Small tropical forest trees have a greater capacity to adjust carbon metabolism to longterm drought than large canopy trees

The response of small understory trees to long-term drought is vital in determining the future composition, carbon stocks and dynamics of tropical forests. Long-term drought is, however, also likely to expose understory trees to increased light availability driven by drought-induced mortality. Relatively little is known about the potential for understory trees to adjust their physiology to both decreasing water and increasing light availability. We analysed data on maximum photosynthetic capacity ( J max , V cmax ), leaf respiration ( R leaf ), leaf mass per area (LMA), leaf thickness and leaf nitrogen and phosphorus concentrations from 66 small trees across 12 common genera at the world’s longest running tropical rainfall exclusion experiment and compared responses to those from 61 surviving canopy trees. Small trees increased J max , V cmax , R leaf and LMA (71%, 29%, 32%, 15% respectively) in response to the drought treatment, but leaf thickness and leaf nutrient concentrations did not change. Small trees were significantly more responsive than large canopy trees to the drought treatment, suggesting greater phenotypic plasticity and resilience to prolonged drought, although differences among taxa were observed. Our results highlight that small tropical trees have greater capacity to respond to ecosystem level changes and have the potential to regenerate resilient forests following future droughts. of two the TFE (1 50% of canopy throughfall has at m since control (1 <50 m The two trenched around the to a depth of 1-2 horizontal throughflow. Both have continuously maintained size after accounting for canopy shading. The TFE effect on small tree J max and V cmax was reduced compared to large trees by 100.3% (Δ18.88 ± 4.56 μmol m -2 s -1 ) and 88.1% (Δ6.69 ± 2.58 μmol m -2 s -1 ), respectively. We found no significant effect of canopy shading, TFE treatment or tree size on leaf thickness. A significant genus effect was present for all leaf traits except R leaf , showing these leaf traits are not solely determined by the environment, but also vary with taxonomy. For


