The β-barrel finder (BBF) program, allowing identification of outer membrane β-barrel proteins encoded within prokaryotic genomes

Abstract

Many outer membrane proteins (OMPs) in Gram-negative bacteria possess known β-barrel three-dimensional (3D) structures. These proteins, including channel-forming transmembrane porins, are diverse in sequence but exhibit common structural features. We here report computational analyses of six outer membrane proteins of known 3D structures with respect to (1) secondary structure, (2) hydropathy, and (3) amphipathicity. Using these characteristics, as well as the presence of an N-terminal targeting sequence, a program was developed allowing prediction of integral membrane β-barrel proteins encoded within any completely sequenced prokaryotic genome. This program, termed the β-barrel finder (BBF) program, was used to analyze the proteins encoded within the Escherichia coli genome. Out of 4290 sequences examined, 118 (2.8%) were retrieved. Of these, almost all known outer membrane proteins with established β-barrel structures as well as many probable outer membrane proteins were identified. This program should be useful for predicting the occurrence of outer membrane proteins in bacteria with completely sequenced genomes.

Gram-negative bacteria are surrounded by two concentric lipid bilayer membranes. Both membranes contain proteins that facilitate the transport of nutrients, end products of metabolism, toxic substances, ionic species, macromolecules, and other molecular species of biological importance (aSaier 2000a). Although integral membrane transporters of the inner membrane are generally of α-structures, traversing the membrane as α-helices, those of the outer membranes consist largely of β-structures, forming β-barrels (Koebnik et al. 2000; Saier 2000a; Schulz 2000). Structural features may provide targeting signals for these two membranes (Hancock 1991; Buchanan 1999).

Among the outer membrane proteins (OMPs) of Gram-negative bacteria are the oligomeric, often trimeric channel-forming porins, several of which have been structurally characterized by X-ray crystallography (Hancock et al. 1990; Jeanteur et al. 1991; Meyer et al. 1997). These proteins can transport small molecules nonselectively, or they can be highly selective for a single class of molecules (Nieweg and Bremer 1997; Wang et al. 1997; Buchanan 1999). Similar proteins are found in outer membranes of mitochondria and plant plastids (Blachly-Dyson et al. 1990; Fischer et al. 1994; Bathori et al. 2000). They may also be present in the outer mycolic acid-containing membranes of acid-fast Gram-positive bacteria such as species of mycobacteria, corynebacteria, and Nocardia (Riess et al. 1998; Senaratne et al. 1998; Kartmann et al. 1999).

Because of their unique structures and subcellular locations, outer membrane β-barrel porins are classified in their own category in our transporter classification (TC) system (Category 1.B) separately from the α-type cytoplasmic membrane channel proteins (Category 1.A) and from the pore-forming toxins, which are synthesized in cells other than the ones in which they exert their toxic effects (Category 1.C) (Saier 1999a, 1999b, 1999c, 2000a, 2000b; Saier and Tseng 1999). There are currently 35 families classified as integral β-barrel porins in our TC system under Category 1.B (aSaier 2000a; see our transporter classification database, TCDB). Twenty-nine of these protein families are derived from Gram-negative bacteria with one from mitochondria, three from chloroplasts, and two from acid fast Gram-positive bacteria. No significant sequence similarity between members of different families can be detected (see Saier 1994, for consideration of significance levels), and in at least some of these families, known structural differences suggest independent evolutionary origins. Because many families of β-barrel porins are not yet recognized, and the sequences of these kind of proteins are very diverse, recognition and characterization of new members of β-barrel porin families is a challenging and interesting task (Achouak et al. 2001).

In this article, we report an analysis of the sequences and available structural information of all β-barrel porins for which 3D structures had been determined when we initiated these studies. Using this information, we developed an algorithm that can be used to judge if a protein has a high probability of having an integral membrane β-barrel structure. This method can be used to screen whole genomes for proteins of predominantly β-structure and to select candidates that may be outer membrane β-barrel porin proteins. Such studies, applied to Escherichia coli, the organism with the most functionally characterized OMPs, are reported here. The method reported complements an early method described by Neuwald et al. (1995) and applied to mitochondrial proteins by Mannella et al. (1996), as well as a neural network-based approach reported by Jacoboni et al. (2001) and a distinct approach based on amino acid composition and protein architecture reported by Wimley (2002). The latter two reports were published after completion of the work reported here. The BBF program is freely available to academic users upon request to the corresponding author (M.H.S.).

