Plastid DNA variation

The concatenated matrix was 5,425 bp in length, representing 4.4% of the Pinus plastome based on the complete sequence of P. thunbergii. All individuals of P. jeffreyi had a ten bp inversion in the trnH-psbA spacer. Exclusion of the inversion resulted in an alignment with 91 variable but parsimony-uninformative sites and 368 parsimony-informative sites. Deleting the two outgroup sequences resulted in 23 variable and 243 parsimony-informative sites and 95 unique haplotypes (Table 1). A total of 27 indels were inferred, ten in the trnH-psbA spacer, eight in the trnD-trnY-trnE spacer region, and nine in ycf1. The region annotated as “ORF128” adjacent to the trnD-trnY-trnE spacer of P. thunbergii was disrupted by a mononucleotide A/T repeat that varied in length from 8 to 11 bp in subsections Ponderosae and Australes. The mononucleotide repeat was 6 bp in subsection Contortae and 7 bp in the two outgroup sequences.

The ycf1 gene accounted for 46.9% of the alignment and exceeded all other regions in variation, with 74.9% of the variable and informative sites (16 variable and 184 parsimony-informative) in section Trifoliae. It had the highest average nucleotide diversity per site (π) of 0.0112; in comparison, π was 0.00495 for trnD-trnY-trnE, the second most variable region, and was as low as 0.00118 for matK. The latter region had only 10 variable sites in section Trifoliae, 7 of which were parsimony-informative. Variation was comparably low in rbcL, with 11 variable sites, all of which were parsimony-informative (π = 0.00375). For the three coding regions, substitutions were most frequent in the third codon position for matK and rbcL, but in ycf1, substitutions in the first and second codon position exceeded those in the third (Figure 2). The ycf1 gene was the only region to yield more haplotypes (75) than recognized species. Subsections Australes and Ponderosae shared one matK haplotype and two trnH-psbA haplotypes. In contrast, rbcL, ycf1, and trnD-trnY-trnE could all be used to unambiguously assign individuals to a Pinus subsection. Concatenating the matK and rbcL sequences as recommended for use as DNA bar codes yielded 16 haplotypes and permitted the discrimination of four of the 49 (8%) section Trifoliae species.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 2. Variation of the plastid markers in Pinus section Trifoliae.

Above: The proportion of variable sites for each of the five markers. Below: the total number of variable sites per codon position for the three protein coding regions.

https://doi.org/10.1371/journal.pone.0070501.g002

The number of variable sites was 2.2 times greater in subsection Australes (115 variable and parsimony informative sites) than in subsection Contortae (56 variable sites) and subsection Ponderosae (56 variable sites). Subsection Australes also had the most ycf1, trnD-trnY-trnE, and trnH-psbA haplotypes (Table 1).

Phylogenetic relationships

The heuristic search of the alignment with gaps treated as missing recovered 28 most parsimonious trees (MPTs) of length (L) = 606 steps, consistency index (CI) = 0.828, consistency index excluding parsimony uninformative characters (CIexc) = 0.792, and retention index (RI) = 0.979 (not shown). Including 27 gaps using simple indel coding resulted in 394 parsimony informative and 98 variable but uninformative characters, and the heuristic search recovered 247 unique MPTs (Appendix S4; L = 648, CI = 0.826, CIexc = 0.789, RI = 0.978).

The partitioning strategy that divided the data matrix into five subsets, one for both spacer regions, one for matK + rbcL, and three for the three codon positions of ycf1, recovered the best likelihood tree and gave the lowest AIC value (Appendix S5 Figure 3). The optimal tree for the five subset partition was found in one of 50 heuristic search replicates and agreed in all main relationships with the parsimony strict consensus, although the likelihood tree was better resolved near the tips. The best models for all the partitions included either a parameter for proportion of invariant sites (I), for rate variation among sites (G), or both.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 3. Maximum likelihood tree.

