Syst. Biol. 48(1):107–118, 1999

The Fossil Record of North American Mammals: Evidence for a Paleocene Evolutionary Radiation

JOHN ALROY1 Department of Paleobiology, Smithsonian Institution, MRC 121, Washington, D.C. 20560, USA

Abstract.— Paleontologists long have argued that the most important evolutionary radiation of mammals occurred during the early Cenozoic, if not that all eutherians originated from a single common post-Cretaceous ancestor. Nonetheless, several recent molecular analyses claim to show that because several interordinal splits occurred during the Cretaceous, a major therian radiation was then underway. This claim conicts with statistical evidence from the well-sampled latest Cretaceous and Cenozoic North American fossil record. Paleofaunal data conŽrm that there were fewer mammalian species during the latest Cretaceous than during any interval of the Cenozoic, and that a massive diversiŽcation took place during the early Paleocene, immediately after a mass extinction. Measurement data show that Cretaceous mammals were on average small and occupied a narrow range of body sizes; after the Cretaceous–Tertiary mass extinction, there was a rapid and permanent shift in the mean. The fact that there was an early Cenozoic mammalian radiation is entirely compatible with the existence of a few Cretaceous splits among modern mammal lineages. [Body mass; Cenozoic; Cretaceous; diversiŽcation; extinction; Mammalia; molecular clock.]

Over the past few years, there has been an the “Cretaceous radiation” research suffers explosion of interest in the early evolution- from using just one or two “clock” calibra- ary radiation of mammals. Traditional sce- tion points (Hedges et al., 1996; Cooper and narios based mostly on paleontological data Penny, 1997; Janke et al., 1997; Kumar and have been challenged by inferences based on Hedges, 1998). Such calibrations often yield the calibration of molecular phylogenies to anomalous results that are defended by as- numerical time (Hedges et al., 1996; Janke et sertions that all of the conicting, paleonto- al., 1997; Springer, 1997; Cooper and Fortey, logically inferred dates of origin are simply 1998; Kumar and Hedges, 1998). Some of too young. these new molecular studies (e.g., Kumar For example, Hedges et al. (1996: 227) jus- and Hedges, 1998) purport not just to over- tify their decision to use a single Carbonifer- throw traditional higher-order phylogenetic ous calibration point with a non sequitur— groupings, but also to show that a major that there is a “long time span between the diversiŽcation of therian mammals began earliest [mammalian] fossils. . . and the Žrst much earlier than previously thought, per- appearance of the modern orders.” In other haps even in the Early Cretaceous. words, because the modern mammal orders If the fate of some other recent debates seem to Hedges et al. to have appeared long in mammalian molecular systematics is a after they diverged, the authors can justify guide, then some of the novel topologies using a Carboniferous calibration point in- undergirding these results may be incor- stead, and because the Carboniferous point rect. For example, Sullivan and Swofford is reasonable, they can infer that the modern (1997) have shown that a heated debate mammal orders appeared long after they ac- over the possible polyphyly of rodents (e.g., tually diverged. Tautological reasoning like D’Erchia et al., 1996) rested on inadequate this makes it impossible to move on to a rea- analyses. Regardless of topologies, most of soned discussion of the relative reliability of molecular clocks and paleontological data. 1 Present address and address for reprint requests: For example, if the Žrst appearances of mod- National Center for Ecological Analysis and Synthe- ern mammal orders are even vaguely close sis, University of California, 735 State Street, Suite 300, Santa Barbara, California 93101, USA; E-mail: to their true divergence times, this strongly [email protected]. suggests a speed up in the molecular clock

107 108 SYSTEMATIC BIOLOGY VOL. 48 of Hedges et al. (1996) sometime during patterns are of interest in this discussion: the 245 million years (MY) between the changes through time in the overall num- Carboniferous and the Cretaceous–Tertiary ber of species, and changes through time in (K-T) boundary. the distribution of morphologies (or other Even the better studies have crucial aws. attributes) across those species. Despite having employed multiple calibra- The fossil record does provide clear-cut tion points and trying tocorrect for variation evidence regarding both of these patterns. It in the clock speed, Springer (1997) arrived at shows that Cretaceous mammals were tax- a clock rate with alarmingly broad 95% con- onomically depauperate and morphologi- Ždence limits of ± 13% (“XR adjusted”) or cally uniform, and that the most important ± 15% (“MRR adjusted”). Furthermore, all radiation in the history of therian mam- but one of the calibration points fell within mals did occur in the earliest Paleocene. Ac- the Cenozoic, forcing the interordinal diver- cording to all interpretations, this radiation gence times to be based largely on extrapo- must have involved multiple, phylogeneti- lation instead of interpolation. This one Cre- cally independent lineages and therefore is taceous point was a supposed Žgure of 130 far more likely to have involved ecological million years ago (MYA) for the marsupial– release rather than a key evolutionary inno- placental split. However, the original source vation. The bulk