Skip to main content
NCBI home page
Wiley Open Access Collection logoLink to Wiley Open Access Collection
. 2026 Mar 6;35(5):e70287. doi: 10.1111/mec.70287

Population Genomic Structure of Sorghum Landraces Across Landscape, Environment and Culture

Eleanna E Vasquez Cerda 1,✉, Emily S Bellis 2, Aayudh Das 1, Emma R Slayton 3, Geoffrey P Morris 4, Jesse R Lasky 1,5
PMCID: PMC12965196  PMID: 41790003

ABSTRACT

The spread of staple crops to diverse environments over time and their current genetic structure may reflect historical dispersal by humans, sustained human preference for particular traits and adaptation to local environments. Sorghum is a drought‐tolerant crop native to Africa cultivated by hundreds of millions of smallholders globally. Here we examined the ecological context of population‐genomic structure of 1806 sorghum landraces across Africa and Eurasia to infer the relative contribution of environmental and cultural factors to sorghum genetic diversity across different relative time periods. Sorghum landraces were spatially and linguistically structured at a large scale and within subregions, following a pattern of isolation by distance. Within regions, much of the genomic structure was best explained by a mechanistic model of human travel time. In our assessment of hierarchical linguistic structure, we found that language families explain 4% of genomic variation while individual languages explain 13% of genomic variation, suggesting the importance of human culture and relationships in gene flow and selection. Variance partitioning showed that travel time, language and climate explain up to 27% of genomic variation among landraces. We also observed regional differences in the degree of genetic relatedness across space and time in our assessment of shared ancestry. East Africa showed particularly strong geographic turnover in genomic composition and haplotype sharing, while West Africa showed substantial haplotype sharing even over large distances, signifying some rapidly spreading lineages. Thus, space, travel time and culture likely capture important forces controlling sorghum genomic variation, but these factors operate heterogeneously over space.

1. Introduction

The rate of gene flow and geographic patterns of relatedness may be shaped by dispersal ability, geographic distance and population size (Bohonak 1999). In natural plant populations, genetic material like seeds and pollen are dispersed through vectors like wind or animals (Ellstrand 2014; Rousset 2000) and the gene flow distribution has a higher density of events within short distances and a lower density at long distances (Ellstrand 2014; Ibrahim et al. 1996; Rousset 1997; Slatkin 1993). The combination of limited gene flow and genetic drift can generate a pattern of isolation by distance; more distant populations often have increased genetic differentiation (Kimura and Weiss 1964; Sexton et al. 2014; Wright 1943). Among other natural forms of dispersal, human‐mediated dispersal can play a role in gene flow dynamics, particularly in long‐distance dispersal events (> 5 km) (Auffret et al. 2014; Bullock et al. 2018; Ellen and Platten 2011; Wichmann et al. 2008). Human‐mediated dispersal events may be accidental (Wichmann et al. 2008), or intentional through the exchange of seeds among communities (Menamo et al. 2021) and human migration (Boivin et al. 2017; Ellen and Platten 2011).

By modelling human‐mediated dispersal for crops, inferences can be made about plant gene flow dynamics and spatial connectivity (Gutaker et al. 2020). Human movement can be modelled by identifying least cost paths (LCP), which in this context assumes that humans traverse heterogeneous landscapes by considering routes that minimise energy expenditure (Adriaensen et al. 2003; Gowen and de Smet 2020; Rees 2004). Rates of human movement between points may be modelled using estimated travel time, where locations connected by shorter travel times are likely to have more frequent movement. If human movement is the primary vector of crop gene flow, then least cost travel times for humans may better predict the genetic differentiation between crop populations than geographic distance (Gutaker et al. 2020).

Humans also play an essential role in the distribution of crop genetic variation through intentional and unintentional selection. Breeders intentionally cross distinct genotypes to confer beneficial traits (Ohadi et al. 2017) like increased yield and stress tolerance (Swarup et al. 2021). For smallholders, selection may be driven by perceived habitat suitability (Atinkut and Mebrat 2016; Seo and Mendelsohn 2008). Similarly, culture may inform conscious or unconscious preference for varieties not explained by climate (Faye et al. 2019) such as non‐adaptive traits like colour (Lacy et al. 2006). Cultural factors like ethnicity, farmer social structure and language have been associated with population structure in crops like maize, barley and sorghum (Benz et al. 2007; Gilabert et al. 2025; Samberg et al. 2013; Westengen et al. 2014). This association may be explained by the ‘farming‐language codispersal hypothesis,’ which holds that languages have historically dispersed with the diffusion of agriculture, aligning with migration and population expansion events (Diamond and Bellwood 2003). Using language as a proxy for culture may help untangle the confounding relationships affecting genetic diversity such as spatial structuring and environment as well as environment and culture (Leclerc and d'Eeckenbrugge 2012; Perales et al. 2005; Velásquez‐Milla et al. 2011). By incorporating these factors into models of population structure, we may better understand the individual and conjoint effects of spatial connectivity, human movement, language and climate on sorghum genetic structure.

Patterns of genetic relatedness across space may be temporally heterogeneous, as demography and patterns of gene flow across landscapes can change over time. For example, changes over time in the slope of relatedness decay over distance have been observed in European humans (Al‐Asadi et al. 2019; Ralph and Coop 2013). While longitudinal genomic samples over time are often unavailable, identity by descent tracts of different lengths, which are shared portions of the genome passed down over generations from a common ancestor between two haplotypes, can reveal temporal dynamics in demographic history of populations (Browning and Browning 2012). The length of identity by descent tracts (along the genetic map), can be utilised as a relative estimate of timescales where longer segments represent more recent genetic relationships while shorter segments represent older relationships, allowing us to look back in time (Browning and Browning 2013). Pairwise comparison of identity by descent tracts between samples may reveal temporal changes in the degree of shared ancestry and the relationship between genetic relatedness, geographic distance and other predictors (Al‐Asadi et al. 2019; Ralph and Coop 2013).

Sorghum is a self‐pollinating drought‐tolerant crop that is geographically widespread and adapted to a wide range of climatic conditions (Menamo et al. 2021; Morris et al. 2013; OECD 2014; Olatoye et al. 2018). For millennia since its domestication in Africa, sorghum has served as a staple for people across Africa and Asia with versatile use as food, fibre and fuel (OECD 2014; Venkateswaran et al. 2019). Researchers have postulated that sorghum originated in eastern Sudan and western Ethiopia 4600 to 10,000 years ago based on archaeological evidence (Kimber 2000; Smith et al. 2019; Winchell et al. 2017) and more recently, based on spatially explicit genetic analyses (Gilabert et al. 2025). Cultivated sorghum likely diffused along ancient overland and sea trade routes within continental Africa and Asia (Kimber 2000; Venkateswaran et al. 2019). For example, sorghum dispersal from Africa to the Sind‐Punjab region and India may have occurred from ports in East Africa via the Indian ocean and along the Red Sea (Sabaean lane) around 3000 bce and 2000 bce, respectively (Harlan and Stemler 1976; Kimber 2000). Within continental Africa, diffusion of varieties followed North–South and East–West axes around > 3000 bce and 2000 bce, respectively (Kimber 2000). Putative dispersal routes of sorghum varieties (Kimber 2000) may have been determined by landscape and environmental features favouring seed movement (i.e., the primary mechanism of gene flow in a self‐pollinating species).

Culture is associated with sorghum genetic variation regionally (Faye et al. 2019; Westengen et al. 2014) and across Africa (Gilabert et al. 2025). The effects of geography and climate on population structure have also been assessed at a large scale (Lasky et al. 2015). However, the relative effects of language, human movement, geography and environment have not been considered across global sorghum landraces. Gilabert et al. (2025) recently identified genetic clusters in sorghum that were associated with specific climates and language families, which when subdivided exhibited fine‐scale population structure. For example, two genetic clusters in West Africa were associated with the Atlantic‐Congo language family and Mande subclass. By conducting an effective migration surface analysis in Africa, both Gilabert et al. (2025) and Morris et al. (2025) found evidence of geographic structuring, specifically corridors across space with increased gene flow and areas with reduced connectivity such as the Ethiopian highlands. However, previous studies have not assessed the combined influence of language, geography and climate over time across African and Eurasian sorghum.

We build upon findings in previous studies by estimating spatiotemporal and linguistic differences in genetic structure as well as the relative effect of spatial connectivity, human movement, language and climate on sorghum genomic diversity. We estimate hierarchical genetic differentiation in 18 language families and 97 individual languages. By conducting a variance partitioning analysis, we disentangle the portion of genetic variation explained by language families relative to geographic distance, climate and human movement. Here we highlight both geographic distance and human‐induced dispersal patterns as distinct drivers of population structure using isolation by distance and travel time models across continents, regions and language families. To infer temporal differences in predictors of population structure we estimate the distribution of shared ancestry among paired landraces across space, climate and language, an approach that to date has not been implemented in landscape genetics. Here we ask:

  1. Is sorghum genetic variation better explained by geographic distance or a human movement model? Dispersal limitation and genetic drift likely result in a positive linear relationship between genetic distance and geographic distance or travel time. However, travel time may be a better predictor of genetic diversity as geographic barriers that constrain human‐mediated dispersal are considered.

  2. How does the hierarchy of human language diversity explain sorghum genomic variation? Shared preferences for traits or similar agronomy and environment among individuals sharing the same language may act as selective forces that drive genetic differentiation between landraces associated with different languages.

  3. What is the relative contribution of geographic isolation, language and climate in explaining sorghum genomic variation? While climate may explain a portion of genomic diversity due to local adaptation to climate, we expect a large proportion of this relationship is collinear with geography due to spatial autocorrelation in climate gradients (Coelho et al. 2023; Ruiz Miñano et al. 2022; Sork et al. 2010). Collinearity between climate and language may reflect overlap between climate and human‐mediated selection, as well as genetic drift, where climates and language groups function as barriers to gene flow.

  4. How do the above patterns change among regions and over time? Processes like spatiotemporal habitat heterogeneity may shift patterns of gene flow across space and over time. This may be reflected in differences in the spatial distribution of landraces across regions and time. If sorghum landraces follow a pattern of isolation by distance, we expect the number of shared identity by descent segments to decay with increasing geographic distance between paired landraces. Furthermore, if the processes we described above operate heterogeneously over time, we expect larger identity by descent tracts to display the strongest isolation by distance pattern as more recent relationships may be confined across geographic space (Ralph and Coop 2013).

