THEO MURPHY (AUSTRALIA) HIGH FLYERS THINK TANK

Preventative health: Science and technology in the prevention and early detection of disease

University of Sydney (Eastern Avenue Complex), Thursday 6 November 2008

Statistics means never being able to say that you are sure...
Professor Christopher Goodnow FAA

Christopher Goodnow trained in molecular and cellular immunology at Stanford University, at the Walter and Eliza Hall Institute of Medical Research and at the University of Sydney after completing a BSc(Vet) and veterinary medicine degree at the University of Sydney. From 1990 to 1997, he headed a laboratory at Stanford University Medical School as an assistant investigator of the Howard Hughes Medical Institute. Since 1997, he has been professor of immunology and genetics at the John Curtin School of Medical Research where he is currently division head. Christopher was the founding director of the Australian Phenomics Facility – a major national research facility for mouse molecular genetics. He served on the founding scientific advisory board of Illumina Inc – now a leading genetic analysis technology company – and was founder and chief scientific officer for Phenomix Corp, a private biotechnology company with treatments for diabetes and infection in clinical development. He has authored many papers in Nature, Science and Cell, and serves on the editorial advisory boards of the Journal of Experimental Medicine, Immunity, Genome Biology, and Mammalian Genome. His honours and awards include the University Medal from Sydney University, Assistant Investigator of the Howard Hughes Medical Institute, Searle Scholar, American Association of Immunologists/Pharmingen Investigator Award, the Gottschalk Medal of the Australian Academy of Science, Commonwealth Centenary Medal, Fellow of the Australian Academy of Science, Federation Fellow of the Australian Research Council, and the Australian Health Minister's Prize for Excellence in Medical Research.

Christopher has illuminated the mechanism of immunological self-tolerance through innovative integration of mouse molecular genetics with cellular immunology. His discoveries have changed our concepts of how self-tolerance is acquired and autoimmune diseases are prevented, by revealing that self-reactive lymphocytes are controlled by a series of mechanisms serving as checkpoints at each step along the process of antibody formation. He has elucidated how these checkpoints achieve self-non-self discrimination, through an ability of antigen receptors to switch between signalling lymphocyte proliferation or triggering tolerance responses via qualitative changes in the intracellular second messengers elicited.

I will talk very much from the science side of things. Paul's opening really set out the macro kind of problems, and I would like to take you into the micro. I will talk of things from personal experience. So I thought I had better start with this slide, just to acknowledge that I'll draw some specific examples from my own group and my colleagues. I won't go through all of this but will just acknowledge that some very talented people have generated some of the examples that I'll use, purely to frame the kinds of thinking and discussions that you might have from the genes side and some of the problems we have there.

One of the things that I would like to put from one perspective – which I think is really one of the other huge drivers that is going to be shaping any kinds of strategies in preventative, diagnostic or treatment health – obviously has to do with this ability that we now have to sequence every person's genome in this room or every person that checks in to hospital. I thought I would try and lay out for you some of the problems that we face getting past the hype, if you like, in dealing with trying to drink from that fire hose. Really, these numbers set that out. We like to think of each of us as being essentially normal or a wild type, as we would say. But, every time we reproduce, the error-prone nature replicating our chromosomes is such that one in every four offspring inherits a deleterious defective gene.

(Click on image for a larger version)

Most of the time we don't see the effect of that, because there is always a normal copy from the other parent. But those mutations have accumulated every generation since we came out of Africa. Some of those variates are common amongst all of us because they happened a long time ago. These are what we call common variates or common SNPs [single nucleotide polymorphisms], and there has been a lot of use of those in recent times in genome-wide association studies. But many of them will be much more unique to individual population groups, individual families; they will go back to that time when your mother or father walked off the ship at Botany Bay or off the plane from Italy – or the United States, in my case.

Just to put that in even more stark contrast, the mutations do not stop there. Every time our cells replicate, there again is a well-defined error­prone mutation rate. Of course, what that means is that, by the time you are of the age that each of us are now, every base pair, every nucleotide of your genome sequence has probably been mutated and changed at least once in one of your cells. So it is amazing that our bodies hold together as well as they do over our life times and over the generational lifetimes that we have.

