Tip:
Highlight text to annotate it
X
Male Speaker: I have a question. So, yeah. You talked about
the -- finding the common variants and rare variants and made a strong argument that there's
a lot more to find.
Sort of coming from a Drosophila background, I'm always wondering, you know, what kind
of special role model organizations might play. One thing you might do -- I mean, in
Drosophila study -- perhaps like G by E, or G by G, or other -- where will you not succeed
in, and what opportunities might there be, then, to transfer that to organisms where
you can maybe manipulate them better?
Male Speaker: Yeah. So, I think the human is struggling
-- I think it will succeed to identify the genes. The human is a terrible setting to
identify the interactions. The scale is just wrong.
So, we actually have certain advantages over flies with regard to gene identification,
namely a pretty small population size, a bunch of bottlenecks, things like that that may
simplify allelic spectrum in the human. But interaction, both G by G and G by E, is very
difficult to do in the human, and if you do the calculations of how many people you would
have to look at, I don't think it's a feasible thing. So, I think there's tremendous interaction
with model organisms, and when we ask at the end of the day, can we account for all of
the heritability, it's very clear from a number of papers that what we count as narrow sense
heritability includes within it components of interaction, and we're never going to get
to the end without understanding them. So, I think it's -- the model organisms have a
very important role to play.
Male Speaker: Can I just add that we -- there's a very robust
model organism, and, you know, organism-specific genome activity still ongoing at all of the
centers as well as the Microbiome Project. We just chose not to focus on those today
because of time constraints.
Male Speaker: So, we would argue NHGRI should indeed continue
to -- we focus on the human just because that was our assignment.
Male Speaker: So, Richard, I'm curious about -- so, you
guys can clearly do a lot of sequencing, go after lots of samples. How do you see the
implementation into the clinic?
Richard Gibbs: Well, we do have a bottleneck right now in
delivering full annotated information to all comers. But I think we're growing that very
rapidly, and that what we see though is a -- you know, necessity is the mother of invention.
So, once we create the real demand for the speedy and effective annotation that is clearly
palatable and digestible, both from the research front and the clinical front, then it comes.
We see that both in the private sector but we also see it academically.
So, in a way, it's pushing the edge here, what has driven the development of the things
that enable us to get to the next level. So...
Male Speaker: Juan [spelled phonetically]?
Male Speaker: So, both Eric and Richard, you talked about
this dimension of case control studies or basic research moving into the clinic. And
I guess for someone who's really impatient, I'd like to know when the economics really
transfer -- convert that around to when we factor in the logistics of re-contacting individuals
in the logistics of sequencing and bringing all of -- when do we skip the case controls
and go straight to the population, and then the nest the cases in there? And then what
does sequencing look like when we do that? Who's doing it and where?
Eric Lander: I think you can --
Richard Gibbs: Go first.
Eric Lander: Yeah. Well, I think you can do just an arithmetic
calculation. If you want 25,000 cases, assuming you have a perfect background control rate,
and it's a 1 percent disease, then you would've sampled 2.5 million people. If it was a 5
percent disease, it'd be five-fold less. So -- and then you're going to need a replication
study.
So, I think the answer is, depending on the frequency of the disease, at some point, it
makes sense to get it out of a large population. The question, of course, is do you have a
really good clinical characterization of those patients. So, it'd better be a large population
with really good clinical care and good health records.
So, I -- that's your answer. And it'll cross over for different diseases at different times,
and it'll be driven by the cost of doing this in clinical settings.
Male Speaker: So, that's the single disease answer. But
what if we're looking at 150, like in GWAS? Then, at some point, it's not just -- you
know, that 2.5 million will apply to a lot of diseases.
Eric Lander: Absolutely. If you're up for, say, 2.5 million
for the discovery set and another 2.5 million for the confirmation set, if we could contemplate
today the 5 million sample set, and we had incredibly good clinical characterization
of those patients, skip the case controls. Just go for it.
I don't think we're close to that yet either in cost or characterization of the patients
so my guess is that in terms of the layered portfolio of this institute, you're going
to be thinking about a next five-year period where we take these very good case control
sets, drive them to ground, understand the issues in them, and begin to segue into very
large population studies where we take on many more. So, we might not take on 200 diseases
by case control studies, but I think we darn well better take on 30 or 50 to really understand
them.
But you're absolutely right. It will cross over at the point where we can do it on extremely
large populations well characterized.
Richard Gibbs: So, a big part of the answer, Lon, is the
mode with which we move these data into routine clinically accessible medical records, and
the degree to which we solve the issues of the ontology and the phenotyping standards
and the communicability -- communication between the different aggregators of the data.
