[MUSIC PLAYING] [APPLAUSE] ROBERT PLOMIN: So
thank you, [? Jan, ?] and thank you all for
coming on your lunch hour. It's great to talk
to a group like this. I usually talk to academic
groups or people in education, so I'll be curious to see
what you make of all of this. But I've been
informed I'm not here to tell you things that
are relevant to what you do at Google. It's more about general interest
in genetics and behavior. When I started in graduate
school in the early '70s, you might find it
hard to believe, but it was dangerous to
even talk about genetics. Psychology was so dominated
by environmentalism-- the view that you
are what you learn-- that it was dangerous
professionally and sometimes personally even to
mention genetics. The textbooks at the time-- this is the 1970s--
would tell you if you look up schizophrenia,
it would tell you that schizophrenia
is caused by what your mother did to you in the
first three years of life. And since then, a
mountain of evidence has convinced most scientists
that genetics is a major factor in most aspects of behavior. And today, I'd like to just
because I want to leave time for questions,
I'm going to focus on one aspect of behavior,
school achievement, in part because this is a talk
I gave at New Scientist Live on Saturday. So what do you think is
responsible for how well you did at school? So it's easy to think about
tangible environmental factors like your parents, your
teachers, your schools, but what I'd like
to convince you is that something
intangible, DNA, is the major factor responsible
for differences between children in their performance at school. The environment is important
too, but the way it works is very different from the way
we thought it worked before we took genetics into account. Previous studies of the
environment are confounded because things that run in
families could run in families for reasons of
nature-- genetics-- as well as nurture. But if you ignore the nature
and assume everything's nurture, you make some very
wrong conclusions about the environmental
influences on behavior. And this is all a lot more
relevant and important now with the DNA revolution. So I think that'll probably
interest you even more. And I'll try and get through the
first part of my talk quickly in order to spend more
time on the DNA revolution. So this is my book, "Blueprint." I'm very sorry that Penguin
didn't get it here in time, but I bet you it'll come right
after my talk [INAUDIBLE].. But Penguin is usually
very good at these things. The first half of the
book is about genetics. Basically, how do we know
genetics is so important? And what have we learned
about the environment by studying it in
genetically-sensitive designs, and then also some of
the implications of it. And as I say, instead
of psychopathology, like psychiatric
disorders or personality, I'm going to just focus on
one aspect of this huge area, school achievement. Then the second part of
the talk and the book is about the DNA
revolution-- identifying specific DNA
differences that account for differences in behavior. And that's what's going
to change everything. So I think it's very important. It's the reason I
wrote the book now is to say we really have to
discuss some of these things because they're happening now. So just to make sure
we're on the same page, you know, genetics,
heritability-- these are difficult concepts
for people to grasp. But when you ask people,
how heritable is eye color-- that is how much is it due to
inherited DNA differences-- I've done these surveys
and people, on average, would say it's
about 90% heritable. That means it's almost
entirely due to inherited DNA differences. But when you turn
to complex traits like weight, how heritable
do you think weight is? And if you do
population surveys, probably not of groups as
smart as this one, but people would say about 20%, 30%. They think there might be
some genetic influence, but it's mostly environmental. But in fact, it's 70% heritable
across a wide range of studies, including DNA studies. And that's really important
because it means that the differences between
people in this room-- you all seem to be
skinny, except for me-- the differences are substantial,
but the major factor is inherited DNA
differences, which is really important in our
society with fat blaming and, you know, fat shaming, the
sorts of stuff that goes on. We need to recognize
people like me find it a lot easier to put on
weight than you skinny people and a lot harder to lose weight. So when you turn to
school achievement though, people underestimate
genetic influence even more. They think it's about
maybe a little bit of genetic influence,
but not much because achievement means-- the etymology of it is-- by dint of effort. As opposed to ability, which
people say, oh, that's genetic. But in fact, school
achievement-- tested school achievement-- in
English in schools in England is about 60% heritable,
meaning 60% of the differences between kids in their
tested school achievement is due to inherited
DNA differences. And you can see
how important it is to know that because we tend to
assume it's all environmental. It's all due to
how good a school you went to, how much your
parents push you on this, but parents don't have
nearly as much control as they think they do. So that's one of the
messages I want to get to. The most misunderstood
concept in genetics is heritability--
these six syllables. And so it's important
just to spend a minute talking about it. We're only talking about
what makes people different-- why some of us are heavier
than others, why some of us do better at school than others. Of our three billion
base pairs of DNA in the double helix of DNA,
more than 99% of those DNA bases are the same for
everybody, but 1% differs. And the 99% is what makes
us human, but the 1% is what makes us
different genetically. So what we're asking
is to what extent does the 1% of DNA that
differs between us make us different in traits
like school achievement? We're only talking
about describing what is rather than what could be. So we study like twins,
and adoptees, and DNA in a representative
English sample, but that only tells
us about what is now. Say in terms of weight,
given our genetic differences and our environmental
differences now, how much do genetic
differences make a difference? If you change the
environment, if you study a different population
in a different time, that would change. But that's a good
descriptive statistic that is sensitive to the
environment in which you study it. And we're only describing the
normal range of variation. I mean, sure, it's
