Physiological responses to light explain competition and facilitation in a tree diversity experiment

Ecologists often invoke interspecific facilitation to help explain positive biodiversity-ecosystem function relationships in plant communities, but seldom test how it occurs. One mechanism through which one species may facilitate another is by ameliorating abiotic stress. Physiological experiments show that a chronic excess of light can cause stress that depresses carbon assimilation. If shading by a plant’s neighbors reduces light stress enough, it may facilitate that plant’s growth. If light is instead most often a limiting factor for photosynthesis, shading may have an adverse, competitive effect. In a temperate tree diversity experiment, we measured stem growth rates and photosynthetic physiology in broadleaf trees across a gradient of light availability imposed by their neighbors. At the extremes, trees experienced nearly full sun (monoculture), or were shaded by nearby fast-growing conifers (shaded biculture). Most species had slower growth rates with larger neighbors, implying a net competitive effect. On the other hand, the two most shade-tolerant species (Tilia americana and Acer negundo) and the most shade-intolerant one (Betula papyrifera) had faster stem growth rates with larger neighbors. The two most shade-tolerant species had large increases in photoinhibition (reduced dark-acclimated Fv/Fm) across the gradient of increasing light availability, which suggests they are more vulnerable to chronic light stress. While most species had lower carbon assimilation rates in the shaded biculture treatment, T. americana had rates up to 25% higher. T. americana also dropped its leaves 3-4 weeks earlier in monocultures, curtailing its growing season. We conclude that although large neighbors can cause light limitation in shade-intolerant species, they can also increase growth through abiotic stress amelioration in shade-tolerant species. Finally, in shade-intolerant B. papyrifera, we find a pattern of stem elongation in trees with larger neighbors, which may suggest that a shade avoidance response accounts for the apparent positive trend in stem volume. Synthesis: Both positive and negative species interactions in our experiment can be explained in large part by the photosynthetic responses of trees to the light environment created by their neighbors. We show that photosynthetic physiology can help explain the species interactions that underlie biodiversity-ecosystem function relationships. The insights that ecologists gain by searching for such physiological mechanisms may help us forecast species interactions under environmental change.


Introduction
Methods 134

Experimental design 135
The FAB experiment was planted in 2013 at Cedar Creek Ecosystem Science Reserve in central 136 Minnesota, USA. Cedar Creek has a continental climate with cold winters and warm summers; the mean 137 annual temperature is about 7 °C and the mean annual precipitation is about 660 mm. The site lies on the 138 Anoka Sand Plain, where upland regions tend to have well-drained, sandy soils (Johnston et al. 1996). 139 FAB comprises 142 4 × 4 m plots, each planted with 64 trees in a 0.5 × 0.5 m grid. The 140 experiment has three blocks, each with 49 plots arranged in a 7 × 7 square. Neighboring plots within a 141 block share a boundary. Each plot is planted with one-or two-year-old seedlings of 1, 2, 5, or 12 species 142 from a pool of 12 species total, including four evergreen conifers with needle-like leaves and eight 143 winter-deciduous angiosperms with broad leaves (Table 1). Except for five-species plots, each distinct 144 species composition is replicated in each block. All species in a plot have nearly equal frequency and are 145 placed at random. Grossman et al. (2017) describes the experimental design in greater detail. The mean 146 relative growth rate in monoculture during the first three years varied dramatically among species-it was 147 highest in conifers and Betula papyrifera, and lowest in Acer negundo (Table 1). 148 149

Tree growth 150
