Genotypic Characterization of the U.S. Peanut Core Collection

Cultivated peanut (Arachis hypogaea) is an important oil, food, and feed crop worldwide. The USDA peanut germplasm collection currently contains 8,982 accessions. In the 1990s, 812 accessions were selected as a core collection on the basis of phenotype and country of origin. The present study reports genotyping results for the entire available core collection. Each accession was genotyped with the Arachis_Axiom2 SNP array, yielding 14,430 high-quality, informative SNPs across the collection. Additionally, a subset of 253 accessions was replicated, using between two and five seeds per accession, to assess heterogeneity within these accessions. The genotypic diversity of the core is mostly captured in five genotypic clusters, which have some correspondence with botanical variety and market type. There is little genetic clustering by country of origin, reflecting peanut’s rapid global dispersion in the 18th and 19th centuries. A genetic cluster associated with the hypogaea/aequatoriana/peruviana varieties, with accessions coming primarily from Bolivia, Peru, and Ecuador, is consistent with these having been the earliest landraces. The genetics, phenotypic characteristics, and biogeography are all consistent with previous reports of tetraploid peanut originating in Southeast Bolivia. Analysis of the genotype data indicates an early genetic radiation, followed by regional distribution of major genetic classes through South America, and then a global dissemination that retains much of the early genetic diversity in peanut. Comparison of the genotypic data relative to alleles from the diploid progenitors also indicates that subgenome exchanges, both large and small, have been major contributors to the genetic diversity in peanut.

Peanut is now an important source of protein and oil worldwide. In 2017, the 718,570 hectares in the U.S. produced 47,097,498 metric tons; and worldwide, 28 million hectares produced 47 million metric tons (https://www.nass.usda.gov). As a nitrogen-fixing legume, peanut is also important as a rotation crop that restores soil nitrogen.
The USDA peanut germplasm collection provides an essential source of diverse genetic material for breeders. The collection, representing peanut landraces and introductions from around the world, and most of the 80 diploid Arachis wild relatives, currently contains 8,982 accessions, which are maintained by the USDA Plant Genetic Resources Conservation Unit in Griffin, GA. As a recent polyploid that experienced a domestication bottleneck, genetic variation across peanut landraces is expected to be low. This lack of diversity presents breeding challenges, particularly as peanut is susceptible to a wide range of pathogens; breeding for disease resistance is of paramount importance. Other traits are important breeding targets, including agronomic traits such as time to maturity and pod-fill, and nutritional and market traits such as seed size, oil quality, and flavor.
The U.S. Peanut Core Collection was developed using geographic origin and phenotypic characteristics to select a representative set of accessions from the US collection that span the diversity of cultivated peanut (Holbrook et al. 1993). The development of the Affymetrix SNP array, 'Axiom_Arachis2' (Clevenger et al. 2018;Korani et al. 2019) enabled low-cost analysis of this core set through genotyping. The resulting data set will serve multiple purposes: to assess the genetic diversity of the core collection and its population structure; to provide breeders with genotype data for each accession; and to generate data that can be used for trait association (GWAS) analyses. In addition to these expected outcomes, investigation of the phylogenetic and network characteristics of the collection provide information about the historical spread of peanut diversity globally.
The US Peanut Core Collection was further reduced to create the US Peanut Mini Core Collection in 2005 (Holbrook and Dong 2005). Although the Mini Core has been phenotyped and genotyped in earlier studies (Otyama et al. 2019;Wang et al. 2011;Belamkar et al. 2011), studies on the entire core collection have been limited to phenotype studies, mainly focused on disease resistance and seed content (Holbrook and Anderson 1995;Anderson et al. 1996;Hammond et al., 1997;Franke et al. 1999;Holbrook et al., 2009;Damicone et al. 2010). Here we report on the first comprehensive genotype study of the US Peanut Core Collection.
The specific objectives of this study were to 1) provide genotype data for each accession, 2) assess genetic diversity of the collection, 3) analyze population structure, 4) estimate the incidence of heterogenous or mixed accessions, and 5) assess relationships between genotypic groups and common traits and phenotypic classes.
Each accession was grown to maturity to enable seed collection. The accessions which originated from Africa were grown by the Ozias-Akins lab in Tifton, GA. The remaining accessions were grown by the Chamberlin lab in Stillwater, OK. Additionally, we selected 253 accessions for replicate genotyping to test accession purity. These were grown to seedling stage in Ames, IA. Of the 253 accessions, 35 were selected to test suspected heterogeneity/mixture in the accession, based on information from Dezern (2018), GRIN-Global (https://www.gringlobal.org), and previous evidence of heterogeneity (Otyama et al. 2019). The remaining 218 were randomly selected to evaluate overall homogeneity of the core collection (Supplementary File S1).
For replicate genotyping, seeds were randomly picked from a seed packet of 30 seeds per selected accession. These were then planted in the greenhouse, on a sand bench, or in a growth chamber. Not all selected samples germinated (even after replanting), which limited the number of samples available for replicate genotyping for some accessions. Of the 253 accessions; 193 accessions were genotyped twice, 43 were genotyped three times, 16 accessions had four samples and one had five samples genotyped. In total, 1145 samples were available for genotyping.

