Continuity of eastern Beringian megafauna phylogenetic diversity following deposition of the Late Pleistocene Dawson tephra

Abstract

Volcanic eruptions can deposit ash, or tephra, across the landscape hundreds to thousands of kilometers from the source volcano. The ecological and genetic effects of ancient ashfall events are only known sparsely from the geological record and from modern genomic studies. About 29,000 years ago, a volcanic eruption in the Aleutian archipelago deposited ash across the eastern Beringian mammoth steppe. The resulting Dawson tephra is visible in loess deposits in interior Alaska and Yukon, with thicknesses ranging from 5 to 80 cm at a distance of 1,700 km from its source. To explore the impact of this ashfall event on ecosystem dynamics, we used sedimentary ancient DNA (sedaDNA) approaches to isolate population and community level genomic data from several exposures of Dawson tephra in the Klondike region of Yukon, Canada. We found no significant changes to species assemblages associated with the ashfall event, and no significant change in megafauna phylogenetic diversity after the ashfall. Our results suggest that the Beringian mammoth steppe ecosystem was resilient to ecological disturbance and reinforce the value of sedaDNA to investigate past environmental dynamics.

Introduction

Large volcanic eruptions can have profound long-term and short-term ecological impacts, both locally and regionally (Del Moral and Grishin 1999). Volcanic eruptions can deposit ash across vast distances, sometimes burying distant ecosystems with short-term effects on plants and animals (Ayris and Delmelle 2012) and global impacts on climate (Meronen et al. 2012). How quickly a community can recover from tephra deposition depends on the thickness, grain size, and chemistry of the tephra as well as vegetation type, although plant communities generally recover to pre-eruption conditions within a century (Arnalds 2013; Allen and Huntley 2018). In environments prone to frequent disturbance, such as the cold steppe habitat of ancient Italy (Allen and Huntley 2018), full recovery may occur within a few decades. However, evidence of volcanic impacts may remain encoded in animal genomes for thousands of years (Bemmels et al. 2022). Ecological consequences of ancient eruptions have largely been studied using pollen records to reconstruct the recovery of plant species (Long et al. 2014; Allen and Huntley 2018; Kristensen, Beaudoin, and Ives 2020), but studies on the impact of these ancient volcanic events on animal populations remain limited.

Genetic studies reveal the long-term and variable impacts of volcanic events on animals. Kuhn et al. (2010) found evidence for a partial population replacement in Yukon caribou in the aftermath of the Holocene White River tephra deposition, suggesting that previously distinct populations began to mix after the ashfall. Conversely, brown kiwi populations became more genetically diverse due to range contraction after the Holocene Taupo supervolcano eruption in New Zealand (Bemmels et al. 2022). While past volcanic events caused genetic shifts in animal populations, there are few studies exploring this phenomenon and none that examine extinct species.

An eruption at Emmons Lake in the Aleutian archipelago ~29 ka deposited ash across much of southeastern Alaska and Yukon (Figure 1), which settled into a 5–80 cm thick bed of ash known as Dawson tephra (Naeser et al. 1982; Froese et al. 2002). Analysis of a vegetation mat buried under Dawson tephra revealed that the tephra was deposited when diurnal icings were active, either in late winter or early spring (Froese, Zazula, and Reyes 2006), which may have mitigated some ecological effects of the ashfall, since snow may have transported much of the ash off hillsides and into valley bottoms and waterways quickly after deposition. Froese et al. hypothesized that the tephra accumulated in river valleys and other low-lying areas, resulting in over-thickening of the ash in these areas. Exposed grazing pastures on hillsides may have maintained plant and animal biodiversity even while other nearby areas were buried (Froese, Zazula, and Reyes 2006).

Figure 1.

Map of northwestern North America showing the location of Emmons Lake volcanic center (black triangle), the probable minimal distribution of Dawson tephra (blue shading) adapted from (Froese, Zazula, and Reyes 2006), and the location of the Klondike (Orange box) within Yukon Territory (yellow shading). Inset shows the locations of sites used in this study (red diamonds).

Here, we use sedimentary ancient DNA (sedaDNA) to investigate the long-term ecological and genetic impacts of Dawson tephra deposition on a mammoth steppe ecosystem of Late Pleistocene eastern Beringia that is 1700 km away from the source volcano. We use a target capture approach to compare plant and animal community composition and biodiversity before and after the ashfall in three sites from the Klondike gold fields region of Yukon Territory, Canada. We track changes in population demographics using an approach to quantify diversity in captured mammalian mitochondrial genomes to determine whether the ashfall resulted in observable megafauna migration or die-off. This study showcases sedaDNA as a tool for investigating both paleoecological and demographic questions and provides an additional perspective on the impact of ancient eruptions on steppe ecosystems.

