Membrane composition influences the topology bias of bacterial integral membrane proteins

Biochimica et Biophysica Acta 1828 (2013) 260–270

Contents lists available at SciVerse ScienceDirect

Biochimica et Biophysica Acta journal homepage: www.elsevier.com/locate/bbamem

Membrane composition influences the topology bias of bacterial integral membrane proteins Denice C. Bay, Raymond J. Turner ⁎ Department of Biological Sciences, University of Calgary, Calgary, Alberta, Canada T2N 1N4

a r t i c l e

i n f o

Article history: Received 23 April 2012 Received in revised form 31 August 2012 Accepted 7 September 2012 Available online 13 September 2012 Keywords: Lipid composition Membrane protein folding Positive inside rule Dual topology Small multidrug resistance (SMR) proteins EmrE Paired SMR

a b s t r a c t Small multidrug resistance (SMR) protein family members confer bacterial resistance to toxic antiseptics and are believed to function as dual topology oligomers. If dual topology is essential for SMR activity, then the topology bias should change as bacterial membrane lipid compositions alter to maintain a “neutral” topology bias. To test this hypothesis, a bioinformatic analysis of bacterial SMR protein sequences was performed to determine a membrane protein topology based on charged amino acid residues within loops, and termini regions according to the positive inside rule. Three bacterial lipid membrane parameters were examined, providing the proportion of polar lipid head group charges at the membrane surface (PLH), the relative hydrophobic fatty acid length (FAL), and the proportion of fatty acid unsaturation (FAU). Our analysis indicates that individual SMR pairs, and to a lesser extent SMR singleton topology biases, are significantly correlated to increasing PLH, FAL and FAU differences validating the hypothesis. Correlations between the topology biases of SMR proteins identified in Gram + compared to Gram− species and each lipid parameter demonstrated a linear inverse relationship. © 2012 Elsevier B.V. All rights reserved.

1. Introduction Bacterial membrane compositions vary significantly when comparing the lipids of Gram positive (Gram +) to Gram negative (Gram−) bacteria [1]. Due to this variation, the composition of membrane phospholipids, including acyl chain length, the degree of unsaturation and charged polar head groups are often used to aid bacterial classification. Phospholipid diversity is essential to understanding factors that influence the function, insertion, and topology of integral membrane proteins. Features such as the degree of phospholipid head group charge (as reviewed by [2]), fatty acid chain length and fatty acid unsaturation (as reviewed by [3]) have all been shown to alter the function and folding of membrane proteins, emphasizing the importance of membrane composition with integral membrane protein activity. In particular, the balance of anionic phospholipids within the membrane, namely phosphatidic acid (net − 1 charge), phosphatidyl glycerol (net − 1), and cardiolipin (net − 2 charge), appears to play an important role in the insertion and transmembrane segment arrangements of integral membrane proteins such as leader peptidase (Lep) [4], phenylalanine permease (PheP) [5], gamma-aminobutyric acid permease (GabP) [6], lactose permease Y transporter (LacY) [7] and potassium channel protein KcsA

⁎ Corresponding author at: BI 156 Biological Sciences Building, 2500 University Drive N.W., Department of Biological Sciences, University of Calgary, Calgary, Alberta, Canada T2N 1N4. Tel.: +1 403 220 4308; fax: +1 403 289 9311. E-mail address: [email protected] (R.J. Turner). 0005-2736/$ – see front matter © 2012 Elsevier B.V. All rights reserved. http://dx.doi.org/10.1016/j.bbamem.2012.09.003

[8]. These studies indicate that increasing anionic phospholipid content in the membrane can alter the folded state of the protein and impact overall function. In general, integral membrane protein topology and transmembrane segment (TMS) insertion can be determined from the primary sequence and secondary structure content following ‘the positive inside rule’ (as reviewed by [9]). This rule states that the amount of positively charged residues (lysine; K and arginine; R) will determine the orientation of protein TMS insertion in the membrane [10,11] (Fig. 1). Loops and/or termini with the greatest abundance of positive residues are expected to reside inside the cell, facing the cytoplasm and away from the proton enriched periplasm [12]. Hence, the topology of any membrane protein can be calculated using the equations provided in Fig. 1. This calculation has enabled examination of entire bacterial membrane proteomes and determined that the vast majority of integral membrane proteins adopt a preference or bias for one orientation within the membrane [13,14]. However, a small proportion of membrane proteins can adopt a dual topology orientation, where they show a neutral topology bias that permits protein membrane insertion in either direction (Fig. 1B). An example of a dual topology integral membrane protein, belongs to a family of multidrug transporters known as small multidrug resistance (SMR) proteins. Members of this family confer host resistance to antiseptics and antimicrobials that possess a permanently charged cation(s) referred to as a quaternary cation compound (QCC) though proton motive force (as reviewed by [15]). The SMR protein family is one of 14 other phylogenetically distinct secondary active transporter protein families classified within drug and metabolite

D.C. Bay, R.J. Turner / Biochimica et Biophysica Acta 1828 (2013) 260–270

FAU

Loop1

A

Loop3

PLH FAL

1 2 3 4

Loop2 N B SMR Singletons

C

KR = (+0) “in” KR = (-0) “out” N

OUT

N

C

C

IN

C SMR Pairs OUT

N

KR = (-) out N

C

KR = (+) in

D

C

IN

KR bias = (∑ KR) Termini+ Loop2 - (∑ KR) loops1+3 KR bias > 0 = Termini “in” KR bias < 0 = Termini “out” Net bias = (∑KR -∑ED)Termini+Loop2 - (∑KR- ∑ED)loops1+3 Net bias > 0 = Termini “in” Net bias < 0 = Termini “out”

Fig. 1. Diagrams of various SMR protein topology biases and equations used to calculate insertion biases. A) A diagram of a lipid bilayer containing a SMR protein. A single SMR protein monomer is shown in black, where labelled lines indicate N- or C‐termini and loop (1–3) regions connected to each of the four transmembrane α-helices (numbered cylinders) of the protein. Phospholipids are represented in grey, where polar head groups (circles) are connected to two fatty acids tails. Other polar lipids such as cardiolipin, are represented as oval (polar head group) connected to four fatty acid tails. Each lipid composition parameter examined in this study is indicated by an arrow and shows mean polar lipid head groups (PLH), relative fatty acid length (FAL) and relative fatty acid unsaturation (FAU). Panels B and C show a membrane orientation diagram of an SMR singleton (B) and SMR pair (C) according to the positive inside rule KR bias calculation provided in D. D) Equations used to determine KR topology bias and net topology bias. Each bias estimates the orientation according to the sum of charged amino acid residues, (KR bias calculates K/R residues only; net charge calculates the net charge difference between K/R from E/D) located at each N–C‐termini and in loop region.

transporter (DMT) superfamily [16,17]. Members of the SMR protein family are composed of four highly hydrophobic α-helical TMS linked together by relatively short loops. The genes encoding SMR proteins are frequently located on conserved 3′ regions of mobile genetic elements known as integrons, enhancing SMR transmission to unrelated bacterial species [18,19]. The SMR protein family is divided into three major subclasses; the small multidrug protein (SMP), the suppressor of groEL mutation (SUG), and paired small multidrug resistance protein

