Origins and close relatives of a semi-domesticated neotropical fruit tree: Chrysophyllum cainito (Sapotaceae)^†

Taxonomic sampling

Fieldwork was conducted from November through May in each of the years 2006–2010, primarily during the fruiting season of C. cainito. We collected fresh leaf material in the Greater (Dominican Republic, Jamaica) and Lesser (the U. S. Virgin Islands) Antilles, northern Mesoamerica, including Mexico (Veracruz and the Yucatan Peninsula) and Guatemala, and southern Mesoamerica, including Costa Rica and Panama. All material was geo-referenced, and identifications based on morphology were made by examining herbarium material and with the help of local experts. We deposited voucher specimens at the following herbaria: JBSD, IJ, CICY, and CR, with duplicates deposited at DAV. Herbarium acronyms follow 58.

For phylogenetic analyses of section Chrysophyllum, we included a total of 20 species and 51 individuals (Appendix 1). For six species of Chrysophyllum section Chrysophyllum, we included multiple individuals collected from a broad part of the geographic range: C. cainito (N = 8), C. argenteum (N = 15) C. mexicanum (N = 3) and C. oliviforme (N = 9). Specimen JP57, labeled as C. bicolor, was collected in the U. S. Virgin Islands. Additional taxa, including South American (Brazil, Ecuador, French Guiana, and Guyana), African, and Asian species, were represented in the phylogenetic reconstruction using sequences obtained from GenBank. To test the monophyly of section Chrysophyllum, we included 10 species from four additional sections of Chrysophyllum, as well as two species of Pouteria section Pouteria, since previous studies have shown that both Chrysophyllum and Pouteria are polyphyletic (see citations in 56). One specimen of Manilkara sideroxylon, subfamily Sapotoideae, was included as an outgroup to root the trees.

For the phylogeographic analyses, we included 81 (51 cultivated and 30 putatively wild) individuals of C. cainito from 58 discrete localities, with 1–4 individuals per locality (Table 1, Appendix 2). We categorized each C. cainito locality according to the cultivation status of the trees collected there. We considered trees to be “wild” if they were collected from natural, undisturbed (including both primary and secondary) forest outside of human settlements, where it was unlikely that the trees had been planted and, to the best of our knowledge, they were from nonanthropogenic sources. In some cases, we considered trees to be “wild” when found in human-managed landscapes if they were obviously remnants from what had been a wild-forested area. Examples include a recently cleared coffee planting in Jamaica and mature cacao groves in the Dominican Republic where, because of the practice of “let stand”, wild trees are spared during forest clearing.

We considered trees to be “cultivated” if we collected them from a setting in which they were clearly tended by people, including backyard gardens, small-scale orchards, or mixed perennial plantings such as parcelas. In each case, we asked local participants for the source of the material and whether the trees had been planted, resulting in two subcategories of “cultivated” trees. Trees that had not been intentionally planted but were nonetheless protected and tended were classified as “spontaneous”, while those that had been intentionally planted were designated as “planted”. Therefore, the material found in cultivated environments may come from a variety of sources and may represent remnant wild populations as well as cultivated gene pools.

DNA extraction, amplification, sequencing, and cloning

We performed DNA extractions of both fresh and silica-dried material using the DNeasy Plant Minikit (Qiagen, Valencia, California, USA). Polymerase chain reaction (PCR) amplification of the nuclear ribosomal ITS region was performed using the ITS6 (5′-tcgtaacaaggtttccgtaggtga-3′) and ITS9 (5′-ccgcttattgatatgcttaaac-3′) primers reported in 43. These primers were used in sequencing reactions along with the internal sequencing primers ITS 2 and ITS 3 (62). For the phylogeographic analysis, we amplified the chloroplast regions, trnS-fM and rps16 intron (51). We performed PCR reactions of a 50 µL volume using: 5 µL 10× buffer, 3.8 µL MgCl₂ and 0.5 µL Amplitaq (Applied Biosystems, Foster City, California, USA); 1 µL each of the forward and reverse primers (10 µmol/L), and dNTPs (10 mmol/L), 17 µL of 5 mol/L betaine in 18.7 µL H₂0 and 1 µL genomic DNA (20 ng/µL). PCR amplifications were carried out on an Applied Biosystems 2720 thermal cycler using the following cycling parameters: a 94°C initial denaturing for 2 min; followed by 35 cycles of 94°C for 45 s, 56°C for 45 s and 72°C for 2 min; with a final extension of 72°C for 10 min. We used the Qiaquick gel extraction kit (Qiagen, Valencia, California, USA) to purify amplified fragments, and sequenced them directly using ABI BigDye Terminator v3.1 Cycle Sequencing chemistry on an ABI 3730 Capillary Electrophoresis Genetic Analyzer (Applied Biosystems).

Although each PCR reaction produced only a single band on agarose gels, some samples of C. cainito exhibited mixed sequences in the ITS region when PCR products were sequenced directly. In those cases, the PCR products were excised from agarose gels and were then cloned using the pCR 2.1 TOPO-TA kit (Invitrogen, Carlsbad, California, USA). Five to eight plasmid colonies per individual were selected and purified using the Qiaprep Spin Miniprep Kit (Qiagen).

We assembled the forward and reverse sequences for each accession using Sequencher 3.1.1 (Gene Codes Corp., Ann Arbor, Michigan, USA) and used the program Clustal X (59) to perform multiple alignments, with subsequent manual refinement carried out in the program Se-Al (44). New DNA sequences generated for this study were deposited in GenBank (accession numbers JF912903–JF913194; Appendices 1 and 2). The DNA sequence alignments can also be found in the Supplemental Data (Appendices S1–S3) with the online version of this article.