of this paper will be de- (Novacek, 1993) did not discuss the 130 voted to the empirical problem of establish- MYA date, which was read off of an ar- ing that there was a Paleocene radiation. tistically rendered text Žgure that clearly implied the absence of any concrete evi- FOSSILS VERSUS MOLECULES dence for a split before 98 MY. The younger Cretaceous Splits among Mammal Orders: Do date has been conŽrmed by more recent re- Fossils and Molecules Really Disagree? search (Cifelli et al., 1997). Changing to a 98 Apart from possible lipotyphlan insec- MYA estimate—a minimum Žgure like all tivores (Novacek, 1993), no representa- of the other ones used by Springer (1997)— tives of a modern therian order have ever increases the clock rate, and therefore de- been clearly shown to occur in the Cre- creases all the estimated divergences, by taceous. Claims of Cretaceous xenarthrans 12%. were based on misidentiŽed multituber- Despite such concerns, my purpose is not culates (Krause, 1993). Claims of Cre- to challenge the inference that the basal taceous primates were based on earli- splits among many therian orders occurred est Paleocene Žnds of plesiadapiforms, sometime during the Cretaceous. Instead, which occur together with reworked Cre- I will make three simple points. First, the taceous mammals (Lofgren, 1995); the modern paleontological literature has never afŽnity of plesiadapiforms and primates implied that all therian or even eutherian or- is contentious and the two are at best sister ders diverged from one common ancestor groups (Beard, 1993). Claims of Cretaceous after the Cretaceous. Any claim to the con- marsupials in the broad sense are valid, the trary is a misinterpretation that makes the relevant fossils having been known since molecular results seem more novel than they the 19th century, but the Cretaceous forms really are. Second, most molecular studies have no clear-cut afŽnities with the modern, have failed to deŽne the idea of “radiation” ordinal- level marsupial groupings (Johan- or “diversiŽcation” in a rigorous manner, son, 1996). Claims of Cretaceous “ungulato- leading to inferences from data that are not morphs” are irrelevant because even if un- really relevant to the debate. Finally, clearer gulates are monophyletic, they are a clade deŽnitions imply that only two biological of at least six orders, not a single order, 1999 ALROY—PALEOCENE EVOLUTIONARY RADIATION OF MAMMALS 109 and may easily not have started to diver- the fossil record already has suggested— sify until the earliest Cenozoic (Archibald, not just these Žve orders but also sev- 1996). eral others do date back to the Creta- Nonetheless, traditional, paleontologi- ceous. cally based phylogenies imply that some basal splits among therians did take place What Would It Mean to Demonstrate a during the Cretaceous. The eutherian- Cretaceous DiversiŽcation? metatherian split dates to at least 98 MYA (Cifelli et al., 1997). The Carnivora (Fox and This remarkable lack of substantive dis- Youzwyshyn, 1994) and Mesonychia (a pos- agreement raises a key question: what ex- sible sister group of the Cetacea; Thewis- actly is the problem supposedly addressed sen, 1994) both appear in the earliest Pa- by the latest molecular studies? For exam- leocene. If “Archonta” and “Ungulata” are ple, Cooper and Penny (1997: 1109) claimed truly clades, they too deŽnitely occur at this there were “incremental changes during time (Novacek, 1993). The sister grouping a Cretaceous diversiŽcation of birds and of the Lagomorpha and Rodentia is con- mammals rather than an explosive radi- troversial, but there is substantial paleon- ation in the Early Tertiary,” which might tological evidence for the existence during imply that molecular data allow us to in- the latest Cretaceous of a separate clade fer the tempo of speciation and adaptation. that includes these two orders (Meng et al., But in discussing the Paleocene, Cooper 1994). Several orders are depauperate and and Fortey (1998: 152) declared that “the rarely fossilized (e.g., Tubulidentata, Pholi- explosive phases of evolution so amply dota, Macroscelida, Scandentia, and Der- demonstrated by the fossil record may, in moptera), so there is little to argue against many cases, have been preceded by an speculations that they might have origi- extended period of inconspicuous innova- nated in the Cretaceous. None of these con- tion.” In other words, the molecular data clusions is contentious among paleontolo- only imply “innovations” (equated with di- gists. For example, the widely reproduced vergences among modern orders), that may phylogenies of Romer (1966) and Novacek have had no consequences for the observed (1993) show multiple eutherian lineages sep- taxonomic and morphological diversity of arating from each other in the late Cre- mammals. taceous. Together, the most current pale- This weaker interpretation leaves little to ontological evidence suggests that at least argue about. All parties agree that several 7 or 8, and probably 10 or 20 living the- basal splits occurred during the Cretaceous, rian lineages do extend back into the Cre- which means that no one can argue seri- taceous. ously for the importance of a key evolution- In fact, with appropriate corrections for ary innovation in a single early Paleocene the previously mentioned error in cali- lineage as the driving mechanism for a ra- bration, the most methodologically sound diation. Not only that, but mammalian pa- analysis (Springer, 1997) implies that only leontologists have shown remarkably little Žve eutherian