Download Theoretical studies of Membrane Proteins

Theoretical studies of Membrane Proteins
Properties, Prediction Methods and Genome-wide analysis
PhD Thesis
Erik Wallin
Department of Biochemistry
Stockholm University
Sweden
Email [email protected]
October 21, 1999
Cover illustration
The membrane protein complex bovine cytochrome c oxidase, forms a dimer
in the crystal structure. The picture shows the membrane spanning helices
±10 Å from the center of the membrane, looking down through the membrane.
Only the bonds between Cα backbone atoms are plotted. The color coding
shows the amount of residue variability, calculated from alignment to other
mammalian cytochrome c oxidase sequences. Conserved residues are red and
variable residues blue.
It is clear that the core residues are more conserved (red) and the residues facing
the lipid environment more variable (blue). This is not unexpected since the
lipid interactions are not as specific as the packing in the protein core.
The author can be reached at [email protected] or
http://www.biokemi.su.se/˜erikw
ISBN 91-7265-012-5 Printed
ISBN 91-7265-013-3 CD-ROM
ISBN 91-7265-014-1 Online (PDF) at http://www.sub.su.se/diss/1999/91-7265014-1
2
Contents
Abstract
7
Acknowledgements
8
List of publications
10
Abbreviations
12
1 Introduction
13
1.1
1.2
Classes of membrane proteins . . . . . . . . . . . . . . . . . . . .
13
1.1.1
Helix bundle integral membrane proteins
. . . . . . . . .
14
1.1.2
Architectural overview of helix bundle membrane proteins
14
1.1.3
β barrel integral membrane proteins . . . . . . . . . . . .
14
Experimental studies of helix bundle membrane proteins . . . . .
15
1.2.1
Structural determination of membrane proteins . . . . . .
15
1.2.2
Mutational studies . . . . . . . . . . . . . . . . . . . . . .
15
1.2.3
Topology mapping . . . . . . . . . . . . . . . . . . . . . .
19
2 Prediction of membrane protein topology and structure
2.1
2.2
21
Predictable features . . . . . . . . . . . . . . . . . . . . . . . . .
21
2.1.1
Hydrophobicity scales . . . . . . . . . . . . . . . . . . . .
22
Overview methods for topology prediction . . . . . . . . . . . . .
22
2.2.1
TopPred . . . . . . . . . . . . . . . . . . . . . . . . . . . .
22
2.2.2
TopPred Multi . . . . . . . . . . . . . . . . . . . . . . . .
24
2.2.3
PHDhtm . . . . . . . . . . . . . . . . . . . . . . . . . . .
25
2.2.4
TMAP . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
25
2.2.5
MEMSAT . . . . . . . . . . . . . . . . . . . . . . . . . . .
26
3
2.3
2.2.6
PRED-TMR . . . . . . . . . . . . . . . . . . . . . . . . .
26
2.2.7
TMHMM . . . . . . . . . . . . . . . . . . . . . . . . . . .
26
2.2.8
HMMTOP . . . . . . . . . . . . . . . . . . . . . . . . . .
26
2.2.9
The dense alignment surface method . . . . . . . . . . . .
27
Prediction of membrane protein structure . . . . . . . . . . . . .
27
3 Architecture of helix bundle membrane proteins
3.1
3.2
Intra- and extracellular loops . . . . . . . . . . . . . . . . . . . .
29
30
3.1.1
Arg and Lys content is biased towards the intracellular side 30
3.1.2
Amino acid bias in the N-terminal and the first loop . . .
31
3.1.3
Other amino acid biases . . . . . . . . . . . . . . . . . . .
31
3.1.4
Charge distribution within the loops . . . . . . . . . . . .
31
3.1.5
Loop length and translocation
. . . . . . . . . . . . . . .
32
3.1.6
Loop length in general . . . . . . . . . . . . . . . . . . . .
32
3.1.7
A coil region near the membrane? . . . . . . . . . . . . .
33
Membrane spanning helices . . . . . . . . . . . . . . . . . . . . .
34
3.2.1
Length of transmembrane helices . . . . . . . . . . . . . .
34
3.2.2
Hydrophobicity and exposure to lipids . . . . . . . . . . .
35
3.2.3
Distribution of other residue types . . . . . . . . . . . . .
35
3.2.4
Helix breakers and charged residues . . . . . . . . . . . .
35
3.2.5
Correlation between residue variability and exposure . . .
36
3.2.6
Helix ends . . . . . . . . . . . . . . . . . . . . . . . . . . .
36
4 Genome wide studies of membrane proteins
39
4.1
Membrane protein content . . . . . . . . . . . . . . . . . . . . . .
39
4.2
Number of transmembrane segments . . . . . . . . . . . . . . . .
40
4.3
Size vs number of transmembrane segments . . . . . . . . . . . .
40
5 Summary
43
6 Papers
49
4
Till minne av
Christina
5
6
Abstract
Membrane proteins are a large and important class of proteins. They are responsible for several of the key functions in a living cell, e.g. transport of nutrients
and ions, cell-cell signaling, and cell-cell adhesion.
Despite their importance it has not been possible to study their structure and
organization in much detail because of the difficulty to obtain 3D structures.
In this thesis theoretical studies of membrane protein sequences and structures
have been carried out by analyzing existing experimental data. The data comes
from several sources including sequence databases, genome sequencing projects,
and 3D structures. Prediction of the membrane spanning regions by hydrophobicity analysis is a key technique used in several of the studies. A novel method
for this is also presented and compared to other methods.
The primary questions addressed in the thesis are: What properties are common
to all membrane proteins? What is the overall architecture of a membrane
protein? What properties govern the integration into the membrane? How
many membrane proteins are there and how are they distributed in different
organisms? Several of the findings have now been backed up by experiments.
An analysis of the large family of G-protein coupled receptors pinpoints differences in length and amino acid composition of loops between proteins with and
without a signal peptide and also differences between extra- and intracellular
loops. Known 3D structures of membrane proteins have been studied in terms
of hydrophobicity, distribution of secondary structure and amino acid types, position specific residue variability, and differences between loops and membrane
spanning regions.
An analysis of several fully and partially sequenced genomes from eukaryotes,
prokaryotes, and archaea has been carried out. Several differences in the membrane protein content between organisms were found, the most important being
the total number of membrane proteins and the distribution of membrane proteins with a given number of transmembrane segments. Of the properties that
were found to be similar in all organisms, the most obvious is the bias in the
distribution of positive charges between the extra- and intracellular loops.
Finally, an analysis of homologues to membrane proteins with known topology
uncovered two related, multi-spanning proteins with opposite predicted orientations. The predicted topologies were verified experimentally, providing a first
example of “divergent topology evolution”.
7
Acknowledgements
Several other people have contributed to this project. I am grateful to all of
them, listed in no particular order.
Gunnar for the support and encouragement, and for getting me hooked on
science. You were always there to answer questions (and had the answers).
Gunnar is the most efficient person I know of, next to my father.
Arne for interesting discussions, computers and being able to answer the questions when Gunnar was not around. And not the least for helping out to organize
a good computer environment.
Fredrik Hedman, who is also responsible for bringing me into science, and letting
me run *the* computer.
Stefan Nordlund, for letting me join the department and for being one of the
best “VD” I’ve seen.
My parents, Sven-Erik and Christina for giving me a good start, and for encouraging and recognizing my studies.
Charbel, Max, John, and Matti, the oldest friends I have.
Jens Lagergren and Joe DePierre for critical evaluation of my predissertation
seminar and the selection of papers.
Henrik Nielsen
Alessandra Devoto for pushing me to finish my part of an interesting study and
doing most of the hard work.
Elisabeth Ekner and Helen Rabo for all the work behind the electronic publication.
Ann, Kicki, Anki, Bogos och Eddie for making sure the important things work
in the department.
Christian Wettergren was the one I did my first scientific project with (my
masters thesis).
Donald Knuth, Leslie Lamport, and everyone else who has developed TeX into
what it is today. They are responsible for the look and feel of this page. I’ve
only written the text, TeX did the rest.
Miklos, for an enthusiastic effort to keep moving forward.
8
Joacim Halen, Abelsson & Sussman for teaching me how to *really* write computer programs.
Arne Sandin, Leif Paulsson, Lars Svensson, Hillevi Gavel for doing a good job
in my basic teaching. I wish every teacher could be like them.
Bertil Andersson, for building a very good department.
Bo Lundgren for fixing computers that I couldn’t, and discussing problems with.
Peter Nemeth, Niklas Wretman, Jonas Källström, and Johan Arvidsson for reminding me about what I could do (and what I could earn) if I ever finish school,
and lots of fun moments, mostly involving beers and talking about computers.
