Introduction

The seasonally dry tropical (SDT) forest of Central America ranks among the most threatened forest ecosystems in the tropical lowlands and is even more endangered than the tropical rainforests of Amazonia (Janzen, 1988; Pennington et al., 2006). Formerly extending over very large areas along the Pacific side of Central America, these SDT forest habitats have been so seriously impacted by human activities that nearly all of them have been lost. Historically, the first clearings of these forests probably date back to at least the Archaic Agricultural Period (ca. 6000–5550 BP) and destruction intensified after the 16th century Spanish Conquest (Janzen, 1988; Piperno, 2011). Nonetheless, up to the early-to-mid 20th century, there were still ca. 550 000 km2 of SDT forest left, extending in an almost continuous strip along the Pacific side of Central America, from south-western Mexico to north-western Costa Rica, and disjunctively to Panama’s Azuero Peninsula (Janzen, 1988; Olson et al., 2001; Griscom and Ashton, 2011). It is now estimated that only 2.0–1.7% of the original SDT forest in Central America escaped deforestation, degradation and conversion into agricultural land and/or pastures, especially between the 1930s and 1970s (Griscom and Ashton, 2011), and only <0.1% of this remaining forest is protected (Miles et al., 2006). This massive habitat loss and disturbance of Central American SDT forests is well recognised today as a major threat to the region’s high biodiversity and endemism (Janzen, 1988; Sánchez-Azofeifa et al., 2005; Linares-Palomino et al., 2011). However, relatively little is known about the impact of this forest degradation on regional-scale genetic structure of constituent tree species.

Especially for outcrossing tree species it has long been assumed that population isolation and decreases in population size associated with anthropogenic forest fragmentation can lead to genetic erosion and increased population differentiation through increased genetic drift, inbreeding and/or reduced gene flow. In turn, such genetic erosion may negatively affect the long-term viability and adaptive potential of remnant populations (Aguilar et al., 2008; Frankham et al., 2010). However, there is increasing evidence that a large number of forest tree species (from both temperate and tropical regions) are able to ameliorate the population genetic consequences of recent habitat fragmentation (reviewed in Lowe et al., 2005, 2015). Potential mechanisms underlying such ‘resilience’ include a flexible mating system (Ward et al., 2005), extensive gene flow via pollen and/or seed (Bacles et al., 2006) and/or longevity coupled with overlapping generations (Petit and Hampe, 2006; Davies et al., 2010; but for exceptions see, for example, Kettle et al., 2007; Dubreuil et al., 2010).

In Central America, genetic surveys of SDT forest trees are relatively scarce, highly heterogeneous (for example, in terms of geographical scale of sampling, time or severity of impact, species’ life history/dispersal characteristics) and mostly limited to heavily exploited timber trees (Lowe et al., 2005; Ratnam et al., 2014). Nonetheless, the few broad-scale studies available typically identified low-to-moderate spatial genetic structure (for example, Chase et al., 1995; Gillies et al., 1999; Novick et al., 2003; Poelchau and Hamrick, 2012; but see Cavers et al., 2003, 2013), whereas more fine-scale surveys usually found little or no significant impact of recent habitat loss on genetic diversity parameters (for example, Dawson et al., 1997; White et al., 2002; Fuchs et al., 2003; Lowe et al., 2005; but see Gillies et al., 1999). Overall, these assessments emphasise the need for more individual case studies, as the genetic impacts of fragmentation likely vary by species, geographical scale, current environmental conditions and/or historical context (Lowe et al., 2005; Poelchau and Hamrick, 2013). Moreover, there is a gap left for phylogeographic studies of SDT forest trees, especially in those countries that continued to experience particular high rates of deforestation up to the last few decades, namely Honduras, Guatemala and Nicaragua (Redo et al., 2012; Hansen et al., 2013). Such regional surveys are also important to guideline how remaining SDT forest populations may be effectively conserved in situ and optimally sampled for ex situ conservation, breeding and restoration programmes (Ceska et al., 1997; Frankham et al., 2010; Diniz-Filho et al., 2012).

Here, we applied dominant amplified fragment length polymorphism (AFLP) markers to assess genetic variation in Nicaraguan populations of Calycophyllum candidissimum (Vahl) DC. (Rubiaceae, Ixoroideae, Condamineeae), one of the most common constituents of Central America’s degraded SDT forest (Faber-Langendoen et al., 2012). This insect-pollinated and wind-dispersed canopy tree is naturally distributed from southern Mexico through Panama to northern South America (Colombia, Venezuela, Ecuador) and also occurs in the Greater Antilles (Grandtner and Chevrette, 2013; Figure 1a; see Methods section for more details). In Nicaragua, where <1% of the area is covered with SDT forest (Stevens et al., 2001), C. candidissimum is valued for its showy inflorescences (for example, sold during religious festivities) and hard, fine-grained wood (used for, for example, house building, carpentry, cabinet work, tool handles, combs, firewood, charcoal; Standley and Williams, 1975; Cordero and Boshier, 2003). Since 1971, it also has received the distinction of being the country’s ‘National Tree’ (Salas, 1993). Apparently for those reasons, this ‘multi-purpose’ tree has often been left when forests were cleared for agricultural practices, timber, livestock grazing and cotton or other plantations (Barrance et al., 2009; Griscom and Ashton, 2011). Although this habitat degradation has occurred only recently (that is, ca.85 years ago; see above), its presumably negative genetic impact (for example, diversity loss, reduced gene flow) might already be discernible in species with rather short generation times, such as C. candidissimum, which only requires ~3–5 years to first flowering under open-grown conditions (R Dolmus/UNAN-Léon; JB Quezada/Universidad Nacional Agraria (UNA), Managua; personal communications). On the other hand, in this wind-dispersed tree, such negative effects might be mitigated through sustained levels of seed flow, but this is something to be tested against molecular data.