That means, though, that we are all mutants; that all of us – once we start sequencing each of our genomes and try and associate that with our likelihood of developing a disease or of being prone to advertising for the latest pizza and Coke deal and the consequences of that – are in for a very difficult problem. One way of thinking about that – and this is the personal angle that I would like to put out to you – goes like this.

(Click on image for a larger version)

So you can think about this in terms of the problem of what we call phenomics. That is, we have the genome sequence on the one hand and we have our phenotype, which is the shape, size and prone to health or disease of our organs and our physiology. So the phenome is what the genome does.

What is really transforming that field from being a sort of micro theme of studying one gene and one biochemical pathway to this sort of integrative process, is the ability to intersequence the whole genome of someone or the genome of a cancer cell and compare it to the normal cells in your body, or these genome­wide association studies that try and associate common variates that arose many, many generations ago in our population with diseases.

The trouble is that, when we start doing that, what we end up with as an output – there has been an enormous burst of this in the last couple of years and it has been tremendously exciting – is something that I always hated more than any other subject when I was an undergraduate here at the University of Sydney, and that is statistics, which means never being able to say you're sure. In fact, some of these statistics are alarmingly small associations.

This is one of the vexing problems that we and each of you over your careers are going to have to deal with, and I will give you some specific examples of that. Many of these associations are so weak that you can only tease them out by studying thousands of individuals with a particular disease. Many of these are disease susceptible variates – 70 per cent of the people in this room have them – and yet their individual contribution to the relevant risk of developing a disease is incredibly small.

This has really created a dilemma amongst every one of us in the field: do we get excited about these outcomes, these little statistical blips? Someone mentioned The Hollowmen the other day – is it a 'blippity blip' or a 'dippity dip'? This is the problem. So we end up with a translation gap, if you like, because it is really hard to know what to make of a little statistical association of the kind I have just described.

My perspective on this – it's a personal one; it may or may not be right – is that one of the important things whenever we deal with complex systems, is the idea of triangulation; that you can never be sure about anything if your answers only come through one set of lenses. This is where experimental research, I think, gives us a separate angle into this matrix. There are others, and you'll hear more about them from some of the other speakers – epidemiology being a key one, obviously.

So the beauty with the mouse is that we can essentially get under the bonnet of the machine of the body, and get into being able to be sure. We can say, 'Well, here's a mouse with a mutation in this gene; it affects these cells and molecular processes in this way.' We can start to say, 'There's a wiring diagram here of X and Y and Z,' which you would like to be simple, although many of them end up looking as complicated as that slide that Paul Zimmet just showed.

But the beauty is that, where these two things intersect, things get very exciting. When you have a statistical association that sits over a gene which, on its own you are not quite sure what that means, but when that gene functions in a pathway that you are absolutely sure about, you can then start to bridge that translation gap because you can go back to people with that risk factor and test cells or physiological markers; what are sometimes called biomarkers. You can test things that are closer to the site of action of that gene and see how much they are really affected by that variant. What you hope for is that the association between the genetic variant and the biochemical process or processes will be a lot stronger than the association with the disease as a whole, because there will be fewer environmental factors interceding. That, in theory, would give you the sort of stepping-stones to bridge this very big gap between the genome that we inherit and the environmental factors to which we subject ourselves, and these clinical outcomes that we want so much to prevent or treat better than we do now.

Let me give you two examples, just to set the stage for that. They both come from this effort of ours to emulate, under controlled conditions, the sort of experiment that each of our family trees have been practising over many generations. The single biggest source of human genetic variation is SNPs – single changes in the genetic code. For each of us, there is one of those about every 1,000 bases, thanks to this legacy of error-prone DNA replication right back to Africa. That is a problem inherently when there is just too much noise in the system.

(Click on image for a larger version)

So what we have been able to do in mice is to essentially introduce in one generation a much lower rate of SNPs using a chemical called ENU [N-ethyl-N-nitrosourea], which essentially changes about one base pair in every million as opposed to one in every 1,000 in each of us.

