Scientist Stories: Eric Lander, Forefront of Genomics

I'm Eric Lander. I'm the director of the Broad Institute of MIT and Harvard.

How did you first get involved in the Human Genome Project?

Oh, how did I first get involved in the Human Genome Project? Well, If I go back, I met David Botstein in the spring of 1985, and I had been working in— well, I'd been on the faculty of the Harvard Business School teaching managerial economics while moonlighting learning molecular biology in my spare time, working in flies and worms. But that particular day, after the Tuesday colloquium at MIT, Barbara Meyer introduced me to David Botstein in a way I hated to be introduced. Here's this mathematician who's learning biology, because of course most biologists ran the other way when you said anything that involved mathematician in it. But Barbara knew whereof she spoke because Botstein immediately cottoned on to the idea that this was a mathematician, because Botstein in 1980 had published this paper about mapping human disease genes with restriction fragment length polymorphisms, and he was frustrated that he was unable to apply it beyond simple Mendelian traits. And he wanted to talk to some person who knew mathematics and genetics, but he hadn't really managed to interest anybody, find anybody, interest anybody. So Botstein immediately, like, latched on and started talking about this question of could we ever map complex traits, polygenic traits. And we had so much fun arguing about that. David said you couldn't do things, and I said you could do things. He's from the Bronx, I'm from Brooklyn. We had a great time arguing. We had so much fun, we argued again the next day. And then we went to— he had to give a talk at the Human Gene Mapping 5 meeting in Helsinki, Finland, but he hadn't written anything else or done any other research since his 1980 paper. And so he really didn't know what he was going to give his keynote on. So, we worked madly for several months on new ways to map more complex traits. And David gave a talk on this, he invited me to Helsinki. That was my first ever exposure to the human genetics community and the human gene mapping community. I came back and then David Baltimore gave me a position as a Whitehead Fellow while I was still teaching at Harvard. And sometime in the early spring of 1986, I got an invitation to Cold Spring Harbor to attend the 1986 symposium on the molecular biology of Homo sapiens, the famous symposium where the Human Genome Project was debated. And I attended. I gave a talk in the first session, I think the third talk of the meeting. And then they had the famous debate on should we have a Human Genome Project. And how did I get involved in the Human Genome Project? I raised my hand twice during the debate and made comments from the point of view of a young scientist and what I was, you know, what the issues were for thinking about this project. And Maxine Singer and Paul Berg pulled me over afterwards as we were all leaving and said, "Oh, those were really interesting comments. Can we go have dinner together at Blackford at Cold Spring Harbor?" And we sat down and we did and we talked. And then a couple months later, a meeting was getting organized on a possible human genome project, and they realized they didn't have anybody who knew anything about data and analysis and computation. And so they said, "Oh, there's that guy at Cold Spring Harbor who raised his hand in the in the debate. And so, I got invited to kind of co-chair that thing. And anyway, through a series of incredibly lucky accidents of meeting David Botstein, who wanted to talk about human genetics, and going to the 1986 Cold Spring Harbor meeting, and, as I always tell my students, raise your hand. Raising my hand, I then got just sort of drawn into this nascent idea of a genome project in 1986, which was still very ill-formed. It was still Wally Gilbert saying, "Let's put up a big factory and sequence everything," and using Southern blots and whatever. And there were all the tensions and David Botstein and David Baltimore being very skeptical. So, through no planning whatsoever, I stumbled into this incredible ferment. And it changed my life.

Why was sequencing the whole human genome important?

The genome, I think, traces intellectually most strongly back to David Botstein's 1980 paper, which said if you could build a whole genome genetic map of polymorphisms, you could use it to trace inheritance for any Mendelian disorder. That was the first time biology had hypothesis-free discovery. Everything else in biology, the cartoon story is you have a hypothesis and you test your hypothesis. But nobody ever asked the children, where do hypotheses come from? And what if you're not smart enough to have a hypothesis of what causes cystic fibrosis or Huntington's disease? Amazing about genetic mapping, and of course known to people who did flies and worms but not known to human geneticists, is if you could do genetic mapping, you could find where a gene was without knowing what a gene was. Well, that necessarily meant we needed the whole genome because we weren't looking for anything in particular. We needed markers everywhere. We needed all the genes. And so if you were taking seriously the Botstein idea of Mendelian mapping, mapping of Mendelian traits, you needed a whole genome map. There were attempts to kind of cut corners. There were, you know, later Craig Venter would say, "Oh, I've got the expressed sequence tags. Let's just get the— let's get the genic regions." There was an editorial that Miranda Robertson wrote in Nature in 1986, coming out of the Cold Spring Harbor meeting, in which she expressed very well the skepticism of many in the scientific community. Why do we need all this genome and this non-coding stuff and everything when we can't even really understand a handful of genes? At the meeting, I guess, the genes for chronic granulomatous disease and Duchenne muscular dystrophy had been found and were reported. And they looked really hard and complicated. So there was a kind of view that, "Well, we shouldn't go try to get more genes until we could understand these genes." And it's really worth looking at that paragraph near the end of the— this news and views. And that was a perfect example of kind of old-style biology thinking, which hung on for decades, but was an old view that to understand something, you had to pick one example and go deep. And that's great. That's one way to understand things. But what the genome taught us was that there were complementary ways. That you could get aerial views and learn things that you would never learn by focusing on individual things. That if you were a geologist and you looked at each rock outcropping on its own, you could learn a lot about that rock outcropping. On the other hand, you would miss huge big pictures, whole layers that existed in many parts of the world, or continental drift from the continents fitting together. And so, it was the moment in which I think biology came to grips with the fact that the big picture taught you things that the little picture didn't, and vice versa.

What was your perspective on NIH directing the HGP?

I guess we can now say, you know, why was it a good idea that it was NIH versus the Department of Energy? Because NIH knew biology. It was about biology. The Department of Energy, you know, I think gets credit for having floated this idea as a big science project because back then they did big science projects and NIH didn't do big science projects. But on the other hand, they didn't know any biology. So, it was clear that the NIH had to have the intellectual organizing force in the country. At least it was clear to most biologists that that was the case. The DOE brought technologies and experience in organizing projects, but none of those were anywhere near as important as a feeling for medicine and molecular biology and recombinant DNA. So it was kind of a no-brainer, although it took a little while to kind of sort out.

