Periplasm homeostatic regulation maintains spatial constraints essential for cell envelope processes and cell viability

The cell envelope of Gram-negative bacteria consists of two membranes surrounding a periplasm and peptidoglycan layer. Molecular machines spanning the cell envelope depend on spatial constraints and load-bearing forces across the cell envelope and surface. The mechanisms dictating spatial constraints across the cell envelope remain incompletely defined. In Escherichia coli, the coiled-coil lipoprotein Lpp contributes the only covalent linkage between the outer membrane and the underlying peptidoglycan layer. Using proteomics, molecular dynamics and a synthetic lethal screen we show that lengthening Lpp to the upper limit does not change the spatial constraint, but rather impacts the load-bearing capacity across the outer membrane. Our findings demonstrate E. coli expressing elongated Lpp homeostatically counteracts periplasmic enlargement with a combination of tilting Lpp and reducing Lpp abundance. By genetic screening we identified all of the genes in E. coli that become essential in order to enact this homeostasis, and by quantitative proteomics discovered that very few proteins need to be up- or down-regulated in steady-state levels in order to enact this homeostasis. We observed increased levels of factors determining cell stiffness, decrease membrane integrity, increase membrane vesiculation and a dependance on otherwise non-essential tethers to maintain lipid transport and peptidoglycan biosynthesis. Further this has implications for understanding how spatial constraint across the envelope controls processes such as flagellum-driven motility, cellular signaling and protein translocation

INTRODUCTION 42 43 Gram-negative bacteria have a cell envelope composed of two membranes sandwiching between them an 44 aqueous space called the periplasm, in which an essential structural layer of peptidoglycan (PG) resides. The 45 outer membrane is critical to cell growth and these bacteria face challenges to their cell biology in terms of 46 membrane protein assembly and lipid-transport pathways that must traverse the distance from the inner 47 membrane (IM) to the outer membrane (OM) 1 . Recent work investigating the spatial demands for assembly 48 of proteins into the outer membrane has shown precincts of active protein integration into the membrane can 49 deliver new material to the growing outer membrane 2 and that random planar movement from these precincts 50 drives the observed non-uniform distributions of the major proteins of the outer membrane 3, 4 . By contrast to 51 these protein components that diffuse to the outer membrane, elements of the LPS-transport machinery (eg. 52 the Lpt complex) 5,6,7,8,9 span the OM and IM in order to fulfil their function in delivering lipid components 53 to the outer membrane. 54

55
The PG layer is a fundamental aspect of the cell envelope, and it must be dynamically remodeled to allow 56 growth as well as the assembly and transit of trans-envelope structures. PG synthesis and remodeling is a 57 complex process with high levels of redundancy at various steps, involving at least 50 enzymes in E. coli 10 . 58 The Penicillin-binding proteins (PBPs) are the core components responsible for the periplasmic biosynthesis 59 of peptidoglycan. There are multiple PBP complexes including the two core, semi-redundant PBP complexes 60 PBP1a and PBP1b embedded in the IM which are activated by interactions with lipoproteins LpoA and LpoB 61 embedded in the OM. Thus the activation of PG synthesis by these enzymes is spatially regulated, serving as 62 a self-limiting molecular ruler to modulate PG thickness 11 . Cells must possess either a functional PBP1a or 63 PBP1b system for growth 12,13 . 64 4 To understand how trans-envelope processes in E. coli adapt to the presence of an enlarged periplasm, a 77 combination of phenotypic analysis, proteomics, molecular dynamics and a synthetic lethal screen was 78 employed to identify and characterize factors needed to maintain viability in the Lpp +21 strain of E. coli. The 79 genetic screen demanded synthetic growth phenotypes from an array of mutants each lacking a gene that, 80 while non-essential in wild-type E. coli 19 , is essential in the Lpp +21 strain. These genes fall into three functional 81 categories: PG biosynthesis and remodelling, LPS biosynthesis and PG-outer membrane linkage. We show 82 that previously non-essential proteins involved in bridging the gap between the OM and PG become essential 83 in the context of the Lpp +21 strain background. These include previously known PG binding OM proteins 84 (OmpA and Pal) as well as proteins previously not known to play an active role in linking the OM and PG 85 (TolC and YiaD). We observed a thicker more diffuse or heterogeneous PG layer in the Lpp +21 strain and 86 whole cell proteomics revealed that in response to an increased length of Lpp, E. coli increases the levels of a 87 range of cell envelope proteins involved in PG turnover. We discuss the outcomes in terms of how the PG-88 outer membrane linkage functionalizes the periplasm, the evolutionary constrains in place to maintain this 89 functionality, and the specific activity of Lpp in contributing to the load-bearing function of the OM. 90

