In 2013, we obtained all seeds used in the present study from two open-pollinated populations on Baima Snow Mountain, Yunnan province. The pure diploid population was located on a site at a higher altitude (28°23’38″ N, 98°59’32″ E, 4160 m) than the pure hexaploid population (28°25’38″ N, 98°58’15″ E, 3560 m). In each population, mature fruits and a small amount of leaf tissue were collected from multiple maternal plants. All seeds and leaf tissues were brought to the laboratory at Kunming Institute of Botany within 36 h and kept at 4°C.

We used flow cytometry and root-tip squashes to examine the ploidy of maternal plants. Approximately 1 cm² fresh leaf tissue was chopped in 1.5 ml of pre-chilled WPB buffer (0.2 mol/L Tris⋅HCl, 4 mmol/L MgCl₂⋅6H₂O, 2 mmol/L EDTA Na₂⋅2H₂O, 86 mmol/L NaCl, 10 mmol/L sodium metabisulfite, 1% PVP-10, 1% (v/v) Triton X-100, pH 7.5). Fresh leaves from an Oryza sativa L. inbred line were chopped for use as an external standard (C-value = 0.86 pg/2C). After filtration, centrifugation, re-suspension and storage in the dark at 4°C for 10 min staining with 150 μl propidium iodide, the resulting cell suspensions were analyzed using a FACS-Vantage flow cytometer following the manufacturer’s recommendations (Partec, Germany; Tian et al., 2011). The histograms were analyzed with the FlowMax software package (Version 2.8.2, Partec GmbH, Germany). The FL2-area parameter (integrated fluorescence) was used to quantify DNA content. To examine the relationship between DNA content and ploidy, we carried out chromosome counts on root tips from seeds germinated in a Petri dish containing moist filter paper (Liu et al., 2006).

After identifying the ploidy of each maternal plant, seeds were germinated in plug trays filled with homogeneous humus soil, and placed in a canopied and naturally lit glasshouse at Yunnan Normal University. For each ploidy level, 180 robust seedlings of similar size were transplanted into pots (two seedlings per pot) 1 month later. All pots contained the same weight of a uniform mixture comprising equal volumes of peat and perlite. Soil surfaces were covered with a small quantity (c. 40 g) of perlite to minimize evaporation. Pots were randomly positioned on a single glasshouse bench and watered every other day to maintain saturation for 2 months. At the rosette stage, leaf tissue was sampled from each plant and re-screened using flow cytometry to exclude any seedlings of other ploidy levels.

After 2 months of growth, we began drought stress treatments. Mortality following transplantation reduced the sample sizes for diploids and hexaploids to 78 and 80 pots, respectively. Fifteen ‘empty’ pots were filled with the same amount of the same soil mixture to measure soil evaporation rates. The remaining pots of seedlings and the ‘empty’ pots were divided into low, moderate, and severe water stress treatments. Water stress was applied by watering to 80, 50, and 20% of maximum field capacity (FC). Soil water content was maintained at these levels by weighing the pots every 2 days, recording the amount of water loss and rewatering to the required water levels immediately. The experimental period commenced on July 9, 2014 (day t₁), and continued until August 10, 2014 (day t₂). During this period, no fertilizer was added to any pot. The sides of the glasshouse were always open for aeration throughout the experiment, so that the temperature inside the glasshouse was closely linked to the ambient outside temperature (Yang et al., 2014).

For growth and water use measurements, each pot was treated as a replicate, with its two seedlings being measured together. To estimate biomass increment during the experiment, 5 pots of each cytotype at the beginning of the experiment (t₁) and 15 pots (five pots per treatment) at the end of the experiment (t₂) were harvested. From each pot, the two seedlings were bulked together, and then divided into two parts: roots, and all aboveground parts including stems, leaves, and flowers. The total weight of each part was then determined after drying in an oven at 80°C for 48 h. The dry mass (dm) accumulated during the experimental period in the root [root dry mass (RDM)] and the aboveground parts dry mass (ADM) was calculated by subtracting dm per pot at day t₁ from that at day t₂, for each cytotype and treatment type, and dividing by two to convert from per pot to per plant. Total dry mass (TDM) was the sum of RDM and ADM, and the ratio of root to shoot (R:S ratio) was calculated by dividing RDM by ADM.