orders had split from their interest in the question of when exactly the sister groups by the K-T boundary (65 modern mammal orders diverged. Instead, MYA): Xenarthra, Insectivora, Primates Simpson (1952), Lillegraven (1972), Savage (the only member of “Archonta” included and Russell (1983), and Stucky (1990) all in that study), Lagomorpha, and Ro- focused on the pace of taxonomic diver- dentia. Far from overturning the tradi- siŽcation. These studies directly counted tional view, molecular studies conŽrm what orders, families, and genera in different time 110 SYSTEMATIC BIOLOGY VOL. 48 intervals, living or not. Meanwhile, paleon- trary, it shows that molecular workers have tologists who work on the early evolution already conceded that the fossil record is the of mammal orders have tended to focus on best means of documenting morphological documenting morphological transitions in- radiations. stead of exact dates of origin (e.g., Meng et If they are not directly addressing taxo- al., 1994; Thewissen, 1994). nomic or morphological diversity, what is Thus, the crucial empirical issue that the evolutionary import of molecular clock might be tested by molecular data is not studies? I would suggest that in fact they when exactly the modern mammalian or- have little to say about the theoretical prob- ders split from each other, but whether lem of evolutionary radiations, and instead taxonomic and morphological diversity ex- are of interest mostly to mammalogists who ploded in the early Paleocene instead of the want to know when particular mammalian Cretaceous. However, this traditional pale- clades originated. Nonetheless, it is still im- ontological concern simply has not been ad- portant to show just what the fossil record dressed by the molecular studies. The num- really does say about the evolutionary radi- ber of species that were present at different ation of mammals. times can in theory be inferred from compre- hensive molecular phylogenies (e.g., Nee et TAXONOMIC DIVERSITY al., 1992, 1994). But this requires sampling The most important features of a large either all living species or at least all living evolutionary radiation that can be inferred lineages that are believed to extend beyond from a phylogeny are these: (1) the num- a certain point in time. None of the molec- ber and timing of evolutionary divergences ular analyses of mammalian diversiŽcation and the tempo of taxonomic diversiŽcation have made any effort to guarantee this kind implied by these facts; and (2) the distribu- of comprehensive sampling. Therefore, it is tion of such attributes as morphology, phys- impossible to conclude from these studies iology, behavior, and biogeography across whether the rate of diversiŽcation was the a phylogeny, which might imply the tempo same or different during the Cretaceous and of ecological diversiŽcation. Although none Cenozoic. of these general issues have yet been ad- Meanwhile, molecular workers have dressed directly by molecular studies of the granted that phylogenetic topologies may mammalian radiation, all of them may be say little or nothing about the timing of the addressed by the fossil record. Here I will re- major morphological transitions that distin- analyze augmented versions of two data sets guish living orders (e.g., Cooper and Fortey, to show exactly what the differences were 1998). They even have used this argument between Cretaceous and Cenozoic mammal to suggest that paleontologists have failed faunas in North America. Most of the issues to recognize a greater diversity of surviv- regarding data preparation have been dis- ing lineages in the Cretaceous because Cre- cussed elsewhere (Alroy, 1992, 1994, 1996, taceous mammals had not yet evolved the 1998a, 1998b), so I will focus instead on diagnostic morphological features of their new analyses and results. I Žrst will treat living descendants. Indeed, all Cretaceous the problem of taxonomic diversiŽcation, mammals were terrestrial and ecologically and then discuss morphological evolution generalized, and, as I will show, all of them in terms of body mass distributions. occupied a narrow range of the size spec- trum. The important point is not that this Data might excuse the mismatch between pale- The faunal data used here are an ex- ontological and molecular data. To the con- tension of a previously discussed compi- 1999 ALROY—PALEOCENE EVOLUTIONARY RADIATION OF MAMMALS 111 lation of North American mammalian fos- blown analyses of trends in diversity and sil localities ranging in age from about 98 morphology, but the lists have been retained MYA (mid-Cretaceous) to 0.1 MYA (late for the purpose of constructing a chronolog- Pleistocene), which now number 4385. Be- ical framework. cause most of these localities pertain to a The calibrated event sequence can be used single quarry or a small outcrop, each serves directly to infer counts of the species that as paleontological “snapshot,” represent- existed at any of several arbitrary, evenly ing a short period of time in a restricted spaced moments in geological time. Togeth- geographical area. Each locality is docu- er, such counts would constitute a diversity mented by a taxonomically standardized, curve for the entire interval. Furthermore, species-level faunal list and whenever pos- counts of species appearing or disappearing sible is placed in a local stratigraphic sec- between sampling points can be used to esti- tion (the lists may be reviewed at http: mate speciation and extinction rates. Before // www. nceas. ucsb. edu/~ alroy/nampfd. doing so, however, a