Personalized Medicine: Are We There Yet? by Marylyn Ritchie

[ Silence ] >> So welcome to the 2013 Penn State Lectures
on the Frontiers of Science. This lecture series is a free mini-course for the general
public that we are pleased to be able to offer to you because of the financial support of
the Penn State Eberly College of Science. And also because of the generosity of our
speakers who volunteer their time to give these lectures for us. Our theme this year
is, “Your genes: how they contribute to who you are.” Today’s event is the fourth out
of six lectures in this series. We have two more after this week. Our speaker today is
Marilyn Ritchie who is an associate professor of Biochemistry and Molecular Biology and
she is the Director of the Center for Systems Genomics at Penn State University. Her research
spans the fields of Biology, Genetics, and Statistics. It focuses on identifying and
analyzing genes, interactions between genes, and interactions between the environment and
genes that may increase the susceptibility to common diseases. Some of these diseases
include cancer, diabetes, hypertension, and cardio vascular disease. Among the honors
she has received for her research are an Alfred P. Sloan Foundation Fellowship and A Rising
Young Investigator Award from the Journal of Genome Technology. Before coming to Penn
State, Dr. Ritchie directed two computational genomics programs at Vanderbilt University.
She also has served as a consultant to one of the world’s leading pharmaceutical companies.
The title of her lecture today is, “Personalized Medicine: Are we there yet?” Please join me
in welcoming Dr. Marilyn Ritchie. [ Applause ] >> Thank you very much. Good morning everyone.
Can everybody hear me okay? All right, great — if you can’t hear me, just start waving
your arms and I’ll talk louder. But I tend to speak pretty loudly. And so today, I’m
going to tell you a series of stories about personalized medicine, how I got into the
field will come up a little bit, and talk about where we are today. I get a lot of questions
about, you know, “Is personalized medicine real? Are we actually doing personalized medicine
or is this just a science-fiction thing that scientists like to talk about?” And — and
so I’ll tell you where we are today and you can decide for yourselves if you think we’re
there yet. So what do I mean when I say personalized medicine? It — it means a lot of different
things to different people and some people are now calling this precision-based medicine.
Some clinicians would argue that we’ve been doing personalized medicine since the age
of time. Doctors personalize treatment for the person coming into the clinic. But it’s
become kind of a — a common thing to talk about using genetics in the clinic as personalized
medicine and so, some people are now referring to that as precision-based medicine. So in
case you read that or hear that, it — it means really the same thing to geneticists.
And what we mean by this is looking at the right medicine for the right patient at the
right dose at the right time. All of these things are equally important when you’re thinking
about personalizing treatment based on one’s genetics. It’s not just a matter of knowing
whether somebody should or should not have a drug. We’re also learning things about whether
or not the dose that’s typically prescribed that was learned in clinical trials is the
right dose of the drug or whether the timing of the drug is important. And I’ll give some
examples today that will show genes that we’ve learned about where the type of treatment
is important and where the dose of treatment is important and where the gene actually plays
a role in the timing of the treatment and so, all of these will come up in the talk
today. These are just some common images that you can get off of a Google search. It’s become
a really popular thing to talk about. You’re personalizing drugs where drug design is being
done for a particular person. That is not happening quite as commonly because it’s very
expensive to manufacture a drug. What is more often the case is that drug companies are
starting to do clinical trials where they’ll look at the genetic ancestry of individuals
and try to categorize them into different groups, so that we can try to understand if
the drug works well for certain groups of people rather than certain individuals specifically.
But when we’re thinking of personalized medicine, we do want to be able to find the drug that’s
going to work for each person. And so, I’m going to tell you a little story about how
I got into this field. So when I was a graduate student back in 1999, I went to a lecture.
I had not really heard much about personalized medicine. I was interested in genetics and
helping people in learning about disease. And a scientist, Dr. Dan Roden who was a professor
at Vanderbilt University gave a talk where he showed a picture of this guy. So I was
a grad student in Nashville, Tennessee and so even though Elvis Presley was from Memphis,
everybody in Nashville takes ownership of Elvis and the star that he was. And so he
showed a picture of Elvis and his driver’s license and said, “You know, we have a lot
of information on our driver’s license. Wouldn’t it be cool if on that barcode on the back
of our driver’s license, we could have our DNA code? And when we go to the doctor or
we go to the pharmacy, we could just swipe our driver’s license and they would go in
the computer and say, “Oh, yeah, you should be prescribed this drug.” Or, “Oh no, this
is not the right drug for you. You’re going to have an allergic reaction.”” And he talked
about it in a very kind of way out in the future science-fiction kind of way, and it
comes to be that we’re almost at this point. We’re not there yet, but we’re almost at the
point. So what I thought sounded like the Jetson’s from when I was a kid, you know,
total science-fiction flying cars, we’re almost there at this point. He gave another example.
We go to the doctors and we get our oxygen checked. You put your finger in the little
thing and it will beep and show what your — your oxygen levels are. He said, “Wouldn’t
it be cool if you could go to the doctor and stick your finger in and they could sequence
you on the spot and tell you what your DNA sequence is and what variants in your DNA
predicts what drugs you should get, predict what diseases you’re going to have, predict
what responses you’re going to have to the disease or drug treatment that we put you
on.” Again, this is still somewhat science-fiction, sequencing is not that fast. But it’s almost
that fast. It’s definitely faster than it was back in 1999. And so I grabbed a few news
articles that will be in your slides and you can feel free to go look at them, at your
leisure. But so, this is one just from a few weeks ago talking about personalized cancer
treatment. So, cancer is one area where personalized medicine is happening today, in many, many
places around the United States and around the world. This is another article that I
found from earlier this year, talking about personalized medicine is almost here. I think
it’s here. Maybe the revolution isn’t here, but we are personalizing treatments. I’m going
to show a series of examples but there are drugs that we now know what genetic variations
in our DNA will explain how we respond to those drugs. And so, I would argue that it’s
not almost here, it’s here. We are living in the age of personalized medicine right
now. And I’m going to talk some about kind of the benefits of personalized medicine and
some of the limitations and where we need to be a little bit cautious. So this is an
article or the cover of Time from 2013, “I want to know my future.” So I thought this
was a really cute cover that shows things like obesity and Parkinson’s, asthma — we
are learning a lot about our risk for disease and how our genes explain that risk. But this
is not a new idea. So this is the cover of Time from 1971. So even back — which is before
I was born by the way — even then, we were talking about understanding our genes and
how they make us who we are. So this is not a new thing that Time Magazine has grabbed
onto. And here’s a cover from 10 years ago, 2003. Again, solving the mysteries of DNA
— I don’t think we’re going to stop seeing covers like this because as we learn more,
in some ways we understand more and in other ways we’re understanding less and I’ll explain
what I mean by that. A few other covers that I think are important to note, this is one
from when we were sequencing the human genome for the very first time. There was a race
to sequence the genome. There was a company that was doing the sequencing and there was
a government-sponsored group around the world doing the sequencing. They raced and finished
at the same time, so this is two — these are two covers, Science and Nature Magazines
from 2001. One talked about the government-sponsored project and the other talked about the corporate-sponsored
project. So 2001 is when we had the first draft genome sequence of humans, which is
not so long ago. We’ve learned quite a bit since then. These are some early covers of
Time talking about human cloning. No, we do not clone humans. No, we are not trying to
clone humans. Human cloning is not closer than you think. So if there’re any fears about
that, I just want to — we’re not going to talk about cloning today. But I wanted to
make it clear that knowing our sequence actually is not allowing us to clone ourselves and
some days I wish I could because if I had two or three of me, I could get a lot more
done. But we cannot clone ourselves yet. And why your DNA isn’t your destiny and I think
this is telling. We can learn a lot about ourselves from our DNA, but not everything
is in the genes. Our environment is very important, our diet, our nutrition. There are a lot of
things that come into play. In addition, it’s the complex interactions of our DNA and that
is where we don’t understand a lot yet. So personalized medicine — I’m going to talk
about some actual examples of truth and where we are today, where it’s being used in clinics
right now. So some of you could go to a clinic and be prescribed a drug where you could have
your DNA evaluated and determine your dose or whether it’s the right drug based on your
sequence. I’m going to talk about exactly where the limitations are today and as always
as a scientist, we just keep calm and do the science. There are things we can learn, there
are things we don’t know yet. That’s what makes us scientists. We keep asking the questions
and trying to understand more. So here are some perspectives on why our field has shifted
so fast since 2001. When the first draft sequence of the human genome was done back in 2001,
it cost $100 million and that was to sequence one human genome. Today, we’re able to sequence
genomes for less than $10,000, somewhere on the order of between $5,000 and $10,000. That
drop is faster than Moore’s Law. So it has dropped dramatically really since about 2006,
2007. And so we’re seeing the ability to look at our DNA variations so all of the bases
in our DNA very, very rapidly which is very exciting and it’s giving us a lot of opportunity.
