Retrospective analysis of wood anatomical traits and tree‐ring isotopes suggests site‐specific mechanisms triggering Araucaria araucana drought‐induced dieback

In 2010–2018, Northern Patagonia featured the longest severe drought of the last millennium. This extreme dry spell triggered widespread growth decline and forest dieback. Nonetheless, the roles played by the two major mechanisms driving dieback, hydraulic failure and carbon starvation, are still not clear and understudied in this seasonally dry region. Here, for the 1800–2017 period, we apply a retrospective analysis of radial growth, wood anatomical traits (lumen area, cell‐wall thickness) and δ13C and δ18O stable isotopes to assess dieback causes of the iconic conifer Araucaria araucana. We selected three stands where declining (defoliated) and nondeclining (not defoliated) trees coexisted along a precipitation gradient from the warm‐dry Coastal Range to the cool‐wet Andes. At all sites declining trees showed lower radial growth and lower theoretical hydraulic conductivity, suggesting a long‐lasting process of hydraulic deterioration in their water transport system compared to nondeclining, coexisting trees. Wood anatomical traits evidenced that this divergence between declining and nondeclining trees started at least seven decades before canopy dieback. In the drier stands, declining trees showed higher water‐use efficiency (WUE) throughout the whole period, which we attributed to early stomatal closure, suggesting a greater carbon starvation risk consistent with thinner cell walls. In the wettest stand, we found the opposite pattern. Here, a reduction in WUE coupled with thicker cell walls suggested increased carbon assimilation rates and exposure to drought‐induced hydraulic failure. The δ18O values indicated different strategies of gas exchange between sites, which are likely a consequence of microsite conditions and water sources. Multiproxy, retrospective quantifications of xylem anatomical traits and tree‐ring isotopes provide a robust tool to identify and forecast, which stands or trees will show dieback or, on the contrary, which will likely withstand and be more resilient to future hotter droughts.