Introduction
Climate change can simultaneously affect multiple environmental variables across ecosystems globally (IPCC, 2019). However, little is known about how trees respond to multiple environmental shifts, especially in tropical forests (Niinemets, 2010;Bonal et al., 2016). Both episodic and sustained droughts have been shown to cause mortality of large canopy trees in Amazonia (Nepstad et al., 2007;Phillips et al., 2010;Bennett et al., 2015;Meir et al., 2015;Rowland et al., 2015a), resulting in canopy openings and changes in the understory environment (Chazdon & Fetcher, 1984;Brown, 2009;Pfeifer et al., 2016;Rowland et al. under review). Small understory trees must therefore respond to concurrent reductions in soil moisture availability and increased light availability to survive, compete and grow under drought conditions. The ability of these small trees to adjust their physiology in response to multiple environmental shifts could be critical for predicting the future of tropical forests.
If canopy trees are susceptible to mortality during drought under future climates, it becomes important to understand the response of understory trees to drought conditions within higher light environments, and concomitant changes in air temperature and VPD. However, most of our current knowledge of tropical tree drought responses is focused on either large canopy trees or seedlings (e.g. Poorter & Hayashida-Oliver, 2000;Nepstad et al., 2007;Schuldt et al., 2011;O'Brien et al., 2014;Rowland et al., 2015b;Feldpausch et al., 2016;O'Brien et al., 2017). Some studies have investigated the impacts of short-term drought events on tropical understory trees (e.g. Phillips et al., 2010;Newbery et al., 2011), but, to our knowledge, no studies to date have investigated the effects of prolonged drought in tropical forests on understorey trees also exposed to elevated light conditions. This represents a gap in our knowledge as understory trees tend to have very different resource requirements to adult trees and seedlings (Sterck et al., 2014). If we are to fully predict the fate of tropical forests and improve climate predictions from the latest generation of demography-based vegetation models, it is critical to understand the physiological responses of understory trees in drought-affected forests (Moorcroft et al., 2001;Smith et al., 2008;Fisher et al., 2018;Moore et al., 2018).
Differences in physiology, microenvironments and resource requirements will likely lead to different drought responses in small understory trees compared with large canopy trees (Kitajima & Poorter, 2008). Small trees tend to be particularly sensitive to reductions in soil moisture availability (Ruger et al., 2012;Kitajima et al., 2013;Quevedo-Rojas et al., 2018), because of smaller total carbon reserves (Hartmann et al., 2018) and shallower rooting depths (Stahl et al., 2013;Brum et al., 2018). However, vulnerability to drought may actually be lower in small trees, as carbon reserves per unit biomass may be greater (Hartmann et al., 2018). An ability to maintain a positive balance between carbon assimilation and use will ultimately be critical for these small trees to survive and avoid carbon starvation under drought conditions (O'Brien et al., 2014;McDowell et al., 2018). The adjustment of functional traits related to carbon metabolism, including maximum photosynthetic capacity (J max and V cmax ) and leaf respiration (R leaf ), is important for buffering long-term reductions in soil moisture availability. Under drought stress, some trees have been shown to reduce photosynthetic capacity because of drought-induced impairment or for nutrient re-allocation for stress repair (Damour et al., 2008;Damour et al., 2009), whilst others have been shown to maintain photosynthetic capacity to optimise carbon assimilation during wetter periods (Rowland et al., 2015b). In response to this, R leaf can increase under drought conditions to support stress-related repair or support osmoregulation (Rowland et al., 2015b), however perhaps more typically R leaf has been found to decrease during drought stress in response to reduced photosynthesis (Atkin & Macherel, 2009;Ayub et al., 2011). The extent of plasticity in these carbon metabolic traits could determine the likelihood of small trees This article is protected by copyright. All rights reserved.
surviving long periods of reduced soil moisture availability. However, these traits are not simply controlled by drought stress, but are also highly sensitive to light availability (Poorter et al., 2009;Atkin et al., 2015;Hasper et al., 2017). Consequently, the light environment of a tree should also be considered when understanding responses to drought.
Tropical forests display strong vertical gradients in light availability with small understory trees generally adapted to shade conditions relying on diffuse light and unpredictable, fleeting sunflecks for the majority of their carbon assimilation (Chazdon & Pearcy, 1991;Leakey et al., 2003). Under normal conditions, tropical trees are able to acclimate their leaf physiology to the steep gradient in irradiance experienced through the vertical profile of the canopy, with photosynthetic capacity and LMA increasing with light availability (Meir et al., 2002;Poorter et al., 2009;Cavaleri et al., 2010;Domingues et al., 2010). Trees modify their leaf physiology to optimise the balance between carbon gain and carbon and water loss, in order to maximise growth, reproduction and competitiveness. Both large and small understory trees in tropical forests have been shown to be highly responsive and plastic to changes in their light environment (Kitajima et al., 2013;Quevedo-Rojas et al., 2018), suggesting light is likely to be the most limiting factor for photosynthesis in intact tropical forests. However, if another factor, such as water availability, also becomes limiting, these trees may no longer be able to acclimate to high light levels, potentially leading to negative impacts of increasing light on growth and survival.
Increases in light availability under drought conditions may result in excessive photon flux density, elevated leaf temperatures and elevated VPD, inducing photoinhibition (Kamaluddin & Grace, 1992;Mulkey & Pearcy, 1992;Krause et al., 1995), prolonged stomatal closure (Reynolds-Henne et al., 2010) and xylem embolism. To avoid these negative consequences of concurrent high light and drought stress, small understory trees may need to modify their physiology in different ways compared to canopy trees.
Insights from the same eastern Amazon throughfall exclusion experiment used in this study showed the responses of large trees to drought were indeed influenced by the light environment (Rowland et al. under review). Large trees that had also experienced increased canopy exposure following mortality events, experienced reductions in photosynthetic capacity under drought conditions, whilst those that did not maintained photosynthetic capacity (Rowland et al. under review). However, large trees will have experienced relatively minor shifts in their light environment when compared to small understory trees. These larger trees may also be more exposed to hydraulic stress from the drought, which may limit their ability to respond positively to elevated light (Bittencourt et al., 2020), although they may already be acclimated to higher VPD conditions at the top of the canopy so may not experience hydraulic stress from sudden shifts in VPD that will occur in the understory. Studying how small understory trees adjust their leaf physiology to concurrent shifts in water and light availability and how these responses differ from large canopy trees may also give new insights into the potential for forest regeneration following drought events.
Here, we use data from a 15-year tropical forest drought experiment located in eastern Amazonia, to test how long-term drought affects carbon metabolism and leaf morphology in small understory trees.
By comparing data on maximum photosynthetic capacity (J max and V cmax ), leaf dark respiration (R leaf ), leaf mass per area and leaf thickness from 66 small understory trees (1-10 cm DBH) against those from 61 surviving canopy trees across 12 genera between a throughfall exclusion (TFE) experiment and neighbouring control plot, we test the following hypotheses:

1)
Small understory trees respond to canopy openings following long-term drought stress by increasing photosynthetic capacity (J max and V cmax ) in response to elevated light availability.

