•The mapping of functional traits onto chronograms is an emerging approach for the identification of how agents of natural selection have shaped the evolution of organisms. Recent research has reported fire-dependent traits appearing among flowering plants from 60 million yr ago (Ma). Although there are many records of fossil charcoal in the Cretaceous (65–145 Ma), evidence of fire-dependent traits evolving in that period is lacking.
•We link the evolutionary trajectories for five fire-adapted traits in Pinaceae with paleoatmospheric conditions over the last 250 million yr to determine the time at which fire originated as a selective force in trait evolution among seed plants.
•Fire-protective thick bark originated in Pinus c. 126 Ma in association with low-intensity surface fires. More intense crown fires emerged c. 89 Ma coincident with thicker bark and branch shedding, or serotiny with branch retention as an alternative strategy. These innovations appeared at the same time as the Earth’s paleoatmosphere experienced elevated oxygen levels that led to high burn probabilities during the mid-Cretaceous.
•The fiery environments of the Cretaceous strongly influenced trait evolution in Pinus. Our evidence for a strong correlation between the evolution of fire-response strategies and changes in fire regime 90–125 Ma greatly backdates the key role that fire has played in the evolution of seed plants.
Increasing evidence from paleontological records and phylogenetic analyses suggests that fire is an ancient process that has influenced the history of life on Earth as far back as the origin of land plants (Bowman et al., 2009; Pausas & Keeley, 2009). Many plants have acquired adaptive traits that enable them to persist and reproduce in fire-prone environments (Gómez-González et al., 2011; Keeley et al., 2011; Pausas et al., 2012). The fossil charcoal record dating from over 400 million yr ago (Ma) is a valuable source of information on the historical occurrence of fire (Scott, 2000, 2010; Glasspool et al., 2004). However, the evolutionary impact of fire on plant evolution is difficult to trace because of inherent habitat biases in the fossil record and because fire-related traits are rarely evident among fossils (Midgley & Bond, 2011). Therefore, dating the origin and evolution of fire-adapted traits from time-based molecular phylogenies can provide new insights into the role of fire in plant evolution, especially the identification of the time at which fire started to act as an agent of natural selection.
Therefore, the known paleohistory of Pinaceae provides a unique opportunity to investigate the impact of fire on the evolution of plants from ‘deep time’ to the present. In this study, we link evolutionary trajectories for five fire-adapted traits in the family Pinaceae to paleoatmospheric conditions over the last 250 million yr (My) to determine the time at which fire originated as a selective force on trait evolution in this family, and test the hypothesis that its fire-driven evolution began in the Cretaceous (Berendse & Scheffer, 2009; Bond & Scott, 2010).
Materials and Methods
Phylogenetic reconstruction and fossil calibration
Many species in Pinus are fire dependent, but few species in other genera of the Pinaceae possess fire-related functional traits (Keeley & Zedler, 1998). We therefore first constructed a phylogeny for 101 species of Pinus, one or two species of the 10 other recognized genera and three outgroup species (Araucaria araucana, Podocarpus elatus, Sciadopitys verticillata) by retrieving their DNA sequences (matK and rbcL; Gernandt et al., 2005) from GenBank (http://www.ncbi.nlm.nih.gov/genbank/). Phylogenies for all other multi-species genera (except for Keteleeria, which only has three species) were reconstructed by retrieving DNA sequences from Genbank, and outgroup species were chosen as in the original literature (Supporting Information Notes S1).