Now, for people like me, fortunately the cost of the analysis of the genome has not gone
down. We may get to the point where genomes cost $1,000, which is the stated goal of the
NIH. They want the cost of sequencing to be $1,000 a genome. The analysis is still at
least $100,000, maybe more. And this is a great article that talks about this — it’s
job security for people like me. I love articles like this. But it’s important to note that
just because we can generate the data and we can generate the data very quickly, that
doesn’t mean that we can process, analyze, and understand all of the data quite yet.
Now, there are some things that we know and that’s what I’ll spend some time talking about
and at the end of the talk today, I’ll talk about what we don’t know and where some of
the challenges are that we are facing today. So a few more perspectives — when the human
genome was sequenced, this was the size of sequencing facilities around the world. The
sequencing machines were huge. The rooms were the size of this room. They had dozens of
sequencers and they were all around the world. The computers that they used to analyze them
on each given machine, they had a computer. So they had lots and lots and lots of computers.
But the size of the data has changed so quickly. Now, sequencing can be done on something the
size of a thumb drive that you could plug into a laptop. Now the quality of course,
is not the same with something like this compared to something like this at this point. But
it’s getting there. It’s getting to the point that we can sequence our genome with a flash
drive. Now we can’t store the data on a flash drive. The data has to go into a supercomputing
center because it’s terabytes of data. So these are not machines that you can go to
Cosco or Wal-Mart to buy. These are machines that you have to have a lot of expertise to
work with. So we can generate the data faster and the data are really big, really big. And
so it brings a lot of challenges. But with this information, we have learned a tremendous
amount. So this figure is from the National Human Genome Research Institute and my guess
is that you probably saw this figure for those of you who were here last week. Eric Green
who spoke here last week is the director of NHGRI and their group has spent a tremendous
amount of time cataloguing all of the genome-wide association studies that has been done. So
these are assays where people looked at somewhere between 100,000 and 1 million or more bases
in the DNA or single nucleotide polymorphisms. So these are just single-based changes in
the DNA and they’ve looked to see are any of those associated with a disease. What we’re
showing here are all the chromosomes and each little colored dot is showing a particular
association of a SNP or single nucleotide polymorphism with some class of disease. And
so what you can notice is that there are some chromosomes that have particular locations
that have up to six traits — there are certain regions that have dozens of traits associated
with them. This particular region is of the MHC or the major histone compatibility complex.
This is a region of the genome that’s important for our immune responses and so a lot of autoimmune
diseases and infectious diseases and traits related to those associate with that region
of the genome. And so as of this week, this catalogue includes over 1,500 papers and over
8,000 SNPs that are associated at statistically significant level across the genome. That’s
a lot of information that we’ve learned. And the first genome-wide association study was
only published in 2005. And so in the last seven to eight years, we’ve learned about
8,000 variants and their associations with hundreds of traits. This is very, very exciting.
However, what we’ve also noted is that the distribution of how much of heritability of
these traits, these variants explain is very small. So what I’m showing here are the number
of associations, so the number of SNPs that are associated and the strength of those associations.
So these are odds ratios. An odds ratio of one means that there is no increased odds
of disease if you have the variant or if you don’t. And so you can see that — now, these
data are a little bit old, but the trend is the same and it just adds more data. The median
odds ratio is 1.28. That is very small. So, that means that the variants that we found,
all those pretty colored dots on that plot, for the majority of them may explain very
little of the heritability of the traits. So, that means we’re explaining very little
of the actual trait itself and how it’s being inherited. They’re associated but they don’t
explain a lot. Now here’s an example of one of those genome-wide association studies.
So this is Alzheimer’s disease. What you’re seeing here is the negative log 10 of the
P-value which is a measure of the statistical significance and what you want are things
above eight. So we’ve drawn a red line at eight. Anything above eight is statistically
significant across the genome. And you’ll notice, these are all of the chromosomes along
here and each dot is one SNP’s association with Alzheimer’s disease. What you can see
here is there’s this nice peak down here on Chromosome 19. These are above that red line.
So that’s indicative that we have found a signal for Alzheimer’s disease that is statistically
significant across the whole genome. We looked at the whole genome and this is the signal
that we found. Now, I can tell you that that region of the genome is Apo-E. This was a
region — a gene that we knew was associated with Alzheimer’s disease for many, many years,
back to the ’80s. We didn’t learn about it in a genome-wide association study although
this is a good proof of concept that this type of study can work. The odds ratio of
something like Apo-E is very, very large. It’s going to be out here in this region of,
you know, five to 10 to higher depending on the size of the dataset and the type of dataset
that you’re working with. So effects like that are — are fairly easy to find in a genome-wide
association study. A lot of other traits that we’ve looked for like Type II Diabetes, Crohn’s
Disease, Obesity have much, much smaller effects which means that while we know lots of our
DNA variations that are associated with these traits, they’re not explaining or predicting
a lot of the trait. That said, there are a few that we’ve learned about for Pharmacogenetics
or for personalized medicine. So Pharmacogenetics is looking at drug response due to genetic
variation and that’s what we’re using for personalized medicine. This is a figure from
a paper that my PhD mentor published that I really, really like. So it shows this low-hanging
fruit. Some would argue that what we have found so far in all of these genome studies
that we’re doing are the low-hanging fruit. We’re finding things that have the biggest
effects on their own. We’re finding things that we’re able to find given the current
analytic technologies that we’re working with which is why my lab spends so much time working
on new techniques to work with the data and to analyze the data. We want to be able to
find all the fruit. We don’t want the Apo-E fruit. We want all the rest of the fruit.
We want to understand and be able to predict who was going to develop this disease and
who’s not. So like I said, some of these low-hanging fruit for personalized medicine we have found
and they’ve been very fruitful. Here’s an example of a genome-wide association study
plot, so again, these are the P-values. The negative log-based 10 — we want things above
eight and I’m showing in red here, this nice peak down on Chromosome 16. This is a genome-wide
study that I was involved with and the phenotype or the trait that we were looking at is drug
response to the drug Warfarin or Coumadin. This is a blood thinner. So when individuals
develop this condition where they’re unable to maintain the proper level of blood thinning,
they go on Warfarin or Coumadin. It prevents people from going into a clotting issue or
from going into a hemorrhaging issue. It maintains that appropriate level of blood. So what I’m
going to show here are two hits, VKORC1 or Vitamin K Epoxide Reductase is here. This
is a new hit that we learned about in genome-wide association studies. Down here in blue is
CYP2C9. This gene has been known to be associated with Warfarin dose for a decade. It’s way
down here. It’s not significant in a genome-wide study but we know about it. It’s functional.
It actually changes protein levels and it actually predicts what dose a person should
be on to maintain the right level of blood thinning. So why is it not statistically significant
in this genome-wide study? This stunned us and it worried us quite a bit. So we went
on — this is the publication in case anybody wants to look for it. It’s — it’s in the
handout. A few — a subsequent study went back and did the study again with a much,
much bigger dataset and when they did that, that CYP2C9 variant was genome-wide significant.
