From mallees to mountain ash, specific leaf area is coordinated with eucalypt tree stature, resprouting, stem construction, and fruit size

Study taxa and traits

Traits were primarily measured on field surveys and collections by the authors and others over several years associated with distinct campaigns (Pollock et al. 2012, 2018; Vesk et al. 2021; Vesk et al. unpubl. data). For this study, we used a subset of this dataset, including measurements from 1942 trees across 164 eucalypt taxa, following Brooker (2000) and Nicolle (2019), including Angophora (2 taxa), Corymbia (5 taxa) and Eucalyptus (157 taxa), along with two other Myrtaceae tree species important in the south-eastern forests, Lophostemon confertus and Syncarpia glomulifera subsp. glomulifera. These were collected predominantly around the southern Great Dividing Range in south-eastern Australia, but also included smaller collections in the Gariwerd Grampians, Millewa Mallee region, and Koi Kyenunu-ruff Stirling National Park in Western Australia (Fig. 1). This included several regions of high species richness and endemism for the genus Eucalyptus and mostly corresponds to the alpine and mainland subregions within the South-east bioregion defined by González-Orozco et al. (2014), though some samples also fall into the South-west bioregion.

Fig. 1.

Collecting locations for samples of 164 eucalypt taxa for which we collected trait data. Regions sampled include alpine, mallee and coastal environments as well as regions of species richness and endemism in South-western WA, Gariwerd Grampians and South-Eastern NSW.

Sampled traits included those in the Leaf-Height-Seed scheme (Westoby 1998) but extended to associated traits for each of these strategy axes including leaf size, bark thickness, fruit mass and fruit wall width, wood density and stature. Details on trait measurement or extraction can be found in Table 1 and, although this was predominantly consistent across collecting campaigns, fruit wall width data were not available for 14 taxa collected in the Millewa Mallee region. Some fruit and seed masses for the Millewa Mallee also were made from herbarium material (Pollock et al. 2018). Relative height, relative bark thickness, stem density and fruit wall width data were also not available for some 11 taxa sampled exclusively in the Millewa Mallee region. The remaining traits were measured in almost all taxa. We added Maximum Height based on values extracted from the EUCLID database (Slee et al. 2015).

For all continuous traits, we characterised each taxon with trait medians, first taking the median by individual tree and then by taxon. Where taxa in the trait dataset were not resolved to subspecies level, we did so by reference to distribution information (Western Australian Herbarium 1998; VicFlora 2021). We calculated the median maximum height of all subspecies. The remaining six taxa were assigned maximum heights based on that of their closest relatives or other databases such as the current accessible EUCLID website (Slee et al. 2015) or local flora databases such as Florabase and VicFlora (Western Australian Herbarium 1998; VicFlora 2021). A list of the taxa for which maximum height was inferred and a more detailed description of how this was done in each case can be found in Supplementary Table S1. To aid in comparison, all trait data were log-transformed, mean-centred and scaled by dividing by one standard deviation prior to analysis.

Fire-response strategies were added for each taxon designating it either an obligate seeder, lignotuber-only (or basal-only) resprouter, stem-only (epicormic only) resprouter, or combination (basal and epicormic) resprouter based upon the definitions and census conducted by Nicolle (2006). Eleven taxa identified by Nicolle to be resprouters of ‘Variable’ type, were included within the combination resprouter category. To include relationships between fire response strategy and other traits in our network analysis, we also defined basal-only resprouting (BOS) behaviour as a binary variable with basal-only resprouters coded as 1 and species known to exhibit other strategies combined and coded as 0. Before scaling to match the variance of the other variables, the mean and standard deviation were 0.25 and 0.43, respectively. We believe this dichotomy was the most useful among fire response strategies due to preliminary data showing that basal-only resprouters had the greatest difference in their traits compared to the remaining categories (see Fig. S2) and our dataset contained enough species exhibiting these strategies to have reasonable statistical power. Further discussion and the full justification of fire-response categorisation is in Supplement 2.

