Main

Our limited understanding of the diversity and trait partitioning within Coffea liberica constrains its utilization and development. The taxonomic delimitation and identification of C. liberica continues to confound researchers and coffee value-chain stakeholders, with inconsistent and confusing use of scientific and vernacular names in published research, agriculture and the media. The current consensus of taxonomic and systematic study4,5,6,7 is that C. liberica is a single species, divided into two botanical varieties: var. liberica and var. dewevrei8. While this classification is generally accepted, it is also argued that it does not fully account for the morphological9,10, and potential molecular variation4,11,12, within the species, and thus requires further critical study2,3. An alternative viewpoint is that C. liberica represents a single species with no infraspecific taxa13. Simply put, what is Liberica coffee and does it represent one, or more, species?

A revised species delimitation for Liberica

Here, we demonstrate congruence across genomic, morphological and spatial analyses, and elucidate distinct evolutionary lineages14, supporting the division of C. liberica into three distinct species: C. liberica (Liberica), C. dewevrei (excelsa) and C. klainei, following the rules of nomenclatural priority15. C. klainei is a poorly known species, previously considered to represent a synonym of C. liberica5,16. With C. dewevrei and C. klainei reinstated, the total number of known Coffea (coffee) species increases from 131 (ref. 8) to 133. Cameroon gains two species (now 18 species in total) and becomes the African country with the highest number of indigenous species, followed by Tanzania (17 species) and second only to Madagascar with 67 species8.

Phylogenomic analyses

We used the Angiosperms353 target capture kit17,18 to elucidate phylogenomic relationships within C. liberica sensu lato and related species. This genomic tool resolves relationships at various levels of taxonomic hierarchy in flowering plants19,20, including those at the population scale21. We sequenced 353 nuclear genes from 55 accessions (Supplementary Table 1). Across all accessions, we recovered 68.7–91.2% (mean 86.5%) of the 353 genes, representing 196,392–262,188 (mean 248,453) bp per sample recovered across the sample set. The mean number of gene sequences missing per species in the final alignments was 0.17, ranging from 1–10. Overall, there were 3.17% missing data in the alignments. These statistics exclude C. magnistipula, for which only small percentage of the 353 genes were recovered (see below). Inferred relationships for C. liberica sensu lato and allied species are shown in Fig. 1, based on a species tree obtained from a multispecies-coalescent ASTRAL-III analysis. Pie charts are placed at the nodes, showing the quartet scores (QSs) (maximum of 1.00), informing on the agreement between genes in the most likely phylogenomic topologies reconstructed, and the local posterior probability (LPP) scores (maximum of 1.00). The bootstrap (BS) values are provided in a supermatrix tree in Supplementary Fig. 1.

Fig. 1: Inferred relationships for C. liberica (Liberica), C. dewevrei (excelsa), C. klainei and allied species based on sequencing of Angiosperms353 nuclear genes.
figure 1

An ASTRAL tree. The pie charts show the QSs informing on the agreement between genes with LPP scores. For BS values, see Supplementary Fig. 1. C, cultivated (accessions from farms or germplasm collections); W, wild (accessions from natural (indigenous) populations). Accession information and country codes are provided in Supplementary Table 1.

Full size image

ASTRAL-III analysis grouped all C. liberica sensu lato accessions into a single clade (QS 0.63, LPP 0.36, BS 99), which subdivides into three monophyletic clades: C. liberica (QS 0.35, LPP 0.50, BS 100), C. klainei (QS 0.39, LPP 0.93, BS 99) and C. dewevrei (QS 0.60, LPP 1.00, BS 100). C. liberica and C. klainei were retrieved as a clade (QS 0.50, LPP 1.00, BS 60), sister to the C. dewevrei clade. These results support the phylogenomic distinction of the three species. The relatively low QS and LPP values are addressed in the single-nucleotide polymorphism (SNP) analyses section below. The remaining Coffea species fall outside the aforementioned clades, including a sister clade of five endemic species from Cameroon (C. bakossii, C. rizetiana, C. leonimontana, C. mapiana and C. montekupensis), C. humilis (indigenous to Guinea, Sierra Leone, Liberia and Ivory Coast) and C. magnistipula (indigenous to Cameroon and Gabon). The outgroups (C. stenophylla, C. brevipes, C. eugenioides and C. mannii) fall into positions that are congruent with published phylogenetic analyses of Coffea6,12,22,23. We were only able to recover a small percentage (19.7%) of 353 genes for the sample of C. magnistipula, and so this species was not included in the ASTRAL tree (Fig. 1). A separate ASTRAL-III analysis shows that this species falls between C. mapiana and C. humilis (Supplementary Figs. 1 and 2), as anticipated based on morphological, geographical and ecological similarities with C. mapiana24. No direct comparison can be made with previous molecular phylogenetic studies for C. liberica relatives6,22,23,25 owing to either their limited taxon sampling or low levels of sequence data or both.