How was the HGP an example of big science? How was it not?

The most important thing was it was hypothesis-free. That was, to me— and that's intellectual. The big science aspect of it is minor by comparison to the intellectual shift that biology had ways to answer questions without hypothesis. We still are digesting that, and it's still the case there are huge numbers of people in biology who have not fully digested this point that you have ways into problems without prior hypotheses. Potstein used to joke that, not joke, I mean he meant this quite seriously. Study sections would always be complaining about, oh well, this isn't a hypothesis-driven grant, and we need hypothesis-driven science. That's what most of the reviewers said. And Potstein began referring to this as, oh yes, yes, they want hypothesis-limited science. And the idea that, of course, hypothesis-driven was hypothesis-limited. Was correct. So that was the big idea. But then, of course, also, yes, it had to get organized. It had to be something that— well, David Baltimore, who I have infinite respect for and was the person at the Whitehead— the founder of the Whitehead Institute and hired me and all that, did have this marvelous thousand flowers bloom moment where he was thinking, oh, maybe we should get the Human Genome Project done with high school students all over the country and use it as a combination generating data and a learning experience for everybody and teach people about biology and wouldn't it be so cool if everybody was involved? And I sort of shot a hole in that by responding to David, yeah, yeah, the human genome, it's 3 billion nucleotides. There's a billion people in China. Everybody could just do a trinucleotide and we'd be done. And I think David Games realized that the logistic aspects of chopping up the genome into a lot of pieces and doing it in small, non-quality-controlled, unorganized ways didn't make sense. So it forced biology to also say, if we're going to do things, we've got to confront quality control. And quality control only exists if you're willing to do scale.. But scale only makes sense if the labs that are doing that are also committed to sharing. So many things flowed out of the recognition that problems have a natural scale. The genome had a natural scale. In the end, 16 groups around the world collaborated, 3 large groups. That was the natural scale for this problem. Had you tried to do it with 100 groups or 1,000 groups, You'd never do it. One group, not so good, because then you can go home at 5 o'clock and you get lazy. So, a little bit of friendly competition is good. But, you know, it had a natural scale, and I think we've come to understand since that there are problems that have different scales, and that's okay, and not all scales are the R01 scale.

What were the early rate-limiting factors facing the HGP?

Well, early '90s sequencing wasn't the issue. Mapping was the issue. It really was about, could we build genetic and physical maps? Sequencing was not even, you know, a serious consideration at scale. Getting a megabase done was, like, huge. So, it was about building genetic maps and physical maps and doing this in human and mouse. So, you know, in those days, it was a question of, should you build maps by getting a lot of clones and making restriction enzyme digestion like Maynard Olson and John Sulston had done respectively in the yeast and the worm. And there came to be an understanding that that was a terrible idea because the actual clone was not like a permanent landmark. And the idea that sequence tags, sequence tag sites, STSs, would give us the anchor points and genetic maps would give us the anchor points Before we could start building a whole sequence, we had to go figure out how to generate the anchor points against which we could— to which we could attach it to be able to assemble a sequence. Because in those days, you know, now you can shotgun sequence a ton of things because you can generate so much data you can overwhelm the problem. But in those days, being able to get 7 to 10x coverage was probably the limit and you really were going to need the anchor points. So from 1990 to 1995 or so, focus was get us the anchor points and the maps. And again, going back to Botstein in 1980, those maps would let us localize disease genes. We could tell they mapped to this section of a chromosome and having physical maps we could begin to march in and positionally clone them. Today you can double click on the phone and look up everything that's there. But I think what was brilliant in the late 1980s, and this the Bruce Alberts Committee gets the credit for, was converting this from a single monolithic genome corp model that Wally Gilbert had proposed to this staged event of genetic maps, physical maps, sequence maps, and model organisms coming along. And I think this has always been the brilliance of the the biological community and the NIH community in the U.S. is to think about— well, you will remember at the same time, late '80s or '90s, somewhere around then, the Superconducting Super Collider was going on. And eventually it died of its own weight. It didn't even get completed because it was very expensive. But the trick there was Half a superconducting supercollider doesn't superconduct or supercollide. It does nothing. Whereas a genetic map, even if it's not the whole genome sequence, starts getting you stuff. And a physical map starts getting you stuff. And so what was brilliant was to get a series of things, each one of which paid dividends and justified continuing to the next step. And that's why the Human Genome Project got completed and there's just a big hole in Waxahachie, Texas. I think the Genome Project was organized— you know, while it sounded like a monolithic project, it was organized to have accountable deliverables along the way. That's where the centers come in. That's where data sharing comes in. Look, I mean, this whole concept of freely sharing your data before publication was unheard of. And I believe it was first called for in the Alberts Report, where they said, because we're going to need centers, we're going to need that kind of sharing. It was— the Genome Project was an intellectual rethinking, and it was an organizational rethinking, sociological rethinking of how to do biomedicine. That's what's so interesting about it, because that intellectual point of view, hypothesis-free biology, and this organizational sociological notion of things at scale and and sharing of data we see play out afterwards. All of those have had this amazing legacy long after we declared victory on the Genome Project.

What was the significance of the earlier work of the model organism community?

It was crucial. I mean, you know, all these things went in parallel. So just as in the Human Genome Project, it was the genetic and physical maps that paid returns. Being able to get sequence in yeast— and in worm paid returns to the biological community too. So you've got these two strands, the model organisms and the kind of medical genetics that were coming up. And then around '96 they meet. Sulston and Waterston just working on worms, you know, me, others working on the medical genetics approach. And then they come together and finally it's time to take on how are we going to actually sequence the human. So, you know, we'll undoubtedly get to the fun and games of the Celera era, but, you know, when Craig Venter says, "Oh my God, they haven't really sequenced much of the human genome yet. I have to come in and do this." Well, of course, it wasn't part of the plan to be sequencing yet. It kind of would have been dumb to be sequencing from 1990 to 1996. You had to build all the infrastructure first. So, I think it was done in a very thoughtful way, and then we came to sequencing stage around '96, when it made sense to start talking about it, and by '97, '98, get it off the ground.

