Multifactorial Inheritance, by Dr. Bruce Korf.
In collaboration with the University of Alabama at Birmingham. My name is Bruce Korf. I'm a Medical Geneticist
at University of Alabama at Birmingham. This talk will focus on the principles of multifactorial
inheritance. We'll look at the evidence that supports a multifactorial mode of inheritance,
some of the models that explain multifactorial inheritance, and then talk about what is known
about the genetics of common disorders. The paradigm that u
nderlies the integration
of genetics in medical practice is that we're all born with a genetic liability, sometimes
overwhelmingly so, causing a genetic disorder like sickle cell anemia. But most of the time,
much more subtle. And it requires the passage of time and exposure to environmental factors
in order to transition from what might be described as a pre-symptomatic state ultimately
to a disease state. The hope is that if we could identify the genes that contribute to
this liability, we mig
ht be able to help avoid the exposure to environmental factors and
reduce the likelihood of transition to disease. Or if that transition should occur, understand
better how and why disease has occurred, and perhaps have better approaches to treatment. The evidence that multifactorial inheritance
is occurring is that a trait has a tendency to recur within families more frequently than
might be expected due to chance, but on the other hand, does not follow the principles
of Mendelian genetics. For
example, a 50% recurrence risk for a dominant, or a 25% risk
of having affected children if both parents are carriers for a recessive. Multifactorial,
as the name implies, involves a combination of the action of multiple genes interacting
with one another and/or with environmental factors. In most cases, the specific genes that underlie
multifactorial traits are not known, and genetic counseling for multifactorial traits is based
on empirical data. These are fairly typical data for congenital a
nomalies that are attributed
to multifactorial inheritance where you see a recurrence risk in a first degree relative,
that is to say where a parent or sibling is affected, is in the range of 3%. And you'll
note that the risk dilutes very quickly as one goes to more distant relatives. What kind of evidence would support multifactorial
inheritance? One would consist of identification of familial clustering. Geneticists use the
variable lambda to indicate the risk of relatives affected with a trai
t compared with the population
risk. Lambda sub R is the generic case where relatives of type R are compared with the
population risk. Lambda sub S is a commonly used variable, in which we're looking at the
ratio of the risk in sibs compared with the population risk. In the case of cystic fibrosis, which is,
of course, an autosomal recessive trait, the risk in sibs if both parents are carriers--
that is, if a child has already been born with CF-- would be 0.25, or 1 in 4. The
risk in the popula
tion, at least of northern European descent, is 0.0004 and hence, lambda
sub S is a very high number. For Huntington disease, an autosomal dominant, the risk in
sibs, of course, is 0.5. The risk in the population is about 0.0001, so lambda sub S is about
5,000. The table shows several examples of congenital
anomalies or other multifactorial traits, when you see that the lambda sub S is in the
tens, as low as 16, as high as 49. Nowhere near as high as the autosomal recessive or
autosomal dominant
examples that we've shown. But of course, the risk would be one if the
risk is the same in sibs as in the general population, which it isn't for these disorders. Another form of evidence that would support
multifactorial inheritance involves comparison of the rate of concordance of a trait in monozygotic
twins as compared with full sibs. About 70% of twins are dizygotic, coming from two separate
fertilization events, and 30% are monozygotic. The monozygotic twins, of course, are genetically
ide
ntical. Sometimes, but not always, one can tell the difference based on the various
fetal membranes. The table at the right shows percent concordance,
either in monozygotic twins or sibs, for a variety of disorders. You note two things.
One, that the concordance rate for twins is in all cases higher than it is for sibs, and
second, in no case is it 100%. Therefore, these traits are not completely determined
genetically because if they were, monozygotic twins should have 100% concordance. But on
the other hand, the substantial increase
in the rate of concordance in monozygotic twins compared to sibs supports that there
is a genetic component to these conditions. Another approach, which is confined to measurable
traits like height or blood pressure, is the measurement of heritability. This is a statistical
concept in which the variance in the phenotype of a trait, which is V sub P, is partitioned
into genetic variance, which itself consists of additive genetic variance-- how particular
g
enetic traits add to one another-- and then deviation of additivity due to dominance and
epistasis. Environmental variance is partitioned into strict environmental variance and interaction
between environmental factors. Then there's a covariance of genes in the environment,
and the measurement variance. Heritability in the narrow sense is defined
as the ratio of additive genetic variants to phenotypic variance. Heritability in the
broad sense would be the ratio of all genetic variance to phenoty
pic variance, and it's
the latter that more typically is used in human genetics. One can measure the degree
of correlation of a quantitative trait in various individuals with specific degrees
of relationship. For monozygotic twins, the correlation would be a direct measurement
of heritability. For either comparison of two sibs or dizygotic twins, heritability
is twice the correlation coefficient because these sibs will share about half their genome. The same applies to comparing a parent and
an
offspring. Comparing a parent average between mother and father and an offspring is R over
the square root of 0.5. Comparing first cousins is 8 times R, and uncle-nephew, for example,
would be 4 times R. Well, how does one conceptualize the various models for multifactorial
inheritance? This example shows a simple model referred to as the additive polygenic model.
