0:15

Pushkin. I'm

0:27

Jacob Goldstein and this is What's Your Problem, the

0:29

show where I talk to people who are trying

0:31

to make technological progress. My

0:33

guest today is Manola's. Kellis

0:36

Manola's is a professor of computer

0:38

science at MIT, and he works

0:40

in computational biology. It's

0:43

a field where researchers take giant

0:45

data sets relating to things like genetics

0:48

and health outcomes and try and understand

0:50

basically what's going on, things

0:53

like what are the cellular mechanisms

0:55

of disease and how can we intervene to

0:57

keep people healthy. In particular, Minola's

1:00

research focuses on genomics and

1:02

a related field called epigenomics.

1:05

Here's how Manola's explains.

1:06

What that means. What's

1:10

extraordinary with genomics is

1:12

that we can see beyond the

1:15

limits of human imagination. We're

1:18

talking about millions of cells across

1:20

hundreds of people, across thousands of genes,

1:23

and now we can now look at how

1:25

the single genome manifests

1:28

in every cell type of the human body

1:31

in a slightly different way to create

1:33

this extraordinary symphony

1:36

that is the human life, that is human

1:38

thought, that is human understanding, cognition, and

1:41

every biological process that ability

1:44

to now start understanding the building

1:47

blocks of how this human

1:49

genome manifests

1:51

into all of these myriad of cell types

1:54

and their interactions and their combinations

1:57

and their coordination and their communication

2:01

is what we can do for the first time. They're also

2:03

giving us the entry points for

2:05

understanding the basis

2:08

of human variation, the basis of human

2:10

disease, and the basis for reversing

2:13

human disease.

2:14

So that is the very big picture

2:16

view of what Manola's does. In

2:18

our conversation, we got into a lot more

2:20

detail. For one thing, Manola's

2:22

talked about his work on obesity, and

2:25

that work is based on epigenomics,

2:27

which is basically the way in which

2:30

different genes are turned on and

2:32

off, and this turns out to be a really

2:34

big deal. Manola's and I also

2:36

talked about his work on Alzheimer's

2:38

disease. In that part of the conversation,

2:41

he talked about how he and his colleagues are trying

2:43

to find these key biological

2:45

pathways that contribute to lots

2:48

of different diseases, and how they're trying

2:50

to come up with drugs to target those

2:52

pathways. We started our

2:54

conversation by talking about Manola's

2:56

early work on the human genome, which

2:59

led to the work he's doing now.

3:00

So the human genome was mapped by K ninety

3:02

nine or two thousand and three, depending on how you count.

3:05

And then we had all of the nucleotides,

3:07

all of the letters through into billion letters.

3:11

Then the hard part begins, how

3:14

do you make sense of that book? So that was the Book

3:16

of Life. So we had all of the letters, how do you make

3:18

sense of the book? My own

3:20

PhD was developing evolutionary

3:23

signatures for understanding systematically

3:26

the human genome. So how do you recognize

3:30

where are the protein coding parts? What are

3:32

the parts that code for protein? We didn't even

3:34

know.

3:34

And just to be clear, sort

3:36

of non intuitively, most

3:39

of the human genome is

3:41

not protein coding, right, Like there's

3:43

this very basic idea that

3:45

like, oh, sure the genome, that's what codes for proteins,

3:47

but in fact, most of the genome is not doing

3:50

that.

3:51

Ninety eight percent of the

3:53

human genome does not code for protein.

3:55

It's wild. That is so nonintuitive,

3:58

correct.

3:59

So in that mysterious

4:01

ninety eight percent of the genome lie

4:04

control regents that are

4:06

responsible for turning genes on and off. And

4:10

that's where ninety

4:12

three percent of the disease associated

4:15

genetic variants are sitting.

4:17

Huh, it's not the genes

4:19

that actually code for proteins, it's the genes

4:21

that control when are proteins

4:24

made, when are they not made, how much are they made.

4:26

That's exactly right.

