Introduction to Genetics and Evolution is a college-level class being offered simultaneously to new students at Duke University. The course gives interested people a very basic overview of some principles behind these very fundamental areas of biology. We often hear about new "genome sequences," commercial kits that can tell you about your ancestry (including pre-human) from your DNA or disease predispositions, debates about the truth of evolution, why animals behave the way they do, and how people found "genetic evidence for natural selection." This course provides the basic biology you need to understand all of these issues better, tries to clarify some misconceptions, and tries to prepare students for future, more advanced coursework in Biology (and especially evolutionary genetics). No prior coursework is assumed.
Genome-wide Association Studies (S)
Loading...
Del curso dictado por Duke University
Introduction to Genetics and Evolution
532 calificaciones
De la lección
Genetics III
This module delves even more deeply into the complexities of the genetics underlying traits,the origin of genetic variation, and how "complex" traits (ones controlled by multiple genes) are studied genetically.
Conoce a los instructores
Dr. Mohamed NoorEarl D. McLean Professor and Chair,
Biology
Hello, and welcome back to introduction to genetics and evolution.
We've been talking so far about mapping complex traits.
We first talked about some of the principles associated with such mapping of
complex traits.
And we've also talked about how you do it using a cross, and
we used this trajectory of associations string.
Now lets look at an alternative approach.
You know with simple genetic traits we've talked about mapping
using data from overall populations.
Well we can do the same thing here, in the case of complex traits.
Now you may be thinking first, why would we wanna do this?
Well, when your QTL mapping across a pedigree,
what happens is you're localizing the genes affecting this difference
between the original strains or people.
So let's say you're using a pedigree, you're looking at where are the genes
contributing to breast cancer predisposition in this particular family.
So let's say you're mapping a gene and
let's say in this case it's diabetes rather than breast cancer in one family.
And you map it to chromosome 18 arm p location 22.
Let's say there's a very strong effect next to a marker.
Then you look at another family that has diabetes and
you see absolutely no affect to this region.
Even though they have a lot of diabetes and
even though it is clearly running in the family, so it has a genetic component.
Why would you see this dichotomy?
Why would it have a strong effect in this family and no effect in this family?
Well the answer is, again,
there are many different genes that can contribute to disease predispositions.
And you may see in one family there's a particular mutation that's contributing
to diabetes predisposition,
and there's a completely different one in another family.
So when you study a trait in one single family,
you're only identifying what's causing that variation within that family.
It's important, now, to look at the broader population,
if you want to understand what causes diabetes more broadly.
So the other approach that people use is this mapping
in populations by association study.
And a common variant of this is referred to as the genome wide association study.
This has become extremely popular lately with the ability
to examine markers across the whole genome simultaneously.
And of course with having a human genome sequence.
We discussed this in general in the context of simple traits.
And the same principle applies as before.
If a marker is close to the disease causing gene,
then individuals having one allele will be more likely to have the disease than
individuals having the other allele.
So there is an association between genotype at the markers, and
likelihood of disease, or whatever the phenotype is.
The same principle is true not only in simple genes and things, but
also in multiple gene traits, such as disease.
So what people do, especially with this genome-wide association study aspect is
they test many, many, many markers, right?
Not just three, five, ten, 100, or even 1,000, but many, many markers.
Well you'll find is that some will have an allele associated with the disease.
Now they're not gonna have huge associations kind of like some of
the things we talked about before.
You may see something like this.
This would be a very large association in this case.
That big A big A individuals have a probability of getting the disease of 10%.
Little a little a individuals have a probability of getting the disease of 1%.
That actually is a very large effect relative to most of these sorts of things.
Now what happens is you do some sort of statistical test such as maybe
a chi square.
And this will say that this difference is unlikely to have happened by chance
given that you've sampled hundreds or thousands of individuals and
you see this big of an association.
In contrast, many other markers will have no association.
So, this b marker, for example, may have probability of gaining disease of 1.5%.
Little b, little b has a probability of 1.6%.
So, the difference between big B, big B and little b,
little b is pretty trivial in terms of likelihood of getting the disease.
Now, we discussed LOD plots when we were talking about genetic crosses,
same thing would apply with pedigrees.
We use something very similar with association studies.
We have these sorts of plots, excuse me, of association.
But just like with simple traits,
you can actually be much more fine scale because you're looking
at the scale of the distance between these recombination hot spots, just as before.
So, this is an example.
This is a mapping study from February 2012 and
published in Nature Genetics of Azospermia in Chinese men.
So, these are people who are having sperm difficulties.
And you see here this LOD score on the y axis and
position of the chromosome on the x axis.
And we see here there's a bunch of markers that have a strong association
with this trait.
Whereas out here, there are markers that don't have any associations.
So if we're going to do something along the lines of this A and
B, this might be the big A, big A individuals, right, versus little A,
little A where's there's a strong association.
This might be where B is where's basically no association of the trait.
Okay.
Now what often happens,
you find, since this is a complex trait, not one but many possible factors.
So this is from that same issue of nature genetics.
This is just a title of paper, a genome wide association study in Han Chinese
identifies multiple susceptibility loci for IgA nephropathy.
And you see there's one on chromosome 8, one on chromosome 17,
one on chromosome 22.
Again we're looking at the case of complex traits, where there are many, many genes
that have allelic variation, or genotypes that are associated with the phenotype.
Now there's a problem that comes out with genome-wide association studies.
This problem is the issue of multiple comparisons.
Now those of you who are at Duke, have seen this in lab, but
let me tell you more generally.
The chi square test will give you a probability of getting
an association by chance.
I just mentioned that in a previous slide as well.