Phylogenetic analysis of ITS sequence data

For maximum parsimony (MP) analyses, heuristic searches were implemented in the program PAUP* version 4.0b10 (57) using 2000 replicates of random taxon addition sequence with tree-bisection-reconnection (TBR) branch swapping, saving 100 trees/replicate. We treated all characters as unordered and with equal weight. Gaps were coded as missing data for the ITS region. We conducted a bootstrap analysis (15) to test the relative support for clades, using 1000 bootstrap replicates and 10 random taxon addition sequence replicates per bootstrap replicate. To reduce time spent in swapping on large numbers of suboptimal trees, we held no more than 10 trees per random sequence addition replicate. Bootstrap support values of less than 50% at a node are not reported. We consider 75–89% bootstrap support to be moderate support of a clade and 90–100% as strong support.

Model selection for likelihood-based analyses was implemented in the program jModeltest 0.1.1 (41; see 18), which evaluates 88 models, estimating a maximum likelihood (ML) tree under each model. We used the Akaike information criterion (AIC) (2) to choose among models and the generalized time reversible model with rate variation among sites (GTR +G) was selected as the best-fit model.

We conducted Bayesian analyses in the program MrBayes 3.2 (46), with two simultaneous independent runs. In each analysis, we ran four Markov chain Monte Carlo (MCMC) chains (three heated and one cold) for 2 to 10 million generations, starting with a random tree topology. Trees were sampled every 1000th generation. Heated chain temperature was set at the default value. We assessed MCMC chains as having converged when the log-likelihood values of the cold chain fluctuated within a stable range. Within a single analysis, we considered a value of <0.01 for the average standard deviation of split frequencies as evidence that the two runs had converged to a stationary distribution (47). We plotted the log-likelihood values of the generations to determine burn-in, which was roughly 10% of the total number of trees per run. We conducted four to six independent analyses to confirm that the log-likelihood scores had converged around an equilibrium value and that the tree topologies being recovered were consistent. A majority-rule consensus tree was generated using the postburn-in trees. Posterior probability was assessed as the frequency of a node among the sampled trees.

Phylogeographic analysis of Chrysophyllum cainito and congeners based on ITS and plastid sequence data

We further examined the inter- and intraspecific relationships resulting from the phylogenetic analysis (Fig. 1) by constructing haplotype networks of plastid and nuclear DNA sequence data using a statistical parsimony framework implemented in the program TCS ver.1.21 (10). We included 93 individuals, comprising 81 individuals of C. cainito (see above), plus at least one individual for each identified specific and intraspecific taxon from each of the clades within section Chrysophyllum from the phylogenetic analysis of ITS data (clades A1, A2, A3, and clade B from Fig. 1): C. argenteum subsp. panamense (N = 1), C. argenteum subsp. argenteum (N = 2), C. argenteum subsp. auratum (N = 1), C. oliviforme subsp. oliviforme (N = 2), C. oliviforme var. picardae (N = 1), C. mexicanum (N = 2), C. pauciflorum, (N = 1), C. bicolor (N = 1) and Chrysophyllum sp. Costa Rica (N = 1). Each of these individuals is indicated by an asterisk in Fig. 1.

For the ITS data set, we treated gaps as missing data. The plastid data were combined and treated as a single locus. Gaps in the plastid haplotype networks were coded as simple insertion/deletion events (indels) (52).

Phylogenetic analysis of ITS sequence data

Phylogeographic analyses

We calculated haplotype networks for ITS and combined plastid DNA sequences based on data from the 81 individuals of C. cainito and 12 additional individuals, representing six congeneric species, for a total of 93 individuals representing seven species (Table 2).

Haplotype network of C. cainito and congeners based on ITS sequence data

We analyzed 676 bp of ITS sequence, treating gaps as missing data. Within the C. cainito samples, three sites were polymorphic and parsimony-informative (sites 342, 515, and 545). Several individuals of wild C. cainito trees from Panama were polymorphic at one, two, or three of these sites (a C/T polymorphism at site 342; a G/C polymorphism at site 515; and a T/C polymorphism at site 545), exhibiting a double chromatogram peak on both the forward and reverse strands. In such individuals, the only variable sites were all ambiguous. When inferring haplotypes from sequences where the variable sites are all ambiguous, the program TCS (9) randomly assigns each of these sequences into one of the existing haplotypes that contains the same variable but nonambiguous sites.

We recovered a highly bifurcating network of 14 haplotypes using a 95% connection limit and fixing the connection limit to 100 steps (Fig. 2). This network was highly congruent with the topology of the phylogenetic reconstruction (Fig. 1). The network construction contains three haplotypes that were not sampled in our data set and were inferred as missing intermediates (42) and one closed loop due to a homoplasious nucleotide substitution. Each species sampled had at least one species-unique haplotype (Chrysophyllum sp. ITS-VII, C. pauciflorum ITS-XIII, C. bicolor ITS-XIV) and intraspecific variation was represented by additional unique haplotypes in cases where multiple individuals were included (C. cainito ITS-I, ITS-II; C. argenteum ITS-III, ITS-IV, ITS-V, ITS-VI; C. oliviforme ITS-VIII, ITS-IX, ITS-X; C. mexicanum ITS-XI, ITS-XII).

As noted already, in the results of direct sequencing of PCR products, some individuals exhibited overlapping peaks at one or more of the polymorphic sites. To determine whether these represented additional unique haplotype variation, we chose three individuals that exhibited the mixed sequence (JP142, JP188, and JP201) and cloned and sequenced five to eight colonies per individual. The cloned sequences were added to an ITS alignment of C. cainito and manually inspected. We did not find any new haplotype variation within the sequenced clones. Rather, all of the cloned sequences corresponded to either the ITS-I or ITS-II haplotype, indicating that each of the individuals (JP142, JP188, and JP201) is effectively heterozygous, carrying both the ITS-I and ITS-II haplotypes. We refer to individuals that carry both the ITS-I and ITS-II haplotypes as carrying the ITS-I/II genotype. In the network construction, each C. cainito sequence with ambiguous sites was assigned randomly into either the ITS-I or the ITS-II haplotype (see above).