261

(PSMR) subclasses. Both SMP and SUG subclasses confer host resistance as a single protein, known as a ‘singleton’ based on previous SMR topology studies [13,14]. The majority of experimental studies of SMP subclass members have focused on the archetypical protein, EmrE in Escherichia coli, which confers host resistance to a broad range of structurally diverse QCC. SUG subclass members, such as E. coli SugE, confer host resistance to a narrow subset of QCC in comparison to SMP [20] but have a broader bacterial distribution than SMP [21]. Finally, PSMR subclass members, MdtI (YdgE) and MdtJ (YdgF) in E. coli, confer host resistance to QCC by the expression of two distinct SMR genes [22]. PSMR proteins reside within the membrane as a paired heterooligomeric complex where each protein adopts a fixed antiparallel topology/ orientation from the other [14]. Topological analyses of other SMR pairs, have mainly focused on the Gram positive Bacillus subtilis SMR pair EbrA and EbrB, which also demonstrates an antiparallel orientation [23–27]. The focus of this study is to determine if the host bacterial lipid composition influences integral membrane protein topology and determine what lipid parameter(s) are specifically involved. Our working hypothesis is that fixed/single orientation integral membrane proteins are expected to have topology biases that shift in value with respect to membrane composition. By extension, if a dual topology orientation is an essential feature for protein function then the neutral topology bias should also vary in response to changing membrane compositions to maintain neutrality. The SMR protein family is an ideal candidate for such an investigation since members of this family exist as either dual topology (or antiparallel topology) singletons or as single/fixed topology pairs. To test this hypothesis, SMR topology biases were determined from a bioinformatic analysis of 1320 bacterial SMR family protein sequences from diverse taxa representing major bacterial phyla. A matrix of positively (and negatively) charged amino acid residues (Lys and Arg) found within Nand C‐termini and loops 1–3 of each protein was assembled (Fig. 1). In parallel, a phospholipid composition dataset was collected for bacteria known to encode an SMR sequence. This lipid dataset summarized the proportion of polar lipid head group charges (PLH), the relative fatty acid tail length (FAL), and the proportion of fatty acid unsaturation (FAU) in various bacterial plasma membranes (Fig. 2). After statistical assessment of the topology-lipid matrices by correlational analyses and hierarchical clustering, we propose that all three lipid composition parameters influence the positive inside rule [12] and thus support the overall hypothesis. 2. Materials and methods 2.1. SMR protein sequence topology bias dataset acquisition and matrix assembly A total of 1320 bacterial SMR protein sequences (as of May 2011) were collected for this study by expanding upon a SMR protein dataset of 685 sequences assembled from a previous study [21]. All additional SMR protein sequences were identified by performing tBLASTn searches of the NCBI microbial genome sequencing database using either E. coli EmrE (P23895) or E. coli SugE (AAC46453) protein sequences. All SMR protein sequences were initially aligned using ClustalW (2.0) and then manually edited using the multiple alignment programme Jalview [28,29]. Aligned SMR proteins were separated into five groups based on SMR homology with known representatives from each subclass (SMP, SUG, and PSMR: YdgEF, EbrAB, YkkCD) using the phylogenetic Neighbour Joining analysis method available in the PHYLIP software package [30]. A summary of SMR distribution within each representative phylum included in this study is provided in Supplementary Table 1. Estimated TMS regions in each protein within the SMR protein dataset were predicted using the TMHMM v2.0 programme [31,32] and the online SOSUI server [33]. All SMR protein sequences in the

262

D.C. Bay, R.J. Turner / Biochimica et Biophysica Acta 1828 (2013) 260–270

Mean % Phospholipid head group charge (PLH)

A 100% +1 PLH

75%

-2 PLH

50%

-1 PLH

La At Cf Ba De Bc ε

Cl α

β

δ

γ

Pl Cy

Phylum

Mean fatty acid length (FAL) (Å)

B

60 50 40 30 20 10 0

α

La Cf At Bc

ε

Ba

γ

Cl

β

δ

De

δ

ε

Bc

At

Ba

Phylum

Mean % fatty acid Unaturation (FAU)

C

100% 80% 60% 40% 20% 0%

2.2. Assembly of membrane composition matrices used for this study

0 PLH

25% 0%

amino acid residues (KR and ED) in loops 1–3 and N- and C‐termini were also sorted according to host taxonomy. Topology biases for either KR or net biases sorted according to the parameters described above were used to build the datasets/matrices described in the results.

α

Cf

Cl

γ

La De

β

Phylum Fig. 2. A summary of bacterial membrane phospholipid compositions determined for this study. Bacterial phyla are shown on the x‐axis of all panels (A–C) and are abbreviated as follows: acinetobacteria (At), bacilli (Ba), lactobacilli (La), clostridia (Cl), chloroflexi (Cf), chlorobia (Ch), deinococci (De), planctomycetia (Pl), bacteroidetes (Bc), acidobateria (Ac), α-proteobacteria (α), β‐proteobacteria (β), γ‐proteobacteria (γ), δ-proteobacteria (δ), and ε-proteobacteria (ε).In all panels phyla are sorted from least to greatest percentage or value. A) The mean percentage of each phospholipid (PL) net polar head group charge (PLH) according to each bacterial phylum surveyed. PLH net charges are represented by neutral/zwitterionic (0 PLH; black), net −1 negatively charged head groups (−1 PLH; light grey), and net −2 charged head groups (−2 PLH; dark grey). B) The mean fatty acid length (FAL) in angstroms (Å) of various bacterial phyla. Standard error within phyla is represented as vertical bars. C) The mean fatty acid tail saturation (light grey) and unsaturation (dark grey) percentage for bacterial phyla surveyed in this study.

alignment possessed a total of four α-helical TMS by either prediction method. Overlapping TMS regions determined by each programme were used to set the boundaries for NH4- and CO2‐termini, loops 1, 2 and 3 in the overall SMR alignment. All positively charged (Lys and Arg) and negatively charged (Asp and Glu) amino acids were tallied in each terminus and loop region for each protein. Charged amino acid residue totals from each of the five SMR termini/loop regions were used to calculate the positive inside rule's KR topology bias (K and R residues only). Additionally, a net topology bias (where K + R residues are subtracted from E + D) was also calculated according to the equations shown in Fig. 1. Mean values for KR and net biases are provided in Supplementary Table 2. The accuracy of our KR topology bias (KR bias) calculations were confirmed by comparing them to previously published KR bias values for E. coli SMR proteins EmrE (− 2), SugE (− 1), MdtI/YdgE (− 6) and MdtJ/YdgF (+ 5) [13]. KR and net bias values were sorted according to bacterial phylum, class, genus and species to calculate the mean and median bias values for each taxonomic category. The amount of charged

The bacterial membrane composition dataset used for this analysis was assembled using published experimentally determined fatty acid composition and/or phospholipid or polar lipid head group composition data as provided in Supplementary Tables 3 and 4. This lipid composition dataset was constructed by collecting fatty acid content (by mol%) and polar lipid content (by relative % abundance) from characterized plasma membranes of 171 bacterial species known to possess SMR proteins (Supplementary Table 4). Lipid compositions specific to the plasma membrane only were used in this study to prevent possible lipid contamination by the outer membrane of Gram negative organisms in particular. One exception to this rule were all Planctomycete species included in the analyses, which possess an intracytoplasmic membrane or ‘nuclear membrane’ that is typically found in membrane isolations since it is continuous to the plasma membrane during its cell cycle [34]. Due to the limited or incomplete characterization of most bacterial plasma membranes, the lipid parameters examined in this study cannot reflect asymmetrical lipid distributions that are present in many organisms [35]. Fatty acid composition information was separated into two categories for this study: hydrocarbon length (ranging from C12 to C22) and hydrocarbon unsaturation. Each fatty acid category was used to determine the relative fatty acid length (FAL) and the relative fatty acid unsaturation (FAU) of each bacterial plasma membrane. FAL values were calculated in angstroms (Å) based on the mean carbon chain length of the two most abundant fatty acids present by mol%. The length of each fatty acid chain was determined by the following equation [(∑(cos60°· 1.54 Å · 2)C\Cn) + (cos60°· 1.10 Å)C\H]; where, n is the length of the hydrocarbon chain (ranging from C12 to C22), a C\C bond length of 1.54 Å, C\H length of 1.10 Å and a C\C bond angle of 120°. For unsaturated fatty acid chain length, the equation above was modified as follows: [(∑(cos60°· 1.54 Å · 2)C\Cn−x) + (cos60°· 1.37 Å) + (cos30°(cos60°· 1.54 Å · 2)C\Cx + (cos60°· 1.10 Å)C\H)]; where x indicates the number of carbon atoms after the C_C double bond from the carboxyl end. Both calculations assume that the membrane of each bacterial species was in a fully hydrated state, as a result this value may not represent the experimentally determined hydrophobic thickness of bacterial plasma membranes. The relative FAU of each bacterial species was calculated as a percentage and determined from the sum of the mol% unsaturated fatty acid values divided by the total fatty acid mol% value. The polar lipid head group charge (PLH) of a bacterial membrane was determined from 171 published membrane compositions of various bacterial species as detailed in Supplementary Table 4. PLH values were separated according to the net charge, where − 2 included diphosphatidylglycerol/cardiolipin; − 1 included phosphatidyl‐glycerol, phosphatidyl‐serine, phosphatidyl‐inositol, phosphatidyl‐inositolmannoside; 0 included phosphatidylethanolamine, lysophosphatidyl‐ethanolamine, phosphatidylmannosyl‐ethanolamine, phosphatidyl‐choline, aminoglycolipid; and +1 included lysylphosphatidyl‐glycerol, sphingolipid, aminoglycolipid. The relative proportion of PLH groups present in the membrane was provided as a relative percentage. When approximate PLH amounts were not directly stated in published articles or tables, PLH values were normalized to 100% based on thin layer chromatography (TLC) or by high performance/pressure liquid chromatography (HPLC) data by dividing the quantified intensity of a particular lipid type by the total quantified values. In some cases, polar lipid data was provided qualitatively by a “+”; if TLC images were

