Frequent fires prime plant developmental responses to burning

Abstract

Coping with temporal variation in fire requires plants to have plasticity in traits that promote persistence, but how plastic responses to current conditions are affected by past fire exposure remains unknown. We investigate phenotypic divergence between populations of four resprouting grasses exposed to differing experimental fire regimes (annually burnt or unburnt for greater than 35 years) and test whether divergence persists after plants are grown in a common environment for 1 year. Traits relating to flowering and biomass allocation were measured before plants were experimentally burnt, and their regrowth was tracked. Genetic differentiation between populations was investigated for a subset of individuals. Historic fire frequency influenced traits relating to flowering and below-ground investment. Previously burnt plants produced more inflorescences and invested proportionally more biomass below ground, suggesting a greater capacity for recruitment and resprouting than unburnt individuals. Tiller-scale regrowth rate did not differ between treatments, but prior fire exposure enhanced total regrown biomass in two species. We found no consistent genetic differences between populations suggesting trait differences arose from developmental plasticity. Grass development is influenced by prior fire exposure, independent of current environmental conditions. This priming response to fire, resulting in adaptive trait changes, may produce communities more resistant to future fire regime changes.

1. Introduction

Fire is a major and ancient environmental perturbation to which plants have adapted through changes to their functional traits [1,2]. There is growing evidence of woody plant adaptation to fire, where different fire regimes create intraspecific trait variation that is genetically determined and thus subject to natural selection [3–5]. However, whether exposure to fire has a priming effect on plastic responses remains unexplored. An expanding body of evidence shows that plants may respond to extreme climatic events differently depending upon their previous experiences [6–9], and these primed responses can have positive effects on plant performance during environmental perturbations and avoid the costs associated with maintaining a constant high level of phenotypic plasticity [10].

Fire-prone savannah grasses are an ideal study system for exploring the role of fire on priming, as these species persist through the most frequent fire regimes on Earth [11] even though fires usually remove all above-ground biomass and can kill mature grass plants [12]. Grasses use two main mechanisms for persisting through fire: resprouting from protected organs or recruiting from a fire-resistant seed bank. The majority of grasses that inhabit fire-prone grasslands and savannahs are perennial, resprouting species [13,14]. For these species, traits that allow an individual to resprout quickly after fire, such as high specific leaf area (SLA) and large below-ground reserves, are likely to be favoured [15,16] as they allow access to the sunlight- and nutrient-rich environment immediately after fires [17,18]. While successful recruitment in many perennial grasslands is infrequent because seedlings must compete with established plants [19,20], traits that enhance recruitment after fire, such as the stimulation of flowering and germination [21,22] can help seedlings to access the high post-fire resources.

Here we test the hypothesis that fire causes divergences in recruitment and regeneration traits that favour persistence in fire-prone savannahs. Our work sampled four savannah grasses from experimental field plots that had been either unburned or annually burned for greater than 35 years. Cuttings from these plants were grown in a common environment for 1 year, after which traits were remeasured to determine whether differences between populations persisted independently of the current environment. In comparison to unburnt plants, we predicted that annually burnt plants would have traits advantageous under recurrent fires, including rapid post-fire resprouting (high investment in below-ground biomass, high regrowth rate, high SLA) and recruitment (fire-stimulated flowering). To evaluate whether phenotypic differences had a genetic component or arose from plasticity, we tested for allelic divergence among populations.

