Analysis of SMALP co-extracted phospholipids shows distinct membrane environments for three classes of bacterial membrane protein

Biological characterisation of membrane proteins lags behind that of soluble proteins. This reflects issues with the traditional use of detergents for extraction, as the surrounding lipids are generally lost, with adverse structural and functional consequences. In contrast, styrene maleic acid (SMA) copolymers offer a detergent-free method for biological membrane solubilisation to produce SMA-lipid particles (SMALPs) containing membrane proteins together with their surrounding lipid environment. We report the development of a reverse-phase LC-MS/MS method for bacterial phospholipids and the first comparison of the profiles of SMALP co-extracted phospholipids from three exemplar bacterial membrane proteins with different topographies: FtsA (associated membrane protein), ZipA (single transmembrane helix), and PgpB (integral membrane protein). The data showed that while SMA treatment per se did not preferentially extract specific phospholipids from the membrane, SMALP-extracted ZipA showed an enrichment in phosphatidylethanolamines and depletion in cardiolipins compared to the bulk membrane lipid. Comparison of the phospholipid profiles of the 3 SMALP-extracted proteins revealed distinct lipid compositions for each protein: ZipA and PgpB were similar, but in FtsA samples longer chain phosphatidylglycerols and phosphatidylethanolamines were more abundant. This method offers novel information on the phospholipid interactions of these membrane proteins.

Membrane proteins have a central place in the function of the cell: they mediate key processes such as the transfer of ions, nutrients and signals into and out of the cell, and in microorganisms contribute to cell membrane biosynthesis and cell division. In view of the urgent need to develop novel antimicrobial agents, much interest has focused on proteins involved in the latter two processes, often using the Gram-negative bacterium Escherichia coli as a model system. For example, ZipA and FtsA act as anchors for the cytoskeletal bacterial tubulin homolog FtsZ, are part of the divisomal apparatus, and migrate to the division septum 1 . ZipA is a bitopic membrane protein harbouring a short N-terminal transmembrane helix 2 , whereas FtsA is thought to attach peripherally to the cell membrane by a short C-terminal amphipathic helix motif 3 . Another important membrane protein, PgpB, is an integral membrane lipid phosphatase with a key role in membrane biosynthesis by generation of phosphatidylglycerol (PG), but additionally with the ability to dephosphorylate the glycan lipid carrier undecaprenyl pyrophosphate 4 . The activity of PgpB depends on the acyl chain length of the lipid substrates and has been reported to be partially inhibited by the zwitterionic phospholipid phosphatidylethanolamine (PE), leading to the suggestion that PE acts as a regulator of PgpB activity to balance the amount of zwitterionic and anionic phospholipids in the membrane necessary for the survival and growth of E. coli 5,6 . Previous studies have shown that areas of high membrane curvature, such as the septum, contain specific lipids that are essential for correct membrane morphology Scientific RepoRts | (2019) 9:1813 | https://doi.org/10.1038/s41598-018-37962-0 during division 7 , and for FtsA and ZipA as well as PgpB, the local membrane lipid environment is likely to be critical for their correct structure and function. However, until recently it has been difficult to study them in a native condition or determine the associated lipids, because of the extraction and reconstitution methods commonly used for membrane proteins. Historically, studies of membrane protein structure and function have typically involved extraction using detergents such as n-dodecyl-β-D-maltopyranoside (DDM) that strip away membrane lipids to leave the protein in a detergent micelle. However, over the last 10 years there has been a re-evaluation of this method, as it is becoming clear that studies of membrane proteins should take into account their local environment, including the immediately surrounding lipids 8 . A growing number of studies have shown that membrane proteins associate closely with specific lipids in the membrane, and that this association is important for fundamental aspects of their function, including their folding, activity and interactions [9][10][11][12] , but data on which lipids a membrane protein is specifically associated with is still very limited; co-purified membrane lipids have only been detected in a small number of studies of detergent-purified membrane protein, as usually the detergent displaces these lipids 13 . A further difficulty in studying membrane proteins relates to their reconstitution from a detergent micelle into lipid bilayer. Apart from the problem that these artificial liposomes may not accurately reproduce the native lipid environment of the protein, such reconstitutions are often complex and yield samples that can be difficult to study. For example, the presence of lipid bilayers can produce scattering artefacts which require substantial instrumental modifications for many optical techniques, particularly those, like circular dichroism (CD), that operate in the ultraviolet 14 . In addition, native mass spectrometry is developing as a promising method for the detailed analysis of closely associated lipids 15,16 , including in lipoprotein nanodisc systems described below 17 , and there is a growing number of laboratories that have the advanced technology and experience to perform this.
