The classic work on evolutionary medicine is the 1996 volume Why we get sick: the New Science of Darwinian Medicine, by Williams and Nesse. It's not classic enough that I've read it yet, so I may be rehashing some of their arguments. However, on reflection it seems there are four general ways to explain illness.
1) Noise. Entropy is positive; things break, and once you're past the age where your genes have assured their survival into another generation, you're largely expendable. (There are very important wrinkles to this in meme-rich species like us, but the general idea holds.) The idea seems clear enough, until you're explaining to your grandmother who's dying of heart failure that it doesn't matter because her DNA has passed on.
2) Pathogens. They evolve and new ones appear from elsewhere to which organisms have no resistance.
3) Mismatch. Environments can shift rapidly. Mismatch hypothesis was conceived about humans especially, who exist in dramatically different environments from that of our paleolithic ancestors a mere hundred centuries ago, owing almost entirely to physical changes we have made to our own surroundings. (Agriculture, living in cities, reading, mating and dominance patterns, etc.)
4) Heterozygote advantage.
The long and short is that there are a lot of schizophrenia candidate genes and no solid reason to link any of them to specific behaviors or treatment responses; only now are we starting to correlate their response-to-treatment with drugs. The classic case of heterozygote advantage is beta-thalassemia as a form of malaria resistance, but the explanation has since been expanded to cover SNP variants that result in rare monogenic diseases with uneven geographic patterning (like PKU) to controversial behavioral phenotypes (homosexuality) to putative recent natural selection of cognitive abilities (Tay-Sachs).
The second half of the aughts saw a growing list of candidate genes predisposing to schizophrenia, and Doi et al argue in this article that these candidates should be investigated as possible heterozygote advantage variants. However, behavioral phenotypes are difficult to model precisely because it's easy to imagine that the context of a behavior would dramatically change its advantageousness. For example, Jones et al found that while homozygotes for a COMT variant were more likely to behave aggressively than controls, while heterozygotes behaved less aggressively. Arguably a good polymorphism for modern humans; a good idea during the paleolithic? Other candidates show poorer fertility in siblings and other relatives of schizophrenia patients (which flies in the face of a possible heterozygote advantage). Heterozygotes for a variant in KCNH2, a primate-specific isoform of a potassium channel found in brain and heart identified by gene screen, were shown to have lower IQs and reaction speeds than controls. The picture is far from complete precisely because of the number of gene candidates identified during screening over the past five years, on the order of a dozen.
An increasing number of psychiatrists and neuroscientists believe that schizophrenia is not one disease, that in fact it is a spectrum of related disease processes (like cancer) that stem from mutations in any of these candidates. Viewed in this light, the clinical picture of this devastating and prevalent disease (1% of the population) becomes clearer. For any anti-psychotic, only 50-60% of the population of schizophrenics will respond. (Imagine if a new ibuprofen-analog came out, and 40% of people who took it got absolutely no pain relief. Not only would you consider the drug a poor one, you would want to find out what's different about that 40% that keeps them from getting relief.) Regardless of the involvement of any of the candidate genes, schizophrenia is doubtless a multi-component disease process, so we should expect that there are multiple targets.
It has been argued by Harpending, Cochran, Hawks and others that not only has there been recent selection changing the frequency of genes between populations in historical or near-historical times, but that some of these variants identified are for genes affecting the central nervous system. For this reason it is all the more interesting that a variant of a schizophrenia gene candidate identified as strongly heritable in Europeans (FXYD6, an ion channel regulator) is not associated with schizophrenia in Han Chinese (Zhang et al).
A logical next step would be to start genotyping schizophrenia patients for the candidate genes, then record efficacy and specific symptom alleviation for different drug therapies. Professor Brian Roth at Duke has done exactly this for the pharmaceutical half of it, though to my meager knowledge, no one has as yet sat down to try to correlate the effectiveness of drugs in treating the positive and negative symptoms of schizophrenia provided by people who have different candidate mutations. It seems that we may be trying to treat different diseases with the same drug, and then wondering why it's only 60% effective. This is what I'm considering doing for my independent research project (a mini PhD required of medical students); if we can improve outcomes for schizophrenics with the therapies already out there, that's a big win for schizophrenia patients and their families, and conceivably a step forward in understanding how the architecture of the human brain results in consciousness.
Consciousness and how it got to be that way