Resilience and growth of the Lpp +21 strain 94
A phylogenetic assessment of Lpp lengths across diverse bacterial lineages showed a very narrow window of 95 protein size (Fig. 1A), with Lpp being 78 residues in most species of bacteria including E. coli. Lpp lengths 96 of 99 residues or more are at the upper end of the natural range for this protein and, in nature, these longer 97 Lpp proteins are found in the genus Geobacter. The previously described Lpp +21 isoform expressed in E. coli 98 therefore sits near the upper physiological range observed among bacteria. We introduced a gene encoding 99 the Lpp +21 isoform into an E. coli background suitable for genetic screens (Fig. S1) and confirmed the size of 100 the protein by SDS-PAGE of bacterial cell extracts (Fig. 1B). Enteric bacteria like E. coli can thrive in 101 hyperosmotic environments compared to most laboratory growth media, such as the human gut by maintaining 102 the periplasm and the cytoplasm in an iso-osmotic state 20, 21 . This is achieved by adjusting the solute 103 concentration in the cell compartments by influx or efflux of water. By doing so, osmolality contributes 104 immensely to the architectural aspects of cellular compartments 21,22 . In this study sorbitol was used to mimic 105 these physiological osmotic conditions. Previous studies have demonstrated that the periplasmic volume 106 increased rapidly in response to increased osmolyte levels in the external medium. This phenomenon can 107 reduce the cytoplasmic volume by about 30%, thereby constricting the IM inwards, and substantially 108 increasing the periplasmic volume by around 300% 21, 22, 23 . The Lpp +21 isoform had little impact on growth of 109 E. coli. On minimal growth medium, growth rates of the Lpp +21 strain of E. coli were equivalent to the isogenic 110 wild-type E. coli (Fig. 1C). This was likewise true on growth media osmotically balanced with concentrations 111 of sorbitol up to 1.0M (Fig. 1D), and on rich (LB) medium with or without sorbitol (Fig. 1D). 112 113 To establish the extent to which the periplasm had been remodeled in the Lpp +21 strain of E. coli, we compared 114 the periplasms of WT and Lpp +21 strains using electron cryotomography in which cells are preserved in a 115 frozen-hydrated, near-native state. To discern peptidoglycan, which was indistinct in previous studies, we 116 increased the signal to noise ratio in images by calculating subtomogram averages instead of inspecting 117 individual tomograms. As expected, the average distance from the middle of the IM density to the middle of 118 the OM density was increased in both strains under hyperosmotic conditions when compared to previously 119 reported data where cells were grown in standard laboratory media 18 . As shown in Fig. 1E, the averaged 120 distance from the IM to the OM were somewhat greater in the Lpp +21 strain (32-36nm) compared to the 121 isogenic wild-type strain (30-32nm) in line with previous Lpp+21 periplasmic width measurements. Previous 122 studies could not discern peptidoglycan; whereas here, we were able discern peptidoglycan in subtomogram 123 averages. The distance to the center of the PG density from the OM was slightly increased in the Lpp +21 strain 124 (Fig. 1F), although by less than the anticipated ~3 nm. The PG morphology was also qualitatively different: 125 in the wild-type strain a uniform dark PG layer could be observed in the images, whereas in the Lpp +21 strain 126 6 a broader and more diffuse PG layer was evident, suggesting heterogeneity in both the density and thickness 127 of the   S1) were grown over 24 hours. The growth medium is M9, containing the indicated concentration of sorbitol as an osmolyte. (D) Growth rates for the same strains were measured in rich (LB) growth media with and without sorbitol over 20 hours. (E) The periplasmic width distribution of the indicated strains in hyperosmotic conditions. While PG layer in the wild-type strain is a uniform thin electron dense layer, the PG layer in the Lpp+21 strain is more diffuse and thicker. (F) Subtomogram averages of cell envelopes in hyperosmotic conditions. Measurements from EM views evaluate the distance between OM and PG in the Lpp+21 strain micrographs. The histogram depicts the frequency with which a given distance is observed between the OM and PG.

