Seed collection and germination

Experimental design: Glasshouse conditions

In March 2011 (autumn), 370 seedlings were transplanted from the nursery into glasshouse conditions with ambient light conditions and a temperature set to cool alpine conditions of 20/10 °C day/night with a natural light cycle. Seedlings were potted into 175 mm diameter/2.8 l pots containing a mix of 80:10:10; commercial potting mix: field soil: and washed river sand with added slow release osmocote for natives (Debco Pty Ltd., Victoria, Australia). Field soil was used to provide the seedlings with a natural inoculum, and was collected from a site within the Kosciuszko National Park where all three species co-occur. Initial leaf number was counted and seedlings of uniform size were selected from each maternal line to minimize variance within and between lines for each species. Following time to adjust to transplanting, pairs of seedlings were allocated to three blocks in a stratified design (May 2011); one member of each pair was randomly assigned to a well-watered treatment, and the other to a dry treatment. Seedlings were placed on a bench in a randomized design (minimum n = 2 per block), however, due to variation in germination and establishment success the design was not perfectly balanced (Table 1).

High and low water treatments were applied using a Water-Pro vapour pressure deficit (VPD) controlled system (MicroGrow GreenHouse Systems, Temecula, California), with each plant watered individually by a dripper at soil level [33]. Drippers were calibrated to a standard flow rate that was checked at the beginning and end of the experiment. Watering events were triggered when pre-set VPD targets were reached. High water plants received ~160 ml per watering (enough to saturate the soil); low water plants received ~80 ml per watering. Watering occurred roughly once per week for low water plants and every 2–3 days for high water plants, depending on VPD in the glasshouse. Under this regime, low water plants received ~25% of that which the high water plants received.

Physiological and leaf chemistry measurements

In order to understand how the soil moisture treatments affected plant physiology, we measured changes in gas exchange rates, leaf morphology, and tissue chemistry on a representative subset of plants, the number of which varied among species (Table 1). Gas exchange was measured using a LiCor 6400 portable photosynthesis system (LiCor, Lincoln, Neb) after 6 months of growth (26/11/2011 to 16/12/2011). On each plant, a young fully expanded leaf was selected for measurement. Measurements were taken at two time points; on the morning following a watering event (wet cycle) when the plants were at their least water-limited, and on the final day of a watering cycle (dry cycle) when the low water plants were at the driest point in the cycle. Block temperature was set at ~20 °C, CO₂ was 400 ppm, relative humidity was maintained around 60% and light levels were 500 μmol m² s. Preliminary assessment indicated that this light level was saturating. Photosynthetic measurements were done between 8:00 till 13:00 h, before stomatal closure.

Leaf samples were taken to measure leaf size and specific leaf area (SLA, cm⁻² g⁻¹) at the time of gas exchange measurements. Collected leaves were placed with petioles positioned in saturated florist foam and kept at 4 °C in the dark overnight before being weighed to determine saturated weight, scanned on a flatbed scanner, and dried at 60 °C for 72 h before being weighed again to determine dry weight [34]. Leaf size was calculated as an area (cm²/per leaf), and specific leaf area was calculated as area/dry mass (cm⁻² g⁻¹).

For each species by elevation combination that had been measured for photosynthesis, replicates from three to four maternal lines were selected for isotope analysis. From these lines, three replicates from each of the high and low water treatments were chosen to be measured from each block (n = 112). Carbon isotopic (δ¹³C) and elemental analysis (%N) were performed on an isochrome continuous flow stable isotope ratio mass spectrometer (Micromass, Middlewich, England).

Harvest measurements

In November 2011 (late-spring) all plants were harvested. The total number of leaves on each plant was counted. Plant leaves, stems and roots were separated and dried in an oven at 60 °C for 72 h before weighing using a Mettler-Toledo AB304-S Analytical Balance (Mettler-Toledo, Switzerland). For the few plants that flowered, total flower number was counted and reproductive biomass weighed. Above-ground biomass was calculated as the sum of leaves and stem dry weight (g), with reproductive biomass included for those that flowered.

Statistical analysis

Trait means were compared using a restricted maximum likelihood (REML) model that included elevation (categorical), treatment, and their interaction as fixed factors. Random factors were block and maternal line. Individual replicates were nested within maternal line, which itself was nested within elevation. For traits where there was not sufficient replication at a given hierarchical level (eg. block or maternal line) to fit the model, the term was excluded.. Each analysis was conducted independently for each of the three species. Comparisons between wet and dry cycle traits revealed no significant difference, so only dry cycle results are presented henceforth.

Traits were examined for conformity to the assumptions of normality, and transformed as required. Data outliers were excluded when associated with data entry, or where for example drippers had failed to deliver the allocated water dose for the treatments [35].