SNP analyses

To investigate the genetic structure and relationships between and within C. liberica, C. klainei and C. dewevrei, we utilized 2,240 SNPs from the exon regions21 of 37 accessions (Supplementary Table 1). A genetic distance phylogenetic tree reconstructed with pairwise genetic distances supports the monophyly of the three species (BSs of 100, 91 and 100, respectively; Fig. 2a). C. liberica and C. klainei form a clade (BS 99) sister to C. dewevrei, consistent with the relationship recovered in the phylogenomic analysis (Fig. 1). Samples 23C16 (Nigeria) and 23H79 (Ivory Coast) form a clade (BS 92) sister to other C. liberica samples. On principal coordinate analysis (PCoA) (Fig. 2c) PC1 (28.2% variance) clearly separates the three species, with C. klainei positioned between C. liberica and C. dewevrei, with PC2 (6.4%) showing two outliers (23C16 and 23H79) for C. liberica. On the STRUCTURE26 analysis (Fig. 2b) we set the K value to K = 3, to match the phylogenomic analysis (Fig. 1), genetic distance tree (Fig. 2a), PCoA analyses (Fig. 2c) and geographical distribution (Fig. 3), representing three clusters (C. liberica, C. dewevrei and C. klainei). For K = 3, in C. liberica there is admixture from the C. klainei genetic group: 47.8% for 23C16 (Nigeria), 47.7% for 23H79 (Ivory Coast), 13% for 23C13 (Ghana) and 24% for 23C17 (Ivory Coast) (Fig. 2b). In K = 4 and K = 5, these four admixtures are not from C. klainei (Supplementary Fig. 3). These outliers require further investigation. K = 2 was identified as the most likely number of K genetic clusters26,27,28 (ΔK value of 3,605.7, compared with <ΔK of 37.54 for the remaining K values assessed; Supplementary Fig. 3). K = 2 was rejected on the grounds that it underrepresented population genetic structure for our study group29 and was incongruent with the known biological information available for the study species29.

Fig. 2: Genetic structure and relationships between and within C. liberica (Liberica), C. dewevrei (excelsa) and C. klainei, based on 2,240 exon region SNPs.
figure 2

a, A genetic distance phylogenetic tree reconstructed with pairwise genetic distances (the proportion of loci that are different). The figures above branches indicate the BS values (BS values <50 not shown). b, STRUCTURE26 analysis with the K value set to K = 3 to match the phylogenomic analysis (Fig. 1), genetic distance tree (a), PCoA analyses (c) and geographical distribution (Fig. 3), representing three clusters (C. liberica, C. dewevrei and C. klainei; see the main text and Supplementary Fig. 3 for alternative K values and details). c, PCoA analysis. PC1 (28.2% variance) separates the three species, with C. klainei intermediate to C. liberica and C. dewevrei. Accession information is provided in Supplementary Table 1.

Full size image
Fig. 3: Distribution map for indigenous (wild) C. liberica (Liberica), C. dewevrei (excelsa) and C. klainei.
figure 3

The colours of the symbols are matched to the genetic distance phylogenetic tree (Fig. 2a). Linked symbols represent a Rapoport’s mean propinquity assessment29 using a barrier distance of 500 km, which reveals a robust population separation between C. liberica versus C. dewevrei and C. klainei, but not between C. dewevrei and C. klainei.

