Introduction

Climate-driven ocean contact zones, caused by range shifts of reproductively compatible but differentiated genetic lineages, provide evolutionary opportunities by promoting rapid genome recombination and novel selective variants1,2. Species can undergo rapid divergence when isolated during climate extremes and subsequent secondary contact can result in outcomes ranging from complete panmixia to reproductive isolation3,4,5,6.

The genomic consequences of contact zones may be especially relevant for species with restricted dispersal abilities. In marine environments, currents can shape and restrict dispersal, causing limited connectivity that favours evolutionary processes at local scales, giving isolated populations time to diverge. This is the case for kelps, which form massive underwater forests across many temperate and Arctic coastal regions7, but have been undergoing distributional range shifts and contractions as ocean temperatures change8,9,10,11. While range shifts can result in the loss of populations and total kelp biomass7,12,13, they may also be associated with an increase in evolutionary potential when contact zones between different lineages of the same, or closely related species, are created. For example, hybridization may accelerate local adaptation of seaweed to previously unsuitable conditions14,15. Several examples of kelp hybridization have been described either experimentally16,17,18,19 or in the field20,21,22,23,24. In changing oceans, it is important to understand how shifts in kelp distribution may result in niche overlap between closely related species, putatively resulting in admixture between genetically distinct populations and species. Given the morphological plasticity of many kelps25,26,27,28, molecular data are essential for distinguishing between cryptic populations with atypical morphologies and hybrids as well as for the identification of populations with unique genetic diversity.

Climate-driven distributional changes have been observed within the genus Ecklonia (E. maxima and E. radiata) along the coast of southern Africa. The range of Ecklonia maxima has shifted eastward8, while the loss of a unique northern hemisphere population of E. radiata was recently reported from Oman, potentially due to the impact of climate change12. Ecklonia radiata is among the most dominant kelps in the southern hemisphere, with a warm-temperate distribution ranging from New Zealand and Australia to the south and eastern coasts of southern Africa12,29 (upper thermal tolerance range 21.2–26.5 °C29). Ecklonia radiata is generally small (40 cm up to 2 m long), occurring typically in the intertidal to subtidal depths of 30 m. Rarer and deeper populations have also been recorded at depths reaching 80 m, under suitable conditions30,31. It forms extensive beds in Australasia and usually sparser populations in sheltered subtidal and intertidal regions in most of its South African distribution8,12,29. In Africa, E. radiata extends eastwards from Cape Agulhas to southern Mozambique12 in addition to a single population in the Cape Peninsula21. There are also records of previously undescribed deep populations in northeastern South Africa and Mozambique and past populations were recorded in similar habitats as far north as Oman12. A second species, Ecklonia maxima is abundant on the west coast of South Africa in the cool-temperate Benguela region. Ecklonia maxima also occurs in the biogeographical overlap region between the west coast and warm temperature south coast and evidence suggests that gametophytes and small sporophytes are capable of surviving in nutrient-replete cultures at temperatures up to 22 °C16. It forms large marine forests, often growing to several meters in length, has a long, hollow stipe (maximum 15 m32), with a float at the top holding the fronds32. It is the main canopy-forming kelp in < 6 m depth in this region, although it can be found down to 30 m under certain conditions33. Its northern limit is in southern Namibia (Anderson et al.34, pers obs MMR) and it has recently expanded ~ 70 km eastwards around the southern tip of South Africa, reaching the De Hoop Nature Reserve, ca. 20068.

The two species can be found in contact zones along the South African coast. Previous genetic studies have shown that E. radiata and E. maxima are sister species that form a clade separate from the east Asian Ecklonia21,25,35. Interestingly, the taxonomic affinity of E. radiata from the Cape Peninsula was unclear (based on nuclear ITS and organellar markers), with individuals presenting atypical morphologies, generally E. maxima-like, but with E. radiata-like rugose and serrated fronds, solid stipes, and with discordant nuclear and mitochondrial gene trees, suggesting a potential hybrid origin21. Ecklonia radiata and E. maxima are able to produce viable hybrid sporophytes in laboratory experiments16 raising the question of whether similar events occur in nature. Since both species can display morphological plasticity that makes morphology-based identification difficult, particularly for E. radiata8,21,29, molecular data from multiple markers are essential to identify lineages.

This study aims to understand the impact of niche overlap on the genetic diversity of the two Ecklonia species and investigate possible hybridization events. To achieve this, we sampled the entire known distribution of Ecklonia in southern Africa, including some of the northernmost localities of the known distribution for both species, the southernmost point, as well as deep offshore populations from the south Atlantic and southern South Africa. Nuclear and mitochondrial markers were compared to identify possible introgression in contact zones and unique genetic diversity in edge and deep populations. In addition, we modelled the distribution of both species to evaluate niche differentiation and the probability of range overlap. Our analyses reveal marked geographic patterns of genetic structure in both Ecklonia species and indicate that the deep populations of Ecklonia in southern Africa are important refugia of unique genetic diversity. Our findings also provide evidence of hybridization in the contact zones between the two species and raise the hypothesis that hybrids are more likely to be adapted to the broader niche of E. radiata.

