Field sampling was carried out in 2012 in the Campolino Natural Reserve (northern Apennines) (Table 1), which is the southernmost known autochthonous population of Norway spruce [Picea abies (L.) H. Karst.] in Italy (Chiarugi, 1936; Magri et al., 2015; Avanzi et al., 2019). Three plots were established along an altitudinal transect on the same slope. The high altitude plot (CAM-H, ∼1,730 m a.s.l.) is a pure, dense and relatively young spruce forest recolonizing an old pasture. The intermediate plot (CAM-E, ∼1,615 m a.s.l.) is a mature mixed forest with Norway spruce, silver fir (Abies alba Mill.) and European beech (Fagus sylvatica L.). The lowest plot (CAM-L, ∼1,475 m a.s.l.) is a similar mixed forest, but with less spruce (Table 1). In each plot sampled, trees were labeled and georeferenced with a geographic positioning system device (GPS; Garmin, 2 m resolution). Stem diameter at breast height (DBH) and tree height were measured, and one increment core was collected with a 5 mm Pressler borer at DBH in cross-slope direction to avoid compression wood.

The three plots fall in the same spatial grid, so monthly mean temperatures and total precipitation for the 2008–2012 period were obtained from the CRU TS V.4.0 (land) 0.5° grid (KNMI Climate Explorer¹).

A total of 10 trees per plot (Table 1, Supplementary Figure S1) were selected to maximize variation in tree height, age and DBH (Supplementary Figure S2). Wood microsections were prepared to collect high-resolution anatomical images following the standard protocols (Gärtner and Schweingruber, 2013; von Arx et al., 2016; Prendin et al., 2017). From each core, 12–15 μm thick cross-section from the last five tree rings (dated 2008–2012) were cut using a WSL lab-microtome (Gärtner et al., 2015a, b). All sections were double-stained using Safranin and Astra blue, dehydrated with ethanol, and permanently fixed on a microscope slide with Canada balsam. Overlapping images of cross-sections were captured at 100 times magnification using a digital camera (Canon EOS 650D, Canon Inc., Tokyo, Japan) through an Olympus BX41 microscope (Olympus Corp., Tokyo, Japan). PTGUI v10 (New House Internet Services B.V., Rotterdam, NL, United States) was used to stitch single images and produce a high-resolution composite image (2.36 pixels/μm). These images were then processed with the software ROXAS (v3.0.88, von Arx and Carrer, 2014), which provides a large set of measurements of tree-ring width (TRW) and wood anatomical traits (Table 2). Image analysis and anatomical data collection was restricted to a width of one tangential millimeter within each tree ring.

The wood anatomical traits used in this study were direct measurements and derived from ROXAS output (Table 2). The analysis was performed within a tree ring, and in the earlywood (EW) and latewood (LW). TRW, EW and LW width (EWW and LWW), and their percentage (EW% and LW%) calculated as the ratio of EWW (LWW) to TRW, as well as their ratio (LW/EW) were measured. To tell apart EW and LW tracheids, ROXAS automatically computes the Mork’s index which is the ratio between twice the double CWT and the lumen diameter, both measured in radial direction (Denne, 1988) (Table 2). Additionally, the ratio between radial and tangential diameter of tracheid lumen was applied to discriminate between EW (ratio > 1) and LW (ratio < 1) tracheid shape. The number of tracheids in tangential direction was manually counted in the LW, and tracheid number in radial direction was computed by dividing the total number of tracheids in each tree ring (ROXAS output) by the number of tracheids in tangential direction within 1 mm strip. Wood anatomical traits were grouped in traits related to (i) tree-ring growth, (ii) cell growth, (iii) hydraulic efficiency and hydraulic safety, and (iv) mechanical support and carbon investment (Table 2).

Tree-ring width was positively correlated with the number of tracheids per radial row (r = 0.99). TRW is EW width dependent (R² = 0.993, P < 0.001), showing on average an EW% higher than 80% (Figure 2A). TRW was also positively related to the ratio between radial and tangential diameter (Drad/Dtan) of tracheid lumen (R² = 0.435, P < 0.001) (Figure 2B). The lumen size maintained a proportional increase in radial and tangential diameter only in EW but not in LW tracheids (Supplementary Figure S3).