Within each species, we sought to determine how an individual's stem growth rates were 151 influenced by the size of its neighbors. We assumed that neighbors' size affects the focal individual's 152 growth rates, but not vice versa. This assumption of exogeneity was warranted because nearly all of the 153 cross-species variation in growth rates is explained by species identity, so the size of a focal individual's 154 neighbors is explained much more by their identity than by their interactions with the focal individual. 155 We surveyed tree growth in late fall of each year. For all living trees, we measured diameter and 156 height to the tallest leader. We measured basal diameter (5 cm from the ground) for trees less than 1.37 m 157 tall and diameter at breast height for trees more than 1.37 m. For the first year that a tree crossed the 1.37 158 m threshold, we measured both diameters. To predict a tree's basal diameter when it was not measured, 159 we used species-specific linear relationships (R 2 = 0.634-0.784) predicting basal diameter additively from 160 height in the same year and the basal diameter in 2016, which was the last year that the basal diameter of 161 all trees was measured. For the conifer Pinus banksiana, basal diameter was measured on few enough 162 trees in 2018 that we could not apply this approach; we instead estimated stem volume under the 163 assumption that relative growth rate (RGR) was the same from 2017 to 2018 as from 2016 to 2017. 164 Because we did not consider Pinus banksiana growth as a response variable, this approach avoided 165 circularity and allowed us to get reasonable estimates of size in 2018. 166 In all other cases, we estimated woody volume as = 2 ℎ, where h is height and r is the basal 167 radius. This equation assumes each tree's woody volume can be approximated as a cylinder, an 168 assumption that has been used and justified in other tree diversity studies in the absence of system- should be interpreted conservatively. We also tested whether any species showed non-monotonic 184 responses to neighbor size using the two-lines test (Simonsohn 2018), without random effects. Some species showed a positive response in stem growth to neighbor size. Besides facilitation, 186 such a response could also be caused by shade avoidance, which may cause plants to increase shoot 187 biomass at the expense of root biomass in order to compete for light (Shipley & Meziane 2002). We 188 aimed to discern whether such a shade avoidance syndrome could contribute to the positive stem growth 189 response. Lacking root biomass sorted by species, we could not test this idea directly, but we could test 190 another symptom of a typical shade avoidance syndrome: stem elongation, which prioritizes vertical 191 growth over lateral growth (Henry & Aarssen 1999). We considered height and basal diameter in 2016,192 the last year in which basal diameter was measured directly on all trees. For all species that showed a 193 positive stem growth response, these two variables were linearly correlated. We used mixed-effects linear 194 models with plot as a random intercept to predict diameter from height in each species, then tested 195 whether the residuals-the observed diameters' deviations from the predictions for a tree of equal 196 height-were associated with average neighbor size. If trees with larger neighbors show a shade 197 avoidance syndrome, we would expect them to have smaller diameters for a given height. We obtained 198 similar conclusions when considering the log-transformed diameter-to-height ratio as the response 199 variable rather than the residuals of the linear diameter-height relationship. ('monoculture'); (2) bicultures comprising one broadleaf and one conifer species ('shaded biculture'); and 207 (3) twelve-species plots (see Fig. S1 for sample images). For each species in each treatment, we measured 208 physiological parameters on six individuals-two in each of three plots. Neither A. negundo nor Q. 209 macrocarpa was planted with a conifer in any biculture. We used B. papyrifera as a shaded biculture 210 partner for Q. macrocarpa because it is a fast-growing species and creates shade. We omitted the shaded biculture treatment for A. negundo. When possible, we chose the same trees within a plot for each kind of 212 physiological measurement so that we could compare these aspects of physiology within individuals. 213 214 Photosynthetic light-response curves 215 We measured photosynthetic light-response curves from the four focal species during July 2018 216 using an LI-6400 gas exchange system (LI-COR BioSciences, Lincoln, NE, USA) with a 6400-40 Leaf 217 Chamber Fluorometer head. From each tree (n = 72 total), we selected a fully expanded upper leaf with 218 no visible sign of disease or herbivory. We noted the angle of each leaf relative to horizontal (to the 219 nearest 15˚) before beginning measurements. Each curve had nine steps in descending order of brightness: 220 2000, 1500, 1000, 500, 200, 100, 50, 20, and 0 µmol m -2 s -1 . We also measured the relative electron 221 transport rate (ETR) using chlorophyll fluorescence at each light level. We maintained favorable 222 conditions inside the chamber during each measurement. Following each curve, we removed the leaf to 223 measure leaf mass per area (LMA) using a balance and a flatbed scanner. Further details about light-224 response curve procedures and ETR calculations can be found in the Supplemental Materials (Appendix 225 1). Finally, we estimated parameters like the light-saturated carbon assimilation rate (Asat) by fitting a 226 non-rectangular hyperbolic model (Johnson & Thornley 1984) to each light-response curve using R code 227 written by Nick Tomeo, available at: https://github.com/Tomeopaste/AQ_curves. 