Copyright (C) 2006, GPCRDB.
About 8 months ago a membrane protein meeting took place in Leeds, and after that meeting I wrote the following general remarks:
Several things became very clear during this meeting:
Bob Bywater started with a summary of the problems that GPCR- hobbyists are facing. Bob welcomed all guests and summarised the ideas of the GPCR workshop series. He mentioned among other things the problems upon alignments, the choice of the template for modelling 7TMs, the total lack of information about the loops, the potential of functional importance of mobility, the tilts and kinks in helices, the mechanism of signal transduction (charge relay? rotamer flips? charge relay?), Thue Schwartz' idea that there are two forms, an active form, stabilised by agonists, and an inactive form, stabilised by antagonists, the role of internal waters, the lack of input from biology in the modellers thinking.
Wilma Kuipers discussed the observed correlations between
certain residue types at certain positions in the GPCRs, and the affinity of
these GPCRs for ligands. The idea is that pure modelling leads to models
of insufficient quality to start docking studies. Without mutation data it
seemed impossible to get an idea about potential ligand binding sites
(especially for antagonists, were docking by homology is less often
possible, GV). However, if high/low affinity binding of a ligand to a
series of homologous receptors is correlated with the presence/absence
of one particular residue, then this information is almost as powerful as
mutation data. Of course, in the real world things are not always as
simple and straightforward as on paper, but in Wilma's hands in about
half of all cases where sufficient data was available new knowledge
about the ligand-GPCR interaction was obtained. The main problem with
the method (see the appendix of these notes for an abstract of the
contribution to the Leeds'94 conference book that explains how the
method works) seems to be false positives. If a small molecule binds with
high affinity on all serotonin1,2 receptors, but not on any other serotonin
receptors, then a binding versus residues correlation analysis will give
all residues important for binding plus all residues that separate the
serotonin 1 and 2 classes from the other serotonins.
Wilma showed four examples where the correlated mutation
analysis enhanced our understanding of the mode of ligand binding. A
problem that she ran into, however, was that different labs use different
assays, and thus find different values for the same binding constants.
The full details of her work will be published shortly. Two of the
examples were extremely nice. It was impressive to see that the
presence of an asparagine on helix VII is correlated 100% with pindolol
binding (in all amine receptors for which data exists). Another example
clearly indicated the importance of a proline two turns of a helix above
the well-known PFF motif in helix VI.
Gerd Krause gave a short overview of the problems that arise
when one decides to model non-rhodopsin-like GPCRs such as calcitonin,
parathyroid, secretin or metabotropic glutamate receptors. For these
receptors no mutation data at all is available. There is no sequence
homology with bacteriorhodopsin. The only homology with the
rhodopsin-like (Class 1) receptors is the presence of the Cys-Cys bridge
between the top of helix III and the second external loop. These receptors
do not even have the famous DRY motif at the bottom of helix III, neither
do they posses the NPXXY motif in helix VII. All Class 2 receptors
(calcitonin receptor, etc.) have long N-termini, and it seems likely that
the natural agonists bind just there. Gerd decided to base his modelling
on the bacteriorhodopsin framework. He described a threading-like
algorithm for modelling the helices. Rather than looking at the
threading of one helix at a time, he built almost all possible combinations
of differently threaded helices, and used the complementarity of
polarity, hydrophobicity, etc., of the contacting helical faces to reduce
the total number of models to a manageable small set. Strikingly, in all
models many residues located near the top of the transmembrane helices
are small (ser, thr, ala, etc.). This creates space where a part of the
ligand could be docked. The residues near the bottom of this 'cavity' are
very conserved.
Gerd went on to discuss his idea about the ligand interaction. It is
obvious that there must be some interaction with the N-terminus and/or
with the external loops. The rather high degree of conservation in
cysteines, polarity and predicted secondary structure in the large N-
terminal arms strongly suggests this area to be involved in ligand
binding. This whole modelling study is about to be published. He went on
to discuss some of their future work, like mutation studies, NMR studies,
etc. The discussion concentrated on modes of binding. Could it be that
the more conserved part of all the ligands interacts with the more
conserved part of the receptors? If so, is there any cross reactivity
between different class 2 receptors? Are the N-terminal residues of the
ligand responsible for activation of the receptor? If so, why don't we see
any changes when the first three residues are removed?
One of the aims of the 7TM club is to collect, sort and store data,
and to make it available to the public. Karl Aberer is a specialist in the
object oriented database field. Because of the heterogeneous nature of
GPCR related data object oriented databases (OODBs) provide the best
tool for GPCR data storage and retrieval.