2. Methods

2.1. Genomic and Climatic Data

We obtained genotype by sequencing (GBS) data for 1806 sorghum landraces collected across Africa and Eurasia (Figure 1) from the study by (Hu et al. 2019). For most of the analysis below, we excluded the relatively isolated Chinese landraces because their status as geographic outliers could overly influence some analysis. For wavelet analysis, which allows localised estimation of population structure, we included the Chinese landraces (see below).

FIGURE 1.

FIGURE 1

Open in a new tab

(A) Map of 1806 sorghum landraces used in regional analyses (East, West and South Africa and South Asia) and the Middle East which was included in global analyses. The waypoints used to connect intercontinental landrace pairs are denoted by symbols. (B) scatterplot of geographic and genetic distance globally, (C) scatterplot of travel time in subregions East Africa, South Africa, South Asia and West Africa. Spline (white line) added to show distribution pattern. Density bar indicates concentration of landraces (blue‐low, red‐high).

We considered temperature and moisture as variables due to their effects on crop performance and evidence that crops are adapted to these aspects of the environment (Bannayan et al. 2011; Gutaker et al. 2020; Riha et al. 1996; Slingo et al. 2005; Sloat et al. 2020; Verón et al. 2015). We extracted climate variables for each georeferenced landrace using CHELSA v.2.1 data with 30 arc‐second resolution (Brun et al. 2022). To account for highly correlated climatic variables pertinent to moisture and temperature, we conducted a PCA using the princomp and fviz_pca_var in R and the factoextra package (Kassambara and Mundt 2020), respectively. We also generated a pairwise correlation matrix. For temperature we used four variables: temperature seasonality, maximum temperature of the warmest month, minimum temperature of the coldest month and mean temperature of the driest quarter. For moisture we considered four variables: annual precipitation, precipitation of the driest month (log(x + 1) transformed to adjust for skew), precipitation seasonality and precipitation of the warmest quarter.

2.2. SNP Filtering

To avoid likely read mapping errors, we removed SNPs with more than 5% heterozygotes in a custom script and converted to ped/map format using gdsfmt v.1.38.0 and SNPRelate v.1.36.0 in R v.4.3.3. We also removed rarer variants (MAF ≤ 0.05) using PLINK v.1.9.0‐b.7.7 leaving a total of 135,085 SNPs.

2.3. Genetic Distance

To consider the extent to which geographic distance and travel time influence genetic divergence between landraces, we estimated genetic distance for each pairwise landrace (Hamming distance, PLINK v.1.9.0‐b.7.7 (Purcell et al. 2007)). We found Hamming genetic distance (Hamming 1950), was the best fit between genetic and geographic distance (R 2 = 0.185) compared to Rousset's (1/1 − F ST) (R 2 = 0.000) (Rousset 1997), pairwise F ST (R 2 = 0.003) (Ochoa and Storey 2021; Wright 1943), Nei's (R 2 = 0.184) (Nei 1972) and Euclidean genetic distance (R 2 = 0.1672) measures. Hamming distance, which is based on coding theory (Hamming 1950), in this context is a measure of the number of nucleotide substitutions (allele counts) in sequences of equal length. While similar to Hamming distance, allele shared distance (ASD) between individuals is measured as the number of shared alleles in a locus (genotype), where two shared alleles are equal to 0, one allele is 1 and no shared alleles is 2 (Gao and Martin 2009).

2.4. Geographic Distance

To test for signatures of isolation by distance, we calculated geographic distance between pairs of landraces using the Vincenty ellipsoid, using the distm function from geosphere v.1.5‐18 (Hijmans 2024). This method follows an ‘as the crow flies’ route finding (i.e., irrespective of barriers) for each pairwise landrace.

2.5. Travel Time

As an attempt to more accurately estimate rates of sorghum gene flow between populations, we modelled human‐mediated dispersal of sorghum landraces by calculating travel time of the least cost path (LCP) between landraces. To do so, we first modelled human movement across terrain of heterogeneous slope, where steep slopes slow the speed of movement (Tobler 1993). Speed as determined with Tobler's hiking function assumes that the maximum walking speed of 6 km/h is reached on a slight decline (−0.05° slope) while the speed of walking across flat land is 5 km/h (Goodchild 2020; Tobler 1993). To obtain slopes we used elevation data from Worldclim v.1.4 with 30 s resolution (Hijmans et al. 2005). Over sea we simulated low‐technology sailing (Gutaker et al. 2020). For travel over water, we estimated three nautical miles per hour or 1.54 m/s (Irwin et al. 1990; Slayton 2018; Surface‐Evans and White 2012), causing travel across water to be 1.1× faster than across flat land (Gutaker et al. 2020).

For each pair of landraces, we estimated travel time by generating a conductance layer, which we use to estimate resistance across a given geographic range using gdistance v.1.2‐2 (van Etten 2017). The conductance layer contains values equal to the reciprocal of travel time between a focal cell and its neighbouring eight cells. Using the costDistance function from gdistance v.1.6.2 in R v.4.2.1, we calculated the number of hours for the LCP between each pair of landraces; this function computes the reciprocal of the conductance matrix values (1/conductance = distance/speed) between two sets of coordinates so that travel time = speed/distance for movement over land and sea (Gutaker et al. 2020; Slayton 2018). Due to computational constraints to working with a single conductance layer across Africa and Eurasia at the desired resolution, we generated separate conductance layers for Africa and South Asia and then connected intercontinental landrace pairs via putative important waypoints. We included seven potential waypoints representing both ports and centers of ancient trade: Suez (Egypt), Mombasa/Fort Jesus (Kenya), Saylac/Zeila (Somalia), Kilwakisiwani (Tanzania), Sabaean Lane/Al Luḩayyah (Yemen), Stone Town/Zanzibar (Tanzania) and Mogadishu (Somalia) based on historical sorghum transportation or ancient trade routes (Harlan and Stemler 1976; Murdock 1959) (Table S1). For example, the Sabaean Lane in modern‐day Yemen was an important overland and maritime connection between Africa and South Asia (Harlan and Stemler 1976). To estimate travel times for pairs of African and Eurasian landraces, we first calculated the least cost path from all landraces within each continent to each waypoint. Then, for each unique landrace combination between different continents, we extracted the waypoint with the shortest combined travel time. To calculate travel time for distant landraces or computationally intensive regions within Africa, we generated additional conductance layers by constraining layers to the smallest size possible within 5° of the minimum and maximum longitude and latitude. We also calculated travel time for landraces with unique locations by rounding the coordinates to 0.25° and assigning the output to the remaining corresponding landraces.

2.6. Spatial Model Selection

To assess which model (geographic distance or travel time) was a better predictor of genetic distance, we extracted AIC values from a linear mixed effects model where individual genotypes have random effects (Clarke et al. 2002) using the MLPE.lmm function from ResistanceGA v.4.2‐10 (Peterman 2018) in R v.4.2.1. We compared transformations of geographic distance, specifically the square root, log and untransformed version and found that the untransformed version was the best fit to genetic distance based on AIC. As an alternative to AIC, we conducted a Mantel test (Spearman's rank correlation with 9999 permutations) for geographic distance and travel time matrices using the vegan package in R (Oksanen et al. 2012).

2.7. Language Structure

To evaluate whether landrace genomic variation is linguistically structured, we first obtained language data from the glottolog database (Hammarström et al. 2022), which contains georeferenced language and language family data across the world. First, we assigned each landrace to a language and language family based on the geographically closest language, as geographic distance is correlated with linguistic typological distance (Auer et al. 2013). The assigned language (family) represents the geographically closest language to each landrace, which does not necessarily represent all languages present in the proximate area. To estimate population structure among language families and languages for 135,085 SNPs, we first filtered for language families with more than five languages, leaving a total of 785 landraces. We then used the varcomp.glob function from hierfstat (Goudet 2005) v. 0.5‐11 to calculate hierarchical F ST. To estimate F ST among language families across all 1806 landraces, we used Weir and Cockerham's estimate from hierfstat.

2.8. Regional Analyses

The processes that generate population genomic differences among landraces, such as human‐mediated gene flow, may operate differently in different regions. Thus, we stratified our isolation by distance analyses by subregions based on putative dispersal patterns of sorghum varieties mentioned in (Kimber 2000; OECD 2014). We conducted SNP filtering and calculated genetic distance for each subregion in isolation. We then tested the relationship between geographic and genetic distance within each region to estimate geographic change in isolation by distance.

To further understand geographic patterns of population structure across spatial scales and locations, we implemented wavelet analysis of genome‐wide landrace differences. Wavelet analysis allows estimation of spatial scale‐specific genomic differences among genotypes and can be used to localise changes in these patterns across the landscape (Lasky et al. 2024). We calculated genome‐wide wavelet dissimilarity, which is an estimate of scale‐specific genetic differentiation, at a range of spatial scales from ~50 m to ~8100 km for all landraces from Africa and Eurasia. We compared the observed wavelet genetic dissimilarity at each scale with a null distribution generated from permuting landrace locations (i.e., a null model of panmixia, used for comparison).

2.9. Temporal Changes in Population Genomic Structure

While some models may present evidence of geographic structuring, it would be difficult to disentangle which genetic relationships are more recent across geographic space, particularly if the distribution of individuals or pattern of isolation by distance is expected to vary with time (Al‐Asadi et al. 2019; Ralph and Coop 2013). Temporal differences in geographic and genetic structure can be measured through comparison of identity by descent tracts, opening a potential window into past relatedness (Lawson et al. 2012). Genotype sharing of identity by descent segments across segments of different lengths can represent a range of timescales: smaller shared identity by descent segments representing older relationships and larger shared segments arising from recent relationships.

To estimate identity by descent sharing, we first filtered SNPs and individuals in the VCF file to match other analyses using VCFtools (Danecek et al. 2011) v.0.1.16. For the remaining SNPs, we used the BEAGLE v. 5.5 phasing algorithm (27Feb25.75f) (Browning et al. 2021) along with a sorghum genetic map (Marla et al. 2025) to estimate recombination rate. To identify identity by descent segments across haplotypes between paired landraces, we used the software Refined IBD v. 17Jan20.102 (Browning and Browning 2013) along with the phased data and genetic map. Following a script from Al‐Asadi et al. (2019), we created an identity by descent sharing matrix for different length bins (1–3, 3–5 and > 5 cM). For each paired landrace within each region, we calculated the mean number of shared identity by descent segments across haplotype pairs, where chromosomes are treated separately for each landrace. To assess geographic distance decay in the sharing of identity by descent segments, and regional differences, we tested a Poisson weighted regression model of geographic distance with region as an interaction versus pairwise identity by descent sharing, using regional sample sizes as weights (Ralph and Coop 2013).

