Tip:
Highlight text to annotate it
X
David Wheeler: I hope that's not the whole talk but -- [laughs]
So I want to thank the meeting organizers for this opportunity to speak to you today
on what is becoming a very intriguing story in the colorectal and endometrial cancer.
As you've heard from several of the other speakers during this symposium, we, of us
who are studying mutations, kind of have a Goldilocks principle where we don't want too
few mutations because we can't discern what's broken, but on the other hand, if we have
too many mutations, the background rates go way up, and it's then very difficult to discern
which mutations are causing a disease. So this is going to be a story about the ultra-high
mutation rate where it's very difficult to tell from one patient to another which are
the driver genes. But the polymerase that may be underlying that is turning out to be
very interesting.
So -- let's see. Okay. So I'll pick up where we had left off with the marker paper, the
colorectal cancer, where we're showing mutation frequencies here. The blue line are the frequencies
of non-synonymous mutations, and you can see that most of the patients have a low rate
between one and 10 mutations per megabase; they're microsatellite stable. And then there's
another group of patients that are recognized as hypermutated, that have also microsatellite
instability, as indicated down on the panel below. The microsatellite instability is associated
with a very high rate of MLH1 silencing through CpG island methylation.
And what we observed was that there is a small group of patients with the highest mutation
rates, these are mutation rates greater than 100 per megabase, that did not have microsatellite
instability and did not have MLH1 silencing apparent. And, in fact, the MLH1 track up
here shows that they weren't even mutated in MLH1. And interestingly, they all had mutations
in the polymerase E. And so polymerase E is one of the two major replicative enzymes that
replicates the genome at S-phase, and this came to our attention, and we dubbed these
ultramutated.
This came to our attention when we looked at all the mismatched repair systems across
these patients, and what's shown here are the different groupings of DNA repair genes.
The green are mutation frequencies in patients with greater than 100 mutations per mega base,
the ultramutated. The red are the hypermutated microsatellite instable. And the blue bars
represent the low mutation rate patients. And generally, you can see that all of the
-- or many of the genes are increasing in mutation frequency as you go to these ultra-high
mutation rate. However, interestingly, not all genes show that trend. In particular,
this blue bar here is P53, whose mutation rate actually goes down as you go to higher
mutation frequency.
So, interestingly, in the polymerases, there was a single gene that was mutated in all
of them. And, however, a single gene mutated at 100 percent where n=6 is not all that much
to write home about. But when we looked at the locations of those mutations in the polymerase,
we saw that they clustered mainly in the exonuclease domain. Now all of these mutations are only
from the ultramutated patients, and so there are more than six here, and that's because
some of these patients are mutated multiple times. And what was really intriguing was
not only the clustering in the exonuclease domain but the fact that S459F had been seen
twice. And as this was coming together, a paper came out from Japan by Yoshida and colleagues,
where they had discovered, in a single patient, the F367S mutation. And so -- which had also
been seen in this study. So this recurrence at these two sites in such a small dataset
seemed strongly suggestive. But again, with n=6, we had a very hard sell here.
So we had about 300 more patients to go in the colorectal project overall, and so we
went back to sequencing, hoping that we would see more of these. And so let me just mention
why Yoshida referred to this in a title of that paper as the proofreading function. These
polymerases have been studies to the last two or three decades in great detail. And
this shows the results of some experiments with the T4 polymerase. And in the exonuclease
domain of this polymerase, mutations are known to cause a very high rate of mutation. And
this mutator phenotype, as it's become called, has been seen in bacteria, now in yeast; it's
been studied extensively. Interestingly, mutations in the polymerase domain over here sometimes
actually improved the fidelity of the polymerase, and it's thought that the polymerase has to
slow down when the polymerase domain is mutated, and that lets the exonuclease domain operate
more efficiently. So what you have here is something like a modern day word processor,
where you're typing in Microsoft Word and you mistype a letter and push the space bar
and the error is corrected.
So we can also look at recent experiments in mice where the exonuclease domain has been
mutated and knocked out in mice, these mice are viable, and they die quickly. Here is
a polymerase E homozygous mutant; compared to the wild-type, they die must faster and
they're dying of cancer. The POLD1, which is the sister polymerase of POLE, also had
a very high rate of death. These are also dying of cancer. And elsewhere in this study
they show that these mice had a mutator phenotype.
So with all this together, we were very excited about this and went on sequencing, and this
now shows the results of sequencing across 500 patients in colorectal cancer. And you
can see that we have replicated mutations in sites we saw previously at P286R with a
variety of different amino acids at that position, V411L. Here is the F367S which we haven't
seen again. And then the S459F, which has not replicated.
So our colleagues at Memorial Sloan-Kettering who are working on this with us also worked
on the endometrial paper where -- or the endometrial project where they also have microsatellite
instable patients with hypermutator phenotypes, and so they were able to quickly confirm that
this same phenomenon occurs in the endometrial cancer. So here you see replication of P286R
and V411L across their patients. And these patients have the same hypermutated -- ultramutated
phenotype. In addition, these ultramutated patients show a very dramatic skewing in the
relative frequencies of CA mutation, relative to the hyper and the low mutation rate microsatellite
instable.
So we don't know for sure the origin of this yet. The mutations that we see in the patients
are a combination of mistake by the polymerase and whatever replication -- sorry, whatever
repair processes are going on. So we don't know for sure whether this just results in
a -- results from a inefficient repair of this kind of mutation. However, there are
some early suggestions now from further work by Nils and Niki at Memorial Sloan-Kettering
that different mutation hotspots lead to slightly different frequencies of these mutations,
suggesting that it might be arising from the enzyme itself.