Bootstrap values greater than 50% are shown at branches. A. Pinus subsection Australes. Images from top to bottom are of P. lumholtzii, P. herrerae, P. oocarpa, P. chihuahuana, and P. attenuata. B. Pinus subsections Ponderosae and Contortae. Images from top to bottom are P. scopulorum, P. douglasiana, P. pseudostrobus var. apulcensis, P. coulteri, and P. contorta. The branch leading to the outgroups has been truncated.

https://doi.org/10.1371/journal.pone.0070501.g003

The three subsections, Australes, Contortae, and Ponderosae, were recovered as monophyletic with high support, with subsection Contortae sister to subsection Australes + Ponderosae. In the subsection Australes clade, the California closed-cone pines, sometimes classified separately as subsection Attenuatae (indicated as “Attenuata” in Fig. 3a), formed a well-supported clade, but nested within it were three individuals of P. glabra from the southeastern U.S; the Attenuata clade was sister to the rest of subsection Australes. Another well-supported clade (labeled as “Leiophylla” in Fig. 3a) included P. greggii, P. chihuahuana, P. leiophylla, and a single P. lumholtzii individual. The remaining sequences formed a poorly supported clade (“Taeda”). Other groups of subsection Australes sequences that received bootstrap support greater than 70% included P. cubensis (including individuals corresponding to "P. maestrensis" here considered as a synonym of P. cubensis), P. georginae (a recently described species), P. palustris, P. caribaea, P. occidentalis, P. praetermissa, P. tecunumanii (three of four individuals), and a clade of all individuals of the southeastern U.S. species P. rigida, P. serotina, P. pungens, and P. taeda.

In subsection Ponderosae (Fig. 3b), a "Sabiniana" clade included P. coulteri sister to P. torreyana + P. sabiniana. Six P. jeffreyi individuals were sister to this clade. Identical sequences from two individuals of P. ponderosa from California were in an unresolved trichotomy with the Sabiniana + Jeffreyi clade and a clade of the remaining subsection Ponderosae sequences. Another seven identical P. ponderosa sequences from more northern coastal and interior localities occurred in a trichotomy with a "Devoniana" and a “Montezumae” clade. The Devoniana clade included southwestern U.S. and northern Mexico taxa P. scopulorum and P. arizonica vars. arizonica and cooperi, and the Mexican and Central, American species P. devoniana, P. durangensis, P. engelmannii, P. maximinoi, P. douglasiana, and P. yecorensis. The Montezumae clade comprised species with distributions in Mexico and Central America (P. pseudostrobus, P. montezumae, P. hartwegii, and a single individual each of P. maximinoi, and P. douglasiana); an individual from the Sierra Madre Oriental in northeast Mexico with unusual combinations of morphological characters, tentatively identified as P. aff. montezumae, was in a clade with P. pseudostrobus and P. montezumae, also from the Sierra Madre Oriental.

In subsection Contortae, sequences from two P. banksiana individuals were monophyletic and sister to a clade of the remaining three species. Two P. contorta var. murrayana sequences formed a clade that was successively paraphyletic to a P. contorta var. latifolia sequence and two P. virginiana individuals that were in turn paraphyletic to a clade of three P. clausa individuals.

Plastid DNA based species delimitation

Allelic monophyly was observed for 23 of the 49 recognized species (47%; Table 2), and paraphily was observed for another nine species. In subsection Contortae, haplotypes for two species (P. banksiana and P. clausa) formed monophyletic groups. Alleles were monophyletic for five of the 16 species of subsection Ponderosae (P. coulteri, P. jeffreyi, P. sabiniana, P. torreyana, and P. yecorensis) and for 16 of the 29 species of subsection Australes. Sequences were paraphyletic for the other two species of subsection Contortae and for another seven species of subsection Australes. Of the remaining 17 species with polyphyletic sequences (six in subsection Australes and eleven in subsection Ponderosae), two were considered diagnosable because all of their haplotypes were unique compared to those of all other species. One of these species was P. ponderosa, which had biphyletic but unique haplotypes. Two caveats applied in this case, first, it depended on our assignment of most morphologically similar interior populations to P. scopulorum, and second, one individual with a P. ponderosa haplotype was an artificial hybrid with P. jeffreyi. In subsection Australes, sequences of P. tecunumanii were also polyphyletic but diagnostic. The remaining 15 species had at least one haplotype that was identical to a haplotype from another species.