An approach that has made a step-change in the study of membrane proteins is the use of Styrene Maleic Acid/ Lipid Particles (SMALPs). The SMALP method uses an amphipathic polymer, styrene maleic acid co-polymer (SMA), to extract membrane proteins directly from native membranes 8 . In contrast to detergent, the SMA reagent extracts the protein within the portion of membrane that surrounds it, thereby in theory preserving the native lipid environment 18 . This assertion has been strengthened by the first high resolution cryo-EM structure of a SMALP solubilised protein, bacterial alternative complex III, which showed the electron densities of several phospholipids tightly bound to the protein, although the precise phospholipid class was not confirmed 19 . Extraction of the lipids from SMALP particles therefore generates the perfect sample for studies of the local lipid environment surrounding a protein. Several studies utilizing SMALP-extracted proteins have involved some form of lipid analysis, including phosphate quantification 20-23 , thin layer chromatography [22][23][24][25][26][27] , gas chromatography 25,28 , and mass spectrometry (MS) [28][29][30][31] . Some studies have reported the enrichment of certain types of lipids with the membrane proteins of interest 22,25,[28][29][30] , highlighting the usefulness of the SMALP approach in studying the protein-associated lipidome. Most studies provided limited detail on the exact lipid species, but some recent work using mass spectrometry has enabled a more thorough profiling of the phospholipids present in SMALPs 30,31 , illustrating the power of this approach to address the question of local lipid environment of membrane proteins.
Therefore, the aim of this study was to carry out the first SMALP-based extraction of ZipA, FtsA, and PgpB from E. coli membranes, and develop an effective reverse phase liquid chromatography tandem mass spectrometry (LC-MS/MS) method to characterise and compare their co-extracted lipids. A key goal was to determine whether the integral membrane protein (PgpB) associates with different membrane phospholipids to the bitopic protein ZipA and peripheral membrane protein FtsA.

Results
Development of a target lipidomics method for use on sMALp samples. The composition of wild type E. coli K-12 membrane lipids has previously been characterised by LC-MS. Oursel et al. used reverse-phase LC-MS/MS to separate PE and phosphatidylglycerol (PG), but cardiolipin (CL; sometimes called di-PG) was not reported 32 , whereas in a later study a normal-phase LC-MS method was employed to examine CL alone 33 . To improve on this, we developed a reverse-phase LC-MS/MS method to separate simultaneouly the three main classes of E. coli membrane phospholipids: PE, PG and CL. The main challenge was to find a column and solvent system capable of separating PG and CL, which are structurally similar, while also obtaining good ionisation of the phospholipids. This was achieved using a gradient of tetrahydrofuran (THF)/methanol (MeOH)/water (H 2 O) containing 10 mM ammonium acetate on a Luna ® C8(2) column. The sensitivity of the method was tested using an exponential dilution (in the concentration range of 0.039-2.5 ng on column, i.e. per injection, per individual lipid) of a lipid mixture containing equal amounts of three synthetic phospholipids: PE(16:0/18:1), PG(16:0/18:1) and CL(18:1/18:1/18:1/18:1). The limit of quantification (LoQ) was determined to be between 0.039-0.078 ng for both PE(16:0/18:1) and PG(16:0/18:1), while CL(72:4) showed a LoQ approximately 10-fold higher at 0.31-0.63 ng (Supplementary Table S1). The LC-MS/MS method developed was also appropriate for the separation and detection of phosphatidylcholines (PC) (Supplementary Fig. 1), which are abundant in eukaryotic membranes and appear to be present in some bacterial species 34 , demonstrating the versatility of this reverse-phase LC-MS/ MS method.
