A total of 1,582 proso millet germplasm resources (including 1,516 landraces, 32 cultivars, 9 wild varieties from China, and 25 foreign varieties, with the accessions from China specifically sourced from Inner Mongolia, Heilongjiang, Jilin, Liaoning, Hebei, Shandong, Gansu, Qinghai, Xinjiang, Shanxi, Shaanxi, and Ningxia), were obtained from the Crop Science Institute of the Chinese Academy of Agricultural Sciences ( Figure 1 , Supplementary Tables S1 , S2 ).

Leaf samples were collected at the three-leaf stage, and plant DNA was extracted from them using a modified CTAB method. DNA quality was assessed by 1% agarose gel electrophoresis, while DNA concentration was determined using a nucleic acid detector (NanoDropOne).

SSR locus detection for the transcriptome unigenes of salt-tolerant varieties was performed using MISA software. The minimum repeat thresholds were set to 10, 6, 5, 5, 5, and 5 for mono-, di-, tri-, tetra-, penta-, and hexa-nucleotide repeat sequences, respectively. The SSR sequences obtained were then utilized for primer design with Primer 3.0 software. Eighty SSR primers ( Supplementary Table S4 ) were synthesized by Beijing TSINGKE Biological Co., Ltd.

Twenty varieties with significant phenotypic differences ( Supplementary Table S5 ) were used for polymorphism screening. Fifteen primers ( Supplementary Table S6 ) yielded clear amplification bands that revealed stable polymorphisms, making them suitable for subsequent genetic diversity analysis of 147 identified superior proso millet germplasm resources ( Supplementary Tables S7 , S8 ). PCR amplification was conducted using the EasyCycler PCR system, and PCR products were separated by 8% polyacrylamide gel electrophoresis, followed by silver staining for band detection.

Microsoft Excel 2021 was employed to compile phenotypic data, while SPSS 27.0 and Origin 2025SR1 were employed to analyze and visualize phenotypic data. Electrophoresis results were manually scored to create a binary data matrix. Genetic diversity parameters were calculated using PopGene 1.32 and PowerMarker 3.25, and cluster analyses were performed using MEGA 6.06. Population genetic structure and principal component analyses were conducted with Structure 2.34 and NTSYSpc 2.11. Based on SSR marker scoring results, data were sequentially encoded to construct DNA molecular IDs. QR code online technology (https://cli.im/text) facilitated the conversion of basic information into QR code DNA molecular IDs.

The genetic diversity indices of 14 agronomic traits, 5 yield traits, and 3 grain traits were evaluated ( Figure 2 , Supplementary Table S9 ). The genetic diversity index was lowest for inflorescence color (0.15) and highest for stem diameter (2.10) among the 14 agronomic traits (comprising 10 qualitative and 4 quantitative traits). Notably, there were two traits with genetic diversity indices exceeding 2, both quantitative in nature. The differences in the frequency distribution of various qualitative traits are presented in Figure 2C . For the 1,582 germplasms, inflorescence color was predominantly green (96.5%), with purple accounting for merely 3.5%. The dense trichome type (59.5%) was the main trichome type. Regarding panicle type, the lateral type was predominant (52.6%), whereas scattered and compact types represented 23.6% and 23.8% of accessions, respectively. The main axis shape was primarily slightly bend (54.1%), with erect and bend types accounting for 23.3% and 22.6% of accessions, respectively. Inflorescence density was chiefly intermediate (41.7%) and slightly dense (37.4%). Most varieties lacked projections on branch bases (66.8%) and branching habits (72.6%), with minor angle deviations present in most varieties (60.6%). Short, intermediate, and long-branched spike length were evenly distributed, accounting for 32.7%, 32.7%, and 34.6% of accessions, respectively. Furthermore, the majority of varieties exhibited weak shatter resistance (44.4%), while strong shatter-resistant varieties were the least frequent (24.8%). As shown in Figure 2D , the average number of main stem nodes among the 1,582 germplasms was 6.34, with a variation coefficient of 16.8%. Stem height ranged from 33.67 to 236.67 cm, with an average of 129.57 cm, reflecting nearly a 7-fold disparity between the extrema, accompanied by a variation coefficient of 20.09%. The average stem diameter was 6.79 mm, with a variation coefficient of 22.12%. Straw weight per plant spanned from 1.53 to 110.10 g, averaging 32.15 g, with a 72-fold disparity between extrema, marked by the largest variation coefficient (60.03%).