Phylogenetic tree

Calculation of phylogenetic signal requires a phylogenetic tree of all taxa with included trait data to measure genetic distance between them. We constructed a phylogenetic tree from several published phylogenies, supplemented with taxonomic information. We used a recent phylogeny of the eucalypts based on Bayesian analysis of both nuclear (internal and external transcribed spacer, ITS and ETS) and plastid (matK and psbA-trnH) DNA regions, and which included many of the taxa in our trait data set (Thornhill et al. 2019). We note that the tree of Thornhill et al. (2019) (their fig. 4) was forced by their methodology to be fully resolved and not all nodes would have had character support. We accessed the tree associated with the Thornhill et al. (2019) paper from Knerr and Thornhill (2019), pruned this tree to the taxa from our trait dataset or their closest relatives (159 taxa in total), and combined it with a smaller (39 taxa), but well-supported, chloroplast-DNA phylogeny (Bayly et al. 2013) in order to confer greater statistical support to deeper divergences. Before the trees were combined, taxon names were verified against the Australian Plant Census (Council of Heads of Australasian Herbaria 2021) and branch lengths removed to avoid inconsistencies between different sources, as root depths were different. We then merged the two phylogenies using supertree methods (implemented with R package phangorn; Schliep 2011; R Core Team 2021) using Matrix Representation Parsimony (Baum 1992; Ragan 1992). To avoid the effects of random resolution of polytomies in the output supertrees (either originating from inherent phylogenetic uncertainty within the input trees or from regions of disagreement between them) we ran the supertree function 100 times and used the strict consensus (with ape package; Paradis and Schliep 2019), which was considered the most conservative approach.

We assumed species to be monophyletic for the purposes of positioning taxa, and so, although only 113 taxa in the trait dataset had an exact representative in either input tree, 43 taxa could be positioned on the final tree based on the position of members of the same species or by identifying orthographic variants. When the trait dataset and phylogeny used names that were not both resolved down to subspecies level, we used other subspecies or the species name to position these taxa. We positioned the remaining 13 taxa on the tree based on the positions of their proposed closest relatives. Although various phylogenetic and taxonomic sources disagreed considerably about which taxa were closest relatives, testing taxa with the most extreme trait values in different potential positions revealed only a marginal effect on phylogenetic signal compared to its uncertainty (see Table S3.1). Hence, even large uncertainty in taxon positions is unlikely to significantly affect results. Where closest relatives were determined using other published phylogenies (Jones et al. 2016; Fahey et al. 2022), only the sections of those phylogenies that were not yet resolved by previous methods (i.e. the immediate sister taxa to those of interest) were considered. Any disagreements these other phylogenies had regarding the relative positions of other taxa we had already resolved were ignored. More detail on the justification of the position of each such taxon can be found in Table S3.2 along with any other taxa whose position was similarly inferred.

Trait data for four taxa were ultimately excluded from this analysis due to a lack of phylogenetic information to position them on the tree, resulting in a final dataset of 164 taxa (Supplement 4). Finally, to enable computation of phylogenetic distance, we recalculated branch lengths for the resulting tree using the ‘Grafen’ method (Grafen 1989), which allows branch length calculation for purely topological trees. The final version of the tree used in this analysis can be found in Fig. S5.