4:26

Okay, so I get that in a broad

4:29

sense. That's

4:31

sort of the state of affairs when you're coming

4:33

into the.

4:34

Field's exactly right. So I wrote a series

4:36

of papers, both as a student and

4:38

as a faculty member that sought to then

4:41

uncover how to even parse

4:43

the genome, how to even start understanding reading

4:45

that book of life. So that's one part.

4:48

The second part is where the regulatory

4:50

motifs are. What are regulatory motifs.

4:52

They are the short words of the

4:54

language of DNA that are

4:57

bound by regulators

5:00

to turn genes on and off. So there's these

5:02

regulatory regions, and within these

5:04

regions lie these words

5:07

which are the regulatory mode.

5:09

And just to be clear, the regulatory

5:12

motifs are part of what

5:14

determine sort of when and how

5:16

much different genes express different proteins.

5:19

That's exactly right, that's exactly right. And that's

5:21

where the human epigenome comes in. So

5:23

what we needed to now understand is how

5:25

that genome turns to life. So

5:28

you can think of the epigenome as the living genome,

5:30

as the genome. There's the genome itself

5:32

is static. It's just the book the tablets, if

5:35

you wish that Moses brought down from the mountain,

5:37

and then the epigenome is the

5:40

music that gets played from the orchestra. The

5:42

epigenome tells you which

5:44

parts are active in the brain and the

5:46

liver, and the heart and the muscle and so and so forth.

5:49

So your work on the epigenome is really

5:52

interesting to me. And I know you've done

5:54

some work on obesity, and the

5:56

epigenome tell me

5:58

about that.

5:59

The strongest genetic association with obesity

6:01

sits in one gene called

6:04

FTO, and FTO

6:06

was renamed fat and obesity

6:09

associated after that discovery, and

6:11

it remained mysterious for seven years.

6:14

People had no idea how that gene

6:16

works.

6:16

You just saw correlate.

6:17

There was a correlation.

6:18

There was a correlation.

6:19

Just the problem of genetics and the beauty

6:21

of genetics. The beauty of genetics is that it tells

6:23

you what region is responsible

6:26

for disease. Regardless of how it functions.

6:29

The downside is that it after

6:31

he tells.

6:31

You it's the same thing.

6:33

After it tells you that he has a role, you

6:36

have no idea how it functions. And

6:39

what we showed in

6:41

our work is that

6:43

that region doesn't affect

6:45

the FTO gene at all.

6:47

So like in the middle of a gene,

6:50

there is this whatever series

6:52

of nucleotides,

6:54

but those those nucleotides are just randomly

6:57

in the middle of that gene and actually have nothing to do

6:59

with that gene. I didn't even know you could do that.

7:01

Fairly, you can't. So there

7:04

are eighty nine differences, eighty

7:07

nine common variants, common

7:10

genetic variants that are all coinherited. If

7:12

you get a here, you get all

7:14

of the other you know, actage,

7:16

you get that passage. If

7:19

you get that package, it spans fifty thousand

7:21

letters. But there are only eighty nine differences

7:23

in these fifty thousand letters.

7:25

Wow, and these will increase

7:28

your body weight

7:30

by one standard deviation, which

7:33

is like how much it's like nine pounds, Like it's a

7:35

lot, okay. So so basically what

7:38

that does is that it functions

7:41

somehow to increase your risk for a basits, it's

7:43

like the strongest genetic association before.

7:46

And what we reason

7:49

is, how could it be acting. It could be acting in your

7:51

brain to decide whether you like sweets or

7:53

salting. It could be acting your muscle to make

7:55

you more fit or less fit. It could be asking

7:57

in your digestives. So we

7:59

basically said, okay, well that's speculation.

8:02

Let's look at the data. And we looked at the data

8:04

and we found that there was this massive

8:06

control region that was active

8:09

in mesenchymal stem

8:11

cells what are mesimo cells and sells. They

8:13

are the progenitors of brown

8:16

fat and white fat.