The phylogeny with branch lengths proportional to time was derived using Bayesian inference with fossil calibrations. When radiometric dates were unavailable, fossil ages were assigned on the basis of a recent geological timescale and the midpoint of the age was used. Gernandt et al. (2008) and Willyard et al. (2007) performed extensive analyses of the fossil record for Pinaceae and suggested the optimal calibration regime for phylogenetic dating. In constructing the phylogeny with branch lengths proportional to time (chronogram) of Pinus, we assumed that Pityostrobus berinissartensis (Alvin, 1957) is the sister group to Pinus and calibrated the node corresponding to the most recent common ancestor of Pinus, Picea and Cathaya at 123 Ma (Gernandt et al., 2008). Within Pinus, Gernandt et al. (2008) delineated two subgenera (Pinus and Strobus) using chloroplast DNA (matK and rbcL), which is also supported by morphological, biogeographical and biochemical character analyses (Kaundun & Lebreton, 2010). A subgeneric divergent calibration at 85 Ma was therefore used (Willyard et al., 2007). A minimum age calibration was also set at the node for the most recent common ancestor of Pseudolarix, Tsuga and Nothotsuga at 156 Ma, given the radiometric date of 156 ± 0.76 Ma (mid-Oxfordian) for the oldest Pseudolarix fossil, P. erensis (Keller & Hendrix, 1997). Based on the macrofossil record for individual genera, a minimum age was set: Abies, 45.5 Ma (Erwin & Schorn, 2005); Larix, 45.0 Ma (Schorn, 1994); Picea, 45.5 Ma (Erwin & Schorn, 2005); Tsuga, 44.9 Ma (Erwin & Schorn, 2005); Pseudotsuga, 32 Ma (Schorn, 1994); Cedrus, 54.8 Ma (Blokhina, 1998).
The chronograms were derived using Bayesian inference in BEAST ver. 1.6.1 (Drummond & Rambaut, 2007); the input file was generated in BEAUTi (Drummond & Rambaut, 2007). We used a Yule process to describe speciation, and an uncorrelated log-normal model to describe the relaxed molecular clock. For each calibration point, a normal prior distribution, with a standard deviation of 5% to account for temporal uncertainty, was used in a Markov Chain Monte Carlo (MCMC) procedure, as the normal prior distribution best allows for uncertainty in the calibration estimates. Posterior distributions of parameter values and the tree topology were estimated by MCMC sampling. Five replicate MCMC runs were performed for each genus phylogeny, with the tree and parameter values sampled every 5 × 103 generations over a total of 1 × 108 steps. MCMC diagnostics were evaluated in Tracer v1.5 (Drummond & Rambaut, 2007). The consequent tree file containing 1000 plausible phylogenies with branch lengths (divergence times) was used for the reconstruction of the ancestral state as described below. Posterior samples from the five independent runs were combined and the trees and parameter values were summarized. A maximum clade credibility tree was identified using TreeAnnotator (Drummond & Rambaut, 2007). The topologies of all subsequently dated phylogenies resembled the original phylogeny, but with improved resolution as a result of Bayesian inference not used in the original reconstruction. For illustrative purposes, a supertree of the Pinaceae family was constructed by grafting the maximum clade credibility tree of individual genera onto the Pinus phylogeny (maximum clade credibility tree) that included all genera in the backbone.
Trait data and fire regime
We define two fire regimes for fire-prone pine forests: a surface fire regime, in which fire burns the ground stratum and litter but does not reach the crown stratum (canopy), and a crown fire regime, in which fire also burns the canopy of the dominant vegetation (Gill & Allan, 2008). Surface and crown fires not only have different effects on ecosystem processes, but also exert different selective pressures on the evolution of plant traits (Keeley & Zedler, 1998). We define any landscape as nonfire prone if the mean fire interval is > 150 yr as that is considered to be the lower estimate of longevity for many pines (Knowles & Grant, 1983).
Bark thickness Thick bark in Pinus has been demonstrated to insulate the cambium from heat generated by surface fires (Jackson et al., 1999) and thus has a significant effect on plant survival after fire, especially from surface fires (Jackson et al., 1999; Brando et al., 2012). Bark thicker than 15 mm is sufficient to insulate the cambium from low-intensity surface fires (temperature of 400°C for up to 3 min; Vines, 1968; Wright & Bailey, 1982; van Mantgem & Schwartz, 2003). Thicker bark (> 30 mm) provides protection from intense surface fires (temperatures over 800°C for > 10 min; van Mantgem & Schwartz, 2003). Alternative explanations for the ecological function of thick bark have been proposed in pines, but all lack generality. Thick bark could be an adaptation to protect trees from bark beetles, but there is evidence that they infest most conifers indiscriminately over a wide range of bark thicknesses (Wood, 1982). Bark has been reported to thicken in Taxodium distichum when flooded (Yamamoto, 1992), but most extant pine species prefer dry habitats (except for P. krempfii which grows in rainforest). Thick bark could be a protection from extreme cold, but, among 39 pine taxa in tropical areas, 30 have thick bark (Myers & Rodríguez-Trejo, 2009), implying that other functions usually apply to thick bark. Bark thickness is related to tree age (Jackson et al., 1999), so that, whenever possible, we compiled bark thickness data at breast height for adult trees.