D.C. Bay, R.J. Turner / Biochimica et Biophysica Acta 1828 (2013) 260–270

263

included in these sources lipid spots were quantified manually using KODAC 1D gel imaging software. If TLC or HPLC data was not included in the published sources using “+” polar lipid designation, all data was excluded from the lipid dataset. FAL, FAU, and PLH values determined from each bacterial species membrane were sorted into groups according to their bacterial phylum, class, and species and used to build matrices/data frames for further statistical analysis. An agglomerative hierarchical clustering analysis of the lipid dataset itself is provided in Supplementary Fig. 1. A summary of lipid parameter distributions across various bacteria phyla is shown in Fig. 2.

titech.ac.jp/~shimo/prog/pvclust/) and the function ‘pvclust(data, method.hclust = “ward”, method.dist= “euclidean”)’. The significance of Ward hierarchical clusters (shown at dendrogram nodes on the x-axis) were assessed from bootstrapped (100 or 1000 replicates) p-values (α = 0.90) and the significance of each cluster is shown as a percentage beside each node (Supplementary Figs. 1, 3, and 4). Similar clustering divisions of the major nodes were also obtained using a maximum likelihood based model for distance relationships in hierarchical clustering analysis performed for this study (data not shown).

2.3. Statistical analysis of SMR diversity, SMR topology bias, and lipid composition

3.1. The topology bias of SMR family members varies across bacterial phyla

Statistical analysis of SMR topology-lipid matrices were performed using ‘R’ version 2.14.1 including ‘R studio’ software (http://www.rproject.org) and Microsoft Excel 2007 edition. Since all protein topology biases are integers by calculation, a mean topology bias value for either the KR or net topology bias was used to represent each of 171 total bacterial phyla surveyed in the dataset. Mean KR and net topology bias datasets were subdivided according to phylum, class, and genus for matrix assembly using R statistics ‘data.matrix(dataset)’ function (http:// stat.ethz.ch/R-manual/R-patched/library/base/html/data.matrix.html). Pearson and Spearman rank correlations of mean KR or net topology biases and each of the three membrane composition parameters (PLH, FAU, FAL) were determined in pairwise combinations within the matrix. Pearson moment correlations (r) were calculated by equation r = (1/ n − 1)∑[(x − X)(y − Y) / (SxSy)] where n is the number of pairs of data, X and Y are the sample means of the topology bias (x) and lipid parameter (y) respectively, and Sx and Sy are the sample standard deviations of the x and y values respectively. Spearman's ranked correlation (ρ or rs) calculations were also determined using Pearson correlation coefficients calculated above between ranked KR bias and lipid variables. All matrix correlations were determined using the R statistics package ‘library(Hmisc)’ (http://cran.r-project.org/ web/packages/Hmisc/index.html) and ‘rcorr(x, type=)’, where x refers to the matrix, and type= designates the specific correlation equation either Pearson or Spearman that was used and provided its associated p-value significance based on the adjusted α level value (http://www.statmethods.net/stats/correlations.html). Correlations were also determined using Pearson and Spearman equations using Excel software data analysis functions. The p-values determined after hypothesis testing in this study were adjusted α values determined based on either the Bonferroni and/or Holm's methods to control for Type 1 family wise error rates that may occur during matrix correlation multiple testing (these p-values set α=b0.05 to b0.01). Correlograms shown in Figs. 4–6 (and Supplementary Figs. 5 and 6) of the results section were generated using the R statistics ‘library(corrgram)’ package and the function ‘corrgram(x, order = TRUE, panel = panel.pie)’ where x refers to the lipid-topology matrix analysed, order = TRUE permitted variables to be ordered using principal component analysis of the correlation matrix, and panel = panel.pie displayed correlation values between +1.0 to −1.0 as pie charts (http://www.statmethods.net/ advgraphs/correlograms.html). Correlations with confidence intervals of >90–99% were included in this analysis based on its value determined from the population mean according to the t-distribution. Agglomerative hierarchical clustering analysis of topology-lipid matrices were calculated by first determining the number of clusters present within the matrix using the K-means partitioning method using the R statistics ‘kmeans’ clustering function (http://stat.ethz. ch/R-manual/R-patched/library/stats/html/kmeans.html). The number of clusters determined was used to generate hierarchical groupings based on the Ward-linkage method and a Euclidean distance metric was used to determine similarities/dissimilarities within the matrix using R statistics library(pvclust) package (http://www.is.

Before performing a comparison between SMR topology biases and the host membrane lipid composition, the KR and net topology biases of SMR singletons (SMP and SUG) and pairs (PSMR) from a single phylum were assessed to determine if topology biases were equivalent between phyla and measure the extent of the variance (Fig. 3). If the hypothesis is valid, we expect that SMR protein KR and/or net topology biases will differ in value across bacterial phylum/species. Previous studies have demonstrated that positively charged residues (Lys and Arg) are the major contributor to membrane protein topology [12] and represented as a KR topology bias (Figs. 1 and 3). In this study, an additional net topology bias calculation, measuring the difference between positively charged and negatively charged amino acid residues (Glu and Asp) present within the same loop/terminus, was also determined to confirm that negatively charged residues did not influence or associate preferentially to a lipid composition (Figs. 1 and 3). Similar to previous findings by Bay and Turner, the SUG subclass has the broadest species distribution of all the SMR subclasses [21]. The SMP subclass is a close second to SUG based on species distribution, while the PSMR subclass has the least diverse species distribution. As expected, all SMR singleton (SUG or SMP) KR and net topology biases had median topology biases that were relatively close to neutral (0 ± 3 units). However, the median topology bias value of SMR singletons demonstrated a gradual linear increase by 4 to 5 units when each phylum was sorted from lowest to the highest median value (Fig. 3). The overall difference in median KR and net topology biases confirmed that neutral dual topology values differed between phyla. A comparison between KR and net SMR singleton topology biases (Fig. 1), demonstrated that net bias calculations of topology appear to reflect a more neutral mean topology bias (net 0) than KR bias values (KR − 1), suggesting that negative charges may be a useful measure of dual topology biases. The KR and net topology bias distributions among phyla encoding SMR pairs representing Gram + only (EbrA/EbrB), Gram− only (MdtI/MdtJ), and from both Gram+/− (YkkC/YkkD) bacteria were examined as topology control groups. SMR pairs represented the majority of integral membrane proteins that adopt a ‘single’ KR “termini in” or KR “termini out” topology (Fig. 3A, B, and C). SMR pairs are known to function as antiparallel heterooligomers, where one member of a pair has a KR (+) “in” bias (EbrB, YdgF, and YkkC) compared to the opposite KR (−) “out” topology (EbrA, YdgE, and YkkD). SMR pair topology bias analysis confirmed that the median topology bias (either KR or net) of at least one member of each pair was well above or below neutral (0). Similar to SMR singletons, SMR pairs exhibited a gradual decrease in median KR or net topology biases across phyla indicating that the SMR topology bias also differs in various bacterial phyla. As expected for SMR pairs, the charges of paired topology biases (either + or −) were diametrically opposed to each other, but unexpectedly showed a gradual decrease in the overall median topology bias values of each pair within a bacterial phylum (Fig. 3A, B, and C). SMR pair topology biases also demonstrated

3. Results

264

D.C. Bay, R.J. Turner / Biochimica et Biophysica Acta 1828 (2013) 260–270

A

15

KR

Gram+ compared to Gram − (Fig. 3A). Furthermore, phylogenetic analysis of Gram + YkkC and YkkD proteins compared to Gram − YkkC and YkkD pairs by neighbour joining and maximum likelihood distance methods confirmed that topology biases of each individual member of the pair maintained their overall orientation bias (either KR + or −) confirming that KR “out/in” between Gram + and Gram − protein sequences are not exchanged. This outcome is interesting since it implies that topology biases of SMR pairs do not flip or exchange when moving from Gram + to Gram − backgrounds.