The other thing which of course we can do in the mice, is to have these random variates that have been introduced in one generation, segregated in family trees that we have laws against in our society – where daughters or sons breed with their fathers or mothers, or brothers with sisters – so that these mutations come to homozygosity. So what are often very subtle, weak effects when only one of the copies is changed, turn into very prominent obvious Mendelian disorders in the homozygotes. The beauty is that in the mouse, of course, we can then see these gene–phenotype associations and we can then relate them to cells and start to put them into pathways. We then map them on to chromosomes and start to characterise them.

(Click on image for a larger version)

Just as a diversion – and to link into what Paul said – quite a number of the variant mice that come out of these pedigrees are obese spontaneously. They are not exposed to anything particularly appetising. They get a high-quality diet that is relatively lower in fat, but they still put on a lot of weight. What is interesting about that is two-fold. One is that these obese mutants – if you do provide them with essentially a McDonald's diet – become morbidly obese very early, illustrating the gene–environment interaction.

But the most interesting thing to me is that many of these obesity gene mutations that we have identified – and I will not talk about them – are novel genes that we do not know anything about yet in man, although they are highly concerted in man. All of them seem so far to be behavioural modifiers: they all work in the brain on satiety.

I think that is a theme that we have to deal with. We understand very little about the neurophysiological circuitry of what tells us when to stop eating, but evolution has put massive stock into controlling how much we eat because whether you survive or don't survive a migration, hinges on how much extra food you put on at the start of it. Probably at no time in our evolution have we ever selected for not eating too much. Even now, of course, we are not selecting for it until our economies fall over.

The examples that I want to give you come from the immune system, which has the advantage that we can get at those cells much easier than we can at the cells that control food intake.

(Click on image for a larger version)

The first example has to do with what is a fairly central and important question in immune regulation: whether or not cells which have receptors for a particular antigen – that we might breathe in, or be infected with, or be part of our body – whether or not those cells are triggered to turn on genes that start the cells to divide and grow and make many clones of themselves. If this is a virus, you want that to happen. But, if it is a pollen grain or part of your own tissues, there are very sophisticated mechanisms to stop this growth from happening. So, all of immune regulation essentially boils down to growth control. The most sinister example of a failure of growth control in the immune system is lymphoid cancers, such as myeloma, leukaemia and lymphoma.

(Click on image for a larger version)

A PhD student in my lab was interested in trying to understand one of the key growth control pathways in lymphocytes, an ancient transcription factor that turns on many of the important growth genes in lymphocytes. It is called NFкB and it can be traced right back to fruit flies. What we did not understand is how anthurium receptors were able to either couple or not couple to this growth factor, depending on whether the lymphocytes should or should not be growing. This story started with one of these pedigrees that I have described to you and where mutant animals had a single gene variate that made them unable to respond to whooping cough vaccine.

(Click on image for a larger version)

So this slide shows immunisation with heat-killed bacteria, and the mutants make no antibody compared to the controls. They could make some other antibodies, but I won't go into that. But the beauty in this generation is that we have the genome sequence of the mouse, just like man, and 99 per cent of the genes in the mouse are conserved between mouse and man. Their [genetic] spelling is no more different than the spelling between modern English and Geoffrey Chaucer's English. So, although we look very different from mice, cells in the mice are reading from the same dictionary and playing out the same code.

(Click on image for a larger version)

Jesse Jun a PhD student, mapped this mutation to a small interval on a particular chromosome, and then sequenced the genes and found a SNP that was unique to this pedigree.

(Click on image for a larger version)

It changed one amino acid in a novel protein, a key linker protein that brings many other signalling molecules together to drive lymphocyte growth. When that is mutated, these lymphocytes are unable to grow when they are stimulated by particular growth stimuli.

(Click on image for a larger version)

That story has been played out again more recently where Anselm Enders and his team – Anselm is here in the audience – identified another mouse with almost the same kind of defect of the B cells being unable to grow in another protein. Again it was a single amino acid change in a protein, but in this case it was an enzyme. What was exciting is that this enzyme is already being worked on as a drugable target – but not in immune disease. It was originally chosen by the pharmaceutical company Lilly because of the potential relevance of this enzyme for complications of type 2 diabetes. It is a protein kinase, and so there are already drugs against it.