Statistical analysis

Effect of phylogeny on individual traits

A summary of the scaled trait values for continuous traits and fire-response categories and how they are distributed across the phylogeny is presented in Fig. 2. Being a primarily molecular phylogeny, many taxonomic groups delineated within well-known classifications were not preserved. Yet, many groups such as genera Angophora and Corymbia, all subgenera of Eucalyptus, and various sections within subgenus Symphyomyrtus (Brooker 2000) could still be recognised. The genera Corymbia and Angophora could be seen with characteristically large, heavy fruits and seeds along with our single eudesmid (E. subgenus EudesmiaBrooker 2000), Eucalyptus pleurocarpa, and the other basal monocalypts (Eucalyptus subgenus EucalyptusBrooker 2000). Early diverging monocalypts (mainly Eucalyptus Sect. Longistylus Brooker 2000) generally had lower values for Maximum Height, Specific Leaf Area, Relative Bark Thickness, Relative Height, Leaf Area and Leaf Mass along with greater values of Stem Sapwood Density. A similar phenomenon was noted within the symphyomyrts (E. subgenus Symphyomyrtus), where section Dumaria (Brooker 2000) and section Bisectae (Brooker 2000) had similar clustering of basal-only resprouting coupled with lower values of Maximum Height, Specific Leaf Area, Leaf Area, Relative Bark Thickness and Relative Height, and higher values of Stem Density. There were only five obligate seeders in our dataset, all of which are monocalypts, not resolved as closely related in the phylogeny, and also classified in different sections. Most taxa were combination resprouters, with basal-resprouting taxa nested within the same clades.

Closely related eucalypt taxa tended to have similar trait values, but to varying degrees. Phylogenetic signal for each of the ten continuous traits (i.e. excluding fire response) ranged from weak in the case of Relative Bark Thickness (λ = 0.37), Leaf Mass (λ = 0.41) and Relative Height (λ = 0.45) to moderately strong for Specific Leaf Area (λ = 0.82) and Maximum Height (λ = 0.80) (Fig. 3). An entirely Brownian model of trait evolution, whereby present-day trait values could be correlated perfectly to ancestry, was rejected for all traits (λ < 1, P < 0.0001). For all traits except Relative Height, we could reject an absence of phylogenetic signal (H₀: λ = 0, P < 0.05).

Fig. 3.

Phylogenetic signal measured by Pagel’s lambda (±95% confidence interval) for each continuous trait.

Trait relationships

Overall, we found evidence for considerable coordination among traits. No traits were entirely independent – all traits correlated with at least one other trait with at least moderate strength |r| > 0.3 in both cross-species and phylogenetic analyses. Strengths of relationships varied widely (r²: <0.001–0.82) and effect sizes (slope magnitudes) ranged from 0.04 to 0.908 for cross-species analyses and from 0.009 to 1.03 for phylogenetic analyses respectively. Stronger relationships had larger effect sizes (slopes closer to one). Phylogenetic signal in trait relationships varied considerably (λ = 0.000001–0.864) even between models with the same dependent variable. A full summary of all regression coefficients for all trait relationships along with a matrix plot of correlation strength is available in Supplement 7.

Correlation networks (Fig. 4) displayed the number and strength of correlations between traits of at least moderate strength with thicker lines for stronger relationships and the position of the trait reflecting the overall strength of all correlations. Correlation networks showed several similarities in the trait relationships found using cross-species analyses (Fig. 4a) and those using phylogenetic analyses (Fig. 4b). Specific Leaf Area was consistently a hub in the networks, being the most, or among the most, connected traits, showing eight (of 10 possible) moderate correlations in the cross-species analyses (Fig. 4a). Resprouting mode was coordinated with each of leaf, stem and stature traits, indicating the importance of fire response to ecological strategies of eucalypts.

Fig. 4.

Trait network diagrams of bivariate correlations among eucalypts summarising (a) cross-species correlations and (b) phylogenetic correlations. Network diagrams represent traits as nodes in a network connected by lines (or arcs) of correlations. Stronger correlations appear as thicker lines and shorter distances between traits. The pattern in the network reflects the overall strength of correlations among all traits. So traits with more, stronger correlations with other traits appear connected by more lines and close to other traits. Hence traits that may be considered ‘hubs’, appear near the centre of the network, connected by lines to multiple traits, e.g. specific leaf area. Traits with weaker and/or fewer correlations appear at the edges of the network. Here, line thickness = ±√r² from regression models, henceforth referred to as r. To aid visualisation, only correlations with |r| > 0.3 are displayed in the network (adjusted P < 0.05), though position of traits is preserved. Abbreviations follow Table 1. Sample sizes for each regression model for each pairwise comparison ranged from 136 to 159. Note that |r| is not comparable across networks, only within a given network.