What are the key moments for entering the sequencing phase of the HGP?

The key moments— are arguments about sequence quality. That's really it. Maynard Olson just going on and on and on and on about quality. And, you know, he was just a broken record, but, but with the right song, going on about how important it was to have sequence quality because there were many, many ways that people wanted to cut corners, get low-hanging fruit, get a rough this and that. And Maynard pushed very hard, as did others, but I remember personally, you know, Maynard making a big point about this, that until we knew we could get highly accurate sequence, there was no point launching a big project because it could all be a big pile of junk. And so, a bunch of time was spent thinking about that. How do we get accurate sequence? How do we know it's accurate? How do we measure accuracy? All these things again that, that people may not pay attention to. And I think again, when Celera came in with the idea we're going to shotgun everything, um, you know, we'll come to it, but we felt it was okay then to go shift to shotgunning everything because we knew what it took to finish. If we didn't know what it took to to clean up and finish, we shouldn't have done such a thing. So then it came to scale was the other big thing. So I will talk about scale because that was an important thing. There were two philosophies. John Solstin, who sadly has passed away just last year, built the Sanger Center. And the idea was as many people as possible would be employed to make plasmid preps and do things. And at our place, Whitehead MIT Center for Genome Research, it wasn't an option. So, we very much took the point of view of automate this thing. Now, in those days, the automation you know, look kind of Henry Ford-ish. So, when we were building physical maps of the genome, we built this bromdig-nagian massive genomatron, which allowed us to set up plastic carts with 1,536 PCR reactions and seal them and then put them into these huge water baths and take them out. And then we had this this medieval bed of nails, the needles that pierced it and sucked it down on. Anyway, we automated the hell out of this so that we could run this with relatively small groups and we could scale arbitrarily. And so we had these competing philosophies. And that was, again, another strength of the Genome Project was there was nobody in charge of the Genome Project. It wasn't run by the NIH or the DOE or the Sanger or the Wellcome Trust or the Germans or whatever. It was run by a confederation of these, and the fact that there was nobody who could actually give orders meant that some degree of diversity reigned, and that was really good. And yet everybody was committed to the common goal. So the great strength— so the great strength of people pursuing different angles on it, but everybody sharing their perspectives was great. So, you know, we took the perspective of this was going to have to be automated and quality controlled and all that. And it went back even to the genetic maps we built. It went to the, you know, where— I mean, these are all so laughable in these days, but we had, you know, different kinds of multi-headed pipettors to load things. And so, from— and actually, in the early '90s, we got the the first automated pipetting robots in. They weren't used anywhere. So we felt that that was going to be really important. And then we took it all the way up finally to sequencing with the Genome Project. This picture that I think appears in the 2001 genome paper of our entire production floor of conveyor belts going everywhere. And that production floor could be run by 8 people. You know, it produced huge amounts of sequence, quarter of a million clones a day sequenced with 8 people running the production floor. But it was great because we argued with each other. We had different perspectives. And that's what made the project strong. People pioneered different pieces of it.

What was Francis Collins' role in the midst of all this discussion, confederation, and mediation?

And it wouldn't make sense for Francis to stand up and bang his fist on the table and say, "We're gonna do it this way." But also leadership is not just let them all figure it out. Real leadership is to listen and shape discussion, to recognize who's got good ideas and who's got wacky ideas and shape discussion. When people argue, to find ways to bring them together. So, I think, you know, it is not command and control, and it's not passivity. It's take a group of essentially 3 to 5 really good centers, and, you know, we had our genome call 11:00 a.m. on Friday mornings for like 5 or 6 years. Bring out the best in that group. By listening, encouraging. There were events, events like most notably when Craig Venter announces Celera a couple days before the Cold Spring Harbor meeting in '98, I guess. John Solston and Bob Waterston and Francis and I go out to an Italian restaurant in Huntington for a long discussion about how we're going to respond. And John goes on about, we should ignore Craig and just stick to finishing as we go. And I said, we will be dead if we do that. —because Craig will do 1x coverage of the genome and declare that he's now covered 63% of the human genome, because that's 1 minus e to the minus 1 by Poisson statistics. Then he'd do 2x coverage of the genome, and he would declare that he'd done, you know, all but a seventh of the genome. By random, and that those would be— and I just sort of like recited the press releases that Salero would issue. And basically, by the time we were done with dinner, mostly John and John taking the purest position and my saying, "We have no choice but to first do a draft coverage," John then extracted a promise that we would go back and finish. He said, "If you do this, you're not gonna finish." I said, "We will finish. I will sign in blood that you will finish, that we will finish." And at that dinner at the Italian restaurant in Huntington, with Francis playing a really good mediating role, we reached an agreement that we're producing a finished genome, but there's nothing wrong. We now know how to finish. We're just gonna do a draft coverage, and then we're gonna go back and finish. And we signed in blood— there wasn't actually blood, I think, but we would have if we had to— that that was the plan. And we went back and decided that's how we were going forward. That was really just 4 of us at a table, but it was a very good example of how it worked. Nothing was imposed. It was very different points of view. People coming together for an impassioned but respectful argument and then reaching a consensus. Francis managed that social and scientific sort of process better than I've ever seen anybody manage it, so that he managed to, you know, not be overbearing on it, and he managed not to do what people do of just be passive about it. So he gets much credit in my book for, like, managing to come down the middle of the highway in the right way to help bring out the best in a group. And that's what real leadership is.

What were Craig Venter's contributions to the Human Genome Project?

