Evolution of woodcutting behaviour in Early Pliocene beaver driven by consumption of woody plants

Modern beavers (Castor) are prolific ecosystem engineers and dramatically alter the landscape through tree harvesting and dam building. Little is known, however, about the evolutionary drivers of their woodcutting behaviour. Here we investigate if early woodcutting behaviour in Castoridae was driven by nutritional needs. We measured stable carbon and nitrogen isotopes (δ13C and δ15N) of coeval subfossil plants and beaver collagen (Dipoides sp.) from the Early Pliocene, High Arctic Beaver Pond fossil locality (Ellesmere Island), in order to reconstruct Dipoides sp. diet. Isotopic evidence indicates a diet of woody plants and freshwater macrophytes, supporting the hypothesis that this extinct semiaquatic beaver engaged in woodcutting behaviour for feeding purposes. In a phylogenetic context, the isotopic evidence implies that woodcutting and consumption of woody plants can be traced back to a small-bodied, semiaquatic Miocene castorid, suggesting that beavers have been consuming woody plants for over 20 million years. We propose that the behavioural complex (swimming, woodcutting, and consuming woody plants) preceded and facilitated the evolution of dam building. Dam building and food caching behaviours appear to be specializations for cold winter survival and may have evolved in response to late Neogene northern cooling.

www.nature.com/scientificreports/ Late Early Pliocene mean global temperature was 2-3 °C above modern 28 and high latitude regions experienced amplified warming. Pliocene Arctic mean annual temperature was near freezing, which is ~ 15-22 °C warmer than present day, and tree line was ~ 2000 km further north 15,24,[29][30][31][32][33] . Summer temperatures at the Beaver Pond site reached 20 °C, while winter temperatures were more moderate than present day, with a low of ~ -12 °C 24 . Despite the relatively mild conditions, the Beaver Pond site still experienced total darkness and subzero temperatures during the winter months.
There is no modern analog for the ecological community found at the Beaver Pond site, although the flora of present-day Labrador (Canada) is considered to be similar 22,[34][35][36] . The Beaver Pond macrofossil assemblage indicates a larch forest, although birch, alder, spruce, pine, cedar, and cold-adapted woody shrubs were also present (see Matthews and Fyles 17 for a complete list of identified flora). The Beaver Pond site supported higher faunal biodiversity than any modern near-tree line communities.
The remains of a complex faunal community were discovered at the Beaver Pond site, including beaver (Dipoides sp., see below), three-toed horse, bear, badger, "deerlet", water fowl, fish and a rabbit relative 10,22 . Pliocene-age sites with similar fauna and flora community composition are very rare-Idaho in mid-continent North America, and the high altitude Yushe Basin, in northeastern China are the only known sites with similar (but not equivalent) faunal assemblages 37,38 . Dipoides ecology. The most common vertebrate remains at the Beaver Pond site belong to Dipoides, an extinct genus of beaver known from the Neogene of Eurasia and North America, represented by 12 different species 7,[39][40][41] . Although not directly related to the extant Castor, both genera share semiaquatic and woodcutting behaviours 9,11,42 .
The Beaver Pond site is the only known locality with sufficiently well-preserved plant macrofossils to record evidence of Dipoides sp. woodcutting behaviour. The peat deposits are hypothesized to be the remnants of an ancient beaver pond due to the presence of many beaver-cut sticks, and even a cluster of cut sticks, cobbles, and silt that resemble the core of a beaver dam 10 .
Here we use stable isotope data to understand Dipoides sp. diet and elucidate the purpose of their woodcutting behaviour. Our study aims to describe the relative contributions of woody vegetation and aquatic plants to the diet of Dipoides sp. using stable carbon and nitrogen isotope analysis of contemporary sub-fossil skeletal and plant macrofossil material from the Beaver Pond site. This new information is used to better interpret (i) the ecological impact of Dipoides sp. on the Pliocene landscape, (ii) Dipoides sp. potential for winter survival strategies such as underwater food caching, and (iii) the evolutionary context of tree-exploitation within the Castoridae family.
