Genomic-environmental associations in wild cranberry (Vaccinium macrocarpon Ait.)

Abstract

Understanding the genetic basis of local adaptation in natural plant populations, particularly crop wild relatives, may be highly useful for plant breeding. By characterizing genetic variation for adaptation to potentially stressful environmental conditions, breeders can make targeted use of crop wild relatives to develop cultivars for novel or changing environments. This is especially appealing for improving long-lived woody perennial crops such as the American cranberry (Vaccinium macrocarpon Ait.), the cultivation of which is challenged by biotic and abiotic stresses. In this study, we used environmental association analyses in a collection of 111 wild cranberry accessions to identify potentially adaptive genomic regions for a range of bioclimatic and soil conditions. We detected 126 significant associations between SNP marker loci and environmental variables describing temperature, precipitation, and soil attributes. Many of these markers tagged genes with functional annotations strongly suggesting a role in adaptation to biotic or abiotic conditions. Despite relatively low genetic variation in cranberry, our results suggest that local adaptation to divergent environments is indeed present, and the identification of potentially adaptive genetic variation may enable a selective use of this germplasm for breeding more stress-tolerant cultivars.

Introduction

Genetic variation in natural populations is influenced by heterogeneous environmental conditions, which may act as selective pressures that lead to local adaptation. Studying the genetics of this adaptation is relevant for many disciplines, including human health, evolution, and plant and animal breeding. In crop improvement, wild populations that have been subject to the forces of local adaptation have often been targeted as sources of novel genetic variation (Brozynska et al. 2016). While a handful of successful cases of unimproved-to-elite introgressions exist (e.g. Singh et al. 2009), crop wild relatives in many systems have been less characterized. A better understanding of the genetic variation influencing local adaptation would provide more raw material for breeders to develop cultivars adapted to changing environmental conditions.

Many approaches have been developed to dissect the genetic architecture of local adaptation and the genetic loci that may be under selection. Those that examine high degrees of genetic differentiation between populations, such as outlier tests based on F_ST, are ideal for detecting positive selection on single beneficial genetic variants (i.e. selective sweeps; Walsh and Lynch 2018). An alternative approach is to correlate environmental factors with allele frequency, or environmental association analysis (EAA). Those EAA methods that use explicit environmental information (the potential selective pressures driving local adaptation) and correct for population structure (which may be correlated with environmental gradients; e.g. Coop et al. 2010; Eckert et al. 2010; Frichot et al. 2013) may be most suited to detect subtle changes in allele frequency (Hancock et al. 2010). A useful EAA method is to deploy the same modeling framework as genome-wide association (GWA) studies, substituting an environmental variable for the quantitative trait. This approach has been successfully used to understand local adaptation in Medicago truncatula (Yoder et al. 2014), cultivated soybean (Glycine max L.) and its wild progenitor (Glycine soja L.; Anderson et al. 2016; Bandillo et al. 2017), and barley (Hordeum vulgare L.; Lei et al. 2019).

Understanding the genetic basis of local adaptation in natural plant populations such as crop wild relatives could be immensely useful for plant breeding. Crop wild relatives may harbor novel and favorable genetic variation, including resistance to pathogens or tolerance to environmental stresses, and these traits may be the result of local adaptation to such biotic or abiotic conditions; however, assessing collections of unimproved germplasm for these and other traits of interest is resource-intensive and often impractical. Thus, just as GWA or quantitative trait loci (QTL) mapping approaches can help screen wild or landrace germplasm for favorable alleles influencing traditional traits of interest (e.g. Muñoz-Amatriaín et al. 2014; McCouch et al. 2016), identifying the genomic regions underlying local adaptation could be useful in prioritizing germplasm to introgress potentially useful genetic variation for biotic or abiotic stress tolerance from crop wild relatives into improved cultivars.

The American cranberry (Vaccinium macrocarpon Ait.), a long-lived woody perennial, is a unique cropping system in which to study the use of unimproved germplasm in a recent domesticate. More intensive cranberry production, beginning as early as 1810 and accelerating through the late-19th and early-20th centuries, was based on the cultivation of “native selections” (i.e. landraces) identified by growers from natural stands (Eck 1990; Vorsa and Zalapa 2019). Intentional cranberry breeding started in 1929, yet initial and even modern cultivars are at most 2 generations removed from wild and native selection germplasm (Vorsa and Zalapa 2019). Unimproved wild cranberry is native to eastern North America from the mid-Atlantic United States to the maritime Canadian provinces and west to the Great Lakes region (Vander Kloet 1988), and is readily found within many of the current major cranberry production regions (i.e. Wisconsin, Massachusetts, and New Jersey). Though the occurrence of wild cranberry is generally restricted to acidic low-nutrient soils, heterogeneous environments across the native range are associated with phenotypic variation in natural populations (Vander Kloet 1983; Vorsa and Johnson-Cicalese 2012).

