Seeds are a key pathway for plant population recovery following disturbance. To prevent germination during unsuitable conditions, most species produce dormant seeds. In fire-prone regions, physical dormancy (PY) enables seeds to germinate after fire. The thermal niche, incorporating seed dormancy and mortality temperature responses, has not been characterised for PY seeds from fire prone environments.

We aimed to assess variation in thermal thresholds between species with PY seeds and if the pyro-thermal niche is aligned with seed mass, ecosystem type or phylogenetic relatedness.

We collected post heat-shock germination data for 58 Australian species that produce PY seeds. We applied species-specific thermal performance curves to define three critical thresholds (DRT_50, dormancy release temperature; T_opt, optimum dormancy release temperature and LT₅₀, lethal temperature), defining the pyro-thermal niche. Each species was assigned a mean seed weight and ecosystem type. We constructed a phylogeny to account for species relatedness and calculated phylogenetic signal (h²) for LT_50, T_opt, and DRT₅₀.

Seeds of Pomaderris (Rhamnaceae) had the highest T_opt and LT₅₀, and Pomaderris bodalla having the highest DRT₅₀ of 101.3°C. Seeds from species within this family exhibited higher temperature thresholds than those from Fabaceae. Seed mass was only influential in explaining LT₅₀ variation.

Species selection and data acquisition.

We set out to acquire seed germination data following heat-shock for as many species as possible across temperate Australia. However, to provide accurate estimates of threshold conditions, we applied three rules that allowed data from multiple sources to be brought together; 1) seeds needed to be heated from ~40°C through to at least 120 °C (or higher if there was no evidence of seed mortality at 120°C),  2) seeds had to be heated for either 5 mins and/or 10mins and 3) there was > two non-zero germination data points within the treatments (e.g., seeds treated at 40 °C, 60 °C, 80 °C, 100 °C and 120 °C, but only recorded ~ 0%  germination at 80 and 100 would be discarded). Where there was insufficient data points to model the response, we removed these datasets were removed, as fitting response curves to two data points induces significant error into hypothetical responses. However, this was a rare occurrence, and only appeared once across all sampled datasets. Using these rules, we were able to acquire seed germination post heat-shock data for 58 species, gathered from either previously published articles (namely Auld and O'Connell (1991), Palmer et al. (2018) and Chen et al. (2022), Figure 2), or from newly acquired data. Protocols for heat-shock and subsequent germination trials for each data set are outlined in Auld and O'Connell (1991), Palmer et al. (2018) and Chen et al. (2022), heat-shock and germination methods within these publications are like those presented below. Raw data from Auld and O’Connell (1991) and Chen (2022) was provided by authors, while data from Palmer et al. 2018 was extracted directly from published figures using the package metadigitise ver 1.0.1 (Pick et al. 2019), within the statistical platform R version 4.2.2 (R Core Team, 2023). Due to the nature of the data presented in Palmer et al. (2018), only the species-specific means and standard errors for each species/treatment combination was able to be integrated. All germination data is provided in Table S1.

Newly acquired data was generated using similar methodologies to Auld and O'Connell (1991) and Palmer et al. (2018) but differed slightly depending on whether the species and seeds were from Western Australia (WA) or New South Wales (NSW; Table S2).

Fresh seeds from an additional 19 WA species were selected as a supplement the dataset. Seed fill was indicated by a full, intact embryo and for WA seeds, viability of all seeds was checked with X-ray (MX-20 Cabinet X-ray Unit, Faxitron, Wheeling, USA). Unfilled seeds were discarded. Viable seeds were exposed to heat pulses between 40°C - 120 °C at 10°C increments for 5 and/or 10 min (Table S2). Seeds were exposed to 140 °C if insufficient seed mortality was recorded at 120 °C (i.e., >50% survival) and seeds were available. To ensure a uniform and consistent heating environment between replications of heating, all seeds were heated inside stainless-steel mesh bags (2‐μm diameter mesh holes) and the bags were inserted into a bed of dry, bleached white silica sand that had been preheated to the required temperature in a laboratory oven (WA oven, Contherm, Korokoro, New Zealand; NSW Oven, D170fs, Steridium, Queensland, Australia). Each of the replicates were heat‐treated separately and sand temperature was monitored independently of the oven settings using a K‐type thermocouple attached to a digital thermometer (Model TX-1004X-SP, ThermoWorks, Utah, USA). After heat treatment, the seeds were quickly removed from the bags to accelerate cooling at room temperature and humidity. Following heat treatment, seed viability status was assessed via germination testing. For WA seeds, seeds were surface sterilised using 3% calcium hypochlorite (CaOCl2) for 20 min under vacuum, and each replicate was plated on a Petri dish containing 0.7% (w/v) water agar. Dishes were incubated at 15°C under a 12/12 hr light/dark regime (30 μM m−2 s−1, 400–700 nM, Model 620RHS Growth Chamber, Contherm Scientific, Wellington, New Zealand). This temperature regime was chosen to match seasonal germination requirements for seeds from southwestern Australia (Merritt et al. 2007).

