The ecological drivers of growth form evolution in flowering plants

Abstract

1. In flowering plants (angiosperms), the herbaceous habit has evolved repeatedly from the ancestral woody state and herbs evolved repeatedly back to woody plants. Yet, how common these transitions were and which ecological conditions promote the herbaceous habit is poorly known. Several hypotheses exist, postulating an advantage of the herbaceous growth form to better cope with frost, drought, fire and shade and in allowing a fast life strategy, but their evaluation has been hitherto limited and support equivocal. We aim to evaluate these hypotheses by testing the difference between woody plants and herbs for a set of variables related to these hypotheses.
2. We compiled and integrated data for up to 21,581 species representing 359 families from public databases. We estimated the minimum number of evolutionary transitions between both growth forms. We assembled data on frost, drought, fire and shade tolerances, clonality and specific leaf area and we tested individual hypotheses by comparing herbaceous and woody angiosperm growth forms globally and within selected biomes and clades using phylogenetic comparative analyses.
3. We found 1656 evolutionary transitions from woody towards herbaceous growth form and 2111 transitions in the opposite direction. In agreement with our expectations, herbs were more tolerant to frost and shade than woody plants and had higher specific leaf area. However, the growth forms did not differ in their fire tolerance and clonality. Furthermore, contrary to our expectation, woody plants were more drought tolerant than herbs. The majority of the differences were robust to the choice of biome or clade.
4. Synthesis. Both herbaceous and woody habits evolved many times making the evolution of growth forms a well-replicated event and suggesting that conditions favourable for either of the growth forms emerge often and plants respond to them. Apart from standard explanation by low temperatures, the success of herbs was likely enabled also by biotic interactions—by their fast life strategy, which is beneficial in seasonal and early successional habitats, and by their ability to tolerate shade.

1 INTRODUCTION

Today, about half of all species of flowering plants (angiosperms) are herbaceous (FitzJohn et al., 2014) often ecologically dominating their habitat, for instance in tropical savannas, temperate grasslands and alpine areas (Moles et al., 2009). Herbs are phylogenetically derived in angiosperms, since the most recent common ancestor of all angiosperms was likely a woody plant (Smith et al., 2010). Evolutionary shifts towards a herbaceous growth form have occurred repeatedly in multiple distantly related evolutionary lineages as well as inverse transitions back to the woody state (Judd et al., 1994; Smith & Donoghue, 2008). Since the genetic changes required for transitions between both growth forms are probably simple (Groover, 2005), the repeated evolutionary transitions and subsequent diversifications suggest adaptive values of both growth forms. Yet, the ecological driver(s) of their evolution and persistence, which enabled them to proliferate alongside each other, remains little examined and largely unknown.

We understand as an ecological driver of the evolution of a growth form a condition under which differences between the two growth forms led to a competitive advantage of one growth form over the other. Herbs and woody plants differ in many ways including, among others, their position on the slow–fast continuum (which describes the trade-off between high growth and rapid reproduction vs. slow growth and limited fecundity; Díaz et al., 2016). Herbs are on the faster and woody plants on the slower side of this spectrum (Salguero-Gómez et al., 2016; Silvertown et al., 1993). Both growth forms also differ in their stem construction, which is perennial for woody plants and short-lived or annual for most herbs (Zanne et al., 2014).

The ecological drivers of the evolution of both growth forms could be biotic or abiotic. An abiotic condition is likely to be disturbance or stress related since woody plants usually dominate in terms of biomass under disturbance-free or stress-free conditions, while many herbs thrive under opposite conditions (Bucini & Hanan, 2007; Dantas et al., 2016). Based on the ability of herbs to repeatedly lose and regrow their above-ground biomass, freezing temperatures have long been proposed to be the key driver of their evolution (Sinnott & Bailey, 1915). This proposition is in agreement with the observed increase in the proportion of herbs with increasing latitude (i.e. increasing exposure to freezing temperatures, Moles et al., 2009). However, this hypothesis has been contested by phylogenetic reconstructions showing that in 58% of the cases the evolutionary transition to herbaceous habit preceded the exposure to freezing temperatures (Zanne et al., 2014).

