Our dataset of 1770 woody plant species, supplemented by the precise information on latitude, longitude, elevation, temperature and precipitation of 13,028 species-site combinations (Methods), resembles the global distribution of species-specific differences in microscopically assessed DCWL (Fig. 3a). DCWL is a dimensionless index of the fraction of lignified cell wall area relative to the total cell wall area, which ranges from ‘0’ in almost completely blue stained (near un-lignified) anatomical images to ‘1’ in almost completely red (near fully lignified) images. Our approach allows the extent of cell wall lignification to be estimated over undisturbed areas of stem cross-sections that include all different cell and tissue types, such as fibres, parenchyma, and vessels. While upright growing stems of trees and shrubs always exhibit some DCWL (Fig. 3b), the stems of much smaller herbs range from nearly un-lignified to almost fully lignified. Evidence for higher DCWL in trees and shrubs, and a much more variable DCWL in herbs, is further exhibited by each of the major plant families (Fig. 3c). Though herbs with low DCWL are common, DCWL increases with plant height, and trees taller than circa ten metres always have an estimated DCWL above 0.5 (Fig. 4a). Most importantly, poorly lignified trees and upright growing shrubs cannot exist, because their self-supporting stems must be able to cope with static and dynamic stressors, as well as long-distance root-to-leaf water transport.Fig. 3

a Black dots show the location of the 198 sampling sites on six continents between 40°S and 80°N, and between sea level and around 6150 m asl, superimposed on corresponding species’ occurrence data (green dots) obtained from the Global Biodiversity Information Facility (GBIF)⁵¹. b Violin plots describe the distribution of DCWL in trees and shrubs (dark blue), as well as in herbs (herbaceous dicots; orange), which ranges from 0 to 1 (from almost un-lignified to almost completely lignified). The shape of the violin plot refers to the proportion of data. Black boxes within each violin show the interquartile range and whiskers the 95% confidence intervals. Red is the mean and white dot is the median of the data. Three asterisks (***) indicate significant (p < 0.001) differences of a paired-sample t-test between the mean DCWL in trees/shrubs and herbs (scores for trees and shrubs: mean = 0.81, standard deviation = 0.17 and for herbs: mean = 0.50, standard deviation = 0.31; t (25.7) = 1115.1). c Box and whisker plot of DCWL of the 17 plant families with more than 25 species split into trees and shrubs (dark blue), as well as herbs (orange). Box width is proportional to the number of species, box length indicates the interquartile range, whiskers the 5th and 95th percentiles, and horizontal black lines the median.

The herein observed variability of DCWL in the stem sections of herbs contradicts the common belief that small plants are always poorly lignified. This finding also challenges the misperception that herbs are non-woody plants. Although one would expect smaller plants to experience less mechanical strength and thus to require less lignified cell walls, this size-driven assumption is not always true. Different herb species can adapt to different environmental conditions^29,30, and the various roles lignin play for their wood, such as resisting negative pressure in the vascular system in dry areas, is reflected by the observed DCWL variability in small plants. DCWL depends on both, the anatomical proportion and arrangement of different wood cell types, as well as their cell wall chemical composition, which may differ within and between plants (Figs. 1 and 2). Starch-storing parenchyma cell walls always stain blue in the sapwood, while DCWL may vary considerably between fibre cells and conducting vessels that provide structural and hydraulic functions, respectively. The wood anatomical diversity of our dataset includes a wide range of plant species and site conditions (Figs. 3 and 4), such as the fibreless wood of some small herbs, in which the mechanical function is relocated to thick-walled vessels (e.g., Pulsatilla), or the heavily lignified and thick-walled fibre-rich wood of other herbs (e.g. Convolvulus) (Fig. 4b).

DCWL in terrestrial plants is generally lower at sites with cold growing season temperatures (Fig. 5a), which are usually located towards higher elevations and higher latitudes. This pattern is most pronounced in herbs but not in trees and shrubs that were not found at the network’s climatic extremes (Fig. 5b, c). While similar results have been reported for gymnosperm wood at much smaller spatial scales³¹, systematic large-scale evidence has been missing so far. To account for the potential effect of an increasing proportion of herbs towards colder sites, we analysed the relative contribution of elevation, latitude, temperature, precipitation, and plant height for DCWL (Supplementary Table 1). Correlations between DCWL and the climatic variables were quantified for all species and each lifeform separately (Supplementary Table 2). Each variable significantly affected DCWL (except for elevation, where the net effect is only marginally significant). The strongest effect was found for temperature. No relationship was found between phylogeny and the elevation and latitude of our samples because those plant families growing at extreme high-elevation and high-latitude sites (e.g. Asteraceae, Brassicaceae, Saxifragaceae and Caryophyllaceae) also include individuals from less extreme environments (Supplementary Fig. 1). When species occurrences are plotted against annual precipitation totals and mean growing season temperatures (Supplementary Fig. 2), less lignified, small plants are predominantly found in colder climates, especially where snow cover persists for most of the year. In contrast, more lignified herbs, shrubs, and trees are widespread across warmer climates, with overall higher DCWL at lower elevations and lower latitudes where growth conditions are favourable (Fig. 5). The relationships between DCWL and temperature, elevation and latitude are confirmed by subsets from natural habitats, all samples from botanic gardens, and all herbs of smaller ten cm vertical height (Supplementary Fig. 3). Our results imply a generalised picture of temperature-induced DCWL distribution supportive of a new theory about the global treeline position (Fig. 6). Under warmer climates at lower elevations and lower latitudes, the stem cell walls of trees and shrubs are always lignified. However, the stem cell walls of herbs are generally less lignified at high elevations and high latitudes, which is particularly the case when herbs exceed the potential treeline position (Supplementary Fig. 4).Fig. 5