an issue you all face that you can only talk about-- you can't generalize beyond the
sample that you've described. So these samples
that we study may be representative of
95% of the population, but they don't
include, say, families where parents are abusing the
kids because those people don't participate in studies. Nor does it include the
genetic extremes of single-gene mutations that are very
severe and debilitating, but they're very, very rare-- 1 in 100,000, 1 in 200,000. So if you have a sample of a
few tens of thousands of people, you don't have
those in the sample. And lastly, I want to
emphasize we're talking about genetic influences. These are nudges,
probabilistic propensities. They're not innate,
they're not immutable, and they're not deterministic. So I have to whip through
this, but those are really important points. And you scratch
the surface of what people know about
genetics, and they somehow think you're talking about
innate, immutable influences, which is the case for
single-gene disorders, which is how most of us learn
about genetics from Mendel. There are thousands of
single-gene disorders. As I say, they're very rare. But if you have the
gene on chromosome 4-- the form of a gene
called an allele-- for Huntington's disease, you
will die from Huntington's. It's hardwired, deterministic. It doesn't matter what
your environment is or anything else. The problem is when we talk
about complex traits, not just psychological traits, but
most of the medical burden in society is caused
by common disorders. And they're not caused
by single genes. They're influenced
genetically, but by many, many genes' small effect. That makes them probabilistic
rather than deterministic. And that's a really difficult
jump for people to make. So one of the methods
we used for 100 years to study the extent to which
genetic influence is important is a sort of
biological experiment where there are
two types of twins. I'm sure you all know. 1% of all births are twins. 1/3 of those are identical
twins called monozygotic. It's a single zygote--
that is a fertilized egg-- that for reasons unknown in
the first few days of life divides in two. And those are clones
of one another. They're genetically identical
if you sequence their DNA. So they're 100%
alike genetically. The rest of the twins are
like any brother and sister who happened to be
born at the same time because their mother had two
eggs in the womb at the time and were fertilized. So they share 50%
of their genes. They're 50% similar. So you'd predict that any
traits, say school achievement, that is influenced
genetically, you'd have to predict identical
twins will be more similar than fraternal twins. And you can use the
extent to which that's true to estimate heritability. And so when I came
to England in 1994, I began the Twins Early
Development Study, which is the world's largest
study of development of twins. We started with about
15,000 pairs of twins. And we studied them 14 times
through young adulthood, most recently at 22 years of age. And about 10,000 pairs
continue to participate. Importantly, as
I'll mention later, we also have DNA on them. So they led the way in terms
of some of the DNA analysis I talk about. I was interested
in studying things that hadn't been studied
much before, like language development in the early
child years or behavior problems which developed
very early on and are highly heritable in early childhood,
like attention deficit problems. But when they got
to school because I was interested in
cognitive development, I really wanted to study
kind of the business end of cognitive ability
and that school achievement. So this summarizes 15 years
of work in the Twins Early Development Study. These are the
heritability estimates based on identical and
non-identical twin correlations across all the key stages,
which key stage one at 7 going to 9 to 12. 16 is GCSE. 18 is A Levels. And you can see that
the average heritability exceeds 60% at all ages,
including the very first grade. A lot of people would have
expected that in first grade, your performance is
more a function of what your parents did, but it's
actually just as much genetic as A Levels or GCSE. And in the behavioral sciences,
explaining 5% of the variance is a very big deal. Explaining 60% of
the variance, meaning differences between
people, is off the scale. So it's a huge effect. And oops. My N got misplaced. It's not just the
twin method, there's many other methods
that are used. If the twin method is like
a biological experiment, the adoption method is
like a social experiment. So family members share genes,
as well as environment-- nature and nurture. We've known forever,
everything runs in families. Parents who do
well at school have kids who do well at school. For decades, people
have said, no problem. That's just nurture. Those parents provide a better
environment for their kids. But when you start
thinking about genetics, you say, well, but they
share 50% of their genes. Could it be nature,
not just nurture? The adoption design
separates that by studying parents
who are related genetically to their kids-- these are birth parents
who relinquish their kids for adoption at birth, so they
share nature, but not nurture-- and then adoptive parents
who adopt those kids early in life and share
nurture, but not nature. So the book also
describes a study I've done for 40 years
called the Colorado Adoption Project, which is about 250
families-- adoptive families-- where we had data on the birth
parents, the adoptive parents, and the adopted
kids longitudinally through adulthood from
infancy to adulthood. And this summarizes
20 years of research. We also had matched
non-adoptive families who share genes and
environment with their kids. So general learning
ability, which we call g to avoid the word
intelligence, which is just like a red flag to
a bull for people, but there is this construct of
general cognitive ability which is one of the better
measured traits. So just don't have a
knee-jerk reaction against it. But what's been known for a
long time is it runs in families and that parents
and their children increasingly resemble each
other as the kids grow up. So that shows you the
parent-offspring correlation when parents share genes and
environment with their kids goes up to about 0.3-ish
by late adolescence. So is it nature or nurture? For a very long time, it
was assumed to be nurture. And it's not an
unreasonable hypothesis. But if it's nurture, that
is the parents are giving the kids the environment they
need to develop cognitively, you'd have to predict that
