Why me? The mutagenic origins of cancer for individual tumors and tumor types

December 01, 2022

Information

Yale Cancer Center Grand Rounds | November 29, 2022
Presentation by: Dr. Jeffrey Townsend

ID9206

To CiteDCA Citation Guide

00:00This is my it's my pleasure to
00:03introduce Jeffrey Townsend as
00:04the today's grand rounds speaker.
00:07Jeff is the Elio professor of
00:10Biostatistics and Professor of
00:11Ecology and evolutionary biology
00:13and the Co leader of the Genetics,
00:15genomics and epigenetics program
00:17at Yale Cancer Center.
00:19He received his PhD in organic chemistry
00:23and evolutionary biology at Harvard
00:25University and in 2019 received
00:27the prestigious membership in the
00:29Connecticut Academy of Sciences and.
00:31Engineering for his work in
00:33developing innovative tools for
00:36to study population biology,
00:38including evolution of
00:39antimicrobial resistance,
00:41disease evolution and transmission and
00:43evolution of of tumor biology tumorigenesis.
00:46His research enabled curtailment of
00:49pathogen evolution outbreak mitigation
00:51and used to inform therapeutic
00:54approaches in cancer metastasis.
00:56So in recognition of his
00:59prominence in the field,
01:00in 2021 Jeff was selected as the Co
01:03Chair elect of the Cancer Revolution
01:05Working Group by the ACR and his
01:07lab is currently working on on many
01:10projects including buying developing
01:12bioinformatics tools for cancer
01:14genetics epigenetics epidemiology.
01:17And nonlinear mathematical
01:19models of disease epidemiology.
01:22So it's my pleasure to give the
01:23podium to Jeff and we look forward
01:25to hearing your your presentation.
01:29Thank you House for that wonderful
01:32introduction and thank you and and
01:35and Ken for the encouragement to
01:38present today for this audience.
01:40And thank you all for basically the
01:43opportunity to present the kind of
01:45work that we've been doing in my lab.
01:47The title of. My talk is why me?
01:51The mutagenic origins of cancer
01:53for individual tumors and tumor
01:56types and I'm going to spend some
01:58time talking about that title.
02:00But first let me just go
02:01by my disclosure slide.
02:02I have done consulting for Black Diamond
02:05Therapeutics and Agios Pharmaceuticals.
02:08And so this title, why me?
02:11Just was inspired by the fact that
02:14as I started working on this work
02:16originally largely with Vincent Kantaro,
02:18who you'll see a picture of later,
02:20we realized that what we were doing to
02:22try to understand just what individual
02:24variants were contributing to cancer.
02:26Actually to some degree and the
02:28degree to which it addresses it,
02:30I'd love for you to think about,
02:32as I give this talk answers the
02:34question for an individual patient,
02:36what the causation of their individual.
02:38The answer was and I'll go through
02:40a lot of detail about that,
02:41but that that gets down to
02:43the mutagenic origins.
02:45Again not the physiological origins
02:47but mutagenic origins of cancer for
02:49individual tumors and tumor types.
02:51And I think this is a very it's obviously
02:53of interest to anyone who studies cancer,
02:56what the mutagenic origins of cancer
02:58are and certainly of interest in one
03:01way or another to patients who have
03:03have a have come down with cancer.
03:08It has been widely reported that one of
03:10the most difficult questions that patients
03:12and doctors struggle with upon diagnosis
03:14of cancer is the question, why me? Why?
03:16Why was I struck with this ailment?
03:19And it's natural for patients to want
03:21to understand the causes behind their
03:23calamities, and it's difficult to hear only
03:26statistics and probabilities as a response.
03:28So the traditional way that you answer
03:30this question of why me is to say,
03:31well, did you smoke that elevates
03:33your process probability.
03:34Do you have this genetic predisposition
03:36that elevates your probability?
03:37Um, you know, how old are you?
03:41What is your ethnic background?
03:43There's lots of different predictors
03:44for whether or not someone
03:46might come down with cancer.
03:47But those aren't answers about why
03:48you came down with your cancer.
03:49Those are answers about
03:51generalizations about your life.
03:53So to date,
03:55these statistics and probabilities are
03:57nearly the only answer that science
03:59and medicine has been able to give.
04:02And one answer that's sort of
04:04straightforward and obvious and if you
04:07are proponent of sort of the genetic
04:09evolutionary model of what makes
04:11cancer happen is that mutations happen
04:13and that's why you have your cancer.
04:15It's a very general answer,
04:17though it's not terribly satisfying,
04:20but it can be broken down into a
04:21lot of different kinds of mutations.
04:23So there are clock like endogenous
04:26mutations and processes that fuel
04:28mutation throughout the body over a lifetime.
04:31So as your body ages,
04:33you get these mutations that happen
04:36simply because the cellular processes
04:38that reproduce your DNA are not perfectly
04:41designed to reproduce it perfectly,
04:44and they can't be just because of
04:46the third law of thermodynamics.
04:48So they're endogenous processes
04:49that fuel mutation in your
04:52body throughout your lifetime.
04:53There are also mutational processes
04:56that are fueled by exogenous sources,
04:58such as viral infection
05:00inducing applebach activity.
05:01So viral infection can cause your
05:03cell to react in certain ways,
05:06maybe for cellular defense.
05:08And those mutations that are
05:10brought about as sort of secondary
05:12consequences of your response to viral
05:15infection can also lead to cancer.
05:17And the third category is exogenous
05:20mutagenic sources such as tobacco
05:21smoke that may affect your lungs.
05:24For your head and neck or UV
05:26radiation that can affect your skin.
05:29So these are all sources of mutation that
05:31we know about and probabilistically we
05:33can tell patients about the fact that,
05:35you know,
05:35exposing yourself to a lot of sun
05:37may increase your risk for Melanoma.
05:38If you have a Melanoma,
05:39it may be partly due to the fact that
05:41you exposed yourself to a lot of sun
05:43light at some point during your life.
05:46Now one of the I think quite
05:48revolutionary sort of discoveries
05:50of the of recent times was that we
05:53can actually trace all a lot of
05:56those sources I should say certainly
05:58all the sources I just mentioned,
06:01but many others as well to when we
06:04when we sequence a tumor for instance
06:07we can trace signatures of those different
06:10sources in the DNA mutations that happen.
06:13So certain DNA mutations are more frequent.
06:16And I'll explain this in
06:17more detail a little later.
06:18Certain mutations are a little
06:20more frequent when you have a UV
06:23mutagenesis and other mutations are
06:24more frequent when you have just
06:26simple aging processes, etcetera.
06:28And it turns out that there are
06:30enough mutations in the typical
06:31tumor that you can do a sort of
06:33machine learning deconvolution.
06:34And I won't go into the detail
06:36about that to sort of figure out
06:38for a given tumor what were the
06:40different sources that contributed
06:41these mutations and this is really,
06:44really extraordinary.
06:46That we can figure that out the one and
06:49and just to give you a little bit more
06:51of a a sort of a a more detail on that.
06:54So here's S1 which is typically it's
06:56called emanation of five methyl cytosine
06:58and that's considered to be sort of
07:01an endogenous aging process that sort
07:03of occurs without any particular cause
07:05other than our other time passing
07:08for our body through development.
07:10S2 is is one of two signatures
07:13that we associate with apobec
07:15activity there's defective.
07:16From August recombination DNA repair,
07:18which may be mutation based
07:20and therefore endogenous,
07:22but related to a very specific
07:23process that might be treatable
07:25tobacco smoke which you can see of
07:27course largely affects lung cancer,
07:29but you can also see some for for liver,
07:34head etcetera, kidney,
07:35there's some other sources tobacco smoking.
07:38So an S5 which also you see
07:40is large circles here.
07:42That's another signature that has
07:44been traced essentially to aging.
07:46Processes.
07:47Although it's a little less well understood
07:49what the what the underlying basis of it is,
07:52it's very clear that age is highly correlated
07:56with the amount of SS5 mutation you get.
07:59Defective DNA,
08:00mismatch repair,
08:01ultraviolet light etcetera.
08:02And you can see these distribute themselves
08:03differently for different types of cancer.
08:05And so again this is very consistent
08:07what we knew already in a lot,
08:08very consistent with what we generally
08:10did which was say predictably like
08:11if you have a lot of exposure to sun,
08:13you're more likely to get UV exposure
08:15and that UV exposure then is more
08:18likely to translate to mutations that
08:20that may or may not cause Melanoma.
08:22But but once you have those mutations there,
08:24you know they may, they may cause that.
08:26So this is great we've we've got, we've got.
08:28The ability to see the,
08:29the sort of the trace or exposure
08:32in cells to these mutagens.
08:35The,
08:35the one thing that's missing though is,
08:38is that the extent to which each of
08:41those processes actually contribute to
08:43tumorigenesis still remains unknown.
08:44So we can look at what mutations
08:46are in the genome.
08:47But if I count up mutations in the genome,
08:49here's one, here's one,
08:49here's one, here's one.
08:50That doesn't tell me how much of those,
08:52each of those mutations are actually
08:54contributing to tumor Genesis.
08:55In fact, most of those mutations are
08:57not contributing to tumor Genesis.
08:58And most analysis find that there's only
09:00a few mutations that are contributing
09:02at a significant level sort of at
09:04this SNV single nucleotide variant,
09:06level two to two tumor Genesis.
09:10So we really need to have more in our,
09:13you know, another tool in our
09:15plate to figure out.
09:17What the level each of these
09:19endogenous and exogenous processes
09:21are contributing to a given cancer,
09:23and here's just a schematic
09:24for this right you know,
09:26so the mutation 1, mutation 2,
09:28up to mutation N However many there
09:30are that are really affecting cancer,
09:32they can cause increased cellular
09:34proliferation and survival.
