Comparisons of biomass, water use efficiency and water use strategies across five genomic groups of Populus and its hybrids

Genetic improvement and hybridization in the Populus genus have led to the development of genotypes exhibiting fast growth, high rooting ability and disease resistance. However, while large biomass production is important for bioenergy crops, efficient use of resources including water is also important in sites lacking irrigation and for maintaining ecosystem water availability. In addition, comparison of water use strategies across a range of growth rates and genetic variability can elucidate whether certain strategies are shared among the fastest growing and/or most water use efficient genotypes. We estimated tree water use throughout the second growing season via sapflow sensors of 48 genotypes from five Populus taxa; P. deltoides W. Bartram ex Marshall × P. deltoides (D × D), P. deltoides × P. maximowiczii A. Henry (D × M), P. deltoides × P. nigra L. (D × N), P. deltoides × P. trichocarpa Torr. & Gray (D × T) and P. trichocarpa × P. deltoides (T × D) and calculated average canopy stomatal conductance (GS). We regressed GS and atmospheric vapor pressure deficit (VPD) wherein the slope of the relationship represents stomatal sensitivity to VPD. At the end of the second growing season, trees were harvested, and their dry woody biomass was used to calculate whole tree water use efficiency (WUET). We found that D × D and D × M genotypes exhibited differing water use strategies with D × D genotypes exhibiting high stomatal sensitivity while retaining leaves while D × M genotypes lost leaf area throughout the growing season but exhibited low stomatal sensitivity. Across measured taxa, biomass growth was positively correlated with WUET, and genotypes representing each measured taxa except D × N and T × D had high 2‐year dry biomass of above 6 kg/tree. Overall, these data can be used to select Populus genotypes that combine high biomass growth with stomatal sensitivity and WUET to limit the negative impacts of bioenergy plantations on ecosystem water resources.


| INTRODUCTION
Woody biomass has the potential to supply increasing amounts of renewable energy in the U.S. and abroad (U.S. Department of Energy, 2016) and interest from the various transportation sectors for renewable liquid fuels should increase investment in woody bioenergy fuels (Gegg et al., 2014) in the future. Populus species and their hybrids have the potential to supply the increasing demand for renewable energy sources due to their historical use (and associated knowledge base and infrastructure) for pulpwood and low-cost hardwood timber (Ridge et al., 1986;Saxena et al., 2009;Spinelli & Hartsough, 2006;Stanton et al., 2002), ability to hybridize readily (Knox et al., 1972), adventitious rooting which allows for inexpensive stand establishment from unrooted cuttings and their demonstrated high yields (Stanton et al., 2002). Eastern cottonwood (Populus deltoides W. Bartram ex Marshall) is native to the southeastern U.S. with a distribution that extends from southern Canada to northeastern Mexico and is one of fastest growing species in the Populus genus (Isebrands et al., 2014). Genetic improvement programs have developed eastern cottonwood genotypes that exhibit high rootability, disease resistance and growth (Zalesny et al., 2016), and it tends to be one of the parent species in top performing hybrid poplar taxa. As such, greater potential gains for the southeastern U.S. might be possible through interspecific hybridization of selected eastern cottonwood with non-native Populus species including Japanese poplar (Populus maximowiczii A. Henry), black poplar (Populus nigra L.), and black cottonwood (Populus trichocarpa Torr. & Gray) particularly in drier upland areas where eastern cottonwood is less adapted. In most cases, these are F 1 hybrids as the resulting genotypes will exhibit heterosis, whereas the F 2 generation shows greater variability of performance. Studies differ on which hybrid taxa perform best likely due to differing site conditions (Guo & Zhang, 2010;Heilman & Stettler, 1985;Trnka et al., 2008), but hybrid genotypes tend to perform better than either of their parental types (Coyle et al., 2006;Rousseau et al., 2013) in field trials on upland sites where Septoria stem canker (Sphaerulina musiva Peck formerly Septoria musiva Peck) is absent. However, eastern cottonwood can outperform hybrid taxa on alluvial sites and/or locations where Septoria stem canker is present (Rousseau et al., 2018).
Differing water use strategies influence characteristics of Populus genotypes including total water use, water use efficiency and response to drought, and in turn, their impacts on the water availability of an ecosystem (Bloemen et al., 2017;Maier et al., 2019;Sevigne et al., 2011). In general, water use strategies across species or individuals can be compared along an isohydricity spectrum. At one end of the spectrum, strictly isohydric species maintain constant midday leaf water potentials under both well-watered or drought conditions through tight control of stomatal conductance (i.e., high stomatal sensitivity to environmental conditions). At the other end of the spectrum, strictly anisohydric species exhibit lower stomatal sensitivity to environmental conditions causing their midday leaf water potentials to vary in response to the changing environment (Attia et al., 2015;Klein, 2014) and potentially risking hydraulic dysfunction (Tardieu & Simonneau, 1998). Each of these strategies possess pros and cons and can, therefore, be desirable based on environmental conditions and objectives. For example, genotypes with high stomatal sensitivity may perform better on sites that are more drought prone while greater water use from genotypes with lower stomatal sensitivity may be useful for phytoremediation purposes in terms of removing water-soluble chemicals that could degrade surrounding water quality (O'Neill & Gordon, 1994). Particularly for woody bioenergy crops whose production should not compete with food crops and irrigation may not be financially feasible, future poplar plantations may occur on less fertile, more water-limited upland sites (U.S. Department of Energy, 2016) necessitating the selection of genotypes that exhibit conservative, drought-tolerant strategies (Monclus et al., 2006).
