High frequency of character transformations is phylogenetically structured within the lichenized fungal family Graphidaceae (Ascomycota: Ostropales)

ABSTRACT Graphidaceae is a large family of over 2000 predominantly tropical, lichenized fungal species encompassing a remarkable range of morphological and chemical diversity. The majority of species belongs in subfamily Graphidoideae, which also exhibits the greatest amount of variation. Various phenotype characters have traditionally been used for classification at the genus and species levels, but their correlations with phylogenetic clades are poorly known. Using a multilocus approach, we reconstructed a phylogeny for 224 taxa, representing all main genera within subfamily Graphidoideae, and employed ancestral character reconstruction and character transformation analyses to understand the evolution of morphological, anatomical and chemical characters within this group. In addition, we examined the changes of habitat and photobiont types over the phylogeny. For this purpose, we focused on 10 characters, including thallus and ascoma features and chemistry. Since previous studies have shown that results may differ depending on the reconstruction method used, both Maximum-parsimony and Maximum-likelihood approaches were employed and multistate coding of characters was used. We reconstructed the ancestral states for 64 well-supported major clades in the family and found support for the ancestor of Graphidoideae being a tropical species with a trentepohlioid photobiont, apothecioid, solitary ascomata lacking both a columella and lateral paraphyses, and having non-amyloid ascospores. The frequency of transformations of morphological and chemical characters over the phylogeny of Graphidaceae was computed, resulting in a high frequency of reversible transformations for some characters, such as secondary chemistry, whereas other characters, such as photobiont, hymenial persistence or ascoma aggregation, exhibited low frequency of transformations. However, we found that even in the character with the highest number of transformations, secondary chemistry, the shifts were highly structured phylogenetically, suggesting that the evolution of the character, rather than the character state itself, can be used to predict phylogenetic relationships with certain accuracy.

Graphidaceae, which has recently been expanded to include the previously separated families Asterothyriaceae, Gomphillaceae, Solorinellaceae and Thelotremataceae (Mangold et al., 2008b;Baloch et al., 2010;Rivas Plata & Lumbsch, 2011;Rivas Plata et al., 2012a, is one of the largest families of lichen-forming fungi with over 2000 accepted species (L€ ucking et al., 2013;Rivas Plata et al., 2013). The classification within this large fungal clade has dramatically changed over the last decade. The traditional circumscription of the family and genera was largely based on ascoma (rounded vs. lirellate or pseudostromatic) and ascospore types (septation and pigmentation) (M€ uller, 1887;Wirth & Hale, 1963Hale, 1974Hale, , 1978. The use of these characters was long perceived as artificial and replaced by a classification among thelotremoid taxa based on excipular structures (Salisbury, 1972a(Salisbury, , 1972b(Salisbury, , 1978Hale, 1980Hale, , 1981. However, major systematic revisions only started after seminal treatments on graphidoid and thelotremoid taxa respectively by the school of Klaus Kalb (Staiger, 2002;Frisch et al., 2006) with a more refined classification based on a combination of phenotypic characters. Molecular studies have further changed the classification with the identification of additional clades that are accommodated in new or resurrected genera and a revised concept at the family level (Staiger et al., 2006;Mangold et al., 2008aMangold et al., , 2008bNelsen et al., 2010;Rivas Plata et al., 2010a, 2013Berger et al., 2011;, 2013Rivas Plata & Lumbsch, 2011;C aceres et al., 2012;Parnmen et al., 2012aParnmen et al., , 2013. Within Graphidaceae, up to four subfamilies are currently accepted (Hodkinson, 2012;Rivas Plata et al., 2012a;L€ ucking et al., 2013). Subfamily Graphidoideae constitutes by far the largest clade of this family of crustose, primarily tropical lichen-forming fungi (Rivas Plata et al., 2012a). The species currently placed in this subfamily represent the core of the former families Graphidaceae and Thelotremataceae. This apparent incongruence of traditional, morphology-based classifications and molecular phylogenies suggest that these fungi are extremely variable with regard to phenotype characters, with a high probability of characters evolving in parallel in unrelated clades, as has been shown for this family and for various other lineages in the Ascomycota (Blanco et al., 2004(Blanco et al., , 2006Crespo et al., 2007;Tehler & Irestedt, 2007;Mugambi & Huhndorf, 2009;Lumbsch et al., 2010b;Parnmen et al., 2010;Muggia et al., 2011;Rivas Plata et al., 2011;Rivas Plata & Lumbsch, 2011).