The two haplotypes recovered in C. cainito, ITS-I and ITS-II, were separated by three nucleotide substitutions. ITS-I was the more common haplotype and was carried by both wild and cultivated trees collected from throughout the range (N = 60). All of the trees collected in Jamaica, the Dominican Republic, and Guatemala, and all but one tree each from Mexico and Costa Rica carried haplotype ITS-I, as did four trees collected in Panama, one tree from Colombia, and two trees from Papua, Indonesia. Ten individuals (eight from Panama, one from Costa Rica and one from the Yucatan Peninsula) had the ITS-II haplotype. The ITS-II haplotype was recovered in wild trees collected in Panama (N = 5) and five cultivated trees, three from Panama, and one each from Costa Rica and the Yucatan Peninsula of Mexico (Table 1).

Haplotype networks of C. cainito and congeners based on plastid sequence data

The combined plastid DNA (rps16 intron and trnS-fM intergenic spacer) sequence alignment was 1893 bp in length and contained seven informative nucleotide substitutions; we added 10 coded indels to the matrix, seven of which were parsimony informative.

We recovered 13 unique haplotypes (Table 2). Four indels and six nucleotide substitutions were homoplasious, which resulted in a network with one closed loop (Fig. 3). Unique haplotypes were recovered for all species and subspecies, with two exceptions: Chrysophyllum bicolor shared the same haplotype (cpA¹) as some samples of C. cainito, and the two individuals of C. mexicanum shared the same haplotype (cpA²) as some C. cainito samples from Panama (N = 2) and Jamaica (N = 2). Three haplotypes were recovered for Chrysophyllum argenteum subspecies: subsp. auratum (cpB¹), subsp. panamense (cpB²), and subsp. argenteum (cpB³) (Fig. 3, Table 2). Chrysophyllum oliviforme trees carried three haplotypes: C. oliviforme var. picardae carried the cpC¹ haplotype and C. oliviforme subsp. oliviforme carried two haploytypes, cpC² and cpC³. Chrysophyllum pauciflorum and Chrysophyllum sp. from the Osa Peninsula both carried unique haplotypes (cpD and cpE, respectively).

Haplotypes carried by C. argenteum and C. cainito were distributed throughout the network and did not form distinct, species-unique lineages. In contrast, haplotypes carried by C. oliviforme were connected to each other in the network, indicating that these haplotypes were more closely related to each other than they were to haplotypes carried by other species, though the two haplotypes carried by C. oliviforme subsp. oliviforme were shown to be independently derived from that of C. oliviforme var. picardae. The centermost haplotype in the network was carried by C. argenteum subsp. panamense and had four points of connection.

Chrysophyllum cainito individuals carried five haplotypes (cpA¹, cpA², cpA³, cpA⁴, and cpA⁵), three of which (cpA³, cpA⁴, and cpA⁵) were unique to the species (Table 2). The cpA¹ haplotype, which was the most common haplotype carried by C. cainito (N = 66), had two points of connection. It was carried by C. cainito trees from throughout the range and was shared by both wild and cultivated trees from all geographic areas sampled (Panama, Costa Rica, Guatemala, Mexico, Jamaica, and the Dominican Republic) (Fig. 3; see Table 1 for a complete list of C. cainito populations included). Haplotype cpA¹ was found in all wild and cultivated trees from northern Mesoamerica (N = 15) and the Antilles (N = 36), with the exception of two trees from one wild population in Jamaica, as well as one cultivated tree from Colombia and two from Papua, Indonesia. The southern Mesoamerican trees that carried the cpA¹ haplotype were predominantly cultivated trees (N = 10), with two wild trees from Panama also sharing the cpA¹ haplotype.

Of the haplotypes recovered in C. cainito trees, cpA⁴ had the most points of connection (three) and was located in the interior of the network. The cpA⁴ haplotype was recovered exclusively in wild trees collected in Panama (N = 4). This haplotype was connected to the cpA⁵ haplotype, which was recovered by wild trees (N = 2) from Panama, the cpA² haplotype, which was recovered in caimito trees collected in Panama (one wild and one cultivated tree) and Jamaica (two wild), as well as two individuals of C. mexicanum (Table 2), and to the lineage leading to the cpD, cpB³, and cpE haplotypes.

Geographic structure of C. cainito haplotypes and agreement between nuclear and plastid data

There is a strong pattern of geographic structuring of the ITS genotypes, with ITS-I being the only genotype recovered in the single tree that we sampled from South America and in the trees from the Antilles and northern Mesoamerica, with the exception of one tree from Muna, Yucatan, which carried the ITS-II genotype (Fig. 4A). The cultivated trees from Indonesia also carried the ITS-I genotype (Table 3). Twenty percent of the caimito trees from southern Mesoamerica carried the ITS-I genotype, 36% carried ITS-II, and 44% carried the ITS-I/II genotype (Table 3); the latter individuals appeared to carry both the ITS-I and ITS-II haplotypes.

Table 3. Frequency (percentage) of plastid haplotypes and ITS genotypes in Chrysophyllum cainitobased on cultivation status (wild vs. cultivated) and geographic region. Empty cells indicate 0%; dashes indicate no samples for a particular cultivation status/geographic region.

