The transition to agricultural cultivation of neo-crops may fail to account for wild genetic diversity patterns: insights from the Cape Floristic Region

Target taxa and sampling

The species selected for evaluation (C. intermedia E.Mey, C. subternata Vogel., and C. longifolia Vogel.) represent the most widely cultivated Honeybush taxa, cultivated in the Western Cape and Eastern Cape provinces of South Africa (Joubert et al., 2011; McGregor, 2017). Consequently, these three species have likely experienced the greatest extent of redistribution outside of their natural range through cultivation (Fig. 1; Fig. 3 in Joubert et al., 2011)—placing them at high risk of genetic pollution. Additionally, these species represent three distinct distribution and life history patterns (Schutte, 1997): a widespread obligate resprouter occurring at altitudes between 500–1700 m (C. intermedia), a widespread coastal lowland obligate seeder (C. subternata ) and a critically endangered Eastern Cape endemic, riparian specialist with a mixed post fire response of facultative seeding and sprouting (C. longifolia). Since life history traits and range size impact a species’ demographic history (Ellegren & Galtier, 2016), these three taxa are likely to exhibit different genetic diversity patterns that need to be accounted for during translocation and cultivation.

Two of these species (C. Intermedia and C. Subternata) were subject of recent phylogeographic studies (Galuszynski & Potts, 2020a; Galuszynski & Potts, 2020c) and these previously published data sets were bolstered with additional samples to ensure that 24 individuals were available from each populations for analysis. Samples were collected from four geographically separated wild populations across the natural range of each species. The intention was to (a) maximise the genetic variation detected among populations, and (b) provide a representative reference of haplotypes to describe the origins of material under cultivation. Cultivated material was sampled from Honeybush farms identified remotely from internet searches rather than relying on existing farmer networks, and were located in different mountain ranges (details of the locations are provided in Fig. 1 and Table 1). This approach was employed to avoid potentially redundant sampling of cultivated material originating from seed exchange between farmers located in close proximity to one another, however, the true origin of the commercial seed remains uncertain (see discussion). The cultivators included in the study are situated near to three of the four major Honeybush nurseries reported by Joubert et al. (2011) (Fig. 1). Three cultivated populations of each species were sampled. From all populations (wild and cultivated), a total of 24 plants were sampled with a minimum of 5 m distance between sampled individuals. The final data set consisted of 504 samples collected across 21 (12 wild, 9 cultivated) populations of three commercially important Cyclopia species, population locations are mapped in Fig. 1. Fresh leaf material was collected from a healthy growing tip of each individual and placed into silica desiccating medium for a minimum of two weeks prior to DNA extraction. All sampling was approved by the relevant landowners involved and permitting agencies, Cape Nature (Permit number: CN35-28-4367), the Eastern Cape Department of Economic Development, Environmental Affairs and Tourism (Permit numbers: CRO 84/ 16CR, CRO 85/ 16CR), and the Eastern Cape Parks and Tourism Agency (Permit number: RA_0185).

DNA extraction and haplotype detection

The DNA extraction and haplotype detection protocol followed the approach previously described in Galuszynski & Potts (2020c) and a brief overview of the approach is provided here. Whole genomic DNA was extracted using a modified CTAB DNA extraction approach, adapted from Doyle & Doyle (1987). Extracted DNA was quantified using a NanoDrop 2000c Spectrophotometer (Thermo Fisher Scientific, Wilmington, DE19810r Scientific, USA) and diluted to 5 ng/µL for PCR amplification and subsequent HRM analysis.

High Resolution Melt analysis involves the gradual heating of PCR products amplified in the presence of a DNA saturating dye. As the double stranded DNA is heated it dissociates at a rate based on the binding strength of the nucleotide sequence under analysis. As such, different nucleotide sequences should produce a distinct melt curve when plotting sample differences in measured fluorescence against change in temperature.

Three DNA fragments from two non-coding cpDNA regions (atpI-atpH intergenic spacer and ndhA intron) were amplified using Cyclopia specific primers and subsequently screened for nucleotide variation via HRM curve analysis. Samples were run in duplicates and HRM clustering was conducted on a single population basis following the recommendations of Dang et al. (2012). This was achieved by grouping populations using the ‘well group’ option in the CFX Manager Software (Bio-Rad Laboratories, Hercules, California, U.S.A.) and running the HRM clustering on these predefined population well groups. All reactions (PCR amplification and subsequent HRM) took place in a 96 well plate CFX Connect (Bio-Rad Laboratories, Hercules, California, U.S.A.). Haplotype melt curve grouping was achieved using the automated clustering algorithm of the High Precision Melt software (Bio-Rad Laboratories, Hercules, California, U.S.A.) (ΔTm =0.05, curve shape sensitivity = 70%, temperature correction = 20). HRM cluster to haplotype confirmation was achieved by unidirectional sequencing, as described in Galuszynski & Potts (2020c). The chloroplast regions targeted by HRM were sequenced for a subset of individuals per HRM cluster from populations not previously studied by Galuszynski & Potts (2020a); Galuszynski & Potts (2020c)). A total of 39 individuals for the atpI-atpH intergenic spacer and 36 individuals for the ndhA intron were PCR amplified using the reverse primers and following PCR protocols of (Shaw et al., 2007) and sequenced. The PCR and HRM conditions and details of the primers used in this study, are provided in S1.