DNA extraction and genotyping
For all accessions, whether grown to maturity or to seedling stage, leaf tissue was sampled between 2 and 4 weeks after germination and immediately frozen in liquid nitrogen. DNA was extracted using Qiagen (Germantown, MD) DNeasy 96 Plant Kits (#69181) and 3 mm Tungsten Carbide Beads (#69997) as recommended by the manufacturer. Initial concentration and purity of 12 DNA samples/ plate was estimated using a Thermo Fisher Scientific NanoDrop ND-1000 Spectrophotometer (Thermo Fisher Scientific, Waltham, MA, USA). Concentrations ranged from 26 to 75 ng/ul, with an average 43 ng/ul. A260/A280 ratios ranged from 1.882 to 1.984, with an average ratio of 1.931. A260/A230 ranged from 1.84 to 2.681, with an average ratio of 2.206. Samples were then shipped to Thermo Fisher for additional quality control and genotyping. DNA concentration and quality for all samples was confirmed using a 'PicoGreen' assay. Average DNA concentration was about 47 ng/mL for 926 highquality samples. The remaining 219 samples had a concentration of 15 ng/ul and were considered of sufficient quality and quantity for genotyping. Samples were then genotyped using the 48k Thermo Fisher 'Axiom_arachis2' SNP array. Of the 1,145 samples, 25 replicate samples were not successfully genotyped.
Raw SNP intensities from Affymetrix were analyzed using the 'Best Practice Workflow' available in the Axiom Analysis Suite. A total of 47,837 SNPs was obtained, of which 14,430 were categorized as 'Poly High Resolution', 15,528 were 'Mono High Resolution', 11,008 were 'No Minor Homozygote', and the remaining 6,871 were of low quality. Poly High Resolution SNPs were processed into a standard VCF format (Supplementary Files S2 and S3) using custom bash scripts for downstream analyses (https://github.com/ cannongroup/peanut_core_collection_genotyping).
Diversity, phylogenetic, and network analysis Several aspects of diversity analysis were carried out on variant data in FASTA format -i.e., with SNP variants represented as DNA bases, positioned in the genomic order of the loci. A FASTA-format sequence representation of the 'Axiom_Arachis2' SNP array variant data were generated by converting genotype calls in the array to DNA base calls from the Axiom_Arachis2 VCF file generated by Thermo-Fisher, using custom shell scripts that converted AA/BB calls to A, T, C, G, or "-" (scripts are available at https://github.com/cannongroup/ peanut_core_collection_genotyping). SNPs from Affymetrix were called using Best Practices Genotyping Analysis Workflow algorithm on Axiom analysis Suite. This classifies SNPs into following categories: MonoHighResilution, PolyHighResolution, NoMinorHom, Other, Call-RateBelowThreshold and OTV. SNPs from only the PolyHighResolution category were selected and these were the high-confidence SNPs. The matrix contains the 14,430 high-confidence SNPs, for 1,120 samples. Relative positions of the SNPs were also determined from the consensus genomic locations from five Arachis genome assemblies, as described below. This sequence representation is available as Supplementary Files S4 and S5.
Base-calls were also derived computationally for four sequenced Arachis genomes: A. duranensis, A. ipaënsis (Bertioli et al. 2016) and A. hypogaea varieties Tifrunner (Bertioli et al. 2019), Shitouqi (Zhuang et al. 2019), and Fuhuasheng . Base-calls from the genomic sequences were made by aligning flanking sequences plus the variant base, using two sequences per variant per locus, to the respective genome, using blastn (Altschul et al. 1990). Per-locus SNPs were called when the flanking+variant sequence matched at 100%, over at least 65 of 71 bases, to only one location in the genome (i.e., full-length alignments were not required, but perfect match was required within the alignment).
The genome-derived SNPs were added to a version of the sequence variant-call file (Supplementary Files S4 and S5) with the A. duranensis and A. ipaënsis calls combined into one "synthetictetraploid" accession. Base calls that were absent in that accession were removed from the merged file, giving an alignment 10,278 bases wide, by 1,123 samples (after removal of sample PI493562_1, which appears to have had a label tracking error). Approximate genomic locations of SNPs were determined as: the location in the respective diploid chromosomes were present; otherwise, the location in Tifrunner; otherwise in Shitouqi; otherwise the location in Fuhuasheng, as shown in Supplementary File S6. Two reduced alignments were also generated (Supplementary Files S7 and S8), consisting of representative "centroid" sequences from clusters at the 98% and 99% identity levels, using the cluster_fast method in the vs.earch suite, version 2.4.3 (Rognes et al. 2016).
The phylogenetic tree in Figure 1 and Supplementary File S9 was calculated based on maximum likelihood, using FastTreeMP, version 2.1.8 (Price et al. 2010), with default parameters. The network diagram in Figure 2 was calculated on the 99%-identity centroid alignment, using the Neighbor-Net algorithm in the SplitsTree package, version 4.15.1 (Huson and Bryant 2006).

Replicate analysis
To assess the genetic similarity among multiple samples from an accession, a list of all possible pairs of replicates per accession was calculated, giving "N choose 2" combinations for an accession with N samples: 3 combinations for an accession with 3 samples; 6 combinations for an accession with 4 samples, etc. For each possible combination, the sequence identity was calculated between the sequence pairs (using blastn), and then scored as "similar" if .= 98% identity and "dissimilar" otherwise. These results are shown in the "rep analysis" worksheet of Supplementary File S1.