Stable isotopes and palaeodiet. The stable carbon (δ 13 C) and nitrogen (δ 15 N) isotope compositions of an animal's bodily tissues correlate closely with that of its diet, when adjusted for 13 C-and 15 N-enrichment that www.nature.com/scientificreports/ occurs during collagen formation and with each successive trophic level 43,44 . Well-preserved bone collagen is therefore a useful integrator of an animal's diet. In addition, sufficient context is required to accurately describe the nutrient flow between subsequent trophic levels of an ecosystem. In particular, the diet of an organism must be interpreted within the context of an appropriate dietary baseline. This baseline is composed of isotopically defined food or "menu-items" available to the organism. The isotopic composition of primary producers at the base of the food chain control the δ 13 C and δ 15 N of the dietary baseline for herbivores. The δ 13 C and δ 15 N of primary producers depend on physiology (i.e. which photosynthetic pathway the plant employs) and the isotopic composition of bioavailable sources of C and N (i.e. atmospheric CO 2 ). Casey and Post 45 provide a thorough review of how primary producer δ 13 C and δ 15 N vary with physiology and local terrestrial and aquatic environmental conditions. A particular challenge in many forested-wetland environments, however, is that the carbon and nitrogen isotope range of terrestrial and freshwater plants overlap. There are, however, sufficient differences between the δ 13 C and δ 15 N of terrestrial vegetation utilizing the C3-photosynthetic pathway and vascular freshwater plants (macrophytes) for them to serve as useful endmembers of herbivore diet in such environments (see Methodology).
Another challenge is that the isotopic composition of regional and global C and N baselines (and subsequently, that of primary producers that use them) can change over time 46,47 . Hence, reconstructing the diet of herbivores that lived thousands or millions of years ago can be problematic when using isotopic data, as is very rare to find sufficiently preserved coeval plant material and faunal remains from the same geologic locality. Typically, isotopic data for modern plants are all that are available in palaeodiet studies. Fortunately, much of this concern is alleviated at the Beaver Pond site, given the excellent organic preservation of coeval plant and animal tissues.
Atomic C:N ratio and carbon and nitrogen contents (wt%) were used to assess collagen preservation for Dipoides skeletal material (Table 1). All specimen parameters are within the accepted range for well-preserved Plant macrofossil species diversity. By volume, the Beaver Pond peat sample examined in this study consisted of 85% bryophytes, 10% wood and twigs, and 5% macrofossils. Eleven genera were identified, representing a diverse assemblage of terrestrial and freshwater plants ( Table 2). Seven taxa were analyzed for stable carbon and nitrogen isotope compositions (Scorpidium scorpioides, Larix, Betula, Stuckenia filiformis, Scheuchzeria sp., Cornus sericea, Menyanthes trifoliata) ( Table 2). The dominant moss type was Scorpidium scorpioides (hooked scorpion moss). Larix (larch-a deciduous conifer) was the only conifer species identified from this peat sample (although many other tree species have been previously identified from the Beaver Pond site-see Introduction). Plant macrofossils from multiple genera (Myrica, Shepherdia, Potamogeton, and Hippuris) and from three species of Carex (Carex aquatilis, Carex diandra, and Carex maritime) were also recognized, but in insufficient quantities for stable isotope analysis.
Plant macrofossil stable carbon and nitrogen isotopes. Plant macrofossil stable isotope results are presented in Table 2 and Fig. 4. Macrofossil δ 13 C, δ 15 N, C (wt%), N (wt%), C/N (wt%), and atomic C:N ratios are all within the range expected for terrestrial and freshwater plants ( , and environment conditions. For these reasons, atomic C:N ratio and carbon and nitrogen content are not considered to be infallible indicators of organic preservation in subfossil plants 49 . That said, the plant macrofossil C (wt%) values obtained in the present study are within or close to the mean carbon content for modern plants (between ~ 40 and 47%) (Metcalfe and Mead 49 , and references therein). Plant macrofossil N (wt%) values are lower than the mean nitrogen contents of modern plants (between ~ 1 and 3%) but are not outside the reported range for modern plants.
Stable isotope analysis in R (SIAR) mixing model. Next, we evaluate Dipoides sp. δ 13 C col and δ 15 N col within the context of the isotopic dietary baseline composed of coeval terrestrial and freshwater vegetation from the High Arctic Beaver Pond fossil site. The faunal and plant macrofossil isotope data were incorporated into a Bayesian mixing model to determine the relative input of terrestrial versus freshwater plants to Dipoides sp. diet (Fig. 6). This also allowed us to better assess the connection between Dipoides sp. woodcutting behaviour and its consumption of woody plants.