2)
Small understory trees increase leaf dark respiration and LMA in response to long-term drought stress.

3)
Leaf physiological traits (J max , V cmax , R leaf and LMA) are more responsive to reduced soil moisture availability and canopy openings following prolonged soil moisture deficit in small understory trees than large trees.
This article is protected by copyright. All rights reserved.

Study site
This study was carried out at the world's longest running TFE experiment in Caxiuanã National Forest Reserve, Para, Brazil (1°43′S, 51°27'W). The experiment is located in seasonally dry terra firme forest with an annual precipitation of 2000-2500 mm. Here, a pronounced dry season occurs between June  (Fisher et al., 2007;da Costa et al., 2010;Rowland et al., 2015b;Meir et al., 2018). Reduced soil moisture availability on the TFE plot compared with the control plot has been shown previously with lower mean pre-dawn leaf water potentials on the TFE plot (SI Fig. 1) (Bittencourt et al., 2020).

Sampling
From August to September 2017, we sampled 66 small trees (1-10 cm diameter at breast height; 1.3m DBH; 2.7-23.0 m height) across the two plots: 30 from the TFE and 36 from the control. We selected individuals from 12 of the most common genera within the two plots (Duguetia, Eschweilera, Inga, Iryanthera, Licania, Manilkara, Minquartia, Ocotea, Protium, Tetragastris, Swartzia, Vouacapoa) in accordance with a corresponding study on large tree carbon metabolism and storage (SI Table 1a; Rowland et al. under review). All species represent canopy or emergent trees when mature. In order to minimise edge effects within the plots, we sampled trees located within one quarter of each 1 ha plot (i.e. 0.25 ha), with all trees located at least 20 m from the plot perimeter. For each genus, individual tree selection was designed to cover a range of sizes from 1 to 10 cm DBH. Two people independently assessed the relative canopy position of each tree and assigned it as either shaded or in a canopy gap according to whether the leaves experienced vertical shading or not. All canopy position assessments were recorded at approximately the same time each day, and the presence of leaves directly above the target tree were used to minimise bias from different sun positions at the time of measurement.
We selected an additional 61 large trees (>20 cm DBH: range 20.2-67.9 cm) from a parallel study across eight corresponding genera (Eschweilera, Inga, Licania, Manilkara, Minquartia, Protium, Swartzia, Tetragastris) to compare responses of large trees against a subset of 48 small trees for the same eight genera (SI Table 1b). For more details about large tree sampling, see Rowland et al. (under review).

Gas exchange measurements
During the peak dry season (August-September 2017), we collected a branch of approximately 1 m length from the top of the crown of each tree, using pole pruners. For the majority of the trees (> 95%), the branches were cut between 09:00 and 10:00 hrs, but on some occasions branches were cut between 10:00 and 13:00. Once harvested, branches were immediately placed in water and were cut twice underwater to restore water supply to the leaves (Domingues et al., 2010). The branches were subsequently left to stabilise in full sunlight for a minimum of 30 minutes. Following stabilisation, we selected non-senescing, fully formed leaves to be measured using two cross-calibrated portable photosynthesis systems (LI-6400XT and LI-6800, LI-COR, Nebraska, USA). For each tree, we measured one leaf for estimates of photosynthetic capacity and one neighbouring leaf for dark-adapted leaf respiration (R leaf ).
We performed photosynthetic CO 2 response curves (A-C i ) to estimate maximum photosynthetic capacity. We placed a leaf within the leaf chamber of a portable photosynthesis system and measured net photosynthetic assimilation (A) and leaf internal carbon dioxide (CO 2 ) concentration (C i ). We generated A-C i curves by manipulating CO 2 concentrations within the leaf chamber (400, 200, 75, 400, 800, 1200, 2000 ppm), whilst providing a photosynthetic active radiation (PAR) of 1500 μmol m -2 s -1 , a temperature of 28 ˚C and a relative humidity of 60-70%. Light response curves carried out on a subset of the samples (n = 8) ensured 1500 μmol m -2 s -1 PAR represented saturating light conditions for the leaves (data not shown). In order to maintain data quality, A-C i curves were aborted if stomatal conductance (g s ) dropped below 0.03 mol m -2 s -1 (following Rowland et al., 2015b). No difference between measurements of photosynthetic capacity or R leaf on cut versus uncut branches have been found in previous studies at this site (Rowland et al., 2015b), suggesting our measurements reflect true values of in situ leaves and differences between the two plot treatments. We waited for steadystate conditions within the leaf chamber to be reached before any gas exchange measurements were recorded. Using these data, we estimated the maximum rate of electron transport (J max ) and the maximum rate of carboxylation (V cmax ) standardised to 25 ˚C following the equations from the C 3 photosynthesis model of Farquhar et al. (1980) as in Sharkey et al. (2007) using the optim function from the stats package in R statistical software (R Core R Core Team, 2013). Ten trees were removed from our sample where A-C i curves could not be fitted.
Dark-adapted leaf respiration (R leaf ) was measured on a leaf adjacent to that used to measure maximum photosynthetic capacity. Leaves were wrapped in aluminium foil for a minimum of 30 minutes before the portable photosynthesis system was used to measure R leaf . We maintained stable leaf chamber conditions of 400 ppm CO 2 concentration, 0 μmol m -2 s -1 PAR and 28 ˚C for all respiration measurements. Following stabilisation within the leaf chamber, three measurements of respiration were recorded at five second intervals, then standardised to 25 ˚C using a Q10 value of 2.2 following Rowland et al. (2015b) and a mean value was calculated.

