lilt.J . Peprrde Prorein Res. 42, I Y Y 3 .

4 Y l - 4 3 / Cupyri,'riyhr 0 Munksgaard 1993

Pnrired in Belgium - all rights reserved INTERNATIONAL JOURNAL OF
['EPTIDE & PROTEIN RESEARCH
ISSN 0367-8377

Prediction of transmembrane P-strands from hydrophobic

characteristics of proteins

M. MICHAEL GROMIHA and P.K. PONNUSWAMY

Depurtnietit of Phrsics. Bhnrathirlasriti C'riiversitj.. Tiriichirupalli. Tamil Nadir. India

Received 14 December 1992, accepted for publication 17 March 1993

The assembly of outer-membrane proteins consisting of 8-strands as transmembrane segments is somewhat

more complex when compared to the assembly of inner membrane proteins having a-helices as transmem-
brane parts. This is probably due to the difference in the amino acid sequences of the transmembrane part
strands and helices. Because of this feature, most predictive schemes which are successful in predicting
transmembrane helical segments fail to predict transmembrane strand segments. Here we propose a new
predictive scheme for forecasting the transmembrane strand segments in outer-membrane proteins with the
use of the general surrounding hydrophobicity scale developed both for the globular and membrane proteins
in the preceding article. Two major features of the scheme are (i) that it does not solely depend on the
amphipathic character of a sequence segment while identifying it as a transmembrane strand, and it is ca-
pable of predicting strands in varied lengths, a facility to reflect the variation in the membrane surfaces. This
scheme predicts the transmembrane P-strands in porin from R . capsulutus at 76"" accuracy (giving due weights
to over- and under-predictions when compared to X-ray results). The predicted 0-structure contents in OmpA,
porin from E. coli and maltoporin compared with the Raman spectroscopic results at 95% level. These ac-
curacy levels are far superior than the levels obtained from other existing methods. Apart from the above four
proteins for which experimental informations are available, ten other outer-membrane proteins, for which there
is no information about their secondary structure, are considered in the forecast. The results are analysed and
the coninion features in the folds of the set of fourteen outer-membrane proteins are deduced.
0 Munksgaard 1993.

During the last ten years we have been witnessing ex-
citing advances in the field ofmembrane proteins. These
advances relate to the accumulation of their amino acid
sequences, their common folds, the process of insertion
of their segments into the membrane matrix, transport
of species across the membrane etc. These advances in
our general understanding of membrane proteins pose
many interesting questions about the basic functional
characteristics of these very complexed macromolecul-
ar systems. However, answers to these questions can-
not be found unless high-resolution three-dimensional
structures of a statistically significant number of mem-
brane proteins are available. Until recently n o crystal
structures of membrane proteins have been solved at
high resolution except the photosynthetic reaction cen-
tre (PRC) from R . viridis (1) and porin from R . coppsii-
lrrfzrs(2). The lack of enough crystal structures, and the
availability of a statistically significant number of pro-
tein sequences. paved way for intensive structure-
predictive schemes of various kinds, and there contin-
ues to be much discussion and debate over their
420
formulations and practical applications. Of these pre-
dictive schemes. hydrophobicity analysis (hydropho-
bicity profile and hydrophobicity moment studies) re-
mains the central focus (3-7). Unfortunately, until
recently only the hydrophobicity scales developed for
the family of globular proteins were exploited in the
case of membrane proteins and obviously the outcome
was not very satisfactory (3, 8). In our preceding article
(9) we developed a hydrophobicity scale which is
equally applicable to both globular proteins and mem-
brane proteins and used it to formulate a scheme to
predict trans-helical segments in a large set of mem-
brane proteins. This scheme worked with a high degree
of success compared to other existing procedures. The
trans-strand segments in membrane proteins differ in
sequence nature and length comparcd to trans-helical
segments, and hence they warrant a separate scheme to
predict them. In this article we make such an attempt
using the more general hydrophobicity scale already
developed in ref. 9.
The outer-membrane proteins are now classified as

members with P-strands as transmembrane segments.
This class of proteins seem to follow a topology simi-
lar to the one exhibited by porin, a topology much
simpler when compared to those seen in the family of
globular proteins (10-12). The molecule of porin from
R. cupsulutus in the crystal consists of 16 anti-parallel
P-strands of different lengths (6- 17 residues) which
form a right-hand twisted P-barrel. A notable feature of
the amino acid sequences of the transmembrane seg-
ments of porin and other outer-membrane proteins is
that they are not long stretches of hydrophobic resi-
Transmembrane P-strands
dues, as is the case with transmembrane helical seg-
ments of other membrane proteins; it is probably this
property that makes the usual hydrophobicity-based
helix-predicting procedures fail in the case of trans-
strand predictions.
Paul & Rosenbusch (13) tried to predict transmeni-
brane strands by eliminating 8-turns and selecting a
minimal length of six residues for a strand; this is an
indirect method. Vogel & Jahnig (14), on the other
hand, tried to predict transmembrane strands on the
basis of the amphipathic character of segments, and

Porln ( R . capsulatus)