Cell envelope response to elongated Lpp 131
To determine the adaptive response to changing the distance constraint between OM and PG, quantitative 132 whole cell proteomics was applied to evaluate the Lpp +21 strain. Triplicate samples of the wild-type and Lpp +21 133 strains were processed for analysis by mass spectrometry and we sought to identify those proteins where the 134 steady-state level increases or decreases three-fold or more (Log2 fold change of ± 1.6) in the Lpp +21 strain 135 ( Fig. 2A; Table S1). The level of the oligopeptide transporter subunits (OppB, OppC, OppD and OppF) are 136 substantially increased in the Lpp +21 strain compared to the wild-type (Table S2). This suggests an increase in 137 PG turnover and an overall increased capacity to recycle PG components, and is consistent with the 138 concomitant increase in AmiC, one of the two major amidases involved in PG remodeling. In addition, 139 proteins implicated in diverse stress-responses (cold shock proteins CspG, CspA, CspI and YdfK, as well as 140 envelope stress protein ZraP and redox stress protein YfcG) were observed at increased steady-state levels in 141 Lpp +21 strain ( Fig. 2A, Table S2). The greatest decreases were seen in the steady state levels of the GatZABCD 142 proteins involved in galactitol phosphotransferase system and DHAP synthesis ( Fig. 2A). The gatABCD genes 143 have been shown to be responsive to factors that change E. coli cell surface tension 24 and Lpp +21 has been 144 reported to significantly decrease cell stiffness 25 . 145 146 A decrease was seen in the steady-state level of the Lpp +21 isoform in the mutant, to approximately one-eighth 147 the level of Lpp in the wild-type strain (Fig. S2). This is consistent with the relative abundance of Lpp and 148 Lpp +21 observed in SDS-PAGE analysis of cell extracts from the two strains (Fig. 1F). However, despite the 149 relative decrease, Lpp +21 remains as a highly abundant component of the OM-PG linkage factors given that 150 Lpp is present at up to 10 6 protein molecules per wild-type cell 1, 26 . 151 152 The mass spectrometry data was processed to allow for an analysis of sub-cellular proteomes 27 ( Fig S3). An 153 initially puzzling observation was that the Lpp +21 strain has a 12% overall reduction of total periplasmic 154 protein compared to wild type (Fig. 2B). This was calculated as the proportion of the summed intensity from 155 identified proteins predicted to reside in the periplasm in the STEPdb: G, E, F2, F3, I annotations 27 ( Fig S3). 156 Both Lpp +21 strains and null Lpp strains of E. coli are softer as previously adjudged by atomic force 157 microscopy 25 , and factors that increase the softness of E. coli also increase outer membrane vesicle (OMV)  158 production 17,28,29,30 . To address whether the measured depletion of periplasmic content reflects an increased 159 production of OMVs, extracts measuring the amount of total protein in the OMV fraction were normalized to 160 OD 600 (Fig 2C, Fig. S3). This confirmed that the presence of Lpp +21 promotes approximately 10-fold more 161 total protein associated with the OMV fraction, reflecting increased OMV production. The overall level of 162 OM proteins associated with the cells was maintained constant (Fig. S3) but a small OM integrity defect was 163 evident from an increased sensitivity to SDS (Fig. 2D

Lpp +21 can be accommodated in the periplasm, but other factors become essential 167
To directly address the altered phenotype induced by Lpp +21 , we established a synthetic genetic array for 168 factors in E. coli that become essential in order to maintain viability of the Lpp +21 strain ( Fig S4). The screen 169 demands synthetic growth phenotypes from an array of mutants each lacking a single gene that, while non-170 essential in wild-type E. coli, become essential in the Lpp +21 strain of E. coli. The endogenous lpp gene was 171 replaced by a gene encoding Lpp +21 in a isogenic library of 3818 E. coli mutants, each of which lacks a non-172 essential gene. Growth on rich medium allowed the rescue of the new library for array into a format suitable 173 for high-throughput screening with a Singer RoToR robotics platform ( Fig S4). Phenotypic analysis was 174 thereafter scored for growth by comparing the growth of the isogenic mutants in the Lpp background ( Fig.  175 3A) with the equivalent mutants in the Lpp +21 background (Fig. 3B). Each of the genes that displayed a 176 noticeable phenotype in these analyses are presented in Table 1. 177 178 The only cytoplasmic factor identified in our screen, YraN is predicted to be a Holliday-junction resolvase 179 related protein, and we therefore speculate that this mutant failed to resolve the merodiploid condition transient 180 in the introduction of the lpp +21 condition to the background strain, making the yraN mutant a technique-181 relevant artefact of the screen. This being the case, only functions performed in the periplasm were recovered 182 as essential to viability for the Lpp +21 strain. 183 184 Most of the components of the LPS biosynthetic machinery are essential genes in E. coli and are thus not 185 represented in the library of non-essential genes. Several non-essential genes in the LPS biosynthetic pathway 186 that are in the library, become essential to the Lpp +21 strain (Table 1): the core LPS biosynthesis factors GalU, 187 GmhB and RfaD were shown to be essential in the Lpp +21 strain..