Full size image

For K = 3 and K = 2, 4 and 5, there is either zero or minimal admixture between C. liberica and C. dewevrei. Low-to-moderate admixture at K = 3 for the accessions from Ivory Coast, Ghana and Nigeria (Fig. 2b) suggest features of historical evolutionary processes and shared common ancestry (with C. klainei), rather than recent introgression. C. liberica and C. klainei are distinctly allopatric, with populations separated by ca. 800 km (Fig. 3). There is no evidence of any introductions of C. klainei to upper West Africa, which rules out the possibility of human-assisted introgression. One of the cultivated samples (23H79; Nigeria, Lagos, 1895, Millen 192) with the highest admixture (Fig. 2b) predates the introduction of Coffea germplasm into upper West Africa (Supplementary Text). Sample 23H79 and 23C17 (Ivory Coast, Abidjan, 1963, De Wilde 156) are both cultivated, but samples 23C16 (Nigeria, Omo, 1946, Jones & Onochie 17214) and 23C13 (Ghana, Atewa Range Forest Reserve. 1963, Enti & Jenik 36571) are of wild origin. For K = 4 and K = 5, these four accessions retain admixture (Supplementary Fig. 3) but not with C. klainei. The admixtures observed for K = 3 (Fig. 2b) probably account for the low QS (0.35) and LPP (0.50) scores in the C. liberica clade from the phylogenomic analysis (Fig. 1), despite the clade achieving a high BS value (100) (Supplementary Fig. 1).

Morphological delimitation

We found that C. liberica, C. dewevrei and C. klainei are readily distinguishable using morphological characteristics (Table 1, Extended Data Figs. 1–4 and Supplementary Table 2). Compared with C. liberica, C. dewevrei has longer, broader leaves, more flowers (and thus more fruits) per leaf axil and node, fewer corolla lobes and flower parts per flower (usually five-merous flowers, versus six to nine-merous or more in C. liberica), smaller fruits with a thinner pulp (mesocarp), a thinner parchment (endocarp) and smaller seeds2 (Table 1, Extended Data Figs. 2–4 and Supplementary Table 2). The micromorphology of the seed epidermis and seed chemistry (diterpenes) may provide additional support for the separation of these two species10. C. klainei has a greater morphological affinity with C. liberica, but differs in having sessile, unbranched (as opposed to cymose and branched), inflorescences, with few flowers/fruits per inflorescence (usually one to three, versus two to six) and ellipsoid to narrowly ellipsoid fruits (versus spherical to ellipsoid). Experienced coffee professionals (for example, farmers and coffee buyers) can readily distinguish Liberica and excelsa based on the leaf dimensions (mainly overall size and shape and width), the fruit and seed size and yield (few or many per branch), in agreement with the morphological data presented here (Table 1 and Extended Data Figs. 1–4).

Table 1 Morphological characters distinguishing C. liberica (Liberica), C. dewevrei (excelsa) and C. klainei
Full size table

Indigenous distribution and elevation

After a thorough appraisal of ground point data (mostly for herbarium specimens) and focusing on the removal of cultivated and spontaneous (that is, self-sown into various habitats, but originating from cultivation) records, we demonstrate that the indigenous (wild) distributions of C. liberica, C. klainei and C. dewevrei are specific and allopatric (Fig. 3). C. liberica occurs in upper West Africa (Sierra Leone, Liberia, Ivory Coast, Ghana and Nigeria); C. klainei occurs in West-Central Africa (Cameroon, Gabon, the Republic of Congo and Angola (Cabinda)); and C. dewevrei in Central Africa (Republic of the Congo, Cameroon, the Democratic Republic of Congo, Central African Republic, South Sudan and Uganda). A Rapoport’s mean propinquity assessment30 using a barrier distance of 500 km demonstrates a robust population separation between C. liberica and C. dewevrei plus C. klainei (bold red line), but not between C. dewevrei and C. klainei (Fig. 3). The revised indigenous geographical range for C. liberica is comparable to that of two other Coffea species: C. humilis (Guinea, Sierra Leone, Liberia and Ivory Coast) and C. stenophylla (Guinea, Sierra Leone and Ivory Coast). This may infer shared drivers of Coffea species distributions in upper West Africa. C. liberica and C. klainei are predominantly located at low elevations (mean values of 386 m and 273 m, respectively), whereas C. dewevrei typically inhabits mid-elevations (mean of 653 m) (Supplementary Table 3).