In addition to considering water use strategies and drought tolerance, water use efficiency or the amount of water used per unit biomass can vary widely across genotypes (Attia et al., 2015) and has the potential to greatly impact surrounding hydrology including groundwater recharge and streamflow (Petzold et al., 2011). For example, the large-scale production of genotypes which are efficient in their water use will likely place lower demands on local water sources and/or require lesser amounts of irrigation from outside sources than less efficient individuals (Petzold et al., 2011) while producing the same amount of harvestable biomass. Implementation of water use efficient genotypes, in turn, could reduce losses in local water budgets and potentially expand the profitable range where poplar can be grown for biofuel feedstocks into areas where lower annual rainfall levels are common. On the other hand, Populus genotypes that exhibit greater water use per unit aboveground biomass may rely on more extensive root systems and thereby achieve greater belowground carbon sequestration (Silim et al., 2009) in areas where ecosystem water budgets are not an issue. In addition, it is unclear if water use efficiency is correlated with overall aboveground productivity with some studies finding a positive correlation (King et al., 2013;Monclus et al., 2009;Renninger et al., 2021;Zhang et al., 2004) and other studies finding no correlation between water use efficiency and biomass production (Bonhomme et al., 2008;Dillen et al., 2011;Marron et al., 2005;Monclus et al., 2005Monclus et al., , 2006Monclus et al., , 2009Rae et al., 2004).
In this study, we sought to compare water use strategies across 48 different genotypes of Populus from eastern cottonwood (P. deltoides × P. deltoides; D × D) and hybrid poplar (P. deltoides × P. maximowiczii D × M; P. deltoides × P. nigra D × N; P. deltoides × P. trichocarpa D × T; P. trichocarpa × P. deltoides T × D) taxa. Specifically, we wanted to determine (1) if water use patterns in terms of canopy conductance differed among taxa across soil moisture and atmospheric vapor pressure deficit gradients throughout the growing season, (2) if water use strategies in terms of stomatal sensitivity versus leaf area loss differed among taxa, (3) if parameters indicative of water use strategies were correlated with aboveground biomass production across taxa, and (4) which water use parameters were correlated with overall tree-level water use efficiency. We hypothesized that genotypes will vary in water use strategies based on their taxa and that genotypes within taxa can be identified that exhibit both high aboveground biomass as well as water use strategies that are indicative of water conservation and potential drought tolerance. On the other hand, genotypes that exhibit high water use can be identified and targeted for phytoremediation applications. Finally, identifying water use strategies and parameters for individual Populus genotypes can provide information for future breeding focused on the development of new genotypes suitable for biomass production in the southeastern U.S. The potential for defining and expanding the physiological strategies of Populus genotypes may provide the capacity to expand production into drier and more marginal areas as well as enable the ability of future Populus production to be adapted to changing climate conditions.

| Genotypes, study area and environmental measurements
The overall study consisted of 100 genotypes arranged in two-tree row plots (i.e., two ramets per genotype per block) in 12 replicated blocks with each block containing four sub-plots of specific taxa to provide similar, within taxa, competition in terms of rooting and crown architecture among genotypes. Genotypes of improved eastern cottonwood (P. deltoides × P. deltoides; D × D; 25 genotypes) and hybrid poplar taxa including P. deltoides × P. maximowiczii (D × M; 46 genotypes), P. deltoides × P. nigra (D × N; 11 genotypes), P. deltoides × P. trichocarpa (D × T; 13 genotypes) and P. trichocarpa × P. deltoides (T × D; 4 genotypes) were selected for planting. Populus maximowiczii has a native range from north-central China to east Siberia and northern Japan and is a rapidly growing, long-lived Populus species (Isebrands et al., 2014). Populus nigra is native to Europe, northern Africa, and western Asia and can grow well in both riparian and well-drained sites as well as locations with dry summers but wetter spring and fall periods (Isebrands et al., 2014). Populus trichocarpa is native to the western U.S. and Canada is the largest Populus species in North America growing best in humid, coastal sites with limited drought conditions (Isebrands et al., 2014).