The method was then applied to investigate the extraction of E. coli BL21(DE3) membrane lipids. A representative total ion chromatogram (TIC) illustrating the effective separation of the membrane lipids is shown in Fig. 1a, and it can be seen that the phospholipids eluted based on hydrophobicity with the elution order: PG > PE > CL. Figure 1b-d show the variety of molecular phospholipid species that were detected in each region of the chromatogram. The identity of the phospholipids were initially assigned using their nominal mass-to-charge ratio (m/z), with PE occurring at even m/z values, and PG and CL occurring at odd m/z values; the CL species were doubly charged and could be distinguished from the PG species by the spacing of the isotopic peaks. To confirm the identity and provide structural characterisation of the different molecular species, tandem MS in negative ion mode was employed; an example spectrum is shown in Fig. 1e for one of the most abundant PE species (PE(34:1)) at  Fig. 1f and a full list of phospholipid species and their m/z is given in Supplementary Table S2).
Do lipids extracted using sMALps match those found in the membrane? An important consideration for the use of SMALPs to sample the lipids in a membrane is the assumption that the SMA-extraction process has no selectivity for any particular lipid. To determine whether this was the case and therefore that SMALPs are a suitable tool for lipidomic studies, E. coli BL21(DE3) membranes were separated into two fractions, one of which was subjected to SMA-treatment to generate generic membrane SMALPs prior to lipid extraction, and one in which the lipids were extracted directly without SMA-treatment. The lipids extracted from both fractions were analysed using LC-MS/MS. As MS is not inherently quantitative, three non-natural lipid internal standards, PE(12:0/13:0), PG(12:0/13:0) and CL(14:1/14:1/14:1/15:1), were added to the samples prior to the lipid extraction to allow downstream data normalisation and relative quantification of each phospholipid class. Figure 2 demonstrates that the overall phospholipid profiles for extracted BL21(DE3) membrane phospholipids with or without SMA treatment were highly comparable, with no major depletion of specific lipid types in the SMA-treated sample. The only significant difference was a small decrease in PG(33:1) in the SMALP sample. This suggests that,  (Fig. 3d). The profile of lipids co-extracted with ZipA in SMALPs from the overexpressing cells was also investigated. SMALP-purified ZipA in this system showed an overall class profile most similar to the membrane from uninduced cells, but within the PE and PG classes the profile was most similar to that of the ZipA induced membrane, whereas the CL profile remained most similar to the uninduced membrane. This suggests that individual proteins associate preferentially with a subset of the lipids present in the overall membrane, which can be co-extracted with the protein in SMALPs.

Comparison of the SMALP lipidome of ZipA, FtsA and PgpB.
To investigate whether the phospholipids preferentially extracted with other membrane proteins were the same, SMALP co-extracted membrane lipids from FtsA and PgpB were examined using the same LC-MS/MS method and compared to those found with SMALP-ZipA. For all three proteins, good enrichment after IMAC was achieved ( Supplementary Fig. 2). At the level of the phospholipid classes, no significant differences between the three E. coli membrane proteins were observed (Fig. 4a). Nevertheless, significant differences were observed between the phospholipid profiles in the PE and PG classes. The phospholipid profiles of ZipA and PgpB were similar, showing a higher abundance in monounsaturated phospholipid species in both the PE and PG classes. ZipA and FtsA showed substantial differences in their phospholipid profiles: PE(30:0), PE(32:1), and PG(32:1) were more abundant in ZipA, especially PE(32:1) and PG(32:1). In contrast, PE(35:2), PE(35:1), PE(36:1), PE(37:2), PG(36:2), and PG(36:1) were observed to have higher abundance in FtsA.