(Click on image for a larger version)

To cut a long story short, what we were able to do in knock-out mice was to piece together a pathway that we previously did not know about where this enzyme – protein kinase C beta – phosphorylates this linker-adapter protein and then brings together a whole suite of other proteins to initiate B cell growth. If you mutate many of the components in this pathway, you prevent B cell growth; but some of the members of this pathway were originally identified by cancer.

(Click on image for a larger version)

The reason for walking you through that is that it illustrates this intersection between human resequencing and mouse studies. Earlier this year a colleague of mine who heads a big group at the National Cancer Institute in Bethesda, based on what had been figured out in the mouse about this CARMA-1 or CARD 11 linker protein, sequenced a whole series of human lymphomas, where B cells were growing out of control. He sequenced the gene encoding this protein to see if there were any mutations in it. About eight per cent of lymphomas of this class that he looked at had mutations in the same region as the region where our mutation was found. But, whereas our mutation – the change of this amino acid – interferes with the function of a protein and so prevents B cell growth, these mutations stimulate B cell growth and stimulate this signalling pathway.

Because this pathway has been worked out in so much detail, it has been possible for him to fingerprint these lymphomas; in every case where a lymphoma has this mutation it causes a particular fingerprint of gene expression that is unique to that kind of lymphoma. So you can be confident that, although this occurs only in eight per cent of lymphomas, the changes that have occurred somatically in the tumour cells are indeed contributing to the B cell growth. They are giving you this gene expression signature that tells you that this NFкB pathway has been driven continuously. Then they were able to do some elegant stuff to show that the lymphomas are addicted to these particular changes.

That illustrates how, if you just sequenced these lymphomas' genome and found these mutations and did not have that framework of information and Sailor assays to do on the lymphomas, you would just say to me, 'So what? There are lots of genes that are mutating in these cancers. I can't be sure about that; it's statistics.' What makes this powerful is the intersection between things that you are sure about that come from basic research, and things that are very tantalising statistical associations in man. The reason again for choosing this example, is that it illustrates that once you are sure like that, you can instantly start to translate.

I mentioned that this drug, enzastaurin, which is a protein kinase beta inhibitor, had been through phase 1 safety trials and was in phases 2 and 3 efficacy trials for type 2 diabetes. The realisation that this pathway was critical for growth control of B cells made it possible to initiate what are now promising phase 2 studies, to test this drug as a much more specific inhibitor for treating these kinds of otherwise very poor prognostic groups of lymphoma.

(Click on image for a larger version)

Let me just briefly give you one other example. This has to do with a gene in another pathway called PTPN22 in man. This is an example of the kind of excitement that is coming out of some of these genome-wide association studies. A number of years ago – and this has been reiterated in many, many studies around the world – a particular single nucleotide change in this gene is strongly associated with a range of autoimmune diseases: rheumatoid arthritis, type I diabetes and thyroiditis. About 70 per cent of people in this room have the susceptible allele of this gene, and the association between the gene and the allele can be made statistically confident only if you look at hundreds and hundreds of people. So its relative contribution or relative risk to diseases is quite small. Yet there is absolutely no question now after it has been replicated in many, many populations, that it is a component of the genetic risk to type 1 diabetes or rheumatoid arthritis.

But, again, what do you do with statistics? The thing that has made this a very exciting story is that this particular statistical peak happens to fall in a gene that we already know a lot about from basic research in a pathway. The pathway that it functions in again has to do with anthurium receptors on lymphocytes stimulating growth.

(Click on image for a larger version)

Without going through a lot of detail, there is a cascade of tyrosine kinases and this gene encodes a tyrosine phosphatase that negatively regulates this cascade. It was subsequently found that the susceptible variate of this gene causes not a loss of function of the gene but rather increased activity of this inhibitor. When you inherit the susceptible allele, your anthurium receptor signalling is actually diminished.