09:36And sunlight may be contributing to UV
09:39radiation may be contributing to some
09:41of those mutations more than others
09:43because certain mutations are caused by
09:45sunlight and other ones are not similarly.
09:47Aging may contribute to some of
09:49those mutations more than others.
09:50And what I've got right here is,
09:52you know,
09:53if you take nothing else from this lecture,
09:55this is the main thing that I
09:57want to emphasize is that there's
09:58sort of two stages to this.
09:59One is, you know,
10:00what mutagens have you been exposed
10:02to and contributing to the set of
10:04mutations that are causing your cancer?
10:06And the 2nd is how much do each of those
10:08mutations actually contribute to the
10:11increased cellular proliferation and
10:13survival that is the phenotype of cancer.
10:17And there's a way to figure this out.
10:20But to figure it out we we need
10:21to sort of deconvolve something.
10:23And this is an old idea and I'm
10:24going to go through it in some
10:26detail just to make sure that it's
10:27clear to everyone that cancers are
10:29the outcome of an evolutionary
10:31process that's driven by mutation,
10:33consequent genetic variation
10:35created by that mutation,
10:36and natural selection for
10:38the more oncogenic variants.
10:40This is from Peter Knowles
10:421976 science article,
10:43a very well known article where he
10:45just went through the idea that,
10:46you know, it's an evolutionary.
10:47Process that actually produces
10:50malignancies and in this depiction
10:53you can see a cellular lineages
10:57differentiating and dividing.
10:59You see a lot of lineages that are
11:01hashed out here meaning they go
11:03extinct and that's the selective
11:05process in operation.
11:06You know most of our our cells
11:08are all dying at the same rate as
11:10we're as they're dividing typically
11:12as in as an adult.
11:13So there's a lot of death going on we
11:15don't usually emphasize that but but.
11:17So that death may be going on and
11:19what happens is that at some point
11:21you get lineages that are reproducing
11:22a lot more than they are dying.
11:24And those ones,
11:25in the case that they cause
11:28difficulties for your life are
11:30usually referred to as malignancies,
11:32especially if they can then
11:34migrate to other locations.
11:35And this.
11:37So these later evolved lineages are
11:40usually the product of a series
11:42of mutations that come along
11:44during this evolutionary process
11:46and what's happening with those.
11:48Patience is they're actually enabling the
11:51cells to survive and proliferate better,
11:53so they're selected as the terminology
11:55we use in evolutionary biology,
11:57and they persist.
11:59And that arising of those mutations
12:02within individual cells within
12:04cancer lineages is what we need to
12:07sort of understand because there's
12:08two processes going on here.
12:09One is the appearance of these
12:11mutations and then there's the amount
12:13that they actually increase the
12:14survival and replication of the cells.
12:16So to quantify cancer effect size,
12:19which is what I typically call
12:20this the effect on,
12:22you know,
12:22on cells of actually leading to cancer,
12:25which in evolutionary biology we
12:27just call a selection coefficient.
12:30We need to understand what the
12:31prevalence in a population,
12:33patient population is of a tumor
12:34and we need to deconvolve that
12:36prevalence into two factors because
12:38when we see a certain mutation
12:41very commonly in a kind of cancer,
12:43that doesn't mean it's causing
12:44a lot of the cancer.
12:46It may just be that the mutation
12:47rate is very high and we've seen
12:49that very frequently in instances
12:51where we have genes that are very
12:52large or have very high mutation
12:54rates that show up frequently
12:55when we sequence tumors,
12:56but are not significant causes of cancer.
12:59And so we really need to understand,
13:01you know,
13:01which ones are actually contributing
13:02cancer and which ones are
13:03just typically contributing to
13:04prevalence because of an
13:05underlying mutation rate.
13:06So to quantify the cancer effective size,
13:09we have to do a fairly straightforward thing,
13:11which is take that prevalence,
13:12how frequent we see them in
13:13patients and deconvolve it into
13:15the baseline mutation rate.
13:16How frequently the mutations are
13:18occurring in the lineage and into
13:21the degree of selection for that
13:23mutation in the cancer lineage.
13:26And if we can differentiate those two things,
13:28then we can better understand how
13:29much is because how much is that
13:31mutation there is because of the
13:33underlying mutations that are happening
13:34and across your whole genome that
13:36aren't necessarily relevant and how
13:37much is due to those individual
13:39mutations actually increasing the
13:41proliferation and survival of the cell.
13:43So here's just a schematic of that.
13:45This is just basic evolutionary biology.
13:48one-on-one mutation creates variation
13:50symbolized by the different shades
13:52of Gray there unfavorable mutations
13:55are selected against.
13:56Reproduction and mutation occur,
13:58and the favorable mutations are more
14:01likely to survive and reproduce,
14:03and the point of this is that it
14:05both the mutation rate and the
14:08extent to which they contribute
14:10to survival and reproduction.
14:13Contribute to what you see at as an
14:16end product of the process of cellular
14:19differentiation, especially into cancers.
14:22All right.
14:23So how do we figure out
14:24that baseline mutation rate?
14:25Well, it's already been done for me anyway.
14:28It was a lot of the work was already done,
14:29which is really great by by
14:33Lawrence and and others.
14:35This is a 2013 paper quite a
14:36while ago where they showed that
14:38mutation rate varies widely across
14:40the genome and correlates with DNA
14:43replication time and expression level.
14:45So there's these covariates.
14:45I'm not going to go into a lot
14:47of detail about this.
14:48I've talked about this before
14:50with various audiences here, but.
14:51That mutation rate varies and correlates
14:54with DNA replication time and
14:56expression level with heterochromatin marks.
14:58A bunch of other correlates that we can
15:01actually get about individual tumors.
15:03Those allow us to ask questions
15:04about you know a given gene and
15:06whether or not it's got a very high
15:09mutation rate or a low mutation rate.
15:11By using those correlates to help us
15:13predict that along with synonymous
15:15changes in the genome which we can
15:17presume don't have any effect on the
15:20proliferation and survival of of cells.
15:22So for instance,
15:23olfactory receptors,
15:24which early on were this bugaboo that would
15:26show up when we did these tumor sequencing,
15:28happened to be in a part of
15:29the genome that gets a very,
15:30very high mutation rate.
15:31It's basically not expressed
15:33and not expressed.
15:34Parts of the genome don't have
15:35transcription enabled repair,
15:36etcetera.
15:37CSMD 3 is another example where
15:39there's very high levels,
15:40high correlates and also very
15:42high mutation rate.
15:43And typically it's not considered
15:44to be a driver even though you see
15:46it a lot in cancer tumor sequencing
15:48and you can do regressions on
15:49this and then I'm just going to
15:51very quickly mention that.
15:52This wonderful work was done by Lawrence,
15:54but typically that work was only applied
15:57to the question of whether or not
15:59genes were overburdened with mutations.
16:01So in other words, they got these
16:02mutation rates and they just said,
16:03well, is it more than we expect.
16:05And then they calculated P value for
16:06whether we should put this gene in the
16:08category of mutated or not and then
16:09they leave behind that mutation rate
16:11and then just look at prevalence in
16:13most of the analysis that were done
16:14from 2013 through 2018 or so. So.
16:18So typically that was sort of left
16:20behind at that point and that's what.
16:23Vincent Kintaro and I in 2018 sort
16:24of picked up on and said look,
16:25this mutation rate is more important
16:27than for just calculating P values.
16:28It's actually important for
16:30calculating the effect.
16:31You know in the biostatisticians mind
16:33P value is sort of a secondary thing.
16:36First you calculate the effect of
16:37the thing you're looking at and
16:39then you calculate that you see
16:40whether you should trust that effect.
16:42And so that's what Vincent Cantara
16:44and I did and just here's a sort of a
16:46brief introduction to how we do that
16:48calculation by convolving the gene
16:49based rates from the silent sites and
16:51covariates with they're trying to die.
16:53Context.
16:53So you can just go through tumor
16:55sequence data and you can look at
16:57what the underlying mutation rate
16:59is using basically that Lawrence at
17:01all approach that I talked about
17:02with the covariance,
17:03you can sort of look at every gene
17:04in the genome and calculate what
17:06the mutation rate is.
17:06And this is just one of these plots
17:08that's just scatter plot on one axis
17:10of what the different gene rates are.
17:11And you can see there's quite a
17:13wide range here.
17:14And I think that's the most important
17:16lesson of this little image is that
17:18the mutation rate varies quite
17:19extensively from gene to gene from 10
17:21to the minus two to 10 to the minus 4.
17:23In this particular instance,
17:25so that's two orders of magnitude
17:27rate variation in mutation rates.
17:28So when you see,
17:29you know one gene mutated in a
17:32cancer tumor pop cohort at 100,
17:33you know,
17:34100 copies out of 1000 and another
17:36at 10 out of 1000,
17:37that's only one order of magnitude
17:39difference in prevalence and you
17:40can explain that by just half
17:41of this mutation rate diagram.
17:43In other words,
17:43mutation rate can explain a lot of
17:46the differences in how prevalent genes
17:48are when you look in a patient population.
17:50So you shouldn't take that
17:52prevalence as an indicator.
17:54As a strong indicator of how
17:56important a gene is in the cancer,
17:58you really need to basically understand
18:01this underlying mutation rate as well.
18:03And so then you can take different
18:06genes that are on that that diagram
18:09and you can look at each individual
18:11tumor and you can map out what
18:14the trinucleotide rate rates are.
18:15So. So this rate is,
18:16the rate above is just the rate at
18:18which the gene itself gets mutated.
18:20But if we want to look at every given site,
18:23the important thing is that the
18:25different mutational processes that I
18:27mentioned earlier in this talk affect
18:29different sites at different frequencies.