228 We aimed to determine how realized carbon assimilation rates vary with the photosynthetic 229 photon flux density (PPFD) at the top of the canopy, a measure of light availability in the absence of 230 shading. While all trees received the same range of light intensities inside the instrument chamber, some 231 trees received much less light than others in situ because their neighbors shaded them more. At any 232 moment, trees that are more shaded are farther to the left on their light-response curve than trees that are 233 less. For example, consider a tree crown whose topmost leaves only receive 50% of the light reaching the 234 top of the canopy. For this tree crown, the realized assimilation rate we estimate at a top of the canopy 235 PPFD of 2000 µmol photons m -2 s -1 would be the assimilation rate at a leaf-level PPFD of 1000 µmol m -2 236 s -1 . To visualize the impacts of shading by neighbors, we followed the approach of Howell et al. (2002): 237 we estimated how much light each leaf receives as a constant fraction of light at the top of the canopy, 238 and rescaled each leaf's fitted light-response curve along the x-axis by this fraction's reciprocal. We 239 estimated the constant rescaling factor for each individual as RLA × cos(θ), where RLA is relative light 240 availability (see Light availability) and θ is the leaf angle from horizontal. (Taking the cosine of θ 241 approximately corrects for the reduction in horizontally projected leaf area in steeply inclined leaves 242 when the sun is directly overhead, as simplified from Ehleringer & Werk [1986].) We used this procedure 243 on each fitted curve in order to compare across treatments in a way that accounts for the varying fractions 244 of light that individuals receive due to differences in relative height and vertical light transmission. 245 Finally, we used the rescaled light-response curves to estimate carbon assimilation by top leaves 246 throughout July. We used an hourly averaged time series of solar radiation from Carlos Avery Wildlife 247 Management Area in Columbus, MN, 13.6 km away from the study site. Following Udo and Aro (1999), 248 we assumed a conversion rate of 2.08 μmol photons m -2 s -1 of PAR per Watt m -2 of solar radiation. By 249 using these data as inputs to the rescaled light-response curves, we estimated carbon assimilation over the 250 month of July. This procedure assumes that the photosynthetic response to light remains constant and 251 unaffected by factors like leaf temperature and stomatal closure. While this assumption is never quite 252 true, we consider it a useful way of estimating the consequences of realistic fluctuations in light. 253 We mainly express assimilation per unit of leaf dry mass because we aim to determine the plants' 254 return on the carbon invested in leaf construction. As expected, leaves growing in low light tended to on constructing a leaf with high LMA, which means that the leaf has to assimilate more carbon to recoup 257 its construction cost. Expressing data on a mass basis accounts for these large differences in construction 258 cost, although we also discuss area-based assimilation rates when they provide additional physiological 259 insight. We also express stomatal conductance (gs) and electron transport rate (ETR) on a mass basis to 260 compare them with mass-based assimilation rates. 261 262

Instantaneous chlorophyll fluorescence and spectral reflectance 264
In all eight broadleaf species, we measured chlorophyll fluorescence parameters using an FMS2 265 pulse-modulated fluorometer (Hansatech Instruments Ltd., Norfolk, UK) over two days in late July (n = 266 138 total). We measured dark-and light-acclimated parameters at the same spot on the same leaves. We 267 attached opaque clips to leaves in the evening before measuring dark-acclimated Fv/Fm within two hours 268 after sunrise. This parameter describes the maximum quantum yield of PSII and is a general index of 269 photoinhibition (Murchie & Lawson 2013). We then took light-acclimated measurements between 12:00 270 and 14:00 each day. The protocol involved the following steps: actinic light at 1000 µmol m -2 s -1 for 15 271 seconds, a saturating pulse, and two seconds of far-red light. We exposed all leaves to the same actinic 272 light because we aimed to assess photoprotective capacity under comparable light conditions, even though 273 each tree had a different light environment. 