2.10. Variance Partitioning

To assess the unique and redundant prediction of population genomic structure by language families, geographic distance, travel time, temperature and moisture, we conducted variance partitioning with redundancy analysis (RDA). RDA is an ordination of multivariate responses (SNPs) with multivariate predictors (e.g., language, climate) that allows estimation of the proportion of genetic variation explained by sets of factors. We reduced the dimensions for geographic distance and travel time matrices by filtering for unique coordinates and extracting a set of spatial eigenvectors with positive values, which we isolated with the pcnm function from vegan following (Bauman et al. 2018). This resulted in 158 and 421 eigenvectors for the geographic distance and travel time models, respectively across 1403 landraces. We used the varpart function from vegan to conduct variance partitioning (Oksanen et al. 2012) v.2.6‐10 in R v.4.3.3.

Additionally, we estimated changes in the processes driving population structure over time by conducting distance‐based variance partitioning for pairwise sharing of different identity by descent segment bins (1–3, 3–5 and > 5 cM). To reduce skew and convert the haplotype sharing matrix into a measure of genetic distance, we applied a sqrt((1/(1 + x))2) transformation, where x is the mean number of shared identity by descent segments for a pair of landraces. We then used this matrix in a distance‐based RDA (Legendre and Anderson 1999) and variance partitioning with the same covariates described above. This represents a novel inference strategy for understanding variation in pairwise haplotype sharing for segments of different lengths.

3. Results

3.1. Pattern of Isolation by Distance and Travel Time

In our analysis of pairwise geographic and genetic distance, we found strong evidence of isolation by distance in 1806 sorghum landraces across Africa and Eurasia (Figure 1A). We observed an increase in genetic distance as geographic distance increased (R 2 = 0.185) (Figure 1B). This pattern indicates that landraces that are geographically proximate are more genetically similar than those that were further apart. Similarly, we observed a pattern of isolation by travel time, specifically an increase in genetic distance as travel time hours increased (R 2 = 0.188, Figure S1). Contrary to our expectations, AIC favoured a geographic distance model (AIC = 2.1086e+07) over travel time (AIC = 2.1089e+07) globally, suggesting some limitations to our global travel time model. Similarly, a simple Mantel test indicated the geographic distance model (r = 0.46, p = 1e−04) was favoured over travel time (r = 0.45, p = 1e−04).

3.2. Regional Differences in Spatial Structuring

In contrast to global models, we found that within regions, the travel time model was most favoured. AIC estimates indicated the travel time model was a better fit in East Africa (AIC_geodist = 1,447,729, AIC_travel = 1,447,668, n = 367), West Africa (AIC_geodist = 1,825,684, AIC_travel = 1,825,608, n = 410) and South Asia (AIC_geodist = 1,209,018, AIC_travel = 1,208,859, n = 338), whereas in South Africa (AIC_geodist = 65,035, AIC_travel = 650,371, n = 329) the favoured model was geographic distance. The difference in model selection results globally versus within regions may be due to differences in the magnitude of distances, and the domain over which our model worked best. Globally, 59% of pairwise distances were over 3000 km and travelled through our waypoints between continents, while regionally only 0.6% were over 3000 km. Additionally, our waypoints and intercontinental travel may have some inaccuracies.

We observed an approximately linear increase in genetic distance with increasing travel time hours and geographic distance within each region, at least for the first few hundred kilometres and hours (Figure 1C and Figure S2). However, while all regions follow a pattern of isolation by distance, the patterns of spatial genetic structuring were variable. Notably, landraces in West Africa contained the most genetically distinct landraces within short distances. The distinct division between two clusters and plateau in genetic distance in West Africa is indicative of highly diverged landraces and limited gene flow between the two groups. These genetic clusters correspond to the genetically divergent guinea margaritiferum variety (S5–7) and the more common, guinea, durra and caudatum type present in the region (Gilabert et al. 2025; Morris et al. 2013, 2025). We observed a similar pattern in South Asia, though to a lesser degree, indicating the divergence between the genetically distinct guinea and durra varieties that dominate opposite wet and dry sides of the Indian subcontinent. Regional analyses indicate that 11%, 13% and 14% of genetic variation was explained by travel time constraints in West, South and East Africa, respectively. On the other hand, only 7% was explained by the same model in South Asia, which suggests that the pattern of spatial structuring may be stronger within Africa (where sorghum is native) compared to South Asia.

There were also major differences revealed among regions in scale‐specific genetic differentiation (wavelet analyses), that allowed us to identify particular locations of major genomic turnover (Figure 2). At the ~387 km scale there were still a few regions with significantly low genetic differentiation (i.e., high genetic similarity). In particular, Chinese landraces had significantly low genetic differentiation at this scale, potentially reflecting their more recent colonisation (< 2 kya, versus e.g., ~5 kya for colonisation of Punjab) (Kimber 2000; Qingshan and Dahlberg 2001) and the rapid spread of closely related genotypes across China. By contrast, at the same 387 km scale there was particularly high genetic differentiation along the Rift Valley in Ethiopia, a region of high sorghum diversity and great topographic and climate heterogeneity (Lasky et al. 2015), as well as in eastern India, a region where the very distinct guinea and durra landraces meet (Wang et al. 2020). At the scale of ~1039 km, we found significantly high genetic differentiation nearly everywhere, especially West Africa (Burkina Faso to Nigeria) and SE Africa (Zambia to Mozambique), which both correspond to the early axes of sorghum spread out of east Africa (Kimber 2000) and are regions of turnover in major genetic clusters (Wang et al. 2020), as well as western India and Pakistan, which correspond to sharp rainfall gradients along which sorghum landraces may be locally adapted (Lasky et al. 2015).

FIGURE 2.

FIGURE 2

Open in a new tab

Scale‐specific and localised population genomic structure in sorghum landraces identified using wavelet analyses, with blue indicating less and red more genomic turnover than panmixia, respectively. (top panel) The global mean wavelet genetic differentiation (dissimilarity) increases with spatial scale, becoming greater than a null expectation from panmixia (grey) at ~100 km. (bottom three panels). (bottom panels) Genetic differentiation at a range of spatial scales, with significantly low differentiation (e.g., northeast Chinese landraces at 387 km scale) shown in blue and high differentiation shown in red. Non‐significant locations lack colours.

3.3. Sorghum Landraces Are Linguistically Structured

In our analysis of language structure, we used hierarchical F ST (HFST), metrics to estimate the degree to which languages and language families explained population genetic structure in sorghum landraces in 785 sorghum landraces (i.e., those landraces assigned to language families with at least 5 languages) and associated 18 language families. We found that 4% of genomic variation was among language families while 13% was among individual languages (Figure 3A). Additionally, 10% of genetic variation within language families was among languages. Our estimate of global F ST across 1806 sorghum landraces and 42 associated language families similarly indicated that only 4% of genetic variation was explained by language families. When we stratified this analysis across continents, we found that 3% and 2% of genomic variation was explained by language families within Africa and South Asia, respectively. In our estimate of HFST, we observed weaker linguistic structuring in South Asia relative to Africa: in South Asia there was no genetic variation among language families while 8% of genetic variation was both within individual languages, and languages within language families. On the other hand, in Africa, 2% of genomic variation was among language families while 12% was within individual languages, and 10% of variation within language families was within languages. These findings indicate that sorghum landraces are linguistically structured, but the portion of genetic variation explained by individual languages is greater than that of language families at a large scale and across continents. Comparatively, the degree of linguistic structuring was stronger in Africa than in South Asia.

FIGURE 3.

FIGURE 3

Open in a new tab

(A) Diagram of population and subpopulation hierarchies and the associated HFST. Out of 785 landraces, 13% of genetic variation is among language groups, 10% of variation within families is among individual languages and 4% is among language families. (B) Map of language families observed in assessment of linguistic structuring. Points represent the geographically closest language family to each landrace. (C) Scatterplot of geographic and genetic distance across individual language families. Spline (white line) added to show distribution pattern. Density bar indicates concentration of landraces (blue‐low, red‐high).

3.4. Linguistic Regions and Isolation by Distance

While the pattern of isolation by distance was present globally, this pattern varied across language families (Figure 3B,C). For example, although the Nubian and Austroasiatic language families had comparable sample sizes and geographic extents (Sudan and eastern India, respectively), the linear models indicated that 0% and 28% of genetic distance was explained by geographic distance for each respective language, perhaps due to the role of large‐scale rapid diffusion of sorghum along the Nile River in Sudan. Within the Atlantic‐Congo language family, which had the largest sample size, 37% of genetic distance was explained by the travel time model while only 11% and 13% were explained by the model within the Afro‐Asiatic and Indo‐European language families, respectively. Thus, among language families there was variation in patterns of isolation by distance. Although generally there was an increase in genetic distance as travel time increased across language families, the rate of increase in genetic distance was variable (slopes): in the Indo‐European language family genetic distance steadily increased and began to plateau around 500 h, while in the Afro‐Asiatic language family the increase was nonmonotonic and steep with a decline around 500 h. Conversely, within the Afro‐Asiatic language family, which was comparatively geographically constrained, genetic dissimilarity increased rapidly as travel time increased and then decreased steadily after 500 h. On the other hand, landraces within the Mande language family were predominantly constrained to West Africa yet exhibited evidence of genetic divergence at relatively short travel times. Just as in our regional analysis, this divergent genetic cluster may be associated with the genetically distinct guinea margaritiferum variety and the more common durra varieties. This finding follows the hypothesis of an independent domestication event within West Africa by individuals from the Mande linguistic group around 4500 bce (Murdock 1959). Linguistic differences in spatial structuring across the remaining language families may be attributable to variation in geographic extent within language families, regional differences in variety preference within language families, or variety‐specific differences in adaptation to particular climates, which may be similar at extreme distances or highly variable within short distances/travel hours.