From both empty pots and those containing seedlings, water loss was measured as the difference between the weight of each pot just after watering and that just before the next watering event, 48 h later. These measurements were taken throughout the experimental period. Within each watering treatment, the amount of water transpired per pot per day for each cytotype was determined by deducting mean daily water loss per empty pot (evaporation) from mean daily water loss per pot with plants (evaporation plus transpiration). From this, total transpired water use (TWU) per plant was calculated as the total water transpired per pot between day t₁ and day t₂ divided by two.

Long-term water use efficiency (WUE_L) per plant, defined as the ratio between biomass production and water consumption for transpiration, was calculated, for each cytotype and treatment, as TDM/TWU (Ma et al., 2010).

Five to six pots for each treatment per ploidy level were randomly selected to measure gas exchange characteristics. For each pot, the fifth leaf down from the top, which was fully opened and matured, was selected on one plant, and the maximum photosynthetic rate (A_max), stomatal conductance (g_s), intercellular CO₂ concentration (C_i) and transpiration rate (E) were measured simultaneously using a LI-6400 XT infrared gas-analyser (LI-Cor Inc., Lincoln, NE, USA) for that plant. Measurements were taken between 9:30 and 12:00 h during sunny weather. Light levels were maintained at 1600 μmol m^-2 s^-1 (light-saturation points were derived from light response curves determined before the experiment) using artificial light provided by an LI-6400-02B LED light source (LI-COR Biosciences). The external CO₂ concentration was maintained at 400 μmol mol^-1 using portable CO₂/air mixture tanks whose output was controlled by a LI-6400-01 CO₂ injector (LI-COR Biosciences). Temperature and relative humidity were maintained at 24–26°C and 23–29%, respectively. Due to the lanceolate leaves of C. angustifolium usually couldn’t cover the leaf chamber (6 cm²), leaves were cut and scanned using a Canon Scan Lide 110 after the measurements had been taken. Then the leaf areas were analyzed by Scion Image (Version 4.0.3, National Institutes of Health, USA) so that leaf gas exchange parameters could be calculated on a per area basis. Instantaneous water use efficiency (WUE_i) was defined and calculated as A_max/E (Sapeta et al., 2013).

Chlorophyll fluorescence parameters were measured between 6:00 and 7:00 h on leaves that had been dark adapted for 10 h. These measurements were taken on the same day as the leaf gas exchange measurements. The maximum quantum yield of photosystem II [PSII; F_v/F_m = (F_m-F_o)/F_m] was measured using a LI-6400-40 leaf chamber fluorometer (LI-COR Biosciences). Five plants in each treatment per cytotype were analyzed (Mena-Petite et al., 2000).

To determine leaf water status, leaves were collected from other five different plants in one treatment. One fully mature leaf of each selected plant was stored in plastic bag on wet tissues until required for measurement of leaf area. The method used for analyzing leaf area was as described in the gas exchange section. The fresh mass (fm) of the measured leaves was then determined. Before measuring the saturated mass (sm) of the leaves, we allowed them to become turgid by resting them in water for 1 h. Leaves were oven-dried at 80°C for 48 hours before dm was determined. Relative water content (RWC) was expressed as 100% [(fm-dm)/(sm-dm)] (Yamasaki and Dillenburg, 1999). Leaf dry mass per unit area (LMA) was also calculated.

Leaves harvested from each of five different plants in one treatment were preserved in malondialdehyde (MDA). The 5 mm × 5 mm fragments from the adaxial side of the leaf and the midrib were used for embedding. The sample was incubated successively in the following solutions: 0.1 M PBS for 10 min (three times), 30% ethanol for 1 h, 50% ethanol for 1 h, 70% ethanol for 1 h, 80% ethanol for 1 h, 90% ethanol for 1 h, 95% ethanol for 1 h, 100% ethanol for 1 h, 100% ethanol:histoclear (1:1) for 4 h, histoclear for 1 h (two times), histoclear:paraffin (1:1) for 12 h (40°C), histoclear:paraffin (1:1) for 4 h (60°C), paraffin for 4 h (60°C), and paraffin for 12 h (60°C), and leaf tissues were then embedded in paraffin. Transverse cross-sections of the embedded samples were obtained with a microtome (Leica RM 2015) equipped with a freshly produced glass knife (Leica 819). The sections were then placed on glass slides.

For histochemical analyses, 12-μm-thick sections were stained with 1% Safranin for 12 h and 1% Fast Green for 10 s. In order to examine the lignified cell walls in midribs, 50-μm-thick sections were stained for 5 min with 1% phloroglucinol in 6 N HCl (Zhou et al., 2009). Stained cross sections were scanned using a brightfield microscopy (Leica DM 1000; Sun et al., 2013).