Progress toward critical breeding goals in cranberry could benefit from the targeted use of genetic variation from natural populations. Cultivation of this fruit crop is hampered by many abiotic and biotic stresses, including extreme temperatures (i.e. frost damage and heat stress), water availability (both drought and flooding), and viral and fungal disease pressure (Eck 1990; Vorsa and Johnson-Cicalese 2012; Vorsa and Zalapa 2019). While natural populations may possess favorable genetic variation imparting tolerance to these stresses, knowledge of the genomic location and effect of potential stress tolerance loci would aid in informing the selection of wild cranberry individuals as parents in crosses. This is particularly important given the long juvenility of cranberry and the extended time required to establish yield trials.

The objective of this research was to assess genetic variation in a collection of wild cranberry accessions and, using a combination of outlier-based tests and EAA, identify genomic regions that potentially underly tolerance to abiotic stresses and could serve as targets for selection in improving cultivated cranberry.

Materials and methods

Plant material and marker genotyping

The plant material used for this study included 111 accessions of wild cranberry from a germplasm collection maintained by Rutgers University. These accessions were collected from natural stands across the native range of cranberry, centered in the northeastern United States and eastern Canada (Vander Kloet 1988). The individuals originated from 17 locations in 10 states/provinces (DE, MA, ME, MI, NB, NJ, NY, PA, WI, and WV; Fig. 1), and the number of individuals from each sampling location ranged from 1 to 38, with a median of 4.

Fig. 1.

Wild cranberry populations were sampled from 17 sites in the native range (darker shaded states/provinces; Vander Kloet 1988). The number of individuals per population ranged from 1 to 38, with a median of 4. Colored points are used only to separate populations.

Genotyping-by-sequencing (GBS) was used to obtain SNP marker genotypes. The DNA extraction, library preparation, and read generation steps were described previously (Diaz-Garcia et al. 2020). We used the TASSEL pipeline (v. 5.2.75; Glaubitz et al. 2014) to process and align reads and call SNP variants. Reads were aligned to a cranberry physical reference genome of an inbred line derived from the native selection “Ben Lear” (Kawash et al. 2021) using Bowtie2 (v. 2.3.4; Langmead and Salzberg, 2012). Variants were subsequently filtered for only biallelic SNPs, a minimum genotype quality score of 40, a minimum average read depth of 10, and a minimum genotype read depth of 7. As the restriction enzymes used in GBS preferentially cut in gene-rich regions (Elshire et al. 2011), which are often limited in transposable elements (Lippman et al. 2004; Li et al. 2017), no maximum read depth was applied. Given the preferentially outcrossing habit and high heterozygosity of cranberry, we phased marker haplotypes and imputed genotypes using the approach implemented in Beagle (v. 5.0; Browning et al. 2021), with an initial effective population size estimate of 1,000, 6 burn-in iterations, and 25 subsequent iterations.

Pairs of SNPs in complete linkage disequilibrium (LD) (i.e. r² = 1) were identified, and the SNP with the greater minor allele frequency was retained. The remaining 7,706 SNPs were then used to construct a realized additive relationship matrix (K) among the 111 individuals via the R package rrBLUP (v. 4.6.1; Endelman 2011). These marker genotypes were further filtered on a minimum minor allele count of 10 (minimum minor allele frequency of 10/111 = 0.0901) and maximum pairwise LD of r² = 0.90 (again retaining the marker in each pair with greater minor allele frequency), leaving 4,457 SNPs for environmental association and genetic diversity statistics. These marker genotypes were coded as 1 for reference allele homozygous, 0 for heterozygous, or -1 for alternate allele homozygous.

Bioclimatic and soil variables

Approximate geographic coordinates of the wild cranberry sampling locations were used to query climate and soil data. Though the precise latitude and longitude information were not available, location descriptions were specific enough to infer geographic coordinates. We used this information to query the WorldClim databased for 19 bioclimatic variables (Hijmans et al. 2005) and the ISRIC SoilGrids database for 10 biophysical soil variables. To account for the uncertainty in the geographic coordinates of the cranberry sampling locations, we obtained the 19 bioclimatic variables at a resolution of 5 arcminutes (∼47 km²) and the 10 soil variables at a resolution of 2.5 arcminutes (∼12 km²). The soil data were available at 6 depths: 0–5, 5–15, 15–30, 30–60, 60–100, and 100–200 cm. Given the relatively shallow (≤15 cm) root system of cranberry (Eck 1990), we aggregated these data into topsoil (≤15 cm) and subsoil (>15 cm) layers using a weighted mean, where the weights were proportional to the size of the original layer depths (Anderson et al. 2016). This resulted in a total of 20 soil variables. The names and descriptions of the 39 bioclimatic and soil variables are available in Supplementary Table 1.