Structure and Principal Component Analysis (PCA)
To define subpopulations based on genomic sequences, a structure analysis and PCA was performed on high confidence Axiom_Arachis2 SNP array variant data. Structure analysis was performed using a Bayesian inference algorithm implemented in fastSTRUCTURE (Raj et al. 2014).
All 13,410 SNP sequences were used for a representative set of 518 "unique" accessions, selected based on sequence identity at 98%. Clusters and group membership were determined for arbitrary groups ranging from K 1 to 10 with settings: -prior = logistic,-cv = 0,-tol = 10e-6, default otherwise. Structure was visualized as proportionally colored bar plots representing global ancestry estimates (Q values) using an R package, Pophelper version 2.3 (Francis 2017).
To avoid the strong influence of SNP clusters in principal component analysis (PCA) and relatedness analysis, only SNPs in approximate linkage equilibrium with each other (r 2 = 0.2) were used. The R package, SNPRelate (Zheng et al. 2012), was used for LD pruning on 1120 samples. snpgdsLDpruning in the SNPRelate package, was used to recursively remove biallelic SNPs in LD within a sliding window of 1Mb. LD threshold was specified at r 2 = 0.2. Monomorphic SNPs were also removed along with uncommon SNPs filtered at MAF , 5% leaving a final set of 2,063 markers in approximate linkage equilibrium with each other.
PCA was performed using snpgdsPCA from the SNPRelate package at default settings and plotted using ggplot2 for defined groups. Groups were defined according to: whether or not they flowered on the main stem, their botanical variety defined in GRIN-Global, agronomic type (market group), growth form, pod type, and country from which seed was originally collected.

Population differentiation analysis
To evaluate differentiation between and among accession groups, we calculated F ST for selected accession groups defined as above under the Structure and PCA methods section. SNPs were first pruned to reduce SNPs in strong LD with one another, as described above. The F ST analysis was performed using the R package Hierfstat, (Goudet 2005) at default settings. Pairwise F ST values were calculated using pairwise.WCfst according to (Weir and Cockerham 1984). A heatmap of pairwise F ST values was plotted using ggcorrplot (Kassambara 2016), for defined groups.

Geographical distribution
A plot of the geographical distribution of peanut accessions by clade ( Figure 6) was generated using the germplasm Geographical Information System (GIS) utility at PeanutBase.org (Dash et al. 2016), with the "add your data" tool. To display the five germplasm categories identified in Figures 1 and 2, we used the following column labels, which are interpreted by the GIS tool: accession_id, trait_observation_value, trait_descriptor, taxon, trait_is_nominal.

Analysis of subgenome invasions
To track possible instances of subgenome interactions, 16 accessions were selected from across the clades identified in Figures 1 and 2 and alleles were examined relative to those identified in the diploid accessions (Supplementary Files S6 and S13). Alleles for each accession were then marked as being the same as the A-genome allele and not the B-genome allele (A-like), or same as the B-genome allele and not the A-genome allele (B-like), or other conditions (invariant in the diploids, different from both diploids, or missing in one or more of the tetraploid or diploid accessions). The results are shown in Figure  Figure 1 Phylogenetic tree for 1122 samples from 791 accessions of the U.S. peanut core collection. For reference, five clades have been assigned (1-4 and a transitional group, 3.2). These clade designations are also used in the network plot ( Figure 2) and in the PCA analysis ( Figure 4). Several accessions and samples of particular interest are highlighted: blue squares indicating the three sequenced tetraploid accessions and duranensi-s+ipaensis (SNP states derived from genome sequence); and red-lined green squares indicating five samples from the heterogeneous accession PI 442786. 7, with red indicating identity with the respective subgenome (A-like for chromosomes 1-10, and B-like for chromosomes 11-20).