And so we think in our initial study, our sample size to the number of people in our
study was just too small and we couldn’t get that level of significance that we needed
to be considered statistically significant. However, we know that it’s real. We know that
it’s actually functionally associated with dosing of Warfarin. And so, that led to a
lot of additional studies. We published one in the New England Journal of Medicine back
in 2008 where we specifically looked at VKORC1 which is on the top panel and CYP2C9 which
is on the bottom. On the left, you’re looking at the time to the first INR in a therapeutic
range. INR is a clinical test that they do. When you go on Warfarin, you get an INR measurement
and that’s going to measure kind of how well that drug is titrating in your system. There’s
a certain range that your INR needs to be in for it to be safe and be considered a stable
dose. And if you’re not in that range, they either raise the dose or lower the dose because
they don’t want you to bleed out or to clot. And so it’s really done by a trial and error
and getting multiple blood tests done once you go on the drug. And then what we’re showing
on the right is the time for the first INR greater than four which is way out of range.
So that’s a bad thing. So what we learned is that VKORC1 is associated with the initial
dosing of the drug. So to make sure that you get on the right dose early, the VKORC1 variant
predicts what your dose should be. CYP2C9 is not associated with that initial dose.
It’s associated with the maintenance dose, the dose that you stay on over time to make
sure that you stay in the right range. And so they’re both important but this is a case
where the timing of the dosing and the timing of the experiment plays a role in understanding
which genetic variation is important. So VKORC1 is important at initiation of the dose, CYP2C9
is important for the maintenance dose. Both are associated now — both are actually implemented
in the clinic and we’ll talk more about that. Because of all the studies done on Warfarin,
the FDA or Food and Drug Administration changed the label on Warfarin so it’s actually on
the label now that genetic testing can and should be done if you’re going to go on Warfarin.
Not all clinics are doing that but many clinics around the country are. It’s a test that can
be ordered online. I found a website where you can call now and send in a sample, and
they can genotype you and tell you what your genotypes are so that you could know what
your Warfarin dose is. And I actually know some people who have done not this particular
test but have had their genotyping done in case they had to go on Warfarin later and
then they went into their clinic and said, “Oh, by the way, doc, here’s what my dose
should be.” And — which is a little scary because they’re not physicians. But — but
it’s important to know if you might go on Warfarin because I personally would rather
go on the right dose to start with and not have to get pricked seven times and go up
and down, and up and down, and have a bleeding event and have a clotting event until they
get it right. So we’re definitely able to use these variants in the clinic today. Another
story — clopidogrel — so clopidogrel is a drug that individuals go on, when they’ve
had a heart attack. Typically, you’ll get a stint placement and you’ll go on clopidogrel
to prevent a subsequent heart attack or myocardial infarction. It was learned just a few years
ago that CYP2C19 has a variant that indicates whether a person is actually responsive to
clopidogrel. So if you have a variant in this gene, your body has no response to the drug,
the drug is not metabolized properly and so it’s as though you’re not taking it at all.
So this is their genome-wide association study plot. So again, these are the P-values. We’re
looking for things above eight. The negative log-base 10 of the P-value — let me correct
myself. We’re looking for things above eight in all the chromosomes. And here’s that cluster
of CYP2C19, their functional variants in that region and one particular variant that has
been shown to explain the function of clopidogrel in individuals. And this study was actually
done in an Amish population, an older Amish population I believe from Central and Eastern
Pennsylvania and then, it was replicated or seen again in another dataset of unrelated
cases and controls, so people who responded to clopidogrel and people who did not. And
so this is a known variant that we have found in genome-wide studies and has been validated
and is being used in the clinic. Now a little bit of a cautionary tale — so I told you
earlier that CYP2C9 was not genome-wide significant. A study I was involved with after the genome-wide
study tried to validate the results of CYP2C19. We validated it, but barely. Our P-value was
.003. To be considered significant in a genome-wide study, you need to have seven 0’s before that
three. So it should be .000 — 10 to the minus 8. Our result was not significant if we had
done a genome-wide study, which troubled me, a little bit. So at this point, we were talking
about implementing this in the clinic at Vanderbilt University Medical Center which is where I
was working at the time. And I thought, “The result is not that significant in our study.
I don’t know that we should put this in the clinic and actually treat patients based on
this information because it’s not that statistically significant.” So I was very, very uncomfortable
with it. That said, it was replicated in many other studies and the final decision was to
actually implement that in the clinic. And so this is a project called, “Predict.” They
have actually implemented three drugs at Vanderbilt University, clopidogrel, Warfarin, and Symvastatin.
I’m going to show you a little video clip and some of you may have already seen this
because the media blitz from Vanderbilt has been pretty significant on this topic. [ Video Playing ] >> [Background music] I became a cardiologist
so I can help well people stay well. The reason I work at Vanderbilt is because they have
a commitment to prevention and to personalized medicine. Statin medicines are an important
part of that. Using a patient’s DNA profile, we’re now able to predict which patients might
have serious side effects or complications to a very common statin. That means I can
customize treatment for each patient. This is truly an exciting time to practice medicine. [ End of Video ] >> Okay. So as you can see, statins are another
example. I’m not going to show you all of the data from the statins but basically, with
statin treatment and Symvastatin in particular, there’re people that develop a severe adverse
reaction called rhabdomyolysis, really severe muscular pain. And there’s a certain variant
that can predict who will have that response. And it’s great to know that in advance. We
don’t give those patients Symvastatin. They’re given a different statin — there’s a whole
class of statin drugs and they don’t all respond in the same way. They’re all metabolized differently.
And so knowing this information, we’re able to change the drug that one individual will
get and hopefully prevent one of those toxic side effects from happening. So as I said,
all three of these drugs are currently implemented in the clinic at Vanderbilt and I know a lot
about that because I worked there for eight years. But they’re also being implemented
at the Mayo Clinic. They’re being implemented at Duke University. Mount Sinai is starting
an implementation; Geisinger Clinic is starting an implementation in Danville. They’re — a
number of them — Marshfield Clinic in Wisconsin — so lots of healthcare facilities are starting
to implement these variants that we know are predictive of response. And so it’s a very,
very exciting time. Now, I want to talk a little bit more about the clopidogrel story.
So this was an article that came out in the Pharmacogenomics Reporter back in 2010. So
this is when the Predict study initiated at Vanderbilt and I shouldn’t call this a study
— it was a Predict program. So this was not a research project. This was change in clinical
care based on DNA variation. So these patients were genotyped even before they needed the
drugs. So these were people who were going to clinic for a lot of different conditions
and they recognized that these people were likely to need Warfarin, clopidogrel, or Symvastatin
at some point in their clinical care. And they just started genotyping them and putting
it into their medical record before they needed it. So this is a screen shot of the medical
record at Vanderbilt and you can see it has things like your different diagnosis, what
procedures you’ve had done, the medications you’re on. What they’ve done now is added
another column of genotypes and they’re actually putting the genotypes that we know are predictive
of drug response into the record. So right now, there are only three going into the record
— or four, VKORC1, CYP2C9, CYP2C19, and SLCO1B1 which is the one for the statins. These are
four genes, four variants that are being put in and that’s all. I’ve listed others because
the hope is to be able to put a lot of other variants into the record. But we’re not quite
at that point yet. So they have a few of them in the record. As a scientist, where the P-values
weren’t really significant, it made me very, very nervous and I was very uncomfortable
with it. But for the clinicians, they said, “You know, we practice evidence-based medicine.”
And in some cases, the statistical evidence from your genetic studies is more than we
have for the PSA testing that we’re doing or mammogram screening. Your evidence is just
as statistically significant as much of that and we’re doing that all the time. And the
other thing to keep in mind is, “Do no harm.” So the variants that we’re learning about
have alternative treatments that are safe and FDA approved. And so it’s not as though
learning these variations is causing individuals to have no treatment or to have a treatment
that is risky. It’s just a different treatment. And so, I’ll talk some more about some of
those. So this is the first individual that we know personalized medicine worked for in
the study at Vanderbilt. So this woman had a heart attack back in 2010 and she received
a stint placement and went on clopidogrel. She had another in-stint thrombosis, was re-stinted
and then again, and again, and again, and again. And this woman had eight stint placements,
eight. I started to ask the question whether the course of treatment was appropriate to
begin with. How can someone have this many stint placements and they stay on this drug
that is clearly not working? This person is on Medicare, this is the generic drug, this
is the drug — it’s the first-line treatment. It is what everyone goes on when they have
this event. So they did exactly what they were supposed to do. So late in 2010, this
lady was genotyped as part of Predict and found to be homozygote rare variant. So that
means she has two copies of the variant that says it is as though she doesn’t have the
drug at all which I could have predicted based on her response. However, they now knew this
and so they switched her to a different drug called “Prasugrel.” Prasugrel functions in
a similar way but is metabolized very differently. It is still patented which means it’s not
— there’s no generic form. It’s much more expensive than clopidogrel which is why clopidogrel
is the first-line treatment. She went on the prasugrel and as of two years later, had not
had another event and so she was thrilled. Personalized medicine worked for her; absolutely
unequivocally prevented her from having another heart attack and stint placement. And so even
though this variant in our study at Vanderbilt had a P-value of .003 and was not so significant,
it clearly was significant for this woman and worked for her. And so I was at this point
convinced that this was not a bad thing. This is also being done at a lot of other clinics.
