<?xml-model href='http://www.tei-c.org/release/xml/tei/custom/schema/relaxng/tei_all.rng' schematypens='http://relaxng.org/ns/structure/1.0'?><TEI xmlns="http://www.tei-c.org/ns/1.0">
	<teiHeader>
		<fileDesc>
			<titleStmt><title level='a'>Defoliated trees die below a critical threshold of stored carbon</title></titleStmt>
			<publicationStmt>
				<publisher></publisher>
				<date>10/01/2021</date>
			</publicationStmt>
			<sourceDesc>
				<bibl> 
					<idno type="par_id">10312329</idno>
					<idno type="doi">10.1111/1365-2435.13891</idno>
					<title level='j'>Functional Ecology</title>
<idno>0269-8463</idno>
<biblScope unit="volume">35</biblScope>
<biblScope unit="issue">10</biblScope>					

					<author>Audrey Barker Plotkin</author><author>Meghan Blumstein</author><author>Danelle Laflower</author><author>Valerie J. Pasquarella</author><author>Jennifer L. Chandler</author><author>Joseph S. Elkinton</author><author>Jonathan R. Thompson</author>
				</bibl>
			</sourceDesc>
		</fileDesc>
		<profileDesc>
			<abstract><ab><![CDATA[1. Carbon starvation posits that defoliation-and drought-induced mortality results from drawing down stored non-structural carbohydrates (NSCs), but evidence is mixed, and few studies evaluate mortality directly. We tested the relationships among defoliation severity, NSC drawdown and tree mortality by measuring NSCs in mature oak trees defoliated by an invasive insect, Lymantria dispar, across a natural gradient of defoliation severity.2. We collected stem and root samples from mature oaks (Quercus rubra and Q. alba) in interior forests (n = 34) and forest edges (n = 47) in central Massachusetts, USA.Total NSC (TNC; sugar + starch) stores were analysed with respect to tree size, species and defoliation severity, which ranged between 5% and 100%.3. TNC stores declined significantly with increasingly severe defoliation. Forest edge trees had higher TNC stores that were less sensitive to defoliation than interior forest trees, although this may be a result of differing defoliation history. Furthermore, we observed a mortality threshold of 1.5% dry weight TNC. 4. Our study draws a direct link between insect defoliation and TNC reserves and defines a TNC threshold below which mortality is highly likely. These findings advance understanding and improve model parametrization of tree response to insect outbreaks, an increasing threat with globalization and climate change.]]></ab></abstract>
		</profileDesc>
	</teiHeader>
	<text><body xmlns="http://www.tei-c.org/ns/1.0" xmlns:xsi="http://www.w3.org/2001/XMLSchema-instance" xmlns:xlink="http://www.w3.org/1999/xlink">
<div xmlns="http://www.tei-c.org/ns/1.0"><head n="1">| INTRODUC TI ON</head><p>Trees rely on stored sugars and starches, collectively termed nonstructural carbohydrates (NSCs), for metabolism, transport, defence, osmoregulation and as a buffer during stressful events when photosynthesis is impaired <ref type="bibr">(Dietze et al., 2014;</ref><ref type="bibr">Hartmann &amp; Trumbore, 2016;</ref><ref type="bibr">Mart&#237;nez-Vilalta et al., 2016;</ref><ref type="bibr">O'Brien et al., 2014;</ref><ref type="bibr">Sapes et al., 2021)</ref>. Abiotic and biotic disturbances or stressors that limit carbon gain should deplete NSC reserves and lead to death when below a critical level, a process referred to as 'carbon starvation' <ref type="bibr">(McDowell et al., 2008)</ref>. However, experimental and observational evidence for carbon starvation is mixed. <ref type="bibr">Piper and Paula (2020)</ref> synthesized 16 recent studies of NSC response to drought, defoliation and fire under natural conditions. They found lower NSCs in drought-affected Pinus species, but few studies and inconsistent results for other stressors. Similarly, in a review of drought mortality (19 studies synthesized in <ref type="bibr">Adams et al., 2017)</ref>, NSCs sometimes were depleted in trees that died, but often NSC stores were maintained and mortality was better explained by hydraulic failure.</p><p>Since the carbon starvation hypothesis was originally framed and usually examined in the context of drought, it is challenging to disentangle direct water limitation from carbon depletion, leading to the conclusion that trees typically die of thirst before starvation <ref type="bibr">(Adams et al., 2017)</ref>.</p><p>While drought is an increasing threat to many parts of the globe under climate change, especially the American West <ref type="bibr">(Cohen et al., 2016;</ref><ref type="bibr">Millar &amp; Stephenson, 2015)</ref>, insect outbreaks are an increasing threat to other forests, especially in the US Northeast and upper Midwest <ref type="bibr">(Aukema et al., 2010;</ref><ref type="bibr">Fei et al., 2019;</ref><ref type="bibr">Liebhold et al., 2013)</ref>.