You know, elements of it were just plain annoying because Craig would just make statements that were kind of absurd. Oh, you know, the NIH hasn't actually sequenced much. Of course it wasn't part of the plan. Yeah, I can guess, even in this current era, this current Trumpian era, where it's much more common to hear people make these absurd statements that are just designed to tweak you and poke you. But at the time, you wanted to just strangle him and say, "That actually hasn't been the plan." And when he says, "Oh, we're going to shotgun everything," we're going to say, "And the quality is not going to be very good. The plan was to get a high-quality genome." So, it was kind of moving all the goalposts around and complaining about things. And so, you know, and the NIH should go do the mouse. It was clearly an economic plan to monopolize the genomic data. Craig had been running a genome center at TIGR, the Institute for Genomic Research, and it hadn't been very successful. There was an article in Science magazine that looked at the efficiencies of the different genome centers and Craig's was just not successful. So when ABI came along and said, "We have these new sequencers for you," for him it was a way to to now just get out of the pack and do it his way alone. And the notion was he was going to basically try to prevent the NIH from going forward by going to his local congressperson and saying the NIH should shut down because it's interfering with private industry and we shouldn't have that. The government should not interfere when industry can do it. They're really kind of bumbling idiots. They haven't gotten much done, and they don't have a plan, and I can do all this. That was extremely annoying stuff, and to my mind, not enormously helpful. Now if I put all of that aside, the idea of let's run an experiment to do it another way, of let's do all this shotgun stuff, perfectly useful thing. What it did was it did provoke from us a very serious reaction saying, "Ah, this is nonsense. We're prepared to rearrange and accelerate the order of certain things." The plan had been finished by 2005. Genome Project got finished by the middle of 2003. So, about a year and a half earlier than had been projected in 1990. At most, Celera's entry shortened things by about 18 months because of the competition. But what it really did— and here's where I give it great credit— it got a lot of funds liberated for the NIH. Because of the competition or the threat of competition, NIH funding for this went up and the genome centers could finally scale. Because until that point, there had been about 16 to 20 smaller genome centers and The NIH at that point bit the bullet and said, "We actually need to, like, scale up those people who know what they're doing." And so larger investments were made, and that helped a great deal. I believe the Genome Project would have finished by 2005 anyway. It finished by 2003. But it kind of forced it into taking scale more seriously. And that's what Venter did. Because he was going to have a whole army of these API machines, you know, suddenly the NIH got over its reluctance to have larger centers. And Celera provoked the need to make those choices. So, all told, it was a fascinating experience while everybody was— you know, it stirred the pot in interesting ways. The shotgun sequence that Celera produced turned out not to be particularly good. It was actually a pretty crappy sequence when you were done with it. And it then pushed us— in the process, it forced one other thing. Were we serious about releasing our data? Because it was clear that Celera was going to use all the Human Genome Project's data to compete with us and try to claim that they had sequenced the genome. And we had at the decide, were we going to restrict them from doing it? And we talked, and we decided no, we were committed to total release, including to people who had knives out for us and wanted to compete. And that's what happened. So in the end, it also helped reinforce values. So I think a whole lot of complex things came out of this process that made people more comfortable with pushing scale, reinforced values, you know, demonstrated a certain kind of flexibility, and it certainly got public attention.

What were some of the key things that came out of the analysis of the draft human genome sequence?

Again, to me, it's about the aerial perspective. So, you know, so I spent 6 months on the 7th floor of the Whitehead in the faculty lounge locked up writing the paper, because I mostly wrote the words in that paper. It's about 65 pages in Nature, so I didn't do anything else for this whole period. John Sulston, back in, I guess, June, John said, "Well, let's just write a half-page description in Nature and be done with it and go on back to the finished sequence." And we had another of these arguments where I said, "No, this does require, like, an incredibly serious paper." And John said, "Well, in that case, you write it." So, to me, the point was we saw the whole continent laid out. I think David Baltimore wrote a perspective in which he says, "While the 65-page Nature paper on the Human Genome Project doesn't hold a candle to the Watson-Crick double helix, it still sent shivers down my spine, I think was how he described it. Because up until that moment, the idea of a human genome was theoretical. It was someday. And I used to give talks in the late '90s where I talked about, you know, someday we'll be able to double-click and up will pop the human genome. And that was a laugh line. People laughed because And it was sort of intended as that, that you would double-click and there would be the human genome. Because people didn't, like, imagine it. And then it was there. And it fell together pretty quickly. It fell together in the first part of 2000 with assembling all these pieces. And you suddenly could answer questions like, you know, where all— you know, where where are these coding regions, kinases, repeat elements, distances. It was— you saw the whole thing. It's as if you'd heard tales of a continent far off, and then suddenly the fog cleared and there was the continent. And that was what the point of the paper was. You suddenly realized we were going to know everything. We were going to— we weren't there yet. We had 90% of the sequence. We had an overestimate of the number of genes because we couldn't really fully get rid of the pseudogenes and— but to me, the amazing thing was we were going to have it. We were going to have polymorphism because the companion paper from the SNP Consortium, we had 1.42 million polymorphic SNPs. A mental turning point. So imagine that you lived in the 1400s and maps of the world just had these regions where it said, "Here be dragons." And then suddenly you're in the 1970s and you've got these pictures of the Earth from the moon and you see the whole You can't imagine how different those worldviews are. To live in a world where you didn't know what was over the horizon and what the whole world looked like, to a world, you know, where today we have GPS down to, you know, a few feet. It's a mental shift. That's, to me, what that genome paper was about, was there was a mental shift that said for all of medicine, we were going to have this foundation. Stay tuned, there were bits to be filled in, but nobody thought about the genome the same way after that because it just made it so concrete. And I think in the writing of the paper, we try to express this. We try to express, you know, the ways that knowing all the kinases, knowing all the thises, knowing all the polymorphism, that, you know, seeing evolutionary constraint, that that it would just shift the way we approach biology.

What avenues of research immediately followed the HGP?

