Manolis Kellis: Biology of Disease | Lex Fridman Podcast #133

00:00:00.000 | The following is a conversation with Manolis Kellis,

00:00:03.040 | his third time on the podcast.

00:00:05.360 | He's a professor at MIT

00:00:07.480 | and head of the MIT Computational Biology Group.

00:00:11.200 | This time we went deep on the science, biology and genetics.

00:00:16.200 | So this is a bit of an experiment.

00:00:19.800 | Manolis went back and forth between the basics of biology

00:00:23.360 | to the latest state of the art in the research.

00:00:26.320 | He's a master at this.

00:00:28.120 | So I just sat back and enjoyed the ride.

00:00:31.080 | This conversation happened at 7 a.m.

00:00:33.440 | So it's yet another podcast episode

00:00:36.400 | after an all-nighter for me.

00:00:38.160 | And once again, since the universe has a sense of humor,

00:00:42.000 | this one was a tough one for my brain to keep up,

00:00:45.400 | but I did my best and I never shy away from good challenge.

00:00:50.240 | Quick mention of each sponsor,

00:00:51.680 | followed by some thoughts related to the episode.

00:00:55.600 | First is SEMrush, the most advanced SEO optimization tool

00:01:00.320 | I've ever come across.

00:01:02.200 | I don't like looking at numbers,

00:01:04.240 | but someone probably should.

00:01:06.040 | It helps you make good decisions.

00:01:08.360 | Second is Pessimist Archive.

00:01:10.740 | They're back.

00:01:11.840 | One of my favorite history podcasts

00:01:13.760 | on why people resist new things,

00:01:16.100 | from recorded music to umbrellas, to cars,

00:01:18.600 | chess, coffee, and the elevator.

00:01:22.120 | Third is 8sleep, a mattress that cools itself,

00:01:25.200 | measures heart rate variability, has an app,

00:01:28.160 | and has given me yet another reason

00:01:30.160 | to look forward to sleep,

00:01:31.400 | including the all-important power nap.

00:01:34.760 | And finally, BetterHelp, online therapy

00:01:37.800 | when you want to face your demons

00:01:39.440 | with a licensed professional,

00:01:41.280 | not just by doing the David Goggins-like physical challenges

00:01:45.000 | like I seem to do on occasion.

00:01:47.900 | Please check out these sponsors in the description

00:01:50.160 | to get a discount and to support this podcast.

00:01:54.120 | As a side note, let me say that biology in the brain

00:01:58.080 | and in the various systems of the body fill me with awe

00:02:01.300 | every time I think about how such a chaotic mess

00:02:04.900 | coming from its humble origins in the ocean

00:02:07.400 | was able to achieve such incredibly complex

00:02:10.640 | and robust mechanisms of life that survived

00:02:14.200 | despite all the forces of nature that want to destroy it.

00:02:18.440 | It is so unlike the computing systems

00:02:20.200 | we humans have engineered that it makes me feel

00:02:22.680 | that in order to create artificial general intelligence

00:02:26.120 | and artificial consciousness,

00:02:27.920 | we may have to completely rethink

00:02:30.440 | how we engineer computational systems.

00:02:33.480 | If you enjoy this thing, subscribe on YouTube,

00:02:36.040 | review it with 5 Stars on Apple Podcasts,

00:02:38.280 | follow on Spotify, support on Patreon,

00:02:41.000 | or connect with me on Twitter @LexFriedman.

00:02:44.800 | And now, here's my conversation with Manolis Callas.

00:02:48.700 | So your group at MIT is trying to understand

00:02:52.120 | the molecular basis of human disease.

00:02:54.760 | What are some of the biggest challenges in your view?

00:02:57.360 | - Don't get me started.

00:02:59.320 | I mean, understanding human disease

00:03:01.440 | is the most complex challenge in modern science.

00:03:06.100 | So because human disease is as complex as the human genome,

00:03:11.100 | it is as complex as the human brain,

00:03:14.200 | and it is in many ways even more complex

00:03:18.720 | because the more we understand disease complexity,

00:03:22.240 | the more we start understanding genome complexity

00:03:24.920 | and epigenome complexity and brain circuitry complexity

00:03:28.920 | and immune system complexity and cancer complexity

00:03:31.160 | and so on and so forth.

00:03:32.180 | So traditionally, human disease

00:03:37.000 | was following basic biology.

00:03:39.320 | You would basically understand basic biology

00:03:41.040 | and model organisms like mouse and fly and yeast.

00:03:45.920 | You would understand sort of mammalian biology

00:03:48.960 | and animal biology and eukaryotic biology

00:03:52.280 | in sort of progressive layers of complexity,

00:03:56.480 | getting closer to human phylogenetically.

00:03:59.640 | And you would do perturbation experiments in those species

00:04:03.600 | to see if I knock out a gene, what happens?

00:04:07.800 | And based on the knocking out of these genes,

00:04:10.280 | you would basically then have a way to drive human biology

00:04:14.440 | because you would sort of understand

00:04:15.880 | the functions of these genes.

00:04:16.840 | And then if you find that a human gene, locus,

00:04:20.880 | something that you've mapped from human genetics

00:04:23.680 | to that gene is related to a particular human disease,

00:04:26.960 | you'd say, aha, now I know the function of the gene

00:04:29.240 | from the model organisms.

00:04:31.440 | I can now go and understand the function

00:04:34.000 | of that gene in human.

00:04:35.440 | But this is all changing.

00:04:38.200 | This is dramatically changed.

00:04:39.360 | So that was the old way of doing basic biology.

00:04:41.720 | You would start with the animal models,

00:04:43.360 | the eukaryotic models, the mammalian models,

00:04:46.120 | and then you would go to human.

00:04:48.360 | Human genetics has been so transformed

00:04:51.060 | in the last decade or two that human genetics

00:04:55.420 | is now actually driving the basic biology.

00:04:58.300 | There is more genetic mutation information

00:05:01.960 | in the human genome than there will ever be

00:05:04.480 | in any other species.

00:05:06.120 | - What do you mean by mutation information?

00:05:08.400 | - So perturbations is how you understand systems.

00:05:11.360 | So an engineer builds systems,

00:05:14.080 | and then they know how they work from the inside out.

00:05:16.240 | A scientist studies systems through perturbations.

00:05:19.300 | You basically say, if I poke that balloon,

00:05:22.280 | what's gonna happen?

00:05:23.120 | And I'm gonna film it in super high resolution,

00:05:24.760 | understand, I don't know, air dynamics or fluid dynamics,

00:05:27.280 | if it's filled with water, et cetera.

00:05:28.840 | So you can then make experimentation by perturbation.

00:05:32.160 | And then the scientific process is sort of building models

00:05:35.420 | that best fit the data, designing new experiments

00:05:40.000 | that best test your models and challenge your models

00:05:42.120 | and so on and so forth.

00:05:43.280 | That's the same thing with science.

00:05:44.520 | Basically, if you're trying to understand

00:05:45.840 | biological science, you basically want to do perturbations

00:05:49.560 | that then drive the models.

00:05:54.560 | - So how do these perturbations

00:05:56.080 | allow you to understand disease?

00:05:58.280 | - So if you know that a gene is related to disease,

00:06:02.360 | you don't wanna just know that it's related to the disease.

00:06:04.760 | You wanna know what is the disease mechanism

00:06:07.320 | because you wanna go and intervene.

00:06:09.900 | So the way that I like to describe it

00:06:11.800 | is that traditionally, epidemiology,

00:06:16.800 | which is basically the study of disease,

00:06:19.280 | sort of the observational study of disease,

00:06:21.820 | has been about correlating one thing with another thing.

00:06:25.720 | So if you have a lot of people with liver disease

00:06:28.240 | who are also alcoholics, you might say,

00:06:30.240 | well, maybe the alcoholism is driving the liver disease,

00:06:33.720 | or maybe those who have liver disease

00:06:35.120 | self-medicate with alcohol.

00:06:36.760 | So the connection could be either way.

00:06:39.880 | With genetic epidemiology,

00:06:42.520 | it's about correlating changes in the genome

00:06:45.600 | with phenotypic differences,

00:06:47.680 | and then you know the direction of causality.

00:06:50.040 | So if you know that a particular gene

00:06:52.120 | is related to the disease,

00:06:55.160 | you can basically say, okay,

00:06:57.760 | perturbing that gene in mouse

00:07:00.080 | causes the mice to have X phenotype.

00:07:02.720 | So perturbing that gene in human

00:07:06.280 | causes the humans to have the disease.

00:07:08.180 | So I can now figure out

00:07:09.360 | what are the detailed molecular phenotypes

00:07:11.680 | in the human that are related

00:07:15.240 | to that organismal phenotype in the disease.

00:07:18.760 | So it's all about understanding disease mechanism,

00:07:21.280 | understanding what are the pathways,

00:07:22.840 | what are the tissues,

00:07:24.440 | what are the processes that are associated with the disease

00:07:27.320 | so that we know how to intervene.

00:07:28.920 | You can then prescribe particular medications

00:07:31.320 | that also alter these processes.

00:07:33.300 | You can prescribe lifestyle changes

00:07:35.260 | that also affect these processes, and so on and so forth.

00:07:38.020 | - That's such a beautiful puzzle to try to solve,

00:07:40.960 | like what kind of perturbations

00:07:42.400 | eventually have this ripple effect

00:07:43.960 | that leads to disease across the population.

00:07:47.000 | And then you study that for animals or mice first,

00:07:50.340 | and then see how that might possibly connect to humans.

00:07:54.520 | How hard is that puzzle of trying to figure out

00:07:57.600 | how little perturbations might lead to,

00:08:01.260 | in a stable way, to a disease?

00:08:03.960 | - In animals, we make the puzzle simpler

00:08:08.340 | because we perturb one gene at a time.

00:08:10.920 | That's the beauty of it, it's the power of animal models.

00:08:13.460 | You can basically decouple the perturbations.

00:08:15.780 | You only do one perturbation,

00:08:17.860 | and you only do strong perturbations at a time.

00:08:21.140 | In human, the puzzle is incredibly complex

00:08:25.180 | because, I mean, obviously,

00:08:26.700 | you don't do human experimentation.

00:08:28.500 | You wait for natural selection

00:08:30.780 | and natural genetic variation

00:08:33.180 | to basically do its own experiments,

00:08:34.880 | which it has been doing for hundreds and thousands of years

00:08:38.960 | in the human population,

00:08:40.720 | and for hundreds of thousands of years

00:08:43.240 | across the history leading to the human population.

00:08:48.240 | So you basically take this natural genetic variation

00:08:52.700 | that we all carry within us.

00:08:54.300 | Every one of us carries six million perturbations.

00:08:57.320 | So I've done six million experiments on you,

00:09:00.480 | six million experiments on me,

00:09:01.880 | six million experiments on every one

00:09:03.940 | of seven billion people on the planet.

00:09:06.220 | - What's the six million correspond to?

00:09:08.620 | - Six million unique genetic variants

00:09:11.960 | that are segregating the human population.

00:09:14.620 | Every one of us carries millions of polymorphic sites,

00:09:19.620 | poly, many, morph, forms.

00:09:22.420 | Polymorphic means many forms, variants.

00:09:25.020 | That basically means that every one of us

00:09:26.900 | has single nucleotide alterations

00:09:29.420 | that we have inherited from mom and from dad

00:09:31.780 | that basically can be thought of

00:09:33.560 | as tiny little perturbations.

00:09:36.000 | Most of them don't do anything,

00:09:38.720 | but some of them lead to all of the phenotypic differences

00:09:42.620 | that we see between us.

00:09:43.900 | The reason why two twins are identical

00:09:46.040 | is because these variants completely determine

00:09:49.100 | the way that I'm gonna look at exactly 93 years of age.

00:09:52.500 | - How happy are you with this kind of data set?

00:09:54.680 | Is it large enough of the human population of Earth?

00:09:59.420 | Is that too big, too small?

00:10:01.680 | - Yeah, so is it large enough is a power analysis question.

00:10:06.680 | And in every one of our grants,

00:10:08.260 | we do a power analysis based on what is the effect size

00:10:11.540 | that I would like to detect

00:10:13.540 | and what is the natural variation in the two forms.

00:10:18.540 | So every time you do a perturbation,

00:10:20.220 | you're asking I'm changing form A into form B.

00:10:23.160 | Form A has some natural phenotypic variation around it

00:10:27.460 | and form B has some natural phenotypic variation around it.

00:10:31.140 | If those variances are large

00:10:33.140 | and the differences between the mean of A

00:10:34.860 | and the mean of B are small,

00:10:37.200 | then you have very little power.

00:10:38.860 | The further the means go apart, that's the effect size,

00:10:43.020 | the more power you have,

00:10:44.420 | and the smaller the standard deviation,

00:10:47.300 | the more power you have.

00:10:48.700 | So basically when you're asking is that sufficiently large,

00:10:52.620 | certainly not for everything,

00:10:54.260 | but we already have enough power

00:10:56.100 | for many of the stronger effects

00:10:59.220 | in the more tight distributions.

00:11:01.700 | - So that's a hopeful message

00:11:02.860 | that there exists parts of the genome

00:11:07.300 | that have a strong effect that has a small variance.

00:11:12.300 | - That's exactly right.

00:11:14.140 | Unfortunately, those perturbations

00:11:16.100 | are the basis of disease in many cases.

00:11:18.700 | So it's not a hopeful message.

00:11:20.540 | Sometimes it's a terrible message.

00:11:22.640 | It's basically, well, some people are sick,

00:11:24.660 | but if we can figure out

00:11:26.740 | what are these contributors to sickness,

00:11:28.980 | we can then help make them better

00:11:30.700 | and help many other people better

00:11:32.420 | who don't carry that exact mutation,

00:11:34.900 | but who carry mutations on the same pathways.

00:11:38.900 | And that's what we like to call the allelic series of a gene.

00:11:42.780 | You basically have many perturbations

00:11:45.380 | of the same gene in different people,

00:11:49.180 | each with a different frequency in the human population

00:11:52.500 | and each with a different effect

00:11:54.520 | on the individual that carries them.

00:11:56.540 | - So you said in the past,

00:11:58.460 | there would be these small experiments

00:12:00.740 | on perturbations in animal models.

00:12:03.780 | What does this puzzle-solving process look like today?

00:12:08.300 | - So we basically have something

00:12:10.860 | like seven billion people in the planet,

00:12:12.420 | and every one of them carries

00:12:13.660 | something like six million mutations.

00:12:16.660 | You basically have an enormous matrix

00:12:19.660 | of genotype by phenotype

00:12:22.960 | by systematically measuring the phenotype

00:12:26.020 | of these individuals.

00:12:27.780 | And the traditional way of measuring this phenotype

00:12:30.540 | has been to look at one trait at a time.

00:12:33.660 | You would gather families,

00:12:35.540 | and you would sort of paint the pedigrees

00:12:38.260 | of a strong effect,

00:12:40.100 | what we like to call Mendelian mutation.

00:12:42.840 | So a mutation that gets transmitted

00:12:46.060 | in a dominant or a recessive,

00:12:48.680 | but strong effect form,

00:12:50.220 | where basically one locus plays a very big role

00:12:53.340 | in that disease.

00:12:54.540 | And you could then look at carriers versus non-carriers

00:12:56.900 | in one family,

00:12:58.220 | carriers versus non-carriers in another family,

00:13:01.100 | and do that for hundreds, sometimes thousands of families,

00:13:04.300 | and then trace these inheritance patterns,

00:13:06.580 | and then figure out what is the gene that plays that role.

00:13:09.500 | - Is this the matrix that you're showing

00:13:11.340 | in talks or lectures?

00:13:14.340 | - So that matrix is the input

00:13:18.940 | to the stuff that I show in talks.

00:13:20.980 | So basically that matrix

00:13:22.060 | has traditionally been strong effect genes.

00:13:24.940 | What the matrix looks like now

00:13:26.780 | is instead of pedigrees, instead of families,

00:13:29.540 | you basically have thousands,

00:13:31.820 | and sometimes hundreds of thousands

00:13:34.100 | of unrelated individuals,

00:13:36.100 | each with all of their genetic variants,

00:13:38.020 | and each with their phenotype,

00:13:39.820 | for example, height, or lipids,

00:13:42.740 | or whether they're sick or not for a particular trait.

00:13:46.180 | That has been the modern view.

00:13:49.740 | Instead of going to families,

00:13:51.300 | we're going to unrelated individuals

00:13:53.260 | with one phenotype at a time.

00:13:55.780 | And what we're doing now,

00:13:56.980 | as we're maturing in all of these sciences,

00:14:00.460 | is that we're doing this in the context

00:14:02.300 | of large medical systems,

00:14:04.780 | or enormous cohorts that are very well phenotyped

00:14:08.180 | across hundreds of phenotypes,

00:14:10.820 | sometimes with our complete electronic health record.

00:14:13.780 | So you can now start relating

00:14:15.500 | not just one gene, segregating one family,

00:14:18.220 | not just thousands of variants,

00:14:20.820 | segregating with one phenotype,

00:14:22.860 | but now you can do millions of variants

00:14:24.900 | versus hundreds of phenotypes.

00:14:27.020 | And as a computer scientist,

00:14:28.300 | I mean, deconvolving that matrix,

00:14:30.540 | partitioning it into the layers of biology

00:14:35.340 | that are associated with every one of these elements

00:14:38.500 | is a dream come true.

00:14:39.780 | It's like the world's greatest puzzle.

00:14:42.860 | And you can now solve that puzzle

00:14:45.340 | by throwing in more and more knowledge

00:14:48.540 | about the function of different genomic regions

00:14:52.740 | and how these functions are changed across tissues

00:14:56.300 | and in the context of disease.

00:14:58.100 | And that's what my group and many other groups are doing.

00:15:00.740 | We're trying to systematically relate

00:15:02.340 | this genetic variation with molecular variation

00:15:05.900 | at the expression level of the genes,

00:15:08.380 | at the epigenomic level of the gene regulatory circuitry,

00:15:12.700 | and at the cellular level

00:15:14.260 | of what are the functions that are happening in those cells,

00:15:17.020 | at the single cell level, using single cell profiling,

00:15:20.340 | and then relate all that vast amount of knowledge

00:15:23.620 | computationally with the thousands of traits

00:15:27.500 | that each of these of thousands of variants are perturbing.

00:15:31.460 | - I mean, this is something we talked about,

00:15:32.900 | I think, last time.

00:15:34.220 | So there's these effects at different levels that happen.

00:15:36.460 | You said at a single cell level,

00:15:38.780 | you're trying to see things that happen

00:15:40.460 | due to certain perturbations.

00:15:42.700 | And then, it's not just like a puzzle

00:15:45.740 | of perturbation and disease.

00:15:49.460 | It's perturbation, then effect at a cellular level,

00:15:53.420 | then at an organ level, at a body,

00:15:55.780 | like how do you disassemble this

00:16:00.020 | into what your group is working on?

00:16:02.420 | You're basically taking a bunch of the hard problems

00:16:05.340 | in the space.

00:16:06.500 | How do you break apart a difficult disease

00:16:09.660 | and break it apart into puzzles

00:16:14.140 | that you can now start solving?

00:16:15.440 | - So there's a struggle here.

00:16:17.100 | Computer scientists love hard puzzles.

00:16:19.660 | And they're like, "Oh, I wanna build a method

00:16:21.820 | that just deconvolves the whole thing computationally."

00:16:24.620 | And that's very tempting and it's very appealing,

00:16:28.420 | but biologists just like to decouple

00:16:31.540 | that complexity experimentally,

00:16:32.980 | to just like peel off layers of complexity experimentally.

00:16:35.940 | And that's what many of these modern tools

00:16:37.780 | that my group and others have both developed and used.

00:16:41.540 | The fact that we can now figure out tricks

00:16:44.500 | for peeling off these layers of complexity

00:16:46.780 | by testing one cell type at a time,

00:16:49.780 | or by testing one cell at a time.

00:16:52.740 | And you could basically say,

00:16:54.100 | what is the effect of these genetic variants

00:16:55.880 | associated with Alzheimer's on human brain?

00:16:58.140 | Human brain sounds like, oh, it's an organ, of course,

00:17:02.520 | just go one organ at a time.

00:17:04.200 | But human brain has, of course,

00:17:05.920 | dozens of different brain regions.

00:17:08.420 | And within each of these brain regions,

00:17:10.200 | dozens of different cell types.

00:17:12.500 | And every single type of neuron,

00:17:14.660 | every single type of glial cell,

00:17:16.820 | between astrocytes, oligodendrocytes, microglia,

00:17:20.540 | between all of the neural cells and the vascular cells

00:17:25.540 | and the immune cells that are co-inhabiting the brain

00:17:29.520 | between the different types of excitatory

00:17:31.660 | and inhibitory neurons that are sort of interacting

00:17:34.360 | with each other between different layers of neurons

00:17:37.340 | in the cortical layers.

00:17:39.020 | Every single one of these has a different type of function

00:17:44.020 | to play in cognition, in interaction with the environment,

00:17:49.380 | in maintenance of the brain, in energetic needs,

00:17:54.140 | in feeding the brain with blood, with oxygen,

00:17:58.300 | in clearing out the debris that are resulting

00:18:01.460 | from the super high energy production

00:18:03.780 | of cognition in humans.

00:18:06.820 | So all of these things are basically

00:18:09.580 | potentially deconvolvable computationally,

00:18:15.200 | but experimentally, you can just do single cell profiling

00:18:19.140 | of dozens of regions of the brain

00:18:21.060 | across hundreds of individuals, across millions of cells.

00:18:24.420 | And then now you have pieces of the puzzle

00:18:28.100 | that you can then put back together

00:18:30.060 | to understand that complexity.

00:18:33.180 | - I mean, first of all, I mean, the human brain,

00:18:35.100 | the cells in the human brain are the most,

00:18:37.740 | okay, maybe I'm romanticizing it,

00:18:39.500 | but cognition seems to be very complicated.

00:18:42.340 | So separating into the function,

00:18:45.900 | breaking Alzheimer's down to the cellular level

00:18:52.780 | seems very challenging.

00:18:54.560 | Is that basically you're trying to find a way

00:18:59.540 | that some perturbation in genome results

00:19:05.220 | in some obvious major dysfunction in the cell?

00:19:10.220 | You're trying to find something like that.

00:19:14.340 | - Exactly.

00:19:15.180 | So what does human genetics do?

00:19:16.900 | Human genetics basically looks at the whole path

00:19:19.580 | from genetic variation all the way to disease.

00:19:22.260 | So human genetics has basically taken thousands

00:19:26.660 | of Alzheimer's cases and thousands of controls

00:19:31.700 | matched for age, for sex, for environmental backgrounds

00:19:36.700 | and so on and so forth.

00:19:38.320 | And then looked at that map where you're asking

00:19:41.640 | what are the individual genetic perturbations

00:19:44.540 | and how are they related to all the way

00:19:46.940 | to Alzheimer's disease?

00:19:48.360 | And that has actually been quite successful.

00:19:51.200 | So we now have more than 27 different loci,

00:19:54.820 | these are genomic regions that are associated

00:19:57.780 | with Alzheimer's at this end to end level.

00:20:02.340 | But the moment you sort of break up that very long path

00:20:05.180 | into smaller levels, you can basically say from genetics,

00:20:09.700 | what are the epigenomic alterations

00:20:12.380 | at the level of gene regulatory elements?

