Over the last several years, as the human genome has been mapped and analyzed, there has been a movement toward personalized medicine in the form of drugs formulated for and targeted to individuals based on their genes. In his lab, David Goldstein, a professor of molecular genetics and microbiology and director of the Center for Population Genomics at Duke's Institute for Genome Sciences and Policy (IGSP), studies how genetic variations influence individuals' response to specific treatments. Goldstein has served informally on a research team that has used population-genetics analysis to study Jewish history. He wrote a book about the work called Jacob's Legacy: A Genetic View of Jewish History.
In Jacob's Legacy, you combine a traditional telling of history with your own genetic research. Can you talk about the intersection between those fields?
The only way you can actually do genetic history is when you get the questions that you address from somewhere else. If you try to use genetics to learn something about human history cold, you will get nowhere. There are just too many different historical scenarios that are consistent with any particular pattern of genetic variation. What that means is that you have to get your ideas of possible alternative histories from somewhere.
And ideally there are only two choices.
That's exactly right. The fewer the options, the more likely you are to be able to make progress using genetics. Jewish genetic history is a good example of that, where there were some [areas where] the historical context gives you extremely polarized alternative possibilities.
One thing you investigate is the idea of a Jewish priesthood descended through the years.
So what did you find in that particular study?
There are two things that stand out in terms of the Y chromosome genetic makeup of the cohenim. The one thing is that the composition is different from the general Jewish population, and more homogeneous.
The other feature is that the set of chromosomes among the cohenim that are similar can actually all be traced to an origin, and you can get a rough estimate of the date of that origin. With statistical reliability, it's more than 500 years ago. But the best guess of its age is almost like it's a purposeful construction. It goes right to the temple period, 3,000 years ago.
The key there is it's old. The estimate of the date is uncertain statistically, but we can absolutely rule out a recent origin to it. Five hundred years gets us out of the surname period. And that's really important, because there's a lot of evidence that there's a connection between surnames and Y chromosomes.
How do you calculate the age of the common Y chromosomes?
Generation to generation, genomes are not copied perfectly. There are mistakes that are made. And some sites in the genome are harder to copy accurately than others. The Y chromosome has a number of these hard-to-copy sites, and these change relatively frequently. So you take all of the chromosomes that appear related to one another, and you ask: How much variation is there at these quickly evolving sites? Then you develop a model to ask: How long would it take for that much variation to develop?
Y chromosomes obviously limit you to studying genetic traits passed down through males of each generation. But in the book, you presented another study using mitochondrial DNA to look at the maternal line.
The basic idea was that sometime after the beginning of the Diaspora period, when Jews were outside of the area of ancient Israel, they developed a custom that you'd be considered Jewish if your mother was Jewish. So you might expect, in a naïve way, that the Y chromosomes of the distinct Jewish populations would not necessarily be similar to one another, but the mitochondrial DNA would be, because you don't get any input of females from outside of the population.
We got what could be viewed as almost the opposite to the expected pattern, where the Y chromosomes were relatively similar among Jewish populations, and generally looked Near Eastern, Middle Eastern in origin. The mitochondria, on the other hand, were hugely different from one population to the next, and you couldn't tell where they came from because they were so different from anything else.
What does that suggest about the way the populations were formed?
My guess is that it reflects an origin of Jewish men who came from the Middle Eastern areas and established Jewish populations with local women—and in the beginning, not that many of them. And then at some very later point, the community said, "That's it, no more women." Barriers go up, and that's what fixes the mitochondrial composition.
At the IGSP, you study the genetic components of neurological disease, as well as treatments. You recently told The New York Times that you believe many of the predictions about the role of decoding the human genome in improving medical treatments were oversold.
A whole machinery was established to represent common [genetic] variation and connect it to diseases. That machinery was systematically applied to every single important common disease. We can view the output of all of that work as having found most of the common variants that are important for disease. And what we got out of it was very, very little.
Genomics is really extremely complicated crank turning—and this is not a criticism, because we had to do it this way. But what you do is you just churn through the genome and see what's in it. And that's turned out some things, but a lot of the biggest, most important things actually were already found before we got to the age of turning the crank systematically.
Today, having discovered eighteen or nineteen new gene variants that are linked to type 2 diabetes, can we do a significantly better job of predicting who is going to get type 2 diabetes than we could before, when we had family history, an individual's weight and so on? No. Can we do better than before in any disease? Probably with a few cancers we can do better than before. And certainly we can do better than before for age-related macular degeneration and glaucoma. That's probably it.
So what is next?
What we now have is a situation that a lot of people in the community are starting to refer to as a "dark matter" problem. Basically, when you assess what the genetic component of disease is, for disease after disease after disease, it is really high. Now that we have done these comprehensive studies for common variation, we can ask: How much of the genetic component have we explained? Usually the answer is only a few percent. So we have something else that we know is there, but we can't find what it is.
How can you tell there's a genetic component if you can't find the specific markers?
Things like, to what extent does the disease run in families? You can assess that by asking questions like, if there is an individual with schizophrenia, what is the probability that a first-degree relative will have schizophrenia? The answer is it is around 10 percent, whereas in the general population, it is around one percent. So the elevated risk if you have got a sibling [with the disease] is a factor of ten. You can take out the environmental contribution by using adoption studies for siblings, and there clearly is an overwhelming genetic component. In fact, for schizophrenia, it is estimated that 80 percent is genetic. But big studies have been done looking for common variants that influence schizophrenia, and we find nothing. No genetic contribution for common variants at all.
There are now all sorts of indications that a lot, and perhaps even the extreme majority, of the genetic control of these conditions is due to rare things that are kept rare in the population because they're bad.
If these genetic factors are that rare, how do you create targeted treatments?
The truth is we don't know. There is a concern, in some cases anyway, that personalized medicine could turn out to be too personalized.
The hope was, in the past, that you might break patients off into a handful of different groups that had different treatment requirements. [But] if almost every patient with schizophrenia or epilepsy has a different underlying genetic risk factor, you're not actually breaking them up into a manageable group that would benefit from alternative treatments.
It might be that we can take these different rare causes and class them together. For example, we are seeing in some of the work that we're doing that there's convergence on some relatively obvious pathways involved in epilepsy of rare causes. So maybe all of those individuals [who] converge on one pathway would benefit from one treatment.
Certainly, our job has gotten harder than it looked before.
What do you think about ongoing studies, including the Harvard study one of your IGSP colleagues is participating in, where subjects are submitting to personal genome mapping and making that information publicly available via the Internet?
For the rarer things that are being studied, people are finding things like gaping holes in the genome, where you've got a whole chunk of the genome that's missing for one of the chromosomes, and for fifty genes, you have only half of the complements of what most people have—you get only a copy from one parent and not the other.
Some of those are just bad, bad, bad. And that's no joke to have it and to have people know about it. And so I have some concern about having all of this stuff up for anyone to look at and what the implications might be for an individual. I have some concern about the movement toward full public disclosure. I don't know that we're actually fully prepared to deal with it, when the genetic differences turn out to mean more than we think they might right now.
Q & A: Genetic Pasts and Futures
January 31, 2009