A central problem in evolutionary biology is identifying factors that promote the evolution of reproductive isolation. Among mosses, biogeographic evidence indicates that the potential for migration is great, suggesting that biological factors other than geographic isolation may be critical for speciation in this group. The moss Physcomitrella patens (Funariaceae) has long been used as a model for interspecies hybridization and has recently emerged as an important model system for comparative genomics. We report genealogical analyses of six loci from several populations of P. patens and related species in the genus Physcomitrium. These results unambiguously indicate that the so-called genus Physcomitrella arose at least three times from distinct ancestors within the genus Physcomitrium. In spite of the evidence for natural hybridization in the Physcomitrella–Physcomitrium complex, genealogical and experimental hybridization data indicate that the taxonomically defined species are reproductively isolated. However, these analyses suggest that Physcomitrium eurystomum was formed from a hybridization event between two early diverging lineages in the complex, and that the ancestral population size of these lineages was much smaller than the current population sizes. We discuss these findings in the context of the inferred mating system in the Physcomitrella–Physcomitrium complex and patterns of speciation and diversification.

Geographic isolation has long been viewed as a critical step in the process of speciation and the generation of biodiversity (Mayr 1963; Barraclough and Vogler 2000; Coyne and Orr 2004). Within many moss species, however, mounting evidence from floristics (van Zanten 1978; Miller and McDaniel 2004; Hutsemekers et al. 2008), molecular population genetics (Shaw et al. 2003; Van der Velde and Bijlsma 2003; McDaniel and Shaw 2005; Vanderpoorten and Long 2006; Heinrichs et al. 2007; Huttunen et al. 2008; Szovenyi et al. 2008), and phylogenetics (Shaw et al. 2005a) indicates that geographic barriers may not cause a long-term impediment to migration. This observation suggests that genetic, demographic, or ecological factors, in addition to geographic isolation, cause the cessation of gene flow between nascent species. The nature of these biological factors, however, is largely unknown. The moss Physcomitrella patens and its relatives have long been model systems for studies of hybrid interfertility (von Wettstein 1924; Bryan 1957). P. patens is now emerging as a model plant for comparative genomics, and studies including P. patens have provided insights into the evolution of land plant gene function and genome structure (Nishiyama et al. 2003; Lang et al. 2005; Rensing et al. 2005; Stenoien 2005; Richardt et al. 2007; Rensing et al. 2008; reviewed in Quatrano et al. 2007). As a consequence, P. patens also holds promise as a model for understanding the process of speciation (von Stackelberg et al. 2006; Kamisugi et al. 2008).

Phylogenetic studies have shown that the monotypic genus Physcomitrella is closely related to the genus Physcomitrium, based on analyses of DNA sequences from the chloroplast (Goffinet and Cox 2000; Goffinet et al. 2007; Werner et al. 2007). The two genera are most easily distinguished by the morphology of the diploid sporophyte; the Physcomitrella sporophyte is nestled among the gametophyte leaves, rather than elevated on a seta like in Physcomitrium and most other mosses, and lacks the specialized abscission layer around the opening of the sporangium. Both genera are cosmopolitan in distribution, but Physcomitrium is more diverse than Physcomitrella, with a total of seven species found in Europe and temperate North America (Crum and Anderson 1981; Hill et al. 2006). Populations containing species of both genera are frequent in some habitats, and hybrids are widely reported (Britton 1895; Andrews 1918; Loeske 1929; Andrews 1942; Pettet 1964; Tan 1978). Therefore, the evolutionary relationships within the Physcomitrella–Physcomitrium complex may not be captured by a single gene or organelle, because ongoing hybridization may cause different parts of the genome to have different histories.

In spite of the evidence for hybridization, the species of Physcomitrella and Physcomitrium are phenotypically distinct, even in sympatric sites (e.g., Crum and Anderson 1981), suggesting that the evolutionary effects of hybridization may be limited. In a landmark series of experiments, von Wettstein (1924, 1928, 1932) and colleagues (Schmidt 1931; Bauer and Brosig 1959) generated artificial crosses among members of the Funariaceae, including P. patens and three European species of Physcomitrium (the cosmopolitan P. pyriforme, and the endemics P. sphaericum and P. eurystomum). In general, the recombinant haploid progeny (i.e., spores that develop into gametophytes) of most intergeneric hybrids were inviable or developmentally abnormal, making it unlikely that natural recombinants would contribute to subsequent generations (reviewed in Bryan 1957). However, occasionally a few recombinants were viable and vigorous, suggesting that the fortuitous segregation of species-difference loci (either karyotypic differences or epistatically interacting factors) may allow for the production of fertile progeny. Von Wettstein (1932) reported that the frequency of such vigorous recombinants was greatest in crosses in which P. patens was the maternal parent, although this frequency also depended upon the paternal species. These data have engendered a considerable amount of speculation regarding the role of introgressive hybridization in the evolution of the Funariaceae, but neither additional crossing studies nor molecular population genetics analyses of hybridization have been undertaken in the PhyscomitrellaPhyscomitrium complex.

In principle, the signature of introgressive hybridization is readily observable by comparing the genealogical patterns at multiple independent loci (Sang and Zhong 2000; McBreen and Lockhart 2006). Ongoing gene flow should cause allele sharing between distinct species and elevated levels of genetic variation within the hybridizing species, because allelic variants coalesce before the origin of either species. Historical gene flow may be indicated by conflicting genealogical relationships among loci. However, genealogical conflict due to ancient introgression can be difficult to distinguish from the retention of ancestral polymorphism (incomplete lineage sorting). This problem is particularly acute where multiple speciation events have occurred over a short period of time relative to the amount of segregating variation in the ancestral population (Maddison 1997). The coalescence times in large populations are long, increasing the probability that the branching order in a particular gene genealogy will not reflect the true order of speciation events. Moreover, because a given genealogy may arise under a broad range of demographic processes, inferences based on genealogical shape alone are generally insufficient to rule out particular demographic histories (Kuhner 2009). Recently, however, researchers have developed approaches that estimate historical demographic parameters over a range of probable genealogies, providing a means to explicitly distinguish gene flow from incomplete lineage sorting (Hey and Nielsen 2004, 2007).

