Conventional paleontological models 1234 of North American mammoth evolution posit that at least two species occupied the continent during the late Pleistocene (150,000 to 10,000 years ago: Mammuthus primigenius (woolly mammoths (WMs)) evolved in Eurasia and immigrated to North America in the late Pleistocene, whereas Mammuthus columbi (Columbian mammoths (CMs)) evolved locally from an earlier Pleistocene immigrant ancestor (Mammuthus meridionalis 12 or Mammuthus trogontherii 34). The species are morphologically differentiated by physical size (CMs were some 25% taller than WMs 5), molar complexity (CMs displayed more 'primitive' crown height and lamellar configuration), and skull morphology (CMs possessed a more downturned mandibular symphysis and more laterally oriented tusk alveoli) 15. Some of these traits are considered adaptations to their disparate habitats: WMs inhabited cold and arid periglacial regions, while CMs inhabited the temperate regions of the southern latitudes. Continental populations of both species went extinct during the Pleistocene-Holocene transition some 10,000 years ago.

Recent paleontological reconsiderations 678 and mitochondrial DNA (mtDNA) phylogeographic studies of predominantly Beringian mammoths 910111213 reveal a complex evolutionary history (Figure 1a). Their populations harbored diverse genetic lineages, two of which, haplogroups A and C, were endemic to Eurasia and North America, respectively. Certain population dynamics - including major immigration/replacement events and regional genetic introgression - have been offered as explanations for this complexity 1011, but its precise origins have proven difficult to define within the broader context of Pleistocene biogeography and paleoecology. This is the case for at least two reasons: first, key coalescent dates remain difficult to measure, in large part due to lack of sequence breadth and methodological shortcomings 1415; and second, almost nothing is known about the mtDNA phylogeny of Mammuthus beyond Beringian late Pleistocene mammoths (and thus probably exclusively M. primigenius). One potential solution to both problems - and means to hone conceptions of Pleistocene mammoth evolution in general - is to sequence DNA from one or more closely related but distinct mammoth species and use it as a temporal and taxonomic calibration tool within the mammoth gene tree. Owing to their apparently separate evolutionary history (Figure 1a) and reasonably well-dated recent divergence from WMs about 1 to 2 million years ago 16, CMs are excellent candidates for this role. To this end we targeted CM remains for mitogenomic sequencing.

We selected the Huntington Mammoth 17 for this purpose on account of its secure morphological identification, direct radiocarbon date (11,220 ± 110 ¹⁴Cya (radiocarbon years ago), exceptional biomolecular preservation 18 and geographic provenience (Fairview, UT, USA), far south of the Wisconsinan glaciers. Typical strategies for DNA sequencing of paleontological specimens would employ a pre-sequencing targeted enrichment approach, through the use of either labor-intensive PCR or hybridization techniques 19. However, following serial extraction and library preparation of our specimen, a quantitative PCR-based metric projected a sufficient ratio of target to non-target DNA to warrant a shotgun-based metagenomic sequencing approach, for which we employed the Illumina platform (see Materials and methods; Additional file 1).

Of the over 27 million reads longer than 50 bp obtained from the Huntington sample library, between 6,000 and 9,000 (0.02 to 0.03%) mapped to a WM reference mitogenome [GenBank: NC007596.2] 20 depending on software assembly parameters (Table S3 in Additional file 2). This provided an average unique read depth of approximately 23 × for the entire mitochondrial genome, excluding the VNTR region (positions 16,157 to 16,476). Roughly 2 million reads also mapped to the African elephant (Loxodonta africana) nuclear genome, providing approximately 0.03 × coverage of the entire nuclear genome of the animal, and bringing the total likely mammoth DNA read count to approximately 7% of all sequences. Such a proportion of total endogenous DNA is consistent with taphonomic models for DNA preservation in temperate burial contexts 21, as well as experimental data from other non-permafrost remains 2223. The coverage depth ratio we observe between mitochondrial and nuclear reads (approximately 800 ×) also falls within the range estimated in other mammoth specimens (245 to 17,000 × 24). This low nuclear read coverage depth also lends evidence that potential Numts make no significant contribution to the consensus generated from the mitochondrial assembly.