Molecular Dating and Lineage Delimitation Using GMYC

The relaxed molecular clock derived from the 95 unique sequences and 5415 sites and using only the secondary calibration of the MRCA of section Trifoliae and of subsections Australes-Ponderosae resulted in a crown group divergence for section Trifoliae of 18.0 ma with a 95% highest posterior density (HPD) of 14.7–21.4, a 14.0 ma (95% HPD 11.0–16.0) age for the MRCA of Australes-Ponderosae, and crown group ages of 8.8 (95% HPD 4.8–13.6), 10.3 (95% HPD 6.8–13.8), and 7.7 ma (95% HPD 4.3–11.6) for subsections Contortae, Australes, and Ponderosae, respectively (Appendix S6). Inclusion of the fossil calibration point for the Attenuata clade (Figure 3) gave the same age estimates for section Trifoliae (18.0 ma; 95% HPD 14.8–21.4; Figure 4), and for the MRCA of Australes-Ponderosae (14.0 ma; 95% HPD 11.0–17.0). This also resulted in ages of the same three subsectional crown nodes of 9.0 (95% HPD 4.9–13.5), 10.2 (95% HPD 6.9–13.7), and 7.7 ma (95% HPD 4.3–11.4; Figure 4).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 4. Chronogram for Pinus section Trifoliae using three calibration points and lineage-through-time plot.

Chronogram for section Trifoliae. The outgroup has been removed. The two secondary calibration points inferred from a previous relaxed molecular clock are indicated with an asterisk (*), and a fossil calibration point attributed to the California closed-cone pines is indicated with a double asterisk (**). The corresponding lineage-through-time plot is given on the upper left. The transition from speciation to coalescence branching processes gave nonsignificant test results.

https://doi.org/10.1371/journal.pone.0070501.g004

The trees resulting from both calibrations gave similar GMYC results (Appendix S7). Here we report the statistics for the tree that included both the secondary and fossil calibration points. The lineage-through-time plot did not show a sudden increase in slope indicative of a transition from (Yule) speciation branching rates to coalescent branching rates (Figure 4). The single threshold model did not result in a significant improvement in likelihood over the null coalescent model (likelihood ratio = 3.17, test statistic = 0.21, not significant). Nevertheless, it gave a threshold time estimate of 2.51 ma, resulting in 20 clusters (C.I. estimated as two log-likelihood units from the maximum likelihood solution = 1–25), and an additional eight lineages consisting of single sequences, giving a total of 28 entities (C.I. = 1−36). Comparison of the multiple to the single threshold method gave a likelihood ratio of 10.08 (test statistic = 0.0065, significant). The multiple method gave threshold times of 7.69, 3.22, and 2.05 ma, resulting in 14 clusters (C.I. = 10−20) and 18 entities (C.I. = 12−27). Application of multimodel AICc (delta AICc = 7; 76 best of 142 total models) yielded an estimate of 14.27 clusters (var. = 4.84) and 22.92 entities (var. = 9.19).

We also examined the cluster probabilities (the probability that two tips of the tree belong to the same cluster) assigned to the tree nodes based on the multimodel AICc. A probability threshold of 0.95 identified a total of 34 entities, four for subsection Contortae (three clusters and one singleton), 11 for subsection Ponderosae (5 clusters and 6 singletons), and 19 for subsection Australes (14 clusters and 5 singletons). Seventeen of these entities corresponded to a single species, ten corresponded to two species, six corresponded to three species, one corresponded to four species, and one corresponded to five species. Lowering the cluster probability threshold to 0.50 reduced the number of entities to 25.

Prospects for plastid DNA bar codes of pines

DNA sequences offer great potential for providing fast and accurate species identifications [75], but the short matK and rbcL fragments chosen for plant DNA bar codes are known to have low substitution rates and thus low discriminatory power [14]. Low variation in the matK and rbcL of Pinus section Trifoliae was documented previously [24,35], but not specifically as DNA bar code markers. The matK and rbcL fragments corresponding to the DNA bar codes were the least variable of the five regions evaluated. Although matK and rbcL were variable enough to place an unknown sample to section Trifoliae, only the rbcL fragment was capable of placing all individuals to one of the three subsections. Both matK and rbcL were usually identical for closely related species and this was reflected in a low success rate for species discrimination. Evaluation of seven plastid markers across land plants reported an average of 72% species discrimination when using matK + rbcL in combination [14]. This same combination gives only 8% species discrimination in Pinus section Trifoliae, a decidedly poor performance for some of the world’s most ecologically and economically important trees.