Well, first thing was clearly comparative sequence. That, you know, apart from open reading frames, we just couldn't figure out what was going on elsewhere. And even for open reading frames, it wasn't that easy because of pseudogenes. So, for the few years it was straightforward. It was get the mouse and get the dog and get the rat and get, you know, aardvarks and other things because evolutionary constraint is an incredibly good functionator. We just didn't have any other general-purpose way to add function to the genome that was like a factor of 10 as good as what evolution had done through natural selection. So that was clear first. Then it was also clear we needed all the polymorphism. So, going back to the HapMap, the idea that we were going to need not just the Botsteinian map that had enough markers to trace inheritance, but we were going to want the infinitely dense map so that when we came to a region for polygenic traits, which is a whole long discussion to have on its own, we would know that whatever the causal nucleotide was, it was in view. And that took a while to get to that point where today, if I give you a region and I've got association to that region, and I got 12 SNPs in that region that show pretty high scores, it's almost surely one of those 12, and the question is which. Whereas before that point is, it could be many things we have never seen. Seen. And so again, to move to complete picture. So to me, that period in the early 2000s was extract evolutionary information and extract human genetic variation. Those were the two things that we needed. And then of course we had to think about, given all the human genetic variation, how are we going to genotype all of it? How are we going to do this in, at the time, what sounded like scary numbers, hundreds of people? So, then it began to diversify. But I would say those two things, comparative genomics and human genetic variation, drove it through about 2005.

What was your view about NHGRI continuing as an institute after the completion of the HGP?

Oh, I mean, it was silly. It had already been decided because the NHGRI had been been changed from NCHGR to NHGRI with the idea that changing from a center, which is inherently a temporary structure or a light structure, to an institute was going to ensure permanence. And so, I feel like that had been taken before. But many people who just thought genomics was going to end with having a human genome sequence, you know, it was very clear that it wasn't because of these things I told talk about comparative genomics, human genetic variation. And the categorical institutes just weren't going to do that. They couldn't do that. Which one should— is diabetes responsible for making the map of the human genetic variation and the haplotype map? No, the NHGRI— human genome research was not done by a long shot. It was clear it was the ground floor to have the genome sequence.

How did plans for the HapMap Project develop?

Oh, I think it was pretty clear by the time of Mark Daly's work in 1999, which didn't get written up until about 2001, but it was clear that we needed all the variation and how it correlated with each other. We needed all haplotypes in the human genome. So, how we were going to get there was the art of the practical. If you said, "We must have every variant," when we didn't know you know, how much that would entail, people would just shoot you. So, you declared some intermediate goal. And, you know, how many populations should you look at? And how many people should you sequence? How dense should you go? It was driven by what could be done. If sequencing had been 10 times cheaper, we'd have declared different goals. But it wasn't, so we didn't. And like with all the other steps of the Human Genome Project, build something that pays returns and you'll get to do the next stage. That's the art here, is have a scientific impact, show that it's worth something, and you'll eventually be able to go all the way. If you, you know, were to declare such an ambitious, expensive goal, you'll never get there. What we didn't know about the scope was much more on the side of how many people would we need to genotype to to find the alleles underlying common disease, because there was no way to predict genetic architecture to that level. We knew it was polygenic, but how polygenic? That was a shot in the dark. The building of the infrastructure, you could very well, just from Mark Daly's paper in, you know, his work on this 5q21, you knew how dense you had to go to be able to detect of ancestral segments. So even if you didn't have every polymorphism, if there was a causal SNP or a causal variant, it would be an LD with something you had genotyped with probability 90%, which was good enough to know you were going to get most things. And you knew that. So, there was like a theorem guarantee that we could detect most common variants down to a certain level. So, that, now of course, who understood that? Francis understood that, I understood that, a few people understood it. I'm not sure most people understood it. In fact, there was much resistance to the idea of doing this. There's this hilarious paper that was published by Joe Terwilliger entitled An Utter Refutation of the Fundamental Theorem of the Haplotype Map. And, 'cause it was just, people thought, why are you wasting your time? 'Cause I don't think most people appreciated that, you know, the only way to do human genetics like this was to move past families, family structure where you're just looking at meiotic transmission, to use the power of all of human population history. And in effect, turn the whole human population into some massive pedigree you can learn from. And I don't think people understood that it was a new paradigm for genetic mapping. They thought it was, you know, like a lot of stuff and a lot of, you know— but it was like when people said, "Let's sequence the genome." People said, "What are we going to learn from it?" And I think the people who understood what human genetics was going to have to do to take on common complex disease, Francis, you know, I think it was pretty clear at that point that what that path would look like for quite some time. So the next steps, Phase 2, Phase 3, 1000 Genomes, they were just, you know, new names to come up with the next step of what was clearly an intellectual program.

What does a comprehensive catalog of human variation mean?

It means now that we can say, let's say if we wanted to find SNPs down to 1% allele frequency, you know, we have 99.9% of them. In a given population. That's a comprehensive catalog. Comprehensive doesn't mean complete. There's always a little bit at the end. You heard my point. It's a big mistake to obsess over complete. Comprehensive means for all practical purposes you got 99+ percent. It's not what's limiting you anymore. The rate-limiting step is not that you don't have the SNPs. You might not have them for 1% of the problems, but, you know, you can count on largely having them So, we know down to even a tenth of a percent frequency in European populations, we do pretty well being able to impute them. They're in our catalog, they've been genotyped, we can impute them. So it means, for all practical purposes, we have all common genetic variation that we can follow. Now, peace to, you know, folks who point out, "Oh, there's still complex regions and funny deletions." and all that's true, and it constitutes a percent or so, and we should find ways to solve it. But to find the basis of common disease, you know, having these— this much of the variation now tells us how far we're going to get with that, and it really pushes us on to what are the next challenges, even as we clean up behind to achieve completeness. Completeness, we already know what this problem looks like and we can move on to the next problem. And that's the art of leapfrogging understanding forward.

Do you think problems like missing heritability will become more tractable?