18:31Have a question right there.
18:342nd normalized for length.
18:37Is the mutation rate itself?
18:39In this case it is, yes.
18:43So, so, so these different mutational
18:46processes contribute to differently.
18:49So in this case,
18:49I'm looking at lung cancer,
18:50which is why we can be carriers
18:52and EGFR highlighted here.
18:54And in lung cancer,
18:55you get a lot of these RCA mutations
18:57that are preceded by a T and
18:59followed by an A and also ones that
19:02are preceded by and followed by
19:03an A&amp;C and an A and an A and an A.
19:05So, so, so all of these bright red
19:08trinucleotide context get much
19:10more mutation than other ones.
19:11And again I just want to
19:13emphasize that the coloration.
19:14Here is scaled to how often we see it.
19:16And so you see almost an order of magnitude,
19:19sometimes more,
19:20with some cancer types of variation,
19:22again in how frequently given
19:25sites get mutated over other sites.
19:27So when you combine this
19:29plus the gene by gene rates,
19:30you're talking about 3 orders of magnitude,
19:32maybe even four in some cases,
19:34between a given site and
19:36another site in the genome,
19:37and how frequently gets mutated.
19:38So this is a really important
19:39factor to take into consideration
19:41when wondering whether or not
19:42a given site is important for.
19:44Driving cancer and what you can
19:45do is you can basically tape this
19:47map and look at each gene and
19:49here I've just look,
19:50I'm looking at like an excerpt of a
19:52tiny little part of the of the genome.
19:54Sorry.
19:55This is this is site 850 to 870 and EGFR,
19:59here's site 1 to 20 in K Ras and here's
20:03site 30 to 50 in cutting and B1.
20:06And I just want to mention that if
20:07you you know you take these rates
20:09and then you make sure that the
20:10individual site rates are accommodated
20:12by ensuring that you know TCA is much more.
20:15Frequent then see CCG chaining 2
20:19and A and and and do all of the math
20:22that's very straightforward here
20:24but a bit of a lot of accounting
20:26bioinformatics ally and then map it
20:29through the the actual genetic code.
20:31So you're looking at every single
20:32site in that protein and saying well
20:34how likely is this 850 histidine to
20:36change based on its three code on
20:38sites into a tyrosine or a proline
20:41or a phenylalanine etcetera etcetera.
20:42And some sites of course some
20:44changes of course can't really
20:46happen through a single.
20:47Nucleotide mutation,
20:48others can in multiple ways, etcetera.
20:50So there's a lot of addition to add up here.
20:52But once you add it all up,
20:53this diagram tells you how likely
20:55each different change is to happen
20:57by neutral mutation.
20:58That is when we just expect new
21:00mutations to be sprayed on there
21:01and have no difference in the
21:03replication and survival.
21:04So then we get this diagram of how
21:06much each amino acid position would
21:08be expected to be mutation mutated,
21:10and then we can compare that
21:12to what's actually observed.
21:18Um, what's actually observed is much,
21:22much more rarified set of mutations
21:24than what you actually expect
21:25based on neutral evolution.
21:27And that's because when we sample tumors,
21:29we're sampling tumors that have been
21:31under selection for very specific
21:32mutations and because right here I've
21:34selected sites that actually do have an
21:37effect on proliferation and survival.
21:38So on the top EGFR 858,
21:42Lucine is a very well known mutational site.
21:45The KSG 12 is also a very well known one.
21:48And then this.
21:49Part of continuing 1B1,
21:50it's a domain that is known to be
21:53oncogenic when it gets mutated slightly
21:54lower level in terms of the others.
21:56But the whole region across here is sort
21:59of known to be important to to oncogenesis.
22:02And So what you can basically do is
22:04take the prevalence that we see and and
22:06this is in a very crude terms but and
22:09there's some corrections that are involved,
22:11I'm not going to go into but essentially
22:14divide the expectations the observed
22:16here by the expected block on the same.
22:19On the same plot on the left and that
22:21gives you a metric for the cancer factor.
22:23How strongly that that site
22:25is mutated that sorry,
22:27how strongly that site is
22:29selected once it is mutated.
22:32And as I said these are well known
22:34sites in these particular cancers.
22:36And if you do that across all the
22:38different sites that you can look at
22:39what you see is a is a distribution
22:40that looks like this where on the
22:42X axis is the cancer effect size.
22:45It ranges from 10 to the zero to 10 to the
22:476th maybe even a little bit more typically.
22:49And why is that?
22:50Why,
22:51what does this range mean?
22:53The range is what it is because
22:54that has it's it's complicated and
22:56I don't want to go into a lot of
22:58detail but population genetically
22:59it has to do with the population.
23:01Size of the cancer,
23:02the reproductive population size.
23:03How many cells in the cancer
23:05could possibly reproduce?
23:06I'm not going to go into more
23:07saying about that,
23:08but that's why it exists across
23:10this wide range.
23:11The density here is just I'm just going
23:13to density a plot across cancer effect
23:15size of these different mutations.
23:16So most of the mutations lie at this
23:18very low range where it's not even
23:21clear necessarily if they're under any
23:23selection below say 10 to the four or so.
23:26And in blue I show you the
23:28synonymous mutations and in red
23:30the non synonymous mutations.
23:31So there's just a slight,
23:33a slight bias over the synonymous
23:36mutations of NONSYNONYMOUS
23:38mutations to be oncogenic.
23:39But the really important mutations
23:40are all out on this tail here,
23:42and I've just shown 2 here for reference.
23:44Here's a P53 mutation that's quite common.
23:46Here's an NF2L2 mutation is quite
23:48common in lung squamous cell carcinoma.
23:51So these tail mutations are the
23:52ones that are important.
23:53And this harks back to
23:54what I was saying earlier
23:55when we say, oh, lots of mutations
23:57are happening in the genome because
23:59of UV light or something like that.
24:01If they're not these key
24:03mutations out here on the tail,
24:04they're not contributing much to cancer.
24:06So we really need that component to be
24:09included if we want to ask the question
24:12what is causing cancer in an individual.
24:14Tumor in a digital patient.
24:17You can do this diagram
24:18not just for lung cancer,
24:19but for lots of different cancers,
24:21and we see very much the same pattern.
24:31OK. Now just to provide you a little
24:33bit of perhaps validation that this
24:36cancer effect you know is meaningful,
24:38probably many of you are
24:39familiar with GLENVAR variants,
24:41variants that have been attributed over
24:44time with some clinical significance.
24:47And these by the way these are Clint
24:49Barbarians that were attributed significance,
24:50not potential, not ones that
24:53weren't attributed significance.
24:54And on the X axis we've sort of divided them,
24:58those Clint Barbarians up and
24:59some categories I'll talk about,
25:00but on the Y axis.
25:01Is the scale selection coefficient,
25:03and generally there's basically
25:052 comparisons.
25:06I really want to emphasize here.
25:07If we look at glenvar single nucleotide
25:09variants that are recurrent within
25:11cancer type and compare it to other
25:14single nucleotide variants that
25:15are recurrent within cancer type,
25:17we see that the GLENVAR variants have a much,
25:20much higher distribution of selection
25:21coefficient than the ones that are other SNV.
25:24So in other words, there's,
25:26you know this literally,
25:28this is saying that Glenvar
25:30predicts cancer effect.
25:31But the opposite is true and I'll
25:33show you that in the next slide.
25:34And then we can also compare Glenvar STD's
25:37that are a single hit within a cancer type.
25:40That is ones that we only see once when
25:43they're clean bar single nukite variance
25:46versus other SNB's that are single hit.
25:48And you can see that the cancer
25:51affect size of those ones that are
25:53you know known oncogenic are believed
25:56oncogenic variants have a much higher
25:58cancer effect than the ones that
26:00are not believed to be oncogenic.
26:02And this is a highly significant from
26:04a a statistical science point of view.
26:07By the way, this is work of Jeffrey Mandel,
26:08who's sitting over here in the audience,
26:10a grad student in my lab.
26:12And so that should be reassuring.
26:17Furthermore,
26:17if you take the mean or the top
26:20cancer effect of a given variant,
26:23they're much stronger predictions of
26:25glenvar status than the SIFT score,
26:28the Polyphen 2 score,
26:30or variant prevalence.
26:31Any of these measures that are
26:32typically used to try to say whether
26:34a variant is important or not.
26:36So really,
26:36you should be using cancer effects if you
26:38want to know whether variance important.
26:41This is also work by Jeff Mandell.
26:45OK, so hopefully it persuaded you that
26:48cancer effect is a measure that you
26:50should be thoughtful about and use
26:52and and in the research you're doing.
26:54But what we wanted to get to from
26:55the beginning of this talk was
26:57the extent to which each of those
26:58processes contribute to tumorigenesis.
27:00So if you'll if you'll at least walk
27:01with me on the idea that cancer affect
27:03quantifies the degree to which a given
27:05variant contributes to tumorigenesis,
27:07then that apply that gives us the
27:09key to finish that association.
27:11I said. So we know, you know,
27:14from Alexandra's work.
27:15The degree to which, no, sorry.
27:17We know from this work the degree to which
27:19mutations contribute to the increased
27:21cellular perforation and survival.
27:23And we know from Alexandra's work and others,
27:26some strain in Xanal and others
27:28what the contribution of various
27:31mutagenic processes toward creating
27:33those mutations are.
27:34And so by putting those two things together,
27:36we can understand the relationship between
27:39these increased cellular proliferation
27:41and survival and the actual processes
27:43underlying these mutational effects.
27:46So just going back again to Alexandra's work,
27:49we know each signature contributes
27:51differentially to mutation counts
27:52observed in each cancer type.