Results

Transmembrane signal sequence analysis

Systematic hydropathy analyses of all protein sequences encoded within completely sequenced genomes enabled us to rule out most of the cytoplasmic and integral inner membrane proteins because most outer membrane protein precursors have N-terminal hydrophobic transmembrane signal sequences. Because of the inaccuracies of the hydropathy analyses, there is a possibility that more than one transmembrane hydrophobic peak will be identified for a given sequence. We have used this program to analyze all proteins classified as β-barrel proteins in the TC system (aSaier 2000a) as reported on our Web site. Of the 80 outer membrane proteins with (putative) β-barrel structures thus identified, 49 had only one predicted TMS, 22 had two predicted TMSs, and 9 proteins had three or four predicted TMSs. None of these proteins displayed more than four putative TMSs. Eighty-three percent of the proteins analyzed had transmembrane segments near their N-termini (first 50 residues). Applying the BBF program to 4290 sequences encoded within the E. coli genome, the MEMSAT program predicted 1730 sequences to have at least one transmembrane segment within their N-terminal 50 residues. Among them, 1073 sequences had only this one TMS, 266 sequences had two putative TMSs, and 191 sequences had three or four predicted TMSs. All of these 1730 sequences were candidates for further analysis.

Outer membrane protein screen program

All identified candidate protein sequences were analyzed for potential β-barrel structures by combining three programs. A hydrophobic peak should coincide with an amphipathic peak in a region predicted by Jnet to be a β-strand. Of the 1730 sequences screened, our program retrieved 70 sequences with one predicted TMS, 35 sequences with two predicted TMSs, 9 sequences with three predicted TMSs, and 4 sequences with four predicted TMSs. The total number of sequences is 118. Thus, 2.8% of the sequences in E. coli were selected by our method as outer membrane protein candidates.

Family identification using BLAST

Using the protein sequences obtained, we performed a systematic BLAST search against the databases for sequences exhibiting clear similarity. We then identified known or putative outer membrane proteins based on the database annotations. The results are presented in Table 1. As the results show, 47 sequences (40%) are either known outer membrane proteins with β-barrel structures, or are putative outer membrane proteins that exhibit sequence similarity with known outer membrane proteins. Other proteins retrieved by the program include 12 extracytoplasmic fimbrial chaperone proteins and 7 extracytoplasmic lipoproteins, both probably largely of β-structure (Choudhury et al. 1999) (Table 2). Fifty-two additional proteins, most of unknown structure and function, are also tabulated. Several of these are known to be extracytoplasmic.

Table 1.

(Putative) E. coli OMPs retrieved by the BBF program

Family
TC #	Abbreviation	Known OMPs and proteins with high similarity to known OMPs	Database description
—	—	PldA	Phospholipase A1 precursor (EC 3.I.1.32)
1.B.1	GBP	OmpF	Outer membrane protein F precursor
1.B.1	GMP	PhoE	Outer membrane pore protein E (phosphoporin)
1.B.1	GMP	NmpC	Outer membrane porin protein NmpC precursor
1.B.1	GMP	OmpC	Outer membrane protein C precursor
1.B.1	GMP	b1377	Outer membrane protein C precursor homolog
1.B.3	LamB	LamB	Maltoporin precursor (lambda receptor protein)
1.B.6	OOP	OmpA	Outer membrane protein A precursor
1.B.6	OOP	YiaD	Hypothetical lipoprotein
1.B.9	FadL	FadL	Long-chain fatty acid transport protein precursor
1.B.10	Tsx	Tsx	Nucleoside-specific channel-forming protein precursor
1.B.11	FUP	b0532	Outer membrane usher protein SfmD precursor
1.B.11	FUP	b0718	Hypothetical outer membrane usher protein (YbgQ)
1.B.11	FUP	b0940	Hypothetical outer membrane usher protein (YcbS)
1.B.11	FUP	FimD	FimD protein
1.B.11	FUP	YehB	Hypothetical outer membrane usher protein
1.B.11	FUP	b3046	Hypothetical outer membrane usher protein (YqiG)
1.B.12	AT	b1169	Hypothetical protein
1.B.12	AT	b1202	Hypothetical protein
1.B.12	AT	b1170	(AidA) Adhesin AidA-I precursor
1.B.12	AT	YdbA	Hypothetical protein
1.B.14	OMR	FhuA	Ferrichrome-iron receptor precursor
1.B.14	OMR	FepA	Ferrienterochelin receptor precursor
1.B.14	OMR	FhuE	Ferric-coprogen receptor protein precursor
1.B.14	OMR	b1451	Probable TonB-dependent receptor (YncD)
1.B.14	OMR	CirA	Colicin I receptor precursor
1.B.14	OMR	b0805	Probable TonB-dependent receptor (YbiL)
1.B.25	Opr	b0681	Hypothetical protein (YbfM)
Proteins with low similarity to known OMPs
1.B.6	OOP	OmpW
1.B.6	OOP	YciD
1.B.9	FadL	b1834
1.B.12	AT	YehA
1.B.12	AT	b1381
1.B.12	AT	b2974
1.B.14	OMR	b1722
1.B.14	OMR	Imp
1.B.17	OMF	b2332
1.B.21	OmpG	YjhA
1.B.25	Opr	b0371
1.B.25	Opr	YbhC
1.B.26	CDP	b3875
—	—	b0964	Homolog of OmpH from Photobacterium SS9
—	—	b1428	Homolog of probable outer membrane protein PMP6 in Chlamydophila pneumoniae
—	—	b1782 (YeaF)	Homolog of porin-like membrane protein Omp26La of Listonella anguillarum
—	—	YfeN	Homolog of outer membrane protein OmpK in Vibrio parahaemolyticus
—	—	YiaT	Homolog of porin-like membrane protein Omp26La of Listonella anguillarum
—	—	YtfM	Homolog of putative outer membrane proteins in several Gram-negative bacteria

