Spatial and evolutionary predictability of phytochemical diversity

Abstract

To cope with environmental challenges, plants produce a wide diversity of phytochemicals, which are also the source of numerous medicines. Despite decades of research in chemical ecology, we still lack an understanding of the organization of plant chemical diversity across species and ecosystems. To address this challenge, we hypothesized that molecular diversity is not only related to species diversity, but also constrained by trophic, climatic, and topographical factors. We screened the metabolome of 416 vascular plant species encompassing the entire alpine elevation range and four alpine bioclimatic regions in order to characterize their phytochemical diversity. We show that by coupling phylogenetic information, topographic, edaphic, and climatic variables, we predict phytochemical diversity, and its inherent composition, of plant communities throughout landscape. Spatial mapping of phytochemical diversity further revealed that plant assemblages found in low to midelevation habitats, with more alkaline soils, possessed greater phytochemical diversity, whereas alpine habitats possessed higher phytochemical endemism. Altogether, we present a general tool that can be used for predicting hotspots of phytochemical diversity in the landscape, independently of plant species taxonomic identity. Such an approach offers promising perspectives in both drug discovery programs and conservation efforts worldwide.

Keywords: alpine habitat; chemical ecology; diversity hotspots; landscape ecology; plant secondary metabolites.

Fig. 1.

Phylogenetic, abiotic, and biotic factors together predict phytochemical richness across species. (A) Frequency distribution of unique molecular families across all plant species sampled. (B) Number of unique molecular families as a function of the increasing number of species sampled. The gray shaded area represents the variance around molecular families’ prediction based on 1 million randomly sampled plant species assemblages covering 2 to 100 plant species. The different colored lines represent the average relationship for conditioned sampling based on three different elevational bands (Low = 400–900 m a.s.l.; Mid = 900–1,600 m a.s.l., and High = 1,600–3,000 m a.s.l.). Finally, the red dotted line represents the same sampling but constrained by phylogenetic proximity of species. (C) Phylogenetic tree of all plant species sampled (n = 416), where branch tips are colored by molecular family richness (low: blue, high: red). (D) Association between observed and predicted values of molecular family richness (Pearson correlation: n = 416, r = 0.49, P < 0.001) derived from an ensemble machine learning algorithm with variable selection. (E) Variable importance for the same machine learning algorithm, colored by variable class (red: phylogenetic variables [Phylo], Nn = number of intervening nodes, C1/C2 = first/second axes of cophenetic distance matrix; yellow: soil variables, pH = soil pH 95% quantile, Rh = soil moisture 5% quantile; blue: climate variables [Clim], Te = average annual temperature 95% quantile, Ra = median solar radiation 95% quantile; green: vegetation productivity [Prod], ND = mean NDVI 95% quantile).

Fig. 2.

Geographic mapping of phytochemical diversity. (A) A geographic representation of phytochemical diversity across ∼20,000 km² in Switzerland. Colors are as described for B, with white and black additionally showing snowy summits and lakes, respectively. (B) PC1 and PC2 scores from a PCA of phytochemical diversity data extrapolated from a random sample of 20,000 geographic pixels (∼100-m resolution) from A. Arrows represent loadings for the environmental variables annual mean temperature (Temperature), annual precipitation sum (Precipitation), annual sum of solar radiation (Solar radiation), topographic position index (aspect, or hills versus depressions in the landscape), topographic roughness index (slope and terrain roughness), topographic wetness index, inner forest density (Forest height Q25), mean NDVI, soil moisture, and soil acidity. Point colors represent phytochemical richness (low: blue, high: red).

Fig. 3.

Elevational distributions of phytochemical families. (A) Optimal distribution of 6,012 phytochemical families found in the Swiss Alps along the elevation (km a.s.l.). Distribution is based on the niche of each plant species identified through spatial modeling plus the presence of the molecule in the plant species. Colors reflect probability of occurrence (red, maximal [1]; blue, absent [0]), with the dotted white line showing the average probability of occurrence considering all molecular families. (B) Inferred major classes of metabolites ordered by average elevation optima, with the gray line representing the global median value.