With the availability of molecular data, we are now able to study the evolution of phenotypic characters previously used in the classification of Graphidaceae. While most species in the family are tropical crustose lichens, some species also occur in subtropical habitats, such as Redonographa (L€ ucking et al., 2013) or have their distribution centre in non-tropical areas, such as Diploschistes (Lumbsch, 1989). Two main types of photobionts are found in the family, chlorococcoid and trentepohlioid, correlating with ecological features of the involved lichens (Friedl & G€ artner, 1988;Nelsen et al., 2011). While numerous species have corticated thalli (Hale, 1981;Staiger, 2002;Frisch et al., 2006), others lack a cortex (Lumbsch, 1989). A cortex is a distinct layer of fungal hyphae covering the upper and/or lower side of the thallus. The ascomata can be either roundish (apothecioid) or elongate (lirellate), but there is no evidence regarding the evolutionary function of ascoma shape. A few lineages form mazaedioid ascomata which consist of spore masses that accumulate on top of the ascomata. These are highly adapted to wind dispersal of the ascospores , while the majority of species in the family are characterized by ascomata with persistent hymenia. A sterile tissue within the hymenium, the so-called columella, which is often partially or totally carbonized, occurs in many tropical species; it is assumed to protect the hymenium against fungivores (Rivas Plata & Lumbsch, 2011). Sterile hyphae that grow from the margin into the central cavity of the ascoma, so-called lateral paraphyses (Henssen, 1995) or periphysoids (Hale, 1981) occur in several lineages, but their function is unknown. Ascospores in the family show different reaction to iodine, being either amyloid, hemiamyloid or non-amyloid (Baral, 1987;Rivas Plata & Lumbsch, 2011), depending on the presence and absence and chemical nature of internal wall substances.
We are now able to address questions on the possible functions of these characters in terms of ecology and adaptations of species to particular (micro-) habitats and niches. Generally, fast-evolving characters can be assumed to be ecologically modified, whereas highly conserved characters have low levels of correlation with ecological parameters, unless the clade in question is also ecologically uniform (Felsenstein, 1985;Harvey & Purvis, 1991;Coddington, 1994;Ackerly, 2003;Kraichak, 2012). In order to understand the patterns of the character evolution in this group of lichenized fungi, we assembled a dataset of four loci including 224 taxa representing all major clades of subfamily Graphidoideae, Fissurinoideae and Redonographoideae and outgroups for phylogenetic analysis. We then performed ancestral character state reconstructions of seven representative morphological and anatomical characters, secondary chemistry, the type of photobiont and the vegetation type zone. The main objectives of this study were: (1) to characterize the hypothetical ancestor of the subfamily Graphidoideae, to better understand the evolution of phenotypic and ecological characters in the core group of the family Graphidaceae; (2) to identify the number and phylogenetic signal of character state transformations of these characters along the reconstructed phylogeny; and (3) to understand the impact of ecology on character transformations.

Materials and methods
Taxon sampling and molecular methods The taxon sampling included the major clades of subfamilies Graphidioideae, Redonographoideae, and Fissurinoideae in Graphidaceae, plus five taxa of the genus Gyalecta as outgroup, based on previous molecular studies (Lumbsch et al., 2010a;Parnmen et al., 2010). Two hundred and twenty-four species were included in the analyses (Table 1). We selected taxa for this study to represent the morphological and chemical diversity in the group and to include all major genera, except for the taxa in subfamily Gomphillioideae, from which a limited number of materials and DNA sequences were available for the current study. We included only species for which we obtained at least two of the four loci studied: nuclear LSU rDNA, mitochondrial SSU rDNA, and the protein-coding RPB1 and RPB2 genes. New sequences were generated for this study using the Sigma REDExtract-N-Amp Plant PCR Kit (St. Louis, Missouri, USA) for DNA isolation following the manufacturer's instructions, except that 40 mL of extraction buffer and 40 mL dilution buffer were used. DNA dilutions (5Â) were used in PCR reactions of the genes coding for the nuLSU, mtSSU, RPB1 and RPB2, respectively. Primers and PCR amplification condition were the same as described previously (Parnmen et al., 2012a(Parnmen et al., , 2012bSchmitt et al., 2012;Rivas Plata et al., 2013). One hundred and fifteen new sequences were generated for this study (21 mtSSU, 16 nuLSU, 54 RPB1 and 24 RPB2).

