The evolutionary loss of aerenchyma limits both realized and fundamental ecohydrological niches in the Cape reeds (Restionaceae)


Correspondence author. E-mail:


  1. The basic physiological mechanisms that determine the niche limits, and so the local ranges, of species are often poorly understood. Here, we assessed whether the presence of root aerenchyma can account for both the fundamental and realized ecohydrological niche in the Restionaceae.
  2. We documented the ability to make aerenchyma of almost 300 of the c. 350 South African Restionaceae species with the help of field observations, cultivation experiments and herbarium collections, and correlated this information with the ecohydrological niche of the species. We used a diversity of sampling approaches to account for variation in both aerenchyma and habitat at individual, population and species level. Tolerance of waterlogging in species with and without aerenchyma was investigated experimentally.
  3. To account for phylogenetic relatedness, the evolution of both aerenchyma and the ecohydrological niche was explored in the context of an almost complete species-level phylogeny using both parsimony and maximum likelihood optimization methods.
  4. We showed that the presence of aerenchyma was correlated with the ecohydrological niche at population and at species level, and after phylogenetic correction. Most individuals of non-aerenchymatous species died when waterlogged, whereas most species with aerenchyma were waterlogging tolerant. This indicates that waterlogging acts as an environmental filter that excludes non-aerenchymatous species from these conditions. The absence, and to a certain degree also the presence, of aerenchyma predicts both the field-observed realized and experimentally determined fundamental ecohydrological niche.
  5. Aerenchyma was ancestrally present as inferred by both parsimony and maximum likelihood reconstruction and showed a strong phylogenetic signal with frequent losses but few gains. The ecohydrological niche was evolutionary more labile, with niche changes in many evolutionary lineages. Changes from wetland to dry habitats were reconstructed in species with aerenchyma, whereas the reverse change was never inferred for species that had lost aerenchyma.
  6. Synthesis. Our study provides evidence that functional traits can effectively predict species niches, and evolution of these may constrain the habitats available to the clades. We underpin the importance of understanding the causal driver of the local distribution of species for making robust predictions of species range shifts under climate change.


Niches are a central paradigm in biology (Wiens et al. 2010) and are used to explore many questions in conservation biology (Wiens & Graham 2005; Boogert, Paterson & Laland 2006), species composition and richness patterns (Silvertown et al. 1999; Silvertown 2004) and species responses to global change (Thomas et al. 2004; Thuiller et al. 2005; Broennimann et al. 2006). The fundamental niche describes where species are physiologically able to survive, or better, where they show a positive population growth rate (Chase & Leibold 2003). The realized niche (Hutchinson 1957) is an often smaller environmental space after accounting for biotic interactions. The realized niche is what we know most about for most species, and it is used in correlative approaches in which current species ranges are used to infer future ranges under climate change. However, such approaches might perform poorly when species encounter novel environmental combinations (Davis et al. 1998). Knowledge about the fundamental niche and the physiological–anatomical processes and capabilities that limit species ranges could increase the robustness of species range models. Mechanistic approaches have greatly advanced this field, in which physiological constraints are used to infer species fundamental niches and translate these to explicit spatial ranges (Kearney et al. 2009; Gallien et al. 2010). It still remains a major challenge to identify the functional traits that are linked to performance and survival and therefore limit species ranges.

Ecohydrology is an important niche parameter (Laan et al. 1989; Silvertown et al. 1999; Visser et al. 2000; Silvertown 2004; Araya et al. 2011). It is yet unknown which plant ecophysiological attributes segregate species along this niche axis. Shoot adaptations might be important for transpiration regulation, and root adaptations may allow plants to deal with anoxic soil. There have been several surveys of root architecture (Richards 1986; Poot & Lambers 2008; Hinsinger et al. 2009) and many investigations into root aerenchyma (Laan et al. 1989; Visser et al. 2000), which is a well-known adaptation to waterlogged conditions (Justin & Armstrong 1987). Aerenchymatous tissue consists of interconnected air-filled gas spaces, which facilitate internal ventilation (Justin & Armstrong 1987). This enables escape from anoxia that arises due to slow oxygen diffusion in waterlogged soils (Vartapetian & Jackson 1997). Developmentally, aerenchyma can result from one of several processes, including lysogeny, schizogeny and expansigeny (Seago et al. 2005). Aerenchyma is a common feature in many distantly related flowering plants (Seago et al. 2005). Its evolutionary origin predates the angiosperms (Seago et al. 2005; Jung, Lee & Choi 2008). However, many plant species are not able to make aerenchyma, some species induce it when waterlogged and others express it constitutively (Justin & Armstrong 1987; Jackson & Armstrong 1999; Visser et al. 2000; Seago et al. 2005). It is suggested that the position of a species along the soil hydration gradient might be related to its possession of root aerenchyma (Laan et al. 1989).