To test for introgressive hybridization, we examined patterns of genetic variation in primarily European isolates of Physcomitrium and Physcomitrella, using molecular evolutionary analyses of the DNA sequence variation from one chloroplast spacer (atpB-rbcL), the nuclear ribosomal internal transcribed spacers (ITS), and intron-containing portions of four nuclear protein-coding loci (adenosine kinase, adk; adenosine 5′-phosphosulfate reductase, apr; phosphoadenosine–phosphosulphate reductase, papr; and heme oxygenase, ho). We included isolates from geographically widespread populations of four species of Physcomitrium, and four subspecies of P. patens, focusing in particular on the species where fertility estimates from interspecific hybrids were available in natural (Pettet 1964) and experimental crosses (von Wettstein 1924, 1928). Our results indicate that P. patens has arisen at least three times from within the genus Physcomitrium, and although the species shows no history of introgression, other species within Physcomitrium show a clear signature of hybridization. We discuss these findings in the context of the patterns of interfertility in the Physcomitrella–Physcomitrium complex, and patterns of speciation and diversification in general.

Materials and Methods


We obtained cultures from multiple isolates of Physcomitrella patens (four from subsp. patens, two from subsp. californica, and one each from subsp. magdalenae and readeri) and from four species of Physcomitrium (two from P. collenchymatum, three from P. eurystomum, six from P. pyriforme, and three from P. sphaericum; Table 1, Fig. 1). This represents a large portion of the distributions of P. patens, P. eurystomum, and P. sphaericum, and a relatively small portion of the distribution of the cosmopolitan P. pyriforme (and although P. collenchymatum is narrowly distributed, we have only a single accession of this species). Live cultures of all samples are accessioned in Freiburg, Germany (see and have been deposited in and are available from the International Moss Stock Center (

Table 1.  Species and localities of isolates of the Physcomitrella–Physcomitrium species complex used in this study.
 Species and localityLat.Long.VoucherIMSC no.
Physcomitrella patens subsp. patens
  1Gransden Wood, Cambridge, U.K.52°12′N00°11′EWhitehouse 196240001
  2Reute, Baden-Württemberg, Germany48°00′N07°51′EStackelberg and Lüth 04.01.200640040
  3Villersexel, Haute Saône, France47°33′N06°26′ELüth 17.10.200340012
Physcomitrella patens subsp. californica
  4del Valle Lake, CA37°37″N121°45′WMishler 24.10.200440039
  5Saitama, Honshu, Japan36°25′N139°30′EHiguchi 4424940016
Physcomitrella patens subsp. readeri
  6Melton Reservoir, Victoria, Australia37°43′S144°32′EMEL 218301340022
Physcomitrella patens subsp. magdalenae
  7Mt. Bisoke, Ruhengeri, Rwanda01°28′N29°30′EBuchbender RWA-VB-010740025
Physcomitrium sphaericum
  8Grosshartmannsdorf, Sachsen, Germany47°08′N15°55′EFrahm 491940018
  9Imsbach-Aue, Saarland, Germany49°29′N07°04′ESaarland-Herbar Moose 620240047
  10Vellescot, Territore-de-Belfort, France47°34′N07°01′ETeichufer, Belegnr 428340043
Physcomitrium eurystomum
  11Neukirch, Baden-Württemberg, Germany48°01′N08°11′ESchäfer-Verwimp 28.08.200140048
  12Neustadt, Thüringen, Germany50°45′N11°45′EEckstein 313940052
  13Schleiz, Thüringen, Germany50°34′N11°50′EEckstein 376740051
Physcomitrium pyriforme
  14Bischofswerda, Sachsen, Germany51°11′N14°16′EEckstein 265540053
  15Nordhausen, Thüringen, Germany51°32′N10°39′EEckstein 284940054
  16Gera, Thüringen, Germany50°51″N12°11″EEckstein 394740055
  17Haardtrand, Rheinland-Pfalz, Germany49°17′N08°07′EHerbar-Nr P17.23640056
  18Oevergran, Uppland, Sweden59°51′N17°38′ELönnell 88240057
  19Madeira, Portugal32°50′N17°13′EEckstein 47240059
Physcomitrium collenchymatum
  20Shaw Nature Reserve, MO38°29′N90°23′WHomberg 115540061
  21Shaw Nature Reserve, MO (no ITS)38°29′N90°23′WHomberg 115540062
Funaria hygrometrica
 Germany (apr, ITS, atpB-rbcL)  Laboratory strain40017
Figure 1.

Geographic origins of isolates of the PhyscomitrellaPhyscomitrium complex used in this study. Numbers correspond to those in Table 1.

Genomic DNA was extracted from axenic tissue of each haploid accession grown on standard media as described in Bierfreund et al. (2003), using modified versions of a cetyltrimethyl ammonium bromide (CTAB) extraction (following McDaniel and Shaw (2005) for atpB-rbcL spacer, adk, and ho, and von Stackelberg et al. (2006) for ITS, apr, and papr). The six loci were amplified using the primers listed in Table S1. The atpB-rbcL spacer, adk, and ho were amplified in 16-μL polymerase chain reactions (PCR) containing 8-μL GoTaq MasterMix (Promega, Madison, WI), 1 μL 10 μM forward and reverse primers, and 10–20 ng of genomic DNA. The PCR cycling conditions were 94°C denaturing temperature for 30 sec, 50°C annealing temperature for 30 sec, and 72°C extension for 1 min 30 sec for 30 cycles. The PCR products were cleaned with QIAquick columns (Qiagen, Hilden, Germany) and sequenced using 10–20 ng of product in a 10-μL cycle-sequencing reaction (BigDye Terminator Reaction, version 3.1, Applied Biosystems, Foster City, CA). The chromatograms were edited in Sequencher 4.8 (GeneCodes Corp., Ann Arbor, MI) and aligned using Se-al 2.0 software (

Although we were generally able to generate clean sequences directly from the PCR products, to test for the presence of paralogs due to polyploidy, which are common in the genus (Rensing et al. 2007), we cloned sequences of adk and ho from an accession of each species and all three accessions of P. eurystomum. The PCR products were cloned into an E. coli vector using a TOPO-TA cloning kit (Invitrogen, Carlsbad, CA), and colonies were screened after 24 h of growth. Colonies with inserts present were amplified in a 12-μL PCR reaction, using M13 universal vector primers and reagents described above, under the following cycling conditions: initial 94°C for 4 min, 94°C denaturing temperature for 1 min, 55°C annealing temperature for 1 min, and 72°C extension for 2 min, for 30 cycles. The PCR products were cleaned and sequenced using the same protocols as described above. Four high-quality sequences were generated for each extraction that was cloned.