| INTRODUC TI ON
Hotter droughts, that is, droughts accompanied by warmer temperatures, in combination with altered disturbance regimes, have caused forest dieback, and increased tree mortality rates, modifying the structure and distribution of forests worldwide (Allen et al., 2010(Allen et al., , 2015Anderegg et al., 2013;McDowell et al., 2020). In addition, climatic models have predicted the frequency, intensity, and duration of such hotter droughts will increase in the next decades, and such events will be more severe in areas subjected to seasonal drought (Jia et al., 2019;Trenberth et al., 2014).
Hotter drought has negative impacts on physiological processes in forest ecosystems, constraining growth and promoting transpiration, inducing hydraulic failure via xylem embolism, and limiting the carbon sink capacity via stomatal closure (Bréda et al., 2006). Trees could become prone to growth decline, crown dieback, and death from two nonexclusive physiological mechanisms: hydraulic failure or carbon starvation (Adams et al., 2017;Hammond et al., 2019;McDowell et al., 2008). To better understand the long-term changes in growth and vigor and to improve our predictive capacity of dieback and tree death in response to hotter droughts in the future, it is crucial to identify which are the physiological mechanisms that allow trees to withstand and cope with severe water shortage (Brodribb et al., 2020;McDowell et al., 2020). The drought-induced cascading changes in tree function (e.g., loss in hydraulic conductivity, impaired carbon uptake) leading to dieback and tree mortality can be reconstructed by analyzing longterm changes in tree radial growth (Cailleret et al., 2016(Cailleret et al., , 2019, woodanatomical traits (e.g., conduits dimension) (Pellizzari et al., 2016), and intrinsic water-use efficiency (WUE i ), (Colangelo et al., 2017). These are some of the proxies of tree functioning, which may allow reconstructing and inferring the relative roles played by the two proposed mechanisms leading to dieback.
In the case of wood anatomy, traits such as lumen area (LA) (a proxy of theoretical hydraulic conductivity; Cuny et al., 2015) and cell-wall thickness (CWT) (a proxy for xylem carbon use ;Castagneri et al., 2017Castagneri et al., , 2018 allow quantifying the long-term tree response to drought stress (Fonti et al., 2010). The evaluation of WUE i , that is, the cost of fixing carbon in terms of water loss (Ehleringer et al., 1993), may be achieved by measuring carbon isotope discrimination in wood or cellulose (δ 13 C), which provides information about the ratio of net photosynthetic CO 2 assimilation rate (A) to stomatal conductance rate (g s ), which depends of temperature and vapor pressure deficit among other factors (Farquhar et al., 1982). Furthermore, the contribution of g s in driving variation of WUEi is inferred by assessing the variation of δ 18 O in tree rings, which depends on leaf transpiration and the evaporative enrichment of oxygen isotope ratios (Saurer et al., 2004;Scheidegger et al., 2000;Seibt et al., 2008). The combination of this dual-isotope approach has been adopted to better understand water and carbon limitations in drought-induce dieback studies (Cherubini et al., 2021;Colangelo et al., 2017;Petrucco et al., 2017;Timofeeva et al., 2017).
Hence, a multiproxy study of tree-ring width (TRW), wood anatomy, and WUE i can provide new insights to disentangle the physiological mechanisms under drought conditions and to reconstruct long-term water-use strategies in trees (Gessler et al., 2018;Hentschel et al., 2014). Nonetheless, until now, this multifaceted approach has been hardly applied, and a clear and univocal interpretation of drought-induced dieback in many tree species is still lacking.
For instance, some authors have observed in declining trees the reduction of LA, and this has been linked to reduced hydraulic conductivity (Liang et al., 2013;Pellizzari et al., 2016). Others have found the opposite, interpreting that fast-growing trees, producing wider lumens in stressful conditions, were more prone to xylem embolism and declined (Petrucco et al., 2017;Voltas et al., 2013). Related to tree growth, Cailleret et al. (2019) compared radial growth trends and rates in coexisting surviving and dead trees, and they observed in gymnosperms a drastic reduction of radial growth 20 years before mortality; however, other studies found the opposite, that is, declining trees were fairly fast-growing before the drought started (Hentschel et al., 2014;Voltas et al., 2013). Finally, recent studies have shown an increase of WUE i in declining trees (Timofeeva et al., 2017). However, until now, there is no consensus about the mechanisms triggering changes in growth, vigor and causing dieback and mortality (Brodribb et al., 2020;Gessler et al., 2018;Trugman et al., 2021). It is, therefore, possible that the relative importance of mechanisms leading to dieback is reliant on site-specific conditions, which requires studying sites subjected to contrasting climatic conditions. In South America, several temperate tree species showed abrupt growth reductions and mortality events in response to 20th-century warming and drying trends across Patagonia (Cailleret et al., 2016;Villalba et al., 2012). Several studies carried out in Argentina and Chile have shown die-off phenomena affecting Nothofagus species and some conifers such as Austrocedrus chilensis (Amoroso et al., 2012;Miranda et al., 2020;Serrano-León & Christie, 2020;Suarez et al., 2004). Here, we argue how a multiproxy approach, combining analysis of wood anatomical traits and stable isotopes in tree-rings, can be leveraged for a comprehensive interpretation of the physiological mechanisms involved in forest dieback. This is the first study to investigate, in the iconic and threatened conifer species Araucaria araucana (Molina) K. Koch, the causes of drought-induced dieback by retrospectively analyzing an assemblage of wood traits, that is, longtime series of tree growth, wood-anatomical traits, and tree-ring stables isotopes in coexisting nondeclining and declining trees, and along a precipitation gradient in Northern Patagonia. In the last decades, this region experienced very dry conditions in 1998, linked to La Niña event (Muñoz et al., 2016), and the longest drought (2010-2018) of the last millennium (Garreaud et al., 2019). These and previous dry spells impacted Northern Patagonia forests in several ways by increasing canopy dieback incidence, tree mortality rates and enhancing wildfire occurrence (Kitzberger et al., 2001;Villalba & Veblen, 1998). Here, we hypothesize that declining and dying trees, showing higher percentage of crown defoliation, could be characterized by reduced tree-ring growth and lower hydraulic conductivity than coexisting nondeclining trees. Specifically, we expect that declining trees will produce smaller LA (mirroring water shortage during tracheid enlargement) and thinner CWT (reflecting depletion of carbon reserves due to reduced photosynthetic and cell-wall lignification rates) than nondeclining trees. We also expect that declining trees will present a higher WUE i , due to reduced water loss per unit carbon gain, compared to nondeclining trees. Investigating how trees respond to long-term warming conditions and drought will help us to disentangle the physiological mechanisms and WUE i strategies to infer causal links between drought, dieback, and tree death. This knowledge has the potential to predict which tree populations will significantly decline or, on the contrary, which stands will be able to withstand and be more resilient to future climate warming (Pellizzari et al., 2016). Such tools are urgently needed for threatened species such as A. araucana, with highly endangered provenances inhabiting the driest sites of its distribution area (Premoli et al., 2013).  (Veblen et al., 1995).