key problem must be html). solved: Each interval is represented by a dif- Instead of being pegged into a traditional ferent number of fossils, as shown by vari- time scale, the faunal lists are subjected to a ation in the number of faunal lists per MY multivariate ordination that is constrained (Alroy, 1996, 1998b). Such variations are ex- by the stratigraphy. In previous studies, the actly the “obvious deŽciencies” that make a ordination was governed by a parsimony “literal reading of the fossil record” so dan- criterion aimed at minimizing the number of gerous (Cooper and Fortey, 1998). temporal overlaps between species and/or Far from being an intractable handicap, genera. In the current analysis, I have mod- however, variation in sampling intensity iŽed the parsimony analysis into a maxi- may be removed. The best method is to stan- mum likelihood approach. The new method dardize sampling in each interval by draw- seeks to Žnd the combination of age- range ing faunal lists at random until reaching a boundaries and “nuisance” sampling pa- predeŽned limit. The limit is set by the total rameters for each taxon that is most likely number of faunal lists that are drawn; the to predict the observed pattern of overlap. cutoff is made as high as possible, given that Although the algorithmic details are of in- all or nearly all of the intervals should be terest, the parsimony and likelihood ap- able to reach it. After lists are drawn in each proaches yielded such similar results that interval and the temporal durations of taxa the exact choice of an ordination method is are recomputed on the basis of these lists, not relevant here. the procedure is iterated 100 times to yield The arrangement of lists implies a relative average diversity and turnover rates. “event sequence” of taxonomic Žrst and last In this paper the subsampling procedure appearances that is numbered from oldest to keeps track of the number of faunal lists, in- youngest and calibrated to numerical time stead of taxonomic records, that are drawn by use of geochronological age estimates in each trial. This is because alpha (within- (Alroy, 1992, 1994, 1996, 1998b). The current locality) diversity is much higher during version of the data set includes 198 strati- the Cenozoic than during the Cretaceous, graphic sections, 1223 genera, 3234 species, such that in the Cenozoic sampling the same and 153 geochronological calibration points. number of fossils is likely to yield a larger As previously, the statistical analyses focus number of distinct taxonomic records. Con- only on the relatively well-sampled western versely, 10 records in the Cenozoic are likely region of the United States and Canada. Lists to represent far fewer actual fossils than 10 older than 80 MYA are too few to allow full- records in the Cretaceous, but any 10 lists 112 SYSTEMATIC BIOLOGY VOL. 48 in each interval are likely to represent about the same number of fossils. In a previous analysis using an earlier ver- sion of the data set, I analyzed the Cenozoic data only and separated the diversity counts by 1.0 MY (Alroy, 1996). Because the calibra- tion of the time scale is poorer in the Creta- ceous, in the present study I used a longer bin size, 2.5 MY. The latest Cretaceous bin and all but one of the Cenozoic bins were able to meet a standardized sampling cutoff of 60 lists per bin (24 per MY). The under- sampled Cenozoic bin is for 42.5–40.0 MYA, for which only 22 lists were available. In the earlier study, I set a cutoff of 85 taxonomic records per 1.0 MY. Because each Cenozoic list averaged about 6.8 taxonomic records, that was equivalent to only 12.5 lists per MY. The current study’s relatively intense sampling is likely to recover any substan- tial differences between the Cretaceous and Cenozoic. Unfortunately, even though the 60-list sampling level is adequate for the last temporal bin of the Cretaceous, it is not for any of the preceding intervals. However, this lack of data should have little effect on the results because (1) my discussion will fo- cus on contrasts between the last, fully sam- pled Cretaceous interval and the Cenozoic; (2) all available lists from all of the Creta- ceous intervals before the last one will be sampled; and (3) computing full temporal ranges for the taxa (“ranging through”) will extend ranges into this ultimate Cretaceous interval for some taxa that were present but FIGURE 1. North American mammalian diversity, not directly sampled. origination (new appearance) rates, and extinction rates for the late Cretaceous and Cenozoic. Data are based on Results multivariate ordination and standardized sampling of faunal lists. Data for lists going back to 98 MYA are The diversity data (Fig. 1) establish three available but are not shown because they are not suf- key patterns. First, regional standing diver- Žcient for analytical purposes. (a) Standing diversity. sity was much lower during the Cretaceous The y-axis is log transformed to show the lack of ei- ther a log-linear (exponential) or asymptotic (simple than at any point during the Cenozoic (Fig. logistic) pattern; instead, an offset between two logis- 1A). For example, standing diversity was tic curves at about 65 MYA is indicated. (b) Origination about 23 species per 60 faunal lists across rates. Anomalously high rate for the 40.0–37.5 MYA in- the time plane at 67.5 Ma, but the total was terval is not shown because it is an artifact of under- sampling in the preceding interval. (c) Extinction rates. never lower than 41 after 65 MYA, and aver- Anomalously low rate for the 42.5–40.0 MYA interval aged 86 across the 25 Cenozoic time planes. is not shown because it is a sampling artifact. 1999 ALROY—PALEOCENE EVOLUTIONARY RADIATION OF MAMMALS 113