So St. Jude Children’s Research Hospital has a huge implementation project in their leukemia
area. They have changed the face of leukemia in kids around the world. The way that they’re
treating leukemia now based on an individual’s genetic profile is just remarkable and it’s
working tremendously well. Cancer is another area. So this is Vanderbilt’s website but
I was looking — you can find these on Cleveland Clinic, Mayo Clinic. They are all starting
to personalize cancer treatment based on genetics. We can look at what variations either in our
germ line or our DNA or the cancer-tumor DNA and predict whether the cancer will respond
to this chemotherapy, chemotherapeutic, or that chemotherapeutic, whether it will respond
to radiation or not. And so we’re learning a lot about cancers as well. So how do you
decide what to implement? And this — I was in a lot of meetings and had a lot of discussions
because like I said, I was very nervous about the statistical significance. And the physicians
explained there are a lot of things that we use in terms of evidence. So we look at the
effect size. So the effect size is — is that odds ratio or that heritability measure. How
much of the trait does this variant explain? If it explains a good amount which that clopidogrel
variant does, then it’s worth implementing. How statistically significant does it replicate
or validate in other studies? Do we know that it’s functional in cells? Do we know that
it actually changes proteins? All of those things become important. And the other pit
— the other tidbit that is key is what is the alternative course of action. If it’s
just a matter of modifying the dose, prescribing a different drug, or just having a different
therapeutic plan, then why not? If we know that this alternative is also safe, it’s just
not first-line therapy because it’s more expensive, then there’s no reason not to implement these
variants. And I would agree. And so the Warfarin dosing and clopidogrel are being done. They
have a clinical variable equation that is typically used to estimate dose. This has
things like our ancestry, our gender, our age, our body mass index. That’s typically
how a doctor prescribes Warfarin. And so right now, they are using this equation and comparing
it to using the genetics to see does using the genetics get us to that stable dose faster?
And I haven’t seen the results of that study yet, but they are definitely trying to do
that. But the rest of the treatment is staying the same. So you’re monitored with that INR
whether you’re going on a clinical dose or the genetic-predicted dose. And so it’s not
as though — if you’re on the wrong dose and your genetics was wrong, it’s going to be
dangerous for you. The doctor’s going to treat you in the same way, that he would treat you
if they did the clinical dose. And so we should see in another year or so, kind of the actual
statistical results of that analysis. Similarly with clopidogrel, all their doing is switching
to prasugrel instead of clopidogrel, and looking at outcomes over time. And so again, because
this just started a few years ago, we don’t have any data yet to know, did it actually
change the face of therapy for the population of patients? We should know that soon. We
know it changed the therapy or the course for that one person who, you know, came forward
and was interviewed about it. So we should see in another year or two the data to support
whether or not this is working broadly. Okay. So I want to spend a few minutes to talk about
kind of the — the not — the scary part of the truth. So I gave you examples, personalized
medicine is here and it is working for certain variants. But there’s a lot that we don’t
know. So if most of the effects that we’re finding are really small and they explain
something like one to five to 10% of the heritability traits that we know from twin studies done
back in the ’70s and ’80s that the heritability estimates for some of these traits are 40%,
50% or 70% and we have found the genes that explain 10%, where is the rest of the heritability?
So this is a really popular paper that talks about where the heritability might be. This
is the Case of the Missing Heritability. It’s a huge problem that a lot of people are talking
about and working on. We’ve done hundreds of thousands of genomes and we don’t have
the rest of the heritability. Where is it? Well, the paper talks about a lot of places
where it might be, so under our nose is indicating that it’s in the data. We just didn’t do the
analysis quite right and if you’d look at the data slightly differently, you’ll find
the variants that are important. Out of sight is the idea that you know, these genome-wide
chips are only measuring a million variants across our genome. And we have 6 billion and
a lot of them are very rare and so they’re unique to our person or our family or they’re
just not very common in the population. Those variants were not assayed. They weren’t looked
at or investigated in all those genome-wide studies from that NHGRI catalogue. And so
there’s a lot of variation that we just haven’t looked at yet and maybe all the heritability
is in those variants. Maybe it’s in the architecture which is the idea that we have structural
variation in our genes. We have copy number variants, so genes where we have multiple
copies; this is something that you’ll hear a lot about in the sixth lecture of this series.
Dr. Scott Selleck works a lot on copy number variants and I’m sure he’ll talk about those
and also epigenetics and these are kind of changes in the DNA chromatin structure and
histone modifications. These are molecular biology things that we know happen to the
DNA, how the DNA is structured and wound together and how it comes un — how it unwinds before
transcription and translation and become you know, DNA to RNA to protein. There’s a lot
in that structural variation that we don’t fully understand yet and maybe, some of the
heritability should be measured in there. Underground networks which is my personal
favorite and what I’ll spend a few minutes talking about is the concept that underlying
Biology is complex and we have lots of networks and pathways that interact. And if we look
one gene at a time, we’re going to miss all of those complex networks and interactions.
Lost in diagnosis is a concern that we all have and we hope that it’s not the case. But
a lot of these studies are done where people are collected and they are put into bins.
These people have a disease and these people don’t. And pick a disease like Type II Diabetes.
Well, if you know people with Type II Diabetes, you know that phenotypically or symptomatically,
they’re all very different. But in a genetics study, we put them all together. Similarly
with Alzheimer’s disease, everybody with Alzheimer’s goes into one study as cases and then controls
are people who don’t have Alzheimer’s. We know in looking at Alzheimer’s individuals,
they’re very different. Some have very violent tempers. Some have a lot of depression. Some
revert to childlike behaviors and mannerisms. Not all, so is the genetic ideology for all
of those different symptoms the same or different? Probably different and so if it’s lost in
diagnosis and part of the problem is that our studies should have been sliced and diced
into different case sets, our sample sizes are going to get much smaller. But maybe that’s
where we need to go to figure out the genes that explain each of these sets of symptoms.
And then the great beyond — as a geneticist, we tend to ignore the environment because
the environment is so hard to measure. Our DNA is easy. We can get a blood sample or
spit in a tube and get our DNA and it doesn’t change. The DNA that we are born with is the
DNA that we die with, super easy to measure. Our environment changes throughout the day,
throughout the week, throughout the month, over the years. If you ask people, “What environmental
toxins have you been exposed to? What was your diet? What — what toxins have you ingested?
Have you worked near heavy metals? Have you ever been exposed to lead?” We don’t know
that stuff. We don’t remember. I can’t remember what I ate for dinner last night so I’m certainly
not going to be able to give you my last week’s dietary questionnaire information. And so
collecting the environment is really hard. But I would bet that a lot of our genes are
only important in the context of the environment that we’re putting them in which is why you
have some people who smoke and never develop lung cancer and other people who smoke and
they do. And then other people who never smoke get lung cancer. Somehow smoking interacts
with certain genomes and not with others. And so understanding the environment is going
to be key to understanding some of that heritability. But what I spend most of my time working on
and what keeps me up at night is the complexity of Biology, and the fact that we simplify
it in all of our studies and I’m guilty as anyone. Biology is hard and it’s complex and
if we don’t simplify it, it’s too hard to do the analysis. And so if you look — do
the Google search for Biology and Pathways, you’ll find hundreds of images like this that
have genes that explain or interact with other genes. Genes interact with pathways. We have
feedback loops. We have compensatory mechanisms. We have all sorts of things that happen in
Biology that when we look one gene at a time and try to find the one gene that explains
a really complex disease like schizophrenia, that’s why we’re not finding the gene. There
probably isn’t one gene. It’s probably a dozen genes or 100 genes. And so how do we find
the dozen or the 10 or the 50? And the other problem is that Molecular Biology is very
complex. So we’re only focused on DNA. Everything I’ve talked about has been variation and DNA.