To ensure the authenticity of the mitogenome sequence, we amplified, cloned and sequenced PCR products of WM haplotype-defining regions of the cytochrome b gene and hypervariable region from multiple extractions of the Huntington mammoth in two separate ancient DNA facilities. These all yielded consensus sequences 100% identical to the shotgun consensus where they overlapped. Furthermore, we sequenced the same loci from PCRs of another securely identified M. columbi (the Union Pacific mammoth, University of Wyoming 6368, found near Rawlins, WY, USA 2526), which yielded identical sequences to those acquired for Huntington. Finally, to control for ascertainment bias in assembly of the whole mitogenome, we mapped the Illumina sequencing reads to an Asiatic elephant (Elephas maximus) mitogenome [GenBank: DQ316068] and obtained a 99.98% identical consensus sequence where it overlapped with the WM assembly consensus. Thus, we are confident that the final Huntington mammoth mitogenome sequence derives from the genuine endogenous mtDNA of the animal.

Bayesian phylogenetic analysis demonstrates that the Huntington mammoth mitogenome is largely indiscernible from those of endemic North American WMs (Figure 1b). For all model and parameter variants (Table S7 in Additional file 2, Figures S3, S4, S5, S6, S7, and S8 in Additional file 3), the sequence sorts securely within haplogroup C, a subclade additionally represented by dozens of WMs from Alaska and the Yukon 11. To test for this relationship at the entire mitogenomic level, we also sequenced the first complete mitogenome of a WM from this haplogroup (IK-99-70, from the Alaskan North Slope, USA), which confirmed Huntington's phylogenetic position within haplogroup C (Figure 1b).

At first glance, these results would suggest that, contrary to a strict interpretation of traditional paleontological models for their evolution, CMs and WMs did not descend from populations that were wholly separate since the early Pleistocene. One interpretation could be that mitochondrial haplogroup C corresponds to descendants of immigrant mammoth populations that ultimately gave rise to M. columbi. But without expansion, this interpretation would fail to explain why haplogroup C belongs to mammoths with both CM and WM morphologies. Indeed, certain paleontological interpretations have already suggested that CMs and WMs were more closely related than typically thought, even 'geoclinal or chronoclinal variants' 27 descending from a very recent common ancestor. We find that our results also warrant consideration of an alternative scheme, one that operates within existing paleontological models but that accommodates incomplete reproductive barriers between CMs and WMs during some period(s) of their evolutionary history.

mtDNA phylogenies are often inconsistent with species phylogenies 28, especially for populations with sex-biased dispersion and breeding patterns. This is particularly true for extant elephants 2930, which exhibit male-mediated gene flow between matriarchal herds, rendering their mtDNA phylogenies incomplete representations of breeding history. For example, Asiatic elephant and WM populations both harbor(ed) at least two highly divergent mitochondrial lineages without corresponding morphological differentiation 9101131. Between CMs and WMs, we observe the opposite situation, where their morphological distinction appears to have little mitochondrial genetic correlation. One potential explanation for this is that incomplete lineage sorting (ILS) resulted in the maintenance in CM populations of what ultimately became more WM-like mitochondrial lineages. However, if this were the case, we would expect the CM-WM most recent common ancestor (MRCA) to be positioned much deeper in the cytochrome b/hypervariable region phylogeny than observed. Our and previous 11 dual-calibrated estimates for the MRCA for the entirety of haplogroup C dates to the middle Pleistocene (Table S7 in Additional file 2), with the CM-WM MRCA necessarily occurring much more recently, long after their purported species divergence. That said, the haplogroup C full mitochondrial dataset is too small to completely rule out ILS during CM-WM speciation as a plausible explanation.