Both networks displayed a general pattern of stem and stature traits (Maximum Height, Relative Height, Stem Density mostly) being positioned on the opposite side of the network to the reproductive traits (Fruit Wall Width, Fruit Mass and Seed Mass) with leaf traits linking the two. This opposition reflects that not only are the stem and stature traits not correlated with reproductive traits, but the patterns of correlation that they each show with the leaf traits differ between those two trait groups. The reproductive traits, Fruit Wall Width, Fruit Mass and Seed Mass showed the greatest correlations (0.7 < r < 0.9) among traits reflecting physical constraints: to make big seeds or to have thick fruit walls, requires larger fruits. Unsurprisingly, Leaf Area and Leaf Mass were also consistently strongly correlated (r > 0.85), being two measures of leaf size. Being connected to the fewest traits, Seed Mass was at the outside in the cross-species networks, and Relative Height and Stem Density were least correlated with other traits in the phylogenetic network. However, all three traits consistently appeared in relatively peripheral positions on both networks, regardless of their number of connections. Full network diagrams for both analyses, showing either all relationships or only those statistically significant, are in Fig. S8.

Although networks of trait relationships were generally consistent between cross-species and phylogenetic methods, they differed in detail. The network based on phylogenetic analyses generally had fewer connections (Fig. 4) and weaker pairwise trait correlations (based on effect size and correlation strength) than the cross-species network (Fig. 5). Relationships between stem, stature and leaf traits, reflecting taller trees being lanky with thick bark and big leaves but low stem density, were particularly weakened in phylogenetic analyses, resulting in a less pronounced trait cluster. The connectedness of Specific Leaf Area became even more apparent in the phylogenetic network as other trait relationships weakened.

Fig. 5.

Difference in effect strength (indicated by the colour scale) between the cross-species and phylogenetic regressions (|phylogenetic slope| – |cross-species slope|) for each model testing a possible trait pair. The triangle icon on a model indicates a change in effect size that accompanied a change in statistical significance (P > or < 0.05). Note, the phylogenetic models used here for each pair of traits were those that generated the highest lambda value of the two options of assigning dependent and independent variables (see Table S7.2). Trait abbreviations follow Table 1.

Although stronger and more numerous moderate relationships (|r| > 0.3) emerged in the cross-species analysis, the relationships between leaf traits (Specific Leaf Area, Leaf Area, and Leaf Mass) and reproductive traits (Fruit Wall Width, Fruit Mass, and Seed Mass) had stronger effects when measured using phylogenetic analyses (Fig. 5). Five (of all possible 55) trait–trait correlations were no longer significant (P > 0.05) with phylogenetic analyses, whereas three became significant with phylogenetic analyses. This can also be seen in the networks (Fig. 4) where the stem, stature and leaf traits became farther apart and more sparsely correlated, whereas leaf size traits appear much closer to the reproductive trait cluster and positive correlations between Leaf Mass and Seed Mass as well as Leaf Area and Fruit Mass exceeded r = 0.3.