The SIAR model is a statistical tool that uses biotracers (stable isotopes) to estimate the relative input of different sources to a product or mixture. In (palaeo)ecology, mixing models use the stable isotope compositions    www.nature.com/scientificreports/ of different food sources to infer their relative contributions to the composition of overall diet, and assess the probability that the inferred proportions are correct. There are systematic differences in the isotopic composition between a consumer's collagen and its diet, both for carbon and for nitrogen. Hence, a correction factor must be applied to render data for consumers and possible diet items directly comparable. Plant macrofossils were divided into three sources, or functional types: terrestrial woody plants, vascular freshwater macrophytes, and bryophytes (mosses). These groupings were created to assess how terrestrial and freshwater resources contributed separately to Dipoides sp. diet. The functional types were statistically defined and their relative contribution to diet was assessed using scripts from SIAR V4 in R Studio 3.1.2 (Stable Isotope Analysis in R: An Ecologist's Guide). Based on existing literature, respective bone collagen-to-diet offsets of + 4.2‰ and + 3.0‰ were subtracted from Dipoides sp. δ 13 C col and δ 15 N col when incorporated into the mixing model [50][51][52][53] .

Discussion
Dipoides sp. palaeoecology. The Bayesian mixing model indicates that Dipoides sp. consumed both woody plants and freshwater macrophytes in approximately equal proportions (Figs. 6 and 7), although it relied slightly more on freshwater macrophytes. This suggests that Dipoides sp. spent a greater proportion of time feeding in the water than on land.
The distribution of Dipoides sp. δ 13 C col and δ 15 N col is not entirely enclosed within the three primary producer functional groups analyzed (Fig. 6). This is likely the result of the relatively small plant macrofossil sample size. Submerged aquatic macrophytes, for example, are under-represented in the plant macrofossils available for stable isotope analysis. Macrophytes have highly variable δ 13 C and may have contributed more to Dipoides sp. diet than the mixing model suggests. Submerged macrophytes can be highly enriched in 13 C because of physiological differences (primarily the use of 13 C-enriched dissolved bicarbonate) or environmental conditions in the water column (i.e. boundary-layer effect) [54][55][56] . In addition, tree bark is more enriched in 13 C than tree foliage 57 and may have been a key resource for Dipoides sp.
The results of the dietary mixing model support the interpretation that woody plants were an important contributor to Dipoides sp. diet. It is likely that Dipoides sp. also used shrubs and trees as a source of construction material 10,11 , but more evidence is needed to confirm this. Similar to extant Castor, Dipoides sp. may have also demonstrated regional differences in diet, where northern and southern populations utilized different resources according to their availability.
Nitrogen content and C/N as indicators of forage quality. Plant macrofossil nitrogen content (N wt%) and C/N are indicators of forage quality and may be used to interpret the relative nutrition of dietary inputs. Plants with high N (wt%) contain more protein and energy-likewise, low N (wt%) correlates with low plant digestibility, high fiber and high lignin compound content 58 . Beaver Pond plant macrofossil N (wt%) and C/N are highly variable (Table 2, Fig. 5). Although there is considerable variability in C/N ratios depending upon which plant part was analyzed (i.e. seeds versus woody tissue), woody vegetation tends to have higher C/N ratios than macrophytes, and thus tends to be of lower food quality. However, the increased structural tissues in woody plants may have rendered them more effective winter cache foods.
In extremely seasonal environments such as the High Arctic, herbivores must use plant resources in a highly efficient manner. Herbivores must consume the highest quality forage possible during the brief growing season to maximize nutrient and energy gain. High quality forage typically includes young leaves with high nitrogen content, minimal structural (fibrous) tissues, and low defense compound content 59,60 . www.nature.com/scientificreports/ Within the Beaver Pond macrofossil assemblage, pod grass (an emergent macrophyte) and birch have the highest nitrogen content and lowest C/N (Fig. 5). A larger sample set is necessary to confirm this observation; however, current data supports the conclusion that emergent macrophytes and deciduous broadleaf trees were among the more nutritious types of forage available to Dipoides sp. at the Beaver Pond site. It should be noted that forage quality is not the only factor that governs herbivore feeding behaviour. Animals may preferentially target plants with higher biomass to minimize energy expenditure traveling between forage sites or select plants that grow in locations that minimize the risk of predation.
The C/N of high Arctic shrubs decreases over the course of the growing season 58 . As there is no timeconstraint on macrofossil deposition at the Beaver Pond site, variation in C/N may also be due to differences in plant phenological stage at time of incorporation into the peat layer. The incorporation of senescent plants into the peat deposit at the end of each growing season may in part account for the lower than expected macrofossil N (wt%) values reported from this site.
Beaver Pond site flora δ 13 C and δ 15 N. The Beaver Pond macrofossil assemblage contains a diverse range of terrestrial and freshwater plant species. The identified plant species in this study concur with previous interpretations that this was an open-forest landscape interspersed with shallow wetlands. Larch trees and coolclimate woody shrubs dominated the forest community. The wetlands supported both vascular macrophytes and dense assemblages of bryophytes.
