Functional processes predict structural colour in Gram-negative bacteria from sequencing data. Bacterial structural colour

We collected a set of 87 structurally coloured (SC) bacterial isolates, and 30 related strains lacking SC, including Gammaproteobacteria that were not previously known to show SC. Next, we used genome wide association approaches to identify genes associated with SC. We found that SC bacteria shared certain signature functions that have also been shown to function as SC enhancing compounds in butterflies and birds, including uroporphyrin and molybdopterin biosynthesis clusters. Involvement of shared carbohydrate utilization proteins is in agreement with observations that SC depends on the selected carbon source. Methionine metabolism and acetolactate metabolism were also implicated in SC, as were gliding associated proteins for the Flavobacteriia. Hidden Markov Models were constructed using 199 SC-associated protein sequences identified using both GWAS and mutational approaches and a machine-learning model was constructed using Random Forest to predict SC in untested strains. The HMM models were used to analyze publically available genomes and environmental metagenomes to predict the occurrence of SC across the tree of life and in various ecosystems. We predicted, and subsequently experimentally confirmed that structural colour exists within the Alpha- and Betaproteobacteria, suggesting that photonic structures are found widely within Gram-negative bacteria. A broader process of validation, scoring 80 strains not involved in creating the classifier for SC, supported the predictive power of the method.