Functional acclimation across microgeographic scales in Dodonaea viscosa

We studied a native Australian shrub—Dodonaea viscosa, or sticky hop bush—in the wild and in a gardening experiment and found that the species can readily adapt to different environments. Our findings are interesting because the plants we used came from sites with quite different environmental conditions, although they were only short distances apart. Our findings indicate that the potential risks associated with moving plants between sites with different environmental conditions are not likely to cause negative outcomes for restoration projects using this species, which is commonly used for restoration in southern Australia.


Introduction
Exploring variation in functional traits across environmental gradients provides mechanistic insights into the persistence and function of widespread species (Ramírez-Valiente et al. 2010;Bolnick et al. 2011;Funk et al. 2017). Similarly, such studies can be combined with fitness measurements to infer adaptive capacity of populations, and can therefore be used to predict the success of plant populations and derive seed sourcing options for revegetation under climate change (Laughlin 2014;Breed et al. in press). However, most studies that explore intraspecific trait variation do so in systems where geographic and environmental distances co-vary (e.g. climate distance correlates with geographic distance) (Hulshof et al. 2013;Carlson et al. 2016). Such co-variation makes it difficult to determine the specific abiotic driver(s) of trait variation and limits the utility of this type of research (Meirmans 2015;Caddy-Retalic et al. 2017). While there are statistical ways to account for this issue post hoc (Dormann et al. 2007;Hothorn et al. 2011), it remains preferable to associate environmental and trait variation in systems where environmental distance varies independently of geographic distance.
Studies that seek to make conclusions on plant adaptive capacity can usefully explore trait variation across environmental gradients at microgeographic scales (e.g. tens of kilometres for trees and shrubs). These smaller spatial scales provide greater potential for gene flow to occur between populations, enabling a test of whether the influx of genes from non-local populations is 'swamping' local gene pools. An influx of non-local genes could increase trait variation, and potentially prevent local adaptation (Richardson et al. 2014). Including a common garden component to these studies helps to control for the environmental influence on traits, allowing an understanding of whether trait variation is under genetic vs. plastic control (Poorter et al. 2016). When studies are conducted in such a way, it should be possible to quantify species trait variability in the presence of gene flow. Similarly, such studies (supported by reproductive traits), and combined with fitness measurements, can be used to infer adaptive capacity of populations. This information is useful in a conservation context as it can help to predict the impacts of habitat fragmentation, which tends to reduce interpopulation gene flow, disrupt plant mating systems and reduce population sizes (Young et al. 1996;Lowe et al. 2005Lowe et al. , 2015Vranckx et al. 2012;Breed et al. 2013Breed et al. , 2015. These impacts can each reduce the adaptive capacity to climate change of plant populations (Breed et al. 2012;Christmas et al. 2016b). However, there are surprisingly few fine-scale studies of trait variation reported (Richardson et al. 2014).
Dodonaea viscosa (Sapindaceae) (hereafter Dodonaea) occurs in Australia across a broad and variable environment, where it displays clinal variation in a number of functional traits (Guerin et al. 2012;Guerin and Lowe 2013;Hill et al. 2015;Baruch et al. 2017). In the Mt Lofty Ranges (South Australia), Dodonaea occurs continuously across a short latitude span (34.6 to 35.5°S = ca. 100 km), but across a variety of elevations (from sea level up to 300 m). This short geographic distance displays large variation in aridity, precipitation and temperature ( Fig. 1; Table 1). Dodonaea also displays low levels of population genetic structure across this region, indicating considerable gene flow (as observed in other plant species; McCallum et al. 2014), but selection appears to have been sufficiently strong to maintain adaptive genetic variation (Christmas et al. 2016a. To the best of our knowledge, no detailed trait-based studies have been conducted in any plant taxon in this region. As such, this environment is an excellent system to study functional trait variation, and investigate whether trait variation still occurs despite likely gene flow across population in this area. Here we assessed trait variability of Dodonaea adult field populations, and compared these data with trait values of offspring grown in a greenhouse common garden trial. We measured functional traits associated with plant architecture, competitive ability and dispersal capacity (i.e. height, wood density, growth rate and seed mass). We also characterized leaf traits that modulate leaf temperature, transpiration and photosynthesis, as well as resource use efficiency on short and long timescales (i.e. specific leaf area (SLA), nitrogen content per unit mass and area, carbon and nitrogen isotope ratios, stomatal size and density). We used these data to explore the following study goals: (i) quantify trait variability of parents and offspring, and link this variation to the parental environments; (ii) determine which traits best explain population differences; and (iii) compare parent and offspring trait-trait relationships.