Sequences alignments and phylogenetic analyses
Alignments were done in Geneious Pro 5.5.2 (Drummond et al., 2012). Ambiguously aligned portions were removed manually. The single-locus and concatenated alignments were analysed by maximum likelihood (ML) and a Bayesian approach (B/MCMC). To test for potential conflict, ML bootstrap analyses (with 2000 pseudoreplicates) were performed on the individual datasets, and resulting singletree trees were examined for conflict, i.e. incongruences with at least75% bootstrap support (Lutzoni et al., 2004).
The ML analysis of the concatenated alignment was performed with the program RAxML-HPC2 (version 7.3.1) on XSEDE (Stamatakis, 2006) using the default rapid hill-climbing algorithm. The model of nucleotide substitution chosen was GTRGAMMA, according to the results from the model selection by jModelTest2 (Guindon & Gascuel, 2003;Darriba et al., 2012). The dataset was partitioned into eight parts (mtSSU, nuLSU and each codon position of RPB1 and RPB2), and each gene partition was treated as independent. Introns in RPB1 and RPB2 sequences were removed from the analysis. Bootstrap estimates were carried out using 2000 pseudoreplicates (Stamatakis et al., 2008).
The B/MCMC analysis was conducted using MrBayes 3.1.2 (Huelsenbeck & Ronquist, 2001), with the same substitution model as in the ML analysis. Two parallel runs with 10 000 000 generations each, starting with a random tree and employing four simultaneous chains, was executed. No molecular clock was assumed. Heating of chains was set to 0.2. Posterior probabilities were approximated by sampling trees using a variant of Markov Chain Monte Carlo (MCMC) method. To avoid autocorrelation, only every 1000th tree was sampled. The first 4000 trees were discarded as burn in. We used AWTY (Nylander et al., 2007) to compare splits frequencies in the different runs and to plot cumulative split frequencies to ensure that stationarity was reached. A majority-rule consensus tree with average branch lengths was calculated from the remaining 6000 sampled trees using the sumt option of MrBayes. Posterior probabilities were obtained for each clade. Clades with bootstrap support above 70% under ML and Bayesian posterior probabilities above 0.95 were considered as strongly supported. Phylogenetic trees were visualized using the program TreeView (Page, 1996).

Ancestral character state reconstruction
Characters for the analysis including habitat, photobiont, cortex, six ascomatal characters, and secondary chemistry were analysed using multistate character coding with characters treated as unordered (Table 2 and Table S1, see online supplemental material, which is available from the article's Taylor & Francis Online page at http://dx.doi. org/10.1080/14772000.2014.905506). The coding was done based on our own observations of the material and recently published studies , 2013Mangold et al., 2009;Lumbsch et al., 2010a;Rivas Plata et al., 2010b, 2013Berger et al., 2011;Rivas Plata & Lumbsch, 2011;Parnmen et al., 2012aParnmen et al., , 2013Sipman et al., 2012;). For chemistry, instead of coding each substance separately, we coded depsidones according to the groups at positions 1 and 6, which might be either carboxyl groups (COOH), hydroxyl group (OH) or methyl group (ME). This was done to emphasize chemical relationships between substances. The three types of depsidones distinguished here include stictic acid for (1 0 -6 0 -OH) depsidones, protocetraric acid for (1 0 -COOH-6-ME) depsidones, and psoromic acid for (6 0 -COOH) depsidones.