189
An essential role for keeping the OM-PG distance 190 Several genes encoding proteins that could play roles in anchoring the PG within the cell envelope were 191 identified as essential in the Lpp +21 background. Independently, none of the major proteins bridging the OM 192 and PG are essential for growth in E. coli 19 and all are therefore represented in the library. In a Lpp +21 193 background the genes encoding the β-barrel protein OmpA and the lipoprotein Pal become essential (Table  194 1). PG-binding domain PF00691 is common to these proteins: appended to a beta-barrel in OmpA, but to a 195 lipoyl anchor in Pal, and is also conserved in other proteins across diverse Gram-negative bacteria ( Fig 3C). 196 In E. coli there are 4 additional proteins containing this PG-binding domain and these were mapped in a 197 sequence similarity network analysis ( Fig 3D). A protein of unknown function, YiaD, is present ( Fig 3D) and 198 it too is essential in a Lpp +21 background (Table 1). We suggest, therefore, that this protein plays a substantive 199 role in OM-PG linkage. The remaining three proteins: MotB, LafU and YfiB, are more divergent to the 200 OmpA/Pal/YiaD cluster. Neither motB, lafU nor yfiB displayed a synthetic phenotype with Lpp +21 , and it has 201 been suggested previously that motB, lafU and yfiB are not expressed at detectable levels under laboratory 202 conditions 31 . 203 204 Detecting genes encoding drug-efflux pumps as important for growth of the Lpp +21 strain was initially 205 surprising. Either the absence of the inner membrane proteins AcrB or the OM component TolC caused a 206 reduction of growth in the Lpp +21 genetic background (Table 1). When antibiotic selection was removed by 207 plating the mutants on medium without chloramphenicol, the synthetic growth defects were observed in the 208 absence of drug selection (Fig. 3D), indicating that this synthetic phenotype is not the result of a decreased 209 drug efflux activity. The trans-envelope AcrAB-TolC multidrug efflux pump has been shown to traverse 210 through the PG and interact directly with PG at several defined sites 32,33,34,35,36 , and as loss of the core AcrB 211 and TolC components became essential, we suggest that this system could be acting as an additional OM-PG 212 linkage that becomes essential in a Lpp +21 background. Together with the observation that OmpA, Pal, YiaD 213 and TolC are also essential in the Lpp +21 genetic background, these data suggest that functions that maintain 214 local areas of closer contact between the OM and PG are essential for viability. 215 216    coli is a triple coiled-coil that is anchored to the inside face of the outer membrane by its N-terminal acyl 226 group with a length equating to approximately 7.5 nm 39 . The experiments were established to test the scenario 227 for Lpp trimers or Lpp +21 trimers in the absence or presence of an OmpA tether between the patch of OM and 228 patch of PG (Fig. 4). These MD simulations only represent a portion of the membrane, and as such we could 229 not account for potential heterogeneity in the protein and LPS concentrations between a WT and Lpp +21 cell. 230 For three of the systems: Lpp only, Lpp with OmpA, and Lpp +21 only, tilting was marginal, with tilt angles of 231 76.9±4.7°, 75.5±4.7°, and 82.8±2.9°, respectively (all numbers from the last 100 ns of the 200-ns trajectory). 232 These angles are in agreement with previous simulations of Lpp alone (~80°) and Lpp with an OmpA 233 monomer (~75°) PG 36,40,41 . In initial simulations of Lpp +21 with OmpA, the non-covalent connection between 234 OmpA and PG was quickly disrupted as Lpp +21 extended from its kinked state. Therefore, the simulation was 235 repeated with an enforced OmpA-PG connection. Lpp +21 was observed to both straighten and tilt within the 236 first 100 ns; the tilt angle measured for the last 100 ns was 49.4±2.3°. 237 . 238 The distance between the OM inner leaflet phosphorus atoms and the PG sugars was measured in each 239 scenario. In the presence of Lpp, the distances with and without OmpA were similar at 8.3±2.1 nm and 8.1±1.2 240 nm, respectively. This was not true for the other scenarios where the distance for Lpp +21 alone was 11.6±1.3 241 nm, but for Lpp +21 with OmpA, the distance was reduced significantly to 8.7±2.7 nm. Thus, we observe that 242 PG-binding proteins like OmpA can counteract the increased distance imposed by Lpp +21 , inducing it to tilt 243 significantly in accommodation. We also compared our simulations to the distances that were measured by 244 EM (centre of the OM to centre of the PG). In wild-type E. coli (i.e. Lpp+OmpA), the centre-centre distance 245 in the simulations is 10.7 +/-2.2 nm (Fig 4B), similar to the 9.7 -10.8 nm measured in intact cells (Fig. 1F). 