Post-fire regeneration strategies overlapped in trait values, although some notable differences were found especially in stem and leaf traits (Fig. 6) and most of these results were similar between the cross-species and phylogenetic analyses. Obligate seeders tended to fall within the trait ranges of the combination resprouters and were few, so we focus on the resprouters here. Supplementary results and discussion relating to obligate seeders and their traits can be found in Supplement 9. On average, basal-only resprouters differed significantly from combination resprouters; they were shorter (Maximum Height −1.62 ± 0.12 (cross-species effect size ± s.e.); −1.52 ± 0.16 (phylogenetic effect size ± s.e.)), more stout (Relative Height −0.38 ± 0.18; −0.38 ± 0.18), with denser wood (Stem Density 0.89 ± 0.18; 0.65 ± 0.21), thinner bark (Relative Bark Thickness −1.12 ± 0.18; −1.12 ± 0.18), smaller leaves (Leaf Area −1.40 ± 0.15; −1.15 ± 0.19), which were thicker and denser (Specific Leaf Area −1.41 ± 0.15; −0.91 ± 0.19), and lighter (Leaf Mass −0.74 ± 0.18; −0.71 ± 0.21), with thicker fruit walls (Fruit Wall Width 0.76 ± 0.19; 0.54 ± 0.23) in both analyses. The greatest differences between basal-only resprouters and combination resprouters and those where fire response explained the most trait variation, were that basal-only resprouters were shorter (Maximum Height: cross-species r² = 0.56; phylogenetic r² = 0.40), had lower Specific Leaf Area (0.37; 0.14), had smaller leaves (Leaf Area: 0.36; 0.20), and thicker bark (Relative Bark Thickness: 0.23; 0.22) (Fig. 6).

Fig. 6.

Boxplots with data plotted traits showing greatest differences between basal-only and combination resprouters. Distribution shown of (a) Maximum Height, (b) Relative Bark Thickness, (c) Specific Leaf Area, and (d) Fruit Wall Width within obligate seeders, basal-only resprouters, and combination resprouters for 151 eucalypt taxa for which fire-response data was available. Note, though scaled data was used for analysis, the x-axis labels are on the original scale. Whiskers extend from interquartile range (IQR) to the largest value no further than 1.5 times the IQR.

Most trait differences between post-fire regeneration strategies weakened when phylogenetic analysis was used, though without crossing the threshold of P = 0.05. Only in the cross-species analysis did basal-resprouters also have significantly, though weakly, heavier seeds (0.42 ± 0.19) than combination resprouters. This difference was found to be non-significant in the phylogenetic analysis. The full summary of plots showing the distribution of each trait within each fire-response category can be found in Fig. S2.

Broad trait relationships that were robust to phylogeny

Our results suggest that different dimensions of plant ecological strategy relate to one another and the whole plant does represent an integrated phenotype. Here we discuss those relationships that were detectable by both cross-species and phylogenetic analyses, indicating they are robust to phylogeny. We found strong coordination among reproductive structures, as expected considering that accessory costs of reproduction are known to scale proportionally with seed size (Henery and Westoby 2001; Lord and Westoby 2006). We also found strong coordination between both measures of leaf size and all fruit and seed traits. These associations support ‘Corner’s Rules’ in eucalypts (Corner 1949; Wright et al. 2007), suggesting taxa with twigs capable of bearing large, heavy leaves would also be able to have larger and more elaborate fruits. The robust position of Specific Leaf Area as a hub, highly connected to other traits throughout the phylogeny or across its tips, indicates the important role that the leaf economics spectrum (Wright et al. 2004) plays in multiple dimensions of variation within extant eucalypt taxa studied here and through their evolution.

We find that, compared to other analyses, Specific Leaf Area is a highly connected trait acting as a hub in our networks as well as relatively strongly correlated with other traits. Notably, Specific Leaf Area was positively correlated with Maximum Height and negatively with Stem Sapwood Density; these effects remained when considering evolutionary divergences. Compared to several studies in the Neotropics (Wright et al. 2007; Baraloto et al. 2010; Fortunel et al. 2012) and globally (Díaz et al. 2016; Flores-Moreno et al. 2019), we found stronger evidence that species of taller trees supported flimsier leaves and that species with more robust leaves had more dense stems. Why might this be? Two aspects of environment may contribute. Compared to the global climates, our sites do not experience strong seasonality, nor very low temperatures, where deciduousness becomes a profitable strategy. Neither is rainfall sufficiently seasonal to justify seasonal drought deciduousness, as is found in wet–dry tropics (Williams et al. 1997; Eamus et al. 1999). On the other hand, rainfall ranges widely across our sites (e.g. 325 mm mean annual rainfall at Murrayville in northwestern Victoria, to 1368 mm at Falls Creek in the Bogong High Plains of Victoria, data from http://www.bom.gov.au/climate/data/index.shtml). Possibly lower water availability drives a coordinated shift in each of lower Specific Leaf Area, higher Stem Density and shorter Maximum Height.

