Corrected placement of Mus-Rattus fossil calibration forces precision in the molecular tree of rodents

Time calibration derived from the fossil record is essential for molecular phylogenetic and evolutionary studies. Fossil mice and rats, discovered in the Siwalik Group of Pakistan, have served as one of the best-known fossil calibration points in molecular phylogenic studies. Although these fossils have been widely used as the 12 Ma date for the Mus/Rattus split or a more basal split, conclusive paleontological evidence for the nodal assignments has been absent. This study analyzes newly recognized characters that demonstrate lineage separation in the fossil record of Siwalik murines and examines the most reasonable nodal placement of the diverging lineages in a molecular phylogenetic tree by ancestral state reconstruction. Our specimen-based approach strongly indicates that Siwalik murines of the Karnimata clade are fossil members of the Arvicanthini-Otomyini-Millardini clade, which excludes Rattus and its relatives. Combining the new interpretation with the widely accepted hypothesis that the Progonomys clade includes Mus, the lineage separation event in the Siwalik fossil record represents the Mus/Arvicanthis split. Our test analysis on Bayesian age estimates shows that this new calibration point provides more accurate estimates of murine divergence than previous applications. Thus, we define this fossil calibration point and refine two other fossil-based points for molecular dating.

The rationale for considering the first appearance of Progonomys as the Mus/Rattus split is based on the simplified phylogenetic hypothesis of Siwalik murine rodents by Jacobs 33 and Jacobs and Downs 10 , which proposed dichotomous lineages (the Progonomys clade and the Karnimata clade), deriving from Antemus (Fig. 1A, see SI Discussion for details). Mus was confidently placed in the Progonomys clade based on observation of gradual morphological change from older Progonomys to younger Mus auctor in Siwalik murines 10,33,34 . On the other hand, Rattus was placed in the Karnimata clade with some uncertainty 10 , along with extant genera of the Arvicanthini (Arvicanthis, Pelomys, Mylomys, and Golunda), heavily relying on the anterior displacement of the anterostyle 10,33,34 (dental terminology in Fig. 1B). No further morphological evidence is found for the evolutionary relationship between Rattus and Siwalik species of the Karnimata clade 35 .
In more recent studies, we have shown that diverging tooth morphology in the Siwalik lineages was initiated by 11.2 Ma, with decreasing morphological overlap through time 34 between the Progonomys and Karnimata clades (Fig. 1C), and that the 11.2 Ma stratigraphic occurrence can be considered as the evidence of lineage separation under the initial split criterion 36 . In this study, we examine the most reasonable placement of this separation event in a molecular phylogenetic tree by presenting new qualitative characters that define the lineage separation and tracing ancestral states of the characters in a molecular phylogenetic tree of extant species.
Here we show that progressive acquisition of new dental characters in the Karnimata clade is strong evidence for Siwalik murine rodents representing a younger node, the Mus/Arvicanthis split, than previous applications. The finding of this study constrains the timing of the Mus/Arvicanthis split to ~11.2 Ma as a stratigraphic occurrence and to 11.36 Ma by a molecular divergence estimate and further allows revision of calibration dates for two additional clades. These fossil-based dates are exceptionally highly  resolved, compared to any other estimate thus far, because the divergence event is morphologically tracked in geologic time and because of well-constrained age estimates of Siwalik localities within both biostratigraphic and paleomagnetic frameworks.

Results
Frequency distributions of the size and inclination of the metacone ( Fig. 2A-E) and the presence of the posterostyle were evaluated for each Siwalik species (Fig. 3, Tables S1-S3, Dataset S1). A large metacone that is posteriorly inclined parallel to the paracone ( Fig. 2A) is plesiomorphic for the Murinae. In the Karnimata clade, ?Karnimata conserves this plesiomorphic condition in 60% of the specimens at 11.2 Ma (Table S2). By 8.2 Ma, this plesiomorphic condition disappears from the Karnimata clade. Instead, a new combination of character states, a small metacone in vertical orientation (Fig. 2E), appears for the first time in 24% of Karnimata darwini at 9.2 Ma (Fig. 3, Table S2). The frequency of the derived condition (small, vertical) exceeds half of the specimens at 8.8 Ma and reaches 85% by 7.4 Ma. In contrast, the plesiomorphic condition of the metacone persisted in the Progonomys clade (Fig. 3). The frequency of the plesiomorphic condition continuously increases as the frequency of the 'large, slightly inclined' state decreases through time. By 8.8 Ma, more than 90% of the specimens have this plesiomorphic condition. None of the specimens in the Progonomys clade have the derived condition of the metacone (Table S3). As expected from the observations in each clade, the correlation between the two characters (size, inclination) is stronger in the Karnimata clade (Goodman-Kruskal's gamma = − 0.76) than in the Progonomys clade (Goodman-Kruskal's gamma = − 0.39). The null hypothesis of independence between the variables was rejected for the Karnimata clade (M 2 = 23.9, df = 2, p < 0.0001) but was accepted for the Progonomys The analysis of ancestral state reconstruction shows that a combination of the 'small' and 'vertical' states of the metacone is concentrated in the Arvicanthini (Fig. 4A, Fig. S2 for black-and-white). Maximum likelihood unambiguously placed it as the most probable ancestral state (93.0%) of the metacone for the Arvicanthini. In the node of the Arvicanthini and Otomyini, a small metacone in vertical orientation is predicted as the most probable ancestral state (80.3%) but is marginally insignificant over the second most probable ancestral state, a large metacone in parallel orientation (12.2%, difference of log likelihood = 1.89). For the node of the Arvicanthini, Otomyini, and Millardini (hereafter the Arvicanthini-Otomyini-Millardini clade), prediction for vertical orientation of the metacone (71.9%) is greater than that for parallel orientation (26.3%).