Net

10 5 0 -5 At Ba La α β ε γ δ

Ba At La α β γ δ ε

-10

Phylum

B

KR

10

C

Net

KR

10

5

5

0

0

-5

-5

Net

D

Phylum

KR

At De α β δ ε γ

-10

At De α δ β ε γ

Phylum

At Ba La Cl

At Ba La Cl

-10

Net

5

0

Ba Pl Cf La Cl Ch Cy At α γ β δ Bc De Ac ε

Pl La Cl At Cy Bc β Ch α γ δ Ba De Cf Ac ε

-5

Phylum

E

KR

5

Net

0

Pl Ba De At δ Bc γ α β La Cf Cl ε Cy Ch Ac

De Bc Ba At δ Cf Ch α β γ Cy ε Cl Pl La Ac

-5

Phylum Fig. 3. A summary of SMR protein singleton and pair KR and net (KR−ED) topology bias variation within different bacterial phyla. All panels (A–E) show the KR (left side of y‐axis) and net (right side of y-axis) topology biases as boxplot diagrams of the 25th, median, and 75th percentiles for a particular bacterial phylum (x-axis). The order of phylum provided on the x‐axis was sorted according to the median value of the topology bias. In all panels, bacterial phyla were abbreviated as follows: acinetobacteria (At), bacilli (Ba), lactobacilli (La), clostridia (Cl), chloroflexi (Cf), chlorobia (Ch), deinococci (De), planctomycetia (Pl), bacteroidetes (Bc), acidobateria (Ac), α-proteobacteria (α), β‐proteobacteria (b), γ-proteobacteria (γ), δ-proteobacteria (δ), and ε-proteobacteria (ε). In all panels, red and blue colouring indicates Gram negative and Gram positive phylum respectively. Panels A–C show the KR and net topology biases for bacterial phyla encoding SMR pairs A) YkkC (black) and YkkD (grey), B) EbrA (black) and EbrB (grey), and C) YdgF (black) and YdgE (grey). Panels D–E show boxplots of KR and net topology biases for SMR singletons, SMP (D) and SUG (E).

increasing median KR and net bias values but these values appear to be separated according to Gram + and Gram − bacteria. An example of this is seen with the pair YkkC and YkkD, found within both Gram+ and Gram − phyla. Comparison of YkkCD to other pairs found only in Gram + (EbrAB) or Gram − (MdtIJ) bacteria revealed that the individual pair YkkD (a KR (−) “out” topology) demonstrated a inverse parabolic median KR and net bias relationship when surveying Gram+ to Gram− phyla in opposition to YkkC (a KR (+) “in”). This indicates that the topology bias of SMR pairs inverts in

3.2. SMR pair topology biases correlate inversely to lipid parameters according to Gram+ and Gram− bacteria Since SMR protein KR and net topology biases differed when examining Gram-negative and Gram-positive bacterial phyla, the SMR topology bias dataset was compared to the host bacterial membrane lipid composition. SMR singleton or pair KR and net topology bias datasets were fused together with lipid parameter (PLH, FAL, and FAU) datasets forming correlation matrices to determine if a particular lipid composition is linked to topology bias (Figs. 4 and 5). The outcome of this analysis revealed that significant correlation coefficient values (>+0.65 or b− 0.65) were present for singletons and pairs to minimally one of the three lipid parameters, thus supporting our hypothesis that lipid composition changes may influence KR or net topology bias values. Pairwise correlograms (represented as coloured pie charts) are provided for SMR pair and singleton KR and net topology bias comparisons to PLH (− 2, − 1, 0, +1), FAL and FAU lipid parameters in Figs. 3 and 4 respectively. Interestingly, in almost all cases SMR protein topology bias and lipid parameter correlations became more significant when net topology biases were analyzed suggesting again that negatively charged residues may play a greater role in lipid-topology associations among diverse bacteria. Another important trend identified from this analysis showed that correlations of anionic PLH (− 1 and −2) values were opposite to neutral 0 PLH charges for individual SMR pairs (Fig. 4). Based on previous studies of anionic lipids and membrane protein folding variations [9,36], this type of trend was expected in this analysis and supports our main hypothesis that lipid alteration, in this case polar head group charge, influences the positive inside rule. Individual pairs that were KR (+) “in” versus KR (−) “out” demonstrated a reversal in the type of correlations observed for anionic PLH and neutral PLH or to FAL and FAU (Fig. 4). For example, EbrA KR (−) demonstrated strong anti-correlations to anionic PLH in comparison to positively correlated 0 PLH, while opposite PLH correlations were observed for EbrB KR (+) (Fig. 4). This opposite orientation topology trend was also observed in individual members of YdgEF and for YkkCD only when they were separated according to Gram + or Gram − phyla (Fig. 4). Although correlations between only Gram + or only Gram− pair YkkC and YkkD topology biases and individual PLH values demonstrated some improvement, most correlations were still below significant thresholds with the exception of YkkC Gram − net bias and the YkkD Gram + KR bias values. This specificity for PLH and Gram type may indicate that each member of the YkkCD pair is differentially influenced by PLH in each Gram type. Unfortunately, no experimental data currently exists for the YkkCD pair in both Gram+ and Gram − species to confirm this explanation. Unlike PLH, lipid parameters FAL and FAU, generally appear to have similar significant correlations (where both are either positive or both negative in value) when they are compared to the topology bias of either Gram + or Gram − singletons or individual members of the YkkCD pair. Individual members of SMR pair topology biases showed similar correlations for FAL and FAU where either both parameters have positive r values or both parameters have negative r values. The correlation values of both FAL and FAU parameters were opposite when both were compared to KR (+) in, versus KR (−)

D.C. Bay, R.J. Turner / Biochimica et Biophysica Acta 1828 (2013) 260–270

Lipid composition

Gram-/+

Gram-

Gram+

Mean SMR Topology Bias

A

265

Lipid composition

Mean SMR Topology Bias

EbrA KR bias (-)

SUG KR both

EbrA Net bias (-)

SUG KR Gram+

EbrB KR bias (+)

SUG KR Gram-

EbrB Net bias (+)

SUG Net both

YdgE KR bias (-)

SUG Net Gram+

YdgE Net bias (-)

SUG Net Gram-

YdgF KR bias (+)

SMP KR both

YdgF Net bias (+)

SMP KR Gram+

YkkC KR bias (+)

SMP KR Gram-

YkkC Net bias (+)

SMP Net both

YkkD KR bias (-)

SMP Net Gram+

YkkD Net bias (-)

SMP Net Gram-

YkkC KR (+) Gm+

-1.0 0.0 +1.0 Correlation (r)

YkkC KR (+) GmYkkC Net (+) Gm+

B SMR protein

YkkC Net (+) Gm-

Distribution

YkkD KR (-) Gm+

Singleton/ species

YkkD KR (-) Gm-

Pair/ species

YkkD Net (-) Gm+

Total SMR/ species

YkkD Net (-) Gm-

-1.0 0.0 +1.0 Correlation (r) -1.0 0.0 +1.0 Correlation (r)

Fig. 4. A correlogram of mean SMR pair KR and net topology biases within Gram+ and Gram− bacterial phyla and lipid composition. A correlation matrix between individual SMR pair KR and net topology biases compared to each lipid composition parameter. Pie charts shown were calculated using a pairwise Pearson correlation (r) matrix of mean SMR subclass members KR and net topology biases in each phylum surveyed and its mean lipid composition parameter. The shade of each colour, shown on the bottom legend, indicates the extent of the correlation in pie chart format, as either anti-correlations (counter-clockwise wedge rotation) or positive correlations (clockwise wedge rotation). Charges (+/−) provided in parentheses after each SMR member name indicate the overall membrane orientation, where (+) corresponds to termini “in” (Nin–Cin) and (−) to a termini “out” (Nout–Cout) topology. Correlation pie charts in either panel that possessed significant adjusted p‐values at α b 0.05 are shown in black bold.

out (Fig. 4). This suggests that FAL and FAU may also play a role in influencing SMR topology biases and should be considered as another influential factor on topology in addition to PLH charges.

3.3. Gram+ and Gram− SMR singleton topology biases correlate inversely to PLH As observed for SMR pair topology correlations to various lipid parameters, significant correlations between SMR singletons to any of the three lipid parameters examined were only identified when singleton datasets were separated according to Gram+ and Gram− bacterial phyla

Fig. 5. Correlograms of mean SMR singleton KR or net topology biases within Gram+ and Gram− bacteria compared to different lipid compositions. Legends provided in each panel indicate the extent and value of a correlation. All pie charts represent pairwise Pearson correlation coefficients. A) The relationship of SMR singleton KR and net biases (y-axis) determined for Gram+ and Gram− bacterial phyla to each lipid composition (x-axis). B) A correlation matrix of mean SMR singleton, pair and total SMR sequence distributions (y-axis) per phylum compared to each lipid composition parameter (x-axis). Correlation pie charts in each panel with significant p‐values at α b 0.05 are circled in black.