Replicate analysis
For the 253 accessions with replicates, a maximum of 428 pairings from same-accession groupings were expected. For example, an accession with one replicate (A and B) has one expected pairing (A-B), while an accession with two replicates (A, B, C) has three expected pairings (A-B, A-C, B-C), and an accession with three replicates has six expected pairings. A missed pairing means that one or more samples for an accession are genetic outliers, and that the accession is not homogeneous. Among the 253 accessions with replicates, 218 accessions genotyped in replicate were selected at random from the core, and 35 accessions were selected as being potentially mixed or in which the seeds appeared to be visibly heterogeneous, based on assessment in Dezern (2018), GRIN-Global (https://www.grin-global.org), and previous evidence of heterogeneity (Otyama et al. 2019).
Of the 428 expected pairings among replicates (with .70% sequence identity across all SNP locations), 368 pairings were observed (86%). The observed pairings had an average identity of 94.4% and a median of 98.7%. The 60 instances of a sample that did not match to a replicate for that accession occurred among 42 accessions, meaning that some accessions had more than one "missing" match for a replicate.
Of the 35 accessions selected as "probably mixed" based on seed color or identified in Dezern (2018), most (27/35 = 77%) were indeed mixed genotypically, with at least one sample per accession being unlike the others at the 98% identity threshold. Eight of these accessions had all of the replicates close to identical (.=98%) across all replicates, despite being identified as "probably mixed" based on seed phenotype. We speculate that this apparent discrepancy might occur for two primary reasons: first, the assessment of whether an accession is mixed is subjective (is seed coat color different or not?); and second, our sampling from each accession was random and limited (from two to four replicates per accession), and may not have captured variation when it existed.
We describe one of the "probably mixed" accessions in some detail, as it appears to indicate substantial diversity within a single line. For accession PI 442786 (from Zimbabwe), five individuals were selected for genotyping, based on observed variation in seed phenotype. Seeds varied in both shape and color: darker or lighter red, and small/round or elongated. Sequence identity among these five replicates is at 87% or below, and the replicates appear in several intermediate positions in the phylogenetic tree and the network plot (Figures 1 and 3 respectively). In Figure 1, the samples from this accession appear basally in both clade 4 and clade 3.2 (leaf nodes indicated with red-bordered green squares) suggesting hybridization between lines from different clades. Similarly, in the network plot ( Figure 2), these five PI 442786 samples appear basally in clade 4 and also intermediate between clades 3 and 4.
Of the 218 accessions selected at random for replicate genotyping, most accessions (124; 56%) were NOT mixed genotypically: in these accessions, all replicates were close to identical (.=98%) across all replicates. Nevertheless, the high rate of apparent genotypic heterogeneity in accessions suggests that the core collection will require further subdivisions or selections to generate material that is well suited for analyses such as QTL and GWAS.

Diversity analysis: phylogenetic analysis
The core collection contains considerable phenotypic diversity, but also displays high genotypic similarity among many accessions, as apparent in Figure 1, where many accessions are near identical in the phylogeny. The 1,122 samples (791 accessions) in this study fall into 671 clusters at an identity threshold of 99% (Supplementary File S1, worksheet "clusters"). The largest clusters at 99% identity have 139, 49, 27, and 25 samples (112, 42, 21, and 22 distinct accessions), and the cluster sizes fall progressively to the singletons, of which there are 560. The existence of large clusters of nearly identical accessions suggests that diversity in the core could be represented by a smaller number of accessions (671, specifically, if 99% identity were used as the identity cutoff).
The phylogenetic tree of accession diversity shows four primary clades of accessions, numbered 1-4 in Figure 1, with an intermediate group (3.2) also indicated. These clade numbers are also used in the network diagram Figure 2. Although some accessions occur on early branches in these clades (rather than nested tightly in terminal clusters), the clades are nevertheless mostly distinct in both the phylogeny and the network plot. The clade designations also generally correspond with the Structure plot at cluster-number K = 5 Figure 3. The Structure plot is ordered by the tree order from Figure 1.
A top-level summary of the cluster and trait correspondences demonstrates that most accessions, including all named cultivars, fall into three large clades (1, 2, and 3), but those clades don't correspond cleanly with typical peanut classifications (e.g., growth habit, botanical variety, market type, or pod type). Traits categories are shown superimposed on the clades, in the PCA plots in Figure 4. A smaller clade (4) does correspond with these typical classification traits (Figures 1 and 4). Clade 4 has exclusively erect growth habit, with pod-types of hypogaea, valencia, or mixed pods, but frequently having strong, linear reticulation, and including the aequatoriana botanical variety of subspecies fastigiata, as exemplified by PI 497426 from this clade (Supplementary File S14).
For each cluster, counts and proportions of phenotypic characters and collection region are given in supplementary file S1, worksheet "clade summary". The clusters have some correspondence with growth-habit traits and with countries of seed origin, as described below. (In this section, all counts are given per accession rather than per sample, as some accessions were genotyped multiple times).
Next, we describe characteristics and compositions of the clades, working from the bottom of the tree (Clade 4) to the top (Clade 1).
Clade 4 (Figures 1 and 4; at the bottom in Figure 1; 104 samples, 84 accessions) is the most distinctive and consistent phenotypically: most accessions (68.4%) have upright growth habit, per Holbrook's phenotype evaluations (Holbrook and Dong 2005). The pod type is more varied, with accessions scored as hypogaea, fastigiata, or mixed (36.8, 31.6%, 31.6%) (Holbrook and Dong 2005). Growth type was scored as fastigiata for seven accessions and two as aequatoriana. The aequatoriana type is a botanical variety of the subspecies fastigiata (Krapovickas et al. 2007). Pod images from GRIN-Global for this clade show pods frequently having strong reticulation and widelyspaced veins running the length of the pod (Supplementary Figure  S14) -which is of interest as these characteristics are seen in precolonial archaeological finds in Peru, Chile, and Argentina (Supplementary Figure S15). Most of the cluster 4 accessions originate from west-central South America (Figure 6), primarily from Bolivia, Peru, Ecuador, and Argentina (38, 17, and 9, and 8 accessions, respectively). Interestingly, the inferred genotype for A. duranensis and A. ipaensis (consisting of alleles at loci corresponding with the marker flanking sequences from the SNP array) also falls solidly within cluster 4, with 100% bootstrap support on several subtending branches in this clade.
Clade 3 (Figures 1 and 4; third from bottom in Figure 1; 275 samples, 215 accessions) shows general phenotypic consistency: most accessions have the fastigiata botanical variety, upright growth habit, and fastigiata pod type (92.6%, 66.7%, 89.5%, respectively). Both characteristics distinguish this group from Cluster 4. The most frequent South American accession origins for Cluster 3 are Bolivia, Argentina, and Brazil (40, 5, 5, respectively), with one each from Peru and Ecuador. The most frequent non-South American countries for cluster 3 are Zambia, Nigeria, and Zimbabwe (12, 6, and 6, respectively).
Clade 2 (Figures 1 and 4; second from top in Figure 1; 291 samples, 216 accessions). In this clade, most accessions have the fastigiata botanical variety, fastigiata growth habit, and fastigiata pod type (83.3%, 87.1%, 83.3%, respectively). The Clade 2 accessions also have the widest geographic spread. Also cosmopolitan in terms of country of origin. The most frequent South American countries for these accessions are Brazil, Argentina, Cuba, and Uruguay (10, 9, 6, 5, 5, respectively). Non-South American countries are the predominant sources for these accessions, however; Zambia, Zimbabwe, India, and Sudan are the most frequent sources (34, 17, 13, 13, 13, respectively). Because the highest-frequency countries of origin are Brazil in South America and Zambia, Zimbabwe and Sudan in Africa suggests early movement of this germplasm through the slave and other colonial trade.