Table 2.

E. coli proteins retrieved by the BBF program that lack sequence similarity to known OMPs

(Probable) extracytoplasmic fimbrial chaperones
b0531	Fimbrial chaperone precursor (SmfC)	b0717	Hypothetical fimbrial chaperone (YbgP)
b0939	Hypothetical fimbrial chaperone (YcbR)	YehC	Hypothetical fimbrial chaperone
YcbF	Hypothetical fimbrial chaperone	YraI	Hypothetical fimbrial chaperone
b2336	Hypothetical fimbrial chaperone (YfcS)	YhcA	Hypothetical fimbrial chaperone
FimC	Type 1 fimbrial chaperone	b3047	Hypothetical fimbrial chaperone (YqiH)
b0716	Hypothetical protein (YbgO)	EcpD	Fimbriae biogenesis protein homologue
(Probable) lipoproteins
b0986	YmcC, hypothetical lipoprotein	NlpA	Lipoprotein-28 precursor
YhiU	Hypothetical protein	YjbF	Hypothetical lipoprotein
LolB/HemM	Outer membrane lipoprotein	YjbH	Hypothetical protein
YaeF	Hypothetical lipoprotein	b2512	Homologue of probably lipoprotein in Ralstonia solanacearum
Others
YacK	Probable blue-copper protein	GalM	Aldose 1-epimerase (EC 5.1.3.3) (mutarotase)
b0574	Hypothetical protein (YlcD)	b0837	Hypothetical protein (YliI)
b0819	Hypothetical protein	b1424	Hypothetical protein
b0947	Hypothetical protein	b1487	Periplasmic dipeptide transport protein (dipeptide) binding protein (Dbp) (DppA)
FlgA	Flagellar basal body p-ring formation protein precursor	b1588	Dimethylsulfoxide reductase chain A (DmsA)
b1113	Hypothetical protein	b2043	Colananic acid biosynthesis glycosyl transferse WcaA
b1440	Hypothetical protein	NapG	Ferredoxin-type protein NapG
b1452	Hypothetical protein	MepA	Penicillin-insensitive murein endopeptidase precursor
b1598	Hypothetical protein	YffE	Hydrogenase-4 component A (HyfA)
b1678	Hypothetical protein	YraK	Hypothetical protein
b1780	Unknown protein from 2D-page (YzzQ)	b3219	Hypothetical protein (YhcF)
ErfK	Protein ErfK/SrfK precursor	YihV	Hypothetical sugar kinase
AtoA	Acetyl CoA-transferase β subunit	b0941	Hypothetical protein (YcbT)
SufI	SufI protein precursor	b1834	Hypothetical protein
AcrE	Acriflavin resistance protein E precursor (EnvC) (Membrane Fusion Protein TC #8.A.1)	b3524	Hypothetical ABC transporter ATP-binding protein (YhiG)
YiaS	Putative hexulose-6-phosphate isomerase	YihR	Hypothetical protein
YjbP	Acid phosphatase (AphA/NapA)	YjhT	Hypothetical protein
CpdB	2`,3`-cyclic-nucleotide 2`-phosphodiesterase precursor	FimH	FimH protein
YagV	Hypothetical protein	b2225	Hypothetical protein
UshA	UDP-sugar hydrolase precursor	b2466	Hypothetical protein
YeiP	Hypothetical protein	YicH	Hypothetical protein
EndA	Endonuclease	YijF	Hypothetical protein
NrfC	NrfC protein	YjaH	Hypothetical protein
YadE	Hypothetical protein (same as YadJ)	YagW	Hypothetical protein
YadC	Hypothetical fimbrial-like protein	b1585	Hypothetical protein (YnfC)
TolB	TolB protein