The ITS, apr, and papr regions were amplified in a 20-μL PCR mix containing 2 μL of 10 × RED-Taq-PCR buffer, 0.1 mM dATP, dCTP, dGTP, and dTTP, 5 pmol each of two primers, 0.5 Units RED-Taq-Polymerase (SIGMA-Aldrich, St. Louis, MO), and 4 ng of genomic DNA. Cycling was carried out in T1 thermal cyclers (TGradient, Biometra, Goettingen, Germany) starting with an initial DNA denaturation at 95°C for 2 min. The first cycle consisted of 30 sec denaturation at 94°C, primer annealing for 30 sec at 62°C, and elongation for 60 sec at 72°C. In each of the 13 subsequent cycles, the annealing temperature was decreased by 0.7°C. The final 24 cycles consisted of 30 sec denaturation at 94°C, 30 sec primer annealing at 52°C, and 70 sec elongation at 72°C. A final elongation was performed for 4 min at 72°C. We were unable to directly sequence ITS, apr, and papr in some accessions. We therefore cloned the amplified fragments prior to sequencing. PCR products were directly ligated using the TOPO-TA cloning kit (Invitrogen, Carlsbad, CA). Vectors were transformed into an E. coli XL1-Blue strain and selected using ampicillin resistance and blue–white screening. The plasmids were isolated using the E.Z.N.A. Plasmid Miniprep Kit II (Peqlab, Erlangen, Germany) and the inserts were sequenced using vector-specific standard M13 primers. Sequences were quality clipped and filtered using the filtering module of the Paracel Transcript Assembler (Striking Development Inc., Los Angeles, CA) and assembled using Vector NTI Advance 9 (Invitrogen). Initial alignments were performed using Muscle 3.51 (Edgar 2004) and adjusted by eye.


To understand the relationships among the sampled taxa, a genealogy was constructed from the aligned sequences from each locus separately using branch-and-bound searches under the maximum parsimony criterion with PAUP* 4.0b10 (Sinauer Associates, Inc., Sunderland, MA). Characters were treated as unordered and equally weighted, and indels were coded as binary characters. Branches were collapsed if the minimum length was zero. Bootstrap analysis used 1000 pseudoreplicates subject to branch-and-bound searches. The atpB-rbcL spacer, ITS, apr, and papr genealogies were rooted with a sequence from Funaria hygrometrica. To test the monophyly of P. patens, we performed searches with the species constrained to be monophyletic, and compared this to the unconstrained topology using a Shimodaira–Hasegawa test (Shimodaira and Hasegawa 1999) implemented in PAUP* assuming the HKY model of evolution.

To assess the significance of the interlocus phylogenetic conflict, we compared the phylogenetic signal among loci (including the indels) using the partition homogeneity test (Farris et al. 1994) as implemented in PAUP* under parsimony with branch-and-bound searches. We then conducted additional branch-and-bound parsimony analyses on subsets of the complete dataset, pruning isolates that were strongly supported in different topological locations across loci. We continued pruning isolates until the partition homogeneity test yielded a nonsignificant result (i.e., no statistically significant topological conflict). Because we found evidence of significant conflict when P. eurystomum, P. collenchymatum, and several isolates of P. pyriforme were included in the analysis, we did not estimate a genealogy from the combined data, but rather constructed a species tree based on combinable splits in the individual gene trees.

To evaluate whether the topological conflict resulted from introgressive hybridization or incomplete lineage sorting, we generated maximum likelihood estimates (MLEs) of current and ancestral population sizes, migration rates, and divergence times, using the program IMa (Hey and Nielsen 2007). This software implements a Markov Chain Monte Carlo (MCMC) search strategy to identify MLEs of demographic parameter values given in the sampled data. Flat likelihood surfaces for divergence time or migration rates between lineages, combined with large ancestral population sizes, would indicate that we could not reject lineage sorting as the source of a topological conflict. Narrow MLEs for divergence and migration, and a small estimated ancestral population size, in contrast, would suggest that incomplete lineage sorting was unlikely to cause topological conflict. This approach, however, assumes a bifurcating isolation-with-migration speciation model. P. eurystomum appears to violate this bifurcating model, as it contained alleles closely related to P. sphaericum and P. pyriforme at all nuclear loci, suggesting that it might be a hybrid species. We therefore indirectly evaluated the likelihood that the genetic composition of P. eurystomum resulted from introgressive hybridization, rather than incomplete lineage sorting, by estimating demographic parameter values for the two putative parental lineages, sphaericum and pyriforme (see the Results section).

We used IMa to perform a pairwise comparison between the inferred parental lineages of the putative hybrid species, excluding the taxa that showed evidence of hybridity (P. eurystomum and P. collenchymatum). We conducted an initial run of the program to establish parameter maxima where the MLEs reached zero probability, and an appropriate burn-in period. We then conducted three additional runs of 10 million steps, each with a burn-in of 10,000 steps, a different seed number, and narrower parameter maxima, and checked convergence among the three runs. To obtain joint-parameter estimates, we used IMa in L-mode, an approach that compares nested demographic hypotheses using likelihood ratio tests. We present the joint parameter estimates from this run.