The trnH-psbA and trnD-trnY-trnE spacers were of comparable length but more variable than the matK and rbcL fragments. However, they were not variable enough to discriminate most species within Pinus subsections, particularly in subsection Ponderosae. The trnH-psbA spacer can be amplified with universal primers designed originally for flowering plants, which makes it an attractive option for adding discriminatory power when using multiple loci for DNA bar coding taxonomically diverse floras. However, two trnH-psbA haplotypes were shared across subsections Australes and Ponderosae and we were unable to obtain bidirectional reads for subsection Contortae.

Only ycf1 yielded on average more than one haplotype per species. The greater relative length of the sequenced fragment (ca. 2400 bp vs. 600–800 bp) was an important factor, but ycf1 was also by far the most variable on a per-site basis (Figure 2), and therefore was more useful for discriminating plastid lineages, even when considering only a 600-800 bp fragment that can be sequenced with typical bidirectional Sanger sequencing reads. The greater variation and an excess of nonsynonymous compared to synonymous substitutions are consistent with positive selection acting on ycf1 in Pinus [76]. Primers have been designed to amplify ycf1 throughout Pinaceae [77], but these may not work in other conifer families, much less for more distantly related land plants. Therefore, although ycf1 has already proved useful for identifying pine species [78], it does not fulfill the universality requirement of a DNA bar code. Nevertheless, if a sample is already known to belong to Pinus or Pinaceae (e.g., based on morphology or on an rbcL sequence) then genus or family specific PCR primers could be used to determine its ycf1 haplotype, which in turn is a useful proxy for species identification.

Phylogenetic relationships within Pinus section Trifoliae

Here we provide the first phylogenetic analysis of plastid DNA for North American hard pines that includes all recognized species and multiple individuals per species. Parsimony and maximum likelihood analyses of plastid DNA recovered three principal lineages of North American hard pines, subsections Australes, Contortae, and Ponderosae, in agreement with previous plastid studies with less taxonomic sampling [24,32,34,76,79]. Morphological synapomorphies are unknown for section Trifoliae, and although these subsections coincide in some respects with the influential classification of Little and Critchfield [31] based on morphology and crossability, emphasis on a limited subset of morphological characters resulted in a classification that conflicts with the plastid tree in several respects. If we take the classification of Little and Critchfield to illustrate this point, these authors did not recognize the species of section Trifoliae as a natural group, instead classifying hard pine species with deciduous fascicle sheaths, P. leiophylla, P. chihuahuana (as a variety of P. leiophylla), and P. lumholtzii in a separate, morphologically heterogeneous section that also included some Eurasian pines. Subsection Australes as recognized here was classified by them into subsections Australes and Oocarpae, the former characterized by multinodal spring shoots, and mostly symmetrical cones, and the latter characterized by mostly oblique, serotinous cones. Pinus lawsonii and P. teocote, two species endemic to Mexico with symmetrical nonserotinous cones were classified by Little and Critchfield in subsection Ponderosae but are now classified in subsection Australes based on crossability data and DNA sequences. Finally, Little and Critchfield classified the "California big-cone pines" in subsection Sabinianae, separate from subsection Ponderosae. These differences are discussed more specifically below.

