Clinical Implementations of Cancer Susceptibility Genes - Nazneen Rahman

Uploaded by GenomeTV on 16.12.2011

Nazneen Rahman: So good morning. Thank you very much for the
invitation. I very much enjoyed this meeting. For those of you who don’t know me, which
I think is pretty much everyone, I’m predominantly a researcher. My group identified genes that
are predisposed to cancer. I’m also a clinical geneticist, and I run the Cancer Genetic Service
at the Royal Marsden Hospital. And I’m really going to speak to that today. We are particularly
passionate about trying to realize the potential of new sequencing technologies to maximize
patient benefit and to bring that into mainstream practice.
So when we go the pre-questionnaire, I was amused I think to see that on the front page
it said that there weren’t going to be questions about the BRCA genes because the clinical
interpretation of those sequence variants was pretty much sorted. So I sort of first
laughed, then I cried, then I started foaming at the mouth.
And I really wish that that were true, but the reality is that despite the fact that
it’s been over 15 years since those genes were discovered and they are some of the most
intensively investigated genes both with respect to clinical practice and scientifically, we’re
still very far from having a consistent way of dealing with sequence variance in those
genes. And I think the potential for chaos and harm as more BRCA mutation results are
available is quite hi.
So, this graph will be known to everybody. With respect to breast cancer, the situation
is that there are about 24 common variants found through GWAS with small affects. There
are BRCA1 and 2 which have very high risks of disease. And then my group has been particular
involved in identifying genes in this area of -- [off mic] -- I shouldn’t do that because
I can’t move. You can use your imagination -- oh God, now got trouble. [laughs] -- in
this intermediate region. So these genes are very like BRCA1 and 2. They are characterized
by rare inactivation mutations, but instead of having risks of 10- to 15-fold, they have
risks of about 2- to 6-fold. And the discovery of these genes has really impacted on how
I think about missense variants and VUSs, so I just wanted to take a couple of slides
just to describe that to you.
So the way that we find these genes is by undertaking large-scale familial case control
resequencing studies. So we sequence the full gene in typically 1,000 to 2,000 familial
cases and the same number of controls or more. And when you do that kind of experiment for
this type of gene, there’s a very clear signal between your cases and controls with
respect to truncating mutations. But when you look at the missense mutations, the data’s
really quite interesting. So you will always see multiple rare, usually singleton missense
mutations that are only present in your cases. And these are just the type of missense variants
that are called VUSs or in gene discovery papers where controls are not analyzed, they
will often to be assumed to be pathogenic because you found them in your cases. But
if you are sequencing the full gene in controls, what you will see is -- and what we have seen
for all of the genes that we’ve found of this type is that there are multiple singleton
rare missense mutations that are in controls only. And if you try and work out which may
be pathogenic and which aren’t, what you find or what we have found is that there is
no significant difference between those two groups. You can’t pull them apart either
in terms of their frequency or their position or their conservation or their predictive
functional impact. And it’s very different from those truncating mutations.
So what is that telling us? Well, I think it’s telling us the same as lots of other
types of data, the data that Jonathan presented yesterday and Les also would, I think, speak
to this. And that’s that most missense variants in genes, particularly those in which we know
that the pathogenic mutations are inactivating are not disease-causing. Also, most missense
variants that are predicted in silico to be deleterious in some way are not disease-causing.
And that cauterization is deleterious, which is often assumed to be the same as pathogenic,
just isn’t. It’s incredibly difficult to pick out disease-causing missense mutations
even in genes which we know cause disease. And so I think the sort of bottom line is
the VUSs should be considered, and we consider them, as innocent until proven guilty. And
I think that everybody does know that when you talk about it, but actually instinctively,
a lot of people, even the most sort of expert in these fields, are actually when they see
those variants, they actually manage them as guilty until proven innocent. And I think
that that’s a problem.
So, with respect to clinical utility, we’re not using the GWAS common variants. And I
don’t think that we’ll be doing that in a meaningful way anytime soon. With respect
to these intermediate penetrance genes, I think once sequencing makes that cheap and
quick, we will be able to use them. But BRCA1 and 2 is clinically usable and in many ways
offers all that you could hope for from a disease gene.
So I think with respect to unaffected women, this is -- people are very on top of this
and they understand this. These women are at high risk of breast cancer, ovarian cancer.
We give them increased surveillance. They will often have their ovaries removed after
completing their family. And many of them will also have bilateral mastectomies. I think
surprisingly it’s underappreciated just how important and how helpful it is to know
the BRCA status in somebody who has cancer. So if you have breast cancer and you’re
BRCA positive, you’re at high risk of having bilateral breast cancer. You’re at high
risk of having ovarian cancer. Increasingly there are genetic-tailored treatments such
as PARP inhibitors which are close to being in widespread use, that are tailored to that
genetic defect. And this is making a sizable contribution to this cancer incidence. So
if you take unselected ovarian cancer just walking into the clinic, over 10 percent of
those will have a BRCA mutation. And all of those are at high risk of breast cancer if
nothing else. It’s a smaller proportion of unselected breast cancer, though in certain
subgroups such as triple-negative, it’s over 10 percent. And I would certainly like
to see, and what I hope the new sort of technology will allow us to do is to offer BRCA testing
to all women with breast and ovarian cancer.