Discussion
Siwalik murines of the Karnimata clade demonstrate progressive acquisition of a derived condition of the metacone (a combination of small size and vertical orientation Fig. 2E), which appeared as a minor variation at 9.2 Ma and became a dominant character state by 8.8 Ma. In contrast, this condition does not occur in any individuals of the Siwalik Progonomys clade. Ancestral state reconstruction suggests that a small meta-cone in vertical orientation is a synapomorphic character for the Arvicanthini-Otomyini-Millardini clade (Fig. 4A). Inside that clade, three arvicanthine genera (Golunda, Mylomys, Stochomys) and the Otomyini secondarily lost the character in acquiring their specialized tooth morphologies (  10,11,33 , it is most logical to conclude that the progressive morphological divergence between the Progonomys and Karnimata clades represents the Mus/Arvicanthis split, which is more internal than the Mus/Rattus split at the tribal level. The independent acquisition of a vertically-oriented small metacone in the Apodemini within the Progonomys clade (Fig. 4A) is consistent with the paleontological hypothesis that European species of Progonomys is ancestral to Apodemus 37 . Absolute nodal ages were estimated with the 11.2 Ma fossil date of the Mus/Arvicanthis split using the published data of Fabre et al. 29 (Figs 4B and S4) and were then compared with the results of Fabre et al. 29 (Tables S4). In three separate analyses of Fabre et al. 29 , the 12 Ma fossil date (i.e., the first appearance of Progonomys) was applied to two older nodes (the Mus/Rattus split and the Phloeomys/core Murinae split, respectively) and was excluded from analysis for cross-validation of the fossil constraint. Newly calibrated ages in this study are consistently older than the three applications of Fabre et al. 29 , and our molecular estimates are more congruent with paleontological evidence compared to any previous studies 9,22,26,29,30 (Table S4). For example, the posterior mean age for the Phloeomys/core Murinae split (i.e., the node of the crown Murinae) estimated to be 13.6 Ma matches with the fossil evidence (13.8 Ma) of the first definitive murine rodents, which is outside the 95% CI of Fabre et al. 29 (Fig. 4). This new calibration point for the most diverse group of modern mammals is unique in that a lineage separation event is identified based on morphological divergence, rather than the stratigraphic occurrence of the   Figure 3. Percent frequency distribution of the size and inclination of the metacone in Siwalik murine rodents. Species that are likely in ancestor-descendant relationships 10,33 are connected by a solid line. Open symbols correspond to those in Fig. 1A, and colors of the 'size and inclination' legend correspond to those in Fig. 4. Different from the previous phylogenetic hypothesis (Fig. 1A), Progonomys hussaini is interpreted to appear after the initiation of lineage separation 34 . See Tables S1-S3 for numeric percentages and the number of individuals observed in this study, and SI Discussion for age determination. All localities but DP 13 have Y as a prefix. most basal taxa, and that the rich fossil record within a magnetostratigraphic framework constrains the bounding of the molecular divergence dating.
Based on our results, we explicitly define the lineage separation event of the Siwalik murines as a fossil calibration point of the Mus/Arvicanthis split and newly refine two fossil-dates for the Arvicanthini-Otomyini-Millardini clade and for the Murini, using Siwalik fossils. We follow Parham et al. 38 , who proposed a standard system for introducing new fossil calibrations, including a series of steps with an emphasis on the use of museum specimens to clarify the phylogenetic position of the fossil calibration.    28 . Pie charts indicate the probability of ancestral states of the combined characters of the metacone at a given node. Systematic nomenclature of the tribes follows Lecompte et al. 22 . The "?" state is assigned for the species in which the inclination of the metacone is not recognizable because the metacone is fused with the paracone or hypocone. (B) Maximum clade credibility tree from the BEAST analysis of the Fabre et al. 29 data using the 11.2 Ma calibration point for the Mus/Arvicanthis split. Node bars indicate the 95% credible interval of the posterior density of divergence times. Number on the nodes represents the posterior mean of divergence times.

