Profiling the E . coli Membrane Interactome Captured in Peptidisc Libraries 1 2

14 Protein-correlation-profiling (PCP), in combination with quantitative proteomics, has emerged as 15 a high-throughput method for the rapid identification of dynamic protein complexes in native conditions. 16 While PCP has been successfully applied to soluble proteomes, characterization of the membrane 17 interactome has lagged, partly due to the necessary use of detergents to maintain protein solubility. Here, 18 we apply the peptidisc, a ‘one-size fits all’ membrane mimetic, for the capture of the Escherichia coli cell 19 envelope proteome and its high-resolution fractionation in the absence of detergent. Analysis of the 20 SILAC-labeled peptidisc library via PCP allows generation of over 4900 possible binary interactions out 21 of >700,000 random associations. Using well-characterized membrane protein systems such as the SecY 22 translocon, the Bam complex and the MetNI transporter, we demonstrate that our dataset is a useful 23 resource for identifying transient and surprisingly novel protein interactions, some of them with profound 24 biological implications, and many of them largely undetected by standard detergent-based purification. 25 The peptidisc workflow applied to the proteomic field is a promising novel approach to characterize 26 membrane protein interactions under native expression conditions and without genetic manipulation. 27 28 29 INTRODUCTION 30 Proteins control biological systems in a cell. While many perform their functions independently, 31 the majority of proteins interact with others to achieve their full biological activity. Characterizing 32 protein-protein interaction networks (the interactome) has traditionally been accomplished by methods 33 such as affinity purification coupled to identification by mass spectrometry (AP/MS) (Arifuzzaman et al. 34 2006; Hu et al. 2009; Babu et al. 2012, 2018), protein fragment complementation assays (Rochette et al. 35 2015; Tarassov et al. 2008), or yeast two-hybrid screening (Y2H) (Rajagopala et al. 2014). While high36 throughput, these methods are quite often limited in their scope by poor scalability because bait proteins 37 must be independently tagged. The addition of these tags can also have uncontrolled effects on proteins 38 such as disrupting binding sites, altering localization, stability, and thereby the accurate prediction of the 39 interactome. Co-fractionation methods, such as protein-correlation-profiling (PCP) in combination with 40 quantitative proteomics methods, such as label free quantitation (LFQ) or stable isotope labelling of 41 amino acids in cell culture (SILAC), are therefore emerging as an attractive alternative to identify protein 42 complexes under native expression conditions and without genetic manipulation (Kristensen et al. 2012; 43 Scott et al. 2017; Havugimana et al. 2012). Fractionation of a proteome under these native conditions, 44 followed by quantitative proteomic analysis of co-fractionation profiles, allows identification of protein 45 complexes through a principle of “guilt-by-association”. This method can generate thousands of potential 46 interactions in a single experiment, and incorporation of SILAC multiplexing allows simultaneous 47 comparison of multiple states of the interactome (Kristensen et al. 2012; Scott et al. 2017). 48 49 While co-fractionation has been successfully applied to soluble proteomes, characterization of the 50 membrane proteome has lagged. This is largely due to the hydrophobic nature of membrane proteins and 51 their sequestration in the lipid membrane. To extract this water-insoluble proteome, it is necessary to 52 solubilize the lipid bilayer with the aid of detergents or amphipathic co-polymers such as styrene maleic 53 acid (SMA) (Dörr et al. 2014). When mild detergents are employed, membrane protein complexes can be 54 directly detected following their separation by techniques such as size exclusion chromatography, density 55 gradient centrifugation (McBride et al. 2017) or blue-native gel electrophoresis (Scott et al. 2017; Heide 56 et al. 2012). However, even the mildest detergents tend to decrease protein stability while increasing 57 protein aggregation (Yang et al. 2014). In fact, prolonged exposure to those detergents tends to delipidate 58 proteins and alter their conformation, which can have confounding effects on membrane protein complex 59 stability. As an additional drawback, micelles of detergent must be removed from all samples before 60 analysis by mass spectrometry, which often decreases protein identification (Yeung and Stanley 2010; 61 Bechara et al. 2015; Bao et al. 2013; Yang et al. 2014). Thus, while a great deal of useful data has been 62 generated using detergent-based proteomics analysis, there is still a pressing need for novel methods that 63 are unencumbered by detergent side-effects. 