Peter Hammarlund for all those computer assignments we did at the KTH, and
all the great parties we’ve been to.
Marie, Guro, Ismael, Paul, IngMarie, Susana, Magnus, Susanne, Annika, Hans,
Thomas, (and Rosie), Jan-Willem, Helena A., Chen-ni and everyone else who
has been in Gunnars group. It’s always been a very fun and social place to be.
Lennart Nilsson, who showed me how to look at structures and for all the
interesting computer discussions.
Erik Lundgren for keeping the computers in good shape during my time at
Novum.
The people at PDC, Johan, Assar, Erik A., Gert och Brita.
BG, for setting up my workstation at the KTH. I guess I still owe him a “semla”.
Lars Arvestad, for all the discussions about algorithms.
The organizers of the workshop on Patterns in Biological Sequences at the Aspen
Center for physics, which provided me with a very good start in my bioinformatic
studies. The results on TopPred Multi partly originated there.
Robert, Olof, Mats, Jakob, Amin, Juma, Jeanette, Anders L., Daniel, Pål and
all the short term guests in Arne’s and Gunnar’s groups.
Zhang for taking over my computer responsibilities.
The Journal club.
The Cake club.
Everyone at the department of Biochemistry.
Helena, my sister.
My grandparents who have given me lots of encouragement through the years
and given me an opportunity to take my mind off science a few weeks every
summer.
Keng-Ling, my everything, the best thing that happened to me.
Anyone that I’ve forgotten to mention is purely unintentional.
Thanks!
9
List of publications
This thesis is based on the following articles, which are included in chapter 6.
They will be referred to in the text by Roman numerals.
I.
E. Wallin and G. von Heijne. Properties of N-terminal tails in G-protein
coupled receptors - a statistical study. Protein Engineer, 8(7):693–698,
1995, Medline, WWW.
II. M. Cserzö, E. Wallin, I. Simon, G. von Heijne, and A. Elofsson. Prediction
of transmembrane α-helices in prokaryotic membrane proteins: the dense
alignment surface method. Protein Engineer, 10(6):673–676, 1997, Medline,
WWW.
III. E. Wallin, T. Tsukihara, S. Yoshikawa, G. von Heijne, and A. Elofsson. Architecture of helix bundle membrane proteins: An analysis of cytochrome c
oxidase from bovine mitochondria. Protein Sci, 6(4):808–815, 1997, Medline, WWW.
IV. K. Seshadri, R. Garemyr, E. Wallin, G. von Heijne, and A. Elofsson. Architecture of β-barrel membrane proteins: Analysis of trimeric porins. Protein
Sci, 7(9):2026–2032, 1998, Medline, WWW.
V. E. Wallin and G. von Heijne. Genome-wide analysis of integral membrane
proteins from eubacterial, archaean, and eukaryotic organisms. Protein Sci,
7(4):1029–1038, 1998, Medline, WWW.
VI. A. Sääf, M. Johansson, E. Wallin, and G. von Heijne. Divergent evolution of membrane protein topology: The Escherichia coli RnfA and RnfE
homologues. Proc Natl Acad Sci USA, 96(15):8540–8544, 1999, Medline,
WWW.
Other publications
The following publications were also produced in the course of my Ph.D. studies,
but are not included in this thesis.
10
• E. Wallin, C. Wettergren, F. Hedman, and G. von Heijne. Fast NeedlemanWunsch scanning of sequence databanks on a massively parallel computer.
CABIOS, 9(1):117–118, 1993, Medline.
• J.F. Tomb, O. White, A.R. Kerlavage, R.A. Clayton, G.G. Sutton, R.D.
Fleischmann, K.A. Ketchum, H.P. Klenk, S. Gill, B.A. Dougherty, K. Nelson,
J. Quackenbush, L. Zhou, E.F. Kirkness, S. Peterson, B. Loftus, D. Richardson, R. Dodson, H.G. Khalak, A. Glodek, K. McKenney, L.M. Fitzegerald,
N. Lee, M.D. Adams, E.K. Hickey, D.E. Berg, J.D. Gocayne, T.R. Utterback,
J.D. Peterson, J.M. Kelley, M.D. Cotton, J.M. Weidman, C. Fuji, C. Bowman,
L. Watthey, E. Wallin, W.S. Hayes, M. Borodovsky, P.D. Karp, H.O. Smith,
C.M. Fraser, and J.C. Venter. The complete genome sequence of the gastric
pathogen Helicobacter pylori. Nature, 388:539–47, 1997, Medline, WWW,
PDF.
• G. Schneider, S. Sjöling, E. Wallin, P. Wrede, E. Glaser, and G. von Heijne. Feature-extraction from endopeptidase cleavage sites in mitochondrial
targeting peptides. Proteins, 30(1):49–60, 1998, Medline, WWW.
• A. Sääf, E. Wallin, and G. von Heijne. Stop-transfer function of pseudorandom amino acid segments during translocation across prokaryotic and eukaryotic membranes. Eur J Biochem, 251(3):821–829, 1998, Medline, WWW.
• A. Devoto, P. Piffanelli, IM. Nilsson, E. Wallin, R. Panstruga, G. von Heijne,
P. Schulze-Lefert. Topology, subcellular localization and sequence diversity
of the Mlo family in plants. J Biol Chem, in press, 1999.
11
Abbreviations
3D three dimension
A adenine
aa amino acid
Ala alanine
Arg arginine
Asn asparagine
C cytosine
DAS dense alignment surface
G guanine
GPCR G-protein coupled receptor
Gln glutamine
Gly glycine
K arginine
Leu leucine
Lys lysine
NH nitrogen-hydrogen amino group
NMR nuclear magnetic resonance
OMPLA name of a gene
ORF open reading frame
Ompa name of a gene
Phe phenylalanine
PhoA name of a gene
Pro proline
R lysine
RnfA name of a gene
RnfE name of a gene
SDS-PAGE sodium dodecyl sulfate- polyacrylamide gel electrophoresis
T thymine
TM transmembrane
Trp trptophan
Tyr tyrosine
WWW world wide web
YrbG name of a gene
Å Ångström
12
Chapter 1
Introduction
The stage for this thesis is set in the membrane of living cells. These membranes
are composed of a lipid bilayer which serves to separate different compartments
of the cell, or in the case of lower organisms, the cell from its environment.
Membranes are impermeable to polar molecules (soluble in water) and ions.
Their function is to localize reactions and keep the cell’s molecular machinery
and supply of nutrients from escaping.
Membranes are built up from lipids which are hydrocarbon chains attached pairwise to a phosphate headgroup. The hydrocarbon chains are hydrophobic (not
soluble in water), and the headgroups charged and polar (water soluble). The
name bilayer comes from the fact that the hydrocarbon chains of two monolayers of lipids are fused together and forms a single sheet with phosphate headgroups on both sides - a bi-layer. This effectively makes the membrane soluble,
but not permeable.
From this description one might get the impression that the membrane is simply
a dull monotonous layer, an inactive component of a cell. On the contrary, it
is full of protein molecules that perform many of the important functions that
the cell needs. Several studies (including [paper V] in this thesis) suggest that
around 25 % of all protein types in a cell are membrane proteins.
1.1
Classes of membrane proteins
The first distinction to make in studying membranes is between integral membrane proteins and membrane associated proteins. The latter consist of proteins
which are only anchored to the surface of the membrane, but where the bulk
of the protein consists of a globular (soluble) domain. This class will not be
discussed further in this thesis.
Integral membrane proteins span the bilayer. The focus here will be on one of
its two subclasses, the helix bundle integral membrane proteins.
13
1.1.1
Helix bundle integral membrane proteins
Helix bundle membrane proteins are the class that has been studied most in
this thesis. These proteins are present in the plasma membrane of prokaryotes
and in all membranes of eukaryotic cells. In the rest of this thesis they will be
referred to simply as membrane proteins.
A helix bundle integral membrane protein is built up from α-helices that span
the lipid bilayer, connected by loop regions that extend into the cyto- or
periplasm. These are normally referred to as membrane spanning regions and
loops. The loops may very well fold into secondary structures, not to be mistaken for coil regions in globular proteins which are sometimes also referred to
as loops.
These proteins are responsible for a wide range of functions, such as photosynthesis, transport of ions and molecules across the membrane, attachment points
for skeletal structures of the cell, and secretion of proteins.
1.1.2
Architectural overview of helix bundle membrane
proteins
As 3D structures of membrane proteins are rare, most membrane proteins can
so far only be described by their topology. The membrane spanning segments
and loops are identified and drawn in a cartoon of a membrane, [paper I,fig 1].