The Cape Floristic Region (CFR) is remarkably species rich (Goldblatt & Manning 2000) and is characterized by high species turnover (beta diversity) as well as a relatively high local or community richness (alpha diversity), resulting in a very high regional (gamma) diversity (Cowling, Holmes & Rebelo 1992). The high diversity is a consequence of very high species richness in a small number of clades (Linder 2003). Consequently, most regional floras contain large numbers of congeneric species, indicating a very high degree of niche specialization. Elucidating factors behind the evolution and coexistence in these clades will cast much light on the origins and maintenance of the regional diversity. The high species richness, in particular the high turnover rates between communities, has been attributed to niche segregation along several axes (Linder et al. 2010) such as fire-survival mode (Schutte, Vlok & van Wyk 1995), pollinator specialization (Johnson 2010), steep rainfall gradients and changes in bedrock (Richards, Cowling & Stock 1997). Recently, Araya et al. (2011) demonstrated the importance of the ecohydrological gradients in the Cape flora.

High local species richness in a clade may be the result of the evolution of the ability to occupy diverse habitats, thus an adaptive radiation. This may be a consequence of the evolution of particular structures, such as fruit type in the Hawaiian lobeliads (Givnish et al. 2009) or leaf venation patterns in the Angiosperms (Walls 2011). Although there are numerous case studies demonstrating a phylogenetic influence in the evolution of these structures, there are also many instances where the taxonomic distribution of traits does not show a phylogenetic signal (Losos 2011). Apparently adaptive structures may also predate the conditions to which they are supposed to be adaptive, as illustrated by the leaf traits in chaparral (Ackerly 2004). Niche evolution often (e.g. Prinzing et al. 2001; Smith & Donoghue 2010; Wiens et al. 2010), but not always (Losos 2011), shows a phylogenetic signal. Tracing the evolution of putative adaptive structures may reveal when they evolved and correlating them with the niches of the species can corroborate the relationship between the niche and structure (Arakaki et al. 2011). This structure–function correlation can provide the context for interpreting the evolutionary history of the clade.

Restionaceae constitute an important component of the Cape flora. The African Restionaceae all belong to the monophyletic subfamily Restionoideae (Briggs & Linder 2009); the remaining subfamilies are Australasian with one South American species. The Restionoideae radiated in the CFR, in which 340 of 350 species occur (Linder 2003). The phylogeny of the Restionoideae has been studied extensively, resulting in an almost completely sampled, well-resolved molecular tree (Hardy, Moline & Linder 2008). Not only is the subfamily one of the largest clades in the flora (Linder 2003), but it is also ecologically dominant over much of the typical fynbos vegetation (Rebelo et al. 2006). Restionoideae are evergreen, rush-like plants, which are entirely wind pollinated and mostly dioecious (Linder 2000). Detailed ecological information has been compiled for all species (Linder 2001). Restionoideae occur both in wetlands and on well-drained soils (Rebelo et al. 2006) and segregate neatly along ecohydrological gradients (Araya et al. 2011).

In this study, we explore the ecological impact of variation in the ability to produce aerenchyma in the African Restionaceae, and its potential macro-evolutionary consequences. Firstly, we asked whether the realized ecohydrological niche is correlated to the presence of aerenchyma. We specifically tested the null hypothesis that the proportion of species with and without aerenchyma is the same in well-drained and wetland habitats. To this end, we documented the distribution of aerenchyma among the African species of Restionaceae and correlated this (with phylogenetic correction) with habitat data at the population (from 547 plots, and from six sites with quantified ecohydrological status) and at the species level for almost all species across the whole Cape flora. Additionally, we tested for an association at individual level between ecohydrological status and the presence of aerenchyma at one site. Secondly, we asked whether the presence of aerenchyma indicated a fundamental waterlogged niche. We tested the null hypothesis that species with aerenchyma had the same capacity to survive waterlogging as species without aerenchyma using a pot cultivation experiment with a small subset (12) of species. Finally, we asked whether the ability to make aerenchyma influenced the evolutionary history of the clade. Specifically, we tested the null hypothesis that the likelihood to gain aerenchyma was the same as loosing aerenchyma and that these are independent of the habitats occupied, by mapping both the presence of aerenchyma and the ecohydrological niche over an almost complete phylogeny of the clade.

Materials and methods

Taxonomic Distribution of Aerenchyma

To document the distribution of aerenchyma in the African Restionaceae, we sampled 298 of the 350 South African Restionaceae species in the field and from the Zurich Herbaria and the Bolus Herbarium (University of Cape Town). We used the taxonomy of Linder & Hardy (2010). In the field one to two individuals per species were sampled. Roots were fixed in FAA [70% ethyl alcohol : glacial acetic acid : formaldehyde (85 : 10 : 5)] and stored in 70% ethanol. In the Zurich Herbaria, all specimens with roots were sampled, whereas one individual per species was sampled in the Bolus Herbarium.