Crossreferencing the TC system

The final step was to screen the E. coli database for members of recognized outer membrane barrel protein families tabulated in the TC system (aSaier 2000a) using the BLAST search tool. Thirteen out of the 28 Gram-negative bacterial families were found to be represented in E. coli. Among them, 10 were recognized by the BBF program. The results are presented in Table 3. Column A tabulates the Gram-negative bacterial porin families that are listed in TC category 1.B; column B reveals families in E. coli that were obtained by our program (marked by ×); column C lists the families that are represented in E. coli but were not retrieved by our program (marked with a star) and column D tabulates those families that have no recognized or predicted homologues in E. coli (marked with an open circle).

Table 3.

Retrieval of E. coli proteins included within 28 Gram-negative bacterial outer membrane protein (OMP) families (TC Category 1.B)

Family	A	B	C	D	Family	A	B	C	D	Family	A	B	C	D	Family	A	B	C	D
GBP	1	×			FadL	9	×			SAP	16			○	CBP	23			○
CP	2			○	Tsx	10	×			OMF	17		★		OPr	25	×
SP	3	×			FUP	11	×			OMA	18		★		CDP	26			○
BRP	4			○	AT	12	×			OprB	19			○	HOP	27			○
POP	5			○	AEP	13			○	TPS	20			○	MomP	31			○
OOP	6	×			OMR	14	×			OmpG	21		★		FomP	32			○
RPP	7			○	RafY	15			○	Secretin	22		★		VC/NP	33			○

A: OMP family TC number in category 1.B (from our Web site: www-biology.ucsd.edu/~msaier/transport/).

B: OMP families represented in E. coli based on the results obtained with the BBF program (×).

C: OMP families represented in E. coli but not recognized by the BBF program (★).

D: OMP families that have no predicted or known homologs in E. coli (○).

Known OMPs not retrieved by the BBF program

Family 17, the Outer Membrane Factor (OMF) family (TC #1.B.17) is one of the families not identified by our program. The 3D structure of one member of this family, TolC of E. coli, has been solved (Koronakis et al. 2000). TolC has a distinctive and previously unknown fold with three TolC protomers assembled to form a continuous channel tunnel over 140 Å long that spans both the outer membrane and the periplasmic space. Export of a transported substrate is brought about by the reversible interaction of the substrate-specific inner membrane protein with an outer membrane protein, thus bypassing the intervening periplasm. Only a small part of each chain has an antiparallel β-sheet structure, and each chain contributes only one-third of the β-barrel in the outer membrane. A large part of the protein has an α-helical structure. Because the β- and α-structures are interspersed, the BBF program did not identify it. Another two families, the Outer Membrane Auxillary (OMA family; TC #1.B.18) (Paulsen et al. 1997; Arrecubieta et al. 2001) and the Outer Membrane Secretin (Secretin) family (TC #1.B.22) (Nguyen et al. 2000) were also not retrieved, possibly for similar reasons. Predictions of the Jnet program suggest that these proteins have relatively large percentages of α-structure, higher than the preselected threshold values used in the BBF program. The OmpG porin (OmpG) family (TC #1.B.21) is the fourth family that was not identified by our program. The single functionally characterized protein in E. coli was suggested to have a structure with a 16 β-stranded barrel lacking the large external loop, L3, that constricts the pores in other porins (Conlan et al. 2000). This protein was not retrieved by our program because its hydrophobic signal sequence begins beyond the first 50 N-terminal residues, at position 55.