Sequences were assembled using CondonCode Aligner [v2.0.1] (CodonCode Corp, http://www.codoncode.com). Each base-call was assigned a quality score using the PHRED base-calling program (Ewing et al., 1998). Sequences were then automatically aligned using ClustalW (Thompson, Higgins & Gibson, 1994) and visually inspected. All indels that were difficult to score (due to homopolymer repeats that are prone to alignment errors) were removed. The cpDNA regions under investigation are maternally inherited in tandem and not subject to recombination (Reboud & Zeyl, 1994), and were therefore combined for subsequent analysis. A custom R script (provided with a minimum working example online: https://doi.org/10.6084/m9.figshare.12624620.v1) was then used to assign each sample its respective haplotype identity based on HRM clustering.

Haplotype diversity analysis

All analyses were performed in R (V 3.5.1) (R Core Team, 2018). The genealogical relationships among haplotypes were established using a Statistical Parsimony (SP) network (Fig. 2) constructed in TCS (v 1.2.1) (Clement, Posada & Crandall, 2000). As TCS treats each base pair in an indel as an evolutionary event, all indels were reduced to a single base pair prior to analysis with default options selected for network construction. Haplotype diversity and differentiation was compared between wild and cultivated individuals grouped by species and origin. Differences in gene diversity (GD) between wild and cultivated populations were tested via Mantel tests (Mantel, 1967); using Hs.test function implemented in the adegenet (v 2.1.1) library, (Jombart & Ahmed, 2011). Genetic differentiation between wild and cultivated populations were tested via an Analysis of Molecular Variance (AMOVA) (Excoffier, Smouse & Quattro, 1992) using the poppr.amova function with 999 permutations implemented in the poppr [v2.8.3] library (Kamvar, Tabima & Grünwald, 2014). Three additional population differentiation measures were calculated: two fixation indices, pairwise Gst (Nei, 1973) and G””st (Hedrick, 2005), and a measure of genetic divergence, Jost’s D (Jost, 2008), using the pairwise_Gst_Nei, pairwise_Gst_Hendrick , and pairwise_D functions respectively, all from the mmod [v1.3.3] library (Winter, 2012). Population clustering was inferred from a Neighbor Joining tree constructed using Prevostis pairwise population genetic distance (Prevosti, Ocaña & Alonso, 1975), calculated using the prevosti.dist function in the poppr library). This distance measure treats alignment gaps as evolutionary events and all gaps were reduced to a single base pair prior to analysis. Support for population clustering was assessed via a bootstrap analysis with 9999 replicates, implemented using the aboot function implemented in poppr.

Haplotype detection

Of the 504 samples screened for haplotype variation, seven (six wild and one cultivated) failed to PCR amplify despite repeated efforts and the final dataset consisted of 497 samples. High Resolution Melt analysis with haplotype confirmation by sequencing revealed 17 cpDNA haplotypes with 100% specificity for all three loci (i.e., no cases of different haplotypes being grouped into the same HRM cluster were detected). The final concatenated dataset consisted of 794 bp (457 bp from the atpI-atpH intergenic spacer and 339 bp from the ndhA intron) with an overall GC content of 28.1%. The alignment contained 22 polymorphic sites including nine transitions, ten transversions, and three indels (two of 7 bp and one of 71 bp). Haplotype frequency within populations and nucleotide variation among haplotypes are summarized in Tables 1 and 2, respectively.

Genetic diversity analysis

The SP network (Fig. 2) revealed relatively low divergence among haplotypes, with all haplotypes diverging from a central variant. Of the 17 haplotypes detected, only four were present in cultivated populations. These four haplotypes (in addition to a fifth haplotype, detected only in wild C. subternata and C. intermedia populations from Garcia’s Pass, GAR Fig. 1) were shared among species. Two of these haplotypes were detected in C. intermedia populations (wild and cultivated) and cultivated C. subternata populations, but were not present in any wild C. subternata populations screened.