Material and methods

Sampling, DNA extraction, genotyping and sequencing

Populations of Ecklonia were sampled from shallow and deeper reefs spanning the entire known distribution of the genus off the coast of southern Africa and Vema Seamount (roughly 1000 km offshore from the west coast of South Africa; closest to Doringbaai: DB on Fig. 1). Further sampling effort was focused on the two identified contact zones (BB, DH) and on deeper populations (VM, RB), where individuals with atypical morphologies were identified. These samples were identified as either E. maxima (EMDH) or E. radiata (ERHOO, ERDH, ERSDH) (see Table 1 for sample site abbreviations) in the field, based on morphological traits considered to be diagnostic, such as hollow stipe for E. maxima or spiny and serrated blades for E. radiata. Individuals with atypical morphologies were also collected such as smooth bladed E. radiata (ERHOO, ERSDH) and plants similar to those described by Rothman et al.21 as “E. cf. radiata/maxima” (ERhBOT, EXBB), a putative hybrid. Additionally, four populations of E. radiata were included from shallow waters off the coast of New South Wales, Australia, for comparison with South African samples and to provide a better representation of the species’ genetic diversity (Table 1).

Figure 1
figure 1

Sampling locations of Ecklonia in southern Africa and genetic structure inferred from mtDNA and microsatellite genotypes. Warm colors (orange–yellow) identify E. radiata haplogroups and clusters while cold colors (blue–purple) identify E. maxima. Colored bars above STRUCTURE plots indicate individuals’ haplotype. Black diamond marks denote individuals for which there was a discordant species attribution between mitochondrial and nuclear markers. A) Geographic distribution of COX1 haplotypes (pie charts) and STRUCTURE clusters (borders) for K = 8 and B) haplotype network showing haplotype genealogy. C) Structure plots assuming K = 2 (top) and K = 4 (middle) and species specific (bottom) genetic clusters and D) FCA scatter plot based on the clusters defined K = 4, circles colored according to mtDNA haplotypes present in the cluster. Map was created using a custom R (https://cran.r-project.org/) script on version 4.3.0 and edited using Affinity Designer, version 2.3.1.

Full size image
Table 1 Sampling events of Ecklonia populations in southern Africa and Australia, ordered east to west.
Full size table

Silica-dried tissue from South African samples was macerated with 3 mm tungsten beads in a Tissue lyser II (Qiagen, Hilden, Germany) for 3 min. Genomic DNA was extracted using the NucleoSpin® 96 Tissue kit (Macherey–Nagel, Duren, Germany), following an adapted version of the supplier’s protocol. Extracted DNA was diluted 1:100 before PCR amplification. DNA from Australian samples was extracted by grinding 50 mg of frozen tissue, followed by an extraction and cleaning of total genomic DNA using the DNeasy plant DNA & Pro clean up kit (Qiagen). A total of 30 previously described microsatellite primers were tested, 12 E. cava primers36, 10 E. radiata primers37 and 8 E. radicosa primers38. Of these, 8 were polymorphic with consistent amplification in both E. maxima and E. radiata and were used to produce a multi-locus genotype matrix for all samples (Primer sequences and amplification are detailed in Table S1). After amplification, fragment size was analyzed on an ABI PRISM3130xl (Applied Biosystems) with GeneScan Liz 500 size standard (Applied Biosystems). Allele sizes were manually scored using STRAND (https://vgl.ucdavis.edu/STRand), binned with the R (R Core Team, 2023) package MsatAllele39 and manually reviewed to resolve ambiguities.

Complementary COX1 sequences for selected samples were obtained using the GAZF2 and GAZR2 primers previously published by40 (primer sequences and PCR conditions are detailed in Table S1). Sequencing was performed on an ABI PRISM3130xl (Applied Biosystems) automated sequencer at CCMAR. A total of 77 sequences were generated for southern African Ecklonia; 45 E. maxima sequences from 14 sites in South Africa and 3 from Namibia, 26 E. radiata sequences from 9 sites in South Africa and 3 from Mozambique, and 6 Ecklonia spp. sequences from 2 deep populations from Vema Seamount (n = 3) and Rocky Bank (n = 3) (Supplementary Table S2). Additionally, 128 E. radiata sequences from Australia and 4 from Oman were obtained from GenBank and added to the dataset (Supplementary Table S2).