Uncharacterized proteins retrieved by the BBF program

The BBF program identified many proteins not known to be β-barrel outer membrane proteins. The functions of most of these proteins are unknown, and they are not classified in the TC system. BLAST searches with these proteins gave either just one hit (to themselves) or multiple hits. The BBF program suggests that these proteins contain large proportions of β-structure, and most of the predicted β-sheet regions contain regions with corresponding hydrophobic and amphipathic peaks. For example, YfeN, a hypothetical 29.2 KD protein, like many other proteins retrieved, displays no functionally characterized homologs when screened with Ψ-BLAST (Altschul et al. 1997). Nevertheless, the secondary structure predictions suggest a large percentage of β-structure, and most of the β-structural regions reveal amphipathic peaks corresponding to hydrophobic peaks. Although the function is unknown, the annotation in Genport notes that it exhibits similarity to an outer membrane protein in V. parahaemolyticus, OmpK. This protein may function as a receptor for the broad host-range vibriophage KVP40 (Inoue et al. 1995). All of the information available suggests that this protein is an outer membrane β-barrel protein. Similarly, a TC-BLAST search (Zhai et al. 2002) revealed that YfeN shows greatest similarity (24% identity; 36% similarity in 150 residue positions) to OmpK of Vibrio parahaemolyticus (Table 1, bottom).

YeaF (Itoh et al. 1996) and YiaT (Sofia et al. 1994) represent two additional proteins retrieved. They have few homologs as indicated using the Ψ-BLAST program. These two proteins are homologs of each other and the outer membrane protein OmpV in V. cholerae (Pohlner et al. 1986). Ψ-BLAST also reveals that these proteins are homologous to a porin-like outer membrane protein Omp26La in L. anguillarum (Table 2; Suzuki et al. 1998). The functions of YiaT and YeaF are currently unknown, but the analyses of our program show that they may be of β-barrel structure. TC-BLAST also revealed that these proteins exhibit short stretches with significant similarity to proteins of the outer membrane receptor (OMR) family (TC #1.B.14).

The outer membrane protein W precursor, OmpW (Stoltzfus et al. 1988), has many homologs. A TC-BLAST search revealed that it exhibits 25% identity and 42% similarity with PorF of Pseudomonas aeruginosa, a member of the OOP family (TC #1.B.6) (Table 1, bottom).

YtfM (Burland et al. 1995) is a hypothetical 64.8 kD protein. Ψ-BLAST results showed that it has similarity with other potential outer membrane proteins from various Gram-negative bacteria (Parkhill et al. 2000a, 2000b). By sequence analysis, it appears that these proteins all have β-barrel structures.

Discussion

In this article we summarize analyses of six families of outer membrane proteins with known β-barrel structures, thereby deriving parameters that delineate individual transmembrane β-strands. We show that most of these β-strands are characterized by peaks of both hydrophobicity and amphipathicity when the angle is set at 180°. This results because of the occurrence of an increased proportion of hydrophobic residues and of alternating hydrophilic and hydrophobic residues in the transmembrane region, respectively. Based on these characteristics of outer membrane barrel proteins, we designed a program that combined three preexisting programs. After screening for an N-terminal signal sequence, we first predict the secondary structure of a protein sequence, and then we calculate the hydropathy and amphipathicity values for the regions predicted to form β-strands. Proteins that meet the assigned criteria are selected as candidates for outer membrane proteins with β-barrel structures. All such proteins are then examined for homologs, one or more of which may have been functionally or structurally characterized, or which may have been localized to the outer membrane of the Gram-negative bacterial envelope.

This method has been designed to automatically screen whole genome protein sequence databases. In this report, we use the method to systematically screen the protein sequences encoded within the E. coli genome. Members of nine of the 13 outer membrane protein families known to be represented in E. coli that had been classified in the TC system were retrieved. The program still has limitations resulting from predictive inaccuracies and unusual positions of transmembrane signal segments as well as inadequate secondary structure predictions. Improvement of these methods should increase the accuracy of the predictions.

We use a sliding window size of seven residues to calculate hydropathy and amphipathicity values, and these values are plotted as a function of protein length. This relatively small window size creates a substantial amount of noise. This problem can be minimized by calculating average hydropathy and amphipathicity values for several aligned sequences (Zhai and Saier 2001b). Using this method together with other bioinformatic tools, we may be able to retrieve a greater proportion of the proteins that are likely to have β-barrel structures. These studies provide a guide to further analyses into the structure/function relationships of outer membrane proteins.