(Fig. 5A). Unlike SMR pairs, FAL and FAU correlations of SMR singleton topology biases were positively correlated (as topology biases increase FAL/FAU also increase) and are statistically significant at a greater frequency in Gram + bacteria compared to Gram − phyla (Fig. 5A). A possible explanation for this may be due to the extensive variation in both fatty acid length and lipid unsaturation among Gram+ bacteria in comparison to Gram− bacteria in the study (Fig. 2). Correlations between SMR singleton topology biases from Gram + and Gram− hosts to PLH were less frequently observed. When significant correlations were observed these generally favoured an inverse correlation between anionic (− 1 or − 2) and neutral PLH (Fig. 5A). Closer examination of SMR singletons separated by subclass (SMP and SUG) revealed different associations between SMP and SUG topology biases and PLH. SUG KR and net topology biases appear to follow a similar inverse correlation to anionic and neutral PLH in Gram+ and Gram − phyla. These SUG–PLH trends are similar to the relationship observed between SMR pairs and PLH (Fig. 4). The most significant correlations identified for SUG KR biases and PLH

266

D.C. Bay, R.J. Turner / Biochimica et Biophysica Acta 1828 (2013) 260–270

were for 0 PLH and − 2 PLH (Supplementary Figs. 2 and 3). The specific association of 0 PLH and −2 PLH to SUG topology biases but not − 1 PLH may be due in part to the function of SUG. In E. coli SUG over-expression experiments, SUG subclass members confer a limited resistance to only cationic detergents [20]. This substrate selectivity may demand SUG protein associations with lipid head groups of greater anionic net charge like cardiolipin but further analysis is needed to confirm this possibility. In contrast to SUG, SMP net topology biases of Gram+ and Gram− phyla show much lower correlations between all PLH values with the exception of −1 PLH which demonstrated a positive correlation across Gram+ and Gram− bacterial phyla and only for Gram− SMP net bias to 0PLH (Fig. 5A). It is uncertain why the SMP singleton trend is opposite to the trend observed for SUG topology bias and PLH. Besides differences in their substrate profile, another possible explanation for this outcome may be due to frequency of SMP homologue identification from horizontally transferred integrons and plasmids. The SMP subclass is the most frequently transferred SMR sequence compared to any other SMR subclass member [15,21]. Since horizontal gene transfer of multidrug resistance containing integrons is favourable in Gram− bacteria [37,38], we examined if an association between the number of SMR singletons, pairs, and total SMR sequences indentified for a particular species/phylum and each lipid parameter existed. Interestingly, only the amount of SMR singletons/phylum showed significant correlations to PLH and opposite between −1 PLH and 0 PLH (Fig. 5B). This may explain why SMP biases and PLH differences do not completely conform to correlation trends observed for SUG or SMR pairs and PLH values. Additional support for the apparent division between Gram+ and Gram− singleton topology biases and PLH is presented in Supplementary Figs. 4 and 5, which show agglomerative hierarchical clustering analysis of individual bacterial PLH compositions compared to SUG and SMP KR and net topology biases respectively. In both analyses, a division of two major clades is observed, one showing an enrichment of predominately Gram− bacterial species and the other enriched within Gram+ and small clusters of Gram− bacterial species. This additional division supports the separation of singleton topology bias according to Gram+ and Gram− bacteria as suggested by our correlational analyses. 3.4. Charged residues in particular SMR loops are strongly correlated to membrane composition changes and favour regions important for topology bias Since the correlation analysis of whole protein SMR topology biases with lipid parameters was successful, examination of charged amino acid residues (K, R, E, and D) in each terminus (N and C) and all loops (1–3) of each SMR subclass member (individual pairs and singletons) was determined. The goal of this analysis was to test if a particular loop or terminus was subject to greater lipid composition parameter influence than others and how consistent this influence was between pairs and singletons. Two types of SMR charged residue termini and loop datasets were prepared, the first relating the number of K/R residues (KR loop dataset) per species (or mean KR/ phylum) only and the other measured the net charge difference between total KR residues from total E/D residues (net loop dataset) by species and/or mean net difference per phylum. Correlation matrices for both datasets were prepared using the lipid parameter dataset and the results are presented as correlograms in Fig. 6 (KR bias only) and Supplementary Fig. 6 (net bias only). Correlograms of individual SMR pair KR and net loop datasets compared to PLH, FAL and FAU datasets indicate that the most significant correlations are found in loop and termini regions that are involved in maintaining the KR (+) “in” or KR (−) “out” topology of that particular pair member (Fig. 6, Supplementary Fig. 5). Although, this correlation pattern is not strictly limited to termini and loops specific for maintaining a particular in/out orientation in all three

N

EbrA(-)

1

L1

2

L2

3

L3

4

C

EbrB(+)

N

N

L1

L1

L2

L2

L3

L3

C

C

YdgE (+)

YdgF (-)

N

N

L1

L1

L2

L2

L3

L3

C

C

YkkC(-)

YkkD (+)

N

N

L1

L1

L2

L2

L3

L3

C

C

SUG N

SMP N

L1

L1

L2

L2

L3

L3

C

C -1.0 0.0 +1.0 Correlation (r)

Fig. 6. A correlogram of mean SMR positively charged residues found in loops and termini compared to various lipid compositions. The cartoon above the correlogram represents each TMS (cylinders 1–4), loop (lines between cylinders marked L1–L3), and terminus (lines marked N and C) of a SMR protein and provides a reference for abbreviated loops and termini (N or C) listed in the correlogram matrix (y-axis). The shade of each loop and terminus indicates regions influencing KR (+) “termini in” (black) and KR (−) “termini out” (light grey) topology biases. Pie charts represent pairwise Pearson correlations determined for the mean number of positively charged residues (K and R) found in each loop or terminus (N or C) across various bacterial phyla (y-axis) compared to a single lipid composition parameter (x-axis). Correlation pie charts with significant adjusted p‐values at α b 0.05 are outlined in black.

pairs surveyed, it did identify a greater number of significant correlations in loops and termini with anionic PLH and FAL/FAU that were important for determining the overall topology bias orientation of each individual pair. This result indicates that all three lipid parameters, PLH, FAL and FAU, influence the amount of charged residue maintained in loops and termini of SMR pairs as noted for the analysis of whole SMR protein topology bias and lipids. Additionally, the value of the correlation (+/−) typically inverted between 0 PLH and anionic PLH in the loops and termini of pairs found only Gram − phyla compared to pairs found only in Gram+ hosts. Charged residues, determined by either KR or net calculations, within SMR singleton (SUG or SMP) loops and termini across both Gram+ and Gram − species showed relatively poor correlations to lipid parameters except for loop1 (Fig. 6, Supplementary Fig. 7). Analysis of SMR singleton charged residues within loop and termini regions separated according to their host (Gram + or Gram −) greatly enhanced correlation significance for − 1 PLH, 0 PLH, and FAL in loops 1–3 and the N-terminus. A greater proportion of significant

D.C. Bay, R.J. Turner / Biochimica et Biophysica Acta 1828 (2013) 260–270

0 PLH 1 PLH -2 PLH

OUT

Gram + PM

Gram - PM

PLH (%) KR (+) in

PLH (%)

C

KR (+) “in”

KR (+) in

KR (+) in

IN

KR (+) in

N

FAL

OUT

IN

4. Discussion

FA (%)

C

KR (-) out

N

KR (-) out

FA (%)

PLH (%)

PLH (%) KR (-) out

KR (-) out

KR (-) “out”

FA (%)

FA (%)

C

PLH (%)