So, I think we're all headed for the same world. Whether it's 10 years to 15 years,
or whether it's just five, really depends -- we believe -- if we can mount some programs
that press on all the right points that allow those problems to be solved properly, so when
the data are truly flowing and truly ubiquitous that they are lodged in a way that you can
do those kind of experiments like an epidemiologist, not necessarily like somebody who's scrounging
up samples to do one study.
Female Speaker: So, do you have any examples of where, at
your institutions, the data that you're generating from these sequencing approaches have actually
been implemented and are being delivered to patients and you have some examples of -- upon
which to build?
Richard Gibbs: Yes, we just passed 2,500 in our Mendelian
diseases diagnostic group, which mainly receives samples from children and families with -- they
have a developmental disorder, or usually it's neural -- has neurological involvement
but also metabolic and skeletal effects, the full range of metabolic disorders for which
we now gain that figure of 25 percent solution rate because there's 25 percent of the cases
we can identify a gene which is known to be pathogenic. But all of those families receive
back a report which describes different tiers of mutation, different levels of significance,
different levels of actionability. And so we built the infrastructure to allow that
to happen, including the genetic counseling and the follow-up with the families, and also
the trickling back of the samples that are not solved into the research arena so we can
do more discovery to boost that 25 percent.
Male Speaker: At my institution, cancer is where that's
starting to happen, more than anything else. So, really, within the next year, we expect
that all patients that come into the cancer center with acute myeloid leukemia will be
sequenced, at the least at the level of exome and transcriptome as we also move to trying
to get whole genome sequencing into the mix.
Male Speaker: Tony, did you have a?
Male Speaker: Yeah, one of the rationale for the sequencing
at scale was the missing heritability argument. Can you make an estimate? If you had 25,000
cases and controls sequence, what would you gain? All of it? Part of it? What does that
mean for other things we should be investing in?
Male Speaker: So, look, it depends on what true genetic
architecture of each disease is. At 25,000, you can certainly say you have 90 percent
power to detect loci with a given selection coefficient and a given effect size. It gets
you a reasonable effect size in the neighborhood of three or four fold increases.
If it turns out that the natural world has put an awful lot of the heritability at such
loci, you'll find them. If it's the case that the natural world has lots of things of smaller
effects still intermediate between the common variants but kind of weak rare variants, you
don't find them as much. I think what we're seeing already, though, is that in each of
the diseases where we've increased the scale, even though we know the power is exceedingly
low, there are one or two interesting results coming out. That suggests that the number
of really interesting results that will emerge with an increased scale will be in the neighborhood
of scores to 50 or 100 such results.
How much of that will add up to the heritability, I don't know and I almost don't care because
if I get handed drug targets, that is, knock outs that are protective, that drive pharmaceutical
development, that may be okay, even if it, for example -- like common variants look like
they about 30 percent. Maybe these might get another 30 percent, and maybe another 30 percent
may go to interactions and other things that we can't nail down.
So, I don't think you can estimate it. But what you can tell is your power to find it
given the structure of it. And we can tell that, just like with that schizophrenia example
with common variants, there's a lot more that will be found with larger scale but I can't
promise you how far it will get. Only that there's a lot of biology that will still be
found.
Yeah.
Male Speaker: So, Rick, I liked you slide. It was something
like big questions in emerging technology.
Male Speaker: Insufficient --
Male Speaker: Insufficient technology. At the time.
Male Speaker: I wonder, you know, in the last hour, we've
seen hesitancy to move from genotyping to exomes and the community has some hesitancy
moving from exomes to genomes. And I wonder what do you think the community needs to be
doing to push this field so that mantra up there doesn't read "the discovery of the exomic
basis of human disease" but the whole genome basis of disease? What do you think we, as
a community, need to be doing to push that envelope besides cost?
Male Speaker: I think the simple answer is we simply have
to commit that more projects are going to be -- have whole genome sequencing as their
primary activity. So, the cost is obviously an important factor but it's also much more
difficult, much more computationally intense, and hence, more expensive to do those analyses
-- much deeper dataset.
But, you know, we've really skimmed the surface. So, TCGA, I think, may be 10 percent of all
samples at whole genome sequencing. And in some of those cases, that really drove some
discovery for certain tumor types. And if we thought about some large projects where
we got much deeper into whole genome, I think that would drive us forward.
Male Speaker: It's worth noting that the power to detect
in a whole genome is dramatically lower than the power to detect in an exome. In an exome,
you aggregate all the mutations in a gene. And they're all pretty relevant, or at least
many of them are pretty relevant.
