Species- and community-level patterns in fine root traits along a 120 000-year soil chronosequence in temperate rain forest


Correspondence author. E-mail: holdawayr@landcareresearch.co.nz


1. Below-ground plant functional traits regulate plant–soil interactions and may therefore strongly influence ecosystem responses to global change. Despite this, knowledge of how fine-root functional traits vary among plant species and along environmental gradients has lagged far behind our understanding of above-ground traits.

2. We measured species- and community-level root and leaf trait responses for 50 temperate rain forest species from 28 families of ferns, woody and herbaceous angiosperms and conifers, along a soil chronosequence in New Zealand that exhibits a strong gradient in soil nutrient availability. Relationships among species traits (both above- and below-ground) and their distribution along the chronosequence were tested using phylogenetic generalized least-squares regression to account for plant relatedness.

3. Distinctive root trait syndromes were observed; they were closely linked to species’ distribution along the chronosequence. Species growing in the strongly P-limited late stages of the chronosequence had relatively high specific root length (SRL), thin root diameter, high root tissue density, high levels of root branching and low root nutrient concentrations compared to intermediate stages. Species on the youngest site also had high SRL, but had low root tissue density, thick root diameter and high root nutrient concentrations.

4. Species root and leaf nutrient concentrations were positively correlated, reflecting the strong underlying gradient in soil fertility. In contrast, the relationship between SRL and SLA was more complex; there was a weak positive correlation between SRL and SLA, but this conflicted with stronger patterns of increasing SRL and declining SLA with increasing site age.

5. Community-averaged trait values calculated using presence/absence data showed similar trends to the species-level patterns. In contrast, community averages calculated using species abundance-weighted data showed weaker relationships with site age, particularly for morphological traits. This suggests that much of the variation in morphological traits between sites was driven by shifts in the presence of subordinate or ‘rare’ species rather than by changes in the dominant species.

6.Synthesis. Our study demonstrates co-ordinated species- and community-level changes in root traits along a soil chronosequence. These results highlight the influence of soil nutrition on plant functional traits and contribute to our understanding of the drivers of community assembly in a changing environment.


Plants are subject to a fundamental trade-off between maximizing their growth rates in nutrient-rich environments versus maximizing their ability to tolerate and persist in environments that are nutrient-poor (Reich, Walters & Ellsworth 1997; Reich et al. 1998; Westoby & Wright 2006). This trade-off is reflected in species’ functional traits; in particular in traits promoting high growth rates (e.g. high nitrogen concentrations, high rates of photosynthesis) versus traits promoting longevity (e.g. low nutrient concentrations, low respiration rates) (Aerts & Chapin 2000). Most plant functional trait research has focused on above-ground components such as leaves, and the traits associated with photosynthetic capacity (e.g. Reich, Walters & Ellsworth 1997; Richardson et al. 2005; Jung et al. 2010) rather than more logistically difficult below-ground tissues (but see Comas & Eissenstat 2004; Freschet et al. 2010). Whether or not similar trade-offs apply below-ground and therefore the extent to which they can be generalized into whole-plant strategies is uncertain (Craine et al. 2005; Tjoelker et al. 2005; Freschet et al. 2010). This is a key area of research given the increasing awareness of the role that below-ground processes play in regulating plant–soil interactions and determining ecosystem responses to global change (Coomes & Grubb 2000; Wardle et al. 2004; Orwin et al. 2010).

Variation in below-ground plant traits among species, communities and along environmental gradients is poorly understood. Specific root length (SRL, length of root per unit mass) is thought to determine soil nutrient uptake (Fitter 1985), and roots with high SRL have been associated with other root traits such as thin diameter, high levels of branching, low tissue density, low phenol concentrations and high N concentrations (e.g. Craine et al. 2005; Comas & Eissenstat 2009). There is some evidence linking this suite of traits to high RGR (Wright & Westoby 1999; Wahl & Ryser 2000; Comas & Eissenstat 2004), high respiration rates (Reich et al. 1998; Tjoelker et al. 2005), short root life span (Eissenstat & Yanai 1997) and low mycorrhizal dependency (Baylis 1975; Seifert, Beaver & Maron 2009). Nonetheless, exceptions are common: studies of northern hemisphere temperate tree species found no correlation between SRL and root N concentration (Pregitzer et al. 2002; Comas & Eissenstat 2004) or between root life span and SRL or root diameter (Withington et al. 2006), and no significant differences in root tissue density or root respiration rates between slow- and fast-growing species (Comas & Eissenstat 2009). These results highlight the need for primary data on root trait variability among species in different environments.

