Range shifts in butternut, a rare, endangered tree, in response to past climate and modern conditions

Range shifts are a key process that determine species distributions and genetic patterns. A previous investigation reported that Juglans cinerea (butternut) has lower genetic diversity at higher latitudes, hypothesized to be the result of range shifts following the last glacial period. However, genetic patterns can also be impacted by modern ecogeographical conditions. Therefore, we re‐investigate the genetic patterns of butternut with additional northern population sampling, hindcasted species distribution models and fossil pollen records to clarify the impact of glaciation on butternut.


| INTRODUC TI ON
Determining the causes and consequences of species' range movements is key to understanding their ecology and evolution because range shifts determine community composition, coevolution, distribution of genetic diversity and niche requirements (Arenas et al., 2012;Blois et al., 2011). Major questions around range movements include migration speed, variation and peak timing (Ordonez & Williams, 2013); the degree to which species tolerate changing climate within their range (Zhu et al., 2012); and how much species and communities will shift ranges in response to modern climate change (Ricklefs, 2004). These topics are also relevant for conservation planning because as climates warm, species will need to shift their current ranges to track suitable climate (Lenoir & Svenning, 2015).
An important period of climate change, and resulting organismal range shifts, occurred during and following the last glacial period in the Pleistocene, and in particular the Last Glacial Maximum (LGM).
This period largely affected modern-day species occurrences and genetic patterns (Hewitt, 2004;Petit et al., 2003), and also exhibited rapid climate change, sometimes as rapid as modern-day climate change . The last glacial period lasted from approximately 130 ka (thousand years before present) to 20 ka, and the LGM lasted from about 26.5 to 19 ka (Hewitt, 2000), during which glaciers extended far south of their previous extent in the northern hemisphere. As glaciers expanded, species were pushed southward into glacial refugia, typically defined as small, distinct regions of suitable habitat near or surrounded by ice sheets during glacial cycles (Bennett & Provan, 2008). Subsequently, climate warming and glacial retreat allowed species to migrate north into their current ranges in a series of founder events, during which bottlenecks likely led to a reduction in genetic diversity with increasing latitude, while species maintained genetic structure indicative of shared glacial refugia during the LGM (Fuller & McGlaughlin, 2019;Hewitt, 2004). For example, Petit et al. (2003) determined that several woody European species exhibit genetic signatures of being pushed into southern glacial refugia, as demonstrated by clines in genetic diversity (higher diversity in the south and less diversity in the north of ranges) and shared structure representative of shared refugia in the major peninsulas of Europe (in Italy, Iberia and the Balkans). However, genetic patterns appear to be less consistent in eastern North American tree species (Lumibao et al., 2017).
Interpreting such migration histories is challenging because genetic patterns may result from multiple evolutionary forces. If species are more impacted by current climatic and geographical factors than past glaciation, species may exhibit genetic patterns more similar to the central-marginal hypothesis or isolation by distance (Hoban et al., 2010). The central-marginal hypothesis predicts that species will lose genetic diversity as distance increases from the range center, which is likely the most ecologically suitable habitat (Eckert et al., 2008). Range edges are where individuals of a species are expected to encounter marginal habitat conditions, leading to small population sizes and reduced genetic diversity from bottleneck effects (Lira-Noriega & Manthey, 2014). Species may also exhibit genetic differentiation based on modern geographical barriers or isolation by distance (Wright, 1943). For example, Euptelea pleiospermum, a Cenozoic relic tree native to China, exhibits high genetic diversity closer to the centre of its range and genetic structure consistent with current geographical barriers (Wei et al., 2016), consistent with expectations of the central-marginal hypothesis. In eastern North America, 14 tree species exhibited little genetic evidence of differentiation based on glaciation (Lumibao et al., 2017), suggesting that some species in this region may not have suffered severe effects of glaciation and eastern North American tree species should be examined for genetic patterns of the central-marginal hypothesis.
The challenges associated with interpreting genetic data can be partly overcome with the addition of other forms of data when trying to elucidate a species' past range movements. Large fossil pollen databases and resolved palaeoclimate models can be useful for contributing distinct lines of evidence for past species ranges and migration routes (Hao et al., 2018;Hoban et al., 2018). Pollen released by trees can be deposited in lakes, bogs or peatlands, where sedimentary layers accumulate over time. By extracting sediment cores, palaeobiologists can identify and count fossil pollen grains within radiocarbon-dated sedimentary layers. Presence of fossil pollen at specific locations and time points can reveal where tree populations occurred thousands of years ago. For example, Davis (1983) charted the migration of 12 major tree genera using fossil pollen records, while others have quantified the speed of range shifts during post-LGM warming for trees in eastern North America (Ordonez & Williams, 2013).
Species distribution models (SDMs) can reconstruct past suitable habitat range by coupling modern species habitat suitability with palaeoclimate models. Knowledge of modern ecological preferences can be hindcast into past climate conditions to predict past locations of suitable habitat. Reconstructions of Fagus orientalis distribution were used to identify past suitable habitat that was supported by fossil pollen records (Dagtekin et al., 2020). Hindcasting current climate-based species distribution models is particularly useful because climate has been a large determinant of plant distributions since the LGM (Ordonez & Williams, 2013). Along with reconstructing past species responses to rapid climate change, distribution modelling can identify migration routes, locations of glacial refugia, and biotic velocity through time.
Using multiple lines of evidence, we reinvestigate the postglacial history of butternut (Juglans cinerea, L., Juglandaceae), a coldtolerant relative of black walnut (J. nigra) that is currently threatened by habitat fragmentation and butternut canker disease (Pike et al., 2021). Previous studies on butternut showed reduced genetic diversity along a south-to-north gradient, which was interpreted as supporting rapid northward migration following glacial retreat (Hoban et al., 2010). Hoban et al. (2010) also identified strong genetic distinctiveness in New Brunswick populations, which was interpreted as evidence for a cryptic northern glacial refugium during the LGM. These results may have been influenced by undersampling at the northern edge of butternut's distribution; therefore, further investigation with more sampling in butternut's northern range is needed to elucidate the root of these genetic patterns.
We use newly available genetic data from butternut populations at the northern range edge and supplement our investigation with fossil pollen data and SDMs. Using three lines of evidence, we evaluate whether: 1. Butternut's genetic diversity declines along a south-to-north gradient, supporting a primary role of postglacial colonization in the genetic diversity structure of the species.
2. New Brunswick populations reveal distinct genetic signatures, suggesting long-term separation from the 'main' populations (e.g. as in a separate glacial refuge).
3. SDMs and fossil pollen data can identify possible suitable habitat and pollen (respectively) near New Brunswick, implying the presence of a cryptic refuge.