In the whole genome, you have to figure out where to aggregate. If you aggregate over
too little a region, you don't have enough events, and your power goes to hell, and you
need 20 times more samples. If you aggregate over a big region, most of which is not functional,
your effect size is diluted, and your power goes to hell, and you need 20 times more samples.
So, this is a case where things like NHGRI's ENCODE project is really crucial because you
actually have to aggregate over functionally-meaningful units to find anything. Otherwise we're not
talking about 25,000 samples, we're talking about a half a million samples. So, this is
going to be an interaction with the biology and the analysis, and I completely agree with
Rick that we want to do some whole genome projects to begin to explore it, but we probably
still want to have an excess of exome projects because we're better powered.
Carl.
Male Speaker: Yeah, no, I actually want to follow up on
that point because I actually feel though that's actually incredibly important. There's
a whole sort of layer of functional biology that, in my mind at least, the three centers
are involved with that I didn't hear a lot about here. There's a lot on the sequencing
of large cohorts and what you could hope to identify there, but there's a whole spectrum
of other activity -- the CRISPR-Cas9 stuff that you guys have been doing and so on -- that
would be great to hear more about and sort of a vision for more of an integration that
could help us go the sort of base pair to bedside that you guys keep talking about because
one big concern that people have is we're going to go wholesale to sequencing, they're
going to be sequencing and then there's not a lot of information you're going to be able
to return back because we will have missed the boat in that [inaudible] the functional
biology.
Male Speaker: Couldn't agree more. Our assignment was to
talk about this. If you care to have us back, the functional end of it would be really important
to talk about as well.
Female Speaker: Hi. Yeah. Along the lines of the technology
development, and we've seen how sequencing technology is really driven and enabled many
of these kinds of studies, and you elaborated some on single cell technologies and others,
could you elaborate more on if you had your wish list of technology development needs
that would help enable this -- your primary goal here?
Male Speaker: So, of -- sorry, are we on? Of -- I'm not
sure I completely got your question, Dede, of sort of functional characteristic of annotation
or --
Male Speaker: Go for it.
Female Speaker: Whether technology developments to help this
could be in the proteomic space or whatever. But what other -- if you had your dream list
of technology development efforts that would help -- be enabling like we've seen sequencing
technologies --
Male Speaker: I mean, in addition to the nucleic acid methods
-- cheaper, faster, better sequencing -- functional essays that can migrate through vast datasets
and petition potentially interesting variants into those that you really want to study more
into those that you don't need to study so closely. That's the next big frontier, Dede.
We've got to be able to do functional studies not one at a time.
We're still in the mode where if you find a new gene, it's practically a career investment
for someone to do -- make a model organism, do a bunch of cellular studies, you know,
get their cell paper, et cetera. We need to accelerate that part of the process. But we
shouldn't be competing that with cheaper and more ubiquitous nucleic acids data either.
Male Speaker: And indeed, I think it's the marriage between
the nucleic acids as a readout because when something's getting cheaper, figure out how
to exploit it, and these other technologies. So, the single cell RNAseq that Rick was describing
that Veve Reggif [spelled phonetically] has been doing and it's up to 15,000 cells now,
she's also thinking a lot about single cell proteomics. How do you do that? Then how do
you do 100 antibodies simultaneously? Well, you do it -- bar-coding with nucleic acids
and being able to pull things down and cross-label. I think these CRISPRs -- the throwing in 70,000
CRISPRs simultaneously and using nucleic acid readouts.
So, I think there's a whole set of functional tools that we need, and they're at this intersection
between fancy technology, like single flow cytometry for these single cell projects,
nucleic acid labeling, new tools like CRISPR, and I think the other half of what NHGRI needs
to be thinking about is, yes, find the bases of disease, and then have a crank to be able
to turn to functionate them, and I think there's a whole lot of important technologies that
you can see being developed -- even a year and a half ago single cell sequencing was
not on the horizon. And now, it's totally clear, there are cell types in the immune
system we've completely missed and they emerge.
There are cancers, glioblastomas, which we've classified into four flavors: A, B, C, D.
They're not really four flavors. There are four cell types, A, B, C, D, and they occur
at different proportions in different tumors, and therefore we call the tumors A, B, C,
D, but we should really be calling them according to their proportion of the cell types.
There's a huge number of things about interpreting developmental responses by cells. Up to now
when we study cells at different time points, we're getting a gamish [spelled phonetically]
average of asynchronized cells. I mean, they're synchronized perhaps to you, but to them,
they're working at different time periods. When you look at single cells, you can reconstruct
the dynamics of that with much greater precision. I think there's a world of engineering here
and a world of computation that'll come together around these sorts of things.
