Genomics of sorghum local adaptation to a parasitic plant

Abstract

Host-parasite coevolution can maintain high levels of genetic diversity in traits involved in species interactions. In many systems, host traits exploited by parasites are constrained by use in other functions, leading to complex selective pressures across space and time. Here, we study genome-wide variation in the staple crop Sorghum bicolor (L.) Moench and its association with the parasitic weed Striga hermonthica (Delile) Benth., a major constraint to food security in Africa. We hypothesize that geographic selection mosaics across gradients of parasite occurrence maintain genetic diversity in sorghum landrace resistance. Suggesting a role in local adaptation to parasite pressure, multiple independent loss-of-function alleles at sorghum LOW GERMINATION STIMULANT 1 (LGS1) are broadly distributed among African landraces and geographically associated with S. hermonthica occurrence. However, low frequency of these alleles within S. hermonthica-prone regions and their absence elsewhere implicate potential trade-offs restricting their fixation. LGS1 is thought to cause resistance by changing stereochemistry of strigolactones, hormones that control plant architecture and below-ground signaling to mycorrhizae and are required to stimulate parasite germination. Consistent with trade-offs, we find signatures of balancing selection surrounding LGS1 and other candidates from analysis of genome-wide associations with parasite distribution. Experiments with CRISPR-Cas9-edited sorghum further indicate that the benefit of LGS1-mediated resistance strongly depends on parasite genotype and abiotic environment and comes at the cost of reduced photosystem gene expression. Our study demonstrates long-term maintenance of diversity in host resistance genes across smallholder agroecosystems, providing a valuable comparison to both industrial farming systems and natural communities.

Keywords: environmental niche modeling; genotype–environment association analysis; species distribution modeling.

Fig. 1.

A subset of global sorghum landraces are from parasite-prone areas. (A) S. hermonthica HS scores across Africa based on MaxEnt SDM. To account for areas with suitable habitat where parasites have not been recorded, HS for accessions more than 200 km from any S. hermonthica record was set to zero; transparent colors demarcate areas of this HS masking. (B) Geographic distribution and frequency histogram of HS scores at locations of all georeferenced and genotyped sorghum landrace accessions. Landraces with available GBS (n = 2,070) and WGS (n = 143) data are shown. (C) Geographic distribution and frequency histogram of HS scores at locations of 1,369 S. hermonthica occurrence records. HS, habitat suitability; SDM, species distribution model; GBS, genotyping-by-sequencing; WGS, whole genome sequencing.

Fig. 2.

LGS1 loss-of-function alleles are broadly distributed within parasite-prone regions. (A and B) Schematic of large deletion variants (A) and frameshift mutation (B) impacting sorghum LGS1, a locus involved in resistance to S. hermonthica. Gray shading indicates position of gene models (A) or coding regions (B). Vertical black bars indicate the position of SNPs in the GBS dataset, and horizontal black lines denote 5-kb flanking regions used to impute deletion calls. In B, vertical white bars show the frameshift mutation (position 69,986,146) and the SNP at position 69,985,710 that tags the frameshift in the GBS dataset. Chr., chromosome. (C) Geographic distribution of LGS1 alleles in sorghum landraces. (D) Distribution of parasite HS scores at locations of sorghum accessions with lgs1-2 (n = 25), lgs1-3 (n = 34), frameshift (n = 131), or intact LGS1 (n = 785). HS, habitat suitability.

Fig. 3.

Sorghum genome-wide associations with parasite distribution implicate cell-wall and SL-signaling genes. (A) Genome-wide association with parasite HS score, based on 317,294 SNPs with MAF > 0.01 in 2,070 sorghum landraces. SNPs in genomic regions linked to SL biosynthesis/signaling (red), resistance through formation of a mechanical barrier (light blue), or low S. hermonthica germination (orange) are indicated. Germ., germination; mech. res., mechanical resistance. The dashed line represents significance threshold at a FDR of 5%. (B) Geographic distribution of reference and alternate alleles for an SNP (S2_21521798) in a pectinesterase gene (MAF = 0.275). (C) Distribution of parasite HS scores for sorghum accessions segregating for S2_21521798. (D) Geographic distribution of reference and alternate alleles for an SNP (S7_14459084) in a gene homologous to SMAX1 (MAF = 0.014). (E) Distribution of parasite HS scores for sorghum accessions segregating for S7_14459084. Alt, alternate; Ref, reference.

Fig. 4.

Efficacy of LGS1 loss-of-function varies across abiotic and biotic gradients. (A) Germination of two S. hermonthica populations in response to root exudate from 3-wk-old wild-type Macia and CRISPR–Cas9-edited sorghum seedlings with lgs1 deletion alleles. Plants were grown under well-watered high-nutrition (WWHN), well-watered low-nutrition (WWLN), or drought low-nutrition (DLN) conditions. Germination was 0% in response to diH₂O and 66% (Kibos) or 78% (Siby) in response to 0.2 μM GR24. (B) Origin of two tested S. hermonthica populations. (C) Germination of two S. hermonthica populations in response to synthetic SLs, (+)5-deoxystrigol (5-DS) or (±)orobanchol (ORO).