Population structure, LD, and genetic diversity

We investigated population structure among the wild cranberry individuals using a principal component (PC) analysis based on the spectral decomposition of the K matrix. In addition, pairwise SNP LD (r²) measurements were used to fit an LD decay curve for each chromosome based on the nonlinear regression model described in Abecasis et al. (2001), with an assumed recombination rate of 1 cM/Mb. These fitted models were used to determine the LD half-life and the physical distance at which LD decayed below r² = 0.2.

F _ST was measured using the variance approach of Weir and Cockerham (1984) and implemented in the R package hierfstat (Goudet 2005). Given the smaller sample sizes of each of the populations and the lack of apparent axes along which to partition the populations for comparisons, we calculated overall F_ST for each marker and a single F_ST estimate for the entire collection of accessions. To detect genomic regions exhibiting elevated spatial allele frequency gradients, we conducted a spatial ancestry analysis using the SPA software (v. 1.13; Yang et al., 2012) using default settings and the marker genotypes and sampling location coordinates (i.e. latitude and longitude) as inputs. Outlier SNPs were identified as those with F_ST or SPA scores exceeding the 0.999 quantile of each statistic.

Genetic distance was calculated between pairs of individuals using the R package ape (v. 5.5; Paradis and Schliep 2019). We summarized the distance estimates by calculating the average distance within and between populations. In addition, we calculated observed heterozygosity, within-population genetic diversity (expected heterozygosity), and among-population genetic diversity using the R package hierfstat (v. 0.5.7; Goudet 2005).

Environmental association analysis

A mixed-model approach was used to identify associations between the bioclimatic and soil variables and SNP markers (Yoder et al. 2014; Anderson et al. 2016; Lei et al. 2019). The models were structurally similar to traditional GWA models, but with the bioclimatic or soil variables used as the response variable (instead of the phenotypic values of a quantitative trait). The general model form was $\mathbf{y}=\mathbf{Qw}+\mathbf{X}\mathrm{\beta }+\mathbf{Zu}+\mathbf{e}$, where y is a vector of bioclimatic or soil data, Q the incidence matrix of PCs (if modeling population structure), w the vector of fixed PC effects, X the SNP marker genotype matrix, β the fixed SNP effect being tested, Z the incidence matrix of individuals, $\mathbf{u}\sim \mathit{MVN}(0,\mathrm{ }\mathbf{K}{\sigma }_u^2)$ the vector of random polygenic background effects, and $\mathbf{e}\sim \mathit{MVN}(0,\mathrm{ }{\sigma }_e^2)$ the vector of random residuals. We tested the following model variations: one with no fixed PC effects and only the polygenic background (i.e. K), one with the fixed effect of the first 3 PCs from the aforementioned PCA (i.e. K + Q), one with the fixed effect of latitude (i.e. K + latitude), one with PCs and latitude (i.e. K + Q + latitude), one with latitude and the fixed effect of longitude (i.e. K + latitude + longitude), and one with PCs, latitude, and longitude (i.e. K + Q + latitude + longitude). Models were fitted on a per-marker basis via the R package rrBLUP using the population parameters previous defined (P3D) option. Subsequently, the 6 models were compared visually using Q–Q plots (Supplementary Fig. 1) and the P-value inflation factor (Λ) via the R package QCEWAS (v. 1.2.2; Van Der Most et al. 2017). This comparison showed that the K and K + Q models led to similar results (Supplementary Fig. 1); we therefore used to the simpler K model in the analysis of all environmental variables. A per-variable false discovery rate (FDR) threshold of 0.20 was used to selected markers with a statistically significant association with the environmental variables.

The F_ST, SPA, genetic distance, and environmental association analyses were conducted using SNP marker data of only those individuals from populations with at least 3 samples (104 total individuals from 12 populations/locations).

Candidate loci characterization

SNPs identified as significant in the EAA or outliers in the SPA or F_ST analysis were used to search for nearby predicted genes based on the genomic annotations developed by Kawash et al. (2021). We limited our search to those predicted genes within 17 kb of a given SNP (i.e. the average distance at which LD decayed to r² = 0.2; Fig. 2). For each significant SNP, we determined the number of nearby predicted genes, the distance to the nearest predicted gene, and any inferred Arabidopsis orthologs.