Materials and methods

Sequence analyses of β-barrel proteins with known 3D structures

Three-dimensional structures of several outer-membrane β-barrel proteins, belonging to seven families, had been solved using the technique of X-ray crystallography when we initiated these studies. Five families include proteins that are known to form channels across the outer membranes of Gram-negative bacteria. According to our TC system (aSaier 2000a), these families are as follows: (1) the general bacterial porin (GBP) family (TC# 1.B.1) including OmpF (Cowan et al. 1995) and phoE (Peuptit et al. 1991) of E. coli; (2) the Rhodobacter PorCa Porin (RPP) family (TC #1.B.7) (Kreusch et al. 1994; Kreusch and Schulz 1994; Schulz 2000); (3) the OmpA-OmpF Porin (OOP) family (TC# 1.B.6 ) including OmpA of E. coli (Movva et al. 1980; Pautsch and Schulz 2000); (4) the Sugar Porin (SP) family (TC# 1.B.3) including maltoporin (LamB) of E. coli (Schirmer et al. 1995) and the sucrose porin, ScrY of Salmonella typhimurium (Forst et al. 1998); and (5) the Outer Membrane Receptor (OMR) family (TC# 1.B.14) including FhuA of E. coli (Locher et al. 1998; Koronakis et al. 2000). Two other families that include members that have not been shown to be channels but for which the 3D structures have been solved and shown to be β-barrels include phospholipase A (Snijder et al. 1999) and the OmpX protein (Vogt and Schulz 1999), both of E. coli. We analyzed the sequences and structures of these proteins, and from the results derived a program for identifying candidates with an increased probability of being outer membrane proteins with β-barrel structures.

The GBP family (TC# 1.B.1) and the RPP family (TC #1.B.7)

The general PorCa porin from Rhodobacter capsulatus was the first outer membrane porin for which a 3D X-ray structure was solved (Weiss et al. 1991). Soon thereafter, two other porins, OmpF and PhoE from E. coli, were obtained (Cowan et al. 1992). These porins form homotrimeric structures in the outer bacterial membrane with 16 β-strands spanning the membrane in each monomer. Figure 1 ▶ shows the hydropathy and amphipathicity plots for the mature OmpF protein of E. coli (lacking its hydrophobic leader) using a sliding window of seven residues and an angle of 180° as is appropriate for assessing amphipathicity for β-strands (Kyte and Doolittle 1982; Le et al. 1999). The solid curve is the hydropathy plot, while the dotted curve is the amphipathicity plot (Zhai and Saier 2001a). The lines at the bottom of the figure reveal the positions of established transmembrane β-strands. As can be seen from the figure, the hydropathy values of these strands are generally less than those for a transmembrane α-helical segment (TMS) in a strongly hydrophobic cytoplasmic integral membrane protein, as expected. Outer membrane proteins would probably not cross the cytoplasmic membrane if they exhibited regions other than the cleavable signal segments capable of inserting permanently in the cytoplasmic membrane.

Fig. 1.

Hydropathy and amphipathicity plots for OmpF of the GBP family. The PDB code for OmpF is 2OMF. Solid lines indicate hydropathy; dotted lines indicate amphipathicity; dashed lines indicate predicted β-structures. The format of presentation used here is also used in Figures 2–6 ▶ ▶ ▶ ▶ ▶.

Almost every transmembrane (TM) β-strand corresponds in position to a peak of hydrophobicity and one of amphipathicity. This fact is in agreement with the property that OmpF and other proteins of this family form aqueous transmembrane channels, allowing the diffusion of small hydrophilic molecules across the membrane. A peak of amphipathicity results from side chains in opposite orientation alternately exhibiting hydrophilic and hydrophobic character (Eisenberg et al. 1982).

In the OmpF protein, the first hydrophobic/amphipathic peak corresponds to the first transmembrane β-strand. The high peak of amphipathicity just preceding the peak of hydrophobicity is due to the side chain orientations of hydrophilic residues K10, D12, and K16 versus hydrophobic residues V11, L13, and G15. The second transmembrane β-strand also corresponds to hydrophobic and amphipathicity peaks, generated by hydrophilic residues R42, K46, E48, and Q50 and hydrophobic residues A41, L43, F45, and G47. The third hydrophobic/amphipathic peak occurs at the edge of the third TM β-strand. The amphipathicity is caused by hydrophilic residues Q60, E62, N64, and Q66 and hydrophobic residues G59, W61, Y63, and F65. All remaining pairs of β-strands exhibit hydrophobic peaks in which the peak of the second β-strand merges with or is immediately adjacent to that of the first.

As we compare the hydrophobic and amphipathic peaks with the 3D structure of the protein, we note that the side chains of the hydrophobic residues in TM β-strands all point toward the lipid bilayer while hydrophilic side chains point towards the interior of the pore, as expected. From the figure, we notice that the amphipathicity peaks of transmembrane β-strands 5 and 6 are not obvious. This is because loop L3, connecting strands 5 and 6, is folded back in the barrel, allowing these two strands to be shorter than normal. The short β-sheet in this region and the effect of a coiled structure near them diminishes the amphipathic peaks. Nevertheless, hydrophilic residues K89 in β-strand 4, D97 in β-strand 5, and R140 in β-strand 6 all face toward the inner channel of the barrel.