We'll consider a quantitative trait locus, in this case for height, considering two hypothetical
gene loci, A and B, with dominant an
d recessive alleles depicted as either big A or little
a, big B or little b. We'll make the assumption that all individuals
in the population would have a baseline height of 150 centimeters. And then, depending on
how many dominant alleles they have-- and it doesn't matter if it's an A or a B-- 2
additional centimeters of height are added. If we assume that the allele frequencies for
big A and little a or big B little b are each 50/50, then we can say that there will be
relatively few individual
s who are homozygous-- little a little a, little b little b-- and
they don't get any addition to their 150 centimeters, and so will be the shortest individuals. To the far right in the diagram there will
be also relatively few, but some individuals, who are homozygous dominant for both A and
B. They'll get the largest boost to height of altogether 8 centimeters for having four
dominant alleles. The most common will be having two dominant alleles. It could either
be one A one big B, it could be t
wo big Bs or two little a's. Either way, 154 centimeters
will be the height. And there will be some individuals who either get a big A or a big
B, and they'll be 152 centimeters, and some who get three dominant alleles, one big A
and two big Bs, or two big As and one big B. Well, you can see how you get the semblance
of a bell-shaped curve, even in this simple model. So it is possible, especially as you
add additional genes to the model, to get a pretty good semblance of a bell-shaped curve
that
would approximate the kind of data one might see in the field. Well, this works for
a strictly quantitative trait, like height for example. But what about a trait where
it is essentially an all or nothing phenotype? A good example would be spina bifida or cleft
lip, for example, where there aren't degrees of spina bifida, but rather it either occurs
or it doesn't occur. Now, there can be differences in severity, but the fact that it occurs or
not really is an all or none phenomenon. The thresho
ld model has been formulated to
account for this kind of situation, and what it posits is that there is a more or less
bell-shaped curve of liability towards the trait in the population. Some individuals
have very low liability, most somewhere in between, and then there are some individuals
who have a very high liability. Now, this liability can be accounted for by any combination
of genetic and non-genetic factors. The model posits that there is a threshold, and that
if that threshold of liabil
ity-- which would be contributed by both parents to a particular
child-- is exceeded, then the trait occurs. So in other words, the combined liability
of mother and father, if it's to the left of this line, will not result, say, in spina
bifida, but if it exceeds this threshold then that trait will occur. Exactly how it occurs
in any particular pregnancy does not need to be the same from one to another. It could
be a very substantial environmental exposure but relatively small amount of genetic
liability,
or it could be overwhelming genetic risk and relatively little environmental contribution.
This model mixes genetic and environmental factors into one bin, which is referred to
as liability, but it does explain how an all or none trait could result from a multifactorial
inheritance model. Well, these are all theoretical models, but
how do we get to the point of actually identifying the genes that are associated with the common
disorders that might be of greatest interest in study of m
ultifactorial inheritance conditions
like diabetes, or hypertension, or asthma? Well, for a long time these have been elusive,
but in recent years substantial progress has been made and it is based on the so-called
common disease-common variant hypothesis, which posits that common diseases are accounted
for by genetic variants that are found in 1% to 5% of the population. Now, we don't necessarily have to identify
the specific genes that are responsible for these liabilities to common disease. T
he phenomenon
of linkage disequilibrium says that for two very closely linked markers-- let's imagine
the red one is really the marker associated with disease-- if you have a closely linked
marker, say the blue one, it may typically travel together with the red marker in the
population and serve as a kind of surrogate marker for it because it is very closely linked
with the actual gene in question. And why this is important is that one does not necessarily
need to study genetic markers that, in
themselves, are the ones that are accounting for risk
of common disease. You can also study something very nearby and
have it serve as a surrogate marker for the marker that you're actually looking for. The
typical way that one searches for the association of a genetic marker, whether it's one nearby
the one that's responsible for disease or the one responsible itself, it's through an
association study and a case control type study. And in recent years it's become possible
to identify genetic ma
rkers all along the genome called single nucleotide polymorphisms,
typically abbreviated as SNPs, as it said. An example would be having an A or a C at
this particular base in whatever genetic region this might be. And one expects to find a single
nucleotide polymorphism in an individual roughly once in every 1,000 bases or so. In this case,
we can define having an A as allele 1 and a C as allele 2 here, and then do a case
control study. We hypothesize that allele 2 is associated with asthma. So
we look
at 100 people with asthma and 100 who do not have asthma, and we determine whether allele
2 is present or not present in individuals with asthma or without. We notice that 30% of those with asthma had
allele 2, but only 10% of those who do not have asthma have allele 2, which would accord
with our hypothesis. Well, many associations have been achieved for type 2 diabetes based
on this kind of approach and for, indeed, many other disorders, as you'll see in a moment.