Only four additional NSW species were used to acquire new data. For these, seed fill was assessed via a cut-test, whereby a firm, fully intact embryo was considered ‘filled’. Following heat-shock, each replicate was plated on moistened filter paper in a petri dish, and wetted regularly using reverse osmosis water to ensure seeds did not dry out during germination (except for Acacia melanoxylon which was plated on 0.7 % w/v water agar). Dishes were incubated at alternating 11-25°C temperature regime under a 12/12 hr light/dark regime (30 μM m−2 s−1, 400–700 nM). This temperature regime was chosen to match seasonal germination requirements for species from the Sydney basin region (Mackenzie et al. 2016). For both experiments, seed germination was recorded for a minimum of six weeks, or until no further germination was recorded for at least two weeks. At the end of the germination period, total germination for each replicate was recorded. For NSW seeds, post-germination viability was checked using a cut test (Ooi et al. 2004).

Phylogenetic relationships and tree development.

As we assessed species with differing degrees of relatedness, we wanted to test if this influenced the arrangement and patterning in the pyro-thermal niche amongst species. To achieve this, we assembled a hypothetical phylogenetic tree using data from published papers (Lavin et al. 2005, Boatwright et al. 2008, Bruneau et al. 2008, Murphy et al. 2010, Miller et al. 2011, Särkinen et al. 2012, Davies et al. 2013, Nge et al. 2021, Zhao et al. 2021) and built the tree using TreeGraph2.15.0-887 (Stöver and Müller 2010). Where divergence times were uncertain or not readily available for a target species, we used published tree topology and species arrangements, along with estimated median divergence times. This was based on comparisons between the target species and known outgroups using TimeTree version 5 (TToL5, (Kumar et al. 2022). For species without sequence data or phylogenetic information, divergence times were assigned to be equal with the divergence length of the node, indicating separation from sister taxa. If we lacked resolution to distinguish divergence times within clades, resulting in a polytomy, we separated taxa by allocating a divergence time of 1 million years (approximately equal to the minimum known divergence time) to each subsequent closest related species. Divergence times were trimmed to a common time ~ 131 MYA from root to tip, before divergence times were rounded using “force.ultrametric” in  phytools ver 1.9-10  (Revell 2012), implementing the “nnls” method from the package phangorn ver 2.11.1 (Schliep 2010) which computes the edge lengths that result in a minimized sum-of-squares distance between the patristic distance of the output and input trees. The resulting tree is ultrametric, which was subsequently scaled to unit length (Hadfield and Nakagawa 2010). Phylogenetic trees and the covariates were visualised using ggtree ver 3.10.1 (Yu et al. 2017).

Data analysis

All analysis was conducted within the statistical platform R version 4.2.2 (R Core Team, 2023). Raw data from Auld and O’Connell (1991) and Chen et al 2022 were combined with mean and standard errors of data reported in Palmer et al. (2018). These were combined with the newly acquired data to form a dataset of 58 species (Figure 2A), of which 28 species had been subject to both 5 and 10-minute heating durations, 23 subject to only 10-minute heating duration and only 7 species were subjected to 5-minute exposures only.

Figure 2: Visual depiction of analytic pipeline. A) Data sources of germination data B) Generating thermal performance curves for each species using rTPC. C) Extracting seed traits from AusTraits and combining them with ecosystem types D) Implementing MCMCGLMM models that integrate phylogenies as random effects.

Total germination (final germination proportion at the end of the germination trial) at each  heat treatment was used to build species-specific thermal performance curves (TPCs) using the rTPC package ver 1.0-2 (Padfield et al., 2021). TPCs were used to define the species-specific T_opt, DRT₅₀ and LT₅₀ (Figure 2B). 