The study site was located on the Gulf Coastal Plain region in northeastern Mississippi (33°51′16″N 88°17′27″W). The climate of this area is humid subtropical, characterized by mild winters and long, hot summers. Mean annual temperature for the site was 18.7°C with a mean maximum temperature of 25.1°C and a mean minimum temperature of 12.4°C (National Oceanic and Atmospheric Administration & National Centers for Environmental Information, 2021). Rainfall is fairly evenly distributed throughout the year, with the region, on average, receiving about 144 cm of rainfall per year (National Oceanic and Atmospheric Administration & National Centers for Environmental Information, 2021). The soil of the site is characterized as a Prentiss silty loam soil, which is moderately well-drained with a seasonal water table perched at 0.61-0.76 m (Natural Resources Conservation Service, 2021).
Prior to Populus planting in April, 2018, the site underwent limited management but was in peanut production. Before planting, the site was tilled and subsoiled to a depth of 45.7 cm along planted rows that were 2.74 m apart. Unrooted cuttings approximately 40.6 cm long for eastern cottonwoods and 22.9 cm long for hybrid poplars were planted in subsoiled rows approximately 1.83 m apart within rows. Dormant, unrooted hybrid poplar cuttings were provided by GreenWood Resources and were placed in a walk-in freezer upon receipt. Eastern cottonwood cuttings were collected from the Mississippi State University Department of Forestry Cottonwood Cutting Orchard, soaked in water and insecticide (Admire Pro; Bayer Corp.) and placed in plastic bags in a walk-in freezer until planting. Hybrid poplar cuttings were also soaked in Admire Pro insecticide approximately 3 days before planting. The entire test site was sprayed immediately after planting and prior to bud break with a pre-emergent Goal 2XL (Dow AgroSciences LLC) and Pendulum 3.3EC (BASF). Following the breakdown of the Goal and Pendulum, additional post emergent herbicide of Select (Valent USA LLC.) was applied to control Johnson grass (Sorghum halepense [L.] Pers.). Combined mechanical weed control (tilling and hand weeding) was performed throughout the first growing season to control competition and provide "free to grow" conditions to the Populus cuttings. No competition control was needed during the second growing season. The test was periodically assessed for cottonwood leaf beetle (Chrysomela scripta Fabricius) feeding and received applications of Admire Pro (either ground applications or aerially depending on tree height) in August 2018, July 2019, and September 2019.
Sensors measuring photosynthetic photon flux density (PPFD; LI190SB Quantum sensor; LI-COR Biosciences Inc.), air temperature and relative humidity (HMP60 temperature and relative humidity probe; Vaisala), and rainfall (TE525MM tipping bucket rain gauge; Texas Electronics) were located next to the Populus plantation in an open field. Eight soil moisture (30 cm long CS616 time domain reflectometry probes; Campbell Scientific Inc.) sensors were installed vertically throughout the study site and located between subsoiled rows. Environmental sensors were installed shortly after planting in May 2018 and connected to a Campbell Scientific CR1000 datalogger (and AM16/32B multiplexer [Campbell Scientific Inc.] for soil moisture sensors). Sensor data were collected every 30 s and half hourly averages were recorded by the datalogger. Air temperature and relative humidity were used to calculate atmospheric vapor pressure deficit (VPD).

| Sapflow and canopy conductance (G S )
In the beginning of the second growing season (May 2019), laboratory made, thermal dissipation sap flow sensors (Granier, 1987) were inserted radially into the sapwood on the north side of 48 genotypes in 2 of the 12 replicate blocks for a total of 96 trees measured (two per genotype). Of the 48 genotypes measured, 12 were eastern cottonwood (D × D), 22 were D × M genotypes, 5 were D × N genotypes, 7 were D × T genotypes and 2 were T × D genotypes. Information about measured genotypes can be found in Table S1. In general, 2-cm-long sensors were used unless tree diameters were too small and required 1-cm-long sensors instead. Sensors were placed about 0.5 m above ground level and covered with reflective insulation. Thermal dissipation sensors were connected to one of three AM16/32B multiplexers that were each connected to a central datalogger (CR1000) which measured sensor data every 30 s and logged averages every half hour. Sapflow sensors, dataloggers, and environmental sensors were powered by marine deep cycle batteries and solar panels.
Sapflow sensor output (mV differential voltages) were converted to sapflow rates (J S ; kg m −2 sapwood area s −1 ) using Baseliner 4 software (Oishi et al., 2008(Oishi et al., , 2016 and the calibration equation described in Granier (1987). Missing data were gapfilled using data from an individual of the same genotype (if available) or data from another genotype from the same taxa. Half-hourly data were integrated to daily and monthly values and calculated on a per tree level by multiplying sapflow rates by the sapwood area of each individual. To determine sapwood areas, trees measured for sapflow in one replicate block were harvested in October of the second growing season (see below) and ratios of stem diameter to bark depth were determined for each taxa (Table S2). All wood was determined to be functional on these 2-year-old trees. These taxa-specific regression equations were used to calculate sapwood areas from diameter measurements of unharvested trees (of the other replicate block) and from May 2019 diameter measurements made when sensors were installed. Sapwood areas were assumed to increase linearly throughout the study period and all sapwood area was assumed to exhibit the same flow rate measured by the sapflow sensor (Samuelson et al., 2007).