The macrofossil δ 13 C are all within the range expected for primary producers utilizing the C 3 photosynthetic pathway and accessing either ambient or dissolved atmospheric CO 2 as their dominant carbon source. The δ 15 N of the macrofossils are also within the expected range for a riparian ecosystem in a cool climate biome.
While Dipoides sp. most likely consumed leafy tree branches and woody tissues (cambium), it is worth noting that plant seeds and cone bracts were analyzed in this study due to ease of macrofossil taxonomic identification. Leaf δ 13 C is typically lower than that of other plant parts 61  Environmental conditions dictate moss δ 13 C rather than species-specific physiological differences. Peat mosses can grow partially or fully submerged in water. Given that mid-Pliocene atmospheric CO 2 concentration levels were similar to modern (~ 400 ppm) 15,65,66 , moss exposed to the atmosphere would preferentially have used the abundant 12 CO 2 , resulting in low δ 13 C. Alternatively, low δ 13 C in peat mosses can also indicate an underwater growing environment rich in 13 C-depleted respired CO 2 from surrounding plants 62 .
Moss macrofossil δ 15 N is relatively high for a photosynthetic organism (mean = + 4.8‰). This is indicative of either the presence of 15 N-enriched sources of bioavailable N (i.e. dissolved nitrates, organic proteins such as urea or amino acids), or increased nutrient availability 45,61 . Unlike vascular plants, mosses do not uptake compounds through their roots. Rather, they obtain nutrients from wet or dry deposition through their leaves 67,68 . Today, beaver ponds are considered to be N sinks, with elevated rates of bacterially mediated denitrification 69 . These bacterial processes result in 15 N-enriched products that are readily dissolved and used by plants (including moss) living in an aqueous environment. Decomposition processes also increase plant δ 15 N over time 47 and remineralized organic debris decomposing in wetlands may be particularly 15 N-enriched.
Beaver Pond bulk peat samples and moss macrofossils show a similar isotopic pattern (low δ 13 C and high δ 15 N), which suggests that hooked scorpion moss contributed substantially to peat biomass accumulation at the Beaver Pond site.
Macrophytes. Beaver Pond macrophyte δ 13 C fall well within the albeit very wide known range for modern freshwater plants (− 50 to − 11‰, see Osmond et al. 70 , Keeley and Sandquist 54 , Mendonça et al. 55 , and Chappuis et al. 56 ). It is reasonable to assume, however, that the very small sample size in this study hides the potential extent of the carbon isotope variability of macrophytes at the site.
Pod grass and bogbean are classified as emergent macrophytes (they grow rooted in water-logged substrates, but their leaves are exposed to the atmosphere), while pondweed grows entirely submerged. Submerged macrophytes become more enriched in 13 C as the dissolved CO 2 pool (the dominant carbon source) becomes increasingly limited 54 .
The Beaver Pond site pondweed δ 13 C is relatively low (− 26.5‰) for a submerged macrophyte. This indicates that it grew in an aquatic environment with adequate dissolved CO 2 . This is in keeping with the interpretation that the Beaver Pond was a fen (near neutral pH, cool water temperature) during the Pliocene. A low δ 13 C may also indicate high influx of terrestrial organic biomass or mosses (with low δ 13 C) into the water that subsequently remineralized and contributed to the dissolved inorganic carbon pool.
Environmental conditions strongly influence aquatic plant δ 15 N. Beaver Pond macrophyte δ 15 N (range = + 0.2 to + 2.7‰) indicate interspecific access and use of a variety of different sources of bioavailable N within the water column and substrate. The most likely N sources are microbial-fixed atmospheric N 2 (which ranges from -2 to + 2‰), the products of nitrification/denitrification processes ( 15 N-enriched NH 4 + or NO x ), and remineralized 15 N-enriched organic material (either terrestrial or aquatic) 71-73 . Larch. Larch (the extinct species Larix groenlandii) is the most common vascular plant species in this macrofossil assemblage. www.nature.com/scientificreports/ There is an offset of ~ 2‰ between the δ 13 C of (i) larch shoots/buds (which bear the needles) and cone bracts, and (ii) larch seeds. Larch shoots and cones (δ 13 C range = -25.4 to − 25.1‰; mean = − 25.3‰) are more depleted of 13 C than larch seeds (δ 13 C range = − 23.3 to − 22.7‰, mean = − 23.1‰). This could be indicative of seasonal physiological or environmental conditions experienced by larch trees at the Beaver Pond site. The cones and shoots of extant larch trees begin growing in the early spring and have lower δ 13 C, whereas their seeds (higher δ 13 C) do not develop and mature until mid to late summer 74 .