246 The centre-centre distance measured in the simulation of Lpp +21 tilted by the presence of OmpA (11.0 +/-2.6 247 nm), fits the observed distances of 10.  We observed that over a range of osmotic conditions, and in nutrient-rich or nutrient-poor media, growth rates 253 of the Lpp +21 strain of E. coli were equivalent to the isogenic wild-type E. coli, suggesting bacteria can adapt 254 to the presence of the extended Lpp +21 . We observed a moderate increase to the OM-IM space similar to that 255 previously reported 17, 18, 25 and reviewed 25,42 . Three adaptive features were expressed as phenotypes in the 256 Lpp +21 strain: (i) the steady-state level of the Lpp +21 tether was reduced eight-fold compared to the level of 257 Lpp in the isogenic wild-type strain, and other tethers that enforce a wild-type distance: OmpA, Pal, YiaD and 258 TolC, became essential factors in the Lpp +21 condition, (ii) structures that depend on a wild-type OM-IM 259 distance, such as the LPS transport system, continued to function but key components of the system became 260 essential for cell viability, and (iii) the PG network took on characteristics of dysregulated synthesis and all 261 components of an otherwise redundant PG biosynthesis pathway become essential to viability. In response to 262 Lpp elongation we demonstrated a reduction in Lpp levels, as shown by the quantitative proteomic presented 263 here, and the apparent tilting of the elongated Lpp as suggested by our molecular dynamic simulation. The 264 reduction in the copy number of the elongated Lpp isoform observed in the quantitative proteomics was not 265 due to differences in transcriptional regulation, as RNAseq data demonstrated a moderate increase in the level 266 of lpp +21 transcripts in the mutant (Table S6). Taken  and mrcB (as well as their OM lipoprotein partners) are essential for viability, indicative of a compromised 321 capability to build the PG-layer. Mutations designed to impact on these interactions lead to transient deposition 322 18 of "high-density PG" and "multi-layered PG" through dysregulation of the synthetases 11 . The broader and 323 more diffuse morphology of the PG-layer observed by electron microscopy in the Lpp +21 strain could suggest 324 a similar phenotype associated with transient or local impacts on the OM-PG distance 11 . It's worth noting 325 Δlpp ΔmrcB mutants have previously shown a moderate growth defect 52 and that this defect could be 326 alleviated with the introduction of the lpp +21 gene 18 , whereas in the present study we observed a synthetic 327 lethal phenotype in the ΔmrcB lpp +21 mutants. This discrepancy could be due differences in growth conditions 328 used (LB vs hyperosmotic minimal media) imparting additional periplasmic stress or differences in the strain 329 background (DH300 vs BW25113). 330 331 An essential requirement was also placed on PG-layer remodeling, whereby the PG-binding factor NlpD was 332 found to be essential in Lpp +21 cells: its function is in modulating the activity of the amidase AmiC 53, 54 to 333 remodel PG strands, and AmiC was observed at increased steady-state levels in the Lpp +21 strain. Taken  334 together with the increase in oligopeptide transporter subunits (OppB, OppC, OppD and OppF) in the Lpp +21 335 strain to recycle PG precursors across the IM, our results suggest a clearance of the malformed PG caused by 336 dysregulation of the PG synthases is a crucial adaptation in the Lpp +21 strain. 337 338

339
The concept of bacterial cell stiffness has emerged as a means to understand the physical parameters that 340 define how readily bacteria can respond to major environmental changes 30, 55, 56 . Measurements by AFM have 341 revealed a characteristic stiffness in Gram-negative bacterial cells that is contributed by load-bearing outer 342 membrane and its attachment to the underlying PG layer 57 . In E. coli, mutants lacking Lpp, Pal or OmpA are 343 softer than wild-type cells 57 , and cells expressing the Lpp +21 isoform are also softer than wild-type cells 25 . 344 Taken together with our observation that the Lpp +21 isoform display a broader and more diffuse PG layer 345 morphology in the subtomogram averages, this further supports the proposition that the OM is a major 346 contributor to cell stiffness 55 (Fig. S4). The mutation was 387 verified by PCR described (Fig. S1) and sequencing. 388 389

SDS-PAGE and immunoblotting
