A total of 775 USDA peanut accessions were phenotyped for growth habit using a binary score 1: spreading, score 2: bunch, and score 3: erect, with the data obtained from the USDA GRIN public data available at https://npgsweb.ars-grin.gov/gringlobal/search .

The Arachis_Axiom2 SNP array was used to genotype the peanut accessions, and this data was made available at https://agdatacommons.nal.usda.gov/ [24] . In the study a total of 13,306 SNPs were used to genotype theaccessions. STRUCTURE 2.3.4 was used to conduct population structure analysis [20] . A total of 10 independent runs, with the Markov Chain Monte Carlo (MCMC) length of burn-in period set to 50,000 and the number of MCMC iterations was 50,000, were conducted to infer population structure (K). The algorithm, developed by Evanno et al. (2005), which STRUCTURE Harvester was established from, was used to identify the optimal K value [25] [26] . Each genotype was assigned to a Q-group using a Q-matrix which contained K-vectors corresponding to the optimal K value with a cutoff probability of 0.55. STURCTURE PLOT in STRUCTURE 2.3.4 and the option “Sort by Q” was used to analyze the population structure [20] .

Prior to conducting GWAS, SNPs were filtered based on the following criteria: heterozygosity <10%, missing data <10%, and minor allele frequency >5%. After filtering, a total of 13,306 SNPs were used for GWAS analysis in TASSEL 5 [27] . A total of five GWAS models were used for this analysis. The first model used was a single-marker regression model (SMR). The second model was the generalized linear model with principal component (PCA) was added as a covariate (GLM_PCA). The third model used was the generalized linear model with the Q matrix, from the population structure analysis, added as a covariate (GLM_Q). A mixed linear model with Kinship (K) was added to the GLM_PCA model (MLM_PCA+K) as the fourth model. For the last model, a mixed linear model with population stratification being controlled by the Q matrix and Kinship (K) (MLM_Q+K) was used. TASSEL 5’s in-built functions were used to estimate both K and PCs. A LOD threshold of > 3.5 was used to identify SNPs significantly associated with plant growth habit phenotype [28] . A candidate gene search was then conducted within a 10-kb region containing a significant SNP and conducted using Peanut Base ( https://www.peanutbase.org/taxa/arachis/ ).

A ridge regression best linear unbiased predictor (rrBLUP) model was then used to compute the genomic-estimated breeding values (GEBVs) [29] . The package “rrBLUP” was used to run the model in R, and the rrBLUP equation was:

γ = WGβ + ℇ

where the y vector phenotype, W the incidence matrix relating the genotype to the phenotype, G the genetic matrix, B the marker effect with $\beta ~N\left(0,I\sigma z_{\beta }^2\right)$, and e the random error. The solution of this equation was:

β˄ = (Z^TZ + Iℷ)⁻¹1Z^Ty

where Z = WG. The ridge parameter was defined as ℷ = ${\sigma }_e^2/{\sigma }_{\beta }^2$, with ${\sigma }_{\beta }^2$ as the marker-effect variance and ${\sigma }_e^2$ as the residual variance.

The effect of the training population size on genomic selection accuracy was evaluated using a 2-fold, 3-fold, 4-fold, 5-fold, and 6-fold cross-validation which corresponded to population sizes of 388, 517, 581, 620, and 646 individuals. A total of 100 replications were used for each cross-validation. Then the accuracy of genomic selection was assessed using Pearson’s correlation coefficient between the observed phenotypes in the population and GEBVs [30] .

Table 1 shows the significant SNPs associated with peanut growth habit using the SMR model. Results from the SMR model indicated that there were 181 SNPs significant to peanut growth habit. These SNPs were located on chromosome A01-A10 and B01-B10, with the majority found on chromosomes B06 and B04. LOD values ranged from 5.43 to 7.65, and the R-square values ranged from 2.74% to 4.46%, indicating that plant growth habit can be controlled by multiple genes with minor effects. The top 10 SNPs with the highest LOD values were AX-147227941 (LOD = 6.9, R² = 4.0%), AX-176811670 (LOD = 6.9, R² = 4.0%), AX-176822503 (LOD = 6.9, R² = 3.6%), AX-176822914 (LOD = 7.0, R² = 3.6%), AX-176820260 (LOD = 7.1, R² = 4.1%), AX-176820577 (LOD = 7.2, R² = 4.2%), AX-176808560 (LOD = 7.2, R² = 3.7%), AX-176823020 (LOD = 7.3, R² = 4.2%), AX-176821681 (LOD = 7.6, R² = 4.0%), and AX-176806956 (LOD = 7.7, R² = 4.5%). These SNPs are located on chromosomes A07 (18,840,166 bp), A06 (95,180,807 bp), B06 (29,039,075 bp), B06 (39,793,516 bp), B06 (108,518,747 bp), B06 (129,766,082 bp), B03 (53,474,187 bp), B03 (122,608,001 bp), B06 (122,581,860 bp), and A06 (96,492,176 bp), respectively ( Table 1 ).

Figure 1(A) and Figure 1(B) show the Manhattan and QQ plot for the SMR model. A total of 85 significant SNPs were found on the A genome, and 96 significant SNPs were found on the B genome. For the A genome, the chromosomes A01, A02, A03, A04, A05, A06, A07, A08, A09, A10 have 4, 7, 6, 31, 5, 10, 6, 6, 2, and 6 SNPs, respectively ( Table 1 ). For the B genome, the chromosomes B01, B02, B03, B04, B05, B06, B07, B08, B09, and B10 have 2, 2, 2, 27, 1, 37, 7, 3, 2, and 13 SNPs, respectively ( Figures 1-5 , Table 1 ).