Figure 1
figure 1

(a) Simplified map of the geographical distribution of C. candidissimum in Central America, the Greater Antilles and northern South America based on geo-referenced specimen records (N=595) available at TROPICOS (http://www.tropicos.org/Name/27903157 (accessed 20 October 2016)) and the Global Biodiversity Facility (GBIF.org (6th September 2016) GBIF Occurrence Download http://doi.org/10.15468/dl.wc19nm). (b) Geographical locations of 13 populations of C. candidissimum in Nicaragua represented in this study for AFLP analysis (red circles), plus additional specimen records (N=90) according to TROPICOS/GBIF (small white circles). The light-brown areas represent the original distribution of seasonal dry forest derived from the World Wildlife Fund (WWF) ecoregion dataset (modified from Olson et al. (2001)). The dashed lines delimitate the physiographic regions of Nicaragua (Pacific, Central Highland, Atlantic) according to Taylor (1963). See Table 1 for identification of population codes.

Because of its massive habitat loss, C. candidissimum essentially lacks primary forest populations that could serve as ‘controls’ for pre-fragmentation conditions. However, a priori expectations about its genetic architecture might be derived from AFLP data available for a close relative from the western Amazonian rainforest, C. spruceanum (Benth.) Hook.f. ex K.Schum. (Russell et al., 1999). Based on nine populations from the Peruvian Amazon Basin, this study revealed relatively high genetic diversity within and moderate variation among populations, without any major influence of river structure (viz. hydrochory) or geographic distance among sites on genetic distance measures (for example, there was no evident pattern of ‘isolation by distance’ IBD; Wright, 1943). Overall, these data were indicative of high levels of gene flow among tree populations (mostly via insect-transferred pollen and/or wind-dispersed seed) in a non-fragmented rainforest setting (Russell et al., 1999). For C. candidissimum in Nicaragua, therefore, the null hypothesis would be that recent SDT forest fragmentation has resulted in much lower genetic diversity within populations (for example, owing to drift and/or inbreeding), and concomitantly higher genetic variation viz. ‘insularization’ among populations, in comparison with C. spruceanum.

We investigated geographical patterns of AFLP variation in 13 populations of C. candidissimum in Nicaragua, which cover much of the country’s original dry forest vegetation zone (Figure 1b), and also differ in their degree of disturbance and/or status of legal protection (Table 1). Our major aim was to gain first insights into the relative roles of contemporary vs historical factors in shaping the genetic variation of this emblematic tree species of Central America’s highly endangered SDT forest. Concomitantly, we anticipated that the results obtained might help assist future conservation measures that aim at preserving the genetic diversity of this species in Nicaragua. We addressed the following specific questions: (1) to what extent has human-driven landscape fragmentation affected levels of genetic diversity within populations and patterns of gene flow/dispersal (viz. the connectivity) between them? (2) Is there evidence that other factors such as past population dynamics and/or environmental dispersal barriers had a confounding effect on the genetic patterns observed? And (3) what is the conservation value of these populations (for example, are protected ones sufficiently representative of the species’ genetic diversity in Nicaragua, or is a broad-scale coverage of conservation measures advisable)?

Table 1 Information on sample locations for 13 populations of C. candidissimum from Nicaragua surveyed for AFLP variation

Materials and methods

The study species

C. candidissimum, also called ‘Lemonwood’ or ‘Madroño’, is a diploid, light-requiring pioneer tree (with 2n=2 × =34 chromosomes; Kiehn, 1995), which grows as tall as 15–20 (30) m (Cordero and Boshier, 2003; Gonzáles-Rivas et al., 2009; Lorence, 2012). Above the glossy leaves, the small, white and bisexual flowers are clustered in large inflorescences (thyrses) in which some flowers bear a large, white-to-greenish calyx lobe or ‘calycophyll’ (Delprete, 1996). The sweet-scented, protogynous flowers are primarily visited by bees and butterflies (Villegas-Durán et al., 2003; Ramírez-Arriaga et al., 2011; A Dávila-Lara, personal observation) and develop into small, septicidal capsules releasing numerous small (3–5 mm), short-winged seeds that are wind-dispersed (Lorence, 2012; Hilje et al., 2015). Although Bawa (1974) stated that C. candidissimum is self-compatible, no concrete experimental evidence was provided due to ‘the lack of adequate numbers’. We are unaware of any report indicating that this species regularly reproduces clonally (for example, via root-sprouting).

In Nicaragua, C. candidissimum mainly occurs along shady hill sides and along water streams within fragments of secondary and/or riparian SDT forest in the ‘Pacific Region’ (sensu Taylor, 1963) and, more rarely, in isolated stands of semi-deciduous (for example, Quercus-Pinus) forest in the ‘Central Highland Region’ (Harvey et al., 2006; Lorence, 2012; see Figure 1b). The former region has high temperatures and moderate-to-high rainfall, but with strong seasonal drought, whereas the latter has a sub-tropical climate (cf. Taylor, 1963). Throughout its distribution range, C. candidissimum is still considered to be locally abundant (Stevens et al., 2001; Lorence, 2012; Kappelle, 2016) and neither listed by the CITES (Convention on Endangered Species; see http://www.cites.org) or the International Union for Conservation of Nature (IUCN, 2016). However, in Nicaragua, it is classified as ‘endangered’ (threat index ‘4’) according to the Food and Agriculture Organisation of the United Nations (FAO; Roa, 2003).