Karl started by explaining how an informatics expert could end up
in molecular biology. He gave a clear introduction about object OODBs.
His aim is to combine PDB, Swiss-Prot, PIR, OWL, EMBL, etc., databases
with the Cambridge small molecule database, several mutation
databases, multiple sequence alignments, GPCR models, human added
information about potential docking sites, small molecule binding
constants, literature, etc., in one OODB. The questions that one can get
answered can sound like: "For which receptors are multiple small
molecule binding constants available as well as mutations that alter
binding, whereas in homologous GPCRs mutation data is available for
potentially contacting residues". he discussed some of the major
problems that he encountered so far. One of the major problem is that
GPCR specialists hardly ever agree on models, alignments or even on
real data. Therefore the raw data need to be stored, which leads to high
internal complexity if high query speeds are requested. It also leads to
the problem that for one data item multiple views exist, or in other
words, for one GPCR model there are multiple coordinate sets, or for one
binding constant there are multiple values, all with their own literature
attached. Karl discussed five important points for OODBs:
The Swiss-Prot database is of the larger databases used in the 7TM
club clearly the cleanest. Brigitte Boeckman works on the Swiss-Prot
database and was invited to tell us how the Swiss-Prot people work,
and what they do to maintain high quality standards. Brigitte discussed
first what quality of data actually means. Every data item must have as
its core: the sequence, the source, a reference to the origin of the data
and a unique identifier (the accession code). Beyond this, other data can
be present, but where data is of similar nature, they aim for similarity in
completeness and format too. Even free text like comments by the
annotators should be as rigid and standardised as possible. Additional
quality is provided by cross pointers to other databases. This allows for
automatic error detection. They even make sure that the database is free
from spelling mistakes. In general databases should be correct,
complete, and up to date.
The big problem Swiss-Prot is facing at this moment is
completeness. Brigitte discussed some techniques they use to speed up
the data processing, but since much information still comes from reading
literature by the annotators, dramatic improvements in processing
speed can not be obtained by further automation of the data processing.
Therefore a need exists for so-called external experts. If for example
external specialists would take care of the GPCRs, Swiss-Prot people
would not have to worry about this class of proteins, and could spend
more time on the other molecules. Also, the annotators at Swiss-Prot
normally only read the original article, and can of course not know all
the extra information that a specialist on one molecular class can gather
over the years.
In Germany there is a group of industries and institutes that collaborates under the name RELIWE. Friedrich Rippmann explained that the aims of this group is to design an integrated in computo drug design environment. One of the RELIWE partners is the Lengauer group in Bonn, where novel docking techniques are being developed. Their methods work so amazingly well that Friedrich decided to present first results at this meeting. The ligand is docked in a stepwise procedure. First, ligands are split into fragments (these fragments are allowed to have internal degrees of freedom). A so-called 'base fragment' is defined by the user which should be small, have little internal freedom, and should preferably contain one or more hydrogen bond donor or acceptor groups. The second step is a rather cleverly designed spatial matching algorithm which places the base fragment in a preliminary position in the predefined receptor site. More fragments are added taking care of preferred torsion angles and favourable interactions with the receptor site until the molecule is complete. In this way e.g. methotrexate is docked successfully into dihydrofolate reductase. At present this docking method still treats the receptor as a rigid molecule, but it does not seem very difficult to incorporate limited flexibility. The incorporation of model uncertainty, however, still seems a bit further away.
Paul Sigler thinks a bit differently about the interaction between
the G protein and the receptor. Most 7TM club members ask the question:
"How does the G protein bind my receptor"? Paul solved the Xray
structure of a transducin, and now started the second day of the
workshop by asking the question: "How does the receptor bind my G
protein"? Paul discussed extensively the transducin structure complexed
with guanine nucleotides. This beautiful molecule was recently published
in Nature, so it does not need a further description here.
More importantly, he discussed the difficulties of crystallising
GPCRs, and the differences, from a point of view of crystallisation,
between GPCRs and, for example, the photo reaction centre. These
differences mainly lie in the ratio of lipid embedded over water facing
surface.
Paul also asked the question why the relatively small difference
between GDP and GTP leads to such dramatic effects. The answer to this
question could be found in the crystal structure of the GDP-AlF4
transducin complex. This structure made clear that the role of the third
phosphate group is to attract residues in its environment so much that
large rearrangements take place. Thus, parts the transducin structure
look different in the presence of GDP or GTP. However, the crucial
question how the receptor activates the G protein could not yet be
answered, and therefore Paul has set his mind to crystallising the
rhodopsin-transducin complex.