Leaf morphological traits
After completing leaf gas exchange measurements, we removed each of the leaves used to measure A-C i curves and R leaf from the branch and placed them in a sealed airtight plastic zip-lock bag. Moist cotton wool was placed in the bag to maintain high levels of humidity and prevent any water loss from the leaf. We scanned leaves using a flatbed scanner (CanoScan LiDE 120, Canon Inc., Tokyo, Japan) and calculated the area of each leaf using ImageJ software (Schneider et al., 2012). Leaves were then placed in an oven to dry for 24 hours at 70 ˚C, to constant mass. We measured dry leaf mass using a precision balance and calculated leaf mass per area (LMA, g m -2 ) by dividing dry leaf mass by leaf area.
We used LMA from the leaves used to measure A-C i curves and R leaf to estimate maximum photosynthetic capacity and R leaf , respectively, on a mass basis. We measured leaf thickness at three different points on fully hydrated leaves using digital callipers avoiding any major veins and calculated the mean. We averaged the LMA and thickness for the A-C i and R leaf leaves to generate an estimate for the overall branch.

Leaf nutrient analyses
We collected an additional sample, depending on leaf size, of 3-20 leaves adjacent to the leaves used for the gas exchange measurements to quantify leaf nitrogen and phosphorus concentrations ([N] leaf , [P] leaf ). Following collection, leaves were dried in an oven for 24 hours at 70 ˚C to constant mass before being combined with the leaves used for gas exchange measurements. Major veins were then removed and the remaining foliar matter was ground to a fine powder using a ball mill. Nitrogen concentrations were measured using the semi-micro Kjeldahl method (Malavolta, 1997), whilst phosphorus concentrations were measured using a Femto 600+ Spectrophotometer using the ammonium metavandate method (Malavolta, 1997). All analyses were tested against laboratory standards. We divided the gas exchange parameters on a mass basis by leaf nutrient concentrations to estimate leaf photosynthetic and respiratory nutrient use efficiency.

Data analysis Small tree physiological responses
We used linear mixed effects models, using the package lme4 (Bates et al., 2014), to test for treatment (TFE vs control) and taxonomic effects on leaf gas exchange, morphological, nutrient content and nutrient-use efficiency traits in small trees (n = 66). Taxonomy was included as a random intercept effect in our models, whilst the TFE treatment was incorporated as a fixed effect. We tested for a genus and a species nested within genus taxonomic effect by comparing the full linear mixed effects model to a generalised least squares model, following Zuur et al. (2009)  Since traits may be coordinated, we additionally used standardised major axis regression (SMA) to test the effect of the TFE trait trade-offs. We tested for differences in the slopes and intercepts of the bivariate trait-trait relationships between small trees on the different treatments using a Wald test in the package smatr (Warton et al., 2012). We investigated relationships between the gas exchange measurements and leaf morphological variables and the relationship between Jmax and V cmax .