The membrane spanning helices can also be drawn with the horizontal position
of each amino acid corresponding to its rotational angle in the helix (a helical
net representation). This enables identification of residues that are on the same
face of the helix but in different turns.
These topology cartoons are often used to illustrate results of topology predictions and experimental topology mappings.
1.1.3
β barrel integral membrane proteins
In addition to the helix bundle membrane proteins, another class of integral
membrane proteins exists, where the membrane spanning segments consist of
β-strands organized in a barrel.
The so-called porins belong to this class. They consist of β-barrels with 1618 βstrands spanning the membrane, usually connected by short loops. The residues
in the strands alternate between pointing towards the lipid bilayer and the
interior of the protein. This interior forms a pore with a polar environment. The
side chains of the strand residues are thus alternately polar and hydrophobic.
Porins are found in the outer membranes of Gram negative prokaryotes and in
the outer mitochondrial membrane of eukaryotic cells. They are responsible for
diffusional transport of molecules through these membranes.
Structures of a few porins have been successfully determined. In this thesis they
have been analyzed in [paper IV], figure 1.2.
14
Other types of β-barrel integral membrane proteins have been found, for example OmpA (Pautsch and Schulz, 1998) and OMPLA (Snijder et al., 1999),
see figure 1.1. They contain fewer transmembrane spanning strands and do not
form pores in the membrane.
1.2
Experimental studies of helix bundle membrane proteins
Despite the difficulties of obtaining structural information (see section 1.2.1),
membrane proteins have been extensively studied. Largely these studies have
focused on the biological functions of particular membrane protein families.
These functions cover many important areas, such as transport of nutrients, ions
and electrons, protein secretion, receptors for the immune system, hormones and
drugs, and even photosynthesis, which takes place in specialized membranes of
plants.
Fewer groups have been working on the architecture and properties of the membrane proteins themselves. What is required for a protein to form a stable
structure in the membrane? How does it get integrated and folded there?
1.2.1
Structural determination of membrane proteins
Unfortunately there are very few 3D structures determined of membrane proteins. There are several reasons for this. First, membrane proteins are hard to
overexpress. They are often toxic to the organism that overexpresses them and
they often form inclusion bodies - large aggregates of protein - in the cell. The
inclusion bodies have to be denatured in order to solubilize the protein, which
then has to be refolded. Second, membrane proteins are difficult to purify in an
active form, and need to be solubilized in detergents.
The third reason is that they are hard to crystallize, because of their hydrophobic nature and the detergent micelles present around the hydrophobic regions.
Unfortunately studies using NMR is not very promising either. Micelles formed
around the hydrophobic part of the protein increases the size. Since NMR has
a limit on the size of the molecule that can be studied, only small membrane
proteins are possible to study, even if they can be successfully overexpressed
and purified.
Despite the above, there are a handful of structures available. The structure of
these are shown in figure 1.3.
1.2.2
Mutational studies
Many interesting results on the architecture of membrane proteins have been
obtained by mutational studies. So far, such studies have mainly focused on
problems related to membrane protein topology and assembly.
Several detection systems have been used. Antibodies against the loop regions
15
Figure 1.1: Structure of a non pore forming β barrel membrane protein, OmpA.
Image generated with [Halaska et al., in preparation]
16
17
Figure 1.2: Structure of a porin. Top image is a view looking down the membrane of a trimer. This is the native form of the complex. Bottom picture is a
side view of one porin subunit, looking along the plane of the membrane. Image
generated with [Halaska et al., in preparation]
Figure 1.3: Cartoon image of some of the structurally determined helix bundle
membrane proteins. From the top left, cytochrome bc1 complex, cytochrome c
oxidase, photisynthetic reaction center, bacteriorhopdopsin. Note the regions of
irregular secondary structure between the transmembrane helixes and the polar
domains.
18
is one way of determining if a loop has been translocated across the membrane.
Proteolytic enzymes can be used to cut loop regions on the outside of the membrane, leaving the regions protected by the membrane still intact. Such a cut
can be seen as a size-shift when the protein is analyzed by SDS-PAGE.
Another common method is fusion of globular proteins/domains to loop regions.
The location of these intra- or extracellular domains can then be detected by
various biochemical assays.
1.2.3
Topology mapping
Even if it is not possible to determine the exact structure of a membrane protein
without crystals, one can still determine the topology using ordinary experimental studies. Many such mapping studies have been performed, and they are the
basis for the evaluation and development of topology prediction methods.
Mapping a topology can be done in several ways. The most common method
today is based on fusion proteins. A protein domain that can be detected when
it is translocated across the membrane is fused to the (predicted) loop regions,
one at a time. Each of the fusion proteins are expressed in a cell line. The cells
are then analyzed with the assay for detection of the fusion domain.
One such system is the PhoA fusion. PhoA is an enzyme that obtains the
correct structure only if translocated to the periplasm, where the appropriate
disulphide bridges are formed. Active PhoA can be detected by a colorimetric
assay. PhoA fusions were used in [paper VI] to map two related proteins that
turn out to have opposite topology.
19
20
Chapter 2
Prediction of membrane
protein topology and
structure
As noted above, integral membrane proteins have so far only been found in two
fold classes - helix bundle and β-barrel. This chapter will focus on the helix
bundle class.
A single chain of a membrane protein spans at least three different compartments
at the same time, the intracellular and extracellular spaces, and the interior
of the membrane. Not only do the environment in these compartments vary,
but the protein has to be able to insert correctly into them during synthesis.
Furthermore, the constraints imposed by the membrane strongly reduces the
number of possible conformations of a membrane protein.
Because of the strict constraints, there are large similarities between membrane
proteins of different function and size, even if they come from different organisms
or are localized to different types of membranes. These similarities are strong
enough to be used for prediction of important structural features from the amino
acid sequence alone.
2.1
Predictable features
The most obvious feature of a membrane protein that can be predicted is the
number of membrane spanning segments and their (approximate) location in
the sequence. Since the interior of the lipid bilayer is highly hydrophobic, the
membrane spanning segments have to have hydrophobic side chains interfacing
the lipids. As with globular proteins, the packed interior of the helix bundle
is also hydrophobic, and thus the entire length of each membrane spanning
segment is hydrophobic.
Several other features are known to be common to all membrane proteins and
21
can be used for prediction. The positive inside rule (von Heijne, 1992), see
sections 3.1.1 and 2.2.1, is the second most useful feature. Unfortunately the
bias is not in general strong enough to determine the location of individual loops.
Instead the prediction has to be based on the average of all loops on each side
of the membrane. Prediction of the orientation of a membrane protein is thus
generally dependent on the correct prediction of the number of transmembrane
segments, but the two predictions can be combined to give an even better result
as described in section 2.2.1.
Other predictable features involve structural information and have been discovered primarily by studying the few structures available. As described in sections
3.2.2 and 3.2.5, the membrane spanning helices, especially the ones facing the
lipid environment, differ in hydrophobicity and residue variability between the
side buried in the helix bundle and the side facing the lipids. This can be used
to predict the lipid facing side of helices from the sequence.
2.1.1
Hydrophobicity scales
Amino acids can be classified in several ways according to their chemical and
physical properties. Their preference to be in contact with polar or non-polar
environments is referred to as their hydrophobicity (hydro-phobia - fear of water). A high hydrophobicity value indicates a preference to be in a non-polar
environment, such as the membrane. Such scales can be derived in several
ways. Directly measuring the chemical preferences (Karplus, 1997; Nozaki and
Tanford, 1971; Damodaran and Song, 1986; Fauchere and Pliska, 1983; Radzicka and Wolfenden, 1988) is one way. Analysis of known protein structures
for the preference of each amino acid to be on the surface (exposed to polar
environment) or in the core is another.
In [paper III], known structures of membrane proteins were analyzed to determine the preferences for each amino acid to be in the membrane or the loop
regions. The correlation to hydrophobicity scales derived with different methods is good, but with some notable exceptions: Met and Val are more preferred
to be in the membrane than would be expected from their hydrophobicity as
measured by other scales.
2.2
2.2.1
Overview methods for topology prediction
TopPred
TopPred (von Heijne, 1992) uses hydrophobicity analysis to identify potential
transmembrane helices, by generating a hydrophobicity profile. The profile is
smoothed by a hat shaped window and the highest peaks are identified as transmembrane segments.