Maximum likelihood (ML) reconstructions were carried out on each individual tree using an unrestricted  (2-parameter) model of character evolution. In order to include topological uncertainty into the ancestral state reconstruction, we used the 'Trace character over trees' method of Mesquite (Maddison & Maddison, 2011). One thousand trees were randomly sampled from the postburning of the Bayesian sampling (described above) of the concatenated dataset using the program RT (http://www.lutzonilab.net/downloads/), and Mesquite displayed a summary for the probability for each node and each character, indicating the probability for the different states, and also taking into account ambiguous reconstructions and the percentage of Bayesian trees in which the given node was present (Table S2, see supplemental material online). In contrast to ML, maximum parsimony (MP) does not take into account branch lengths when reconstructing ancestral states. The reconstructions were performed over the same 1000 randomly sampled trees as in the ML analysis. We used Mesquite 2.75 (Maddison & Maddison, 2011) to carry out both ML and MP reconstructions for the character datasets.   Table 2. Character states of each trait analyzed in the current study and the average number of transformations in discrete morphological and chemical characters across a posterior sample of trees of lichenized fungi Graphidaceae under maximum parsimony with 95% equal-tail credible interval in parentheses.

Character state transformations
We used MP to estimate the number of character state transformations over the tree using Mesquite (Maddison & Maddison, 2011). Parsimony potentially underestimates the true amount of changes in a character over a tree (Huelsenbeck & Lander, 2003) but it gives the minimum amount of changes. To test for phylogenetic signals of the characters studied, we used two different approaches. First, we employed a randomization test (Archie, 1989) in Mesquite to calculate the number of parsimony steps from 999 randomizations of character states on the optimal ML tree. The mean number of parsimony steps and standard deviation were calculated. Then the numbers from the randomization and the observed number of parsimony steps were used to calculate the probability that the observed number of parsimony steps is higher or equal than expected from the randomization. Second, we used the fit-Discrete program in the R package 'geiger' (Yang, 2006) to calculate Pagel's λ value (Pagel, 1999). For this, the likelihood values of the tree with the actual data were calculated and a tree transformed to have no phylogenetic signal (λ ¼ 0). High phylogenetic signal of a character is indicated by λ ¼ 1. Significant departure from lack of phylogenetic signal was calculated using a likelihood ratio test.

Phylogenetic analysis
The aligned 4-gene matrix contained 3274 unambiguously aligned nucleotide positions (945 in nuLSU, 836 in mtSSU, 607 in RPB1 and 886 in RPB2), with a total of 1009 constant characters. Topologies of single-locus analyses did not show conflict and hence combined analyses were performed. Since the topologies of the ML and B/ MCMC analyses did not show any supported conflict, only the phylogram obtained from the ML analysis is shown with branches in bold that received strong support in Bayesian analyses (i.e. PP above 0.95 and ML bootstrap values equal or above 70% indicated at branches) (Figs S1-S3, see supplemental material online). The topology obtained from the 4-gene dataset is similar to previously published phylogenies (L€ ucking et al., 2013;Rivas Plata et al., 2013). All subfamilies, tribes and genera as previously delimited were recovered (Fig. 1).

Ancestral character state reconstructions
The characters studied and their states are listed in Table 2.
Results of the ancestral character state reconstructions are listed in Tables S2-S4 (see supplemental material  In the ancestral character reconstruction for the character 'habitat' (Fig. 2), most ancestral nodes were reconstructed as being tropical, with the exception of Redonographa, which was reconstructed as subtropical, and the ancestral nodes of the genera Diploschistes, Schizotrema and Topeliopsis, which were reconstructed as being non-tropical. The reconstructions estimated a tropical habitat as ancestral for Graphidaceae and subfamily Graphidoideae. All reconstruction methods estimated a trentepohlioid photobiont as ancestral state for all nodes, with the exception of the base of the genus Diploschistes (Fig. S4, see supplemental material online), for which a chlorococcoid photobiont was reconstructed.
Regarding the presence or absence of cortex, the ancestral character state reconstruction for the basal nodes in Graphidaceae, with the exception of node 1 (corticated) did not yield conclusive results (Fig. 3). Within tribes Graphideae and Ocellularieae, most nodes were reconstructed as being corticate. In contrast, in tribe Thelotremateae, most nodes were reconstructed as being ecorticate, including the basal nodes of the genera Chapsa, Diploschistes, Leucodecton, Pseudochapsa and Thelotrema. The ancestral state for the genera Acanthotrema, Chroodiscus and the clade including Gintarasia and Pseudoramonia were reconstructed as corticate.