274 From these data, we estimated qN, a parameter that indicates how much a plant relies on On separate top leaves of the same trees, we measured reflectance spectra (350-2500 nm) using a 284 PSR+ 3500 field spectroradiometer (Spectral Evolution, Lawrence, MA, USA). We used these spectra to 285 calculate the photochemical reflectance index (PRI), calculated as PRI = (R531 -R570) / (R531 + R570), 286 where Rn is the reflectance at a wavelength of n nm. PRI shows a negative correlation with carotenoid : We monitored the timing of leaf abscission for T. americana in monoculture and shaded biculture 296 plots to assess whether shade could delay senescence and abscission, extending the period for carbon gain 297 five axillary buds lacked a leaf. From each of these trees, we marked a single leaf and returned every 1-2 300 weeks to monitor its senescence. We assessed each leaf on a binary basis (senesced or not) based on 301 whether it had at least 50% remaining green leaf area. We chose this criterion as a simple proxy for the  We also performed a one-time measurement of 60 (30 per treatment) A. rubrum plants on October 307 1, using the same protocol we used to do our initial early August survey of T. americana. We aimed for 308 this survey to help test whether our results hold across species. 309 310 Water potential 311 Water deficits-even moderate ones-can arrest tissue growth (Tardieu et al. 2014) and reduce 312 carbon gain by causing stomata to close (Brodribb et al. 2003). To see how our treatments affected water 313 status, we measured leaf water potential using a Scholander pressure bomb. We measured pre-dawn water 314 potential (ΨPD) in all eight broadleaf species (n = 96) and midday water potential (ΨMD; between 12:00 315 and 14:00) in only the four focal species (n = 48). We included monoculture and shaded biculture 316 treatments, except in A. negundo, where we used the twelve-species treatment in place of the absent 317 shaded biculture treatment. We removed leaves and immediately placed them in impermeable plastic bags 318 with wet paper towels. We carried leaves in coolers to be measured indoors within 90 minutes. 319 Because we only intended to test whether there was a general tendency across species for water 320 potential to vary with light, we included species as a random intercept and light availability as a fixed 321 effect in mixed-effect regression models for both ΨPD and ΨMD. We calculated Rm 2 and p-values as in 322 models for growth rate. 323 324

Leaf angles 325
Leaf angle is a key control on light exposure because more vertical leaves intercept less light 326 when the sun is directly overhead, and may allow light to penetrate deeper into the canopy (Posada et al.  Because angles of leaves within individuals may be non-independent, we tested whether 335 treatments differ using a mixed-effects model with treatment as a fixed effect and individual tree as a 336 random intercept. Deviations from horizontal in either direction can reduce light interception, so we used 337 the absolute value of leaf angle as the dependent variable. We again calculated p-values using 338 Satterthwaite's degrees of freedom method. 339

Light availability 342
On a cloudy day with diffuse light conditions, we measured the available light above each tree 343 selected for this study (n = 138) using an AccuPAR LP-80 ceptometer (METER Group, Pullman, WA, 344 USA). We took the mean of two to four PAR measurements in the open and the mean of two 345 measurements directly above the topmost leaf of each tree. By calculating a ratio of these values, we 346 could calculate the percent of light transmitted to the top of each tree. We called this value relative light 347 availability (RLA). This value usually correlates well with the percentage of light a tree can access over 348 much longer time-scales (Parent & Messier 1996). For nine days during early September, we set out four Thermochron iButtons (model DS1921G; 359 Maxim Integrated Products, Sunnyvale, CA, USA) in shaded plots and two in twelve-species plots, 360 setting them to log air temperature hourly. Each iButton was suspended on mesh inside a 1 m-tall PVC 361 pipe that was capped with an elbow to shield against solar irradiance. We compared these data to air 362 temperature measured hourly under open, sunny conditions at the Cedar Creek LTER weather station 363 about 0.77 km away. We assumed that data from this station would be representative of air temperature in 364 non-shaded areas. We ignored nighttime and early morning (20:00-08:00 h) readings because they 365 seemed to be influenced by water condensation and evaporation.