KR (0) N

C

PLH (%) SMP KR

N

SUG KR

SUG KR

The main goal of this study was to determine whether the positive inside rule of an integral membrane protein family was influenced by the changing lipid compositions of its host membranes using preexisting protein sequences and experimentally determined lipid composition data. Even before lipid compositions were tested, the extent of SMR pair and singleton KR and net topology bias variation within a given phylum and the gradual change between phyla confirmed that SMR topology biases varied depending on phylum or species (Fig. 3). The bioinformatic analysis performed herein appears to validate the overall hypothesis, by demonstrating that a linear correlation exists between SMR protein KR and net topology biases to each of the three major lipid composition parameters (PLH, FAL and FAU) we examined. Unexpectedly, the correlations were most significant only when SMR proteins were separated according to bacterial membrane type (Gram+/Gram−) indicating that bacterial membrane architecture and/or protein sequence origin plays an additional factor in determining protein topology and orientation within the membrane. Enhanced correlation significance was generally observed for net SMR topology values and all lipid parameters examined in this study in comparison to KR topology biases. This outcome was unexpected since the distribution of positively charged amino acids within loops and termini are known to influence topology (as reviewed by [9]). It should be noted that studies of the positive inside rule have noted some exceptions, where negatively charged residues appears to influence topology [5,6,39]. Our results suggest that negatively charged residues contributing to the overall net topology bias are an important measure to consider when comparing to the lipids of organisms representing diverse phyla. In addition, the neutral topology bias of dual-topology integral membrane proteins, as represented by SMR singletons in this study, demonstrated a linear relationship to lipid composition parameters. Similar to the analysis of SMR pairs, this relationship appears to support an inverse correlation between PLH lipids and singleton topology biases in bacteria representing Gram + versus Gram − membranes. However, specific differences also exist between SUG and SMP subclass correlations and PLH, suggesting that SMR singleton dual topology has additional pressures influencing their topology bias (Fig. 7). Overall, these observations support the hypothesis that the neutrality of dual topology proteins fluctuates (Fig. 7) as particular lipid parameters change (Fig. 2). An overall summary of the lipid composition influences on SMR protein topology biases from this study in Gram + versus Gram− bacteria is provided in Fig. 7. The analysis of the SMR protein family, with respect to PLH and topology bias, fits well with the overall findings from in vivo experiments evaluating TMS topology and folding within E. coli membranes lacking lipid biosynthetic genes [7,36,40]. Such work suggested that protein conformation and insertion alters as anionic lipids increase. The topology of SMR protein family members, such as SMR pair E. coli YdgEF [13,14], and Bacillus subtilis EbrAB [23–27], has also been experimentally characterized and confirm that individual SMR pairs have opposite but fixed topologies (KR bias termini in and KR bias termini out) that match the findings of our analysis.

FAU

SMP KR

correlations were observed between singleton loops and termini of Gram− bacteria in comparison to Gram+ phyla. The difference of correlation significance between Gram − and Gram + singleton charged residues in loops and termini regions appears to be in agreement with the trend showing an increase in the number of singletons as bacterial membranes composed of increasing 0 PLH (Fig. 5B). This suggests that the topology bias of SMR singletons changes across phyla but the selective pressures maintaining these charged amino acid residues within loops and termini of SMR proteins in Gram + and Gram− bacteria are inversely related.

267

PLH (%)

PLH (%)

Fig. 7. A summary diagram of SMR protein topology bias and the influence of bacterial lipid composition. The plasma membranes (PM) of Gram+ and Gram− bacteria are provided to indicate the overall lipid composition differences (according to colours provided in the upper left hand legend box) between each membrane type. The orientation or sidedness of each lipid membrane is indicated as “in” (cytoplasmic side) and “out” (periplasmic/external side) and serves as a reference for SMR protein diagrams shown below. SMR proteins (refer to Fig. 1 for cartoon description) are provided in each quadrant of the diagram showing KR (+) “termini in”, KR (−) “termini out”, and KR (0) dual topology correlation trends with each lipid parameter shown in the legend as a linear arrows in a chart. The slopes of each arrow shown in all charts only reflects overall positive or negative (anti-) correlations in the analyses. Dashed lines separate Gram+ and Gram− results from each topology bias determined for pairs and singletons.

The proposed dual topology (also known as antiparallel topology) of SMR singletons is still actively debated [41,42]. Confusion arises from studies of the archetypal SMR member EmrE protein, where experimental studies also support a functional single orientation parallel topology [43–45]. Nonetheless, a growing number of SMR structural studies focusing on the topology and multimerization of EmrE make an increasingly convincing argument for SMR singletons functioning as dual topology homo oligomers [27,46]. Various in vitro

268

D.C. Bay, R.J. Turner / Biochimica et Biophysica Acta 1828 (2013) 260–270

EmrE folding and/or transport experiments strongly indicate that anionic lipids [47–49] or anionic detergents [50,51] alter the folded conformation or transport activity of this protein. The results from this study support these findings but also indicate that singleton KR and net topology biases possess inverse correlations to anionic lipid head group charges which again inversely differ depending on membrane type (Gram+/Gram−) (Fig. 6). Based on our results and given the Gram− E. coli membrane composition (− 2 PLH 3%; − 1 PLH 19%; 0 PLH 74%) [52], we would predict that EmrE protein should demonstrate an altered folding conformation (or potentially a biased single topology insertion) if the protein inserted into an anionic lipid enriched membrane of a Gram+ bacteria, for example Staphylococcus aureus (− 2 PLH 5%; − 1 PLH 57%; 0 PLH 0%; 38% +1 PLH) [53] (Supplementary Figs. 3 and 4). An opposite PLH influence would be expected for the S. aureus EmrE homologue, Smr, if it were to insert in E. coli membranes. To date, there are no experiments examining the topology of SMR singleton homologues expressed in varying host membranes, however the SMR pair EbrAB from B. subtilis has been characterized in E. coli and adopts the expected orientation predicted by the positive inside rule [26,27]. An important aspect identified from this study that should be considered in future work was inherent issues that emerged from limitations in the lipid datasets as well as current bacterial taxonomy. The correlations in this study improved when specific samples were omitted and is highlighted in Supplementary Figs. 2 and 3. However we chose to show all data without sampling omissions, to prevent omission bias and to show the robustness of the analysis prior to sampling. Data availability, abundance, and taxonomic limitations hampered the statistical analysis to generate reliable/relevant standard errors for statistical analysis. There was a relatively low amount of detailed lipid data (~ 80–90 diverse species) available for all sequenced bacteria that encoded SMR members within each phylum (in total ~ 400 unique diverse bacterial species). Since individual KR values are all integers, mean SMR topology values were essential to draw meaningful correlations between lipid values within each phylum. SMR singleton and pair subclass distributions were often unevenly represented within the currently available genomes. This restricted our analysis to the bacterial phylum level (as opposed to class level) to avoid decreasing the available unique protein sequence sample size (n) (refer to Supplementary Table 1 for SMR distribution/coverage). Hence, we relied on phylum level mean topology values for all calculated correlations. In some cases, topology values and lipid compositions of some bacteria were closer to unrelated phyla and shifted mean topology bias values within a given phylum. We used agglomerative hierarchical clustering (Supplementary Figs. 1, 4, and 5) to assess the significance of phylum groupings at the species level and trim SMR topology datasets. After excluding outlying species that outgrouped based on hierarchical clustering analyses, some correlation improvements can be seen (Supplementary Fig. 3). Herein we only show the outcome for the untrimmed complete analysis, since this prevents sampling bias and highlights the natural variation of lipid compositions at the phylum level and would be the outcome for those who would use the analysis. Regardless of these limitations, exciting correlations could be observed within this system and serve as an example that demonstrates the importance of the influence of lipids on integral membrane proteins. Another finding from our results may provide an explanation for the peculiar distribution of SMR protein sequences within different bacterial phylum based on previous bioinformatic analyses [21]. To date, SMR pairs and singletons are unevenly distributed within various bacterial phyla based on SMR sequence homology searches. These SMR searches indicated that Gram − proteobacteria had a greater than average number of singletons per species than in Gram+ acinetobacteria, bacilli, and lactobacilli [21]. The ecological niche and routine exposure to QCC by bacteria that encode SMR sequences were originally suggested to account for the selective