A lot happens from DNA, to RNA, to protein, to cells, to organs, to organisms. We ignore
all of that other variability because it’s challenging and for years we weren’t able
to assay the RNA and protein expression well enough to do the studies. Today we are. The
new sequencing technologies have enabled us to actually get whole genome measures of genome
expression and of protein expression and of metabolomics and add “omics” onto any word
and we can pretty much measure it now. And so a lot of work is being done to try to integrate
all of these, “OMs,” the genome, the proteome, the phenome, the transcriptome, the metabolome,
the microbiome, altogether to try to understand if we can understand more about these complex
traits. So actually treating them in a complex way which is probably how we should be doing
it since they are underlying complex traits. So as a very naïve graduate student in 1999,
when we were talking about how one day, we will be able to measure 500,000 SNPs or single
nucleotide polymorphisms across the genome, and won’t that be amazing and then we’ll be
able to understand these complex traits because at that time, studies were very small. So
my PhD project was on 25 SNPs. That was it. 25 SNPs, I studied breast cancer and that
was a big study back in ’99. My grad students I think laugh at me when I tell them my studies
were 25 SNPs because theirs are 13 million and you know, they could do the 25 SNP analysis
on their iPhone or their Android. But still, that was a big study back in ’99. And so I
said, “Well, there’s going to come a time that we’ll have 500,000 but if we think that
interactions are important, can’t we just exhaustively look for all the combinations
of genes?” Because with my 25 SNPs, I was looking for all of the pair-wise interactions
and all the three-way and all the four-way and I just ran exhaustive tests of all of
the effects to see what I could find. My PhD mentor said, “Well, I don’t know. Can you
exhaustively look at 500,000 SNPs?” So I went and did the math, so this is the number of
combinations. If you’re looking at single SNP models, so one SNP at a time, that’s 500,000
tests. And at the time, we were thinking, “Well, maybe for something like cancer, it’s
five SNPs.” So five different variants that are important, that would be two times 10
to the 26th tests. Can you do that? Well back then, we could do about one computation per
second. It was going to take me 10 to the 21th days. And I did not want to be a grad
student for that long. So I said, “This isn’t going to work.” I recently redid the math
because we now have five million SNPs in a lot of studies. We’re up to 10 to the 20th
days. We cannot, even with computers that can do a million tests per second because
our computers have gotten much, much faster. We still cannot exhaustively do all the combinations
in the genome and that’s just looking at five SNPs. Do I really think that schizophrenia
is due to five SNPs? Probably not — it’s probably two dozen or 50 or 100. I don’t know
how many but we can’t look at all of them. So how long is 10 to the 20th days? Just to
give some perspective, the Big Bang Theory, not the show which by the way — this is what
my whiteboard in my office looks like and every time my mother-in-law comes in, she
said, “Oh, you look like Sheldon.” [Audience laughter] And — and I don’t. Anyway, Big
Bang Theory — 10 to the 12th days, that was the beginning of time and it’s going to take
10 to the 20th to exhaustively look at five million SNPs in five-way combinations. It’s
not going to happen. We need to be smarter than that which is fortunate for people like
me who want to come up with smart, interesting algorithms. And so this is a toolbox and what
most people in my field do is they grab the hammer. They do their standard genome-wide
associations study because it works. It found VKORC1. It found CYP2C19. If we use a hammer,
we will find the low-hanging fruit which is true. Everybody in my lab knows how to use
the hammer or the standard statistical test. When it works, it’s fantastic. But it doesn’t
always work. And so we’re trying other tools. We’re developing what we call Meta-Dimensional
Analysis. These are the ideas that we integrate DNA, RNA, and protein all into the analysis
in a more systems biology or a systems genomics kind of way. That’s one approach that we’re
taking in addition to many other labs around the world. We’re doing pathway analysis. So
instead of considering any one gene, we’re looking at the pathway as a whole and we’re
looking to see are there certain pathway effects that we observe more in people with a disease
than you would expect just in a random subset of people without the disease. We’re looking
at genomic convergence. This is the idea that regions of the genome that show evidence of
association from SNP tests where they show evidence in linkage studies which are studies
that were done in families back in the ’80s and ’90s and early 2000s. Do we see regions
of the genome that have increased expression? You know, can we look for those regions that
keep coming up? So that Chromosome 6 region that comes up in a lot of traits — we focus
on that region quite a bit because we see it come up from multiple different types of
data. The biofilter is a tool that we work on trying to integrate information from the
literature. So there are decades of biological information out there in Pub Med and in Journals
and in databases. Can we pull all that information and use that to more intelligently look at
the data rather than looking at the genome broadly and try to test everything which we
can’t do. Let’s focus our analysis to things that we know interact with one another in
biological systems. Do any of those interactions then, associate with our trait? So that’s
another approach we’re taking. And then polygenic modeling as a statistical technique that people
are using to try to look at combinations of variants. And this has been done quite a bit
in — on schizophrenia and in bipolar and in multiple sclerosis and they’re finding
if you take these genome-wide chips that have a million variants and you try to find subsets
that fit in a polygenic model, they can find models that have up to 20,000 SNPs that they
can be sure that the functional ones are in there. They can’t tell you which of the 20,000
they are, but at least they have filtered it down to a smaller set. And if we had 20,000,
we could start to look for interactions there and exhaustively look for interactions. That’s
a much easier space. And so that’s another technique that some people are using to try
to get the data into something much more manageable. Okay. I have two slides that aren’t in your
handout. I decided to add these to have you do a little bit of a thought experiment. And
so one of the challenges that we have is that we don’t know what types of models we’re looking
for. We think they’re complex. That’s the whole reason we’re developing more complex
techniques. We’ve tried the simple stuff, it didn’t work. And so we’re trying something
more complex. If the fitness landscape which is the space of all possible models looks
something like this where you have all models that you could test. These are statistical
models of combinations of genes or of pathways and the fitness or how well these models explain
the heritability of the trait looks like this, it’s a pretty simple landscape to find the
solution. So if you imagine — so everybody close your eyes and imagine that you’re in
a helicopter and someone is going to drop you out of the helicopter safely with a blindfold
on. And your challenge is to get to the peak of this mountain and so you take a step forward.
If you go up, you’re going in the right direction. If you go down, you’re clearly going in the
wrong direction. And so you turn and you keep taking steps in one direction or another only
moving in the direction where you’re going to go up because you’re trying to find the
peak of the mountain. So if you climb and you take your blindfold off and you — anyway
you step, you go down, you’re reached the peak of the landscape. This is how most statistical
step-wise methods works. We make tiny changes to models and see which one improves. We put
this gene in, it makes it better — we keep it. We put this gene in, it improves, we keep
it. Put this gene in, it makes it worse — we take it back out. And we make little tiny
step-wise changes in the hopes that we’ll find the model that explains our trait, disease,
phenotype. Unfortunately, I think the reality of biological landscapes is that they look
like this. So if you were dropped out of the helicopter into [inaudible] Canyon safely,
hopefully you’ll find a peak. But if you took your blindfold off here, you would see that
you’re still in the valley. There are many, many, many other peaks much higher than you,
not to mention there isn’t only one peak. There are dozens of peaks and so if complex
traits are actually like this and there are multiple models, that makes sense. And there
is no one solution. These analyses where we’re making tiny incremental changes and looking
for one model are never going to get us there. We have to do something smarter where we search
more of the landscape and we allow for the — the possibility that there are a dozen
models that explain Type II Diabetes, or 15 models that explain obesity. This idea of
looking for only one is not going to get us there. Now, for things like Warfarin and clopidogrel,
it did. That was a landscape like this. We have found those variants which is fantastic
and its changing lives today. However we’re not quite there for everything and so that
brings us to, what does this mean for you. So, I like to talk about “23 and Me” which
is a company that allows an individual, so anybody in the audience can go home today
and go to this website and order a kit and get your own whole genome genotyping done.