Serotiny Serotiny in pines (apart from Pinus, only Picea mariana and, perhaps, some populations of Larix gmelinii are serotinous in Pinaceae) is marked by the retention of mature cones that open promptly only after exposure to high temperatures generated by fire. This mechanism ensures prolific seed release after fire when conditions for seedling establishment are highly favorable, that is, when resources, such as light, moisture and nutrients, are maximal and competition with adults is minimized (Lamont et al., 1991; Keeley et al., 2011). Mass seed release may also allow the satiation of predators (Lamont et al., 1991). Much research has shown that serotiny is tied to fire environments and, in the absence of fire, there is usually negligible recruitment, making clear the fitness benefits of serotiny in fire-prone systems, especially among fire-killed species (Lamont et al., 1991; He et al., 2011; Keeley et al., 2011).
Branch shedding Some Pinus species have the ability to shed their lower branches once their foliage has died (self-pruning or branch shedding), thus creating a fuel gap between the forest floor and the crown. This gap limits the ability of surface fires to climb into the canopy, and thus branch shedding is typically linked to surface fire ecosystems and the absence of serotiny, whereas branch retention is linked to crown fire systems and the presence of serotiny (Keeley & Zedler, 1998; Schwilk & Ackerly, 2001). Dead branch retention has been shown to increase fire intensity in shrub systems (Schwilk, 2003).
Grass stage A curious life-history trait in Pinus is delayed trunk development (‘grass stage’), whereby internode elongation is suppressed for the first 5–10 yr of development, with the resulting juveniles persisting in a bunchgrass growth form (Keeley & Zedler, 1998; Rodríguez-Trejo & Fulé, 2003). During this stage, needles are decumbent and accumulate as a mantle over the apex and insulate the apical bud against fire heat (Rodríguez-Trejo & Fulé, 2003).
Trait states for the above five fire-adapted traits and the fire regime within which each species lives were collated from the literature, personal observations and contact with local ecologists (Table S1).
Ancestral state reconstruction and trait evolution correlation through time
The times of origin of functional traits and their alternative states were identified by the reconstruction of ancestral traits within the studied taxa. We first reconstructed the ancestral trait state for each of the five fire-adapted traits for Abies, Picea, Tsuga, Pseudotsuga, Cedrus and Larix. The consequent ancestral traits of the above genera were then used in the ancestral trait reconstruction for Pinaceae, which included 101 species of Pinus and one or two apparent ancestral species in all other genera. For ancestral trait reconstruction, the agglomerative technique assumes that the current trait state for each species or lineage remains unchanged from the node until fusion with its sister lineage, when the probability of that trait is retested by simultaneously comparing all previous probabilities linked to that node (Hardy, 2006). Bayesian MCMC methods were used to derive posterior distributions of log-likelihoods and the probability values of traits at the nodes, and were implemented in BayesTrait (Pagel & Meade, 2007). For each dataset, the tree file containing 1000 plausible phylogenies with the branch length (node age) generated in BEAST and the trait files containing the discrete state of the trait for each species were used for further analysis.
A continuous-time Markov model, in which the trait state can evolve repeatedly between the two possible states at any branch of the phylogenetic tree (Pagel et al., 2004), was used to construct the ancestral state at each internal node, and a reversible-jump (RJ-)MCMC was adopted (Pagel & Meade, 2007). A few Markov chains were run for each trait and the acceptance rate of newly proposed values of the rate parameters was monitored to determine a suitable value for the ratedev parameter that generated an acceptance rate of 0.20–0.40 (Pagel & Meade, 2007). A hyperprior seeding of an exponential distribution from a uniform 0–30 distribution was applied to the Markov chain runs. The Markov chains were run for 1 × 108 iterations with a burn-in of 5 × 104, and every 5 × 103 iteration was logged. The ancestral state was defined as the trait state with the higher average posterior probability determined by a t-test (P <0.05).