| Tree species and geographical distribution
Araucaria araucana is a dioecious conifer, which can reach up to 40 m in height, forming a thick and straight cylindrical trunk up to 2 m in diameter. Nowadays, the species is classified as endangered, due to extensive logging and human-induced fires (Premoli et al., 2013).

| Study sites and climate
In January 2019 (austral summer), we selected three sites dominated by A. araucana affected by the 2010-2018 drought (Garreaud et al., 2019) with a high percentage of canopy defoliation, leaves discoloration and branch dieback and abundant dying trees.
Sites were selected across the natural distribution of A. araucana in the Chilean Coastal range and the Andes range in the border between Chile and Argentina ( Figure 1; Table 1). The sites were Trongol Alto (hereafter abbreviated as CHI1), Altos de Pemehue National Reserve (hereafter abbreviated as CHI2), both located in Chile, and Tromen Lake situated in Argentina (hereafter abbreviated as ARG1).
All sites were located within conservation areas. Sites selection allowed the comparison along a marked precipitation gradient since rainfall increases eastwards and upwards. Moving from north to south, the average annual temperature is 13.0° to 11.7℃ and the mean yearly precipitation ranges from 1091 to 3002 mm ( Figure S1).
All locations show the typical temperate rainforest climate characterized by wet winters and hot summers with a more marked summer drought at CHI1. In the Andes CHI2 and ARG1 sites, the soil is shallow, composed by volcanic ash and scoria characterized by low water holding capacity and deficient nutritional levels. On the other hand, in the coastal range site (CHI1), soils are usually shallow but poor-drained and develop on granitic and metamorphic bedrocks (CIREN, 2010). All sites present signs of not recent (more than 50-100 years) natural or anthropogenic disturbances mostly in terms of fires and logging.

| Field sampling
The three sites were dominated by A. araucana old-growth forests with abundant mature trees and heterogeneous vertical and horizontal structures. We randomly selected trees firstly defining their health status according to crown condition by estimating the percentage of crown defoliation of each tree using binoculars. Trees presenting crown defoliation <50% were categorized as nondeclining individuals, while trees with crown defoliation ≥50% were F I G U R E 1 Araucaria araucana natural distribution (purple areas). Location of the three study sites in temperate rainforest in northern Patagonia, including southern Chile and south-western Argentina (red stars considered declining . Then, after having identified a declining individual, we selected a healthy one with similar dimensions as close as possible to the first one. We finally selected 54 trees (27 couples) and for each tree, after defining the defoliation status, we measured diameter at breast height (measured at 1.3 m) and height (using a laser rangefinder), and collected two 10mm cores per tree with an increment borer (Table 2).

| Climate data
Daily climate records of rainfall, minimum, mean, and maximum temperature, for each site were obtained from the Chilean Directorate of water resources (DGA), and National Weather Service (DMC; Table S1 and Figure S1). All the station data are available from the Research Center for Climate Science and Resilience (http://explo rador.cr2.cl).
In addition, to quantify drought severity, we calculated the

| Processing of wood samples
Once back in the laboratory, cores were air dried, polished with successive grains to make the ring boundaries visible at best, and then ring widths were measured to a precision of 0.01 mm using the TSAP measuring device (Rinntech). Then the tree-ring series were visually cross-dated using standard dendrochronological methods (Stokes & Smiley, 1968), and checked for dating accuracy and measurement errors with the COFECHA program (Holmes, 1983). For dating purposes, we followed the convention for the Southern Hemisphere, which assigns to each ring the date of the year in which growth started (Schulman, 1956). Finally, we defined the age of each tree as the sum of the number of years from the cross-dated cores plus the number of rings estimated with pith-offset (the difference in years between the innermost dated ring in a core and the pith at sampling level) (Norton et al., 1987).  (Table 2).