Methods

Sediment core collection and subsampling

We collected nineteen samples from three sites at permafrost exposures containing Dawson tephra in the Klondike region of Yukon Territory, Canada (Figure 1 and 2). These sites—Hunker (64.008, −139.097), Little Blanche (63.842, −139.118), and Eureka (63.560, −138.891) creeks—are all sites of active placer gold mining located in the Klondike goldfields of Yukon Territory and on the traditional territories of the Tr’ondëk Hwëch’in First Nation. At Little Blanche and Eureka creeks, the tephra was less than 5 cm thick, whereas it was slightly more than 10 cm thick at Hunker Creek.

Figure 2.

Age models and permafrost sections at Eureka and Hunker Creeks. Age models for Eureka Creek (A) and Hunker Creek (D) were generated in OxCal (Bronk Ramsey 2009). Stratigraphy of Eureka Creek (B) and Hunker Creek (E) are shown as described when the cores were taken in 2022. Permafrost cores were taken below, through, and above Dawson tephra at Eureka Creek (C) and Hunker Creek (F).

In the summer of 2022, we took multiple lateral permafrost cores below, above, and through Dawson tephra at each site using a 3-in coring barrel and a gas-powered drill (Figure 2 and S1). After collection, cores were kept frozen and transported to the Permafrost Archives Science Laboratory at the University of Alberta, where they were stored at −20°C. We sampled a 10–15 cm segment of each core in a dedicated cold room, and then transported the frozen core segments to a dedicated sediment sampling room at the Paleogenomics Laboratory at the University of California, Santa Cruz. We used two sterilized chisels to chip away outer layers of sediment and used a third chisel to collect sediment from the center of the core to minimize risk of contamination. Between sampling each core, we cleaned tools with a 3 percent bleach solution and rinsed them with 70 percent ethanol. After collection, sediment subsamples were stored in sterile 50 mL falcon tubes at −20°C until processing.

Chronology

Tephrochronology

Tephra samples from Little Blanche Creek (SC22-LB-6), Hunker Creek (SC22-HC-13), and Eureka Creek (EC22-2-27) were prepared and geochemically analyzed on a JEOL 8900 R electron microprobe by wavelength dispersive spectrometry at the University of Alberta using standard methodologies (e.g., Jensen et al. 2019). We confirmed the correlation to Dawson tephra via concurrent analysis of a known Dawson tephra reference sample. See Table S2 for further methodological details and analytical results.

Radiocarbon dating

We prepared two samples from Hunker Creek and three samples from Eureka Creek for radiocarbon dating by completing pre-treatment at the University of Alberta using the acid-base-acid method (e.g., Reyes, Froese, and Jensen 2010). Suitable plant macrofossils for radiocarbon dating were not available from the Little Blanche site. The samples were subsequently frozen, freeze-dried, and stored in airtight sterilized vials. We then shipped the sample to the Keck-Carbon Cycle Accelerator Mass Spectrometry facility at UC Irvine (UCIAMS) where CO₂ production, graphitisation, and measurement of radiocarbon abundance was completed. We calibrated the radiocarbon dates using OxCal version 4.4 (Bronk Ramsey 2009) and the IntCal20 calibration curve (Reimer et al. 2020).

Age models

Age models were developed for both the Hunker and Eureka Creek sites using Oxcal version 4.4 (Bronk Ramsey 2009) and the IntCal20 radiocarbon calibration curve (Reimer et al. 2020). At Eureka and Hunker creeks, the sedimentary records are vertically continuous with no clear breaks in the stratigraphy. As a result, we developed separate P_Sequence depositional models for each site which used the seven radiocarbon dates available from Eureka Creek and the two radiocarbon dates from Hunker Creek. For both sites, the P_Sequence model was run with a variable k parameter (Bronk Ramsey and Lee 2013) with a General_Outlier model which accounts for a 5 percent probability that individual radiocarbon ages are statistical outliers.

Sedimentary ancient DNA

Extraction

We performed all extractions in a dedicated ancient DNA facility at UC Santa Cruz. We extracted each of the 19 permafrost samples in triplicate using a cold spin extraction method (Murchie et al. 2021), for a total of 57 extractions. We processed triplicates for each sample to maximize recovery of rare taxa (Shirazi, Meyer, and Shapiro 2021) and to obtain enough reads for phylogenetic analyses. We included three extraction controls with no sediment. Input sediment for each extraction was 0.25 g, which we added to 0.7 mm garnet PowerBead tubes. We added 750 uL PowerBead Solution and 500 uL digestion solution to each tube. Then, we vortexed the samples at high speed for 15 minutes, briefly centrifuged them, added 15.63 uL of 20 mg/mL proteinase K solution to each tube, and placed them in a 35°C oven, where they incubated while rotating for 19 hours.

After 19 hours, we centrifuged the samples for 5 minutes at 10,000 g. We carefully removed the supernatant for each sample to avoid disturbing the pellet and added it to a 50-mL centrifuge tube containing 16.25 mL of a guanidine binding buffer. The concentration of sodium acetate in the binding buffer was 0.12 M as in Rohland et al. (2018), which is slightly higher than the buffer in Murchie et al. (2021). We centrifuged the samples and binding buffer at 4°C at 2500 g for 20 hours to pellet inhibitors.