So this is then the score card in colorectal cancer, and so what this shows, that across
our microsatellite stable low mutations rate patients, of which there are 412 in colorectal
cancer, only four of them have mutations in POLE. None of those are in the exonuclease
domain and none at the recurrent sites. In the hypermutated, we actually get 19 mutations
in the hypermutated, three in the exonuclease domain, but none of those in the recurrent
sites, and this suggest that these are just passenger mutations resulting from the hypermutated
phenotype. And then in the ultramutated, across the whole molecule, we have 23 mutations,
which is more than the number of patients because there are multiple mutations per patient;
100 percent in the exonuclease domain, and about 80 percent in the recurrent sites. The
phenomenon looks very similar in endometrial, except that in endometrial, the frequency
of microsatellite instable is higher, and the frequency of the ultramutated is correspondingly
higher. But we come down to very close to 80 percent of the patients with mutation in
the recurrent sites.
So this shows the -- that there is actually detectible similarity between T4 phage and
human POLE. These systems are -- for DNA replication was pretty much solved once in evolutionary
history and is now recognizably similar across all -- most species, and not only the polymerases
but the other components of the replication machinery. So this enables us to easily map
the mutation locations onto a x-ray crystalography structure of T4 polymerase, and the gray domain
here is the polymerase domain, you can see the double-helical DNA moving through here.
This purple-ish is the exonuclease domain, and the red and yellow are amino acids that
are mutated in our dataset, and so they're all clustering in this one area of the exonuclease
domain.
So earlier I mentioned that there are actually two polymerases, one is POLE and the other
is POLD; these have been known for decades. And over the last five or six years, studies
in yeast, where yeast origins of replication are well-known, have knocked out the exonuclease
domain. And looking at the skewed ratios of mutation arising from POLE mutants in yeast,
researchers have been able to show that the POLE is -- functions on replication of the
leading strand. And likewise experiments with mutation in POLD show that POLD functions
on the lagging strand. So there's this asymmetry in function of these two polymerases.
So there's been a recently published collection of origins of replication in human, and so
Nils Weinhold looked at the mutation skewing in our polymerase e mutants, which are effectively
our yeast experiment in our tumors, and found a 60:40 bias in CA on the leading strand,
suggesting that the POLE is operating on the leading strand even in humans. This would
be the first time that human have replicated this -- well, what's known in yeast, although
it's widely assumed to hold in yeast as well -- hold in human as well.
So just to remind you of this high mutation rate again, and cancer in the mice, these
cancers are different from POLE and POLD. In the mouse, the POLE mutants lead to primarily
intestinal cancers. So that kind of leads to the question, well, what about POLD in
these cancers? Do we ever see POLD mutated? And this oncoprint from the cBio portal shows
that for all these POLE mutants we never see a mutation in the nuclease domain of POLD.
So this suggests that the POLD may be required for some essential function in these tumors,
and though we don't know what that function is yet, this asymmetry is very intriguing.
We've also looked at the rate of mutation as a function of the expression levels, and
you can see that as expression goes up, the mutation rate goes down, suggesting that transcription
coupled repair is operating in these patients. When you look at the hypermutated, this line
is flat across the expression levels.
And finally, we are just getting the first look at progression free survival, and Doug
Levine showed this slide this morning showing that the patients that are ultramutated have
the best -- have a better prognosis than patients that are hypermutated. Similarly, in colorectal
cancer, the microsatellite instable patients are known to have a better prognosis, and
so now it's become a very urgent question to find out whether this is a generalizable
feature, that high mutation rate leads to better prognosis.
So, in conclusion, the rare exonuclease mutation in POLE leads to an ultra-mutator phenotype
in colorectal and endometrioid tumors. The ultra-mutator phenotype finds a new subtype
of these tumors that may have unique prognostic features and interesting biological properties.
And so at this point, we're gathering with our colleagues, Gordon Mills and Stan Hamilton
at MD Anderson, cohorts of patients that will be able to test -- that have outcomes that
we'll be able to verify what the prognosis is. The ultra-mutator patients exhibit a signature
of transcription coupled repair, and the absence of POLD1 mutators suggests that it may perform
an essential functional in this new subtype of colorectal and endometrioid cancers; maybe
that's transcription coupled repair, but it will be interesting to try to figure that
out. The strand-specific mutation pattern associated with putative origins of replication
in humans is the first suggestive evidence for confirmation of the yeast model of replication
in a higher eukaryote. And so we are now sequencing whole genome where we'll be able to look at
more origins of replication and get out of the transcribed regions where things could
be biased, and get a better look at this phenomenon.
So, with that, I'd like to thank all my collaborators, especially Nils Weinhold and Niki Schultz
at Memorial Sloan-Kettering, and the rest of the crew at the Baylor Genome Center, the
Wash U Sequencing Center, who sequenced the endometrial and many of the colon cancers;
my colleagues at the Broad as well. Thank you.
[applause]
Lou Staudt: One question.
Male Speaker: It was a very interesting fact that you didn't
see POL-delta mutations but you saw POL-epsilon mutations. And I want to bring up here the
analogy with yeast. Many of your samples, they actually aren't defective in mismatched
repair, either by MLH1 or by combination of MSH2 or MSA3, MSH6, and it is known in yeast
that POL-delta proofreading deficiency in combination with mismatched repair deficiency
just kills the yeast cell because of hyper -- extreme hypermutability. However, POL-epsilon
proofreading deficiency is, in combination with mismatch repair deficiency, is hypermutable
but still can live. So that may be another factor that can be included into all considerations
here, and we can talk about it later.
David Wheeler: Okay. Thank you very much.