In the early days th
is was done by taking candidate genes. Genes that were, for whatever
reason, physiological evidence, for example, assumed to be associated with a particular
trait. And one could determine through a case control study if they were or were not associated. This did show some success, but the problem
with that approach was that one had to know in advance of which genes might be most likely
to be associated, which usually that list was relatively small. And furthermore, in
effect you really weren't l
earning very much that you didn't already know. A few years
ago, though, as the cost of genotyping plummeted, it became possible to look at markers all
along the genome initially spaced about 1,000 bases apart. But ultimately, it became clear
that there are blocks of genetic information maybe 10,000 or more bases in length that
tend to segregate together from generation to generation. And this reduced the amount
the genotyping necessary. And all of this brought the task of doing
so-called genome
-wide association studies into the realm of affordability. This is a
list of some of the various genes that have been found to be associated with type 2
diabetes. And indeed, for both this condition and many others, the list is getting longer
and longer. This is a diagram maintained by the National Human Genome Research Institute
showing examples of genome-wide associations. This was last updated in December of 2012.
At least in the slide the different colors correspond to different disorders, a
nd then
they are mapped to regions of the chromosome where genetic association through case control
studies has been identified. You can see that the gene map is very densely
populated now, and very large numbers of traits have been attributed to particular associations.
One can use this kind of data now to estimate the odds ratio of disease. So in this case--
and this is the same data set you saw just a few minutes ago, with 30% percent of individuals
having allele 2 and asthma, and only 10% pe
rcent of non-asthmatics having allele 2--
we'll define the odds as the ratio of the probability that an event will happen over
the probability that it will not happen. For example, that an allele carrier gets asthma
or doesn't get asthma. So the odds of an allele carrier having asthma
would be 30 over 40, which is 30 plus 10, compared with the odds of an allele carrier
not having asthma, which is 10 over 40. And then the odds of a non-carrier having asthma
would be 70 over 70 plus 90, which is 1
60, compared with 90 over 160. And from that you
can calculate an odds ratio, which is simply the ratio of 3 over 0.77 in this example,
giving you an odds ratio of 3.86, if one identifies the presence of having allele 2. One needs to be very careful in interpreting
these odds ratios. Imagine that you have a maker that is associated with a 52% increase
in odds of disease. So this would correspond to a 1.52 increase in odds of disease. And
here we show hypothetical numbers of the frequency of dise
ase in individuals who have the trait
in question compared to the population. A 3.5 per 1,000 risk over a population risk
of 2.3 per 1,000 would account for 52% increase in odds. If you say it as 52% increase in
odds, it sounds pretty impressive. If you say that your risk has changed from 2.3 per
1,000 to 3.5 per 1,000, I think for most people that does not sound terribly impressive. And
therefore, one needs to be very careful in how one expresses the data and how one interprets
it. Well, the ho
pe has been that one could use
GWAS data as a basis for predicting risk of common disorders. But one of the problems
has been that a relatively small proportion of estimated heritability is accounted for
by the so far discovered examples of association. So in these five different examples, type
2 diabetes, Crohn's disease, lupus, height, and early myocardial infarction, the blue
sector is the proportion of heritability that is accounted for by GWAS studies done so far.
The green sector is the so
-called missing heritability. So you need to realize that
all of the work done so far is only chipping away at a relatively small proportion of the
heritability that has been estimated from things like familial clustering analyses,
or twin studies, or analysis of variance. What accounts for that? Well, various things
could. There may be non-SNP genetic variants like copy number changes that are not being
included in the studies. It could be that the common variant-common disease hypothesis
does
not apply in all cases, and that some of the genetic risk factors actually are very
rare, and would be hard to detect in the case control studies that have been done so far.
And finally, it's possible that the actual heritability estimates themselves may have
been overestimated. In conclusion, multifactorial inheritance
involves effects of multiple genes interacting with one another and with the environment.
Recurrence risk counseling for most multifactorial traits is based on empirical data. Ge
nome-wide
association studies are revealing genetic factors that contribute to risk of common
disorders, and prediction of risk of common disorders can be challenging since most genetic
markers account for only a small contribution of heritability. Please help us improve the content by providing
us with some feedback.
Comments