Tree propagation and pre-treatment culture
For Experiment 1, root cuttings were collected from various aspen genotypes in two populations, the Intermountain West (collected in Utah) and the Great Lakes (collected in Wisconsin). The cytotype of each genotype was determined via flow cytometry as described in Mock et al. (2012) and triploid cytotypes were confirmed using mitotic chromosome counts. We used micropropagation techniques (similar to Donaldson and Lindroth (2004)) to generate ramets of aspen genotypes for each experiment. We rooted micro-cuttings of individual ramets in flats of potting soil (Metro-Mix 350, Sun Gro Horticulture, Agawam, MA, US) and then transferred them to 66-mL cone-shaped cells (RLC4 Cone-tainer cells, Stuewe & Sons Inc., Tangent OR, USA). Ramets were grown under these conditions for six weeks during the spring of 2015. In early June of 2015, we out-planted 30 ramets from each of 32 aspen genotypes at the West Madison Agricultural Research Station. We planted eight diploid and eight triploid genotypes from each population. We placed 64 ramets, two per genotype, at randomly chosen locations within each of 15 mesocosms. Mesocosms were 80 cm in diameter, had a raised soil bed (15 cm elevation) surrounded by plastic garden edging, and were covered with a sheet of white plastic (0.025 cm thickness) in order to retain moisture. Plots were watered by hand and by seasonal precipitation.
For Experiments 2 and 3, we used the same Wisconsin root material collected for Experiment 1. In October 2017, root cuttings from all genotypes were planted in 4-l plastic flats filled with equal parts of torpedo sand and silt-loam field topsoil (Keleney Topsoil, Madison, Wisconsin, USA). Once root cuttings produced new shoots, we cut them into sections of ca. 10 cm length. Shoots were individually grown in 0.5-l pots (60% torpedo sand, 40% field topsoil) and received 2 g of 6-month-release fertilizer (Nutricote total, Type 100 Blend, NPK 13-13-13, Arystra Lifescience, Broadway, NY, USA). By May 2018, shoots had developed into ~40 cm tall trees. These were planted into outdoor mesocosms (1.5 × 1.5 m) in a 20 cm deep layer of 60% torpedo sand and 40% field topsoil. One tree of each genotype was planted in each mesocosm (i.e., eight different diploids and eight different triploids per mesocosm). A randomly selected triploid tree was always planted next to a randomly selected diploid tree. To minimize potential edge effects, a border of single “buffer trees” (P. tremuloides) was planted around the experimental trees of each mesocosm. All trees were planted at a 20 x 20 cm spacing. One week after planting, each tree received 3.5 g of 6-month-release fertilizer. Trees were watered frequently and allowed to grow for an additional eight weeks before applying the experimental treatments.
 