A number of physiological and environmental conditions could be responsible for this offset between needle/bearing structures and seeds. Atmospheric vapor pressure deficit (aridity) induces stomatal closure in vascular plants 75 . This restricts not only the rate of water leaving the needle/tree, but also that of atmospheric CO 2 entering it. Stomatal closure reduces CO 2 entry and results in less discrimination against 13 CO 2 . High levels of solar irradiance in the summer increase the rate of CO 2 assimilation. Plants growing at very high latitudes experience 24-h of daylight during the summer. This creates a greater demand for CO 2 to maintain photosynthesis and less discrimination against 13 CO 2 . Both aridity and increased light levels could contribute to why Beaver Pond larch tissues grown late in the summer are more 13 C-enriched than those grown in the early spring/the previous fall.
Alternatively, trees can use water and carbon (in the form of sugars) stored during the previous year to promote new growth during the early spring when leaves are absent and light levels are low. Tissues that develop early in the growing season (i.e. needle-bearing buds and shoots) can therefore reflect the δ 13 C of photosynthetic conditions from the previous growing season 76,77 . In addition, differences in the macromolecular (lipid, protein, sugar) composition of larch buds/needles versus seeds could account for their offset in δ 13 C (i.e. lipids are typically more 13 C-depleted than proteins).
Larch δ 15 N (mean = + 2.7‰) indicate that these conifer trees had access to N sources other than "light" fixed atmospheric N 2 . Given the proximity of wetlands, the root systems of larch trees may have had access to 15 N-enriched dissolved nitrates in the surrounding water-logged soils. Increasing foliar N concentration due to atmospheric N deposition also drives up plant δ 15 N 78,79 .
Aridity may also have influenced terrestrial plants growing at the Beaver Pond site. Higher rainfall is inversely correlated with δ 15 N, where rainier ecosystems tend to produce more 15 N-depleted plants 80 .
Similar to δ 13 C, there is an offset in δ 15 N (and N wt%) between larch needle-bearing structures (mean = + 3.8‰; 0.9%) and larch seeds (mean = + 2.1‰; 0.3%). This could indicate differences in the macromolecular composition of these different tissue types (where high N content typically indicates higher tissue protein content).  (Table 1), and late Pleistocene Castoroides ohioensis (n = 11) ( Table 1) δ 13 C col and δ 15 N col are compared in Fig. 3. A correction for the Suess effect was first necessary render the δ 13 C of all three genera comparable. The carbon isotope composition of atmospheric CO 2 has changed over time with global climatic conditions. More recently, anthropogenic burning of fossil fuels that has rapidly released CO 2 enriched in 12 C into the atmosphere 46,81 . Hence, a correction is needed when comparing δ 13 C of organic samples from different time periods to account for this isotopic variation in the primary carbon source of photosynthetic organisms at the base of the food web.
Suess effect corrections of + 2.02‰ and − 0.1‰ were applied to the δ 13 C col of modern C. canadensis (collected in 2013 and 2014) and Castoroides (late Pleistocene in age), respectively. These corrections were based on the average δ 13 C of atmospheric CO 2 (δ 13 C CO2 ) calculated from Pliocene dual-benthic and planktonic foraminifera proxy records, spanning from ~ 4.1 to 3.8 Ma (average δ 13 C CO2 = − 6.55‰) 76 . These foraminifera proxy records are approximately contemporary with the Beaver Pond site. Average δ 13 C CO2 for 2014 (− 8.57‰) was compiled from the Scripps CO 2 monitoring program. Average δ 13 C CO2 for the late Pleistocene (− 6.45‰) was compiled using ice core data from Schmitt et al. 82 .
Plants growing during these three different time periods (Pliocene, late Pleistocene, and modern/2014) would reflect the δ 13 C of contemporary atmospheric CO 2 . Therefore, changes in δ 13 C CO2 help explain differences in δ 13 C between Dipoides sp. and modern C. canadensis. Additional factors, however, are important in explaining the wide range of δ 13 C and large enrichment in 13 C measured for Castoroides.
In comparison with Castoroides, both Dipoides sp. and C. canadensis have a relatively small range of δ 13 C col and δ 15 N col (Table 1) (Fig. 3). Dipoides sp. mean δ 13 C col and δ 15 N col are higher than those of modern C. canadensis (Fig. 3). This is attributable to either variation in diet between the two species, or changes in global C and N baselines over geologic time.