Study species and field sampling
Dodonaea is a 1-4 m tall woody shrub, with upright, narrow, tough and sticky leaves covered by reflective wax. Pollination is by wind and insects, and seeds are dispersed by wind and ants. Dodonaea is widely distributed throughout the southern half of Australia, predominantly on well-drained soils, and it can re-sprout after fire (Hodgkinson 1998). Locally, it forms sparse to dense cover in shrublands and in open woodlands as a recognizable shrub layer (Hodgkinson and Oxley 1990).
We selected eight parent populations in the Mt Lofty Ranges, South Australia. Local climates varied considerably due to differences in altitude and its effects on precipitation, aridity and air temperature. All populations (henceforth designated by their initials in Fig. 1 and Table 1) were located in remnant vegetation within nature reserves ( Fig. 1; Table 1). Between October and December 2015, we collected five sun-exposed terminal branch segments from 10 non-neighbouring parent plants (henceforth families), separated by >5 m, which were stored at 4 °C until processing. Concurrently, we also harvested mature fruits from across the canopy of each parent plant, which were stored in paper bags in dry, room temperature conditions.

Parent populations handling and traits
We measured 11 traits in the parent cohort for our analyses (Table 1). Five of the youngest, fully expanded leaves of each parent plant were excised to estimate leaf area by scanning and measuring with ImageJ (Schneider et al. 2012). Each leaf was then oven dried at 65 °C for 48 h and weighed. Specific leaf area (the ratio of leaf area to mass) was calculated (n = 400; 5 leaves × 10 parent plants (families) × 8 populations). Wood density was determined by the dimensional method, described in Pérez-Harguindeguy et al. (2013), from the proximal end of one branch segment per parent from each population. Uniform ca. 3 cm segments (close to cylindrical and care was taken to avoid rough sections) were cut and their length and diameter measured with digital callipers, to obtain volume. These segments were then oven dried at 85 °C for 72 h and weighed. Wood density is the ratio of dry weight and volume (n = 80; 10 families × 8 populations). Stomatal size and density were determined on 1 cm 2 leaf segments excised from the middle of the abaxial lamina of one leaf per parent (n = 80). The cuticle preparations and measurements were the same as described in Hill et al. (2015).
Carbon and nitrogen isotope (δ 13 C; δ 15 N) ratios and element content (C mass ; N mass ) were determined on dried leaves of parent plants (n = 80). Leaves were ground in 2 mL screw cap micro centrifuge tubes with a Qiagen TissueLyzer (Dusseldorf, Germany) adaptor for a Retsch ball mill (Mixer Mill 400, Haan, Germany). Ground material (ca. 2.5 mg) was weighed in tin capsules. Samples were analysed at the University of Adelaide on a EuroVector Euro EA inline (Pavia, Italy) with a Nu Instruments Continuous Flow Isotope Ratio Mass Spectrometer (CF-IRMS, Wrexham, UK). Internal isotope standards run alongside were glycine and glutamic acid. Certified reference material for elemental concentration was triphenyl amine (TPA; C:N). The uncertainty of 13 C and 15 N measurements was less than ±0.09 and ±0.12 ‰, respectively. The C:N ratio was assessed as well as the N content per unit leaf area (N area ), which was calculated from N mass and SLA. Fruits were manually crushed to liberate seeds, and seeds were weighed in sets of 50 per parent to obtain mean seed weight.

