Deep learning-driven insights into super protein complexes for outer membrane protein biogenesis in bacteria

Abstract

To reach their final destinations, outer membrane proteins (OMPs) of gram-negative bacteria undertake an eventful journey beginning in the cytosol. Multiple molecular machines, chaperones, proteases, and other enzymes facilitate the translocation and assembly of OMPs. These helpers usually associate, often transiently, forming large protein assemblies. They are not well understood due to experimental challenges in capturing and characterizing protein-protein interactions (PPIs), especially transient ones. Using AF2Complex, we introduce a high-throughput, deep learning pipeline to identify PPIs within the Escherichia coli cell envelope and apply it to several proteins from an OMP biogenesis pathway. Among the top confident hits obtained from screening ~1500 envelope proteins, we find not only expected interactions but also unexpected ones with profound implications. Subsequently, we predict atomic structures for these protein complexes. These structures, typically of high confidence, explain experimental observations and lead to mechanistic hypotheses for how a chaperone assists a nascent, precursor OMP emerging from a translocon, how another chaperone prevents it from aggregating and docks to a β-barrel assembly port, and how a protease performs quality control. This work presents a general strategy for investigating biological pathways by using structural insights gained from deep learning-based predictions.

Atomic Models of All Major Trans-Envelope Complexes Involved in Lipid Trafficking inEscherichia ColiConstructed Using a Combination of AlphaFold2, AF2Complex, and Membrane Morphing Simulations

“…Obviously, the method rests heavily on the assumption that AlphaFold2derived methods such as AF2Complex can produce accurate predictions of the structures of protein complexes. We think that this has largely been illustrated by the Skolnick group previously (Gao et al, 2022a;Gao et al, 2022b), but an important result obtained here is that -at least for complicated combinations of subunits -repeated predictions may be required to find the most reasonable model. The best example of this is provided by our results for the PqiB6:PqiC6 interaction, for which the piTM score for one predicted structure was superior to 49 other predicted structures by a large margin (Figure 3F).…”

“…As such, we used metrics that are automatically generated by the method as our primary means of identifying the "best" predictions. In general, we found that both the piTM score and the so-called interface score (Gao et al, 2022a;Gao et al, 2022b) performed well, with the former appearing more useful especially when we needed to identify the homo-oligomeric state of a protein, as it seemed less dependent on the number of chains included in the prediction. However, since neither score explicitly monitors nor penalizes the presence of steric clashes, we encountered several cases where predicted structures with promising scores were riddled with clashes due to an absurd placement of identical subunits on top of each other.…”

“…Before proceeding, it is worth noting two general features of our use of AF2Complex. First, while several metrics have been proposed for estimating the reliability of structures predicted by AF2Complex, we focus here on ranking predictions by the predicted interface TM (piTM) score proposed by the Skolnick group (Gao et al, 2022a;Gao et al, 2022b) because we found this to be generally the most reliable in the current setting. However, we also found that even this score can occasionally provide a misleading view of the plausibility of a predicted structure: in some cases, we obtained very favorable piTM scores from predicted structures that placed copies of the same chain directly on top of each other, i.e.…”

“…AF2Complex provides several metrics for assessing the confidence of predicted protein complexes in addition to the original metrics provided by the underlying AlphaFold2 source code (Jumper et al, 2021), and it has already been applied to predict several new interactions for proteins that make up the cell wall of E. coli (Gao et al, 2022b).…”

Interrogation and validation of the interactome of neuronal Munc18-interacting Mint proteins with AlphaFold2

“…These predictions provide useful insights into those interactions that are likely to be directly mediated by the Mint proteins. Our results provide an example of how AlphaFold2 (and similar algorithms) can be used to provide confidence in the plausibility of complexes identified in larger proteomic datasets (Burke et al, 2023; Gao et al, 2022a; Gao et al, 2022b; Humphreys et al, 2021; Sifri et al ., 2023; Yu et al, 2023). Such an approach has the potential to inform and accelerate subsequent experimental validation of molecular complexes detected in high-throughput screens, by providing much greater confidence as to which hits represent specific interactions.…”