| Wood anatomy data
Both TRW and anatomical trait series were standardized by removing the size/age trends using a 32-year cubic smoothing spline, to stabilize the variance and preserve high to decadal frequency variability potentially related to climate (Carrer et al., 2015;Cook & Kairiukstis, 1990). To accomplish this step, individual series were detrended fitting the spline curve to measured data series and then Defoliation (%) 28.3 ± 3.5 a 9.4 ± 5.1 a 27.9 ± 3.7 a 67.9 ± 4.6 b 79.2 ± 7.3 b 73.2 ± 4.3 b Different letters indicate significant differences (p < .05) between nondeclining and declining trees of the same site based on Mann-Whitney test.
dividing observed by expected values. The resulting dimensionless indexes were then averaged at site level for each vigor class using a biweight robust mean (Cook & Kairiukstis, 1990). Then, raw and standardized chronologies for anatomical traits of CN, LA, LD, K h , D h , CD, CWT, CWA, and CTA were computed with dplR package in R (Bunn, 2008;Bunn et al., 2021;R Core Team, 2021). We assessed the common variability between trees by calculating, for the time series of the selected variables (TRW, LA, K h , and CWT), the mean correlation (rbar) values and the common variance considering the common period 1900-2017 (Table S2).
Later, for the selection of the most representative xylem traits that can be highly correlated among them, we performed at site

| Stable isotope analyses
The carbon and oxygen isotope analyses were performed on the same trees used for wood-anatomical analyses. We used rings spanning the period 1800-2017 and split the cores with a scalpel under a binocular in decadal wood segments (e.g., 1800-1809, 1810-1819, …, excepting 2010-2017). In total, we obtained 22 samples per tree.
The same chronological decadal segments were pooled together at site level, milled to a fine powder with a ball mill (ZM 1000; Retsch) and weighed 0.05-0.06 mg and 0.08-0.09 mg of wood for carbon and oxygen isotope analyses, and, finally, encapsulated in tin and silver capsules, respectively.
The isotope composition was measured at the IRMS laboratory of the University of Campania "Luigi Vanvitelli" by using mass spectrometry with continuous flow isotope ratio (Delta V plus Thermo electron Corporation). Standard deviation for repeated analysis of an internal standard (commercial cellulose) was better than 0.1‰ for carbon and better than 0.3‰ for oxygen. The δ 13 C series were corrected for the fossil fuel combustion effect (Francey et al., 1999).

| WUE i inferred from carbon isotope discrimination
Measurements of δ 13 C were used to calculate intrinsic WUE i for nondeclining and declining trees. WUE i is defined as the ratio between the net photosynthesis rate (A) and the stomatal conductance to water vapor (g s ) rate (Ehleringer et al., 1993) and computed as: where c a is the concentration of CO 2 in the atmosphere; c i is the concentration of CO 2 in the intercellular spaces; and 1.6 is the ratio of water and CO 2 diffusivity in the atmosphere.
For A. araucana, WUE i can be calculated starting from the δ 13 C of the plant material (δ 13 C tree ), which is related to atmospheric δ 13 C (δ 13 C atm ) and to the ratio c i /c a , according to Farquhar and Richards (1984) and Farquhar et al. (1982): where a is the fractionation factor due to 13 CO 2 diffusion through stomata (4.4‰), and b is the fractionation factor due to Rubisco enzyme during the process of carboxylation (27‰). Therefore, we can calculate c i by using the formula: Finally, the WUE i can be calculated as follows: However, the WUE i should not be considered as equivalent to instantaneous WUE, which is the ratio of assimilation (carbon gained) to transpiration (water lost) and takes into account the atmospheric water demand (Pacheco et al., 2020;Seibt et al., 2008). Thus, the equation used is the "simple" form of isotopic discrimination that does not include effects due to mesophyll conductance and photorespiration, which were not available for the study species.
We used δ 13 C atm values for the period 1850-2017 from Belmecheri and Lavergne (2020). We obtained the atmospheric concentration of CO 2 from the Mauna Loa station data (available at http://www.esrl.noaa.gov/) and for the period before 1850 from the ice cores obtained at Law Dome, East Antarctica (http://cdiac.ornl. gov/trend s/co2/lawdo me.html).