Leaf blade thickness, palisade parenchyma thickness, lacunar parenchyma thickness, leaf central vein diameter, xylem conduit diameter, leaf central vein total area, and xylem cell area were then measured from digital photographs with the Image J software (Version 1.45, National Institutes of Health, USA).

Data for all measured variables were analyzed using the general linear model (PROC GLM) to test the effects of cytotype, treatments, and their interactions. Significant differences among treatments for each cytotype were compared using one-way analysis of variance (ANOVA), and an independent-samples t test was used to compare the differences among cytotypes for each treatment. The homogeneity of variances was tested before analysis. Separate one-way ANOVAs were performed which assume independence between dependent variables and multiple traits measured for individual plants, but due to insufficient statistical power, correction of family-wise error rates for trait functional groups or individuals could not be performed. All statistical analyses were carried out using the SPSS statistical software package (Version 19.0, IBM, USA).

Flow cytometry analyses revealed two DNA ploidy levels in our samples: DNA-diploid (1.081 ± 0.009 pg/2C) and DNA-hexaploid (3.008 ± 0.084 pg/2C). The hexaploids thus had triple the DNA content of the diploids, a finding which was confirmed by chromosome counts.

Diploids and hexaploids of C. angustifolium exhibited different biomass allocation strategies in response to drought stress, and diploids generally accumulated more biomass during the experiment across all water gradients than hexaploids. The differences between cytotypes were significant for TDM accumulation and ADM accumulation under high water stress, and for RDM accumulation under low water stress (Table 1). Although TDM apparently declined in both cytotypes as the available soil water decreased, both ADM and RDM decreased in diploids, whereas hexaploids only showed a reduction in ADM. Due to the difference in dry weight allocation, water stress did not result in a significant difference for the R:S ratio in diploids, but it caused a significant increase in hexaploids (Figure 1).

Total TWU decreased significantly with decreasing soil water content in both cytotypes (Table 1). From low to severe water stress, diploids experienced a significant increase in WUE_L but hexaploids did not. Furthermore, diploids had higher WUE_L than hexaploids under severe water stress (Table 1). Transpiration rate (E) and WUE_i differed significantly among the three soil water contents (Tables 1 and 2). E decreased significantly with increasing water stress in both cytotypes, leading to a significant increase in WUE_i (Figure 2). Of the two ploidy levels, diploids had significantly lower E and higher WUE_i than hexaploids under low water stress.

Maximal photosynthetic rate (A_max), intercellular CO₂ concentration (C_i) and stomatal conductance (g_s) decreased significantly as water stress increased (Table 1), and there were significant differences between cytotypes in these three variables under specific drought stress treatments (Tables 1 and 2). Diploids had a higher value of A_max than hexaploids in all drought stress treatments and the difference was significant under 20% maximum FC (Table 1). The values of C_i and g_s were significantly higher in hexaploids than in diploids under low stress, whereas these differences were reduced and no significant difference could be detected under medium or high water stress (Table 1).

Maximum quantum yield of PS II (F_v/F_m) was similar between cytotypes for 80% FC and 50% FC, but it was significantly lower in hexaploids than in diploids under severe drought stress treatment (Table 1).

A significant decrease in leaf RWC was observed with increasing water stress for diploids, but there was no significant difference for hexaploids (Table 1). The value of RWC was significantly lower in diploids than in hexaploids under 20% FC. Leaf mass per unit area (LMA) varied significantly among water stress treatments and between ploidy levels. Drought stress gave rise to the increase in LMA from 50% FC to 20% FC for both cytotypes, but diploids showed significantly higher LMA than hexaploids under 80% FC and 20% FC (Figure 3).

Cross sections of leaves indicated that the leaf blade was significantly thicker in diploid than in hexaploid plants because of the thicker palisade parenchyma under extreme water stress (Table 3; Figure 4). It is also worth noting that diploids always had a higher percentage of palisade parenchyma than hexaploids (Table 3; Figure 4). Consistent with the higher LMA observed for diploids under all soil water conditions and the compact cell packaging expected in leaves with higher LMA values, the palisade parenchyma cells in diploids were closely arranged, but those in hexaploids were more sparse (Figure 4). Hexaploids had significantly wider xylem conduits for low and medium stress and the xylem cell area in the leaf central vein was larger for diploids than that for hexaploids under severe water stress (Table 3; Figure 5).