0
12.5-

+
.-
._
0
20 40 60 80 100 120 140

160 180 200 220 240 260 280

Amina acid seauence

a 20 40 60 80 100 120 140

2
(b)
I

I I
160 180 200 220 240 2 60 280

Amino acid sequence

FIGURE 1
Structure prediction plots for porin ( R . rupsulutus). (a) Continuous line: 12-residue surrounding hydrophobicity profile; filled squares: ob-
served peaks in the profile; dotted lines: the first seven sites of the 6-residue hydrophobicity profile for each low peaks; filled circles: observed
peaks in the 6-residue profile. (b) Portions of single residue hydrophobicity profile to fix the ends of strands.
42 1
M.M. Gromiha and P.K. Ponnuswamy
this procedure seems to be the popular one so far (14- essence, HD is simply the average of the surrounding
17). Neither all the transmembrane helices nor all the hydrophobicities of the residues constituting the strand.
transmembrane strands need to be amphipathic in char- The facial components f i and f 2 of the hydrophobic
acter. There are quite a few transmembrane sequences character help us to define an index to measure the
which exhibit either negligible or zero amphipathicity amphipathicity of the strand as
(2,9, 18, 19). It is therefore more appropriate to develop
a scheme which does not depend solely on the amphi-
pathicity of the segment in question. In this article we
develop such a general procedure to predict transmem- Prediction of' transnienibrane fl-strands
brane strands in outer-membrane proteins, and com- Structure predictions from amino acid sequence were
pare the results with those already reported by work- made from a plot of Hij from eqn. (1) against sequence
ers using other methods. In essence, the present method, number for a protein, with a pre-set value of m. In such
based on the concept of 'surrounding hydrophobicity' a hydrophobicity profile, the value of HBat a residue site
proposed in our laboratory. identifies the transmem- represents the hydrophobic character of a segment of
brane 8-strands to an accuracy of 76",, a much higher chosen residues starting from that residue number.
predictive level than the existing methods. The method This hydrophobicity profile will consist of mountains
is applied to porin from R . capsulatus, porin from E . coli. and valleq s about the average hydrophobicity line,
segment 1-177 of OmpA, maltoporin and 10 other which in turn is obtained from the hydrophobicity val-
outer-membrane proteins. ues of all the amino acid residues in the sample set of
proteins. The sample set included in the present work
consists of porin from R . capsulafus, porin from E. coli,
the N-terminal 1-177 segment of OmpA, maltoporin,
MATERIAL AND METHODS and other 10 outer-membrane proteins.
The mountains of the H D profile of a protein have
The surrounding hvdrophobicitj?scale been used to identify the secondary structural elements
The development of a general surrounding hydropho-
in I t . The mountains were classified into two groups,
bicity scale (for the 20 amino acid residues) applicable
those of high peaks and those of low peaks: peaks
to both the globular protein and membrane protein within a height of 0.1 kcal from the average line are
families has been described in ref. 9. This scale has been
considered as low peaks, and those having heights
obtained by judiciously combining the high-resolution
above the 0.1 kcal limit are considered to be high peaks.
crystal structure data of a set of 64 globular proteins
The high peaks were used to identify transmembrane
and of the two membrane proteins PRC and porin from
scgments with a length oft)?residues, and the low peaks
R . ccipsulatus. The same scale is used in the following
were used to identify strands which have lengths less
hydrophobicity profile computations for the outer-
than nz residues.
membrane proteins.
In any attempt to frame rules to predict transmeni-
brane strands one has to treat properly the three im-
H,vdrophobic character of ,&strands portant factors, namely the length of the strand, the
The hydrophobic character of a 8-strand has been de- termini of the strand and the role of the special residue,
termined by following the procedure proposed earlier proline. From their IR and X-ray results of porin from
by us (20). In this procedure a 8-strand segment is E. coli, Kleffel et al. (2 1) have suggested that the trans-
considered to have two faces, one of residues i, i + 2. membrane 8-strands may have an average length of
i + 4, ... i + ni - 1, and the other of rcsidues i + 1, i + 3, 10-12 residues. Using Raman data on segment 1-177
(

i + 5 , ..., i + nz, i being the N-terminal residue, and I I I of OmpA, Vogel & Jahnig (14) concluded that the av-
being the number of residues in the segment. The av- erage length of a strand may be 13 residues. Indicating
erage surrounding hydrophobicity, which is taken to be the uneven character of the surfaces of membrane pro-
a measure of the hydrophobic character of the segment, teins (due to factors such as the presence of channels),
is given by Paul & Rosenbusch (13 ) have chosen a minimum length
of 6 residues for a strand. Stoorvogel et al. (16) sug-
gested a length of 9 or 10 residues. Interestingly, the
molecule of porin from R. capsulatus contains 16
with 8-strands of lengths ranging from 6 to 17 residues in its
crystal form (2, 22). The above theoretical inferences
"f; = 1 (11, ,I,
f (2) and experimental results strongly support the view that
the segments traversing the membrane matrix may be
where h is the surrounding hydrophobicity index of the of varied lengths, but they are neither too long nor too
residue (H,,, in Table 1 of ref. 9), n is the total number short. Accordingly, in our predictive scheme we assume
of residues in the face, and j increases at an interval of that the lengths fall around two values, namely, 12 and
2 from 0 to in - 1 if i = 1, and from 1 to tn if i = 2. In 6 residues.
422