Herbarium root samples were rehydrated in soapy water at 60 °C for 3 days before sectioning. Both rehydrated and fixed roots were sectioned by hand to check for the presence of aerenchyma. Sections were stained for 10 min in 2 : 1 Alcian Blue (1% in distilled water, stains non-lignified cellulose): Safranin (1% in 70% ethanol, stains lignified cell walls). Sections were washed in water, dehydrated in an ethanol series and mounted in histomount. To illustrate the root anatomy, a few freshly fixed roots were embedded in Technovit 7100, sectioned by HM 355S microtome, stained for 1 min in ruthenium red (stains cellulose) and for 3 min in toluidine blue (stains lignified tissues) and mounted in histomount on glass slides.

Aerenchyma was inferred as present in a species if detected in at least one root. Because absence of aerenchyma may be due to a plastic response to well-drained conditions, we assessed the discovery rate of aerenchyma for the species in a given genus. The discovery rate (r) is the ratio of aerenchymatous roots to all roots investigated for each species and averaged for each genus (except in the case of Restio, which was partially separated into the subgenera). Only species for which several root samples were available, and in which at least one root sample was aerenchymatous, were used. If the discovery rate approaches one, then even a small sample of that species will give a confident result, whereas a small ratio suggests that the chances of missing the ability to make aerenchyma are very large. The necessary sample size (n) to detect aerenchyma for a selected significance level (α) can be calculated with the following equation:

display math

To exclude the possibility of failing to detect aerenchyma in well-drained soils due to a plastic response, we investigated whether discovery rate differs between wetland and well-drained habitats, for each genus, as noted above, as well as for the whole data set, using a Kruskal–Wallis rank sum (KW) test.

Waterlogging Experiment

A waterlogging experiment was conducted to test whether aerenchyma formation can be induced in species in which we had not seen aerenchyma and to assess the fundamental niche by investigating survival under well-drained and waterlogged conditions for aerenchymatous and non-aerenchymatous species. Five aerenchymatous species restricted to marshes or streambanks (‘wetland species’: Cannomois grandis, Elegia fistulosa, Elegia cuspidata, Elegia tectorum, Restio paniculatus) and seven non-aerenchymatous species that are restricted to well-drained soils (‘dry land species’: Elegia ebracteata, Rhodocoma arida, Rhodocoma gigantea, Thamnochortus bachmannii, Thamnochortus insignis, Thamnochortus lucens, Thamnochortus platypteris) were used. Five watering regimes with increasing aeration stress were applied: (i) normal (2× weekly) watering, (ii) 2 days waterlogging, 1 day dry, (iii) 6 days waterlogging, 1 day dry, (iv) 13 days waterlogging, 1 day dry, (v) constant waterlogging. For each watering regime, two mature plants (not clones) of each species (except one specimen only in C. grandis, E. cuspidata, E. ebracteata, E. tectorum) were potted in easy-draining soil (sand and peat mix) in 19-cm pots with draining holes and cultivated over the summer 2010 (April–October). After 3 months, three actively growing roots from the base of the pots were cut without disturbing the main root system and fixed in FAA. After 5 months, the root system was exposed and investigated for the presence of actively growing roots from the base of the tiller, and one to three of them were fixed in FAA. Species were scored as waterlogging susceptible if more than half of the individuals of the waterlogging treatments (ii–v) were dead or dying at the end of the experiment; otherwise, waterlogging resistance was inferred. The experiment was conducted in a greenhouse located at the Botanical Garden of the University of Zurich (47°39′ N, 8°54′ E). The association between aerenchyma and habitat, and aerenchyma and waterlogging tolerance, was investigated with a Fisher's exact test.

Habitat–Aerenchyma Correlations

We used four sampling approaches to take account of potential within-species variation in root anatomy as well as ecological variability of the species (Table 1). This allowed us both to assess variability within species as well as broad extrapolation over the whole African Restionaceae.

Table 1. Sampling scheme to test for ecohydrological niche–aerenchyma correlation at increasing levels of generalization and at different levels of quantification
Ecohydrological nicheAerenchyma recorded at level of
Quantified at individual level Ecohydrology sites
Estimated at population levelRiverlands 
Estimated at population level Co-occurrence plots
Estimated at species level Global associations

To test the association between aerenchyma and soil drainage capacity, we assessed 23 circular plots with a diameter of 10 m at Riverlands (Western Cape of South Africa, c. 50 km north of Cape Town, 33°3′ S, 18°4′ E). These plots were placed subjectively to cover as wide a range of soil and vegetation types as possible and to ensure that all Restionaceae species in the area were sampled. The presence of aerenchyma in each species of Restionaceae was recorded from the first two plots on which it was found. Habitats were classified as wetland if either an impervious layer was present in the upper 100 cm with no possibility of lateral ground water drainage or if ground water was present in the upper 100 cm; otherwise, they were categorized as well-drained. No quantification of soil moisture was undertaken. The aerenchyma status that was inferred for each individual was correlated with the habitat in which the individual was collected with a Fisher's exact test.