It is well known that coordinated sets of traits or ‘syndromes’ exist for above-ground plant tissues, particularly leaves. For example, plants with high leaf N content tend to have low SLA and a short leaf life span (Reich, Walters & Ellsworth 1997; Reich et al. 1998). However, the relationship between leaf trait syndromes and root traits is poorly understood (Craine et al. 2005; Tjoelker et al. 2005; Westoby & Wright 2006). Positive correlations between SRL and SLA have been found for nine boreal tree species (Reich et al. 1998), but Tjoelker et al. (2005) found no relationship between SRL and SLA in 39 grassland and savanna species despite strong correlations between leaf and root N content. Craine et al. (2005) investigated root and leaf traits of 90 grassland species from 77 sites and found distinctive leaf and root trait syndromes, but only tissue N concentration was consistently positively correlated among both leaves and roots. The life span of leaves and roots was also found to be uncorrelated for 11 northern temperate tree species (Withington et al. 2006). A likely reason for a decoupling of leaf and root traits is their differing responses to changes in soil nutrient availability (Wright & Westoby 1999). Most root trait studies have been intentionally restricted to glasshouse or common-garden experiments (e.g. Reich et al. 1998; Withington et al. 2006; Orwin et al. 2010), have considered only a small subset of co-occurring species (e.g. Comas & Eissenstat 2009) and have purposefully minimized environmental gradients (e.g. Orwin et al. 2010). There is therefore a need to examine root trait variation among species and communities and how this changes along environmental gradients (e.g. Craine et al. 2005; Liu et al. 2010), particularly of soil fertility (Coomes & Grubb 2000; Lambers et al. 2008).

Soil nutrient availability changes predictably during primary succession, with initial N limitation of primary production shifting to P limitation over long-term time-scales (Walker & Syers 1976; Vitousek et al. 2010), potentially resulting in ecosystem retrogression (Wardle, Walker & Bardgett 2004; Peltzer et al. 2010). The response of root traits to directional variation in soil nutrient availability has important implications for understanding the drivers of vegetation change during all phases of ecosystem development. We quantified variation in fine root traits (SRL, diameter, nutrient content, branching intensity, tissue density and abundance of hairs) and the relationship between root traits and leaf traits (SLA, tissue density and nutrient content) for 50 common temperate rain forest species along a long-term soil chronosequence characterized by strong changes from N to P limitation (Walker & Syers 1976; Richardson et al. 2004; Walker et al. 2010). We used this study system to resolve three outstanding questions about root traits and how they vary with soil fertility: (i) Do identifiable root trait syndromes exist? (ii) Are root traits strongly linked to leaf trait syndromes? (iii) Do community-level root traits vary systematically in relation to changing soil nutrients along the long-term soil chronosequence? We hypothesize that root traits associated with a fast-growing, resource acquiring plant strategy should be more common on younger, more fertile sites (e.g. high SRL, fine diameter and high nutrient concentrations), while those associated with slow growth, slow turnover and persistence (e.g. low SRL, thick diameter and low nutrient concentrations) should be more common on older, more nutrient-limited surfaces.

Materials and methods

Study site description

The Franz Josef soil chronosequence in southern New Zealand is a series of schist outwash surfaces ranging in age from 10 to c. 120 000 years created by repeated advance and retreat of the Franz Josef Glacier between its current terminal face (43.42°S 170.17°E) and the Tasman Sea (43.25°S 170.18°E; Stevens & Walker 1970; Almond, Moar & Lian 2001). Rainfall in the area is high, ranging from 6.5 m in the glacial valley to 3.5 m at the older sites nearer the coast. Soils are generally nutrient-poor with poor structure due to acid dissolution resulting from extreme leaching (Almond 1996). Soil P levels show strong declines with increasing soil age, with total available soil P (ratio of organic P to total C) declining from c. 5 mg kg−1 (60 years) to c. 1 mg kg−1 (120 000 years) (Richardson et al. 2004, 2005). Soil available N (aerobic mineralisable N) increases very rapidly at the early stages of succession (0–60 years), but then also declines with soil age from c. 110 mg kg−1 (60 years) to c. 15 mg kg−1 (120 000 years) (Richardson et al. 2004; Table S1 in Supporting Information). Leaf N and P concentrations decline strongly with site age (Richardson et al. 2004), and these are well correlated with a decline in maximum photosynthetic and leaf respiration rates (Turnbull et al. 2005; Whitehead et al. 2005). These studies have demonstrated that the Franz Josef soil chronosequence represents a strong gradient of soil fertility that is reflected in leaf traits and physiological processes.