From an analysis of the terminal residues of the 16
P-strands in porin from R . cnpsulatus, we note that the
N - 1 and C + 1 positions are occupied mainly by the
three residues, Asp, Gly and Ala, but there was no
particular preference by any residue in the N and C
positions. However, a careful study of the hydropho-
bicity patterns of three or four residues preceding resi-
due N, and three or four residues succeding residue C,
indicate that out of 32 ends, 21 show a pattern of low
and high hydrophobicity values about the average line,
a typical P-strand pattern (23, 24). This indicates the
fact that most of the transmembrane strands acquire
amphipathic character at their terminal parts, even
though they are not so in total. In view of this obser-
vation, we make a specific rule in our predictive scheme
to look for the amphipathic character of the terminals.
Proline, owing to its peculiar backbone structure, is
not a favoured residue in the P-strands of normal globu-
lar proteins (25). It has been indicated by Deber et al.
(26) that prolines create thermodynamically less stable
structures in membranes. Hence we do not permit pro-
line to occupy positions within the length of a trans-
membrane strand, except at the ends. With these re-
strictions we constructed a predictive scheme consisting
of the following steps.
The predictive scheme
Step 1 . With the use of eqn. (l), construct hydro-
phobicity profiles with control lengths of in = 12, 6 and
1.
Step2. Consider the mountains in the t n = 12 profile.
Identify the top peak in each mountain, and classify
such selected peaks into high peaks and low peaks
according to their heights above the average hydropho-
bicity line: high peaks will have heights above 0.1 kcal
from the average line, and low peaks will have heights
equal to or below 0.1 kcal above the average line. As-
sume that each high peak corresponds to a transmem-
brane P-strand of around 12 residues whose ends will
be adjusted in step 4.
Step 3 . For each low peak, consider the first seven
sites of the m = 6 hydrophobicity profile, select the high-
est peak in it, and assume that it represents a trans-
membrane P-strand of the relevant six residues, whose
end residues will be fixed in step 4; if no peaks are
discernable in any of the seven sites, compute amphip-
athy indices Ag [from eqn. (3)] for the same first seven
six-residue segments, pick up the segment which has
the largest A0 value, and assume that this segment is a
transmembrane P-strand, whose end residues will be
fixed again in step 4.
Step 4 . List the predicted 12-residue segments (from
step 2) and 6-residue segments (from step 3); check
each of the 12-residue and 6-residue segments thus se-
lected, to see whether the single-residue (tn = 1) hydro-
phobicity profile continues beyond the end residues in
a zig-zag way, following a pattern characteristic of the
two faces of a strand, adopting high and low hydro-
Transniembrane P-strands
phobicity values about the average line: if so, extend the
length until the above pattern breaks.
Step 5 . Check whether Pro occurs inside the seg-
ments predicted in step 4; if so, cut the segment, and
select the largest of the resultants.
RESULTS AND DISCUSSION
Prediction of transriienibrnne D-strnrid.~iii poriii froin
R. capsulatus
As a working example, here we apply the above steps
to predict the transmembrane strands in the protein
porin from R. cnpsulntus:
Step I (compute the profile): Figure l a is the t ? i = 12
hydrophobicity profile for this protein.
Step2 (select the peaks and assign the longer strands):
16 peaks are marked on the in = 12 profile; of these 11
are high and 5 are low peaks; these peaks were selected
as follows: in the first mountain, sites 2-5 and 8-12
register surrounding hydrophobicities well above the
average line; as site 4 registers the highest hydropho-
bicity, it is taken to be the high peak in this mountain;
hence the corresponding 12 residues, viz. 4-15, are the
first segment to be selected. In the second mountain,
site 19 registers the highest hydrophobicity, and hence
residues 19-30 are to be selected as the relevant seg-
ment. All other peaks and the respective initial seg-
ments can be selected in the same way. The 16 initial
segments thus selected are given in Table 1.
Step 3 (assign shorter strands): Site 41 is the first low
peak in the porin profile (Fig. la). We have to look at
the nz = 6 hydrophobicity profile (not shown here)
around the seven six-residue segments, namely, 4 1-46,
42-41,43-48,44-49,45-50.46-5 1 and 41-52. We do
not observe any high peaks for these segments, and
hence we resort to computing the amphipathy indices
for the seven segments; of these, segment 41-46 has the
largest A Dvalue, and hence it is selected as a possible
transmembrane strand (computational details not pro-
vided).
Step 4 (fix the end residues): Consider the first seg-
ment 4-15, and look at the single-residue hydropho-
bicity profile (Fig. lb) for residues 1-4 and for residues
15-18: it shows that residues 1, 2, 3 and 4 alternate in
hydrophobicity about the average line; hence residues
1-3 can be appended, making residue 1 as the N -
terminus of the segment. Residues beyond 15 do not
alternate in their hydrophobicity values (not shonm):
hence, residue 15 itself can become the C-terminus of
the segment. Accordingly, residues 1 - 15 form the pre-
dicted transmembrane strand segment from the first
high peak of the profile. Consider the second segment
19-30. The single residue profile for residues 16-19
and 30-33 indicates that residues 18-31 alternate in
their hydrophobicities: hence residues 18 and 31 can
become, respectively, the N- and C-terminal ends of the
segment. Accordingly, residues 18-31 form the final
strand segment predicted from the second peak. Simi-
423
M.M. Gromiha and P.K. Ponnuswamy
TABLE 1 TABLE 1
Predicriori of rr(ztisr?ienibrune8-srmrids iri porirr (R. capsulatus) (continued)

Experimental (X-ray) results (2,22.27) Theoretical prcdictions

1-15 (15) 161-171 (11) Lenghts Lengths Final Over- under-

18-35 (17) 181-192 (12) from from predicted predi- predi-
39-46 (8) 195-206 (12) step 2 step 3 segmenis ction ction
59-65 (7) 227-130 (14)
68-74 (7) 243-254 (12) Vogcl & Jahnig (14)
118-125 (8) 258-271 (14) 1-14 (14) 0 0
128-135 (8) 275-285 (1 1 ) 20-34 (15) 1 1
148-158 (11) 292-301 (10) 38-47 (10) I 1
61-72 (13) 2 2
74-83 (10) 10 0
Theoretical predictions 0 8
128-139 (12) 4 0
Lengths Lengths Final Over- Under- 149-158 (10) 0 1
from from predicted predi- predi- 0 11
step 2 step 3 segments ction ction 183-192 (10) 0 2
0 12
Present work 210-219 (10) 10 0
4-15 1-15 (15) 0 0 222-231 (10)
19-30 18-31 (14) 0 4 233-242 (10) 8 1
41-46 39-47 (9) 1 0 243-252 (10) 0 2
59-70 59-72 (14) -1 0 256-265 (10) 2 4
72-77 72-77 (6) 3 0 277-286 (10) 2 7
I