Genetic analyses

Genetic diversity per site for the nuclear markers was estimated with allele frequencies, mean allelic richness (A), Nei’s gene diversity (HE), observed heterozygosity (HO), and inbreeding coefficient (FIS) using GENETIX 4.0541. Normalized allelic richness (Â) and private alleles (PÂ) were also calculated using the approach described in5). Values were standardized for a sample size of 20 using 1000 randomizations. Genetic differentiation was estimated between populations using Jost’s D and pairwise FST (θ42) as implemented in the R package “diveRsity”43 with 10,000 bootstrap replicates.

Genetic structure was estimated using the ParallelStructure package implementation of STRUCTURE in R44 without any prior population assignments. Population clusters for the full microsatellite dataset were tested by setting sequential K from 1 to 10, with 10 replicates for each K, for a total of 2,000,000 Markov Chain Monte Carlo (MCMC) iterations with a 100,000 burn-in each. The optimal number of clusters (K) was selected based on multiple factors. STRUCTURE HARVESTER Web v0.6.9445 was used to calculate the ΔK criterion46 and a factorial correspondence analysis (FCA) was run as implemented in GENETIX 4.0.541. Initial exploratory analyses showed that the nuclear markers were able to clearly separate the two Ecklonia species, with K = 2 being the optimal choice according to the ΔK criterion. With the objective of analyzing genetic structure within each species and the possible hybrids, the dataset was separated into putative species (E. maxima and E. radiata) and into putative species plus putative hybrids (E. maxima, E. radiata and putative hybrids from BB and DH). A ParallelStructure analysis was run for each set with the same previous parameters. Runs for the selected optimal Ks were averaged and the results were plotted with CLUMPAK47. The number of optimal clusters for E. maxima (K = 3) and E. radiata (K = 2) was selected based on the ΔK criterion while the best K (K = 4) for the analysis focusing on both species and putative species was selected based on a combination of ΔK and morphological identification of putative hybrids. Results for the three approaches were carefully evaluated manually and are presented here.

The 77 COX1 sequences from South Africa, 127 from Australia and 4 from Oman were used to generate a 207 sequence alignment. Sequence editing, trimming and quality assessment was performed in Geneious 4.8.5 (www.geneious.com), and subsequently aligned with MAFFT v7.47048 using the default setting, resulting in a 574 bp alignment. Haplotypes and genetic diversity metrics for each species and region were evaluated as nucleotide diversity (π), number of haplotypes (H), haplotype diversity (Hd), number of private haplotypes (PH), segregating sites (S) and average number of nucleotide differences (K) using DNASP v649. Geographic distribution and genealogy of COX1 haplotypes were mapped and inferred using a median-joining approach as implemented in POPART v1.750. Given the geographic distribution of COX1 haplotypes, genetic pairwise distance between relevant population clusters (see Fig. 1 and Table 2) was calculated in MEGA X51.

Table 2 Genetic diversity values for the mitochondrial marker COX1 across the different species and regions.
Full size table

Species distribution modelling

Species distribution modelling (SDM) was used to estimate niche overlap between the sister species E. radiata and E. maxima. This approach used Boosted Regression Trees (BRT), a machine learning algorithm with high predictive performance, owing to its ability to fit non-linear relationships and complex interactions between predictor variables, and overfitting reduction through hyperparametrization and forcing of explicit monotonic responses52,53.

Biologically meaningful predictors for both species54,55 were extracted from the dataset Bio-ORACLE v.256 for the benthic realm (i.e., layers with information for the bottom of the ocean). Predictors were chosen to reflect essential resources (nutrients, as nitrate) and factors affecting physiology (salinity and maximum and minimum temperatures). Presence records for the species were extracted from the fine-tuned dataset of marine forest species57. The same number of pseudo-absences as presences were generated for both species in random sites where no presences were recorded58. The negative effect of spatial autocorrelation in the models developed for both species was reduced by building a correlogram to estimate the correlation of predictors within increasing distances of occurrence records. Records were then pruned by randomly selecting one record from the group of records found within the minimum distance at which predictors were significantly correlated, e.g.59.

A tenfold cross-validation framework using independent blocks54,60 was used to find the optimal combination of BRT hyperparameters., specifically, number of trees (from 50 to 1000, step 50 trees), tree complexity (from 1 to 6) and learning rate (0.1, 0.01, 0.005 and 0.001). This also allowed inference of model performances with independent data, as well as their potential for temporal and spatial transferability54,61 by using the area under the curve (AUC) of the receiver operating characteristic curve and sensitivity (true positive rate62). Models were forced to fit strict monotonic responses, negative for maximum temperature and positive for minimum temperature, nitrate and salinity53,54,63. Thermal tolerance limits were determined from individual response functions (i.e., partial dependency functions) produced for maximum temperature, while accounting with the average effect on the models produced by all alternative predictors54,64.