Discussion
In this work, we showed that E. coli membrane proteins ZipA, FtsA and PgpB are amenable to the solubilisation and purification using the 2:1 SMA copolymer ( Supplementary Fig. 2); to the best of our knowledge, the work on PgpB represents the first report to-date of an integral membrane lipid phosphatase directly extracted and examined in a detergent-free membrane mimetic system. The solubilisation efficiency of the 2:1 SMA copolymer was previously reported to be comparable to DDM (around 55%) for several other bacterial membrane proteins overexpressed in E. coli, namely ZipA, BmrA (multidrug efflux pump from Bacillus subtilis), and LeuT (amino acid:Na + symporter from Aquifex aeolicus) 38 . Previously, lower overall yields of purified protein have been noted when employing the SMALP technology 18 , suggesting a requirement for future improvement in the protein solubilisation and purification protocol. Several other variants of the SMA copolymers which differ in the S:MA ratio and molecular weight were also evaluated by Morrison et al. 38 , but the 2:1 variant gave the best yield, purity and functional conservation of target membrane proteins upon a single step of IMAC purification. Importantly, in this work the SMA copolymer was found to not preferentially solubilise specific types of E. coli membrane phospholipids. Similar observations have also been reported for the cell membranes of R. sphaeroides 23 , S. cerevisiae mitochondria 26,27 and synthetic PC/PE large unilamellar vesicles 39 using the SMA copolymer. This fact highlights the utility of the SMA copolymer as an alternative to detergents for biological membrane solubilisation, which facilitates subsequent membrane lipid characterisation without selective lipid bias, at least for E. coli, which is a common host for recombinant protein production.
A key finding of this work was that significant differences were observed between the phospholipid profiles in the PE and PG classes for the three E. coli membrane proteins. The phospholipid profiles of ZipA and PgpB were similar, showing a higher abundance in monounsaturated phospholipid species in both the PE and PG classes,  41 . The interaction between phospholipids and FtsA was suggested to be essential for rapid ATP hydrolysis in vitro. FtsA and other peripheral membrane proteins, such as SpoVM and DivIVA (two curvature-sensitive proteins in B. subtilis), have been suggested to associate with the cell membrane by inserting the hydrophobic residues of their proposed amphiphatic helix into the cell membrane 41,42 . Possibly the higher abundance of phospholipids with longer fatty acyl chains in FtsA (35-37 carbons vs 32 carbons in ZipA) might be required to facilitate this. Given that the sum of each of the phospholipid classes were similar for these three membrane proteins, yet significant differences were observed in terms of different fatty acyl profiles, this suggests a clear preference for different lipid chain lengths and/or degree of unsaturation (rather than headgroup) for these bacterial membrane proteins, which is an important novel finding.