The genetic diversity indices of yield traits ranged from 1.95 to 2.08, with tiller number having the lowest index at 1.95, while grain weight per plant had the highest at 2.08 ( Supplementary Table S9 ). The genetic diversity indices for panicle length, panicle weight per plant, and thousand-grain weight all exceeded 2. As shown in Figure 3 , the average tiller number among 1,582 germplasms was 2.26, with a considerable variation coefficient (31.54%). The average panicle length was 37.05 cm, exhibiting a variation coefficient of 21.43%. Panicle weight per plant ranged between 0.88 and 28.83 g, averaging 9.97 g, with a 33-fold difference between extrema and a substantial variation coefficient (35.64%). Grain weight per plant varied from 0.61 to 19.11 g, averaging 7.20 g, displaying a 32-fold difference between extrema and a high variation coefficient (37.37%). The average thousand-grain weight was 7.09 g, with a variation coefficient of 14.94%. The above results collectively indicate there was substantial variation in yield traits among proso millet accessions, with considerable genetic diversity.

The genetic diversity indices for grain length, grain width, and length-to-width ratio were 2.04, 2.02, and 2.00, respectively, indicating considerable genetic diversity in these grain traits ( Figure 4 , Supplementary Table S9 ). As shown in Figure 4B , the average grain length and width among 1,582 accessions were 2.85 mm and 2.16 mm, respectively. Grain length exhibited the lowest variation coefficient at 5.22%, whereas grain width had a variation coefficient of 6.61%. The average grain length-to-width ratio was 1.33, with a variation coefficient of 5.48%. Genetic diversity indices for grain traits were higher than those of certain agronomic and yield traits, while their variation coefficients are comparatively lower than those of all agronomic and yield traits.

Figure 5 presents the correlation analysis between four primary qualitative traits and eight representative quantitative traits. Among the qualitative traits, a highly significant positive correlation was observed between panicle type and inflorescence density (r = 0.50), while a highly significant negative correlation was observed between panicle type and branched spike length (r = -0.24). No correlations were detected among other qualitative traits. In examining the quantitative traits, the tiller number demonstrated a positive correlation with the main stem nodes (r = 0.07) and a negative correlation with panicle length (r = -0.13), yet showed no significant correlation with other traits. The number of main stem nodes was not correlated with panicle length but was highly significantly and positively associated with stem height (r = 0.61), panicle weight per plant (r = 0.37), grain weight per plant (r = 0.40), straw weight per plant (r = 0.44), and thousand-grain weight (r = 0.12). Panicle length was not correlated with the number of main stem nodes, was negatively correlated with the tiller number (r = -0.13), and was positively correlated with the stem height (r = 0.16), panicle weight per plant (r = 0.14), grain weight per plant (r = 0.10), straw weight per plant (r = 0.10), and thousand-grain weight (r = 0.27). Stem height, panicle weight per plant, grain weight per plant, straw weight per plant, and thousand-grain weight exhibited no correlations with tiller number. Stem height was highly significantly and positively associated with panicle weight per plant (r = 0.59), grain weight per plant (r =0.58), straw weight per plant (r = 0.70), and thousand-grain weight (r = 0.37). Panicle weight per plant was highly significantly and positively associated with grain weight per plant (r = 0.91), straw weight per plant (r = 0.56), and thousand-grain weight (r = 0.43). Grain weight per plant was highly significantly and positively associated with straw weight per plant (r = 0.54) and thousand-grain weight (r = 0.39). Straw weight per plant was highly significantly and positively associated with thousand-grain weight (r = 0.41). Among the relationships between qualitative and quantitative traits, panicle type was not correlated with the tiller number, but was negatively correlated with panicle length (r = -0.18) and positively correlated with the number of main stem nodes (r = 0.28), stem height (r = 0.37), panicle weight per plant (r = 0.24), grain weight per plant (r = 0.26), straw weight per plant (r = 0.34), and thousand-grain weight (r = 0.13). Inflorescence density was negatively correlated with panicle length (r = -0.23), and positively correlated with tiller number (r = 0.16), the number of main stem nodes (r = 0.31), stem height (r = 0.27), panicle weight per plant (r = 0.27), grain weight per plant (r = 0.27), straw weight per plant (r = 0.27), and thousand-grain weight (r = 0.13). Branched spike length was highly significantly positively correlated with the tiller number (r = 0.16) and panicle length (r = 0.22) and highly significantly negatively correlated with stem height (r = -0.08), while also exhibiting a significant negative correlation with straw weight per plant (r = -0.05); however, it showed no significant correlations with the remaining traits. Shatter resistance exhibited no correlation the tiller number, panicle length, or grain weight per plant, yet was highly significantly positively correlated with the number of main stem nodes (r = 0.06), stem height (r = 0.12), panicle weight per plant (r = 0.11), straw weight per plant (r = 0.13), and thousand-grain weight (r = 0.14).