To determine half-hourly canopy-averaged stomatal conductance (G S ) tree-level sapflow rates (kg s −1 ) were divided by tree-specific leaf areas to determine leaf-specific transpiration rates (E L ; kg m −2 leaf area s −1 ). Tree leaf areas were determined during biomass harvest (see below) and taxa-specific regression equations between tree basal area × height and leaf area were developed (Table S2) and used to estimate leaf areas for unharvested trees as well as leaf areas for trees at the beginning of the sapflow measurement period (May 2019). Because some D × T, T × D and D × M taxa begin losing leaves in late July and August (well before biomass harvest occurred), the regression equation for D × D genotypes was used to estimate leaf areas in May for D × M, D × T, and T × D taxa. Previous research of leaf area index (LAI) measured in July found that a D × M and T × D genotypes had similar or slightly higher LAI than D × D genotypes (Renninger et al., 2021). A linear relationship between leaf areas estimated for May and leaf areas measured or estimated in October was assumed and differences between measured leaf areas in October and calculated leaf areas in May were used to estimate leaf area loss. For each individual and half-hourly transpiration measurement, average canopy stomatal conductance (G S ; m s −1 ) was calculated as follows: where K G is a coefficient calculated as 115.8 + 0.4226 × air temperature (Phillips & Oren, 1998). Canopy stomatal conductance was further scaled to units of mol m −2 s −1 by dividing values by the density of water vapor scaled based on air temperature (0.0224 × air temp[K]/27).
Half-hourly G S values for time periods when VPD >0.6 kPa were regressed versus environmental parameters for each measured individual to determine genotypeand/or taxa-specific responses. Relationships between G S and volumetric soil moisture content (θ; cm 3 cm −3 ) were nonlinear, therefore, linear regressions were fitted separately for θ < 0.22 cm 3 cm −3 and θ > 0.22 cm 3 cm −3 and slopes of these regressions for each individual were estimated using Sigmaplot version 13 (systat Software Inc.).
To estimate each individual's stomatal sensitivity to VPD (Oren et al., 1999), linear relationships were fitted between G S and the natural log of VPD (lnVPD) with the sapflow dataset split into low soil moisture (<0.22 cm 3 cm −3 ) and average to high soil moisture (>0.22 cm 3 cm −3 ) categories to determine if stomatal sensitivity differs by soil moisture regime. For each G S versus lnVPD linear regression equation, parameters were estimated including G S,ref (yintercept or G S at VPD = 1 kPa), stomatal (G S ) sensitivity (negative slope [m] of the regression line) and scaled G S sensitivity (negative slope/y-intercept).

| Biomass, leaf area and whole-tree water use efficiency
Toward the end of the second growing season (October 2019) but before significant leaf drop in genotypes that hold leaves until fall dormancy, biomass was determined for half (48 individuals from one replicate block) of the individuals measured for sapflow. Ground line diameter (GLD) and diameter at breast height (DBH) were measured, then trees were cut and stem length was measured to estimate tree height. From the cut stem, sapwood diameter was estimated in two azimuthal directions. If trees were multi-stemmed, all stems greater than 2.53 cm at their base were measured and harvested although only the height of the tallest stem was recorded. Small branches containing leaves were removed and placed into plastic bags for transport back to the lab. Defoliated branches and the main stem were weighed in the field to determine fresh green weight. A wood sample (with bark) from the main stem (about 10 cm long) was cut, its fresh field weight measured, and it was transported back to the lab. The volume of the wood sample was determined using volume displacement and it was placed in a drying oven at 105°C and its dry weight was determined. For each individual, all collected leaves were removed from stems and their fresh weight was determined. A subsample of approximately 20 leaves was randomly collected and its fresh area (using a LI-COR 3100 leaf area meter; LI-COR Biosciences Inc.) and fresh weight were obtained and used to calculate total leaf area from the total leaf fresh weight.
Fresh weight of twigs with leaves removed were added to the weight of biomass samples measured in the field and dry biomass (kg) was estimated using the ratio of fresh to dry weight of the wood sample. Whole-tree water use efficiency (WUE T ; g biomass/kg water used) was calculated for each individual measured for sapflow for the period in which sapflow rates were measured. Monthly sapflow data for each individual were summed to calculate seasonal water use. For individuals that were not harvested as well as to determine biomass at the beginning of the sapflow period, linear or non-linear regression equations between dry biomass and basal area × height were determined for eastern cottonwood and hybrid poplars respectively (Table S2). The difference between dry biomass at the end and beginning of the sapflow period was divided by seasonal water use for each individual to calculate WUE T .

| Statistical analysis
Linear and non-linear regression relationships were fitted, and parameters for equations estimated, using Sigmaplot software. Biomass and water use parameters were compared across taxa using linear mixed effect models (nlme package; Pinheiro et al., 2014) in R version 3.6.3 (R Core Team, 2014) including replicate block as a random effect in models. Tukey post-hoc tests were performed in R using the multcomp package (Hothorn et al., 2008). Principal component analysis was performed in R using the prcomp function with all variables scaled and centered on the origin.