</p><p>However, compared to drought, there are few studies of mature tree NSC responses to insect herbivory <ref type="bibr">(Piper &amp; Paula, 2020)</ref>. Like NSC response to drought, defoliation studies have also found mixed evidence for mortality due to storage depletion. Tree species vary in defoliation tolerance <ref type="bibr">(Kulman, 1971)</ref>, and many produce a second flush of leaves with high assimilation rates that can compensate for the foliage lost <ref type="bibr">(Heichel &amp; Turner, 1983;</ref><ref type="bibr">Herms et al., 1992)</ref>. This partially explains why many studies have found that NSC stores remained relatively stable or rebounded quickly following defoliation <ref type="bibr">(Anderegg &amp; Callaway, 2012;</ref><ref type="bibr">Palacio et al., 2012;</ref><ref type="bibr">Piper et al., 2015;</ref><ref type="bibr">Schmid et al., 2017;</ref><ref type="bibr">Wiley et al., 2013)</ref>, although <ref type="bibr">Landh&#228;usser and Lieffers (2012)</ref> found that total NSC root reserves took more than 2 years to recover after herbivory. Notably, none of these defoliation studies included trees that ultimately died or examined the impacts of non-native pest outbreaks, preventing them from directly addressing the role of NSC depletion in mortality and the NSC store level below which recovery was not possible. Understanding what this mortality threshold is for trees is critical to predicting future forest response to infestation.</p><p>Here we leverage a natural gradient of defoliation severity and tree mortality to investigate the relationships among defoliation by non-native insect Lymantria dispar L. (formerly known as gypsy moth), NSC reserves and tree mortality. L. dispar is among the most destructive invasive forest insects in North America <ref type="bibr">(Elkinton &amp; Liebhold, 1990;</ref><ref type="bibr">Lovett et al., 2016)</ref>. It was a major defoliator in New England during the 20th century but has been quiescent for nearly 30 years. Early work identified a possible threshold for mortality of 2% root starch in young oaks defoliated by L. dispar <ref type="bibr">(Wargo, 1981)</ref>, but no work has been completed since to examine total NSC stores or the impacts to mature trees. We take advantage of a surprisingly severe L. dispar outbreak that began in 2016 across southern New England to investigate how depletion of NSC stores may lead to mortality. The outbreak resulted in dramatic mortality of oak (Quercus spp.) across thousands of forested hectares by 2018.</p><p>Lymantria dispar and its preferred hosts, Quercus spp., provide an ideal study system to examine how insect defoliation drives NSC reserve drawdown and mortality. Quercus is a globally important genus (Cavender-Bares, 2019) and a dominant in New England forests <ref type="bibr">(Finzi et al., 2020)</ref>. Furthermore, ring-porous oaks initiate growth and build new xylem prior to bud break each year <ref type="bibr">(Barbaroux &amp; Br&#233;da, 2002;</ref><ref type="bibr">Foster, 2017)</ref>, making them strongly dependent on reserves of stored NSCs in the spring. L. dispar persists at low population densities, punctuated by periodic outbreaks <ref type="bibr">(Elkinton &amp; Liebhold, 1990;</ref><ref type="bibr">Liebhold et al., 2000)</ref>. Oaks typically can survive 1 or 2 years of defoliation and can produce a second flush of leaves after L. dispar larvae pupate in early July <ref type="bibr">(Heichel &amp; Turner, 1983)</ref>, but mortality risk increases with successive years of defoliation <ref type="bibr">(Davidson et al., 1999)</ref>, as observed in our study. Infestation levels vary among trees in a stand and among stands, resulting in a natural gradient of defoliation severity <ref type="bibr">(Pasquarella et al., 2018)</ref>. We measured survival and NSC stores in oaks that experienced varying defoliation severity and predicted that:</p><p>1. NSC stores are depleted with increasingly severe defoliation and 2. There is a critical NSC threshold below which mortality occurs.</p><p>Following the collapse of the L. dispar outbreak in 2018, we measured survival and NSC stores during Winter 2018-2019 (after the final year of defoliation), and again during Winter 2019-2020 to test whether NSC stores recovered in trees that experienced severe defoliation. While some studies have cited rapid recovery of NSCs following defoliation <ref type="bibr">(Palacio et al., 2012;</ref><ref type="bibr">Wiley et al., 2013)</ref>, other studies observed that recovery took more than 2 years <ref type="bibr">(Landh&#228;usser &amp; Lieffers, 2012;</ref><ref type="bibr">Wiley et al., 2017)</ref>. We included trees from two distinct habitats-interior forest and roadside trees-because the ecological and economic consequences of pest outbreaks differ between trees in human-dominated landscapes versus forests <ref type="bibr">(Lovett et al., 2016)</ref>. Forest edge trees receive higher radiation than trees in the forest, which allows for higher leaf area and growth rates <ref type="bibr">(Reinmann &amp; Hutyra, 2017;</ref><ref type="bibr">Reinmann et al., 2020;</ref><ref type="bibr">Smith et al., 2018)</ref>.</p><p>Therefore, we predicted that forest edge trees would have higher NSC levels and that NSC stores of forest edge trees would be less sensitive to defoliation than interior forest trees.</p></div>