After centrifugation, we removed the supernatant and passed it through a high-volume silica column, washed twice with 750 uL PE buffer, and eluted in 50 ul of EBT buffer (Rohland et al. 2018). As some of the final extracts were dark in color, we performed a qPCR dilution assay to test for inhibition before continuing with library preparation. There was no detectable inhibition even in the most concentrated samples, so we continued to library preparation without further dilution or cleaning of the samples.

Library preparation, targeted capture, and sequencing

We prepared sequencing libraries using a single-stranded library protocol optimized for ancient and degraded DNA with 2ng of total DNA input for each sample (Kapp, Green, and Shapiro 2021). We then performed a hybridization capture on the libraries using the PalaeoChip Arctic v1.0 bait set, which includes sequences from extinct and extant Arctic animal mitochondrial genomes and plant metabarcoding loci (Murchie et al. 2021). We followed the myBaits v4.1 capture protocol (Daicel Arbor Biosciences) with a single 24-hour capture incubation at 60°C. After capturing, we quantified each library with a Qubit assay and determined the average fragment length of each with a fragment analyzer and then pooled all libraries equimolarly. We sequenced each library to an average depth of 2.3 M reads on a NextSeq 550 on a 2 × 75 bp paired-end kit (Illumina) at the UC Santa Cruz Paleogenomics sequencing facility.

Bioinformatics

Preprocessing

We used bcl2fastq (RRID:SCR_015058) to demultiplex raw sequencing reads, and then removed adapters, merged paired-end reads, and filtered for read quality with fastp (Chen et al. 2018). We used PRINSEQ-lite (Schmieder and Edwards 2011) to remove duplicate sequences and FastQC (Andrews 2010) to confirm the quality of the resulting FASTQ files.

Taxonomic profiling

To obtain read assignments for phylogenetic placement, we mapped filtered FASTQ files against all of NCBI-nt and RefSeq (downloaded March 2023) (Sayers et al. 2022) using Bowtie2 (Langmead and Salzberg 2012) (parameters: bowtie2 -q -k 100 – no-unal) and ran ngsLCA (Wang et al. 2022), a least-common-ancestor program that uses the NCBI taxonomy for taxonomic assignments, with default parameters. We used bamdam (De Sanctis et al. 2025) to visualize and authenticate the ngsLCA-assigned reads.

We merged triplicates for each sample before mapping so that we obtained one output for each sample that could be easily compared with other samples from the same site. One out of three extraction controls yielded no mapped reads after capture. The reads present in the other controls mapped to human (Homo sapiens) and carp (Cyprinius carpio), both of which are common contaminants in sedaDNA datasets (Armbrecht et al. 2021). A list of contaminant taxa from these controls with associated read counts is available in Table S8.

We excluded taxa from our analyses based on taxonomic rank, the number of mapped reads, DUST complexity scores, absence in the Paleochip 1.0 bait set, and ecological plausibility. First, we filtered our initial dataset to exclude reads that mapped above family level and below genus level to mitigate uninformative and erroneous results. Next, we excluded animal taxa with less than 3 uniquely aligned reads and plant taxa with less than 8 uniquely aligned reads. Then, we excluded results with DUST complexity scores greater than 5 to remove low-complexity reads that are frequently mismapped (Morgulis et al. 2006). Finally, we removed taxa that were not present in the Paleochip 1.0 bait set or were ecologically implausible.

Since many Arctic plant taxa are absent from major nucleotide databases, pairwise alignment can erroneously assign plants to taxa that are neither present in the Palaeochip 1.0 bait set nor ecologically rational, like cultivated food and exclusively tropical taxa (Pach et al. 2024). We removed these taxa (a list is given in Table S7) from before continuing with biodiversity and paleoecological analyses. Most of these taxa were assigned to fewer than 50 reads or were present in fewer than 5 samples. Those reads that were present at higher abundances primarily mapped to Eurasian and South American grass genera and likely represent reads from grass species that are not present in genomic databases. We removed these reads from biodiversity analyses (Figure 3C), but retained them at the family level for our community composition report (Figure 3A). We used BLAST-n (Camacho et al. 2009) to further inspect suspected misaligned taxa that received more than 20 reads in a sample and determined that all contained reads from conserved regions of related plant taxa that were present in the Palaeochip 1.0 bait set. Despite its absence from the bait set, we chose to retain reads that mapped to Taraxacum (dandelion) since it is a taxon known from the region (Zazula et al. 2006).

Figure 3.