Absence of secondary metabolites was reconstructed for the basal nodes in subfamily Fissurinoideae and tribe Graphideae and the base of Acanthotrema and Topeliopsis (Fig. 4). Due to the variability of presence of secondary metabolites among species, there was a higher incidence of inconclusive results compared with other characters. However, presence of (1 0 -6 0 -OH) depsidones, e.g. stictic acid and related substances, was reconstructed for the base of Chroodiscus, Leucodecton, Phaeographis, Pseudochapsa, Redonographa, Wirthiotrema and the clades including Asteristion, Austrotrema, Nadvornikia and Mytriotrema peninsulae, as well as this group plus Wirthiotrema, corresponding to tribe Thelotremateae. The presence of (6 0 -COOH) depsidones, mainly psoromic acid, was reconstructed as ancestral trait for the base of the genera Compositrema, Myriotrema, Stegobolus and the genus Rhabdodiscus. For all examined nodes in Ocellularia s. lat. and s. str., presence of (1 0 -COOH-6 0 -ME) depsidones, such as protocetraric acid, was reconstructed as the ancestral state.
The ancestral character state for ascoma persistence for all nodes studied was reconstructed as non-mazaediate with persistent hymenium (Fig. S5, see supplemental material online), strongly suggesting that the mazaediate ascomata found in the genera Nadvornikia and Schistophoron originated independently from ascomata with persistent hymenium. Ascoma shape for the basal node of the family and the clade including the two subfamilies Graphidoideae and Redonographoideae was not reconstructed with certainty, whereas the base of Graphidoideae was reconstructed as apothecioid (Fig. 5). Within Graphidoideae, most nodes in tribes Ocellularieae and Thelotremateae were reconstructed as apothecioid, except for Acanthothecis. In contrast, most nodes within tribe Graphideae were reconstructed as lirellate, except for the Phaeographis lobata þ P. spondaica clade.  Figs. 5-7. Ancestral state reconstruction at the major nodes for (5) shape of ascoma, (6) the presence of lateral paraphyses and (7) ascospore amyloidity of lichenized fungi family Graphidaceae. Node colours represent the reconstructed states (see legend). Inconclusive reconstructions are indicated in grey circles.
Almost all reconstructions suggested solitary ascomata as ancestral state (Suppl. Fig. S6). Only for the base of Compositrema (tribe Ocelluarieae) the analysis reconstructed pseudostromatic ascomata as the ancestral character state. For all nodes outside tribe Ocellularieae, the analyses reconstructed absence of a columella as ancestral states (Fig. S7, see supplemental material online). Within Ocellularieae, the bases of the Macropyrenium and Stigmagora groups, Ocellularia s. str. and related taxa, and Stegobolus were reconstructed as having a columella.
The ancestral character state reconstructions for lateral paraphyses showed these to be absent at numerous basal nodes, as well as all nodes in Fissuroinoideae, Ocellularieae and Graphideae, except for Schizotrema and Topeliopsis (Fig. 6). In addition, the analysis suggests absence of lateral paraphyses at certain nodes within tribe Thelotremateae, including the clades containing the genera Chroodiscus, Gintarasia, Leucodecton and Pseudoramonia. Presence of lateral paraphyses was reconstructed for the bases of the genera Acanthotrema, Acanthothecis, Astrochapsa, Chapsa, Diploschistes, Pseudochapsa, Schizotrema, Thelotrema, Topeliopsis and the clade including the genera Astrochapsa, Pseudochapsa and Pseudotopeliopsis. In general, species with a columella lack lateral paraphyses.The ancestral state of ascospore amyloidity was reconstructed as non-amyloid for the base of Graphidaceae (Fig. 7) and several basal nodes in the family, e.g. subfam. Graphioideae and Redonographoideae. Non-amyloidity was also reconstructed as ancestral state for the base of Thelotremateae, the genera Acanthothecis, Acanthotrema, Astrochapsa, Chroodiscus, Diploschistes, Gintarasia and Pseudoramonia. Ascospores were reconstructed as being amyloid for the base of the genera Diorygma, Graphis, the Graphis scripta group in Graphideae, all nodes in Ocellularieae and two nodes in Thelotremateae (Leucodecton and Thelotrema). Within Phaeographis s. lat., all nodes were reconstructed as hemi-amyloid.

Character state transformations
The minimal numbers of transformations under MP are shown in Table 2. The most conserved characters were photobiont, lateral paraphyses, ascoma aggregation and persistence, and habitat, whereas by far the fastest evolving was chemistry.