| RESULTS
At low soil moistures (<0.22 cm 3 cm −3 ), canopy stomatal conductance (G S ) increased at a faster rate for a given increase in soil moisture than that exhibited at average to high soil moistures (>0.22 cm 3 cm −3 ) with slope terms that were about five times greater under low versus high soil moisture conditions (Table 1). At average to high soil moistures, G S was much more consistent across a large soil moisture range with a low or near zero rate of change ( Figure 1). Taxa did not differ significantly in their responses of G S to soil moisture exhibiting statistically similar slope terms under both low and average to high soil moistures (Table 1) although both D × M and D × T/T × D genotypes exhibited a broader range in G S at average to high soil moistures than D × N and D × D genotypes (Figure 1). Given the different responses of G S at low and average to high soil moistures, relationships between the natural log of vapor pressure deficit (lnVPD) and G S were examined under each soil moisture regime. For D × D, D × N and D × T/T × D genotypes, slopes (−m) of the relationship between G S and lnVPD at low and average to high soil moistures did not differ significantly, while for D × M genotypes, slope terms were about 35% more negative (p < 0.001) under average to high soil moisture compared with low soil moisture conditions (Figure 2).
Because only one taxa exhibited significant differences based on soil moisture content, data were pooled across soil moisture categories to compare relationships between G S and lnVPD across taxa and genotypes. D × M genotypes had the largest reference G S (G S,ref ) and greatest stomatal sensitivity (regression slope; −m) differing significantly from D × D and D × N genotypes (Table 1). Because individuals with larger G S,ref will likely have greater calculated stomatal sensitivity given their larger conductances at low lnVPD, scaled stomatal sensitivity (−m/G S,ref ) was calculated to better compare across genotypes and taxa. D × D, D × T and T × D genotypes had scaled stomatal sensitivities that were significantly larger than the D × M genotypes, with D × N genotypes being statistically similar to all other taxa (Table 1). The D × D genotypes in October had significantly larger leaf areas than any of the hybrid taxa with the D × M genotypes also having significantly lower leaf areas than D × T genotypes (Table 1). Compared T A B L E 1 Height, biomass, and physiological parameters related to water use strategies for Populus deltoides W. Bartram ex Marshall (D) as well as hybrids of P. deltoides crossed (×) with P. maximowiczii A. Henry (M), P. nigra L. (N), and P. trichocarpa Torr. & Gray (T). Ref. G S , reference canopy conductance (G S ) at vapor pressure deficit (VPD) = 1 kPa; G S sensitivity, slope of regression between G S and lnVPD; scaled G S sensitivity, G S sensitivity/ref. G S ; G S slope low/high θ, slope of regression between G S and soil moisture (θ) under low or high θ; WUE, water use efficiency. Means with different superscript letters differed significantly at p < 0.05 Year 2 height (m) 6.6 (0.2) ab 6.7 (0.2) a 5.9 (0.4) b 6.1 (0.2) ab 6.0 (0.4) ab Year 2 dry biomass (kg) 5.

(a) (b) (c) (d)
with calculated leaf areas based on October height and diameter, the D × M genotypes exhibited leaf losses that were about 1.9 times greater than D × N and D × T genotypes and about eight times greater than D × D genotypes. At the end of the first and second growing seasons, D × D and D × M genotypes tended to be taller than D × N and D × T genotypes although only D × M genotypes differed significantly (Table 1). Dry biomass at the end of two growing seasons was statistically similar across D × D, D × M, D × T, and T × D genotypes with D × N genotypes having significantly lower biomass (Table 1). Biomass growth in the second growing season did not differ significantly across taxa (Table 1). Several D × M genotypes were at, or above, the 75th percentile in terms of 2-year dry biomass including '7388', '9711', '9225', '14591' with '14507', '13724', '8002' and '9709' being above the 90th percentile ( Figure 3a). For D × D genotypes, 2-year dry biomass of both '6-5' and '112107' was at, or above, the 75th percentile and '6-4' was near the 90th percentile (Figure 3a). For D × T genotypes, 2-year dry biomass of both '7938' and '10029' was above the 75th percentile (Figure 3a).