Because tree height measurements prior to the year of sampling were not available, only Drad and Dtan of the outermost ring (2012) can be related to tree height. Dtan was more related to tree height than Drad, and these relationships were strongest for EW (EW Dtan: R² = 0.536, EW Drad: R² = 0.390, and LW Dtan: R² = 0.357, P < 0.001) (Figure 3A). The Drad/Dtan scaled with tree height: taller trees had proportionally larger Dtan in EW (R² = 0.530, P < 0.001), meaning that the tangential diameter increased more with tree height than Drad (Figure 3B). As a consequence, the tangential number of tracheids was negatively related to tree height (R² = 0.473, P < 0.001).

Tracheid diameter affected both cell density (CD), the number of tracheids per mm², and their lumen area (LA). Both CD and LA were related to tree height but with opposite directions: CD decreases with tree height, while LA increases (Supplementary Figure S4).

Both mean hydraulic diameter (Dh) and specific hydraulic conductivity (Ks) were positively related to tree height (Supplementary Figure S5) while, theoretical hydraulic conductivity (Kh) was not. Although Kh is positively related to Drad (R² = 0.503), total number of tracheids (CNO) (R² = 0.556), EW% (R² = 0.615), basal area increment (BAI) (R² = 0.672), TRW (R² = 0.707), and EW width (R² = 0.722) (all relationships are significant, P < 0.001) (Figures 4A,B). In addition high EW% corresponded to larger values of “hydraulic carbon use efficiency” index (HCUE) (R² = 0.600, P < 0.001) (Supplementary Figure S6). HCUE increased linearly with TRW (R² = 0.772, P < 0.001) and decreased exponentially with age (R² = 0.358, P < 0.001), meaning that younger trees had a high carbon investment per unit of water conductivity.

Earlywood percentage (LW%) had a negative (positive) effect on cell wall reinforcement index measured in EW tracheids ([t/b_₀₅]²) (Figure 5A). Trees with denser wood (due to larger LW%, R² = 0.638, P < 0.001 – Supplementary Figure S7) showed higher values of [t/b]² in the EW portion (Figure 5B). The relative wood density (RWD) was also negatively related to Drad and Dtan (R² = 0.618 and R² = 0.316, respectively, P < 0.001) (Supplementary Figure S7), as well as positively related to CWT (R² = 0.526, P < 0.001).

Cell wall thickness in radial cell walls was positively correlated with the CWT in tangential cell walls (Figure 6). In the LW portion of the ring, this relationship showed a higher intercept, as opposed to the 1:1 relationship in the EW, with the increase in tangential direction stronger than in the radial direction. A weak but positive relation between CWT and tree age was also found (R² = 0.281, P < 0.001), whereas no relation was observed between tree height and CWT.

A PCA on wood anatomical traits relative to the 2012 tree ring showed that the first three PCs explains the 36.8, 28.5, and 13.3% of variation, respectively (Figure 7). The hydraulic traits Kh and HCUE were correlated positively with PC1, while [t/b]² and [t/b_{_05}]² negatively (Figure 7, Supplementary Figure S8). The remaining hydraulic traits (Dh and Ks) were positively correlated with PC2, meaning that they were orthogonal with respect to the other hydraulic traits. All mechanical and carbon-allocation traits (Table 2, Supplementary Figure S8) showed a negative correlation with PC1, whereas PC2 was highly correlated with some wood anatomical traits connected to tangential diameter in both EW and LW, and the LA of the EW portion (Figure 7, Supplementary Figure S8).

The results of the multiple regression analysis between the first three PCs and tree height, age and plot were reported in Supplementary Table S1. Scores of PC1 increased with tree height (ß = 0.470, P = 0.031) and decreased with age (ß = −0.082, P = 0.006; Supplementary Figure S9A). Scores of PC2 increased with tree height (ß = 0.546, P < 0.001; Supplementary Figure S9B). Scores of PC3 increased with tree height (ß = 0.218, P = 0.037) and were significantly determined by the plot (P = 0.006; Supplementary Figure S9C). Scores of PC3 were significantly different between CAM-H and CAM-L (adj P = 0.023). Pairwise geographic distances and pairwise differences calculated on PC scores were positively correlated only in CAM-H (r = 0.357, P = 0.019) (Supplementary Figure S10). This means that, in CAM-H, trees that are spatially closer to each other have similar wood anatomical traits.