The geographic distribution of genotypes
All parts of the phylogenetic tree are dominated by accessions from South America, but all clades also have interspersed accessions from many parts of the world (Table 1). This pattern of broad geographical dispersal, with heavy representation in South America, confirms that peanut had fully diversified into modern cultivar types prior to dispersal through colonial shipping and trade. Influence of the slave and spice trade is suggested by adjacent appearance in the phylogenetic tree of widespread geographical locations. For example, accessions from Portugal are interspersed among accessions from countries in west Africa, south Asia, and the Caribbean and eastern South America (in Clades 1, 2, 3, and 3.2) or Spain and countries in Africa, the Middle East, and Asia (middle of Clade 1).
Clade 4 is much less mixed geographically, coming predominantly from central and western South America ( Figure 6). Peanut's geographic origin (through the initial instance of tetraploidy) has been convincingly established as having occurred in southeastern Bolivia/northwestern Argentina (Bertioli et al. 2016;Bertioli et al. 2019). It is therefore noteworthy that the combined diploid progenitors (A. duranensis and A. ipaensis) fall into the Bolivia-dominated Clade 4. This clade contains hypogaea and fastigiata varieties, including the uncommon aequatoriana variety, which is classified (Krapovickas et al. 2007) as A. hypogaea subsp. fastigiata var. aequatoriana. The aequatoriana variety is generally not widely used Figure 3 Genetic structure of 518 samples selected as representatives at .= 98% sequence identity. Accessions are grouped into five clusters represented by distinct colors: Green = cluster 1, light blue = cluster 2, deep blue = cluster 3, red = cluster 4, and yellow = cluster 5. The X-axis represents accessions ordered according to their positions in the phylogenetic tree analysis (Figure 1). Correspondences with clades in Figure 1 are indicated with colored bars below the accessions. The Y-axis represents proportions of cluster assignment based on Q values from fastSTRUCTURE analysis.
in cultivation outside of the landrace occurrence in these regions in South America. Krapovickas (1995) describes A. hypogaea subsp. hypogaea var. hirsute, A. hypogaea subsp. fastigiata var. peruviana, and A. hypogaea subsp. fastigiata var. aequatoriana as being important in ancient times, and still important locally, being found in Peruvian markets, for example. These highly reticulated pod types are also seen in multiple archaeological sites on the coast of Peru and Chile, and Argentina (Masur et al. 2018;Krapovickas 1995), as well as in early European herbarium specimens. This pod form is depicted in the royal tombs of Sipán, in northern Peru, dating ca. 250 AD, associated with the Moche culture (Krapovickas 1995;Masur et al. 2018). The peanut form in the necklace, sculpted clearly in gold and silver, is identified by Krapovickas (1995) as A. hypogaea, subsp. fastigiata var. peruviana (Supplementary Figure S15).
The identification of southeastern Bolivia, as the center of origin of cultivated peanut relies on several lines of evidence. Both ancestral diploid species A. duranensis and A. ipaensis are found close to Villa Montes, in the Province of Tarija Krapovickas and Gregory 1994). These species are strongly prostrate, lack flowers on the main stem, have dark green leaves, and small, two seeded pods Krapovickas and Gregory 1994). Also in Tarija are found a large number of var. hypogaea landraces including the archetypal primitive cultivated peanut, "Rastrero colorado de dos granos," which combines the most primitive characteristics being, a strongly prostrate variety, with dark green leaves, lacking flowers on the main stem, and most importantly, it has two seeded pods with small seeds Krapovickas and Gregory 1994). This combination of prostrate habit and small seeds is very rare. The sample studied here did not include Rastrero colorado de dos granos, but it is notable that Clade 4 includes other landraces with these Tarija primitive characteristics. These include Sara Maní (PI 468280), from nearby Cochabamba Province, which has pods that are very similar to Rastrero colorado de dos granos, except with a slightly less prominent beak ). Also very notably, Clade 4 contains all nine of the smallest seeded var. hypogaea types (prostrate and lacking flowers on the main stem): PI 336978, PI 442768, PI 210831, PI 497342, PI 331337, PI 471986, PI 288210, PI 221068, and PI 468280. Network and structure analysis The fastSTRUCTURE analysis resulted in five clusters (K = 5), which are shown in Figure 3 and in Supplementary Files S10 and S11. Among the 5 clusters identified by structure and network analysis, clusters 1 and 2 had the most membership and cluster 3 the least (166, 164, 19) (Supplementary Files S10 and S11). Based on the global ancestry estimates on all genomic SNP sequences (Raj et al. 2014), accessions were colored in accordance with cluster assignment. An accession that could not be assigned to a definitive cluster was painted admixed with colors representative of each cluster with which it proportionally shared genomic sequences.
Overall, 240 accessions were exclusively assigned to a single group and 278 were assigned, in admixed proportions, to two or three groups: with 221 assigned to two, and 57 to three clusters. Of the 12 commercial cultivars genotyped, eight were assigned to cluster 4, along with Tifrunner. Of these, only Jupiter was exclusively assigned to a single cluster with the remaining seven, including Tifrunner, sharing admixed proportions with more than one cluster. Fuhuasheng and Shitouqi were assigned to cluster 2, same as cultivars Olin and Tamnut OL 06. The synthetic tetraploid sequence "duranensis_ipaensis" was assigned to cluster 5 -the only cluster without any commercial cultivars assigned.
Clustering accessions via a phylogenetic analysis is overly simplistic as it suggests a one-dimensional source for sequence similarity or dissimilarity between a pair of accessions. Network analysis provides a more representative and explanatory relationship between given accessions. In Figure 2, accessions with similar sequence characteristics cluster near each other in the network. The further apart accessions are in the network, the more different they are in sequence characteristics ( Figure 2). Four main clusters were defined representing accessions that were more similar to each other and distinct from those in other clusters. Even though most accessions cluster in close correspondence to phylogenetic cluster definitions, exceptions show that a bifurcating tree representation of sequence similarity may not represent the true underlying nature of relatedness among accessions.
Overall, we found groups defined on phylogenetic clade membership to correspond with groups defined by structure and network analyses. These groups showed high genetic differentiation. Clade 1 was genetically distinct from Clades 2, 3, 3.2 and 4 (F ST : 0.74, 0.75, 0.67, 0.51). Clades 3 and 3.2 were not much different from each other (F ST 0.22). Clade 3.2 was also not strongly distinct from Clade 2 (F ST 0.3). Genetic clustering via PCA confirmed the main groups as distinct clusters ( Figure 4A, Figure 5A).

