Of the 340 models examined, eight fell within 2 LnL units of the optimal score, forming a well-constrained “credible set” of models (Additional file 1: Table S4). Six of these models estimated separate model parameters for each of the four major sloth subgroups, while Models #315 and #320 estimated a single trended random walk (fully parameterized) model for a combined clade of Megalonychidae, Megatheriidae, and Nothrotheriidae (Additional file 1: Tables S4 and S5). The predominant reconstructed evolutionary modes in the credible set were trended random walks (43% of subgroup reconstructions: Additional file 1: Table S5), followed by driven trends and Brownian motion (23% each). Stasis was reconstructed in only three cases (10%). Reconstructed modes were not evenly distributed across sloth subgroups. All but one of the best-supported models supported a trended walk for the basal group (Mylodontidae + Bradypodidae + Pseudoglyptodontidae), with the other model supporting a driven trend. The Hapalopidae + Pelycodontidae grouping returned Brownian motion in all but two model structures, with the remaining models supporting trended walks. Megatheriidae + Nothrotheriidae results were evenly split, with four models supporting a driven trend towards large body sizes, while the other four supported a trended random walk (two in combination with Megalonychidae). Finally, Megalonychidae, which includes the modern two-toed sloths, showed varied results, with the only observed reconstructions of evolutionary stasis occurring in this clade.
We calculated model-averaged parameter estimates  for the credible set of 8 models (Table 1). The model-averaged estimates for the mean drift parameter (α) and the variance inflation term (β) vary greatly among the four groups, reflecting different inferred modes of body size evolution. The basal group displayed high reconstructed values for the mean drift parameter (α), emphasizing a pronounced trend to larger body sizes. The inferred mean body size increase over the observed range of this group is approximately 11.6 kg/million year (hereafter, My). The model-averaged estimate of β was also high, indicating a trend to increased variance.
Both of the linear drift parameters were slightly negative, with CI’s excluding zero, for the Hapalopidae + Pelycodontidae group in the model-averaged estimates (Table 1), indicating decrease in mean body size and reduced variance through time for this group, although the absolute value of these terms was quite low. Thus, although stasis was not supported for this group, the model-averaged parameters do not indicate large changes in the body size distribution through time for Hapalopidae + Pelycodontidae (inferred change in mean body mass was only −0.11 kg/My for this group).
The model-averaged estimate for α for Megatheriidae + Nothrotheriidae was strongly positive (Table 1), indicating a pronounced trend to increased size in this group, with an inferred increase in mean body mass of 129 kg/My (range: 42 to 384 kg/My, based on estimated CIs for α; Table 1) for this clade. Estimates for β were generally low, although positive. Both α and β were low, but also positive, for the Megalonychidae, with an inferred increase in body size of about 2.6 kg/My (range: 0.7 to 5.3 kg/My) across its range.
Despite the coarse phylogenetic resolution currently available for fossil sloths, the combination of fossil and extant taxa into a single analysis has permitted reconstruction of an interesting and surprisingly complex pattern of body size evolution. A single set of rate parameters cannot describe the evolution of body size in this clade, and reconstructing these patterns would not have been possible absent data from the fossil record. Our analysis support trended walks as the predominant model of sloth body size evolution, with driven trends and random walks receiving some support as well. All four sloth subgroups were separately parameterized in 6 of the 8 best-supported models, reflecting distinct body size distributions for each group. Moreover, each clade is reconstructed with markedly difference ancestral body masses (Additional file 1: Table S6) that increase from earlier to later diverging groups, demonstrating a general trend towards larger body size through their evolution.
Previous studies of mammal body size evolution focusing only on extant taxa have reconstructed small-bodied ancestral body sizes for sloths, with a low variances , which is not surprising given the observed body size distribution in extant sloths. The problem is that, in the particular case of sloths, the extant sample represents such a minor fraction of their evolutionary diversity, both taxonomically and morphologically, that inferences drawn from living taxa are unlikely to accurately reflect true evolutionary processes. Indeed, inferences based on a depauperate clade of small-bodied folivores would be hard-pressed to reconstruct anything along the lines of giant megatheriid ground sloths ,,,. This issue is clearly reflected in the ancestral body mass estimates for all of the sub-groups when fossil taxa are incorporated in the analysis (Additional file 1: Table S4).