00:20:14.240 | Where that genetic variant perturbs

00:20:16.660 | the control region nearby.

00:20:19.060 | That effect is much larger.

00:20:20.580 | - You mean much larger in terms of

00:20:23.500 | this down the line impact?

00:20:25.460 | - It's much larger in terms of the measurable effect.

00:20:28.100 | This A versus B variance is actually so much cleanly defined

00:20:33.100 | when you go to the shorter branches.

00:20:35.660 | Because for one genetic variant to affect Alzheimer's,

00:20:39.380 | that's a very long path.

00:20:40.740 | That basically means that in the context of millions

00:20:42.940 | of these six million variants that every one of us carries,

00:20:45.500 | that one single nucleotide has a detectable effect

00:20:49.340 | all the way to the end.

00:20:51.300 | I mean, it's just mind boggling that that's even possible.

00:20:54.620 | - That's crazy. - But indeed, yeah.

00:20:55.940 | But indeed, there are such effects.

00:20:57.540 | - So the hope is, or the most, scientifically speaking,

00:21:01.420 | the most effective place where to detect

00:21:03.900 | the alteration that results in disease

00:21:07.460 | is earlier on in the pipeline, as early as possible.

00:21:10.700 | - It's a trade off.

00:21:12.620 | If you go very early on in the pipeline,

00:21:14.860 | now each of these epigenomic alterations,

00:21:17.680 | for example, this enhancer control region,

00:21:20.460 | is active maybe 50% less, which is a dramatic effect.

00:21:25.420 | Now you can ask, well, how much does changing

00:21:27.340 | one regulatory region in the genome

00:21:29.340 | in one cell type change disease?

00:21:31.260 | Well, that path is now long.

00:21:32.700 | So if you instead look at expression,

00:21:37.640 | the path between genetic variation

00:21:39.140 | and the expression of one gene

00:21:40.420 | goes through many enhancer regions,

00:21:42.700 | and therefore it's a subtler effect at the gene level.

00:21:45.500 | But then now you're closer because one gene is acting

00:21:49.620 | in the context of only 20,000 other genes,

00:21:52.300 | as opposed to one enhancer acting

00:21:53.700 | in the context of 2 million other enhancers.

00:21:56.320 | So you basically now have genetic,

00:22:00.020 | epigenomic, the circuitry, transcriptomic,

00:22:03.140 | the gene expression level, and then cellular,

00:22:06.580 | where you can basically say,

00:22:07.500 | I can measure various properties of those cells.

00:22:10.980 | What is the calcium influx rate

00:22:15.060 | when I have this genetic variation?

00:22:17.380 | What is the synaptic density?

00:22:19.580 | What is the electric impulse conductivity?

00:22:22.700 | And so on and so forth.

00:22:24.380 | So you can measure things along this path to disease,

00:22:29.380 | and you can also measure endophenotypes.

00:22:32.560 | You can basically measure your brain activity.

00:22:37.380 | You can do imaging in the brain.

00:22:39.580 | You can basically measure, I don't know,

00:22:41.180 | the heart rate, the pulse, the lipids,

00:22:43.140 | the amount of blood secreted, and so on and so forth.

00:22:46.540 | And then through all of that,

00:22:48.740 | you can basically get at the path to causality,

00:22:52.300 | the path to disease.

00:22:53.480 | - And is there something beyond cellular?

00:22:57.540 | So you mentioned lifestyle interventions

00:23:01.340 | or changes as a way to,

00:23:03.340 | or like be able to prescribe changes in lifestyle.

00:23:07.740 | Like what about organs?

00:23:09.240 | What about like the function of the body as a whole?

00:23:13.220 | - Yeah, absolutely.

00:23:14.060 | So basically, when you go to your doctor,

00:23:16.020 | they always measure your pulse.

00:23:18.140 | They always measure your height.

00:23:19.260 | They always measure your weight, your BMI.

00:23:21.380 | So basically, these are just very basic variables.

00:23:23.980 | But with digital devices nowadays,

00:23:26.300 | you can start measuring hundreds of variables

00:23:27.900 | for every individual.

00:23:29.500 | You can basically also phenotype cognitively

00:23:32.980 | through tests, Alzheimer's patients.

00:23:37.180 | There are cognitive tests that you can measure,

00:23:38.940 | that you typically do for cognitive decline,

00:23:43.620 | these mini mental observations

00:23:46.060 | that you have specific questions to.

00:23:48.420 | You can think of sort of enlarging

00:23:49.940 | the set of cognitive tests.

00:23:51.860 | So in the mouse, for example,

00:23:53.020 | you do experiments for how do they get out of mazes?

00:23:55.580 | How do they find food?

00:23:57.100 | Whether they recall a fear,

00:23:59.160 | whether they shake in a new environment,

00:24:01.300 | and so on and so forth.

00:24:02.340 | In the human, you can have much, much richer phenotypes,

00:24:04.900 | where you can basically say,

00:24:06.340 | not just imaging at the organ level,

00:24:10.680 | but, and all kinds of other activities at the organ level,

00:24:13.740 | but you can also do at the organism level,

00:24:17.540 | you can do behavioral tests.

00:24:19.360 | And how did they do on empathy?

00:24:20.980 | How did they do on memory?

00:24:22.740 | How did they do on long-term memory

00:24:24.780 | versus short-term memory?

00:24:26.020 | And so on and so forth.

00:24:26.860 | - I love how you're calling that phenotype.

00:24:28.740 | I guess it is. - It is.

00:24:30.900 | - But like your behavior patterns

00:24:32.500 | that might change over a period of a life.

00:24:36.060 | Your ability to remember things,

00:24:38.700 | your ability to be, yeah, empathetic or emotionally,

00:24:42.620 | or your intelligence, perhaps even.

00:24:44.940 | - Yeah, but intelligence has hundreds of variables.

00:24:46.940 | You can be your math intelligence,

00:24:48.380 | your literary intelligence,

00:24:49.460 | your puzzle-solving intelligence, your logic.

00:24:51.300 | It could be like hundreds of things.

00:24:52.780 | - And all of that, we're able to measure that

00:24:55.660 | better and better.

00:24:56.500 | And all of that could be connected to the entire pipeline.

00:24:58.820 | - We used to think of each of these as a single variable,

00:25:01.180 | like intelligence.

00:25:02.020 | I mean, that's ridiculous.

00:25:03.180 | It's basically dozens of different genes

00:25:06.600 | that are controlling every single variable.

00:25:10.780 | You can basically think of,

00:25:12.340 | imagine us in a video game

00:25:14.060 | where every one of us has measures of strength,

00:25:17.620 | stamina, energy left, and so on and so forth.

00:25:20.840 | But you could click on each of those five bars

00:25:23.060 | that are just the main bars,

00:25:24.120 | and each of those will just give you then hundreds of bars.

00:25:27.020 | And you can basically say,

00:25:27.980 | "Okay, great, for my machine learning task,

00:25:31.180 | "I want someone who, a human,

00:25:34.280 | "who has these particular forms of intelligence.

00:25:37.200 | "I require now these 20 different things."

00:25:40.500 | And then you can combine those things

00:25:42.240 | and then relate them to, of course,

00:25:44.580 | performance in a particular task.

00:25:46.180 | But you can also relate them to genetic variation

00:25:49.420 | that might be affecting different parts of the brain.

00:25:52.700 | For example, your frontal cortex

00:25:54.260 | versus your temporal cortex

00:25:55.460 | versus your visual cortex, and so on and so forth.

00:25:57.920 | So genetic variation that affects expression of genes

00:26:01.060 | in different parts of your brain

00:26:02.540 | can basically affect your music ability,

00:26:05.120 | your auditory ability, your smell,

00:26:07.460 | your, you know, just dozens of different phenotypes

00:26:11.340 | can be broken down into, you know,

00:26:14.100 | hundreds of cognitive variables,

00:26:15.780 | and then relate each of those to thousands of genes

00:26:19.060 | that are associated with them.

00:26:20.880 | - So somebody who loves RPGs, role-playing games,

00:26:24.640 | there's too few variables that we can control.

00:26:28.340 | So I'm excited, if we're in fact living in a simulation,

00:26:31.180 | this is a video game,

00:26:32.560 | I'm excited by the quality of the video game.

00:26:35.700 | The game designer did a hell of a good job,

00:26:39.620 | so we're impressed.

00:26:41.020 | - So I don't know, the sunset last night

00:26:42.460 | was a little unrealistic.

00:26:43.620 | (both laughing)

00:26:45.060 | - Yeah, yeah, the graphics.

00:26:47.020 | - Exactly.

00:26:47.860 | - Come on, Nvidia.

00:26:48.860 | To zoom back out, we've been talking about

00:26:50.900 | the genetic origins of diseases,

00:26:53.980 | but I think it's fascinating to talk about

00:26:57.140 | what are the most important diseases to understand,

00:27:00.780 | and especially as it connects to the things

00:27:03.360 | that you're working on.

00:27:05.220 | - So it's very difficult to think about

00:27:07.460 | important diseases to understand.

00:27:08.820 | There's many metrics of importance.

00:27:10.260 | One is lifestyle impact.

00:27:13.100 | I mean, if you look at COVID,

00:27:14.300 | the impact on lifestyle has been enormous.

00:27:16.380 | So understanding COVID is important

00:27:18.500 | because it has impacted the wellbeing

00:27:22.040 | in terms of ability to have a job,

00:27:24.620 | ability to have an apartment,

00:27:25.660 | ability to go to work,

00:27:26.920 | ability to have a mental circle of support,

00:27:30.580 | and all of that for millions of Americans,

00:27:33.540 | like huge, huge impact.

00:27:35.460 | So that's one aspect of importance.

00:27:36.980 | So basically mental disorders,

00:27:38.780 | Alzheimer's has a huge importance

00:27:41.060 | in the wellbeing of Americans.

00:27:43.740 | Whether or not it kills someone,

00:27:46.000 | for many, many years, it has a huge impact.

00:27:48.140 | So the first measure of importance is just wellbeing.

00:27:51.900 | - Impact on the quality of life.

00:27:53.540 | - Impact on the quality of life, absolutely.

00:27:55.780 | The second metric, which is much easier to quantify,

00:27:58.420 | is deaths.

00:27:59.860 | - What is the number one killer?

00:28:01.860 | - The number one killer is actually heart disease.

00:28:04.700 | It is actually killing 650,000 Americans per year.

00:28:08.940 | Number two is cancer, with 600,000 Americans.

00:28:14.100 | Number three, far, far down the list, is accidents.

00:28:17.440 | Every single accident combined.

00:28:19.520 | So basically you read the news, accidents,

00:28:22.100 | like there was a huge car crash, all over the news.

00:28:25.700 | But the number of deaths, number three by far, 167,000.

00:28:30.700 | Lower respiratory disease, so that's asthma,

00:28:33.660 | not being able to breathe, and so on and so forth, 160,000.

00:28:37.900 | Alzheimer's, number five, with 120,000.

00:28:42.300 | And then stroke, brain aneurysms, and so on and so forth,

00:28:44.940 | that's 147,000.

00:28:46.700 | Diabetes and metabolic disorders, et cetera, that's 85,000.

00:28:51.080 | The flu, a 60,000.

00:28:53.540 | Suicide, 50,000.

00:28:56.060 | And then overdose, et cetera,

00:28:58.140 | you know, goes further down the list.

00:29:00.080 | So of course, COVID has creeped up

00:29:01.720 | to be the number three killer this year,

00:29:04.820 | with more than 100,000 Americans, and counting.

00:29:09.820 | And, you know, but if you think about sort of

00:29:14.500 | what do we use, what are the most important diseases,

00:29:17.020 | you have to understand both the quality of life,

00:29:20.460 | and the just sheer number of deaths,

00:29:22.820 | and just numbers of years lost, if you wish.

00:29:25.100 | - And each of these diseases you can think of as,

00:29:28.820 | and also including terrorist attacks,

00:29:30.900 | and school shootings, for example,

00:29:32.600 | things which lead to fatalities,

00:29:36.600 | you can look at as problems that could be solved.

00:29:41.060 | And some problems are harder to solve than others.

00:29:44.900 | I mean, that's part of the equation.

00:29:46.740 | So maybe if you look at these diseases,

00:29:48.620 | if you look at heart disease, or cancer, or Alzheimer's,

00:29:53.100 | or just like schizophrenia, and obesity,

00:29:56.700 | that'd be like, not necessarily things that kill you,

00:29:59.620 | but affect the quality of life.

00:30:01.620 | Which problems are solvable, which aren't,

00:30:04.940 | which are harder to solve, which aren't?

00:30:07.300 | - I love your question, because it puts it in the context

00:30:09.780 | of a global effort, rather than just a local effort.

00:30:14.780 | So basically, if you look at the global aspect,

00:30:19.580 | exercise and nutrition are two interventions

00:30:22.620 | that we can, as a society, make a much better job at.

00:30:26.080 | So if you think about sort of the availability

00:30:28.900 | of cheap food, it's extremely high in calories,

00:30:32.900 | it's extremely detrimental for you,

00:30:34.780 | like a lot of processed food, et cetera.

00:30:36.860 | So if we change that equation, and as a society,

00:30:40.900 | we made availability of healthy food much, much easier,

00:30:44.340 | and charged a burger at McDonald's,

00:30:48.500 | the price that it costs on the health system,

00:30:51.520 | then people would actually start buying more healthy foods.

00:30:56.260 | So basically, that's sort of a societal intervention,

00:30:58.660 | if you wish.

00:30:59.500 | In the same way, increasing empathy, increasing education,

00:31:03.980 | increasing the social framework and support

00:31:08.020 | would basically lead to fewer suicides,

00:31:09.960 | it would lead to fewer murders,

00:31:11.800 | it would lead to fewer deaths overall.

00:31:15.740 | So that's something that we as a society can do.

00:31:18.540 | You can also think about external factors

00:31:20.800 | versus internal factors.

00:31:21.900 | So the external factors are basically communicable diseases,

00:31:24.700 | like COVID, like the flu, et cetera.

00:31:26.960 | And the internal factors are basically things like cancer

00:31:31.780 | and Alzheimer's, where basically your genetics

00:31:34.260 | will eventually drive you there.

00:31:36.580 | And then of course, with all of these factors,

00:31:41.620 | every single disease has both a genetic component

00:31:44.480 | and environmental component.

00:31:46.100 | So heart disease, huge genetic contribution.

00:31:49.580 | Alzheimer's, it's like 60% plus genetic.

00:31:55.620 | So I think it's like 79% heritability.

00:31:58.960 | So that basically means that genetics alone

00:32:01.560 | explains 79% of Alzheimer's incidence.

00:32:06.040 | And yes, there's a 21% environmental component

00:32:10.240 | where you could basically enrich your cognitive environment,

00:32:15.240 | enrich your social interactions, read more books,

00:32:19.420 | learn a foreign language, go running,

00:32:21.900 | sort of have a more fulfilling life.

00:32:24.680 | All of that will actually decrease Alzheimer's,

00:32:26.640 | but there's a limit to how much that can impact

00:32:29.360 | because of the huge genetic footprint.

00:32:31.200 | - So this is fascinating.

00:32:32.040 | So each one of these problems have a genetic component

00:32:36.720 | and an environment component.

00:32:38.760 | And so when there's a genetic component,

00:32:41.240 | what can we do about some of these diseases?

00:32:43.000 | What have you worked on?

00:32:44.760 | What can you say that's in terms of problems

00:32:47.160 | that are solvable here or understandable?

00:32:50.400 | - So my group works on the genetic component,

00:32:52.800 | but I would argue that understanding the genetic component

00:32:56.040 | can have a huge impact even on the environmental component.

00:32:59.560 | Why is that?

00:33:00.560 | Because genetics gives us access to mechanism.

00:33:03.440 | And if we can alter the mechanism,

00:33:05.520 | if we can impact the mechanism,

00:33:07.680 | we can perhaps counteract

00:33:09.600 | some of the environmental components.

00:33:11.000 | - Oh, interesting.

00:33:11.960 | - So understanding the biological mechanisms

00:33:15.720 | leading to disease is extremely important

00:33:18.300 | in being able to intervene.

00:33:20.700 | But when you can intervene,

00:33:22.920 | the analogy that I like to give is, for example, for obesity.

00:33:27.320 | Think of it as a giant bathtub of fat.

00:33:30.280 | There's basically fat coming in from your diet

00:33:32.920 | and there's fat coming out from your exercise.

00:33:36.640 | That's an in-out equation.

00:33:40.120 | And that's the equation that everybody's focusing on.

00:33:42.100 | But your metabolism impacts that bathtub.

00:33:47.360 | Basically, your metabolism controls

00:33:50.460 | the rate at which you're burning energy.

00:33:52.980 | It controls the rate at which you're storing energy.

00:33:55.620 | And it also teaches you about the various valves

00:34:01.180 | that control the input and the output equation.

00:34:03.900 | So if we can learn from the genetics, the valves,

00:34:08.020 | we can then manipulate those valves.

00:34:09.920 | And even if the environment is feeding you a lot of fat

00:34:14.100 | and getting a little of that out,

00:34:15.960 | you can just poke another hole at the bathtub

00:34:18.180 | and just get a lot of the fat out.

00:34:19.900 | - Yeah, that's fascinating.

00:34:21.680 | Yeah, so we're not just passive observers of our genetics.

00:34:25.700 | The more we understand,

00:34:26.860 | the more we can come up with actual treatments.

00:34:29.540 | - And I think that's an important aspect to realize.

00:34:33.080 | When people are thinking about strong effect

00:34:36.180 | versus weak effect variants.

00:34:37.980 | So some variants have strong effects.

00:34:39.520 | We talked about these Mendelian disorders

00:34:41.300 | where a single gene has a sufficiently large effect,

00:34:43.980 | penetrance, expressivity, and so on and so forth,

00:34:46.400 | that basically you can trace it in families

00:34:50.760 | with cases and not cases, cases not cases,

00:34:53.160 | and so on and so forth.

00:34:54.960 | But even the, you know, but,

00:34:58.320 | so these are the genes that everybody says,

00:35:01.280 | oh, that's the genes we should go after

00:35:02.920 | because that's a strong effect gene.

00:35:04.920 | I like to think about it slightly differently.

00:35:06.600 | These are the genes where genetic impacts

00:35:11.040 | that have a strong effect were tolerated.

00:35:13.940 | Because every single time we have a genetic association

00:35:17.500 | with disease, it depends on two things.

00:35:20.040 | Number one, the obvious one,

00:35:21.880 | whether the gene has an impact on the disease.

00:35:24.500 | Number two, the more subtle one,

00:35:26.220 | is whether there is genetic variation

00:35:30.300 | standing and circulating and segregating

00:35:33.180 | in the human population that impacts that gene.

00:35:36.400 | Some genes are so darn important

00:35:41.000 | that if you mess with them even a tiny little amount,

00:35:44.360 | that person's dead.

00:35:46.360 | So those genes don't have variation.

00:35:48.940 | You're not gonna find a genetic association

00:35:51.100 | if you don't have variation.

00:35:52.960 | That doesn't mean that the gene has no role.

00:35:55.380 | It simply that the gene,

00:35:56.460 | it simply means that the gene tolerates no mutations.

00:35:59.060 | - So that's actually a strong signal

00:36:00.420 | when there's no variation.

00:36:01.380 | That's so fascinating. - Exactly.

00:36:03.260 | Genes that have very little variation are hugely important.

00:36:06.740 | You can actually rank the importance of genes

00:36:08.900 | based on how little variation they have.

00:36:10.700 | And those genes that have very little variation

00:36:13.700 | but no association with disease,

00:36:16.740 | that's a very good metric to say,

00:36:18.140 | oh, that's probably a developmental gene

00:36:19.840 | because we're not good at measuring those phenotypes.

00:36:22.780 | So it's genes that you can tell evolution

00:36:25.100 | has excluded mutations from,

00:36:27.280 | but yet we can't see them associated

00:36:29.940 | with anything that we can measure nowadays.

00:36:32.040 | It's probably early embryonic lethal.

00:36:34.380 | - What are all the words you just said?

00:36:36.000 | Early embryonic what?

00:36:37.140 | - Lethal.

00:36:38.260 | - Meaning?

00:36:39.100 | - Meaning that--

00:36:39.920 | - If you don't have that, then--

00:36:40.760 | - If you don't have that, you'll die.

00:36:41.580 | - Okay.

00:36:42.420 | There's a bunch of stuff that is required

00:36:45.980 | for a stable functional organism--

00:36:48.380 | - Exactly. - Across the board.

00:36:49.700 | - Exactly. - For our entire,

00:36:51.900 | for an entire species, I guess.

00:36:53.840 | - If you look at sperm, it expresses thousands of proteins.

00:36:58.540 | Does sperm actually need thousands of proteins?

00:37:01.000 | No, but it's probably just testing them.

00:37:04.060 | So my speculation-- - Early on.

00:37:06.460 | - Is that misfolding of these proteins

00:37:09.500 | is an early test for failure.

00:37:11.880 | So that out of the millions of sperm that are possible,

00:37:16.440 | you select the subset that are just not

00:37:19.120 | grossly misfolding thousands of proteins.

00:37:21.840 | - So it's kind of an assert that this is folded correctly.

00:37:25.560 | - Correct.

00:37:26.400 | - Yeah, this, just because if this little thing

00:37:29.520 | about the folding of a protein isn't correct,

00:37:32.380 | that probably means somewhere down the line

00:37:34.520 | there's a bigger issue.

00:37:35.560 | - That's exactly right, so fail fast.

00:37:37.480 | So basically if you look at the mammalian investment

00:37:41.080 | in a newborn, that investment is enormous

00:37:45.880 | in terms of resources.

00:37:47.600 | So mammals have basically evolved mechanisms for fail fast.

00:37:52.600 | Where basically in those early months of development,

00:37:56.400 | I mean, it's horrendous, of course, at the personal level

00:37:59.680 | when you lose your future child,

00:38:03.320 | but in some ways there's so little hope

00:38:08.320 | for that child to develop and sort of make it

00:38:11.100 | through the remaining months that sort of fail fast

00:38:12.920 | is probably a good evolutionary principle.

00:38:16.360 | - Yeah, from an evolutionary perspective.

00:38:17.480 | - For mammals, and of course humans have a lot

00:38:22.100 | of medical resources that you can sort of give

00:38:24.780 | those children a chance.

00:38:26.720 | And we have so much more success in sort of giving folks

00:38:32.560 | who have these strong carrier mutations a chance,

00:38:35.400 | but if they're not even making it

00:38:36.640 | through the first three months, we're not gonna see them.

00:38:39.740 | So that's why when we say what are the most important genes

00:38:43.580 | to focus on, the ones that have a strong effect mutation

00:38:46.840 | or the ones that have a weak effect mutation,

00:38:49.200 | well, the jury might be out,

00:38:51.280 | because the ones that have a strong effect mutation

00:38:53.920 | are basically not mattering as much.

00:38:58.640 | The ones that only have weak effect mutations,

00:39:01.900 | by understanding through genetics

00:39:04.720 | that they have a weak effect mutation

00:39:07.120 | and understanding that they have a causal role

00:39:08.860 | on the disease, we can then say, okay, great,

00:39:11.280 | evolution has only tolerated a 2% change in that gene.