So what are the challenges to the implantation of that? Well, we need cheap, quick testing.
And this should be achievable. Obviously there’s also the issue of patents which I’m not
going to talk about today, partly because I’m not able to talk about it unless I’ve
drunk quite a lot of gin.
So, [laughs] but I’m assuming that that bit will be solvable. We need to have quick,
simple report of the results. And that will have to be readily understandable by non-genetic
experts. And we will need that to be then triaged into what the clinical actions should
So where are we with that? We’re a long way off from that. So, we saw this slide yesterday
from Les. This is one of the ways in which classification systems for BRCA variants are
used. We don’t use this. Why don’t we use it? I think there are a number of different
problems. It is a fairly arbitrary system based on various assumptions. That sounds
like a criticism. I don’t mean it as a criticism. I think that sort of taking your best guess
is the sort of bedrock of clinical practice and science. I don’t have any problem with
that. I’m not sure that if you are guessing that you should quantify that to three decimal
places. And I think that if you are having an arbitrary, pragmatic system, you should
try and have it as simple as possible. And it should be tailored to the aims you’re
trying to achieve. And I think this is very laborious, and many variants you just don’t
have the data in order to classify it.
But I think the main problem is there are five variant classes, but there are effectively
only two management things. You either can manage it as a positive or a negative. And
therefore, you have to -- if you’re trying to have this system, you have to be able to
tailor that into your management strategy. So I get daily emails, I think, from people
around the world where they’ve got variants that are in classes two, three, and four,
and they don’t know what to do about them. And, again, going back to these things that
most people sort of think things as being guilty until proven innocent, there have been
hundreds of women who have had their breasts removed on the basis of variants that are
highly, highly likely to not be increasing their risk of breast cancer.
So, in part to manage my inbox, and in part because we wanted to think about trying to
make this more user-friendly, over the last year, we’ve been using a different system.
And this is the system that we use. So we have a variant category, but it’s very much
tailored towards going to a management category. And then the management category, it’s either
negative or it’s positive. And I think the key difference, which was a slight moment
of epiphany in my group, I think, was that instead of trying to classify every variant
individually, there was a default category that, when you first have a variant, if you
can’t do anything else to it, it goes into what we call our variant 2 category. So if
there’s a missense variant that’s never been seen before, you actually can’t move
it into pathogenic. You can’t move it into full [spelled phonetically]. And that just
goes straight -- that’s in 2 and it goes straight out to getting our report. If a variant
has been seen before and has been previously categorized, obviously it goes into that category.
If a variant is a frame-shifting mutation, you can just code that. That will go straight
into pathogenic because we know that it doesn’t matter whether you’ve seen it before or
not. If it’s causing premature protein truncation, it is pathogenic. And then we have the possibly
assignable is for the group where there is some information in the literature, but -- or
available, but it’s not sufficient to cause -- to say it’s pathogenic. And here, we’re
using genetic information because at least for the BRCA genes, there is no robust in
silico predictive information or functional information unless it’s showing that it’s
the equivalent to a frame shifting mutation that will allow you to be -- to call that
pathogenic. So those are the categories that we use.
This is -- so following Howard’s example, this is one that I did this morning because
I had one in my inbox. So, this is a variant here which has not been reported before. In
fact, the in silico predictions and conservation wouldn’t suggest that it was pathogenic.
And so that goes straight out to our variant 2. And this is what those reports say, manages
a negative BRCA test. The consultee should be informed that the BRCA test was negative.
I think the next bit would be a point of discussion, possibly a point for transatlantic difference.
A discussion of variant is not required. There’s a management recommendation. Breast surveillance
should be recommended appropriate to family history, and we have that goes to protocol.
We have a number of protocols on there. The only rule of thumb is that no protocol can
be more than one page. No predictive testing should be offered for this variant. No additional
analyses are currently required. So is this model transferrable or scalable? I think it
probably is both.
At the moment, our reports are going back to geneticists or genetic counselors because
those are the people who are doing those tests, but I could see how it could go into oncology
if the oncology people were doing the tests. The negatives would just be negatives. In
terms of the positives, the report would be slightly different. You would have the information
that the oncologist potentially would be using, and then there would be another thing that
would say “Refer to genetics.” So here, instead of genetics having to see everybody,
we’re seeing the positive people. We will do the genetic information. We will be doing
the cascading. And we can manage that in terms of hours. What we can’t manage is seeing
everybody who’s going to be negative. I think there probably are other genes and systems
for which it is also transferable.
What -- we’ve often been asked, what would you need? What I would really like, I said
yesterday, I think we need more data. I think we need population data. One of the major
areas where we’re doing mismanagement is when you see the variant that’s been seen
in cases from a particular ethnic group. It’s assumed to be pathogenic because it’s been
seen several times from a particular ethnic group, but the equally plausible and I would
say more plausible thing is that that is a population-specific polymorphism. And unless
we have really good data on these low-hanging fruit key genes from lots of populations,
we are going to keep making that mistake. And I’m not clear that we’re just going
to get that data from unless we directly try and get it.