	Plastid haplotypes					ITS genotypes
Samples	cpA1	cpA2	cpA3	cpA4	cpA5	ITS-I	ITS-II	ITS-I/II
Cultivated
Antilles	100					100
Northern Mesoamerica	100					93.3	6.7
Southern Mesoamerica	83.4	8.3	8.3			41.7	33.3	25.0
South America	100					100
Old World	100					100
(All cultivated)	96.0	2.0	2.0			84.3	9.8	5.9
Wild
Antilles	87.5	12.5				100
Northern Mesoamerica	100					100
Southern Mesoamerica	15.4	7.7	30.8	30.8	15.4		38.5	61.5
South America	—	—	—	—	—	—	—	—
Old World	—	—	—	—	—	—	—	—
(All wild)	56.7	10.0	13.3	13.3	6.7	56.6	16.7	26.7
All samples (cultivated and wild)
Antilles	94.6	5.4				100
Northern Mesoamerica	100					93.8	6.2
Southern Mesoamerica	48.0	8.0	20.0	16.0	8.0	20.0	36.0	44.0
South America	100
Old World	100
(All samples)	81.5	4.9	6.2	4.9	2.5	74.1	12.3	13.6

The plastid haplotypes show similar geographic structuring (Fig. 4B). Almost all individuals in the Antilles and northern Mesoamerica carried haplotype cpA¹ (94.6%, 100%, respectively) (Table 3). All of the trees from South America and Indonesia also shared the cpA¹ haplotype. The frequency of haplotypes carried by southern Mesoamerican trees was variable. Forty-eight percent of the trees carried the cpA¹ haplotype, 8% carried cpA², 20% carried cpA³, 16% cpA⁴, and 8% the cpA⁵ haplotype (Table 3).

We observed a strong level of concordance between the nuclear ribosomal ITS genotypes and the plastid haplotypes within caimito. The vast majority of the individuals from the Antilles and northern Mesoamerica, which carried the cpA¹ haplotype, also carried the ITS-I genotype (100% and 93.3%, respectively) (Table 4). One exception was found in the cultivated tree from Muna, Yucatan mentioned above, which carried the cpA¹ and ITS-II combination. There was a different pattern in the south, where the plastid haplotypes (cpA¹, cpA², cpA³, cpA⁴, cpA⁵) and ITS genotypes were found in various combinations (Table 4). The one Antillean population that carried the cpA² plastid haplotype (Slipe, Jamaica, represented by individuals JP115 and JP116) carried the ITS-I genotype.

Table 4. Percentage concordance between plastid haplotypes and nuclear ribosomal ITS genotypes within cultivated and wild individuals of Chrysophyllum cainito. Empty cells indicate 0% and dashes indicate no data for a given cultivation status/geographic region.

Samples	cpA1- ITSI	cpA1- ITSII	cpA1- ITSI/II	cpA2- ITSI	cpA2 - ITSI/II	cpA3- ITSI	cpA3 - ITSII	cpA3- ITSI/II	cpA4 - ITSII	cpA4- ITSI/II	cpA5- ITSI/II
Cultivated
Antilles	100
Northern Mesoamerica	93.3	6.7
Southern Mesoamerica	33.3	33.3	16.7		8.3	8.3
South America	100
Old World	100
(All cultivated)	82.4	9.8	3.9	0	2.0	2.0	0	0	0	0	0
Wild
Antilles	87.5			12.5
Northern Mesoamerica	100.0
Southern Mesoamerica			23.1				15.4	15.4	23.1	7.6	15.4
South America	—	—	—	—	—	—	—	—	—	—	—
Old World	—	—	—	—	—	—	—	—	—	—	—
(All wild)	50.0	0	10.0	6.7	0	0	6.7	6.7	10.0	3.2	6.7
All samples (cultivated and wild)
Antilles	94.6			5.4
Northern Mesoamerica	93.8	6.2
Southern Mesoamerica	16.0	16.0	20.0		4.0	4.0	8.0	8.0	12.0	4.0	8.0
South America	100.0	—	—	—	—	—	—	—	—	—	—
Old World	100.0	—	—	—	—	—	—	—	—	—	—
(All samples)	70.3	6.2	6.2	2.5	1.2	1.2	2.5	2.5	3.7	1.2	2.5

Phylogenetic relationships in Chrysophyllum and close relatives of C. cainito

Our results provide insights into the taxonomy of Chrysophyllum and the origins of C. cainito, an important Mesoamerican fruit tree. Several inferences can be drawn from our analysis of phylogenetic relationships in Chrysophyllum based on ITS sequence data (Fig. 1). (1) The genus Chrysophyllum is paraphyletic with respect to Pouteria, as found in previous studies (56). (2) The monophyly of sect. Chrysophyllum is supported in both the MP and the Bayesian analyses and, with the increased taxonomic sampling compared to previous studies (e.g., 56), is strongly supported as a clade. (3) The probable sister group to sect. Chrysophyllum is a clade of Old World origin (Africa–Australia), a result consistent with the finding that Chrysophyllum originated in the Old World and was subsequently dispersed to South America (5).

Our results indicate that there are two main lineages within sect. Chrysophyllum and that three subgroups can be identified among accessions of C. cainito and C. argenteum. Chrysophyllum cainito, the type species of the genus, forms a clade (clade A1 in Fig. 1), but is nested within C. argenteum, suggesting that the former species is derived from the latter. Our results indicate that C. argenteum represents several lineages, which is reflected in observed morphological variation and the recognition of five subspecies.

Several morphological characteristics support the distinction between C. argenteum and C. cainito. In C. cainito, the length of the corolla lobes are equal to or exceed the length of the corolla tube, while in C. argenteum they are shorter than the tube (38). The combination of vegetative characters and number of seeds per fruit also differ between the two species. In C. cainito, the abaxial leaf surface is golden or golden-reddish brown (ferruginous) sericeous, and the fruit contains 3–10 seeds (38; 36). In C. argenteum, several combinations exist (38). Fruits of subsp. panamense have several seeds, but the leaves are subglabrous with whitish hairs, whereas in subspecies nitidum, whose fruits are unknown, the leaves are glabrous. In the other three subspecies, the fruits are single-seeded, and the abaxial leaf surface may be sericeous to sparsely appressed-puberulous or subglabrous with whitish hairs (subsp. argenteum), sericeous with golden hairs (subsp. auratum), or densely sericeous-tomentose with golden or ferruginous hairs (subps. ferrugineum). These distinctions hold for both cultivated and wild caimito trees compared to C. argenteum trees. Thus, we believe that the distinction between C. cainito and C. argenteum is warranted, but further investigation of the taxonomy of these species is needed; it appears that recognition of multiple species within C. argenteum may be most appropriate.