Plant material and sampling

Between April and August 2013 an extensive survey of the distribution of C. candidissimum in Nicaragua was carried out by one of us (AD-L) within the UNAN-León/Department of Biology project ‘Conservation and sustainable use of timber and non-timber tree species’ in cooperation with the MARENA (Ministry of Environment and Natural Resources)/Nicaragua (MARENA, 2002). Based on this, 13 natural populations were chosen as sampling sites to cover the full geographic range of the species in Nicaragua (see Figure 1b and Table 1): two populations (SO, QU) from the Central Highland Region (407–905 m above sea level), and 11 from the lowland and premontane areas of the Pacific Region (1–432 m above sea level), including one island population (OM) of Lake Nicaragua. Estimated pairwise geographical distances among sampling sites ranged between 25 km (MO–TE) and 287 km (SO–OS) with a mean (±s.d.) of 139 (±63.8) km (data not shown). These 13 sites differed in their degree of disturbance level and/or status of legal protection (Table 1). Five populations (QU, MO, CH, OM, EO) were located in relatively undisturbed, more or less-continuous forest (A Dávila-Lara, personal observation), of which three (MO, CH, OM) were protected; by contrast, all the remaining populations were from at least partially logged or highly degraded areas, and only two of those (TE, SC) were protected. At each site, young leaves were collected from, if possible, up to 20 mature trees that were sampled according to the line transect method of González (2006), with a minimum distance of 100 m between trees to avoid selecting closely related individuals (Gillies et al., 1999; Inza et al., 2012). To ensure sufficient individual sampling, the sizes of the areas surveyed (in total ca. 2638 ha) had to be adjusted accordingly, ranging from 61 ha (for the relatively dense QU population) to 453 ha (for the sparse LV and ES populations) (Table 1). Logistic constraints and the sampling scheme used did not allow for a detailed investigation of the tree density of C. candidissimum at any individual site, but it provided the basis for an estimation of the number of sampled trees per hectare at all sites (Table 1). This, together with visual assessments during sampling, allowed for approximating the disturbance level of populations, as indicated above. In total, leaf materials of 250 individuals were collected, dried with silica gel, and stored at room temperature.

DNA extraction and AFLP genotyping

Total genomic DNA was extracted from the dried leaf tissue using a modified cetyltrimethyl ammonium bromide protocol (Doyle, 1991). The AFLP analysis was performed according to Vos et al. (1995), with minor modifications as described by Jaros et al. (2016) for the use of fluorescent-dye-labelled primers. For a preliminary AFLP screen, we selected one individual each from four populations (LE, CH, OM EO), using 18 selective primer pair combinations with two or three selective nucleotides (data not shown). Based on this, the following primer combinations that produced the best results with respect to polymorphism and clarity of AFLP profiles were chosen for the complete survey: EcoRI-AAC/MseI-CA (NED), EcoRI-AC/MseI-CTC (VIC) and EcoRI-AT/MseI-CTG (6-FAM). Each PCR plate (48 slots) contained two positive controls, one negative control and one blind sample. The selective PCR products were purified using Sephadex G-50 Superfine (GE Healthcare BioSciences, Uppsala, Sweden) following the manufacturer’s instructions. AFLP fragments were separated on a MegaBACE 1000 (GE Healthcare Biosciences, Pittsburgh, PA, USA), using the ET400-R-Rox-MegaBACE sizing standard, and manually scored in DAX version 8.0 (Van Mierlo Software Consultancy, Eindhoven, The Netherlands) as described in Bendiksby et al. (2011). AFLP products were scored as presence (1) or absence (0), whereby a fragment was considered polymorphic if at least one individual showed a variant pattern. Separate test runs with 16 individuals (representing all 13 populations) were used to obtain an AFLP error rate, calculated as the percentage of differently scored loci over all replicates and the three primer pair combinations (Bonin et al., 2004). In total, 220 individuals were finally available for data analysis because 30 individuals failed to amplify in at least one of the three primer combinations. Accordingly, the mean (±s.d.) sample size was 17.0 (±2.6) individuals per population (see Table 2).

Table 2 Summary of AFLP variation for 13 populations (220 individuals) of C. candidissimum from Nicaragua

Data analysis

Using the program GENALEX version 6.5 (Peakall and Smouse, 2012), three different measures of genetic diversity were calculated for each locus and then averaged across all loci to represent a single estimate for each population: the mean number of observed alleles per locus (NA), the percentage of polymorphic fragments (PPF), and Nei’s (1987) unbiased gene diversity (HE) for small sample sizes. In addition, we used the AFLPDAT R package (Ehrich, 2006) to estimate frequency down-weighted marker (DW) values (or ‘rarity-indices’), with 95% confidence intervals obtained by bootstrapping over all loci (9999 replicates), and significance tested by a permutation procedure (A Tribsch and D Ehrich, unpublished data) implemented in AFLPDAT (1000 replicates). Significantly high vs low values of DW may indicate long-term population isolation vs genetic drift and/or founder effects (Skrede et al., 2009). Nonparametric Mann–Whitney U-tests were performed to determine whether there was a significant difference in genetic diversity and rarity estimates between populations differing in disturbance level or protection status. Likewise, Spearman’s rank correlations were employed to test for a potential relationship between these genetic parameters and various extrinsic factors (latitude, longitude, precipitation, temperature; see Table 1). All these statistical tests were made using the ‘Social Science Statistics’ online calculators (http://www.socscistatistics.com/).