| Overview of methods
To determine whether previous results in butternut (Hoban et al., 2010) were influenced by sampling gaps, we used newly sampled individuals from butternut's northern range. In addition, we attempted to more directly model butternut's past movements using fossil pollen records and species distribution modelling. In addition, in the previous study, population differentiation and substructure analysis identified New Brunswick populations as highly differentiated from all other butternut populations, potentially indicative of cryptic glacial refugia near New Brunswick during the LGM. To test the hypothesis of a cryptic refugium for butternut during the LGM, we needed to (1) use genetic structure analysis to determine whether, with more extensive sampling, we identified similar patterns, (2) examine fossil pollen in areas near glacial refugia shortly after the LGM and (3) identify the degree of past suitable habitat near these proposed refugia using hindcast species distribution models. Lastly, fossil pollen records and past species distribution models were used to calculate biotic velocity (i.e. rate of range shift), which we then interpret in light of current genetic diversity.

| Study species
Juglans cinerea L. (butternut) is a medium-sized, short-lived hardwood tree with a life span around 75 years (Rink, 1990). It is native to eastern North America ( Figure 1a) and is an outcrossing, windpollinated tree. Butternut is threatened by butternut canker disease, caused by Ophiognomonia clavigignenti-juglandacearum (Oc-j), an exotic fungus introduced to North America in the early 20th century (Pike et al., 2021). Butternut, now locally extirpated in many forests, is listed as a 'species of concern' in many areas of the United States (U.S.) and is a Regional Forester Sensitive species in 13 of the 16 National Forests in the eastern U.S. (Morin et al., 2018) and 23 out of 31 states list butternut as vulnerable, imperilled or critically imperilled (NatureServe Explorer, 2021). Oc-j has rapidly spread throughout its naturalized range and in 1997 was found in New Brunswick, where it is disjunct and considered the most northeastern part of butternut's range (Harrison et al., 1998). In the U.S., butternut canker disease results in high tree mortality (Morin et al., 2018). In Canada, Oc-j has infected 90% + of trees in Quebec and Ontario (COSEWIC, 2017) and over 70% of trees from New Brunswick (Williams et al., 2020). In Ontario, an annualized mortality rate of 5.4% was calculated over a 7-year period on 60 sites (COSEWIC, 2017). This rapid decline has led to butternut being listed as endangered at the federal level in Canada and imperilled or critically imperilled in three Canadian provinces (COSEWIC, 2017).