Vegetation composition

The vegetation along the chronosequence is temperate rain forest with decreasing angiosperm cover and increasing conifer cover with increasing site age (Richardson et al. 2004). To quantify changes in species composition and abundance, we established three plots at each of eight sites ranging in age from 60 to 120 000 years in February 2009 (further site details given in Table S1). The youngest site (10 years, site 1 in Richardson et al. 2004) was not sampled as it had not yet been colonized by common forest plant species. Plot size scaled with stem diameter and was 5 × 10 m at the youngest two sites, and 10 × 10 m for all other sites. We recorded the foliar cover of all vascular species (excluding epiphytes) for each plot using an ordinal scale (1 = <1% cover, 2 = 1–5%, 3 = 5–25%, 4 = 25–50%, 5 = 50–75%, 6 = 75–100%) in five height classes (<0.3 m, 0.3–2 m, 2–5 m, 5–12 m, 12–25 m and 25+ m). To produce a single site-level percentage cover score for each species, we converted ordinal scores to percentage cover using the geometric mean for each cover class, averaged across all height classes and plots within a site. Foliar cover (expressed as a percentage of the total observed cover) ranged from 17 to 81% for woody angiosperms, 0–30% for conifers, 8–40% for ferns and 1–21% for herbaceous angiosperms (Table S2). Most species were present at multiple sites along the chronosequence; however, some species were restricted to either the youngest site, e.g. Melicytus ramiflorus (Violaceae) and Coriaria arborea (Coriariaceae), or the oldest site, e.g. Manoao colensoi (Podocarpaceae) and Gahnia procera (Cyperaceae).

Root traits

We sampled root traits at four sites along the chronosequence (sites 2, 5, 7 and 9 from Richardson et al. 2004) during January–March 2009, sampling all non-epiphytic species with >2% cover at any site, producing 50 species and 74 ‘site–species’ combinations of root traits. Sampled species accounted for >90% of total cover at all sites (Table S2). Most sampled species (including all conifers) were arbuscular mycorrhizal (Table S3). We collected fine roots located in the top 20 cm of the soil layer from three individuals of each target species at each site. Species identity was determined by one of three approaches: direct removal of individual plants for small and numerous species; root tracing for rare species or trees with difficult-to-identify roots; and removing a bulk soil sample close to an adult plant and using post-sampling identification to separate roots to species level (easily identifiable species only). Conservative root characteristics used to distinguish species in the field included colour, smell, diameter and the presence or absence of podocarp-type nodules (see Table S3). Unwashed root samples were taken back to the laboratory, kept cool (4 °C), and washed within 5 days of collection with tap water to remove adhering soil particles and organic debris, and to isolate roots of the target species. We used two criteria to functionally define fine roots for each species (Guo et al. 2008; Xia, Guo & Pregitzer 2010). We first selected roots <2 mm diameter, and then sampled all the non-woody ephemeral root orders within that diameter class. This typically resulted in the first 2–3 root orders being sampled (Table S3).

We divided each individual fine root sample (= 222) into three uniform subsamples of c. 0.5–1 g (wet weight). Each subsample consisted of several intact fine root clusters. We analysed scanned images (400 DPI, Epson Expression 10000XL, Seiko Epson Corporation, Nagano, Japan) of the first subsample (including root nodules where present; Dickie & Holdaway 2011) using WinRhizo Pro (Regent Instruments Inc., Quebec, Canada, 2000) to determine total root length and number of tips. The number of tips was adjusted to take into account the apex of each root cluster within each subsample. We then oven-dried subsamples at 60 °C for 72 h to determine dry weight. We calculated SRL as the fresh length of root per unit dry weight. Branching intensity was calculated as the number of tips divided by the total length of the sample (Comas & Eissenstat 2009). We dried the second subsample for analyses of N, P, potassium (K), calcium (Ca) and magnesium (Mg) concentrations (using Kjeldahl acid-digest and calorimetric methods; Blakemore, Searle & Daly 1987). The third subsample was stored in 70% ethanol for analyses under a dissecting microscope. We cut roots into 1–2 cm sections, selected eight at random, and measured segment length, root diameter, number of hairs, average hair length (intact hairs only) and number of nodules and nodule size (length + width). We calculated the total length of root hairs per centimetre of root length (hair index) and fine root tissue density for each species using the general formula tissue density = 1/(SRL × π × diameter2/4). For conifers, we included the volume and length contributions of podocarp-type (non-N-fixing) root nodules in the root tissue density and SRL calculations (Russell, Bidartondo & Butterfield 2002; Dickie & Holdaway 2011). For species sampled at multiple sites, root trait values were averaged across different sites. This was done because intraspecific variability among sites was non-significant for most of the species tested (Table S4).