Tests for correlated evolution between binary functional traits (or with their putative selective environment) were implemented by comparing the fit of two continuous-time Markov models: one that presumes that the two traits evolve independently on the tree (independent model), and one that allows the traits to evolve together (dependent model). The log-Bayes factor is nominally distributed as a chi-square with degrees of freedom equal to the difference in the number of parameters between the two models, and a log-Bayes factor > 9.49 indicates strong support (P <0.05) for an evolutionary correlation between the two traits (or with a particular fire regime). In each model run, 5 × 106 iterations were implemented and the results were logged every 2 × 103 iterations. The overall results were summarized by the log-Bayes factor, which is twice the difference between the two harmonic means derived from the dependent and independent models of evolution (Pagel & Meade, 2007).
Reconstruction of biomass burning probability
Burn probabilities over the last 350 My were estimated using the FIREOX model of Belcher et al. (2010). The FIREOX model is a cellular automaton model, analogous to a lattice cellular automaton of a SIR-type epidemic (Karafyllidis & Thanailakis, 1997). FIREOX simulates the propagation of a smouldering fire (Stott, 2000; Drysdale, 2011) as a contact invasion process when ignited from a localized external source at a given atmospheric O2 concentration. The model has two parameters, the probability of local fire spread (β) and the probability of local fire extinction (μ), and is driven by spread rate and burn duration information from experiments undertaken at different concentrations of atmospheric O2 (Belcher et al., 2010). The model generates burn probabilities for different values of β and μ, and refers to the amount of area burned in the grid of the cellular automata. The modeled relationship between burn probability and atmospheric O2 concentration is then used to estimate the burn probability over the past 350 Ma according to modeled estimates of paleoatmospheric O2 (in this case, the COPSE model of Bergman et al., 2004).
We constructed a dated molecular phylogeny for Pinaceae, showing that the evolutionary history of crown Pinaceae can be traced to 237 Ma (189–293 Ma; 95% highest posterior probability, HPD) in the mid-Triassic (Fig. 1). The Pinus clade, the largest genus in the Pinaceae, originated 126 Ma (105–147 Ma, 95% HPD), with a crown age at 89 Ma (80–96 Ma), when the subgenera Pinus and Strobus diverged (Fig. 1). Key fire-adapted traits are well represented in Pinus (Figs 1, 2). Thin bark, seed release at maturity and branch retention are ancestral (Figs 1, 2, S2). We first tested whether the proliferation of the main fire-related traits in Pinus is correlated with the two standard fire regimes, surface and crown fires. The selected traits were the presence of thick bark and canopy seed storage (serotiny). We found that thick bark (> 15 mm) is closely correlated through time with the surface fire regime in Pinus (P <0.01), and the correlation is even stronger for very thick bark (> 30 mm) (P <0.001, Fig. S1), supporting the postulated evolution of bark thickness as a response to surface fires. Serotiny is strongly adaptive in the presence of crown fires and is closely associated with the crown fire regime in pines (P <0.001).
To determine the time of origin of these fire-adapted traits, we reconstructed the ancient state of thick bark (> 15 mm), very thick bark (> 30 mm) and serotiny on the dated Pinaceae phylogeny. We found that thick bark originated up to 126 Ma (105–147 Ma) (stem age) in Pinus (posterior probability = 0.63, P <0.05; Fig. 1), suggesting that surface fires existed then and were already shaping the direction of trait evolution in Pinus. Very thick bark (> 30 mm), coupled with branch shedding (correlated evolution: P <0.001), appeared up to 89 Ma (80−96 Ma) (stem age) in subgenus Pinus (posterior probability = 0.67, P <0.05; Fig. S1). Serotiny occurs in sections Pinaster, Contortae and Australes of subgenus Pinus. It was traced back to 89 Ma (posterior probability = 0.68, P <0.05; Fig. 2) (stem age) and was correlated with lower branch retention (P <0.05).
The other fire-related traits examined, fire-tolerant grass stage and ability to resprout, are rare and clearly derived among pines; their lineages were traceable to < 40 Ma (Fig. S3). These two traits are correlated with recurrent fire through time (P <0.01). Thick bark and serotiny, but not grass stage and branch shedding, are also present in other genera of Pinaceae, but are rare and recent, and appear to represent convergent evolution among those species in fire-prone environments (Figs 1, 2, S1–S3).