The Mus/Arvicanthis split
Paleomagnetic and biostratigraphic studies in the Siwalik group designate each small mammal fossil locality as belonging to a chronologically controlled stratigraphic bin of 100,000 years maximum duration (see SI Discussion for details). The bins were constructed by subdividing stratigraphic intervals between magnetochron boundaries 41,42 . This dating precision to ≤ 100,000 years, in combination with the fine-scale fossil record, is the great strength of Siwalik Muridae for molecular clock calibration studies, providing narrow intervals for the new fossil calibrations.
While the usage of molecular divergence dating is vast, a priori evaluation of fossil calibration quality is crucial for accurate Bayesian estimates of divergence times [2][3][4][5][6] . A comparative simulation study shows that an a posteriori cross-validation approach may select highly precise but inaccurate calibration points that are not consistent with fossil evidence 5 . Our results better corroborate paleontological estimates and greatly reduce the range of the prior age for the calibration point from the 95% credible interval of ~10 million years 29 , an interval nearly equaling to most of Murine evolutionary history (~14 million years), to that of ~2 million years. We further emphasize the fundamental importance of a well-dated and documented fossil record for evolutionary questions concerning modern organisms that rely on molecular time estimates. Such questions include but are not limited to diversification rates, evolutionary patterns, and biogeographic events, which are influenced by geological processes and climate changes through time.
Scientific RepoRts | 5:14444 | DOi: 10.1038/srep14444 Methods Fossil specimens (n = 272) of upper first molars (M1), ranging in age from 14.3 to 6.5 Ma, were examined in this study (Dataset S1). They were collected from the Potwar Plateau, northern Pakistan, in the 1970's to 2000, and are currently on long-term loan from the Geological Survey of Pakistan at the Peabody Museum of Archaeology and Ethnology, Harvard University. We follow Kimura et al. 36 for systematic classification of Siwalik murines. Ages of the Siwalik localities derived from paleomagnetic stratigraphy are based on Geomagnetic Polarity Time Scale 2004 43 . Modern specimens (n = 500) examined in this study (Dataset S2) include individuals of 70 genera and 79 species, comprising 54% (70/130) of the total genera of Murinae and 79% (31/39) of the genera within the Mus/Arvicanthis split 22,29,44 . These specimens are curated in the Museum of Comparative Zoology (MCZ), Harvard University, and in the Smithsonian Institution National Museum Natural History (USNM). Digital images of upper molars of these species are provided in Fig. S3.
First, we examined change in the frequency distribution of the size and inclination of the metacone relative to the paracone on M1 of the Siwalik fossil species. Schematic diagrams of the character states, to which each specimen was referred for scoring characters, are shown in Fig. 2. The size of the metacone was observed as the width of the metacone on the labial side of the tooth relative to that of the paracone, scored as (a) as large as the paracone or (b) smaller than the paracone ( Fig. 2A,B). The inclination of the metacone was scored as the axis of the metacone is (c) inclined posteriorly parallel to that of the paracone, (d) slightly inclined posteriorly but not parallel to that of the paracone, and (e) not inclined posteriorly (= vertical) (Fig. 2C-E). The size of the posterostyle was scored relative to that of the enterostyle in occlusal view. These character states were also scored for modern species. We preferred the qualitative assessments, rather than quantitative measurements, in this study because large-scale systematic differences were expected to appear as fixed characters. For the Karnimata and Progonomys clades, independence and correlation of the two metacone characters were tested and measured by a generalized Cochran-Mantel-Haenszel (CMH) test and Goodman-Kruskal's gamma.
Then, we traced ancestral states of these metacone characters on the maximum likelihood (ML) tree of Fabre et al. 28 by the ML method of ancestral state reconstruction in Mesquite 2.75 45 . Topologies of molecular trees are largely congruent for Murinae 22,24,27,28,30 , but the chronogram of Fabre et al. 28 most comprehensively covers murine species to date. The Markov k-state 1 parameter model (Mk1) was chosen, which takes the rate of character change as a single parameter and assumes equal probability for any particular character change. The likelihood decision threshold of 2.0 was adopted as a cutoff for the significance of the likelihood ratio between two character states. Eight genera (Thamnomys, Dephomys, Hapalomys, Papagomys, Echiothrix, Lenothrix, Margaretamys, Pithecheir) were excluded from the ancestral character state reconstruction due to low bootstrap values (< 70%) on their nodes.
Finally, we applied the new fossil calibration point (the Mus/Arvicanthis split) to a published dataset to test whether the new calibration point gives more accurate estimates of absolute nodal ages than previous applications. The data of Fabre et al. 29 were chosen for the test analysis because they provide detailed comparisons of estimated ages obtained by different applications of the 12 Ma fossil date. In our analysis, we replaced these calibration points with the Mus/Arvicanthis split. The fossil date of the Apodemurini was excluded because its credible interval (CI) greatly overlaps with the newly proposed date for the Mus/Arvicanthis split. All other conditions were identical to those in Fabre et al. 29 . The analysis was conducted in BEAST (v 1.8.0) 46 via the CIPRES Science Gateway 47 . We bounded the lower and upper limits of this prior (median age: 11.63 Ma, 95% CI: 11.16-14.03 Ma) based on the stratigraphic occurrence of ?Karnimata and Progonomys hussaini at the lower bound and the stratigraphic occurrence of "near Progonomys" at the upper bound, respectively, as described in the Discussion. We set the minimum age of the fossil locality in the 5% quantile of the lognormal curve. SI Materials and Methods provide further details of the methods used in this study.