So you can get up to a million variants on yourself. They’ll send you a kit, you spit
in the test tube. They has mouthwash you switch it around and spit in the tube and send it
back and in a number of weeks or months they will send back an email with a code and you
can log on and download your DNA sequence at this million position. So it’s not full
sequence, it’s just a million of the DNA variants. Right now it’s $99. That is less than my car
insurance, so I could go and get my genome done and thousands and thousands of people
have done this, which is super exciting and there used to be a couple of companies that
did it and primarily this is the one now. When you’re on there and this is an older
screen shot, it used to be $299. I think when it first came out it was more like $699 so
it’s really dropped quite a bit. You can learn about your ancestry, and you can learn about
your health, and you get a lot of information by doing something like this. You can also
get your whole genome sequence done, so there are a few companies complete genome sequence
known. So there are a few companies complete genomic and know that you can do a similar
process and get your full genome sequence it costs much more like $10,000. Ozzy Osborne
is one individual who has his done. I would love to get my hands on that data because
he’s a very interesting character that I would love to understand what makes him tic. But
that data is not currently publicly available. But you can do this, today, you could go home
and order a kit and get your genome sequenced, as well. So this is often called “Direct to
Consumer Sequencing” or “Direct to Consumer GWAS Genome Wide Association Study,” what
do you get? So I’m going to — I’m going to kind of give you the pros and cons. It can
be dangerous and it can be very interesting, and it’s all a matter of how you use the information.
So 23 and Me will give back information, it’ll tell you 48 traits that you are the starier
carrier of. That means that you are carrying a variant for a particular trait but most
of these are traits that you need two copies to actually present with a disease. And so
these are really important for people who are in their reproductive age thinking about
having kids and whether if they are a carrier and their spouse is a carrier, then what is
the risk that their offspring will then have the trait? There are 120 disease risks and
conditions that they’ll give you information for. So it’ll tell you how what at risk you
are for cardiovascular disease, Type II Diabetes, male pattern baldness, prostate cancer, ovarian
cancer. A lot of these are in there, they’ll give you 21 drug responses, clopidogrel and
Warfarin and Symvastatin are in there among others and 57 traits. These are things like
freckling, height, red hair — those things, most people already know what the risk is
because you can look in the mirror and figure out what you are. But you can find out if
your genetics explained who you have become and the redhead gene is on there and I’ve
been told by many, it’s a particular mutation that causes the disease of red hair which
I would argue is not a disease, but that’s neither here nor there. Most of these associations
that’s are there are from genome-wide association studies. So remember I showed you the odds
ratios were teeny-tiny that don’t explain a lot of the traits. And so the odds ratio
estimates or the risks that they give you are estimated from data that when it was found
was very, very small. That said as well most of the GWA studies that have been done are
in European-American populations, or European populations and so people who are of other
ancestries, the risk predictions either can’t be done or they are not as accurate because
our DNA varies quite a bit from population to population. So this is a screen shot of
Type II Diabetes risk, this is from their website. This is not my own, I have not had
mine. I don’t have my sequence I don’t have my genome-wide study done. I come from a long
line hypochondriacs and the last thing I need is more data to support that and tell me what
I’m at risk for. I know based on family history and that is sufficient for me personally,
though many of my colleagues have done this and they think it’s awesome and a lot of fun.
But here’s the Type II Diabetes risk. So green is showing decrease risk and red is showing
increase risk and they are showing a number of variants from genome studies for Type II
Diabetes risk, so this particular person is a decrease risk for a bunch of markers and
an increase risk for a bunch of markers. So are you at risk or are you not? Some markers
say you are, some markers say you’re not. What do you do with that? It’s going to give
a risk prediction. It’s going to give you a number and say based on these variants your
risk is two times that of the population, or 1% of that of the population. But those
are hard to estimate given — they’re not conclusive, they’re not all trending in the
same direction. Now, there are many, many stories [inaudible] — there are many stories
of these risk predictions that have come out that they either work or they don’t and it’s
very individual. So I had a colleague that did this. He was told that he was at decreased
risk for male pattern baldness and he was shiny since he was 30. His head is shiny as
can be and so it was wrong. He was also told that he was at increase risk for ovarian cancer
which he found interesting because as far as he knows, he doesn’t have ovaries. This
is a problem. I know other individuals who were told that they were at increase risk
for cardiovascular events. They went in to the doctor and had some tests run and sure
enough they had an increase in Atherosclerotic Plaques and they needed to go on treatment.
They had no symptoms yet but they were nervous. They had a family history, they went in they
got tested and they needed it. I know another person who had to go on Warfarin and he took
this to his physician and said, “Look at this. Please prescribe my dose based on 23 and Me.
I know this data is accurate, use it.” There are lots of stories like this. So it’s not
all bad for some traits it works, for others it doesn’t. This does have things in here
about breast cancer risk, Parkinson’s risk, and Alzheimer’s risk. Now all three of these
are harder to get to, they’re not in your initial report. You have to actually kind
of click extra buttons to access that. Those are some of those trigger diseases that people
don’t want to know. You know, if you ask 100 people, everybody would love to know their
risk for Type II Diabetes or obesity or some of these things that we can prevent although,
we should be eating right and exercising and not smoking or drinking and getting plenty
of sleep, yet not many people don’t do that anyway. So I’m not sure that our DNA profile
telling us that we’re at increased risk for diabetes is going to change our diet. But
I’ve heard one story of a faculty member at Stanford that it did. It showed he was at
risk for diabetes; he changed his diet lost ten pounds. His glucose levels went from slightly
out of normal back to normal. I argued with him that he should have done that anyway;
he didn’t his DNA to tell him. But he used the DNA evidence to make the change in his
lifestyle which made a difference in his — in his outcome. But like I said, people are doing
this also because it’s very interesting. You learn about your ancestry. So one of my colleagues,
her mom is from Mexico; her dad is from the U.S. could see where her mitochondrial genome
came from, from her mother and where it came from originally. And she could look to see
what regions of Europe her dad’s Y — her dad’s regions of his genome came from because
he was also not. You could learn about your family structure. A word of caution there
is that families that do this can learn about non-paternity. So in any genetic study kind
of on average the report is that when we study families, about 10% of our participants we
find evidence of non paternity. But it’s not reported we see it in the DNA. We can tell
that dad is not dad. We don’t ever report that. But I know a lot of families that are
getting this done like for Christmas that’s not the type of gift you want to open. [Audience
laughter] So be careful if you know something — this is going to tell — this can tell
everybody in your family. So if you know just don’t do it. Just have other reasons to do
the analysis. But I’ve heard horror stories about that happening. It can tell you about
dietary risks. So you learn things about lactose intolerance, or your risk for obesity, or
how you can process different nutrients and so that might be useful. You might learn that
you have risks for Celiac Disease and change eat less gluten and maybe that would change
some of your outcome. Maybe you don’t have full Celiac Disease yet but changing your
diet some would prevent it from happening. And then there’s a lot of pharmacogenetics.