Apart from freezing temperatures, the ability of herbs to renew above-ground biomass is likely to be beneficial under other disturbances that primarily damage above-ground biomass and thus suppress woody plants. One such condition proposed as a potential driver of the evolution of herbs is drought (Sinnott, 1916). Herbs could avoid drought periods by persisting as dormant below-ground structures (or as seeds) and thus gain an advantage in comparison with woody plants (Sinnott, 1916). This hypothesis has been suggested for Echium, where the evolution of a herbaceous growth form has been associated with dry habitats (García-Maroto et al., 2009). Another condition primarily damaging above-ground biomass is fire. As a result of their ability to renew above-ground biomass, herbs can deal better with recurrent fire regimes than woody plants. Their advantage in landscapes with frequent fires is well-documented (Bond et al., 2005; Kraaij et al., 2017; Robertson & Hmielowski, 2014) and fire likely played a major role in the evolution of angiosperms (Bond & Midgley, 2012).

Besides the abiotic drivers, the advantage of one growth form over the other may be due to biotic interactions (Klimeš, Koubek, et al., 2021). Herbs actively compete for light with young woody plants of comparable stature both in understorey and in open landscape, and the result of this competition can strongly influence habitat structure (Pausas & Bond, 2020). Two characteristics related to lower carbon demand of herbs could give them an advantage in interactions with woody plants. The first is tolerance to shaded conditions. In general, herbs need less carbon than woody plants to create above-ground structures of the same stature (Graves et al., 2006) leading to a higher proportion of biomass invested in leaves (Niinemets, 2010). Nevertheless, shade tolerance of herbs seems to be very variable, ranging from shade intolerant ephemerals and gap specialists in tropical forests, to shade tolerant understorey vegetation in temperate forests (Augspurger & Salk, 2017; Bierzychudek, 1982; De Keersmaeker et al., 2011; Dirzo et al., 1992). The second is a high growth rate of inexpensive (less carbon demanding) structures, which would be an advantage for herbs that compete with young woody plants for light. Indeed, the relative growth rate of herbs was found to be higher than in trees (linked to differences in lignin formation; Poorter, 1989) and their growth is likely more plastic allowing them to effectively respond to surrounding conditions (Klimeš, Koubek, et al., 2021) which might further increase their advantage when shaded. Fast growth can even compensate for lower shade tolerance in some herbaceous groups such as ephemerals, which complete their life cycles before late-spring canopy shade increases (Augspurger & Salk, 2017). Also, the life cycle of herbs is mostly dominated by growth in contrast to woody plants for which the prevailing vital rate is survival (Salguero-Gómez et al., 2016; Silvertown et al., 1993).

Here, we use global data on angiosperm growth forms, traits and environmental variables related to their distribution to estimate the number of evolutionary transitions between both growth forms and to test predictions arising from five hypotheses on the adaptive value of herbaceous growth. Several studies have reported transitions within angiosperms (in both directions; Judd et al., 1994; Smith & Donoghue, 2008; Beaulieu et al., 2013; Lens et al., 2013), but to our knowledge no large-scale quantification has yet been attempted. We fill this gap by estimating the number of transitions between herbs and woody plants in both directions.

Furthermore, using phylogenetically informed comparison of present day herbaceous and woody species in their differential ability to deal with frost, drought, fire and shade, we infer to what extent these presumed drivers are associated with parallel evolutionary events (assuming their constancy - uniformitarianism; Hutton, 1788). Specifically, we test, in relation to woody plants:

Hypothesis 1 (H1).Herbs are overall more tolerant to freezing. We use three proxies and have the following expectations about them: fewer frost-free days in the geographical range of herbs, lower minimum temperature in the range of herbs and higher freezing exposure (encountering freezing temperatures) in the geographical range of herbs.

Hypothesis 2 (H2).Herbs are overall more tolerant to drought. We expect them to have higher measured drought tolerance than woody plants.

Hypothesis 3 (H3).Herbs are overall more tolerant to fire. We expect higher recorded fire tolerance for herbs and also higher clonality (clonality is the ability to produce offspring as vegetative rooting units), a trait beneficial under such disturbances since it facilitates regrowth or vegetative reproduction from below-ground buds on clonal organs (Bellingham & Sparrow, 2000; Klimešová & Klimeš, 2003).

Hypothesis 4 (H4).Herbs are overall more shade tolerant. We expect higher recorded shade tolerance, a characteristic that summarises various light-harvesting strategies (these span from leaf physiology and morphology to architecture of plants; Niinemets, 2010), for herbs.

Hypothesis 5 (H5).Herbs have overall a higher growth rate of inexpensive structures. The cost of leaves is traditionally assessed by specific leaf area (SLA) which is often positively related to growth rate, namely at early ontogenetic stages (Reich et al., 1992). Therefore, we expect higher SLA for herbs.