This approach assumes free recombination between loci, no recombination within loci, an absence of selection on the surveyed loci, and random mating. The software dnaSP 4.0 (Rozas et al. 2003) was used to estimate the quantity θ (4Neμ, the genetic effective population size) for each locus, historical recombination in each locus (Hudson and Kaplan 1985; Hudson 1987), and deviations from neutral-equilibrium expectations in mutation frequency spectrum and fixation of amino acid changing substitutions. We also tested for recombination in the concatenated matrix of the four nuclear loci in the entire data matrix, and various subsets of the taxa based on lineages found in the phylogenetic analyses. The output from this analysis was visually inspected to identify sequences containing putative recombination events. In cases where evidence of recombination was found, we divided the sequences into segments that did not violate the four-gamete test, based on the output from dnaSP. It also is not clear whether the sphaericum and pyriforme lineages undergo random mating, as assumed by the IM approach, because members of the Physcomitrium-Physcomitrella complex are self-fertile. We conducted the IMa analyses with various partitions of our data and got similar results, leading us to suspect that any additional structure within these lineages does not substantially alter parameter estimates.

IMa generates parameter estimates expressed in units of 4Neμ, tμ, and m/μ, where μ is the neutral mutation rate per generation, t is the divergence time in generations, and m is the migration rate per generation. To convert these to Ne, t, and m, we assumed that Physcomitrella and Physcomitrium have an annual life cycle (one generation per year). We used a rate of 9.4 × 10−9 neutral mutations per site per generation, following Rensing et al. (2007). Given the uncertainty in estimated rates of molecular evolution, we intend these parameter estimates to be taken only as a rough guide.



To understand the evolutionary patterns in the PhyscomitrellaPhyscomitrium complex, we estimated genealogies from six loci including 21 isolates from within the complex. The individual genealogies are shown in Figure 2, and the details regarding the loci and reconstruction of these genealogies are shown in Table 2. The position of the root, determined using F. hygrometrica as an outgroup, is shown for the apr, papr, ITS, and atpB-rbcL genealogies, although alternate rootings could not be rejected for the latter two genealogies. The alignments for the four nuclear loci (adk, apr, papr, and ho) were unambiguous, phylogenetic analyses of these loci resulted in well-supported relationships, and all four loci exhibited similar numbers of segregating sites (Fig. 2, Table 2). In contrast, in the ITS alignment, it was difficult to assess confidently positional homology, whereas the atpB-rbcL alignment contained too little variation to distinguish the genealogical relationships among all of the species. Therefore, we principally focus on relationships produced from analyses of the nuclear protein-coding loci, referring to the ITS and the atpB-rbcL spacer genealogies where they specifically support particular relationships.

Figure 2.

Figure 2.

Maximum parsimony genealogies of members of the PhyscomitrellaPhyscomitrium complex from (A) adk; (B) apr; (C) papr; (D) ho; (E) ITS; and (F) atpB-rbcL spacer. Thick lines indicate branches with > 70% bootstrap support, and double-thick lines indicate branches with >85% support; the light gray lines in the ITS genealogy reflect a lack of confidence in the reconstruction due to ambiguous alignment. The apr and papr genealogies were rooted with F. hygrometrica; the remaining genealogies are midpoint rooted. Light gray shaded areas, labeled S and P, correspond to the sphaericum and pyriforme lineages, respectively, with bootstrap support indicated on the branch between the two lineages (see the text; absent in the ITS genealogy); dark gray shaded taxa show the three origins of P. patens; white regions bounded by gray lines circle the putative hybrid species, P. eurystomum and P. collenchymatum. Species localities correspond with Table 1, and details regarding the sequence length and phylogenetic reconstructions are given in Table 2.

Figure 2.

Figure 2.

Maximum parsimony genealogies of members of the PhyscomitrellaPhyscomitrium complex from (A) adk; (B) apr; (C) papr; (D) ho; (E) ITS; and (F) atpB-rbcL spacer. Thick lines indicate branches with > 70% bootstrap support, and double-thick lines indicate branches with >85% support; the light gray lines in the ITS genealogy reflect a lack of confidence in the reconstruction due to ambiguous alignment. The apr and papr genealogies were rooted with F. hygrometrica; the remaining genealogies are midpoint rooted. Light gray shaded areas, labeled S and P, correspond to the sphaericum and pyriforme lineages, respectively, with bootstrap support indicated on the branch between the two lineages (see the text; absent in the ITS genealogy); dark gray shaded taxa show the three origins of P. patens; white regions bounded by gray lines circle the putative hybrid species, P. eurystomum and P. collenchymatum. Species localities correspond with Table 1, and details regarding the sequence length and phylogenetic reconstructions are given in Table 2.

Table 2.  Tree lengths and consistency indices for parsimony reconstructions.
LocusAligned lengthPolymorphic sitesTree lengthInformative sitesIndels1Homoplasy indexMPTs
  1. 1Only two-state (0/1) indels where the rarer type was found in more than one isolate were scored.

  2. 2F. hygrometrica was not included in the analysis of tree length statistics.

apr2874118144 8310.193  2
papr978108123 7610.143  4
adk874126167 8340.313  8
ho714 73 83 5530.049 45
atpB-rbcL2572 19 43 1360.167 84

In all six loci, three distinct lineages of P. patens were evident (Fig. 2). To test whether these represented multiple statistically distinguishable origins of this species, we compared the most parsimonious trees for each locus separately to trees constrained to Physcomitrella monophyly using the Shimodaira–Hasegawa test. The constrained topologies were significantly longer for all loci except the atpB-rbcL spacer (Table 3). In the unconstrained trees, the European and North American P. patens, including an accession from Del Valle Lake, California, of the subspecies californica, were strongly supported as sister to the three P. sphaericum isolates. The subsp. californica from Saitama, Japan, was sister to the isolate of subsp. readeri from Australia (Melton) in all genealogies. These Austro-Asian P. patens isolates were either nested within P. pyriforme (adk, papr, Fig. 2A,C) or sister to the rest of the PhyscomitrellaPhyscomitrium species complex (apr, ho, atpB-rbcL, Fig. 2B,D,E). The subsp. magdalenae from Mt. Bisoke, Rwanda, was sister to an isolate of P. pyriforme from Madeira in all genealogies.