A similar plastid DNA data set for Pinus subsection Ponderosae was reported previously [35]. The main differences here are the use of a slightly longer and contiguous fragment of ycf1, an increase from 67 to 86 individuals, and the use of broader taxonomic concepts for four species. As with the earlier version of the plastid data, the three California big-coned pines, P. coulteri, P. sabiniana, and P. torreyana were monophyletic (the Sabiniana clade; Figure 3b). A close relationship between these species was recognized based on similarities in growth form, leaf, cone, and seed morphology, and their ability to form natural or artificial hybrids [80]. This group thus provides an example of congruence between plastid DNA relationships and other sources of evidence. Furthermore, both P. coulteri and P. torreyana can hybridize with P. jeffreyi, a species which also occurs in California, and both P. jeffreyi and P. coulteri can form hybrids with Californian populations of P. ponderosa. This genetic link between California big-coned pines and other members of subsection Ponderosae is reflected in the plastid tree, which recovers P. jeffreyi as the sister group to the California big-coned pines, and two Californian collections of P. ponderosa (one is actually a hybrid with P. jeffeyi), in an unresolved trichotomy with all other species of subsection Ponderosae.

Ponderosa pine, or P. ponderosa, is one of the most taxonomically challenging taxa in subsection Ponderosae. In the broad sense, P. ponderosa also includes P. scopulorum as a variety (P. ponderosa var. scopulorum Lemmon) [27,28,80]. Only minor, possibly clinal, morphological differences separate these taxa (e.g., P. scopulorum has higher proportions of two, rather than three leaves per fascicle, and P. ponderosa has longer leaves and larger ovulate cones and seeds). In contrast to their morphological similarity, sequences from P. ponderosa and P. scopulorum were polyphyletic, occurring as three divergent lineages both here and with nearly complete plastomes [34]. One of these lineages is distributed mainly in the southern Rocky Mountains and northern Mexico (“southern interior ponderosa pine” or P. scopulorum), and is related to other taxa from the same geographic region like P. arizonica, which has also been synonymized with P. ponderosa by some workers [31]. The other two lineages correspond to “northern interior ponderosa pine” (P. ponderosa in a restricted sense) and “Pacific coastal ponderosa pine” (P. ponderosa var. benthamiana or P. benthamiana) [34]. Our analysis included three individuals from the Rocky Mountains of Utah, but only two of these grouped with P. scopulorum, while the third was morphologically atypical (e.g., it had longer leaves), and grouped with northern P. ponderosa. Populations of ponderosa pine from Utah and Nevada are extremely variable in the proportion of needles in fascicles of three, typical of northern and Pacific P. ponderosa, and needles in fascicles of two, here treated as P. scopulorum [81]. This may be a result of hybridization, but more study is needed on these ecologically important and emblematic taxa to characterize their morphological and genetic variation.

Whereas we departed from recent taxonomic treatments [27,28] by recognizing P. scopulorum as separate from P. ponderosa, we followed them in treating P. donnell-smithii as a synonym of P. hartwegii and P. nubicola as a synonym of P. pseudostrobus var. apulcensis. Nevertheless, even with these broader species concepts, P. hartwegii, P. pseudostrobus, and P. montezumae had very low sequence divergence and shared haplotypes, consistent with previous studies reporting interspecific gene flow [7,82].

The broad circumscription of Pinus subsection Australes is based primarily on plastid DNA and to a lesser extent on the internal transcribed spacer of nuclear ribosomal DNA. In earlier classifications these species were divided into three subsections, Attenuatae, Australes, and Oocarpae [26,31], but none of these proposed groups was monophyletic in the plastid trees reported here or elsewhere [24,34,79]. Pinus subsection Attenuatae (“the California closed-cone pines”), was erected for P. attenuata, P. muricata, and P. radiata, the only three far western United States and Baja Californian species in this subsection, all with serotinous cones [26,83]. The three species formed a well-supported group, but included all three P. glabra sequences (Figure 3b). The latter species is also nested within the California closed-cone pines with nearly complete plastomes [34]. This species’ cones are non-serotinous, and it is geographically disjunct from the three Californian species, occurring in the southeastern United States. The position of P. glabra is one of the best possible examples of genealogical discordance of plastid DNA in Pinus section Trifoliae.

The ten species in subsection Australes as originally circumscribed by Little and Critchfield are distributed in eastern North America, the Caribbean, and Central America, and have symmetrical, non-serotinous cones and multinodal spring shoots. Four of these (P. pungens, P. rigida, P. serotina, and P. taeda) formed a well-supported clade, but nested among species of Little and Critchfield’s subsection Oocarpae. The placement of the other four species was not robust. Therefore, our results do no support the recognition of subsection Oocarpae but corroborate other molecular studies and the resulting subsectional classification of this group [24,34]. The nonmonophyly of Little and Critchfield’s subsections needs to be confirmed with independent data such as from another genomic compartment or morphology.