390 Cells grown in M9 minimal media (0.5 M sorbitol) and normalized by OD600nm, lysed in Laemmli SDS-loading 391 dye and separated in 15% acrylamide gels and transferred to 0.45 μm hydrophobic Immobilon-P PVDF 392 membrane (Merck Millipore). Immunoblotting was as described previously 19 . Rabbit primary antibodies; α-393 Lpp antibody (kindly provided by T. Silhavy) and α-OmpA were diluted 1:400,000 and 1: 30,000, respectively 394 in 5% skim milk, TBST. The membranes were incubated with goat, α-rabbit IgG, HRP-conjugated secondary 395 antibody (Sigma; 1: 20,000 in 5% skim milk, TBST), and washed with TBST. Detection was by enhanced 396 chemiluminescence with ECL prime western blotting detection reagent (GE Healthcare Life Sciences), 397 visualized using Super RX-N film (Fujifilm). 398 399 Outer membrane vesicle purification and quantification 400 Overnight cultured cells, grown in LB without antibiotics, were washed twice in 1 x M9 salts then subcultured 401 in 500 ml M9 minimal media supplemented with 0.5 M sorbitol (1:1000 dilution). The strains were grown to 402 late logarithmic phase without antibiotics, OD 600 ≈ 0.9 and spun down to collect culture supernatant. Collected 403 culture supernatant were then processed for OMVs isolation and purification using differential 404 ultracentrifugation technique as discussed previously 63 . OMVs were washed twice in PBS to remove sorbitol 405 then quantified using a bicinchoninic acid assay kit (Thermo Scientific CST#23225 and protein concentration was determined with Bicinchoninic Acid assay (BCA, Thermo Fisher). SDS was 418 removed with chloroform/methanol, the protein was digested by trypsin overnight and the digested peptides 419 were purified with ZipTips (Agilent). Using a Dionex UltiMate 3000 RSLCnano system equipped with a 420 Dionex UltiMate 3000 RS autosampler, an Acclaim PepMap RSLC analytical column (75 µm x 50 cm,  421 nanoViper, C18, 2 µm, 100Å; Thermo Scientific) and an Acclaim PepMap 100 trap column (100 µm x 2 cm, 422 nanoViper, C18, 5 µm, 100Å; Thermo Scientific), the tryptic peptides were separated by increasing 423 concentrations of 80% ACN / 0.1% formic acid at a flow of 250 nl/min for 120 min and analyzed with a 424 QExactive Plus mass spectrometer (Thermo Scientific). The instrument was operated in the data dependent 425 acquisition mode to automatically switch between full scan MS and MS/MS acquisition. Each survey full scan 426 (m/z 375-1575) was acquired in the Orbitrap with 60,000 resolution (at m/z 200) after accumulation of ions 427 to a 3 x 10 6 target value with maximum injection time of 54 ms. Dynamic exclusion was set to 30 seconds. 428 The 20 most intense multiply charged ions (z ≥ 2) were sequentially isolated and fragmented in the collision 429 cell by higher-energy collisional dissociation (HCD) with a fixed injection time of 54 ms, 15,000 resolution 430 and automatic gain control (AGC) target of 2 x 10 5 . 431 432

22
The raw data files were analyzed using MaxQuant software suite v1.6.5.0 64 against Andromeda search engine 433 65 for protein identification and to obtain their respective label-free quantification (LFQ) values using in-house 434 standard parameters. The proteomics data was analyzed using LFQ-Analyst 66 and the analysis of the data 435 quality analysis is presented in Fig S6. Due to the 21 amino acid insertion in the Lpp +21 isoform, the relative 436 levels of Lpp in the mutant had to be assessed manually. Only the unique peptide (IDQLSSDVQTLNAK)  437 shared between the two isoforms was used to quantify the levels of Lpp and Lpp +21 in the wild-type and mutant 438 strain, respectively (Fig S2). 439 440 To estimate the total relative amount of proteins from the various subcellular compartments, the raw intensities 441 from peptides identified from proteins from different subcellular locations were summed and divided by the 442 total summed intensity from all peptides. Sub-cellular locations annotations were applied from the STEPdb 443 2.0 27 where proteins designated F1, A, R, and N were classified as cytoplasmic; B was designated inner 444 membrane; H, X, and F4 were designated outer membrane / extracellular; and I, G F2, F3, and E were 445 designated as periplasmic. 446 447 Sequence similarity network analysis

448
Proteobacterial proteins containing the Pfam domain (PF00691) were extracted from the Representative 449 Proteome 35% co-membership rpg-35 group 67 and a sequence similarity network was generated with the EFI 450 Enzyme Similarity Tool 68 . This network was visualized with Cytoscape 69 with a similarity score cutoff of 451 30. Each protein is represented by a colored circle node and each similarity match above the similarity score 452 cutoff is represented by an edge between nodes with the length determined by the similarity score. 453 454 Lpp length distribution across bacterial species

455