Fig. 2.

a) PC analysis using SNP marker data indicated population structure among wild cranberry accessions (colors indicated different sampling regions), with accessions from the MA1, NJ1, and DE1/MI1/WI1 locations forming the most distinct clusters. b) LD (r²) between pairs of SNPs decayed quickly, as summarized using nonlinear regression. The pattern of decay was consistent for most chromosomes (colors), though chromosomes 3 and 8 were exceptions. Average r² was negligible after 200–300 kb (plot window).

Results

Variation in climate and soil variables

The locations sampled for wild cranberry populations varied across temperature, precipitation, and soil variables. Many variables reflected the conditions known to be suitable to cranberry and Vaccinium species at large, including low soil pH (∼4.5–5.7), high sand content (∼30–70%), and relatively high annual precipitation (∼800–1,400 mm yr⁻¹; Supplementary Fig. 2). We observed generally moderate to high correlations (mean: 0.57, range: −0.052 to 0.97) between the topsoil and subsoil layers of the soil variables (Supplementary Fig. 3). A PC analysis of all environmental variables indicated that 6 PCs were important for explaining the variation, after which additional PCs individually explained very little of the variance (<2.5%). Together, the 6 PCs explained about 94% of the variation. The variables most highly correlated with these first PCs (i.e. with the highest-magnitude loadings) were annual precipitation, elevation, mean temperature of the wettest quarter, subsoil clay content, maximum temperature of the warmest month, and subsoil cation exchange capacity (Supplementary Table 2). These first 6 PCs were included as separate variables in the EAA.

Population structure, LD, and genetic diversity

Population structure, as visualized by PC analysis, was apparent among the 111 wild cranberry individuals used for our study (Fig. 2a). This analysis confirmed our expectations of geographic distinction of genetic variation, separating individuals into one cluster of NJ origin (n = 14), one of MA origin (n = 39), and the remaining individuals clustered closer together, with individuals of shared geographic origin grouped together. The first PC explained nearly 40% of the variance and largely separate MA individuals from the rest of the populations; the second PC explained only 7.7% of the variance and primarily separated NJ individuals from the remaining individuals. The DE1, MI1, and WI1 populations also appeared to form separated clusters.

We observed that LD (r²) decayed quickly in the wild cranberry germplasm. The average LD half-life per chromosome was 23 kb with a range of 10–37 kb, and on average, LD reached r² = 0.2 after 17 kb. Genetic diversity was elevated within populations relative to among populations. The average pairwise genetic distance within populations (of more than 2 individuals) was 0.43 and ranged from 0.34 (WI1, n = 4) to 0.49 (WV1, n = 8; Table 1). Populations from NJ and MA had the greatest range in genetic distances (NJ1: 0.19–0.51; MA1: 0.24–0.54), in agreement with the higher separation visualized by the PCA (Fig. 2a). The average among-population pairwise distance was 0.50 and ranged from 0.47 (MI1 vs WI1) to 0.53 (NJ1 vs MA1). The overall observed heterozygosity was 0.28, within-population diversity (H_S) was 0.31, and among-population diversity (D_ST) was 0.047.

Table 1.

Wild cranberry sampling locations, population sizes, and average within-population genetic distance (with range in parentheses).

Location	State	Abbreviation	N Ind a	Mean genetic distance (minimum, maximum)
Cranberry Mountain	WV	WV1	8	0.488 (0.416, 0.515)
Susquehanna	PA	PA1	6	0.486 (0.472, 0.500)
St Charles	NB	NB1	3	0.468 (0.463, 0.475)
Pilgrim Lake	MA	MA1	35	0.458 (0.239, 0.541)
Fire Island West	NY	NY2	4	0.437 (0.365, 0.488)
Cranberry Lake	MI	MI1	9	0.436 (0.321, 0.492)
Island Beach State Park	NJ	NJ1	14	0.433 (0.186, 0.514)
Sandy Neck	MA	MA2	4	0.419 (0.352, 0.454)
Amagansett	NY	NY3	3	0.414 (0.391, 0.444)
Lewes	DE	DE1	8	0.400 (0.352, 0.439)
Flanders	NY	NY1	6	0.383 (0.326, 0.435)
Oneida County	WI	WI1	4	0.342 (0.303, 0.360)

^a N _Ind, Number of individuals per population.