The commonly reported positive relationship between Maximum Height and Seed Mass (Moles et al. 2005; Wright et al. 2007; Díaz et al. 2016) did not appear in either the cross-species or phylogenetically-informed analysis in the eucalypts we studied. Murray and Gill (2001) found a negative correlation between height and fruit size, in a dataset primarily comprising southwestern WA eucalypts. It is possible that a complex interplay of aridity, nutrient poverty, fire frequency, fire severity, plant height, and recruitment dynamics has affected trait expression in the eucalypts (Murray and Gill 2001; Hopper 2021). Further investigation should help clarify what is driving this difference between eucalypts and other plant groups. Notably, this study is biased in sampling to south-eastern Australia. This bias is present in both the environments sampled, but also phylogeny. We have very few Corymbia, which are predominantly a northern Australian clade. Further work should endeavour to cover more species from northern and western Australia. We also note that our sampling of mallee growth form eucalypts strongly represents subgenus Symphyomyrtus section Bisectae (Brooker 2000) and Eucalyptus sect. Dumaria (L.D.Pryor & L.A.S.Johnson ex Brooker 2000), characteristic of dry regions, whereas those from subgenus Eucalyptus section Eucalyptus (Brooker 2000) characteristic of wetter regions, were less sampled.

Fire response strongly associated with traits and trait shifts through the phylogeny

Our results reveal that post-fire regeneration strategies and the functional traits of eucalypts are strongly intertwined. We focus on combination resprouters and basal-only resprouters and their traits and differences, with discussion of the few obligate seeders in Supplement 9. All patterns of traits differing between fire responses were robust to phylogeny, except the differences in fruit and seed mass of basal-only and combination resprouters. This suggests that differentiation between these fire-response categories in extant taxa is underlain by repeated correlated shifts in fire response and almost all traits.

Several traits were particularly strongly associated with fire responses. Aside from the few obligate seeding species (Supplement 9), our results support an interpretation that tall maximum height is benefitted by epicormic resprouting, distinctive among the eucalypts (Burrows 2013), whereas basal-only resprouting is associated with shorter maximum heights. This is likely because, at greater heights, the time difference to re-establish a canopy after fire by resprouting directly from an already-elevated trunk, compared with having to regrow the entire trunk from the ground, is much greater (Burrows 2013; Pausas and Keeley 2017). The multi-stemmed habit (mallee) common amongst basal-only resprouters also means that for the same investment in woody tissue, the maximum height of the whole plant must be shorter (Midgley 1996; Kruger et al. 1997). Basal-only resprouting needs no thick bark as the stems are regrown following fire, whereas epicormic resprouters (and obligate seeders) benefit from protective bark (Burrows 2002; Vesk and Westoby 2004a; Lawes et al. 2011, 2013; Clarke et al. 2013; Pausas 2015). This also provides a potential explanation for the tallest eucalypts, surprisingly, having the thickest bark overall, especially given the resource-intensive nature of maintaining thick bark as protection (Lawes et al. 2011, 2021; Pausas 2015). It is possible that such tall taxa compromise in other areas such as lower wood density. However, the link between tall species and thick bark may also be artefactual, given that bark thickness was only measured at a single height and many tall taxa (e.g. Eucalyptus regnans, Eucalyptus pilularis) have a ‘skirt’ of thick, rough bark at the base with much thinner bark above. This results in such tall species driving a positive correlation between Maximum Height and thick bark. How representative the thickness of the bark of the skirt is of bark on higher branches is unclear and would be valuable information.

