Fig. 2. PCA of landscape factors of Taihang Mountains.
PCA analysis was performed on 160 individual landscape factors with VIF<5.
Finally, eight landscape factors were retained, namely average precipitation in August, average precipitation in October, average precipitation in November, built-up land (residential and infrastructure), rain-fed cultivated land, workability (restricted site management), solar radiation in August, and soil PH.

2.6 FST outliers filtered and selected SNPs identified

F ST outliers generally show the selected genes or loci among populations. To identify the selected SNPs in the twoOpisthopappus species, BAYESCAN 2.1 software was used to filter the F ST outliers (Fischer et al., 2011; Foll et al., 2010; Ruan et al., 2021). Prior odds of the selection model were set at 10,000 to reduce false-positive results under a variety of demographic events. A logarithmic scale for model choice of selection over neutrality was defined as: substantial (log10PO > 0.5, 0-0.05), strong (log10PO > 1.0, 0.05-0.15), very strong (log10PO > 1.5, 0.15-0.25) and decisive (log10PO > 2, F ST>0.25). A gene or locus with log10PO > 0.5 was considered as a potential selective outlier under natural selection (Feng et al., 2015). Finally, 29 genes/loci identified based on the BayeScan were considered as putative SNPs under selection. These SNPs were retained for the subsequent landscape features association analysis. Then the filtered SNPs were extracted from the VCF file.

2.7 Association of SNPs with landscape factors

The SNP associations with landscape factors were assessed using Samβada v.0.9.0 and latent factor mixed model (LFMM) software (Chien et al., 2020; Feng et al., 2015; Ruan et al., 2021). Samβada builds logistic regressions to estimate an individual’s probability of presenting a particular molecular marker depending on the landscape factors that characterize its sampling site (Li et al., 2019; Vargas-Mendoza et al., 2016).
In order to accurately describe the landscape factors of each population, the eigenvalues of the first four principal components of principal component analysis (PC1-4) were chosen, which explained 77.04% of the total landscape features. In Samßada, the effect of each landscape factor was tested by adding one factor at a time to the population landscape factors (dimensionP+1), and the more likely model was assessed (without or with the landscape factors). For each test model, Samßada created an output file containing the model parameters, logarithmic likelihood, G score, Wald score, AIC, and BIC. To ensure the model’s accuracy, all the models were screened according to the AIC value (Mahtani-Williams et al., 2020; Stucki et al., 2017). Then the first 29 valid models were selected with the smallest AIC value, and the proportion of each factor in these 29 models was counted. And the 29 models involved a total of three genes among selected SNPs. These genes were subsequently subjected to carry KEGG annotation (https://www.genome.jp/kegg/).
LFMM is a hierarchical Bayesian hybrid model, which considers the background of population structure as the random effect of population history and isolation by distance model, and through the potentialK value of population structure (Frichot et al., 2013; Wang et al., 2017). In LFMM, the genetic data matrix was tested based on a z-score as a fixed effect. The number of possible factor K was set to 2 (according to the Structure results). LFMM ran 5 times with 10,000 iterations in the Gibbs sampling algorithm and a burn-in period of 5,000 cycles for each K value. Z-scores from five independent replicate runs were combined using Fisher–Stouffer method, and the P values were adjusted using the genomic inflation factor (λ). P values were further adjusted based on an FDR correction of 1 % using the R ‘qvalue’ package to get Q values (Li et al., 2019).

3. Result

3.1 Genetic characteristics about two species

For the all 17 populations, the HLT population had the largest genetic diversity values, Ar, H o and H s were 1.262, 0.138, 0.104 respectively (Table 2). SLD population had the minimum values, Ar and H S were 1.198 and 0.082 respectively. However, WML population showed a lowestH O value of 0.116.