At present, however, we suspect that hybridization between CMs and WMs may be a more parsimonious explanation for our observations. Under one conception, haplogroup C could have been a predominantly CM haplogroup that introgressed into WM populations, at such a frequency that it came to dominate the North American mitochondrial gene pool of that species. The fact that both CMs sequenced here are haplogroup C would lend some support to this hypothesis. Another possibility is that introgression occurred in the opposite direction, such that WM-typical haplogroup C introgressed into CM populations (Figure 2a). From a behavioral perspective, this configuration is perhaps more likely, especially in light of phenomena documented in extant African forest (Loxodonta cyclotis) and savanna (L. africana) elephants (Figure 2b). These living species are morphologically distinct and deeply divergent at many nuclear loci 32333435, but are known to interbreed at forest-savanna ecotones 3637. The result is 'cytonuclear dissociation' 38 between genomes in hybrid individuals, such that forest-typical mitochondrial haplotypes occur at low frequency in savanna populations. Hypothetically, this is driven by savanna males reproductively out-competing physically smaller forest males 38, producing unidirectional backcrossing of hybrid females into savanna populations. Since mammoths were probably very similar to modern elephants in social and reproductive behavior 427, it is conceivable that WMs and the physically larger CMs engaged in a similar dynamic when they encountered each other. Indeed, hybridization between CMs and WMs has already been suggested by others 39, and genetic exchange may explain mammoths bearing CM-WM intermediate morphologies. Such mammoths are frequently found in areas where CMs and WMs overlapped in time and space, such as the Great Lakes region 2. Some of these apparent intermediates have been formally named (for example, Mammuthus jeffersonii), but their taxonomic identity is questionable. Indeed, the large number of synonyms currently registered for North American mammoths 40 is at least partly a function of efforts by earlier systematists to come to grips with the large amount of morphological variation expressed within Mammuthus (or Elephas). Although the Huntington mammoth exhibits no such morphological intermediacy and was found quite distant from the documented WM range, its status as a genetic hybrid would not be inconsistent with the modern analog: forest haplotype-bearing savanna elephants can be found several thousands of kilometers from modern ecotones, bearing no phenotypic indication of hybridism 38.

Both the ILS and introgression hypotheses discussed above provide straightforward testable predictions. First, under a WM-CM introgression scenario, some presently unidentified and distinct mitochondrial haplogroup should characterize a significant percentage of CM lineages, rendering their mitogenomes polyphyletic, as they are in L. africana (Figure 2). While we also observe a likely C haplotype in short sequences from one other well-identified terminal Pleistocene M. columbi, only a broad population-level survey of CM genetic diversity can rigorously test this prediction. Second, under the introgression hypothesis, CMs with WM-type mitogenomes should possess nuclear genes that are significantly more divergent from WMs than all haplogroup C mammoths are from each other. On the other hand, an ILS scenario would predict that CM and WM nuclear genes should show a similar degree of divergence as is detected between haplogroup C mitogenomes. Though we did recover several million nuclear sequences from the Huntington DNA library, the very low coverage depth provided by these reads is not sufficient for reliable nuclear divergence estimates between CMs and WMs. However, we anticipate that targeted enrichment techniques 4142 prior to high-throughput sequencing will provide the necessary coverage depth to test these hypotheses in the near future.

The revealed mitochondrial phylogenetic position of M. columbi does not immediately clarify complexities and chronological uncertainties previously observed in mammoth mtDNA phylogeny. Instead, it emphasizes that the unique reproductive behavior of elephantids necessitates a multi-genomic approach to characterizing their evolutionary history, as has been so effectively used in studies of living elephants. Their very recent mitochondrial common ancestry strongly suggests that CMs and WMs interbred at some point, most likely post-dating their morphological divergence, and in a fashion that confounds simple correlation of mtDNA phylogeny to evolutionary models derived from mammoth morphology alone. However, the precise mode and setting of genetic interchange between WMs and CMs are elusive, and therefore all hypotheses explaining our observations warrant testing. The possibility that hybridization explains our data is particularly tantalizing, since in many animals, interspecific hybridization accompanies population displacement and/or expansion resulting from habitat reconfiguration 4344. Thus, interbreeding between extinct late Pleistocene taxa - especially keystone herbivores like mammoths - could serve as an indicator of major ecological events, including those surrounding the megafaunal extinctions. Our results demonstrate that the use of next-generation sequencing technologies holds promise in rigorously testing such hypotheses using full ancient genomic data, even from non-cave, non-permafrost Pleistocene depositional contexts.

¹⁴Cya, Radiocarbon years ago; CM, Columbian mammoth, Mammuthus columbi; ILS, incomplete lineage sorting; MRCA, most recent common ancestor; mtDNA, mitochondrial DNA; PCR, polymerase chain reaction; WM, woolly mammoth, Mammuthus primigenius.

The authors declare that they have no competing interests.

JE, RD, DO, DF, RM and HP planned and designed the project. JE, RD, AD and CK performed wet laboratory work. JE, RD, AD, TT, and SS performed assembly and analysis of the sequence data. JE wrote the paper with assistance from all authors.