Community composition of plants (A) and animals (B) at all three sites. Read counts are represented as percentage of plant or animal community, after off-target taxa were removed. Plant families with fewer than 2 percent of assigned reads are aggregated together into “other” categories for each plant type. Some animal taxa, including carnivores (Felidae, Canidae, and Ursidae), hare (Leporidae), and ground squirrel (Sciuridae) are present in low abundances (<2 percent). All read counts, including those for low abundance taxa and excluded taxa, can be found in Tables S3–7. Gray dashed lines represent the location of Dawson tephra at each site. Plant (C) and animal (D) biodiversity are shown as NMDS plots of Bray-Curtis dissimilarity. Circles represent samples taken from below Dawson tephra and triangles represent those taken from above. The dotted line indicates the spread of samples from below Dawson tephra and the solid line indicates the spread of samples from above. Vectors (gray lines) were produced using the envfit function in the vegan package (Oksanen et al. 2024). The ten strongest vectors are represented in the plant biodiversity plot, whereas the five strongest family-level vectors are represented in the animal plot.

We also excluded reads that mapped to ecologically improbable species in our animal profiles. As with the plants, we first removed taxa that did not appear in the Paleochip 1.0 bait set. Some reads in our samples mapped to genera that were present in the bait set, are not ecologically feasible, and are closely related to species that were present in high abundances in our samples. These taxa include Bos (cattle) and Odocoileus (white-tailed deer), which are genera that are not known to be present in Late Pleistocene east Beringia. Given the relatedness between Bos and Bison, we considered that these Bos reads could be mismapped Bison reads. We used BLAST-n to check these assignments and they were identified as Bison priscus, an expected Bovidae species for this time and place. Since reads assigned to Bos were a small proportion of total Bovidae reads, we excluded all of them at the genus level. Using the same procedure, we determined that the Odocoileus reads were mismapped Rangifer reads and excluded them at the genus level. Finally, some of our samples yielded Ursus (bear) reads. Given the presence of the related Arctodus (short-faced bear), we examined all Ursidae reads with BLAST-n and found that all aligned either to Arctodus or to highly conserved genomic regions. Therefore, we excluded reads that mapped to Ursus at the genus level. All excluded reads can be found in Table S7.

To investigate the relations between the ashfall and site biodiversity, we used the vegan community ecology R package (Oksanen et al. 2024) to estimate both plant and animal beta diversity with Bray-Curtis dissimilarity and plot it on a non-metric dimensional scalar plot (Figure 3C,D). To mitigate bias due to differences in sequencing depth, we calculated the biodiversity metrics based on the proportion of total plant or animal reads assigned to a given taxon. For taxonomic ranks above genus, we reported unaggregated reads that did not include assignments to lower taxonomic levels. Since we removed species-level assignments, we reported aggregated reads at the genus level. Absolute read counts used to calculate read proportions for the non-metric multidimensional scaling plots are reported in Tables S3 and S4. Aggregated family-level reads are reported in Tables S5 and S6.

Phylogenetic placement

We next performed phylogenetic placement to determine how our most common ancient mammals, Equus, Mammuthus, Bison, and Rangifer, fell within the mitochondrial diversity of each taxon. To contextualize our Equidae, Elephantidae, and Bison reads, we obtained existing mitochondrial multiple sequence alignments and corresponding phylogenies (Murchie et al. 2022). Our Equidae multiple sequence alignment contained 237 reference genomes, Elephantidae contained 133 reference genomes, and Bison contained 136 reference genomes. From these, we used the R phytools and ape libraries (Paradis, Schliep, and Schwartz 2019; Revell 2024) to remove the lower-coverage samples from Murchie et al. (2022), leaving only reference sequences in both the alignments and phylogenies. We also rerooted, ladderized, and balanced the root branches for all three taxa, and removed highly diverged outgroup sequences in the case of Equus. For Rangifer, we obtained a published reference mitochondrial sequence alignment of 83 mitochondrial caribou sequences (Hold et al. 2024). Since this alignment did not include an outgroup, we obtained an additional outgroup sequence for Odocoileus virginianus (NC_015247.1), used rotate (Durbin, De Sanctis, and Blumer 2023) to rotate it to the same starting position as the existing alignment, and added it to the existing multiple sequence alignment using muscle (Edgar 2004). We then created a mitochondrial phylogeny using the BEAST suite with default settings (version 1.10.4) (Suchard et al. 2018) and 1 million burn-in states. Lastly, we used Figtree (http://tree.bio.ed.ac.uk/software/figtree/) to convert to a newick tree file, and checked that the phylogeny matched well to that shown in the original article (Hold et al. 2024). We used Seqotron (Fourment and Holmes 2016) to visually confirm that all four multiple sequence alignments looked as expected. To prepare for phylogenetic placement, we ran snp-sites followed by a custom R script fix_vcf (https://github.com/miwipe/KapCopenhagen/blob/main/scripts/fix_vcf.R) to obtain pseudo-haploid vcf files for each reference panel and called a consensus sequence across the full multiple sequence alignment in each case using the Bio.Align Python package (Cock et al. 2009).

