Investigation of the Molecular Landscape of Bacterial Aromatic Polyketides by Global Analysis of Type II Polyketide Synthases

Abstract

Aromatic polyketides biosynthesized by microorganisms are a prominent class of natural products widely used in clinical treatments. Although genome mining approaches have accelerated the discovery of these molecules, the molecular diversity, abundance, and distribution of bacterial aromatic polyketides at a global scale remain elusive. Here, we provide a global atlas of bacterial aromatic polyketides based on large‐scale analysis of type II polyketide synthases. We first established the chain length factor protein as a marker that can predict both chemical class and molecular uniqueness of the biosynthetic product, and analyzed the abundance, taxonomic distribution, estimated structural diversity, and the total number of aromatic polyketides in bacteria. We further show the identification of oryzanaphthopyrans with an unprecedented angular naphthopyran scaffold from a rare actinobacterium by genome mining guided by the global atlas. These results serve as a compass for exploiting the entire type II polyketide synthase‐derived aromatic polyketides in bacteria.

Protein language model-based end-to-end type II polyketide prediction without sequence alignment

“…This approach differs from traditional unsupervised learning for clustering (44), as it strives to strike a balance between the sequence embeddings and the compound class labels to improve the model's accuracy. For example, T2PK AQ-256-8 consists of 8 building blocks, but its KSβ is confirmed as ancestral nonoxidative, which differs from other KSβs that involve the biosynthesis of T2PKs with 8 building blocks (6). Clearly, the state-of-the-art performance of the model trained with 9 refined class labels suggests that the classification effect is unsatisfactory when simply using five biosynthetic building blocks as labels.…”

“…T2 polyketide synthase is a family of single heterodimeric ketosynthases that iteratively catalyzes the elongation of the polyketide chain structure, leading to our inability to precisely predict T2PK structures. As introduced previously, despite the multiple sequence alignment approaches based on KSβ (5,6), incorporation of new sequences into the evolutionary model may alter the structure of the original phylogenetic tree and therefore compromise the accuracy of the predictions. To address Fig.…”

“…In contrast, KSβ sequences without corresponding chemical structures were classified as 'unlabeled'. To further curate the unlabeled KSβ, the 164 labeled KSα and KSβ sequences were used to obtain 20 kb putative T2PK minimum BGCs following the inhouse pipeline described by Chen et al (6). For the criterion of a reliable T2PK gene cluster, KSα and KSβ sequences should be identified in the same contig.…”

“…The core skeleton of T2PKs is highly correlated with the KSα/KSβ protein structure. Hillenmeyer et al observed correlations between KSβ protein phylogeny and the building blocks of T2PK skeletons (5), while Chen et al utilized KSβ as a biomarker to construct a coevolutionary statistical model (phylogenetic tree) to expand the T2PK biosynthetic landscape (6). However, the above and other (7)(8)(9) methods frequently rely on multiple sequence alignments, which are time-consuming and do not effectively represent protein structural information (10,11).…”