To test whether the presence of aerenchyma is associated with a quantitative measure of soil moisture, we examined species recorded in eight ecohydrology sites (Araya et al. 2011). In these sites, species composition in a large number of plots was correlated with water availability. The water table fluctuation was measured during 12 months and drought and aeration stress were modelled based on these results (Silvertown et al. 1999). A sum exceedence value for soil drought (SEVd) was calculated by cumulating values in periods in which the moisture tension of the surface soil was higher than 5 kPa, which could potentially induce stomatal closure (Henson, Jensen & Turner 1989). A sum exceedence value for aeration (SEVa) was assessed by cumulating values during periods in which the soil air-filled porosity fell below 10% by volume, which indicates waterlogging (Wesseling & van Wijk 1957). Correlations between the presence of aerenchyma (inferred to species level) and aeration stress were analysed with KW test.

To test whether qualitatively assessed ground water conditions at population level correlate with the ratio of aerenchymatous species (as determined at species level), we used 547 plots (‘co-occurrence plots’) from across the whole CFR, which cover all different habitat types. Circular, 10-m-diameter plots were placed subjectively in the centre of homogenous vegetation communities. At each plot, all species of Restionaceae were recorded, and the conditions described as ‘cliff’, ‘cliff-seep’, ‘convex-seep’, ‘impeded drainage’, ‘marsh’, ‘seepage’, ‘streambank’, ‘valley-bottom’ and ‘well-drained’, terms that describe different types of ground water conditions, following Linder (2005). Plots in azonal habitats (e.g. streambanks, cliff-seep) were still of the same size, but differed in shape, adapted to the habitat available. Cliff and well-drained habitat were categorized as ‘well-drained’ and the remaining as ‘wetland’. The correlations between both the ratio of aerenchymatous species and soil drainage capacity, and Restionaceae species richness and soil drainage capacity, were analysed with KW test. Pairwise comparisons were made with the Wilcoxon test with Bonferroni correction for multiple comparisons.

We tested the correlation between the species' inferred presence of aerenchyma and the ecohydrological niches of the species with Fisher's exact test.

Geographical Patterns

To test whether the correlation of aerenchyma and habitat is confounded by a geographical pattern, we tested the proportional distribution of species with and without aerenchyma in the Cape flora. We recorded the number of species with each type from the regions used by Linder (2001). The observed ratio of aerenchymatous species to all species in each region was tested against the expected frequency (ratio of aerenchymatous species in the CFR) in a chi-square test.

Phylogenetic Analysis

Since correlative approaches can overestimate the link between two traits, we explored the evolutionary history of both aerenchyma and the ecohydrological niche. We tested whether evolution of both aerenchyma and the ecohydrological niche showed a phylogenetic signal, inferred the ancestral states of these characters and investigated the correlations between them after phylogenetic correction. A sample of 500 Bayesian trees, generated by a Bayesian analysis of a DNA sequence variation data set of (Hardy, Moline & Linder 2008), were used for these analyses. All characters were scored with binary coding. Aerenchyma and the ecohydrological niche information were acquired of 234 of the 297 species represented in the phylogenetic tree. Niche information was derived from Linder (2001) and scored as well-drained if the species never occurs on waterlogged soils, otherwise waterlogged habitat was inferred.

Phylogenetic signal was investigated using a randomizing approach as implemented in mesquite 2.74 (Maddison & Maddison 2010). The number of evolutionary steps required to fit each character most parsimoniously over an arbitrary tree from the Bayesian set was calculated. This was compared with a test distribution generated by randomizing each character over the selected tree 1000 times. Phylogenetic signal was inferred if the observed number of steps was in the lowest 5% of the test distribution.

The ancestral states for the presence of aerenchyma and the ecohydrological niche were estimated using both maximum likelihood (ML) and parsimony (MP) approaches. Topological variation was accounted for by using 500 randomly selected Bayesian trees. In the ML analysis, different parameters for forward and backward evolution were allowed. We classified the basal optimization for each tree into significantly 0, 1 or ‘ambiguous’ if there was less than a 2 log-likelihood difference between the two states, which is arbitrarily taken as the cut-off for a significant difference (Pagel 1999). All analyses were conducted in mesquite. Ancestral state reconstruction was additionally conducted with BayesTraits ( (Barker & Pagel 2005; Pagel & Meade 2006). States were reconstructed empirically and the algorithms were optimized 25 times per run.

To test whether gains and losses were equally likely, the transition rates were investigated with the 500 Bayesian trees using ML approach as implemented in BayesTraits. States were reconstructed empirically and the algorithms were optimized 25 times per run. A chi-square test of the likelihood ratio (LR) of the ‘two rates model’ (gains and losses vary independently) and the ‘equal rate model’ (gains = losses) was applied (LR = 2 × log-likelihood(better fitting model) − log-likelihood(worse fitting model)) with degree of freedom (d.f.) = difference in d.f. of the better fitting model minus d.f. of the worse fitting model. Significance was accepted if the better fitting model was in 95% of the trees significantly better (α = 0.05) than the worse fitting model. We tested whether models allowing only gains or only losses were significantly better than either the two rates or the equal rate model for aerenchyma and ecohydrological niche evolution, depending on which was selected as the better model. Significance was accepted if LR > 2.