Testing hypotheses on these drivers requires comprehensive datasets to average out numerous evolutionary contingencies in individual lineages or ecological groupings. Although we use tolerance variables, these are based on plant occurrences and therefore describe tolerances in the broad sense, including the escape strategy of herbs below-ground or in the form of seeds. We compare species that are by consensus classified into either herbaceous or woody growth form, as there are multiple definitions of woodiness and unequivocal classification of some species is not possible (Schweingruber & Büntgen, 2013). We also focus on perennial plants, since the adoption of an annual life cycle represents a distinct herbaceous strategy from that of perennials, with its own drivers and tolerances, and may therefore obscure the direct comparison of the two growth forms.

2 MATERIALS AND METHODS

We combined information on plant growth form (woody/herbaceous) of angiosperm species globally with trait information and occurrence based environmental variables on tolerances to abiotic conditions (tolerance to frost, drought and fire), two variables related to biotic interactions (specific leaf area and shade tolerance) and clonality.

We compiled woodiness data from Engemann et al. (2016), Zanne et al. (2014) and the CLO-PLA database (Klimesova et al., 2017) assigning herbs and vines as herbs; trees, shrubs and lianas as woody and excluding ambiguous species and epiphytes. To ensure that repeated transitions between both growth forms took place in the evolution of angiosperms, and thus our comparison of both growth forms is informative about possible drivers of their evolution (Maddison & FitzJohn, 2015), we estimated the number of transitions in four major clades of eudicots—Fabidae, Malvidae, Campanulidae and Lamiidae. We used an ‘ARD’ (all rates different) continuous-time reversible Markov model for the evolution of growth forms and then simulated stochastic character history. We also calculated the parsimony score which is the minimum number of transitions. We used the simulated history to estimate the proportion of primary woody species in each of the clades.

We obtained data on frost tolerance in the form of (a) ‘frost-free days’, which is a variable from the PLANTS database (USDA NRCS, 2017) defined as the minimum average number of frost-free days within the plant's known geographical range; (b) ‘minimum temperature’, again from PLANTS database, defined as the lowest temperature recorded in the plant’s historical range; and (c) ‘freezing exposure’, which is a binary variable denoting if a species encounters freezing temperatures in its area of distribution—from Zanne et al. (2014). We obtained data on drought tolerance from PLANTS and The Ecological Flora Database (Fitter & Peat, 1994). These tolerances are classified into ordered categories (e.g. low, medium, high), which we substituted by integer values. As these categories probably differ between datasets, we combined the datasets using a phylogenetic linear model with data on overlapping species (for details see supplementary methods). The model was used to predict values on the scale of the PLANTS database for species missing in this dataset and present in other datasets. In the analyses, we used not only the combined dataset but also the largest original dataset (which was from PLANTS). We obtained data on fire tolerance from PLANTS and substituted its categories by integer values in the same way as for drought tolerance. We compiled clonality as a binary trait from multiple sources (see supplementary methods). We considered a species to be clonal if at least one source documented it as clonal. We obtained shade tolerance from PLANTS (with states tolerant, intermediate and intolerant), Wirth and Lichstein (2009) and The Ecological Flora Database. We processed the data and combined the datasets in the same way as for drought tolerance. We obtained SLA from the BIEN database (Enquist et al., 2009). The assembled dataset does not allow a combined analysis of all variables due to low overlap of species for individual traits. Therefore, to uncover differences between herbs and woody plants in individual variables, we analysed each variable separately. Variables SLA and number of frost-free days were transformed using square root prior to the analyses to satisfy the assumption (approx. Normal distribution of residuals) of the model.

To exclude annual plants from the analyses, we assembled information about plant life span from multiple sources (see supplementary methods). A species was marked as annual if at least one source documented it as annual (or belonging to a category which could contain annual species—e.g. ‘annual and biennial’). Therefore, the category annuals in our life span data covers most true annuals and probably also some perennials. To explore the effects of this classification on results, we ran analyses with and without the exclusion of annuals. At the same time, we explored the effect of exclusion of (woody as well as herbaceous) epiphytes.