Modeled (Bergman et al., 2004) and proxy (Glasspool & Scott, 2010) estimates of atmospheric O2 concentrations over the past 350 My show that exceptionally high concentrations of atmospheric O2 (> 26%) occurred c. 100 Ma in the Cretaceous (Fig. 3). Such high O2 levels are estimated to have led to increased burn probabilities in the Earth’s forests, and inertinite showed a local peak at that time. Atmospheric CO2 concentrations and temperatures were also elevated during this time (Fig. 3).
Fire is known to have been a regular occurrence in Cretaceous conifer ecosystems, and the Pinaceae is well represented in the fossil charcoal record at this time (Batten, 1998; Collinson et al., 2000; Falcon-Lang et al., 2001; Glasspool & Scott, 2010), but ascertaining the evolutionary impact of fire on early seed plants has been elusive (Pausas & Keeley, 2009; Pausas & Schwilk 2012). Our analysis shows that, although the Pinaceae has a 237-My evolutionary history, fire-adapted traits only appeared in this family from 126 Ma. The development of thick bark coincided with increases in atmospheric O2 and burn probabilities (Fig. 3). Initial low-intensity surface fires, having a minor effect on plant traits, were replaced by highly intense crown fires by 89 Ma (possibly as early as 96 Ma), and resulted in the evolution of two alternative suites of highly successful, fire-adapted strategies: very thick bark with branch shedding, and serotiny with branch retention. We conclude that the change to a more intense fire regime has been the main selective force in the evolution of these fire-related traits in pines.
The Cretaceous fossils assignable to subgenus Pinus (which best expresses all fire-adapted traits) are widespread throughout America and Eurasia at mid- to high latitudes (Myers & Rodríguez-Trejo, 2009), suggesting that recurrent fire might have existed on a broad scale at that time. The trait innovations of thick bark, serotiny and branch shedding appeared at the same time as the Earth’s atmosphere experienced record high O2 concentrations that would have led to greatly increased plant flammability. We also note that many pinaceous fossils cannot be assigned to any modern groups (Gernandt et al., 2008), suggesting extinction of those taxa unable to adapt to these novel fire regimes. For the nucleus of Pinaceae that did adapt to fires originating as far back as 126 Ma, speciation was both strong and widespread across the Northern Hemisphere until pines became, and remain, the dominant group there.
Fuel characteristics and weather are the two most important drivers of fire in modern fire-prone ecosystems (Archibald et al., 2009; Krawchuk & Moritz, 2011; Pausas & Paula, 2012). However, the ignition and spread of fire are also strongly dependent on O2 concentration in the atmosphere (Belcher & McElwain, 2008), such that burn probabilities are known to increase abruptly above 24% O2 (Belcher et al., 2010). Because atmospheric O2 concentrations have fluctuated throughout Earth’s history (Bergman et al., 2004; Glasspool & Scott, 2010), O2 has been a major factor controlling fire activity in the past (Scott, 2000, 2010; Pausas & Keeley, 2009; Belcher et al., 2010). During the Cretaceous, high atmospheric O2 concentrations at c. 125 Ma would have greatly increased the flammability of the Earth’s ecosystems, as indicated by the elevated biomass burning probabilities (Fig. 3). Higher atmospheric CO2 concentrations at a similar time would have led to an increase in primary productivity (LaMarche et al., 1984) and, consequently, greater fuel loads. Huang et al. (2007) reviewed evidence from CO2 enrichment experiments and dendrochronological studies, and concluded that atmospheric CO2 fertilization increases photosynthetic rates, above- and below-ground biomass. As the carboxylating enzyme has a much greater affinity for CO2 than O2, and the increase in CO2 was an order of magnitude greater than the increase in O2 at any time, these trends would have been unaffected by the higher levels of O2 during this period (Fig. 3). Combined with the globally warm temperatures during this period (Fig. 3; Spicer & Herman, 2001; Spicer, 2003) all these conditions probably made the Cretaceous one of the most flammable periods in the Earth’s history.