Male Speaker: Eric, I want to come back to your statement
about the exomes. So, I'm a little bit surprised because I would think that the big genome
sequencing centers, one of the great assets you could provide is really getting the data
on the whole genome because by whole genome sequencing, we still get the exome data, and
the challenge we have is that we leave the rest of it on the table and here's a chance
to actually look for the aggregate by sequencing more and finding where these regions are important.
So, I'm a little bit surprised by --
Eric Lander: Okay. Easy to explain. At the same price,
whole genomes any day. Now, the question is, do I get 25,000 samples in their exomes and
have 90 percent power, or 5,000 samples in their genomes and have about half a percent
power? Your choice.
I would say that I would rather have the power to make discoveries first while the cost of
whole genome sequencing is coming down. Now, if whole genome sequencing is down at $10
or $100, no question. But at the point where you're not yet indifferent to an extra factor
of five, then you're making a real tradeoff. You may say I get to see the whole genome,
and I might get to see -- and you get to discover nothing. We need a portfolio of both. I'm
in favor of a portfolio in both. But I want to be really clear. There is a cost to deciding
to look in an unbiased way to a genome that does have biases to it.
Male Speaker: But this reminds me of the discussion early
on. There is a cost to sequence the whole genome, right, and the point was, was to figure
out how to do it, and you guys are the ones that figured out how to do this.
Eric Lander: So, if you mean, is our goal total genome
sequencing on everybody, you bet. All samples, total genome sequencing. If you mean, is the
next step over the next three, four, five years prove that you get meaningful biological
results and compromise by focusing in there because it's going to be more cost-effective
to get more? Yeah, that's important.
There's no doubt that's a temporary measure. I'm still up for whole genome, and that's
why we have to have -- even in the whole genome project, we had a long-term goal, and there
are a lot of very useful things along the way: genetic maps, which I recall you made
some of, and physical maps before one got to whole sequences. And I think we still have
to use that logic here of prove value at each stage while not forgetting about the full
goal.
Eric Green: In that sort of discussion of the broad portfolio
projects, we just heard the H3Africa update, the fact that they're doing genotyping in
diverse African populations. What role do you believe genome-wide associations and common
variants and understudy populations will play in the coming years?
Eric Lander: In non-European populations?
Eric Green: In non-European -- in non-U.S. populations.
Eric Lander: In non-U.S. populations. Well, I think it's
absolutely clear that different populations have amazing advantages that we must take
advantage of. So, for example, as you well know, in Mexico, there was this recent discovery
about diabetes, killer allele present at 30 percent frequency in Mexico that causes a
significant increase in risk diabetes but it's only half a percent in Europe.
So, that's one where it's a rareish variant in Europe and it's a common variant in Mexico.
And if we decided to focus on Europeans, it wouldn't just be politically incorrect, it
would scientifically incorrect. When you look at deCODE and the discovery of the particular
variant associated with APP, you know, there, you can pick some things up in populations
that have undergone bottlenecks.
So, the world gives us a whole lot of really interesting population genetic variation having
to do with its history. The fact that it diverged [spelled phonetically] and has different allelic
spectra, that it has different bottlenecks that have scattered things, different frequencies.
The right way to do this would be a partnership that involves, certainly, you know, our country,
but countries like Finland, that have a really interesting bottleneck, parts of Africa that
have bottlenecks, parts of India have all sorts of bottlenecks due to endogamy. We ought
to be thinking that way because if we really want to complete the discovery of human disease,
we're going to use the whole human population, not just what's convenient.
Eric Green: Any other questions from council? Okay. Rudy,
you're going to take over.
Rudy Pozzatti: I need a microphone.
There are a few last things to do in the open session. We're going to move now to the council-initiated
discussion. For the new members, this is basically your opportunity to bring things to our attention.
We've been in charge of the agenda up to this point, but if there are important issues you
want to bring to our attention, or perhaps suggest topics for future reports that we
could give to the council in coming meetings, you have the floor.
Female Speaker: Rudy.
Rudy Pozzatti: Yes.
Female Speaker: So, relevant to a lot of the things that both
Richard was saying and also in the some of the things in Teri's report, the Global Alliance
is having their big meeting in -- I think it's March, isn't it, Eric?
Eric Green: Yeah.
