Ancient mitogenomics elucidates diversity of extinct West Indian tortoises

We present 10 nearly complete mitochondrial genomes of the extinct tortoise Chelonoidis alburyorum from the Bahamas. While our samples represent morphologically distinct populations from six islands, their genetic divergences were shallow and resembled those among Galápagos tortoises. Our molecular clock estimates revealed that divergence among Bahamian tortoises began ~ 1.5 mya, whereas divergence among the Galápagos tortoises (C. niger complex) began ~ 2 mya. The inter-island divergences of tortoises from within the Bahamas and within the Galápagos Islands are much younger (0.09–0.59 mya, and 0.08–1.43 mya, respectively) than the genetic differentiation between any other congeneric pair of tortoise species. The shallow mitochondrial divergences of the two radiations on the Bahamas and the Galápagos Islands suggest that each archipelago sustained only one species of tortoise, and that the taxa currently regarded as distinct species in the Galápagos should be returned to subspecies status. The extinct tortoises from the Bahamas have two well-supported clades: the first includes one sample from Great Abaco and two from Crooked Island; the second clade includes tortoises from Great Abaco, Eleuthera, Crooked Island, Mayaguana, Middle Caicos, and Grand Turk. Tortoises belonging to both clades on Great Abaco and Crooked Island suggest late Holocene inter-island transport by prehistoric humans.

Processing of subfossil Chelonoidis samples. All samples were processed according to established guidelines 22 in the aDNA facility of the Senckenberg Natural History Collections Dresden, which is physically separated from the main molecular genetic laboratory in another building. Negative controls (water blanks) were included during DNA extraction and library preparation and screened for evidence of contamination. Most bones were sampled using a Proxxon Micromot 50/E multitool equipped with 2-4 mm metal or stone drilling bits. Only the samples of C. lutzae/C. cf. lutzae were received as bone powder. Approximately 50 mg of bone powder of each sample was processed according to a protocol optimised for the recovery of short DNA fragments (Table S1) 23 . Then, up to 14 ng of DNA was converted into single-indexed, single-stranded Illumina  5,6,[9][10][11]21 , yellow circles: samples unsuccessfully processed for aDNA, red circles: samples successfully processed for aDNA. Grey areas show the shelf to a depth of 200 m. Map was created using ArcGIS 10.4 (https ://www.esri.com/en-us/arcgi s/about -arcgi s/overv iew) and Adobe Illustrator CS6 (http://www.adobe .com/produ cts/illus trato r.html).  25 , including the removal of uracil residues by uracil-DNA glycosylase (UDG) treatment. In order to increase the amount of endogenous mitochondrial DNA in the libraries, two-rounds of in-solution hybridization capture 26,27 were performed in a dedicated capture-only workspace in the main laboratory using DNA baits generated from long-range PCR products of C. carbonarius, C. chilensis, C. denticulatus, and C. vicina at an equimolar rate. Details for long-range PCR, primer sequences, and PCR conditions are explained in the Supplementary Information. Sequencing was performed in-house on an Illumina MiSeq platform, generating 75 bp-long paired-end reads.
Mitogenome sequence assembly and sequence annotation. Assembly of mitogenome sequences from the enriched libraries involved adapter trimming with Skewer 0.2.2 28 , read merging (minimum length 35 bp), quality filtering (minimum Q-score 20) and duplicate removal with BBmap-suite 37.24 (https ://sourc eforg e.net/proje cts/bbmap /) 29 . The remaining reads were screened for contamination using FastQ Screen 0.11.4 30 and a set of predefined mitochondrial genomes (Table S2). The identified non-target reads were excluded from the readpools, which are subsequently referred to as readpools 1. Then, all reads mapping to a reference genome of Chelonoidis were copied into a second readpool (readpools 2). Genome assembly was achieved with MITObim 31 , a two-step baiting and iterative mapping approach, with an allowed mismatch value of 2 and a starting seed according to Table S1. For each sample, readpool 2 was used for the initial building of a reference genome, whereas readpool 1 was used for the actual assembly. Resulting scaffolds were visualised and checked for assembly artefacts in Tablet 32 . Assembly artefacts were manually removed from the assembled contigs and all positions with a coverage below threefold masked as ambiguous (N) using the maskfasta subcommand of BEDTools 2.29.2 33 . Sequence length distribution of mapped reads was calculated with a customised awk command and Microsoft Excel.