Genetic structure analysis was performed in STRUCTURE version 2.2 with a Bayesian clustering approach based on a recessive allele model, as applicable for dominant markers (Pritchard et al., 2000; Falush et al., 2007). We run this analysis after pruning markers that proved uninformative (that is, monomorphic) across the entire AFLP dataset (ca. 35%; see Results) and assumed both recent ancestry and some current gene flow among populations (that is, admixture and correlated allele frequencies). The number of clusters (K) was set to vary from 1 to 15. For each value of K, we performed 10 runs with a burn-in length of 50 000 and a run length of 100 000 Markov chain Monte Carlo replications. To identify the most probable value of K, post processing STRUCTURE runs were performed in STRUCTURE HARVESTER version 0.6.94 (Earl and vonHoldt, 2012) to produce graphical files representing the final posterior probability [lnP(D)] (Pritchard et al., 2000) and ΔK (Evanno et al., 2005) per K and repeated run. The CLUMPP program version 1.1.2 (Jakobsson and Rosenberg, 2007) was used to align the membership coefficients from the 10 replicates for each K based on the LargeK Greedy algorithm.

Analyses of molecular variance (AMOVAs) were performed in ARLEQUIN version 3.5 (Excoffier and Lischer, 2010) to quantify the partitioning of AFLP variance (1) among populations (ΦST) and within populations; and (2) between regional groups of populations (ΦCT) and among populations within such groups (ΦSC). Significance levels of variance components were evaluated by a permutation approach (1023 replicates). We also used GENALEX to calculate values of ΦPT (an analogue of ΦST) for all pairs of populations and to infer their significance with 9999 random permutations. Genetic relationships among populations were visualised by subjecting their pairwise ΦPT distances to a principal coordinates analysis (PCoA) in GENALEX. In addition, we constructed a Neighbour-Joining (NJ) phenogram in PHYLIP version 3.69 (Felsenstein, 2005) based on Nei’s (1987) unbiased genetic distances calculated using the fragment-frequency approach in AFLPSURV version 1.0 (Vekemans, 2002). The NJ phenogram was midpoint-rooted, with BS (bootstrap support) obtained by resampling with replacement over loci (9999 replicates). Isolation by distance (IBD) was assessed by regressing values of ΦPT/(1−ΦPT) against the logarithm (log10) of geographical distance (in km) for all pairs of populations (Rousset, 1997), and a Mantel test was employed to evaluate significance with 9999 permutations using IBDWS version 3.23 (Jensen et al., 2005).

Results

Population genetic diversity

For the 220 individuals (13 populations) of C. candidissimum, the three AFLP primer combinations resolved a total of 226 unambiguous fragments, ranging from 87 to 411 base pairs (bp), of which 147 markers were polymorphic (PPF=65%; Table 2). Different primer pairs generated a variable number of fragments (‘NED’: 88; ‘VIC’: 49; ‘6-FAM’: 89), with an average of 75.3 (±22.8 s.d.) fragments per primer combination. The overall error rate calculated on the basis of 16 replicates (1.16%) is at the lower end of the range reported for other AFLP studies in plant species (Bonin et al., 2004; Poelchau and Hamrick, 2012).

Estimates of genetic diversity (in terms of NA, PPF and HE) for each population (Table 2) were positively correlated to each other (all P<0.001), but not with the number of individuals (N) sampled per population (r=0.29–0.36, P=0.140–0.675). Values of these diversity parameters were similar across populations and generally low (Table 2). For example, within-population gene diversity (HE) averaged 0.125 (±0.015 s.d.), with the lowest value (0.111) observed in a Central Highland population (OU) and the highest (0.166) in the southernmost population from the Pacific Region (EO). Although populations in disturbed forest displayed on average slightly lower gene diversity (mean HE=0.121) than those in undisturbed forest (0.132), this difference was not significant (U-test: Z=−0.732, P=0.465), and the same was true for NA and PPF (P=0.418 and 0.509, respectively). Similarly, there were no significant differences between populations from unprotected vs protected sites in any of the three diversity parameters (all P0.660). Levels of allelic rarity (in terms of DW) varied among populations but were generally low, averaging 0.669 (±0.113 s.d.) across all populations (Table 2). Based on permutation tests, one Central Highland (SO) and two Pacific North/Central populations (TE/SL) had significantly low DW values, whereas the southernmost Pacific population (EO) had a significantly high value (Table 2). Notably, the proportion of rare markers was on average significantly lower in populations from disturbed (0.617) vs undisturbed forest (0.754; U-test, two-tailed significance: Z=−2.269, P=0.023). However, no such difference was found between unprotected vs protected populations (0.689 vs 0.638; P=0.826). Correlation analyses revealed a negative and significant relationship between latitude and both NA (r=−0.72, P=0.006) and PPF (r=−0.65, P=0.016), whereas the same trend was marginally significant for HE (r=−0.53, P=0.061). None of the other factors examined (that is, longitude, precipitation, temperature) showed a significant effect (P0.05) on genetic diversity or allelic rarity within populations (data not shown).

Population genetic structure and relationships

In the STRUCTURE analysis, log probabilities of the data [lnP(D)] showed some tendency to plateau at K≈4, whereas the ad hoc statistic ΔK clearly showed the highest likelihood at K=2 (see Supplementary Appendix Figure S1). Based on the latter model, all populations from the Central Highland (SO, QU) and Pacific North Region (LV, ES, TE, MO, LE) were fixed for cluster I (‘black’ in Figure 2), while cluster II (‘grey’) was predominant in the three Pacific South populations (CH, OM, EO). By contrast, populations from the Pacific Central Region were differently admixed with clusters I vs II, whereby SL and SC showed higher membership proportions to the northern cluster I and ST to the southern cluster II. At K=4, the Pacific Far South population (EO) became clearly distinguishable (cluster III), whereas few samples (mostly from QU and LV) were erratically assigned to a fourth ‘gene pool’, without reflecting a geographical pattern (see Supplementary Figure S2). We therefore retained K=2 as the likely more robust model, albeit without dismissing the genetic distinctiveness of EO (see also below).