Clustering of Cyclopia populations, based on pairwise population genetic distance resulted in weak grouping of species. Cultivated populations, however, exhibited little differentiation and generally clustered together based on species (Fig. 3). Similarly, wild C. longifolia populations exhibited little genetic differentiation and all wild and cultivated populations of this species formed a single group (Fig. 3). Wild C. intermedia and C. subternata populations tended to exhibit higher levels of genetic divergence. All cultivated C. subternata populations were clustered with two wild populations originating from the Tsitsikamma mountains (Kareedouw Pass KAR, and Bloukrans Bridge BKB). The remaining wild C. subternata populations were more divergent and did not cluster with other C. subternata populations. The C. intermedia and C. subternata populations sampled from Garcia’s pass (GAR) exhibited no genetic differentiation (both fixed for haplotype F). In general, wild C. intermedia populations tended to be genetically distinct, supporting unique haplotypes and formed no clear clusters in the Neighbor Joining population tree (Fig. 3). Cultivated C. intermedia populations did not exhibit this variability. Rather, all cultivated C. intermedia populations form a single cluster with the wild C. intermedia population sampled from the Langkloof (LK).

In the cases of C. subternata and C. intermedia, genetic structuring was detected in wild populations, with 60.1% and 83.3% of genetic variation detected among populations for the two species, respectively (associated Fst values significant, p < 0.005, Fst =0.093 and 0.023, respectively). In contrast, no structuring was detected in cultivated populations of these species, with 4.3% and 1.7% of variation structured within cultivated populations for C. intermedia and C. subternata, respectively (p < 0.005, Fst =0.0001 for both species). No genetic structuring was found for wild or cultivated C. longifolia populations, as all populations shared the same common haplotype and only two rare haplotypes (N and M, Fig. 2) were detected in wild populations. Gene diversity and genetic differentiation followed a similar pattern, with wild C. intermedia and C. subternata populations having higher mean diversity than cultivated populations (p < 0.01), but no differences in gene diversity was detected between wild and cultivated C. longifolia populations. Mean genetic differentiation (Gst, G”st and Josts D) was higher in all wild populations than in cultivated populations. All population differentiation and diversity measures are summarized in Table 3.

Origin of cultivated genetic diversity

Despite initial Cyclopia breeding material originating from multiple wild populations (Joubert et al., 2011, Fig. 1), screening for individuals with commercially desirable traits has likely removed much of the haplotype richness from commercial breeding stock. The transition to Honeybush cultivation therefore represents a genetic bottleneck—under representing rare haplotypes and homogenizing the cultivated genepool.

Based on the NJ clustering of populations (Fig. 3), cultivated C. intermedia have likely originated from populations located in the Langkloof (LK); C. subternata populations from the Tsitsikamma and/or Outeniqua mountains (these wild populations share the same common haplotype found in cultivated populations); while C. longifolia originates from its only known wild source, the Van Stadens River system. These findings are consistent with those from recent microsatellite analysis of C. subternata (Niemandt et al., 2018). They compared wild populations from the Tsitsikamma and Outeniqua mountains to the Agricultural Resource Council’s (ARC) commercial genebank (an important source of commercially traded Honeybush seed), revealing no genetic differentiation between the two wild populations sampled and the genebank accessions. The lack of haplotype diversity detected in cultivated populations is therefore unlikely to be a byproduct of failing to detect variation in the slow evolving chloroplast genome (Schaal et al., 1998), as greater divergence was detected among populations originating from the Tsitsikamma or Outeniqua mountains using cpDNA screening via HRM than in the microsatellite based study.

The history of the movement of cultivated seed remains speculative and conversations with cultivators during sampling revealed that seed is largely sourced from existing farmer networks in their respective areas (with the initial origin of seed unknown). However, C. subternata cultivated in Harkerville (H) was confirmed to have been established from commercial seed and did not differ in dominant haplotypes from other cultivated populations. Furthermore, Harkerville shared a rare haplotype with cultivated material from Uniondale (U)—where putative hybrids were detected (NC Galuszynski, pers. obs., 2018, leaf material from these individuals was collected and is stored at the Nelson Mandela University in Port Elizabeth, South Africa). This rare haplotype was only detected in wild C. intermedia populations located in the Swartberg Mountains (SWB) and may be evidence of possible chloroplast capture (Hansen, Siegismund & Jørgensen, 2003) resulting from interspecific crosses taking place during the initial breeding trials (Joubert et al., 2011) or under field cultivation. The reasoning behind this argument is two fold. First, the chloroplast regions screened exhibit phylogeographic structuring in both species (Galuszynski & Potts, 2020a; Galuszynski & Potts, 2020c) and it is therefore unlikely that a wild C. subternata population will support this rare haplotype outside of the Swartberg mountains (where C. subternata does not naturally occur). Secondly, C. intermedia and C. subternata have been found to successfully produce hybrid offspring with other members of the genus under experimental conditions, and C. intermedia material from the Swartberg was included in the initial Honeybush cultivation trials (Joubert et al., 2011, Fig. 1). However, interspecific hybridization should be investigated through additional molecular work targeting the nuclear genome, which is subject to recombination and provides more insight into introgression history.