To calculate plasticity across watering treatments, plants were paired within maternal line x treatment combinations within each block. This was performed per each species. Then, we calculated the plasticity index (PI) as: (largest trait value – smallest trait value)/largest trait value. Not all plants could be paired, e.g. if the individual in the other treatment had died. Where more than one pair was possible within a maternal line*block, plants were paired by proximity on the glasshouse bench to minimize differences in conditions. As a result of the study design the number of pairs varied across the three species (A. glacialis = 25, O. eriopoda = 62 and W. ceracea = 58).

The potential for adaptive, maladaptive, or neutral plasticity was analyzed using selection gradient analysis, where the average fitness of plants across environments (or a proxy thereof) is regressed against the index of trait plasticity [15]. We standardized plasticity indices and trait values both to a mean of zero. We used total biomass as a fitness proxy and calculated relative fitness as the mean total biomass for a pair of plants divided by the mean of all pairs, and then subsequently log (e) transformed [24, 36]. Selection differentials were assessed separately for each species. Using a REML model (lmer, package lme4, R) [37], the selection gradient model assesses relative fitness regressed against trait plasticity as a fixed factor, with maternal line nested within elevation as a random factor. We also assessed plasticity against relative fitness for each water treatment individually to investigate which treatment, if any, may be driving adaptive plasticity. Here, relative fitness was again regressed against plasticity, using the maternal line as a random factor. Using the same models as above, we also investigated how trait values regressed against fitness to see if there was selection upon the trait mean itself.

Responses to water availability

Reducing water availability to 25% of saturated conditions affected many morphological and physiological traits across the three species. As predicted, total biomass was significantly lower under water limitation for O. eriopoda and W. ceraceae whilst in A. glacialis this was found only to be true for the high elevation plants (Fig. 1a–c). In conjunction with reduced growth, a greater investment into water-sourcing root relative to shoot biomass was observed in all species under water limitation and was significant for both A. glacialis and W. ceracea (Fig. 1d, f). Total leaf number was lower under water limitation in all but the A. glacialis low elevation plants (Fig. 1g–i). Reductions in leaf size were also observed under water limitation and were significant for A. glacialis (Fig. 1j). We predicted that SLA would also be lower under water limitation, but changes in SLA were varied across the three species, with no clear main treatment response observed (Fig. 1 m–o; Table 2).

	Treatment			Elevation			Treatment x Elevation
Plant Functional Traits	Ag	Oe	Wc	Ag	Oe	Wc	Ag	Oe	Wc
Morphological
Total Biomass		●	●				●
Root Mass Ratio	●		●			●
Total Leaf Count	●	●	●			●
Leaf Area	●	●
Specific Leaf Area							●
Physiological
Photosynthesis
Conductance	●	●	●			●
Water Use Efficiency
Tissue Chemistry
% Nitrogen	●	●
δ13			●		●	●	●	●

Significant responses are shown by black dots (p < 0.05)

Physiological and tissue chemistry responses to water limitation were more varied. As predicted, all three species had significantly lower stomatal conductance (gs) under water-limiting conditions (Fig. 2a–c). Rates of photosynthesis were also lower under water-limiting conditions, though these differences were not statistically significant (Fig. 2d–f). Consequently, instantaneous water use efficiency (WUEi), being the ratio of photosynthesis to stomatal conductance, while slightly greater under water limitation was not significantly so (Fig. 2g–i). Similarly, this pattern was also observed across species using the longer term indicator of water use efficiency, δ¹³C, although less negative values, indicating increased water use efficiency, were observed for water-limited plants within W. ceracea (Fig. 2l). Leaf nitrogen content was lower under water limitation for A. glacialis (Fig. 2m).

Elevation effects and interactions

We found little evidence of intra-specific variation in traits or response to water limitation along the elevation gradient (Table 2), in keeping with our alternative hypothesis. Wahlenbergia ceracea plants sourced from low elevation produced fewer leaves throughout the experiment (Fig. 1i), and in general, these leaves had higher levels of stomatal conductance (Fig. 2c), and correspondingly more negative δ¹³C (higher water use efficiency) (Fig. 2l). Low elevation sourced O. eriopoda plants also displayed more negative δ¹³C values (Fig. 2k).

We predicted that low elevation sourced plants would show greater plasticity than high elevation plants but found few significant interaction terms (Table 2). For A. glacialis, low elevation sourced plants consistently maintained homeostasis in total biomass across treatments, whereas high elevation plants showed a decline in biomass under water-limiting conditions, our proxy for fitness (see below) (Fig. 1a, g). For SLA, high elevation sourced plants maintained homeostasis between treatments (Fig. 1m). The same pattern was also observed for δ¹³C, whereby low elevation sourced plants displayed a more negative δ¹³C and a lower water use efficiency across treatments in comparison to high elevation plants (Fig. 2j). Finally, for O. eriopoda we found a significant treatment by elevation (TxE) interaction for SLA, whereby high elevation sourced plants increased SLA under low water conditions, while low elevation sourced plants decreased SLA under low water conditions (Fig. 1n).