Now, if you roll two dice,
imagine you have something that's just standard six sided dice.
You roll them and you get snake-eyes twice in a row.
Are they loaded?
Well you may think the answer is yes because,
the probability of getting snake-eyes twice in a row is very low.
In fact it's about 0.1% chance that you'd get snake-eyes twice in a row.
It's not zero, but it's very, very unlikely.
So let's think of this in the context of looking at
markers across the whole genome.
How likely is it that you get something very improbable?
Well, if you rolled the dice a million times, what is the chance you get
snake eyes twice in a row sometime over the course of that million?
Well the answer is it's about 99.99% chance that you should get that
somewhere in there.
This is a problem associated with genome wide association studies.
That if you're testing a million SNPs,
a million of these markers, for an association with the disease of interest
you're very likely to get an association that looks strong by chance,
when you do the statistical test, but there's no biological basis to it.
That, in fact, this region has no effect, whatsoever.
There's no key TL there that's contributing to the trait of interest.
But since you're testing so many times, the odds of you getting something
that seems unlikely enough to be suggestive of a real effect is very high.
So as a result instead of using the standard five percent probability cutoff
that people often use in statistics,
people often use things like a .00001% probability as a cut off.
This is a way to work around this problem.
So again at that point if you see an association you have
very high confidence that there's probably some sort of gene near there.
The downside to this is that it requires a very strong association.
Probably, and this probably happens a lot in genome-wide association studies,
you actually see some associations, but you throw them out because you say
not strong enough to be statistically convincing.
So we end up losing a lot of good associations because we can't
distinguish the good ones from the bad ones except in these very strong cases.
So let's contrast these two types of studies, Pedigree cross versus Population.
Again with a Pedigree across you are finding genetic factors that
cause differences in that family.
So you have high power for a small number of genes because probably within
a particular family there is not tons of genes with allude variation contributing
to diabetes, breast cancer, and nephropathy, whatever.
In contrast, a population association study is trying to find
genetic factors causing variation across the whole population.
Now typically here,
you have much lower power because you're looking at different causes throughout.
But you can find potentially more genes and,
if you do find them, you can localize them much more precisely as I mentioned before.
You may be wondering why is it that you have so much lower power?
Aside from the statistical aspect, why do you have lower power?
Well, there's another facet in here.
In crosses and pedigrees,
you're limited to mapping the genes with effects in that crosses.
You're limited to mapping genes with effects in that cross.
But, the alleles you're mapping are initially abundant, right?
That if you're crossing, say, an individual big Z,
little z, cross the little z, little z, and
let's say this is the one that actually causes the mutation.
In that initial cross, the fraction of individuals that have it is 25, or
the fraction of chromosomes, excuse me, that have the mutation is 25%.
So your ability to detect it is very high.
Let's say in a population though you can theoretically map alleles at
any gene causing this effect.
And there's probably a lot of them doing it.
But you may have many different
very rare variants causing the same effect or disease.
So in this case, let's say for example this capital A and
this capital Z both cause the disease.
They're each initially very rare, so it's very hard to see their effects.
It's because of this rarity.
Now genome-wide association studies work well
when you're mapping common disease variants.
But an open question is are most disease variants common?
In fact, several people have argued that in fact
most disease causing variants are quite rare.
Those often different rare mutations that cause the same disease.
I refer you in this context to an interview that I have online with
Professor David Goldstein from Duke University about this idea,
about rare variants causing disease.
There's another confounding factor that comes along with both of these
kinds of studies.
That is you can see differences among ethnic groups for associations.
Sometimes a marker allele's associated with a disease in one ethnic group, but
it's not associated when look at another ethnic group.
That if you map something with the Hahn Chinese,
you may see an effect associated with this particular region.
In contrast when you look at say, a Latin American population,
you may not see that association.
Let me give you a real example of this.
This is from Type II Diabetes.
In this case, the initial study that was done, was done using East Asian subjects.
What they found is that there were four SNPs that were identified,
that have a pretty strong association with this diabetes,
in this set of East Asian subjects.
In contrast when they looked in European populations there was one particular SNP.
One on chromosome 11 that had no detectable effect in Europeans.
So clearly there was some sort of difference between these two groups.
There are many possible explanations for this.
One of them is we mentioned before epistasis.
There's interactions between genes that, for example,
another variant may be very common in Europeans that makes it so
this particular gene variant that's near this SNP that's causing diabetes in
East Asians, doesn't have that effect in Europeans.
That's one possibility.
Another possibility is the allele at that gene which is causing diabetes in
East Asians just isn't found in Europeans populations, or is very rare.
There's a lot of possible explanations out there.
But importantly, we can't immediately extrapolate and say oh, there's the gene
that's causing this disease, and just assume that that's causing it across
different families, across different ethnic groups, across different economies.
So, mapping is often a very complex task, but
it's obviously very worthwhile, because we have made in roads in
identifying many genes that are associated with disease.
Some of them with fairly large effects,
though many with much more moderate effects.
These are the genes that are actually used in those studies that you get from
companies like 23 and Me, that takes examples, or
takes samples of your saliva, and tell you your disease predisposition.
They're using this sort of data to tell you what your likelihood
is of getting the.
Thank you very much.
I hope you enjoyed this video.
Explora nuestro catálogo
Inscríbete de manera gratuita y obtén recomendaciones personalizadas, actualizaciones y ofertas.
Coursera brinda acceso universal a la mejor educación del mundo, al asociarse con las mejores universidades y organizaciones, para ofrecer cursos en línea.
© 2018 Coursera Inc. Todos los derechos reservados.