The role temperature, elevation and latitude play for DCWL in a all species, b herbs and c trees and shrubs (black dots), with the red lines being locally weighted polynomial Loess Regressions fitted to mean DCWL values. Green dots represent species-site combinations, while black dots are mean values of DCWL computed in each class of 1 °C temperature of the warmest quarter, of 100 m elevation, and of 2° latitude. Similar results are obtained when plotting the data separately for specimens collected in natural growing conditions, in botanic gardens, and for herbs smaller than 10 cm in vertical height (Supplementary Fig. 3).

We quantify the relative degree of stem cell wall lignification (hereinafter DCWL) in 1770 different woody plant species from 198 disjunct sampling sites (Fig. 3a). Species collection was carried out over the last 30 years from natural and undisturbed sites, as well as from botanic gardens and cultivated lands, mostly across the Northern Hemisphere. At each sampling site, we selected adult individuals that appeared undamaged and healthy. During harvest, we sampled one adult individual per species that was usually the tallest individual of that species at that site and then recorded the plant height of that individual, as well as elevation and coordinates of the sites (from a GPS device) and sites name (from the nearest geographic feature). Small plants were dug out from the ground to expose the root collar, whereas for shrubs and trees, a stem disk was cut with a saw, or a wood core was extracted with an increment borer from the upright stem (only from the main stem in multi-stem shrubs) to avoid tension wood. The sampled portions of stem and cores were stored in sealed plastic bags to which we added several drops of 40% ethanol and kept it at 3–4 °C until the samples were sectioned. We defined the plant lifeform as a functional type, independently from any taxonomic unit. While all trees and shrubs are self-supporting, herbs comprise herbaceous flowering plants other than grasses⁴⁸. All terrestrial plants in our dataset are dicotyledonous angiosperms capable of secondary growth that were collected on six continents either in their natural habitats or in botanic gardens. Considering only samples from one mature individual per species, the vertical plant heights of our collection range from 2 cm to around 40 m, and the specimens vary in age from annual plants to several decades⁴⁹. Though weighted towards Europe and North America, our sampling sites cover the extreme margins of plant growth at around 6,150 m asl in the Himalayan Ladakh¹¹, as well as in coastal east Greenland circa 73° North¹⁰. A large Saxifraga oppositifolia L. cushion plant from circa 4500 m asl in the central Swiss Alps most likely represents one of the coldest sites of the collection⁵⁰.

Our dataset is dominated by herbs (78.7%), followed by trees (16.9%) and shrubs (4.4%). We used the Global Biodiversity Information Facility (GBIF)⁵¹ to extract actual distribution coordinates of the 1770 plant species collected and analysed (Fig. 3a). Species-specific geographical information was based on occurrence data available from GBIF. A maximum of ten occurrence data points was downloaded for each species via the R package ‘spocc’ software⁵². Bioclimatic attributes and elevation data were downloaded from the WorldClim Global Climate Data⁵³ at 2.5 minutes resolution. For each species occurrence data point, 12 BIOCLIM variables, representing temperature and total annual precipitation attributes of a given locality, as well as elevation, were extracted using the R package ‘raster’ software⁵⁴. For each occurrence point, we estimated elevation, as well as annual and seasonal temperature means and precipitation totals. Covering all climatic zones and terrestrial biomes, the resulting GBIF network of 13,028 species-specific locations is denser in western Eurasia and North America compared to all other continents. The 1770 woody plant species from 118 families were classified into two main lifeforms: (i) trees, shrubs and dwarf shrubs, as well as (ii) herbs that comprise herbaceous flowering plants with vascular cambium. Ferns, grasses and graminoids, as well as climbers, stem succulents, epiphytes and aquatic plants, were not included in this study.