We then extracted all reads from our ngsLCA output which mapped to Cervidae, Elephantidae, Equidae, and Bison using samtools (Li et al. 2009). To minimize reference bias, we re-mapped these taxa-specific reads with bowtie2 default parameters against the consensus sequences in each case. Because some reads erroneously mapped to Bos, we chose to restrict our analyses to reads that matched the Bison genus rather than the Bovidae family. We filtered bam files using samtools (Li et al. 2009) and Picard’s MarkDuplicates (https://broadinstitute.github.io/picard/), and ran pathPhynder (Martiniano et al. 2022) to obtain phylogenetic placements. To verify that the target sequences were ancient, we analyzed the reads that mapped to each megafauna species using bamdam (De Sanctis et al. 2025) to visualize the frequency of cytosine deamination at the ends of DNA strands and average fragment length, which are characteristic features of post-mortem damage (Dabney, Meyer, and Pääbo 2013) (Figure S2).

Results

Chronology

The age models (Table 1 and Figure 2A) indicate a sedimentation rate at Eureka Creek at ~0.04 cm/year and the median time between samples ranges from 100 to 600 years. However, the margin of error for these median dates is hundreds of years, so the actual time between samples at Eureka Creek may be larger or smaller. At Hunker Creek (Figure 2D), the time between samples is less clear due to the rapid sedimentation rate and large uncertainties on the associated dates. At this site, the date of our oldest sample was 25,400 ¹⁴C years BP and our youngest date was 25,380 ¹⁴C years BP, both with an uncertainty of 180 years. Given how close these dates are to one another, the sedimentation rate of 0.4 cm/yr estimated in OxCal is likely higher than reality. The time between samples is also difficult to estimate as the modeled ages of the samples are very close to one another. Dates were estimated in OxCal based on our age models (Figure 2A,D) and are available in Table 1.

Table 1.

Radiocarbon and modeled dates of samples taken from Eureka Creek and Hunker Creek.

							Modeled age (cal yr BP)
Site	Sample ID	Depth relative to Dawson tephra (cm)	Sampled material	14C age (BP)	+/−	Sedimentation rate (cm/yr)	median	from_95_4	to_95_4
Eureka Creek	EC22-2-22	72.5	Misc. plant	23350	150	0.0406	27550	27780	27310
	EC22-2-23	51				0.0407	28075	28408	27727
	EC22-2-24	39				0.0406	28370	28711	28037
	EC22-2-25	19	Misc. plant	24610	170	0.0413	28855	29110	28645
	EC22-2-26	6				0.0421	29165	29462	28899
	EC22-2-27	−1				0.0404	29325	29641	29035
	EC22-2-28	−15				0.0395	29680	30146	29260
	EC22-2-29	−39				0.0396	30275	30965	29708
	EC22-2-30	−62				0.0396	30845	31678	30193
	EC22-2-31	−106.5	Misc. plant	28020	560	0.0395	31950	32950	31160
Hunker Creek	SC22-HC-22	0.41	Fecal pellet	25400	180	0.41	29479	29749	29245
	SC22-HC-21	0.275				0.411	29478	29749	29245
	SC22-HC-20	0.14				0.412	29478	29748	29244
	SC22-HC-19	0				0.413	29477	29748	29244
	SC22-HC-18	−0.22				0.415	29477	29748	29244
	SC22-HC-17	−0.38				0.416	29476	29747	29244
	SC22-HC-16	−0.52	Misc. plant	25380	180	0.417	29476	29747	29244

Sedimentary ancient DNA

We generated 129 M total sequencing reads across 57 permafrost capture libraries, which encompassed laboratory triplicates for each of the 19 permafrost samples. Of the 129 M total reads, 1.1 M were taxonomically assigned to 12 on-target animal families and 40 on-target plant families. One sample from Hunker Creek (SC22-HC-19) was composed entirely of Dawson tephra and yielded 3.88 M total sequencing reads, 15.7k of which were assigned to 4 animal families and 12 plant families. In comparison, another permafrost sample from the same site (SC22-HC-16) yielded 6.75 M reads across all replicates, 85.4k of which were assigned to 9 animal families and 36 plant families. It is interesting that DNA was preserved in the tephra itself, but unclear whether the limited number of taxa recovered from this sample is due to poor DNA preservation conditions in tephra or a true reflection of the ecosystem composition of the ash-covered landscape, so we excluded the sample from subsequent analyses. Tephra deposition is a short-term event, so the tephra sedaDNA sample likely covers a shorter amount of time relative to the permafrost samples that have a higher deposition rate, and thus encompass a greater period of time. This discrepancy in deposition rate between tephra and permafrost may also help to explain the low read counts in the tephra sample. Reads recovered from the SC22-HC-19 can be found in Tables S3 and S4.

Grasses (Poaceae) dominate the plant communities at each site, followed by forbs (Asteraceae, Rosaceae, and others) and shrubs (primarily Salicaceae) (Figure 3A). Within Asteraceae, sage (Artemisia) is particularly abundant. Other taxa include mosses (Bryaceae and others) and sedges (Cyperaceae). Megafauna such as bison (Bison), mammoth (Mammuthus), and horse (Equus) consistently dominate the animal community at each site (Figure 3B), with smaller proportions of caribou (Rangifer), sheep (Ovis), lemming (Lemmus and Dicrostonyx), ptarmigan (Lagopus), ground squirrel (Urocitellus), and fox (Vulpes). At very low rates (fewer than 10 reads), we detected short-faced bear (Arctodus), lion (Panthera), and muskox (Ovibos). The relative abundances of the animal species varied across space and time. Mammoth, horse, bison, and caribou sequences showed patterns of terminal cytosine to thymine transitions typical of ancient DNA damage (Figure S2).