Environmental association and candidate loci evaluation

We detected 217 marker–environment associations between 104 markers and 34 environmental variables (Fig. 3, Supplementary Table 3). The number of significant markers (FDR threshold of 0.20) per variable ranged from 1 to 28, and the number of variables associated with a given significant marker ranged from 1 to 14. We identified 107 associations with 16 soil variables, 14 with 5 temperature variables, 50 with 5 precipitation variables, and 18 with 2 geographic variables (elevation, latitude, and longitude). The frequency of the minor allele at significant EAA markers in all wild individuals (mean: 0.21, SD: 0.12) was significantly lower than the genome-wide average (mean: 0.26, SD: 0.12; P-value <1 × 10⁻⁵, Mann–Whitney U-test). Using genotypes from the same SNP markers profiled in “native selections” (i.e. landraces; n = 81) and breeding selections (n = 108; Diaz-Garcia et al. 2020), we compared allele frequencies of our significant EAA SNPs across populations representing different levels of genetic improvement. We found that for 76 (73%) of the EAA SNPs, the minor allele in our wild cranberry population was at elevated frequency in the native selection or breeding selection germplasm (Supplementary Table 3).

Fig. 3.

Chromosomal location of markers detected as significant in the EAA or as outliers in F_ST or spatial ancestry (SPA) analyses. EAA SNPs are separated by the class of environmental variable (Geo., geography; Prec., precipitation; Temp., temperature). SNPs detected for more than one variable or analyses, or in close proximity, are annotated.

A handful of significant associations in the EAA are worth highlighting. Two SNPs, one on chromosome 4 and one on chromosome 5 were significantly associated (or nearly so) with multiple temperature-related environmental variables (Figs. 3 and 4), including those related to average or cold temperature (i.e. annual mean temperature, mean temperature of coldest quarter, and minimum temperature of coldest month) and temperature ranges (i.e. temperature seasonality and annual temperature range). The marker on chromosome 4 (“S04_24963382”) mapped within an intron of the predicted gene Vmac_021589, the Arabidopsis homolog of which is ALA3 (Fig. 4b). The SNP on chromosome 5 (“S05_24981052”) also mapped within a predicted gene (Vmac_025878) with the Arabidopsis homolog MOS14 (Fig. 4b). The minor alleles at both SNPs were found at higher frequency in the WI1 and NB1 populations, but remained at much lower frequency in nearly all others (Fig. 4, c and e). Individuals with this minor allele were native to locations that were, on average, colder and more variable in temperature (Fig. 4, d and e).

Fig. 4.

Identifying 2 candidate loci for cold temperature adaptation. a) Two SNPs in the EAA were consistently associated with bioclimatic variables related to temperature. b) Both SNPs were located within predicted cranberry genes; the Arabidopsis homolog of the predicted gene on chromosome 4 was ALA3, while that on chromosome 5 was MOS14, both known to be involved in chilling sensitivity. c, e) The minor alleles at both SNPs were generally at higher frequency in northern (WI1 and NB1) populations. d, f) Individuals (points) with these minor alleles were collected from location with colder and more variable temperatures.

An SNP on chromosome 3 (“S03_4582157”) was associated with multiple precipitation and soil variables, including precipitation of the driest month, driest quarter, and coldest quarter, and both topsoil and subsoil pH (Figs. 3 and 5, Supplementary Table 3). This SNP was found within the annotated gene Vmac_014687 (Fig. 5b), the highest matching Arabidopsis homolog of which was At3g51120 (48% similarity), putatively encoding a DNA-binding protein that may be involved in hormone signaling and environmental response. The minor allele at this marker was found at elevated frequencies primarily in the MI, PA, and WI populations and coincided with drier conditions and more alkaline soils. Another strong marker association with soil variables was found on chromosome 2 (“S02_38027635”) and correlated with soil texture (i.e. silt, sand, and clay content; Fig. 3, Supplementary Table 3). Individuals with the minor allele at this marker were found in locations with reduced sand content (−22.4 percentage points) and elevated silt (13.6 percentage points) and clay (8.83 percentage points). This marker was found within 1.7 kb of the predicted gene Vmac_013899, the Arabidopsis homolog of which was PIP2-8, an aquaporin gene responsible for transmembrane water transport.

Fig. 5.

Identifying a candidate locus for water–soil quality adaptation. a) A significant SNP in the EAA was detected for reduced precipitation and soil pH. b) This SNP was located within a predicted gene, and c) the minor allele (orange) was enriched in inland populations (PA1, MI1, WI1). d) Individuals (points) with this allele were found in drier locations with more alkaline soils.