Tree growth 368
Species varied in their relationship between neighbor stem volume and focal individual relative 369 growth rate (RGR; mixed-effects ANCOVA, neighbor volume × species interaction; p < 10 -15 ; F11,7279 = 370 12.440). For individual species, relationships between RGR and neighbor volume were noisy but often 371 highly significant (Fig. 2). In most species, individuals with larger neighbors had lower RGR, including in (2) removed trees that died from analyses. The rate of mortality was low between 2016 and 2018 (~7.5%), 387 and accounting for mortality makes little qualitative difference when considering absolute growth rate as 388 the response variable ( Fig. S3; Supplemental Materials, Appendix 2). 389 Although the individual-level effects are noisy, they reveal strong tendencies in average stem 390 growth across the full range of neighbor size. Imagine two T. americana trees that begin at the same size: 391 One in monoculture, where the average neighbor in 2018 is 370 cm 3 , and one growing in a shaded biculture with Pinus strobus, where the average neighbor is 2120 cm 3 . The individual-level regression 393 (with log-transformation) predicts that after two years, the average T. americana in the shaded biculture 394 would be 26.7% larger because of the difference in mean RGR. Or consider Q. rubra, which had the 395 strongest negative response to neighbor size: We predict that an average tree in the shaded biculture 396 (mean neighbor size 1988 cm 3 ) would be 40.2% smaller than in monoculture (mean neighbor size 160 397 cm 3 ). These figures assume that individuals across treatments start at the same size at the initial point for 398 RGR estimates, so they may underrepresent the true differences in size that have developed. The absolute 399 growth rate gives some insight into these differences.  We can also aggregate these observations to the plot scale-for example, by considering the RGR 406 of summed stem volume of each species in each plot from 2016 to 2018 (Fig. 2). (This approach is 407 distinct from an average of individual RGRs, which assigns equal weight to small and large individuals.) 408 Although aggregation leaves less statistical power to detect relationships, it also reduces noise, such that 409 the relationships we find explain much more of the variation in growth across plots. As average neighbor 410 size increases, RGR declines in Q. alba (R 2 = 0.297, p < 0.005, t(28) = -3.642) and Q. macrocarpa Finally, for the three species that showed positive individual-level stem growth responses to 413 neighbor size (T. americana, A. negundo, and B. papyrifera), we performed a follow-up analysis to test 414 whether they showed a shade avoidance response. We found that trees with larger neighbors had smaller 415 diameters than expected based on their height in B. papyrifera (R 2 = 0.009, p = 0.008, t(795) = -2.645), 416 but not T. americana and A. negundo (p > 0.05). 417

Photosynthetic physiology 419
Photosynthetic light-response curves 420 As a function of the chamber light level, mass-based assimilation rates were always higher in 421 shaded biculture and twelve-species treatments than in monoculture in three out of the four focal species 422 (Q. ellipsoidalis, A. rubrum, and T. americana; Fig. 3, left). In the early-successional species B. 423 papyrifera, mass-based assimilation rates in shaded biculture were lower than in the other two treatments. 424 Area-based assimilation rates showed similar trends in T. americana, but varied less among treatments in 425 B. papyrifera and Q. ellipsoidalis-and in A. rubrum, they were lower in shaded biculture than the other 426 treatments (Fig. S4) as a result of large differences among treatments in LMA (Fig. S2). 427 In all broadleaf species except B. papyrifera and T. americana, LMA was higher in the 428 monoculture than the other two treatments (Fig. S2). Across the four focal species, the light-saturated 429 assimilation rate Asat on an area-basis was positively correlated with LMA (R 2 = 0.431; p < 10 -9 ; t(69) = The rate of photosynthesis is often limited either by ETR or by the RuBP carboxylation rate; the 435 latter may in turn be limited by stomatal diffusion of CO2 (Farquhar et al. 1980). In our data, both ETR 436 and stomatal conductance (gs) increased with light availability (Fig. S5). (ETR tended to decline at very 437 high light levels, especially in shaded treatments, perhaps because of acute photoinhibition.) Within each 438 species, the rank-order of treatments in carbon assimilation rates, ETR, and gs were broadly congruent. 439 One exception is that T. americana in shaded biculture had much higher gs than in twelve-species or 440 monoculture plots, despite having slightly lower assimilation rates than in twelve-species plots. 