We were not able to identify all of our C. argenteum samples to the subspecific level. In some cases, there were no reproductive characters present at the time of collection, and not all of the sequences that we obtained from GenBank included subspecific identifications. Sample numbers JP724 and JP725, which fall into clade A3 in these analyses (Fig. 1), were collected in the Dominican Republic; they most likely represent C. argenteum subsp. argenteum and were labeled as such in this study. These trees had similar vegetative characters to those reported for subsp. argenteum, although the fruits from our collections were round, while subsp. argenteum fruits are typically ellipsoid (38). These trees therefore appear to represent additional morphological variation to what has been published previously for this taxon.

Chrysophyllum mexicanum is not supported as being the progenitor to C. cainito, but rather is closely allied with, and may even be derived from, C. oliviforme, a result consistent with 38 treatment. This result is notable because C. mexicanum is commonly considered by both local people and field biologists to be the wild form of C. cainito due to the similarity between the two species based on vegetative characters. Chrysophyllum mexicanum also has edible fruits, which are locally consumed (J. Petersen, personal observation).

Chrysophyllum bicolor appears to represent a unique lineage rather than part of the variation found within C. cainito as indicated by 38 listing of the former species as a synonym of the latter. Chrysophyllum bicolor is considered, along with C. pauciflorum, to be endemic to the Greater Puerto Rico Basin (1). The few individuals of C. bicolor that are known occur primarily in moist secondary forest on the islands of Puerto Rico and Hispaniola (1). JP57, a 10-m tall heritage tree, was collected on the Island of St. John, U. S. Virgin Islands and identified with the help of Eleanor Gibley, a local expert on the flora of St. John. It is interesting to note that in the original description by Poiret published in 1811, Chrysophyllum bicolor is described as an “arbrisseau,” meaning a woody plant that branches from the base, i.e., a shrub (25). Pedro 1 in the Flora of St. John, U. S. Virgin Islands states that C. bicolor is a tree, 6–15 m in height. In other aspects, the original description by Poiret and that found in the Flora of St. John appear to correspond with our collection (JP57), fitting the species description of 1. The fruit of C. bicolor is a 1-cm diameter, asymmetrically ellipsoid berry with a single seed (1), quite different from the 4- to 6-cm diameter, depressed-globose, multi-seeded berry of C. cainito (38). Acevedo considers these differences in fruit type between C. cainito and C. bicolor to be sufficient to warrant recognition of the two as separate species (Flora of St. John, p. 401). These observations, combined with the unique ITS sequence reported here, support recognition of C. bicolor as a distinct species, a decision that may have consequences for conservation of this taxon with an extremely restricted present-day distribution.

Patterns of divergence and distribution of caimito and close relatives using phylogenetic and phylogeographic approaches

Evolutionary relationships at the interspecific level are assumed to be hierarchical and are generally understood using tree-based phylogenetic approaches. Long divergence times are expected to lead to distinct gene pools between species due to the long-time accumulation of mutations and the subsequent fixation of unique alleles in the different lineages. These relationships can be depicted in bifurcating tree topologies (42). Using tree-based approaches to infer evolutionary relationships is considered appropriate when the following conditions are met: ancestral and descendent alleles do not occur simultaneously, there is no gene flow between the species, and there is no genomic recombination (42). However, these assumptions are sometimes violated, and evolutionary relationships can be obscured, particularly in studies involving plants, because of processes such as interspecific hybridization and the presence of ancestral and descendent alleles within a data set (49).

In contrast, relationships at the intraspecific level are considered nonhierarchical due to population-level forces such as gene flow and its associated possibility for recombination, as well as fewer and more recent mutation events. Multifurcating networks constructed using sequence variation (haplotypes) are therefore often better suited for inferring relationships below the species level (42). Phylogeographic approaches that examine haplotype diversity over a geographic landscape (4) have become increasingly useful for understanding past events that have shaped a species's current distribution pattern, particularly when the close relatives of the species under study are included (40).

We constructed haplotype networks to understand the patterns of divergence and distribution of caimito and its close relatives. We constructed separate haplotype networks of nuclear ribosomal ITS sequence data, which is biparentally inherited, and plastid sequence data, which is most commonly maternally inherited in angiosperms (19). The haplotype network based on ITS sequence (Fig. 2) has a highly bifurcating topology and reveals relationships very similar to those recovered in the phylogenetic tree (Fig. 1), with all species having at least one species-unique haplotype.

Although the species sampled appear to be well diverged based on the ITS data, there is minimal geographic structuring of C. cainito and congeners as inferred from either the haplotype network or the phylogenetic relationships of clades A and B (Fig. 1). Clade A contains species that are distributed in the Antilles, northern and southern Mesoamerica, and South America, while clade B contains species from the Antilles and northern and southern Mesoamerica. We undertook biogeographic analyses in DIVA (45) using a fully resolved topology from the phylogenetic reconstruction (Fig. 1), but we were unable to resolve the ancestral areas of the clades (data not shown). This is in part due to our sampling; to derive a robust hypothesis of the biogeographic history of sect. Chrysophyllum, we would need to sample all the species in the section including the nine species from South America, the putative ancestral area of the clade, and we did not have access to material of those species.

Nonetheless, using a network approach, our current sampling does allow insights into the historical biogeography within sect. Chrysophyllum. In the network constructed using ITS data, we see a lack of geographic structure among haplotypes of C. cainito and C. argenteum, reflecting the close relationship between these lineages and suggesting that coalescence of the haplotypes likely preceded the diversification of these species. We do, however, observe that the Antillean taxa, C. oliviforme subsp. oliviforme, C. oliviforme var. picardae, C. pauciflorum, and C. bicolor, share a close genealogical relationship with the northern Mesoamerican species, C. mexicanum. The haplotype of C. mexicanum appears to be derived from C. oliviforme (Fig. 2), which would suggest an Antilles to northern Mesoamerica dispersal event.