F _ST and spatial ancestry analysis

F _ST and SPA were used to assess more extreme allele frequency differentiation among populations. The genome-wide F_ST among all individuals was 0.14, and the range in among-population F_ST scores on a per-SNP basis was 0–0.81. Individual locus SPA scores ranged from 0.0033 to 2.9, with a median of 0.44. We found no overlap between the 5 SNPs with outlier F_ST scores (F_ST > 0.60) and the 5 SNPs with outlier SPA scores (SPA > 2.2; Fig. 3, Supplementary Table 3).

Significant SNPs from the EAA often exhibited elevated or outlier F_ST scores. The average F_ST score of significant EAA SNPs was 0.32 (SD = 0.11) compared to a genome-wide average of 0.13 (SD = 0.095), a statistically significant difference (Mann–Whitney U-test, P-value < 1 × 10⁻¹⁵). Two of the 5 F_ST outlier SNPs overlapped with significant EAA SNPs (Fig. 3, Supplementary Table 3).

SPA outlier SNPs primarily separated the populations by geographic coordinates. The 5 outlier SNPs (Supplementary Table 3) displayed allele frequency gradients that separated the NY, MA, ME, and NB populations from the remaining populations (Supplementary Fig. 5). All SPA outlier SNPs were found to be in proximity of predicted genes (<6.5 kb); however, not one of these had a high-matching Arabidopsis homolog (Supplementary Table 3). One SPA outlier SNP (“S03_17446409”) was significant in the EAA, but was only associated with elevation and longitude (Supplementary Fig. 5). Overall, EAA SNPs displayed elevated SPA scores (mean = 0.80, SD = 0.58) compared to the genome-wide average (mean = 0.55, SD = 0.39), a significant difference (Mann–Whitney U-test, P-value = 1 × 10⁻⁵).

Discussion

Crop wild relatives, including those of cranberry, may be sources of novel and useful genetic variation for plant breeding programs. Assessing the genetic potential of unimproved germplasm through phenotypic screening can be time-consuming, especially for long-lived perennials; therefore, an understanding of the genome-wide footprints of adaptation to local environmental conditions can aid in more quickly identifying promising germplasm for crop improvement. Here, we build on previous assessments of diversity in wild cranberry (Bruederle et al. 1996; Rodríguez-Bonilla et al. 2019, 2020) by conducting a genome-wide scan for signatures of selection for local adaptation, providing useful information for conserving (Khoury et al. 2020) and exploiting the diversity of wild cranberry for future cultivar development.

Genetic diversity and LD

Genetic diversity was moderate to high in our wild cranberry germplasm. Compared to other outcrossing or partially selfing perennial crops, within-population diversity (or expected heterozygosity) in cranberry was similar, though slightly lower, than that reported in SNP-based surveys of grape (Vitus vinifera L.; Laucou et al. 2018) and cassava (Manihot esculenta Crantz.; Ferguson et al. 2019). The levels of diversity we observed were elevated relative to other studies of wild cranberry populations using allozyme markers (Bruederle et al. 1996), but reduced compared with recent measurements using SSR markers (Zalapa et al. 2015; Schlautman et al. 2018; Rodríguez-Bonilla et al. 2019, 2020). Given that neutral SNP markers mutate less rapidly than SSRs but more so than highly conserved allozymes, this result is not entirely surprising. Nevertheless, both SSR- and allozyme-based comparisons with other Vaccinium species indicate reduced genetic diversity in V. macrocarpon (Bruederle et al. 1996; Vorsa and Johnson-Cicalese 2012; Rodríguez-Bonilla et al. 2019, 2020). One proposed hypothesis is that V. macrocarpon is a more recent species, possibly undergoing a genetic bottleneck in the Pleistocene epoch, and neutral mutations at conserved allozyme markers—unlike the more quickly mutating neutral SNPs and, especially, SSRs—have been slow to accumulate (Bruederle et al. 1996).

Consistent with previous studies in wild cranberry (Bruederle et al. 1996; Rodríguez-Bonilla et al. 2019, 2020), genetic diversity was greater within populations than among populations. Though this pattern is apparent in other perennial crops, such as lowbush blueberry (Vaccinium augustifolium Ait.; Beers et al. 2019) and cassava (Ferguson et al. 2019), the historical use of cranberry may help to further explain this observation. Cranberry was, and still is, an important part of Native American culture, being valued as a source of food and dye and as a peace symbol by peoples from the present-day eastern United States and Canada to the Great Lakes region (Eck 1990). Cranberry fruit was among the items exchanged through trade among native peoples in these regions (Turner and Von Aderkas 2012), and this movement may help explain the high degree of within population genetic variation compared to that among populations.