To be able to detect or rule out helices that have borderline hydrophobicity,
peaks with a hydrophobicity value in a certain range are considered putative
transmembrane helices. The region has been calibrated by selecting the least
22
Residue
Phe
Ile
Leu
Ala
Trp
Met
Val
Thr
Gly
Ser
Tyr
Cys
His
Pro
Asn
Asp
Arg
Lys
Glu
Gln
Structural
preference
1.30
0.86
0.86
0.56
0.52
0.44
0.33
0.18
0.16
0.16
-0.39
-0.41
-0.70
-1.30
-1.71
-2.05
-2.10
-2.69
-2.78
-3.00
EngelmanSteitz scale
3.7
3.1
2.8
1.6
1.2
3.4
2.6
1.2
1.0
0.6
-0.7
2.0
-3.0
-0.2
-4.8
-9.2
-12.3
-8.8
-8.2
-4.1
KyteDoolittle scale
2,8
4,5
3,8
1,8
-0,9
1,9
4,2
-0,7
-0,4
-0,8
-1,3
2,5
-3,2
-1,6
-3,5
-3,5
-4,5
-3,9
-3,5
-3,5
Table 2.1: Preference scale for an amino acid to be in the membrane as opposed
to being in the polar regions. Comparison between the scale obtained in [paper III] and two traditional hydrophobicity scales; Engelman-Steitz (Engelman
et al., 1986), and Kyte-Doolittle (Kyte and Doolittle, 1982)
23
hydrophobic segment from a collection of membrane proteins as the lower bound
(in order not to miss any real transmembrane segments) and the most hydrophobic segment from a collection of globular proteins as the upper bound.
The positive inside rule is then used both to determine the most probable of
all possible topologies, with and without each putative transmembrane helix,
by choosing the topology that has the highest bias of positive charges. The
orientation of the protein is also identified at the same time.
A WWW server and source code for TopPred is available at
http://www.biokemi.su.se/˜server/toppred2.
2.2.2
TopPred Multi
Several problems can be identified when running TopPred on single sequences.
Charged residues sometimes occur in the hydrophobic transmembrane helices.
With the result is that the segment is not properly detected by the hydrophobicity profile. Long loops are another problem. They make the orientation
prediction harder since they “dilute” the effect of the positive inside rule. This
affects both orientation prediction and the selection of which putative transmembrane segments to include through the positive inside rule. Short loops
are also a problem. If they do not have charged or strongly polar residues,
the hydrophobicity analysis will identify the two surrounding transmembrane
segments as one.
Unfortunately there is not enough experimental data available to incorporate
changes based on the observations above without the risk of overfitting the
model.
There is one problem that is possible to address however. The identification
of transmembrane spanning segments through the hydrophobicity profile is a
critical step. One way improve this is to apply TopPred to an aligned family of
membrane proteins. The hydrophobicity and charge at each residue is averaged
and the resulting profiles are then used in the same way that a single sequence
would be predicted.
The advantage is that when a protein that belongs to a family has an extreme hydrophobicity, either in a transmembrane segment or a loop, the other
members of the family may have a more normal value for the corresponding
region. As discussed above in section 2.2.1 there is an overlap between the most
hydrophobic segment in globular proteins and the least hydrophobic transmembrane segment. By averaging several sequences we can avoid the extremes of
these distributions and increase the chance of being able to separate them correctly.
Several issues have to be addressed when using families instead of a single sequence. Bad alignments and areas of low similarity may cause the prediction to
fail on examples where the single sequence prediction is correct. Examples have
also been found where homology searching pick up proteins with more or fewer
membrane spanning segments than the target for the prediction.
Predictions on families using TopPred Multi improve the results slightly [un24
published results]. However, the increase in the absolute number of correct
predictions is small, and the statistical significance is questionable.
An unexpected problem was revealed when evaluating TopPred Multi. RnfA, a
membrane protein which has an experimentally determined topology and orientation showed homology to RnfE, a membrane protein that TopPred (on a single
sequence) predicted to have the reversed orientation. This was discovered since
TopPred Multi predicted the family containing the RnfA and RnfE proteins to
have the wrong orientation compared to the experimental results from RnfA.
In [paper VI] this divergent evolution of orientation for the two proteins was
experimentally verified. Fortunately, a sequence analysis of the E. coli genome
did not reveal any other examples. However, it was recently discovered that
YrbG, an inner membrane protein with 10 predicted transmembrane segments is
likely to be the result of a gene duplication, where the two halves of the protein
have different orientations in the membrane and thus must have undergone
divergent topology evolution.
In summary, multiple sequence alignments can improve the prediction slightly,
but also introduce new difficulties.
2.2.3
PHDhtm
The PHDhtm (Rost et al., 1996) method uses multiple sequence alignments
to do a consensus prediction of a target protein. A neural network is used to
compute the preferences for each residue to be in a transmembrane segment or
in a loop. This preference profile is then used to compute the most probable
transmembrane segments using dynamic programming.
In the last step PHDhtm uses the positive inside rule to compute the most
probable orientation.
A PHDhtm server is available at
http://dodo.cpmc.columbia.edu/predictprotein.
2.2.4
TMAP
The TMAP (Persson and Argos, 1994) method uses multiple sequence alignments to produce a preference profile for the consensus sequence. Preference
scales, derived from statistical analysis of database annotations of membrane
proteins, are used to calculate preferences for a residue to be in the hydrophobic core region of membrane spanning segments and the two cap regions.
This profile is used to locate transmembrane segments, through a procedure
that include steps to avoid the problem of single conserved charged or polar
residues. It also includes a method for splitting long hydrophobic regions into
pairs of transmembrane helices. Failure to split long hydrophobic segments is a
common problem in other methods.
TMAP is available as a WWW-service at http://www.mbb.ki.se/tmap.
25
2.2.5
MEMSAT
MEMSAT (Jones et al., 1994) is similar to the preference profile method used
in TMAP. It is more detailed in that it has separate preference scales for intraand extracellular loops and helix ends, in addition to the central helix region.
It also uses different preferences for single and multispanning proteins.
The sequence of the protein of interest is matched, using a dynamic programming method, to a model with the highest preferences in each region.
A PC version of MEMSAT is available from
http://globin.bio.warwick.ac.uk/˜jones/memsat.html
2.2.6
PRED-TMR
PRED-TMR (Pasquier et al., 1999) is a mixture of hydrophobicity analysis
and a weight-matrix method. Transmembrane segments are identified using
a hydrophobicity profile, but they are further discriminated by a propensity
matrix for the helix ends.
2.2.7
TMHMM
This method is based on a Hidden Markov Model. This is a common tool for
modeling sequence data. It basically consists of matching a string against a set
of states. Each state has probabilities for matching different letters. The states
are connected with different transfer probabilities. Dynamic programming is
commonly used to match a sequence against the model in order to find the
most probable match.
In TMHMM (Sonnhammer et al., 1998) a Hidden Markov Model is used to
model the different regions of a membrane protein. The model is more detailed than the previously described approaches. Separate states with preference
probabilities are defined for the helix core, helix end caps on either side of the
membrane, and loop regions as described in 3.1.7. Globular regions on the cytoplasmic and extracellular side are defined differently, and on the extracellular
side, long and short coil regions are encoded by different states.
The parameters in TMHMM are fixed, and come from training on a set of
proteins with known topology. Another larger set was used to test the method
after training.
TMHMM is available as a WWW-service at
http://www.cbs.dtu.dk/services/TMHMM-1.0
2.2.8
HMMTOP
As the name suggests HMMTOP (Tusnady and Simon, 1998) is another method
based on a hidden Markov model. One difference to TMHMM is that only one
26
loop state is used on the cytoplasmic side. A different approach is used for
training the HMM. HMMTOP trains the model on the protein being predicted.
The idea is that it should fine tune the difference in composition between the
various regions for each protein or protein family. The fine tuning/training
procedure can be improved by including proteins homologous to the target of
the prediction.
2.2.9
The dense alignment surface method
DAS [paper II] is a method based on computing a dot plot between the query
protein and one or several reference protein(s). The reference proteins should
be known to be membrane proteins, but do not have to have known topology.
A special comparison matrix is used in computing the dot plot. This matrix
gives higher values when matching hydrophobic than polar pairs of residues. An
average is then computed of the rows/columns in the dot plot. This returns a
hydrophobicity profile that turns out to be somewhat better than those obtained
with other methods.
A WWW server for DAS is available at
http://www.biokemi.su.se/ server/DAS.
2.3
Prediction of membrane protein structure
No definite method for prediction of the structure or packing of transmembrane proteins exist today. The main obstacle is the difficulty to test different
approaches, since so few experimentally determined structures are available.
Several studies have pointed out the existence of useful predictors. The most
important ones are the hydrophobic and variability moments, observed by several studies of both sequences and structures (Rees et al., 1989a; Donnelly et al.,
1993; Eisenberg et al., 1984), [paper III]. In [paper III] data is presented that
could be used to predict the interfaces between subunits in a complex. Similar
results were also found in [paper IV], where porin structures were studied.