| Sampling for genetic analyses
In this study, individuals from a previous butternut analysis (see Hoban et al., 2010) were combined with a subsequent sampling effort to better characterize genetic patterns in butternut. The initial sampling effort sampled 1004 trees from 29 populations, largely from U.S. populations. The second sampling effort sampled 757 butternut trees from northern populations, mostly in Canada (Ontario, Quebec and New Brunswick).

| Geographical data preparation
We organized butternut individuals into cohesive population units to assess how geographical locations were related to genetic diversity. Individuals were sampled at different times, which resulted in disparate sample sizes that could bias genetic analyses. Therefore, we organized butternut individuals into populations by geographical region to represent breeding groups ( Figure  Quebec populations were below this threshold and were removed from our main analysis, but all analyses were repeated with these populations (see Supplemental). Longitude, latitude and distance to range edge were then calculated for each population. Mean population latitude and longitude were calculated in the R using the 'sp' package 1. 4-5 (Pebesma & Bivand, 2005). We created butternut's range extent shapefile in an R script (see Figure S2) and calculated the distance to the range edge for each population using its mean location and the dist2Line function in the R package 'geosphere' 1.5-10 (Hijmans, 2019). Individuals were organized into 24 populations but are visualized by geographical region for clarity ( Figure 1a).

| Genetic data preparation and analyses
Following population organization, we extracted and genotyped these individuals' DNA according to Hoban et al. (2008) using the microsatellite loci: B114, B159, WGA, A5_2, B157, B212_2, B121, B147, B249, B262 and B264. Individuals with more than 25% missing data were removed, leaving a total of 1635 individuals for genetic analyses. Since genotypes were scored at different times, and microsatellite genotypes can vary due to the equipment and fragment size detection software used, loci were assessed for differences based on year scored. Details of these analyses are described in-depth in the supplement ( Figure S3). Loci were assessed for deviations from Hardy-Weinberg Equilibrium, linkage disequilibrium and null allele frequency, and then diversity statistics (allelic richness and expected heterozygosity) were calculated by population. Hardy-Weinberg Equilibrium, linkage disequilibrium and expected heterozygosity were assessed in the R package 'poppr' 2.9.2 (Kamvar et al., 2014) and allelic richness and population pairwise Fsts were assessed in the 'hierfstat' R package 0.5-7 (Goudet, 2004). Null alleles were identified using the R package 'PopGenReport' 3.0.4 (Adamack & Gruber, 2014).

| Regression analyses
We then regressed each populations' mean genetic diversity levels with their mean geographical location to identify support for either past range shifts affecting butternut's diversity distribution or modern climatic conditions. A negative linear relationship between mean population latitude and a genetic diversity statistic (allelic richness or to the centre of butternut's range. A significant, positive linear relationship between genetic diversity and distance to range edge would also support the central-marginal hypothesis, since genetic diversity would be higher at the centre of the species range and lower at the range margins. Regressions were also performed with and without Wisconsin (just above our minimum population sample size threshold) and Quebec (just below our minimum population size threshold) populations since they had much smaller population sizes and diversity may be affected more by small effective size than range-wide diversity patterns. We also tested for isolation by distance (IBD), where populations become more genetically distinct with distance, which may be another indication of modern climate largely affecting butternut, by determining whether there was a significant linear relationship between the mean geographical locations of each population and pairwise Fst between populations.