Climate parameters

Identifying the climate parameters essential for growth, yield and plant health is critical for optimizing the cultivation and development of crop species. C. liberica (Liberica) and C. dewevrei (excelsa) are recent introductions to agriculture (<200 years old)2, although wild gathering and local use may be date back centuries and perhaps millennia. At best, these two species are minimally domesticated; C. klainei is undomesticated and has only been cultivated in germplasm collections on a few occasions (Supplementary Text). Climate variable data for these species, over the natural distributions, provides a useful starting point for understanding climate requirements in cultivation31,32. Summary data for the 19 Bioclims33 are given in Supplementary Table 3.

The following narrative focuses on notable similarities and differences for the two crop species C. liberica (Liberica) and C. dewevrei (excelsa), as indicated in Supplementary Fig. 4 and Supplementary Table 4 (with P values). The annual mean temperature (Bio1) values are nearly identical (24.6 versus 24.4 °C), although Liberica has a lower mean diurnal range (Bio2; 7.9 versus 8.9). The main differences between C. liberica and C. dewevrei are for precipitation (the respective mean values are given): mean annual precipitation (Bio12; 2,215 versus 1,678 mm), precipitation of the wettest month (Bio13; 376.6 versus 230 mm); precipitation of the driest month (Bio14; 15.9 versus 36.2 mm); precipitation seasonality (Bio15; 66.9 versus 49.3); precipitation of the wettest quarter (Bio16; 988.9 versus 643.8 mm); precipitation of the driest quarter (Bio17; 80 versus 141.1 mm); and precipitation of the coldest quarter (Bio19; 961.7 versus 597.7 mm). Although the mean annual precipitation for C. dewevrei is lower than for C. liberica, C. liberica experiences lower precipitation in the dry season (Bio14 and Bio17). The higher mean annual precipitation for C. liberica is due to higher precipitation in the wet season (Bio16), which corresponds to the wetter coldest quarter (Bio 19). These patterns, along with precipitation seasonality (Bio15), are consistent with mean annual temperature and total annual precipitation climate bar charts. Upper West Africa (C. liberica) has a longer and more severe dry season with proportionally higher precipitation during the wetter/cooler months of the year, whereas much of the distribution range of C. dewevrei (in central Africa) has a shorter dry season, or dry seasons (if annual rainfall is bimodal), with a more even annual distribution of precipitation (Supplementary Tables 3 and 4). Comprehensive field trials for C. liberica and C. dewevrei are required to ascertain the precipitation requirements in cultivation and particularly to test whether C. liberica is better adapted to a more seasonal rainfall pattern and is perhaps more drought tolerant than C. dewevrei.

C. dewevrei has the greatest range for most of the Bioclims, which may infer either, or a combination of, the following: (1) that this species has greater climate plasticity, (2) that there is a wider range of climate tolerance over the entirety of the metapopulation or (3) simply that the considerably larger distribution (compared with C. liberica and C. klainei) encompasses a greater range of data values. C. dewevrei frequently occurs in riverine and gallery forest types within savanna woodland landscapes2,3, where populations may gain access to belowground or perhaps even surface water, at least during certain times of the year. Water availability in these habitats may enable this species to exist in areas of lower precipitation (for example, <1,000 mm yr−1), biasing precipitation values (for example, those for this species (that is, below Q1; Supplementary Fig. 4 and Supplementary Tables 3 and 4)). In the wild, C. dewevrei occurs in both open-canopy and closed-canopy forest3.