64 65 We recently developed the peptidisc as a novel membrane mimetic scaffold to keep membrane 66 proteins water-soluble (Carlson et al. 2018). The peptidisc is formed when multiple copies of the 4.5kDa 67 amphipathic scaffold NSPr (also called Peptidisc peptide) wrap around the solubilized membrane 68 proteins. Reconstitution occurs spontaneously upon removal of detergent, incorporating both endogenous 69 lipids and solubilized membrane proteins into detergent-free particles. The number of scaffolds adapts to 70 fit the diameter of the protein target without bias toward large protein complexes. The end result is 71 peptidisc particles that are stable, free of detergent effects, and soluble in aqueous solution (Carlson et al. 72 2018). Our previous work has shown that the peptidisc is able to stabilize both inner and outer membrane 73 proteins of E.coli. 74 75 In this study, we apply the peptidisc to the trapping of the bacterial cell envelope proteome into 76 water soluble particles. This is performed by reconstituting the heterogeneous membrane protein mixture 77 immediately after its extraction from the cell envelope with mild detergent. This process minimizes 78 protein dissociation and denaturation because it limits exposure to detergent and thereby protein 79 delipidation. The membrane proteome trapped in the peptidisc library is water-soluble and stable during 80 prolonged incubations. This library is then fractionated by high-resolution size exclusion chromatography 81 (SEC) in the total absence of detergent. Application of the PCP workflow, which includes stable isotope 82 labelled amino acids in cell culture (SILAC) and mass spectrometry (LC/MS-MS), allows us to precisely 83 characterize the content of the peptidisc library across the various fractions. When the peptidisc library 84 from the raw E. coli cell envelope is analyzed this way, we identify and quantify 1209 unique proteins, of 85 which 591 are predicted to be directly membrane integrated. From these 1209 proteins, we predict 4911 86 binary interactions each characterized by a degree of precision. Our interaction list is hereafter called the 87 peptidisc interactome. 88 89 To computationally validate the precision of the peptidisc interactome, we benchmark the dataset 90 against the recently published E.coli cell envelope interactome (“CE”) (Babu et al. 2018) and two other 91 unpublished interactomes collected for that earlier study (“validating interactomes”). We also measure the 92 biological plausibility of the peptidisc interactome by determining enrichment for shared gene ontology 93 terms, binding domains, and correlation of growth phenotype (Erickson et al. 2017; Mosca et al. 2014). 94 We also compare the peptidisc-reconstituted membrane proteome against a membrane proteome prepared 95 using the SMA polymer instead of detergent. We find however that large membrane protein complexes 96 are better preserved in the peptidisc workflow. 97 98 Guided by the peptidisc interactome datalist, we select three well-characterized membrane protein 99 complexes in order to discover novel interactions. With the Sec translocon, we validate association of 100 SecY with the membrane chaperones YfgM and PpiD. This interaction can be isolated in detergent but 101 only when all subunits are simultaneously over-produced. Remarkably, we also discover significant 102 correlation between certain subunits of Sec and Bam complexes, suggesting an astonishing network of 103 protein associations across the entire bacterial cell envelope. We confirm this observation using SILAC 104 AP/MS, thereby providing direct evidence for the Bam-Sec super-complex. Continuing with the Bam 105 complex, we show that all 5 subunits are captured in peptidisc in addition to two other interactors RcsF 106 and OmpA. These interactions were previously inferred from genetic and indirect cross-linking 107 experiments, but direct association was not formally demonstrated (Hart et al. 2019; Konovalova et al. 108 2014). Accordingly, these interactions are much less apparent in detergent. Finally, with the ABC 109 transporter MetNI, we find that binding of the substrate binding protein MetQ depends on its N-terminal 110 lipidation The importance of the MetQ lipid anchor is novel and is unique case among Type I ABC 111 transporters. Moreover, we identify NlpA, also called lipoprotein 28, as a bona fide novel interactor of 112 the MetNI complex. 113 114 Altogether, this work validates the peptidisc workflow as an efficient method for capturing and 115 stabilizing the membrane proteome into soluble particles. The method enables high-throughput detection 116 of detergent-sensitive membrane protein interactions. When combined with rigorous experimental 117 validation, the peptidisc interactome is revealing novel and transient interactions, many of them of 118 fundamental importance to the transport process and biogenesis mechanism of the cell envelope. 119