Experimental treatments
All trees were afforded access to similar resource levels, and remained undisturbed, for the entire duration of Experiment 1.  In Experiment 2, eggs of white-marked tussock moths (Orgyia leucostigma) were obtained from the Canadian Forest Service (Sault Ste. Marie, Ontario, Canada). Eggs and hatched caterpillars were kept in climate chambers (24 °C, 50–70% humidity, 16:8 L:D light cycle) and were reared on wheat germ based Bell diet (Bell et al. 1981) to the fourth larval stadium. On July 17-18, 2018 we placed four fourth-stadium caterpillars onto each of the experimental trees in each of ten mesocosms. Trees in the remaining 12 mesocosms received no larvae. Caterpillars were confined to trees in nylon mesh bags and allowed to feed on the entire tree for two days. Larvae were then removed and the defoliation was manually completed to a consistent 50% level across all leaves on all trees, using pinking shears.  This method allowed us to incorporate the cues derived from natural herbivory with a standardized amount of foliar tissue removal (Stevens et al. 2007). Control trees (no defoliation) were covered with empty mesh bags for two days.
Several strategies were implemented to control the soil water content of the mesocosms in Experiment 2. To intercept rainfall, we installed transparent, plastic covers over all 22 mesocosms. To prevent infiltration of water from surrounding soil, the soil layer in each mesocosm was elevated on top of a 7.5-cm high layer of perforated plastic flats. A layer of root barrier fabric was also placed between the soil layer and the supporting plastic flats. Finally, the soil surface in each mesocosm was covered with a 6-cm layer of leaf mulch to reduce evaporative drying. The drought stress treatment was applied to five mesocosms of defoliated trees and six mesocosms of undefoliated trees. Water was withheld for 7-12 days until most trees began to wilt, whereupon the mesocosm was watered to soil saturation. This procedure was repeated for 37 days, until the end of the study. Control mesocosms were watered to saturation every 2-4 days.
In Experiment 3, trees were defoliated (beginning on July 25, 2018, in five of the ten additional mesocosms) and soil water content was controlled using the same approaches outlined above for Experiment 2, except that, beginning July 27, 2018, all mesocosms were subjected to an extended drought stress treatment. Namely, water was withheld until all leaves from at least 12 of the 16 trees in a mesocosm had turned completely brown and the shoot tip meristems had died, upon which the mesocosm was watered to soil saturation. Consequently, mesocosms containing nondefoliated trees were re-watered 12-17 days after drought stress initiation, whereas mesocosms containing defoliated trees were re-watered 17-21 days after drought stress initiation. Starting with the initial re-watering date, mesocosms were watered frequently for an additional 42 days, at which time all trees were harvested (see below). Because of flooding by an extreme rain event in the middle of the experiment, two mesocosms of the defoliation treatment were eliminated.
Measurements of tree traits
Tree growth
To account for the typically substantial variation in initial tree size, stem height (h, from soil surface to apical bud/meristem) and basal diameter (d) of every tree were measured at the beginning of each experiment. These data were used to calculate initial d2h, which is a metric highly and linearly correlated with tree biomass in young aspen (Stevens et al. 2007). Observed genotype-level relationships between initial d2h and tree weight afforded estimates of the latter for all non-harvested trees at the outset of growth measurement in Experiment 1. This size proxy also served as an effective covariate in analyses of variation in aboveground tree weight, which was quantified at the end of Experiments 2 and 3 using a destructive harvest. At harvest, all tissues were separated, roots (only harvested in Experiment 1) were rinsed free of soil, and fresh leaf area (without petioles) was measured using an LI-3100 (Li-Cor Biosciences, Lincoln, NE, USA).
In Experiment 1, initial h and d (2 cm above soil surface) were measured on July 23, and 10 trees per genotype (across six mesocosms) were harvested (including roots) on September 15. Specific leaf area (SLA, total fresh leaf area/ total leaf dry weight), leaf weight ratio (LWR, total leaf dry weight/ total tree dry weight [including roots]) and relative growth (RG, the ln of final tree dry weight [including roots] – the ln of initial tree dry weight [estimated from initial d2h]). In Experiment 2, d (4.5 cm above soil surface) and h were measured just prior to defoliation and the onset of drought treatment (July 17). Stem height was also measured on each tree just prior to aboveground destructive harvest on August 27. SLA, total leaf dry weight and final tree dry weight (without roots) were measured. In Experiment 3, d (4.5 cm above soil surface) and h were measured just before the onset of defoliation and drought treatments. On September 18, d and h were again measured for all trees. All trees were harvested 42 days after initial re-watering. In this case, intact stem sections, newly grown tissue (new side shoots, petioles and leaves that developed after the stress treatment application), and old leaves that developed before the stress treatments were separated and dried. Before drying, the surface area of a randomly selected subset of 15-21 newly grown leaves (without petioles) per tree was measured again using the LI-3100 and the SLA was calculated. All leaves were then dried and total leaf weight and tree weight (without roots) were calculated.
 
Phytochemistry
To compare cytotype and genotypic variation in phytochemistry and its response to environmental stress, leaves were harvested in roughly the middle of Experiments 1 and 2. In Experiment 1, we collected all foliage from the crowns of 12 trees per genotype (across five mesocosms) on July 23, 2015. In Experiment 2, leaves were collected evenly across the different treatments on August 17-24, 2018. From each tree, three fully expanded, young leaves were collected from the upper third of the tree crown. All leaves collected from defoliated trees showed signs of damage. In both experiments, leaves were collected under moist soil conditions and were vacuum-dried to constant mass and then stored at -20°C to preserve labile phytochemical constituents. Concentrations of condensed tannins were analyzed using the acid butanol method of Porter et al. (1985) standardized against purified condensed tannins from P. tremuloides. Levels of the salicinoid phenolic glycosides tremulacin, tremuloidin, salicin and salicortin were quantified using ultra-high-performance liquid chromatography-mass spectrometry as described by Rubert‐Nason et al. (2017). Leaf nitrogen concentration was measured using near-infrared reflectance spectroscopy (Rubert-Nason et al. 2013) verified against a subset of the samples analyzed using combustion gas chromatography (Thermo Flash EA1112 elemental analyzer [Thermo Finnigan, Milan, Italy]).
 