Under the revised circumscription and reassessment of wild occurrences, the mean annual temperature value for C. liberica is 0.7 °C higher (at 24.6 °C) than previously reported for C. liberica sensu lato (23.9 °C)32, and C. dewevrei is 0.4 °C higher (at 24.4 °C) than previously reported (23.9 °C)32. The modelled mean annual temperature for C. liberica is 5.9 °C and 0.9 °C higher than naturally occurring Arabica (C. arabica: 18.7 °C) and robusta (C. canephora: 23.7 °C)32, respectively; and for C. dewevrei (5.7 °C and 0.7 °C, respectively). The modelled mean total annual precipitation for C. dewevrei (1,678 mm) is similar to Arabica (1,614 mm) and robusta (1,699 mm)32, although field observations suggest that C. dewevrei exhibits a considerable measure of drought tolerance, particularly compared with C. canephora2,3. Ultimately, multilocation field trials over a range of climates, using a range of genotypes, would be required to more thoroughly compare climate tolerances for these four crop species.

Implications for crop use and development

The species delimitations proposed in this study have implications for coffee crop development. Our genomic (Figs. 1 and 2) and phenotypic data (Table 1) reveal that C. liberica (Liberica) and C. dewevrei (excelsa) possess distinct alleles and unique allelic combinations of genes, as well as specific phenotypic (morphological) and climate characteristics. These attributes offer resources and utility for coffee breeding programmes.

The robust species delimitation identified enables the unambiguous partitioning of attribute data. For example, compared with C. liberica, C. dewevrei exhibits a higher yield2 owing to the number of fruits produced per tree, a higher outturn (that is, the conversion ratio of fresh fruit to clean (unroasted) coffee, mainly attributed to its thinner pulp2 and thinner parchment (endocarp)) (Table 1). In addition, the smaller seeds of C. dewevrei (Table 1 and Supplementary Table 2), which are similar in size and shape to Arabica2, make them more amenable to existing post-harvest, and preconsumption (roasting, packaging and coffee making) processes, as used for Arabica and robusta. Liberica and excelsa have contrasting coffee flavour profiles2, which influences consumer preferences and supports market differentiation. In addition to previously reported agronomic traits for these species, we report here that the parchment (endocarp) of C. dewevrei is conspicuously thinner than C. liberica (mean values of 0.31 versus 0.57 mm; Table 1, Extended Data Fig. 1 and Supplementary Table 2), which, in combination with a thinner pulp (mesocarp), improves outturns and ultimately the profitability of the harvested crop compared with Liberica.

Our climate analyses show that indigenous C. dewevrei is adapted to a lower mean annual rainfall, compared with C. liberica (Supplementary Fig. 4 and SupplementaryTables 3 and 4), but that C. liberica might be better adapted to higher precipitation seasonality and thus longer dry seasons, with periods (1–4 months) of relatively low precipitation. C. liberica occurs at lower elevations (mean of 386 m), whereas C. dewevrei is located at mid- to high elevations (mean of 386 versus 654 m; Supplementary Table 3). In cultivation, Liberica is mostly farmed at low elevations (10–500 m), in warm to hot (for example, a mean annual temperature of 24–27 °C) and wet (for example, mean annual precipitation of 2,000–4,000 mm) habitats, with low precipitation seasonality (short or indistinct wet season(s)), such as those in lowland regions of Malaysia, the Philippines and Indonesia. In contrast, excelsa is principally farmed at mid-elevations (500–1,200 m), with cooler temperatures (a mean annual temperature of 22–25 °C and with lower mean annual precipitation (1,500–1,800 mm), for example, in Guinea, South Sudan and Uganda. Notably, these species are rarely cultivated together or in overlapping agroecological zones. In tropical Central Africa34,35, Peninsula Malaysia and Sarawak (K. Lee Wing Ting, D. Jitam & A. Clayre, personal communication) and Java36 cultivated C. liberica flowers and fruits throughout the year. It is not certain whether this is the same for wild populations of C. liberica. By contrast C. dewevrei has a distinct flowering and fruiting seasons, although it is not known whether phenology would be disrupted under the conditions stated above for C. liberica.