Leaf traits

In January 2002 we collected leaf samples from 28 woody species including two species of tree ferns (see Richardson et al. 2004, 2005 for more details). We collected whole branches from five individuals of each species at the nine sites, sampling fully expanded sunlit leaves where possible. We kept leaves on branches until fresh leaf area could be measured (LiCor Area Meter, Model Li-3100; Lincoln, NA, USA), then dried and weighed them and measured N and P concentrations (Blakemore, Searle & Daly 1987). We measured leaf thickness using digital callipers at three points on leaves from five individuals at each site, and calculated leaf tissue density as leaf mass per unit area divided by leaf thickness.

Data analysis

To test for relationships among traits, we used phylogenetic least-squares (PGLS) regressions to account for phylogenetic dependence, following Duncan, Forsyth & Hone (2007). We constructed a hypothesis of the species phylogeny by grafting recently published phylogenetic trees (Fig. S1 in Supporting Information). We fitted two separate models to the data, one using ordinary least-squares (OLS), which assumes that species evolved independently and that there is no phylogenetic signal, and the other using PGLS, which assumes a Brownian-motion model of evolution. The most appropriate model was chosen using the Akaike Information Criterion (AIC). We calculated the phylogenetic correlation, lambda (λ), using maximum likelihood methods (Freckleton, Harvey & Pagel 2002; Duncan, Forsyth & Hone 2007). Lambda values range between 0 and 1, with values close to 1 indicating strong phylogenetic dependence in the data. A PGLS model with λ = 0 has no phylogenetic dependence and is equivalent to an OLS model, except it has a higher AIC due to extra model parameters. All trait data were log-transformed to control for heteroscedacity and log-normal distributions of trait values. As PGLS model parameters depend upon the order of the model (X on Y or Y on X), we modelled all trait combinations in both directions and, to be conservative, only report those that are significant in both cases.

We performed CCA on site-level percentage cover scores for each species constrained by log (site age). This was done to reduce the multivariate species distribution data to a single floristic axis describing the chronosequence, and to generate a value for each species representing its position along the chronosequence. The major axis of vegetation change arising from the CCA (axis 1, eigenvalue = 0.81) was strongly correlated with increasing site age (= 0.96, < 0.001), as well as with underlying declines in total P (= −0.94, = 0.001), inorganic P (= −0.97, < 0.001) and aerobic mineralisable N (= −0.77, = 0.026). To test for relationships among root traits and the distribution of species along the chronosequence, we analysed the CCA axis 1 species scores against the species-level trait data using the PGLS and OLS methods described above.

To assess the relative effects of species turnover and changes in species abundance on the variation in community-level root traits, we calculated community-level mean trait values for each of the eight sites, using two methods. First, to assess the effects of species turnover, we calculated community-level means on the basis of presence/absence of each species at each site. Secondly, to incorporate changes in species abundance, we calculated community-level means using trait values for each species weighted by their site-level percentage cover scores. Above-ground plant cover was used as a surrogate for root abundance due to the difficulty in assessing community-level root abundance for each species in situ. Preliminary data from root cores taken at each site showed strong correlations between total root length and above-ground abundance for two easily identified species groups, woody angiosperms (r = 0.97) and conifers (r = 0.83), with no difference in slopes (P > 0.2). The exception were monocots, which did show a significantly different slope from other easily identified groups (P = 0.016), but monocots were an insignificant component of root mass and cover at most sites. As root traits were sampled only from a subset of these sites, both methods used a single mean trait value for each species across all sites (Table S3) and thus assumed no within-species plasticity. See Table S4 for further information regarding within-species plasticity in our data set. Relationships between community-level mean trait values and site age were tested for using Pearson’s product-moment correlation on log-transformed data. All statistical analyses were performed in R version 2.9 (R Development Core Team 2009).