2. Material and methods

(a). Plant collection and establishment

Plants of four Poaceae species (Cymbopogon pospischilii, Digitaria eriantha, Melica racemosa and Themeda triandra) were collected on 7 September 2015, from experimental burn plots (based at the University of Fort Hare Research Farm, Eastern Cape, South Africa; 32°47′ S, 26°52′ E) experiencing two contrasting fire frequency treatments: annual burn and no burn. The latter had not been burned in the 35 years since the plots were created in 1980 ([23]; see the electronic supplementary material, table S1 for details about the plots including climate data). Both of these treatments represent a departure from the natural fire return interval of the semi-arid savannah site (approx. 15–20 years; W.S.W. Trollope, 2017, personal communication). Each treatment is replicated twice with the 100 × 50 m plots arranged in a randomized block design alongside four other fire frequency treatment plots that were not sampled here. The site varied little in slope and soil chemical and physical properties [24,25]. The four species occurred abundantly in all treatment plots and are perennial, resprouters from three grass subfamilies (electronic supplementary material, table S2). Based on the reported longevity of these species, it is likely that the populations have undergone several rounds of reproduction and recruitment in the 35 years of treatment [26]. Thirty-five mature individuals of each species were dug up from open areas of grassland, minimizing root damage, from across the two replicate plots (n = 17 or 18 from each plot). Within 48 h of collection, a clump of five tillers was removed from each individual. The roots were washed carefully to remove soil and limit the effects of any soil nutrient differences on plant growth. The clumps were subsequently planted into 10 l pots containing locally sourced topsoil. A voucher specimen of each species was created (see the electronic supplementary material, table S2 for specimen details).

To determine whether there were differences in plant traits between annual-burn and no-burn populations at the time of sampling from the experimental burn plots, traits were measured on 14 plants per treatment per species (n = 7 from each plot). For this, the remainder of biomass (after the five-tiller clump had been removed) from each plant was used to measure plant height and above-ground dry biomass (after drying for 48 h at 70°C).

The plants were grown for 12 months (July 2015–July 2016) in a common environment (a naturally lit polytunnel at Rhodes University, South Africa) in a fully randomized block design and were weeded and watered regularly. In the polytunnel, average monthly temperatures ranged from 14°C (July) to 26°C (January) and average relative humidity was 68% (as recorded by thermochron data loggers: i-buttons, model DS1923, Maxim Integrated Products, California, USA). A 12-month growth period was chosen so that plants could become well established in the pots and any environmental effects carrying over from the different field treatments could be minimized. Thirty-seven of the 280 plants died during this period, but mortality was not associated with a particular species, treatment or plot (ANOVA: p > 0.05). Watering was reduced and eventually stopped two weeks prior to burning to imitate the winter dry season and to force the plants into a phenological stage most relevant to burning.

(b). Pre-fire traits

On the day prior to the experimental burns, the number of flowering tillers was recorded and a sample of above-ground biomass (approx. one-quarter of the total biomass) was removed for all plants. The harvested biomass was dried (for 48 h at 70°C) and measured. For each species, eight annual-burn and no-burn plants were also randomly chosen, destructively harvested and used to measure above- and below-ground dry biomass. Roots were carefully washed over a fine sieve and then dried at 70°C for 7 days. Root dry mass was measured and expressed as a proportion of the total dry plant biomass.

(c). Experimental burn and post-fire regrowth

Plants were burned in a random order on a warm day with little wind (4 July 2016). An area of land was cleared of vegetation and a series of holes were dug. Each plant was carefully removed from its pot and lowered into a hole, the depth of which was adjusted to ensure plants sat flush with the soil surface and thus burning was even. Each plant was burned sequentially in a controlled way (see the electronic supplementary material, figure S1 for diagram of the set-up). After burning, plants were returned to their pots (with any ash on the soil surface removed to standardize any fertilizing effect) and were returned to the polytunnel in a randomized block design and watered.

Most plants had initiated regrowth 6 days after the burns. For each individual, the length of five regrowing leaves was measured using digital callipers 6 days after the burn and on a further four occasions (each being 5–7 days apart), with the final measurement taken 30 days after the burn. Daily average temperatures were slightly higher (2.7°C on average) in the polytunnel than outside (electronic supplementary material, figure S3), thus the plants experienced conditions similar to early spring without late season frosts. Ten of 47 M. racemosa and six of 44 T. triandra failed to resprout within 30 days, but mortality was not associated with treatment (ANOVA: p > 0.05).

After the last measurement was taken, all regrown biomass was removed from each plant and stored in sealed plastic bags containing moist tissue paper. Total regrown leaf area was then measured within 72 h using digital images and the program WinDIAS (Delta-T Devices, Cambridge, UK). The regrown material was subsequently dried at 70°C for 48 h and the dry mass was determined. SLA was calculated by dividing the regrown leaf area by the regrown dry mass.