Figure 2
figure 2

Distribution of the two major AFLP gene pool clusters (I and II) within and among 13 populations (220 individuals) of C. candidissimum from Nicaragua as identified by STRUCTURE based on the ad hoc statistic ΔK (see also Supplementary Figure S1). The bar-plot displays the assignment of individuals to the two clusters, with the smallest vertical bar representing one individual. The y axis presents the estimated membership coefficient (Q) for each individual in the two clusters. The x axis corresponds to population codes as identified in Table 1. The pie charts at the top represent the average proportion of cluster membership computed across individuals per site. Populations are ordered from left to right according to decreasing latitude (see also Table 1 and Figure 1b).

The non-hierarchical AMOVA (Table 3) indicated that most of the total genetic variance resided within populations (89.08%) even though the among-population component was highly significant (ΦST=0.109, P<0.001). Pairwise ΦPT values (as measures of connectivity between pairs of sites) ranged from 0.005 to 0.229, whereby 93.6% (73/78) were significant (P<0.05) based on permutation tests (see Supplementary Appendix Table S1 and Supplementary Figure S3). The PCoA plot of the first two coordinates (Figure 3a) revealed a more or less continuous north-south alignment of most populations along the first coordinate (44.3% of the total variance), which nonetheless allowed their grouping into: (1) a well-delimited ‘North’ group (SO, QU, LV, ES, TE, MO, LE); (2) the genetically heterogeneous ‘Pacific Central’ group (SL/SC and ST); and (3) the ‘Pacific South’ group (CH, OM). In addition, the second coordinate (15.6%) separated the Pacific Far South population (EO) from the latter group. Hierarchical AMOVA (Table 3) attributed a relatively high percentage of the total variance (9%) to differences among these four population groups, compared with 4.5% among populations, and 86.5% within populations. The PCoA for the first and third coordinates (Figure 3b) suggested a further subdivision of the ‘North’ group into ‘Central Highland’ and ‘Pacific North’ populations. However, the corresponding AMOVA based on five groups resulted in the same amount of among-group partitioning (9%; Table 3). Consistent with the PCoA, the midpoint-rooted NJ phenogram (Figure 4) placed the three populations from the Pacific Central Region (SL, SC, ST) at intermediate branches between the strongly supported ‘North’ group (BS 95%) and the genetically more distinct southern populations (CH, OM, EO; BS 87%), as would be expected for admixed groups (Verdu et al., 2014). Not unexpectedly, therefore, the Mantel regression of pairwise ΦPT/(1−ΦPT) values against the geographical distances among all 13 populations (Figure 5) revealed a positive and highly significant relationship (r=0.652, R2=0.425, P<0.001), indicating a strong pattern of IBD.

Table 3 Non-hierarchical and hierarchical analyses of molecular variance (AMOVAs) for AFLP variation based on several groupings of C. candidissimum populations sampled from Nicaragua
Figure 3
figure 3

Principal coordinates analysis (PCoA) for 13 populations of C. candidissimum from Nicaragua based on pairwise ΦPT values of AFLP data. Plots of (a) the first and second, and (b) the first and third coordinates. Percentages of total variance explained by each coordinate are noted in parentheses. Population symbols and colours indicate, respectively, regional affiliation (Table 1; Figure 1b) and gene pool (I–III) structure (black, I; grey, II; grey/black, I and II; see Figure 2; and white, III; see Supplementary Figure S2). Note in (b) the subdivision between ‘Central Highland’ and ‘Pacific North’ populations. Population codes are identified in Table 1.

Figure 4
figure 4

Midpoint-rooted Neighbour-Joining (NJ) phenogram depicting AFLP-derived Nei’s (1987) unbiased genetic distances between 13 populations of C. candidissimum from Nicaragua. Numbers above branches indicate bootstrap values (70%) based on 9999 replicates. Populations are labelled according to their geographic origin. Population codes are identified in Table 1.

Figure 5
figure 5

Regression of AFLP-derived ΦPT/(1−ΦPT) values against the logarithm (log10) of geographic distance (km) for 78 pairwise comparisons of 13 Nicaraguan populations of C. candidissimum (r=0.651, R2=0.425; Mantel probability (9999 permutations) P<0.001).

Discussion

Our analysis of AFLP variation within and among 13 populations of C. candidissimum from fragmented SDT forest in Nicaragua revealed three major findings. (1) Genetic diversity is low in all investigated populations and negatively correlated with latitude. (2) Genetic variation is partitioned into two major regional (latitudinal) clusters and a cline, resulting in an overall moderate level of population differentiation (ΦST=0.109). (3) The IBD pattern found may indicate spatially restricted gene flow, but this interpretation requires careful examination. In the following sections, we discuss these results and comment on their implications for conservation.

Low levels of genetic diversity in populations of C. candidissimum in Nicaragua

Tropical trees in general, and especially those that are outcrossing, and pollinated and/or dispersed by wind, tend to harbour high levels of genetic diversity at local and regional spatial scales (for example, Hamrick et al., 1993; Chase et al., 1995; Alvarez-Buylla et al., 1996; Collevatti et al., 2014). However, comparisons across such studies have to be taken with caution owing to discordances in sampling scale, genetic markers and/or data analysis. Nonetheless, our data suggest that populations of C. candidissimum in Nicaragua exhibit unexpectedly low diversity (in terms of NA, PPF, HE; Table 2). For example, if we focus on values for gene diversity (mean HE=0.125, range: 0.111–0.166), those are much lower than for populations of the likewise insect-pollinated and wind-dispersed C. spruceanum (mean: 0.284, range: 0.249–0.349) surveyed for the same marker system in the Peruvian Amazon Basin (Russell et al., 1999). The same discrepancy emerges if we consider the averages of within-population diversity (H) reported by Kremer et al. (2005) for 10 Neotropical outcrossing tree species using AFLPs under the assumption of Hardy–Weinberg equilibrium (mean: 0.23, range: 0.14–0.32) or full selfing (0.24, 0.17–0.38). Although realising that dominant AFLPs cannot provide estimates of inbreeding, we do not consider a highly selfing mating system to be a compelling explanation for the reduced genetic diversity of C. candidissimum. First, there is no firm experimental evidence to suggest that this species is truly self-compatible, as surmised by Bawa (1974; see also Methods section). And second, our global estimate of population differentiation (ΦST=0.109; Table 3) is much lower when compared to selfers (ΦST=0.65; Nybom, 2004) yet more characteristic of a highly outcrossing species (see below). This is in further agreement with field observations and melissopalynogical research indicating that C. candidissimum is a highly melliferous tree and important nectar/pollen resource, especially for butterflies (Villegas-Durán et al., 2003; Ramírez-Arriaga et al., 2011).