27:53I showed this slide earlier and
27:55here's here's the slide where
27:57you can you can sort of like.
27:59Fade out for a moment if you want,
28:01and then come back in a moment.
28:02It's only saying.
28:03What it's it's this is I'm going to
28:06narrate through for those of you
28:07who are really interested how we
28:09actually calculate this process.
28:11But if if you've understood everything
28:13before, there's nothing new here.
28:14It's just the bookkeeping of how we
28:17calculate this process and the the
28:19point is that forget for each for each.
28:21A source of mutation.
28:23Here's deamination with age apobec tobacco,
28:26and then unload clock like signature,
28:28which were the four sources that
28:30came out of the deconvolution for
28:32a particular tumor in the TCA data
28:35set that turned out to be useful
28:37for illustration of this.
28:39For each of those processes,
28:40there's a weight of mutation that
28:42they contribute to given trying time
28:44nucleotides that are listed down here.
28:47So deamination with age really
28:49focuses on these AC to TG mutations.
28:52That's what they cause for the most part.
28:54But then there's a few other ones
28:55here that are quite frequent.
28:57Apobec really focuses on TCA or
29:01TCC or TCG or TCT changing to T,
29:05and tobacco has a broader distribution
29:08of neurogenic.
29:09In fact and this unknown clock
29:11like signature,
29:12there's another aging signature has a
29:15generally quite broad distribution as well.
29:18So we deconvolve that tumor into these
29:22different signatures to understand
29:23how much each one is contributing.
29:25That gives us a signature weight
29:27for every signature here.
29:29And I'm just emphasizing that,
29:30you know,
29:31we can do lots of uncertainty analysis by.
29:34Bootstrapping the signature,
29:35deconvolution,
29:35that's what all these dots are many
29:37bootstraps on and given tumor,
29:38just saying how much of that
29:40signature do we really believe
29:41is contributing to that cancer.
29:43So we do do that and then you
29:45can also and then in addition to
29:47understanding how much signature
29:48is contributed to cancer,
29:49we look at the probability that
29:51each source created each variant.
29:53And we know that because we know
29:54what the sources are and we can
29:55just look at the relative height
29:56of these bars essentially to give
29:58us the probability that each source
29:59contributed to a given variant and
30:02then that probability comes out of that.
30:04Just by multiplying those together
30:06essentially and that gives US4P53
30:08here KF5 and this odorant receptor
30:11which doesn't have much cancer effect,
30:13what the probably each source
30:15contributed to creating each variant.
30:17And then we take that effect size
30:18that I just described to you,
30:20which is very high for this P53 variant,
30:23quite a bit lower for KF5,
30:24but still there and is basically
30:26nonexistent for the odorant receptor
30:28mutation. So.
30:29So this is a really important variant,
30:31this is a less important variant
30:32and this is not important.
30:34Fall and then we can just sort of
30:36multiply through each variant by the
30:38probability that each source created
30:39that variant and that gives us this
30:42final thing which is the proportional
30:43mutation source effect size.
30:45That's a mouthful.
30:46But what we're just trying to say
30:48is how much of this given variant
30:51was caused by the particular
30:53mutational process and or sorry,
30:56how much of the selection for oncogenesis
30:57was caused by that particular process.
30:59So the TP50 were bar the TP 53 bars.
31:03Are much higher than the ones in
31:05KF5 and are those are way higher
31:07than anything in order receptor
31:09because the odorant receptor in
31:11fact doesn't do anything for cancer.
31:13So the average then you can then
31:16you can look across all of those,
31:19all of the variants,
31:20not just these ones to look at what
31:22the average attributable effect size
31:23is in a given tumor and you get this
31:25distribution which says oh for this
31:27tumor you know most of the oncogenic
31:30cause came from deamination with age.
31:33And for this tumor?
31:35You know the second most common process
31:37that was creating mutations that led to
31:40oncogenesis was this light Gray which
31:41is this unknown clock like SIEGENER 5.
31:43So.
31:44So this is a largely aging driven
31:46tumor and there's a little bit
31:48of Apple back here and a little
31:50bit of tobacco smoke and and you
31:52can follow it through like that.
31:53So this is one example for a
31:55given tumor and then that result,
31:57you know it basically tells you what
31:59at least with the knowledge we have
32:01right now what the effect size by
32:03mutational source for this tumor was,
32:05this is a lung cancer tumor.
32:06By the way.
32:07Now you can look at this not just at,
32:10you know,
32:11you can sort of understand that
32:13for a given site,
32:14but then you can also look at
32:18what a set of sites all look like.
32:21So this is just a diagram where we do that.
32:22Again,
32:23a little bit complex,
32:24but hopefully this everyone can
32:26follow along directly with.
32:28If you look across the genome,
32:29there's an average mutational weight.
32:31So tobacco smoke is causing a
32:33certain number of the mutations,
32:34certain proportion of the mutations
32:36and then a number of others.
32:37And in these diagrams,
32:38I've sort of put the major mutagenic
32:40cause on the left and then stacked
32:42all the other causes on the right,
32:44just because it helps you really
32:46see the differential effect
32:47of these different processes.
32:49So tobacco smoking is the major cause
32:52of of loads in general in terms of
32:56the underlying genomic mutation.
32:58But if you look at from site to site,
33:01each site has a different probability
33:03of being caused by tobacco smoke.
33:05So here's.
33:07KSG 12C very, very,
33:09very strong caused by tobacco smoke,
33:11maybe that's not surprising it in
33:13lung cancer, we see that variant very,
33:15very frequently.
33:16We very rarely see care SG12C in
33:19other cancers like pancreatic cancer,
33:20other K rosterman cancers.
33:22So why is that?
33:23Well, it's just because that site is
33:25hit a lot more in terms of mutations.
33:28It's not a doesn't appear to have anything to
33:31do from our calculations with its particular
33:34cancer effect relative to other variants.
33:37And, and in contrast, here's EGFR LA58R.
33:41It's a long known fact that you
33:43rarely see those in individuals who
33:45are non-smokers, are smokers.
33:47You see that in non-smokers and the
33:49reason is it's not caused by smoking.
33:51So when you see a patient with these Fr
33:55mutation, they typically are not smoker.
33:57There's a lot more individuals coming
34:00in with EGFR who are not smokers then
34:03are smokers relatively speaking.
34:05So you can do this for lung adenocarcinoma.
34:09You can look at other variants of
34:11course lung squamous cell carcinoma.
34:12Here you see PI3 kinase largely driven
34:15in lung squamous cell carcinoma by other
34:18effects, mostly apobec but in fact.
34:22Not at all driven by tobacco smoke.
34:23Again, that's an empirical observation
34:25that people have noted many times
34:27that individuals with lung squamous
34:29cell carcinoma who have PI3 kinase
34:30mutation are rarely are less frequently
34:32smokers than than other mutations.
34:34And P3 mutations, on the other hand,
34:37are diverse,
34:37some of them likely to be created by
34:39tobacco smoke, some of them less likely.
34:42OK, so we can look at this on an
34:44individual basis and then we can
34:45look at some other cancers.
34:46So here's bladder cancer and cervical cancer.
34:49I just added these because.
34:51Maybe this is a little less well known,
34:54but a lot of the mutation in both
34:56bladder cancer and cervical cancer
34:58is is caused at least by the.
35:00This deconvolution approach appears
35:02to be attributable to APOBEC mutation.
35:05Apobec is this apolipoprotein B.
35:09Gene that enzymatically we know
35:11mutates DNA and appears to be a
35:14viral defense protein.
35:15And what we see is that a lot of
35:18the mutagenic cause in the in
35:20the genome is created by apobec,
35:22some of it's by aging and bladder
35:24cancer and cervical cancer.
35:25There's a little bit of defective homologous
35:28recombination as a source there as well.
35:30But as you can see for a number
35:32of these mutations,
35:33the some P3 mutations for FGFR 3 for KSG 12D,
35:37we see almost no cause from APOBEC.
35:39But on the other hand,
35:40this other FGFR 3 mutation very
35:42likely to be caused by apobec,
35:45PI3 kinase,
35:45again very likely to be caused
35:47by APOBEC mutation.
35:48Cervical cancer are the same thing.
35:50All right,
35:50so we can look at the interval variance here.
35:52Let's get back to the main theme
35:55that this talk hopefully is.
35:57Presenting to you,
35:58which is that once we understand for every
36:00one of these variants what the causes
36:02are and how much they're causing cancer,
36:04we can then look at tumor
36:06causation by tumor type.
36:08And this isn't the best way to contrast them,
36:10I'll show you another that maybe
36:11contrast it a little bit better.
36:12But here we have all the different
36:15signatures on the Y axis and all
36:17the different cancers on the X
36:19axis and the red is the amount that
36:22the tumor type is caused by that
36:25particular signature and the Gray.
36:27It is or black is, the amount that you
36:29see mutation for due to that signature.
36:32And there's some big differences,
36:34say in signature 5 here for thyroid
36:36cancer where you see an enormous amount
36:39of cause but much less mutation.
36:42But it's a little hard to read that dot plot.
36:45Down below we have just these
36:47bar plots showing the can't,
36:49the weight of mutation.
36:51How much? Mutation was caught,
36:54which of the mutation in the genome was
36:57caused by a given mutational process.
36:58And on the right the effects and these
37:01may look pretty similar but let I'll
37:02show you the contrast that shows you
37:04how they're different in a moment.
37:06The thing I want to emphasize right
37:07now is we've given colors for all
37:09of those exogenous sources that
37:11may in principle be things that
37:13we could interfere on to stop.
37:15So UV light, defective,
37:18homologous recombination, presumably,
37:19maybe there be a way to do that,
37:21apobec perhaps if we.