441 Compared to the unscaled chamber light-response curves, the picture that emerged from rescaled 442 light-response curves is more complex (Fig. 3, right). In B. papyrifera, the assimilation rate was still 443 lowest across light levels in the shaded biculture treatment. In late-successional T. americana, the assimilation rate was higher in shaded biculture and twelve-species treatments by up to 25% compared to 445 monoculture across most light levels. In the other two species, the mass-based assimilation rate was 446 highest in monoculture at low light, but twelve-species and (in Q. ellipsoidalis) shaded biculture plots 447 intersected and surpassed monocultures when enough light was available. A. rubrum and Q. ellipsoidalis 448 grew less and had lower light availability in shaded bicultures than T. americana (Fig. S2), so their mass-449 based assimilation rates dropped considerably more in the rescaled light-response curves. The area-based 450 assimilation rate in these two species was lower in shaded bicultures (and in Q. ellipsoidalis, twelve-451 species plots) than in monocultures across the full domain of light availability (Fig. S4), but otherwise, 452 mass-based and area-based rates showed similar patterns across treatments within species. 453 Using a time series of solar radiation, we estimated total mass-based assimilation rates in July constant across light environments (Fig. 4). Consequently, we found that shade-tolerant species also had 469 smaller rises in qN with RLA ( Fig. 5; R 2 = 0.619; p = 0.022; t(5) = -3.276). PRI declined as RLA 470 increased, with statistically indistinguishable slopes among species (Fig. 4). 471 To see why shade-tolerant species showed such large increases in photoinhibition with light, 472 consider ϕPSII, ϕNPQ, and ϕNO-the quantum yields of PSII photochemistry, non-photochemical quenching, 473 and non-regulated dissipation. In most species, ϕNO decreased with light in the growth environment ( In T. americana, leaf abscission in monoculture began in early August, and more than half of all 497 leaves had senesced by September 3 (Fig. 6). In shaded bicultures, more than 90% of leaves remained by 498 September 3, and no survey found greater than 50% senescence until October 1. We assigned the 499 senescence date of each leaf as the date of the first survey by which it had senesced; using this response 500 variable, leaves in monoculture senesced 22 days earlier than those in shaded bicultures (t-test; p < 10 -11 ; We removed two data points, one in each treatment, with ΨPD more negative than -0.8 MPa on the 509 grounds that they may have been subject to measurement error. Across treatments, the mean ΨPD and ΨMD Among each of the three species whose leaf angles we measured, leaves were nearer to horizontal 518 in the shaded biculture treatment (Fig. 7). The size of this fixed effect varied among species: 11.0˚ in Q.

Competition and shade-intolerant species 545
Individuals of all four Quercus species and A. rubrum had lower stem growth when surrounded 546 by larger neighbors (Fig. 2). This growth pattern implies that neighborhood interactions were dominated by competition, potentially both above-and belowground. (The fact that soil moisture was lowest in 548 shaded biculture plots suggests an important role for belowground competition.) 549 The physiological data can help us make sense of these growth patterns. We measured light-550 response curves of two species in this group: A. rubrum and Q. ellipsoidalis. In the untransformed 551 chamber light-response curves, both had lower mass-based assimilation rates in monoculture than in the 552 other two treatments (Fig. 3, left). But on an area basis, A. rubrum had its lowest assimilation rates in the 553 shaded biculture treatments, and Q. ellipsoidalis was similar across all three treatments. The discrepancy 554 between area-and mass-based results is explained by variation in LMA across treatments (Fig. S2). area-based Asat, which may imply that much of the variation in LMA could be driven by the mass of non-560 photosynthetic structural tissue. But for A. rubrum in particular, there was a strong negative correlation 561 between LMA and dark-acclimated Fv/Fm, two variables jointly affected by light availability (Fig. S2; Fig.  562 4). As a result, a potential positive relationship between LMA and Asat in A. rubrum could have been 563 masked by photoinhibition. We also found evidence that photoinhibition lowers Asat in Q. ellipsoidalis 564 and T. americana. Since photoinhibition occurs through damage or downregulation of PSII, it is usually 565 thought to depress photosynthesis by lowering ETR. But both ETR and stomatal conductance mirrored 566 assimilation across treatments in chamber light-response curves (Fig. S5), so both may have some role in 567 explaining why most species had lower assimilation in monoculture at a given chamber light level. 