It is known that the annular lipid belt of proteins can undergo exchange with the bulk lipids of the biological membrane. In the absence of any selectivity between the lipids and protein, the annular belt largely reflects the composition and properties of the surrounding membrane 43 . Recent publications on the exchange of lipids in SMALPs (without membrane proteins) have suggested that conclusions cannot be drawn on particular lipid enrichment of the target proteins unless they are tightly bound to the proteins 39,44,45 . The 2:1 SMA copolymer was found to have a much slower collisional lipid transfer than the 3:1 counterpart, owing to higher coloumbic repulsion 46 . So far, no evidence of lipid transfer has been reported for SMALP-protein nanodiscs. In the future, different SMA variants could be investigated to tune the size of the SMALP nanodiscs that encapsulate varying amount of membrane lipids, to examine the lipids that are most closely associated with the target membrane proteins. Alterntively, complementary studies using native MS could be employed to screen for the most tightly bound lipids upon controlled delipidation using detergents 47 . An interesting additional finding of our work was that changes in total membrane lipid composition occurred when ZipA was overexpressed in E. coli. However, the observed increase in CL content did not seem to be locally associated with ZipA, as there was no significant difference between the CL profile for the SMALP-purified ZipA and the uninduced membrane control. The overexpression of ZipA 48 and FtsA 41 has been reported to result in multilayered membrane inclusions within the cytoplasm, and similar extra membrane structures have been observed following overexpression of several other membrane proteins, including the F 1 F o ATP synthase, fumarate reductase, glycerophosphate acyltransferase PlsB, mannitol permease MltA, alkane hydrolase AlkB, chemotaxis receptor Tsr, and lipid glycosyltransferases 49 . Thus overexpression of the target membrane proteins using conventional IPTG-induction techniques may represent a limitation of this work, as the co-extracted lipids might not represent the exact spatio-temporal picture of the local lipid environment of the target proteins when in vivo in wild-type cells. Nevertheless, specific lipid enrichment has been reported in other SMALP studies 25,30 , indicating that the SMALP co-extracted lipidomes are not merely a snapshot of the native biological membrane at the time of solubilisation, but can indeed reflect specific protein-lipid interactions. This supports the use of SMALP technology as a valid means to probe protein-lipid interactions.
Together, the data reported here show how the combination of MS lipidomic analysis and SMALP encapsulation provides an analytical method that is capable of both qualitative and quantitaive analysis of the protein-associated lipidome for any sample that can be SMA solubilised and enriched using a suitable affinity method (e.g. recombinant tags or protein specific antibodies), could be applied to any membrane protein sample under investigation. Moreover, it demonstrates for the first time that 3 membrane proteins of different types associate preferentially with specific phospholipid profiles.

Methods and Materials
Unless stated otherwise, all chemicals and reagents used were purchased from Sigma-Aldrich (Dorset, UK), Fisher Scientific (Leicestershire, UK), or Melford Biolaboratories (Suffolk, UK). All organic solvents used in the MS experiments were of HPLC grade and LC-MS grade H 2 O was used.
Membrane protein overexpression and purification. E. coli ZipA and FtsA were overexpressed and purified based on methods of Lee et al. 18 and PgpB based on Touzé et al. 6 . The ZipA and FtsA proteins were purified from E. coli BL21(DE3) cells following overnight incubation at 25 °C post IPTG induction, while the PgpB was purified from E. coli C43(DE3) cells after 4 h incubation at 37 °C post IPTG induction. In outline, the bacterial cells were lysed using steps 1-5 of the protocol of Harder and Fotiadis 50 . Cells were lysed using a French press, unlysed cells removed by centrifugation at 10,000 g, and the membranes then pelleted by centrifugation at 100,000 g. Membranes were washed once in lysis buffer and once with buffered 300 mM NaCl (to remove adherent proteins) by resuspension and homogenization followed by centrifugation at 100,000 g. The pellet was then fully resuspended (using a Potter homogenizer) in the buffer for SMALP extraction. The overexpressed ZipA, FtsA and PgpB were then solubilised by addition of SMA 2000 to a final concentration of 2.5% (prepared as per Lee et al. 