Our results support not only the pivotal role of fire in the evolution of Pinus, but also provide a place for fire in the evolution of flowering plants (Bond & Scott, 2010). Angiosperms diversified from c. 135 Ma and rapidly increased their ecological importance from c. 100 Ma (Crane & Lidgard, 1989). The earliest angiosperms (Magnoliids and pre-Magnoliids) are believed to have been ‘weedy’, understory herbs, shrubs and small trees in ‘dark, damp and disturbed’ habitats (Feild & Arens, 2007). Although mostly of tropical origin, it is possible that these early angiosperms contributed to the initial low-intensity surface fire regime in the early Cretaceous conifer forests, together with cycads, ferns and seed-ferns. By 90–100 Ma, the monocots and early eudicots had arisen, diverged and migrated into conifer-dominated forests of the Northern Hemisphere (Stevens, 2012), including such fire-tolerant, fine-leaved rhizomatous and cormaceous clades as Liliales (stem Melanthiaceae at 109 Ma), Asparagales (Iridaceae at 103 Ma, Aparagaceae at 91 Ma), Commelinales (Haemodoraceae at 98 Ma; Lamont & Downes, 2011) and Poales (Cyperaceae at 88 Ma, Poaceae at 89 Ma; Janssen & Bremer, 2004), ‘weedy’ Ranunculaceae (> 100 Ma; Krassilov & Volynets, 2008) and shrubby Berberidaceae. Thus, the highly flammable understory angiosperms would have greatly expanded during the second wave of fire-related evolution among pines, and promoted equivalent fire-related responses among them as well (Bond & Scott, 2010). Thus, our evidence for a strong correlation between the evolution of fire-response strategies and increase in fire intensity at 90 and 125 Ma among pines greatly backdates and extends the key role played by fire in the evolution of seed plants more generally.
Bond & Scott (2010) noted the marked drop in fire-derived inertinite records for Northern Hemisphere mires during the early Eocene (40–55 Ma) and attributed this to the suppression of fire by the now-dominant angiosperm forests during one of the wettest periods known. This raises the question of the fate of fire-dependent species at that time (and thus the validity of our findings). First, atmospheric O2 and CO2 levels, temperatures and burn probabilities were still much higher than currently (Fig. 3); second, inertinite levels have remained low (Fig. 3), despite the fact that most pine forests are currently fire prone, highlighting the inherent bias against recording dryland fire at wet sites; third, there is corroborating evidence of fire-prone floras and species existing or arising among pines (Stockey, 1984: subgenus Pinus for which fire-adapted traits are diagnostic; Figs 1, 2, S1) and elsewhere (Itzstein-Davey, 2004; Lamont & Downes, 2011) during the early Eocene; fourth, some parts of the world received only moderate and highly seasonal rainfall at that time (Macphail, 2007), and there is no reason to believe that the vast area currently covered by pine forests was any different; fifth, Keeley (2012) has postulated that pines migrated to drier, and hence more fire-prone, uplands at that time. We conclude that Pinus, and other fire-adapted clades, were well placed to survive the Eocene.
By analyzing the evolutionary history of the Pinaceae, we have obtained compelling evidence that fire has been an effective agent of natural selection since c. 126 Ma. Whether fire was equally potent before this remains an open question. From our analysis, fire appears to have been ineffective among the Pinaceae for the first 100 My of its existence. Cycads are a much older clade and are universally fire tolerant with fire-stimulated coning (Lamont & Downes, 2011) – they would seem to be the ideal candidate for further research, but for the recent findings that extant cycads originated in the early (Nagalingum et al., 2011) to mid- (Crisp & Cook, 2011) Cenozoic, with long unbranched stems, greatly reducing the probability of following fire-adapted traits into the Cretaceous, let alone beyond. Ginkgo is older than the crown cycads, but there is no evidence of fire in its history that might explain its ability to resprout when pruned. Modern tree and rhizomatous ferns may resprout after fire and extend back to the Cretaceous (no earlier than Pinus; McLoughlin, 2001; Schneider et al., 2004), and ferns are often represented in fossil charcoal and may become abundant immediately postfire (Collinson et al., 2000), but these instances are too poorly represented or documented among extant species to be able to impute their fire history using molecular methods. The tracing of fire adaptations beyond the Lower Cretaceous appears to be a major challenge for future research.
T.H. acknowledges support from the Australian Research Council (LP100100620 and DP120103389), J.G.P. from the VIRRA project (CGL2009-12048/BOS, Spanish Government) and C.M.B. from a European Union Marie Curie Intra-European Fellowship (FILE-PIEF-GA-2009-253780). We thank M. Bonilla, T. Nangchuk, M. Panayotov, D.A. Rodríguez-Trejo, W. Zheng, S. Li and Z. Zhang for providing trait data, and three referees for their helpful comments.