the adoptive parents ought to be just about as similar
to their adopted kids. And the correlations
are actually zero. So that suggests it's
not nurture in the sense that we've thought about
it as systematic effects of the parents, say. Well, then is it really nature? Could it really be that
these birth parents who don't share environment,
but share genes are correlated as much as
parents who rear the kids? And the answer is yes. So it's a powerful demonstration
of the importance of nature and the unimportance
of nurture as we've defined it in terms
of systematic effects of the family environment. The environment is important
because heritability is not 100%. It's more like 50% on average,
but it's a very different sort of environment than
anyone ever thought about from Freud onwards. So the first part of the book-- and I'm whipping
to the conclusions here because I want to get
onto the molecular genetics-- it talks about how do we
know genetics is important and just how important are they. But then some of the
most important findings are about the environment
when you control for genetics. So the second big
finding is this one I just alluded to that the
environmental effects are important, but
they're not nature in the sense of systematic
effects of the environment. Adoptive children-- a
third of adoptive families adopt a second child,
and they're genetically unrelated to each other. They correlate zero. Whereas siblings who grow
up together and share genes, they correlate substantially-- 0.2 for personality and 0.3
for, say, cognitive development. So for 30 years,
people have been trying to find out what are
these mysterious factors that make two kids in a family
different from one another? And it could be lots of things--
accidents, or illnesses, or different peers, or
the parents treating them differently-- but after
30 years of research, no systematic factors
have been found. And so I've come
to the conclusion, called the dark hypothesis,
that the effects are essentially idiosyncratic,
stochastic, random, in a word, chance. So they're unsystematic. It could be like Bill
Clinton in his biography, he talks about why did
he go into politics. And he says, it's because
at 16, he shook JFK's hand. See, that would be a
good example of this. I mean, that's not a
systematic variable you could measure very well. And maybe, you know,
knowing Bill's history here, you can't bet a lot on
its veracity, I suppose. But that's the sort
of thing you get-- these idiosyncratic experiences. Like why are you doing
what you're doing? I know in my case, it's all
these chance sorts of events, just little nudges
in one direction or another that snowball. So that's what we
think it's about. And the other third
finding is what we-- that's what we call
non-shared environment. It's not environment
shared by kids growing up in the same family, going
to the same schools. The third finding is called
the nature of nurture. It's the idea that what
looks like systematic effects of the environment are
actually mediated genetically. So you know, correlations
don't imply causation, right? Everyone knows that. But if you see a correlation,
like I do once a week in the papers, parents do this
and the kids are like that. So parents who read
out loud to kids have kids who read better
when they go to school. It's so hard to resist an
environmental interpretation, but correlations
don't imply causation. And if you start saying,
what about genetics, it'll drive you mad
because, you know, they share 50% of their genes. It could just be
parents who read have kids who read,
but more increasingly, it's sort of the
nature of nurture-- that the correlation goes in the
opposite direction from the way we think it goes. Parents are responding to
differences in their kids, which I really see as a
grandparent with six kids. One grandchild does what
I thought grandchildren are supposed to do. She'd let you read
to her all day long. She loves words and reading. But then the first one I
had, she didn't want to read. She wanted to kick
a ball around. She wanted to be active. It almost would
have been abusive if I said, no,
you're a grandchild. I'm a grandparent. You sit there and I read to you. That's what you're
supposed to do. But it doesn't. You know, we're responding
to differences in the kids and that's as it should be. And also in education,
we should be recognizing that kids differ, try to
minimize the weaknesses, maximize the strengths,
which I'll try to get onto. Oops. So I know I went
through this quickly, but if you look at these
three things together, they lead to the title of
the book, which I agree is provocative and misleading. But what I'm trying
to say is that DNA is the major systematic
force making us who we are as individuals. It's systematic in a sense. I try to emphasize that because
the environment's important, but it's not systematic. And so what I'm saying
to tie these together is if you were cloned and your
clone was reared, obviously, in a different woman
prenatally, grew up in a different family
with different parents, went to a different school,
had different friends, had a different job,
that clone would be very similar to
who you are now, not just in school
achievement, but in personality and psychopathology. In fact, that clone would be
as similar as identical twins reared together. So with school
achievement, they correlate about 0.7, identical
twins reared together. Being reared apart doesn't
make you any less similar. So you can see
that that's talking about the importance of
nature, the unimportance of the systematic
family environment. And it's not just a
hypothetical experiment. I don't know if you
know who this is. Nope? Yeah. You'll know in just a minute. This is Bobby, who
grew up in Long Island in a very wealthy family. He went to university
in Upstate New York. And on the first day,
everyone's calling Bobby, Eddie. Girls are coming
up and giving him a hug and a kiss saying, oh,
Eddie, it's so good to see you. So he's thinking,
psychology experiment, looking for the cameras. But then he met Eddie. And they quickly worked out
they had the same birth date. They were adopted from the same
adoption agency in New York. And the publicity
that came from that led to the rare circumstance
of a third identical triplet because I told you that
the zygote separates for reasons unknown. Sometimes, one of those
zygotes separates again. So these are clones
of one another. They have the same DNA sequence. And the film-- this is
a documentary film that won the Sundance
award last year called "Three Identical Strangers,"