<div xmlns="http://www.tei-c.org/ns/1.0"><head n="2">| MATERIAL S AND ME THODS</head></div>
<div xmlns="http://www.tei-c.org/ns/1.0"><head n="2.1">| Study sites and L. dispar outbreak characteristics</head><p>In 2016, a major L. dispar outbreak began in southern New England, USA, and by 2018 had caused notable oak mortality. Defoliation severity and the number of consecutive years of defoliation were patchy at the regional, site and individual scales. In our central Massachusetts study region, defoliation was widespread in 2017 but more localized in 2016 and 2018 <ref type="bibr">(Elkinton et al., 2019;</ref><ref type="bibr">Pasquarella et al., 2018)</ref>.</p><p>During the mid-late growing season of 2018, we established two study sites. We installed plots in the oak-hardwood forests of the Quabbin Watershed Forest (42.5&#176;N; 72.3&#176;W, hereafter 'Forest').</p><p>Northern red oak (Quercus rubra L.) was the most common oak species, with some white oak (Q. alba L.) and a few black oak (Q. velutina Lam.) present. Although L. dispar is a generalist herbivore, nearly all the defoliation impacts were observed on Quercus and not the other species in the plots. We chose plots with a range of defoliation severity, based on a published Landsat time series analysis of canopy condition change <ref type="bibr">(Pasquarella, 2018;</ref><ref type="bibr">Pasquarella et al., 2017)</ref>. Within plots, defoliation of oaks varied. That is, even in lightly defoliated areas, we observed a few oaks that were severely defoliated. We also tagged 85 individual oak trees from roadsides and hedgerows in and within 15 km of the town of Amherst, Massachusetts (42.4&#176;N; 72.5&#176;W, hereafter 'Town'). Q. rubra and Q. alba were the major oak species present in this site as well. At both sites, for each tree we recorded diameter at breast height (DBH), species and canopy position (1 = full sun; 2 = partially shaded; 3 = mostly/fully shaded). In the Forest plots, we recorded defoliation class (1 = 0%-25% defoliated; 2 = 25%-50% defoliated; 3 = 50%-75% defoliated; 4 = 75%-100% defoliated). For the Town trees, we estimated defoliation to the nearest 5%. These two sites were established independently and came together in Winter 2018-2019 for this study.</p><p>To estimate defoliation intensity throughout the L. dispar <ref type="bibr">outbreak (2016-2018)</ref>, we used a Landsat-based condition assessment product (a more recent iteration of the product we used to choose the Forest plots; <ref type="bibr">Pasquarella et al., 2018)</ref> that was assessed using a large plot-level defoliation dataset in central Massachusetts <ref type="bibr">(Pasquarella et al., 2021)</ref>. Change in condition was based on comparison to the best model identified by <ref type="bibr">Pasquarella et al. (2021)</ref>: a harmonic regression model using a tri-annual frequencies, the Tasseled Cap Greenness spectral index, the full set of Landsat observations (Landsats 5, 7 and 8), a 2005-2015 baseline period and a 1 May-30 September monitoring period. For each tree, we extracted the average condition change score for the 2016, 2017 and 2018 monitoring periods for the Landsat pixel within which the tree was located. The spatial grain of the condition change product (30 m) may encompass canopies of multiple trees, and for the forest edge trees, may include other cover types. Therefore, while this product produces reliable site-level estimates of defoliation intensity, it may not scale well to individual trees.</p></div>
<div xmlns="http://www.tei-c.org/ns/1.0"><head n="2.2">| NSC sampling</head><p>We selected 34 Forest trees (from four plots) and 47 Town trees for NSC sampling, representing a range of defoliation severity in 2018.</p><p>All selected trees were alive in late summer 2018, although some were predicted to die within a year based on a visual assessment of crown dieback. Secondary selection criteria included sampling a range of tree size and species. Diameters of the sample trees ranged from 15 to 69 cm at the Forest site (mean 49 cm) and from 12 to 66 cm at the Town site (mean 35 cm). Most of the oaks at the two sites were red oak (Q. rubra, n = 66) but we included white oak (Q.</p><p>alba, n = 15) to the extent it was available.