To determine the amino acid length distribution of Lpp in Gammaproteobacteria (Table S4), amino acid 456 sequences were sourced from the InterPRO database (version 81.0) 70 using the Interpro Family tag -Murein-457 lipoprotein (IPR016367). Filtered Lpp sequences were then concatenated into representative nodes (at least 458 >90% sequence similarity) using the online available amino acid Initiative-Enzyme Similarity Tool (EFI-EST) 459 68 . 460 461 Synthetic genetic interaction array

462
The Lpp +21 isoform was transferred to each of the Keio collection clones by conjugation as described (Fig.  463  S4). First, the Hfr chloramphenicol resistant Lpp +21 strain was arrayed in 384-colony density on LB agar 464 containing chloramphenicol using the Singer rotor HAD (Singer Instruments, United Kingdom). Similarly, 465 the Keio collection arrayed in 384-colony density was pinned on LB agar plates containing kanamycin and 466 incubated overnight at 37 °C. Using the Singer rotor HDA, the Hfr Lpp +21 strain and the Keio collection clones 467 from the 384-colony density were then co-pinned onto LB agar plates and incubated at 37 °C for 16 hours. 468 Following conjugation, the colonies were transferred to LB agar with kanamycin (selection 1) at the same 469 colony density and incubated at 37 °C for 16 hours. To select for double mutants (selection 2), colonies from 470 the intermediate selection were pinned on LB agar with both kanamycin and chloramphenicol and incubated 471 at 37 °C for 14 hours. For assessment of synthetic genetic interaction in nutrient-limited media, the double 472 mutants generated were replica pinned in M9 minimal media at the same density and incubated at 37 °C for 473 25-30 hours. Images were acquired using Phenobooth (Singer Instruments, United Kingdom) for analysis. 474 Images were manually screened to cross-reference recipient plate images to the final double antibiotic 475 selection plates images. Candidate synthetic lethal or growth-compromised mutants were then subjected to 476 another round of screening in the same conditions as previously identified (mini-screen) for validation. Four 477 biological replicates were included that were further arrayed in four technical replicates. Mutants were 478 confirmed by PCR. Where further phenotypic screening of mutants was conducted, independent isogenic 479 knock-out mutants were generated using the PCR and λ-red based homologous recombination method 61 . 480 481 Since the Keio collection yiaD mutant has been identified as containing a potential duplication event 71 , the 482 candidate yiaD synthetic lethal interaction was confirmed through independently constructing a yiaD mutant 483 in the BW25113 strain background (Fig. S5). The lpp +21 variant was subsequently generated in this mutant as 484 described above. As with the Keio yiaD mutant, this strain demonstrated synthetic lethality on M9 media. 485 486 Two colony PCR reactions 61, 71 (Fig. S4) confirmed the identity of all candidate double mutants, using a set 487 of primers flanking the lpp gene, and a set of primers flanking the kanamycin gene (Table S5). 488 489

490
Strains were grown aerobically in M9 minimal media (0.5 M sorbitol) until an OD600 of 0.6 was reached. Cells 491 were collected by spinning at 6000xg for 5 minutes and resuspended to an OD600 of ≈ 12. Cryo-EM, data 492 collection and analysis were performed similarly to previous studies 17, 18 , except using 3-D subtomogram 493 averages derived from whole-cell cryotomograms instead of projection images so as to discern peptidoglycan. 494 Tilt series of WT and Lpp +21 strains was acquired on an FEI Krios operating at 300 keV with a Gatan K2 495 direct detector and energy filter with a 20eV slit with a tilt range of +/-60° using 3° increments and 496 reconstructed using IMOD. Subtomograms were picked manually using 3DMOD along the length of all non-497 polar periplasm and averaged using PEET. 498 499 Generation of simulation systems 500 Initially, two systems were generated: the OM and PG, as previously detailed 56 , with two copies of wild-type 501 Lpp and with two copies of Lpp +21 . For wild-type Lpp, we used the homo trimer from PDB 1EQ7 39 . For 502 Lpp +21 , a monomer was first built using I-TASSER 72 . Next, the trimer of Lpp +21 was built using the wild-type 503 Lpp trimer as a template, further optimized using Targeted Molecular Dynamics (TMD) for 1 ns. For both 504 24 Lpp and Lpp +21 , the proteins were anchored in the OM via N-terminal acylation while the C-terminus of one 505 copy from each trimer was covalently linked to the PG. The systems generated were prepared for equilibration 506 using the following steps for 1 ns each: 1) minimization for 10,000 steps, 2) melting of lipid tails, 3) restraining 507 only the PG and the protein, and 4) restraining the PG and the protein backbone. Both systems were 508 equilibrated for 200 ns. 509 For each of the two systems (Lpp and Lpp +21 ), a new system was constructed with one Lpp trimer removed 510 and OmpA inserted into the OM. The full-length OmpA structure was taken from Ortiz-Suarez et al. 73  Scholarship. J.C.G. acknowledges support from the US National Institutes of Health (R01-GM123169). 