Functional Genomics: Systematic Approaches for Mapping the Cell • iBiology

00:00:06.13 Hi.
00:00:07.26 I'm Manuel Leonetti,
00:00:09.12 and I'm a group leader at the Chan Zuckerberg Biohub,
00:00:11.11 and in this lecture we're gonna be talking about
00:00:13.29 systematic approaches to map the architecture of human cells.
00:00:18.18 I really love this painting by David Goodsell.
00:00:20.12 This is a mycoplasma bacterium.
00:00:22.00 This is one of the simplest unicellular organisms
00:00:24.01 that we know about.
00:00:25.13 This guy is tiny.
00:00:27.07 It's about a third of a micron wide
00:00:29.05 and has about 500 genes.
00:00:30.16 But I just find it fascinating that we understand its architecture
00:00:33.12 so well that an artist like David
00:00:35.08 can actually paint a very realistic representation of it.
00:00:38.07 So essentially, the focus of my research
00:00:40.18 is to be able to generate data
00:00:43.03 and understand a human cell
00:00:44.21 with the same level of detail.
00:00:46.02 Now, here's the challenge.
00:00:48.01 A human cell is five orders of magnitude
00:00:50.24 bigger and more complex than mycoplasma.
00:00:54.19 Okay.
00:00:56.03 So, we still have much to learn about how it is organized.
00:00:58.11 But in the postgenomic era,
00:01:00.27 we have, you know, a great chance,
00:01:02.09 which is that the sequence of the gen...
00:01:04.21 of the genome gives us the list of parts,
00:01:06.23 the list of all the different proteins
00:01:08.16 that make up a human cell.
00:01:11.01 Now, the problem becomes,
00:01:13.17 can we understand how the proteins are coming together
00:01:17.09 and essentially, you know, decipher the blueprint
00:01:19.14 of cellular architecture?
00:01:21.16 So, today, we're gonna be talking about
00:01:23.19 key methods to tackle this very problem.
00:01:25.28 We're gonna take a very protein-centric view
00:01:28.15 of the cell,
00:01:30.00 and we're gonna be addressing two main questions.
00:01:31.16 Where are the proteins localized in the cell?
00:01:34.22 And how are they all wired together?
00:01:36.24 Okay.
00:01:37.28 And really, my goal here is to give you an overview
00:01:39.23 of the different approaches that have been deployed,
00:01:41.24 or could be deployed,
00:01:43.10 at genome-wide scale
00:01:45.08 to get the complete map of a human cell.
00:01:47.20 So, let's start with localization.
00:01:49.24 Obviously, fluorescence microscopy
00:01:52.02 is the key technique here.
00:01:53.21 And there are too main ways to localize a protein
00:01:56.13 under a fluorescent microscope:
00:01:58.01 immunofluorescence using labeled antibodies,
00:02:00.16 or fusing a protein with fluorescent proteins,
00:02:03.22 such as GFP.
00:02:05.23 So, I think it's fair to say that antibody-based methods
00:02:09.09 are the most advanced.
00:02:10.19 And the best example is the Human Protein Atlas project,
00:02:14.08 which managed to raise polyclonal antibodies
00:02:17.09 against the majority of human proteins.
00:02:20.00 And in a real tour de force experiment,
00:02:21.25 Emma Lundberg and her colleagues
00:02:23.21 have used this collection
00:02:25.19 to map the localization of most proteins
00:02:27.19 in human cell lines.
00:02:29.04 They classify localization between
00:02:31.23 30 different cellular structures.
00:02:33.02 A few of them are shown here.
00:02:35.02 And they've shown that most proteins localize
00:02:37.14 to multiple structures at the same time.
00:02:40.00 I really think that their website
00:02:42.04 is a mine of information,
00:02:43.16 and I encourage you to have a look at it.
00:02:45.28 Now, antibodies usually require cells
00:02:48.15 to be fixed and frozen in time,
00:02:49.22 which is obviously a limitation.
00:02:51.12 But it has one main advantage,
00:02:53.18 which is that the same sample can be bound with antibodies,
00:02:57.00 imaged,
