So, did you jump from looking at the genome from a SNP level to this more macroscopic view?
We were never looking at the genome from a SNP level. We were looking at cancers and we were comparing cancers to sometimes unrelated people—normal tissue from unrelated people and sometimes normal tissue from the same person—and we saw more differences when compared to an unrelated person than when compared to a normal person. And that was really our first indication.
Nobody was paying attention to variation of copy number. There was a lot of talk about SNPs. There was this sub, almost below the ground, discussion of the concept of an "in / del," so when the sequencers were trying to assemble the genome, they were having a great bit of difficulty at certain places and they developed the concept of an in / del. An in / del is either insertion or deletion, they don't know which, they don't know which version of the genome to believe, so they postulated a small insertion or deletion. It was always small, because again in sequencing, you can't see the bit. So, there was this vague sense that the genome had types of variation that were not simply SNPs.
Nobody was studying that seriously and nobody had a sense of how common it was and whether it would be easy to study. And it was through our comparative cancer work that we realized it would be easy to study, that it was fairly common and then we devoted a study to it that resulted in the 2003 Science paper. That's its historical route.
And then how did this move to autism?
Autism was an example of what people call a complex, genetic disorder that was failing to be conquered by Mendelian SNP association studies. People were really breaking their teeth on this. To groups would rarely come up with the same conclusion. And it was fairly easy for me to believe that this was not the right approach to autism or to other complex genetic diseases, like schizophrenia or obesity—any large number of things. There were two things that I thought were being missed: one was the possible role of spontaneous mutation. There were really three things: the possibility of spontaneous mutation; the possibility of rare variants that don't exist in the population for very long because they're eliminated quickly; the possibility of there being many loci that could contribute to the disorder. And those three things were generally missed, and the way the Mendelians tried to deal with this was to say: "These are complex disorders caused by the alignment of the planets;" that there would be four or five loci and that if you got the wrong allele configuration at these four or five loci, you would have the disorder. This was sort of the hypothesis you heard to explain why they had failed.
And it was a very unsatisfying hypothesis because first of all it's not testable. Second of all, it gave them reason to hope that they could continue to use their methods, just scale them up and eventually they'd get [a] signal. So, it was sort of self-serving. So, I really didn't like it. I really disliked it. There were huge amounts of money that were going into supporting giant efforts of that type.
So, it seemed to me that there were simpler hypotheses. All we had to do was admit to the possibility that there multiple loci and multiple mutations, each one of strong impact and high penetrance, and you could still get the failure of the Mendelian methods.
Given that schizophrenia, diabetes and other diseases may also work in this manner, was there something that tipped the scales in favor of autism?
I thought for a long time that autism was the right disease to do this on, but didn't have the funding for a study of the size that would be needed. So, it was just a hypothesis.
How long had you had the idea to train your methodology on autism?