Principal component analysis was employed to standardize the relative values of various traits, effectively eliminating factors with minimal impact but significant interference, to facilitate a comprehensive evaluation of proso millet germplasms ( Supplementary Table S10 ). The first principal component accounted for 41.25% of the variance, the second for 13.59%, the third for 9.66%, and the fourth for 9.27%, with these four components collectively explaining 73.78% of the variance. The eigenvalues of these four principal components each exceeded 1, confirming their suitability for a comprehensive evaluation of proso millet germplasm. The highest loading values for the first principal component were associated with stem height, panicle weight per plant, and grain weight per plant. The second principal component was most positively correlated with thousand-grain weight, indicating that this trait heavily contributed to the component. The third principal component displayed the highest positive loadings for grain length and grain length-to-width ratio, suggesting it mainly reflects the overall condition of the grain. The fourth principal component exhibited higher absolute values for the tiller number and panicle length, showing its dependency on these traits.

Based on the score coefficient matrix of the four principal components, the comprehensive score (F) for each variety was calculated to rank proso millet germplasms by excellence ( Supplementary Table S11 ). A higher F value indicates a better overall phenotype for that variety. The top 10 performing varieties were Hongruanmi, Hongruanmi, Laohuangmi, Getami, Ruantiaozhoumi, Zhuyeqing, Limashu, Mizi, Nuomizi, and Hongnianmi. Among these, Ruantiaozhoumi and Limashu are from Shanxi province; the remaining eight high-performing varieties are all from Shaanxi province.

Cluster analysis of the 1,582 germplasms identified five distinct groups ( Figure 6 ). The first group includes 22 varieties, including Etou and Sansui, mainly originating from Heilongjiang, accounting for 1.4% of accessions. The second group comprises 292 varieties, including Zaohongmi and Anchunwei, primarily sourced from Heilongjiang and Inner Mongolia, along with most foreign varieties, representing 18.5% of the germplasms. The third group consists of 528 varieties, including Daqihuangmizi and Taolibai, largely obtained from Shanxi and Inner Mongolia, comprising 33.4% of accessions. The fourth group contains 736 varieties, including Ruantiaozhuomi and Ziganmi (shu), predominantly sourced from Shanxi and Shaanxi, followed by Gansu and Inner Mongolia, constituting 46.4% of accessions. The fifth group consists of only four varieties, namely Hongruanmi, Hongruanmi, Getami, and Laohuangmi, all originating from Shaanxi, accounting for just 0.3% of accessions. Based on the clustering analysis of phenotypic traits, the germplasms within the five groups were ranked in descending order of comprehensive F values and compared against qualitative traits. This approach resulted in the identification of 147 representative superior germplasms, which were systematically selected for subsequent genetic diversity analysis.

Cluster analysis of 147 proso millet germplasms was conducted by unweighted pair group method with arithmetic mean analysis (UPGMA), which indicated that the germplasms could be divided into four groups ( Figure 7A , Table 4 ). Group 1 encompasses one variety, sourced from Heilongjiang, representative of the Northeast region. Group 2 includes one variety, sourced from Shandong, representative of the Northern region. Group 3 comprises 31 varieties with 8 from Shanxi, 6 from Qinghai, 3 from Gansu, 3 from Ningxia, 3 from Hebei, 2 from Heilongjiang, 1 from Shaanxi, and 5 from abroad, mainly from the Loess Plateau and Northwest regions. Group 4 contains 114 varieties, with contributions from Shanxi (22), Shaanxi (22), Heilongjiang (19), Inner Mongolia (14), Gansu (9), Jilin (7), Ningxia (7), Xinjiang (5), Hebei (4), Qinghai (2), Shandong (2), and Liaoning (1), primarily from the Loess Plateau, Northeast, and Northwest regions. Additionally, Figure 7A and Table 4 collectively illustrate that all foreign varieties are consolidated within Group 3.

Three-dimensional principal component analysis (PCA) was conducted on the 147 proso millet superior germplasms, as shown in Figure 7B . The first three principal components, denoted as PC1, PC2, and PC3, accounted for 21.37%, 18.87%, and 17.39% of the total variance, respectively, with a cumulative explanation of variance reaching 57.63%. The 147 germplasms were thus classified into six groups: 14 from the Inner Mongolia region, 30 from the Northeast region, 10 from the Northern region, 25 from the Northwest region, 63 from the Loess Plateau region, and 5 from abroad, with each group corresponding to their geographical sources.