568 Rescaling the light-response curves allowed us to visualize the costs of shading-the fact that 569 some trees receive less light and face more frequent photosynthetic light limitation than others. Compared 570 to monoculture, A. rubrum and Q. ellipsoidalis in shaded biculture had lower mass-based carbon 571 assimilation under low light at the top of the canopy, but higher mass-based carbon assimilation under high light (Fig. 3, right). For these two species, shading often appears severe enough to cause light 573 limitation of growth and photosynthesis, especially under otherwise low light. 574 Given that light availability fluctuates-diurnally, seasonally, and with cloud cover-the 575 treatment that performs best may also vary from moment to moment. We used a time-series of solar 576 radiation to estimate total carbon assimilation throughout July. We found that A. rubrum and (only on a 577 mass basis) B. papyrifera had lower carbon assimilation in shaded biculture than in the other treatments 578 (Fig. S6). PPFD was under 500 µmol m -2 s -1 more than 60% of the time during the month; it is under such 579 dim conditions that trees in monoculture have the greatest photosynthetic advantage over those in shaded 580 biculture, no matter what benefits shading may confer at sunnier times ( Fig. 3; Fig. S4). Trees in twelve-581 species plots had similar assimilation rates to those in monoculture in all species. The intermediate light 582 environment in twelve-species plots may have allowed trees to avoid photoinhibition without becoming 583 frequently light-limited (Fig. S2), although we caution that we only measured top leaves, so we cannot 584 directly compare whole-plant carbon gain. In T. americana, A. negundo, and B. papyrifera, having larger neighbors increases stem growth 597 ( Fig. 2; Fig. S3). The former two are the most shade-tolerant broadleaf species in the experiment, while the last is the most shade-intolerant. Moreover, these three species showed divergent photosynthetic 599 responses to the light environment, which suggests that different mechanisms may explain the positive 600 stem growth response in B. papyrifera than in the other two species. 601 We first discuss the positive trend in T. americana and A. negundo before returning to B. 602 papyrifera in the following section. In these two species, we interpret this trend mainly as a result of 603 facilitation, reinforcing prior results that showed that shade-tolerant species may often have facilitative 604 responses to being shaded (Montgomery et al. 2010). In particular, these two species' growth trends are 605 best fit by a model where neighbor size is log-transformed (Fig. 2, Appendix 2). Increasing neighbor size 606 thus has positive but diminishing marginal benefits. This outcome may result if neighbor size has a non-607 linear influence on light availability, or if deep shade has escalating costs that begin to offset its benefits. 608 Here, the physiological data may help us explain the growth patterns. T. americana was among 609 the four focal species, and the only one of the four that had the lowest carbon assimilation rates in the 610 monoculture treatment across most light levels in both chamber and rescaled light-response curves (Fig.  611 3). The steep decline in dark-acclimated Fv/Fm with increasing light (Fig. 4) suggests that the low Asat in 612 monoculture results from photoinhibition (Fig. 5). This pattern aligns with T. americana's role as a late-613 successional dominant species throughout much of its range (Braun 1950). Our finding that carbon gain 614 in T. americana can increase under shade is reinforced by Carter and Cavaleri (2018), who found that its 615 assimilation rates may increase down vertical gradients from the upper canopy to the sub-canopy. 616 Nevertheless, we found no significant differences among treatments in estimated assimilation across the 617 whole month of July (Fig. S6), mainly because under low light, monocultures still had higher assimilation 618 rates in the rescaled light-response curves. The amelioration of photoinhibition is very likely to contribute 619 to the positive effect of neighbor size on growth in T. americana, but it cannot fully explain it. 620 We next discuss two mechanisms of facilitation that could also contribute to positive growth 621 responses to neighbor size, but which analyses based on light-response curves do not account for: 622 Delayed senescence and microclimatic effects. Both may contribute to potential positive effects of 623 shading by neighbors in the species that show such responses. the shaded biculture (Fig. 6). We also found that A. rubrum in monoculture had earlier leaf senescence. 