(Click on image for a larger version)

That is a problem for us in immunology because we are mainly used to knockouts that completely disrupt a pathway. So we know that both mice and rare humans with knockouts in this pathway have severe combined immunodeficiency and not autoimmunity. Paradoxically, you would expect an allele that gives you what you think would be a mild version of the knockout phenotype to cause deficiency; yet all of the epidemiology is telling us that this risk factor is giving you too much immunity and is directed at the wrong targets at self-tissues.

To cut a long story short, we have been able to find mice not only with knockouts in this pathway but also with imaginative names that would mean things only to people in this room who watched a lot of TV in the 1970s.

(Click on image for a larger version)

Some of these alleles are not knockouts; they are the kind of SNPs that occur in each of us. They are changing individual amino acids in the catalytic side of this central kinase in the pathway. Some of them, such as this allele, are really quite subtle reductions in function.

(Click on image for a larger version)

The real surprise came when Owen Siggs, an honours student in the lab, started to breed some of the severe alleles with the mild alleles in an allelic series, progressively decreasing the strength of TCR signalling. He found that, if you had a very severe reduction of TCR signalling, you had immunodeficiency and very few T cells. If you had a very mild allele, the mice were fine, at least in this environment, although there is evidence that if you place the mice into an environment where they are challenged with particular infectious agents, you can bring out a latent susceptibility to autoimmunity that these mice would have.

(Click on image for a larger version)

But right at this intermediate allele we got the same paradoxical outcome, where too little signalling did not cause immunodeficiency but actually an excess of production of the allergic type of antibody IgE and an excess of antibodies against self tissues.

(Click on image for a larger version)

The reason why is because of this other very big problem that we face in trying to relate genome sequences and gene variates to phenotypes and disease, which is what geneticists call pleiotropy.

(Click on image for a larger version)

This pathway, with this gene and enzymes at ZAP-70, does not do just one thing; it actually underpins a whole series of different functional outcomes in the immune system, some of which are critical for immunity. Others are critical for shutting down immune responses that we do not want to make, such as against allergic compounds or against self.

What we're learning is that if you knock the pathway out, you knock out everything and immunodeficiency reigns; but, if you inherit subtle alleles that just reduce the function, these kinds of processes titrate out before these kinds of processes, so you end up with the exact opposite outcome. The key thing is that this is a very complex circuit; there are many, many genes involved in it. It is a microcosm of the kind of problem that we face right through physiology where we really need to have these wiring diagrams figured if we are going to bridge this translation gap.

(Click on image for a larger version)

Another way of thinking about it is this: the genome encodes all of these switches and buttons and we would like to be able to get down to the console and know which ones do what and what settings each of us inherit at different stages. But we need more than statistical associations if we are going to do that. Thanks very much for your attention; I look forward to the discussion.

Discussion

Chair (Phil Kuchel): Chris is director of the National Phenomics Facility at the ANU and basically adopts a rather large-scale screening approach to identifying the bases of disease.

I have been intrigued by your number of the rate of mutation. Do you have a feeling for what, in our lifetime, would be due to photons and chemicals and what would be virally induced mutations? What would be the fraction of all of those?

Chris Goodnow: Oxygen is our worst enemy, unfortunately, I think. Every time our cell has to fold a disulfide bond, it has to spin out a whole lot of reactive oxygen species and, if they are not scattered, they are a fantastic damaging agent for DNA.

Chair (Phil Kuchel): Ian Frazer, this year's Prime Minister's Science Prize winner, reminded us that 10 per cent of the cancer load in the world is due to papilloma virus. I just wonder whether viruses are… (inaudible).

Chris Goodnow: I think to some extent. But most of these mutations, when you trace them back, are single nucleotide changes, so they are not viruses hopping in and out. It gets back to the environmental issue because many, many studies show that longevity is extended dramatically in various organisms if you limit food intake and/or introduce mutations that diminish metabolism and growth in the AKT pathway, for example – PI3 kinase pathway. Part of that extension of longevity may well be the fact that, by diminishing metabolism, you are decreasing the load of reactive oxygen species that are damaging your DNA somatically.