Delimiting species with plastid DNA

Twenty-three of the 49 Pinus subsection Trifoliae species exhibited allelic monophyly, another nine were paraphyletic, two had polyphyletic but unique haplotypes, and 15 shared at least one plastid haplotype with another species. One explanation for shared haplotypes among species is that the morphological characters that distinguish these species are minor and insufficient to justify species status. For example, P. serotina is treated as a subspecies of P. rigida (P. rigida subsp. serotina) by Eckenwalder [27]. Here the four individuals sampled for these taxa (two each) had identical haplotypes. Also, P. douglasiana and P. maximinoi are recognized as distinct species in recent taxonomic treatments, but they are difficult to distinguish morphologically, and several of their sequences are identical. However, introgression is probably the most important cause of discordance between plastid DNA trees and species trees in plants [2]. Controlled crosses have thoroughly documented the weak intrinsic barriers to gene flow within Pinus subsections [80,84,85], and plastid DNA introgression in natural populations has been reported for all three subsections of North American hard pines [6,7,82,86,87]. Recent diversification via allopatric speciation followed by secondary contact may have promoted interspecific gene flow in this group. The range of intra- and interspecific morphological variation in species such as P. arizonica, P. durangensis, P. ponderosa, and P. scopulorum (subsection Ponderosae), and P. tecunumanii, P. lawsonii, P. patula, and P. pringlei (subsection Australes), is imperfectly understood, and natural hybridization is often suspected based on morphological intermediates. In these species, morphological and plastid DNA variation is relatively high, and cases of haplotype sharing occurs among different clades. If interspecific gene flow is a rare but taxonomically widespread phenomenon in Pinus section Trifoliae, then more examples of shared plastid DNA lineages should be found with increased intraspecific sampling.

The assignment of individuals to plastid DNA haplogroups has intrinsic and practical value for providing preliminary species identifications and studying species limits, but even in the absence of introgression, estimates of interspecific variation and the proportion of nonmonophyletic species should increase as intraspecific sampling increases [88]. Consequently, the exclusive reliance on plastid DNA haplotypes for identifying pine species would require us to accept a high error rate. The accuracy of both species identification and delimitation would be greatly improved by the development of additional morphological and genetic markers that can more accurately assess variation and interspecific gene flow.

Lineage estimation using GMYC

Studies in animals and bacteria have concluded that the GMYC method yields reasonable DNA based species number estimates, and that in animals these are in line with estimates based on morphology [18,19,21,89]. A potential advantage of the GMYC method is that a priori assignment of haplotypes to taxa is irrelevant for quantifying the number of species, thus the method may be more suited to analyzing single locus data sets in taxonomic groups where introgression or incomplete lineage sorting are suspected to have occurred.

Given that there is some disagreement in the number of species in Pinus section Trifoliae, we explored whether a molecular estimate might favor one taxonomic treatment over another. However, over half of the sequences we obtained were identical, no increase was observed in the slope of the lineage-through-time plot, and the GMYC method gave exceptionally low point estimates with broad confidence intervals. In particular, the multimodel point estimates detected 22 or 23 lineages of North American hard pines, or 45-47% of the 49 species that we recognize based on our admitedly more subjective evaluation of morphology and plastid DNA variation. The relaxed molecular clocks gave Miocene crown diversification times for each subsection and the transition time estimates from Yule to coalescent processes were very recent (Pliocene or Pleistocene). The GMYC method has been reported to be less appropriate for recent radiations, and in simulation studies, the method’s accuracy decreased with increasing effective population size [90]. However, the disparity between taxonomic and GMYC estimates found here may be due mainly to low plastid DNA sequence variation and consequently few unique haplotypes per species, thus the application of GMYC with plant plastid DNA may give better estimates as longer sequences are used. Empirical studies are needed to determine whether greater range-wide taxonomic sampling and inclusion of more plastid sequence data will improve the ability of the GMYC methods to recognize more lineages.