Direction of these changes varies among the characters (Table 3). A few characters were reconstructed to change unidirectionally, such as the switch from trentepohlioid to chlorococcoid photobiont, from persistent to mazaediate ascomata, and solitary to pseudostromatic ascomata. In contrast, transformations from apothecioid to lirellate and vice versa were reconstructed as being almost equally common (53% vs. 46%).
We also tested whether the observed number of transformations was significantly different from expectation based on randomized data. Table 4 summarizes the results and show that for all 10 characters, both the randomization test of parsimony steps and Pagel's λ suggest high phylogenetic signal of the characters, except for ascoma persistence.
In addition, the nature of transformations was found to be highly phylogenetically structured even for the fastest evolving character, secondary chemistry. Transitions between no substances and (1 0 6 0 OH) depsidones and vice versa were almost entirely restricted to tribes Graphideae and Thelotremateae, and the subfamily Redonographoideae, whereas transitions between no substances and (1 0 COOH 6 0 ME) and (6 0 COOH) and vice versa were restricted to tribe Ocellularieae in this dataset.

Discussion
Our phylogenetic study is based on a broad taxon sampling of Graphidaceae including the entire range of morphological and chemical diversity in three of the four subfamilies currently accepted, even if the species studied here represent only slightly more than 10% of the total species diversity in this lineage. The overall phylogeny is congruent with previously published analyses which are discussed elsewhere in detail (Rivas Plata et al., 2013). Hence, we anticipate that including a larger number of species will not change the overall structure of the results, with the exception of possible new lineages to be discovered. Rather, we project that with increased taxon and gene sampling, higher resolution and backbone support will resolve ancestral character reconstruction for most of the presently unresolved nodes. In the current study the subfamily Gomphilloideae was not included and this might potentially influence the results on character evolution, especially the basal nodes of Graphidaceae and hence we have refrained from discussing the character states obtained for node 1 at the base of the family in detail.
We found support for the ancestor of subfamily Graphidoideae to be a tropical species with a trentepohlioid photobiont, apothecioid, solitary ascomata with persistent hymenium lacking a columella and lateral paraphyses, and having non-amyloid ascospores. Among extant lineages, the taxon that comes closest to this circumscription is Acanthotrema. This genus indeed appears to be the relict of an ancient lineage going back almost 100 million years and usually is positioned close to the base of the subfamily (L€ ucking et al. 2013).
Reconstructions of the ancestral character state of the cortex and chemistry did not yield conclusive results for the ancestor of the subfamily. Regarding ecology, our results suggest that subtropical and non-tropical Table 3. Proportion of changes between character states under maximum parsimony summing to 1 for ten studied characters in lichenized fungi family Graphidaceae. The second line contains maximum a posteriori value of number of transitions and 95% equal-tail credible interval in the parenthesis. species in this lineage are derived from tropical taxa. However, none of these lineages re-adapted to strictly tropical rain forest vegetation, where the majority of Graphidaceae occurs. This is consistent with a previous study with a smaller taxon sampling (L€ ucking et al., 2013). A chlorococcoid photobiont is restricted to the nontropical genera Diploschistes and Xalocoa (formerly D. ocellatus; Kraichak et al., 2014). Within those clades, there was no transformation back to trentepohlioid photobionts, which are the dominant type of photosynthetic partners in this family, including other extratropical groups within the family such as Schizotrema and Topeliopsis (Nakano, 1988;Nelsen et al., 2011). Species of Diploschistes differ from other Graphidaceae in their peculiar ecology, being most common and diverse in subtropical, semi-arid regions in both hemispheres and growing mostly on soil and rock substrata. Other, unrelated lichenized fungi sharing these habitats almost exclusively have chlorococcoid photobionts. Since Diploschistes species do not produce vegetative propagules but exclusively propagate by means of ascospores, new thalli can only be formed through resymbiosis with the appropriate photobiont, and it is highly unlikely in these habitats to encounter Trentepohlia algae. This is recognized by some species of Diploschistes being juvenile parasites on other lichen species (Hawksworth, 1982;Friedl, 1987;Friedl & G€ artner, 1988). We therefore conclude that the switch to a chlorococcoid photobiont enabled Diploschistes to diversify in these unusual habitats for Graphidaceae and with no reversal to a trentepohlioid photobiont in this lineage. Notably, subfamily Gomphilloideae, which was not included in this study, also is characterized by a chlorococcoid photobiont. However, this subfamily is predominantly tropical and mostly foliicolous, sharing niches with other lichens that do have trentepohlioid photobionts, so the advantage of the photobiont in this lineage is unclear.  