Genetic diversity correlates with subspecies and botanical types:
Principal Coordinates 1 and 2 (PC1 and PC2), which together explained 59.75% of the genetic variation in the collection, differentiated between the two subspecies and corresponding botanical varieties. PC1 separated ssp. hypogaea from ssp. fastigiata, while PC2 delineated between the two ssp. fastigiata varieties; separating var. fastigiata from var. vulgaris. PC1 also corresponded with Virginia and Runner type accessions while PC2 separated Spanish types from Valencia types ( Figures 4B,C).
These results suggest a pattern, consistent with the biology of subspecies and botanical variety classification, as the most important correlates of the genetic diversity in the collection. Previous studies using a subset of this collection, the Mini Core, have suggested the presence of between four to five sub-populations (Otyama et al. 2019;Wang et al. 2011;Belamkar et al. 2011). These results recapitulate and add support to these findings, further linking biology to the landscape of genetic stratification in the U.S. peanut core collection.
Growth form and pod shape did not correspond well with PCA even though both traits are key determinants of agronomic type classification, and, by extension, subspecies groups ( Figures 4D,E). Pod shape considers the constriction, reticulation, and the number of seeds per pod to define five main groups: vulgaris, fastigiata, peruviana, hypogaea and hirsuta. Spanish and Valencia types are classified as "bunch" for their upright growth form while Virginia and Runner types are classified as "runners" for their prostrate (flat) growth form. Several Virginia varieties are also classified as "decumbent", for their intermediate growth form between "runner" and "bunch" (Pittman 1995). The lack of a clear correspondence between growth form and pod shape with genetic diversity, begs for more studies with special emphasis on accurate phenotyping, to help establish their contribution to genetic stratification and diversity in peanut collections. Figure 5 Plots of FST (fixation index) values among genetic groupings, to determine stratification in the core collection. Cluster identities are as shown in the phylogenetic and PCA analyses. The pairwise population differentiation (FST index) was calculated using Hierfstat for a set of unlinked markers and plotted as heatmaps. Accessions were classified into groups of: A. clade membership as defined in the phylogenetic and network analyses, B. botanical varieties, C. market type, D. growth Habit, E. pod shape, and F. continent of seed origin.