Principal component analysis was performed on the dataset relative to each tree ring (from 2008 to 2012) to test the stability of wood anatomical traits through time. Procrustes tests show consistently that individual PC scores were highly correlated from year to year (r ≥ 0.8, P = 0.001; Figure 8A). Only 22 out of 65 anatomical traits displayed a mean CV larger than 10% and a range of individual CVs larger than 20% (Figure 8B). The largest inter-annual and inter-individual variation was displayed by [t/b]² and LA of the 1st and 5th percentiles (LW portion) (Table 2, Figure 8B).

Besides confirming that taller trees have larger tracheids in the basal portion of their stem, our results take the discussion forward. EW tracheids in tall trees become larger by increasing both in radial and tangential directions proportionally. However, the tangential diameter of tracheids is partly constrained by the diameter of cambial mother cells and by adjacent tracheids. Despite previous works confirming that tracheid size increases primarily in the radial direction, whereas the tangential diameter and length remain almost constant (Skene, 1972; Cuny et al., 2014), here we suggest that the radial tracheid diameter, and even more their tangential diameter, are primarily influenced by tree height and to a lesser extent by the altitudinal position along the transect. This confirms recent results on conifers growing at the treeline: once the LA of EW tracheids is standardized for tree height, this trait is not related to temperatures (Kašpar et al., 2019).

All 65 wood anatomical traits in this study are grouped according to their wood characteristics and function in tree-ring growth, cell growth, hydraulic efficiency and safety, and mechanical support and carbon investment. PCA appears as an efficient analysis for summarizing the information embedded within a large set of wood anatomical traits and shows a high degree of variation both between and within plots (Supplementary Figure S11). We identify two main wood anatomical “strategies” associated with the first and second PCs, respectively (Figure 6). On the one hand, high PC1 scores describe trees with high EW percentages and hydraulic conductivity, whereas low PC1 scores describe trees with high LW% and high RWD, thicker tracheid walls, and high cell wall reinforcement index, traits linked to hydraulic safety. On the other hand, PC2 describes whether each tree invests in tracheid number or in their lumen diameter. High PC2 scores identify trees with few large tracheids, while low PC2 scores identify trees with numerous small ones. Since PCs are orthogonal, this second anatomical strategy is independent to the first one. PC1 is negatively related to tree age showing that older trees tend to have a higher proportion of LW, thicker cell wall in both radial and tangential direction, and higher LW RWD than younger trees, which may limit their water transport capacity, increase the investment in carbon and increase their resistance to embolism (Domec and Gartner, 2002; Rosner et al., 2014) as explained in the following section. PC2 is strongly, positively related to tree height, meaning that taller trees tend to have fewer but larger tracheids to maximize water conductivity per unit area. Tall trees have a high hydraulic tracheid diameter and Ks at breast height (Rosner et al., 2008; Anfodillo et al., 2012; Olson et al., 2014), whereas the Kh is mainly related to the amount of EW: the higher the EW percentage, the higher is the conductivity.

Recent studies have shown a plant size-dependent sensitivity to drought (Bennett et al., 2015; Fajardo et al., 2018; Stovall et al., 2019). Bennett et al. (2015) found that drought has more detrimental impacts on the stem diameter growth of large trees. Stovall et al. (2019) reported that tree height is the most important predictor of tree death during drought episodes, with tall trees (>30 m) dying at a rate of more than double than short trees (<15 m). However, tall or large trees are not always old trees, and the lack of such correlation depends on species characteristics (e.g., shade-tolerant vs light-tolerant species) and within-population structure, such as local site conditions, tree’s disturbance history, growth form, as well as individual genotype (Lindenmayer and Laurance, 2016). Finally, the site conditions can play an important role in wood anatomical trait variability. A significant plot-effect is found on PC3; in particular, a statistically significant difference between PC3 scores of CAM-H and CAM-L. This difference may suggest an altitudinal effect as these two plots are the most contrasting in terms of elevation (1,730 vs 1,475 m). However, PC3 explained just 13.3% of the total inter-individual wood anatomical variation, suggesting that the environment is, in any case, less relevant than tree height and age in affecting wood anatomical trait variation. This could be partially related to the limited extension of the altitudinal gradient explored (∼300 m) that, however, encompasses the elevation gradient of spruce in the study area. For this reason, it could be worth exploring the inter-individual variation of wood anatomical traits along larger and replicated altitudinal transects and in more limiting environmental conditions. A weak but statistically significant spatial effect is found in CAM-H, where neighbor trees display more similar anatomical traits. In this case, spatial proximity is considered as a proxy for micro-environmental heterogeneity. Avanzi et al. (2019) demonstrated the relevance of micro-environmental heterogeneity of growth patterns on a larger dendrochronological dataset of trees sampled in the same plots. In this study, we analyzed 10 trees per plot, not a high number for fully capturing the micro-environmental heterogeneity.