00:39:14.700 | Pharmaceutically, I can go in and induce a 70% change

00:39:19.480 | in that gene.

00:39:20.320 | And maybe I will poke another hole at the bathtub

00:39:24.800 | that was not easy to control in many of the studies

00:39:30.500 | in many of the other strong effect genetic variants.

00:39:35.060 | - So there's this beautiful map across the population

00:39:39.980 | of things that you're saying strong and weak effects,

00:39:42.660 | so stuff with a lot of mutations

00:39:44.880 | and stuff with little mutations, with no mutations.

00:39:48.260 | And you have this map and it lays out the puzzle.

00:39:51.320 | - Yeah, so when I say strong effect,

00:39:53.420 | I mean at the level of individual mutations.

00:39:56.020 | So basically genes where,

00:39:58.900 | so you have to think of first the effect of the gene

00:40:03.120 | on the disease.

00:40:03.960 | Remember how I was sort of painting that map earlier

00:40:07.100 | from genetics all the way to phenotype.

00:40:09.060 | That gene can have a strong effect on the disease,

00:40:14.140 | but the genetic variant might have a weak effect

00:40:16.880 | on the gene.

00:40:18.840 | So basically when you ask what is the effect

00:40:22.440 | of that genetic variant on the disease,

00:40:24.880 | it could be that that genetic variant impacts the gene

00:40:27.200 | by a lot, and then the gene impacts the disease by a little,

00:40:30.920 | or it could be that the genetic variant

00:40:32.440 | impacts the gene by a little,

00:40:33.600 | and then the gene impacts the disease by a lot.

00:40:35.800 | So what we care about is genes that impact the disease a lot,

00:40:40.800 | but genetics gives us the full equation.

00:40:43.560 | And what I would argue is if we couple the genetics

00:40:48.120 | with expression variation,

00:40:51.960 | to basically ask what genes change by a lot,

00:40:54.720 | and which genes correlate with disease by a lot,

00:41:01.240 | even if the genetic variants change them by a little,

00:41:04.560 | then those are the best places to intervene.

00:41:07.760 | Those are the best places where pharmaceutically,

00:41:10.360 | if I have even a modest effect,

00:41:12.680 | I will have a strong effect on the disease.

00:41:15.240 | Whereas those genetic variants

00:41:16.280 | that have a huge effect on the disease,

00:41:18.200 | I might not be able to change that gene by this much

00:41:20.480 | without affecting all kinds of other things.

00:41:22.760 | - Interesting.

00:41:23.600 | So yeah, okay, so that's what we're looking at.

00:41:25.320 | And what have we been able to find

00:41:27.720 | in terms of which disease could be helped?

00:41:31.920 | - Again, don't get me started.

00:41:34.960 | This is, we have found so much.

00:41:38.840 | Our understanding of disease has changed so dramatically

00:41:43.520 | with genetics.

00:41:46.040 | I mean, places that we had no idea would be involved.

00:41:48.880 | So one of the worst things about my genome

00:41:51.720 | is that I have a genetic predisposition

00:41:53.520 | to age-related macular degeneration, AMD.

00:41:56.440 | So it's a form of blindness that causes you

00:41:59.080 | to lose the central part of your vision

00:42:01.680 | progressively as you grow older.

00:42:03.400 | My increased risk is fairly small.

00:42:06.240 | I have an 8% chance.

00:42:07.560 | You only have a 6% chance.

00:42:09.480 | - You, I'm an average.

00:42:11.120 | - Yeah.

00:42:12.080 | - By the way, when you say my, you mean literally yours.

00:42:14.440 | You know this about you.

00:42:15.880 | - I know this about me, yeah.

00:42:17.880 | - Which is kind of, I mean, philosophically speaking

00:42:22.600 | is a pretty powerful thing to live with.

00:42:26.240 | Maybe that's, so we agreed to talk again, by the way,

00:42:29.280 | for the listeners to where we're gonna try to focus

00:42:32.480 | on science today and a little bit of philosophy next time.

00:42:35.920 | But it's interesting to think about

00:42:40.480 | the more you're able to know about yourself

00:42:42.600 | from the genetic information in terms of the diseases,

00:42:45.960 | how that changes your own view of life.

00:42:48.760 | - Yeah, so there's a lot of impact there.

00:42:51.960 | And there's something called genetic exceptionalism,

00:42:55.960 | which basically thinks of genetics

00:42:58.120 | as something very, very different than everything else

00:43:01.120 | as a type of determinism.

00:43:03.760 | And, you know, let's talk about that next time.

00:43:07.280 | So basically-- - That's a good preview.

00:43:08.880 | - Yeah.

00:43:10.000 | So let's go back to AMD.

00:43:11.640 | So basically with AMD, we have no idea what causes AMD.

00:43:16.640 | You know, it was a mystery

00:43:20.080 | until the genetics were worked out.

00:43:23.600 | And now the fact that I know that I have a predisposition

00:43:27.160 | allows me to sort of make some life choices, number one.

00:43:30.480 | But number two, the genes that lead to that predisposition

00:43:34.920 | give us insights as to how does it actually work.

00:43:37.360 | And that's a place where genetics

00:43:40.800 | gave us something totally unexpected.

00:43:42.920 | So there's a complement pathway,

00:43:46.400 | which is an immune function pathway

00:43:48.840 | that was in, you know,

00:43:52.400 | most of the loci associated with AMD.

00:43:55.880 | And that basically told us that, wow,

00:43:57.600 | there's an immune basis to this eye disorder

00:44:02.520 | that people had just not expected before.

00:44:05.120 | If you look at complement,

00:44:06.760 | it was recently also implicated in schizophrenia.

00:44:10.640 | And there's a type of microglia

00:44:13.960 | that is involved in synaptic pruning.

00:44:16.960 | So synapses are the connections between neurons.

00:44:19.960 | And in this whole use it or lose it view

00:44:22.680 | of mental cognition and other capabilities,

00:44:26.840 | you basically have microglia, which are immune cells

00:44:31.200 | that are sort of constantly traversing your brain

00:44:33.720 | and then pruning neuronal connections,

00:44:36.160 | pruning synaptic connections that are not utilized.

00:44:40.080 | So in schizophrenia,

00:44:42.600 | there's thought to be a change in the pruning

00:44:47.040 | that basically if you don't prune your synapses

00:44:49.480 | the right way,

00:44:50.960 | you will actually have an increased role of schizophrenia.

00:44:53.720 | This is something that was completely unexpected

00:44:56.320 | for schizophrenia.

00:44:57.160 | Of course, we knew it has to do with neurons,

00:44:58.960 | but the role of the complement complex,

00:45:01.280 | which is also implicated in AMD,

00:45:03.760 | which is now also implicated in schizophrenia,

00:45:05.400 | was a huge surprise.

00:45:06.280 | - What's the complement complex?

00:45:07.920 | - So it's basically a set of genes, the complement genes,

00:45:11.160 | that are basically having various immune roles.

00:45:15.320 | And as I was saying earlier,

00:45:16.440 | our immune system has been co-opted

00:45:18.520 | for many different roles across the body.

00:45:21.040 | So they actually play many diverse roles.

00:45:23.400 | - And somehow the immune system is connected

00:45:26.360 | to the synaptic pruning process.

00:45:28.200 | - Exactly. - The process of it.

00:45:29.120 | - Exactly.

00:45:29.960 | So immune cells were co-opted to prune synapses.

00:45:33.080 | - How did you figure this out?

00:45:34.280 | (laughing)

00:45:35.640 | How does one go about figuring this intricate connection,

00:45:40.200 | like pipeline of connections out?

00:45:42.040 | - Yeah, let me give you another example.

00:45:43.840 | So Alzheimer's disease,

00:45:45.960 | the first place that you would expect it to act

00:45:48.200 | is obviously the brain.

00:45:49.680 | So we had basically this

00:45:51.280 | roadmap epigenomics consortium view of the human epigenome,

00:45:56.800 | the largest map of the human epigenome

00:45:59.080 | that has ever been built,

00:46:01.760 | across 127 different tissues and samples

00:46:05.680 | with dozens of epigenomic marks

00:46:07.800 | measured in hundreds of donors.

00:46:10.440 | So what we've basically learned through that

00:46:13.520 | is that you basically can map

00:46:15.840 | what are the active gene regulatory elements

00:46:18.040 | for every one of the tissues in the body.

00:46:20.200 | And then we connected these gene regulatory active maps

00:46:25.080 | of basically what regions of the human genome

00:46:28.360 | are turning on in every one of different tissues.

00:46:31.920 | We then can go back and say,

00:46:34.440 | where are all of the genetic loci

00:46:37.480 | that are associated with disease?

00:46:39.160 | This is something that my group,

00:46:41.720 | I think was the first to do back in 2010

00:46:45.160 | in this Ernst Nature Biotech paper.

00:46:48.640 | But basically we were for the first time able to show

00:46:50.960 | that specific chromatin states,

00:46:53.160 | specific epigenomic states,

00:46:54.560 | in that case enhancers,

00:46:56.160 | were in fact enriched in disease associated variants.

00:47:00.640 | We pushed that further

00:47:02.240 | in the Ernst Nature paper a year later,

00:47:05.440 | and then in this roadmap epigenomics paper,

00:47:08.920 | a few years after that.

00:47:11.320 | But basically that matrix that you mentioned earlier

00:47:15.120 | was in fact the first time that we could see

00:47:17.320 | what genetic traits have genetic variants

00:47:21.160 | that are enriched in what tissues in the body.

00:47:26.040 | And a lot of that map made complete sense.

00:47:28.640 | If you looked at a diversity of immune traits,

00:47:31.440 | like allergies and type 1 diabetes and so on and so forth,

00:47:34.720 | you basically could see that they were enriching,

00:47:37.800 | that the genetic variants associated with those traits

00:47:40.120 | were enriched in enhancers,

00:47:42.960 | in these gene regulatory elements,

00:47:44.600 | active in T cells and B cells

00:47:46.520 | and hematopoietic stem cells and so on and so forth.

00:47:49.040 | So that basically gave us a confirmation in many ways

00:47:54.560 | that those immune traits were indeed

00:47:56.880 | enriching in immune cells.

00:47:58.840 | If you looked at type 2 diabetes,

00:48:02.480 | you basically saw an enrichment in only one type of sample,

00:48:06.000 | and it was pancreatic islets.

00:48:07.880 | And we know that type 2 diabetes

00:48:10.760 | sort of stems from the dysregulation of insulin

00:48:14.680 | in the beta cells of pancreatic islets.

00:48:17.200 | And that sort of was spot on, super precise.

00:48:19.920 | If you looked at blood pressure,

00:48:22.800 | where would you expect blood pressure to occur?

00:48:25.760 | You know, I don't know, maybe in your metabolism,

00:48:27.800 | in ways that you process coffee or something like that,

00:48:30.160 | maybe in your brain, the way that you stress out

00:48:32.360 | and increases your blood pressure, et cetera.

00:48:34.240 | What we found is that blood pressure localized specifically

00:48:37.240 | in the left ventricle of the heart.

00:48:40.280 | So the enhancers of the left ventricle in the heart

00:48:42.680 | contained a lot of genetic variants

00:48:44.160 | associated with blood pressure.

00:48:45.760 | If you look at height,

00:48:48.320 | we found an enrichment specifically

00:48:50.480 | in embryonic stem cell enhancers.

00:48:52.280 | So the genetic variants predisposing you

00:48:54.800 | to be taller or shorter

00:48:56.400 | are in fact acting in developmental stem cells.

00:48:59.320 | Makes complete sense.

00:49:01.280 | If you looked at inflammatory bowel disease,

00:49:03.400 | you basically found inflammatory, which is immune,

00:49:06.640 | and also bowel disease, which is digestive.

00:49:09.760 | And indeed, we saw a double enrichment,

00:49:12.000 | both in the immune cells and in the digestive cells.

00:49:15.680 | So that basically told us that,

00:49:16.880 | aha, this is acting in both components.

00:49:18.880 | There's an immune component to inflammatory bowel disease

00:49:21.360 | and there's a digestive component.

00:49:23.320 | And the big surprise was for Alzheimer's.

00:49:25.840 | We had seven different brain samples.

00:49:28.120 | We found zero enrichment in the brain samples

00:49:33.120 | for genetic variants associated with Alzheimer's.

00:49:36.080 | And this is mind boggling.

00:49:37.880 | Our brains were literally hurting.

00:49:39.760 | What is going on?

00:49:41.560 | And what is going on is that the brain samples

00:49:44.520 | are primarily neurons, oligodendrocytes, and astrocytes

00:49:49.520 | in terms of the cell types that make them up.

00:49:52.920 | So that basically indicated that genetic variants

00:49:56.480 | associated with Alzheimer's were probably not acting

00:50:00.520 | in oligodendrocytes, astrocytes, or neurons.

00:50:04.400 | So what could they be acting in?

00:50:06.240 | Well, the fourth major cell type is actually microglia.

00:50:09.560 | Microglia are resident immune cells in your brain.

00:50:13.680 | - Oh, nice.

00:50:14.520 | The immune, oh, wow.

00:50:17.320 | - And they are CD14+, which is this sort of

00:50:21.280 | cell surface markers of those cells.

00:50:24.080 | So they're CD14+ cells, just like microphages

00:50:27.000 | that are circulating in your blood.

00:50:29.080 | The microglia are resident monocytes

00:50:33.960 | that are basically sitting in your brain.

00:50:35.560 | They're tissue-specific monocytes.

00:50:38.280 | And every one of your tissues, like your fat, for example,

00:50:41.440 | has a lot of macrophages that are resident.

00:50:43.680 | And the M1 versus M2 macrophage ratio

00:50:46.720 | has a huge role to play in obesity.

00:50:49.120 | And so basically, again, these immune cells are everywhere,

00:50:51.760 | but basically what we found,

00:50:53.480 | through this completely unbiased view

00:50:56.080 | of what are the tissues that likely

00:50:58.280 | underlie different disorders,

00:51:00.440 | we found that Alzheimer's was humongously enriched

00:51:05.040 | in microglia, but not at all in the other cell types.

00:51:08.960 | - So what are we supposed to make of that

00:51:11.080 | if you look at the tissues involved?

00:51:14.720 | Is that simply useful for indication

00:51:18.040 | of propensity for disease?

00:51:20.920 | Or does it give us somehow a pathway of treatment?

00:51:24.560 | - It's very much the second.

00:51:26.040 | If you look at the way to therapeutics,

00:51:31.040 | you have to start somewhere.

00:51:33.760 | What are you gonna do?

00:51:34.600 | You're gonna basically make assays

00:51:36.960 | that manipulate those genes and those pathways

00:51:41.960 | in those cell types.

00:51:43.680 | So before we know the tissue of action,

00:51:46.640 | we don't even know where to start.

00:51:49.120 | We basically are at a loss.

00:51:50.960 | But if you know the tissue of action,

00:51:52.520 | and even better, if you know the pathway of action,

00:51:55.240 | then you can basically screen your small molecules,

00:51:58.160 | not for the gene, you can screen them directly

00:52:00.360 | for the pathway in that cell type.

00:52:03.080 | So you can basically develop

00:52:04.560 | a high throughput multiplexed robotic system

00:52:08.880 | for testing the impact of your favorite molecules

00:52:12.960 | that you know are safe, efficacious,

00:52:14.800 | and sort of hit that particular gene, and so on and so forth.

00:52:18.800 | You can basically screen those molecules

00:52:20.800 | against either a set of genes that act in that pathway,

00:52:25.800 | or on the pathway directly by having a cellular assay.

00:52:29.640 | And then you can basically go into mice and do experiments

00:52:32.360 | and basically sort of figure out ways

00:52:33.800 | to manipulate these processes

00:52:36.280 | that allow you to then go back to humans

00:52:38.520 | and do a clinical trial that basically says,

00:52:40.200 | okay, I was able indeed to reverse these processes in mice,

00:52:43.880 | can I do the same thing in humans?

00:52:46.080 | So the knowledge of the tissues

00:52:49.080 | gives you the pathway to treatment.

00:52:51.280 | But that's not the only part.

00:52:52.640 | There are many additional steps

00:52:54.760 | to figuring out the mechanism of disease.

00:52:56.960 | - I mean, so that's really promising.

00:52:58.760 | Maybe to take a small step back,

00:53:01.760 | you've mentioned all these puzzles

00:53:03.560 | that were figured out with the Nature paper for,

00:53:07.720 | I mean, you've mentioned a ton of diseases,

00:53:10.540 | from obesity to Alzheimer's, even schizophrenia,

00:53:14.800 | I think you mentioned.

00:53:15.840 | What is the actual methodology of figuring this out?

00:53:20.640 | - So indeed, I mentioned a lot of diseases,

00:53:22.920 | and my lab works on a lot of different disorders.

00:53:25.760 | And the reason for that is that

00:53:29.280 | if you look at the,

00:53:31.680 | if you look at biology,

00:53:36.360 | it used to be zoology departments

00:53:39.560 | and botanology departments and virology departments

00:53:42.560 | and so on and so forth.

00:53:43.400 | And MIT was one of the first schools

00:53:45.160 | to basically create a biology department,

00:53:47.080 | like, oh, we're gonna study all of life suddenly.

00:53:49.520 | Why was that even a case?

00:53:51.320 | Because the advent of DNA and the genome

00:53:55.240 | and the central dogma of DNA makes RNA makes protein,

00:53:58.520 | in many ways, unified biology.

00:54:00.620 | You could suddenly study the process of transcription

00:54:04.360 | in viruses or in bacteria

00:54:06.920 | and have a huge impact on yeast and fly

00:54:10.120 | and maybe even mammals,

00:54:12.040 | because of this realization

00:54:14.600 | of these common underlying processes.

00:54:16.440 | And in the same way that DNA unified biology,

00:54:22.400 | genetics is unifying disease studies.

00:54:27.080 | So you used to have,

00:54:28.460 | you used to have, I don't know,

00:54:34.560 | cardiovascular disease department

00:54:36.640 | and neurological disease department

00:54:40.120 | and neurodegeneration department

00:54:41.960 | and basically immune and cancer and so on and so forth.

00:54:46.960 | And all of these were studied in different labs

00:54:50.520 | because it made sense,

00:54:52.640 | because basically the first step

00:54:54.240 | was understanding how the tissue functions

00:54:56.440 | and we kind of knew the tissues involved

00:54:57.960 | in cardiovascular disease and so on and so forth.

00:55:00.560 | But what's happening with human genetics

00:55:01.840 | is that all of that,

00:55:03.520 | all of these walls and edifices

00:55:05.600 | that we had built are crumbling.

00:55:08.280 | And the reason for that is that genetics

00:55:11.680 | is in many ways revealing unexpected connections.

00:55:16.440 | So suddenly we now have to bring the immunologists

00:55:19.080 | to work on Alzheimer's.

00:55:20.300 | They were never in the room.

00:55:22.600 | They were in another building altogether.

00:55:25.800 | The same way for schizophrenia,

00:55:28.040 | we now have to sort of worry about

00:55:30.440 | all these interconnected aspects.

00:55:33.060 | For metabolic disorders,

00:55:34.300 | we're finding contributions from brain.

00:55:36.380 | So suddenly we have to call the neurologist

00:55:38.920 | from the other building and so on and so forth.

00:55:41.200 | So in my view, it makes no sense anymore

00:55:46.000 | to basically say, oh, I'm a geneticist

00:55:48.400 | studying immune disorders.

00:55:50.640 | I mean, that's ridiculous because,

00:55:52.440 | I mean, of course, in many ways,

00:55:54.280 | you still need to sort of focus.

00:55:56.260 | But what we're doing is that we're basically saying,

00:55:59.240 | we'll go wherever the genetics takes us.

00:56:02.080 | And by building these massive resources,

00:56:05.580 | by working on our latest maps, now 833 tissues,

00:56:10.280 | sort of the next generation of the epigenomics roadmap,

00:56:13.740 | which we're now called EpiMap, is 833 different tissues.

00:56:17.880 | And using those, we've basically found enrichments

00:56:21.140 | in 540 different disorders.

00:56:24.460 | Those enrichments are not like, oh, great,

00:56:26.200 | you guys work on that and we'll work on this.

00:56:29.180 | They're intertwined amazingly.

00:56:31.840 | So of course there's a lot of modularity,

00:56:34.500 | but there's these enhancers that are sort of broadly active

00:56:37.120 | and these disorders that are broadly active.

00:56:39.500 | So basically some enhancers are active in all tissues

00:56:41.820 | and some disorders are enriching in all tissues.

00:56:45.060 | So basically there's these multifactorial

00:56:47.740 | and these other class,

00:56:48.580 | which I like to call polyfactorial diseases,

00:56:51.420 | which are basically lighting up everywhere.

00:56:54.140 | And in many ways, it's sort of cutting across these walls

00:56:59.140 | that were previously built across these departments.

00:57:01.640 | - And the polyfactorial ones were probably

00:57:03.980 | the previous structure departments

00:57:06.180 | wasn't equipped to deal with those.

00:57:08.620 | I mean, again, maybe it's a romanticized question,

00:57:12.860 | but there's in physics, there's a theory of everything.

00:57:16.780 | Do you think it's possible to move towards

00:57:19.740 | an almost theory of everything of disease

00:57:22.520 | from a genetic perspective?

00:57:24.020 | So if this unification continues, is it possible that,

00:57:27.940 | like, do you think in those terms,

00:57:29.460 | like trying to arrive at a fundamental understanding

00:57:33.060 | of how disease emerges, period?

00:57:35.460 | - That unification is not just foreseeable, it's inevitable.

00:57:40.460 | I see it as inevitable.

00:57:43.440 | We have to go there.

00:57:45.140 | You cannot be a specialist anymore if you're a genomicist,

00:57:49.660 | you have to be a specialist in every single disorder.

00:57:53.740 | And the reason for that is that the fundamental understanding

00:57:58.260 | of the circuitry of the human genome

00:58:00.780 | that you need to solve schizophrenia,

00:58:03.800 | that fundamental circuitry is hugely important

00:58:08.020 | to solve Alzheimer's.

00:58:09.500 | And that same circuitry is hugely important

00:58:11.420 | to solve metabolic disorders.

00:58:13.020 | And that same exact circuitry is hugely important

00:58:16.940 | for solving immune disorders and cancer

00:58:19.340 | and every single disease.

00:58:22.180 | So all of them have the same sub task.

00:58:26.580 | And I teach dynamic programming in my class.

00:58:29.580 | Dynamic programming is all about sort of

00:58:31.380 | not redoing the work, it's reusing the work that you do once.

00:58:36.380 | So basically for us to say, oh, great,

00:58:40.140 | you guys in the immune building,

00:58:41.780 | go solve the fundamental circuitry of everything.

00:58:44.140 | And then you guys in the schizophrenia building

00:58:46.120 | go solve the fundamental circuitry

00:58:47.380 | of everything separately, is crazy.

00:58:50.040 | So what we need to do is come together

00:58:52.700 | and sort of have a circuitry group,

00:58:55.580 | the circuitry building that sort of tries

00:58:57.500 | to solve the circuitry of everything.

00:58:59.300 | And then the immune folks who will apply this knowledge

00:59:03.840 | to all of the disorders that are associated

00:59:06.620 | with immune dysfunction.