Well, it's not so missing anymore, right? I mean, we, you know, we, I think we know where most, or at least where much of it is. It's clear that the larger the sample sizes get, more of it gets found, and moreover, the indirect methods that originally Sean Purcell and then much more completely Peter Vischer developed, lets us estimate that for many diseases we have half of the heritability in the common SNPs. So, we're now down to, okay, what's the rest? And my guess is we haven't even begun to include genetic interactions. Interactions. In no organism I know does everything act linearly. So, there's probably a lot. We don't have any tools to get it. And then there's what will rare variants contribute. But it's not as mysterious. The 2008-9 kind of period when people were agonizing over missing heritability, you know, there were all sorts of wacky ideas going around, you know, epigenetic thises and that. And, you know, it's power. It's insufficient power in those early studies. The power has begun to show us what's there. We still need more power in the rare variants. We don't know how to get the power on interactions. But my guess is most heritability is going to come down to, you know, perfectly respectable variants and their interactions. So, at the moment, when like with schizophrenia we have 250 loci that are genome-wide significant, the issue is no longer missing heritability. If you can't make real progress with 250 regions, it's not because heritability is missing, it's because we lack right now the tools to pin down which variant is it in the region, what gene does it affect, and what cell type. And what's exciting to me is I think human genetics is now entering a new phase in which it's going to take on these questions that, you know, I might call going from maps to mechanisms in a systematic way. Right now we go from maps to mechanisms like one at a time, you know, heroic efforts to wrestle this variant into submission. We need to bring the power of the kind of things that tend to get built within HDRI to make maps to mechanisms like falling off a log. I think there's a head of steam around it. I think I feel the same ferment that there was in the late '90s, early 2000s in human genetics. I think we're entering a phase when human genetics is about to undergo another such explosion. So I think there's a tremendous amount of just scale of human genetics will explode and sequencing costs will come down to $100 or so. Next is this variance to function or variance to phenotype notion. I think we can now see paths to build catalogs of everything. And here the Human Cell Atlas is enormously influential in that because this can only be done at the single-cell level. In the last 3 or 4 years, we've now cracked open this ability to look at the single-cell level. So I think there are just technological— and then being able to edit by CRISPR and prove things. So between single-cell sequencing, CRISPR, large biobanks and disease studies, you know, many other kinds of ways to assay cells, new kinds of analyses. Like all the pieces that are needed are appearing. And I think they will lock into place in the next 5 years. It will be a really generative period.

What are some technologies that have changed the DNA sequencing landscape?

Certainly sequencing technology has changed. Pretty dramatically since the official end of the Human Genome Project in 2003. The advent of massively parallel sequencing on slides has made a tremendous difference. You know, it's also the case that the ability to do accurate exomes Selection has made a huge difference, because we can focus in on 1 or 2% of the genome. These are probably the two big things, is just incredible density that can be achieved by sequencing by synthesis on slides, and the factor of whatever fold you want to call it. Concentration on a percentage of the genome, although you have to get higher coverage of it, so— but still a very significant factor that comes from being able to focus on particular regions. Then I guess I would also add, more recently, the ability to do single-cell sequencing. The ability to do things like droplet sequencing of individual cells and read out in that fashion the expression patterns of cells have been tremendous as well. So, I'd say maybe those are the three biggest changes in sequencing technologies.

What sequencing technologies did the Broad Institute invent or help perfect?

You know, we developed the exome selection protocol that's used around the world today. And so, that was a question of the tech dev group in the sequencing center working very hard to do this, because we recognized the importance for finding the genetic basis of common disease, that if we're going to look for rarer mutations, we're going to want to do that in exomes, which is where the vast majority of those mutations are going to occur, rather than in non-coding regions. So, that was a 2- or 3-year effort. There had been multiple attempts to do selection that had been done by other groups on chips on other things. We set out from the beginning to make a production-scale exome protocol and did so, and then it spread. So, that was not a case of our adapting to technology development outside, it was a question of driving a technology development. And I think it really helped to have a strong tech dev team at the NHGRI Genome Center. Similarly, the single-cell sequencing in droplets. The drop-seq and related kinds of protocols were, again, developed at The Broad by a whole bunch of people. I think of Aviv Regev and Steve McCarroll and others who did this. So, this, too, is a question of— and in an environment that was both biological and technological, it was possible to draw on lots of expertise to do it. The third of the technological changes in sequencing, which was certainly the most dramatic in many ways, was the sequencing by synthesis on arrays. There, what we did was we were in very close touch with the companies, from 454 in the early days, and then particularly to Solexa in England, which was eventually acquired by Illumina.— and were the earliest alpha and beta sites for these technologies from the beginning. And so we ended up working closely with the companies to beat up on their technologies. And I think the companies wanted to work with us because they would find out what didn't work with their technologies, because we did things at larger scale and higher quality control than the companies themselves what they could do. So, these were great partnerships, and for us it meant that we started early and got down the learning curves and made lots of changes and suggestions to these protocols. I think having centers' expertise with running large-scale operations meant that they had a lot of expertise with quality, with recognizing problems. So, it's a great environment in which to develop new methods, because you can— you come with the right kind of skepticism about methods and what you have to do to kick the tires on them.

Were there technologies that didn't succeed in the high-throughput sequencing environment?

Well, I mean, the 454 methodology turned out just not to have legs to be able to scale in the right way. We used it because it was the first thing available. But we discarded it when it was clear that it wasn't going to be able to scale in effective ways. We experiment with every possible method. You know, there was the flirtation with PacBio, who thought they'd be able to do genome sequencing as opposed to a niche application. They were just never able to keep up with the cost. Cost drives a lot. If you don't have the most cost-effective technology, it's hard to get the scale of datasets, and the scale of datasets drive a tremendous amount of discovery. Every technology is a failure? Every technology fails. I mean, you know, it's not a question of— of course it does. That's how you make it better. It's our job is to find that every way it can fail, every failure mode something has. So it's not a question of, oh, Oh, some technology showed up and it was great and didn't fail. Technologies fail. That's what they're supposed to do. Because, you know, there's 100 problems with any new technology, and you don't meet them until you take them out on a drive, until you start demanding which sequences can it sequence. Does it do well in GC-rich regions? How does it do on homopolymers? You know, what are the— you know, how sensitive is it to various sorts of How do you recognize when there's a problem? You want to have signatures of a problem. For us, a lot of it has been developing the kind of procedures that allow us to recognize when something's going wrong. Because something might not matter a lot if a student in a lab is going to go run one thing. If you run a lot of something, well, you care about whether it's right, and you also have the data to be able to learn whether it's right.