| Genetic structure analyses
Initial analyses of butternut genetic structure indicated divergence between New Brunswick butternut and all other populations, so to determine whether these patterns were consistent with the addition of new individuals, we ran clustering analyses. However, to ensure these analyses would not be biased by familial groups, individuals with greater than 25% (half-sibling) relatedness were removed, leaving 993 individuals for these analyses. Relatedness was assessed in the R package 'Demerelate' 0.9-3 (Kraemer & Gerlach, 2017) using the Loiselle statistic, which corrects for small population sizes. Genetic clustering patterns were examined in STRUCTURE (Pritchard et al., 2000) and principal coordinate analysis (PCoA).
Structure results were assessed for the best supported K using the Evanno method (Evanno et al., 2005) as implemented in STRUCTURE HARVESTER (Earl & von Holdt, 2012). PCoA was run in the R package 'adegenet' (Jombart, 2008).

| Fossil pollen analysis
To better resolve butternut's past geographical locations and range shifts, we examined butternut pollen records over time since the LGM. Pollen records were downloaded on August 7, 2020 from the Neotoma Paleoecology Database using the neotoma package in R (Goring et al., 2015) which is documented in more detail in the supplement ( Figures S4, S5). Pollen age estimates that were not already converted to calendar years were converted using the 'Bchron' 4.7.6 package in R (Haslett & Parnell, 2008). We retained only butternut pollen counts located in North American east of −110 degrees longitude (east of the Rocky Mountains), our study area of interest.
We visualized pollen records in 1000-year time increments along with glacial coverage for each time period, which is appropriate because temporal resolution of fossil pollen data in eastern North America during this time period is about 500 years .
We show nine time points here for simplicity and because several time points were similar (See Figure S6 for all time points). Glacial coverage was visualized by downloading polygon shapefiles of North American ice sheets from time periods most closely aligned with 1000-year intervals from 21 to 1 ka (Dalton et al., 2020).
Additionally, we calculated the distance of butternut pollen occurrences to the glacier by averaging the distance from each pollen occurrence to the glacier for each time period to provide us with an average distance to the glacial margin. Distance to the glacier margin was calculated using the dist2Line function in the 'geosphere' package in R  and all pollen maps were constructed using the 'ggplot2' package in R (Wickham & Chang, 2012).
To determine migration rate based on pollen records, we first calculated the location of the centroid of the pollen records within each of 21 time periods by taking the mean of all x-, y-coordinates falling within the same time bin. We then calculated the Euclidean distance between each centroid and the centroid of the subsequent time bin using the pointDistance function in the 'raster' R package (Hijmans, 2020). Finally, we divided each of the distances (in meters) between the centroids by 1000 years.

| Species distribution modelling
In addition to pollen models and genetic data, butternut's movements before, during and since the LGM were assessed by projecting current distribution models into past climate scenarios. This process involved the following: (1) determining butternut's modern climate niche using boosted regression trees (BRT), (2) predicting the locations of butternut populations in the past by projecting butternut's modern ecological niche into eight past time periods, and (3) visualizing possible locations of past butternut populations by mapping the projected habitat suitability across time. BRT were used to predict butternut's modern occurrence because they can use many types of data and handle missing data and outliers well (Elith et al., 2008). In summary, we compared 19 WorldClim variables (Fick & Hijmans, 2017) with butternut presence and selected the five variables most correlated with presence and least autocorrelated, and generated a prediction of suitable habitat (see Figure S7, Table S1).
The function uses as input a series of rasters and each raster's time point. In our case, we used the eight SDM output rasters, each of which represents a different time range from 130 ka to present. For the vector of times, we took the middle (i.e. the mean) of the SDM time ranges. Results were converted to meters per year.

| Genetic diversity overview
Overall, 1635 butternut individuals were successfully genotyped and retained for final genetic diversity and structure analyses. Diversity was relatively high in populations, with 233 alleles observed across 11 loci (mean per locus = 21.2, mean per population = 126.2) and the number of alleles within populations ranging from 83 to 168 (Table S2). Allelic richness and expected heterozygosity ranged from 6.47 to 9.23 and 0.751 to 0.852, respectively. Wisconsin populations had the lowest diversity of all populations (allelic richness lower than 7, expected heterozygosity lower than 0.78) (Table S2). Deviation from HWE expectations was locus and population specific (Table S3), likely indicating close relatives in some populations, and linkage disequilibrium was observed between some loci (Table S4). Null alleles were infrequent in most loci, near or exceeding mean frequency of 10% only in B157, B212_2 and B264 (means = 9.00%, 13.1% and 16.8%, respectively) (Table S5).