Genetic differentiation among groups (F ST fixation index):
The genetic difference between varieties belonging to contrasting subspecies was relatively high. Accessions classified as var. vulgaris appeared genetically distinct from those classified as var. hypogaea with F ST 0.59. The difference was comparatively low for varieties of the same subspecies, var. vulgaris and var. fastigiata accessions, F ST 0.198 ( Figure 5B). This provides clear evidence for the genetic distinction between subspecies and corresponding botanical varieties.
Interestingly, a comparison between var. aequatoriana and var. fastigiata showed a surprisingly high level of differentiation, F ST 0.40. Since both varieties are classified as ssp. fastigiata, genetic differentiation was expected to be much smaller. Contrastingly, we observed low genetic separation for an inter-subspecies comparison between var. aequatoriana and var. hypogaea, F ST 0.2 ( Figure 5B). This result suggests a possible misclassification of var. aequatoriana accessions, which share greater similarity to ssp. hypogaea than the ssp. fastigiata group to which they are assigned. Evidence for misclassification was first suggested by (He and Prakash 2001;Raina et al. 2001;Ferguson et al. 2004;Tallury et al. 2005;Freitas et al. 2007;Cuc et al. 2008) and later alluded to by (Bertioli et al. 2011). However, like their studies, this present analysis suffers from a low number of var. aequatoriana accessions. Additionally, only 159 samples representing accessions in the core, have been classified. Of these, 114 are classified as var. fastigiata, 43 as var. hypogaea and two as var. aequatoriana (Data source: GRIN). Since within-population diversity has been shown to affect F ST as an estimate of genetic differentiation among populations (Hedrick 1999;Bird et al. 2011), we recommend cautious interpretation of these results, especially where they conflict with known peanut biology.
Market types, Spanish and Virginia, showed evidence of genetic differentiation (F ST 0.4), as did Valencia and Virginia (F ST 0.4), and Valencia and Spanish (F ST 0.3) ( Figure 5C). Indeed, accessions marked as "mixed" showed low pairwise genetic differentiation with main groupsas would be expected from a phenotypically ambiguous group. As expected, Runner accessions were more similar to Virginia accessions (F ST 0.027) compared to Valencia (F ST 0.29), and Spanish types (F ST 0.26) ( Figure 5C). Classification studies place Valencia and Spanish types under the same subspecies, ssp. fastigiata, but different botanical varieties -var. fastigiata and var. vulgaris, respectively. Virginia types are classified under a different subspecies altogether -ssp. hypogaea var. hypogaea. This result supports Runner types as a hybrid between the two peanut subspecies as classified by Krapovickas (1969).
Non-distinct phenotypes like pairwise comparisons of growth forms: "spreading-bunch", "spreading", "bunch" and "mixed", which are affected by environmental conditions, resulted in less pronounced genetic separation among groups. The contrast was true with phenotypically distinct groups for pairwise comparisons between growth forms: "spreading" and "prostrate" (F ST 0.55), "spreading" and "erect" (F ST 0.39), "spreading-bunch" and "erect" (F ST 0.28) ( Figure 5D). This suggests a good prediction of phenotypic diversity by genetic variation. Groups defined under pod shape were not distinct from each other, suggesting phenotypic ambiguity in these classes ( Figure 5E).
Collectively, these results suggest a level of stratification that is consistent with subspecies groups and botanical variety classification. Overall, we found accessions were similar within botanical varieties and subspecies groups, but genetic separation increased evidently between group comparisons. This carries important implications for studies using this collection for genetic associations. Treating the collection as a homogenous group may obscure association results and if not properly accounted for, population stratification may cause studies to fail due to lack of significant results or overwhelming false-positive signals.
Geographic origin does not generally correspond with genetic diversity On the whole, the country of seed origin was not an important contributor to structure in the collection. There was little genetic differentiation between peanuts based on where seed was originally collected. African and North American accessions appeared genetically similar (F ST 0.02), as did Asian and African accessions (F ST 0.01) ( Figure 5F). We also found the country of seed origin to be a poor correlate of genetic structure, even though the core collection is predominated by accessions from South America and Africa, which together make up 74.6% of the entire collection. The peanuts collected from Bolivia and South America were not so distinct as to cluster around a recognizable pattern or separate from those collected from other continents. This may suggest that not many independent mutations have arisen in the different continental subgroups to cause significant genetic separation. It is also known that peanuts had completely differentiated into subspecies and botanical varieties prior to being dispersed from their center of origin by early explorers and traders (Simpson et al. 2001).
The mini core is representative of the genetic diversity in the core collection The mini core collection was created to further define a small manageable sub collection representative of the diversity in the germplasm collection. The need was driven by a reliance on lowthroughput markers, like RFLPs and SSRs, which are difficult and costly to assay in large collections and some agronomic traits being quite difficult and costly to measure (Holbrook and Dong 2005). We used genetic clustering via PCA to define how well the mini core represents the diversity in the core collection.
Results show remarkable representation spanning the entire spread of the genetic diversity in the core collection ( Figure 4F). Thus, clustering on select morphological characteristics followed by sampling within defined clusters likely resulted in the selection of a well representative set. The main weakness of the Mini Core is its relatively small size (94 available accessions), which weakens the ability to identify novel marker-trait associations in genome-wide association studies (Otyama et al. 2019). However, the Mini Core collection has proven to be of much utility for identifying germplasm with desirable characteristics for breeding pipelines and for verifying identified marker-trait associations (Holbrook and Dong 2005;Dean et al. 2009;Wang et al. 2011).
Subgenome exchanges are a significant source of diversity in tetraploid peanut An enduring puzzle regarding peanut evolution is that the diversity in the crop appears to have arisen quickly, from a severe genetic bottleneck at the time of the tetraploidization event roughly 10,000 years ago, likely involving a rare, single plant in an early horticulturalist's garden (Bertioli et al. 2019). The diploid progenitors, A. duranensis and A. ipaensis, separated approximately 2 million years ago (Bertioli et al. 2016), and the best evidence is that the mergers of these diploids has occurred only once in pre-modern times (Bertioli et al. 2016;Bertioli et al. 2019). To put the question simply: how did so much genotypic and phenotypic diversity arise in modern peanut varieties?
One source of the diversity was identified by (Bertioli et al. 2019), with the reporting of the high-quality Tifrunner genome sequence. Specifically, exchanges between corresponding chromosomes of the A and B genomes were seen -on scales both small (on the gene-scale), and large (on the scale of multiple megabases, at chromosome ends). We used the genotyping data from the current project to independently assess the patterns of subgenome exchanges.
In the variation data from the Affymetrix array, we found evidence of both widespread small-scale exchanges between subgenomes, and apparent large-scale "invasions" of one subgenome to the other. These patterns are evident in Figure 7, shown in red, whereas gray indicates loci where subgenome exchange either was not observed or there was insufficient evidence regarding exchange. One pattern to n■ Table 1 Counts of genetically unique samples, relative to geographic regions. Unique samples and countries and regions are listed in Supplementary File S1, worksheet "uniques." Detailed counts (per country) are given in Supplementary File S1, worksheet "clade summary." Columns labeled 1-4 indicate clades, as identified in Figure 1, and listed in File S1, worksheet "uniques." Summary counts by continent are given at the bottom, corresponding with F ST results in Figure 5F Region \ clade  1  2  3  3.2  4   Africa -central  2  3  7  0  0  Africa -north  1  2  0  0  0  Africa -northeast  12  13  4  2  0  Africa -northwest  63  31  6  3  1  Africa -south  82  71  38  15  14  Australia  1  6  1  0  0  China  26  10  4  0  0  Europe -east  0  1  3  0  0  Europe -south  3  2  1  1  0  India  29  13  0  3  note is that different accessions show different patterns. Each of the 16 diverse accessions used for comparison is represented along a vertical slice next to each chromosome. At high resolution, many between-accession differences can be seen -for example, at the top of A01, where the first two accessions show an exchange, and the middle accessions do not. Also noteworthy are regions that were reported, in the Tifrunner genome paper, to show invasion (and replacement) of one subgenome by the other. In these locations (marked in green along the chromosome backbones), most alleles are either all red, indicating that the chromosomal segment was contributed by the other subgenome; or all gray, indicating that the chromosomal segment was contributed by the "cis" subgenome. This is evident at the top of A05 and B05, for example. Of the 10,829 SNP positions for which it was possible to evaluate subgenome exchanges (as data were present for all tested lines), there was evidence of exchanges in at least one accession for 1,068 positions (9.8%). This is likely a highly conservative estimate, as many positions are ambiguous with respect to subgenome origin -for example, when the reference SNPs from the diploids may be from the other allele (not represented in the genome sequence).
Our interpretation is that a substantial fraction (.10%) of alleles have arisen through subgenome exchanges; and further, that these exchanges appear to be ongoing, as there are numerous differences between accessions, in the subgenome allele status at a given locus.