To estimate the effect of the anomalous extant sloths on reconstructions of evolutionary trends in this clade, we reanalysed the dataset excluding the two genera of extant sloths. Here, only two models were strongly supported (Additional file 1: Table S6), and the model-averaged results were largely similar for the two groups with no extant membership (Megatheriidae + Nothrotheriidae and Hapalopidae + Pelycodontidae, see Table 1). However, model-averaged estimates of α and β for Megalonychidae indicated a substantial trended random walk to larger body size with larger variance for this clade. Estimates for both parameters were approximately four times greater than those estimated when including extant Choloepus and fall outside the estimated CIs for both paramters (Table 1). Thus, not only would an analysis of the extant taxa alone provide an oversimplification of sloth body size evolution, but extant sloths appear to actively obscure an otherwise strong signal in the fossil record.
The ability of fossil data to improve inferences of character evolution is a function of the amount of the data the fossil record can bring to bear on the reconstructions. While sloths were more diverse in the past than they are today, sloths have never been a highly speciose clade. The available sample size, even incorporating fossil data is somewhat small (57 taxa spanning ~35 million years, Additional file 1: Table S2). More specifically, it is important to note that the method we employ here is unable to resolve finer scale trends within individual lineages, and more complex processes of body size evolution may be masked by a lack of phylogenetic resolution, low sample size, and the resulting reduction in resolving power, particularly for clades such as Megalonychidae. For example, the sole extant megalonychid genus, Choloepus, is the smallest megalonychid taxon (mean mass of ~5.5 kg), nested within a clade with a mean body mass of 236 kg and a reconstructed ancestral body mass of 50 kg. It is therefore almost certain that an isolated shift to smaller body size occurred in within megalonychids, but we currently lack the fossil data to resolve this trend.
Another caveat of this method is that it is reliant on a priori groupings. For example, the extant genus Bradypus is one of the earliest-diverging lineages in the phylogeny (Figure 1) , and is not itself nested within a large-bodied clade. It is possible that the bradypodids diverged from other sloths prior to the onset of the trended walk towards larger body size that is reconstructed for the basal group in which it was included, and is therefore not part of this larger pattern. If we were to group Bradypodidae and Pseudoglyptodontidae in a basal group apart from the Mylodontidae, this would result in a partition of six species in two genera: one genus (two species) from >30 Mya and one Recent genus (four species). This very large gap between essentially two data points will almost inevitably resolve as a trended walk, simply as an artefact of the temporal distance between the genera, rather than a real evolutionary trend. In an attempt to bridge the 30 million year gap between these families, we included Mylodontidae, a speciose family of medium- to large-bodied sloths spanning nearly the entire temporal gap between pseudoglyptodontids and bradypodids. However, it is possible that the reconstruction of a trended body size increase in the basal group is actually a punctuated event across the small-to-large gap between pseudoglyptodontids and bradypodids on the one hand and mylodontids on the other.
If we analyse Mylodontidae on its own, we reconstruct a strong positive trend in mean body size, comparable to that reconstructed for the entire basal group: the model-averaged estimate for α was 0.113 for Mylodontidae, as compared to 0.120 for the basal group in Table 1. However, reconstructed β for Mylodontidae alone was −0.162, suggesting a variance decrease through time in this large-bodied clade when analysed separately. Similarly, ancestral mass reconstructions for the basal group combine the small-bodied Bradypodidae and Pseudoglyptodontidae (median body mass ~4 kg) with the much larger Mylodontidae (median mass of ~920 kg). As such, the reconstructed ancestral mass of approximately 66 kg (Table 1) for the basal group seems rather large, in light of the branching order of the cladogram (Figure 1). Analysing Mylodontidae to the exclusion of the small bodied basal clades produces estimates of the ancestral mean body mass more in accord with observed masses in each group (~6 kg for the basal group and ~31 kg for mylodontids). As such, there was likely an increase in body size along the branch between bradypodids + pseudoglyptodontids and mylodontids. Whether this was a punctuated event or a trend is ambiguous given the current data, but we can be confident of a trend of increasing body size within the Mylodontidae, and that this is not an artefact of our a priori grouping.