626 Given that leaves remained photosynthetically active until shortly before abscission (as suggested by Although we emphasize the role of light stress, other microclimatic factors could contribute to 636 variation in assimilation rates among treatments. For example, an alternate perspective on photoinhibition 637 is that it mainly results not from photodamage but from carbon sink limitation, which leads plants to 638 downregulate light capture and electron transport as a photoprotective response (Adams III et al. 2013). 639 From this perspective, photoinhibition seldom causes carbon limitation or reduces growth rates in nature; 640 rather, slow growth rates cause photoinhibition. A potential cause of sink limitation that could affect 641 monocultures of T. americana and A. negundo more than shaded bicultures is water limitation, which 642 could physically constrain the expansion of tissues (Tardieu et al. 2014). We found that leaf ΨPD was 643 slightly more negative in shaded bicultures, while ΨMD was more negative in monocultures (Fig. S8). But from this perspective, it is hard to explain why this sink limitation would particularly matter 649 in T. americana and B. papyrifera among the four focal species; for example, A. rubrum also has more negative ΨMD in monoculture (Fig. S8). Moreover, trends in ϕNO suggest that T. americana and A. 651 negundo in high light are especially vulnerable to photodamage (Fig. 4). Photoinhibition due to source-652 sink imbalance is usually accompanied by compensatory upregulation in protective non-photochemical 653 quenching (Adams III et al. 2013), but these two species fail to increase qN under high light (Fig. 4). 654 These findings reinforce the idea that damage, not just downregulation, contributes to the steep declines 655 in dark-acclimated Fv/Fm these species show under increasing light. Sink limitation may contribute to the 656 photoinhibition and the growth patterns we observe in these species, but we believe it is not the main 657

explanation. 658
The microclimate could also alter photosynthesis in ways our light-response measurements did 659 not capture because we controlled the chamber microclimate. Our water potential measurements show 660 that water stress was not very severe in any treatment, but even modest differences could change the 661 treatments. Because it tended to overtop even its largest neighbors, the top leaves received nearly the full 676 amount of available light in all treatments (Fig. S2) and showed few signs of photoinhibition (Fig. 4). 677 These results suggest that the increase in stem volume is unrelated to photosynthetic physiology. 678 One possibility is that rather than (or in addition to) representing an increase in total growth due 679 to facilitation, the positive trend in stem volume in B. papyrifera is driven by a competitive shade 680 avoidance response, which could increase allocation to shoot biomass at the expense of root biomass. 681 Such responses tend to be especially strong in shade-intolerant, early-successional species like B.

Mechanisms of photoprotection 695
Plants' strategies to avoid damage from the stress of excess light can be classified broadly into 696 biochemical and structural strategies. We show that trees in full sun allocate more to biochemical 697 photoprotection, as indicated by lower PRI in all species and higher qN in all species but T. americana 698 and A. negundo (Fig. 4). Our leaf angle survey is also consistent with the idea that plants may steeply 699 incline their leaves to avoid intercepting excess light-a form of structural photoprotection (Fig. 6). 700 Leaves that are steeply inclined intercept less light per unit area, particularly during midday, when solar 701 radiation is otherwise most intense. The dramatic trend in leaf angles suggests that gross structural 702 characteristics allow these species use to regulate their light absorption. (2017) further showed that conifer partners in these species combinations also grew faster than they did in 758 monoculture, resulting in community overyielding. 759 The two broadleaf species that show the strongest evidence for facilitation by shade had low 760 tolerance to excess light, and likely benefited from the milder microclimate their larger neighbors created. 761 The conifer species, which are fast-growing and less shade-tolerant (except for P. strobus), may also have 762 grown faster in bicultures with broadleaf species because they faced less competition for light than in 763 monoculture. Other researchers have found that heterogeneity in shade tolerance explains the effect of 764 tree diversity on productivity, but they attributed this result to higher stand-level light capture enabled by    Table 1.