pressures influencing SMR pair versus singleton distributions. This study now provides an additional explanation by suggesting that membrane composition, specifically the amount of anionic phospholipids may be an additional variable accounting for the predominance of SMR singletons in Gram− versus Gram+ species. Since SMR singletons are the likely progenitors of SMR pairs, based on previous phylogenetic [21,54] and SMR site directed mutagenesis topology studies [14,27], it is tempting to speculate that the variation in membrane lipid compositions also creates a bias selecting more SMR singletons in Gram − rather than in Gram+ hosts. SMR singletons are the most frequently identified SMR member on in the 3′ conserved regions of mobile genetic elements known as integrons [55] and these elements have thrived in Gram − bacteria [38,56]. Hence, lateral gene transfer of SMR singleton genes to Gram+ bacterial hosts that possess greater anionic PLH content (in many cases little or no 0 PLH content) may exert greater pressures on singleton gene sequences to adapt by either gaining/losing additional residues or undergoing gene duplication to maintain a fixed antiparallel two protein paired complex. It is important to note that SMR pairs are found in both Gram+ and Gram− bacteria, but the greatest number of SMR pair sequences are identified in bacilli, lactobacilli and ε-proteobacteria [21]. This study also identified SMR topology bias connections to FAL and FAU suggesting that other lipid parameters may also bias membrane protein topology and insertion (Fig. 7). Experimental studies relating lipid thickness to membrane protein topology are limited but studies of a few bacterial membrane proteins such as, Lactobacillus lactis Leu-H or E. coli melibiose-cation cotransporter indicate that the hydrophobic mismatch between the acyl chain and TMS plays a role in the folding, stability, and conformation of a membrane protein (as reviewed by [57]). As TMS exceed the lipid bilayer acyl chain length or thickness, alterations to TMS tilt and/or TMS leaflet exclusion result and demonstrated from in silico modelling experiments of hydrophobic peptide TMS [58]. The influence of the bilayer hydrophobic core (FAL and FAU) has been characterized in much greater detail for the E. coli β-barrel outer membrane protein A (OmpA) protein and studies of this protein indicate that elements such as fatty acid chain length and unsaturation also alter lipid packing in a bilayer [59,60]. Associations between the positive inside rule and bilayer thickness are currently focused on phospholipid head group composition, which is understandable considering the difficulty of reliably regulating, altering, and assaying acyl chain variation in vivo. Hence, this study suggests that the positive inside rule of integral membrane proteins may also be influenced by the thickness of the membrane (FAL) and the relative acyl chain packing (FAU). Examination of charged amino acid residues within individual SMR pair termini and loops demonstrated significant correlations between PLH, FAL and FAU to the amount of residues in loops/termini important for maintaining an overall topology bias (either KR (+) “in” or KR (−) “out”). This outcome was expected based on the overall hypothesis of this study and the positive inside rule. By extension, SMR singletons also demonstrated significant correlations in loops and termini but did not demonstrate the distinct trend observed for individual SMR pairs. This result was also expected for a dual topology protein and in support of the positive inside rule. A final outcome from this examination is the opposing influence of lipid compositions on topology biases between Gram+ and Gram− membrane types. The lipid composition changes dramatically when surveying proteobacteria to acinetobacteria/bacilli (Fig. 2). Thus far, most theory explaining lipid influence on the positive inside rule has been determined from studies of E. coli membranes and varying the amount of PE (0 PLH) and PG (− 1 PLH) content [9,36]. Positively charged residues (K + R) within a protein act as the major determinants of topology since the effects of negatively charged residues (E + D) in insertions are likely neutralized by the presence of PE in Gram− membranes. In Gram+ membranes, 0 PLH lipids are less

D.C. Bay, R.J. Turner / Biochimica et Biophysica Acta 1828 (2013) 260–270

abundant likely explaining why our results, using the net (∑KR − ∑ED) residue charge enhanced the significance of most correlations, particularly in Gram+ net topology bias comparisons. As discussed above, selective pressures originating from lateral gene transfer, those maintaining substrate recognition/function, and pressures from the host environment are likely variables influencing SMR singletons and pairs. Now, a fourth pressure can be attributed to lipid composition based on this study. An alternative explanation for the Gram+/Gram− division lipid division may also be due to the function of individual SMR pairs exchanging between membrane types. Studies of the B. subtilis SMR pair EbrA (KR −) and EbrB (KR +) proteins accumulated in E. coli membranes demonstrate a functional non-equivalence, where only EbrB is functionally dominant, between each member of the pair during transport [25] but the predicted topologies of both were as expected based on the positive inside rule [26,27]. In this event, the dominance of one member, EbrB over EbrA, may also influence SMR pair topology. Overall, our analysis demonstrates the influential role of the lipids towards membrane protein structure and insertion orientation prediction. This type of analysis is also suitable for other integral membrane proteins and may provide insights regarding lipid targeting, folding and evolution in different types of membranes. Further understanding of these relationships may lead to the development of phylum specific antimicrobials and potentially explain why particular bacterial membrane proteins are only found in particular hosts. Acknowledgements We would like to thank Dr. Elmar Prenner for his assistance with lipid dataset resources, Sean Booth for his assistance and advice using R statistical analysis, and Veerle De Wever for manuscript discussions. We also acknowledge the funding for this work by NSERC Discovery and NSERC Accelerator grants awarded to R.J.T. Appendix A. Supplementary data Supplementary data to this article can be found online at http:// dx.doi.org/10.1016/j.bbamem.2012.09.003. References [1] C. Ratledge, S.G. Wilkinson, Microbial Lipids, Academic Press, San Diego, CA, 1988. [2] W. Dowhan, E. Mileykovskaya, M. Bogdanov, Diversity and versatility of lipid-protein interactions revealed by molecular genetic approaches, Biochim. Biophys. Acta 1666 (2004) 19–39. [3] P.A. Janmey, P.K. Kinnunen, Biophysical properties of lipids and dynamic membranes, Trends Cell Biol. 16 (2006) 538–546. [4] W. van Klompenburg, I. Nilsson, G. von Heijne, B. de Kruijff, Anionic phospholipids are determinants of membrane protein topology, EMBO J. 16 (1997) 4261–4266. [5] W. Zhang, M. Bogdanov, J. Pi, A.J. Pittard, W. Dowhan, Reversible topological organization within a polytopic membrane protein is governed by a change in membrane phospholipid composition, J. Biol. Chem. 278 (2003) 50128–50135. [6] W. Zhang, H.A. Campbell, S.C. King, W. Dowhan, Phospholipids as determinants of membrane protein topology. Phosphatidylethanolamine is required for the proper topological organization of the gamma-aminobutyric acid permease (GabP) of Escherichia coli, J. Biol. Chem. 280 (2005) 26032–26038. [7] M. Bogdanov, J. Xie, W. Dowhan, Lipid-protein interactions drive membrane protein topogenesis in accordance with the positive inside rule, J. Biol. Chem. 284 (2009) 9637–9641. [8] M. Raja, The role of phosphatidic acid and cardiolipin in stability of the tetrameric assembly of potassium channel KcsA, J. Membr. Biol. 234 (2010) 235–240. [9] G. von Heijne, Membrane-protein topology, Nat. Rev. Mol. Cell Biol. 7 (2006) 909–918. [10] G. von Heijne, Control of topology and mode of assembly of a polytopic membrane protein by positively charged residues, Nature 341 (1989) 456–458. [11] G. von Heijne, Membrane protein structure prediction. Hydrophobicity analysis and the positive-inside rule, J. Mol. Biol. 225 (1992) 487–494. [12] G. von Heijne, The distribution of positively charged residues in bacterial inner membrane proteins correlates with the trans-membrane topology, EMBO J. 5 (1986) 3021–3027. [13] M. Rapp, E. Granseth, S. Seppala, G. von Heijne, Identification and evolution of dual-topology membrane proteins, Nat. Struct. Mol. Biol. 13 (2006) 112–116.