Table 4: Observed numbers of parsimony steps of transitions and phylogenetic signals of ten studied characters in the lichenized fungal family Graphidaceae from randomization tests (Archie 1989) and Pagel's Lambda (1999). Mean random and SD random refer to the mean and standard deviation of the number of parsimony steps from 999 randomizations of character states. P random is the probability that the observed number is higher than expected from the randomization. Thus, a low P random suggests high phylogenetic signal. Lambda ¼ 1 indicates high phylogenetic signal, and P LTR is the result from the Likelihood Ration Test when compared the likelihoods of data against the tree transformed to lambda ¼ 0 (no signal). Therefore, a low P LTR indicates a significant deviation from the model with no signal. Our study suggests that a cortex was gained twice as often as lost, and that this character has switched repeatedly over the evolution of the family (15-29 times). As discussed elsewhere (Lakatos et al., 2006;Rivas Plata & Lumbsch, 2011), a dense cortex could be an adaptation to avoid oversaturation with water in understorey crusts (the cortex is water-repellent) or to provide protection against high light levels and damaging UV radiation, as well as against herbivores and fungivores. However, ecorticate thalli might also be water-repellent, but respond to hydration differently (Lakatos et al. 2006). Given these multiple functions of the cortex and the fact that closely related lineages may show different habitat preferences, it is not surprising that a cortex has been gained or lost several times independently. Despite the fact that we coded secondary metabolites as substance classes (Huneck & Yoshimura, 1996), rather than individual substances, we reconstructed a high number of transformations. Secondary metabolites are well-known in lichenized fungi, with over 1000 substances recorded (Huneck & Yoshimura, 1996;Lumbsch, 1998;Nash, 2008) and their potential ecological importance, such as sun screens, protection against herbivores and control of carbon diffusion, has been discussed in a number of studies (Lawrey, 1983;Golojuch & Lawrey, 1988;Emmerich et al., 1993;Giez et al., 1994;Rikkinen, 1995;Lange et al., 1997;Nybakken et al., 2004Nybakken et al., , 2010Gauslaa, 2005;McEvoy et al., 2007;Solhaug et al., 2009). Our study showed that the majority of the transformations (86%) are between having no substance and producing a given substance class, while transformations between different substance classes are much less common. In addition, substance classes appear to be largely conserved within major clades. Thus, basal nodes of tribe Ocellularieae (e.g. Rhabdodiscus, Myriotrema, Stegobolus) are reconstructed as having (6 0 -COOH) depsidones, i.e. psoromic acid and relatives, while the majority of clades within Ocellularia are reconstructed as having (1 0 -COOH-6 0 -ME) depsidones, i.e. protocetraric acid and relatives. The high level of conservation among major clades suggests that transformations between substance classes are the result of constrained evolution. The pattern of occurrence and absence of secondary metabolites within each clade suggests that there is no strong selective advantage in having medullary secondary substances, since species with and without substances often grow side by side in the same conditions, e.g. in the genera Graphis and Ocellularia.
Our analyses suggest that the transformation of ascomata with persistent hymenium to mazaediate ascomata happened at least twice in Graphidaceae and that there was no reversal from this trait back to persistent ascomata. Mazaediate ascomata are characteristic for the genera Nadvornikia and Schistophoron Tehler et al., 2009). Both genera have few species, with rather deep phylogenetic relationships, with lack of evidence for radiation after the transformation, contrary to other mazaediate groups such as calicioid species in Physciaceae, which are very species-rich. Mazaedia are considered an adaptation to wind dispersal of ascospores, and species with such ascomata often grow in open microhabitats and particularly well on old trees with weathered bark or on wood. This is also the case with species of Nadvornikia and Schistophoron, hence representing a case of convergent evolution with unrelated lineages also producing mazaedia, such as Heterocyphelium and Tylophoron (Arthoniales), Mazaediothecium and Pyrgillus (Pyrenulales), and many of the species in Caliciales.