8:20

Now, white fat is white because

8:22

it's full of lipids. That's

8:24

where all the calories are stored. Brown

8:27

fat is brown because of all of the iron

8:29

in the mitochondria. That's where the calories

8:31

are burned. So it turns out that

8:33

our fat cells make a developmental

8:36

decision in their first three days of differentiation

8:39

to go down the white path lineage or

8:41

the brown path lineage. And

8:44

what the white fat does is it stores

8:46

energies and brown

8:49

burns energies. So

8:51

it turns out that I'm actually homozygous risk for

8:54

the store calories position, which

8:57

is the obesity risk.

8:58

So you have the obesity.

9:00

I have two copies of the obesity risk. My

9:02

wife has zero. I can tell you,

9:05

you know, we look very different. Fair So

9:10

we basically realize that it sits

9:12

in the progenitor cells of white and

9:14

brown flat and then

9:16

we could show that

9:18

the true target was not the ftogene

9:20

at all. It was instead two

9:22

other genes that are sitting one point

9:25

two million letters away from

9:28

this region and six hundred thousand

9:30

letters away, and those genes turned

9:32

out to be master controllers

9:35

of thermogenesis. They

9:37

are basically switching your

9:40

metabolic state. So my

9:43

cells are stuck on the store

9:45

position and

9:47

my wife cells are stuck on the burn position.

9:50

And so what is the relationship between

9:52

the genes that are acting

9:55

here and this

9:57

this you know, package variant that is

9:59

far away from them.

10:00

It comes back to the epigena. So

10:03

our DNA is stored

10:06

inside a tiny little space. The

10:09

way that gene regulation works is that you have

10:12

these control regions that are scattered

10:14

throughout the region of

10:16

every gene that are linked together

10:18

to that gene in three dimensions. So

10:20

they do around and.

10:22

So it's it's far away. If you think of it

10:24

as a strand but in three dimensional space,

10:27

right there, three dimension pats right, Ah,

10:30

that's satisfying.

10:32

And when we took these genes and

10:34

we modulated them, we show

10:37

that you can turn off one

10:39

gene in mouse, in

10:42

specifically the adipocytes of mouse

10:45

with a dominant negative cus of

10:47

fat cells with a dominant negative

10:50

construct, and that turned

10:52

the mouse fifty percent leaner. They

10:55

eat the same amount, they exercise

10:57

the same amount, but they burn calories

11:00

when they're awake and they burn calories

11:03

when they're sleeping. And

11:05

what's really fascinated with that story

11:07

is that the variant associated

11:09

with obesity is at two percent

11:12

frequency in Africa, but forty

11:14

two percent frequency in Europe and

11:17

forty four percent frequency in Southeast Asia.

11:20

So it rose from two percent to forty

11:23

four percent maybe

11:25

because of positive selection. Maybe

11:28

it was beneficial to be able to

11:30

store every kind of.

11:31

Places where food is, where you have food

11:33

is scarce in moments of famine, exactly.

11:36

In the out of Africa event, this may have

11:38

been selected for. Or in the you know,

11:40

ice ages, it may have been selected for. And

11:42

it's only after World War two that

11:45

this variant became associated

11:47

with obesity.

11:48

Because food became so abundant.

11:50

And we stopped exercising as much. So

11:53

it's fascinating to see how the environmental

11:55

shift led to a new genetic

11:58

association which is now plaguing

12:00

our society, and of course

12:02

the hope that by understanding the circuit

12:05

systematically, we can now

12:08

solve so many different

12:11

circuits and ultimately so many

12:13

different pathways and ultimately so

12:15

many different disorders.

12:19

In a minute, Manola's describes how

12:21

he and his colleagues are trying to turn

12:23

their genomic research into new

12:25

medicines. That's

12:35

the end of the ads.

12:36

Now we're going back to the show.

12:39

Another area where Manola's and his colleagues have

12:41

done a lot of work is on Alzheimer's

12:43

disease. They looked at a common

12:45

genetic variant called apo E four.

