments older than its first appearance in the fossil record.

Normal distributions were centered on midpoint age of the period of time to which the fossil-bearing formation has been referred. The standard deviation was set so that the 95% probability distribution reached the upper and lower bound of the age of the lithostratigrapic unit (see Calibrated Nodes and Table 1). This prior distribution makes a strong assumption on the absence of representatives of the calibrated node in older sediments, which may be appropriate in some cases but not in others [98].

Gamma distributions were used to produce asymmetric probability distributions that place the highest prior probability along the interval of time represented by the lithostratigrapic unit that has yielded the oldest member of the clade being calibrated. The long tail of the distribution extends towards older ages, gradually decreasing the probability (maximum soft-bound), and including in the 95% prior probability distribution the age of the most recent sediments in which representatives of the node being calibrated are absent, but numerous remains of its stem-group or other caviid lineages are known (see Calibrated Nodes and Table 1). The presence of abundant remains of outgroup taxa is a taphonomic-preservation control using ecological/taxonomic equivalents [116] and resembles the criteria advocated for calibrating nodes by some authors [98], [112], [114], [115], [117].

The 30 exploratory MCMC runs were conducted using different combinations of these distributions for the four calibration points and varying the number of nodes calibrated with the fossil record (from one to four calibrations; Table 1). The evidence used to define the prior distribution of ages of the four calibrated nodes is given in Calibrated Nodes and further details on the parameters used and the results obtained are given in Table 1.

Node 1 (Cavioidea; see Figs.1, 4).

This node represents all cavioids and includes representatives of Dasyproctidae, Cuniculidae, and Cavioidea sensu stricto. The most ancient fossil that has been assigned to this clade is Andemys termasi, known from a mandibular fragment found in the Tinguiririca fauna from the Abanico Formation of Central Chile [24], [29], [118], dated at 31.5–37.5 Ma [61]. Its affinities have never been tested in a phylogenetic analysis, but the combination of plesiomorphic and apomorphic features identified in previous studies [24], [29]; mesodont crowns but with a deep hypoflexid that would have projected nearly to the lingual margin of the molars when unworn) indicates this specimen can be tentatively regarded as a basal member of Dasyproctidae and therefore useful in calibrating the age of the node Cavioidea.

Calibrating this node with Andemys termasi [29] from the Tinguiririca fauna involves two different sources of uncertainty. First, there is an uncertainty in the radiometric dates of the Abanico Formation. Wyss et al. [24] provided Ar/Ar dates of the fossiliferous horizons and concluded that the fossils are at least as old as 31.5 Ma but the deposition of lower levels of the unit must have started about 37.5 Ma. Therefore, the 31.5–37.5 Ma uncertainties should be taken into account when calibrating the age of this node. Second, as noted above, this fossil is the among the oldest cavioid rodent known from South America, being the earliest member of this large clade of caviomorph rodents that evolved after the arrival of ancestral hystricognaths to this continent. The timing of the arrival of rodents to South America is, however, poorly constrained given the scarce rodent fossil record prior to the Tinguirirican deposits. The recent discovery of the oldest rodent fauna in the Yahuarango Formation [27] lacks cavioid rodents and has been radioisotopically dated by Ar/Ar at 43.44±2.5 Ma, suggesting an upper bound for the origin of Cavioidea. This finding lies within the range of previous molecular clock estimates for the initial diversification of caviomorph rodents (after their arrival to South America) that ranged between 30.7 and 55 Ma [20], [56], [59]. Therefore, the actual time of diversification of Cavioidea is uncertain due to the poor record of Eocene caviomorphs and the error associated to radioisotopic dates of the oldest rodent faunas, as there is a minimum of 3.5 my and a maximum of 14.5 my between their appearance of cavioids in the fossil record (Tinguiririca) and the oldest rodent record from South America, in which cavioids are absent (Yahuarango).

We have explored the use of two different prior probability distributions to account for these sources of uncertainty while calibrating this node. The first approach uses a normal distribution whose 95% probability density encompasses the radioisotopic ages published for Tinguiririca (31.5–37.5 Ma). This distribution largely ignores the second source of uncertainty (i.e., lack of adequate early fossil record of Caviomorpha and Cavioidea) and places a strong belief in that the dasyproctid from Tinguiririca is actually very close (in time) to the divergence time of Cavioidea. The second approach uses a gamma distribution, with a hard minimum bound at 31.5 Ma and its long tail extends the 95% probability density back to 45 Ma, representing the upper bound of the available dates of oldest record of caviomorphs from South America [27], as well as the inferred date for the diversification of Caviomorpha obtained in the most densely sampled molecular clock analysis [59].