Question: Alex McLellan from the University of Otago, Dunedin. A lot of the mice that were involved did mutate and develop obesity. Is it possible that, as a population, we are also accumulating mutations, which would explain the high obesity?

Chris Goodnow: That's a great question. We definitely are. Basically, those mutational rates that I described are essentially fixed constants – at least in the germ line for the most part, as far as we know. The variable that determines how many mutations are in the population is population size. So, essentially, our population has been logarithmically expanding for a long, long time. There are bottlenecks that have probably weeded out some deleterious mutations, but most of these mutations are probably not under much purifying selection. Because their association with phenotype is so subtle, there is no way that selection can weed them out in any serious number of generations. You'd be tempted to say, 'Oh, well, they're all recessive because they are all covered by a wild-type allele.' But, when we look in the mice, it is actually the exception rather than the rule to have a really recessive mutation. When you start doing the cellular phenotyping, we find that in almost every case what looks like a recessive phenotype at the whole-animal level is really interfering in the heterozygote state already with the pathway. So it is a bit like a card game: if you inherit enough small cuts at a pathway – to mix metaphors – you end up with a problem.

For example, in my area of type 1 diabetes, the incidence of that has doubled almost every decade. That is not genetic; that is environment. The big question, as we know, is obesity. The issue though is that we are all potentially exposed to this increasing risk environment. In a type 1 diabetes analogy, it is our exposure or lack of exposure, as children to germs. We don't know and we will not really tease those things out until we can identify subsets of people who are at risk genetically for type 1 diabetes, and we can then dovetail that with the epidemiology. Otherwise, we just do not have the power, because the environmental risk factors are likely to be such ubiquitous things that, if you have a ubiquitous risk factor and then only one or two per cent of people that are really at risk, there is a lot of noise and not much signal.

Question: Michael Valenzuela, School of Psychiatry, University of New South Wales. I could really sympathise with these complexity issues and I guess my question is more about computational methods for understanding complex systems. If we think about neuro-imaging, which is something that I do, we get excited if we see a blob that correlates with some kind of activity in the scanner. You could do another activity and see the same blob or you could see different blobs that correlate with that blob. So we start talking about neuro context where certain parts of the brain are only important when they are in relationship – both spatially and temporally – with other parts of the brain as mediating behaviour. But then it becomes an issue of how you characterise and analyse that computationally. I think we are struggling. I think, when we start talking about multi-pleiotropic mechanisms that may be underlying very complex disease states, like diabetes or, in my field dementia, then I think we are really struggling to handle that from a data analysis point of view. Do you think new methods are emerging to handle these issues?

Chris Goodnow: That's a great question and it is another example, I think, of that same thing – of an integrated approach like that, which is becoming possible. As you say, you end up with these associations that are statistically compelling, but then how do you go the next step to tease it apart? I am not sure to what extent it's a mathematical thing. My gut feeling says that it is a matter now of the stepping stones that are needed, which are the older traditional disciplines of biochemistry and cell biology. For example, there were two big papers associating coffee nut invariance with schizophrenia in a number of studies. Those associations were incredibly weak; they were statistically very strong but very weak.

So the key issue there is going to be, I guess, some kind of combining neuro-imaging with some kind of Sailor or biochemical assay that can say, 'Well, the genes that are halved in gene dose in individual A with schizophrenia are indeed causing cell X of the amygdala to run by chemical pathway Y at path-X rate.' So how we bridge that sort of biochemical rigour with integrated physiology has always been a problem. I think in the past we have just avoided it. We can't avoid it any longer and I think that is why this is a timely issue to tease out. I think it is an area where Australia can make a contribution because sometimes the advantage of being smaller is that people in different disciplines are forced to talk to each other rather than sticking in their own groups.

Question: Bill Warren, James Cook University. Chris, I really enjoyed your talk. I just wonder whether you might be able to briefly comment on genetic background effects. When you do studies on mice, they are an inbred population; humans are an outbred population and the amount of genetic diversity is much greater in the human population. It wasn't all that long ago when you did a mouse knockout and you got the phenotype and you got a great paper. If you change the genetic background and change the strain, you get a quite different phenotype. I just wonder how applicable the findings in the mouse are to the human population, particularly when you put on top of that epigenetic modification, so heritable changes to the chromatin?