Small and large tree comparisons
We tested for differences in individual tree-level responses to the TFE treatment for large canopy (n = 61) and small understory trees (n = 48), for the same set of species and genera in which data were available (SI Table 1b). We used linear mixed effect models to test the effect of the TFE treatment, tree size (large vs small), canopy shading (canopy gap vs fully shaded) and the interactions between treatment and tree size, and between treatment and canopy shading on leaf gas exchange and morphological traits. We also tested for taxonomic effects by including genus and species nested within genus as random effect variables and comparing to a generalised least squares model, following the same protocol used for small tree analyses (Bartón, 2018). Within this paper, all data represent the mean and associated errors denote standard errors of the mean. All data analyses were undertaken on individual tree-level data in R (R 3.5.1, R Core Team 2018).

Physiological responses to long-term drought in small understory trees
We found a significant positive effect of the TFE, relative to the control plot, on both mean J max (71.1%; Δ14.18 ± 2.65 μ mol m -2 s -1 , P < 0.001; Fig. 1a) and V cmax (29.2%; Δ3.99 ± 1.40 μ mol m -2 s -1 , P < 0.01; trees found on the TFE plot (P < 0.01; Fig. 1d), but mean leaf thickness did not significantly differ between the two treatments (Table 1). We found no significant differences in mean leaf nitrogen and phosphorus concentrations ([N] leaf and [P] leaf ) on a mass-basis between the control and TFE treatments ( Fig. 1 e-f, Table 1). We found a significant effect of genus on the intercept for all of the traits except J max and R leaf , but no significant species-nested-within-genus effect for any traits (SI Table 2).
Using standardised major axis (SMA) regression to compare differences in the relationships between J max , V cmax , R leaf , and LMA across the two experimental plots, we found no significant trait-trait relationships, except for J max -V cmax (SI Table 3). An increase in the log 10 -transformed electron transport rates relative to carboxylation on the TFE treatment compared with the control was observed (TFE = 1.72 ± 0.07 , Control = 1.36 ± 0.08, Wald = 8.503, df = 1, P = 0.004; Fig. 2), but no difference in the slope of the relationship (P = 0.757).
We found no relationship between J max , V cmax or R leaf with [N] leaf or [P] leaf on either plot (SI Table 3).
However, the TFE trees significantly increased J max on a mass basis per unit N leaf and unit P leaf relative to the control (Fig. 3). J max increased per unit nitrogen and unit phosphorus in the TFE trees by 0.006 ± 0.002 μmol g -1 s -1 and 0.227 ± 0.068 μmol g -1 s -1 , respectively (J max /N leaf : P = 0.018; J max /P leaf : P < 0.01; Fig. 3). Overall, neither V cmax nor R leaf increased per unit N leaf or P leaf in the TFE trees relative to the control (Fig. 3).

Comparison of responses to long-term drought between large canopy and small understory trees
We compared the responses of leaf traits in large (>20 cm DBH) and small (1-10 cm DBH) trees to the TFE treatment and to canopy shading using linear mixed models. We found a significant effect of tree size and the TFE treatment on J max , V cmax , R leaf , and LMA, except for a non-significant effect of the TFE treatment on V cmax and R leaf ( Fig. 4; Table 1). J max , V cmax , and R leaf were, respectively, 33.8% (Δ18.82 ± 3.01 μmol m -2 s -1 ), 25.3% (Δ7.59 ± 2.18 μmol m -2 s -1 ) and 36.5% (Δ0.23 ± 0.05 μmol m -2 s -1 ) lower in small trees after accounting for canopy shading in the models. Canopy shading had a significantly negative effect on J max and V cmax , but not R leaf or LMA, with maximum photosynthetic capacity higher for trees positioned in the canopy or a canopy gap compared to trees shaded by the canopy (Table 1).
In addition, a significant interaction between the TFE treatment and tree size for J max and V cmax was found, showing the response of maximum photosynthetic capacity to the TFE treatment to be dependent on tree size after accounting for canopy shading. The TFE effect on small tree J max and V cmax was reduced compared to large trees by 100.3% (Δ18.88 ± 4.56 μmol m -2 s -1 ) and 88.1% (Δ6.69 ± 2.58 μmol m -2 s -1 ), respectively. We found no significant effect of canopy shading, TFE treatment or tree size on leaf thickness. A significant genus effect was present for all leaf traits except R leaf , showing these leaf traits are not solely determined by the environment, but also vary with taxonomy. For example, Inga and Swartzia had the highest J max and V cmax , respectively, whilst Protium had the lowest J max and V cmax of the genera studied, according to the intercepts in our mixed effect models. See Table   1 and SI Table 4 for effect sizes, standard errors, significance and R 2 values.
When the above results were repeated separately for the five most common genera in our dataset, we discovered the effect of tree size and the TFE treatment varied for different genera (Fig. 5). We show Eschweilera and Swartzia to have a positive response to the TFE treatment in small trees for J max , with Swartzia having the largest (positive) response for V cmax (Fig. 5). In contrast our data show Inga and Protium to exhibit little response to the TFE treatment in small trees for all traits. Different genera also exhibit different degrees of variation in their leaf morphology, showing that the responses to the TFE treatment vary by taxa (Fig. 5). Despite relatively small sample sizes, our data indicate similar or greater intraspecific than interspecific variation in leaf traits of small trees, especially photosynthetic capacity on the TFE, with the standard deviation in trait values within species exceeding the standard deviation in species-level means on the TFE in 8/9, 6/9, 4/10, and 2/10 species for J max , V cmax , R leaf and LMA respectively (SI Table 5).