269

[14] M. Rapp, S. Seppala, E. Granseth, G. von Heijne, Emulating membrane protein evolution by rational design, Science 315 (2007) 1282–1284. [15] D.C. Bay, K.L. Rommens, R.J. Turner, Small multidrug resistance proteins: a multidrug transporter family that continues to grow, Biochim. Biophys. Acta 1778 (2008) 1814–1838. [16] D.L. Jack, N.M. Yang, M.H. Saier Jr., The drug/metabolite transporter superfamily, Eur. J. Biochem. 268 (2001) 3620–3639. [17] M.H. Saier Jr., I.T. Paulsen, Phylogeny of multidrug transporters, Semin. Cell Dev. Biol. 12 (2001) 205–213. [18] Y. Boucher, M. Labbate, J.E. Koenig, H.W. Stokes, Integrons: mobilizable platforms that promote genetic diversity in bacteria, Trends Microbiol. 15 (2007) 301–309. [19] K. Hegstad, S. Langsrud, B.T. Lunestad, A.A. Scheie, M. Sunde, S.P. Yazdankhah, Does the wide use of quaternary ammonium compounds enhance the selection and spread of antimicrobial resistance and thus threaten our health? Microb. Drug Resist. 16 (2010) 91–104. [20] Y.J. Chung, M.H. Saier Jr., Overexpression of the Escherichia coli sugE gene confers resistance to a narrow range of quaternary ammonium compounds, J. Bacteriol. 184 (2002) 2543–2545. [21] D.C. Bay, R.J. Turner, Diversity and evolution of the small multidrug resistance protein family, BMC Evol. Biol. 9 (2009) 140. [22] K. Higashi, H. Ishigure, R. Demizu, T. Uemura, K. Nishino, A. Yamaguchi, K. Kashiwagi, K. Igarashi, Identification of a spermidine excretion protein complex (MdtJI) in Escherichia coli, J. Bacteriol. 190 (2008) 872–878. [23] T. Kikukawa, T. Nara, T. Araiso, S. Miyauchi, N. Kamo, Two-component bacterial multidrug transporter, EbrAB: mutations making each component solely functional, Biochim. Biophys. Acta 1758 (2006) 673–679. [24] Y. Masaoka, Y. Ueno, Y. Morita, T. Kuroda, T. Mizushima, T. Tsuchiya, A two-component multidrug efflux pump, EbrAB, in Bacillus subtilis, J. Bacteriol. 182 (2000) 2307–2310. [25] Z. Zhang, C. Ma, O. Pornillos, X. Xiu, G. Chang, M.H. Saier Jr., Functional characterization of the heterooligomeric EbrAB multidrug efflux transporter of Bacillus subtilis, Biochemistry 46 (2007) 5218–5225. [26] T. Kikukawa, S. Miyauchi, T. Araiso, N. Kamo, T. Nara, Anti-parallel membrane topology of two components of EbrAB, a multidrug transporter, Biochem. Biophys. Res. Commun. 358 (2007) 1071–1075. [27] I. Nasie, S. Steiner-Mordoch, A. Gold, S. Schuldiner, Topologically random insertion of EmrE supports a pathway for evolution of inverted repeats in ion-coupled transporters, J. Biol. Chem. 285 (2010) 15234–15244. [28] M. Clamp, J. Cuff, S.M. Searle, G.J. Barton, The Jalview Java alignment editor, Bioinformatics 20 (2004) 426–427. [29] A.M. Waterhouse, J.B. Procter, D.M. Martin, M. Clamp, G.J. Barton, Jalview version 2—a multiple sequence alignment editor and analysis workbench, Bioinformatics 25 (2009) 1189–1191. [30] J. Felsenstein, PHYLIP — phylogeny inference package (Version 3.2), Cladistics 5 (1989) 164–166. [31] A. Krogh, B. Larsson, G. von Heijne, E.L. Sonnhammer, Predicting transmembrane protein topology with a hidden Markov model: application to complete genomes, J. Mol. Biol. 305 (2001) 567–580. [32] E.L. Sonnhammer, G. von Heijne, A. Krogh, A hidden Markov model for predicting transmembrane helices in protein sequences, Proc. Int. Conf. Intell. Syst. Mol. Biol. 6 (1998) 175–182. [33] T. Hirokawa, S. Boon-Chieng, S. Mitaku, SOSUI: classification and secondary structure prediction system for membrane proteins, Bioinformatics 14 (1998) 378–379. [34] K.C. Lee, R.I. Webb, P.H. Janssen, P. Sangwan, T. Romeo, J.T. Staley, J.A. Fuerst, Phylum Verrucomicrobia representatives share a compartmentalized cell plan with members of bacterial phylum Planctomycetes, BMC Microbiol. 9 (2009) 5. [35] H. Goldfine, Bacterial membranes and lipid packing theory, J. Lipid Res. 25 (1984) 1501–1507. [36] W. Dowhan, M. Bogdanov, Lipid-dependent membrane protein topogenesis, Annu. Rev. Biochem. 78 (2009) 515–540. [37] D.A. Rowe-Magnus, A.M. Guerout, P. Ploncard, B. Dychinco, J. Davies, D. Mazel, The evolutionary history of chromosomal super-integrons provides an ancestry for multiresistant integrons, Proc. Natl. Acad. Sci. U. S. A. 98 (2001) 652–657. [38] D.A. Rowe-Magnus, D. Mazel, Integrons: natural tools for bacterial genome evolution, Curr. Opin. Microbiol. 4 (2001) 565–569. [39] Y. Gavel, G. von Heijne, The distribution of charged amino acids in mitochondrial inner-membrane proteins suggests different modes of membrane integration for nuclearly and mitochondrially encoded proteins, Eur. J. Biochem. 205 (1992) 1207–1215. [40] M. Bogdanov, E. Mileykovskaya, W. Dowhan, Lipids in the assembly of membrane proteins and organization of protein supercomplexes: implications for lipid-linked disorders, Subcell. Biochem. 49 (2008) 197–239. [41] S. Schuldiner, Controversy over EmrE structure, Science 317 (2007) 748–751 (author reply 748-751). [42] S. Schuldiner, Parallel or antiparallel, who cares? Science 328 (2010) 1. [43] H.S. McHaourab, S. Mishra, H.A. Koteiche, S.H. Amadi, Role of sequence bias in the topology of the multidrug transporter EmrE, Biochemistry 47 (2008) 7980–7982. [44] M. Soskine, S. Mark, N. Tayer, R. Mizrachi, S. Schuldiner, On parallel and antiparallel topology of a homodimeric multidrug transporter, J. Biol. Chem. 281 (2006) 36205–36212. [45] S. Steiner-Mordoch, M. Soskine, D. Solomon, D. Rotem, A. Gold, M. Yechieli, Y. Adam, S. Schuldiner, Parallel topology of genetically fused EmrE homodimers, EMBO J. 27 (2008) 17–26. [46] T. Nara, T. Kouyama, Y. Kurata, T. Kikukawa, S. Miyauchi, N. Kamo, Anti-parallel membrane topology of a homo-dimeric multidrug transporter, EmrE, J. Biochem. (Tokyo) 142 (2007) 621–625.

270

D.C. Bay, R.J. Turner / Biochimica et Biophysica Acta 1828 (2013) 260–270

[47] K. Charalambous, D. Miller, P. Curnow, P.J. Booth, Lipid bilayer composition influences small multidrug transporters, BMC Biochem. 9 (2008) 31. [48] D. Miller, P.J. Booth, The use of isothermal titration calorimetry to study multidrug transport proteins in liposomes, Methods Mol. Biol. 606 (2009) 233–245. [49] D. Miller, K. Charalambous, D. Rotem, S. Schuldiner, P. Curnow, P.J. Booth, In vitro unfolding and refolding of the small multidrug transporter EmrE, J. Mol. Biol. 393 (2009) 815–832. [50] D.C. Bay, R.A. Budiman, M.P. Nieh, R.J. Turner, Multimeric forms of the small multidrug resistance protein EmrE in anionic detergent, Biochim. Biophys. Acta 1798 (2010) 526–535. [51] D.C. Bay, R.J. Turner, Spectroscopic analysis of small multidrug resistance protein EmrE in the presence of various quaternary cation compounds, BBA — Biomembranes (2012) 1–14. [52] S.G. Wilkinson, Gram-negative bacteria, in: C. Ratledge, S.G. Wilkinson (Eds.), Microbial Lipids, vol. 1, Academic Press, San Diego, CA, 1988, pp. 299–488. [53] W.M. O'Leary, S.G. Wilkinson, Gram-positive bacteria, in: C. Ratledge, S.G. Wilkinson (Eds.), Microbial Lipids, vol. 1, Academic Press, San Diego, CA, 1988, pp. 117–201.

[54] Y.J. Chung, M.H. Saier Jr., SMR-type multidrug resistance pumps, Curr. Opin. Drug Discov. Devel. 4 (2001) 237–245. [55] I.T. Paulsen, T.G. Littlejohn, P. Radstrom, L. Sundstrom, O. Skold, G. Swedberg, R.A. Skurray, The 3′ conserved segment of integrons contains a gene associated with multidrug resistance to antiseptics and disinfectants, Antimicrob. Agents Chemother. 37 (1993) 761–768. [56] W.H. Gaze, N. Abdouslam, P.M. Hawkey, E.M. Wellington, Incidence of class 1 integrons in a quaternary ammonium compound-polluted environment, Antimicrob. Agents Chemother. 49 (2005) 1802–1807. [57] O.S. Andersen, R.E. Koeppe 2nd, Bilayer thickness and membrane protein function: an energetic perspective, Annu. Rev. Biophys. Biomol. Struct. 36 (2007) 107–130. [58] M. Venturoli, B. Smit, M.M. Sperotto, Simulation studies of protein-induced bilayer deformations, and lipid-induced protein tilting, on a mesoscopic model for lipid bilayers with embedded proteins, Biophys. J. 88 (2005) 1778–1798. [59] H. Hong, L.K. Tamm, Elastic coupling of integral membrane protein stability to lipid bilayer forces, Proc. Natl. Acad. Sci. U. S. A. 101 (2004) 4065–4070. [60] J.H. Kleinschmidt, M.C. Wiener, L.K. Tamm, Outer membrane protein A of E. coli folds into detergent micelles, but not in the presence of monomeric detergent, Protein Sci. 8 (1999) 2065–2071.