Approaches to use these prediction parameters for the packing of helices have
been made. In addition to traditional structure prediction methods, they use
the hydrophobic- and variability moments to predict the lipid exposed residues
of helices (Taylor et al., 1994; Alkorta and Du, 1994; Donnelly et al., 1993).
27
28
Chapter 3
Architecture of helix bundle
membrane proteins
The overall architecture of a membrane protein consists of a number of helical
regions spanning the membrane and loops that connect these helices on either
side of the membrane. Loops have secondary structure elements [paper III, fig
1 and 2] in similar proportion as in globular proteins.
A membrane protein can be viewed as two globular domains anchored together
by a helix bundle. A major difference compared to a globular proteins is that
the segments that build up the two globular domains are spread out in the
sequence to a much greater extent than is normally found in the domains of a
globular protein.
The folding process of a membrane protein is of course also very different from
globular proteins, since the transmembrane helices first have to be inserted into
the membrane before the final structure can form. This puts extra constraints
on the structure and possible configurations. Several features of the polypeptide chain, e.g. the positive inside rule, appear to be present mainly to guide
insertion, and not to support the stability of the protein.
Properties, common to all membrane proteins, can in many cases be detected by
statistical analysis, even if they are not strong enough to be of use for prediction
of structure or topology. Many of the results in this chapter are of this kind.
They are still valuable since they provide a basis for further, experimental and
theoretical studies.
29
3.1
3.1.1
Intra- and extracellular loops
Arg and Lys content is biased towards the intracellular side
The most well known property of loops in membrane proteins is the positive
inside rule (von Heijne, 1992). Short loops are enriched in K and R residues on
the inside and depleted on the outside. This has been studied in more detail
in [paper I] and [paper V].
It was found that, while present in all organisms studied [paper V], the difference
in K+R content is not equally strong in all of them. Eukaryotic cells have a
normal K+R frequency of 10 % in extracellular loops, i.e. the same as for
soluble proteins (von Heijne and Gavel, 1988), but about a twofold increase on
the inside [paper I]. In prokaryotes on the other hand, the difference results
from both depletion on the outside (5 %) and enrichment on the inside (20 %)
(von Heijne and Gavel, 1988).
The reason for enrichment in positive charges on the intracellular side is believed
to be that translocation needs to be arrested so that the inside loop is not pulled
through the membrane. Experimental studies have been performed that support
this view. (Kuroiwa et al., 1991) The fact that all organisms seem to have this
charge bias suggests that the same mechanism is involved.
In addition the K and R content may of course be affected by other factors. One
such example is the C. acetobutylicym, where only Lys residues are enriched
[paper V,fig 4]. The reason for this is not known. It could be a result of
environmental conditions for this organism or a bias in GC content since Lys is
coded by AAG and AAA (no GC) and Arg by CGx, AGA and AGG (all contain
G). Nucleotide frequencies in the C. acetobutylicym are 34 % A, 35 % T, 16 %
C, 15 % G, compared to E. coli which has an even distribution nucleotides1 .
The difference between organisms seen in [paper V] could partly be due to random fluctuations, and differences in the environment for the organisms. Lipid
composition could also be one factor. Experiments show that translocation
of loops with positive charges are affected by the amount of lipids with negatively charged headgroups (van Klompenburg et al., 1997; van Klompenburg
and de Kruijff, 1998).
Membrane potential has been shown to affect the translocation of positively
charged loops and also the orientation of membrane proteins (Andersson and
von Heijne, 1994; Schuenemann et al., 1999). Interestingly in (van de Vossenberg
et al., 1998) it was found that in organisms with reversed membrane potential
the positive inside rule is still valid.
As seen in [paper I, fig 3], shorter loops on the intracellular side have a tendency
to be more enriched in K+R than longer loops. This suggests that it is not so
much the frequency, but rather the number of charges that produces the effect of
arresting translocation. Short loops should also be more at risk of being pulled
through along with their neighboring helices and thus be in need of a stronger
1 Nucleotide statistics for C. acetobutylicum (AE001438): 65548 A, 30229 C, 29117 G,
67106 T, and E. coli (U00096): 1142136 A, 1179433 C, 1176775 G, 1140877 T.
30
anchoring to the intracellular side.
3.1.2
Amino acid bias in the N-terminal and the first loop
Earlier studies (Hartmann et al., 1989; Sipos and von Heijne, 1993) have reported that N-terminal tails have biased distributions of both positive and negative charges [paper I, fig 4]. These biases are anticorrelated and thus give a net
charge bias over the first transmembrane segment. This is strongly pronounced
in eukaryotic membrane proteins, but can not be observed with any statistical
significance in prokaryotes. There is no obvious bias of negative charges in any
of the other loops.
An overall bias of negative charges can be seen in some genomes when looking
at whole genome statistics. It is statistically significant only in C. elegans, H.
sapiens, and C. acetobutylicum. [paper V] This could of course be the result
from only a bias in the first segment.
3.1.3
Other amino acid biases
Several studies also indicate that aromatic residues are enriched in extracellular
loops. In [paper I] this is found in the G-protein coupled receptor family, and
in (Nakashima and Nishikawa, 1992) in several eukaryotic proteins. This does
however not seem to be a general property if one looks at the entire membrane
protein content of genomes [paper V].
Such an enrichment could be specific to certain families of membrane proteins
and be a result of the function of these families rather than a property common
to all membrane proteins.
3.1.4
Charge distribution within the loops
Charges are not only different between loops on either side of the membrane.
In the GPCR family they are also unevenly distributed along the length of the
intracellular loops. [paper I] The first charged residue in the loop, counting from
the transmembrane segment, is often a positive charge. This was verified to be a
statistically significant deviation by comparing the observed number of positive
and negative charges to what would be expected from the overall frequencies
in the intracellular loops. It has not been studied if this is the case in other
membrane protein families or organisms, but experimental results on proteins
with mutated loops support this (Andersson and von Heijne, 1991).
One reason for having a positive charge close to the membrane spanning segment
would be to arrest translocations in order not to pull any part of the loop through
the membrane. It has been suggested that positive charges in intracellular loops
interact with negative charges on the headgroups of the lipids.
The need for positive charges to arrest translocation could possibly be dependent
on the hydrophobicity of the transmembrane segment. The rationale would be
that a strongly hydrophobic segment would stick in the membrane by itself.
31
A less hydrophobic segment (for example one embedded in the helix bundle,
section 3.2.2) would need a positive charge close to its end to stop it from being
pulled out into the extracellular domain. Such a correlation has been shown in
(Kuroiwa et al., 1991).
3.1.5
Loop length and translocation
In [paper I] a comparison between proteins from the same family (G-protein
coupled receptors) with and without signal sequences was performed. The average length of the N-terminal tail was found to be strongly dependent on the
presence or absence of a signal peptide. It was also shown that the K+R content
in the N-terminal tail was lower for proteins without a signal peptide.
This indicates that in order to be translocated across the membrane a long Ntail needs the help of a signal peptide. If it is short enough and contain few
positive charges it can be translocated on its own. There is a “gray zone” where
proteins with and without signal peptide has similar N-tail length [paper I, table
I]. A closer look at these proteins revealed that short N-tails with many positive
charges need a signal peptide. Similarly, long loops can be translocated without
a signal peptide if they contain many negative charges.
Recent studies on genome wide data combined with experiments [Sääf et al.
in progress] suggest that there is a bias in loop length between the intra- and
extracellular side. It appears that there is a strong correlation between the
side with most positive charges and the longest average loop length.2 A similar
selection and prediction to the one used in [paper V] was used. This seems to
hold for all 12 organisms studied so far. See figure 3.1.
3.1.6
Loop length in general
Another finding is that the relative number of residues in helices and loops seem
to be quite constant in the large membrane protein families. When plotting the
protein length against the number of transmembrane segments for all membrane
proteins in an organism [paper V, fig 3], the bulk of the proteins fall around
an axis with 35 residues per transmembrane segment. Since transmembrane
segments are on the order of 20-25 amino acids long, this makes the average
loop 10-15 amino acids long.
One explanation for this is that there may be strong restraints on the length of
the loops. A loop has to be long enough to span the distance between the helices
it is connecting. A short loop between two helices would have to be compensated
by a long loop elsewhere unless all helices in the bundle are packed next to their
sequential neighbor. Not only does the loop need to be long enough to connect
the helices, when they are packed but also during insertion into the membrane.
An upper bound, especially for the extracellular loops, would be that secondary
structure may start to form before the loop has a chance to translocate to the
2 The
side with most positive charges is very likely to be the intracellular side, but the
investigation is done on proteins from whole genomes and thus no experimentally determined
orientations.