</p><p>Non-structural carbohydrate reserves tend to be most stable during the dormant season <ref type="bibr">(Landh&#228;usser &amp; Lieffers, 2003)</ref>, prior to seasonal drawdown for early stem growth and leaf-out <ref type="bibr">(Furze et al., 2018)</ref>. Therefore, we sampled NSCs in February 2019, and again in February 2020. From each sample tree, we collected a 2 cm (under bark) sample of wood from (a) the stem, at ~1.3 m height and (b) a coarse root, at ~20 cm from the base of the tree. Samples were collected using a standard 4.3 mm diameter increment borer (Hagl&#246;f Company Group), stored on ice in the field, and then stored in a -80&#8451; freezer until processing. We chose to exclude the inner bark, which includes the phloem, as we were largely interested in how longer-term stores are drawn down by defoliation and ultimately impact tree survival rather than current-year photosynthates. We chose to sample from the two largest storage pools in oaks, stems and roots <ref type="bibr">(Furze et al., 2018)</ref>. While branches are also a significant and dynamic store, we were limited in the location and height of the trees from sampling these tissues.</p></div>
<div xmlns="http://www.tei-c.org/ns/1.0"><head n="2.3">| Tree defoliation, dieback and mortality assessments</head><p>We assessed defoliation and/or dieback four times over the course of the study. At the initial defoliation assessment in late summer 2018, some of the Forest trees exhibited branch dieback and browning leaves. We recorded crown dieback and mortality status at leafout in May 2019, during June-August 2019 and in July-August 2020.</p><p>A tree was presumed dead if it did not produce leaves; no tree that failed to leaf out in May 2019 produced leaves later that season or in 2020. Crown dieback was rated visually on a five-point scale</p><p>In 2019, we recorded defoliation by L. dispar during their larval period from late May-early July.</p><p>Despite high egg-mass counts in Fall 2018, high larval mortality in 2019 led to very low defoliation <ref type="bibr">(Brown, 2019)</ref>. There were few L. dispar larvae and minimal defoliation in 2020.</p></div>
<div xmlns="http://www.tei-c.org/ns/1.0"><head n="2.4">| NSC analysis</head><p>Total NSC reserves were extracted from freeze-dried and ground woody tissue samples following the protocol of <ref type="bibr">Landh&#228;usser et al. (2018)</ref>. In brief, after grinding the wood samples using a Wiley mill and then a ball grinder, we extracted sugars from 30 mg of dried tissue using 80% hot ethanol, then read them using a phenolic colorimetric assay and a spectrophotometer at 490 nm (ThermoFisher Scientific Genesys 180 UV-Vis Spectrophotometer). The methodology extracted and quantified all soluble sugars in the samples, including glucose, sucrose, fructose, and other oligosaccharides and glucans <ref type="bibr">(Landh&#228;usser et al., 2018)</ref>. We included two internal laboratory standards (Q. rubra stem-wood) with each set of samples, with sets within 10% of the laboratory estimate (42 mg/g) accepted.</p><p>Starch was extracted from the tissue remaining after sugar extraction using two digestive enzymes: alpha-amylase and amyloglucosidase. Enzymes came in powder form and the amount dissolved into solution depended on the recorded purity of the enzyme batch, to standardize the enzyme activity units per run (Sigma Aldrich code A4551 and ROAMYGLL). We solubilized the tissue in NaOH, then incubated it at 85&#8451; for 30 min with alpha-amylase dissolved into deionized water, then 55&#8451; for 30 min with amyloglucosidase dissolved in NaOAc buffer solution (25 mM, pH 4.6); both digest starch into glucose. Solutions were assayed using a PGO-colour reagent solution (Sigma Chemicals) and read on the spectrophotometer at 525 nm. Starch concentrations (mg glucose-starch-equivalent per g dry wood) were calculated based on a glucose standard curve (Sigma Chemicals).</p></div>
<div xmlns="http://www.tei-c.org/ns/1.0"><head n="2.5">| Data analysis</head><p>We used TNC in our analyses, rather than separating the total into the starch and sugar components. We chose to use TNC because we were interested in how the total supply is affected by defoliation, and because sugar and starch content is highly dynamic as they can hydrolyse back and forth in the tree on the order of hours <ref type="bibr">(Tixier et al., 2018)</ref>. Two of 329 TNC values were negative, with all other values ranging between 0.01% and 11% dry weight. These negative values are the result of zero values with minor error from the laboratory process. Because negative TNC values are biologically impossible, we set the negative TNC values to 10 -8 %.