3.5. Temporal Differences in Spatial Population Genomic Structure

We used different identity by descent tract lengths to represent the degree of shared ancestry at different timescales where sharing of smaller (1–3 cM) and larger (> 5 cM) represents older and more recent relationships, respectively. The mean tract length in our dataset was 3.9 cM, the median was 3 cM while the first and third quantile were 2.1 cM and 5 cM, respectively. We found the mean number of shared identity by descent segments across length bins decreased with geographic distance (Figure 4A), consistent with isolation by distance. Poisson regression models across all length bins indicate that geographic distance, regions, and the interaction between geographic distance and regions are statistically significant (p < 10−16) predictors of the mean number identity by descent segments. A likelihood ratio test and model selection with AIC both had significant region by distance interactions, indicating regions differed in the rate of distance decay in haplotype sharing (likelihood ratio test across all bins, p < 10−16; AIC: 1–3 cM, model1 with interaction: 4.7365e+08, model2 without interaction: 4.7674e+08; 3–5 cM model1: 6.6756e+08, model2: 6.7161e+08; > 5 cM model1: 8.1954e+08, model2: 8.2428e+08). The distance decay within haplotype sharing was particularly strong in East Africa despite having the smallest geographic extent. This pattern may reflect reduced effective migration rates or reduced gene flow between highland and lowland landraces in Ethiopia primarily aided by adaptation to higher altitudes as described in a recent pangenome study (Morris et al. 2025).

FIGURE 4.

FIGURE 4

Open in a new tab

(A) Identity by descent (IBD) haplotype sharing between pairs of landraces averaged over all single chromosome pairs between individuals (log scale), compared to geographic distance. Landraces with no shared ancestry are excluded. Panels separate different IBD length bins, which correspond to relationships at different timescales in the past. The 1–3 cM bin represents older relationships, 3–5 cM intermediate and > 5 cM recent relationships. The trend lines correspond to predicted weighted poisson regression values for each region across all pairs (weight by sample size). (B) Variance partitioning for predictor variables moisture, temperature, geographic distance, language across identity by descent length bins. The adjusted R2 value by each set of predictor variables represents the degree of shared ancestry explained solely by that category of predictor. Ellipse size corresponds to the proportion of variation explained by variables in each category, with the ellipse overlap indicating collinear variation in predictors that explained variation in shared ancestry.

Distance decay in haplotype sharing was strongest for > 5 cM segments followed by 3–5 cM and 1–3 cM as the degree of shared ancestry (i.e., the average number of IBD segments per pair) increased over time. Increased geographic decay among recent relationships relative to older relationships is consistent with migration models where the geographic distance from an individual's genetic ancestor is expected to increase at a rate of √t, looking backward in time (Kelleher et al. 2016; Ralph and Coop 2013). In other words, genotypes are more likely to have genetic ancestors in close geographic proximity in recent times than in the distant past.

3.6. Geography, Language and Climate Explain Sorghum Genetic Variation

To assess the extent to which geographic distance (or travel time), language and climate explained genetic variation in landraces, we used variance partitioning with redundancy analysis. We found most SNP variation was individually explained by geographic distance and language (R 2 adj = 0.223 and 0.107, respectively), followed by temperature and moisture (R 2 adj = 0.068 and 0.057, respectively) (Figure 5). Together, the climatic variables explained 9% of genetic variation, while geographic space collinear with language explained 24% of SNP variation. Temperature collinear with geographic distance as well as moisture collinear with geographic distance explained 23% of SNP variation. Using travel time in place of geographic distance gave similar results: travel time and language (R 2 adj = 0.228 and 0.107, respectively) individually explained the most genetic variation in sorghum landraces, followed by temperature and moisture (R 2 adj = 0.068 and 0.057, respectively) (Figure S3). Most SNP variation (25%) was explained by travel time collinear with language, followed by moisture collinear with travel time (R 2 adj = 0.242) and temperature collinear with travel time (R 2 adj = 0.239). Moisture collinear with temperature explained 9% of SNP variation. Comparatively, the travel time model explained 27% of genetic variation in sorghum landraces, while geographic distance model explained 26%.

FIGURE 5.

FIGURE 5

Open in a new tab

Variance partitioning for sets of predictor variables moisture, temperature, geographic distance, language using SNPs as the multivariate response variables. Ellipse size corresponds to the proportion of variation explained by variables in each category, with the ellipse overlap indicating collinear variation in predictors that explained variation in SNPs.

When we considered the relationship between genomic variation in these predictors across time, we found the relative explanatory power of individual and combined factors increased for more recent relationships, with 41%, 46% and 48% of shared ancestry explained by predictors for the 1–3, 3–5 and > 5 cM bins, respectively (Figure 4B). For simplification, we will use 1, 3 and 5 to represent 1–3, 3–5 and > 5 cM bins below. Similar to our variance partitioning analysis utilising SNPs as the multivariate response factor (Figure 5), geographic distance (1: 34%, 3: 39%, 5: 41%) followed by language (1: 15%, 3: 18%, 5: 19%) independently explained the most shared ancestry across all timescales. When considering the effect of climate on shared ancestry, we found that temperature explained more shared ancestry than moisture across all timescales: (1: 10%, 3: 11%, 5: 12%) (1: 8%, 3: 10%, 5: 10%), for temperature and moisture respectively. The degree of shared ancestry explained by collinearity between geographic distance and language increased over time (1: 38%, 3: 43%, 5: 45%), which is comparable to moisture collinear with geographic distance (1: 37%, 3: 42%, 5: 44%).

4. Discussion

Understanding the patterns of geographic structuring and genetic relatedness across space and time as well as the relative contribution of factors to genetic diversity may allow us to uncover sources of genetic variation, which may help the conservation of diversity for improvement of staple crops globally. Presently, crop‐related research studies have focused on reconstructing the origin and dispersal patterns of staple crops as well as the molecular, genetic and environmental mechanisms contributing to plant resilience and adaptation (Alam et al. 2021; Caproni et al. 2023; Silva et al. 2015; Zhang et al. 2021). Geographic space and climatic factors like temperature and precipitation in particular have been utilised to model gene flow patterns and explain genetic differentiation within and between populations (Gutaker et al. 2020; Lasky et al. 2015; Olatoye et al. 2018). Nevertheless, the factors that explain genetic variation outside of climate or space as well as the elements within climate and space that contribute to fine‐scale population structure have received little attention until recently. Human‐mediated gene flow, represented as human movement over landscapes, as well as cultural factors like language have gained recognition as contributors to crop genetic structure, contextualising historical diffusion of crops across space (Benz et al. 2007; Gilabert et al. 2025; Gutaker et al. 2020; Samberg et al. 2013). However, few studies have resolved the degree to which geographic space, environment, human movement and culture conjointly and independently explain genetic variation. And despite the importance of geographic space in modelling genetic relationships, it has been difficult to discern which relationships are more recent, or how the pattern of shared ancestry changes across geographic space. This study expands upon previous findings through an integrative examination of broad and fine‐scale patterns of population structure to unveil the population genomic history of sorghum. By modelling intercontinental, regional and linguistic patterns of geographic structure using paired landraces, we build upon analyses of effective migration surfaces to reveal a strong pattern of isolation by distance at a large‐scale, which when stratified by region and language families reveals further structuring. By employing identity by descent segments, we observed how the pattern of genetic relatedness across space changed over time and across regions, which expands upon models of early diffusion.

4.1. Isolation by Distance and Travel Time Explain Sorghum Diversity and Distribution

The spatial distribution of genomic variation in sorghum landraces across Africa and Eurasia (Figure 1A,B) presents a pattern of isolation by distance (Wright 1943) and travel time, where gene flow between landraces is reduced as geographic distance or travel time increases. Globally, the plateau in genetic distance at ~10,000 km may reflect the rare occurrence of long‐distance dispersal events between populations (Ibrahim et al. 1996; Jordano 2017). Contrary to our expectations, globally the geographic distance model was a better fit to sorghum genomic variation than the estimated travel time. By contrast, Gutaker et al. (2020) found that travel time was a better fit than geographic distance for japonica rice landraces (though not indica rice). However, when we compared regional models, we found the travel time model was a better fit in East Africa, West Africa and South Asia while geographic distance was a better fit in southern Africa. This result suggests that our travel time model may capture relative variation in gene flow at the within‐region level, and that this model performed better at modelling shorter travel routes. While improvements could be implemented in our travel time model, other biological or cultural factors like variety‐specificity and/or differences in geographic constraint among linguistic groups may explain why there are scale‐dependent differences in model fit. Limitations for the travel time model could be related to the implementation of waypoints and more specifically the exclusion of waypoints in South Asia, which may have better represented human movement between continents. Future modifications may involve a systematic division of conductance layers into demes and the inclusion of a desert‐specific travel model.

4.2. Sorghum Genetic Structure Explained by Hierarchical Linguistic Differences

We found evidence of linguistic structuring within language families and individual languages. Language families, which may be a proxy for cultural preferences and cultivation practices as well as gene flow mediated by shared culture, explained up to 4% of genetic variation in both our estimate of hierarchical F ST and multicontinental F ST. Individual languages explain 13% of total genetic variation and 10% of total SNP variation within language families (Figure 3A). Westengen et al. (2014) and more recently, Gilabert et al. (2025) provided systematic evidence for ethnolinguistic population structure, but only within Africa. Linguistic structuring of sorghum landraces in Africa was observed among language families and subclasses (Gilabert et al. 2025). Our analysis of hierarchical F ST corroborates these findings across multiple continents and language families. In other words, a majority of SNP variation explained by language families can also be found within individual languages but there is also a small portion within language families that only individual languages can explain, which suggests that individual languages explain more genetic diversity than language families. This difference may be attributable to differences in granularity between individual languages and language families. For example, a large geographically widespread language family like Atlantic‐Congo has a diversity of individual languages which in themselves may exhibit differing patterns of spatial structuring (i.e., variation in the degree of genetic differentiation based on geographic proximity). Thus, the SNP variation explained by language families, particularly for larger language families with a large geographic extent, may not fully capture fine‐scale population structure.

Like population genomic similarity, researchers have also found that language variation is spatially autocorrelated (Barbieri et al. 2022). Linguistic spatial autocorrelation may thus drive differences in geographic structure across language families due to selection and shared gene flow of sorghum driven by human groups sharing language. Our analysis of spatial structuring across the language families with the largest sample sizes showed that the amount of genetic diversity, geographic extent and distribution across space differs despite the consistent isolation by distance pattern (Figure 3B). By contrast, sorghum landraces attributed to the Mande language family exhibited evidence of genetic divergence, with two distinct genetic clusters; this division may be a result of reproductive isolation between the highly diverged guinea margaritiferum variety and the more common guinea and durra variety in West Africa, which we elaborated below.