Discussion
We demonstrate that small understory trees (1-10 cm DBH) are more responsive than large canopy trees (>20cm DBH) to prolonged drought conditions. Canopy openings driven by drought-induced mortality of large trees (Rowland et al., 2015a) have allowed small understory trees to increase photosynthetic capacity (J max & V cmax ), leaf respiration (R leaf ) and leaf mass per area (LMA) in spite of reduced soil moisture availability and no changes in leaf nutrient concentrations ( Fig. 1 & SI Fig. 1 &   2). These small understory trees show greater capacity to adjust their leaf physiology than large canopy trees following 15 years of through-fall exclusion (TFE; Fig. 4), suggesting responses to drought are dependent on tree size. Measured values of J max , V cmax , R leaf and LMA in this study remained lower in small trees compared to large trees for both treatments, but values for small understory trees under the TFE treatment approach those of large trees for J max and V cmax after accounting for canopy position (canopy gap vs shaded understory) in our statistical models (Table 1). This indicates small trees are sufficiently plastic to increase photosynthetic capacity with higher light availability, even under drought conditions.
Maximum photosynthetic capacity was highly responsive to the TFE treatment in small understory trees with a 71% increase in the maximum rate of electron transfer (J max ) and a 29% increase in the maximum rate of carboxylation (V cmax ; Fig. 1). The increase in light availability in the droughted forest and the release from extreme light limitation is likely to be driving these physiological adjustments as light represents an important control on both traits (Hasper et al., 2017). Typically, very little light penetrates the canopy to the understory in tropical forests because of the high leaf area index characteristic of tropical rainforests (Asner et al., 2003). With the opening of the canopy, increased light availability may mean understory trees are no longer dependent on sunflecks for the majority of their photosynthesis (Chazdon & Pearcy, 1991;Leakey et al., 2003). As a consequence, the magnitude of the change in the light environment may be sufficiently great that light acts as a strong selection pressure on photosynthetic capacity in these small trees, even when exposed to long-term drought conditions.
The greater plasticity of J max compared to V cmax and resultant increase in the J max / V cmax ratio on the TFE plot (Fig. 2) is highly indicative of a response to light, as the electron transfer reactions are directly involved in capturing light energy (Farquhar et al., 1980;Sharkey et al., 2007). A similar shift in the ratio in tropical montane cloud forests compared to lowland tropical forests has been attributed to lower total daily light availability caused by cloud cover, but with periods of intermittent intense light (van de Weg et al., 2012). Following the mortality of canopy trees, the understory may experience similar sporadic periods of intense light, as a consequence of increased sunfleck occurrence and duration. A larger upregulation of J max compared to V cmax may allow these trees to maximise light capture during these prolonged sunflecks, with carboxylation reactions occurring subsequently (Pearcy, 1990). The division of the light dependent and light independent reactions may allow small understory trees to increase overall carbon assimilation whilst avoiding the additional maintenance costs of elevated V cmax .
We show small trees can increase J max , but not V cmax , without additional leaf nitrogen or phosphorus ( Fig. 1), most likely by increasing nutrient use efficiency (Fig. 3). This may occur via a potential reallocation of nitrogen and phosphorus to optimise photosynthetic capacity (Hasper et al., 2017;Mo et al., 2019). The carboxylation reactions have greater nutrient demand for enzymes, such as RuBisCO, compared to those in the electron transport chain (Evans, 1989;Xu et al., 2012;Raven, 2013). This may allow J max to be more plastic and responsive to changes in light availability than V cmax , without increasing leaf nutrient concentrations. Overall, the reduction in light limitation of photosynthesis in understory trees we observe here could facilitate increased wet-season growth rates as observed previously at this experiment in larger understory trees (Metcalfe et al., 2010;Rowland et al., 2015a).
Consequently, the ability of small understory trees to respond to light whilst under soil moisture deficit may allow them to regenerate and recover biomass faster than currently predicted by dynamic vegetation models (Fisher et al., 2007).