Offspring population handling and traits
We measured 13 offspring traits, 9 of which were also measured in the parent plants (Table 1). Seed dormancy was broken by soaking seeds for 5 min in recently boiled water (Baskin et al. 2004). Five replicates of 20 seeds per family were set to germinate on moist filter paper in Petri dishes. These dishes were arrayed on a greenhouse bench, moistened as required with demineralized water, and their position in the greenhouse randomized fortnightly. The protrusion of the radicle indicated germination, and its presence was scored three times per week over 2 months (from February to April 2016), and these seedlings were then discarded. During this germination trial, mean air temperature was 16.5 °C (range: 3.7-40.5 °C) and mean relative humidity was 77.3 % (range: 21.1-99.9 %) recorded with a Senonics Minnow data logger (MicroDAQ, USA). Shade cloth covered the roof of the greenhouse to decrease photosynthetic active radiation (PhAR) irradiance, which was between one-third and half of that outside as measured with a quantum sensor Q 34327 and LI-1400 data logger (Li-COR, Lincoln, NE, USA). In January 2016, 20 replicates per family per population (20 × 10 × 8; n = 1600) were sown each with five seeds over commercial native potting mix (Grow Better) in standard forestry tubes (5 × 5 × 12 cm; see Supporting Information- Fig. S1 for plant images). These were distributed randomly into trays whose position in the greenhouse was randomized fortnightly. Spray watering was provided three times per week, for 10 min at 3-h intervals. Pots were fertilized once with 0.2 g slow release Osmocote Native Mix. To minimize selection for fitness, if more than one seed germinated per pot, the most central seedling was chosen and the rest were removed. Seedling height was scored twice: (t 1 ) at 96 days after sowing and (t 2 ) on 173 days after sowing. Instantaneous relative growth rate was calculated as relative growth rate (RGR) = (log e HEIGHT 2 − log e HEIGHT 1 )/(t 2 − t 1 ). We employed the non-destructive, seedling height as a proxy for plant weight. Survival on both dates was recorded.
In September 2016 (234 days after sowing), we sorted the seedlings from every family by height and selected the four closest to the median (n = 320 seedlings: 4 replicates × 10 families × 8 populations). These were transplanted to larger polyethylene pots (8.5 × 8.5 × 17.5 cm) with the same potting mix, and transferred to a greenhouse bench where mean air temperature was 17.89 °C (range: 4.75-31.53 °C) and mean relative humidity was 70.80 % (range: 35.23-90.25 %). Photosynthetic active radiation was about half of that outside, and both temperature and PhAR were monitored and recorded as above. Plants were watered for 10 min twice per day, four times per week, and top-dressed with 0.6 g of the same fertilizer. Thirteen months after sowing, 80 young but fully expanded leaves (one leaf per randomly selected individual per family) were excised to measure SLA, stomatal density and size, as well as isotope ratios and elemental content as above. Leaf thickness of offspring was determined on 80 mature leaves (10 families × 8 populations) as the mean from 10 measurements per leaf at ~1 mm from the mid-vein at the centre of the leaf on the same 80 individuals.

Environmental variables
We included seven environmental variables in our analyses, including the aridity index (AI), mean annual precipitation (MAP), mean annual temperature (MAT), and soil nitrogen (N), phosphorus (P) and clay content (see Table 2 for further details). Environmental data were sourced from the Atlas of Living Australia at 0.01° (~1 km) resolution (http://www.ala.org.au; accessed 15 December 2016) (Williams et al. 2012). Geographic linear distances between populations were estimated from GPS coordinates and suitable software (http://www. boulter.com/gps/distance/).