and it's available on streaming. And I really recommend it
as a dramatic illustration of the points I'm
trying to make, but it is just an anecdote
and an illustration. And just watch the first half. The second half is
a very bad story because did you think, why were
they separated and didn't know about each other's existence? Why was one put in
a lower-class home, one in a middle-class home,
one in an upper-class home? A nutty psychiatrist, who was
actually out to prove nurture is important. He thought it was going to
be a definitive experiment. Identical triplets-- put one
in a lower-class family, one in a middle-class family,
one in an upper-class family. They already had adopted
another kid from the agency, so they knew what the
parenting styles were like. So they made them as
different as possible. And he was a Freudian. You know, this was in the '50s. He just assumed the environment,
it's all about nurture. But then when the results
started coming out, he actually buried them. So they're under lock and key at
Yale Medical School till 2066, never been published. But this film exposes it. And there's probably
at least half a dozen, if not more, identical
twins who have been separated as a result of this. He was a psychiatrist
to that adoption agency. Right, so it is
just an anecdote, but behind it is systematic
data on twins reared apart. I did a study in Sweden
of over 100 pairs of identical twins reared
apart, as well as other adoption designs like biological parents
and their adopted-away kids, adoptive siblings, biological
siblings adopted apart, and increasingly DNA. You can use DNA itself
in unrelated individuals to estimate heritability,
but it would take too long to explain that now, but it's
kind of the hot new thing. So they all converge on the
conclusions that I mentioned. So I would have been-- well, I'll just mention
implications of that research. And the main thing to emphasize
is no necessary policy implications. This is an old-fashioned view
that policy depends on values, as well as knowledge. And I'm increasingly
cynical about this-- that better decisions are made
with knowledge than without. More often, I think if your
data, your research agrees with the values of the
government, they'll use it, but it doesn't actually
inform the policy very much. But here's the one thing that
got quite a bit of attention in terms of schools. If you see what I have
been talking about, can you see how
this makes sense? Schools matter, but they
don't make a difference. So schools matter a lot. Kids have to learn basic
skills of literacy, and numeracy, and
enculturation, but they don't make a difference. Kids going to the same school
aren't any more similar than if they had gone
to different schools. And part of that is because
60% of the differences are genetic anyway and the
environmental effects are not systematic. So one quick fact that's
important for you guys because you look like you're
at risk of child-bearing age. And you know how in England,
we have this crazy selective system for secondary schools. Parents spend hundreds
of thousands of pounds to get their kid
into a better school. A better school, why? Mostly based on OFSTED
ratings of the schools. People don't ask about the
effect size of the OFSTED ratings. So they cost about
10 grand each. You know, they're really
good ratings of schools-- atmosphere, teacher support,
bullying, the whole schmear. But then they are the primary
difference between schools in the league tables. So the question is how
much of the variance-- how much of the
differences between kids and their GCSE scores is
accounted for by OFSTED ratings of school quality? So some kids go to schools where
they have very high ratings and some go to schools where
there are very low ratings. So heritability counts for 60%. OFSTED ratings
account for 4%, which is not a noticeable difference. So you get these
mean differences, but you've got to ask
about the effect size? And the effect
size is very small. If you correct for
socioeconomic status because kids aren't randomly
assigned to schools, it goes down to 1%. That's not a
difference you can even detect with your experience. You need very large samples
and statistics to detect it. And then finally, what looks
like systematic effects of schools are often
genetic effects in disguise. So one quick example of
this is that you probably know that there is a big
GCSE difference between kids in selective schools and
non-selective schools. It's a whole grade difference. That's a correlation between
going to a selective school and how well you do on GCSEs. But it's hard to
avoid interpreting that environmentally. The selective schools
have more resources, better playing fields,
probably maybe better teachers, but it isn't. They're selecting for
genetically-influenced traits of school achievement
and ability. And it's a
self-fulfilling prophecy. If you select the kids
who do the best at school and have the greatest
ability, they're going to do the best at school
and have the greatest ability. If you correct-- you
merely just correct GCSE scores for what
the schools select on-- earlier achievement
and ability-- there's no difference
in performance. There's no added value
of selective schools. Now, a lot of parents will say,
but in their cups at least, well, I'm not just
sending them there to get better
school achievement. You know, they get
better contacts. Half of the judges in
the UK are from the 7% of selective schools. So maybe, there is an access
difference or whatever, but if you're really just
thinking about achievement, the evidence is that
it doesn't matter. It doesn't make a difference. Now, it might matter. They might be
nicer places to be. I'm not even convinced of that. There's greater self-harming
at these selective schools. My grandson got
a lot of pressure because my son wanted
to send his kid to one of these selective schools-- nearly 30,000 a year-- when he had a perfectly
good comprehensive. And what was I going to say? Well, just I resisted
that because I don't believe it makes a difference. Oh, yeah. And a kid like him-- Tristram, no less-- is
a star in his school-- the comprehensive
elementary school. But when these kids get to
these high-pressure selective schools, they're
no longer a star. They're lucky to
be kind of average. And there's a lot of
pressure, and they kick kids out if they don't do
well on the pretests and stuff. I mean, it's a crazy,
stressful system that I think destroys learning
and the enjoyment of learning. So I think we need to
think about education in a different way and
thinking about it genetically. Most of what's going on is kids