Finally, given the scarcity of the available material to constrain the age of this node and the lack of an explicit phylogenetic analysis including Andemys termasi [24], [29] Tinguiririca mandible, we have also tested the use of this fossil for calibrating the deepest node of our phylogenetic tree (Fig. 4). The results, however, are largely similar in terms of the molecular clock estimates for most nodes within Cavioidea and therefore we conducted all analyses using Andemys termasi to calibrate Cavioidea.

Node 2 (Cavioidea s.s. + Cuniculus paca; see Figs. 1, 4).

This node of the phylogenetic tree represents all forms of Cavioidea s.s. plus the lineage of the family Cuniculidae. The fossil record of Cuniculidae is extremely poor and only starts in the Pleistocene [119], [120]. Cavioidea s.s., in contrast, has an extremely rich fossil record [53], especially in the southern region of South America (Patagonia). The most ancient definitive members of Cavioidea s.s. are Asteromys punctus and Chubutomys simpsoni, both known from few specimens found in the late Oligocene beds of Patagonia (Deseadan SALMA) of the Sarmiento Formation [49], [121]. The phylogenetic position of both taxa within Cavioidea s.s. is strongly supported in by the morphological data of the phylogenetic analysis presented here (Fig. 1), as in previous phylogenetic studies [47]–[49]. In our analysis, the two most basal nodes of Cavioidea s.s. have high Bremer support (i.e., 4 and 5) and bootstrap and jackknife frequencies above 96% (Fig. 2; see Document S1).

The age of Deseadan deposits that yielded Asteromys and Chubutomys is therefore critical for calibrating this node. Specimens of Asteromys were found in the localities Cabeza Blanca and Laguna de los Machos [49] and the type Chubutomys simpsoni only in the former locality. These fossils were found together with a mammalian faunal assemblage characteristic of the Deseadan age [52]. No radiometric dates are available for these two localities but other Deseadan beds have been dated between 24.5 and 29 Ma [61]–[62]. The range of radiometric ages obtained for other Deseadan localities is the first source of uncertainty that should be taken into account for calibrating this node. The possibility that these early cavioid fossils are younger than the cladogenetic event that separated the cuniculid lineage from the lineage of Cavioidea s.s. is the second source of uncertainty that should be considered. Pre-Deseadan rodent assemblages from Patagonia are known from La Cantera locality, which lacks forms of Cavioidea s.s. ([65]; see below) and has been dated between 29.5 and 31.1 Ma [62].

As with the calibration of Node 1, we explored the use of two different prior probability distributions for this node. The first was a normal distribution whose 95% probability density encompasses the range of radioisotopic ages published for the Deseadan SALMA (24.5–29 Ma), ignoring the second source of uncertainty. The second approach uses a gamma distribution, which puts a hard minimum bound at 24.5 Ma and its long tail extends the 95% probability density back in time to the age of the youngest rodent assemblage that lacks forms of Cavioidea s.s. (i.e. 29.5–31.1Ma; [62]). Although we have explored both strategies we believe the second option more accurately represents the uncertainties of the early fossil record of cavioid rodents.

Node 3 (Kerodon + Hydrochoerus; see Figs. 1, 4).

This node of the molecular phylogeny represents all forms of Hydrochoerinae (including the lineage leading to the Rock cavy Kerodon). The fossil record of Kerodon is only known from scarce material of the late Pleistocene of Brazil [122]–[124]. Fossil representatives of the lineage leading to Hydrochoerus (see Fig. 1), in contrast, are relatively abundant in some Miocene-Pliocene deposits of South America [125]. The oldest definitive records of this lineage are found in the Arroyo Chasicó Formation of Central Argentina (Chasicoan SALMA) and include Cardiatherium chasicoense [126], [127], Cardiomys cavinus [128], and Procardiomys martinoi [129]. The phylogenetic position of Cardiomys cavinus and Cardiatherium chasicoense within Hydrochoerinae is supported by the morphological data of the phylogenetic analysis presented here (Fig. 1). Although the clades within Hydrochoerinae have low to moderate Bremer support values and bootstrap and jackknife frequencies are below 60% (Fig. 2; see Document S1), forcing the exclusion of both Cardiomys and Cardiatherium outside Hydrochoerinae requires five extra steps in the parsimony analysis, demonstrating their inclusion within Node 3 is relatively well supported by the morphological data.