There is some heterogeneity in the plant and animal communities between sites, although they remain homogenous through time (Figure 3C,D). When only considering plant taxa, samples from all sites fall close to each other on an NMDS plot, although samples tend to cluster closely with those from the same site (Figure 3C). Samples taken from below and above the tephra overlap in NMDS space, indicating that there is not a significant difference in plant community composition between the two groups. Animal communities are more heterogeneous among sites (Figure 3D). Similar to the plant communities, the animal communities before and after the ashfall show significant overlap, indicating that the animal communities did not change in response (Figure 3D).

Phylogenetics

Horse mitochondrial diversity appears to have not been affected by ecosystem-level changes associated with the ashfall, with low overall diversity (Figure 4B; full version available in Figure S3). We used a mitochondrial reference phylogenetic tree built from 237 equid mitochondrial genomes generated by Murchie et al. (2022) to investigate the demographic history of the horse reads isolated from our samples. We recovered enough variants from fourteen of our eighteen samples to place in the phylogeny. All horse samples cluster with east Beringian E. caballus and E. caballus lambei (Yukon horse) genomes and exhibit little diversity within this eastern Beringian clade with no evidence for population replacement following the ashfall.

Figure 4.

Map of eastern Eurasia and western North America shaded to show the locations of sample origins (A). Placements of sediment-derived mitochondrial sequences from Eureka Creek (EC), Hunker Creek (HC), and Little Blanche Creek (LB) using pathPhynder (Martiniano et al. 2022) within reference phylogenies of horses (B), bison (C), and mammoths (D) from Murchie et al. (2022). Full versions of the bison and horse trees include additional species and can be found in Figures S3 and S4. Placements of caribou sequences were also examined and a phylogenetic tree is available in Figure S5.

Bison mitochondrial diversity also appears to have not been affected by ecosystem-level changes associated with the ashfall (Figure 4C; full version available in Figure S4). Seventeen out of eighteen samples contained enough bison variants to be placed within a phylogeny of 136 bovid mitochondrial genomes generated by (Murchie et al. 2022). Of these, four only contained enough information to be identifiable as Bison sp. while the other thirteen clustered with east Beringian B. priscus genomes, with no apparent structure due to either the ashfall or site location. The bison reads were placed within bison mitochondrial clades known to be present in the region (Heintzman et al. 2016; Froese et al. 2017; Murchie et al. 2022).

Similar to bison, mammoth mitochondrial diversity was unchanged after the ashfall (Figure 4D). We recovered enough variants from all eighteen of our samples to place in a tree of 133 mammoth mitochondrial genomes generated by Murchie et al. (2022), although two (SC22-HC-17 and EC22-2-23) were only resolved as basal to clades 1DE and 1C, probably indicating that there was not enough information to place them deeper within either clade. Most of the mammoth samples recovered here belong to clades 1C or 1DE, reflecting typical genetic diversity of east Beringian mammoths from this period (Palkopoulou et al. 2013). Mammoths from both clades are present on this landscape both before and after the ashfall and evidence for genetic turnover is lacking. Again, we do not observe population structure across sites or between samples taken before versus after the ashfall, indicating that this region supported a genetically diverse population of mammoths across space and time.

Finally, we investigated the demographic history of the caribou reads isolated from our samples using a mitochondrial reference phylogenetic tree built from 83 Rangifer sp. mitochondrial genomes (Hold et al. 2024) (Figure S5). We recovered enough variants from sixteen out of our eighteen samples to place in the tree. All of these sixteen samples belong to the Euro-Beringian lineage, although eight were only resolved as basal to the other genomes in the clade. Like the other megafauna, there is no structure across sites or between pre- and post-ashfall samples.

Discussion

Our results indicate that the Dawson tephra ashfall did not have obvious long-term effects on mammoth steppe plant and animal communities. We did not detect changes in either plant or animal community composition at any site (Figure 3), and the horse, bison, and mammoth mitochondrial diversity also remained unchanged (Figure 4). Our taxonomic profiles for plants and animals (Figure 3A,B) reflect a steppe-tundra ecosystem that remains constant across time, although there is some site-specific heterogeneity, with samples from the same site clustering closely together in NMDS space when considering only the plant communities (Figure 3C). Horse, bison, and mammoth mitochondrial diversity also remain constant over time (Figure 4). The megafauna present at these sites reflect expected Beringian lineages, with horses containing lower mitochondrial diversity than bison and mammoths (Figure 4B). The bison populations present at these sites contained variants associated with two separate, prominent clades within the Beringian and North American bison phylogeny, indicating that bison maintained a genetically diverse population both before and after the ashfall. Mammoths from both Siberian (1DE) and North American (1C) clades were also present, the result of genetic exchange across the Bering Land Bridge at this time. This observed lack of ecosystem and genetic change may be due to the inherent resilience of the mammoth steppe, the seasonality of the ashfall and landscape topography, or some combination of these.