Relationships among traits

Species-level root traits were strongly correlated (Table 1). Species with high SRL tended to have roots with thin diameters, high levels of branching, high tissue density, fewer hairs, low P concentrations and high N-to-P ratios (N:P). Root N and P concentrations were strongly correlated, with higher nutrient concentrations being associated with thicker root diameter, lower root tissue density, more root hairs and lower branching intensity. There was a significant negative relationship between SRL and root P, but not root N. Root nutrient concentrations were positively correlated with leaf nutrient concentrations (Fig. 1c,d, P < 0.001 in both cases) and SLA (Table S5). There was also a weak positive correlation between SRL and SLA (Fig. 1a, P = 0.041), and between root and leaf tissue density (Fig. 1b, = 0.017). Although root morphological traits were generally not related to leaf N or P, they were related to leaf N:P, with high leaf N:P associated with smaller-diameter roots with high SRL, and high root tissue density (Table S5). Root traits were broadly conserved within phylogenetic groups (Fig. S1, Table S6), and in most cases the PGLS model provided a significantly better fit to the data than the OLS model (see Table S7 for full model outputs).

Table 1.   Relationships among species-level root traits [slope, significance and phylogenetic correlation (λ)]
 SRLRoot diam.Root TDBranch. int.Hair indexRoot NRoot P
  1. Slope values are taken from the best-fitting phylogenetic least-squares (PGLS) or ordinary least-squares (OLS) model assessed using Akaike Information Criterion (AIC). All trait data were log-transformed. For full model details see Table S7. SRL, specific root length; root diam, root diameter; branch int, branching intensity; TD, tissue density; hair index, length of hairs per unit root length.

  2. *< 0.05, **< 0.01, ***< 0.001.

Root diam.−1.37***0.711           
Root TD0.51*0.35−0.73***0.001         
Branch. int.0.85*0.60−0.62***0.690.180.321       
Hair index−0.07*0.630.05***0.78−0.04**0.790.000.341     
Root N−0.280.460.36**0.69−0.45***0.79−0.26**0.221.95*0.411   
Root P−0.39**0.600.38***0.75−0.36***0.76−0.23**0.132.01**0.450.64***0.001 
Root N : P0.87**0.71−0.61***0.690.300.690.32*0.00−3.02*0.32−0.380.00−1.38***0.00
Figure 1.

 Relationships between species root and leaf traits: (a) Specific root length (SRL) and SLA, (b) root and leaf tissue density, (c) root and leaf N concentrations and (d) root and leaf P concentrations. Legend shows phylogenetic correlations (λ, phylogenetic least-squares model), and Pearson correlations (r, ordinary least-squares model), along with their respective significance (*< 0.05, **< 0.01, ***< 0.001). Fitted lines taken from the model with the lowest Akaike Information Criterion (#). See Table S7 for full model details. Closed circles indicate woody angiosperms, open circles indicate conifers and open squares indicate tree ferns.

Species-level shifts in traits along chronosequence

The relative positions of species along the chronosequence (represented by the axis 1 scores in the vegetation CCA) were significantly correlated with their root traits (Fig. 2). Species typical of older sites (i.e. those with higher CCA axis 1 scores) tended to have fine roots with higher SRL (P = 0.002, phylogenetic correlation λ = 0.72), thinner diameter (< 0.001, λ = 0.84), higher tissue density (< 0.001, λ = 0.81), increased branching intensity (= 0.002, λ = 0), less hair length per centimetre of root length (= 0.006, λ = 0.39), lower nutrient concentrations (< 0.001 in all cases, λ = 0.47–0.61) and higher N:P ratios (< 0.001, λ = 0.26) than younger sites. The relationships between both SRL and root N:P and site age were better described by quadratic PGLS relationships, with moderate to high SRL and root N:P ratios occurring in both the very early and very late-successional species (Fig. 2). Leaf traits were also correlated with CCA axis 1 scores, with species typical of older sites having lower SLA (= 0.004, λ = 0), higher leaf tissue density (= 0.046, λ = 0.27) and lower leaf N and P concentrations (< 0.001, λ = 0, and < 0.001, λ = 0.28, respectively).

Figure 2.