32
Organism
A. fulgidus
B. subtilis
B. burgdorferi
B. taurus mitochondrion
C. elegans
C. acetobutylicum
E. coli
H. influenzae
H. pylori
H. sapiens
M. thermoautotrophicum.
M. jannaschii
M. genitalium
M. pneumoniae
N. tabacum chloroplast
S. cerevisiae
Synechocystis Sp.
Correlated
61
95
14
3
165
62
150
44
28
37
26
26
9
4
4
16
36
Total
61
97
14
3
169
63
152
44
28
43
26
29
9
4
4
18
40
Fraction correlated
1
0.98
1
1
0.98
0.98
0.99
1
1
0.86
1
0.90
1
1
1
0.89
0.90
Table 3.1: Number of proteins where the average loop length in the predicted
topology is longer on the intracellular side (actually the side with more positive
charges). The same prediction and selection scheme as in [paper Vtable 2, First
selection] was used.
outside, thus hindering translocation.
3.1.7
A coil region near the membrane?
In cytochrome c oxidase there is a distinct region between the end of the helices
and the globular part of the loops, 20-30 Å from the center of the membrane.
[paper IIIfig 1] In this region there is very little secondary structure. Further
out from the membrane, about 30 Å from the center, helices and β-strands are
found in the same proportions as in globular proteins. This was not examined
systematically in other structures, but the same phenomena can be seen by
visual inspection of the photosynthetic reaction center and the cytochrome bc1
complex, figure 1.3.
When seen from a projection in the plane of the membrane, these unstructured
loop regions are quite striking. At least in cytochrome c oxidase they constitute
a very distinct border between the three domains. If this is a general feature
of membrane proteins, it would be worth examining further. It should help
structure prediction if it is the case. This remains to be seen when more large
membrane protein complexes are crystallized.
33
3.2
Membrane spanning helices
In the previous section, results were mostly derived from statistical analysis
of sequences. In the studies of this thesis, the results on transmembrane helices come from examining the structure of cytochrome c oxidase, the largest
membrane protein complex structurally determined so far [paper III].
There are of course properties of helices that can be derived from sequence
analysis alone, and a lot of work has already been done in this area.
Currently only a few membrane proteins have been structurally determined.
This limits the amount of analysis that can be performed based on structure.
However, it seems that most membrane proteins are not that different in terms
of general architecture. (Bowie, 1997) Analysis of the few structures available
should therefore give information relevant to the whole class of membrane proteins.
3.2.1
Length of transmembrane helices
The membrane spanning segments in cytochrome c oxidase [paper III] span the
membrane in an entirely helical conformation. In only a few cases do they
extend into the globular domain (the loop regions) of the protein. In the ±12
Å central region there is almost only helix conformation. This is also the most
hydrophobic region of the transmembrane segments. The helices end 15-25 Å
from the center of the membrane.
The length of the hydrophobic stretch is also possible to examine from sequence
analysis. It has therefore been known for quite a long time that the membrane
spanning regions are on the order of 18-25 residues long. This does however not
imply that the helices end at the membrane border. Such evidence can only be
found from structure studies as [paper III] and (Yeates et al., 1987; Rees et al.,
1989b; Bowie, 1997).
Helices do not seem to extend into the globular domains. One reason for this
could be that it should be possible for the helices in the membrane to form
independently of secondary structures in the loop regions. It could also be
important that when forming the helix bundle the helices are able to shift around
during the integration into the membrane. This would be hindered if parts of
the helices were sticking out of the membrane into the polar domains, where
secondary structures have already packed into a rigid structure.
Another hypothesis why most of the helices end at the membrane border is that
the negatively charged headgroups of the lipids interact with the N-terminal
end of the helices. Because of the helix dipole moment the N-terminus has a
positive charge of about 0.5-0.7 units, and also contain NH groups that could
assist in binding a phosphate group from the lipids. Phosphate binding to the
N-terminal end of helixes is suggested in (Branden and Tooze, 1999, fig 2.3,
page 16).
34
3.2.2
Hydrophobicity and exposure to lipids
The distribution of hydrophobic residues is one of the most striking features of
membrane proteins. It can be seen clearly from a simple sequence analysis that
the transmembrane segments have hydrophobic sidechains. The same pattern
with a central ±10 − 12 Å hydrophobic region can be seen [paper III, fig 8] in
all the structures studied there.
By classifying residues as buried in the helix bundle, intermediately exposed,
or exposed to the lipid environment, one can get a more detailed picture. The
exposed class has higher average hydrophobicity than the buried class, when
looking at the central ±10 Å region. [paper III, fig 9]. A similar analysis was
done earlier in (Rees et al., 1989a) by using Fourier methods to analyze the
transmembrane helices in the photosynthetic reaction center.
This is contrary to what one sees in soluble proteins, where the core is always
more hydrophobic than the protein surface. Unfortunately the difference in
membrane proteins is not very strong. It can not be used efficiently in structure
prediction by itself.
Using the same classification it seems from the analysis of cytochrome c oxidase
that not all hydrophobic sidechains have the same preference for being in contact
with the lipids. Leu and Ala are more common in exposed positions, whereas
Phe, Trp, and Tyr are more common in buried positions.
3.2.3
Distribution of other residue types
In addition to the central hydrophobic region, two flanking regions of the helices
around ±12 − 15 Å are enriched in polar aromatic residues (Trp and Tyr) and
amidated residues (Asn and Gln). Further out, around ±15 − 20 Å charged
residues appear. Finally there is an increased concentration of Pro residues
where the helices end around ±15−25 Å away from the membrane center [paper
III, fig 4]. This data is again from cytochrome c oxidase. Similar results were
also obtained for another large structure, the photosynthetic reaction center, in
(Schiffer et al., 1992).
The “aromatic belt” has also been observed before in porin structures (Weiss
et al., 1991) and in sequence studies of helix bundle membrane proteins (von
Heijne, 1994; Pawagi et al., 1994). It is believed to stabilize the helices in
the membrane by interacting with the lipid headgroups. Recently experimental
studies have also been performed on this subject (de Planque et al., 1999; Braun
and von Heijne, 1999).
3.2.4
Helix breakers and charged residues
Helix breakers (Gly and Pro) and charged residues are rare in the membrane,
most likely because of the energetic disadvantage of breaking the helix or introducing a charged residue in the hydrophobic lipid environment. When they are
present they are almost always facing the interior of the protein. [paper III]
35
Charged residues are a problem when predicting topology from sequence alone.
(Pro and Gly are hydrophobic so it is not a problem from a prediction point.)
At least one method has tried to overcome this. In TMAP (section 2.2.4) single
occurrences of charged residues are allowed in a transmembrane segment if they
are conserved throughout the protein family being predicted.
3.2.5
Correlation between residue variability and exposure
By aligning several sequences to a known structure a measure of the sequence
variability in each position can be calculated. This was done in [paper III]
with cytochrome c oxidase, and the information content (Schneider et al., 1986)
was calculated and used as a measure. When looking at the central ±10 Å of
the membrane domain it is clear that exposed residues display a higher degree
of sequence variation (lower information content), whereas buried residues are
more conserved. [paper III, fig 6] This has also been observed in studies of the
photosynthetic reaction center and bacteriorhodopsin (Rees et al., 1989a).
Cytochrome c oxidase is a large complex made up of a core formed by three
mitochondrially encoded chains. In mammals, several small nuclear encoded
subunits are attached to the core complex. This buries residues on the surface
of the core subunits. These interface residues between small subunits and the
core complex turn out to be more conserved than the lipid exposed residues.
This is no surprise since they are involved in a specific interaction just as the
core residues. Variability could thus be useful as a prediction tool for identifying
putative protein-protein interface regions.
The variability can also be calculated from sequence data. It is quite certain
that transmembrane segments, which can be roughly predicted from sequence,
are in helix conformation. One can therefore assign buried and exposed helix
faces from variability calculations. This was done in (Rees et al., 1989a) in
addition to the structural study, and it has been done in purely sequence based
studies (Donnelly et al., 1993)
3.2.6
Helix ends
Helix ends in membrane proteins have the same characteristics as in globular
proteins. This is no surprise since they end outside the membrane environment.
Pro residues are found on the N-terminal side and in the first two N-terminal
positions of the helix (MacArthur and Thornton, 1991). Asp, Glu, Asn and Ser
are often found in the first non-helix position on the N-terminal side, which is
typical also for globular proteins (Richardson and Richardson, 1988; Doig and
Baldwin, 1995). On the C-terminal side, Gly is enriched, also this a common
preference in globular proteins (Richardson and Richardson, 1988; Aurora et al.,
1994). [paper III]
This result is promising since it means that similar methods as those used to
predict helix ends in globular proteins can be used. This should be useful
both for structure prediction, but also for topology prediction where a better
36
identification of helix ends would definitely help to distinguish between single
long helices and closely spaced pairs of helices.