Chris Goodnow: In fact, different strains of mice are equally diverse; a base pair changes every thousand. As Anselm will tell you, it is our biggest problem, because to map these mutations we do need to cross to another strain and then we have essentially a small version of the human population. Many of our complex phenotypes hold their system diseases, whether it is diabetes on top of obesity or an immunological disorder; they often disappear entirely in the complexity of those additional factors. That has actually been a very interesting thing. The cellular and biochemical phenotypes are often much more robust to those modifying effects; whereas, with things that affect the whole physiology, there are many more moving parts and so there are many more parts that are affected both by other genes variants and by environment.

I think the mouse is exactly the same as human in that regard. That has been a huge negative experience for us, because it often means that we have a really exciting mutant that then disappears into the sand in the crosses, and that is someone's PhD project down the toilet. But the lesson from that, I think, comes back to what we are going to do in the human population. I think it gets back to that same question – that we need more precise cellular and biochemical phenotyping measures to act as stepping stones between these goal standard disease end points that we are trying to deal with, and both what the genes are doing and what the environment is doing with them.

Question: Paul Korner, Physiology, Sydney. One of the points that you made earlier on was the very small association between some genetic mutations and particular phenotypes in diseases. Nowhere is this more apparent than in the hypertension field, which is certainly an area that merits consideration in a seminar on preventative medicine. One of the things it seems to me that you may have done – because you are a geneticist, you start with a gene and hope for the best, so to speak. As you have given in a couple of examples, sometimes the best comes through but, in the hypertension field, it really hasn't – and we have good animal models there. One of the things that I have been doing for quite a while now is starting at the physiological end, because you have a large environmental component, and examining which parts of the control system really are different in these. You can find a few, especially in the nervous system. The question really is: if we are to make progress in these areas, shouldn't geneticists give physiology a bigger role than they tend to do?

Chris Goodnow: I agree entirely. In fact, I sort of hoped that was my message. I am not really a geneticist; I am an immunologist and a physiologist. The thing that attracted me to immunology was really that, like neuroscience, it was a way to go from the macro to the micro. I agree that at the moment we have such an emphasis in biology on this bottom­up approach and with a lot of hope that, starting at the genes with resequencing and so on, somehow it will all become clear.

I think with these genome-wide association studies – this is sort of the message that I was trying to give – in fact, it doesn't become clearer. It gives you some dippy-dips and blippy-blips and they cost $10 million a pop. We absolutely have to have a top-down integrated approach and the stepping stones in between; the cellular biology and the biochemistry. I think the challenge is: how do we create a training environment and a research funding environment that allows people to come from all of those different angles and not be a sort of monoculture of 'genomicists', for example.

Question: Naomi Wray from the Queensland Institute of Medical Research, and I am a statistical geneticist. I completely agree with what you are saying about understanding function and that will lead to treatment. But, in terms of preventative medicine, I think statistical association, even from the genome-wide association studies that are going on now, will lead to prediction of genetic risk much more quickly than the very long time it is going to take to understand every gene and its function.

Chris Goodnow: Absolutely. I shouldn't be too negative on that point. I am glad that you injected that because it is absolutely true. The thing that we must not forget – this is the reason why certainly I always come back to genetics – is that the problem with physiology, immunology or anything else, is that anything else that we measure in the body that varies between a healthy person and an ill person, we are always caught with this causality problem: is it a cause or an effect of their disease? Immunology is the worst field for describing this cytokine and that cytokine being up in this disease and that disease and it all being epi or secondary effects.

The power of genetics, which you point out, is that the inherited genome is only a cause, not an effect. That is why, as you say, in principle it has enormous power to get to the underlying causes of disease. The problem I think is – certainly in type 1 diabetes – that the goal there has been to identify children early on who are at risk of type 1 diabetes. The MHC has helped a lot in terms of identifying people at risk, but it is still not clear that any of the other associated SNPs really are having any useful utility in terms of predictive power; it is just that they are so small.