| Statistical analyses
To compare mean values of the analyzed variables (TRW, wood anatomy traits, WUE i , and δ 18 O) between the two vigor classes at each site we used the Mann-Whitney U tests and the paired-sample t test.
The statistical significance between tree growth, wood-anatomical traits and isotopes data were tested using Pearson correlations. We used the Wilcoxon rank-sum test to check if the changes through time of growth (TRW) and wood anatomical variables or isotope ratios differed between nondeclining and declining trees (Hentschel et al., 2014).

| Traits selection
PCA was performed with the chronologies of growth, and xylem anatomical traits were resulted for all the sites in two significant principal components (PC1 and PC2): together they explained 80.2%, 82.4%, and 83.1% of the variance for CHI1, CHI2, and ARG1 sites, respectively (Figure 2). The ordination along the two axis revealed recurrent patterns with three groups corresponding to (i) hydraulic traits (LA, LD, K h , D h , and CTA) showing high PC1 scores, (ii) growth traits (TRW and CN) showing high PC2 scores, and (iii) traits related to carbon use (CWT R AD , CWT TAN , CWT ALL and CWA) with low PC1 scores. According to these outcomes, we selected TRW, LA and K h , and CWT ALL as representative traits for each of these groups.

| Climate trends and drought variability
Annual precipitation showed a significant reduction ( Figure S2) at sites CHI1 and CHI2 from 1950 to 2017. At the same time, mean temperature showed a significant increase in the CHI2 and ARG1 sites located at the Andes range.
Drought severity significantly intensified in all sites (Figure 3).
The SPEI drought index showed that the driest coastal range site CHI1 was severely impacted by the 2010-2018 drought (Figure 3), whereas the wettest site ARG1, was mostly affected by the severe 1998 drought (Figure 3).

| Biometric parameters, growth, and wood anatomical traits
Declining trees presented significantly narrower rings than nondeclining trees (Table 2). Growth rate (TRW) usually showed higher mean intertree correlation (rbar) and common variance than woodanatomical variables such as LA, K h , and CWT (Table S2). We found a drastic growth reduction starting more than 120, 50, and 80 years ago in declining trees from CHI1, CHI2, and ARG1 sites, respectively ( Figure 4a). The TRW and its cell number were positively related.
Nonetheless, in all sites, trees of the two vigor classes were similar in terms of diameter, height, and age (Table 3) and also showed similar year-to-year variability in the TRW chronologies ( Figure S3).
During the 1800-2017 period, as for TRW, CN is significantly lower for declining trees in all sites (Table 3; Figure 4b), whereas tracheids in declining trees presented a clear smaller LA than nondeclining trees in CHI1 and ARG1 sites. This pattern was more evident at the driest site CHI1. In CHI2 site, LAs were rather similar.

F I G U R E 2
View of Araucaria araucana tree ring and illustration of the growth (a, TRW, tree ring width; CN, cell number along the radial direction) and anatomical traits (b, LA, lumen area; LD, lumen diameter, Dh, hydraulic diameter; Kh, theoretical hydraulic conductivity; CWTRAD, radial cell wall thickness; CWTTAN, tangential cell wall thickness, CWTALL, total cell wall thickness; CWA, cell wall area; and CTA, cell total area) considered. Scatter plots of the scores for the first (PC1) and second (PC2) principal components calculated on growth and anatomical traits chronologies for the CHI1, CHI2 and ARG1 sites (c1-c3) [Colour figure can be viewed at wileyonlinelibrary.com] Consequently, the theoretical hydraulic conductivity (K h ) was also lower in declining trees, indicating a theoretically lower water transport capacity than nondeclining trees (Table 3; Figure 4c). LA in the CHI2 site presented lower mean values compared with CHI1 and ARG1 sites (Table 3; Figure 4b,c).
Finally, CWT was thinner in declining trees at the CHI1 and CHI2 sites throughout the whole period, indicating a potential limitation in their carbon use to build cell walls. On the contrary, in the ARG1 site, we observed an increase in CWT starting in the 20th century (Figure 4d).
Considering all growth and anatomical traits, Pearson correlations were positive among traits related to LA (LD, LA, D h , CTA, K h ), but they were negatively related to cell-wall traits (CWT, CTA) ( Figure S4). These results were in agreement with the PCA (Figure 2).