Global trait data suffer from uneven sampling of species (Kattge et al., 2020). Stratification of global data into smaller units (which would be more evenly sampled and would not differ so much in the proportion of herbaceous species) could reduce potential biases and show robustness of results. Therefore, we stratified the data according to biomes. Consequently, we used biome as a covariate (we only included biomes with more than 20 herbs and woody species in all analyses) and also explored separately two biomes with a high number of species (temperate mixed broadleaf forest biome, further referred to as ‘temperate forest’; tropical and subtropical coniferous forest biome, further ‘tropical forest’). We assigned species to one of the world's biomes (Olson et al., 2001) based on the information of species distribution obtained from the Global Biodiversity Information Facility (GBIF) data portal (see supplementary methods; GBIF, 2021). Exploration of differences within biomes has an advantage of more even geographical completeness of observations but with the disadvantage of a lower and non-random cover of phylogeny.

Since the data we assembled are about species and these species share varying degrees of evolutionary history, they do not represent independent observations from a statistical perspective (Felsenstein, 1985). To account for their phylogenetic relatedness, we used a phylogenetic linear model which assumes that covariance of residuals corresponds to the proportion of shared evolutionary history between species. We used the largest up-to-date phylogenies ALLOTB and GBOTB (Smith & Brown, 2018), which combine genetic data from GenBank with phylogenetic data and taxa from the Open Tree of Life project (Hinchliff et al., 2015). The GBOTB phylogenetic tree contains only species for which genetic data were available.

To explore the differences between herbs and woody plants, we modelled individual variables (Response) using a phylogenetic linear model:

$$$$ Response={\beta}_0+{\beta}_1\times Biome+{\beta}_2\times Herb+{\beta}_3\times Biome\times Herb+\varepsilon . $$$$

$$$$ \varepsilon \sim Multivariate\ Normal\left(0,\lambda \times phy\right). $$$$

where β are the vectors of regression coefficients, Biome is the biome to which the species was assigned, Herb is a binary variable denoting whether the species is herbaceous, phy is the phylogenetic variance–covariance matrix and λ is a parameter multiplying off-diagonal elements of phy (so called Pagel's lambda (Pagel, 1999) used to account for phylogenetic relatedness to the appropriate degree (Freckleton et al., 2002; Revell, 2010)). The response variables were standardised to a mean of 0 and standard deviation of 1 prior to the analysis. Model adequacy was checked visually. For binary response variables (exposure to freezing temperatures and clonality) we used phylogenetic logistic regressions with the same predictors and with logit link function (Ives & Garland, 2010). We standardised estimates from the logistic regression to be in units of standard deviations of the response variable and thus be approximately comparable to estimates for non-binary responses. We tested all predictors (whether their effect is different from zero) using Wald tests.

In large-scale phylogenetic models, trait evolution is likely to differ in different branches of the tree, meaning that estimation of a single phylogenetic signal might not be appropriate. To assess the robustness of our results to the assumption of a consistent phylogenetic signal across branches, we ran the model separately for phylogenetic subsets—the orders Fabales and Lamiales, for which we had the best data coverage (for number of species see Table S1) and for larger clades of Rosidae and Asteridae—Fabidae, Malvidae, Campanulidae, Lamiidae (each encompassing seven or eight orders; APG IV, 2016). In these cases, biomes were not included in the model due to a low number of observations.

In summary, we ran the default model where we excluded annuals and epiphytes, used the ALLOTB phylogenetic tree and included biomes as a predictor (and interactions between herbaceousness and biomes). Then to explore whether our results were sensitive to these choices, we ran three models, which differed from the default model: the first model used the GBOTB phylogenetic tree instead of ALLOTB, the second included annual herb species and epiphytes, and the third excluded biomes as predictors. Finally, we ran the model for subsets of our data—two biomes (temperate and tropical forests) and six clades (orders Fabales and Lamiales, Fabidae, Malvidae, Campanulidae and Lamiidae)—using ALLOTB phylogenetic tree, excluding annuals and epiphytes and not using biomes as a predictor.

We carried out all analyses in r (version 4.0.0; R Core Team, 2020) using the packages nlme (version 3.1-148; Pinheiro et al., 2019) and phylolm (version 2.6.2; Ho & Ané, 2014) for phylogenetic linear models and logistic regressions, speciesgeocodeR (Töpel et al., 2017) to assign species to biomes, msm to calculate confidence intervals for visualisation based on the delta method (version 1.6.8; Oehlert, 1992; Jackson, 2011), aod for Wald tests (version 1.3.1; Lesnoff & Lancelot, 2021), phytools (version 0.7-80; Revell, 2012) for Markov model of evolution and stochastic history simulation and bien (version 1.2.4; Maitner, 2017) to access the database of the same name. Trait values and environmental variables, apart from traits from the BIEN and CLO-PLA databases, were accessed via the TRY database (Kattge et al., 2020).