Second, there was an abrupt transition Cretaceous turnover rates were on average between the two diversity levels. Diver- lower than Cenozoic turnover rates. There sity surged shortly after the K-T bound- also is signiŽcant evidence that origination ary, reached a plateau by no later than rates are negatively correlated with stand- 45 MYA, and then uctuated dynamically ing diversity levels. This monotonic rela- within relatively narrow limits. The lack of tionship underlies the Cenozoic’s dynamic any true net diversiŽcation after this point equilibrium and partially explains why orig- can be shown in a simple way by focus- ination rates are much more variable than ing on the 18 data points from 45 MYA on, extinction rates. Finally, the major Cenozoic for which no signiŽcant correlation between North American mammal orders had dif- time and standing diversity can be demon- ferent diversity trajectories, suggesting that strated (Spearman’s rank-order correlation they were obeying different dynamic rules rs = –0.253; t = 1.045; n.s.). (Alroy, 1996, 1998b). The S-shaped pattern seen in this semi- The important point, however, is not the log plot is consistent with neither a simple exact nature of the evolutionary dynamic, exponential growth model, which predicts but rather the fact that these abundant and a linear curve, nor a simple logistic model, essentially unbiased data refute the two which predicts an asymptotic curve with no main inferences one might wish to draw visible lag in this kind of plot. Moreover, the from some of the recent molecular phyloge- pattern is not an artifact of poor sampling netic studies. First, diversiŽcation was not during most of the Cretaceous: Better sam- a slow and steady process: The early Pale- pling would only raise the Žrst several data ocene witnessed an immense evolutionary points relative tothe fully-sampled (but very radiation that went unmatched anywhere low) 67.5–65 MYA data point, which would else in mammalian history. This remarkable create the appearance of a Cretaceous de- pattern is visible even in the coarse 2.5 MY cline in the diversity and therefore make the bin data reported here. The closest matches subsequent Paleocene diversiŽcation seem to the 1.51 originations/species per 2.5 MY even more dramatic. Further evidence for rate for the earliest Paleocene are rates of just a dynamic Cenozoic equilibrium has been over 1 origination/species per 2.5 MY for presented previously (Alroy, 1996, 1998b). the intervals beginning at 62.5, 57.5, 30, and This study’s additional data suggest that 5 MYA. The remaining 18 Cenozoic inter- distinct Cretaceous and Cenozoic equilibria vals average only 0.45 originations/species were offset by the Paleocene radiation. per 2.5 MY, even if one includes the anoma- Finally, both origination and extinction lously high rate for the 40–37.5 MYA bin, rates increased dramatically around the which is an artifact of undersampling in K-T boundary: there were 0.75 extinc- the preceding time interval. In other words, tions/species per 2.5 MY bin just before 65 the early Paleocene origination rates were at MYA, and 1.51 originations/species per 2.5 least three times greater than background. MY just afterwards. Both curves remained Data computed with Žner bin sizes (Alroy, high during the Paleocene (roughly 65–55 1996, 1998b) yield essentially the same re- MYA); the three relevant data points aver- sult, and furthermore strongly suggest that age 0.60 extinctions and 1.07 originations per the most rapid phase of the radiation was species per 2.5 MY. However, these high val- conŽned to the very Žrst 1.0 MY of the Pa- ues do little to obscure the singular nature of leocene. Because only a fraction of Žrst ap- the K-T event. pearances in the Paleocene record might rea- Several further details could be discussed. sonably be attributed to anagenesis instead For example, there is weak evidence that of cladogenesis (Archibald, 1993), there is no 114 SYSTEMATIC BIOLOGY VOL. 48 way to discount this result as a taxonomic morphological disparity (sensu Foote, 1993), artifact. but this quantity is hard to measure across Second, contrary to some speculations the entire range of mammalian orders be- based on molecular data (e.g., Cooper and cause of the great anatomical differences Penny, 1997), there was in fact a mam- among them. For example, the only cheek malian mass extinction at the K-T bound- tooth that is found in every toothed mam- ary. The lengthy 2.5 MY bins used in this mal is the Žrst lower molar. Thus, it would study obscure the intensity of the event be nearly impossible to construct a disparity and also intensify some later (background) measure for all groups of mammals based rates that do not pertain to short-term on homologous morphological features of extinction episodes (e.g., for the 60–57.5 cheek teeth (but see Jernvall et al., 1996, for MYA bin). Nonetheless, the bin that includes “ungulates”). Worse, these very cheek teeth the K-T boundary still has virtually the high- are the only easily preserved and identiŽed est extinction rate in the time series (Fig. parts of the mammalian skeleton. Construct- 1C). Moreover, in the best available strati- ing morphospaces for, say, the postcranium graphic section, fully characteristic latest would therefore not be feasible in a general Cretaceous faunas from just meters below analysis of all