There are some limitations to assigning the geographically closest language (family) to a landrace, given that there are multilingual regions where sorghum has evolved alongside and followed the migration of linguistic groups. For example, Ethiopia functions as a multilingual case study, where Sudanese migrants from the Chari‐Nile language group in western Ethiopia with a preference for the caudatum variety coincide with members of the Afro‐Asiatic family, who prefer durra (Harlan and Stemler 1976; Kimber 2000). On the other hand, the biogeographic distribution of a southern sorghum population follows the migration pattern of Bantu farmers associated with the Niger‐Congo family, from East Africa to the south, demonstrating how the expansion of a linguistic group coincides with the history of sorghum (Westengen et al. 2014). These examples reflect the complex history and codistribution of language and domesticated sorghum; thus, our model is a static representation of this dynamic as spatiotemporal changes are not necessarily considered. Despite the simplicity of language family groupings in analyses of population structure, they can still be utilised to assess differing patterns of spatial structuring on a global scale. And while we only considered language as a proxy for culture, other factors like ethnicity (Faye et al. 2019) and farmer social dynamics may play a role (Leclerc and d'Eeckenbrugge 2012; McGuire 2007). Future projects may consider implementing these factors to represent cultural preference at any geographic scale. Ultimately, our findings show that language can be a predictor for sorghum genetic structure and that language families in themselves exhibit differing patterns of isolation by distance.

4.3. Spatiotemporal Differences in Genetic and Geographic Structure

Although landraces exhibited a pattern of isolation by distance and travel time across all regions, we observed considerable regional variation in patterns of genetic differentiation across space (Figures 1 and 2), which suggests there have been regional differences in the processes driving population differentiation. Among the considered subregions, some West African landraces exhibited high genetic divergence, which may be evidence of reproductive isolation. Gilabert et al. (2025) and Morris et al. (2013) identified a highly genetically divergent clade within West Africa associated with the guinea margaritiferum variety, which may have been domesticated independently or carry introgression from wild sorghum. Our assessment of regional and linguistic differences in spatial structuring was consistent with these findings (Figures 1C, 2 and 3B). The sorghum landraces assigned to the Mande language family, which were found mostly in Southwest Mali and Sierra Leone, included two highly divergent groups of genotypes (one of which represented guinea margaritiferum, Figure 3) and those in Sierra Leone showed high differentiation in comparison to other landraces only ~400 km away (Figure 2).

South Asian landraces also exhibited substantial geographic genetic divergence, which may follow the East–West division of the Indo‐European and Dravidian language families. In wavelet analysis, we found strong genetic differentiation at the ~400 km scale in eastern India, where guinea varieties in the wetter regions dominate but then give way to durra adapted to drier climates (Figures 2 and 3) (Lasky et al. 2015; Morris et al. 2013).

Landraces in East Africa showed some of the strongest patterns of isolation by distance, reflected in the relatively steep increase in genetic distance from 0 to 250 h travel time, and the peak in wavelet genetic differentiation at ~400 km scales (Figures 1 and 2). These patterns may be caused by Ethiopia's diverse topography, climate, linguistic groups, limiting gene flow across space and potentially selecting for distinct genotypes in different regions. As a hotspot for sorghum diversity, admixture within East Africa may have been bolstered by cultural exchange (e.g., seed sharing and emigration) (Menamo et al. 2021), which could have impacted the degree of genetic differentiation.

In our assessment of temporal differences in spatial structuring, we observed regional differences in the relationship between geographic proximity and shared ancestry over time (Figure 4A). Our findings suggest that landraces that are geographically proximate are more likely to share genetic ancestry than those that are more distant. In East Africa, the steep pattern of geographic decay may be attributable to differences in genetic structure and adaptation along altitudinal gradients in Ethiopia, as found in previous studies in sorghum (Gilabert et al. 2025; Morris et al. 2025) as well as in barley (Hadado et al. 2010). The conjoint effects of altitude and differential farmer practices also exhibited strong genetic differentiation in Ethiopian barley (Samberg et al. 2013), which could play a role in this context. By contrast, many West African landraces presented a notable pattern of high shared ancestry for many pairs of landraces at larger distances, which may be attributable to the rapid spread of some sorghum genotypes across long latitudinal belts of similar climate in this region (Morris et al. 2013). Over time with longer chromosomal segments, the degree of shared ancestry increased across all regions, particularly in the > 5 cM bin, a pattern that is expected under limited migration models where distance from an ancestor increases backwards through time (Kelleher et al. 2016; Ralph and Coop 2013). Greater shared ancestry at closer distances in recent times may have also been a result of limited dispersal, local adaptation to climates and differential preferences among communities, which have likely strengthened the pattern of isolation by distance among more recent relationships. Regional differences in the pattern of genetic relatedness across space and time may further contextualise the regional differences in the pattern of spatial structuring we observed (Figure 1C).

Our variance partitioning analysis showed that the relative explanatory power of geographic distance, climate and language as well as their conjoint effects increased when comparing larger haplotypes representing more recent relationships (Figure 4B). Consistent with our assessment of the effect of these predictors on SNP variation (Figure 5), we found that geographic distance and language consistently explained the most identity by descent sharing over time.

The explanatory power of individual and conjoint factors was relatively similar over time, suggesting the processes determining large scale population structure have been largely consistent through time. However, the degree of shared ancestry explained by temperature was slightly higher than moisture across all timescales and increased over time, whereas the explanatory power of moisture remained the same in the intermediate and recent timescales. Additionally, geography collinear with language was higher than geography collinear with moisture over time. This suggests that there are some slight differences in the importance of variables over time. For example, in the earliest timescale, geography collinear with language may have been a more important contributor to shared ancestry than geography collinear with moisture due to limited dispersal favouring communities sharing the same language family, and reduced gene flow between different climates in the early diffusion period of sorghum landraces.

4.4. Multicollinearity of Spatial and Human‐Related Factors Explain Most Genetic Variation

Our variance partitioning analysis suggests that geographic space and travel time individually explained the most genetic variation in sorghum landraces (Figure 5). However, 24% and 25% of genomic variation was explained by collinearity between language and space or travel time, respectively, indicating that detailed study of human history and preferences may be required to better disentangle culture from isolation by distance to further contextualise crop spatial dynamics. We also observed 24% collinearity between geographic distance and the climatic factors moisture and temperature, which aligns with findings in previous studies that consider space and environment (Faye et al. 2019; Lasky et al. 2015; Olatoye et al. 2018). These findings suggest that SNP variation in multicontinental sorghum landraces can be explained by geography and climate as established regionally in Nigerian (Olatoye et al. 2018) and Senegalese (Faye et al. 2019) sorghum, as well as globally (Lasky et al. 2015). Collinearity between geographic distance and climate may be partially explained by a pattern of isolation by distance and spatial autocorrelation in climate conditions. Here we found that geographic distance and language explained more genetic variation than climate alone or space and climate, highlighting the importance of components of human culture like language in sorghum population structure.

5. Conclusion

Climate change in certain regions may limit the variety of crops that can be cultivated, making drought‐tolerant crops like sorghum essential for future agriculture (Rosenow et al. 1983; Tuinstra et al. 1996). Understanding the factors that influence dispersal patterns of staple crops may allow us to predict how barriers to gene flow, like climate, alter the frequency and distribution of genomic diversity across space, particularly under temporal, environmental and habitat variability. In turn, these insights may reveal potential sources of diversity for breeding climate‐resilient crops (Campbell, Castañeda‐Álvarez, et al. 2025). Furthermore, the patterns identified here help place in context knowledge developed about the history of specific adaptive quantitative trait loci (Morris et al. 2025). Our assessment of the ecological and environmental factors that affect sorghum diversity suggests that spatial proximity, human movement, culture and climate differentially impact genetic structure relative to each other as well as over space and time. The pattern of isolation by distance and travel time indicates that gene flow between landraces is limited by geographic distance and human‐mediated dispersal, particularly between more distant varieties. We also observed considerable variation in spatial structuring across language families and regions and time, which highlights the importance of these factors in determining genetic variation. Crop breeding and/or conservation efforts may consider the effects of culture and language as partial predictors for genetic diversity. Additionally, this study may encourage researchers to explore beyond language families, as more genetic variation may lie within languages alone. By leveraging nearly 2000 GBS georeferenced sorghum landraces across a large geographic range and associated climatic data we were able to assess population structure across different spatial and temporal scales, highlighting the utility of agricultural landscape genomics (Campbell, Bedford, et al. 2025). Ultimately, this research may provide a historical framework for sorghum geneticists and further understanding of evolutionary drivers of sorghum diversity.

Author Contributions

Eleanna Vasquez Cerda designed the study, analysed the data and wrote the paper. Jesse Lasky designed the study, analysed the data and contributed to writing. Emily Bellis, Aayudh Das, Emma Slayton and Geoffrey Morris contributed to design, interpretation and writing.

Funding

This work was supported by the National Science Foundation (DGE1255832), Gates Foundation (INV‐030574) and National Institutes of Health (R35GM138300).

Conflicts of Interest

The authors declare no conflicts of interest.

Supporting information

Figure S1: Scatterplot of travel time and genetic distance 1806 landraces.

Figure S2: Scatterplot of travel time in subregions East Africa (R 2 = 0.140), South Africa (R 2 = 0.131), West Africa (R 2 = 0.111) and South Asia (R 2 = 0.078).

Figure S3: Variance partitioning for predictor variables moisture, temperature, travel time, language using SNPs as the multivariate response factor.

Figure S4: Map of 1806 landraces and associated 42 unique language families.

Figure S5: Neighbour‐joining tree depicting genetic relatedness (Hamming distance) among West African landraces.

Figure S6: Scatterplot of geographic distance and genetic distance for 11 varieties present among 491 West African landraces without guinea margaritiferum varieties.

Figure S7: Scatterplot of geographic distance and genetic distance for four varieties present in West Africa.

Figure S8:Scatterplot of geographic distance (km) and travel time (hours) for paired sorghum accessions. Points are stratified by continent.

Table S1: Metadata/coordinates for waypoints used in travel time calculations.

MEC-35-e70287-s001.pdf (2.7MB, pdf)

Acknowledgements

Travel time calculations were run using Bridges‐2 resources from the Pittsburgh Supercomputing Center through allocation BIO230186 from the Advanced Cyberinfrastructure Coordination Ecosystem: Services & Support (ACCESS) program, which is supported by National Science Foundation grants #2138259, #2138286, #2138307, #2137603 and #2138296. This work was supported by the National Science Foundation Graduate Research Fellowship Program under Grant No. DGE1255832 to EEVC, Gates Foundation grant ‘Mining useful alleles for climate change adaptation from CGIAR gene banks (INV‐030574)’ to G.P.M. and J.R.L., and by the National Institutes of Health grant R35GM138300 to J.R.L.

