A Taxi-driver writes

Your high-throughput drug-discovery programme and machines that go "PING"

My main reactions to the adverts inviting me to buy an extremely expensive machine (a machine that goes “PING!”) to help with my high-throughput drug-discovery programme are (a) I haven’t got a high-throughput drug-discovery programme and (b) I’m not convinced that high-volume genotyping would necessarily help me if I had.

This kind of advert (that admittedly subsidises a lot of the journals we read) often inhabits a strange world in which drugs are discovered via automated genotyping of human DNA samples. This genotyping will (in some vague, unspecified way) be connected with the discovery of a causative gene, and hence of drugs to combat the problem. The money made on the drug will easily repay the original investment of the punter’s money in the machine that goes “PING!”.

The argument, in a little more detail, generally goes as follows. Important common disorders, such as hypertension or schizophrenia, may have genetic components attributable to combined genotypes of “dangerous” or “protective” alleles at a series of key genes. In and around any such gene there are numerous single nucleotide polymorphisms (“SNPs”), which can be typed in affected individuals, and their allele frequencies compared with unaffected controls. Even if we are not lucky enough to be typing the actual causative variants, neighbouring variants should be in linkage disequilibrium and hence still show us where the action is - even if we subsequently have to narrow the search to get to the real culprit gene. This may identify the protein players in the pathways deranged in the disorders, the nature of the derangement caused by the genetic variation and lead to the design of specific drugs to rectify the imbalances caused by the sufferer’s adverse genotype.

Despite the obvious successes, most heroically in the study of genetic factors involved in type I diabetes, the problems with this line of reasoning as a general approach to the discovery of genetic determinants of human ill-health are many and various. Firstly, culprit genes have no obligation to place “dangerous” or “protective” alleles on a distinctive set of haplotypic backgrounds; only if they do will the gene be visible by an association screen. Secondly, making a simple step from strength of allelic association to physical distance assumes that recombination is smoothly distributed along human chromosomes (which it isn’t) and that stochastic noise from historical recombinations is negligible (which it isn’t). Thirdly, the success stories in narrowing down an allelic association to an assigned cause have been spectacular efforts involving huge numbers of samples (and researchers!), and in the end usually have some catch showing us how complicated life really is: the gene identified predisposes to the disorder, but only in Mexicans, or only if transmitted from the father. Fourthly, even if we establish such a causative variant, what do we make of the information we would get from the corresponding test? (imagine, for example, you are told you have a genotype which increases your risk of schizophrenia 2.6-fold — what should you do, if anything? How should your insurance company react, if at all?). Lastly, even if we have a gene, know what the protein does, and know exactly what’s wrong with it, will we be able to design a drug to solve the problem? That’s really a question for pharmacologists, but I would simply observe that despite knowing all about the human globin genes and their pathology for decades, the best treatment for the thalassaemias remains blood transfusion, and the main new drugs introduced in that time have been new iron chelators designed to combat the side-effects of that blood transfusion therapy.

So what’s the problem? I guess my main gripe with this kind of world is that it encourages anyone who can genotype a biallelic SNP to believe that if only they pick the right candidate locus they will find the variant which forms the genetic basis of tennis elbow, and hence design, patent and get rich on the sales of wonder-drug Elboleum, available in simple, easy-to-take capsules. On a larger scale, the pharmaceutical world seems also persuaded, though again in a vague way, that there is cash to be made by using machines that go “PING!”. In that world, it is perhaps not the desire for progress as the fear of being left behind by their competitors that is uppermost.

I have no principled objection to people collecting large amounts of data that may turn out to be useless — after all, everyone should have a hobby, and if someone wants to spend their weekends cataloguing every vole in Berkshire, or making scale models of Salisbury cathedral out of matchsticks, we should leave them to it. What I object to is the simplistic way in which the whole scheme is sometimes presented. Serious investigators of complex human disorders don't need to be told that real life is immeasurably more complicated that the naïve hype would lead us to believe. So who is the hype aimed at? Are the review articles, which have only recently (for example, Weiss and Terwilliger, Nature Genetics 26, 151-157) begun to dissent from the wide-eyed optimism that we are about to enter a beautiful new land, flowing with multifactorial milk and heterogeneous honey, at least partially aimed at funding agencies or R&D departments, for the purchase of more machines that go “PING!”? After all, to hear some people talk, it’s almost better to have the machine than the understanding, more important to secure the prestigious funding than actually do the research.