A regrowth rate was calculated using the leaf length and air temperature data. Daily minimum and maximum temperature values were used to calculate growing degree-days (GDD,°C-day) for each time period between measurements using the equation:

$$\mathrm{GDD}=\left[\frac{T_{\mathrm{MAX}}+T_{\mathrm{MIN}}}{2}\right]-T_{\mathrm{BASE}},$$

where T_MAX and T_MIN are daily maximum and minimum air temperature, respectively. 10°C was selected for T_BASE (the base temperature for growth), which represents an intermediate value of published temperate and tropical grass T_BASE values [27–29]. Individual average rates of leaf length regrowth were calculated by fitting linear models to the cumulative leaf length and GDD data.

To convert the rate of leaf length regrowth to a rate of leaf biomass regrowth, the fresh length and dry mass of three leaves of each individual were measured. The relationship between leaf length and dry mass was determined for each species by fitting linear models to the log-transformed data. The fit of the models to the data was good (R² values > 0.87 for all species; electronic supplementary material, figure S2), and the slopes of these relationships were used to convert leaf length regrowth rate into leaf biomass regrowth (in mg GDD⁻¹).

(d). Statistical analysis

All analyses were performed using R (v. 3.4.1; [30]). The effect of fire frequency on plant traits (all biomass measurements, plant height, number of flowering tillers, regrowth rate, regrown leaf area and SLA) was determined by fitting a linear mixed-effects model to the data (‘lme4’ package; [31]). The fixed effects were ‘treatment’ (annual burn versus no burn) and ‘species’, and an interaction term between these effects was added if it improved the quality of the model (as indicated by the Akaike information criterion value). ‘Plot’ (i.e the replicate plot the plant was taken from) was added as a random effect. ‘Plant size’ (the sampled subset of above-ground biomass taken from each plant before being burned) was also added as a fixed effect for models in which the trait is probably influenced by plant size (number of flowering tillers, height and some regrowth traits after fire). To determine whether fire frequency was significantly influencing plant traits, this model was then compared to a grand mean model using a parametric bootstrapping method (‘pbkrtest’ package, [32]) with 10 000 simulated generations.

(e). DNA extraction and restriction-site associated DNA sequencing

For each species, total genomic DNA was extracted from leaf material for a subset of individuals (n = 3–5) per treatment (using the DNeasy Plant Mini Kit, Qiagen) and double-digested restriction-associated DNA libraries were built (following [33]). DNA extract (150–350 ng DNA) was double-digested using EcoRI and MseI after which barcoded adaptors were ligated to the EcoRI side and a common adaptor to the MseI side (following [34]). The 34 libraries were pooled with 62 libraries from different projects and the library pool was gel size selected (300–600 bp), purified (using QIAquick Gel Extraction kit) and paired-end sequenced on an Illumina HiSeq2500 lane at Edinburgh Genomics (University of Edinburgh, UK), following standard protocols.

Sequencing data were cleaned such that adaptor and primer sequences were removed and low-quality (less than 3) bases were trimmed from 3′ and 5′ ends, as well as bases with a minimum quality of 15 in a four-base sliding window. Reads shorter than 36 bases after trimming were removed. The library pool was de-multiplexed and the barcodes were removed (using the program ‘Stacks’, [35]). Nuclear reads were selected (see the electronic supplementary material, table S3 for details on chloroplast read removal) and used to de novo assemble nuclear Restriction-site associated DNA (RAD) sequencing loci in ipyrad (v. 0.7.2; [36]). A cluster threshold (sequence similarity for homology) of 0.85 was used and only loci with a cluster depth below 100 and less than 50% missing data were output.

One random single nucleotide polymorphism (SNP) with a minor allele count of three was extracted using VCFtools [37] from each of the assembled RAD loci. The SNPs were then used in a principal component analysis (R package ‘adegenet’; [38]) to test whether the two treatments were genetically distinct. An analysis of similarity was used to evaluate the significance of sample clustering (R package ‘vegan’; [39]). Signatures for genetic differences were further evaluated by calculating the genetic distances between the treatments for each species. Pairwise F_ST for each SNP were calculated in VCFtools, and an average F_ST across all SNPs was estimated. Jackknifing was used to evaluate the significance of average F_ST values, by randomly sampling individuals for each species with replacement into 1000 two-population comparisons and the average F_ST between the two populations calculated. Significance was evaluated as the percentage of the jacknifed F_ST values that were greater than or equal to the observed F_ST. The number of SNPs showing extreme F_ST values (greater than 0.8) was also assessed. The p-values for each SNP were calculated as the proportion of jacknifed F_ST values above the observed F_ST. Comparisons of observed and expected p-values were then used to evaluate the power of the genetic data to detect the differentiation between treatments.