In the following, therefore, we discuss three alternative factors that may have acted additively and/or synergistically in causing this reduced diversity over time: biogeographic history, Pleistocene range dynamics, and recent anthropogenic deforestation.

First, this reduced diversity may reflect a signature of the species’ colonisation of Central America from South America. This scenario gains support from two congruent lines of evidence. (1) Considering Calycophyllum DC. in its traditional circumscription (ca. 10–15 spp.; Delprete, 1996; Lorence, 2012), C. candidissimum is the only Central American representative of an otherwise South American genus. (2) Although the most recent molecular phylogeny of tribe Condamineeae renders the genus polyphyletic (Kainulainen et al., 2010), it still supports a clade comprised of C. candidissimum (the type species) as sister to the South American C. multiflorum Griseb. and C. spruceanum, which together are likely sister to a clade consisting of genera from mainly South America (that is, the monotypic Bothriospora Hook.f., Dialypetalanthus Kuhlm., Wittmackanthus Kuntz (also found in eastern Panama) and Alseis Schott. (ca. 18 spp., three present in Central America)). Although the phylogeny and taxonomy of the entire group requires more work (Delprete and Jardim, 2012), all current evidence seems to indicate that C. candidissimum arrived in Central America from South America, perhaps facilitated by the most recent uplift of the Panama land bridge (ca. 3 million years ago), as proposed for many other constituents of the Central American lowland flora (for example, Burnham and Graham, 1999; Novick et al., 2003; Antonelli et al., 2009; Poelchau and Hamrick, 2013; but see Cavers et al., 2013). Thus, the overall depletion of genetic diversity in Nicaraguan C. candidissimum could, at least in part, reflect a historical legacy of the species’ colonisation from South America (see also Novick et al., 2003). This also concurs with a significant south-to-north trend of declining diversity (in terms of NA and PPF), consistent with successive (‘serial’) founder events in the direction of spread (Hewitt, 2004) and the nested (that is, potentially derived) position of northern populations in the NJ phenogram (Figure 4; see also below).

Second, other historical events that may have further contributed to the loss of genetic diversity in Nicaraguan C. candidissimum are repeated range shifts in response to the climatic fluctuations of the Quaternary (2.5 million years ago), which likely had a major impact on the current genetic structure of many Neotropical plant species (reviewed in Ramírez-Barahona and Eguiarte, 2013). Some authors have hypothesised that the ranges of SDT forest plant species in Central America were not restricted during the cooler and drier periods of the Pleistocene (Piperno, 2006) but rather attained a more continuous distribution (for example, Prado and Gibbs, 1993; Pennington et al., 2000; Novick et al., 2003). This hypothesis seems intuitively obvious as such plants are tolerant to drought resulting from seasonal rainfall or drought spells during rainy seasons (Pineda-García et al., 2013). In support of this, a recent palaeodistribution modelling of multiple Central American tree species (Poelchau and Hamrick, 2013) found little evidence to suggest that, during the LGM (Last Glacial Maximum; ca. 21 000 years ago; and possibly earlier cold periods), constituents of SDT forest (including C. candidissimum but also those of wet-topical forest) retreated into isolated humid refugia (for example, extending from southern Nicaragua to northern Panama), as postulated previously (Brown, 1987); instead, this modelling approach inferred a decrease of potential range size since the LGM for at least some of the SDT forest species studied, including C. candidissimum. Hence, following its initial colonisation of Central America, this species may have experienced subsequent reductions in historical population sizes owing to inter-/postglacial range contractions, accentuating the loss of diversity through drift.

Finally, recent anthropogenic forest fragmentation may have also contributed to the genetic erosion of Nicaraguan C. candidissimum. In general, however, molecular studies to date have found only little evidence for such an effect on levels of genetic diversity in forest trees from the Neotropics and elsewhere, which is usually ascribed to their long generation times relative to the time since deforestation (Lowe et al., 2005; Kramer et al., 2008; Ratnam et al., 2014; but for exceptions see, for example, Gillies et al., 1999; André et al., 2008; Inza et al., 2012). Yet another problem is that measures of genetic diversity differ in their sensitivity to detect such recent impact. For example, while gene diversity (HE) should be appropriate for assessing long-term (for example, historical) loss of diversity (including common alleles), it often fails to register reductions in diversity in recently bottlenecked populations owing to the loss of rare alleles (Luikart et al., 1998; Aguilar et al., 2008). This is in line with our findings in C. candidissimum populations from disturbed vs undisturbed forest in Nicaragua (Table 2): although average levels of HE (as well as NA and PPF) were reduced in the former, this difference was not significant (for example, HE=0.121 vs 0.132; P=0.248). By contrast, populations from disturbed forest exhibited on average lower levels of rare makers compared with those from undisturbed sites (DW=0.617 vs 0.754; P=0.026). We therefore conclude that the record of intensified deforestation in Nicaragua (since about the 1930’s; Griscom and Ashton, 2011) has not yet impacted overall levels of diversity (NA, PPF, HE), but has been sufficiently long (compared to the relatively short generation time of our study species) for genetic drift having caused a significant loss of rare alleles in disturbed forest fragments.