Data Availability Statement

GBS data from Hu et al. (2019) can be found in the DRYAD repository (https://doi.org/10.5061/dryad.63h8fd4). A zipped file with the data and scripts pertaining to language family, environment, and SNP data is available in the DRYAD repository (https://doi.org/10.5061/dryad.37pvmcw0f).

References

  1. Adriaensen, F. , Chardon J. P., De Blust G., et al. 2003. “The Application of ‘Least‐Cost’ Modelling as a Functional Landscape Model.” Landscape and Urban Planning 64, no. 4: 233–247. 10.1016/S0169-2046(02)00242-6. [DOI] [Google Scholar]
  2. Alam, O. , Gutaker R. M., Wu C.‐C., et al. 2021. “Genome Analysis Traces Regional Dispersal of Rice in Taiwan and Southeast Asia.” Molecular Biology and Evolution 38, no. 11: 4832–4846. 10.1093/molbev/msab209. [DOI] [PMC free article] [PubMed] [Google Scholar]
  3. Al‐Asadi, H. , Petkova D., Stephens M., and Novembre J.. 2019. “Estimating Recent Migration and Population‐Size Surfaces.” PLoS Genetics 15, no. 1: e1007908. 10.1371/journal.pgen.1007908. [DOI] [PMC free article] [PubMed] [Google Scholar]
  4. Atinkut, B. , and Mebrat A.. 2016. “Determinants of Farmers Choice of Adaptation to Climate Variability in Dera Woreda, South Gondar Zone, Ethiopia.” Environmental Systems Research 5, no. 1: 6. 10.1186/s40068-015-0046-x. [DOI] [Google Scholar]
  5. Auer, P. , Hilpert M., Stukenbrock A., and Szmrecsanyi B.. 2013. Space in Language and Linguistics: Geographical, Interactional, and Cognitive Perspectives. De Gruyter Brill. [Google Scholar]
  6. Auffret, A. G. , Berg J., and Cousins S. A. O.. 2014. “The Geography of Human‐Mediated Dispersal.” Diversity and Distributions 20, no. 12: 1450–1456. 10.1111/ddi.12251. [DOI] [Google Scholar]
  7. Bannayan, M. , Sadeghi Lotfabadi S., Sanjani S., Mohamadian A., and Aghaalikhani M.. 2011. “Effects of Precipitation and Temperature on Crop Production Variability in Northeast Iran.” International Journal of Biometeorology 55, no. 3: 387–401. 10.1007/s00484-010-0348-7. [DOI] [PubMed] [Google Scholar]
  8. Barbieri, C. , Blasi D. E., Arango‐Isaza E., et al. 2022. “A Global Analysis of Matches and Mismatches Between Human Genetic and Linguistic Histories.” Proceedings of the National Academy of Sciences of the United States of America 119, no. 47: e2122084119. 10.1073/pnas.2122084119. [DOI] [PMC free article] [PubMed] [Google Scholar]
  9. Bauman, D. , Drouet T., Fortin M.‐J., and Dray S.. 2018. “Optimizing the Choice of a Spatial Weighting Matrix in Eigenvector‐Based Methods.” Ecology 99, no. 10: 2159–2166. 10.1002/ecy.2469. [DOI] [PubMed] [Google Scholar]
  10. Benz, B. , Perales H., and Brush S.. 2007. “Tzeltal and Tzotzil Farmer Knowledge and Maize Diversity in Chiapas, Mexico.” Current Anthropology 48, no. 2: 289–300. [Google Scholar]
  11. Bohonak, A. J. 1999. “Dispersal, Gene Flow, and Population Structure.” Quarterly Review of Biology 74, no. 1: 21–45. http://www.jstor.org.ezaccess.libraries.psu.edu/stable/2664376. [DOI] [PubMed] [Google Scholar]
  12. Boivin, N. , Crassard R., and Petraglia M.. 2017. Human Dispersal and Species Movement: From Prehistory to the Present. Cambridge University Press. 10.1017/9781316686942. [DOI] [Google Scholar]
  13. Browning, B. L. , and Browning S. R.. 2013. “Improving the Accuracy and Efficiency of Identity‐By‐Descent Detection in Population Data.” Genetics 194, no. 2: 459–471. 10.1534/genetics.113.150029. [DOI] [PMC free article] [PubMed] [Google Scholar]
  14. Browning, B. L. , Tian X., Zhou Y., and Browning S. R.. 2021. “Fast Two‐Stage Phasing of Large‐Scale Sequence Data.” American Journal of Human Genetics 108, no. 10: 1880–1890. 10.1016/j.ajhg.2021.08.005. [DOI] [PMC free article] [PubMed] [Google Scholar]
  15. Browning, S. R. , and Browning B. L.. 2012. “Identity by Descent Between Distant Relatives: Detection and Applications.” Annual Review of Genetics 46, no. 1: 617–633. 10.1146/annurev-genet-110711-155534. [DOI] [PubMed] [Google Scholar]
  16. Brun, P. , Zimmermann N. E., Hari C., Pellissier L., and Karger D. N.. 2022. “CHELSA‐BIOCLIM+ A Novel Set of Global Climate‐Related Predictors at Kilometre‐Resolution.” EnviDat. 10.16904/envidat.332. [DOI] [Google Scholar]
  17. Bullock, J. M. , Bonte D., Pufal G., et al. 2018. “Human‐Mediated Dispersal and the Rewiring of Spatial Networks.” Trends in Ecology & Evolution 33, no. 12: 958–970. 10.1016/j.tree.2018.09.008. [DOI] [PubMed] [Google Scholar]
  18. Campbell, Q. , Bedford J. A., Yu Y., et al. 2025. “Agricultural Landscape Genomics to Increase Crop Resilience.” Plant Communications 6, no. 2: 101260. 10.1016/j.xplc.2025.101260. [DOI] [PMC free article] [PubMed] [Google Scholar]
  19. Campbell, Q. , Castañeda‐Álvarez N., Domingo R., et al. 2025. “Prioritizing Parents From Global Genebanks to Breed Climate‐Resilient Crops.” Nature Climate Change 15, no. 6: 673–681. 10.1038/s41558-025-02333-x. [DOI] [Google Scholar]
  20. Caproni, L. , Lakew B. F., Kassaw S. A., et al. 2023. “The Genomic and Bioclimatic Characterization of Ethiopian Barley ( Hordeum vulgare L.) Unveils Challenges and Opportunities to Adapt to a Changing Climate.” Global Change Biology 29, no. 8: 2335–2350. 10.1111/gcb.16560. [DOI] [PubMed] [Google Scholar]
  21. Clarke, R. T. , Rothery P., and Raybould A. F.. 2002. “Confidence Limits for Regression Relationships Between Distance Matrices: Estimating Gene Flow With Distance.” Journal of Agricultural, Biological, and Environmental Statistics 7, no. 3: 361–372. 10.1198/108571102320. [DOI] [Google Scholar]
  22. Coelho, M. T. P. , Barreto E., Rangel T. F., et al. 2023. “The Geography of Climate and the Global Patterns of Species Diversity.” Nature 622, no. 7983: 537–544. 10.1038/s41586-023-06577-5. [DOI] [PMC free article] [PubMed] [Google Scholar]
  23. Danecek, P. , Auton A., Abecasis G., et al. 2011. “The Variant Call Format and VCFtools.” Bioinformatics 27, no. 15: 2156–2158. 10.1093/bioinformatics/btr330. [DOI] [PMC free article] [PubMed] [Google Scholar]
  24. Diamond, J. , and Bellwood P.. 2003. “Farmers and Their Languages: The First Expansions.” Science 300, no. 5619: 597–603. 10.1126/science.1078208. [DOI] [PubMed] [Google Scholar]
  25. Ellen, R. , and Platten S.. 2011. “The Social Life of Seeds: The Role of Networks of Relationships in the Dispersal and Cultural Selection of Plant Germplasm.” Journal of the Royal Anthropological Institute 17, no. 3: 563–584. 10.1111/j.1467-9655.2011.01707.x. [DOI] [Google Scholar]
  26. Ellstrand, N. C. 2014. “Is Gene Flow the Most Important Evolutionary Force in Plants?” American Journal of Botany 101, no. 5: 737–753. 10.3732/ajb.1400024. [DOI] [PubMed] [Google Scholar]
  27. Faye, J. M. , Maina F., Hu Z., Fonceka D., Cisse N., and Morris G. P.. 2019. “Genomic Signatures of Adaptation to Sahelian and Soudanian Climates in Sorghum Landraces of Senegal.” Ecology and Evolution 9, no. 10: 6038–6051. 10.1002/ece3.5187. [DOI] [PMC free article] [PubMed] [Google Scholar]
  28. Gao, X. , and Martin E. R.. 2009. “Using Allele Sharing Distance for Detecting Human Population Stratification.” Human Heredity 68, no. 3: 182–191. 10.1159/000224638. [DOI] [PMC free article] [PubMed] [Google Scholar]
  29. Gilabert, A. , Deu M., Champion L., et al. 2025. “Integrating Ethnolinguistic and Archaeobotanical Data to Uncover the Origin and Dispersal of Cultivated Sorghum in Africa: A Genomic Perspective.” Peer Community Journal 5: e96. 10.24072/pcjournal.613. [DOI] [Google Scholar]
  30. Goodchild, M. F. 2020. “Beyond Tobler's Hiking Function.” Geographical Analysis 52, no. 4: 558–569. 10.1111/gean.12253. [DOI] [Google Scholar]
  31. Goudet, J. 2005. “Hierfstat, a Package for r to Compute and Test Hierarchical F‐Statistics.” Molecular Ecology Notes 5, no. 1: 184–186. 10.1111/j.1471-8286.2004.00828.x. [DOI] [Google Scholar]
  32. Gowen, K. M. , and de Smet T. S.. 2020. “Testing Least Cost Path (LCP) Models for Travel Time and Kilocalorie Expenditure: Implications for Landscape Genomics.” PLoS One 15, no. 9: e0239387. 10.1371/journal.pone.0239387. [DOI] [PMC free article] [PubMed] [Google Scholar]
  33. Gutaker, R. M. , Groen S. C., Bellis E. S., et al. 2020. “Genomic History and Ecology of the Geographic Spread of Rice.” Nature Plants 6, no. 5: 492–502. 10.1038/s41477-020-0659-6. [DOI] [PubMed] [Google Scholar]
  34. Hadado, T. T. , Rau D., Bitocchi E., and Papa R.. 2010. “Adaptation and Diversity Along an Altitudinal Gradient in Ethiopian Barley (Hordeum vulgare L.) Landraces Revealed by Molecular Analysis.” BMC Plant Biology 10: 121. [DOI] [PMC free article] [PubMed] [Google Scholar]
  35. Hammarström, H. , Forkel R., Haspelmath M., and Bank S.. 2022. Glottolog 4.7. Max Planck Institute for Evolutionary Anthropology. 10.5281/zenodo.15525265. [DOI] [Google Scholar]
  36. Hamming, R. W. 1950. “Error Detecting and Error Correcting Codes.” Bell System Technical Journal 29, no. 2: 147–160. 10.1002/j.1538-7305.1950.tb00463.x. [DOI] [Google Scholar]
  37. Harlan, J. R. , and Stemler A.. 1976. “The Races of Sorghum in Africa.” In Origins of African Plant Domestication, 1976, edited by Harlan J. R., De Wet J. M. J., and Stemler A. B. L., 465–478. De Gruyter Mouton. 10.1515/9783110806373.465. [DOI] [Google Scholar]
  38. Hijmans, R. J. 2024. “Geosphere: Spherical Trigonometry (R Package Version 1.5–20).” https://github.com/rspatial/geosphere.
  39. Hijmans, R. J. , Cameron S. E., Parra J. L., Jones P. G., and Jarvis A.. 2005. “Very High Resolution Interpolated Climate Surfaces for Global Land Areas.” International Journal of Climatology 25, no. 15: 1965–1978. 10.1002/joc.1276. [DOI] [Google Scholar]
  40. Hu, Z. , Olatoye M. O., Marla S., and Morris G. P.. 2019. “An Integrated Genotyping‐By‐Sequencing Polymorphism Map for Over 10,000 Sorghum Genotypes.” Plant Genome 12, no. 1: 180044. 10.3835/plantgenome2018.06.0044. [DOI] [PMC free article] [PubMed] [Google Scholar]
  41. Ibrahim, K. M. , Nichols R. A., and Hewitt G. M.. 1996. “Spatial Patterns of Genetic Variation Generated by Different Forms of Dispersal During Range Expansion.” Heredity 77, no. 3: 282–291. 10.1038/hdy.1996.142. [DOI] [Google Scholar]
  42. Irwin, G. , Bickler S., and Quirke P.. 1990. “Voyaging by Canoe and Computer: Experiments in the Settlement of the Pacific Ocean.” Antiquity 64: 34–50. [Google Scholar]
  43. Jordano, P. 2017. “What Is Long‐Distance Dispersal? And a Taxonomy of Dispersal Events.” Journal of Ecology 105, no. 1: 75–84. 10.1111/1365-2745.12690. [DOI] [Google Scholar]
  44. Kassambara, A. , and Mundt F.. 2020. “Factoextra: Extract and Visualize the Results of Multivariate Data Analyses.” https://github.com/kassambara/factoextra.
  45. Kelleher, J. , Etheridge A. M., Véber A., and Barton N. H.. 2016. “Spread of Pedigree Versus Genetic Ancestry in Spatially Distributed Populations.” Theoretical Population Biology 108: 1–12. 10.1016/j.tpb.2015.10.008. [DOI] [PubMed] [Google Scholar]
  46. Kimber, C. T. 2000. “Origins of Domesticated Sorghum and Its Early Diffusion to India and China.” In Sorghum: Origin, History, Technology and Production, edited by Smith W.. John Wiley & Sons, Inc. [Google Scholar]
  47. Kimura, M. , and Weiss G. H.. 1964. “The Stepping Stone Model of Population Structure and the Decrease of Genetic Correlation With Distance.” Genetics 49, no. 4: 561–576. 10.1093/genetics/49.4.561. [DOI] [PMC free article] [PubMed] [Google Scholar]
  48. Lacy, S. M. , Cleveland D. A., and Soleri D.. 2006. “Farmer Choice of Sorghum Varieties in Southern Mali.” Human Ecology 34, no. 3: 331–353. 10.1007/s10745-006-9021-5. [DOI] [Google Scholar]
  49. Lasky, J. R. , Takou M., Gamba D., and Keitt T. H.. 2024. “Estimating Scale‐Specific and Localized Spatial Patterns in Allele Frequency.” Genetics 227, no. 3: iyae082. 10.1093/genetics/iyae082. [DOI] [PMC free article] [PubMed] [Google Scholar]