Selection gradient analyses

While slightly more cases of selection upon plasticity than would be expected by chance alone, likely arising from false significant effects due to multiple testing, the 5/22 traits displaying significant results provide only very weak evidence that water response traits and plasticity therein are under selection in these species.

Why elevation doesn’t affect water response patterns?

The use of elevation gradients within ecological studies is increasingly common, particularly as a method of assessing responses to future climate change conditions. Our first elevation based hypothesis predicted that low elevation sourced plants would display greater variation in trait mean, and therefore increased plasticity as a result of occupying a more heterogeneous environment [12, 38]. For the high elevation populations it was predicted that little evidence of plasticity would be observed [27, 39], though see Frei et al. [40] and Pohlman et al. [41] for opposing perspectives. We found relatively few significant elevation responses, and, in many instances within our study low elevation sourced plants displayed a greater degree of trait homeostasis across our water treatment (Figs. 1 and 2). In contrast, previous studies have demonstrated differences in plastic responses along elevation gradients and associated these with differing selection pressures at each elevation, such as increased temperature [24, 25, 42], increased frost risk due to earlier melting snow [43–45], and competition from upward shift in low elevation alpine plants [46]. Likewise, past research has assessed elevation based changes in traits such as; growth rate, biomass [29], leaf size [20], stomatal conductance [30] and carbon isotope discrimination [47].

Generally within alpine regions, decreasing soil depth and vegetation cover with increasing elevation may mean that soil moisture deficits, particularly during summer, may be higher at high elevations [18, 48]. However, others suggest that because water availability within an alpine landscape may be more likely to vary over micro-scales than across an elevation gradient per se we considered an alternative hypothesis: that patterns of plasticity in response to water availability might be consistent across elevations, and in this differ from plasticity patterns in response to temperature. Thus, we would predict plenty of plasticity in response to water availability, but not differentiation in the pattern of that plasticity across elevations. Water availability within alpine ecosystems is not only a function of precipitation, or evapotranspiration, but may be determined by prevailing weather conditions, snow melt, localised topography, aspect, soil structure, and/or vegetation cover [18, 49]. In our tall alpine herb field, heavy vegetation cover may act to minimize evaporative moisture loss during the summer period such that variation with elevation is moderated [18]. Thus, it appears that these alpine plants have plasticity in response to water in expected directions, but that there has been no selective pressure for differentiation in that plasticity across elevation [23, 49].

Water availability is just one component of microhabitat heterogeneity that may influence the evolution of adaptive plasticity in physiological and morphological traits, consequently conflict in selective pressures are likely to arise. Shifts in abiotic factors may co-occur and act synergistically, or alternatively antagonistically, such that the fitness landscape may be highly complex [27]. Given the significant differentiation in distribution of adaptive plasticity in response to water between populations of different elevation found by Nicotra et al. [24] and Briceno et al. [25], we expected to detect adaptive plasticity in water response, and potentially elevation differences as well. The paucity of such differentiated responses and limited evidence of adaptive plasticity in response to water thus could reflect such conflicts [26, 50]. This finding highlights the fact that adaptive plasticity in response to a given trait may reflect a neutral or even maladaptive response for another, thereby limiting the development of a phenotypic optimum in response to any one environmental trait [26, 51–53]. Furthermore, given that phenotypic plasticity is predicted to evolve when environmental cues are reliable [54], it is important to consider both spatial and temporal scales of different environmental parameters [51]. Follow-up experiments could therefore investigate the simultaneous effects of water and temperature, perhaps incorporating elements of frequency and duration of deficit events to fully understand whether and how selective conflicts and trade-offs exist within this system.

We found little evidence that the observed plasticity in water-use traits was adaptive (Tables 3, 4), and this may reflect both evolutionary history and the different ways phenotypic plasticity may influence fitness. If selection for plasticity in water response has been strong and consistent across the landscape, inability to detect an adaptive signal may simple reflect loss of variation in that plasticity, not that the plastic response is not adaptive per se. Observations of positive correlations between fitness and plasticity may indicate an ability to capitalize upon favourable conditions, or alternatively, observations of high plasticity under unfavourable, or stressful conditions may result in the maintenance of fitness homeostasis [38, 55]. Furthermore, such responses are likely to be trait and species specific. It is also important to note that plasticity may not always be advantageous [13, 26]. There are costs and limits associated with plasticity that may limit not only its development, but also its extent and effectiveness. It is also the case that costs and adaptive value of plasticity are not always easy to assess under experimental settings, or on naturally occurring genotypes [15, 26, 51, 56–58]. Finally, plasticity may operate in a non-linear manner and it is possible that our study represents a smaller portion of a much larger and complex reaction norm [59]. Thus, further research to identify the genetic architecture of the observed plastic response to water availability might improve our understanding of its history and adaptive role.