Lets hope that he is successful real soon.
Bob Bywater tries to design sub-type specific dopamine
antagonists. He told us that very similar receptors such as the dopamine
1-5 receptors can elicit very different responses. This mainly has to do
with the localisation of these receptors. Drugs however can not be
administered in a localised manner, and that creates most of his
problems. He used the correlated mutation method that was described
earlier by Wilma to analyse the differences between the dopamine
receptor sub-type sequences. This showed several residues for which the
importance in ligand binding was already known, but also revealed
some new insights. Since he wanted to map these residues on a three
dimensional model, he built models as reliable as he thought possible at
this moment.
Bob, like many other, started mapping the sequence on the
bacteriorhodopsin framework using the alignments from the TM7
server. However, he measured the helix-helix distances from the
published low resolution opsin model, and used these distances as
restraints in a otherwise straightforward vacuum molecular dynamics
run. His models will be made available via the TM7 server soon. Using
these models he found an alternative way to dock dopamine (and other,
similar ligands) in the receptors. In his docking the two famous serines
on helix V are less important, and a more prominent role is put aside for
a backbone C=O group that is one helical turn away from a proline in
helix VI. The position of this proline is not conserved among the
dopamine sub-types, and thus, this C=O could very well explain some of
the observed differences in antagonist binding that he observed.
Helmut Reilaender discussed all the problems that were encountered over the last years in his attempts to clone, overexpress and purify large amounts of GPCR protein for crystallisation purposes. Looking at the complete world wide lack of expression systems that could give milligram quantities of material, I think we can safely assume that his problems are everybodies problems. He started by listing the topics to be thought about:
Mark Scheidler started with a more realistic look at the
membrane, and what one can all find in there. After all, the membrane is
the only thing our seven helical friends ever see in their active life. We
need to look at many things, the cytoskeleton, peripheral proteins. And
for in vitro experiments knowledge is needed about stoechiometry,
expression, purification, reconstitution, etc. Mark discussed two systems
he likes to work with. The yeast alpha factor receptor is small.
Glycosilation is not important and it produces well. Unfortunately, it
does not have a small molecular agonist. He also liked the E Coli
glycerol P acyltransferase because it overproduces in crystalline form,
but since this molecule does not have seven helices, he did not elaborate
much about it, except for a warning that this molecule became less
inactivated because of the very high density in the overexpression
system...
He discussed levels of (im)purity. For crystallisation a small
amount of impurity often is not a catastrophe, but a small amount of
impurity that is similar to the product is terrible. This puts other
conditions on the purification depending on what one wants to do with
the protein afterwards. Putting a tag on the overexpressed GPCR helps
to avoid that mixtures of GPCRs are purified.
Mark sees purification as three steps: Solubilization (avoid
denaturation), stabilisation and the assay. He discussed the advantages
and disadvantages of putting tags on the GPCRs.
Detergents are very important in the GPCR Solubilisation. A good
detergent is easy to remove afterwards, does not change its behaviour
as function of concentration, pH, salt concentration or temperature, and
they do not denature the protein. Some detergents always form the
same size micel, others don't. No one has made a detergent yet that
combines all good characteristics, but mixtures of detergents sometimes
seem to be very nice to the GPCRs.
Natural mammalian membranes normally consist of the major
components PE, PC, PS and cholesterol, and the minor components PG, CL,
PI, Galactolipids, DAG and Ceramide. (PG, PC, CL and PS form vesicles,
the others don't). Several aspects of the influence of the membrane
constitution on the GPCR functioning were discussed. Often, a pure
brain lipid extract provides the best membranes for reconstitution and
stabilisation of the receptors. However, the amount of lipids needed for
reconstitution were always higher than expected.
Mark finished his talk with a set of examples of poor or well
working lipid mixtures. Sometimes the wrong lipid leads to a five fold
reduction in reconstitution, but a badly chosen set of lipids can prohibit
reconstitution completely. Also, many lipids oxidise upon purification.
What this means for our experiments is today one of the many questions
that we desperately need to answer.
Susanne Trump-Kallmeyer used, just like Bob , the low resolution
data of opsin to place the helices. She built vasopressin and oxytocin
receptor models using the opsin helical framework. Rather than using
molecular dynamics, followed by drug docking and sequence analysis
techniques, as described by Bob, she decided to verify and improve the
models using mutagenesis data.