| Growth and wood anatomical traits associations with climate variables
Growth and wood anatomical traits of the two vigor classes showed significant correlations with climate variables and SPEI in all study sites ( Figure S5). In general, TRW, LA, and K h were negatively correlated with growing season temperature (September-April), and positively related with summer precipitation and SPEI (December-February, Figure S5). At CHI1, growth (TRW) of declining trees was negatively correlated with maximum temperature from November to February and positively correlated with February precipitation in both tree groups. Similarly, December precipitation was positively related to LA and K h in declining trees, and this relationship was stronger in nondeclining than in declining trees. The same pattern was observed with SPEI during summer ( Figure S5).
At CHI2, TRW of nondeclining and declining trees was negatively associated with minimum and maximum temperatures during the growing season ( Figure S5). In this site, TRW, LA, and K h showed the weakest correlations with precipitation and SPEI ( Figure S5). At ARG1, TRW showed a positive association with precipitation (December and March) and growing-season SPEI. LA and K h of nondeclining trees were negatively related to the minimum temperature of February but positively related to precipitation of October, November, and February. LA and K h were correlated with growing season SPEI ( Figure S5).
Overall, CWT in nondeclining trees showed positive correlations with growing season maximum temperature, whereas in declining trees there were positive correlations with maximum temperature at the beginning of the growing season (October) in CHI1 and ARG1 sites. On the contrary, CWT showed a negative correlation with spring precipitation and summer SPEI, which was particularly strong in the ARG1 site ( Figure S5).
In CHI1 and ARG1 sites, the highest correlations between treering and anatomical variables and climate were found for SPEI ( Figure S5). Overall, LA and K h , showed a greater positive influence of growing season precipitation and SPEI at CHI1 and ARG1 sites, suggesting a higher sensitivity to soil moisture conditions. On the other hand, growing season temperature was positively related to TRW at the CHI2 site, suggesting a moderate water shortage constraining growth at this site.

| Patterns of WUE i and δ 18 O
In all sites, the increasing trend in WUE i for both tree vigor classes started in 1850 ( Figure 5). The increase of WUE i in the period was linearly associated with the increase in atmospheric CO 2 concentrations F I G U R E 3 Standardized 9-month SPEI at the three study sites (CHI1, CHI2 and ARG1) for the 1950-2017 period. Negative (red) and positive (blue) values indicate drier and wetter conditions, respectively [Colour figure can be viewed at wileyonlinelibrary. com] (p < .05). The mean WUE i derived in declining trees presented significant higher values than nondeclining trees in the drier CHI1 and CHI2 sites (Table 4). In the more humid ARG1 site, declining trees showed higher WUE i only at the beginning of the 19th century, but this pattern reversed after the 1980s, although the mean WUE i values in both vigor classes did not present significant differences (Table 4).
Declining trees in all the sites and for the whole 1800-2017 period featured significant higher δ 18 O values than nondeclining trees. This pattern was more evident in the CHI1 and CHI2 sites ( Figure 5).
The relationship between δ 13 C and δ 18 O was always positive for both vigor classes; however, the linear trend was not significant for both groups in the ARG1 site and just for the nondeclining trees in the CHI2 site. In the CHI1 site we found the strongest δ 13 Cδ 18 O association ( Figure 6).
The precipitation of February was positively correlated with δ 13 C in declining trees in all sites ( Figure S6). The δ 13 C and δ 18 O series of two vigor classes were negatively correlated with temperature in CHI2 and ARG1 sites during the growing season. Instead, in the CHI1 site nondeclining trees presented a negative correlation with δ 13 C during summer and a positive correlation with δ 18 O ( Figure S6).
Different letters indicate significantly different values (p < .05) between nondeclining and declining trees of the same site based on Mann-Whitney tests.
to similar stressful climatic conditions. Our findings confirm that the drought-induced dieback can be the final step to the very long aftermath of the hydraulic deterioration and growth declining process.