00:59:09.080 | And the schizophrenia folks will basically interact

00:59:12.500 | with both the immune folks and with the neuronal folks.

00:59:15.460 | And all of them will be interacting

00:59:16.780 | with the circuitry folks and so on and so forth.

00:59:18.940 | So that's sort of the current structure of my group,

00:59:21.500 | if you wish.

00:59:22.340 | So basically what we're doing is focusing

00:59:24.180 | on the fundamental circuitry.

00:59:25.700 | But at the same time, we're the users of our own tools

00:59:31.160 | by collaborating with many other labs

00:59:34.860 | in every one of these disorders that we mentioned.

00:59:37.460 | We basically have a heart focus on cardiovascular disease,

00:59:41.060 | coronary artery disease, heart failure, and so on and so forth.

00:59:44.220 | We have an immune focus on several immune disorders.

00:59:48.840 | We have a cancer focus on metastatic melanoma

00:59:53.840 | and immunotherapy response.

00:59:55.520 | We have a psychiatric disease focus on schizophrenia,

01:00:00.360 | autism, PTSD, and other psychiatric disorders.

01:00:04.060 | We have an Alzheimer's and neurodegeneration focus

01:00:06.880 | on Huntington's disease, ALS, and AD-related disorders

01:00:11.780 | like frontotemporal dementia and Lewy body dementia.

01:00:14.380 | And of course, a huge focus on Alzheimer's.

01:00:16.660 | We have a metabolic focus on the role of exercise and diet

01:00:21.200 | and sort of how they're impacting metabolic organs

01:00:26.200 | across the body and across many different tissues.

01:00:28.980 | And all of them are interfacing with the circuitry.

01:00:33.980 | And the reason for that is another computer science principle

01:00:38.100 | of eat your own dog food.

01:00:40.640 | If everybody ate their own dog food,

01:00:44.900 | dog food would taste a lot better.

01:00:46.600 | The reason why Microsoft Excel and Word and PowerPoint

01:00:51.720 | was so important and so successful

01:00:55.140 | is because the employees that were working on them

01:00:58.340 | were using them for their day-to-day tasks.

01:01:01.460 | You can't just simply build a circuitry and say,

01:01:04.420 | "Here it is, guys, take the circuitry, we're done,"

01:01:07.060 | without being the users of that circuitry

01:01:09.380 | because you then go back.

01:01:11.220 | And because we span the whole spectrum

01:01:13.420 | from profiling the epigenome, using comparative genomics,

01:01:17.420 | finding the important nucleotides in the genome,

01:01:19.800 | building the basic functional map

01:01:21.740 | of what are the genes in the human genome,

01:01:24.500 | what are the gene regulatory elements of the human genome.

01:01:27.620 | I mean, over the years, we've written a series of papers

01:01:30.260 | on how do you find human genes in the first place,

01:01:32.740 | using comparative genomics.

01:01:34.060 | How do you find the motifs that are the building blocks

01:01:36.940 | of gene regulation, using comparative genomics?

01:01:39.620 | How do you then find how these motifs come together

01:01:43.060 | and act in specific tissues, using epigenomics?

01:01:46.220 | How do you link regulators to enhancers and enhancers

01:01:50.980 | to their target genes,

01:01:52.220 | using epigenomics and regulatory genomics?

01:01:55.220 | So through the years, we've basically built

01:01:57.500 | all this infrastructure for understanding

01:02:00.380 | what I like to say, every single nucleotide

01:02:03.700 | of the human genome and how it acts in every one

01:02:07.420 | of the major cell types and tissues of the human body.

01:02:10.500 | I mean, this is no small task.

01:02:12.020 | This is an enormous task that takes the entire field.

01:02:15.460 | And that's something that my group has taken on,

01:02:18.060 | along with many other groups.

01:02:19.500 | And we have also, and that sort of, I think,

01:02:23.100 | sets my group perhaps apart,

01:02:24.780 | we have also worked with specialists

01:02:26.940 | in every one of these disorders

01:02:28.820 | to basically further our understanding

01:02:30.700 | all the way down to disease.

01:02:32.460 | And in some cases, collaborating with pharma

01:02:34.420 | to go all the way down to therapeutics

01:02:36.580 | because of our deep, deep understanding

01:02:39.980 | of that basic circuitry

01:02:41.140 | and how it allows us to now improve the circuitry,

01:02:46.500 | not just treat it as a black box,

01:02:49.060 | but basically go and say, okay,

01:02:50.300 | we need a better cell type specific wiring

01:02:53.620 | that we now have at the tissue specific level.

01:02:56.420 | So we're focusing on that because we're understanding

01:02:59.500 | the needs from the disease front.

01:03:01.780 | - So you have a sense of the entire pipeline.

01:03:04.060 | I mean, one, maybe you can indulge me,

01:03:07.940 | one nice question to ask would be,

01:03:09.780 | how do you, from the scientific perspective,

01:03:14.340 | go from knowing nothing about the disease

01:03:17.340 | to going, you said, to going through the entire pipeline

01:03:22.060 | and actually have a drug or a treatment

01:03:25.260 | that cures that disease?

01:03:26.740 | - So that's an enormously long path

01:03:30.100 | and an enormously great challenge.

01:03:32.460 | And what I'm trying to argue is that

01:03:34.940 | it progresses in stages of understanding

01:03:38.820 | rather than one gene at a time.

01:03:40.900 | The traditional view of biology was

01:03:42.740 | you have one postdoc working on this gene

01:03:44.860 | and another postdoc working on that gene.

01:03:47.460 | And they'll just figure out everything about that gene.

01:03:50.300 | And that's their job.

01:03:51.940 | What we've realized is how polygenic

01:03:54.500 | the diseases are.

01:03:56.260 | So we can't have one postdoc per gene anymore.

01:03:58.140 | We now have to have these cross-cutting needs.

01:04:03.140 | And I'm gonna describe the path to circuitry

01:04:07.460 | along those needs.

01:04:10.140 | And every single one of these paths,

01:04:12.900 | we are now doing in parallel across thousands of genes.

01:04:16.860 | So the first step is you have a genetic association.

01:04:21.660 | And we talked a little bit about sort of the Mendelian path

01:04:24.740 | and the polygenic path to that association.

01:04:27.660 | So the Mendelian path was looking through families

01:04:30.020 | to basically find gene regions and ultimately genes

01:04:34.500 | that are underlying particular disorders.

01:04:36.740 | The polygenic path is basically looking at

01:04:40.580 | unrelated individuals in this giant matrix

01:04:43.220 | of genotype by phenotype,

01:04:44.940 | and then finding hits where a particular variant

01:04:47.940 | impacts disease all the way to the end.

01:04:51.300 | And then we now have a connection,

01:04:53.900 | not between a gene and a disease,

01:04:56.700 | but between a genetic region and a disease.

01:05:00.100 | And that distinction is not understood by most people.

01:05:03.460 | So I'm gonna explain it a little bit more.

01:05:05.580 | Why do we not have a connection between a gene

01:05:10.020 | and a disease, but we have a connection

01:05:11.340 | between a genetic region and a disease?

01:05:13.380 | The reason for that is that 93% of genetic variants

01:05:19.980 | that are associated with disease

01:05:21.900 | don't impact the protein at all.

01:05:25.220 | So if you look at the human genome, there's 20,000 genes.

01:05:29.940 | There's 3.2 billion nucleotides.

01:05:32.100 | Only 1.5% of the genome codes for proteins.

01:05:38.140 | The other 98.5% does not code for proteins.

01:05:46.100 | If you now look at where are the disease variants located,

01:05:49.460 | 93% of them fall in that outside the genes portion.

01:05:55.700 | Of course, genes are enriched,

01:05:57.340 | but they're only enriched by a factor of three.

01:06:00.500 | That means that still 93% of genetic variants

01:06:03.460 | fall outside the proteins.

01:06:05.700 | Why is that difficult?

01:06:08.140 | Why is that a problem?

01:06:09.420 | The problem is that when a variant falls outside the gene,

01:06:14.220 | you don't know what gene is impacted by that variant.

01:06:16.860 | You can't just say, "Oh, it's near this gene.

01:06:19.220 | Let's just connect that variant to the gene."

01:06:20.860 | And the reason for that is that the genome circuitry

01:06:24.340 | is very often long range.

01:06:26.740 | So you basically have that genetic variant

01:06:30.740 | that could sit in the intron of one gene.

01:06:34.500 | And an intron is sort of the place

01:06:36.300 | between the exons that code for proteins.

01:06:38.140 | So proteins are split up into exons and introns,

01:06:41.300 | and every exon codes for a particular subset

01:06:43.740 | of amino acids, and together they're spliced together,

01:06:46.860 | and then make the final protein.

01:06:49.180 | So that genetic variant might be sitting

01:06:50.660 | in an intron of a gene.

01:06:51.820 | It's transcribed with the gene,

01:06:53.620 | it's processed and then excised,

01:06:55.500 | but it might not impact this gene at all.

01:06:57.140 | It might actually impact another gene

01:06:59.500 | that's a million nucleotides away.

01:07:01.060 | - So it's just riding along,

01:07:02.380 | even though it has nothing to do

01:07:03.540 | with this nearby neighborhood.

01:07:05.820 | - That's exactly right.

01:07:07.580 | Let me give you an example.

01:07:09.580 | The strongest genetic association with obesity

01:07:12.380 | was discovered in this FTO gene,

01:07:15.660 | fat and obesity associated gene.

01:07:18.340 | So this FTO gene was studied ad nauseum.

01:07:23.340 | People did tons of experiments on it.

01:07:26.700 | They figured out that FTO is in fact

01:07:29.140 | RNA methylation trans rays.

01:07:32.980 | It basically, it sort of impacts something

01:07:36.060 | that we know that we call the epitranscriptome,

01:07:38.740 | just like the genome can be modified,

01:07:40.740 | the transcriptome, the transcripts of the genes

01:07:43.580 | can be modified.

01:07:44.660 | And we basically said, oh great,

01:07:46.620 | that means that epitranscriptomics

01:07:48.500 | is hugely involved in obesity

01:07:49.980 | because that gene FTO is clearly

01:07:53.740 | where the genetic locus is at.

01:07:55.620 | My group studied FTO in collaboration

01:08:00.180 | with a wonderful team led by Melina Klausnitzer.

01:08:04.140 | And what we found is that this FTO locus,

01:08:08.460 | even though it is associated with obesity,

01:08:11.740 | does not implicate the FTO gene.

01:08:14.860 | The genetic variant sits in the first intron

01:08:19.420 | of the FTO gene, but it controls two genes,

01:08:23.500 | IRX3 and IRX5, that are sitting 1.2 million nucleotides away,

01:08:28.500 | several genes away.

01:08:31.860 | - Oh boy.

01:08:34.500 | When am I supposed to feel about that?

01:08:36.140 | Isn't that like super complicated then?

01:08:38.620 | - So the way that I was introduced at a conference

01:08:40.740 | a few years ago was, and here's Manolis Kellis

01:08:43.780 | who wrote the most depressing paper of 2015.

01:08:46.940 | (laughs)

01:08:48.620 | And the reason for that is that

01:08:49.780 | the entire pharmaceutical industry was so comfortable

01:08:52.180 | that there was a single gene in that locus.

01:08:56.020 | Because in some loci, you basically have three dozen genes

01:08:58.500 | that are all sitting in the same region of association.

01:09:01.340 | And you're like, oh gosh, which ones of those is it?

01:09:04.020 | But even that question of which ones of those is it,

01:09:06.780 | is making the assumption that it is one of those,

01:09:09.860 | as opposed to some random gene just far, far away,

01:09:12.860 | which is what our paper showed.

01:09:14.820 | So basically what our paper showed is that

01:09:16.860 | you can't ignore the circuitry.

01:09:19.060 | You have to first figure out the circuitry,

01:09:21.060 | all of those long range interactions,

01:09:23.180 | how every genetic variant impacts the expression

01:09:25.780 | of every gene in every tissue imaginable

01:09:28.620 | across hundreds of individuals.

01:09:30.780 | And then you now have one of the building blocks,

01:09:33.980 | not even all of the building blocks,

01:09:35.660 | for then going and understanding disease.

01:09:39.340 | - So okay, so embrace the wholeness of the circuitry.

01:09:44.340 | - Correct.

01:09:45.500 | - But what, so back to the question of starting

01:09:48.060 | knowing nothing to the disease and going to the treatment.

01:09:51.660 | So what are the next steps?

01:09:53.460 | - So you basically have to first figure out the tissue

01:09:56.100 | and then describe how you figure out the tissue.

01:09:57.980 | You figure out the tissue by taking

01:09:59.260 | all of these non-coding variants

01:10:01.300 | that are sitting outside proteins,

01:10:03.220 | and then figuring out what are the epigenomic enrichments.

01:10:06.700 | And the reason for that, thankfully,

01:10:10.220 | is that there is convergence,

01:10:12.420 | that the same processes are impacted in different ways

01:10:17.060 | by different loci.

01:10:18.180 | And that's a saving grace for our field.

01:10:23.060 | The fact that if I look at hundreds of genetic variants

01:10:26.140 | associated with Alzheimer's,

01:10:27.700 | they localize in a small number of processes.

01:10:31.820 | - Can you clarify why that's hopeful?

01:10:34.580 | So they show up in the same exact way

01:10:36.860 | in the specific set of processes?

01:10:40.180 | - Yeah, so basically there's a small number

01:10:42.460 | of biological processes that underlie,

01:10:44.980 | or at least that play the biggest role in every disorder.

01:10:48.460 | So in Alzheimer's, you basically have

01:10:51.380 | maybe 10 different types of processes.

01:10:53.900 | One of them is lipid metabolism.

01:10:56.180 | One of them is immune cell function.

01:10:58.740 | One of them is neuronal energetics.

01:11:02.300 | So these are just a small number of processes,

01:11:04.100 | but you have multiple lesions,

01:11:06.940 | multiple genetic perturbations

01:11:08.580 | that are associated with those processes.

01:11:10.860 | So if you look at schizophrenia,

01:11:12.740 | it's excitatory neuron function,

01:11:14.300 | it's inhibitory neuron function,

01:11:15.620 | it's synaptic pruning, it's calcium signaling,

01:11:17.780 | and so on and so forth.

01:11:18.820 | So when you look at disease genetics,

01:11:22.100 | you have one hit here and one hit there

01:11:24.580 | and one hit there and one hit there,

01:11:26.300 | completely different parts of the genome.

01:11:28.100 | But it turns out all of those hits

01:11:30.100 | are calcium signaling proteins.

01:11:32.340 | - Oh, cool.

01:11:33.180 | - You're like, aha, that means

01:11:35.060 | that calcium signaling is important.

01:11:37.300 | So those people who are focusing on one docus at a time

01:11:39.940 | cannot possibly see that picture.

01:11:42.540 | You have to become a genomicist.

01:11:44.500 | You have to look at the omics, the om,

01:11:46.940 | the holistic picture to understand these enrichments.

01:11:51.220 | - But you mentioned the convergence thing.

01:11:53.500 | So whatever the thing associated

01:11:56.820 | with the disease shows up.

01:11:58.300 | - So let me explain convergence.

01:11:59.700 | Convergence is such a beautiful concept.

01:12:01.780 | (Zubin laughs)

01:12:03.460 | So you basically have these four genes

01:12:07.380 | that are converging on calcium signaling.

01:12:11.340 | So that basically means that they are acting

01:12:15.260 | each in their own way, but together in the same process.

01:12:18.900 | But now in every one of these loci,

01:12:22.540 | you have many enhancers controlling each of those genes.

01:12:27.500 | That's another type of convergence

01:12:29.540 | where dysregulation of seven different enhancers

01:12:32.180 | might all converge on dysregulation of that one gene,

01:12:35.980 | which then converges on calcium signaling.

01:12:39.220 | And in each one of those enhancers,

01:12:41.020 | you might have multiple genetic variants

01:12:43.580 | distributed across many different people.

01:12:46.540 | Everyone has their own different mutation,

01:12:49.820 | but all of these mutations are impacting that enhancer,

01:12:52.860 | and all of these enhancers are impacting that gene,

01:12:55.140 | and all of these genes are impacting this pathway,

01:12:57.540 | and all of these pathways are acting in the same tissue,

01:12:59.980 | and all of these tissues are converging together

01:13:02.420 | on the same biological process of schizophrenia.

01:13:04.700 | - And you're saying the saving grace

01:13:07.820 | is that that convergence seems to happen

01:13:09.660 | for a lot of these diseases.

01:13:11.060 | - For all of them.

01:13:11.900 | Basically that for every single disease

01:13:14.620 | that we've looked at,

01:13:15.500 | we have found an epigenomic enrichment.

01:13:18.380 | How do you do that?

01:13:19.580 | You basically have all of the genetic variants

01:13:22.540 | associated with the disorder,

01:13:24.020 | and then you're asking for all of the enhancers

01:13:26.140 | active in a particular tissue.

01:13:27.940 | For 540 disorders,

01:13:30.860 | we've basically found that indeed there is an enrichment.

01:13:33.740 | That basically means that there is commonality,

01:13:37.020 | and from the commonality, we can just get insights.

01:13:40.620 | So to explain in mathematical terms,

01:13:43.300 | we're basically building an empirical prior.

01:13:47.060 | We're using a Bayesian approach to basically say,

01:13:50.020 | "Great, all of these variants are equally likely

01:13:53.540 | "in a particular locus to be important."

01:13:55.780 | So in a genetic locus,

01:13:58.420 | you basically have a dozen variants that are co-inherited,

01:14:02.660 | because the way that inheritance works in the human genome

01:14:05.460 | is through all of these recombination events during meiosis.

01:14:10.140 | You basically have, you know,

01:14:12.700 | you inherit maybe three, chromosome three, for example,

01:14:16.740 | in your body.

01:14:17.780 | It's inherited from four different parts.

01:14:20.140 | One part comes from your dad,

01:14:21.860 | another part comes from your mom,

01:14:23.060 | another part comes from your dad,

01:14:24.380 | another part comes from your mom.

01:14:25.860 | So basically, the way that it,

01:14:28.060 | sorry, from your mom's mom.

01:14:30.180 | So you basically have one copy that comes from your dad

01:14:33.140 | and one copy that comes from your mom.

01:14:34.540 | But that copy that you got from your mom

01:14:36.580 | is a mixture of her maternal and her paternal chromosome.

01:14:40.980 | And the copy that you got from your dad

01:14:42.260 | is a mixture of his maternal and his paternal chromosome.

01:14:45.620 | So these breakpoints that happen

01:14:48.300 | when chromosomes are lining up

01:14:51.180 | are basically ensuring, through these crossover events,

01:14:54.780 | they're ensuring that every child cell

01:14:59.780 | during the process of meiosis,

01:15:02.540 | where you basically have, you know,

01:15:04.460 | one spermatozoid that basically couples with one ovule

01:15:08.220 | to basically create one egg,

01:15:10.020 | to basically create the zygote.

01:15:12.180 | You basically have half of your genome that comes from dad

01:15:15.780 | and half your genome that comes from mom.

01:15:17.300 | But in order to line them up,

01:15:19.540 | you basically have these crossover events.

01:15:21.100 | These crossover events are basically leading

01:15:23.740 | to co-inheritance of that entire block

01:15:27.700 | coming from your maternal grandmother

01:15:30.260 | and that entire block coming from your maternal grandfather.

01:15:33.780 | Over many generations,

01:15:35.620 | these crossover events don't happen randomly.

01:15:38.780 | There's a protein called PRDM9

01:15:41.460 | that basically guides the double-stranded breaks

01:15:45.140 | and then leads to these crossovers.

01:15:48.180 | And that protein has a particular preference

01:15:50.820 | to only a small number of hotspots of recombination,

01:15:54.180 | which then lead to a small number of breaks

01:15:57.740 | between these co-inheritance patterns.

01:15:59.900 | So even though there are 6 million variants,

01:16:01.820 | there are 6 million loci,

01:16:03.300 | this variation is inherited in blocks.

01:16:09.580 | And every one of these blocks has like two dozen

01:16:11.940 | genetic variants that are all associated.

01:16:14.220 | So in the case of FTO, it wasn't just one variant,

01:16:17.420 | it was 89 common variants

01:16:19.740 | that were all humongously associated with obesity.

01:16:23.180 | Which ones of those is the important one?

01:16:26.780 | Well, if you look at only one locus, you have no idea.

01:16:29.620 | But if you look at many loci,

01:16:32.140 | you basically say, "Aha, all of them are enriching

01:16:36.700 | in the same epigenomic map."

01:16:40.060 | In that particular case, it was mesenchymal stem cells.

01:16:44.140 | So these are the progenitor cells that give rise

01:16:47.100 | to your brown fat and your white fat.

01:16:50.660 | - Progenitor is like the early on developmental stem cells?

01:16:54.020 | - So you start from one zygote,

01:16:56.060 | and that's a totipotent cell type, it can do anything.

01:16:59.100 | You then, that cell divides, divides, divides,

01:17:04.340 | and then every cell division is leading to specialization,

01:17:09.340 | where you now have a mesodermal lineage,

01:17:13.060 | an ectodermal lineage, an endodermal lineage,

01:17:15.620 | that basically leads to different parts of your body.

01:17:19.220 | The ectoderm will basically give rise to your skin.

01:17:22.300 | Ecto means outside, derm is skin.

01:17:25.740 | So ectoderm, but it also gives rise to your neurons

01:17:28.740 | and your whole brain, so that's a lot of ectoderm.

01:17:31.060 | Mesoderm gives rise to your internal organs,

01:17:33.780 | including the vasculature and your muscle

01:17:37.220 | and stuff like that.

01:17:38.300 | So you basically have this progressive differentiation,

01:17:43.060 | and then if you look further, further down that lineage,

01:17:45.940 | you basically have one lineage that will give rise

01:17:47.900 | to both your muscle and your bone, but also your fat.

01:17:51.760 | And if you go further down the lineage of your fat,

01:17:55.840 | you basically have your white fat cells.

01:17:58.900 | These are the cells that store energy.

01:18:01.580 | So when you eat a lot, but you don't exercise too much,

01:18:03.940 | there's an excess set of calories, excess energy.

01:18:07.540 | What do you do with those?

01:18:08.620 | You basically create, you spend a lot of that energy

01:18:11.140 | to create these high-energy molecules, lipids,

01:18:14.660 | which you can then burn when you need them on a rainy day.

01:18:19.660 | So that leads to obesity if you don't exercise

01:18:22.580 | and if you overeat, because your body's like,

01:18:26.420 | oh, great, I have all these calories, I'm gonna store them.

01:18:28.620 | Ooh, more calories, I'm gonna store them too.

01:18:30.420 | Ooh, more calories.

01:18:31.460 | And 42% of European chromosomes

01:18:36.020 | have a predisposition to storing fat,

01:18:40.140 | which was selected probably in the food scarcity periods.

01:18:45.140 | Like basically as we were exiting Africa

01:18:50.140 | before and during the ice ages,

01:18:52.660 | there was probably a selection to those individuals

01:18:54.380 | who made it north to basically be able to store energy,

01:18:58.120 | a lot more energy.

01:19:00.820 | So you basically now have this lineage

01:19:03.860 | that is deciding whether you want to store energy

01:19:07.220 | in your white fat or burn energy in your beige fat.