Over their indigenous ranges, both species are likely to include populations with adaptations to regional climate differences, other abiotic factors (for example, soil), and various biotic interactions (for example, pest and disease incidence and resistance). This may particularly be the case for C. dewevrei, which has a large natural distribution range across tropical Central Africa. Importantly, Liberica and excelsa hold substantial potential for developing coffee farming in areas that are unsuitable for Arabica or robusta2,32, particularly those at low elevations in hotter and wetter climates (with higher mean annual temperatures and different annual precipitation patterns, see above). They may also have potential as a replacement coffee crop in areas that are becoming climatically unsuitable for Arabica and robusta. Excelsa has been used to replace robusta in some areas of Uganda, probably as result of climate change, for example, refs. 2,3.

Given the close phylogenomic relationship between C. liberica and C. dewevrei (Fig. 1) the production of fertile interspecies hybrids is likely37, either artificially (by hand or close-proximity cross pollination) or by chance38. C. dewevrei × C. liberica hybrids have been reported to be of outstanding vigour and yield38, although the existence of hybrids has not yet been verified by molecular methods. The use of either species in interspecies breeding programmes with other species may hold promise3.

Online sources regularly state that ‘Liberica’ (that is, C. liberica and C. dewevrei) provides 1–2% of the global coffee supply. This is incorrect, as these percentages are based on figures from the late-nineteenth century, when C. liberica stood with Arabica as the second most important coffee of commerce2,39. By the mid-twentieth century, Liberica was reported to represent about40 or less than41 1% of global production, respectively. Today, global production of C. liberica and C. dewevrei is probably less than 1,000 metric tons (mt). Based on the figures for global exports of Arabica and robusta, which combined was around 10 million mt for 20241, an estimate of production of 1,000 mt would represent 0.01% of global coffee exports. Despite this seemingly insignificant figure, Liberica and excelsa production is now being upscaled2, particularly in Uganda, South Sudan, India, Vietnam, Malaysia, the Philippines, Indonesia and even the Pacific.

Extinction risk

Under our revised taxonomic circumscription, and refinement of indigenous distribution (Fig. 3), the extent of occurrence (EOO) for C. liberica decreases from 6,812,900 km2, as per the existing International Union for Conservation of Nature (IUCN) Red List assessment42, to 352,310 km2, which represents a reduction of 94.8%. The area of occurrence (AOO), is similarly affected, decreasing from 736 km2 to 52 km2, a reduction of 92.9%. The current IUCN Red List assessment reports that C. liberica occurs naturally in 17 countries42; our revised species delimitation reduces this to five: Sierra Leone, Liberia, Ivory Coast, Ghana and Nigeria. In all these countries, except Liberia, forest loss has been ongoing and in some cases severe, even over the past two decades43. It should be made clear, that the conservation metrics given above are based on historical records only, mainly from 1900–1980. Over the past 45 years, many of the populations and subpopulations recorded during that period have been extirpated or reduced in area and health owing to deforestation and other land-use changes. For example, in Sierra Leone, targeted searches for wild coffee species at Kasewe Hills Forest Reserve in 2023 and 2024 failed to locate C. liberica (Lebbie personal communication, 2024). To our knowledge, it was last recorded at Kasewe Hills in 1913 (herbarium specimen: Lane-Poole 128, 1913, K). Conversely, there are likely to be additional populations and subpopulations in Liberia, which has considerably more natural forest than the other four countries and is under-botanized. Given an AOO of less than 2,000 km2 (52 km2 for C. liberica), evidence of severe fragmentation (via Google Earth imagery) and continuing declines in AOO and EOO, C. liberica may warrant reassessment as a species threatened with extinction, shifting from its current classification of ‘Least Concern’42 to ‘vulnerable’ (VU B2 (a,b(i–v)))44. C. klainei has an AOO of 76 km2, and might also qualify as ‘Vulnerable’ under IUCN Red List criteria44, although further data would be required, for both species, before confident IUCN extinction threat assessments could be made. With an EOO of 2,464,990 km2 and an of AOO of 456 km2, C. dewevrei falls into the ‘Least Concern’ category, even with the observed forest loss evident in many areas of its distribution3. Regardless of the conservation metrics presented here, enhanced conservation measures are urgently needed for all three species, and particularly for C. liberica, to ensure their survival in the wild and potential role in global coffee sustainability.