| Genetic diversity highest near the centre of butternut's range
Regression of population geographical metrics (e.g. mean latitude, distance to range edge) with genetic diversity metrics (e.g. allelic richness, expected heterozygosity) support the central-marginal hypothesis rather than the postglacial migration hypothesis, contrary to our expectations. To test for the postglacial colonization hypothesis, we regressed mean population latitude with genetic diversity metrics (e.g. allelic richness and expected heterozygosity) and these relationships were significant but not well fit to the data (Table 1). To test for the central-marginal hypothesis, we performed quadratic regressions between genetic diversity metrics and mean population latitude and also a linear relationship between population distance to range edge, as lower diversity at range edges is more indicative of the central-marginal hypothesis. We found a significant quadratic relationship between mean population latitude and a significant, positive linear relationship between genetic diversity metrics and population distance to range edge, without Wisconsin populations (Table 1). When Wisconsin populations were included in regressions, the relationships were either not significant or well-fit to the data (Table 1) population size than range edge effects. All of these patterns are the same when Quebec populations are added (Tables S7). Isolation by distance would be supported by a significant linear regression with pairwise Fst and distance; this regression was significant but explained little of the variance in the data ( Figure S9).

| Genetic structure is distinct in New Brunswick
Genetic structure analyses were run to determine whether genetic structure remained distinct between New Brunswick and other

| Fossil pollen is found near New Brunswick during the LGM
Fossil pollen records were examined to determine whether there was evidence for butternut persisting near New Brunswick, close to glacial margins. Butternut pollen was not identified near New Brunswick during the LGM, but some terrestrial environments exposed during the LGM are currently under ocean water. Butternut pollen first appeared in New Brunswick, Nova Scotia and southern Quebec at 14 ka, when ice sheets were first breaking apart during deglaciation and habitat was becoming available between smaller ice sheets. With further glacial retreat, pollen occurrences became more numerous within the current butternut range (Figure 3). Butternut pollen was commonly found close to the glacial margin at the LGM and tracked glacial retreat closely until about 10 ka ( Figure S12), when butternut reached its present-day range (Figure 3, Figure S6). Migration rates calculated from pollen records ranged from under 50 m/year to 450 m/year (at 17 ka), oscillating between 50 and 200 m/year from 16 to 2 ka ( Figure S14).

| Hindcast distribution models support the presence of suitable habitat near New Brunswick
We used SDMs to predict the presence of suitable habitat for butternut during and following the LGM. Butternut's modern ecological niche was best predicted by the following variables: precipitation of the wettest month, mean diurnal range, mean temperature of the driest quarter, mean temperature of the wettest quarter and seasonal precipitation (Table S1) Younger Dryas period (12.9-11.7 ka) ( Figure S13).

| DISCUSS ION
In this study, we aimed to use additional sampling as well as fossil pollen and hindcast distribution models to better understand the impact of glaciation and postglacial range shifts on the genetic patterns of the rare species, Juglans cinerea (butternut).

Specifically, we reassessed previous interpretations of genetic
patterns after further scrutiny of the northern range edge. Our results provide a more detailed picture of the range shifts of but-

| Genetic diversity is best explained by centralmarginal hypothesis, not postglacial expansion
A previous examination in Hoban et al. (2010) found that butternut's northernmost populations had the lowest diversity, similar to observations in 21 woody species in Europe, likely due to successive postglacial migration bottlenecks (Petit et al., 2003). However, with the addition of more sampling near butternut's northern range edge, genetic diversity patterns are more similar to the patterns predicted by the central-marginal hypothesis, rather than postglacial migration or isolation by distance. The previous findings that genetic diversity declined with increased latitude may have been partly due to sampling strategy. In Hoban et al. (2010), fewer individuals were sampled from butternut's northern range and most northern range individuals were sourced from Wisconsin populations. Wisconsin forests have been strongly impacted by modern land-use conversion and have small butternut populations, leading to unusually low genetic diversity in these populations, which do not fit well to the species-wide relationship between genetic diversity and ecology (Table 1). Including the smaller populations from Quebec, butternut populations fit the pattern of higher diversity closer to the centre of the range, where habitat is higher quality (Gullberg et al., 1998).
Overall, with improved sampling, we suggest butternut's genetic diversity patterns are more likely shaped by modern ecological marginality than past range shifts.