Female Speaker: And I presume someone is coming from NHGRI
Eric Green: I can answer that very quickly because I was
going to mention it later --
Female Speaker: No. I was just going to say that it would
be nice to get some kind of report on your reactions to that at our next meeting before
Eric Green: Already was thinking about it. There'd be
-- there's like five of us going from NIH. Mark Guyer is going --
Female Speaker: I figured.
Eric Green: I'm going, Jim Ostell, Phil Bourne, Harold
Varmus. So, there's five of us going. And absolutely I will update you on what I learn
Female Speaker: Yeah. Because I think that's very important
and very relevant to what we've been discussing today.
Eric Green: Completely agree.
Rudy Pozzatti: Other issues? Okay. So I'm going to draw your
attention to a couple of items of interest. They're linked in the open session agenda.
There's an article from the Atlantic titled "When Will We Cure Cancer." It's an interview
between James Fallows and Eric Lander. I'm pretty sure Eric didn't send us that article.
Someone else did.
Male Speaker: He did not.
Rudy Pozzatti: Yeah, okay. [laughs]
And there's an update from a recent activities provided by the National Society of Genetic
Counselors.
Every year, we bring to you the Statement of Understanding between the Council and NHGRI.
Internally, we call this the Memorandum of Understanding. This is basically -- it's a
relatively short document, and it describes how we will conduct business between NHGRI
and the council. So, we are required to bring it to your attention every year. We're monitoring
policy changes that go on at the NIH, and we will introduce those into the MOU, those
that are relevant to council activity.
So, Special Council Review was something that got implemented two years ago. There are no
substantive changes to the MOU. We went through it as staff. We changed a few "whiches" to
"thats." I'm not going to bother to point that stuff out to you.
I'll just quickly race through and give you some highlights. And again, this is largely
for the new members. There are specific type of applications that we're required to bring
to you. Things like program projects, institutional trading grants, cooperative agreements; there's
a list in the MOU.
Council can take four different actions when they're asked to review or evaluate those
applicants. You can concur with the IRG. You can defer for re-review if you believe, or
a PI has pointed out perhaps, that there's a flaw in the review. You can recommend for
high or low program priority, or you can defer because there's additional information that
council needs.
There's a description of a process called expedited council concurrence. There's a subcommittee
of council, and I think Jim is on it, Dede, and Howard. So, these three individuals get
a list of the SBIR and STTR applications about five weeks before the council meeting, and
they're asked to review and approve those applications. This simply allows NHGRI to
move forward faster on the award process for those.
There is, if you look in the ECB, the Electronic Council Book, there is a report of the early
council concurrence so the larger member -- the larger body of council members can see the
report of those applications.
Staff administrative authorities. NHGRI can take action on applications without council
approval. We can negotiate award amounts after they've come through the council. We can make
supplement awards, and there are limits to that: $150,000 or 25 percent of the total
cost that were approved by council when the application came to council, whichever is
greater. In the case of very large awards, we're capped at a million dollars. Whatever
actions we take, we are obligated to report back to the council at the earliest opportunity.
So, those are the features or highlights. It's a four-page document. I would encourage
the new members to read it. It's pretty easy going.
Any questions or comments about the MOU? Okay. Then I need a motion to accept it. And a second.
All in favor? Anybody opposed? Thank you very much.
Okay. One last piece of business, and that's to read the conflict of interest statement
to you.
"You must leave the meeting when" -- this refers to applications that are going to be
reviewed in the closed session. "You must leave the meeting room when applications submitted
by your organization are being individually discussed. In the case of state, higher education,
or other systems with multiple campuses geographically separated, 'own organization' is intended
to mean the entire system except where a determination has been made that the components are separate
organizations for the purpose of conflict of interest determination. You should avoid
situations that would give rise to charges of conflict of interest, whether real or apparent.
For example, you should not participate in the deliberations and actions on any application
from or involving your spouse or child, a recent student, recent teacher, a professional
collaborator with whom you've had -- with whom you have worked closely, a close personal
friend, or a scientist with whom you have had a longstanding scientific or personal
difference. The NHGRI staff will determine the appropriate actions based on recency,
frequency, and strength of such associations or interests either positive or negative and
will instruct you accordingly. In council actions in which you vote on a block of applications
without discussing any individual, the so-called 'on-block action,' your vote will not apply
to any application from any institution fulfilling the criteria noted above that constitutes
conflict of interest."
There should be a conflict of interest form at the table. Please sign them and Comfort
will pick them up.
I think if you wield your gavel, we're then done with the open session.
Eric Green: How's that?
Male Speaker: Very good.
Sorry. We'll take about a 10-minute break to clear the room and disconnect the cameras,
and then we'll reconvene for about an hour to continue the closed session discussion.
Okay.