Susanne started by discussing the properties of the ligands.
Oxytocins all are nonapeptides with an intra molecular cys-cys bridge.
They all have one polar part. The receptor models all showed a polar
cavity about 10 A deep in the membrane, where the polar part of the
oxytocin molecules would fit very well. Susanne did an alanine scan on
all potentially important residues. All but two significantly reduced the
binding of the ligand. Several other residues that were mutated into
alanine influenced other aspects of the receptor function, but not the
ligand binding. The conclusion must be that the residues involved in
ligand binding were predicted correctly from the model, but the exact
mode of binding is not yet clear. Several new mutations can now be
suggested, and Susanne intends to produce those in the near future.
The second example was the substance P receptor. She concludes
that the N-terminus is only needed for binding, and not for specificity,
whereas activation takes place between the helices. (This contradicts the
conclusions by T. Schwartz, who measured that substance P does not
bind between the helices). Some chimeras were discussed to prove this.
The third example, was the bombesin receptor. Here also the N-
terminus is needed for binding, and the C-terminus for activation. She
showed that also bombesin docks nicely between the helices, even a bit
deeper in the membrane than substance P. The model was used to
identify key residues involved in binding.
Jan Kelder concentrated on 5HT2C receptor mutants and binding studies that were designed to shed light on the quality of his serotonin receptor model. He listed the headaches till psychoses that result from poorly functioning receptors in this class. (Some participants recognised their own mental state after a day of GPCR modelling.... GV). Jan used correlation analysis on sequence identities to determine alignments, and combined this with all mutation data. He built his models directly from bacteriorhodopsin. His main question was "How does serotonin bind to these receptors", but of course, a better model was also an important goal. Jan started by showing that all previously existing mutation data would lead to two potential docking possibilities for serotonin. In the one model the two famous serines on helix V were playing a crucial role, in the other model a more prominent role was reserved for the polar residues on helix IV. Jan showed a beautiful set of data. In many earlier meetings it has been discussed that mutations that abolish binding do not provide enough evidence, but if mutations in the receptor could be compensated by modifications of the ligand, one would know how the ligand really docks. Well, Jan showed such a set of data. Despite the large amount of mutations and compensating ligand alterations, no real hard conclusion could be drawn. Too many binding constants showed only small or surprising differences. Nevertheless, the data tends to favour the binding mode in which the serines on helix V are important. More data will be produced to create greater certainty about this final conclusion.
I personally think that the presentations by Susanne and Jan brings us one step further, or actually, they move the problem one step further. About a year ago, many discussion took place at GPCR meetings about the ideal mutation studies or ligand binding studies to elucidate how the agonists bind. Now that two of these studies have been done, we know that this is not yet the solution. We need better assays, we need to understand the system better before we can interpret the results of receptor or ligand modifications. The ball no longer seems to be in the modellers or the mutators court....
Ad IJzerman discussed Adenosine 2A receptors. These receptors
play a role in the extracellular fate of adenine nucleosides and
nucleotides. There are four subclasses known, that are creatively named
subclasses one till four. Ad quickly discussed the model that he already
published, and indicated the, from a modelling point of view, important
residues. Two histidines (on helix VI and VII) are important, although
not crucial for binding. In Ad's model these two histidines, obviously,
point inwards, and they provide strong constraints for the docking of the
ligands.
He discussed how certain groups on the ligand were obligatory for
binding, but other groups could be changed for almost everything
without affecting binding. This led to the conclusion that this part of the
ligand was pointing outwards. Ad also discussed a set of smart
modifications in his ligands to prove which of the possible
conformations was the one that bound. Most GPCR antagonists are
larger than agonists. For the A2 receptor, however, an antagonist
exists that is smaller than the endogenous agonist. This made the in
computo docking more difficult.
Ad finished by asking the question what the agonist actually does
with the receptor, or in other words, how is the signal generated? He
studied his A2 model, and realised that upon docking the adenosine, an
OH group of the sugar ring points to one of the histidines in a similar
manner as the serine in a serine protease points towards it histidine.
Even more, the negative charge which in the proteases is provided by
the aspartic acid, is also observed in his model; the glutamic acid that is
conserved in helix I could be modelled just at the right position for
completion of the triad. (Laerte Oliveira has observed several similar
triads in other GPCRs, GV) Ad's idea is now that part of the signal
generation or transduction takes place via charge relay systems similar
as those seen in proteases.