3 RESULTS

In four major clades of eudicots, we found a total of 1656 evolutionary transitions from woody towards herbaceous growth form and 2111 transitions in the opposite direction (Figure 1). The parsimony score indicated a minimum of 2747 growth form transitions (in both directions combined and given the woodiness and phylogeny information). The proportion of primary woody species in these clades was estimated to be 71% for Fabidae, 78% for Malvidae, 28% for Campanulidae and 9% for Lamiidae.

FIGURE 1

Reconstructions of transitions between growth forms in the evolution of four major eudicot clades. Simulated stochastic histories are visualised. Diagrams in the upper-right corner show estimated number of transitions between both growth forms and parsimony score in the middle (minimum possible total number of transitions). Number of species used in each reconstruction is in the upper-left corner

We explored differences between herbs and woody plants in variables related to the five hypotheses on the ecological conditions that favour herbaceous growth (H₁–H₅). Differences were assessed in eight variables, each of them available for 1411 up to 21,581 species representing 132 up to 359 families (from 438 known families of angiosperms; Figure 2). Results were robust to selection of higher clades—Fabidae, Malvidae, Campanulidae and Lamiidae (for results see Table S2).

FIGURE 2

Phylogenetic representativeness of data about variables used for evaluation of the hypotheses of evolution of herbaceous habit (H₁–H₅). Non-dated family tree (Gastauer & Meira-Neto, 2017) with colour marks for families present in the dataset for each variable is visualised. Variables are ordered (from Centre) as in Table 1 and Figure 3. ‘P.’ stands for the PLANTS database (USDA NRCS, 2017). Adapted function trait.plot from package diversitree (FitzJohn, 2012) was used to plot the figure

Concerning frost tolerance proxies (H₁), we found that herbaceous plants were exposed to a lower number of frost-free days (the estimate was 205.3 days for woody plants and 132.7 days for herbs, p-value < 0.001), had a lower minimum temperature within their distributional ranges (−18.8°C for woody plants and −29.6°C for herbs, p-value < 0.001), and a higher freezing exposure of species (45.8% for woody plants and 55.2% for herbs, p-value < 0.001), all in agreement with our expectations. The exceptions were the model focused on temperate forests and the model of only Fabales where the differences in minimum temperature variable between both growth forms were not significant (p-values = 0.432 and 0.096 respectively; Table 1, Figure 3).

TABLE 1. Differences between herbaceous and woody plants in selected variables related to hypotheses of the evolution of herbs. Arrows show the direction of the difference (up: Higher values of the variable for herbs than for woody plants). The numbers are p-values for the difference being zero (based on Wald tests). The first three columns represent the tests when biome was included in the analysis as an explanatory variable. The significant difference between herbs and woody plants among biomes (an interaction of growth form and biomes) is represented by an asterisk (*). Significant results that are in agreement with our expectations are highlighted in blue. Significant results that contrast with our expectations are highlighted in red. Default model means the model based on the ALLOTB phylogenetic tree (Smith & Brown, 2018), without annuals and epiphytes and with biomes as predictors. Exp., expectation; GBOTB, phylogenetic tree (Smith & Brown, 2018); PLANTS, database (USDA NRCS, 2017); SLA, specific leaf area