As the director of the Broad Institute, how do you foster innovation and collaboration?

You hire extraordinary people, and then you respect them. So, I think many places talk about having core facilities. We never have ever used the word core facility at the Broad, because as I say, only slightly tongue-in-cheek, a core facility is like your dry cleaner. You drop stuff off at You pick it up at 5. But you wouldn't say you collaborate with your dry cleaner. And people don't have deep respect for their dry cleaner. They're turning the crank. I think, you know, too many people imagine that that's what the partnership is between biology and technology. It's not. And we try at the Broad in our platforms— and we call them platforms because that's what they are, they're scientific platforms. To hire extraordinary people who are every bit as world-leading in the technology they do and the quality control they do as the scientists we hire are in their science.

And so, it's meant to be a truly equal partnership. Would you categorize the science being done at the Broad as big science?

I think there's some notion growing out of the Human Genome Project Project, the Broad is a centrally directed big science kind of place. Nothing of the sort. I mean, the Broad is a place that's unafraid to do whatever it takes to solve a problem. And if that means to be able to operate at large scale, because that's the scale necessary to solve the problem, we'll figure out how to do that collectively as a community. If it means bringing people together across disciplines or across institutions or across countries, we'll do that. Most of the problems that get solved at the Broad get solved at small scale. There are labs, lots of people with labs that are perfectly recognizable to a typical university setting. It's just that in most laboratories, people feel like they can't do something bigger if that's what's needed. So, I think what's liberating about the Broad for young scientists is they're able to say, "What does the problem require?" Not, "What can I do with this, you know, with artificial limitations put on me?" If the problem requires we do something like invent a new technology, that's great. If the problem requires— I don't know, I mean, outside of genomics— large-scale scale chemical screening, you don't have to develop it in your own lab. You can collaborate with somebody else who does it. If the problem requires creating a network of patients across the United States, and people would say, "Oh, you can't do that because how are you going to get IRB approval across all those institutions, and how would you reach the patients," and all that, I think the approach of the Broad is those are great questions to ask. Those are not ways to pour cold water on something. You say, "Okay, if we need IRB approval across the whole country, how do you do it?" And the answer is, there are ways. If we need to reach patients all over the country, hey, you know, this is the internet, there are ways. If we need to scale up, you know, sample handling or sequencing or anything else, I think what the Broad is about is how do you release the shackles from smart young people to let them take on the problem they want. And so, you know, I think going back to the days of the Human Genome Project, the biggest issue then was being able to do large amounts of sequencing, large science. Today, that's certainly important, but, you know, I have individual students in my lab who do, you know, a quarter of a million single-cell sequences. I suppose you might call that large science, but it's like a student who does it. It's not. It's how do we put the full power of genomics in the hands of individuals and groups? And it means we have to figure out how to collaborate.

Do you think there is a dichotomy between individual investigator-initiated projects and large center-driven projects?

Yeah, it's nonsense. It's just silly. I mean, maybe there was a time that was the case, but investigators want to initiate all sorts of things. Now, if you ask, is there a difference between hypothesis-driven science and hypothesis-free science? Yes, there clearly is. David Botstein famously used to refer to it as hypothesis-limited of science. So there's certainly a class of science where you go in with a hypothesis and all you're doing is testing it. It's pretty rare these days. For the most part, the ability to do discovery and discover things you weren't looking for is becoming more and more common. When people do a genetic study, they don't have a prior hypothesis other than that there's going to be some genetics. When people try to understand the cell types, they don't necessarily have a prior hypothesis other than to say there'll be cell types. I think the ability to take comprehensive views has pretty much burned down the line between the cartoon version of a hypothesis-driven experiment where you're only allowed to go in with an idea. And you test your idea and it succeeds or it fails. No, you go in and you actually look at what's going on in a biological system. You look at the, you know, the full range of genetic variation across a lot of patients, or the full range of gene expression, or the full range of proteomic expression, and then you make— and then you'll test those conclusions, but those conclusions are part of a natural cycle between unbiased discovery, test, and back. And so I think these are old-fashioned ways of thinking. Any scientist today who's not doing both discovery science and then testing the hypotheses that emerge— you know, last thing I'll say, the cartoon version of scientists test hypotheses and that's what they do, just ignored the question, where do hypotheses come from? They don't fall out of the sky. They come because we engage in organized looking. So science is today hypothesis generation, hypothesis testing, hypothesis generation, hypothesis testing. It's the way it should be. So I think all these Boundaries and distinctions and things, you know, don't really describe how it's done today.

With all of these reference cohorts being developed, what are the principal issues?

Well, the world's moving fast. If we're looking at where we are today, the cost of sequencing will surely fall to $100. It may fall well under $100. When that happens, we're talking about the amortized cost of sequencing a a person's genome as being on the order of $1 per year of life. Given medical costs in the United States, the thought that we might allocate $1 of the medical cost to getting the data means that, of course, we're going to have sequence on tons of people. We will end up, I suspect, with reference cohorts that aggregate together to 50 million people or more, and we'll have vastly more in the medical systems. The issue increasingly is not going to be that. It's going to be how do we manage to combine data across these cohorts. That's going to be the hard thing as we go forward. I mean, imagine a world where sequencing costs fall to $10. That's not inconceivable, although $100 seems like the next stop. We're going to look at everybody in Finland and and say, "Oh, this variation is common in Finland, rare in the rest of Europe. What's the, you know, what's it associated with?" We're going to do, just as before we did unbiased genome-wide associations for individual phenotypes, we'll simultaneously do phenome-wide associations across a zillion phenotypes. So, we'll be looking at all genetic variants and any phenotypes in the medical record, and there'll be a tremendous amount of such discovery. Now, yes, that's not precision medicine. That's the discovery that underlies precision medicine.

What will it take to implement precision medicine into the clinic?