</p><p>Data visualization and analysis were conducted in R version 3.6.1 (R Core Team, 2019). We used a generalized linear mixed modelling (GLMM) approach to analyse the data, using the gLm-mTMB package <ref type="bibr">(Brooks et al., 2017)</ref>. To test whether TNC declined with increasing defoliation severity, we analysed the Town and Forest sites separately because we expected edge trees to have higher reserves and therefore more capacity to tolerate defoliation. Fixed effects included 2018 defoliation severity, tree diameter, tree species and canopy exposure. We included diameter to control for the fact that larger trees have larger NSC stores <ref type="bibr">(Sala &amp; Hoch, 2009)</ref>, and therefore may be more tolerant of defoliation.</p><p>We included species as a factor because of past evidence for interspecific differences in defoliation effects. For example, a classic paper by <ref type="bibr">Campbell and Sloan (1977)</ref> found that Q. alba was more heavily defoliated and more likely to die than Q. rubra. Although we expected well-lit canopies to have higher resources and more tolerance to defoliation than shaded canopies, canopy exposure was not a significant predictor for either the Forest or Town site, so was not included in the final models. Random effects included year and nested plot/tree/tissue (all Amherst trees were considered one 'plot'). Because TNC is measured as a proportion (percent dry weight), we chose a GLMM with a beta distribution and a logit link <ref type="bibr">(Douma &amp; Weedon, 2019)</ref>.</p><p>To test whether TNC levels predicted tree mortality, we fit a logistic GLMM. We included the Town and Forest sites in the same model because we expected mortality to depend on a critical threshold of TNC regardless of site. Fixed effects included TNC percent and tissue (stem or root). Random effects included year and nested site/ plot. For all models, we assessed model fit by examining standardized residuals using the package DHARMA <ref type="bibr">(Hartig, 2020)</ref>, and extracted model predictions using the package ggEffEcTs <ref type="bibr">(L&#252;decke, 2018)</ref>. Total NSC values were higher on average in the Town than the Forest site and declined less in response to defoliation (Figure <ref type="figure">2</ref>). We observed higher TNC in 2019 than in 2020, and root tissue TNC tended to be higher than stem tissue TNC, especially in 2019 (Table <ref type="table">1</ref>). Sugars comprised a larger share of TNC than did starch in our mid-winter sampling (Table <ref type="table">1</ref>).</p></div>
<div xmlns="http://www.tei-c.org/ns/1.0"><head n="3.2">| Non-structural carbohydrate stores are depleted with increasingly severe defoliation</head><p>TNC stores declined with increasing defoliation severity at both sites, but the strength of the relationship varied between sites, as did the influence of tree size and species. The final model for the Forest site showed that TNC stores declined with increasing percent defoliation (Table <ref type="table">2</ref>; Figure <ref type="figure">2a</ref>). Tree diameter was a significant covariate; larger trees had higher TNC. However, there was no significant difference between TNC stores in Q. rubra and Q. alba.</p><p>The effect size of defoliation severity was larger than that of tree diameter. The final model for the Town site also showed that TNC stores declined with increasing percent defoliation, but the effect size was small compared to the Forest site (Table <ref type="table">2</ref>; Figure <ref type="figure">2b</ref>). Tree diameter was not a significant covariate in the Town model, but Q. rubra had significantly higher TNC stores than Q. alba.</p><p>We also observed interannual variation in average TNC stores; 2019 TNC stores on average were higher than 2020, regardless of defoliation severity. This pattern indicates that NSC stores are also responding to differences in climatic conditions between the two years. 2018 was an unusually wet year <ref type="bibr">(Boose &amp; Gould, 2019)</ref>, which may explain the overall higher TNC in Winter 2018-2019 than in Winter 2019-2020. We found no evidence that TNC variation between years related to defoliation severity and no evidence of TNC recovery; an interaction term between defoliation and year was nonsignificant, indicating that TNC difference between 2019 and 2020 was not mediated by defoliation severity.</p></div>
<div xmlns="http://www.tei-c.org/ns/1.0"><head n="3.3">| There is a critical non-structural carbohydrate threshold below which mortality occurs</head><p>All the trees that died had TNC levels measured the prior February below 1.5% dry weight in root and stem samples (Figure <ref type="figure">3</ref>). Across sites and years, TNC was a strong predictor of mortality (Table <ref type="table">2</ref>; Figure <ref type="figure">4</ref>) with a sharp increase in survival probability above the 1.5% TNC threshold. This did not differ significantly between stem and root tissue (Table <ref type="table">2</ref>). Surprisingly, eight trees with TNC levels that dropped below the 1.5% threshold survived (Figure <ref type="figure">3</ref>), all from the Forest site. In 2019, four surviving trees had stem (but not root) TNC levels below 1.5%. One of these three trees again had stem TNC &lt;1.5% in 2020, but the other three trees' stem TNC levels were &gt;1.5%. In 2020, four additional trees had stem and/or root TNC levels &lt;1.5%. These </p></div>