12:48

People with two copies of this variant have a

12:50

much much higher risk of getting Alzheimer's,

12:53

and Manola's and his colleagues were trying to figure out

12:55

why. They found that having

12:57

this Apoe four variant was linked to

13:00

problems with moving cholesterol

13:02

around in the brain, a process

13:05

called cholesterol transport.

13:07

Then they did experiments and mice that

13:09

found that drugs that restore cholesterol

13:12

transport actually restored

13:14

cognition in the mice. Now

13:17

that's in mice, and Alzheimer's

13:19

is a notoriously difficult disease

13:21

to treat in humans. So I

13:23

asked Minolas what it will take to

13:26

move his research from mice to humans,

13:28

and his answer was really interesting. It

13:30

pointed not only two ideas about treating

13:33

Alzheimer's, but to bigger ideas

13:35

about treating human disease more generally.

13:38

The way that I'm thinking about this, the way that our team

13:40

is thinking about these, is how

13:42

do we enable personalized medicine

13:45

and precision medicine. Namely,

13:48

Alzheimer's is not going to be only about transport.

13:51

It's going to be a combination. Every person

13:53

has some combination of these regulations.

13:56

A point four is the strongest genetic

13:59

risk, but there are many others.

14:01

And the question is how do we now

14:04

systematically take a person with Alzheimer's,

14:07

or take a family with risk, develop

14:09

treatments that are either directly addressing

14:12

the root causes rather than

14:14

treating the symptoms, and

14:16

are not only preventative but

14:20

adapted to every family and every

14:22

person.

14:22

And just to be clear, like having you know,

14:24

two copies of the APO four lil

14:28

is neither necessary nor sufficient to get Alzheimer's.

14:31

Right, that's exactly both of them and not get it. You

14:33

can have neither of them and get it. So it's exactly

14:36

so complicated hard.

14:37

So, as with everything with human disease, genetics

14:40

is not destiny. Genetics is

14:42

a predisposition, and there are environmental

14:45

factors. There are behavioral factors, there

14:47

are nutritional exercise factors,

14:49

there are socio economic factors. There's so many

14:52

other factors that are affecting how your

14:54

genetics will manifest ultimately

14:56

into disease. But

14:59

now the question is for every person,

15:01

how do we create a drug?

15:04

And it's not going to be feasible

15:06

economically or in any other

15:08

way to create one

15:10

pill for each person. The way that

15:12

we're going to enable personalized medicine is by

15:15

understanding what are the hallmarks of

15:17

disease, what are the hallmarks of Alzheimer's,

15:19

the wholemarks of obesity, the whole moods of diabetes,

15:22

the hallmarks of cardiac disorders, and

15:25

develop therapeutics for every one

15:27

of those hallmarks. So think of it as an

15:29

arsenal of twelve or twenty

15:32

different drugs for Alzheimer's that

15:34

you're going to be taking a combination of it.

15:37

Seems like oncology is already

15:39

some way down that road, right, I

15:41

mean, you know her two positive

15:43

breast cancers have certain drugs that target

15:46

them that sort of thing, right, is that the model?

15:48

That's exactly the model. So the hallmarks

15:50

of cancer have been the way

15:53

of thinking about cancer for twenty plus years

15:55

now. And the difference

15:57

in cancer is the following. Cancer

16:00

is subject to positive

16:02

selection. What does that mean? That

16:04

means that because it's a replicative

16:07

disorder where the cell, the

16:09

cancer cells make more of themselves.

16:12

If a cell acquires a mutation

16:14

that allows it to replicate faster, you

16:17

will have more of that cell. So

16:20

you are subject to positive selection

16:22

where the bad mutations are

16:25

increasing in frequency

16:27

in every generation of the cancer. By

16:30

contrast, most complex

16:33

disorders are subject to purifying

16:35

selection, where the mutations that

16:37

are responsible for them are maintained

16:39

at low frequency by evolution.