Similar to the relationships based on the ITS network, there is little geographic structuring in the haplotype network constructed using plastid data. Based on coalescent theory, we expect that ancestral haplotypes will have a high number of descending lineages as evidenced by the number of points of connection to other haplotypes (42). In our network based on plastid data, the cpB² haplotype has four points of connection and appears to be the oldest allele, ancestral to the others (Fig. 3). The cpB² haplotype is carried by C. argenteum subsp. panamense, which has a southern Mesoamerican and northern South American distribution, and is connected to haplotypes that are carried by four taxa (Fig. 3). The cpB² haplotype is connected to cpA¹, which is carried by cultivated and wild trees of C. cainito collected from throughout the present-day range and by the narrow endemic species, C. bicolor. In addition, cpB² is connected to cpC¹, carried by C. oliviforme var. picardae, cpA², carried by individuals of C. cainito and C. mexicanum, and cpA³, carried by individuals of C. cainito. The species that carry these haplotypes are distributed over a large geographic area that includes the Antilles as well as northern and southern Mesoamerica.

The plastid haplotypes are mostly species-unique, but in the case of C. cainito and C. argenteum, each haplotype is not most closely related to other haplotypes carried by the same species. This pattern suggests that the most recent coalescent event may have preceded the divergence of these two species, the same conclusion we reached based on ITS data. Incomplete lineage sorting and/or hybridization have likely played important roles in establishing this pattern, although the relative roles of these various processes are notoriously difficult to ascertain.

We also observe that the only two caimito trees from northern Mesoamerica and the Antilles that carry the cpA² haplotype were two putative wild caimito trees from Slipe, Jamaica. Neither fruits nor flowers were present on the trees at the time of collection. However, these trees appear to exhibit a fruit morphology that is intermediate between C. cainito and C. oliviforme, and local participants refer to these trees as “pencil star apples”. The fruits are said to be longer and more slender than typical C. cainito fruits (“star apple” is the common name of the species in Jamaica). These trees carry the ITS-I genotype, which was recovered in C. cainito trees and which predominates in northern Mesoamerican and Antillean caimito populations, and the cpA² haplotype, which was recovered in southern Mesoamerican caimito populations and in C. mexicanum. The combination of the C. cainito “northern” ITS genotype with the “southern” cpA² haplotype found in these trees may be due to incomplete lineage sorting within the plastid genome or local introgression from sympatric populations of C. oliviforme. Chrysophyllum cainito and C. oliviforme have not been reported to hybridize, although some other possibly intermediate forms have been observed on Jamaica (J. Petersen and D. Potter, personal observation). 31 reported a similar situation where the limits of species boundaries were obscured between a neotropical domesticate, Spondias purpurea L., and its wild relatives. Further detailed studies are required to parse out the relative extent to which these observed patterns in C. cainito haplotype diversity are due to incomplete lineage sorting, hybridization, and/or lack of resolving power of the DNA markers used in the study.

5 reported that the lineages leading to the clades containing C. cainito and C. oliviforme likely diverged from a common ancestor in the early Oligocene, between 29–19 Ma. This result would suggest a particularly long coalescence time for a group of plants, and it may be partly due to limited sampling of species in sect. Chrysophyllum in the 5 study. However, the occurrence of ancient polymorphisms being maintained in plastid regions has also been observed at the intraspecific level in cases such as Coreopsis L. (Asteraceae) (30), at both the intra- and interspecific levels in Senecio L. (Asteraceae) (11), and at the interspecific level in Hordeum L. (Poaceae) (23).

Our data suggest that there was a large ancestral effective population size of the species representing clades A and B (Fig. 1) as indicated by the fact that some plastid haplotypes are shared by more than one species (C. cainito and C. bicolor; C. cainito and C. mexicanum) and that the plastid haplotypes carried by a given species are not all more closely related to one another than they are to haplotypes carried by other species, with the exception of the haplotypes recovered in C. oliviforme. This observed pattern suggests that haplotype divergence evolved before species divergence and may also be due in part to interspecific hybridization. The tree and network based on nuclear ribosomal ITS haplotypes (Figs. 1, 2) reveal greater divergence among recognized taxa in sect. Chrysophyllum and each haplotype is species-unique; nonetheless, the ITS haplotypes of some species (C. oliviforme, C. argenteum) do not form exclusive lineages. The nuclear ribosomal DNA (ITS) is likely tracking more contemporary diversification patterns due to the process of concerted evolution of the ribosomal DNA (21; 61). We hypothesize that this process has partially erased patterns indicating gene flow and incomplete lineage sorting that occurred earlier in the evolution of these taxa, but these patterns are evident in the network based on plastid DNA sequence data, which tracks more ancient events.

Based on these observations, we hypothesize that the putative migration out of South America of the common ancestor of clades A and B (Fig. 1) happened early in the neotropical diversification of the group. The putative large ancestral effective population size may also explain the relatively high levels of lineage diversification (speciation) in both the Antilles and Mesoamerica. The distribution of several species occurring either on the Antillean chain or in Mesoamerica, but not both, suggests that there were possibly two migration routes out of South America; one via the Antillean Chain and the second up through Mesoamerica. In this scenario, repeated uni- or bidirectional dispersal between the Antilles and continental America would explain the lack of exclusivity of the relatedness among plastid haplotypes that we observed for many of the species. These emerging hypotheses regarding the history of lineage diversification and expansion of neotropical Chrysophyllum will benefit from further taxonomic and genomic sampling.

