Premise of research. Evergreen forests in eastern South Africa have high biodiversity but are limited in extent and have a highly fragmented distribution. Populations of forest plants are thus geographically isolated, and fine-scale evolutionary studies of these lineages might yield important insights into the history and assembly of the forests themselves. Despite their morphological diversity, a prior study showed that three South African Memecylon taxa in Melastomataceae (M. natalense, M. bachmannii, and M. australissimum) had almost identical nuclear ribosomal spacer sequences. Our study investigates phylogenetic relationships within this clade using multiple samples collected across populations and a next-generation phylogenomic approach.

Methodology. We used 87 low-copy nuclear (LCN) loci to examine relationships among these taxa using both concatenated and coalescent methods. We further used LCN loci to estimate phylogenetic networks and SNPs derived from LCN genes for STRUCTURE analysis of South African Memecylon populations. Finally, we employed two approaches (flow cytometry and SNPs) to infer the ploidy levels of these three taxa.

Pivotal results. Our investigations showed discordance among gene trees and the species tree, and low statistical support for relationships indicating that species monophyly cannot be recovered from this phylogenomic analysis. Phylogenetic networks and population structures showed that the South African Memecylon clade may be affected by gene flow and reticulate evolution. Flow cytometry and SNP-based estimations provided evidence for polyploidy within this group.

Conclusions. We found no evidence of monophyly for species within the South African Memecylon clade, which we infer to be the consequence of reticulation and recent and rapid evolution. More cytological studies and genomic data are needed to elucidate the evolutionary history of this group. Additionally, our study identifies priority populations within the South African Memecylon clade for conservation.

The DNA sequences were obtained from silica-dried tissue samples of Memecylon that were collected from South Africa (KwaZulu-Natal, Eastern Cape, and Mpumalanga provinces). These samples include Memecylon natalense, M. bachmannii, and M. australissimum. Additional samples of outgroups, M. soutpansbergense and M. kosiense, were collected from Limpopo and KwaZulu-Natal provinces. Target capture was employed to enrich genomic regions of interest for the samples. The probe set from Jantzen et al.(2020) was used to capture low-copy nuclear (LCN) genes. Total genomic DNAs were extracted following a modified CTAB extraction protocol for Melastomataceae (Jantzen et al. 2020). Preparation and quantification of double-barcoded genomic libraries and target enrichment were performed by RAPiD Genomics, Gainesville, Florida, USA. All accessions were processed as described in Jantzen et al. (2020). Cleaned reads were assembled with HybPiper v.1.3.1 using the template sequences used for probe design as references (Johnson et al. 2016). Potential paralogs identified by HybPiper scripts were removed (Jantzen et al. 2020). The LCN sequences were individually aligned using MAFFT v7.215 (Katoh and Standley 2013), with the default gap opening penalty. The alignment was then trimmed using trimAl v1.2 (Capella-Gutierrez et al. 2009) with a gap threshold of 0.9. Gene trees were generated for each of the individual gene alignments using Maximum Likelihood (ML) in RAxML v8 (Stamatakis 2014), with 500 rapid bootstrap pseudoreplicates combined with a ML tree search using the GTRGAMMA model. The remaining genes were concatenated, and a partition scheme was generated using Phyx v8.2.0 (Brown et al. 2017). Although maximum Likelihood analyses of the concatenated dataset were repeated for supercontigs, exons, and introns using the RAxML parameters as described above, we archived phylogenies only from introns here because 87 individual introns were informative at the population level and resulted in phylogenies with higher bootstrap support (BS) compared to the individual supercontig and exon datasets. To infer a coalescent species tree, 87 ML gene trees (derived from introns) generated from RAxML were input to ASTRAL-III v5.0.3 (Zhang et al. 2018).

The alignments and phylogenies included in this dataset are generated only from 87 introns because they were informative at the population level of this South African Memecylon. Cleaned sequence reads used to generate this data have been deposited on the NCBI Sequence Read Archive (PRJNA576018).