00:02:58.11 and then these antibodies can be washed away,
00:03:00.00 leaving the sample ready for the next round of imaging.
00:03:02.07 So, if we can do this process iteratively,
00:03:04.08 we can localize many proteins
00:03:07.05 in the same sample, over and over again.
00:03:09.27 Okay.
00:03:10.29 And I really think that these multiplexing approaches
00:03:12.19 are gonna become very powerful
00:03:14.11 in the near future.
00:03:15.20 And if you want to see a great example from the
00:03:18.21 Pelmans lab that was published recently,
00:03:20.27 you can check out this paper.
00:03:23.24 Now, again, antibodies cannot capture dynamic information.
00:03:27.13 And this is where protein tags become really interesting,
00:03:29.29 because they allow us to localize proteins in live cells.
00:03:33.25 And as you can see, in these videos
00:03:35.25 that have been taken by Rachel
00:03:37.26 -- who is a student visiting my lab --
00:03:39.14 not only are proteins dynamic within cells,
00:03:41.19 but the cells themselves are dynamic.
00:03:44.11 So, I think it's really important
00:03:46.22 to capture this time information.
00:03:48.17 And the power of fluorescent protein tagging
00:03:50.29 has really been exemplified by the work
00:03:54.06 that's been done in the yeast Saccharomyces cerevisiae,
00:03:56.23 where many teams
00:03:58.28 -- and in particular Erin O'Shea and Jonathan Weissman,
00:04:01.02 when they were at UCSF --
00:04:02.26 have been able to create a genome-wide collection
00:04:05.12 of GFP-tagged yeast lines.
00:04:07.04 So, this image is a composite illustration
00:04:10.11 of many different yeast lines.
00:04:12.00 And by now, the vast majority of
00:04:14.17 the 6,000 or so genes in the budding yeast, cerevisiae,
00:04:17.03 have been tagged with GFP.
00:04:19.29 And beyond informing about localization,
00:04:21.25 the yeast library has really provided us a ton of information
00:04:24.15 about proteins.
00:04:26.17 It can be used in particular to
00:04:29.23 accurately quantify protein levels
00:04:31.17 and understand how the proteome changes
00:04:34.05 across conditions.
00:04:35.09 Okay.
00:04:36.17 Every one of these lines is also essentially
00:04:38.22 a potential reporter for phenotypic screening
00:04:40.23 and so on.
00:04:42.05 So, one of the main reasons why this work has been done in yeast
00:04:46.12 is because we've had methods for a long time
00:04:49.07 that allow us to rewrite the genome of cerevisiae
00:04:51.27 fairly simply
00:04:53.14 and introduce these GFP fusions.
00:04:55.27 Now, the big revolution in human cells comes from,
00:04:59.02 you know, the era of CRISPR methods,
00:05:01.06 which were developed a few years ago.
00:05:03.28 Essentially, by cutting the genome at a specific locus,
00:05:07.22 CRISPR allows us to catalyze homologous recombination
00:05:10.22 and insert fluorescent proteins,
00:05:13.00 in frame, directly into a protein coding gene.
00:05:15.28 And so, now we can do the exact same experiments
00:05:18.24 that we were doing in yeast,
00:05:20.15 this time in human cells.
00:05:22.00 And so, many projects have started building fluorescent libraries
00:05:24.27 of human cells,
00:05:26.25 and I think a pioneer in this field is
00:05:29.17 The Allen Institute of Cell Science.
00:05:31.01 So, they're using GFP tagging in stem cells
00:05:33.29 to unveil cellular architecture.
00:05:35.17 They've tagged dozens of targets.
00:05:37.29 And they've used these cell lines
00:05:39.27 -- together with machine learning --
00:05:41.27 to build an integrated model of cellular architecture.
00:05:44.01 And here, again,
00:05:45.27 I'm gonna encourage you to check out their website
00:05:47.25 for up-to-date information.
00:05:50.09 In a separate project,
00:05:52.01 Jan Ellenberg and his team at EMBL
00:05:53.25 are using the fact that fluorescence
00:05:56.12 gives us very quantitative information about protein levels,