| Growth and wood anatomical traits: earlywarning signals of dieback
Through the retrospective examination of more than 200 years of growth and wood-anatomical traits, we showed that drought severity, expressed as SPEI, was the strongest climatic driver of tree growth, LA and K h long-term variability both for nondeclining and declining trees. It is well known that during the growing season cell expansion highly depends on water availability and cell turgor, and therefore a drought period can strongly affect this crucial phase of cell development (Cuny et al., 2018). Warm spring-summer temperatures had a positive effect on CWT, as already observed in conifers (Björklund et al., 2021;Carrer et al., 2018). Hence, the increase of drought severity and intensity will have a first detrimental effect with the reduction of LA that will reduce hydraulic conductivity and constrain water transport capacity and growth (Pellizzari et al., 2016;Puchi et al., 2019), predisposing to dieback and mortality (McDowell et al., 2008).
Nonetheless, not all the trees within a stand responded in the same way, and, currently, declining trees resulted more sensitive to drought producing narrower tree rings and tracheids of smaller LA.
The parallel drop of these two traits induced a dramatic reduction in the theoretical hydraulic conductivity (K h ). In addition, we demonstrated that the divergence in TRW and K h between nondeclining and declining trees occurred in all sites many decades earlier than the recent 2010-2018 drought (Garreaud et al., 2019), and before any sign of decline such as leaf shedding or canopy dieback Pellizzari et al., 2016). Therefore, these two traits could be used as useful early-warning signals of tree death for this species.
Such growth declines and the related loss in hydraulic conductivity anticipating a widespread forest dieback seems a recurring attribute at the driest distribution limits of many conifers sensitive to water shortage: a similar pattern was, for example, found in Pinus sylvestris and Abies alba in Spain (Pellizzari et al., 2016). In addition, previous studies in Patagonian temperate forests found a drastic reduction

| Isotope signals for assessing forest dieback
Araucaria araucana is an extreme isohydric species (Zimmer et al., 2015), i.e. this species reduces stomatal conductance rates as soil water potential decreases and atmospheric conditions dry (Brodribb et al., 2004;McDowell et al., 2008). As we expected, declining trees presented higher WUE i than nondeclining trees, but this was observed only at the driest Chilean sites. In Chile, declining trees had lower growth rates and K h , together with higher WUE i and δ 18 O. This indicates the closing of stomata to prevent water loss (Hentschel et al., 2014), but also incurs in an increased risk of carbon starvation (Gessler et al., 2018). The thinner cell walls of declining trees would agree with this interpretation indicating a lower carbon availability to build thicker walls. Indeed, although growth is not fully supplied by recent photosynthates, but also by carbon fixed in previous years Kagawa et al., 2006;Kuptz et al., 2011), several investigations confirmed that conifers rely much less and for less years on carbon reserves than broadleaves for xylem formation Soudant et al., 2016;Zweifel et al., 2006).
Furthermore, the simultaneous analysis of tree-ring carbon and oxygen isotopes may help understanding the ecophysiological responses of trees in different sites, in particular to discriminate whether changes observed in WUE i originated from a modification of A or g s (Barbour, 2007;Dawson et al., 2002). Moreover, the positive correlation between WUE i and δ 18 O found in the Chilean sites suggests that declining trees experienced very tight stomatal control of leaf water loss during drought periods (Drake & Franks, 2003;Moreno-Gutiérrez et al., 2012;Petrucco et al., 2017). Therefore, the δ 18 O increases in tandem with WUE i , should confirm stomatal closure as the key driver of growth decline in the dry Chilean sites (Scheidegger et al., 2000). However, the different δ 18 O values reported between declining and nondeclining trees could be linked to different water uptake ability, with declining trees being relatively better in exploiting the evaporative δ 18 O-enriched water from shallow soil layers (Battipaglia et al., 2010;Moreno-Gutiérrez et al., 2012). In contrast, nondeclining trees could uptake deeper water sources by forming deeper or more efficient root systems (Ripullone et al., 2020).
On the other hand, the WUE i trends of declining and nondeclining trees were similar in the ARG1 site, suggesting that here photosynthetic activity controls WUE i rather than stomatal closure (Scheidegger et al., 2000). The increase of CWT in declining trees in ARG1 and the decrease in K h would confirm this picture: here, trees could not grow or expand their tracheids, but they continued thickening their walls due to sufficient carbon availability, and increased their photosynthetic rates in response to drought. Moreover, these sites appeared as the mostly affected by the drought spell related to the 1998 drought. Following the conceptual model of Gessler et al. (2018)  poor plastic adaptation of the species under rapid climate change (Camarero & Fajardo, 2017;Camarero et al., 2018;Urrutia-Jalabert et al., 2015). Similar results were also found in the Erica arborea in Italy since this Mediterranean shrub species, under different soil moisture, highlighted different WUE i to cope with divergent water stress conditions (Battipaglia et al., 2014).
Recent studies across the Chilean Andes evidenced an increment in WUE i and δ 18 O in Nothofagus pumilio treelines, linked to reduce stomatal conductance, and associated to a reduction in growth (Fajardo et al., 2019;Lavergne et al., 2017). Likewise, long-term low growth and high WUE i in declining trees was found in P. sylvestris in Switzerland, where the reduction in carbon availability triggered dieback (Timofeeva et al., 2017). Another hydraulic strategy to survive has been observed in declining Pinus nigra trees in Italy after extreme drought. There, declining trees produced larger tracheids (more vulnerable to cavitation), lower WUE i (higher g s ), and high δ 18 O (high transpiration rates) than nondeclining trees (Petrucco et al., 2017), a similar pattern to that observed in some P. sylvestris declining trees in Spain (Voltas et al., 2013). In addition, in southern Italy, the same pattern, with lower δ 18 O values in nondeclining trees was observed in declining Quercus frainetto trees, and the authors considered hydraulic failure as the potential cause of dieback (Colangelo et al., 2017). Finally, the observed trends in tree-ring stables isotopes are seemingly associated to a twofold process: the common increase in WUE i is likely related to the global plant responses to the increasing atmospheric CO 2 concentration (Walker et al., 2020), but here the improvement in WUE i was more marked in declining trees from the drier sites indicating local, physiological responses. On the other side, the different responses in terms of δ 13 C and δ 18 O between declining and nondeclining trees across sites, with a stronger coupling in drier sites, together with the significant reduction in TRW and K h started several decades before dieback symptoms such as leaf shedding were observed (Figure 4), might represent a common ground in the long-lasting deterioration of their hydraulic functioning as already observed (Gessler et al., 2018;Pellizzari et al., 2016).
Despite the lack of common patterns, our study infers different site-contingent mechanisms of dieback and represents a new step forward for better understanding the xylem trait and physiological mechanisms that trigger drought induced dieback in temperate rainforests at long-term scales.