Our analysis on inter-annual variation of wood anatomical traits investigates which traits are likely to show no or small changes through time. PCA results are highly correlated among years, meaning that overall trees have similar wood anatomical traits in their last formed five rings. This result emphasizes, over a limited temporal window, how trees exhibit globally coherent wood anatomical patterns, even with different mean temperatures and total precipitation of the growing season (Supplementary Figure S12). Nonetheless, the coefficients of variation for each single wood anatomical trait reveal that the magnitude of such inter-annual variation changes among traits and individuals. We found that the cell wall reinforcement index of the entire ring, the LA of LW, and LW% are the most variable traits both at inter- and intra-individual level. LW is probably the wood anatomical trait mostly involved in potentially adaptive responses to climate and environmental changes. According to Larson (1969), EW to LW ratio is influenced by tree crown size that, in turn, depends on stand density and competition. When growing conditions are ideal, height growth is enhanced and the LW% is lowered (Domec and Gartner, 2002). In our study site, the EW percentage is on average >80%, pointing toward a weak effect of climate on individual tree growth dynamics, as shown by a climate-growth correlation analysis performed on a larger dataset of trees sampled from the same plots (Avanzi et al., 2019).

Latewood percentage affects RWD and also the hydraulic properties of coniferous wood (Hacke et al., 2001; Domec and Gartner, 2002). In Norway spruce, EW tracheids are the most effective water-conducting pipes with higher numbers of pits per conduit length unit compared to LW tracheids (Rosner, 2013). Wood density shows moderate to high values of heritability in conifers (e.g., Hannrup et al., 2004; Martinez Meier et al., 2008; Chen et al., 2018) and, despite that this trait is not directly linked to hydraulic vulnerability, it is involved in mechanisms of drought tolerance and hydraulic failure, with higher wood density being associated with lower vulnerability to cavitation (e.g., Dalla-Salda et al., 2009; Rosner, 2013; Rosner et al., 2016; Fernández et al., 2019). As in previous studies (Bouche et al., 2014; Bouriaud et al., 2015), the present results show that RWD in LW is more related to tree age than to tree height, and that trees with higher RWD in LW feature a higher cell wall reinforcement index. We used the cell wall reinforcement index as a proxy for hydraulic vulnerability (e.g., Hacke et al., 2001; Rosner, 2013), concluding that trees with high RWD are more resistant to tracheid embolism. Studies on LW, EW, and transition-wood tracheids suggest that, when a water-deficit episode takes place, cavitation first occurs and expands in the LW tracheids. Only when water deficit further increases, embolism spreads in the EW and finally reaches the transition wood (Dalla-Salda et al., 2014). However, the loss in conductivity is not related to where the initial cavitation occurs, but to the wood anatomical structure (e.g., RWD and EW to LW ratio) that shapes the species vulnerability to drought events (Luss et al., 2019).

In conclusion, our findings seem to consistently demonstrate that it is crucial in any wood anatomical, ecological, and evolutionary genetics study to record and consider tree height to correctly investigate inter-individual wood anatomical variation. By doing so, wood anatomical traits can be standardized based on tree height as opposed to tree diameter which is not always directly related to tree age. Additionally, we also consider crucial to account for age-related trends that might affect wood anatomical traits with mechanical support functions, such as CWT, RWD, and lumen diameter of LW tracheids. Such traits can ultimately affect tree resistance to embolism. We recommend that analyzing wood anatomical traits within few tree rings but from a large number of individuals is a good method for capturing most of the variance in both wood anatomical traits and microenvironmental conditions. This strategy may avoid the “noise” created by difference in tree height, and age. Finally, the creation of a common garden would be extremely useful to determine whether and how individual genotypes drive wood anatomical variation, minimizing microsite heterogeneity.