Table 3.  Likelihood difference between the optimal tree for each locus and trees constrained to a monophyletic Physcomitrella using the Shimodaira–Hasegawa test.

In genealogies from the four nuclear protein-coding loci, P. pyriforme was paraphyletic with respect to the P. patens isolate from Rwanda, and those from Japan and Australia (Fig. 2A–D). For the purpose of grouping the accessions for population genetic analyses, we refer to isolates of P. pyriforme, P. patens ssp. magdalenae, P. patens ssp. readeri, and P. patens ssp. californica (Japan) collectively as the pyriforme lineage (P in Fig. 2). The isolates of P. sphaericum and the European and North American isolates of P. patens were well resolved as distinct from the pyriforme lineage in all loci. We refer to these isolates collectively as the sphaericum lineage (S in Fig. 2). The isolates of P. eurystomum grouped with both the pyriforme and sphaericum lineages, whereas the P. collenchymatum grouped with one or the other, depending upon the locus. These accessions were therefore not included in demographic analyses of the sphaericum and pyriforme lineages.


Among isolated lineages, we expect that all single-locus genealogies will represent approximations of a single underlying evolutionary history. The genealogical placement of P. eurystomum represents a clear departure from this expectation. In genealogies of each of the nuclear loci, alleles from P. eurystomum were closely related to both P. sphaericum and P. pyriforme. The placement of particular isolates of P. eurystomum varied among the loci we sampled; in the apr genealogy, the isolates from Neustadt and Neukirch were nested within the pyriforme lineage (Fig. 2B), whereas in the papr genealogy these isolates were closely related to the sphaericum lineage (Fig. 2D). In the ITS genealogy, the P. eurystomum isolates were all closely related and distinct from both the sphaericum and pyriforme lineages (Fig. 2E), whereas in the atpB-rbcL genealogy, all P. eurystomum isolates were very closely related to chloroplast haplotypes from P. sphaericum (Fig. 2F).

In the ho genealogy (Fig. 2C), four isolates of P. pyriforme (Oevergran, Bischofswerda, Nordhausen, and Haardtrand) each contained two alleles, in all cases one identical to the Madeira allele and one identical to the Gera allele. This finding suggests that two divergent lineages of P. pyriforme hybridized and the descendants retained both copies of the ho gene or the surrounding genomic region. We refer to these as the “hybrid”P. pyriforme isolates. Because we did not find multiple alleles at any other locus in these isolates, we suspect that this is a locus-specific phenomenon, rather than an allopolyploidy event. Nevertheless, we removed the hybrid isolates from population genetic analyses involving the ho locus because the presence of multiple alleles complicates inferences regarding effective population size and recombination.

To assess the significance of the genealogical conflict among loci for the remaining isolates, we used the partition homogeneity test. The ITS data were excluded from this analysis due to alignment ambiguity. The test indicated that the five regions have significantly different phylogenetic signals (P < 0.001; Table 4). To test whether particular taxa were a significant source of genealogical conflict, we iteratively pruned individuals from the analysis and performed the test again. When the P. eurystomum sequences were removed, the remaining data continued to show significant discordance across the five loci (P= 0.001; Table 4). This finding indicated that additional taxa were strongly supported in distinct places in different genealogies. P. collenchymatum in particular was strongly supported as sister to P. sphaericum in the adk, apr, and atpB-rbcL genealogies (Fig. 2A,B,E), but sister to P. pyriforme in the the papr, ho, and ITS genealogies (Fig. 2C,D,F). Accordingly, a nonsignificant partition homogeneity test (i.e., no significant conflict) was achieved when P. collenchymatum, P. eurystomum, and the hybrid P. pyriforme isolates were pruned from the alignment (Table 4).

Table 4.  Partition homogeneity tests among five loci1 (apr, papr, adk, ho, atpB-rbcL) for pruned subsets of the Physcomitrella–Physcomitrium complex.
Included dataP
  1. 1ITS not included due to alignment uncertainty.

  2. 2P. pyriforme isolates from Bischofswerda, Hardtrand, Nordhausen, and Oevergran were excluded because each single isolate contained two ho alleles, suggesting hybridity.

All taxa0.001
P. eurystomum excluded0.001
P. eurystomum and P. collenchymatum isolates excluded0.001
P. eurystomum, P. collenchymatum, and “hybrid”P. pyriforme2 isolates excluded0.906


The P. patens isolates from Europe and North America, the Western Pacific, and Africa have different origins and were analyzed separately. The four Euro-American isolates of P. patens contained few segregating sites at all six loci (Table 5). Similarly, the three isolates of P. sphaericum contained similar or identical alleles, as did the accessions of P. patens from Japan and Australia, and a subset of the P. pyriforme isolates (Bischofswerda, Gera, Haardtrand, Oevergran) at the atpB-rbcL, ITS, adk, and apr loci. P. pyriforme showed greater nucleotide diversity than the Euro-American P. patens or P. sphaericum, but P. eurystomum consistently had the largest number of segregating sites, because it contained sphaericum and pyriforme-type alleles. We found neither a deviation from a neutral-equilibrium mutation frequency spectrum nor an elevation in nonsynonymous substitutions, and therefore no evidence of natural selection acting on any locus (data not shown).

Table 5.  Nucleotide diversity in members of the Physcomitrella–Physcomitrium complex.
  1. 1ho values for P. pyriforme not calculated due to the presence of multiple copies.

  2. 2ITS total value not calculated due to alignment uncertainty.