Data analysis
We used one-way ANOVAs to test for environmental differences and parent and offspring trait differences between populations. We used linear mixed-effects models and ordinations to explore variation in traits, trait-trait correlations and trait-environment associations. All traits were treated as continuous response variables. Supporting Information-Appendix S1 describes our statistical analyses in more detail.
To reduce redundancy in the environmental predictor variables, we ran a principal component analysis (PCA) to obtain an integrated, composite environmental ordination for each population. Since soil N, P and clay content were very different at one population (ASCP; Table 1), the PCA was only run on climatic variables, which were all strongly associated with elevation. Parent mean trait and offspring traits were correlated against their position along the environmental PCA1 and PCA2 axes. We also ran PCAs of parent and offspring functional traits to assess their relative influence within these two cohorts. To test for the association and significance of the multidimensional parent and offspring trait spaces, their respective Euclidean interpopulation distance matrices were computed and then correlated with the Mantel test (Mantel 1967). To account for the effect of environmental differences and geographical distances, partial Mantel tests (Smouse et al. 1986 To explore trait-trait correlations, we ran linear regressions in SYSTAT (2002) for each pair of traits within parent and offspring cohorts. Traits shared between parents and offspring were also regressed. We used sequential Bonferroni adjustment of the P threshold for multiple testing. We fitted linear mixed-effects models in R (R Core Team 2017) using the package lme4 (Bates et al. 2015) to test for differences in offspring relative growth rate (Gaussian distribution) and height (Gamma distribution) between populations, including family as a random effect nested within population. We generated P values and 95 % confidence intervals for the effect of population on relative growth rate and height using parametric bootstrapping methods (1000 iterations) implemented in the R package 'afex' (Singmann et al. 2017).

Environmental and geographic distances
Dodonaea populations were separated by 1.4 to 96.8 km ( Fig. 1; see Supporting Information- Table S1) at elevations that ranged from 17 to 299 m above sea level (Table 1). As expected, elevation regulated local climates, where more arid and warmer sites were at lower elevations (ASCP, FHRP, BHLO, SCCP) and the less arid and cooler sites were at higher elevation (MBCP, BENP, BHHI, WACP). Aridity and air temperature were higher at sites below 200 m above sea level, while rainfall was greater at higher elevation sites (Table 1). Soil texture, N and P content did not differ between these two groups, but soils from ASCP were unlike the other populations, with considerable less clay and more nutrients ( Table 1).
The ordination of populations within environmental space clearly separated low and high elevation populations ( Fig. 2A), where PCA1 was strongly correlated with all climatic variables [see Supporting Information- Table S2], and explained 79.9 % of the total variance. The AI and MAP were associated negatively with environmental PCA1 and MAT was associated positively with environmental PCA1. Environmental and geographical distances were uncorrelated in the ordination plot ( Fig. 2A), confirmed by the lack of significant multivariate association between them (Table 2).

Trait differences in parent and offspring cohorts
Apart from N mass , all traits differed significantly across parent populations (Fig. 3A-H Fig. S2) and relative growth rate (χ 2 = 28.09, P < 0.001; see Supporting Information- Fig. S2). For height and relative growth rate, linear mixed-effects models (assuming normal error distribution) showed that family (random effect) explained very little variation in either trait. Survival on day 173 after sowing was notably low in ASCP and BHHI populations (50.5 and 62.5 %, respectively), where survival in all other populations was >90 %. Table 2. Geographic coordinates, elevation, climate and soil variables of the eight Dodonaea viscosa subsp. angustissima study populations. C.P. = Conservation Park; N.P. = National Park; R.P. = Recreation Park; AI = aridity index = an inverse scale of aridity, where high values indicate less arid climates; MAP = mean annual precipitation; MAT = mean annual temperature; N and P = pre-European plant available soil N and P stores. Mean leaf area and SLA were substantially greater in offspring than in parent populations (ca. 2.0-and 3.5fold, respectively). Conversely, mean N mass , N area and the δ 13 C ratio were higher in the parents (1.9-, 1.6-and 2.1fold, respectively) ( Table 2; Figs 3 and 4). Dodonaea germination was uneven, FHRP, MBCP and SCCP were the first to germinate after 26 days, and also displayed the highest germination percentages, whereas ASCP had the poorest germination (Fig. 4A).