| Genetic structure analyses support distinction of New Brunswick
Range-wide analyses of genetic structure provide additional information about a species' biogeographical history by suggesting which populations had more genetic connectivity. Supporting our hypothesis, structure and PCoA analyses suggest butternut has two distinct genetic clusters, with the U.S. and Ontario populations forming one cluster and New Brunswick populations forming a second (Figure 2, Figure   S12, Table S6). We recognize that structure is biased towards selecting K = 2 as the optimal K (Cullingham et al., 2020); however, other K values similarly identified New Brunswick as distinct from other populations ( Figure S11). Quebec populations cluster differently based on clustering methods, as they cluster with New Brunswick in structure analysis and cluster independently in PCoA analysis ( Figure S13). This could be due to small population sizes, limited spatial sampling in Quebec

| Fossil pollen and SDMs support the possibility of a cryptic refugia in New Brunswick
Butternut's pollen records and species distribution modelling largely support our hypothesis and the genetic evidence of a cryptic glacial  Figure   S6). While suitability scores were low (20-30%) near Nova Scotia and New Brunswick in certain time periods (e.g. 17.9-12 ka), suitable habitat is still present in those areas during all time periods (Figure 4, Figure S8). We identified a lake off the coast of modern-day Maine that separated two portions of butternut's suitable habitat 20 ka which could have led to butternut's two genetic clusters.
Assessing butternut's dispersal ability also allows us to assess the likelihood of a cryptic glacial refugium near New Brunswick. Tree squirrels are butternut's main seed dispersal vector and generally disperse below 100 m (Laricchia et al., 2015), while seeds of butternut's relative, Juglans nigra, were observed to be moved as far as 2000 m by crows (Cristol, 2001). If squirrels or crows dispersed butternut seeds, the biotic velocity would be 100-2000 m/generation,

| Caveats
Each data source used in this paper has deficiencies (Hoban et al., 2018). Genetic methods represent multiple ecogeographical patterns and therefore provide imperfect evidence of the past. New

F I G U R E 4
Habitat suitability and glacial coverage (blue polygon) for butternut at several past time points. Areas with (somewhat) suitable habitat are present in areas currently under ocean water, indicating the possibility that butternut could have been present there in a glacial refuge. Maps were created using Albers equal area projection Brunswick populations demonstrate a slight reduction in genetic diversity, consistent with patterns representative of the centralmarginal hypothesis, but could also represent a bottleneck in New Brunswick populations following postglacial migration. Therefore, while we provided support for modern ecogeographical processes influencing genetic patterns in this species the most, it is important to note all mentioned ecological processes are likely influencing butternut to some degree, similar to results examining the genetic patterns of other tree species from this region (Bemmels & Dick, 2018).
Fossil pollen data have known limitations as well. Pollen can be carried by wind, even hundreds of kilometres from a population (McLachlan & Clark, 2004) and so the presence of fossil pollen at a site does not necessarily mean a tree population existed at that location. To avoid interpreting a single pollen occurrence as evidence of the presence of a butternut population, we examined maps including many pollen occurrences ( Figure S5) and interpreted patterns with SDM and genetic evidence. Species distribution models assume that climate data alone provide an accurate picture of ecological niches and butternut's fundamental niche has not changed during the last 130,000 years. Solar radiation, available canopy cover and soil characteristics also shape butternut's distribution and habitat preferences (Crystal & Jacobs, 2014), but were not included in our SDMs.
However, comparisons of past range models and pollen distributions have found similarity among past climate modelling and past range distributions (Ordonez & Williams, 2013) and between multiple climate models (Beyer et al., 2020). Overall consistency between genetic patterns, fossil pollen records and suitable habitat suggests the usefulness of this modelling technique and combining multiple lines of evidence.