are selecting, and modifying, and creating
environments in part correlated with their
genetic propensities. So kids in the
same classroom can experience different
environments based on their genetic propensities. They just ask more questions. They follow up on stuff. And I think we need to move
from this passive model of imposed instruction, which is
the-- you know, instruction is from Latin, [LATIN],,
to shove in, which is the way we think of it. We're shoving in this national
curriculum in their head. Instead, we need to move to
an active model of shaped environments in which
kids select environments that are conducive,
that are fitting with their propensities. You know, not everyone--
this might be heresy here-- but not everyone needs to know
advanced math, for example. So we need to think about what
kids are good at and maximize their strengths and
minimize their weakness, and mostly have them learn to
learn because it's well-known-- I mean, everybody says-- I don't if this is
true-- but that the kids in elementary
school of today are going to be doing jobs
that don't exist now. They've got to learn to learn. It's not a matter of specific
skills that they learn. They need to learn to learn. And mostly, they need
to enjoy learning. But our test-obsessed
high-stakes testing culture is really destroying any sort
of enjoyment of learning. OK, I'll get off the soapbox. And I would be happy-- that was 35 years
of my research, and I would have been happy if
it ended there, but then along came the DNA revolution. And as I say, I think it's
going to change everything. It'll allow us to predict from
DNA alone, problems and promise from birth because your DNA
doesn't change throughout life. And that allows us to
move towards prevention. So rather than waiting
until problems occur, like waiting till
kids go to school, and then they fail
at reading, you can predict reading problems. And if you can predict
them, you can prevent them. And prevention has got
to be a better way to go. That's the way all of medicine
is moving towards prevention. Don't wait till people have
heart attacks because we're not very good at fixing things
like that, or obesity, or alcoholism, or schizophrenia. Let's predict who's
got the problems, find out what those
processes are, and intervene to
prevent the problems. And then it really is
transforming science already. Almost all large studies
now are including DNA. And it will transform
society, parenting, as well as education-- I'm writing my next book on
the genetics of parenting-- and then how we
understand ourselves. The end of the book has the
world's first polygenic profile for psychological
traits, and it's for me. And so I describe
what does it mean that I'm at the 94th
percentile for BMI-- Body Mass Index? People say, oh, then you're
just going to give up, say you're a genetic fatty. But I know it's not the case. For all of these problems,
it's motivating to say, OK, I've got to work harder at it. I've got to change
my environment to make it less easy for me
to eat junk food, for example. So it's important to know
about the DNA revolution. And that's what I'd
like to talk about, but I don't have to worry about
getting techy a little bit with you guys, I suppose. The first step is to get DNA. If you've done any
direct-to-consumer testing-- ancestry.com, 23andMe--
you spit in a tube. You can get DNA from
any cell in your body because a remarkable thing-- you may not appreciate
this-- before you start life as a single cell-- half the DNA from your mother,
half from your father-- that unique set of
DNA is the same DNA in every cell in your body-- trillions of cells. So you can get DNA from
any cell in your body. Saliva actually
doesn't have cells, but it does have cells
that are sloughed off from inside your mouth. That's partly what
saliva is doing. So all you need is one cell. If you drink from a cup,
MI5 can get your DNA because you leave a cell or
so on the lip of the cup. So once you get DNA, then
you genotype the DNA. And there's, as I say,
only 1% of our DNA differs. The most common type of DNA
difference is called a SNP-- Single Nucleotide Polymorphism. Polymorphism is
just a difference. And a nucleotide, it
refers to the bases of DNA. So you know the DNA code is
written in a four-letter code of A's, C's, T's, and G's. So what this shows is
your two chromosomes. And if you can see
the little letters, you'll see that
those two chromosomes have exactly the same nucleotide
bases, except for one. And that would be true for us. We're 99% similar, 1% differs. And what we'll do is so
instead of having a C, some people have a
T. Those are called alleles-- alternate
forms of DNA. And once we get
that difference, we can just simply
correlate it with traits, like academic achievement. They call it
association in genetics, but it's just a correlation
between whether you have zero, one, or two
C's, for example, in that. And here's the
first one that was discovered using these new
techniques that I'll describe. It was published in "Science." And it was discovered using this
atheoretical approach that I'll describe in just a
minute, but I just wanted to show you what
a SNP looks like really. This is a SNP where
we-- humans all used to have TT at this
one particular locus spot on the chromosome. Then some guy had a mutation. DNA is incredibly reliable
in its replication, but when you've got three
billion base pairs of DNA, every time your cells divide,
you're duplicating that. Once in a great while,
you get a difference. So this guy had an
A instead of a T. He could have had
a C or a G, but he had an A. It turns
out it rapidly spread through human
populations, especially in Europe where it
originally developed, because it helps
you store body fat. And back in the Stone Age,
that was a very good thing. But you can imagine now
in a fast food nation, it's not such a good thing to
be efficient at storing fat. And so now, 40% of the
population has an A allele. And if you have two
A alleles, as I do, you're six pounds
heavier on average than people who have
no A alleles, and then people with one A
allele is in between. So that's what we mean
by an association. It's literally a correlation
between the zero, one, or two A alleles and the trait--
in this case, body mass index. Now, the thing that's
changed everything is a technological
advance called a SNP chip. So this is a DNA array the
size of a postage stamp that consists of synthesized
short fragments of DNA that surround a particular SNP. And the method is
the same method we've always used
to detect SNPs. You denature-- you raise