118-129 118-129 (12) 2 0 292-301 (10) 0 0

129-134 129-134 (6) 0 1 Total 40 47
150- 16 1 149-162 (14) 2 1
162-167 162-168 (7) 0 3
Accuracy: 49.1 <:
186-197 183-199 (17) 2 I
Stoorvogel er ul. (16)
0 7
2 14-225 214-219 (6) 2-11 (10) 0 4
6 0
231-242 227-243 (17) 20-29 (10) 0 7
2 0
243-254 243-255 (13) 37-46 (10) 2 0
1 0
260-27 1 58-67 (10) 4 0
259-271 (13) 0 1
27 1-282 271-283 (13) 68-76 (10) 3 0
2 2
2 92 -2 9 7 292-301 (10) 106-115 (10) 10 0
0 0
Total 0 8
23 20
128-137 (10) 2 0
Accuracy: 75.8", 148-157 (10) 0 1
158-167 (10) 3 4
Paul & Rosenbusch (13) 182-191 (10) 0 2
8-13 (6) 0 8 192-201 (10) 3 5
0 17 227-236 (10)
39-57 (19) 11 0 237-246 (10) 2 8
61-89 (28) 18 1 264-273 (10)
109-114 (6) 6 0 274-283 (10) 5 6
0 8 292-301 (10) 0 0
125-134 (10) 2 1 Total 34 45
139-158 (20) 8 0
Accuracy: 5 3.8 O 0
163-170 (8) 0 3
176-206 (31) 7 0
210-227 (18) 16 0 larly, considering the segment 4 1-46, the relevant por-
231-236 (6) 0 6
tions of the single residue hydrophobicity profile indi-
240-245 (6) 2 3
219-261 (13) 3 8
cate that residues 39-41 and 45-47 have the pattern of
275-282 (8) 0 -1 a strand; hence the length of this strand can be fixed as
291-301 (11) 1 0 39-47. We could fix all the lengths in a similar way. In
Total 74 57 Table 1 we list the sites of the 16 peaks (1 1 high peaks
and 5 low peaks), the correspondingly selected 12- or
Accuracy: 29.8",
6-residue initial segments and the final predicted trans-
membrane strands for porin.
424
 Transmembrane p-strands
TABLE 2 TABLE 2
Predicted trutistnenibrutie j-strunds it1 OtnpA. porin (E. coli). niulfoiporiti (continued)
arid OmpX protein
Presently predicted Amphipathy Previously suggested
Presently predicted Amphipathy Previously suggested scgmcnts (AD ) segments
scgment s ($' 7) segmcnts
OmpX ref. 16
Omp A (1-177) segment ref. 14 1-6 (6) 1.23 1-10 (10)
6-19 (14) 0.85 7-18 (12) 23-32 (10)
34-46 (13) 0.76 33-44 (12) 36-46 (11) 0.67 37-46 (10)
49-64 (16) 1.50 49-61 (13) 60-68 (9) 0.28 61-70 (10)
76-86 ( 1 1 ) 1.46 74-87 (14) 74-86 (13) 0.13 76-85 (10)
90-105 (16) 0.26 91-104 (14) 108-118 (11) 0.49 107-116 (10)
121-132 (12) 0.61 119-131 (13) 122-134 (13) 0.47 122-131 (10)
138-154(17) 1.17 135-146 (12) 135-149 (15) 0.09 140-149 (10)
160-171 (12) 0.59 159-170 (12) j-content 52.3:0 53.7""
/?-content 62.7y0 57.6";
p-content from Raman data: 687;
In Table 1 we include the theoretical results provided
Porin ( E . coli) ref. 13 by three other groups for comparison with our present
10-24 (15) 0.78 17-23 (7) results. In their procedure Paul & Rosenbusch (13) first
54-66 (13) 1.82 38-51 (14) identified turns in a segment of three or more residues;
83-94 (12) 0.12 59-66 (8) then, by eliminating the turns, they predicted the mem-
95-106 (12) 0.31 71-76 (6)
brane Ij-strands as a long stretch of (minimum 6) amino
116-121 (6) 0.87 82-96 (15)
128-141 (14) 0.09 103-109 (7)
acid residues. In an alternative way, Vogel & Jahnig
148-160 (13) 0.65 128-138 (11) (14) computed the amphipathicity of segments with 9
169-183 (15) 0.78 144-158 (15) or 10 residues and selected those lengths which had as
184-195 (12) 0.06 178-194 (17) high an amphipathicity as the probable Ij-strands.
213-225 (13) 0.80 199-205 (7) Stoorvogel et al. (16), in their procedure of selecting the
225-236 (12) 0.34 211-218 (8) transmembrane strands, included two additional rules,
255-269 (15) 0.58 222-229 (8) along with amphipathicity, (i) that the hydrophilic
271-281 (11) 0.73 255-269 (15) maxima of the protein are exposed at the cell surface
291-303 (13) 0.61 280-303 (24) and (ii) that these maxima are separated by about 40
306-317(12) 0.45 311-318 (8)
residues.
326-340 (1 5) 1.25 332-340 (9)
50",
We note from Table 1 that the Ij-turn elimination
8-content 57.3%
j-content from Raman data: 60% procedure of Paul & Rosenbusch (13) has predicted
131 residues wrongly (74 over-predictions and 57
MaltoporIn ref. 28 under-predictions) in the 16 /?-strands, the accuracy of
1-14 (14) 1.15 19-24 (6) prediction being just around 30% [the accuracy is com-
63-77 (15) 1.19 33-53 (21) puted using eqn. (4) of ref. 91. The results of Vogel &
97-102 (6) 0.33 64-76 (13) Jahnig (14) show 164 residues in the trans p-strands,
112-124 (13) 0.03 80-87 (8) with 40 over-predictions and 47 under-predictions; this
131-144(14) 0.27 113-122 (10) method thus performs at an accuracy level of 49%. The
149-160 (12) 0.22 127-144 (18) method of Stoorvogel et al. (16) has a slightly higher
170-182 (13) 0.68 166-181 (16)
accuracy level, namely 54%. On the other hand, the
186-193 (8) 1.92 186-191 (6)
204-210 (7) 0.97 202-208 (7)
present method provides results at an accuracy level of
213-224 (12) 0.97 213-235 (23) 767;.
224-234 ( I 1) 0.96 252-263 (12) The data of Table 1 also show that the present
248-254 (7) 0.11 268-273 (6) method predicts 2 1 terminal residues correctly within
267-278 (12) 0.23 284-297 (14) an error of 2 residues; on the other hand, the method
284-298 (15) 0.56 303-326 (24) of Vogel & Jahnig predicts 20 terminal residues, the
304-31 1 (8) 0.57 331-336 (6) method of Stoorvogel etal. predicts 18 terminal resi-
318-327(10) 0.02 342-353 (12) dues, and the method of Paul & Rosenbusch predicts
342-354 (13) 1.45 362-376 (15) 12 terminal residues, correctly within the same error
361-366 (6) 0.44 388-397 (10)
396-403 (8) 0.65
limit.
410-421 (12)
410-421 (12) 1.83 Perdiction of transmembrane P-strmds ,for other oufer-
8-content 51.37; 55";
membrane proteins
8-content from Raman data: 49.5%
In addition to R . capsulatus porin, we selected 13 other
outer-membrane proteins for the purposes of predicting
425
M.M. Gromiha and P.K. Ponnuswamy
TABLE 3 TABLE 3
Predicred ~raiisiv~enibraiie
srrarrds 111 iiriie orher oiiter iireitibraiie prore firs (continued)