37:23You know, avoided viral infection,
37:25tobacco certainly interventional alcohol,
37:28definitely something we can intervention on.
37:30Mutagenic chemical exposures definitely
37:32something we can do intervention on anyway.
37:34All those interventional ones are the
37:36colored ones and the aging ones are the
37:38Gray ones and then the unknown processes.
37:40The process is that we haven't figured out
37:42what they're associated with are in black.
37:44So this diagram actually tells you
37:46a lot about what you can do now
37:49to understand more about cancer,
37:50right because or to intervene we can
37:53intervene a lot on these cancer on these
37:55cancer types for which we see a lot of color.
37:58We there's much less we can
37:59do for the ones we don't.
38:01So for instance glioma very a lot
38:03of aging not a lot of other things,
38:05thyroid cancer, a lot of apobec,
38:07but other than that aging glioblastoma.
38:10Again, a lot of aging prostate cancer,
38:13a lot of aging,
38:14just a little bit of apobec and
38:17defective homologous recombination.
38:18So there's some we don't have much
38:20way to intervene on skin cancer,
38:22extremely easy to intervene to
38:25reduce the number of mutations,
38:27lung cancer, a lot of tobacco,
38:30a lot of defective,
38:31longest recombination and Applejack.
38:32So there's a lot we can do in terms
38:34of stopping those and then also where
38:36there's a lot more to understand.
38:38So for instance, breast cancer,
38:39ER 9 minus breast cancer.
38:40Like nearly half of the mutations,
38:42we don't know why they're being
38:44caused process wise.
38:45So this is something to be investigated
38:47because if we could figure it out,
38:48maybe there are these interminable
38:51processes that we could do something about.
38:55Etcetera. So you can sort
38:56of look at the black.
38:56That gives you an idea of how much
38:58we still need to learn and the Gray
39:00tells you and the idea of like,
39:01how much more.
39:04How much aging versus other processes
39:06seem to be causing that given cancer?
39:09And of course the cancers that
39:11are most age-related are at
39:12the bottom of this diagram,
39:13and the ones that are least
39:15age-related tend to be higher up.
39:18So this is just a bigger diagram of
39:21of that same picture in case Umm,
39:23you can see it better.
39:25And then I'm going to show you,
39:26I'm not going to show you
39:27the actual cancer types,
39:28but just an animation that actually
39:30Vincent Cantero made that helps
39:32you see the difference between
39:33the cancer mutation and effects.
39:35So this just varies between how
39:37much mutation is causing the given
39:39cancer and how much of the cancer
39:41affect by those mutations is causing
39:43the cancer and allows you to sort
39:45of see how different they are.
39:49You know, for ones like skin cancer,
39:52it doesn't change that much because nearly
39:53all the mutations are caused by UV anyway.
39:57All right. So as we said,
40:00as I said earlier,
40:01the extent to which the processes
40:02contribute determines tumor
40:03Genesis has been unknown.
40:04But now we can link it together
40:07with this process.
40:08And now I wanted to go back to
40:10this slide because I'm going to
40:11show you a bunch of diagrams and
40:13they're pretty complicated diagrams.
40:15But on the left hand side is going
40:16to be a bar plot that's respect
40:18reflecting like how much each process
40:20is contributing to the mutations.
40:22It's this left hand sign sign
40:24and on the right of each plot
40:26is going to be another bar.
40:27Plot that shows you how much each
40:29mutation is contributing to the increased
40:32cellular proliferation and survival.
40:34For four different cancers,
40:36here's primary skin cancer.
40:38Sorry, primary and metastatic skin cancer.
40:42Colorectal cancer.
40:45Actually this color,
40:46colon cancer,
40:47HPV negative head,
40:48neck cancer and thyroid cancer and the
40:51diagrams are this bar versus this bar.
40:54So the bar on the left is how
40:56much of a gift for a specific
40:58tumor was contributed by a given
41:00process and then how much of the
41:02oncogenesis for that tumor was
41:04caused by that particular process.
41:06And I've lined these up so that
41:08what I'm showing you is just
41:10five examples here from TCJ and
41:12this is actually from some data
41:14gathered here at Yale on Melanoma.
41:16But anyway,
41:17we've looked across these
41:19different cancer type,
41:20these different tumors and the
41:21question is are these different
41:23or are these similar?
41:24Like is the basic mutagenic effect
41:25and the cancer effect similar or is
41:27it different and you see they're very
41:29similar for two these two tumors,
41:31very similar for this third
41:32getting a little different here
41:34and getting quite different here.
41:35And these are arranged at the
41:38quartiles of the distribution.
41:40So it sort of represents the
41:41range of what you see in patients.
41:43So most of the time mutagenic
41:45effect and cancer causation.
41:46Are aligned very closely in primary
41:48skin cancer and that's because UV is
41:50causing almost all these mutations
41:51and changing things in colon cancer.
41:53As you extend from the more
41:55similar to the more different,
41:56you see a lot more heterogeneity
41:58from patient to patient in terms of
42:01whether or not the causative factors
42:02are the same as the myogenic factors.
42:05And that gets even more extreme
42:06with HPV negative head,
42:07neck cancer and even more
42:10extreme with thyroid cancer.
42:12So, but let me just emphasize again,
42:15these measures are for individual patients.
42:17So in principle this calculation can
42:19be done on any tumor sequence from
42:22an individual patient to tell you.
42:25What the causation of their cancer was,
42:28at least to the level that we
42:30are able to analyze this now,
42:31there's a bunch of things that
42:33are that we would love to also
42:35be able to incorporate into this.
42:36This is only single nucleotide mutations.
42:38It doesn't take into account
42:39copy number variation.
42:40It doesn't take into account
42:42epigenetic changes.
42:42And as I said at the outset,
42:44none of this has to do with physiological
42:46things like whether you exercise and have
42:47good autophagy in your you know it doesn't.
42:49It's not that physiological
42:51question of why you got cancer,
42:52but it is the mutagenic
42:53answer of why you got cancer.
42:55Down at the SMV level. Uh. And.
43:00So it reveals that and so that, so.
43:04So I think we're very good there.
43:05I would argue that the logic behind
43:07this is right and that we can apply
43:10that same logic to epigenetics to
43:11to copy number changes etcetera.
43:13There's just a lot of understanding.
43:14We still need of the basic underlying
43:16mutation rate for those things
43:18in order to actually do that and
43:19we're trying to work on that now.
43:23Now I just this, this is basically
43:25the the the end of the major talk but
43:27I just want to emphasize that this
43:28doesn't just apply to the origin of
43:30cancer in the early tumor genesis the
43:33same the same processes are going on
43:36in patients as we treat them as well.
43:38So there is you know so if you have a
43:41patient where you take out a a biopsy or
43:43a resection and then they undergo some
43:46sort of treatment and have recurrence
43:48there are ways to figure out exactly
43:50what the underlying processes that are
43:52contributing to the mutations that.
43:54Because that recurrence are.
43:56So that should be of interest to all of
43:58us who are interested in figuring out
44:00what's causing recurrence in cancer.
44:02So there's a clinical as well as the
44:04sort of more public health side that
44:05I was talking about with regard to
44:07these mutations and clinical side of
44:08how we might be able to apply this.
44:10And just to give you 2 material
44:13examples of this, here are two,
44:15I'm going to show you two sort of
44:18tree studies of individual patients.
44:20These were led by Nick Fisk in my lab.
44:24And here's a patient who was diagnosed
44:29with stage 3B lung cancer.
44:32They had an EGFR exon 19 deletion
44:35and their tumor was resected.
44:37They were given cisplatin and permatex
44:40bib and this there's a little pipe
44:42part and this is a phylogenetic tree
44:44relating their metastatic tumors to
44:46the primary tumor and it's been dated.
44:48We have all these techniques in my
44:49lab to date that based on the when
44:51the primary tumor was and how many
44:53mutations we see etcetera, etcetera.
44:54These were extracted at a later
44:56date than this,
44:57and so that gives us a way
44:58to calibrate the time.
44:59And what you see here in these pie
45:01charts is I've I've made it simpler.
45:03I just am looking at all other
45:05kinds of mutagenic sources.
45:07And one specific source
45:08that I'm interested in.
45:10And in this particular case,
45:11the source I'm interested in
45:12is the effect of cisplatin,
45:14which we know has a mutagenic
45:16effect on tumors even as it.
45:19The you know applies its own selective
45:21effect killing tumor cells and what
45:23you can see here is that the cisplatin
45:26mutations on this branch so this
45:28is independently determined right.
45:30This isn't because cisplatin is here
45:31we just did the deconvolution and boom
45:33here are all these despite mutations.
45:35This white pie piece here almost
45:37you know a bit less than 1/4 of or
45:40around 1/5 of the the mutations in
45:42this tumor are now cisplatin derived
45:44mutations and we can deconvolve
45:46that by doing this tree and seeing
45:48OK on this branch right here.
45:49That's how many are are that kind of
45:51mutation and then and then that of
45:53course that tumor continued to evolve
45:55and the reason it continued to evolve
45:57it was the patient was given or alot nib.
46:00Unfortunately a lot nib wasn't
46:01very successful in this case
46:03because they got the EGFR T790M
46:05resistance mutation on this branch as
46:07well and the tumor differentiated into
46:12these metastatic metastatic tumors and
46:15another metastatic tumor in the pancreas
46:18and and the point is here just that.
46:20The proportion of cisplatin was
46:21discontinued and so the proportion of
46:24mutations in subsequent branches is
46:25actually lower out of the total because
46:27new mutations are being added but they
46:29aren't cisplatin related mutations.
46:31So all of this deconvolution and
46:34understanding of the underlying mutagenic
46:36causation occurs during this treatment
46:39process that that that patients receive.