| CON CLUS IONS
This research confirms long-term retrospective quantifications of tree-ring growth, anatomical traits and tree-ring stable isotopes as relevant tools to forecast tree dieback. We proved this by comparing these variables in declining versus nondeclining A. araucana trees in Northern Patagonia. Dieback in this iconic South American conifer was triggered by the earlier deterioration in hydraulic functional traits, that leads to a reduce capacity of declining trees, with their lower K h , to withstand extreme droughts respect to nondeclining trees. However, this pattern was not observed in all sites and depended on local climate conditions. Retrospective wood anatomical and isotope analyses provide multiproxy frameworks to identify and forecast which species, tree populations, and individuals will be more prone to drought-induced dieback, and which will be more resilient in response to warming and future drying trends. This information will be particularly valuable for the conservation of endangered tree species as A. araucana, one of the most emblematic tree of South American temperate rainforests. Nonetheless, this research raises many questions about the role of microsite topography and soil moisture conditions that could determine the differences in tree water use strategies to cope with drought and mechanisms leading to drought-induced dieback. Future research, considering a larger number of trees and sites, should also investigate if genetic variability (e.g., Bansal et al., 2015), sex (e.g. Hadad et al., 2021), or microsite features (e.g., soil depth) could contribute to better explain which individuals and populations are able to tolerate and adapt better under stressful water shortage.

ACK N OWLED G M ENTS
We are grateful to Cristián Frêne, Alejandro Martinez Meier, and Anne-Sophie Sergent for the support in the collection of Araucaria araucana samples. We thank Raffaella Dibona and Michele Colangelo for helping with measurements and analyses of wood microsections. We thank Claudia Puchi for the cell design in Figure 2.
Paulina F. Puchi was also supported by The Rufford Foundation for Conservancy, project N°23677-1. We acknowledge Chilean Forest Service CONAF for permission N°851540 to collect the tree-ring samples in Altos de Pemehue National Reserve. We also thank the Argentinian National Parks Administration for permission to collect dendrochronological samples in Lanin National Park. Finally, we thank Forestal MININCO S.A. for the authorization to sample in Trongol Alto.