Figure 1(A) shows clusters of significant SNPs located on the following chromosomes: A04 (76,469,234 bp to 119,296,024 bp), A05 (91,855,042 bp to 101,331,168 bp), A06 (1,982,008 bp to 6,960,258 bp and 9,5180,807 bp to 105,425,114 bp), A08 (1,732,429 bp to 24,455,877 bp), A09 (19,151,716 bp to 20,677,308 bp), B04 (16,211,778 bp to 27,486,400 bp, 75,330,522 bp to 96,903,250 bp, and 100,494,295 bp to 124,286,278 bp), B06 (916,391 bp to 45,645,879 bp and

102,807,076 bp to 129,766,082 bp), and B10 (108,714,204 bp to 134,914,334 bp). The genomic regions where SNP clusters are found suggest that a quantitative trait locus affecting plant growth habit can be found in these areas.

Results from the GLM PCA model identified one SNP, AX-176821681, as significantly associated with peanut growth habit. This SNP is located on chromosome B06 (122,581,860 bp), has a LOD score of 7.1, and an R-square value of 3.6%.

The GLM Q model identified 108 SNPs as significantly associated with peanut growth habit. The SNPS were located on chromosomes A02-A10, B03-B04, and B06-B10, most frequently occurring on chromosomes A04, B04, and B06. LOD values ranged from 5.4-7.9, and R-Square values ranged from 2.7% to 4.6%. The ten SNPS with the highest LOD values are AX-176799357 on A02 (76,275,147 bp, LOD = 7.0, and R² = 4.1%), AX-14722704 on A07 (18,840,166 bp, LOD = 7.0, and R² = 4.1%), AX-176822704 on B06 (9,163,091 bp, LOD = 7.1, and R² = 4.1%), AX-176808560 on B03 (53,474,187 bp, LOD = 7.1, and R² = 3.6%), AX-17682250 on B06 (29,039,075 bp, LOD = 7.1, and R² = 3.6%), AX-176822914 on B06 (39,793,516 bp, LOD = 7.2, and R² = 3.7%), AX-176820577 on B06 (129,766,082, LOD = 7.4, and R² = 4.3%), AX-176823574 on B06 (125,992,228 bp, LOD = 7.4, and R² = 4.3%), AX-176806956 on A06 (96,492,176 bp, LOD = 7.9, and R² = 4.6%), and AX-176821681 on B06 (122,581,860 bp, LOD = 7.9, and R² = 4.1%). Of the SNPS identified, 52 were located on the A genome and 56 were located on the B genome. For the A genome, the chromosomes A02, A03, A04, A05, A06, A07, A08, A09, A10 had 3, 4, 23, 3, 8, 3, 3, 2, and 3 SNPS respectively. For the B genome, the chromosomes B03, B04, B06, B07, B08, B09, and B10 had 2, 22, 25, 2, 1, 2, and 2 respectively.

Clusters of SNPs significantly associated with plant growth habit were found on the following chromosomes: A03 (128,328,312 bp to 134,513,642 bp), A04 (76,790,192 bp to 118,900,581 bp), A08 (24,455,849 bp to 24,455,877 bp), B04 (75,330,522 bp to 115,634,391 bp), B06 (22,822,411 bp to 29,039,075 bp and 108,518,747 bp to 129,766,082 bp), and B10 (119,790,764 bp to 134,914,334 bp).

The MLM PCA found one SNP (AX-176821681) to be significantly associated with peanut growth habit on chromosome B06 (122,581,860 bp) with an LOD score of 5.6 and R² square of 2.9%.

The MLM Q model found 10 SNPs to be significantly associated with peanut growth habit. Located on chromosomes A01 and A06, and B03, B06, and B08, the ten SNPS identified are AX-176800551 on A01 (69,524,629 bp, LOD = 3.1, and R² = 1.9%), AX-177642631 B08 (39,306,016 bp, LOD = 3.2, and R² = 1.9%), AX-176807751 on A06 (2,763,696 bp, LOD = 3.2, and R² = 1.9%), AX-176806956 on A06 (96,492,176 bp, LOD = 3.4, and R² = 2.0%), AX-147241123 on B02 (23,631,160 bp, LOD = 3.4, and R² = 2.0%), AX-147253437 on B06 (12,153,774 bp) (LOD = 3.5, and R² = 1.7%), AX-176813106 B06 (134,300,181 bp) (LOD = 3.6, and R² = 2.1%), AX-176806377 B06 (122,321,048 bp, LOD = 3.8, and R² = 1.9%), AX-176808560 on B03 (53,474,187 bp, LOD = 3.9, and R² = 1.9%), AX-176821681 B06 (122,581,860 bp, LOD = 4.9, and R² = 2.5%).

Only a single cluster of SNPs significantly associated with plant growth habit was found on chromosome B06 from 121,537,744 bp to 134,300,181 bp.

Figure 6 shows the genomic selection accuracy of plant growth habit using different training population sizes. The results indicated that a larger training population provided better genomic selection accuracy. The highest accuracy (r = 0.61) was obtained for the training population 646, whereas the lowest accuracy was recorded for the training population size 388 (r = 0.23). These results demonstrates that genomic prediction can be used as a selection tool for plant growth habits in peanuts.