Alignment and phylogenetic, divergence time and biogeographic analyses. The newly generated mitogenomes of Chelonoidis were merged with our previously used annotated alignment 34 for tortoises, with the exception that only one sequence per Cylindraspis species was included. This alignment also contained previously obtained data for a subfossil specimen of Chelonoidis alburyorum from Sawmill Sink, Great Abaco, Bahamas 19 . We also included 15 recently published sequences of Chelonoidis spp. from South America and Galápagos 35 and removed a GenBank sequence for a Galápagos tortoise of unknown provenance (accession number JN999704) from the original alignment 34 .
To briefly summarise the approach, an automated preliminary alignment was generated using Clustal W 1.4 36 and default parameters, as implemented in BioEdit 7.0.9.0 37 . This alignment was adjusted manually, and sequences were annotated using MITOS 38 and several published tortoise mitogenomes (GenBank/ENA accession numbers: AF069423, DQ080042, DQ080048, FJ469674, KT613185, LT599485). Each coding region was  20 . Specimens are in the collection of the National Museum of The Bahamas (NMB), except for Grand Turk and Middle Caicos, which are in the Turks and Caicos National Museum and for which we provide field provenience. See Table S1 for further details. *Sequenced in a previous study 19 . **Two age ranges due to natural atmospheric variation through time in abundance of 14  Phylogenetic relationships of the mitogenomes were inferred with Maximum Likelihood (ML) and Bayesian Inference (BI) approaches using RAxML 8.0.0 40 and MrBayes 3.2.6 41 . The best evolutionary models and partitioning schemes (Tables S3, S4) were determined with PartitionFinder2 42 applying the greedy search scheme and the Bayesian Information Criterion. For ML, 20 independent searches were carried out using the GTR + G substitution model, different starting conditions, and the rapid bootstrap option. Subsequently, 1000 non-parametric thorough bootstrap replicates were calculated and the values plotted against the best tree. For BI, four parallel runs (each with eight chains) were performed with 2 million generations (burn-in 0.25; print frequency 1000; sample frequency 1000). Calculation parameters were analysed using Tracer 1.7.1 43 . In addition, uncorrected p distances were calculated in MEGA X 39 using the pairwise deletion option. Divergence times were estimated using the uncorrelated lognormal relaxed clock models implemented in BEAST 1.84 44 and constrained by four fossil calibration points following Kehlmaier et al. 34 (Table 2). Further details for analyses are explained in the Supplementary Information.

Results
Ten of the 17 subfossil Chelonoidis samples produced high-quality data representing nearly the entire mitochondrial genome (15,350 bp length, coverage: ninefold to 482-fold; Table S1). This is an excellent yield for aDNA from tropical environments 19,45 , especially when it is considered that some specimens were from open unsheltered sites (Table 1).
For the mitogenomes, only the control region and part of adjacent DNA coding for tRNAs could not be reconstructed. All successfully assembled mitogenomes belonged to extinct tortoises from the Bahamas. The holotype of Chelonoidis dominicensis from the Dominican Republic, the material from Argentina (holotype of C. lutzae, three samples of C. cf. lutzae), the single species-undetermined Chelonoidis from Venezuela, and one Chelonoidis sample from Long Island, Bahamas, did not yield sufficient endogenous DNA. Assembly details of individual samples and blanks as well as genetic diversity indices and substitution rates for selected clades are provided in the Supplementary Information.
Our phylogenetic analyses and the molecular clock calculation, including the 10 new and one previously published mitogenomes of West Indian Chelonoidis and additional data for Chelonoidis species from Galápagos and South America, produced general tree topologies consistent with our previous studies 19,34 . Of particular interest are the relationships of the crown clade containing C. chilensis and the Chelonoidis species from the Bahamas and Galápagos. These taxa were placed in a maximally supported clade, although the branching pattern within that clade was weakly resolved. Accordingly, the ML and BI trees suggested that C. chilensis is sister to a weakly supported clade consisting of the two island clades (Fig. 2), whereas our time tree reflected the weakly supported topology of our previous studies, with C. chilensis as sister taxon of the Galápagos tortoises (Fig. 3). Thus, the divergence time inferred for this node should be taken with caution. In contrast, the two clades containing giant tortoises from the Bahamas and Galápagos were both maximally supported (Fig. 2).