One possible explanation for our results is that the species that lived in Late Pleistocene east Beringia were particularly well adapted to life in a harsh environment (Guthrie 2001), and thus perhaps resilient to ecosystem disturbance. Allen and Huntley (2018) studied the effects of tephra on ancient Italian landscapes and found that cold steppe environments were more resilient than wooded steppe habitats. They suggested that the cold steppe vegetation was better adapted to frequent disturbance and thus was able to recover quickly. The vegetation of the cold Beringian mammoth steppe was certainly adapted to endure disturbance. Most of the biomass of many plant species, such as grasses and forbs, was stored underground as deep root systems, allowing for quick regrowth following snow melts (Guthrie 2001). These adaptations may have made the plants resilient to even major ecological disturbance, particularly volcanic ashfall which covers the landscape similarly to snow. Large grazing animals, as well as the predators that hunted them, needed vast ranges to consume enough to support their populations (Guthrie 2001). The huge ranges and population connectivity of mammoth steppe megafauna provided them a safeguard against population collapse in the wake of major ecological disturbance. During Marine Isotope Stage 3, Beringian megafauna populations were decreasing, likely due to shrub expansion during relatively warmer interstadial conditions (Debruyne et al. 2008; Lorenzen et al. 2011; Mann et al. 2015). Our findings suggest that despite decreasing population sizes, horses, mammoths, and bison maintained genetic diversity without experiencing population turnover or bottlenecks, highlighting that factors besides population size were in play to confer resilience to these populations. Population and habitat connectivity bolstered these animals’ ability to resist significant declines in genetic diversity and population size. Increased connectivity is associated with higher ecosystem resilience after disturbance (Pearson et al. 2021) while community isolation due to habitat fragmentation is associated with decreased biodiversity (Steiner and Asgari 2022) and loss of genetic diversity in previously diverse populations (Robinson et al. 2019). We found that bison and mammoth populations were genetically diverse, indicating that there was gene flow and connectivity among Beringian populations (Figure 4C,D).

Another possibility is that the seasonality of the ashfall mitigated negative ecological effects. Previous research indicates that Dawson tephra was likely deposited in late winter or early spring, when snow was present (Froese, Zazula, and Reyes 2006). These authors hypothesized that the ecological impact of the ashfall would be mitigated due to mobilization and reworking of the ash with snow melt. As part of the northern Cordillera, the landscape of the Klondike Plateau consists of rolling hills and valleys. During spring thaw, the ash would have moved off hillsides and into valley bottoms and waterways with the melting snow, thus exposing new grazing pastures. The resilient characteristics of mammoth steppe plants and animals would have allowed them to quickly take advantage of these new ash-free areas as fast-growing plants became available for wide-roaming herds of grazers. Thus, it is the seasonality of the ashfall, landscape characteristics, and the resilience of the mammoth steppe ecosystem that buffered it from destruction following the ashfall.

The relatively shallow depth of Dawson tephra at these sites is another factor that may have contributed to the observed phylogenetic continuity of the megafauna. At all three sites, the tephra was less than 15 cm thick, and the primary thickness may have been shallower due to reworking of the tephra with snowmelt (Froese, Zazula, and Reyes 2006). In many studies on ancient volcanic eruptions, plant communities completely recovered to pre-eruption conditions within just a few decades after thin tephra deposition, if they were impacted at all (Egan et al. 2016; Allen and Huntley 2018; Pickarski, Kwiecien, and Litt 2023), which would in turn lessen the impact on herbivores. For example, Letts et al. (2012) found that the caribou of the Northwest Territories exhibited genetic continuity after the deposition of the White River tephra, which triggered a partial genetic turnover in southern Yukon populations. In this region, the White River ash layer was less than 5 cm, which was likely thin enough for the caribou to continue to forage successfully. Similarly, the thinness of Dawson tephra at these sites, particularly when combined with the seasonality and the hilly landscape, may have allowed the megafauna to graze and browse as normal, without any major disruptions.

This study provides insight not only into the ecological implications of Dawson tephra, but also into the utility of sedaDNA as a paleoproxy to study complicated ecological questions. We can compare our sedaDNA taxonomic profile with one previously constructed from pollen and macrofossils. The plant and animal taxa recovered from these sites are typical of a mammoth steppe environment and are consistent with the community recovered from plant matter buried under Dawson tephra at Goldbottom Creek, another site in the Klondike goldfields proximal to those examined here (Froese, Zazula, and Reyes 2006; Zazula et al. 2006). This previous study recovered riparian taxa alongside steppe-tundra taxa, suggesting that the site was a mosaic landscape of mesic lowland and arid upland communities. We similarly find a combination of riparian taxa, such as horsetail, willow, and sedges, and steppe-tundra taxa, such as graminoids and sage. Although they could not be resolved to the species level, the willow reads present in these samples likely belong to a dwarf willow species that is common on the tundra today, rather than a larger species that would require more moisture than was available on the mammoth steppe, even in a riparian area (Zazula et al. 2006). Indeed, willow macrofossils from the buried vegetation were found to belong to a dwarf species (Zazula et al. 2006). In addition, sedaDNA detected the presence of some taxa not recovered in this earlier macrofossil-based study, such as poppy (Papaver) and buttercup (Ranunculus), both of which made up a considerable portion of the total plant reads at Eureka Creek. However, these species are known from other macrofossil studies in the region (Zazula et al. 2007).