To determine the most suitable model for each species-specific response, we applied a limited suite of non-linear (nls) models. The nls models used are described in Johnson and Lewin 1946, O’Neill et al. 1972, Lynch and Gabriel 1987, and Angilletta Jr 2006. We applied all the nls models as a set to the germination data for each species and fitted curves to each species independently. Curve fitting suitability was assessed using Akaike information criterion (AIC; Wagenmakers and Farrell, 2004; Bedrick and Tsai 1994), with the lowest ranking AIC model chosen for each species to provide the best fit. Model fit was further assessed visually, and obvious visual errors were used to remove spurious overfitted models (See data availability statement). The most suitable model was reconstructed using the minpack.lm ver 1.2-3 package (Elzhov et al. 2016), which provides functionality for bootstrapping within the car ver 3.1-2 package (Fox et al. 2012). Residual bootstrapping was undertaken for 100 iterations for each species-specific model to create 95% confidence intervals around modelled parameters and the mean DRT₅₀, T_opt, and LT₅₀ (Supplementary Data S2). We implemented residual bootstrapping to ensure predictor variables remain fixed during resampling in this method (Davison and Hinkley 1997), which is consistent with design of the germination data.

To assess how the T_opt, DRT₅₀ and LT₅₀ varied between species and ecosystems we compiled a simple set of co-variables, specifically seed mass and ecosystem type (Figure 2C). Seed mass was extracted from the AusTraits database (Falster et al. 2021). In cases where there were multiple records for seed mass for a single species, we calculated the mean seed mass and standard error for each species. Where there was no data in AusTraits and we had seed available, we calculated seed mass by weighing 10 replicates of 100 seeds. All seed masses are presented as single seed mass (mg). Ecosystem type was assigned based on the seed collections outlined in Auld and O'Connell (1991), Palmer et al. (2018) and Chen et al (2022). Newly acquired seed germination data was assigned an ecosystem type based on seed collection locations for that seed lot (Supplementary Data S2). Ecosystem types were restricted to high level categorisation, with three levels: Forest, Woodland, or Heath.

To assess whether the metrics of the pyro-thermal niche correlate with seed mass and ecosystem type, we developed phylogenetic mixed models using Markov Chain Monte-Carlo generalized linear mixed models (MCMCglmm) within a Bayesian framework, inside the MCMCglmm package (Hadfield 2010). We tested each pyro-thermal niche metrics (i.e., DRT₅₀, T_opt, and LT₅₀) against seed mass and ecosystem type while controlling for phylogeny (Figure 2D).

Once we had defined models for DRT₅₀, T_opt, and LT₅₀ for each species, we compiled co-variables, seed mass and ecosystem type, with phylogeny included as a random factor. To reduce the chance of autocorrelation, we ran 1,000,000 iterations with a burn-in of 1000 and thinning of 10. Effective sample size was ~99,900. Model diagnostics were assessed using the ‘plot.MCMCglmm’ function, with autocorrelation between iterations and overall model convergence visually assessed using trace plots, and the posterior density of each parameter assessed using density plots.

To assess whether pyro-thermal niche metrics were influenced by seed mass, we used the MCMCglmm models with a residual covariance structure included to associate each data point with a unique residual (Hadfield and Nakagawa 2010). To assess whether ecosystem type in isolation or through an interaction with seed mass shaped the pyro-thermal niche, we used the same approach but included ecosystem type into the model as a secondary covariable, specifically interacting with seed mass. Ecosystem Type was restricted to a 3-level factor. Seed mass was log transformed in all MCMC models. Again, we included a residual covariance structure to associate each data point with a unique residual. In both cases we included an inverse phylogenetic relatedness matrix as a random continuous variable. Posterior means (pmeans) are reported along with lower and upper 95 % credible intervals (LCI and UCI) in every instance and a partial MCMC value (pMCMC) was reported only where pMCMC ≤ 0.05. For each model we also report residual variance (ςe). 

To assess how the metrics of the pyro-thermal niche are aligned with phylogenetic relatedness, we report phylogenetic signal (via Lynch’s phylogenetic heritability h²: Housworth et al. (2004)). Heritability (h²) of a trait can be calculated by examining the proportion of the variance in the trait that is explained by the relationship among taxa given the phylogeny, and is defined by Equation 1 (Housworth et al. 2004), which compares the phylogenetic variance  with the variance explained by the residual variance or environmental variance  (Housworth et al. 2004, de Villemereuil and Nakagawa 2014). As h² is a ratio of phylogenetic variance and environmental variance, values of h² occur exclusively between zero and one, with values close to zero indicating trait expression in the absence of phylogenetic structure, while values close to one indicate complete phylogenetic structuring in line with Brownian motion evolution (Pearse et al. 2023). For each pyro-thermal metric we report the estimated posterior mean for h² along with lower and upper 95 % credible intervals (LCI and UCI).

                                                                                      Equation 1 h^2= σ_p^2/(σ_p^2+σ_e^2)