Leaf physiology
In Experiment 1, we measured light-saturated rates of net photosynthesis (Aarea, µmol m-2 s-1) and stomatal conductance (gs, mol m-2 s-1) on fully expanded, sunlit leaves from four ramets per genotype, using an LI-6400XT portable photosynthesis system with a broadleaf cuvette (Li-Cor Biosciences, Lincoln, NE, USA), during several mornings under clear to partly cloudy skies in late July, 2015. Cuvette temperature was set at 25°C, photosynthetic photon flux density (PPFD) was 2000 mmol m-2 s-1, and the reference pCO2 was 40 Pa. Relative humidity in the sample cuvette was between 60-70%.
In Experiment 2, pre-dawn leaf water potential (ΨPD), light-saturated Aarea, gs, and the maximum and operating efficiencies of photosystem II were measured, using the LI-6400XT portable photosynthesis system with a pulse amplitude modulation (PAM) fluorometer attached to the broadleaf cuvette, between July 21 and August 26, 2018. All traits were measured during 12 sessions. Each session consisted of two steps that were executed on the same day. First, ΨPD was measured from each of three trees growing in one mesocosm of each treatment combination. The selected trees were evenly distributed within a mesocosm and ΨPD from at least one diploid and one triploid tree was measured per mesocosm. ΨPD was measured between 0400-0530 h using a pressure chamber (PMS Instrument Company, Albany OR, USA). In one additional session, we quantified ΨPD and leaf physiological traits from only two mesocosms, of which one was exposed to drought stress and defoliation and the other was exposed to drought stress only. During the experiment, every mesocosm was selected two or three times for measurements. Mesocosms subjected to drought stress were measured at different stages of soil dry down. In a second step, leaf gas-exchange and variable chlorophyll fluorescence were measured on fully expanded, young leaves from all trees growing in the previously selected mesocosms (i.e., four mesocosms, 16 trees per mesocosm). During measurements, leaves in the cuvette were exposed to a photosynthetic photon flux of 1800 µmol m-2 s-1. The reference CO2 partial pressure was 40 Pa, and the cuvette temperature was set at 25 °C. We did not control vapor pressure deficit between leaf and cuvette air, which ranged from 1.4–1.9 kPa. Measurements were taken on sunlit foliage between 0900-1300h. From August 18 through August 25, gas exchange was also measured at a cuvette CO2 partial pressure (pCO2) ranging from 7.5 to 25 Pa on sunlit foliage from all trees of three mesocosms that were neither drought stressed nor defoliated. The observed relationship between photosynthesis(Aarea) and intercellular pCO2 (Ci) was used to estimate Vcmax (µmol CO2 m-2 s-1), the maximum velocity of RuBP carboxylation, employing a trend-fitting method that minimized the total sums of squares for differences between observed versus predicted A (Long and Bernacchi 2003). Estimates relied on Michaelis–Menten constants for CO2 (Kc) and oxygen (Ko), as well as photosynthetic compensation pCO2 (Γ*), derived using formulae from Long and Bernacchi (2003). Because we did not account for the influence of mesophyll conductance on CO2 diffusion into the chloroplast (Dillaway and Kruger 2010), our derived Vcmax values are “apparent” (Bernacchi et al. 2013) and based on Ci rather than chloroplastic pCO2 (Cc).
 
Statistical analyses
For Experiment 1, Linear Mixed Models (LMMs) were used to evaluate the impacts of the explanatory variables ploidy, geographic origin (population), and their interaction on the various growth, allocational, morphological and leaf physiological traits. We accounted for genotypic variation by treating the variable genotype as random intercepts. In addition, block was included as a random intercept in each LMM. Individual RG values were normalized for initial tree weight (estimated from initial d2h values). The normalized values were then used as model response variables. Individual LWR values were normalized for tree dry weight. Prior to any normalization, the relationship between the response and the normalizing variable was explored and, if necessary, ln- or square root transformed to meet the model assumptions of normality and homoscedasticity.
 
For Experiment 2, LMMs were used to explore the roles of the explanatory variables ploidy, drought, defoliation and their interactions on the various growth, allocational, morphological phytochemical and leaf physiological traits. To account for the effect of genotypic variation, genotype and all possible genotype × stress interactions were modeled as random intercepts. In addition, we also included block as a random intercept in our models. Individual values of final above ground tree dry weights were normalized for variation in individual initial d2values. The normalized values were then used as model response variables. To assess aboveground biomass allocation to foliage, total leaf weights were normalized for final tree dry weights. To quantify how much stem mass a tree allocated to height growth (i.e., how "lanky" a tree was) final height values were normalized for variation in final stem dry weight.Response and the normalizing variables were ln- or square root transformed if necessary to meet the model assumptions of normality and homoscedasticity.
The impacts of ploidy, genotype, and defoliation on leaf physiology in Experiment 2 were explored with LMMs as described above. Drought-stress treatments resulted in various soil-dry down stages at different points in time which could not be reliably modeled with our LMMs models. Therefore, we focused on only well-watered trees in our analyses. Because physiological trait measurements were taken multiple times from each tree, average values across all measurements per tree were calculated and used for statistical analyses. For drought-stressed trees, we examined the relationship between Aarea and ΨPD for each genotype in both defoliation treatments using a three-parameter, Weibull-type vulnerability curve with the equation: 
\(fi(y)=A_maxe^(-(x/b)^c)\)