Methods

DNA sampling

We sampled 12 accessions of C. liberica and 22 of C. dewevrei, sourced from wild and cultivated populations (farmed or germplasm collections), and 15 accessions representing all eight closely related species, as identified by prior molecular phylogenetic analyses6,22,25 and taxonomic study5,16,24. Material representing the geographical range of validly published synonyms for the two foci species were included in the DNA sampling (Supplementary Table 1 and Supplementary Text 1 and 2). Outgroup taxa from within Coffea (five accessions) were selected based on previous molecular phylogenetic analyses6,12,22,25. DNA was extracted from 26 herbarium leaf tissue samples, 13 silica-gel dried samples and 16 seed samples. Sampling details, other accession information and sequence information are provided in Supplementary Table 1. Accepted botanical names and authorities follow the International Plant Names Index (https://www.ipni.org).

DNA sequencing and phylogenomic analysis

Total DNA was extracted using a modified CTAB protocol for herbarium specimens45. Sequence target capture data were generated using the universal Angiosperms353 target capture kit developed to retrieve 353 nuclear genes across the angiosperms17,18. Genomic libraries were constructed using an optimized protocol46 for half volumes of the NEBNext Ultra II DNA Library Prep kit for Illumina (New England Biolabs) and purified using AMPure XP magnetic beads and multiplexed using NEBNext Multiplex Oligos for Illumina (Dual Index Primer Sets I and II). Pools containing 55 genomic libraries mixed in equimolar conditions were enriched with half reactions of the Angiosperms353 probe kit following the myBaits kit manual v.3.02 (Arbor Biosciences), using an optimized protocol47. The DNA concentration and fragment size distribution were calculated using a Quantus fluorometer (Promega Corp.) and an Agilent 4200 TapeStation (Agilent Technologies), respectively. Sequencing was performed on a HiSeq (Illumina Inc.) by Macrogen, producing 2× 150 bp paired-end reads. Raw reads were submitted to the European Nucleotide Archive (https://www.ebi.ac.uk). ID codes are given in Supplementary Table 5. Trimmomatic v.0.35 (ref. 48) was used to discard low-quality reads and trim adaptors based on the reports generated by FastQC v.0.11.7 (ref. 49) and HybPiper v.2.3.0 (ref. 50) to retrieve the 353 nuclear loci using a combination of map to reference and de novo assembly methods for all samples. Alignments were generated in MAFFT 7.305b (ref. 51) using the command ‘auto’, then edited with trimAL v.1.4.rev22 (ref. 52) using the ‘automated1’ parameter.

Phylogenetic analyses were conducted for the concatenated and partitioned dataset of nuclear data (that is, the supermatrix approach) and by estimating a species tree from individual phylogenetic trees reconstructed for each nuclear locus independently (that is, multispecies-coalescent approach). Phylogenetic trees were reconstructed using RAxML-NG53 and IQ-TREE54 with 1,000 BS replicates. ModelFinder determined the optimal substitution model (-m MFP), selecting GTR + GAMMA55 on the best BIC score. Concatenated datasets were built with FASconCAT-G v.1.04 (ref. 56). ASTRAL‐III57 was used to construct a species tree based on the independent nuclear gene trees and the ASTRAL-III phylogenetic topology as input for the supermatrix approach. Support values were assessed using LPPs, with branches deemed supported if their LPP exceeded 0.95. To evaluate incongruence among gene trees, a quartet-based analysis was performed using the -t 8 option in ASTRAL-III, enabling the identification of the proportion of genes supporting alternative topologies at each node.

SNP production and analyses

To generate SNP data for C. liberica, C. dewevrei and C. klainei (38 accessions), we used the framework developed by DePristo et al.58 using GATK59 following the pipelines established for Angiosperms353 data21,60. This process involved combining aligned and unaligned reads to a reference built with the longest exon obtained for all samples of the three species. We removed duplicate sequences and performed joint genotype calling for all samples after initially generating variants for each sample individually61 in a variant call format (VCF) file. The initial VCF file was processed with a stringent filter (QD < 5.0 || FS > 60.0 || MQ < 40.0 || MQRankSum < −12.5 || ReadPosRankSum < −8.0), removing indels and SNPs with missing data using GATK and eliminating linked SNPs with PLINK62. Base quality score recalibration was performed in GATK, followed by a repeated variant calling step.