539 Computational resources were provided through XSEDE (TG-MCB130173), which is supported by the US 540 National Science Foundation (NSF; ACI-1548562). This work also used the Hive cluster, which is supported 541 by the NSF (1828187)  showing that length is shown on the y-axis. The location of Lpp and the lengthened Lpp +21 are indicated. (B) 548 Whole cell lysates were prepared from the indicated strains and subject to SDS-PAGE and immunoblot 549 analysis with anti-Lpp antibodies and anti-OmpA antibodies. OmpA serves as a loading control. (C) The 550 JW5028 -Keio BW25113 strain with kan gene replacing a pseudogene background and isogenic Lpp +21 strain 551 (Fig. S1) were grown over 24 hours. The growth medium is M9, containing the indicated concentration of 552 sorbitol as an osmolyte. (D) Growth rates for the same strains were measured in rich (LB) growth media with 553 and without sorbitol over 20 hours. (E) The periplasmic width distribution of the indicated strains in 554 hyperosmotic conditions. While PG layer in the wild-type strain is a uniform thin electron dense layer, the PG 555 layer in the Lpp +21 strain is more diffuse and thicker. (F) Subtomogram averages of cell envelopes in 556 hyperosmotic conditions. Measurements from EM views evaluate the distance between OM and PG in the 557 Lpp +21 strain micrographs. The histogram depicts the frequency with which a given distance is observed 558 between the OM and PG. Lpp +21 mutant has an overall reduction in the level of periplasmic proteins.  PCR confirmation of the lpp +21 mutant strain (Methods). 596 597 Figure S2. Quantitation of Lpp and Lpp +21 isoforms. The sequence of the 21 residues inserted to create the 598 Lpp +21 isoform is also indicated. Mass spectrometry data for Lpp vs Lpp +21 was reanalyzed after extraction 599 from the whole cell proteomic data. Given the different tryptic peptides generated from the two isoforms of 600 Lpp, a shared peptide (red) was used to quantify the relative levels of each Lpp isoform in each of the strains. 601 The graphs document the relative levels of the peptide and show that the presence or absence of sorbitol in 602 the growth medium has no effect on the level of Lpp +21 relative to Lpp. wild-type and Lpp +21 strain for cell lysate and extracted outer membrane vesicles. Loading of each technical 608 replicate was normalized to OD 600. Representative data are shown from experiments performed in biological 609 triplicate. 610 611 Figure S4. A synthetic lethal screen to determine genes essential to Lpp +21 E. coli. An Hfr donor strain 612 carrying a selectable marker (cat) fused to lpp +21 , replacing the lpp ORF, is mated on agar plates with arrayed 613 Frecipients (384) per plate carrying a selectable marker (kan) replacing other ORF. Upon mating, cells are 614 subjected the first round of selection (intermediate selection) using antibiotic kanamycin and then further 615 subjected to a second round of selection using both antibiotics; (A) depicts images of representative plates 616 generated in each step of the procedure with imaging and manual analysis step, cross-referencing of single 617 gene knock outs and double recombinants, included. (B) depiction of the strains as cartoons generated in each 618 step of the procedure. (C) A representative mini screen of manually selected genes from the main synthetic 619 lethal screen. Sterility controls were included on each mini screen. The mini screen was performed in 384-pin 620 density with each clone arrayed in four biological replicates, each having four technical replicates (blue 621 boxes). Synthetic lethal mutants identified from the mini screen were further verified by PCR to confirm the 622 presence of both gene modifications and rule out partial duplication events. 623 624 Figure S5. Construction and characterization of the validation yiaD mutant. A kanamycin resistance 625 cassette was amplified from pKD4 using primers with overhangs complementary to upstream and downstream 626 29 of yiaD. The PCR fragment was electroporated in BW25113 cells harbouring the λ -red recombineering 627 plasmid (pKD46). Transformants were selected on kanamycin-resistant plates and verified by PCR (methods). 628 Primers flanking the yiaD gene confirm replacement of yiaD with kanamycin cassette and primers amplifying 629 lpp confirm lpp +21 replacement of lpp. The sequence information for all primers used are included in Table  630 S5. 631 632 Figure S6.