So these are those genes that explain drug response they’re in there and for that it’s
very useful because a lot of those are actually predictive and work. So there, I’m going to
wrap up and I want to acknowledge work like this is done by teams of people. So I’m not
even sure all of the teams at Vanderbilt. It was many people that I collaborated with
for a decade and I also, got my graduate degree and my first [inaudible] position there and
so I was there for 12 years and worked a lot with various people. But my lab here at Penn
state has a number of people, graduate students, some undergrads, software developers, data
analysts, program managers, kind of doing this complex analysis. And methods development
is a lot of work and so I just want to acknowledge this is Team Science, genomics is always Team
Science and, you know, I get to be the one who gets up and talks about all the fun stuff
we do. But there’s a lot that goes on behind the scenes that I’m very thankful and grateful
that people do all that work. And so I will stop there. I like to show this slide. A lot
of people criticize the idea of these complex modeling tools because we don’t have good
examples of complex models yet. So why are you looking for them? And I would argue just
because we haven’t found it doesn’t mean it’s not there. That’s why I’m a scientist to try
to discover things and learn things and find things that we didn’t previously know. And
so we use techniques that do evolution in computers, it’s a technique called genetic
programming and we try to evolve computer programs to solve our problems. It’s a really
fascinating area of computer science and it’s a lot of fun. And here’s my lab website and
my email address, if you have questions or want to look up more information about what
we do and I’ll stop there and Barbara can take the questions, thank you. [ Applause ] >> Hold your questions up please so they can
be collected for you. Dr. Ritchie, let’s see — oops, sorry — first question, “In addition
to the medical conditions in your presentation, are there other medical conditions that now
are already being treated with personalized genomic medicine and also what medical conditions
do you think might be added to this list soon?” >> Let’s see, so a lot of different cancers,
some of the infectious diseases are being researched quite a bit. So HIV is one that
we are now starting to understand who has variation in their DNA that even when they
are exposed to the virus, they will never go on to develop AIDS. They are essentially
protected from the infection and so they are treated a little bit differently because they
never actually develop the disease. And then the HIV drug treatment is another one I didn’t
talk about. But we’ve learned about various adverse reactions to HIV drug that are really,
really terrible. HIV drugs are almost worse than the disease in terms of the side effects.
And so we now understand why people are developing peripheral neuropathy. So they essentially
lose feeling in their fingers and toes and in extremities and so they’re given different
treatments. We’re learning about people that have liver failure due to HIV drugs and they
are getting different treatment. And then also on this its hyper sensitivity where it’s
called “Stevens – Johnson syndrome,” it happens from a couple of different drugs. You could
Google it if you want — I thought about putting up the images and decided in case anybody
had a snack with them, it’s not appropriate. It’s a condition where your body breaks out
in these red splotches and hives that they’re in your eyelids, in your mouth, on your lips,
the skin, their hands swell up. It’s a horrible, horrible side effect and we know what generic
variation and HLA predicts it. And so now they are starting to genotype people for that,
so that they don’t give them this particular Abacavir is the name of the drug that has
this horrible, horrible side effect. So, there are a number of areas where it is also being
done. Where do I think it’ll go next? I think more of the cardiovascular drugs because there’s
a lot of research going on there, a lot of the statin-related drugs. I think we’re going
to see more and more of it in cancer because we are understanding more about the cancer
genome which is allowing us to — or I should say actually the cancer genomes, every cancer
has its own genome. But we’re able to sequence them fast enough now that we can actually
kind of learn more about the cancer and treat it appropriately. And then I think we’re going
to see a lot more of the pharmacogenetic trait. So there’s a network in the U.S. called the
“Pharmacogenetics or Pharmacogenomics Research Network” that I’m a part of, and they’re doing
a lot of studies in a variety of drugs to try to understand drug responses. And I think
collectively, we have kind of decided that is where we should emphasize our time because
understanding people’s risk for disease while interesting, if you can’t alter something
to change it, it’s really just interesting or scary. And so we’re spending more of our
time trying to find drug treatment responses that we could change. So depression is one,
there’s a lot going on. When people come into the clinic and they are depressed, they go
on a drug and then they observe. And then they go on another drug and observe and if
it didn’t work they go another drug and it’s just a titration of drugs, they keep trying
different things. And if you’ve ever seen commercials for depression drugs they’ll say
you may become tired, lethargic, sad, suicidal, diarrhea, stomach — it has all these side
effects that if you were depressed to start, it would only make you more depressed. [Audience
laughter] So that’s an area where they are spending a lot of time, trying to understand,
Can we just figure out what you should go on based on your DNA,” and just start there.
And so I think we will see that in the next decade I hope because that’s only a clinical
condition that’s only been increasing and the treatments have not gotten much better. >> National Geographic also does genome sequencing.
How is their report the same or different from 23 and Me’s report. If I want to do only
one which one would you suggest? >> Good question. I have not seen the raw
data from the National Geographic sequencing. My guess is that it’s more focused on population,
variation, and ancestry just based on the types of things that they publish. But if
they are using the same scientific literature that 23 and Me is using, if they are doing
the same type of clinical profiling then I would say do whichever one is cheaper because
the data or the data and for certain things it’s predictive and for other things it’s
not. And so I would — if you want to do it, spend less and get the accurate and inaccurate
information that you can process at your leisure. >> Because you have a complex landscape to
analyze, can you not use Chaos Theory? >> Oh yeah, and a lot of investigators are.
So it’s not something I’ve done but a lot of people are using fractiles and Chaos Theory
in doing, in kind of really sophisticated Computer Science modeling and trying to use
that to understand that landscape. That’s a great question and it is being done just
not by me. >> It seems that there is a great need for
genome data for future research is it feasible or desirable to develop a genome bank either
pre or post mortem that would allow people to donate their genomes for research? >> Oh, great question. So yes it is definitely
desirable and it’s happening all over the place actually. So Francis Collins who is
currently the director of NIH has been talking for years that we need the Million American
Bio Bank. So we should try to collect a million Americans and put them in a bio bank and that’s
a really expensive project to do. And so what’s happened instead is that healthcare facilities,
many of them in the U.S., are starting their own bio banks, linked to electronic medical
records. So a lot of the work that I talked about from Vanderbilt, we did because we had
a DNA bank of patients and it’s all linked to their electronic health record. And so,
I’m part of a network in the U.S. called the “Emerge Network: which is electronic medical
records in genomics. So they are comprised of Vanderbilt, Northwestern, Geisinger, Mount
Sinai, Marshfield, Group Health, and two more. Who am I forgetting — the Mayo Clinic and
— I’m forgetting the last one. Anyway there are seven in the U.S. and now actually there
are children’s hospitals. So Children’s Hospital Cincinnati, Philadelphia, and Boston are also
now part of this network and they are all creating bio-banks linked to their electronic
health records. And each one I think collectively, now they are somewhere on the order of 250,000
samples in this bio — all of the bio banks collectively. And there are new ones starting.
So Geisinger Clinic has one. Currently it’s only running out of Danville. But I imagine
that over time, it will span to all of their satellite clinics and Penn State Purdue Medical
Center started their bio bank and their Institute for Personalized Medicine this fall. And so
they are starting to accrue patients into a bio bank and then all of their data will
be linked to their electronic health record. And so I think through these medical systems
doing bio banks, if they team up, we will then get the million-person bio bank and have
decades of health information from their health records. So yeah, it’s a great, great idea,
we are currently doing it — we need it because we need big data in order to find some of
these complex models. >> If the P-value is not significant, what
pushes you to apply the results in the field? Is it the gene sequencing company or the pharmaceutical
company? >> No, it’s definitely not the sequencing
company or the pharmaceutical company. I think what is pushing — the things that are not
so significant that are being implemented, I think the reason is that if you go back
into the lab and do the functional validations, these variants have actual functions. So they
actually do something to change the protein. And so, they can demonstrate in the lab that
even though the statistics didn’t show that it was really strongly statistically significant,
the effect is real and has a function. And so I think they’re using more biological evidence
rather than statistical evidence to push those into the clinic. >> I am one of seven siblings. My father’s
family has a history of some Type I Diabetes. Three of the seven siblings were diagnosed
with Type I within two months, at ages nine, 16 and 27 and 37 years ago — at ages nine,
16 and 27 and that happened 37 years ago. The other four siblings never contracted to
the disease. What happened? >> Well, my prediction would be that Type
I Diabetes like most complex traits are complex. It’s not only one gene; it’s a combination
of genes. Type I is less — we think is less due to environment because it does happen
typically at an early age where our environment is fairly controlled. So children don’t have
all of those toxins that we adults expose ourselves to. And so typically I would say
it’s, you know, a complex array of genes that explains Type I and, you know, siblings share
on average, the same percentage of their genome. But what chunks of the genome are very different?
And so it sounds like some of the individuals in the family got the wrong combination of
alleles or SNPs that led to the traits happening and the other individuals were protected.