Parent and offspring trait-environment correlations
Traits of parent and offspring displayed distinctive correlations with environment PCA1 and PCA2 ( Fig. 5; see Supporting Information-Tables S4 and S5). For parents, leaf area and SLA were negatively correlated to PCA1, implying that they increase with elevation, precipitation and lower aridity. The δ 15 N ratio was positively correlated to PCA1, indicating that it increased at lower elevation in warmer and more arid sites. In offspring, only relative growth rate was correlated to PCA1 ( Fig. 5D; see Supporting Information-Tables S4 and S5).

Parent and offspring in multi-trait space
The ordination of parent populations in trait space was quite different to the environmental space ordination that showed environmental disparity between high and low elevation populations (Fig. 2B). Leaf area, SLA, stomata size, stomata density, both isotope ratios and N mass had the largest influence on the ordination along PCA1 (50.6 % of total variance explained), and only mean seed weight was associated to PCA2 (19.5 % of total variance explained) (Fig 2B; see Supporting Information- Table S2). A negative correlation between stomata size and stomata density was evident [see Supporting Information- Table S2]. The ordination of offspring was less structured than parents as both ordination PCA axes captured equally low percentages of total variance (29.4 and 23.5 %, respectively). Leaf area, SLA, leaf thickness and stomata density defined the ordination along PCA1, whereas relative growth rate and N mass defined the ordination along PCA2 ( Fig. 2C; see Supporting Information- Table S2). In this offspring trait ordination, populations from cooler and wetter sites appeared to be more clustered than those from warmer and drier sites. The distinct ordinations of parent and offspring populations in trait space were confirmed by the lack of correlation between their respective distance matrices, even when controlling for geographic and environmental distances using a partial Mantel test (Table 3).

Parent and offspring trait-trait correlations
Trait-trait correlations differed in parent and offspring populations. In parents, apart from the obvious links (e.g. leaf area with SLA, and N mass with C:N and N area ), SLA was strongly and negatively correlated with N area and δ 13 C ratio ( Fig. 6A and B; see Supporting Information-Tables S6 and S7). Wood density also correlated with δ 13 C ratio [see Supporting Information- Table S7]. In offspring, SLA was strongly and negatively correlated with N area (Fig. 6C; see Supporting Information-Tables S6 and S7). Stomata size and stomata density were negatively correlated in both parents and offspring ( Fig. 6D

Information-Tables S6 and S7
). Mean seed weight did not influence germination, height or relative growth rate [see Supporting Information- Table S7].