Variable	Hypo-thesis	Exp.	Default model	GBOTB	With annuals	Without biomes	Temperate forest	Tropical forest	Fabales	Lamiales
Frost-free days	H1	↓	↓ <0.001*	↓ <0.001*	↓ <0.001*	↓ <0.001	↓ <0.001	↓ <0.001	↓ 0.004	↓ <0.001
Minimum temperature		↓	↓ <0.001*	↓ <0.001*	↓ <0.001*	↓ <0.001	↓ 0.432	↓ <0.001	↓ 0.096	↓ <0.001
Freezing exposure		↑	↑ <0.001*	↑ <0.001*	↑ <0.001*	↑ <0.001	↑ <0.001	↑ <0.001	↑ <0.001	↑ <0.001
Drought tolerance	H2	↑	↓ <0.001*	↓ <0.001	↓ <0.001*	↓ <0.001	↓ <0.001	↓ <0.001	↓ 0.020	↓ 0.004
Drought tolerance (PLANTS)		↑	↓ <0.001	↓ <0.001*	↓ <0.001	↓ <0.001	↓ <0.001	↓ <0.001	↓ 0.007	↓ 0.018
Fire tolerance	H3	↑	↑ 0.898*	↓ 0.881*	↓ 0.248*	↑ 0.872	↓ 0.010	↑ 0.709	↑ 0.541	↓ 0.854
Clonality		↑	↓ 0.085*	↓ 0.023*	↓ 0.007*	↓ <0.001	↓ 0.087	↓ <0.001	↓ 0.028	↑ 0.059
Shade tolerance	H4	↑	↑ <0.001	↑ <0.001	↑ <0.001	↑ <0.001	↑ 0.562	↑ 0.027	↑ 0.445	↑ <0.001
Shade tolerance (PLANTS)		↑	↑ 0.022	↑ 0.044	↑ 0.018	↑ 0.003	↓ 0.600	↑ 0.770	↓ 0.327	↑ 0.003
SLA	H5	↑	↑ <0.001*	↑ <0.001*	↑ <0.001*	↑ <0.001	↑ <0.001	↑ <0.001	↑ 0.246	↑ <0.001

FIGURE 3

Standardised effects of herbaceous habit for selected variables related to our hypotheses of the evolution of herbs. Estimates and 95% confidence intervals (computed using the delta method via package msm version 1.6.8; Oehlert, 1992; Jackson, 2011) are visualised for the default model (without annuals and with ALLOTB phylogenetic tree). Thin lines denote individual biomes and thick lines denote average effects. Blue colour denotes agreement with the hypothesis, red colour denotes disagreement and black colour is for no significant difference between herbs and woody plants. Numbers at the left side show the number of species used in each analysis. Visualised and analysed biomes are (from top): Tropical and subtropical moist broadleaf forests; tropical and subtropical dry broadleaf forests; tropical and subtropical coniferous forests; temperate broadleaf and mixed forests; temperate coniferous forests (and for ‘specific leaf area’ and ‘freezing exposure’: Boreal forests; tropical and subtropical grasslands, savannas and shrublands; temperate grasslands, savannas and shrublands; only for ‘freezing exposure’: Flooded grasslands and savannas; montane grasslands and shrublands; tundra; Mediterranean forests, woodlands and scrubs)

We found a lower tolerance to drought (H₂) in herbs compared to woody plants (p-value < 0.001), contrasting with our expectation (Table 1). The only exception was Malvidae where the tolerance of both growth forms was not significantly different (p-value = 0.182; Table S2).

Concerning the predicted higher fire tolerance of herbs (H₃), we found no difference in most models in fire tolerance between the growth forms (p-value = 0.898, with exception of temperate forests where woody plants had higher fire tolerance, p-value = 0.010), and even the direction of the effect of growth form differed among the analyses (Table 1).

For clonality, the difference between the growth forms was variable, in most models contrasting with our expectation of herbs being more clonal than woody species (78.9% for woody plants and 76.9% for herbs, p-value = 0.085; Table 1).

For shade tolerance (H₄), our expectation of herbs being more tolerant than woody species was confirmed in most of the analyses (p-value < 0.001). The three exceptions were for Fabales, the temperate forest biome and Malvidae, where the differences between the growth forms were not significant but herbs had still higher estimated tolerance (p-values = 0.445, 0.562 and 0.193 respectively). Furthermore, when analysing the PLANTS data subset, our expectation was confirmed for most of the models (p-value = 0.022; Tables 1, Figure S2).

Our expectation of higher SLA values (H₅) for herbs than for woody species was supported by all analyses (estimated 10.0 m²/kg for woody plants and 19.6 m²/kg for herbs, p-value < 0.001), with the exception of Fabales (p-value = 0.246; Table 1).

4 DISCUSSION

Our results show that there have been more than 1600 evolutionary transitions towards the herbaceous habit and more than 2100 transitions in the opposite direction, suggesting that the evolution of both growth forms is a well-replicated event. We identified five hypotheses explaining the adaptive value of the herbaceous habit and we tested whether differences between herbs and woody plants correspond to the expectations of these hypotheses. Using global data, we found that herbaceous species are more tolerant to frost, have higher shade tolerance and higher specific leaf area than woody species in agreement with the hypotheses. Both growth forms do not differ in fire tolerance and clonality and woody species are more drought tolerant than herbs which contrasts with the hypothesis.