Niche overlap analyses

To assess niche overlap between the two species, the probabilistic method of Swanson et al.65 was used. This provides directional estimates of overlap and accounts for the specific distribution of species in niche spaces. The niche region (NR) of each species, inferred with SDM, was defined as the 95% probability region of their multivariate space. The method determined overlap as the probability of an individual from a species being found in the NR of the alternative species, within a Bayesian framework to account for uncertainty.

Niche similarity between species was also addressed following hypothesis testing as proposed by Warren et al.66. This approach used two metrics of niche overlap, namely the Warren’s I66 and the Schoener’s D (ranging between 0 and 1, from no overlap to complete overlap), and tested for niche similarity by asking whether niches are more similar to one another than expected by chance. The D and I values were compared to a null distribution of 104 overlap values produced by linking the niche of one species to a niche generated with random occurrences drawn from the geographic space of the alternative species.

Results

Genetic diversity

A total of 181 alleles were identified in 773 individuals at all 8 microsatellite loci. The total number of alleles per locus ranged from 9 to 37, while allele number per population varied between 19 and 71, with 1.54% of missing data. Overall, allelic richness and normalized allelic richness (A, Â) were higher in populations located near the centre of each species distribution, with lower values in populations near the edges (DP, PN and KZ for E. maxima, PE, PA and SL for E. radiata). The number of private alleles was low overall, with the notable exception of the Australian east coast E. radiata populations (CBS, E, PS, SH) and the E. maxima populations of Millers Point (MP) and St James (SJ). The deeper population of Rocky Bank (RB) also exhibited low values of allelic richness, while the population of Vema Seamount (VM) had amongst the highest values for E. maxima. Expected heterozygosity (HE) and observed heterozygosity (HO) varied between 0.218–0.687 and 0.108–0.629 respectively and followed a similar pattern to allelic diversity. Most populations showed a significant deficit of observed heterozygosity revealed by significantly high inbreeding coefficient (FIS) values (Table 1).

A total of 22 COX1 haplotypes were found (Supplementary Table S3). Haplotype diversity was high within each species and higher for E. radiata from South Africa compared to Australia. Nucleotide diversity was low on average due to a few mutations separating each haplotype (Table 2, Fig. 1).

Genetic structure

Both nuclear and mitochondrial markers clearly separated Ecklonia maxima and E. radiata, with a few exceptions (Fig. 1, diamond symbols). The first STRUCTURE analysis including the full microsatellite data set revealed two very well supported genetic clusters (Fig. 1C, K = 2) that separated E. maxima and E. radiata, though some populations had individuals with high admixture levels, most notably VM and ERSDH. Similar results were found in the haplotype network, with E. maxima and E. radiata clearly separated (Fig. 1B). Both marker types also supported strong geographic patterns of genetic structure for Ecklonia in South Africa.

When separating the nuclear data into 4 genetic clusters, subdivision within species becomes more apparent while also highlighting unique characteristics for the deep-water populations and putative hybrids (Fig. 1C, K = 4). Both species are divided into two genetic groups, E. maxima having one group that encompasses populations from DP to HK and another from DB to DH and E. radiata presenting a genetic split between clusters at DW, where samples show great admixture between clusters. The most remarkable results though are those pertaining to the deeper population from VM and RB, as well as the putative hybrids from BB (EXBB, ErHBOT) and DH (ERSDH). Deeper populations present high admixture between both E. maxima clusters and even with E. radiata (VM with K4 and RB with K5, Fig. 1C), while putative hybrids from BB show significant admixture between both species and hybrids from DH keep the same pattern as before. The FCA analysis supports the possibility of 4 diversifying clusters, with a separation between E. radiata groups, a group with both deeper populations and DH putative hybrids at the edges of the major E. maxima cluster and a diverging group made up of the putative BB hydrids (Fig. 1D).

A second level of subpopulations was revealed by the analyses focusing on each specific species. Ecklonia maxima was further divided into 3 genetic clusters, with clear geographic distributions for each one (Fig. 1A,C). The K1 cluster ranged from Diaz Point in Namibia (DP) southwards to Hondeklipbaai (HK), K2 from Doringbaai (DB) to Cape Agulhas (CA), with K3 found only in De Hoop (DH). Haplotype distributions for E. maxima support similar geographic patterns (Fig. 1A). A western group comprised of a private haplotype at Diaz Point (H12), closely related to the haplotype (H11) was found in locations sharing the same nuclear cluster (PN up to HK) and in populations further south (DB, KM). Another private haplotype (H13) was found in this region, only in Jacobs Bay (JB). A second geographic group was found between populations from Millers Point (MP) to Quoin Point (QP), with haplotype H10 being shared and corresponding to nuclear cluster K2. Finally, a third group of closely related haplogroups (H19, H20) is distributed from Muizenberg (MZ) to De Hoop (DH), with a private haplogroup present in Cape Agulhas (H21).