LD decayed quickly in the wild cranberry germplasm (average half-life of 23 kb; Fig. 2), consistent with previous analyses in this germplasm (Diaz-Garcia et al. 2020). As a comparison, this rate of LD decay is faster than in southern highbush blueberry (Vaccinium spp.; Ferrão et al. 2018) and intermediate wheatgrass [Thinopyrum intermedium (Host) Barkworth and D.R. Dewey] (Zhang et al. 2016), both obligately outcrossing polyploids, but slower than the outcrossing sunflower (Helianthus annuus L.; Liu and Burke 2006) and partially selfing grape (Myles et al. 2011) and maize (Zea mays L.; Tenaillon et al. 2002; Yu and Buckler 2006). The intermediate levels of LD decay that we observed in cranberry are consistent with its partially selfing sexual and stoloniferous asexual reproductive behaviors (Eck 1990; Roper and Vorsa 1997) and suggest that it should be possible to identify and select for useful variation even in smaller populations.

Marker associations with environmental variables and potential candidate loci

Our EAA used the same mixed-effect modeling framework developed for GWA and identified 217 marker–environment associations (Fig. 3, Supplementary Table 3). This approach, as in other studies (Yoder et al. 2014; Anderson et al. 2016; Bandillo et al. 2017; Lei et al. 2019), included covariates for kinship, thereby reducing the likelihood of detecting SNPs simply associated with background genetic relatedness (Yoder et al. 2014). Significant markers were identified on all 12 chromosomes and were detected for 34 of the 42 geographic, precipitation, soil, and temperature variables considered, along all 6 PCs computed from this set (Supplementary Table 3). The large number of variables associated with marker loci may indicate that local adaptation is being driven by multiple dimensions of climate and soil. Furthermore, many markers (42 of 104) were associated with 2 or more variables, and while much of these were likely due to highly correlated environmental variables (Supplementary Fig. 3), it is possible that some genetic loci are mediating adaptation to multiple different climate and soil conditions.

Patterns of allele frequency and overlap with markers identified in the outlier tests further suggest a role of local adaptation along climate or soil gradients. The overlap between SPA/F_ST outlier markers and those in the EAA (Supplementary Table 3), along with a significant elevation in SPA/F_ST scores of EAA marker relative to the genome-wide average, indicates that allele frequency differentiation among the populations may be strongly driven by specific environmental gradients. This provides genomic evidence to support previous suggestions that phenotypic variation of wild cranberry is influenced, in part, by environmental heterogeneity (Vander Kloet 1983; Bruederle et al. 1996; Vorsa and Zalapa 2019). Interestingly, we observed no overlap between SPA and F_ST outliers, both measures of allele frequency differentiation. This observation may be due to potentially different signals of adaptation detected by each method, where F_ST measures overall variation in allele frequencies using discrete populations and SPA measures allele frequency gradients along continuous geographic space. Previous studies with much larger sample sizes have found overlap and a positive correlation between F_ST and SPA (Anderson et al. 2016; Bandillo et al. 2017), and it is possible that the lack of overlap in our study may thus be due to reduced sampling of genetic and geographic space. Nevertheless, these signals of differentiation do not always imply adaptation, but provide key target loci for further exploration.

The colocation of significant markers in the EAA with predicted genes, with some particularly notable examples, helped to increase the confidence that true associations were detected. Nearly all (100 of 104) significant markers were located near (<17 kb) a predicted gene, and 27 (26%) were located within a gene (Supplementary Table 3). Several of the annotated genes near significant markers had Arabidopsis homologs that implicated a role in environmental adaptation. For example, the locus on chromosome 4 associated with several cold temperature-related variables (Fig. 4) was located within a predicted cranberry gene with high similarity to ALA3 in Arabidopsis. This gene has a well-documented role in adaptability to temperature stresses, specifically chilling tolerance (Poulsen et al. 2008; McDowell et al. 2013). Furthermore, another SNP on chromosome 5 associated with cold temperature mapped to a predicted gene with high similarity to MOS14 in Arabidopsis. This gene has a more direct role in immune response (Xu et al. 2011) and interacts with another Arabidopsis gene with a role in freezing stress tolerance (Xu et al. 2016). On chromosome 3, a marker associated with different temperature, soil, and precipitation variables (Fig. 5) was also located within a predicted gene; the most similar Arabidopsis homolog of this gene (At3g51120) contains a nucleotide-binding domain and is putatively involved in hormone signaling and environmental response (Wang et al. 2008), suggesting a transcriptional regulation role that may explain the association with many different environmental variables. Several other significant markers tagged genes in cranberry with putative Arabidopsis homologs implicated in the response to abiotic and biotic conditions (Supplementary Table 3), both of which are likely impacted by climate or soil.