 Species root traits against the vegetation CCA axis 1 scores. Note log scale on y-axes. The x-axis is positively correlated with site age (Pearson’s correlation = 0.95, < 0.001). SRL, specific root length; TD, tissue density and hair index is length of hairs per unit root length. Legend shows phylogenetic correlations [λ, phylogenetic least-squares (PGLS) model], and Pearson correlations (r, ordinary least-squares model), along with their respective significance (*< 0.05, **< 0.01, ***< 0.001). Linear fitted lines taken from the model with the lowest Akaike Information Criterion (#). See Table S7 for full model details. For both SRL and root N:P, the dotted line is the significant linear PGLS model, while the solid line is a better fitting polynomial PGLS model [log(SRL) = 0.35*CCA12 + 0.29*CCA1 + 7.32, λ = 0.89, < 0.002, AIC = 89.2; log(N:P) = 0.28*CCA12 + 0.13*CCA1 + 2.3, λ = 0.63, < 0.001, AIC = 4.18]. Closed circles indicate woody angiosperms, open circles indicate conifers, closed squares indicate herbaceous angiosperms and open squares indicate ferns.

Community-level shifts in traits along chronosequence

Community-averaged trait values for morphological traits calculated using abundance-weighted data showed no relationship with site age or site-level soil variables (solid circles in Fig. 3a–e, Table S8). In contrast, community average trait values calculated using species presence/absence data showed significant relationships with site age for root tissue density and branching intensity (Fig. 3d,e), and there were strong correlations between presence/absence-based morphological traits and site-level soil variables, particularly total P (Table S8). A wide range of root trait values were present at any particular site, especially for morphological traits (Fig. 3, Table S9). The youngest site (60 years) stood out as having unusually high SRL, thin-diameter roots and high root N concentrations for its age (Fig. 3). These results were driven by the dominance of the angiosperms Schefflera digitata (Araliaceae), M. ramiflorus (Violaceae), Aristotelia serrata (Elaeocarpaceae) and the N-fixer C. arborea (Coriariaceae); Tables S2 and S3). This unusually high SRL (Fig. 3a, solid circles) meant that a better fit was obtained using a nonlinear quadratic model for the relationship between community-level SRL and SLA [SRL (mg−1) = 118 − 1.58 × SLA + 0.0062 × SLA2, < 0.003 for all terms]. Abundance-weighted diameter and hair index traits were unusually high at the second youngest site due to the dominance of the woody angiosperm Griselinia littoralis (Cornaceae) and the herbaceous angiosperm Astelia fragrans (Liliaceae). Root N and P concentrations decreased with increasing site age for both presence/absence and abundance-weighted community-level means (Fig. 3f,g), reflecting the narrower range of root nutrient concentrations found at any particular site (Fig. 2). Root nutrient concentrations were positively correlated with declining inorganic soil P, total soil P and mineralisable soil N (Table S8).

Figure 3.

 Community-level average trait values for (a) specific root length, (b) root diameter, (c) root tissue density, (d) branching intensity, (e) hair length per unit root length, (f) root N, (g) root P and (h) root N:P ratio, against log site age (Table S8). Values calculated using either even-weighting for all species present at each site (open circles), or abundance-weighted trait values (closed circles). In both cases, a single mean trait value for each species across all sites was used. Legend shows Pearson correlations (r) and significance (**< 0.01, ***< 0.001) of each relationship.


This study demonstrates that community-level plant root traits change in a predictable and coordinated way along a long-term soil chronosequence. Early successional species growing in fertile young sites had relatively thick-diameter roots, low tissue density, high nutrient concentrations and thin leaves (high SLA), whereas late-successional species growing on older P-limited sites had thin root diameter, high tissue density, low nutrient concentrations and thick dense leaves (low SLA, Fig. 2). Although the relationship between SRL and successional stage was more complex, the highest SRL were consistently associated with late-successional species growing in P-limited sites. Many of these relationships were stronger once phylogenetic dependence had been accounted for in statistical analyses. For example, the conifers in this study became more abundant with increasing site age (Table S2; Richardson et al. 2004), yet they all had relatively thick root diameter and low SRL. Within the conifers, however, there was a shift towards species with smaller root diameter and higher SRL with increasing site age, reflecting the general community-level trend. These results support the idea that root traits are coordinated into generalized syndromes (Eissenstat 1992; Westoby & Wright 2006), and they highlight the importance of underlying soil nutrient gradients on species- and community-level variation in plant functional traits (Wright & Westoby 1999; Lambers et al. 2008).