It happens a lot. So these are those traits, they are complex. They don’t perfectly segregate
in families and that makes it a lot — a lot harder but more interesting to study. >> With shortages of specifics drugs in the
market now that have produced measurable increase in mortality because alternative treatments
were not as successful, how will the pharmaceutical market need to change to increase personalized
care? >> That’s a great question. So what a lot
of companies have started doing recently because they’ve been fighting personalized medicine
for a long time. You know, drug companies want to design a drug that can be marketed
to the largest number of people possible. And that’s how they make their money by having
huge market. Personalized medicine is the complete opposite of that where we have drugs
that are for individuals or small subsets. But, what they’ve realized is that a lot of
drugs are failing during clinical trials. So it takes something like 15 years from like
initiation of a drug design through all the clinical trials and FDA approvals and it’s
somewhere between seven and 15 years of different numbers. And so what they found out is that
if they had done the clinical trials differently and they had subsetted their patients based
on genetics, some drugs would not have failed clinical trials. So they found — you know,
there are certain people that clopidogrel for example, if it doesn’t work for people
who have that variant, if those are the people who are in the trial for efficacy or how well
the drug works, it’s going to fail because it doesn’t work for them. If you happen to
be looking at the side effects, you know, the phase two where we’re looking for adverse
events, then if you have people who have the risk variants that’s going to make them have
this hypersensitivity reaction, the drug is going to fail. If you can genotype people
and then do the analysis of the clinical trial based on genotype and figure out it doesn’t
work for these people, toxic side effect for these people but for these people it works,
more drugs could get to market and so a lot of pharmaceutical companies, GlaxoSmithKline,
Roache, Bristol Meyers Squibb — they’re all starting to do pharmacogenetic-based clinical
trials, so that they can figure out which drugs will work for what subsets of their
patient population. Because they’ve realized even though the target market will be smaller,
more of their drugs will make it through because it’s a multimillion dollar industry to go
from drug design through marketing like to actually pass the trials. So — so they’ve
realized sort of the bang for buck is to find who it will work for and market it for them.
And so I think it’s already been changing and it will continue to change. Now, I don’t
get think we’ll get to the point that — I’ve heard some people talk about — you know,
they’ll sequence the genome and then they’ll go in and design the drug for that person.
Companies aren’t super excited about that. [Audience laughter] The market of one is really
cost-prohibitive. But the market for, you know, this age group, this ancestry, this
body mass index, then we can start to get to large enough groups of people that would
matter. >> Since we know that the key to personalize
medicine will be in gene environment interactions, what is the current status and strategies
for gene environment research? >> Great question, so a lot of epidemical
logic studies are trying to do a better job of collecting the data so that people can
do the analysis. So a lot of the statistics are already really well developed to do the
analysis. We just don’t have the data so there are studies like the [inaudible] which is
running out of the CDC, the Women’s Health Initiative, the Nurses’ Health Study, the
Physicians’ Health Study — lots of studies were there and they were collecting a lot
of environmental data. But where I think it’s going to work and I’m super excited that Penn
State Hershey, our sister campus listened to me. I’ve been saying this for years and
when I was at Vanderbilt, nobody would ever listen to me. I think what we need to do — so
we’re building these bio banks relate to our health records. Our health records have nothing
about our environment, right, other than maybe the doctor will write, “57, female smoker.”
But then if they say quit smoking, if you run an algorithms to find smoking, it’ll find
it. And then you have to find an algorithms to look to see did they quit or did they start.
So, my idea was that I go to the doctor and I sit in the waiting room forever, bored to
death waiting to see the doctor and then, I have to get a lab run. And so I go wait
in the lab for another half an hour and I’m just sitting. And sometimes I remember to
bring something to read and sometimes I don’t. Well if we’re banking these peoples DNA anyway
and we have their health record, why not give them an IPad with questionnaires that they
can just sit there and go, “I ate this. I work outside. I sit at a computer. I painted
with lead this week, whatever,” and you could answer the questions about yourself every
time you go to the clinic. It would give us something to do and we would get environmental
data linked to our DNA and our health record. And so Penn State Hershey is actually implementing
that in the clinics. They have tablets that they are giving people who are enrolled in
the bio bank that they can answer they’re health questionnaires, dietary questionnaires,
family history questionnaires, and they’re going to capture all that information so that
it will be linked to the DNA and to the health record. And so I think that’s where we need
all the medical facilities doing this banking to switch to that or do the online web forms.
And most health centers now have the [email protected] or [email protected] They have a website
where you can log in and message your doctor, and do your appointment, and look at your
health record. Some people have free time and they would love to log in and enter stuff
about their diet or enter stuff about their exercise that week. And so if we give participants
either the ability to give us the data, just make it easy for them. Then we’ll get the
information and then we can do the genome-environment analysis much more effectively. So I’m thinking
the more we see of that, the better studies will get. >> This question is about information flow
in the opposite direction. What is the mechanism for providing clinical useful new medical
genomics discoveries to doctors who treat patients? >> That varies quite a bit from facility to
facility. It’s a big concern at the moment because most physicians today were trained
in a time where we didn’t know much about the genetics. so most of the clinical training
up until about maybe three years ago when the medical schools curriculums changes, the
genetics that they learned were modillion genetics, cystic fibrosis, Mannington’s, these
really rare diseases, Down’s syndrome and they learned how to predict risk based on
that. This new stuff, this is new. So, even the medical curriculum that they changed three
years ago doesn’t have all of this in it yet. So, one thing that will change is the training
of doctors will shift so that the doctors get more of this information. that said, doctors
are thrown so much information about treatments, about new therapeutic regiments, how to change
diet and exercise and they have thrown so much information that to expect them to stay
up with literature for about all these genetics findings is just unrealistic. And so what
a lot of places are working on now are these implementation procedures in their electronic
medical record systems that will alert the doctors about particular risk or genetic variants.
so I know at Vanderbilt and I’m not meaning to talk too much about Vanderbilt it’s just
the one that I know the most about since I worked there, when somebody goes on or is
going to be prescribed clopidogrel for example, if they were genotyped and the doctor goes
in and they click around, they say, “Heart attack, prescribe clopidogrel this dose.”
they hit enter, a big red warning box pops up on the screen. It says, Warning this patient
has been genotyped. Their data show that they will not respond to this drug. Would you like
to prescribe prasugrel instead?” And it has a link that they can click and read more if
they want which most of the time they don’t have time because they have nine minutes per
patient. And so, they’ll just make a decision to accept it or not and in most physicians,
at least in Vanderbilt because in the clinics, they prescribe clopidogrel. They’ve had a
training where they said, Here’s the evidence. Here’s the data.” Our informatics tool is
going to tell you what to do. It’s up to you to do it or not. But the message is being
given to physicians in a really easy to understand way at the point of care and I think Geisinger
are working on that, Mount Sinai is working on that, [inaudible] is working on that. how
do we get the information to the doctor about a particular variants that we want to implement
and I think that’s the only way that it will work, because the field is changing so fast
that you can’t expect the physicians to get it any other way other than the research,
once it’s decided that it’s going to be implemented put it in the system they’re already working
in and it can pop up in their face here, flashing lights do this don’t do this treat the patients
differently so I think that it’s going to happen through this medical informatic tools
that are being developed. >> Have researchers been able to analyze traits
related to granular interaction in the metabolism, heritable thyroid disease for example? >> Yeah, a lot has been done in thyroid disease,
I actually was involved in one with hypothyroidism, there are variations that are starting to
be identified that explain kind of risk for hypothyroidism and hyperthyroidism. If you
— so whoever asked that question, if you go to the GWAS catalogue website you can search
by disease and it has a big box that you can just scroll through and look for the particular
diseases of interest. And then you can click on that and it will pull out all the papers,
another place to look that I really like, it’s really easy to use it’s called “Snpedia,”
it’s S-N-P-E-D-I-A. It’s really user friendly on the web and you can enter a disease, and
it will give you back all of the genetic associations that have been found recently in the last
10 years or so. >> As usual we’ve had so many good questions
and we’ve run out of time, if you haven’t gotten your question answered, you can ask
Dr. Richie come up front ask her after we give her a big thank you. [Applause] Dr. Marilyn
Richie. >> [Applause] Thank you very much. [ Silence ]

Leave a Reply

(*) Required, Your email will not be published