00:05:58.18 and using these, you know, tagged cell lines
00:06:01.13 to understand how the cell gets remodeled
00:06:04.16 during cell division,
00:06:06.02 and doing that, again, very precisely and quantitatively
00:06:08.06 so that we are able to plug in, you know,
00:06:11.14 physical models in these kind of experiments.
00:06:14.02 And in my own lab at the Biohub,
00:06:16.12 we're using a very similar approach.
00:06:18.14 In a project that we call OpenCell,
00:06:21.00 we're actually using a split GFP system
00:06:23.06 to facilitate high-throughput CRISPR tagging,
00:06:25.23 and you can read this paper
00:06:28.23 if you want to know more about this technique.
00:06:30.25 Essentially, that allows us to build
00:06:33.07 libraries of hundreds, and soon thousands,
00:06:35.13 of human cell lines tagged endogenously
00:06:38.03 with fluorescent proteins.
00:06:39.17 So, stay tuned.
00:06:41.00 We're approaching the final stages
00:06:42.15 of our first data release,
00:06:43.29 and I'll be very excited to tell you,
00:06:45.24 in the future,
00:06:47.15 what we learned from these experiments.
00:06:49.25 So, in a nutshell,
00:06:51.22 it's a very exciting time for cell engineering.
00:06:53.13 And I think it's just, you know,
00:06:55.07 a question of months or years
00:06:57.04 before all of us together
00:06:58.25 will be able to create genome-wide collections
00:07:01.15 of fluorescent cell lines in human cells,
00:07:04.08 just like we had in yeast.
00:07:06.14 Okay.
00:07:07.28 So, before I finish this part,
00:07:09.18 let me just mention that imaging
00:07:11.08 is not the only method
00:07:13.12 that allows us to profile localization.
00:07:14.29 Spatial proteomics,
00:07:16.16 which is a biochemical method, for example,
00:07:19.08 can localize cells by first isolating different organelles
00:07:22.07 in a cell lysate
00:07:24.04 by ultracentrifugation
00:07:25.13 and identifying proteins in each fraction
00:07:27.20 in a very quantitative manner
00:07:30.03 using mass spectrometry.
00:07:31.16 And that allows us to create organelle-level maps
00:07:33.27 of protein localization.
00:07:35.21 And here, again, is a great paper
00:07:37.27 that you can check out
00:07:40.09 to learn more about this.
00:07:41.20 So, overall, we have methods today
00:07:43.12 -- mostly using imaging, antibodies,
00:07:45.13 and CRISPR cell lines --
00:07:46.27 to be able to localize any protein in a human cell.
00:07:50.01 Now, the second step is going to be
00:07:53.03 understanding of how these different building blocks
00:07:55.00 are wired together to give the blueprint of a cell.
00:07:58.13 The first stage of wiring
00:08:00.11 happens at the level of protein complexes,
00:08:01.29 for example ATP synthase,
00:08:03.23 which is shown here,
00:08:05.19 which contains more than 20 different polypeptides
00:08:07.10 from 8 different subunits.
00:08:09.06 So, a lot of these proteins
00:08:10.29 are acting not on their own
00:08:12.18 but as part of, like,
00:08:14.04 these kind of big assemblies of protein complexes.
00:08:17.00 Now, to profile protein-protein interactions,
00:08:22.01 the most powerful method by far
00:08:24.00 is immunoprecipitation mass spectrometry
00:08:25.28 or IP/mass-spec.
00:08:27.14 So, in this method,
00:08:28.29 we're gonna take cells,
00:08:30.20 we're gonna break them open,
00:08:32.05 and very quickly go fishing for a protein of interest
00:08:34.29 -- usually using antibodies --
00:08:36.17 and use mass spectrometry
00:08:39.19 to identify which other proteins are coming along for the ride.
00:08:42.28 There are a few large-scale data sets in human cells
00:08:45.28 that are available using this method.
00:08:48.09 And a reference in this field is the Bioplex data set
00:08:51.07 by Steve Gygi, Wade Harper, and their colleagues at Harvard,
00:08:53.23 which have used overexpression