Mitochondrial DNA from deeper populations diverged from coastal ones in both the haplotype network (Fig. 1B) and in terms of average genetic distance (Table 3). Vema Seamount (VM) and Rocky Bank (RB) presented private haplotypes that diverged from both E. maxima and E. radiata (H15–17 and H22, respectively). VM haplogroups were closely related to those found in the Cape Town region, (i.e. H10 at Millers Point, Bordjiesrif and Saint James; H14 at Muizenberg). In contrast, putative hybrids from deeper waters at BB (ERHBOT, EXBB) were found with E. radiata haplotype (H5), showing incongruence between nuclear and mitochondrial markers.

Table 3 Pairwise mitochondrial sequence divergence (%) at the COX1 locus, between species and geographic groups of Ecklonia based on COX1 sequences.
Full size table

In the same STRUCTURE analysis, E. radiata was split into two genetic clusters with a clear geographic break (Fig. 1C). A southern K4 cluster grouped populations from Port Elizabeth (PE) to Dwesa (DW) and a second K5 cluster grouped populations from northeastern South Africa (MK and SL), Mozambique (MOZ) and Australia (AUS). The distribution of haplotypes in E. radiata were not completely consistent with these discrete nuclear genetic clusters. One haplotype (H1) is shared across most of the distribution range of E. radiata, spanning from Australia (AUS) to Port Elizabeth (PE). Only four populations do not share this haplotype: St Lucia (SL), Three Sisters (TS), Port Alfred (PA) and Abrolhos Island (AB) displayed private haplotypes. Additionally, a unique haplotype was found in Oman. A single haplotype (H5) is shared between Port Elizabeth and De Hoop. The sample from De Hoop (ERSDH03) with this haplotype exhibited a high level of genome admixture from E. maxima and E. radiata in the K = 2 STRUCTURE analysis and was placed within the E. radiata cluster in the FCA analysis (Fig. 1A–D). Geographic patterns for both species are corroborated by population differentiation as measured with FST and Jost’s D (Supplementary Table S4).

Specimen identification

The combination of nuclear and mitochondrial markers was essential for the identification of individuals sampled in interspecific contact zones and in deeper offshore populations because there were clear instances of mito-nuclear discordance as well as between morphological and genetic species identification. Deeper individuals in BB (ERHBOT, EXBB) had E. radiata haplotypes (H5) but clustered within E. maxima in the STRUCTURE analysis with K = 2, showed great admixture between species at K = 4 and were a diverging group in the FCA (Fig. 1B–D). A similar nuclear-mitochondrial discordance was found for one individual from the De Hoop contact zone (ERSDH03). De Hoop clustered as a unique genetic group within E. maxima (K3, Fig. 1C) but one individual of this nuclear cluster had an E. radiata haplotype (H5, Fig. 1B). Genetic distances between these individuals with discordant genomes and other Ecklonia populations confirm them having a nuclear genome of E. maxima and mitochondrial genome of E. radiata, with low FST and Jost’s D relative to E. maxima and inversely low pairwise genetic distance to E. radiata for COX1. Both observations suggest hybridization in the two contact zones. Most samples from De Hoop were conclusively recovered as E. maxima with both nuclear and mitochondrial markers, regardless of their morphology (Fig. 1). The only exceptions were the 6 individuals labeled as ERSDH, for which nuclear markers showed high admixture (Fig. 1C), low genetic distances from both species (Supplementary Table S4), and this population had the only sequenced individual with an E. radiata haplotype (Fig. 1A,B; H5). Similar mito-nuclear discordance was found for the atypical samples from Bordjiesrif (EXBB, ERHBOT), which were recovered as part of the major E. maxima nuclear cluster (Fig. 1C, K  = 2) but possessed a unique E. radiata haplotype (Fig. 1A,B; H5).

Deeper populations from Vema Seamount (VM) and Rocky Bank (RB) produced distinct results. While both were recovered as E. maxima in the K = 2 analysis, with both having high levels of admixture between E. maxima clusters and some with E. radiata in the K = 4 analysis. The mtDNA analyses revealed unique haplotypes for VM which are closely related to E. maxima (H15–17, Fig. 1B). In contrast, the haplotype from RB was intermediate with respect to the two species, and pairwise genetic distances, both between species and geographic groups, confirm this. The Ecklonia sp. from RB is divergent from both E. maxima and E. radiata (Table 3), possibly representing a distinct Ecklonia lineage.