<div xmlns="http://www.tei-c.org/ns/1.0"><head n="4">| D ISCUSS I ON</head><p>This study demonstrates that defoliation drives the drawdown of NSC reserves and eventual mortality of trees when defoliation is severe, especially for interior forest trees. These results stand in contrast to studies that have shown weak and transient NSC drawdown in response to defoliation <ref type="bibr">(Palacio et al., 2012;</ref><ref type="bibr">Piper et al., 2015;</ref><ref type="bibr">Piper &amp; Paula, 2020)</ref>. Furthermore, all the trees that died had root and stem TNC levels &lt;1.5%. Few past studies of NSC response to defoliation included trees that died <ref type="bibr">(Wargo, 1981;</ref><ref type="bibr">Webb, 1981)</ref>; our study identifies a critical threshold below which mortality is highly likely, directly addressing the role of NSC depletion in mortality.</p></div>
<div xmlns="http://www.tei-c.org/ns/1.0"><head n="4.1">| NSC stores are depleted with increasingly severe defoliation, especially in interior forest</head><p>We found a significant relationship between defoliation and TNC in both Town and Forest trees, and we found no evidence of recovery one full year after the L. dispar outbreak ended. We expect recovery of TNC to take several years and to be influenced by climate and other disturbances <ref type="bibr">(Anderegg, Schwalm, et al., 2015;</ref><ref type="bibr">Landh&#228;usser &amp; Lieffers, 2012)</ref>. Our study agrees with earlier findings of NSC drawdown following defoliation, including root starch in oaks defoliated by L. dispar <ref type="bibr">(Wargo, 1981)</ref>, stemwood starch in experimentally defoliated oaks <ref type="bibr">(Wiley et al., 2017</ref>, though root starch recovered quickly), Douglas-fir defoliated by tussock moth <ref type="bibr">(Webb, 1981)</ref> and aspen defoliated by native insects <ref type="bibr">(Landh&#228;usser &amp; Lieffers, 2012)</ref>. However, experimental defoliation of aspen <ref type="bibr">(Anderegg &amp; Callaway, 2012)</ref>, defoliation of pine by pine processionary moth <ref type="bibr">(Palacio et al., 2012)</ref> and defoliation of Nothofagus by Ormiscodes amphimone <ref type="bibr">(Piper et al., 2015)</ref> found only transient drawdown of NSCs. It is intriguing that most of the studies examined the effects of herbivory by native insects; perhaps non-native insects have a more severe effect on tree carbon stores because the hosts lack an evolutionary history with them <ref type="bibr">(Gandhi &amp; Herms, 2010)</ref>.</p><p>It is unclear why findings differ among studies, especially within the same species. Defoliation history and additional stressors such as drought and competition likely play a part. As conceptualized by the tree mortality spiral, decline and death is a complex, multi-factor process <ref type="bibr">(Franklin et al., 1987;</ref><ref type="bibr">Manion, 1981)</ref>. Many species are tolerant of a single defoliation event <ref type="bibr">(Kulman, 1971)</ref>. Most (but not all, e.g. <ref type="bibr">Palacio et al., 2012;</ref><ref type="bibr">Wargo, 1981)</ref> prior studies measured NSC drawdown after a single defoliation. In our study, although the defoliation severity of individual trees in 2018 was a much stronger predictor of NSC drawdown than the site-level defoliation history 2016-2018, widespread defoliation in 2017 likely contributed to slow recovery of NSCs. In addition, it is well documented that mortality is greater when L. dispar defoliation follows drought <ref type="bibr">(Campbell &amp; Sloan, 1977;</ref><ref type="bibr">Davidson et al., 1999)</ref>. A drought in 2016 <ref type="bibr">(Koch &amp; Coulton, 2018)</ref> as the L. dispar outbreak began in central New England likely exacerbated the effects of defoliation in 2016-2018 and contributed to NSC depletion and tree mortality. We F I G U R E 2 Total non-structural carbohydrates (TNC) declined with increasing defoliation severity of oak (Quercus rubra and Q. alba) by Lymantria dispar at the Forest (a) and Town (b) sites. The response was stronger for the Forest site (interior forest trees) than Town site (forest edge trees). The lines and shading show how the predicted TNC amount and 95% CI change with percent defoliation. The points (jittered) show the data did not begin this study until 2018, so could not assess this directly, but a related study (currently in review) found dramatically lower TNC in oaks and other tree species in central New England following the 2016 drought, building on work that showed that NSCs vary among years <ref type="bibr">(Furze et al., 2018;</ref><ref type="bibr">Landh&#228;usser &amp; Lieffers, 2012;</ref><ref type="bibr">Richardson et al., 2013)</ref> and that climate is a major driver of interannual variation <ref type="bibr">(Landh&#228;usser &amp; Lieffers, 2012)</ref>. While 2016 was an unusually dry year in our study area, 2018 was an unusually wet year, which may explain the overall higher TNC in Winter 2018-2019 than in Winter 2019-2020. Recovery may also have been complicated by the fact that trees redistribute nitrogen following a defoliation event <ref type="bibr">(Lovett et al., 2002)</ref>.