00:08:56.07 of epitope-tagged proteins
00:08:58.03 to profile the interactome of tens of thousands of targets.
00:09:00.09 So, this is an image from their first paper,
00:09:03.03 but they've kept building a great resource over the years,
00:09:06.13 and all of their data is available online.
00:09:10.05 Now, what I think is really interesting
00:09:12.14 is that it turns out that fluorescent proteins such as GFP
00:09:15.12 are also a formidable handle for immunoprecipitation.
00:09:18.10 And that's because we have very good antibodies,
00:09:20.14 or nanobodies,
00:09:22.02 against GFP.
00:09:23.13 So essentially, we can use the same cells
00:09:26.00 to localize proteins and define their interactions.
00:09:29.04 And this is an approach that for example has been used
00:09:31.24 by Matthias Mann, Tony Hyman, Marco Hein,
00:09:33.15 and their colleagues a few years ago
00:09:35.18 to define the interactome
00:09:38.11 of over 1000 different proteins
00:09:40.26 with very high precision.
00:09:42.11 And today, my lab is working with Matthias
00:09:44.22 to use the same techniques and profile the interactome
00:09:47.25 in our collection of CRISPR GFP cell lines.
00:09:51.03 And finally, beyond profiling just direct interactions,
00:09:54.06 there are methods that are available today
00:09:56.27 to measure proximity of proteins within a cell.
00:09:59.22 And in these methods
00:10:01.23 -- many of them were pioneered by Alice Ting in her lab --
00:10:04.19 instead of tagging proteins with fluorescent proteins,
00:10:08.13 we can fuse them with chemically active reporters.
00:10:12.01 And under these conditions,
00:10:13.19 if we treat the ce...
00:10:15.12 if we treat cells with the right molecules,
00:10:17.05 these chemical... these chemical reporters become active,
00:10:21.24 which leads to chemical marks being deposited
00:10:25.04 on proteins in the proximity of our target.
00:10:27.26 And because the radius of labeling, here,
00:10:30.24 is fairly small, just tens of nanometers,
00:10:33.01 that allows us actually to map proximity with fairly high precision.
00:10:36.02 Okay?
00:10:37.02 And we can use mass spec to identify
00:10:39.06 these chemical marks.
00:10:40.20 So, this is a method that is more recent
00:10:42.14 but has a lot of potential.
00:10:44.04 And the first large-scale study,
00:10:46.16 by Anne-Claude Gingras' lab,
00:10:48.21 just got published on bioRxiv,
00:10:51.08 and they just released the website.
00:10:53.24 So, we have a lot of mass spec methods
00:10:56.13 to profile wiring at the level of protein complexes.
00:10:58.26 And now the next,
00:11:02.09 and the final, stage of understanding how a cell is wired
00:11:05.09 is to be able to put, you know,
00:11:07.17 all of these functional units back into the context
00:11:09.15 of functional pathways,
00:11:11.03 for example signaling pathways
00:11:12.20 that shape cellular behavior.
00:11:14.22 And here, the method of choice
00:11:16.25 is gonna be to turn genes off or on
00:11:19.03 and measure how the cells react to it.
00:11:21.09 And we can cluster genes
00:11:23.08 based on the similarity of their phenotypes.
00:11:25.02 Okay?
00:11:26.05 For example, if the loss of...
00:11:27.23 the loss of gene A and the loss of gene B
00:11:30.20 elicit very similar phenotypes,
00:11:32.26 it's gonna be likely that these genes
00:11:35.06 are gonna be part of the same pathway.
00:11:36.26 Now, to take this a step further,
00:11:38.23 we can study what's called genetic interactions.
00:11:41.10 So, let's assume that the phenotype we're interested in here
00:11:45.26 is cell growth, just to keep things simple.
00:11:47.16 Okay?
00:11:49.07 We can take cells, we can turn off gene A,
00:11:51.14 measure its phenotype,
00:11:52.28 we can turn off gene B,
00:11:54.11 measure its phenotype
00:11:56.17 on its own, okay?
00:11:57.21 And then we can turn off both genes