With sedaDNA, we can reconstruct animal communities at these sites, which has not been possible with the previous pollen study (Figure 3B). In addition to the megafauna, each site hosted ptarmigan (Phasianidae) and lemming (Cricetidae), both of which are members of the modern tundra. Hare (Leporidae) and ground squirrel (Sciuridae) were found in a portion of the sites and at lower abundances, indicating that they were either not as widespread on the landscape or that, as animals with small biomass, they leave less total DNA behind. Carnivores such as short-faced bear (Arctodus), wolves (Canis), foxes (Vulpes), and cave lion (Panthera) were also present at low abundances. These species inhabited higher trophic levels and thus their population sizes were relatively lower compared to species in lower trophic levels. As a result, they are relatively rare in the fossil record. We are able to uncover these cryptic taxa of the mammoth steppe using sedaDNA. See Tables S4 and S6 for lists of the read counts for each species.

Unexpectedly, several samples contained relatively high proportions of caribou, which are mixed feeders that are generally associated with mesic environments (Mann et al. 2013) and are less commonly associated with the arid mammoth steppe. Caribou diet is variable, but they are known to feed heavily on woody plants, which may have been in short supply on the graminoid-dominated mammoth steppe (Guthrie 2001). However, we recovered high proportions of willow at these sites, particularly at Hunker and Little Blanche where we also found the most caribou reads. Riparian zones within the broader steppe-tundra biome may have supported caribou populations by providing a habitat for their preferred food sources. This result again highlights the use of sedaDNA for recovering taxa that are cryptic in the fossil record and uncovering previously unknown species interactions.

This study provides valuable insight into the paleoecology of Beringia and the effects of ecosystem disturbance, but with some limitations. First, the relationship between the number of species reads recovered from sedaDNA and actual biomass is currently unknown. Studies in modern systems suggest that for some species, read abundance and biomass are closely related (Rourke et al. 2022), but how age and preservation conditions may affect this relationship in ancient systems is unknown. Therefore, while the relative abundance of species within samples is useful to understand community structure, conclusions based simply on taxa presence are more robust. Second, since the sequences used to infer phylogenetic placement for bison, horses, caribou and mammoths were recovered from environmental samples, they are made up of genetic material of many individuals. Using this method with aggregate genomes may mask the genetic diversity found within samples in favor of examining the diversity between samples. Another caveat is that the deposition rates for Hunker and Eureka Creeks were estimated based on dates from the tops and bottoms of the section since each sample was not dated. As a result, the sedimentation rates may include some degree of error, particularly at Hunker Creek where the radiocarbon ages of the samples are within the same age range and confidence interval. Here, the sedimentation rate reported is likely higher than reality due to these overlapping age ranges. In addition, dates for Little Blanche Creek were not available so no age model was constructed. However, the site was geographically close to the others and likely faced a similar depositional environment of wind-blown loess and loess resedimentation. Finally, plant communities are known to recover from tephra deposition within decades (Allen and Huntley 2018). Our coarse sampling resolution may limit our ability to detect any changes in vegetation or animal community composition, even if they did occur. However, we would still be able to detect shifts in phylogenetic diversity if the megafauna experienced negative impacts due to the ashfall, as these signals persist through time. Future studies using records with higher temporal resolution are necessary to investigate post-ashfall succession on the mammoth steppe.

Conclusion

Despite its ubiquity and thickness 1700 km from its volcanic source, the deposition of Dawson tephra had no detectable long-term genetic consequences on the megafauna species that inhabited the mammoth steppe ecosystem of east Beringia. The seasonality of the ashfall on a hilly landscape may have decreased any immediate ecological impacts and the life histories of mammoth steppe species may have bolstered overall ecosystem resilience. Here, we demonstrate the utility of sedaDNA as a paleoproxy for environmental reconstructions and as a means for exploring population demographics. Our results give insight into the resiliency of the mammoth steppe community and underscore the severity of the climatic factors at the Pleistocene-Holocene transition that led to the total collapse of this resilient ecosystem. In modern contexts, our findings add to a robust body of research indicating that population and habitat connectivity are vital for perseverance of large grazers in the wake of large-scale environmental disturbance.

Supplementary Material

Supplemental data for this article can be accessed online at https://doi.org/10.1080/15230430.2025.2606444