The age of Chasicoan deposits yielding this diverse assemblage of hydrochoerines is important for calibrating this node. The stratigraphic levels that contain these taxa belong to the upper section of the Arroyo Chasicó Formation (Lithofacies 3 sensu Zárate [130] or Las Barrancas Member sensu Ugarte [131]. The age of these levels is certainly younger than 9.07 Ma, which is the age of radiometric dates of the underlying lithofacies 1 and 2 of the Arroyo Chasicó Formation [130]. Unfortunately, radiometric dates for the fossiliferous levels are lacking, but Deschamps [132] considered that lithofacies 3 of the Arroyo Chasicó Formation can be biostratigraphically correlated with the Loma de Las Tapias Formation (northwestern Argentina), which has a radiometric date of 7±0.9 Ma. A conservative approach, therefore, is to consider the age of the fossiliferous levels of the Arroyo Chasicó Formation as ranging between 6.1 and 9.07 Ma.

Although these fossils from the Chasicó Formation are the oldest well preserved and definitive record of hydrochoerines, it is difficult to determine if earlier members of this node were present before the deposition of this unit. In fact, the fossil record of cavioid rodents has a noticeable gap before the Chasicoan. The only older rodent remains that could be related to hydrochoerines come from the Mayoan SALMA (11.5 Ma; [133]) and consists of a single isolated tooth from the Río Frias Formation that has been referred with doubts to Cardiomys [80], [134]. This isolated tooth, however, lacks diagnostic features to determine their phylogenetic placement either as part of Node 3 or as a close relative of this clade. The caviid fossil record before the Mayoan is found in sediments of the Laventan SALMA (11.8–13.5Ma; [69]) and consists of the basalmost members Caviidae, which are definitively placed outside Hydrochoerinae (e.g., Prodolichotis pridiana; see below and Fig. 1), placing a maximal bound for the origin of this node.

As with the other nodes, we explored two different prior probability distributions for the age of Node 3. The first used a normal distribution whose 95% probability density encompasses the range of radioisotopic ages that bracket the fossiliferous levels of the Arroyo Chasicó Formation (6.1–9.07 Ma); this ignores the phylogenetic uncertainty of the fragmentary remains with possible hydrochoerine affinities in older sediments (e.g., Mayoan SALMA). The second used a gamma distribution, with a hard minimum bound at 6.1 Ma and a long tail to extend the 95% probability density back to the age of the Laventan SALMA, which is the youngest rodent assemblage that is well sampled and lacking forms that could potentially belong to Node 3 (i.e., 11.8–13.5 Ma). As in previous cases, we have explored both approaches but consider the second option better represents the uncertainties in the fossil record of early hydrochoerines.

Node 4 (Microcavia + Cavia; see Figs. 1, 4).

This node of the molecular tree represents all forms of Caviinae closer to Microcavia or Cavia than to Galea. The earliest fossil members of Cavia come from the San Andrés Formation (late Pliocene; [135]) and are based on scarce mandibular material. In contrast, fossils referred to Microcavia or other genera possibly related to Microcavia are relatively abundant in some Late Miocene to early Pliocene deposits of South America [124], [136]. The phylogenetic analysis performed here includes four fossil taxa that have been regarded as primitive members of Caviinae: Allocavia chasicoense, Paleocavia impar, Dolicavia minuscula, and Microcavia chapalmalensis [124], [134], [137], [138]. The last three taxa are retrieved as more closely related to the extant Microcavia australis than to Cavia and provide adequate information for calibrating this node.

M. chapalmalensis is recovered as the sister taxon of M. australis in the phylogenetic analysis (Fig. 1) and comes from the Chapadmalal Formation of Central Argentina (Chapadmalan SALMA). The sister group relationship of these two species of Microcavia are well supported by the morphological data of our phylogenetic analysis. The support values for the node of the genus Microcavia are low (Bremer  = 2 and bootstrap and jackknife frequencies around 60%; Fig. 2; see Document S1), but this is due to the unstable behavior in suboptimal (or bootstrap) trees of some fragmentary taxa (e.g., Allocavia). Forcing M. chapalmalensis to be positioned outside the Node 4 requires a minimum of five extra steps demonstrating the strong support for its inclusion within this node. The minimum age of the Chapadmalal Formation has been dated at 3.27 Ma using K-Ar radioisotopes [139], whereas the maximum age of these levels is usually regarded as 4 Ma [140], [141]. Fragmentary remains found in older sediments of Monte Hermoso Formation (Montehermosan SALMA) and Aconquija Formation (late Miocene – early Pliocene) have been referred to Microcavia [142], [143] but these cannot be identified at the species level and lack unambiguous synapomorphies of this genus. These remains could belong to the stem of Microcavia (i.e., being part of Node 4) or have an even more basal position, but more complete remains are needed to place them confidently.