Presently Amphi- Prescntlj Amphi- Presently Amphi- Presently 4mphi-

predicted Path) predicted path) predicted pathy predicted pathy
segments (A01 segments (Air) segnients ( AJI segments ('40)

OmpV protein, ref. 33 162-174(13) 0.94 404-4 15( 12) 0.60

1-14 (14) 0.36 113-125(13) 0.75 177-192(16) 0.13 439-457( 19) 0.20
15-27 (13) 0.79 133-147 (15) 0.41 fl-content 52 9",
35-48 (14) 0.3 1 149-161 (13) 0.08
49-65 (17) 0.47 186-195 (10) 0.80 PI. ref 39
67-79 (13) 0.20 206-212 (7) 0.11 5-14 (10) 0.22 180-1 91( 12) 0.45
97-111 (15) 1.64 227-238 (12) 0.97 19-32 (14) 1.43 221-232(12) 0.14
8-content 65.5"" 52-60 (9) 0.61 235-246( 12) 0.02
72-85 (14) 1.13 250-264( 15) 0.49
MOMP, ref. 34 96-109 (14) 0.07 267-280( 14) 0.43
10-21 (12) 0.13 182-194 (13) 0.03 121-128(8) 0.32 282-295( 14) 1.13
31-43 (13) 0.33 202-215 (14) 0.06 130-142 (13) 0.00 318-329(12) 0.33
47-53 (7) 0.39 247-261 (15) 1.20 158-169(12) 0.23 334-348( 15) 0.07
71-76 (6) 0.61 264-278 (15) 0.02 8-content 57 5",,
97- 109 ( 13) 0.13 279-285 (7) 1.39
110-125 (16) 0.73 300-312 (13) 1.50 PhoE. ref. 10
126-137 (12) 0.06 314-327 (14) 0.03 10-25 (16) 1.05 172-187( 16) 0.06
153-162 (10) 0.07 335-347 (13) 0.17 35-41 (7) 1.40 2 13-227(
1 5 ) 0.60
168-182 (15) 0.00 356-370 (15) 0.31 49-60 (12) 1.50 262-269(8) 0.24
8-content 60.1 93-105 (13) 0.56 280-292( 13) I .04
126-134 (9) 0.45 296-309( 14) 0.71
Protein H1, ref. 35 137-152 (16) 0 . I4 3 13-325(
1 3 ) 0.76
6-18 (13) 0.23 113-126 (14) 0 . I5 []-content 46 I",
27-39 (13) 0.11 137-150(14) 0.85
55-67 (13) 0.97 155-166 (12) 0.83 OmpT. ref 40
75-81 (7) 1.10 184-200 ( 17) 0.96 1-12 (12) 0.25 158-170( 13) 0.60
96-105 (10) 0.65 16-28 (13) 0.07 185- 192(8) 1.OO
8-content 56.5"" 31-43 (13) 0.98 199-2 10( 12) 0.85
67-80 (14) 1.17 21 7-226( 10) 0.91
Protein P2. ref. 36 83-92 (10) 0.36 249-262( 14) 0.61
4-16 (13) 0.30 193-206 (14) 0.35 92-103 (12) 0.53 264-278( 15) 0.90
30-42 (13) 1.23 213-229(17) 0.65 139- I44 (6) 1.41 300-3 17(18) 0.25
56-73 (18) 0.63 236-248 ( 1 3 ) 0.25 8-contcnt 53 6""
78-84 (7) 0.84 250-263 0.65
101-113 (13) 0.68 279-285 1.13
126-133 (8) 0.74 31 1-323 0.54 their transmembrane strands. Of these, the amounts of
138-144 (7) 0.85 325-336 0.44 secondary structures present are known from Raman
161-173 (13) 0.36 350-361 0.81 spectroscopy for three proteins, namely fragment 1-177
/?-content 53.5"" of OmpA (14), porin from E. coli (13) and maltoporin
(28). The secondary structure of OmpX protein was
Protein F. ref 37 predicted by Stoorvogel et 01. (16). We first applied our
25-38 (14) 0.07 149-161 1.22 technique to these four proteins and compared our re-
40-48 (9) I .46 206-213 (8) 0.63
sults with the transmembrane segments already indi-
71-80 (10) 0.42 271-282 (12) 0.60
89-105 (17) 0.5 I 283-295 (13) 0.72
cated. The results pertaining to these four proteins are
109-124 (16) 0.58 3 12-326 ( 15) 2.02 given in Table 2. The surrounding hydrophobicity pro-
8-content 39 0"" files and the resultant folding patterns for segment
1-177 of OnipA, porin ( E . coli) and maltoporin are
Protein PI, rcf. 38 presented in Figs. 2-4. The predicted results for the
7-18 (12) 0.18 192-204 (13) 0.30 rest of the nine outer-membrane proteins for which no
28-41 (14) 0.41 240-251 (12) I .07 information is available concerning their secondary
43-54 (12) 0.27 257-272 (16) 1.34 structures are prescnted in Table 3 .
64-81 (18) 0.66 313-324 (12) 0.55
83-95 (13) 0.25 350-362 (13) 0.62
118-123 (6)
A few biological studies carried out on mutants of
0.15 375-386 (12) 0.44
130- I46 (17) 1.25 391-403 (13) 0.54
OmpA and maltoporin provide information about the
exposure of certain regions of the polypeptide chain
426