Potential impacts of cultivated genetic material on wild populations

Higher levels of genetic diversity were detected in wild Cyclopia populations than cultivated populations, with multiple cases of near complete haplotype turnover among wild populations (Table 1)—cultivated populations exhibited nearly no differentiation (Table 3). This level of haplotype turnover and genetic structuring of cpDNA in the wild was expected and has been described in detail for C. intermedia (Galuszynski & Potts, 2020a) and C. subternata (Galuszynski & Potts, 2020c). This suggests that there may be a tendency for cultivated populations to fail to account for natural phylogeographic patterns in regions where genetic structuring of plant populations occurs, including regions that may play a significant role in the discovery of neo-crops (e.g., Mesoamerica, Ornelas et al., 2013; South America, Turchetto-Zolet et al., 2013; Australia, Byrne, 2008; the Mediterranean basin, Feliner, 2014; and the Cape of South Africa, Galuszynski & Potts, 2020b; Tolley et al., 2014).

Genetic diversity differed between species (Table 3), suggesting different demographic histories among these closely related taxa (Ellegren & Galtier, 2016). Notably, rare and locally endemic taxa are predicted to have low levels of genetic variation due to their restricted distributions and small population sizes (Ellstrand & Elam, 1993; Gitzendanner & Soltis, 2000; Segarra-Moragues, Torres-Díaz & Ojeda, 2012), evident in wild C. Longifolia populations. The redistribution of genetic lineages via neo-crop cultivation needs to account for differences in demographic histories. Local endemics, for instance, may be at greater genetic risk from cultivated variants due to naturally low levels of genetic variation being more susceptible to genetic pollution by foreign lineages (Levin, Francisco-Ortega & Jansen, 1996; Wolf, Takebayashi & Rieseberg, 2001). Commercial production of narrow endemics may therefore require periodic supplementing of cultivated stands with locally sourced seed material in order to promote the preservation of rare haplotypes.

Genetic pollution can only occur if genetic material is able to escape into the wild. Seed dispersal is limited to a few meters in Cyclopia, and one would expect the chances of seed escape to be low. However, unmonitored spillover of cultivated seed into adjacent natural habitat does occur (NC Galuszynski, pers. obs., 2017) in addition to rare cases of cultivated plants intentionally established in natural vegetation (NC Galuszynski, pers. obs., 2017; G McGregor, pers. com., 2016; S Nortje, pers. com., 2019). The chloroplast genome is maternally inherited with no recombination in the majority of angiosperms (Mogensen, 1996) and is, therefore, subject to directional selective sweeps. Thus, the introduction of vigorous maternal lineages from commercial seed lots could disrupt local haplotypes diversity in wild crop plants. While the impacts of foreign haplotypes establishing in wild Honeybush populations is unknown, the introduction of foreign Spartina alterniflora (Poaceae) haplotypes from the United States to native Chinese populations resulted in a hybrid swarm that disrupted local cpDNA genome frequency and ecological processes due to vigorous growth displacing native plant species (Qiao et al., 2019).

Mass flowering of cultivated populations forms a powerful attractant to pollinators, increasing local pollinator density (Holzschuh et al., 2011; Westphal, Steffan-Dewenter & Tscharntke, 2003). This facilitates the spread of genetic material from cultivated populations into the wild via pollen flow, particularly in outcrossing species (Ellstrand, Prentice & Hancock, 1999), such as members of Cyclopia (Koen et al., 2021). By altering local allele frequencies through the introduction of large genetically depauperate commercial plantations, a landscape wide genetic bottleneck may result, promoting erosion of wild genetic diversity. The current state of neo-crop cultivation, relying on low genetic diversity breeding stock that is redistributed outside of its natural range (and possibly containing interspecific hybrid taxa as observed in the Honeybush populations in Uniondale, but requiring further research), represents a genetic threat that should be acknowledged and mitigated.

Until the genetic risks are better understood, formal guidelines should be developed to facilitate sustainable cultivation of neo-crops. In the case of Honeybush, an ecology-centric approach, as the one outlined by Potts (2017), may be desired due to the conservation value of many Cyclopia species; including the commercially important taxa: C. longifolia (critically endangered), C. genistoides (near threatened), C. maculata (near threatened), C. plicata (endangered), and C. sessiliflora (near threatened). However, it is unlikely that all neo-crops will exhibit the levels of phylogeographic structuring, genetic diversity and, conservation requirements of Cyclopia, and more work is required to define the extent to which wild genetic resources are at risk during neo-crop cultivation practices, particularly in species rich landscapes such as the Cape of South Africa.