01:19:11.580 | Turns out that your fat is,

01:19:14.340 | we have such a bad view of fat.

01:19:18.540 | Fat is your best friend.

01:19:19.900 | Fat can both store all these excess lipids

01:19:23.140 | that would be otherwise circulating through your body

01:19:26.500 | and causing damage, but it can also burn calories directly.

01:19:29.900 | If you have too much of energy,

01:19:33.180 | you can just choose to just burn some of that as heat.

01:19:36.220 | So basically when you're cold,

01:19:38.020 | you're burning energy to basically warm your body up

01:19:41.180 | and you're burning all these lipids

01:19:42.340 | and you're burning all these calories.

01:19:44.500 | So what we basically found is that across the board,

01:19:48.740 | genetic variants associated with obesity

01:19:50.540 | across many of these regions were all enriched repeatedly

01:19:54.260 | in mesenchymal stem cell enhancers.

01:19:58.340 | So that gave us a hint as to which of these genetic variants

01:20:02.020 | was likely driving this whole association.

01:20:05.940 | And we ended up with this one genetic variant

01:20:09.340 | called RS1421085.

01:20:13.340 | And that genetic variant out of the 89

01:20:17.340 | was the one that we predicted to be causal for the disease.

01:20:20.780 | So going back to those steps,

01:20:23.180 | first step is figure out the relevant tissue

01:20:25.900 | based on the global enrichment.

01:20:27.740 | Second step is figure out the causal variant

01:20:30.780 | among many variants in this linkage disequilibrium

01:20:34.900 | in this co-inherited block

01:20:37.140 | between these recombination hotspots,

01:20:39.660 | these boundaries of these inherited blocks.

01:20:42.500 | That's the second step.

01:20:43.900 | The third step is once you know that causal variant,

01:20:47.660 | try to figure out what is the motif

01:20:50.020 | that is disrupted by that causal variant.

01:20:52.540 | Basically, how does it act?

01:20:54.140 | Variants don't just disrupt elements,

01:20:56.060 | they disrupt the binding of specific regulators.

01:20:59.500 | So basically the third step there was

01:21:01.020 | how do you find the motif that is responsible,

01:21:04.660 | like the gene regulatory word,

01:21:07.020 | the building block of gene regulation

01:21:09.020 | that is responsible for that dysregulatory event.

01:21:12.420 | And the fourth step is finding out

01:21:13.980 | what regulator normally binds that motif

01:21:17.140 | and is now no longer able to bind.

01:21:19.900 | - And then once you have the regulator,

01:21:21.340 | can you then try to figure out how to,

01:21:23.740 | what, after it developed, how to fix it?

01:21:27.140 | - That's exactly right.

01:21:28.140 | You now know how to intervene.

01:21:30.260 | You have basically a regulator,

01:21:32.140 | you have a gene that you can then perturb.

01:21:34.140 | And you say, well, maybe that regulator

01:21:36.140 | has a global role in obesity.

01:21:38.740 | I can perturb the regulator.

01:21:40.180 | - Just to clarify, when we say perturb,

01:21:43.180 | like on the scale of a human life,

01:21:46.300 | can a human being be helped?

01:21:48.380 | - Of course, of course.

01:21:49.940 | Yeah, so perturb. - I guess understanding

01:21:51.180 | is the first step.

01:21:52.180 | - Exactly.

01:21:53.340 | No, no, but perturbed basically means

01:21:55.020 | you now develop therapeutics,

01:21:56.420 | pharmaceutical therapeutics against that.

01:21:59.140 | Or you develop other types of intervention

01:22:01.140 | that affect the expression of that gene.

01:22:03.660 | - What do pharmaceutical therapeutics look like

01:22:06.780 | when your understanding's on a genetic level?

01:22:11.340 | - Yeah. - Sorry if it's

01:22:12.180 | a dumb question. - No, no, no.

01:22:13.180 | It's a brilliant question,

01:22:14.260 | but I wanna save it for a little bit later

01:22:15.900 | when we start talking about therapeutics.

01:22:17.700 | - Perfect. - We've talked about

01:22:18.620 | the first four steps.

01:22:20.200 | There's two more.

01:22:21.540 | So basically the first step is figure out,

01:22:23.780 | I mean, the zeroth step,

01:22:25.060 | the starting point is the genetics.

01:22:26.740 | The first step after that is figure out

01:22:29.220 | the tissue of action.

01:22:31.060 | The second step is figuring out the nucleotide

01:22:34.300 | that is responsible or set of nucleotides.

01:22:36.900 | The third step is figure out the motif

01:22:38.700 | and the upstream regulator, number four.

01:22:40.800 | Number five and six is what are the targets?

01:22:44.260 | So number five is great.

01:22:45.740 | Now I know the regulator, I know the motif,

01:22:48.260 | I know the tissue, and I know the variant.

01:22:51.380 | What does it actually do?

01:22:53.380 | So you have to now trace it to the biological process

01:22:56.980 | and the genes that mediate that biological process.

01:23:00.420 | So knowing all of this can now allow you

01:23:03.340 | to find the target genes.

01:23:05.300 | How?

01:23:06.260 | By basically doing perturbation experiments

01:23:08.920 | or by looking at the folding of the epigenome

01:23:13.300 | or by looking at the genetic impact

01:23:16.180 | of that genetic variant on the expression of genes.

01:23:19.420 | And we use all three.

01:23:21.500 | So let me go through them.

01:23:22.500 | Basically, one of them is physical links.

01:23:26.300 | This is the folding of the genome onto itself.

01:23:29.780 | How do you even figure out the folding?

01:23:32.100 | It's a little bit of a tangent,

01:23:33.380 | but it's a super awesome technology.

01:23:35.220 | Think of the genome as, again,

01:23:39.120 | this massive packaging that we talked about

01:23:41.440 | of taking two meters worth of DNA

01:23:43.540 | and putting it in something that's a million times smaller

01:23:48.500 | than two meters worth of DNA, that's a single cell.

01:23:51.700 | You basically have this massive packaging,

01:23:53.540 | and this packaging basically leads to the chromosome

01:23:57.060 | being wrapped around in sort of tie-tight ways,

01:24:01.320 | in ways, however, that are functionally capable

01:24:04.100 | of being reopened and reclosed.

01:24:05.940 | So I can then go in and figure out that folding

01:24:11.460 | by sort of chopping up the spaghetti soup,

01:24:14.860 | putting glue, and ligating the segments

01:24:17.900 | that were chopped up but nearby each other,

01:24:20.740 | and then sequencing through these ligation events

01:24:23.660 | to figure out that this region of this chromosome,

01:24:25.940 | that region of the chromosome were near each other,

01:24:28.300 | that means they were interacting,

01:24:30.020 | even though they were far away on the genome itself.

01:24:32.620 | So that chopping up, sequencing, and re-gluing

01:24:37.540 | is basically giving you folds of the genome that we call.

01:24:42.540 | - Sorry, can you backtrack?

01:24:44.580 | - Of course. - How does cutting it

01:24:46.420 | help you figure out which ones were close

01:24:49.220 | in the original folding?

01:24:50.500 | - So you have a bowl of noodles.

01:24:52.440 | - Go on.

01:24:54.460 | - And in that bowl of noodles,

01:24:56.100 | some noodles are near each other.

01:24:59.860 | So throwing a bunch of glue,

01:25:01.560 | you basically freeze the noodles in place,

01:25:05.580 | throwing a cutter that chops up the noodles

01:25:07.500 | into little pieces,

01:25:08.560 | now throwing some ligation enzyme

01:25:13.820 | that lets those pieces that were free

01:25:17.260 | re-ligate near each other.

01:25:19.060 | In some cases, they re-ligate what you had just cut,

01:25:22.740 | but that's very rare.

01:25:24.020 | Most of the time, they will re-ligate

01:25:26.700 | in whatever was proximal.

01:25:28.620 | You now have glued the red noodle

01:25:32.940 | that was crossing the blue noodle to each other.

01:25:35.380 | You then reverse the glue, the glue goes away,

01:25:40.260 | and you just sequence the heck out of it.

01:25:42.860 | Most of the time, you'll find red segment with,

01:25:46.140 | you know, a red segment,

01:25:47.700 | but you can specifically select for ligation events

01:25:50.180 | that have happened that were not from the same segment

01:25:52.500 | by sort of marking them a particular way,

01:25:54.420 | and then selecting those,

01:25:56.540 | and then you sequence and you look for red with blue matches

01:26:00.180 | of sort of things that were glued

01:26:01.860 | that were not immediate proximal to each other,

01:26:04.500 | and that reveals the linking of the blue noodle

01:26:07.780 | and the red noodle.

01:26:08.600 | You're with me so far? - Yeah.

01:26:09.740 | - Good.

01:26:10.580 | So we've done these experiments.

01:26:11.420 | - That's the physical.

01:26:13.340 | - That's the physical.

01:26:14.180 | That's step one of the physical,

01:26:15.660 | and what the physical revealed

01:26:17.180 | is topologically associated domains,

01:26:18.900 | basically big blocks of the genome

01:26:20.820 | that are topologically, you know, connected together.

01:26:23.960 | That's the physical.

01:26:26.260 | The second one is the genetic links.

01:26:29.060 | It basically says, across individuals

01:26:32.660 | that have different genetic variants,

01:26:35.520 | how are their genes expressed differently?

01:26:39.140 | Remember before I was saying

01:26:40.300 | that the path between genetics and disease is enormous,

01:26:43.020 | but we can break it up to look at the path

01:26:44.620 | between genetics and gene expression.

01:26:47.460 | So instead of using Alzheimer's as the phenotype,

01:26:50.220 | I can now use expression of IRX3 as the phenotype,

01:26:54.180 | expression of gene A,

01:26:55.300 | and I can look at all of the humans

01:26:59.380 | who contain a G at that location

01:27:01.260 | and all the humans that contain a T at that location

01:27:03.980 | and basically say, wow,

01:27:05.220 | turns out that the expression of this gene is higher

01:27:07.420 | for the T humans than for the G humans at that location.

01:27:10.580 | So that basically gives me a genetic link

01:27:12.660 | between a genetic variant, a locus, a region,

01:27:16.220 | and the expression of nearby genes.

01:27:18.740 | Good on the genetic link?

01:27:20.980 | - I think so. - Awesome.

01:27:22.240 | So the third link is the activity link.

01:27:25.380 | What's an activity link?

01:27:26.420 | It basically says,

01:27:27.260 | if I look across 833 different epigenomes,

01:27:31.060 | whenever this enhancer is active,

01:27:34.180 | this gene is active.

01:27:35.980 | That gives me an activity link

01:27:37.540 | between this region of the DNA and that gene.

01:27:40.840 | And then the fourth one is perturbations

01:27:43.940 | where I can go in and blow up that region

01:27:46.940 | and see what are the genes that change in expression,

01:27:49.340 | or I can go in and over-activate that region

01:27:51.980 | and see what genes change in expression.

01:27:53.980 | - So I guess that's similar to activity?

01:27:57.620 | - Yeah, yeah, so that's basically similar to activity.

01:28:00.380 | I agree, but it's causal rather than correlational.

01:28:03.180 | - Again, I'm a little weird.

01:28:05.380 | - No, no, you're 100% on.

01:28:07.060 | It's exactly the same as activity,

01:28:08.380 | but perturbation. - But the perturbations.

01:28:09.700 | - Where I go and intervene,

01:28:11.500 | I basically take a bunch of cells.

01:28:13.700 | So you know CRISPR, right?

01:28:15.860 | CRISPR is this genome guidance and cutting mechanism.

01:28:20.860 | It's what George Church likes to call genome vandalism.

01:28:24.640 | So you basically are able to, (laughs)

01:28:28.020 | you can basically take a guide RNA

01:28:31.700 | that you put into the CRISPR system,

01:28:34.380 | and the CRISPR system will basically use this guide RNA,

01:28:36.740 | scan the genome, find wherever there's a match,

01:28:39.380 | and then cut the genome.

01:28:41.420 | So I digress, but it's a bacterial immune defense system.

01:28:47.500 | So basically bacteria are constantly attacked by viruses,

01:28:50.880 | but sometimes they win against the viruses,

01:28:55.300 | and they chop up these viruses,

01:28:56.820 | and remember as a trophy inside their genome,

01:28:59.820 | they have these loci, these CRISPR loci,

01:29:02.740 | that basically stands for clustered,

01:29:04.460 | repeats, interspersed, et cetera.

01:29:06.380 | So basically it's an interspersed repeats structure,

01:29:10.140 | where basically you have a set of repetitive regions,

01:29:13.220 | and then interspersed where these variable segments

01:29:16.760 | that were basically matching viruses.

01:29:19.540 | So when this was first discovered,

01:29:21.820 | it was basically hypothesized

01:29:23.260 | that this is probably a bacterial immune system

01:29:25.560 | that remembers the trophies of the viruses

01:29:27.940 | that manage the kill.

01:29:29.060 | And then the bacteria pass on,

01:29:32.100 | you know, they sort of do lateral transfer of DNA,

01:29:34.460 | and they pass on these memories

01:29:36.340 | so that the next bacterium says,

01:29:37.660 | "Ooh, you killed that guy.

01:29:38.940 | "When that guy shows up again, I will recognize him."

01:29:41.580 | And the CRISPR system was basically evolved

01:29:44.020 | as a bacterial adaptive immune response

01:29:47.300 | to sense foreigners that should not belong,

01:29:50.920 | and to just go and cut their genome.

01:29:53.060 | So it's an RNA guided, RNA cutting enzyme,

01:29:57.480 | or an RNA guided DNA cutting enzyme.

01:30:00.220 | So there's different systems.

01:30:02.140 | Some of them cut DNA, some of them cut RNA,

01:30:04.340 | but all of them remember this sort of viral attack.

01:30:09.340 | So what we have done now as a field is, you know,

01:30:13.660 | through the work of, you know, Jennifer Doudna,

01:30:16.700 | Manuel Carpentier, Feng Zhang, and many others,

01:30:19.340 | is co-opted that system of bacterial immune defense

01:30:24.300 | as a way to cut genomes.

01:30:26.460 | You basically have this guiding system

01:30:30.780 | that allows you to use an RNA guide

01:30:33.380 | to bring enzymes to cut DNA at a particular locus.

01:30:37.660 | - That's so fascinating.

01:30:39.140 | So this is like already a natural mechanism,

01:30:41.900 | a natural tool for cutting that was useful

01:30:45.580 | in this particular context.

01:30:47.260 | And we're like, well, we can use that thing to actually,

01:30:49.940 | it's a nice tool that's already in the body.

01:30:51.860 | - Yeah, yeah.

01:30:52.700 | It's not in our body, it's in the bacterial body.

01:30:55.380 | It was discovered by the yogurt industry.

01:30:58.300 | They were trying to make better yogurts,

01:31:01.660 | and they were trying to make their bacteria

01:31:03.560 | in their yogurt cultures more resilient to viruses.

01:31:06.800 | And they were studying bacteria,

01:31:10.060 | and they found that, wow, this CRISPR system is awesome.

01:31:12.420 | It allows you to defend against that.

01:31:14.700 | And then it was co-opted in mammalian systems

01:31:16.920 | that don't use anything like that as a targeting way

01:31:21.380 | to basically bring these DNA-cutting enzymes

01:31:23.980 | to any locus in the genome.

01:31:25.740 | Why would you want to cut DNA to do anything?

01:31:29.580 | The reason is that our DNA has a DNA repair mechanism,

01:31:33.860 | where if a region of the genome gets randomly cut,

01:31:36.580 | you will basically scan the genome for anything that matches

01:31:40.180 | and sort of use it by homology.

01:31:43.420 | So the reason why we're diploid

01:31:45.140 | is because we now have a spare copy.

01:31:47.200 | As soon as my mom's copy is deactivated,

01:31:49.220 | I can use my dad's copy.

01:31:50.580 | And somewhere else, if my dad's copy is deactivated,

01:31:52.540 | I can use my mom's copy to repair it.

01:31:55.180 | So this is called homologous-based repair.

01:31:58.640 | - So all you have to do is the cutting,

01:32:02.020 | and you don't have to do the fixing.

01:32:04.020 | - That's exactly right.

01:32:04.860 | You don't have to do the fixing.

01:32:06.140 | - 'Cause it's already built in.

01:32:07.220 | - That's exactly right.

01:32:08.400 | But the fixing can be co-opted

01:32:10.780 | by throwing in a bunch of homologous segments

01:32:14.340 | that instead of having your dad's version,

01:32:16.780 | have whatever other version you'd like to use.

01:32:20.540 | - So you then control the fixing

01:32:22.460 | by throwing in a bunch of other stuff.

01:32:23.980 | - That's exactly right.

01:32:24.820 | And that's how you do genome editing.

01:32:26.380 | - So that's what CRISPR is.

01:32:27.780 | - That's what CRISPR is.

01:32:28.620 | - In popular culture, people use the term.

01:32:30.900 | I've never, wow, that's brilliant.

01:32:32.540 | That's just an awesome explanation.

01:32:34.340 | - Genome vandalism followed by a bunch of Band-Aids

01:32:38.420 | that have the sequence that you'd like.

01:32:39.900 | - And you can control the choices of Band-Aids.

01:32:42.740 | - Correct.

01:32:43.660 | And of course, there's new generations of CRISPR.

01:32:46.300 | There's something that's called prime editing

01:32:48.400 | that was sort of very, very much in the press recently,

01:32:51.860 | that basically instead of sort of making

01:32:53.160 | a double-stranded break, which again is genome vandalism,

01:32:57.240 | you basically make a single-stranded break.

01:33:00.800 | You basically just nick one of the two strands,

01:33:03.560 | enabling you to sort of peel off

01:33:06.180 | without sort of completely breaking it up,

01:33:08.740 | and then repair it locally using a guide

01:33:11.820 | that is coupled to your initial RNA

01:33:16.000 | that took you to that location.

01:33:18.440 | - Dumb question, but is CRISPR as awesome

01:33:22.200 | and cool as it sounds?

01:33:23.900 | I mean, technically speaking, in terms of like,

01:33:27.760 | as a tool for manipulating our genetics

01:33:31.840 | in the positive meaning of the word manipulating,

01:33:35.920 | or is there downsides, drawbacks,

01:33:38.680 | in this whole context of therapeutics

01:33:40.360 | that we're talking about, or understanding and so on?

01:33:42.840 | - So when I teach my students about CRISPR,

01:33:46.580 | I show them articles with the headline,

01:33:49.480 | Genome Editing Tool Revolutionizes Biology,

01:33:53.000 | and then I show them the date of these articles,

01:33:55.440 | and they're 2004, like five years

01:33:57.920 | before CRISPR was invented.

01:33:59.320 | And the reason is that they're not talking about CRISPR.

01:34:02.260 | They're talking about zinc finger enzymes

01:34:04.680 | that are another way to bring these cutters to the genome.

01:34:08.880 | It's a very difficult way of sort of designing

01:34:12.040 | the right set of zinc finger proteins,

01:34:13.800 | the right set of amino acids

01:34:15.480 | that will now target a particular long stretch of DNA,

01:34:19.160 | because for every location that you want to target,

01:34:22.160 | you need to design a particular regulator,

01:34:25.360 | a particular protein that will match that region well.

01:34:29.200 | There's another technology called talons,

01:34:31.760 | which are basically just a different way of using proteins

01:34:36.760 | to sort of guide these cutters

01:34:39.300 | to a particular location in the genome.

01:34:41.140 | These require a massive team of engineers,

01:34:44.620 | of biological engineers,

01:34:45.680 | to basically design a set of amino acids

01:34:48.100 | that will target a particular sequence of your genome.

01:34:51.420 | The reason why CRISPR is amazingly, awesomely revolutionary

01:34:55.720 | is because instead of having this team of engineers

01:34:58.760 | design a new set of proteins

01:35:00.640 | for every locus that you want to target,

01:35:02.560 | you just type it in your computer,

01:35:04.560 | and you just synthesize an RNA guide.

01:35:06.740 | The beauty of CRISPR is not the cutting.

01:35:10.000 | It's not the fixing.

01:35:11.000 | All of that was there before.

01:35:12.800 | It's the guiding, and the only thing that changes

01:35:15.860 | is that it makes the guiding easier

01:35:17.980 | by sort of just typing in the RNA sequence,

01:35:21.940 | which then allows the system to sort of scan the DNA

01:35:24.860 | to find that.

01:35:25.860 | - So the coding, the engineering of the cutter is easier

01:35:29.740 | in terms of SV. - Exactly.

01:35:32.180 | - That's kind of similar to the story of deep learning

01:35:34.260 | versus old school machine learning,

01:35:36.660 | is some of the challenging parts are automated.

01:35:39.540 | Okay, so, but CRISPR's just one cutting technology.

01:35:44.540 | And then there's, that's part of the challenges

01:35:47.120 | and exciting opportunities of the field

01:35:49.600 | is to design different cutting technologies.

01:35:52.960 | - So now, this was a big parenthesis on CRISPR,

01:35:56.080 | but now, when we were talking about perturbations,

01:36:00.760 | you basically now have the ability

01:36:02.380 | to not just look at correlation between enhancers and genes,

01:36:05.800 | but actually go and either destroy that enhancer

01:36:09.900 | and see if the gene changes in expression,

01:36:11.800 | or you can use the CRISPR targeting system

01:36:15.260 | to bring in not vandalism and cutting,

01:36:19.860 | but you can couple the CRISPR system with,

01:36:24.500 | and the CRISPR system is called usually CRISPR-Cas9

01:36:27.500 | because Cas9 is the protein that will then come and cut.

01:36:30.760 | But there's a version of that protein called dead Cas9

01:36:34.580 | where the cutting part is deactivated.

01:36:36.700 | So you basically use dCas9, dead Cas9,

01:36:39.640 | to bring in an activator or to bring in a repressor.

01:36:44.640 | So you can now ask, is this enhancer changing that gene?

01:36:48.640 | By taking this modified CRISPR,

01:36:51.860 | which is already modified from the bacteria

01:36:53.580 | to be used in humans,

01:36:54.740 | that you can now modify the Cas9 to be dead Cas9,

01:36:57.700 | and you can now further modify to bring in a regulator,

01:37:00.960 | and you can basically turn on or turn off that enhancer

01:37:03.980 | and then see what is the impact on that gene.

01:37:06.580 | So these are the four ways of linking the locus

01:37:10.100 | to the target gene, and that's step number five.

01:37:12.700 | Step number five is find the target gene,

01:37:16.100 | and step number six is what the heck does that gene do?

01:37:19.540 | You basically now go and manipulate that gene

01:37:22.180 | to basically see what are the processes that change.

01:37:26.620 | And you can basically ask, well, in this particular case,

01:37:31.060 | in the FTO locus, we found mesenchymal stem cells

01:37:34.660 | that are the progenitors of white fat and brown fat,

01:37:38.100 | or beige fat.

01:37:39.500 | We found the RS1421085 nucleotide variant

01:37:43.020 | as the causal variant.

01:37:44.820 | We found this large enhancer, this master regulator.

01:37:49.820 | I like to call it OB1 for obesity one,

01:37:53.300 | like the strongest enhancer associated with it.

01:37:55.620 | And OB1 was kind of chubby as the actor,

01:37:57.180 | I don't know if you remember him.

01:37:58.500 | (laughing)

01:38:00.420 | - Yeah.

