Genetic exchanges between closely related groups of organisms with different adaptations have well-documented beneficial and detrimental consequences. In plants, pollen-mediated exchanges affect the sorting of alleles across physical landscapes, and influence rates of hybridisation. How these dynamics affect the emergence and spread of novel ecological strategies remains only partially understood. Here, we use phylogenomics and population genomics to retrace the origin and spread of two geographically overlapping ecotypes of the African grass Alloteropsis angusta. Besides an ecotype inhabiting wetlands, we report the existence of a previously undescribed ecotype inhabiting miombo woodlands and grasslands. The two ecotypes are consistently associated with different nuclear groups, which represent an advanced stage of divergence with secondary low-level gene flow. However, the seed-transported chloroplast genomes are consistently shared by distinct ecotypes inhabiting the same region. These patterns suggest that the nuclear genome of one ecotype can reach the seeds of the other via occasional pollen movements with sorting of nuclear groups in subsequent generations. The contrasting ecotypes of A. angusta can thus use each other as a gateway to new locations across a large part of Africa, showing that hybridisation can facilitate the geographical dispersal of distinct ecotypes of the same grass species.

Samples of Alloteropsis angusta were obtained from herbaria, or from fieldwork conducted in Uganda, Tanzania and Zambia. When A. angusta and A. semialata grew together, both were sampled. Whole genomes of A. angusta were sequenced here as 150-bp or 250-bp paired-end Illumina reads or retrieved from previous studies together with those of A. semialata and A. cimicina (Table S1). Raw reads obtained from fresh samples were cleaned using NGS QC Toolkit v.2.3.3 [30] to remove reads with more than 20% of bases with a quality score below Q20, and those which had ambiguous bases. Bases with quality score below Q20 were further trimmed from the 3’ end of reads. Adaptors were removed using NxTrim. New sequences can be found on the NCBI SRA under project number PRJNA715711. Cleaned reads were aligned to the A. semialata reference using bowtie2 with default settings for paired end reads. and read processing scripts can be found at https://github.com/evcurran/Angusta-Pop-Genomics.

The genomes of population-level samples were scanned using restriction site associated DNA (RAD) sequencing. In brief, the DNA of up to five individuals of each population (more in ZAM1930 where the two morphs occurred) were double digested and then pooled before sequencing 72 – 96 individuals per lane of Illumina HiSeq 2500. Our final dataset included 196 A. angusta individuals. Raw RAD-sequencing reads were trimmed with Trimmomatic [40] to remove adaptor and other Illumina-specific sequences and bases with a low quality score (Q < 3) from the 5’ and 3’ ends. Reads were further clipped when the average quality within a sliding window of four bases dropped below 15. Reads were de-multiplexed using the module ‘process_radtags’ of STACKS [41], and they were mapped to the A. semialata reference genome using Bowtie2 with default settings for paired-end reads.

The alignment files are provided in the BAM format, the sample name is in the filename, and the relevant individual sample information can be found in Table S1.