Water availability as a growth limiting factor was demonstrated in the present study, since it caused significant reductions in TDM, RDM, and ADM, but the decreases in TDM and ADM in hexaploids were larger than those in diploids (Table 1). In addition, drought effects were more pronounced for above- than below-ground biomass in hexaploids, leading to an increase in R:S ratio, as would be generally expected (Fernandez and Reynolds, 2000; Otieno et al., 2005). Curiously, there was no detectable change in R:S ratio for diploid cytotype (Figure 1), indicating the existence of an intrinsic trait for coping with drought stress in diploid C. angustifolium (Ma et al., 2010; Ma et al., 2014). Plant growth is generally inhibited by water deficit before photosynthesis and maintenance of respiration, leading to the increase of non-structural carbon concentrations in tissues (Muller et al., 2011), which is commonly interpreted as osmotic adjustment (Hasibeder et al., 2015). During desiccation, parallel degree of osmotic adjustments-related assimilates allocation in roots and leaves may lead to the stable R:S ratio for perennial grasses to maintain water balance between organs, such as Helianthus annuus (Sobrado and Turner, 1986). This could explain the constant R:S ratio of diploids C. angustifolium across different gradients of water stress.

Drought can affect plant growth by influencing leaf gas exchange rates (Sapeta et al., 2013). Early responses to water deficit involve stomatal closure and a subsequent reduction in stomatal conductance (Chaves et al., 2002). The resulting reduction in leaf diffusive capacity then causes a simultaneous decline in CO₂ uptake and transpiration during desiccation (Cornic and Massacci, 1996; Tezara et al., 1999). The magnitudes by which g_s was reduced (85.20% vs. 66.23%), C_i (25.24% vs. 15.88%), and E (74.86% vs. 56.06%) were greater in the hexaploids than in the diploids (Table 1). Lower C_i mediated by a reduction in leaf conductance may thus be inhibiting carbon metabolism in the face of drought stress (Downton et al., 1988; Maroco et al., 2002; Flexas et al., 2004) in both cytotypes (Table 1). Comparatively, diploids had a significantly higher A_max than hexaploids under severe water stress, and the reduction in A_max caused by drought in diploids (39.30%) was lower than that in hexaploids (53.94%; Table 1). Similarly, a higher photosynthetic rate during desiccation was found in diploids of Mercurialis annua (Buggs and Pannell, 2007) due to the case that polyploidy cytotype had lower transpiration rates and CO₂ exchange rates under drought stress. Taking these results together, the fact that gas exchange was less depressed in diploids than in hexaploids would suggest that the former is better able to resist drought. Another reliable diagnostic indicator of plant photosynthetic activity is F_v/F_m and its value often experiences a reduction under environmental stress (Maxwell and Johnson, 2000). Our results revealed that F_v/F_m was significantly decreased by drought stress in hexaploids, to below the optimal value of 0.8 (Table 1) (Mena-Petite et al., 2000), indicating the sensitivity of hexaploids to drought stress.

WUE_i and WUE_L were enhanced from 50% FC to 80% FC by increasing drought stress for both cytotypes (Table 1; Figure 2); similar results have been reported in other species (Zhang and Marshall, 1994). Plants with high water use efficiency should have greater abilities to survive drought stress than those with lower water use efficiency (Lauteri et al., 1997). Our results indicated that WUE in diploids increased relative to that in hexaploids under severe water stress, reflecting a water use strategy adapted to growing in environments where drought stress was frequent.

The leaf is a major bottle-neck in the whole plant hydraulic conductivity system (Sack et al., 2003). Anatomical analysis revealed that diploid C. angustifolium possessed thicker leaves and palisade tissue (Table 3; Figure 4), and such morphological adjustments may enable the diploid to be more drought resistant than hexaploid (Li et al., 2009). Species that occur in dry areas are able to maintain living tissue at low RWC (Baltzer et al., 2008) and usually have thicker leaves (Rhizopoulou and Psaras, 2003). Accordingly, LMA tends to increase with leaf density and a high LMA is generally being considered as an adaptation to drought (Niinemets, 2001). Comparisons of LMA and RWC between ploidy levels showed that diploids had significantly higher LMA and lower RWC than hexaploids under extreme drought stress (Table 1; Figure 3), implying that the dense leaf tissues of diploids can withstand a low water content and thus slow water loss from the whole plant (Cunningham et al., 1999; Bucci et al., 2004).