P. patens (Europe and North America)0.00130.00970.003100.00450.0101
P. sphaericum0.00230.0022000.00220.0029
P. eurystomum0.02650.02440.03880.02950.00650.0022
P. pyriforme0.01740.01410.0099 0.06120.0061

We estimated the recombination parameter R and the minimum number of recombination events for each locus separately, in the sphaericum lineage, among the P. pyriforme isolates (except at the ho locus), in the pyriforme lineage, and in the entire sample (Table 6). We also estimated the minimum number of recombination events (or assortment events, as the nuclear loci are on separate P. patens genomic scaffolds and thus on potentially different chromosomes) for the entire concatenated matrix. In the isolates that showed evidence of a genealogical conflict, we scanned by eye for recombination breakpoints where multiple polymorphic sites support one genealogical placement in one part of a sequence, but a distinct placement in a second portion of the sequence. The alignment for the apr gene showed the clearest evidence of a recombination breakpoint between a sphaericum-type sequence and a pyriforme-type sequence in one of the P. eurystomum alleles. Indeed, a phylogenetic analysis using the 5′-side of the breakpoint resulted in this isolate being strongly supported as sister to P. sphaericum, whereas the sequence 3′ of the breakpoint strongly supported a placement within P. pyriforme (data not shown).

Table 6.  Inferred intra- and intergenic recombination among nuclear loci in the sphaericum and pyriforme lineages. The top number is inferred recombination rate per site (Hudson 1987), and the bottom number is the minimum number of recombination events (Hudson and Kaplan 1985).
  1. 1P. sphaericum, P. patens subsp. patens, and P. patens var. californica (USA).

  2. 2P. pyriforme, P. patens var. magdalenae, P. patens subsp. californica (Japan), and P. patens var. readeri.

  lineage10200 2
P. pyriforme0.00120.00740.09940.0001 
 0220 7
All species0.05170.04280.03510.0449 

The IMa analyses indicated that the current Ne for the pyriforme and sphaericum lineages are approximately 4.92 million (HPD90 = 3.34 × 106–7 × 106) and 1.69 million (HPD = 1.05 × 106–2.73 × 106), respectively, with an ancestral population size nearly 20 times smaller (Na= 84,000, HPD90 = 19,100–1.39 × 106; Fig. 3a). The MLE of migration from sphaericum to pyriforme (m= 0.084, HPD90 = 0–0.474) was larger than that from pyriforme to sphaericum (m= 0.0009, HPD90 = 0–0.375; Fig. 3b). We estimate the divergence between the sphaericum and pyriforme lineages found ∼11.4 million years ago (MYA), with a 90% highest probability density (HPD90) including 8.37–12.8 MYA (Fig. 3c). Using a higher mutation rate would obviously decrease these divergence times and effective population sizes. The best-fit joint parameter estimates generated by IMa in L-mode matched the values above, except that the migration rate from pyriforme to sphaericum was fixed at 0 (Table 7). We could not reject additional models (at the P= 0.05 level) with symmetrical but either low migration rates or no migration between the sphaericum and pyriforme lineages.

Figure 3.

Marginal posterior probability distributions for IM model parameters between sphaericum and pyriforme lineages.

Table 7.  Testing nested demographic models for the pyriforme and sphaericum lineages.
  1. 1q1, pyriforme effective population size; q2, sphaericum Ne; qa, ancestral Ne; m1, migration rate from pyriforme into sphaericum; m2, migration rate from sphaericum into pyriforme; t, divergence time between pyriforme and sphaericum; P, model probability.

  2. 2Because of a parameter fixed at the boundary of the parameter space, the expected distribution is a mixture when the null model is true; 2LLR should be asymptotically distributed as a random variable that takes the value 0 with probability 0.5 and takes on a value from a χ2 distribution with probability 0.5 (Hey and Nielsen 2007).

  3. 3Models under which the probability of achieving the test statistic by chance under the null model is >0.05 are shown in bold.

q1, q2, qa, m1, m210.7358  2.7134
q1, q2, qa, m1 = m210.4581  1.457412.512
q1, q2, qa, m1, m2 = 010.6624  2.7113120.0043
q1, q2, qa, m1 = 0, m210.048  0.641912 4.143
q1, q2, qa, m1 = m2 = 010.05  0.6419224.143
q1 = q2, qa, m1, m210.3944 −2.7979111.0226
q1 = q2 = qa, m1, m210.8539 −7.3736220.174
q1 = q2, qa, m1 = m210.601 −3.2943212.0154
q1 = q2, qa, m1 = m2 = 010.3637 −4.67873214.7842
q1 = q2 = qa, m1 = m2 9.2436 −8.3774322.1815
q1 = q2 = qa, m1 = m2 = 0 8.163−10.98794227.4027
q1 = qa, q2, m1, m2 9.4584 −0.09851 5.6238
q1 = qa, q2, m1 = m2 9.1366 −1.31542 8.0576
q1 = qa, q2, m1 = m2 = 0 9.1271 −2.187332 9.8014
q1, q2 = qa, m1, m210.7344 −3.6297112.6863
q1, q2 = qa, m1 = m2 8.5912 −5.1609215.7486
q1, q2 = qa, m1 = m2 = 0 8.3555 −6.95013219.327



Taxonomists have long noted the gametophytic similarity between the genera Physcomitrella and Physcomitrium, although the two genera are readily distinguished based on the morphology of the diploid sporophyte (Crum and Anderson 1981; Schwartz 1994, 1997). Our results, like those from earlier molecular phylogenetic studies based on chloroplast DNA sequences (Goffinet and Cox 2000; Goffinet et al. 2007; Werner et al. 2007) support this inference. Strikingly, P. patens has arisen at least three times from distinct ancestors within the genus Physcomitrium, a result strongly supported by all nuclear loci (P < 0.001, Table 3; Fig. 4). Importantly, there was no topological conflict or evidence of recombination in the alignment that included the three lineages of P. patens and their immediate sister isolates (i.e., P. sphaericum, the Madeira and Gera isolates of P. pyriforme; Fig. 1, Table 4) suggesting that this arrangement is not an artifact of hybridization. The fact that members of the three P. patens lineages have nonoverlapping distributions suggests that allopatry may play a role in the persistence of these taxa.

Figure 4.

A model of the PhyscomitrellaPhyscomitium complex species tree. The width of the branch corresponds to the relative effective population size of the species (Table 5). The dotted and dashed lines indicate hybridization events leading to the formation of the putative hybrid species P. collenchymatum and P. eurystomum, respectively.