Species distribution models

The marine forests dataset retrieved a total of 37,105 and 466 occurrence records for E. radiata and E. maxima (Table S5), which after pruning resulted in 272 records. The models using the best combination of hyperparameters showed good potential for transferability (Cross-validation AUC > 0.83; Cross-validation Sensitivities > 0.92; Table S5) and the final predictions largely matched the known distribution of the species (Final AUC > 0.9; Final Sensitivities > 0.97; Table S5; Fig. 2), with the exception of E. maxima presence at De Hoop.

Figure 2
figure 2

Potential distribution of Ecklonia radiata and Ecklonia maxima predicted with Species Distribution Modelling. Predictions aggregated in equal area hexagons for better visualization. The model correctly predicted the presence of both species in the Buffels Bay area but was not able to recover the same result for De Hoop.

Full size image

Niche overlap analyses

The probability of an individual of E. radiata being found within the ecological niche of E. maxima was 29.72%, while the contrary, finding an individual of E. maxima within the niche of E. radiata was 66.63% (Fig. S1). Species distribution modelling showed E. radiata to have a higher tolerance to maximum temperature when compared to E. maxima (21.3 °C vs. 26.2 °C; Fig. S2).

The similarity test did not reject the null hypothesis of the background test; thus, indicating that both niches are less similar than expected from random niches drawn (Tables S6 and S7). Overall, the Bayesian probabilistic niche overlap analyses and the similarity tests provide evidence for niche differentiation between E. radiata and E. maxima, while indicating that E. radiata has a much broader ecological niche than E. maxima. Putative hybrids EXBB and ER-h-BOT niche was more likely to overlap with that of E. radiata, the same being true for the deep samplings from VM and RB, while ERSDH niche was more similar to that of E. maxima (Table S8).

Discussion

Our results reveal the presence of discrete genetic groups within both Ecklonia species along distinct geographical sectors and depths of the southern African coastlines spanning from Namibia to Mozambique. Besides general population differentiation, the molecular species assignment in some cases differs from expectations based on morphology. Furthermore, a few instances of mito-nuclear incongruence were found in the contact zones between the two Ecklonia species, accompanied by the presence of atypical morphological traits, suggesting the occurrence of species hybridization. Notably our analysis of nuclear and mitochondrial data indicates the existence of a putative third Ecklonia lineage at Rocky Bank, separate from coastal populations. Below we discuss the implications of these findings for the characterization of southern African Ecklonia.

Phylogeography of southern African Ecklonia

Ecklonia maxima and E. radiata were confirmed to have distinct and mostly discrete distributions along the coast of southern Africa. The northernmost extant distribution of E. maxima was confirmed at Diaz Point (Lüderitz) in Namibia, extending along the west coast of South Africa as far east as Cape Agulhas. We also discovered morphologically atypical E. maxima populations in deep offshore habitats (Vema Seamount) for the first time. Additionally, our molecular data confirms previous reports of E. maxima in Koppie Alleen, in the De Hoop Nature Reserve8,21 confirming this site as a contact zone between the species.

Ecklonia maxima showed marked genetic structure along its distribution, with three nuclear genetic groups and 12 mitochondrial haplotypes. The geographic distribution of this diversity is very pronounced, in line with some previously described phylogeographic breaks on the coast of South Africa67,68,69,70. Cluster K1 (with haplotypes H11 and H12) was found in the cool-temperate region of Namaqua, from Namibia southwards to the southwestern Cape bioregion71,72. The break between nuclear clusters K1 and K2 between Hondeklipbaai (HB) and Doringbaai (DB) is similar to that described for another dominant kelp species, Laminaria pallida73. The private haplotype of E. maxima in southern Namibia, and larger population differentiation values between this and the closest South African populations suggest limited connectivity, although the possibility that sampling was insufficient to detect most haplotypes is an alternative explanation. The formation of cyclonic eddies due to the convergence of currents near Lüderitz74,75 may contribute to the isolation of this population. The oceanographic conditions and marginal location of this population is also a possible cause for localized, non-random mating, thereby explaining the high inbreeding coefficient.

Cluster K2 is found in the southwestern Cape bioregion71,72, along with haplotypes H10–11, H13–14, and H18–20, respectively. This is where most of the diversity within E. maxima is concentrated, with populations displaying the highest overall levels of allelic richness, gene diversity and observed heterozygosity (Table 1). A small number of private alleles, low FST values and haplotype sharing (H10) point to significant population connectivity in the region. The Benguela current, coupled with rare long distance dispersal events involving fertile floating thalli76, can also explain the presence of K2 and H11 from Kommetjie (KM) up to Doringbaai (DB).