16:42

Huh.

16:43

So it's working at the opposite

16:45

ends of the evolutionary spectrum. So

16:48

cancer has a small number of genes

16:50

that drive the disorder. Complex

16:53

traits have thousands of genes that are

16:55

maintained at low frequency or

16:58

at weak effects.

16:59

Except that sounds much

17:01

harder. It's harder to figure out what's going on harder.

17:05

But the saving grace is that even

17:07

though you have extreme heterogeneity

17:10

in the number of drivers, for every one of these

17:12

disorders, they coalesce,

17:16

they cluster, they converge

17:19

in a small number of recurrent

17:21

pathways, and these are the hallmarks.

17:24

Huh.

17:25

So you can find multiple genes associated

17:27

with lipid transport, you can find multiple

17:30

genes associated with new inflammation with DNA

17:32

damage, so.

17:33

You target the sort of pathways where they

17:35

converge.

17:35

That's exactly right. So we're not going to make a drug

17:38

for Alzheimer's that we might make a drug

17:40

for DNA damage, a drug for

17:43

lipid metabolism, a drug for cholesterol

17:45

transport, et cetera. And that's

17:47

what we're working.

17:48

That's satisfying. That's a satisfying

17:50

explanation.

17:51

It basically says that it is a

17:53

limited number. There's a billion people

17:55

in the planet. We're not going to have a billion drugs.

17:58

What we're going to have it's a small number of drugs,

18:01

one for each pathway, and these

18:03

are sometimes going to be actually reused

18:05

between different disorders. So we

18:08

work on cardie disorders, we're finding

18:10

the same genes underlying

18:12

Alzheimer's, and specifically

18:14

the lipid and cholesterol component are

18:16

in fact reused in the heart disease.

18:19

And again it's about lipids. It's about

18:22

saturation of the fat

18:24

stores of an individual and now the

18:26

lipid escaping into the blacks into the

18:28

bloodstream, forming these plaques

18:31

that will then cause heart

18:34

you know, failure and heart damage and so and so forth.

18:36

So that's where we're at.

18:39

So is there. I

18:41

mean, the dream is that there is some dysfunction

18:44

that is common to all these different diseases

18:47

that you could target, right, Like, I

18:50

mean, the naive dream is find

18:52

the cure for everything, or not everything, but find the

18:54

cure for a lot of things, or at least find

18:57

a drug that will reduce risks of

18:59

many different bad things, right, I

19:01

mean, is that plausible or am I just

19:03

naive in going there? From what you're saying.

19:06

So you're right

19:08

that some of the time these

19:11

pathways that we're finding are going to be helping

19:13

in multiple frauds, And

19:15

then that's absolutely the dream. We should basically

19:18

start not with what is the worst disease, but

19:20

maybe what is the best pathway that if

19:22

we fix that one, we're going to have an impact on

19:24

most diseases.

19:25

Right, like the highest return on investments

19:28

for example.

19:28

Like, Yeah, that's a great way to think about it. But

19:32

the way that I would say is that for

19:35

each person, this might be a different

19:37

molecule.

19:39

So now I'm not hopeful.

19:43

But that with a small number

19:45

of these molecules, say one hundred, one hundred

19:47

and fifty two hundred molecules.

19:48

When you say molecule, you mean drug.

19:50

I mean trust, might I mean drust. Yeah, Basically

19:52

that there's going to be a small number of pathways and

19:55

a small number of these modulators,

19:58

and that those are going to be mixed and

20:00

matched in each person to then

20:02

target a communatorially large number

20:04

of people.

20:05

Yeah, it just got hard. I know, I know biology

20:07

is hard, but I got up to for a

20:09

second.

20:10

There's not going to be a single silver bullet for all

20:13

of those. In fact, for any one of these diseases,

20:15

there's no silver bullet. But the moment you

20:17

build your panelbly of fifty silver

20:19

bullets, then you're going to be hitting two hundred

20:21

diseases. That's the beauty of it.