If you really mean by precision medicine bringing that to the clinic, that's going to take a tremendous amount of work. If you're going to take it to the clinic in the form of, here is a standard of care action you should take, when person X has this variant, well, you're going to have to say, "When person X has this variant, in what context? When exposed to what? At what age? With what genetic background?" You know, we have a tremendous amount to learn. Now, that said, will we end up in a world where people are going to act probabilistically and say, "Look, I don't have all those data, but we're going to make our best model based on what we've got." And there it'll have an awful lot to do with what are the downsides, what are the risks you're exposing somebody to. If you have a completely harmless, certified, completely safe treatment, you'll use it when you think there's even a chance that somebody might benefit. Few treatments are of that sort. So what precision medicine is going to do is it's going to work in a layered fashion. It's going to say, "All right, we can recognize what pathway—" We'll put diseases in their own categories. That's right. Then we're going to figure out which pathways, particularly in those diseases, are deranged. And what it'll really do, I think, is drive treatment. Treatment, that the discovery of new therapeutics will be based on recognitions of key pathways and key diseases. We see this already, but it'll become far more routine. We'll see, for rare Mendelian diseases, gene therapies, including genome editing, as ways to approach things. There's just a vast range of ways medicine is going to change. Probably rethinking therapeutics to being much more modular things, delivery vehicles and information within the vehicle. But, you know, it's certainly not one thing and it's going to be incredibly messy because the information that we're going to be able to glean is going to be of of so many different sorts of certainty and so many different types of actionability.

What are your thoughts regarding germline genome editing?

Everybody agrees we shouldn't be doing clinical germline editing today. The world is only divided by those two people, those people who are too squeamish to use the word moratorium and those people who want to just call it a temporary prohibition, which the dictionary says is a moratorium. So, you know, the moratorium is easy. It's insane to be doing clinical germline editing today because, you know, issues ranging from safety to societal consent— none of the key issues have been addressed. The big issue that the article in— the comment in Nature addressed was what comes afterwards. We have a moratorium. Under what circumstances would we ever consider lifting it? Given that this is a technology that affects the human population as a whole, once you make genetic changes, you don't have any way to recall them. You can try to persuade people not to reproduce and pass them on, but we don't have any way to change it. It means one wants to be pretty careful before one goes off and does things. And I think the notion is, at least the piece we wrote said, there needs to be some international governance framework. We proposed a pretty, I think, moderate kind of framework, which does not have a treaty that bans it. Each nation is sovereign and can do what it wants to, but on the other hand, they should be transparent about what they're doing. They should be willing to consult before going ahead and doing it. They should be at least able to say with a straight face that they have broad societal consensus in favor of it. So we tried to lay out what would a world look like in which nations acted recognizing that they had authority to make whatever choice they want, but acted with due respect to the opinions of humankind. One way or the other, I think we're going to need such a thing. Because the questions around germline editing don't end with should it ever be done. Even if people decide there may be circumstances where it should be done, there are so many proposals, including nutty proposals, about what you could do with it. You know, any crackpot can come up with some idea of how they'd like to improve the human species. And it turns out it's really tough to know what will be the implications of of that. So, we have to separate serious medical applications from crackpot applications, from well-meaning things in the middle that turn out to be very hard to evaluate because human genetics is hard, and being able to predict these things. So, balancing what benefits you're going to get, you know, I've had— I know people who made suggestions to me about things that we should edit in the human population. Knocking out the FUT2 gene because it's a receptor for noroviruses. You know, the idea that everybody would therefore not have to get these cruise ship noroviruses sounds so attractive. On the other hand, it would dramatically increase, you know, by maybe a quarter million cases, the number of inflammatory bowel disease cases and colon cancer. It's kind of nutty. And in response, the person said to me, "Oh, but I didn't know that." Yeah, well, that's the point. That's why we proceed with some humility about it. And so I don't want to say whether we should or shouldn't ever cross the threshold. I do want to say that we ought to have some very clear framework for how we have these conversations. I'm all in favor of somatic therapy. We want to be able to take a patient here who's suffering from a disease and make sure we can treat them. And for that, full steam ahead on genome editing in somatic tissue. It's really the permanent enhancements to the human population. Currently, there is no situation where you can't prevent a genetic disease from passing to the next generation. If you know that the parents have it. In extremely rare cases, 100% of the kids will get it, get the disease. In that case, the way you'd have to prevent it is by a sperm donor, an egg donor, adoption, things like that. But in the vast majority of cases, you know, in vitro fertilization with genetic testing is sufficient to prevent it. So the number of cases where you're talking about preventing the birth of some new person with a genetic disease by using germline editing, pretty limited. The number of ideas for how somebody might think they're going to make the human species better, unlimited. And many of them very poorly thought out. Making genetically modified babies, one should think really, really hard about. It's not something to be done lightly, because it will in the long run have major effects on society if we willy-nilly just decide it's okay to edit the human genome. I'm not saying we might not someday choose to do it. I'm just saying that is an important enough thing, it better be done in a deeply thoughtful fashion. And currently, we have no international framework, no ongoing framework. We're in a situation where we have a committee meets in 2015 to organize a meeting and writes a 2-page statement and dissolves. A National Academy committee is constituted, writes a report with a chapter on this, it dissolves.

This is not a serious, grown-up, ongoing framework. How can society participate in the ethical debates related to genome editing?

Once you have transparency, you have the ability for people to object. So, you know, I think the transparency is very important because if a a country wants to go forward with something, others might say, "Well, it's a terrible idea for the following reason." And they might say, "Oh, you're right," or they might say, "No, we disagree." But the transparency begins debate. Debate is very important to this. What we don't want to do is just, you know, go down thoughtlessly and decide that, "Oh, there's one compelling case here which we should do," a parent who's homozygous for Huntington's and wants to have a and therefore we're going to say, "Well, germline editing is fine." We should recognize that that might be okay and the next 10 ideas might not be okay. Look what happened with genetically modified foods. GMOs are deeply, deeply treated with deep suspicion by large portions of the population, in no small part because it went forward without attempting to achieve broad societal consensus. Senses in attempting to listen. It's okay to take some time to listen.