37
38
Chapter 4
Genome wide studies of
membrane proteins
Genome data open up an interesting new field to study the distribution of
membrane proteins in different genomes. Thanks to the sequencing efforts of
the genome projects, a large number of microbial genomes are available. The
sequencing of a few higher organisms have been finished as well.
However, when looking statistically at proteins one has to keep in mind that each
individual cell or organism only uses a finite number of proteins. For example
M. pneumoniae has only 300 ORFs and a predicted number of 50-80 membrane
proteins. This suggests that we should not go into too much statistical detail,
simply because there might not be enough data points to draw more than rough
conclusions. An analysis that requires a hundred membrane proteins to be
statistically significant will not be relevant if it is applied to a small genome as
M. pneumoniae.
4.1
Membrane protein content
A natural question to ask is how many different membrane proteins there are
in a cell. Does this vary with the type of organism?
These questions have been addressed in several studies, [paper V], (Jones, 1998;
Boyd et al., 1998). The approach has been to use an algorithm to classify each
protein in a genome as membrane or non-membrane. In [paper V] this was
done using the method described in section 2.2.1. Membrane protein prediction
is not 100 % accurate, especially not for proteins with very few transmembrane
segments. In an attempt to put upper and lower bounds on the numbers, the
assignment was done using three different parameter sets.
The result of the analysis is plotted in [paper V, fig 1]. It indicates that organisms with smaller genomes require proportionally fewer membrane proteins.
This proportionality was, however, not apparent in two other studies, which
were done around the same time (Jones, 1998; Boyd et al., 1998). The method
39
used to identify membrane proteins, TopPred, is optimized for predicting the
topology of membrane proteins, but not for discriminating between membrane
proteins and soluble proteins. This could have resulted in incorrect assignments.
This is especially a problem when looking at membrane proteins with few predicted membrane spanning segments, as these could easily have been picked up
from hydrophobic segments from the core of a globular protein.
A more recent study [Krogh et al, personal communication] however indicates
that there is a difference, not between organisms of different size, but between
single- and multicellular organisms. In particular, in the worm C. elegans Gprotein coupled receptors account for 5 % of the ORFs, thereby increasing the
percentage of membrane proteins.
The reason for this difference is suggested to be that multicellular organisms
require communication and adhesion. They still have the same basic machinery
for the internal working of the cell, and thus do not need to introduce so many
new functions in terms of soluble proteins. It is also possible that higher organisms have a greater need to localize reactions to different compartments of the
cell, and this requires more membrane proteins.
4.2
Number of transmembrane segments
In [paper V] the number of transmembrane segments were predicted using TopPred, section 2.2.1, for each protein in a whole genome. The result was plotted
in a histogram with the number of proteins of each type (2 TM, 3 TM, etc.).
One can clearly discern important families of each genome. For example in lower
organisms, 6 TM and 12 TM families are overrepresented. This corresponds to
proteins that are necessary to transport nutrients and small molecules. In higher
organisms on the other hand, 7 TM proteins (i.e. G-protein coupled receptors)
are highly enriched, which shows the importance of communication to these
cells.
A similar study (Jones, 1998), agrees with these observations, even though a
different prediction method was used. However, in (Arkin et al., 1997) the
families are less visible. In this study a simpler method was used for prediction.
Sharp peaks could therefore have been smoothed out by errors in the prediction.
4.3
Size vs number of transmembrane segments
Another view of the data from section 4.2 is to build a height profile of the proteins with the height corresponding to the fraction of proteins with a particular
number of transmembrane segments and size. A contour plot of this landscape
shows where the important families are and also reveals another interesting
feature: Most membrane proteins have the same number of residues per transmembrane segment. The peaks representing protein families are scattered along
a line with about 35 residues per transmembrane segment. In addition there
is a large collection of proteins with only a few predicted transmembrane segments. This is believed to be mostly false positive predictions, but also globular
40
proteins that are anchored to the membrane with one or two transmembrane
segments.
Recent studies [Larsson, et al, unpublished results] show that TopPred does
have a higher rate of false positives than TMHMM. This would explain the
large number of membrane proteins with few transmembrane segments and the
disagreement with other studies of the number of membrane proteins per organism described in section 4.1.
41
42
Chapter 5
Summary
In this thesis, membrane proteins have been studied by statistical and computational methods based on data from a collection of G-protein coupled receptors,
complete genome sequences of several organisms, and structures of several multispanning integral membrane proteins.
The positive inside rule was verified to hold in most of the studied organisms, but
with differences in the degree of depletion of positive charges on the extracellular
side between eukaryotes and prokaryotes. Short intracellular loops seem to be
more enriched in positive charges than long ones.
Statistical studies of G-protein coupled receptors with and without signal peptides suggested that long N-tails require a signal peptide to be translocated
across the membrane. N-tails without a signal peptide are depleted in positive charges. A bias in negative charges towards the extracellular side was seen
across the first transmembrane segment.
Statistical studies of genome sequences were performed by predicting the topologies of all ORFs. It was found that the relative number of membrane protein
differs between uni- and multicellular organisms, that 12TM proteins are particularly abundant in unicellular organisms and 7TM protein in multi-cellular
organisms, and that most multi-spanning proteins have roughly 35 residues per
transmembrane segment.
A study of 3D structures from three helix bundle membrane proteins showed
that the general architecture of these consist of a helical region extending +15-25 Å from the center of the membrane. The helices are connected to the
globular domains by a coil region, without secondary structure elements, about
+-20-30 Å from the membrane center. It was confirmed that the previously
observed aromatic regions around 12-15 Å from the center of the membrane is
a conserved feature of helix bundle membrane proteins.
An example of divergent evolution of the topology of a membrane protein was
found when evaluating a new method for topology prediction on aligned protein
families. The RnfA protein in E. coli was found to have the opposite orientation
to its homolog RnfE. This was experimentally verified by mapping the topology
of both RnfA and RnfE.
43
44
References
Alkorta, I. and Du, P. (1994). Sequence divergence analysis for the prediction
of seven-helix membrane protein structures: II. A 3-D model of human
rhodopsin. Protein Eng, 7(10):1231–1238, Medline. 2.3
Andersson, H. and von Heijne, G. (1991). A 30-residue-long ”export initiation
domain” adjacent to the signal sequence is critical for protein translocation
across the inner membrane of Escherichia coli. Proc Natl Acad Sci USA,
88(21):9751–9754, Medline. 3.1.4
Andersson, H. and von Heijne, G. (1994). Membrane protein topology: effects of
∆µH+ on the translocation of charged residues explain the ’positive inside’
rule. EMBO J, 13(10):2267–2272, Medline. 1
Arkin, I., Brunger, A., and Engelman, D. (1997). Are there dominant membrane
protein families with a given number of helices? Proteins, 28(4):465–466,
Medline. 4.2
Aurora, R., Srinivasan, R., and Rose, G. (1994). Rules for alpha-helix termination by glycine. Science, 264(5162):1126–1130, Medline. 3.2.6
Bowie, J. (1997). Helix packing in membrane proteins. J Mol Biol, 272(5):780–
789, Medline, WWW. 3.2, 3.2.1
Boyd, D., Schierle, C., and Beckwith, J. (1998). How many membrane proteins
are there? Protein Sci, 7(1):201–205, Medline. 4.1, 4.1
Branden, C. and Tooze, J. (1999). Introduction to Protein Structure. Garland,
second edition, WWW. 3.2.1
Braun, P. and von Heijne, G. (1999). The aromatic residues Trp and Phe
have different effects on the positioning of a transmembrane helix in the
microsomal membrane. Biochemistry, 38(30):9778–9782, Medline. 3.2.3
Damodaran, S. and Song, K. (1986). The role of solvent polarity in the free
energy of transfer of amino acid side chains from water to organic solvents.
J Biol Chem, 261:7220–7222, Medline. 2.1.1
de Planque, M., Kruijtzer, J., Liskamp, R., Marsh, D., Greathouse, D., Koeppe,
R. n., de Kruijff, B., and Killian, J. (1999). Different membrane anchoring positions of tryptophan and lysine in synthetic transmembrane alphahelical peptides. J Biol Chem, 274(30):20839–20846, Medline, WWW. 3.2.3
45
Doig, A. and Baldwin, R. (1995). N- and C-capping preferences for all 20 amino
acids in alpha-helical peptides. Protein Sci, 4(7):1325–1336, Medline. 3.2.6
Donnelly, D., Overington, J., Ruffle, S., Nugent, J., and Blundell, T. (1993).