Wood samples from the outermost mature and functional stem tissues (Figs. 1 and 2), close to the root collar, were prepared for wood anatomical analyses. Stem sections for anatomical observation were cut to include fully developed secondary wood and the outermost complete growth ring, though excluding the cambial region and dead cells of the heartwood. Samples enable the estimation of DCWL free from ongoing xylogenesis processes and only represent the condition of the physiologically active xylem tissues. Thin sections were bleached with sodium/potassium hypochlorite to decolour cell walls and remove undesired cell content. Thin sections were double-stained for 3–5 min with a 1:1 blend of freshly prepared Safranin (1% in water) and Astra Blue (0.5% in water with 0.2% acetic acid), which is known to stain lignified cell walls red and less-lignified cell walls blue^55,56. Although the absolute changes in lignin content that are visualised by the double-staining method remain unknown, there is evidence for qualitative similarities between double-staining, autofluorescence imaging and Raman spectroscopy^37,58. In contrast to polarised light microscopy that does not visualise lignin, the staining with Safranin and Astra Blue can reveal different degrees of lignin content in wood cell walls⁵⁷. Thin sections were washed with water and dehydrated with ethanol at increasing concentration, finally embedded in Canada balsam under the cover glass. Permanent thin sections were dried for 12–24 h at 60 °C. One microscopic image of the outermost wood, excluding still developing and not yet functional cells near the cambium and dead cells of the heartwood (Fig. 2), was digitised at 100x or 200x magnification under a transmitted light optical microscope. White areas reflect cell lumina and intercellular spaces, whereas the red parts indicate more, and the blue parts indicate less lignified cell walls, respectively. A dimensionless area index of DCWL was calculated to quantify the fraction of lignified (red-stained) cell wall area relative to the total cell wall area. DCWL ranges from ‘0’ in almost completely blue stained (near un-lignified) anatomical images to ‘1’ for almost completely red (near fully lignified) images. Our staining technique captures the lignin and holocellulose components of the secondary cell walls of all cell types in each wood sample⁴⁷. Standard wood anatomical features of each species were previously described^58–60. Sections were imaged using a digital camera mounted on a Nikon Eclipse E400 compound microscope, equipped with an Olympus C5050 camera. Before picturing the wood sections, we adjusted the white balance on the camera to avoid variations in colour hue from picture to picture. Magnification enabled a representation of the general xylem features for the cross-section of the species and excluded cracks or artefacts on the slide, such as transverse (end) walls in axial parenchyma cells as visible in the cross-section.

In each anatomical picture, the area occupied by cell lumina and intercellular spaces (white), as well as the area occupied by lignified (red) and less lignified (blue) cell walls, were estimated with the image-analysis programme ImageJ⁶¹. The colour area in a picture was measured using the macro colour analysis setting colour thresholds for hue, saturation, and brightness of each colour range (red, blue and white) (Supplementary Table 3). The three colour ranges capture the variability within each colour (e.g., 'red' capture all shades from light to dark red) while ensuring a clear separation from the other two. Colour settings were defined by direct observation below the microscope and comparison with image analyses output. To calculate the degree of cell wall lignification (DCWL), we divided the lignified cell wall area (red) by the total cell wall area (red and blue): DCWL = (Lignified cell wall area (red))/(Total cell wall area (red and blue)). The degree of cell wall lignification (DCWL) distribution was graphically visualised within each lifeform and geographically visualised on an elevation vs latitude plot. Relationships between DCWL, plant height and temperature of the warmest quarter, elevation and latitude were graphically represented. We investigated the relative importance of abiotic (elevation, latitude, temperature of the warmest quarter) and biotic (plant height) control of DCWL variability in trees, shrubs, and herbs. Relative abundance data of trees and herbs were subjected to variance partitioning using partial Redundancy Analyses. This analysis was performed to overcome the problem associated that rare species (e.g., trees beyond treeline) might have a much larger influence on the analysis than more common species. Three quantitative environmental variables extracted for each occurrence point-species combination (elevation, latitude, temperature of the warmest quarter) and plant height measured from the collected plant were used as explanatory variables. The significance of the variables was tested with a Monte-Carlo permutation test performed in CANOCO 5.0 software^62,63 (Supplementary Table 1). All tests were done using 999 permutations and a significance threshold of p < 0.05. Both the dependent and explanatory variables were mean-cantered and variance-scaled prior to the analyses. For data visualisation, we plotted DCWL vs elevation, latitude, and temperature of the warmest quarter with each of the analysed variables divided into classes (ten cm plant height, 1° Celsius degree temperature, 100 metres elevation, and 2° latitude) and a mean DCWL value was computed in each class. Loess Regression represents a locally weighted polynomial curve fitting the mean values of each variable and visualised the trends. Pearson’s correlation coefficients between DCWL and the individual bioclimatic variables were quantified for all species and for each lifeform separately (Supplementary Table 2).

Further information on research design is available in the Nature Research Reporting Summary linked to this article.