To test whether aerenchyma and habitat are also correlated after phylogenetic correction, we investigated the co-evolution of aerenchyma and habitat with the 500 Bayesian trees using ML approach as implemented in BayesTraits. States were reconstructed empirically and the algorithms were optimized 25 times per run. A chi-square test of the LR of a dependent and independent evolutionary model was applied (see above). If the dependent model was in 95% of the trees significantly better than the independent model, this was interpreted as co-evolution of the two traits. Additionally, co-evolution was tested using an MP approach. The proportionate number of losses and gains (or no change) of aerenchyma was compared with the optimized habitat conditions that occur at the same time, and the significance of the differences tested with Fisher's exact test (Sillén-Tullberg & Temrin 1994).

All analyses in this paper were carried out in r (R Development Core Team 2010) if not indicated differently.


Taxonomic Distribution of Aerenchyma

Of the 298 species investigated, 180 species had the ability to make aerenchyma, no aerenchyma was detected in 93 species, and for 25 species no confident results could be obtained. Aerenchyma can be recognized by broad cortical fan-shaped gas spaces, which greatly differed in size and width (Fig. 1). In very few roots, large gas spaces scattered in the cortex were detected and these were also categorized as aerenchyma (‘scattered aerenchyma’) (see Table S1 in Supporting Information).

Figure 1.

Microtome root cross-sections with aerenchyma: (a) Elegia tectorum, (b) Elegia recta, (c) Elegia vaginulata, (d) Elegia hookeriana, (e) Elegia verreauxii, (f) Staberoha distachyos; and without aerenchyma: (g) Restio unispicatus, (h) Restio gaudichaudianus, (i) Thamnochortus fruticosus. Scale bar = 500 μm. AER, aerenchyma.

Although the discovery rate did not differ among habitats (KW rank sum test, see Table 2), our sampling in Rhodocoma, Willdenowieae and Restio (except subgenus Simplicaulos) was not adequate to detect aerenchyma with > 95% confidence. Whereas nearly all the species in the Rhodocoma and Willdenowieae are aerenchymatous, more than half of the species in Restio (except subgenus Simplicaulos) did not have aerenchyma.

Table 2. Discovery rate (r) (=ratio of aerenchymatous to all roots investigated for each species with more than one root sample and at least one aerenchymatous root, averaged for each genus or tribe) of aerenchyma in dry land (DL) and wetland (WL) habitat
Genusn (spec)r (DL)r (WL)P-valuemin nn (roots)
  1. Discovery rates approaching one indicate a high probability of detecting the ability to make aerenchyma. P-value refers to a Kruskal–Wallis rank sum test between r in DL and WL. Discovery rate did not differ between habitats. Minimum sample size for a 95% confidence of detection of aerenchyma (min n) and mean number of roots (n (roots)) and number of species (n (spec)) were investigated according to genus. Cells of P-value are empty if no P-value could be calculated. Restio 1, Restio except subgenus Simplicaulos; Restio 2, Restio subgenus Simplicaulos.

Askidiosperma21.001.00 11.2
Platycaulos1No spec1.00 11.1
Restio 1210.660.690.8531.8
Restio 21No spec1.00 13
Staberoha61.001.00 12.9
Thamnochortus0No specNo spec 12.6

Waterlogging Experiment

The absence of aerenchyma was positively correlated with dry land species and waterlogging intolerance, whereas the presence of aerenchyma was associated with wetland species and waterlogging tolerance (Table 3). A significantly higher proportion of aerenchymatous to non-aerenchymatous species were waterlogging tolerant, compared with the proportion that was waterlogging intolerant (Fisher's exact test, P = 0.044). Wetland species more often had aerenchyma than dry land species (Fisher's exact text, P = 0.001). Cannomois grandis was the only wetland species that showed mortality with progressive waterlogging, although all individuals of the species survived under the normal watering regime (Table S2).

Table 3. Waterlogging tolerance of dry and wetland species after 5 months of waterlogging. Most dry land species were waterlogging intolerant and did not have the ability to make aerenchyma (AER), whereas most wetland species were waterlogging tolerant and did have the ability to make this structure
Waterlogging intolerantWaterlogging tolerant
Without AERWith AERWithout AERWith AER
  1. R., Rhodocoma; T., Thamnochortus; E., Elegia; C., Cannomois.