To examine genetic differentiation patterns, we used STRUCTURE26 to determine the optimal number of genetic clusters in the dataset. STRUCTURE was run for five potential clusters (K), corresponding to the number of assumed genetic groups plus two. Each K was analysed with ten replicates, using 100,000 burn-in iterations followed by 1,000,000 Markov chain Monte Carlo repetitions. The most likely number of clusters was identified using Structure Harvester27, which implemented the method of Evanno et al.28 to calculate ΔK values. The K with the highest ΔK value was selected as one probable assessment of the number of clusters, and K = 3 was examined as an alternative, based on the number of clades obtained with the phylogenomic analysis (Fig. 1), genetic distance phylogenetic tree (Fig. 2a) and geographical distribution, heeding the issues raised regarding K = 2 and recommendations for exploring population subdivision29. K = 2, 4 and 5 were also examined. PLINK outputs were converted into STRUCTURE-compatible files using PGDSpider63. The results from STRUCTURE were visualized with StructRly64. Genetic differentiation was explored using a genetic distance-based phylogenetic tree using upgma with poppr 2.9.6, adegenet, ape 5.7.1 and RColorBrewer packages (Fig. 2a) and PCoA in R 4.3.0 (ref. 65), with the adegenet and ggplot2 packages, retaining three principal components to investigate genetic groupings (Fig. 2c).

Morphological study

Morphological characters (Table 1) were measured from herbarium specimens (held at The Natural History Museum, London, UK (BM); Meise Botanic Garden, Belgium (BR); Royal Botanic Gardens, Kew, UK (K); Muséum National d’Histoire Naturelle, Paris, France (P) and Naturalis Biodiversity Center, Leiden, Netherlands (WAG)66) and living plants. More than 700 herbarium specimens, encompassing wild and cultivated C. liberica, C. dewevrei and C. klainei, were examined. Material representing all validly published synonyms of the three taxa were comprehensively studied (Supplementary Text 1 and 2). Living plants were studied in the wild and in cultivation settings across Africa, Madagascar and Asia. Parchment measurements were taken using a Mitutoyo no. 193-111, 0–25 mm (0.001 mm) micrometer. Seed measurement data were taken from published work2,36.

Distribution and conservation assessment metrics

Data for the production of the distribution map, climate profiling (see ‘Climate profiling’ section) and conservation metrics for C. liberica, C. dewevrei and C. klainei were gathered from occurrence data points representing indigenous (wild) locations derived from herbarium specimens (BM, BR, K, MHU, P and WAG66) and field surveys. Georeferencing was performed for locations lacking coordinates, followed by manual validation and correction using Google Earth imagery. The dataset comprised 311 records, including 19 for C. liberica, 267 for C. dewevrei and 25 for C. klainei; after the removal of duplicate locations (within 1 km of each other), this was reduced to 152 data points (13, 119 and 20, respectively). The distribution map (Fig. 3) was produced in ArcGIS Pro 3.2.0 (Environmental Systems Research Institute) using Natural Earth (https://www.naturalearthdata.com) and their terrain and country boundaries dataset (version 5.1.1). IUCN Red List conservation metrics44, were produced using ShinyGeoCAT67, with default settings aligned to IUCN methodology and criteria44, providing the EOO (that is, a minimum convex polygon enclosing all occurrences) and AOO based on at least one occurrence in a 2 × 2 km grid cell (that is, the IUCN default44). A Rapoport’s mean propinquity assessment30 with a barrier distance of 500 km was used to test for population and subpopulation separation.

Climate profiling

To understand the key climate parameters for each species, the statistics package R65 was used to sample the same dataset as above, against 19 Bioclim variables33 from the CHELSA dataset68. We reviewed all 19 Bioclim variables (Supplementary Fig. 4 and Supplementary Tables 3 and 4). For validation purposes, the modelled Bioclim data were compared against publicly available mean annual temperature and total annual precipitation climate bar charts.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.