Root trait syndromes and changing soil nutrients

Root trait syndromes have been proposed on the basis of a fundamental trade-off between high RGR and decreased tissue longevity (Eissenstat 1992; Wright & Westoby 1999; Westoby & Wright 2006). In particular, roots with high SRL, thin diameter, low tissue density and high nutrient concentrations are thought to be associated with high levels of root proliferation, high RGR, but short longevity (Eissenstat 1992; Reich, Walters & Ellsworth 1997; Wright & Westoby 1999). Our results caution against the generality of such statements; for example, although high SRL was found to occur in the early stages of succession where the fastest-growing tree species were present (Whitehead et al. 2005), thin-diameter roots with high SRL also dominated the most infertile sites supporting the slowest-growing species (Figs 2 and 3). This suggests that the scale of comparison may play an important role in determining the relationship between SRL and RGR (Liu et al. 2010). We propose that for large-scale comparisons such as ours across strong fertility gradients, high SRL is likely to be associated with sites of low fertility, resulting in a negative relationship between SRL and RGR. Within specific environments, however, high SRL is likely to be associated with high RGR due to increased capacity for nutrient uptake (e.g. Comas & Eissenstat 2004).

The increase in SRL with increasing nutrient limitation is consistent with literature demonstrating root proliferation in infertile sites (Fitter 1985; Grubb 1994) and a recent study by Zangaro et al. (2008) that investigated community-level fine root traits using soil cores extracted from grassland, secondary forest and mature forest sites in Brazil. In contrast to our primary chronosequence, Zangaro et al. (2008) found that soil nutrient concentrations (both N and P) increased along their secondary successional sequence. This led to an increase in root nutrient concentrations and decrease in SRL in the more fertile mature forest sites. They argue that thin-diameter, high-SRL roots observed in the comparatively infertile grassland and secondary forest sites represent a more effective strategy for increased soil exploration in nutrient-poor environments by increasing the total absorbing length per unit carbon invested (Eissenstat 1992; Wright & Westoby 1999; Zangaro et al. 2008). These results are consistent with our observations of increasing SRL with decreasing soil fertility during the retrogressive phase of ecosystem development along our chronosequence (Fig. 2a; Peltzer et al. 2010), and they highlight the importance of taking into account variation in soil nutrients when making generalizations about changes in root traits during succession.

Relationship between leaf and root traits

The strong, positive correlations between leaf and root nutrient concentrations are in line with previous results for temperate grasses (Craine et al. 2005) but not for northern hemisphere trees (Withington et al. 2006). This is most likely due to differences in the relative scales of investigation. In our study, as in Craine et al. (2005), root and leaf traits were sampled from a range of environments of contrasting fertility. In such cases, optimal nutrient allocation in terms of both nutrient uptake (roots) and productivity (leaves) is likely to be strongly constrained by the underlying gradient in soil fertility, resulting in strong correlations between leaf and root nutrient concentrations. In contrast, for species grown in a common environment (e.g. Withington et al. 2006), nutrient allocation to roots (in terms of optimizing nutrient uptake) is likely to reflect a different set of constraints and trade-offs to those involved in leaf nutrient allocation (in terms of optimizing canopy productivity), and this is likely to result in a lack of direct correlation between root and leaf nutrient concentrations.

A positive relationship between SRL and SLA has been predicted on the basis that both metrics reflect the same basic trade-off between metabolic rate and tissue longevity (Eissenstat & Yanai 1997; Reich, Walters & Ellsworth 1997). Our study found a weak positive correlation between species-level SRL and SLA (Fig. 1a, = 0.041), but also stronger relationships of increasing SRL (= 0.002) and declining SLA (= 0.004) with increasing site age. These findings add to uncertainty of the relationships between tissue longevity, SRL and SLA (e.g. Wright & Westoby 1999; Tjoelker et al. 2005). One explanation for this uncertainty is that SRL may not be as strongly coupled to root longevity as SLA is to leaf longevity (Withington et al. 2006). This is supported by other studies suggesting that morphological traits, such as the thickness of outer tangential walls of the exodermis, may be better predictors of root longevity than SRL (Withington et al. 2006). A second explanation is that root longevity is not necessarily related to leaf longevity, since these two traits are subjected to different physical, chemical and biological selection pressures (Espeleta, West & Donovan 2009; Hobbie et al. 2010). Further data, particularly on root life span, is required to distinguish between these possible explanations.