  50. Lasky, J. R. , Upadhyaya H. D., Ramu P., et al. 2015. “Genome‐Environment Associations in Sorghum Landraces Predict Adaptive Traits.” Science Advances 1, no. 6: 1–13. 10.1126/sciadv.1400218. [DOI] [PMC free article] [PubMed] [Google Scholar]
  51. Lawson, D. J. , Hellenthal G., Myers S., and Falush D.. 2012. “Inference of Population Structure Using Dense Haplotype Data.” PLoS Genetics 8, no. 1: e1002453. 10.1371/journal.pgen.1002453. [DOI] [PMC free article] [PubMed] [Google Scholar]
  52. Leclerc, C. , and d'Eeckenbrugge G. C.. 2012. “Social Organization of Crop Genetic Diversity. The G × E × S Interaction Model.” Diversity 4, no. 1: 1–32. 10.3390/d4010001. [DOI] [Google Scholar]
  53. Legendre, P. , and Anderson M. J.. 1999. “Distance‐Based Redundancy Analysis: Testing Multispecies Responses in Multifactorial Ecological Experiments.” Ecological Monographs 69, no. 1: 1–24. 10.1890/0012-9615(1999)069[0001:DBRATM]2.0.CO;2. [DOI] [Google Scholar]
  54. Marla, S. R. , Olatoye M., Davis M., et al. 2025. “Mining Sorghum Pangenome Enabled Identification of New dw3 Alleles for Breeding Stable‐Dwarfing Hybrids.” G3 (Bethesda, Md.) 15, no. 5: jkaf054. 10.1093/g3journal/jkaf054. [DOI] [PMC free article] [PubMed] [Google Scholar]
  55. McGuire, S. J. 2007. “Vulnerability in Farmer Seed Systems: Farmer Practices for Coping With Seed Insecurity for Sorghum in Eastern Ethiopia.” Economic Botany 61, no. 3: 211–222. 10.1663/0013-0001(2007)61[211:VIFSSF]2.0.CO;2. [DOI] [Google Scholar]
  56. Menamo, T. , Kassahun B., Borrell A. K., et al. 2021. “Genetic Diversity of Ethiopian Sorghum Reveals Signatures of Climatic Adaptation.” Theoretical and Applied Genetics 134, no. 2: 731–742. 10.1007/s00122-020-03727-5. [DOI] [PubMed] [Google Scholar]
  57. Morris, G. P. , Harder A. M., and Jerry W.. 2025. “Developing Future Resilience From Signatures of Adaptation Across the Sorghum Pangenome.” BioRxiv Preprint. 10.1101/2025.08.01.667986. [DOI] [Google Scholar]
  58. Morris, G. P. , Ramu P., Deshpande S. P., et al. 2013. “Population Genomic and Genome‐Wide Association Studies of Agroclimatic Traits in Sorghum.” Proceedings of the National Academy of Sciences of the United States of America 110, no. 2: 453–458. 10.1073/pnas.1215985110. [DOI] [PMC free article] [PubMed] [Google Scholar]
  59. Murdock, G. P. 1959. Africa: Its Peoples and Their Culture History. McGraw‐Hill Book Company. https://babel.hathitrust.org/cgi/pt?id=uc1.b4346933&seq=9. [Google Scholar]
  60. Nei, M. 1972. “Genetic Distance Between Populations.” American Naturalist 106, no. 949: 283–292. 10.1086/282771. [DOI] [Google Scholar]
  61. Ochoa, A. , and Storey J. D.. 2021. “Estimating FST and Kinship for Arbitrary Population Structures.” PLoS Genetics 17, no. 1: e1009241. 10.1371/journal.pgen.1009241. [DOI] [PMC free article] [PubMed] [Google Scholar]
  62. OECD . 2014. “Consensus Document on the Biology of Sorghum (Sorghum bicolor (L.) Moench).” Series on Harmonisation of Regulatory Oversight in Biotechnology 62, no. 57: 1–27. [Google Scholar]
  63. Ohadi, S. , Hodnett G., Rooney W., and Bagavathiannan M.. 2017. “Gene Flow and Its Consequences in Sorghum spp.” Critical Reviews in Plant Sciences 36, no. 5–6: 367–385. 10.1080/07352689.2018.1446813. [DOI] [Google Scholar]
  64. Oksanen, J. , Blanchet F. G., Kindt R., et al. 2012. “Vegan: Community Ecology Package.”
  65. Olatoye, M. O. , Hu Z., Maina F., and Morris G. P.. 2018. “Genomic Signatures of Adaptation to a Precipitation Gradient in Nigerian Sorghum.” G3: Genes, Genomes, Genetics 8, no. 10: 3269–3281. 10.1534/g3.118.200551. [DOI] [PMC free article] [PubMed] [Google Scholar]
  66. Perales, H. R. , Benz B. F., and Brush S. B.. 2005. “Maize Diversity and Ethnolinguistic Diversity in Chiapas, Mexico.” Proceedings of the National Academy of Sciences of the United States of America 102, no. 3: 949–954. 10.1073/pnas.0408701102. [DOI] [PMC free article] [PubMed] [Google Scholar]
  67. Peterman, W. E. 2018. “ResistanceGA: An R Package for the Optimization of Resistance Surfaces Using Genetic Algorithms.” Methods in Ecology and Evolution 9, no. 6: 1638–1647. 10.1111/2041-210X.12984. [DOI] [Google Scholar]
  68. Purcell, S. , Neale B., Todd‐Brown K., et al. 2007. “PLINK: A Tool Set for Whole‐Genome Association and Population‐Based Linkage Analyses.” American Journal of Human Genetics 81, no. 3: 559–575. 10.1086/519795. [DOI] [PMC free article] [PubMed] [Google Scholar]
  69. Qingshan, L. , and Dahlberg J. A.. 2001. “Chinese Sorghum Genetic Resources.” Economic Botany 55, no. 3: 401–425. 10.1007/BF02866563. [DOI] [Google Scholar]
  70. Ralph, P. , and Coop G.. 2013. “The Geography of Recent Genetic Ancestry Across Europe.” PLoS Biology 11, no. 5: e1001555. 10.1371/journal.pbio.1001555. [DOI] [PMC free article] [PubMed] [Google Scholar]
  71. Rees, W. G. 2004. “Least‐Cost Paths in Mountainous Terrain.” Computers & Geosciences 30, no. 3: 203–209. 10.1016/j.cageo.2003.11.001. [DOI] [Google Scholar]
  72. Riha, S. J. , Wilks D. S., and Simoens P.. 1996. “Impact of Temperature and Precipitation Variability on Crop Model Predictions.” Climatic Change 32, no. 3: 293–311. 10.1007/BF00142466. [DOI] [Google Scholar]
  73. Rosenow, D. T. , Quisenberry J. E., Wendt C. W., and Clark L. E.. 1983. “Drought Tolerant Sorghum and Cotton Germplasm.” Agricultural Water Management 7, no. 1: 207–222. 10.1016/0378-3774(83)90084-7. [DOI] [Google Scholar]
  74. Rousset, F. 1997. “Genetic Differentiation and Estimation of Gene Flow From F‐Statistics Under Isolation by Distance.” Genetics 145, no. 4: 1219–1228. 10.1093/genetics/145.4.1219. [DOI] [PMC free article] [PubMed] [Google Scholar]
  75. Rousset, F. 2000. “Genetic Differentiation Between Individuals.” Journal of Evolutionary Biology 13, no. 1: 58–62. 10.1046/j.1420-9101.2000.00137.x. [DOI] [Google Scholar]
  76. Ruiz Miñano, M. , While G. M., Yang W., Burridge C. P., Salvi D., and Uller T.. 2022. “Population Genetic Differentiation and Genomic Signatures of Adaptation to Climate in an Abundant Lizard.” Heredity 128, no. 4: 271–278. 10.1038/s41437-022-00518-0. [DOI] [PMC free article] [PubMed] [Google Scholar]
  77. Samberg, L. H. , Fishman L., and Allendorf F. W.. 2013. “Population Genetic Structure in a Social Landscape: Barley in a Traditional Ethiopian Agricultural System.” Evolutionary Applications 6, no. 8: 1133–1145. 10.1111/eva.12091. [DOI] [PMC free article] [PubMed] [Google Scholar]
  78. Seo, S. N. , and Mendelsohn R.. 2008. “An Analysis of Crop Choice: Adapting to Climate Change in South American Farms.” Ecological Economics 67, no. 1: 109–116. 10.1016/j.ecolecon.2007.12.007. [DOI] [Google Scholar]
  79. Sexton, J. P. , Hangartner S. B., and Hoffmann A. A.. 2014. “Genetic Isolation by Environment or Distance: Which Pattern of Gene Flow Is Most Common?” Evolution 68, no. 1: 1–15. 10.1111/evo.12258. [DOI] [PubMed] [Google Scholar]
  80. Silva, F. , Stevens C. J., Weisskopf A., et al. 2015. “Modelling the Geographical Origin of Rice Cultivation in Asia Using the Rice Archaeological Database.” PLoS One 10, no. 9: e0137024. 10.1371/journal.pone.0137024. [DOI] [PMC free article] [PubMed] [Google Scholar]
  81. Slatkin, M. 1993. “Isolation by Distance in Equilibrium and Non‐Equilibrium Populations.” Evolution 47, no. 1: 264–279. [DOI] [PubMed] [Google Scholar]
  82. Slayton, E. R. 2018. Seascape Corridors: Modeling Routes to Connect Communities Across the Caribbean Sea. Sidestone Press. [Google Scholar]
  83. Slingo, J. M. , Challinor A. J., Hoskins B. J., and Wheeler T. R.. 2005. “Introduction: Food Crops in a Changing Climate.” Philosophical Transactions of the Royal Society of London. Series B, Biological Sciences 360, no. 1463: 1983–1989. 10.1098/rstb.2005.1755. [DOI] [PMC free article] [PubMed] [Google Scholar]
  84. Sloat, L. L. , Davis S. J., Gerber J. S., et al. 2020. “Climate Adaptation by Crop Migration.” Nature Communications 11, no. 1: 1243. 10.1038/s41467-020-15076-4. [DOI] [PMC free article] [PubMed] [Google Scholar]
  85. Smith, O. , Nicholson W. V., Kistler L., et al. 2019. “A Domestication History of Dynamic Adaptation and Genomic Deterioration in Sorghum.” Nature Plants 5, no. 4: 369–379. 10.1038/s41477-019-0397-9. [DOI] [PubMed] [Google Scholar]
  86. Sork, V. L. , Davis F. W., Westfall R., et al. 2010. “Gene Movement and Genetic Association With Regional Climate Gradients in California Valley Oak ( Quercus lobata Née) in the Face of Climate Change.” Molecular Ecology 19, no. 17: 3806–3823. 10.1111/j.1365-294X.2010.04726.x. [DOI] [PubMed] [Google Scholar]
  87. Surface‐Evans, S. L. , and White D. A.. 2012. Least Cost Analysis of Social Landscapes: Archaeological Case Studies. University of Utah Press. 10.1353/book41407. [DOI] [Google Scholar]
  88. Swarup, S. , Cargill E. J., Crosby K., Flagel L., Kniskern J., and Glenn K. C.. 2021. “Genetic Diversity Is Indispensable for Plant Breeding to Improve Crops.” Crop Science 61, no. 2: 839–852. 10.1002/csc2.20377. [DOI] [Google Scholar]
  89. Tobler, W. 1993. Three Presentations on Geographical Analysis and Modeling. NCGIA, University of California. [Google Scholar]
  90. Tuinstra, M. R. , Grote E. M., Goldsbrough P. B., and Ejeta G.. 1996. “Identification of Quantitative Trait Loci Associated With Pre‐Flowering Drought Tolerance in Sorghum.” Crop Science 36, no. 5: 1337–1344. 10.2135/cropsci1996.0011183X003600050043x. [DOI] [Google Scholar]
  91. van Etten, J. 2017. “R Package Gdistance: Distances and Routes on Geographical Grids.” Journal of Statistical Software 76, no. 13: 1–21. 10.18637/jss.v076.i13.36568334 [DOI] [Google Scholar]
  92. Velásquez‐Milla, D. , Casas A., Torres‐Guevara J., and Cruz‐Soriano A.. 2011. “Ecological and Socio‐Cultural Factors Influencing In Situ Conservation of Crop Diversity by Traditional Andean Households in Peru.” Journal of Ethnobiology and Ethnomedicine 7: 40. 10.1186/1746-4269-7-40. [DOI] [PMC free article] [PubMed] [Google Scholar]
  93. Venkateswaran, K. , Elangovan M., and Sivaraj N.. 2019. “Chapter 2—Origin, Domestication and Diffusion of Sorghum bicolor .” In Woodhead Publishing Series in Food Science, Technology and Nutrition, edited by Aruna C., Visarada K. B. R. S., Bhat B. V., and Tonapi V. A. B., 15–31. Woodhead Publishing. 10.1016/B978-0-08-101879-8.00002-4. [DOI] [Google Scholar]
  94. Verón, S. R. , de Abelleyra D., and Lobell D. B.. 2015. “Impacts of Precipitation and Temperature on Crop Yields in the Pampas.” Climatic Change 130, no. 2: 235–245. 10.1007/s10584-015-1350-1. [DOI] [Google Scholar]
  95. Wang, J. , Hu Z., Upadhyaya H. D., and Morris G. P.. 2020. “Genomic Signatures of Seed Mass Adaptation to Global Precipitation Gradients in Sorghum.” Heredity 124, no. 1: 108–121. 10.1038/s41437-019-0249-4. [DOI] [PMC free article] [PubMed] [Google Scholar]
  96. Westengen, O. T. , Okongo M. A., Onek L., et al. 2014. “Ethnolinguistic Structuring of Sorghum Genetic Diversity in Africa and the Role of Local Seed Systems.” Proceedings of the National Academy of Sciences of the United States of America 111, no. 39: 14100–14105. 10.1073/pnas.1401646111. [DOI] [PMC free article] [PubMed] [Google Scholar]
  97. Wichmann, M. C. , Alexander M. J., Soons M. B., et al. 2008. “Human‐Mediated Dispersal of Seeds Over Long Distances.” Proceedings of the Royal Society B: Biological Sciences 276, no. 1656: 523–532. 10.1098/rspb.2008.1131. [DOI] [PMC free article] [PubMed] [Google Scholar]
  98. Winchell, F. , Stevens C. J., Murphy C., Champion L., and Fuller D. Q.. 2017. “Evidence for Sorghum Domestication in Fourth Millennium BC Eastern Sudan: Spikelet Morphology From Ceramic Impressions of the Butana Group.” Current Anthropology 58, no. 5: 673–683. 10.1086/693898. [DOI] [Google Scholar]
  99. Wright, S. 1943. Isolation by Distance. University of Chicago Press. [Google Scholar]
  100. Zhang, F. , Wu J., Sade N., et al. 2021. “Genomic Basis Underlying the Metabolome‐Mediated Drought Adaptation of Maize.” Genome Biology 22, no. 1: 260. 10.1186/s13059-021-02481-1. [DOI] [PMC free article] [PubMed] [Google Scholar]