46:41And we can figure it out.
46:44There's one more point that
46:44I just want to make here,
46:45which is that it turns out,
46:48and I don't have a plot for this,
46:49but the T790M mutation is a mutation
46:52that is very likely to be caused
46:54just like those other ones I
46:57showed you by cisplatin mutation.
46:59So this is a poor ordering clinically
47:02for these treatments to be given
47:04because this despite mutation,
47:06creates a bunch of that genetic
47:07variation that is exactly what we
47:09don't want to have if we're going
47:11to put them on our lot and later.
47:13And very likely they had that mutation
47:14right when they were put on their lot nib,
47:16which is why there's a very little
47:19duration of of of benefit for the patient.
47:22So this is a great example for
47:24for in terms of a clinical or
47:26exogenous source of mutation,
47:27the cisplatin treatment
47:28that they were receiving.
47:30And let me give you another example that's
47:32about an endogenous change that has an
47:34interesting effect in a very similar way.
47:36So here's another lung cancer case.
47:38This patient was put on her right resection.
47:41They had a P53 mutation
47:44already after resection,
47:45but over a much longer period of time.
47:47They were never treated
47:48with cisplatin a much,
47:50much longer time later they did receive.
47:52They did get an ESR T790M mutation and
47:56you can see these plots are solid here
47:59meaning that the mutational process of
48:01interest that I wanted to talk about you
48:03know hasn't happened at all here yet.
48:05And then you can see unfortunately
48:08later on they were they were moved
48:09to Avastin or not unfortunately
48:11movement they were moved to Avastin.
48:12It wasn't unfortunate necessarily
48:15but Erlotinib was discontinued.
48:17They were given permatex have been
48:19carboplatin late in latent therapy but.
48:22But the divergence,
48:24but they're metastatic tumors
48:26started genetically diverging about
48:29two years before their death.
48:30And the thing that I just want to emphasize
48:32is here is a continuum B1 mutation S37C,
48:35which is known to induce.
48:39Yeah.
48:42Defects in homologous recombination
48:44based mutations and you can see in the
48:47descendant lineages the increased amount
48:49of that kind of mutation occurring after
48:52continuing 1B stinging B1 mutation.
48:54So this is an endogenous process that
48:56was started by a mutation that we can
48:58then track again down to the individual
49:00branch where the mutations are occurring
49:03and how many of them are caused.
49:04And then from then on there's a lot of
49:07cutting and B1 mutation in these tumors,
49:09but not in the spleen.
49:10Interesting.
49:11That's an interesting story here
49:12is that this could be 1 mutation
49:15occurred and led to all the metastases
49:18to all these other tissues.
49:20The one tissue that had a
49:21metastasis that was not continuing
49:23to be 1 mutated was the spleen,
49:25which is very interesting because Canadian
49:27B1 mutation causes vascularization.
49:29The spleen is already quite
49:30highly vascularized,
49:31so it may not have been
49:32needed for the spleen,
49:33whereas it may have been more important
49:35to the cancer for the rest of the tumors.
49:38All right,
49:38I've sort of gone through all of
49:40what I wanted to talk about.
49:42Today, in terms of introducing you to
49:44this way of actually trying to understand,
49:47you know,
49:47why an individual tumor is
49:50has been made oncogenic.
49:52I hope that I've at least been able
49:53to argue that the logic behind what
49:55we're doing is sound and that the
49:57process that we're doing is a sound way
49:59of sort of attributing that cancer effect,
50:01at least as regards those
50:02single nucleotide variants,
50:03which are what we mostly focus on.
50:05But where we don't, you know,
50:07we don't know whether that's 10% or 90%
50:09of the reason why genetically tumors are.
50:12Because we don't know that yet.
50:13But regardless,
50:14if we look at that single nucleotide effect,
50:18we now can sort of deconvolve that.
50:20And I'm very curious if anyone
50:22has thoughts to share with me.
50:24You know what how this information could
50:27be used for the benefit of patients,
50:30for the knowledge of patients,
50:31but also as I mentioned in the later
50:33part of my talk in in understanding
50:35better what our therapies are doing to
50:38patients over time as well and ways
50:40that we can ideally order our therapies.
50:43So that we avoid the evolution of the
50:46resistance that we're trying to avoid,
50:48which is so clearly evidenced in
50:50that one case with the cisplatin
50:53and relative treatment.
50:54Thanks very much for your attention.
50:56I want to thank very much.
50:58Vincent Cantaro who was a postdoc in my
51:00lab is now a professor at Emmanuel College,
51:02remains a a vibrant and
51:05wonderful collaborator of mine.
51:07I really like working with him.
51:09It's been incredibly productive to continue
51:11to do so and I'm really delighted that.
51:13He's been able to do so despite a very
51:16heavy teaching load at Emmanuel College.
51:18Jeff Mandel over here in the audience,
51:21graduate student working on the
51:22cancer effect size calculations
51:24and the machinery underlying that.
51:26Nick Fisk who's worked on a lot
51:28of the tree
51:28based analysis in my lab,
51:30including those last ones that I mentioned.
51:32Everyone else in the town's lab,
51:34they're also great.
51:34I didn't specifically mention their work.
51:36We've got interesting work on epistasis
51:38and all kinds of other things that
51:40that I really think is outstanding
51:42and should be really interesting.
51:44But so all the Members,
51:45it's a great group.
51:46Also I want to thank the NIH, NIDCR,
51:48Yale support and head and neck is
51:50a great community and I really
51:52enjoy being part of it and and it
51:54also provides a substantial amount
51:56of my funding for cancer work.
51:58So thank you very much and I
51:59would love to take any questions
52:01or hear any thoughts, comments,
52:02etcetera from anyone in the audience.
52:11Jeff, I think you actually have
52:13an amazing talent to make really
52:15complex things, explain them in
52:17a very simple and logical way.
52:19Do we have a microphone so,
52:22so that people could actually ask questions?
52:27Oh, OK.
52:37OK. Amazing work.
52:39Jeff I just wanted to ask if you could
52:42address how you handle commutations
52:44when you look at cancer effect size.
52:46And you know I'm thinking of
52:48this finding that we haven't had
52:50neck cancer where P53 mutation is
52:51truncating if you don't have CDK into
52:54a mutated and when you when you have
52:56a mutation in the DNA binding domain,
52:59it seems like you need the second mutation.
53:01So how do you handle that and then I guess
53:04also when you have P53 mutations or.
53:07CDK into a mutations or whatever.
53:10Does that muddy your signatures at all?
53:13You know to the to the mutational effects of.
53:18Losing control of of cell cycle
53:20and DNA repair.
53:23This is a little more straightforward.
53:25Does it muddy things when things get changed?
53:27Yes, it does in the sense that there
53:29is a temporal difference, right,
53:31between what was happening before that
53:33change happened and what happened afterwards.
53:35And our resolution for understanding those
53:38temporal differences is somewhat weak, right.
53:40So generally, if we have a lot of
53:42samples like we had in those two cases,
53:44we can sort of piece apart when things
53:46happened in a nice way and we'll be able
53:48to understand those sorts of differences.
53:50But when we're just looking at tumor
53:52Genesis to resection and we have.
53:53This association then we have to use
53:55very large numbers to get sort of
53:58statistical associations to understand
53:59that sort of ordering process.
54:01Which gets me to your second question or
54:04your first question of of commutation which
54:08I have a strong opinion on everything.
54:09So my strong opinion on this is that that
54:12the general genomics approach towards
54:15looking at commutation is is flawed in a
54:18way that is not apparent when you read
54:21all the papers on it and the argument that.
54:23I want to make is that when
54:24you look at commutation,
54:25you're typically looking at a
54:27very observational thing,
54:27which is like how often is this
54:29one mutated and this one mutated.
54:31So for the same reasons that I
54:34outlined in my talk today,
54:36that there are two reasons why
54:38you see things mutated, you know,
54:41underlying mutation or selection.
54:43There are two reasons why things
54:45might be commutated.
54:46They might be commutated because
54:47when you get one,
54:48the other one is selected and it
54:50really creates a great benefit to
54:52the cell to survive and replicate.
54:53The other reason might be because
54:56they both have the same underlying
54:59mutational process.
55:00And when you have four orders
55:01of magnitude of difference in
55:02mutational process site to site,
55:04that can be a very big reason
55:06why you see commutation.
55:07So commutation is not the signature we
55:09like it to be to say these things are
55:12selected together because sometimes
55:13they may not be even though they're
55:15strongly come come mutated in a data set.
55:18So then how do I deal with it?
55:20Well,
55:20we can take all of the approaches
55:22I told you and we're working on
55:24you know even more sophisticated
55:26approaches now to try to do this.
55:27I think I have some slides on it,
55:29so I would love to take the time to.
55:31Just quickly introduce it since
55:33they're they're way down here though.
55:36Ah.
55:37Yeah.
55:38So this is the point that you
55:39that I was just answering to you
55:41which is mutual exclusivity and Co
55:42occurrence are patterns that are
55:44caused by either commutation or
55:45what I call selective epistasis.
55:47Again I'm using the terminology from my
55:49background and evolutionary biology.
55:50Epistasis meaning 1 gene is having an
55:52effect on another or the mutation in
55:54one gene is having an effect on another.
55:56So typical approaches have not
55:57acknowledged the possibility
55:58of commutation
55:59which is a common underlying mutational bias.
56:01That's what I just said to you.
56:02This is a typical slide from I
56:04don't mean to be you know casting
56:05aspersion on this as I said this
56:07is what everyone pretty much.
56:08Does but but they look for whether
56:10cancers have sequential mutations
56:12developed or commutation but we can
56:15actually take those same analysis same
56:17the same data and and and deconvolve with
56:20some fairly sophisticated mathematics
56:22that Jorge Alfaro Murillo and I did on
56:25the fluxes mutation rates and scale.