The dominant current in this region also helps to explain the presence of E. maxima in Vema Seamount and its genetic nature. Agulhas rings eddies shed by the Agulhas Current move towards the South Atlantic, and are known to reach Vema Seamount77. Together with the Benguela Current, favoring northern and offshore dispersal, it is very likely that gametophytes or rafting individuals of both E. maxima genetic clusters reached Vema Seamount and settled there. Indeed, E. maxima drift plants have been found up to 600 km east of its distribution8,32. However, the high nuclear divergence of this group and the presence of 3 private haplotypes in the area (despite the possibility that sampling was insufficient to detect most haplotypes) is striking and suggests an ancient and rare colonization of this region, where it has evolved independently for a substantial period of isolation from coastal populations.

Cape Agulhas acts as an important phylogeographic break for E. maxima, marking the split between nuclear clusters K2 and K3 and harboring a private haplotype. This break is well-established and marks the transition between the Southwestern Cape and the warm-temperate Agulhas bioregion67,69,70,78. The existence of a unique nuclear cluster (K3) at De Hoop suggests the isolation of this population. Currents near Cape Agulhas are complex, with strong westwards flow and weak and variable flow to the east79,80,81, connectivity between De Hoop and the nearest sampled populations (CA, PE) can be relatively low. Adding to this, De Hoop is mostly dominated by warm-water algae82 and species distribution modelling only predicted the presence of the warm-adapted E. radiata, meaning local conditions could be driving the divergence of this cold-adapted E. maxima cluster.

Genetic structure in E. radiata was less pronounced compared to E. maxima. Nonetheless, the two genetic clusters recovered have well-defined geographical distributions. Cluster K4 is distributed along the eastern limit of the warm-temperate Agulhas region and in the Wild Coast transition zone83,84,85, extending up to Mkhambathi, near the border of Eastern Cape and KwaZulu Natal provinces. While the boundary between the warm-temperate Agulhas region and the Natal bioregion has been difficult to define67,68,70,83, the break between nuclear clusters K4 and K5 corresponds to the northern limit of the Wild Coast transition region67,83,84. Cluster K5 has a sub-tropical to tropical distribution86,87,88, extending to Mozambique, and even to the eastern coast of Australia. The shared nuclear genetic cluster between South Africa and Australia is congruent with common E. radiata COX1 haplotypes between these regions12,89. Indeed, broad scale haplotype sharing (H1, from South Africa to Oman, western and eastern Australia, and Tasmania), suggests recent and/or rapid expansion in E. radiata. The presence of E. radiata in tropical and sub-tropical locations such as Mozambique, St Lucia and Moreton Island, Australia is noteworthy and may be facilitated by pockets of cooler deep reef oceanic habitats where light and temperature conditions are suitable for the species12,29, as known for other kelps5,6,76,90,91. These deeper populations might have played a critical role in the dispersal and equatorial crossing of E. radiata, as evidenced by extinct populations in Oman12. Genetic clustering and the haplotype distribution of E. radiata in South Africa can be attributed to the effect of the Agulhas Current, promoting gene flow along the eastern shore into suitable habitats, and subsequent genetic divergence through drift in small, isolated populations. Genetic pairwise differences between populations are mostly consistent with the previously described patterns, with FST and Jost’s D values increasing with population distance from southwest to northeast, and a significant increase at Mkhambathi, marking a phylogeographic break for E. radiata.

Introgression and speciation along the southern African coast

Our results challenge some previous concepts about southern African Ecklonia and further explain some unresolved questions. Molecular data confirmed reports of E. maxima at Koppie Alleen, in the De Hoop Nature Reserve8,21 and suggests it is currently the dominant kelp at this location. Although in general, both nuclear and mitochondrial markers identified only E. maxima in this region, a subset of individuals had clearly admixed nuclear genomes and one individual had an E. radiata haplotype. A similar pattern of mito-nuclear discordance was found for the atypical deeper samples from Bordjiesrif in the Cape Peninsula, which were part of the major E. maxima nuclear cluster but on further analysis presented a highly admixtured genotype and a unique E. radiata haplotype. These cases of marker discordance suggest that hybridization occurs when the species are in contact, as shown in nature between other Ecklonia species23 and experimentally between E. maxima and E. radiata16.

Multiple hybridization events have likely occurred, because putative hybrids from De Hoop have high admixture and do not diverge much from either parental species, while the atypical samples from Bordjiesrif seem to be evolving in isolation. The latter possess a unique haplotype (H5) and forms a unique nuclear cluster in the FCA. Additionaly, pairwise genetic distances were relatively high between each of these populations with each other and with all others, further indicating their isolation and the lack of gene flow with other Ecklonia genetic entities. Although we cannot exclude that geographical isolation and/or depth at Bordjiesrif might be driving this isolation, differences between putative hybrid populations here and at De Hoop may have a temporal basis, with hybridization being more recent in the range contact zone of De Hoop than in the deeper distinct population of Bordjiesrif. Alternatively, a process of genetic hitchhiking might also be responsible for the increased divergence of these populations, facilitating the fixation of the results from an ancient hybridization.