fossil mammals (but see Janis the K-T boundary are replaced just above and Wilhelm, 1993, for large mammals). the boundary by almost completely differ- Fortunately, there is one very easily quan- ent Paleocene faunas (Archibald, 1982). The tiŽed morphological feature of overwhelm- best of these Cretaceous assemblages (Flat ing ecological importance for mammals: Creek 5) includes 24 species, which is an body mass, which is correlated strongly not almost complete inventory of the regional just with every linear measurement of the fauna. Of these species, just three have well- mammalian skeleton but also with dietary, established Paleocene records (Mesodma for- locomotor, and life history variables. De- mosa, M. thompsoni, Cimolestes incisus), and spite important residual variation among Žve more have possible Paleocene descen- species in such features, body mass distri- dants: M. hensleighi, C. propalaeoryctes, C. stir- butions do capture a considerable amount toni, Batodon tenuis, and Gypsonictops illu- of ecological information (Legendre, 1986). minatus (Lillegraven, 1969; Archibald and In this section I will outline Cretaceous and Bryant, 1990; Fox and Youzwyshyn, 1994). Cenozoic trends in body mass distributions, Therefore, even the most liberal estimate showing again that a profound biotic transi- suggests that as many as two-thirds of all tion did occur at the K-T boundary. mammal species went extinct during a rela- Data tively short interval bracketing the K-T im- pact event. The raw data discussed here have been described previously (Alroy, 1998a), so I will omit many details of data preparation. The MORPHOLOGICAL DISPARITY latest version of the data set consists of body If taxonomic diversity was lower during mass estimates for 1815 North American fos- the Cretaceous than during the Cenozoic, sil mammal species based on a compilation and if this transition was rapid and con- of published measurements for 19,363 in- Žned mostly to the early Paleocene, then dividual lower Žrst molars (m-1-s). Sepa- the only remaining arena for a possible rate regression equations are available for “Cretaceous radiation” would have to be each of the major mammalian orders (e.g., ecology. The best way to capture ecological Legendre, 1986; Damuth, 1990; Bloch et al., variation among fossil forms is to measure 1998), and for the others I used a generic all- 1999 ALROY—PALEOCENE EVOLUTIONARY RADIATION OF MAMMALS 115 mammal equation (Legendre, 1986). All of these equations have very high r-squared values; most of them relate the log of m- 1 length times width to the log of body mass, although for ungulates the log of m-1 length was used as the independent variable (see Damuth, 1990). Although no account is taken of such factors as sexual dimorphism, geographic variation, or within-species ana- genetic change, these all are inconsequential in light of the study’s shrew-to-mammoth size range. The data were used tocompute body mass distributions for 1.0-MY-long temporal bins. The taxonomic age-ranges that were used to determine the presence and absence of species in bins were based on the same multivariate ordination of faunal lists de- scribed earlier. Species were considered to be present in a bin if they ranged anywhere into it, which does occasionally result in the lumping together of species that never actually co-existed. The relatively short 1.0 MY bin length largely avoids this problem, but the data for the 66–65 MYA interval do lump classic latest Cretaceous faunas to- FIGURE 2. Trends through time in North American gether with a few small, earliest Paleocene mammalian body mass distributions. All species falling (“Pu0”) faunas. into each 1.0 MY bin are considered. (a) Mean body Trends in the size distribution were quan- mass. (b) Standard deviation of body mass. tiŽed in two ways. First, I computed the mean body mass across all species in each bin (Fig. 2A), which is an important vari- able because there is a strong trend toward such locality-speciŽc data for sampling ef- size increase (Alroy, 1998a). Second, I com- fects, one arrives at almost exactly the same puted the standard deviation of the same patterns that are seen in the lumped data. body mass values (Fig. 2B), which is impor- Second, additional features of the distri- tant because it is a direct measure of dispar- bution also might be quantiŽed, including ity (this latter term is typically equated with the skewness and kurtosis. However, these such measures of morphologicalvariation as statistics are noisy and add little to the key the range, variance, or standard deviation; conclusions. Finally, these trends apparently are governed by a complex, nonrandom dy- Foote, 1993). namic operating within evolutionary lin- Some caveats are in order. First, it is pos- eages (Alroy, 1998a). However, I will con- sible to compute all of these statistics sepa- strain my discussion to the pattern itself, rately for individual fossil localities, which instead of the underlying evolutionary pro- would avoid the problem of lumping species cesses, because the raw data alone are sufŽ- together in a temporal bin. However, prelim- cient to show that a Paleocene radiation oc- inary analyses indicate that after correcting curred. 116 SYSTEMATIC BIOLOGY VOL. 48