Our 10 new samples from the Bahamas, and the previously sequenced specimen 19 , represented morphologically highly divergent populations from six islands 11,16 . Nevertheless, the genetic divergences were shallow and resembled those among Galápagos tortoises (Fig. 2). Our molecular clock estimated a mean of 2.34 × 10 -3 substitutions per site per million years (95% HPD: 2.11 × 10 -3 -2.78 × 10 -3 ). Divergence among tortoises from the Bahamas commenced approximately 1.5 million years ago (mya), while divergence among tortoises from the Galápagos Islands began approximately 2 mya (Fig. 3). Two samples of the congeneric species Chelonoidis carbonarius from South America were more different genetically (with a divergence time of ~ 3.7 mya), as were two other pairs of congeneric tortoises (the extinct Cylindraspis indica from Réunion and its sister species C. inepta from Mauritius, that diverged ~ 4 mya, and the sister species C. peltastes and C. vosmaeri from Rodrigues, which diverged ~ 4.3 mya). In contrast, the divergence between two samples of Chelonoidis denticulatus was shallow and estimated to be ~ 0.6 million years old.
The inter-island divergences of tortoises from within the Bahamas and within the Galápagos Islands are much younger (0.09-0.59 mya, and 0.08-1.43 mya, respectively) than the genetic differentiation between any other congeneric pair of tortoise species (in Figs. 2 and 3 from top to bottom: Psammobates (14.6 mya), Astrochelys  0 mya). The mitochondrial divergence of subspecies of Testudo graeca (7.5 mya) also resembles or exceeds that between many sister species. Within the extinct tortoises from the Bahamas, there are two well-supported clades (Figs. 2, 3), each of which died out from ~ 900 to 700 years ago (Table 1). One clade includes sequences from one sample from Great Abaco and two tortoises from Crooked Island; the other clade contains sequences corresponding to other tortoise specimens from Great Abaco and Crooked Island, as well as ones from Eleuthera, Mayaguana, Middle Caicos, and Grand Turk. The Galápagos Islands have six or seven clades of tortoises with genetic divergences similar to those of the two Bahamian clades.

Discussion
Our study provides evidence that the extinct tortoises from the Bahamas had diverged very little genetically, despite pronounced morphological differences 11,16 . For example, while all of the Bahamian tortoises were large, the one from Lost Reel Cave on Great Abaco was by far the largest, approaching if not matching in size the largest extant tortoises from the Galápagos and Aldabra. The various Bahamian forms also displayed major differences (not age-related) in the shape, rugosity, and relative size of the entoplastron, epiplastron, hypoplastron, and xiphiplastron. It was upon these differences that Steadman et al. 20 tentatively recognized that as many as seven distinct species of Bahamian tortoises existed. Subsequently, two of these forms were described as subspecies of Chelonoidis alburyorum, namely C. a. sementis from Middle Caicos and C. a. keegani from Grand Turk 11 (Table S5). According to our molecular clock, the Bahamian radiation commenced approximately 1.5 mya, i.e., about 500,000 years after the onset of the radiation of the giant tortoises on the Galápagos Islands. Another recently published estimate for the age of the Galápagos tortoise radiation is slightly younger than ours (1.5 mya instead of 2.0 mya) 35 , resembling our molecular clock results for the Bahamas.