It is worth noting that many of the traits most related to fire response are also known to vary along environmental gradients (Wright et al. 2005; Liu et al. 2019; Lawes et al. 2021) and many of our basal-only resprouters were sampled in more arid environments. This would reduce both Specific Leaf Area and Maximum Height and could account for some of the observed associations between these traits and fire responses. Although numerous studies relate traits to environment (e.g. Wright et al. 2005; Pollock et al. 2012; 2018) and other studies relate traits to fire-response (e.g. Pausas 2015; Lawes et al. 2021), we are not aware of work that does both simultaneously. Further work should aim to clarify the relative effects of fire response and environment on traits.

Traits displayed phylogenetic signal but with varied strength

Our results show that some traits are strongly related to patterns of heredity, but others are less so. The distributions of continuous traits across the phylogeny were also neither purely Brownian nor absent of any phylogenetic influence. Hence, although the evolution of these functional traits is most certainly affected by some form of selection pressure, ancestry remains an important factor driving many aspects of the functional ecology of extant Eucalyptus species.

Although phylogenetic signal for the categorical trait of resprouting could not be calculated, the uniformity of basal-only resprouting within certain lineages is notable, as broad studies across plant phylogeny have found that resprouting in the broad sense is an ancient trait and is labile across broader phylogenies (Vesk and Westoby 2004b; Pausas and Keeley 2014; Lawes et al. 2022). Epicormic resprouting in the eucalypts appears to have evolved from ancestors unable to resprout epicormically (see Crisp et al. 2011). One possible explanation is that the evolution of epicormic resprouting within the eucalypts and relatives presented by Crisp et al. (2011) refers to combination (i.e. epicormic and basal) resprouters, not epicormic resprouting only. Then, basal-only resprouting has repeatedly emerged after subsequent loss of epicormic resprouting ability while retaining basal-resprouting ability.

Highly conserved traits (Maximum Height, Specific Leaf Area, Seed Mass and Fruit Mass) indicated that the eucalypt taxa we studied had most similar maximum height, leaf economics strategies, and fruits and seeds to that of their ancestors and close relatives. Hence, related species are also more likely to have similar tendencies within ‘height growth’ strategies (Moles et al. 2009), leaf economics strategies (Wright et al. 2004), and seed and seedling provisioning strategies (Muller-Landau 2010). Although this does not necessarily indicate slow evolutionary rates for these traits (Ackerly 2009), it may indicate greater phylogenetic inertia (Felsenstein 1985) within these traits.

The associations of traits such as Maximum Height, Specific Leaf Area, and Seed Mass with environmental gradients have been the focus of numerous studies (in eucalypts see e.g. Schulze et al. 2006; Pollock et al. 2012, 2018; Liu et al. 2019). They reveal varied, and apparently scale-dependent patterns of association, with the climatic associations of height being the clearest (Moles 2018). If environment does influence the relative performance of different trait values, then we might expect strong environmental driving forces to overcome phylogenetic inertia more readily, resulting in relatively a weak phylogenetic signal. Yet, we observed strong phylogenetic signals in these traits (see also Ackerly 2004; Pollock et al. 2015). This may indicate that environmental trait optima themselves have phylogenetic signal (Hansen et al. 2008). Alternatively, adaptation may be overridden by community assembly via dispersal, abiotic and biotic filtering of taxa from a wider species pool (Keddy 1992), as has been shown to influence traits of co-occurring Eucalyptus taxa in Gariwerd Grampians (Pollock et al. 2015).

Furthermore, Leaf Mass and Relative Height, which had among the lowest phylogenetic signal, have not been especially associated with environmental gradients, relating more to allometry (Corner 1949) and perhaps light availability and adaptation to herbivory (Dantas and Pausas 2013) respectively. Bark thickness has been shown to be quite labile within species, likely related to fire regime (Lawes and Neumann 2022), which would accord with low phylogenetic signal. Variance partitioning analysis on a large subset of the taxa studied also suggests that Relative Height varied more within than between species (data not shown). Also, stature is well known to be plastic as the form of trees growing in open paddocks compared to dense plantings readily demonstrate. Hence, Relative Height may not be particularly stable for a given species and its use in cross-species studies may require greater intra-specific replication.