Discussion
Dodonaea displayed significant variation in a broad variety of established functional traits associated with climate over microgeographic distances (from 1.4 to 96.8 km). This variation was likely as a result of acclimation to elevation and/or aridity, and not as a result of genetic differentiation of traits between populations. By studying trait variability across short geographic distances and in a system where environmental and geographic distances were uncorrelated, our results increase our understanding of studying functional trait variability, which is most often done across much larger geographic areas (hundreds of kilometres) (Guerin et al. 2012;Hill et al. 2015;Baruch et al. 2017), and where geography often confounds environmental distance (Meirmans 2015;Caddy-Retalic et al. 2017). We conclude that most observable variation in Dodonaea functional traits was likely a result of environmental acclimation rather than genetic adaptation or differentiation. As such, Dodonaea displays considerable adaptive capacity to acclimate to climate change in situ. We therefore suggest that future work on climate adaptation in woody plants should further consider the large amount of trait variation present over short geographic distances, and account for the confounding effect geography has on associating environmental with trait variation. Such robust study designs allow more confident statements about the drivers of adaptive capacity, which is essential to better inform conservation and restoration management strategies, such as designing in situ conservation areas and selecting seed for restoration under climate change.
The parent plants in our study system showed great variation in functional traits, with 8 out of 11 differing among populations. Only leaf N content and related did not differ significantly between populations, possibly due to relatively homogeneous site soil N content (Ordoñez et al. 2009). A large portion of the phenotypic variability was lost when offspring were grown in the common garden, implying strong plasticity of most traits (only 4 out of 13 of the tested traits differed significantly between offspring populations). Running our common garden experiments in a greenhouse with uniform and favourable growth conditions may have contributed to this high plasticity (Poorter et al. 2016). Further experimental work would also assist with teasing plasticity from genetic effects in this system, such as reciprocal transplant studies across the environmental gradient, and comparing trait vs. neutral genetic differentiation (Breed et al. 2018). It is also possible that differences in ontogenetic stage between mature individuals and plantlets contributed to their trait differences. Our results were consistent with previous work on Protea repens, a sclerophyllous shrub in South Africa, in terms of trait responses to aridity (Carlson et al. 2016). We also observed a doubling of SLA and leaf area in the offspring grown in the greenhouse trial, indicating considerable trait plasticity, as previously reported (Zhao et al. 2012). Indeed, lower field SLA was probably caused by higher irradiance in conjunction with higher temperatures (Poorter et al. 2009(Poorter et al. , 2014. However, leaf N mass was significantly lower in offspring; a dilution effect that can potentially be attributed to by rapid growth and expansion of leaves under the favourable growing conditions of the greenhouse. Despite the climatic and elevational gradients in our study being shorter than those previously sampled (Baruch et al. 2017), leaf area and SLA responded in a consistent fashion, where both became smaller under more arid and warmer conditions. This is a common response in plants (Poorter et al. 2009;Guerin et al. 2012;Guerin and Lowe 2013;Carlson et al. 2016), but such a parallel response in leaf area and SLA along short and long environmental clines emphasizes the importance of studying both traits. Small leaves offer better control of temperature, while low SLA provides more efficient use of resources in stressful habitats (Poorter et al. 2009). Average seed mass was higher in wetter sites at higher elevation, which was an expected response to more resources and less stressful conditions. Wood usually becomes denser under more stressful environments as narrower xylem vessels control better water flow through the stem (Martínez-Cabrera et al. 2009). However, we did not observe a change in wood density across sites, possibly because our regional drought stress cline was relatively weak. Liu and Noshiro (2003) have also previously reported a lack of altitudinal and latitudinal difference in D. viscosa wood anatomy.  The leaf δ 13 C ratios indicated less carbon isotope discrimination (i.e. more positive δ 13 C) as climate became more arid, indicating a higher long-term water use efficiency (Farquhar et al. 1989). This is the typical response of plants along drought clines (Prentice et al. 2011;Schulze et al. 2014). Leaf N mass and N area did not respond to the environmental cline, possibly because of an absence of variation in soil N content, as discussed above. However, leaf δ 15 N ratios were positive in the less arid sites, which tend to have higher soil N availability. Nevertheless, δ 15 N is a complex trait that also scales with short-term variation in N cycling and the availability of N supply (Craine et al. 2015;Dong et al. 2017).
Along our environmental cline, stomatal density decreased but their size increased under cooler and wetter conditions, which is also a common, but environmentally dependant, response in plants (Franks et al. 2009;Hill et al. 2015;Carlson et al. 2016). Small and dense stomata promote transpiration cooling under hotter environments. Adjusting stomatal size and density is a powerful plastic mechanism for thermal regulation and gas exchange in local populations (Franks et al. 2009).
Most of the traits surveyed here are part of the 'leaf economics spectrum' and, for parent populations, showed the trade-offs consistent with longer and steeper environmental gradients (Wright et al. 2004). As such, we were able to compare our results in a wellestablished empirical and theoretical context. Specific leaf area is prevalent and central among functional traits because it scales positively with growth and photosynthetic rate, and negatively with leaf longevity (Wright et al. 2004). Parent SLA was strongly and inversely correlated to the δ 13 C ratio, which is associated with carbon assimilation and long-term water use efficiency (Farquhar et al. 1989). Populations with higher SLA displayed higher δ 13 C discrimination, as anticipated for plants under more favourable conditions at higher elevation. Similar results have been observed along aridity gradients (Prentice et al. 2011). Also, we found a negative correlation between SLA and N area , which is characteristic of fast-growing individuals (Wright et al. 2004;Onoda et al. 2017). Further, low SLA in dry environments is usually associated with high N area .
The positive association between parent δ 13 C and N area was expected, as high N area enhances photosynthetic capacity reducing internal CO 2 concentration (ci:ca) and increasing δ 13 C (Cernusak et al. 2013). The δ 13 C ratio and stomatal density were positively correlated through the direct association of δ 13 C to ci:ca, which is regulated by stomatal conductance (Cernusak et al. 2013). Therefore, it was expected that the δ 13 C ratio associated with stomatal traits, such as density or size, as well as between stomatal traits and N area .
Specific leaf area and stomatal density were negatively but weakly correlated, as anticipated from their respective associations with δ 13 C discussed above. Parent populations differed significantly in SLA, which was inversely (although weakly) correlated to wood density, supporting the relationship expected for these traits where faster growing individuals (with high SLA) had low wood density (Chave et al. 2009). Also, the δ 15 N ratio and N area were positively correlated, following the trend observed in other studies and attributed to high N availability (Craine et al. 2015;Dong et al. 2017). In offspring populations, only the correlations between SLA and N area , and stomatal density and stomatal size, persisted. The absence of correlations in plants under favourable conditions indicates the importance of resource limitation for maintaining the web of coordinated functional leaf trait responses, and also the high plasticity Dodonaea displays for these traits. At the whole-plant level, the relationship between the reproductive traits (seed mass and germination) and seedling performance (height and relative growth rate) was either absent or reduced, possibly as a result of the environmental uniformity of the common garden.