Previous mixing model studies predict that extant C. canadensis diet is composed of approximately equal proportions of woody terrestrial plants and aquatic macrophytes 13 . However, this can vary by latitude and season. For example, extant C. canadensis in the Canadian subarctic vary their winter diet significantly depending on habitat 83 . It is worth noting that extant C. canadensis does not occur north of 70° latitude and High Arctic Dipoides sp. living at 78° latitude may have employed different dietary strategies.
Dipoides sp. may have relied more heavily than C. canadensis on underwater stores of tree branches to survive the long, dark polar winter. Tree bark is more 13 C-enriched than leafy vegetation 57 and increased consumption could account for the higher δ 13 C col seen in Dipoides sp. Variation in the quantity and type of macrophytes consumed by each beaver species could also account for this difference (i.e. emergent and floating macrophytes are, on average more 15 N-enriched than submerged macrophytes).
Scientific RepoRtS | (2020) 10:13111 | https://doi.org/10.1038/s41598-020-70164-1 www.nature.com/scientificreports/ Changes in the isotopic composition of the C and N baseline between the Pliocene and the present could also account for the isotopic offset between beaver species. Further investigation of possible changes in the δ 15 N baseline of flora in terrestrial high latitude environments during the Pliocene would be a valuable avenue of future research.

Dipoides sp. behaviour and evolutionary implications. Evidence from the Beaver Pond site has
implications for our understanding of Dipoides sp. ecology. These data also contribute to our understanding of the evolution of behavioural transitions within Castoridae. In particular, how Castor's distinctive complex of behavioural traits (tree harvesting, underwater food caching, and construction behaviour) may have evolved. A new hypothesis of behavioural evolution in castorids based on evidence from the fossil record (i.e. fossil burrows, cut wood, and stable isotope measurements) and skeletal-dental morphology is mapped onto a simplified phylogenetic tree in Fig. 8 42,[84][85][86][87] .
Castoridae is a group of herbivorous rodents comprising roughly two dozen genera. Most fossil castorids fall within two major groups: a clade of fossorial specialists (Palaeocastorinae) and a semiaquatic clade 42,84,86,[88][89][90] . The latter includes Castor and Dipoides. Members of the fossorial clade (~ 7 genera) possess striking specializations such large digging claws, extremely reduced tails, and broad, procumbent incisors for digging. In some cases, specimens have been found within fossil burrows (i.e. Palaeocastor, or "The devil's corkscrew" burrows discovered in the plains of North America 88 ). The semiaquatic group comprises two subfamilies, Castorinae (~ 6 genera, including Steneofiber and the extant Castor), and Castoroidinae (~ 7 genera, including Dipoides and the giant beaver, Castoroides). The oldest definitive Castorinae in the fossil record is Steneofiber eseri from the early Miocene (France, MN2, ~ 23 Ma). S. eseri shows evidence of living in family groups and swimming specializations 91 . This, in combination with aDNA evidence 12 , suggests Castorinae and Castoroidinae are derived from a semiaquatic ancestor in the early Miocene.
Digging behaviour was not just characteristic of the fossorial group and appears within the semiaquatic clade as well. Castor, though not morphologically highly specialized for the task, digs bank burrows and creates extensive canal systems 92 . In addition, the extinct semiaquatic beaver Steneofiber eseri was found within a  Rybczynski 9 , which used a matrix of 88 morphological characters and 38 taxa. The origination of dam building is a minimum age (~ 7-8 Ma), corresponding to the time of divergence of Castor canadensis and C. fiber, inferred from molecular evidence 96 and supported by fossil evidence 97 . Legend: CIRCLE-taxa that burrowed (Dipoides and Castoroides may have burrowed, but direct fossil evidence is currently lacking); WP-taxa with significant woody plant contribution to their diet; NWP-taxa that did not generally consume woody plants (the terrestrial burrowing clade is associated with open plains and unforested habitat, and therefore assumed to have not consumed significant amounts of woody plants); Plio-Pliocene; Q-Quaternary. Age range sources: Castor 96 www.nature.com/scientificreports/ burrow 91 . Considering the phylogenetic distribution of burrowing behaviour within the Castorid tree (Fig. 8), it is likely that the common ancestor of the fossorial and semiaquatic clades also burrowed. Thus, the appearance of burrowing behaviour within Castor and Steneofiber are seen as a retention of a primitive trait 9 . If burrowing behaviour in semiaquatic castorids is the primitive condition, it is likely Dipoides burrowed as well, as seen in other semiaquatic rodents today such as Castor, but also Crossomys (earless water rat), Myocastor (nutria), and Ondatra (muskrat) 92 . It is also possible that Dipoides constructed lodges. Extant Castor and Ondatra are known to construct burrows and lodges, depending on the characteristics of the habitat. Bank burrows are associated with stream environments, whereas lodges are better suited to calmer waters 92 . Unlike Castor, extant Ondatra construct their push-up lodges using cattails and other fibrous vegetation rather than wood. The abundance of cut wood at the Beaver Pond site 11 suggests that Dipoides sp. had the option to incorporate wood into their nesting structures, and possibly built lodges.
