Results:

Complete host specialization is rare.
We first characterized levels of host specialization observed inS. hermonthica populations across Africa. Degree of host specialization was highly variable across the 27 populations studied (Fig. 2). Populations were characterized by high emergence on either millet or sorghum only, high emergence on two hosts, or high emergence on all three hosts (Fig. 2). Hierarchical clustering with a cut-point at 0.5 showed five distinct groups: millet specialists (n=2), millet/sorghum generalists (n=3), sorghum specialists (n= 7), virulent generalists (n =3), and maize/sorghum specialists (n= 10; Fig. 2). Besides the ‘virulent generalists’, all groups included populations from at least two studies, suggesting that differences among studies did not strongly bias clustering. These results suggest that although parasites may become locally adapted to commonly cultivated host species, complete specialization is rare.
Distribution of host communities shapes specialization.
We then investigated whether the distribution of host communities across environments might shape the evolution of host specialization (Q1).S. hermonthica- prone regions covered approximately 628 million hectares, of which 80% was estimated to have nonzero production of maize, millet, or sorghum. Crop production patterns generally followed rainfall gradients, with greater land area planted to millet in more arid regions of the Sahel, sorghum dominant at lower latitudes and in eastern Sudan, and maize most common in more mesic regions (Fig. 3). In 2000, sorghum or maize were the dominant crop in most S. hermonthica- prone areas (38% sorghum; 38% maize), with 24% of areas characterized by millet as the major cereal crop.
Specialization observed in experiments was strongly associated with spatial variation in host crop communities. For parasites on all three hosts, relative host crop area harvested within a 50 km radius was a strong predictor of relative emergence in experimental studies. This finding was especially pronounced for millet(p millet<0.001; βharvest_area=0.85; GLMM; Table S2-S3). Relative crop area harvested was also a statistically significant predictor of relative emergence on sorghum (p sorghum=0.03; βharvest_area=0.25; GLMM) and maize (p maize=0.02; βharvest_area=0.25; GLMM; Table S2-S3). These results are consistent with the conclusion that parasites adapt to the most abundant host in a particular region and also suggest that relative parasite emergence is a reasonable proxy for fitness (Fig. 1).