Results mean mass had increased by only another The body mass curves (Fig. 2) show four 1.70 ln g. clear-cut patterns: (1) Cretaceous mammals were small and occupied a very narrow range of the size spectrum (67–66 MYA bin: DISCUSSION n = 20, mean ± SD mass = 4.40 ± 1.98 ln One possible criticism of the preceding [natural log] g); (2) there was little change in results is that they pertain to just a single the Cretaceous fauna during a period of at continent. After all, only the North Amer- least 9 MY (76–75 MYA bin: n = 20, mass = ican record is relatively complete and well 3.86 ± 1.44 ln g, i.e., about the size of an ele- studied through both the latest Cretaceous phant shrew); (3) there was an abrupt shift and all of the Cenozoic. Thus, one could in the mean just around the K-T boundary, argue that molecular studies imply that a which resulted from the extinction of many signiŽcant Cretaceous radiation did take small mammals and the addition of many place, but happened not to do so on this medium-sized mammals (65–64 MYA bin: n particular continent. But even apart from = 40, mass = 6.42 ± 1.88 ln g); and (4) there the fact that the best molecular clock data was a steady expansion in the size range merely imply a few interordinal splits in- throughout the Cenozoic; mean body mass stead of a true evolutionary radiation, it is was forced to track this upwards trend be- not true that the bulk of living eutherian di- cause the lower limit to size was static (Fig. versity has its deep roots entirely outside of 1 in Alroy, 1998a). North America. Basal members of the Car- Was the K-T shift a true evolutionary nivora, Insectivora, Primates, Artiodactyla, event, or was it merely the side-effect of Perissodactyla all appeared in North Amer- immigration (speciŽcally from Eurasia) in ica during the early Cenozoic. Basal lago- the wake of a major mass extinction? Im- morphs, rodents, and cetaceans fail to do migration might seem important because, so, but on the other hand the fossil record for example, mean mass is 4.29 ln g in demonstrates that these groups originated the Flat Creek 5 assemblage (latest Creta- in Asia during the Paleocene, not the Creta- ceous) and 5.60 ln g, already much higher, ceous, and that some major lineages within in the apparently earliest Paleocene compo- these orders (e.g., Geomyoidea, Sciuromor- nent of the Bug Creek Anthills fauna (see pha) did originate in North America. There Lofgren, 1995). A few of the larger Bug are many minor groups that do seem to have Creek species are indeed most likely immi- originated in times and places where the fos- grants (e.g., Catopsalis joyneri, Stygimys kusz- sil record is not strong, but it simply is not mauli, and one or more of the three ungu- reasonable to suggest that North America lates: Archibald, 1982, 1993; Archibald and witnessed less than its fair share of mam- Bryant, 1990). However, some in situ spe- malian evolution. Instead, as a typical con- ciation may already have occurred at this tinent that long served as a crossroads for point, and in any case mean mass was still migrating lineages, North America is well- 0.8 ln g short of the average for the whole qualiŽed to serve as a testing ground for 65–64 MYA interval. Given that the later, models of mammalian evolutionary radia- post-Bug Creek Anthills shift was almost tion. certainly due to a rapid, in situ radiation, If we accept even such a limited defense of the overall transition seems to have been the North American data, there are few ways less an effect of immigration than of trends to avoid this study’s major conclusions: In within lineages combined with differential terms of both taxonomic diversity and body speciation of large forms. Regardless of the mass distributions, the single most impor- exact cause, the Cretaceous–Paleocene shift tant radiation of mammals occurred not dur- certainly had a more important long-term ing the Cretaceous, but during the early Pa- effect than any other sudden transition dur- leocene. Far from being a “literal reading of ing the Cenozoic: By the 1–0 MYA interval, the fossil record,” these are statistically ro- 1999 ALROY—PALEOCENE EVOLUTIONARY RADIATION OF MAMMALS 117 bust results based on standardized sampling brates; evidence from northeastern Montana. Geol. regimes. They show the folly of denying the Soc. Am. Spec. Pap. 247:549–562. Paleocene radiation on the basis of loosely BEARD, K. C. 1993. Phylogenetic systematics of thePri- matomorpha, with special reference to Dermoptera. calibrated molecular clocks. A more fruitful Pages 129–150 in Mammal phylogeny: Placentals (F. line of inquiry would be to take advantage of S. Szalay, M. J. Novacek, and M. C. McKenna, eds.). this obviously important event by exploring Springer-Verlag, New York. its impact on molecular evolution. For ex- BLOCH, J. I., K. D. ROSE, AND P. D. GINGERICH. 1998. New species of Batodonoides ample, if the extraordinary early Paleocene (Lipotyphla, Geolabididae) from the early Eocene burst of speciation and morphological evo- of Wyoming: Smallest known mammal? J. Mammal. lution was correlated with intense selection 79:804–827. at the molecular level, that may have created CIFELLI, R. L., J. I. KIRKLAND, A. WEIL, A. L. DEINO, 40 39 a simultaneous speed-up in the molecular AND B. J. KOWALLIS. 1997. High-precision Ar/ Ar geochronology and the advent of North America’s clock across most or all early Paleocene lin- Late Cretaceous terrestrial fauna. Proc. Natl. Acad. eages. In turn, such an effect might account Sci. USA 94:11163–11167. for some of the much-touted, but biologi- COOPER, A., AND R. FORTEY. 1998. Evolutionary explo- cally inconsequential discrepancies between sions and the phylogenetic fuse. Trends Ecol. Evol. molecular and paleontological data. 13:151–156. COOPER, A., AND D. PENNY. 