</p><p>Nitrogen redistribution has complex feedbacks on L. dispar <ref type="bibr">(Lindroth et al., 1997)</ref> and tree performance. In a study that included some of the same tree and forest plot sites, Conrad-Rooney et al. (2020) found lower total soil and leaf N associated with sites defoliated more intensively; low N could limit tree recovery in addition to TNC <ref type="bibr">(Piper et al., 2015)</ref>. Another possible explanation for the inconsistency in NSC drawdown to stress, not tested in this study, is active regulation of NSCs.</p><p>Although carbon storage is typically modelled as a simple source-sink process <ref type="bibr">(Dietze &amp; Matthes, 2014;</ref><ref type="bibr">Dietze et al., 2014)</ref>, there is increasing evidence that storage is actively regulated and may be prioritized over growth when carbon is limiting <ref type="bibr">(Herms et al., 1992</ref>; Mart&#237;nez-Vilalta Forest (n = 34) Town (n 2019 = 47; n 2020 = 45) Root Stem Root Stem 2019 mean % dry weight (SD) Sugars 3.35 (1.79) 2.00 (1.38) 4.99 (1.18) 3.55 (0.78) Starch 1.00 (1.15) 0.56 (0.59) 1.61 (1.10) 1.71 (0.85) Total (TNC) 4.34 (2.56) 2.56 (1.81) 6.60 (1.72) 5.26 (1.24) 2020 mean % dry weight (SD) Sugars 1.81 (1.24) 1.53 (1.03) 3.46 (0.73) 2.74 (0.77) Starch 0.80 (0.82) 0.62 (0.66) 1.06 (0.82) 1.58 (0.59) Total (TNC) 2.61 (1.65) 2.15 (1.43) 4.52 (1.15) 4.32 (1.09)  <ref type="bibr">et al., 2016;</ref><ref type="bibr">Wiley et al., 2017)</ref>. Genetic variation in NSC regulation <ref type="bibr">(Blumstein &amp; Hopkins, 2020)</ref> could underlie the equivocal support for the carbon starvation hypothesis. In addition, many studies focus on seedlings <ref type="bibr">(Galvez et al., 2013;</ref><ref type="bibr">O'Brien et al., 2014;</ref><ref type="bibr">Sapes et al., 2021)</ref>, which likely differ from adult trees in resource acquisition and storage.</p><p>Total NSC stores were lower in the interior forest trees (Forest) than in the forest edge and street trees (Town), and the effect of defoliation on TNC levels was stronger for the Forest trees. This could be because defoliation was milder at the Town site than the Forest, as indicated by the Landsat condition change scores for 2017-2018</p><p>(Figure <ref type="figure">1</ref>). However, the Landsat condition change scores are appropriate for the plot (30 m) scale but did not predict individual-tree defoliation (Figure <ref type="figure">1g</ref>). Forest edge pixels often include a mix of cover types, which may lead to underestimated defoliation severity if part of the pixel contains cover that is not subject to defoliation. An alternate explanation is that because forest edge trees have higher leaf area and access to light, they have higher NSC levels that are less sensitive to the effects of defoliation than interior forest trees. Furthermore, trees receiving more light may induce more defence compounds such as jasmonic acid <ref type="bibr">(Elderd et al., 2013;</ref><ref type="bibr">Kazan &amp; Manners, 2011)</ref>, which could add to the edge trees' advantage. If so, our findings suggest that NSC depletion may be a greater risk for interior forests than forest edges, and adds to our growing understanding of how forest edges differ from interior forests <ref type="bibr">(Reinmann &amp; Hutyra, 2017;</ref><ref type="bibr">Reinmann et al., 2020;</ref><ref type="bibr">Smith et al., 2018)</ref>. In addition, this may help explain why some previous studies (e.g. <ref type="bibr">Wiley et al., 2013)</ref> that drew inference from trees growing in a tended garden setting did not find NSC drawdown with defoliation.</p></div>