SNPs were concatenated to an alignment and used to estimate a maximum-likelihood phylogenetic tree for each species using RAxML v. 8.2.11 [40] under a GTR + G substitution model and 100 fast bootstrap replicates were used to evaluate node support.

3. Results

(a). Traits in plants sampled from the field

Plants in the annual-burn and no-burn populations differed significantly in their initial (field-state) traits. In comparison to annual-burn plants, no-burn plants were taller (+29.6%; likelihood ratio test (LRT) = 35.1, d.f. = 1, p < 0.001) and had higher above-ground dry biomass (+33%; LRT = 62.5, d.f. = 1, p < 0.001; all model coefficients are given in the electronic supplementary material, table S4).

(b). Traits in a common environment

After plants had been reduced to a small, uniform number of tillers and grown in a common environment for 12 months, significant trait differences persisted between the no-burn and annual-burn populations. After this growth period, all plants were well established and had greatly increased in size (approx. 500–700% increase from the initial number of tillers, data not shown). Past fire frequency had a significant effect on the number of flowering tillers, with annual-burn plants having 50% more flowering tillers on average than no-burn plants (LRT = 11.11, d.f. = 1, p < 0.001; figure 1a). Annual-burn plants also invested significantly more of their total biomass below-ground (+23% on average; LRT = 19.98, d.f. = 1, p < 0.001; figure 1c) than no-burn plants. However, the treatment had no effect on total (above- and below-ground) dry biomass (LRT = 0.62, p = 0.43; figure 1b) or on plant height (LRT = 0.09; d.f. = 1, p = 0.77; model coefficients in the electronic supplementary material, table S5).

Figure 1.

Grass traits differ in populations exposed to contrasting fire frequencies. Annual-burn plants had significantly more flowering tillers (p < 0.001 (a)) than no-burn plants. Total dry biomass did not differ between treatments (b) but the allocation of biomass differed significantly (p < 0.001 (c)) with annual-burn plants investing a higher proportion of their dry biomass below ground in comparison to no-burn plants. After burning all individuals, there was no overall effect of treatment on total regrown dry biomass ((d) although, for two species, annual-burn plants did regrow significantly more biomass than no-burn plants; p < 0.05). After burning, there was also no overall effect of treatment on regrowth rate (e) or the SLA of regrown leaves (f) C., Cymbopogon; D., Digitaria; M., Melica; T., Themeda. *p < 0.05; **p < 0.01; ***p < 0.001.

(c). Traits after experimental fire

Populations with a different fire history behaved similarly after the experimental fire. All regrowth traits differed significantly between species (ANOVA: p < 0.05) but were unaffected by the fire frequency previously experienced in the field. Treatment had no significant effect on tiller-scale regrowth rate (LRT = 0.69, d.f. = 1, p = 0.41; figure 1e), regrown leaf area (LRT = 0.11; df = 1, p = 0.73) or regrown leaf SLA (LRT = 1.22, d.f. = 1, p = 0.27, figure 1f; all model coefficients in the electronic supplementary material. table S6). Overall, there was no treatment effect on regrown dry biomass across species (LRT = 0.46; d.f. = 1, p = 0.49) but significant intraspecific differences existed within C. pospischilii and T. triandra populations where annual-burnt plants regrew a larger above-ground biomass after fire in comparison to no-burn plants (p < 0.05; figure 1d).