The ability of small understory trees to respond to increased light availability in the face of drought, increased leaf temperatures and elevated VPD may have various explanations. Firstly, it is possible that despite the reduced soil moisture concentrations (SI Fig. 1), our small trees are less stressed than the adult trees by the drought treatment. This could be a consequence of reduced competition for water following the mortality of large trees (Rowland et al., 2015a), or a result of physiological adjustments to their hydraulic architecture or stomatal conductance. Small trees may be able to avoid drought stress by having stronger stomatal regulation or greater resistance to xylem embolism (Anderegg et al., 2018;Bittencourt et al., 2020). Reduced physiological and architectural constraints also allow small trees to explore more trait combinations than canopy trees that may allow them to be more effective at resisting drought (Reed et al., 2011;Damián et al., 2017). Alternatively, small trees may be able to avoid drought stress by reducing non-maintenance related metabolic activity, such as growth, in the dry season when drought conditions are most pronounced, and maximising growth during the wet season when the soil moisture deficit is reduced. This hypothesis is consistent with previous observations of high wet season and lower dry season diameter growth rates in the trees of smaller size classes on the TFE relative to the control (Metcalfe et al., 2010;Rowland et al., 2015a). Whichever mechanism small trees use, our results suggest that these trees are likely to be able to minimise the impact of the drought, relative to larger trees, in order to facilitate the upregulation of photosynthetic capacity and a release from extreme light limitation.
In addition to upregulation of maximum photosynthetic capacity, we found R leaf to increase in small trees following long-term drought by 32% (Fig. 1). These elevated respiratory rates likely reflect the additional maintenance costs of higher photosynthetic capacity under elevated light conditions (Atkin et al., 2015). However, we do not find a direct relationship between R leaf and V cmax or J max in these trees (SI Table 3), suggesting elevated R leaf is not simply a consequence of increased photosynthetic capacity.
Instead, elevated R leaf may reflect a response to increased stress under reduced water availability, potentially representing a means of drought resistance through increasing carbon metabolism for maintenance respiration (Rowland et al., 2018). Under water limitation, particularly when combined with increased irradiance, leaves can accumulate harmful reactive oxygen species (ROS) that cause cell damage (Wang & Vanlerberghe, 2013). An upregulation of respiration can provide a mechanism for leaves to purge these harmful ROS and avoid damage to the photosynthetic metabolic machinery (Atkin & Macherel, 2009). These negative effects of ROS are likely to be strongest during the dry season when water availability is particularly low. Here, small understory trees may have increased their R leaf in the dry season to avoid drought stress and facilitate responses to increased light availability. It must also be acknowledged that our results may be influenced by changes in canopy temperature profiles between the plots, with a smaller expected vertical gradient in temperature in the more open canopy of the TFE plot, but measuring this was beyond the scope of this study. We do find large intra-generic variations in R leaf and low explanatory power of the TFE treatment in our statistical models (SI Table 2), suggesting this trait is not simply influenced by long-term drought, but also by a range of other unaccounted for factors that are likely species-specific.
An increase in light availability in the understory of the droughted forest is also likely to be the main explanation for the 15% increase in LMA in small trees on the TFE we observe here. LMA is most strongly controlled by light (Poorter et al., 2009), as leaves invest more in proteins that catalyse photosynthesis and produce more carbohydrates under elevated light conditions (Niinemets et al., 1998;Poorter et al., 2009). Higher concentrations of these high molecular weight metabolites will increase the density of cells and hence LMA. However, it should be noted that elevated LMA may also occur in response to reduced water availability, as a water conservation strategy (Wright et al., 2004;Poorter et al., 2009). Unlike photosynthetic capacity and R leaf , LMA has been shown to increase in large trees at the same experiment via increased investment in spongy mesophyll for water storage (Rowland et al., 2015b;Binks et al., 2016;Rowland et al. under review), indicating increased LMA in small trees may also be directly associated to lower water availability in the TFE.
We show small trees were more responsive to prolonged drought conditions than large canopy trees, suggesting small trees experience different selection pressures or possess a greater ability to respond to ecosystem level changes in the physical environment. Maximum photosynthetic capacity of small trees increased in response to the TFE treatment -the exact opposite to the response of large trees to the TFE once differences in canopy position have been accounted for in our models (Table 1). In fact, only canopy shading prevents small trees from matching the capacity of large trees. Under prolonged drought, small trees experience changes to both their light and water availability, whereas larger canopy trees will predominantly only experience substantial shifts in water availability. The different responses to prolonged drought suggest small trees are responding most strongly to changes in their light environment, whilst large trees are responding to reductions in soil moisture availability.
These shifts may reflect changes in the most limiting resource from the understory to the top canopy, with understory trees strongly light limited and canopy trees strongly water or nutrient limited (Chazdon & Fetcher, 1984;Sterck & Schieving, 2011