Given the occurrence of woodcutting and woody plant consumption within both subfamilies of semiaquatic castorids (represented by Castor and Dipoides in Fig. 8), it seems likely these behaviours appeared in the common ancestor of the semiaquatic group. Woody plant consumption may have preadapted castorids to exploit colder environments that arose during and after the late Miocene. Castor canadensis does not hibernate, but builds and sink rafts of branches and foliage to use as a source of fresh food during the winter months 1,93 . Dipoides sp. may have also engaged in this behaviour and used underwater caches of branches as a primary food source to survive the consecutive months of darkness during the high latitude winter when plants become dormant. The use of woody plants in this way may have been key to allowing beavers to disperse between North American and Eurasia, which required crossing the Bering Isthmus 94 , a high latitude landmass. Curiously, given that a diet rich in woody plants appears to be the primitive condition of semiaquatic castorids, the absence of woody plant consumption seen in the Pleistocene giant beaver Castoroides 13 must be interpreted here as an evolutionary loss and potentially a leading factor in their extinction (Fig. 8).
Among living mammals, Castor's dam construction is a unique and highly derived behaviour -an evolutionary puzzle, associated with a set of innate behavioural specializations 95 . For example, dam construction is well known to be triggered by the sound of running water alone 95 . The presence of such "hard-wired" behaviours may be associated with the ancient origins of this behaviour. Molecular and fossil occurrence records indicate that the split between Eurasian and North American Castor arose around 7.5 Ma ago 96,97 , implying that dam building behaviour itself is at least as old.
Definitive fossil evidence for dam building by an extinct beaver is currently lacking. Consequently, dam building behaviour is shown as possibly arising only on the lineage leading to Castor. Hypothetically, dam building may have arisen from beavers collecting branches near their burrow/lodge for feeding purposes and the accumulations of sticks could have dammed streams by happenstance. The effects may have been multifold. A deeper pond is an effective defense mechanism and provides a safe refuge from predators. Raised water levels also create more favourable conditions for underwater food caching of branches in sub-freezing winter conditions because the deeper water would prevent an underwater food cache from being locked in ice. As such, natural selection would have favoured animals that maintained the dam, presumably as an extension of their pre-existing nesting behaviour such as lodge building. In this scenario, the climate cooling that started around 15 Ma ago and continued into the Pleistocene would have provided an interval where behaviours promoting over-wintering survival, such as underwater food caching branches and dam building, would have been increasingly reinforced by natural selection.
It seems unlikely that the common ancestor of all semiaquatic beavers was a dam-builder. Extant Castor is a large powerful rodent weighing 12-25 kg, with some individuals as large as 40 kg 92 . Its body size is one factor that allows the animal to harvest branches and whole trees to build lodges and maintain dams over multiple years. The Beaver Pond site Dipoides sp. was also a large rodent and was roughly two-thirds the size of an average extant Castor. In contrast, the less-derived semiaquatic beavers, such as the Miocene Eucastor tortus (Castoroidinae) and Steneofiber eseri (Castorinae) were small (~ 1 kg, or less), suggesting that the common ancestor of the semiaquatic lineage was also small bodied. Although the common ancestor of the semiaquatic beaver lineage is inferred to have consumed woody plants (this study), and may have used branches in creating food piles and wood for lodge construction, it would have been too small to have had the capacity to build and maintain dams. As such, if Dipoides sp. did exhibit dam building behaviour, it would be the result of parallel evolution within the Castoroidinae and Castorinae lineages.

conclusions
Here, we reconstruct Pliocene High Arctic Dipoides sp. palaeodiet from bone collagen δ 13 C and δ 15 N within the context of an isotopic dietary baseline composed of coeval ~ 4 Ma old terrestrial and freshwater plant macrofossil remains. The Beaver Pond site provides a very rare opportunity for such a palaeodiet reconstruction using coeval herbivore and plant remains. A Bayesian mixing model indicates that Dipoides sp. diet was composed of approximately equal proportions of woody plant material and freshwater macrophytes, with slightly more emphasis on macrophyte consumption. Dipoides sp. dietary preferences lie somewhere in between those of other North American late Cenozoic semiaquatic beavers (extant Castor and extinct Pleistocene giant beaver, Castoroides).