In terms of whole-tree water use efficiency (WUE T ), D × M genotypes had significantly lower WUE T than D × D, D × N or D × T genotypes (Table 1). For the D × M genotypes, only genotype '13700' was above the 75th percentile in terms of WUE T , whereas D × D genotypes 'S13C20' and  '6-1' were at, or above, the 75th percentile and '112107', '120-4' and '6-4' were above the 90th percentile in terms of WUE T (Figure 3). For D × N genotypes, '433' and '11867' were above the 75th percentile and '11822' was at the 90th percentile. For D × T genotypes, '10029' and '8729' were above the 75th percentile and '8717' and '7903' were at, or above, the 90th percentile (Figure 3). WUE T did not exhibit a significant correlation with 2-year dry biomass (p = 0.84; Figure 3a) but was significantly correlated with biomass growth occurring in the second growing season (p = 0.004; r 2 = 0.13), which is the year that seasonal water use was estimated ( Figure 3b). The D × M genotype '9189', D × D genotypes '47-5', 'S13C20' and '6-5', D × N genotypes '433' and '11822', D × T genotype '8717' and T × D '9755' were above the 50th percentiles in both 2-year dry biomass and WUE T (Figure 3a). D × T genotype '10029' was above the 75th percentile in both parameters and D × D genotypes '112107' and '6-4' were above the 90th percentile in WUE T and above the 75th percentile in 2-year dry biomass ( Figure 3a). All measured parameters (Table 1) were incorporated into a principal component analysis to determine how parameters varied with one another and where individuals genotypes were located based on multi-parameter principal components. PC1 captured the interaction between G S , leaf area loss and WUE T wherein trees with high PC1 scores had high reference G S (G S,ref ) and stomatal sensitivity, high leaf area loss and low October leaf areas and WUE T (Figure 4; Table S3). Scaled G S sensitivity exhibited a significant negative correlation with calculated leaf area loss of trees across taxa (p < 0.0001; r 2 = 0.47; Figure 5) wherein trees that maintain leaf area throughout the growing season tend to exhibit high stomatal sensitivity. PC2 mainly captured variability in biomass parameters, scaled stomatal sensitivity, October leaf areas and changes in G S with soil moisture at low volumetric soil moistures (G S slope, low θ). Trees with high PC2 scores were smaller with low scaled stomatal sensitivity, October leaf areas and a low response of G S to decreasing soil moisture levels (Figure 4) Figure 6a) and positively correlated with scaled stomatal sensitivity (p < 0.0001; r 2 = 0.29; Figure 6b). In addition, WUE T was positively and non-linearly correlated with leaf areas measured in October (p = 0.0002; r 2 = 0.31; Figure 6c) and negatively correlated with calculated leaf area losses (p < 0.0001; r 2 = 0.33; Figure 6d). Similarly to F I G U R E 6 Whole-tree water use efficiency (WUE T ; g biomass/kg water used) versus (a) Reference canopy conductance (G S ; mol m −2 s −1 estimated at vapor pressure deficit (VPD) = 1 kPa), (b) scaled G S sensitivity to VPD (slope of relationship/reference G S ), (c) leaf areas measured in October (m 2 ), and (d) differences between leaf areas calculated for May and measured in October (m 2 ) for Populus deltoides W. Bartram ex Marshall × P. deltoides (D × D; red), P. deltoides × P. maximowiczii A. Henry (D × M; green), P. deltoides × P. nigra L. WUE T , biomass growth in the second growing season was positively correlated with reference G S (G S,ref ; p = 0.002; r 2 = 0.18; data not shown) and scaled stomatal sensitivity (p = 0.01; r 2 = 0.12) across genotypes and taxa (Figure 7a). Genotypes '11822' (D × N), '13724' (D × M) and eastern cottonwood genotypes '47-5' and '6-5' were at, or above, the 75th percentile for biomass growth and above the 50th percentile for scaled G S sensitivity. D × T genotypes '7938' and '10029' as well as eastern cottonwood genotypes '6-4' and '112107' were above the 90th percentile for biomass growth and at, or above, the 50th percentile in terms of scaled G S sensitivity with D × D genotypes being above the 90th percentile in both categories (Figure 7a). Biomass growth was positively correlated with changes in G S with increasing soil moisture at low soil moisture contents (G S slope, low θ; p = 0.02; r 2 = 0.12) across taxa and genotypes ( Figure 7b). However, genotypes that maintain G S with decreasing soil moisture (and hence exhibit low slope terms) could be considered to be more drought tolerant and several genotypes have both high biomass growth and lower changes in G S with changes in soil moisture. For example, D × M genotypes '7388' and '9225' were above the 75th percentile in terms of biomass growth and below the 50th percentile in terms of G S response to low soil moisture while eastern cottonwood genotypes '112107' and '6-4' as well as D × M genotype '8002' were above the 90th percentile in terms of biomass growth with a lower than average G S response to soil moisture (Figure 7b). Populus taxa differed in their relationships between year two biomass growth and leaf area dynamics. For D × M genotypes, biomass growth was positively correlated with both October leaf areas (p = 0.009; r 2 = 0.29; Figure 7c) and calculated leaf area losses (p = 0.003; r 2 = 0.37; Figure 7d). However, for D × D genotypes, biomass growth was positively correlated with October leaf areas (p < 0.0001; r 2 = 0.85; Figure 7c) and marginally negatively correlated with leaf area losses (p = 0.05; r 2 = 0.33; Figure 7d). D × N genotypes exhibited a positive correlation between biomass growth and October leaf areas (p = 0.01; r 2 = 0.90; Figure 7c) and no correlation between biomass growth and leaf area loss (p = 0.50; Figure 7d). For D × T and T × D genotypes, neither October leaf areas (p = 0.34; Figure 7c) nor leaf area losses (p = 0.11; Figure 7d) were significantly correlated with biomass growth.