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.

Supplementary Materials

Figure S1: Scatterplot of travel time and genetic distance 1806 landraces.

Figure S2: Scatterplot of travel time in subregions East Africa (R 2 = 0.140), South Africa (R 2 = 0.131), West Africa (R 2 = 0.111) and South Asia (R 2 = 0.078).

Figure S3: Variance partitioning for predictor variables moisture, temperature, travel time, language using SNPs as the multivariate response factor.

Figure S4: Map of 1806 landraces and associated 42 unique language families.

Figure S5: Neighbour‐joining tree depicting genetic relatedness (Hamming distance) among West African landraces.

Figure S6: Scatterplot of geographic distance and genetic distance for 11 varieties present among 491 West African landraces without guinea margaritiferum varieties.

Figure S7: Scatterplot of geographic distance and genetic distance for four varieties present in West Africa.

Figure S8:Scatterplot of geographic distance (km) and travel time (hours) for paired sorghum accessions. Points are stratified by continent.

Table S1: Metadata/coordinates for waypoints used in travel time calculations.

MEC-35-e70287-s001.pdf (2.7MB, pdf)

Data Availability Statement

GBS data from Hu et al. (2019) can be found in the DRYAD repository (https://doi.org/10.5061/dryad.63h8fd4). A zipped file with the data and scripts pertaining to language family, environment, and SNP data is available in the DRYAD repository (https://doi.org/10.5061/dryad.37pvmcw0f).

Articles from Molecular Ecology are provided here courtesy of Wiley

RESOURCES