Modeling alpha-helical transmembrane domains: the calculation and use
of substitution tables for lipid-facing residues. Protein Sci, 2(1):55–70,
Medline. 2.3, 2.3, 3.2.5
Eisenberg, D., Schwarz, E., Komaromy, M., and Wall, R. (1984). Analysis of
membrane and surface protein sequences with the hydrophobic moment
plot. J Mol Biol, 179(1):125–142, Medline. 2.3
Engelman, D., Steitz, T., and Goldman, A. (1986). Identifying nonpolar transbilayer helices in amino acid sequences of membrane proteins. Annu Rev
Biophys Biophys Chem, 15:321–353, Medline. 2.1
Fauchere, J.-L. and Pliska, V. (1983). Hydrophobic parameters pi of amino acid
side chains from the partitioning of N-acetyl-amino-acid amides. Eur J Med
Chem, 18:369–375. 2.1.1
Hartmann, E., Rapoport, T., and Lodish, H. (1989). Predicting the orientation of eukaryotic membrane-spanning proteins. Proc Natl Acad Sci USA,
86(15):5786–5790, Medline. 3.1.2
Jones, D. (1998). Do transmembrane protein superfolds exist?
423(3):281–285, Medline. 4.1, 4.1, 4.2
FEBS Lett,
Jones, D., Taylor, W., and Thornton, J. (1994). A model recognition approach
to the prediction of all-helical membrane protein structure and topology.
Biochemistry, 33(10):3038–3049, Medline. 2.2.5
Karplus, P. (1997). Hydrophobicity regained. Protein Sci, 6(6):1302–1307,
Medline, WWW. 2.1.1
Kuroiwa, T., Sakaguchi, M., Mihara, K., and Omura, T. (1991). Systematic
analysis of stop-transfer sequence for microsomal membrane. J Biol Chem,
266(14):9251–9255, Medline. 3.1.1, 3.1.4
Kyte, J. and Doolittle, R. (1982). A simple method for displaying the hydropathic character of a protein. J Mol Biol, 157(1):105–132, Medline. 2.1
MacArthur, M. and Thornton, J. (1991). Influence of proline residues on protein
conformation. J Mol Biol, 218(2):397–412, Medline. 3.2.6
Nakashima, H. and Nishikawa, K. (1992). The amino acid composition is different between the cytoplasmic and extracellular sides in membrane proteins.
FEBS Letters, 303(2,3):141–146. 3.1.3
Nozaki, Y. and Tanford, C. (1971). The solubility of amino acids and two
glycine peptides in aqueous ethanol and dioxane solutions. J Biol Chem,
246:2211–2217, Medline. 2.1.1
Pasquier, C., Promponas, V., Palaios, G., Hamodrakas, J., and Hamodrakas, S.
(1999). A novel method for predicting transmembrane segments in proteins
based on a statistical analysis of the SwissProt database: the PRED-TMR
algorithm. Protein Eng, 12(5):381–385, Medline. 2.2.6
46
Pautsch, A. and Schulz, G. (1998). Structure of the outer membrane protein A
transmembrane domain. Nat Struct Biol, 5(11):1013–1017, Medline. 1.1.3
Pawagi, A., Wang, J., Silverman, M., Reithmeier, R., and Deber, C. (1994).
Transmembrane aromatic amino acid distribution in P-glycoprotein. A
functional role in broad substrate specificity. J Mol Biol, 235(2):554–564,
Medline. 3.2.3
Persson, B. and Argos, P. (1994). Prediction of transmembrane segments in
proteins utilising multiple sequence alignments. J Mol Biol, 237(2):182–
192, Medline. 2.2.4
Radzicka, A. and Wolfenden, R. (1988). Comparing the polarities of the amino
acids: side-chain distribution coefficients between the vapor phase, cyclohexane, 1-octanol, and neutral aqueous solution. Biochemistry, 27:1664–
1670. 2.1.1
Rees, D., DeAntonio, L., and Eisenberg, D. (1989a). Hydrophobic organization
of membrane proteins. Science, 245(4917):510–513, Medline. 2.3, 3.2.2,
3.2.5, 3.2.5
Rees, D., Komiya, H., Yeates, T., Allen, J., and Feher, G. (1989b). The bacterial
photosynthetic reaction center as a model for membrane proteins. Annu
Rev Biochem, 58:607–633, Medline. 3.2.1
Richardson, J. and Richardson, D. (1988). Amino acid preferences for specific locations at the ends of alpha helices. Science, 240(4859):1648–1652,
Medline. 3.2.6, 3.2.6
Rost, B., Fariselli, P., and Casadio, R. (1996). Topology prediction for helical
transmembrane proteins at 86% accuracy. Protein Sci, 5(8):1704–1718,
Medline. 2.2.3
Schiffer, M., Chang, C., and Stevens, F. (1992). The functions of tryptophan
residues in membrane proteins. Protein Eng, 5(3):213–214, Medline. 3.2.3
Schneider, T., Stormo, G., Gold, L., and Ehrenfeucht, A. (1986). Information
content of binding sites on nucleotide sequences. J Mol Biol, 188(3):415–
431, Medline. 3.2.5
Schuenemann, T., Delgado-Nixon, V., and Dalbey, R. (1999). Direct evidence
that the proton motive force inhibits membrane translocation of positively
charged residues within membrane proteins. J Biol Chem, 274(11):6855–
6864, Medline. 1
Sipos, L. and von Heijne, G. (1993). Predicting the topology of eukaryotic
membrane proteins. Eur J Biochem, 213(3):1333–1340, Medline. 3.1.2
Snijder, H. J., Ubarretxena-Belandia, I., Blaauw, M., Kalk, K. H., Verheij,
H. M., Egmond, M. R., N., D., and W., D. B. (1999). Structural evidence for
dimerization-regulated activation of an integral membrane phospholipase.
Nature, 401:717–721, WWW. 1.1.3
47
Sonnhammer, E., von Heijne, G., and Krogh, A. (1998). A hidden Markov
model for predicting transmembrane helices in protein sequences. Intell.
Syst. Mol.Biol., 6:175–182, PDF. 2.2.7
Taylor, W., Jones, D., and Green, N. (1994). A method for α-helical integral
membrane protein fold prediction. Proteins, 18(3):281–294, Medline. 2.3
Tusnady, G. and Simon, I. (1998). Principles governing amino acid composition
of integral membrane proteins: application to topology prediction. J Mol
Biol, 283(2):489–506, Medline. 2.2.8
van de Vossenberg, J., Albers, S., van der Does, C., Driessen, A., and van
Klompenburg, W. (1998). The positive inside rule is not determined by the
polarity of the delta psi. Mol Microbiol, 29(4):1125–1127, Medline, WWW.
1
van Klompenburg, W. and de Kruijff, B. (1998). The role of anionic lipids in
protein insertion and translocation in bacterial membranes. J Membr Biol,
162(1):1–7, Medline, WWW. 1
van Klompenburg, W., Nilsson, I., von Heijne, G., and de Kruijff, B. (1997). Anionic phospholipids are determinants of membrane protein topology. EMBO
J, 16(14):4261–4266, Medline, WWW. 1
von Heijne, G. (1992). Membrane protein structure prediction. Hydrophobicity
analysis and the positive-inside rule. J Mol Biol, 225(2):487–494, Medline.
2.1, 2.2.1, 3.1.1
von Heijne, G. (1994). Membrane proteins: from sequence to structure. Annu
Rev Biophys Biomol Struct, 23:167–192, Medline. 3.2.3
von Heijne, G. and Gavel, Y. (1988). Topogenic signals in integral membrane
proteins. Eur J Biochem, 174(4):671–678, Medline. 3.1.1, 3.1.1
Weiss, M., Abele, U., Weckesser, J., Welte, W., Schiltz, E., and Schulz, G.
(1991). Molecular architecture and electrostatic properties of a bacterial
porin. Science, 254(5038):1627–1630, Medline. 3.2.3
Yeates, T., Komiya, H., Rees, D., Allen, J., and Feher, G. (1987). Structure of
the reaction center from Rhodobacter sphaeroides R-26: membrane-protein
interactions. Proc Natl Acad Sci USA, 84(18):6438–6442, Medline. 3.2.1
48
Chapter 6
Papers
49
I
This article is unfortunately only available in the printed thesis. Look at the
reference in section and follow the links from there.
II
III
IV
V
VI
This article is unfortunately only available in the printed thesis. Look at the
reference in section and follow the links from there.