01:38:01.260 | - So you basically are using this Jedi mind trick

01:38:02.980 | to basically find out the--

01:38:04.580 | - Thank you.

01:38:05.420 | (laughing)

01:38:06.300 | - The location of the genome that is responsible,

01:38:09.100 | the enhancer that harbors it, the motif,

01:38:12.540 | the upstream regulator, which is ARID5B,

01:38:15.540 | for AT-rich interacting domain 5B.

01:38:18.100 | That's a protein that sort of comes and binds normally.

01:38:20.960 | That protein is normally a repressor.

01:38:23.100 | It represses this super enhancer,

01:38:25.060 | this massive 12,000 nucleotide

01:38:26.900 | master regulatory control region,

01:38:28.740 | and it turns off IRX3,

01:38:32.500 | which is a gene that's 600,000 nucleotides away,

01:38:34.940 | and IRX5, which is 1.2 million nucleotides away.

01:38:38.340 | So those--

01:38:39.180 | - And what's the effect of turning them off?

01:38:40.620 | - That's exactly the next question.

01:38:42.260 | So step six is, what do these genes actually do?

01:38:45.420 | So we then ask, what does IRX3 and IRX5 do?

01:38:48.580 | The first thing we did is look across individuals

01:38:50.940 | for individuals that had higher expression of IRX3

01:38:53.740 | or lower expression of IRX3.

01:38:55.380 | And then we looked at the expression

01:38:56.700 | of all of the other genes in the genome.

01:38:58.860 | And we looked for simply correlation.

01:39:01.500 | And we found that IRX3 and IRX5

01:39:03.500 | were both correlated positively with lipid metabolism

01:39:08.340 | and negatively with mitochondrial biogenesis.

01:39:12.680 | You're like, what the heck does that mean?

01:39:16.300 | - Doesn't sound related to obesity.

01:39:17.940 | - Not at all, superficially.

01:39:20.320 | But lipid metabolism should,

01:39:22.840 | because lipids is these high energy molecules

01:39:26.620 | that basically store fat.

01:39:28.420 | So IRX3 and IRX5 are negatively correlated

01:39:32.380 | with lipid metabolism.

01:39:33.700 | So that basically means that when they turn on,

01:39:36.100 | lipid metabolism, or positively,

01:39:37.740 | when they turn on, they turn on lipid metabolism.

01:39:41.020 | And they're negatively correlated

01:39:42.460 | with mitochondrial biogenesis.

01:39:45.820 | What do mitochondria do in this whole process?

01:39:49.220 | Again, small parenthesis, what are mitochondria?

01:39:51.740 | Mitochondria are little organelles.

01:39:56.140 | They arose, they only are found in eukaryotes.

01:40:00.860 | Euk means good, karyote means nucleus.

01:40:03.900 | So truly, like a true nucleus.

01:40:05.820 | So eukaryotes have a nucleus.

01:40:07.740 | Prokaryotes are before the nucleus.

01:40:09.900 | They don't have a nucleus.

01:40:11.200 | So eukaryotes have a nucleus.

01:40:13.500 | Hmm, compartmentalization.

01:40:15.180 | Eukaryotes have also organelles.

01:40:18.620 | Some eukaryotes have chloroplasts.

01:40:22.700 | These are the plants, they photosynthesize.

01:40:25.860 | Some other eukaryotes, like us,

01:40:28.580 | have another type of organelle called mitochondria.

01:40:33.140 | These arose from an ancient species that we engulfed.

01:40:38.140 | This is an endosymbiosis event.

01:40:43.660 | Symbiosis, bio means life, sym means together.

01:40:47.260 | So symbiotes are things that live together.

01:40:49.860 | Endosymbiosis, endo means inside,

01:40:51.980 | so endosymbiosis means you live together,

01:40:53.780 | holding the other one inside you.

01:40:55.980 | So the pre-eukaryotes engulfed an organism

01:41:00.980 | that was very good at energy production,

01:41:05.620 | and that organism eventually shed most of its genome

01:41:11.140 | to now have only 13 genes in the mitochondrial genome.

01:41:14.740 | And those 13 genes are all involved in energy production,

01:41:21.060 | the electron transport chain.

01:41:23.340 | So basically, electrons are these massive,

01:41:25.620 | super energy-rich molecules.

01:41:27.280 | We basically have these organelles that produce energy,

01:41:33.340 | and when your muscle exercises,

01:41:35.700 | you basically multiply your mitochondria.

01:41:37.780 | You basically sort of use more and more mitochondria,

01:41:41.900 | and that's how you get beefed up.

01:41:43.740 | So basically, the muscle sort of learns

01:41:45.740 | how to generate more energy.

01:41:47.820 | So basically, every single time,

01:41:49.180 | your muscles will, you know, overnight regenerate

01:41:51.580 | and sort of become stronger

01:41:52.460 | and amplify their mitochondria and so forth.

01:41:55.180 | So what do the mitochondria do?

01:41:56.460 | The mitochondria use energy to sort of do any kind of task.

01:42:01.460 | When you're thinking, you're using energy.

01:42:04.920 | This energy comes from mitochondria.

01:42:06.740 | Your neurons have mitochondria all over the place.

01:42:09.800 | Basically, this mitochondria can multiply its organelles,

01:42:12.180 | and they can be spread along the body of your muscle.

01:42:14.900 | Some of your muscle cells have actually multiple nuclei.

01:42:17.340 | They're polynucleated,

01:42:18.500 | but they also have multiple mitochondria

01:42:20.380 | to basically deal with the fact

01:42:22.580 | that your muscle is enormous.

01:42:24.340 | You can sort of span this super, super long length,

01:42:26.780 | and you need energy throughout the length of your muscle.

01:42:29.320 | So that's why you have mitochondria throughout the length,

01:42:31.420 | and you also need transcription through the length,

01:42:32.820 | so you have multiple nuclei as well.

01:42:35.000 | So these two processes, lipids store energy.

01:42:40.000 | What do mitochondria do?

01:42:42.020 | So there's a process known as thermogenesis,

01:42:45.660 | thermal heat genesis generation.

01:42:48.020 | Thermogenesis is the generation of heat.

01:42:50.380 | Remember that bathtub with in and out?

01:42:55.060 | That's the equation that everybody's focused on.

01:42:57.100 | So how much energy do you consume?

01:42:58.800 | How much energy do you burn?

01:43:00.920 | But in every thermodynamic system,

01:43:03.280 | there's three parts to the equation.

01:43:05.940 | There's energy in, energy out, and energy lost.

01:43:10.780 | Any machine has loss of energy.

01:43:14.580 | How do you lose energy?

01:43:15.660 | You emanate heat.

01:43:17.540 | So heat is energy loss.

01:43:18.980 | So there's--

01:43:24.740 | - Which is where the thermogenesis comes in.

01:43:26.660 | - Thermogenesis is actually a regulatory process

01:43:30.140 | that modulates the third component

01:43:32.100 | of the thermodynamic equation.

01:43:33.940 | You can basically control thermogenesis explicitly.

01:43:37.180 | You can turn on and turn off thermogenesis.

01:43:39.340 | - And that's where the mitochondria comes into play.

01:43:40.980 | - Exactly.

01:43:41.820 | So RX3 and RX5 turn out to be the master regulators

01:43:46.280 | of a process of thermogenesis

01:43:49.060 | versus lipogenesis, generation of fat.

01:43:52.300 | So RX3 and RX5 in most people burn heat,

01:43:56.940 | burn calories as heat.

01:43:58.660 | So when you eat too much,

01:43:59.820 | just burn it off in your fat cells.

01:44:02.660 | So that bathtub has basically a sort of dissipation knob

01:44:07.660 | that most people are able to turn on.

01:44:11.200 | I am unable to turn that on

01:44:13.880 | because I am a homozygous carrier

01:44:16.340 | for the mutation that changes a T into a C

01:44:20.420 | in the RS1421085 allele, a locus, a SNP.

01:44:24.620 | I have the risk allele twice,

01:44:26.940 | from my mom and from my dad.

01:44:28.240 | So I'm unable to thermogenize.

01:44:30.760 | I'm unable to turn on thermogenesis through RX3 and RX5

01:44:35.220 | because the regulator that normally binds here,

01:44:38.100 | RX5B, can no longer bind

01:44:39.780 | because it's an AT-rich interacting domain.

01:44:42.620 | And as soon as I change the T into a C,

01:44:44.720 | it can no longer bind because it's no longer AT-rich.

01:44:47.940 | - But doesn't that mean

01:44:48.780 | that you're able to use the energy more efficiently?

01:44:51.700 | You're not generating heat?

01:44:53.580 | Or is that--

01:44:54.420 | - That means I can eat less and get around just fine.

01:44:56.940 | - Yes.

01:44:57.780 | - Yeah, so--

01:44:58.620 | - That's a feature, actually.

01:44:59.820 | - It's a feature in a food-scarce environment.

01:45:02.140 | - Yeah, but--

01:45:02.980 | - If we're all starving, I'm doing great.

01:45:05.080 | If we all have access to massive amounts of food,

01:45:07.140 | I'm obese, basically.

01:45:08.980 | - That's taken us through the entire process

01:45:11.040 | of then understanding why mitochondria and then the lipids

01:45:15.540 | are both, even though distant, are somehow involved.

01:45:18.420 | - Different sides of the same coin.

01:45:20.580 | You basically choose to store energy

01:45:22.460 | or you can choose to burn energy.

01:45:23.860 | - And that all of that is involved in the puzzle of obesity.

01:45:27.640 | - And that's what's fascinating, right?

01:45:29.580 | Here we are in 2007,

01:45:31.460 | discovering the strongest genetic association with obesity

01:45:34.580 | and knowing nothing about how it works for almost 10 years.

01:45:39.360 | For 10 years, everybody focused on this FTO gene.

01:45:42.660 | And they were like, "Oh, it must have to do something

01:45:44.620 | "with RNA modification."

01:45:48.140 | And it's like, "No, it has nothing to do

01:45:49.280 | "with the function of FTO.

01:45:50.580 | "It has everything to do with all of this other process."

01:45:53.700 | And suddenly, the moment you solve that puzzle,

01:45:56.420 | which is a multi-year effort, by the way,

01:45:58.380 | a tremendous effort by Melina and many, many others.

01:46:01.780 | So this tremendous effort basically led us

01:46:04.180 | to recognize this circuitry.

01:46:07.100 | You went from having some 89 common variants associated

01:46:10.860 | in that region of the DNA, sitting on top of this gene,

01:46:13.800 | to knowing the whole circuitry.

01:46:16.040 | When you know the circuitry, you can now go crazy.

01:46:21.060 | You can now start intervening at every level.

01:46:24.420 | You can start intervening at the RX5B level.

01:46:27.180 | You can start intervening with CRISPR-Cas9

01:46:28.980 | at the single SNP level.

01:46:31.200 | You can start intervening at RX3 and RX5, directly there.

01:46:34.760 | You can start intervening at the thermogenesis level

01:46:36.900 | because you know the pathway.

01:46:38.340 | You can start intervening at the differentiation level,

01:46:41.500 | where the decision to make either white fat or beige fat,

01:46:46.500 | the energy-burning beige fat,

01:46:47.980 | is made developmentally in the first three days

01:46:51.900 | of differentiation of your adipocytes.

01:46:53.940 | So as they're differentiating, you basically can choose

01:46:56.220 | to make fat-burning machines or fat-storing machines,

01:46:59.180 | and sort of that's how you populate your fat.

01:47:02.180 | You basically can now go in pharmaceutically

01:47:04.460 | and do all of that.

01:47:05.560 | And in our paper, we actually did all of that.

01:47:09.300 | We went in and manipulated every single aspect.

01:47:12.220 | At the nucleotide level,

01:47:13.620 | we use CRISPR-Cas9 genome editing

01:47:16.080 | to basically take primary adipocytes

01:47:18.220 | from risk and non-risk individuals,

01:47:20.660 | and show that by editing that one nucleotide

01:47:23.720 | out of 3.2 billion nucleotides in the human genome,

01:47:26.580 | you could then flip between an obese phenotype

01:47:29.660 | and a lean phenotype like a switch.

01:47:31.420 | You can basically take micelles that are non-thermogenizing

01:47:34.900 | and just flip into thermogenizing cells

01:47:36.580 | by changing one nucleotide.

01:47:38.540 | It's mind-boggling.

01:47:39.900 | - It's so inspiring that this puzzle

01:47:41.920 | could be solved in this way,

01:47:43.140 | and it feels within reach to then be able

01:47:46.020 | to crack the problem of some of these diseases.

01:47:50.380 | What are, so it's 2007 you mentioned,

01:47:54.260 | what are the technologies, the tools that came along

01:47:58.100 | that made this possible?

01:47:59.780 | Like what are you excited about,

01:48:01.780 | maybe if we just look at the buffet of things

01:48:03.620 | that you've kind of mentioned.

01:48:05.340 | Is there, what's involved, what should we be excited about,

01:48:09.460 | what are you excited about?

01:48:11.540 | - I love that question because there's so much ahead of us.

01:48:13.940 | There's so, so much.

01:48:15.180 | There's, so basically solving that one locus

01:48:20.980 | required massive amounts of knowledge

01:48:23.740 | that we have been building across the years

01:48:25.500 | through the epigenome, through the comparative genomics

01:48:28.320 | to find out the causal variant

01:48:29.820 | and the controller regulatory motif

01:48:33.380 | through the conserved circuitry.

01:48:35.260 | It required knowing this regulatory genomic wiring.

01:48:38.500 | It required high C of these sort of topologically

01:48:41.500 | associated domains to basically find

01:48:42.940 | this long range interaction.

01:48:44.500 | It required EQTLs of this sort of genetic perturbation

01:48:48.460 | of these intermediate gene phenotypes.

01:48:51.060 | It required all of the arsenal of tools

01:48:53.200 | that I've been describing was put together for one locus.

01:48:57.060 | And this was a massive team effort,

01:48:59.380 | huge investment in time, energy, money,

01:49:03.980 | effort, intellectual, everything.

01:49:06.180 | - You're referring to, I'm sorry,

01:49:08.380 | just for the obesity one. - This one paper, yeah.

01:49:09.860 | This one paper. - This one single paper.

01:49:11.460 | - This one single locus.

01:49:12.500 | I like to say that this is a paper

01:49:14.340 | about one nucleotide in the human genome,

01:49:16.460 | about one bit of information,

01:49:17.940 | C versus T in the human genome.

01:49:20.400 | That's one bit of information

01:49:21.660 | and we have 3.2 billion nucleotides to go through.

01:49:25.240 | So how do you do that systematically?

01:49:29.240 | - I am so excited about the next phase of research

01:49:32.320 | because the technologies that my group

01:49:34.940 | and many other groups have developed

01:49:36.540 | allows us to now do this systematically,

01:49:39.040 | not just one locus at a time,

01:49:41.540 | but thousands of loci at a time.

01:49:44.980 | So let me describe some of these technologies.

01:49:47.920 | The first one is automation and robotics.

01:49:52.300 | So basically, we talked about how you can take

01:49:56.420 | all of these molecules and see

01:49:58.240 | which of these molecules are targeting each of these genes

01:50:00.660 | and what do they do.

01:50:02.100 | So you can basically now screen

01:50:03.420 | through millions of molecules,

01:50:06.300 | through thousands and thousands and thousands of plates,

01:50:09.140 | each of which has thousands and thousands

01:50:10.820 | and thousands of molecules,

01:50:12.700 | every single time testing all of these genes

01:50:17.500 | and asking which of these molecules perturb these genes.

01:50:21.960 | So that's technology number one, automation and robotics.

01:50:25.420 | Technology number two is parallel readouts.

01:50:29.200 | So instead of perturbing one locus

01:50:31.080 | and then asking if I use CRISPR-Cas9 on this enhancer

01:50:35.920 | to basically use dCas9 to turn on or turn off the enhancer,

01:50:39.720 | or if I use CRISPR-Cas9 on the SNP

01:50:42.140 | to basically change that one SNP at a time,

01:50:44.780 | then what happens?

01:50:46.520 | But we have 120,000 disease-associated SNPs

01:50:51.160 | that we wanna test.

01:50:52.360 | We don't wanna spend 120,000 years doing it.

01:50:55.620 | So what do we do?

01:50:58.820 | We basically develop this technology

01:51:01.300 | for massively parallel reporter assays, MPRA.

01:51:06.300 | So in collaboration with Tarjan Michelson, Eric Lander,

01:51:09.980 | I mean Jason Durey's group has done a lot of that.

01:51:12.020 | So there's a lot of groups

01:51:13.460 | that basically have developed technologies

01:51:15.980 | for testing 10,000 genetic variants at a time.

01:51:21.340 | How do you do that?

01:51:22.280 | We talked about microarray technology,

01:51:26.040 | the ability to synthesize these huge microarrays

01:51:29.120 | that allow you to do all kinds of things

01:51:30.720 | like measure gene expression by hybridization,

01:51:33.740 | by measuring the genotype of a person,

01:51:36.260 | by looking at hybridization with one version with a T

01:51:38.680 | versus the other version with a C,

01:51:41.560 | and then sort of figuring out

01:51:42.640 | that I am a risk carrier for obesity

01:51:45.560 | based on these hybridization,

01:51:47.380 | differential hybridization in my genome that says,

01:51:49.240 | oh, you seem to only have this allele

01:51:51.120 | or you seem to have that allele.

01:51:52.800 | Microarrays can also be used

01:51:54.340 | to systematically synthesize small fragments of DNA.

01:51:59.340 | So you can basically synthesize

01:52:01.360 | these 150 nucleotide long fragments

01:52:03.840 | across 450,000 spots at a time.

01:52:08.380 | You can now take the result of that synthesis,

01:52:14.120 | which basically works through all of these sort of layers

01:52:16.760 | of adding one nucleotide at a time.

01:52:18.680 | You can basically just type it into your computer

01:52:20.600 | and order it, and you can basically order 10,000

01:52:25.120 | or 100,000 of these small DNA segments at a time.

01:52:30.560 | And that's where awesome molecular biology comes in.

01:52:33.360 | You can basically take all these segments,

01:52:35.280 | have a common start and end barcode or sort of ligator,

01:52:39.960 | like just like pieces of a puzzle,

01:52:42.040 | you can make the same end piece

01:52:44.600 | and the same start piece for all of them.

01:52:47.920 | And you can now use plasmids,

01:52:50.760 | which are these extra chromosomal,

01:52:53.280 | small DNA circular segments

01:52:56.560 | that are basically inhabiting all our genomes.

01:53:00.520 | We basically have plasmids floating around.

01:53:03.040 | I mean, bacteria use plasmids for transferring DNA,

01:53:06.840 | and that's where they put a lot

01:53:07.940 | of antibiotic resistance genes.

01:53:10.640 | So they can easily transfer them

01:53:12.080 | from one bacterium to the other.

01:53:14.120 | So one bacterium evolves a gene to be resistant

01:53:17.040 | to a particular antibiotic.

01:53:19.680 | It basically says to all its friends,

01:53:20.880 | "Hey, here's that sort of DNA piece."

01:53:24.680 | We can now co-opt these plasmids into human cells.

01:53:28.400 | You can basically make a human cell culture

01:53:30.900 | and add plasmids to that human cell culture

01:53:34.080 | that contain the things that you want to test.

01:53:38.080 | You now have this library of 450,000 elements.

01:53:41.240 | You can insert them each into the common plasmid

01:53:45.280 | and then test them in millions of cells in parallel.

01:53:48.280 | - And the common plasmid is all the same

01:53:50.160 | before you add any-- - Exactly.

01:53:51.520 | The rest of the plasmid is the same.

01:53:53.220 | So it's called an epizomal reporter assay.

01:53:57.240 | Epizome means not inside the genome.

01:53:59.640 | It's sort of outside the chromosomes.

01:54:01.520 | So it's an epizomal assay

01:54:03.860 | that allows you to have a variable region

01:54:05.720 | where you basically test 10,000 different enhancers,

01:54:09.200 | and you have a common region

01:54:10.760 | which basically has the same reporter gene.

01:54:13.680 | You now can do some very cool molecular biology.

01:54:16.560 | You can basically take the 450,000 elements

01:54:19.360 | that you've generated,

01:54:20.760 | and you have a piece of the puzzle here,

01:54:22.720 | piece of the puzzle here, which is identical,

01:54:24.360 | so they're compatible with that plasmid.

01:54:27.020 | You can chop them up in the middle

01:54:28.740 | to separate a barcode reporter from the enhancer,

01:54:32.600 | and in the middle, put the same gene,

01:54:34.360 | again, using the same pieces of the puzzle.

01:54:36.840 | You now can have a barcode readout

01:54:39.860 | of what is the impact of 10,000 different versions

01:54:43.080 | of an enhancer on gene expression.

01:54:45.560 | So we're not doing one experiment.

01:54:47.880 | We're doing 10,000 experiments.

01:54:49.560 | And those 10,000 can be 5,000 of different loci,

01:54:55.480 | and each of them in two versions, risk or non-risk.

01:55:00.360 | I can now test tens of thousands--

01:55:02.040 | - These are little hypotheses.

01:55:03.600 | - Exactly.

01:55:04.440 | - And then you can do 10,000, and wait--

01:55:06.840 | - You can test 10,000 hypotheses at once.

01:55:08.880 | - How hard is it to generate those 10,000?

01:55:12.240 | - Trivial, trivial.

01:55:13.800 | - But it's biology.

01:55:14.920 | - No, no, generating the 10,000 is trivial

01:55:16.880 | because you basically add, it's by technology.

01:55:20.680 | You basically have these arrays

01:55:22.720 | that add one nucleotide at a time at every spot.

01:55:26.480 | - So it's printing, and so you're able to control.

01:55:29.880 | - Yeah.

01:55:31.400 | - Super costly, is it?

01:55:33.160 | - 10,000 bucks.

01:55:34.360 | - So this is in millions?

01:55:35.800 | - 10,000 bucks for 10,000 experiments.

01:55:37.360 | Sounds like the right, you know.

01:55:39.640 | - I mean, so that's super, that's exciting,

01:55:42.040 | 'cause you don't have to do one thing at a time.

01:55:44.000 | - You can now use that technology,

01:55:45.440 | these massively parallel reporter assays,

01:55:47.320 | to test 10,000 locations at a time.

01:55:49.880 | We've made multiple modifications to that technology.

01:55:54.880 | One was SHARPER MPRA, which stands for, you know,

01:55:59.880 | basically getting a higher resolution view

01:56:04.600 | by tiling these elements.

01:56:09.480 | So you can see where along the region of control

01:56:14.480 | are they acting.

01:56:16.000 | And we made another modification called HYDRA

01:56:18.440 | for high, you know, definition, regulatory annotation

01:56:23.440 | or something like that,

01:56:25.360 | which basically allows you to test 7 million of these

01:56:29.880 | at a time by sort of cutting them directly from the DNA.

01:56:32.800 | So instead of synthesizing,

01:56:34.320 | which basically has the limit of 450,000

01:56:36.720 | that you can synthesize at a time,

01:56:38.240 | we basically said, "Hey, if we wanna test

01:56:40.080 | all accessible regions of the genome,

01:56:42.440 | let's just do an experiment that cuts accessible regions.