Ascoma shape was traditionally used to distinguish the families Graphidaceae and Thelotremataceae, now included in a single family, as well as groups of genera within Graphidaceae. Our study confirms previous results demonstrating that ascoma shape does not constitute a synapomorphic trait characterizing monophyletic groups within Graphidaceae (Staiger et al., 2006;Mangold et al., 2008b;Rivas Plata et al., 2013). Transformations from apothecioid to lirellate ascomata and vice versa are almost equally probable in the family. Despite the number of transformations in this character, most nodes -except the majority of basal nodes -were reconstructed with strong support as being either apothecioid or lirellate. While the character is variable at a deeper phylogenetic level, it is mostly invariable within genera and tribes (Rivas Plata et al., 2012a). This suggests a high level of evolutionary plasticity at the more basal nodes but subsequent stabilization at nodes leading to major lineages. This is consistent with the observation that this character is more variable in smaller clades, such as Acanthothecis, which includes species with both types of ascomata (perfectly round and strongly lirellate ascomata). The reason for the observed variation and its relatively high level of conservation is unknown, but likely this character is somehow involved in protecting the hymenium and ascospores from fungivores or exposure to UV radiation. Since most species reproduce through ascospores, reproductive success depends in part on the number of ascospores produced, which is in turn a function of the hymenium surface. In lineages with rounded ascomata, the hymenium has to expand radially in order to produce a large number of ascospores, exposing a large surface area to the environment and thus making it more vulnerable. To still protect the hymenium, one solution is to maintain the ascomata relatively closed, which is observed in many lineages in tribe Ocellularieae and in the genera Thelotrema and Leucodecton (Thelotremateae). Another solution is to cover the hymenium with dead hyphal material and crystals for protection, as found in species with chroodiscoid ascomata in tribe Thelotremateae. Another alternative is to maintain the ascomata completely closed, in which case a lirellate shape is of advantage since the ascoma can grow without the necessity to expose the hymenium.
Although we found evidence for at least 10 transformations in ascoma aggregation, we found no evidence for transformations from pseudostromatic to solitary ascomata. This is consistent with results in other groups that in part produce pseudostromatic ascomata, such as Arthoniales .
The higher number of transformations in gaining a columella and lateral paraphyses in comparison to losing them can be interpreted as evidence for an adaptive value of these structures. A columella is a sterile tissue in ascomata that can sometimes cover large parts of the ascomata, potentially preventing fungivores from feeding on hymenial structures (L€ ucking & Bernecker-L€ ucking, 2000;Rivas Plata & Lumbsch, 2011). The adaptive value of lateral paraphyses, however, is currently not understood. The character might represent a conserved relict of the ascoma ontogeny, rather than an ecological adaptation, since the lateral paraphyses are ontogenetically a part of the generative hymenium (Henssen, 1976(Henssen, , 1995. It appears that in lineages forming lateral paraphyses, the hymenium originally develops throughout most of the cavity, and upon maturity and opening of the ascoma, the lateral portions of the hymenium that originally developed in the upper parts of the cavity remain vertical and sterile. Ascospore amyloidity showed a high number of transformations (16-28), especially gains and losses of amyloidity. Similar to what we observed for secondary chemistry, in spite of the high number of transformations, this character is usually conserved at the generic and often also the tribe level, with the notable exception of tribe Thelotremateae (and subfamily Fissurinoideae). Amyloid ascospores are also found in other groups of lichen-forming and non-lichenized ascomycetes, but the ecological importance of this character is not known.
In summary, it appears that while there are a large number of transitions between phenotype character states in Graphidaceae, more than any other crustose lichen group with the exception of Arthoniales, these transitions are highly structured phylogenetically and also provide evidence to formulate hypotheses on ecological functions versus evolutionary constraints. Our results are consistent with previous studies on character evolution in the family focusing on striking cases of parallel evolution and evolutionary plasticity in closely related lineages (Rivas Plata & Lumbsch, 2011). However, our extended sampling of loci and taxa improved the confidence in the phylogenetic estimate (Rivas Plata et al., 2013) and allowed for a more solid statistical approach, also demonstrating an even higher number of character transformations than previously assumed. This study hence provides an ideal base for addressing the question whether and how specific traits are correlated with ecological conditions and how these traits go along with diversification, in order to test hypotheses of adaptive radiations.