Conclusions
Results from our common garden experiment demonstrate that a range of key functional traits in Dodonaea were highly plastic (e.g. SLA, leaf nitrogen content, carbon and nitrogen isotope ratios, stomatal size and density). We sampled populations over microgeographic distances with strong environmental gradients and observed no clear genetic differentiation in most functional traits. As such, Dodonaea has great adaptive potential via functional acclimation, at least in this region. However, in the presence of increased barriers to gene flow with habitat fragmentation (Young et al. 1996), as is likely for many Dodonaea populations in this region (Guerin et al. 2016), there is a greater chance that the observed plastic trait divergence could accumulate a genetic basis as gene flow is reduced between populations leading to genetic assimilation (Waddington 1953;Pfennig et al. 2010). If this process plays out, these populations would likely have a reduced capacity to adapt to climate change due to increased genetic basis in these functional traits and reduced interpopulation gene flow. Future studies in this study system should therefore assess trait plasticity in populations from more fragmented and isolated habitat patches, rather than from the large and intact populations we studied here. In addition, it should be noted that high plasticity, as observed here, may indeed be adaptive in the presence of high environmental variation. Further, since Dodonaea is commonly used for restoration in southern Australia (Monie et al. 2013;Pickup et al. 2013), and that local populations likely house considerable trait variability via acclimation, the potential risks of transferring seed (Breed et al. 2013) across this study region are not likely to be a significant issue.

Supporting Information
The following additional information is available in the online version of this article-Appendix S1. Table of statistical analyses performed throughout the study. Figure S1. Images of Dodonaea viscosa (A) seedlings at 3-month-old, (B) at 12-month-old and (C) an adult female bearing ripe fruit in the field. Figure S2. Box plots of offspring traits that displayed significant variation across populations and tested with linear mixed-effects models, with family nested as a random effect within population. Table S1. Linear geographic distances between populations. Table S2. Pearson correlation coefficients of environmental variables and traits of parents and offspring after principal component analysis (PCA) ordination. Table S3. ANOVA table for trait differences between  populations for parent and offspring populations.  Table S4. Coefficients and significance of the regressions between traits and environmental composite axis PCA1. Table S5. Coefficients and significance of the regressions between traits and environmental composite axes PCA1 and PCA2. Table S6. Trait-trait correlations in parents, offspring and parent-offspring shared traits. Table S7. Pearson correlation coefficients of trait-trait correlations.