Longitudinal wastewater sampling in buildings reveals temporal dynamics of metabolites

Direct sampling of building wastewater has the potential to enable precision public health observations and interventions. Temporal sampling offers additional dynamic information that can be used to increase the informational content of individual metabolic features, but few studies have focused on high-resolution sampling. Here, we sampled three spatially close buildings, revealing individual metabolomics features, retention time (rt) and mass-to-charge ratio (mz) pairs, that often possess similar stationary statistical properties, as expected from aggregate sampling. However, the temporal profiles of features-providing orthogonal information to physicochemical properties-illustrate that many possess different feature temporal dynamics (fTDs) across buildings, with large and unpredictable single day deviations from the mean. Internal to a building, numerous and seemingly unrelated features, with mz and rt differences up to hundreds of Daltons and seconds, display highly correlated fTDs, suggesting non-obvious feature relationships. Data-driven building classification achieves high sensitivity and specificity, and extracts building-identifying features found to possess unique dynamics. Analysis of fTDs from many short-duration samples allows for tailored community monitoring with applicability in public health studies. Author summary Understanding a community's wastewater profile may allow for specific and targeted interventions. Untargeted wastewater metabolomics provides a rich data source, but one that is high dimensional, noisy and difficult to understand. We analyze building-to-building differences and through-time patterns from temporal wastewater metabolomics data, obtained directly from three buildings. We develop and apply computational techniques to extract building-specific temporal and stationary properties for each small molecule feature. Stationary properties are predominantly conserved, but by studying the temporal dynamics, we find distinct, building-specific signatures and metabolite patterns. Interestingly, using clustering techniques and temporal similarity metrics, we find that for each building there exist groups of small molecules that possess highly similar temporal dynamics, despite having vastly different molecular properties (e.g. molecular weight or chromatographic retention time). These findings may suggest similar generative processes for such small molecules, which may lead to increased biological understanding. Additionally, our computational methods link putatively identified small molecules with unknown features. This produces a list of unknown compounds, with community-specific temporal dynamics for follow up experimental analysis and targeted discovery to better understand a community of interest.