facilitating functions, such as antibody binding, phage
adsorption etc. These regions, being mostly hydrophilic,
should fall on the surface or loop regions of the folded
protein. Morona et ul. (29, 30) studied the mutants of
OmpA, and indicated that the regions enclosing sites
25, 70, 110 and 154 are exposed on the cell surface.
From our folding model of 1-177 OmpA (Fig. 2b) we
note that residues 25, 70 and 110 are exposed on the
cytoplasmic surface, each within a large loop segment;
residue 154 is also very near the surface (Fig. 2b).
Gehring et a]. (3 1) indicated that in maltoporin the
residue positions 18 and 259 are phage-resistant, resi-
due position 333 is an antibody binding site, and re-
gions near residues 150 and 380 are on the exterior
surface. Ferenci et ul. (28) hypothesised that in the
starch-binding mutants of maltoporin, residues 163 and
245, are accessible from outside of the pore. Shenkman
et ul. (32), from their proteolysis studies of LamB pro-
tein, suggested that sites near residues 155,203 and 380
Transmembrane 8-strands
are accessible from the outside of the membrane. A11
these suggestions are acceptable in our folding model
(Fig. 4b): except for residue 155, none of these residues
falls inside the predicted trans 8-strand regions.
Our method has predicted eight trans 8-strands in
1-177 OmpA, the average length of these strands being
14 residues. The predicted /?-strand residues constitute
63 % of the total residues, which is close to the predic-
tion (68%) from Raman data; the method of Vogel &
Jahnig (14) predicts only 58% of the residues to be in
the 8-structure. From Table 2 and Fig. 3a we note that
the present method predicts 16 trans /J-strands for the
molecule of porin ( E . coli), most of them ranging in
length from 6 to 15 residues; the results show that about
57% of the residues are in strands, as against 6070
predicted from Raman data; the method of Paul &
Rosenbusch (13) predicts only 5 0 x of the residues in
strands. For maltoporin, 20 transmembrane strands
are predicted, which constitute about 51 O 0 of the total

I 1
20 40 60 eo 100 120 140 160
Amino acid sequence

OmpA protein

llOV y G K
1 “ N
F H R
V E T
G P D P
7

T T 30 G T G 0
H G
)D I
V
s I20 T

H
A150

D
G Y

M I
(b) ~~ GT
Y lo a N
N
T
W
0
Y ~170
140
W v G

T G 50 L 0
P

‘a
-
N
D “
V
E -
T
R G
E
P I
P K A A pi77

FIGURE 2
(a) Structure prediction plot for the segment 1-177 of OmpA protein. (b) Folding model of the OnipA fragment 1-177; the transmembrane
/&strands are shown in boxes.

427
M.M. Gromiha and P.K. Ponnuswamy

.$
r
0
".'I . , , , , , , , j
a
n 20 40 60 80 I00 120 140 I60

I 180 200 220 240 2 60 2 80 300 320 340

Amino acid sequence

DG Ty40
G A
G R Porin ( ~ . c o I i )
L
S G
F
30 N
a G
K Q P L G N
G cT
N ' S_
ti
250 N K
G
N E A
R2@%
A
E
K
K210 K
r
- D L
G
GK a 50 L A
N G D 320S -
T K I V

