Data and analysis files from: Oaxacan green dent maize is not from Oaxaca

“Oaxacan Green Dent” is a maize (Zea mays L.) cultivar marketed as an introduced Mexican heritage cultivar adapted to the higher latitudes of the USA. Its adaptation and appearance contradict an origin in Oaxaca, Mexico, however, and no indigenous cultivars in Oaxaca are known to have the unique kernel colors of Oaxacan Green Dent. We compared phenotypes and genotypes of Oaxacan Green Dent sampled from three different sources along with several Corn Belt cultivars and 15 landrace accessions collected from a wide range of geography, altitude, and cultural groups in Oaxaca. Multivariate analysis of 13 phenotypic traits measured in a field experiment suggested that Oaxacan Green Dent is more closely related to Corn Belt Dents than to Oaxacan cultivars. Genomic analysis from DNA sequencing demonstrated unambiguously that Oaxacan Green Dents are even more distantly related to Oaxacan cultivars than typical USA Corn Belt Dent cultivars are. Phenotypic, genetic, and historical data indicate that Oaxacan Green Dent is almost certainly directly derived from Ernest Strubbe’s Green Dent cultivar, which he developed in Minnesota from crosses between a Corn Belt Dent cultivar and an intensely colored popcorn cultivar, with no contribution from Oaxacan cultivars.

Methods
Germplasm

The term “cultivar” is used throughout the paper as a generic and neutral term to refer to open pollinated maize populations with distinct names. The genetic materials used in this study include open-pollinated maize populations that have been variously considered heirloom cultivars, heritage cultivars, farmer’s varieties, or landraces (Preston et al., 2011; Curry, 2022).

We obtained seeds of Oaxacan Green Dent from Seed Savers Exchange, Restoration Seed Company, and Sand Hills Preservation Center. Seeds used in the field experiment were directly from the seeds obtained from suppliers, except for the Seed Savers Exchange source, which was increased by controlled cross-pollination among 100 plants in the field (Woore et al., 2024). We also obtained seeds of five USA Corn Belt cultivars and 15 Oaxacan cultivars from the National Plant Germplasm System maize collection (Volk et al., 2023). The USA cultivars included two populations of Reids Yellow Dent and one population of Minnesota 13, which represent important germplasm sources that contributed to modern commercial corn hybrid pedigrees (Troyer, 1999; Smith et al., 2022). In addition, we included ES Green Dent and Black Beauty, a black popcorn cultivar donated to the Plant Introduction collection by Ernest Strubbe, which may have been a source of the kernel anthocyanin colors in ES Green Dent.

The 15 Oaxacan landrace accessions were selected to represent a range of previously described Mexican native maize groups, latitudes, and elevations within the Zapotec areas of Oaxaca, principally the Valles Centrales, the largest area of Zapotec culture, as well as the Sierra Norte where the Zapotecos de la Sierra live, and the area of the Istmo of Tehuantepec where the Zapotecos del Istmo are located (Table 1; Figure S1). Most of these collections were not assigned to landrace groups in the USDA Germplasm Resources Information Network, but the sampled cultivars represent at least Bolita, Negrito, Olotillo, Tepecintle, Tuxpeño, and Zapalote Chico groups (Table 1).

Field Experiment

The 23 cultivars were evaluated in a randomized complete block design with two replications in Clayton, NC in 2021. Experimental units were single 3.7-m rows, seeded at a rate of 25 seeds per plot with rows spaced 0.97 m apart. Seedlings were thinned as needed to ensure interplant distances of at least 0.1 m within rows. N was applied at 53.2 kg ha−1 pre-planting and then at 190.5 kg ha−1 in a split application during the growing season. Irrigation was applied as needed up to 25 mm of water per week. Days to anthesis (DTA) and days to silk (DTS) were recorded as the number of days from planting to 50% of plants within a plot shedding pollen or with emerged silks, respectively. Anthesis-silk interval was computed as the number of days between anthesis and silk emergence. Five plants per plot were measured for ear height, plant height, ear position (ear height relative to plant height), number of primary tassel branches, tassel branched part length (the length of the tassel between uppermost and lowest primary branches). Up to six plants per plot were measured for cob length, ear diameter, cob diameter, number of kernel rows, cob color, kernel width, and weight of 100 kernels. Details of most trait measurements are given in Woore at al. (2024). Traits measured were previously demonstrated to have high heritability and relatively low influence from environment and genotype-by-environment interaction effects (Sánchez G. et al., 1993; Woore et al., 2024), such that single-environment data on multiple traits are sufficient for measuring phenotypic relationships.