We should point out that our sampling of the genus Physcomitrium is sparse; although additional samples would not affect our inferences of multiple origins, other pygmy species in the Funariaceae, such as Physcomitrium immersum and Aphanorhegma serratum, may represent either independent cases of sporophytic reduction or diversification within P. patens-like lineages. Multiple origins of the P. patens morphology from Physcomitrium-like ancestors parallel cases of recurrent sporophytic reductions in other bryophyte families (Buck et al. 2000; Goffinet and Shaw 2002). The repeated evolution of this phenotype suggests that it may be adaptive under certain circumstances—potentially like the evolution of cleistogamy in flowering plants—but we cannot eliminate neutral explanations for the sporophytic reduction.

The extreme differences in sporophyte morphology between Physcomitrella and Physcomitrium make hybrids between these genera easy to detect, and indeed the study of hybridization in the Funariaceae has a long history relative to that in other mosses (Natcheva and Cronberg 2004). Individuals with morphologically intermediate, presumably F1, sporophytes found on otherwise pure-species gametophytes (i.e., maternal gametophytes bearing sporophytes produced with heterospecific sperm) have been reported between P. patens and P. sphaericum (Loeske 1929; Pettet 1964), P. patens and P. pyriforme (Andrews 1918, 1942; Loeske 1929; Pettet 1964; Tan 1978; Crum and Anderson 1981), and A. serratum and P. pyriforme (Britton 1895; Crum and Anderson 1981). However, we found no genealogical evidence of hybridization between species with extreme differences in sporophyte morphology. Nor do we have evidence that the species with reduced sporophytes arose following a hybridization event, as the phylogenetic placement of all three lineages of P. patens was consistent in all nuclear genealogies. We have repeated many intergeneric crosses produced by von Wettstein (1924, 1928, 1932), as well as generated new crosses between European and Asian isolates of P. patens, and confirmed that in all cases < 1:1000 spores developed beyond a few cells (S. F. McDaniel and P.-F. Perroud, unpubl. data). The nearly complete lack of recombinant progeny precludes us knowing whether the incompatibility involves nuclear–cytoplasmic interactions, as suggested by von Wettstein (1932), and whether the mutations responsible for the differences in sporophyte development have pleiotropic effects on other aspects of the reproductive biology of P. patens.


The consistent genealogical placement of P. eurystomum alleles between the well-differentiated sphaericum and pyriforme lineages (Fig. 2), and the concomitant increase in nucleotide diversity in the species relative to other sampled taxa (Table 5), could result from taxonomic error, incomplete lineage sorting, or introgressive hybridization. We do not, however, believe that the P. eurystomum individuals have been misidentified. First, all three P. eurystomum isolates had multilocus genotypes with alleles from both the sphaericum and pyriforme lineages, which could not be explained by misidentification. Second, the ITS sequences contained obvious species–specific patterns that nearly perfectly matched the traditional taxonomic boundaries in the PhyscomitrellaPhyscomitrium complex, including the genetic similarity of the three accessions of P. eurystomum (but with the exception of the multiple origins of P. patens, Fig. 2E).

Our results also suggest that the close relationships between P. eurystomum and both the pyriforme and sphaericum lineages are unlikely to result from incomplete lineage sorting. Incomplete lineage sorting is most likely to cause genealogical conflict among loci where population sizes are large, divergence times are recent, and therefore coalescence times may long precede the speciation event. The highest posterior density intervals for the ancestral population size in the PhyscomitriumPhyscomitrella complex was 20 times smaller than the current estimated population sizes (HPD90 = 19,100–1.39 × 106; Fig. 3A), and those for the MLEs of divergence did not include zero (HPD90 = 8.37–12.8 MYA; Fig. 3C). Thus, the most likely demographic parameters (Table 7) do not favor the retention of ancestral polymorphism (although we could not reject a model with no gene flow, which would obviously be necessary in the case of hybridization). These analyses, however, are based on a very simple demographic model that may not capture the actual demographic history of the PhyscomitrellaPhyscomtrium complex, in part because we could not include the putative hybrid species. We found some preliminary biological confirmation of these patterns by studying the fertility of interspecies hybrids. The MLEs for migration between the sphaericum and pyriforme lineages were asymmetrical and most evident in the chloroplast atpB–rbcL gene (Fig. 3B, see also Fig. 2F); all P. eurystomum alleles were closely related and identical or very similar to sphaericum alleles. This pattern is consistent with the asymmetry in crossing success between P. patens and P. pyriforme (von Wettstein 1924, 1928) where crosses were only successful with P. patens as the maternal parent. The close relationships among P. sphaericum, P. eurystomum, and P. collenchymatum at the chloroplast atpB-rbcL locus suggest cytoplasmic introgression, a common phenomenon in angiosperms (Tsitrone et al. 2003) and potentially mosses (Shaw and Goffinet 2000; Shaw et al. 2005b; Natcheva and Cronberg 2007).

The consistent presence in P. eurystomum of alleles at the four nuclear loci from both the sphaericum lineage and pyriforme lineage (Fig. 2) suggests that the species may represent not a case of adaptive introgression but rather hybrid speciation. Natcheva and Cronberg (2004) cite several well-documented cases of allopolyploid hybrid species in mosses, and indeed the entire PhyscomitrellaPhyscomitrium species complex may have originated from an Eocene allopolyploidy event (Rensing et al. 2007). The few reported chromosome counts from P. eurystomum are variable, including n= 9, 26, 52, and 54 (Fritsch 1991; Kapila and Kumar 1997). Polyploid series are frequent in mosses—chromosome numbers from n= 9 to n= 72 have been reported for P. pyriforme—and preliminary flow-cytometric data suggest that the genome size of P. eurystomum is about twice that of P. patens or P. sphaericum (S. A. Rensing, unpubl. data). If P. eurystomum were an allopolyploid, we would presumably be able to amplify two parental homeologous alleles of the nuclear loci in this species. However, we found no evidence of multiple paralogs (i.e., multiple peaks in the direct sequencing reactions, or distinct sequences in the cloned PCR products as we found for the four “hybrid” isolates of P. pyriforme at the ho locus), suggesting either that the isolates we sampled have undergone a complex pattern of gene loss and divergence (Ku et al. 2000; Vanderpoorten et al. 2004; Lukens et al. 2006; Town et al. 2006) or that P. eurystomum represents a homoploid hybrid species with a large genome (Ungerer et al. 2006).