Dry land species
 R. arida   E. ebracteata  
 R. gigantea    
 T. bachmannii    
 T. insignis    
 T. lucens    
 T. platypteris    
Wetland species
  C. grandis   E. fistulosa
    E. cuspidata
    E. tectorum
    R. paniculatus

Habitat–Aerenchyma Correlations and Geographic Analysis

In Riverlands, aerenchymatous individuals were more often found in wetland than well-drained habitats (Fisher's exact test, P = 0.014, Table S3). In the ecohydrology sites, aerenchymatous species were found under higher aeration stress (wSEVa value) than non-aerenchymatous species (KW test: P = 0.008; n(aerenchymatous) = 34; n(non-aerenchymatous) = 9) (Fig. 2). In the analysis of 547 co-occurrence plots from across all habitat types of the CFR, wetland plots contained a significantly higher ratio of aerenchymatous to non-aerenchymatous species than well-drained plots (KW test: P < 10−4, wetland: n = 166, mean(ratio aerenchymatous species) = 0.845, SD = 0.237; well-drained: n = 381, mean(ratio aerenchymatous species) = 0.584, SD = 0.272; SD = standard deviation) (Fig. 3). Well-drained plots were more species rich than wetland plots (KW test: P < 10−4, well-drained: n(plots) = 381, mean(number of species) = 5.01, SD = 2.57; wetland: n(plots) = 166, mean(number of species) = 3.745; SD = 2.35). Across the whole CFR, the ability of species to make aerenchyma was associated with wetlands as inferred at species level (Fisher's exact test, P < 10−4, see Table S4). The ratio of aerenchymatous species did not differ between regions of the CFR (χ2 = 2.98, d.f. = 8, P-value = 0.94, see Table S5).

Figure 2.

Scatterplot of aerenchymatous and non-aerenchymatous species according to dry (wSEVd) and aeration (wSEVa) stress. Aerenchymatous species had a higher aeration stress than non-aerenchymatous species. One data point represents one species recorded in the ecohydrology sites from Araya (2011).

Figure 3.

Box and Whisker plot of the ratio of non-aerenchymatous to total number of species according to habitat. Wetland habitats (cliff-seep, convex-seep, impeded drainage, marsh, seepage, streambank and valley-bottom) had a higher percentage of aerenchymatous species than well-drained habitats (cliff, well-drained). Habitats with the same lower-case letters are not significantly different from each other. The number (n) of sites per habitat is given in italic.

Phylogenetic Analysis

Both aerenchyma and habitat showed a strong phylogenetic signal (Fig. 4). Observed MP character step length for both characters was in the lowest 0.001% of the randomly reshuffled trees.

Figure 4.

One randomly selected phylogeny with aerenchyma (left) and ecohydrological niche (right) plotted on it with maximum likelihood approach. Both aerenchyma and ecohydrological niche showed a strong phylogenetic signal and co-evolved.

Aerenchyma was resolved as ancestrally present in all of the 500 trees in a MP analysis. Maximum likelihood reconstruction optimized aerenchyma as ancestrally present in 467 trees, whereas 33 trees were equivocal. In a BayesTraits ML reconstruction, the mean probability of ancestral aerenchyma across 500 trees was 0.7 (SD = 0.03). Parsimony analysis showed that aerenchyma was gained 2–13 times (average seven times) and lost 18–29 times (average 24 times). This was confirmed by the ML analysis, in which the two rates model was significantly better (α = 0.05) than the equal rate model in more than 95% of the trees. The two rates model was significantly better (α = 0.001) than the models in which only loss or gain was allowed in all 500 trees.

Ancestral habitat was reconstructed as wetland in all trees in the MP approach, whereas the ML reconstruction always gave equivocal results. With BayesTraits ML reconstruction, the mean probability of a wetland ancestral habitat was 0.51 (SD = 0.001). Shifts towards well-drained soils occurred 19–43 times (average 30 times), and shifts towards wetlands happened 14–37 times (average 26 times). Transition rates between the two states are not significantly different: the two rates model was not significantly better (α < 0.05) in any of the 500 trees than the equal rate model. The equal rate model was significantly better (LR > 2) in all 500 trees than the models in which only gain or losses were allowed.

The dependent evolutionary model was significantly better (α < 0.001) than the independent evolutionary model in all of the 500 trees, indicating co-evolution of the two traits. Transition rates show the evolutionary direction of the transitions (Fig. 5): aerenchyma was never lost when habitat remained wet, but was frequently gained under these conditions. Changes from well-drained to wetland habitats showed a higher transition rate in lineages with aerenchyma than in lineages lacking aerenchyma. Transition rates from wetland to well-drained habitats were higher for non-aerenchymatous lineages than the other way round. Well-drained soils were more frequently associated with the gain than the loss of aerenchyma; however, transition rates were low in both directions.

Figure 5.

Illustration of the transition rates of the independent evolutionary model. First number aerenchyma (0: absent; 1: present); second number, after full-stop, ecohydrological niche (0: well-drained; 1: waterlogged).

Maximum parsimony analysis showed a strong association between evolutionary changes of aerenchyma and habitat (Fisher's exact test, P < 10−4). The number of branches with aerenchyma absent or lost in waterlogged habitat was lower than expected. Similarly, the number of branches with aerenchyma or with gain of aerenchyma in well-drained habitat was also lower than expected (Table 4).