the temperature of DNA, and it separates. And DNA doesn't like
to be separated. It wants to find its mate. You chop it up. You put a fluorescent tag on
those little fragments of DNA. And you wash them over this
plate that has these probes-- millions of probes-- for SNPs. So what this cartoon at
the right-hand corner is showing you, if you
have the right SNP, it will hybridize because
DNA wants to hybridize. But if you have the
wrong allele there, it won't be able to hybridize. So what you're left with
then is fluorescence indicating whether or not you
have that particular allele. And if you have a
strong signal, you can see in this array on the
lower left, some of the dots are brighter than others. If it's dark, it means
you didn't have any. If it's medium, it
means you had one. And if it's very bright,
you had two of that allele. And so you can study
millions of these SNPs this way and very cheaply. It started out with
thousands of pounds for this. And now, the real costs
are more like 40 pounds to do whole millions of SNPs
on this one little chip. And it's very reliable, so
it's an amazing technological advance that's really
transformed the life sciences. So now, instead of correlating
one SNP with a trait, you can correlate millions
of SNPs with a trait. And it's atheoretical. You don't just look at
a few candidate genes, like serotonin because you think
it's important in depression, which was a dead end. Instead, you can take an
atheoretical approach just saying, let's look at millions
of SNPs across the genome and see if any of them are
associated with a trait. That's called
genome-wide association. And this is the thing
that's changed my career. In the last year, there was a
genome-wide association study published of
educational attainment with over a million people. And the reason
that's important is that they could detect
very tiny differences. Early on, these studies
weren't successful because they weren't powered
to detect small effects. Everyone thought
heritability is going to be due to a few
genes of big effect, but we never found those. Instead, what you find in this
study and throughout the life sciences is that
above that line is genome-wide
significance corrected for a million multiple testings. And each dot is a SNP. And SNPs close together
on a chromosome are correlated with
each other, therefore they should also be
correlated with the trait. So these peaks suggest the
most significant results. But the point is
thousands of these SNPs are significantly
associated with a trait-- in this case,
educational attainment, which is merely
years of education. And this is what's found
throughout the life sciences-- there are no big effects. That SNP I showed you
for body mass index, people thought, well,
it's just a tiny effect-- 1% of the variance. It turns out it's one
of the bigger effects. Most of the effects
are very much smaller. So that's been a
startling revelation. It means you need huge samples
to detect these effects. But how are you going
to use them then, if there's so many tiny effects? If you want to study
gene-brain behavior pathways, good luck because these
are such tiny effects. It's hard to see them. But I'm interested in
predicting behavior. And you can do that by
adding up these SNPs, just like you add
up items on a scale. And you have to get them
in the right direction. And you weight them
by the effect size of their association,
just because a SNP like these count for more
than some of these SNPs that are less associated,
but it's just simply adding them up. And that's a polygenic score. And this is what's
transforming everything because they're 100% reliable. They're unbiased. They're cheap. But unlike any other
predictor we have, you can predict just
as well from birth as you can from later in life. And most of what we know
about prevention in medicine, as well as in psychiatry,
even think of obesity, it's earlier preventions
that work best, but we can't predict earlier. But we can now with DNA. And it's one of the
few correlations that do an imply causation
in the limited sense that there is no
reverse causation. You inherit your DNA sequence
and nothing changes that. And before you ask me about
epigenetics or gene expression, yeah, these SNPs
have to be expressed, but if they're associated with
a trait like school achievement, they were expressed. We don't have to know anything
about the pathways in between. And nothing in behavior,
environment, or the brain changes your DNA sequence. So this is really happening. People worried about
this 10 years ago, but then direct-to-consumer
companies like 23andMe, ancestry.com came along. And 25 million people have
voted with their checkbook to do this-- to pay
for it themselves. Mostly what you get are
single-gene disorders. And most people
do it for ancestry because it is fascinating. You might think you know
your ancestry, but you don't. We're all mongrels and we
come from diverse parts of the world. And we have a lot of relatives
out there we didn't know about. There's some great stories. I just saw that BBC documentary
last month about diblings-- donor-inseminated siblings-- who
don't find out till they're 18? There's one father who's
inseminated 65 kids, so they're actually
half-siblings who didn't know about each other. So it's a wild west out there. And what they're not doing
yet is these polygenic scores, but that's the big thing now. They're all struggling to do it. A new company came
up this week that's trying to sell polygenic scores. 23andMe won't do it because they
were burned by FDA for reasons we can go into. So they allow you,
if you do 23andMe, with one push of a button, you
download all your genotypes. And these other companies
with one push of a button, you upload them, and
then they give you these polygenic scores. And so the big news is
it's not announced yet, but it's really cool that
the NHS, the government has just given 80 million
pounds to make genotyping free on the NHS. So when you go to the
hospital beginning next year and they take blood,
you'll be asked, do you want to do this and how
much information do you want? And when they've done this
in Finland and Estonia, they're oversubscribed
right away. 85% of the people want to
do it because in a way, it's so much better than
direct-to-consumer companies. In terms of data
confidentiality for example, they're not going to
sell it on to pharma as 23andMe does anonymously. And also if you do 23andMe
and you're in the unlucky 1% of the population who has two
alleles for this recessive trait for Alzheimer's, you could