56:26So selection coefficients for up to five
56:28genes and look at what the likelihood
56:30of individual genes are are to get
56:32mutated what the likelihood Karras is
56:34going to be muted after P53 etcetera.
56:36So we can look at all of these.
56:38Figure out how frequently those happen.
56:40So this is the flux which is a
56:42measure of commutation,
56:43essentially the underlying mutation
56:44rates and then the scaled selection
56:46coefficient for the new mutation.
56:48So these are how likely is P53,
56:50how likely is KSB mutated after PHP 53
56:53and then how likely is it how selected
56:56is it to have KRS after PMI 50?
56:58P 53 is a separate measure,
56:59so we can basically take all of those
57:01and look at all of those different
57:03things for up to five or six.
57:05And again there are constraint
57:06is usually the amount of data.
57:08We need massive amounts of data
57:10to understand,
57:11like 3 way effects or four way effects.
57:13So you need to have examples of every
57:16possible combination in that data set
57:18and that rapidly exhausts our samples.
57:19But on their hand,
57:20we're getting a lot more data
57:22now and so we're able to do this
57:23with more and more data sets.
57:24Now this is lung cancer and we were
57:26able to do it for these five genes.
57:30P53 KSDK, 11 RL, RP1B and.
57:33And figure out all their
57:34relations with each other.
57:36This is maybe an easier way to
57:37see this instead of a big table,
57:39which is just what's the
57:40evolutionary trajectory of them.
57:41And again, this is all epistatic,
57:43like it's all taking into account
57:45that commutation factor and the
57:46width of the bar is the flux,
57:48or how frequently you go from normal
57:50to say P53 in this particular case,
57:53or LPV one or K Ras or SDK 111.
57:56And then you can see that if you KS isn't
57:58actually that frequent as a first mutation,
58:00but if you do get it,
58:01then you're very likely to get LRP 1B.
58:04Or SDK 11.
58:05If you get P53,
58:06you're very likely to then
58:08get LRP 1B as well.
58:10You're you're you know some probability,
58:12but it's not so high of getting curious.
58:13After that you're very likely to get a KRS
58:16mutation if you have P53 and LRP we want to.
58:19One LRP 1B together et cetera.
58:21So you can you can look at what the
58:23likely trajectory for a given patient is.
58:25You could even look at where they are on
58:26this trajectory and we haven't done this,
58:28but presumably you can figure out what
58:29their prognosis was based on where
58:31they were on this diagram etcetera.
58:33And we have basically a a map of what's
58:35actually happening to these these patients.
58:37And then down below in the smaller diagrams,
58:39I've just divided this up because
58:41this is all the fluxes again,
58:42but let's divide it up into mutation
58:44rates and selection coefficients
58:46and what you see is the mutation
58:47rates are here are quite.
58:49Symmetrical because we haven't
58:51accounted for things like.
58:53Containing 1B mutation,
58:54changing the mutation rate etcetera.
58:56In this particular analysis,
58:57although in principle we can do
58:59that and then on the right are
59:01so there's a LRP 1B particularly
59:03has a very high mutation rate.
59:05So it's relatively high frequency
59:06is not that big a deal,
59:07although it does seem to have
59:09some selective effect as well.
59:10And then over here we see the
59:11selective effects and you can see
59:13there's very strong selection for
59:14P53 initially is the major selection
59:16and yet that exists after LRP 1B
59:19as well but after after P53 or.
59:24LRP 1B and P33,
59:25then we're very likely to get
59:27this Karas mutation, etcetera.
59:28So you can really understand what the
59:31relative effect of each of these is.
59:34Trajectories after the sample size.
59:37That's a good question.
59:39I haven't done the study that I'd like
59:41to do to answer that, which would be
59:43like do some very massive analysis.
59:45It's actually a lot of computation to like
59:47do that 1000 times subsampling etcetera.
59:49But what I have done is just do the analysis,
59:51you know, with one data set and then
59:53add more data sets and it seems
59:55quite stable from that perspective.
59:57That's not really the same because
59:58we're not subtracting out the
59:59first data set when we do that.
01:00:01But but it's not like it varies all over
01:00:03the place and the stability of course is
01:00:06proportional to the prevalence, right?
01:00:07Of that particular mutation,
01:00:09the mutations that are really
01:00:10highly prevalence, you know,
01:00:11they stay very stable because we've got a lot
01:00:13of examples of them with the other genes.
01:00:15As soon as you get the lower prevalence,
01:00:17it's it's a lot iffier.
01:00:19So.
01:00:19So really this can only be used right
01:00:21now for these for the most prevalent
01:00:23kinds of mutations that you see.
01:00:25And typically we are for instance
01:00:27assembling all the mutations in a given
01:00:30gene as one kind of mutation because
01:00:32we need that sample size to do that,
01:00:34which is something that in my other
01:00:35research I usually avoid because I think
01:00:37it's really important to understand it.
01:00:39Different sites have different effects.
01:00:41So
01:00:42one thing that that I didn't
01:00:44see certain probably this.
01:00:46So you can calculate an additional process
01:00:49contribution to to the privatization
01:00:52in particular individual cases.
01:00:54But what happens if you caused the
01:00:56cases and obviously you should be able
01:00:57to sell it off lung cancers related
01:00:59to smoking and those who don't and
01:01:01that would be a trial thing to do.
01:01:03But could you do the same and create
01:01:05a new classification for example for
01:01:07initial cancer, breast cancer that
01:01:09are going to aging and the other?
01:01:11By looking at them separately,
01:01:12you might get some idea about
01:01:15what's actually causing.
01:01:16The.
01:01:17The Unknown edition signature.
01:01:22Yeah, I definitely think you
01:01:23could cluster them. I think you
01:01:25know the you're reducing the
01:01:28dimensionality of the data when you
01:01:31go from the raw data back to the
01:01:34processes and so you have a reduced
01:01:36dimensionality of that raw data.
01:01:38And then you're and then if you were
01:01:40to cluster on the basis of this,
01:01:41you would be taking that reduced
01:01:42dimensionality data and trying to
01:01:43say does that predict something.
01:01:44So I I think from a machine learning
01:01:46standpoint you might want to just go
01:01:48back to that broad data in some way,
01:01:49but there might be some way
01:01:51of thinking about it.
01:01:51That I say that,
01:01:52but then I also think there's
01:01:53a second part of that,
01:01:54which is that I do think you do better
01:01:57looking at actual biological processes,
01:01:59even if it involves some reduction
01:02:01of the data,
01:02:02because it simplifies the data in
01:02:03a way that means you don't go off
01:02:05on these random tangents of all
01:02:06the noisy stuff you're looking at.
01:02:08So, so there's, there's a,
01:02:09I guess there's a tension I think
01:02:11you should be wary of in doing that,
01:02:13but I don't see any reason you couldn't
01:02:14do that and and it would probably
01:02:16be highly predictive in some cases.
01:02:17You're probably going to see
01:02:19most skin cancers very easily,
01:02:20you know, predictive that way.
01:02:21Because they're just UV all over the place.
01:02:25Some other cancers are probably
01:02:26quite hard to distinguish one from
01:02:29the other just by the mutational
01:02:31processes that underlie their cause,
01:02:33and so I could imagine doing that.
01:02:36We haven't done anything like that.
01:02:41Any other comments?
01:02:43To ask questions, then the audience on
01:02:46there, there was, I thought,
01:02:48Q&amp;A, but I there it is.
01:02:51We have time. Ohh. Yeah.
01:02:53We've got some questions here,
01:02:54but maybe one more for you and then
01:02:55I'll go to the online questions.
01:02:56Yes, OK. Thank you.
01:02:58Thanks, Jeff. Fantastic work.
01:03:03I think your methodology is on the right
01:03:06track and nothing to worry about at all.
01:03:09The opposite is true.
01:03:12My only concern is availability
01:03:15of data in the future,
01:03:18especially for new types of cancers.
01:03:21Are we asking the right questions?
01:03:24Are we collecting the right data?
01:03:27Be meaning human as humans. And.
01:03:34I'd like us humans to ensure
01:03:37that this data is available,
01:03:39it's it's open source and it's reliable
01:03:44and what are your thoughts on that?
01:03:47Yeah, so that's a great question.
01:03:49I mean I think that the volume of data
01:03:51sets on like tumor Genesis for section
01:03:53kind of data is going to increase very
01:03:55well on its own like we don't need
01:03:57to pay attention to that question.
01:04:00The the datasets that I think I would like
01:04:02to see more of are these multi sample data.
01:04:04That's from individual patients.
01:04:06Back in 2016,
01:04:07I was lucky to be funded by Gilead
01:04:09to actually sequence these large
01:04:10numbers of metastatic and primary
01:04:12tumors and they were really there.
01:04:14The potential of those data
01:04:16sets is really high,
01:04:17especially if they have a
01:04:19clinical annotations alongside.
01:04:19So you can map it to to understand what
01:04:21was happening for the patient at the
01:04:23same time as what was happening genetically.
01:04:25That data set though was
01:04:27heterogeneous by cancer type, right?
01:04:29And I haven't seen similar sized data
01:04:31sets on individual cancer types gathered.
01:04:34And it's not, you know,
01:04:36it's a lot of money like it's a couple
01:04:38$1,000,000 to do that sequencing,
01:04:39but you could do that for
01:04:41every cancer type for.
01:04:43You know,
01:04:43$30 million or something like that.
01:04:45And I think that would be so worth
01:04:47it because we would learn so much
01:04:49about the evolutionary trajectory of
01:04:50each of these cancer types by looking
01:04:52at multi sample data like that.