The genetic structure of 147 proso millet accessions was analyzed using Structure. Peaks in the number of genetic populations (K) were identified at K = 3, K = 5, and K = 8 ( Figure 7C ), prompting analysis under both models. For K = 3, the germplasms were categorized into three groups as illustrated in Figure 7D . Group 1, marked in red in Figure 7D (73 varieties), predominantly sourced from the Loess Plateau, and Inner Mongolia regions. This includes varieties from Shanxi (19), Shaanxi (13), and Inner Mongolia (12). Group 2, marked in green in Figure 7D (39 varieties), primarily consists of varieties mainly originating from the Loess Plateau, Northwest regions, and abroad (5), including varieties specifically from Qinghai (7), Shanxi (6), Shaanxi (6), and abroad (5). Group 3, marked in blue in Figure 7D (35 varieties), encompasses varieties mainly from the Loess Plateau, and Northeast regions, including Shanxi (5), Shaanxi (4), and Heilongjiang (14).

For K = 5, the germplasms were divided into five groups. Group 1, marked in red in Figure 7D (27 varieties), consists of varieties predominantly from the Northeast, and Loess Plateau regions, comprising Heilongjiang (11), Shanxi (5), and Shaanxi (3). Group 2, marked in green in Figure 7D (20 varieties), includes varieties mostly sourced from the Loess Plateau region, featuring Shanxi (6), Ningxia (3), and Shaanxi (2). Group 3, marked in blue in Figure 7D (16 varieties), encompasses varieties primarily from the Northwest region, specifically Qinghai (5). Group 4, marked in yellow in Figure 7D (20 varieties), primarily comprises varieties mainly from the Loess Plateau region, and abroad, including Shaanxi (4), Ningxia (3), and abroad (5). Group 5, marked in purple in Figure 7D (64 varieties), contains varieties primarily from the Loess Plateau, and Inner Mongolia regions, comprising Shanxi (15), Shaanxi (12), and Inner Mongolia (10).

For K = 8, the germplasms were categorized into eight groups as illustrated in Figure 7D . Group 1, marked in red in Figure 7D (18 varieties), predominantly sourced from the Northeast, and Northwest regions. This includes varieties from Heilongjiang (8), and Gansu (4). Group 2, marked in green in Figure 7D (14 varieties), primarily consists of varieties mainly originating from the Loess Plateau region, including varieties specifically from Shanxi (6), Shaanxi (2), and Ningxia (2). Group 3, marked in blue in Figure 7D (16 varieties), encompasses varieties mainly from the Loess Plateau, Inner Mongolia, and Northwest regions, including Shanxi (3), Ningxia (3), Qinghai (3), and Inner Mongolia (3). Group 4, marked in yellow in Figure 7D (25 varieties), mainly comprises varieties from the Loess Plateau, and Northeast regions, with varieties specifically from Shanxi (6), Shaanxi (6), Ningxia (3), Heilongjiang (3), and Jilin (3). Group 5, marked in purple in Figure 7D (17 varieties), contains varieties primarily from the Loess Plateau, and Northeast regions, comprising Shaanxi (3), and Heilongjiang (5). Group 6, marked in aqua in Figure 7D (11 varieties), consists of varieties predominantly from the Northeast, and Loess Plateau regions, comprising Heilongjiang (4), and Shanxi (5). Group 7, marked in golden in Figure 7D (31 varieties), includes varieties mostly sourced from the Loess Plateau, Northwest, and Inner Mongolia regions, featuring Shanxi (8), Xiangjiang (4), Shaanxi (3), and Inner Mongolia (6). Finally, Group 8, marked in crimson in Figure 7D (15 varieties), primarily includes varieties from the Loess Plateau, and Northwest regions, specifically Qinghai (4), Shanxi (3), and Shaanxi (2).

To facilitate the conservation of the genetic resources of proso millet varieties, fingerprint maps were constructed for all 147 evaluated superior varieties on the basis of the SSR marker electrophoresis results ( Figure 8 ). The electrophoresis results were digitally encoded, and following the principle of distinguishing the maximum number of varieties with the minimum number of primers, all amplification outcomes of the assayed varieties were combined in the following order: BLF41, BLF4, BLF47, BLF8, BLF9, BLF52, BLF58, BLF21, BLF66, BLF-27, BLF-79, and BLF80. This procedure produced a distinct 24-character string for each variety ( Supplementary Table S12 ). These strings of DNA molecular IDs were then augmented with additional information such as Name, Unicode, Source, and Classification to create QR code DNA molecular IDs using online QR code generation technology. QR code DNA molecular IDs for varieties 1, 2, 3, and 4 are presented in Figure 9 , while those for the remaining 143 varieties are detailed in Supplementary Figure S1 .