The frequency and extent of hybridization between these species is an important question. Hybridization may result in heterosis (increased fitness compared to one or both parental species92, as recently demonstrated in some kelp species18,93. Experimental E. maxima/E. radiata hybrids showed broad thermal optima for growth16, possibly giving them an adaptive advantage. Our niche similarity modelling indicated that putative hybrids from Bordjiesrif are more likely to occupy a niche more similar to E. radiata than E. maxima, while putative hybrids from De Hoop have a niche overlapping that of E. maxima. Scenarios in which more adapted hybrids outcompete the parental species may lead to the loss of unique genetic diversity.

The presence of a distinct genetic group in the De Hoop region indicates that these populations have had time to diverge from the nearest sampled E. maxima and thus may have been established for longer than previously suspected8. Ecklonia species exhibit considerable morphological plasticity, making species identification and delimitation based on morphological traits challenging23,25,94. However, while morphological plasticity has been widely documented in E. radiata, leading to multiple taxonomic updates and the synonymization of other species with E. radiata29,94,95,96,97, E. maxima has been considered morphologically stable across most of its distribution27. Recent modeling studies have demonstrated size reduction in E. maxima from shallow populations exposed to higher temperatures98. Changes in stipe morphology, such as hollow versus solid, correlating with abiotic factors were described in another South African kelp, Laminaria pallida27,98, previously considered a diagnostic character separating L. pallida and (now synonymized) Laminaria schinzii32. Lower growth rates have been associated with nutrient-poor conditions99,100,101, which could be expected in the warm shallows and intertidal pools of De Hoop. Furthermore, although species distribution modelling provided evidence for niche differentiation between the species, the presence of E. maxima at De Hoop was not predicted. Changes in morphology due to abiotic factors and/or hybridization may therefore have played a role in the appearance of atypical E. maxima individuals at De Hoop.

Finally, we provide evidence for the existence of a likely third lineage of Ecklonia along the South African coast. Although individuals from the deep Rocky Bank population were part of E. maxima by K = 2 (species) clustering and were at the edge of the E. maxima cluster in FCA analysis, samples from Rocky Bank had a unique mixed genotype when considering a K = 4. The mitochondrial haplotype from Rocky Bank (H22) was intermediate and divergent from both species, while pairwise mitochondrial genetic distances between Rocky Bank Ecklonia and either E. maxima or E. radiata were comparable to intraspecific divergence in Ecklonia21,25,35. Further support for this distinct genetic entity comes from elevated pairwise nuclear genetic distances between Rocky Bank and populations of either species. However, it remains unclear whether the RB population represents a previously described species or a new diverging species. Further investigation using additional molecular markers and multispecies coalescent approaches could provide a better insight into the nature of the divergence between the South African Ecklonia lineages and their delimitations.

An important discovery is the significant genetic diversity and divergence from coastal Ecklonia observed in populations sampled from deeper habitats at Vema Seamount (VM) and Rocky Bank (RB), rendering them a critical genetic component of the genus in Southern Africa. These and previously identified deep populations12,29,102 add to the growing realization of deep reefs as important genetic refugia5,6,76,91,103,104. Understanding the functional genetic divergence between coastal and deeper populations is of utmost importance considering the predicted trends and pressures on this unique biodiversity105.

Conclusions

We present the first comprehensive analysis of genetic diversity and population structure of the kelps Ecklonia maxima and E. radiata in southern Africa. Our findings confirm the presence of a deep E. maxima population located 1000 km offshore of the west coast of South Africa, on Vema Seamount. Furthermore, our results suggest an earlier-than-presumed expansion of E. maxima eastward to the De Hoop, potentially hindered in detection by morphological plasticity and hybridization in contact zones. This shift may have led to the replacement of the previously dominant E. radiata by E. maxima in the region. The known and predicted distribution of E. radiata was confirmed, notably deep populations in the Natal and Delagoa bioregions in South Africa and Mozambique. Intriguingly, our data also reveal the existence of a possible third Ecklonia lineage in the deep population of Rocky Bank, where both nuclear and mitochondrial markers exhibit unexpectedly high levels of divergence. These results underscore the importance of molecular data in identifying cryptic diversity, particularly in morphologically plastic genera capable of hybridization23,25,106,107. Further analysis that integrates taxonomy and genetics will be key for understanding the complex dynamics of hybridization in a region where ocean conditions are predicted to undergo complex and unique changes.