Conclusion
This article is protected by copyright. All rights reserved.
In this study, we highlight the important role of forest structural changes and light in determining physiological responses to long-term drought. Small trees (<10 cm DBH) relative to large trees (>20 cm DBH) display sufficient phenotypic plasticity in leaf morphology and carbon metabolism traits to allow them to respond to increases in light availability despite long-term drought. The ability of small trees to increase their photosynthetic capacity may facilitate increased growth and consequently partial recovery of forest aboveground biomass following earlier drought-induced mortality of large trees (Rowland et al., 2015a). This capacity of small trees to show positive responses to ecosystemlevel changes in water and light availability could ultimately allow a more resilient forest to establish and potentially moderate the negative impacts of climate change on the forest ecosystem.
This article is protected by copyright. All rights reserved.  represents a 1:1 ratio. A significant difference in the elevation (p < 0.01) and a shift in the data (p < 0.001) between the treatments were found, but no significant difference between the slopes was found (p = 0.757). changed between the control and TFE treatment for nitrogen (a, c, e) and phosphorus (b, d, f). NUE was calculated by dividing gas exchange parameters on a mass basis by leaf nutrient concentrations.
Asterisks represent significant differences from linear mixed effects models between the two treatments (ns: non-significant; * p < 0.05; ** p < 0.01; *** p < 0.001). Boxes represent the interquartile range with a horizontal line for the median and the whiskers represent 1.5*interquartile range or the maximum and minimum point. Dots represent points outside the extent of 1.5*interquartile range.  Control (C) and TFE (T) treatment for large (>20 cm DBH) and small (1-10 cm DBH) trees across the five most common genera: Eschweilera, Inga, Licania, Protium, Swartzia. All genera with n ≥ 2 for each of the four categories are presented. White dots represent the median, thick grey lines represent the interquartile range and thin grey lines represent the mean ± 1.5 standard errors. Tables   Table 1: Parameter estimates for the minimal adequate models explaining J max , V cmax , R leaf , leaf mass per area (LMA) and leaf thickness for large (>20 cm DBH; n = 61) and small (1-10 cm DBH; n = 48) trees. The intercept represents the trait value when the fixed effects are zero: large control trees that are not shaded. Canopy position represents the difference when a tree is shaded, plot represents the difference between the TFE treatment compared to the control and tree size represents the difference for small trees compared to large trees. Coefficient estimates ± the standard error are presented for each fixed effect and refer to the differences between the factor levels indicated in brackets for each column and the overall intercept. The genus intercept is given whenever the genus included as a random intercept effect improved the model fit. Species nested within genus was not significant for any traits, so is not included here. Random effect variance for genus ± standard deviation is presented. Total (conditional) R 2 represents the total variation explained by the model and is partitioned into the variation explained by the fixed effects (marginal R 2 ) and fixed plus random-effects (conditional R 2 ). Asterisks represent the significance level of each variable: ns P > 0.05; * P < 0.05; ** P < 0.01; *** P < 0.001. Dashes represent variables that were not retained in the minimal adequate model but were tested in the initial fully saturated model. See Methods and SI Table 3 for more details about model selection.
This article is protected by copyright. All rights reserved.