Transitions between both growth forms were inferred to be common throughout the evolutionary history of angiosperms, which confirms previous suggestions about common transitions back to the woody growth form (Lens et al., 2012, 2013; Nürk et al., 2019). Since we do not have data for all angiosperms, our estimated number of transitions is certainly lower than the actual number of transitions and serve as a lower estimate. Although ancestrally woody, only a minority of plants in angiosperms kept this primary woody state—many of them are herbaceous or are secondary woody (woody species that evolved from herbs). Secondary woody species are known to have different woody structure than primary ones (Dulin & Kirchoff, 2010) but their ecology, especially in terms of perenniality of above-ground biomass which is relevant for our hypotheses, is similar.

Frost tolerance as the universal driver of the evolution of a herbaceous habit was rejected by a previous study, which showed that many plant lineages evolved a herbaceous habit before they encountered freezing temperatures (Zanne et al., 2014). Also, young plants of both growth forms deal similarly well when exposed to freezing temperatures (Klimeš, Weiser, et al., 2021). Our results show that herbs do have consistently higher frost tolerance than woody plants. This is unsurprisingly in agreement with the current global distribution of herbs—their proportion (in terms of species richness) increases towards the temperate regions and decreases towards the tropics (Moles et al., 2009). However, this could also result from a stronger exclusion of woody plants during cooler periods, such as Pleistocene ice ages, followed by dispersal limitation—slower dispersal of woody plants facilitated further radiations and expansions of herbaceous plants in temperate regions (Murphy et al., 2016; Svenning & Skov, 2007; Willson et al., 1990). Also, the slightly stronger effect of growth form on the frost-free days (less frost-free days for herbs) variable compared to the minimum temperature variable in our dataset suggests that seasonality, which is probably better captured by the first variable, promotes advantage of herbs (as suggested by Sinnott, 1916).

We hypothesised a higher drought tolerance as advantageous for herbs due to the negative impact of drought on above-ground biomass, which is more easily renewed in herbs than in woody plants. In contrast to our expectation, we found woody plants to be more drought tolerant than herbs. Woody plants on average have longer and thicker roots than herbs (Valverde-Barrantes et al., 2020) which enable them to reach deeper sources of below-ground water when soil moisture is deficient (although woody seedlings do not have this advantage). Our results suggest that the ability of many herbs to escape short-term drought in a dormant stage, below-ground, does not balance out this advantage. This is in agreement with the observation of secondary woody species being associated with dry localities (Dória et al., 2018; Lens et al., 2013; van Huysduynen et al., 2021) and with vulnerability to drought embolism being negatively correlated with plant woodiness (Dória et al., 2019).

Fire tolerance of growth forms did not differ in our dataset. However, the fire tolerance variable we used is defined as an ability of plants to resprout or regrow after fire and as such is likely studied mainly in plants that encounter fire. Nevertheless, this would probably lead to an underrepresentation of woody plants (in case they would be less fire tolerant) in the data, which was not the case for most biomes (with exception of Tropical and subtropical moist broadleaf forests; Figure S1). Due to efficient adaptations for resprouting or, in the case of woody plants, in bark traits, both growth forms are able to survive conditions with frequent fire (Keeley et al., 2011; Zizka et al., 2014). However, the mechanism likely differs between growth forms—woody plants, especially trees, thanks to their size and bark can tolerate many fire events when mature (Barlow et al., 2003). On the other hand, herbs rely on resprouting ability and avoidance thanks to their fast life cycle (Keeley et al., 1981). It is also possible that differences between woody and herbaceous plants are restricted to specific biomes (tropical and subtropical moist broadleaf forests in our dataset—Figure 3 or savannas: Giroldo et al., 2017) or are more subtle and depend on fire frequency or intensity (Robertson & Hmielowski, 2014), which were not captured by our data.