Trait data analysis

Traits with one measurement per plot were analyzed with the following linear mixed model:

Y_ij=μ+R_i+C_j+ε_ij, where Y_ij is the trait measurement on the ith replicate block and jth cultivar, R_i is the random effect of block, C_j is the fixed effect of cultivar, and ε_ij is the residual effect. Traits with more than one measurement per plot were analyzed with the following linear mixed model:

Y_ijk=μ+R_i+C_j+ε_ij+w_ijk, where terms are the same as the previous model except that Y_ijk is the trait measurement on the kth plant in the ith replicate block and jth cultivar, ε_ij is the random experimental error effect, and  w_ijk is the residual effect that represents the effect of individual plant measurements within a plot.

The null hypothesis of no cultivar effects was tested with a Wald F-test and cultivar best linear unbiased estimates (BLUEs) were estimated. Mixed model analyses were fit with ASReml-R version 4.2 (Butler et al., 2017).

Cultivar BLUEs were used to estimate pairwise correlations among traits. Trait pairs with correlation greater than r = 0.95 were pruned, dropping the trait with more missing data. Data were missing for various reasons, including poor germination in a few plots and plants not producing ears. The resulting matrix of cultivar-by-trait means had 10% missing data. Multivariate analyses required a complete data matrix, so missing cultivar-trait means were imputed. First, Black Beauty was missing ear and plant height data (because of decomposition at time of height measurements), but this cultivar was evaluated in a common trial grown in the same location in summer 2024 with the other temperate maize cultivars used in this experiment, so its missing ear height value was imputed based on relative differences to other cultivars observed. Oaxaca 321 was missing data for silking date, anthesis-silk interval, and ear height because of its photoperiod sensitivity which resulted in no ear development. Because it was very tall and very late (based on plant height and anthesis observations), we imputed these missing values as equal to the maximum mean values among all entries in the experiment. All remaining missing values were imputed with the mice package in R (van Buuren and Groothuis-Oudshoorn, 2011).

The complete matrix of cultivar-by-trait means was subjected to principal components analysis. Trait loadings on the first two principal components and cultivar scores on the first four principal components were visualized with the ggplot2 package in R (Wickham, 2016).

DNA sequencing

Five plants per cultivar were sampled and equal amounts of tissue for each plant were bulked for DNA extractions. Several cultivars were represented by two independent samples of five plants each. Inbreds B73 and CML311 were also sampled multiple times to serve as quality control checks for sequencing. Oaxaca 17 seeds did not germinate well enough to include in the DNA sequencing study. A cultivar named “Strubbes Green” from Sand Hill Preservation Center was also included in the sequencing study. Multiplexed reduced representation sequencing was used to sequence all cultivars following the library preparation methods of Manching et al. (2017) using an Illumina NovaSeq 6000. Sequences were aligned to B73 version 5 reference genome (Hufford et al., 2021). SNPs were called using bcftools mpileup and call functions (Danecek et al., 2021). SNPs with more than 23% missing data were filtered out, leaving 39,708 SNPs remaining. A distance matrix was computed in TASSEL version 5 (Bradbury et al., 2007). Cluster analysis and multidimensional scaling were performed on the resulting distance matrix using functions in base R (R Core Team, 2019). A dendrogram was visualized using factoextra package in R (Kassambara, 2024).