Table 4. Number of evolutionary transitions of aerenchyma and ecohydrological niche in a parsimony analysis
AerenchymaEcohydrological niche
  1. Ecohydrological niche: DL, dry land; WL, wetland; abs, absent; pres, present.

abs=>abs & pres=>abs7625
pres=>pres & abs=>pres77174


We show that the absence, but not presence, of aerenchyma is a reasonable predictor of both realized and fundamental ecohydrological niche in the Restionaceae. The absence of aerenchyma was highly correlated with well-drained conditions at individual and species level, and after phylogenetic correction, revealing a strong realized niche prediction. Although non-aerenchymatous species are very rare in wetlands, aerenchymatous species occur both in wetlands and on well-drained soils. Cultivation experiments of 12 species showed a significant positive correlation between the presence of aerenchyma and the ability to survive waterlogging. We suggest that environmental filtering leads to an exclusion of most non-aerenchymatous species from waterlogged conditions. This leads to species segregation along soil moisture gradients, as the fundamental ecohydrological niche of species lacking aerenchyma tends to be constrained to well-drained soils. The species pool for well-drained habitats is larger than that for waterlogged soils, as the former habitat can be occupied both by aerenchymatous as well as non-aerenchymatous species, whilst the latter is restricted to aerenchymatous species. This could be responsible for the greater species richness in dry land conditions.

Loss of Aerenchyma Definite

We observed the absence of aerenchyma in 93 of the 273 species investigated. These absences could be the result of experimental error, developmental flexibility or due to genetic modification. Experimental error could be due to sectioning roots, which are too young, or in which the aerenchyma was destroyed. Developmental flexibility could be due to roots in well-drained soils not making aerenchyma, although they have the ability to do so. Such facultative absence of aerenchyma has commonly been reported (Justin & Armstrong 1987; Jackson & Armstrong 1999; Visser et al. 2000).

There are several reasons why we infer that most of our absences of aerenchyma are due to genetic modification rather than phenotypic plasticity. First, the absence of aerenchyma is neatly clustered on the phylogeny, as is evident from the strong phylogenetic signal. False absences should give a phylogenetically random signal, unless the aerenchyma absences are due to a phylogenetic clustering of unsuitable habitats. However, discovery rates were the same in well-drained and wetland habitats, indicating that habitat should not lead to a clustering of false absences. Secondly, the species without aerenchyma, which we subjected to the waterlogging experiment, failed to produce aerenchyma. It seems that at least in these species the inability to make aerenchyma might be genetic rather than phenotypic. However, this experiment was subsequently undermined by the observation of aerenchymatous roots on herbarium specimens of two of the seven non-aerenchymatous species that were included in the experiment. It is not clear whether the inability to make aerenchyma is variable within these species, or whether our waterlogging experiment was too rapid, not allowing the plants enough time to make aerenchyma.

It has been proposed that cortical cell arrangement can be used to infer the ability to make aerenchyma, with hexagonally arranged cells less likely to make aerenchyma than a cortex of cubically arranged cells (Justin & Armstrong 1987). However, we did not observe such a correlation, and so could not use the anatomical structure of the roots to infer the probability of making aerenchyma. The sum of the phylogenetic analyses, the discovery rate test and the experimental test indicates that whilst occasional missed aerenchyma cannot be discounted, these are likely to be in a minority.

Can Absence of Aerenchyma Predict the Ecohydrological Niche?

We can to a certain extent predict the ecohydrological niches of Restionaceae species on the basis of their root anatomy, specifically whether Restionaceae have aerenchyma or not. Species lacking aerenchyma were rare in wetland habitats. This correlation was consistent when individuals in the local communities were investigated, as well as when presence of aerenchyma and habitat conditions were generalized to species level. However, aerenchyma might not be the only important trait that facilitates waterlogging resistance – E. ebracteata did not have aerenchyma but survived 5 months of waterlogging. In wetland habitats, most species had aerenchyma (84.5%, Fig. 3), whilst the proportion of species with aerenchyma was down to 58.4% in well-drained habitats. In the waterlogging experiment, most individuals without aerenchyma died when waterlogged. This does not mean that all species with aerenchyma can survive waterlogging and occur in this habitat – in our experiments, the aerenchymatous C. grandis, which is typical of alluvial soils along drainage lines, died when waterlogged. We only used a small number of species to assess the fundamental niche, which might limit the conclusions that can be drawn from this subset. However, for those species in which both the fundamental niche (waterlogging experiment) and realized niche (field observations) were investigated, the two niches were mostly similar.