Population genetic differentiation, spatial structure and IBD

Despite the highly fragmented landscape occupied by C. candidissimum in Nicaragua, we found little evidence for strong genetic structuring across the 13 sampled populations. Although the great majority (93.6%) of estimated pairwise ΦPT values (ranging from 0.005 to 0.229) were significant (Supplementary Table S1), AMOVA revealed an overall ΦST value of 0.109 (Table 3), which is generally considered to indicate ‘moderate’ genetic differentiation (Hollingsworth and Ennos, 2004; Freeland et al., 2011). While this value is much lower than the corresponding averages reported for outcrossing (0.27) or wind-dispersed (0.25) species (Nybom, 2004), it is comparable to those of some other insect-pollinated Neotropical trees (for example, Swietenia macrophylla, ΦST=0.126, Gillies et al., 1999; Cedrela lilloi, ΦPT=0.130, Inza et al., 2012; Ficus insipida, ΦPT=0.140, Poelchau and Hamrick, 2012) and almost identical to that of Amazonian C. spruceanum (ΦST=0.094; Russell et al., 1999). Furthermore, although the STRUCTURE analysis (Figure 2) inferred that individuals do cluster into two major population groups that correspond to geographic regions (‘North’ vs ‘Pacific South’), several geographically intermediate ‘Pacific Central’ populations (SL, SC, ST) had individuals with membership to either the ‘North’ or ‘Pacific South’ group. These populations most likely reflect a continuous genetic gradation or admixture of these neighbouring groups (Rosenberg et al., 2005). This interpretation also accords with the among-population relationships detected in the PCoA and NJ analyses (Figures 3 and 4). When coupled with the strong pattern of IBD (Figure 5), these results seem to suggest that population structure in this highly fragmented landscape is still mediated through an equilibrium between genetic drift and gene flow structured by space (Wright, 1943; Slatkin, 1993).

In turn, this would imply ongoing gene flow between presently isolated populations, yet more commonly between neighbouring ones. There are three arguments in support of such a process, recently called 'IBDL' (isolation by dispersal limitation; Orsini et al., 2013). First, our estimates of population differentiation (ΦST, ΦPT) are probably not strongly biased towards pre-fragmentation conditions (Lowe et al., 2005). Although exact measurements of tree DBH (diameter at breast height) were not undertaken during this study, most sampled individuals had a DBH of ca. 23–80 cm, with only few (mostly from population OM) approaching 110 cm (A Dávila-Lara, personal observation). Assuming a DBH growth rate (mean±s.d.) of ca. 1.9±1.0 cm per year for C. candidissimum (Caughlin et al., 2016), this may indicate that at least a sizable portion of trees we have genotyped were probably not present prior to the onset of intensified deforestation around the 1930’s. Second, ensuing loss of SDT forest in Nicaragua must have greatly reduced the effective population size (Ne) of C. candidissimum there, leaving migration rate (m) viz. gene flow as the predominant, if not sole factor responsible for counteracting drift to reach, and probably maintain current equilibrium (see also Galbusera et al., 2004). And finally, as typical for a SDT forest tree, the small, winged seeds of this species are released during the dry season (January to April, when mean wind-gust speeds are at their highest in Nicaragua; Stevens et al., 2001), and thus may be dispersed by wind several kilometres, facilitating gene flow beyond the local neighbourhood (Clark et al., 2002). By comparison, pollen flow via bees or butterflies (as main pollinators) might be expected to decay more rapidly with distance from paternal trees, particularly in disturbed tropical forest canopies (for example, Hamrick and Apsit, 2004; Bonebrake et al., 2010; Duminil et al., 2016). In sum, this would suggest that the studied populations still exchange genes (mostly via seed) at a rate dependent upon the geographic distance between habitat fragments.

However, there are two potential caveats to this classical interpretation of IBD. First, estimators of population genetic differentiation (for example, FST, ΦST) only represent an integrated indirect measure of the number of migrants across generations (Nem), and therefore must be treated cautiously when used to infer levels of contemporary gene flow (Whitlock and McCauley, 1999). And second, there is increasing awareness that IBD can also be generated through a particular history of ‘serial’ colonisation and drift (reviewed in Orsini et al., 2013), unless being confounded by environmental conditions and/or local adaptation (Wang and Bradburd, 2014). At first glance, this variant scenario of ‘isolation by colonisation’ (IBC; sensu Orsini et al., 2013) seems to fit the notion that C. candidissimum is a Central American colonist (see above). However, under any persisting signature of IBC, one would expect highest levels of genetic differentiation among the most recently colonised populations (Boileau et al., 1992; Slatkin, 1993; Orsini et al., 2013; Spurgin et al., 2014), but this is not observed. Instead, the potentially derived populations of Nicaraguan C. candidissimum from the ‘Pacific North’ and Central Highland are genetically more similar to each other than to the remainder (see Figures 2,–4; Supplementary Table S1), and in spite of their dissimilar elevation ranges and (macro-)climatic conditions (Table 1; Taylor, 1963). Hence, the current data provide no conclusive evidence for IBC or particular environmental (for example, physiographic, climatic) factors influencing spatial population divergence, but also cannot exclude the possibility that other processes than IBDL (for example, historic-demographic, adaptive) underlie the observed pattern of IBD. Disentangling those phenomena, even though difficult, may be achievable through comparisons of correlations between FST and geographic and ecological distances for, respectively, neutral loci, non-neutral loci and genotypic trait values (see Orsini et al., 2013 for details) and/or coalescent-based approaches to model colonisation (including bottleneck) scenarios (for example, Henn et al., 2012; Spurgin et al., 2014; Dewoody et al., 2015).