01:56:45.400 | Let's take those accessible regions,

01:56:47.640 | put them all with the same end joints of the puzzles,

01:56:51.360 | and then now use those to create a much, much larger array

01:56:56.360 | of things that you can test.

01:56:59.360 | And then tiling all of these regions,

01:57:01.200 | you can then pinpoint what are the driver nucleotides,

01:57:04.000 | what are the elements, how are they acting

01:57:05.760 | across 7 million experiments at a time.

01:57:07.400 | So basically, this is all the same family of technology

01:57:11.080 | where you're basically using these parallel readouts

01:57:14.040 | of the barcodes.

01:57:15.680 | And then, you know, to do this,

01:57:18.000 | we used a technology called StarSeq

01:57:20.240 | for self-transcribing reporter assays,

01:57:23.560 | a technology developed by Alex Stark,

01:57:25.600 | my former postdoc, who's now a PI over in Vienna.

01:57:30.000 | So we basically coupled the StarSeq,

01:57:32.680 | the self-transcribing reporters,

01:57:35.520 | where the enhancer can be part of the gene itself.

01:57:39.280 | So instead of having a separate barcode,

01:57:40.920 | that enhancer basically acts to turn on the gene

01:57:43.520 | and is transcribed as part of the gene.

01:57:45.960 | - So you don't have to have the two separate parts.

01:57:47.320 | - Exactly, so you can just read them directly.

01:57:49.000 | - So there's a constant improvement in this whole process.

01:57:52.560 | By the way, generating all these options,

01:57:54.840 | is it basically brute force?

01:57:56.920 | How much human intuition is--

01:57:58.560 | - Oh gosh, of course it's human intuition

01:58:01.000 | and human creativity and incorporating

01:58:03.160 | all of the input data sets,

01:58:05.760 | because again, the genome is enormous, 3.2 billion.

01:58:09.200 | You don't wanna test that.

01:58:10.520 | Instead, you basically use all of these tools

01:58:12.520 | that I've talked about already.

01:58:14.120 | You generate your top favorite 10,000 hypotheses,

01:58:18.080 | and then you go and test all 10,000.

01:58:19.760 | And then from what comes out,

01:58:21.080 | you can then go to the next step.

01:58:22.720 | So that's technology number two.

01:58:25.760 | So technology number one is robotics, automation,

01:58:28.720 | where you have thousands of wells

01:58:30.200 | and you constantly test them.

01:58:31.880 | The second technology is instead of having wells,

01:58:34.680 | you have these massively parallel readouts

01:58:37.320 | in sort of these pooled asses.

01:58:39.920 | The third technology is coupling CRISPR perturbations

01:58:44.680 | with these single-cell RNA readouts.

01:58:50.680 | So let me make another parenthesis here

01:58:53.400 | to describe now single-cell RNA sequencing.

01:58:56.480 | So what does single-cell RNA sequencing mean?

01:58:59.560 | So RNA sequencing is what has been traditionally used,

01:59:04.560 | or well, traditionally, the last 20 years,

01:59:07.600 | ever since the advent of next-generation sequencing.

01:59:10.120 | So basically, before, RNA expression profiling

01:59:12.840 | was based on these microarrays.

01:59:14.560 | The next technology after that was based on sequencing.

01:59:17.400 | So you chop up your RNA

01:59:18.960 | and you just sequence small molecules,

01:59:21.440 | just like you would sequence a genome,

01:59:23.720 | basically reverse transcribe the small RNAs into DNA,

01:59:27.040 | and you sequence that DNA

01:59:29.000 | in order to get the number of sequencing reads

01:59:32.920 | corresponding to the expression level

01:59:35.600 | of every gene in the genome.

01:59:37.400 | You now have RNA sequencing.

01:59:39.520 | How do you go to single-cell RNA sequencing?

01:59:42.240 | That technology also went through stages of evolution.

01:59:45.760 | The first was microfluidics.

01:59:48.000 | You basically had these, or even chambers,

01:59:51.280 | you basically had these ways of isolating individual cells,

01:59:54.160 | putting them into a well for every one of these cells.

01:59:57.240 | So you have 384 well plates,

01:59:59.280 | and you now do 384 parallel reactions

02:00:02.320 | to measure the expression of 384 cells.

02:00:05.520 | That sounds amazing, and it was amazing,

02:00:08.200 | but we wanna do a million cells.

02:00:11.200 | How do you go from these wells to a million cells?

02:00:14.040 | You can't.

02:00:15.520 | So what the next technology was after that

02:00:18.640 | is instead of using a well for every reaction,

02:00:21.520 | you now use a lipid droplet for every reaction.

02:00:26.200 | So you use micro droplets as reaction chambers

02:00:30.320 | to basically amplify RNA.

02:00:32.080 | So here's the idea.

02:00:34.480 | You basically have microfluidics,

02:00:36.480 | where you basically have every single cell

02:00:38.680 | coming down one tube in your microfluidics,

02:00:41.440 | and you have little bubbles getting created in the other way

02:00:44.560 | with specifical primers that mark every cell

02:00:47.440 | with its own barcode.

02:00:49.240 | You basically couple the two,

02:00:50.680 | and you end up with little bubbles that have a cell

02:00:54.320 | and tons of markers for that cell.

02:00:57.280 | You now mark up all of the RNA for that one cell

02:01:00.280 | with the same exact barcode,

02:01:01.680 | and you then lyse all of the droplets,

02:01:06.240 | and you sequence the heck out of that,

02:01:08.200 | and you have for every RNA molecule

02:01:10.280 | a unique identifier that tells you what cell was it on.

02:01:12.720 | - That is such good engineering, microfluidics,

02:01:16.320 | and using some kind of primer to put a label on the thing.

02:01:21.320 | I mean, you're making this sound easy.

02:01:24.040 | I assume it's-- - It's beautiful, right?

02:01:26.160 | - But it's gorgeous, yeah.

02:01:27.280 | - So there's the next generation.

02:01:29.440 | So that's the second generation.

02:01:30.840 | Next generation is, forget the microfluidics altogether.

02:01:33.800 | Just use big bottles.

02:01:35.560 | How can you possibly do that with big bottles?

02:01:37.840 | So here's the idea.

02:01:39.280 | You dissociate all of your cells,

02:01:41.160 | or all of your nuclei from complex cells like brain cells

02:01:44.200 | that are very long and sticky, so you can't do that.

02:01:47.640 | So if you have blood cells,

02:01:49.040 | or if you have neuronal nuclei or brain nuclei,

02:01:51.960 | you can basically dissociate, let's say, a million cells.

02:01:56.040 | You now want to add a unique barcode,

02:01:58.800 | a unique barcode in each one of a million cells

02:02:01.440 | using only big bottles.

02:02:03.240 | How can you possibly do that?

02:02:04.240 | Sounds crazy, but here's the idea.

02:02:06.000 | You use 100 of these bottles.

02:02:08.720 | You randomly shuffle all your million cells,

02:02:13.320 | and you throw them into those 100 bottles,

02:02:15.040 | randomly, completely randomly.

02:02:17.040 | You add one barcode out of 100 to every one of the cells.

02:02:21.440 | You then, you now take them all out,

02:02:23.440 | you shuffle them again,

02:02:24.760 | and you throw them again into the same 100 bottles,

02:02:27.200 | but now in a different randomization,

02:02:30.640 | and you add a second barcode.

02:02:33.840 | So every cell now has two barcodes.

02:02:35.760 | You take them out again,

02:02:38.000 | you shuffle them, and you throw them back in.

02:02:40.240 | Another third barcode is adding,

02:02:43.120 | randomly, from the same 100 barcodes.

02:02:47.240 | You've now labeled every cell probabilistically

02:02:51.920 | based on the unique path that it took

02:02:53.520 | of which of 100 bottles did it go for the first time,

02:02:55.760 | which of 100 bottles the second time,

02:02:57.240 | and which of 100 bottles the third time.

02:02:59.200 | 100 times 100 times 100 is a million unique barcodes

02:03:04.080 | in every single one of these cells

02:03:07.200 | without ever using microfluid.

02:03:09.520 | - Very clever.

02:03:10.360 | - It's beautiful, right? - From a computer science

02:03:11.600 | perspective, that's very clever.

02:03:12.720 | - Yeah, so you now have the single cell

02:03:14.960 | sequencing technology.

02:03:16.000 | You can use the wells, you can use the bubbles,

02:03:18.680 | or you can use the bottles.

02:03:20.040 | You have ways--

02:03:22.360 | - The bubbles still sound pretty damn cool.

02:03:23.640 | - The bubbles are awesome,

02:03:24.600 | and that's basically the main technology that we're using.

02:03:26.480 | So the bubbles is the main technology.

02:03:29.040 | So there are kits now that companies just sell

02:03:32.360 | to basically carry out single cell RNA sequencing

02:03:34.840 | that you can basically, for $2,000,

02:03:37.480 | you can basically get 10,000 cells from one sample,

02:03:40.520 | and for every one of those cells,

02:03:44.800 | you basically have the transcription of thousands of genes.

02:03:48.040 | And of course, the data for any one cell is noisy,

02:03:52.680 | but being computer scientists,

02:03:54.200 | we can aggregate the data from all of the cells together

02:03:57.360 | across thousands of individuals together

02:03:59.320 | to basically make very robust inferences.

02:04:01.600 | So the third technology is basically single cell

02:04:05.560 | RNA sequencing that allows you to now start asking

02:04:08.520 | not just what is the brain expression level difference

02:04:12.680 | of that genetic variant,

02:04:14.240 | but what is the expression difference

02:04:15.960 | of that one genetic variant

02:04:17.440 | across every single subtype of brain cell?

02:04:21.520 | How is the variance changing?

02:04:23.320 | You can't just, you know, with a brain sample,

02:04:27.240 | you can just ask about the mean.

02:04:28.840 | What is the average expression?

02:04:30.680 | If I instead have 3,000 cells that are neurons,

02:04:35.280 | I can ask not just what is the neuronal expression,

02:04:38.320 | I can say for layer five excitatory neurons,

02:04:41.960 | of which I have, I don't know, 300 cells,

02:04:44.080 | what is the variance that this genetic variant has?

02:04:46.960 | So suddenly, it's amazingly more powerful.

02:04:50.960 | I can basically start asking

02:04:52.040 | about this middle layer of gene expression

02:04:54.200 | at unprecedented levels.

02:04:56.080 | - And when you look at the average,

02:04:57.080 | it washes out some potentially important signal

02:05:01.400 | that corresponds to ultimately the disease.

02:05:03.720 | - Completely.

02:05:04.560 | - Yeah.

02:05:05.400 | - So that, I can do that at the RNA level,

02:05:07.720 | but I can also do that at the DNA level for the epigenome.

02:05:11.600 | So remember how before I was telling

02:05:13.200 | about all this technology that we're using

02:05:14.320 | to probe the epigenome?

02:05:15.760 | One of them is DNA accessibility.

02:05:18.000 | So what we're doing in my lab is that

02:05:19.440 | from the same dissociation of, say, a brain sample,

02:05:23.280 | where you now have all these tens of thousands

02:05:24.960 | of cells floating around,

02:05:26.600 | you basically take half of them to do RNA profiling,

02:05:30.080 | and the other half to do epigenome profiling,

02:05:31.920 | both at the single cell level.

02:05:34.000 | So that allows you to now figure out

02:05:35.440 | what are the millions of DNA enhancers

02:05:39.840 | that are accessible in every one

02:05:41.920 | of tens of thousands of cells.

02:05:43.640 | And computationally, we can now take the RNA

02:05:47.440 | and the DNA readouts and group them together

02:05:50.560 | to basically figure out how is every enhancer

02:05:55.200 | related to every gene.

02:05:57.400 | And remember these sort of enhancer gene linking

02:05:59.400 | that we were doing across 833 samples?

02:06:02.320 | 833 is awesome, don't get me wrong,

02:06:05.000 | but 10 million is way more awesome.

02:06:08.080 | So we can now look at correlated activity

02:06:10.120 | across 2.3 million enhancers and 20,000 genes

02:06:14.560 | in each of millions of cells

02:06:16.520 | to basically start piecing together

02:06:18.080 | the regulatory circuitry of every single type of neuron,

02:06:22.520 | every single type of astrocytes, oligodendrocytes,

02:06:24.840 | microglial cell inside the brains

02:06:27.560 | of 1,500 individuals that we've sampled

02:06:30.400 | across multiple different brain regions

02:06:33.480 | across both DNA and RNA.

02:06:36.120 | So that's the dataset that my team generated last year alone.

02:06:39.520 | So in one year, we've basically generated 10 million cells

02:06:43.600 | from human brain across a dozen different disorders,

02:06:48.040 | across schizophrenia, Alzheimer's,

02:06:49.880 | frontotemporal dementia, Lewy body dementia, ALS,

02:06:53.560 | Huntington's disease, post-traumatic stress disorder,

02:06:58.640 | autism, bipolar disorder, healthy aging, et cetera.

02:07:04.280 | - So it's possible that even just within that dataset

02:07:08.000 | lie a lot of keys to understanding these diseases

02:07:13.000 | and then be able to directly lead to then treatment.

02:07:18.360 | - Correct, correct.

02:07:19.520 | So basically we are now-- - Motivating.

02:07:22.040 | - Yeah, so our computational team is in heaven right now

02:07:24.720 | and we're looking for people.

02:07:25.720 | I mean, if you have listeners who are--

02:07:27.080 | - They inspire, yeah, yeah.

02:07:28.000 | - Super smart computational people.

02:07:29.640 | - So this is a very interesting kind of side question.

02:07:32.960 | How much of this is biology, how much of this is computation?

02:07:36.160 | So you had the computational biology group,

02:07:38.600 | but how much of, should you be comfortable with biology

02:07:43.600 | to be able to solve some of these problems?

02:07:48.320 | If you just find, if you put several of the hats

02:07:51.440 | that you wear on, fundamentally,

02:07:53.640 | are you thinking like a computer scientist here?

02:07:56.400 | - You have to.

02:07:57.400 | This is the only way.

02:07:58.680 | As I said, we are the descendants

02:08:01.480 | of the first digital computer.

02:08:02.640 | We're trying to understand the digital computer.

02:08:04.920 | We're trying to understand the circuitry,

02:08:06.240 | the logic of this digital core computer

02:08:11.080 | and all of these analog layers surrounding it.

02:08:14.040 | So the case that I've been making

02:08:17.320 | is that you cannot think one gene at a time.

02:08:19.720 | The traditional biology is dead.

02:08:21.920 | There's no way, you cannot solve disease

02:08:23.760 | with traditional biology.

02:08:24.840 | You need it as a component.

02:08:27.000 | Once you figured out RX3 and RX5,

02:08:29.480 | you now can then say, hey,

02:08:31.120 | have you guys worked on those genes

02:08:32.360 | with your single gene approach?

02:08:33.760 | We'd love to know everything you know.

02:08:35.440 | And if you haven't,

02:08:36.400 | we now know how important these genes are.

02:08:38.760 | Let's now launch a single gene program

02:08:40.760 | to dissect them and understand them.

02:08:43.280 | But you cannot use that as a way to dissect disease.

02:08:46.640 | You have to think genomically.

02:08:48.480 | You have to think from the global perspective

02:08:50.720 | and you have to build these circuits systematically.

02:08:53.280 | So we need numbers of computer scientists

02:08:56.520 | who are interested and willing to dive into these data,

02:09:00.040 | you know, fully, fully in and sort of extract meaning.

02:09:04.800 | We need computer science people

02:09:06.920 | who can understand sort of machine learning and inference

02:09:10.080 | and sort of, you know, decouple these matrices,

02:09:12.960 | come up with super smart ways of sort of dissecting them.

02:09:16.240 | But we also need computer scientists who understand biology,

02:09:20.320 | who are able to design the next generation of experiments.

02:09:24.400 | Because many of these experiments,

02:09:26.320 | no one in their right mind would design them

02:09:28.400 | without thinking of the analytical approach

02:09:30.200 | that you would use to deconvolve the data afterwards.

02:09:32.840 | Because it's massive amounts of ridiculously noisy data.

02:09:35.640 | And if you don't have the computational pipeline

02:09:39.680 | in your head before you even design the experiment,

02:09:42.600 | you would never design the experiment that way.

02:09:44.640 | - That's brilliant, so in designing the experiment,

02:09:47.120 | you have to see the entirety of the computational pipeline.

02:09:50.040 | - That drives the design.

02:09:52.120 | That even drives the necessity for that design.

02:09:55.600 | Basically, you know, if you didn't have

02:09:57.880 | a computer scientist way of thinking,

02:10:00.120 | you would never design these hugely combinatorial,

02:10:03.480 | massively parallel experiments.

02:10:05.440 | So that's why you need interdisciplinary teams.

02:10:09.080 | You need teams.

02:10:10.240 | And I wanna sort of clarify that,

02:10:12.280 | what do we mean by computational biology group?

02:10:15.120 | The focus is not on computational,

02:10:16.760 | the focus is on the biology.

02:10:18.800 | So we are a biology group.

02:10:20.800 | What type of biology?

02:10:22.040 | Computational biology.

02:10:24.160 | That's the type of biology that uses the whole genome.

02:10:27.640 | That's the type of biology that designs experiments,

02:10:30.400 | genomic experiments, that can only be interpreted

02:10:33.120 | in the context of the whole genome.

02:10:34.720 | - Right, so it's philosophically

02:10:37.320 | looking at biology as a computer.

02:10:39.120 | - Correct, correct.

02:10:40.640 | - So which is, in the context of the history of biology,

02:10:45.080 | is a big transformation.

02:10:46.640 | - Yeah, yeah, you can think of the name as,

02:10:48.920 | what do we do?

02:10:50.040 | Only computation, that's not true.

02:10:51.880 | But how do we study it?

02:10:53.760 | Only computationally, that is true.

02:10:55.520 | So all of these single cell sequencing

02:10:58.560 | can now be coupled with the technology

02:11:00.480 | that we talked about earlier for perturbation.

02:11:02.880 | So here's the crazy thing.

02:11:04.320 | Instead of using these wells and these robotic systems

02:11:07.840 | for doing one drug at a time,

02:11:10.560 | or for perturbing one gene at a time in thousands of wells,

02:11:14.680 | you can now do this using a pool of cells

02:11:18.200 | and single cell RNA sequencing.

02:11:20.040 | How?

02:11:20.920 | You basically can take these perturbations using CRISPR,

02:11:25.920 | and instead of using a single guide RNA,

02:11:29.320 | you can use a library of guide RNAs

02:11:31.320 | generated exactly the same way using this array technology.

02:11:34.440 | So you synthesize a thousand different guide RNAs.

02:11:37.560 | You now take each of these guide RNAs

02:11:42.240 | and you insert them in a pool of cells

02:11:45.400 | where every cell gets one perturbation.

02:11:48.160 | And you use CRISPR editing or CRISPR,

02:11:51.360 | so with either CRISPR-Cas9 to edit the genome

02:11:56.520 | with these thousand perturbations.

02:11:57.720 | - Or the activation one. - Or with the activation

02:11:59.920 | or with the repression.

02:12:01.320 | And you now can have a single cell readout

02:12:04.400 | where every single cell has received

02:12:06.840 | one of these modifications.

02:12:09.520 | And you can now, in massively parallel ways,

02:12:12.000 | couple the perturbation and the readout

02:12:17.040 | in a single experiment.

02:12:18.280 | - How are you tracking which perturbations

02:12:20.200 | each cell received?

02:12:21.560 | - So there's ways of doing that,

02:12:23.560 | but basically one way is to make that perturbation

02:12:26.200 | an expressible vector so that part of your RNA reading

02:12:30.120 | is actually that perturbation itself.

02:12:33.120 | So you can basically put it in an expressible part,

02:12:36.440 | so you can self-drive it.

02:12:37.680 | So the point that I wanna get across

02:12:40.320 | is that the sky's the limit.

02:12:42.040 | You basically have these tools,

02:12:43.560 | these building blocks of molecular biology.

02:12:46.440 | You have these massive datasets of computational biology.

02:12:50.240 | You have this huge ability to sort of use machine learning

02:12:54.200 | and statistical methods and linear algebra

02:12:57.080 | to sort of reduce the dimensionality

02:12:59.400 | of all these massive datasets.

02:13:01.680 | And then you end up with a series of actionable targets

02:13:06.680 | that you can then couple with pharma

02:13:11.080 | and just go after systematically.

02:13:13.280 | So the ability to sort of bring genetics to the epigenomics,

02:13:18.280 | to the transcriptomics, to the cellular readouts

02:13:21.440 | using these sort of high-throughput perturbation technologies

02:13:24.320 | that I'm talking about,

02:13:25.440 | and ultimately to the organismal

02:13:27.800 | through the electronic health record endophenotypes,

02:13:31.440 | and ultimately the disease battery of assays

02:13:35.480 | at the cognitive level, at the physiological level,

02:13:38.200 | and every other level,

02:13:41.760 | there is no better or more exciting field, in my view,

02:13:45.120 | to be a computer scientist then

02:13:46.960 | or to be a scientist in period.

02:13:48.920 | Basically, this confluence of technologies,

02:13:51.200 | of computation, of data, of insights,

02:13:53.960 | and of tools for manipulation

02:13:56.160 | is unprecedented in human history.

02:13:58.760 | And I think this is what's shaping the next century

02:14:02.640 | to really be a transformative century

02:14:04.720 | for our species and for our planet.

02:14:09.360 | - So you think the 21st century will be remembered

02:14:12.000 | for the big leaps in understanding

02:14:15.560 | and alleviation of biology?

02:14:18.640 | - If you look at the path

02:14:20.120 | between discovery and therapeutics,

02:14:22.720 | it's been on the order of 50 years.

02:14:24.760 | It's been shortened to 40, 30, 20,

02:14:27.040 | and now it's on the order of 10 years.

02:14:29.380 | But the huge number of technologies

02:14:31.960 | that are going on right now for discovery

02:14:35.480 | will result undoubtedly

02:14:37.920 | in the most dramatic manipulation of human biology

02:14:41.200 | that we've ever seen in the history of humanity

02:14:44.040 | in the next few years.

02:14:45.160 | - Do you think we might be able to cure some of the diseases

02:14:47.560 | we started this conversation with?

02:14:49.640 | - Absolutely, absolutely.

02:14:51.320 | It's only a matter of time.

02:14:54.040 | Basically, the complexity is enormous,

02:14:55.880 | and I don't want to underestimate the complexity,

02:14:58.360 | but the number of insights is unprecedented,

02:15:01.360 | and the ability to manipulate is unprecedented,

02:15:03.880 | and the ability to deliver these small molecules

02:15:07.520 | and other non-traditional medicine perturbations.

02:15:10.920 | There's a lot of sort of new,

02:15:13.080 | there's a new generation of perturbations

02:15:15.600 | that you can use at the DNA level, at the RNA level,

02:15:18.440 | at the microRNA level, the epigenomic level.

02:15:22.640 | There's a battery of new generations of perturbations.

02:15:26.440 | If you couple that with cell type identifiers

02:15:30.320 | that can basically sense when you are in the right cell

02:15:32.880 | based on a specific combination

02:15:34.080 | and then turn on that intervention for that cell,

02:15:37.360 | you can now think of combinatorial interventions

02:15:40.000 | where you can basically sort of feed

02:15:42.040 | a synthetic biology construct to someone

02:15:44.400 | that will basically do different things in different cells.