find out you're at a 60% risk for having Alzheimer's. And what do you get from them? A link saying you might
want to find out more about Alzheimer's, but there's
nothing you can do about it. So if you had it
with the NHS and you have NICE-- the National
Institute for Clinical Excellence-- they could
decide you can't do everything for everybody. There's some things you
can't do anything about. So probably, the
standard procedure will be do you want to just
know the stuff that NHS thinks you ought to know about? Certainly, heart attacks
because you can predict those from early in life and you
can prevent heart attacks. Alcoholism? You know, it's not that
good a polygenic score, but you cannot become alcoholic
unless you drink a lot of alcohol for a long time. If you drink as much
as your friends, they might not be at risk for
alcoholism, but you might. So there are things like
that that you might really want to know about. And this is 160
million are behind this if it works well with the
first five million people. So the only thing
is you have to agree to make your NHS electronic
records available for research anonymized because
the idea is to get bigger and bigger
samples to detect smaller and smaller effects. So here are the big
ones in psychiatry. A lot of work has gone into
severe mental disorders. This is bipolar and
major depression. And with schizophrenia,
you can predict 7% of the liability
towards schizophrenia. Now, it's maybe 50% heritable,
so it's a long way to go, but this is all just
in the last few years once we recognized that the
biggest effects are very small. But the star, and
the thing that's really been amazing
to me, is that we use the educational attainment
genome-wide association study to create polygenic
scores for kids in my TEDS project-- the twins project. So 7,000 kids, we
create their score for educational
attainment, which means years of schooling completed. Well, they're 16, so they
haven't completed school, right? But I was interested in saying,
how much of the variance will we predict in tested
school achievement? And the answer is we
predict 15% of the variance in school achievement, more
than you predict for the target trait of educational attainment
in adults, which is 10%. And this is the
strongest prediction yet in the behavioral sciences,
explaining 15% of the variance. School achievement
is 60% heritable, so we've got a long way to go. I have no doubt in a few
years, we'll be at 30%. It's a technical
problem to get to 60%. We're going to need
whole-genome sequencing because the SNPs
we use are great and we've got millions
of them, but there's a lot of the genome we're
not tagging with them. So that's the next big thing--
whole-genome sequencing. But right now, 15%
of the variance is a lot of variance to explain. You think of OFSTED
ratings of school quality explaining 4% of the variance. This is better
prediction than you can make from income and
SES sort of variables of the family. And I just wanted to show you
this in a bit more detail. If you take this correlation
of a 0.4 for the sample, now we take the sample now
and divide the 7,000 kids into equal groups
of 100 deciles. And you can see that there
is a linear relationship. The y-axis is GCSE scores. The higher the polygenic score,
the higher the GCSE scores. But there's a big
difference at the extremes, even though we're only
explaining 15% of the variance. The average grade at
the bottom decile is a C and the average grade of the
upper decile is an A-minus, but it's only 15%
of the variance. So an important point
to emphasize here is-- I don't if you can
see the yellow dots-- this is a box plot. So it means the kids
in the lowest decile, 75% have a grade of C or lower,
but 25% have a higher grade and some of them have
grades in the A's. It's not a perfect prediction. Conversely, the kids in
the highest group, 75% have an A-minus or greater,
but 25% have a lower grade. But that's what will always be
the case because heritability isn't 100%, so we'll
never explain it all. And of real-world
significance even now is 25% of the kids
in the lowest group go to university and 75%
of those in the top group go to university. So this is a real difference. And we need to think
about how we're going to deal with this in
education and in parenting. So right now, all we've
got is this coarse variable of educational
attainment, but there's a lot of work being done on
more specific polygenic scores for reading, for STEM
subjects, and for ADHD. And I think it might help
with personalized education, where we don't just have a
one-size-fits-all educational system. We recognize that
kids are different, and we try to go with that. And if you can predict
problems as in medicine, I think that will lead
towards more preventative work rather than waiting
until problems occur. And here's something
people don't recognize is siblings
in a family are 50% similar genetically. That means they're
50% different. And people don't really
realize how different that is. So you can have kids
in a family where one kid did very well at
school and the other one is not doing well, but it could
be they have a very different polygenic score. You know, it could be
all over the place. You could have one that's
two standard deviations above the mean and one that's
two standard deviations below. And this is the thing
that always comes up in education is selection. And my view, just
based on values, is we shouldn't have selection. But if you do, I don't
see any logical reason why you could argue against
using polygenic scores to supplement tests because
at least they're not biased and you can't get a better score
if you buy an expensive tutor, for example. So they add something
to the prediction. And I'm particularly
interested in the possibility that they could be useful in
the most socially disadvantaged families, where the kids don't
have the [INAUDIBLE] to deal with academic training. So what I've tried to say is
that inherited DNA differences are the major systematic
force making us who we are, being
responsible for how our children do at school. The environment is important,
but it's not systematic. It's these random chance
variables, not systematic effects of families or schools. And that polygenic scores
will transform science. It's already happening. All big studies
now are including DNA to add a genetic component
to what they're doing-- society, parenting,
schools, and also how we understand ourselves. Thank you. [APPLAUSE] SPEAKER 1: Thank you very much. We do have about 10
minutes for questions. Please wait for a microphone. AUDIENCE: Hey. Thanks for the talk. So on the genome-wide