The consumption of woody plant material suggests that a proportion of the assemblage of the wood cut by Dipoides sp. at the Beaver Pond fossil site was the result of harvesting for consumption, possibly as part of an underwater winter food cache. The results also suggest that the early Miocene ancestor of the semiaquatic beaver lineage engaged in woodcutting and consumed woody plants as part of its diet. Swimming, woodcutting, and a diet of woody plants could have set the stage for the evolution of dam building behaviours-advantageous traits that may have been selected for by the cooling climate of the late Neogene, and which have resulted in Castor's modern role as a keystone species and ecosystem engineer.

Methods
The Dipoides sp. skeletal material and the plant macrofossils used in this study originated from the Beaver Pond fossil site, Unit III, as defined by Mitchell et al. 21 . Unit III is a peat layer that yielded the majority of the beavercut sticks and vertebrate faunal remains discovered at the site. It is interpreted to have been a rich fen connected to open water, within a larch-dominated forest ecosystem 21 .
Plant macrofossil preparation. Plant macrofossils were isolated from bulk samples of Unit III peat and identified to taxon. Macrofossils were extracted from the peat using a combination of water-flotation and wetsieving. Organic material greater than 0.425 mm was retained for further cleaning. Adhered sediment and moss were removed from the macrofossils using surgical forceps and repeated ultrasonic water baths. Cleaned macrofossils were dried at 26 °C for 24 h and identified to taxon using a binocular microscope.
Stable isotope analysis. Dipoides sp. bone collagen was extracted and its δ 13 C and δ 15 N measured at the Alaska Stable Isotope Facility (UAF). Collagen extraction was performed using a modified Longin 98 method of gelatinization. Organic contaminants were removed using XAD-2 resin 99,100 and collagen purification was performed according to methods developed by Matheus 101 . Non-soluble collagenous portions were rinsed to neutral pH, but not subjected to base treatment. Collagen was gelatinized in weak HCl (pH 3) under N 2 gas at 105° C until dissolved (2 to 6 h). The solution was centrifuged and filtered with a 0.45 µm syringe-type PTFE filter and the supernatant containing dissolved collagen was lyophilized and weighed to determine the percent collagen yield. Lyophilized collagen was then hydrolyzed in 6 N HCL under N 2 gas for 4 h at 120 °C. The hydrolyzates were passed by gravity flow through 2 cc of compacted Serva XAD-2 HPLC resin in syringe columns to extract humates and other long-chain organic contaminants that can adhere to fossil collagen. The hydrolyzates were passed through a 0.45 µm PTFE filter placed at the distal end of each syringe column and were dried by rotary evaporation. Stable carbon and nitrogen isotope analysis of the hydrolyzed collagen was performed using a GC-Isolink gas chromatography combustion system coupled to a Thermo Scientific Delta V Plus isotope ratio mass spectrometer operated in continuous flow mode, using helium as the carrier gas. Plant macrofossils were powdered using a ball-bearing mill and weighed into tin capsules (0.38 ± 0.02 mg). Stable carbon and nitrogen isotope analysis of macrofossil remains was conducted at the LSIS-AFAR facility at the University of Western Ontario (London, Canada). Samples were analyzed in continuous flow mode using a Costech elemental analyzer (ECS 4010), coupled to a Thermo Scientific ConFlo IV and Delta V Plus isotope ratio mass spectrometer in continuous flow mode, using helium as the carrier gas. One method duplicate (complete duplication of sample preparation and isotopic analysis) and one analytical duplicate (separate isotopic analysis of sample powder) were included for every ten samples. The carbon and nitrogen isotope measurements of the plant macrofossils were completed in separate analytical sessions. The first session was used to determine δ 13 C and nitrogen content (weight percent, N wt%); values of δ 15 N were determined in the second session, using individually tailored weights based on each sample's N wt%.
All isotopic results are reported in δ-notation in per mil (‰) relative to international standards. Collagen δ 13 C and δ 15 N were calibrated to VPDB and AIR, respectively. Analytical accuracy and precision were 0.0‰ for δ 13 C measurements, and 0.2‰ for δ 15 N measurements.
The Dipoides sp. and plant macrofossil isotopic results were incorporated into a statistically-based Bayesian mixing model (SIAR V4). This approach provides a statistically robust means of evaluating the relative dietary contributions of woody plants and aquatic primary producers to Dipoides sp. Diet 102 .