Ploidy issues aside, the fertility of natural and experimental hybrids between P. eurystomum and either P. patens, P. pyriforme, and even the more distant F. hygrometrica, was generally similar to that found in other crosses (von Wettstein 1924, 1928). We have found that the viability of recombinant spores from F1 hybrids between P. patens and P. eurystomum was similar to that of spores from P. patens and P. sphaericum crosses, where the genome size is similar between the two species—many recombinants showed signs of early viability, but none survived past the two-cell stage (S. F. McDaniel and P.-F. Perroud, unpubl. data). This similarity in viability suggests that karyotypic differences are not the only causes of hybrid sterility or inviability in this group.

In addition to the topological conflict caused by P. eurystomum, the phylogenetic position of the North American species, P. collenchymatum, also varied among loci (Fig. 2, Table 4). In the original description of the species, Gier (1955) suggested that P. collenchymatum may be a hybrid between P. pyriforme and A. serratum. We have no experimental crossing data involving P. collenchymatum, and our sample size for this species is limited to a single locality, so this inference requires further validation. We are now actively engaged in genetic analyses to more rigorously test whether the genomic structure of the putative hybrid species is similar to recombinants from either P. patens or P. sphaericum×P. pyriforme F1 hybrids in karyotype and allelic composition.


Our data suggest that the mating system in the PhyscomitrellaPhyscomitrium complex is mixed. Members of the Funariaceae have long been known to be self-fertile under laboratory conditions (von Wettstein 1924). The limited genetic variation within species (e.g., P. sphaericum, or the Euro-American isolates of P. patens) and the small inferred ancestral population size together could result from frequent selfing, consistent with expectations based on the proximity of the gametangia and with previous allozyme studies (Eppley et al. 2006). However, among the accessions of the pyriforme lineage, we find evidence of inter- and intragenic recombination. In P. pyriforme, recombination is evident in the fixed heterozygosity at the ho locus and the distinct genealogical patterns among the other loci (Table 6). In P. eurystomum, the Neustadt isolate in the apr genealogy appears to be a natural recombinant between a sphaericum-type and a pyriforme-type sequence, with the 5′-end of the sequence sphaericum-like, and the 3′-end pyriforme-like. In the aggregate, the available data indicate that outcrossing occurs in self-fertile mosses, even among partially reproductively isolated relatives (Stenoien and Sastad 2001; Eppley et al. 2006).

Theory suggests that both population divergence and hybrid speciation may occur rapidly in self-fertile lineages (McCarthy et al. 1995; Barraclough et al. 2003; Coyne and Orr 2004). The evidence for more rapid speciation in self-fertile mosses than outcrossing mosses in general, however, is inconclusive. Nucleotide variation at the adk locus within the outcrossing species Ceratodon purpureus (McDaniel and Shaw 2005) among partially reproductively isolated populations (McDaniel et al. 2007, 2008) was nearly equal to that among the six species we sampled in the PhyscomitrellaPhyscomitrium complex (θ= 0.030 and 0.033, respectively). The difference in number of species between these genera suggests that the diversification rate, relative to the mutation rate, is higher in the self-fertile lineage in this one comparison. Consistent with this inference, experimental crossing studies showed that the hermaphroditic nematode C. elegans exhibits outbreeding depression at smaller genetic distances than its outcrossing congener, C. remanei (Dolgin et al. 2007). However, the situation is less clear in Sphagnum spp., where the mating system was not obviously correlated with patterns of genetic variation or reproductive isolation (Shaw and Cox 2005; Natcheva and Cronberg 2007), or Polytrichum spp., where F1 postzygotic barriers isolate the closely related and obligate outbreeders P. commune and P. uliginosum (van der Velde and Bijlsma 2004).

The differences between P. pyriforme and P. patens in nucleotide diversity and evidence of past recombination (Tables 5 and 6) also suggest that factors other than the arrangement of the gametangia may be critical for structuring variation within and among populations in the PhyscomitrellaPhyscomitrium complex; in fact, our estimates for P. pyriforme are almost certainly too low, given our limited sampling of this species. Species in the complex have evolved different temperature preferences and reproductive phenologies (Nakosteen and Hughes 1978; Furness and Grime 1982). Adaptation to these and other ecological factors, such as humidity or substrate preference, may be critical for limiting gene flow among diverging lineages, because the causative alleles participate in genetic incompatibilities or promote differential habitat specificity. Indeed, the putative hybrid P. eurystomum is rare and appears to be endemic to a marginal habitat which is different from that of its inferred parental species (Hill et al. 2006), similar to hybrid species of sunflowers and butterflies (Rieseberg et al. 2003; Gompert et al. 2006) and consistent with theoretical predictions (Buerkle et al. 2000; Gross and Rieseberg 2005). We anticipate that uniting the molecular techniques now well established in P. patens (Quatrano et al. 2007) with classical genetics (von Wettstein 1924, 1928, 1932) will provide important insights into the key genetic, ecological, and demographic factors that generate new species.

Associate Editor: J. Vamosi


Numerous collectors from the bryological community kindly provided live material or spores to the International Moss Stock Center. J.-P. Frahm, P.-F. Perroud, B. L. Gross, J. Strasburg, K. M. Olsen, and three anonymous reviewers provided many thoughtful comments that clarified and focused this manuscript. This work was funded by the generous support of the German Research Foundation (Re 837/10-2) and the German Federal Ministry of Education and Research (0313921, FRISYS) to R. Reski and S. A. Rensing, and a Pilot Sequencing Grant from the Washington University Genome Sequencing Center (to S. F. McDaniel and R. S. Quatrano). S. F. McDaniel was supported by an NIH–NRSA fellowship (F32 GM075606).