01:04:53But I haven't managed to sort
01:04:55of put together the argument to
01:04:57get funding to do that.
01:04:58I encourage you to elevate that, you know.
01:05:03Definitely. Yeah. Thanks. Like.
01:05:08Just. Comments. I'm sorry.
01:05:12OK. First, I enjoy your talk.
01:05:14Thank you.
01:05:14But I'm not so sure.
01:05:17Given the tumor heterogeneity.
01:05:20Your math, just the tumor cell.
01:05:22We don't even talk about
01:05:24the microenvironment.
01:05:25Math sequence will really be useful.
01:05:29With all the other tools.
01:05:32You know, otherwise you're going to.
01:05:34For instance,
01:05:35you just mentioned about the cluster.
01:05:37Approach.
01:05:39You can have a mutation in different
01:05:41tumor cells within the tumor mass.
01:05:46When you do the analysis,
01:05:47you put them all together.
01:05:51Does that make sense?
01:05:53I think I might need to talk to you
01:05:55at more length to sort of fully
01:05:56understand your question, but but I
01:05:58guess what I would comment is just that.
01:06:00And I say this is the kind of data we need.
01:06:02I'm mostly talking about for the kind of
01:06:04work that I'm talking about rather than
01:06:06for everything to solve cancer, of course.
01:06:08So, but but in order to understand the
01:06:09underlying selective coefficients and
01:06:11understand the mutational processes,
01:06:12I do think large amounts of.
01:06:15Tumor resection data which will
01:06:16be gathered anyway,
01:06:18but also more of this multi sample data
01:06:20so that we can understand dynamically
01:06:21over time what's happening which we can't.
01:06:24We can do, I said in a probabilistic way,
01:06:26but never in a very satisfying way with
01:06:28just the tumor genesis resection data.
01:06:32It makes the noise that the tumor
01:06:34cellularity differences we bring in and
01:06:37I think it also remains you are gorgeous
01:06:39question about the copy number changes.
01:06:42So how do you adjust,
01:06:43what is that you know if it has
01:06:4517 copies of imitation it has that
01:06:48signature that will be amplified.
01:06:50And it's not necessarily black
01:06:52would be the actual sometimes
01:06:54higher prevalence of contribution
01:06:55of the particular audition process,
01:06:58but it's just that the gene.
01:07:01I see these questions about the the.
01:07:04So the adjacent normal tissues
01:07:07requires mutations and they
01:07:09actually introduce noise, right?
01:07:13Yes. So both of those are sources of noise
01:07:17in the sense that on average as we look at,
01:07:21so the say talk about a gene amplification
01:07:23for instance is a great example.
01:07:24When you get a gene amplification,
01:07:26you know the the mutation itself may not
01:07:28be contributing the cancer effect size that
01:07:30we analyze when we get this kind of data.
01:07:32But what is true is that those
01:07:34mutations and the amount of copy number
01:07:37amplification that they typically have
01:07:39contributes this amount because we're just
01:07:41looking at whether or not we see these.
01:07:44The patients and whatever other processes
01:07:45are going on, we're averaging over. So.
01:07:47So the cancer effect size is still I
01:07:49would say it's still the measure of how
01:07:51much that mutation is contributing to it.
01:07:54But the means by which it contributes
01:07:55we don't really know from this analysis.
01:07:57It's a it's just that wider question of
01:07:59how much is this variant contributing
01:08:01and and if it needs amplification
01:08:03as part of that process,
01:08:05well then we need to do a more
01:08:07detailed analysis that looks both at
01:08:08amplification and the and the mutation
01:08:10and then we'll be able to say like how
01:08:12important that mutation is in terms of.
01:08:14Cancer affect how important the amplification
01:08:15vacation is in terms of cancer effect
01:08:18compared to the mutation itself.
01:08:19That's not something we've
01:08:20been able to do yet,
01:08:21but it's something on our agenda.
01:08:23It's very difficult but I think
01:08:25it's achievable but very difficult.
01:08:29I think I better quickly ask,
01:08:30I feel sorry for the people
01:08:32who ask questions online.
01:08:34The one question is,
01:08:36is mutation a biochemical reaction to
01:08:39TR GRC a substitute of T or G or C?
01:08:42The mutations I'm talking about
01:08:43in this entire study were all
01:08:45single new type mutations.
01:08:46In the context of A3,
01:08:49what I meant by trinucleotide context
01:08:50is the 3 mutations in the central one.
01:08:53How was that mutated to
01:08:54another single nucleotide?
01:08:56There are ways to look at doublets
01:08:57there are ways to look at.
01:08:58Some other more complicated indels which
01:09:00we have in the lab almost implemented,
01:09:03but other mutation types we don't
01:09:06have actually looked at Yuval
01:09:08Kluger's question I think thank you.
01:09:10You have echoed that for for me on low.
01:09:13So I believe I answered that.
01:09:18That you know basically it's true
01:09:20that we don't know the specific,
01:09:23you know when we talk about this mutation
01:09:25and how much is cancer effect sizes,
01:09:27that's in the context of everything that
01:09:29happens to that mutation in cancers
01:09:30and it's the average across that.
01:09:34But Tim Robinson has a question,
01:09:36which is, can the spectrum of mutations
01:09:38tell us about the chance that the
01:09:40tumor will respond to treatment?
01:09:45It may well, so for instance you know this,
01:09:47the fact that there were cisplatin
01:09:49mutations is going to tell you that it's
01:09:52likely to have an EGFR T790M resistant
01:09:54mutation sort of sitting there waiting
01:09:55to come out when you give it a lot.
01:09:57So in a sense that spectrum could
01:09:59tell us about the chance that a
01:10:00tumor was bound to treatment.
01:10:02But in general if I could I would
01:10:04rather look at look for EGFR
01:10:06T790M itself directly for example,
01:10:09if the vast majority of mutations are in
01:10:11Melanoma and Melanoma are B rap 600 and the.
01:10:13The vast memory of cancer
01:10:15causation by mutation is there.
01:10:17Does that inform the chance that the tumor
01:10:19will respond to directed therapy to be wrap?
01:10:24Umm, I think the, you know, the number
01:10:27of mutations I don't think does at all.
01:10:28I think that what's important to
01:10:30understand about Viraf E7 and E and
01:10:32it's cancer effect size, which by the
01:10:34way is a very high cancer effect size,
01:10:36is that if you can get a therapy that
01:10:38treats the rap fee 600 effectively,
01:10:40it will be a very effective therapy.
01:10:42And there's a good example of that.
01:10:44And there's a caveat to that example also,
01:10:46which is that the raffish under 600 E,
01:10:48as many people know, there's vemurafenib,
01:10:50which is a very effective therapy
01:10:52for skin cancer.
01:10:52The only problem is there's.
01:10:54Very rapid evolution of resistance.
01:10:56Nothing about cancer effect tells you
01:10:58how quickly resistance will be evolved,
01:11:00and in that case this also interplays
01:11:02with CNV's because at least one
01:11:04of the explanations for why that
01:11:06rapid rises occurs is that you get
01:11:08amplification of the variant BRAF
01:11:10V600E that basically overwhelms the
01:11:12treatment of vemurafenib and means that
01:11:14you and that's a very fast process.
01:11:17Amplification of a gene in a genome
01:11:19is not hard to do as a high mutation
01:11:22rate happens very quickly.
01:11:23Some cells have.
01:11:24More of it somehow is less than
01:11:25those ones with more selected.
01:11:26It's very easy to select on that basis.
01:11:28So so it I think it informs you
01:11:30about how likely a treatment is
01:11:32to have a big effect at the moment
01:11:35you apply the treatment.
01:11:36How quickly you evolve resistance
01:11:38is another question.
01:11:41Umm. And already moustaki the
01:11:45sources of mutations smoking,
01:11:47UV infection affect the normal non
01:11:50transform tissues. Yes they do.
01:11:52Can you use your approach to calculate
01:11:53the cancer effect mutations on the tumor
01:11:55micro movement have on tumorigenesis.
01:11:57One might argue a lot of these mutation
01:11:59sources act on the environment reducing
01:12:00the fitness of a normal cell allowing the.
01:12:02This is a really interesting question.
01:12:03We are working on this.
01:12:05So the the bottom line is that
01:12:07and I'll be very quick with this
01:12:09answer that once we are able to
01:12:11figure out these cancer effects.
01:12:13Then we can ask it to the extent
01:12:15that we have annotated data on
01:12:17this tumor was exposed to this
01:12:19given treat this given environment,
01:12:21we can ask how does that environment
01:12:24affect the cancer effect.
01:12:25So we can ask if you're,
01:12:27if you have different ages,
01:12:28not just what mutations are caused by aging,
01:12:30but how much does the cancer effect of a
01:12:32given mutation change as someone ages.
01:12:34So there's ways to do that with
01:12:35the kind of data we have.
01:12:37Again it requires bigger sample
01:12:40sizes in general,
01:12:41but we're looking at that right now with.
01:12:43Regard to smoking,
01:12:44because smoking of course can have a
01:12:46direct effect of mutating individual genes,
01:12:48but it can also have a physiological
01:12:51effect of degrading the normal cells
01:12:53in general in the lung ecosystem.
01:12:55And because you have degraded normal cells,
01:12:57that could increase your chance
01:12:58of getting cancer.
01:12:59Or it could mean that certain
01:13:00mutations are more likely to be
01:13:02able to make cancer proliferate and
01:13:03survive better than other mutations.
01:13:05So.
01:13:05So the Physiology could be very important,
01:13:07and there are ways to get at that.
01:13:09But you need to know this first,
01:13:10and then you can ask the question
01:13:12about Physiology affecting things.
01:13:14And I think I'm out of time.