Bioclimatic data, in particular rainfall, cannot account for the proportional presence of aerenchyma. The geographic regions in the CFR differ greatly in both total and seasonal distribution of rainfall, but had equal proportions of aerenchymatous species. We propose that impeded drainage leading to local wetland conditions is a main driver of local species distribution patterns, as has been reported for Rumex (Laan et al. 1989). The precise habitat within a particular local site is equivalent to what Silvertown called the alpha niche. This can be contrasted to the beta niche, which reflects the climatic variation patterns, and so is over a larger geographical scale (Silvertown 2004). This means that soil properties and other local modifiers might be very informative and should be used in combination with broad bioclimatic data when predicting local species distribution.

Niche Evolution

Aerenchyma optimizes as ancestral for the African Restionaceae in both MP and ML analysis. This is consistent with the widespread presence of aerenchyma in the Australian Restionaceae (Cutler 1969; Meney & Pate 1999), within which the African Restionaceae are embedded. It may indicate that Restionaceae ancestrally occupied a wetland habitat. This was already inferred on the basis of the culm anatomical variation, where the ancestral Restionaceae lack many of the adaptations typical of arid habitats (Linder 2000). It is also consistent with an evolutionary model of the Poales as a whole, suggesting initial diversification in the late Cretaceous in wetland habitats, with radiation into dry land habitats occurring mostly in the Neogene (Cutler 1969; Linder & Rudall 2005). Although aerenchyma is also found in dry land species, and so the presence of aerenchyma is not conclusive evidence for this ancestral habitat, the phylogenetic reconstructions are consistent with the previous data suggesting diversification of the Restionaceae in wetland habitats. Our attempts to optimize habitat directly onto the tree did not result in conclusive reconstructions. Parsimony optimization did indeed suggest an ancestral wetland habitat, but this could be affected by the great evolutionary lability of this character. Likelihood optimization, which is more robust to frequent changes, did not lead to significant result. Structural data may be more conservative than niche data, thus allowing more confident postdictions of ancestral habitats. If this is so, then it implies that the ancestral restioid niche may have been at least seasonally wet.

We speculate that with the development of seasonal dry climates during the Neogene (Zachos et al. 2001; Bytebier et al. 2010), wetland habitats in southern Africa became much rarer. This reduction in the extent of wetland habitat could also been accentuated by the middle Miocene and Pliocene uplifts of the subcontinent, which significantly steepened the river valleys and led to a loss of flat, peneplained landscapes (Partridge 1998; Partridge & Maud 2000). The reduction in wetland meant that lineages, which had lost the ability to survive in wetlands (by loss of aerenchyma), were no longer disadvantaged, and so could survive.

Aerenchyma was lost 18–29 times, but was much less often regained (on average seven times), and a model forcing equal rates of gains and losses is significantly worse than a model allowing variable rates. This suggests that loss of aerenchyma is an easier and more common process in the Restionaceae. The patterns of habitat change, however, were quite different, with numerous gains and losses, almost in equal frequency, of wetland and dry land habitat. However, in lineages in which aerenchyma was lost, there is hardly any further reversal to a wetland ecohydrological habitat. Niche evolution is therefore partially linked to the aerenchyma evolution, and loss of aerenchyma results in a loss of potential niche space. We do not know whether the loss of aerenchyma is an adaptive or a neutral process, and there is little information in the literature (Lambers, Atkin & Millenaar 2002; Ryser, Gill & Byrne 2011). Possible costs of aerenchyma include reduced mechanical stability (Striker et al. 2007) and lower competitive ability (Carter & Grace 1990) compared with non-aerenchymatous plants. However, the loss of aerenchyma could also be evolutionary neutral. An example of evolutionary loss of features under conditions in which they are not used are cave animals, whose eyes were reduced during the evolution in darkness (Poulson & White 1969). Such losses impact on the potentially available niche space and might therefor constrain the evolution of a clade.

A simple binary niche parameter, like the presence or absence of aerenchyma and the associated differentiation along the ecohydrological gradient, can explain only a small fraction of the remarkable diversity of the Cape flora. A more complete explanation will require the exploration of functional and evolutionary attributes of some of the other niche axes. Of these, maybe the more important are rootstock adaptations that allow resprouting (Schutte, Vlok & van Wyk 1995; Bell, Pate & Dixon 1996), leaf and culm adaptations that enable the evergreen plants to survive the long, hot, dry summers (Yates et al. 2010), and root adaptations to the very low soil nutrients (Lamont 1982; Lambers et al. 2008). We are still a long way from understanding why there are so many species in the Cape flora, and how they can coexist.


We are grateful to Mike Cramer, Rupert Koopman, Eliane Furrer and Franziska Perl for their help in the field, to Markus Meierhofer and Melanie Ranft for their commitment in the greenhouse, and Carmen Hiltebrand and Jolanda Steiner for their meticulous anatomical work. We thank Jonathan Silvertown and Yoseph Araya for the data from the ecohydrology sites. Fieldwork was funded by the Swiss Academy of Science. Cape Nature gave permission to conduct fieldwork in Riverlands and provided logistic support. We particularly thank Rupert Koopman for his input and assistance in this project, and David Ackerly and two anonymous referees for detailed and constructive input.