(d). Population genetic analyses

The species M. racemosa was excluded from the genetic analyses because sample failure resulted in a low sample size. The number of assembled RAD loci and retained SNPs differed between the remaining species (table 1), as expected owing to variation in sequencing quality and depth, and divergence between individuals within each species. We observed deviation from the null hypothesis with an excess of SNPs with low p-values. At the whole genome level, there is, therefore, power in our genetic data to detect the differentiation between treatments (electronic supplementary material, figure S4). However, we found no significant clustering of individuals within species based on treatment (figure 2, analysis of similarity: C. pospischilii, p = 0.22; D. eriantha, p = 0.42; T. triandra p = 1), and none of the species showed a significant genetic difference (as estimated by average F_ST and number of F_ST outliers) between the treatments (table 1). Furthermore, with a few exceptions, the bootstrap support in the maximum-likelihood trees was generally low (less than 95), indicating that there is no significant phylogenetic clustering in the investigated species.

Table 1.

Assembly statistics for the analysis of genetic differences between populations of three savannah grass species under contrasting fire regimes (annual burn (AB) and no burn (NB)). (RAD, restriction-site associated DNA sequencing; SNPs, single nucleotide polymorphisms. Melica racemosa was removed from this analysis because the failure of some samples resulted in a low sample size.)

species	number of individuals (NB : AB)	number of assembled nuclear RAD loci	number of SNPs	average FST (p-value)	no. outlier FST (p-value)a
Cymbpogon pospischilii	10 (5 : 5)	21 649	5528	0.083 (0.074)	33 (0.136)
Digitaria eriantha	9 (5 : 4)	11 716	4611	0.095 (0.107)	22 (0.361)
Themeda triandra	8 (5 : 3)	40 031	9977	0.107 (0.076)	104 (0.122)

^aF_ST > 0.80.

Figure 2.

No genetic differentiation between grass populations under contrasting fire frequencies. Phylogenies and principal component analysis (PCA) plots reveal no clustering based on treatment for individuals of Cymbopogon pospischilii (a), Digitaria eriantha (b) and Themeda triandra (c). Analysis of similarity (anosim) results are indicated in the top left of the PCA plots. Values on nodes represent support evaluated with 100 bootstrap replicates (only support values greater than 50 are shown). PCAs are based on all SNPs.

4. Discussion

This study of grass functional traits under differing fire frequencies supports the hypothesis that fire has strong direct effects upon plant structure and function [2]. Previous studies have found evidence of a genetic basis for fire-related traits such as serotiny in pines [3] and flammability in a Mediterranean shrub [5]. However, we found no detectable genetic differences between plants that had experienced one or the other of the two fire regimes. Given the statistical power of our test, this is strong evidence that the selective pressure imposed by the past fire regime has not led to consistent genetic differences between the two treatments. Previous evidence of selection for fire-related traits is from obligate seeder species [5]. Such species are expected to experience stronger selection pressures for fire adaptations than resprouting species, such as those studied here, owing to their short and non-overlapping generations, and the higher cost of being burned. The absence of detectable genetic differentiation between the annual-burn and no-burn grasses may, therefore, be a result of their resprouting mode of persistence through fire. Alternatively, as grasses are wind-pollinated, gene flow among populations in the different fire treatments may have counteracted the effects of selection.

The trait differences observed between the contrasting fire treatments could potentially be explained by environmental effects carried over from the long-term treatments into the common environment, rather than by differential developmental responses to these treatments. However, this is unlikely for three reasons. Firstly, we washed the roots of the soil before potting the plants. Fire causes a release of nutrients into the soil and may result in increased soil fertility and faster plant growth in burnt areas [41,42]. However, any possible carry-over fertilization effects were limited by soil removal from the roots. Secondly, we found no significant difference in the total (above- and below-ground) biomass between the treatments after 1 year in a common environment, implying that any carry-over of internal resource stores from annually burnt plots did not enable plants to grow larger. Finally, the long period of growth in a common environment resulted in the initial transplanted biomass (five tillers) constituting only a small fraction of the final plant biomass (30–40 tillers). While efforts were made to limit variation in the age of plants removed from the burn plots (by selecting plants of a similar basal diameter), we are unable to directly determine age and whether this differed by treatment. Individuals in the frequently burned plots could be younger and therefore differ in allocation strategies. However, as plants were standardized by tiller number before being grown in a common environment, and plant size (above-ground dry biomass) was included as a fixed effect in appropriate analyses, there can only be age effects and not size effects. Furthermore, many age-related changes in allocation strategy and growth can be explained by size [43].