The OmpA-OmpF porin (OOP) family (TC# 1.B.6)

The large OOP family includes the functionally well-characterized OmpA porin of E. coli (Sugawara and Nikaido 1994) as well as the OmpF porin of Pseudomonas aeruginosa (Sugawara et al. 1996). These proteins and their many homologs are believed to form structures consisting of eight transmembrane β-strands (Baldermann et al. 1998). OmpA provides a model system for studying the mechanism of insertion, folding and assembly of constitutive integral membrane proteins both in vivo and in vitro. The function of OmpA is not currently well understood, but channel formation has been demonstrated (Arora et al. 2000).

Figure 2 ▶ shows the hydropathy and amphipathicity plots for the mature form of OmpA. Secondary structural information, based on the 3D structure (Pautsch and Schulz 1998, 2000) is shown at the bottom of the figure. As for the results presented in Figure 1 ▶, each TM β-strand generally corresponds to one hydrophobic peak and one amphipathic peak. The hydrophilic residues that contribute to each of these amphipathic peaks are K13 in β-strand 1, K35 in β-strand 2, E53 and D57 in β-strand 3, Q79 and K83 in β-strand 4, D93 and Q97 in β-strand 5, E129 in β-strand 6, R139, E141, Q143, and N147 in β-strand 7, and R170 in β-strand 8. They all face inwards, and are important for formation of the hydrophilic ion channel.

Fig. 2.

Hydropathy and amphipathicity plots for OmpA of the OOP family. The PDB code for OmpA is 1BXW.

The sugar porin (SP) family (TC #1.B.3)

The SP family includes the well-characterized maltooligosaccharide-specific maltoporin of E. coli, LamB, and the sucrose-specific porin, ScrY, of S. typhimurium. The 3D structures of both of these two proteins have been solved (Schirmer et al. 1995; Forst et al. 1998). Figure 3 ▶ shows our amphipathicity/hydrophobicity analysis for the mature form of the E. coli maltoporin. The 18 transmembrane β-strands are shown at the bottom of the figure. Most of the hydrophilic side chains contributing to the amphipathic peaks point inwards, lining the channel. These residues are, for example, R8 in β-strand 1, E43, K45, and E49 in β-strand 2, E61 and N63 in β-strand 3, E83, N85, Q87, and K89 in β-strand 4, R105 in β-strand 5, and so on. These residues are important for the channel characteristics as they form a hydrogen bonded network with the hydroxyl groups of the substrate sugar molecules.

Fig. 3.

Hydropathy and amphipathicity plots for LamB of the SP family. The PDB code for LamB is 1MAL.

The outer membrane receptor (OMR) family (TC #1.B.14)

The OMR family includes a large number of sequenced Gram-negative bacterial outer membrane proteins that form transmembrane pores and transport relatively large molecules from the external milieu to the periplasm in an energized process. As for the other families discussed above, no OMR member has been identified in a Gram-positive bacterium, an archaeon or a eukaryote. Energization of transport across the outer membrane requires a heterotrimeric complex of proteins, the TonB-ExbB-ExbD complex (Moeck and Coulton 1998) or in some cases, the TolA-TolQ-TolR complex (Locher et al. 1998). Energization requires the proton motive force (pmf) across the cytoplasmic membrane. In the absence of a pmf or one of the three energy-coupling proteins of the complex, the receptor binds its substrate, but transport does not occur.

The proteins in this family form a C-terminal 22-stranded β-barrel and an N-terminal plug domain. The plug is located inside the barrel and thus obstructs the channel interior. This domain tightly binds the barrel by more than 60 hydrogen bonds and nine salt bridges (Locher et al. 1998). Figure 4 ▶ shows the sequence analysis of the β-barrel domain of the TonB-dependent receptor, TolC. Similar to Figures 1–3 ▶ ▶ ▶, most of the transmembrane β-strands correspond to hydrophobic/amphipathic peaks, and most of the hydrophilic residues that contribute to the amphipathic peak face inwards to assist in the formation of hydrogen bonds and salt bridges. For example, E163, Q165, and K167 in β-strand 1, Q175 and D179 in β-strand 2, and R199 in β-strand 3 all line the channel.

Fig. 4.

Hydropathy and amphipathicity plots for FhuA of the OMR family. The PDB code for FhuA is 1BY3.