Geographic origins of wild caimito populations

In C. cainito populations, we see strong geographic structuring along a north–south gradient based on the nuclear genotype and plastid haplotype diversity (Fig. 4), similar to genetic structuring of Mesoamerican populations reported for other neotropical species such as Cedrela odorata L. (7 and references therein) and mahogany (Swietenia macrophylla King) (35, but see also 27). Our data support a southern Mesoamerican origin for wild caimito populations. We did not recover any nuclear or plastid haplotypes unique to northern Mesoamerica or the Antilles. Rather, all were a subset of haplotypes present in populations from the south. In northern Mesoamerica and the Antilles, we observe a partial reduction in overall diversity compared to southern Mesoamerica; 40% of the plastid diversity and 100% of the ITS diversity were recovered there. In addition, we see a strong skew toward particular haplotypes, as 94.3% of all northern Mesoamerican and Antillean populations carry the cpA¹ and ITS-I combination. Of the remaining trees from that region, 1.9% carry the cpA¹ ITS-II combination and 3.8% carry the cpA² ITS-I combination.

As all of the genetic diversity of C. cainito is present on the Isthmus of Panama, we hypothesize this to be the origin or center of diversity of the species. We cannot, however, definitively rule out a northern South American origin of the species, where wild trees possibly occur (C. Baraloto, UMR EcoFoG, personal communication), but we did not have access to any samples of such trees. We hypothesize a migration of the ITS-I genotype and the cpA¹ haplotype from more southern areas, including the Isthmus of Panama, into northern Mesoamerica and the Greater Antilles. This migration may have been due to natural causes such as adaptation to environmental factors associated with historic climate change, with a concomitant reduction in frequency of the cpA¹ haplotype in the more mesic environment within Panama. A similar situation has been hypothesized for the central and southern plastid haplotype lineages of Cedrela odorota, a common timber tree in Central America (7).

Coupled with the strong geographic structuring of plastid haplotypes, we observe a shallow phylogeographic signal as indicated by the relatively few nucleotide substitutions that differentiate the haplotypes (Fig. 3). This shallow signal may in part reflect slow rates of molecular evolution, particularly with regard to the plastid genome, which have also been observed in studies done at the subfamilial level where seven plastid regions were necessary to resolve higher-level relationships within the Chrysophylloideae (56). The shallow phylogeographic signal may also reflect a fairly recent northward geographic expansion of the species into a newer range. Recent expansions into new geographic ranges have been observed in Gliricidia sepium (Jacquin) Steudel, where founder and source populations differed by only a few molecular changes (26).

Genetic differentiation of wild and cultivated trees and the origins of cultivated material

The cultivated caimito trees that we sampled retain high levels of the total diversity sampled in wild caimito trees. In addition to finding the highest overall diversity in southern Mesoamerica, we observe that the cultivated trees that we sampled from that region are also highly diverse and retain 60% of the plastid diversity present in local wild populations and 100% of the nuclear ribosomal diversity sampled.

The wild and cultivated trees sampled from southern Mesoamerica have higher levels of diversity than the wild and cultivated trees in northern Mesoamerica and the Antilles. The frequencies of the cpA², cpA³, cpA⁴, and cpA⁵ haplotypes are much higher in southern Mesoamerica where the cpA¹ haplotype is found in roughly the same frequency (48%) as the remaining plastid haplotypes (cpA², cpA³, cpA⁴, and cpA⁵) combined (52%), as compared to all of the trees (100%) carrying the cpA¹ haplotype in northern Mesoamerica and the vast majority (94.6%) carrying the cpA¹ haplotype in the Antilles. The majority of southern wild trees carry the cpA², cpA³, cpA⁴, or cpA⁵ haplotype, with cpA³ and cpA⁴ being the most common. In southern Mesoamerica, the cpA¹ haplotype greatly increases in frequency in cultivated populations (83.4% vs. 15.4% in wild populations sampled). In addition, some southern Mesoamerican cultivated populations, such as Arraijan Barriada 2000 and Chilibre, are polymorphic and trees there carry either the cpA¹ or the southern Mesoamerica-specific haplotype cpA³. These patterns suggest that cultivated material in Panama may have been derived from multiple local wild populations.

A similar pattern is observed in the nuclear ribosomal genotypes. In the trees that we sampled from southern Mesoamerica, the ITS-I/II genotype was most common (44%). This genotype predominates in the southern Mesoamerican wild trees (61.5%) but is only carried by 25% of the southern Mesoamerican cultivated trees. Cultivated trees in southern Mesoamerica predominantly carry the ITS-I genotype (41.7%). We also see an increase in the association of particular plastid haplotypes and nuclear genotypes under cultivation, with one third of the cultivated trees in southern Mesoamerica carrying the cpA¹ ITS-I combination and one-third carrying the cpA¹ ITS-II combination, both of which are absent in the local wild populations that we sampled. The patterns observed in the haplotype frequencies and distributions suggest that the commonly cultivated form of caimito may have been brought into cultivation in Panama from wild populations such as Clayton, where the trees carry the cpA¹ and ITS-I/II combination. Subsequently, due to population genetic forces such as local drift or selection (natural or anthropogenic), the cpA¹ ITS-I and cpA¹ ITS-II combinations both increased in frequency under cultivation.

Within Panama, the cultivated trees retain high levels of diversity. For example, trees collected in the community of Chilibre retain high levels of local haplotype diversity with trees there carrying the cpA¹ and cpA³ haplotypes as well as ITS-I and ITS-II (including the less common combinations, cpA¹ ITS-II and cpA³ ITS-I). These may represent selections made from local surrounding forests such as Madden, Venta de Cruces, and Pipeline, where trees carry the cpA³ and ITS-I/II combination, and Clayton, where trees carry the cpA¹ ITS-I/II combination.

It is interesting to note that we did not sample any of the ITS-I genotype (ITS-I in homozygous form) in wild populations from southern Mesoamerica, yet 41.7% of the cultivated trees in the area carry this genotype. The ITS-I haplotype is found in heterozygous form (ITS-I/ITS-II genotype) in wild trees in this area. There are three possible, and not mutually exclusive, hypotheses that may explain these observations: (1) The ITS-I genotype is in fact present in some wild trees in southern Mesoamerica, but we did not sample them in our study. (2) Many wild trees in southern Mesoamerica retain ancestral polymorphisms in ITS. (3) The ITS-I genotype, which has increased in frequency under cultivation, is at least partly maintained in heterozygous form in southern Mesoamerican populations due to historic and/or ongoing gene flow between cultivated and wild trees.