The Galápagos Islands formed approximately 4 mya 46 , whereas the age of the Bahamas, from the standpoint of being able to support terrestrial vertebrates, is estimated not to exceed 400,000 years because of interglacial flooding during marine isotope stage 11 [47][48][49] . This situation implies for the Galápagos Islands that the current diversity of giant tortoises resulted from a single colonization event and a local radiation on the archipelago 35 . In contrast, the two tortoise clades from the Bahamas seem to be too old for having diverged on the islands. This suggests that the Bahamas may have been colonized twice from other landmasses. If that was the case, then the two colonizers must have been very similar genetically, given the low amount of genetic divergence of tortoises across www.nature.com/scientificreports/ the island group. As far as known, the extinct giant tortoises from the Greater Antilles, which would seem to be the likely source region of the Bahamian tortoises, are morphologically clearly distinct 11,21 . We cannot exclude, however, that these differences reflect morphological plasticity, a phenomenon well known from many other tortoise taxa [50][51][52][53] . Alternatively, the ancestral taxa on the Greater Antilles may still be undiscovered or known only by material too fragmentary to discern the crucial morphological characters. (The majority of tortoise fossils from the Greater Antilles are represented only by very incomplete material 21 ). Unfortunately, the only sample from the Greater Antilles that we studied (ulna from the holotype of C. dominicensis, radiocarbon-dated to the early Holocene) did not yield aDNA sequences, so that genetic evidence must await further investigation. Nevertheless, C. dominicensis remains valuable for morphological studies, such as its possession of an epiplastral shelf, which is characteristic of the Galápagos and South American clades of Chelonoidis but not the Bahamian clade 21 .
An unexpected result of our study was that we found tortoises belonging to the two Bahamian clades on the same islands. One of these clades was represented only by two tortoises from Crooked Island (1702 Cave) and the previously sequenced specimen from Great Abaco (Sawmill Sink) 19 . The two specimens from Crooked Island are ~ 2600 years old, and therefore pre-cultural, whereas that from Abaco is ~ 950 years old (Table 1). Human arrival in the Bahamas took place ~ 1200 to 1000 years ago 17,20,54,55 . This situation suggests prehistoric human transport of tortoises from Crooked Island to Great Abaco, which lies on a different (and distant) bank.
The second Bahamian clade contained one specimen each from these same two islands, namely from McKay's Bluff Cave on Crooked Island, and from Lost Reel Cave on Great Abaco, as well as specimens from Eleuthera, Mayaguana, Middle Caicos, and Grand Turk. This yields a total of six islands on six different banks (Fig. 1). Because the late Holocene specimens post-date the arrival of humans (Lucayans) in the Bahamas, the inter-island mixing of clades is strong evidence that early people were moving tortoises among islands, which could have also contributed to hybridization and an increase in morphological variation.
A similar situation was discovered recently with aDNA of the Bahamian hutia (Geocapromys ingrahami), again involving prehistoric human transport between Great Abaco and Crooked Island 56 . Whether tortoises or hutias, their inter-island transport by people did not prevent their eventual extinction on both Great Abaco and Crooked Island. We have no evidence that indigenous tortoises survived in the Bahamas beyond 800-700 cal BP 20 , which is several centuries before European contact.
In any case, compared to other tortoises (Figs. 2, 3; Table S6), the shallow mitochondrial divergences of the tortoise radiations on the Bahamas and the Galápagos Islands suggest that each archipelago harboured only one species and that the many taxa currently regarded as distinct species 1,35 should be returned to subspecies status. This also is in agreement with the weak nuclear genomic divergence of Galápagos tortoises 57,58 .
Conspecificity is further supported when mitogenomic divergences within other Chelonoidis species are compared to those of the two island radiations. Two samples of the widely distributed South American species C. carbonarius were estimated in our molecular clock calculation to have diverged ~ 3.7 mya, and these samples differed by an uncorrected p distance of 1.7% compared to maximum values of 0.7% within the Bahamian radiation and 0.9% within the Galápagos radiation. The shallower divergence between two samples of another widely distributed South American species, C. denticulatus (~ 0.6 mya; 0.2% uncorrected p distance), is in line with a previous study 59 that found the savannah species C. carbonarius more differentiated than its forest-dwelling sister species C. denticulatus.
Even though the genetic divergences among the tortoises of the Bahamas (and, for that matter, the Galápagos Islands) are small, each of the populations had a distinctive morphology with which it interacted in its environment. Both archipelagos have gradients of temperature and precipitation, yielding distinctive environments and vegetation types on different islands. Giant tortoises are known to play important roles for the vegetation structure and composition on other islands 60,61 , and undoubtedly once did the same on the Bahamas.