We also noted that traits with high phylogenetic signal were not necessarily correlated and the clades in which taxa had most similar trait values were sometimes different. This indicates that any phylogenetic inertia or lag on trait evolution imposed by phylogeny (Felsenstein 1985) was not specific to any one lineage and trait shifts likely occurred at different divergences in the tree.

Phylogenetic signal depends upon both the scale and accuracy of the phylogeny. Depending on the traits of other genera and families, our findings regarding relative degrees of phylogenetic signal of different traits may not apply at higher levels (Cornwell et al. 2014). Hence, the phylogenetic signal discussed here should be interpreted in the context of the eucalypt clade, and predominantly within Eucalyptus and its main subgenera and sections. All assessments of phylogenetic signal (and adjusted correlations depending on it) are also subject to phylogenetic uncertainty. The method by which the phylogeny was built assumed monophyly of species to place the most taxa on the tree. Monophyly may not be reasonable, especially given that different samples of the same eucalypt taxon collected at different locations can form a paraphyletic or even polyphyletic group (Jones et al. 2016). We also assumed bifurcating speciation and divergence, which does not consider the well-established ability of eucalypts to hybridise and undergo more reticulate modes of evolution (Griffin et al. 1988; Jackson et al. 1999; Pollock et al. 2013). We also combined two phylogenies (Bayly et al. 2013, Thornhill et al. 2019) to resolve deep divergences and only used a single tree from Thornhill et al.’s (2019) study and many of the relationships of sections and series within subg. Eucalyptus (Monocalypts), e.g. series Pachyphloia (stringybarks) appear spurious in light of longstanding morphological classification (Brooker 2000) and remain uncertain. We suggest that similar analyses using alternative phylogenies would be a fruitful avenue for future work.

Pattern of correlated evolution generally weaker when accounting for phylogeny, but not always

The strengths of most (but not all) relationships were found to weaken upon accounting for shared ancestry. This was most notably observed in the relationships between Specific Leaf Area, Stem Density, Maximum Height, Relative Bark Thickness and Leaf Area (Fig. 4). Within this subset of traits, a consistent pattern of correlated evolution of shifts to shorter trees were accompanied by shifts to thinner bark, lower Specific Leaf Area and smaller leaves, likely associated with the overall safety vs height dimension, allometry, or environmental affinities (Wright et al. 2005; Liu et al. 2019; Lawes et al. 2021). However, several of the correlations among leaf and stem traits weakened or became negligible among phylogenetically independent divergences on the tree. Thus, many of those associations occurred in more closely related taxa rather than in many independent lineages and are less likely to be generalisable to other clades.

Interestingly, the positive relationships between leaf and fruit (also seed) size strengthened upon accounting for phylogeny (Fig. 5). This suggests coordinated evolution of these traits throughout the evolution of the eucalypts, more than would be suggested by the prevalence of this trait combination within extant taxa. In addition to not being uniform in its amplification, shared ancestry can also obscure certain trait correlations.

Together these results suggest that although details of relationships within trait networks were more vulnerable to effects of shared ancestry, especially if they were weak, changes in statistically significant relationships were observed in only a minority of cases. Only one trait relationship with |r| > 0.3 became non-significant (P > 0.05) after accounting for phylogeny – that between Specific Leaf Area and Leaf Area. This suggests that for eucalypts we can have reasonable confidence that the broad evolutionary relationships between these traits with moderate strength in extant taxa are likely to generalise to other clades and do represent correlated evolution. However, we caution interpreting marginal results or those from small studies (see also Wright et al. 2005).

We hope that this study inspires further work to clarify the evolution of the eucalypt clade and provides a basis for systematic investigation of trait–environment relationships in trees at a subcontinental scale.