| DISCUSSION
By selecting genotypes that exhibit high biomass and biomass growth, high water use efficiency, stomatal sensitivity and relatively low changes in canopy conductance (G S ) with decreasing soil moisture, the most drought tolerant genotypes for the southeastern US can be selected for future testing and breeding. In addition, genotypes that maintain G S at decreasing soil moisture may also sequester more belowground carbon (Sanchez et al., 2007;Stanton et al., 2002) in roots compared with individuals that are more responsive to changes in soil moisture. Two eastern cottonwood genotypes, '112107' and '6-4', scored high on all metrics being above the 90th percentile in terms of water use efficiency, scaled G S sensitivity and were below average in terms of their G S response to low soil moisture. Eastern cottonwood genotypes '47-5' and '6-5' also exhibited higher than average water use efficiency and stomatal sensitivity. Giovannelli et al. (2007) also found that an eastern cottonwood genotype was more drought tolerant than a D × N genotype. For hybrid poplar genotypes, D × N '11822' was near the 90th percentile in terms of water use efficiency and above the 50th percentile in terms of stomatal sensitivity, and D × T '10029' had above average stomatal sensitivity and was above the 75th percentile in terms of water use efficiency. Bonhomme et al. (2008) found that one D × N genotype, 'Soligo', had high water use efficiency in both alluvial and non-alluvial sites compared with T × D genotypes. D × M genotypes tended to be below the 50th percentile in terms of water use efficiency, but genotypes '8002', '7388', and '9225' were better than average in terms of smaller changes in G S with decreasing soil moisture suggesting drought tolerance. In addition, if water is not limited or high water use (low water use efficiency) is desired in terms of phytoremediation (Bonhomme et al., 2008), several D × M genotypes exhibited high biomass growth (at or above the 75th percentile), and below average water use efficiency and low scaled G S sensitivity including genotypes '9709', '14507', '9225', '8002', and '7388'. In support of our hypothesis, we found that the tested Populus taxa exhibited differing water use strategies, with eastern cottonwood and D × M genotypes specifically differing from one another. While eastern cottonwoods tended to maintain high leaf areas and strong stomatal sensitivity to control water use and achieve high WUE T (Muries et al., 2019), D × M genotypes exhibited larger leaf area losses with less stomatal sensitivity and lower WUE T . Attia et al. (2015) found that anisohydric poplars that exhibit low stomatal control had lower water use efficiency than isohydric poplar genotypes that exhibited higher water use efficiency. Several other studies found that hybrid poplars exhibit leaf area loss as a drought avoidance strategy (Arango-Velez et al., 2011;Giovannelli et al., 2007;Monclus et al., 2006Monclus et al., , 2009 which could reallocate nutrients and resources to newly created leaves (Arango-Velez et al., 2011;Chaves et al., 2003). We also found that leaf area losses and scaled stomatal sensitivity were inversely correlated across genotypes wherein trees that lost significant leaf area throughout the growing season exhibited less stomatal sensitivity suggesting differing distinct strategies to deal with water stress. D × N, D × T, and T × D genotypes tended to fall between the eastern cottonwood and D × M genotypes in terms of leaf area losses and stomatal sensitivity. P. trichocarpa tends to be less drought tolerant and water use efficient compared with eastern cottonwood (Bassman & Zwier, 1991) and hybrids of the two species should exhibit more drought tolerance than P. trichocarpa (Bassman & Zwier, 1991;Maier et al., 2019). In fact, Zalesny et al. (2019) found that eastern cottonwood genotypes exhibited relatively low WUE compared with backcross (T × D) × D genotypes, which exhibited high WUE and Ghezehei et al. (2019) showed that growth of eastern cottonwood genotypes was more dependent on favorable site conditions compared with T × D and D × M genotypes likely because eastern cottonwood improved genotypes were selected from field trials on fertile, alluvial sites. Marron et al. (2007) reported that D × T genotypes were more productive than D × N genotypes, with our study also finding that D × N genotypes exhibited the lowest biomass and biomass growth at the southeastern U.S. site.