1997. Mass survival of birds across the Cretaceous–Tertiary boundary: Molecular evidence. Science 275:1109–1113. ACKNOWLEDGMENTS DAMUTH, J. 1990. Problems in estimating body masses This research was supported by the Evolution of Ter- of archaic ungulates using dental measurements. restrial Ecosystems program at the Smithsonian. I thank Pages 229–253 in Body size in mammalian paleobi- D. Archibald,G. Eble, J. Hunter, M. Springer, P. Waddell, ology (J. Damuth and B. J. MacFadden, eds.). Cam- P. Wagner, and P. Wilf for comments on the manuscript. bridge Univ. Press, Cambridge, England. D’ERCHIA, A. M., C. GISSI, G. PESOLE, C. SACCONE, AND U. ARNASON. 1996. The guinea pig is not a rodent. REFERENCES Nature 381:597–599. FOOTE, M. 1993. Contributions of individual taxa ALROY, J. 1992. Conjunction among taxonomic distri- to overall morphological disparity. Paleobiology butions and the Miocene mammalian biochronology 19:403–419. of the Great Plains. Paleobiology 18:326–343. FOX, R. C., AND G. P. YOUZWYSHYN. 1994. New primi- ALROY, J. 1994. Appearance event ordination: A new tive carnivorans (Mammalia) from the Paleocene of biochronologic method. Paleobiology 20:191–207. western Canada, and their bearing on relationships ALROY, J. 1996. Constant extinction, constrained di- of the order. J. Vertebr. Paleontol. 14:382–404. versiŽcation, and uncoordinated stasis in North HEDGES, S. B., P. H. PARKER, C. G. SIBLEY, AND S. KUMAR. American mammals. Palaeogeogr. Palaeoclimatol. 1996. Continental breakup and the ordinal diversi- Palaeoecol. 127:285–311. Žcation of birds and mammals. Nature 381:226–229. ALROY, J. 1998a. Cope’s ruleand thedynamics of body JANIS, C. M., AND P. B. WILHELM. 1993. Were there mass evolution in North American mammals. Sci- mammalian pursuit predators in the Tertiary? ence 280:731–734. Dances with wolf avatars. J. Mammal. Evol. 1:103– ALROY, J. 1998b. Equilibrial diversity dynamics in 125. North American mammals. Pages 232–287 in Biodi- JANKE, A., X. XU, AND U. ARNASON. 1997. The complete versity dynamics: turnover of populations, taxa and mitochondrial genome of the wallaroo (Macropus communities (M. L. McKinney and J. Drake, eds.). robustus) and the phylogenetic relationship among Columbia University Press, New York. Monotremata, Marsupalia, and Eutheria. Proc. Natl. ARCHIBALD, J. D. 1982. A study of Mammalia and ge- Acad. Sci. USA 94:1276–1281. ology across the Cretaceous–Tertiary boundary in JERNVALL, J., J. P. HUNTER, AND M. FORTELIUS. 1996. Mo- GarŽeld County, Montana. Univ. Calif. Publ. Geol. lar tooth diversity, disparity, and ecology in Cenozoic Sci. 122:1–286. ungulate radiations. Science 274:1489–1492. ARCHIBALD, J. D. 1993. The importance of phyloge- JOHANSON, Z. 1996. Revision of the late Cretaceous netic analysis for the assessment of species turnover: North American marsupial genus Alphadon. Palaeon- A casehistory of Paleocenemammals in North Amer- togr. Abt. A Palaeozool.-Stratigr. 242:127–184. ica. Paleobiology 19:1–27. KRAUSE, D. W. 1993. Vucetichia (Gondwanatheria) is ARCHIBALD, J. D. 1996. Fossil evidence for a Late a junior synonym of Ferugliotherium (Multitubercu- Cretaceous origin of “hoofed” mammals. Science lata). J. Paleontol. 67:321–324. 272:1150–1153. KUMAR, S., AND S. B. HEDGES. 1998. A molecular ARCHIBALD, J. D., AND L. J. BRYANT. 1990. Differential timescale for vertebrate evolution. Nature. 392:917– Cretaceous/Tertiary extinctions of nonmarine verte- 920. 118 SYSTEMATIC BIOLOGY VOL. 48

LEGENDRE, S. 1986. Analysis of mammalian communi- ROMER, A. S. 1966. Vertebrate paleontology. Univer- ties from the late Eocene and Oligocene of southern sity of Chicago Press, Chicago. France. Palaeovertebrata 16:191–212. SAVAGE, D. E., AND D. E. RUSSELL. 1983. Mammalian LILLEGRAVEN, J. A. 1969. Latest Cretaceous mammals paleofaunas of the world. Addison-Wesley, Reading, of upper part of Edmonton Formation of Alberta, Massachusetts. Canada, and review of marsupial–placental di- SIMPSON, G. G. 1952. Periodicity in vertebrate evolu- chotomy in mammalian evolution. Univ. Kans. Pa- tion. J. Paleontol. 26:359–370. leontol. Contrib. 50:1–122. SPRINGER, M. S. 1997. Molecular clocks and the tim- LILLEGRAVEN, J. A. 1972. Ordinal and familial diversity ing of the placental and marsupial radiations in re- of Cenozoic mammals. Taxon 21:261–274. lation to the Cretaceous–Tertiary boundary. J. Mam- LOFGREN, D. L. 1995. The Bug Creek problem and mal. Evol. 4:285–302. the Cretaceous–Tertiary transition at McGuire Creek, STUCKY, R. K. 1990. Evolution of land mammal diver- Montana. Univ. Calif. Publ. Geol. Sci. 140:1–200. sity in North America during the Cenozoic. Pages MENG, J., A. R. WYSS, M. R. DAWSON, AND R. 375–432 in Current mammalogy, Volume 2 (H. H. ZHAI. 1994. Primitive fossil rodent from Inner Genoways, ed.). Plenum, New York. Mongolia and its implications for mammalian phy- SULLIVAN, J., AND D. L. SWOFFORD. 1997. Are guinea logeny. Nature 370:134–136. pigs rodents? The importance of adequate models in NEE, S., R. M. MAY, AND P. H. HARVEY. 1994. The recon- molecular phylogenetics. J. Mammal. Evol. 4:77–86. structed evolutionary process. Philos. Trans. R. Soc. THEWISSEN, J. G. M. 1994. Phylogenetic aspects of Lond. B 344:305–311. cetacean origins: A morphological perspective. J. NEE, S., A. Ø. MOOERS, AND P. H. HARVEY. 1992. Tempo Mammal. Evol. 2:157–184. and mode of evolution revealed from molecular phy- logenies. Proc. Natl. Acad. Sci. USA 89:8322–8326. NOVACEK, M. J. 1993. Reections on higher mam- Received 21 July 1998; accepted 15 September 1998 malian phylogenetics. J. Mammal. Evol. 1:3–30. Associate Editor: P. Waddell