Dolicavia minuscula was recovered in a basal polytomy within the lineage leading to the genus Microcavia together with Paleocavia impar (Fig. 1). Numerous and well-preserved remains of Dolicavia are known from the Chapadmalal Formation of Central Argentina ([134], [144]; 3.27–4 Ma). No remains of this taxon have been found in older deposits (e.g., Monte Hermoso Formation). In contrast to the strong support for the position of Microcavia chapalmalensis, the position of Dolicavia is only poorly supported, as it takes a single extra step to place this taxon basally within Caviinae and only 20% of the bootstrap and jackknife trees place this taxon within Node 4.

Paleocavia impar is known from multiple specimens with skull and mandibles found in the Monte Hermoso Formation [145]. As in the case of Dolicavia, the inclusion if Paleocavia impar within Node 4 is not strongly supported. Topologies with a single extra step places this taxon more basally within Caviinae and only 33% of the bootstrap and jackknife trees place Paleocavia allied with the genus Microcavia. The maximum age of this unit has been determined as 5.3 Ma based on radiometric dates [146] and the minimum age was estimated as 4 Ma [140], [141]. The presence of Paleocavia in older sediments (Huayquerian SALMA; 5.3–6.1 Ma; [147]) of the Ituzaingó Formation [148] and of the Cerro Azul Formation [149] have been reported in faunal lists of these units, but these specimens have not been described and we cannot test at the moment their phylogenetic affinities. If the presence of Paleocavia in the Ituzaingó and Cerro Azul formations were confirmed, it would push the first appearance of Node 4 back to the Huayquerian, but for the moment the Montehermosan record of Paleocavia is the oldest record of Node 4 (Microcavia+Cavia).

The rodent fossil record of the older sediments of the Chasicó Formation (Chasicoan SALMA; 6.1–9.07 Ma; see above) provides confident information to place a maximal bound for the origin of this node. The Chasicoan fossil record contains remains of caviine-like caviids (e.g., Allocavia) as well as numerous forms of Dolichotinae and Hydrochoerinae [53], [81], [150]. All these forms are recovered in the parsimony analysis outside the derived Node 4 of Caviinae (see Fig. 1). None of the hundreds of rodent remains known from Chasicoan deposits can be allied to the node formed by Microcavia and Cavia (Node 4).

As with other nodes, we explored two different prior probability distributions for the age of Node 4 based on the position of fossil taxa in the most parsimonious trees of our analysis. The first used a normal distribution whose 95% probability density encompasses the range of radioisotopic ages that bracket the fossiliferous levels of the Monte Hermoso Formation (4–5.3 Ma); this ignores the uncertainty associated with the possible presence of Paleocavia in older sediments (i.e., Huayquerian). The second approach used a gamma distribution, with a hard minimum bound at 4 Ma and a long tail that extended the 95% probability density back to the Chasicoan, which is the youngest well-known rodent assemblage lacking taxa that belong to Node 4 (i.e., 6.1–9.07 Ma). As in previous cases, we have explored both approaches but believe the second option better represents the uncertainties in the fossil record of early caviines.

Finally, given the oldest member of this clade (Paleocavia; Montehermosan SALMA) is only poorly supported within Node 4, we have also tested alternative calibrations for this node. We have conducted exploratory runs of the Bayesian analysis calibrating this node with the age of Microcavia chapalmalensis (Chapadmalan SALMA), which is the only fossil of this clade that is robustly supported by the morphological data of the phylogenetic analysis (see above). The results of this analysis are largely similar in terms of the molecular clock estimates for Caviidae (mean age  = 17.5 Ma) and place the diversification of dolichotines and Galea within the Chasicoan SALMA. Therefore, the estimates of interest for our purposes do not seem to be sensitive to the alternative ages that can potentially be used for calibrating Node 4.