Clonality was included in our analyses as a trait expected to be beneficial under fire disturbance. It could also be beneficial under other conditions damaging above-ground biomass, such as herbivory (Bellingham & Sparrow, 2000; Klimešová & Klimeš, 2003). Consequently, we hypothesised that herbs would be more clonal than woody plants, which was not confirmed. Previous studies have shown that the relationship between clonality and disturbance is quite complex, however, as clonality is probably not beneficial when the disturbance is severe or when the plants affected by disturbance are young (Herben et al., 2018; Martínková et al., 2020). Furthermore, our dataset contained a much higher proportion of clonal species (in some biomes over 80%) than expected, suggesting bias towards clonal plants (Figure S3; 60% of plants are clonal in central Europe where clonality is well studied; Klimesova et al., 2017). Clonality in woody plants is studied mainly in relation to resprouting after disturbance (see supplementary methods) and thus may often not be active clonality (i.e. some plants marked as clonal do not clonally reproduce unless disturbed). Although the dataset might not fully represent plant clonality, our analyses are based on the best large-scale data we currently have. In many biomes, the proportion of clonal species is higher for woody plants than for herbs, which was opposite to our expectation (Figure S3). Therefore, we consider our results as evidence against the proposed difference in clonality (but there are floras where the proportion of clonal plants is considerably higher for herbs than for woody plants; see Aarssen, 2008).

We expected greater shade tolerance and inexpensive leaves with faster growth (traits related to SLA; Reich et al., 1992) in herbs than in woody plants. Indeed, our results confirmed herbs to be more shade tolerant and to have higher specific leaf area. A notable exception was Fabales, in which we found no significant differences. Many Fabales have the capacity to obtain nitrogen via symbiosis (Fabaceae), and hence may be subject to different cost–benefit and carbon allocation strategies than the species of most other plant families. The difference in shade tolerance was generally weaker than the difference observed in frost tolerance but consistent among biomes (Figure 3; Figure S2). Smaller stature and faster growth of herbs in comparison with woody plants enable them to occupy early successional habitats. In combination with greater shade tolerance, herbs can effectively compete with young trees and shrubs in the understorey (Tsvuura et al., 2012). Small size of herbs might be an advantage in these habitats as it can be associated with weaker competition due to better niche differentiation (facilitated by e.g., greater growth plasticity of herbs; Klimeš, Koubek, et al., 2021) in comparison with larger woody plants (physical-space-niche hypothesis; Aarssen et al., 2006). Small size of herbs also enables larger proportional investment into reproduction solidifying their success in macroevolution (fecundity allocation premium hypothesis; Aarssen et al., 2006; Dombroskie & Aarssen, 2010).

Our evaluation of differences between woody plants and herbs is limited by data availability and quality, although tests on subsets of the data (Table 1; Table S2) show that the results are robust. This is of particular concern in some variables studied here, such as clonality and additional potentially relevant variables, such as herbivory tolerance. On the other hand, for some variables (e.g. frost and shade tolerances) the sampling has much better coverage and hypotheses can be effectively evaluated, as we have demonstrated.

5 CONCLUSIONS

We show that evolutionary transitions between herbs and woody plants were common in angiosperms and that transitions can be linked to environmental conditions. We found that herbaceous and woody life-forms differ in frost, drought and shade tolerances and in SLA. In particular, woody plants are unexpectedly more drought tolerant than herbs, implying that drought is unlikely to be an ecological condition promoting the herbaceous habit, but it probably promotes the woody one. Based on these findings, we suggest frost and the competition for light expressed in SLA (with herbs having an advantage over young woody plants) to be the most likely drivers associated with the evolution of growth forms. Our results pave the way for further experimental research to test for an ecological advantage of one growth form over the other.

ACKNOWLEDGEMENTS

We thank members of the Antonelli Lab and Jitka Klimešová for discussions and feedback and Rhian Smith for scientific editing. This work was supported by the Czech Science Foundation [19-13231S], Ministry of Education, Youth and Sports of the Czech Republic [LTT20003] and long-term research development project of the Czech Academy of Sciences [No. RVO 67985939]. A.Z. is thankful for funding by iDiv via the German Research Foundation (FZT-118, DGF), specifically through sDiv, the Synthesis Centre of iDiv. A.A. acknowledges financial support from the Swedish Research Council, the Swedish Foundation for Strategic Research and the Royal Botanic Gardens, Kew.

CONFLICT OF INTEREST

The authors declare they have no conflict of interest.

AUTHORS' CONTRIBUTIONS

A.K., T.H. and I.S. planned and designed the research; A.K. assembled the data with the help of I.S., A.Z. and A.A.; A.K. analysed the data and wrote the manuscript with contributions of all other authors.

Open Research

PEER REVIEW

The peer review history for this article is available at https://publons.com/publon/10.1111/1365-2745.13888.

DATA AVAILABILITY STATEMENT

All used datasets are publicly available (see Section 2). Composed dataset and code for analyses is published on GitHub 10.5281/zenodo.6389681 (Klimeš et al., 2022).