<div xmlns="http://www.tei-c.org/ns/1.0"><head n="4.2">| There is a critical NSC threshold below which mortality occurs</head><p>Trees survived with very low TNC reserves. Only trees with root TNC levels below 1.5% dry weight died, and some died with fully depleted (i.e. zero) reserves. This is in contrast to drought <ref type="bibr">(McDowell &amp; Sevanto, 2010;</ref><ref type="bibr">Nelson et al., 2018)</ref> or defoliation <ref type="bibr">(Anderegg &amp; Callaway, 2012)</ref> studies that showed that mortality occurred without carbon depletion, but bolsters and builds on past work that showed that starch reserves were depleted in trees that died following insect defoliation <ref type="bibr">(Wargo, 1981;</ref><ref type="bibr">Webb, 1981)</ref>. Most studies of NSC response to defoliation do not include trees that died <ref type="bibr">(McDowell &amp; Sevanto, 2010;</ref><ref type="bibr">Piper &amp; Paula, 2020)</ref>, and therefore miss threshold dynamics as shown in the present study. <ref type="bibr">Wiley et al. (2016)</ref> examined NSC dynamics of lodgepole pines killed by mountain pine beetle, but since the mortality agent was a phloem borer, the mortality process likely differs from defoliation. Unlike early studies that measured only starch <ref type="bibr">(Wargo, 1981;</ref><ref type="bibr">Webb, 1981)</ref>, our study demonstrates that both starch and sugars, in both stems and roots, were drawn down to near-zero levels before mortality occurred, indicating full depletion of TNC reserves. Whereas starch functions as a storage reservoir, sugars are critical to maintain functions such as osmoregulation and freezing tolerance (Mart&#237;nez-Vilalta et al., 2016; Sapes et al., 2021), and therefore sugars are a critical component of determining carbon starvation thresholds.</p><p>Surprisingly, some surviving trees had TNC reserves below the 1.5% threshold. Similarly, <ref type="bibr">Wargo (1981)</ref> found that root starch levels &lt;2% were a necessary, but not sufficient, condition for oak mortality; that is, some trees below that threshold survived. It appears that trees can maintain basic metabolism and osmoregulation at very low levels of TNC, at least for short periods of time. In the present study, however, most of these trees exhibited some canopy dieback and may continue to decline.</p></div>
<div xmlns="http://www.tei-c.org/ns/1.0"><head n="5">| CON CLUS IONS</head><p>Most ecophysiologically based forest models formalize the concept of carbon starvation to simulate tree mortality in response to non-physical disturbances, such as drought, defoliation and pathogens <ref type="bibr">(Anderegg, Hicke, et al., 2015;</ref><ref type="bibr">Dietze &amp; Matthes, 2014)</ref>. This includes widely used dynamic vegetation models (DVMs), such as the Community Land Model <ref type="bibr">(Lawrence et al., 2019)</ref>, PnET <ref type="bibr">(Aber &amp; Federer, 1992;</ref><ref type="bibr">De Bruijn et al., 2014)</ref> and the Ecosystem Demography Model <ref type="bibr">(Medvigy et al., 2009)</ref>, that scientists and policymakers rely on to forecast ecosystem responses to global change (e.g. <ref type="bibr">Thompson et al., 2020)</ref>. These models all assume that tree mortality is dependent on a full drawdown of carbon stores, an assumption that has until now lacked empirical evidence. Our finding of a 1.5% TNC threshold can improve mortality functions in DVMs and our ability to predict impacts of disturbance.</p><p>This study leveraged a unique natural defoliation gradient to demonstrate that defoliated trees drew down their NSC stores, sometimes to zero. Mortality occurred below a threshold of 1.5%</p><p>TNC. The strong relationships among defoliation, NSC depletion and mortality are in accord with the carbon starvation hypothesis, although, as posited by the tree mortality spiral concept <ref type="bibr">(Franklin et al., 1987;</ref><ref type="bibr">Manion, 1981)</ref>, carbon starvation may not be the sole cause of death. Because we were unable to assess tree hydraulics, we could not say that hydraulic failure did not contribute to tree death. The 2016 drought may have caused some cavitation (the blockage of water flow in vessels by air bubbles) and weakened the trees. Since oaks are ring-porous and routinely re-build xylem each spring <ref type="bibr">(Barbaroux &amp; Br&#233;da, 2002)</ref>, any cavitation that occurred in 2016 would have been unlikely to continue to hamper trees' hydraulic function in subsequent years. However, NSCs are critical for hydraulic function <ref type="bibr">(O'Brien et al., 2014;</ref><ref type="bibr">Sapes et al., 2021)</ref>, so the NSC depletion we documented here may have impaired hydraulic function. Secondary infection by native insects or diseases is another alternative cause of death of trees weakened by defoliation <ref type="bibr">(Davidson et al., 1999;</ref><ref type="bibr">Wargo, 1996)</ref>. However, since NSCs are involved in tree defence <ref type="bibr">(Herms et al., 1992)</ref>, carbon starvation may result in vulnerability to typically innocuous pests <ref type="bibr">(Anderegg &amp; Callaway, 2012)</ref>. The in situ conditions of this study that included climate stress and more than one year of defoliation demonstrate a realistic scenario in which compounding stressors and disturbance led to decline and mortality. As the direct cause of death or as a precondition, we show that carbon starvation is a key part of the tree mortality spiral.</p></div></body>
		</text>
</TEI>