Applications for plant breeding

Crop wild relatives have long been exploited by breeders as a source of favorable genetic variation for different traits (Hajjar and Hodgkin 2007). In addition, since the advent of genetic mapping, populations of wild or landrace germplasm or those generated from wild × elite crosses have been used to identify, among other traits, loci conferring potential abiotic stress tolerance (e.g. Hartman et al. 2014; Honsdorf et al. 2014; Arms et al. 2015). Environmental association analyses may serve as another tool to identify novel variation in exotic germplasm that may be useful for crop improvement programs (Anderson et al. 2016; Bandillo et al. 2017; Lei et al. 2019).

The domestication history of cranberry suggests that loci identified from wild germplasm using EAA may be of more immediate utility for breeding than in other crops. Few generations separate modern cranberry cultivars from unimproved germplasm (Vorsa and Zalapa 2019). While this short improvement history has nevertheless led to notable phenotypic changes, genetic evidence suggests only modest divergence of cultivars and breeding selections from wild germplasm (Diaz-Garcia et al. 2020). As measured by expected heterozygosity in our plant material, cranberry breeding germplasm has retained about 88% of the genetic diversity found in the wild populations (data not shown). Indeed, many of the putatively favorable alleles at loci we identified using EAA were already segregating at moderate frequency in cranberry breeding germplasm (Supplementary Table 3). This is consistent with the pattern in other perennial crops, where improved germplasm retains a high proportion of the genetic diversity found in their wild progenitors (Gross et al. 2014). Given this preservation of genetic variation, it may be possible to use the results of EAA analogous to traditional QTL mapping or GWA: marker–environment associations tagging putative abiotic stress tolerance loci may be discovered in natural populations, and validation and marker-assisted selection can be conducted in more improved populations. As in traditional marker-assisted selection, the benefits of this approach may be more likely realized in the case of large-effect genes and minimal genetic background effects. Yet, this approach may be broadly appealing for perennial crop breeding by avoiding the need for repeated cycles of backcrossing (and the associated risks of excess homozygosity and inbreeding) required to introgress exotic variation into elite genetic backgrounds. Additional research is required to examine the feasibility of this approach.

Successful utilization of putative abiotic stress tolerance loci could address several environmental-related limitations to cranberry cultivation. Depending on the region, abiotic stresses such as heat, frost, soil pH, and water availability pose serious challenges for cranberry production (Eck 1990; Vorsa and Zalapa 2019). While breeding stress-tolerant cultivars is a more durable solution, it is extremely time-consuming due to the lengthy juvenile period and the time required to establish and evaluate germplasm under divergent field conditions or controlled environments. Early-stage screening of unimproved germplasm or progeny from wild × elite crosses could reduce the breeding cycle length, and our analysis identified several markers potentially related to abiotic stress tolerance that could be applied in a breeding program. The improvement in genomic resources for cranberry and related wild species (Diaz-Garcia et al. 2021; Kawash et al. 2021) will be instrumental in this effort.

Limitations of our analysis

Though EAA was a useful approach for identifying signals of local adaptation in our collection of wild cranberry, there are limitations to our analysis that are worth discussing. First, the spatial quality of the sampled wild cranberry populations is reduced, due in part to both imprecise geographic coordinates and limited sampling of the native range (Fig. 1). As a result of this imprecision, we obtained climate and soil data at a course resolution, which introduces error in the association analysis. Second, along with limited geographic sampling, the size of our population is low compared to similar studies (Yoder et al. 2014; Anderson et al. 2016; Lei et al. 2019), reducing power and inflating association effect sizes. This limitation suggests that we are only able to detect those variants with large effect and at moderate frequency, missing much of the small-effect alleles that are likely underlying much of genetic architecture of local adaptation. Third, while we detected several associations using our 4,457-SNP marker set, the rapid decay of LD in the germplasm (Fig. 2b) suggests that additional loci may have gone undetected due to insufficient marker density. Given the encouraging results of our study, larger and more geographically continuous sampling of wild cranberry, along with higher marker density, should enable the detection of additional genetic loci with potential influence on environmental adaptation.

Data availability

The input georeferenced population metadata and marker genotypes, as well as supplemental figures and tables, are available from the figshare repository: https://doi.org/10.25387/g3.20067062. Scripts used to replicate the analyses and figures in this paper are available from the GitHub repository https://github.com/neyhartj/CranberryGermplasmEAA.