00:12:00.01 in the same cell at the same time.
00:12:01.14 If the two genes are not interacting,
00:12:03.16 we're gonna expect that, you know,
00:12:05.06 the growth rate of this cell is essentially
00:12:08.19 gonna be the product of the two individual phenotypes.
00:12:10.09 Now, things start getting interesting
00:12:11.20 when the measurement deviates from this prediction.
00:12:14.17 For example, the double mutant
00:12:17.05 can be actually doing better than what we expect.
00:12:19.15 This is often the case when A and B
00:12:22.09 are actually part of the same linear pathway.
00:12:24.22 And the idea here is...
00:12:26.15 you know, it's the pathway that the cell is caring about.
00:12:28.27 So, you know, if we turn off A,
00:12:30.23 then there is not much of an additional phenotype
00:12:34.03 to turn off B at the same time,
00:12:36.09 because the pathway is already dead.
00:12:38.00 So, conversely,
00:12:40.04 the double mutant can be actually
00:12:43.12 doing far worse than what we expect.
00:12:44.18 Okay?
00:12:45.25 And that usually tells us that A and B
00:12:47.26 are acting into two parallel pathways
00:12:49.28 that are important for the same response,
00:12:51.26 so that the cell can cope with the loss of one of these branches,
00:12:55.08 but if both of them are turned off at the same time,
00:12:58.16 the cell really cannot... cannot...
00:13:00.22 cannot cope anymore.
00:13:02.24 Okay?
00:13:04.06 So, these are very simplified examples,
00:13:05.16 but that gives you kind of the idea of how genetic interactions
00:13:08.07 are very powerful at very precisely mapping functional pathways.
00:13:12.05 And again, one of the best examples
00:13:14.27 illustrating the power of these approaches
00:13:17.01 comes from the yeast Saccharomyces cerevisiae,
00:13:19.27 especially the seminal work of Charlie Boone,
00:13:22.10 Brenda Andrews,
00:13:24.01 and their colleagues in Toronto,
00:13:26.03 which have been able to map,
00:13:27.29 using an extremely large collection of double mutants,
00:13:30.03 every... or close to every pairwise combination
00:13:32.16 of genetic interactions in the yeast.
00:13:35.15 And they've used this information
00:13:37.06 to show how powerful genetic interactions
00:13:39.05 can be at grouping proteins
00:13:41.13 that belong to the same functional units.
00:13:44.04 And you can check one of their recent papers,
00:13:46.04 and especially one of their recent reviews,
00:13:48.22 to learn more about how genetic interactions
00:13:51.07 are a powerful way to understand pathways.
00:13:54.05 So, here again,
00:13:57.12 the reason why yeast has been so powerful
00:13:59.02 is that we know how to turn genes off or on.
00:14:01.20 In cerevisiae, we've been able to do this for a long time.
00:14:05.07 In the human cell,
00:14:07.06 the CRISPR revolution has also transformed our ability
00:14:09.20 to do that recently.
00:14:11.00 And especially, we now have tools
00:14:12.19 -- called CRISPRi for inactivation
00:14:15.05 or CRISPRa for activation --
00:14:16.19 that allow us to very specifically
00:14:19.13 turn genes on or off in human cells.
00:14:21.06 And most of these tools rely on
00:14:24.03 a catalytically inactive version of the Cas9 protein
00:14:26.11 that essentially acts as a protein platform
00:14:29.29 to recruit transcriptional repressors or transcriptional activators
00:14:34.15 to a gene of interest
00:14:36.07 to turn it off or to turn it on.
00:14:38.08 And a lot of this work has been developed in Jonathan Weissman's lab,
00:14:41.00 Stanley Qi’s and Luke Gilbert’s
00:14:42.29 at UCSF.
00:14:44.13 So, these methods are extremely powerful
00:14:46.28 and very easy to scale,
00:14:48.13 and so the first studies are coming out
00:14:50.16 that allow us to do very large data sets
00:14:53.26 of functional genomics in human cells.
00:14:56.00 And at the same time,