For hydraulic transport to be efficient, there should be a compromise between the ability to cope with water stress and the ability to grow at high rates under more favorable water conditions (Piñol and Sala, 2000; Martínez et al., 2002). As an instance, the lower xylem hydraulic conductivity found in higher ploidy levels of Atriplex canescens may be a major constraint counteracting the beneficial effects of their better drought tolerance (Hao et al., 2013). In contrast, the higher hydraulic conductivity of drought-sensitive diploid A. canescens may endow it with a higher growth rate (Stutz and Sanderson, 1983), and thus make it more competitive in environments with relatively high water availability (Sperry and Hacke, 2002). In C. angustifolium, higher hydraulic conductivity was detected in tetraploids than in diploids (Maherali et al., 2009). According to the above mentioned compromise rule, tetraploids may be much more drought-susceptible but competitive compared with diploids. Surprisingly, vulnerability to water stress induced cavitation was not found to differ across cytotypes (Maherali et al., 2009). The biomass of tetraploids was more negatively impacted by the imposition of water limitation than that of diploids, and both cytotypes had equal competitive abilities when water was limited (Thompson et al., 2015). Our experiments showed that both diploids and hexaploids of C. angustifolium were apparently affected by water shortage (as indicated by reductions in biomass accumulation and photosynthetic rate). Nevertheless, we found partial support for the hypothesis that diploids might be better able to cope with drought conditions than the polyploid cytotypes, since they showed physiological (e.g., gas exchange rate) and morphological (e.g., leaf architecture) adjustments appropriate for enduring water loss.

Both water deficit and freeze–thaw cycles can lead to xylem cavitation (Pockman and Sperry, 1996), and thus cytotypes with higher ploidy levels and larger conduits may be more vulnerable to drought stress due to the positive relationship between xylem conduit size and the risk of cavitation (Sperry and Sullivan, 1992; Hacke and Sperry, 2001; Martínez and Pockman, 2002). In C. angustifolium, polyploids generally had wider hydraulic vessel diameters than diploids (Table 3; Figure 5). Consequently, diploids, with their smaller xylem conduits, should be better able to avoid hydraulic disruption formed upon embolism (Pockman and Sperry, 1996), and a greater susceptibility to freezing-induced cavitation may exclude polyploids from sites at high altitudes (Maherali et al., 2009). In addition to the smaller xylem conduits, the vessel area of the leaf central vein in diploids was larger than that in hexaploids (Table 3; Figure 5), and the dense vascular bundle found in diploids was consistent with those observed in a few other cases (Maherali et al., 2009; Allario et al., 2011). A high density of major veins such as that found in diploids can provide a large number of parallel xylem pathways for water transport per leaf area, contributing to drought tolerance by routing water around embolized conduits (Scoffoni et al., 2011; Sack et al., 2012).

Two scenarios have been proposed to explain differences in patterns of cytotype distribution (Manzaneda et al., 2012). The adaptive evolutionary scenario suggests that shifts in ploidy level could result in differential stress tolerances (Levin, 2002; Baack, 2004; Kubátová et al., 2008; Sonnleitner et al., 2010; McAllister et al., 2015), which may underlie the ecological divergence and adaptation of cytotypes to novel environments (Ramsey, 2011). In contrast, the environmentally independent explanations (‘non-adaptive scenarios’) posit that exclusion of minority cytotypes (Levin, 1975) and historical processes (Sonnleitner et al., 2010; McAllister et al., 2015) may be the driving force behind the observed distribution patterns. Diploid C. angustifolium inhabits higher altitudes than polyploids, and a similar pattern has also been observed in Senecio carniolicus (Sonnleitner et al., 2010; Huelber et al., 2015) and Centaurea jacea (Hardy et al., 2000). The results of our study illustrate the distinct natures of physiological tolerance in the different ploidy levels, with diploids being less sensitive to drought stress than hexaploids. Thus, the occurrence of diploids in more open habitats at higher altitudes may be the result of adaptation that provides greater resistance to abiotic stress (Körner, 2003; Sonnleitner et al., 2010), whereas the predominance of polyploids in dense and nutrient-rich vegetation at lower altitudes may be due to greater competitiveness compared with the surrounding vegetation (Schönswetter et al., 2007; Ståhlberg, 2009). An earlier study indicated that differences in physiological tolerances in C. angustifolium probably evolved through natural selection acting on plant water relations after polyploidization (Maherali et al., 2009) and thus induced the adaptation of cytotypes to their native habitats (Martin and Husband, 2013). We therefore consider eco-physiological differentiation to be an important adaptive factor underlying the origin of the geographical separation and divergence in climatic niche (Thompson et al., 2014), although other, environmentally independent, factors cannot be excluded.