The gene flow analysis results did not identify a single domestication center for proso millet, as cluster 4 did not clearly originate from cluster 5. Our results could reflect either single or multiple domestication centers, as population structural analysis cannot assume that K = 1 [9]. Our gene flow analysis results were insufficient to demonstrate that cluster 4 originated from cluster 5; they only showed that gene flow from cluster 5 to cluster 4 was greater than that from cluster 4 to cluster 5. According to the local diversity of each cluster (Table S1) and the gene flow analysis results, we determined the primary or secondary center of origin of each clusters. Nepal (and its surrounding areas) was the secondary center of origin for cluster 6. Our phylogenetic, population structure, and gene flow analysis results indicated that cluster 6 originated from cluster 5, and that the secondary center of origin was in Nepal and its surrounding areas, in high-latitude and high-altitude areas. This is similar to the three gene microcenters detected in Turkey, where the cultivation environment was exceptionally well-suited to wheat domestication [24]. In South Korea, cluster 1 was found to have derived from cluster 5, which is a unique germplasm resource. A previous study also identified Japan as an origin point for proso millet cluster 1 [9].

Cluster 1 is considered to be a unique proso millet germplasm resource that was domesticated in South Korea. Although one germplasm resource was collected in Russia, and one in Thailand, while two germplasms were collected in India, gene flow analysis showed that they were all derived from the Korean germplasm. Cluster 2 is distributed only in Asia. Gene flow analysis showed that these two clusters were introduced from China to South Korea, Mongolia, Uzbekistan, and India, and then from India to Russia and Thailand. Cluster 3 is widely distributed in Eurasia, with a diversity index of 0.143174. However, our gene flow results showed that cluster 3 originated in the fertile crescent of Turkey and then spread to France and India, from which it subsequently spread to Central Asia (Uzbekistan and Turkey) and finally to East Asia. In cluster 4, the highest genetic diversity was detected in germplasms native to Turkey; however, gene flow analysis showed that Turkey’s seed resources migrated from China to Uzbekistan, and then to Turkey. Cluster 4 originated in China and was introduced via Mongolia to Russia and North Korea. Germplasms of China and Mongolia also spread to Uzbekistan, and then from Uzbekistan to India, India to Thailand, Uzbekistan to Turkey, and Turkey to Ukraine. Cluster 5 occurred frequently in East, Southeast, and South Asia, and evolved according to a network pattern. Cluster 5 had the highest diversity; gene flow analysis pointed to China as the center of domestication, from which it migrated to Europe (Ukraine) via different routes, and then to Mongolia and finally back to China. Cluster 6 showed high abundance in high-latitude and high-altitude regions; it originated in Nepal and was then introduced into Ukraine, Russia, and Mongolia, and then into South Korea from Ukraine.

By combining the phylogeny, diversity, population structure, M, and gene flow analysis results, we obtained expansion paths for each cluster of proso millet that were consistent with a previously established archaeological map of the agricultural origins and migration of Neolithic and formative cultures [25]. In this study, we provided SSR markers combination of each cluster to identify the cluster where an individual is located (Table S2). The highest diversity was found in clusters 4 and 5, and the longest branches (indicating the longest genetic distance from other germplasms) were found in clusters 3 and 5. Regions with high diversity were not identified as regions of origin, likely because gene introgression from other proso millet clusters resulted in high diversity. In a future study, we will amplify and sequence these genotypes to identify the markers with the greatest diversity (SSR-203, SSR-232, or SSR-365) as plant DNA barcodes, to detect other known antioxidants. Then, we will use these markers to accurately determine genotype composition and compare it with a neutral model to shed light on the population expansion associated with bottleneck events.

In this study, the population genetic diversity is low, and the genetic differentiation is high. We believe that the reason is that the proso millet is an inbred plant. Self-crossing reduces the effective population size and effective recombination rate. Compared with outcrossing, it directly leads to a decrease in polymorphism and an increase in linkage disequilibrium [26]. The increase in isolation between populations also directly stems from selfing or indirectly from evolutionary changes, leading to greater differentiation of molecular markers than during outcrossing. The lower effective recombination rate increases the possibility of free-riding and further reduces the internal diversity of inbreds, thereby increasing their genetic differentiation [27].

In addition, in the process of agronomic traits data collation, we observed that the standard deviation within each sub-cluster was very high (Table 4). The degradation of cultivars of proso millet to wild species may be the main reason for the large standard deviation. In this study, the population of 578 accessions is composed of wild species and landraces. Landraces degenerate into wild species, some genotypes are preserved, and may still be clustered with local landraces, but agronomic traits show differences. Feral derivatives of crop varieties may show a similar phenotype to that of the crop ancestor [28].

The results showed high-level LD. This may be due to the mating system (selfing) of proso millet affecting the pattern of LD [29]. We arranged the TPC and SOD data corresponding to each accession on the outer circle with a simple bar (Figure S3). We observed that two red clades had higher TPCs, with average values of 25.4 and 23.2 μg/g. Blue clades had lower TPCs, with average values of 25.4 and 23.2 μg, respectively. Each clade can be clustered together because some SSR markers have the same amplification length. Therefore, we speculate that certain markers may be linked to the quantitative trait loci (QTL) associated with TPC. Comparison of the genotypes in these 4 clades showed high TPC in the 261-bp SSR-195 marker, and low TPC in the 269-bp SSR-195 marker (Supplementary Dataset 4). To further explore the relationships between these phenotypes and markers, we performed an association analysis of the phenotype and genotype data. Because proso millet is a selfed species, in the results of genetic differentiation analysis, a high degree of differentiation was observed in the population. This may lead to false positives in the association results. We choose to use the mixed linear model (MLM) for correlation analysis. In the MLM (Q + K) model, the population structure matrix (Q) and the kinship matrix (K) are used as random effects to control the false positives. SSR-31 associated with TPC explained >7.1% of the variance in TPC, indicating that there may be QTLs on both sides of the marker.

There have been a large number of reports on the antioxidant mechanism of total phenols. The number and position of hydroxyl groups in phenolic compounds, and the nature of the substituents on the aromatic ring, determine the antioxidant capacity of plant extracts [30]. Whole-genome sequencing of the proso millet genome has been completed [31]. In future research, we will map the positions of SSR-31 identified in this study on the proso millet genome and develop flanking markers to gradually narrow the target range and determine the main proso millet genes influencing TPC and oxidation capacity, and clarify the types of phenolic compounds and antioxidant mechanisms in this species. This will provide new opportunities for high-quality proso millet breeding.