OmpX and phospholipase A

Two other outer membrane β-barrel proteins for which 3D structures are available are OmpX (Vogt and Schulz 1999) and phospholipase A (Snijder et al. 1999). OmpX belongs to a family of proteins that are important for virulence. They are believed to neutralize host defense mechanisms. Similar to OmpA, OmpX has eight transmembrane β-strands but they are different in shear number (a measure of the stages of the strands in the β-sheet (Koebnik et al. 2000), one of the important parameters of β-barrel structures). Phospholipase A, an enzyme that hydrolyzes phospholipids, has 12 transmembrane β-strands arranged in a barrel. These two proteins are similar to outer membrane β-barrel porins with hydrophilic residues facing inwards to form a hydrogen-bonded network. When analyzed for hydrophobicity and amphipathicity, most transmembrane β-strands in these proteins yield hydrophobic/amphipathic peaks (Figs. 5 and 6 ▶ ▶). It is not known if these β-barrels form transmembrane aqueous channels.

Fig. 5.

Hydropathy and amphipathicity plots for OmpX. The PDB code for OmpX is 1QJ8.

Fig. 6.

Hydropathy and amphipathicity plots for phospholipase A. The PDB code for Phospholipase A is 1QD5.

β-Barrel structural analyses: conclusions

From the hydropathy and amphipathicity analyses of the outer membrane β-barrel proteins of known 3D structure discussed above, we found that (1) most of the transmembrane strands correspond to a peak of hydrophobicity, although the hydrophobic values of these peaks are generally not as high as those of the transmembrane α-helices of cytoplasmic integral membrane proteins as expected; and (2) most of the transmembrane β-strands exhibit peaks of amphipathicity caused by the alternating hydrophilic residues located inside the barrel and the hydrophobic residues located outside the barrel. These properties of outer membrane β-barrel proteins have been noted previously by several investigators including our study of the proteins of the autotransporter (AT) family (TC #1.B.12) (Loveless and Saier 1997).

Whole genome sequence screening

Based on the sequence analyses described above, we identified criteria for searching for outer membrane β-barrel proteins in any protein sequence database, and based on these criteria, we developed a program for identifying such proteins. We combine three sequence analysis methods. The first is secondary structure prediction; the second is hydropathy analysis; and the third is amphipathicity analysis. For secondary structure prediction, we use the program Jnet, developed by the Barton group (Cuff et al. 1998; Cuff and Barton 2000). This program uses a two-level neutral network algorithm and gives better predictive results than the other programs we have examined. After obtaining the secondary structure results, we calculate the hydropathy and amphipathicity values using a window size of 7, which we found to be optimal for transmembrane β-sheets. Each predicted β-strand that also exhibits a peak of hydrophobicity and a peak of amphipathicity is recorded as a transmembrane β-strand. We can define an overall value based on these three parameters. The higher the value, the higher the probability that the region is in a true transmembrane β-strand, and the presence of multiple such regions increases the probability that the protein has a β-barrel structure. A specified threshold value can arbitrarily be assigned to allow the program to count potential TM β-strands.

Calculation method

We have used our program to screen proteins encoded within the E. coli genome (Blattner et al. 1997) for predicted transmembrane β-barrel structures. The whole genome sequence analysis includes three steps as follows: (1) automated hydropathy analysis of the protein sequences. Because there should be a transmembrane signal sequence for any protein precursor located in the outer membrane or periplasm, this step is used to rule out most cytoplasmic and integral inner membrane proteins. The program we used is a modified version of MEMSAT using default values for prediction of transmembrane α-helices (Jones et al. 1994, modified by us). Due to the inaccuracies of the program, we selected proteins that have at least one but no more than four putative TMSs with the first TMS within the N-terminal 50 residues. (2) Automated secondary structure, hydropathy, and amphipathicity predictions for each sequence in the genome database. Proteins that are selected exhibit 70% or greater of predicted transmembrane β-strands either for the whole protein if the full-length protein comprises the β-barrel, or for that domain that does comprise the β-barrel in the case of multidomain proteins. The cutoff point hydropathy value for β-strands is 1.5, and the amphipathicity value must be ≥1.0. However, if the hydropathy value is less than 1.5 but greater than 0.5, an amphipathicity value of ≥2.0 will compensate for the lower hydropathy value and the strand will be counted. (3) Automated similarity search. This third step is performed for each sequence to find homologs, some of which may be known as outer membrane proteins. Database searches are performed using the BLAST search tool (Altschul et al. 1990).