association studies, you mentioned that
these are correlational, but that generally
implies causation. And it's true there's no
reverse causation here, but don't things like
assortative mating increase those, bias those
estimates upward? And I guess the
extreme case would be you could imagine a
genome-wide association study that would test for
a country of birth, right? Like, and is it
possible that genetics would have impact [INAUDIBLE]? ROBERT PLOMIN: Yeah. Well, this is like condensed
from three hours of talk and it's discussed in the book. That's a very good point that
as I said at the beginning, we only describe
particular populations at particular times. And the populations that
are in these big genome-wide association studies are almost
entirely Northern European, American, Australian. They're Caucasian
populations, so these scores describe that
population pretty well. So it describes
people in England, but it doesn't describe other
ancestry groups for example. And the way you
deal with this is you take out principal
components from these analyses. So this is all the life
sciences doing this stuff and it's some of
the brightest people around, so that's well-covered. The issue of assortative
mating is a tricky one because if you're interested
in predicting how well kids do at school, part
of their genetics is whether or not there's
assortative mating. You know, like begets like. Opposites attract. Uh-uh. It's only like begets like. It's assortative-- positive
assortative mating. So if you don't believe in
intelligence, be single, go to a singles bar. And in a few minutes, the
first thing you're picking up is how bright someone is,
especially verbal intelligence. So for personality,
the correlation is 0.1 between spouses. For non-verbal
ability, it's 0.4. And for verbal
ability, it's 0.6. I mean, you don't really know
somebody's spatial ability very well when you talk to them,
but you very quickly pick up on whether they're
going to be worth talking to in the morning. So that's assortative mating,
and it's part of the genetics. It's getting at mechanisms. If you want to
understand the mechanisms by which this occurs, great. But if you're interested
in prediction, it's OK-- assortative mating. And it really is only
for cognitive abilities that you get this
assortative mating. But it's a great
question and there's tons of stuff about this. And so it'd be nice to be able
to go on about it some more, but I think we'll have to
leave it there for this. AUDIENCE: Thanks. ROBERT PLOMIN: Thank you. AUDIENCE: Thanks for the talk. So in the study about
the polygenic scores and essentially
the effect sizes, if you have a million people
and you have a lot of SNPs, how can you, like, detect
the effects and interactions between different
small effect sizes? ROBERT PLOMIN: Great question. I'll repeat it because that
mic is only for the recording. It was a question about if
you have these small effects, how do you correct for
looking at millions of SNPs? And the second part was? AUDIENCE: Essentially, if
you consider interactions between small effects, right? ROBERT PLOMIN:
Interactive effects. So again, a lot
to say about that. That's a very
pertinent question. But as I showed you, we're
correcting for a million tests. So that line of significance
corrects for a million tests. Even though you have
more than a million SNPs, those are linkage groups. So it's thought that
correcting for a million test works, and it does
because what you do is you create these
polygenic scores from the genome-wide
association data in one study, but then you apply them
in independent samples. And what people realized right
away is if you take 100 SNPs, it doesn't predict
nearly as well as 1,000, doesn't predict nearly as
well in independent samples as 10,000. But as you rightly
are intuiting, these are additive
effects of each SNP. And most people think genetic
effects must interact. Well, they do at a
mechanistic level, but fortunately for
us, they don't when it comes to adding up
the effects of genes. You get most of the
effect of heritability from adding up these genes. And in a way, that's the way
selection works in evolution. It works on additive
genetic variance. But you know, it's
hard to believe, but quantitative genetic
studies, twin and adoption studies, and animal
selection studies are all consistent
with the notion that genetic effects
are largely additive. Because the problems of power
would be so much greater if you have to deal with not
just interactions between two genes, but what about
10 genes or 100 genes? AUDIENCE: Thank
you for the talk. I'm a physicist
by training, so I have to ask somewhat
sort of trade, some of the more
fundamental perspective of, if I understand
correctly, DNA determines what proteins
are produced by the body. Which means that if we
inject those proteins which make people in quotes, smarter,
then that study potentially, this study could be potentially
interpreted like today you can do a test even in
the UK before the child is born for severe
genetic disorders and potentially terminate
a pregnancy early. And one can imagine
that some people won't want kids which have
predisposed to be less performant at school. But at the same
time, these studies could be interpreted
as a way to develop medicines, which make people's
cognitive abilities higher. It does make sense. ROBERT PLOMIN: Yeah, so
you started off by saying, you're talking about single
gene, simple gene protein sort of relationships. And it's all a lot
more complicated. All these genes do many,
many different things. And each thing,
even in the brain, is influenced by
many, many genes. So it's going to be hard to take
like a gene editing approach for these complex disorders and
traits influenced by thousands of DNA differences. Where people are really
thinking about it though, are in terms of
single gene disorders. That's where gene
editing comes in. But you have to, there was this
crazy BBC documentary on beauty and how we could do gene
editing for beautiful. But you see, you've
got trillions of cells. How are you going to change
all of your cells as an adult? All you can do with gene
editing is get in there at the first stages. But you'd have to
change every cell. But people are trying to do
that for single gene disorders with gene editing, this CRISPR
technology, which is amazing. But I don't think
it's going to work when you're dealing with
these thousands of tiny DNA differences. But another good question. SPEAKER 1: For
one more question. No one has one? We can finish here? So, thanks a lot again. ROBERT PLOMIN: My pleasure. Thank you all. [APPLAUSE]