Implications for conservation

Measures to conserve the genetic resources of C. candidissimum in Nicaragua have been called upon repeatedly (Roa, 2003; Weaver et al., 2003; Noguera-Talavera et al., 2011), mainly because of its socio-economic importance. However, apart from serving as important pollen/nectar resource for insects, this SDT forest tree also functions as foraging, nesting, shelter, and/or resting site (and so on) for a diversity of other animals (for example, howler monkeys; Garber et al., 1999; birds, bats: Harvey et al., 2006). The present data, therefore, have implications for devising conservation plans for Nicaragua’s ‘National Tree’ in the context of a community-based conservation policy also to preserve the integrity of plant-animal interactions (Ruiz-Mallén and Corbera, 2013). Currently, all populations of C. candidissimum in this country, regardless of their protection status, are facing risks of losing even more of their already low genetic diversity through reductions in population size, whether due to increased urbanisation/agricultural intensification, especially in the Pacific Region (for example, Redo et al., 2012), or stochastic factors (for example, natural or man-made fires; de Dixmude, 2001), with likely detrimental consequences for their long-term fitness and adaptation (Frankham et al., 2010). In fact, when compared with their protected counterparts (TE, MO, SC, CH, OM), the eight populations analysed here from outside nature reserves do not harbour significantly lower genetic diversity (Table 2), thus emphasising the need to extend close monitoring, preventive action, and in situ conservation to them as well. In view of limited resources, and based on our ΦST estimate of 0.109, one might argue to focus ex situ sampling efforts (for example, for provenance trials) on just two populations, representing >95% of the total genetic variation (that is, 1−(ΦST)2=0.99; see Ceska et al. (1997)). However, our data show that Nicaraguan C. candidissimum is comprised of two major regional genetic clusters (I and II), and a remarkable cline (between I and II; Figure 2), calling for a broad coverage of both in situ and ex situ conservation measures. Hence, ‘North’ (I) vs ‘Pacific South’ (II) populations should be valued separately under conservation programmes (for example, by avoiding human assisted gene exchange between them), whereas those from the Pacific Central Region (SL, SC, ST) likewise warrant special attention, as they are likely stepping-stones for natural gene flow between these groups. The same should apply to the southernmost EO population, given its unusually high diversity/rarity (Table 2), divergent gene pool (detectable at K=4; Supplementary Figure S2), and location in a transition zone between SDT and seasonal moist forest (Taylor, 1963; Olson et al., 2001). Notably, too, this divergence occurs in a region (southern Isthmus of Rivas/north-western Costa Rica) where previous large-scale phylogeographic studies have registered genetic discontinuities in dry-forest trees (Cedrela odorata: Cavers et al., 2003, 2013; Brosimum alicastrum: Poelchau and Hamrick, 2012) and various animals (for example, ants, bats, birds, reptiles; reviewed in Bagley and Johnson, 2014). Hence, only a wider sampling could clarify whether EO represents a genetically unique population or just delimitates the northernmost edge of a perhaps more widespread southern lineage.

Conclusions and research gaps

This study has assessed for the first time genetic variation for a SDT forest tree in Nicaragua. When compared with C. spruceanum from continuous Amazonian rainforest (Russell et al., 1999), Nicaraguan C. candidissimum exhibits lower genetic diversity and, despite similarly moderate overall population differentiation (ΦST), stronger geographical structure (IBD). We propose this reduced diversity more likely reflects population history (that is, the species’ colonisation from South America; Pleistocene (inter-/postglacial) range contractions) rather than recent human impact, whereas the strong pattern of IBD is probably best explained by ‘dispersal limitation’ rather than ‘isolation by serial colonisation’ (sensu Orsini et al., 2013). The spatial genetic patterns observed thus conform to expectations of a subdivided meta-population (Wright, 1943) in which drift–migration equilibrium is still maintained by ongoing gene flow, possibly owing to the species’ efficient seed dispersal capacity. Overall, these results underscore that even tree species from one of the world’s most decimated forest regions may be genetically resilient to habitat fragmentation due to species-typical dispersal characteristics that ensure nearest-neighbouring population connectivity and viability (Bacles et al., 2006; Kramer et al., 2008; Lowe et al., 2015). However, future research is required to arrive at more conclusive evidence about the recurrent and historical factors shaping genetic variation in C. candidissimum.

First, our inference of ongoing gene flow should be re-examined using approaches that can provide more direct estimates of pollen and seed movement at both local and regional scales, such as parentage and/or paternity analyses based on nuclear microsatellite loci (for example, Gerber et al., 2014). Second, such co-dominant markers are also critical to examine whether inbreeding (due to selfing and/or sib mating) is increased by the small size and isolation of these populations (that is, in terms of FIS), and whether levels of genetic diversity differ between adult trees and seedlings (for example, Davies et al., 2010; Duminil et al., 2016). Third, experimental and field data are required to assess whether the genetic processes leading to the loss of rare alleles, as found in some of the populations studied here, also contribute to decreased survival, reproductive output and/or progeny fitness of individual trees, as indicated by a recent monitoring of C. candidissimum in Southwest Nicaragua (Noguera-Talavera et al., 2011). Fourth, more integrated approaches using both neutral and non-neutral genetic markers in relation to spatial and environmental factors would be useful to test for local adaptation in C. candidissimum across its entire range in Central/northern South America (Orsini et al., 2013; Wang and Bradburd, 2014). And finally, this Neotropical tree and its allies should provide a promising study model for phylogenetic and phylogeographic research as indicated by the biogeographic and population demographic scenarios we have invoked here.

Data archiving

Data available from the Dryad Digital Repository: http://dx.doi.org/10.5061/dryad.3274n.