0 _I
R N E ~ 2 9 0D V
Y
- 01 L " c _ _

I
P G
A T
3
r A S
13C
D
33(
I G

Y R
G V
10
G
N G G
Y
Y V
T A
W
V T V

V Y Y
141
Y E
R
T
- Y
N Y G
D -
K80 s L
N N
G
K E T F F
N G

' AE
IY A D AE

FIGURE 3
(a) Structure prediction plot for porin ( E . coli). (b) Folding model of porin ( E . coli)

residues. This is in good agreement with the Raman From the predicted results of the other nine outer-
spectroscopic results, which indicate that 50% of resi- membrane proteins (10, 33-40) in Table 3, we observe
dues are in strands; the method of Ferenci etal. (28) that thc p-strands traverse many times inside the mem-
predicts 5 5 % residues in strands. In OmpX, although brane in all these proteins: OmpT has 14 strands, OmpV
the predicted p-content is approximately equal to that has 12 strands, MOMP and P1 each have 18 strands
predicted in the present method, the lengths of all the etc. The predicted P-strands are even in number in most
strands are equal in the method of Stoorvogel e t u l . , of the proteins (12 out of 14 proteins), including porin
whereas they adopt different lengths in the present ( R . cupsulutus), for which the X-ray structure is known.
method. Interestingly, the total number ofthe predicted P-strands
428
 Transmembrane /I-strands

.-
.-0
D
c
P 20 40 60 80 100 120 140 160 180
1
200

220 240 260 280 300 320 340 360 380 400 420
Amino acid sequence
40

Ma1t o w r i n
G W
T K
E
G
D

1
K
S
K
Y
F 60 D
G
To R H l10
R
s
S
‘“3s N

P
s
Y K W
M T

M270

W
H
w s
W G Y
N N O
N =ON
L G
~130 A O300 v
‘ N I N
D N90 A
L YT
P G G N E
80 A 0
F v
P E A N

FIGURE 4
(a) Structure prediction plot for maltoporin. (b) Folding model of maltoporin.

and the total number of residues in 8-strands do not strands; 16 8-strands are predicted both in porin
depend on the total number of residues in a protein. ( R . capsufatus)and in protein P2, even though they dif-
Raman data show that porin ( E . coli) with 340 residues fer by 60 residues. Similarly, OmpV protein, with 238
has a 8-strand residue content of 60%, whereas m d - residues, has 12 strands, but only 10 strands are pre-
toporin, with 421 residues, has only 49.5% of them in dicted in protein F, which has 326 residues.
429
M.M. Gromiha and P.K. Ponnuswamy
TABLE 4 ticle (which is equally applicable to globular proteins,
Aniphipathic characrers of ~-straridsiti porirr / R. capsulatusi inner-membrane proteins involving trans-helices and
outer-membrane proteins involving trans-strands), and
Observed Amphi- Observed ‘Amphi- a few well defined steps to determine the lengths and
p-strand pathy /3-strand pathy the end residues of strands, we were able to predict
(X-ray) AP (X-ray) A /I successfully the transmembrane strand segments in 14
1-15 (15) 1.37 161-171 ( 1 1 ) 0.07 outer-membrane proteins: the predicted result for porin
18-35 (17) 0.83 181-192(12) 0.45 from R . c~apszriatzisis at an accuracy level of 76% with
39-46 (8) 1.54 195-206 ( I ? ) 0.30 the X-ray structure, and those for the segment 1-177
59-65 (7) 0.50 227-240 (14) 0.03 of OmpA, porin ( E . coli) and maltoporin come to an
68-74 (7) 1.42 243-254 (12) 0.10 accuracy level of 95”, with Raman results. These ac-
118-125 (8) 0.24 3 8 - 2 7 1 (14) 0.35 curacy levels are far superior to those obtainable from
128-135 (8) 0.55 275-285 ( I I ) 0.14 methods available in the literature.
148-158 (11) 0.66 292-301 (10) 1 .jj