00:14:57.18 our ability to measure very complex phenotypes
00:14:59.12 -- high dimensional phenotypes --
00:15:01.09 is rapidly increasing.
00:15:02.21 And the rule of thumb here is that
00:15:05.10 the deeper the phenotype we can measure,
00:15:06.26 the more specific the clustering approach
00:15:09.06 to identify genes that belong to the same pathway becomes.
00:15:13.05 So, there's a lot of interest
00:15:15.25 in using the transcriptome of perturbed cells
00:15:17.26 as a very high-dimensional phenotype.
00:15:20.24 And there are methods that exist today
00:15:23.17 that can couple CRISPR-based perturbation
00:15:25.17 and single-cell RNAseq
00:15:27.02 to read perturbation and transcriptome
00:15:29.17 in the same very high-throughput experiment.
00:15:32.05 And if you want to learn more about
00:15:34.09 how these kinds of approaches are very useful,
00:15:36.21 especially in the context of genetic interactions,
00:15:38.21 here's a recent paper
00:15:41.19 that you can read.
00:15:43.10 And finally, there are image-based methods
00:15:46.07 that allow us to also get
00:15:48.28 very high-dimensional information about cell state.
00:15:51.07 I'm gonna call that measuring "cell morphology", okay?
00:15:54.12 We can use multiplex labeling approaches,
00:15:56.16 such as what we discussed a few slides ago,
00:15:59.13 but also label-free imaging,
00:16:01.04 for example brightfield or phase...
00:16:03.05 or phase imaging,
00:16:04.24 to get, you know, a fairly comprehensive view
00:16:07.10 of what a cell looks like,
00:16:08.20 and use machine learning to interpret
00:16:10.12 what that tells us about the state of a cell,
00:16:12.06 and transform that into a high-dimensional phenotype.
00:16:15.22 Okay.
00:16:17.20 So, now we have these methods,
00:16:19.14 that are kind of folding together,
00:16:19.15 that allow us for the first time
00:16:21.26 to do these very large-scale perturbation studies
00:16:22.26 in human cells as well.
00:16:25.09 Okay?
00:16:27.02 So, I'm gonna leave you on this,
00:16:28.23 and I think I've shown you that we have, now,
00:16:30.19 methods to study protein localization
00:16:32.22 or molecular interactions
00:16:34.19 and map functional pathways
00:16:37.17 with unprecedented... unprecedented scale and precision.
00:16:40.17 So, the next challenge
00:16:42.11 is going to be being able to bring all of these different approaches together
00:16:44.29 to get the complete wiring diagram of a human cell.
00:16:48.25 And I think it's a very exciting time for cell biology,
00:16:51.11 and I can't wait to see the next generation of studies.
00:16:54.24 Let me finish by thanking my lab
00:16:57.15 and my collaborators at the Biohub and beyond,
00:17:00.04 and thank you for watching.

This Talk

Speaker: Manuel Leonetti

Audience:

Educators of Adv. Undergrad / Grad
Researcher
Educators

Recorded: December 2019

Talk Overview

What if we could understand the human cell in such detail that we could paint an accurate representation of a cell’s molecular organization? In this lecture, Dr. Manuel Leonetti outlines the different genome-wide approaches that scientists are using to build a complete map of the human cellular architecture. Understanding protein networks and localization could aid our quest to understand human biology and disease.

Speaker Bio

Manuel Leonetti

Manuel Leonetti

Dr. Manuel Leonetti is a group leader at the Chan Zuckerberg Biohub. Leonetti completed his bachelor’s degree in Chemistry and his master’s degree in Interdisciplinary Life Sciences at the Ecole Normale Superieure in Paris. He continued his graduate studies in Molecular Neurobiology at The Rockefeller University under the supervision of Dr. Rod MacKinnon. In 2013,… Continue Reading

Comments

Dirk Schäfer says

July 14, 2021 at 1:41 am

An excellent introduction to cell biology ans its complexity!

Reply

Leave a Reply Cancel reply