CONCLUSIONS
Genotype data for each accession in the U.S. peanut core collection will benefit peanut breeders in multiple ways: providing SNP data for use in marker-trait association studies to identify SNPs associated with important traits, describing the population structure of the core, Figure 7 Plot of inferred subgenome origins. Each colored region (gray or red) indicates data at a SNP location. At each position, values are shown for 16 diverse accessions. In chromosomes A01-A10 (left half), red indicates that alleles are the same as the B-genome assembly (A. ipaensis) and different than the A-genome assembly (A duranensis), at the respective locations (determined by perfect correspondence of flanking sequence). In chromosomes B01-B10 (right half), red indicates that alleles are the same as the A-genome assembly (A. duranensis) and different from the B-genome assembly (A ipaensis). Green marks on the chromosome backbones (e.g., tops of A05 and B05) show the locations of large-scale subgenome invasion, observed in the Tifrunner genome assembly (Bertioli et al., 2019). and enabling breeders to work with smaller groups of accessions by selection through both phenotypic and genotypic characteristics. A probable ancestral genotypic group is identified, with most such accessions still coming from near the geographical origin of tetraploid peanut. The data also provides information about the ongoing rapid changes in the peanut genome through subgenome exchanges, and supports theories about the origin, early cultivation, and dispersion of peanut throughout the world.

ACKNOWLEDGMENTS
This project was supported by the Agriculture and Food Research Initiative Competitive Grant no. 2018-67013-28138 co-funded by the USDA National Institute of Food and Agriculture and the National Peanut Board. Mention of trade names or commercial products in this publication is solely for the purpose of providing specific information and does not imply recommendation or endorsement by the U.S. Department of Agriculture. USDA is an equal opportunity provider and Employer. We thank Dr. Josh Clevenger for insights regarding subgenome exchanges, Dr. Guillermo Seijo for information regarding the origins of cultivated peanut, Dr. H. Eric R. Olsen regarding Portuguese and Spanish colonial trade of peanut, Dr. Shyam Tallury for assistance with germplasm, and Dr. Naveen Puppala for permitting the use of NuMex-01 before its public release.