The first residue of the very first P-strand segment is
nearer to the N-terminus, and the last residue of the
final segment is nearer to the C-terminus in most of the
outer-membrane proteins. Also, the strands are having
lengths ranging from 6 to 16 residues, but most of them
have residues mainly around 12 residues. One strand in
protein P1 is 19 residues in length. An analysis of the
P-strand contents in all 14 outer-membrane proteins
studied shows that their 8-contents range from 39 to
66 yo,the average being 54 a/, .
Amphipathie character of P-strands in the
outer-membrane proteins
The amphipathic character of the fi-strand segments in
the crystal of porin ( R . ciipsulatus),computed using eqn.
(3), are given in Table 4. We considered the segments
having a amphipathicity index Ag> 1.0 as highly am-
phipathic. From Table4 we note that only the four
segments, 1-14, 39-46, 68-73 and 292-301 are highly
amphipathic. Interestingly, segments 161-17 1, 227-
240,243-254 and 258-269 do not have any noticeable
amphipathicity ( A D < 0.1). A study of the amphipathic-
ity of the P-strand segments in the rest of the 13 pro-
teins follows what we have seen in porin ( R . cnpsuia-
tus): only 25 7; of the segments are highly amphipathic.
CONCLUSIONS
The membrane assembly of the fi-barrel proteins of
bacterial outer membranes seems to be governed by
rules that are more complex than those for helical mem-
brane proteins. Accordingly, the methods which were
successful in predicting trans-helices fail to predict cor-
rectly the membrane P-strands. The major difference
between the trans-helix scquences and trans-strand se-
quences is their hydrophobic character: trans-helices
have a stretch of continuous hydrophobic residues,
whereas in the trans-strands certain hydrophilic resi-
dues intervene between the hydrophobic members. This
warrants some complicated steps in formulating rules
to predict transmembrane strands. Using the surround-
ing hydrophobicity scale developed in the previous ar-
ACKNOWLEDGEMENTS
M.M.G thanks the CSIR, Governmcnt of India, for providing fi-
nancial support to carry out this work.
REFERENCES
1. Deisenhofer. J . & Michel, H. (1989) Science 245, 1463-1473
2. H’ciss, M.S., Abele, U., Weckessor, J., Welle, W. & Schulz. G.E.
(1991) Srierice 254. 1627-1630
3. Kyte, J . & Doolittle. R.F. (1982) J . Mol. Biol. 157. 105-132
4. Argos. P.. Rao. J.K.M. & Hargrave, P.A.H. (1981) Eur. J . Bio-
c,herji. 128, 565-575
5 , Engelman, D.M.. Steitz, T.A. & Goldman, A. (1986)Anrlu. Rev.
5iophI.s. BiophJ,.y.Cherii. 15, 321-353
6. Esposti. M.D., Crimi. M. & Venturoli, G. (1990) Eur. J . Bio-
cher~i.190. 207-219
7 . Eisenbcrg. D.. Schwarz. E., Konaromy, M . & Wall, R. (1984)J.
Mo1. Bid. 179, 125-142
8. Fasman, G.D. & Gilbert, W.A. (1990) Trends. Blochem. Sci. 15,
89-92
9. Ponnuswamy, P.K. & Gromiha, M.M. (1993) hit. J . Peptide
Proreiri Rex. 42. 326-343
10. Overbeeke. N., Bergmans, H., van Mansveld, F. & Lugtenberg,
B. (1983)J. Mol. Bid. 163, 513-532
11. Inokunchi. K., Mutoh, N., Matsuyama, S.I. & Mizushima, S .
(1982) A”v’rc1. Acids Res. 10, 6957-6958
12. Clement, J.M. & Hofnung, M . (1981) CeII 27. 507-514
13. Paul, C. & Rosenbusch, J.P. (1985) EMBO J . 4, 1593-1597
14. Vogel. H . & Jahnig. F. (1986) J . hfol.Bid. 190, 191-199
15. Charbit, A,, Gehring. K., Nikaido, H., Ferenci, T. & Hofnung,
M. (1988) J . Mo1. Bid. 201. 487-496
16. Stoorvogel, J.. van Bussel, M.J.A.W.M., Tommassen, J. & van
de Klundert, J.A.M. (1991) J . Bacreriol. 173, 156-160
17. Jahnig, F. (1990) Trends. Biocher~t.Sci. 15, 93-95
18. Nakashinia, Y.. Frangioro, B., Wiseman, R.L. & Konigsberg,
W . H . (1981) J . B i d Chenr. 256, 5792-5797
19. Yaniamotto, T.. Davis, C.G.. Brown, M.S., Schneider. W.J.,
Caesy, M L . , Goldstein, J.L. & Russel, D.W. (1984) Ce// 39,
28-37
20. Muthusamy, R. & Ponnuswamy, P.K. (1990) I r i r . J . Peptide Pro-
reiri Res. 35, 378-395
21. Kleffel, B.. Garnvito, R.M., Baumeistcr, W. & Rosenbusch. J.P.
(1985) E.MBOJ. 4. 1589-1592
22. Wciss. M.S., Kreusch, A,, Schiltz, E., Nestel. U., Wcltc, W..
Wcckesser, J . & Schulz, G.E. (1991) FEBS Lett. 280, 379-382
23. Cid. H.. Bunstcr, M.. Arriagada, E. & Campus, M. (1982) FEBS
Lerr. 150, 247-254

430
 Transmembrane P-strands
24. Ponnuswamy, P.K. (1993) Progr. Biophjs. Mol. Biol. 59, 57-103 36. Munson, R., Jr. & Tolan, R.W., Jr. (1989) Infect. Imniw. 57,
25. Mac Arthur. M.W. & Thornton, J.M. (1991)J. Mol. Biol. 218, 88-94
397-412 37. Duchene. M . . Schweizer, A., Lottspeich, F.. Krauss. G., Marget,
26. Deber, C.M.. Glibowicka. M. & Woolley, G.A. (1990) Biopoly- M., Vogel. K. von Specht, B.L. & Domdey, H . (1988) J . Bac-
tner.~29, 149- 157 feriol. 170, 155-162
27. Weiss, M.S. & Schulz. G . E . (1992) J . Mol. Biol. 227, 493-509 38. Munson. R., Jr. & Grass, S . (1988) Infect. fniwui. 56. 2235-
28. Ferenci, T., Saurin, W. & Hofnung, M. (1988)J. Mol. Biol. 200, 2242
493-496 39. Gotschlich, E.C., Seiff, M.E.. Blake, M.S. & Koomey, M. (1987)
29. Morona. R., Klose, M . & Henning, U.(1984) J . Bacteriol. 159, Proc. Natl. Acad. Sci. USA 84, 8135-8139
570-578 40. Grodberg. K., Lundrigan, M.D., Toledo, D.L., Mangel, W.F. &
30. Morona, R., Kramer, C. & Henning, U. (1985) J. Bacteriol. 164, Dunn. J.J. (1988) Nucl. Acids Res. 16, 1209
539-543
31. Gehring, K.. Charbts, E. & Hofnung, M. (1987)J. Bacteriol. 169,
2103-2 I06 Address:
32. Shenkmans, S., Tsugiya, A., Schwartz, M. & Rosenbusch, J.P. Dr. P . K . Ponnusw~atny
(1984) J. Biol. Clieni. 259, 7570-7576 Professor and Head
33. Pohlner, J., Meyer, T.F., Jalajakumari, M.B. & Menning, P.A. Department of Physics
(1986) Mol. Gen. Genet. 205, 494-500 Bharathidasan University
34. Pickett. M.A., Ward, M.E. & Clarke, I.N. (1987) FEMS Micro- Tiruchirapalli 620 024
biol. Lett. 42, 185-190 Tamil Nadu
35. Bell, A. & Hancock, R.E.W. (1989)J. Bacteriol. 171,3211-3217 India

43 1