I think it also does give you better information about the spectrum of mutations across there.
And because [unintelligible] into our old genes, there’s virtually no sequence data
from control individuals in BRCA1 and 2 because it was sort of done before, and it’s still
quite a herculean enterprise, but it certainly was then.
We clearly need much, much better predictive in silico methods. That’s a given. What
I’d really like to see is some standards for deciding for us as a group or some group
to decide if genes or variants are definitely pathogenic, so really thinking about which
ones we classify into that group. I appreciate that all the others we need to sort out as
well, but at least if we can sort that out.
And I want to just finish. On a sort of cautionary tale to exemplify that, so RAD51C was a gene
that was published in “Nature Genetics,” very high profile, about 18 months ago. And
it was published as a high penetrance breast/ovarian gene comparable to BRCA1 and 2. It’s got
[unintelligible], BRCA3. I’m aware of one woman who’s had prophylactic mastectomies
because they have a mutation. It was a little unusual, this paper, because in it, the truncating
mutations were only found in breast cancer cases if they were in families with both breast
and ovarian cancer. And if you looked at families with just breast cancer only, there were no
mutations. And that’s obviously odd for a high penetrance breast cancer gene. Subsequent
data, nobody was finding any truncating mutations, but they were finding missense mutations.
So they thought, all right, the missense mutations are causing breast cancer. That wasn’t the
original hypothesis, and no controls were studied.
For a variety of things I haven’t time to go, and we were working on another gene which
is an ovarian cancer gene. Anyway, we’ve looked at this, and this is hopefully shortly
going to be published, but it is also consistent with all of the available literature, including
the original paper I should say. So we did a large case control resequence study looking
at both cases and controls. RAD51C is an ovarian cancer gene. It has a decent size risk of
ovarian cancer, about nine percent lifetime risk, but it doesn’t predispose to breast
cancer. So what’s going on? I think it’s fairly old fashioned to ascertain [unintelligible]
bias. If you’ve got a gene that causes phenotype A, in this case ovarian cancer, and you look
at phenotype B, but the -- which is breast cancer in this case, but the only breast cancer
cases you’re looking at are relatives of your phenotype A, you are bound to see an
association unless you correct for that. But if you look up for phenotype B on its own,
which they did, you’re not going to see an association.
So 18 to 24 months after that first gene report, you know, the cancer is clarified. It’s
not a high-risk breast cancer gene. In fact, it doesn’t cause breast cancer. But you
can see a potential where if we’d had the whole genome data in the electronic patient
records and that data was already available and the gene comes out and you’re thinking,
right, it’s high-risk breast cancer. You can see there’s a situation where a lot
of women might have thought, right, I’m going to have prophylactic surgery or act
on that. And I think we’re in the era where there’s going to be lots and lots of new
gene discoveries. When you’re writing a paper, it is perhaps natural that one’s
optimistic about the clinical utility of that. And there’s often in that paper, but I think
there needs to be standards and that -- or principle by which people can look to see
whether it has passed a bar where it is ready to be used in the clinic. And I think we need
to think about that now. And I’ll finish then.
Female Speaker: All right, I think we used up our question
time, unless somebody has -- Heidi, do you have like a quick one?
Heidi: Just a quick question. You mentioned recommending
no management in cases where there’s a non-assignable BRCA1 variant. You know, we study cardiomyopathy
which is a very similar paradigm in terms of dominant inheritance and high risk for
sudden cardiac death and a need to manage those family members. And we would always
recommend familial testing to look at segregation in order to determine the clinical significance
of those VUSs. And once you get a high enough LOD score, you can say with statistical confidence
that that variant is pathogenic and then use it in a predictive fashion. So I’m a little
bit concerned about not encouraging those physicians to pursue familial studies which
can be extremely informative in terms of pathogenicity.
Nazneen Rahman: So, I think that with respect to those Class
3 ones, we certainly do do that. All the Class 3 ones, we then would recruit into our research
and we would do that more broadly. I think this does speak to things what should you
do with just any VUS, and it then also speaks to the fact that are they innocent until proven
guilty or guilty until proven innocent. Now, the one I talked to you about, in fact, all
the in silico things would say that it’s benign anyway. And we know that they’re
mostly benign. We’re not doing that for synonymous variants. So I think in terms of
going -- often you can’t do the segregation. The families are too small. The chance of
being able to really sort that out in terms of a LOD score for breast cancer, which has
got a high phenocopy rate, which also impacts in that so it makes it very difficult. It’s
quite difficult. So I think the problem that has as people go back to them and you’re
telling them, if people knew that it was innocent rather than proven guilty and they were doing
it on a purely exploratory fashion, I wouldn’t have so much problem with it. But I think
what happens is people are assuming that if the doctor’s telling you this or the doctor
is actually telling that it is likely to be pathogenic, that we have a problem there.
And so, currently, we’re doing more harm than good. So it’s part of that balance,
but I appreciate it’s a difficult issue.
Female Speaker: Great, thank you. And we’ll definitely have
time after the panel’s all done for more questions.