44 ). The clarified SMALP preparations were centrifuged at 100,000 g at 4 °C for 45 min to removed the insoluble fraction and produce a purified SMALP preparation. Samples were taken at this stage for analysis of total SMALP-extracted membrane lipid. The SMALP extracted ZipA and FtsA were subsequently purified by nickel-nitrilotriacetic acid immobilised metal affinity chromatography (Ni-NTA IMAC) (Generon, UK or Qiagen, USA) followed by size exclusion chromatography, while only Ni-NTA IMAC was used to purified the SMALP extracted PgpB. The purified proteins were analysed by 12% sodium dodecyl sulfate polyacrylamide gel electrophoresis (SDS-PAGE) ( Supplementary Fig. 2). Protein concentrations were estimated using either the A280 method on a NanoDrop ND-1000 spectrophotometer or the Pierce bicinchoninic acid protein assay (Thermo Scientific, USA). Lipid extraction was carried out based on a modified Folch method 51 . Firstly, 200 µL ice-cold MeOH (with 0.005% w/v BHT) was added to 400 µL sample (containing IS), followed by 400 µL ice-cold chloroform (CHCl 3 ) and lastly 150 µL ice-cold ultrapure H 2 O. Each sample was vortexed for 20 s and sonicated in ice bath for 15 min, after each solvent addition to facilitate rigorous mixing and extraction. The samples were centrifuged at 16,000 × g for 5 min at 4 °C to achieve phase separation. The lipid-containing (lower) organic layer was collected and the remaining aqueous phase was subjected to a second extraction by adding 400 µL ice-cold CHCl 3 . The samples were centrifuged again at 16,000 × g for 5 min at 4 °C and the organic layer was collected. The resulting lipid extract was dried under a gentle stream of N 2 gas and stored at −20 o C until analysis. Liquid chromatography-tandem mass spectrometry (LC-Ms/Ms). Samples were reconstituted in 1 mL of 9:1 v/v MeOH:CHCl 3 , followed by appropriate dilution (typically 2 or 5-fold) in the chromatography starting solvent detailed below. Ten µL of reconstituted lipid extract was injected onto a Luna C8(2) 100 Å column, 150 × 1 mm (length × internal diameter), 3 µm (particle size) (Phenomenex, USA) at a flow rate of 50 µL/ min solvent A. Lipid separation was achieved using a gradient of the following mobile phases: THF:MeOH:H 2 O (Solvent A: 3:2:5 v/v/v; Solvent B: 7:2:1 v/v/v) containing 10 mM ammonium acetate. The gradient was formulated as follows: 40% B for 4 min, 40% B to 60% B over 6 min, 60% B to 100% B over 15 min, and hold 100% B for 5 min, 100% B to 40% B over 2 min, then reequilibrated to 40% B over 13 min. A QTRAP 5500 mass spectrometer (AB Sciex, UK) was utilised with instrument parameters as follows: curtain gas: 35 psi, ion spray voltage: −4.5 kV (negative ion mode); temperature: 150 °C; nebuliser gas (GS1): 13 psi; auxiliary/turbo gas (GS2): 0 psi; declustering potential: 50 V; entrance potential: 10 V; collision energy: 10 eV. Information-dependent acquisition (IDA) was employed to obtain MS/MS spectra during the LC run for structural characterisation of the lipid species in the negative ion mode. A general survey scan in the EMS mode (with dynamic fill time) at m/z 400-1000 Da was used to trigger IDA on one to two most intense ion peak(s) based on the following selection criteria: over m/z of 400 Da, 1000 cps, and 20 s; maximum occurrence of 2; mass tolerance of 250 mDa; excluding isotopes within 4 Da. This was followed by enhanced product ion (EPI) scans using the linear ion trap with a collision energy of 45 eV in negative ion mode.

Data analysis. Mass spectra inspection and data extraction involving relevant peak integration with
Gaussian smoothing width of 0-2 points on each extracted ion chromatogram (XIC) were performed manually using the PeakView 2.2 software (AB Sciex, UK). The identification and assignment of each phospholipid species and corresponding molecular structures were facilitated by the IDA LC-MS/MS analysis with subsequent data matching to the LIPID MAPS Structure Database http://www.lipidmaps.org/data/databases.html 52 . A two-step normalisation approach was employed. Firstly, the peak intensity for each phospholipid species was normalised against the peak intensity for the corresponding class-specific internal standard. This data was then further normalised against the total phospholipid ion intensity per sample/replicate or amount of protein for which lipid extraction was performed to facilitate cross-sample comparison.
statistical analysis. Statistical analysis as implemented in either in Microsoft Excel or Prism 7 (GraphPad Software, USA), including unpaired t-tests: single or multiple (with Bonferroni correction) and one-way ANOVA with Tukey's multiple comparisons test.