Metabolic processes shape microbial interaction distributions
Microbial communities are largely driven by metabolic interactions, whereby species compete for shared resources and exchange byproducts. Such interactions are commonly classified by their sign alone. Yet, theory shows that the strength of interactions and the shape of their distribution impact how many and which species coexist. What this shape looks like in microbial communities and why has rarely been examined. Starting from a consumer-resource model of resource competition and cross-feeding, we find that these metabolic processes generate a skewed distribution with "many weak, few strong" interactions - a pattern previously documented in the food webs of larger organisms. This skew emerges whenever species have different substrate preferences and when metabolite leakage is sufficiently high. Across nine microbial datasets spanning diverse community origins, empirical interactions not only qualitatively display this skewed pattern but quantitatively follow the relationship between higher-order statistics predicted by the model. Generalized Lotka-Volterra (gLV) models, the standard framework for predicting community diversity, typically assume Gaussian interaction strengths. Instead, we show that the observed "many weak, few strong" interaction distribution is better captured by a lognormal distribution than the symmetric Gaussian distribution. Sampling interactions in gLV models from a lognormal distribution yields more stable communities in simulations and more accurately predicts the diversity observed in the two datasets where community-assembly experiments were performed. Together, these results reveal a metabolic origin for the "many weak, few strong" pattern of microbial interactions and show that using empirically grounded interaction distributions improves predictions of community diversity. SignificanceHow strongly do microbes affect each others growth? The distribution of their interaction strengths shapes how many species coexist in a community and how stable they are. Across nine datasets, we show that most interactions are weak, with rare strong outliers, mirroring the macro-organism "many weak, few strong" interaction pattern. Using mathematical models, we show that competition for shared resources and metabolite cross-feeding, both prevalent in microbes, suffice to generate this pattern. Accounting for the full distribution of interaction strengths, rather than only their mean and variance, improves predictions of community diversity in simulations and empirical datasets. This work connects microbial metabolic processes to community-level structure and provides a route for predicting diversity when interactions cannot be exhaustively measured.