Both the concurrent increase in root tissue density (Fig. 2) with decreasing root diameter and the presence of aerenchyma in five of the 50 species (Table S3) would further confound the predicted relationships between SRL and root longevity, and therefore between SRL and SLA. The fact that four of the species with aerenchyma were only found at the oldest, most infertile site (Table S2) is likely to be a response to increased water logging that can occur in conjunction with declining P during primary succession (Gaxiola, McNeill & Coomes 2010). These findings suggest that the component variables of SRL (root diameter and tissue density) should be measured independently in order to fully understand the trade-offs involved between metabolic rate and tissue longevity in different environments.

Community-level patterns

For morphological root traits (i.e. those involved in soil exploration), the relationship between community-level root traits and site age differed depending on whether species-level traits were weighted by relative abundance or not. Significant relationships similar to the observed species-level patterns were found for unweighted community-level traits, while abundance-weighted community traits showed little or no relationship with site age or soil variables (Fig. 3, Table S8). These results suggest that the observed directional variation in morphological fine root traits along the chronosequence is driven by changes in the presence of low-biomass ‘rare’ species, rather than by shifts in abundance of the dominant (i.e. high above-ground cover) species. In contrast, strong declines in community-level root N and P concentrations with site age were observed for both presence/absence and abundance-weighted trait values. These reflect the strong underlying gradient in soil fertility, but also suggest that chemical plant traits may be inherently more influenced by environmental drivers than morphological traits (Craine et al. 2005).

Unweighted community mean values for root N and P were consistently higher than the abundance-weighted values in the most nutrient-limited sites (Fig. 3). This suggests that low-biomass subordinate or ‘rare’ species had relatively higher tissue nutrient content compared to dominant or ‘common’ species. This result is consistent with other findings that low-biomass species can have a significantly different set of functional traits than dominant species, and can exert a disproportionate influence on ecosystem properties through functional distinctiveness (Peltzer et al. 2009). Whether species are rare because they have unique functional traits (e.g. due to environmental filtering), or have unique functional traits because they are rare (i.e. ‘rareness’ is part of their particular life-history strategy) is an interesting ecological question that requires further attention.

Are root traits adaptive features?

It is unclear whether thick root diameter and low SRL are adaptive features or simply phylogenetic basal root traits (Baylis 1975; Brundrett 2002). In this study, roots of basal angiosperms (from the Magnoliidae clade) did have relatively thick diameters and low SRL, but these traits were also found in roots of more phylogenetically-distal species such as Griselinia littoralis (Cornaceae) (Fig. S1, Table S3). Large-diameter roots may be favoured because of their ability to rapidly penetrate a large soil volume (Eissenstat 1992) and their increased cortex volume available for mycorrhizal colonization (Brundrett 2002). Our results show decreasing root diameter with increasing soil P limitation, which appears contrary to the prediction that thicker roots could represent greater dependence on mycorrhizas (Brundrett 2002; Craine et al. 2005). This may be because mycorrhizal activity is uniformly high across all of the plant species studied (Hurst, Turnbull & Norton 2002; Dickie pers. obs.). Along the Franz Josef chronosequence, species having large-diameter roots and low SRL were most common at intermediate-aged sites (Fig. 2, Tables S2 and S3), suggesting that they might be out-competed in high-nutrient situations by more distal angiosperms with higher SRL, thin diameter and less dense root systems, and that they are also displaced during the low-nutrient retrogressive stages by species that have high SRL but also high root tissue density and low root nutrient concentrations.

Concluding remarks

Our study demonstrates predictable changes in root trait syndromes during the progressive and retrogressive stages of ecosystem development using a long-term soil chronosequence. Changes in these trait syndromes reflect underlying shifts in soil nutrients and the resulting selection pressures for different soil nutrient acquisition strategies. Much of the variation in morphological traits between sites was driven by shifts in the presence of subordinate or ‘rare’ species rather than by changes in the dominant species. Future work investigating the physiological mechanisms behind variation in root traits in relation to leaf traits, root longevity and RGR across sites of different fertility has the potential to provide a general framework for understanding the distribution of species along environmental gradients such as those that occur during both primary and secondary succession.


We thank A. Merkel, R. Avery, S. Orchard, D. Hood and L. Young for field assistance, K. Boot for laboratory support, A. Sparrow and E. Tanner for their help with planning and results interpretation, and L. Comas and two referees for their helpful advice. This project was funded by the Brian Mason Scientific and Technical Trust, the Woolf Fisher Trust (R.J.H.), and the New Zealand Foundation for Research, Science and Technology Ecosystem Resilience Outcome-Based Investment (Contract C09X0502; S.J.R., D.A.P., I.A.D. & R.J.H).