A distinct cultivated gene pool, represented by the cpA¹ ITS-I combination, is being maintained under cultivation. We hypothesize that the cpA¹ ITS-I combination was selected from several local wild populations in Panama. Cultivated material outside of Panama, which almost exclusively bears that combination, appears to be derived from a subset of the locally cultivated material found within Panama. This combination was found in the material we sampled from South America and Indonesia, as well as the Antilles and northern Mesoamerica. The collection from Muna, Yucatan, which was the only cultivated tree from outside of Panama that did not carry the cpA¹ ITS-I combination, may represent diversity that is retained in very low frequency outside the putative area of domestication.

In addition, some of the observed ITS-I in southern Mesoamerican cultivated trees may be the result of a secondary recolonization of caimito from the north to the south by human-mediated dispersal. Several of these ITS-I trees were collected from communities such as Chilibre that are heavily dominated by people of Afro-Antillean extraction whose ancestors may have brought caimito selections with them when they arrived in Panama to build the railroad and canal.

Based on our data, we hypothesize that after trees were brought into cultivation in Panama, a genetic bottleneck occurred due to either anthropogenic or natural causes (such as those discussed above), and the material that migrated to northern Mesoamerica and the Antilles was primarily the cpA¹ ITS-I combination. It is possible that there are some traits associated with cultivation or domestication that are carried by individuals that have the cpA¹ ITS-I combination.

We observe strong phenotypic, chemical, and ecological differences between wild and cultivated caimito populations sampled in Panama, which we documented in an earlier study (36). However, these differences do not have a corresponding genetic component based on the nuclear and plastid markers used in these analyses. It appears that the markers that we used, which are designed to track neutral variation, are not tracking genomic regions associated with the observed differences between wild and cultivated fruit and seed traits.

Previous studies that used molecular markers to evaluate domestication processes of neotropical fruit trees found that cultivated populations retain significant levels of the genetic diversity present in wild populations (i.e., 22) but that there is not always a clear demarcation between cultivated and wild populations when using neutral genetic markers, particularly in areas where wild and cultivated trees are sympatric, as in the case of peach palm (Bactris gasipaes Kunth) (13). Similar to the pattern that we observe in caimito, other instances have been reported where cultivated or domesticated neotropical fruit trees appear to be derived from multiple wild populations, such as the cases of jocote (Spondias purpurea L.) (32, 33) and Inga edulis Mart. (14).

Conclusions and future directions

In our phylogenetic analysis of ITS data, we see a well-supported set of relationships, which support the monophyly of Chrysophyllum sect. Chrysophyllum as well as the evolutionary distinctiveness of most of the species that we sampled within the section. The notable exceptions are that C. cainito is nested within C. argenteum and C. argenteum, as currently circumscribed, represents multiple lineages. We observe similar relationships between C. cainito and C. argenteum based on the network constructed using plastid sequence data. Both results from the phylogenetic and phylogeographic analyses suggest that factors such as incomplete lineage sorting and/or hybridization may be at play. The lack of concordance in the inferred relationships based on ITS and plastid data of C. mexicanum and C. bicolor also appears to best be explained by incomplete lineage sorting. Based on our current sampling, it is unclear whether the taxonomic delineation of C. oliviforme subspecies adopted by 38 reflects evolutionary history, and additional studies will help elucidate whether the variation within C. oliviforme warrants recognition of intraspecific taxa.

Clades A and B based on ITS data (Fig. 1), which represent clades containing C. cainito and C. oliviforme, respectively, are estimated to have diverged 29–19 Ma. It is possible that repeated migration and dispersal events within the neotropical range may in part be responsible for the maintenance of ancestral alleles along with descendent alleles in the different species. The ancestral and descendent alleles also occur at the intraspecific level in C. cainito populations in southern Mesoamerica. These observed patterns may also be caused by hybridization, as evidenced by both the sharing of alleles between sympatric congeners and also by the nonconcordance between nuclear ribosomal ITS and plastid alleles. The importance of using multiple individuals per species to infer evolutionary relationships in phylogeny reconstruction is elucidated, as is the benefit of using both network and tree-based approaches, both of which have been useful approaches in previous studies.

Understanding the biogeographic history of sect. Chrysophyllum will not only help in understanding how the aforementioned factors shape present-day diversity and distribution patterns; it is also essential for developing a robust hypothesis of the origin and history of evolution and domestication of C. cainito. To further unravel the evolutionary history of this group, we recommend including multiple individuals for each species and subspecies of sect. Chrysophyllum collected from throughout the neotropical range, particularly from South America and the to-date unsampled areas in Mesoamerica and the Antilles.

Within C. cainito, we see a strong level of geographic structuring on a north–south gradient, a result that is consistent between both plastid and nuclear ribosomal ITS data. In both data sets, there are higher levels of diversity in southern Mesoamerica in comparison to the Greater Antilles, the presumed origin of the species. Our data suggest a possible southern Mesoamerican origin of the species and also point to southern Mesoamerica as the likely center of domestication of caimito. Using these neutral plastid and nuclear ribosomal ITS markers, we do not detect any specific haplotypes associated with a tree's cultivation status, but rather we see an increase in frequency of the cpA¹ ITS-I combination. While our results do not preclude a historic distribution range that may include both the Antilles and southern Mesoamerica, it appears that an expansion into the northern distribution range of caimito may have occurred via human-mediated migration through cultivation. We plan to use highly variable microsatellite markers developed for C. cainito to test our “Out of Panama” hypothesis. Additional evolutionary and biogeographic studies within sect. Chrysophyllum will also provide context in our efforts to shed light on the present and historical processes that have thus far shaped the distribution patterns of an important neotropical tree species.