We found that while 2-year biomass was not correlated with WUE T , biomass growth in the second growing season was significantly, positively correlated with WUE T . Several other studies report a positive correlation between biomass or growth rate and water use efficiency parameters Monclus et al., 2009;Pilipović et al., 2022;Renninger et al., 2021) while other studies found no correlation (Dillen et al., 2011;Marron et al., 2005;Monclus et al., 2005Monclus et al., , 2006Rae et al., 2004;Zhao et al., 2021). Across all genotypes, we found a mean WUE T of about 4.14 g biomass/kg H 2 O with 10% of genotypes falling below 2.29 g/kg and 10% of genotypes falling above 5.98 g/kg. These WUE T estimates compare well with other whole tree or whole stand estimates of water use efficiency in Populus which range from 1.7 to 7.9 g biomass/kg water (assuming 50% C for studies that report gC/kg H 2 O) for studies that estimated water use efficiency using sapflow and biomass data (Kim et al., 2008;Maier et al., 2019) and eddy covariance (Jones et al., 2017;Migliavacca et al., 2009;Xu et al., 2020) techniques. The similarity in ranges of WUE T across studies suggests that our whole tree water use estimates are not under-or overestimated and that assumptions of constant sapflow with sapwood depth are appropriate. However, for taxa including D × M, low WUE T may be due to an overestimation of water use based on the assumption of completely functional sapwood in the cross section and constant sapflow with sapwood depth. More research is needed to determine if sapflow rates decline with depth in D × M genotypes.
In this study, we found that the most water use efficient genotypes had the lowest reference canopy conductance and highest stomatal sensitivity and tended to have high leaf areas that were retained throughout the growing season. Marron et al. (2005) reported that carbon isotope discrimination, an indirect measure of water use efficiency, was also correlated with leaf area in terms of the total numbers of leaves on trees. Genotypes exhibiting high WUE may be less susceptible to drought stress compared with genotypes with low WUE (Yin et al., 2005); however, genotypes may have other strategies for coping with drought stress including leaf area reductions (Monclus et al., 2006). We determined that biomass growth was also positively correlated with stomatal sensitivity and the rate of change in canopy conductance with changes in soil moisture content under low soil moistures. Maier et al. (2019) also reported that slow growing Populus genotypes tended to exhibit lower stomatal sensitivity compared with more productive genotypes. However, Attia et al. (2015) found that anisohydric genotypes which potentially exhibit lower stomatal sensitivity were more productive than more sensitive, isohydric genotypes under well-watered conditions. In response to low soil moisture or soil water stress conditions, we showed that genotypes that were more responsive in terms of stomatal closure to decreasing soil moisture had greater biomass growth. However, Arango-Velez et al. (2011) reported that stomatal responsiveness to drought conditions did not explain differences in growth for measured hybrid poplars and balsam poplars although a smaller number of genotypes were tested.
We found that biomass growth was positively correlated with leaf area measured in October for both D × D and D × M genotypes even though these taxa exhibited differing water use strategies with D × D genotypes having higher stomatal sensitivity and WUE T and D × M genotypes having greater leaf area loss and lower sensitivity. Several other studies also reported that leaf area was positively correlated with biomass Monclus et al., 2005;Pellis et al., 2004;Pregitzer et al., 1990;Rae et al., 2004). However, D × D and D × M genotypes had opposing relationships between biomass growth and calculated leaf area losses throughout the growing season, wherein there was a positive correlation for D × D genotypes, but a negative correlation in D × M genotypes. Similar to D × D genotypes, Verlinden et al. (2013) showed that leaf area duration was positively correlated with biomass production in 12 hybrid poplar genotypes. For D × M genotypes, this negative relationship may be driven by the tendency for larger trees with the potential for greater biomass growth to have more leaf area to lose compared with less productive genotypes.

| CONCLUSIONS
Overall, we found that taxa differed in their water use strategies, particularly D × D and D × M genotypes, however certain D × D, D × M and D × T genotypes exhibited high 2-year biomass (around 6 kg/tree) and biomass growth across the 2-year study period. D × D genotypes exhibited high stomatal sensitivities and water use efficiency and held leaves longer into the growing season while D × M genotypes were less water use efficient, lost significant leaf area throughout the growing season, and tended to exhibit less stomatal sensitivity. Across all taxa, genotypes with lower reference canopy conductance, higher scaled stomatal sensitivity, large October leaf areas, and low leaf area loss throughout the growing season tended to be the most water use efficient on a whole-tree level. Across taxa, biomass growth during the second growing season was positively correlated with whole tree water use efficiency, scaled stomatal sensitivity, October leaf area, and the rate of change in G S with decreases in soil moisture under dry conditions. Taken together, these data support our hypothesis and shed light on genotypes that may show promise on drier sites given their high biomass growth coupled with stomatal sensitivity, water use efficiency and low stomatal response under dry soil conditions. Likewise, genotypes with high biomass growth coupled with high water use and low stomatal sensitivity may be optimal for wetter sites and/or for phytoremediation purposes. However, future research is necessary to determine the longer-term viability of identified genotypes in terms of continued high productivity, disease resistance (particularly to Septoria stem canker; Kaczmarek et al., 2013), coppice production potential and disease resistance following coppice. Overall, genotype-specific data on water use strategies across individuals with large genetic variability provide ranges for Populus physiological parameters which can be used to refine model estimates of Populus production and ecosystem impacts under a variety of climate conditions (Wang et al., 2013) and inform future Populus breeding efforts.