02:15:47.560 | So basically for cancer, this is one of the therapeutics

02:15:50.160 | that our collaborator Ron Weiss is using

02:15:52.400 | to basically start sort of engineering the circuits

02:15:54.800 | that will use microRNA sensors of the environment

02:15:57.120 | to sort of know if you're in a tumor cell,

02:15:59.080 | or if you're in an immune cell,

02:16:00.280 | or if you're in a stromal cell, and so on and so forth,

02:16:02.040 | and basically turn on particular interventions there.

02:16:04.800 | You can sort of create constructs

02:16:07.440 | that are tuned to only the liver cells,

02:16:10.240 | or only the heart cells, or only the brain cells,

02:16:14.680 | and then have these new generations of therapeutics

02:16:18.760 | coupled with this immense amount of knowledge

02:16:21.840 | on the sort of which targets to choose

02:16:23.920 | and what biological processes to measure

02:16:25.920 | and how to intervene.

02:16:27.560 | My view is that disease is gonna be fundamentally altered

02:16:31.800 | and alleviated as we go forward.

02:16:35.020 | - Next time we talk,

02:16:37.300 | we'll talk about the philosophical implications of that

02:16:39.700 | and the effect of life,

02:16:40.860 | but let's stick to biology for just a little longer.

02:16:44.180 | We did pretty good today.

02:16:45.100 | We stuck to the science.

02:16:47.200 | What are you excited in terms of the future of this field,

02:16:52.200 | the technologies in your own group, in your own mind?

02:16:58.580 | You're leading the world at MIT in the science

02:17:01.780 | and the engineering of this work,

02:17:04.340 | so what are you excited about here?

02:17:06.460 | - I could not be more excited.

02:17:08.580 | We are one of many, many teams who are working on this.

02:17:12.580 | In my team, the most exciting parts are many folds.

02:17:16.940 | So basically, we've now assembled

02:17:18.980 | these battery of technologies.

02:17:20.180 | We've assembled these massive, massive data sets,

02:17:22.420 | and now we're really sort of in the stage

02:17:25.020 | of our team's path of generating disease insights.

02:17:30.340 | So we are simultaneously working on a paper

02:17:34.420 | on schizophrenia right now

02:17:35.980 | that is basically using

02:17:37.020 | the single-cell profiling technologies,

02:17:38.580 | using this editing and manipulation technologies

02:17:40.980 | to basically show how the master regulators

02:17:45.260 | underlying changes in the brain

02:17:47.500 | that are sort of found in schizophrenia

02:17:49.900 | are in fact affecting excitatory neurons

02:17:52.220 | and inhibitory neurons in pathways

02:17:54.520 | that are active both in synaptic pruning,

02:17:57.740 | but also in early development.

02:17:59.180 | We've basically found this set of four regulators

02:18:01.660 | that are connecting these two processes

02:18:03.180 | that were previously separate in schizophrenia

02:18:06.460 | in sort of having a sort of more unified view

02:18:10.220 | across those two sides.

02:18:12.660 | The second one is in the area of metabolism.

02:18:15.420 | We basically now have a beautiful collaboration

02:18:17.420 | with the Goodyear Lab

02:18:18.460 | that's basically looking at multi-tissue perturbations

02:18:23.460 | in six or seven different tissues across the body

02:18:28.540 | in the context of exercise

02:18:30.300 | and in the context of nutritional interventions

02:18:33.220 | using both mouse and human,

02:18:35.860 | where we can basically see

02:18:37.180 | what are the cell-to-cell communications

02:18:39.980 | that are changing across them.

02:18:42.300 | And what we're finding is this immense role

02:18:44.940 | of both immune cells,

02:18:46.620 | as well as adipocyte stem cells

02:18:49.140 | in sort of reshaping that circuitry

02:18:51.100 | of all of these different tissues,

02:18:52.700 | and that's sort of painting to a new path

02:18:54.500 | for therapeutical intervention there.

02:18:56.820 | In Alzheimer's, it's this huge focus on microglia,

02:19:00.780 | and now we're discovering different classes

02:19:02.620 | of microglial cells

02:19:04.220 | that are basically either synaptic or immune,

02:19:08.220 | and these are playing vastly different roles

02:19:13.300 | in Alzheimer's versus in schizophrenia.

02:19:15.980 | And what we're finding is this immense complexity

02:19:19.100 | as you go further and further down

02:19:21.420 | of how, in fact, there's 10 different types of microglia,

02:19:25.580 | each with their own sort of expression programs.

02:19:28.260 | We used to think of them as, oh, yeah, they're microglia,

02:19:30.900 | but in fact, now we're realizing

02:19:32.500 | just even in that sort of least abundant of cell types,

02:19:36.220 | there's this incredible diversity there.

02:19:38.260 | The differences between brain regions

02:19:41.420 | is another sort of major, major insight.

02:19:43.940 | Again, one would think that,

02:19:46.100 | oh, astrocytes are astrocytes no matter where they are,

02:19:48.620 | but no, there's incredible region-specific differences

02:19:52.420 | in the expression patterns

02:19:53.940 | of all of the major brain cell types

02:19:56.100 | across different brain regions.

02:19:57.780 | So basically there's the neocortical regions

02:19:59.420 | that are sort of the recent innovation

02:20:00.980 | that makes us so different from all other species.

02:20:03.460 | There's the sort of reptilian brain sort of regions

02:20:06.500 | that are sort of much more,

02:20:08.100 | very extremely distinct.

02:20:10.500 | There's the cerebellum.

02:20:11.660 | Each of those basically is associated

02:20:15.180 | in a different way with disease,

02:20:17.380 | and what we're doing now is looking into

02:20:19.300 | pseudotemporal models for how disease progresses

02:20:23.740 | across different regions of the brain.

02:20:25.740 | If you look at Alzheimer's,

02:20:27.060 | it basically starts in this small region

02:20:29.260 | called the entorhinal cortex,

02:20:30.500 | and then it spreads through the brain,

02:20:33.380 | and through the hippocampus,

02:20:35.500 | and ultimately affecting the neocortex.

02:20:39.460 | And with every brain region that it hits,

02:20:41.860 | it basically has a different impact on the cognitive

02:20:46.860 | and memory aspects, orientation,

02:20:50.500 | short-term memory, long-term memory, et cetera,

02:20:52.780 | which is dramatically affecting the cognitive path

02:20:56.700 | that the individuals go through.

02:20:58.220 | So what we're doing now is creating

02:21:00.980 | these computational models for ordering the cells

02:21:04.620 | and the regions and the individuals

02:21:07.060 | according to their ability to predict Alzheimer's disease.

02:21:10.460 | So we can have a cell-level predictor of pathology

02:21:14.740 | that allows us to now create a temporal time course

02:21:18.140 | that tells us when every gene turns on

02:21:19.780 | along this pathology progression,

02:21:22.580 | and then trace that across regions

02:21:25.060 | and pathological measures that are region-specific,

02:21:27.780 | but also cognitive measures, and so on and so forth.

02:21:30.260 | So that allows us to now sort of, for the first time,

02:21:33.140 | look at, can we actually do early intervention

02:21:35.580 | for Alzheimer's, where we know that the disease

02:21:38.220 | starts manifesting for 10 years

02:21:39.980 | before you actually have your first cognitive loss?

02:21:43.100 | Can we start seeing that path to build new diagnostics,

02:21:47.980 | new prognostics, new biomarkers

02:21:50.140 | for this sort of early intervention in Alzheimer's?

02:21:53.460 | The other aspect that we're looking at is mosaicism.

02:21:56.940 | We talked about the common variants and the rare variants,

02:21:59.780 | but in addition to those rare variants,

02:22:01.860 | as your initial cell that forms the zygote

02:22:06.540 | divides and divides and divides,

02:22:08.260 | with every cell division,

02:22:09.740 | there are additional mutations that are happening.

02:22:12.340 | So what you end up with is your brain being a mosaic

02:22:16.220 | of multiple different types of genetic underpinnings.

02:22:19.060 | Some cells contain a mutation that other cells don't have.

02:22:23.180 | So every human has the common variants

02:22:27.300 | that all of us carry to some degree,

02:22:29.900 | the rare variants that your immediate tree

02:22:33.420 | of the human species carries,

02:22:35.260 | and then there's the somatic variant,

02:22:37.260 | which is the tree that happened after the zygote

02:22:40.460 | that sort of forms your own body.

02:22:44.100 | So these somatic alterations

02:22:47.100 | is something that has been previously inaccessible

02:22:50.100 | to study in human post-mortem samples.

02:22:53.060 | But right now,

02:22:54.220 | with the advent of single-cell RNA sequencing,

02:22:57.220 | in this particular case,

02:22:58.100 | we're using the well-based sequencing,

02:22:59.740 | which is much more expensive,

02:23:00.780 | but gives you a lot richer information

02:23:02.860 | about each of those transcripts.

02:23:04.420 | So we're using now that richer information

02:23:06.500 | to infer mutations that have happened

02:23:09.700 | in each of the thousands of genes

02:23:12.060 | that sort of are active in these cells,

02:23:15.380 | and then understand how the genome relates to the function,

02:23:20.380 | this genotype-phenotype relationship

02:23:24.780 | that we usually build in GWAS,

02:23:26.420 | between, in genome-wide association studies,

02:23:28.620 | between genetic variation and disease.

02:23:31.260 | We're now building that at the cell level,

02:23:33.780 | where for every cell,

02:23:34.980 | we can relate the unique specific genome of that cell

02:23:38.620 | with the expression patterns of that cell,

02:23:41.100 | and the predicted function,

02:23:42.940 | using these predictive models that I mentioned before,

02:23:44.980 | on dysregulation for cognition,

02:23:47.340 | for pathology in Alzheimer's, at the cell level.

02:23:50.820 | And what we're finding is that the genes that are altered,

02:23:53.940 | and the genetic regions that are altered in common variants

02:23:56.820 | versus rare variants versus somatic variants,

02:23:59.060 | are actually very different from each other.

02:24:01.140 | The somatic variants are pointing to neuronal energetics,

02:24:05.500 | and oligodendrocyte functions

02:24:08.580 | that are not visible in the genetic legions

02:24:10.660 | that you find for the common variants,

02:24:12.500 | probably because they have too strong of an effect

02:24:15.180 | that evolution is just not tolerating them

02:24:17.500 | on the common side of the allele frequency spectrum.

02:24:20.860 | - So the somatic one,

02:24:22.060 | that's the variation that happens after the zygote,

02:24:24.820 | after a new individual. - Correct.

02:24:26.980 | - I mean, this is a dumb question,

02:24:28.220 | but there's mutation and variation, I guess,

02:24:31.140 | that happens there,

02:24:32.460 | and you're saying that through this,

02:24:35.260 | if we focus in on individual cells,

02:24:37.100 | we're able to detect a story that's interesting there,

02:24:40.060 | and that might be a very unique

02:24:42.300 | kind of important variability that arises for,

02:24:46.620 | you said neuronal, or something,

02:24:49.140 | that would sound-- - Energetics.

02:24:50.260 | - Energetics,

02:24:51.100 | that's not a really cool term. - Energetics.

02:24:51.940 | So you're, I mean, the metabolism of humans

02:24:55.060 | is dramatically altered from that of nearby species.

02:24:59.020 | You know, we talked about that last time,

02:25:00.580 | that basically we are able to consume meat

02:25:03.240 | that is incredibly energy-rich,

02:25:05.900 | and that allows us to sort of have functions

02:25:09.800 | that are, you know,

02:25:11.620 | meeting this humongous brain that we have.

02:25:14.660 | So basically, on one hand,

02:25:15.660 | every one of our brain cells is much more energy-efficient

02:25:18.340 | than our neighbors, than our relatives.

02:25:20.420 | Number two, we have way more of these cells,

02:25:23.140 | and number three, we have, you know,

02:25:26.420 | this new diet that allows us to now feed all these needs.

02:25:29.980 | That basically creates a massive amount of damage,

02:25:33.420 | oxidative damage, from this huge,

02:25:36.280 | super-powered factory of ideas and thoughts

02:25:40.220 | that we carry in our skull.

02:25:41.980 | And that factory has energetic needs,

02:25:44.980 | and there's a lot of sort of biological processes

02:25:47.620 | underlying that, that we are finding are altered

02:25:51.060 | in the context of Alzheimer's disease.

02:25:52.820 | - That's fascinating that,

02:25:54.180 | so you have to consider all of these systems

02:25:57.220 | if you wanna understand even something like diseases

02:26:00.300 | that you would maybe traditionally associate

02:26:02.700 | with just the particular cells of the brain.

02:26:05.260 | - Yeah.

02:26:06.100 | - The immune system.

02:26:08.760 | - The metabolic system.

02:26:09.720 | - The metabolic system.

02:26:11.120 | - And these are all the things that makes us uniquely human.

02:26:13.320 | So our immune system is dramatically different

02:26:15.720 | from that of our neighbors.

02:26:16.960 | Our societies are so much more clustered.

02:26:19.520 | The history of infections that have plagued

02:26:21.680 | the human population is, you know,

02:26:24.040 | dramatically different from every other species.

02:26:26.320 | The way that our society and our population

02:26:28.640 | has sort of exploded has basically put unique pressures

02:26:31.800 | on our immune system, and our immune system

02:26:34.060 | has both coped with that density and also been shaped by,

02:26:37.360 | as I mentioned, the vast amount of death

02:26:40.080 | that has happened in the Black Plague

02:26:41.800 | and other sort of selective events in human history,

02:26:44.560 | famines, ice ages, and so forth.

02:26:47.080 | So that's number one on the sort of immune side.

02:26:49.840 | On the metabolic side, you know, again,

02:26:52.360 | we are able to sort of run marathons.

02:26:54.760 | You know, I don't know if you remember

02:26:56.240 | the sort of human versus horse experiment

02:26:58.400 | where the horse actually tires out faster than the human,

02:27:00.840 | and the human actually wins.

02:27:03.120 | So on the metabolic side, we're dramatically different.

02:27:05.780 | On the immune side, we're dramatically different.

02:27:07.460 | On the brain side, again, you know,

02:27:09.880 | no need to sort of, you know, it's a no-brainer

02:27:12.380 | of how our brain is like just enormously more capable.

02:27:16.700 | And then, you know, in the side of cancer,

02:27:19.500 | so basically the cancers that humans are having,

02:27:21.700 | the exposures, the environmental exposures,

02:27:23.980 | is again dramatically different.

02:27:25.660 | And the lifespan, the expansion of human lifespan

02:27:29.060 | is unseen in any other species

02:27:31.960 | in, you know, recent evolutionary history.

02:27:35.520 | And that now leads to a lot of new disorders

02:27:39.080 | that are starting to, you know, manifest late in life.

02:27:43.760 | So, you know, Alzheimer's is one example

02:27:46.240 | where basically, you know, these fast energetic needs

02:27:49.000 | over a lifetime of thinking

02:27:52.080 | can basically lead to all of these debris

02:27:54.320 | and eventually saturate the system

02:27:56.360 | and lead to, you know, Alzheimer's in the late life.

02:27:59.420 | But there's, you know, there's just such a dramatic

02:28:04.700 | set of frontiers when it comes to aging research

02:28:08.660 | that, you know, will, so what I often like to say

02:28:11.740 | is that if you want to engineer a car

02:28:14.740 | to go from 70 miles an hour to 120 miles an hour,

02:28:17.380 | that's fine, you can basically, you know,

02:28:19.080 | fix a few components.

02:28:20.380 | If you want it to now go at 400 miles an hour,

02:28:22.740 | you have to completely redesign the entire car.

02:28:25.720 | Because the system has just not evolved to go that far.

02:28:30.720 | Basically, our human body has only evolved

02:28:33.660 | to live to, I don't know, 120.

02:28:36.140 | Maybe we can get to 150 with minor changes.

02:28:39.180 | But if, you know, as we start pushing these frontiers

02:28:41.460 | for not just living, but well living,

02:28:45.100 | the F-Zine that we talked about last time.

02:28:48.120 | So to basically push F-Zine into the 80s and 90s

02:28:51.460 | and 100s and, you know, much further than that,

02:28:54.380 | we will face new challenges that have, you know,

02:28:58.780 | never been faced before.

02:29:00.300 | In terms of cancer, the number of divisions,

02:29:02.220 | in terms of Alzheimer's and brain related disorders,

02:29:05.120 | in terms of metabolic disorders, in terms of regeneration.

02:29:08.220 | There's just so many different frontiers ahead of us.

02:29:10.800 | So I am thrilled about where we're heading.

02:29:14.300 | So basically I see this confluence in my lab

02:29:16.580 | and many other labs of AI, of, you know,

02:29:20.480 | sort of, you know, the next frontier of AI for drug design.

02:29:23.100 | So basically these sort of graph neural networks

02:29:26.460 | on specific chemical designs that allow you to create

02:29:31.460 | new generations of therapeutics.

02:29:34.520 | These molecular biology tricks for intervening

02:29:38.980 | at the system at every level.

02:29:41.140 | These personalized medicine prediction, diagnosis,

02:29:45.380 | and prognosis using the electronic health records

02:29:49.420 | and using these polygenic risk scores,

02:29:51.900 | weighted by the burden, the number of mutations

02:29:55.980 | that are accumulating across common, rare

02:29:57.740 | and somatic variants, the burden converging

02:30:01.220 | across all of these different molecular pathways,

02:30:05.560 | the delivery of specific drugs and specific interventions

02:30:09.460 | into specific cell types.

02:30:10.840 | And again, you've talked with Bob Langer about this.

02:30:12.700 | There's, you know, many giants in that field.

02:30:14.620 | And then the last concept is not intervening

02:30:18.500 | at the single gene level.

02:30:20.620 | And I want you to sort of conceptualize the concept

02:30:23.300 | of an on target side effect.

02:30:27.540 | What is an on target side effect?

02:30:29.140 | An off target side effect is when you design a molecule

02:30:31.780 | to target one gene and instead it targets another gene

02:30:34.520 | and you have side effects because of that.

02:30:36.580 | An on target side effect is when your molecule

02:30:38.900 | does exactly what you were expecting,

02:30:40.940 | but that gene is pleiotropic.

02:30:43.580 | Pleio means many, tropos means ways, many ways.

02:30:46.860 | It acts in many ways.

02:30:48.100 | It's a multifunctional gene.

02:30:49.980 | So you find that this gene plays a role in this,

02:30:52.500 | but as we talked about, the wiring of genes to phenotypes

02:30:56.580 | is extremely dense and extremely complex.

02:30:58.980 | So the next stage of intervention will be intervening,

02:31:03.000 | not at the gene level, but at the network level.

02:31:05.620 | Intervening at the set of pathways and the set of genes

02:31:08.620 | with multi input perturbations to the system,

02:31:12.100 | multi input modulations, pharmaceutical

02:31:14.820 | or other interventional.

02:31:17.660 | That basically allow you to now work

02:31:20.220 | at the sort of full level of understanding,

02:31:23.900 | not just in your brain, but across your body,

02:31:26.300 | not just in one gene, but across the set of pathways

02:31:29.020 | and so on and so forth for every one of these disorders.

02:31:31.900 | So I think that we're finally at the level of systems

02:31:35.340 | medicine of basically instead of sort of medicine

02:31:38.140 | being at a single gene level,

02:31:40.340 | medicine being at the systems level,

02:31:42.180 | where it can be personalized based on a specific set

02:31:45.260 | of genetic markers and genetic perturbations

02:31:47.100 | that you are either born with,

02:31:49.460 | or that you have developed during your lifetime,

02:31:52.940 | your unique set of exposures,

02:31:54.660 | your unique set of biomarkers,

02:31:56.940 | and your unique set of current set of conditions

02:32:00.860 | through your EHR and other ways.

02:32:04.420 | And the precision component of intervening

02:32:09.980 | extremely precisely in the specific pathways

02:32:12.980 | and in specific combinations of genes

02:32:14.620 | that should be modulated to sort of bring you

02:32:16.620 | from the disease state to the physiologically normal state,

02:32:20.620 | or even to physiologically improved state

02:32:22.620 | through this combination of interventions.

02:32:25.620 | So that's in my view, the field

02:32:26.980 | where basically computer science comes together

02:32:29.060 | with artificial intelligence, statistics,

02:32:31.220 | all of these other tools,

02:32:32.420 | molecular biology technologies and biotechnology

02:32:34.980 | and pharmaceutical technologies

02:32:36.140 | that are sort of revolutionary in the way of intervention.

02:32:39.380 | And of course, this massive amount of molecular biology

02:32:41.820 | and data gathering and generation and perturbation

02:32:44.500 | in massively parallel ways.

02:32:46.300 | So there's no better way, there's no better time,

02:32:49.660 | there's no better place to be sort of,

02:32:52.700 | you know, looking at this whole confluence of ideas.

02:32:56.740 | And I'm just so thrilled to be a small part

02:32:58.980 | of this amazing, enormous ecosystem.

02:33:01.300 | - It's exciting to imagine what humans of 100,

02:33:04.260 | 200 years from now, what their life experience is like,

02:33:08.660 | because these ideas seem to have potential

02:33:12.300 | to transform the quality of life.

02:33:15.060 | And I think that when they look back at us,

02:33:18.220 | they probably wonder how we were put up

02:33:21.140 | with all the suffering in the world.

02:33:23.500 | Manolis, it's a huge honor.

02:33:25.180 | Thank you for spending this early Sunday morning with me.

02:33:29.220 | I deeply appreciate it.

02:33:30.700 | See you next time.

02:33:31.540 | - Sounds like a plan.

02:33:32.380 | Thank you, Lex.

02:33:33.740 | Thanks for listening to this conversation

02:33:35.300 | with Manolis Kellis, and thank you to our sponsors.

02:33:38.860 | SEMrush, which is an SEO optimization tool,

02:33:43.220 | Pessimist Archive, which is one

02:33:45.060 | of my favorite history podcasts,

02:33:47.380 | 8Sleep, which is a self-cooling mattress

02:33:50.100 | with smart sensors and an app,

02:33:52.580 | and finally, BetterHelp, which is an online therapy service.

02:33:57.140 | Please check out these sponsors in the description

02:33:59.340 | to get a discount and to support this podcast.

02:34:03.300 | If you enjoy this thing, subscribe on YouTube,

02:34:05.560 | review it with Five Stars on Apple Podcast,

02:34:07.760 | follow on Spotify, support on Patreon,

02:34:10.460 | or connect with me on Twitter @LexFriedman.

02:34:14.380 | And now, let me leave you some words from Haruki Murakami.

02:34:18.820 | Human beings are ultimately nothing

02:34:20.460 | but carriers, passageways for genes.

02:34:24.500 | They ride us into the ground like racehorses

02:34:27.620 | from generation to generation.

02:34:30.020 | Genes don't think about what constitutes good or evil.

02:34:34.060 | They don't care whether we're happy or unhappy.

02:34:37.420 | We're just means to an end for them.

02:34:39.940 | The only thing they think about

02:34:42.540 | is what is most efficient for them.

02:34:44.780 | Thank you for listening, and hope to see you next time.

02:34:49.200 | (upbeat music)

02:34:51.780 | (upbeat music)

02:34:54.360 | [BLANK_AUDIO]

Manolis Kellis: Biology of Disease | Lex Fridman Podcast #133

Chapters