This study constitutes the first documentation, as far as we are aware, of plants having a primed response to fire, as found for some other abiotic stresses such as drought and inundation (e.g. [10,44,45]). Traits relating to flowering and growth allocation differed across all of the species according to prior fire exposure. These differences continued at least until the end of the study providing an example of a persistent phenotypic change but could be maintained for one or more generations as has been found in other cases of environmentally induced carry-over effects [46–48]. While fire is a major disturbance to plants, these developmental changes may mean current performance can be maximized through improved tolerance and/or responses to future fires, while avoiding the potential costs of maintaining a lifelong high-fire-suited phenotype. The priming mechanism is not addressed here, but such responses could involve epigenetic, metabolic, physiological or morphological changes [7–9]. The roles of epigenetic and chromatin modifications are particularly recognized in plant stress responses, and therefore, represent a likely mechanism for the traits differences seen between plants with and without prior fire exposure.

Similar to findings on the effect of crown fires on woody species [49–51], this study shows that fires cause trait divergence in the above- and below-ground allocation strategies of herbaceous plants. Annually burned plants invested more of their biomass below ground compared to no-burn plants, which probably equates to them having greater stored energy reserves to initiate and support early resprouting. The greater frequency of disturbance experienced by the annual-burn plants means they are regularly subjected to the near-complete removal of above-ground biomass and frequently encounter the competitive, post-fire environment. Thus, greater investment below ground results in a smaller proportion of total plant biomass being consumed by fire. However, a higher proportional investment in root biomass in annual-burn plants did not cause the faster initial resprouting rate compared to no-burn plants that we expected. Similarly, SLA, a trait indicative of resource acquisition [52], did not differ between annual-burn and no-burn plants. The recurrent fire could instead select for more vigorous resprouting (i.e. greater resprouted biomass, as found for two of the four species examined here, [53]) rather than a faster rate of resprouting at the tiller scale. Such differences could be owing to a greater number of resprouting tillers rather than a faster rate of regrowth per tiller. Interestingly, the two species in which annually burnt populations regrew significantly more biomass after fire than unburnt populations both belong to the monophyletic group Andropogoneae. Similarly, in a previous comparative analysis of grass fire responses of different lineages [54], regrowth was stimulated by fire only in the Andropogoneae species studied. In fire-prone areas, the rapid creation of a large, flammable fuel load by these shade-intolerant species may aid in the maintenance of an open canopy by burning off standing dead and woody biomass [55].

Grasses showed plasticity in reproduction, dependent upon previous fire experience, that is likely to be adaptive in fire-prone environments. A history of high-fire frequency favours grass traits relating to vigorous post-fire recruitment, with the heightened flowering in annual-burn plants suggesting that flowering and seed production is stimulated by fire. Fire-stimulated flowering has been demonstrated in other savannah grass species [56], but this study represents, to our knowledge, the first documentation of fire having a priming effect on grass flowering. In many perennial grasslands, successful recruitment is a rare event [19,20], but fire may enhance seedling establishment through reduced below-ground competition with resprouters [57].

Fire-prone savannahs are vulnerable to global change drivers [58], with fire regimes changing in frequency and intensity [59]. As fire behaviour influences plant traits, a consequence of such changes may be transformed community functional diversity. However, the finding here that grasses may have a primed response to fire, resulting in adaptive trait changes, may lead to a community composition that is more resistant to future fire regime changes [8].

Data accessibility

Trait data: individual-level trait values uploaded available in the Dryad Digital Repository: https://doi.org/10.5061/dryad.7qr55jn [60]. Sequence data: All sequences are available in the Sequence Read Archive (SRA) database under the accession SRP126993.

Authors' contributions

All authors contributed to the design of the study. K.J.S., J.K.O. and B.S.R. generated the data. K.J.S., J.K.O. and C.P.O. analysed and interpreted the data. K.J.S. wrote the manuscript with all authors contributing critically to drafts.

Competing interests

We declare we have no competing interests.

Funding

Research support was provided by a Natural Environment Research Council studentship (1371737) to K.J.S. and a European Research Council grant to support J.K.O. (ERC-2014-STG-638333). Edinburgh Genomics is partly supported through core grants from NERC (R8/H10/56), MRC (MR/K001744/1) and BBSRC (BB/J004243/1).