Genetic evidence for a western Chinese origin of broomcorn millet (Panicum miliaceum)

Abstract

Broomcorn millet (Panicum miliaceum) is a key domesticated cereal that has been associated with the north China centre of agricultural origins. Early archaeobotanical evidence for this crop has generated two major debates. First, its contested presence in pre-7000 cal. BP sites in eastern Europe has admitted the possibility of a western origin. Second, its occurrence in the 7th and 8th millennia cal. BP in diverse regions of northern China is consistent with several possible origin foci, associated with different Neolithic cultures. We used microsatellite and granule-bound starch synthase I (GBSSI) genotype data from 341 landrace samples across Eurasia, including 195 newly genotyped samples from China, to address these questions. A spatially explicit discriminative modelling approach favours an eastern Eurasian origin for the expansion of broomcorn millet. This is consistent with recent archaeobotanical and chronological re-evaluations, and stable isotopic data. The same approach, together with the distribution of GBSSI alleles, is also suggestive that the origin of broomcorn millet expansion was in western China. This second unexpected finding stimulates new questions regarding the ecology of wild millet and vegetation dynamics in China prior to the mid-Holocene domestication of millet. The chronological relationship between population expansion and domestication is unclear, but our analyses are consistent with the western Loess Plateau being at least one region of primary domestication of broomcorn millet. Patterns of genetic variation indicate that this region was the source of populations to the west in Eurasia, which broomcorn probably reached via the Inner Asia Mountain Corridor from the 3rd millennium BC. A secondary westward expansion along the steppe may have taken place from the 2nd millennium BC.

Keywords: China; Loess Plateau; Panicum; agricultural origins; broomcorn millet; domestication; early Holocene; semi-arid.

Figure 1.

(a) Interpolated surface of correlation coefficient values between genetic diversity (unbiased heterozygosity of Chinese broomcorn millet microsatellite data recorded in kernels) and geographic distance. Red colour shows negative correlation values, gradually turning blue the more positive the correlation values become. Since genetic diversity is expected to decrease with geographic distance from the origin of an expansion, regions yielding more negative correlation values represent more plausible locations for the source of spread of broomcorn millet. Green dots show the sample locations. White stars indicate the locations of Dadiwan (1) and Xinglonggou (2). (b) Interpolated surface of correlation coefficient values between genetic diversity (unbiased heterozygosity of panEurasian broomcorn millet microsatellite data recorded in kernels) and geographic distance. Red colour shows negative correlation values, gradually turning blue the more positive the correlation values become. Since genetic diversity is expected to decrease with geographic distance from the origin of an expansion, regions yielding more negative correlation values represent more plausible locations for the source of spread of broomcorn millet. Green dots show the sample locations. White stars indicate the locations of Sokol’tsy (1) and Xinglonggou (2).

Figure 2.

(a) Comparison of the observed difference in Pearson’s correlation coefficients (red line) between Dadiwan (‘1’ in Figure 1a) and Xinglonggou (‘2’ in Figure 1a) generated with Chinese dataset, to the distribution of those generated by permuting (randomly distributing) the site data among sample sites 1000 times (blue line). The p values represent the probability of obtaining the observed difference in correlation values under the null hypothesis of no geographic structure in the genetic data. This can be interpreted as a measure of how well the data favour one site over the other as a location for the source of spread of broomcorn millet, given the assumption that genetic diversity decreases with geographic distance from the origin of expansion. (b) Comparison of the observed difference in Pearson’s correlation coefficients (red line) between Sokol’tsy (‘1’ in Figure 1b) and Xinglonggou (‘2’ in Figure 1b) generated with panEurasian dataset, to the distribution of those generated by permuting (randomly distributing) the site data among sample sites 1000 times (blue line). The p values represent the probability of obtaining the observed difference in correlation values under the null hypothesis of no geographic structure in the genetic data. This can be interpreted as a measure of how well the data favour one site over the other as a location for the source of spread of broomcorn millet, given the assumption that genetic diversity decreases with geographic distance from the origin of expansion.

Figure 3.

Principal components analysis output with samples coloured according to the K genepools from Instruct output (sample majority allocation). The axes represent the first two principal components in each case: (a) 195 Chinese samples, coloured according to K = 3 (see below), (b) 341 panEurasian samples, coloured according to K = 3 and (c) 341 panEurasian samples, coloured according to K = 6.

Figure 4.

Proportional assignments of each landrace sample to ancestral genepools inferred using Instruct (Gao et al., 2007). Each sample is represented as a pie chart, mapped according to its origin as provided by the accession data supplied by the germplasm banks. Different colours of the pie slices represent the K genepools modelled by Instruct. Colours of the genepools are chosen to correspond with previously published analyses of related datasets (Hunt et al., 2011, 2013). The pie charts show the relative membership of the K genepools for each sample. The most realistic inferred values of K are shown: (a) 195 Chinese samples, for K = 3, (b) 341 panEurasian samples, under K = 3 and (c) 341 panEurasian samples, under K = 6.

Figure 5.

Geographical distribution of GBSSI genotypes for 195 Chinese landrace samples. (a) GBSSI-S locus. Samples shown as green points are homozygous wild type, that is, both alleles in the individual are the non-waxy S0. Samples shown as dark blue points are homozygous waxy, that is, both alleles are the mutant S-15. Samples shown as cyan points are heterozygous, that is, both alleles have one wild type (S0) and one waxy (S-15). The S0 allele is dominant, so heterozygous individuals are phenotypically wild type. (b) GBSSI-L locus. Samples shown as red points are homozygous for the wild-type (LC) allele. Samples shown as dark blue and yellow points are homozygous for different waxy mutations (LY and Lf, respectively). The three heterozygous combinations (LC/LY, LC/Lf and LY/Lf) are shown as cyan, orange and green points, respectively.