| CON CLUS IONS
We add support to several studies (Bemmels & Dick, 2018;Lumibao et al., 2017;Soltis et al., 2006) that have identified weak or inconsistent genetic diversity impacts (weak clines in genetic diversity with latitude, and structure is relatively idiosyncratic) from postglacial migrations in eastern North American tree species, contrasted to the strong structure in Europe (Petit et al., 2003), possibly due to contemporary ecological marginality. It is certainly challenging to distinguish the central-marginal pattern from other processes, and our results based on adding more complete sampling to previous data suggest that other previous studies may warrant revisiting with more complete range-wide data (McLachlan et al., 2005;Victory et al., 2006). The sufficiency of geographical range coverage may also be useful to consider in meta analyses such as those evaluating the central-marginal hypothesis and biotic velocity (e.g. Hamann et al., 2015). We conclude that more robust sampling has provided better evidence of past genetic patterns and contributed to a better understanding of past demographic histories of butternut.
Additionally, we emphasize the importance of leveraging evidence from multiple data sources when evaluating species biotic velocity, or migration speed, and when trying to locate species' glacial refuges. Genetic evidence alone is inconclusive on the possible New Brunswick refuge and biotic velocity. Examining pollen fossils with spatiotemporal resolution and niche models provides strongly suggestive, though not fully conclusive, evidence of this northern glacial refuge for butternut. Comparing the findings from the distribution models and pollen evidence also gives a better sense of uncertaintythese two lines of evidence are only partly in agreement on overall velocity or on periods of more or less rapid movement. Uncertainty in such estimates is important for determining contributing factors to biotic velocity, and for reaching conclusions about forests' ability to track climate change (Hoban et al., 2019). Further improvement to our work can be made using next-generation sequencing and Approximate Bayesian Computation to more formally integrate genetics with other lines of evidence to understand glacial refugia and migration speed (Bemmels et al., 2019).

| Conservation applications
First, we note that butternut's habitat may be under threat due to climate change. SDMs generated for this study suggest butternut's range is largely predicted by seasonal precipitation, patterns of which are changing rapidly due to climate change (Pendergrass et al., 2017), particularly in northern areas where precipitation is already quite variable (Konapala et al., 2020). Previous analyses have also found that butternut is also generally found in mesic habitat, and changes to seasonal precipitation are also predicted to change localized stream and river flow (Nkhonjera et al., 2020). Bemmels et al. (2019) identified that many modern cold-tolerant species specialized on mesic environments could have persisted very close to glacial margins, which seems possible for butternut due to the locations predicted by suitable habitat models and fossil pollen records, but this, in turn, makes them particularly under threat by these climate-change-related effects. Areas that may be suitable for this species now may not be suitable for this species in 50 years (Barnes & Delborne, 2019), so using climate change modelling and carefully considering habitat needs of this species are essential when planning restoration (Pike et al., 2021).

Second, the geographical and genetic distinction of New
Brunswick butternut indicates these populations may warrant special protection and possibly their own taxonomic unit for conservation. In Canada, the Committee on the status of endangered wildlife in Canada (COSEWIC) has the mandate to evaluate if a species should require its own taxonomic unit or designatable unit (DU) below the species level. In order to qualify, DUs need to be recognized as discrete but also represent evolutionarily significant units when compared to other populations. We have shown that New Brunswick populations are genetically distinctive from the rest of the species range. Discreteness is also supported geographically and climatically. New Brunswick populations are naturally disjunct, and may have been for a long time, and appear within their own ecozonethe Acadian Forest ecozone. Of course, genetic differentiation could be further supported using more slowly evolving DNA markers that show either a deep intraspecific phylogenetic divergence or strong natural selection due to environmental conditions, which has not been established (Laricchia et al., 2015). Although more data and analysis are required to fully characterize the significance of New Brunswick populations as a DU, the data herein support the need to follow through with specific conservation activities that will capture the maximum amount of remaining genetic diversity from both groups before it is lost to butternut canker.

| Summary
Our work better resolved past locations and movements of butternut using multiple lines of evidence, and we suggest there is potential for re-evaluating past conclusions on other eastern North American taxa. A better understanding of more species' past range shifts in eastern North America is critical to comparing biogeographical understanding compared to European taxa, and to better understand how these forests may respond to future changes in climate.