20:24

Fifty bronze bo there's no silver bullet,

20:26

but maybe.

20:26

You can find it for hearts exactly right.

20:30

We'll be back in a minute with the lightning round.

20:43

Now, let's get back to the show. I

20:45

read that you have been an author on more than

20:48

two hundred and thirty papers, which

20:50

is a lot. Which one was the most fun?

20:52

Oh? You know what, don't I tell you about my very first one?

20:54

Sure?

20:56

And the very first paper was published

20:59

in c graph and it now has like two thousand

21:01

citations, And it was about how do we reconstruct

21:04

the surface of an object

21:06

from a cloud of points? So

21:09

you can basically use laser scanning to sort of figure

21:11

out points in three D and then

21:13

the question is what is the surface that goes between

21:15

them. I've always been fascinated with three D

21:17

space, so it was very fun for me to

21:19

just like you know, as a kid, basically as

21:22

as a freshman at to

21:24

work on such a project and then showing

21:26

up at the conference. He was in Disneyland,

21:29

so it was my first time in Disneyland as an author of

21:31

a vapor.

21:31

Sounds relevant for motion capture, not

21:34

knowing anything about it. When I think of, like, you

21:36

know, people, the

21:38

way they make movies now exactly as they put a bunch

21:40

of censors on people and they move around and

21:42

then you can turn them into a dragon or whatever

21:44

you want.

21:45

Yeah, that's exactly right. So you

21:47

know that paper has been quite influential

21:49

and used for a lot of a lot of different things.

21:52

What's the most overrated Greek island?

21:54

Oh my god, I can tell you about the most underrated

21:56

Santorini. Definitely not overrated tons

21:58

of people, but worth every time. I

22:00

can tell you about my first day in Santorini,

22:03

which is I walked out on this balcony

22:05

and I asked the owner of the restaurant if I can

22:07

take a look at the view and I'm not order anything.

22:10

He said, please be my guest,

22:12

and I walked out, and ten minutes later, I'm like, I can't

22:14

leave. I'm gonna have to order. He

22:17

tells me, ten years ago, I came here to look at

22:19

the view.

22:19

I want you to throw a little bit of shade. I

22:21

want you to get in a little bit of drug.

22:23

Can't.

22:24

What's one place in Greece I should not cannot.

22:28

It's not possible. I

22:31

mean, you know, if you keep insisting, I'll give you another

22:33

twenty amazing places to visit.

22:35

Well, that's fair, that's fair. I did

22:37

what I could do. If everything goes

22:39

well, what problem will you be trying to solve

22:41

in five years?

22:43

I think what I'm trying to solve now of

22:46

actually creating

22:48

these drugs in such a modular, AI

22:51

driven, personalized, reusable

22:54

way, centered on pathways.

22:57

That's going to keep me busy for a long time. And

23:00

I hope that in five years we

23:02

have actually sold a

23:05

big chunk of the platform and

23:07

that we have a few drugs

23:09

in clinical trials. So you know, my dream needs

23:12

to take all of these circuits that we have uncovered

23:15

and make a difference for humanity, make

23:17

a difference for you know, my fellow beings.

23:19

That's my big goal.

23:21

Great, it's fun to talk to you.

23:23

I learned a lot, such a pleasure, thank you, and

23:26

I love that you're fearless. You're like, well,

23:29

we're gonna jump into this new topic and find

23:31

it all about it.

23:36

Man nola's Kellis is a professor of computer

23:38

science at MIT. Today's

23:41

show was produced by Edith Russelo, edited

23:43

by Karen Chakerji, and engineered

23:46

by Sarah Bruguer. You can email

23:48

us at problem at pushkin dot FM.

23:51

I'm Jacob Goldstein. One last thing

23:53

we are going to be taking a break for a couple

23:55

of weeks, but we'll be back with new shows

23:57

in early twenty twenty four. Thanks

24:00

for listening, Happy New Year, that

24:10

t